Publication

CAN’T FIND THE MOUSE MODEL YOU NEED?
BUILD IT WITH
CRISPR!
From simple indel KOs to more complex conditional KOs and reporter insertions,
we’ve created a wide range of models using CRISPR/ Cas9 technology.
If we aren’t able to generate your CRISPR model, you don’t pay.
TO LEARN MORE,
VISIT JAX.ORG/CRISPR.
1.800.422.6423
1.207.288.5845
is
n ce ies
e g
i o sci nolo
r s e B Tech
o t
e ah gilen
S A
w
no
A MAJOR LEAP FORWARD IN PHENOTYPING

Determine the functional impact of genes, proteins
or compounds on metabolism in an hour
XF Cell Energy Phenotype Profile
Agilent Seahorse XFp Cell Energy Phenotype Test Kit Aerobic Energetic
Oxygen Consumption Rate (OCR)
Using as few as 15,000 live cells and an Agilent Seahorse XFp Analyzer,
Mitochondrial Respiration
Stressed
Phenotype
the Seahorse XFp Cell Energy Phenotype Test Kit identifies your cells’
l
ia
nt
te
Po
metabolic phenotype as well as their metabolic et
ab
ol
ic
M
potential — the utilization of mitochondrial
respiration and glycolysis. Baseline
Phenotype
LEARN MORE at www.agilent.com Quiescent Glycolytic
Glycolysis
Extracellular Acidification Rate (ECAR)

For Research Use Only. Not for use in diagnostic procedures.
Make Cell Biology Research A Piece of Cake
BioLegend’s Cell Biology products cover research areas including
cellular metabolism, structure, communications, signaling, epigenetics, Featured Products
and the cell cycle, and applications such as western blotting, Golgi: Giantin Antibody
immunoprecipitation, and immunofluorescence microscopy. Centrosome: Pericentrin Antibody
Cytoskeletal Filaments: Flash Phalloidin™ Localization Probes
Nucleus: Nuclear Pore Complex Proteins
Total protein extracts from Jurkat cells were probed with either and Lamin Antibodies
Direct-Blot™ HRP anti-STAT1 (clone 10C4B40) or purified anti-STAT1 Mitochondria: MitoSpy™ Localization Probes
followed by secondary antibody.
Functional Probes: TetraZ™ Cell Counting Kit
Receptors and gp130 Antibody
Signaling Proteins: STAT3 Antibody
Learn more at: Cell Tracking: Tag-it™ Proliferation and Cell Tracking Probes
biolegend.com/cell_bio_cake
BioLegend is ISO 9001:2008 and ISO 13485:2003 Certified

Toll-Free Tel: (US & Canada): 1.877.BIOLEGEND (246.5343)
Tel: 858.768.5800
biolegend.com
08-0061-06
World-Class Quality | Superior Customer Support | Outstanding Value

Identifying and Isolating Mesenchymal Stem Cells
Reliable Purification of Using a 100 μm nozzle tip, the Dual Sorting Modes on the a reduction in CD106 expression
BM-Derived MSCs S3e Cell Sorter was able to analyze S3e Cell Sorter May Further may indicate compromised stem
MSCs are rare in their resident tissues; and identify BM-derived MSCs from Improve Purity and Yield cell multipotency during in vitro
from Bone Marrow Cultures the frequency of MSCs in BM is 0.001–

0.01%, diluted in a heterogeneous mix
a mixed BM culture and sort target
cells at satisfying purity and efficiency.
Cells were sorted at up to 1,500 eps
While non-MSCs were sorted at
close to 100% purity, CD90+ CD73+
culture (Boxall and Jones 2012,
Liu et al. 2008). This could be caused
by a serum component in the culture
of hematopoietic cells (Bernardo et and CD90+ CD106+ dual-positive
Reliably Sort Stem Cell Populations with the Bio-Rad S3e Cell Sorter ™
al. 2009). When MSCs are isolated, with 30 psi sheath pressure and MSCs achieved only 95 and 92% medium, which induces stem cell
a heterogeneous pool of cells with 41 kHz drop drive frequency, and purity, respectively. There are two differentiation, or simply by the fact
various proliferation and differentiation the sort efficiency ranged from possible reasons for this: a majority that cells with less potential outgrow
Xiaoti Guo, PhD, Bio-Rad Laboratories, Inc., 2000 Alfred Nobel Drive, Hercules, CA 94547 91 to 98%. Note that the in vitro the multipotent MSC subset. As an
potentials can be obtained (Bianco et al. of the CD90+ cells are larger in
Bone marrow (BM) mesenchymal stem cells (MSCs) are multipotent adult stem cells with the capacity to differentiate into 2001). The rareness and heterogeneity expanded bone marrow MSCs are size (SSChigh) whereas CD90– cells alternative approach, single cell sorting
cells that compose adipose, bone, cartilage, and muscle tissue, thus presenting a wide potential for cell-based therapies. together make it difficult to identify, relatively large (20–35 μm) for the make up the majority of the SSClow using Single mode on the S3e Cell
According to the minimal criteria published by the International Society for Cellular Therapy and multiple studies (Boxall and isolate, and purify MSCs. The adherent 100 μm orifice. Therefore, to minimize population. The large cells (relative Sorter can be used to separate the
Jones 2012, Dominici et al. 2006), CD73, CD90, and CD105 are the primary markers used to identify MSCs across species, culture or panning approach is able to disturbance of the break-off point to the orifice diameter of the nozzle different stem cell subpopulations. The
and their expression levels tend to be stable during culture. In contrast, CD106 (VCAM-1) and STRO-1 have been shown to eliminate some of the hematopoietic and fanning of deflected streams, tip) likely have compromised sort single colony–derived cells can then
represent the most potent and undifferentiated MSCs, and they exhibit reduced expression levels after multiple passages cells; however, flow cytometry–based sort performance at a higher speed purity due to insufficient drop charge be cultured and evaluated separately.
(Boxall and Jones 2012, Halfon et al. 2011, Liu et al. 2008). For now, cell sorting by flow cytometry remains the most efficient cell sorting remains the most reliable was not assessed at current settings. and excessive side stream fanning Conclusions
approach to isolate MSCs from a highly heterogeneous environment like bone marrow. In this study, the S3e Cell Sorter was approach to isolating a pure population A nozzle tip with a larger diameter (Arnold and Lannigan 2010, Petersen With the use of appropriate surface
used to analyze and purify prospective MSCs from passage 1 (P1) and passage 3 (P3) rat bone marrow cultures. Sorting of MSCs for downstream cellular and would be more suitable for MSC sorts et al. 2003). Another possible factor markers, the S3e Cell Sorter is
performance was evaluated and results were compared between P1 and P3. The data presented here show the capability of molecular characterization. in future studies. contributing to the suboptimal purity capable of identifying bone marrow
the S3e Cell Sorter to isolate rare MSCs from a mixed bone marrow culture with satisfying sort efficiency and postsort purity. is that dual-positive cells are found MSCs from a mixed BM culture
with lower frequency (3% of target and sorting target cells at satisfying
population) than dual-negative ones purity and efficiency. Compared
(40–50%), and sorting low abundance to P1 cells, P3 rat MSCs display
Efficient Isolation of MSCs from CD106+ from 3 to 92% (Figure 2E). A B C
Table 1. Sort performance.*
populations can result in lower purity
a Heterogeneous Culture Non-MSCs were enriched from 40 to 256 256 256 Presort Frequency Postsort Purity Sort Efficiency, Theoretical
a more homogeneous phenotype
R1 R2 R3
Cell Marker Subset (% of parent) (% of parent) % Efficiency, %** and yield unless they are pre-enriched with an increased level of CD90 but
To identify potential MSCs and to test 99% (Figures 2F and 2G). (Arnold and Lannigan 2010, Kuka
the capability of the S3e Cell Sorter
192 192 192 CD90+ 40 99 98 98.2 decreased levels of CD90+ CD106+
The sorting performance for each and Ashwell 2013). The mixed mode
FSC height
and CD90+ CD73+ stem cell subsets,
SSC area
FSC area
CD90+
to purify MSCs from heterogeneous population is summarized in Table 1, 128 128 128 CD73+
15 95 91 97
sorting feature of the S3e Cell Sorter
BM culture, P1 rat BM cells were CD90–
suggesting that prolonged culture
which represents results from three CD73–
40 97 94 97.8 allows sorting of abundant populations may help remove contaminating cells
stained for CD90, CD73 or CD106, independent experiments. Theoretical
64 64 64
CD90+ in Purity mode and rare populations in
and CD45, and then with the viability CD106+
3 92 96.2 96.6 but also could compromise stem cell
efficiencies are calculated for a 0 0 0 Enrich mode. Hence, presorting rare phenotype and possibly affect their
CD90–
dye VivaFix™ 498/521. Stained cells 100 μm nozzle tip at 30 psi sheath
0 64 128 192 256 0 64 128 192 256 10 0 101 102 103 104
CD106–
50 99 92 98.4
populations in the Enrich followed by
were analyzed on the S3e Cell Sorter
FSC area FSC area CD45 FITC and VivaFix 498/521
* The size of rat MSCs range from 20 to 35 µm. P1 MSCs were sorted in Purity mode at 1,000–1,500 eps with a 100 µm nozzle tip,
pluripotency. An improved long-
pressure and various input rates and 30 psi sheath pressure, and 41 kHz drop drive frequency. Data represent three independent experiments.
Purity modes will likely help improve term cell culture strategy should be
equipped with a 100 μm nozzle tip and target cell frequencies with CytoCalc
Fig. 1. Rat MSC gating strategy. Rat bone marrow MSCs (P1) were stained with CD90 PerCP-Cy5.5, CD106 PE
** Theoretical efficiencies are calculated with CytoCalc Software v3.2 for a 100 µm nozzle tip at 30 psi sheath pressure at various MSC purity and yield.
30 psi sheath pressure. Viable MSCs
or CD73 ReadiLink 555/570, and CD45 FITC and VivaFix 498/521, then acquired on the S3e Cell Sorter. After input rates and target cell frequencies. investigated to minimize stem cell
Software v3.2. excluding debris and doublets (A–B), the fluorescence channel 1 (FL1) dim population was selected for further
differentiation before therapeutic
and non-MSCs were then sorted on analysis (C). FITC, fluorescein isothiocyanate; FSC, forward scatter; PE, phycoerythrin; PerCP, peridinin A B C Better Cell Culture and Isolation
the S3e Cell Sorter and assessed P3 Cells Reveal Less chlorophyll protein; SSC, side scatter. Rat MSCs: P1 Rat MSCs: P3
Methods Needed to Minimize applications of MSCs can be realized.
for postsort purity. Representative Phenotypic Heterogeneity A B C 80 CD90+
56.2 150
CD90+
86.4
120 CD73+
23.3 Loss of Stem Cell Phenotypes Acknowledgements
FL2 FMO Control CD90/CD73 Sorting CD90/CD106 Sorting
graphs in Figure 1 illustrate the To test whether rat bone marrow 104
FL2+ CD90+
104
CD90+ CD73+
104
CD90+ CD106+ 60 90 The P1 cells used in this study The author would like to acknowledge Drs. Haixiang
Liang and Daniel A Grande (The Feinstein Institute for
MSCs are able to maintain their 100
were the first passage of Sprague-
CD73 ReadiLink 555/570
following gating strategy. Since P1
Count
103 103 103 60
Medical Research) for kindly providing P1 and P3 rat
40
bone marrow displays a relatively high phenotypic characteristics after Dawley (SD) rat BM, in which a MSCs, Drs. Qing Xiang and Ralph J Garippa (Memorial
CD106 PE
50 Sloan Kettering Cancer Center) for their generous

level of heterogeneity, a broad gate in vitro expansion, P1 and P3 bone 20 30 significant amount of contaminating
FL2
102 102 102 support with equipment, Michael Kissner (Propel Labs)
was applied to include both SSClow marrow MSCs were stained and FL2– CD90– CD90– CD73–
CD90– CD106– 0 0
hematopoietic cells remained in the for his helpful technical discussions, and Brandon
0 Williams and Melissa Ma (Bio-Rad Laboratories) for
and SSChigh populations (Figure 1A). analyzed side by side on the S3e Cell 101 101 101 103 104 105 10 6 103 104 105 10 6 102 103 104 105 10 6
culture, contributing mainly to the their help with manuscript preparation.
CD90 (gated on live cells) CD90 (gated on live cells) CD73 (gated on live CD90+ cells)
Following doublet exclusion (Figure 1B) Sorter to compare their expression CD90– population. After two more
10 0 10 0 10 0 D E F References
viable (VivaFixdim) and CD45neg of MSC-related surface markers 10 0 101 102 103 104 10 0 101 10 2 103 104 10 0 101 102 103 104 passages in 5 weeks, most of the Arnold LW and Lannigan J (2010). Curr Protoc Cytom
CD90 PerCP-Cy5.5 CD90 PerCP-Cy5.5 CD90 PerCP-Cy5.5 120
(Figure 3 and Table 2). Compared 300 Chapter 1:Unit 1.24.1–1.24.30.
populations were gated for stem cell CD73+ CD106+
250
CD106+ hematopoietic cells were removed
0.92 6.41 0.018
D E Bernardo ME et al. (2009). Ann N Y Acad Sci 1176,
surface marker expression (Figure 1C). with P1 MSCs, P3 cells revealed 90 200
from the culture, which greatly 101–117.
104 104
CD90+ CD73+ CD90+ CD106+
As shown in Figure 2, about 15% less heterogeneity in morphology as 200
increased the frequency of the CD90+ Bianco P et al. (2001). Stem Cells 19, 180–192.
Count
150
60
of gated cells stain positive for both determined by microscopy, and the 103 103 population in P3. But the expression Boxall SA and Jones E (2012). Stem Cells Int 2012,
100 100 975871.
SSChigh population increased from
CD106 PE
CD90 and CD73 (Figure 2B) and 30 of two other stem cell markers, CD73
102 102 50 Dominici M et al. (2006). Cytotherapy 8, 315–317.
3% coexpress CD90 and CD106 50% in P1 to 90% in P3 (data not CD90– CD73– CD90– CD106–
and CD106, drastically decreased
0 0 0
Halfon S et al. (2011). Stem Cells Dev 20, 53–66.
(Figure 2C), as determined by the shown). These more homogeneous 101 101 102 103 104 105 10 6 102 103 104 105 10 6 102 103 104 105 10 6 in P3 cells. This finding is consistent Kuka M and Ashwell JD (2013). J Immunol Methods
phycoerythrin (PE) fluorescence P3 cells expressed a greater level Fig. 2. Rat MSC sorting. Rat bone
CD73 (gated on live CD90+ cells) CD106 (gated on live CD90+ cells) CD106 (gated on live CD90+ cells)
with other reports that showed that 400–401, 111–116.
10 0 10 0
minus one (FMO) control (Figure 2A). of CD90 compared to P1 cells marrow MSCs (P1) were stained
10 0 101 102 103 104 10 0 101 102 103 104 Fig. 3. Characterization of P1 vs. P3 rat bone marrow MSCs. Rat bone marrow MSCs were stained with CD106 expression in MSCs was Liu F et al. (2008). J Bone Miner Metab 26, 312–320.
with CD90 PerCP-Cy5.5, CD106 PE Petersen TW et al. (2003). Cytometry A 56, 63–70.
The dual-positive cells are prospective (86 vs. 56%), but the CD73 and CD106 or CD73 ReadiLink 555/570, and
CD90 PerCP-Cy5.5 CD90 PerCP-Cy5.5 anti-CD90 PE-Cy7, CD106 PE or CD73 ReadiLink 555/570, and CD45 FITC and VivaFix 498/521, then downregulated by tenfold after
acquired on the S3e Cell Sorter. Live CD90+ (A–B), CD90+ CD73+ (C–D), and CD90+ CD106+ (E–F)
MSCs and the dual-negative ones are expression within the CD90+ subset CD45 FITC and VivaFix 498/521, F G
subsets were compared between P1 and P3 MSCs. Gray traces indicate FMO control for the PE channel. 2 weeks of osteogenic differentiation Cy is a trademark of GE Healthcare. CytoCalc is a
then sorted on the S3e Cell Sorter trademark of Applera Corporation. ReadiLink is a
likely contaminating hematopoietic was drastically reduced (from 23 to for CD90+ CD73+, CD90– CD73–
104
CD90+ CD73+
104
G CD90+ CD106+
The frequency of each cell marker subset is shown in Table 2. (Liu et al. 2008). Even without trademark of AAT Bioquest, Inc.
0.9% for CD73+; 6.4 to 0.02% for induction, the expression of CD106
and stromal cells. Both prospective (B) and CD90+ CD106+, CD90–
103 103 Table 2. Frequency of P1 vs. P3 rat bone marrow MSCs.
CD106– (C) cell marker subsets
MSCs and non-MSCs were then CD106+). These data suggest that Cell Marker Subset P1 Frequency, % P3 Frequency, % could also decrease at a later passage
CD106 PE
within FL1 dim/neg populations.

sorted on the S3e Cell Sorter at up to although in vitro expansion may help Gates were set according to PE 102 102 CD90+ 56 86 compared with an earlier culture
eliminate contaminating cells from the FMO control (A). Cells were sorted CD90– CD73– CD90– CD106– CD90+
CD106 PE
1,500 events per second (eps) using CD73+

23 0.9 (Halfon et al. 2011). As it is a surface
in Purity mode and reanalyzed 101 101
Purity mode. After sorting, a CD90+ culture, it may also compromise stem for purity (D–G; Table 1). FL2,
CD90+
CD106+
6.4 0.02 marker considered to be linked with
Bio-Rad Laboratories, Inc.
CD73+ population was enriched from cell phenotype by downregulating fluorescence channel 2; PE,
10 0 10 0 stem cell differentiation potential,
phycoerythrin; PerCP, peridinin 10 0 101 102 103 104 10 0 101 102 103 104 Visit bio-rad.com/info/S3eMSC-ASCB for
15 to 95% (Figure 2D), and CD90+ stem cell marker expression. chlorophyll protein. CD90 PerCP-Cy5.5 CD90 PerCP-Cy5.5 more information.
16-8119_CBD_S3e_MSC_Advertorial_Elsevier_Cell_ASCB_FINAL.indd All Pages 11/17/16 1:34 PM
Job # 16-8119 Publication Elsevier Cell, ASCB 2016 Trim Size 210 mm x 280 mm Run Date 12/01/16
Identifying and Isolating Mesenchymal Stem Cells
Reliable Purification of Using a 100 μm nozzle tip, the Dual Sorting Modes on the a reduction in CD106 expression
BM-Derived MSCs S3e Cell Sorter was able to analyze S3e Cell Sorter May Further may indicate compromised stem
MSCs are rare in their resident tissues; and identify BM-derived MSCs from Improve Purity and Yield cell multipotency during in vitro
from Bone Marrow Cultures the frequency of MSCs in BM is 0.001–

0.01%, diluted in a heterogeneous mix
a mixed BM culture and sort target
cells at satisfying purity and efficiency.
Cells were sorted at up to 1,500 eps
While non-MSCs were sorted at
close to 100% purity, CD90+ CD73+
culture (Boxall and Jones 2012,
Liu et al. 2008). This could be caused
by a serum component in the culture
of hematopoietic cells (Bernardo et and CD90+ CD106+ dual-positive
Reliably Sort Stem Cell Populations with the Bio-Rad S3e Cell Sorter ™
al. 2009). When MSCs are isolated, with 30 psi sheath pressure and MSCs achieved only 95 and 92% medium, which induces stem cell
a heterogeneous pool of cells with 41 kHz drop drive frequency, and purity, respectively. There are two differentiation, or simply by the fact
various proliferation and differentiation the sort efficiency ranged from possible reasons for this: a majority that cells with less potential outgrow
Xiaoti Guo, PhD, Bio-Rad Laboratories, Inc., 2000 Alfred Nobel Drive, Hercules, CA 94547 91 to 98%. Note that the in vitro the multipotent MSC subset. As an
potentials can be obtained (Bianco et al. of the CD90+ cells are larger in
Bone marrow (BM) mesenchymal stem cells (MSCs) are multipotent adult stem cells with the capacity to differentiate into 2001). The rareness and heterogeneity expanded bone marrow MSCs are size (SSChigh) whereas CD90– cells alternative approach, single cell sorting
cells that compose adipose, bone, cartilage, and muscle tissue, thus presenting a wide potential for cell-based therapies. together make it difficult to identify, relatively large (20–35 μm) for the make up the majority of the SSClow using Single mode on the S3e Cell
According to the minimal criteria published by the International Society for Cellular Therapy and multiple studies (Boxall and isolate, and purify MSCs. The adherent 100 μm orifice. Therefore, to minimize population. The large cells (relative Sorter can be used to separate the
Jones 2012, Dominici et al. 2006), CD73, CD90, and CD105 are the primary markers used to identify MSCs across species, culture or panning approach is able to disturbance of the break-off point to the orifice diameter of the nozzle different stem cell subpopulations. The
and their expression levels tend to be stable during culture. In contrast, CD106 (VCAM-1) and STRO-1 have been shown to eliminate some of the hematopoietic and fanning of deflected streams, tip) likely have compromised sort single colony–derived cells can then
represent the most potent and undifferentiated MSCs, and they exhibit reduced expression levels after multiple passages cells; however, flow cytometry–based sort performance at a higher speed purity due to insufficient drop charge be cultured and evaluated separately.
(Boxall and Jones 2012, Halfon et al. 2011, Liu et al. 2008). For now, cell sorting by flow cytometry remains the most efficient cell sorting remains the most reliable was not assessed at current settings. and excessive side stream fanning Conclusions
approach to isolate MSCs from a highly heterogeneous environment like bone marrow. In this study, the S3e Cell Sorter was approach to isolating a pure population A nozzle tip with a larger diameter (Arnold and Lannigan 2010, Petersen With the use of appropriate surface
used to analyze and purify prospective MSCs from passage 1 (P1) and passage 3 (P3) rat bone marrow cultures. Sorting of MSCs for downstream cellular and would be more suitable for MSC sorts et al. 2003). Another possible factor markers, the S3e Cell Sorter is
performance was evaluated and results were compared between P1 and P3. The data presented here show the capability of molecular characterization. in future studies. contributing to the suboptimal purity capable of identifying bone marrow
the S3e Cell Sorter to isolate rare MSCs from a mixed bone marrow culture with satisfying sort efficiency and postsort purity. is that dual-positive cells are found MSCs from a mixed BM culture
with lower frequency (3% of target and sorting target cells at satisfying
population) than dual-negative ones purity and efficiency. Compared
(40–50%), and sorting low abundance to P1 cells, P3 rat MSCs display
Efficient Isolation of MSCs from CD106+ from 3 to 92% (Figure 2E). A B C
Table 1. Sort performance.*
populations can result in lower purity
a Heterogeneous Culture Non-MSCs were enriched from 40 to 256 256 256 Presort Frequency Postsort Purity Sort Efficiency, Theoretical
a more homogeneous phenotype
R1 R2 R3
Cell Marker Subset (% of parent) (% of parent) % Efficiency, %** and yield unless they are pre-enriched with an increased level of CD90 but
To identify potential MSCs and to test 99% (Figures 2F and 2G). (Arnold and Lannigan 2010, Kuka
the capability of the S3e Cell Sorter
192 192 192 CD90+ 40 99 98 98.2 decreased levels of CD90+ CD106+
The sorting performance for each and Ashwell 2013). The mixed mode
FSC height
and CD90+ CD73+ stem cell subsets,
SSC area
FSC area
CD90+
to purify MSCs from heterogeneous population is summarized in Table 1, 128 128 128 CD73+
15 95 91 97
sorting feature of the S3e Cell Sorter
BM culture, P1 rat BM cells were CD90–
suggesting that prolonged culture
which represents results from three CD73–
40 97 94 97.8 allows sorting of abundant populations may help remove contaminating cells
stained for CD90, CD73 or CD106, independent experiments. Theoretical
64 64 64
CD90+ in Purity mode and rare populations in
and CD45, and then with the viability CD106+
3 92 96.2 96.6 but also could compromise stem cell
efficiencies are calculated for a 0 0 0 Enrich mode. Hence, presorting rare phenotype and possibly affect their
CD90–
dye VivaFix™ 498/521. Stained cells 100 μm nozzle tip at 30 psi sheath
0 64 128 192 256 0 64 128 192 256 10 0 101 102 103 104
CD106–
50 99 92 98.4
populations in the Enrich followed by
were analyzed on the S3e Cell Sorter
FSC area FSC area CD45 FITC and VivaFix 498/521
* The size of rat MSCs range from 20 to 35 µm. P1 MSCs were sorted in Purity mode at 1,000–1,500 eps with a 100 µm nozzle tip,
pluripotency. An improved long-
pressure and various input rates and 30 psi sheath pressure, and 41 kHz drop drive frequency. Data represent three independent experiments.
Purity modes will likely help improve term cell culture strategy should be
equipped with a 100 μm nozzle tip and target cell frequencies with CytoCalc
Fig. 1. Rat MSC gating strategy. Rat bone marrow MSCs (P1) were stained with CD90 PerCP-Cy5.5, CD106 PE
** Theoretical efficiencies are calculated with CytoCalc Software v3.2 for a 100 µm nozzle tip at 30 psi sheath pressure at various MSC purity and yield.
30 psi sheath pressure. Viable MSCs
or CD73 ReadiLink 555/570, and CD45 FITC and VivaFix 498/521, then acquired on the S3e Cell Sorter. After input rates and target cell frequencies. investigated to minimize stem cell
Software v3.2. excluding debris and doublets (A–B), the fluorescence channel 1 (FL1) dim population was selected for further
differentiation before therapeutic
and non-MSCs were then sorted on analysis (C). FITC, fluorescein isothiocyanate; FSC, forward scatter; PE, phycoerythrin; PerCP, peridinin A B C Better Cell Culture and Isolation
the S3e Cell Sorter and assessed P3 Cells Reveal Less chlorophyll protein; SSC, side scatter. Rat MSCs: P1 Rat MSCs: P3
Methods Needed to Minimize applications of MSCs can be realized.
for postsort purity. Representative Phenotypic Heterogeneity A B C 80 CD90+
56.2 150
CD90+
86.4
120 CD73+
23.3 Loss of Stem Cell Phenotypes Acknowledgements
FL2 FMO Control CD90/CD73 Sorting CD90/CD106 Sorting
graphs in Figure 1 illustrate the To test whether rat bone marrow 104
FL2+ CD90+
104
CD90+ CD73+
104
CD90+ CD106+ 60 90 The P1 cells used in this study The author would like to acknowledge Drs. Haixiang
Liang and Daniel A Grande (The Feinstein Institute for
MSCs are able to maintain their 100
were the first passage of Sprague-
following gating strategy. Since P1
Count
103 103 103 60
Medical Research) for kindly providing P1 and P3 rat
40
bone marrow displays a relatively high phenotypic characteristics after Dawley (SD) rat BM, in which a MSCs, Drs. Qing Xiang and Ralph J Garippa (Memorial
CD106 PE
50 Sloan Kettering Cancer Center) for their generous

level of heterogeneity, a broad gate in vitro expansion, P1 and P3 bone 20 30 significant amount of contaminating
FL2
102 102 102 support with equipment, Michael Kissner (Propel Labs)
was applied to include both SSClow marrow MSCs were stained and FL2– CD90– CD90– CD73–
CD90– CD106– 0 0
hematopoietic cells remained in the for his helpful technical discussions, and Brandon
0 Williams and Melissa Ma (Bio-Rad Laboratories) for
and SSChigh populations (Figure 1A). analyzed side by side on the S3e Cell 101 101 101 103 104 105 10 6 103 104 105 10 6 102 103 104 105 10 6
culture, contributing mainly to the their help with manuscript preparation.
CD90 (gated on live cells) CD90 (gated on live cells) CD73 (gated on live CD90+ cells)
Following doublet exclusion (Figure 1B) Sorter to compare their expression CD90– population. After two more
10 0 10 0 10 0 D E F References
viable (VivaFixdim) and CD45neg of MSC-related surface markers 10 0 101 102 103 104 10 0 101 10 2 103 104 10 0 101 102 103 104 passages in 5 weeks, most of the Arnold LW and Lannigan J (2010). Curr Protoc Cytom
CD90 PerCP-Cy5.5 CD90 PerCP-Cy5.5 CD90 PerCP-Cy5.5 120
(Figure 3 and Table 2). Compared 300 Chapter 1:Unit 1.24.1–1.24.30.
populations were gated for stem cell CD73+ CD106+
250
CD106+ hematopoietic cells were removed
0.92 6.41 0.018
D E Bernardo ME et al. (2009). Ann N Y Acad Sci 1176,
surface marker expression (Figure 1C). with P1 MSCs, P3 cells revealed 90 200
from the culture, which greatly 101–117.
104 104
CD90+ CD73+ CD90+ CD106+
As shown in Figure 2, about 15% less heterogeneity in morphology as 200
increased the frequency of the CD90+ Bianco P et al. (2001). Stem Cells 19, 180–192.
Count
150
60
of gated cells stain positive for both determined by microscopy, and the 103 103 population in P3. But the expression Boxall SA and Jones E (2012). Stem Cells Int 2012,
100 100 975871.
SSChigh population increased from
CD106 PE
CD90 and CD73 (Figure 2B) and 30 of two other stem cell markers, CD73
102 102 50 Dominici M et al. (2006). Cytotherapy 8, 315–317.
3% coexpress CD90 and CD106 50% in P1 to 90% in P3 (data not CD90– CD73– CD90– CD106–
and CD106, drastically decreased
0 0 0
Halfon S et al. (2011). Stem Cells Dev 20, 53–66.
(Figure 2C), as determined by the shown). These more homogeneous 101 101 102 103 104 105 10 6 102 103 104 105 10 6 102 103 104 105 10 6 in P3 cells. This finding is consistent Kuka M and Ashwell JD (2013). J Immunol Methods
phycoerythrin (PE) fluorescence P3 cells expressed a greater level Fig. 2. Rat MSC sorting. Rat bone
CD73 (gated on live CD90+ cells) CD106 (gated on live CD90+ cells) CD106 (gated on live CD90+ cells)
with other reports that showed that 400–401, 111–116.
10 0 10 0
minus one (FMO) control (Figure 2A). of CD90 compared to P1 cells marrow MSCs (P1) were stained
10 0 101 102 103 104 10 0 101 102 103 104 Fig. 3. Characterization of P1 vs. P3 rat bone marrow MSCs. Rat bone marrow MSCs were stained with CD106 expression in MSCs was Liu F et al. (2008). J Bone Miner Metab 26, 312–320.
with CD90 PerCP-Cy5.5, CD106 PE Petersen TW et al. (2003). Cytometry A 56, 63–70.
The dual-positive cells are prospective (86 vs. 56%), but the CD73 and CD106 or CD73 ReadiLink 555/570, and
CD90 PerCP-Cy5.5 CD90 PerCP-Cy5.5 anti-CD90 PE-Cy7, CD106 PE or CD73 ReadiLink 555/570, and CD45 FITC and VivaFix 498/521, then downregulated by tenfold after
acquired on the S3e Cell Sorter. Live CD90+ (A–B), CD90+ CD73+ (C–D), and CD90+ CD106+ (E–F)
MSCs and the dual-negative ones are expression within the CD90+ subset CD45 FITC and VivaFix 498/521, F G
subsets were compared between P1 and P3 MSCs. Gray traces indicate FMO control for the PE channel. 2 weeks of osteogenic differentiation Cy is a trademark of GE Healthcare. CytoCalc is a
then sorted on the S3e Cell Sorter trademark of Applera Corporation. ReadiLink is a
likely contaminating hematopoietic was drastically reduced (from 23 to for CD90+ CD73+, CD90– CD73–
104
CD90+ CD73+
104
G CD90+ CD106+
The frequency of each cell marker subset is shown in Table 2. (Liu et al. 2008). Even without trademark of AAT Bioquest, Inc.
0.9% for CD73+; 6.4 to 0.02% for induction, the expression of CD106
and stromal cells. Both prospective (B) and CD90+ CD106+, CD90–
103 103 Table 2. Frequency of P1 vs. P3 rat bone marrow MSCs.
CD106– (C) cell marker subsets
MSCs and non-MSCs were then CD106+). These data suggest that Cell Marker Subset P1 Frequency, % P3 Frequency, % could also decrease at a later passage
CD106 PE
within FL1 dim/neg populations.

sorted on the S3e Cell Sorter at up to although in vitro expansion may help Gates were set according to PE 102 102 CD90+ 56 86 compared with an earlier culture
eliminate contaminating cells from the FMO control (A). Cells were sorted CD90– CD73– CD90– CD106– CD90+
CD106 PE
1,500 events per second (eps) using CD73+

23 0.9 (Halfon et al. 2011). As it is a surface
in Purity mode and reanalyzed 101 101
Purity mode. After sorting, a CD90+ culture, it may also compromise stem for purity (D–G; Table 1). FL2,
CD90+
CD106+
6.4 0.02 marker considered to be linked with
Bio-Rad Laboratories, Inc.
CD73+ population was enriched from cell phenotype by downregulating fluorescence channel 2; PE,
10 0 10 0 stem cell differentiation potential,
phycoerythrin; PerCP, peridinin 10 0 101 102 103 104 10 0 101 102 103 104 Visit bio-rad.com/info/S3eMSC-ASCB for
15 to 95% (Figure 2D), and CD90+ stem cell marker expression. chlorophyll protein. CD90 PerCP-Cy5.5 CD90 PerCP-Cy5.5 more information.
16-8119_CBD_S3e_MSC_Advertorial_Elsevier_Cell_ASCB_FINAL.indd All Pages 11/17/16 1:34 PM
Job # 16-8119 Publication Elsevier Cell, ASCB 2016 Trim Size 210 mm x 280 mm Run Date 12/01/16
Impress Yourself
The new Eppendorf Cell Culture Consumables
The all new line of Eppendorf Cell > Unsurpassed quality, clarity, purity
Culture Consumables will truly delight and sterility, providing reliable cell
your cells. The outstanding design, culture conditions
reliability and purity is based on more > Significantly improved design for
than 50 years of experience. Products more safety and consistency
created by experts, developed for > Maximum safety and confidence
perfectionists. Impress yourself! during storage and transportation
ccc.eppendorf.com • 800-645-3050
052.A1.0101.D © 2017 Eppendorf AG.
Full page journal ad Cell - CEL

052.A1.0101.D.US-CEL.indd 1 11/22/2016 6:51:41 AM
RnDSy-lu-2945
Expedite Your Research

by partnering with our industry-renowned product development and biomarker
testing scientists to deliver unique reagents and reliable data.
Custom Services
• Get custom products tailored
to your exact specifications
• Outsource troubleshooting and

QA oversight to the experts
• Deliver your project on time

and on budget
Setting the standard in quality

research reagents and support
for over 30 years.
Learn more | rndsystems.com/customservices
Bosto_std_4C
RnDSy-2945
Tocri-2945
Global info@bio-techne.com bio-techne.com/find-us/distributors TEL +1 612 379 2956

North America TEL 800 343 7475 Europe | Middle East | Africa TEL +44 (0)1235 529449
China info.cn@bio-techne.com TEL +86 (21) 52380373 bio-techne.com
Foreword
Four years ago, Cell Press launched the “ Best of” reprint collections across a number of our journals, including Cell. We are
pleased to now bring you a new Best of Cell that focuses on articles published over the course of 2016. For this edition, we made
our selection by looking at review and research articles with the most full-text HTML and PDF downloads since publication, then
choosing a representative group from each of the four published volumes. We’ ve also included a selection of frontmatter articles
and a collection of SnapShots that highlight some of the most impactful biology that we’ ve had the privilege to publish this year.
This collection covers a broad range of topics and includes papers that feature some of the exciting experimental approaches,
such as genome engineering with CRISPR, disease modeling with organoids, an examination of the cellular phase of Alzheimer’ s,
and breaking cryo-EM resolution boundaries to facilitate drug discovery, that have helped drive science forward in recent years.
We recognize that no one measurement can be indicative of the “ best” papers over a given period of time, especially when the
articles are relatively new and their true significance may still need time to be established. We’ ve selected several papers that were
published later in the year and that may not have had the time to develop quite as significant usage statistics but show promise as
groundbreaking in their own right. We hope that this combination of approaches to highlighting articles will give you an overview
of different perspectives on the science that we published during 2016.
We hope that you will enjoy reading this special collection and that you will visit www.cell.com/cell to see the latest findings that we
have had the privilege to publish. Also, be sure to visit www.cell.com to find other high-quality articles published in the full collection
of Cell Press journals. If you prefer to receive your updates via social media, you may want to follow us on Twitter (@ CellCellPress)
or like our Facebook page (www.facebook.com/CellCellPress/).
Last but definitely not least, we are grateful for the generosity of our many sponsors, who helped to make this reprint collection
possible.
For information for the Best of Series, please contact:

Jonathan Christison
Program Director, Best of Cell Press
e: jchristison@ cell.com
p: 617-397-2893
t: @ cellPressBiz
Discover what’s
beneath.
Explore the depths of amplicon sequencing with the
NEXTflex™ Amplicon Panels. The NEXTflex Amplicon
Panels are complete solutions for your targeted
resequencing needs. These panels are designed
to detect and screen relevant germline and
somatic mutations using either Illumina or Ion
Torrent-based sequencing platforms. Visit
BiooScientific.com/panels to learn how the
ready-to-use predesigned amplicon panels
or validated NEXTflex™ custom solutions
simplify targeted sequencing protocols.
AMPLICON PANELS
Custom
Mediterranean Fever
BRCA 1/2
Obesity
Colon
TP53
CEBPA
Phenylketonuria
Your Discovery Awaits.

Visit biooscientific.com/panels to learn more.
For research use only. Not for use in diagnostic procedures.
THE NGS EXPERTS™

PipetmaX® + qPCR Assistant
Now everyone in your lab can
automate qPCR protocols
 Easy to set up, no programming required
 Create flexible, single and multiplex
qPCR methods
 Eliminate data entry errors with sample import
and export
 Save time by combining multiple methods
into one run
 Standard Amplification Curves for

COX gene. Standard cDNA samples
were serially diluted 5-fold and
amplified in 5 replicates (dilutions left
to right: 5x, 25x, 125x, 625x and 3125x).
instruments • accessories • software • solutions  Meet PipetmaX and your new qPCR Assistant at www.gilson.com
Visit www.gilson.com to download Speaking Volumes, our interactive publication for your tablet!
16GIP054_PIPETMAXqPCR_ad_Cell_asm1.indd 2 6/20/16 1:36 PM

NEW!
See More Than Before

Introducing the Eclipse Ti2, Nikon’s all-new inverted
microscope platform for advanced imaging.
Nikon Instruments has raised the bar for core imaging capability
with the release of the new Eclipse Ti2 Inverted Research
Microscope. With its unprecedented 25mm field of view, you
can capture twice as much data with a single image. The fourth
generation Perfect Focus System combined with a completely
redesigned Z-drive provide a super-stable imaging platform with
high-precision focusing for demanding applications. Built-in sensors
and intelligent Assist Guide eliminate user error while the internal
detector enables you to see the back-aperture for alignments.
There’s so much more! For more information, contact your local

Nikon sales representative, call 1-800-52-NIKON (1-800-526-4566)
or visit www.nikon-ti2.com. The new microscope will also be
featured at the Nikon booth at the ASCB Annual Meeting.
nikon-instruments-inc. nikoninstrumentsinc
Nikon Instruments Inc. n www.nikoninstruments.com n nikoninstruments@nikon.net ®
NIK-5959 Ti2 Full Page Ad_R5.indd 1 11/18/16 1:50 PM

BIOSPA SYSTEM F O R L I V E C E L L I M A G I N G
More Than an IncuCyte®

BioSpa™ System IncuCyte® Zoom
Live cell imaging
Temperature, humidity, gas control Integrated, monitored Incubator required
Labware capacity 8 6
Objective capacity 6 1
Magnification range 1.25x to 60x 4x to 20x
Fluorescence colors available >15 2
Z-stacking for 3D biology
Laser autofocus
Advanced cell-level analysis
Washing, media replacement
Reagent dispensing
*BioSpa system includes BioSpa 8 and Cytation 5
Learn more. Visit www.biotek.com/biospasystem
IncuCyte® ZOOM is a registered trademark of Essen BioScience www.biotek.com
BioSpa System ad-Cell Press.indd 1 10/17/16 2:44 PM

PHYSIOLOGICAL
RUSKINN
CELL CULTURE
WORKSTATIONS
CULTURE AS
NATURE INTENDED
I n v i v O2
TOGETHER
WE CAN MAKE
MORE THAN
A DIFFERENCE
WE CAN
MAKE HISTORY
Since 1998 Baker Ruskinn have

been working with pioneering
scientists around the world to make
breakthroughs in cell research.
Get in touch today to hear how we can work with you: U.S.: (800) 992-2537 | sales@ruskinn.com | www.bakerruskinn.com
3D cell culture system
Nano Culture Plate

Magnify Your Nanoculture
Gel or
Biological-
matrix
Free Easy
High Handling
Cell Viability
& HTS
Motility Compatible
Suitable for
Wide Variety in Cell Imager
Sizes of Plate &
Widespread Scanner
Applications
depending on
your target
EMT Induction
ORGANOGENIX
mblintl.com
EMT Inhibition
MC-RUO-044
Aurora B Phospho pT232 Antibody
Lilian Kabeche, PhD.
Massachusetts General Hospital
Do You Know Where Your Antibodies Are Coming From?

Rockland is a frequently referenced • Extensive line of antibodies and reagents
antibody manufacturer and collaborates • Non-normalized antibodies for higher
with leading organizations such as the concentrations
National Cancer Institute, St. Jude • Multi-assay validated products
Children’s Research Hospital, and
• Custom antibody production and assay
Lankenau Institute for Medical Research
to develop high-quality antibodies and development
antibody based tools. Rockland is a • Published staff scientists
name synonymous with reliability, • USDA accredited animal facilities
specificity, and sensitivity. • Trial size antibodies available
Protect your experiments with

Rockland antibodies.
www.rockland-inc.com 1-800-656-7625
Best of 2016
SnapShots
Transcription Elongation Christine E. Cucinotta and Karen M. Arndt
Expanding the Ubiquitin Code Krishnaraj Rajalingam and Ivan Dikic
Epigenomic Assays Martin Krzywinski and Martin Hirst
Signaling in Symbiosis Sharon R. Long
The Bacterial Cytoskeleton Gero Fink, Andrzej Szewczak-Harris, and Jan Löwe
Neuronal Regulation of Aging Heather J. Weir and William B. Mair
Microglia in Disease Simon Beggs and Michael W. Salter
Types of Ubiquitin Ligases Francesca Ester Morreale and Helen Walden
Reviews
Biology and Applications of CRISPR Systems: Addison V. Wright, James K. Nuñez, and Jennifer A. Doudna
Harnessing Nature’s Toolbox for Genome Engineering
The Basis of Oncoimmunology A. Karolina Palucka and Lisa M. Coussens
Modeling Development and Disease with Organoids Hans Clevers
The Cellular Phase of Alzheimer’s Disease Bart De Strooper and Eric Karran
Articles and Resources

Mitochondrial Dynamics Controls T Cell Fate through Michael D. Buck, David O’Sullivan, Ramon I. Klein Geltink,
Metabolic Programming Jonathan D. Curtis, Chih-Hao Chang, David E. Sanin, Jing
Qiu, Oliver Kretz, Daniel Braas, Gerritje J.W. van der Windt,
Qiongyu Chen, Stanley Ching-Cheng Huang, Christina M.
O’Neill, Brian T. Edelson, Edward J. Pearce, Hiromi Sesaki,
Tobias B. Huber, Angelika S. Rambold, and Erika L. Pearce
Genomic and Transcriptomic Features of Response to Willy Hugo, Jesse M. Zaretsky, Lu Sun, Chunying Song,
Anti-PD-1 Therapy in Metastatic Melanoma Blanca Homet Moreno, Siwen Hu-Lieskovan, Beata Berent-
Maoz, Jia Pang, Bartosz Chmielowski, Grace Cherry,
Elizabeth Seja, Shirley Lomeli, Xiangju Kong, Mark C. Kelley,
Jeffrey A. Sosman, Douglas B. Johnson, Antoni Ribas, and
Roger S. Lo
(continued)
Progressive Loss of Function in a Limb Enhancer Evgeny Z. Kvon, Olga K. Kamneva, Uirá S. Melo, Iros Barozzi,
during Snake Evolution Marco Osterwalder, Brandon J. Mannion, Virginie Tissières,
Catherine S. Pickle, Ingrid Plajzer-Frick, Elizabeth A. Lee,
Momoe Kato, Tyler H. Garvin, Jennifer A. Akiyama, Veena
Afzal, Javier Lopez-Rios, Edward M. Rubin, Diane E. Dickel,
Len A. Pennacchio, and Axel Visel
Domestication and Divergence of Saccharomyces Brigida Gallone, Jan Steensels, Troels Prahl, Leah Soriaga,
cerevisiae Beer Yeasts Veerle Saels, Beatriz Herrera-Malaver, Adriaan Merlevede,
Miguel Roncoroni, Karin Voordeckers, Loren Miraglia, Clotilde
Teiling, Brian Steffy, Maryann Taylor, Ariel Schwartz, Toby
Richardson, Christopher White, Guy Baele, Steven Maere,
and Kevin J. Verstrepen
Zika Virus Infection during Pregnancy in Mice Causes Jonathan J. Miner, Bin Cao, Jennifer Govero, Amber M. Smith,
Placental Damage and Fetal Demise Estefania Fernandez, Omar H. Cabrera, Charise Garber,
Michelle Noll, Robyn S. Klein, Kevin K. Noguchi, Indira U.
Mysorekar, and Michael S. Diamond
Microbial Reconstitution Reverses Maternal Shelly A. Buffington, Gonzalo Viana Di Prisco, Thomas A.
Diet-Induced Social and Synaptic Deficits in Offspring Auchtung, Nadim J. Ajami, Joseph F. Petrosino, and Mauro
Costa-Mattioli
Breaking Cryo-EM Resolution Barriers to Alan Merk, Alberto Bartesaghi, Soojay Banerjee, Veronica
Facilitate Drug Discovery Falconieri, Prashant Rao, Mindy I. Davis, Rajan Pragani,
Matthew B. Boxer, Lesley A. Earl, Jacqueline L.S. Milne, and
Sriram Subramaniam
Programmable RNA Tracking in Live Cells David A. Nelles, Mark Y. Fang, Mitchell R. O’Connell, Jia L.
with CRISPR/Cas9 Xu, Sebastian J. Markmiller, Jennifer A. Doudna, and Gene W.
Yeo
High-Throughput, High-Resolution Mapping of Takayasu Mikuni, Jun Nishiyama, Ye Sun, Naomi Kamasawa,
Protein Localization in Mammalian Brain by and Ryohei Yasuda
In Vivo Genome Editing
Ultra-High Resolution 3D Imaging of Whole Cells Fang Huang, George Sirinakis, Edward S. Allgeyer, Lena
K. Schroeder, Whitney C. Duim, Emil B. Kromann, Thomy
Phan, Felix E. Rivera-Molina, Jordan R. Myers, Irnov Irnov,
Mark Lessard, Yongdeng Zhang, Mary Ann Handel, Christine
Jacobs-Wagner, C. Patrick Lusk, James E. Rothman, Derek
Toomre, Martin J. Booth, and Joerg Bewersdorf
THE BENCHMARK
IN ANTIBODIES
Since the day it was founded, Proteintech has been making all
of its products to the highest standards possible while taking
complete responsibility for the quality of each product.
• Proteintech makes every single antibody in its
12,000-strong catalog.
• Each Proteintech product is unique and cannot be bought
under a different label.
• Antibodies tested in siRNA knockdown experiment to
demonstrate specificity.
• It works in any species and any application or get a full
money-back refund.
www.ptglab.com
2017
SCIENTIFIC
CONFERENCES
Presenting the most significant research on cancer etiology, prevention, diagnosis, and treatment
Precision Medicine Series: EACR-AACR-SIC Special Conference 2017:

Opportunities and Challenges of The Challenges of Optimizing Immuno-and Targeted
Exploiting Synthetic Lethality in Cancer Therapies: From Cancer Biology to the Clinic
Cochairpersons: René Bernards, Cochairpersons: Anton J.M. Berns,
William C. Hahn, and Louis M. Staudt Nancy E. Davidson, and Silvia Giordano
January 4-7, 2017 • San Diego, CA June 24-27, 2017 • Florence, Italy
AACR International Conference Third CRI-CIMT-EATI-AACR International

on New Frontiers in Cancer Research Cancer Immunotherapy Conference
Cochairpersons: Peter A. Jones and Frank McCormick September 6-9, 2017 • Mainz/Frankfurt, Germany
January 18-22, 2017 • Cape Town, South Africa
Advances in Modeling Cancer in Mice:
AACR Annual Meeting 2017 Technology, Biology, and Beyond
Program Committee Chairperson: Kornelia Polyak Cochairpersons: Cory Abate-Shen, Kevin M. Haigis,
April 1-5, 2017 • Washington, DC Katerina A. Politi, and Julien Sage
September 24-27, 2017 • Orlando, FL
Head and Neck Cancer Conference: Optimizing
Survival and Quality of Life Through Basic, Clinical, Tenth AACR Conference on The Science of
and Translational Research Cancer Health Disparities in Racial/Ethnic
Chairperson: Jeffrey N. Myers Minorities and the Medically Underserved
Cochairpersons: Jennifer Rubin Grandis, Cochairpersons to be announced
J. Silvio Gutkind, and Quynh-Thu Le September 25-28, 2017 • Atlanta, GA
April 23-25, 2017 • San Diego, CA
AACR-NCI-EORTC Molecular
AACR International Conference Targets and Cancer Therapeutics
on Translational Cancer Medicine Cochairpersons to be announced
Cochairpersons: Carlos L. Arteaga October 26-30, 2017 • Philadelphia, PA
and Carlos Gil M. Ferreira
May 4-6, 2017 • São Paulo, Brazil Prostate Cancer
Conference Cochairpersons:
Hematologic Malignancies: Johann S. de Bono, Karen E. Knudsen,
Translating Discoveries to Novel Therapies Peter S. Nelson, and Mark A. Rubin
Conference Chairperson: Jonathan D. Licht December 2-5, 2017 • Orlando, FL
Conference Cochairpersons: Lucy A. Godley,
Louis M. Staudt, and Catherine J. Wu San Antonio Breast Cancer Symposium
May 6-9, 2017 • Boston, MA Presented by CTRC-AACR-BCM
Codirectors: Carlos L. Arteaga,
International Conference Virginia G. Kaklamani, and C. Kent Osborne
on Malignant Lymphoma (ICML) December 5-9, 2017 • San Antonio, TX
ICML President: Franco Cavalli
Chairperson, Local Organizing Committee:
Michele Ghielmini
June 14-17, 2017 • Lugano, Switzerland
Learn more and register at

AACR.org/Calendar
1608100_UpcomingConf_Dec16_Cell_8.375x10.875.indd 1 10/25/16 8:43 AM

SnapShot: Transcription Elongation
Christine E. Cucinotta and Karen M. Arndt
Department of Biological Sciences, University of Pittsburgh, Pittsburgh, PA 15260, USA
Pol II pause, release, elongation rate, and processivity factors
Paused Pol II Super Elongation Complex (SEC) Protein Function

P-TEF-b/Cdk9/CycT1/ Cyclin-dependent kinase, releases Pol II from
AFF1/4 AF9/ (S.c.:Bur1/Bur2) pausing and phosphorylates Pol II, NELF, and DSIF
NELF DSIF P-TEFb ENL
DSIF (S.c.:Spt4-Spt5) Stabilizes Pol II pausing, recruits elongation
Y P-TEFb EAF
1 ST P S factors, stimulates elongation
4 S5 7 Pol II
2
ELL NELF Negative elongation factor
PARP-1 Poly[adenosine diphosphate (ADP-ribose)]
Little Elongation Complex (LEC) polymerase, transfers ADP-ribose to NELF
to inhibit NELF
EAF BRD4 Recruits and activates P-TEFb
Released Pol II ICE1 TFIIS Resumes Pol II elongation from backtracked-arrest
ELL
TFIIF Prevents transient pausing of Pol II
Recruited by: ICE2
PARP-1 P-TEFb Elongin Increases transcription elongation rate
SEC
BRD4 Polymerase-associated Super elongation complex Recruits P-TEFb, increases Pol II elongation rate
NELF
factor 1 complex Little elongation complex Required for snRNA expression in metazoans
ADP-R P Paf1C Associated with Pol II and Spt4-Spt5, recruits
DSIF Cdc73 Ski8 chromatin remodelers, histone chaperones,
Y P P and modifiers
1 S T P Elongation
2 S5 S7 Pol II Rtf1 Ctr9
4 GDOWN1 Tightly associated with Pol II, stabilizes pausing
RNA processing factors CCR4-NOT, THO/TREX, Xrn2: mRNA processing
Paf1 Leo1
and export factors that also regulate elongation
Pol II CTD phosphorylation Histone modifications involved in elongation

P P H2B K123 H3 S10
YS P YS P Y S P monoubiquitylation cascade phosphorylation cascade
1 T S
2 4 S5 7
1 T S S7 1
2 T S S
Pol II
Pol II 2
4 5 Pol IIII
Pol 4 5 7 PolIIII
Pol
Cdc73 Ski8 PIM1
Ac P
Kinase Phosphatase Rtf1 Ctr9 K9 S10
Ser2-P Cdk9 (S.c: Bur1) Fcp1
Cdk12 (S.c: Ctk1) H3
Paf1 Leo1 A H4
H2
Ser5-P Cdk7 (S.c: Kin28) Rtr1, Ssu72 H2
B
Ser7-P Cdk7 (S.c: Kin28) Ssu72

Cdk9 (S.c: Bur1)
Thr4-P Plk3, Cdk9 ?
Tyr1-P ? Rtr1 (S.c.)
H3
Rad6 H2
A H4
14-3-3 MOF
Histone chaperones and chromatin remodelers H2
B
Bre1 Ub Ac P Ac
Nucleosome disassembly Nucleosome spacing K9 S10 K16
and reassembly
Ub
H3
Ubp8 H2
A H4
Pol II H 2 B
COMPASS Dot1
K Me Me
Pol II 4 K BRD4
79
H3 H3
Protein Function Role in elongation A H4 A 4
H2 H2 B H
B
2 H2
FACT Histone chaperone complex Displaces H2A-H2B dimer in the wake of transcribing Pol II, UbH Ub
of H.s.: hSPT16, Ssrp1, S.c.: evicts nucleosome; reassembles nucleosomes, regulates
Spt16, Nhp6, Pob3 H2BK123ub, prevents cryptic transcription P-TEFb
Spt6 Histone chaperone Prevents cryptic transcription, required for proper histone
occupancy during elongation
Asf1 Histone chaperone Controls H3 exchange during transcription H3 K36 methylation
Nap1 Histone chaperone Binds H2A-H2B and forms hexasome structures through RSC, Me HAT
promotes nucleosome assembly SET2
Ac
K4
Rtt106 Histone chaperone Binds H3-H4 and promotes transcription-coupled H3 K14 Me

deposition, prevents cryptic transcription Ac K
36
Chd1 Chromatin remodeler Controls nucleosome spacing and histone exchange, HDAC H3
promotes Pol II promoter escape (mammals) A H4
H2 H3
2 B
ISW1 Chromatin remodeler Controls nucleosome spacing and histone exchange Ub
H
H2
A H4
2 B
H
RSC Chromatin remodeler Helps Pol II passage through nucleosomes and maintains
proper nucleosome occupancy
See online version for

Cell 166, August 11, 2016 ©2016 Elsevier Inc. DOI http://dx.doi.org/10.1016/j.cell.2016.07.039 legend and references.
CELL-FINAL-MESSAGE-3.pdf 1 6/29/2016 10:05:01 AM
QUALIFIED ANTIBODIES | MADE IN THE USA | VALIDATED IN THE USA
Oh My Gosh!—Great Reactions Start with Really Good Antibodies

Really Good Antibodies That’s because every antibody is manufactured and validated on-site using a
unique process that verifies target specificity with paired antibodies raised
against distinct protein epitopes. We’ve heard our customers’ great reactions
for over 40 years and now we’d like to hear what you’ll say.
Request a free trial size antibody by visiting bethyl.com/1613

Terms & Conditions Apply. Please see webpage for complete details. ©2016 Bethyl Laboratories, Inc. All rights reserved.
SnapShot: Expanding the
Ubiquitin Code
Krishnaraj Rajalingam1 and Ivan Dikic2
1
Molecular Signaling Unit, FZI, Institute of Immunology, University Medical Center Mainz, JGU, Langenbeck Str 1, Mainz, Germany
2
Institute of Biochemistry II, Goethe University, Theodor-Stern-Kai 7, 60590 Frankfurt am Main, Germany; Buchmann Institute for
Molecular Life Sciences, Max-von-Laue-Straße 15, 60438 Frankfurt am Main, Germany
1. The UB code 2. Expansion of the UB code
NH2 Multi-monoubiquitination
Ub
Substrate “Writers” Substrate Ub
E3 ligases Poly-ubiquitination
(enzymatic HECT-type; Ub
E1 / E2 non-enzymatic Ub
Ub enzymes U-Box or Ub
Ub Ub Ub Ub
Ub RING-type) Ub Ub
Ub Ub Ub Ub Ub Ub Ub
Ub Ub
NH Ub Ub Ub Ub Ub Ub Ub Ub Ub
Substrate Substrate Substrate Substrate Substrate Substrate Substrate Substrate Substrate
O
K48 K6-linkage K11-linkage K27-linkage K29-linkage K33-linkage K48-linkage K63-linkage M1-linkage
DUBs
Post-translationally Ub Homotypic chains
“Erasers” M1 Ub
modified ubiquitin P Ub Ub Ub (8 distinct types)
K27 K29 Ub P
K63 Ub
A Ub Ub
K6 Ubiquitin Ub Ub Ub Ub
... Heterotypic chains
molecule Substrate Substrate Substrate Substrate
C-terminal Gly K33 (numerous variations)
attached to substrate lysine K11 Acetylated S65-phosphorylated Branched Mixed
3. Reading and interpretation of the UB code
Ub-binding domains (UBDs) I44 patch UBD Specificity Ub surface Examples

L8, I44, H68, V70 CUE mUb Ile44 Cue2, Vps9
recognize different surface
DUIM mUb Ile44 Hrs
patches on Ub, linker regions, C-terminal GAT mUb Ile44 GGA3, TOM1
F4 patch
or the (free) C-terminal tail tail Q2, F4, T14 GLUE mUb Ile44 EAP45
I36 patch TEK-box MIU mUb Ile44 Rabex-5
I36, L71, L73 K6, K11,T12, T14, E34 NZF mUb, K63, Ile44 NpI4, VPS36, TAB2, TAB3,
K29, K33 HOIP, HOIL-1L, TRABID
Endocytosis DNA repair PFU mUb, polyUb Ile44 Doa1, PLAA
MultimonoUb, K63 K6, K27, K33, K63 Ub receptor
Substrates: Endocytic cargo Substrate: H1-type histone UBD PRU mUb, K48 Ile44 Rpn13
Ub receptors: Eps15, Epsin, Ub-receptor: Rnf186, Rap80, SH3 mUb Ile44 Sla1, CIN85, amphiphysin
Sts-1/2, Hrs, Stam2 BRCA1, 53BP1 UBA mUb, K48 Ile44 PLIC1/2, HHR23A/B,
Ddi1, p62, NBR1, Cbl-b,
Xenophagy USP5, HERC2
K63 UBAN M1-diUb Ile44, Phe4, linker NEMO, optineurin,
Substrates: Proteins Ub ABIN1-3
surrounding pathogen
Ub receptors: autophagy Ub UBC mUb Ile44 UbcH5C
receptors p62, NBR1, optineurin Ub UBM mUb Leu8 Polymerases iota and
Rev1
Substrate
Mitophagy Link to pathway- Cell-cycle regulation UBZ mUb, polyUb Ile44 Polymerases eta and
K6, K11, K63, pS65-K63 specific effectors K11/48 branched kappa, FAN1, NDP52,
pS65-monoUb Substrate: Nek2A TAX1BP1, WRNIP1
Substrates: MOM proteins Receptors: UEV mUb Ile44 VPS23, TSG101
Ub receptors: autophagy Proteasome UIM mUb, K48, Ile44 S5a, Vps27, USP28,
adaptors p62, NBR1, optineurin K63, K33 ataxin-3, EPS15, STAM,
RAP80
Post-golgi trafficking NF-kB signaling Proteasomal degradation ERAD VHS mUb Ile44 STAM, Hrs
K33 K63 and M1 K48, K11, K29, (K63) K11 and K48
Substrate: Substrates: Substrates: Proteins Substrates: WD40 repeat mUb Ile44 Doa1/UFD3
Cm7 RIP, Nemo destined for degradation Misfolded proteins propeller
Ub receptor: Ub receptors: Receptors: Rpn10, Rpn13, Ub receptor: ZnF A20 mUb, K63 Asp58 A20, Rabex-5
Eps15 Nemo, Tab2/3 Dss1, Rad23, Dsk2, and Ddi1 Proteasome ZnF UBP Free polyUb Leu8, Ile36, tail USP5, HDAC6, BRAP2
4. Pharmacological exploitation of the UB system

A) Inhibiting E1, E2, E3 or DUBs C) Directing E3 ligases D) Inhibiting Ub code recognition
to desired substrates Ub Ub
Substrate Signal Signal
recognition E3 ligase
Active
centers CRL4 CRL4 CRL4
Ub propagation Ub X propagation
(CRBN) (CRBN) (CRBN) Ub Ub
E3 ligase E3 ligase E3 ligase Substrate Substrate
Substrate Substrate
Ub Native Bifunctional
thalidomide Targeting agent Cellular target Desired outcome Indication Stage
substrate
DUB Ub Thalidomide JNJ-26854165 MDM2 E3 activity Stabilization of p53, Refractory solid Phase I
tumor suppression tumors
Ub
Nutlins; R05503781; MDM2-p53 inter- Stabilization of p53, AML, Phase I
CRL4 Harmful R05045337 action tumor suppression retinoblastoma,
New substrate neuroblastoma
(CRBN) protein
B) Inhibiting the proteasome E3 ligase GDC-0152; LCL-161; IAP proteins IAP auto-ubiquitination Breast, ovar- Phase I/II
CRL4 TL32711; HGS1029; and degradation, cell ian, lung cancer,
Ub Native CRL4 (CRBN) AT-406 death melanomas, ALL,
Ub CLL
Ub Ub substrate (CRBN) E3 ligase
E3 ligase AEG-35156 XIAP Cell death AML Phase II
Ub Ub P5091; P22077; HBX USP7 Stabilization of p53, CLL, multiple Pre-clinical
Ub 19, 818; HBX 41, 108 tumor suppression myeloma
Substrate Ub
IU1, b-AP15 USP14, UCHL5 Increased proteasomal Solid tumors Pre-clinical
Ub Ub Ub Ub activity
Ub Ub
Bortezomib Ub Ub Bortezomib Protein degradation Tumor cell death Multiple myeloma Approved
Ub Ub
(Velcade ®) Ub Ub Carfilzomib/PR171 Hematol. malig- Phase III
nancies
CEP-18770 Multiple myeloma Phase I
NPI-0052 Multiple myeloma Phase I
Proteasomal elimination ONYX 0912 Multiple myeloma Phase I
Cell 164, February 25, 2016 ©2016 Elsevier Inc. DOI http://dx.doi.org/10.1016/j.cell.2016.02.019 See online version for legends and references
MP Premium Dextran Sulfate Sodium Salt (DSS)
At least 3,000 Scientists Prove

MP DSS is the Most Effective
in the Industry!
Let us it to you!
High Purity, Reproducible Quality! Best Results!

• Validated by 3,000+ scientific publications
• Highest purity (99%) and stability
• Offers Highest Sulfur content (18-20%)
< 0.2% free sulfate
• Highest efficiency
• Best value in the industry.
The mice were fed normal chow and 2.0% wt/wt DSS in water for 1 week and
then sacrificed. The weight of individual mice was monitored daily. The data
represent means=SD (data obtained from Bamba et.al., 2012)
FREE SAMPLE
($180 value) Just Ask!
Get a 10g sample of our MP Gold Standard DSS for colitis research absolutely FREE.
Use promo code: DSSCP when placing your order.
www.mpbio.com/DSS or call 800.854.0530
SnapShot: Epigenomic Assays
Martin Krzywinski1 and Martin Hirst1,2
Canada’s Michael Smith Genome Sciences Centre, BC Cancer Agency Research Centre,
1
BC Cancer Agency, Vancouver, BC V5Z 1L3, Canada; 2Department of Microbiology and Immunology,
Michael Smith Laboratories Centre for High-Throughput Biology, University of British Columbia, Vancouver, BC V6T 1Z4, Canada
3 4
1
2
10 bp
0.1–1 kb 10–100 kb
150 bp
1 DNA methylation analysis 2 Chromatin immunoprecipitation sequencing

Cytosine modification is a feature Histones are decorated with
of the mammalian genome. chemical modifications.
Sonicate Shear DNA

Histone Sonicate Digest Shear DNA
CpG mCpG hmCpG
Modified histone
5-methyl-cytosine (5mC) and its
oxidative derivatives (e.g., 5hmC) Add linkers DNA wound
are measured genome-wide around histone
using enrichment- and Enrich
conversion-based methodologies
Genomic positions of modified histones are Antibody
followed by massively parallel
sequencing. Bisulfite conversion measured genome-wide by chromatin
provides quantitative measure- Bilsulfite Enrich immunoprecipitation followed by massively
ments of 5mC but is unable to treatment parallel sequencing (ChIP-seq). Histones can
distinguish 5hmC. Antibody Antibody be liberated from the genome by sonication,
enrichment provides qualitative enzymatic digestion, or by transposon Remove
measurement of 5mC and 5hmC. insertion (not shown). If sonication is used, histone and
Bisulfite-converted or -enriched UCU CU CU chromatin must first be chemically cross-linked add linkers
DNA is purified, subjected to (see Step 4). Following histone liberation,
library construction and clonally Sequence specific chemical modifications are enriched by
sequenced. Specialized and analyze immuno-absorption. DNA associated with Sequence
algorithms are required to align enriched histones is purified, subjected to and analyze
bisulfite-converted reads to a library construction, clonally sequenced, and
reference genome. aligned to a reference genome.
3 Open chromatin analysis 4 3D chromatin capture

Nucleosome-depleted regions are Chromatin loop contacts reveal
enriched in regulatory elements. distal regulatory elements and
structural domains.
Genomic positions of open chromatin are measured genome-wide by
massively parallel sequencing of a collection of DNA fragments liberated Genomic positions of long-range
Digest
from intact chromatin by either transposon insertion, enzymatic digestion, chromatin contacts are measured
or sonication. The resulting DNA fragments are subjected to size selection genome-wide by massively parallel
or phenol-chloroform extraction to deplete nucleosome-associated DNA. sequencing a collection of DNA
The resulting DNA is purified, subjected to library construction, clonally fragments generated by proximity
sequenced, and aligned to a reference genome. ligation. Intact chromatin is
cross-linked to physically link
genomically distal nucleosomes that
Biotin Ligate ends
are adjacent in 3-dimensional space.
Transposon at Digest specific

open regions to open regions
Sonicate Enrich with

streptavidin
Cross-linking Digest pull-down
Size select
Cross-linked chromatin is enzymatically
digested and the resulting DNA end
labeled with biotin and subjected to Add linkers
Add linkers proximity ligation. Ligated DNA is sheared
by sonication or enzymatic digestion and
the ligated junctions enriched by Sequence
Sequence streptavidin pull-down. The resulting DNA and analyze
and analyze is purified, subjected library construction,
clonally sequenced, and aligned to a
reference genome.

Cell 167, November 17, 2016 ©2016 Elsevier Inc. DOI http://dx.doi.org/10.1016/j.cell.2016.11.015 legends and references
ALL-IN-ONE FLUORESCENCE MICROSCOPE
BZ-X700
ALL-IN-ONE
FLUORESCENCE
CONVENTIONAL SETUP MICROSCOPE
KEYENCE optimizes the imaging process with a fully automated, easy-to-use system that
captures high- resolution images and allows researchers to achieve their desired results faster.
REAL-TIME OVERLAY PHASE CONTRAST BRIGHTFIELD
TIME-LAPSE OPTICAL SECTIONING CELL COUNT
For more information, visit: www.keyence.com/bz-x700

SnapShot: Signaling in Symbiosis
Sharon R. Long
Department of Biology, Stanford U niversity, Stanford, CA 94 305 , U SA
Initial interaction Infection thread formation Symbiosome formation

Rhizobium
Root hair
LE G U M E RO O T
Cell 167, October 6, 2016 ©2016 Elsevier Inc.

Lu teol in OH Plant leghemoglobin
OH RHIZOBIUM
O (Lb) lowers free O2
OH
Ex oR Ex oS
OH O
N odD
protein
Kinase
L L*
GENOME Inactive Active
PO4 H is-kinase Fix L
nodD1 nodABC Transcription senses low O2
N odu l ation ( nod) g enes Chvl factor
Enzymes that
synthesize Nod factor Fix J response regulator J J * Active NCR
+ - activates N -related genes peptides
BA C T E RIA
EPS gene Motility
OSO3H
CH OH CH 2
expression
CH OH 2
CH OH 2 O
2
O O
O O O Transcription of
O OH
H O
N H H O N H NifA N-related genes
H O N H H O
H O N H CO CO
CO
CO CH 3 CH 3
CH 3
CH
H C
(CH 2 5
)
CH
N od F actor CH N od F actor E P S
(CH 2 5
) RpoN N2ase genes
CH 3
DOI http://dx.doi.org/10.1016/j.cell.2016.09.046
Flotillins and remorins promote PLANT CELL WALL
interaction between N F receptors
PLANT PM
N F receptors required and cytoskeleton
for early signaling Plant membrane surrounds bacterium;
together, these are the “ symbiosome”
E P S
E S R3 Membrane
Unknown differentiation
secondary
C E LL
signals Flotillins
NFR5
NFR1 Pump Nuclear
SymRK pore Golgi
Calcium NFR1 I nfection CYTOSKELETON NCRs
spiking
P LA N T
Symbiosome-
Transcription CaM specific vesicle
factors trafficking
CCaMK Remorin
CYCLOPS
Leghemoglobin,
Symbiosis-associated N CRs, other
gene ex pression O2 O2 O2 nodule-specific
Lb leghemoglobin transcrpts
CYTOPLASM NUCLEUS CYTOPLASM NUCLEUS
legend and references.

Make scientific
breakthroughs faster with
SciFinder®, the most
complete source of chemistry
and related information.
CAS scientists thoroughly analyze the literature indexed in SciFinder so

you can quickly search the authoritative CAS databases for substances,
reactions, and patent and journal references, giving you confidence you
aren’t missing important results.
l Save time with powerful, intuitive search features that allow you
to quickly find the most relevant answers to your research
questions
l Build your plan, design a synthesis and easily manage results
with insightful organization and planning tools
l Stay current in your field by easily creating alerts for timely
updates on the latest research
l Affordable pricing programs are available for organizations of
all types and sizes
SciFinder is designed by and for scientists like you to enhance your
discovery efforts every step of the way.
Learn more about SciFinder at www.cas.org/SciFinder.
sf_scifinder flyer_0316_8.5x10.5 0.125in bleed.indd 1 9/22/16 11:12 AM

SnapShot: T he Bacterial Cytoskeleton
Gero Fink, Andrzej Szewczak-H arris, and Jan Lö we
MRC Laboratory of Molecular Biology, Cambridge CB2 0Q H , U K
Cell 166, July 14, 2016 ©2016 Elsevier Inc.

Your neurons are
talking.
Are you listening?
MaestroTM
microelectrode array (MEA)
in vitro neural activity analysis
Record network-level neural activity directly on the
benchtop with Maestro. Over minutes or months,
gain unprecedented access to electrical network
function from cultured neural populations.
Straightforward and easy to use, any researcher

can measure activity from neurons in 12-, 48-, or
96-well plates.
Connect to your network.

Maestro MEA platform with AxISTM software showing real-time
neural activity across a 96-well plate.
Learn more: axionbio.com/maestro

SnapShot: Neuronal Regulation of Aging
H eather J. W eir and W illiam B. Mair
Department of Genetics and Complex Diseases, H arvard T.H Chan School of Public H ealth, 665 H untington Avenue, Boston, MA 02115 , U SA
C. elegans
N eu ronal m ech anism s D ow nstream m ech anism s ( p erip h eral tissu es)
S ensory p ercep tion ( ex trinsic) C el l u l ar stress ( intrinsic) Respiration

Nutrient availability Proteostasis UPR ER (XBP-1s) Xenobiotic detoxification FMO-2
Dietary restriction SKN-1 Chronic stress HSF-1
Nutrient type NMUR-1 Mitochondrial stress Stress resistance
Cell 166, July 28, 2016 ©2016 Elsevier Inc.

Low O 2 HIF-1 Pro-longevity gene transcription
Olfactory and gustatory UPR ER, UPR mt
inputs
O ctop am ine
ILP s Long ev ity
D A F -1 6
M etabol ic h om eostasis activ ity
N eu roendocrine
Nucleus U P R Mitochondrial metabolism
Reduced I I S (rI I S):
D A F -1 6 , H S F -1 Mitochondrial dynamics
Age-1/daf-2 mutations
AMPK
Cytoplasm Serotonin synthesis Reduced insulin-like signaling
CRTC-1 Stress resistance
Drosophila M ou se
S ensory p ercep tion S ensory p ercep tion Glucose homeostasis
Fat deposition Pain Respiration
Olfactory receptors: Nucleus
Stress resistance
Or 83b Gr 63a CRTC1 Cytoplasm
X X CGRP
D ILP s Autophagy Pancreas
Gustatory inputs TR PV1
CRTC1
X
I nsulin signaling
M etabol ic Y outhful mitochondrial

h om eostasis C el l u l ar Exogenous
function and morphology
stress Hsp70 Neurogenesis,
AMPK autophagy
cognitive improvement
Altered fat storage
N eu roendocrine Tissue homeostasis
N eu roendocrine Altered growth
rllS: Irs2/IGF-1R
rllS: dFOXO, I mmune inhibition:
reduced DILP production
IKKβ/ NF-κB
X SIRT1 ox2r
Long ev ity M etabol ic Long ev ity
GnRH h om eostasis Neural activation
legends and references

Knock Out
Any Gene!
CRISPR/CAS 9
Genome Editing Kits
OriGene’s system for genome disruption and gene
replacement delivers pre-designed plasmids and all the
vectors needed to knock out any human or mouse gene.
Knockout is as simple as 1-2-3:
1 Search the gene symbol on origene.com and order
2 Follow the simple protocol for transfection and Puro selection

Scan to view
3
CRISPR/Cas-9 Video
Validate the knockout
Come to OriGene,
the trusted molecular biology
Kit Components Include: expert, for your CRISPR needs.
• 2 guide RNA vectors to ensure efficient cleavage
• Donor vector with predesigned homologous arms
• Knockin GFP-Puro for selection
• Scramble gRNA as negative control included
www.origene.com/CRISPR-CAS9
SnapShot: Microglia in Disease
Simon Beggs1,2 and Michael W . Salter1
1
Program in N eurosciences & Mental H ealth, H ospital for Sick Children, Toronto ON M5 G 1X 8 , Canada
2
Developmental N eurosciences Programme, U CL I nstitute of Child H ealth, London W C1N 1EH , U K
C3
S y nap tic p ru ning
CR3
D E N D RIT E
Synaptic
terminal
N E U RO N
Pre-synaptic
Microglia
Cell 165, May 19, 2016 ©2016 Elsevier Inc.

M icrog l ia
S y nap tic density across l if etim e
A S D
M icrog l ia
Synaptic density
S C H IZ O A D
Birth Adolescence Adult Aged
A U T IS M S P E C T RU M D IS O RD E RS S C H IZ O P H RE N IA A LZ H E IM E R’ S D IS E A S E P A IN
H eal th y H ig h risk S ch iz op h renia

Amyloid
plaque C S F 1 P ain sig nal ing
to brain
C S F 1 R S p inal
P 2 X 4
inh ibition
BD N F
P 2 X 4
D ecreased C 3 ex p ression
Retention of im m atu re sy nap ses
Increased C 3 w ith ag e
A bnorm al connectiv ity Increased m icrog l ial activ ity and p ru ning M icrog l ia draw n to p l aq u e dep osits P 2 X 4 R+ m icrog l ia in sp inal dorsal h orn

Reliable Quantification
High-Sensitivity Screening Assays
Rapidly and Reliably Quantify Second Messengers with High-Sensitivity Assays

As our understanding of neurodegenerative disease origins increases, so does the need for innovative, high-quality screening assays for identifying
potential therapeutics. Enzo’s expertise in assay development and labeling and detection provide a unique collection of innovative research tools to
monitor neural signaling networks.
cAMP/cGMP ELISA Kits FLUOFORTE® Calcium Assays Cyclic Nucleotide Phosphodiesterase
Most sensitive and reliable ELISA kits for Brighter, more robust fluorescent calcium Assay
detection of cAMP/cGMP mobilization assays Unique colorimetric assay used to screen
inhibitors and modulators of cyclic
nucleotide PDE activity
L ot- to- L ot Consistenc y O b tain R ob u st Signal I ntensity H igh Sensitiv ity
O D 6 2 0 nm
5′-AMP (nmol)
Life Sciences
scientists enabling scientists.TM
www.enzolifesciences.com/neuroscience © 2016 Enzo Life Sciences
CELL_Neruo_FP.indd 1 9/26/16 2:59 PM

SnapShot: T ypes of U biq uitin Ligases
Francesca Ester Morreale and H elen W alden
MRC Protein Phosphorylation and U biquitylation U nit at U niversity of Dundee, Scotland DD1 5E H , U K
U BIQ U IT IN A T IO N
O O
HN
SH Ubiquitin ligase
HN
O (E3) U b
O O O U b
HN
HO U b E2 Protein degradation by the
SH S U b S U b E3 U b S u bstrate
proteasome, cell-cycle
E1 E1 E2 S u bstrate O progression, transcriptional
U b
HN
ATP U b regulation, DNA repair,
Ubiquitin activating AMP SH SH U b
Ubiquitin conjugating signal transduction
enzyme (E1) enzyme (E2)
E1 E2 S u bstrate
RIN G E 3 s ~ 6 0 0 in h u m ans
T y pe s of RING E 3s
Ubiqu itin transfe r M onom eric RIN G C u l l in-RIN G l ig ases

E 2 S u bstrate S u bstrate
Example: c-CBL
O O RIN G E 3 E 2 S R
HN
S U b U b RING G -box
RIN -box p prrotein
otein E 3 A dap tor
H om odim eric RIN G Cu lin
E2 S u bstrate E 2 S u bstrate
RIN G E3 Examples: cIAP, RNF4, RIN G E 3

BIRC7, IDOL, TRIM5α RIN G E 3 Composed of:
S u bstrate recep tor ( S R) - example: F-box protein Skp2
A dap tor p rotein - example: Skp1
H eterodim eric RIN G C u l l in - example: Cul1
Ex am ple s of k now n regu la tion m ech anism s Ring -box p rotein ( Rbx ) - example: Rbx1
Examples: E 2 S u bstrate
Neddylation (e.g., CRLs) BRCA1–BARD1,
Phosphorylation (e.g., c-CBL, APC, Mdm2) RIN G E 3 O th er m u l tisu bu nit E 3 s
Mdm2–MdmX,
Small molecule: E3 interaction (e.g., TRAF-2:sphingosine-1-phosphate) RING1B–Bmi1 RIN G E 3
Apc7A
The anaphase
promoting complex/ Apc7B
M onom eric U -box
cyclosome (APC/C)
6
Ch aracteristic dom ains E 2 S u bstrate
Apc1
Apc3A Apc3B S u bstrate
Example: E4B
U -box E 3 Apc10
RING (Really Interesting New Gene): E 2
N
Zn binding domain, recruits the Apc6A Cdh1
Apc1 Apc11 RIN G
Ub-charged E2 H om odim eric U -box A
Apc13
Zn2+ Zn2+
S u bstrate c12
E 2 Ap Apc8B
Apc2
U-box : same fold as RING without Examples: CHIP, Prp19 U -box E 3 Apc15
Zn coordination Apc5
C U -box E 3 Apc8A Apc4
H E C T E 3 s ~ 3 0 in h u m ans
HE C T F am ilie s
Ubiqu itin transfe r Ch aracteristic dom ain
NED4 fa m ily
O 1 O C2 W W W W W W W W HE C T
HN
HECT (homologous to E6-AP

2
S U b O
U b carboxyl terminus) Examples: Smurf1 (2 WW motifs),
Smurf2 (3 WW motifs), Itch (4 WW motifs)
E2 S U b S u bstrate Contains a
SH HER C fa m ily
catalytic cysteine
HECT E3 C lobe Small HERCs RLD HE C T
N lobe
Large HERCs RLD SPRY W D40 RLD HE C T
Ex am ple s of k nown regu la tion m ech anism s Flexible tether allows
changes in relative
Recruits the
Autoinhibition through intramolecular interactions released by: orientations of the
Ub-charged E2 Ot h er HECT s
phosphorylation (e.g., Itch), neddylation (e.g., Smurf1), protein two lobes
interaction (e.g., Smurf1, Smurf2) HE C T
Example: E6AP
RBR E 3 s ~ 1 2 in h u m ans
E x am ple s of RBR E 3 s
Ubiqu itin transfe r Ch aracteristic dom ain
O 1 O P ark in U bl RIN G0 RBR
HN
2 IBR (InBetweenRING)
S U b U b
O or BRcat
P arc DOC Cullin RBR
E2 S U b S u bstrate SH 1000AA
R B R E3 R B R RNF 14
( A/ B) RBR
NZF
RING1, recruits RING2 or Rcat,
Ex am ple s of k nown regu la tion m ech anism s HOIP PU B B-box U BA RBR LDD
the Ub-charged E2 contains a
catalytic cysteine
Autoinhibition through intramolecular interactions released by:
phosphorylation/phosphoubiquitin interaction (Parkin), binding HAR I U BA RBR Ariadne
to neddylated CRLs (e.g., HHARI), protein interactions (e.g., HOIP)

Cell 165, March 24, 2016 ©2016 Elsevier Inc. DOI http://dx.doi.org/10.1016/j.cell.2016.03.003 legend and references.
Quantify with Confidence
Ever wonder about the quantitative nature of image data?
We do. Introducing the new ORCA-Flash4.0 V3 and the
W-VIEW GEMINI-2C dual channel, dual camera optics.
ENGINEERED FOR SUPER RESOLUTION — READY FOR ANYTHING.
Society for Neuroscience Booth 1613

American Society for Cell Biology Booth 309
908-231-0960 hamamatsucameras.com
Leading Edge
Review
Biology and Applications of CRISPR Systems:

Harnessing Nature’s Toolbox for Genome Engineering
Addison V. Wright,1 James K. Nuñez,1 and Jennifer A. Doudna1,2,3,4,5,6,*
1Department of Molecular and Cell Biology, University of California, Berkeley, Berkeley, CA 94720, USA
2Howard Hughes Medical Institute HHMI, University of California, Berkeley, Berkeley, CA 94720, USA
3Department of Chemistry, University of California, Berkeley, Berkeley, CA 94720, USA
4Center for RNA Systems Biology, University of California, Berkeley, Berkeley, CA 94720, USA
5Innovative Genomics Initiative, University of California, Berkeley, Berkeley, CA 94720, USA
6Physical Biosciences Division, Lawrence Berkeley National Laboratory, Berkeley, Berkeley, CA 94720, USA
*Correspondence: doudna@berkeley.edu
http://dx.doi.org/10.1016/j.cell.2015.12.035
Bacteria and archaea possess a range of defense mechanisms to combat plasmids and viral infec-
tions. Unique among these are the CRISPR-Cas (clustered regularly interspaced short palindromic
repeats-CRISPR associated) systems, which provide adaptive immunity against foreign nucleic
acids. CRISPR systems function by acquiring genetic records of invaders to facilitate robust inter-
ference upon reinfection. In this Review, we discuss recent advances in understanding the diverse
mechanisms by which Cas proteins respond to foreign nucleic acids and how these systems have
been harnessed for precision genome manipulation in a wide array of organisms.
CRISPR-Cas (clustered regularly interspaced short palindromic stage involves CRISPR locus transcription, often as a single
repeats-CRISPR associated) adaptive immune systems are pre-crRNA, and its subsequent processing into mature crRNAs
found in roughly 50% of bacteria and 90% of archaea (Makarova that each contain a single spacer. In the interference stage, an
et al., 2015). These systems function alongside restriction-modi- effector complex uses the crRNA to identify and destroy any
fication systems, abortive infections, and adsorption blocks to phage or plasmid bearing sequence complementarity to the
defend prokaryotic populations against phage infection (Labrie spacer sequence of the crRNA.
et al., 2010). Unlike other mechanisms of cellular defense, which These steps are carried out primarily by Cas proteins, which
provide generalized protection against any invaders not pos- are encoded by cas genes flanking the CRISPR arrays. The spe-
sessing countermeasures, CRISPR immunity functions analo- cific complement of cas genes varies widely. CRISPR-Cas sys-
gously to vertebrate adaptive immunity by generating records tems can be classified based on the presence of ‘‘signature
of previous infections to elicit a rapid and robust response genes’’ into six types, which are additionally grouped into two
upon reinfection. classes (Figure 1B) (Makarova et al., 2011b; Makarova et al.,
CRISPR-Cas systems are generally defined by a genomic lo- 2015; Shmakov et al., 2015). Types I–III are the best studied,
cus called the CRISPR array, a series of 20–50 base-pair (bp) while Types IV–VI have only recently been identified (Makarova
direct repeats separated by unique ‘‘spacers’’ of similar length and Koonin, 2015; Makarova et al., 2015; Shmakov et al.,
and preceded by an AT-rich ‘‘leader’’ sequence (Jansen et al., 2015). The signature protein of Type I systems is Cas3, a protein
2002; Kunin et al., 2007). Nearly two decades after CRISPR with nuclease and helicase domains that functions in foreign
loci were first identified in Escherichia coli, spacers were found DNA degradation to cleave DNA that is recognized by the
to derive from viral genomes and conjugative plasmids, serving multi-protein-crRNA complex Cascade (CRISPR-associated
as records of previous infection (Bolotin et al., 2005; Ishino complex for antiviral defense). In Type II systems, the signature
et al., 1987; Mojica et al., 2005; Pourcel et al., 2005). Sequences cas9 gene encodes the sole protein necessary for interference.
in foreign DNA matching spacers are referred to as ‘‘proto- Type III systems are signified by Cas10, which assembles into
spacers.’’ In 2007, it was shown that a spacer matching a phage a Cascade-like interference complex for target search and
genome immunizes the host microbe against the corresponding destruction. Type IV systems have Csf1, an uncharacterized
phage and that infection by a novel phage leads to the expansion protein proposed to form part of a Cascade-like complex,
of the CRISPR array by addition of new spacers originating from though these systems are often found as isolated cas genes
the phage genome (Barrangou et al., 2007). without an associated CRISPR array (Makarova and Koonin,
CRISPR immunity is divided into three stages: spacer ac- 2015). Type V systems also contain a Cas9-like single nuclease,
quisition, CRISPR RNA (crRNA) biogenesis, and interference either Cpf1, C2c1, or C2c3, depending on the subtype (Shmakov
(Figure 1A) (Makarova et al., 2011b; van der Oost et al., 2009). et al., 2015; Zetsche et al., 2015a). Type VI systems have
During spacer acquisition, also known as adaptation, foreign C2c2, a large protein with two predicted HEPN (higher eu-
DNA is identified, processed, and integrated into the CRISPR karyotes and prokaryotes nucleotide-binding) RNase domains
locus as a new spacer. The crRNA biogenesis or expression (Shmakov et al., 2015). Type I, III, and IV systems are considered
Cell 164, January 14, 2016 ª2016 Elsevier Inc. 29

Figure 1. Function and Organization of
CRISPR Systems
(A) CRISPR immunity occurs in three stages. Upon
introduction of foreign DNA, the adaptation ma-
chinery selects protospacers and inserts them into
the leader end of the CRISPR locus. During crRNA
biogenesis, the CRISPR locus is transcribed and
sequence elements in the repeats direct process-
ing of the pre-crRNA into crRNAs each with a
single spacer. The crRNA then assembles with Cas
proteins to form the effector complex, which acts
in the interference stage to recognize foreign
nucleic acid upon subsequent infection and
degrade it.
(B) CRISPR systems are extremely diverse but can
largely be classified into six major types. Repre-
sentative operons for each type are shown here.
Genes only present in some subtypes are shown
with dashed outlines. Genes involved in interfer-
ence are colored red, those involved in crRNA
biogenesis are colored yellow, and those involved
in adaptation are colored blue. Type IV systems are
notable for their frequent occurrence in the
absence of CRISPR loci.
Acquisition: Creating Genetic

Records of Past Infections
CRISPR immunity begins with the detec-
tion and integration of foreign DNA into
the host cell’s chromosome. In the Strep-
tococcus thermophilus Type II-A system,
where acquisition was first detected
experimentally, new spacers from bacte-
riophage DNA are inserted into the leader
end of the CRISPR locus, causing dupli-
cation of the first repeat to maintain the
repeat-spacer architecture (Figure 1A)
(Barrangou et al., 2007). Subsequent
studies using the E. coli Type I-E system
verified that Cas1 and Cas2 mediate
spacer acquisition (Datsenko et al.,
2012; Swarts et al., 2012; Yosef et al.,
2012). The selection of new protospacer
sequences is nonrandom and, in most
systems, depends on the presence of a
2–5 nucleotide protospacer adjacent
motif (PAM) found next to the protospacer
Class 1 systems based on their multi-subunit effector com- sequence (Deveau et al., 2008; Mojica et al., 2009). PAM-specific
plexes, while the single-subunit effector Type II, V, and VI sys- selection of protospacers is critical for immunity, as crRNA-
tems are grouped into Class 2 (Makarova et al., 2015; Shmakov guided interference in most systems depends on the PAM
et al., 2015). sequence for foreign DNA detection and destruction, which
The study of CRISPR biology has revealed enzyme mecha- avoids self-targeting at the PAM-free CRISPR locus. Interest-
nisms that can be harnessed for precision genome engineering ingly, spacers originating from the host genome are present in
and other applications, leading to an explosion of interest in almost 20% of CRISPR-containing organisms, suggesting alter-
both native CRISPR pathways and the use of these systems native roles of the CRISPR-Cas machinery in directing other pro-
for applications in animals, plants, microbes, and humans. In cesses such as endogenous gene regulation and genome evolu-
this Review, we discuss recent advancements in the field that tion (Westra et al., 2014). Spacer acquisition has been observed
reveal unexpected divergence, as well as unifying themes under- experimentally in various systems across Types I–III. Here, we
lying the three stages of CRISPR immunity. In each case, we focus on recent mechanistic studies of acquisition in Type I-E
highlight the ways in which these systems are being harnessed and Type II-A systems, in which the most comprehensive studies
for applications across many areas of biology. have been done.
30 Cell 164, January 14, 2016 ª2016 Elsevier Inc.

Figure 2. Protospacer Selection and Inte-
gration in Adaptation
(A) The selection of protospacers for acquisition is
poorly understood, but studies suggest at least
three distinct mechanisms for the selection of
substrates for integration. In Type I systems,
primed adaptation occurs when Cascade binds a
partially mismatched target. The nuclease/heli-
case Cas3 is recruited to the target site and then
likely translocates along the target DNA to a new
site. The new location is then selected as a pro-
tospacer to be used by Cas1-Cas2 in the integra-
tion reaction. In E. coli, naive adaptation involves
the nuclease/helicase RecBCD. The degradation
products appear to serve as substrates for Cas1-
Cas2, but how the variable-length single-stranded
products of RecBCD activity are converted into
double-stranded protospacers of appropriate size
is unknown. In Type II systems, Cas9 recognizes
PAM sites and likely recruits Cas1-Cas2 to acquire
the flanking sequence.
(B) Cas1-Cas2 act as an integrase to insert pro-
tospacers into the CRISPR locus as new spacers.
The complex with protospacer bound recognizes
the leader-adjacent repeat and catalyzes a pair of
transesterification reactions. The 30 OH of each
protospacer strand makes a nucleophilic attack on
the repeat backbone, one at the leader-side and
one at the spacer side. The resulting gapped
product is then repaired, causing duplication of the
first repeat.
mentary strand, suggesting that Cas1

recognizes PAM sites on potential proto-
spacers before they are processed for
integration (Wang et al., 2015).
After a spacer is acquired from a new
invader, the resulting crRNA assembles
with Cas proteins to form Cascade, the
interference complex capable of targeting
PAM-adjacent DNA sequences matching
the spacer sequence of the crRNA
(Brouns et al., 2008; Jore et al., 2011; Lint-
Type I Acquisition ner et al., 2011). Upon target binding, the helicase/nuclease
Acquisition in E. coli occurs via two mechanisms—naive and Cas3 is recruited to the site and processively degrades the
primed (Figure 2A). Naive acquisition initiates upon infection by foreign DNA (Hochstrasser et al., 2014; Mulepati and Bailey,
previously unencountered DNA and relies on the Cas1-Cas2 2011; Sinkunas et al., 2011; Sinkunas et al., 2013; Westra
integrase complex to recognize and acquire new spacers from et al., 2012). Strikingly, when Cascade encounters a mutant
foreign DNA. Overexpression of Cas1 and Cas2 in the absence PAM or protospacer that prevents Cas3 degradation, hyperac-
of other Cas proteins leads to the acquisition of 33 bp spacers tive spacer acquisition from the targeted plasmid or genome is
at the leader-proximal end of the CRISPR array (Datsenko triggered in a process called ‘‘priming’’ (Figure 2A) (Datsenko
et al., 2012; Yosef et al., 2012). The PAM of the E. coli et al., 2012; Li et al., 2014; Richter et al., 2014; Savitskaya
CRISPR-Cas system was identified as 50 -AWG-30 , with the et al., 2013; Swarts et al., 2012). Priming increases the host’s
G becoming the first nucleotide of the integrated spacer (Dat- repertoire of functional spacers, allowing the host to adapt to in-
senko et al., 2012; Dı́ez-Villaseñor et al., 2013; Levy et al., vaders that evade the CRISPR-Cas system by mutation.
2015; Nuñez et al., 2014; Savitskaya et al., 2013; Shmakov Cascade is capable of binding escape mutant target sites, and
et al., 2014; Swarts et al., 2012; Yosef et al., 2012; Yosef et al., recent single-molecule studies showed that the presence of
2013). In addition to the PAM, a dinucleotide motif, AA, found Cas1 and Cas2 allows for the recruitment of Cas3 to these sites
at the 30 end of the protospacer was also shown to be present (Blosser et al., 2015; Redding et al., 2015; Richter et al., 2014).
in a disproportionately large number of spacers (Yosef et al., The recruited Cas3 can then translocate in either direction, in
2013). A recent crystal structure of the Cas1-Cas2 complex contrast to the unidirectional movement observed at perfect tar-
bound to an unprocessed protospacer revealed sequence-spe- gets, without degrading the target DNA (Redding et al., 2015).
cific contacts with the 50 -CTT-30 sequence on the PAM-comple- Cas1 and Cas2 may accompany the translocating Cas3 and

be activated for protospacer selection, allowing for robust acqui- isms have evolved to cooperate with these evolutionarily distinct
sition on either side of the target site. machineries remains to be tested.
Primed acquisition has also been shown experimentally in the Mechanism of Protospacer Integration
P. atrosepticum Type I–F system, in which Cas2 and Cas3 are Cas1 and Cas2 play central roles in the acquisition of new
naturally fused as a single polypeptide that associates with spacers, where they function as a complex (Nuñez et al.,
Cas1, as well as in the Haloarcula hispanica Type I-B system, 2014). Crystal structures of Cas1 and Cas2, with or without
where naive acquisition was not experimentally observed (Li bound protospacer, revealed two copies of a Cas1 dimer
et al., 2014; Richter et al., 2014; Richter et al., 2012). Acquisition bridged by a central Cas2 dimer (Nuñez et al., 2014; Nuñez
in H. hispanica also requires Cas4, a 50 /30 exonuclease found in et al., 2015b; Wang et al., 2015). Cas1 functions catalytically,
most Type I subtypes as well as Type II-B and Type V systems, while Cas2 appears to serve a primarily structural role (Arslan
and which might be involved in generating 30 overhangs on pro- et al., 2014; Datsenko et al., 2012; Nuñez et al., 2014; Yosef
tospacers prior to integration (Lemak et al., 2013; Li et al., 2014; et al., 2012).
Makarova et al., 2015). Although Cas1 and Cas2 may be the min- The first insight into the mechanism of protospacer integration
imal proteins required for spacer acquisition in some systems, was gained by Southern blot analysis of the genomic CRISPR
the association of Cas1, Cas2, and the interference machinery locus of E. coli cells overexpressing Cas1 and Cas2 (Arslan
allows the host to coordinate robust adaptive immunity in et al., 2014). This revealed integration intermediates consistent
Type I systems. with two transesterefication reactions, where each strand of
Self- versus Non-Self-Recognition the protospacer is integrated into opposite sides of the leader-
The mechanism underlying the preference for foreign over self proximal repeat (Figure 2B). This integrase-like model was
DNA during protospacer selection remained poorly understood further bolstered by the in vitro reconstitution of protospacer
until a recent study on spacer acquisition during naive acquisi- integration into a plasmid-encoded CRISPR locus using purified
tion. Spacer acquisition in E. coli was shown to be highly depen- Cas1-Cas2 complex (Nuñez et al., 2015a). The integration reac-
dent on DNA replication, and foreign-derived spacers were tion required double-stranded DNA protospacers with 30 -OH
preferred over self-derived spacers by about 100- to 1,000- ends that are integrated into plasmid DNA via a direct nucleo-
fold (Levy et al., 2015). Analysis of the source of self-derived philic transesterification reaction, reminiscent of retroviral inte-
spacers demonstrated that protospacers were acquired largely grases and DNA transposases (Engelman et al., 1991; Mizuuchi
from genomic loci predicted to frequently generate stalled repli- and Adzuma, 1991).
cation forks and double-stranded DNA breaks (Levy et al., 2015). Although deep sequencing of in vitro integration products re-
Such harmful dsDNA breaks are repaired by the helicase/ vealed preferential protospacer integration adjacent to the first
nuclease RecBCD complex, which degrades the broken ends repeat, confirming that Cas1-Cas2 directly recognize the
until reaching a Chi-site, after which only the 50 end is degraded CRISPR locus, integration also occurred at the borders of every
(Dillingham and Kowalczykowski, 2008). Due to the lower repeat at varying levels (Nuñez et al., 2015a). This contrasts
frequency of Chi sites in foreign DNA, RecBCD is predicted to with spacer acquisition only occurring at the first repeat in
preferentially degrade plasmids and viral DNA, resulting in the E. coli in vivo (Datsenko et al., 2012; Swarts et al., 2012; Yosef
generation of candidate protospacer substrates for Cas1 and et al., 2012). To determine if the Cas1-Cas2 complex has
Cas2 (Levy et al., 2015) (Figure 2A). RecBCD degrades DNA sequence specificity for the leader-repeat sequence, a recent
asymmetrically, yielding single-stranded fragments ranging study took advantage of the Cas1-catalyzed disintegration reac-
from tens to hundreds of nucleotides long from one strand and tion, a reversal of the integration reaction also observed with
kilobases long from the other (Dillingham and Kowalczykowski, retroviral integrases and transposases (Chow et al., 1992; Rollie
2008). It is unclear how Cas1-Cas2 substrates, which are et al., 2015). Disintegration activity was stimulated when using
33 bp long and partially double stranded with 30 overhangs, are the correct leader-repeat border sequences, highlighting intrinsic
generated from RecBCD products (Nuñez et al., 2015a; Nuñez sequence-specific recognition by Cas1. Furthermore, disintegra-
et al., 2015b; Wang et al., 2015). It is possible that ssDNA prod- tion was faster at the leader-repeat junction compared to the
ucts re-anneal to produce partial duplexes, followed by process- repeat distal end (Rollie et al., 2015). Taken together, protospacer
ing to 33 bp by an unknown mechanism prior to integration into integration likely begins at the leader-repeat junction via
the CRISPR locus. Recent crystal structures of Cas1-Cas2 with sequence-specific recognition by Cas1, followed by a second
bound protospacer reveal that the complex defines the length of nucleophilic attack at the repeat distal end. This ensures precise
the duplex region of the protospacer via a ruler mechanism and duplication of the first repeat, as observed in vivo, after DNA
may cleave the 30 overhangs to their final length (Nuñez et al., repair by host proteins. The integration mechanism is hypothe-
2015b; Wang et al., 2015). The involvement of a helicase/ sized to be highly specific, as almost all acquired spacers with
nuclease in both Type I-E primed and naive acquisition (Cas3 a corresponding AAG PAM are oriented with the 50 -G at the
and RecBCD, respectively), as well as in Cas4-containing sub- leader-proximal end, leading to functional crRNA-dependent tar-
types, hints at a conserved mechanism for protospacer genera- geting by Cascade and Cas3 (Shmakov et al., 2014). A preference
tion. It is also worth noting that RecBCD is conserved primarily for integration in the proper orientation was observed in vitro
in Gram-negative bacteria, while Gram-positive bacteria when protospacers with a 50 -G were used (Nuñez et al., 2015a);
and archaea rely on AddAB and HerA-NurA, respectively, to fill however, inclusion of part of the PAM in spacers has only been
a similar role (Blackwood et al., 2013; Dillingham and Kowalczy- observed in E. coli, raising the question of how Cas1-Cas2 in
kowski, 2008). Whether CRISPR-Cas systems in these organ- other systems properly orient the integration reaction.

Type II Acquisition the E. coli Type I-E system, where Cas1-Cas2 recognize PAM
While most mechanistic work on acquisition has been performed sequences independently.
in Type I systems, recent studies in Type II systems have also Intriguingly, expression of dCas9 results in the acquisition of
shed light on key aspects of spacer acquisition. One generaliz- primarily self-targeting spacers, suggesting that many acquisi-
able finding in Type II systems is the dependence of acquisition tion events lead to self-targeting and suicide (Wei et al.,
on infection by defective phage (Hynes et al., 2014). A significant 2015b). Microbial populations may rely on a few individuals to
problem with CRISPR immunity is the time required for foreign acquire phage resistance while the rest succumb to infection
DNA to be identified, integrated into the CRISPR locus, tran- or CRISPR-mediated suicide. Some systems, such as that found
scribed, processed, and assembled into an interference com- in E. coli, may evolve to use host processes to bias acquisition
plex that must then begin the search for appropriate targets. away from self-targeting. Alternatively, S. thermophilus might
Since lytic phage can kill cells within 20 min, providing insuffi- have mechanisms of self-non-self-discrimination that were
cient time for this multi-step process, Hynes and colleagues masked in the strain overexpressing CRISPR proteins. Phage
tested the hypothesis that initial immunization takes place from challenge experiments with wild-type S. thermophilus revealed
infection by a defective phage. Supplementation of active phage that some sequences were acquired as spacers disproportion-
with UV-irradiated phage or phage susceptible to a restriction- ately often across multiple experiments, suggesting that the
modification system stimulated spacer acquisition compared Type II acquisition machinery has preferences in addition to
to that observed with active phage alone (Hynes et al., 2014). Cas9-dependent PAM selection, though no clear pattern
The authors speculate that acquisition from compromised phage emerged with respect to the genomic location or sequence of
might also represent the dominant mode of acquisition in wild protospacers that indicated a basis for the preferences (Paez-
populations, allowing for a small subset of the population to ac- Espino et al., 2013).
quire resistance and escape without needing to outpace a Additionally, it was demonstrated that the four proteins of the
rapidly reproducing phage. S. pyogenes CRISPR system (Cas1, Cas2, Csn2, and Cas9) form
Type II Acquisition Machinery a complex, suggesting that Cas9 directly recruits the acquisition
Type II systems are subdivided into II-A, II-B, and II-C based on proteins to potential targets (Heler et al., 2015). While drawing
the presence or absence of an additional cas gene alongside the comparisons between the involvement of Cas9 in acquisition
minimal complement of cas1, cas2, and cas9. Type II-A systems and primed acquisition in Type I systems is tempting, neither
contain csn2, while Type II-B systems, which are least commonly group saw evidence that acquisition was affected by the pres-
found, contain cas4 (Chylinski et al., 2014; Makarova et al., ence of existing spacers matching or closely matching the infect-
2011b). Type II-C systems comprise only the minimal three ing phage or plasmid (Heler et al., 2015; Wei et al., 2015b). In
genes. Csn2 has been shown to be essential for acquisition in addition, while the trans-activating crRNA (tracrRNA) that forms
several Type II-A systems (Barrangou et al., 2007; Heler et al., a complex with Cas9 and the crRNA is necessary for acquisition,
2015; Wei et al., 2015b). It forms a tetramer with a torroidal archi- it is unclear whether a corresponding crRNA is also required (He-
tecture that binds and slides along free DNA ends, though its ler et al., 2015; Wei et al., 2015b). Future mechanistic work will be
function in CRISPR systems is unclear (Arslan et al., 2013; required to shed light on the similarities between Cas9-mediated
Ellinger et al., 2012; Koo et al., 2012; Lee et al., 2012). Cas4, spacer acquisition and the primed acquisition in Type I systems.
discussed above, is likely involved in acquisition in Type II-B sys- Type II Protospacer Integration
tems. Type II-C systems, which constitute the majority of identi- The sequence requirements for protospacer integration in Type
fied Type II systems (Chylinski et al., 2014; Makarova et al., II-A systems were recently demonstrated in S. thermophilus
2015), are possibly functional for acquisition in the absence of (Wei et al., 2015a). Similar to E. coli, the leader and a single
auxiliary acquisition factors, though in the case of the Campylo- repeat were sufficient to direct integration. Furthermore, only
bacter jejuni system, acquisition was only observed following the ten nucleotides of the leader proximal to the first repeat are
infection by phage encoding a Cas4 homolog (Hooton and Con- required to license the integration of new spacers, in contrast
nerton, 2014). to the 60 nt minimal requirement in E. coli (Wei et al., 2015a; Yo-
Recently, two simultaneous studies demonstrated that, in sef et al., 2012). A limited mutational study of the repeat showed
addition to Cas1, Cas2, and Csn2, Cas9 plays a necessary role that the first two nucleotides are necessary for acquisition, while
in the acquisition of new spacers in Type II systems (Heler the final two nucleotides can be mutated without consequence
et al., 2015; Wei et al., 2015b). Both groups, one working with (Wei et al., 2015a). Thus, Cas1-Cas2-catalyzed integration at
the CRISPR1 Type II-A system of S. thermophilus, the other the leader-repeat junction is sequence specific, while the attack
with the Type II-A system of Streptococcus pyogenes and the at the repeat-spacer junction is determined by a ruler mecha-
CRISPR3 system of S. thermophilus, also Type II-A, showed nism, in agreement with observations from experiments in the
that wild-type or catalytically inactive Cas9 (dCas9) supported E. coli system (Dı́ez-Villaseñor et al., 2013). Together, these find-
robust spacer acquisition, whereas deletion of Cas9 abolished ings support the functional conservation of the Cas1-Cas2 inte-
spacer acquisition. It is proposed that Cas9 serves to recognize grase complex despite divergent mechanisms of protospacer
PAM sites in potential protospacers and mark them for recogni- selection between Types I and II CRISPR-Cas systems.
tion by Cas1 and Cas2 (Figure 2A). This hypothesis was CRISPR Integrases as Genome-Modifying Tools
confirmed by mutating the PAM-interacting residues of Cas9, re- As with many other Cas proteins, the Cas1-Cas2 integrase
sulting in complete loss in PAM-specificity in the newly acquired complex shows promise for use in modifying genomes. While
spacers (Heler et al., 2015). This presents a striking contrast to Cas1-Cas2 catalyze a reaction similar to that of many integrases

and transposases, they exhibit several fundamental differences teins in Haloferax volcanii (Cas6b), E. coli and T. thermophilus
that make them uniquely suited to certain applications. Cas1- (Cas6e), and Pseudomonas aeruginosa (Cas6f) remain stably
Cas2 complexes lack sequence specificity for the DNA substrate bound to the 30 handle and eventually become part of the
to be integrated, a property that could make the system ideal for Cascade complex (Brendel et al., 2014; Brouns et al., 2008; Ges-
barcoding genomes. Genome barcoding allows for tracking lin- ner et al., 2011; Haurwitz et al., 2010; Sashital et al., 2011).
eages originating from individual cells, facilitating studies of pop- Type I-A, I-B, III-A, and III-B repeat sequences are non-palin-
ulation evolution, cancer, development, and infection (Blundell dromic and predicted to be unstructured in solution (Kunin et al.,
and Levy, 2014). Cas1-Cas2 complexes integrate short DNA se- 2007). Thus, the respective Cas6 is thought to rely on sequence
quences, in contrast with current techniques based on recombi- for specificity rather than structure. Interestingly, a crystal struc-
nases that integrate entire plasmids, resulting in potential fitness ture of the Type I-A Cas6 bound to its cognate RNA structure re-
costs and unwanted negative selection (Blundell and Levy, veals Cas6 inducing a 3 bp hairpin in the RNA that positions the
2014). Interestingly, in vitro integration of DNA substrates into scissile phosphate in the enzyme active site (Shao and Li, 2013).
plasmid targets revealed integration into non-CRISPR sites (Nu- It remains unknown whether other Cas6s that recognize non-
ñez et al., 2015a), suggesting that Cas1-Cas2 can be harnessed palindromic repeats have a similar mechanism of RNA stabiliza-
to integrate into a wide array of target sequences. A greater un- tion. Following or concurrent with the maturation of the crRNAs,
derstanding of the minimal functional recognition motif for the Cas proteins involved in interference assemble into the final
various Cas1-Cas2 integrases will facilitate the development of effector complex that functions to recognize and destroy targets
this technology. bearing sequence complementarity to the crRNA. In systems
where Cas6 remains bound to the crRNA, it may serve to
crRNP Biogenesis: Generating Molecular Sentinels for nucleate the assembly of the subunits that constitute the effector
the Cell complex backbone along the crRNA. In type III systems, the
CRISPR immune systems use RNA-programmed proteins to pa- number of backbone subunits defining the complex length is var-
trol the cell in search of DNA molecules bearing sequences com- iable, and any unprotected crRNA remaining is degraded (Hale
plementary to the crRNA. Assembly of these molecular sentinels et al., 2008; Staals et al., 2014).
begins with transcription of the CRISPR locus to generate long, Processing in Type II Systems
precursor CRISPR RNAs (pre-crRNAs), followed by processing Type II systems rely on a different mechanism to process pre-
into short crRNA guides (Brouns et al., 2008; Carte et al., crRNAs. In Types II-A and II-B, pre-crRNA cleavage specificity
2008). The promoter is embedded within the AT-rich leader is aided by a tracrRNA that has sequence complementarity to
sequence upstream of the repeat-spacer array, or sometimes the CRISPR repeat sequence (Deltcheva et al., 2011). The gene
within the repeat sequences (Zhang et al., 2013). Here, we briefly encoding the tracrRNA is typically located either proximal to or
review the processing of pre-crRNAs catalyzed by the Cas6 en- within the CRISPR-cas locus (Chylinski et al., 2014). Upon
doribonuclease family in Type I and III systems and a distinct crRNA:tracrRNA base pairing, which is stabilized by Cas9,
processing pathway in Type II systems that involves endoge- endogenous RNase III cleaves the pre-crRNA at the repeat. The
nous RNase III, Cas9, and a tracrRNA. The crRNA biogenesis reliance on RNase III, which is not found in archaea, may explain
pathway has been extensively reviewed elsewhere (Charpentier why Type II systems are limited to bacteria (Garrett et al., 2015).
et al., 2015; Hochstrasser and Doudna, 2015). An unknown nuclease trims the 50 end of the crRNA to remove
Processing by Cas6 Endoribonucleases the flanking repeat sequence and portions of the spacer. In
Type I and Type III systems employ Cas6 endoribonucleases to S. pyogenes, the 30 nt spacer sequence is trimmed to the 20 nt
cleave pre-crRNAs sequence specifically within each repeat that base-pairs with complementary foreign sequences during
(Brouns et al., 2008; Carte et al., 2008; Haurwitz et al., 2010). interference (Deltcheva et al., 2011; Jinek et al., 2012).
Although Cas6 homologs are variable in sequence, they share In the Neisseria meningitidis and C. jejuni Type II-C systems,
a conserved cleavage mechanism that results in crRNA guides each repeat sequence encodes a promoter, resulting in varying
comprising an entire spacer sequence flanked by portions of lengths of pre-crRNAs depending on the transcription start site
the repeat sequence on the 50 and 30 ends. Mature crRNA guides (Dugar et al., 2013; Zhang et al., 2013). Although RNase III-medi-
consist of an 8 nt 50 handle derived from the repeat sequence ated pre-crRNA processing can still occur, RNase III is dispens-
and variable lengths of the repeat at the 30 handle, which is able for interference in these systems (Zhang et al., 2013). Thus,
further trimmed by as-yet-unidentified cellular nuclease(s) in Cas9 is able to complex with the pre-crRNA and unprocessed
Type III systems (Hale et al., 2008). A notable exception is in tracrRNA for functional target interference without further pro-
Type I-C systems, which utilize a Cas5 variant for crRNA processing of the pre-crRNAs.
cessing, leaving an 11 nt 50 handle and 21–26 nt at the 30 end Cas6 as a Biotechnology Tool
(Garside et al., 2012; Nam et al., 2012b). In other Type I systems, The Cas6 homolog from Type I-F systems, Cas6f (also known as
Cas5 subunits serve a non-catalytic role capping the 50 end of Csy4), was the first Cas protein to be repurposed as a tool.
the crRNA in Cascade complexes. Following demonstration of the sequence specificity of Cas6f
In Type I-C, I-D, I-E, and I-F systems, the repeats form stable binding and cleavage, the protein has been used for the purifica-
hairpin structures that allow for structure- and sequence-spe- tion of tagged RNA transcripts from cells (Haurwitz et al., 2010;
cific cleavage by Cas6 at the base of the hairpin (Gesner et al., Lee et al., 2013; Salvail-Lacoste et al., 2013; Sternberg et al.,
2011; Haurwitz et al., 2010; Sashital et al., 2011). After cleavage, 2012). Subsequent studies showed that Cas6f could be used
the hairpin constitutes the 30 handle of the crRNA. The Cas6 pro- to alter the translation and stability of tagged mRNAs, allowing

Figure 3. Interference by Class 1 Systems
(A) Interference in Type I systems is carried out by
Cascade and Cas3. Cascade is a large complex
composed of the crRNA, bound at either end by
Cas5 and Cas6, multiple Cas7 subunits along the
crRNA, a large subunit (Cse1, Csy1, Cas8, or
Cas10), and sometimes small subunits (Cse2 and
Csa5). The Type I-E complex is schematized here.
The large subunit recognizes the PAM in foreign
DNA and initiates unwinding of the target DNA and
annealing to the crRNA. Cas3 is recruited to the
resulting R-loop and makes a nick. It then trans-
locates along the displaced strand and proc-
essively degrades it.
(B) Type III systems contain either Csm or Cmr
complexes, which share a similar architecture. The
Csm complex from Type III-A systems is shown
here. The crRNA is bound at either end by Csm5/
Cmr1 and Csm4/Cmr3, which have homology to
Cas6 and Cas5, respectively. Csm3/Cmr4 form the
backbone of the complex, Cas10 serves as the
large subunit, and Csm2/Cmr5 are the small sub-
unit. These complexes can target both RNA and
actively transcribed DNA. Cas10 catalyzes cleav-
age of target DNA, while the backbone subunit
catalyzes cleavage of the target RNA at every sixth
base, which is unpaired with the crRNA. Rather
than recognizing a PAM sequence, these com-
plexes only cleave if the 50 and 30 handles of the
crRNA do not anneal to the target.
for post-transcriptional regulation of protein expression (Borch- a distinct complement of Cascade components and, in some
ardt et al., 2015; Du et al., 2015; Nissim et al., 2014). Cas6f has cases, significant variation of the cas3 gene (Makarova et al.,
also been used alongside Cas9 to process multiple guide 2011b).
RNAs from a single transcript, greatly facilitating multiplexed ed- The E. coli Cascade complex has served as the model system
iting (Tsai et al., 2014). for understanding the mechanism of Type I interference. In addi-
tion to the central 61 nt crRNA bearing the 32 nt spacer, the com-
Interference: Precise, Programmable DNA Binding and plex comprises five proteins in different stoichiometries: (Cse1)1,
Cleavage (Cse2)2, (Cas5)1, (Cas7)6, and (Cas6)1. The Cas7 subunits form
Implementation of CRISPR systems to provide immunity in- the ‘‘backbone’’ that polymerizes along the crRNA and deter-
volves RNA-guided recognition and precision cutting of DNA mines the crescent-shaped, semi-helical architecture seen in
molecules, a property that makes them useful for genome engi- all structurally characterized Cascade complexes (Hochstrasser
neering and control of gene expression. The extreme diversity of et al., 2014; Jackson et al., 2014; Jore et al., 2011; Mulepati et al.,
the crRNP targeting complexes is largely responsible for the vari- 2014; Wiedenheft et al., 2011a; Zhao et al., 2014). Cas6 (Cas6e in
ability observed in different CRISPR types. Whereas Types I and Type I-E systems) remains bound to the 30 hairpin following
III use multi-protein complexes, Types II and V rely on a single CRISPR maturation, while Cas5 binds the 50 handle (Brouns
protein for interference. Extensive studies have elucidated the et al., 2008; Jore et al., 2011). A ‘‘small subunit’’ (Cse2 in Type
mechanisms and structures of several complexes from each of I-E) is often found in two copies forming the ‘‘belly’’ of the struc-
the three major types, revealing the commonality of target bind- ture and helps stabilize the crRNA and target DNA (Jackson
ing through crRNA base-pairing and high divergence in the ma- et al., 2014; Mulepati et al., 2014; Zhao et al., 2014). A ‘‘large sub-
chineries and modes of target cleavage. For more in-depth unit’’ (Cse1 in Type I-E, Cas8 in most other subtypes) binds at the
recent reviews focused exclusively on CRISPR interference, 50 end of the crRNA and recognizes the PAM sequences and re-
refer to Tsui and Li (2015) and Plagens et al. (2015). cruits Cas3 to an authenticated target (Figure 3A) (Hochstrasser
Type I Interference et al., 2014; Sashital et al., 2012). While Cas6 does not always
In Type I systems, the roles of target DNA recognition and degra- remain with the complex and the small subunit is often found
dation are segregated into two distinct components. The crRNA- as a fusion with the large subunit, the overall architecture of
guided Cascade complex binds and unwinds the DNA target Cascade complexes is generally conserved (Makarova et al.,
sequence (Brouns et al., 2008) and then recruits Cas3 to degrade 2011b; Plagens et al., 2012; Sokolowski et al., 2014).
the target in a processive manner through the combined action Cascade pre-arranges the spacer segment of the crRNA in six
of its HD nuclease and helicase domains (Figure 3A) (Makarova five-base segments of pseudo A-form conformation, with the
et al., 2011b; Mulepati and Bailey, 2013; Sinkunas et al., 2013; sixth base flipped out and bound by a Cas7 subunit (Jackson
Westra et al., 2012). Each Type I subtype (I-A through I-F) has et al., 2014; Mulepati et al., 2014; Zhao et al., 2014). To initiate

Cse1 subunit (Hochstrasser et al., 2014; Mulepati et al., 2014).
Cas3 binds and nicks the displaced strand using its catalytic
center of the HD nuclease domain (Gong et al., 2014; Huo
et al., 2014; Mulepati and Bailey, 2013; Sinkunas et al., 2013;
Westra et al., 2012). The ATP-dependent helicase activity of
Cas3 is then activated, causing metal- and ATP-dependent
30 /50 translocation and processive degradation of the non-
target strand (Gong et al., 2014; Huo et al., 2014; Westra et al.,
2012). Cas3 initially degrades only 200–300 nt of the nontarget
strand, though it continues translocating for many kilobases
(Redding et al., 2015). Exposed ssDNA on the target strand
may then become a substrate for other ssDNA nucleases or an
additional Cas3 molecule to complete the degradation of foreign
DNA (Mulepati and Bailey, 2013; Redding et al., 2015; Sinkunas
et al., 2013). In addition to the PAM, target interference also relies
on a seed region at the 30 end of the spacer segment of the
crRNA (Semenova et al., 2011; Wiedenheft et al., 2011b). Single
point mutations of the seed region of the E. coli Cascade com-
plex, at the 1 to 5 and 7 to 8 position of the spacer, is enough
to decrease target DNA binding and subsequent interference
(Semenova et al., 2011).
Differences in the cas3 gene among Type I subtypes suggest
some variability in interference mechanism. In some Type I-E
species, Cas3 is fused to Cse1 by a linker that allows it to stably
associate with the Cascade complex (Westra et al., 2012). In
Type I-A systems, the Cas3 helicase and nuclease domains exist
as separate polypeptides that both associate with the Cascade
complex (Plagens et al., 2014). In Type I-F systems, Cas3 is
fused to Cas2, lending further genetic support for the interaction
between the interference and acquisition machinery during
primed acquisition (Makarova et al., 2015; Richter and Fineran,
2013; Richter et al., 2012). How these fusions and domain sepa-
rations affect the processive degradation observed in Type I-E
Figure 4. Interference by Class 2 Systems
systems requires further study.
(A) In Type II systems, Cas9 forms the effector complex with a crRNA and a
tracrRNA. Cas9 is composed of the nuclease lobe and the a-helical lobe. The Type II Interference
nuclease lobe contains both the HNH and RuvC-like nuclease domains as well In contrast to the multi-subunit effector complexes seen in Type I
as the PAM-interacting domain. The 30 hairpins of the tracrRNA bind the and Type III systems (but similar to Cpf1 of Type V systems), the
nuclease lobe, while the stemloop and spacer line the channel between the
two lobes. Binding to a matching, PAM-adjacent target causes the HNH Type II signature protein Cas9 functions as an individual protein,
domain to move into position to cleave the annealed strand, while the dis- along with a crRNA and tracrRNA, to interrogate DNA targets
placed strand is fed into the RuvC active site for cleavage. and destroy matching sequences by cleaving both strands of
(B) Cpf1 is the effector protein in Type V-A systems, the best characterized
Type V subtype. It binds the crRNA alone. The structure of Cpf1 is unknown,
the target (Figure 4A) (Gasiunas et al., 2012; Jinek et al., 2012).
but it contains an active RuvC-like nuclease domain for target cleavage. Cpf1 Extensive studies on Cas9 have yielded a range of structures
recognizes a PAM and makes two staggered cuts in a matching sequence. It of S. pyogenes Cas9 in different substrate-bound states, as
has been proposed that Cpf1 acts as a dimer, with each monomer providing a well as structures of several orthologs (Anders et al., 2014; Jiang
RuvC active site, though there may be another unidentified nuclease domain.
et al., 2015; Jinek et al., 2014; Nishimasu et al., 2015; Nishimasu
et al., 2014). Many of these structures, as well as the mechanism
interference, Cascade first recognizes trinucleotide PAM sites in of Cas9 target search and recognition, are reviewed elsewhere
the target strand of foreign DNA through specific interactions (van der Oost et al., 2014); here, we focus on the most recent
with Cse1 (Sashital et al., 2012). Upon PAM binding, the DNA advances.
target is unwound starting at the PAM-proximal end of the proto- Structures of Cas9 have revealed two distinct lobes, the
spacer to form an R loop structure (Hochstrasser et al., 2014; nuclease lobe and the a-helical or REC lobe (Anders et al.,
Rollins et al., 2015; Rutkauskas et al., 2015; Sashital et al., 2014; Jinek et al., 2014; Nishimasu et al., 2015; Nishimasu
2012; Szczelkun et al., 2014; van Erp et al., 2015). Each stretch et al., 2014). The nuclease lobe is composed of the HNH
of five exposed bases in the crRNA is free to bind the target nuclease domain, which cleaves the target strand, a RuvC-like
DNA, leading to a stable but highly distorted and discontinuous nuclease domain, which cleaves the non-target strand and is
crRNA:target strand duplex (Mulepati et al., 2014; Szczelkun separated into three distinct regions in the primary sequence
et al., 2014). Cascade undergoes a conformational change by the intervening a-helical lobe and the HNH domain, and a
upon target binding that enables recruitment of Cas3 to the C-terminal PAM-interacting domain (Anders et al., 2014; Jinek

et al., 2014; Nishimasu et al., 2015; Nishimasu et al., 2014). The Cas9 from the Type II-C CRISPR system of S. aureus was
a-helical lobe contains an arginine-rich ‘‘bridge helix,’’ which crystallized in complex with sgRNA and a single-stranded DNA
connects the two lobes and interacts with the guide RNA, and target sequence, providing insight into the structural variation
is the most variable region of Cas9, with insertions or deletions between more distantly related Cas9 (Nishimasu et al., 2015).
accounting for much of the wide variation in size seen in Cas9 S. aureus Cas9 is significantly smaller than the Cas9 of
orthologs (Chylinski et al., 2014; Jinek et al., 2014; Nishimasu S. pyogenes (1,053 versus 1,368 amino acids) and recognizes
et al., 2014). a significantly different guide RNA and PAM site. The S. aureus
Cas9 initiates its target search by probing duplexed DNA for Cas9 structure revealed a smaller a-helical lobe, with domains
an appropriate PAM before initiating target unwinding (Sternberg in the middle and PAM-proximal side notably absent, while the
et al., 2014). The target unwinds from the seed region, the first nuclease lobe is largely conserved (Nishimasu et al., 2015).
10–12 nucleotides following the PAM, toward the PAM-distal The authors proposed a new domain designation, the wedge
end (Szczelkun et al., 2014). A perfect or near-perfect match domain, which diverges significantly between the two proteins
leads to cleavage of both DNA strands, with mismatches being and appears integral to determining guide RNA orthogonality.
more tolerated outside of the seed region (Cong et al., 2013; Another small Cas9, that from Actinomyces naeslundi, was pre-
Jiang et al., 2013; Jinek et al., 2012; Sternberg et al., 2014). viously crystallized in the apo form, but the absence of bound
The mechanism by which mismatches distant from the cleavage substrate and significant disordered regions limited detailed
site prevent cleavage appears to rely on the structural flexibility exploration of the differences between the orthologs (Jinek
of the HNH domain, which has yet to be crystallized in proximity et al., 2014). Other recent work with Type II-C Cas9 proteins
to the scissile phosphate (Anders et al., 2014; Nishimasu et al., from N. meningitidis and Corynebacterium diphtheriae, among
2015; Nishimasu et al., 2014). FRET assays show that the HNH other Type II-C orthologs, revealed that these enzymes have a
domain swings into a catalytically competent position only reduced ability to unwind dsDNA compared to S. pyogenes
upon binding to a cognate double-stranded DNA substrate, Cas9 and exhibit efficient PAM-independent and in some cases
underscoring the multiple steps of conformational control of tracrRNA-independent cleavage of ssDNA (Ma et al., 2015;
Cas9-catalyzed DNA cleavage (Sternberg et al., 2015). The Zhang et al., 2015). This activity may allow for more efficient
RuvC domain is in turn allosterically regulated by the HNH interference with ssDNA plasmid or phage or represent a more
domain. Cleavage of the non-target strand requires movement ancestral activity that predates the expansion of the a-helical
of the HNH domain into an active position, even when the mis- lobe to facilitate more robust DNA unwinding.
matched substrates allow full unwinding of the non-target strand Type III Interference
(Sternberg et al., 2015). Type III systems are classified into Type III-A and Type III-B
Recent crystal structures of S. pyogenes Cas9-sgRNA surveil- based on their effector complexes (Type III-C and III-D have
lance complex and of the smaller Staphylococcus aureus Cas9 in also been identified, but not yet characterized) (Makarova
a target-bound state provided new insights into Cas9 function et al., 2015). The former is constituted by the Csm complex,
(Jiang et al., 2015; Nishimasu et al., 2015). The sgRNA-bound and the latter by the Cmr complex (Makarova et al., 2011b).
structure revealed how binding of sgRNA shifts Cas9 from the Phylogenetic studies suggested that some csm and cmr genes
auto-inhibited state observed in the apo form to a conformation are distant homologs of cas genes that compose the Cascade
competent for target search (Jiang et al., 2015; Jinek et al., complex of Type I systems, and subsequent structural studies
2014). As previously observed in low-resolution electron micro- have revealed a striking structural conservation between
scopy structures, a nucleic acid binding cleft is formed between Cascade and the Csm and Cmr complexes (Hochstrasser
the two lobes upon sgRNA binding (Jinek et al., 2014). Further- et al., 2014; Jackson et al., 2014; Makarova et al., 2013; Mulepati
more, two PAM-interacting arginine residues are pre-positioned et al., 2014; Osawa et al., 2015; Staals et al., 2014; Taylor et al.,
to allow for scanning of potential target DNA, a finding that may 2015; Zhao et al., 2014). For a detailed discussion of the struc-
explain the necessity of tracrRNA in directing PAM-dependent tural similarities between these complexes, refer to Jackson
spacer acquisition. Surprisingly, while the 30 hairpins of the and Wiedenheft (2015). Briefly, Csm3 (in III-A systems) or Cmr4
tracrRNA have been shown to provide nearly all of the binding (in III-B) polymerizes along the crRNA as a helical backbone,
energy and specificity for Cas9, the repeat-anti-repeat region analogously to Cas7, while Csm2 or Cmr5 take the role of
of the sgRNA as well as the seed sequence were required to Cse2 as the small subunit (Figure 3B) (Jackson and Wiedenheft,
induce the conformational rearrangement (Briner et al., 2014; 2015). Similar to Cascade, the crRNA is pre-arranged for binding
Jiang et al., 2015; Wright et al., 2015). The seed sequence of with kinks every six nucleotides. The target nucleic acid (RNA in
the sgRNA was also found to be pre-ordered in an A-form helix, all solved Type III structures) binds in a distorted manner, form-
analogous to the pre-ordered seed region of guide RNA ing five-nucleotide helical stretches with the sixth base flipped
observed in eukaryotic Argonaute structures and the Type I out to allow for the extreme deviation from helical nucleic acid
and Type III effector complexes, where the entire crRNA is pre- observed in all structures (Osawa et al., 2015; Taylor et al.,
arranged in a target-binding-competent state (Jackson et al., 2015). Cmr3 and Csm4 bind the 50 crRNA handle, while Cas10
2014; Kuhn and Joshua-Tor, 2013; Mulepati et al., 2014; (also referred to as Csm1 and Cmr2) serves as the large subunit
Osawa et al., 2015; Taylor et al., 2015; Zhao et al., 2014). The (Makarova et al., 2011a; Osawa et al., 2015; Staals et al., 2014;
observed pre-ordering of the guide RNA provides an energetic Taylor et al., 2015). Csm5, Cmr6, and Cmr1 also share homology
compensation for the unwinding of the target duplex to facilitate with Cas7 and cap the helical backbone at the 30 end of the
binding. crRNA. In Type III-B systems, two major crRNA species are

generally observed, differing by six nucleotides (Juranek et al., for interference with targets that somehow avoid recognition
2012; Staals et al., 2014). Cryo-electron microscopy captured by Cascade (Makarova et al., 2011b).
two Cmr complexes of different sizes, with one complex having Type III systems are also unique in their lack of a PAM. Rather
one fewer Cmr4 and Cmr5 subunit, suggesting that the different than recognizing a distinct motif to avoid auto-immunity at the
crRNA lengths are the result of different complex sizes, or vice CRISPR locus, the Csm and Cmr complexes instead check for
versa (Taylor et al., 2015). complementarity between the repeat-derived region of the
Despite the structural similarities, the Type III interference crRNA with the target and do not cleave if a full match is detected
complexes function quite distinctly from Cascade. The substrate (Marraffini and Sontheimer, 2010; Samai et al., 2015; Staals
specificity of Csm and Cmr complexes has only recently been et al., 2014; Tamulaitis et al., 2014). The specificity of Type III
clarified. Early in vivo genetic experiments suggested Csm tar- effector complexes for single-stranded targets might provide a
geted DNA, while in vitro studies of Cmr showed binding and rationale for their distinct mode of target authentication. For
cleavage activity against RNA only (Hale et al., 2009; Marraffini Type I and Type II effector complexes, which target dsDNA,
and Sontheimer, 2008), leading to a model wherein the two PAM recognition allows for an initial binding event to facilitate
subtypes had evolved distinct and complementary substrate subsequent unwinding of the target to probe for complemen-
preferences. This simple model was soon complicated by the tarity to the crRNA (Hochstrasser et al., 2014; Rollins et al.,
observation that Csm complexes in vitro also bind and cleave 2015; Sternberg et al., 2014; Szczelkun et al., 2014; Westra
RNA while exhibiting no activity against DNA (Staals et al., et al., 2012). Type III complexes can immediately probe a poten-
2014; Tamulaitis et al., 2014). Meanwhile, the in vivo DNA- tial single-stranded target for complementarity to their bound
targeting activity of III-A systems was shown to depend on crRNA without a need to license initial unwinding, and the
transcription at the target site, in contrast to the transcription-in- exposed nature of a single-stranded target facilitates the check
dependent targeting seen in Type I and Type II systems, and a for complementarity to the repeat-derived region of the guide.
similar activity was observed for a III-B system in vivo (Deng Type V Interference
et al., 2013; Goldberg et al., 2014). These observations were Type V systems have only recently been classified, but initial
reconciled by the discovery that the Csm complex from Staphy- work demonstrated that these systems are functional for inter-
lococcus epidermidis exhibits both RNA cleavage and DNA ference (Makarova et al., 2015; Zetsche et al., 2015a). The sys-
cleavage when directed against the non-template strand of tems appear most similar to Type II systems, possessing only
actively transcribed DNA (Samai et al., 2015). the acquisition machinery and a single additional protein (Makar-
DNA and RNA interference are carried out by distinct subunits ova et al., 2015; Schunder et al., 2013; Vestergaard et al., 2014).
of the Type III complexes. RNA interference is mediated by the Three subtypes of Class V systems have been identified with
backbone subunit Csm3 (or Cmr4 in III-B systems), which widely varying interference proteins (Shmakov et al., 2015).
cleaves the target every six nucleotides in the active site of a Type V-A, V-B, and V-C are characterized by the presence of
separate subunit by activating the ribose 20 OH for nucleophilic Cpf1, C2c1, and C2c3, respectively (Shmakov et al., 2015). All
attack in a manner typical of metal-independent RNases (Osawa three proteins are evolved from the same family of transposon-
et al., 2015; Staals et al., 2014; Tamulaitis et al., 2014; Taylor associated TpnB proteins as Cas9 and have a C-terminal
et al., 2015). Cas10 cleaves DNA exposed by a transcription RuvC domain and arginine-rich bridge helix (Shmakov et al.,
bubble using a single catalytic site in its palm polymerase 2015). However, the proteins show little similarity to each other,
domain (Samai et al., 2015). The details of DNA targeting by and the phylogenetic grouping of the associated cas1 genes with
Cmr have not been independently confirmed, but the conserva- various branches of Type I and Type III cas1 genes suggests that
tion of Cas10 and evidence for transcription-dependent plasmid each of these subtypes originated from distinct recombination
clearing supports a similar mechanism (Deng et al., 2013; Makar- events between CRISPR systems and tpnB genes (Shmakov
ova et al., 2011b). et al., 2015).
The distinct behavior of Type III systems provides the host While some Type V-B systems have an identifiable tracrRNA
microbe with the ability to tolerate temperate phages (Goldberg necessary for activity, Type V-A and V-C systems lack both a
et al., 2014). While Type I and Type II systems target and degrade tracrRNA and Cas6 or Cas5-like endonuclease, making it un-
any protospacer-containing DNA, Type III systems ignore clear how the crRNA is processed (Makarova et al., 2015; Shma-
foreign DNA until transcription begins that poses a threat to the kov et al., 2015). The crRNA of Type V-A systems has a
cell. This has the advantage of allowing cells to acquire ad- conserved stem-loop and can be processed to a functional
vantageous genes contained in prophages, such as antibiotic form when transcribed in E. coli in the presence of Cpf1 (Zetsche
resistance genes, and causing cell suicide in the event that a et al., 2015a). Whether Cpf1 is also required for processing and
lysogenic phage becomes lytic and begins transcribing genes the potential involvement host factors remains unknown. The
with matching spacers (Goldberg et al., 2014). However, the Cpf1 from Francisella novicida can successfully interfere with
strand-specific nature of both the RNA targeting and transcrip- transformed plasmids and recognizes a 50 -TTN-30 PAM at the
tion-dependent DNA targeting imposes an additional restriction 50 end of the protospacer sequence, similar to the PAM location
on the integration step of acquisition, as only one direction of of Type I systems and in contrast to the 30 PAM observed in
integration will yield productive interference. The means by Type II systems. The enzyme makes a double-strand break, re-
which this apparent limitation is overcome are unclear. Type III sulting in five-nucleotide 50 overhangs distal to the PAM site
systems are also frequently found coexisting with Type I sys- (Figure 4B). Mutation of catalytic residues in the RuvC active
tems, in which case their distinct target specificity might allow site prevents cleavage of either strand (Zetsche et al., 2015a).

The authors propose that Cpf1 might act as a dimer, with each is another challenge that must be addressed to enable clinical
monomer providing a RuvC active site but only one recognizing applications (D’Astolfo et al., 2015; Gori et al., 2015; Howes
the target. If this is the case, whether one or both monomers has and Schofield, 2015; Lin et al., 2014; Zuris et al., 2015). As the
a bound crRNA is unclear. Alternatively, an as-of-yet undiscov- field continues to advance rapidly, clinical trials may occur within
ered active site might be present in the protein, in which case a few years, with therapies possibly following within a decade.
its activity must be tightly coupled to that of RuvC to explain Engineering of crop plants with Cas9 is already underway; regu-
the phenotype observed for the RuvC mutant. A C2c1, which latory rulings have so far considered knockout plants not to be
also has only one identifiable nuclease domain, has also been genetically modified organisms, but the regulatory fate of other
shown to be active for cleavage in vivo and in vitro, where it rec- modifications is currently being considered (Servick, 2015).
ognizes a 50 -TTN-30 PAM and requires a tracrRNA (Shmakov
et al., 2015). Many mechanisms in these newly discovered sys-
Concluding Remarks
tems, both Type V and the essentially uncharacterized Type VI,
Despite the rapid progress of the field since the first demonstra-
remain unknown and open for future study.
tion of CRISPR immunity in 2007, many mechanistic questions
Interference Complexes as Genome Editing Tools
remain unanswered. Fundamental aspects of acquisition, such
Most tool development of Cas proteins has focused on exploit-
as how substrates for Cas1-Cas2-mediated integration are gener-
ing the programmable sequence-specific DNA recognition of
ated and the mechanism and extent of self- versus non-self-
interference complexes. Cas9 from S. pyogenes in particular
discrimination in different CRISPR subtypes, are still a mystery.
has proven enormously useful for genome engineering. The
While crRNA biogenesis and interference are reasonably well un-
ability to render Cas9 a two-component system by fusing the
derstood for certain model subtypes (Type I-E, Type II-A), the
crRNA and tracrRNA into a single guide RNA (sgRNA) has al-
sheer diversity of CRISPR systems means that many subtypes
lowed for its easy use for genome editing, transcriptional con-
with potentially distinct mechanisms remain unexplored. Type V
trol, RNA targeting, and imaging (for recent reviews, see Jiang
and VI systems have only begun to be analyzed, and Type IV sys-
and Marraffini, 2015; Sternberg and Doudna, 2015). Cas9 has
tems, bearing some familiar cas genes but no identifiable CRISPR
been used in various cell types and organisms ranging from
locus, have yet to be characterized experimentally and almost
mice and monkeys to primary human T cells and stem cells,
certainly rely on mechanisms distinct from those of traditional
as well as plants, bacteria, and fungi (Jiang and Marraffini,
CRISPR systems (Makarova and Koonin, 2015).
2015; Sternberg and Doudna, 2015). Recent work has focused
Other aspects of CRISPR-Cas systems lie beyond the scope
on developing various chemical- and light-inducible Cas9 con-
of this Review. We have not discussed the non-immune func-
structs to allow for greater spatiotemporal control and on em-
tions of CRISPR-Cas systems, some of which appear to have
ploying Cas9 orthologs with different PAM sequences and
evolved to serve regulatory rather than defense roles (for re-
smaller sizes, allowing for easier packaging in adeno-associ-
views, see Westra et al., 2014, and Ratner et al., 2015). Phage
ated virus vectors (Davis et al., 2015; Nihongaki et al., 2015;
evasion of CRISPR immunity is another active area of research,
Polstein and Gersbach, 2015; Ran et al., 2015; Zetsche et al.,
with identified mechanisms including DNA modification, special-
2015b).
ized anti-CRISPR proteins, and mutational escape (Bondy-Den-
Other interference complexes have already been used or
omy et al., 2013; Bondy-Denomy et al., 2015; Bryson et al., 2015;
have the potential to be useful for genome manipulation as
Deveau et al., 2008; Paez-Espino et al., 2015; Pawluk et al.,
well. Although the multi-subunit composition of Cascade
2014). The context-dependent regulation of CRISPR-Cas sys-
makes it less tractable for genome engineering compared to
tems in response to phage infection and stress signals has
Cas9, its large size and stable binding has been used for tran-
also been explored but requires further study (Bondy-Denomy
scriptional silencing in E. coli (Rath et al., 2015). No published
and Davidson, 2014; Garrett et al., 2015; Kenchappa et al.,
work has shown the application of Csm or Cmr complexes,
2013; Patterson et al., 2015; Pul et al., 2010). The rapid develop-
but either could likely be used for various RNA modulation ap-
ment of technology derived from CRISPR-Cas systems, most
plications in cells. Two Cpf1 homologs, out of 16 that were
notably Cas9 but also Cas6f/Csy4, Cascade, and Cpf1, has fu-
tested, have been shown to facilitate genome editing in human
eled intense interest in the field. The arms race between bacteria
cells (Zetsche et al., 2015a). The alternate PAM specificity of
and bacteriophage has generated powerful molecular biology
Cpf1 may prove useful for targeting sites without an appro-
tools, from restriction enzymes that enabled recombinant DNA
priate PAM for Cas9, and the staggered cuts might prove to
technology to Cas9, which started the ‘‘CRISPR revolution’’ in
favor distinct pathways of DNA repair. However, a thorough
modern genome engineering. CRISPR systems haven proven
investigation of the efficiency and off-target editing of Cpf1
to be both fascinating and enormously useful. Further study of
will be needed to determine if this protein will see significant
bacterial immune systems, both CRISPR systems and those
use alongside Cas9.
yet undiscovered, promises to yield further unforeseen discov-
While Cas9 has already seen extensive use in the research
eries and exciting new technologies.
setting, challenges remain for its application in the clinic. While
making programmed cuts has become largely trivial, biasing
ACKNOWLEDGMENTS
DNA repair toward homology-directed repair rather than non-ho-
mologous end joining remains a challenge (Chu et al., 2015; Mar- This work was funded by US National Science Foundation grant number
uyama et al., 2015). Delivery of Cas9, either as an RNP or on a 1244557 to J.A.D. A.V.W. and J.K.N. are NSF Graduate Research Fellows.
plasmid or viral vector, to particular tissues in whole organisms Megan Hochstrasser provided valuable input on the manuscript.

REFERENCES Chow, S.A., Vincent, K.A., Ellison, V., and Brown, P.O. (1992). Reversal of inte-
gration and DNA splicing mediated by integrase of human immunodeficiency
Anders, C., Niewoehner, O., Duerst, A., and Jinek, M. (2014). Structural basis virus. Science 255, 723–726.
of PAM-dependent target DNA recognition by the Cas9 endonuclease. Nature Chu, V.T., Weber, T., Wefers, B., Wurst, W., Sander, S., Rajewsky, K., and
513, 569–573. Kühn, R. (2015). Increasing the efficiency of homology-directed repair for
Arslan, Z., Wurm, R., Brener, O., Ellinger, P., Nagel-Steger, L., Oesterhelt, F., CRISPR-Cas9-induced precise gene editing in mammalian cells. Nat. Bio-
Schmitt, L., Willbold, D., Wagner, R., Gohlke, H., et al. (2013). Double-strand technol. 33, 543–548.
DNA end-binding and sliding of the toroidal CRISPR-associated protein
Chylinski, K., Makarova, K.S., Charpentier, E., and Koonin, E.V. (2014). Classi-
Csn2. Nucleic Acids Res. 41, 6347–6359.
fication and evolution of type II CRISPR-Cas systems. Nucleic Acids Res. 42,
Arslan, Z., Hermanns, V., Wurm, R., Wagner, R., and Pul, Ü. (2014). Detection 6091–6105.
and characterization of spacer integration intermediates in type I-E CRISPR-
Cong, L., Ran, F.A., Cox, D., Lin, S., Barretto, R., Habib, N., Hsu, P.D., Wu, X.,
Cas system. Nucleic Acids Res. 42, 7884–7893.
Jiang, W., Marraffini, L.A., and Zhang, F. (2013). Multiplex genome engineering
Barrangou, R., Fremaux, C., Deveau, H., Richards, M., Boyaval, P., Moineau, using CRISPR/Cas systems. Science 339, 819–823.
S., Romero, D.A., and Horvath, P. (2007). CRISPR provides acquired resis-
D’Astolfo, D.S., Pagliero, R.J., Pras, A., Karthaus, W.R., Clevers, H., Prasad,
tance against viruses in prokaryotes. Science 315, 1709–1712.
V., Lebbink, R.J., Rehmann, H., and Geijsen, N. (2015). Efficient intracellular
Blackwood, J.K., Rzechorzek, N.J., Bray, S.M., Maman, J.D., Pellegrini, L., delivery of native proteins. Cell 161, 674–690.
and Robinson, N.P. (2013). End-resection at DNA double-strand breaks in
Datsenko, K.A., Pougach, K., Tikhonov, A., Wanner, B.L., Severinov, K., and
the three domains of life. Biochem. Soc. Trans. 41, 314–320.
Semenova, E. (2012). Molecular memory of prior infections activates the
Blosser, T.R., Loeff, L., Westra, E.R., Vlot, M., Künne, T., Sobota, M., Dekker, CRISPR/Cas adaptive bacterial immunity system. Nat. Commun. 3, 945.
C., Brouns, S.J., and Joo, C. (2015). Two distinct DNA binding modes guide
dual roles of a CRISPR-Cas protein complex. Mol. Cell 58, 60–70. Davis, K.M., Pattanayak, V., Thompson, D.B., Zuris, J.A., and Liu, D.R. (2015).
Small molecule-triggered Cas9 protein with improved genome-editing speci-
Blundell, J.R., and Levy, S.F. (2014). Beyond genome sequencing: Lineage ficity. Nat. Chem. Biol. 11, 316–318.
tracking with barcodes to study the dynamics of evolution, infection, and
cancer. Genomics 104, 417–430. Deltcheva, E., Chylinski, K., Sharma, C.M., Gonzales, K., Chao, Y., Pirzada,
Z.A., Eckert, M.R., Vogel, J., and Charpentier, E. (2011). CRISPR RNA matura-
Bolotin, A., Quinquis, B., Sorokin, A., and Ehrlich, S.D. (2005). Clustered regu-
tion by trans-encoded small RNA and host factor RNase III. Nature 471,
larly interspaced short palindrome repeats (CRISPRs) have spacers of extra-
602–607.
chromosomal origin. Microbiology 151, 2551–2561.
Deng, L., Garrett, R.A., Shah, S.A., Peng, X., and She, Q. (2013). A novel inter-
Bondy-Denomy, J., and Davidson, A.R. (2014). To acquire or resist: the com-
ference mechanism by a type IIIB CRISPR-Cmr module in Sulfolobus. Mol.
plex biological effects of CRISPR-Cas systems. Trends Microbiol. 22,
Microbiol. 87, 1088–1099.
218–225.
Deveau, H., Barrangou, R., Garneau, J.E., Labonté, J., Fremaux, C., Boyaval,
Bondy-Denomy, J., Pawluk, A., Maxwell, K.L., and Davidson, A.R. (2013).
P., Romero, D.A., Horvath, P., and Moineau, S. (2008). Phage response to
Bacteriophage genes that inactivate the CRISPR/Cas bacterial immune sys-
CRISPR-encoded resistance in Streptococcus thermophilus. J. Bacteriol.
tem. Nature 493, 429–432.
190, 1390–1400.
Bondy-Denomy, J., Garcia, B., Strum, S., Du, M., Rollins, M.F., Hidalgo-
Dı́ez-Villaseñor, C., Guzmán, N.M., Almendros, C., Garcı́a-Martı́nez, J., and
Reyes, Y., Wiedenheft, B., Maxwell, K.L., and Davidson, A.R. (2015). Multiple
Mojica, F.J. (2013). CRISPR-spacer integration reporter plasmids reveal
mechanisms for CRISPR-Cas inhibition by anti-CRISPR proteins. Nature 526,
distinct genuine acquisition specificities among CRISPR-Cas I-E variants of
136–139.
Escherichia coli. RNA Biol. 10, 792–802.
Borchardt, E.K., Vandoros, L.A., Huang, M., Lackey, P.E., Marzluff, W.F., and
Asokan, A. (2015). Controlling mRNA stability and translation with the CRISPR Dillingham, M.S., and Kowalczykowski, S.C. (2008). RecBCD enzyme and the
endoribonuclease Csy4. RNA 21, 1921–1930. repair of double-stranded DNA breaks. Microbiol. Mol. Biol. Rev. 72, 642–671.
Brendel, J., Stoll, B., Lange, S.J., Sharma, K., Lenz, C., Stachler, A.E., Maier, Du, P., Miao, C., Lou, Q., Wang, Z., and Lou, C. (2015). Engineering transla-
L.K., Richter, H., Nickel, L., Schmitz, R.A., et al. (2014). A complex of Cas tional activators with CRISPR-Cas system. ACS Synth. Biol. http://dx.doi.
proteins 5, 6, and 7 is required for the biogenesis and stability of clustered org/10.1021/acssynbio.5b00130.
regularly interspaced short palindromic repeats (crispr)-derived rnas (crrnas) Dugar, G., Herbig, A., Förstner, K.U., Heidrich, N., Reinhardt, R., Nieselt, K.,
in Haloferax volcanii. J. Biol. Chem. 289, 7164–7177. and Sharma, C.M. (2013). High-resolution transcriptome maps reveal strain-
Briner, A.E., Donohoue, P.D., Gomaa, A.A., Selle, K., Slorach, E.M., Nye, C.H., specific regulatory features of multiple Campylobacter jejuni isolates. PLoS
Haurwitz, R.E., Beisel, C.L., May, A.P., and Barrangou, R. (2014). Guide RNA Genet. 9, e1003495.
functional modules direct Cas9 activity and orthogonality. Mol. Cell 56, Ellinger, P., Arslan, Z., Wurm, R., Tschapek, B., MacKenzie, C., Pfeffer, K.,
333–339. Panjikar, S., Wagner, R., Schmitt, L., Gohlke, H., et al. (2012). The crystal struc-
Brouns, S.J., Jore, M.M., Lundgren, M., Westra, E.R., Slijkhuis, R.J., Snijders, ture of the CRISPR-associated protein Csn2 from Streptococcus agalactiae.
A.P., Dickman, M.J., Makarova, K.S., Koonin, E.V., and van der Oost, J. (2008). J. Struct. Biol. 178, 350–362.
Small CRISPR RNAs guide antiviral defense in prokaryotes. Science 321, Engelman, A., Mizuuchi, K., and Craigie, R. (1991). HIV-1 DNA integration:
960–964. mechanism of viral DNA cleavage and DNA strand transfer. Cell 67, 1211–
Bryson, A.L., Hwang, Y., Sherrill-Mix, S., Wu, G.D., Lewis, J.D., Black, L., 1221.
Clark, T.A., and Bushman, F.D. (2015). Covalent Modification of Bacterio- Garrett, R.A., Shah, S.A., Erdmann, S., Liu, G., Mousaei, M., León-Sobrino, C.,
phage T4 DNA Inhibits CRISPR-Cas9. MBio 6, e00648. Peng, W., Gudbergsdottir, S., Deng, L., Vestergaard, G., et al. (2015). CRISPR-
Carte, J., Wang, R., Li, H., Terns, R.M., and Terns, M.P. (2008). Cas6 is an en- Cas Adaptive Immune Systems of the Sulfolobales: Unravelling Their
doribonuclease that generates guide RNAs for invader defense in prokaryotes. Complexity and Diversity. Life (Basel) 5, 783–817.
Genes Dev. 22, 3489–3496. Garside, E.L., Schellenberg, M.J., Gesner, E.M., Bonanno, J.B., Sauder, J.M.,
Charpentier, E., Richter, H., van der Oost, J., and White, M.F. (2015). Biogen- Burley, S.K., Almo, S.C., Mehta, G., and MacMillan, A.M. (2012). Cas5d
esis pathways of RNA guides in archaeal and bacterial CRISPR-Cas adaptive processes pre-crRNA and is a member of a larger family of CRISPR RNA en-
immunity. FEMS Microbiol. Rev. 39, 428–441. donucleases. RNA 18, 2020–2028.

Gasiunas, G., Barrangou, R., Horvath, P., and Siksnys, V. (2012). Cas9-crRNA Jiang, W., Bikard, D., Cox, D., Zhang, F., and Marraffini, L.A. (2013). RNA-
ribonucleoprotein complex mediates specific DNA cleavage for adaptive im- guided editing of bacterial genomes using CRISPR-Cas systems. Nat. Bio-
munity in bacteria. Proc. Natl. Acad. Sci. USA 109, E2579–E2586. technol. 31, 233–239.
Gesner, E.M., Schellenberg, M.J., Garside, E.L., George, M.M., and Macmil- Jiang, F., Zhou, K., Ma, L., Gressel, S., and Doudna, J.A. (2015). STRUCTURAL
lan, A.M. (2011). Recognition and maturation of effector RNAs in a CRISPR BIOLOGY. A Cas9-guide RNA complex preorganized for target DNA recogni-
interference pathway. Nat. Struct. Mol. Biol. 18, 688–692. tion. Science 348, 1477–1481.
Goldberg, G.W., Jiang, W., Bikard, D., and Marraffini, L.A. (2014). Conditional Jinek, M., Chylinski, K., Fonfara, I., Hauer, M., Doudna, J.A., and Charpentier,
tolerance of temperate phages via transcription-dependent CRISPR-Cas tar- E. (2012). A programmable dual-RNA-guided DNA endonuclease in adaptive
geting. Nature 514, 633–637. bacterial immunity. Science 337, 816–821.
Gong, B., Shin, M., Sun, J., Jung, C.-H., Bolt, E.L., van der Oost, J., and Kim, Jinek, M., Jiang, F., Taylor, D.W., Sternberg, S.H., Kaya, E., Ma, E., Anders, C.,
J.-S. (2014). Molecular insights into DNA interference by CRISPR-associated Hauer, M., Zhou, K., Lin, S., et al. (2014). Structures of Cas9 endonucleases
nuclease-helicase Cas3. Proc. Natl. Acad. Sci. USA 111, 16359–16364. reveal RNA-mediated conformational activation. Science 343, 1247997.
Gori, J.L., Hsu, P.D., Maeder, M.L., Shen, S., Welstead, G.G., and Bumcrot, D. Jore, M.M., Lundgren, M., van Duijn, E., Bultema, J.B., Westra, E.R., Wagh-
(2015). Delivery and Specificity of CRISPR-Cas9 Genome Editing Technolo- mare, S.P., Wiedenheft, B., Pul, U., Wurm, R., Wagner, R., et al. (2011). Struc-
gies for Human Gene Therapy. Hum. Gene Ther. 26, 443–451. tural basis for CRISPR RNA-guided DNA recognition by Cascade. Nat. Struct.
Hale, C., Kleppe, K., Terns, R.M., and Terns, M.P. (2008). Prokaryotic silencing Mol. Biol. 18, 529–536.
(psi)RNAs in Pyrococcus furiosus. RNA 14, 2572–2579. Juranek, S., Eban, T., Altuvia, Y., Brown, M., Morozov, P., Tuschl, T., and Mar-
Hale, C.R., Zhao, P., Olson, S., Duff, M.O., Graveley, B.R., Wells, L., Terns, galit, H. (2012). A genome-wide view of the expression and processing pat-
R.M., and Terns, M.P. (2009). RNA-guided RNA cleavage by a CRISPR terns of Thermus thermophilus HB8 CRISPR RNAs. RNA 18, 783–794.
RNA-Cas protein complex. Cell 139, 945–956. Kenchappa, C.S., Heidarsson, P.O., Kragelund, B.B., Garrett, R.A., and Poul-
Haurwitz, R.E., Jinek, M., Wiedenheft, B., Zhou, K., and Doudna, J.A. (2010). sen, F.M. (2013). Solution properties of the archaeal CRISPR DNA repeat-
Sequence- and structure-specific RNA processing by a CRISPR endonu- binding homeodomain protein Cbp2. Nucleic Acids Res. 41, 3424–3435.
clease. Science 329, 1355–1358. Koo, Y., Jung, D.K., and Bae, E. (2012). Crystal structure of Streptococcus
Heler, R., Samai, P., Modell, J.W., Weiner, C., Goldberg, G.W., Bikard, D., and pyogenes Csn2 reveals calcium-dependent conformational changes in its ter-
Marraffini, L.A. (2015). Cas9 specifies functional viral targets during CRISPR- tiary and quaternary structure. PLoS ONE 7, e33401.
Cas adaptation. Nature 519, 199–202. Kuhn, C.-D., and Joshua-Tor, L. (2013). Eukaryotic Argonautes come into
Hochstrasser, M.L., and Doudna, J.A. (2015). Cutting it close: CRISPR-asso- focus. Trends Biochem. Sci. 38, 263–271.
ciated endoribonuclease structure and function. Trends Biochem. Sci. 40, Kunin, V., Sorek, R., and Hugenholtz, P. (2007). Evolutionary conservation of
58–66. sequence and secondary structures in CRISPR repeats. Genome Biol. 8, R61.
Hochstrasser, M.L., Taylor, D.W., Bhat, P., Guegler, C.K., Sternberg, S.H., No- Labrie, S.J., Samson, J.E., and Moineau, S. (2010). Bacteriophage resistance
gales, E., and Doudna, J.A. (2014). CasA mediates Cas3-catalyzed target mechanisms. Nat. Rev. Microbiol. 8, 317–327.
degradation during CRISPR RNA-guided interference. Proc. Natl. Acad. Sci. Lee, K.-H., Lee, S.-G., Eun Lee, K., Jeon, H., Robinson, H., and Oh, B.-H.
USA 111, 6618–6623. (2012). Identification, structural, and biochemical characterization of a group
Hooton, S.P., and Connerton, I.F. (2014). Campylobacter jejuni acquire new of large Csn2 proteins involved in CRISPR-mediated bacterial immunity. Pro-
host-derived CRISPR spacers when in association with bacteriophages teins 80, 2573–2582.
harboring a CRISPR-like Cas4 protein. Front Microbiol 5, 744. Lee, H.Y., Haurwitz, R.E., Apffel, A., Zhou, K., Smart, B., Wenger, C.D., Lader-
Howes, R., and Schofield, C. (2015). Genome engineering using Adeno-Asso- man, S., Bruhn, L., and Doudna, J.A. (2013). RNA-protein analysis using a
ciated Virus (AAV). Methods Mol. Biol. 1239, 75–103. conditional CRISPR nuclease. Proc. Natl. Acad. Sci. USA 110, 5416–5421.
Huo, Y., Nam, K.H., Ding, F., Lee, H., Wu, L., Xiao, Y., Farchione, M.D., Jr., Lemak, S., Beloglazova, N., Nocek, B., Skarina, T., Flick, R., Brown, G., Po-
Zhou, S., Rajashankar, K., Kurinov, I., et al. (2014). Structures of CRISPR povic, A., Joachimiak, A., Savchenko, A., and Yakunin, A.F. (2013). Toroidal
Cas3 offer mechanistic insights into Cascade-activated DNA unwinding and structure and DNA cleavage by the CRISPR-associated [4Fe-4S] cluster con-
degradation. Nat. Struct. Mol. Biol. 21, 771–777. taining Cas4 nuclease SSO0001 from Sulfolobus solfataricus. J. Am. Chem.
Hynes, A.P., Villion, M., and Moineau, S. (2014). Adaptation in bacterial Soc. 135, 17476–17487.
CRISPR-Cas immunity can be driven by defective phages. Nat. Commun. 5, Levy, A., Goren, M.G., Yosef, I., Auster, O., Manor, M., Amitai, G., Edgar, R.,
4399. Qimron, U., and Sorek, R. (2015). CRISPR adaptation biases explain prefer-
Ishino, Y., Shinagawa, H., Makino, K., Amemura, M., and Nakata, A. (1987). ence for acquisition of foreign DNA. Nature 520, 505–510.
Nucleotide sequence of the iap gene, responsible for alkaline phosphatase Li, M., Wang, R., Zhao, D., and Xiang, H. (2014). Adaptation of the Haloarcula
isozyme conversion in Escherichia coli, and identification of the gene product. hispanica CRISPR-Cas system to a purified virus strictly requires a priming
J. Bacteriol. 169, 5429–5433. process. Nucleic Acids Res. 42, 2483–2492.
Jackson, R.N., and Wiedenheft, B. (2015). A Conserved Structural Chassis for Lin, S., Staahl, B.T., Alla, R.K., and Doudna, J.A. (2014). Enhanced homology-
Mounting Versatile CRISPR RNA-Guided Immune Responses. Mol. Cell 58, directed human genome engineering by controlled timing of CRISPR/Cas9 de-
722–728. livery. eLife 3, e04766.
Jackson, R.N., Golden, S.M., van Erp, P.B., Carter, J., Westra, E.R., Brouns, Lintner, N.G., Kerou, M., Brumfield, S.K., Graham, S., Liu, H., Naismith, J.H.,
S.J., van der Oost, J., Terwilliger, T.C., Read, R.J., and Wiedenheft, B. Sdano, M., Peng, N., She, Q., Copié, V., et al. (2011). Structural and functional
(2014). Structural biology. Crystal structure of the CRISPR RNA-guided sur- characterization of an archaeal clustered regularly interspaced short palin-
veillance complex from Escherichia coli. Science 345, 1473–1479. dromic repeat (CRISPR)-associated complex for antiviral defense (CASCADE).
Jansen, R., Embden, J.D., Gaastra, W., and Schouls, L.M. (2002). Identifica- J. Biol. Chem. 286, 21643–21656.
tion of genes that are associated with DNA repeats in prokaryotes. Mol. Micro- Ma, E., Harrington, L.B., O’Connell, M.R., Zhou, K., and Doudna, J.A. (2015).
biol. 43, 1565–1575. Single-stranded DNA cleavage by divergent CRISPR-Cas9 enzymes. Mol.
Jiang, W., and Marraffini, L.A. (2015). CRISPR-Cas: New Tools for Genetic Cell 60, 398–407.
Manipulations from Bacterial Immunity Systems. Annu. Rev. Microbiol. 69, Makarova, K.S., and Koonin, E.V. (2015). Annotation and Classification of
209–228. CRISPR-Cas Systems. Methods Mol. Biol. 1311, 47–75.

Makarova, K.S., Aravind, L., Wolf, Y.I., and Koonin, E.V. (2011a). Unification of Nuñez, J.K., Harrington, L.B., Kranzusch, P.J., Engelman, A.N., and Doudna,
Cas protein families and a simple scenario for the origin and evolution of J.A. (2015b). Foreign DNA capture during CRISPR-Cas adaptive immunity.
CRISPR-Cas systems. Biol. Direct 6, 38. Nature 527, 535–538. http://dx.doi.org/10.1038/nature15760.
Makarova, K.S., Haft, D.H., Barrangou, R., Brouns, S.J., Charpentier, E., Hor- Osawa, T., Inanaga, H., Sato, C., and Numata, T. (2015). Crystal structure of
vath, P., Moineau, S., Mojica, F.J., Wolf, Y.I., Yakunin, A.F., et al. (2011b). Evo- the CRISPR-Cas RNA silencing Cmr complex bound to a target analog. Mol.
lution and classification of the CRISPR-Cas systems. Nat. Rev. Microbiol. 9, Cell 58, 418–430.
467–477.
Paez-Espino, D., Morovic, W., Sun, C.L., Thomas, B.C., Ueda, K., Stahl, B.,
Makarova, K.S., Wolf, Y.I., and Koonin, E.V. (2013). The basic building blocks Barrangou, R., and Banfield, J.F. (2013). Strong bias in the bacterial CRISPR
and evolution of CRISPR-CAS systems. Biochem. Soc. Trans. 41, 1392–1400. elements that confer immunity to phage. Nat. Commun. 4, 1430–1437.
Makarova, K.S., Wolf, Y.I., Alkhnbashi, O.S., Costa, F., Shah, S.A., Saunders, Paez-Espino, D., Sharon, I., Morovic, W., Stahl, B., Thomas, B.C., Barrangou,
S.J., Barrangou, R., Brouns, S.J.J., Charpentier, E., Haft, D.H., et al. (2015). An R., and Banfield, J.F. (2015). CRISPR immunity drives rapid phage genome
updated evolutionary classification of CRISPR-Cas systems. Nat. Rev. Micro- evolution in Streptococcus thermophilus. MBio 6, e00262–e00215.
biol. 13, 722–736.
Patterson, A.G., Chang, J.T., Taylor, C., and Fineran, P.C. (2015). Regulation of
Marraffini, L.A., and Sontheimer, E.J. (2008). CRISPR interference limits hori- the Type I-F CRISPR-Cas system by CRP-cAMP and GalM controls spacer
zontal gene transfer in staphylococci by targeting DNA. Science 322, 1843– acquisition and interference. Nucleic Acids Res. 43, 6038–6048.
1845.
Pawluk, A., Bondy-Denomy, J., Cheung, V.H.W., Maxwell, K.L., and Davidson,
Marraffini, L.A., and Sontheimer, E.J. (2010). Self versus non-self discrimina- A.R. (2014). A new group of phage anti-CRISPR genes inhibits the type I-E
tion during CRISPR RNA-directed immunity. Nature 463, 568–571. CRISPR-Cas system of Pseudomonas aeruginosa. mBio. 5, e00896-14.
Maruyama, T., Dougan, S.K., Truttmann, M.C., Bilate, A.M., Ingram, J.R., and
Plagens, A., Tjaden, B., Hagemann, A., Randau, L., and Hensel, R. (2012).
Ploegh, H.L. (2015). Increasing the efficiency of precise genome editing with
Characterization of the CRISPR/Cas subtype I-A system of the hyperthermo-
CRISPR-Cas9 by inhibition of nonhomologous end joining. Nat. Biotechnol.
philic crenarchaeon Thermoproteus tenax. J. Bacteriol. 194, 2491–2500.
33, 538–542.
Plagens, A., Tripp, V., Daume, M., Sharma, K., Klingl, A., Hrle, A., Conti, E., Ur-
Mizuuchi, K., and Adzuma, K. (1991). Inversion of the phosphate chirality at the
laub, H., and Randau, L. (2014). In vitro assembly and activity of an archaeal
target site of Mu DNA strand transfer: evidence for a one-step transesterifica-
CRISPR-Cas type I-A Cascade interference complex. Nucleic Acids Res. 42,
tion mechanism. Cell 66, 129–140.
5125–5138.
Mojica, F.J., Dı́ez-Villaseñor, C., Garcı́a-Martı́nez, J., and Soria, E. (2005).
Plagens, A., Richter, H., Charpentier, E., and Randau, L. (2015). DNA and RNA
Intervening sequences of regularly spaced prokaryotic repeats derive from
interference mechanisms by CRISPR-Cas surveillance complexes. FEMS
foreign genetic elements. J. Mol. Evol. 60, 174–182.
Microbiol. Rev. 39, 442–463.
Mojica, F.J., Dı́ez-Villaseñor, C., Garcı́a-Martı́nez, J., and Almendros, C.
(2009). Short motif sequences determine the targets of the prokaryotic Polstein, L.R., and Gersbach, C.A. (2015). A light-inducible CRISPR-Cas9 sys-
CRISPR defence system. Microbiology 155, 733–740. tem for control of endogenous gene activation. Nat. Chem. Biol. 11, 198–200.
Mulepati, S., and Bailey, S. (2011). Structural and biochemical analysis of Pourcel, C., Salvignol, G., and Vergnaud, G. (2005). CRISPR elements in Yer-
nuclease domain of clustered regularly interspaced short palindromic repeat sinia pestis acquire new repeats by preferential uptake of bacteriophage DNA,
(CRISPR)-associated protein 3 (Cas3). J. Biol. Chem. 286, 31896–31903. and provide additional tools for evolutionary studies. Microbiology 151,
653–663.
Mulepati, S., and Bailey, S. (2013). In vitro reconstitution of an Escherichia coli
RNA-guided immune system reveals unidirectional, ATP-dependent degrada- Pul, U., Wurm, R., Arslan, Z., Geissen, R., Hofmann, N., and Wagner, R. (2010).
tion of DNA target. J. Biol. Chem. 288, 22184–22192. Identification and characterization of E. coli CRISPR-cas promoters and their
silencing by H-NS. Mol. Microbiol. 75, 1495–1512.
Mulepati, S., Héroux, A., and Bailey, S. (2014). Structural biology. Crystal
structure of a CRISPR RNA-guided surveillance complex bound to a ssDNA Ran, F.A., Cong, L., Yan, W.X., Scott, D.A., Gootenberg, J.S., Kriz, A.J., Zet-
target. Science 345, 1479–1484. sche, B., Shalem, O., Wu, X., Makarova, K.S., et al. (2015). In vivo genome ed-
iting using Staphylococcus aureus Cas9. Nature 520, 186–191.
Nam, K.H., Haitjema, C., Liu, X., Ding, F., Wang, H., DeLisa, M.P., and Ke, A.
(2012b). Cas5d protein processes pre-crRNA and assembles into a cascade- Rath, D., Amlinger, L., Hoekzema, M., Devulapally, P.R., and Lundgren, M.
like interference complex in subtype I-C/Dvulg CRISPR-Cas system. Structure (2015). Efficient programmable gene silencing by Cascade. Nucleic Acids
20, 1574–1584. Res. 43, 237–246.
Nihongaki, Y., Kawano, F., Nakajima, T., and Sato, M. (2015). Photoactivatable Ratner, H.K., Sampson, T.R., and Weiss, D.S. (2015). I can see CRISPR now,
CRISPR-Cas9 for optogenetic genome editing. Nat. Biotechnol. 33, 755–760. even when phage are gone: a view on alternative CRISPR-Cas functions from
Nishimasu, H., Ran, F.A., Hsu, P.D., Konermann, S., Shehata, S.I., Dohmae, the prokaryotic envelope. Curr. Opin. Infect. Dis. 28, 267–274.
N., Ishitani, R., Zhang, F., and Nureki, O. (2014). Crystal structure of Cas9 in Redding, S., Sternberg, S.H., Marshall, M., Gibb, B., Bhat, P., Guegler, C.K.,
complex with guide RNA and target DNA. Cell 156, 935–949. Wiedenheft, B., Doudna, J.A., and Greene, E.C. (2015). Surveillance and pro-
Nishimasu, H., Cong, L., Yan, W.X., Ran, F.A., Zetsche, B., Li, Y., Kurabayashi, cessing of foreign DNA by the Escherichia coli CRISPR-Cas system. Cell 163,
A., Ishitani, R., Zhang, F., and Nureki, O. (2015). Crystal Structure of Staphylo- 854–865.
coccus aureus Cas9. Cell 162, 1113–1126. Richter, C., and Fineran, P.C. (2013). The subtype I-F CRISPR-Cas system in-
Nissim, L., Perli, S.D., Fridkin, A., Perez-Pinera, P., and Lu, T.K. (2014). Multi- fluences pathogenicity island retention in Pectobacterium atrosepticum via
plexed and programmable regulation of gene networks with an integrated RNA crRNA generation and Csy complex formation. Biochem. Soc. Trans. 41,
and CRISPR/Cas toolkit in human cells. Mol. Cell 54, 698–710. 1468–1474.
Nuñez, J.K., Kranzusch, P.J., Noeske, J., Wright, A.V., Davies, C.W., and Richter, C., Gristwood, T., Clulow, J.S., and Fineran, P.C. (2012). In vivo protein
Doudna, J.A. (2014). Cas1-Cas2 complex formation mediates spacer acquisi- interactions and complex formation in the Pectobacterium atrosepticum sub-
tion during CRISPR-Cas adaptive immunity. Nat. Struct. Mol. Biol. 21, type I-F CRISPR/Cas System. PLoS ONE 7, e49549.
528–534. Richter, C., Dy, R.L., McKenzie, R.E., Watson, B.N., Taylor, C., Chang, J.T.,
Nuñez, J.K., Lee, A.S., Engelman, A., and Doudna, J.A. (2015a). Integrase- McNeil, M.B., Staals, R.H., and Fineran, P.C. (2014). Priming in the Type I-F
mediated spacer acquisition during CRISPR-Cas adaptive immunity. Nature CRISPR-Cas system triggers strand-independent spacer acquisition, bi-di-
519, 193–198. rectionally from the primed protospacer. Nucleic Acids Res. 42, 8516–8526.

Rollie, C., Schneider, S., Brinkmann, A.S., Bolt, E.L., and White, M.F. (2015). Sternberg, S.H., Redding, S., Jinek, M., Greene, E.C., and Doudna, J.A. (2014).
Intrinsic sequence specificity of the Cas1 integrase directs new spacer acqui- DNA interrogation by the CRISPR RNA-guided endonuclease Cas9. Nature
sition. eLife 4, 4. 507, 62–67.
Rollins, M.F., Schuman, J.T., Paulus, K., Bukhari, H.S., and Wiedenheft, B. Sternberg, S.H., LaFrance, B., Kaplan, M., and Doudna, J.A. (2015). Confor-
(2015). Mechanism of foreign DNA recognition by a CRISPR RNA-guided sur- mational control of DNA target cleavage by CRISPR-Cas9. Nature 527,
veillance complex from Pseudomonas aeruginosa. Nucleic Acids Res. 43, 110–113.
2216–2222. Swarts, D.C., Mosterd, C., van Passel, M.W., and Brouns, S.J. (2012). CRISPR
Rutkauskas, M., Sinkunas, T., Songailiene, I., Tikhomirova, M.S., Siksnys, V., interference directs strand specific spacer acquisition. PLoS ONE 7, e35888.
and Seidel, R. (2015). Directional R-Loop Formation by the CRISPR-Cas Sur-
Szczelkun, M.D., Tikhomirova, M.S., Sinkunas, T., Gasiunas, G., Karvelis, T.,
veillance Complex Cascade Provides Efficient Off-Target Site Rejection. Cell
Pschera, P., Siksnys, V., and Seidel, R. (2014). Direct observation of R-loop
Rep. 10, 1534–1543.
formation by single RNA-guided Cas9 and Cascade effector complexes.
Salvail-Lacoste, A., Di Tomasso, G., Piette, B.L., and Legault, P. (2013). Affinity Proc. Natl. Acad. Sci. USA 111, 9798–9803.
purification of T7 RNA transcripts with homogeneous ends using ARiBo and Nwokeoji,
Tamulaitis, G., Kazlauskiene, M., Manakova, E., Venclovas, C.,
CRISPR tags. RNA 19, 1003–1014.
A.O., Dickman, M.J., Horvath, P., and Siksnys, V. (2014). Programmable
Samai, P., Pyenson, N., Jiang, W., Goldberg, G.W., Hatoum-Aslan, A., and RNA shredding by the type III-A CRISPR-Cas system of Streptococcus ther-
Marraffini, L.A. (2015). Co-transcriptional DNA and RNA Cleavage during mophilus. Mol. Cell 56, 506–517.
Type III CRISPR-Cas Immunity. Cell 161, 1164–1174.
Taylor, D.W., Zhu, Y., Staals, R.H.J., Kornfeld, J.E., Shinkai, A., van der Oost,
Sashital, D.G., Jinek, M., and Doudna, J.A. (2011). An RNA-induced conforma- J., Nogales, E., and Doudna, J.A. (2015). Structural biology. Structures of the
tional change required for CRISPR RNA cleavage by the endoribonuclease CRISPR-Cmr complex reveal mode of RNA target positioning. Science 348,
Cse3. Nat. Struct. Mol. Biol. 18, 680–687. 581–585.
Sashital, D.G., Wiedenheft, B., and Doudna, J.A. (2012). Mechanism of foreign Tsai, S.Q., Wyvekens, N., Khayter, C., Foden, J.A., Thapar, V., Reyon, D.,
DNA selection in a bacterial adaptive immune system. Mol. Cell 46, 606–615. Goodwin, M.J., Aryee, M.J., and Joung, J.K. (2014). Dimeric CRISPR RNA-
Savitskaya, E., Semenova, E., Dedkov, V., Metlitskaya, A., and Severinov, K. guided FokI nucleases for highly specific genome editing. Nat. Biotechnol.
(2013). High-throughput analysis of type I-E CRISPR/Cas spacer acquisition 32, 569–576.
in E. coli. RNA Biol. 10, 716–725. Tsui, T.K.M., and Li, H. (2015). Structure Principles of CRISPR-Cas Surveil-
Schunder, E., Rydzewski, K., Grunow, R., and Heuner, K. (2013). First indica- lance and Effector Complexes. Annu. Rev. Biophys. 44, 229–255.
tion for a functional CRISPR/Cas system in Francisella tularensis. Int. J. Med.
van der Oost, J., Jore, M.M., Westra, E.R., Lundgren, M., and Brouns, S.J.
Microbiol. 303, 51–60.
(2009). CRISPR-based adaptive and heritable immunity in prokaryotes. Trends
Semenova, E., Jore, M.M., Datsenko, K.A., Semenova, A., Westra, E.R., Biochem. Sci. 34, 401–407.
Wanner, B., van der Oost, J., Brouns, S.J., and Severinov, K. (2011). Interfer-
van der Oost, J., Westra, E.R., Jackson, R.N., and Wiedenheft, B. (2014). Un-
ence by clustered regularly interspaced short palindromic repeat (CRISPR)
ravelling the structural and mechanistic basis of CRISPR-Cas systems. Nat.
RNA is governed by a seed sequence. Proc. Natl. Acad. Sci. USA 108,
Rev. Microbiol. 12, 479–492.
10098–10103.
van Erp, P.B.G., Jackson, R.N., Carter, J., Golden, S.M., Bailey, S., and Wie-
Servick, K. (2015). SCIENCE POLICY. U.S. to review agricultural biotech reg-
denheft, B. (2015). Mechanism of CRISPR-RNA guided recognition of DNA tar-
ulations. Science 349, 131.
gets in Escherichia coli. Nucleic Acids Res. 43, 8381–8391.
Shao, Y., and Li, H. (2013). Recognition and cleavage of a nonstructured
Vestergaard, G., Garrett, R.A., and Shah, S.A. (2014). CRISPR adaptive im-
CRISPR RNA by its processing endoribonuclease Cas6. Structure 21,
mune systems of Archaea. RNA Biol. 11, 156–167.
385–393.
Wang, J., Li, J., Zhao, H., Sheng, G., Wang, M., Yin, M., and Wang, Y. (2015).
Shmakov, S., Savitskaya, E., Semenova, E., Logacheva, M.D., Datsenko, K.A.,
Structural and mechanistic basis of PAM-dependent spacer acquisition in
and Severinov, K. (2014). Pervasive generation of oppositely oriented spacers
CRISPR-Cas systems. Cell 163, 840–853.
during CRISPR adaptation. Nucleic Acids Res. 42, 5907–5916.
Shmakov, S., Abudayyeh, O.O., Makarova, K.S., Wolf, Y.I., Gootenberg, J.S., Wei, Y., Chesne, M.T., Terns, R.M., and Terns, M.P. (2015a). Sequences span-
Semenova, E., Minakhin, L., Joung, J., Konermann, S., Severinov, K., et al. ning the leader-repeat junction mediate CRISPR adaptation to phage in Strep-
(2015). Discovery and functional characterization of diverse Class 2 tococcus thermophilus. Nucleic Acids Res. 43, 1749–1758.
CRISPR-Cas systems. Mol. Cell 60, 385–397. Wei, Y., Terns, R.M., and Terns, M.P. (2015b). Cas9 function and host genome
Sinkunas, T., Gasiunas, G., Fremaux, C., Barrangou, R., Horvath, P., and sampling in Type II-A CRISPR-Cas adaptation. Genes Dev. 29, 356–361.
Siksnys, V. (2011). Cas3 is a single-stranded DNA nuclease and ATP-depen- Westra, E.R., van Erp, P.B.G., Künne, T., Wong, S.P., Staals, R.H.J., Seegers,
dent helicase in the CRISPR/Cas immune system. EMBO J. 30, 1335–1342. C.L.C., Bollen, S., Jore, M.M., Semenova, E., Severinov, K., et al. (2012).
Sinkunas, T., Gasiunas, G., Waghmare, S.P., Dickman, M.J., Barrangou, R., CRISPR immunity relies on the consecutive binding and degradation of nega-
Horvath, P., and Siksnys, V. (2013). In vitro reconstitution of Cascade-medi- tively supercoiled invader DNA by Cascade and Cas3. Mol. Cell 46, 595–605.
ated CRISPR immunity in Streptococcus thermophilus. EMBO J. 32, 385–394. Westra, E.R., Buckling, A., and Fineran, P.C. (2014). CRISPR-Cas systems:
Sokolowski, R.D., Graham, S., and White, M.F. (2014). Cas6 specificity and beyond adaptive immunity. Nat. Rev. Microbiol. 12, 317–326.
CRISPR RNA loading in a complex CRISPR-Cas system. Nucleic Acids Res. Wiedenheft, B., Lander, G.C., Zhou, K., Jore, M.M., Brouns, S.J., van der Oost,
42, 6532–6541. J., Doudna, J.A., and Nogales, E. (2011a). Structures of the RNA-guided sur-
Staals, R.H.J., Zhu, Y., Taylor, D.W., Kornfeld, J.E., Sharma, K., Barendregt, veillance complex from a bacterial immune system. Nature 477, 486–489.
A., Koehorst, J.J., Vlot, M., Neupane, N., Varossieau, K., et al. (2014). RNA tar- Wiedenheft, B., van Duijn, E., Bultema, J.B., Waghmare, S.P., Zhou, K., Bare-
geting by the type III-A CRISPR-Cas Csm complex of Thermus thermophilus. ndregt, A., Westphal, W., Heck, A.J., Boekema, E.J., Dickman, M.J., and
Mol. Cell 56, 518–530. Doudna, J.A. (2011b). RNA-guided complex from a bacterial immune system
Sternberg, S.H., and Doudna, J.A. (2015). Expanding the Biologist’s Toolkit enhances target recognition through seed sequence interactions. Proc. Natl.
with CRISPR-Cas9. Mol. Cell 58, 568–574. Acad. Sci. USA 108, 10092–10097.
Sternberg, S.H., Haurwitz, R.E., and Doudna, J.A. (2012). Mechanism of Wright, A.V., Sternberg, S.H., Taylor, D.W., Staahl, B.T., Bardales, J.A., Korn-
substrate selection by a highly specific CRISPR endoribonuclease. RNA 18, feld, J.E., and Doudna, J.A. (2015). Rational design of a split-Cas9 enzyme
661–672. complex. Proc. Natl. Acad. Sci. USA 112, 2984–2989.

Yosef, I., Goren, M.G., and Qimron, U. (2012). Proteins and DNA elements Zhang, Y., Heidrich, N., Ampattu, B.J., Gunderson, C.W., Seifert, H.S.,
essential for the CRISPR adaptation process in Escherichia coli. Nucleic Acids Schoen, C., Vogel, J., and Sontheimer, E.J. (2013). Processing-independent
Res. 40, 5569–5576. CRISPR RNAs limit natural transformation in Neisseria meningitidis. Mol. Cell
Yosef, I., Shitrit, D., Goren, M.G., Burstein, D., Pupko, T., and Qimron, U. 50, 488–503.
(2013). DNA motifs determining the efficiency of adaptation into the Escheri- Zhang, Y., Rajan, R., Seifert, H.S., Mondragón, A., and Sontheimer, E.J. (2015).
chia coli CRISPR array. Proc. Natl. Acad. Sci. USA 110, 14396–14401. DNase H Activity of Neisseria meningitidis Cas9. Mol. Cell 60, 242–255.
Zetsche, B., Gootenberg, J.S., Abudayyeh, O.O., Slaymaker, I.M., Makarova, Zhao, H., Sheng, G., Wang, J., Wang, M., Bunkoczi, G., Gong, W., Wei, Z., and
K.S., Essletzbichler, P., Volz, S.E., Joung, J., van der Oost, J., Regev, A., et al. Wang, Y. (2014). Crystal structure of the RNA-guided immune surveillance
(2015a). Cpf1 Is a Single RNA-Guided Endonuclease of a Class 2 CRISPR-Cas Cascade complex in Escherichia coli. Nature 515, 147–150.
System. Cell 163, 759–771. Zuris, J.A., Thompson, D.B., Shu, Y., Guilinger, J.P., Bessen, J.L., Hu, J.H.,
Zetsche, B., Volz, S.E., and Zhang, F. (2015b). A split-Cas9 architecture for Maeder, M.L., Joung, J.K., Chen, Z.Y., and Liu, D.R. (2015). Cationic lipid-
inducible genome editing and transcription modulation. Nat. Biotechnol. 33, mediated delivery of proteins enables efficient protein-based genome editing
139–142. in vitro and in vivo. Nat. Biotechnol. 33, 73–80.

Leading Edge
Review
The Basis of Oncoimmunology

A. Karolina Palucka1,2,* and Lisa M. Coussens3,*
1The Jackson Laboratory for Genomic Medicine, 10 Discovery Drive, Farmington, CT 06032, USA
2Baylor Institute for Immunology Research, Dallas, TX 75204, USA
3Department of Cell, Developmental & Cancer Biology, Knight Cancer Institute, Oregon Health and Science University, 3181 SW Sam Jackson
Park Road, Mail Code L215, Room 5508, Richard Jones Hall, Portland, OR 97239-3098, USA
*Correspondence: karolina.palucka@jax.org (A.K.P.), coussenl@ohsu.edu (L.M.C.)
Cancer heterogeneity, a hallmark enabling clonal survival and therapy resistance, is shaped by
active immune responses. Antigen-specific T cells can control cancer, as revealed clinically by
immunotherapeutics such as adoptive T-cell transfer and checkpoint blockade. The host immune
system is thus a powerful tool that, if better harnessed, could significantly enhance the efficacy of
cytotoxic therapy and improve outcomes for cancer sufferers. To realize this vision, however, a
number of research frontiers must be tackled. These include developing strategies for neutralizing
tumor-promoting inflammation, broadening T-cell repertoires (via vaccination), and elucidating the
mechanisms by which immune cells organize tumor microenvironments to regulate T-cell activity.
Such efforts will pave the way for identifying new targets for combination therapies that overcome
resistance to current treatments and promote long-term cancer control.
Introduction constituting the ‘‘tumor stroma’’; collectively, they fuel neoplastic

Cancer is an insidious disease traditionally classified by cell evolution (Hanahan and Coussens, 2012). In other words, recip-
and tissue type of origin. Cancer has historically been treated rocal interactions between accessory cells, their mediators, struc-
according to a ‘‘one size fits all’’ approach based on broad path- tural components of the extracellular matrix (ECM), and genetically
ologic criteria and involving various regimens of cytotoxic ther- altered neoplastic cells regulate all aspects of tumorigenicity.
apy. With the advent of modern sequencing methodologies, These realizations fueled the development of anti-cancer agents
however, we now appreciate that significant genomic, transcrip- targeting the vasculature (Kerbel, 2011) and, more recently,
tomic, and epigenetic heterogeneity exists within individual propelled clinical investigations into the efficacy of immune
tumor types; this recognition has enabled subclassification of therapeutic approaches that neutralize tumor-promoting chronic
tumors of common origin. This, in turn, has led to improved out- inflammation and/or embolden or unleash cytotoxic activities of
comes for some cancer types, as response rates to targeted and antigen-specific T cells (Coussens et al., 2013; Pardoll, 2012).
cytotoxic therapies increase when patients are stratified based Indeed, cancer is visible to the immune system, i.e., immuno-
on the molecular characteristics of their tumors. Examples genic, during early neoplasia. Classic studies from Schreiber and
include imatinib in chronic myelogenous leukemia (Druker colleagues in mice with carcinogen-initiated sarcomas revealed
et al., 2006), HER2-targeted therapies for HER2-positive breast that the immune system could recognize and reject cancerous
cancer (Shepard et al., 1991), and estrogen antagonists for cells (Dunn et al., 2004). Elimination can be explained by cyto-
estrogen-receptor-positive breast cancers (Heiser et al., 2012). toxic antigen-specific T cells responding to relatively high muta-
These molecular advances helped to usher in a new era of tional burdens induced by carcinogens and thus providing
precision medicine that is reshaping clinical treatment across neo-antigens for T-cell priming; these findings established the
the cancer spectrum. However, there remain significant frac- principles of elimination, equilibrium, and eventually escape
tions of patients that do not respond to ‘‘designer’’ therapies when neoplastic cells become invisible to the immune system
even when their tumors are classified based on molecular and (Dunn et al., 2004). Neoplastic cells in part escape when tumor
pathologic criteria. Additional tumor or systemic characteristic(s) arises out of chronically inflamed tissues—there, chronic infiltra-
are thus unaccounted for that not only impact neoplastic growth tion of tissue by leukocytes (e.g., type 2 cytokine-activated
and dissemination, but also impact response to therapy. myeloid cells and immune-suppressive B, T, and myeloid sub-
Recent seminal in vivo studies revealed that neoplastic cells rely sets) subvert T-cell-directed elimination and thus aid tissue-
on the diversity of normal resident and recruited accessory cells to based programs, e.g., angiogenesis, lymphangiogenesis, matrix
support their evolution (Hanahan and Coussens, 2012). Accessory remodeling, etc., supporting neoplastic progression (Coussens
cells are now recognized as ‘‘neoplastic cell-extrinsic hallmarks et al., 2013).
of cancer’’ and include those forming the hematogenous and Mounting observations in humans support the concept that
lymphatic vasculature, tissue-specific mesenchymal support cancer initiation and progression are significantly impacted by
cells, and myeloid and lymphoid-lineage immune cells. Accessory altered or misled immune responses (Figure 1). Individuals
cells integrate with the dynamic soluble and insoluble matrices suffering from chronic inflammatory conditions are at increased
Cell 164, March 10, 2016 ª2016 Elsevier Inc. 1233

Figure 1. The Makings of Tumor Immunity
The communication between cancer and the im-
mune system is a dynamic process, reminiscent of
a balance. When immunity to cancer is ‘‘up’’ and
the suppressive processes are ‘‘down,’’ cancer is
under control. However, a strong anti-tumor im-
mune response will trigger largely physiological
processes designed to dampen effector T cells to
prevent tissue damage and maintain tissue ho-
meostasis. Given that the immunity might have
evolved mainly to maintain self, to establish
coexistence with environment, and to occasionally
protect self from external threats, the suppression
prevails. Multiple pathways of suppression are at
play in tumor microenvironments, including cells
such as TH2-polarized macrophages, immature and suppressive monocytes, regulatory B cells, and regulatory T cells, as well as molecules such as checkpoints
that control T-cell differentiation (for example, CTLA-4 and IDO) and effector function (such as PD-1). Pharmacological blockade of these inhibitory pathways can
tip the balance toward anti-cancer effector T cells. The latter ones can be primed or boosted by antigen-presenting cells (DCs) and/or by co-stimulatory signals
(for example, CD137 ligands). Recent studies demonstrate that thymus-independent neo-antigens generated in adult life by somatic mutation or post-trans-
lational regulation (for example, phosphorylation) might be more immunogenic (or perhaps linked with less suppression) than shared tumor antigens. Neo-
antigens can occur as random results of somatic mutation, as well as a by-product of anticancer treatments, e.g., chemotherapy (CTX) or radiation therapy (RT),
or by targeting epigenetic control mechanisms or drugs intervening with DNA repair pathways. They can be presented to T cells in exogenous vaccines, as well as
endogenously via DCs that captured dying neoplastic cells. When T cells specific to defined antigens kill neoplastic cells, such a process can enable generation of
responses to other antigens, so called epitope spreading. A critical factor in the balance between immunogenicity and suppression is inflammation (which, in turn,
is impacted by the microbiome); indeed, the type of inflammation (tumor-destructing TH1 or tumor-promoting TH2 and TH17) should become a part of TNM
grading, along with pathology, microbiome phenotype, and immune infiltrate assessment.
risk for developing cancer (Thun et al., 2004). Incidence of viral more amenable to control than ‘‘cold’’ tumors, i.e., tumors
(DNA tumor virus) and carcinogen-associated cancers is with diminished T-cell infiltrates, thus driving modern cancer
increased in immune-compromised individuals, even as the rela- medicine to investigate how to reprogram the tumor microenvi-
tive risk of cancer types lacking viral or carcinogen etiology is ronment (TME) to attract the right type of immune infiltrate. This
diminished (reviewed in de Visser et al., 2006). Age-related im- topic, along with other open questions in the field of oncoimmu-
munosenescence likely plays a role in increased incidence of nology, are discussed here.
malignancy in aged individuals (Campisi et al., 2011). The advent
of some biologic therapies impacting how tissues activate and The Makings of the Immune Response to Cancer
resolve inflammation, e.g., tumor necrosis factor (TNF) blockade Tumors are organized tissues with numerous reciprocal local
(Bongartz et al., 2006), also skews cancer incidence metrics. and systemic connections with immune cell populations of
However, the role(s) that immune pathways play in driving malig- both the myeloid and lymphoid lineages. Here, we summarize
nancy remains to be clarified. How does the immune system the key myeloid and lymphoid populations regulating the im-
recognize tissue-specific mediators triggering and maintaining mune response to cancer and how the fundamental physiolog-
chronic inflammatory responses? What oncogenic events and ical processes that they govern are harnessed for neoplastic
altered metabolic states lead to the generation of neo-antigens progression and tumor formation.
that in turn induce T-cell responses? What physiological mech- The Myeloid Compartment
anisms regulate immune homeostasis such that (acute) inflam- Myeloid cells have multiple homeostatic functions that are co-
mation can be resolved as rapidly as it is activated (a critical opted by evolving neoplasms; these can be roughly summarized
control program to thwart autoimmunity)? What is the role of as: (1) antigen capture for degradation (macrophages) or presen-
the host microbiota in regulating systemic immune responses tation (dendritic cells [DCs]); (2) tissue repair (macrophages), and
to neoplasia? How do neoplastic cells survive immune attack (3) effector functions (mast cells, monocytes, and granulocytes).
by T cells? These questions are in need of answering to effec- Neoplastic cells can alter the steady-state activity of all myeloid
tively move cancer research and cancer medicine forward. cells present in the TME, including tissue-resident and blood-
A common feature of all cancers, regardless of origin, is derived cells, by secreting factors such as interleukin (IL)-6 or
prominent presence of diverse assemblages of immune cells granulocyte-macrophage colony-stimulating factor (GM-CSF),
(Coussens et al., 2013). The consequences of such infiltrates that increase recruitment and proliferation of immature myeloid
on the fate of cancerous cells are diverse (Figure 2). For cells atypical under physiological conditions (Gabrilovich et al.,
example, under continual immune pressure, i.e., antigen pre- 2012).
sentation to T cells, neoplastic cells become ‘‘immune-edited’’ An important feature of myeloid cells is their functional plas-
to escape immune surveillance (Dunn et al., 2004) and instead ticity in response to environmental signals. This property can
co-opt immune cells to favor their sustained proliferation (Balk- dictate such opposite outcomes as antigen degradation or anti-
will et al., 2005). Nonetheless, recent studies demonstrate that gen presentation when macrophages acquire DC capabilities
the presence of lymphoid aggregates is linked with improved re- (Banchereau et al., 2000), tissue repair rather than inflammation
sponses to cancer therapies—for example, standard cytotoxic when macrophages are polarized toward type 2 states, and pro-
therapies, vaccine-based treatments, or immune checkpoint tective or non-protective T-cell immunity when programmed by
blockade (Topalian et al., 2015). Such ‘‘hot’’ tumors are thus cancer-derived factors (Balkwill et al., 2005). Thus, plasticity
1234 Cell 164, March 10, 2016 ª2016 Elsevier Inc.

Figure 2. Immune-Mediated Landscape
The yin and yang implications of tumor-immune
system communications form the basis for dis-
ease pathophysiology and, at the same time, tar-
gets for therapeutic intervention. The disease
landscape emerging from these multi-factorial in-
teractions is orchestrated by the three compart-
ments, i.e., the cancer, the immune system, and
the host. The outputs are numerous and dramati-
cally opposite, as well as both local and systemic,
and include: immunity that might control cancer;
chronic inflammation that can be linked with tissue
remodeling processes and metabolic changes
that support neoplastic cell survival and primary
tumor development; angiogenesis and lym-
phangiogenesis that can also support metastatic
dissemination; as well as systemic consequences
for the host including cachexia. Clearly, therapy
going forward will require a well-timed and
orchestrated combination of therapies, targeting
multiple modes of communication and effect, to
combat this multi-factorial disease, taking into
account the patient’s steady-state commensal
bacteria complexity and load and how that is
impacted by therapy.
DCs express numerous pattern recog-

nition receptors, including lectins, Toll-
like receptors (TLRs), NOD-like receptors
(NLRs), and helicases, through which
they can sense microbes and tissue dam-
and communication within the myeloid compartment and be- age (cancer) such as increased pericellular nucleic acids (Pulen-
tween myeloid and other immune cells and stromal components dran, 2015). If DCs do not receive maturation signals, such as
is critical for tumor formation. when exposed to high levels of IL-10 (Ruffell et al., 2014), they
Cancer Antigen Presentation and Dendritic Cells. Cancer anti- remain immature and antigen presentation instead leads to T-
gens are presented to T cells either at tumor sites or in draining cell suppression. DC plasticity in response to extrinsic signals,
lymph nodes by DCs (Figure 3). Cancer antigens, soluble and together with the existence of discrete subsets with unique func-
cell borne, are transported to lymph nodes via lymphatic vessels. tions, empowers DCs as key initiators and regulators of the im-
Soluble antigen is captured by lymph-node-resident DCs, while mune response (Pulendran, 2015). We will illustrate this point
tissue-resident DCs capture antigen at tumor sites; either popu- briefly; mouse and human DC subset biology was recently re-
lation can present antigen locally or migrate through lymphatic viewed elsewhere (Merad et al., 2013).
vessels to present in lymph nodes (Steinman, 2011). DCs display The diversity of human DC subsets was revealed by studies
protein antigens in the context of classical major histocompatibil- of blood and skin DCs. Three main cell-surface markers
ity (MHC) class I and MHC class II molecules or lipid antigens in distinguished human-blood-circulating DC subsets: CD303
the context of non-classical CD1 molecules that allow selection (BDCA-2) on plasmacytoid DCs (pDCs), CD1c (or BDCA-1) ex-
of rare antigen-specific T lymphocytes, including CD8+ T cells, pressed on the majority of circulating DCs, and CD141/
CD4+ T cells, and NK T cells. Compared with other antigen-pre- BDCA-3 expressed on a small fraction (Merad et al., 2013). Hu-
senting cells (APCs), DCs are extremely efficient in their ability man CD141+CD1c DCs uniquely express TLR3, produce IL-12,
to induce antigen-specific T-cell responses, justifying their and efficiently cross-prime CD8+ T cells when activated with
name ‘‘professional APCs’’ (Lanzavecchia and Sallusto, 2001). poly I:C (Joffre et al., 2012). However, other human DCs, such
Naive CD8+ T cells differentiate into cytotoxic T lymphocytes as epidermal Langerhans cells and CD1c+ DCs, also cross-
(CTLs) in lymphoid organs upon encounter with DCs, presenting present antigens to CD8+ T cells. Indeed, our studies have
tumor-derived peptides in the context of co-stimulation through unraveled the basic principles by which human DC subsets
CD80, CD70, and 4-1BB, as well as through DC-derived cyto- differentially regulate CD8+ T cells (Klechevsky et al., 2008).
kines such as IL-12, type I interferon, and IL-15 (Steinman, Thus, human Langerhans cells are highly efficient at priming
2012). The priming of new T-cell repertoires during tumorigenesis cytotoxic CD8+ T cells, while CD14+ dermal DCs prime type 2
may be critical for clinical success of therapeutic agents aiming to cytokine-secreting CD8+ T cells (Klechevsky et al., 2008). Blood-
unleash antigen-specific CTL activities. Naive CD4+ T cells can and tissue-resident CD1c+ DCs, but not CD141+ DCs, exposed
give rise to helper cells with distinct cytokine profiles or to Fox- to live-attenuated influenza virus promote CD103 (aE integrin)
P3+ regulatory T cells (Treg), whose role is to dampen CTL activity expression on CD8+ T cells and their accumulation in epithelia
and avoid autoimmune responses (Zhu and Paul, 2008). (Yu et al., 2013).

Figure 3. The Priming of Cancer Immunity
The cycle of anti-tumor immunity starts presumably with presentation of cancer antigens liberated in the process of cell turnover; this same pathway can be
followed for vaccination, as illustrated herein. Antigens are sensed and captured either by tissue-resident DCs or by DCs in draining lymph nodes (LNs). DCs
initiate an immune response by presenting captured antigens, in the form of peptide-major histocompatibility complex (MHC) molecule complexes, to naive (that
is, antigen inexperienced) T cells in lymphoid tissues. When compared with other APCs, such as macrophages, DCs are extremely efficient and can elicit very low
numbers of T cells to respond. Naive CD8+ T cells differentiate into CTLs in lymphoid organs upon encounter with DCs presenting tumor-derived peptides in the
context of co-stimulation through CD8, CD70, and 4-1BB, as well as DC-derived cytokines such as IL-12 and IL-15. Naive CD4+ T cells can give rise to helper cells
(e.g., TH) with distinct cytokine profiles or to regulatory T cells (Treg) whose role is to dampen the immune response. T cells migrate through blood and lymphatics.
Upon arrival in tumor beds, CD8+ T cells must confront numerous barriers including: (1) intrinsic regulators, for example, CD28-CTLA-4, PD1-PDL1, and ILTs, as
well as extrinsic regulators cells such as Tregs, Bregs, or myeloid cells; (2) a corrupted TME with pro-tumor inflammation; (3) impaired cross-presentation due to
TME-based DC inhibition; (4) antigen loss and immune evasion of tumor target; and (5) tissue-specific alterations such as fatty cells in breast cancer or des-
mofibrosis in pancreatic cancer stroma. Killing of tumor cells either via T cells or by standard therapy can lead to endogenous antigen release and DC activation,
so called ‘‘endogenous vaccination,’’ thereby closing the cycle. Inevitable to this is the induction of tissue resistance mechanisms, for example, expression of PD-
L1 on neoplastic or other immune cells, as the result of powerful effector immunity, including actions of IFNg. Thus, future immunotherapy approaches will be
based on combinations of different therapeutics targeting distinct components of this cycle, for example, via intratumoral delivery of activating agents able to
reprogram the function of infiltrating leukocytes.
The Lymphoid Compartment for T cells), the MHC class I molecule (for NK cells), or surface
The lymphoid compartment in tumors includes natural killer (NK) proteins (for B-cell products, i.e., antibodies) that can be recog-
cells, gd T cells, NK T cells, CD4+ T cells, CD8+ T cells, and B nized in a specific manner. In addition, lymphoid cells can
cells. Their functional activity depends upon expression of re- be induced to secrete different types of cytokines based on
striction elements, including peptide-MHC complexes (pMHC; effector functions. For example, following an activating stimulus,

TH1-polarized CD4+ T cells secrete IL-2, TNFa, and IFNg; in An important concept recently proposed by Mellman and col-
conjunction with cytotoxic CD8+ T cells, they promote macro- leagues is the cancer-immunity cycle (Chen and Mellman, 2013).
phage cytotoxic activity (Stout and Bottomly, 1989) and can It becomes apparent that any effective immune response against
induce upregulation of antigen processing and expression of cancer will generate resistance via physiological pathways that
MHCI and II molecules in professional APCs (i.e., macrophages evolved to protect tissue homeostasis. Here, we discuss how
and DCs). In contrast, expression of IL-4, -5, -6, -10, and -13 by this cycle is altered in cancer pathogenesis and how it can be
TH2-polarized CD4+ T cells can induce T-cell anergy and loss of harnessed therapeutically. Clearly, combination therapies that
T-cell-mediated cytotoxicity, enhance humoral immunity, and intervene at several distinct pathways within the cancer-immu-
regulate the tumor-promoting activities of macrophages (De- nity cycle are needed to achieve cancer control.
Nardo et al., 2009).
CD8+ T cells are considered the major anti-cancer effector Chronic Inflammation and Alterations of Leukocyte
cells, as they can give rise to CTLs that kill neoplastic cells pre- Compartments in Cancer
senting a specific pMHC complex (Appay et al., 2008). CTLs can Basic Principles
be generated through either the priming of naive T cells or re- Unabated inflammation is a hallmark of cancer and is mediated
programming of memory T cells. Naive CD8+ T cells differentiate by immune cells attracted to or residing at sites of neoplastic
into CTLs in lymphoid organs upon encounter with APCs pre- transformation (Balkwill et al., 2005). Indeed, immune cells are
senting tumor-derived peptides in the context of appropriate selectively recruited into early neoplastic tissues, likely in
co-stimulation and cytokine help. The ideal properties of anti- response to hard-wired pathways utilized by all tissues to
cancer CD8+ T cells include: high affinity for pMHC on tumor resist/repair damage caused by bacterial, viral, or other patho-
cells; high levels of cytotoxic mediators, e.g., granzymes A and genic assaults. When successful, ‘‘initiated’’ pre-neoplastic cells
B and perforin; expression of surface molecules, allowing traf- are purged by the immune system (Dunn et al., 2004). When the
ficking into the tumor; and extended longevity and memory, immune system fails, neoplastic cells are retained in ‘‘damaged’’
thus enabling CTL generation upon antigen re-exposure (Appay TMEs and provide a survival advantage resulting from abundant
et al., 2008). bioavailable mediators liberated as a function of tissue remodel-
Memory T cells have long been described as two circulating ing (Hanahan and Coussens, 2012). Ensuing neoplastic progres-
populations: (1) central memory T cells that migrate between sion requires sustained presence of select immune subtypes
the secondary lymphoid organs and are capable of mounting that, combined with ongoing host-derived programs (angiogen-
proliferative responses on pathogen re-encounter and (2) esis, matrix, and tissue remodeling, etc.), contribute to cancer
effector memory T cells that traffic between blood and extralym- progression (Hanahan and Coussens, 2012) (Figure 2).
phoid compartments for peripheral immune surveillance (Mueller The classic view that immune cells merely facilitate tumor
et al., 2012). Tissue-resident memory T cells are a third and rejection has been supplanted by a more complex view of leuko-
phenotypically distinct category. Studies in mice and humans cytes having both tumor-promoting and tumor-inhibiting proper-
have revealed that this latter population can be superior to circu- ties (Coussens et al., 2013). This is best explained by the
lating central memory T cells at providing rapid long-term pro- existence of (at least) two types of inflammation with opposing
tection against re-infection (Sheridan and Lefrançois, 2011). effects on tumors: chronic inflammation, which promotes
Therefore, an active mechanism of peripheral T-cell retention neoplastic cell survival, angiogenesis, tissue remodeling, and
likely exists not only to facilitate clearance of infected cells, but metastasis, and acute inflammation that triggers neoplastic cell
also to promote accumulation at sites having cleared an infec- destruction. While chronic inflammation is often linked with the
tious virus. CD103/b7 integrin endows peripheral CD8+ T cells presence of TH2 responses, acute inflammation associated
with a unique capacity to access epithelial compartments. with cancer destruction is linked to TH1 responses.
Expression of CD103 on CTLs mediates adherence to E-cad- As neoplastic cells escape elimination, some become less
herin and appears to be important in the final stages of immunogenic by downregulating MHC molecules; however,
neoplastic cell lysis and rejection (Le Floc’h et al., 2007). Indeed, most if not all also activate intrinsic gene-expression programs
for mucosal cancer vaccines, homing to and retention of CD8+ that are inherently T-cell suppressive and myelo-stimulatory,
T cells in mucosa is critical for efficacy (Sandoval et al., 2013). e.g., TH2 responses. Cytokines implicated in these scenarios
Upon arrival in tumor beds, CD8+ T cells must confront include transforming growth factor b (TGFb); IL-4, -13, -8,
numerous barriers, including intrinsic checkpoint regulators, and -10; thymic stromal lymphopoietin (TSLP); and indole-
such as CD28-CTLA-4, PD1-PD-L1, and immunoglobulin-like amine 2,3-dioxygenase (IDO) (Coussens et al., 2013). This en-
transcript receptors (ILTs) (Pardoll, 2012); extrinsic checkpoint ables recruitment of FoxP3+CD4+ Treg cells, TH2-CD4+ T cells,
regulators, such as Treg cells (Fehérvari and Sakaguchi, 2004) TH2-polarized macrophages and monocytes, and B regulatory
or myeloid cells (Gabrilovich et al., 2012); a corrupted TME cells (Bregs). In response to TH2-mediated activation, myeloid
with protumor inflammation (Coussens et al., 2013); antigen cells commonly increase synthesis of angiogenic (e.g., VEGF),
loss and immune evasion of tumor targets (Klebanoff et al., growth and/or survival (e.g., EGF, TNFa) factors that directly
2011); and tissue-specific alterations, such as fatty cells in regulate epithelial cell proliferation, as well as tissue-remodeling
breast cancer or desmofibrosis in pancreatic cancer stroma. enzymes (e.g., metallo-, cysteine, and serine proteases). These
Defining strategies for bypassing these obstacles and improving activities are remarkably pro-tumorigenic in that they nurture
the clinical efficacy of T-cell therapies is the object of intense a TME favoring neoplastic cell survival and sustained prolif-
study. eration (Balkwill et al., 2005). Simultaneous TH2 activation of

macrophages and monocytes also increases expression of mol- vironment in the lung that resists neoplastic progression and
ecules, e.g., inducible nitric oxide synthase or Arginase 1, that metastatic dissemination (Fridlender and Albelda, 2012).
directly and indirectly suppress CD8+ T-cell proliferation and cy- Eosinophils, like other myeloid lineage cells, can exert cyto-
tokines such as IL-10 that inhibit DC maturation and antigen toxic immune-effector activities. Tumor-associated tissue eosin-
cross-presentation to T cells (Ruffell et al., 2014). Thus, TH2- ophilia (TATE) is associated with improved prognosis for a
type immune microenvironments are both tumor promoting number of malignancies, including gastrointestinal, bladder,
and immune suppressive. Notably, in the colon, tumor-promot- and prostate cancers; in contrast, TATE is associated with
ing immunity via IL-17 (TH17)-mediated activation of myeloid poor outcome in Hodgkin’s lymphoma, cervical carcinoma,
and lymphoid cells has been reported (Wang et al., 2009; Wu and oral squamous cell carcinoma (Davis and Rothenberg,
et al., 2009). 2014). Eosinophils have been associated with degranulation
Tumor-Promoting Activities of the Myeloid and release of cytotoxic proteins that mediate tumor rejection;
Compartment recent results also reveal their role in normalizing the vasculature
Owing to their established role in wound healing, we investigated to improve CD8+ T-cell trafficking associated with tumor regres-
the ability of myeloid cells infiltrating early benign tissues to fos- sion (Carretero et al., 2015).
ter malignancy. In mice prone to squamous carcinogenesis, Monocytes, once in tissues, can differentiate into macrophages
mast cells and macrophages activate pro-neoplastic angiogenic and DCs. Two circulating monocyte populations have been iden-
and tissue-remodeling programs (Coussens et al., 1999). In other tified: classical inflammatory monocytes that are CCR2HIGH and
studies of mice bearing mammary carcinomas, macrophages non-classical patrolling monocytes that are CX3CR1HIGH (Geiss-
could regulate neoplastic cell dissemination and metastasis via mann et al., 2003). Recruitment of inflammatory monocytes into
EGF-mediated paracrine interactions with neoplastic epithelial tissues is normally guided by the CCR2-CCL2 axis in response
cells (Lin et al., 2001). In human cancers, multiple studies to parasitic or bacterial infections; in tumors, when CCR2HIGH
have reported that the presence of macrophages in stroma cor- monocytes are recruited, they can promote neoplastic cell sur-
relates with aggressive disease and outcome (Komohara et al., vival and extravasation through VEGF and CSF1 production
2014). Macrophages are recruited into tumors following activa- (Qian et al., 2011). CCR2HIGH monocytes promote survival of met-
tion of colony-stimulating factor 1 receptor (CSF1R) by either astatic cells through a CCL3-dependent mechanism (Kitamura
CSF1 or IL-34, two high-affinity ligands for CSF1R; the chemo- et al., 2015). CX3CR1HIGH monocytes instead patrol capillaries
kine CCL2 may also facilitate macrophage recruitment (Qian in response to the CXC3R1-CX3CL1 axis; in these locales, they
et al., 2011). A CSF1-response gene-expression signature has are positioned to scavenge particles and debris and thus are
been identified in 17%–25% of breast cancers associated with more likely to be found in wounds when inflammation is resolving.
decreased estrogen receptor and progesterone receptor At sites of metastasis, CX3CR1HIGH monocytes recruit NK cells
expression (Beck et al., 2009); serum concentrations of CSF-1 that, in turn, kill metastatic cells, thereby providing a potent sur-
correlate positively with breast tumor size and predict poor survival advantage (Hanna et al., 2015). In pancreatic adenocarci-
vival (Aharinejad et al., 2013). In addition, in two independent nomas, activation of the Ras oncogene leads to increased
breast cancer cohorts, intratumoral macrophage presence was expression of GM-CSF and recruitment of immature monocytes
correlated with potentially prognostic tumor features (high- that subsequently suppress CD8+ T-cell proliferation to enhance
grade, hormone receptor negativity; basal-like subtype; and tumor progression (Pylayeva-Gupta et al., 2012), analogous to
increased risk of death) (Komohara et al., 2014). Macrophages other tumor systems (Gabrilovich et al., 2012).
therefore serve as promising targets for novel therapeutic inter- Mast cells, present in all vascularized tissues, respond to
ventions, particularly for patients with high-risk disease. diverse stimuli by either secreting or releasing (via degranulation)
Conversely, favorable prognosis has been associated with biologically active compounds, e.g., proteolytic enzymes, hepa-
some tumor types exhibiting increased macrophage infiltration, rin, histamine, prostaglandins, cytokines, and chemokines. Mast
e.g., non-small-cell lung cancer, prostate, colorectal, and gastric cells are key for maintaining tissue homeostasis and are best
cancers (Komohara et al., 2014). Whether these distinctions known for their effector functions following IgE-stimulated
reflect true differences in macrophage biology and function or allergic responses and anaphylaxis (Metz et al., 2007). Mast cells
arise due to discordant detection methodologies is unclear. have been implicated in the vascularization of a multitude of
Neutrophils, on the other hand, are typically less abundant than solid human tumor types, likely owing to their proteolytic prod-
macrophages in solid tumors, but their presence correlates with ucts and high VEGF expression following activation (Coussens
reduced survival in head and neck and breast cancers, and similar et al., 1999; Marichal et al., 2013) following CCL2-mediated
to macrophages, neutrophils develop polarized phenotypes that recruitment where their bioactive mediators promote neoplastic
either favor or restrict tumor progression (Fridlender and Albelda, progression.
2012). Recent studies identified granule products that suppress Tissue Specificity of Myeloid Programming
T-cell function (Sippel et al., 2011). Neutrophil expansion in mam- While it is conceptually unclear how some myeloid cells adopt a
mary carcinomas of mice bearing mutant p53 alleles is driven TH2 or protumorigenic state to support neoplastic progression,
by T-cell-derived IL-17; this results in systemic granulocyte some clues have emerged in recent genetic studies. Several
colony-stimulating factor (G-CSF)-dependent expansion and groups revealed that lymphocytes drive initial myeloid cell pro-
polarization toward a T-cell-suppressive phenotype that facili- gramming to foster chronic inflammation in a tissue-specific
tates metastatic dissemination and colonization (Coffelt et al., manner. For example, during mammary branching morpho-
2015). In contrast, neutrophils create a tumor-restrictive microen- genesis and ductal carcinogenesis, cytokines derived from

TH2-CD4+ T cells, e.g., IL-4 and -13, activate macrophages and inhibitor of CSF1R kinase increased progression-free survival
monocytes infiltrating mammary tissue (DeNardo et al., 2009; and improved outcomes as a monotherapy (Tap et al., 2015).
Plaks et al., 2015). In neoplastic scenarios, signaling down- The macrophage presence in tumors has also been therapeuti-
stream of IL-4 receptors on monocytes and macrophages cally manipulated by targeting the macrophage signaling protein
triggers protumorigenic TH2 gene-expression programs that acting through its transmembrane receptor kinase RON, wherein
activate tissue-remodeling cascades via expression and activa- activation of RON in macrophages favors conversion of micro-
tion of cathepsin proteases and immune-suppressive programs metastatic lesions to overt metastases by suppressing antitumor
via upregulation of IL-10 and immune-checkpoint molecules immune responses. Functional RON blockade in preclinical
(DeNardo et al., 2009; Gocheva et al., 2010; Mitchem et al., models potentiates tumor-specific CD8+ T-cell responses, indi-
2013; Ruffell et al., 2014). Mast cells and macrophages (as well cating that RON inhibitors may also improve outcomes for can-
as other myeloid cell types) are TH2 programmed in early squa- cer patients (Eyob et al., 2013).
mous and pancreatic carcinomas by a diversity of pathways, Bruton’s tyrosine kinase (BTK) is an attractive target, as BTK is
which also include activation of immunoglobulin receptors activated downstream of the B-cell receptor and FcgR and PI3Kg
(FcgRs) by immune complexes (ICs) (Affara et al., 2014; Andreu in some myeloid subsets (Smith et al., 2001). In vitro, neoplastic
et al., 2010). ICs are composed of antigen-specific antibodies cell challenge via co-culture with splenic cells from B-cell-defi-
and complement proteins that variably activate FcR and comple- cient versus B-cell-proficient mice revealed that IFNg release
ment receptors depending on composition of IC and status of from CD8+ and NK cells is increased when B cells were absent,
the myeloid cell being activated (Karsten and Köhl, 2012). While whereas presence of B cells or B-cell-derived IL-10 was associ-
these humoral immune-regulated paracrine programs were ated with reduced IFNg (Inoue et al., 2006). Though these in vitro
known to shape outcomes in chronic inflammatory diseases, studies indicate that B cells can direct T-cell responses, the role
recognition of their significance in solid tumors was paradigm of myeloid cells as mediators of these responses is now clear and
shifting and highlighted the significance of hard-wired tissue- indicates that therapies targeting common pathways in B cells
specific programs shaping host response to disease. These and/or myeloid cells, such as SYK, BTK, PI3Kg, may be effica-
data illustrate the diversity of pathways utilized by innate immune cious in solid tumors, analogous to efficacy observed for BTK
cells to propel cancer by directly enhancing tissue-based pro- and PI3Kd inhibitors in B-cell malignancies (Hendriks et al.,
grams favoring survival of neoplastic cells, in concert with direct 2014). This concept was recently validated preclinically, whereby
and indirect activation of programs to extinguish cytotoxic im- BTK inhibition enhanced survival of mice bearing pancreatic
mune responses aiding immune escape (Figure 2). adenocarcinoma (Gunderson et al., 2016; Massó-Vallés et al.,
2015), neuroendocrine cancers (Soucek et al., 2011), and other
TH2-Based Targets for Anti-cancer Therapy subcutaneous tumors (Sagiv-Barfi et al., 2015) in which a com-
The collective evidence not only supports a protumorigenic role mon feature was reduced inflammation and inflammatory des-
for chronic inflammation in cancer, but also indicates that inflam- moplasia with evidence of macrophage repolarization.
mation is malleable, akin to the healing of acute wounds during If these preclinical findings are any indication, immune thera-
which immune cells toggle between TH1 and TH2 states. Thus, pies targeting macrophages and/or other protumorigenic
the hypothesis that TH2-driven myeloid cells could be re-pro- immune cells could alter the human tumor immune microenvi-
grammed, or at least neutralized, to reduce the presence or ronment in a way that fosters cytotoxic properties of CD8+
immunosuppressive status of macrophages, trigger anti-tumor T cells. As immune-checkpoint inhibitors of pathways regulating
immunity, and/or suppress tumor growth has been tested in T-cell activity are proving efficacious for subsets of cancer
several tissue-specific cancer models. We and others have eval- patients, we predict that combining these two immune-based
uated CSF1-neutralizing monoclonal antibodies (aCSF1 mAB) approaches represents a compelling clinical opportunity. How-
and small-molecule CSF1R inhibitors for their ability to suppress ever, it is likely that not all tumors will respond; thus, identifying
macrophage survival and/or presence in tumors, in combination predictive biomarkers and correlates of therapeutic response
with chemotherapy (CTX) or radiation therapy (RT) (DeNardo is a top priority. Based on preclinical data evaluating macro-
et al., 2011; Ruffell et al., 2014; Shiao et al., 2015). These studies phage antagonists and checkpoint inhibitors, we predict that
reveal increased chemo- and radiation sensitivity associated biomarkers of response will also be reflected by changes in pe-
with anti-tumor immune responses directed by CD8+ T-cell infil- ripheral blood. Such biomarkers will form the basis for simple,
tration of tumors, culminating in reduced primary tumor growth non-invasive diagnostic and/or prognostic screens aiding early
and metastasis with increased survival. Other preclinical studies detection in susceptible populations (Figure 4).
revealed that CSF1/CSF1R blockade, as monotherapy or com- In preclinical models, regardless of tumor or approach, TH1
bined with CTX/RT, improved outcomes for glioma, prostate, immunity emerges when dominant TH2-driver pathways are
and pancreatic adenocarcinoma, as well as melanoma, whereas attenuated; when concomitant with cytotoxic therapy, tumor
CSF1R blockade improved antitumor efficacy of immune check- growth stalls or regresses by CD8+ T-cell-dependent mecha-
point blockade and adoptive T-cell therapy (reviewed in Ruffell nisms. These findings highlight the importance of neutralizing
and Coussens, 2015). Importantly, administration of RG7155, a pro-tumor inflammation as a therapeutic strategy and indicate
CSF1R-blocking mAb, in patients with diffuse-type giant cell that tumor-infiltrating CTLs can be mobilized in tumors with
tumors, reduced CSF1R+CD163+ macrophage levels; this trans- low mutational burdens. These data also highlight the clinical
lated into objective clinical responses (Ries et al., 2014). Treat- need for biomarkers that identity tissue-specific programs
ment of tenosynovial giant-cell tumors with a small-molecule driving TH2 immune responses; such data is needed to inform

Figure 4. Multi-Modal Biomarker-Based Approach for Optimal Im-
mune-Mediated Tumor Control
Cancer medicine is evolving. Going forward, individuals with cancer will be
evaluated for biomarkers enabling stratification to determine most optimal
combinations for therapy based on tumor-based and systemic biomarkers.
Eventually, all patients with cancer will be treated with checkpoint inhibitors,
either directly or after interventions targeting inflammation (for example, with
TH2-blockade therapies, radiation therapy, or epigenetic modulation), or
vaccination via DCs to boost T-cell repertoires, or adoptive T cell transfer.
Based on the known tissue-embedded programs empowered to control auto-
immunity, it is reasonable to anticipate that a majority of patients will develop
acquired resistance followed by immune escape; this will lead to the next cycle
of treatments incorporating multi-modal biomarkers (e.g., based on micro-
biome phenotype, circulating cell-free DNA [cfDNA], circulating cytokine
levels) and perhaps NK cells recognizing loss of MHC class I by neoplastic
cells, thus rendering them invisible to T cells. Cytotoxic treatments such as
with NK cells or standard cytotoxic therapy (CTX or RT) or oncolytic viruses will
release neo-antigens that can be used for generation of the next round of
effector T cells. Whole-exome sequencing (WES) of tumor samples as well as
cfDNA will yield information on mutational load that can, in turn, be used as one
class of neo-antigens for vaccination and priming of new T-cell repertoires.
T-cell receptors (TCR) can be assessed using genomic approaches enabling
sequencing of TCRb chains to assess repertoire diversity. Given the impor-
tance of T-cell specificity for relevant antigens, strategies enabling paired
sequencing of a and b TCR chains will be invaluable, as will high-throughput
tetramer analysis. In addition, RNA-seq and epigenetic analysis of tumors and
their infiltrates will enable assessment of the type and flavor of inflammation.
Future studies will incorporate metabolomics to this biomarker portfolio.
precision medicine strategies employing TH2 blockade, in con-

cert with other immune, targeted, or cytotoxic approaches
(Figure 4).
Immune-Targeted Therapies Focused on T Cells

Basic Principles
Cancer immunotherapy historically relied on two principal mech-
anisms of action: (1) ‘‘passive’’ immunotherapy via provision of
anti-tumor antibodies, e.g., Trastuzumab (aHER2 mAB) or Ritux-
imab (aCD20 mAB), or adoptive transfer of cytotoxic T and NK
cells; and (2) ‘‘active’’ immunotherapy that mobilizes the pa-
tient’s immune cells via checkpoint blockade, i.e., administration
of antibodies directed against immune-regulatory checkpoint
molecules expressed on T cells or via vaccines that expand
antigen-specific T cells. In all circumstances, T cells are the
drug; we are learning that T cells have the ability to clinically
control some cancers (Postow et al., 2015). T cells can be tar-
geted in three major ways: (1) by being liberated by checkpoint
inhibitors; (2) through adoptive transfer when T cells are missing,
as validated by the clinical success of genetically engineered
T-cell therapies; and (3) through induction in vivo by vaccination
or endogenous mechanisms subsequent to other anti-cancer
therapies (e.g., CTX, targeted therapies, or anticancer anti-
bodies) (Palucka and Banchereau, 2013).
Immune Checkpoint Blockade
Immune checkpoint blockade (ICB) can unleash the power of
naturally occurring T cells by eliminating negative signals that
block T-cell function (Pardoll, 2012). ICB has produced durable
clinical responses and improved survival across a variety of can-
cers (Topalian et al., 2015). CTL expansion and function are care-
fully regulated by cytotoxic-T-lymphocyte-associated protein 4
(CTLA-4), programmed cell death protein 1 (PD-1), and other
molecules so as to maintain a delicate balance between the res-
olution of infection, the elimination of infected cells, the protec-
tion of tissue homeostasis, and the prevention of autoimmune

attack. CTLA-4 is a cell-surface receptor expressed by activated Cancer Vaccines
T cells with homology to the T-cell costimulatory molecule CD28. Patients may fail or resist checkpoint therapy owing to a lack of
Although CD28 and CTLA-4 are both ligands for B7-1 (CD80) and pre-existing T cell infiltrates. Therefore, vaccination and adoptive
B7-2 (CD86), they serve opposing roles in regulating T-cell acti- transfer strategies to first induce and expand the breadth of
vation. CD28 provides costimulatory signals required for T-cell endogenous T-cell responses could prove useful. Vaccines are
activation, whereas CTLA-4 negatively modulates T-cell re- composed of antigens and adjuvants. Responses to vaccination
sponses by raising the activation threshold for T-cell priming; and adjuvants involve DCs that capture and present vaccine an-
thus, CTLA-4 is likely most important during priming. PD-1 tigens, thereby facilitating differentiation of lymphocytes and
binds programmed death ligand 1 (PD-L1; a.k.a., B7-H1 or subsequent immunity (Figure 3). DCs also integrate the adjuvant
CD274) expressed by neoplastic cells, various immune cells, signals and determine the quality of induced immune responses.
mesenchymal support cells, and vascular cells; this interaction Several phase III clinical trials testing various cancer vaccine
negatively regulates T-cell activation when engaged with an platforms, including DC-based and viral-vector-based vaccines,
APC and/or effector function when engaged with other PD-L1- are ongoing. These exogenous vaccine platforms will need to be
positive cells. Indeed, binding of PD-L1 to its receptors sup- accompanied by high-throughput genomics to incorporate
presses T-cell migration and proliferation and restricts cancer personalized cancer-specific mutations and candidate peptide
cell killing (Topalian et al., 2015); thus, PD-1 is important in regu- antigens. Indeed, proof-of-concept trials in patients with
lating effector functions after CD8+ T cells are activated. PD-1 advanced melanoma demonstrated that naturally occurring
and CTLA-4 regulate distinct phases of T-cell differentiation neo-antigen-specific immunity was enhanced by vaccination
and function, and their inhibition might need to be optimally with DCs loaded with patient-specific peptides (Carreno et al.,
phased for maximum efficacy. This concept needs to be incor- 2015). Another concept is endogenous vaccination based on
porated in the next generation of clinical trials of combination antigen released upon standard CTX/RT or oncolytic viruses
therapy regimens, especially when combined with vaccines or (i.e., viruses that preferentially infect and kill cancer cells); this,
agents that reprogram myeloid cells to foster a TH1-type activa- however, requires effective antigen presentation to generate
tion state. therapeutic T-cell immunity (Palucka and Banchereau, 2013).
Indeed, combination therapies targeting the two checkpoints, DCs are skewed by tumors toward pro-tumor immunity; thus, re-
i.e., CTLA-4 and PD-1, further increase progression-free survival programming to foster TH1-skewed mature functionality in vivo is
in patients with metastatic melanoma (Larkin et al., 2015); how- critical for success of endogenous vaccination. Our understand-
ever, in other cancers, these responses are present in fewer ing of cancer genomics, the biology of antigen presentation, and
patients. Resolving the natural and acquired resistance to T-cell biology has advanced impressively and continues to in-
checkpoint inhibition therapy represents the next frontier in basic crease; this knowledge will feed into the development of the
research and clinical development (Figure 4). As the effector next-generation cancer vaccines that, combined with check-
arm of checkpoint inhibition, T cells could underpin the major point inhibitors, hold promise for improving outcomes for cancer
resistance mechanisms for checkpoint blockade. Thus, non-re- patients (Figure 4).
sponding patients might actually lack naturally occurring T cells DC vaccines can be exploited as anti-cancer therapeutics
with specificity against neoplasias, and/or their T cells could be through various strategies, including: non-targeted peptide/
held hostage and rendered dysfunctional in TMEs via pathways protein and nucleic-acid-based vaccines captured by DCs
other than checkpoints, such as immune-suppressive microen- in vivo; vaccines composed of antigens directly coupled to
vironments directed by TH2-activated myeloid, Treg, or Breg anti-DC-antibodies; or vaccines composed of ex-vivo-gener-
cells (Coussens et al., 2013). Links between treatment resistance ated DCs loaded with tumor antigens (Palucka and Banchereau,
and T-cell shortage are supported by recent findings that tumor- 2013). DCs are also engaged in response to complex vaccine
specific mutations generate neo-antigens that, in turn, may drive preparations such as GVAX, a tumor-cell-based vaccine in
anti-tumor responses. Indeed, whole-exome sequencing of ma- which neoplastic cells genetically modified to express GM-CSF
lignant melanomas from patients treated with CTLA-4 blockers attract and activate DCs (Le et al., 2010). Other vaccine plat-
demonstrated an association between mutational load and de- forms are based on recombinant-attenuated Listeria monocyto-
gree of clinical benefit (Snyder et al., 2015); however, in other genes (Lm), an intracellular bacterium that targets DCs in vivo
melanoma cohorts, recurrent neo-antigen peptide sequences and utilizes MHCI and II antigen-processing pathways (Le
were not found to predict responder populations (Van Allen et al., 2012), as well as intratumoral delivery of oncolytic viruses;
et al., 2015). In non-small-cell lung cancers treated with aPD-1 these can be modified to express GM-CSF to attract DCs and
mAb, higher mutation burden in tumors was associated with du- lymphocytes at the lysed tumor site (Russell et al., 2012). Finally,
rable clinical benefit and progression-free survival (Rizvi et al., pioneering studies from Ralph Steinman and Michel Nussenz-
2015). Neo-antigens arising as products of somatic mutations weig demonstrated the principle of targeting antigens to DCs
are not presented in the thymus; thus, they can be recognized in vivo through coupling of antigens to antibodies specific to
by the immune system as foreign antigens, similar to viral anti- DC surface receptors such as DEC205 or DCIR (Bonifaz et al.,
gens or organ transplants, because the T cells have not been 2002; Soares et al., 2007). Importantly, in the absence of adju-
eliminated or tolerized. These concepts were put forward early vants, targeting antigens to DEC205+ DCs in vivo induces anti-
(Srivastava, 2000) but validated only recently (Schumacher and gen-specific tolerance (Hawiger et al., 2001). Administration of
Schreiber, 2015), thanks to the availability of massively parallel these complex vaccines with DC activators such as TLR3,
sequencing. TLR7-8, or CD40 agonists enables maturation of DCs and

consequent establishment of immunity rather than tolerance Taxanes, broadly used microtubule inhibitors, and vinca alka-
(Bonifaz et al., 2002; Soares et al., 2007). It remains to be seen loids promote polyploidization due to mitotic interference, thus
which vaccine platform will be most effective at priming and leading to endoplasmic reticulum stress responses favoring cal-
boosting T cells in patients; this clearly represents the next fron- reticulin exposure and immune-mediated elimination (Senovilla
tier in research. et al., 2012). Clinically, docataxel, vinorelbine, and cisplatin all
lead to increased abundance of circulating CTLs and decreased
T-Cell-Dependent Nature of Cytotoxic and Targeted presence of Treg cells and immature myeloid cells harboring
Therapy T-cell-suppressive activity; this latter effect is also shared by
Cancer medicine evolved largely based on the principle that gemcitabine, a common CTX for pancreatic adenocarcinomas,
rapidly proliferating malignant cells can be eradicated by cyto- and 5-fluoruracil. Interestingly, paclitaxel is also a TLR4 ligand
toxic regimens (CTX or RT) or by targeted drugs attacking attri- and thus enhances T-cell priming by DCs (Pfannenstiel et al.,
butes of mutationally corrupted cells. As discussed above, the 2010).
recent advent and remarkable efficacy of immune-checkpoint Cyclophosphamide also provokes relocalization of intestinal
inhibitors revealed the clinical potential of harnessing endoge- gram-positive bacteria to secondary lymphoid organs, resulting
nous mechanisms of anti-tumor immunity in tumors harboring in generation of TH17 cells secreting IL-17 and IFNg that promote
significant mutational burdens. Upon reflection, however, it is anti-tumor immune responses (Viaud et al., 2013). In murine tu-
appreciated that conventional cytotoxic approaches modulate mor models, therapies targeting TH2-based programs (e.g.,
the composition and functional bioactivity of intratumoral leuko- CSF1R or RON antagonists, BTK or SYK inhibitors, B-cell deple-
cytes, in addition to effects on neoplastic cells (Galluzzi et al., tion, aIL-4 or aIL-13 mAbs) enhance efficacy of either CTX or RT
2015). Furthermore, in some scenarios, the efficacy of neoadju- by T-cell-dependent mechanisms (Ruffell and Coussens, 2015).
vant CTX correlates with increased presence of intratumoral im- Perhaps the most compelling evidence is that provided by recent
mune-effector T cells (Galluzzi et al., 2015). These correlations data revealing that immune-checkpoint blockade, when com-
are not limited to cytotoxic regimens—the tyrosine kinase inhib- bined with CTX, improves overall survival in several cancer types
itor imatinib also leads to increased presence of CTLs and NK beyond CTX alone (Topalian et al., 2015). The ability of these
cells in gastrointestinal tumors in a manner that correlates with agents to activate adaptive stress-response pathways and
disease outcome (Kroemer et al., 2013), while efficacy against send danger signals operative as immunologic adjuvants inher-
chronic myelogenous leukemia can be reversed by co-adminis- ently increases the antigenicity of tumors even when mutational
tration of type I IFN (Galluzzi et al., 2015). A recent study in breast burden is low. These untoward effects can be capitalized upon to
cancer also revealed that efficacy of transtuzumab emtansine is improve outcomes for individuals with cancer.
linked to elicitation of anti-tumor immune responses (Müller
et al., 2015). A Role for the Microbiome in Regulating Systemic
Malignant cells can emit danger signals, albeit distinct from Cancer Risk and Response to Therapy
those of normal tissue, that are sensed by immune cells and If the precision medicine equation wasn’t sufficiently compli-
thus are antigenic. Increased antigenicity is linked to either muta- cated by neoplastic cell genomics, epigenomics, host immune
tional burden, where peptides from mutant proteins are pre- responses, and the TME, mounting evidence points to an addi-
sented by MHC molecules (Gubin et al., 2014), or to ectopic tional consideration when attempting to stratify patients and pre-
expression of cancer testis or oncofetal antigens typically only dict therapy response: the host microbiome. The context and
expressed during embryonic or fetal development (Whitehurst, composition of common microorganisms living in the gut not
2014). The increased adjuvanticity of neoplastic cells is linked only shapes local immune responses, but also regulates sys-
to metabolic stress caused by their sustained proliferation and temic immunity and thus impacts the risk of and progression to
to their ability to adapt and survive in hypoxic TMEs (Krysko malignancy and the response to anti-cancer therapies. Intra-
et al., 2012). Furthermore, preclinical data have emerged sup- abdominal infections and use of antibiotics has long been asso-
porting the proposition that tumors treated with conventional ciated with increased incidence of colorectal cancer (Wang et al.,
CTX engage antigenic and adjuvant immune-mediated mecha- 2014). In mouse models, attenuating or selectively altering the
nisms. In murine tumor models, the anti-neoplastic effects of composition of gut microorganisms influences both the inci-
anthracyclines are significantly reduced when either gd or dence and progression of cancer (Zitvogel et al., 2015). Intestinal
CD8+ T cells are depleted, but not when B or NK cells are microorganisms not only impact local risk of tumorigenesis, but
absent, DC infiltration is blocked or corrupted, immune-stimula- also influence neoplastic progression distally by altering inflam-
tory type I IFNs or IL-17 are lacking, or DAMP-mediated matory and metabolic circuitry. These experimental results
recruitment and activation of effector cells is thwarted (Kroemer correlate with epidemiologic data revealing increased incidence
et al., 2013). Cyclophosphamide, oxaliplatin, and bortezomib of breast cancer in women with significant history of antibiotic
similarly rely on immune-mediated mechanisms for their effi- use (Zitvogel et al., 2015).
cacy; these commonly used cytotoxics elicit effector cell activity Gut microbiota composition is dramatically impacted by com-
via plasma membrane exposure of calreticulin and release of the mon anti-neoplastic drugs, including RT, allogeneic stem-cell
chromatin-binding protein high-mobility group box 1 (HMGB1). transplantation, and select CTXs, notably 5-fluorouracil and iri-
This, in turn, fosters DC maturation and TLR4 and RAGE activa- notecan (Zitvogel et al., 2015). Along these lines, the gut micro-
tion (Apetoh et al., 2007), thus increasing adjuvanticity of malig- biota affects the amenability of some tumor types to therapy
nant cells. by impacting regulatory aspects of the immune response.

Examples include translocation of gut microbiota across the in- cancers represents a new frontier for biomedical research.
testinal epithelium in response to lympho-depleting irradiation Such topology has been reported by Galon and colleagues to
in which DCs are inadvertently activated, leading to altered be clinically meaningful in colorectal cancer, where a statistically
serum cytokines and improved responses to adoptively trans- significant correlation between immune cell density and patient
ferred CTLs; these beneficial effects are abated by antibiotics outcome was revealed (Galon et al., 2006). Moreover, develop-
(Paulos et al., 2007). Similarly, cyclophosphamide alters compo- ment of TLS in individuals with pancreatic adenocarcinoma
sition of gut microbiota, resulting in translocation of gram- treated with vaccines correlated with improved clinical out-
positive bacteria into secondary lymphoid organs wherein comes (Le et al., 2015). Furthermore, compared with single-re-
pathogenic TH17 and memory TH1 cells are activated; tumors gion analysis, combined analysis of the tumor core and invasive
grown in germ-free mice, or antibiotics tropic for gram-positive margins improved the accuracy of survival prediction in different
bacteria exhibit reduced TH17 responses and tumor resistance patient groups (Galon et al., 2006). These early results form the
to cyclophosphamide (Viaud et al., 2013). Antibiotic eradication basis for immune stratification of patients, or the so-called Im-
of gram-positive bacteria also impairs the efficacy of CpG-oligo- munoscore, and its coordinated assessment in the clinic (As-
nucleotide immunotherapy and platinum CTX by altering myeloid cierto et al., 2013). An international consortium has been initiated
cells within the TME (Iida et al., 2013). Bifidobacterium occu- to validate and promote the Immunoscore in routine clinical set-
pancy supports anti-tumor immunity against melanoma and im- tings (Ascierto et al., 2013); results of this international effort may
proves the efficacy of aPD-L1 and aCTLA-4 mAb therapy by lead to implementation of the Immunoscore as a new classifica-
altering DC activity, leading to improved antigen-specific CD8+ tion metric, designated TNM-I (TNM-Immune).
T cell function—these effects were reduced by ampicillin, Will the Immunoscore provide enough additional information
colistin, or streptomycin but were enhanced by vancomycin to prospectively predict response to therapy? Likely not. We
due to preferential enhancement of Bacteroidales colonization predict that integrating the Immunoscore with additional metrics
(Vétizou et al., 2015). These data underscore the impact of gut will be critical for guiding patient stratification and phasing of
commensals on therapeutic responsiveness. combinatorial therapies. Such metrics will include genomic
Could selectively manipulating the gut microbiota impact risk and exomic features of neoplastic cells (through sequencing
of developing cancer, limit incidence of select tumor types, of neoplastic cells themselves or cfDNA, tracking the expansion
and/or improve activity of some anti-cancer therapies? Zitvogel of tumor-specific CD4+ and CD8+ T cells, monitoring serum
and colleagues have proposed four distinct approaches for cytokine fluxes, and evaluating the composition and health of
manipulating the gut microbiota to boost cancer therapy: (1) pref- commensal bacteria (Figure 4). Serum cytokines have long
erential use of antibiotics selective for untoward bacterial been enigmatic due to their labile nature and the detection
species; (2) increased use of probiotics; (3) increased use of limitations of conventional methodologies. That said, serum
prebiotics to stimulate healthy gut colonization; and (4) use of biomarker signatures are now able to discern asymptomatic
postbiotics, nonviable products of microbiota that exert biolog- early stage pancreatic cancer from healthy controls with 96%
ical activities in hosts (Zitvogel et al., 2015). Prospective stool accuracy (Ghatnekar et al., 2013) and can be used to monitor
analysis and monitoring in cancer patients receiving therapy the pharmacodynamics of CSF1R-targeted therapies (Butowski
will surely reveal biomarkers that, if harnessed, could improve et al., 2015). Moreover, transcriptional profiling of blood mono-
patient stratification and/or support new microbiota-based stra- cytes in renal cell carcinoma identifies biomarkers correlating
tegies for boosting therapeutic responsiveness, e.g., fecal trans- with tumor staging (Chittezhath et al., 2014), and mRNA
plant of beneficial species. sequencing of tumor-educated blood platelets distinguishes
cancer patients from healthy individuals with 96% accuracy
Multi-Modal Tumor and Systemic Biomarkers for (Best et al., 2015). Thus, multi-modal functional diagnostic stra-
Stratification and Resistance Monitoring tegies integrating the tumor, host, and commensals will likely
A major clinical goal is to understand the multi-modal tissue- forge the advent of next-generation precision bioinformatics to
based and systemic pathways regulating therapy responses so match patients with appropriately combined and phased anti-
as to minimize resistance and maximize efficacy of cancer med- cancer therapies.
icine (Figure 4). Whether therapies target tumor-intrinsic path-
ways, host pathways, or commensal microbiota, it is clear that Oncoimmunology Treatment Paradigm
understanding non-genomic mechanisms of resistance from Future immunotherapies will be based on cycles of interventions
an integrated standpoint is critical. designed to boost and modulate anti-cancer immunity (Figure 4).
Understanding which immune cell types are present in and Indeed, as we rediscover and refine the fundamental principles
around a tumor currently provides invaluable retrospective infor- of tumor immunology, it is increasingly clear that curing cancer
mation regarding tumor ecology and/or tumor response to might not be a realistic goal. Rather, aiming for a continuum of
therapy. However, we must improve our ability to integrate infor- treatment cycles designed and based on mechanistic in vivo
mation on not only the complexity of leukocytes in tumors, but studies and in-depth analysis of each patient’s tumor will be
also their geography in tumor nests and stroma. Immune cells necessary for optimizing outcomes. Clinical trials with check-
are scattered in tumor core and within tumor stroma, in invasive point inhibitors teach us that in situ immune infiltration is critical
margins, and in organized lymphoid structures often distant from for tumor regression; however, not all immune infiltrates are
neoplastic cells. Investigating the mechanisms governing forma- equal, and as discussed throughout this article, the quality of im-
tion of tertiary lymphoid structures (TLSs) found in numerous mune response is a critical factor for therapeutic success. This,

in turn, is determined by underlying inflammation, which we Apetoh, L., Ghiringhelli, F., Tesniere, A., Criollo, A., Ortiz, C., Lidereau, R., Mari-
assert must become a staging parameter, along with classical ette, C., Chaput, N., Mira, J.P., Delaloge, S., et al. (2007). The interaction be-
tween HMGB1 and TLR4 dictates the outcome of anticancer chemotherapy
pathology-based schemas and the Immunoscore. It will also
and radiotherapy. Immunol. Rev. 220, 47–59.
need to be established to what extent inflammation, which
Appay, V., Douek, D.C., and Price, D.A. (2008). CD8+ T cell efficacy in vacci-
clearly plays a role in epithelial tumors, impacts other tumor
nation and disease. Nat. Med. 14, 623–628.
types, e.g., melanoma or sarcomas. Additional parameters
Ascierto, P.A., Capone, M., Urba, W.J., Bifulco, C.B., Botti, G., Lugli, A., Mar-
pertain to the specificity of infiltrating T cells against cancer
incola, F.M., Ciliberto, G., Galon, J., and Fox, B.A. (2013). The additional facet
antigens, as again, the infiltrate with passenger T cells might
of immunoscore: immunoprofiling as a possible predictive tool for cancer
not be therapeutically useful and should be tested. Eventually treatment. J. Transl. Med. 11, 54.
all patients will be treated with checkpoint inhibitors, either
Balkwill, F., Charles, K.A., and Mantovani, A. (2005). Smoldering and polarized
directly or after interventions targeting inflammation, by vaccina- inflammation in the initiation and promotion of malignant disease. Cancer Cell
tion to boost T-cell repertoires or by adoptive T-cell transfer. The 7, 211–217.
majority of patients will subsequently develop acquired resis- Banchereau, J., Briere, F., Caux, C., Davoust, J., Lebecque, S., Liu, Y.J., Pu-
tance followed by immune escape; this will lead to the next lendran, B., and Palucka, K. (2000). Immunobiology of dendritic cells. Annu.
cycle of treatments incorporating multi-modal biomarkers (e.g., Rev. Immunol. 18, 767–811.
based on microbiome phenotype, cfDNA, circulating cytokine Beck, A.H., Espinosa, I., Edris, B., Li, R., Montgomery, K., Zhu, S., Varma, S.,
levels) and perhaps NK cells recognizing loss of MHC class I Marinelli, R.J., van de Rijn, M., and West, R.B. (2009). The macrophage colony-
by neoplastic cells, thus rendering them invisible to T cells. stimulating factor 1 response signature in breast carcinoma. Clin. Cancer Res.
Cytotoxic treatments, such as with NK cells, standard CTX/RT, 15, 778–787.
or oncolytic viruses, will release neo-antigens that can be Best, M.G., Sol, N., Kooi, I., Tannous, J., Westerman, B.A., Rustenburg, F.,
used to generate the next round of effector T cells. To this Schellen, P., Verschueren, H., Post, E., Koster, J., et al. (2015). RNA-Seq of
Tumor-Educated Platelets Enables Blood-Based Pan-Cancer, Multiclass,
latter point, we must fully understand the rules of T-cell priming
and Molecular Pathway Cancer Diagnostics. Cancer Cell 28, 666–676.
in vivo in humans, identify the most effective ways to utilize
DCs for priming, and develop strategies for mobilizing the Bongartz, T., Sutton, A.J., Sweeting, M.J., Buchan, I., Matteson, E.L., and
Montori, V. (2006). Anti-TNF antibody therapy in rheumatoid arthritis and the
naive T-cell repertoire from the thymus in adults (Sportès et al.,
risk of serious infections and malignancies: systematic review and meta-anal-
2008). In later rounds of therapy, the scope of neo-antigens will ysis of rare harmful effects in randomized controlled trials. JAMA 295, 2275–
likely be broadened as, in addition to somatic mutations, neo- 2285.
antigens can be generated via epigenetic and post-translational Bonifaz, L., Bonnyay, D., Mahnke, K., Rivera, M., Nussenzweig, M.C., and
regulation. Last but not least, the role of Tregs, so well estab- Steinman, R.M. (2002). Efficient targeting of protein antigen to the dendritic
lished in murine cancer, will need to be redefined in humans. cell receptor DEC-205 in the steady state leads to antigen presentation on ma-
Resolving all of these challenges will surely keep us busy for a jor histocompatibility complex class I products and peripheral CD8+ T cell
long while. tolerance. J. Exp. Med. 196, 1627–1638.
Butowski, N., Colman, H., De Groot, J.F., Omuro, A.M., Nayak, L., Wen, P.Y.,
ACKNOWLEDGMENTS Cloughesy, T.F., Marimuthu, A., Haidar, S., Perry, A., et al. (2015). Orally
administered colony stimulating factor 1 receptor inhibitor PLX3397 in recur-
The authors thank members of the Palucka and Coussens laboratories for rent glioblastoma: an Ivy Foundation Early Phase Clinical Trials Consortium
critical insight and discussions; Amanda Lund and Jeffrey Nolz for critical phase II study. Neuro. Oncol. Published online October 8, 2015. http://dx.
feedback; Anna Lisa Lucido for help with editing the manuscript; and all doi.org/10.1093/neuonc/nov245.
authors contributing to studies discussed herein but not mentioned due Campisi, J., Andersen, J.K., Kapahi, P., and Melov, S. (2011). Cellular senes-
to space consideration. A.K.P. thanks Jacques Banchereau, a long-time cence: a link between cancer and age-related degenerative disease? Semin.
collaborator. The authors acknowledge support from the NIH/NCI and Susan Cancer Biol. 21, 354–359.
B. Komen Foundation (A.K.P. and L.M.C.), Baylor Research Foundation
Carreno, B.M., Magrini, V., Becker-Hapak, M., Kaabinejadian, S., Hundal, J.,
(A.K.P.), DOD BCRP Era of Hope Scholar Expansion Award, Breast Cancer
Petti, A.A., Ly, A., Lie, W.R., Hildebrand, W.H., Mardis, E.R., and Linette,
Research Foundation, a Stand Up To Cancer–Lustgarten Foundation
G.P. (2015). Cancer immunotherapy. A dendritic cell vaccine increases the
Pancreatic Cancer Convergence Dream Team Translational Research Grant
breadth and diversity of melanoma neoantigen-specific T cells. Science 348,
(SU2C-AACR-DT14-14), and Brenden-Colson Center for Pancreatic Care
803–808.
(to L.M.C.).
Carretero, R., Sektioglu, I.M., Garbi, N., Salgado, O.C., Beckhove, P., and
Hämmerling, G.J. (2015). Eosinophils orchestrate cancer rejection by normal-
REFERENCES
izing tumor vessels and enhancing infiltration of CD8(+) T cells. Nat. Immunol.
16, 609–617.
Affara, N.I., Ruffell, B., Medler, T.R., Gunderson, A.J., Johansson, M., Born-
stein, S., Bergsland, E., Steinhoff, M., Li, Y., Gong, Q., et al. (2014). B cells Chen, D.S., and Mellman, I. (2013). Oncology meets immunology: the cancer-
regulate macrophage phenotype and response to chemotherapy in squamous immunity cycle. Immunity 39, 1–10.
carcinomas. Cancer Cell 25, 809–821. Chittezhath, M., Dhillon, M.K., Lim, J.Y., Laoui, D., Shalova, I.N., Teo, Y.L.,
Aharinejad, S., Salama, M., Paulus, P., Zins, K., Berger, A., and Singer, C.F. Chen, J., Kamaraj, R., Raman, L., Lum, J., et al. (2014). Molecular profiling re-
(2013). Elevated CSF1 serum concentration predicts poor overall survival in veals a tumor-promoting phenotype of monocytes and macrophages in hu-
women with early breast cancer. Endocr. Relat. Cancer 20, 777–783. man cancer progression. Immunity 41, 815–829.
Andreu, P., Johansson, M., Affara, N.I., Pucci, F., Tan, T., Junankar, S., Korets, Coffelt, S.B., Kersten, K., Doornebal, C.W., Weiden, J., Vrijland, K., Hau, C.S.,
L., Lam, J., Tawfik, D., DeNardo, D.G., et al. (2010). FcRgamma activation reg- Verstegen, N.J., Ciampricotti, M., Hawinkels, L.J., Jonkers, J., et al. (2015). 17-
ulates inflammation-associated squamous carcinogenesis. Cancer Cell 17, producing gammadelta T cells and neutrophils conspire to promote breast
121–134. cancer metastasis (IL: Nature).

Coussens, L.M., Raymond, W.W., Bergers, G., Laig-Webster, M., Behrendt- Hanna, R.N., Cekic, C., Sag, D., Tacke, R., Thomas, G.D., Nowyhed, H., Herr-
sen, O., Werb, Z., Caughey, G.H., and Hanahan, D. (1999). Inflammatory ley, E., Rasquinha, N., McArdle, S., Wu, R., et al. (2015). Patrolling monocytes
mast cells up-regulate angiogenesis during squamous epithelial carcinogen- control tumor metastasis to the lung. Science 350, 985–990.
esis. Genes Dev. 13, 1382–1397. Hawiger, D., Inaba, K., Dorsett, Y., Guo, M., Mahnke, K., Rivera, M., Ravetch,
Coussens, L.M., Zitvogel, L., and Palucka, A.K. (2013). Neutralizing tumor-pro- J.V., Steinman, R.M., and Nussenzweig, M.C. (2001). Dendritic cells induce
moting chronic inflammation: a magic bullet? Science 339, 286–291. peripheral T cell unresponsiveness under steady state conditions in vivo.
J. Exp. Med. 194, 769–779.
Davis, B.P., and Rothenberg, M.E. (2014). Eosinophils and cancer. Cancer
Immunol. Res. 2, 1–8. Heiser, L.M., Sadanandam, A., Kuo, W.L., Benz, S.C., Goldstein, T.C., Ng, S.,
Gibb, W.J., Wang, N.J., Ziyad, S., Tong, F., et al. (2012). Subtype and pathway
de Visser, K.E., Eichten, A., and Coussens, L.M. (2006). Paradoxical roles of
specific responses to anticancer compounds in breast cancer. Proc. Natl.
the immune system during cancer development. Nat. Rev. Cancer 6, 24–37.
Acad. Sci. USA 109, 2724–2729.
DeNardo, D.G., Barreto, J.B., Andreu, P., Vasquez, L., Tawfik, D., Kolhatkar, Hendriks, R.W., Yuvaraj, S., and Kil, L.P. (2014). Targeting Bruton’s tyrosine ki-
N., and Coussens, L.M. (2009). CD4(+) T cells regulate pulmonary metastasis nase in B cell malignancies. Nat. Rev. Cancer 14, 219–232.
of mammary carcinomas by enhancing protumor properties of macrophages.
Iida, N., Dzutsev, A., Stewart, C.A., Smith, L., Bouladoux, N., Weingarten, R.A.,
Cancer Cell 16, 91–102.
Molina, D.A., Salcedo, R., Back, T., Cramer, S., et al. (2013). Commensal bac-
DeNardo, D.G., Brennan, D.J., Rexhepaj, E., Ruffell, B., Shiao, S.L., Madden, teria control cancer response to therapy by modulating the tumor microenvi-
S.F., Gallagher, W.M., Wadhwani, N., Keil, S.D., Junaid, S.A., et al. (2011). ronment. Science 342, 967–970.
Leukocyte complexity predicts breast cancer survival and functionally regu-
Inoue, S., Leitner, W.W., Golding, B., and Scott, D. (2006). Inhibitory effects of
lates response to chemotherapy. Cancer Discov. 1, 54–67.
B cells on antitumor immunity. Cancer Res. 66, 7741–7747.
Druker, B.J., Guilhot, F., O’Brien, S.G., Gathmann, I., Kantarjian, H., Gatter- Joffre, O.P., Segura, E., Savina, A., and Amigorena, S. (2012). Cross-presen-
mann, N., Deininger, M.W., Silver, R.T., Goldman, J.M., Stone, R.M., et al.; tation by dendritic cells. Nat. Rev. Immunol. 12, 557–569.
IRIS Investigators (2006). Five-year follow-up of patients receiving imatinib
Karsten, C.M., and Köhl, J. (2012). The immunoglobulin, IgG Fc receptor and
for chronic myeloid leukemia. N. Engl. J. Med. 355, 2408–2417.
complement triangle in autoimmune diseases. Immunobiology 217, 1067–
Dunn, G.P., Old, L.J., and Schreiber, R.D. (2004). The three Es of cancer immu- 1079.
noediting. Annu. Rev. Immunol. 22, 329–360.
Kerbel, R. (2011). Anti-angiogenesis in cancer; met and unmet goals - an inter-
Eyob, H., Ekiz, H.A., Derose, Y.S., Waltz, S.E., Williams, M.A., and Welm, A.L. view with Robert Kerbel by Francesco Bertolini. Int. J. Dev. Biol. 55, 395–398.
(2013). Inhibition of ron kinase blocks conversion of micrometastases to overt Kitamura, T., Qian, B.Z., and Pollard, J.W. (2015). Immune cell promotion of
metastases by boosting antitumor immunity. Cancer Discov. 3, 751–760. metastasis. Nat. Rev. Immunol. 15, 73–86.
Fehérvari, Z., and Sakaguchi, S. (2004). CD4+ Tregs and immune control. Klebanoff, C.A., Acquavella, N., Yu, Z., and Restifo, N.P. (2011). Therapeutic
J. Clin. Invest. 114, 1209–1217. cancer vaccines: are we there yet? Immunol. Rev. 239, 27–44.
Fridlender, Z.G., and Albelda, S.M. (2012). Tumor-associated neutrophils: Klechevsky, E., Morita, R., Liu, M., Cao, Y., Coquery, S., Thompson-Snipes, L.,
friend or foe? Carcinogenesis 33, 949–955. Briere, F., Chaussabel, D., Zurawski, G., Palucka, A.K., et al. (2008). Functional
Gabrilovich, D.I., Ostrand-Rosenberg, S., and Bronte, V. (2012). Coordinated specializations of human epidermal Langerhans cells and CD14+ dermal den-
regulation of myeloid cells by tumours. Nat. Rev. Immunol. 12, 253–268. dritic cells. Immunity 29, 497–510.
Komohara, Y., Jinushi, M., and Takeya, M. (2014). Clinical significance of
Galluzzi, L., Buqué, A., Kepp, O., Zitvogel, L., and Kroemer, G. (2015). Immu-
macrophage heterogeneity in human malignant tumors. Cancer Sci. 105, 1–8.
nological Effects of Conventional Chemotherapy and Targeted Anticancer
Agents. Cancer Cell 28, 690–714. Kroemer, G., Galluzzi, L., Kepp, O., and Zitvogel, L. (2013). Immunogenic cell
death in cancer therapy. Annu. Rev. Immunol. 31, 51–72.
Galon, J., Costes, A., Sanchez-Cabo, F., Kirilovsky, A., Mlecnik, B., Lagorce-
Pagès, C., Tosolini, M., Camus, M., Berger, A., Wind, P., et al. (2006). Type, Krysko, D.V., Garg, A.D., Kaczmarek, A., Krysko, O., Agostinis, P., and Vande-
density, and location of immune cells within human colorectal tumors predict nabeele, P. (2012). Immunogenic cell death and DAMPs in cancer therapy.
clinical outcome. Science 313, 1960–1964. Nat. Rev. Cancer 12, 860–875.
Lanzavecchia, A., and Sallusto, F. (2001). The instructive role of dendritic cells
Geissmann, F., Jung, S., and Littman, D.R. (2003). Blood monocytes consist of
on T cell responses: lineages, plasticity and kinetics. Curr. Opin. Immunol. 13,
two principal subsets with distinct migratory properties. Immunity 19, 71–82.
291–298.
Ghatnekar, O., Andersson, R., Svensson, M., Persson, U., Ringdahl, U., Zeilon,
Larkin, J., Chiarion-Sileni, V., Gonzalez, R., Grob, J.J., Cowey, C.L., Lao, C.D.,
P., and Borrebaeck, C.A. (2013). Modelling the benefits of early diagnosis of
Schadendorf, D., Dummer, R., Smylie, M., Rutkowski, P., et al. (2015). Com-
pancreatic cancer using a biomarker signature. Int. J. Cancer 133, 2392–2397.
bined Nivolumab and Ipilimumab or Monotherapy in Untreated Melanoma.
Gocheva, V., Wang, H.W., Gadea, B.B., Shree, T., Hunter, K.E., Garfall, A.L., N. Engl. J. Med. 373, 23–34.
Berman, T., and Joyce, J.A. (2010). IL-4 induces cathepsin protease activity Le, D.T., Pardoll, D.M., and Jaffee, E.M. (2010). Cellular vaccine approaches.
in tumor-associated macrophages to promote cancer growth and invasion. Cancer J. 16, 304–310.
Genes Dev. 24, 241–255.
Le, D.T., Dubenksy, T.W., Jr., and Brockstedt, D.G. (2012). Clinical develop-
Gubin, M.M., Zhang, X., Schuster, H., Caron, E., Ward, J.P., Noguchi, T., Iva- ment of Listeria monocytogenes-based immunotherapies. Semin. Oncol. 39,
nova, Y., Hundal, J., Arthur, C.D., Krebber, W.J., et al. (2014). Checkpoint 311–322.
blockade cancer immunotherapy targets tumour-specific mutant antigens.
Le, D.T., Wang-Gillam, A., Picozzi, V., Greten, T.F., Crocenzi, T., Springett, G.,
Nature 515, 577–581.
Morse, M., Zeh, H., Cohen, D., Fine, R.L., et al. (2015). Safety and survival with
Gunderson, A.J., Kaneda, M.M., Tsujikawa, T., Nguyen, A.V., Affara, N.I., Ruf- GVAX pancreas prime and Listeria Monocytogenes-expressing mesothelin
fell, B., Gorjestani, S., Liudahl, S.M., Truitt, M., Olson, P., et al. (2016). Bruton’s (CRS-207) boost vaccines for metastatic pancreatic cancer. J. Clin. Oncol.
Tyrosine Kinase (BTK)-dependent immune cell crosstalk drives pancreas can- 33, 1325–1333.
cer. Cancer Discov. Published online December 29, 2015. http://dx.doi.org/
Le Floc’h, A., Jalil, A., Vergnon, I., Le Maux Chansac, B., Lazar, V., Bismuth, G.,
10.1158/2159-8290.
Chouaib, S., and Mami-Chouaib, F. (2007). Alpha E beta 7 integrin interaction
Hanahan, D., and Coussens, L.M. (2012). Accessories to the crime: functions with E-cadherin promotes antitumor CTL activity by triggering lytic granule po-
of cells recruited to the tumor microenvironment. Cancer Cell 21, 309–322. larization and exocytosis. J. Exp. Med. 204, 559–570.

Lin, E.Y., Nguyen, A.V., Russell, R.G., and Pollard, J.W. (2001). Colony-stimu- Ruffell, B., Chang-Strachan, D., Chan, V., Rosenbusch, A., Ho, C.M., Pryer, N.,
lating factor 1 promotes progression of mammary tumors to malignancy. Daniel, D., Hwang, E.S., Rugo, H.S., and Coussens, L.M. (2014). Macrophage
J. Exp. Med. 193, 727–740. IL-10 blocks CD8+ T cell-dependent responses to chemotherapy by sup-
Marichal, T., Tsai, M., and Galli, S.J. (2013). Mast cells: potential positive and pressing IL-12 expression in intratumoral dendritic cells. Cancer Cell 26,
negative roles in tumor biology. Cancer Immunol. Res. 1, 269–279. 623–637.
Massó-Vallés, D., Jauset, T., Serrano, E., Sodir, N.M., Pedersen, K., Affara, Russell, S.J., Peng, K.W., and Bell, J.C. (2012). Oncolytic virotherapy. Nat. Bio-
N.I., Whitfield, J.R., Beaulieu, M.E., Evan, G.I., Elias, L., et al. (2015). Ibrutinib technol. 30, 658–670.
exerts potent antifibrotic and antitumor activities in mouse models of pancre- Sagiv-Barfi, I., Kohrt, H.E., Czerwinski, D.K., Ng, P.P., Chang, B.Y., and Levy,
atic adenocarcinoma. Cancer Res. 75, 1675–1681. R. (2015). Therapeutic antitumor immunity by checkpoint blockade is
Merad, M., Sathe, P., Helft, J., Miller, J., and Mortha, A. (2013). The dendritic enhanced by ibrutinib, an inhibitor of both BTK and ITK. Proc. Natl. Acad.
cell lineage: ontogeny and function of dendritic cells and their subsets in the Sci. USA 112, E966–E972.
steady state and the inflamed setting. Annu. Rev. Immunol. 31, 563–604. Sandoval, F., Terme, M., Nizard, M., Badoual, C., Bureau, M.F., Freyburger, L.,
Metz, M., Grimbaldeston, M.A., Nakae, S., Piliponsky, A.M., Tsai, M., and Galli, Clement, O., Marcheteau, E., Gey, A., Fraisse, G., et al. (2013). Mucosal
S.J. (2007). Mast cells in the promotion and limitation of chronic inflammation. imprinting of vaccine-induced CD8+ T cells is crucial to inhibit the growth of
Immunol. Rev. 217, 304–328. mucosal tumors. Sci. Transl. Med. 5, 172ra20.
Mitchem, J.B., Brennan, D.J., Knolhoff, B.L., Belt, B.A., Zhu, Y., Sanford, D.E., Schumacher, T.N., and Schreiber, R.D. (2015). Neoantigens in cancer immu-
Belaygorod, L., Carpenter, D., Collins, L., Piwnica-Worms, D., et al. (2013). notherapy. Science 348, 69–74.
Targeting tumor-infiltrating macrophages decreases tumor-initiating cells, re- Senovilla, L., Vitale, I., Martins, I., Tailler, M., Pailleret, C., Michaud, M., Gal-
lieves immunosuppression, and improves chemotherapeutic responses. Can- luzzi, L., Adjemian, S., Kepp, O., Niso-Santano, M., et al. (2012). An immuno-
cer Res. 73, 1128–1141. surveillance mechanism controls cancer cell ploidy. Science 337, 1678–1684.
Mueller, S.N., Gebhardt, T., Carbone, F.R., and Heath, W.R. (2012). Memory Shepard, H.M., Lewis, G.D., Sarup, J.C., Fendly, B.M., Maneval, D., Mordenti,
T cell subsets, migration patterns, and tissue residence. Annu. Rev. Immunol. J., Figari, I., Kotts, C.E., Palladino, M.A., Jr., Ullrich, A., et al. (1991). Mono-
31, 137–161. clonal antibody therapy of human cancer: taking the HER2 protooncogene
Müller, P., Kreuzaler, M., Khan, T., Thommen, D.S., Martin, K., Glatz, K., Savic, to the clinic. J. Clin. Immunol. 11, 117–127.
S., Harbeck, N., Nitz, U., Gluz, O., et al. (2015). Trastuzumab emtansine Sheridan, B.S., and Lefrançois, L. (2011). Regional and mucosal memory
(T-DM1) renders HER2+ breast cancer highly susceptible to CTLA-4/PD-1 T cells. Nat. Immunol. 12, 485–491.
blockade. Sci. Transl. Med. 7, 315ra188.
Shiao, S.L., Ruffell, B., DeNardo, D.G., Faddegon, B.A., Park, C.C., and Cous-
Palucka, K., and Banchereau, J. (2013). Dendritic-cell-based therapeutic can- sens, L.M. (2015). TH2-Polarized CD4(+) T Cells and Macrophages Limit Effi-
cer vaccines. Immunity 39, 38–48. cacy of Radiotherapy. Cancer Immunol. Res. 3, 518–525.
Pardoll, D.M. (2012). The blockade of immune checkpoints in cancer immuno- Sippel, T.R., White, J., Nag, K., Tsvankin, V., Klaassen, M., Kleinschmidt-
therapy. Nat. Rev. Cancer 12, 252–264. DeMasters, B.K., and Waziri, A. (2011). Neutrophil degranulation and immuno-
Paulos, C.M., Wrzesinski, C., Kaiser, A., Hinrichs, C.S., Chieppa, M., Cassard, suppression in patients with GBM: restoration of cellular immune function by
L., Palmer, D.C., Boni, A., Muranski, P., Yu, Z., et al. (2007). Microbial translo- targeting arginase I. Clin. Cancer Res. 17, 6992–7002.
cation augments the function of adoptively transferred self/tumor-specific Smith, C.I., Islam, T.C., Mattsson, P.T., Mohamed, A.J., Nore, B.F., and Vihi-
CD8+ T cells via TLR4 signaling. J. Clin. Invest. 117, 2197–2204. nen, M. (2001). The Tec family of cytoplasmic tyrosine kinases: mammalian
Pfannenstiel, L.W., Lam, S.S., Emens, L.A., Jaffee, E.M., and Armstrong, T.D. Btk, Bmx, Itk, Tec, Txk and homologs in other species. BioEssays 23,
(2010). Paclitaxel enhances early dendritic cell maturation and function 436–446.
through TLR4 signaling in mice. Cell. Immunol. 263, 79–87. Snyder, A., Wolchok, J.D., and Chan, T.A. (2015). Genetic basis for clinical
Plaks, V., Boldajipour, B., Linnemann, J.R., Nguyen, N.H., Kersten, K., Wolf, Y., response to CTLA-4 blockade. N. Engl. J. Med. 372, 783.
Casbon, A.J., Kong, N., van den Bijgaart, R.J., Sheppard, D., et al. (2015). Soares, H., Waechter, H., Glaichenhaus, N., Mougneau, E., Yagita, H., Mize-
Adaptive Immune Regulation of Mammary Postnatal Organogenesis. Dev. nina, O., Dudziak, D., Nussenzweig, M.C., and Steinman, R.M. (2007). A sub-
Cell 34, 493–504. set of dendritic cells induces CD4+ T cells to produce IFN-gamma by an IL-12-
Postow, M.A., Callahan, M.K., and Wolchok, J.D. (2015). Immune Checkpoint independent but CD70-dependent mechanism in vivo. J. Exp. Med. 204,
Blockade in Cancer Therapy. J. Clin. Oncol. 33, 1974–1982. 1095–1106.
Pulendran, B. (2015). The varieties of immunological experience: of pathogens, Soucek, L., Buggy, J.J., Kortlever, R., Adimoolam, S., Monclús, H.A., Allende,
stress, and dendritic cells. Annu. Rev. Immunol. 33, 563–606. M.T., Swigart, L.B., and Evan, G.I. (2011). Modeling pharmacological inhibition
Pylayeva-Gupta, Y., Lee, K.E., Hajdu, C.H., Miller, G., and Bar-Sagi, D. (2012). of mast cell degranulation as a therapy for insulinoma. Neoplasia 13, 1093–
Oncogenic Kras-induced GM-CSF production promotes the development of 1100.
pancreatic neoplasia. Cancer Cell 21, 836–847. Sportès, C., Hakim, F.T., Memon, S.A., Zhang, H., Chua, K.S., Brown, M.R.,
Qian, B.Z., Li, J., Zhang, H., Kitamura, T., Zhang, J., Campion, L.R., Kaiser, Fleisher, T.A., Krumlauf, M.C., Babb, R.R., Chow, C.K., et al. (2008). Adminis-
E.A., Snyder, L.A., and Pollard, J.W. (2011). CCL2 recruits inflammatory mono- tration of rhIL-7 in humans increases in vivo TCR repertoire diversity by prefer-
cytes to facilitate breast-tumour metastasis. Nature 475, 222–225. ential expansion of naive T cell subsets. J. Exp. Med. 205, 1701–1714.
Ries, C.H., Cannarile, M.A., Hoves, S., Benz, J., Wartha, K., Runza, V., Rey- Srivastava, P.K. (2000). Immunotherapy of human cancer: lessons from mice.
Giraud, F., Pradel, L.P., Feuerhake, F., Klaman, I., et al. (2014). Targeting tu- Nat. Immunol. 1, 363–366.
mor-associated macrophages with anti-CSF-1R antibody reveals a strategy Steinman, R. (2011). Cancer therapeutics: time to swim downstream? Oncol-
for cancer therapy. Cancer Cell 25, 846–859. ogist 16, 1479–1480.
Rizvi, N.A., Hellmann, M.D., Snyder, A., Kvistborg, P., Makarov, V., Havel, J.J., Steinman, R.M. (2012). Decisions about dendritic cells: past, present, and
Lee, W., Yuan, J., Wong, P., Ho, T.S., et al. (2015). Cancer immunology. Muta- future. Annu. Rev. Immunol. 30, 1–22.
tional landscape determines sensitivity to PD-1 blockade in non-small cell lung Stout, R.D., and Bottomly, K. (1989). Antigen-specific activation of effector
cancer. Science 348, 124–128. macrophages by IFN-gamma producing (TH1) T cell clones. Failure of IL-4-
Ruffell, B., and Coussens, L.M. (2015). Macrophages and therapeutic resis- producing (TH2) T cell clones to activate effector function in macrophages.
tance in cancer. Cancer Cell 27, 462–472. J. Immunol. 142, 760–765.

Tap, W.D., Wainberg, Z.A., Anthony, S.P., Ibrahim, P.N., Zhang, C., Healey, Wang, L., Yi, T., Kortylewski, M., Pardoll, D.M., Zeng, D., and Yu, H. (2009). IL-
J.H., Chmielowski, B., Staddon, A.P., Cohn, A.L., Shapiro, G.I., et al. (2015). 17 can promote tumor growth through an IL-6-Stat3 signaling pathway. J. Exp.
Structure-Guided Blockade of CSF1R Kinase in Tenosynovial Giant-Cell Med. 206, 1457–1464.
Tumor. N. Engl. J. Med. 373, 428–437.
Wang, J.L., Chang, C.H., Lin, J.W., Wu, L.C., Chuang, L.M., and Lai, M.S.
Thun, M.J., Henley, S.J., and Gansler, T. (2004). Inflammation and cancer: an (2014). Infection, antibiotic therapy and risk of colorectal cancer: a nationwide
epidemiological perspective. Novartis Found. Symp. 256, 6–21, discussion nested case-control study in patients with Type 2 diabetes mellitus. Int. J. Can-
22–28, 49–52, 266–269. cer 135, 956–967.
Topalian, S.L., Drake, C.G., and Pardoll, D.M. (2015). Immune checkpoint Whitehurst, A.W. (2014). Cause and consequence of cancer/testis antigen
blockade: a common denominator approach to cancer therapy. Cancer Cell activation in cancer. Annu. Rev. Pharmacol. Toxicol. 54, 251–272.
27, 450–461. Wu, S., Rhee, K.J., Albesiano, E., Rabizadeh, S., Wu, X., Yen, H.R., Huso, D.L.,
Van Allen, E.M., Miao, D., Schilling, B., Shukla, S.A., Blank, C., Zimmer, L., Brancati, F.L., Wick, E., McAllister, F., et al. (2009). A human colonic
Sucker, A., Hillen, U., Foppen, M.H., Goldinger, S.M., et al. (2015). Genomic commensal promotes colon tumorigenesis via activation of T helper type 17
correlates of response to CTLA-4 blockade in metastatic melanoma. Science T cell responses. Nat. Med. 15, 1016–1022.
350, 207–211. Yu, C.I., Becker, C., Wang, Y., Marches, F., Helft, J., Leboeuf, M., Anguiano, E.,
Vétizou, M., Pitt, J.M., Daillère, R., Lepage, P., Waldschmitt, N., Flament, C., Pourpe, S., Goller, K., Pascual, V., et al. (2013). Human CD1c+ dendritic cells
Rusakiewicz, S., Routy, B., Roberti, M.P., Duong, C.P., et al. (2015). Anticancer drive the differentiation of CD103+ CD8+ mucosal effector T cells via the cyto-
immunotherapy by CTLA-4 blockade relies on the gut microbiota. Science kine TGF-b. Immunity 38, 818–830.
350, 1079–1084. Zhu, J., and Paul, W.E. (2008). CD4 T cells: fates, functions, and faults. Blood
112, 1557–1569.
Viaud, S., Saccheri, F., Mignot, G., Yamazaki, T., Daillère, R., Hannani, D.,
Enot, D.P., Pfirschke, C., Engblom, C., Pittet, M.J., et al. (2013). The intestinal Zitvogel, L., Galluzzi, L., Viaud, S., Vétizou, M., Daillère, R., Merad, M., and
microbiota modulates the anticancer immune effects of cyclophosphamide. Kroemer, G. (2015). Cancer and the gut microbiota: an unexpected link. Sci.
Science 342, 971–976. Transl. Med. 7, 271ps1.

Leading Edge
Review
Modeling Development and Disease with Organoids

Hans Clevers1,*
1Hubrecht Institute/Royal Netherlands Academy of Arts and Sciences, Princess Maxima Centre and University Medical Centre Utrecht,
3584CT Utrecht, The Netherlands
*Correspondence: h.clevers@hubrecht.eu
Recent advances in 3D culture technology allow embryonic and adult mammalian stem cells to
exhibit their remarkable self-organizing properties, and the resulting organoids reflect key struc-
tural and functional properties of organs such as kidney, lung, gut, brain and retina. Organoid tech-
nology can therefore be used to model human organ development and various human pathologies
‘in a dish.’’ Additionally, patient-derived organoids hold promise to predict drug response in a
personalized fashion. Organoids open up new avenues for regenerative medicine and, in combina-
tion with editing technology, for gene therapy. The many potential applications of this technology
are only beginning to be explored.
In 1975, James Rheinwald and Howard Green described the first witnessed a revival of the organoid, yet in a somewhat different
long-term culture of normal human cells (Rheinwald and Green, guise: an organoid is now defined as a 3D structure grown
1975). For this, they combined freshly isolated keratinocytes from stem cells and consisting of organ-specific cell types that
with irradiated mouse 3T3 fibroblasts, established in the same self-organizes through cell sorting and spatially restricted line-
lab years earlier. As in stratified skin, cell division was confined age commitment (after Eiraku and Sasai, 2012; Lancaster and
to the basal layer of the growing clones, while superficial Knoblich, 2014).
layers consisted of terminally differentiating keratinocytes that Organoids can be initiated from the two main types of stem
gradually developed a cornified cell envelope. Successive cells: (1) pluripotent embryonic stem (ES) cells and their synthetic
improvements in the methodology allowed the cultivation of induced pluripotent stem (iPS) cell counterparts and (2) organ-
large confluent sheets of epidermis grown from relatively small restricted adult stem cells (aSCs). Both approaches exploit the
numbers of primary proliferative keratinocytes (for which the seemingly infinite expansion potential of normal stem cells in
term ‘‘stem cell’’ was not applied). Green and co-workers per- culture. For ES and iPS cells, here collectively termed pluripotent
formed the first successful treatment of two third-degree burn stem cells or PSCs, this potential has been an essential prereq-
patients with cultured autologous keratinocyte sheets at the uisite for their discovery. By contrast, aSCs—with the exception
Peter Bent Brigham Hospital in 1980 (O’connor et al., 1981). In of Green’s skin cells—were long believed to be incapable of sig-
a particularly dramatic demonstration of the potential of the nificant proliferation outside of the body. Yet, recent years have
method, they showed, in the summer of 1983, that this approach witnessed the rapid development of growth factor cocktails that
was life-saving for the 5-year-old Jamie Selby and his 6-year-old mimic the various organ stem cell niches. When PSCs and aSCs
brother Glen, who had both sustained burns over >95% of their are allowed to differentiate in culture, they display an uncanny
body surface (Gallico et al., 1984). capacity to self-organize into structures that reflect crucial as-
In his own lab, Rheinwald built on this work to establish a pects of the tissues to which they are fated.
comparable method for culturing another stratified squamous
epithelium, the cornea (Lindberg et al., 1993). De Luca and Pel- Organoids Derived from Pluripotent Stem Cells
legrini applied this technology for the treatment of corneal blind- Ever since pluripotent ES and iPS cell lines were established,
ness with a high rate of success, as reported upon long-term scientists have applied insights from developmental biology to
follow-up of 112 patients. Their procedure was straightforward: derive differentiated cell types from these stem cells (Chen
a 1–2 mm biopsy from the limbal region of the healthy eye was et al., 2014; Cherry and Daley, 2012) (Figure 2). Yoshiki Sasai
grown in culture on 3T3 feeder cells, and the resulting sheet and his colleagues were the first to take this one step further
was grafted onto the injured eye (Pellegrini et al., 1997; Rama by asking whether such an in vitro system could recapitulate
et al., 2010). While the term ‘‘organoid’’ was not used in these some of the robust regulatory systems of organogenesis—in
pioneering studies, Rheinwald and Green were the first to recon- terms of not only cell differentiation, but also spatial patterning
stitute 3D tissue structure from cultured human stem cells. and morphogenesis. In a remarkable tour de force, they devel-
Organoids revealed their first popularity in the years 1965– oped methods to generate brain structures, retina, and pituitary
1985, shown by an increase in the PubMed search term ‘‘orga- ‘in a dish’ (Eiraku and Sasai, 2012).
noids’’ (Figure 1), mostly in classic developmental biology Brain Organoids
experiments that sought to describe organogenesis by cell The central nervous system derives from the neural ectoderm.
dissociation and reaggregation experiments (for an overview, Set up first as the neural plate, it is then shaped into the neural
see Lancaster and Knoblich, 2014). The past 7–8 years have tube through folding and fusion. Morphogen gradients in this
1586 Cell 165, June 16, 2016 ª 2016 Elsevier Inc.

Figure 1. Citations to the Search Term
‘‘Organoids’’ in PubMed
organoids containing representations of

several different brain regions (Lancaster
et al., 2013). Like the Sasai method, the
approach starts with floating embryoid
bodies, but growth factors are not added
to drive particular brain region identities.
Instead, the aggregates are embedded
in a laminin-rich extracellular matrix
secreted by the Engelbreth-Holm-Swarm
tumor cell line (Matrigel). This allows
outgrowth of large neuroepithelial buds,
which spontaneously develop into various
brain regions. Cerebral organoids can
reach sizes of up to a few millimeters
when grown in a spinning bioreactor.
A spectacular variety of brain regions,
tube establish a dorsal-ventral axis (roof, alar, basal, and floor including retina, dorsal cortex, ventral forebrain, midbrain-hind-
plate) and a rostral-caudal axis (tel-, di-, mes-, and rhomb- brain boundary, choroid plexus, and hippocampus, is observed
encephalon and spinal cord). Neurons are generally generated in these cultures (Figure 4A).
from neural stem cells that reside near the ventricles. These A subsequent study applied single-cell RNA sequencing to
stem cells initially increase their numbers through symmetric di- compare gene expression programs of cells within cerebral
visions. During neurogenesis, stem cells switch to asymmetric organoids to those of fetal human neocortex development
divisions to yield temporal waves of distinct self-renewing (Camp et al., 2015). It was thus found that gene expression
progenitors and differentiated cell types, such as neurons and programs of cortical cells in organoids are remarkably similar
intermediate progenitors, that migrate outward to generate re- to those of the corresponding fetal tissue, underscoring that as-
gion-specific stratified structures such as the medulla, the optic pects of human cortical development can be studied in organoid
tectum, and the cerebral cortex (see Eiraku and Sasai, 2012; culture.
Lancaster and Knoblich, 2014). Retinal Organoids
Spontaneous neural differentiation occurs in ES culture in the The retina is of neurectodermal origin and constitutes the light-
absence of inhibitors of neural differentiation (such as BMP, receptive neural region of the eye. The optic vesicle forms as a
Nodal, and Wnts), consistent with a neural-default state for pseudostratified, cystic outgrowth of the diencephalon. The front
ES cells. Based on this notion, Sasai and colleagues designed of the vesicle then moves inward to form the two-layered optic
SFEBq: serum-free floating culture of embryoid body-like aggre- cup, consisting of the outer retinal pigment epithelium and the
gates with quick reaggregation (Eiraku et al., 2008). In this culture inner neural retina. The neural retina continues to stratify into
system, ES cells isolated from growth factor-free 2D cultures are layers of photoreceptors and supportive cell types, such as
reaggregated in 96-well non-adhesive culture plates. The re-ag- horizontal cells, bipolar cells, and amacrine cells.
gregates are kept in serum-free medium containing no or mini- To mimic this process in vitro (Eiraku et al., 2011), Sasai and
mal growth factors for 7 days, after which they are replated in colleagues again generated floating embryoid bodies from
adhesion plates. When lumens form, the ES cells differentiate re-aggregated murine ES cells in growth factor-free medium to
and polarize to form a continuous neurectoderm-like epithelium generate neuroectoderm. Matrigel, dissolved in the medium, al-
that subsequently generates stratified cortical tissues contain- lowed the formation of more rigid neuro-epithelial tissues. This
ing cortical progenitors, deep cortical-layer neurons, superficial resulted in the formation of buds of retinal primordial tissue
cortical-layer neurons, and so called Cajal-Retzius cells. In resembling the optic vesicle. Isolated buds were then maintained
the absence of growth factors, the generated cortical tissue in a medium supporting retinal tissue identity. The morphological
spontaneously adopts a rostral hypothalamic fate. Regional tissue shape changes were reminiscent of the stepwise evagina-
identity (e.g., olfactory bulb, rostral and caudal cortices, hem, tion and invagination of the optic cup in vivo. Retinal stratification
and choroid plexus) can be selectively controlled by addition of with proper apical-basal polarity occurred, and markers of neural
specific patterning factors such as Fgf, Wnt, and BMP. It was retina and pigment epithelium were expressed in a spatially cor-
thus shown that, under controlled conditions, ES cells can reca- rect manner. More recently, optic cup organoids were generated
pitulate some of the spatial and temporal events leading to the from human PSCs (Nakano et al., 2012). These human retinal or-
formation of layered neural structures in the brain. ganoids resembled their mouse counterparts but encouragingly
Lancaster and Knoblich took this approach to a next level also displayed human-specific features: they are larger in size
by generating cerebral organoids, or ‘‘mini-brains’’: single neural (yet still small relative to the ‘‘real thing’’), and the formed neural
Cell 165, June 16, 2016 1587

Figure 2. Schematic of the Various Organoids that Can Be Grown from PSCs and the Developmental Signals that Are Employed
Adapted from Lancaster and Knoblich, 2014.
retina grows into a thick multi-layered tissue containing both caudal brain structures, Fgf2 was added soon after initiation
rods and cones, whereas cones were rarely observed in mouse of the culture. To dorsalize the caudalized brain organoids, a
organoid cultures. Hedgehog inhibitor was added during the second week. These
Adenohypophysis Organoids conditions recapitulated early cerebellar plate development,
The adenohypophysis secretes multiple systemic hormones. eventually leading to the formation of mature Purkinje cells (Mu-
During early mammalian development, its anlage originates as guruma et al., 2010). In a subsequent study, the investigators
a placode in the non-neural head ectoderm near the anterior reported that the addition of Fgf19 and SDF1 to this protocol
neural plate. The thickened placode invaginates and detaches allows human ES cells to generate a polarized structure reminis-
from the oral ectoderm, forming a hollowed epithelial vesicle, cent of the first trimester cerebellum (Muguruma et al., 2015).
Rathke’s pouch. This process depends on poorly defined Hippocampus
cross-signaling between ectoderm and developing neural tube. The hippocampus develops from the dorsomedial telencephalon
Sasai’s group sought to recapitulate the inductive microenviron- through a precursor structure termed the medial pallium. A
ment of this morphogenetic field in order to promote the simulta- final protocol developed by Sasai and coworkers involved the
neous generation of both tissues within the same aggregate in vitro generation of a reliable source of hippocampal tissue
of SFEBq-cultured ES cells. Three-fold larger cell aggregates from human ES cells (Sakaguchi et al., 2015). SFEBq served
were required, compared to the above protocols. Hedhehog once again as the starting material. Stimulation by BMP and
and Notch antagonists were added to block neural fate in the Wnt induced choroid plexus, the dorsomedial-most part of
outer layers and to allow the subsequent development of all the telencephalon. Careful titration of BMP and Wnt exposure
major hormone-producing linages, respectively. Under these allowed the self-organization of tissue resembling the medial
conditions, ES cells differentiated into head ectoderm and hypo- pallium, located adjacent to choroid plexus in the developing
thalamic neuroectoderm in adjacent layers within the aggregate. brain. Following long-term dissociation culture, granule neurons
Rathke’s-pouch-like structures arose at the interface of these and pyramidal neurons were formed, both of which were electri-
two epithelia, and the various endocrine cell types were subse- cally functional within connected networks.
quently formed. Upon transplantation under the kidney capsule In addition to these CNS organoids, protocols have also been
of hypophysectomized mice, the aggregates partially rescued developed to grow various endodermal organoids from PSCs.
systemic glucocorticoid level and prolonged survival of the mice. Formation of the endoderm germ layer during gastrulation re-
Cerebellar Organoids quires Nodal signaling. Definitive endoderm presents as a 2D
The initial phase of cerebellar development depends on the sheet of cells, which is subsequently patterned along the ante-
function of the isthmic organizer located at the midbrain-hind- rior-posterior axis and folded into a primitive gut tube, from
brain boundary. Sasai and colleagues focused on the induction which all endodermal organs arise. The foregut forms the ante-
of isthmic development in an attempt to create functional Pur- rior section of this tube and generates, e.g., the thyroid, lungs,
kinje cells, the beautiful key output cells of the cerebellum. Again, stomach, liver, and pancreas. The mid- and hindgut develop
they started from a mouse SFEBq culture. In order to produce into small intestine, colon, and rectum. Insights into the signals
1588 Cell 165, June 16, 2016

that control these fate decisions in vivo can be exploited in vitro. Initial attempts to create thyroid organoids involved forced
Exposure to Nodal or its mimetic Activin directs differentiation expression of the lineage-specific transcription factors NKX2.1
of PSCs into definitive endoderm and serves as a common start- and PAX8 and encouragingly resulted in the formation of mouse
ing point of these protocols. Exposure to subsequent inductive and human thyroid follicles in vitro and upon transplantation (An-
signals can then induce the various endodermal organ identities tonica et al., 2012; Ma et al., 2015). Kotton and colleagues
(reviewed in Sinagoga and Wells, 2015). applied an improved version of their ‘‘all soluble factor’’ protocol
Stomach Organoids (Longmire et al., 2012), followed by sequential treatment by
The stomach develops from the posterior foregut. Wells and col- BMP4/FGF2 and induced maturation by 3D plating in the pres-
leagues used activin treatment of human PSCs to generate defin- ence of thyroid-stimulating hormone in Matrigel. The resulting
itive endoderm (McCracken et al., 2014). Subsequent addition of fully mature murine thyroid follicular organoids secreted thyroid
BMP inhibitors and of FGF and Wnt activators instructed the cells hormones in vivo upon transplantation and rescued hypothyroid
toward a foregut fate. When retinoic acid was applied, the orga- mice. The same protocol allowed the derivation of human thyroid
noids were specified toward a posterior foregut fate. Finally, high progenitors from iPS cells (Kurmann et al., 2015).
concentrations of EGF then converted these into human gastric Small Intestinal Organoids
organoids, progressing through molecular and morphogenetic Wnt and FGF signals are known to specify definitive endoderm
stages that resembled those of the developing antrum of the toward mid-/hindgut fates (‘‘posteriorization’’). To generate in-
mouse stomach. Organoids contained primitive gastric gland- testinal organoids (McCracken et al., 2014; Spence et al.,
and pit-like domains, proliferative zones with Lgr5+ stem cells, 2011), Activin-treated human PSCs were cultured with FGF4
mucous cells, and a host of gastric endocrine cells. and WNT3a. Mid/hindgut spheroids budded off from the 2D
Lung and Thyroid Organoids monolayer epithelium and were further cultured in Matrigel along
The lung and the thyroid arise from Nkx2-1+ progenitors in the with a pro-intestinal growth factor cocktail, defined previously for
developing ventral foregut endoderm. An initial study demon- expansion of adult crypt cultures (Sato et al., 2009; see below).
strated the directed differentiation of primordial lung and thyroid The organoids expanded over 1–3 months to give rise to a polar-
progenitors from ESCs. The protocol involves activin-induced ized intestinal epithelium patterned into villus-like structures and
definitive endoderm and treatment with TGFb /BMP inhibitors, crypt-like proliferative zones and containing all major epithelial
followed by BMP/FGF stimulation, and results in a relatively cell types. Intestinal mesenchyme (presumably derived from
pure population of progenitors that recapitulate early develop- mesodermal remnants after endoderm induction) surrounded
mental milestones of lung/thyroid development (Longmire the epithelial structures and consisted of myofibroblasts and
et al., 2012). This study has been the stepping stone for subse- smooth muscle cells (McCracken et al., 2014; Spence et al.,
quent attempts to create organoids representing mature ver- 2011). Transplantation of these organoids into immunodeficient
sions of the two organs. mice yielded human epithelium and laminated human mesen-
A first description of the generation of lung organoids from iPS chyme, supported by mouse vasculature. The transplanted
cells was reported by Rossant and colleagues and involved at its tissue was functional, as shown by permeability and peptide
last stage air-liquid interphase culture. The protocol was applied uptake tests (Watson et al., 2014).
to CFTR mutant iPS cells as a proof of concept for modeling Liver Organoids
cystic fibrosis (Wong et al., 2012). Snoeck and colleagues de- During early hepatogenesis, progenitor cells delaminate from the
signed an improved four-stage, 50-day protocol (Huang et al., foregut endoderm to form a condensed tissue mass termed the
2014). First, definitive endoderm was induced by Activin A. Sub- liver bud, which is vascularized soon thereafter. Taniguchi and
sequently, anterior foregut endoderm was induced by sequential co-workers exploited cross-signaling between endodermal
inhibition of BMP, TGF-b, and Wnt signaling. The cells were then epithelial, mesenchymal, and endothelial progenitors in an effort
ventralized by Wnt, BMP, FGF, and RA to obtain lung and airway to generate tissues reminiscent of the human liver bud. Human
progenitors. Finally, epithelial cell types (basal, goblet, Clara, PSCs were induced into hepatic endodermal cells in 2D culture
ciliated, type I and type II alveolar epithelial cells) were matured (a protocol involving activin treatment, followed by bFGF/
using Wnt, FGF, c-AMP, and glucocorticoids. Spence and col- BMP4). The human PSC-derived hepatic cells were mixed with
leagues (Dye et al., 2015) similarly started from Activin-treated mesenchymal stem cells and endothelial cells. Plated at high
human PSCs but then followed a slightly different trajectory. density on a layer of Matrigel, 3D aggregates spontaneously
Subsequent addition of TGFb/BMP inhibitors, FGF4, and Wnt formed (Takebe et al., 2013). These liver bud-like aggregates
activators instructed the cells toward an anterior foregut fate. contained blood vessels that, upon transplantation into mice,
When the Hedgehog pathway was simultaneously activated, connected to the host vessels within 48 hr. Liver-specific func-
organoids were ventrally specified toward a lung fate. Upon tions such as protein production and human-specific drug meta-
embedding in Matrigel and prolonged exposure to Fgf10, mature bolism became evident over time. Furthermore, mesenteric liver
lung organoids arose. The cultures could be maintained for bud transplantation rescued recipient mice from drug-induced
several months and resembled proximal airways, containing lethal liver failure.
basal cells, ciliated cells, and Clara cells. The endodermal airway
tissues were found to be often surrounded by smooth muscle The Mesodermal Kidney
actin (SMA)-positive mesenchymal cells. Early markers of the The kidney, with its more than 20 specialized cell types, exhibits
distal (alveolar) airways were expressed early in culture but the highest architectural complexity of all organs outside of the
were lost later. CNS. The adult kidney, or metanephric kidney, arises from the
Cell 165, June 16, 2016 1589

Figure 3. Schematic of the Various Regions of the Body that Can Be Cultured as aSC-Derived Organoids
posterior end of the embryonic intermediate mesoderm, which in this, the cells are pelleted and cultured as 3D organoids for up
turn derives from the primitive streak (presomitic mesoderm). to an additional 3 weeks. Numbers of nephrons are strongly
The intermediate mesoderm generates the two key kidney increased upon a brief (1 hr) exposure to a Wnt agonist at the
progenitor populations: the ureteric epithelium and the meta- start of organoid culture. A complex multicellular kidney orga-
nephric mesenchyme. Through reciprocal interactions, these noid results that contains fully segmented nephrons and is sur-
form the collecting ducts and nephrons (i.e., the epithelia of rounded by endothelia and renal interstitium (Figure 4B). Kidney
glomeruli and proximal and distal renal tubules), respectively. organoids may contain >500 nephrons with defined glomeruli
Until recently, the complex spatial and temporal control of organ- comprising a Bowman’s capsule with podocytes and connected
ogenesis has stood in the way of a detailed molecular under- to proximal tubules. Occasionally, glomeruli show evidence of
standing of specification of individual cell types. Despite this, endothelial invasion.
rapid progress has been made in establishing protocols for While remarkably complete, further improvements of the pro-
differentiation of human PSCs into virtually complete ‘‘mini-kid- tocol will focus on tubular functional maturation, more extensive
neys.’’ glomerular vascularization, and the formation of a contiguous
First, in 2013, it was shown how to induce intermediate meso- collecting ductal tree ‘‘with a single exit path for urine’’ (Takasato
derm from PSCs under defined media conditions (Mae et al., et al., 2015).
2013). One of the renal precursor tissues that derives from the in-
termediate mesoderm, the ureteric epithelium, can be generated Organoids Derived from Adult Stem Cells
from human PSCs in 2D via a similar mesodermal specification While PSC-based organoids exploit developmental processes
step (Xia et al., 2013). Upon aggregation with dissociated mouse for their establishment, aSCs can be coerced to form organoids
embryonic kidney, these progenitors self-organize into 3D by creating conditions that mimic the stem cell niche environ-
ureteric bud structures. The second renal precursor tissue, the ment during physiological tissue self-renewal or during damage
metanephric mesenchyme, can be created from human and repair (Figure 3). As first described for gut stem cells (Korinek
mouse embryoid bodies through sequential exposure to defined et al., 1998), the Wnt pathway has emerged as the major driver
soluble factors. Coculturing of the resulting metanephric mesen- of epithelial aSCs (Clevers et al., 2014). Lgr5 (a receptor for the
chyme with spinal cord tissue, a nephric inducer, produces well- secreted Wnt-amplifying R-spondins and itself encoded by a
organized nephric tubules and nascent glomeruli (Taguchi et al., Wnt target gene) marks active aSCs in many, if not all, epithelia.
2014). It is not surprising that Wnt activators (Wnt3A, R-spondins, or the
Little and colleagues managed to balance the two divergent small molecule GSK3 inhibitor CHIR) are key components of
commitment paths to produce both principal lineages of the kid- most aSC culture protocols and that Lgr5+ stem cells invariably
ney simultaneously (Takasato et al., 2014). Their original protocol appear in such cultures. Below, I discuss the establishment of
involves the application of Activin A and Bmp4 to human PSCs feeder layer/serum-free, fully defined 3D culture conditions for
cultured in 2D to generate primitive streak identity. Fgf9 drives a rapidly growing list of epithelial tissues.
these cells toward an intermediate mesoderm identity, after Small Intestine and Colon
which they spontaneously develop further into ureteric bud and The small intestinal epithelium displays an extremely short turn-
metanephric mesenchyme. The cells display 3D morphologies over time of 5 days. Actively proliferating Lgr5+ intestinal stem
when grown at low density in 2D or when cocultured with mouse cells reside at the crypt base (Barker et al., 2007). Their rapidly
kidney reaggregates. In both cases, structures resembling dividing, transit-amplifying (TA) daughter cells occupy the
ureteric epithelium and proximal tubules appear. In a spectac- remainder of the crypts and, upon differentiation, move onto
ular follow-up study, the protocol was further refined and simpli- the flanks of the villi to eventually die at the villus tips. Differ-
fied: human PSCs are cultured in 2D in the presence of Wnt entiated cell types include absorptive enterocytes, multiple
signals for 4 days followed by 3-day exposure to Fgf9. After secretory cell types (Paneth cells, goblet cells, enteroendocrine
1590 Cell 165, June 16, 2016

cells, and tuft cells), and the M cells that cover Peyer’s patches and endodermal origin. The essential components appear to
(Clevers, 2013). be: (1) a potent source of Wnt, (2) a potent activator of tyrosine
Crypt stem cells are tightly controlled by four signaling path- kinase receptor signaling like EGF, (3) inhibition of BMP/Tgfb sig-
ways. Wnt constitutes the key pathway to maintain stem cell nals, and (4) Matrigel. It is not essential to start from purified
fate and drive proliferation of stem and TA cells. Notch helps to Lgr5+ aSCs. Small fragments of primary tissue serve well as
maintain the undifferentiated state of proliferative stem and starting material, possibly due to the fact that the culture condi-
TA cells: when Notch signaling is blocked, the cells instantly tions mimic a damage response, which in many tissues can re-
differentiate into goblet cells. Epidermal growth factor (EGF) sig- cruit committed cells back to a stem cell state (Clevers, 2015)
nals exert strong mitogenic effects on stem and TA cells. And Stomach
finally, BMP signals are active in the villus compartment, and Rapidly proliferating Lgr5 stem cells are located at the base of
their inhibition is crucial to create a crypt-permissive environ- pyloric glands of the adult mouse stomach. With slight modifica-
ment (Clevers, 2013). tions to the mini-gut culture system, single Lgr5 cells efficiently
Encouraged by the observation that Lgr5 crypt stem cells can generated long-term, continuously expanding organoids closely
go through thousands of cell divisions in vivo, we established resembling mature pyloric epithelium (Barker et al., 2010). At the
a culture system that allows growth of epithelial organoids base of glands of the gastric corpus, Troy marks specialized
(‘‘mini-guts’’) from a single Lgr5 stem cell (Sato et al., 2009). Chief cells. In a remarkable example of cellular plasticity, these
Whole crypts or single Lgr5 stem cells are suspended in Matrigel Chief cells spontaneously dedifferentiate to act as multipotent
and are cultured in serum-free medium supplemented with three epithelial stem cells in vivo, particularly upon damage. Single
recombinant proteins: R-spondin-1 (a Wnt signal amplifier and Troy+ chief cells can be cultured to generate long-lived gastric
ligand of Lgr5), EGF, and the BMP inhibitor Noggin. For colon organoids, containing the various cell types of corpus glands
crypt culture, Wnt3a is additionally required because colon (Stange et al., 2013). Very similar conditions have allowed
epithelium itself makes little, if any, Wnt. The organoids strictly long-term culturing of human stomach organoids that maintain
consist of a simple highly polarized epithelium, tightly closing many characteristics of the original tissue (Bartfeld et al., 2015).
off a central lumen. Crypt-like structures project outward. The Liver and Pancreas
basal side of the cells is oriented toward the surrounding Matri- The two cell types of the liver proper (the hepatocyte and the bile
gel. Enterocyte brush borders form the luminal surface, while duct cell) turn over at a slow, pedestrian rate. In agreement, Lgr5
secretion by Paneth and goblet cells occurs toward the lumen. is not expressed at appreciable levels in the healthy adult mouse
All cell types of the epithelium are represented at normal ratios liver. Yet, we observed that small Lgr5+ cells appear near bile
(Grün et al., 2015; Sato et al., 2009). The organoids can be ducts upon toxic damage, coinciding with robust Wnt pathway
passaged weekly at a 1:5 ratio for years and are remarkably activation. These damage-induced Lgr5+ cells generated hepa-
stable, both genetically and phenotypically (Sato and Clevers, tocytes and bile ducts in vivo. When cultured in a modified
2013). version of the mini-gut medium, single Lgr5+ cells could be
Addition of Wnt3A to the combination of growth factors al- clonally expanded as organoids, consisting largely of progen-
lowed seemingly indefinite growth of mouse colon organoids. itor cells expressing early bile duct and hepatocyte markers.
Addition of nicotinamide, along with a small molecule inhibitor Removal of mitotic stimuli and simultaneous inhibition of Notch
of Alk and an inhibitor of p38, was required for long-term culture signals led to hepatocyte lineage differentiation. Upon transplan-
of human small intestine and colon organoids (Jung et al., 2011; tation, these organoids matured into functional hepatocytes
Sato et al., 2011). Given that Lgr5 protein expression is vanish- (Huch et al., 2013b). In a follow-up study, we defined conditions
ingly low, other stem cell markers have been explored to initiate for long-term expansion of adult bipotent progenitor cells from
intestinal organoid cultures, including CD24 (von Furstenberg human liver. Somewhat surprisingly, one-third of all mature bile
et al., 2011), EphB2 (Jung et al., 2011), and CD166+/GRP78 duct cells could initiate clonal liver organoid growth. Deep
(Wang et al., 2013). As proof of stability upon culture, a large sequencing of clonal organoids derived at different intervals of
batch of organoids was grown from a single Lgr5 colon stem culture revealed a highly stable genome at the structural level,
cell and transplanted per anum into multiple mice with experi- while single base changes occurred at very low rates. Again,
mental colitis. The organoids readily integrated as functional the cells could be converted into functional hepatocytes
epithelial patches that were indiscernible from the surrounding in vitro and upon transplantation into mice (Huch et al., 2015).
host epithelium (Yui et al., 2012). The same protocols allowed long-term expansion of canine liver
In a different approach, fragments of neonatal mouse intestine progenitor cells that could be differentiated toward functional
containing epithelial and mesenchymal elements were grown in hepatocytes (Nantasanti et al., 2015).
serum-containing medium (without specific growth factors) in Like their liver counterparts, the exocrine/acinar, ductal, and
collagen with air-liquid interface. The expanding cystic struc- endocrine cell types of the adult pancreas turn over very slowly.
tures consisted of a simple epithelium in which all cell types Wnt signaling is inactive and Lgr5 is not expressed under phys-
were discernible. The structures were surrounded by myofibro- iological conditions, yet the Wnt pathway is robustly activated
blasts and were responsive to R-spondin and to Notch inhibition upon injury, concomitant with induced Lgr5 expression in regen-
(see above) (Ootani et al., 2009). erating pancreatic ducts. Under modified mini-gut conditions,
The mini-gut culture system has since been adapted for the single isolated duct cells could be cultured long-term as pancre-
generation of organoids representing the epithelial compart- atic progenitor organoids. Clonal pancreas organoids differ-
ments of a series of mouse and human tissues of ecto-, meso- entiated along ductal and endocrine lineages when grafted
Cell 165, June 16, 2016 1591

in vivo in a developing pancreas, indicative of bipotentiality Taste Buds
(Huch et al., 2013a). Similar observations were made for human Previous studies had shown that Lgr5 marks adult stem cells in
pancreatic organoids (Boj et al., 2015). Grompe and colleagues the rapidly self-renewing taste buds of the tongue. Using the
addressed the identity of the organoid-initiating epithelial cell original mini-gut culture protocol, single isolated stem cells
from mouse pancreas and liver using a set of cell surface from taste tissue generated continuously expanding 3D organo-
markers and found that the transcriptomes of the two popula- ids, which phenotypically contained mature taste receptor cells
tions overlapped extensively. Pancreatic organoid cells had the (Ren et al., 2014). To assay functionality of these cells, cultured
unexpected capacity to generate hepatocyte-like cells upon organoids were reseeded in 2D onto laminin-coated coverslips
transplantation in a mouse liver damage model, indicative of in the same culture medium. By calcium imaging assays, dose-
the close kinship of these two progenitor populations (Dorrell dependent responses to tastants were readily documented,
et al., 2014). demonstrating that functional taste cells can be generated
Prostate ex vivo from single Lgr5+ taste bud stem cells. Moreover, it could
The pseudostratified prostate epithelium consists of basal and be concluded that single stem cells generate all taste cell types
luminal cells. We developed a mini-gut-based 3D culture protocol and that the formation of taste cells does not require innervation.
that supports long-term expansion of primary mouse and human Lung
prostate organoids, composed of fully differentiated basal and Hogan and colleagues reported an early bronchiolar lung orga-
luminal cells. Single human luminal as well as basal cells gave noid culture protocol, involving Matrigel supplemented with
rise to organoids, yet luminal-cell-derived organoids more EGF, e.g. Single basal cells isolated from the trachea grew into
closely resembled prostate glands. Stimulation with R-spondin/ ‘‘tracheospheres’’ consisting of a pseudostratified epithelium
Wnt was not essential for continued growth of the organoids with basal cells and ciliated luminal cells. These organoids could
but strongly induced luminal cells, leading to a prostate-like be passaged at least twice. No mature Clara-, neuroendocrine-
pseudostratified structure of the organoids. Long-term cultured or mucus-producing cells were observed (Rock et al., 2009). In
organoids were genetically stable and reconstitute prostate a later study, this clonal 3D organoid assay was used to screen
glands in recombination assays (Karthaus et al., 2014). Indepen- for factors controlling generation of ciliated versus secretory
dently, Shen and colleagues developed a Matrigel/EGF-based cells from basal cells. It was thus found that IL-6 treatment
culture system supplemented with androgens and reported resulted in the formation of multiciliated cells at the expense of
very similar observations (Chua et al., 2014). secretory and basal cells (Tadokoro et al., 2014). Figure 4C de-
Mammary Gland picts a human airway organoid.
This pseudostratified epithelium consists of two major cell Organoids representing the distal airways (‘‘alveolospheres’’)
lineages. The inner (luminal) cells secrete milk, while the con- have been more recently established. The alveoli consist of
tractile outer layer of myoepithelial (basal) cells ejects the milk. gas-exchanging type I and surfactant-secreting type II cells.
No long-term organoid protocol has been reported yet. While both cell types originally derive from a common progenitor,
However, encouragingly, freshly isolated human mammary it appears that, later in life, a rare self-renewing type II cell acts
epithelial cells have been cultured for two to three passages as the stem cell to regenerate the alveolar epithelium. Indeed,
in floating collagen gels in the presence of a Rho-associated sorted type II cells remained proliferative in short-term culture
kinase inhibitor to form branching ducts with alveoli at their and could generate type I cells (Desai et al., 2014; Treutlein
tips. Basal and luminal markers were expressed at correct po- et al., 2014). Alternative culture conditions allowed establish-
sitions, and the ducts displayed contractility. Thus, the orga- ment of mouse and human alveolospheres from single type I
noids resembled terminal ductal-lobular units, the functional as well as type 2 alveolar cells, containing both cell types in
units of the mammary gland (Linnemann et al., 2015). Since the same organoid. Having said this, these alveolosphere culture
Wnt signals and Lgr5 have been implied in mammary stem conditions are as yet not fully defined, requiring co-culture with
cell biology (Plaks et al., 2013; Rios et al., 2014), it will be of non-epithelial cells (e.g., mouse lung fibroblasts) (Barkauskas
interest to test the effects of the addition of Wnt/R-spondin et al., 2013; Jain et al., 2015).
to these cultures. Salivary Gland
Fallopian Tube Coppes and colleagues have exploited organoid culture to
The fallopian tube of the uterus is lined by a simple columnar expand single salivary gland cells in vitro into distinct lobular or
epithelium in which secretory cells produce tubular fluid, while ductal/lobular organoids, containing some salivary gland line-
ciliated cells facilitate transport of gametes. Since the epithelium ages. The original short-term culture technology depended on
is exposed to cyclical hormonal changes, self-renewal mecha- FGF, EGF, and Matrigel. The cultured cells were able to effi-
nisms are of critical importance for its integrity. Notably, recent ciently restore radiation-damaged salivary gland function in
evidence has indicated that the fallopian tube epithelium is the transplanted mice (Nanduri et al., 2014). In a follow-up study,
tissue of origin for ovarian cancer. Based on the mini-gut-proto- robust Wnt pathway activation through the addition of Wnt3A
col, long-term, stable 3D organoid cultures were established and R-spondin allowed long-term expansion of the organoids,
from human fallopian tubes. Single epithelial stem cells gave containing all differentiated salivary gland cell types. Transplan-
rise to clonal organoids containing both ciliated and secretory tation of these cells into submandibular glands of irradiated mice
cells, thus establishing an experimental system for the study of robustly restored saliva secretion and increased the number of
the human fallopian tube epithelium in health and disease (Kess- functional acini in vivo (Maimets et al., 2016). Since post-radia-
ler et al., 2015). tion hyposalivation often leads to irreversible and untreatable
1592 Cell 165, June 16, 2016

Figure 4. A ‘‘Mini-Brain’’ Generated from PSCs
(A) A complex morphology with heterogeneous regions containing neural progenitors (SOX2, red) and neurons (TUJ1, green) is apparent (Lancaster et al., 2013).
Courtesy of Madeline Lancaster.
(B) Immunofluorescent image of an entire kidney organoid grown from PSCs with patterned nephrons. Podocytes of the forming glomeruli (NPHS1, yellow), early
proximal tubules (lotus tetragonolobus lectin, pink), and distal tubules/collecting ducts (E-Cadherin, green). Courtesy of Melissa Little.
(C) 3D reconstruction of the midsection of a human aSC-derived lung organoid stained for intermediate filaments of basal cells (green), the actin cytoskeleton
(red), and nuclei (blue) and imaged by confocal microscopy (N. Sachs and H.C., unpublished data).
xerostomia, this condition may present an early opportunity for regulatory factor 7 gene. These organoids produced less type I
the development of organoid technology-based cell therapy. interferon and displayed increased influenza virus replication
Esophagus (Ciancanelli et al., 2015). In another example, human stomach
All examples above represent simple or two-layered epithelia. organoids, grown from PSCs or aSCs, can be productively in-
Lagasse and colleagues showed that the keratinizing stratified fected by Helicobacter pylori (Bartfeld et al., 2015; McCracken
epithelium of the esophagus can also be cultured as organoids et al., 2014).
in ‘‘mini-gut’’ medium (DeWard et al., 2014). Basal cells in the As a striking example, Qian et al. developed a miniaturized
mouse esophagus represent a heterogeneous population of spinning bioreactor to generate forebrain-specific organoids
proliferative cells. When plated as single cells, these give rise from human iPSCs, following the Lancaster/Knoblich protocol.
to organoids that were morphologically similar to normal esoph- These organoids recapitulate many features of cortical develop-
ageal tissue, with small basal-like cells in contact with the extra- ment, including the formation of a distinct human-specific outer
cellular matrix, large flat suprabasal-like cells in the interior, and radial glia cell layer. Infection of these developing forebrain orga-
hardened keratinized material in the center. Expression of spe- noids with Zika virus (ZIKV) resulted in the preferential infection of
cific markers for each of these cell types confirmed the correct neural progenitors, resulting in cell death, decreased prolifera-
layering of the organoid walls. It will be of interest to determine tion, and a reduced neuronal cell-layer volume, thus modeling
whether basal cells from other squamous epithelia (epidermis, ZIKV-associated microcephaly. The authors propose this as a
vagina) will also be amenable to organoid culture. versatile experimental for mechanistic studies as well as for
testing of potential ZIKV antiviral drugs (Qian et al., 2016).
Applications of Organoid Technology
Both PCS- and aSC-based organoids can be initiated from sin- Hereditary Disease
gle cells and cultured long-term and are amenable to essentially Organoids can be used to study and model organ-specific
all cell-biological and molecular analyses that have been devel- monogenic hereditary diseases. Knoblich and colleagues identi-
oped for ‘‘traditional’’ cell lines. As such, they provide a new fied a patient with a mutation in the CDK5RAP2 and severe
window—between cell lines and in vivo studies—to studying microcephaly. The corresponding iPS cells made significant
basic gene functions and cellular processes. In addition to this, smaller ‘‘mini-brains,’’ containing only occasional neuroepithelial
organoid technology also holds great promise for translational regions with signs of remature neural differentiation, a phenotype
research. Below, I give some examples of its translational appli- that could be rescued by reintroducing the CDK5RAP2 protein
cations. (Lancaster et al., 2013).
Infectious Disease Cystic fibrosis (CF) is caused by a spectrum of mutations in the
Since organoids—unlike cell lines—ideally represent all cellular cystic fibrosis transmembrane conductance regulator (CFTR)
components of a given organ, they are theoretically well suited chloride channel that is normally expressed in epithelial cells of
for infectious disease studies, particularly of pathogens that many organs. Mirroring the in vivo situation, surface expression
are restricted to man and are dependent on specialized cell of CFTR was absent in iPS-derived lung organoids from CF
types. In an illustrative application, iPS-derived lung organoids patients but could be restored by treatment with a (then still
were generated from an otherwise healthy child who suffered experimental) small molecule that corrects some of the common
life-threatening influenza and carried null alleles in the interferon CF-processing mutations (Wong et al., 2012). Dekkers and
Cell 165, June 16, 2016 1593

colleagues derived intestinal organoids from rectal biopsies of a more than one culture condition can be applied to efficiently
series of CF patients. Forskolin induces a robust swelling of wild- grow human pancreas cancer organoids (Boj et al., 2015; Huang
type organoids due to fluid transport to the organoid lumen. This et al., 2015).
swelling response is absent in CF organoids yet can be restored In a marriage between organoid and CRISPR/CAS9 technolo-
for the common, temperature-sensitive CFTR-F508del mutant gies, two independent studies have modeled the ‘‘adenoma-car-
by culturing at 27 C and also by the addition of experimental cinoma sequence’’ by introducing four sequential mutations into
CFTR corrector compounds (Dekkers et al., 2013). human colon organoid stem cells. Xenotransplantation revealed
Independently, the Verma lab generated iPS cells from CF the progressive transformation of the wild-type stem cells into
patients and corrected the mutation by CRISPR/Cas9. The cor- adenocarcinomas (Drost et al., 2015; Matano et al., 2015). These
rected iPS cells were subsequently converted to mature airway approaches may allow rapid modeling of novel (combinations of)
epithelial cells demonstrating recovery of normal CFTR function gene mutations, as found in the ongoing genome- and world-
(Firth et al., 2015). wide sequencing efforts on large solid tumor panels.
Liver organoids from alpha 1-antitrypsin deficiency patients
reproduced the deleterious effects of the mutant protein pre- Personalized Medicine
cipitates in hepatocytes, while the absence of mature biliary cells In principle, the aSC-based organoid technology allows rapid
in liver organoids from an Alagille syndrome patient mirrored the ex vivo testing of drug responses on the affected tissue of indi-
in vivo biliary tree abnormalities (Huch et al., 2015). Liver organo- vidual patients. As a first example, the colon organoid-based
ids from dogs deficient in the copper-transporter COMMD1 CF test (Dekkers et al., 2013) can be read out in weeks after
mimicked the disease by accumulating toxic levels of copper, biopsy. The approach has already been applied for identification
which could be salvaged by re-expression of wild-type COMMD1 and successful treatment of patients with very rare CFTR muta-
protein (Nantasanti et al., 2015). tions, who otherwise have no access to the recently introduced
CF drugs (Dekkers, 2016). The feasibility of culturing various
Toxicology solid tumors directly from the patient in the form of tumor orga-
The possibility to grow human organoids representative of the noids (see above) holds a similar promise, yet the applicability
main targets for drug-related toxicity (gut, liver, kidney) opens of such an approach is less clear than in the case of the ‘‘single
up theoretical avenues to complement animal-based toxicology genetic lesion’’ CF organoids. Tumor organoids grow with
with assays performed directly on these vulnerable human tis- unpredictable and often slower kinetics when compared to
sues. In one such example, Little and colleagues have utilized wild-type organoids, and—like the original tumors—display a
human kidney organoids to illustrate that cisplatin acts as a heterogeneous genetic make-up. Ongoing trials will reveal the
nephrotoxicant (Takasato et al., 2015). validity and applicability of tumor organoids in the assessment
of drug response at the level of the individual patient.
Cancer Regenerative Medicine and Gene Therapy
Once culturing protocols for human aSC-based organoids were Proof-of-concept studies have demonstrated the feasibility
established, we have shown the feasibility of growing organoids of expanding organoids from (single) aSCs followed by safe
from primary colon, prostate, and pancreatic cancers (Boj et al., transplantation into animals. This was first done for murine small
2015; Gao et al., 2014; Sato et al., 2011; van de Wetering et al., intestine (Fordham et al., 2013; Yui et al., 2012). When small in-
2015). These cancer organoids provide the unique opportunity testinal organoids are transplanted to colon, they retain original
for functional testing (e.g., for drug sensitivity) and for correlating small intestinal features like villus formation and the presence
such data with the genetic make-up of individual tumors. of Paneth cells, indicative of the phenotypic stability of cultured
Cancer can also be modeled in organoids derived from aSC organoids (Fukuda et al., 2014). Of note, it may be advanta-
wild-type stem cells. Kuo and colleagues probed the metastatic geous to modify the culture conditions to selectively expand
potential of TGFBR2 loss in murine stomach organoids by its stem cells at the cost of differentiated cells (Wang et al.,
shRNA knockdown within Cdh1 / ;Tp53 / organoids. This 2015; Yin et al., 2014) prior to transplantation. We have used
resulted in invasive phenotypes in vitro and in robust metas- CRISPR/Cas9 genome editing to correct the CFTR locus by
tasis in vivo (Nadauld et al., 2014). Using similar shRNA-based homologous recombination in cultured intestinal stem cells of
strategies, the same authors created combinatorial Apc, p53, CF patients. The corrected allele was fully functional, as demon-
KrasG12D, and Smad4 mutations in wild-type murine colon or- strated in clonally expanded organoids (Schwank et al., 2013).
ganoids and observed progressive transformation to an invasive This approach can presumably be used for gene correction in
adenocarcinoma-like histology in vivo, recapitulating the multi- any clonally expandable cell population derived from monogenic
hit model of colorectal cancer (CRC) (Li et al., 2014). In a different hereditary disease patients.
approach, Huang et al. established a three-step culture method
using human PSCs to generate 3D structures closely resembling Conclusions
human fetal exocrine pancreas (Huang et al., 2015). Expression In this Review, I have attempted to describe the state of the art of
of mutant KRAS and/or TP53 in these early pancreas organoids the explosively developing field of PSC- and aSC-based organo-
induced abnormal ductal architecture and nuclear morphology ids. The current versions of organoids have clear limitations,
consistent with neoplastic transformation in culture and in vivo. e.g., innervation, blood vessels, and immune cells are absent,
17 of 20 primary human pancreas cancers could be propagated and as a consequence, disease processes are only partially
under the ‘‘third-step’’ conditions (i.e., EGF and FGF). Thus, recapitulated. Yet it is anticipated that the potent self-organizing
1594 Cell 165, June 16, 2016

properties of organoids may extend beyond their current bound- ease. Life-threatening influenza and impaired interferon amplification in human
aries and allow the proper incorporation of additional cellular IRF7 deficiency. Science 348, 448–453.
(or microbial) elements. From a basic science perspective, Clevers, H. (2013). The intestinal crypt, a prototype stem cell compartment.
PSC-based organoids will by their very nature play a key role Cell 154, 274–284.
in understanding the developmental biology of organs and will Clevers, H. (2015). STEM CELLS. What is an adult stem cell? Science 350,
1319–1320.
thus complement the long tradition of in vivo studies in this field.
Clevers, H., Loh, K.M., and Nusse, R. (2014). Stem cell signaling. An integral
From the same perspective, aSC-based organoids provide basic
program for tissue renewal and regeneration: Wnt signaling and stem cell
insights into the processes that allow aSCs to maintain and control. Science 346, 1248012.
repair established tissues. Yet, because of the ease of produc-
Dekkers, F.e.a. (2016). Identifying potential clinical responders to CFTR-
tion and the close resemblance to human organs in health and modulating drugs using rectal cystic fibrosis organoids. Sci. Transl. Med.,
disease, organoids hold great appeal for translational research in press.
and invite an almost immediate application into the clinic. Dekkers, J.F., Wiegerinck, C.L., de Jonge, H.R., Bronsveld, I., Janssens, H.M.,
de Winter-de Groot, K.M., Brandsma, A.M., de Jong, N.W., Bijvelds, M.J.,
ACKNOWLEDGMENTS Scholte, B.J., et al. (2013). A functional CFTR assay using primary cystic
fibrosis intestinal organoids. Nat. Med. 19, 939–945.
Thanks for comments to the text to Melissa Little, Hans Snoeck, Juergen Desai, T.J., Brownfield, D.G., and Krasnow, M.A. (2014). Alveolar progenitor
Knoblich, and Esther Verheyen and to Janny van Eldik for secretarial help. and stem cells in lung development, renewal and cancer. Nature 507, 190–194.
Apologies to those scientists whose work could not be cited due to space re- DeWard, A.D., Cramer, J., and Lagasse, E. (2014). Cellular heterogeneity in the
strictions. Written as visiting professor at Memorial Sloan Kettering Cancer mouse esophagus implicates the presence of a nonquiescent epithelial stem
Center, New York, NY. H.C. is named as inventor on several patents related cell population. Cell Rep. 9, 701–711.
to Lgr5 stem-cell-based organoid technology. Dorrell, C., Tarlow, B., Wang, Y., Canaday, P.S., Haft, A., Schug, J., Streeter,
P.R., Finegold, M.J., Shenje, L.T., Kaestner, K.H., and Grompe, M. (2014). The
organoid-initiating cells in mouse pancreas and liver are phenotypically and
REFERENCES
functionally similar. Stem Cell Res. (Amst.) 13, 275–283.
Antonica, F., Kasprzyk, D.F., Opitz, R., Iacovino, M., Liao, X.H., Dumitrescu, Drost, J., van Jaarsveld, R.H., Ponsioen, B., Zimberlin, C., van Boxtel, R., Buijs,
A.M., Refetoff, S., Peremans, K., Manto, M., Kyba, M., and Costagliola, S. A., Sachs, N., Overmeer, R.M., Offerhaus, G.J., Begthel, H., et al. (2015).
(2012). Generation of functional thyroid from embryonic stem cells. Nature Sequential cancer mutations in cultured human intestinal stem cells. Nature
491, 66–71. 521, 43–47.
Barkauskas, C.E., Cronce, M.J., Rackley, C.R., Bowie, E.J., Keene, D.R., Dye, B.R., Hill, D.R., Ferguson, M.A., Tsai, Y.H., Nagy, M.S., Dyal, R., Wells,
Stripp, B.R., Randell, S.H., Noble, P.W., and Hogan, B.L. (2013). Type 2 J.M., Mayhew, C.N., Nattiv, R., Klein, O.D., et al. (2015). In vitro generation
alveolar cells are stem cells in adult lung. J. Clin. Invest. 123, 3025–3036. of human pluripotent stem cell derived lung organoids. eLife 4, 4.
Barker, N., van Es, J.H., Kuipers, J., Kujala, P., van den Born, M., Cozijnsen, Eiraku, M., and Sasai, Y. (2012). Self-formation of layered neural structures in
M., Haegebarth, A., Korving, J., Begthel, H., Peters, P.J., and Clevers, H. three-dimensional culture of ES cells. Curr. Opin. Neurobiol. 22, 768–777.
(2007). Identification of stem cells in small intestine and colon by marker Eiraku, M., Watanabe, K., Matsuo-Takasaki, M., Kawada, M., Yonemura, S.,
gene Lgr5. Nature 449, 1003–1007. Matsumura, M., Wataya, T., Nishiyama, A., Muguruma, K., and Sasai, Y.
(2008). Self-organized formation of polarized cortical tissues from ESCs and
Barker, N., Huch, M., Kujala, P., van de Wetering, M., Snippert, H.J., van Es,
its active manipulation by extrinsic signals. Cell Stem Cell 3, 519–532.
J.H., Sato, T., Stange, D.E., Begthel, H., van den Born, M., et al. (2010).
Lgr5(+ve) stem cells drive self-renewal in the stomach and build long-lived Eiraku, M., Takata, N., Ishibashi, H., Kawada, M., Sakakura, E., Okuda, S.,
gastric units in vitro. Cell Stem Cell 6, 25–36. Sekiguchi, K., Adachi, T., and Sasai, Y. (2011). Self-organizing optic-cup
morphogenesis in three-dimensional culture. Nature 472, 51–56.
Bartfeld, S., Bayram, T., van de Wetering, M., Huch, M., Begthel, H., Kujala, P.,
Vries, R., Peters, P.J., and Clevers, H. (2015). In vitro expansion of human Firth, A.L., Menon, T., Parker, G.S., Qualls, S.J., Lewis, B.M., Ke, E., Dargitz,
gastric epithelial stem cells and their responses to bacterial infection. Gastro- C.T., Wright, R., Khanna, A., Gage, F.H., and Verma, I.M. (2015). Functional
enterology 148, 126–136.e6. Gene Correction for Cystic Fibrosis in Lung Epithelial Cells Generated from
Patient iPSCs. Cell Rep. 12, 1385–1390.
Boj, S.F., Hwang, C.I., Baker, L.A., Chio, I.I., Engle, D.D., Corbo, V., Jager, M.,
Fordham, R.P., Yui, S., Hannan, N.R., Soendergaard, C., Madgwick, A.,
Ponz-Sarvise, M., Tiriac, H., Spector, M.S., et al. (2015). Organoid models of
Schweiger, P.J., Nielsen, O.H., Vallier, L., Pedersen, R.A., Nakamura, T.,
human and mouse ductal pancreatic cancer. Cell 160, 324–338.
et al. (2013). Transplantation of expanded fetal intestinal progenitors contrib-
Camp, J.G., Badsha, F., Florio, M., Kanton, S., Gerber, T., Wilsch-Bräuninger, utes to colon regeneration after injury. Cell Stem Cell 13, 734–744.
M., Lewitus, E., Sykes, A., Hevers, W., Lancaster, M., et al. (2015). Human ce-
Fukuda, M., Mizutani, T., Mochizuki, W., Matsumoto, T., Nozaki, K., Sakamaki,
rebral organoids recapitulate gene expression programs of fetal neocortex
Y., Ichinose, S., Okada, Y., Tanaka, T., Watanabe, M., and Nakamura, T.
development. Proc. Natl. Acad. Sci. USA 112, 15672–15677.
(2014). Small intestinal stem cell identity is maintained with functional Paneth
Chen, K.G., Mallon, B.S., McKay, R.D., and Robey, P.G. (2014). Human plurip- cells in heterotopically grafted epithelium onto the colon. Genes Dev. 28,
otent stem cell culture: considerations for maintenance, expansion, and 1752–1757.
therapeutics. Cell Stem Cell 14, 13–26.
Gallico, G.G., 3rd, O’Connor, N.E., Compton, C.C., Kehinde, O., and Green, H.
Cherry, A.B., and Daley, G.Q. (2012). Reprogramming cellular identity for (1984). Permanent coverage of large burn wounds with autologous cultured
regenerative medicine. Cell 148, 1110–1122. human epithelium. N. Engl. J. Med. 311, 448–451.
Chua, C.W., Shibata, M., Lei, M., Toivanen, R., Barlow, L.J., Bergren, S.K., Gao, D., Vela, I., Sboner, A., Iaquinta, P.J., Karthaus, W.R., Gopalan, A., Dow-
Badani, K.K., McKiernan, J.M., Benson, M.C., Hibshoosh, H., et al. (2014). ling, C., Wanjala, J.N., Undvall, E.A., Arora, V.K., et al. (2014). Organoid cul-
Single luminal epithelial progenitors can generate prostate organoids in cultures derived from patients with advanced prostate cancer. Cell 159, 176–187.
ture. Nat. Cell Biol. 16, 951–961. Grün, D., Lyubimova, A., Kester, L., Wiebrands, K., Basak, O., Sasaki, N.,
Ciancanelli, M.J., Huang, S.X., Luthra, P., Garner, H., Itan, Y., Volpi, S., Lafaille, Clevers, H., and van Oudenaarden, A. (2015). Single-cell messenger RNA
F.G., Trouillet, C., Schmolke, M., Albrecht, R.A., et al. (2015). Infectious dis- sequencing reveals rare intestinal cell types. Nature 525, 251–255.
Cell 165, June 16, 2016 1595

Huang, S.X., Islam, M.N., O’Neill, J., Hu, Z., Yang, Y.G., Chen, Y.W., Mumau, derivation of purified lung and thyroid progenitors from embryonic stem cells.
M., Green, M.D., Vunjak-Novakovic, G., Bhattacharya, J., and Snoeck, H.W. Cell Stem Cell 10, 398–411.
(2014). Efficient generation of lung and airway epithelial cells from human Ma, R., Latif, R., and Davies, T.F. (2015). Human embryonic stem cells form
pluripotent stem cells. Nat. Biotechnol. 32, 84–91. functional thyroid follicles. Thyroid 25, 455–461.
Huang, L., Holtzinger, A., Jagan, I., BeGora, M., Lohse, I., Ngai, N., Nostro, C., Mae, S., Shono, A., Shiota, F., Yasuno, T., Kajiwara, M., Gotoda-Nishimura, N.,
Wang, R., Muthuswamy, L.B., Crawford, H.C., et al. (2015). Ductal pancreatic Arai, S., Sato-Otubo, A., Toyoda, T., Takahashi, K., et al. (2013). Monitoring
cancer modeling and drug screening using human pluripotent stem cell- and and robust induction of nephrogenic intermediate mesoderm from human
patient-derived tumor organoids. Nat. Med. 21, 1364–1371. pluripotent stem cells. Nat. Commun. 4, 1367.
Huch, M., Bonfanti, P., Boj, S.F., Sato, T., Loomans, C.J., van de Wetering, M., Maimets, M., Rocchi, C., Bron, R., Pringle, S., Kuipers, J., Giepmans, B.N.G.,
Sojoodi, M., Li, V.S., Schuijers, J., Gracanin, A., et al. (2013a). Unlimited in vitro Vries, R.G.J., Clevers, H., De Haan, G., Van Os, R., et al. (2016). Long-term
expansion of adult bi-potent pancreas progenitors through the Lgr5/R-spon- in vitro expension of salivary gland stem cells driven by Wnt signals. Stem
din axis. EMBO J. 32, 2708–2721. Cell Reports 6, 1–13.
Huch, M., Dorrell, C., Boj, S.F., van Es, J.H., Li, V.S., van de Wetering, M., Sato, Matano, M., Date, S., Shimokawa, M., Takano, A., Fujii, M., Ohta, Y., Wata-
T., Hamer, K., Sasaki, N., Finegold, M.J., et al. (2013b). In vitro expansion of nabe, T., Kanai, T., and Sato, T. (2015). Modeling colorectal cancer using
single Lgr5+ liver stem cells induced by Wnt-driven regeneration. Nature CRISPR-Cas9-mediated engineering of human intestinal organoids. Nat.
494, 247–250. Med. 21, 256–262.
Huch, M., Gehart, H., van Boxtel, R., Hamer, K., Blokzijl, F., Verstegen, M.M., McCracken, K.W., Catá, E.M., Crawford, C.M., Sinagoga, K.L., Schumacher,
Ellis, E., van Wenum, M., Fuchs, S.A., de Ligt, J., et al. (2015). Long-term cul- M., Rockich, B.E., Tsai, Y.H., Mayhew, C.N., Spence, J.R., Zavros, Y., and
ture of genome-stable bipotent stem cells from adult human liver. Cell 160, Wells, J.M. (2014). Modelling human development and disease in pluripotent
299–312. stem-cell-derived gastric organoids. Nature 516, 400–404.
Jain, R., Barkauskas, C.E., Takeda, N., Bowie, E.J., Aghajanian, H., Wang, Q., Muguruma, K., Nishiyama, A., Ono, Y., Miyawaki, H., Mizuhara, E., Hori, S.,
Padmanabhan, A., Manderfield, L.J., Gupta, M., Li, D., et al. (2015). Plasticity Kakizuka, A., Obata, K., Yanagawa, Y., Hirano, T., and Sasai, Y. (2010).
of Hopx(+) type I alveolar cells to regenerate type II cells in the lung. Nat. Com- Ontogeny-recapitulating generation and tissue integration of ES cell-derived
mun. 6, 6727. Purkinje cells. Nat. Neurosci. 13, 1171–1180.
Jung, P., Sato, T., Merlos-Suárez, A., Barriga, F.M., Iglesias, M., Rossell, D., Muguruma, K., Nishiyama, A., Kawakami, H., Hashimoto, K., and Sasai, Y.
Auer, H., Gallardo, M., Blasco, M.A., Sancho, E., et al. (2011). Isolation and (2015). Self-organization of polarized cerebellar tissue in 3D culture of human
in vitro expansion of human colonic stem cells. Nat. Med. 17, 1225–1227. pluripotent stem cells. Cell Rep. 10, 537–550.
Karthaus, W.R., Iaquinta, P.J., Drost, J., Gracanin, A., van Boxtel, R., Wongvi- Nadauld, L.D., Garcia, S., Natsoulis, G., Bell, J.M., Miotke, L., Hopmans, E.S.,
pat, J., Dowling, C.M., Gao, D., Begthel, H., Sachs, N., et al. (2014). Identifica- Xu, H., Pai, R.K., Palm, C., Regan, J.F., et al. (2014). Metastatic tumor evolution
tion of multipotent luminal progenitor cells in human prostate organoid and organoid modeling implicate TGFBR2 as a cancer driver in diffuse gastric
cultures. Cell 159, 163–175. cancer. Genome Biol. 15, 428.
Nakano, T., Ando, S., Takata, N., Kawada, M., Muguruma, K., Sekiguchi, K.,
Kessler, M., Hoffmann, K., Brinkmann, V., Thieck, O., Jackisch, S., Toelle, B.,
Saito, K., Yonemura, S., Eiraku, M., and Sasai, Y. (2012). Self-formation of op-
Berger, H., Mollenkopf, H.J., Mangler, M., Sehouli, J., et al. (2015). The Notch
tic cups and storable stratified neural retina from human ESCs. Cell Stem Cell
and Wnt pathways regulate stemness and differentiation in human fallopian
10, 771–785.
tube organoids. Nat. Commun. 6, 8989.
Nanduri, L.S., Baanstra, M., Faber, H., Rocchi, C., Zwart, E., de Haan, G., van
Korinek, V., Barker, N., Moerer, P., van Donselaar, E., Huls, G., Peters, P.J.,
Os, R., and Coppes, R.P. (2014). Purification and ex vivo expansion of fully
and Clevers, H. (1998). Depletion of epithelial stem-cell compartments in the
functional salivary gland stem cells. Stem Cell Reports 3, 957–964.
small intestine of mice lacking Tcf-4. Nat. Genet. 19, 379–383.
Nantasanti, S., Spee, B., Kruitwagen, H.S., Chen, C., Geijsen, N., Oosterhoff,
Kurmann, A.A., Serra, M., Hawkins, F., Rankin, S.A., Mori, M., Astapova, I.,
L.A., van Wolferen, M.E., Pelaez, N., Fieten, H., Wubbolts, R.W., et al. (2015).
Ullas, S., Lin, S., Bilodeau, M., Rossant, J., et al. (2015). Regeneration of
Disease Modeling and Gene Therapy of Copper Storage Disease in Canine
thyroid function by transplantation of differentiated pluripotent stem cells.
Hepatic Organoids. Stem Cell Reports 5, 895–907.
Cell Stem Cell 17, 527–542.
O’connor, N.E., Mulliken, J.B., Banks-Schlegel, S., Kehinde, O., and Green, H.
Lancaster, M.A., and Knoblich, J.A. (2014). Organogenesis in a dish: modeling
(1981). Grafting of burns with cultured epithelium prepared from autologous
development and disease using organoid technologies. Science 345,
epidermal cells. Lancet 1, 75–78.
1247125.
Ootani, A., Li, X., Sangiorgi, E., Ho, Q.T., Ueno, H., Toda, S., Sugihara, H., Fu-
Lancaster, M.A., Renner, M., Martin, C.A., Wenzel, D., Bicknell, L.S., Hurles, jimoto, K., Weissman, I.L., Capecchi, M.R., and Kuo, C.J. (2009). Sustained
M.E., Homfray, T., Penninger, J.M., Jackson, A.P., and Knoblich, J.A. (2013). in vitro intestinal epithelial culture within a Wnt-dependent stem cell niche.
Cerebral organoids model human brain development and microcephaly. Nat. Med. 15, 701–706.
Nature 501, 373–379.
Pellegrini, G., Traverso, C.E., Franzi, A.T., Zingirian, M., Cancedda, R., and De
Li, X., Nadauld, L., Ootani, A., Corney, D.C., Pai, R.K., Gevaert, O., Cantrell, Luca, M. (1997). Long-term restoration of damaged corneal surfaces with
M.A., Rack, P.G., Neal, J.T., Chan, C.W., et al. (2014). Oncogenic transforma- autologous cultivated corneal epithelium. Lancet 349, 990–993.
tion of diverse gastrointestinal tissues in primary organoid culture. Nat. Med.
Plaks, V., Brenot, A., Lawson, D.A., Linnemann, J.R., Van Kappel, E.C., Wong,
20, 769–777.
K.C., de Sauvage, F., Klein, O.D., and Werb, Z. (2013). Lgr5-expressing cells
Lindberg, K., Brown, M.E., Chaves, H.V., Kenyon, K.R., and Rheinwald, J.G. are sufficient and necessary for postnatal mammary gland organogenesis.
(1993). In vitro propagation of human ocular surface epithelial cells for trans- Cell Rep. 3, 70–78.
plantation. Invest. Ophthalmol. Vis. Sci. 34, 2672–2679.
Qian, X., Nguyen, H.N., Song, M.M., Hadiono, C., Ogden, S.C., Hammack, C.,
Linnemann, J.R., Miura, H., Meixner, L.K., Irmler, M., Kloos, U.J., Hirschi, B., Yao, B., Hamersky, G.R., Jacob, F., Zhong, C., et al. (2016). Brain-Region-
Bartsch, H.S., Sass, S., Beckers, J., Theis, F.J., et al. (2015). Quantification Specific Organoids Using Mini-bioreactors for Modeling ZIKV Exposure. Cell
of regenerative potential in primary human mammary epithelial cells. Develop- 165, 1238–1254.
ment 142, 3239–3251. Rama, P., Matuska, S., Paganoni, G., Spinelli, A., De Luca, M., and Pellegrini,
Longmire, T.A., Ikonomou, L., Hawkins, F., Christodoulou, C., Cao, Y., Jean, G. (2010). Limbal stem-cell therapy and long-term corneal regeneration.
J.C., Kwok, L.W., Mou, H., Rajagopal, J., Shen, S.S., et al. (2012). Efficient N. Engl. J. Med. 363, 147–155.
1596 Cell 165, June 16, 2016

Ren, W., Lewandowski, B.C., Watson, J., Aihara, E., Iwatsuki, K., Bachmanov, Takasato, M., Er, P.X., Becroft, M., Vanslambrouck, J.M., Stanley, E.G.,
A.A., Margolskee, R.F., and Jiang, P. (2014). Single Lgr5- or Lgr6-expressing Elefanty, A.G., and Little, M.H. (2014). Directing human embryonic stem cell
taste stem/progenitor cells generate taste bud cells ex vivo. Proc. Natl. differentiation towards a renal lineage generates a self-organizing kidney.
Acad. Sci. USA 111, 16401–16406. Nat. Cell Biol. 16, 118–126.
Rheinwald, J.G., and Green, H. (1975). Serial cultivation of strains of human Takasato, M., Er, P.X., Chiu, H.S., Maier, B., Baillie, G.J., Ferguson, C., Parton,
epidermal keratinocytes: the formation of keratinizing colonies from single R.G., Wolvetang, E.J., Roost, M.S., Chuva de Sousa Lopes, S.M., and Little,
cells. Cell 6, 331–343. M.H. (2015). Kidney organoids from human iPS cells contain multiple lineages
Rios, A.C., Fu, N.Y., Lindeman, G.J., and Visvader, J.E. (2014). In situ identifi- and model human nephrogenesis. Nature 526, 564–568.
cation of bipotent stem cells in the mammary gland. Nature 506, 322–327. Takebe, T., Sekine, K., Enomura, M., Koike, H., Kimura, M., Ogaeri, T., Zhang,
R.R., Ueno, Y., Zheng, Y.W., Koike, N., et al. (2013). Vascularized and func-
Rock, J.R., Onaitis, M.W., Rawlins, E.L., Lu, Y., Clark, C.P., Xue, Y., Randell,
tional human liver from an iPSC-derived organ bud transplant. Nature 499,
S.H., and Hogan, B.L. (2009). Basal cells as stem cells of the mouse trachea
481–484.
and human airway epithelium. Proc. Natl. Acad. Sci. USA 106, 12771–12775.
Treutlein, B., Brownfield, D.G., Wu, A.R., Neff, N.F., Mantalas, G.L., Espinoza,
Sakaguchi, H., Kadoshima, T., Soen, M., Narii, N., Ishida, Y., Ohgushi, M.,
F.H., Desai, T.J., Krasnow, M.A., and Quake, S.R. (2014). Reconstructing line-
Takahashi, J., Eiraku, M., and Sasai, Y. (2015). Generation of functional hippo-
age hierarchies of the distal lung epithelium using single-cell RNA-seq. Nature
campal neurons from self-organizing human embryonic stem cell-derived
509, 371–375.
dorsomedial telencephalic tissue. Nat. Commun. 6, 8896.
van de Wetering, M., Francies, H.E., Francis, J.M., Bounova, G., Iorio, F.,
Sato, T., and Clevers, H. (2013). Growing self-organizing mini-guts from a sin-
Pronk, A., van Houdt, W., van Gorp, J., Taylor-Weiner, A., Kester, L., et al.
gle intestinal stem cell: mechanism and applications. Science 340, 1190–1194.
(2015). Prospective derivation of a living organoid biobank of colorectal cancer
Sato, T., Vries, R.G., Snippert, H.J., van de Wetering, M., Barker, N., Stange, patients. Cell 161, 933–945.
D.E., van Es, J.H., Abo, A., Kujala, P., Peters, P.J., and Clevers, H. (2009). Sin-
von Furstenberg, R.J., Gulati, A.S., Baxi, A., Doherty, J.M., Stappenbeck, T.S.,
gle Lgr5 stem cells build crypt-villus structures in vitro without a mesenchymal
Gracz, A.D., Magness, S.T., and Henning, S.J. (2011). Sorting mouse jejunal
niche. Nature 459, 262–265.
epithelial cells with CD24 yields a population with characteristics of intestinal
Sato, T., Stange, D.E., Ferrante, M., Vries, R.G., Van Es, J.H., Van den Brink, stem cells. Am. J. Physiol. Gastrointest. Liver Physiol. 300, G409–G417.
S., Van Houdt, W.J., Pronk, A., Van Gorp, J., Siersema, P.D., and Clevers, Wang, F., Scoville, D., He, X.C., Mahe, M.M., Box, A., Perry, J.M., Smith, N.R.,
H. (2011). Long-term expansion of epithelial organoids from human colon, Lei, N.Y., Davies, P.S., Fuller, M.K., et al. (2013). Isolation and characterization
adenoma, adenocarcinoma, and Barrett’s epithelium. Gastroenterology 141, of intestinal stem cells based on surface marker combinations and colony-
1762–1772. formation assay. Gastroenterology 145, 383–395.e1–e21.
Schwank, G., Koo, B.K., Sasselli, V., Dekkers, J.F., Heo, I., Demircan, T., Sa- Wang, X., Yamamoto, Y., Wilson, L.H., Zhang, T., Howitt, B.E., Farrow, M.A.,
saki, N., Boymans, S., Cuppen, E., van der Ent, C.K., et al. (2013). Functional Kern, F., Ning, G., Hong, Y., Khor, C.C., et al. (2015). Cloning and variation
repair of CFTR by CRISPR/Cas9 in intestinal stem cell organoids of cystic of ground state intestinal stem cells. Nature 522, 173–178.
fibrosis patients. Cell Stem Cell 13, 653–658.
Watson, C.L., Mahe, M.M., Múnera, J., Howell, J.C., Sundaram, N., Poling,
Sinagoga, K.L., and Wells, J.M. (2015). Generating human intestinal tissues H.M., Schweitzer, J.I., Vallance, J.E., Mayhew, C.N., Sun, Y., et al. (2014).
from pluripotent stem cells to study development and disease. EMBO J. 34, An in vivo model of human small intestine using pluripotent stem cells. Nat.
1149–1163. Med. 20, 1310–1314.
Spence, J.R., Mayhew, C.N., Rankin, S.A., Kuhar, M.F., Vallance, J.E., Tolle, Wong, A.P., Bear, C.E., Chin, S., Pasceri, P., Thompson, T.O., Huan, L.J.,
K., Hoskins, E.E., Kalinichenko, V.V., Wells, S.I., Zorn, A.M., et al. (2011). Ratjen, F., Ellis, J., and Rossant, J. (2012). Directed differentiation of human
Directed differentiation of human pluripotent stem cells into intestinal tissue pluripotent stem cells into mature airway epithelia expressing functional
in vitro. Nature 470, 105–109. CFTR protein. Nat. Biotechnol. 30, 876–882.
Stange, D.E., Koo, B.K., Huch, M., Sibbel, G., Basak, O., Lyubimova, A., Xia, Y., Nivet, E., Sancho-Martinez, I., Gallegos, T., Suzuki, K., Okamura, D.,
Kujala, P., Bartfeld, S., Koster, J., Geahlen, J.H., et al. (2013). Differentiated Wu, M.Z., Dubova, I., Esteban, C.R., Montserrat, N., et al. (2013). Directed dif-
Troy+ chief cells act as reserve stem cells to generate all lineages of the stom- ferentiation of human pluripotent cells to ureteric bud kidney progenitor-like
ach epithelium. Cell 155, 357–368. cells. Nat. Cell Biol. 15, 1507–1515.
Tadokoro, T., Wang, Y., Barak, L.S., Bai, Y., Randell, S.H., and Hogan, B.L. Yin, X., Farin, H.F., van Es, J.H., Clevers, H., Langer, R., and Karp, J.M. (2014).
(2014). IL-6/STAT3 promotes regeneration of airway ciliated cells from basal Niche-independent high-purity cultures of Lgr5+ intestinal stem cells and their
stem cells. Proc. Natl. Acad. Sci. USA 111, E3641–E3649. progeny. Nat. Methods 11, 106–112.
Taguchi, A., Kaku, Y., Ohmori, T., Sharmin, S., Ogawa, M., Sasaki, H., and Yui, S., Nakamura, T., Sato, T., Nemoto, Y., Mizutani, T., Zheng, X., Ichinose,
Nishinakamura, R. (2014). Redefining the in vivo origin of metanephric nephron S., Nagaishi, T., Okamoto, R., Tsuchiya, K., et al. (2012). Functional engraft-
progenitors enables generation of complex kidney structures from pluripotent ment of colon epithelium expanded in vitro from a single adult Lgr5+ stem
stem cells. Cell Stem Cell 14, 53–67. cell. Nat. Med. 18, 618–623.
Cell 165, June 16, 2016 1597

STAR Methods—a new methods section designed with authors
and readers in mind—are now available in Cell and Cell Systems.
Learn more about how we’re empowering methods.
Visit www.cell.com/star-methods today!
Leading Edge
Review
The Cellular Phase of Alzheimer’s Disease

Bart De Strooper1,2,3,* and Eric Karran2,3,4
1VIB Center for the Biology of Disease, VIB-Leuven, 3000 Leuven, Belgium
2Center for Human Genetics, Universitaire Ziekenhuizen and LIND, KU Leuven, 3000 Leuven, Belgium
3Institute of Neurology, University College London, Queen Square, WC1N 3BG London, UK
4Alzheimer Research UK, Cambridge, CB21 6AD, UK
*Correspondence: bart.destrooper@cme.vib-kuleuven.be
The amyloid hypothesis for Alzheimer’s disease (AD) posits a neuron-centric, linear cascade initi-
ated by Ab and leading to dementia. This direct causality is incompatible with clinical observations.
We review evidence supporting a long, complex cellular phase consisting of feedback and feedfor-
ward responses of astrocytes, microglia, and vasculature. The field must incorporate this holistic
view and take advantage of advances in single-cell approaches to resolve the critical junctures
at which perturbations initially amenable to compensatory feedback transform into irreversible,
progressive neurodegeneration.
Introduction in a cascade initiated by Ab deposition and leading progressively

Alzheimer’s disease (AD) impairs cognition, memory, and lan- to Tau pathology, synaptic dysfunction, inflammation, neuronal
guage and causes dementia. In 1907, Aloı̈s Alzheimer described loss, and, ultimately, dementia.
the extensive distribution of neuronal tangles and amyloid pla- The linearity of the cascade remains very controversial. For
ques in brain that, together with astrogliosis, neuronal dystrophy, instance, exploration of the direct link between Ab and neurotox-
neuronal loss, and vascular alterations, constitute the hallmarks icity has resulted in a rather confusing literature with at least ten
of the disorder. different molecular mechanisms and receptors (see Figure 1)
For over 20 years, the ‘‘amyloid cascade hypothesis’’ has pro- (Benilova et al., 2012). It is debatable whether they all equally
vided the main theoretical construct for AD (Hardy and Selkoe, contribute to the disease process in humans, as many of the sup-
2002). It proposes amyloid plaques or their major constituents, porting studies used non-physiologically high concentrations of
the Ab-peptides, as the direct cause of the progressive neurode- Ab or transgenic mice significantly overexpressing APP (see
generation. The concept that a biochemical/biophysical process Nilsson et al., 2014). However, an alternative explanation for
underlies the complex cognitive alterations has transformed AD the richness of these findings is that different mechanisms or
research, replacing earlier descriptive studies with a molecular, even the same mechanism exert multiple effects at different
mechanistic view. This has led to diagnostic and therapeutic stages of disease progression. For instance, the interaction of
trials for a disease believed before to be either incurable or an Ab with the receptor for advanced glycation end products
inevitable consequence of aging. The tremendous increase in (RAGE) causes oxidative stress in neurons but enhances inflam-
knowledge on the molecular biology, pathophysiology, and diag- matory responses in microglia and is involved in endothelial cells
nosis of AD is exciting and holds promise for future prevention in ‘‘reversed’’ transport of Ab across the blood-brain barrier
and therapies but also starts to erode the assumptions of its (BBB) (Deane et al., 2012). Thus, the neuron-centric view has
main theoretical foundation. to be expanded to one that considers the contribution of different
To understand the disease, the well-studied biochemistry cell types, their interactions with each other, and the gradual
needs to be integrated into the complex cellular context of the evolution of the disease.
brain. We examine here how a failing vascular system contrib- Preclinical or prodromal AD stages have been defined by clini-
utes to disease progression; how inhibitory and excitatory cians to underscore the concept that AD starts insidiously many
neurons, neuronal networks, microglia, astroglia, and, finally, decades before full dementia becomes apparent (Carrillo et al.,
oligodendrocytes all contribute to a complex cellular phase of 2013; Dubois et al., 2014). The amyloid cascade hypothesis pro-
the disease evolving over decades, and how initial benign reac- vides no explanation for the silent incubation period of AD. Even
tions ultimately become chronic, resulting in an irreversible dys- in familial AD (FAD), where the role of Ab is established because
homeostasis of the brain. Emerging approaches could provide of causative mutations in APP and Presenilin, the amyloid hy-
us with a more holistic understanding of the spatial, temporal, pothesis is found wanting. Why, for instance, does abnormal
and cellular aspects of the disease process. We start with a brief Ab production, already present from conception, manifest its
critique of the amyloid cascade hypothesis itself. toxicity only in midlife? Clearly the complex compensation
mechanisms in the brain that maintain almost normal cognitive
The Amyloid Hypothesis performance for decades should be accommodated into any
The amyloid hypothesis for AD is a neuron-centric, linear, and theoretical construct of the disease. In our view, AD only starts
quantitative model postulating direct cause and consequences when a pathological cellular reaction is initiated.
Cell 164, February 11, 2016 ª2016 Elsevier Inc. 603

Figure 1. Ab Receptors and Ab Toxicity Mechanisms
All membrane receptors implied in Ab pathogenic mechanisms are indicated. Some receptors are only indirectly affected by Ab, inferred by physiological
alterations of cells exposed to Ab. The second series are receptors shown to directly bind Ab (see Benilova et al., 2012).
The quantitative aspects of the hypothesis (the concept that these proteases whereby the C-terminal region of long Ab is
more amyloid plaques, or, in more recent versions, more Ab42, progressively cleaved. The mutations might destabilize the
is causing disease) imply that quantitative lowering of Ab-pla- three-dimensional structure of the g-secretases, which might
ques or Ab-oligomers should be sufficient to halt progression explain the premature release of longer (insufficiently ‘‘trimmed’’)
of AD (critically discussed in Karran et al., 2011). The discrep- hydrophobic Ab42, Ab43, and, hypothetically, Ab>43, peptides
ancy between amyloid load and disease symptoms in sporadic (Chávez-Gutiérrez et al., 2012; Szaruga et al., 2015; Saito
AD (SAD) has, however, been pointed out in the past: ‘‘it seems et al., 2011). In this view, even lowering of total Ab could still
as if the Ab plaques appear at the wrong time and in the wrong be pathogenic if the Ab produced is longer than Ab40. Such
places with respect to the clinical dementia’’ (Mesulam, 1999). long Ab, even in low amounts, could provide seeds for further
Less appreciated is the fact that the quantitative concept is nucleation and amyloidosis. Thus, in contrast to what is
equally not tenable for FAD. Most Presenilin (PSEN1) mutations assumed in the amyloid hypothesis, the pathological effects of
do not increase Ab generation. PSEN1 is part of the g-secretase the PSEN1 mutations on Ab production are qualitative and not
complex, and mutations affect mainly the ‘‘trimming’’ function of quantitative (Kuperstein et al., 2010; Szaruga et al., 2015).
604 Cell 164, February 11, 2016 ª2016 Elsevier Inc.

The Biochemical Phase in AD Causes Proteostatic et al., 2014) is mild. The effects of age-associated Ab accumula-
Stress and Homeostatic Cellular Responses tion are initially also mild—they are found in many aged people
The ‘‘biochemical phase’’ of AD is characterized by abnormal without causing clinical symptoms of brain dysfunction. Ab
aggregation of Ab and potentially abnormal processing of APP; stress might enhance the progression of Tau pathology (Khan
formation of Tau conformers followed by hyper phosphorylation; et al., 2014), but described molecular links are multiple and indi-
the seeding and propagation in a prionoid fashion of these rect. Alternatively, Ab-stress and PART could be considered in-
abnormally folded proteins (Walker and Jucker, 2015); the gener- dependent stressors that each cause a slowly evolving, relatively
ation of oligomeric amyloids (Abo- and Tau-oligomers); and the benign, age-related neurodegenerative process with no or only
generation of amyloid plaques and neuronal tangles. The mild symptomatology. Only when they come together and recip-
different conformations of Ab and Tau exert ‘‘proteopathic’’ or rocally augment their deleterious effects is a more aggressive
‘‘aggregrate’’ stress on different cells of the brain, i.e., they disease manifested with multiple cellular effects. Even in pure
interact with proteins and membranes and interfere with FAD, this scenario is conceivable: the abnormal Ab generated
signaling and other functions. They also bind APP and Tau, by the mutations becomes only part of a disease process
which disturbs the normal activities of these proteins in neuro- when the ‘‘natural’’ process of progressive accumulation of neu-
transmission (Palop and Mucke, 2010). The temporal sequence rofibrillar tangles has reached a threshold and both pathologies
of these events is undefined. Age-related deficiencies of the pro- start to interact. The rather stereotypical evolution of the disease
teostatic network could cause accumulation of aggregating pro- is compatible with this view (Karran et al., 2011).
teins; conversely, protein aggregates might disturb the complex In the current Review, we try to go beyond the discussion—
cellular molecular network that sustains protein folding (Labba- which is nevertheless important—of the biochemistry and
dia and Morimoto, 2015). The lysosomal/endolysosomal system toxicity of Ab and Tau (De Strooper, 2010). We see AD as a com-
and, in particular, autophagy is a key regulator of this proteo- plex process consisting of different interacting phases encom-
pathic stage (Nixon, 2013). In AD brains, a striking increase in passing feedback and feedforward mechanisms between cells,
classic autophagosomes, featuring a double membrane that en- which, after many years, lead to dementia. Most importantly, in
capsulates heterogenous intracellular components, is seen in an this view, Ab and Tau accumulations in SAD are considered
early stage of the disorder (Nixon, 2013). Deletion of key macro- risk factors—i.e., even if their accumulation is a sign of increased
autophagy components such as ATG5 or ATG7 in neurons in proteopathic stress, additional cellular factors are determinative
mice causes motor deficits, ubiquitinated neuronal inclusions, in the evolution toward dementia. Instead of Ab and Tau, the real
and neurodegeneration (Komatsu et al., 2006). In Niemann- causes of sporadic disease are upstream of these proteopathies
Pick’s disease type C, loss of function of the NPC1 or NPC2 and are likely manifold, with aging being the major driver. Never-
genes leads to late endosomal/lysosomal accumulation of lipids theless, the proteopathic stress of the biochemical phase results
and defective autophagocytosis (Elrick et al., 2012). Affected in the cellular phase of AD. The clinical phase of the disease is
neurons display tau inclusions that are indistinguishable from initiated when the cellular reaction can no longer maintain ho-
those in AD. Thus, abnormalities in autophagy are linked to tau meostasis (Figure 2).
pathology. The genetic support for a role of autophagy in AD
compared to Parkinson disease, for example, remains surpris- Defective Clearance Mechanisms Are Part of the Initial
ingly limited, however (Lambert et al., 2013) Cellular Phase
None of the initial effects of proteopathy should be considered Defective clearance of Ab (and Tau), considered to underlie the
overwhelming or irreversible, as brain cells can apparently majority of sporadic AD cases, illustrates the complexity of the
survive this stress for many years. The elicited responses are cellular response. Many proteases can break down Ab peptides
physiological and can be cell or non-cell autonomous. They in the biochemical phase (Figure 2) (De Strooper, 2010), but, as
can be biochemical, maintaining homeostasis in the proteostatic shown by injecting labeled Ab or Tau in the brain parenchyma
network (Labbadia and Morimoto, 2015), or functional, involving and following its elimination (Iliff et al., 2014), it is now clear
various synaptic plasticity mechanisms in which astroglia may that a large part of normal Ab turnover depends on bulk flow
play crucial roles. Inflammatory responses contribute initially to via the perivascular circulation and the glymphatic system in
sustaining homeostasis. It is only when these compensating the brain (Tarasoff-Conway et al., 2015). Genetic risk associated
mechanisms turn into chronic, irreversible, and pathological pro- with APOE4, PICALM, and APOJ (or CLU) is thought to affect Ab
cesses that disease progresses inexorably. The transition of an clearance (Verghese et al., 2013; Zhao et al., 2015).
early reversible to a chronic, irreversible, autonomous cellular The cellular component of the clearance system is complex.
response, likely no longer dependent on the initial Ab and Tau The BBB consists of tightly connected endothelial cells and as-
aggregate stressors, represents a critical phase in the disease troglia end-feet. Pericytes and smooth muscle cells provide con-
process. tractile elements. The glia barrier is relatively open because of
The separate locations of Tau and amyloid pathology and their gap junctions allowing passage of Ab and Tau. The endothelial
natural progression in the brain (Braak and Braak, 1995; Thal barrier is, however, not permeable, and therefore, specialized
et al., 2002) suggest that the two biochemical pathologies prog- transport proteins (receptors of the LDL receptor family, e.g.,
ress relatively independently from each other (Mesulam, 1999; LRP1&2, RAGE, ABC transporter ABCB1 or P-glycoprotein,
Small and Duff, 2008). Tangles are universally found in all aged ABCA1, and Ab free or bound to proteins such as a2-macroglob-
people and are already present early in life (Braak and Del Tre- ulin, APOJ, or APOE particles) are responsible for Ab (and Tau)
dici, 2011). This primary age-related tauopathy or PART (Crary egress to the circulation. For instance, heterozygote Picalm+/

Figure 2. The Biochemical, Cellular, and Clinical Phases of AD
The transition of biochemical to cellular and cellular to clinical phases are indicated. The cellular phase is complex and evolves gradually over two decades from
reversible physiological reactions upon proteopathic stress to irreversible immune and electrophysiological compensation mechanisms that disturb the normal
homeostasis of the brain. The effects are cell and non-cell autonomous and occur in the context of neurovascular or glioneuronal units (see main text). Processes
are derived from GWAS studies (clearance dysfunction, lipid metabolism, immune response, protein trafficking) and are indicated in different colors. Genes are in
italics when genetic evidence links them to AD. Other genes (LRP1, RAGE and others) are functionally implicated in the disease process. Each gene is associated
to a color code indicating the pathways in which they operate. The picture is a late self-portrait from the German painter William Utermohlen, who suffered from
AD. Image courtesy of Chris Boı̈cos Fine Arts, Paris.
mice show decreased LRP1-mediated clearance of Ab via the in AD are a recurrent theme in this Review and illustrate the
BBB. Although neurons with downregulated Picalm produce importance of placing the biochemical process into the context
less Ab, accelerated amyloidosis and worsening in behavioral of the cellular phase.
parameters is seen in these mice: this could be rescued via In contrast to Ab, much less is known regarding Tau clearance.
adenoviral expression of PICALM in the endothelium (Zhao Previously, Tau pathology was seen as a cell-autonomous phe-
et al., 2015). PICALM proteolysis has also been associated nomenon, and studies of Tau clearance were predominantly
with Tau-tangle formation and disturbed endocytosis in SAD focused on intracellular processes such as autophagy. However,
(Ando et al., 2013). The multifactorial effects of genes involved Tau pathology can progress through the brain in a prion-like

fashion (Walker and Jucker, 2015), and Tau is secreted after Neurons and Neuronal Circuitry as Part of the Cellular
increased electrical activity (Yamada et al., 2014). Some Phase of AD
secreted Tau is cleared by the glymphatic flow (Iliff et al., In the prevailing neuron-centric view, dysfunction of the brain is
2014); thus, it is reasonable to predict that Tau gets cleared mediated by decrements in synaptic plasticity: e.g., long-term
from the brain via similar cell-regulated bulk flow pathways potentiation and depression, changes in homeostatic scaling,
as Ab. However, the role of the vasculature in AD goes far and disruption of neuronal connectivity. These alterations are
beyond clearance. currently often studied by recording synaptic function in
neuronal cell and hippocampal slice models treated with Ab pep-
The Neurovascular Unit and the Vascular Hypothesis tides, either synthetically generated, isolated from ‘‘natural’’
for AD brain material, or expressed in transgenic mouse models. These
The term ‘‘neurovascular unit’’ is used to describe the close models are too acute to mimic the progressive cellular disruption
anatomic and functional interrelationship between the cellular that occurs in AD brains over many years.
components of the brain. The vascular hypothesis proposes The complexity of how neurons are progressively affected by
that initial vascular damage precipitates AD. Hypoperfusion Ab stress has been addressed by a few groups. In a series of
and hypoxia are one aspect of the problem, but the associated landmark publications, Palop and Mucke studied how Ab stress
breakdown of the BBB also results in accumulation of neuro- results in excitatory, (non-convulsive) seizure activity in neuronal
toxic serum proteins in brain, inflammation, vascular, and syn- networks (Palop and Mucke, 2010). The study nicely illustrates
aptic dysfunction that lead secondarily to defects in Ab and Tau how work in preclinical animal models can translate into novel in-
metabolism and clearance, which in turn cause vascular prob- sights relevant to human pathogenesis: epileptic seizures and
lems such as cerebral amyloid angiopathy (CAA) (Zlokovic, neuronal hyperexcitability in AD patients became a recognized
2011). CAA is caused by widespread deposition of Ab in the part of the condition (Bakker et al., 2012). To explain the unex-
vasculature, resulting in fibrinoid necrosis, microaneurysms, pected hyper-reactivity, they looked to cellular alterations and
and bleedings. Co-existence of AD and vascular pathology in- found loss of calbindin expression and ectopic expression of
creases the risk for dementia considerably (Attems and Jellin- the inhibitory neuropeptide Y (NPY) in granule cells and mossy
ger, 2014). fibers, and GABAergic synaptic sprouting in the molecular layer
Many functional studies support the central postulate of the emanating from NPY- and somatostatin positive interneurons.
vascular hypothesis—i.e., the primacy of vascular lesions in They also found a reduction in the voltage-gated sodium channel
driving the AD phenotype (Zlokovic, 2011). Mice that express subunit Nav1.1 in parvalbumin cells, leading to defective inhibi-
the major genetic risk factor for AD, human APOE4 (on a tory input that could be rectified by restoring channel expression
mouse ApoE / background), show a curious degenerative (Verret et al., 2012). Their data suggest that increased Ab stress
response of endothelial cells and pericytes mediated by cyclo- in AD leads to a complex pattern of neuronal actions and
phylin A and NFkb that leads to damage of the BBB (Bell et al., reactions, causing aberrant excitatory network activity and
2012), which is also seen in human APOE4 carriers. Other ge- compensatory inhibitory responses involving learning and mem-
netic manipulations of receptors, transporters, and cells sup- ory circuitry (Palop and Mucke, 2010). These fascinating studies
port the idea that early alterations in the BBB could play a are but a first step to address the complex cellular processes in
major role in the initiation of AD. For example, AD patients AD. For example, recent work suggests that reactive astroglia in
typically display diminished glucose uptake in the brain as these mice secrete GABA, which paradoxically could contribute
measured by 2-[18F]fluoro-2-deoxy-d-glucose positron emis- to seizure-like activity by suppressing inhibitory input on excit-
sion tomography (FDG-PET) (Protas et al., 2013). Interestingly, atory circuitry (Jo et al., 2014).
heterozygous deficiency of GLUT1 (Slc2a1+/ ) in endothelium Other mouse models also show that the neuronal reaction to
accelerated degenerative changes in the APPsw mouse model Tau and Ab proteopathic stress is complex and not just simple
for AD (Winkler et al., 2015). Degeneration of pericytes or cell death. Pyramidal cells in the frontal cortex show phases of
vascular smooth muscle cells under hypoxic stress also leads hypo- and/or hyperexcitability in Tau and Ab mouse models
to neurodegeneration and Ab clearance problems (Winkler (Menkes-Caspi et al., 2015). Oscillatory network activity is
et al., 2014), suggesting that atherosclerosis and hypoperfusion altered in many AD models. Finally, the activity of hippocampal
of the brain might be an upstream cause in a subgroup of pa- place cells that normally show a robust relation to animal loca-
tients. In models of traumatic brain injury, the clearance flow is tion in the environment lose spatial selectivity in AD mice
slowed down, which might contribute to accumulation of (Cheng and Ji, 2013). Importantly, the recorded alterations
secreted Tau (Iliff et al., 2014). Sleep increases, and aging are likely due to a mixture of direct and indirect effects of Ab
and/or stiffening of the arterial wall slows down the glymphatic on neurons and glial cells, synaptic dysfunctions that impact
clearance flow (Tarasoff-Conway et al., 2015). While it seems local network activity, and afferent activity in downstream tar-
that vascular defects lead to amyloid accumulation, it may gets and compensatory mechanisms. Disease progression
also be the case that Ab accumulation can lead to vascular likely entails an initial localized and acute direct or indirect
defects as seen with some clinical APP mutations that manifest effect of Ab species on cell and synaptic function, where
as CAA. Thus, Ab is both cause and consequence of vascular distinct cell types may have a varying degree of sensitivity.
pathology, and this destructive cycle is an important part The early local changes in network activity patterns would
of the cellular action and reaction that underlies the cellular then progressively affect connected brain structures, followed
phase of AD. by the spread of pathology.

Figure 3. Molecular and Cellular Links between Astroglia and AD
Astroglia are central in the cellular phase of AD. They interact directly with Ab and display normal synaptic and metabolic responses in AD. They also display
reactive and atrophic responses in AD. Major pathways are schematically represented.
An important caveat is that functional disturbances in mice dal neurons that become reduced only in the very advanced
are typically occurring in the absence of severe tangle forma- state of AD (Mufson et al., 2015).
tion and neuronal loss that characterizes AD dementia. Some
investigators therefore severely doubt the validity of mouse Astrocytes Are Central Players in the Cellular Phase
experiments: a more moderate and constructive view is that of AD
the alterations in these mice recapitulate early alterations In contrast to neurons, the astroglia population is strikingly
occurring in incipient AD brain (Zahs and Ashe, 2010). In that under-investigated in AD-related studies. Each astrocyte de-
sense, they are very relevant to the study of the early cellular ploys many fine processes to contact up to 140,000 synapses
phase of AD. in the CA1 region of rat hippocampus (Bushong et al., 2002).
Morphological and biochemical studies of human brain mate- With their end feet, they also contact blood vessels, forming
rial are limited to evolved stages of the disease. However, they the glia limitans of the BBB (Figure 2). The astrocytes occupy
too demonstrate the complex cellular phase in AD. Loss of inner- anatomically and functionally discrete micro-domains (Bushong
vation by the entorhinal cortex (EC) of the outer molecular layer of et al., 2002), and arrays of astrocyte and neurons are organized
the dentate gyrus is associated with extensive sprouting of around capillaries, providing the basis for higher-order neuro-
cholinergic innervation (Mufson et al., 2015). Loss of input via vascular units (Allen, 2014).
the Schaffer collaterals is associated initially with increases in The metabolic roles of astroglia are well recognized (Figure 3).
length and complexity of the basal dendritic trees of CA1 pyrami- Importantly, astroglia and oligodendrocytes synthesize all brain

cholesterol and secrete apolipoprotein E and apolipoprotein J expression of intermediate filaments (glial fibrillar acidic protein
that are loaded with cholesterol by ATP-binding cassette trans- [GFAP], vimentin, and nestin). It is a physiological response
porters—in particular, ABCA1. Synaptogenesis is promoted by and should, a priori, be considered a protective reaction; how-
glia-derived cholesterol, but there is also complex overlap be- ever, in AD, it could contribute directly to defective clearance
tween the lipid metabolism function of these various genes and because astrogliosis results in downregulation of Aquaporin 4
the clearance of Ab over the BBB and blood-CSF barriers. Lipo- interfering with normal glymphatic flow. Astrogliosis is a relatively
protein metabolism is strongly associated with risk for AD as early event seen before the appearance of Ab deposits in an AD
repeatedly suggested in GWAS (Karch and Goate, 2015). mouse model (Heneka et al., 2005). PET studies measuring
Astroglia play an active role in synapse formation and synaptic monoamine oxidase B (MaoB) activity in astrocytes confirm
strength regulation and in synchronizing and integrating the that astrogliosis is an early event in patients, with the largest sig-
many synaptic processes under their control. Astroglia possess nals seen in prodromal AD (Carter et al., 2012). The relationship
receptors and cognate signaling machinery for glutamate, between amyloidosis, tau pathology, and astrogliosis remains
GABA, adenosine, noradrenaline, acetylcholine, and endocan- enigmatic.
nabinoids and respond to neurotransmitter release with calcium Experiments in various mouse models suggest that the astro-
signaling (Figure 3). They are thus well equipped to monitor glial reaction does not lead to astroglial scars as seen in injury
neuronal activity. They also actively secrete gliotransmittors models—i.e., the hypertrophy reflects morphological thickening
like ATP, glutamate, and D-serine that modulate activity of of processes and soma, and the astroglia sustain support for the
neuronal receptors (Allen, 2014). In fact, astroglia participate in synapses in their unit. Indeed, attenuation of astrocyte activation
the ‘‘tripartite synapse,’’ providing a third component to the clas- (by deleting Gfap and vimentin genes) accelerates plaque path-
sical ‘‘bipartite’’ flow between pre- and postsynaptic neurons. ogenesis in APP/PS1 mice (Kraft et al., 2013). Moreover, a
Astrocytes thus both sense and modulate synaptic output and marked increase in dystrophic neurites was seen, indicating
may exert an integrating role in the overall computing function the protective nature of the astrogliosis reaction in the mice.
of the brain (Allen, 2014; Gittis and Brasier, 2015). Astrocytes Another group, however, found the opposite effect in similar
are also involved in elimination of synapses via phagocytic re- models (Kamphuis et al., 2015), and whether astrogliosis is
ceptors MEG10 and MERT (Chung et al., 2013) or via guiding beneficial or harmful remains an open question. Indeed, it might
microglial cells to synapses that are pruned involving the com- be that the answer depends upon the stage of the cellular phase
plement pathway. Recently, it was shown how this might operate of AD and the brain region affected (Olabarria et al., 2010). The
in AD: NFkB is upregulated by Ab (Lian et al., 2015) and induces paradoxes in the different studies will probably only be resolved
C3 release from the astrocytes. C3 binding to the G-protein- when the progressive and divergent cellular responses in AD are
linked receptor C3aR on neurons leads to alterations of dendritic systematically and comprehensively delineated.
morphology and network dysfunction. A C3aR antagonist im- Atrophy of astroglia, characterized by reduced branching, is
proves performance of AD mice in the Morris water maze (Lian also seen at 1 month in the entorhinal cortex and at 3 months
et al., 2015). in the prefrontal cortex of triple transgenic AD mice (Kulijewicz-
Astrocytes are intimately involved in Ab catabolism, and Ab Nawrot et al., 2012). This could lead to defective metabolic sup-
peptides affect their metabolic phenotypes. Astroglia are acti- port and shrinking of their territorial domain, leading to synaptic
vated, express inflammatory markers early (Heneka et al., dysfunction. Atrophy of astroglia was independently observed in
2005), and display abnormal synchronous Ca2+ transients over another AD mouse (Beauquis et al., 2014). Studies linking
long distances upon Ab exposure (Kuchibhotla et al., 2009). morphological changes to functional alterations are limited. Jo
Elevated Ca2+ increases 1,4,5-triphosphate (IP3) signaling and et al. (2014) provide insight in how MaoB (which is upregulated
N-cadherin expression, which underlies astrogliosis. Likely in AD patients [Carter et al., 2012]) increases GABA in reactive,
several other signaling pathways are involved as well, and astro- GFAP-positive astroglia in the dentate gyrus of APP/PS1 mice.
glial functionality should be considered when investigating Release of GABA causes a tonic inhibition of perforant path/
mechanisms of Ab toxicity in brain. dentate granule cell synapses. Two other studies link the Gs-pro-
Local ablation of astrocytes in specific domains of the mouse tein-coupled adenosine receptor A2A expressed on astroglia to
brain causes neurodegeneration. This is partially because of memory defects in Tau- and APP-mice (Orr et al., 2015). The
increased excitotoxicity, as glutamate is no longer efficiently A2A-receptor is increased in AD brain, and these new studies
removed. However, the number of astrocytes is not markedly demonstrate that specific antagonists can restore memory
reduced in aging or AD brains (Pelvig et al., 2003) nor in various (and possibly decrease Tau-pathology) in models for AD.
AD mice (Olabarria et al., 2010). The contribution of astrocytes to
AD neurodegeneration is thus clearly more complex than simple Microglia and Inflammation: Beneficial or AD Promoting
‘‘astrodegeneration’’ and encompasses a whole spectrum of Effects?
beneficial and harmful reactions (Figure 3). Microglia, the phagocytic cells of the CNS, comprise between
5% and 12% of the cell population in the adult mouse brain
Reactive, Degenerating, and Atrophic Astrocytes in AD (Lawson et al., 1990) and are a recognized part of the cellular
The ‘‘reactive’’ astrogliosis in AD was noticed by Dr. Alzheimer phase of AD. Seminal observations by McGeer et al. (1988)
himself. Reactive astrogliosis is a complex and dynamic demonstrated their intimate association with senile plaques.
response to brain damage (Burda and Sofroniew, 2014) charac- Since then, their role and, more generally, the role of inflamma-
terized by astrocytic hypertrophy, proliferation, and augmented tory processes (which also involves astrocytes and monocytes)

in the pathogenesis of AD have been intensively investigated berger et al., 2014). An obvious hypothesis is that TREM2 is trig-
[Heneka et al., 2015]). The role of microglia in AD is likely broader gering the phagocytosis of Ab plaques (Kleinberger et al., 2014).
than inflammation alone. These cells constitute a dynamic pop- Another microglia risk gene, CD33, apparently also inhibits up-
ulation that samples the local brain environment continuously take of Ab but via a gain-of-function mechanism, as it is upregu-
with highly ramified processes and interacts intimately with neu- lated in SAD (Griciuc et al., 2013).
rons (Li et al., 2012). Microglia phagocytose supernumerary To assess the effects of Trem2 in vivo, APP/PS1 transgenic
neurons during normal development and can directly induce mice were crossed with Trem2 / mice (Ulrich et al., 2014;
neuronal apoptosis (Salter and Beggs, 2014). More subtle ac- Wang et al., 2015). The results are divergent. In one study (Ulrich
tions, such as synaptic pruning, are also part of their repertoire. et al., 2014), no effects on Ab plaque load, cytokine levels, or mi-
Highly replicable data show complement activation, increases croglial activation were recorded. The only reported difference
in inflammatory cytokines in the CSF, increased levels of reactive was a reduction in the numbers of microglia that clustered
oxygen species, and so on, in AD. Epidemiological studies sug- around plaques. In the second study, which used a different
gest, in addition, a potential inverse correlation between the use model—i.e., the 5XFAD mice—Trem2 knockout resulted in a
of non-steroidal anti-inflammatory drugs and AD prevalence gene dosage increase in brain Ab plaques, accompanied by
(Heneka et al., 2015), although interventional studies with anti- significantly reduced expression of microglial markers of activa-
inflammatory drugs have all been negative to date. Recent tion and inflammatory cytokines. Thus, this work suggests that
genetic data unequivocally place inflammatory processes as Trem2 plays an important role in the microglial inflammatory
being important in AD. A series of GWASs of increasing size re- response: in the absence of Trem2, this response is diminished.
vealed 22 susceptibility loci (Lambert et al., 2013) of which a Apart from their phagocytic capacities, microglia express a
large group clearly could be ascribed to pathways within innate range of cell-surface receptors that activate innate immune re-
immune system regulation (Figure 2): complement receptor 1 sponses. This allows microglia to mount protective responses
(CR1), clusterin, CD33, the MS4A6-MS4A4 cluster, ABCA7, to damage-associated molecular patterns, such as Ab aggre-
CD2AP, EPHA1, HLA-DRB5–DRB1, INPP5D, and MEF2C gates. In particular, a number of studies have implicated the
(Karch and Goate, 2015). However, for most of these associa- CD36 scavenger receptor and the Toll-like Receptor 4 (TLR-4)
tions, it is not possible to ascribe changes in biological function in mediating inflammatory cytokine release in response to aggre-
to the identified gene or, indeed, whether the polymorphisms are gated Ab. Data on the role of microglia and inflammatory
in disequilibrium with other genes that harbor the relevant biolog- mediators on the deposition and clearance of Ab plaques in
ical function. APP transgenic mice are again mixed, with some researchers
A critically important finding from two groups revealed that finding amelioration and others exacerbation of plaque deposi-
rare mutations to the triggering receptor expressed on myeloid tion. Recently, two groups looked at the effects of Interleukin
cells 2 (TREM2) gene significantly increased the risk of late onset 10 (IL10) on amyloid pathology in APP transgenic mice, with
AD by about 3- to 5-fold (Guerreiro et al., 2013; Jonsson et al., very concordant results. Interleukin 10 is a cytokine that inhibits
2013)—an increase in risk on a par with that of APOE4. TREM2 inflammatory cytokine signaling. Increasing its expression in two
mutations can also increase the risk of other neurodegenerative APP transgenic mouse models using adeno-associated virus
diseases, suggesting some commonality in the underlying path- mediated brain expression (Chakrabarty et al., 2015) caused
ological processes. At the same time, a very large study of late an increase of deposited Ab42 and plaque burden. The authors
onset AD patients versus controls using whole-genome gene- postulate that IL10-mediated increases in astrocyte ApoE pro-
expression profiling (Zhang et al., 2013) demonstrated that a duction led, in turn, to an elevation of ApoE binding to Ab plaque
module grouping innate immunity/microglia related genes corre- that reduced microglial phagocytosis—although this appears to
lated best with clinical disease. TYROBP ranked highest as the be at variance with data showing binding of ApoE to Trem2 (Atagi
module’s potential regulator. TYROBP, otherwise known as et al., 2015). In the second study (Guillot-Sestier et al., 2015), the
DAP12, encodes the signaling partner for TREM2. Thus, using investigators crossed APP/PS1 transgenic mice with IL10 /
radically different approaches, microglial function and TREM2 knockout mice. In a similar fashion, but with an opposite polarity,
signaling are found to play a major role in SAD. the IL10 deficiency reduced insoluble, deposited Ab. Another
TREM2 is a type I transmembrane protein that is expressed on study investigated the effects of removing microglia from the
the surface of osteoclasts, immature dendritic cells, macro- brain altogether (Grathwohl et al., 2010). Using different gangci-
phages, and, in the CNS, on microglia. TREM2 ranks as one of clovir dosing regimens, it was possible to remove nearly 100% of
highest expressed receptors on microglia, where its expression the microglia, albeit for a relatively short duration of up to
is >300-fold that of astrocytes (Hickman and El Khoury, 2014). 4 weeks. In no case was the deposition or clearance of amyloid
Homozygous, loss-of-function mutations in TREM2 and its plaques altered.
adaptor molecule DAP12 (TYROPB) cause Nasu Hakola disease Collectively, these examples illustrate the conundrum of in-
that is characterized by dementia, white matter loss, and cystic flammatory processes: in some situations they can be beneficial,
bone lesions. TREM2 binds to a range of poorly defined ligands in others, damaging. A more integrated approach seems neces-
such as phospholipids, bacterial products, and cell debris, and sary to investigate the progressive inflammatory and other
receptor binding mediates microglial phagocytosis and pro- responses, and a more comprehensive analysis of the role of
motes an anti-inflammatory cytokine profile. The greatest risk the different cells (microglia, but also astroglia and infiltrating
for AD is associated with the R47H variant, which causes loss monocytes) over the different stages of the cellular phase of
of function by preventing normal folding of the protein (Klein- AD is needed. Indeed, what is beneficial in the beginning could

become destructive later. Finally, inflammation must be placed to phagocytose Ab while Ab triggers deficits in oligodendrocytes
into the wider context of aging. The AD brain carries the hall- in different mice models (Desai et al., 2010). Other molecular
marks of an inflammatory process, but it is also apparent that links between AD and oligodendrocytes include their central
age itself predisposes to inflammation—a concept that has role in lipid metabolism, discussed above, and the role of
been coined as ‘‘inflammaging.’’ In a fascinating study, an anal- BACE1 in neuregulin processing.
ysis of gene expression profiles of immune- and inflammation-
related genes was conducted across a range of ages of normal Systems Biology Approaches to Study the Cellular
and AD brains (Cribbs et al., 2012). This study demonstrated Responses in AD
that the increase in expression of immune/inflammation-related The complexity emerging from the current analysis of the litera-
genes with age far exceeded that seen in AD versus normal ture stresses the importance of generating a new theoretical
brains. The conclusion from this work is that, with age, the brain framework that encompasses the many parallel processes that
becomes predisposed to inflammatory processes, which might go astray in the brains of AD patients. Typically, systems biology
explain why age is the greatest risk factor for AD. addresses such questions, but the ‘‘big data’’ approaches,
excepting the genetic studies, have not yet yielded spectacular
Oligodendrocytes: A Silent Majority new insights into AD pathogenesis. Available studies charac-
The roles of oligodendrocytes in AD are even less studied than terize global alterations of mRNA (Bennett et al., 2014; Bossers
those of astrocytes, although evidence supports their role in et al., 2010; Matarin et al., 2015; Miller et al., 2010; Zhang
the cellular phase of the disease (Ettle et al., 2015). Oligodendro- et al., 2013) or miRNA expression (Lau et al., 2013), thus aggre-
cytes constitute 75% of the neuroglia cells in the neocortex gating all changes in neurons, interneurons, astroglia, oligoden-
and are therefore the largest group of non-neuronal cells in the drocytes, and microglia of the analyzed brain block with a
brain (Pelvig et al., 2008). They produce myelin and facilitate consequent loss in resolution. Typically, small expression
neuronal transmission via saltatory action potential conductance changes of thousands of genes are recorded. Systems biology
over the nodes of Ranvier. They also provide buffer capacity for uses sophisticated statistical methods to correct for the multiple
potassium, metabolic, trophic, and mechanic support to axons. testing of hypotheses and involves a series of assumptions to
They are considered highly vulnerable cells and show a dramatic increase the power of analysis—for instance, that complex dis-
( 27%) decrease in number in the aging brain (Pelvig et al., eases disturb complex biological pathways and that the genes
2008). This cellular loss is reflected in age-associated myelin in these pathways are co-regulated (Santiago and Potashkin,
breakdown as seen in MRI scans of patients from age 50 on 2014). Other assumptions are that key regulators of biological
(Bartzokis, 2011). Oligodendrocyte progenitor cells, character- networks can be discovered by identifying regulatory patterns
ized by the expression of the proteoglycan nerve-glial antigen of co-expressed genes or by finding gene expression correlation
NG2, provide up to 5% of the cells in the mature brain. They patterns between different samples (differential co-expression
are the main proliferating cells and can differentiate into oligo- network analysis or DCA) and that proteins in pathways interact
dendrocytes, astrocytes, and possibly neurons, although this directly in a protein complex or in a small protein interaction
pluripotency remains an issue of debate (Crawford et al., 2014). network (protein-protein interaction or PPI analysis). In AD, real
Braak suggested a link between vulnerability of neurons and experimental validation of the conclusions of such analyses is
their myelination stage because the spreading of neurofibrillary mostly lacking. A notable exception is the study on transcrip-
tangles recapitulates the pattern of myelination in reverse order tion/neuron-restrictive silencer factor (REST/NRF) (Lu et al.,
(Braak and Braak, 1996). Later differentiating oligodendrocytes 2014). Data mining of transcription profiles from prefrontal
en-sheath more axons with small axon diameters as seen in cortex of aging humans predicted that REST/NRF was activated
late myelinating frontal and temporal lobes (Bartzokis, 2011), in aging brains but less so in AD (Lu et al., 2014). An impressive
and it might be that these make these areas more vulnerable series of experiments in human brain and in mouse and
to all kinds of stresses. Bartzokis goes even as far as to propose C. elegans genetic models demonstrated that REST expression
that the recently evolved extensive myelination of the human is neuroprotective, represses genes that promote cell death, in-
brain explains increased vulnerability to neurodegenerative and creases oxidative stress resistance by de-repressing FOXO1a
neuropsychiatric disorders (Bartzokis, 2011). Age-related myelin and SOD1 expression, and maintains neuronal viability in the
breakdown as assessed by MRI might indeed result in a progres- aging cortex (Lu et al., 2014), providing crucial insight in protec-
sive disconnection of neural networks and presage AD (Bartzo- tive mechanisms that, when malfunctioning, contribute to the
kis, 2011). One of the more compelling arguments to increase risk for neurodegenerative disease.
focus on oligodendrocytes and possible demyelination is the As stated, other available systems biology studies in the field
strongly accentuated age-related breakdown of myelin in AD are more descriptive. Altered regulation of transcription and
and in APOE4 patients (Bartzokis, 2011; Scheltens et al., cell proliferation, changes in energy pathways or mitochondria,
1992). At the cellular level, Ab deposition is linked to focal demy- upregulation of complement activation, inflammation, lipid meta-
elination. In APP/PS1 mice, an early demyelination phenotype is bolism, or downregulation of synaptic transmission are expected
seen that is apparently ‘‘repaired’’ by proliferation of oligoden- (Matarin et al., 2015; Zhang et al., 2013). Further analyses try to
drocyte progenitor cells at 6–8 months (Behrendt et al., 2013). identify molecular players that act as drivers or are critical in the
Such proliferation is not observed in AD brain, suggesting that overall processes identified. One study found downregulation of
this putative ‘‘protective’’ reaction does not occur in patients tumor suppressors and focused also on changes in oligodendro-
(Behrendt et al., 2013). Oligodendrocyte progenitor cells appear cyte growth factors. Another study found significant changes in a

human-specific module (group of highly co-expressed genes) strated major changes between Braak stage II and III, which is
that closely correlated with AD progression. The authors specu- just before or at the onset of plaque and tangle pathology. Unex-
late that the hubs in this network are important parts of the AD pectedly, and in contrast with other genome-wide studies, in-
process (Miller et al., 2010). However, none of those hubs has creases in synaptic plasticity genes came up in this early phase,
been further validated, and one (FLJ12151) has even been with- while downregulation was only seen in the late stage of AD
drawn as a gene by NCBI. Overall, this study found major links (Bossers et al., 2010). The miRNA study demonstrates complex
with oligodendrocytes, astroglia, and microglia, re-enforcing patterns of change in 41 miRNA with a 2- to 3-fold decrease in
the major theme of the current review. As previously mentioned, miRNA132 expression standing out in this analysis (Lau et al.,
TYROBP, was identified as a regulator of an immune- and micro- 2013). Unfortunately, the dataset encompasses only 49 brains
glia-specific module in a study in 549 brains from late-onset pa- at the moment, and therefore, statistical power is limited. Ex-
tients and non-demented subjects (Zhang et al., 2013). Although panding this dataset and providing in addition cellular resolution
only upregulated by 18%, the authors boldly conclude that would be extremely helpful for the field.
TYROBP may be a therapeutic target in prevention of neuronal Real temporal resolution can only come from animal studies
damage in LOAD, supported by the genetic data found in other where brain samples can be taken from cohorts of mice at
studies (vide supra). different stages of the disease. One such study compared
There are many reasons why systems biology approaches are gene expression alterations over time in Ab- and Tau-models
currently not yielding novel insights in our field. Further massive of the disease (Matarin et al., 2015). Mainly immune-related
data gathering is of questionable use unless systematic, quality changes were recorded, with an immune module strongly asso-
controlled, in-depth phenotypic information is associated with ciated to amyloid pathology, and genes related to synaptic func-
the samples (Bennett et al., 2014). However, in addition, tion most altered in the Tau model. Cellular resolution, however,
massively collecting ‘‘omic’’ data from blocks of brain material was lacking.
probably lacks the granularity required to dissect key aspects
of the disease process. The brain is the most complex structure The Future: Single-Cell Biology and Cellular Network
of the human body with an incredible variety of cell types. The Analysis in the Study of AD
lack of cellular resolution makes it very likely that dramatic Breakthroughs in the study of the biochemical phase of AD are
changes in gene expression in particular cell types or groups still needed, and an excellent example was the recent elucidation
of cells will be diluted or lost by the absence of effects in other of the atomic structure of g-secretases, opening a complete new
cell groups. The interpretation of data from brain blocks might perspective on drug discovery for this important AD target (Bai
be very misleading with regard to the cell type affected in the et al., 2015; De Strooper, 2014). However, it is unlikely that mo-
disorder. lecular genetics or molecular cell biology will continue to propel
our understanding of the disorder in as productive a way as it has
Single-Cell Populations and Resolution over Different over the last 20 years. The major challenge for AD research is
Stages of Disease now to understand the complex cellular reaction underlying the
A recent systems biology study in AD mice strikingly illustrated long prodromal phase of AD. Cellular and temporal resolution
the above point by looking separately at astrocytes and micro- will be crucial to map systematically the progressive cellular
glia (Orre et al., 2014). The immune pathways were much more alterations and to generate a comprehensive cellular theory of
affected in astrocytes than in microglia, and interestingly, while the disorder. AD is indeed not a biochemical or molecular prob-
the recorded alterations in astrocytes were remarkably similar lem but a physiological one of disrupted cellular connectivity.
to the alterations recorded in human AD brain, the microglial The disease can therefore only be fully understood in the context
changes showed much less similarity (Orre et al., 2014). Individ- of the complex cellular interactions that maintain homeostasis in
ual changes in expression of genes were much more pro- the brain—i.e., at the level of the neurovascular and glioneuronal
nounced than in the ‘‘whole brain block’’ studies. Tyrobp expres- units discussed above.
sion increased almost 10-fold (compared to the overall 18% Spectacular progress in ‘‘single-cell biology’’ (Sandberg,
increase reported in human AD [Zhang et al., 2013]), and 2014) over the last 3 years makes it possible to map alterations
Trem2 increased more than 9-fold in the astrocytes while both in gene expression in many different cell types in parallel. The
were conspicuously absent in the list of microglia associated strong advantages of a single-cell approach are (1) agnosticism
gene changes (Orre et al., 2014). While Trem2 expression is with respect to cell type (allowing also the discovery of new cell
much lower in astrocytes than in microglia (Hickman and El types) and (2) that information on the variation in responses over
Khoury, 2014), these dramatic and consistent changes in gene a subpopulation of cells can be evaluated—e.g., do all astroglia,
expression provide high confidence that they are real and will interneurons and so on, respond in the same way and at the
translate to biologically significant effects. The study clearly illus- same time to Ab stress? Two recent publications provide proof
trates how important it is to provide cellular resolution when per- of concept that genomic expression data from hundreds of indi-
forming whole-genome analyses of the brain. vidual cells from mouse hippocampus (Zeisel et al., 2015) and
Temporal resolution is also important. In particular, studies on human cortex (Darmanis et al., 2015) can be resolved success-
human tissue tend to pool the data without taking into account fully. In addition, spatial transcriptomics is rapidly evolving, and
disease duration or evolution. A study following global gene spatial resolution of genome-wide gene expression in cells is
(Bossers et al., 2010) and miRNA (Lau et al., 2013) expression becoming possible. A limitation of current approaches is that
in the prefrontal cortex over different Braak stages of AD demon- changes are recorded only at the transcriptional level. However,

cells react by changes in gene expression, and pathway analysis Bossers, K., Wirz, K.T.S., Meerhoff, G.F., Essing, A.H.W., van Dongen, J.W.,
will help to indicate processes affected in the different cells. Houba, P., Kruse, C.G., Verhaagen, J., and Swaab, D.F. (2010). Concerted
changes in transcripts in the prefrontal cortex precede neuropathology in Alz-
A framework could be generated by measuring all changes in
heimer’s disease. Brain 133, 3699–3723.
the different cell types of the brain over the different Braak stages
Braak, H., and Braak, E. (1995). Staging of Alzheimer’s disease-related neuro-
in a few relevant brain areas, which would constitute a valuable
fibrillary changes. Neurobiol. Aging 16, 271–278, discussion 278–284.
resource for the field and propel AD research into the complex
Braak, H., and Braak, E. (1996). Development of Alzheimer-related neurofibril-
biology of the 21st century. Further experimentation and valida- lary changes in the neocortex inversely recapitulates cortical myelogenesis.
tion of hypotheses at the protein and in vivo level will still use Acta Neuropathol. 92, 197–201.
classical reductionist approaches and yield individual insights, Braak, H., and Del Tredici, K. (2011). The pathological process underlying Alz-
but being able to fit such pieces into the time, cell, and network heimer’s disease in individuals under thirty. Acta Neuropathol. 121, 171–181.
dependent atlas that describes the evolution of Alzheimer’s dis- Burda, J.E., and Sofroniew, M.V. (2014). Reactive gliosis and the multicellular
ease will be transformative for the field. Novel ideas will emanate response to CNS damage and disease. Neuron 81, 229–248.
from such an insight and an integrated conceptual framework Bushong, E.A., Martone, M.E., Jones, Y.Z., and Ellisman, M.H. (2002). Proto-
will provide a scientifically coherent basis for targeted therapeu- plasmic astrocytes in CA1 stratum radiatum occupy separate anatomical do-
tics that address different elements of the disease in a stage- mains. J. Neurosci. 22, 183–192.
dependent manner. Carrillo, M.C., Dean, R.A., Nicolas, F., Miller, D.S., Berman, R., Khachaturian,
Z., Bain, L.J., Schindler, R., and Knopman, D.; Alzheimer’s Association
Research Roundtable (2013). Revisiting the framework of the National Institute
ACKNOWLEDGMENTS
on Aging-Alzheimer’s Association diagnostic criteria. Alzheimers Dement. 9,
594–601.
B.D.S. is a consultant for Janssen Pharmaceutics and receives research fund-
ing from this company. E.K. is an employee of ABBVIE. Carter, S.F., Schöll, M., Almkvist, O., Wall, A., Engler, H., Långström, B., and
Nordberg, A. (2012). Evidence for astrocytosis in prodromal Alzheimer
disease provided by 11C-deuterium-L-deprenyl: a multitracer PET paradigm
REFERENCES combining 11C-Pittsburgh compound B and 18F-FDG. J. Nucl. Med. 53,
37–46.
Allen, N.J. (2014). Astrocyte regulation of synaptic behavior. Annu. Rev. Cell
Chakrabarty, P., Li, A., Ceballos-Diaz, C., Eddy, J.A., Funk, C.C., Moore, B.,
Dev. Biol. 30, 439–463.
DiNunno, N., Rosario, A.M., Cruz, P.E., Verbeeck, C., et al. (2015). IL-10 alters
Ando, K., Brion, J.-P., Stygelbout, V., Suain, V., Authelet, M., Dedecker, R., immunoproteostasis in APP mice, increasing plaque burden and worsening
Chanut, A., Lacor, P., Lavaur, J., Sazdovitch, V., et al. (2013). Clathrin adaptor cognitive behavior. Neuron 85, 519–533.
CALM/PICALM is associated with neurofibrillary tangles and is cleaved in Alz-
Chávez-Gutiérrez, L., Bammens, L., Benilova, I., Vandersteen, A., Benurwar, M.,
heimer’s brains. Acta Neuropathol. 125, 861–878.
Borgers, M., Lismont, S., Zhou, L., Van Cleynenbreugel, S., Esselmann, H., et al.
Atagi, Y., Liu, C.-C., Painter, M.M., Chen, X.-F., Verbeeck, C., Zheng, H., Li, X., (2012). The mechanism of g-Secretase dysfunction in familial Alzheimer disease.
Rademakers, R., Kang, S.S., Xu, H., et al. (2015). Apolipoprotein E Is a Ligand EMBO J. 31, 2261–2274.
for Triggering Receptor Expressed on Myeloid Cells 2 (TREM2). J. Biol. Chem.
Cheng, J., and Ji, D. (2013). Rigid firing sequences undermine spatial memory
290, 26043–26050.
codes in a neurodegenerative mouse model. eLife 2, e00647.
Attems, J., and Jellinger, K.A. (2014). The overlap between vascular disease
Chung, W.-S., Clarke, L.E., Wang, G.X., Stafford, B.K., Sher, A., Chakraborty,
and Alzheimer’s disease–lessons from pathology. BMC Med. 12, 206.
C., Joung, J., Foo, L.C., Thompson, A., Chen, C., et al. (2013). Astrocytes
Bai, X.C., Yan, C., Yang, G., Lu, P., Ma, D., Sun, L., Zhou, R., Scheres, S.H.W., mediate synapse elimination through MEGF10 and MERTK pathways. Nature
and Shi, Y. (2015). An atomic structure of human g-secretase. Nature 525, 504, 394–400.
212–217.
Crary, J.F., Trojanowski, J.Q., Schneider, J.A., Abisambra, J.F., Abner, E.L.,
Bakker, A., Krauss, G.L., Albert, M.S., Speck, C.L., Jones, L.R., Stark, C.E., Alafuzoff, I., Arnold, S.E., Attems, J., Beach, T.G., Bigio, E.H., et al. (2014). Pri-
Yassa, M.A., Bassett, S.S., Shelton, A.L., and Gallagher, M. (2012). Reduction mary age-related tauopathy (PART): a common pathology associated with hu-
of hippocampal hyperactivity improves cognition in amnestic mild cognitive man aging. Acta Neuropathol. 128, 755–766.
impairment. Neuron 74, 467–474.
Crawford, A.H., Stockley, J.H., Tripathi, R.B., Richardson, W.D., and Franklin,
Bartzokis, G. (2011). Alzheimer’s disease as homeostatic responses to age- R.J.M. (2014). Oligodendrocyte progenitors: adult stem cells of the central ner-
related myelin breakdown. Neurobiol. Aging 32, 1341–1371. vous system? Exp. Neurol. 260, 50–55.
Beauquis, J., Vinuesa, A., Pomilio, C., Pavı́a, P., Galván, V., and Saravia, F. Cribbs, D.H., Berchtold, N.C., Perreau, V., Coleman, P.D., Rogers, J., Tenner,
(2014). Neuronal and glial alterations, increased anxiety, and cognitive impair- A.J., and Cotman, C.W. (2012). Extensive innate immune gene activation ac-
ment before hippocampal amyloid deposition in PDAPP mice, model of Alz- companies brain aging, increasing vulnerability to cognitive decline and neuro-
heimer’s disease. Hippocampus 24, 257–269. degeneration: a microarray study. J. Neuroinflammation 9, 179.
Behrendt, G., Baer, K., Buffo, A., Curtis, M.A., Faull, R.L., Rees, M.I., Götz, M., Darmanis, S., Sloan, S.A., Zhang, Y., Enge, M., Caneda, C., Shuer, L.M., Hay-
and Dimou, L. (2013). Dynamic changes in myelin aberrations and oligoden- den Gephart, M.G., Barres, B.A., and Quake, S.R. (2015). A survey of human
drocyte generation in chronic amyloidosis in mice and men. Glia 61, 273–286. brain transcriptome diversity at the single cell level. Proc. Natl. Acad. Sci.
Bell, R.D., Winkler, E.A., Singh, I., Sagare, A.P., Deane, R., Wu, Z., Holtzman, USA 112, 7285–7290.
D.M., Betsholtz, C., Armulik, A., Sallstrom, J., et al. (2012). Apolipoprotein E De Strooper, B. (2010). Proteases and proteolysis in Alzheimer disease: a
controls cerebrovascular integrity via cyclophilin A. Nature 485, 512–516. multifactorial view on the disease process. Physiol. Rev. 90, 465–494.
Benilova, I., Karran, E., and De Strooper, B. (2012). The toxic Ab oligomer and De Strooper, B. (2014). Lessons from a failed g-secretase Alzheimer trial. Cell
Alzheimer’s disease: an emperor in need of clothes. Nat. Neurosci. 15, 159, 721–726.
349–357. Deane, R., Singh, I., Sagare, A.P., Bell, R.D., Ross, N.T., LaRue, B., Love, R.,
Bennett, D.A., Yu, L., and De Jager, P.L. (2014). Building a pipeline to discover Perry, S., Paquette, N., Deane, R.J., et al. (2012). A multimodal RAGE-specific
and validate novel therapeutic targets and lead compounds for Alzheimer’s inhibitor reduces amyloid b-mediated brain disorder in a mouse model of Alz-
disease. Biochem. Pharmacol. 88, 617–630. heimer disease. J. Clin. Invest. 122, 1377–1392.

Desai, M.K., Mastrangelo, M.A., Ryan, D.A., Sudol, K.L., Narrow, W.C., and Karran, E., Mercken, M., and De Strooper, B. (2011). The amyloid cascade hy-
Bowers, W.J. (2010). Early oligodendrocyte/myelin pathology in Alzheimer’s pothesis for Alzheimer’s disease: an appraisal for the development of thera-
disease mice constitutes a novel therapeutic target. Am. J. Pathol. 177, peutics. Nat. Rev. Drug Discov. 10, 698–712.
1422–1435. Khan, U.A., Liu, L., Provenzano, F.A., Berman, D.E., Profaci, C.P., Sloan, R.,
Dubois, B., Feldman, H.H., Jacova, C., Hampel, H., Molinuevo, J.L., Blennow, Mayeux, R., Duff, K.E., and Small, S.A. (2014). Molecular drivers and cortical
K., DeKosky, S.T., Gauthier, S., Selkoe, D., Bateman, R., et al. (2014). spread of lateral entorhinal cortex dysfunction in preclinical Alzheimer’s dis-
Advancing research diagnostic criteria for Alzheimer’s disease: the IWG-2 ease. Nat. Neurosci. 17, 304–311.
criteria. Lancet Neurol. 13, 614–629. Kleinberger, G., Yamanishi, Y., Suárez-Calvet, M., Czirr, E., Lohmann, E.,
Elrick, M.J., Yu, T., Chung, C., and Lieberman, A.P. (2012). Impaired proteol- Cuyvers, E., Struyfs, H., Pettkus, N., Wenninger-Weinzierl, A., Mazaheri, F.,
ysis underlies autophagic dysfunction in Niemann-Pick type C disease. et al. (2014). TREM2 mutations implicated in neurodegeneration impair cell
Hum. Mol. Genet. 21, 4876–4887. surface transport and phagocytosis. Sci. Transl. Med. 6, 243ra86.
Ettle, B., Schlachetzki, J.C.M., and Winkler, J. (2015). Oligodendroglia and Komatsu, M., Waguri, S., Chiba, T., Murata, S., Iwata, J., Tanida, I., Ueno, T.,
Myelin in Neurodegenerative Diseases: More Than Just Bystanders? Mol. Koike, M., Uchiyama, Y., Kominami, E., and Tanaka, K. (2006). Loss of auto-
Neurobiol. Published online May 13, 2015. http://dx.doi.org/10.1007/ phagy in the central nervous system causes neurodegeneration in mice.
s12035-015-9205-3. Nature 441, 880–884.
Gittis, A.H., and Brasier, D.J. (2015). NEUROSCIENCE. Astrocytes tell neurons Kraft, A.W., Hu, X., Yoon, H., Yan, P., Xiao, Q., Wang, Y., Gil, S.C., Brown, J.,
when to listen up. Science 349, 690–691. Wilhelmsson, U., Restivo, J.L., et al. (2013). Attenuating astrocyte activation
Grathwohl, S.A., Kälin, R.E., Bolmont, T., Prokop, S., Winkelmann, G., Kaeser, accelerates plaque pathogenesis in APP/PS1 mice. FASEB J. 27, 187–198.
S., Odenthal, J., Radde, R., Eldh, T., Gandy, S., et al. (2010). Formation and Kuchibhotla, K.V., Lattarulo, C.R., Hyman, B.T., and Bacskai, B.J. (2009). Syn-
maintenance of Alzheimer’s disease beta-amyloid plaques in the absence of chronous hyperactivity and intercellular calcium waves in astrocytes in Alz-
microglia. Nat. Neurosci. 12, 1361–1363. heimer mice. Science 323, 1211–1215.
Griciuc, A., Serrano-Pozo, A., Parrado, A.R., Lesinski, A.N., Asselin, C.N., Mul- Kulijewicz-Nawrot, M., Verkhratsky, A., Chvátal, A., Syková, E., and Rodrı́-
lin, K., Hooli, B., Choi, S.H., Hyman, B.T., and Tanzi, R.E. (2013). Alzheimer’s guez, J.J. (2012). Astrocytic cytoskeletal atrophy in the medial prefrontal cor-
disease risk gene CD33 inhibits microglial uptake of amyloid beta. Neuron tex of a triple transgenic mouse model of Alzheimer’s disease. J. Anat. 221,
78, 631–643. 252–262.
Guerreiro, R., Wojtas, A., Bras, J., Carrasquillo, M., Rogaeva, E., Majounie, E., Kuperstein, I., Broersen, K., Benilova, I., Rozenski, J., Jonckheere, W., Debul-
Cruchaga, C., Sassi, C., Kauwe, J.S.K., Younkin, S., et al.; Alzheimer Genetic paep, M., Vandersteen, A., Segers-Nolten, I., Van Der Werf, K., Subramaniam,
Analysis Group (2013). TREM2 variants in Alzheimer’s disease. N. Engl. J. V., et al. (2010). Neurotoxicity of Alzheimer’s disease Ab peptides is induced by
Med. 368, 117–127. small changes in the Ab42 to Ab40 ratio. EMBO J. 29, 3408–3420.
Guillot-Sestier, M.-V., Doty, K.R., Gate, D., Rodriguez, J., Jr., Leung, B.P., Re- Labbadia, J., and Morimoto, R.I. (2015). The biology of proteostasis in aging
zai-Zadeh, K., and Town, T. (2015). Il10 deficiency rebalances innate immunity and disease. Annu. Rev. Biochem. 84, 435–464.
to mitigate Alzheimer-like pathology. Neuron 85, 534–548.
Lambert, J.C., Ibrahim-Verbaas, C.A., Harold, D., Naj, A.C., Sims, R., Bellen-
Hardy, J., and Selkoe, D.J. (2002). The amyloid hypothesis of Alzheimer’s dis- guez, C., DeStafano, A.L., Bis, J.C., Beecham, G.W., Grenier-Boley, B.,
ease: progress and problems on the road to therapeutics. Science 297, et al.; European Alzheimer’s Disease Initiative (EADI); Genetic and Environ-
353–356. mental Risk in Alzheimer’s Disease; Alzheimer’s Disease Genetic Consortium;
Heneka, M.T., Sastre, M., Dumitrescu-Ozimek, L., Dewachter, I., Walter, J., Cohorts for Heart and Aging Research in Genomic Epidemiology (2013). Meta-
Klockgether, T., and Van Leuven, F. (2005). Focal glial activation coincides analysis of 74,046 individuals identifies 11 new susceptibility loci for Alz-
with increased BACE1 activation and precedes amyloid plaque deposition in heimer’s disease. Nat. Genet. 45, 1452–1458.
APP[V717I] transgenic mice. J. Neuroinflammation 2, 22. Lau, P., Bossers, K., Janky, R., Salta, E., Frigerio, C.S., Barbash, S., Rothman,
Heneka, M.T., Carson, M.J., El Khoury, J., Landreth, G.E., Brosseron, F., Fein- R., Sierksma, A.S.R., Thathiah, A., Greenberg, D., et al. (2013). Alteration of the
stein, D.L., Jacobs, A.H., Wyss-Coray, T., Vitorica, J., Ransohoff, R.M., et al. microRNA network during the progression of Alzheimer’s disease. EMBO Mol.
(2015). Neuroinflammation in Alzheimer’s disease. Lancet Neurol. 14, Med. 5, 1613–1634.
388–405. Lawson, L.J., Perry, V.H., Dri, P., and Gordon, S. (1990). Heterogeneity in the
Hickman, S.E., and El Khoury, J. (2014). TREM2 and the neuroimmunology of distribution and morphology of microglia in the normal adult mouse brain.
Alzheimer’s disease. Biochem. Pharmacol. 88, 495–498. Neuroscience 39, 151–170.
Iliff, J.J., Chen, M.J., Plog, B.A., Zeppenfeld, D.M., Soltero, M., Yang, L., Singh, Li, Y., Du, X.F., Liu, C.S., Wen, Z.L., and Du, J.L. (2012). Reciprocal regulation
I., Deane, R., and Nedergaard, M. (2014). Impairment of glymphatic pathway between resting microglial dynamics and neuronal activity in vivo. Dev. Cell 23,
function promotes tau pathology after traumatic brain injury. J. Neurosci. 34, 1189–1202.
16180–16193. Lian, H., Yang, L., Cole, A., Sun, L., Chiang, A.C., Fowler, S.W., Shim, D.J., Ro-
Jo, S., Yarishkin, O., Hwang, Y.J., Chun, Y.E., Park, M., Woo, D.H., Bae, J.Y., driguez-Rivera, J., Taglialatela, G., Jankowsky, J.L., et al. (2015). NFkB-acti-
Kim, T., Lee, J., Chun, H., et al. (2014). GABA from reactive astrocytes impairs vated astroglial release of complement C3 compromises neuronal morphology
memory in mouse models of Alzheimer’s disease. Nat. Med. 20, 886–896. and function associated with Alzheimer’s disease. Neuron 85, 101–115.
Jonsson, T., Stefansson, H., Steinberg, S., Jonsdottir, I., Jonsson, P.V., Snae- Lu, T., Aron, L., Zullo, J., Pan, Y., Kim, H., Chen, Y., Yang, T.-H., Kim, H.-M.,
dal, J., Bjornsson, S., Huttenlocher, J., Levey, A.I., Lah, J.J., et al. (2013). Drake, D., Liu, X.S., et al. (2014). REST and stress resistance in ageing and Alz-
Variant of TREM2 associated with the risk of Alzheimer’s disease. N. Engl. J. heimer’s disease. Nature 507, 448–454.
Med. 368, 107–116. Matarin, M., Salih, D.A., Yasvoina, M., Cummings, D.M., Guelfi, S., Liu, W., Na-
Kamphuis, W., Kooijman, L., Orre, M., Stassen, O., Pekny, M., and Hol, E.M. haboo Solim, M.A., Moens, T.G., Paublete, R.M., Ali, S.S., et al. (2015).
(2015). GFAP and vimentin deficiency alters gene expression in astrocytes A genome-wide gene-expression analysis and database in transgenic mice
and microglia in wild-type mice and changes the transcriptional response of during development of amyloid or tau pathology. Cell Rep. 10, 633–644.
reactive glia in mouse model for Alzheimer’s disease. Glia 63, 1036–1056. McGeer, P.L., Itagaki, S., Tago, H., and McGeer, E.G. (1988). Occurrence of
Karch, C.M., and Goate, A.M. (2015). Alzheimer’s disease risk genes and HLA-DR reactive microglia in Alzheimer’s disease. Ann. N Y Acad. Sci. 540,
mechanisms of disease pathogenesis. Biol. Psychiatry 77, 43–51. 319–323.

Menkes-Caspi, N., Yamin, H.G., Kellner, V., Spires-Jones, T.L., Cohen, D., and itative changes in human g-secretase underlie familial Alzheimer’s disease.
Stern, E.A. (2015). Pathological tau disrupts ongoing network activity. Neuron J. Exp. Med. 212, 2003–2013.
85, 959–966. Tarasoff-Conway, J.M., Carare, R.O., Osorio, R.S., Glodzik, L., Butler, T.,
Mesulam, M.-M. (1999). Neuroplasticity failure in Alzheimer’s disease: bridging Fieremans, E., Axel, L., Rusinek, H., Nicholson, C., Zlokovic, B.V., et al.
the gap between plaques and tangles. Neuron 24, 521–529. (2015). Clearance systems in the brain-implications for Alzheimer disease.
Miller, J.A., Horvath, S., and Geschwind, D.H. (2010). Divergence of human Nat. Rev. Neurol. 11, 457–470.
and mouse brain transcriptome highlights Alzheimer disease pathways. Thal, D.R., Rüb, U., Orantes, M., and Braak, H. (2002). Phases of A beta-depo-
Proc. Natl. Acad. Sci. USA 107, 12698–12703. sition in the human brain and its relevance for the development of AD.
Mufson, E.J., Mahady, L., Waters, D., Counts, S.E., Perez, S.E., DeKosky, S.T., Neurology 58, 1791–1800.
Ginsberg, S.D., Ikonomovic, M.D., Scheff, S.W., and Binder, L.I. (2015). Hippo-
Ulrich, J.D., Finn, M.B., Wang, Y., Shen, A., Mahan, T.E., Jiang, H., Stewart,
campal plasticity during the progression of Alzheimer’s disease. Neuroscience
F.R., Piccio, L., Colonna, M., and Holtzman, D.M. (2014). Altered microglial
309, 51–67.
response to Ab plaques in APPPS1-21 mice heterozygous for TREM2. Mol.
Nilsson, P., Saito, T., and Saido, T.C. (2014). New mouse model of Alzheimer’s. Neurodegener. 9, 20.
ACS Chem. Neurosci. 5, 499–502.
Verghese, P.B., Castellano, J.M., Garai, K., Wang, Y., Jiang, H., Shah, A., Bu,
Nixon, R.A. (2013). The role of autophagy in neurodegenerative disease. Nat. G., Frieden, C., and Holtzman, D.M. (2013). ApoE influences amyloid-b (Ab)
Med. 19, 983–997. clearance despite minimal apoE/Ab association in physiological conditions.
Olabarria, M., Noristani, H.N., Verkhratsky, A., and Rodrı́guez, J.J. (2010). Proc. Natl. Acad. Sci. USA 110, E1807–E1816.
Concomitant astroglial atrophy and astrogliosis in a triple transgenic animal
Verret, L., Mann, E.O., Hang, G.B., Barth, A.M.I., Cobos, I., Ho, K., Devidze, N.,
model of Alzheimer’s disease. Glia 58, 831–838.
Masliah, E., Kreitzer, A.C., Mody, I., et al. (2012). Inhibitory interneuron deficit
Orr, A.G., Hsiao, E.C., Wang, M.M., Ho, K., Kim, D.H., Wang, X., Guo, W., links altered network activity and cognitive dysfunction in Alzheimer model.
Kang, J., Yu, G.-Q., Adame, A., et al. (2015). Astrocytic adenosine receptor Cell 149, 708–721.
A2A and Gs-coupled signaling regulate memory. Nat. Neurosci. 18, 423–434.
Walker, L.C., and Jucker, M. (2015). Neurodegenerative diseases: expanding
Orre, M., Kamphuis, W., Osborn, L.M., Jansen, A.H.P., Kooijman, L., Bossers, the prion concept. Annu. Rev. Neurosci. 38, 87–103.
K., and Hol, E.M. (2014). Isolation of glia from Alzheimer’s mice reveals inflam-
mation and dysfunction. Neurobiol. Aging 35, 2746–2760. Wang, Y., Cella, M., Mallinson, K., Ulrich, J.D., Young, K.L., Robinette, M.L.,
Gilfillan, S., Krishnan, G.M., Sudhakar, S., Zinselmeyer, B.H., et al. (2015).
Palop, J.J., and Mucke, L. (2010). Amyloid-beta-induced neuronal dysfunction
TREM2 lipid sensing sustains the microglial response in an Alzheimer’s dis-
in Alzheimer’s disease: from synapses toward neural networks. Nat. Neurosci.
ease model. Cell 160, 1061–1071.
13, 812–818.
Winkler, E.A., Sagare, A.P., and Zlokovic, B.V. (2014). The pericyte: a forgotten
Pelvig, D.P., Pakkenberg, H., Regeur, L., Oster, S., and Pakkenberg, B. (2003).
cell type with important implications for Alzheimer’s disease? Brain Pathol. 24,
Neocortical glial cell numbers in Alzheimer’s disease. A stereological study.
371–386.
Dement. Geriatr. Cogn. Disord. 16, 212–219.
Pelvig, D.P., Pakkenberg, H., Stark, A.K., and Pakkenberg, B. (2008). Neocor- Winkler, E.A., Nishida, Y., Sagare, A.P., Rege, S.V., Bell, R.D., Perlmutter, D.,
tical glial cell numbers in human brains. Neurobiol. Aging 29, 1754–1762. Sengillo, J.D., Hillman, S., Kong, P., Nelson, A.R., et al. (2015). GLUT1 reduc-
tions exacerbate Alzheimer’s disease vasculo-neuronal dysfunction and
Protas, H.D., Chen, K., Langbaum, J.B., Fleisher, A.S., Alexander, G.E., Lee,
degeneration. Nat. Neurosci. 18, 521–530.
W., Bandy, D., de Leon, M.J., Mosconi, L., Buckley, S., et al. (2013). Posterior
cingulate glucose metabolism, hippocampal glucose metabolism, and hippo- Yamada, K., Holth, J.K., Liao, F., Stewart, F.R., Mahan, T.E., Jiang, H., Cirrito,
campal volume in cognitively normal, late-middle-aged persons at 3 levels of J.R., Patel, T.K., Hochgräfe, K., Mandelkow, E.-M., and Holtzman, D.M. (2014).
genetic risk for Alzheimer disease. JAMA Neurol. 70, 320–325. Neuronal activity regulates extracellular tau in vivo. J. Exp. Med. 211, 387–393.
Saito, T., Suemoto, T., Brouwers, N., Sleegers, K., Funamoto, S., Mihira, N., Zahs, K.R., and Ashe, K.H. (2010). ‘Too much good news’ - are Alzheimer
Matsuba, Y., Yamada, K., Nilsson, P., Takano, J., et al. (2011). Potent amyloi- mouse models trying to tell us how to prevent, not cure, Alzheimer’s disease?
dogenicity and pathogenicity of Ab43. Nat. Neurosci. 14, 1023–1032. Trends Neurosci. 33, 381–389.
Salter, M.W., and Beggs, S. (2014). Sublime microglia: expanding roles for the Zeisel, A., Muñoz-Manchado, A.B., Codeluppi, S., Lönnerberg, P., La Manno,
guardians of the CNS. Cell 158, 15–24. G., Juréus, A., Marques, S., Munguba, H., He, L., Betsholtz, C., et al. (2015).
Sandberg, R. (2014). Entering the era of single-cell transcriptomics in biology Brain structure. Cell types in the mouse cortex and hippocampus revealed
and medicine. Nat. Methods 11, 22–24. by single-cell RNA-seq. Science 347, 1138–1142.
Santiago, J.A., and Potashkin, J.A. (2014). A network approach to clinical inter- Zhang, B., Gaiteri, C., Bodea, L.-G., Wang, Z., McElwee, J., Podtelezhnikov,
vention in neurodegenerative diseases. Trends Mol. Med. 20, 694–703. A.A., Zhang, C., Xie, T., Tran, L., Dobrin, R., et al. (2013). Integrated systems
Scheltens, P., Barkhof, F., Valk, J., Algra, P.R., van der Hoop, R.G., Nauta, J., approach identifies genetic nodes and networks in late-onset Alzheimer’s dis-
and Wolters, E.C. (1992). White matter lesions on magnetic resonance imaging ease. Cell 153, 707–720.
in clinically diagnosed Alzheimer’s disease. Evidence for heterogeneity. Brain Zhao, Z., Sagare, A.P., Ma, Q., Halliday, M.R., Kong, P., Kisler, K., Winkler,
115, 735–748. E.A., Ramanathan, A., Kanekiyo, T., Bu, G., et al. (2015). Central role for
Small, S.A., and Duff, K. (2008). Linking Abeta and tau in late-onset Alzheimer’s PICALM in amyloid-b blood-brain barrier transcytosis and clearance. Nat.
disease: a dual pathway hypothesis. Neuron 60, 534–542. Neurosci. 18, 978–987.
Szaruga, M., Veugelen, S., Benurwar, M., Lismont, S., Sepulveda-Falla, D., Zlokovic, B.V. (2011). Neurovascular pathways to neurodegeneration in Alz-
Lleo, A., Ryan, N.S., Lashley, T., Fox, N.C., Murayama, S., et al. (2015). Qual- heimer’s disease and other disorders. Nat. Rev. Neurosci. 12, 723–738.

OUR
NETWORK
IS YOUR
NETWORK
With Cell Press Webinars, our network is
your network!
Cell Press Webinars give you access to
Need-to-know topics, editorially curated
hot topics in emerging research and the
application of new technology.
World-class presenters, experts in their field
Watch essential, need-to-know webcasts
via live streaming or on demand from the
Moderated by Cell Press editors
comfort and convenience of your office,
lab, or home.
Powered by
people in the know.
Like you.
Tap into our network today!

Visit www.cell.com/webinars Webinars
Article
Mitochondrial Dynamics Controls

T Cell Fate through Metabolic Programming
Michael D. Buck,1,2 David O’Sullivan,1 Ramon I. Klein Geltink,1 Jonathan D. Curtis,1 Chih-Hao Chang,2 David E. Sanin,1
Jing Qiu,1,2 Oliver Kretz,3,4 Daniel Braas,5 Gerritje J.W. van der Windt,6 Qiongyu Chen,2 Stanley Ching-Cheng Huang,2
Christina M. O’Neill,2 Brian T. Edelson,2 Edward J. Pearce,1,7 Hiromi Sesaki,8 Tobias B. Huber,3,9
Angelika S. Rambold,10,11 and Erika L. Pearce1,*
1Department of Immunometabolism, Max Planck Institute of Immunobiology and Epigenetics, 79108 Freiburg, Germany
2Department of Pathology and Immunology, Washington University School of Medicine, St. Louis, MO 63110, USA
3Renal Division, University Medical Center Freiburg, 79106 Freiburg, Germany
4Department of Neuroanatomy, University of Freiburg, 79104 Freiburg, Germany
5University of California Los Angeles Metabolomics Center, Los Angeles, CA 90095, USA
6Academic Medical Center, 1105 AZ Amsterdam, the Netherlands
7Faculty of Biology, University of Freiburg, 79104 Freiburg, Germany
8Department of Cell Biology, Johns Hopkins University School of Medicine, Baltimore, MD 21205, USA
9BIOSS Centre for Biological Signaling Studies, 79104 Freiburg, Germany
10Center for Chronic Immunodeficiency, University Medical Center Freiburg and University of Freiburg, 79106 Freiburg, Germany
11Department of Developmental Immunology, Max Planck Institute of Immunobiology and Epigenetics, 79108 Freiburg, Germany
*Correspondence: pearce@ie-freiburg.mpg.de
SUMMARY elevated OXPHOS (Chang et al., 2013; Sena et al., 2013), the
former of which is characteristic of the Warburg effect shared
Activated effector T (TE) cells augment anabolic path- by tumor cells and unicellular organisms (Vander Heiden et al.,
ways of metabolism, such as aerobic glycolysis, 2009). Once antigen is cleared, most TE cells die, but a subset
while memory T (TM) cells engage catabolic path- of long-lived memory T (TM) cells persists with enhanced
ways, like fatty acid oxidation (FAO). However, sig- mitochondrial capacity marked by a reliance on FAO to fuel
nals that drive these differences remain unclear. OXPHOS, which equips them to rapidly respond should infection
or cancer recur (Pearce et al., 2013). These extensive changes in
Mitochondria are metabolic organelles that actively
phenotype and function of T cells go along with a dynamic meta-
transform their ultrastructure. Therefore, we ques-
bolic range (MacIver et al., 2013; Buck et al., 2015). Failure to
tioned whether mitochondrial dynamics controls engage specific metabolic programs impairs the function and
T cell metabolism. We show that TE cells have punc- differentiation of T cells. As such, T cells represent an amenable
tate mitochondria, while TM cells maintain fused net- system to study changes in cell metabolism that occur as part
works. The fusion protein Opa1 is required for TM, but of normal development and not as a result of transformation.
not TE cells after infection, and enforcing fusion in TE Establishing the precise reasons why and how these and other
cells imposes TM cell characteristics and enhances cells emphasize one particular metabolic pathway over another
antitumor function. Our data suggest that, by altering remains a challenge.
cristae morphology, fusion in TM cells configures Mitochondria are essential hubs of metabolic activity, antiviral
electron transport chain (ETC) complex associations responses, and cell death that constantly remodel their structure
via nuclear-encoded GTPases (Nunnari and Suomalainen,
favoring oxidative phosphorylation (OXPHOS) and
2012). Mitochondrial fission generates discrete and fragmented
FAO, while fission in TE cells leads to cristae expan-
mitochondria that can increase reactive oxygen species (ROS)
sion, reducing ETC efficiency and promoting aerobic production (Yu et al., 2006), facilitate mitophagy (Frank et al.,
glycolysis. Thus, mitochondrial remodeling is a 2012; Toyama et al., 2016), accelerate cell proliferation (Taguchi
signaling mechanism that instructs T cell metabolic et al., 2007), and mediate apoptosis (Youle and Karbowski,
programming. 2005). Dynamin-related protein 1 (Drp1) is a cytosolic protein
that translocates to the outer mitochondrial membrane (OMM)
INTRODUCTION upon phosphorylation to scission mitochondria. Fusion of mito-
chondria into linear or tubular networks limits deleterious muta-
T cells mediate protective immunity against pathogens and can- tions in mtDNA (Santel et al., 2003), induces supercomplexes
cer and possess the unique ability to proliferate at an unparal- of the ETC maximizing OXPHOS activity (Cogliati et al., 2013;
leled rate in an adult organism. In this regard, one naive T (TN) Mishra et al., 2014), and enhances ER interactions important
cell can clonally expand into millions of ‘‘armed’’ effector T (TE) for Ca2+ flux (de Brito and Scorrano, 2008). In addition, mito-
cells in just a few days (Williams and Bevan, 2007). Concomitant chondria elongate as a survival mechanism in response to
with T cell activation is the engagement of aerobic glycolysis and nutrient starvation and stress, linking fusion to cell longevity
Cell 166, 63–76, June 30, 2016 ª 2016 Elsevier Inc. 63

(legend on next page)
64 Cell 166, 63–76, June 30, 2016

(Gomes et al., 2011; Rambold et al., 2011; Friedman and microscopy, we observed that while a day after activation the
Nunnari, 2014). OMM fusion is mediated by mitofusin 1 and 2 mitochondria appeared fused, from days 2–6 the IL-2 TE cells
(Mfn1, Mfn2), while inner membrane fusion is controlled by optic exhibited predominantly punctate mitochondria (Figure 1C). In
atrophy 1 (Opa1). Total deletion in any of these proteins is embry- contrast, once cells were exposed to IL-15, a cytokine that sup-
onically lethal and mutations in the genes that encode them ports TM cell formation (Schluns et al., 2002), the mitochondria
underlie the cause of several human diseases (Chan, 2012; formed elongated tubules. Magnified images from these experi-
Archer, 2014). ments emphasized the marked differences in mitochondrial
Mitochondrial membrane remodeling is acutely responsive to morphology between IL-2 TE and IL-15 TM cells (Figure 1D).
changes in cell metabolism (Mishra and Chan, 2016; Wai and We also examined the protein expression of several critical
Langer, 2016), but whether it instructs metabolic pathway utiliza- regulators of mitochondrial dynamics. We found that by day 6,
tion has been inferred but not extensively studied. In general, fusion mediators Mfn2 and Opa1 were lower in TE cells
deletion of any of the dynamics machinery perturbs OXPHOS compared to TM cells, while fission factor Drp1 was phosphory-
and glycolytic rates at baseline (Liesa and Shirihai, 2013). Tis- lated at its activating site Ser616 in TE cells (Figure 1E) (Mars-
sue-specific deletion of Mfn2 in muscles of mice disrupts boom et al., 2012). Together, these data suggest that mitochon-
glucose homeostasis (Sebastián et al., 2012), and Drp1 ablation dria in TE cells actively undergo fission, while in TM cells, these
in the liver results in reduced adiposity and elevated whole-body organelles exist in a fused state.
energy expenditure, protecting mice from diet-induced obesity
(Wang et al., 2015). A recent study has also suggested a link be- Mitochondrial Inner Membrane Fusion Protein Opa1 Is
tween Drp1-mediated fission and its effect on glycolysis during Necessary for TM Cell Generation
cell transformation (Serasinghe et al., 2015). The central question To test whether mitochondrial fusion was important for TM cell
of whether fission/fusion and associated changes in cristae development, we crossed Mfn1, Mfn2, and Opa1 floxed mice
morphology actively control the adoption of distinct metabolic to OT-I CD4 Cre transgenic mice to conditionally delete these
programs and therefore regulates T cell responses, however, re- proteins in T cells. Peripheral T cell frequencies in these mice
mains unanswered. were grossly normal (data not shown). We differentially cultured
Mfn1 / , Mfn2 / , and Opa1 / OT-I T cells in IL-2 and IL-15
RESULTS and found that only Opa1 / T cells had a survival defect when
cultured in IL-15 (Figure 2A), but not in IL-2. We measured the ef-
Unlike TE Cells, TM Cells Maintain a Fused Mitochondrial ficiency of gene deletion by mRNA and/or protein analyses (Fig-
Network ures S2A–S2C). While Mfn1 and 2 were efficiently deleted, we
We reported that TM cells have more mitochondrial mass than TE found residual expression of Opa1 particularly in IL-15 TM cells,
or TN cells and suggested that mitochondria in these T cell sub- suggesting that cells that retained some expression of Opa1 in
sets are morphologically distinct (van der Windt et al., 2012, IL-15 cultures had a survival advantage (Figure 2A). Assessment
2013). These observations prompted us to assess mitochondrial of mitochondrial ultrastructure revealed that the cristae were
structure in T cells. We infected mice with Listeria monocyto- altered and disorganized in the absence of Opa1 in agreement
genes expressing ovalbumin (OVA) (LmOVA) and isolated TE with published results for other cells (Figure 2B) (Zhang et al.,
and TM cells for ultrastructure analysis by electron microscopy 2011; Cogliati et al., 2013). Consistent with their survival defect,
(EM). We found that TE cells had small, distinct mitochondria Opa1 / IL-15 TM cells had decreased OXPHOS activity,
dispersed in the cytoplasm, while TM cells had densely packed, as measured by O2 consumption rate (OCR, an indicator of
somewhat tubular, mitochondria (Figure 1A). To investigate OXPHOS) to extracellular acidification rate (ECAR, an indicator
these morphological differences, we differentially cultured acti- of aerobic glycolysis) ratio, and spare respiratory capacity
vated OVA-specific T cell receptor (TCR) transgenic OT-I cells (SRC), compared to normal cells (Figure 2C). SRC is the extra
in interleukin-2 (IL-2) and IL-15 to generate IL-2 TE and IL-15 mitochondrial capacity available in a cell to produce energy under
TM cells (Figure S1) (Carrio et al., 2004). These culture conditions conditions of increased work or stress and is thought to be impor-
approximate T cell responses in vivo and allow us to generate tant for long-term cell survival and function (measured as
large numbers of cells amenable to further experimentation OXPHOS activity above basal after uncoupling with FCCP) (Nich-
in vitro (O’Sullivan et al., 2014). We found that IL-2 TE and IL- olls, 2009; van der Windt et al., 2012). To determine whether Opa1
15 TM cells possessed similar mitochondrial ultrastructure as is required for TM cell development in vivo, we adoptively trans-
their ex vivo isolated counterparts (Figure 1B). Using confocal ferred Opa1 / OT-I TN cells into congenic recipients, infected
Figure 1. Effector and Memory T Cells Possess Distinct Mitochondrial Morphologies

(A and B) Effector (TE, CD44hi CD62Llo, 7 days post infection) and memory T (TM, CD44hi CD62Lhi, 21 days post infection) cells (A) sorted from C57BL/6 mice
infected i.p. with 107 CFU LmOVA and IL-2 TE and IL-15 TM cells (B) generated from differential culture of OT-I cells activated with OVA peptide and IL-2 using IL-2
or IL-15 analyzed by EM. Scale bar, 0.5 mm. Representative of two experiments.
(C and D) Mitochondrial morphology in live OT-I PhAM cells before and after aCD3/CD28 activation and differential cytokine culture by spinning disk confocal
microscopy. Mitochondria are green (GFP) and nuclei are blue (Hoechst). Scale bar, 5 mm (C); 1 mm (D). Representative of two experiments.
(E) Immunoblot analysis of cell protein extracts from (C), probed for Mfn2, Opa1, Drp1, phosphorylated Drp1 at Ser616 (Drp1pS616), and b-actin. Representative of
two experiments.
See also Figure S1.
Cell 166, 63–76, June 30, 2016 65

Figure 2. Memory T Cell Development and Survival, Unlike Effectors, Requires Mitochondrial Fusion
(A) Relative in vitro survival ratios of Mfn1, Mfn2, or Opa1-deficient (CD4 Cre+, Cre / ) to wild-type (CD4 Cre , Cre+/+) OT-I IL-2 TE and IL-15 TM cells (*p = 0.0465).
Data normalized from two to three independent experiments shown as mean ± SEM.
(legend continued on next page)
66 Cell 166, 63–76, June 30, 2016

these mice with LmOVA, and assessed TM cell formation in the affected, although a small decrease in KLRG1 and increase in
weeks after infection. Control and Opa1 / OT-I T cells mounted CD25 was measured (Figure S3E). Additionally, we performed
normal TE cell responses (day 7) to infection, while Opa1 / OT-I a genetic gain of function experiment and transduced activated
TM cell formation (days 14–21) was drastically impaired (Fig- IL-2 TE cells with retrovirus expressing Mfn1, Mfn2, or Opa1.
ure 2D). Consistent with diminished TM cell development, a signif- Similar to enforcement of fusion pharmacologically, we found
icantly higher proportion of short-lived effector to memory that cells transduced with Opa1 had more mitochondria (Fig-
precursor effector cells were present within the Opa1 / OT-I ure 3G) and OXPHOS (Figure 3H), than empty vector control or
donor cell population 7 days after infection (Figure S2D) (Kaech Mfn-transduced T cells, as well as increased overall metabolic
et al., 2003). In addition, at day 10 post infection, a time point activity, with a predominant increase in OXPHOS over glycolysis
when TE cells contract while TM cells emerge, Opa1 / T cells iso- (Figure S3F). TM cell-associated markers such as CCR7 and
lated ex vivo had decreased SRC compared to control cells (Fig- CD127 were increased on transduced cells, as well as TE cell
ure S2E), correlating with their decreased survival. To assess proteins, such as PD-1 (Figure S3G). Overexpression of each
whether Opa1 / TM cells existed in too low an abundance to target gene over the control was confirmed by mRNA expression
be discerned by flow cytometry, we challenged these mice with (Figure S3H). Together, our results show that mitochondrial
a second infection. We observed no recall response (day 3 and fusion confers TM cell phenotypes on activated TE cells even in
6 p.c.) from Opa1 / T cells when assessing frequency (Figure 2E) culture conditions that program TE cell differentiation.
or absolute numbers (Figure 2F), while there was considerable
expansion of control donor cells. These data illustrate that T Cell Mitochondrial Fusion Improves Adoptive Cellular
Opa1 is required for TM, but not TE cell generation. Immunotherapy against Tumors
A consideration when designing adoptive cellular immuno-
Mitochondrial Fusion Imposes a TM Cell Phenotype, therapy (ACI) is to improve T cell fitness during ex vivo culture
Even in the Presence of Activating Signals so that when T cells are re-introduced into a patient they are
Genetic loss of function of Opa1 revealed that this protein is crit- able to function efficiently and persist over time (Restifo et al.,
ical for TM cell formation. Given the fused phenotype of mito- 2012; Maus et al., 2014; O’Sullivan and Pearce, 2015). Our
chondria in TM cells, we hypothesized that Opa1-mediated mito- data showed that fusion-promoting drugs created metabolically
chondrial fusion supports the metabolism needed for TM cell fit T cells. We predicted that enforcing fusion would also enhance
development. We used a gain of function approach to enhance the longevity of IL-2 TE cells in vivo. To test this, we adoptively
mitochondrial fusion. Culturing T cells with the ‘‘fusion promoter’’ transferred control and M1 + Mdivi-1-treated OT-I T cells into
M1 (Wang et al., 2012) and the ‘‘fission inhibitor’’ Mdivi-1 (Cas- congenic mice and tracked donor cell survival. We found signif-
sidy-Stone et al., 2008) (Figure 3A) induced mitochondrial fusion icantly more drug-treated T cells in the spleen (Figure 3I) and
in IL-2 TE cells, rendering them morphologically similar to IL-15 lymph nodes (Figure 3J) 2 days after transfer. To determine if
TM cells (Figure 3B). Treatment with these drugs enhanced other the persistence of these cells would be maintained better long
TM cell properties in activated IL-2 TE cells, including increased term than controls, we infected mice with LmOVA >3 weeks later
mitochondrial mass (Figure 3C), OXPHOS and SRC (Figure 3D), and measured T cell responses against the bacteria. We found
CD62L expression (Figure 3E), and robust metabolic activity, as that drug-treated cells selectively expanded in response to
indicated by bioenergetic profiling in response to secondary infection (Figure 3K) and could be recovered in significantly
stimulation with PMA + ionomycin, followed by addition of oligo- greater numbers in the spleen 6 days post challenge (Figure 3L).
mycin (ATP synthase inhibitor), FCCP, and rotenone with antimy- Next, we assessed whether these drugs could be used to pro-
cin A (ETC complex I and III inhibitors), all drugs that stress the mote T cell function in an ACI model. We injected EL4-OVA tu-
mitochondria (Figures 3F and S3A). However, we did not observe mor cells into mice. Then either 5 or 12 days later we adoptively
increased mtDNA in these cells (Figure S3B). We found that transferred IL-2 TE cells that had been treated with DMSO or
ECAR and the OCR/ECAR ratio increased after drug treatment M1 + Mdivi-1. In both settings, mice that received ‘‘fusion-pro-
(Figure S3C), indicating elevated metabolic activity overall, with moted’’ T cells were able to control tumor growth significantly
a predominant increase in OXPHOS over glycolysis. While we better than mice that received control-treated cells (Figures 4A
observed these changes in mitochondrial activity, we did not and 4B). The cytolytic ability (Figure S4A) and proliferation
measure any significant differences in mitochondrial membrane (Figure S4B) of the fusion enforced IL-2 TE cells were similar to
potential or ROS after drug treatment (Figure S3D). The expres- controls, however, they expressed significantly higher levels of
sion of other activation markers were also not substantially IFN-g and TNF-a when restimulated with PMA + ionomycin
(B and C) Mitochondrial morphology of OT-I Opa1 wild-type and Opa1 / IL-2 TE and IL-15 TM cells analyzed by (B) EM (scale bar, 0.5 mm, represents one
experiment) and (C) Seahorse EFA. Bar graphs represent (left) ratios of O2 consumption rates (OCR, indicator of OXPHOS) to extracellular acidification rates
(ECAR, indicator of aerobic glycolysis) at baseline and (right) spare respiratory capacity (SRC) (% max OCR after FCCP injection of baseline OCR) of indicated
cells (*p < 0.03, **p = 0.0079). Data from three experiments are shown as mean ± SEM.
(D–F) 104 OT-I Opa1+/+ or Opa1 / T cells were transferred i.v. into C57BL/6 CD90.1 mice infected i.v. with 107 CFU LmOVA. Blood analyzed by flow cytometry at
indicated times post infection. After 21 days, mice were challenged i.v. with 5 3 107 CFU LmOVA and blood analyzed post challenge (p.c.). (D) % Donor Kb/OVA+
CD90.2+ cells shown in representative flow plots and (E) line graph with mean ± SEM (*p = 0.0238, **p < 0.005). (F) Number of donor Kb/OVA+ cells from spleens of
infected mice shown with mean ± SEM (*p = 0.0126). Representative of two experiments (n = 9–11/group).
See also Figure S2.
Cell 166, 63–76, June 30, 2016 67

Figure 3. Enhancing Mitochondrial Fusion Promotes the Generation of Memory-like T Cells
(A–F and I–L) OVA peptide and IL-2-activated OT-I cells differentiated into IL-2 TE or IL-15 TM cells for 3 days in the presence of DMSO or 20 mM fusion promoter
M1 and 10 mM fission inhibitor Mdivi-1 (M1 + Mdivi-1) as shown (A) pictorially. (B) Representative spinning disk confocal images from three experiments of live
cells from OT-I PhAM mice. Mitochondria are green (GFP) and nuclei are blue (Hoechst). Scale bar, 5 mm. (C) Cells stained with MitoTracker Green and analyzed
by flow cytometry. Relative MFI (left) from 6 experiments (*p = 0.0394, **p = 0.0019) with representative histograms (right). (D) Baseline OCR and SRC from three to
four experiments (*p = 0.0485, ***p < 0.0001), and (E) CD62L expression analyzed by flow cytometry of indicated cells. Relative MFI (left) from seven experiments
(*p = 0.0325, **p = 0.0019, ***p < 0.0001) with representative histograms (right). (F) OCR of indicated cells at baseline and in response to PMA and ionomycin
stimulation (PMA + iono), oligomycin (Oligo), FCCP, and rotenone plus antimycin A (R + A). Represents two experiments. (C–F) Shown as mean ± SEM. (I–L) A total
68 Cell 166, 63–76, June 30, 2016

in vitro (Figure S4C). We also exposed activated human T cells to ure 6A). Furthermore, unlike controls, we observed no additional
M1 + Mdivi-1 treatment in vitro and found that they had visibly drop of OCR in Opa1 / IL-2 TE cells after the addition of oligo-
more fused mitochondria (Figure 4C) and exhibited the bioener- mycin (Figure 6B), suggesting that in the absence of Opa1, only
getic profile (Figure 4D) and surface marker expression (Fig- FAO supports OXPHOS, and that substrates, such as glucose-
ure 4E) characteristic of TM cells, compared to control-treated derived pyruvate, are not used for mitochondrial ATP production
cells. Parameters such as mitochondrial mass (Figure 4E) and in this setting. We cultured Opa1+/+ and Opa1 / IL-2 TE cells
other surface markers (Figure S4D) were not significantly altered. with 13C-labeled glucose and traced 13C into TCA cycle metab-
These data suggest that promoting fusion in T cells may be a olites. While the percent of 13C-labeled pyruvate was higher in
translatable treatment for enhancing human therapy. Opa1 / T cells, the frequency of 13C-labeled TCA cycle inter-
mediates was significantly reduced in Opa1 / T cells compared
Mitochondrial Fusion Promotes TM Cell Metabolism, but to controls (Figures 6C and S6A), a result that is supported by
Opa1 Is Not Required for FAO their higher ECAR (Figure 6A). These data suggested that without
Our data showed that Opa1 was necessary for TM cell formation, mitochondrial fusion, pyruvate is preferentially secreted as
but the question of how Opa1 acted to support TM cells re- lactate, rather than oxidized in the mitochondria. Therefore, we
mained. We hypothesized that mitochondrial fusion, via Opa1 questioned whether FAO is a ‘‘default’’ pathway for mitochondria
function, was needed for FAO, as the engagement of this in a resting or fused state (i.e., Opa1 sufficiency), and the induc-
pathway is required for TM cell development and survival (Pearce tion of aerobic glycolysis is a major downstream effect of fission
et al., 2009; van der Windt et al., 2012, 2013). This hypothesis (i.e., Opa1 deficiency). If this were the case, then a balance
was based on our observations that these two processes between fission and fusion, modulated by proteins such as
seemed to be linked in TM cells and also on a recent report Opa1, could act as a primary signal to dictate the metabolic
that mitochondrial fusion is important for efficient FAO via lipid phenotype of T cells. In support of this idea, T cells from poly-
droplet trafficking under starvation conditions (Rambold et al., clonal T cell-conditional deleted Opa1 mice had higher ECAR
2015). We treated IL-2 TE and IL-15 TM cells with M1 + Mdivi-1 and an increased proportion of CD8 T cells with an activated
or vehicle and then measured OCR in response to etomoxir, a effector phenotype in the basal state based on surface marker
specific inhibitor of mitochondrial long chain FAO (Deberardinis expression (Figure S6B).
et al., 2006), and mitochondrial inhibitors. We found that the Opa1 is critical for inner mitochondrial membrane fusion, but
increased OCR and SRC evident in these cells after M1 + also for other processes like cristae remodeling (Frezza et al.,
Mdivi-1 treatment was due to enhanced FAO (Figures 5A and 2006; Cogliati et al., 2013). We observed major changes in
S5A). IL-2 TE cells transduced with Opa1 also exhibited elevated cristae morphology in Opa1 / T cells (Figure 2B). Given the
OCR that decreased in the presence of etomoxir compared importance of Opa1 in TM cell development (Figure 2), we further
to controls (Figure 5B). Bone marrow-derived macrophages assessed cristae morphology in TE and TM cells isolated ex vivo
(BM-Macs) cultured with M1 + Mdivi-1 increased OCR and after LmOVA infection (Figure 6D), as well as IL-2 TE and IL-15 TM
SRC to levels similar to M2 polarized macrophages, which cells (Figure 6E), and found that TE cells had many cristae with
engage FAO much like TM cells do (Figure S5B) (Huang et al., slightly wider and loosely organized intermembrane space than
2014). Importantly, M1 + Mdivi-1 treatment did not increase TM cells. Opa1 overexpression induces cristae tightening and
OCR (Figure 5C) or affect ECAR (Figure S5C) in Opa1 / IL-2 close association of ETC complexes in the inner mitochondrial
TE cells compared to controls, suggesting a requirement for membrane (Cogliati et al., 2013; Civiletto et al., 2015). Therefore,
Opa1 in augmenting OCR and FAO. However, in contrast we surmised that in the absence of Opa1, cristae disorganization
to what we expected, when we assessed bioenergetics of leads to dissociation of ETC complexes and subsequently less
Opa1+/+ and Opa1 / IL-2 TE (Figure 5D) and ex vivo isolated efficient ETC activity in T cells (Figure 2C). We assessed OCR af-
TE cells (Figure 5E), we found that both are equally responsive ter oligomycin in relation to OCR after rotenone and antimycin A
to etomoxir. Our results show that while Opa1 can promote treatment (i.e., proton leak), which indicates the coupling effi-
FAO in T cells, it is not compulsory for engagement of this meta- ciency of OXPHOS to mitochondrial ATP production. Consistent
bolic pathway. with decreased OXPHOS efficiency, we observed elevated pro-
ton leak in Opa1 / T cells compared to controls (Figure 6F). This
Mitochondrial Cristae Remodeling Signals Metabolic was also true for ex vivo isolated TE cells compared to TM cells
Adaptations in TM and TE Cells (Figure 6G), as well as IL-2 TE and IL-15 TM cells (Figure 6H).
Although Opa1+/+ and Opa1 / IL-2 TE cells could equally Together, these data suggest that there are cristae differences
engage FAO (Figure 5D), we observed that ECAR was signifi- between TE and TM cells, which may contribute to their distinct
cantly increased in Opa1 / cells both in vitro and ex vivo (Fig- metabolic phenotypes.
of 1–2 3 106 IL-2 TE cells cultured with DMSO (gray diamonds) or M1 + Mdivi-1 (blue squares) were transferred into congenic C57BL/6 recipient mice. Cell counts
of donor cells recovered 2 days later from the (I) spleen (***p = 0.005) and (J) peripheral lymph nodes (pLNs, ***p = 0.0006). Dots are individual mice. (K) Blood from
recipient mice analyzed for % donor Kb/OVA+ cells post transfer and challenged with 107 CFU LmOVA by flow cytometry (*p = 0.0150, n = 5/group). (L) Donor Kb/
OVA+ cells recovered from recipient spleens 6 days p.c. (*p = 0.0383). Dots are individual mice. (I–L) Represents two experiments shown with mean ± SEM.
(G and H) OT-I cells were transduced with either empty (Control), Mfn1, Mfn2, or Opa1 expression vectors, sorted, and cultured to generate IL-2 TE cells. (G)
Representative histograms of MitoTracker Deep Red staining from four experiments and (H) basal OCR from two experiments of transduced cells.
See also Figure S3.
Cell 166, 63–76, June 30, 2016 69

Figure 4. Mitochondrial Fusion Improves Adoptive Cellular Immunotherapy against Tumors
(A and B) C57BL/6 mice inoculated s.c. with 106 EL4-OVA cells. (A) After 5 or (B) 12 days, 106 or 5 3 106 OT-I IL-2 TE cells cultured with DMSO or M1 + Mdivi-1
were transferred i.v. into recipients and tumor growth assessed. Represents two experiments shown as mean ± SEM (n = 5/group, *p < 0.05, **p < 0.005).
(C–E) Human CD8+ PBMCs activated with aCD3/CD28 + IL-2 to generate IL-2 TE cells. (C) Confocal images of indicated cells where mitochondria are green
(MitoTracker) and nuclei are blue (Hoechst). Representative images from two of four biological donors. Scale bar, 5 mm. (D) OCR/ECAR ratios and SRC of
indicated cells from 2 separate donors shown as mean ± SEM (*p = 0.0303, **p < 0.005, ***p < 0.0001). (E) MitoTracker Green staining and CD62L, CD45RO, and
CCR7 expression analyzed by flow cytometry shown with representative histograms from four to six biological replicates.
See also Figure S4.
We reasoned that fusion renders tightly configured cristae, that this predominantly occurs in TM cells. However, if electron
which results in closely associated ETC complexes and efficient transport across the ETC became less efficient, caused by phys-
OXPHOS (Patten et al., 2014), producing conditions that favor ical separation of the individual complexes due to cristae remod-
the entrance of pyruvate into the TCA cycle. NADH generated eling via mitochondrial fission, then electrons could linger in the
from the TCA cycle is able to easily donate electrons to complex complexes and imbalance redox reactions. NADH levels would
I, which are passed efficiently along the ETC. Our data suggest build, slowing forward momentum of the TCA cycle. To restore
70 Cell 166, 63–76, June 30, 2016

Figure 5. Fusion Promotes Memory T Cell Metabolism, but Opa1 Is Not Required for FAO
OCR measured at baseline and in response to media, etomoxir (Eto) and other drugs as indicated of (A) IL-2 TE cells cultured in DMSO or M1 + Mdivi-1, (B) control
or Opa1 transduced IL-2 TE cells, (C) Opa1+/+ and Opa1 / IL-2 TE cells cultured in DMSO or M1 + Mdivi-1, (D) or without drugs, and (E) ex vivo donor OT-I Opa1+/+
and Opa1 / day 7 TE cells derived from LmOVA infection. Representative of two experiments shown as mean ± SEM (***p < 0.0001).
See also Figure S5.
redox balance, cells could augment glycolysis and shunt pyru- aCD3/CD28 beads (Figure 6I), or with PMA + ionomycin (Fig-
vate as excreted lactate (i.e., aerobic glycolysis), regenerating ure S6C), in the presence or absence of Mdivi-1 to modulate ac-
NAD+ from cytosolic NADH. We speculate that this occurs in tivity of mitochondrial fission protein Drp1 (Cassidy-Stone et al.,
TE cells. Correlating with this idea, we previously reported that 2008). We observed dramatic changes to cristae morphology by
TE and TM cells have different ratios of NAD+/NADH (i.e., redox EM, with the intermembrane space widening over time in con-
balance) with TM cells maintaining higher NAD+/NADH than TE trols compared to drug-treated cells. These data are consistent
cells. We also showed that NADH levels dramatically rise in TM with the hypothesis that fission-induced mitochondrial cristae
cells compared to TE cells when exposed to rotenone and anti- remodeling supports metabolic reprogramming in T cells.
mycin A, indicating that TM cells consume more NADH for the
purpose of donating electrons to the ETC (van der Windt et al., TM Cells Maintain Tight Cristae with Closely Associated
2012). Together, our data suggest that fission and fusion regulate ETC Complexes
cristae remodeling, which alter ETC efficiency and redox bal- Our data suggested that unlike TE cells, TM cells have tight
ance, ultimately controlling metabolic adaptations in T cells. cristae with closely associated ETC complexes. To investigate
To examine this idea further, we assessed cristae morphology this biochemically, we treated native lysates of IL-2 TE and IL-
in TE and TM cells by EM after TCR stimulation. We hypothesized 15 TM cells with increasing concentrations of digitonin to disrupt
that if cristae remodeling induces aerobic glycolysis, changes in cell membranes (including mitochondrial). The crude mem-
cristae structure could be visualized after T cell activation. TM brane-bound fraction was separated from solubilized proteins
cells rapidly augment aerobic glycolysis when restimulated by centrifugation. Both pellet and soluble supernatants were
(van der Windt et al., 2013). We activated IL-15 TM cells with loaded on a denaturing reducing gel and then probed for various
Cell 166, 63–76, June 30, 2016 71

72 Cell 166, 63–76, June 30, 2016

mitochondrial proteins by western blot. We found that mitochon- show that cristae remodeling and/or fission acts as a signal to
dria in IL-2 TE cells were susceptible to digitonin disruption, indi- drive the induction of aerobic glycolysis and subsequent cell
cated by the fact that ETC complex proteins became less detect- activation via Drp1.
able in the pellet and enriched in the soluble fraction in 0.5%
detergent (Figures 6J and S6D). This was in contrast to IL-15 DISCUSSION
TM cells, where ETC proteins did not solubilize to the same
extent as those in IL-2 TE cells into the supernatant, even when Although TM cells rely on FAO for development and survival, pre-
2% digitonin was used. To investigate whether this phenomenon cisely why TM cells utilize FAO and the signals that drive the
was unique to the mitochondrial compartment, we also probed induction of aerobic glycolysis in TE cells remain unclear. Our
for the ER integral protein calnexin and found that it solubilized data suggest that manipulating the structure of a single organelle
similarly in 0.5% digitonin in both cell types. Overall, these can have profound consequences that impact metabolism and
data suggest that there is more exposed mitochondrial mem- ultimately cell differentiation. We found that Opa1 regulated
brane between proteins in IL-2 TE than IL-15 TM cells and is tight cristae organization in TM cells, which facilitated efficient
consistent with the idea that TM cells have tight cristae that yield ETC activity and favorable redox balance, allowing continued
efficient ETC activity, while TE cells have looser cristae with less entrance of pyruvate into mitochondria. We originally hypothe-
efficient ETC activity, ultimately supporting their distinct meta- sized that Opa1 would be required for FAO. However, we found
bolic phenotypes. that Opa1 / IL-2 TE and ex vivo TE cells generated during infec-
tion utilized FAO to the same level as controls. While this was true
Mitochondrial Fission in Activated Immune Cells for TE cells, this may not be the case for TM cells, whose survival
Facilitates Aerobic Glycolysis is severely impaired in vitro and in vivo when Opa1-deficient. It
Our data support a model wherein cristae remodeling, through is possible that Opa1 / T cells are unable to form TM cells
fission and fusion events, is a mechanism to regulate efficient because they cannot efficiently engage FAO under the metabolic
OXPHOS and FAO in TM cells, as well as the induction of aerobic constraints imposed during TM cell development. Previous
glycolysis in TE cells. To more directly test this idea, we assessed studies point to the existence of a ‘‘futile’’ cycle of fatty acid syn-
ECAR of IL-15 TM cells stimulated with aCD3/28 beads in the thesis (FAS) and FAO in TM cells (O’Sullivan et al., 2014; Cui et al.,
presence or absence of Mdivi-1. We found that when mitochon- 2015) whereby carbon derived from glucose is used to build fat
drial fission protein Drp1 was inhibited with Mdivi-1, T cell activa- that is subsequently burned by mitochondria. TM cells have a
tion did not robustly increase aerobic glycolysis when compared lower overall metabolic rate than TE cells, and tightly configured
to controls (Figure 6K), which correlated with our EM data (Fig- cristae may be important to ensure that any pyruvate generated
ure 6I). Because fission is associated with cell division (Taguchi will efficiently feed into the TCA cycle not only for reducing equiv-
et al., 2007), we tested our idea in a non-proliferating cell that alents, but also for deriving citrate for FAS. Without tight cristae
substantially augments aerobic glycolysis upon stimulation and efficient ETC activity, electrons may loiter in the complexes
(Krawczyk et al., 2010). We stimulated bone marrow-derived causing more ROS that could be damaging, but also provide
dendritic cells (BM-DCs) and macrophages (BM-Macs) with lipo- signals that drive cell activation (Sena et al., 2013).
polysaccharide (LPS) with or without interferon (IFN)-g in the We did not observe a defect in TM cell survival in Mfn1 / or
presence or absence of Mdivi-1 and measured ECAR. Aerobic Mfn2 / T cells, but this does not exclude the possibility that
glycolysis was curtailed in BM-DCs and BM-Macs after stimula- OMM fusion or additional activities ascribed to each protein
tion when Drp1 was inhibited (Figure 6L). The blunted ECAR in are not important. Mfn1 and Mfn2 form homotypic and hetero-
Mdivi-1-treated cells correlated with decreased nitric oxide syn- typic interactions, suggesting that in the absence of one, the
thase 2 (Nos2) protein expression in the BM-Macs (Figure 6M), other can compensate (Chen et al., 2003). Our results show
indicating that their activation was also repressed. These data that unlike Opa1 / T cells, in vitro cultured Mfn1 / or Mfn2 /
Figure 6. Mitochondrial Cristae Remodeling Signals Metabolic Pathway Engagement

(A) Basal ECAR of OT-I Opa1+/+ and Opa1 / IL-2 TE cells (left) and day 7 TE cells isolated ex vivo after adoptive transfer from LmOVA infection (right). Data
combined from two to three experiments (*p = 0.0412, ***p < 0.0001).
(B) OCR at baseline and with indicated drugs, representative of two experiments shown as mean ± SEM.
(C) D-Glucose-13C1,2 trace analysis of OT-I Opa1+/+ and Opa1 / IL-2 TE cells. Each lane represents separate mice with a technical replicate.
(D and E) EM analysis of mitochondrial cristae from (D) TE and TM cells isolated after LmOVA infection and (E) in vitro cultured IL-2 TE and TM cells. Representative
of two experiments. Scale bar, 0.25 mm.
(F–H) Relative proton leak (DOCR after oligomycin and subsequent injection of rotenone plus antimycin A) of (F) Opa1+/+ and Opa1 / IL-2 TE, (G) infection-elicited
TE and TM, and (H) IL-2 TE and IL-15 TM cells. Combined from two to four experiments shown as mean ± SEM (p** < 0.005, ***p < 0.0001).
(I) EM analysis of IL-15 TM cell-mitochondrial cristae before and after aCD3/CD28 bead stimulation over hours. Scale bar, 0.2 mm. Represents one experiment.
(J) Immunoblot analysis of calnexin and ETC complexes (CI-NDUFB8, CII-SDHB, CIII-UQCRC2, CIV-MTC01, CV-ATP5A). Equivalent numbers of IL-2 TE and IL-
15 TM cells lysed in native lysis buffer followed by digitonin solubilization of intracellular membranes with pellet (P) and solubilized supernatant (S) fractions
resolved on a denaturing gel, representative of two experiments.
(K–M) IL-15 TM cell (K), BM-DCs (L), and BM-Macs (M) % ECAR measured at baseline and after media, aCD3/CD28 bead, LPS, or LPS+IFN-g injection as
indicated. Data baselined prior to or right after injection with stimuli. (M) BM-Macs stained for intracellular Nos2 protein by flow cytometry with MFI values (left) and
representative histogram (right). Shown as mean ± SEM and represents two to three experiments (***p < 0.0001).
See also Figure S6.
Cell 166, 63–76, June 30, 2016 73

T cells do not have a survival defect when differentiated in IL-15, to function. In fact, although TE cells have less efficient OXPHOS
even though, like Opa1 / T cells, they are more glycolytic and in terms of coupling to ATP synthesis, TE cells are very metabol-
OXPHOS-impaired compared to controls (data not shown). ically active with high OCR and ECAR (Chang et al., 2013; Sena
Our imaging data showed that TM cells maintained fused mito- et al., 2013). Our experiments involving pharmacological
chondrial networks, suggesting that OMM fusion also has enforcement of mitochondrial fusion promoted OCR and SRC
a compulsory role in TM cell development. However, unlike (and ECAR, albeit to a lesser extent) in IL-2 TE cells. The drug-
Opa1, retroviral expression of Mfn1 and Mfn2 did not confer a modified cells maintained full TE cell function with no effect on
TM cell phenotype in TE cells. In this setting, an increase in cytolytic ability or proliferation, but possessed enhanced cyto-
OMM fusion without a concomitant increase in inner membrane kine expression. Fusion and/or cristae tightening boosted oxida-
fusion may still yield loose cristae morphology and a redox state tive capacity in TE cells, endowing them with longevity and
that by default results in sustained excretion of lactate. persistence, while their higher aerobic glycolysis supported
The question of what initial signals drive T cell mitochondrial increased cytokine production, which may explain their superior
remodeling still remains. In the case of TM cell development, antitumor function.
withdrawal of activating signals and growth factors may induce Our data suggest a model where morphological changes in
fusion, consistent with previous reports that starvation induces mitochondria are a primary signal that shapes metabolic reprog-
mitochondrial hyperfusion (Rambold et al., 2015), an effect we ramming during cellular quiescence or activation. We speculate
also observe in TE cells after IL-2 withdrawal (data not shown). that fission-associated expansion of cristae as a result of TCR
However, pro-survival signals from cytokines like IL-15 or IL-7 stimulation physically separates ETC complexes, decreasing
are needed to sustain TM cell viability and metabolically remodel ETC efficiency. With delayed movement of electrons down the
these cells for FAS and FAO via increased CPT1a (van der Windt ETC, NADH levels rise, slowing forward momentum of the TCA
et al., 2012) and aquaporin 9 expression (Cui et al., 2015). Fac- cycle and cause an initial drop in ATP. To correct redox balance,
tors such as these may enforce fusion and would be consistent cells will export pyruvate to lactate to regenerate NAD+ in the
with our observations that mitochondria in activated T cells sub- cytosol, which can enter the mitochondria through various shut-
sequently cultured in IL-15 fuse over time. Another possibility is tles to restore redox balance (Dawson, 1979) and increase flux
that during metabolic stress, Opa1 is activated via sirtuin 3 through glycolysis to restore ATP levels, all contributing to the
(SIRT3) (Samant et al., 2014). Sirtuins are post-translational Warburg effect in activated T cells. When cristae are tightly
modifiers that are activated by NAD+, directly tying their activity configured, the ETC works efficiently and maintains entrance
to cell metabolism (Houtkooper et al., 2012; Wang and Green, of pyruvate into the mitochondria with a favorable redox balance.
2012). We previously published that there is more available In this case, cristae morphology as a result of fusion directs TM
NAD+ in TM cells (van der Windt et al., 2012), which may correlate cell formation and retains these cells in a quiescent state.
with this scenario. Thus, mitochondrial dynamics controls the balance between
In TE cells a day after TCR stimulation, we saw activation of metabolic pathway engagement and T cell fate.
Drp1 via its phosphorylation site Ser616 prior to seeing a fissed
phenotype. TCR signals instigate Ca2+ flux that promotes the EXPERIMENTAL PROCEDURES
phosphatase activity of calcineurin (Smith-Garvin et al., 2009),
which in turn dephosphorylates Drp1 at Ser637, leading to its See the Supplemental Information for details.
activation (Cereghetti et al., 2008). Initial Drp1 activation could
Mice and Immunizations
facilitate some level of fission and cristae remodeling, tipping
Mice were purchased from The Jackson Laboratory. LmOVA deleted for actA
off aerobic glycolysis via the initial shunting of pyruvate to
was used for primary and secondary immunizations. For tumor studies, EL4
lactate. Although inhibition of Drp1 blocks activation-induced lymphoma cells expressing OVA (EL4-OVA) were injected subcutaneously
ECAR, our preliminary data do not show overt mitochondrial (s.c.) into the flank of mice.
fragmentation in the early hours after TLR stimulation of DC or
macrophages (data not shown). This however, does not exclude Flow Cytometry and Imaging
the possibility that Drp1 is mediating subtle changes to mito- Antibody and H2-KbOVA257–264 (KbOVA) MHC-peptide tetramer staining were
chondrial structure that are not discernable by confocal micro- performed as described (Chang et al., 2015). For electron microscopy (EM),
cells were fixed in 2% paraformaldehyde, 2.5% glutaraldehyde in 100 mM
scopy. Drp1 also affects cristae structure by altering the fluidity
sodium cocodylate containing 0.05% malachite green.
of the mitochondrial membrane (Benard et al., 2007). While
Drp1 has been implicated in mitochondrial positioning at the Metabolism Assays
immune synapse (Baixauli et al., 2011), lymphocyte chemotaxis OCR and ECAR were measured using a 96-well XF or XFe extracellular flux
(Campello et al., 2006), and ROS production during T cell activa- analyzer (EFA) (Seahorse Bioscience). For fission inhibition studies, cells
tion (Röth et al., 2014), our data suggest that in addition to these were plated in assay media with 10 mM Mdivi-1 or vehicle control (DMSO) fol-
processes, cristae remodeling via fission underlies the program- lowed by assay media or aCD3/CD28 bead or 20 ng/ml LPS ± 50 ng/ml IFN-g
injection.
ming of cells to aerobic glycolysis.
We show that mitochondria in IL-2 TE cells are more suscepti-
Glucose Tracing
ble to digitonin disruption than IL-15 TM cells, suggesting a
Cells were activated in glucose-free media (prepped with dialyzed FBS) sup-
more exposed membrane with less densely packed protein plemented with 11 mM glucose. After 3 days, cells were washed and cultured
complexes. This relatively enhanced permeability, however, overnight in media replaced with 11 mM D-[1,213C] glucose. Metabolites were
does not mean that their mitochondria are damaged or unable extracted with MeOH and analyzed by mass spectrometry (MS).
74 Cell 166, 63–76, June 30, 2016

Adoptive Transfers Carrio, R., Bathe, O.F., and Malek, T.R. (2004). Initial antigen encounter pro-
For in vivo TM cell studies, %104 OT-I+ CD8+ donor cells/mouse were trans- grams CD8+ T cells competent to develop into memory cells that are activated
ferred intravenously (i.v.) into congenic C57BL/6 mice. For in vivo survival ex- in an antigen-free, IL-7- and IL-15-rich environment. J. Immunol. 172, 7315–
periments, 1–2 3 106 day 6 IL-2 TE-treated cells/mouse were injected i.v. into 7323.
congenic C57BL/6 mice. For tumor studies, 1–5 3 106 day 6 IL-2 TE-treated Cassidy-Stone, A., Chipuk, J.E., Ingerman, E., Song, C., Yoo, C., Kuwana, T.,
cells/mouse were injected i.v. into previously EL4-OVA-inoculated mice. Kurth, M.J., Shaw, J.T., Hinshaw, J.E., Green, D.R., and Nunnari, J. (2008).
Chemical inhibition of the mitochondrial division dynamin reveals its role in
Retroviral Transduction Bax/Bak-dependent mitochondrial outer membrane permeabilization. Dev.
Activated OT-I splenocytes were transduced with control (empty vector) or Cell 14, 193–204.
Mfn1, Mfn2, Opa1 expressing retrovirus by centrifugation. GFP or human
Cereghetti, G.M., Stangherlin, A., Martins de Brito, O., Chang, C.R.,
CD8 marked retroviral expression.
Blackstone, C., Bernardi, P., and Scorrano, L. (2008). Dephosphorylation by
calcineurin regulates translocation of Drp1 to mitochondria. Proc. Natl.
Statistical Analysis Acad. Sci. USA 105, 15803–15808.
Comparisons for two groups were calculated using unpaired two-tailed Stu-
Chan, D.C. (2012). Fusion and fission: interlinked processes critical for mito-
dent’s t tests and one-way ANOVA followed by Bonferroni’s multiple compar-
chondrial health. Annu. Rev. Genet. 46, 265–287.
ison tests for more than two groups. Comparisons over time were calculated
using two-way ANOVA followed by Bonferroni’s multiple comparison tests. Chang, C.H., Curtis, J.D., Maggi, L.B., Jr., Faubert, B., Villarino, A.V., O’Sulli-
van, D., Huang, S.C., van der Windt, G.J., Blagih, J., Qiu, J., et al. (2013). Post-
transcriptional control of T cell effector function by aerobic glycolysis. Cell 153,
SUPPLEMENTAL INFORMATION
1239–1251.
Supplemental Information includes Supplemental Experimental Procedures Chang, C.H., Qiu, J., O’Sullivan, D., Buck, M.D., Noguchi, T., Curtis, J.D.,
and six figures and can be found with this article online at http://dx.doi.org/ Chen, Q., Gindin, M., Gubin, M.M., van der Windt, G.J., et al. (2015). Metabolic
10.1016/j.cell.2016.05.035. competition in the tumor microenvironment is a driver of cancer progression.
Cell 162, 1229–1241.
AUTHOR CONTRIBUTIONS Chen, H., Detmer, S.A., Ewald, A.J., Griffin, E.E., Fraser, S.E., and Chan, D.C.
(2003). Mitofusins Mfn1 and Mfn2 coordinately regulate mitochondrial fusion
M.D.B., D.O., R.I.K.G., D.E.S., D.B., B.T.E., E.J.P., H.S., T.B.H., A.S.R., and and are essential for embryonic development. J. Cell Biol. 160, 189–200.
E.L.P. designed the research and analyzed data. M.D.B., D.O., R.I.K.G., Civiletto, G., Varanita, T., Cerutti, R., Gorletta, T., Barbaro, S., Marchet, S.,
J.D.C., C.-H.C., D.E.S., J.Q., O.K., D.B., G.v.d.W., S.C.-C.H., and C.M.O. per- Lamperti, C., Viscomi, C., Scorrano, L., and Zeviani, M. (2015). Opa1 overex-
formed experiments. M.D.B. and E.L.P wrote the manuscript. pression ameliorates the phenotype of two mitochondrial disease mouse
models. Cell Metab. 21, 845–854.
ACKNOWLEDGMENTS Cogliati, S., Frezza, C., Soriano, M.E., Varanita, T., Quintana-Cabrera, R.,
Corrado, M., Cipolat, S., Costa, V., Casarin, A., Gomes, L.C., et al. (2013).
We thank David Chan (Caltech) for Mfn1/2 floxed mice, Tom Graeber (DMMP, Mitochondrial cristae shape determines respiratory chain supercomplexes
David Geffen School of Medicine, UCLA Metabolomics Center), Wandy Beatty assembly and respiratory efficiency. Cell 155, 160–171.
(Molecular Microbiology Imaging Facility, Washington University School of
Cui, G., Staron, M.M., Gray, S.M., Ho, P.C., Amezquita, R.A., Wu, J., and
Medicine), Erica Lantelme, Dorjan Brinja, Barbara Joch, Hani Suleiman, Elena
Kaech, S.M. (2015). IL-7-induced glycerol transport and TAG synthesis pro-
Tonc, Tara Bradstreet, and Elizabeth Schwarzkopf for technical assistance.
motes memory CD8+ T cell longevity. Cell 161, 750–761.
This work was funded by the NIH (CA181125 and AI091965 to E.L.P. and
AI110481 to E.J.P.), the Burroughs Wellcome Fund (Investigator in the Patho- Dawson, A.G. (1979). Oxidation of cytosolic NADH formed during aerobic
genesis of Infectious Disease Award to E.L.P. and Career Award for Medical metabolism in mammalian cells. Trends Biochem. Sci. 4, 171–176.
Scientists to B.T.E.), the BMBF (to A.S.R.), the Max Planck Society, and the de Brito, O.M., and Scorrano, L. (2008). Mitofusin 2 tethers endoplasmic retic-
NSF Graduate Research Fellowship DGE-1143954 (to M.D.B.). ulum to mitochondria. Nature 456, 605–610.
Deberardinis, R.J., Lum, J.J., and Thompson, C.B. (2006). Phosphatidylinosi-
Received: December 24, 2015 tol 3-kinase-dependent modulation of carnitine palmitoyltransferase 1A
Revised: April 3, 2016 expression regulates lipid metabolism during hematopoietic cell growth.
Accepted: May 6, 2016 J. Biol. Chem. 281, 37372–37380.
Published: June 9, 2016
Frank, M., Duvezin-Caubet, S., Koob, S., Occhipinti, A., Jagasia, R., Petcher-
ski, A., Ruonala, M.O., Priault, M., Salin, B., and Reichert, A.S. (2012). Mitoph-
REFERENCES
agy is triggered by mild oxidative stress in a mitochondrial fission dependent
manner. Biochim. Biophys. Acta 1823, 2297–2310.
Archer, S.L. (2014). Mitochondrial fission and fusion in human diseases.
N. Engl. J. Med. 370, 1074. Frezza, C., Cipolat, S., Martins de Brito, O., Micaroni, M., Beznoussenko, G.V.,
Rudka, T., Bartoli, D., Polishuck, R.S., Danial, N.N., De Strooper, B., and Scor-
Baixauli, F., Martı́n-Cófreces, N.B., Morlino, G., Carrasco, Y.R., Calabia-
rano, L. (2006). OPA1 controls apoptotic cristae remodeling independently
Linares, C., Veiga, E., Serrador, J.M., and Sánchez-Madrid, F. (2011). The
from mitochondrial fusion. Cell 126, 177–189.
mitochondrial fission factor dynamin-related protein 1 modulates T-cell recep-
tor signalling at the immune synapse. EMBO J. 30, 1238–1250. Friedman, J.R., and Nunnari, J. (2014). Mitochondrial form and function.
Nature 505, 335–343.
Benard, G., Bellance, N., James, D., Parrone, P., Fernandez, H., Letellier, T.,
and Rossignol, R. (2007). Mitochondrial bioenergetics and structural network Gomes, L.C., Di Benedetto, G., and Scorrano, L. (2011). During autophagy
organization. J. Cell Sci. 120, 838–848. mitochondria elongate, are spared from degradation and sustain cell viability.
Buck, M.D., O’Sullivan, D., and Pearce, E.L. (2015). T cell metabolism drives Nat. Cell Biol. 13, 589–598.
immunity. J. Exp. Med. 212, 1345–1360. Houtkooper, R.H., Pirinen, E., and Auwerx, J. (2012). Sirtuins as regulators of
Campello, S., Lacalle, R.A., Bettella, M., Mañes, S., Scorrano, L., and Viola, A. metabolism and healthspan. Nat. Rev. Mol. Cell Biol. 13, 225–238.
(2006). Orchestration of lymphocyte chemotaxis by mitochondrial dynamics. Huang, S.C., Everts, B., Ivanova, Y., O’Sullivan, D., Nascimento, M., Smith,
J. Exp. Med. 203, 2879–2886. A.M., Beatty, W., Love-Gregory, L., Lam, W.Y., O’Neill, C.M., et al. (2014).
Cell 166, 63–76, June 30, 2016 75

Cell-intrinsic lysosomal lipolysis is essential for alternative activation of macro- activates OPA1 to regulate mitochondrial dynamics during stress. Mol. Cell.
phages. Nat. Immunol. 15, 846–855. Biol. 34, 807–819.
Kaech, S.M., Tan, J.T., Wherry, E.J., Konieczny, B.T., Surh, C.D., and Ahmed, Santel, A., Frank, S., Gaume, B., Herrler, M., Youle, R.J., and Fuller, M.T.
R. (2003). Selective expression of the interleukin 7 receptor identifies effector (2003). Mitofusin-1 protein is a generally expressed mediator of mitochondrial
CD8 T cells that give rise to long-lived memory cells. Nat. Immunol. 4, 1191– fusion in mammalian cells. J. Cell Sci. 116, 2763–2774.
1198. Schluns, K.S., Williams, K., Ma, A., Zheng, X.X., and Lefrançois, L. (2002). Cut-
Krawczyk, C.M., Holowka, T., Sun, J., Blagih, J., Amiel, E., DeBerardinis, R.J., ting edge: requirement for IL-15 in the generation of primary and memory an-
Cross, J.R., Jung, E., Thompson, C.B., Jones, R.G., and Pearce, E.J. (2010). tigen-specific CD8 T cells. J. Immunol. 168, 4827–4831.
Toll-like receptor-induced changes in glycolytic metabolism regulate dendritic Sebastián, D., Hernández-Alvarez, M.I., Segalés, J., Sorianello, E., Muñoz,
cell activation. Blood 115, 4742–4749. J.P., Sala, D., Waget, A., Liesa, M., Paz, J.C., Gopalacharyulu, P., et al.
Liesa, M., and Shirihai, O.S. (2013). Mitochondrial dynamics in the regulation of (2012). Mitofusin 2 (Mfn2) links mitochondrial and endoplasmic reticulum func-
nutrient utilization and energy expenditure. Cell Metab. 17, 491–506. tion with insulin signaling and is essential for normal glucose homeostasis.
MacIver, N.J., Michalek, R.D., and Rathmell, J.C. (2013). Metabolic regulation Proc. Natl. Acad. Sci. USA 109, 5523–5528.
of T lymphocytes. Annu. Rev. Immunol. 31, 259–283. Sena, L.A., Li, S., Jairaman, A., Prakriya, M., Ezponda, T., Hildeman, D.A.,
Marsboom, G., Toth, P.T., Ryan, J.J., Hong, Z., Wu, X., Fang, Y.H., Thenap- Wang, C.R., Schumacker, P.T., Licht, J.D., Perlman, H., et al. (2013). Mito-
pan, T., Piao, L., Zhang, H.J., Pogoriler, J., et al. (2012). Dynamin-related pro- chondria are required for antigen-specific T cell activation through reactive ox-
tein 1-mediated mitochondrial mitotic fission permits hyperproliferation of ygen species signaling. Immunity 38, 225–236.
vascular smooth muscle cells and offers a novel therapeutic target in pulmo- Serasinghe, M.N., Wieder, S.Y., Renault, T.T., Elkholi, R., Asciolla, J.J., Yao,
nary hypertension. Circ. Res. 110, 1484–1497. J.L., Jabado, O., Hoehn, K., Kageyama, Y., Sesaki, H., and Chipuk, J.E.
Maus, M.V., Fraietta, J.A., Levine, B.L., Kalos, M., Zhao, Y., and June, C.H. (2015). Mitochondrial division is requisite to RAS-induced transformation
(2014). Adoptive immunotherapy for cancer or viruses. Annu. Rev. Immunol. and targeted by oncogenic MAPK pathway inhibitors. Mol. Cell 57, 521–536.
32, 189–225. Smith-Garvin, J.E., Koretzky, G.A., and Jordan, M.S. (2009). T cell activation.
Mishra, P., and Chan, D.C. (2016). Metabolic regulation of mitochondrial Annu. Rev. Immunol. 27, 591–619.
dynamics. J. Cell Biol. 212, 379–387. Taguchi, N., Ishihara, N., Jofuku, A., Oka, T., and Mihara, K. (2007). Mitotic
Mishra, P., Carelli, V., Manfredi, G., and Chan, D.C. (2014). Proteolytic cleav- phosphorylation of dynamin-related GTPase Drp1 participates in mitochon-
age of Opa1 stimulates mitochondrial inner membrane fusion and couples drial fission. J. Biol. Chem. 282, 11521–11529.
fusion to oxidative phosphorylation. Cell Metab. 19, 630–641. Toyama, E.Q., Herzig, S., Courchet, J., Lewis, T.L., Jr., Losón, O.C., Hellberg,
Nicholls, D.G. (2009). Spare respiratory capacity, oxidative stress and excito- K., Young, N.P., Chen, H., Polleux, F., Chan, D.C., and Shaw, R.J. (2016).
toxicity. Biochem. Soc. Trans. 37, 1385–1388. Metabolism. AMP-activated protein kinase mediates mitochondrial fission in
Nunnari, J., and Suomalainen, A. (2012). Mitochondria: in sickness and in response to energy stress. Science 351, 275–281.
health. Cell 148, 1145–1159. van der Windt, G.J., Everts, B., Chang, C.H., Curtis, J.D., Freitas, T.C., Amiel,
O’Sullivan, D., and Pearce, E.L. (2015). Targeting T cell metabolism for ther- E., Pearce, E.J., and Pearce, E.L. (2012). Mitochondrial respiratory capacity is
apy. Trends Immunol. 36, 71–80. a critical regulator of CD8+ T cell memory development. Immunity 36, 68–78.
O’Sullivan, D., van der Windt, G.J., Huang, S.C., Curtis, J.D., Chang, C.H., van der Windt, G.J., O’Sullivan, D., Everts, B., Huang, S.C., Buck, M.D., Curtis,
Buck, M.D., Qiu, J., Smith, A.M., Lam, W.Y., DiPlato, L.M., et al. (2014). Mem- J.D., Chang, C.H., Smith, A.M., Ai, T., Faubert, B., et al. (2013). CD8 memory
ory CD8(+) T cells use cell-intrinsic lipolysis to support the metabolic program- T cells have a bioenergetic advantage that underlies their rapid recall ability.
ming necessary for development. Immunity 41, 75–88. Proc. Natl. Acad. Sci. USA 110, 14336–14341.
Patten, D.A., Wong, J., Khacho, M., Soubannier, V., Mailloux, R.J., Pilon-Lar- Vander Heiden, M.G., Cantley, L.C., and Thompson, C.B. (2009). Understand-
ose, K., MacLaurin, J.G., Park, D.S., McBride, H.M., Trinkle-Mulcahy, L., et al. ing the Warburg effect: the metabolic requirements of cell proliferation. Sci-
(2014). OPA1-dependent cristae modulation is essential for cellular adaptation ence 324, 1029–1033.
to metabolic demand. EMBO J. 33, 2676–2691. Wai, T., and Langer, T. (2016). Mitochondrial dynamics and metabolic regula-
Pearce, E.L., Walsh, M.C., Cejas, P.J., Harms, G.M., Shen, H., Wang, L.S., tion. Trends Endocrinol. Metab. 27, 105–117.
Jones, R.G., and Choi, Y. (2009). Enhancing CD8 T-cell memory by modulating Wang, R., and Green, D.R. (2012). Metabolic checkpoints in activated T cells.
fatty acid metabolism. Nature 460, 103–107. Nat. Immunol. 13, 907–915.
Pearce, E.L., Poffenberger, M.C., Chang, C.H., and Jones, R.G. (2013). Fueling Wang, D., Wang, J., Bonamy, G.M., Meeusen, S., Brusch, R.G., Turk, C., Yang,
immunity: insights into metabolism and lymphocyte function. Science 342, P., and Schultz, P.G. (2012). A small molecule promotes mitochondrial fusion
1242454. in mammalian cells. Angew. Chem. Int. Ed. Engl. 51, 9302–9305.
Rambold, A.S., Kostelecky, B., Elia, N., and Lippincott-Schwartz, J. (2011). Wang, L., Ishihara, T., Ibayashi, Y., Tatsushima, K., Setoyama, D., Hanada, Y.,
Tubular network formation protects mitochondria from autophagosomal Takeichi, Y., Sakamoto, S., Yokota, S., Mihara, K., et al. (2015). Disruption of
degradation during nutrient starvation. Proc. Natl. Acad. Sci. USA 108, mitochondrial fission in the liver protects mice from diet-induced obesity and
10190–10195. metabolic deterioration. Diabetologia 58, 2371–2380.
Rambold, A.S., Cohen, S., and Lippincott-Schwartz, J. (2015). Fatty acid Williams, M.A., and Bevan, M.J. (2007). Effector and memory CTL differentia-
trafficking in starved cells: regulation by lipid droplet lipolysis, autophagy, tion. Annu. Rev. Immunol. 25, 171–192.
and mitochondrial fusion dynamics. Dev. Cell 32, 678–692. Youle, R.J., and Karbowski, M. (2005). Mitochondrial fission in apoptosis. Nat.
Restifo, N.P., Dudley, M.E., and Rosenberg, S.A. (2012). Adoptive immuno- Rev. Mol. Cell Biol. 6, 657–663.
therapy for cancer: harnessing the T cell response. Nat. Rev. Immunol. 12, Yu, T., Robotham, J.L., and Yoon, Y. (2006). Increased production of reactive
269–281. oxygen species in hyperglycemic conditions requires dynamic change of mito-
Röth, D., Krammer, P.H., and Gülow, K. (2014). Dynamin related protein chondrial morphology. Proc. Natl. Acad. Sci. USA 103, 2653–2658.
1-dependent mitochondrial fission regulates oxidative signalling in T cells. Zhang, Z., Wakabayashi, N., Wakabayashi, J., Tamura, Y., Song, W.J.,
FEBS Lett. 588, 1749–1754. Sereda, S., Clerc, P., Polster, B.M., Aja, S.M., Pletnikov, M.V., et al. (2011).
Samant, S.A., Zhang, H.J., Hong, Z., Pillai, V.B., Sundaresan, N.R., Wolfgeher, The dynamin-related GTPase Opa1 is required for glucose-stimulated ATP
D., Archer, S.L., Chan, D.C., and Gupta, M.P. (2014). SIRT3 deacetylates and production in pancreatic beta cells. Mol. Biol. Cell 22, 2235–2245.
76 Cell 166, 63–76, June 30, 2016

Article
Genomic and Transcriptomic

Features of Response to Anti-PD-1
Therapy in Metastatic Melanoma
Willy Hugo,1,6,9 Jesse M. Zaretsky,2,6,9 Lu Sun,1,6 Chunying Song,1,6 Blanca Homet Moreno,3 Siwen Hu-Lieskovan,3
Beata Berent-Maoz,3 Jia Pang,3 Bartosz Chmielowski,3 Grace Cherry,3 Elizabeth Seja,3 Shirley Lomeli,1,6 Xiangju Kong,1,6
Mark C. Kelley,7 Jeffrey A. Sosman,8 Douglas B. Johnson,8 Antoni Ribas,2,3,4,5,6 and Roger S. Lo1,2,5,6,*
1Division of Dermatology, Department of Medicine
2Department of Molecular and Medical Pharmacology
3Division of Hematology & Oncology, Department of Medicine
4Division of Surgical Oncology, Department of Surgery
5Jonsson Comprehensive Cancer Center
6David Geffen School of Medicine
University of California, Los Angeles, CA 90095-1662, USA

7Department of Surgery, Vanderbilt-Ingram Cancer Center, Nashville, TN 37232, USA
8Department of Medicine, Vanderbilt-Ingram Cancer Center, Nashville, TN 37232, USA
9Co-first authors
*Correspondence: rlo@mednet.ucla.edu
SUMMARY benefits (Hamid et al., 2013; Topalian et al., 2012). This therapeu-
tic approach has also been shown to be active against a growing
PD-1 immune checkpoint blockade provides signifi- list of human malignancies, and clinical testing of combinations
cant clinical benefits for melanoma patients. We of PD-1 (or PD-L1) with other treatment targets has already
analyzed the somatic mutanomes and transcrip- begun (Sharma and Allison, 2015). However, effective clinical
tomes of pretreatment melanoma biopsies to identify use of anti-PD-1 agents is encumbered by a high rate of
factors that may influence innate sensitivity or resis- innate resistance (60%–70%) in advanced metastatic mela-
noma. The mechanistic basis for the variation in response pat-
tance to anti-PD-1 therapy. We find that overall high
terns or in long-term clinical benefits (i.e., survival) remains
mutational loads associate with improved survival, poorly explained.
and tumors from responding patients are enriched In melanoma, the extent of pretreatment and especially treat-
for mutations in the DNA repair gene BRCA2. Innately ment-induced intra-tumoral T cell infiltration correlates with
resistant tumors display a transcriptional signature clinical responses (Tumeh et al., 2014), supporting unleashing
(referred to as the IPRES, or innate anti-PD-1 resis- of tumor-specific T cells as the primary mechanistic basis of
tance), indicating concurrent up-expression of genes anti-PD-1 therapy. Preliminary retrospective analyses of clinical
involved in the regulation of mesenchymal transition, data hinted at prior failure of mitogen-activated protein kinase
cell adhesion, extracellular matrix remodeling, angio- (MAPK)-targeted therapy being a negative factor for subsequent
genesis, and wound healing. Notably, mitogen-acti- response to immune checkpoint blockade in melanoma (Puza-
vated protein kinase (MAPK)-targeted therapy nov et al., 2015, Pigment Cell Melanoma Res., abstract;
Ramanujam et al., 2015, Pigment Cell Melanoma Res., abstract;
(MAPK inhibitor) induces similar signatures in mela-
Simeone et al., 2015, Pigment Cell Melanoma Res., abstract). In
noma, suggesting that a non-genomic form of this context, acquired resistance to MAPK-targeted therapy has
MAPK inhibitor resistance mediates cross-resistance been correlated with depletion of intra-tumoral T cells, exhaus-
to anti-PD-1 therapy. Validation of the IPRES in other tion of CD8 T cells, and loss of antigen presentation (Hugo
independent tumor cohorts defines a transcriptomic et al., 2015).
subset across distinct types of advanced cancer. At the genomic level, the overall mutation load has been corre-
These findings suggest that attenuating the biological lated with clinical responses to anti-PD-1 therapy and linked to
processes that underlie IPRES may improve anti- smoking in non-small-cell lung cancer or mismatch repair defi-
PD-1 response in melanoma and other cancer types. ciency in colon cancer. (Le et al., 2015; Rizvi et al., 2015).
Whether it was non-small-cell lung tumors on anti-PD-1 therapy
or melanoma tumors on anti-CTLA-4 therapy (Snyder et al.,
INTRODUCTION 2014; Van Allen et al., 2015), the range of somatic mutation
and neoepitope loads of the responding pretreatment tumors
PD-1 immune checkpoint blockade therapy induces a high rate overlapped significantly with that of the non-responding tumors,
of anti-melanoma response and provides unprecedented clinical despite statistically significant differences in their medians. As
Cell 165, 35–44, March 24, 2016 ª2016 Elsevier Inc. 35

such, the lack of high mutational loads did not necessarily pre- et al., 2014). However, we did not observe enrichment of this
clude clinical responses, and conversely, the presence of high peptide motif in the pretreatment tumors that responded to
mutational loads does not always correlate with responses. anti-PD-1 therapy (Figure S1D). Likewise, analysis of an inde-
Thus, additional genomic or non-genomic features seem likely pendent cohort of 110 melanoma tumors’ pre-anti-CTLA-4 ther-
to contribute to anti-PD-1 response patterns. Here, we sought apy also did not yield enrichment of this tetrapeptide motif
to assess omic-scale features related to clinical response and among responding tumors (Van Allen et al., 2015).
survival patterns in order to gain insights into potential strategies In addition to examining the relationship between non-synon-
for patient stratification and identification of anti-PD-1 combina- ymous somatic mutational loads in pretreatment tumors and
torial therapies. anti-tumor responses elicited by anti-PD-1 antibodies, we also
examined the potential influences of somatic mutational loads
RESULTS AND DISCUSSION on clinical benefits derived from anti-PD-1 immunotherapy as re-
flected by patient survival. Notably, a mutational load in the top
High Mutational Load Does Not Associate with Tumor third (compared to the bottom third) was significantly associated
Response but Correlates with Improved Patient Survival with improved survival (Figure 1A). We also observed a trend
We analyzed the whole-exome sequences (WES) of 38 pre- toward higher mutational load being associated with better sur-
treatment (pembrolizumab and nivolumab) melanoma tumors vival among melanoma patients not treated with anti-PD-1 anti-
(responding, n = 21; non-responding, n = 17; total 34 of 38 pre- bodies (Cancer Genome Atlas Network, 2015), although this
treatment; 4 of 38 early on-treatment; 14 of 38 patients with prior association did not reach statistical significance (Figure S1E),
MAPK inhibitor (MAPKi) treatment; Table S1A) and patient- suggesting that the prognostic power of a high mutational load
matched normal tissues for germline references. Responding is augmented in the setting of anti-PD-1 therapy. As expected,
pretreatment tumors were derived from patients who went on a positive association between objective tumor responses and
to have complete or partial responses or stable disease control survival was highly statistically significant (Figure 1B). Moreover,
(with mixed responses excluded) in response to anti-PD-1 ther- when we divided each non-responding and responding tumor
apy. Non-responding tumors were derived from patients who group into sub-groups with low or high mutational loads (i.e.,
had progressive disease. These response patterns were based below or above the median total somatic nsSNVs of each
on irRECIST (Hoos et al., 2015; Wolchok et al., 2009). We also response group; Figure 1C), patients with responding tumors
analyzed the transcriptomes through RNA-seq of responding of low mutation loads significantly outlived patients with non-
(n = 15) and non-responding (n = 13) pretreatment tumors (total responding tumors of high mutation loads (Figure 1D). This is
27 of 28 pretreatment; 1 of 28 early on-treatment) with available despite the fact that mutational loads of these two groups
high-quality RNA. WES achieved a median of 1403 coverage in were significantly different, with no overlap across the two
both tumor and normal tissues (Table S1B). We detected a me- distributions (Figure 1C). Hence, factors beyond the mutational
dian of 489 non-synonymous somatic mutations in the 38 tumors load also influence shorter-term tumor response patterns and
(range, 73 to 3,985, which is similar to that in a different set of longer-term patient survival.
melanoma tissues; Van Allen et al., 2015).
We found that responding pretreatment tumors on anti-PD-1 Enrichment for BRCA2 Mutations in Anti-PD-1
therapy harbored more non-synonymous single nucleotide vari- Responsive Melanoma
ants (nsSNVs) compared to the non-responding tumors, albeit We then sought to identify mutations (nsSNVs and small inser-
the statistical significance cutoff was not met (median nsSNVs tion-and-deletions [indels]; Table S1D) that (1) were recurrently
responding = 495 and non-responding = 281, p = 0.30, Mann- and selectively associated with either responding or non-re-
Whitney; Table S1B). Increased predicted HLA (human leuko- sponding tumors (recurrence R 25% in one group and, at
cyte antigen) class I and class II neoepitope loads were also most, one hit in the other group) and (2) occurred in genes at
detected in the responding pretreatment tumors, although these rates higher than background rates (Fisher’s exact test, false
differences were not statistically significant either (median HLA discover rate [FDR]-corrected p % 0.05; Figure 1E; Table S1E).
class I neoepitopes responding = 231 and non-responding = The background mutation rate of each gene was calculated
156, p = 0.41; median HLA class II neoepitopes responding = from the WES data of 469 melanoma tumors (Hodis et al.,
130 and non-responding = 95, p = 0.36, Mann-Whitney; Table 2012; Cancer Genome Atlas Network, 2015). Analysis of copy-
S1B). Even when we considered only expressed nsSNV and number variations (CNVs) did not identify any recurrent alteration
neoepitope loads, the statistical significance of the differences exclusive to either group. BRCA2 harbored nsSNVs in 6 of 21 re-
between the responding versus non-responding tumors was sponding tumors (28%), but only 1 of 17 non-responding tumors
not augmented. The comparison of these two groups of tumors (6%; Figure 1E). With a background mutational rate estimated at
was not likely biased by small differences in mean tumor purities 6% (28 of 469 melanoma tumors), BRCA2 was significantly more
or depth of sequencing (Figures S1A and S1B). The numbers of frequently mutated in the responding tumors than expected
predicted HLA class I and II neoepitopes were strongly corre- (Fisher p = 0.002, odds ratio = 6.2). The pattern of mutations in
lated with the number of nsSNVs (Figure S1C). We did not disparate BRCA2 protein domains suggested loss-of-function
identify any recurrent predicted neoepitope or experimentally mutations (Figure 1F): one in the N-terminal NPM1-interacting
validated neoantigens (Table S1C). Previous work analyzing mel- region, one in the POLH-interacting domain, and four in the heli-
anoma tumors sampled prior to anti-CTLA-4 antibody therapy cal domain critical for FANCD2 interaction. Intriguingly, the so-
had associated responses with a tetrapeptide signature (Snyder matic mutational load of the tumors with BRCA2 nsSNVs was
36 Cell 165, 35–44, March 24, 2016 ª2016 Elsevier Inc.

Figure 1. Mutational Correlates of Innate
Sensitivity to Anti-PD-1 Therapy
(A) Overall survival of anti-PD-1-treated patients
whose melanoma tumors harbored high (top third)
versus low (bottom third) mutational (somatic
nsSNVs) loads; p values, log-rank test.
(B) Overall survival of anti-PD-1-treated melanoma
patients whose pretreatment tumors responded
(n = 20) or did not respond (n = 17); p value, log-rank
test.
(C) Total number of nsSNVs detected in anti-PD-1
responding and non-responding melanoma tumors
harboring high (above the respective group’s me-
dian) or low (below the group’s median) mutational
loads; p value, Mann Whitney test.
(D) Overall survival of anti-PD-1-treated melanoma
patients harboring high (above the group median) or
low (below the group median) mutational loads and
whose pretreatment tumors responded or did not
respond; p value, log-rank test.
(E) Recurrent exomic alterations (nsSNVs and small
INDELs) in pretreatment tumors of responding
versus non-responding patients on anti-PD-1 ther-
apy. Copy-number alterations were annotated for
the same genes as a reference. Top, mutations of
melanoma signature genes. Middle, mutations
recurrent in responding versus non-responding tu-
mors (recurrence in 25% in one group and at most
one occurrence in the opposite group, Fisher’s
exact test, FDR-corrected p % 0.05 on enrichment
against the background mutation frequency). Bot-
tom, the total nsSNV load of each melanoma tumor.
(F) Schematics of impact of missense mutations in
the BRCA2 protein and its domains.
(G) Total number of nsSNVs detected in melanomas
with or without BRCA2 non-synonymous mutations;
p value, Mann Whitney test.
See also Table S1 and Figure S1.

significantly higher than that with wild-type BRCA2 in this cohort noma tumors of patients who derived benefit from CTLA-4 anti-
of tumors (Figure 1G), as well as two additional cohorts of mela- bodies (Van Allen et al., 2015). However, these genes, along with
noma tumors (Figure S1F). Thus, BRCA2 loss-of-function muta- other T cell related genes such as CD8A/B, PD-L1, LAG3 (T cell
tions, which are expected to produce defects in homologous checkpoint genes), and IFNG, did not present higher expression
recombination and double-stranded DNA break repair (Hollo- in anti-PD-1-responsive tumors (Figures 2A, bottom, and S2A).
man, 2011), may produce specific mutational signatures or un- Similarly, we did not observe higher enrichment of multiple inter-
known effects (e.g., induction of cell death) that contribute to feron signatures in the anti-PD-1-responsive group (Figure 2D,
anti-PD-1 responsiveness. bottom). Previously, an interferon gamma signature was found
to be differentially up-expressed in the pretreatment tumor
Co-enriched Transcriptomic Signatures in a Major biopsies from responding patients when a restricted set of im-
Subset of Anti-PD-1 Resistant Melanoma mune genes were analyzed (Ribas et al., 2015, J. Clin. Oncol.,
We then addressed whether transcriptomic features would abstract). However, the technical approach may not be compa-
differentiate between responding (n = 15) versus non-responding rable to our whole tumor transcriptomic approach. We did note
(n = 13) tumors sampled prior to anti-PD-1 therapy (total 27 of 28 that the expression levels of HLA class I genes (HLA-A, -B,
pretreatment tumors and 1 of 28 early on-treatment). We or -C) trended higher among the responding tumors, although
compared the transcriptomes of the two tumor groups using the differences were not statistically significant. Lastly, the com-
two approaches: (1) analysis of differentially expressed genes plete loss of PTEN was reported to promote resistance to im-
(DEGs; Figures 2A, top, and 2B) across the two aggregate mune checkpoint blockade (Peng et al., 2016), but there was
groups (Table S2A) coupled with GO term enrichment analysis only one case of homozygous PTEN deletion (with nearly unde-
of DEGs (Figure 2C) and (2) differential signature enrichment tectable PTEN mRNA expression; Figure S2A) in our cohort (in
based on single-sample gene set variance analysis, or GSVA, the non-responsive sub-group), limiting our ability to draw
scores using publicly available (C2 chemical and genetic pertur- meaningful associations in this dataset. Generally, we did not
bation, C6 oncogenic, and C7 immunologic subsets of the Mo- observe a statistically significant difference in PTEN expression
lecular Signature Database, Broad Institute) and self-curated between anti-PD-1 responding versus non-responding tumors.
(see below) perturbation-induced gene signatures (Figure 2D; Thus, individual gene-based expression analysis suggested
Table S2B). mesenchymal and T cell suppressive inflammatory or angiogenic
From analysis of DEGs (cutoff, 2-fold difference between the tumor phenotypes as being associated with innate anti-PD-1
absolute medians of normalized expressions in the two groups; resistance.
nominal Mann-Whitney p % 0.1), we made observations sug- We then queried biological processes represented by DEGs.
gesting that mesenchymal and inflammatory tumor phenotypes While gene ontology (GO) enrichment analysis of genes up-ex-
may be associated with innate anti-PD-1 resistance. First, 693 pressed among responding tumors produced no significantly
genes were differentially expressed between the responding enriched terms, genes up-expressed among non-responding tu-
versus non-responding pretreatment tumors, and the transcrip- mors were enriched for cell adhesion, extracellular matrix (ECM)
tomes of non-responding tumors were dominated by relative organization, wound healing, and angiogenesis (FDR-adjusted
gene up-expression events compared with the transcriptomes p values of GO gene sets shown in Figure 2C). Using indepen-
of responding tumors (Figure 2A, top, showing only genes whose dently derived perturbation-based transcriptomic signatures
differential expression met nominal Mann-Whitney p % 0.05; (Molecular Signature Database; Table S2C), we sought after
Table S2A). Second, DEGs that were expressed higher in non-re- differentially enriched processes in the responding versus non-
sponding pretreatment tumors included mesenchymal transition responding pretreatment tumors (cutoff, 10% difference be-
genes (AXL, ROR2, WNT5A, LOXL2, TWIST2, TAGLN, FAP), tween the absolute medians of GSVA scores in the two groups;
immunosuppressive genes (IL10, VEGFA, VEGFC), and mono- FDR-corrected Welch t test p % 0.25). Gene sets meeting these
cyte and macrophage chemotactic genes (CCL2, CCL7, CCL8 standard cutoffs formed the core set (Figure 2D, top, in bold)
and CCL13; Figures 2A and 2B). In addition to mesenchymal from which we compiled additional concurrently enriched
genes, genes associated with wound healing and angiogenesis, (nominal Welch t test p % 0.1) and functionally related gene
which are considered T cell suppressive (Motz and Coukos, sets (Figure 2D, top; Table S2B). We considered these statisti-
2011; Schäfer and Werner, 2008; Voron et al., 2014), were ex- cally weaker gene set enrichments biologically meaningful given
pressed higher among non-responding relative to responding the functional coherence of these gene signatures with the core
pretreatment tumors. Interestingly, a recent study using a mouse signatures (Subramanian et al., 2005).
melanoma model showed that VEGFA and CCL2 expression Importantly, a group of 26 transcriptomic signatures were co-
was associated with innate anti-PD-1 resistance (Peng et al., enriched en bloc in 9 of 13 non-responding versus 1 of 15 re-
2016). CDH1, which is typically down-expressed by mesen- sponding pre-anti-PD-1 tumors (see Experimental Procedures).
chymal cancer cells, was also down-expressed by non-respond- Co-enrichment of these signatures, collectively referred to as
ing (versus responding) pretreatment tumors. Third, genes with the innate anti-PD-1 resistance (IPRES) signature, again indi-
putative roles in modulating immune checkpoint sensitivity cated heightened mesenchymal transition, angiogenesis, hyp-
were not differentially expressed between responding versus oxia, and wound healing. The concurrence of a tumor cell
non-responding tumor groups (Figures 2A, bottom, and S2A). mesenchymal phenotype with an angiogenesis- and wound-
GZMA, PRF1 (CD8 T cell cytolytic score), PDCD1LG2 (PD-L2), healing-related inflammatory microenvironment has been docu-
and CTLA4 were expressed higher in the pretreatment mela- mented in the literature (Chen et al., 2014, 2015; Mak et al.,

Figure 2. Transcriptomic Signatures of Innate Resistance to Anti-PD-1 Therapy
(A) Top, heatmap showing differentially expressed genes in the pretreatment tumors derived from patients who responded versus who did not respond to anti-
PD-1 treatment (gene expression with inter-quartile range [IQR] R 2; median fold-change [FC] difference R 2; Mann Whitney p % 0.05). Middle, mRNA
expression levels of genes with hypothetical roles in modulating response patterns to anti-PD-1 therapy. Bottom, overall number of nsSNVs and HLA class 1
and 2 neoepitopes (predicted).
(B) mRNA levels of genes (which control tumor cell mesenchymal transition, tumor angiogenesis, and macrophage and monocyte chemotaxis) that were
differentially expressed between the responding versus non-responding pretreatment tumors; p values, Mann Whitney test.
(C) GO enrichment of genes that were expressed higher in the non-responding tumors.
(D) Heatmap showing the gene set variance analysis (GSVA) scores of gene signatures differentially enriched in the responding versus non-responding pre-anti-
PD-1 tumors (absolute median GSVA score difference R 10%, FDR-corrected Welch t test p % 0.25 or nominal Welch t test p % 0.1). For comparison, enrichment
scores of interferon signatures are also displayed.
(E) Overall survival of anti-PD-1-treated melanoma patients with presence (n = 10) or absence (n = 16) of co-enriched innate anti-PD-1 resistance (IPRES)
signatures; p value, log-rank test.
See also Table S2 and Figure S2.

Figure 3. Co-enrichment of Innate Anti-
PD-1 Resistance-Associated Signatures
Defines a Transcriptomic Subset in Mela-
noma and Multiple Cancers
(A) Heatmap showing GSVA scores of IPRES sig-
natures across four independent RNA-seq data-
sets derived from metastatic melanoma. Cohort 1,
pretreatment (anti-PD-1) tumors; cohort 2, pre-
treatment (anti-CTLA-4) tumors; cohort 3, pre-
treatment (MAPKi) tumors; cohort 4, TCGA
cutaneous melanoma (metastatic only).
(B) Heatmap showing GSVA scores of IPRIM sig-
natures across TCGA RNA-seq datasets (meta-
static melanoma, SKCM; lung adenocarcinoma,
LUAD; colon adenocarcinoma, COAD; kidney
clear cell carcinoma, KIRC; and pancreatic
adenocarcinoma, PAAD).
See also Figure S3.
was an exclusive feature of our cohort, we

queried three additional cohorts of meta-
static melanoma-derived RNA-seq (Hugo
et al., 2015; Cancer Genome Atlas
Network, 2015; Van Allen et al., 2015),
including a cohort consisting of only
V600
BRAF mutant melanomas (cohort 3;
Hugo et al., 2015). We found that IPRES
content signatures co-enriched not only
in the same tumors, but also in about a
third of total samples in each of the
four independent transcriptomic datasets
(cohort 1 from this study, 10 IPRES-en-
riched tumors of 28 total tumors; cohort
2, 15 of 42; cohort 3, 11 of 32; cohort 4,
90 of 282; Figure 3A). Considering 126
among 384 total tumors as the back-
ground prevalence for co-enrichment of
IPRES content signatures in metastatic
melanoma, we determined that this
2016). Interestingly, this set of 26 IPRES signatures included IPRES-enriched transcriptomic subset was over-represented
signatures induced by MAPKi treatment of melanoma tumors among the anti-PD-1 non-responding pretreatment tumors
and cell lines (Table S2C). We have shown recently that MAPKi (Fisher p = 0.013, odds ratio = 4.6) and under-represented
treatment of melanoma cells induces transcriptome-wide re- among the responding pretreatment tumors (Fisher p = 0.04,
programming, leading to concurrent phenotype switches (C.S., odds ratio = 0.15) within cohort 1. In contrast, co-enrichment
L.S., W.H., and R.S.L., unpublished data). Notably, MAPKi- of IPRES signatures was neither over- nor under-represented
induced signatures of mesenchymal-invasive transition, angio- among the responding or non-responding pre-anti-CTLA-4 mel-
genesis, and wound-healing signatures were detected in the anoma tumors in cohort 2 (Figure S2B; Van Allen et al., 2015),
residual melanoma tumors from patients on MAPKi therapy, sug- which suggests that mechanisms of innate resistance to anti-
gesting that induction of these signatures may negatively impact PD-1 and anti-CTLA-4 are not necessarily similar.
responsiveness to combinatorial anti-PD-1/L1 therapy. Furthermore, co-enrichment of the IPRES signatures defined a
transcriptomic subset within not only melanoma, but also all
IPRES Signatures Define a Transcriptomic Subset major common human malignancies analyzed (Figure 3B). The
across Cancers IPRES-enriched transcriptomic subset of certain cancers, such
The observations that IPRES content signatures were co-en- as pancreatic adenocarcinoma, made up the majority of tumors.
riched in the same tumors (Figure 2D) and that MAPKi induced Within a side-by-side comparison, only 6 of 69 primary cuta-
these signatures concurrently (Table S2C) implied co-regulated neous melanomas showed co-enrichment of IPRES signatures,
tumor phenotypes that together define a transcriptomic subset. in contrast to 90 of 282 metastatic (The Cancer Genome Atlas
To evaluate whether co-enrichment of IPRES content signatures [TCGA]) melanomas (p = 3.9e-5, odds ratio = 0.2; Figure S3),

consistent with mesenchymal transition and metastasis gene EXPERIMENTAL PROCEDURES
sets among IPRES signatures. Thus, co-enrichment of IPRES
Tumor Specimens and Profiling
signatures defines a distinct transcriptomic program that exists
All tissues in this study were obtained with the approval of institutional review
across distinct cancer histologies. boards and patients’ consents. All patients received either pembrolizumab or
This study highlights the utility of both exome and transcrip- nivolumab as the anti-PD-1 therapy for their metastatic melanoma. 38 mela-
tome sequencing data generated from pretreatment tumor noma specimens (32 pre-treatment tumors, 2 pretreatment tumor-derived cul-
samples for the identification of potential determinants of tures, 3 early on-treatment tumors without response, and 1 early on-treatment
response to anti-PD-1. Although the overall somatic mutational tumor with response) and their patient-matched normal tissues were analyzed
by WES. Among these 38 samples with WES data, 28 with sufficient RNA qual-
loads of anti-PD-1-responsive melanoma tumors were not
ity were also analyzed by RNA sequencing (RNA-seq). This set included
significantly higher than those of non-responsive tumors, higher
another RNA-seq dataset derived from a second-site, pre-treatment tumor
mutational loads associated significantly with better survival af- biopsy from patient #27. However, this second-site, pre-treatment tumor-
ter anti-PD-1 therapy. This finding is still consistent with the derived WES dataset was excluded in our aggregate mutation analysis to
notion that neoepitopes derived from somatic non-synonymous avoid double counting two tumor exomes from the same patient (Table S1A).
mutations are critical for deriving clinical benefits from anti-PD-1 38 tumor specimens and their respective normal tissues were subjected
therapy in melanoma. However, objective tumor responses, to WES (Table S1A). WES was performed using pair-end sequencing with
read length of 2 3 100 bps based on the Illumina HiSeq2000 platform. RNA
although strongly associated with survival benefits, did not
from a subset of 28 tumors were pair-end sequenced with read length
appear to be driven overwhelmingly by the overall somatic of 2 3 100 bps (Illumina HiSeq2000). We included two tumors from
mutational loads. That is to say, a relatively low mutational patient #27 for transcriptomic analyses, but not for mutation and neoepitope
load did not preclude a tumor response. This is consistent analyses, since the tumors may not share the same transcriptomic profile,
with findings from gastrointestinal cancers where low mutational but they essentially contain the same set of non-synonymous somatic
loads did not preclude tumor infiltration by mutation-reactive, mutations.
class I- and II-restricted T cells (Tran et al., 2015). Thus, overall
Whole Exome Sequencing
somatic or predicted neoepitope loads of pretreatment mela-
We called single nucleotide variant (SNV) and INDEL as reported (Shi et al.,
noma tumors are not enough to predict response patterns to 2014) using a stand-alone version of Oncotator (Ramos et al., 2015). Copy
anti-PD-1 therapy. numbers were called using the intersection of the copy-number calls derived
In our cohort, responsive tumors were significantly enriched from Sequenza (Favero et al., 2015) and VarScan2 (Koboldt et al., 2012).
for (likely) loss-of-function mutations in BRCA2. As one would Tumor purities and ploidies (Table S1B) were calculated based on the calls
predict from the known function of BRCA2 in DNA repair, of Sequenza using WES data with default parameters. The impact of BRCA2
nsSNVs was visualized using the domain information in the INTERPRO protein
BRCA2-mutated melanomas harbored higher mutational loads
domain database (Mitchell et al., 2015).
than BRCA2-wild-type melanomas. Although it is conceivable
that defective BRCA2-DNA repair results in specific mutational HLA Types and Neoepitopes
motifs (rather than or in addition to the general increase in The four-digit HLA class 1 and 2 types of each patient were called using
mutational load) that enhance responsiveness, it is also ATHLATES (Liu et al., 2013) using the WES-sequencing reads from the normal
possible that cellular stress resulting from defective DNA repair tissue. To ensure concordance, we manually compared ATHLATES’ calls of
the normal versus tumor samples and ascertained that there was at least no
could lead to increased cell death and anti-tumor immunity.
two-digit HLA typing discrepancy between any normal tumor pair. For each
Moreover, these data support the notion that tumor cell pheno- non-synonymous coding mutation from a tumor, we predicted its impact
typic plasticity (i.e., mesenchymal transition) and the resultant on the patient’s HLA class I and II binding using the standalone version of
impacts on the microenvironment (e.g., ECM remodeling, cell the programs NetMHCpan v2.8 (Hoof et al., 2009; Nielsen et al., 2007) and
adhesion, and angiogenesis features of immune suppressive NetMHCIIpan v3.0 (Karosiene et al., 2013), respectively. Specifically, for
wound healing) are critical barriers to anti-PD-1 responses. HLA class I binding prediction using netMHCpan v2.8, we tested all
The limited number of patients in our melanoma cohort posed 9-11-mer peptides containing the mutated amino acids for binding to the
patient’s HLA-A, -B, and -C. A peptide was defined as a neoepitope based
certain challenges to our analysis. For example, we relaxed the
on two criteria: (1) predicted binding affinity % 500nM and (2) rank
statistical stringency in single gene-based differential expres- percentage % 2% (default cutoff). For HLA class II binding prediction using
sion analysis (bypassing multiple hypothesis correction) to netMHCIIpan v3.0, we tested the 9-19-mers containing the mutated amino
derive enough genes for GO enrichment analysis. However, acids for binding to the patient-specific, ATHLATES-predicted DPA-DPB,
converging findings from alternative analysis (i.e., GSVA) of DQA-DQB, and DRA-DRB allele pairs. We also applied the same predicted
the transcriptome data helped to mitigate potential caveats. binding affinity and rank percentage cutoff as we did for HLA class I to nomi-
nate the HLA class-II-binding neoepitopes. Expressed non-synonymous mu-
Finally, in separate work, we found that mutation-targeted ther-
tations and neoepitopes were defined based on corresponding genes with
apy (i.e., MAPKi) induces tumor cell-autonomous changes (e.g., normalized expression levels R 1 (in fragments per kilobase of transcripts
mesenchymal transition; C.S., L.S., W.H., and R.S.L., unpub- per million mapped reads [FPKM]). Statistical differences of nsSNV, HLA
lished data) and upregulates anti-PD-1 resistance-associated class I and II neoepitopes, WES coverages, and tumor purities were computed
processes in residual tumors that have regressed in response using two-sided Mann-Whitney test.
to MAPKi treatment. Thus, while our findings in this study
necessitate confirmation in independent tissue cohorts, the Mutation Recurrence
To estimate the statistical significance of the recurrence of gene mutations in
identification of transcriptomic features associated with anti-
the responding or non-responding tumors, we used an independent batch of
PD-1 resistance suggests that mitigation of IPRES-related bio- 469 melanomas’ whole-exome sequence datasets (Hodis et al., 2012; Cancer
logical processes may enhance response rates to anti-PD-1 Genome Atlas Network, 2015) to estimate each gene’s background mutation
(and anti-PD-1 plus MAPKi) therapy. frequency. Significance was computed by Fisher’s exact test followed by FDR

adjustment for multiple hypothesis testing. We listed genes that fulfilled the Figure 2D, top). The IPRES (enrichment) score was defined as the average
criteria: (1) recurrence in at least 25% of the responder/non-responder, Z score across all gene sets in the IPRES signature, and we applied an average
(2) occurrence of at most once in the opposite group, and (3) Fisher’s exact Z score of 0.35 as the cutoff for IPRES signature enrichment in a tumor sample.
test FDR-adjusted p value % 0.05. These genes were illustrated in Figure 1A, This resulted in IPRES co-enrichment in nine non-responding tumors and
and all genes that fulfilled the first and second criteria and tested for multiple one responding tumor in our anti-PD-1 cohort (this cutoff was chosen
hypotheses were listed in Table S1E. The association between BRCA2 because it provided the largest average Z score separation between the
nsSNVs and overall nsSNV counts were tested using two-sided Mann Whitney samples with and without IPRES co-enrichment). Since the IPRES score
test and validated in independent WES datasets (Hodis et al., 2012; Cancer was not comparable across analyses in Figures 3A, 3B, and S3 (since the
Genome Atlas Network, 2015). IPRES score is Z score-base), we used the 90th highest IPRES score in the
TCGA metastatic melanoma cohort as the IPRES score cutoff (since there
RNA-Seq and Gene Set Enrichment were 90 of 282 tumors showing IPRES co-enrichment in this TCGA metastatic
Paired-end transcriptome reads were mapped to the UCSC hg19 reference cohort; Figure 3A) for analyses performed to yield Figures 3B and S3. This
genome using Tophat2 (Kim et al., 2013). Normalized expression levels of allowed for a non-parametric comparison across multiple TCGA datasets at
genes were expressed in FPKM values as generated by cuffquant and cuff- the IPRES co-enrichment level established in our anti-PD-1 cohort.
norm (Trapnell et al., 2012). The program was run with the option ‘‘–frag-bias-
correct’’ and ‘‘–multi-read-correct’’ to improve sensitivity (Roberts et al., Source Data
2011). A gene was defined as differentially expressed between the re- Analysis of differential non-synonymous mutational hits in responders versus
sponding and non-responding tumor groups when its median expression non-responders to ipilimumab was based on the mutation calls as reported
differed by at least 2-fold between the groups with a nominal two-sided (Van Allen et al., 2015). We curated published CD8 T cell exhaustion genes
Mann-Whitney p value % 0.1 (Table S2A). Applying multiple hypothesis (Wherry, 2011) to minimize those likely to be expressed by melanoma cells
correction of FDR p % 0.25 only yielded three differentially expressed genes: by excluding genes whose maximum log2 FPKM was 1 in an in-house mela-
ALDH1L2 and MFAP2 in the non-responding and CDH1 (E-cadherin) in the noma cell-line-derived RNA-seq database (n = 26 cell lines). This resulted in
responding group. As such, the genes meeting the uncorrected, nominal the inclusion of genes for surface receptors PDCD1 (PD-1), LAG3, HAVCR2
Mann-Whitney p value % 0.1 that were expressed higher either in the re- (Tim-3), CD160, and CD244 as well as transcription factors EOMES, PRDM1
sponding or non-responding group were separately analyzed for GO term (Blimp-1), and TBX21 (T-bet). We assessed co-enrichment of IPRES content
enrichments using the online functional annotation tools DAVID (Huang signatures in the (1) anti-CTLA-4 pretreatment cohort (Van Allen et al., 2015),
et al., 2009a). Enriched GO terms were selected from the GO biological (2) MAPKi pretreatment cohort (Hugo et al., 2015; C.S., L.S., W.H., and
process terms in DAVID’s fat database (Huang et al., 2009b). GO terms R.S.L., unpublished data), (3) TCGA melanoma (metastatic and primary
which were highly overlapping, as defined by functional clustering in DAVID’s subsets separately analyzed; Cancer Genome Atlas Network, 2015), (4)
website, were represented by the terms with the best FDR-adjusted TCGA pancreatic ductal adenocarcinoma (TCGA, provisional, see https://
p values. tcga-data.nci.nih.gov/tcga/tcgaHome2.jsp), (5) TCGA lung adenocarcinoma
To calculate single-sample gene set enrichment, we used the GSVA program (Cancer Genome Atlas Research Network, 2014), (6) TCGA colorectal adeno-
(Hänzelmann et al., 2013) to derive the absolute enrichment scores of previ- carcinoma (Cancer Genome Atlas Network, 2012), and (7) TCGA kidney clear
ously experimentally validated gene signatures as follows: (1) the C2 CGP cell carcinoma (Cancer Genome Atlas Research Network, 2013).
(chemical and genetic perturbation sets), (2) the C6 and C7 subset of the
Molecular Signature Database version 4.0 (Subramanian et al., 2005), (3) self- ACCESSION NUMBERS
curated MAPKi-induced gene signatures using cell lines and patient-derived
tumors (C.S., L.S., W.H., and R.S.L., unpublished data), (4) post-operation The accession number for the transcriptome data is GEO: GSE78220.
wound signature (Inkeles et al., 2015), and (5) melanoma invasive/proliferative
signatures (Hoek et al., 2008). To derive the GSVA score of each signature in
each tumor sample, we computed from raw RNA-seq read counts by HTSEQ SUPPLEMENTAL INFORMATION
COUNT program and then normalized them to log2 counts per million (CPM)
values using EdgeR (McCarthy et al., 2012). We removed batch effects using Supplemental Information includes three figures and two tables and can be
the edgeR function RemoveBatchEffect when we combined RNA-seq data found with this article online at http://dx.doi.org/10.1016/j.cell.2016.02.065.
from multiple experiments (Figure 3A). The normalized log2 CPM values were
then passed on as input for GSVA in the RNA-seq mode. Differentially enriched AUTHOR CONTRIBUTION
core gene sets between the responding and non-responding tumor groups
were defined by GSVA score differences of R 10% and FDR-corrected, two- W.H., J.M.Z., L.S., C.S., A.R., and R.S.L. generated, analyzed, and interpreted
sided Welch t test p value % 0.25 (we used t test because the GSVA scores the data. S.H.-L., B.H.M., B.C., G.C., E.S., M.C.K., J.A.S., D.B.J., A.R., and
were normally distributed around zero). Two gene sets, INGRAM_SHH_ R.S.L. evaluated patients and provided tissue reagents. B.B.-M., J.P., S.L.,
TARGETS_DN and WONG_ENDMETRIUM_CANCER_DN, were not included and X.K. processed tissues for analysis. All authors contributed to manuscript
in the core set because they did not specifically point to a cellular process preparation. W.H. and R.S.L. developed the concepts, supervised the project,
and/or relate to the other six gene sets in the core set (Table S2B, top eight). and wrote the paper.
We also collected gene sets that met the GSVA score differences of R 10%
and nominal Welch t test p value % 0.1 (Table S2B) and included those which ACKNOWLEDGMENTS
were concordantly enriched and functionally related to the core gene sets to
make up the full list of IPRES signatures (Figure 2D). This work has been funded by the NIH (R35 CA197633 to A.R., P01 CA168585
To compare co-enrichment of IPRES signatures across multiple melanoma to A.R. and R.S.L., 1R01CA176111 to R.S.L., and K12CA0906525 to D.B.J.),
cohorts, we combined and batch-corrected the log2 CPM values of four mel- the Ressler Family Foundation (to R.S.L. and A.R.), the Grimaldi Family Fund
anoma transcriptome cohorts: (1) our current pre-anti-PD-1 tumors (n = 28), (to A.R. and R.S.L.), the Burroughs Wellcome Fund (to R.S.L.), the Ian Cope-
(2) pre-anti-CTLA-4 tumors (n = 42), (3) pre-MAPKi tumors (n = 32), and land Melanoma Fund (to R.S.L.), the Wade F.B. Thompson/Cancer Research
(4) the metastatic subset of TCGA melanoma (n = 282). We row-normalized Institute CLIP Grant (to R.S.L.), the Steven C. Gordon Family Foundation (to
the GSVA scores of each gene set in the IPRES signature across the samples R.S.L.), the American Skin Association (to W.H.), the Dr. Robert Vigen Memo-
from the four cohorts. For this comparative study, we excluded the gene sets rial Fund (to A.R.), the Garcia-Corsini Family Fund (to A.R.), the ASCO Conquer
JAEGER_METASTASIS_UP, YE_METASTATIC_LIVER_CANCER, KARAKAS_ Cancer Career Development Award (to D.B.J.), and the American Cancer So-
TGFB1_SIGNALING, and JEON_SMAD6_ TARGETS_ DN from the IPRES set ciety Research Professorship (to J.A.S.). Informed consents were obtained
because they showed weaker co-enrichment with rest of the gene sets (see from patients for the research performed in this study.

Received: December 25, 2015 method including all three human MHC class II isotypes, HLA-DR, HLA-DP
Revised: January 30, 2016 and HLA-DQ. Immunogenetics 65, 711–724.
Accepted: February 29, 2016 Kim, D., Pertea, G., Trapnell, C., Pimentel, H., Kelley, R., and Salzberg, S.L.
Published: March 17, 2016 (2013). TopHat2: accurate alignment of transcriptomes in the presence of in-
sertions, deletions and gene fusions. Genome Biol. 14, R36.
REFERENCES Koboldt, D.C., Zhang, Q., Larson, D.E., Shen, D., McLellan, M.D., Lin, L., Miller,
C.A., Mardis, E.R., Ding, L., and Wilson, R.K. (2012). VarScan 2: somatic mu-
Cancer Genome Atlas Network (2012). Comprehensive molecular character- tation and copy number alteration discovery in cancer by exome sequencing.
ization of human colon and rectal cancer. Nature 487, 330–337. Genome Res. 22, 568–576.
Cancer Genome Atlas Network (2015). Genomic Classification of Cutaneous Le, D.T., Uram, J.N., Wang, H., Bartlett, B.R., Kemberling, H., Eyring, A.D.,
Melanoma. Cell 161, 1681–1696. Skora, A.D., Luber, B.S., Azad, N.S., Laheru, D., et al. (2015). PD-1 Blockade
Cancer Genome Atlas Research Network (2013). Comprehensive molecular in Tumors with Mismatch-Repair Deficiency. N. Engl. J. Med. 372, 2509–2520.
characterization of clear cell renal cell carcinoma. Nature 499, 43–49. Liu, C., Yang, X., Duffy, B., Mohanakumar, T., Mitra, R.D., Zody, M.C., and
Cancer Genome Atlas Research Network (2014). Comprehensive molecular Pfeifer, J.D. (2013). ATHLATES: accurate typing of human leukocyte antigen
profiling of lung adenocarcinoma. Nature 511, 543–550. through exome sequencing. Nucleic Acids Res. 41, e142.
Chen, L., Gibbons, D.L., Goswami, S., Cortez, M.A., Ahn, Y.H., Byers, L.A., Mak, M.P., Tong, P., Diao, L., Cardnell, R.J., Gibbons, D.L., William, W.N.,
Zhang, X., Yi, X., Dwyer, D., Lin, W., et al. (2014). Metastasis is regulated via Skoulidis, F., Parra, E.R., Rodriguez-Canales, J., Wistuba, I.I., et al. (2016).
microRNA-200/ZEB1 axis control of tumour cell PD-L1 expression and intra- A Patient-Derived, Pan-Cancer EMT Signature Identifies Global Molecular
tumoral immunosuppression. Nat. Commun. 5, 5241. Alterations and Immune Target Enrichment Following Epithelial-to-Mesen-
Chen, L., Heymach, J.V., Qin, F.X., and Gibbons, D.L. (2015). The mutually reg- chymal Transition. Clin. Cancer Res. 22, 609–620.
ulatory loop of epithelial-mesenchymal transition and immunosuppression in McCarthy, D.J., Chen, Y., and Smyth, G.K. (2012). Differential expression anal-
cancer progression. OncoImmunology 4, e1002731. ysis of multifactor RNA-Seq experiments with respect to biological variation.
Favero, F., Joshi, T., Marquard, A.M., Birkbak, N.J., Krzystanek, M., Li, Q., Nucleic Acids Res. 40, 4288–4297.
Szallasi, Z., and Eklund, A.C. (2015). Sequenza: allele-specific copy num- Mitchell, A., Chang, H.Y., Daugherty, L., Fraser, M., Hunter, S., Lopez, R.,
ber and mutation profiles from tumor sequencing data. Ann. Oncol. 26, McAnulla, C., McMenamin, C., Nuka, G., Pesseat, S., et al. (2015). The InterPro
64–70. protein families database: the classification resource after 15 years. Nucleic
Hamid, O., Robert, C., Daud, A., Hodi, F.S., Hwu, W.J., Kefford, R., Wolchok, Acids Res. 43, D213–D221.
J.D., Hersey, P., Joseph, R.W., Weber, J.S., et al. (2013). Safety and tumor re- Motz, G.T., and Coukos, G. (2011). The parallel lives of angiogenesis and
sponses with lambrolizumab (anti-PD-1) in melanoma. N. Engl. J. Med. 369, immunosuppression: cancer and other tales. Nat. Rev. Immunol. 11, 702–711.
134–144.
Nielsen, M., Lundegaard, C., Blicher, T., Lamberth, K., Harndahl, M., Justesen,
Hänzelmann, S., Castelo, R., and Guinney, J. (2013). GSVA: gene set variation S., Røder, G., Peters, B., Sette, A., Lund, O., and Buus, S. (2007). NetMHCpan,
analysis for microarray and RNA-seq data. BMC Bioinformatics 14, 7. a method for quantitative predictions of peptide binding to any HLA-A and -B
Hodis, E., Watson, I.R., Kryukov, G.V., Arold, S.T., Imielinski, M., Theurillat, locus protein of known sequence. PLoS ONE 2, e796.
J.P., Nickerson, E., Auclair, D., Li, L., Place, C., et al. (2012). A landscape of Peng, W., Chen, J.Q., Liu, C., Malu, S., Creasy, C., Tetzlaff, M.T., Xu, C.,
driver mutations in melanoma. Cell 150, 251–263. McKenzie, J.A., Zhang, C., Liang, X., et al. (2016). Loss of PTEN promotes
Hoek, K.S., Eichhoff, O.M., Schlegel, N.C., Döbbeling, U., Kobert, N., Scha- resistance to T cell-mediated immunotherapy. Cancer Discov. 6, 202–216.
erer, L., Hemmi, S., and Dummer, R. (2008). In vivo switching of human mela- Ramos, A.H., Lichtenstein, L., Gupta, M., Lawrence, M.S., Pugh, T.J., Sak-
noma cells between proliferative and invasive states. Cancer Res. 68, sena, G., Meyerson, M., and Getz, G. (2015). Oncotator: cancer variant anno-
650–656. tation tool. Hum. Mutat. 36, E2423–E2429.
Holloman, W.K. (2011). Unraveling the mechanism of BRCA2 in homologous Rizvi, N.A., Hellmann, M.D., Snyder, A., Kvistborg, P., Makarov, V., Havel, J.J.,
recombination. Nat. Struct. Mol. Biol. 18, 748–754. Lee, W., Yuan, J., Wong, P., Ho, T.S., et al. (2015). Cancer immunology. Muta-
Hoof, I., Peters, B., Sidney, J., Pedersen, L.E., Sette, A., Lund, O., Buus, S., tional landscape determines sensitivity to PD-1 blockade in non-small cell lung
and Nielsen, M. (2009). NetMHCpan, a method for MHC class I binding predic- cancer. Science 348, 124–128.
tion beyond humans. Immunogenetics 61, 1–13. Roberts, A., Trapnell, C., Donaghey, J., Rinn, J.L., and Pachter, L. (2011).
Hoos, A., Wolchok, J.D., Humphrey, R.W., and Hodi, F.S. (2015). CCR 20th Improving RNA-Seq expression estimates by correcting for fragment bias.
Anniversary Commentary: Immune-Related Response Criteria–Capturing Genome Biol. 12, R22.
Clinical Activity in Immuno-Oncology. Clin. Cancer Res. 21, 4989–4991. Schäfer, M., and Werner, S. (2008). Cancer as an overhealing wound: an old
Huang, W., Sherman, B.T., and Lempicki, R.A. (2009a). Systematic and inte- hypothesis revisited. Nat. Rev. Mol. Cell Biol. 9, 628–638.
grative analysis of large gene lists using DAVID bioinformatics resources. Sharma, P., and Allison, J.P. (2015). Immune checkpoint targeting in cancer
Nat. Protoc. 4, 44–57. therapy: toward combination strategies with curative potential. Cell 161,
Huang, W., Sherman, B.T., and Lempicki, R.A. (2009b). Bioinformatics enrich- 205–214.
ment tools: paths toward the comprehensive functional analysis of large gene Shi, H., Hugo, W., Kong, X., Hong, A., Koya, R.C., Moriceau, G., Chodon, T.,
lists. Nucleic Acids Res. 37, 1–13. Guo, R., Johnson, D.B., Dahlman, K.B., et al. (2014). Acquired resistance
Hugo, W., Shi, H., Sun, L., Piva, M., Song, C., Kong, X., Moriceau, G., Hong, A., and clonal evolution in melanoma during BRAF inhibitor therapy. Cancer Dis-
Dahlman, K.B., Johnson, D.B., et al. (2015). Non-genomic and Immune Evolu- cov. 4, 80–93.
tion of Melanoma Acquiring MAPKi Resistance. Cell 162, 1271–1285. Snyder, A., Makarov, V., Merghoub, T., Yuan, J., Zaretsky, J.M., Desrichard,
Inkeles, M.S., Scumpia, P.O., Swindell, W.R., Lopez, D., Teles, R.M., Graeber, A., Walsh, L.A., Postow, M.A., Wong, P., Ho, T.S., et al. (2014). Genetic basis
T.G., Meller, S., Homey, B., Elder, J.T., Gilliet, M., et al. (2015). Comparison of for clinical response to CTLA-4 blockade in melanoma. N. Engl. J. Med. 371,
molecular signatures from multiple skin diseases identifies mechanisms of im- 2189–2199.
munopathogenesis. J. Invest. Dermatol. 135, 151–159. Subramanian, A., Tamayo, P., Mootha, V.K., Mukherjee, S., Ebert, B.L., Gil-
Karosiene, E., Rasmussen, M., Blicher, T., Lund, O., Buus, S., and Nielsen, M. lette, M.A., Paulovich, A., Pomeroy, S.L., Golub, T.R., Lander, E.S., and Me-
(2013). NetMHCIIpan-3.0, a common pan-specific MHC class II prediction sirov, J.P. (2005). Gene set enrichment analysis: a knowledge-based approach

for interpreting genome-wide expression profiles. Proc. Natl. Acad. Sci. USA blockade induces responses by inhibiting adaptive immune resistance. Nature
102, 15545–15550. 515, 568–571.
Topalian, S.L., Hodi, F.S., Brahmer, J.R., Gettinger, S.N., Smith, D.C., Van Allen, E.M., Miao, D., Schilling, B., Shukla, S.A., Blank, C., Zimmer, L.,
McDermott, D.F., Powderly, J.D., Carvajal, R.D., Sosman, J.A., Atkins, M.B., Sucker, A., Hillen, U., Foppen, M.H., Goldinger, S.M., et al. (2015). Genomic
et al. (2012). Safety, activity, and immune correlates of anti-PD-1 antibody in correlates of response to CTLA-4 blockade in metastatic melanoma. Science
cancer. N. Engl. J. Med. 366, 2443–2454. 350, 207–211.
Tran, E., Ahmadzadeh, M., Lu, Y.C., Gros, A., Turcotte, S., Robbins, P.F., Gart-
Voron, T., Marcheteau, E., Pernot, S., Colussi, O., Tartour, E., Taieb, J., and
ner, J.J., Zheng, Z., Li, Y.F., Ray, S., et al. (2015). Immunogenicity of somatic
Terme, M. (2014). Control of the immune response by pro-angiogenic factors.
mutations in human gastrointestinal cancers. Science 350, 1387–1390.
Front. Oncol. 4, 70.
Trapnell, C., Roberts, A., Goff, L., Pertea, G., Kim, D., Kelley, D.R., Pimentel,
H., Salzberg, S.L., Rinn, J.L., and Pachter, L. (2012). Differential gene and tran- Wherry, E.J. (2011). T cell exhaustion. Nat. Immunol. 12, 492–499.
script expression analysis of RNA-seq experiments with TopHat and Cufflinks. Wolchok, J.D., Hoos, A., O’Day, S., Weber, J.S., Hamid, O., Lebbé, C., Maio,
Nat. Protoc. 7, 562–578. M., Binder, M., Bohnsack, O., Nichol, G., et al. (2009). Guidelines for the eval-
Tumeh, P.C., Harview, C.L., Yearley, J.H., Shintaku, I.P., Taylor, E.J., Robert, uation of immune therapy activity in solid tumors: immune-related response
L., Chmielowski, B., Spasic, M., Henry, G., Ciobanu, V., et al. (2014). PD-1 criteria. Clin. Cancer Res. 15, 7412–7420.

Article
Progressive Loss of Function

in a Limb Enhancer during Snake Evolution
Evgeny Z. Kvon,1 Olga K. Kamneva,2 Uirá S. Melo,1 Iros Barozzi,1 Marco Osterwalder,1 Brandon J. Mannion,1
Virginie Tissières,3 Catherine S. Pickle,1 Ingrid Plajzer-Frick,1 Elizabeth A. Lee,1 Momoe Kato,1 Tyler H. Garvin,1
Jennifer A. Akiyama,1 Veena Afzal,1 Javier Lopez-Rios,3 Edward M. Rubin,1,4 Diane E. Dickel,1 Len A. Pennacchio,1,4,*
and Axel Visel1,4,5,6,*
1MS 84-171, Lawrence Berkeley National Laboratory, Berkeley, CA 94720, USA
2Department of Biology, Stanford University, Stanford, CA 94305, USA
3Department of Biomedicine, University of Basel, 4058 Basel, Switzerland
4U.S. Department of Energy Joint Genome Institute, Walnut Creek, CA 94598, USA
5School of Natural Sciences, University of California, Merced, CA 95343, USA
6Lead Contact
*Correspondence: lapennacchio@lbl.gov (L.A.P.), avisel@lbl.gov (A.V.)

SUMMARY the subject of considerable debate (Hoekstra, 2012; Wittkopp

and Kalay, 2011; Wray, 2007).
The evolution of body shape is thought to be tightly In the present study, we utilized a series of recently sequenced
coupled to changes in regulatory sequences, but snake genomes to study the molecular and functional evolution of
specific molecular events associated with major mor- a critical limb enhancer in snakes and examine its possible role in
phological transitions in vertebrates have remained limb loss. Our analysis focuses on one of the best-studied verte-
elusive. We identified snake-specific sequence brate enhancers, the Zone of Polarizing Activity [ZPA] Regulatory
Sequence (ZRS, also known as MFCS1) (Lettice et al., 2003,
changes within an otherwise highly conserved long-
2008, 2012, 2014; Sagai et al., 2004, 2005; Zeller and Zuniga,
range limb enhancer of Sonic hedgehog (Shh). Trans- 2007). The ZRS is a limb-specific enhancer of the Sonic hedge-
genic mouse reporter assays revealed that the in vivo hog (Shh) gene that is located at the extreme distance of nearly
activity pattern of the enhancer is conserved across a one million base pairs from its target promoter. During limb devel-
wide range of vertebrates, including fish, but not in opment, the enhancer is active in the posterior limb bud mesen-
snakes. Genomic substitution of the mouse enhancer chyme (Figure 1A), where its activity is critically required for
with its human or fish ortholog results in normal limb normal limb development in mouse (Sagai et al., 2005). Single-
development. In contrast, replacement with snake nucleotide mutations within the ZRS cause limb malformations,
orthologs caused severe limb reduction. Synthetic such as preaxial polydactyly, in multiple vertebrate species
restoration of a single transcription factor binding including humans (Hill and Lettice, 2013; Lettice et al., 2003,
site lost in the snake lineage reinstated full in vivo 2008; VanderMeer and Ahituv, 2011). Surprisingly, we observed
that the sequence of this limb enhancer is conserved throughout
function to the snake enhancer. Our results demon-
nearly all examined species in the snake lineage. In basal snakes,
strate changes in a regulatory sequence associated which retain vestigial limbs, it is highly conserved, whereas it un-
with a major body plan transition and highlight the derwent a rapid increase in substitution rate in advanced snakes,
role of enhancers in morphological evolution. in which all skeletal limb structures have disappeared. Consistent
with this, we provide evidence that the snake enhancer progres-
sively lost its in vivo function as the body plan evolved from basal
INTRODUCTION to advanced snakes. Finally, we identify a specific subset of
nucleotide changes within the enhancer that contribute to its
Distant-acting transcriptional enhancers are a major class of functional degeneration in snakes and show in a mouse model
tissue-specific regulatory DNA sequences that has been impli- that synthetic reintroduction of just one degraded transcription
cated in morphological evolution in vertebrates (Chan et al., factor binding site is sufficient to recreate the ancestral function
2010; Cooper et al., 2014; Cretekos et al., 2008; Guenther and to rescue normal limb formation in vivo.
et al., 2014; Guerreiro et al., 2013; Indjeian et al., 2016; Jones
et al., 2012; Lopez-Rios et al., 2014; McLean et al., 2011; Prabha-
kar et al., 2008). Sequence changes in non-coding regulatory RESULTS
DNA are hypothesized to be a main driver of changes in body
shape (Britten and Davidson, 1969; Carroll, 2008; King and Wil- A Critical Limb Enhancer Is Evolutionarily Conserved
son, 1975; Wray, 2007), but many aspects of this complex inter- but Highly Diverged in Snakes
play between molecular changes in regulatory sequences and To explore the potential role of the ZRS limb enhancer in snake
morphological adaptations across the vertebrate tree remain evolution, we examined the draft genomes of six snake species
Cell 167, 633–642, October 20, 2016 ª 2016 Elsevier Inc. 633
A B * **** ** * ****** ***** ** ****** **** ** * * * *** ** ** *** *
human T T GT CC T GGT T T A T GT CCC T T T T GGC A A A C T T A C A T A A A AGT GA CC T T GT A C T GT A T T T T A T GA CC AGA T GA C T T
mouse T T GT CC T GGT T T A T GT CGC T T T T GGC A A A C T T A C A T A A A AGT GA CC T T GT A C T GT A T T T T A T GA CC AGA T GA C - T
cow T T GT CC T GGT T T A T GT CCC T T T T GGC A A A C T T A C A T A A A AGT GA CC T T GT A C T GT A T T T T A T GA CC AGA T GA C T T
dolphin T T GT CC T GGT T T A T GT CCC T T T T GGC A A A C T T A C A T A A A AGT GA CC T T GT A C T GT A T T T T A T GA CC AGA T GA C T T
horse
h T T GT CC T GGT T T A T GT CCC T T T T GGC A A A C T T A C A T A A A AGT GA CC T T GT A C T GT A T T T T A T GA CC AGA T GA C T T
megabat T T GT CC T GGT T T A T GT CCC T T T T GGC A A A C T T A C A T A A A AGT GA CCC T GT A C T GT A T T T T A T GA CC AGA T GA C T T
sloth
l th T T GT CC T GGT T T A T GT CCC T T T T GGC A A A C T T A C A T A A A AGT GA CC T T GT A C T GT A T T T T A T GA CC AGA T GA C T T
platypus T T GT CC T GGT T T A T GT CCC T T T T GGC A A A C T T A C A T A A A AGT GA CCC T GT A C T GT A T T T T A T GA CC AGA T GA C T T
hi k
chicken T T GT CC T GGT T T A T GT CCC T T T T GGC A A A C T T A C A T A A A AGT GA CCC T GT A C T GT A T T T T A T GA CC AGA T GA C T T
lizard T T GT CC T GGT T T A T GT CCC T T T T GGC A A A C A T A C A T A A A AGT GA CC A T GT A C T GT A T T T T A T GA CC AGA T GA C T T

basal python
h T T GT CC T GGT T T A T GT CCC T T T T GGC A A A C T T A C A T A A A AGT GA C T GT GT A C T GT A T T T T A T GA CC AGA T GA C T T
snakes boa T T GT CC T GGT T T A T GT CCC T T T T GGC A A A C T T A C A T A A A AGT GA C T GT GT A C T GT A T T T T A T GA CC AGA T GA C T T

viper
ip T T GT CC T A A T T GA T GT T CC T T T T GGC A A A C T T A C A T A A A AGT GA C - - T GC A T T GC A T T T T GT GA T C A A A T GA C T T
advanced rattlesnake T T GT CC T A A T T T A T GT T CC T T T T GGC A A A C T T A - A T A A A AGT GA C - - T GC A T T GT A T T T T GT GA T C AGA T GA C T T
cobra
snakes cornsnake
A T C T CC T GGT T T A T GT T CC T T T T GCC A A A C T T A T A T A A A AGT GA C - - T GC A C T A T A - T T T A T GA T C AGA T GAGT T
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
E11.5 coelacanth T T GT CC T GGT T T A T GT CCC T T T T GGC A A A C T T A C A T A A A AGT GA CCC T GT A C T GT A T T T T A T GA CC AGA T GA C T T

e. shark T T GT CC T GGT T T GC A T CCC T T T T GGC A A A C T T A C A T A A A A C T GA CC A T GT A C T GT A T T T T A CGA CC AGA T GA C T T
C
coelacanth
Fast
Fas
s lizard
corn snake
ZRS evolutionary rate
cobra advanced snakes

viper
rattlesnake
boa
basal snakes
python
chicken
platypus limbless
sloth
Slow megabat
horse
dolphin vestigial hindlimbs
cow
mouse 0.1 subs. /site
human
Figure 1. Evolution of a Limb Enhancer across the Vertebrate Tree

(A) Human ZRS enhancer activity in a mid-gestation (E11.5) mouse embryo. Staining in structures other than limb was not reproducible in additional transgenic
embryos and due to ectopic effects.
(B) Comparison of the core ZRS region across 18 different vertebrate species including two basal (blue) and four advanced (purple) snakes. See Figure S1 for full
alignment.
(C) Phylogeny of vertebrate species used in the study (based on UCSC [https://genome.ucsc.edu/cgi-bin/hgGateway] and Hsiang et al., 2015; Pyron et al., 2013).
Branch length indicates absolute ZRS substitution rate, colors indicate relative ZRS evolutionary rate compared to other embryonic enhancers (see Figure S2 and
Method Details). The schematic illustrations of snake skeletons were drawn using images from Romanes (1892), http://www.zoochat.com/, and http://www.
skullcleaning.com/ as templates.
See also Figures S1 and S2.
including the Burmese python (Python molurus bivittatus) (Cas- snakes displays a substantially increased number of substitu-
toe et al., 2013), boa constrictor (Boa constrictor constrictor), tions compared to other enhancers (p = 0.012, permutation
king cobra (Ophiophagus hannah) (Vonk et al., 2013), speckled test; Figures 1B, 1C, and S2; Table S4). This fast evolutionary
rattlesnake (Crotalus mitchellii pyrrhus), viper (Vipera berus rate clearly distinguishes the ZRS from other limb enhancers,
berus), and corn snake (Pantherophis guttatus) (Ullate-Agote which do not show such an increase in substitutions (Figure S2)
et al., 2014). These species represent different morphological (Infante et al., 2015). Thus, while nearly all snake species exam-
stages within the evolutionary history of snakes (Apesteguı́a ined have a ZRS enhancer, a loss of evolutionary constraint on
and Zaher, 2006; Martill et al., 2015), from basal snakes (boa this enhancer coincides with the complete loss of limb structures
and python) that retained a vestigial pelvic girdle and rudimen- at the transition from basal to advanced snakes.
tary hindlimbs, to advanced snakes (viper, rattlesnake, king
cobra, and corn snake) that completely lost all skeletal limb Loss of Region-Specific Limb Enhancer Activity in
structures and represent the majority (>85%) of all extant snake Snakes
species (Lawson et al., 2005; Pyron et al., 2013). Nearly all of the To systematically examine whether the sequence changes
snake species studied have a ZRS-orthologous sequence (Fig- observed in different snake ZRS orthologs alter the in vivo func-
ures 1B and S1). However, while the ZRS enhancer of basal tion of the enhancer, we used a transgenic mouse enhancer re-
snakes shares 80% nucleotide identity with the orthologous porter assay (http://enhancer.lbl.gov/) (Kothary et al., 1989; Visel
region from limbed lizards and shows a substitution rate similar et al., 2007). We determined ZRS enhancer activity patterns for
to other vertebrate ZRS orthologs, the ZRS of advanced 16 different species covering a wide range of jawed vertebrates,
634 Cell 167, 633–642, October 20, 2016

including cartilaginous and bony fishes (elephant shark and
coelacanth), four snakes (boa, python, rattlesnake, and cobra),
and ten limbed tetrapods at mid-gestation (embryonic day [E]
11.5), a time point when the mouse ZRS is active (Figures 1A
and 2) (Lettice et al., 2003). The orthologs from nine finned or
limbed vertebrates (coelacanth, lizard, chicken, platypus, sloth,
horse, cow, mouse, and human) displayed reproducible patterns
of activity in the posterior limb bud that were indistinguishable
from the activity of the mouse enhancer (Figure 2), confirming
the deep conservation of its function across vertebrates with
paired appendages (Dahn et al., 2007; Lettice et al., 2003; Sagai
et al., 2004). ZRS orthologs from three species were active in
the ZPA of mouse limb buds but also had activity expanded
anteriorly (dolphin and megabat) or proximally (elephant shark).
In contrast, in four out of five basal and advanced snake species
examined, either the enhancer activity in the ZPA or the enhancer
sequence itself was lost (Figure 2). Among them, the rattlesnake
ZRS displayed an ectopic limb activity pattern that did not
include the ZPA and may be related to an 180-bp insertion
specifically found in the viper and rattlesnake lineage (Figures
2 and S1). Only the ZRS of boa, which diverged from python
63–96 million years ago (Esquerré and Scott Keogh, 2016) and
among the examined snakes is the one showing the lowest
nucleotide substitution rate with respect to that of the lizard,
retained activity in the ZPA. Notably, the ZRS from all advanced
snakes examined (rattlesnake and cobra) completely lost ZPA-
specific activity.
Snake Enhancer Knockin Causes Severe Limb

Truncation in Mice
To assess the extent to which the observed activity changes in
transgenic reporter assays affect vertebrate limb morphology
in vivo, we employed CRISPR/Cas9 genome editing to generate
a series of knockin (KI) mice where the functionally critical 1.3-kb
core region of the ZRS (Figure S3) was replaced with the orthol-
ogous sequences of the same length from other species. We
first replaced the mouse ZRS with the orthologs from human
(73% sequence identity to the mouse ZRS) and coelacanth
(57% sequence identity to the human ZRS), whose last common
ancestor lived approximately 400 million years ago. Both the hu-
man and coelacanth orthologs resulted in Shh expression at the
onset of limb bud formation that was indistinguishable from wild-
type and rescued the formation of fully developed limbs (Figures
3 and S4G–S4J), indicating that despite considerable evolu-
tionary distance between mammals and fish, the enhancers of
mouse, human, and coelacanth are largely functionally inter-
changeable. In contrast, replacing the mouse ZRS with the or-
thologous cobra sequence resulted in a complete loss of Shh
expression and a truncated limb phenotype, affecting both the
fore- and hindlimbs, that is indistinguishable from the phenotype
Figure 2. Comparison of Enhancer Activity across Jawed Vertebrates
Enhancer activities for 16 different vertebrate species in the limb buds of
caused by deletion of the mouse enhancer (Figures 3, S3, and
transgenic E11.5 stage mouse embryos. Numbers of embryos with lacZ ac- S4G) (Sagai et al., 2005). This result confirms that despite recog-
tivity in the limb over the total number of transgenic embryos screened are nizable sequence conservation, the cobra sequence lacks limb
indicated. Some species (marked in italics) were active in the ZPA of the limb enhancer function and is therefore unable to support limb devel-
buds but had additional activity expanded anteriorly (dolphin and megabat) or opment. The less diverged python ZRS resulted in a similar but a
proximally (elephant shark). *The rattlesnake ZRS enhancer drives an ectopic
slightly milder phenotype. While most skeletal forelimb and hind-
reporter activity pattern that does not include the ZPA (arrows point to the ZPA
limb elements distal of the stylopod:zeugopod junction were
area without detectable LacZ activity).
also severely affected, the python ZRS resulted in formation of
Cell 167, 633–642, October 20, 2016 635

A B
C D
Figure 3. Limb Phenotypes of Knockin Mice with ZRS Orthologs from Other Vertebrate Species
(A) CRISPR/Cas9-mediated replacement of the mouse ZRS sequence with an orthologous sequence from cobra. Schematic of the mouse Shh locus is shown at
the top. The ZRS is located in the intron of the Lmbr1 gene (intron-exon structure not shown), 850 kb away from the promoter of Shh. A homologous locus from
king cobra with the cobra ZRS enhancer (cZRS) is indicated in purple. A CRISPR/Cas9-modified ‘‘serpentized’’ mouse Shh locus is shown below. See also
Figures S4A–S4F and Method Details. Gene diagram not to scale.
(B) Gross phenotypes of ZRSWT/D (top) and serpentized ZRScZRS/D (bottom) mice. Scale bars, 10 mm.
(C and D) Limb phenotypes of knockin mice with ZRS orthologs from other vertebrate species.
(C) Phylogeny and approximate divergence estimates (Amemiya et al., 2013; Hsiang et al., 2015; Wright et al., 2015) are shown on the left. Schematic mouse Shh
loci with the ZRS replaced by orthologs from human (hZRS), python (pZRS), cobra (cZRS), and coelacanth fish (fZRS) are shown.
(D) Comparative Shh mRNA in situ hybridization analysis in knockin mouse embryos during forelimb bud development (first column). Per knockin line, the
Shh transcript distribution was assessed in at least three independent mouse embryos. See Figure S4G for hindlimb bud analysis of Shh expression.
Corresponding whole-mount E14.5 knockin mouse embryos (second column) and skeletal preparations at E18.5 (third and fourth columns) are shown; s,
scapula; h, humerus; r, radius; u, ulna; fe, femur; fi, fibula; t, tibia; a, autopod. The genotypes of the embryos are ZRSWT/D (mouse), ZRShZRS/D (human), ZRSpZRS/D
(python), ZRScZRS/D (cobra), and ZRSfZRS/D (coelacanth fish). Arrow points to rudimentary digits in ZRSpZRS/D embryos. Bottom embryo shows E14.5
636 Cell 167, 633–642, October 20, 2016

Figure 4. Resurrection of Snake Limb
A
Enhancer Function In Vivo
(A) Snake-specific deletion in the ZRS. An alignment
of the central ZRS region for 18 vertebrates,
including six snakes, is shown. Asterisks indicate
nucleotides that are conserved in limbed tetrapods
and fish.
(B) A 17-bp sequence is able to resurrect python
ZRS enhancer function.
(C) Shown are the wild-type (left) and modified (right)
python ZRS in vivo enhancer activities in the limb
buds of transgenic E11.5 mouse embryos. Numbers
of embryos with lacZ activity in the limb over the
total number of transgenic embryos screened are
indicated.
(D) The resurrected allele is able to rescue limb
development when knocked into the mouse
genome in place of the wild-type ZRS. Shown are
gross phenotypes of ZRSpZRS/D (python, left) and
ZRSpZRS(r)/D (python+, right) mice at 2 weeks of age.
Scale bars, 10 mm.
B (E) Skeletal preparations from E18.5 knockin mice
are shown. See Figures S5B and S5C for more
detailed skeletal phenotypes. Scale bars, 2 mm.
See also Figure S5.
in situ hybridization indeed revealed very

weak levels of Shh transcript in the poste-
rior forelimb bud of python ZRS knockin
mouse embryos (data not shown). Taken
together our data indicate that both snake
D enhancers tested lost their ability to induce
normal limb development in mice despite
the much shorter evolutionary distance be-
tween mammals and snakes than between
mammals and lobe-finned fish.
In Vivo Resurrection of a Distant-

Acting Snake Limb Enhancer
E
To identify specific nucleotide changes
within the enhancer that may have led to
its loss of activity in snakes, we examined
the snake sequences in detail. While multi-
ple nucleotide differences are observed
between snakes and limbed lizards (Fig-
ure S1), one small deletion of 17 bp stood
out because it affected a region of the
ZRS that was highly conserved across all
examined tetrapods and fish (Figure 4A).
two to three rudimentary digits in the forelimb and a slightly Although it represents less than 10% of all sequence changes
enlarged ossification resembling a rudimentary zeugopod (Fig- between the snake and lizard ZRS, this deletion is the only
ure 3D). This result may be due to residual enhancer activity sequence that is deleted in all snakes but present in all examined
that was not detected in transgenic reporter assays (Figure 2). limbed vertebrates and fish (Figures 4A and S1). To directly test
Consistent with this possibility, prolonged staining after RNA whether this small snake-specific deletion contributed to the loss
gross and limb skeletal phenotypes of the ZRSD/D KO mice (see Figure S3 for details). Numbers of embryos that exhibited representative limb phenotype over
the total number of embryos with the genotype are indicated. *Three of five mouse embryos displayed mild digit number variation (see Figures S4H–S4J). Scale
bars, 0.1 mm (left column), 2 mm (columns 3 and 4).
See also Figures S3 and S4.
Cell 167, 633–642, October 20, 2016 637

A of enhancer activity in snakes, we created a partially ancestral
allele by reintroducing the 17-bp deleted sequence into the py-
thon enhancer sequence (Figures 2, 4A, and 4B). In a transgenic
mouse reporter assay, this reintroduction of 17 bp of sequence
alone was sufficient to reinstate full enhancer activity in the
posterior mesenchyme of the limb bud at E11.5 (Figure 4C).
To determine whether this allele could also functionally restore
normal limb development in vivo, we used CRISPR/Cas9
genome editing to replace the endogenous mouse enhancer
with this partially ancestral allele. Consistent with the results of
the transgenic reporter experiments, the resulting knockin mice
with the modified python allele had normal limbs (Figures 4D,
4E, and S5B). These results suggest that a 17-bp snake-specific
deletion contributed to enhancer degeneration and that syn-
thetic reintroduction of this microdeletion is sufficient to recreate
the ancestral function of the ZRS and to rescue limb develop-
B ment in vivo.
To identify specific transcription factors that may be involved
in the loss of enhancer function, we examined potential tran-
scription factor binding sites that may have been affected by
the 17-bp sequence deletion in the snake lineage. We identified
a highly conserved motif within the deleted region whose
sequence matched the binding preference of the ETS1 tran-
scription factor. ETS1 has been suggested to directly activate
the ZRS enhancer by binding to multiple ETS recognition sites
(Lettice et al., 2012). We scanned the ZRS-orthologous se-
quences from 18 vertebrates for the presence of additional
conserved ETS motifs (Figure S5C). In total, five ETS motifs
within the enhancer are conserved across tetrapods, which in-
cludes four ETS binding sites previously identified in the mouse
enhancer (Lettice et al., 2012). Remarkably, all five motifs were
C D also conserved in coelacanth (bony fish), and three were present
in elephant shark (cartilaginous fish, Figures 5 and S5C). In
contrast to this strong conservation of ETS motifs across limbed
vertebrates and fish, and, despite the overall conservation of the
ZRS sequence in basal snakes, all examined snakes have lost
the E0 and E1 ETS motifs. In addition, the E4 motif was lost in rat-
tlesnake, and cobra lost the E2 motif (Figures 5B, S1, and S5C).
More generally, in vertebrates with paired appendages the ETS
sites show increased evolutionary constraint compared to the
rest of the ZRS, whereas in snakes the ETS sites do not stand
out as particularly constrained (Figure 5C). The fact that loss of
the E1 motif in the mouse ZRS is not sufficient to alter limb
bud expression (Lettice et al., 2012) and that the boa ZRS is
active despite the absence of both E0 and E1 motifs indicate
that loss of these motifs alone cannot explain ZRS deactivation
in the snake lineage. We therefore also scanned the ZRS for
other transcription factor motifs that showed a similar snake-
specific loss of evolutionary constraint (Table S5). Interestingly,
binding sites for homeodomain transcription factors, which
Figure 5. Loss of Conserved ETS Binding Sites in the Snake Lineage have also been implicated in ZRS regulation (Capellini et al.,
(A) A detailed view of the E1 ETS binding site alignment for 18 vertebrates 2006; Kmita et al., 2005; Lopez-Rios, 2016), display a similar
including six snakes. ETS1 motif is shown above. Asterisks indicate nucleo-
tides that are conserved in limbed tetrapods and fish.
(B) Distribution of tetrapod conserved ETS motifs in the ZRS enhancer (C and D) Relative substitution rates in the ETS and homeodomain DNA motifs
in different jawed vertebrates. Shown is a schematic alignment of the in the ZRS enhancer in non-snake species (black dots: species from Figure 5A)
ZRS for 16 vertebrates (tree) and the locations of predicted ETS binding and snakes (red dots: boa, python and rattlesnake). Mann-Whitney p value is
sites (E0–E4). Red crosses indicate motifs that were lost. See Figure S5 for shown on top.
details. See also Figure S5.
638 Cell 167, 633–642, October 20, 2016

increase in substitution rate in snakes (Figure 5D). Taken of a fully functional ZRS into a snake genome alone would be suf-
together, our results implicate the loss of the E1 ETS site as ficient to induce the formation of fully or even partially developed
well as potentially other ETS and homeodomain transcription limbs in snakes.
factor binding sites in the loss of function of this limb enhancer While we deliberately focused on a locus with strong pre-exist-
in snakes. ing evidence for function from human disease and mouse ge-
netics studies (reviewed in Hill and Lettice, 2013; VanderMeer
DISCUSSION and Ahituv, 2011), an increasing number of unbiased genome-
wide enhancer data across closely and distantly related animal
In the present study, we demonstrate an increased rate of species (Acemel et al., 2016; Arnold et al., 2014; Cotney et al.,
sequence changes, as well as progressive in vivo loss of function 2013; Eckalbar et al., 2016; Gehrke et al., 2015; He et al.,
for a distant-acting limb enhancer in snakes. Decreased 2011; Prescott et al., 2015; Reilly et al., 2015; Villar et al., 2015;
sequence conservation and loss of enhancer function were Xiao et al., 2012) creates a rapidly growing list of candidate line-
most pronounced in advanced snakes, which have lost all skel- age- and species-specific enhancers. A major challenge is the
etal limb structures. The only snake genome in which no ZRS identification of the subsets of these enhancers that functionally
sequence was detected belonged to the corn snake. Our results contribute to morphological and other phenotypic diversity. Our
indicate that the previously reported loss of the ZRS enhancer in study provides an example how genome editing-enabled
Japanese rat snakes (Sagai et al., 2004), a member of the same enhancer replacement makes it possible to recapitulate the
subfamily (Colubrinae) as corn snakes, is not representative of functional erosion of a regulatory sequence across evolution
snakes in general but affects only a small subset of advanced through in vivo experiments. As genome-editing tools are
snakes where it occurred after the morphological loss of all becoming increasingly available, we expect that this approach
limb structures (Figures 1B and S1). Across the snake species will be useful to routinely study the phenotypes associated
examined, the progressive sequence degeneration of the with evolutionary changes in other regulatory sequences associ-
enhancer correlated with its loss of activity in transgenic reporter ated with morphological adaptations in vertebrates.
assays. In contrast, across all limbed tetrapods and fish exam-
ined, the enhancer activity was highly conserved. Remarkably, STAR+METHODS
even a ZRS ortholog from fish (coelacanth), which shares less
sequence similarity with the human ortholog than with the python Detailed methods are provided in the online version of this paper
ortholog (57% versus 59%), was sufficient for normal limb devel- and include the following:
opment despite the major morphological differences between
mammalian limbs and coelacanth fins. d KEY RESOURCES TABLE
The molecular basis of loss of limbs in snakes as they evolved d CONTACT FOR REAGENT AND RESOURCE SHARING
from their limbed ancestor has been the subject of extensive d EXPERIMENTAL MODEL AND SUBJECT DETAILS
speculation (Apesteguı́a and Zaher, 2006; Cohn and Tickle, d METHOD DETAILS
1999; Di-Poı̈ et al., 2010; Infante et al., 2015; Lopez-Rios, 2016; B Phylogenetic Analysis
Martill et al., 2015; Sagai et al., 2004; Tchernov et al., 2000; Zeller B In Vivo Transgenic Reporter Assays
et al., 2009). Our genomic enhancer replacement experiments in B Generation of Enhancer Knockout and Knockin Mice
mice conclusively demonstrate that the loss of function in a single Using CRISPR/Cas9
enhancer observed in snakes is sufficient to cause severe limb B In Situ Hybridization
reduction in mice, raising the possibility that ZRS deactivation B Skeletal Preparations
contributed to the loss of limbs in the snake lineage. However, B Sample Selection and Blinding
changes in other sequences involved in limb development must B Motif Analysis
also have occurred in snakes. These changes could for example d QUANTIFICATION AND STATISTICAL ANALYSIS
involve regulation of Hox genes that act upstream of Shh (Cohn B Substitution Rates in TF Motifs
and Tickle, 1999; Di-Poı̈ et al., 2010; Head and Polly, 2015), or B Differences in Evolutionary Rates
other genes that are critical for initiation of limb development
(e.g., Min et al., 1998; Rallis et al., 2003; Sekine et al., 1999; Ta- SUPPLEMENTAL INFORMATION
naka et al., 2002). Notably, following the morphological disap-
Supplemental Information includes five figures and six tables and can be found
pearance of limbs, any sequence required exclusively for limb
with this article online at http://dx.doi.org/10.1016/j.cell.2016.09.028.
development is no longer subject to negative selection and is ex- An audio PaperClip is available at http://dx.doi.org/10.1016/j.cell.2016.09.
pected to degrade over time. This is exemplified by the reduction 028#mmc2.
in the transgenic reporter activity of other serpentine limb en-
hancers whose phenotypic impact on limb development remains AUTHOR CONTRIBUTIONS
to be determined (Guerreiro et al., 2016; Infante et al., 2015). In the
case of the ZRS, the enhancer activity observed in a basal snake E.Z.K., D.E.D., E.M.R., A.V., and L.A.P. conceived the project. E.Z.K. and O.K.
performed the phylogenetic analysis. E.Z.K. and U.S.M. cloned transgenic re-
(boa, Figure 2) suggests that the sequence degeneration of the
porter and targeting vectors. E.Z.K., B.J.M., I.P.-F., C.S.P., T.H.G., M.K.,
ZRS in snakes started in conjunction with or, more likely, after E.A.L., J.A.A., and V.A. carried out transgenic validation. E.Z.K. performed
other disruptive molecular events contributing to the loss of the enhancer knockout and knockin studies. V.T., J.L.-R., M.O., and E.Z.K.
limbs. Consequently, we do not expect that the reintroduction performed in situ hybridization (ISH). I.B. and E.Z.K. performed motif analysis.
Cell 167, 633–642, October 20, 2016 639

E.Z.K., A.V., and L.A.P. wrote the manuscript with input from the remaining (2010). Adaptive evolution of pelvic reduction in sticklebacks by recurrent
authors. deletion of a Pitx1 enhancer. Science 327, 302–305.
Cohn, M.J., and Tickle, C. (1999). Developmental basis of limblessness and
ACKNOWLEDGMENTS axial patterning in snakes. Nature 399, 474–479.
Cooper, K.L., Sears, K.E., Uygur, A., Maier, J., Baczkowski, K.-S., Brosnahan,
This work was supported by NIH grants R01HG003988, U54HG006997, and
M., Antczak, D., Skidmore, J.A., and Tabin, C.J. (2014). Patterning and post-
U01DE024427 (to A.V. and L.A.P.) and by the Olga Mayenfisch Foundation
patterning modes of evolutionary digit loss in mammals. Nature 511, 41–45.
and the University of Basel (to J.L.-R). E.Z.K. is supported by a postdoctoral
fellowship from the Helen Hay Whitney Foundation funded by a Howard Cotney, J., Leng, J., Yin, J., Reilly, S.K., DeMare, L.E., Emera, D., Ayoub, A.E.,
Hughes Medical Institute. M.O. was supported by a Swiss National Science Rakic, P., and Noonan, J.P. (2013). The evolution of lineage-specific regulatory
Foundation (SNSF) fellowship. We thank T. Castoe for providing the draft as- activities in the human embryonic limb. Cell 154, 185–196.
sembly sequence of the Boa genome and J. Doudna for providing a plasmid Cretekos, C.J., Wang, Y., Green, E.D., Martin, J.F., Rasweiler, J.J., 4th, and
containing a human-optimized Cas9 gene. We also thank A. Stark for helpful Behringer, R.R. (2008). Regulatory divergence modifies limb length between
suggestions and comments on the manuscript. Research was conducted at mammals. Genes Dev. 22, 141–151.
the E.O. Lawrence Berkeley National Laboratory and performed under Depart- Dahn, R.D., Davis, M.C., Pappano, W.N., and Shubin, N.H. (2007). Sonic
ment of Energy Contract DE-AC02-05CH11231, University of California. hedgehog function in chondrichthyan fins and the evolution of appendage
patterning. Nature 445, 311–314.
Received: June 19, 2016
Darriba, D., Taboada, G.L., Doallo, R., and Posada, D. (2012). jModelTest 2:
Revised: August 7, 2016
More models, new heuristics and parallel computing. Nat. Methods 9, 772.
Accepted: September 15, 2016
Published: October 20, 2016 Di-Poı̈, N., Montoya-Burgos, J.I., Miller, H., Pourquié, O., Milinkovitch, M.C.,
and Duboule, D. (2010). Changes in Hox genes’ structure and function during
the evolution of the squamate body plan. Nature 464, 99–103.
REFERENCES
Echelard, Y., Epstein, D.J., St-Jacques, B., Shen, L., Mohler, J., McMahon,
Acemel, R.D., Tena, J.J., Irastorza-Azcarate, I., Marlétaz, F., Gómez-Marı́n, C., J.A., and McMahon, A.P. (1993). Sonic hedgehog, a member of a family of pu-
de la Calle-Mustienes, E., Bertrand, S., Diaz, S.G., Aldea, D., Aury, J.M., et al. tative signaling molecules, is implicated in the regulation of CNS polarity. Cell
(2016). A single three-dimensional chromatin compartment in amphioxus indi- 75, 1417–1430.
cates a stepwise evolution of vertebrate Hox bimodal regulation. Nat. Genet. Eckalbar, W.L., Schlebusch, S.A., Mason, M.K., Gill, Z., Parker, A.V., Booker,
48, 336–341. B.M., Nishizaki, S., Muswamba-Nday, C., Terhune, E., Nevonen, K.A., et al.
Amemiya, C.T., Alföldi, J., Lee, A.P., Fan, S., Philippe, H., Maccallum, I., (2016). Transcriptomic and epigenomic characterization of the developing
Braasch, I., Manousaki, T., Schneider, I., Rohner, N., et al. (2013). The African bat wing. Nat. Genet. 48, 528–536.
coelacanth genome provides insights into tetrapod evolution. Nature 496, Esquerré, D., and Scott Keogh, J. (2016). Parallel selective pressures drive
311–316. convergent diversification of phenotypes in pythons and boas. Ecol. Lett.
Apesteguı́a, S., and Zaher, H. (2006). A Cretaceous terrestrial snake with 19, 800–809.
robust hindlimbs and a sacrum. Nature 440, 1037–1040. Gehrke, A.R., Schneider, I., de la Calle-Mustienes, E., Tena, J.J.,
Arnold, C.D., Gerlach, D., Spies, D., Matts, J.A., Sytnikova, Y.A., Pagani, M., Gomez-Marin, C., Chandran, M., Nakamura, T., Braasch, I., Postlethwait,
Lau, N.C., and Stark, A. (2014). Quantitative genome-wide enhancer activity J.H., Gómez-Skarmeta, J.L., et al. (2015). Deep conservation of wrist and digit
maps for five Drosophila species show functional enhancer conservation enhancers in fish. Proc. Natl. Acad. Sci. USA 112, 803–808.
and turnover during cis-regulatory evolution. Nat. Genet. 46, 685–692. Gibson, D.G., Young, L., Chuang, R.-Y., Venter, J.C., Hutchison, C.A., 3rd, and
Attanasio, C., Nord, A.S., Zhu, Y., Blow, M.J., Li, Z., Liberton, D.K., Morrison, Smith, H.O. (2009). Enzymatic assembly of DNA molecules up to several hun-
H., Plajzer-Frick, I., Holt, A., Hosseini, R., et al. (2013). Fine tuning of craniofa- dred kilobases. Nat. Methods 6, 343–345.
cial morphology by distant-acting enhancers. Science 342, 1241006. Grant, C.E., Bailey, T.L., and Noble, W.S. (2011). FIMO: Scanning for occur-
Blow, M.J., McCulley, D.J., Li, Z., Zhang, T., Akiyama, J.A., Holt, A., Plajzer- rences of a given motif. Bioinformatics 27, 1017–1018.
Frick, I., Shoukry, M., Wright, C., Chen, F., et al. (2010). ChIP-Seq identification Guenther, C.A., Tasic, B., Luo, L., Bedell, M.A., and Kingsley, D.M. (2014).
of weakly conserved heart enhancers. Nat. Genet. 42, 806–810. A molecular basis for classic blond hair color in Europeans. Nat. Genet. 46,
Britten, R.J., and Davidson, E.H. (1969). Gene regulation for higher cells: A the- 748–752.
ory. Science 165, 349–357. Guerreiro, I., Nunes, A., Woltering, J.M., Casaca, A., Nóvoa, A., Vinagre, T.,
Bronson, S.K., Plaehn, E.G., Kluckman, K.D., Hagaman, J.R., Maeda, N., and Hunter, M.E., Duboule, D., and Mallo, M. (2013). Role of a polymorphism in a
Smithies, O. (1996). Single-copy transgenic mice with chosen-site integration. Hox/Pax-responsive enhancer in the evolution of the vertebrate spine. Proc.
Proc. Natl. Acad. Sci. USA 93, 9067–9072. Natl. Acad. Sci. USA 110, 10682–10686.
Capellini, T.D., Di Giacomo, G., Salsi, V., Brendolan, A., Ferretti, E., Srivastava, Guerreiro, I., Gitto, S., Nóvoa, A., Codourey, J., Nguyen Huynh, T.H., Gonza-
D., Zappavigna, V., and Selleri, L. (2006). Pbx1/Pbx2 requirement for distal lez, F., Milinkovitch, M.C., Mallo, M., and Duboule, D. (2016). Reorganisation
limb patterning is mediated by the hierarchical control of Hox gene spatial dis- of Hoxd regulatory landscapes during the evolution of a snake-like body
tribution and Shh expression. Development 133, 2263–2273. plan. eLife 5, 347.
Carroll, S.B. (2008). Evo-devo and an expanding evolutionary synthesis: A Guindon, S., Dufayard, J.-F., Lefort, V., Anisimova, M., Hordijk, W., and
genetic theory of morphological evolution. Cell 134, 25–36. Gascuel, O. (2010). New algorithms and methods to estimate maximum-likeli-
Castoe, T.A., de Koning, A.P.J., Hall, K.T., Card, D.C., Schield, D.R., Fujita, hood phylogenies: Assessing the performance of PhyML 3.0. Syst. Biol. 59,
M.K., Ruggiero, R.P., Degner, J.F., Daza, J.M., Gu, W., et al. (2013). The Bur- 307–321.
mese python genome reveals the molecular basis for extreme adaptation in He, Q., Bardet, A.F., Patton, B., Purvis, J., Johnston, J., Paulson, A., Gogol, M.,
snakes. Proc. Natl. Acad. Sci. USA 110, 20645–20650. Stark, A., and Zeitlinger, J. (2011). High conservation of transcription factor
Castresana, J. (2000). Selection of conserved blocks from multiple alignments binding and evidence for combinatorial regulation across six Drosophila spe-
for their use in phylogenetic analysis. Mol. Biol. Evol. 17, 540–552. cies. Nat. Genet. 43, 414–420.
Chan, Y.F., Marks, M.E., Jones, F.C., Villarreal, G., Jr., Shapiro, M.D., Brady, Head, J.J., and Polly, P.D. (2015). Evolution of the snake body form reveals
S.D., Southwick, A.M., Absher, D.M., Grimwood, J., Schmutz, J., et al. homoplasy in amniote Hox gene function. Nature 520, 86–89.
640 Cell 167, 633–642, October 20, 2016

Heinz, S., Benner, C., Spann, N., Bertolino, E., Lin, Y.C., Laslo, P., Cheng, J.X., Martill, D.M., Tischlinger, H., and Longrich, N.R. (2015). EVOLUTION. A four-
Murre, C., Singh, H., and Glass, C.K. (2010). Simple combinations of legged snake from the Early Cretaceous of Gondwana. Science 349, 416–419.
lineage-determining transcription factors prime cis-regulatory elements May, D., Blow, M.J., Kaplan, T., McCulley, D.J., Jensen, B.C., Akiyama, J.A.,
required for macrophage and B cell identities. Mol. Cell 38, 576–589. Holt, A., Plajzer-Frick, I., Shoukry, M., Wright, C., et al. (2011). Large-scale dis-
Hill, R.E., and Lettice, L.A. (2013). Alterations to the remote control of Shh gene covery of enhancers from human heart tissue. Nat. Genet. 44, 89–93.
expression cause congenital abnormalities. Philos. Trans. R. Soc. Lond. B McLean, C.Y., Reno, P.L., Pollen, A.A., Bassan, A.I., Capellini, T.D., Guenther,
Biol. Sci. 368, 20120357. C., Indjeian, V.B., Lim, X., Menke, D.B., Schaar, B.T., et al. (2011). Human-spe-
Hoekstra, H.E. (2012). Genomics: Stickleback is the catch of the day. Nature cific loss of regulatory DNA and the evolution of human-specific traits. Nature
484, 46–47. 471, 216–219.
Hsiang, A.Y., Field, D.J., Webster, T.H., Behlke, A.D.B., Davis, M.B., Racicot, Min, H., Danilenko, D.M., Scully, S.A., Bolon, B., Ring, B.D., Tarpley, J.E.,
R.A., and Gauthier, J.A. (2015). The origin of snakes: Revealing the ecology, DeRose, M., and Simonet, W.S. (1998). Fgf-10 is required for both limb and
behavior, and evolutionary history of early snakes using genomics, phenom- lung development and exhibits striking functional similarity to Drosophila
ics, and the fossil record. BMC Evol. Biol. 15, 87. branchless. Genes Dev. 12, 3156–3161.
Indjeian, V.B., Kingman, G.A., Jones, F.C., Guenther, C.A., Grimwood, J., Montague, T.G., Cruz, J.M., Gagnon, J.A., Church, G.M., and Valen, E. (2014).
Schmutz, J., Myers, R.M., and Kingsley, D.M. (2016). Evolving new skeletal CHOPCHOP: A CRISPR/Cas9 and TALEN web tool for genome editing.
traits by cis-regulatory changes in bone morphogenetic proteins. Cell 164, Nucleic Acids Res. 42, W401–W407.
45–56. Ovchinnikov, D. (2009). Alcian blue/alizarin red staining of cartilage and bone in
Infante, C.R., Mihala, A.G., Park, S., Wang, J.S., Johnson, K.K., Lauderdale, mouse. Cold Spring Harb. Protoc. Published online March 2009. http://dx.doi.
J.D., and Menke, D.B. (2015). Shared enhancer activity in the limbs and phallus org/10.1101/pdb.prot5170.
and functional divergence of a limb-genital cis-regulatory element in snakes. Panman, L., Galli, A., Lagarde, N., Michos, O., Soete, G., Zuniga, A., and Zeller,
Dev. Cell 35, 107–119. R. (2006). Differential regulation of gene expression in the digit forming area of
Jolma, A., Yan, J., Whitington, T., Toivonen, J., Nitta, K.R., Rastas, P., Morgu- the mouse limb bud by SHH and gremlin 1/FGF-mediated epithelial-mesen-
nova, E., Enge, M., Taipale, M., Wei, G., et al. (2013). DNA-binding specificities chymal signalling. Development 133, 3419–3428.
of human transcription factors. Cell 152, 327–339. Pennacchio, L.A., Ahituv, N., Moses, A.M., Prabhakar, S., Nobrega, M.A.,
Shoukry, M., Minovitsky, S., Dubchak, I., Holt, A., Lewis, K.D., et al. (2006).
Jones, F.C., Grabherr, M.G., Chan, Y.F., Russell, P., Mauceli, E., Johnson, J.,
In vivo enhancer analysis of human conserved non-coding sequences. Nature
Swofford, R., Pirun, M., Zody, M.C., White, S., et al.; Broad Institute Genome
444, 499–502.
Sequencing Platform & Whole Genome Assembly Team (2012). The genomic
basis of adaptive evolution in threespine sticklebacks. Nature 484, 55–61. Prabhakar, S., Visel, A., Akiyama, J.A., Shoukry, M., Lewis, K.D., Holt, A.,
Plajzer-Frick, I., Morrison, H., Fitzpatrick, D.R., Afzal, V., et al. (2008). Hu-
Katoh, K., and Standley, D.M. (2013). MAFFT multiple sequence alignment
man-specific gain of function in a developmental enhancer. Science 321,
software version 7: Improvements in performance and usability. Mol. Biol.
1346–1350.
Evol. 30, 772–780.
Prescott, S.L., Srinivasan, R., Marchetto, M.C., Grishina, I., Narvaiza, I., Selleri,
King, M.C., and Wilson, A.C. (1975). Evolution at two levels in humans and
L., Gage, F.H., Swigut, T., and Wysocka, J. (2015). Enhancer divergence and
chimpanzees. Science 188, 107–116.
cis-regulatory evolution in the human and chimp neural crest. Cell 163, 68–83.
Kmita, M., Tarchini, B., Zàkàny, J., Logan, M., Tabin, C.J., and Duboule, D.
Pyron, R.A., Burbrink, F.T., and Wiens, J.J. (2013). A phylogeny and revised
(2005). Early developmental arrest of mammalian limbs lacking HoxA/HoxD
classification of Squamata, including 4161 species of lizards and snakes.
gene function. Nature 435, 1113–1116.
BMC Evol. Biol. 13, 93.
Kothary, R., Clapoff, S., Darling, S., Perry, M.D., Moran, L.A., and Rossant, J.
Rallis, C., Bruneau, B.G., Del Buono, J., Seidman, C.E., Seidman, J.G., Nissim,
(1989). Inducible expression of an hsp68-lacZ hybrid gene in transgenic mice.
S., Tabin, C.J., and Logan, M.P.O. (2003). Tbx5 is required for forelimb bud
Development 105, 707–714.
formation and continued outgrowth. Development 130, 2741–2751.
Lawson, R., Slowinski, J.B., Crother, B.I., and Burbrink, F.T. (2005). Phylogeny Reilly, S.K., Yin, J., Ayoub, A.E., Emera, D., Leng, J., Cotney, J., Sarro, R.,
of the Colubroidea (Serpentes): New evidence from mitochondrial and nuclear Rakic, P., and Noonan, J.P. (2015). Evolutionary genomics. Evolutionary
genes. Mol. Phylogenet. Evol. 37, 581–601. changes in promoter and enhancer activity during human corticogenesis. Sci-
Lettice, L.A., Heaney, S.J.H., Purdie, L.A., Li, L., de Beer, P., Oostra, B.A., ence 347, 1155–1159.
Goode, D., Elgar, G., Hill, R.E., and de Graaff, E. (2003). A long-range Shh Romanes, G.J. (1892). Darwin and after Darwin: An Exposition of the Darwinian
enhancer regulates expression in the developing limb and fin and is associated Theory and a Discussion of post-Darwinian Questions (The Open Court Pub-
with preaxial polydactyly. Hum. Mol. Genet. 12, 1725–1735. lishing Company).
Lettice, L.A., Hill, A.E., Devenney, P.S., and Hill, R.E. (2008). Point mutations in Sagai, T., Masuya, H., Tamura, M., Shimizu, K., Yada, Y., Wakana, S., Gondo,
a distant sonic hedgehog cis-regulator generate a variable regulatory output Y., Noda, T., and Shiroishi, T. (2004). Phylogenetic conservation of a limb-
responsible for preaxial polydactyly. Hum. Mol. Genet. 17, 978–985. specific, cis-acting regulator of Sonic hedgehog ( Shh). Mamm. Genome 15,
Lettice, L.A., Williamson, I., Wiltshire, J.H., Peluso, S., Devenney, P.S., Hill, 23–34.
A.E., Essafi, A., Hagman, J., Mort, R., Grimes, G., et al. (2012). Opposing func- Sagai, T., Hosoya, M., Mizushina, Y., Tamura, M., and Shiroishi, T. (2005).
tions of the ETS factor family define Shh spatial expression in limb buds and Elimination of a long-range cis-regulatory module causes complete loss of
underlie polydactyly. Dev. Cell 22, 459–467. limb-specific Shh expression and truncation of the mouse limb. Development
Lettice, L.A., Williamson, I., Devenney, P.S., Kilanowski, F., Dorin, J., and Hill, 132, 797–803.
R.E. (2014). Development of five digits is controlled by a bipartite long-range Sekine, K., Ohuchi, H., Fujiwara, M., Yamasaki, M., Yoshizawa, T., Sato, T.,
cis-regulator. Development 141, 1715–1725. Yagishita, N., Matsui, D., Koga, Y., Itoh, N., and Kato, S. (1999). Fgf10 is essen-
Lopez-Rios, J. (2016). The many lives of SHH in limb development and evolu- tial for limb and lung formation. Nat. Genet. 21, 138–141.
tion. Semin. Cell Dev. Biol. 49, 116–124. Shimodaira, H., and Hasegawa, M. (2001). CONSEL: For assessing the confi-
Lopez-Rios, J., Duchesne, A., Speziale, D., Andrey, G., Peterson, K.A., dence of phylogenetic tree selection. Bioinformatics 17, 1246–1247.
Germann, P., Ünal, E., Liu, J., Floriot, S., Barbey, S., et al. (2014). Attenuated Tanaka, M., Münsterberg, A., Anderson, W.G., Prescott, A.R., Hazon, N., and
sensing of SHH by Ptch1 underlies evolution of bovine limbs. Nature 511, Tickle, C. (2002). Fin development in a cartilaginous fish and the origin of verte-
46–51. brate limbs. Nature 416, 527–531.
Cell 167, 633–642, October 20, 2016 641

Tchernov, E., Rieppel, O., Zaher, H., Polcyn, M.J., and Jacobs, L.L. (2000). A Wittkopp, P.J., and Kalay, G. (2011). Cis-regulatory elements: Molecular
fossil snake with limbs. Science 287, 2010–2012. mechanisms and evolutionary processes underlying divergence. Nat. Rev.
Ullate-Agote, A., Milinkovitch, M.C., and Tzika, A.C. (2014). The genome Genet. 13, 59–69.
sequence of the corn snake (Pantherophis guttatus), a valuable resource for
Wray, G.A. (2007). The evolutionary significance of cis-regulatory mutations.
EvoDevo studies in squamates. Int. J. Dev. Biol. 58, 881–888.
Nat. Rev. Genet. 8, 206–216.
VanderMeer, J.E., and Ahituv, N. (2011). cis-regulatory mutations are a genetic
cause of human limb malformations. Dev. Dyn. 240, 920–930. Wright, A.M., Lyons, K.M., Brandley, M.C., and Hillis, D.M. (2015). Which came
Villar, D., Berthelot, C., Aldridge, S., Rayner, T.F., Lukk, M., Pignatelli, M., Park, first: The lizard or the egg? Robustness in phylogenetic reconstruction of
T.J., Deaville, R., Erichsen, J.T., Jasinska, A.J., et al. (2015). Enhancer evolu- ancestral states. J. Exp. Zool. B Mol. Dev. Evol. 324, 504–516.
tion across 20 mammalian species. Cell 160, 554–566. Xiao, S., Xie, D., Cao, X., Yu, P., Xing, X., Chen, C.-C., Musselman, M., Xie, M.,
Visel, A., Minovitsky, S., Dubchak, I., and Pennacchio, L.A. (2007). VISTA West, F.D., Lewin, H.A., et al. (2012). Comparative epigenomic annotation of
Enhancer Browser—A database of tissue-specific human enhancers. Nucleic regulatory DNA. Cell 149, 1381–1392.
Acids Res. 35, D88–D92.
Yang, H., Wang, H., Shivalila, C.S., Cheng, A.W., Shi, L., and Jaenisch, R.
Visel, A., Blow, M.J., Li, Z., Zhang, T., Akiyama, J.A., Holt, A., Plajzer-Frick, I.,
(2013). One-step generation of mice carrying reporter and conditional alleles
Shoukry, M., Wright, C., Chen, F., et al. (2009). ChIP-seq accurately predicts
by CRISPR/Cas-mediated genome engineering. Cell 154, 1370–1379.
tissue-specific activity of enhancers. Nature 457, 854–858.
Vonk, F.J., Casewell, N.R., Henkel, C.V., Heimberg, A.M., Jansen, H.J., Yang, H., Wang, H., and Jaenisch, R. (2014). Generating genetically modified
McCleary, R.J.R., Kerkkamp, H.M.E., Vos, R.A., Guerreiro, I., Calvete, J.J., mice using CRISPR/Cas-mediated genome engineering. Nat. Protoc. 9, 1956–
et al. (2013). The king cobra genome reveals dynamic gene evolution and 1968.
adaptation in the snake venom system. Proc. Natl. Acad. Sci. USA 110,
Zeller, R., and Zuniga, A. (2007). Shh and Gremlin1 chromosomal landscapes
20651–20656.
in development and disease. Curr. Opin. Genet. Dev. 17, 428–434.
Wang, H., Yang, H., Shivalila, C.S., Dawlaty, M.M., Cheng, A.W., Zhang, F.,
and Jaenisch, R. (2013). One-step generation of mice carrying mutations in Zeller, R., López-Rı́os, J., and Zuniga, A. (2009). Vertebrate limb bud develop-
multiple genes by CRISPR/Cas-mediated genome engineering. Cell 153, ment: Moving towards integrative analysis of organogenesis. Nat. Rev. Genet.
910–918. 10, 845–858.
642 Cell 167, 633–642, October 20, 2016

Article
Domestication and Divergence

of Saccharomyces cerevisiae Beer Yeasts
Brigida Gallone,1,2,3,4,11 Jan Steensels,1,2,11 Troels Prahl,5 Leah Soriaga,6 Veerle Saels,1,2 Beatriz Herrera-Malaver,1,2
Adriaan Merlevede,1,2 Miguel Roncoroni,1,2 Karin Voordeckers,1,2 Loren Miraglia,8 Clotilde Teiling,9 Brian Steffy,9
Maryann Taylor,10 Ariel Schwartz,6 Toby Richardson,6 Christopher White,5 Guy Baele,7 Steven Maere,3,4,*
and Kevin J. Verstrepen1,2,12,*
1Laboratory for Genetics and Genomics, Centre of Microbial and Plant Genetics (CMPG), KU Leuven, Kasteelpark Arenberg 22, 3001 Leuven,
Belgium
2Laboratory for Systems Biology, VIB, Bio-Incubator, Gaston Geenslaan 1, 3001 Leuven, Belgium
3Department of Plant Systems Biology, VIB, 9052 Gent, Belgium
4Department of Plant Biotechnology and Bioinformatics, Ghent University, 9052 Gent, Belgium
5White Labs, 9495 Candida Street, San Diego, CA 92126, USA
6Synthetic Genomics, 11149 North Torrey Pines Road, La Jolla, CA 92037, USA
7Department of Microbiology and Immunology, Rega Institute, KU Leuven, 3000 Leuven, Belgium
8Encinitas Brewing Science, 141 Rodney Avenue, Encinitas, CA 92024, USA
9Illumina, 5200 Illumina Way, San Diego, CA 92122, USA
10Biological & Popular Culture (BioPop), 2205 Faraday Avenue, Suite E, Carlsbad, CA 92008, USA
11Co-first author
12Lead Contact
*Correspondence: steven.maere@psb.vib-ugent.be (S.M.), kevin.verstrepen@biw.vib-kuleuven.be (K.J.V.)

SUMMARY due to the presence of ethanol (Michel et al., 1992; Steensels and
Verstrepen, 2014). Whereas the use of pure cultures started well
Whereas domestication of livestock, pets, and crops after the pioneering work of Pasteur and Hansen in the 19th cen-
is well documented, it is still unclear to what extent tury, early brewers, winemakers, and bakers had already learned
microbes associated with the production of food that inoculating unfermented foods with a small portion of fer-
have also undergone human selection and where mented product resulted in fast and more predictable fermenta-
the plethora of industrial strains originates from. tions. This so-called ‘‘backslopping’’ might have resulted in
yeast lineages that grew continuously in these man-made envi-
Here, we present the genomes and phenomes of
ronments and lost contact with their natural niches, providing a
157 industrial Saccharomyces cerevisiae yeasts. perfect setting for domestication. However, strong evidence
Our analyses reveal that today’s industrial yeasts for this hypothesis is still missing and it remains unclear whether
can be divided into five sublineages that are geneti- industrial yeast diversity is shaped by selection and niche
cally and phenotypically separated from wild strains adaptation (domestication) or neutral divergence caused by
and originate from only a few ancestors through geographic isolation and limited dispersal (Goddard and Greig,
complex patterns of domestication and local diver- 2015; Warringer et al., 2011).
gence. Large-scale phenotyping and genome anal- Domestication is defined as human selection and breeding of
ysis further show strong industry-specific selection wild species to obtain cultivated variants that thrive in man-made
for stress tolerance, sugar utilization, and flavor pro- environments, but behave suboptimally in nature. Typical signs
duction, while the sexual cycle and other phenotypes of domestication, including genome decay, polyploidy, chromo-
somal rearrangements, gene duplications, and phenotypes re-
related to survival in nature show decay, particularly
sulting from human-driven selection, have been reported in
in beer yeasts. Together, these results shed light on crops, livestock, and pets (Driscoll et al., 2009; Purugganan
the origins, evolutionary history, and phenotypic di- and Fuller, 2009). Several studies have recently investigated
versity of industrial yeasts and provide a resource the S. cerevisiae population by sequencing the genomes of hun-
for further selection of superior strains. dreds of different strains, providing a first glimpse of the complex
evolution of this species (Almeida et al., 2015; Borneman et al.,
INTRODUCTION 2011, 2016; Liti et al., 2009; Magwene et al., 2011; Schacherer
et al., 2009; Strope et al., 2015). However, most of these studies
Since prehistoric times, humans have exploited the capacity of focused primarily on yeasts from wild and clinical habitats and
the common baker’s yeast Saccharomyces cerevisiae to convert often include only a limited set of industrial strains, mainly origi-
sugars into ethanol and desirable flavor compounds to obtain nating from wine. Moreover, most studies use haploid deriva-
foods and beverages with a prolonged shelf-life, enriched tives instead of natural strains and can therefore not explore
sensorial palate, improved digestibility, and an euphoriant effect typical patterns of domestication like polyploidy, aneuploidy,
Cell 166, 1397–1410, September 8, 2016 ª 2016 The Author(s). Published by Elsevier Inc. 1397
This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/).
and heterozygosity. The use of haploids also excludes a large exclusively produced by strains of the genetically related
fraction of industrial S. cerevisiae strains that have lost the ability Saccharomyces pastorianus. After de novo assembly of each
to sporulate, such as the vast majority of beer yeasts. Neverthe- of the genomes, we inferred a maximum-likelihood phylogenetic
less, some studies already revealed signs of domestication in tree based on codon alignments for 2,020 concatenated single-
wine strains, such as an increased resistance to copper copy nuclear genes shared by each of the 157 isolates and
(present in grapevine pesticides) and sulfite (used as a preserva- the outgroup species Saccharomyces paradoxus (Figure S1A).
tive in wine) (Pérez-Ortı́n et al., 2002; Warringer et al., 2011). An Additionally, we included a representative set of 24 previously
in-depth investigation of strains originating from other industrial sequenced strains belonging to the main established lineages
niches is still lacking. of the S. cerevisiae phylogeny (Liti et al., 2009; Strope et al.,
Here, we describe the high-quality sequencing, de novo 2015), extending the number of strains to 181 (Figure 1A). Trees
assembly, annotation, and extensive phenotyping of 157 constructed from the original and extended datasets are
S. cerevisiae strains used for the industrial production of beer, congruent and show five main lineages that contain the majority
wine, bread, spirits, saké, and bioethanol, in their natural ploidy. of industrial yeasts: Wine (bootstrap support 100%), Beer 1
Our data reveal that industrial yeasts are genetically and pheno- (86%), Beer 2 (56%), Asia (100%), and a Mixed lineage (99%)
typically distinct from wild strains and stem from only a limited containing yeasts used in different industries. Three of these lin-
set of ancestral strains that have been adapting to man-made eages (Beer 1, Beer 2, and Mixed) were not previously described.
environments. They further diversified into five clades: one Next, we studied the population structure in a filtered set of
including Asian strains such as saké yeasts, one mostly contain- 53,929 polymorphic sites accounting for 2,454,052 SNPs across
ing wine yeasts, a mixed clade that contains bread and other all strains, using the Bayesian model-based clustering approach
yeasts, and two separate families of beer yeasts. While most implemented in fastStructure (Raj et al., 2014) (Figures 1B and
clades lack strong geographical substructure, one of the beer S1B). This analysis yields a population structure that is highly
clades contains geographically isolated subgroups of strains consistent with the major lineages defined in the phylogeny
used in continental Europe (Belgium/Germany), the United and identifies mosaicism in 17% of the strains (in which the esti-
Kingdom, and a recent sublineage of United States beer yeasts mated ancestry Q < 0.8 for K = 8 ancestral populations). The
that diverged from the British subclade during colonization. population structure is further supported by a principal compo-
Interestingly, these beer yeast lineages exhibit clear and pro- nent analysis (PCA) on the same SNP data (Figure 1C).
found hallmarks of domestication, more so than the other line- Further analysis of the phylogeny and population structure
ages. The shift from variable, complex, and often harsh environ- reveals that the evolutionary divergence of industrial yeasts is
ments encountered in nature to more stable and nutrient-rich shaped by both their industrial application and geographical
beer medium favored specialized adaptations in beer yeasts, origin. First, most yeasts cluster together according to the indus-
but also led to genome decay, aneuploidy, and loss of a func- try in which they are used and are clearly separated from the wild
tional sexual cycle. Specifically, we find evidence for active or clinical yeasts that have previously been sequenced. This was
human selection, demonstrated by convergent evolution for effi- further confirmed by constructing a larger phylogeny, based
cient fermentation of beer-specific carbon sources, mainly on nine genomic regions, that includes the vast majority of all
through mutations and duplications of the MAL (maltose) genes, sequenced S. cerevisiae strains, 450 isolates in total (Figure S1C;
as well as nonsense mutations in PAD1 and FDC1, which are Table S2). Wine and saké yeasts cluster in the previously identi-
involved in the production of 4-vinyl guaiacol (4-VG), an unde- fied Wine and Asia lineages (Liti et al., 2009). The majority of beer
sirable off-flavor in beer. Our results further suggest that beer yeasts (85.3%) are found in two main lineages (Beer 1 and Beer
yeast domestication was initiated hundreds of years ago, well 2) that are only distantly related. The Mixed clade harbors 7.8%
after the first reported beer production, but before the discovery of all beer strains (most of which are atypical beer yeasts that
of microbes. Together, our results reveal how today’s industrial are used for bottle refermentation of strong Belgian ales) and
yeasts are the outcome of centuries of human domestication contains all bread strains. Interestingly, spirit strains lack this
and provide a new resource for further selection and breeding clear phylogenetic relationship, as they are highly mosaic and
of superior variants. scattered throughout the tree, suggesting that these strains
might be the result of breeding by modern-day yeast companies
RESULTS that sell yeasts for spirits production. Moreover, because spirit
yeasts are typically not re-used after fermentation, they likely
Niche and Geography Drive Diversification had less opportunity to diverge into a separate clade.
To examine the evolutionary history of industrial yeasts, we Within and between the lineages, we also observed geograph-
sequenced the genomes of 157 S. cerevisiae isolates originating ical patterns. For example, most saké yeasts form a monophy-
from various sources in their natural ploidy to a median coverage letic group and cluster together with wild isolates and bioethanol
of 1353 (min = 263, max = 4033) (for details on data analysis, strains from China, while South American bioethanol strains are
see STAR Methods). This collection includes 102 industrial closely related to strains used to produce cachaça, a Brazilian
beer strains, 19 wine strains, 11 spirit strains, 7 saké strains, 7 sugarcane spirit. Moreover, the Beer 1 clade consists of three
strains isolated from spontaneous fermentations, 5 bioethanol separate subpopulations, each reflecting geographically distinct
strains, 4 bread strains, and 2 laboratory strains (Table S1). Inter- groups: Belgium/Germany, Britain, and the United States. The
estingly, ten of these S. cerevisiae beer strains are used for com- absence of genetic admixture among these subpopulations indi-
mercial production of lager beers, which were believed to be cates that these strains diverged allopatrically after the initial split
1398 Cell 166, 1397–1410, September 8, 2016

A
Beer 2
LA YC3467
BE003
BE039
BE063
BE040
BE080
BE011
BE021
BE085
BE059
BE027
BE0321
WI01 4
BE09 2
Origin
BE00 9
BE0692
NC 2
BE00
BE0 86
BE0013
BE0002
01 0
BE 034
W 18
W
BE 083
SP 145
yjmI016
BE 084
Beer A
BE 004
yj 3 0
W m6 20
SP 06
W L00 82
SP I002 5
W L0 1
Wine
yjm
B I00 0
L L0 03
0
B 03 4
32 574 6
N A0 02
BE 01 4
YC 3 6
Spirits
BE I01 07
C C 2
N C 01
N CY YC3
65
W I0 18
W I0
Saké NC W
W
5
YC WI I01 00 03
Wild 3 0 0 SA A0 04
As
NC W 44 06 S A0 07
I 0 7
YC WI0 17
ia
Bio-ethanol 3 0
S A0 01
S A0 06
B 56 3 S A0 3
Bread NC SAE0208 S I00
B I001
e
YC 00
Laboratory yj 331 2
Win
B I004 89
y m2 9 B m13 445
Clinical NC jm12 44 yj YC3 0
YC 42 NC 140 81
3
S.paradoxus W 570 yjm C35
yjm I015
NA
12 NCY1273
WI0144 yjm C3448
SP01 1 NCY1342
Lineage WI0041 yjm 3449
WI005 NCYC
BE088 WL004
WI002
Beer 1 BE024 yjm1418
M
WI009 yjm1447
Britain SP002 spar
WI013 BE054
US BE033 BE049
WI001 BE101
SP007 BE037
Belgium/Germany SP008 BE048
WI008 BE03
Mixed 6 BE01 6
BE00 1 BE0 0
SP0029 BE0 79
Wine BE0 28 BE0 89
E
B 05 0 BE 64
Beer 2 BE0009 BE 100
SP 038 BE 047
BE 003 WI 009
West Africa (WA) SP 061 B 012
B E07
BE 0234 B E05 0
Mix
tain
Asia BE 00 2 BEE05 3
BRR00 5 B 0 6
B E0 55
ed
B E02 01
Bri
North America (NA) B R0 05 B E0 69
B L0 07 B E0 01
BE E0 99
Malaysia (M) W L0 06 04 50
W L0
B E0 03
B E0 066
5
E0 94
W
B E0 72
B E 60
B R0
BE E09 74
B E0 35
90
B
B 01 3
B 0 7
B E04 7
BEE05 9
Mosaic
BEE00 6
B E01 2
BE 0418
B E10 8
B 09
BE 012
BEE007
BE 073
B 071
BE 087
BE 067
BE 015
BE 044
BE0043
BE 068
BE0 81
BE 82
SP0016
BE0 5
BE02 5
BE0651
BE0226
BE0 6
BE09
BE077
BE097
BE076
BE095
BE075
BE052
BE078
BE042
BE031 BE058
1
er
Be
Be
lgiu
m/G US
erma
ny
0.005
B C
1.00
0.3
Britain
0.50 US
K=2 Belgium/Germany
Estimated ancestry (Q)
0.00 Beer 2
1.00 0.2 Mixed
Wine
Asia
PC2 (9.9%)
0.50
K=4 Mosaic
NS
0.00
1.00
0.1
0.50
K=6
0.00 0.0
1.00
0.50
K=8
0.00
-0.1
Belgium/ −0.15 −0.10 −0.05 0.00 0.05 0.10

Britain US Mixed Wine Beer 2 Asia Mosaic NS
Germany
PC1 (11.7%)
Figure 1. Phylogeny and Population Structure of Industrial S. cerevisiae Strains

(A) Maximum likelihood phylogenetic tree of all S. cerevisiae strains sequenced in this project supplemented with a representative set of 24 previously sequenced
strains (Liti et al., 2009; Strope et al., 2015) and using Saccharomyces paradoxus as an outgroup. Black dots on nodes indicate bootstrap support values < 70%.
Color codes indicate origin (names) and lineage (circular bands). The basal splits of the five industrial lineages are indicated with colored dots. Mosaic strains
identified in this study are indicated with black dots next to the strain codes. Branch lengths reflect the average number of substitutions per site. Scale bar, 0.005
substitutions per site.
(B) Population structure identified in the 157 surveyed strains. The vertical axis depicts the fractional representation of resolved populations (colors) within each
strain (horizontal axis, strains listed in Figure S1C) for K = 2, 4, 6, and 8 assumed ancestral populations (where K = 8 maximizes the marginal likelihood and best
explains the data structure). Mosaic strains (i.e., strains that possess < 80% ancestry from a single population) are visualized as a separate group.
(C) Principal component projection, using the same set of SNPs as in Figure 1B. Colors represent different populations. WA, West Africa; NA, North America; M,
Malaysia; NS, not specified.
See also Figure S1 and Tables S1, S2, and S8.
Cell 166, 1397–1410, September 8, 2016 1399

Table 1. Genetic Diversity within Each Subpopulation of Industrial S. cerevisiae Strains
Subpopulation Number of Strains Analyzed Sites Segregating Sites p Qw
Britain 26 12,018,937 101,881 3.13E-03 1.88E-03
United States 10 11,973,239 72,559 2.31E-03 1.72E-03
Belgium/Germany 18 12,017,007 108,560 3.12E-03 2.19E-03
Mixed 17 12,043,532 132,188 4.35E-03 2.69E-03
Wine 24 12,052,956 114,133 1.59E-03 2.15E-03
Beer 2 21 12,063,361 142,745 2.95E-03 2.77E-03
Asia 10 12,035,745 99,879 2.39E-03 2.36E-03
The number of strains per subpopulation, the amount of analyzed and segregating sites, as well as nucleotide diversity (p) and population mutation rate
(Watterson’s q, qw) are indicated.
(Figure 1B). Moreover, the high nucleotide diversity within each incidence of polyploidy and aneuploidy (R2 0.14, p < 0.001;
of the Beer 1 sublineages exceeds that within the Wine popula- average genome content of 3.52, SD = 0.67, Figures 2A, 2F,
tion, suggesting that the split did not happen recently (Table 1). and 2G), which is linked to extensive chromosomal loss and
Compared to Beer 1, Beer 2 is more closely related to the general genome instability (Sheltzer et al., 2011).
Wine lineage and includes 20.6% of all brewing strains. Howev- CNVs are not uniformly spread across the genome. Consid-
er, in contrast to the Beer 1 group, the Beer 2 lineage lacks ering subtelomere lengths of 33 kb (Brown et al., 2010), on
geographic structure and contains yeasts originating from average 39.7% of subtelomeric nucleotide positions are affected
Belgium, the United Kingdom, the United States, Germany, by CNV events compared to 9.54% of non-subtelomeric nucle-
and Eastern Europe. The presence of two major genetically otide positions (4.1-fold difference, Wilcoxon signed-rank test,
distinct sources of beer yeasts hints toward two independent p < 0.001). However, not all subtelomeres are equally prone to
European domestication events, one of which is at the origin of CNV: most variability is detected in ChrI, ChrVII, ChrVIII, ChrIX,
both the Wine and Beer 2 clade. ChrX, ChrXII, ChrXV, and ChrXVI (Figure 2A). Gene ontology
(GO) enrichment analysis reveals that genes involved in nitrogen
Remarkable Structural Variation in Beer Yeasts and carbon metabolism, ion transport, and flocculation are
Variation in genome structure, such as polyploidy, aneuploidy, most heavily influenced by CNVs (Table S3), which is in line
large segmental duplications, and copy-number variations with previous results (Bergström et al., 2014; Dunn et al.,
(CNVs), have repeatedly been found in association with domes- 2012). Interestingly, some CNVs seem linked to specific environ-
tication and adaptation to specific niches in experimentally ments (Table S4), suggesting that CNVs may underlie niche
evolved microbes (Bergström et al., 2014; Borneman et al., adaptation. For example, many genes involved in uptake and
2011; Dunham et al., 2002; Dunn et al., 2012; Pavelka et al., breakdown of maltose (present in saké medium, main carbon
2010; Rancati et al., 2008; Selmecki et al., 2009; Voordeckers source in beer, but absent from grape must) are amplified in
et al., 2015) and in association with domestication of higher beer and saké-related subpopulations, while they are often lost
organisms (Purugganan and Fuller, 2009). in strains from the Wine subpopulation (false discovery rate
Sequencing the yeast strains in their natural ploidy allowed [FDR] q value < 0.001).
analysis of gross chromosomal rearrangements and aneu-
ploidies (Figure 2A). We detected a staggering 15,288 deletion Relaxed Selection on Sex and Survival in Nature
and amplification events across all strains, covering on average Apart from selection for industrial traits, domestication is also
1.57 Mb per strain. The size of the regions ranges from complete characterized by relaxed selection and potential loss of costly
chromosomes (resulting in aneuploidies) to small local variations traits that are not beneficial in the man-made environment (Dris-
of a few kilobases (kb), all of which we will refer to as ‘‘CNVs.’’ coll et al., 2009). In order to chart the phenome of our collection
The extent of deletions significantly exceeds that of amplifica- and investigate signs of selection for some traits and loss of
tions, respectively 1.07 Mb and 0.50 Mb on average per strain others, 82 phenotypes, such as aroma production, sporulation
(2.15-fold difference, Wilcoxon signed rank test, p < 0.001). We characteristics, and tolerance to osmolytes, acids, ethanol,
observed significant variation among strains originating from and low and high temperatures, were measured in all strains (Fig-
different industries in the total frequency of CNV events (ANOVA ures 3A and S3; Table S5). Hierarchical clustering of the pheno-
F test, p < 0.001) and the fraction of the genome affected types resolves the main phylogenetic lineages and reveals a
(ANOVA F-test, p < 0.001) (Figure S2). Pairwise comparisons of moderate correlation between genotype and phenotype dis-
subpopulations and industries show no significant differences tances between strains (Spearman correlation 0.33), which is
in the load of amplifications between strains from different indus- further increased (Spearman correlation 0.36) when mosaic
tries or subpopulations, but we detected significant differences strains, for which genetic distance has no straightforward evolu-
in the load of deletions between strains from the wine (median = tionary interpretation, are omitted (Figure 3A). Moreover, the
0.51 Mb) and beer (median = 0.94 Mb) industry (Tukey honest clustering splits the collection into two main phenotypic sub-
significant difference [HSD], p < 0.05) (Figures 2B–2E). This groups: one largely overlapping with the Beer 1 clade that con-
high incidence of CNV in beer strains goes together with a high tains the majority of the Belgium/Germany, United States, and
1400 Cell 166, 1397–1410, September 8, 2016

A Origin
1 2 3 4 5
Beer
I II III IV V VI VII VIII IX X XI XII XIII XIV XV XVI
WI002
WL004
Wine
BI004
BI003 Spirits
Asia
BI001
SA001
SA006 Saké
Wild
SA003
SA005
SA007
Bioethanol
SA004
WL002
WL003
Bread
WL001
LA001
SP010
WI016
BE018
LA002
Laboratory
BE002
WI019 S.paradoxus
BE004
BE027
BE085
BE040 >= 2-fold amplification
Beer 2
BE011
BE039 Amplification
BE003 < 2-fold amplification
BE086
BE013
BE034
BE084
BE080
BE021 1 =< deletion < n
BE032 Deletion
BE062
SP006 complete deletion
SP004
BI005
BI002
BE014
BE030
WI014
WI007
WI018
WI010
WI017
WI003
WI006
SA002
Wine
WI011
WI015
BE020
WI004
SP011
WI005
WI009
BE088
BE024
WI001
SP002
WI013
BE033
SP008
SP007
WI008
BE006
SP001
BE028
BE029
BE005
Mixed
BR004
BE023
BE038
SP003
BE025
BR001
BR003
BR002
WL007
WL005
BE074
BE093
BE017
BE046
BE008
BE041
BE087
BE073
Bel/Ger
BE012
BE015
BE043
BE081
BE016
SP005
BE022
BE026
BE077
BE076
BE075
BE078
BE031
BE058
BE042
BE052
BE097
BE096
BE095
BE051
BE065
BE071
BE067
US
BE007
BE102
BE098
BE082
BE068
BE044
BE060
BE035
BE019
BE057
BE066
BE094
BE090
BE099
BE050
BE045
BE069
BE001
BE055
BE056
Britain
BE053
BE047
WI012
BE009
BE089
BE079
BE036
BE010
BE100
BE064
BE048
BE037
BE101
BE054
BE049
1 2 3 4 5
Estimated
ploidy (n)
B C
Beer Bioethanol Bread Saké Spirits Wild Wine Britain US Bel/Ger Beer 2 Mixed Wine Asia Mosaic
2.5 2.5
Amplifications (Mb)
Amplifications (Mb)
2 2
1.5 1.5
1 1
0.5 0.5
0 0
D E
Beer Bioethanol Bread Saké Spirits Wild Wine Britain US Bel/Ger Beer 2 Mixed Wine Asia Mosaic
Deletions (Mb)
Deletions (Mb)
6 6
4 4
2 2
0 0
F G Britain
Estimated ploidy (n)
Estimated ploidy (n)
Beer
Bioethanol US
4 4
Bread Bel/Ger
Lab Beer 2
3 Saké 3 Mixed
Spirits Wine
Wild Asia
2 Wine 2
R2 = 0.14, p < 0.001 R2 = 0.14, p < 0.001 Mosaic
0 2 4 6 8 0 2 4 6 8
CNV load (Mb) CNV load (Mb)
Cell 166, 1397–1410, September 8, 2016 1401

Britain beer yeasts as well as mosaic strains containing major Selection for Industrial Phenotypes
genome fractions of these subpopulations and a second one A key hallmark of domestication is phenotypic adaptation to
where the remaining genetic subpopulations are strongly over- artificial, man-made niches and accentuation of traits desirable
represented (Fisher’s exact test, Bonferroni corrected p < for humans. Phenotypic evaluation of the strains for industrially
0.001). Overall, strains from the Beer 1 clade perform poorly in relevant traits (including aroma production, ethanol production,
general stress conditions that are not usually encountered in and fermentation performance) shows that many strains harbor
the brewing environment (Figure S3; Table S6). In contrast, phenotypic signatures linked to their industrial application.
strains from the Wine subpopulation show superior performance The ability to accumulate high concentrations of ethanol, for
in general stress conditions, which likely reflects the high-sugar example, seems tightly linked to industrial niche. Beer 1 strains
and high-alcohol environments encountered in wine-making, as typically generate only 7.5%–10% v/v of ethanol, while strains
well as survival in potentially nutrient-poor and harsh natural en- used for the production of high-alcohol products like saké,
vironments in between the grape harvest seasons. spirits, wine, and especially bioethanol, can produce up to
Saccharomyces cerevisiae is a facultative sexual organism. 14.5% v/v (Figure 3B; Table S6).
While its main mode of reproduction is clonal, sporadic sporula- With the exception of a few wine yeast characteristics (see
tion can help to survive periods of stress (Briza et al., 1990). It has earlier) (Figures 3C and 3D), it remains unclear whether genetic
been shown that in yeast, sexual reproduction is beneficial when and phenotypic variation between S. cerevisiae lineages is pri-
adapting to new, harsh niches, but plays a lesser role in more marily caused by human-driven selection and domestication,
favorable environments (Goddard et al., 2005; McDonald et al., or if neutral genetic drift or non-human selection are involved.
2016). Our data show that there are large systematic differences To assess this further, we compared the phenotypic behavior
in the reproductive lifestyle of yeasts inhabiting different indus- of different subpopulations for two industrially relevant traits
trial niches: 44.4% of the Beer 1 population is obligate asexual, for which the genetic underpinnings are largely known, namely
while this trait ranges between 0% and 21% in the other popula- maltotriose fermentation and the production of 4-vinyl guaiacol
tions (Figure 4A) and is absent in wild strains. Furthermore, over (4-VG), the main compound responsible for phenolic off-flavors
80% of the non-mosaic Beer 1 strains that are able to sporulate (POF). Beer yeasts show a significantly higher capacity to
show little or no spore viability (Figure 4B). Additionally, beer metabolize maltotriose, a carbon source specifically found in
yeast lineages generally show a high level of heterozygosity, beer medium (Figure 3E; Table S6). Efficient utilization of malto-
especially Beer 1. Compared to the Wine clade for example, triose correlates with the presence of a specific allele (AGT1)
strains from the Beer 1 and Beer 2 clade have on average of the sugar transporter MAL11, known to show high affinity
5.10-fold (Tukey HSD, p < 0.001) and 2.04-fold (Tukey HSD, for maltotriose (phenotypic variability explained by SNPs in
p = 0.06) more heterozygous sites, respectively (Figures 4C, MAL11 77.40%, SE 0.5%). This allele is only present in Beer
S4A, and S4B). The lack of genetic admixture suggests that 1 subpopulations and some mosaic strains, while the complete
this heterozygosity was acquired during long periods of asexual MAL1 locus (including the MAL11 gene) is absent in the Wine
reproduction, rather than through outbreeding. Further analysis subpopulation (Table S7). Interestingly, strains of the Beer 2 sub-
of the correlation between sexual lifestyle and genome structure population are generally able to ferment maltotriose but contain
shows that spore viability is weakly anticorrelated with the het- various frameshift mutations in MAL11 and show a reduced
erozygosity level (R2 0.17; p < 0.001) and the fraction of the CNV for the complete MAL1 locus, suggesting that other, yet un-
genome associated with large (>20 kb) amplifications and dele- known mechanisms facilitate maltotriose uptake in this lineage,
tions (R2 0.16; p < 0.001), while sporulation efficiency is only and maltotriose metabolism evolved convergently in the Beer 1
significantly anticorrelated with the latter (R2 0.19; p < 0.001) and Beer 2 lineages.
(Figures 4D–4G). Yeasts used for the production of alcoholic beverages ideally
Together, this indicates that the genome of beer yeasts, but should not produce undesirable aromas. Although tolerated in
not wine yeasts, show signs of decay and loss of survival skills some specialty beers, the presence of 4-VG, a compound with
outside a specific man-made environment, probably caused a spicy, clove-like aroma, is generally undesired in saké, wine,
by their long (estimated >75,000 generations) and uninterrupted and most beer styles. Two genes, phenylacrylic acid decarbox-
growth in rich medium. ylase (PAD1) and ferulic acid decarboxylase (FDC1), both
Figure 2. Ploidy and Copy-Number Variation in Industrial S. cerevisiae Strains

(A) Genome-wide visualization of copy-number variation (CNV) profiles, with the aggregate profile across all strains depicted on the top. Estimates for the nominal
ploidy (n) values of the strains are represented by a bar chart next to the strain codes. Heat map colors reflect amplification (red shades) or deletion (blue shades)
of genomic fragments. A distinction is made between completely deleted fragments (dark blue) and fragments of which at least one copy is still present (light blue).
Similarly, highly amplified fragments (copy number R2-fold the basal ploidy) are depicted in dark red, while low and moderately amplified fragments (copy
number <2-fold the basal ploidy) are depicted in orange. For strains with no estimated ploidy available, colors are only indicative of the presence of amplifications
(orange) or deletions (light blue). Roman numbers indicate chromosome number. Strains are clustered according to their genetic relatedness as determined in
Figure S1A. Origin (name colors) and population (colored rectangles) are indicated on the figure.
(B–E) Violin plots describing the density of amplifications and deletions across different industries and subpopulations. Triangles indicate the median within each
group.
(F and G) Correlations between levels of CNV load (Mb) and estimated ploidy (n), by industry and subpopulations.
See also Figure S2 and Tables S3 and S4.
1402 Cell 166, 1397–1410, September 8, 2016

A
SP006
BR001
BR003
BR002
BR004
SP003
LA001
SP007
BE038
BE023
WL002
WL003
WL001
BE006
BE014
BE005
BE028
BE029
BI001
BI003
BI004
BE102
BE034
BE025
WI002
WL004
WI016
BI005
BI002
SP004
WI015
WI006
WI014
WI018
WI013
WI010
WI007
WI005
WI011
WI003
SP001
WI009
SP002
WI001
BE033
BE024
SP011
WI004
WI008
LA002
BE018
BE020
WI017
BE086
BE080
BE021
BE011
BE013
BE004
BE032
BE027
BE039
BE040
BE091
BE059
BE063
BE062
BE002
BE003
BE083
BE084
SA002
SA003
SA006
SA001
SA004
SA005
SA007
SP009
SP010
WL007
WL006
WL005
WI019
BE085
BE092
SP008
BE061
SP005
BE075
BE073
BE087
BE012
BE035
BE041
BE078
BE081
BE077
BE030
BE043
BE022
BE026
BE031
BE015
BE100
BE099
BE101
WI012
BE009
BE037
BE036
BE001
BE010
BE008
BE090
BE050
BE045
BE079
BE070
BE064
BE055
BE058
BE049
BE066
BE069
BE048
BE054
BE056
BE047
BE046
BE044
BE053
BE065
BE051
BE057
BE067
BE068
BE071
BE082
BE072
BE074
BE093
BE052
BE088
BE089
BE094
BE076
BE016
BE060
BE017
BE019
BE007
BE098
BE042
KCl (500mM) z-score
KCl (1000mM) 1.5
KCl (1500mM)
NaCl (250mM)
NaCl (500mM)
NaCl (1000mM) 0.5
4°C
16°C
30°C
40°C -0.5
Ethanol (5%)
Ethanol (7%)
Ethanol (9%)
Environmental stress
Ethanol (10%) -1.5

Ethanol (11%)
Ethanol (12%) No Data
Sulfite (1.50mM)
Sulfite (2.25mM)
Sulfite (3.0mM)
Glucose (46%)
Sorbitol (46%)
Glucose (48%)
Sorbitol (48%)
Glucose (50%)
Sorbitol (50%)
Copper (0.075mM)
Cadmium (0.3mM)
Cadmium (0.4mM)
Cadmium (0.5mM)
Actidione (0.2 mg L-1)
Acetic acid (50mM)
Acetic acid (75mM)
Acetic acid (100mM)
Levulinic acid (25mM)
Formic acid (25mM)
Formic acid (50mM)
Formic acid (75mM)
Glycerol (20°C)
Melibiose (20°C)
Sorbitol (20°C)
Ethanol (20°C)
Galactose (20°C)
Nutrient stress
Fructose (20°C)
Sucrose (20°C)
Maltose (20°C)
Maltose (liquid)
Maltotriose (liquid)
Glucose (10°C)
Ethanol (10°C)
Fructose (10°C)
Sucrose (10°C)
Maltose (10°C)
Glucose (39°C)
Ethanol (39°C)
Fructose (39°C)
Sucrose (39°C)
Maltose (39°C)
Lysine
Ethyl propionate
Ehtyl butyrate
Ethyl hexanoate
Ethyl octanoate
Ethyl decanoate
Aroma
Propyl acetate
Isobutyl acetate
Ethyl acetate
Isoamyl acetate
Phenyl ethyl acetate
Phenyl ethanol
Butanol
Isoamyl alcohol
Isobutanol
POF
Acetaldehyde
Other
Flocculation
Ethanol production
Sporulation efficiency
Spore viability
Subpopulation
B C
Britain US Bel/Ger Beer 2 Mixed Wine Asia Mosaic Britain US Bel/Ger Beer 2 Mixed Wine Asia Mosaic
Ethanol production (% v/v)
15.0 100
Copper tolerance (au)
12.5 75
10.0 50
7.5 25
5.0 0
D E
Maltotriose fermentation (au)
100 100
Sulfite tolerance (au)
75 75
50 50
25 25
0 0
Figure 3. Trait Variation of Industrial S. cerevisiae Strains

(A) Heat map representation of phenotypic diversity within industrial S. cerevisiae strains. Phenotypic values are calculated as Z scores (normalized values) and
colored according to the scale on the right. Missing values are represented by gray shadings. Strains are hierarchically clustered based on phenotypic behavior.
Strain names are colored according to geographical origin, as in Figure 1A. The corresponding subpopulation of each strain is indicated by the colored bar below
the figure, according to the color code of Figure 1B.
(B) Ethanol production (depicted as % v/v1) of all strains from different subpopulations in fermentation medium containing 35% glucose.
Cell 166, 1397–1410, September 8, 2016 1403

located as a cluster in the subtelomeric region of ChrIV, control from a wine strain (4-VG production) that is only desirable in spe-
4-VG production. PAD1 encodes a flavin prenyltransferase that cial beer styles.
catalyzes the formation of a flavin-derived cofactor, which is
required by Fdc1 for decarboxylation of the precursor ferulic Creating Superior Hybrid Yeasts through Marker-
acid (White et al., 2015). Pad1 and Fdc1 help to detoxify phenyl- Assisted Breeding
acrylic acids found in plant cell walls (Mukai et al., 2010). There- Apart from yielding insight into the origins of today’s industrial
fore, it would be expected that, unless there is counterselection, yeasts, our results also open new routes for the creation of new su-
activity of these genes is retained. Interestingly, phenotypic perior strains. The availability of genomic data and the increasing
profiling reveals that many industrial strains have lost the ability number of polymorphisms that are known to contribute to indus-
to produce 4-VG, while it is generally retained in wild strains, trially relevant phenotypes enables rapid DNA-based selection of
as well as in bakery and bioethanol strains (Figure 5B). In superior segregants and hybrids in large-scale breeding schemes.
these cases, 4-VG production is likely less detrimental, either Such marker-assisted breeding is already intensively used for
because the flavor disappears during baking, or the product is crop and livestock breeding, because it circumvents labor-inten-
not destined for consumption. Sequence analysis shows that sive and time-consuming phenotyping. As proof-of-concept, we
many industrial strains, especially beer and saké strains, ac- combined our genomic and phenotypic data to obtain new hy-
quired loss-of-function mutations (SNPs and/or frameshift In- brids with altered aromatic properties using marker-assisted
Dels) in PAD1 and/or FDC1, while this was never observed in breeding. Specifically, a 4-VG producing beer strain harboring a
strains from natural environments or bioethanol production heterozygous loss-of-function mutation in FDC1 (strain BE027)
(Figure 5A). Moreover, different sublineages acquired different was selected and sporulated to obtain segregants. Next, the
disruptive mutations, hinting to the presence of diverse conver- FDC1 allele of the segregants was genotyped using mismatch
gent adaptive strategies in response to human selection against PCR. Two segregants, one harboring the loss-of-function allele
4-VG production. and one harboring the functional allele, were crossed with segre-
To investigate the origin and the maintenance of the pheno- gants of SA005, an Asian saké strain with a homozygous non-
typic diversity in 4-VG production, we used Bayesian inference functional FDC1 allele, resulting in hybrids with good beer fermen-
to reconstruct the ancestral phenotypic state in the two key tation characteristics but drastically different aroma profiles
genes PAD1 and FDC1 using BEAST (Drummond et al., 2012) (4-VG+ versus 4-VG) that suit specific beer styles (Figure 5D).
(Figure 5C). Shifts from 4-VG+ to 4-VG and vice versa occurred
frequently after the initial split from S. paradoxus. In both the Domestication Predates Microbe Discovery
PAD1 and the FDC1 trees, an early subclade containing most Despite its wide use in industry and as a model organism, little is
Beer 1 strains acquired loss-of-function mutations at the base known about the ecology and evolutionary history of S. cerevisiae.
of the clade, suggesting that already very early during domesti- Moreover, because early brewers, winemakers, and bakers were
cation of the Beer 1 lineage, a 4-VG variant was derived from unaware of the existence of yeast, there is no record of how yeasts
the 4-VG+ ancestor. Several other loss- and gain-of-function made their way into these processes, nor how yeasts were prop-
mutations occurred across both trees, most notably the loss- agated and shared. As a result, it has proven difficult to estimate
of-function mutation in FDC1 of the Asian saké (but not bio- when specific industrial lineages originated. Moreover, current
ethanol) strains. demographic and molecular clock models of S. cerevisiae employ
Interestingly, a strong incongruence between single gene the experimentally determined mutation rate of the haploid lab
trees and the strain phylogeny is present for three beer strains strain S288c in rich growth medium (Lynch et al., 2008), while it
used in the production of German Hefeweizen beers (BE072, is known that the mutation rate is heavily influenced by the genetic
BE074, and BE093). Hefeweizen (wheat) beer is a traditional background (Filteau et al., 2015), ploidy (Sheltzer et al., 2011),
German beer style and one of the few styles where a high growth speed (van Dijk et al., 2015), and environmental stress
4-VG level is desirable because it contributes to the typical (Voordeckers et al., 2015), factors that are likely very different
smoky, spicy aroma of these beers. Phylogenetically, Hefewei- for industrial, wild, and lab yeasts. However, our dataset, and spe-
zen yeasts cluster within the Beer 1 lineage, but they are shown cifically the Beer 1 clade, provides a strong tool for dating beer
to be highly mosaic, containing genomic fragments of all three yeast divergence. First, given the absence of a functional sexual
Beer 1 subclades (mainly from Belgium/Germany). Only a small cycle and lack of admixture, exclusively clonal reproduction can
fraction (8%–13%) of the genome originates from the Wine be assumed. Second, our data show that United States beer
subpopulation, but this fraction includes the subtelomeric region yeasts are related closest to European beer yeasts, suggesting
of ChrIV, containing a functional PAD1 and FDC1 allele. This that they were imported from Europe during colonization, rather
suggests that hybridization between different domesticated sub- than stemming from indigenous wild United States yeasts. More
populations yielded variants combining the typical traits of beer specifically, United States beer yeasts seem phylogenetically
yeasts, including maltotriose fermentation, with a particular trait most closely related to British beer yeasts (Figure 1A), which is
(C) Growth of all strains from different subpopulations on medium supplemented with 0.075 mM copper, relative to growth on medium without copper.
(D) Growth of all strains from different subpopulations on medium supplemented with 2.25 mM sulfite, relative to growth on medium without sulfite.
(E) Growth of all strains from different subpopulations in medium containing 1% w v1 maltotriose as the sole carbon source, relative to growth on medium with
1% w/v1 glucose. au, arbitrary units; Bel/Ger, Belgium/Germany.
1404 Cell 166, 1397–1410, September 8, 2016

A B
100 100
Sporulation efficiency (%)
Spore viability (%)

75 75
50 50
25 25
0 0
C
I II III IV V VI VII VIII IX X XI XII XIII XIV XV XVI
1.00
Britain
0.50
0.00
1.00
US
0.50
0.00
1.00
Bel/Ger
Level of heterozygosity
0.50
0.00
1.00
Beer 2
0.50
0.00
1.00
Mixed
0.50
0.00
1.00
Wine
0.50
0.00
1.00
Asia
0.50
0.00
D E
80 R2 = 0.17, p < 0.001 80 R2 = 0.03, p = 0.022
Nr. of heterozygous SNPs (x103)
Nr of heterozygous SNPs (x103)
Britain Britain
60 US 60 US
Bel/Ger Bel/Ger
Beer 2 Beer 2
40 40
Mixed Mixed
Wine Wine
20 Asia 20 Asia
Mosaic Mosaic
0 0
0 25 50 75 100 0 25 50 75 100
Spore viability (%) Sporulation efficiency (%)
F G
Amplified/Deleted genome fraction (%)
Amplified/Deleted genome fraction (%)
R2 = 0.16, p < 0.001 R2 = 0.19, p < 0.001

Britain Britain
60 60
US US
Bel/Ger Bel/Ger
40 Beer 2 40 Beer 2
Mixed Mixed
Wine Wine
20 20
Asia Asia
Mosaic Mosaic
0 0
0 25 50 75 100 0 25 50 75 100
Spore viability (%) Sporulation efficiency (%)
Figure 4. The Reproductive Lifestyle of Industrial S. cerevisiae Strains

(A) Violin plots depicting sporulation efficiency of all strains from different subpopulations.
(B) Violin plots depicting spore viability of all sporulating strains from different subpopulations.
(C) Visualization of the level of heterozygosity across the genome of the different subpopulations, calculated as the ratio of heterozygous/homozygous SNPs in
10 kb windows.
(D–G) Scatter plots depicting the correlation between the number of heterozygous loci and spore viability (D) or sporulation efficiency (E), and the correlation
between the fraction of the genome subjected to large (>20 kb) structural variation and spore viability (F) or sporulation efficiency (G). Dot colors indicate
subpopulations similar to the color code of Figure 1B.
See also Figure S4.
confirmed by the average per-site nucleotide divergence (dxy), signed-rank test, p < 0.001) (Table S8). This suggests that the
which is significantly lower between Britain and the United States origin of the United States brewing strains can be traced back
(average dxy = 1.97 3 103) than between Belgium/Germany and to the introduction of beer culture in the United States by early
the United States strains (average dxy = 2.26 3 103) (Wilcoxon 17th century British settlers (Van Wieren, 1995). Third, in contrast
Cell 166, 1397–1410, September 8, 2016 1405

A B
PAD1 FDC1
100% 4-VG- 100%
c.495_496insA
4-VG-
4-VG+
4-VG+
c.864delA
75% 75%
W102*
W497*
Q154*
R309*
Q86*
Y98*
K54*
BE049
BE054
50% 50%
BE101
BE037
BE048
BE064
BE100
25% 25%
BE010
BE036
BE079
BE089
BE009
WI012
0% 0%
Wine
Spirits
Saké
Wild
Bioethanol
Bread
Laboratory
Beer
BE047
Britain
US
Beer 2
Mixed
Wine
Asia
Mosaic
Bel/Ger
BE053
BE070
BE056
BE055
BE001
BE069
BE045
BE050
BE099
BE090
BE094
BE066
BE057
C
BE019
BE035
PAD1 FDC1
BE060
BE068
BE044
BE082
WL003
SP006
WI004
WI011
B E076
BE086
WI015
WI005
WI018
BE081
WI005
SP010
BE031
WI017
BE058
BE078
WI014
WI016
7
BE01
BE061
BE098
WI007
1
6
WI01
LA00
SP00
SP00 8
BE02
BE06
SP00
8
BE01
SP00 3
8
BE03
WI0
WI00
BI0 4
1
WL 02
BE03
SP00 9
BE102
WI0 3
BE 02
BR 09
SP00
WI0 17
8
005
BR 01
SP 88
2
BE 03
05
01
00
BE007
0
4
4
BE 02
BE 001
09
03
WL 003
SP
0
WI0 2
BE 06
SA 021
W 005
0
SP
BE 014
0
W 028
BE067
BE 007
BE 080
SP
0
BE 011
BR 007
0
00
BE 084
BE 07 2
SP 004
B 001
L
BE 07
BE071
BE 010 3
4
L
BR 00 3
6
SP 06 6
B E09 3
BE E02
BR E06 8
BE065
9
B E07 9
BE 00
0
B 02
E0 2
B I00
00 7
W
BE051 B
43
2
B
B E03
W
L
3
3
2 BE E03
B
BE095 W E08 2 02 5 00 9
R
L 3 BE E07 1 BE 026 0 BE I01 6
BE096 BE 003 W I00
B 08 BE 077
BE 062 BE I002 BE 022 W 007
BE097 BE 030 SA 059
W 003 BE 075
BE052 BE 017 SA 059 0 BE 003
BE 7
BE042 BE 027 BE 001 0
8 SA 001
00 BE 58 SA 004
BE058 BE 4 SA 001 0
0 LA 05 BE 76 SA 05
BE 20 03
SA
0
BE031 0 0
SA 6 BE 1
006
BE 11 00 01
SA
BE078 03 SA SP00 6
BE04 9 4
SA00 7 5 BI0
03
BE075 BE0
WI0 0 SA00 BE03
04 BI004
BE076 11 1
BE07 BI00
1 9 BI00
BE04
BE077 BE08
4
WL0
01 0 WL002
6 BE011
BE026 BE043 BI003 WL001
WI014 BE023
BE022 WI010 WL002
WI007 LA001
SP005 BR002 BI004
spar BE020 spar
BE016 SP011
BI002 WI018 BI002
BE081 WI006
BE096 WI010 BE094
BE043 BE034 BE036
BE036 WI013
BE015 BE003 BE037 WI012
BE088
BE091 WI009 BE099
BE012 BE057
SP007 BE049 WI001 BE079
BE073
BE002 BE045 BE033
BE070
BE087 BE01
4 BE08 4 BE06
BE041 3 2 BE02 6
BE09 BE05 2 BE04
SP00 7
BE008 WI0
04 BE0 0 83 BE0
4 55 BE0 9 64
BE01 BE01
BE046 BR00 5 BE02
08 BE 0 BE 0
BE017 BE 097 17
WI0 13
051
005 BE BE
BE093 WL 15 BE 0 6 0
0
BE 27 BE
090
WI0 72 1
BE 00 0
BE 06 BE 054
BE074 0
BE 13 BE 035 0
WL 025 BE 089
BE072 0
BE 03 BE 067 BE 009
I0 6 BE 062
WL006 W BE 056
BE 021 BE 053
00 B 04 B 04
WL005 SP I019 BE E01 1 BE 091 4
W 092 3 BE E05 8
WL007 W 08 5 BE 08 2 BE 04 0
BE E02 5 I0 9
12 BE 03 00 5
BR002 B 02 BE
BE 00 8
1
B E06 4
B E06 0
SA 02
E
E0 9
E0 8
8
BE 080 2
B
B 06
BR003
B 10 5
LA 01
56
46
W I01 2
B
BE 09
BE 05
BE
W 00
6
BE 006
BR001
BE 095
BE 015
SP 004
BE 102
BE 034
BE 067
4
BE 085
BE 069
BE025
BE 102
0
L
4
SP 005
BE 42
BE 05
0
BE 097
BE 08
BE 048
BE 19
SP003
BE 95
0
BE 01
BE
BE 60
BE
BE 82
0
BE 47
BE
057
0
BE
0
BE
BE09 4
BE0 5
0
BE038
BE
9
BE0
87
0
0
04
03
BE0 6
BE01
012
1
2
0
BE09
BE09
6
087
8
BE07
BE04
BE04
BE05
037
8
BE
BE066
BE061
BE009
BE073
BE10
BE04
BE070
BE09
BE007
BE012
BE071
BE053
BE008
B E079
SP008
BE07
BE00
BE065
BE054
BE052
BE068
BE051
65
BE101
BE007
BE098
BE042
44
9
BE023
0
3
9
1
BR004
SP009
BE005
BE029 6.0E-4 5.0E-4
BE028
SP001
BE006
WI008
D
SP007
BE027
SP008
BE033
WI013
SP002
WI001
BE024
BE088
WI009
WI005
SP011
WI004
BE020
WI015 (1)
WI011
SA002
WI006
WI003
WI017 Mating type a Mating type α
WI010
WI018
WI007
WI014
BE030
BE014 (1)
BI002
BI005
SP004
SP006
BE091
BE062
BE032
BE059
BE021
BE080
BE084
BE083
BE034
(2)
BE013 SA005
BE086
BE092
BE003
BE063 120
BE039 BE027-a SA005-α Hybrid 1
BE011
4-VG production (au)
BE040 100
BE085
BE027
x
BE004
WI019 80
BE002 SA005-a BE027-α Hybrid 2 Parental
LA002
BE018
WI016 x
60 Haploid
SP010 (3) Hybrid
LA001
WL001 40
WL003 BE027-a BE027-α Hybrid 3
WL002
SA004 20
SA007 x
SA005
SA003
SA006
0
SA001 SA005-a SA005-α Hybrid 4
BE027
BE027-a
BE027-α
SA005-a
SA005-α
Hybrid 1
Hybrid 2
Hybrid 3
Hybrid 4
SA005
BI001
BI003
BI004 x
WL004
WI002
spar
Figure 5. Production of 4-Vinyl Guaiacol by Industrial S. cerevisiae Strains

(A) Distribution of loss-of-function SNPs and frame-shift mutations in FDC1 and PAD1 of industrial S. cerevisiae strains. Gray boxes indicate the presence of the
loss-of-function mutation, diagonal bars indicate heterozygosity at this site. Strains are clustered according to the strain phylogeny and strain names are colored
according to their origin. Basal splits of the five industrial lineages are indicated with colored dots (see Figure 1A).
1406 Cell 166, 1397–1410, September 8, 2016

to wine, beer is not produced seasonally but throughout the whole stocks due to human selection and trafficking. Specifically, the
year, which provides fermenting yeast with a predictable and thousands of industrial yeasts that are available today seem to
stable growth environment. Yeast cells undergo about three stem from only a few ancestral strains that made their way
doublings during one batch of beer fermentation, which takes into food fermentations and subsequently evolved into separate
1 week. Moreover, brewers typically recycle yeasts from a lineages, each used for specific industrial applications. Within
finished fermentation to inoculate a new batch, which implies each cluster, strains are sometimes further subdivided along
that beer yeasts are continuously growing in their industrial niche. geographical boundaries, as is the case for the Beer 1 clade,
Together, these facts make it possible to estimate the number of which is divided into three main subgroups. However, further
generations to be 150/year. subclustering of beer yeasts according to beer style was gener-
Based on estimates of the number of generations per year and ally not observed, which may not be surprising as it is common
the divergence time between United Kingdom and United States practice for brewers to use only one yeast strain within their
beer strains, we calculated the average mutation rate in a brewing brewery for the production of a wide array of different beers.
environment to be 1.61–1.73E-08/bp/generation. While this value Notable exceptions are yeasts associated with the few beers
differs from previous assumptions, it is similar to the measured that largely depend on very specific yeasts characteristics,
mutation rate in a diploid yeast strain that was subjected to 2 such as Hefeweizen beers. Another exception may include those
years of artificial evolution in a high-ethanol environment (Voor- beers for which production is restricted to a specific geographic
deckers et al., 2015). Moreover, mutations likely also occur in area, such as Belgian Saisons or British Stouts.
the second phase of beer fermentations, when cells are no longer We further show that industrial yeasts were clearly subjected
dividing, which implies that the mutation rate per generation in in- to domestication, which is reflected in their genomes and phe-
dustrial conditions should be higher than what is measured under nomes. Interestingly, domestication seems strongest in beer
conditions where the cells are dividing frequently, as is usually the yeasts, which demonstrate domestication hallmarks such as
case in laboratory experiments (Loewe et al., 2003). Using these decay of sexual reproduction and general stress resistance, as
data, the last common ancestor of the three major Beer 1 sub- well as convergent evolution of desirable traits like maltotriose
clades (Belgium/Germany, United Kingdom, and the United utilization. Yeasts from the Beer 1 clade show the clearest signs
States) is estimated to date from AD 1573–1604, suggesting of domestication, possibly because Beer 2 only diverged more
that domestication started around this time. Interestingly, this co- recently from other sublineages. Many of these domestication
incides with the gradual switch from home-centered beer brew- features may have simply been the result of the yeasts’ adapta-
ing where every family produced their own beer, to more profes- tion to their new industrial niches. However, for some traits, it is
sional large-scale brewing, first in pubs and monasteries and later likely that humans actively intervened, e.g., by selecting strains
also in breweries (Hornsey, 2003). The last common ancestor of that do not produce undesirable off-flavors, which our analysis
Beer 2 is estimated to be more recent, between AD 1645–1671. identifies as PAD1 or FDC1 nonsense mutants.
This suggests that beer yeast domestication started before The presence of a strong domestication signature in beer
the discovery of microbes and the isolation of the first pure yeast yeast genomes agrees well with the common practices in the
cultures by Emil Hansen in the Carlsberg brewery in 1883, but brewing industry. Beer yeasts are typically recycled after each
well after the invention of beer production, estimated to have fermentation batch, and because beer is produced throughout
occurred as early as 3000 BC (Michel et al., 1992). Although it is the year, this implies that beer yeasts are continuously growing
difficult to assess how many different yeast strains were domes- in their industrial niche. By contrast, wine yeasts can only grow
ticated and in which industrial context these domestications in wine must for a short period every year, spending the rest of
occurred, the limited number of clades of industrial yeasts and their lives in and around the vineyards or in the guts of insects
the clear segregation of wild and industrial yeasts suggests that (Bokulich et al., 2014; Christiaens et al., 2014; Stefanini et al.,
today’s industrial yeasts originated from a limited set of ancestral 2016). During these nutrient-poor periods, wine yeasts likely un-
strains, or closely related groups of ancestral strains. dergo few mitotic doublings, yet they may undergo sexual cycles
and even hybridize with wild yeasts (Stefanini et al., 2016). More-
DISCUSSION over, only a very small portion of the yeasts may find their way
back into the grape must when the next harvest season arrives,
Together, our results show that today’s industrial S. cerevisiae while trillions of cells are being transferred to the next batch
yeasts are genetically and phenotypically separated from wild during backslopping in beer production. This results in large
(B) Percentage of strains within each origin (left) and population (right) capable of producing 4-vinyl guaiacol (4-VG). Red, 4-VG; turquoise, 4-VG+.
(C) Phylogenetic trees and ancestral trait reconstruction of PAD1 and FDC1 genes. Branches are colored according to the most probable state of their ancestral
nodes, turquoise (4-VG+) or red (4-VG). Pie charts indicate probabilities of each state at specific nodes, turquoise (4-VG+) or red (4-VG); posterior probability for
the same nodes is indicated by a dot: black dot, 90%–100%; gray, 70%; white, 42%. Branch lengths reflect the average numbers of substitutions per site
(compare scale bars).
(D) Development of new yeast variants with specific phenotypic features by marker-assisted breeding. Two parent strains (BE027 and SA005) were sporulated
and, using genetic markers, segregants with the desired genotype were selected (1). Next, breeding between segregants from different parents (outbreeding) or
the same parent (inbreeding) were performed (2). This breeding scheme yields hybrids with altered aromatic properties that can directly be applied in industrial
fermentations (3). 4-VG production is shown relative to the production of BE027. Yeast genomes are represented by gray bars, loss-of-function mutations in
FDC1 as red (W497*) and blue (K54*) boxes within the gray bars. Error bars represent one SD from the mean.
See also Table S5.
Cell 166, 1397–1410, September 8, 2016 1407

effective population sizes for beer, but not wine, yeasts. The dif- SUPPLEMENTAL INFORMATION
ferences in industrial practices between beer brewing and wine-
Supplemental Information includes four figures and eight tables and can be
making likely had three important consequences. First, beer
found with this article online at http://dx.doi.org/10.1016/j.cell.2016.08.020.
yeasts evolved faster than wine strains (Figure 1A). This resulted An audio PaperClip is available at http://dx.doi.org/10.1016/j.cell.2016.08.
in a large genetic diversity within beer yeasts, while wine yeasts 020#mmc9.
are genetically more homogeneous (Table 1). Second, after the
initial domestication event, some beer yeasts were contained AUTHOR CONTRIBUTIONS
in the brewery and diverged allopatrically, leading to geograph-
ically defined subpopulations mirroring human traffic and coloni- Conceptualization, B.G., J.S., T.P., L.M., S.M., and K.J.V.; Formal Analysis,
zation. Third, beer strains generally lost their ability to reproduce B.G., J.S., L.S., M.R., A.M., and K.V.; Investigation, B.G., J.S., T.P., A.M.,
sexually. This, combined with continuous cultivation in a mild L.S., V.S., B.H.-M., and M.T.; Resources, L.M., B.S., C.T., C.W., and K.J.V.;
Writing, B.G., J.S., S.M., and K.J.V.; Supervision, T.P., T.R., A.S., C.W.,
growth environment, made them susceptible to genetic drift
G.B., S.M., and K.J.V.
and fixation of deleterious alleles that would otherwise be purged
by evolutionary competition in harsh conditions. Hence, these
ACKNOWLEDGMENTS
asexual populations continuously accumulated deleterious mu-
tations in an irreversible manner, a process known as Muller’s We thank all K.J.V. and S.M. laboratory members for their help and sugges-
ratchet (Muller, 1964). We propose that continuous clonal repro- tions. We thank K. Wolfe, S. Oliver, and P. Malcorps for their valuable feedback
duction and relaxed selection for general stress resistance and on the manuscript. We acknowledge D. Brami, A. Bass, J. Kurowski, K. For-
famine likely allowed genome decay in beer yeasts and resulted tmann, and N. Parker for data collection and data review. Additionally we
in yeasts specialized in thriving in a man-made niche like beer acknowledge S. Rombauts, Y. Lin, R. de Jonge, and V. Storme for fruitful dis-
cussions on data analysis. B.G. acknowledges funding from the VIB Interna-
fermentations, but not in natural environments. Both these char-
tional PhD Program in Life Sciences. J.S. acknowledges funding from IWT
acteristics (genome decay and niche specialization) are consid- and KU Leuven. K.J.V. acknowledges funding from an ERC Consolidator grant
ered to be key characteristics of domestication. CoG682009, HFSP program grant RGP0050/2013, KU Leuven NATAR Pro-
Our study does not only provide insight into the domestication gram Financing, VIB, EMBO YIP program, FWO, and IWT. S.M. acknowledges
origin of industrial yeasts, it may also help to select and breed new funding from VIB and Ghent University. The funders had no role in study
superior strains. The genome sequences, phylogenetic tree, and design, data collection and analysis, the decision to publish, or preparation
of the manuscript.
phenome data can be used to set up marker-assisted breeding
schemes similar to those routinely used for the breeding of supe-
Received: March 24, 2016
rior crops and livestock (Takeda and Matsuoka, 2008). Revised: June 8, 2016
Accepted: August 8, 2016
STAR+METHODS Published: September 8, 2016
Detailed methods are provided in the online version of this paper REFERENCES
and include the following:

Almeida, P., Barbosa, R., Zalar, P., Imanishi, Y., Shimizu, K., Turchetti, B.,
Legras, J.L., Serra, M., Dequin, S., Couloux, A., et al. (2015). A population
d KEY RESOURCES TABLE
genomics insight into the Mediterranean origins of wine yeast domestication.
d CONTACT FOR REAGENT AND RESOURCE SHARING Mol. Ecol. 24, 5412–5427.
d EXPERIMENTAL MODEL AND SUBJECT DETAILS
Baele, G., and Lemey, P. (2013). Bayesian evolutionary model testing in
B Strain Collection the phylogenomics era: matching model complexity with computational effi-
d METHOD DETAILS ciency. Bioinformatics 29, 1970–1979.
B DNA Extraction Baele, G., Lemey, P., Bedford, T., Rambaut, A., Suchard, M.A., and
B Library Prep and Whole Genome Sequencing Alekseyenko, A.V. (2012). Improving the accuracy of demographic and molec-
B De Novo Assembly ular clock model comparison while accommodating phylogenetic uncertainty.
B Annotation Mol. Biol. Evol. 29, 2157–2167.
B Core Genome Analysis and Identification of Single Bergström, A., Simpson, J.T., Salinas, F., Barré, B., Parts, L., Zia, A., Nguyen
Copy Genes Ba, A.N., Moses, A.M., Louis, E.J., Mustonen, V., et al. (2014). A high-definition
view of functional genetic variation from natural yeast genomes. Mol. Biol.
B Reference-Based Alignments and Variant Calling
Evol. 31, 872–888.
B Phylogenetic Analyses
Bokulich, N.A., Thorngate, J.H., Richardson, P.M., and Mills, D.A. (2014).
B Population Structure and Diversity Analysis
Microbial biogeography of wine grapes is conditioned by cultivar, vintage,
B Time Divergence Estimate
and climate. Proc. Natl. Acad. Sci. USA 111, E139–E148.
B Copy-Number Variation Analysis
Bolger, A.M., Lohse, M., and Usadel, B. (2014). Trimmomatic: a flexible
B Character Evolution Analysis
trimmer for Illumina sequence data. Bioinformatics 30, 2114–2120.
B Determination of Cell Ploidy
Borneman, A.R., Desany, B.A., Riches, D., Affourtit, J.P., Forgan, A.H., Pretor-
B Phenotypic Analysis ius, I.S., Egholm, M., and Chambers, P.J. (2011). Whole-genome comparison
B Development of Artificial Hybrids reveals novel genetic elements that characterize the genome of industrial
d QUANTIFICATION AND STATISTICAL ANALYSIS strains of Saccharomyces cerevisiae. PLoS Genet. 7, e1001287.
d DATA AND SOFTWARE AVAILABILITY Borneman, A.R., Forgan, A.H., Kolouchova, R., Fraser, J.A., and Schmidt, S.A.
B Data Resources (2016). Whole genome comparison reveals high levels of inbreeding and strain
1408 Cell 166, 1397–1410, September 8, 2016

redundancy across the spectrum of commercial wine strains of Saccharo- Katoh, K., and Standley, D.M. (2013). MAFFT multiple sequence alignment
myces cerevisiae. G3 (Bethesda) 6, 957–971. software version 7: improvements in performance and usability. Mol. Biol.
Briza, P., Breitenbach, M., Ellinger, A., and Segall, J. (1990). Isolation of two Evol. 30, 772–780.
developmentally regulated genes involved in spore wall maturation in Saccha- Kozlov, A.M., Aberer, A.J., and Stamatakis, A. (2015). ExaML version 3: a tool
romyces cerevisiae. Genes Dev. 4, 1775–1789. for phylogenomic analyses on supercomputers. Bioinformatics 31, 2577–
2579.
Brown, C.A., Murray, A.W., and Verstrepen, K.J. (2010). Rapid expansion and
functional divergence of subtelomeric gene families in yeasts. Curr. Biol. 20, Kück, P., and Meusemann, K. (2010). FASconCAT: convenient handling of
895–903. data matrices. Mol. Phylogenet. Evol. 56, 1115–1118.
Capella-Gutiérrez, S., Silla-Martı́nez, J.M., and Gabaldón, T. (2009). trimAl: a Kuhn, R.M., Karolchik, D., Zweig, A.S., Trumbower, H., Thomas, D.J., Thakka-
tool for automated alignment trimming in large-scale phylogenetic analyses. pallayil, A., Sugnet, C.W., Stanke, M., Smith, K.E., Siepel, A., et al. (2007). The
Bioinformatics 25, 1972–1973. UCSC genome browser database: update 2007. Nucleic Acids Res. 35, D668–
D673.
Christiaens, J.F., Franco, L.M., Cools, T.L., De Meester, L., Michiels, J., Wen-
seleers, T., Hassan, B.A., Yaksi, E., and Verstrepen, K.J. (2014). The fungal Lanfear, R., Calcott, B., Ho, S.Y.W., and Guindon, S. (2012). Partitionfinder:
aroma gene ATF1 promotes dispersal of yeast cells through insect vectors. combined selection of partitioning schemes and substitution models for phylo-
Cell Rep. 9, 425–432. genetic analyses. Mol. Biol. Evol. 29, 1695–1701.
Cingolani, P., Platts, A., Wang, L., Coon, M., Nguyen, T., Wang, L., Land, S.J., Legras, J.L., and Karst, F. (2003). Optimisation of interdelta analysis for
Lu, X., and Ruden, D.M. (2012). A program for annotating and predicting the Saccharomyces cerevisiae strain characterisation. FEMS Microbiol. Lett.
effects of single nucleotide polymorphisms, SnpEff: SNPs in the genome of 221, 249–255.
Drosophila melanogaster strain w1118; iso-2; iso-3. Fly (Austin) 6, 80–92. Lemey, P., Rambaut, A., Drummond, A.J., and Suchard, M.A. (2009). Bayesian
Danecek, P., Auton, A., Abecasis, G., Albers, C.A., Banks, E., DePristo, M.A., phylogeography finds its roots. PLoS Comput. Biol. 5, e1000520.
Handsaker, R.E., Lunter, G., Marth, G.T., Sherry, S.T., et al.; 1000 Genomes Li, H., and Durbin, R. (2009). Fast and accurate short read alignment with Bur-
Project Analysis Group (2011). The variant call format and VCFtools. Bioinfor- rows-Wheeler transform. Bioinformatics 25, 1754–1760.
matics 27, 2156–2158. Liti, G., Carter, D.M., Moses, A.M., Warringer, J., Parts, L., James, S.A., Davey,
Dittmar, J.C., Reid, R.J., and Rothstein, R. (2010). ScreenMill: a freely available R.P., Roberts, I.N., Burt, A., Koufopanou, V., et al. (2009). Population genomics
software suite for growth measurement, analysis and visualization of high- of domestic and wild yeasts. Nature 458, 337–341.
throughput screen data. BMC Bioinformatics 11, 353. Liu, Y., Schröder, J., and Schmidt, B. (2013). Musket: a multistage k-mer spec-
Driscoll, C.A., Macdonald, D.W., and O’Brien, S.J. (2009). From wild animals to trum-based error corrector for Illumina sequence data. Bioinformatics 29,
domestic pets, an evolutionary view of domestication. Proc. Natl. Acad. Sci. 308–315.
USA 106 (Suppl 1 ), 9971–9978. Loewe, L., Textor, V., and Scherer, S. (2003). High deleterious genomic muta-
Drummond, A.J., and Suchard, M.A. (2010). Bayesian random local clocks, or tion rate in stationary phase of Escherichia coli. Science 302, 1558–1560.
one rate to rule them all. BMC Biol. 8, 114. Lynch, M., Sung, W., Morris, K., Coffey, N., Landry, C.R., Dopman, E.B., Dick-
Drummond, A.J., Suchard, M.A., Xie, D., and Rambaut, A. (2012). Bayesian inson, W.J., Okamoto, K., Kulkarni, S., Hartl, D.L., and Thomas, W.K. (2008). A
phylogenetics with BEAUti and the BEAST 1.7. Mol. Biol. Evol. 29, 1969–1973. genome-wide view of the spectrum of spontaneous mutations in yeast. Proc.
Natl. Acad. Sci. USA 105, 9272–9277.
Dunham, M.J., Badrane, H., Ferea, T., Adams, J., Brown, P.O., Rosenzweig,
F., and Botstein, D. (2002). Characteristic genome rearrangements in experi- Magwene, P.M., Kayıkçı, Ö., Granek, J.A., Reininga, J.M., Scholl, Z., and Mur-
mental evolution of Saccharomyces cerevisiae. Proc. Natl. Acad. Sci. USA ray, D. (2011). Outcrossing, mitotic recombination, and life-history trade-offs
99, 16144–16149. shape genome evolution in Saccharomyces cerevisiae. Proc. Natl. Acad.
Sci. USA 108, 1987–1992.
Dunn, B., Richter, C., Kvitek, D.J., Pugh, T., and Sherlock, G. (2012). Analysis
of the Saccharomyces cerevisiae pan-genome reveals a pool of copy number McDonald, M.J., Rice, D.P., and Desai, M.M. (2016). Sex speeds adaptation by
variants distributed in diverse yeast strains from differing industrial environ- altering the dynamics of molecular evolution. Nature 531, 233–236.
ments. Genome Res. 22, 908–924. McKenna, A., Hanna, M., Banks, E., Sivachenko, A., Cibulskis, K., Kernytsky,
Eden, E., Navon, R., Steinfeld, I., Lipson, D., and Yakhini, Z. (2009). GOrilla: a A., Garimella, K., Altshuler, D., Gabriel, S., Daly, M., and DePristo, M.A. (2010).
tool for discovery and visualization of enriched GO terms in ranked gene lists. The Genome Analysis Toolkit: a MapReduce framework for analyzing next-
BMC Bioinformatics 10, 48. generation DNA sequencing data. Genome Res. 20, 1297–1303.
Michel, R.H., McGovern, P.E., and Badler, V.R. (1992). Chemical evidence for
Filteau, M., Hamel, V., Pouliot, M.C., Gagnon-Arsenault, I., Dubé, A.K., and
ancient beer. Nature 360, 24.
Landry, C.R. (2015). Evolutionary rescue by compensatory mutations is con-
strained by genomic and environmental backgrounds. Mol. Syst. Biol. 11, 832. Mukai, N., Masaki, K., Fujii, T., Kawamukai, M., and Iefuji, H. (2010). PAD1 and
FDC1 are essential for the decarboxylation of phenylacrylic acids in Saccharo-
Fu, L., Niu, B., Zhu, Z., Wu, S., and Li, W. (2012). CD-HIT: accelerated for clus-
myces cerevisiae. J. Biosci. Bioeng. 109, 564–569.
tering the next-generation sequencing data. Bioinformatics 28, 3150–3152.
Muller, H.J. (1964). The relation of recombination to mutational advance.
Goddard, M.R., and Greig, D. (2015). Saccharomyces cerevisiae: a nomadic
Mutat. Res. 106, 2–9.
yeast with no niche? FEMS Yeast Res. 15, fov009.
Pattengale, N.D., Alipour, M., Bininda-Emonds, O.R., Moret, B.M., and Stama-
Goddard, M.R., Godfray, H.C.J., and Burt, A. (2005). Sex increases the effi-
takis, A. (2010). How many bootstrap replicates are necessary? J. Comput.
cacy of natural selection in experimental yeast populations. Nature 434,
Biol. 17, 337–354.
636–640.
Pavelka, N., Rancati, G., Zhu, J., Bradford, W.D., Saraf, A., Florens, L.,
Hornsey, I.S. (2003). A History of Beer and Brewing (Cambridge: The Royal
Sanderson, B.W., Hattem, G.L., and Li, R. (2010). Aneuploidy confers quanti-
Society of Chemistry).
tative proteome changes and phenotypic variation in budding yeast. Nature
Hutter, S., Vilella, A.J., and Rozas, J. (2006). Genome-wide DNA polymor- 468, 321–325.
phism analyses using VariScan. BMC Bioinformatics 7, 409. Peng, Y., Leung, H., Yiu, S., and Chin, F. (2010). IDBA – a practical iterative de
Huxley, C., Green, E.D., and Dunham, I. (1990). Rapid assessment of S. cere- Bruijn Graph de novo assembler. In Research in Computational Molecular
visiae mating type by PCR. Trends Genet. 6, 236. Biology, B. Berger, ed. (Springer), pp. 426–440.
Cell 166, 1397–1410, September 8, 2016 1409

Pérez-Ortı́n, J.E., Querol, A., Puig, S., and Barrio, E. (2002). Molecular charac- Steensels, J., Meersman, E., Snoek, T., Saels, V., and Verstrepen, K.J. (2014).
terization of a chromosomal rearrangement involved in the adaptive evolution Large-scale selection and breeding to generate industrial yeasts with superior
of yeast strains. Genome Res. 12, 1533–1539. aroma production. Appl. Environ. Microbiol. 80, 6965–6975.
Pfeifer, B., Wittelsbürger, U., Ramos-Onsins, S.E., and Lercher, M.J. (2014). Stefanini, I., Dapporto, L., Berná, L., Polsinelli, M., Turillazzi, S., and Cavalieri,
PopGenome: an efficient Swiss army knife for population genomic analyses D. (2016). Social wasps are a Saccharomyces mating nest. Proc. Natl. Acad.
in R. Mol. Biol. Evol. 31, 1929–1936. Sci. USA 113, 2247–2251.
Purcell, S., Neale, B., Todd-Brown, K., Thomas, L., Ferreira, M.A.R., Bender, Strope, P.K., Skelly, D.A., Kozmin, S.G., Mahadevan, G., Stone, E.A.,
D., Maller, J., Sklar, P., de Bakker, P.I.W., Daly, M.J., and Sham, P.C. Magwene, P.M., Dietrich, F.S., and McCusker, J.H. (2015). The 100-genomes
(2007). PLINK: a tool set for whole-genome association and population-based strains, an S. cerevisiae resource that illuminates its natural phenotypic and
linkage analyses. Am. J. Hum. Genet. 81, 559–575. genotypic variation and emergence as an opportunistic pathogen. Genome
Purugganan, M.D., and Fuller, D.Q. (2009). The nature of selection during plant Res. 25, 762–774.
domestication. Nature 457, 843–848. Suyama, M., Torrents, D., and Bork, P. (2006). PAL2NAL: robust conversion
R Development Core Team (2011). R: A language and environment for statis- of protein sequence alignments into the corresponding codon alignments.
tical computing (R Foundation for Statistical Computing). Nucleic Acids Res. 34 (Web Server issue, suppl 2), W609–W612.
Raj, A., Stephens, M., and Pritchard, J.K. (2014). fastSTRUCTURE: varia- Takeda, S., and Matsuoka, M. (2008). Genetic approaches to crop improve-
tional inference of population structure in large SNP data sets. Genetics 197, ment: responding to environmental and population changes. Nat. Rev. Genet.
573–589. 9, 444–457.
Rambaut, A., Suchard, M.A., Xie, D., and Drummond, A.J. (2014) Tracer v1.6.
van Dijk, D., Dhar, R., Missarova, A.M., Espinar, L., Blevins, W.R., Lehner, B.,
http://beast.bio.ed.ac.uk/Tracer.
and Carey, L.B. (2015). Slow-growing cells within isogenic populations have
Rancati, G., Pavelka, N., Fleharty, B., Noll, A., Trimble, R., Walton, K., Perera, increased RNA polymerase error rates and DNA damage. Nat. Commun. 6,
A., Staehling-Hampton, K., Seidel, C.W., and Li, R. (2008). Aneuploidy under- 7972.
lies rapid adaptive evolution of yeast cells deprived of a conserved cytokinesis
Van Wieren, D.P. (1995). American Breweries II (Eastern Coast Breweriana
motor. Cell 135, 879–893.
Association).
Ranwez, V., Harispe, S., Delsuc, F., and Douzery, E.J.P. (2011). MACSE: Mul-
Voordeckers, K., Kominek, J., Das, A., Espinosa-Cantú, A., De Maeyer, D.,
tiple Alignment of Coding SEquences accounting for frameshifts and stop
Arslan, A., Van Pee, M., van der Zande, E., Meert, W., Yang, Y., et al. (2015).
codons. PLoS ONE 6, e22594.
Adaptation to high ethanol reveals complex evolutionary pathways. PLoS
Schacherer, J., Shapiro, J.A., Ruderfer, D.M., and Kruglyak, L. (2009).
Genet. 11, e1005635.
Comprehensive polymorphism survey elucidates population structure of
Saccharomyces cerevisiae. Nature 458, 342–345. Wang, Q.-M., Liu, W.-Q., Liti, G., Wang, S.A., and Bai, F.Y. (2012). Surprisingly
diverged populations of Saccharomyces cerevisiae in natural environments
Selmecki, A.M., Dulmage, K., Cowen, L.E., Anderson, J.B., and Berman, J.
remote from human activity. Mol. Ecol. 21, 5404–5417.
(2009). Acquisition of aneuploidy provides increased fitness during the evolu-
tion of antifungal drug resistance. PLoS Genet. 5, e1000705. Wapinski, I., Pfeffer, A., Friedman, N., and Regev, A. (2007). Natural history and
Sheltzer, J.M., Blank, H.M., Pfau, S.J., Tange, Y., George, B.M., Humpton, evolutionary principles of gene duplication in fungi. Nature 449, 54–61.
T.J., Brito, I.L., Hiraoka, Y., Niwa, O., and Amon, A. (2011). Aneuploidy drives Warringer, J., Zörgö, E., Cubillos, F.A., Zia, A., Gjuvsland, A., Simpson, J.T.,
genomic instability in yeast. Science 333, 1026–1030. Forsmark, A., Durbin, R., Omholt, S.W., Louis, E.J., et al. (2011). Trait variation
Stamatakis, A. (2014). RAxML version 8: a tool for phylogenetic analysis and in yeast is defined by population history. PLoS Genet. 7, e1002111.
post-analysis of large phylogenies. Bioinformatics 30, 1312–1313. White, M.D., Payne, K.A.P., Fisher, K., Marshall, S.A., Parker, D., Rattray,
Stanke, M., and Morgenstern, B. (2005). AUGUSTUS: a web server for gene N.J.W., Trivedi, D.K., Goodacre, R., Rigby, S.E.J., Scrutton, N.S., et al.
prediction in eukaryotes that allows user-defined constraints. Nucleic Acids (2015). UbiX is a flavin prenyltransferase required for bacterial ubiquinone
Res. 33 (Web Server issue, suppl 2), W465–W467. biosynthesis. Nature 522, 502–506.
Steensels, J., and Verstrepen, K.J. (2014). Taming wild yeast: potential of Zheng, X., Levine, D., Shen, J., Gogarten, S.M., Laurie, C., and Weir, B.S.
conventional and nonconventional yeasts in industrial fermentations. Annu. (2012). A high-performance computing toolset for relatedness and principal
Rev. Microbiol. 68, 61–80. component analysis of SNP data. Bioinformatics 28, 3326–3328.
1410 Cell 166, 1397–1410, September 8, 2016

Article
Zika Virus Infection during Pregnancy

in Mice Causes Placental Damage and Fetal Demise
Jonathan J. Miner,1,8 Bin Cao,5,8 Jennifer Govero,1,8 Amber M. Smith,1 Estefania Fernandez,1,4 Omar H. Cabrera,6
Charise Garber,1 Michelle Noll,1 Robyn S. Klein,1,2,3,4 Kevin K. Noguchi,6 Indira U. Mysorekar,4,5,*
and Michael S. Diamond1,3,4,7,*
1Department of Medicine
2Department of Anatomy and Neurobiology
3Department of Molecular Microbiology
4Department of Pathology and Immunology
5Department of Obstetrics and Gynecology
6Department of Psychiatry
7The Center for Human Immunology and Immunotherapy Programs
Washington University School of Medicine, Saint Louis, MO 63110, USA

8Co-first authors
*Correspondence: mysorekari@wudosis.wustl.edu (I.U.M.), diamond@borcim.wustl.edu (M.S.D.)

SUMMARY occur, including Guillain-Barre syndrome and meningoencepha-

litis (Carteaux et al., 2016; Oehler et al., 2014). Infection in preg-
Zika virus (ZIKV) infection in pregnant women causes nant women is of major concern, as it is linked to catastrophic
intrauterine growth restriction, spontaneous abor- fetal abnormalities including microcephaly, spontaneous abor-
tion, and microcephaly. Here, we describe two mouse tion, and intrauterine growth restriction (IUGR) due to placental
models of placental and fetal disease associated with insufficiency (Brasil et al., 2016). Because of the growing public
in utero transmission of ZIKV. Female mice lacking health concern, there is an urgent need to establish animal
models of intrauterine ZIKV infection that define mechanisms
type I interferon signaling (Ifnar1/) crossed to
of fetal transmission and facilitate testing of therapeutics and
wild-type (WT) males produced heterozygous fe- vaccines. Furthermore, an in utero animal model of ZIKV infec-
tuses resembling the immune status of human fe- tion would establish causality and satisfy the criteria for proof
tuses. Maternal inoculation at embryonic day 6.5 of teratogenicity (Rasmussen et al., 2016).
(E6.5) or E7.5 resulted in fetal demise that was asso- In 2015, Brazil experienced a sharp rise in the number of cases
ciated with ZIKV infection of the placenta and fetal of pregnancy-associated microcephaly, and this was linked to a
brain. We identified ZIKV within trophoblasts of the concurrent epidemic of ZIKV infection. Mounting evidence sug-
maternal and fetal placenta, consistent with a trans- gests that ZIKV infection in pregnant women causes congenital
placental infection route. Antibody blockade of abnormalities and fetal demise (Brasil et al., 2016; Sarno et al.,
Ifnar1 signaling in WT pregnant mice enhanced 2016; Ventura et al., 2016). Initial case descriptions of micro-
ZIKV trans-placental infection although it did not cephaly and spontaneous abortion have been bolstered by
evidence of viral RNA and antigen in the brains of congenitally in-
result in fetal death. These models will facilitate the
fected fetuses and newborns (Martines et al., 2016; Mlakar et al.,
study of ZIKV pathogenesis, in utero transmission, 2016). These findings were substantiated by a prospective study
and testing of therapies and vaccines to prevent of a cohort of symptomatic, ZIKV-infected pregnant women in
congenital malformations. which 29% of fetuses exhibited developmental abnormalities
including microcephaly and IUGR, which in a subset of cases re-
INTRODUCTION sulted in fetal demise or stillbirth (Brasil et al., 2016). Preliminary
reports suggest that ZIKV-induced fetal abnormalities can occur
Zika virus (ZIKV) is a mosquito-transmitted flavivirus that was in all trimesters of pregnancy although the most severe manifes-
first isolated from a febrile rhesus macaque in Uganda in 1947 tations are associated with infections in the first and second tri-
and is related to other globally relevant arthropod-transmitted mesters (Brasil et al., 2016). Congenital abnormalities associated
human pathogens including dengue (DENV), yellow fever with ZIKV infection also have been described in French Polyne-
(YFV), West Nile (WNV), Japanese encephalitis (JEV), and tick- sia (by retrospective analysis) and other Latin American coun-
borne encephalitis viruses (Lazear and Diamond, 2016). Over tries (Cauchemez et al., 2016). These findings suggest that
the last decade, ZIKV has emerged from a relatively obscure sta- ZIKV strains in French Polynesia and Latin America share the
tus to causing large epidemics in Micronesia, French Polynesia, potential to cause disease during pregnancy.
and South and Central America. Although in most instances ZIKV Recently, we and others have developed models of ZIKV path-
infection results in a self-limiting febrile illness associated with ogenesis in adult mice that recapitulated several features of hu-
rash and conjunctivitis, severe neurological phenotypes can man disease (Aliota et al., 2016; Lazear et al., 2016; Rossi et al.,
Cell 165, 1081–1091, May 19, 2016 ª 2016 Elsevier Inc. 1081
2016). Whereas 4- to 6-week-old wild-type (WT) mice did not uated morphologically for size and appearance by measuring the
develop overt clinical illness after infection with a contemporary crown-rump length and the occipito-frontal diameter of the fetal
clinical strain of ZIKV, mice lacking the ability to produce or head, the latter of which establishes microcephaly in human
respond to type I interferon (IFN) (e.g., Ifnar1/ mice) developed fetuses (Chervenak et al., 1987; Staples et al., 2016). By E13.5,
severe neurological disease that was associated with high viral the majority of ZIKV-infected Ifnar1+/ heterozygous fetuses
loads in the brain and spinal cord and substantial lethality. In a had undergone fetal demise and been resorbed, leaving only a
complementary approach using WT mice treated with a blocking placental remnant (Figures 1B, upper, and 1C). The remaining
anti-ifnar antibody (MAR1-5A3), we reported a less severe model intact Ifnar1+/ fetuses exhibited significant IUGR (60.2 mm2
of ZIKV pathogenesis that also resulted in replication of ZIKV in versus 48.7 mm2, p < 0.0001, Figure 1D). In ZIKV-infected preg-
several organs (Lazear et al., 2016). These animals, however, nant women, multiple phenotypes have been described
survived infection and did not develop neurological signs or neu- including fetal demise, IUGR, and microcephaly (Brasil et al.,
roinvasive disease. 2016; Sarno et al., 2016). Although we did not observe isolated
Given the urgent need to understand the basis for in utero microcephaly in this in utero model of ZIKV infection, several
transmission of ZIKV and its pathological consequences, we other abnormalities were visible in ZIKV-infected Ifnar1+/ fe-
developed two models of ZIKV infection during pregnancy using tuses, including pallor and foci of necrotic tissue in the placenta
Ifnar1/ females crossed to WT males as well as pregnant WT (Figure 1B).
females treated with an anti-ifnar-blocking antibody. We found To determine whether direct infection of the placenta and fetus
that ZIKV infects pregnant dams and the placenta, and this re- occurred, we measured ZIKV RNA levels by quantitative real-
sulted in damage to the placental barrier and infection of the time RT-PCR (qRT-PCR) as well as infectious virus by plaque
developing fetus, as well as placental insufficiency and IUGR. assay. High levels of viral RNA and infectious virus were detected
In severe cases, ZIKV infection of Ifnar1/ females led to fetal within the placenta and also within the fetus head by E13.5 (Fig-
demise. When dams were treated with an anti-ifnar antibody, ures 1E and 1F and Figures S1A and S1B). As seen with the fe-
infection of the developing fetus occurred but was less severe tuses from dams infected on E6.5, ZIKV inoculation on E7.5 also
and did not cause fetal death. These findings establish models resulted in fetal demise and resorption by E15.5 as well as
for studying mechanisms of in utero transmission and testing growth restriction (141.8 mm2 versus 79.5 mm2, p < 0.0001, Fig-
of candidate therapies for preventing congenital malformations. ure 1H) and pallor (Figure 1B) of intact fetuses. As expected from
They also highlight the concern that ZIKV infection can occur in prior studies with Ifnar1/ males (Lazear et al., 2016; Rossi
fetuses of otherwise healthy-appearing dams with uncertain et al., 2016), high levels of ZIKV were present in the blood,
neurodevelopmental consequences. spleen, and brain of Ifnar1/ dams at day 7 after infection (Fig-
ures 1I–1K). Of note, the amount of ZIKV RNA within the placenta
RESULTS was 1,000-fold greater than in maternal serum (Figures 1F and
1I), suggesting that ZIKV replicates preferentially within this
Since the type I interferon (IFN) response prevents efficient repli- tissue.
cation of ZIKV in peripheral organs of WT mice (Lazear et al., In our second model of ZIKV infection during pregnancy, WT
2016), we initially used Ifnar1/ mice to facilitate high levels of mice were treated with MAR1-5A3, a blocking anti-ifnar mono-
ZIKV replication during pregnancy. Ifnar1/ female mice were clonal antibody (Sheehan et al., 2006), on E5.5, inoculated
bred with WT males so that resulting fetuses would be heterozy- with ZIKV on E6.5 or E7.5, and fetuses were analyzed on
gous (Ifnar1+/) and thus exhibit a largely intact type I IFN E13.5 or E15.5, respectively (Figure 1A). Although demise was
signaling response. In parallel, we developed a second model not observed, fetuses exhibited evidence of IUGR compared
of ZIKV infection during pregnancy by treating WT pregnant to control mAb-treated and mock-infected animals (62.3 mm2
dams with an anti-ifnar-blocking antibody 1 day prior to infection versus 50.2 mm2, p < 0.005), albeit to a lesser extent than
(Figure 1A). Both sets of pregnant mice were inoculated via a seen in Ifnar1+/ animals (Figure 1D). In contrast, anti-ifnar
subcutaneous route in the footpad with 103 focus forming units mAb-treated mice inoculated subcutaneously with 103 FFU of
(FFU) of a clinical isolate from French Polynesia (H/PF/2013) a clinical DENV serotype 3 (DENV-3) isolate that replicates in
that was passaged in Vero cells. This ZIKV strain is at least mice (Pinto et al., 2015; Sarathy et al., 2015) did not exhibit ev-
97% identical at the nucleotide level to the sequence of an idence of placental or fetal infection by qRT-PCR or signs of
epidemic strain of ZIKV in Brazil (Calvet et al., 2016; Faria IUGR (Figure 1D and data not shown). These results suggest
et al., 2016). We confirmed the sequence of our ZIKV H/PF/ that ZIKV may have greater tropism for placental cells than other
2013 stock by next-generation sequencing (data not shown), flaviviruses.
which also allowed us to exclude the presence of adventitious The levels of ZIKV RNA detected in WT fetuses were affected
pathogens. by the dose of anti-ifnar mAb administered, with the greatest
In the Ifnar1/ model, pregnant dams mated with WT mice amounts of ZIKV RNA present in fetuses receiving 2 or 3 mgs
were inoculated on embryonic days 6.5 (E6.5) and E7.5 and of anti-ifnar mAb (Figure 1E). ZIKV RNA persisted in the anti-ifnar
sacrificed on E13.5 and E15.5, respectively (Figure 1A). To mini- mAb-treated fetal heads and bodies at least through E16.5 (Fig-
mize confounding effects of maternal illness on fetal viability, we ures S1C and S1D), a critical time in early development of the
evaluated pregnant Ifnar1/ mice prior to the onset of disease, mouse brain. The placentas in both the Ifnar1/ and anti-ifnar
which is characterized by hunched posture, fur ruffling, or hind- antibody models exhibited higher levels of infection than the fetal
limb paralysis (Lazear et al., 2016). Individual fetuses were eval- tissues, and ZIKV RNA accumulation in the placenta was
1082 Cell 165, 1081–1091, May 19, 2016

Figure 1. Mortality, Viral Burden, and Size of Mouse Fetuses after Maternal Infection with ZIKV
(A) Schematic depiction of two models of infection during pregnancy. Model 1: WT males were crossed with Ifnar1/ dams. Pregnant dams were infected
subcutaneously with ZIKV (103 FFU) on E6.5 or E7.5 followed by harvest on E13.5, or 15.5, respectively. Model 2: WT males were crossed with WT dams.
Pregnant dams treated with 1 mg of an anti-ifnar antibody on days 1, +1, and +3 relative to ZIKV (103 FFU) or DENV-3 (103 FFU) infection. Mice were sacrificed on
E13.5 or E15.5 and fetuses and placentas were harvested for measurements of fetal size by crown-rump length (CRL) and occipito-frontal (OF) diameter.
(B) E13.5 uteri from ZIKV-infected WT and Ifnar1/ dams. Most Ifnar1+/ fetuses carried by Ifnar1/ dams died in utero and had undergone resorption, leaving
only the residual placenta. In the lower three panels are representative images of fetuses carried by ZIKV-infected WT and mock-infected Ifnar1/ dams, the
latter of which exhibited growth restriction at E15.5.
(C) Fetus survival on E13.5 after infection with ZIKV on E6.5. Mice were either treated with three 1 mg doses of control or anti-ifnar antibody (left two bars) or
untreated mock- or ZIKV-infected Ifnar1/ dams (right two bars). Data are representative of at least 3 independent experiments with 1 pregnant female dam per
experiment. The n for each group is indicated above each bar. ****p < 0.0001.
(D) Fetus size as assessed by CRL x OF diameter in E13.5 fetuses following E6.5 infection of the indicated pregnant dams with either ZIKV or DENV-3. For these
experiments, a total dose of 3 mg of anti-ifnar antibody was used. Bars indicate the mean size of 8-20 fetuses from 2 or 3 independent experiments from fetuses
carried by 2 to 3 pregnant dams. ***p < 0.0005; ****p < 0.0001.
(E and F) Viral burden was measured by qRT-PCR assay from the fetal head and placenta on E13.5 after infection at E6.5. Symbols represent individual fetuses
pooled from several independent experiments with the exception of 4 intact Ifnar1+/ fetal heads that were carried by a single dam. Bars indicate the mean of 4 to
17 mice per group. Dotted lines represent the limit of sensitivity of the assay. *p < 0.05 **p < 0.005; ***p < 0.0005; ****p < 0.0001.
(G) Fetus survival on E15.5 after infection with ZIKV on E7.5. Data are representative of at least 2 independent experiments with 1 pregnant female dam per
experiment. The n for each group is indicated above each bar. ****p < 0.0001.
(H) Fetus size as assessed by CRL x OF diameter in E15.5 fetuses following E7.5 infection of the indicated pregnant dams with ZIKV. ****p < 0.0001.
(I–K) Viral burden was measured by qRT-PCR assay from maternal serum, spleen, and brain at E13.5. Symbols are derived from individual animals and pooled
from 2 or 3 independent experiments. Bars indicate the mean of 4 to 5 mice per group. Dotted lines represent the limit of sensitivity of the assay. See also
Figure S1.
Cell 165, 1081–1091, May 19, 2016 1083

independent of the anti-ifnar mAb dose above 0.5 mg (Figure 1F). ing of pan-cytokeratin, a pan-trophoblast marker, was dimin-
In comparison, mice treated with the isotype control antibody ished in infected Ifnar1/ placentas, consistent with evidence
sustained low levels or no detectable ZIKV infection in the of apoptotic trophoblasts (Figure 3B). Apoptosis in trophoblasts
placenta, fetal heads or maternal tissues (Figure 1E, 1F, and can cause disruption of the placental barrier, which compro-
1I–1K). Collectively, these data suggest that the mouse placenta mises protection against pathogens (Robbins and Bakardjiev,
is vulnerable to infection with ZIKV, and that high-grade infection 2012). Indeed, ZIKV-infected Ifnar1+/ placentas contained large
may cause placental insufficiency, IUGR, and fetal demise, at numbers of nucleated fetal erythrocytes (Figure 3A, blue arrows),
least in Ifnar1+/ animals. Anti-ifnar mAb-treated animals sus- key indicators of fetal stress. Evidence of vascular damage and
tained less infection and no enhanced lethality although a mild fewer blood vessels also was reflected by diminished staining of
IUGR phenotype was observed. vimentin, a marker of fetal blood vessels in mouse placentas
We evaluated ZIKV localization in the placenta to define (Figure 3B).
whether transmission occurred by a trans-placental route. The Histopathological assessment of ZIKV-infected Ifnar1+/ fetal
mouse placenta is comprised of the maternal decidua and the brains demonstrated abundant apoptotic cells within multiple re-
fetal embryo-derived compartments, including the junctional gions at E13.5 (Figures 4A–4D). Activated caspase-3 staining
and labyrinth zones (Figure 2A). Different types of trophoblasts showed low levels of physiological apoptosis in uninfected fe-
with distinct functions reside within all three layers, including tuses (Figures 4E–4H), whereas infected animals had apoptotic
trophoblast giant cells, glycogen trophoblasts, and spongiotro- cells throughout the midbrain and hindbrain (Figures 4B–4D
phoblasts. Within the labyrinth zone, fetal capillaries are lined and 4I). Although we could localize viral RNA in infected pla-
by fetal blood vessel endothelium, which are separated from centas, multiple attempts at RNA FISH staining of ZIKV-infected
maternal sinusoids by a layer of mononuclear trophoblasts fetal brains did not yield a clear pattern of viral RNA expression
and a syncytiotrophoblast bilayer (Figure 2A) (Simmons and (data not shown), despite the recovery of infectious virus (Figures
Cross, 2005; Watson and Cross, 2005). We performed RNA S1A and S1B). Accordingly, we cannot state with certainty
fluorescence in situ hybridization (FISH) coupled with histopath- whether the enhanced apoptosis within ZIKV-infected fetuses
ological analysis in ZIKV-infected Ifnar1+/ placentas and results from infection-induced apoptosis or another process,
confirmed the presence of ZIKV RNA in different trophoblast including ischemia due to placental insufficiency. The presence
cells, including glycogen trophoblasts and spongiotrophoblasts of numerous apoptotic cells within the developing central ner-
(Figure 2B) and to a lesser extent in mononuclear trophoblasts vous system (CNS) coupled with the established neurotropism
and syncytiotrophoblasts (data not shown). These findings are of ZIKV (Lazear et al., 2016), however, suggests direct infection
consistent with cell culture studies demonstrating ZIKV infection may contribute.
of human trophoblast cell lines (Bayer et al., 2016) and suggest
that the mouse model of infection during pregnancy recapitu- DISCUSSION
lates features of human disease including placental tropism of
ZIKV. We independently confirmed ZIKV infection and replica- Epidemiological studies have found that ZIKV infection during
tion in two of three human trophoblast cell lines (Figure S2). pregnancy causes catastrophic neurodevelopmental outcomes
Transmission electron microscopy of placentas revealed multi- in human fetuses, but there currently is no effective treatment
ple 50 nm dense bodies within the endoplasmic reticulum of the or prevention of ZIKV infection other than avoidance of its
mononuclear trophoblasts (Figure 2C, left), consistent with ZIKV mosquito vectors. Given the devastating effects of this rapidly
infection of the maternal placenta. As these bodies resemble emerging infectious disease, small animal models of ZIKV
flavivirus virions (Allison et al., 2003) and were not present in infection during pregnancy are urgently needed to test candi-
uninfected animals, they are suggestive for the presence of date therapeutics and vaccines that could prevent or mitigate
virus. Proximity to non-nucleated maternal erythrocytes (Fig- intrauterine infection with ZIKV. We developed two mouse
ure 2C, left) confirmed the location as within the maternal face models that support ZIKV replication and trans-placental trans-
of the placenta. Consistent with a trans-placental route of infec- mission in pregnant dams: (1) a model of severe disease in
tion, we also observed bodies resembling virions within the pregnant Ifnar1/ dams that resulted in fetal demise; and (2)
endoplasmic reticulum of fetal endothelial cells that lined a less severe model of ZIKV pathogenesis in utero using
damaged fetal capillaries (Figure 2C, right). The cellular and ul- pregnant WT dams that were given anti-ifnar antibody prior
trastructural evidence of ZIKV infection in trophoblasts and fetal to and during infection, which resulted in mild IUGR and
endothelium suggests that maternal viremia leads to compro- viral infection within the fetal head during a key period in
mise of the placental barrier by infecting fetal trophoblasts neurodevelopment.
and entering the fetal circulation. The placenta acts as a barrier against infections, due to multi-
Pathological analysis of ZIKV-infected Ifnar1/ (maternal) and ple unique structural, cellular, and immune properties. The detri-
Ifnar1+/ (fetal) placentas showed severe vascular injury charac- mental effects of congenital viruses on pregnancy and fetal out-
terized by irregularly shaped, reduced fetal capillaries and comes occur in part because of impaired trophoblast function
destruction of the placental microvasculature (Figures 3A and (Arechavaleta-Velasco et al., 2002). Defective placentas can
3B, Ifnar het severe). Infected Ifnar1/ placentas were smaller, lead to severe maternal and fetal morbidity and mortality during
mostly because the labyrinth zone was markedly thinned. In pregnancy, including spontaneous abortion, stillbirth, preterm
addition, apoptotic trophoblasts were evident in ZIKV-infected birth, IUGR, and other complications. We observed profound
placentas (Figure 3A, black arrows). Immunofluorescence stain- pathological changes in ZIKV-infected placentas, including
1084 Cell 165, 1081–1091, May 19, 2016

Figure 2. ZIKV Infects Trophoblasts and Endothelial Cells within Mouse Placentas
Pregnant Ifnar1/ dams were infected on E7.5 with 103 FFU of ZIKV via a subcutaneous route and placentas were harvested on E15.5 for histological analysis.
(A) Schematic representation of the mouse placental structure.
(B) Representative RNA FISH images in uninfected and infected Ifnar1/ placentas. Images in each column correspond to the same field of view generated under
bright-field or confocal microscopy. Higher magnification of images is displayed as inserts. Scale bar, 25 mm.
(C) Transmission electron microscopy images of ZIKV infected Ifnar1/ placentas. ZIKV particles were identified within the endoplasmic reticulum in the maternal
sinus (left), and in the fetal endothelium (right) lining fetal capillaries in the labyrinth layer.
MT = Mononuclear trophoblast; MS = Maternal sinus; FE = Fetal endothelial cell; FC = Fetal capillary; MRBC = Maternal erythrocyte; and ER = Endoplasmic
reticulum. See also Figure S2.
trophoblast apoptosis, abnormal fetal capillary features, and We observed variability in susceptibility to ZIKV infection of
increased fetal nucleated erythrocytes, indicating malfunction different human trophoblast cell lines. Trophoblast cell lines
of mouse placentas caused by ZIKV infection. (JEG-3 and HTR-8) originally cultured from choriocarcinoma
Cell 165, 1081–1091, May 19, 2016 1085

Figure 3. ZIKV Infection Triggers Apoptosis and Vascular Damage in the Placenta
Pregnant dams were infected on E7.5 with 103 FFU of ZIKV via a subcutaneous route and placentas were harvested on E15.5 for histological analysis.
(A) Representative hematoxylin and eosin staining showed pathological features of placentas at E15.5. Labyrinth layers were marked with a solid line on the cross
section of mouse placentas. Black arrows indicate apoptotic trophoblasts. Blue arrows indicate increased number of nucleated fetal erythrocytes in fetal
capillaries.
(B) Immunofluorescence staining of cytokeratin (CK) and vimentin in mouse placentas. CK, a marker for trophoblasts; vimentin, a marker for the endothelium in
fetal capillaries.
See also Figure S2.
explants and first trimester human villous explants, respec- ekar, 2014; Robbins and Bakardjiev, 2012; Zeldovich and Ba-
tively, which exhibit features of extravillous trophoblasts kardjiev, 2012). Thus, it is possible that in early pregnancy,
(EVTs) including high invasive capacity and expression of ZIKV infects EVTs and enters the fetal circulation. Placentas
HLA-G, a MHC class II molecule, were susceptible to ZIKV nearer to term, which have reduced EVTs on the tips of
infection. In contrast, a relatively undifferentiated cytotropho- anchoring villi and a more fully developed placental barrier,
blast cell line (BeWo) was not. Previous studies have shown in general exhibit greater resistance to infection. Indeed, hu-
that EVTs are most susceptible to bacterial infections, partic- man primary trophoblasts of the villous syncytiotrophoblast
ularly during the first and second trimesters (Cao and Mysor- phenotype from term placentas were resistant to ZIKV
1086 Cell 165, 1081–1091, May 19, 2016

Figure 4. ZIKV Infection Is Associated with
Evidence of Apoptosis in the Fetal Brain
Pregnant Infar1/ dams were infected with 103
FFU of ZIKV via a subcutaneous route. Infected
(left) or uninfected (right) Ifnar1+/ E13.5 fetuses
were stained with the apoptotic marker activated
caspase-3 (AC3; red) and the proliferative marker
Ki-67 (green).
(A and E) Sagittal images of representative in-
fected (A) and uninfected (E) fetal heads showing
high expression of Ki-67 along brain regions
adjacent to ventricles indicative of proliferating
neural progenitor cells in the neuroepithelium.
(B-H) Lettered box regions (B-D and F-H) in these
images are magnified in corresponding panels
below. Higher levels of apoptosis can be seen in
the midbrain (B-C) and hindbrain (D) of the infected
Ifnar1+/ fetus. Alternatively, low levels of physio-
logical apoptosis are seen in the absence of
infection (F-H).
(I) Diagram depicting the developing E13.5 fetal
brain in sagittal view including the forebrain
(green), midbrain (blue), and hindbrain (red).
Images are representative of 4-5 sections per
fetus from 2 fetuses.
infection in mice may require infection

later during pregnancy. Since ZIKV infects
mature neurons as well as neural cell pro-
genitors and limits their growth (Tang
et al., 2016), the morphological effects of
ZIKV infection on brain development
may be more apparent in species with
larger cerebral cortices.
Although ZIKV-associated micro-
cephaly has drawn major public health
and media attention, ZIKV infection of hu-
man fetuses does not always cause this
manifestation, and the sequelae of intra-
uterine infection in the absence of gross
morphological abnormalities remain to
be defined. In pregnant WT dams that
infection due to the production of IFN-l in paracrine manner were treated with an anti-ifnar antibody, we observed only mild
(Bayer et al., 2016). growth restriction in the developing fetus although ZIKV RNA
Several aspects of our models of ZIKV infection during preg- was detectable in the fetal head at both E13.5 and E16.5 after
nancy resemble intrauterine infection by ZIKV in humans. Com- infection. Future behavioral studies may define whether intra-
mon features included tropism of ZIKV for the placenta, evidence uterine infection by ZIKV has long-term neurological effects in
of intrauterine infection, and fetal demise. However, infection mice and serves as a model for evaluation of disease in humans
during pregnancy in mice did not recapitulate all aspects of hu- (Staples et al., 2016).
man disease, as we did not detect microcephaly, brain calcifica- We found that the murine placentas can be infected by ZIKV.
tions, or absence of individual brain structures, such as the Infection of Ifnar1/ dams led to severe placental damage
corpus callosum. There are several reasons why ZIKV may not and destruction of the microvasculature, which most likely
have induced these pathological manifestations in our models. limited blood flow to the developing fetus and caused severe
In mice, brain neurogenesis begins around E10 (Finlay and Dar- IUGR, ischemia, and fetal demise. In vivo infection of the
lington, 1995), and the brain of a newborn pup is relatively imma- mouse placenta with ZIKV may provide a model for defining
ture at postnatal day 1, akin to the developmental stage of the host factors required for or that restrict infection, which could
human brain at mid-gestation (Semple et al., 2013). As the devel- suggest a path for developing therapies to limit placental and
opment of the mouse brain includes a major postnatal compo- intrauterine infection. For example, since IFN-l restricts ZIKV
nent, examination of the neurodevelopmental effects of ZIKV replication within human trophoblasts from term placentas
Cell 165, 1081–1091, May 19, 2016 1087

(Bayer et al., 2016), studies are planned to test the effects of ogens. Finally, our animal model of in utero transmission estab-
exogenously administered IFN-l on in utero transmission in lishes causality of a fetal syndrome associated with ZIKV infec-
mice. tion in mice.
Maternal-fetal transmission of pathogens can be mediated
by diverse pathways, with the most common being via an EXPERIMENTAL PROCEDURES
ascending route and hemochorial transmission. Viruses from
the urogenital tract can disseminate into intrauterine space and Ethics Statement
This study was carried out in accordance with the recommendations in the
colonize the fetal membrane or the placenta in an ascending
Guide for the Care and Use of Laboratory Animals of the National Institutes
manner (Edwards et al., 2015). In contrast, viruses from the of Health. The protocols were approved by the Institutional Animal Care and
maternal blood circulation can be transported to the feto- Use Committee at the Washington University School of Medicine (Assurance
maternal space and infect trophoblasts lining the maternal-fetal number A3381-01). Inoculations were performed under anesthesia induced
interface. These trophoblasts include the EVTs, which embed in and maintained with ketamine hydrochloride and xylazine, and all efforts
maternal decidua, endovascular extravillous cytotrophoblasts, were made to minimize animal suffering.
and villous trophoblasts, which are bathed in maternal blood
Viruses and Titration
(Delorme-Axford et al., 2014). Our studies suggest that ZIKV
Vero cells (African green monkey kidney epithelial cells) were maintained in
can infect the placenta through blood-placental transmission DMEM supplemented with 5% fetal bovine serum (Omega) and L-glutamine
and bypass the placental barrier to infect the fetus. Nonetheless, at 37 C with 5% CO2. ZIKV strain H/PF/2013 (French Polynesia, 2013) was pro-
ascending infection routes might be important in sexual trans- vided by the Arbovirus Branch of the Centers for Disease Control and Preven-
mission during pregnancy. ZIKV has been found in human semen tion with permission (X. de Lamballerie, Aix Marseille Université) (Baronti et al.,
(Musso et al., 2015) and mouse testes (Lazear et al., 2016; Rossi 2014). ZIKV stocks were propagated in Vero cells and titrated by focus-forming
assay (FFA). Infected cell foci were detected at 48 hr after infection, following
et al., 2016) and can be transmitted sexually from male to female
fixation with 1% paraformaldehyde and incubation with 500 ng/mL of flavivirus
in humans (Musso et al., 2015). cross-reactive mouse monoclonal antibody E60 (Oliphant et al., 2006) for 2 hr
Intrauterine infection with flaviviruses may be an underappre- at room temperature. After incubation for 1 hr with a 1:5,000 dilution of horse-
ciated phenomenon. Prior to the ZIKV epidemic, there were radish peroxidase (HRP)-conjugated goat anti-mouse IgG (Sigma), foci were
isolated descriptions of trans-placental infection in humans detected by addition of TrueBlue substrate (KPL). Foci were analyzed with a
with other flaviviruses including WNV and JEV (Chaturvedi CTL Immunospot instrument. Studies with ZIKV were conducted under
et al., 1980; Nguyen et al., 2002). These reports suggest that biosafety level 2 (BSL2) and animal BSL3 (A-BSL3) containment.
sporadic cases of flavivirus-induced miscarriage or fetal demise

Mouse Experiments
might have been unrecognized, although this remains specula-
Ifnar1/ mice (Hwang et al., 1995) were backcrossed onto a C57BL/6 back-
tive. We are currently testing whether additional variations in ground. Mice were bred in a specific-pathogen-free facility at Washington Uni-
ZIKV infection (e.g., dose, route of administration, virus strain, versity, or purchased (WT animals) from Jackson Laboratories. Mice were set
time of infection during pregnancy, and time of analysis) in our up for timed-matings and at embryonic days E6.5 or E7.5 were inoculated with
mouse model of in utero transmission can recapitulate other ZIKV by subcutaneous (footpad) route with 103 FFU of ZIKV in 50 mL of PBS.
morphological abnormalities in the CNS that are described in Mice were sacrificed at E13.5, E15.5, or E16.5 depending on the experimental
design. Placentas and fetuses were harvested from the infected mice. Fetus
human disease. Of note, intrauterine infection with Saint
size was measured as the crown-rump length 3 occipito-frontal diameter of
Louis encephalitis virus, a less well studied flavivirus, caused the head. In some experiments, WT mice were treated with indicated doses
severe neurological outcomes in mice. In that model, disease of an Ifnar-blocking mouse mAb (MAR1-5A3) or isotype control mouse mAb
depended on the gestational date of infection (Andersen and (GIR-208) (produced by Leinco Technologies) (Sheehan et al., 2006; Sheehan
Hanson, 1970); mice infected early in gestation survived, et al., 2015) by intraperitoneal injection prior to and after ZIKV or DENV-3
whereas those inoculated later developed neurological malfor- infection.
mations and died as neonates (Andersen and Hanson, 1970;
Measurement of Viral Burden
Andersen and Hanson, 1975). In our studies with DENV and an
ZIKV-infected pregnant mice were euthanized on E13.5, E15.5, or E16.5. Pla-
anti-ifnar-blocking antibody, inoculation of mice did not result centas and fetal heads were weighed and homogenized with zirconia beads in
in placental infection, although it is possible that this was due a MagNA Lyser instrument (Roche Life Science) in 250 or 500 mL of PBS. All
to a diminished ability of DENV to replicate in mice compared homogenized tissues from infected animals were stored at 80 C until virus
to ZIKV. Experiments that test infection of pregnant animals titration. With some samples, viral burden was determined by plaque assay
with additional related viruses may clarify whether placental on Vero cells. Samples were thawed, clarified by centrifugation (2,000 3 g at
4 C for 10 min), and then diluted serially prior to infection of Vero cells. Plaque
infection is an underappreciated clinical manifestation of flavivi-
assays were overlaid with low-melt agarose, and 4 days later were fixed with
rus pathogenesis. 10% formaldehyde and stained with crystal violet (Brien et al., 2013). Tissue
Treatment and prevention of ZIKV infection will likely require samples and serum from ZIKV-infected mice were extracted with the RNeasy
small animal models for testing of vaccines and potential thera- Mini Kit (tissues) or Viral RNA Mini Kit (serum) (QIAGEN). ZIKV RNA levels were
pies. The mouse models described in our study may be relevant determined by one-step quantitative reverse transcriptase PCR (qRT-PCR) on
to studying mechanisms of pathogenesis and determining an ABI 7500 Fast Instrument using standard cycling conditions. Viral burden
whether vaccines given prior to pregnancy can prevent infection was expressed on a log10 scale as viral RNA equivalents per g or per mL after
comparison with a standard curve produced using serial 10-fold dilutions of
in the developing fetus. Mouse models of ZIKV infection during
ZIKV RNA. A published primer set was used to detect ZIKV RNA (Lanciotti
pregnancy also may provide fundamental insights into how the et al., 2008): Fwd, 50 -CCGCTGCCCAACACAAG-30 ; Rev, 50 -CCACTAACGTTC
placental barrier prevents viral infection from the developing TTTTGCAGACAT-30 ; Probe, 50 -/56-FAM/AGCCTACCT/ZEN/TGACAAGCAAT
fetus, and why this process fails in the context of specific path- CAGACACTCAA/3IABkFQ/-30 , (Integrated DNA Technologies).
1088 Cell 165, 1081–1091, May 19, 2016

RNA Fluorescence In Situ Hybridization cultured in 4-well chamber slides (Millipore). Cells were counted and infected
RNA FISH was performed using ViewRNA ISH Tissue 2-Plex Assay kit (Affyme- with ZIKV at a MOI 0.1 for 2 hr, washed twice with warm PBS and cultured with
trix) according to the manufacturer’s instructions. Formalin-fixed paraffin- fresh medium. At indicated time points after infection, supernatants were har-
embedded tissue sections were hydrated, heat-treated for 10 min at 90 C, vested for virus titration, and cells were trypsinized for flow cytometry or fixed
and digested with protease for 20 min. Endogenous alkaline phosphatase for immunofluorescence staining using ZV-2, a ZIKV-specific mAb.
was inactivated with 0.2 M HCl and 300 mM NaCl at room temperature for
15 min before the probe hybridization. The probe targeting ZIKV RNA was de- Data Analysis
signed and synthesized by Affymetrix and was based on the ZIKV French Poly- All data were analyzed with GraphPad Prism software. For viral burden anal-
nesian 2013 genomic RNA sequence (accession number SAMN04592777). ysis, the log titers and levels of viral RNA were analyzed by the Mann-Whitney
Positive controls (probe-targeting housekeeping genes) and no-probe nega- test or ANOVA. A p value of < 0.05 indicated statistically significant differences.
tive controls were included in all samples. Tissues were counterstained with Significance of the survival rates were assessed by Chi-square test. The
DAPI and Gill’s hematoxylin and visualized using standard bright-field and morphological measurements were assessed by a Mann-Whitney test or by
confocal microscopes. a 2-way ANOVA.
Immunohistochemistry and Immunofluorescence Imaging SUPPLEMENTAL INFORMATION

Fetuses were removed by Caesarian-section on E13.5, post-fixed in 4% para-
formaldehyde, and sectioned at 70 mM with a vibratome. Immunohistochem- Supplemental Information includes two figures and can be found with this
istry was performed by immersing sections in blocking solution and incubating article online at http://dx.doi.org/10.1016/j.cell.2016.05.008.
overnight in primary antibodies raised against activated caspase-3 (AC3; Cell
Signaling) and Ki-67 (BD-Biosciences). Subsequently, sections were rinsed AUTHOR CONTRIBUTIONS
and reacted with fluorescent secondary antibodies.
Harvested placentas were fixed in 10% neutral buffered formalin (Fisher) at J.J.M., J.G., B.C., K.K.N., R.S.K., I.U.M., and M.S.D. designed experiments.
room temperature and embedded in paraffin. At least three placentas from J.J.M., B.C., J.G, A.M.S, O.H.C., E.F., C.G., M.N., and K.K.N. performed the
different litters with the indicated treatments were sectioned and stained experiments. J.J.M, B.C, I.U.M., and M.S.D. analyzed the data. J.J.M., B.C.,
with hematoxylin and eosin to assess morphology. For immunofluorescence I.U.M., and M.S.D. wrote the first draft of the paper; all authors edited the
staining of mouse placentas, deparaffinized tissues were blocked in buffer manuscript.
(1% BSA, 0.3% Triton, 1 3 PBS) for 2 hr and incubated with primary antibodies
overnight. The following primary antibodies were used: cytokeratin (1:500, rab-
ACKNOWLEDGMENTS
bit, DAKO Z0622) and vimentin (1:500, rabbit, Abcam ab92547). After rinsing
with PBS, secondary antibody conjugated with Alexa 488 (1:500 in PBS)
This work was supported by grants from the NIH (R01 AI073755 and R01
was applied for 1 hr at room temperature. Samples were counterstained
AI104972 to M.S.D., R01 HD052664 to K.K.N., and R01 NS052632 to
with DAPI (4’6’-diamidino-2-phenilindole, 1:1,000 dilution) for 10 min and
R.S.K.) and the Intellectual and Developmental Disabilities Research Center
mounted in Prolong Gold (Life Technologies). RNA FISH imaging of mouse pla-
at Washington University (NIH/NICHD U54 HD087011). J.J.M., E.F., and
centas was acquired using a TSC SPE inverted confocal microscope (Leica)
D.J.P. were supported by a Rheumatology Research Foundation Scientist
using a 633 objective. Histology images were captured by use of a Nikon
Development Award, NIH Pre-doctoral training grant award (T32 AI007163),
Eclipse microscope equipped with an Olympus DP71 color camera under
and the NIH Research Education Program (R25 HG006687), respectively.
203 and 403 objectives.
The work also was supported by a Preventing Prematurity Initiative grant
from the Burroughs Wellcome Fund and a Prematurity Research Initiative
Transmission Electron Microscopy
Investigator award from the March of Dimes (to I.U.M). We thank Wandy Betty
Mouse placental samples were fixed in 2% paraformaldehyde and 2.5%
for her TEM assistance, Fredrik Kraus for his expertise in placental pathology,
glutaraldehyde (Polysciences) in 100 mM sodium cacodylate buffer for 1 hr
and Justin Richner for advice with establishing the RNA FISH assays. Finally,
at room temperature and overnight at 4 C, washed in cacodylate buffer,
we acknowledge Xavier de Lamballerie (Emergence des Pathologies Virales,
post-fixed in 1% osmium tetroxide (Polysciences) for 1 hr, then rinsed with
Aix-Marseille Université, Marseille, France) and the European Virus Archive
distilled H2O before en bloc staining with 1% aqueous uranyl acetate (Ted
goes Global (EVAg) for consenting to the use of H/PF/2013 ZIKV strain for
Pella) for 1 hr. Samples were dehydrated in a graded series of ethanol and
this study under a material transfer agreement with the EVAg parter, Aix-Mar-
embedded in Eponate 12 resin (Ted Pella). Sections (90 nm thick) were cut
seille Université. M.S.D. is a consultant for Inbios, Visterra, and Takeda Phar-
with a Leica Ultracut UCT ultramicrotome, stained with uranyl acetate and
maceuticals and on the Scientific Advisory Boards of Moderna and OraGene.
lead citrate, and imaged on a JEOL 1200 EX transmission electron microscope
(JEOL USA).
Received: April 24, 2016
Revised: May 2, 2016
mAb Generation
Accepted: May 2, 2016
A more complete description of the anti-ZIKV mAbs will be provided in a forth-
Published: May 11, 2016
coming manuscript (E. Fernandez, D. Platt, and M. Diamond, unpublished data).
Briefly, to create ZV-2, Irf3/ mice were infected and boosted with 103 FFU of REFERENCES
ZIKV (MR-766 and H/PF/2013, respectively) and given a final intravenous boost
with live 106 FFU of ZIKV (H/F/2013) 3 days prior to fusion with P3X63 myeloma
Aliota, M.T., Caine, E.A., Walker, E.C., Larkin, K.E., Camacho, E., and Osorio,
cells. Hybridomas secreting antibodies that reacted with ZIKV-infected Vero
J.E. (2016). Characterization of Lethal Zika Virus Infection in AG129 Mice.
cells were identified by flow cytometry and cloned by limiting dilution. ZV-2
PLoS Negl. Trop. Dis. 10, e0004682.
was purified by protein A affinity chromatography.
Allison, S.L., Tao, Y.J., O’Riordain, G., Mandl, C.W., Harrison, S.C., and Heinz,
F.X. (2003). Two distinct size classes of immature and mature subviral particles
Cell Culture and Infection
from tick-borne encephalitis virus. J. Virol. 77, 11357–11366.
JEG-3 and BeWo cells were obtained from ATCC and cultured in F12/DMEM
media supplemented with 10% FBS (Life Technologies) at 37 C with 5% CO2. Andersen, A.A., and Hanson, R.P. (1970). Experimental transplacental trans-
HTR-8/SVneo cells were provided by C. H. Graham (Queen’s University) and mission of st. Louis encephalitis virus in mice. Infect. Immun. 2, 320–325.
maintained in RPMI 1640 media with 5% FBS. For flow cytometry and viral Andersen, A.A., and Hanson, R.P. (1975). Intrauterine infection of mice with St.
yield assays, trophoblast cell lines were seeded in 24-well plates at 5 3 105 Louis encephalitis virus: immunological, physiological, neurological, and
cells per well. For immunofluorescence staining, trophoblast cell lines were behavioral effects on progeny. Infect. Immun. 12, 1173–1183.
Cell 165, 1081–1091, May 19, 2016 1089

Arechavaleta-Velasco, F., Koi, H., Strauss, J.F., 3rd, and Parry, S. (2002). Viral Lazear, H.M., Govero, J., Smith, A.M., Platt, D.J., Fernandez, E., Miner, J.J.,
infection of the trophoblast: time to take a serious look at its role in abnormal and Diamond, M.S. (2016). A Mouse Model of Zika Virus Pathogenesis. Cell
implantation and placentation? J. Reprod. Immunol. 55, 113–121. Host Microbe, S1931-3128(16)30102-0.
Baronti, C., Piorkowski, G., Charrel, R.N., Boubis, L., Leparc-Goffart, I., and de Martines, R.B., Bhatnagar, J., Keating, M.K., Silva-Flannery, L., Muehlen-
Lamballerie, X. (2014). Complete coding sequence of zika virus from a French bachs, A., Gary, J., Goldsmith, C., Hale, G., Ritter, J., Rollin, D., et al. (2016).
polynesia outbreak in 2013. Genome Announc. 2, e00500–e00514. Notes from the Field: Evidence of Zika Virus Infection in Brain and Placental
Bayer, A., Lennemann, N.J., Ouyang, Y., Bramley, J.C., Morosky, S., Marques, Tissues from Two Congenitally Infected Newborns and Two Fetal Losses -
E.T., Jr., Cherry, S., Sadovsky, Y., and Coyne, C.B. (2016). Type III Interferons Brazil, 2015. MMWR Morb. Mortal. wkly. Rep. 65, 159–160.
Produced by Human Placental Trophoblasts Confer Protection against Zika Mlakar, J., Korva, M., Tul, N., Popovic , M., Poljsak-Prijatelj, M., Mraz, J.,
Virus Infection. Cell Host Microbe 16, 30100–30107. Kolenc, M., Resman Rus, K., Vesnaver Vipotnik, T., Fabjan Vodusek, V.,
Brasil, P., Pereira, J.P., Jr., Raja Gabaglia, C., Damasceno, L., Wakimoto, M., et al. (2016). Zika Virus Associated with Microcephaly. N. Engl. J. Med. 374,
Ribeiro Nogueira, R.M., Carvalho de Sequeira, P., Machado Siqueira, A., 951–958.
Abreu de Carvalho, L.M., Cotrim da Cunha, D., et al. (2016). Zika Virus Musso, D., Roche, C., Robin, E., Nhan, T., Teissier, A., and Cao-Lormeau, V.M.
Infection in Pregnant Women in Rio de Janeiro - Preliminary Report. (2015). Potential sexual transmission of Zika virus. Emerg. Infect. Dis. 21,
N. Engl. J. Med. Published online March 4, 2016. http://dx.doi.org/10. 359–361.
1056/NEJMoa1602412.
Nguyen, Q., Morrow, C., Novick, L., Cambareri, C., Olson, B., Aubry, R., Sne-
Brien, J.D., Lazear, H.M., and Diamond, M.S. (2013). Propagation, quantifica- deker, J., Anand, M., Huang, C., Morse, D., et al.; Centers for Disease Control
tion, detection, and storage of west nile virus. Current protocols in microbi- and Prevention (CDC) (2002). Intrauterine West Nile virus infection–New York,
ology 31, 15D 13 11–15D 13 18. 2002. MMWR Morb. Mortal. wkly. Rep. 51, 1135–1136.
Calvet, G., Aguiar, R.S., Melo, A.S.O., Sampaio, S.A., de Filippis, I., Fabri, A., Oehler, E., Watrin, L., Larre, P., Leparc-Goffart, I., Lastère, S., Valour, F., Bau-
Araujo, E.S.M., de Sequeira, P.C., de Mendonça, M.C.L., de Oliveira, L., et al. douin, L., Mallet, H., Musso, D., and Ghawche, F. (2014). Zika virus infection
(2016). Detection and sequencing of Zika virus from amniotic fluid of fetuses complicated by Guillain-Barre syndrome–case report, French Polynesia,
with microcephaly in Brazil: a case study. Lancet Infect. Dis. S1473-3099(16) December 2013. Euro Surveill. 19, 20720.
00095-5.
Oliphant, T., Nybakken, G.E., Engle, M., Xu, Q., Nelson, C.A., Sukupolvi-Petty,
Cao, B., and Mysorekar, I.U. (2014). Intracellular bacteria in placental basal S., Marri, A., Lachmi, B.-E., Olshevsky, U., Fremont, D.H., et al. (2006). Anti-
plate localize to extravillous trophoblasts. Placenta 35, 139–142. body recognition and neutralization determinants on domains I and II of
Carteaux, G., Maquart, M., Bedet, A., Contou, D., Brugières, P., Fourati, S., West Nile Virus envelope protein. J. Virol. 80, 12149–12159.
Cleret de Langavant, L., de Broucker, T., Brun-Buisson, C., Leparc-Goffart, Pinto, A.K., Brien, J.D., Lam, C.-Y.K., Johnson, S., Chiang, C., Hiscott, J.,
I., and Mekontso Dessap, A. (2016). Zika Virus Associated with Meningoen- Sarathy, V.V., Barrett, A.D., Shresta, S., and Diamond, M.S. (2015). Defining
cephalitis. N. Engl. J. Med. 374, 1595–1596. New Therapeutics Using a More Immunocompetent Mouse Model of Anti-
Cauchemez, S., Besnard, M., Bompard, P., Dub, T., Guillemette-Artur, P., Eyr- body-Enhanced Dengue Virus Infection. MBio 6, e01316–e15.
olle-Guignot, D., Salje, H., Van Kerkhove, M.D., Abadie, V., Garel, C., et al. Rasmussen, S.A., Jamieson, D.J., Honein, M.A., and Petersen, L.R. (2016).
(2016). Association between Zika virus and microcephaly in French Polynesia, Zika Virus and Birth Defects - Reviewing the Evidence for Causality.
2013–15: a retrospective study. The Lancet, epub ahead of print. N. Engl. J. Med. Published online April 13, 2016. http://dx.doi.org/10.1056/
Chaturvedi, U.C., Mathur, A., Chandra, A., Das, S.K., Tandon, H.O., and Singh, NEJMsr1604338.
U.K. (1980). Transplacental infection with Japanese encephalitis virus. Robbins, J.R., and Bakardjiev, A.I. (2012). Pathogens and the placental
J. Infect. Dis. 141, 712–715. fortress. Curr. Opin. Microbiol. 15, 36–43.
Chervenak, F.A., Rosenberg, J., Brightman, R.C., Chitkara, U., and Jeanty, P.
Rossi, S.L., Tesh, R.B., Azar, S.R., Muruato, A.E., Hanley, K.A., Auguste, A.J.,
(1987). A prospective study of the accuracy of ultrasound in predicting fetal
Langsjoen, R.M., Paessler, S., Vasilakis, N., and Weaver, S.C. (2016). Charac-
microcephaly. Obstet. Gynecol. 69, 908–910.
terization of a Novel Murine Model to Study Zika Virus. Am. J. Trop. Med. Hyg.
Delorme-Axford, E., Sadovsky, Y., and Coyne, C.B. (2014). The Placenta as a 16-0111.
Barrier to Viral Infections. Annu Rev Virol 1, 133–146.
Sarathy, V.V., White, M., Li, L., Gorder, S.R., Pyles, R.B., Campbell, G.A., Milli-
Edwards, M.S., Popek, E.J., Wise, B., Hatzenbuehler, L., Arunachalam, A.R., gan, G.N., Bourne, N., and Barrett, A.D.T. (2015). A lethal murine infection
and Hair, A.B. (2015). Ascending in utero herpes simplex virus infection in an model for dengue virus 3 in AG129 mice deficient in type I and II interferon re-
initially healthy-appearing premature infant. Pediatr. Dev. Pathol. 18, 155–158. ceptors leads to systemic disease. J. Virol. 89, 1254–1266.
Faria, N.R., Azevedo, Rdo.S., Kraemer, M.U.G., Souza, R., Cunha, M.S., Hill, Sarno, M., Sacramento, G.A., Khouri, R., do Rosário, M.S., Costa, F., Arch-
S.C., Thézé, J., Bonsall, M.B., Bowden, T.A., Rissanen, I., et al. (2016). Zika vi- anjo, G., Santos, L.A., Nery, N., Jr., Vasilakis, N., Ko, A.I., and de Almeida,
rus in the Americas: Early epidemiological and genetic findings. Science 352, A.R. (2016). Zika Virus Infection and Stillbirths: A Case of Hydrops Fetalis, Hy-
345–349. dranencephaly and Fetal Demise. PLoS Negl. Trop. Dis. 10, e0004517.
Finlay, B.L., and Darlington, R.B. (1995). Linked regularities in the development Semple, B.D., Blomgren, K., Gimlin, K., Ferriero, D.M., and Noble-Haeusslein,
and evolution of mammalian brains. Science 268, 1578–1584. L.J. (2013). Brain development in rodents and humans: Identifying bench-
Hwang, S.Y., Hertzog, P.J., Holland, K.A., Sumarsono, S.H., Tymms, M.J., marks of maturation and vulnerability to injury across species. Prog. Neurobiol.
Hamilton, J.A., Whitty, G., Bertoncello, I., and Kola, I. (1995). A null mutation 106-107, 1–16.
in the gene encoding a type I interferon receptor component eliminates anti- Sheehan, K.C., Lai, K.S., Dunn, G.P., Bruce, A.T., Diamond, M.S., Heutel, J.D.,
proliferative and antiviral responses to interferons alpha and beta and alters Dungo-Arthur, C., Carrero, J.A., White, J.M., Hertzog, P.J., et al. (2006). Block-
macrophage responses. Proc. Natl. Acad. Sci. USA 92, 11284–11288. ing monoclonal antibodies specific for mouse IFN-alpha/beta receptor subunit
Lanciotti, R.S., Kosoy, O.L., Laven, J.J., Velez, J.O., Lambert, A.J., Johnson, 1 (IFNAR-1) from mice immunized by in vivo hydrodynamic transfection. Jour-
A.J., Stanfield, S.M., and Duffy, M.R. (2008). Genetic and serologic properties nal of interferon & cytokine research 26, 804–819.
of Zika virus associated with an epidemic, Yap State, Micronesia, 2007. Sheehan, K.C., Lazear, H.M., Diamond, M.S., and Schreiber, R.D. (2015).
Emerg. Infect. Dis. 14, 1232–1239. Selective Blockade of Interferon-a and -b Reveals Their Non-Redundant
Lazear, H.M., and Diamond, M.S. (2016). Zika Virus: New Clinical Syndromes Functions in a Mouse Model of West Nile Virus Infection. PLoS ONE 10,
and Its Emergence in the Western Hemisphere. J. Virol. 90, 4864–4875. e0128636.
1090 Cell 165, 1081–1091, May 19, 2016

Simmons, D.G., and Cross, J.C. (2005). Determinants of trophoblast lineage and Published online March 4, 2016. http://dx.doi.org/10.1016/j.stem.2016.
cell subtype specification in the mouse placenta. Dev. Biol. 284, 12–24. 02.016.
Staples, J.E., Dziuban, E.J., Fischer, M., Cragan, J.D., Rasmussen, S.A., Can- Ventura, C.V., Maia, M., Bravo-Filho, V., Góis, A.L., and Belfort, R., Jr. (2016).
non, M.J., Frey, M.T., Renquist, C.M., Lanciotti, R.S., Muñoz, J.L., et al. (2016). Zika virus in Brazil and macular atrophy in a child with microcephaly. Lancet
Interim Guidelines for the Evaluation and Testing of Infants with Possible 387, 228.
Congenital Zika Virus Infection - United States, 2016. MMWR Morb. Mortal.
wkly. Rep. 65, 63–67. Watson, E.D., and Cross, J.C. (2005). Development of structures and transport
functions in the mouse placenta. Physiology (Bethesda) 20, 180–193.
Tang, H., Hammack, C., Ogden, S.C., Wen, Z., Qian, X., Li, Y., Yao, B.,
Shin, J., Zhang, F., Lee, E.M., et al. (2016). Zika Virus Infects Human Zeldovich, V.B., and Bakardjiev, A.I. (2012). Host defense and tolerance:
Cortical Neural Progenitors and Attenuates Their Growth. Cell Stem Cell. unique challenges in the placenta. PLoS Pathog. 8, e1002804.
Cell 165, 1081–1091, May 19, 2016 1091

Article
Microbial Reconstitution Reverses

Maternal Diet-Induced Social
and Synaptic Deficits in Offspring
Shelly A. Buffington,1,2 Gonzalo Viana Di Prisco,1,2 Thomas A. Auchtung,3,4 Nadim J. Ajami,3,4 Joseph F. Petrosino,3,4
and Mauro Costa-Mattioli1,2,*
1Department of Neuroscience
2Memory and Brain Research Center
3Alkek Center for Metagenomics and Microbiome Research
4Department of Molecular Virology and Microbiology
Baylor College of Medicine, Houston, TX 77030, USA

*Correspondence: costamat@bcm.edu
SUMMARY the lower gut and lives in a symbiotic relationship with the
host. Maternal obesity has been associated with alterations in
Maternal obesity during pregnancy has been associ- the gut microbiome in offspring in both human and non-human
ated with increased risk of neurodevelopmental dis- primates (Galley et al., 2014; Ma et al., 2014). In addition, some
orders, including autism spectrum disorder (ASD), individuals with neurodevelopmental disorders, including ASD,
in offspring. Here, we report that maternal high-fat co-present with gastrointestinal problems and dysbiosis of the
diet (MHFD) induces a shift in microbial ecology gut microbiota (Bresnahan et al., 2015; Mayer et al., 2014; Parra-
cho et al., 2005). Given the large body of preclinical literature
that negatively impacts offspring social behavior. So-
supporting the notion that a bidirectional communication system
cial deficits and gut microbiota dysbiosis in MHFD between the gut and the brain—known as the gut-brain axis—
offspring are prevented by co-housing with offspring links gut and brain activities (Cryan and Dinan, 2012; Mayer
of mothers on a regular diet (MRD) and transferable et al., 2015), it has been speculated that changes in the gut mi-
to germ-free mice. In addition, social interaction crobiome may be relevant to the development of behavioral
induces synaptic potentiation (LTP) in the ventral symptoms associated with ASD (Hsiao et al., 2013; Mayer
tegmental area (VTA) of MRD, but not MHFD et al., 2014). However, how changes in bacteria that inhabit
offspring. Moreover, MHFD offspring had fewer the intestine could influence brain development and function
oxytocin immunoreactive neurons in the hypothala- remains unknown.
mus. Using metagenomics and precision microbiota Here we report that maternal high-fat diet (MHFD)-induced
reconstitution, we identified a single commensal obesity in mice is associated with social behavioral deficits,
which are mediated by alterations in the offspring gut micro-
strain that corrects oxytocin levels, LTP, and social
biome. Notably, we also found that MHFD-induced changes
deficits in MHFD offspring. Our findings causally in the offspring gut microbiota block long-lasting neural adap-
link maternal diet, gut microbial imbalance, VTA plas- tation in the mesolimbic dopamine reward system (ventral
ticity, and behavior and suggest that probiotic treat- tegmental area [VTA]). Moreover, oral treatment with a single
ment may relieve specific behavioral abnormalities commensal bacterial species corrects oxytocin levels and
associated with neurodevelopmental disorders. synaptic dysfunction in the VTA and selectively reverses social
deficits in MHFD offspring.
INTRODUCTION
RESULTS
Recent epidemiological evidence suggests that exposure to
maternal obesity in utero increases the risk of neurodevelopmen- Social Behaviors are Impaired in MHFD Offspring
tal disorders, such as autism spectrum disorder (ASD) in children To investigate how maternal diet-induced obesity affects
(Connolly et al., 2016; Krakowiak et al., 2012; Sullivan et al., offspring neurodevelopment, female mice were fed either regular
2014). Given the increase in the prevalence of obesity (Skinner diet (RD) or high-fat diet (HFD) for 8 weeks, a standard period
and Skelton, 2014), it is important to understand the neurobio- required to reach a state of diet-induced obesity in mice (Aye
logical mechanism by which maternal obesity affects offspring et al., 2015). Females were then paired with males to produce
behavior and brain function. offspring that were given regular diet after weaning (Figure 1A).
The amount and type of dietary macronutrients strongly influ- As expected, MHFD significantly increased maternal weight (Fig-
ence the intestinal microbiota (Tremaroli and Bäckhed, 2012), ures S1A–S1C). Consistent with reports of more frequent spon-
which consists of a vast bacterial community that resides in taneous abortion in obese mothers (King, 2006), the litter size
1762 Cell 165, 1762–1775, June 16, 2016 ª 2016 Elsevier Inc.
A B C D
E F G
H I
Figure 1. Social Deficits and Dysbiosis of the Gut Microbiota in MHFD Offspring
(A) Schematic of the maternal diet regimen and breeding.
(B) Schematic of the reciprocal social interaction task.
(C and D) MHFD offspring showed reduced reciprocal interaction (C, p < 0.0001, t = 7.90; D, p < 0.001, t = 5.89).
(E) Schematic of the three-chamber social interaction task.
(F–G) In the sociability test, MRD offspring spent more time interacting with a mouse than with an empty wire cage (F, p < 0.0001, t = 8.817), whereas MHFD
offspring showed no preference for the mouse (F, p = 0.48, t = 1.19; maternal diet effect F1,52 = 6.08, p < 0.05). In the social novelty test, unlike MRD (G, p < 0.0001,
t = 6.68), MHFD offspring had no preference for interacting with a novel versus a familiar mouse (G, p = 0.086, t = 2.08; maternal diet effect F1,52 = 34.96,
p < 0.0001).
(H and I) Representative exploratory activity of MRD (H) and MHFD (I) offspring in the three-chamber test.
(J) Principal coordinates analysis (PCoA) of unweighted UniFrac distances from the averaged rarefied 16S rRNA gene dataset (n = 1,000 rarefactions; 7,617
reads/sample) showed that MRD samples clustered separately from MHFD samples (p < 0.001, R2 = 0.37). Plots show mean ± SEM. See also Figures S1 and S2.
was reduced (Figure S1D) and latency to first litter increased in S1G), the time at which behavioral and electrophysiological ex-
female mice fed HFD (Figure S1E). It is noteworthy that there periments were performed.
was no significant difference in offspring weight between Given that maternal obesity has been associated with increased
maternal diet cohorts at 7–12 weeks of age (Figures S1F and risk for neurodevelopmental disorders including ASD in offspring
Cell 165, 1762–1775, June 16, 2016 1763

(Bilder et al., 2013; Krakowiak et al., 2012; Moss and Chugani, four MRD mice (Figure 2B). Fecal samples were collected and so-
2014) and deficient social interactions are a salient behavioral cial behavior in MRD and MHFD offspring was assessed when
feature of ASD (Mefford et al., 2012), we studied social behavior mice were 7–8 weeks old. Strikingly, MHFD mice co-housed
in maternal regular diet (MRD) and MHFD offspring. First, we as- with MRD mice exhibited normal reciprocal social interactions
sessed reciprocal social interactions by recording the amount of (Figures 2C and 2D and S3A–S3C), as well as normal sociability
time a pair of mice, unfamiliar with each other, spent interacting and preference for social novelty, as determined by the 3-cham-
in a neutral arena (Figure 1B). When compared to MRD offspring, ber test (Figures 2E, 2F, S3D and S3E). Thus, co-housing with con-
MHFD offspring had fewer reciprocal interactions (Figures 1C and trol mice corrects social deficits in MHFD offspring.
1D, S1H, and S1I). Next, we used the three-chamber test (Silver- We next examined whether co-housing also corrected the
man et al., 2010) to assess (1) sociability by comparing the time changes in the microbiota of MHFD offspring. Indeed, co-hous-
mice spent interacting with an empty wire cage versus one con- ing caused a shift in the bacterial phylogenetic profile of MHFD
taining a mouse and (2) preference for social novelty by measuring mice to resemble that of MRD or MRD co-housed mice (Fig-
the time mice spent interacting with a familiar versus a stranger ure 2G), thus correcting the MHFD-induced alterations in the
mouse (Figure 1E). Consistent with the results from reciprocal so- commensal microbiota.
cial interactions, MHFD offspring had impaired sociability and In agreement with the idea that the fecal microbiota of MHFD
showed no preference for social novelty (Figures 1F–1I, S1J, offspring lacks one or more beneficial bacterial species required
and S1K). Taken together these data indicate that MHFD offspring for normal social behavior, co-housing one MRD with three
display social deficits. MHFD offspring was sufficient to rescue both the social behav-
iors and microbiota phylogenetic profile of MHFD offspring (Fig-
Dysbiosis of the Gut Microbiota in MHFD Offspring ures S3F–S3L).
A variety of factors could contribute to the etiology of MHFD-
induced social behavioral abnormalities. However, maternal Colonization of Germ-Free Mice with the Microbiota
obesity has been shown to alter the gut microbiome of offspring from MRD, but Not MHFD Mice, Reverses Their Deficient
(Galley et al., 2014; Ma et al., 2014) and individuals diagnosed Social Behavior
with ASD can co-present dysbiosis of the gut microbiota (Bresna- Studies on germ-free (GF) mice have shown that the intestinal
han et al., 2015; Mayer et al., 2014; Parracho et al., 2005). To microbiota can influence brain development and function (Cryan
examine whether MHFD induces alterations in offspring gut mi- and Dinan, 2012). We hypothesized that, if the lack of one or
crobiota, we analyzed the bacterial composition and community more bacterial species in the microbiota of MHFD offspring is
structure in the feces of MRD and MHFD offspring by 16S ribo- responsible for their defective social behavior, GF mice should
somal RNA (rRNA) gene sequencing. The microbial communities also be socially deficient. Confirming this hypothesis and in
in both MRD and MHFD offspring were comprised of a typical keeping with recent results (Desbonnet et al., 2014), social be-
mouse gut microbiota, dominated by Bacteroidetes and Firmi- haviors were impaired in GF mice (Figures 3A–3D and S4A–S4D).
cutes (Figures S2A–S2D). While bacterial diversity computed To identify functional differences between gut microbial com-
based on weighted UniFrac distances (the assessment of communities and determine whether their role is causal, we trans-
munity structure by considering abundance of operational taxo- planted (gavaged) fecal microbiota from adult MRD and MHFD
nomic units [OTUs]) did not differ significantly between the offspring into 4-week- (Figure 3E) and 8-week-old (Figure 3F)
offspring from either diet group (Figure S2E), unweighted analyses GF mice. Interestingly, GF mice that received fecal microbiota
of UniFrac distances (assessment of community structure by from MRD offspring at weaning (4 weeks; Figures 3G, 3H, S4E
considering only OTU presence/absence) revealed a marked dif- and S4F), but not at 8 weeks (Figures 3I, 3J, S4G, and S4H),
ference between the structures of the bacterial communities (Fig- showed normal social behavior. By contrast, GF mice that
ure 1J). Moreover, the diversity of microbiota in MHFD offspring received fecal microbiota from MHFD offspring remained so-
was reduced compared to MRD microbiota (Figure S2F). cially impaired, regardless of the age at which the fecal transfer
Consistent with previous reports (Turnbaugh et al., 2006), an was performed (Figures 3G–3J and S4E–S4H). Moreover, as
HFD regimen in mothers induced a remarkable change in the in the case of the phylogenetic separation of MRD and MHFD
maternal microbiome composition and diversity (Figures S2G microbiota (Figure 1J), the bacterial communities in GF mice
and S2H), which was similar to that observed in their offspring receiving feces from MHFD donor offspring clustered separately
(Figures 1J and S2F). from those of GF mice receiving feces from MRD donor
offspring, irrespective of whether the fecal transplant was per-
Gut Microbiota Mediate MHFD-Induced Social Deficits formed at 4 or 8 weeks (Figures 3K, 3L, and S4I–S4N). These
While microbial communities vary across individuals (Yatsunenko data reveal a neurodevelopmental window during which microbi-
et al., 2012), co-housed family members are known to share their al reconstitution effectively improves social behavior.
microbiota (Song et al., 2013). Since mice are coprophagic and
transfer gut microbiota between each other by the fecal-oral route MHFD Negatively Impacts a Subset of Bacteria in the
(Ridaura et al., 2013), we examined whether co-housing MHFD Intestinal Gut and Selective Re-introduction of
with MRD mice prevents the social deficits in MHFD offspring. Lactobacillus (L.) reuteri Restores Social Deficits in
To this end, at weaning (3 weeks) an MHFD mouse was co-housed MHFD Offspring
with three MRD mice (Figures 2A and 2B). Control groups con- To investigate which bacterial species are absent in the micro-
sisted of individual cages containing either four MHFD mice or biota community of MHFD offspring, we performed metagenomic
1764 Cell 165, 1762–1775, June 16, 2016

A C D
Contact duration (s)

1.5
Interaction time (s)

g 150
*** *** * *
in
pr
fs RD
s
f se
f o ed a ly 100 1.0
o ,f MRD (n=3 pairs)
an
ng g al Co-housed MRD (n=3 pairs)
u si anin i or ed 50 0.5 MHFD (n=3 pairs)
ho e v t Co-housed MHFD (n=3 pairs)
o- t w ha itia
C a Be in 0 0.0
4wks
FD
FD
FD
FD
R
R
R
R
Fe for
H
H
H
M
M
M
M
M
M
M
M
c a 16
ed
ed
ed
ed
ls S
us
us
us
us
am rR
ho
ho
ho
ho
p l NA
o-
o-
o-
o-
es s
C
C
C
co eq
lle
ct
Sociability Social Novelty

B E F
de
250 200

MRD cage ****
4 MRD 200 *** **** 150 *** ***
*** MRD (n=4)
MHFD cage 150 Co-housed MRD (n=6)
100
4 MHFD 100
MHFD (n=4)
Co-housed MHFD (n=6)
Co-housing cage 50 50
3 MRD 1 MHFD 0 0
M se 1
M us 2
M se 1
ou 2
M se 1
M us 2
se 1
2
M mp 1
M mp 1
M mp 1
1
o u ty
o u ty
o u ty
o u ty
ou e
ou e
ou e
ou e
E e
E e
E e
se
M mp
M us
M s
s
s
E
o
M
G
MRD (n=10)
PC2 (11.7% variation explained)
Co-housed MRD (n=5)

0.2 MHFD (n=10)
Co-housed MHFD (n=5)
0.1
0.0
–0.1
–0.2
–0.2 –0.1 0.0 0.1 0.2

Figure 2. Co-housing MHFD with MRD Offspring Rescues Both Social Dysfunction and the Microbiota Phylogenetic Profile of MHFD Mice
(A) Schematic of the co-housing experiment.
(B) MRD and MHFD offspring were weaned into one of three cage compositions.
(C–G) Social interaction time (C, MRD versus MHFD p < 0.001, t = 9.30; MRD versus co-housed MHFD p > 0.99, t = 0.31; MHFD versus co-housed MHFD
p < 0.001, t = 7.99; p < 0.0001, F3,8 = 30.51) and contact duration (D, MRD versus MHFD p < 0.05, t = 4.13; MRD versus co-housed MHFD p > 0.99, t = 0.46; MHFD
versus co-housed MHFD p < 0.05, t = 4.59; p < 0.001, F3,8 = 9.01) in the reciprocal interaction test; social interaction times in the sociability (E, MRD p < 0.001, t =
4.36; MHFD p > 0.99, t = 0.078; Co-housed MRD p < 0.0001, t = 6.33; Co-housed MHFD p < 0.001, t = 4.78; Maternal diet/ Housing/Interaction effect F3,32 = 6.13,
p < 0.01) and social novelty tests (F, MRD p < 0.0001, t = 5.12; MHFD p > 0.99, t = 0.60; Co-housed MRD p < 0.001, t = 4.20; Co-housed MHFD p < 0.001, t = 4.76;
maternal diet/housing/interaction effect F3,32 = 4.37, p < 0.01), as well as UniFrac-based phylogenetic clustering (G, p < 0.001, R2 = 0.552; n = 1,000 rarefactions;
3,390 reads/sample), are all restored in MHFD offspring co-housed with MRD mice. Plots show mean ± SEM. See also Figure S3.
shotgun sequencing of fecal samples from both MHFD and MRD was tested (Figure 4A). Remarkably, treatment with L. reuteri
offspring. Our analysis identified several species whose relative significantly improved sociability and preference for social nov-
abundance was dramatically reduced in the MHFD offspring elty in MHFD offspring (Figures 4B, 4C, 4E, S5A, and S5B). Re-
microbiota (Table 1). Among these, L. reuteri was the most dras- sults from several control experiments underscore the specificity
tically reduced (>9-fold) in the MHFD microbiota population, of L. reuteri-mediated rescue of social behaviors in MHFD
compared to the MRD microbiota (Table 1). offspring. First, drinking water treated with either resuspension
L. reuteri has been shown to promote oxytocin levels (Pouta- media or heat-killed L. reuteri (80 C for 20 min) failed to restore
hidis et al., 2013), a hormone that plays a crucial role in social social behavior in MHFD offspring (Figures 4B–4D, S5A, and
behaviors (Donaldson and Young, 2008). We hypothesized that S5B). Second, drinking water with live L. reuteri did not change
the selective decrease in L. reuteri in the microbiota of MHFD the social behavior of MRD offspring (Figures 4B, 4C, S5A, and
offspring was causally related to their social deficits. To test S5B), presumably because their gut microbiota already contains
this hypothesis, we introduced L. reuteri into the drinking water ample L. reuteri. Finally, addition of L. reuteri to the drinking wa-
of MHFD offspring at weaning for 4 weeks, after which behavior ter had no major effect on bacterial viability and the heat-killing
Cell 165, 1762–1775, June 16, 2016 1765

procedure completely abrogated colony-forming units (Fig- larly rewarding and triggers synaptic potentiation in VTA DA neu-
ure S5C). Importantly, the amelioration of the deficient social rons of birds (Huang and Hessler, 2008), we examined whether
behavior is specific to L. reuteri since similar treatment with direct social interaction evokes long-term potentiation (LTP) of
another Lactobacillus species, L. johnsonii, whose abundance synaptic inputs to VTA DA neurons (Figures S7A–S7C). To
is also reduced in the gut microbiota of MHFD offspring (Table 1), this end, we recorded AMPAR/NMDAR ratios of glutamatergic
failed to rescue social behaviors in MHFD offspring (Figures excitatory postsynaptic currents (EPSCs) in MRD and MHFD
S5D–S5M). offspring 24 hr following a 10 min reciprocal interaction with
MHFD offspring also show other behavioral traits that are either a stranger or a familiar mouse (Figure 5A). In control
associated with ASD, like repetitive behaviors and anxiety (Fig- MRD mice, social interaction with a stranger, but not a familiar
ures S6). Interestingly, while co-housing MHFD with MRD mouse, triggered LTP in VTA DA neurons, as determined
offspring restores social behavior (Figures 2C and 2D), it failed by an increase in AMPAR/NMDAR ratios (Figures 5B and 5D).
to rescue marble burying (Figure S6B), a behavioral task reflect- By contrast, in MHFD offspring, social interaction with a stranger
ing repetitive and perseverative behavior (Thomas et al., 2009). failed to induce LTP in their VTA DA neurons (Figures 5C and 5E).
Accordingly, GF mice also showed increased marble burying, Input-output curves, paired-pulse ratios and miniature EPSCs
and fecal microbial transplants from MRD (or MHFD) offspring (mEPSCs) frequency and amplitude show that the impairment
into GF mice failed to reverse the repetitive behavior (Figure S6B). of LTP induced by social interaction in MHFD offspring cannot
Thus, these data suggest that repetitive behaviors in MHFD be attributed to changes in basal synaptic transmission (Figures
offspring do not depend on changes in the gut microbiome. In S7D–S7H).
addition, L. reuteri treatment had no effect on anxiety in MHFD Mirroring the electrophysiological results, MRD offspring
offspring (Figures S6C–S6H). Taken together, these data indi- spent significantly more time interacting with a stranger than a
cate that L. reuteri reconstitution specifically rescues social, familiar mouse, but MHFD offspring did not (Figures 5F and
but not other behavioral endophenotypes associated with ASD. 5G). Thus, social interaction induces a long-lasting increase in
the activity of the dopaminergic reward system of MRD, but
Oxytocin Levels Are Reduced in the Hypothalamus of not in MHFD, offspring.
MHFD Offspring
There is growing evidence that the neuropeptide oxytocin mod- Treatment with L. reuteri or Oxytocin Reverses Both the
ulates numerous aspects of social behavior (Donaldson and LTP in VTA DA Neurons and Social Behavior in MHFD
Young, 2008) and is implicated in ASD (Lerer et al., 2008; Wu Offspring
et al., 2005). L. reuteri, which rescued social behaviors in We next wondered whether L. reuteri treatment, which restores
MHFD mice (Figures 4B, 4C, and 4E), has been reported to sociability and preference for social novelty in MHFD offspring
increase oxytocin levels (Poutahidis et al., 2013). Because (Figures 4B, 4C, and 4E), would also rescue reciprocal social
oxytocin is primarily synthesized in the paraventricular nuclei interaction and related changes in synaptic strength in the
(PVN) of the hypothalamus, we decided to compare the number VTA. Live (Figures 5H and 5J), but not heat-killed (Figures 5I
of oxytocin-expressing cells in the PVN of MRD and MHFD and 5K), L. reuteri rescued stranger interaction-induced LTP in
offspring. Interestingly, MHFD offspring had significantly fewer the VTA as well as reciprocal social interactions in MHFD
oxytocin immunoreactive neurons compared to MRD offspring offspring (Figure 5L). Thus, L. reuteri restores social interac-
(Figures 4F–4H, 4K, and 4L). The reduction in oxytocin immuno- tion-induced LTP in the VTA of MHFD offspring.
reactivity was not due to an overall decrease in PVN neurons, These findings, together with the fact that L. reuteri treatment
since the total number of neurons was unchanged (as measured increased oxytocin immunoreactivity in the PVN of MHFD
by NeuN staining; Figures 4G, 4H, 4M, and 4N). Notably, in offspring (Figures 4I and 4J, 4O, and 4P), led us to examine
L. reuteri-treated MHFD offspring, the number of oxytocin-ex- whether direct oxytocin application could also reverse the
pressing cells was higher than in control-treated MHFD offspring behavioral and electrophysiological deficits characteristic of
(Figures 4I, 4J, 4O, and 4P). Thus, the number of oxytocin immu- MHFD offspring. To test this idea, we administered oxytocin
noreactive neurons in the PVN is reduced in MHFD offspring but intranasally—a preferred method of administering neuropep-
can be restored by L. reuteri treatment. tides to the brain bypassing more invasive procedures (Peñagar-
ikano et al., 2015)—to MHFD offspring and measured reciprocal
Mesolimbic Dopamine Reward System Function Is social interactions 30 min later. Although either oxytocin alone
Impaired in MHFD Offspring or social interaction alone failed to rescue social interaction-
Brain regions that respond to naturally rewarding stimuli, induced LTP in the VTA, the combination of social interaction
including the ventral tegmental area (VTA) and the nucleus ac- and oxytocin treatment restored LTP in the VTA of MHFD
cumbens (NAc), are crucially involved in social behaviors (Dölen offspring (Figures 6A and 6B), supporting prior work implicating
et al., 2013; Gunaydin et al., 2014). In addition, oxytocin-ex- a synergistic effect of oxytocin and dopamine in the processing
pressing neurons in the PVN project to the VTA (Melis et al., of socially relevant cues (Modi and Young, 2012). Accordingly,
2007). Oxytocin activates VTA neurons in both mice and oxytocin treatment improved reciprocal social interaction (Fig-
humans, influencing the processing of socially relevant cues ures 6C-6F), as well as sociability and the preference for social
(Groppe et al., 2013; Tang et al., 2014) and oxytocin receptor novelty (Figures 6G-6J). Thus, oxytocin administration rescues
blockade in the VTA prevents social attachment in rodents (Ped- social behavior and related neural adaptations in the VTA of
ersen et al., 1994). Given that social stimulation can be particu- MHFD offspring. Collectively, our data show that MHFD impairs
1766 Cell 165, 1762–1775, June 16, 2016

A B C Sociability D Social Novelty

150 1.5 250 250


**** *** 200 **** 200
100 1.0 **
150 150
100 100
50 0.5
50 50
0 0.0 0 0
s y
M m 1
se y
1
M use 1
M us 2
se 1
2
l
l
F
F
tro
tro
o u pt
o u pt
E e
G
o e
ou e
M m
M us
on
on
o
C
M
Control (n=3 pairs) Control (n=7)
Germ free (n=3 pairs) Germ free (n=7)
T
rn
rn
FM
bo
bo
E F
d,
g
d
in
rin
ne
ne
or
or
pr
p
ea
ea
vi
vi
fs
fs
ha
ha
w
of
of
T
FM
F
F
Be
Be
F
F
G
G
G
G
4wks 3wks 4wks 4wks 3wks
FMT of 4-week-old GF mice FMT of 8-week-old GF mice
G Sociability H Social Novelty I Sociability J Social Novelty

200 200
200 200

**** 150 150

150 150 **
100 100 100 100
50 50 50 50
0 0 0 0
M mp 1
M se 1
M us 2
se 1
2
o u ty
o u ty
M mp 1
M se 1
M us 2
se 1
2
o u ty
o u ty
E e
se
M mp
ou e
ou e
E e
se
ou e
ou e
M mp
M us
s
M us
E
E
o
o
M
M
GF MRD Col, 4 wks (n=8) GF MRD Col, 8 wks (n=4)

GF MHFD Col, 4 wks (n=8) GF MHFD Col, 8 wks (n=4)
K FMT of 4-week-old GF mice L FMT of 8-week-old GF mice
GF MRD Col, 8 wks (n=4)

GF MHFD Col, 8 wks (n=4))

0.2
0.1
0.1
0.0 0.0
–0.1
–0.1
–0.2
GF MRD Col, 4 wks (n=4)
GF MHFD Col, 4 wks (n=4)
–0.4 –0.2 0.0 0.2 0.4 –0.3 –0.2 –0.1 0.0 0.1 0.2 0.3
PC1 (42.7% variation explained) PC1 (55.4% variation explained)
24h post-colonization 14d post-colonization 56d post-colonization
7d post-colonization 28d post-colonization
Cell 165, 1762–1775, June 16, 2016 1767

Table 1. Species Whose Abundance Is Reduced in the Gut Microbiota of MHFD Offspring
Species of Interest MRD Representation MHFD Representation Fold Change MRD/MHFD
Lactobacillus reuteri 7.49 ± 3.0 0.879 ± 0.21 9.24 ± 0.65
Parabacteroides distasonis 0.00709 ± 0.0055 0.00126 ± 0.0011 5.63 ± 1.17
Helicobacter hepaticus 7.35 ± 2.4 2.58 ± 1.3 2.84 ± 0.61
Bacteroides uniformis 5.49 ± 2.2 2.07 ± 0.78 2.65 ± 0.56
Olsenella unclassified 0.230 ± 0.064 0.121 ± 0.031 1.90 ± 0.38
Collinsella unclassified 0.0866 ± 0.031 0.0494 ± 0.016 1.75 ± 0.48
Bifidobacterium pseudolongum 19.4 ± 3.3 11.3 ± 2.4 1.71 ± 0.27
Lactobacillus johnsonii 24.5 ± 6.2 17.1 ± 5.2 1.43 ± 0.40
oxytocin-mediated synaptic adaptations in the VTA that underlie phylogenetic profile (Figures 2 and S3). Fifth, GF mice are
social behaviors. socially impaired and fecal microbiota transplanted from MRD
(but not MHFD) offspring rescues GF social behavior (Figures 3
DISCUSSION and S4). Finally, treatment with a single bacterial species,
L. reuteri, which is dramatically reduced in MHFD offspring
Both genetic and environmental factors, and their interactions, (Table 1), selectively restores social behavior in MHFD mice
play a crucial role in the etiology of neurodevelopmental disor- (Figures 4 and S5A–S5C).
ders including ASD (Hallmayer et al., 2011). There is growing We propose a model in which L. reuteri improves social
epidemiological evidence that maternal obesity heightens the behavior by promoting oxytocin-mediated functions. Consistent
risk of neuropsychiatric disorders in offspring (Krakowiak et al., with this model, L. reuteri-treatment enhances oxytocin levels in
2012; Sullivan et al., 2014). Indeed, a recent study reported the PVN of MHFD mice (Figures 4I and 4J) and direct oxytocin-
that mothers with obesity were 1.5 times more likely to have a treatment normalizes the social behavior of MHFD offspring
child with ASD, and the increased risk of children with ASD (Figure 6). Although the precise mechanism by which L. reuteri
was two-fold greater for pregnant mothers with both obesity promotes oxytocin in the brain remains to be determined, we
and gestational diabetes (Connolly et al., 2016). favor the idea that the vagus nerve (Davari et al., 2013) could
While most of the focus in the field has been on inflammation be the main pathway of communication between the gut/
(Bolton and Bilbo, 2014) or epigenetic changes (Mathers and L. reuteri and changes in oxytocin in the PVN. It is known that
McKay, 2009), the biological mechanism by which maternal vagal nerve fibers project to the PVN (Sabatier et al., 2013;
obesity affects offspring neurodevelopment remains to be deter- Uvnäs-Moberg et al., 2014). In addition, neuronal activity in the
mined. Here, we show that that the behavioral dysfunction asso- PVN induced by bacterial colonization is blocked by subdiaph-
ciated with MHFD-induced obesity is induced by alterations in ragmatic vagotomy (Wang et al., 2002). Especially relevant are
the offspring gut microbiota. Several lines of evidence support the reports that the L. reuteri-mediated increase in oxytocin de-
this idea. First, some individuals diagnosed with ASD present pends on the vagus nerve (Poutahidis et al., 2013) and that
dysbiosis of the gut microbiota and gastrointestinal issues (Bres- another Lactobacillus species, L. rhamnosus, reduced stress-
nahan et al., 2015; Mayer et al., 2014; Parracho et al., 2005). Sec- induced anxiety in mice in a vagus-dependent manner (Bravo
ond, maternal obesity leads to alterations in the offspring’s gut et al., 2011).
microbiome in humans and non-human primates (Galley et al., Our results provide new insight into the mechanism by which a
2014; Ma et al., 2014). Third, in mice, the gut microbiota of marked shift in microbial ecology, caused by MHFD, can nega-
MHFD offspring is altered (Figure 1J) by the reduction in specific tively impact social behaviors and related neuronal changes in
bacterial species (Table 1). Fourth, manipulation of the micro- offspring. These neuronal adaptations, which underlie social
biome community by co-housing MHFD with MRD offspring res- behavior by enhancing the salience and rewarding value of social
cues MHFD-induced social deficits and corrects their microbial stimuli, are surprisingly impaired by maternal diet-induced
Figure 3. Fecal Microbiota from MRD, but not MHFD, Offspring Improves Germ-Free (GF) Recipient Social Behavior
(A–D) GF mice show reduced reciprocal social interaction (A, p < 0.0001, t = 22.73; B, p < 0.001, t = 11.31) and deficits in sociability (C, Control p < 0.0001, t = 5.30,
GF p > 0.99, t = 0.39; main group effect F1,24 = 21.98, p < 0.0001) and preference for social novelty (D, control p < 0.01, t = 3.64, GF p = 0.39, t = 1.33; main group
effect F1,24 = 5.29, p < 0.05).
(E and F) Schematic of fecal microbiota transplant (FMT) at 4 (E) and 8 weeks of age (F).
(G and H) FMT from MRD, but not MHFD, offspring at weaning restored both GF sociability (G, GFMRDCol p < 0.0001, t = 6.66; GFMHFDCol p = 0.35, t = 1.40; Donor
effect F1,28 = 32.44, p < 0.0001) and preference for social novelty (H, GFMRDCol p < 0.01, t = 3.60; GFMHFDCol p = 0.81, t = 0.84; Donor effect F1,28 = 9.86, p < 0.01).
(I and J) At 8 weeks, FMT from either MRD or MHFD donors failed to improve sociability (I, GFMRDCol p = 0.51, t = 1.20; GFMHFDCol p = 0.28, t = 1.58; Donor effect
F1,12 = 0.07, p = 0.79) or preference for social novelty in GF mice (J, GFMRDCol p = 0.48, t = 1.23; GFMHFDCol p > 0.99, t = 0.043; Donor effect F1,12 = 0.71, p = 0.42).
(K and L) PCoA of unweighted UniFrac distances based on the 16S rRNA gene sequencing dataset from GF recipients of stools from either MRD or MHFD donors
at four (K, p = 0.001, R2 = 0.83; n = 1,000 rarefactions; 4,628 reads/sample) or eight (L, p < 0.001, R2 = 0.77; n = 1,000 rarefactions; 4,805 reads/sample) weeks of
age. Plots show mean ± SEM.
See also Figure S4.
1768 Cell 165, 1762–1775, June 16, 2016

r
A B Sociability C Social Novelty
-o
D by
ed
D
od
250 250

R

R d
at
fo
d te
to
iti
fe la
in
rn
200 **** ****

ll so
200 ***
****
es
at t s m o
,a ,i **** ****
m j e c da g b
ys
et d
****
di n e
150
al
150
b d rin
er w s
an
al e a
Su -fe ffsp
al
100 100
FD o
or
H ect
vi
ha
bj
50 50
Su
Be
0 0
3wks 4wks
o u ty
o u ty
o u ty
o u ty
o u ty
o u ty
M mp 1
M mp 1
M mp 1
M mp 1
M mp 1
M se 1
M us 2
M se 1
M us 2
M se 1
M us 2
M se 1
M us 2
M se 1
M us 2
se 1
2
M mp
E e
E e
E e
E e
E e
se
ou e
ou e
ou e
ou e
ou e
ou e
L. reuteri treatment
M us
E
o
M
MRD + media (n=10) MHFD + media (n=3)
MRD + Heat-killed L. reuteri (n=10) MHFD + Heat-killed L. reuteri (n=8)
MRD + L. reuteri (n=10) MHFD + L. reuteri (n=8)
D Habituation Sociability Social Novelty E Habituation Sociability Social Novelty

MHFD + Heat-killed
MHFD + L. reuteri
L. reuteri
M1 E M1 M2 M1 E M1 M2
F Anterior Posterior Paraventricular Nucleus of the Hypothalamus

MRD Oxytocin
200µm 200µm 200µm 200µm 200µm
Merged Oxytocin NeuN Merged Oxytocin NeuN Merged Oxytocin NeuN

G
MRD
200µm 200µm 200µm
H
MHFD
Merged Oxytocin NeuN Merged Oxytocin NeuN Merged Oxytocin NeuN

I
HkL. reuteri
MHFD +
200µm 200µm 200µm
J
L. reuteri
MHFD +
K ** M O
100 800 100 *
PVN oxytocin+
PVN oxytocin+
80 MRD (n=3 mice) 80 MHFD + Heat-killed L. reuteri

PVN NeuN+
600
cell count
cell count
cell count
MHFD (n=3 mice) (n=3 mice)

60 60
400 MHFD + L. reuteri
40 40 (n=3 mice)
20 200 20
0 0 0
fluorescence intensity (AU)
L 60 N 80 P 60
**
*
PVN oxytocin
PVN oxytocin
60 MRD (N=4 mice, n=42 cells/mouse) MHFD + Heat-killed L. reuteri

PVN NeuN
40 MHFD (N=4 mice, n=42 cells/mouse) 40 (N=3 mice, 42 cells/mouse)

40 MHFD + L. reuteri
(N=3 mice, 42 cells/mouse)
20 20
20
0 0 0
Cell 165, 1762–1775, June 16, 2016 1769

changes in the gut microbiome (Figure 5). Interestingly, accord- during a second 10-min period in which the subject could interact either with
ing to a recent study, probiotic-based restoration of gut perme- an empty wire cup (empty) or a wire cup containing an age and sex-matched
stranger conspecific (mouse 1). Time spent interacting (sniffing, crawling upon)
ability in a mouse model of ASD can improve some behavioral
with either the empty cup or the stranger mouse contained in the other cup as
abnormalities, but not social behaviors (Hsiao et al., 2013). Given well as time spent in each chamber was recorded using the AnyMaze software,
that we identified a different probiotic candidate, L. reuteri, that by independent observers. Empty cup placement in the left or right chamber
rescues social behavior (Figures 4 and 5), but not other during the sociability period was counterbalanced between trials. Finally, pref-
behavioral endophenotypes associated with ASD (Figure S6), erence for social novelty was assayed by introducing a second stranger mouse
in MHFD mice, we propose that a carefully selected combination (mouse 2) into the previously the empty wire cup. Time spent in each chamber
of probiotics may be useful as a potential non-invasive treat- as well as time spent interacting with either mouse 1 or mouse 2 was recorded
using the automated AnyMaze software by independent observers.
ment for patients suffering from neurodevelopmental disorders
including ASD.
Marble Burying
Marble burying was performed as previously described (Thomas et al., 2009).
EXPERIMENTAL PROCEDURES Briefly, mice were placed in a standard-sized cage containing 20 regularly-
spaced black marbles sitting on fine-wood chipped bedding 5 cm in depth.
Mice and Maternal Diet After 20 min, the mouse was removed and marbles with at least two-thirds
C57Bl6/J mice were obtained from Jackson Laboratories (#000-664) and were of their depth obscured by wood chips were counted as buried.
kept on a 12 hr light/dark cycle and had access to food and water ad libitum.
Females were placed on either a regular diet (RD) consisting of 13.4% kcal
Open Field
from fat, 30% kcal from protein, and 57% kcal from carbohydrates (Lab Diets,
Mice were placed in an open arena (40 3 40 3 20 cm) and allowed to explore
#5001) or HFD consisting of 60% kcal from fat, 20% kcal from protein, and
freely for 10 min while their position was continually monitored using tracking
20% kcal from carbohydrates (Research Diets, #D12492). Maternal weight
software (AnyMaze). Tracking allowed for measurement of distance traveled,
was measured weekly. Maternal total and fat mass were measured using
speed, and position in the arena throughout the task. Time spent in the center
an mq7.5 Minispec NMR body composition analyzer. After 8 weeks on diet,
of the arena, defined as the interior 20 3 20 cm, was recorded.
females were paired with C57Bl6/J adult males to produce subject offspring.
Resulting offspring were weaned at 3 weeks of age and all placed on RD,
regardless of maternal diet (MRD or MHFD). Germ-free mice (C57Bl6/J) 16 s rRNA Gene Sequencing
were maintained in a flexible isolator fed with HEPA-filtered air and provided Methods were adapted from protocols developed for the NIH-Human Micro-
with irradiated food and water. Germ-free offspring were weaned at 4 weeks biome Project (Human Microbiome Project Consortium, 2012a, 2012b). For
of age. All behavioral tests were performed on 7- to 12-week-old male mice. a detailed protocol see Supplemental Experimental Procedures.
Animal care and experimental procedures were approved by Baylor College
of Medicine’s Institutional Animal Care and Use Committee in accordance Whole Genome Shotgun Sequencing
with all guidelines set forth by the U.S. National Institutes of Health. Individual libraries constructed from each sample were loaded onto the HiSeq
platform (Illumina) and sequenced using the 2 3 100 bp pair-end read proto-
Reciprocal Social Interaction col. Illumina paired-end libraries were constructed from total genomic DNA
Mice were placed in a 25 3 25 3 25 cm Plexiglass arena, to which they had not isolated from each sample. The DNA was sheared into approximately 400–
been previously habituated, with either a familiar (cage-mate) or stranger age- 600 bp fragments followed by ligation of Illumina adaptors containing molec-
and sex-matched conspecific. In all experiments, paired mice were matched ular barcodes for downstream de-multiplexing. These products were then
for maternal diet, colonization source, and/or treatment. We recorded the amplified through ligation-mediated PCR (LM-PCR) using KAPA HiFi DNA
time a pair of mice engaged in social interaction (close following, touching, Polymerase (Kapa Biosystems, Wilmington, MA, USA). Following bead purifi-
nose-to-nose sniffing, nose-to-anus sniffing, and/or crawling over/under cation with Agencourt AMPure XP (Beckman Coulter, Brea, CA, USA), quanti-
each other). The human observer was blind to maternal diet and/or treatment fication and size distribution of the LM-PCR product was determined using the
group. Social behavior was analyzed with AnyMaze automated software. LabChip GX electrophoresis system (PerkinElmer, Akron, OH, USA). Libraries
were pooled in equimolar amounts at six samples per pool, and prepared for
Three-Chamber Social Test sequencing with TruSeq PE Cluster Generation Kit (Illumina). Each library
Crawley’s three-chamber test for sociability and preference for social novelty pool was loaded onto one lane of a HiSeq 2000 flow cell spiked with 1%
was performed as described (Silverman et al., 2010). In brief, the mouse first PhiX control library. Sequencing files were de-multiplexed with CASAVA
experienced a 10 min period of habituation during which it was allowed to version 1.8.3 (Illumina).
freely explore a 60 3 40 3 23 cm Plexiglass arena divided into three equally Quality filtering, trimming and de-multiplexing was carried out by a custom
sized, interconnected chambers (left, center, right). Sociability was measured pipeline containing Trim Galore and cutadapt (Martin, 2011)for adaptor and
Figure 4. Selective Treatment with Lactobacillus reuteri Restores Social Deficits and Oxytocin Levels in MHFD Offspring
(A) Schematic of L. reuteri–treatment.
(B and C) Unlike resuspension media (B, p > 0.99, t = 1.03; c, p > 0.99, t = 0.40) or heat-killed L. reuteri (B, p > 0.99, t = 1.35; c, p > 0.99, t = 0.21), administration
of live L. reuteri in the drinking water rescued sociability (B, p < 0.0001, t = 5.98) and preference for social novelty (C, p < 0.001, t = 5.01) in MHFD offspring
(B, treatment effect F1,86 = 87.53, p < 0.0001; C, treatment effect F1,86 = 30.24, p < 0.0001).
(D and E) Representative track plots showing exploratory activity.
(F) Representative images of control oxytocin immunoreactivity at different anteroposterior levels of the PVN.
(G–J) Oxytocin immunoreactivity in the PVN of MRD (G), MHFD (H), heat-killed L. reuteri-treated MHFD (I), and live L. reuteri-treated MHFD offspring (J).
(K–N) Oxytocin immunoreactive cell number (K, p < 0.01, t = 4.76) and oxytocin immunofluorescence intensity (L, p < 0.01, t = 3.80) were reduced in the PVN of
MHFD versus MHFD mice. In the PVN of MRD and MHFD offspring, NeuN cell number immunoreactivity (M, p = 0.34, t = 1.09) and immunofluorescence intensity
(N, p = 0.79, t = 0.28) were similar.
(O and P) Relative to treatment with heat-killed L. reuteri, treatment with live L. reuteri significantly increased oxytocin-positive cell number (O, p < 0.05, t = 2.93)
and oxytocin immunofluorescence intensity (P, p < 0.05, t = 3.09) in the PVN of MHFD offspring. AU: arbitrary units. Plots show mean ± SEM. See also Figures S5
and S6.
1770 Cell 165, 1762–1775, June 16, 2016

A D F
E G
J K L
Figure 5. Reciprocal Social Interaction and Social Interaction-Induced LTP in MHFD Offspring VTA DA Neurons Are Restored by L. reuteri
(A) Schematic of the experimental design.
(B–E) LTP was measure 24 hr following reciprocal interaction with either a stranger or a familiar mouse. Only interaction with a stranger mouse induced LTP in
MRD VTA DA neurons, as determined by increased AMPAR/NMDAR ratios (B and D, Baseline versus Familiar p > 0.99, t = 0.12, baseline versus stranger p < 0.01,
t = 3.79; familiar versus stranger p < 0.05, t = 3.03; F = 8.03, p < 0.01). In MHFD offspring, neither stranger nor familiar reciprocal interaction evoked LTP in VTA DA
neurons (C and E, baseline versus familiar p > 0.99, t = 0.035, baseline versus stranger p = 0.64, t = 1.30; familiar versus stranger p = 0.50, t = 1.45; F2,15 = 1.47,
p = 0.26).
(F and G) Whereas MRD mice spent more time interacting with a stranger than a familiar mouse (familiar versus stranger; F, p < 0.001, t = 4.88; G, p < 0.0001, t =
5.87), MHFD mice did not (familiar versus stranger; F, p = 0.47, t = 1.87; G, p = 0.40, t = 1.96) (F, F3,19 = 13.8, p < 0.0001; G, F3,19 = 18.54, p < 0.0001).
(H–K) Live (H and J, p < 0.01, t = 4.95), but not heat-killed L. reuteri (I and K, p = 0.84, t = 0.20), restored stranger interaction-evoked LTP in the VTA of MHFD
offspring.
(L) Unlike heat-killed L. reuteri (MHFD versus MHFD+Hk-Lr p > 0.99, t = 0.099), live L. reuteri restored reciprocal social interaction (MHFD versus MHFD+Lr
p < 0.05, t = 3.24; F2,9 = 6.45, p < 0.05). Plots show mean ± SEM. See also Figure S7.
quality trimming, and PRINSEQ (Schmieder and Edwards, 2011) for low- Culture and Treatment with L. reuteri and L. johnsonii
complexity filtering and sequence deduplication. In addition, Bowtie2 v2.2.1 Lactobacillus reuteri MM4-1A (ATCC-PTA-6475) and Lactobacillus johnsonii
was used to map reads to MetaPhlAn markers for the classification of bacterial (ATCC 33200) were cultured anaerobically in MRS broth in a 90% N2, 5%
species (Segata et al., 2012). CO2, 5% H2 environment. L. reuteri was heat-killed by keeping the bacteria
at 80 C for 20 min. Bacterial viability was assessed by plating and the efficacy
Colonization of Germ-Free Mice by Fecal Microbiota Transplant of the heat-kill procedure was confirmed by the absence of colony growth
Fresh fecal samples were collected from donor mice/microbiome cohort and following plating. Cultures were centrifuged, washed, and resuspended in
homogenized on ice in sterile PBS under sterile conditions. The resulting slurry anaerobic solution (PBS) and frozen at 80 C until use. PBS, live or heat-killed
was spun at 1,000g for 3 min at 4 C. The supernatants were isolated and diluted L. reuteri were added to the drinking water, which was changed daily to mini-
to 5 3 109 CFU/ml with sterile PBS. Four- or eight-week-old C57Bl6/J germ- mize dosage variability. Whereas the experimental group received live bacte-
free (GF) recipient mice were then immediately colonized by a single gavage ria, one control group received identically prepared cultures of heat-killed
with 0.2 mL solution. Fecal samples were collected from the colonized GF bacteria. A second group of control mice received water treated with PBS
mice at 24 hr, 7 days, 14 days, 28 days, and 56 days following colonization. alone. Live and heat-killed L. reuteri were supplied at a dosage of 1 3 108
Fecal samples were snap frozen and stored at 80 C until prepared for organisms/mouse/day continuously in drinking water. Mice consumed the
sequencing. Behavioral experiments were initiated at 3 weeks post-transplant. treated water ad libitum over the treatment period. The treated drinking water
Cell 165, 1762–1775, June 16, 2016 1771

AMPAR/NMDAR ratio
A B 1.00 **
*
MHFD + OXT MHFD + Vehicle MHFD + OXT 0.75 MHFD + OXT Alone (n=4)
Baseline Stranger Stranger MHFD + Vehicle Stranger (n=6)
0.50 MHFD + OXT Stranger (n=8)
0.25
0.00
Oxytocin + – +
Social interaction – + +
C D G H
125 Number of contacts 100 * 150 150


*
100 * * ***
** 75
100 100
75
50
50
50 50
25 25
0 0 0 0
Oxytocin – – + – – +
M use 1
M us 2
se 1
2
s y
M m 1
se y
1
ou pt
ou pt
o e
ou e
E e
M m
M us
E
o
M
Interaction frequency (s -1)
E F * I J
Time in Chamber (s)
Time in Chamber (s)

1.5 0.15 400 400

**
**
300 300
1.0 0.10
200 200
0.5 0.05
100 100
0.0 0.00 0 0
– – + – – +
s y
M m 1
se y
1
M use 1
M us 2
se 1
2
Oxytocin
ou pt
ou pt
E e
o e
ou e
M m
M us
E
o
M
MHFD Stranger (n=4 pairs) MHFD + Vehicle (n=4)
MHFD + Vehicle Stranger (n=3 pairs) MHFD + Oxytocin (n=4)
MHFD + OXT Stranger (n=4 pairs)
Figure 6. Oxytocin Restores Social Interaction-Induced VTA Plasticity and Social Behavioral Deficits in MHFD Offspring
(A and B) LTP was measure 1–3 hr following a reciprocal interaction. Intranasal oxytocin administration rescued LTP in the VTA of MHFD offspring (B, MHFD+OXT
alone versus MHFD+OXT stranger p < 0.01, t = 3.66; MHFD+Vehicle stranger versus MHFD+OXT stranger p < 0.05, t = 2.86; F2,15 = 7.97, p < 0.01).
(C–F) Oxytocin restored reciprocal social interaction (C, MHFD versus MHFD+Vehicle p = 0.55, t = 1.46; MHFD versus MHFD+OXT p < 0.05, t = 3.62;
MHFD+Vehicle versus MHFD+OXT p < 0.01, t = 4.81; F2,8 = 12.82, p < 0.01; D, MHFD versus MHFD+Vehicle p > 0.99, t = 0.16; MHFD versus MHFD+OXT p < 0.05,
t = 4.075; MHFD+Vehicle versus MHFD+OXT p < 0.05, t = 3.94; F2,8 = 10.97, p < 0.01; E, MHFD versus MHFD+Vehicle, p > 0.99, t = 0.11; MHFD versus
MHFD+OXT p = 0.052, t = 2.99; treatment effect F2,8 = 5.87, p < 0.05; F, MHFD versus MHFD+Vehicle p = 0.2, t = 2.11; MHFD versus MHFD+OXT p < 0.05, t =
3.43; treatment effect F2,8 = 14.58, p < 0.01), sociability (G, MHFD+Vehicle p > 0.99, t = 0.44; MHFD+Oxytocin p = 0.24, t = 1.74; treatment effect F1,8 = 2.37, p =
0.16; H, MHFD+Vehicle p > 0.99, t = 0.29; MHFD+Oxytocin p < 0.01, t = 3.50; treatment effect F1,12 = 5.16, p < 0.05) and preference for social novelty in MHFD
offspring (I, MHFD+Vehicle p = 0.65, t = 1.05; MHFD+Oxytocin p < 0.05, t = 3.54; treatment effect F1,8 = 10.54, p < 0.05; J, MHFD+Vehicle p = 0.50, t = 1.25;
MHFD+Oxytocin p = 0.096, t = 2.34; treatment effect F(1,8) = 6.41, p < 0.05). Plots show mean ± SEM.
for each group was replaced daily 2 hr prior to the onset of the dark cycle to ized using secondary goat anti-rabbit Alexa Fluor 488 (ThermoFisher Scienti-
minimize variation in microbial exposure. Behavioral assays were initiated after fic, #A-11034) and goat anti-mouse Alexa Fluor 594 (ThermoFisher Scientific,
4 weeks of L. reuteri or control treatment. The protocol of the L. johnsonii #A-11032) antibodies (1:1,000 dilution). Slices were incubated in secondary
preparation and administration matched the L. reuteri protocol. Fecal samples antibodies rocking in the dark for 1h at RT. Five minute final washes with
for sequencing and tissue used in the immunofluorescence studies were each of PBTgs, 0.1M PB, and 0.05M PB preceded mounting onto 2% gelatin
collected at the end of the treatment. (Sigma-Aldrich, #G9391)-coated coverslips. Nuclei were visualized using
Vectashield H-1200 with DAPI (Vector Labs, #H-1200).
Immunofluorescence Fluorescent imaging and data acquisition was performed on a Zeiss AxioI-
Mice were deeply anesthetized by inhalation of isoflurane and perfused trans- mager.Z2 microscope (Car Zeiss MicroImaging) mounted with an AxioCam
cardially with 10 mL 0.9% phosphate-buffered saline followed by 30 mL 4% digital camera (Carl Zeiss MicroImaging). Images were captured using
paraformaldehyde in 0.1M phosphate buffer (PFA). Brains were post-fixed in AxioVision acquisition software (Carl Zeiss Microimaging). All images within
4% PFA at 4 C overnight, then cryoprotected in 30% sucrose 0.1M PB over a given dataset were acquired at identical exposure times, within a given chan-
3 days. Coronal slices (30 mm) thick were obtained from frozen tissue using nel, to allow comparison of signal intensity. In some images, contrast and
a sliding blade microtome then transferred to ice cold PBS. Slices were brightness were linearly adjusted using Photoshop (Adobe). Image processing
blocked with 5% normal goat serum, 0.3% Triton X-100 0.1M PB (PBTgs) was applied uniformly across all images within a given dataset. Fluorescence
for 1 hr rocking at RT and then incubated in primary antibodies (rabbit anti-ox- intensity was measured in ImageJ (NIH) by selecting regions of interest
ytocin, ImmunoStar #20068, 1:2,000; mouse anti-NeuN, Millipore, #MAB377, (Oxytocin- and NeuN-positive hypothalamic cell bodies). Hypothalamic
1:2,000) diluted in PBTgs rocking at 4 C for 24 hr. Slices were then washed oxytocin-expressing neuronal population and NeuN+ cell number was as-
three times with 0.3% Triton X-100 0.1M PB. Primary antibodies were visual- sessed in ImageJ using the following operational sequence: (1) open image
1772 Cell 165, 1762–1775, June 16, 2016

file, (2) subtract background, (3) adjust threshold, (4) convert to mask, (5) 0.99.292 (http://www.R-project.org/ [2014]), utilizing the phyloseq package
watershed, (6) analyze particles. Automatic identification of cell boundaries (McMurdie and Holmes, 2013) to import sample data and calculate alpha-
was validated against the source image by an experimenter blind to group and beta-diversity metrics. 16S rRNA gene sequencing data was analyzed
allocation. using Silva 115 and 123 (Quast et al., 2013). Analyses were performed on data-
sets that were rarefied 1,0003, then averaged and rounded. Amplicon
Oxytocin Administration sequences were deposited to the NCBI Sequence Read Archive under acces-
Oxytocin was obtained from Tocris Bioscience (product 1910) and solubilized sion number TBD. Dirichlet Multinomial Mixture modeling was performed in
in 10% dimethyl sulfoxide (DMSO) in PBS. 10% DMSO in PBS was used as mother (Schloss et al., 2009). Significance of categorical variables were deter-
the vehicle control. Mice received oxytocin intranasally (at approximately mined using the non-parametric Mann-Whitney test for two category compar-
200mg/kg) 30 min prior to behavior. 1.25 mL of oxytocin or vehicle solution isons or the Kruskal-Wallis test when comparing three or more categories.
was injected into each nostril from P10 pipette. Oxytocin dose was selected One-way ANOVA followed by Dunnett’s multiple comparisons test was
according to dosages reported to rescue social behavior in genetic models performed using GraphPad’s Prism 6. Correlation between two continuous
phenotypically expressing ASD-like behaviors (Peñagarikano et al., 2015). variables was determined by linear regressions, where p values indicate the
probability that the slope of the regression line is zero. Principal coordinate
Electrophysiology plots employed the Monte Carlo permutation test to estimate p values. All
Recordings were performed as recently described (Huang et al., 2016) and the p values were adjusted for multiple comparisons with the FDR algorithm (Ben-
investigators were kept blind to treatment conditions. Briefly, mice were anes- jamini et al., 2001). No animals or data points were excluded from analyses.
thetized with a mixture of ketamine (100 mg/kg), xylazine (10 mg/kg), and ace-
promazine (3 mg/kg). Horizontal slices (225–300 mm thick) containing the VTA SUPPLEMENTAL INFORMATION
were cut from the brains of C57BL/6J mice. Mice were transcardially perfused
with an ice-cold, oxygenated solution containing (in mM) NaCl, 87; NaHCO3, Supplemental Information includes Supplemental Experimental Procedures
25; KCl, 2.5; NaH2PO4, 1.25; MgCl2, 7; CaCl2, 0.5; dextrose, 25; sucrose, and seven figures and can be found with this article online at http://dx.doi.
75. Horizontal slices were cut with a vibrating tissue slicer (VF-100 Compres- org/10.1016/j.cell.2016.06.001.
stome, Precisionary Instruments, San Jose, CA, or Leica VT 1000S, Leica A video abstract is available at http://dx.doi.org/10.1016/j.cell.2016.06.
Microsystems, Buffalo Grove, IL), incubated at 34 C for 40 min, kept at 001#mmc3.
room temperature for at least 30 min before their transfer to a recording cham-
ber continuously perfused with artificial cerebrospinal fluid (ACSF) at 32 C
AUTHOR CONTRIBUTIONS
and a flow rate of 2–3 mL/min. The recording ACSF contained in mM: 120
NaCl, 3.3 KCl, 1.25 NaH2PO4, 25 NaHCO3, 10 Dextrose, 1 MgCl2 and
Conceptualization and Methodology, S.A.B. and M.C.-M.; Investigation,
2 CaCl2. Recording pipettes were made from thin-walled borosilicate glass
S.A.B. and G.V.D.P.; Analysis, S.A.B., G.V.D.P., T.A.A., N.J.A., J.F.P., and
(TW150F-4, WPI, Sarasota, FL). After filling with intracellular solution
M.C.-M.; Writing – Original Draft, S.A.B. and M.C.-M.; Writing – Review & Edit-
(in mM): 117 CsMeSO3; 0.4 EGTA; 20 HEPES; 2.8 NaCl, 2.5 ATP-Mg 2.0;
ing, S.A.B., G.V.D.P., T.A.A., N.J.A., J.F.P, and M.C.-M.; Visualization, S.A.B.;
0.25 GTP-Na; 5 TEA-Cl, adjusted to pH 7.3 with CsOH and 290 mosmol/L,
Funding Acquisition, M.C.-M.; Supervision, M.C.-M.
they had a resistance of 3–5 MU. Data were obtained with a MultiClamp
700B amplifier, digitized at 20 kHz with a Digidata 1440A, recorded by
Clampex 10 and analyzed with Clampfit 10 software (Molecular Devices). ACKNOWLEDGMENTS
Recordings were filtered online at 3 kHz with a Bessel low-pass filter.
A 2 mV hyperpolarizing pulse was applied before each EPSC to evaluate the We thank members of the Costa-Mattioli laboratory, K. Krnjevic , M. Beau-
input and access resistance (Ra). Data were discarded when Ra was either un- champ, and H. Dierick for comments on the manuscript; A. Swennes, J. Gar-
stable or greater than 25MU, holding current was >200 pA, input resistance cia, D. Quach and J. Auchtung for technical advice. This work was supported
dropped >20% during the recording, or EPSCs baseline changed by >10%. by funding from the NIH (NIMH 096816, NINDS 076708) to M.C.-M and the Al-
After establishing a gigaohm seal (>2GU) and recording stable spontaneous kek Foundation and Baylor College of Medicine to the Alkek Center for Meta-
firing in cell-attached, voltage clamp mode ( 70 mV holding potential), cell genomics and Microbiome Research to J.F.P.
phenotype was determined by measuring the width of the action potential
and the presence of an Ih current. AMPAR/NMDAR ratios were calculated Received: November 18, 2015
as previously described (Huang et al., 2016). Briefly, neurons were voltage- Revised: February 26, 2016
clamped at +40 mV until the holding current stabilized (at <200 pA). Monosyn- Accepted: May 31, 2016
aptic EPSCs were evoked at 0.05 Hz with a bipolar stimulating electrode Published: June 16, 2016
placed 50–150 mm rostral to the lateral VTA. Picrotoxin (100 mM) was added
to the recording ACSF to block GABARA-mediated IPSCs. After recording REFERENCES
the dual-component EPSC, DL-AP5 (100 mM) was bath-applied for 10 min to
remove the NMDAR component, which was then obtained by offline subtrac- Aye, I.L., Rosario, F.J., Powell, T.L., and Jansson, T. (2015). Adiponectin sup-
tion of the remaining AMPAR component from the original EPSC. The peak plementation in pregnant mice prevents the adverse effects of maternal
amplitudes of the isolated components were used to calculate the AMPAR/ obesity on placental function and fetal growth. Proc. Natl. Acad. Sci. USA
NMDAR ratios. Picrotoxin and DL-AP5 were purchased from Tocris Biosci- 112, 12858–12863.
ence and all other reagents were obtained from Sigma-Aldrich. Benjamini, Y., Drai, D., Elmer, G., Kafkafi, N., and Golani, I. (2001). Controlling
the false discovery rate in behavior genetics research. Behav. Brain Res. 125,
Statistical Analysis 279–284.
Data are presented as mean ± SEM. For behavioral experiments, statistics
Bilder, D.A., Bakian, A.V., Viskochil, J., Clark, E.A., Botts, E.L., Smith, K.R., Pi-
were based on the two-sided unpaired Student’s t tests, one- or two-way
mentel, R., McMahon, W.M., and Coon, H. (2013). Maternal prenatal weight
ANOVA with Bonferroni post hoc analysis to correct for multiple comparisons,
gain and autism spectrum disorders. Pediatrics 132, e1276–e1283.
unless otherwise indicated. P, t, and F values are presented in the figure leg-
ends, n values are provided in the figures. p < 0.05 was considered significant. Bolton, J.L., and Bilbo, S.D. (2014). Developmental programming of brain and
*p < 0.05, **p < 0.01, ***p < 0.001, ****p < 0.0001. GraphPad’s Prism 6 (La Jolla, behavior by perinatal diet: focus on inflammatory mechanisms. Dialogues Clin.
CA) software was used to perform statistical analyses and for generating Neurosci. 16, 307–320.
graphical representations of data. For 16S rRNA gene sequencing, analysis Bravo, J.A., Forsythe, P., Chew, M.V., Escaravage, E., Savignac, H.M., Dinan,
and visualization of microbiome communities was conducted in RStudio T.G., Bienenstock, J., and Cryan, J.F. (2011). Ingestion of Lactobacillus strain
Cell 165, 1762–1775, June 16, 2016 1773

regulates emotional behavior and central GABA receptor expression in a relationship to Vineland Adaptive Behavior Scales and cognition. Mol. Psychi-
mouse via the vagus nerve. Proc. Natl. Acad. Sci. USA 108, 16050–16055. atry 13, 980–988.
Bresnahan, M., Hornig, M., Schultz, A.F., Gunnes, N., Hirtz, D., Lie, K.K., Mag- Ma, J., Prince, A.L., Bader, D., Hu, M., Ganu, R., Baquero, K., Blundell, P., Alan
nus, P., Reichborn-Kjennerud, T., Roth, C., Schjølberg, S., et al. (2015). Asso- Harris, R., Frias, A.E., Grove, K.L., and Aagaard, K.M. (2014). High-fat maternal
ciation of maternal report of infant and toddler gastrointestinal symptoms with diet during pregnancy persistently alters the offspring microbiome in a primate
autism: evidence from a prospective birth cohort. JAMA Psychiatry 72, model. Nat. Commun. 5, 3889.
466–474.
Martin, M. (2011). Cutadapt removes adaptor sequences from high-
Connolly, N., Anixt, J., Manning, P., Ping-I Lin, D., Marsolo, K.A., and Bowers, throughput sequencing reads. EMBnetjournal 17, 10–12.
K. (2016). Maternal metabolic risk factors for autism spectrum disorder-An
Mathers, J.C., and McKay, J.A. (2009). Epigenetics - potential contribution to
analysis of electronic medical records and linked birth data. Autism Res. Pub-
fetal programming. Adv. Exp. Med. Biol. 646, 119–123.
lished online January 29, 2016.
Mayer, E.A., Padua, D., and Tillisch, K. (2014). Altered brain-gut axis in autism:
Cryan, J.F., and Dinan, T.G. (2012). Mind-altering microorganisms: the impact
comorbidity or causative mechanisms? BioEssays 36, 933–939.
of the gut microbiota on brain and behaviour. Nat. Rev. Neurosci. 13, 701–712.
Mayer, E.A., Tillisch, K., and Gupta, A. (2015). Gut/brain axis and the micro-
Davari, S., Talaei, S.A., Alaei, H., and Salami, M. (2013). Probiotics treatment
biota. J. Clin. Invest. 125, 926–938.
improves diabetes-induced impairment of synaptic activity and cognitive func-
tion: behavioral and electrophysiological proofs for microbiome-gut-brain McMurdie, P.J., and Holmes, S. (2013). phyloseq: an R package for reproduc-
axis. Neuroscience 240, 287–296. ible interactive analysis and graphics of microbiome census data. PLoS ONE
8, e61217.
Desbonnet, L., Clarke, G., Shanahan, F., Dinan, T.G., and Cryan, J.F. (2014).
Microbiota is essential for social development in the mouse. Mol. Psychiatry Mefford, H.C., Batshaw, M.L., and Hoffman, E.P. (2012). Genomics, intellec-
19, 146–148. tual disability, and autism. N. Engl. J. Med. 366, 733–743.
Dölen, G., Darvishzadeh, A., Huang, K.W., and Malenka, R.C. (2013). Social Melis, M.R., Melis, T., Cocco, C., Succu, S., Sanna, F., Pillolla, G., Boi, A., Ferri,
reward requires coordinated activity of nucleus accumbens oxytocin and se- G.L., and Argiolas, A. (2007). Oxytocin injected into the ventral tegmental area
rotonin. Nature 501, 179–184. induces penile erection and increases extracellular dopamine in the nucleus
Donaldson, Z.R., and Young, L.J. (2008). Oxytocin, vasopressin, and the neu- accumbens and paraventricular nucleus of the hypothalamus of male rats.
rogenetics of sociality. Science 322, 900–904. Eur. J. Neurosci. 26, 1026–1035.
Galley, J.D., Bailey, M., Kamp Dush, C., Schoppe-Sullivan, S., and Christian, Modi, M.E., and Young, L.J. (2012). The oxytocin system in drug discovery for
L.M. (2014). Maternal obesity is associated with alterations in the gut micro- autism: animal models and novel therapeutic strategies. Horm. Behav. 61,
biome in toddlers. PLoS ONE 9, e113026. 340–350.
Groppe, S.E., Gossen, A., Rademacher, L., Hahn, A., Westphal, L., Gründer, Moss, B.G., and Chugani, D.C. (2014). Increased risk of very low birth weight,
G., and Spreckelmeyer, K.N. (2013). Oxytocin influences processing of socially rapid postnatal growth, and autism in underweight and obese mothers. Am. J.
relevant cues in the ventral tegmental area of the human brain. Biol. Psychiatry Health Promot. 28, 181–188.
74, 172–179. Parracho, H.M., Bingham, M.O., Gibson, G.R., and McCartney, A.L. (2005).
Gunaydin, L.A., Grosenick, L., Finkelstein, J.C., Kauvar, I.V., Fenno, L.E., Ad- Differences between the gut microflora of children with autistic spectrum
hikari, A., Lammel, S., Mirzabekov, J.J., Airan, R.D., Zalocusky, K.A., et al. disorders and that of healthy children. J. Med. Microbiol. 54, 987–991.
(2014). Natural neural projection dynamics underlying social behavior. Cell Pedersen, C.A., Caldwell, J.D., Walker, C., Ayers, G., and Mason, G.A. (1994).
157, 1535–1551. Oxytocin activates the postpartum onset of rat maternal behavior in the ventral
Hallmayer, J., Cleveland, S., Torres, A., Phillips, J., Cohen, B., Torigoe, T., tegmental and medial preoptic areas. Behav. Neurosci. 108, 1163–1171.
Miller, J., Fedele, A., Collins, J., Smith, K., et al. (2011). Genetic heritability Peñagarikano, O., Lázaro, M.T., Lu, X.H., Gordon, A., Dong, H., Lam, H.A.,
and shared environmental factors among twin pairs with autism. Arch. Gen. Peles, E., Maidment, N.T., Murphy, N.P., Yang, X.W., et al. (2015). Exogenous
Psychiatry 68, 1095–1102. and evoked oxytocin restores social behavior in the Cntnap2 mouse model of
Hsiao, E.Y., McBride, S.W., Hsien, S., Sharon, G., Hyde, E.R., McCue, T., Co- autism. Sci. Transl. Med. 7, 271ra8.
delli, J.A., Chow, J., Reisman, S.E., Petrosino, J.F., et al. (2013). Microbiota Poutahidis, T., Kearney, S.M., Levkovich, T., Qi, P., Varian, B.J., Lakritz, J.R.,
modulate behavioral and physiological abnormalities associated with neuro- Ibrahim, Y.M., Chatzigiagkos, A., Alm, E.J., and Erdman, S.E. (2013). Microbial
developmental disorders. Cell 155, 1451–1463. symbionts accelerate wound healing via the neuropeptide hormone oxytocin.
Huang, Y.C., and Hessler, N.A. (2008). Social modulation during songbird PLoS ONE 8, e78898.
courtship potentiates midbrain dopaminergic neurons. PLoS ONE 3, e3281. Quast, C., Pruesse, E., Yilmaz, P., Gerken, J., Schweer, T., Yarza, P., Peplies,
Huang, W., Placzek, A.N., Viana Di Prisco, G., Khatiwada, S., Sidrauski, C., J., and Glöckner, F.O. (2013). The SILVA ribosomal RNA gene database proj-
, K., Walter, P., Dani, J.A., and Costa-Mattioli, M. (2016). Translational
Krnjevic ect: improved data processing and web-based tools. Nucleic Acids Res. 41,
control by eIF2a phosphorylation regulates vulnerability to the synaptic and D590–D596.
behavioral effects of cocaine. eLife 5, 5.
Ridaura, V.K., Faith, J.J., Rey, F.E., Cheng, J., Duncan, A.E., Kau, A.L., Griffin,
Human Microbiome Project Consortium (2012a). A framework for human N.W., Lombard, V., Henrissat, B., Bain, J.R., et al. (2013). Gut microbiota from
microbiome research. Nature 486, 215–221. twins discordant for obesity modulate metabolism in mice. Science 341,
Human Microbiome Project Consortium (2012b). Structure, function and diver- 1241214.
sity of the healthy human microbiome. Nature 486, 207–214. Sabatier, N., Leng, G., and Menzies, J. (2013). Oxytocin, feeding, and satiety.
King, J.C. (2006). Maternal obesity, metabolism, and pregnancy outcomes. Front. Endocrinol. (Lausanne) 4, 35.
Annu. Rev. Nutr. 26, 271–291. Schloss, P.D., Westcott, S.L., Ryabin, T., Hall, J.R., Hartmann, M., Hollister,
Krakowiak, P., Walker, C.K., Bremer, A.A., Baker, A.S., Ozonoff, S., Hansen, E.B., Lesniewski, R.A., Oakley, B.B., Parks, D.H., Robinson, C.J., et al.
R.L., and Hertz-Picciotto, I. (2012). Maternal metabolic conditions and risk (2009). Introducing mothur: open-source, platform-independent, community-
for autism and other neurodevelopmental disorders. Pediatrics 129, e1121– supported software for describing and comparing microbial communities.
e1128. Appl. Environ. Microbiol. 75, 7537–7541.
Lerer, E., Levi, S., Salomon, S., Darvasi, A., Yirmiya, N., and Ebstein, R.P. Schmieder, R., and Edwards, R. (2011). Quality control and preprocessing of
(2008). Association between the oxytocin receptor (OXTR) gene and autism: metagenomic datasets. Bioinformatics 27, 863–864.
1774 Cell 165, 1762–1775, June 16, 2016

Segata, N., Waldron, L., Ballarini, A., Narasimhan, V., Jousson, O., and Hutten- Tremaroli, V., and Bäckhed, F. (2012). Functional interactions between the gut
hower, C. (2012). Metagenomic microbial community profiling using unique microbiota and host metabolism. Nature 489, 242–249.
clade-specific marker genes. Nat. Methods 9, 811–814.
Turnbaugh, P.J., Ley, R.E., Mahowald, M.A., Magrini, V., Mardis, E.R., and
Silverman, J.L., Yang, M., Lord, C., and Crawley, J.N. (2010). Behavioural phe- Gordon, J.I. (2006). An obesity-associated gut microbiome with increased
notyping assays for mouse models of autism. Nat. Rev. Neurosci. 11, 490–502. capacity for energy harvest. Nature 444, 1027–1031.
Skinner, A.C., and Skelton, J.A. (2014). Prevalence and trends in obesity and
Uvnäs-Moberg, K., Handlin, L., and Petersson, M. (2014). Self-soothing be-
severe obesity among children in the United States, 1999-2012. JAMA pediat-
haviors with particular reference to oxytocin release induced by non-noxious
rics 168, 561–566.
sensory stimulation. Front. Psychol. 5, 1529.
Song, S.J., Lauber, C., Costello, E.K., Lozupone, C.A., Humphrey, G., Berg-
Lyons, D., Caporaso, J.G., Knights, D., Clemente, J.C., Nakielny, S., et al. Wang, X., Wang, B.R., Zhang, X.J., Xu, Z., Ding, Y.Q., and Ju, G. (2002). Evi-
(2013). Cohabiting family members share microbiota with one another and dences for vagus nerve in maintenance of immune balance and transmission
with their dogs. eLife 2, e00458. of immune information from gut to brain in STM-infected rats. World J. Gastro-
enterol. 8, 540–545.
Sullivan, E.L., Nousen, E.K., and Chamlou, K.A. (2014). Maternal high fat diet
consumption during the perinatal period programs offspring behavior. Physiol. Wu, S., Jia, M., Ruan, Y., Liu, J., Guo, Y., Shuang, M., Gong, X., Zhang, Y.,
Behav. 123, 236–242. Yang, X., and Zhang, D. (2005). Positive association of the oxytocin receptor
Tang, Y., Chen, Z., Tao, H., Li, C., Zhang, X., Tang, A., and Liu, Y. (2014). gene (OXTR) with autism in the Chinese Han population. Biol. Psychiatry 58,
Oxytocin activation of neurons in ventral tegmental area and interfascicular 74–77.
nucleus of mouse midbrain. Neuropharmacology 77, 277–284. Yatsunenko, T., Rey, F.E., Manary, M.J., Trehan, I., Dominguez-Bello, M.G.,
Thomas, A., Burant, A., Bui, N., Graham, D., Yuva-Paylor, L.A., and Paylor, R. Contreras, M., Magris, M., Hidalgo, G., Baldassano, R.N., Anokhin, A.P.,
(2009). Marble burying reflects a repetitive and perseverative behavior more et al. (2012). Human gut microbiome viewed across age and geography.
than novelty-induced anxiety. Psychopharmacology (Berl.) 204, 361–373. Nature 486, 222–227.
Cell 165, 1762–1775, June 16, 2016 1775

Article
Breaking Cryo-EM Resolution

Barriers to Facilitate Drug Discovery
Alan Merk,1,3 Alberto Bartesaghi,1,3 Soojay Banerjee,1,3 Veronica Falconieri,1 Prashant Rao,1 Mindy I. Davis,2
Rajan Pragani,2 Matthew B. Boxer,2 Lesley A. Earl,1 Jacqueline L.S. Milne,1 and Sriram Subramaniam1,*
1Laboratory of Cell Biology, Center for Cancer Research, National Cancer Institute, NIH, Bethesda, MD 20892, USA
2National Center for Advancing Translational Sciences, 9800 Medical Center Drive, Rockville, MD 20850, USA
3Co-first author
*Correspondence: ss1@nih.gov
SUMMARY cesses have now been extended to a wide spectrum of protein

complexes, including several integral membrane proteins (Bai
Recent advances in single-particle cryoelecton et al., 2015b; Du et al., 2015; Liao et al., 2013; Matthies et al.,
microscopy (cryo-EM) are enabling generation of 2016). Structures determined by cryo-EM can now reach resolu-
numerous near-atomic resolution structures for tions as high as 2.2 Å and 2.3 Å, as exemplified by structures of
well-ordered protein complexes with sizes R the 465 kDa b-galactosidase (Bartesaghi et al., 2015) and the
200 kDa. Whether cryo-EM methods are equally 540 kDa AAA ATPase p97 (Banerjee et al., 2016). However, all
of the near-atomic-resolution structures reported have been of
useful for high-resolution structural analysis of
proteins with sizes in the range of 200 kDa or larger, and
smaller, dynamic protein complexes such as those an informal opinion in the field is that cryo-EM technology is
involved in cellular metabolism remains an important primarily suited for analysis of relatively stable proteins with
question. Here, we present 3.8 Å resolution cryo-EM sizes >150 kDa (Thompson et al., 2016). The smallest protein
structures of the cancer target isocitrate dehydroge- for which a cryo-EM structure has been reported using single
nase (93 kDa) and identify the nature of conforma- particle cryo-EM is that of the 135 kDa ABC exporter TmrAB,
tional changes induced by binding of the allosteric at 10 Å resolution (Kim et al., 2015), and the challenges in
small-molecule inhibitor ML309. We also report 2.8- achieving near-atomic resolution for small proteins, even with
Å- and 1.8-Å-resolution structures of lactate dehy- sizes as large as 300 kDa have been noted (Skiniotis and
drogenase (145 kDa) and glutamate dehydrogenase Southworth, 2016; Cheng, 2015; Belnap, 2015). Crystallographic
(334 kDa), respectively. With these results, two analyses are generally very powerful for proteins in this size
range but usually require that flexible regions are either deleted
perceived barriers in single-particle cryo-EM are
or altered. Given that small, dynamic protein complexes are
overcome: (1) crossing 2 Å resolution and (2) obtain- implicated in numerous cellular processes, there is considerable
ing structures of proteins with sizes < 100 kDa, interest in determining whether cryo-EM methods can be also
demonstrating that cryo-EM can be used to investi- applicable for structural analysis of this class of proteins under
gate a broad spectrum of drug-target interactions near-native conditions and at near-atomic resolution.
and dynamic conformational states. A principal reason why small proteins such as those with sizes
<150 kDa have been considered intractable for analysis by cryo-
INTRODUCTION EM is that the errors in alignment of individual projection images
become progressively higher as the size of the scattering entity
Cryo-electron microscopy (cryo-EM) is now firmly established as gets smaller (Henderson et al., 2011). In principle, with a perfect
a central tool in the arsenal of structural biology. The ability to detector that displays minimal falloff in detective quantum effi-
obtain near-atomic resolution structures using cryo-EM was ciency (DQE) even at Nyquist frequency (Henderson, 1995;
shown initially almost three decades ago in the context of elec- McMullan et al., 2014), it should be possible to achieve accurate
tron crystallographic studies of membrane proteins (Henderson alignment of projection images of smaller proteins, but all
et al., 1990). Continued advances in single-particle cryo-EM over currently available detectors still show a significant drop in
the next two decades enabled high-resolution analysis of non- DQE at higher frequencies. The use of phase plates is an option
crystalline samples with high internal symmetry such as icosahe- that may help partially alleviate the problem of image contrast
dral and helical viruses (Ge and Zhou, 2011; Settembre et al., (Danev and Baumeister, 2016), but these developments are still
2011; Yu et al., 2008; Zhang et al., 2010). Large and relatively sta- at an early stage. One way to compensate for the falloff in DQE at
ble complexes such as ribosomes also proved especially higher spatial frequencies is to collect data at higher magnifica-
amenable to analysis using cryo-EM methods, first at medium tion. This strategy, however, lowers contrast and makes the
resolution (Matadeen et al., 1999; Rawat et al., 2003) and more alignment of individual frames collected in movie mode of data
recently at near-atomic resolution (Amunts et al., 2014; Fischer collection more challenging. Experimental approaches to opti-
et al., 2015; Jomaa et al., 2016; Wong et al., 2014). These suc- mize specimen preparation provide an alternative route to
1698 Cell 165, 1698–1707, June 16, 2016 Published by Elsevier Inc.
Table 1. Summary of Data Collection and Image-Processing Parameters for the Density Maps of LDH, IDH1, and GDH Reported in this
Manuscript
Total number Defocus range Starting number Particles used Frames used B-factor for
of micrographs (microns) of particles in final map in final map sharpening (Å2) Final resolution (Å)
LDH-inh 1,707 0.8–2.2 508,402 50,865 (10%) 1–30 150 2.8
IDH-apo 1,506 0.7–3.0 499,184 49,936 (10%) 1–30 180 3.8
IDH-inh 820 0.7–2.6 232,343 46,483 (20%) 1–30 180 3.8
GDH 232 0.8–2.1 45,388 21,818 (48%) 3–9 90 1.8
improve image contrast: it can be easier to minimize the back- (Dragovich et al., 2014), as well as sulfamoylquinoline benzoic
ground scattering from the ice layer for smaller proteins because acid derivatives, which compete with NAD binding to LDH (Bil-
the lower the molecular weight, the thinner the ice layer that is liard et al., 2013). To evaluate the potential of cryo-EM to deter-
required to surround the protein with an aqueous layer. mine structures of a small protein complex such as LDH and
To further test the limits of what is possible with present-day localize the binding sites of potential small-molecule inhibitors,
cryo-EM technology, we have analyzed structures of two small, we carried out cryo-EM analysis of LDH B in complex with
soluble enzymes implicated in cancer metabolism: the 145 kDa GSK2837808A, a 650 Dalton compound in the quinoline 3-sul-
lactate dehydrogenase (LDH B, a tetramer composed of four fonamide series (Billiard et al., 2013). These compounds potently
identical 36 kDa subunits) and the 93 kDa isocitrate dehy- inhibit activity of LDH A and B, as well as isoforms with mixtures
drogenase (IDH1, a dimer composed of two identical 47 kDa of LDH A and B, making them attractive candidates for use in
subunits). In both cases, we tested whether structures can be treatment of solid tumors that rely principally on aerobic glycol-
obtained at high enough resolution to localize bound small-mole- ysis for survival.
cule ligands and to determine the structures of ligand-bound The cryo-EM structure of the LDH B-GSK2837808A complex
complexes. We also carried out structural analysis of the confor- at an overall resolution of 2.8 Å (Figure S1A; Table 1) enables
mationally dynamic enzyme glutamate dehydrogenase (GDH). delineation of the binding site of the inhibitor (Figures 1A–1C).
Earlier crystallographic analyses of apo-GDH, a 334 kDa hex- The location of the compound at the periphery of the protein is
americ protein composed of six identical 56 kDa subunits, in excellent agreement with the placement of inhibitors in the
have been limited to a resolution of 2.7 Å, possibly because two X-ray structures (4QSM and 4QT0 at resolutions of 3.0 Å
some regions of the protein may display a continuum of confor- and 3.2 Å, respectively) that are available for LDH A in complex
mational states even in the context of a crystal lattice. This is a with quinoline 3-sulfonamides (Figures S1B and S1C). The
feature that may well be shared by a large majority of protein density for the inhibitor is not adequate to model in a specific
complexes involved in signaling and metabolism. While it is ex- conformation, which could be a consequence of high conforma-
pected that image averaging will limit resolutions that can be tional flexibility coupled with its location at the outer edge of the
achieved for the conformationally flexible components, it is inter- protein with a high degree of solvent exposure. Nevertheless,
esting to ascertain whether cryo-EM analyses can be used to most regions of the polypeptide display density that enables
obtain atomic resolution for the ordered regions of these types unambiguous construction of an atomic model for the protein
of protein complexes. (Figures 1D and 1E), including localization of numerous water
molecules in the density map (Figures S1D and 1E). The fact
RESULTS that the overall quality and resolution of the structure is compa-
rable to that obtained from X-ray crystallography validates the
Structural Studies of Lactate Dehydrogenase potential of cryo-EM to obtain near-atomic resolution structures
Cancer cells catalyze the production of lactate from pyruvate, of <150 kDa complexes of this kind, without the need for
the final metabolite generated from glucose during glycolysis. crystallization.
In non-malignant cells, pyruvate enters mitochondria to partici-
pate in the Krebs cycle and in oxidative phosphorylation. Under Structural Studies of Isocitrate Dehydrogenase
anaerobic conditions, such as that found in muscle cells, tetra- The conversion of isocitrate to a-ketoglutarate (aKG) is catalyzed
meric LDH mediates the formation of lactate from pyruvate using by NADP+-bound IDH1. IDH1 proteins with mutations at residue
NADH as a cofactor. However, in cancer cells, irrespective of Arg132 are found in cancer cells such as those in glioblastoma.
oxygen availability, glycolysis followed by production of lactate This residue, most often found mutated to a histidine or a
through LDH is the preferred pathway, enhancing the production cysteine, is critical for the maintenance of the active site (Dang
of metabolic precursors required for biosynthesis of cellular et al., 2009; Yang et al., 2010). The R132H and R132C variants
macromolecules. The recognition in a growing number of can- of IDH1 lose affinity for isocitrate; instead, with the cofactor
cers of the central role of both LDH A and B isoforms (Fiume NADPH, these mutants bind aKG and convert it to R( )-2-hy-
et al., 2014; McCleland et al., 2013; Rodriguez et al., 2003) has droxyglutyrate (2HG), an oncometabolite. ML309 is a potent
driven the identification of small-molecule inhibitors, including and selective inhibitor of IDH1 R132H/C, decreasing production
mercaptocyclohex-2-enone derivatives that bind away from of 2HG in glioblastoma cells (Davis et al., 2014). Although a
the NAD-binding pocket and are not competitive inhibitors number of structures of the R132H mutant exist, structures of
Cell 165, 1698–1707, June 16, 2016 1699

Figure 1. Cryo-EM Analysis of LDH B Bound
to Inhibitor GSK2837808A
(A) Cryo-EM density map of complex showing
density for GSK2837808A (red) and selected resi-
dues in the binding pocket.
(B) Ribbon diagram of refined structure of LDH B
showing the location of the bound inhibitor
(marked by arrows) on the periphery of the
tetramer.
(C) Structure of GSK2837808A.
(D and E) Visualization of side-chains in the cryo-
EM density map in an a-helical segment (D) and
loop region (E) of the polypeptide.
See also Figure S1.
variable, with the expected gradient

from regions that are in the well-packed
interior at higher resolution, to those on
the periphery with greater solvent expo-
sure at lower resolution. Comparison of
the structure of apo-IDH1 with that of
ML309-bound IDH1 enables localization
of ML309 in the structure (Figures 2A
and 2D). The location of the inhibitor in
the cryo-EM structure (Figure 2D) in a sol-
vent-exposed region close to the 2-fold
axis and near the catalytic site is similar
to that seen for many IDH1-inhibitor com-
plexes that have been amenable to crys-
tallography (Deng et al., 2015; Okoye-
Okafor et al., 2015; Wu et al., 2015; Zheng
et al., 2013).
Despite the modest resolution of the
maps, the structures provide new insights
into structural aspects of IDH1 and its
interaction with the inhibitor. Cryo-EM re-
constructions carried out without imposi-
tion of 2-fold symmetry resulted in maps
at a resolution of 4.5 Å, similar to those
obtained with the use of 2-fold symmetry
(Figure S2B), indicating that the structure
of the IDH1 dimer in solution is symmetric
at least at this resolution. This is a
different result than that obtained from
crystal structures of IDH1, which are
mostly asymmetric with regions in the
two protomers exhibiting different confor-
IDH Arg132 mutants in complex with the ML309 inhibitor have not mations (Xu et al., 2004; Yang et al., 2010). In our structures, both
yet been determined. Structural analysis of IDH1 by cryo-EM is polypeptide chains that form the IDH1 dimer appear to be equiv-
technically challenging given its small size, low symmetry (C2), alent, suggesting that the apparent asymmetry is a likely conse-
and potential flexibility. quence of crystal lattice formation and is not an intrinsic property
In Figure 2A, we present a cryo-EM structure for the R132C of the dimer, as confirmed by inspection of lattice packing (Fig-
mutant of IDH1 that shows the characteristic butterfly fold of ure S3A). Further, comparison of the apo- and inhibitor-bound
the dimer at a nominal resolution of 3.8 Å (Figure S2A). The poly- structures shows that ML309 binding essentially drives a wedge
peptide backbone and densities for the larger side chains are between the two protomers, leading to an outward movement of
resolved in many regions (Figures 2B and 2C), enabling place- the peripheral domains (Figure 2E) that is evident even when
ment of the polypeptide chain into the density map using prior comparing reconstructions obtained without imposition of sym-
crystal structures of IDH1 as a guide. The quality of the map is metry (Movie S1). This structural change includes displacement
1700 Cell 165, 1698–1707, June 16, 2016

Figure 2. Cryo-EM Analysis of IDH1 in the Absence and Presence of the Inhibitor ML309
(A) Cryo-EM map of the apo-IDH1 (isocitrate dehydrogenase) dimer, colored by subunit.
(B and C) Selected regions of the IDH1 map demonstrating density for side-chains in an a-helical region (B) and a b sheet region (C).
(D) Cryo-EM map of IDH1 in complex with ML309 showing density (red) for the inhibitor (inset) close to the dimer interface.
(E) Superposition of the apo- (gray) and ML309-bound (yellow/blue) IDH1 structures shows the outward movement of the subunits with ML309 binding (density
shown in red). The black arrows indicate the direction of the changes in tertiary structure while the yellow and blue arrows show the overall movements at the level
of quaternary structure.
See also Figures S2 and S3, and Movie S1.
of a stretch of polypeptides spanning residues 270–281. The crystal contacts can freeze specific protein conformations,
density for bound NADPH and the surrounding residues in the enabling structure determination at resolutions of 2 Å or better,
pocket is adequate for placement of the cofactor in the pocket but in other instances, local flexibility can be high enough to limit
(Figure S3B), but the low 3.8 Å resolution of the maps precludes overall resolution. The cancer target and hexameric enzyme
identification of the detailed interactions within the binding GDH is a good example of this problem. GDH displays closed
pocket. Inspection of the asymmetrically reconstructed density and open conformations involving large movements of the nucle-
map also enables the unambiguous identification of additional otide binding domain (NBD), but even when trapped in closed or
density consistent with the binding of a single molecule of open conformations, the highest resolutions reported so far from
ML309 at the dimer interface (Figure S3C), although its precise crystallographic studies and cryo-EM analyses of the open
orientation cannot be determined. Nevertheless, the location of conformation are 2.7 Å (Smith et al., 2002) and 3.3 Å (Borgnia
the catalytic site between the NADPH binding site and the site et al., 2016), respectively. We therefore used GDH to evaluate
of bound ML309 potentially explains how ML309 binding can whether we could further optimize cryo-EM methods to achieve
cause a profound change in the catalytic activity of the enzyme. resolutions of 2 Å or better for the more ordered regions,
These results show that cryo-EM density maps from small, despite the presence of significant conformational flexibility in
<100 kDa dynamic proteins can not only help localize binding other areas of the protein.
sites for small molecule inhibitors, but also potentially predict Using cryo-EM projection images that were selected for
changes in activity that can complement information obtained reduced beam-induced drift, we determined a cryo-EM structure
from crystallographic studies. for GDH in the open state using <22,000 molecular images (Fig-
ures 3, S4, S5; Table 1; Movie S3). Irrespective of whether the
Cryo-EM Analysis of Glutamate Dehydrogenase maps were divided into one, two, or more classes, the density
Smaller size, lower symmetry, and conformational flexibility maps displayed a clear difference between well-ordered central
may all be contributing factors that limit resolution of the IDH1 regions covering the central half of the complex and ‘‘fuzzy’’
structures obtained using cryo-EM methods. Intrinsic flexibility outer regions for the rest, as illustrated by slices through projec-
of proteins also poses a challenge for determining high-resolution images of the density map (Figures 3A and 3B). Based on
tion structures even with X-ray crystallography. In many cases, comparison of the extent of movement determined by classifying
Cell 165, 1698–1707, June 16, 2016 1701

Figure 3. Cryo-EM Analysis of GDH
(A and B) Projection views of two 3D classes from cryo-EM analysis of GDH (glutamate dehydrogenase). In both classes, there is well-defined density at the core,
but it is weakly defined at the peripheral nucleotide binding domain (NBD). The two classes display similar structures in the interior but differ slightly in the
peripheral NBD. The two classes are likely to be subsets of a continuum of states with varying orientations for the outer domain relative to the core.
(C) Ribbon diagrams of the open and closed structures (Borgnia et al., 2016) demonstrating the more extensive NBD movement associated with substrate binding
and catalytic cleft closure.
(D) A selected region of the cryo-EM map of the GDH structure, highlighting high-resolution features such as ‘‘holes’’ in the aromatic rings of Tyr382, Phe383, and
Trp385, water molecules (shaded yellow), and well-resolved densities for carbonyl bonds.
See also Figures 4 and S4 and Movie S2.
the data to produce two discrete 3D structures, the range of twisting motions of the pivot helix in the absence of either ADP
motion in the peripheral NBD (Figures 3A and 3B; Movie S2) is or GTP in solution. Despite the presence of the flexible outer
seen to be small in comparison to the large domain movements domains (Figures 3A and 3B), most of the interior of GDH can
observed for transition from open to closed conformations (Fig- nevertheless be visualized at resolutions better than 2 Å, well
ure 3C) (Borgnia et al., 2016). beyond the 2.7 Å resolution obtained with X-ray crystallographic
The binding of GTP (an inhibitor) and ADP (an activator) to two analysis, in which the flexible regions presumably limit the
allosteric sites in each protomer modulates the transition be- order of the crystals obtained. In comparison to our earlier
tween ‘‘closed’’ and ‘‘open’’ states of GDH. These allosteric work with b-galactosidase (Bartesaghi et al., 2015), in which
modulators tightly control GDH function in vivo. The two states we reported an average resolution of 2.2 Å, the density map
we present in Figures 3A and 3B are both in the ‘‘open’’ confor- for GDH at 1.8 Å shows better-defined and more uniform density
mation with differences likely due to small perturbations arising over most of the ordered central core, spanning the subunit
from the unhindered sampling of conformations due to the interface (Figure 3D), as well as clear density for carbonyl bonds,
1702 Cell 165, 1698–1707, June 16, 2016

Figure 4. Density Representations for Each
of the 20 Amino Acid Types from the 1.8 Å
Resolution Cryo-EM Structure of apo-GDH
Features such as holes in aromatic rings, as
well as the ‘‘zigzag’’ structure of extended side-
chains such as Arg and Lys, are visible in
the density maps. See also Figure S5 and
Movie S3.
fraction of the protein that is conforma-

tionally mobile may preclude achieving
atomic resolution for the more structured
or stable regions of the complex. The fact
that we can visualize densities for each of
the 20 amino acids in the GDH cryo-EM
map (Figure 4), without 3D masking
during refinement, addresses this ques-
tion definitively while also enabling the
crossing of the 2 Å resolution mark in
single particle cryo-EM. These cryo-EM
densities are easily comparable to those
obtained by X-ray crystallography at a
similar resolution (Figure S5).
The maps we have obtained for LDH
display 2.8 Å resolution in the protein
component and densities for numerous
water molecules (Figures S1D and S1E),
but density for the bound ligand is at
lower resolution. One reason that the
density for the bound ligand is less well-
defined than that of the polypeptide may
be because small molecule ligands
bound to the protein periphery are likely
to sample far more conformations in
aqueous solution than what is possible
in the context of a crystal lattice. Most
solvents used for protein crystallization
invariably include additives such as PEG
that dehydrate the crystal by removing
several H-bonded water molecules and well-defined density for excess solvent and tightening the packing of protein molecules
extended Lys side chains (Figures S4B–S4E). while reducing the size of the solvent channels around the
protein. As a result, these additives improve crystal order and
DISCUSSION diffraction resolution, but at the same time, also affect ligand-
protein interactions, as dehydration reduces the number of water
We have previously shown that it is possible to achieve atomic molecules in the binding pocket and may minimize mobility of the
resolution information from an inhibitor-bound protein complex ligand. Thus, cryo-EM studies in aqueous dispersions and X-ray
such as the 465 kDa b-galactosidase (Banerjee et al., 2016; crystallographic studies in ordered 3D crystals provide poten-
Bartesaghi et al., 2015). The inhibitor-bound state of b-galactosi- tially complementary snapshots of ligand-protein interactions
dase is relatively stable, with only a small gradient in temperature that are likely to occur in cellular environments.
factor between the central and peripheral regions, making it Finding the correct orientation of each projection image is cen-
possible to achieve high resolution across the breadth of the tral for the success of the single particle cryo-EM reconstruction
protein. However, many proteins can have flexible domains, as process. The practical size limitation for high-resolution structure
in the case of GDH, even when they are expected to be in a single determination is currently a question of considerable interest
conformational state. While cryo-EM can be used to effectively within the cryo-EM field. Figure 5 provides a comparison of the
mask out flexible or disordered areas of a protein during steps relative sizes of some complexes whose structures have been
in image processing (Amunts et al., 2014; Bai et al., 2015a), it studied at near-atomic resolution, including those reported in
has remained an open question whether the presence of a large the present study. Given the present state of detector and
Cell 165, 1698–1707, June 16, 2016 1703

Figure 5. Relative Sizes of Small Metabolic
Complexes Compared to Other Structures
Solved by Cryo-EM and Variations in Flexi-
bility across Different Regions
(A) Relative sizes of GDH, LDH B, and IDH1 in
comparison to a representative set of other
structures that have been solved by cryo-EM
to high resolution including icosahedral viruses,
ribosomes, and protein complexes such as
b-galactosidase and the AAA ATPase p97.
(B–D) Ribbon diagram of hexameric GDH (B),
tetrameric LDH B (C), and dimeric IDH1 (D),
colored to show B-factor variation from blue
(lowest) to red (highest), respectively.
See also Movie S3.
the entire protein, which poses a chal-

lenge for capturing all of the information
in a single view. Nevertheless, it is clear
that in all three of the protein complexes
that we have analyzed, the outer regions
are significantly more flexible than the
core of the complex (Figures 5B–5D).
Since cryo-EM analyses report on the
structure of the proteins in solution, they
can provide information on quaternary
structures complementary to that ob-
tained from crystal structures, which
microscope technology, a figure of 150 kDa is thought to be a may be influenced by contacts within the crystal lattice. A com-
practical lower limit even for medium resolutions (Thompson mon feature of the cryo-EM density maps we have observed in
et al., 2016), even though theoretical arguments have been structures of complexes of ligands with diverse proteins such
advanced for why much smaller sizes should be accessible as LDH, IDH1, b-galactosidase, and p97 is that the density for
(Glaeser and Hall, 2011; Henderson, 1995). As instrumentation bound ligands, especially in the peripheral regions, is generally
and methods for data analysis have improved, our ability to somewhat weaker and at lower resolution than that of residues
overcome the handicaps of small size, low symmetry, and high in the binding pocket. As noted earlier in the discussion, this is
flexibility have likewise improved, allowing the reconstruction perhaps not surprising given the presence of numerous single
of not just large, highly symmetric complexes such as icosahe- bonds around which rotations are possible, coupled with poor
dral viruses, but now also of smaller, lower symmetry complexes steric and electrostatic fits between ligand and protein. Even in
(Figure 5A). In this study, we have shown that not only is it regions where the density map of the polypeptide is poorly
possible to determine the structure of a protein smaller than defined in small proteins such as IDH1, crystallographic informa-
150kDa to 3 Å resolution, it is also possible to determine the tion can provide an invaluable framework to interpret the cryo-
structure of a protein <100 kDa to near-atomic (<4 Å) resolution, EM structures, enabling identification of sites of inhibitor binding.
coupled with the visualization of the location of bound small- Size, symmetry, and intrinsic flexibility are all parameters that
molecule inhibitors in both cases. From an analysis of all of the play a role in determining the resolutions that can be achieved us-
parameters that influenced final map quality, our conclusion is ing cryo-EM. LDH and b-galactosidase are both tetramers with the
that selecting for specimens with thin ice, retaining only those same symmetry (D2), but LDH is about a third of the size of
images that display the lowest beam-induced motion, and using b-galactosidase. The lower resolutions currently achieved with
a sufficiently small pixel size and the improved XP sensor with LDH are most likely from the reduced accuracy of alignment of
higher DQE performance were the key factors that enabled us projection images, and this comparison is thus a good way to
to obtain high resolution with these smaller protein complexes. assess effect of size on resolution with presently available technol-
As more examples of cryo-EM structures are reported, there ogy. The further drop in resolution observed for IDH1 could be due
are beginning to be noticeable patterns in the differences be- to reduction in size, lower symmetry, and potentially greater flexi-
tween the visual appearance of cryo-EM density maps and those bility of the dimeric enzyme. Reconstruction of IDH1 without appli-
obtained using X-ray crystallography. In visualizing cryo-EM cation of symmetry resulted in only a slightly lower resolution
density maps, different regions of the map are typically explored (4.5 Å), indicating that the trends we observe are likely general
by implementing B-factor sharpening specific to the different in nature. It is possible that, for these small enzymes, improve-
segments of the map, but deposited maps and figures generally ments in resolution could be obtained by the use of one or two
show density maps sharpened with a single value for B-factor for bound Fab fragments (Wu et al., 2012). Fab fragments not only
1704 Cell 165, 1698–1707, June 16, 2016

increase the size of a complex, but also assist with orientation For the LDH B and IDH1 reconstructions, particles were picked from the
assignment and can stabilize certain conformations of highly flex- original micrographs using a Gaussian disk of 42 Å in radius as the search
template. Particles were extracted using a binning factor of 8 and a box size
ible complexes (Kim et al., 2015; Lyumkis et al., 2013b). Since Fab
of 96 pixels and subsequently subjected to 12 rounds of refinement with the
fragments are 50 kDa in size, an important consequence of our program Frealign (Lyumkis et al., 2013a). The numbers of particle images
results is that the presence of one or two bound Fab fragments used for each final reconstruction are shown in Table 1. The subset of particles
may bring almost any protein into a size range suitable for analysis that contributed to the reconstruction in the last iteration was re-extracted
by cryo-EM. Although we have not yet achieved 2 Å resolution from the original micrographs using a binning factor of 4 and a box size of
with small complexes such as IDH and LDH, it is likely that further 192 pixels followed by eight additional iterations of local refinement in Frealign.
In a similar way, particles were re-extracted and further refined two additional
advances in detector technology will contribute toward achieving
times using a binning factor of 2 and a box size of 768 pixels, first using the full
this goal. Ongoing hardware developments such as phase plates
exposure and then using only the first half of the exposure (frames 1–30),
(Danev and Baumeister, 2016), as well as improvements in soft- corresponding to 30 e /Å2. B-factor sharpening was applied to the final
ware for image alignment and reconstruction of heterogeneous maps for purposes of visualization (Table 1).
complexes, will also undoubtedly be necessary to advance Images of GDH were processed in a way similar to that done for LDH B and
cryo-EM frontiers to determine high resolution structures of even IDH1, this time using a Gaussian disk of 75 Å in radius as the search template
smaller and more challenging complexes. for particle picking. The selection of particles that contributed to the recon-
struction at each iteration was done according to the bi-modal distribution
of Frealign scores (Banerjee et al., 2016). Re-extraction of selected particles
EXPERIMENTAL PROCEDURES from the original micrographs using progressively less binning followed by
local refinement was done as described above for the LDH B and IDH1 data-
Proteins and Grid Preparation sets. In addition, particles with a binning factor of 2 and a box size of 768 pixels
Chicken heart LDH B (catalog #59747, Sigma-Aldrich, St. Louis) lyophilized were re-extracted one final time using frames 3–9 and used to produce the final
sample was suspended in 550 ml 13 PBS buffer and centrifuged at reconstruction by applying the latest set of alignments (derived using the first
20,000 3 g for 30 min. Thereafter, it was subjected to gel filtration on a Super- half of the exposure) without further refinement. A B-factor of 90 Å2 was
dex-200 size-exclusion chromatography column connected to a ÄKTA FPLC applied to the final map for purposes of visualization.
apparatus (GE Healthcare Bio-Sciences, Piscataway) with an elution buffer To estimate resolution, we used two approaches, both of which provided
comprised of 13 PBS. The ligand GSK 2837808A (catalog #5189 Tocris mutually consistent results. In one approach, we used the strategy imple-
Bioscience) was dissolved in 100% DMSO and added to LDH (1.5 mg/ml) mented in Frealign to carry out refinement using only low-resolution informa-
to a final inhibitor concentration of 75 mM and 0.5% DMSO. tion (to prevent over-fitting) and then measured the value at which the FSC
Human IDH1 R132C mutant protein was purified following the published value drops to 0.143. The highest resolution we used for refining particle orien-
protocol (Davis et al., 2014). ML309 was dissolved in 100% DMSO and added tations was 6 Å for the IDH1 datasets and 4 Å for the LDH B and GDH datasets.
to IDH1 (2.8 mg/ml) to a final inhibitor concentration of 50 mM and 0.5% DMSO. We also computed the FSC curve between the experimentally derived map
Bovine GDH (catalog #G2626, Sigma-Aldrich, St. Louis) was dialyzed and the refined model, and determined the value at which the correlation drops
overnight against gel-filtration buffer (100 mM potassium phosphate [pH to 0.5.
6.8]) prior to purification by size-exclusion chromatography (SEC) using a
Superdex-200 size-exclusion chromatography column connected to a ÄKTA
FPLC apparatus (GE Healthcare Bio-Sciences, Piscataway). Final protein Atomic Model Refinement
concentration was 3 mg/ml. The starting model used for the refinement of the LDH coordinates was the X-
Prior to preparation of frozen-hydrated specimens, 3–4 ml of a freshly ray structure 1I0Z. Changes were made to the initial model in order to incorpo-
thawed aliquot of purified proteins or inhibitor-protein complexes were depos- rate the differences between the human and chicken LDH amino acid se-
ited on Quantifoil R1.2/1/3 grids and plunge frozen in liquid ethane cooled by quences. The structure of the apo LDH B was initially refined by rigid-body
liquid nitrogen using either a Leica plunge-freeze device (for the LDH and IDH1 refinement followed by real-space refinement using the program Phenix
samples) or a Vitrobot (for the GDH samples). For experiments with added (Adams et al., 2010). Manual adjustments were made using the program
inhibitor, the protein was pre-incubated for 30 min at room temperature under COOT (Emsley et al., 2010). Solvent molecules were added manually followed
the buffer conditions indicated above for each kind of protein complex. by a round of real-space refinement.
The starting model used for the refinement of NADPH-bound IDH1 was
from the coordinates from the structure 3MAP. Prior to refinement, the
Data Acquisition
residue Arg132 was mutated to a cysteine and the isocitrate molecule was
The grids were imaged using a Titan Krios transmission electron microscope
removed from the model. The structure of the NADPH-bound IDH1 was initially
(FEI Company, Hillsboro) aligned for parallel illumination and operated at
refined by rigid-body refinement followed by real-space refinement using the
300 kV, with the specimen maintained at liquid nitrogen temperatures. Images
program Phenix. Manual adjustments were made by refinement in COOT
were recorded on a K2 Summit camera equipped with the XP sensor (Gatan,
following another round of real-space refinement. For the ML-309-bound
Pleasanton) operated in super-resolution counting mode, placed at the end of
structure, the starting model was the NADPH-bound IDH1 obtained using
a GIF Quantum energy filter (Gatan), functioning in zero-energy-loss mode with
the cryo-EM data. The refinement protocol was identical to that used for
a slit width of 20 eV as described previously (Bartesaghi et al., 2015). Images
LDH B.
were typically collected with a defocus range between 0.7 and 3.0 mm. For
The PDB coordinates 3JCZ served as a starting point for the refinement of
LDH and IDH, a dose rate of 5 e /Å2/s (at the specimen plane) was used with
apo-GDH. Initially, rigid-body fitting followed by real-space refinement using
a total exposure time of 12 s. Intermediate frames were recorded every 0.2 s
the program Phenix was used to refine the model. Manual adjustments were
giving an accumulated dose of 60 e /Å2 and a total of 60 frames per image.
made in COOT. Solvent molecules were added automatically using stringent
Pixel size was set to 0.495 Å. For GDH, a dose rate of 2.6 e /Å2/s was used
thresholds followed by manual inspection. Thereafter, the model was refined
and a total exposure time of 15.2 s. Intermediate frames were recorded every
again using real-space refinement as implemented in Phenix.
0.4 s giving an accumulated dose of 40 e /Å2 and a total of 38 frames per im-
age with a pixel size of 0.637 Å.
ACCESSION NUMBERS
Image Processing
Movie frame alignment and contrast transfer function (CTF) estimation for each The accession numbers for the density maps reported in this paper are Elec-
micrograph were carried out as described previously (Bartesaghi et al., 2014). tron Microscopy Data Bank: 8191, 8192, 8193, and 8194 for LDHB, apo- and
Cell 165, 1698–1707, June 16, 2016 1705

ML309-bound IDH1, and GDH, respectively. Each entry includes uncorrected Billiard, J., Dennison, J.B., Briand, J., Annan, R.S., Chai, D., Colón, M.,
and B-factor sharpened maps, as well as the asymmetric reconstructions Dodson, C.S., Gilbert, S.A., Greshock, J., Jing, J., et al. (2013). Quinoline 3-sul-
where applicable. The accession numbers for the corresponding refined fonamides inhibit lactate dehydrogenase A and reverse aerobic glycolysis in
atomic models reported in this paper are PDB: 5K0Z, 5K10, 5K11, and 5K12. cancer cells. Cancer Metab. 1, 19.
Borgnia, M.J., Banerjee, S., Merk, A., Matthies, D., Bartesaghi, A., Rao, P.,
SUPPLEMENTAL INFORMATION Pierson, J., Earl, L.A., Falconieri, V., Subramaniam, S., and Milne, J.L.
(2016). Using Cryo-EM to Map Small Ligands on Dynamic Metabolic Enzymes:
Supplemental Information includes five figures and three movies and can be Studies with Glutamate Dehydrogenase. Mol. Pharmacol. 89, 645–651.
found with this article online at http://dx.doi.org/10.1016/j.cell.2016.05.040. Cheng, Y. (2015). Single-Particle Cryo-EM at Crystallographic Resolution. Cell
161, 450–457.
AUTHOR CONTRIBUTIONS
Danev, R., and Baumeister, W. (2016). Cryo-EM single particle analysis with
the Volta phase plate. eLife 5, e13046.
Specimen preparation was carried out by A.M. and P.R., data collection was
carried out by A.M., image processing was carried out by A.B., data analysis, Dang, L., White, D.W., Gross, S., Bennett, B.D., Bittinger, M.A., Driggers, E.M.,
interpretation, and presentation was carried out by S.B., A.B., A.M., V.F., Fantin, V.R., Jang, H.G., Jin, S., Keenan, M.C., et al. (2009). Cancer-associated
L.A.E., J.L.S.M., and S.S., and samples of purified IDH1 and ML309 were pro- IDH1 mutations produce 2-hydroxyglutarate. Nature 462, 739–744.
vided by M.I.D., R.P., and M.B.B. Final figure preparation was carried out pri- Davis, M.I., Gross, S., Shen, M., Straley, K.S., Pragani, R., Lea, W.A., Popovici-
marily by V.F. with help from A.M. and A.B. Project conception, organization, Muller, J., DeLaBarre, B., Artin, E., Thorne, N., et al. (2014). Biochemical,
and supervision was provided by S.S., who was also primarily responsible for cellular, and biophysical characterization of a potent inhibitor of mutant isoci-
writing the manuscript, with help from S.B., A.B., A.M., L.A.E., V.F., and trate dehydrogenase IDH1. J. Biol. Chem. 289, 13717–13725.
J.L.S.M. Deng, G., Shen, J., Yin, M., McManus, J., Mathieu, M., Gee, P., He, T., Shi, C.,
Bedel, O., McLean, L.R., et al. (2015). Selective inhibition of mutant isocitrate
ACKNOWLEDGMENTS dehydrogenase 1 (IDH1) via disruption of a metal binding network by an
allosteric small molecule. J. Biol. Chem. 290, 762–774.
This research was supported by funds from the Center for Cancer Research,
Dragovich, P.S., Fauber, B.P., Boggs, J., Chen, J., Corson, L.B., Ding, C.Z., Ei-
National Cancer Institute, NIH in Bethesda. We thank Kieran Moynihan, Robert
genbrot, C., Ge, H., Giannetti, A.M., Hunsaker, T., et al. (2014). Identification of
Mueller, and Joe Cometa for technical assistance with electron microscopy;
substituted 3-hydroxy-2-mercaptocyclohex-2-enones as potent inhibitors of
Fred Ulmer, Paul Mooney, and Chris Booth for advice and assistance with opti-
human lactate dehydrogenase. Bioorg. Med. Chem. Lett. 24, 3764–3771.
mizing K2 detector performance; and Rishub Jain, Jierui Fang, and Amy Jin for
assistance with data analysis. This work utilized the computational resources Du, J., Lü, W., Wu, S., Cheng, Y., and Gouaux, E. (2015). Glycine receptor
of the NIH HPC Biowulf cluster (http://hpc.nih.gov). mechanism elucidated by electron cryo-microscopy. Nature 526, 224–229.
Emsley, P., Lohkamp, B., Scott, W.G., and Cowtan, K. (2010). Features and
Received: April 29, 2016 development of Coot. Acta Crystallogr. D Biol. Crystallogr. 66, 486–501.
Revised: May 9, 2016 Fischer, N., Neumann, P., Konevega, A.L., Bock, L.V., Ficner, R., Rodnina,
Accepted: May 9, 2016 M.V., and Stark, H. (2015). Structure of the E. coli ribosome-EF-Tu complex
Published: May 26, 2016 at <3 Å resolution by Cs-corrected cryo-EM. Nature 520, 567–570.
Fiume, L., Manerba, M., Vettraino, M., and Di Stefano, G. (2014). Inhibition of
REFERENCES
lactate dehydrogenase activity as an approach to cancer therapy. Future
Med. Chem. 6, 429–445.
Adams, P.D., Afonine, P.V., Bunkóczi, G., Chen, V.B., Davis, I.W., Echols, N.,
Headd, J.J., Hung, L.W., Kapral, G.J., Grosse-Kunstleve, R.W., et al. (2010). Ge, P., and Zhou, Z.H. (2011). Hydrogen-bonding networks and RNA bases re-
PHENIX: a comprehensive Python-based system for macromolecular struc- vealed by cryo electron microscopy suggest a triggering mechanism for cal-
ture solution. Acta Crystallogr. D Biol. Crystallogr. 66, 213–221. cium switches. Proc. Natl. Acad. Sci. USA 108, 9637–9642.
Amunts, A., Brown, A., Bai, X.C., Llácer, J.L., Hussain, T., Emsley, P., Long, F., Glaeser, R.M., and Hall, R.J. (2011). Reaching the information limit in cryo-EM
Murshudov, G., Scheres, S.H., and Ramakrishnan, V. (2014). Structure of the of biological macromolecules: experimental aspects. Biophys. J. 100, 2331–
yeast mitochondrial large ribosomal subunit. Science 343, 1485–1489. 2337.
Bai, X.C., Rajendra, E., Yang, G., Shi, Y., and Scheres, S.H. (2015a). Sampling Henderson, R. (1995). The potential and limitations of neutrons, electrons and
the conformational space of the catalytic subunit of human g-secretase. eLife X-rays for atomic resolution microscopy of unstained biological molecules.
4, e11182. Q. Rev. Biophys. 28, 171–193.
Bai, X.C., Yan, C., Yang, G., Lu, P., Ma, D., Sun, L., Zhou, R., Scheres, S.H., Henderson, R., Baldwin, J.M., Ceska, T.A., Zemlin, F., Beckmann, E., and
and Shi, Y. (2015b). An atomic structure of human g-secretase. Nature 525, Downing, K.H. (1990). Model for the structure of bacteriorhodopsin based
212–217. on high-resolution electron cryo-microscopy. J. Mol. Biol. 213, 899–929.
Banerjee, S., Bartesaghi, A., Merk, A., Rao, P., Bulfer, S.L., Yan, Y., Green, N., Henderson, R., Chen, S., Chen, J.Z., Grigorieff, N., Passmore, L.A., Ciccarelli,
Mroczkowski, B., Neitz, R.J., Wipf, P., et al. (2016). 2.3 Å resolution cryo-EM L., Rubinstein, J.L., Crowther, R.A., Stewart, P.L., and Rosenthal, P.B. (2011).
structure of human p97 and mechanism of allosteric inhibition. Science 351, Tilt-pair analysis of images from a range of different specimens in single-par-
871–875. ticle electron cryomicroscopy. J. Mol. Biol. 413, 1028–1046.
Bartesaghi, A., Matthies, D., Banerjee, S., Merk, A., and Subramaniam, S. Jomaa, A., Boehringer, D., Leibundgut, M., and Ban, N. (2016). Structures of
(2014). Structure of b-galactosidase at 3.2-Å resolution obtained by cryo-elec- the E. coli translating ribosome with SRP and its receptor and with the trans-
tron microscopy. Proc. Natl. Acad. Sci. USA 111, 11709–11714. locon. Nat. Commun. 7, 10471.
Bartesaghi, A., Merk, A., Banerjee, S., Matthies, D., Wu, X., Milne, J.L., and Kim, J., Wu, S., Tomasiak, T.M., Mergel, C., Winter, M.B., Stiller, S.B., Robles-
Subramaniam, S. (2015). 2.2 Å resolution cryo-EM structure of b-galactosi- Colmanares, Y., Stroud, R.M., Tampé, R., Craik, C.S., and Cheng, Y. (2015).
dase in complex with a cell-permeant inhibitor. Science 348, 1147–1151. Subnanometre-resolution electron cryomicroscopy structure of a heterodi-
Belnap, D.M. (2015). Electron Microscopy and Image Processing: Essential meric ABC exporter. Nature 517, 396–400.
Tools for Structural Analysis of Macromolecules. Curr. Protoc. Protein Sci. Liao, M., Cao, E., Julius, D., and Cheng, Y. (2013). Structure of the TRPV1 ion
82, 17.2.1–17.2.61. channel determined by electron cryo-microscopy. Nature 504, 107–112.
1706 Cell 165, 1698–1707, June 16, 2016

Lyumkis, D., Brilot, A.F., Theobald, D.L., and Grigorieff, N. (2013a). Likelihood- Skiniotis, G., and Southworth, D.R. (2016). Single-particle cryo-electron micro-
based classification of cryo-EM images using FREALIGN. J. Struct. Biol. 183, scopy of macromolecular complexes. Microscopy (Oxf.) 65, 9–22.
377–388.
Smith, T.J., Schmidt, T., Fang, J., Wu, J., Siuzdak, G., and Stanley, C.A. (2002).
Lyumkis, D., Julien, J.P., de Val, N., Cupo, A., Potter, C.S., Klasse, P.J., The structure of apo human glutamate dehydrogenase details subunit
Burton, D.R., Sanders, R.W., Moore, J.P., Carragher, B., et al. (2013b). communication and allostery. J. Mol. Biol. 318, 765–777.
Cryo-EM structure of a fully glycosylated soluble cleaved HIV-1 envelope
Thompson, R.F., Walker, M., Siebert, C.A., Muench, S.P., and Ranson, N.A.
trimer. Science 342, 1484–1490.
(2016). An introduction to sample preparation and imaging by cryo-electron
Matadeen, R., Patwardhan, A., Gowen, B., Orlova, E.V., Pape, T., Cuff, M., microscopy for structural biology. Methods 100, 3–15.
Mueller, F., Brimacombe, R., and van Heel, M. (1999). The Escherichia coli
large ribosomal subunit at 7.5 A resolution. Structure 7, 1575–1583. Wong, W., Bai, X.C., Brown, A., Fernandez, I.S., Hanssen, E., Condron, M.,
Tan, Y.H., Baum, J., and Scheres, S.H. (2014). Cryo-EM structure of the Plas-
Matthies, D., Dalmas, O., Borgnia, M.J., Dominik, P.K., Merk, A., Rao, P.,
modium falciparum 80S ribosome bound to the anti-protozoan drug emetine.
Reddy, B.G., Islam, S., Bartesaghi, A., Perozo, E., and Subramaniam, S.
eLife 3, e03080.
(2016). Cryo-EM Structures of the Magnesium Channel CorA Reveal Symme-
try Break upon Gating. Cell 164, 747–756. Wu, S., Avila-Sakar, A., Kim, J., Booth, D.S., Greenberg, C.H., Rossi, A., Liao,
McCleland, M.L., Adler, A.S., Deming, L., Cosino, E., Lee, L., Blackwood, E.M., M., Li, X., Alian, A., Griner, S.L., et al. (2012). Fabs enable single particle
Solon, M., Tao, J., Li, L., Shames, D., et al. (2013). Lactate dehydrogenase B is cryoEM studies of small proteins. Structure 20, 582–592.
required for the growth of KRAS-dependent lung adenocarcinomas. Clin. Wu, F., Jiang, H., Zheng, B., Kogiso, M., Yao, Y., Zhou, C., Li, X.N., and Song,
Cancer Res. 19, 773–784. Y. (2015). Inhibition of Cancer-Associated Mutant Isocitrate Dehydrogenases
McMullan, G., Faruqi, A.R., Clare, D., and Henderson, R. (2014). Comparison by 2-Thiohydantoin Compounds. J. Med. Chem. 58, 6899–6908.
of optimal performance at 300keV of three direct electron detectors for use in Xu, X., Zhao, J., Xu, Z., Peng, B., Huang, Q., Arnold, E., and Ding, J. (2004).
low dose electron microscopy. Ultramicroscopy 147, 156–163. Structures of human cytosolic NADP-dependent isocitrate dehydrogenase
Okoye-Okafor, U.C., Bartholdy, B., Cartier, J., Gao, E.N., Pietrak, B., Rendina, reveal a novel self-regulatory mechanism of activity. J. Biol. Chem. 279,
A.R., Rominger, C., Quinn, C., Smallwood, A., Wiggall, K.J., et al. (2015). New 33946–33957.
IDH1 mutant inhibitors for treatment of acute myeloid leukemia. Nat. Chem.
Yang, B., Zhong, C., Peng, Y., Lai, Z., and Ding, J. (2010). Molecular mecha-
Biol. 11, 878–886.
nisms of ‘‘off-on switch’’ of activities of human IDH1 by tumor-associated
Rawat, U.B., Zavialov, A.V., Sengupta, J., Valle, M., Grassucci, R.A., Linde, J., mutation R132H. Cell Res. 20, 1188–1200.
Vestergaard, B., Ehrenberg, M., and Frank, J. (2003). A cryo-electron micro-
scopic study of ribosome-bound termination factor RF2. Nature 421, 87–90. Yu, X., Jin, L., and Zhou, Z.H. (2008). 3.88 A structure of cytoplasmic polyhe-
drosis virus by cryo-electron microscopy. Nature 453, 415–419.
Rodriguez, S., Jafer, O., Goker, H., Summersgill, B.M., Zafarana, G., Gillis,
A.J., van Gurp, R.J., Oosterhuis, J.W., Lu, Y.J., Huddart, R., et al. (2003). Zhang, X., Jin, L., Fang, Q., Hui, W.H., and Zhou, Z.H. (2010). 3.3 A cryo-EM
Expression profile of genes from 12p in testicular germ cell tumors of adoles- structure of a nonenveloped virus reveals a priming mechanism for cell entry.
cents and adults associated with i(12p) and amplification at 12p11.2-p12.1. Cell 141, 472–482.
Oncogene 22, 1880–1891. Zheng, B., Yao, Y., Liu, Z., Deng, L., Anglin, J.L., Jiang, H., Prasad, B.V., and
Settembre, E.C., Chen, J.Z., Dormitzer, P.R., Grigorieff, N., and Harrison, S.C. Song, Y. (2013). Crystallographic Investigation and Selective Inhibition of
(2011). Atomic model of an infectious rotavirus particle. EMBO J. 30, 408–416. Mutant Isocitrate Dehydrogenase. ACS Med. Chem. Lett. 4, 542–546.
Cell 165, 1698–1707, June 16, 2016 1707

Resource
Programmable RNA Tracking in Live Cells

with CRISPR/Cas9
David A. Nelles,1,2 Mark Y. Fang,1 Mitchell R. O’Connell,4 Jia L. Xu,1 Sebastian J. Markmiller,1 Jennifer A. Doudna,4,5,6
and Gene W. Yeo1,2,3,*
1Department of Cellular and Molecular Medicine and Institute for Genomic Medicine, University of California, San Diego, La Jolla,
CA 92037, USA
2Materials Science and Engineering Graduate Program, University of California, San Diego, La Jolla, CA 92093, USA
3Department of Physiology, Yong Loo Lin School of Medicine, National University of Singapore, Singapore 1190777, Singapore
4Department of Molecular and Cell Biology and Center for RNA Systems Biology, University of California, Berkeley, Berkeley, CA 94720, USA
5Department of Chemistry, Innovative Genomics Initiative and Howard Hughes Medical Institute, University of California, Berkeley, Berkeley,
CA 94720, USA
6Physical Biosciences Division, Lawrence Berkeley National Laboratory, Berkeley, Berkeley, CA 94720, USA
*Correspondence: geneyeo@ucsd.edu
SUMMARY of DNA targeting for other applications such as gene expression

modulation (Qi et al., 2013) and imaging (Chen et al., 2013). Cas9
RNA-programmed genome editing using CRISPR/ and its associated single-guide RNA (sgRNA) require two critical
Cas9 from Streptococcus pyogenes has enabled features to target DNA: a short DNA sequence of the form
rapid and accessible alteration of specific genomic 50 -NGG-30 (where ‘‘N’’ = any nucleotide) known as the proto-
loci in many organisms. A flexible means to target spacer adjacent motif (PAM) and an adjacent sequence on the
RNA would allow alteration and imaging of endo- opposite DNA strand that is antisense to the sgRNA. By support-
ing DNA recognition with specificity determined entirely by a
genous RNA transcripts analogous to CRISPR/
short spacer sequence within the sgRNA, CRISPR/Cas9
Cas-based genomic tools, but most RNA target- provides uniquely flexible and accessible manipulation of the
ing methods rely on incorporation of exogenous genome. Manipulating cellular RNA content, in contrast, remains
tags. Here, we demonstrate that nuclease-inactive problematic. Whereas there exist robust means of attenuating
S. pyogenes CRISPR/Cas9 can bind RNA in a gene expression via RNAi and antisense oligonucleotides, other
nucleic-acid-programmed manner and allow endog- critical aspects of post-transcriptional gene expression regu-
enous RNA tracking in living cells. We show that lation such as subcellular trafficking, alternative splicing or
nuclear-localized RNA-targeting Cas9 (RCas9) is polyadenylation, and spatiotemporally restricted translation are
exported to the cytoplasm only in the presence of difficult to measure in living cells and are largely intractable.
sgRNAs targeting mRNA and observe accumulation Analogous to the assembly of zinc finger nucleases (Urnov
of ACTB, CCNA2, and TFRC mRNAs in RNA granules et al., 2010) and transcription activator-like effector nucleases
(TALEN) to recognize specific DNA sequences, efforts to recog-
that correlate with fluorescence in situ hybridization.
nize specific RNA sequences have focused on engineered
We also demonstrate time-resolved measurements RNA-binding domains. Pumilio and FBF homology (PUF) proteins
of ACTB mRNA trafficking to stress granules. Our carry well-defined modules capable of recognizing a single base
results establish RCas9 as a means to track RNA each and have supported successful targeting of a handful of
in living cells in a programmable manner without transcripts for imaging and other manipulations (Filipovska
genetically encoded tags. et al., 2011; Ozawa et al., 2007; Wang et al., 2009). PUF proteins
can be fused to arbitrary effector domains to alter or tag target
RNAs, but PUFs must be redesigned and validated for each
INTRODUCTION RNA target and can only recognize eight contiguous bases,
which does not allow unique discrimination in the transcriptome.
Clustered regularly-interspaced short palindromic repeats Molecular beacons are self-quenched synthetic oligonucleotides
(CRISPRs) form the basis of adaptive immune systems in bac- that fluoresce upon binding to target RNAs and allow RNA detec-
teria and archaea by encoding CRISPR RNAs that guide tion without construction of a target-specific protein (Sokol et al.,
CRISPR-associated (Cas) nucleases to invading genetic material 1998). But molecular beacons must be microinjected to avoid the
(Wiedenheft et al., 2012). Cas9 from the type II CRISPR system of generation of excessive background signal associated with en-
S. pyogenes has been repurposed for genome engineering in eu- dosome-trapped probes and are limited to imaging applications.
karyotic organisms (Hwang et al., 2013; Li et al., 2013a; Mali An alternative approach to recognition of RNA substrates is to
et al., 2013; Nakayama et al., 2013; Sander and Joung, 2014; introduce RNA aptamers into target RNAs, enabling specific
Yang et al., 2014) and is rapidly proving to be an efficient means and strong association of cognate aptamer-binding proteins
488 Cell 165, 488–496, April 7, 2016 ª2016 Elsevier Inc.

such as the MS2 coat protein (Fouts et al., 1997). This approach nuclear export of a nuclear-localized RCas9 system. Specif-
has enabled tracking of RNA localization in living cells over time ically, we tested whether mCherry-tagged Cas9 containing a
with high sensitivity (Bertrand et al., 1998) but relies upon labo- nuclear localization signal (NLS) can be co-exported from the
rious genetic manipulation of the target RNA and is not suitable nucleus with an mRNA in the presence of a cognate sgRNA
for recognition of arbitrary RNA sequences. Furthermore, inser- and PAMmer designed to recognize that mRNA (Figure 1A).
tion of exogenous aptamer sequence has the potential to inter- Nuclease null Cas9 (dCas9) was fused to two SV40 NLS se-
fere with endogenous RNA functions. Analogous to CRISPR/ quences at the C terminus with the coding sequence for mCherry
Cas9-based recognition of DNA, programmable RNA recognition and cloned into a mammalian expression vector (dCas9-2xNLS-
based on nucleic acid specificity alone without the need for ge- mCherry, abbreviated as dCas9-mCherry). In a separate expres-
netic manipulation or libraries of RNA-binding proteins would sion vector, a modified sgRNA scaffold with an extended
greatly expand researchers’ ability to modify the mammalian stem-loop structure that improves association with Cas9 and
transcriptome and enable transcriptome engineering. mutations that eliminate a partial transcription termination
Although the CRISPR/Cas9 system has evolved to recognize sequence (Chen et al., 2013) was driven by the U6 small nuclear
double-stranded DNA, recent in vitro work has demonstrated RNA (snRNA) polymerase III promoter. The PAMmer was syn-
that programmable targeting of RNAs with Cas9 is possible by thesized as a mixed DNA and 20 -O-methyl (20 OMe) RNA oligo-
providing the PAM as part of an oligonucleotide (PAMmer) that nucleotide using standard phosphoramidite chemistry and
hybridizes to the target RNA (O’Connell et al., 2014). By taking purified using high-performance liquid chromatography (HPLC)
advantage of the Cas9 target search mechanism that relies on (see Tables S1 and S2 for target mRNA, sgRNA, and PAMmer
PAM sequences (Sternberg et al., 2014), a mismatched PAM sequences). As a proof of concept, we designed a sgRNA-
sequence in the PAMmer/RNA hybrid allows exclusive targeting PAMmer pair to target the 30 UTR of GAPDH mRNA (Figure 1B).
of RNA and not the encoding DNA. The high affinity and specificity As a negative control, we designed a sgRNA-PAMmer pair tar-
of RNA recognition by Cas9 in cell-free extracts and the success of geting a sequence in the l bacteriophage that is absent in human
genome targeting with Cas9 indicate the potential of CRISPR/ cells (‘‘N/A’’ sgRNA and PAMmer and ‘‘l2’’ sgRNA and PAMmer
Cas9 to support programmable RNA targeting in living cells. sequences in Tables S1 and S2). We observed that transiently
To assess the potential of Cas9 as a programmable RNA-bind- transfected dCas9-mCherry co-transfected with the negative
ing protein in live cells, we used a modified sgRNA scaffold with control sgRNA and PAMmer is almost exclusively nuclear with
improved expression and Cas9 association (Chen et al., 2013) only 12% of cells containing greater mCherry signal in the cyto-
with a stabilized PAMmer oligonucleotide that does not form a plasm compared to the nucleus (Figures 1B and 1C). When the
substrate for RNase H. We measured the degree of nuclear negative control PAMmer was replaced with the GAPDH-target-
export of a nuclear localization signal-tagged nuclease-deficient ing PAMmer, the results were identical. Importantly, upon co-
Cas9-GFP fusion and demonstrate that the sgRNA alone is suf- transfection of GAPDH-targeting sgRNA plasmid, we observed
ficient to promote nuclear export of Cas9 without influencing the that 84% of cells had greater mCherry signal in the cytoplasm.
abundance of the targeted mRNA or encoded protein. In order to Interestingly, even with a non-targeting PAMmer, 79% of cells
evaluate whether RNA-targeting Cas9 (RCas9) signal patterns had predominantly cytoplasmic mCherry signal, suggesting
correspond with an established untagged RNA-labeling method, that the sgRNA is the primary determinant of RNA substrate
we compared distributions of RCas9 and fluorescence in situ recognition. Indeed, the sgRNA targeting GAPDH resulted in a
hybridization (FISH) targeting ACTB mRNA. We observed high significant increase in the fraction of cells with cytoplasmic
correlation among FISH and RCas9 signal that was dependent mCherry signal compared to a non-targeting sgRNA (Figure 1C).
on the presence of a PAMmer, indicating the importance of Overall, these results are consistent with previous in vitro
the PAM for efficient RNA targeting. RNA trafficking and sub- RNA pull-down experiments showing that RNA binding by
cellular localization are critical to gene expression regulation Cas9:sgRNA is independent of but strengthened by the PAMmer
and reaction to stimuli such as cellular stress. To address (O’Connell et al., 2014). Thus, we demonstrate that RCas9 RNA
whether RCas9 allows tracking of RNA to oxidative stress- recognition is programmable with a sgRNA and a PAMmer de-
induced RNA/protein accumulations called stress granules, we signed to target a specific, abundant mRNA in live cells.
measured ACTB, TFRC, and CCNA2 mRNA association with
stress granules in cells subjected to sodium arsenite. Finally, Recognition of an mRNA with RNA-Targeting Cas9 Does
we demonstrated the ability of RCas9 to track trafficking of Not Alter RNA Abundance or Amount of Translated
ACTB mRNA to stress granules over time in living cells. This Protein
work establishes the ability of RCas9 to bind RNA in live cells To further characterize the interaction between RCas9 and a
and sets the foundation for manipulation of the transcriptome target mRNA, we directed RCas9 to the 30 UTR of Renilla lucif-
in addition to the genome by CRISPR/Cas9. erase carrying a commonly used RNA tag for RNA tracking
from the MS2 bacteriophage (Fouts et al., 1997) and a sequence
RESULTS targeted by a previously validated sgRNA:PAMmer pair (‘‘l2’’;
see Table S2; O’Connell et al., 2014; Figure 1D). RNA immuno-
RNA-Targeting Cas9 Export from Nucleus in Presence of precipitation with an antibody recognizing EGFP revealed a
sgRNA Targeting GAPDH mRNA 4-fold greater association of luciferase mRNA to dCas9-EGFP
We initially assessed the ability of RCas9 to recognize specific in the presence of a cognate sgRNA and PAMmer compared
mRNA substrates in human cells by evaluating the degree of to non-targeting sgRNA (sense to the l2 RNA sequence) with a
Cell 165, 488–496, April 7, 2016 ª2016 Elsevier Inc. 489

A C D
RNA-targeting Cas9
E F
G
B
Figure 1. Targeting mRNA in Living Cells with RCas9

(A) Components required for RNA-targeting Cas9 (RCas9) recognition of mRNA include a nuclear localization signal-tagged nuclease-inactive Cas9 fused to a
fluorescent protein such as GFP, a modified sgRNA with expression driven by the U6 polymerase III promoter, and a PAMmer composed of DNA and 20 -O-methyl
RNA bases with a phosphodiester backbone. The sgRNA and PAMmer are antisense to adjacent regions of the target mRNA whose encoding DNA does not carry
a PAM sequence. After formation of the RCas9/mRNA complex in the nucleus, the complex is exported to the cytoplasm.
(B) RCas9 nuclear co-export with GAPDH mRNA. The RCas9 system was delivered to U2OS cells with a sgRNA and PAMmer targeting the 30 UTR of GAPDH or
sgRNA and PAMmer targeting a sequence from l bacteriophage that should not be present in human cells (‘‘N/A’’). Cellular nuclei are outlined with a dashed white
line. Scale bars represent 5 microns.
(C) Fraction of cells with cytoplasmic RCas9 signal. Mean values ± SD (n = 50).
(D) A plasmid carrying the Renilla luciferase open reading frame with a b-globin 30 UTR containing a target site for RCas9 and MS2 aptamer. A PEST protein
degradation signal was appended to luciferase to reveal any translational effects of RCas9 binding to the mRNA.
(E) RNA immunoprecipitation of EGFP after transient transfection of the RCas9 system in HEK293T cells targeting the luciferase mRNA compared to non-
targeting sgRNA and PAMmer or EGFP alone. Mean values ± SD (n = 3).
(F and G) Renilla luciferase mRNA (F) and protein (G) abundances were compared among the targeting and non-targeting conditions. Mean values ± SD (n = 4).
p values are calculated by Student’s t test, and one, two, and three asterisks represent p values less than 0.05, 0.01, and 0.001, respectively. See also Figure S1.
scrambled PAMmer or to EGFP protein alone (Figure 1E). We cence-activated cell sorting (FACS) gated on cells positive for
next measured the effect of RCas9 targeting on luciferase both Cas9 and sgRNA transfection. We observed no significant
mRNA abundance by qRT-PCR. We observed no significant dif- differences in GAPDH and ACTB mRNA or protein levels as
ference in the abundance of MS2-tagged luciferase mRNA in the measured by western blot and qRT-PCR among samples with
presence of the targeting or non-targeting RCas9 system or RCas9-targeting ACTB mRNA, GAPDH mRNA, cells transfected
EGFP alone. In contrast, co-expression of EGFP fused to the with GFP instead of dCas9-GFP, and RCas9-targeting sequence
MS2 coat protein (MCP) recognizing the MS2 aptamer had a sig- from l bacteriophage (Figure S1; Tables S1 and S2). Our results
nificant stabilizing effect (Figure 1F). We also considered poten- demonstrate that RCas9 recognition of RNA with a sgRNA and
tial effects of RCas9 targeting on translation of luciferase (Fig- PAMmer does not perturb RNA and encoded protein levels.
ure 1G) and observed that the presence of the targeting
sgRNA and PAMmer caused no significant changes in protein Correlation of RNA-Targeting Cas9 Signal Distributions
levels compared to non-targeting RCas9. To confirm that the with an Established Untagged RNA Localization
RCas9/sgRNA/PAMmer complex does not perturb its endoge- Measurement
nous RNA substrates, we evaluated the influence of RCas9 tar- To assess whether RCas9 signal distributions correlate with an
geting on ACTB and GAPDH mRNAs. We transiently transfected orthogonal method to measure RNA localization, we targeted
the RCas9 system and isolated transfected cells using fluores- the 30 UTR of ACTB (‘‘+’’ sgRNA and ‘‘+’’ PAMmer) and compared

A
Figure 2. ACTB mRNA Localization with RCas9 Compared to FISH

(A) The RCas9 system was delivered to U2OS cells, and the cells were subjected to FISH for ACTB mRNA. RCas9 with sgRNA- and PAMmer-targeting ACTB
mRNA was compared to non-targeting sgRNA and PAMmer antisense to a sequence from l bacteriophage (‘‘–’’ sgRNA and ‘‘–’’ PAMmer). White dotted lines
delineate the cellular boundaries, and black dotted lines delineate cellular nuclei. Scale bars represent 5 microns. Insets (on right) are delineated by white boxes.
(B) Pixel-by-pixel analysis of RCas9 and FISH colocalization using the Manders’ overlap coefficient is summarized with a cumulative distribution of the percent of
cytoplasmic area with overlapping signal in 60–80 cells in each condition. The presence of the PAMmer produces a significantly greater colocalization among
RCas9 and FISH in the presence of the sgRNA-targeting ACTB mRNA (p = 0.035; two-tailed Mann-Whitney U test).
See also Figure S2.
dCas9-GFP signal to RNA FISH for ACTB mRNA (Figure 2A) and (Figure 2A). This result is consistent with weaker binding of
non-targeting sgRNA and PAMmer (‘‘–’’ sgRNA and ‘‘–’’ PAMmer RCas9 with a non-targeting PAMmer observed in cell-free sys-
with sequences corresponding to l bacteriophage). By tems (O’Connell et al., 2014). A non-targeting sgRNA results in
comparing the Manders’ overlap coefficients that describe largely nuclear retention of RCas9 signal with low correlation be-
pixel-by-pixel overlap among FISH and RCas9 (Manders et al., tween cytoplasmic RCas9 signal and FISH (Figures 2A and 2B).
1992; Figures 2B and S2), we determined that the sgRNA pri- We conclude that the localization patterns of single EGFP-fused
marily accounts for colocalization among FISH and RCas9 with RCas9 in live cells correlates surprisingly well with localization
maximal overlap in the presence of both sgRNA- and PAMmer- obtained by FISH using tens of probes per mRNA in fixed cells.
targeting ACTB mRNA. A non-targeting PAMmer results in signif-
icantly less overlap (Figure 2B; p = 0.035; Mann-Whitney U test) Tracking RNA Trafficking to Stress Granules over Time
and produces a diffuse pattern of RCas9 signal in the cytoplasm Many polyadenylated RNAs including ACTB mRNA are known
that contrasts with the highly localized pattern revealed by FISH to localize to stress granules (Unsworth et al., 2010) during

oxidative stress. We simultaneously tracked ACTB mRNA using evaluated the potential of RNA recognition with Cas9 by tar-
RCas9 and RFP fused to the Ras GTPase-activating protein- geting GAPDH mRNA and evaluating degree of nuclear export
binding protein 1 (G3BP1) protein, a well-described marker for of dCas9-mCherry (Figure 1B). Robust cytoplasmic localization
stress granules (Tourrière et al., 2003). Application of sodium of dCas9-mCherry in the presence of a sgRNA-targeting GAPDH
arsenite to induce cellular stress results in an accumulation of mRNA compared to nuclear retention in the presence of a
mRNAs, including ACTB mRNA, into G3BP1-positive granules. non-targeting sgRNA indicated that Cas9 association with the
We observed accumulation of RCas9 signal to G3BP1-positive mRNA was sufficiently stable to support co-export from the
foci in the presence of the RCas9 system targeting ACTB nucleus.
mRNA. To verify that this accretion of RCas9 signal required spe- RCas9 as an RNA-imaging reagent requires that RNA recogni-
cific RNA recognition, we utilized three different sgRNAs and tion by RCas9 does not interfere with normal RNA metabolism.
PAMmers targeting sequences from l bacteriophage (NTC1–3; Here, we show that RCas9 binding within the 30 UTR of Renilla
Figures 3A and 3B). Indeed, in these negative controls, increased luciferase does not affect its mRNA abundance and translation
signal in G3BP1-positive foci was not observed. We also evalu- (Figures 1F and 1G). The utility of RCas9 for imaging and other
ated the ability of RCas9 to track less-abundant transcripts applications hinges on the recognition of endogenous tran-
CCNA2 and TFRC mRNA to stress granules. Notably, we scripts, so we evaluated the influence of RCas9 targeting on
observed statistically significantly higher fractions of stress gran- GAPDH and ACTB mRNAs and observed no significant differ-
ules that accumulate RCas9 signal (39% and 23%, respectively) ences among the mRNA and protein abundances by western
compared to non-targeting controls (Figures 3A and 3C). Next, blot analysis and qRT-PCR (Figure S1). These results indicate
we tracked RCas9 signal in stressed, live cells over time that RCas9 targeting these 30 UTRs does not perturb the levels
(Figure 3B). We observed accumulation of RCas9 signal in of mRNA or encoded protein.
G3BP1-positive foci in a manner dependent on the presence We also evaluated the ability of RCas9 to reveal RNA localiza-
of sgRNA- and PAMmer-targeting ACTB mRNA. We also tion by comparing RCas9 signal patterns to FISH. We utilized a
observed that the rate and degree of RCas9 signal accumulation FISH probe set composed of tens of singly labeled probes tar-
in stress granules is dependent on dosage of the stressor sodium geting ACTB mRNA and compared FISH signal distributions to
arsenite (Figure 3D). These results indicate the potential of a single dCas9-GFP/sgRNA/PAMmer that recognizes the
RCas9 as a means to generate time-resolved RNA localization ACTB mRNA. Our findings indicate that the sgRNA primarily de-
measurements. termines the degree of overlap among the FISH and RCas9 sig-
nals whereas the PAMmer plays a significant but secondary role.
DISCUSSION Importantly, in contrast to other untagged RNA localization
determination methods such as FISH and molecular beacons,
Effective RNA recognition by Cas9 in living cells while avoiding RCas9 is compatible with tracking untagged RNA localization
perturbation of the target transcript relies on careful design of in living cells and can be delivered rapidly to cells using estab-
the PAMmer and delivery of Cas9 and its cognate guide RNA lished transfection methods. We also note that the distribution
to the appropriate cellular compartments. Binding of Cas9 to nu- of ACTB mRNA was visualized using a single EGFP tag per tran-
cleic acids requires two critical features: a PAM DNA sequence script, and higher-sensitivity RNA tracking or single endogenous
and an adjacent spacer sequence antisense to the Cas9-associ- RNA molecule visualization may be possible in the future with
ated sgRNA. By separating the PAM and sgRNA target among RCas9 targeting multiple sites in a transcript or with a multiply
two molecules (the PAMmer oligonucleotide and the target tagged dCas9 protein.
mRNA) that only associate in the presence of a target mRNA, Stress granules are translationally silent mRNA and protein
RCas9 allows recognition of RNA while avoiding the encoding accumulations that form in response to cellular stress and are
DNA. To avoid unwanted degradation of the target RNA, the increasingly thought to be involved with neurodegeneration (Li
PAMmer is composed of a mixed 20 OMe RNA and DNA that et al., 2013b). There are limited means that can track the move-
does not form a substrate for RNase H. Further, the sgRNA ment of endogenous RNA to these structures in live cells (Bertrand
features a modified scaffold that removes partial transcription et al., 1998). In addition to ACTB mRNA, we demonstrate that
termination sequences and a modified structure that promotes RCas9 is capable of measuring association of CCNA2 and
association with Cas9 (Chen et al., 2013). Other CRISPR/Cas TFRC mRNA trafficking to stress granules (Figure 3A). Upon
systems have demonstrated RNA binding in bacteria (Hale stress induction with sodium arsenite, we observed that 50%,
et al., 2009; Sampson et al., 2013) or eukaryotes (Price et al., 39%, and 23% of stress granules featured overlapping RCas9
2015), although these systems cannot discriminate RNA from foci when targeting ACTB, TFRC, and CCNA2 mRNAs, respec-
DNA targets, feature RNA-targeting rules that remain unclear, tively (Figure 3C). This result correlates with the expression levels
or rely on large protein complexes that may be difficult to recon- of these transcripts (Figure S3) as ACTB is expressed about 8 and
stitute in mammalian cells. 11 times more highly than CCNA2 and TFRC, respectively. We
In this work, we demonstrate RCas9-based recognition of also observed that RCas9 is capable of tracking RNA localization
GAPDH, ACTB, CCNA2, and TFRC mRNAs in live cells. Because over time as ACTB mRNA is trafficked to stress granules over a
the U6-driven sgRNA is largely restricted to the nucleus, the period of 30 min (Figure 3B). We noted a dependence of RCas9
NLS-tagged dCas9 allows association with its sgRNA and signal accumulation in stress granules on stressor concentration
subsequent interaction with the target mRNA before nuclear (Figure 3D). This approach for live-cell RNA tracking stands in
co-export with the target mRNA. As an initial experiment, we contrast to molecular beacons and aptamer-based RNA-tracking

A B
C D
Figure 3. Tracking of mRNA Trafficking to Stress Granules with RCas9

(A) The RCas9 system targeting ACTB, TFRC, or CCNA2 mRNAs or one of three non-targeting controls (NTCs) was delivered to HEK293T cells expressing
G3BP1, a protein known to be trafficked to stress granules, fused to RFP. Cells were treated with sodium arsenite; fixed; subjected to FISH for ACTB, TFRC, or
CCNA2 mRNA; and imaged.
(B) RNA trafficking to stress granules was imaged in real time using cells harboring RCas9-targeting ACTB mRNA. At time zero, cells were imaged and sodium
arsenite applied. Sixty minutes later, cells were imaged again, and a comparison of RCas9 and G3BP1-positive stress granules revealed close correlation of foci
only in the presence of sgRNA- and PAMmer-targeting ACTB mRNA.
(C) The fraction of stress granules with RCas9 foci when targeting three mRNAs (ACTB, TFRC, and CCNA2) compared to three non-targeting controls. Error
bars ± SD calculated from 50 cells from each of three biological replicates. p values among the RCas9 system targeting ACTB, TFRC, and CCNA2 mRNA
are < 0.001 when compared to each of the NTC conditions. p values were calculated with Student’s t test.
(D) In a similar experiment, RCas9-targeting ACTB mRNA signal accumulation in stress granules was tracked over time. Eight to eleven stress granules were
tracked in each condition with time points every 8 min for 32 min.
Scale bars represent 5 microns. See also Figure S3.
methods, which suffer from delivery issues and/or require alter- cent proteins or splicing factors adjacent to alternatively
ation of the target RNA sequence via incorporation of RNA tags. spliced exons. Further, the nucleic-acid-programmable nature
Future applications of RCas9 could allow the measurement of RCas9 lends itself to multiplexed targeting (Cong et al.,
or alteration of RNA splicing via recruitment of split fluores- 2013) and the use of Cas9 proteins that bind orthogonal sgRNAs

(Esvelt et al., 2013) could support distinct activities on multiple Cell Lines
target RNAs simultaneously. It is possible that the simple RNA U2OS and HEK293T cells were grown in DMEM supplemented with 10% fetal
bovine serum, Glutamax, penicillin/streptomycin, and non-essential amino
targeting afforded by RCas9 could support the development of
acids (Life Technologies). Cells were passaged every 3 or 4 days with TrypLE
sensors that recognize specific healthy or disease-related gene EXPRESS (Life Technologies) using standard methods and maintained in a
expression patterns and reprogram cell behavior via alteration humidified incubator at 37 C with 5% CO2.
of gene expression or concatenation of enzymes on a target
RNA (Delebecque et al., 2011; Sachdeva et al., 2014). Efforts GAPDH and ACTB mRNA Targeting with RCas9
toward Cas9 delivery in vivo are underway (Dow et al., 2015; U2OS cells cultured as described above were passaged at 80% confluency.
Swiech et al., 2015; Zuris et al., 2015), and these efforts com- Glass-bottom 96-well plates or chamber slides were coated with 20 mg/ml fibro-
nectin in PBS for 2 hr at 37 C and then the fibronectin solution was aspirated and
bined with existing oligonucleotide chemistries (Bennett and
20,000 cells were plated in each well. Sixteen hours later, cells were transfected
Swayze, 2010) could support in vivo delivery of the RCas9 sys- with the sgRNA and dCas9-2xNLS-EGFP plasmids using Lipofectamine 3000
tem for targeted modulation of many features of RNA processing (Life Technologies) according to the manufacturer’s instructions. pCMV-Renilla
in living organisms. luciferase was co-transfected in these experiments so that total transfected pro-
RNA is subject to processing steps that include alternative tein load was the similar among various dosages of sgRNA and dCas9. The
splicing, nuclear export, subcellular transport, and base or mass ratio of sgRNA and dCas9-EGFP plasmids was 5:1 in experiments
described in this work where the amount of dCas9-EGFP was fixed at 10% of
backbone modifications that work in concert to regulate gene
total transfected material. Immediately after plasmid transfection, PAMmers
expression. The development of a programmable means of were transfected using Lipofectamine RNAiMax (Life Technologies) according
RNA recognition in order to measure and manipulate these to manufacturer’s instructions. Twenty-four hours after transfection, cells
processes has been sought after in biotechnology for decades. were washed with PBS and fixed with 3.7% paraformaldehyde in PBS, permea-
This work is, to our knowledge, the first demonstration of bilized with 70% ethanol at 4 C for 1 hr, and mounted using Prolong Gold
nucleic-acid-programmed RNA recognition in living cells with Antifade mounting medium with DAPI (Life Technologies). Confocal microscopy
was conducted using a Zeiss LSM 810 confocal microscope.
CRISPR/Cas9. By relying upon a sgRNA and PAMmer to deter-
Nuclear export of RCas9 in the presence of sgRNA and PAMmer targeting
mine target specificity, RCas9 supports versatile and unambig-
the 30 UTR of GAPDH mRNA was analyzed by measuring the average signal
uous RNA recognition analogous to DNA recognition afforded in the nuclei and cytoplasm of individual cells. Cells with average cytoplasmic
by CRISPR/Cas9. The diverse applications supported by signal greater than the average nuclear signal were considered positive and
DNA-targeted CRISPR/Cas9 range from directed cleavage, counted toward the fraction of cells with cytoplasmic RCas9 signal.
imaging, transcription modulation, and targeted methylation,
indicating the utility of both the native nucleolytic activity of RNA Immunoprecipitation
HEK293T cells cultured as described above were passaged at 80% conflu-
Cas9 as well as the range of activities supported by Cas9-fused
ency, and 600,000 cells were seeded in each well of 6-well tissue culture plates
effectors. In addition to providing a flexible means to track this coated with poly-L-lysine. Sixteen hours later, cells were co-transfected with
RNA in live cells, future developments of RCas9 could include the RCas9 system as described above or plasmids encoding MS2-EGFP or
effectors that modulate a variety of RNA-processing steps EGFP along with a plasmid encoding the model Renilla luciferase mRNA driven
with applications in synthetic biology and disease modeling by a cytomegalovirus (CMV) promoter (Figure 1D). Twenty-four hours later, the
or treatment. growth media was aspirated and the cells were washed with PBS. 1% parafor-
maldehyde in PBS was applied to the cells, incubated for 10 min at room tem-
perature, and then the solution was aspirated and the cells washed twice with
EXPERIMENTAL PROCEDURES cold PBS. Next, the cells were scraped from the wells in cold PBS and the cell
suspension was centrifuged at 800 3 g for 4 min to pellet the cells. The cells
Plasmid Construction, PAMmer Synthesis, and Target Site Choice were washed once more and then resuspended in radioimmunoprecipitation
The dCas9-2xNLS sequence was amplified from pHR-SFFV-dCas9-BFP- assay (RIPA) buffer with protease inhibitors (Roche) and sonicated for 5 min
KRAB (a gift from Stanley Qi and Jonathan Weissman; Addgene plasmid no. in a Bioruptor sonicator (50% duty cycle; 1 min period). Insoluble material
46911), tagged with two SV40 NLSs on the C terminus, and fused to EGFP was pelleted after a high-speed centrifugation, and the supernatant was
or mCherry in pCDNA 3.1 (Life Technologies) using Gibson assembly. To applied to protein G DynaBeads (Life Technologies) coated with mouse anti-
construct the sgRNA scaffold construct, the human U6 polymerase III pro- GFP antibody (Roche). After overnight incubation at 4 C, the bead supernatant
moter with the modified sgRNA scaffold (Chen et al., 2013) was purchased was retained and beads washed three times with RIPA buffer containing
as a gBlock from IDT with two BbsI restriction sites at the 50 end of the sgRNA 0.02% Tween-20 and once with DNase buffer (350 mM Tris-HCl [pH 6.5];
scaffold (see sequence in Table S2) and cloned into the multiple cloning site of 50 mM MgCl2; 5 mM DTT). The beads were resuspended in DNase buffer
pBlueScript II SK (+) (Agilent) using Gibson assembly. Phosphorylated oligonu- and TURBO DNase (Life Technologies) was added to 0.08 units/ml. The
cleotides encoding the sgRNA sequences (with overhangs 50 CACC on the beads were incubated at 37 C for 30 min, and then proteinase K (NEB) was
RNA antisense strand and 50 AAAC on the sense strand) were ligated into added to 0.1 U/ml and incubated with shaking at 37 C for 30 min. Next, urea
BbsI-digested sgRNA scaffold construct to produce sgRNAs targeting the 30 was added to 2.5 M and the beads were incubated with shaking at 37 C for
UTR of GAPDH, ACTB, and Renilla luciferase mRNAs (see Tables S1 and 30 min. The bead supernatant was collected and subjected to two sequential
S2). The luciferase-PEST construct for pull-down and abundance experiments phenol:chloroform:isoamyl alcohol (25:24:1) extractions followed by three
was modified from plasmid pRLuc (gift from Jens Lykke-Andersen, UCSD). chloroform extractions. The RNA was precipitated and reverse transcribed
pCMV-Renilla luciferase is a version of the same construct lacking MS2 and using SuperScript III (Life Technologies) using random hexamer primers, and
RCas9 target sites. relative abundance of Renilla luciferase RNA on the beads was compared to
RCas9 target sites were chosen with a combination of the IDT antisense the supernatant using RT-PCR (see Table S3 for primer sequences).
oligonucleotide design tool and the microarray probe design tools Picky
(Chou et al., 2004) and OligoWiz (Wernersson and Nielsen, 2005). We designed Measurements of Influence of RCas9 on RNA Stability and
PAMmers against high-confidence sites with eight bases on the 50 end beyond Translation
the PAM sequence. PAMmers were composed of mixed 20 OMe RNA and DNA HEK293T cells were cultured as described above, passaged and plated in
bases and purified by HPLC (Integrated DNA Technologies). 96- or 12-well tissue culture plates, and co-transfected 24 hr later with the

RCas9 system as described above and the Renilla luciferase construct car- of each pixel by the average cytoplasmic signal, which reduced cytoplasmic
rying MS2 and RCas9 binding sites in the 30 UTR. In the protein abundance signal to a value of ‘‘1’’ and RCas9-positive foci to a value of ‘‘2’’ or more.
measurements, a small amount of CMV-driven firefly luciferase vector (5% Next, a Gaussian blur was applied to improve foci definition. These images
of total transfected plasmid) was co-transfected as a transfection control. are reported in Figure 3A. The accumulation of RCas9 signal over time in stress
For RNA stability measurements, RNA was isolated 24 hr after transfection, granules was calculated by recording average signal intensity in the RCas9
DNase treated, and reverse transcribed with Superscript III (Life Technologies) channel in areas with overlapping G3BP1-RFP foci at each time point. Average
using dT(20) primers according the manufacturer’s instructions. The amount of signal intensity in the RCas9 channel surrounding G3BP1-RFP foci was re-
Renilla luciferase cDNA relative to GAPDH was then measured using RT-PCR corded as background and subtracted from the previous value to produce
(see Table S3 for primer sequences). For the translation studies, Renilla and the background-adjusted RCas9 signal in stress granules.
firefly luciferase protein were measured with the Dual Luciferase Kit (Promega)
according to the manufacturer’s instructions.
For measuring the influence of RCas9 on GAPDH and ACTB mRNA and pro-
tein abundance, the RCas9 system targeting the 30 UTR of each transcript,
Supplemental Information includes three figures and three tables and can be
GFP alone, or RCas9 with the l2-targeting sgRNA and PAMmer was trans-
found with this article online at http://dx.doi.org/10.1016/j.cell.2016.02.054.
fected. Modified sgRNA vectors carrying pGK-driven blue fluorescent protein
A video abstract is available at http://dx.doi.org/10.1016/j.cell.2016.02.
(BFP) were used in this experiment, and transfected cells were isolated using
054#mmc3.
FACS gated on cells positive for GFP and BFP expression. Transfected cells
were lysed and subjected to western blotting and RT-PCR for ACTB, GAPDH,
and TUBA1A transcripts. AUTHOR CONTRIBUTIONS
FISH Conceptualization, D.A.N., M.R.O., J.A.D., and G.W.Y.; Methodology, D.A.N.

Stellaris FISH Probes recognizing human ACTB, CCNA2, and TFRC mRNAs and M.R.O.; Investigation, D.A.N., M.Y.F., J.L.X., and S.J.M.; Writing – Original
and labeled with Quasar 670 (VSMF-2003-5; Biosearch Technologies) were Draft, D.A.N. and G.W.Y.; Writing – Review & Editing, D.A.N., G.W.Y., and
hybridized to cells 24 hr after transfection with the RCas9 system. Hybridiza- M.R.O.; Funding Acquisition, G.W.Y.; Supervision, G.W.Y.
tion was conducted according to the manufacturer’s instructions. Confocal
microscopy was conducted using an Olympus FV1000 confocal microscope ACKNOWLEDGMENTS
or Zeiss LSM 810 confocal microscope.
We acknowledge members of the G.W.Y. lab, particularly Ashleigh Schaffer,
Overlap Analysis for FISH and RCas9 Stefan Aigner, and Ron Batra for critical comments. This work was supported
Colocalization analysis among FISH and RCas9-targeting ACTB mRNA was by grants from the NIH (HG004659 and NS075449) and from the California
conducted using the Coloc 2 plugin from the image analysis software FIJI Institute of Regenerative Medicine (RB3-05009 and RB4-06045) to G.W.Y.
(Schindelin et al., 2012). The cytoplasm of individual cells with similar D.A.N. is supported by the National Science Foundation Graduate Research
dCas9-EGFP transfection levels was selected, and the Coloc 2 analysis was Fellowship (DGE-1144086) and Powell-Focht graduate fellowship. M.R.O. is
conducted using default parameters. The Manders’ overlap coefficient supported by a CJ Martin fellowship from the National Health and Medical
describing degree of overlap of the FISH signal with RCas9 for more than 60 Research Council (Australia). J.A.D. is a Howard Hughes Medical Institute
cells in each condition was compiled, and p values were calculated with the Investigator. G.W.Y. is an Alfred P. Sloan Research Fellow. J.A.D. is a co-
two-tailed Mann-Whitney U test. founder of Editas Medicine, Intellia Therapeutics, and Caribou Biosciences
and a scientific advisor to Caribou, Intellia, eFFECTOR Therapeutics, and
Tracking ACTB mRNA Trafficking to Stress Granules Driver.
A HEK293T cell line was genetically modified with a fusion of RFP to the C ter-
minus of Ras G3BP1 using CRISPR/Cas9. Briefly, a donor plasmid was Received: November 15, 2015
constructed consisting of the RFP ORF, a puromycin selection cassette, and Revised: February 5, 2016
flanking 1.5-kb homology arms directed at the G3BP1 locus. An sgRNA Accepted: February 24, 2016
sequence targeting the C terminus of G3BP1 was cloned into pSpCas9(BB)- Published: March 17, 2016
2A-GFP (pX458) (gift from Feng Zhang; Addgene plasmid no. 48138) and
co-transfected with the donor plasmid using Fugene HD (Roche) following
the manufacturer’s instructions. Forty-eight hours after transfection, cells REFERENCES
were selected with 1 mg/ml puromycin in growth medium for 14 days and
RFP-positive clones were selected and screened by PCR. Bennett, C.F., and Swayze, E.E. (2010). RNA targeting therapeutics: molecular
A clone with at least one modified allele was plated on glass chamber slides mechanisms of antisense oligonucleotides as a therapeutic platform. Annu.
coated with fibronectin and transfected with the RCas9 system targeting the 30 Rev. Pharmacol. Toxicol. 50, 259–293.
UTR of ACTB, CCNA2, or TFRC mRNA or one of three non-targeting sgRNA Bertrand, E., Chartrand, P., Schaefer, M., Shenoy, S.M., Singer, R.H., and
and PAMmer combinations (l2–l4) as described above (see Table S2 for Long, R.M. (1998). Localization of ASH1 mRNA particles in living yeast. Mol.
sgRNA and PAMmer sequences). Twenty-four hours after transfection, cells Cell 2, 437–445.
were subjected to 200 mM sodium arsenite for 30 min, fixed, and subjected Chen, B., Gilbert, L.A., Cimini, B.A., Schnitzbauer, J., Zhang, W., Li, G.W.,
to FISH and imaged with a Zeiss LSM 810 confocal microscope. In a sepa- Park, J., Blackburn, E.H., Weissman, J.S., Qi, L.S., and Huang, B. (2013).
rate experiment, cells were stressed with varying concentrations of sodium Dynamic imaging of genomic loci in living human cells by an optimized
arsenite and imaged with a Zeiss LSM 810 confocal microscope with a stage CRISPR/Cas system. Cell 155, 1479–1491.
incubator over 32 min. Cells were maintained at 37 C in a humidified atmo-
Chou, H.H., Hsia, A.P., Mooney, D.L., and Schnable, P.S. (2004). Picky: oligo
sphere and 5% CO2 and imaged every 8 min.
microarray design for large genomes. Bioinformatics 20, 2893–2902.
Analysis of ACTB mRNA Trafficking to Stress Granules Cong, L., Ran, F.A., Cox, D., Lin, S., Barretto, R., Habib, N., Hsu, P.D., Wu, X.,
Stress granules with average RCas9 signal at least 2-fold greater than sur- Jiang, W., Marraffini, L.A., and Zhang, F. (2013). Multiplex genome engineering
rounding cytoplasm were determined to be positive for RCas9. The average using CRISPR/Cas systems. Science 339, 819–823.
RCas9 signal in stress granules was calculated using the ‘‘measure’’ function Delebecque, C.J., Lindner, A.B., Silver, P.A., and Aldaye, F.A. (2011). Organi-
in Fiji and compared to the average cytoplasmic RCas9 signal in the sur- zation of intracellular reactions with rationally designed RNA assemblies.
rounding cytoplasm. RCas9-positive foci were visualized by dividing the value Science 333, 470–474.

Dow, L.E., Fisher, J., O’Rourke, K.P., Muley, A., Kastenhuber, E.R., Livshits, Sachdeva, G., Garg, A., Godding, D., Way, J.C., and Silver, P.A. (2014). In vivo
G., Tschaharganeh, D.F., Socci, N.D., and Lowe, S.W. (2015). Inducible in vivo co-localization of enzymes on RNA scaffolds increases metabolic production
genome editing with CRISPR-Cas9. Nat. Biotechnol. 33, 390–394. in a geometrically dependent manner. Nucleic Acids Res. 42, 9493–9503.
Esvelt, K.M., Mali, P., Braff, J.L., Moosburner, M., Yaung, S.J., and Church, Sampson, T.R., Saroj, S.D., Llewellyn, A.C., Tzeng, Y.L., and Weiss, D.S.
G.M. (2013). Orthogonal Cas9 proteins for RNA-guided gene regulation and (2013). A CRISPR/Cas system mediates bacterial innate immune evasion
editing. Nat. Methods 10, 1116–1121. and virulence. Nature 497, 254–257.
Filipovska, A., Razif, M.F., Nygård, K.K., and Rackham, O. (2011). A universal Sander, J.D., and Joung, J.K. (2014). CRISPR-Cas systems for editing, regu-
code for RNA recognition by PUF proteins. Nat. Chem. Biol. 7, 425–427. lating and targeting genomes. Nat. Biotechnol. 32, 347–355.
Fouts, D.E., True, H.L., and Celander, D.W. (1997). Functional recognition of Schindelin, J., Arganda-Carreras, I., Frise, E., Kaynig, V., Longair, M., Pietzsch,
fragmented operator sites by R17/MS2 coat protein, a translational repressor. T., Preibisch, S., Rueden, C., Saalfeld, S., Schmid, B., et al. (2012). Fiji: an
Nucleic Acids Res. 25, 4464–4473. open-source platform for biological-image analysis. Nat. Methods 9, 676–682.
Hale, C.R., Zhao, P., Olson, S., Duff, M.O., Graveley, B.R., Wells, L., Terns, Sokol, D.L., Zhang, X., Lu, P., and Gewirtz, A.M. (1998). Real time detection of
R.M., and Terns, M.P. (2009). RNA-guided RNA cleavage by a CRISPR DNA.RNA hybridization in living cells. Proc. Natl. Acad. Sci. USA 95, 11538–
RNA-Cas protein complex. Cell 139, 945–956. 11543.
Hwang, W.Y., Fu, Y., Reyon, D., Maeder, M.L., Tsai, S.Q., Sander, J.D., Peter- Sternberg, S.H., Redding, S., Jinek, M., Greene, E.C., and Doudna, J.A. (2014).
son, R.T., Yeh, J.R., and Joung, J.K. (2013). Efficient genome editing in zebra- DNA interrogation by the CRISPR RNA-guided endonuclease Cas9. Nature
fish using a CRISPR-Cas system. Nat. Biotechnol. 31, 227–229. 507, 62–67.
Li, D., Qiu, Z., Shao, Y., Chen, Y., Guan, Y., Liu, M., Li, Y., Gao, N., Wang, L., Lu, Swiech, L., Heidenreich, M., Banerjee, A., Habib, N., Li, Y., Trombetta, J., Sur,
X., et al. (2013a). Heritable gene targeting in the mouse and rat using a M., and Zhang, F. (2015). In vivo interrogation of gene function in the mamma-
CRISPR-Cas system. Nat. Biotechnol. 31, 681–683. lian brain using CRISPR-Cas9. Nat. Biotechnol. 33, 102–106.
Li, Y.R., King, O.D., Shorter, J., and Gitler, A.D. (2013b). Stress granules as Tourrière, H., Chebli, K., Zekri, L., Courselaud, B., Blanchard, J.M., Bertrand,
crucibles of ALS pathogenesis. J. Cell Biol. 201, 361–372. E., and Tazi, J. (2003). The RasGAP-associated endoribonuclease G3BP
Mali, P., Yang, L., Esvelt, K.M., Aach, J., Guell, M., DiCarlo, J.E., Norville, J.E., assembles stress granules. J. Cell Biol. 160, 823–831.
and Church, G.M. (2013). RNA-guided human genome engineering via Cas9. Unsworth, H., Raguz, S., Edwards, H.J., Higgins, C.F., and Yagüe, E. (2010).
Science 339, 823–826. mRNA escape from stress granule sequestration is dictated by localization
Manders, E.M., Stap, J., Brakenhoff, G.J., van Driel, R., and Aten, J.A. (1992). to the endoplasmic reticulum. FASEB J. 24, 3370–3380.
Dynamics of three-dimensional replication patterns during the S-phase, ana- Urnov, F.D., Rebar, E.J., Holmes, M.C., Zhang, H.S., and Gregory, P.D. (2010).
lysed by double labelling of DNA and confocal microscopy. J. Cell Sci. 103, Genome editing with engineered zinc finger nucleases. Nat. Rev. Genet. 11,
857–862. 636–646.
Nakayama, T., Fish, M.B., Fisher, M., Oomen-Hajagos, J., Thomsen, G.H., and Wang, Y., Cheong, C.G., Hall, T.M., and Wang, Z. (2009). Engineering splicing
Grainger, R.M. (2013). Simple and efficient CRISPR/Cas9-mediated targeted factors with designed specificities. Nat. Methods 6, 825–830.
mutagenesis in Xenopus tropicalis. Genesis 51, 835–843. Wernersson, R., and Nielsen, H.B. (2005). OligoWiz 2.0–integrating sequence
O’Connell, M.R., Oakes, B.L., Sternberg, S.H., East-Seletsky, A., Kaplan, M., feature annotation into the design of microarray probes. Nucleic Acids Res. 33,
and Doudna, J.A. (2014). Programmable RNA recognition and cleavage by W611–W615.
CRISPR/Cas9. Nature 516, 263–266. Wiedenheft, B., Sternberg, S.H., and Doudna, J.A. (2012). RNA-guided genetic
Ozawa, T., Natori, Y., Sato, M., and Umezawa, Y. (2007). Imaging dynamics of silencing systems in bacteria and archaea. Nature 482, 331–338.
endogenous mitochondrial RNA in single living cells. Nat. Methods 4, 413–419. Yang, D., Xu, J., Zhu, T., Fan, J., Lai, L., Zhang, J., and Chen, Y.E. (2014). Effec-
Price, A.A., Sampson, T.R., Ratner, H.K., Grakoui, A., and Weiss, D.S. (2015). tive gene targeting in rabbits using RNA-guided Cas9 nucleases. J. Mol. Cell
Cas9-mediated targeting of viral RNA in eukaryotic cells. Proc. Natl. Acad. Sci. Biol. 6, 97–99.
USA 112, 6164–6169. Zuris, J.A., Thompson, D.B., Shu, Y., Guilinger, J.P., Bessen, J.L., Hu, J.H.,
Qi, L.S., Larson, M.H., Gilbert, L.A., Doudna, J.A., Weissman, J.S., Arkin, A.P., Maeder, M.L., Joung, J.K., Chen, Z.Y., and Liu, D.R. (2015). Cationic lipid-
and Lim, W.A. (2013). Repurposing CRISPR as an RNA-guided platform for mediated delivery of proteins enables efficient protein-based genome editing
sequence-specific control of gene expression. Cell 152, 1173–1183. in vitro and in vivo. Nat. Biotechnol. 33, 73–80.

Resource
High-Throughput, High-Resolution Mapping

of Protein Localization in Mammalian Brain
by In Vivo Genome Editing
Takayasu Mikuni,1,4 Jun Nishiyama,1,4,* Ye Sun,1,2 Naomi Kamasawa,3 and Ryohei Yasuda1,*
1Neuronal Signal Transduction Group, Max Planck Florida Institute for Neuroscience, Jupiter, FL 33458, USA
2Integrative Program in Biology and Neuroscience, Florida Atlantic University, Jupiter, FL 33458, USA
3Electron Microscopy Core Facility, Max Planck Florida Institute for Neuroscience, Jupiter, FL 33458, USA
4Co-first author
*Correspondence: jun.nishiyama@mpfi.org (J.N.), ryohei.yasuda@mpfi.org (R.Y.)

SUMMARY antibodies with non-targeted proteins; overexpression often

causes mistargeting of the expressed protein and potential
A scalable and high-throughput method to identify changes in cell function. To address some of these issues,
precise subcellular localization of endogenous pro- knockin mice in which a specific protein is tagged with an
teins is essential for integrative understanding of a epitope tag or fluorescent protein can be used (Yang et al.,
cell at the molecular level. Here, we developed a sim- 2009). However, in dense tissue, such as mammalian brain, it
ple and generalizable technique to image endoge- is difficult to obtain images with high contrast in subcellular pro-
cesses when all cells are labeled. To overcome these problems,
nous proteins with high specificity, resolution, and
several methods have been recently developed for single-cell
contrast in single cells in mammalian brain tissue. labeling of endogenous proteins by using recombinant anti-
The technique, single-cell labeling of endogenous body-like proteins or a conditional tag knockin strategy (Fortin
proteins by clustered regularly interspaced short et al., 2014; Gross et al., 2013). However, none of these tech-
palindromic repeats (CRISPR)-Cas9-mediated ho- niques provides rapid, scalable, and high-throughput readouts
mology-directed repair (SLENDR), uses in vivo for the localization of endogenous proteins.
genome editing to insert a sequence encoding an Direct, single-cell manipulation of the genome in vivo to insert
epitope tag or a fluorescent protein to a gene of inter- a tag sequence to a gene of interest would overcome these
est by CRISPR-Cas9-mediated homology-directed limitations, providing rapid, specific, and sparse labeling of the
repair (HDR). Single-cell, HDR-mediated genome gene product. Genome editing based on the clustered regularly
editing was achieved by delivering the editing ma- interspaced short palindromic repeats (CRISPR)-associated
endonuclease Cas9 enables rapid and efficient modification of
chinery to dividing neuronal progenitors through in
the genome (Cong et al., 2013; Doudna and Charpentier, 2014;
utero electroporation. We demonstrate that SLENDR Hsu et al., 2014; Jinek et al., 2012; Sander and Joung, 2014;
allows rapid determination of the localization and dy- Wang et al., 2013; Yang et al., 2013). CRISPR-Cas9 induces tar-
namics of many endogenous proteins in various cell geted DNA double-strand breaks in the genome, which are then
types, regions, and ages in the brain. Thus, SLENDR repaired through either non-homologous end-joining (NHEJ) or
provides a high-throughput platform to map the sub- homology-directed repair (HDR) pathways (Cox et al., 2015;
cellular localization of endogenous proteins with the Doudna and Charpentier, 2014; Hsu et al., 2014; Sander and
resolution of micro- to nanometers in the brain. Joung, 2014; Yang et al., 2013). Although introducing frame-shift
knockout mutations through NHEJ at the single-cell level has
been established (Straub et al., 2014; Swiech et al., 2015), tar-
INTRODUCTION geted insertion of a sequence through HDR has not been
possible in the mammalian brain in vivo (Heidenreich and Zhang,
Precise mapping of a large number of proteins with subcellular 2016; Platt et al., 2014; Xue et al., 2014; Yin et al., 2014). This is
resolution is essential to understand cellular processes. Thus, due to the lack of homologous recombination activity in postmi-
it is critical to develop a rapid and scalable method to determine totic cells and the inefficient delivery of HDR machinery to target
the localization of proteins with high specificity, resolution, and cells (Chu et al., 2015; Cox et al., 2015; Heidenreich and Zhang,
contrast. Conventionally, either immunostaining of endogenous 2016; Hsu et al., 2014; Maruyama et al., 2015; Saleh-Gohari and
proteins or overexpression of proteins fused with epitope tags Helleday, 2004).
or fluorescent proteins have been used to determine protein Here, we developed single-cell labeling of endogenous pro-
localization. These methods, however, have significant prob- teins by CRISPR-Cas9-mediated homology-directed repair
lems: immunostaining often suffers from the lack of specific an- (SLENDR), a technique that allows HDR-mediated genome edit-
tibodies against a protein of interest and the cross-reaction of ing in the mammalian brain in vivo. Since HDR is known to
Cell 165, 1803–1817, June 16, 2016 ª 2016 Elsevier Inc. 1803
A C-terminal labeling C N-terminal labeling
(HA tag) CRISPR-Cas9 (HA tag) CRISPR-Cas9
Target PAM
5’...tcctgccgcagtaaggattctttcttcctgaat...3’
taa 5’...acgccgagcccgatcgccaccgccatggccacc...3’
atg
3’...aggacggcgtcattcctaagaaagaaggactta...5’ 3’...tgcggctcgggctagcggtggcggtaccggtgg...5’
PAM Target
CaMKIIα CaMKIIβ
199 87 27 85 (bases) 199 103 27 69 (bases)

Oligo donor HA Oligo donor HA
Homologous recombination Homologous recombination
CaMKIIα CaMKIIβ
HA HA
Primer set, control Primer set, control
Primer set, recombination CaMKIIα-R1 Primer set, recombination
CaMKIIα-F1 HA-R1 CaMKIIβ-F1 HA-F1 CaMKIIβ-R1
B D HA-CaMKII DAPI HA-CaMKII mEGFP

CaMKIIα-HA DAPI CaMKIIα-HA mEGFP
sgRNA_CaMKIIα
sgRNA_CaMKII
E12 IUE
E12 IUE
cortex cortex
sgRNA_CaMKIIα
sgRNA_CaMKIIα sgRNA_CaMKII
sgRNA_CaMKII
E15 IUE
E15 IUE
E F
CaMKIIα CaMKII sgRNA_CaMKIIα - - +
10 CaMKIIα HA tag
- - 5’-
HA(+) / mEGFP(+) (%)
sgRNA_CaMKIIβ + -3’
ssODN_CaMKIIα - + +
8 Primer set,
recombination
6 Primer set,
control
4 G
sgRNA_CaMKIIβ - - + CaMKIIβ HA tag
2 ** **
sgRNA_CaMKIIα - + - 5’- -3’
** ** ssODN_CaMKIIβ - + + (complementary strand)
0 Primer set,
E12 E12 E15 E12 E12 E15 recombination
Primer set,
L2/3 L4-6 L2/3 L2/3 L4-6 L2/3 control
1804 Cell 165, 1803–1817, June 16, 2016

predominantly occur in the S and G2 phases of the cell cycle double-strand breaks were repaired through the NHEJ pathway,
(Chu et al., 2015; Heidenreich and Zhang, 2016; Hsu et al., we selected the target sequences throughout this study so that
2014; Maruyama et al., 2015; Saleh-Gohari and Helleday, the CRISPR-Cas9 cleavage sites were located either in the
2004), we targeted mitotic neuronal progenitors, which presum- non-coding region upstream of the start codon or within 10 bp
ably have homologous recombination activity. We introduced from the stop codon. We also selected the target sequences
CRISPR-Cas9-based HDR machinery into progenitor cells in and designed ssODNs so that Cas9 could not recognize the
the embryonic mouse brain several days before their final neuro- loci after HDR was completed (see Supplemental Information).
genic divisions using in utero electroporation (IUE) (Nishiyama We introduced these constructs (Streptococcus pyogenes
et al., 2012; Tabata and Nakajima, 2001). We demonstrate that Cas9 or SpCas9 and sgRNA expressing vectors and ssODNs)
a tag sequence for a short epitope or a longer fluorescent protein together with hyperactive piggyBac transposase and piggyBac
can be rapidly and precisely inserted into an endogenous gene of transposon vectors expressing monomeric EGFP (mEGFP) as
interest in vivo. This method is scalable to many species of pro- a marker of transfection (Chen and LoTurco, 2012; Yusa et al.,
teins in diverse cell types, and permits high-resolution imaging 2011) to neuro-progenitor cells using IUE to target pyramidal
with light and electron microscopy both in fixed and live tissue. neurons in the cerebral cortex. The transposon system induces
Thus, SLENDR allows researchers to rapidly and precisely deter- genomic integration of transgenes, preventing the dilution of
mine the localization and dynamics of endogenous proteins with mEGFP during cell divisions. Following IUE at embryonic day
the resolution of micro- to nanometers in various cell types, re- 12 (E12), we performed immunostaining of brain slices at post-
gions, and ages of the brain, providing a powerful tool suitable natal days 14–48 (P14–P48) using anti-HA antibody together
for large-scale analysis on a broad spectrum of proteins. with anti-NeuN antibody to label neurons.
In the stained slices, HA signals were observed in a sparse
RESULTS subset of neurons, suggesting that HDR was successfully
induced in these cells. Immunofluorescence signal was localized
In Vivo Single-Cell Labeling of Endogenous Proteins by mostly in cytosol and excluded from the nucleus, consistent with
Homology-Directed Repair previously reported distribution of CaMKII (Lee et al., 2009). Most
In order to label a specific protein with a tag in non-dividing brain of HA-positive neurons were found in layer 2/3 (Figures 1B, 1D,
cells by HDR-mediated gene editing, it is necessary to introduce and 1E). Among neurons in transfected area in layer 2/3, about
HDR machinery into progenitor cells before their final cell divi- a half of the population was mEGFP-positive (mEGFP/NeuN:
sion. To test whether this strategy, termed SLENDR, can provide CaMKIIa, 48.2% ± 6.7%; CaMKIIb, 40.0% ± 2.9%). Among
the efficiency sufficient for imaging subcellular localization of a these mEGFP-positive cells, a small population of the cells
protein of interest, we first aimed to insert the human influenza were found to be HA-positive (HA/mEGFP: CaMKIIa, 7.5% ±
hemagglutinin (HA) tag into endogenous CaMKIIa and CaMKIIb, 1.2%; CaMKIIb, 4.5% ± 0.8%), providing a few percent of overall
two major subunits of the dodecameric Ca2+/CaM-dependent knockin efficiency (HA/NeuN: CaMKIIa, 3.4% ± 0.2%; CaMKIIb,
kinase II (CaMKII) necessary for neuronal plasticity (Kim et al., 1.8% ± 0.5%). We also found a smaller population of HA-positive
2015; Lee et al., 2009; Lisman et al., 2012). neurons in layers 4–6 (HA/mEGFP: CaMKIIa, 1.8% ± 0.1%;
We designed specific single-guide RNAs (sgRNAs) targeting CaMKIIb, 0.2% ± 0.1%) (Figures 1B, 1D, and 1E). When IUE
the vicinity of the stop and start codon of CaMKIIa and CaMKIIb, was performed at E15, a smaller population of layer 2/3 neurons
respectively. We also designed corresponding single-stranded were labeled with the HA tag (HA/mEGFP: CaMKIIa, 0.7% ±
oligodeoxynucleotides (ssODNs) (200 bases) to integrate the 0.1%; CaMKIIb, 0.7% ± 0.4%, HA/NeuN: CaMKIIa, 0.1% ±
HA tag sequence into the genome just upstream of the stop 0.0%; CaMKIIb, 0.2% ± 0.1%) (Figures 1B, 1D, and 1E; Tables
codon of CaMKIIa and downstream of the start codon of S1 and S2). Given that the superficial cortical layers (layer 2/3)
CaMKIIb (Figures 1A and 1C). To minimize the possibility of are populated with later-born neurons (Chen and LoTurco,
gene knockdown in cells where CRISPR-Cas9-mediated DNA 2012), these findings suggest that SLENDR is more efficient
Figure 1. In Vivo Single-Cell Labeling of Endogenous Proteins by Homology-Directed Repair

(A and C) Graphical representation of the mouse genomic loci of CaMKIIa (A) and CaMKIIb (C) showing the target sites for Cas9, sgRNA, and ssODNs. The sgRNA
targeting sequences are labeled in magenta. The protospacer-adjacent motif (PAM) sequences are labeled in green. The stop and start codons of CaMKIIa (A)
and CaMKIIb (C), respectively, are marked in orange. The Cas9 cleavage sites are indicated by the black arrowheads. PCR primer sets (control and recombi-
nation) for PCR genotyping (E and F) are indicated by the arrows.
(B and D) Confocal microscopic images of the cerebral cortex electroporated at E12 (top and middle) and E15 (bottom) showing the DAPI signal (blue) and
immunoreactivities for mEGFP (magenta) and the HA tag (green) fused to the C terminus of endogenous CaMKIIa (B) and N terminus of endogenous CaMKIIb (D).
Middle panels show negative control experiments in which the sgRNA for CaMKIIa was paired with the ssODNs for CaMKIIb (B) and vice versa (D).
(E) The efficiency of SLENDR for CaMKIIa and CaMKIIb (the ratio of the number of HA/mEGFP double-positive neurons to that of mEGFP-positive neurons).
CaMKIIa, E12 layer (L) 2/3, n = 545 neurons/3 mice; E12 L4–L6, n = 275/3; E15 L2/3, n = 713/3. CaMKIIb, E12 L2/3, n = 716/3; E12 L4–L6, n = 379/3; E15 L2/3,
n = 367/3. **p < 0.01, Dunnett’s test, in comparison with E12 L2/3.
(F and G) Left: PCR genotyping using genomic DNA extracted from the electroporated brain. Recombination primer sets (top: F, CaMKIIa-F1 and HA-R1;
G, HA-F1 and CaMKIIb-R1) and control primer sets (bottom: F, CaMKIIa-F1 and CaMKIIa-R1; G, CaMKIIb-F1 and CaMKIIb-R1) were used for PCR. Right: DNA
sequencing analysis of the PCR products for CaMKIIa-HA (F) and HA-CaMKIIb (G). The HA tag sequence is marked in orange.
Data are represented as mean ± SEM. Scale bars, 50 mm.
Cell 165, 1803–1817, June 16, 2016 1805

A CRISPR-Cas9 C CRISPR-Cas9 E CRISPR-Cas9
Target PAM Target PAM
5’...cgtgaccgagagagttagctgactttacataga...3’
tga 5’...tcttttgtgtcttgatagttcgccatggatgac...3’
atg 5’...aggttgctgtggttccaccaaaatatggaactt...3’
atg
3’...gcactggctctctcaatcgactgaaatgtatct...5’ 3’...agaaaacacagaactatcaagcggtacctactg...5’ 3’...tccaacgacaccaaggtggttttataccttgaa...5’
PAM Target
MeCP2 β-Actin DCX
200 84 27 89 (bases) 192 89 27 76 (bases) 198 86 27 85 (bases)

Oligo donor HA Oligo donor HA Oligo donor HA
B D F
MeCP2-HA MeCP2-HA DAPI HA- -Actin HA- -Actin GFAP DAPI HA-DCX mEGFP DAPI HA-DCX
P7 P28 P28 E14 E14
G CRISPR-Cas9 I CRISPR-Cas9 K CRISPR-Cas9

Target PAM Target PAM
5’...gcccctcagctcctcggccgcgccatgggcacc...3’
atg 5’...agcaacctgtagtcctcagggctggcgagacgc...3’
tag 5’...gaccctgccggagccgctgccgcaatggatgat...3’
atg
3’...cggggagtcgaggagccggcgcggtacccgtgg...5’ 3’...tcgttggacatcaggagtcccgaccgctctgcg...5’ 3’...ctgggacggcctcggcgacggcgttacctacta...5’
PAM Target
Rab11a CaV1.2 14-3-3ε

H J L
HA-Rab11a mEGFP HA-Rab11a CaV1.2-HA CaV1.2-HA mEGFP HA-14-3-3 HA-14-3-3 NeuN
P23 P23 P16 P0 P28
M CRISPR-Cas9 O CRISPR-Cas9 Q CRISPR-Cas9

Target PAM Target PAM Target PAM
5’...cccggcgcgaggacggacgagaagatggaggag...3’
atg 5’...cggagccgcagcaccgacgaccagatggagctg...3’
atg 5’...ggcaagaggtggttggggggaaccatggctgac...3’
atg
3’...gggccgcgctcctgcctgctcttctacctcctc...5’ 3’...gcctcggcgtcgtggctgctggtctacctcgac...5’ 3’...ccgttctccaccaaccccccttggtaccgactg...5’
Fmrp Arc PKCα

N P R
HA-FMRP HA-FMRP HA-Arc DAPI HA-PKCα DAPI
DAPI P7 P29 P9
P7
Figure 2. SLENDR Is Scalable to Various Endogenous Proteins

(A, C, E, G, I, K, M, O, and Q) Graphical representation of the mouse genomic loci of MeCP2 (A), b-Actin (C), DCX (E), Rab11a (G), CaV1.2 (I), 14-3-3ε (K), Fmrp (M),
Arc (O), and PKCa (Q) showing the target sites for Cas9, sgRNA, and ssODNs. The sgRNA targeting sequences are labeled in magenta. The PAM sequences are
labeled in green. The stop (A and I) and start (C, E, G, K, M, O, and Q) codons are marked in orange. The Cas9 cleavage sites are indicated by the black
arrowheads.
(B, F, H, J, N, P, and R) Confocal microscopic images of the cerebral cortex showing the DAPI signal (blue) and immunoreactivities for mEGFP (magenta) and the
HA tag (green) fused to the C terminus of MeCP2 (B) and CaV1.2 (J) and the N terminus of DCX (F), Rab11a (H), FMRP (N), Arc (P), and PKCa (R). The dashed line
represents the shape of the dendrite (H and N).
(D) Images of the cerebral cortex showing the DAPI signal (blue) and immunoreactivities for GFAP (magenta, an astrocyte marker) and the HA tag (green) fused to
the N terminus of b-Actin.
1806 Cell 165, 1803–1817, June 16, 2016

when IUE is performed several days before the final neurogenic sion of the targeted gene in these cells. Indeed, the intensities
divisions of targeted cells. of endogenous MeCP2 detected by an antibody against the
To test the specificity of the construct, we used incorrect C terminus of MeCP2 was similar between mEGFP-positive
ssODNs-sgRNA pairs, like CaMKIIa ssODNs-CaMKIIb sgRNA and -negative cells (Figure S4D; p = 0.19, Student’s t test). This
or vice versa. Under these conditions, no fluorescence signal indicated that the endogenous expression of MeCP2 was not
was detected in brain slices stained with anti-HA antibody (Fig- significantly disturbed by the NHEJ-mediated repair.
ures 1B and 1D). These results indicate that the genome editing Next, we applied SLENDR to insert the HA tag to endogenous
is specific to the sequence of the sgRNA. We further confirmed b-Actin, a major cytoskeletal protein in dendritic spines. As ex-
that the expected genome editing occurred at the DNA level by pected, immunostaining using the anti-HA antibody showed
performing PCR amplification of the targeted locus. Electropho- that HA-b-Actin was highly accumulated in dendritic spines in
retic analysis revealed the presence of recombined PCR product layer 2/3 pyramidal neurons and microfilaments in astrocytes
at a size consistent with the recombined allele in brains trans- in the cortex at P28 (Yuste and Bonhoeffer, 2004) (Figures 2C
fected with the constructs necessary for HDR (Figures 1F and and 2D). In addition, the high contrast images demonstrate
1G). In contrast, neither untransfected control brains nor brains that SLENDR enables the precise evaluation of the number
transfected with incorrect ssODNs-sgRNA pairs showed the and morphology of dendritic spines without the potential
corresponding PCR amplification (Figures 1F and 1G). Further- morphological phenotypes caused by overexpression of fluoro-
more, DNA sequencing of the amplified PCR products indicated phore tagged-actin or actin binding proteins (Riedl et al., 2008).
that the HA tag sequence was integrated as expected (Figures To evaluate the timescale of HDR after introducing SLENDR
1F and 1G). In addition, sgRNAs targeting different sequences constructs, we targeted doublecortin (DCX), a protein that is
in the respective genes produced a similar pattern of HA staining expressed early in brain development. DCX is a microtubule-
(Figure S1), demonstrating flexibility of construct design. Thus, associated protein expressed in postmitotic migrating and differ-
SLENDR enables specific single-cell labeling of endogenous entiating neurons in the developing brain (Gleeson et al., 1999).
proteins by either N- or C-terminal epitope tagging. Intriguingly, HA-DCX was detected in the cortical migrating neu-
rons and accumulated in the growth cone as early as 60 hr after
SLENDR Is Scalable to Various Endogenous Proteins IUE at E12 (Figures 2E and 2F), suggesting that HDR occurred
To validate the applicability of SLENDR to visualize the subcellu- rapidly, possibly within 1 or 2 days after IUE. HA-DCX was de-
lar localization of a broad spectrum of endogenous proteins in tected at both E14 and E18 (Figures 2F and S5B), suggesting
brain tissue, we performed SLENDR to fuse the HA tag to either SLENDR is suitable for studying protein localization in the em-
the N or C terminus of a variety of proteins including nuclear, bryonic brain. In addition, we evaluated the effect of NHEJ on
cytoskeletal, vesicular, cytosolic, and membrane proteins (Fig- the expression of DCX by immunostaining using an antibody
ures 2A–2R, S2, S3, S4, and S5; Table S1). against the C terminus of the protein. The images showed that
We first fused the HA tag to MeCP2, a chromatin-associated 98.7% of the mEGFP-positive neurons exhibited DCX expres-
protein that regulates gene transcription (Chen et al., 2001). sion (n = 151 cells) (Figure S5A). Together with the negligible
MeCP2 was selected for the following reasons. First, the NHEJ effects on MeCP2 expression (Figure S4D), these data
endogenous subcellular localization can be imaged by traditional suggest that our strategy of targeting the Cas9-mediated cleav-
immunostaining approaches using a specific, characterized anti- age at 50 -UTRs minimizes the effect of NHEJ on the expression of
body. Second, MeCP2 is known to accumulate in heterochro- the target gene for N-terminal tagging.
matin in the nucleus, allowing for the contrast needed for staining We further applied SLENDR to a variety of endogenous pro-
of the endogenous protein in tissue. Therefore, double-immuno- teins with distinct subcellular localizations. Rab11a, a small
staining for the SLENDR inserted HA tag and endogenous GTPase involved in the endosomal recycling of proteins, was
MeCP2 enabled further validation of the SLENDR approach. As localized to numerous, small dispersed vesicles throughout the
expected, we found that MeCP2-HA and HA-MeCP2 were soma and dendrites in cortical neurons and astrocytes at P23,
distributed exclusively in the nucleus at P7 (Figures 2A, 2B, consistent with the pattern expected from the localization of
S4B, and S4C). In addition, MeCP2-HA was accumulated in the recycling endosomes (Hutagalung and Novick, 2011) (Figures
heterochromatin regions (labeled with DAPI) (Figure 2B) and 2G, 2H, and S5C). CaV1.2, the a1C subunit of the L-type
well co-localized with endogenous MeCP2 (Figure S4A), confirm- voltage-gated calcium channel, was distributed in clusters on
ing the specificity of SLENDR. We also evaluated potential effects cell bodies and proximal dendrites in layer 2/3 pyramidal neu-
of SLENDR in transfected cells that did not undergo HDR. Since rons at P16 (Figures 2I and 2J) (Hell et al., 1993). Immunofluores-
many mEGFP-positive cells were expected to undergo the error- cence signal of CaV1.2-HA was also detected in the nucleus. This
prone, NHEJ-mediated repair of Cas9-induced DNA double- signal may represent the C-terminal fragment of CaV1.2, which
strand breaks, we designed the targeting sgRNA to place the functions as a calcium channel-associated transcription regu-
cleavage site in the 50 -UTR of MeCP2 gene (Figure S4B). This lator (Gomez-Ospina et al., 2006). 14-3-3ε, a signaling protein
strategy was designed to minimize potential effects on expres- involved in neuronal migration and synaptic plasticity, was
(L) Images of the cerebral cortex at P0 and P28 showing immunoreactivities for NeuN (blue, a neuron marker) and the HA tag (green) fused to the N terminus of
14-3-3ε.
Scale bars, 5 mm (B, D, F, right, H, J, N, P, and R); 50 mm (F, left; L).
See also Figures S2, S3, S4, and S5 and Tables S1 and S3.
Cell 165, 1803–1817, June 16, 2016 1807

diffusely distributed in the cytoplasm both at P0 and P28 in neu- control, 2.6 ± 2.6 particles/mm3; p < 0.001, Student’s t test).
rons and astrocytes (Toyo-oka et al., 2014) (Figures 2K and 2L). These experiments demonstrate that SLENDR is useful for nano-
FMRP, a polyribosome-associated RNA-binding protein that meter scale localization of endogenous proteins.
regulates translation of a large number of mRNAs (Contractor
et al., 2015), was found in puncta, which likely reflect FMRP- SLENDR Is Scalable to Various Cell Types in Various
associated mRNA granules, in the soma as well as dendrites in Brain Regions
pyramidal neurons at P7 and P28 (Figures 2M, 2N, and S5D). We then tested if SLENDR can be applied to different cell types in
Arc, an immediate early gene product involved in synaptic plas- different brain regions by targeting distinct progenitor cells pre-
ticity, was localized both in the nucleus and cytoplasm at P29 sent at different locations and timings in the developing brain
(Figures 2O and 2P) (Korb et al., 2013; Shepherd and Bear, (Borrell et al., 2005; Chen and LoTurco, 2012; Kitazawa et al.,
2011). Finally, we found that a isoform of protein kinase C 2014; Nishiyama et al., 2012; Soma et al., 2009). By adjusting
(PKC), a member of a family of serine/threonine kinases impli- the timing and electroporation angle (Figure 4A), various HA-
cated in a wide range of cellular functions (Steinberg, 2008), tagged endogenous proteins (b-Actin, CaMKIIa, CaMKIIb, and
was distributed mostly on the plasma membrane of the soma MeCP2) were observed in various cell types in widespread brain
and dendrites at P9 (Figures 2Q and 2R). Interestingly, PKCa regions (Table S2), including CA1 pyramidal neurons and dentate
was less accumulated on the plasma membrane at P27, sug- granule cells in the hippocampus (Figures 4B and 4H), spiny stel-
gesting that endogenous PKCa may be more active at the devel- late cells in the subiculum (Figures 4C and 4H), granule cells in
oping stage (Figures S5E–S5G). Thus, SLENDR enables us to the olfactory bulb (Figure 4D), medium spiny neurons in the stria-
determine the distribution pattern of proteins that has been tum (Figure 4E), basolateral amygdala neurons (Figure 4F), and
undefined or controversial by conventional methods. granule and Purkinje cells in the cerebellum (Figures 4G and
Notably, we never observed HA-positive cells when we used 4H). HA-CaMKIIb in parallel fibers originating from cerebellar
incorrect ssODNs-sgRNA pairs (Figures S2 and S4). In addition, granule cells was detected along the dendrites of Purkinje cells
precise genome editing was confirmed by PCR amplification of in the molecular layer (Figure 4G).
the targeted loci followed by DNA sequencing of the PCR prod- As we showed that staining of endogenous HA-b-Actin en-
ucts (Figure S3). Furthermore, for each target protein, all HA- ables clear visualization of dendritic spine morphology in neu-
positive cells in the same region showed similar HA-staining rons (Figure 2D), we imaged HA-b-Actin in various cell types in
pattern. These results collectively demonstrate that SLENDR the brain to compare the structure of dendritic spines. Morpho-
enables us to label endogenous proteins specifically. Taken logical diversity of dendritic spines existed among different cell
together, SLENDR is a highly generalizable technique that can types including dentate granule cells (Figure 4H) and CA1 pyra-
be applied to various species of proteins in the brain from embry- midal neurons (Figure 4B) in the hippocampus, layer 2/3 pyrami-
onic to adult stages for high-quality mapping of subcellular dal neurons in the cortex (Figure 2D), spiny stellate cells in the
localization. subiculum (Figure 4H), and cerebellar Purkinje cells (Figure 4H).
Thus, SLENDR enables comparison of subcellular localization of
Nanometer-Resolution Analysis by SLENDR endogenous proteins in various neuron subtypes and brain re-
Immunoelectron microscopy allows nanoscale visualization of gions across developmental stages, providing regional and
endogenous proteins with defined ultrastructures in cells. How- developmental specific information about protein localization.
ever, the lack of reliable antibodies compatible with electron mi-
croscopy imaging limits its application to a variety of proteins. Labeling Multiple Endogenous Proteins by SLENDR
Thus, we tested whether SLENDR could be applied to ultrastruc- Multiplex labeling of different endogenous proteins with different
tural imaging of endogenous proteins using immunoelectron tags would enable high-resolution co-localization assays. Since
microscopy. To do so, we used cortical tissue in which endoge- CRISPR-Cas9 is able to target multiple genes simultaneously
nous CaMKIIb is fused with HA using SLENDR (Figures 1C and (Cong et al., 2013; Heidenreich and Zhang, 2016; Swiech et al.,
1D). We applied pre-embedding staining technique: we incu- 2015), we tested the ability of SLENDR to target two different
bated the tissue with HA-antibody and secondary antibody con- proteins in the same cell (Figure 5A). We performed simultaneous
jugated with gold, followed by silver enhancement and tissue labeling of CaMKIIa and CaMKIIb with the HA and FLAG tag,
embedding. We then prepared serial thin-section (50 nm) of respectively, by co-introducing SLENDR constructs for HA-
the tissue using the automatic tape-collecting ultramicrotome CaMKIIa and FLAG-CaMKIIb into the developing cortex at E13
(ATUMtome). ATUMtome permits rapid and automated cutting by IUE (Figure 5A). We performed immunostaining of brain slices
and collection of serial thin sections onto a continuous reel of at P14 using anti-HA and anti-FLAG antibodies. HA-CaMKIIa
tape (Kasthuri et al., 2015). We imaged a number of serial and FLAG-CaMKIIb were detected in a number of layer 2/3
sections on the tape by scanning electron microscopy and re- neurons (HA-positive/mEGFP-positive, 4.5% ± 0.4%; FLAG-
constructed three-dimensional images of SLENDR-generated positive/mEGFP-positive, 4.9% ± 0.6%; n = 577 cells), and a
knockin cells (Figures 3A–3D). We found that CaMKIIb was local- significant fraction of neurons exhibited both HA and FLAG
ized near the postsynaptic density (PSD) (mode 80 nm) in signals (HA and FLAG double-positive/mEGFP-positive, 0.7% ±
dendritic spines (Figure 3F). Finally, the specificity of the immu- 0.1%; n = 577 cells) (Figure 5B). This double labeling efficiency
nogold labeling was examined by comparing the labeling density was higher than the simple multiplication of each labeling effi-
in HA-positive and surrounding neurons in the same specimen ciency (4.5% 3 4.9% = 0.2%), demonstrating the practicality
(Figure 3E; HA-positive, 241.3 ± 34.5 particles/mm3; surrounding of the method for co-localization assay. Thus, SLENDR allows
1808 Cell 165, 1803–1817, June 16, 2016

A B
HA-CaMKII
Spines
Surrounding control
C D E
400 ***
Gold particles / m3
300
HA-CaMKII
200
100
0
ve g
siti din
Dendrites
Po r o un ntrol
S ur co
F 25
Gold particles (%)
Surrounding control
20
15
10
5
0
100 200
Distance from cleft surface
(nm)
Figure 3. Nanometer-Scale Analysis of Endogenous Proteins by SLENDR

(A–D) Electron microscopic images of dendritic spines (A) and shafts (C) in the cerebral cortex showing immunogold labeling for the HA tag (arrowheads) fused to
the N terminus of CaMKIIb in HA-positive (top) and surrounding control (bottom) cells. Three-dimensional reconstructions of corresponding spines (B) and
dendrites (D). Analyzed spines and dendrites are marked in purple and presynaptic terminals are marked in green. Presynaptic vesicles and postsynaptic
densities are marked in yellow and red, respectively (B and D).
(E) Density of immunogold particles in HA-positive spines (n = 6) and surrounding control spines (n = 8). ***p < 0.001, Student’s t test.
(F) The frequency distribution of distances between individual gold particles and cleft surfaces (n = 92 particles/6 spines).
Data are represented as mean ± SEM. Scale bars, 500 nm.
See also Table S3.
Cell 165, 1803–1817, June 16, 2016 1809

A (a) Hippocampus (b) Olfactory bulb (c) Striatum (d) Cerebellum
Subiculum Amygdala (E10:PC)
(E13) (E12) (E11-12) (E13:GC)
+
+ - (c)
(a)
-
-
+
LV LV LV
IV +
(b)
+
(d) -
- + + - + - -
B Hippocampus C Subiculum
HA- -Actin DAPI HA- -Actin mEGFP HA- -Actin DAPI
CA1
DG
CA3
D Olfactory bulb E Striatum
CaMKIIα-HA DAPI GL MeCP2-HA DAPI CaMKIIα-HA NeuN CaMKIIα-HA
EPL
MCL
GCL
F Amygdala G Cerebellum
CaMKIIα-HA NeuN HA-CaMKII DAPI HA-CaMKII Calbindin
BLA
Ctx ML
HA-CaMKII PCL
Calbindin GCL
H
HA- -Actin mEGFP
Dentate granule cell in the hippocampus Spiny stellate cell in the subiculum Cerebellar Purkinj
e cell
Figure 4. SLENDR Is Scalable to Various Cell Types in Various Brain Regions

(A) Schematic illustration of IUE for targeting distinct brain regions. The relative position of the electrodes (+, positive pole; , negative pole) are shown in the
transverse section of the brain to target different brain areas (left). The position of electrode paddles and the injected DNA (green) are shown in the coronal section
of the brain to target hippocampus and subiculum (a), olfactory bulb (b), striatum and amygdala (c), and cerebellum (d). LV, lateral ventricle; IV, fourth ventricle; PC,
Purkinje cell; GC, granule cell.
(B) Confocal microscopic images of the hippocampus showing the DAPI signal (blue) and immunoreactivities for mEGFP (magenta) and the HA tag (green) fused
to b-Actin. DG, dentate gyrus.
1810 Cell 165, 1803–1817, June 16, 2016

A Multiplex Protein Labeling B
HA FLAG
knock-in knock-in
(HDR) (HDR)
HA FLAG
CaMKIIα CaMKIIβ HA-CaMKIIα
HA-CaMKIIα FLAG-CaMKIIβ FLAG-CaMKIIβ
DAPI DAPI DAPI
C SLENDR with Single-Cell Knockout D E

GL
HA-β-Actin MeCP2 MeCP2 positive
Number of spines / μm
HA Gene HA 3
knock-in knockout knock-in *
(HDR) (NHEJ) (HDR)
2
MeCP2 negative 1
HA HA
β-Actin MeCP2
eC
C β-Actin MeCP2
0
(+) (-)
CA1 MeCP2
Figure 5. Application of SLENDR by Multiplex Genome Editing

(A) Schematics of multiplex labeling of different endogenous proteins with different tags. The HA and FLAG sequences are inserted to CaMKIIa and CaMKIIb,
respectively, in the same cell through HDR-mediated genome editing.
(B) Multiplex labeling of endogenous CaMKIIa and CaMKIIb. Confocal microscopic images of the cerebral cortex at P14 showing the DAPI signal (blue) and
immunoreactivities for the HA tag (green) and the FLAG tag (magenta) fused to the N terminus of endogenous CaMKIIa and CaMKIIb, respectively. The yellow
arrows indicate HA and FLAG double-positive layer 2/3 neurons.
(C) Schematics of combining SLENDR with NHEJ-mediated gene knockout. The HA sequence is inserted to b-Actin through HDR-mediated genome editing and
a frame-shift mutation is induced in MeCP2 through NHEJ-mediated genome editing in the same cell (left). In our strategy, some cells in the same tissue undergo
only HDR-mediated genome editing (right), allowing comparison of the expression and localization of endogenous proteins within the same brain slice.
(D) Immunofluorescence images of MeCP2 (magenta) and HA-b-Actin (green) in the hippocampus CA1 region. Magenta arrowhead, MeCP2-positive; white
arrowhead, MeCP2-negative. Representative images of secondary dendrites of MeCP2-positive and -negative neurons (right).
(E) The averaged density of spines on secondary or tertiary apical dendrites in MeCP2-positive (n = 527 spines/6 neurons) and -negative (n = 437/5) neurons.
*p < 0.05, Student’s t test.
Data are represented as mean ± SEM. Scale bars, 50 mm (B and D, left); 5 mm (D, right).
See also Table S3.
labeling of two different species of proteins with different tags in gene function and to compare normal and knockout cells in
single cells, providing a valuable tool for co-localization assays the same tissue (Zong et al., 2005). Taking advantage of the
of a pair of endogenous proteins. multiplexity of CRISPR-Cas9, we simultaneously introduced
SLENDR constructs to insert the HA tag to b-Actin and CRISPR
SLENDR in Combination with Single-Cell Knockout constructs to induce NHEJ-mediated gene knockout of MeCP2
The ability to examine endogenous subcellular protein localiza- in progenitors of hippocampal neurons at E13 (Figure 5C) (In-
tion in the context of a knockout of a different protein would contro et al., 2014; Straub et al., 2014; Swiech et al., 2015).
provide functional insight into the interaction between the visu- Visualization of HA-tagged b-Actin by SLENDR enables visual-
alized and deleted proteins. In this regard, combining SLENDR ization of dendritic spine morphology, providing a useful tool to
with NHEJ-based single-cell knockout would be of particular study effects of MeCP2 gene deletion on dendritic spines in
interest, since it would allow for the study of cell-autonomous single cells.
(C–F) Images of the subiculum (C), olfactory bulb (D), striatum (E), and amygdala (F) showing the DAPI signal (C–E, blue) and immunoreactivities for NeuN (E and F,
blue) and the HA tag (green) fused to b-Actin (C), CaMKIIa (D to F), and MeCP2 (E). GL, glomerular layer; EPL, external plexiform layer; MCL, mitral cell layer; GCL,
granule cell layer; BLA, basolateral amygdala; Ctx, cerebral cortex.
(G) Images of the cerebellum showing the DAPI signal (blue) and immunoreactivities for calbindin D-28k (magenta, a Purkinje cell marker) and the HA tag (green)
fused to CaMKIIb. ML, molecular layer; PCL, Purkinje cell layer; GCL, granule cell layer.
(H) Images of the dendrites of a dentate granule cell in the hippocampus, a spiny stellate cell in the subiculum and a Purkinje cell in the cerebellum showing
immunoreactivities for mEGFP (magenta) and the HA tag (green) fused to b-Actin.
Scale bars, 50 mm (B–G); 5 mm (B, right; E, inset; G, inset and right); 1 mm (H).
See also Tables S2 and S3.
Cell 165, 1803–1817, June 16, 2016 1811

A C-terminal labeling (mEGFP) C N-terminal labeling (mEGFP)
CRISPR-Cas9 CRISPR-Cas9
Target PAM
5’...tcctgccgcagtaaggattctttcttcctgaat...3’
taa 5’...acgccgagcccgatcgccaccgccatggccacc...3’
atg
3’...aggacggcgtcattcctaagaaagaaggactta...5’ 3’...tgcggctcgggctagcggtggcggtaccggtgg...5’
PAM Target
CaMKIIα CaMKIIβ
Plasmid-based 818 723 941 (bases) Plasmid-based 930 714 696 (bases)
donor (2.5 kb) mEGFP donor (2.3 kb) mEGFP
Homologous recombination Homologous recombination
CaMKIIα CaMKIIβ
mEGFP mEGFP
B CaMKIIα-mEGFP CaMKIIα-mEGFP DsRed2 D mEGFP-CaMKIIβ mEGFP-CaMKIIβ DsRed2
E CaMKIIα (knock-in) CaMKIIα (overexpression) F 4 *** **
Spine/dendrite ratio
3
2
CaMKIIβ (knock-in) CaMKIIβ (overexpression)
1
0
knock-in OE knock-in OE
CaMKIIα CaMKIIβ
Uncaging
CaMKIIα intensity change (%)
CaMKIIα intensity change (%)
G CaMKIIα knock-in H 250 I 100

Stimulated spine ***
200 Nearby spine 80
Before 30 min Control spine
150 (No stimulation)
60
100
40
50
20
uncaging 0
0
ted
0 5 10 15 20 25 30
arb
ol
ula
-20
ntr
Time (min)
Ne
Co
m
Sti
200 Uncaging
CaMKIIβ intensity change (%)
CaMKIIβ intensity change (%)
J CaMKIIβ knock-in K L 100 ***

Stimulated spine
Nearby spine
Before 30 min 150 Control spine 80
(No stimulation)
100 60
uncaging
40
50
20
0
0
d
0 5 10 15 20 25 30
e
lat
ol
Time (min) -20

arb
mu
ntr
Ne
Co
Sti
1812 Cell 165, 1803–1817, June 16, 2016

While knockout efficiency of NHEJ-mediated genome editing structs together with the transposon vectors expressing DsRed2
has been reported to be very high in neurons (70%–100%) were introduced to progenitors of layer 2/3 neurons through IUE
(Heidenreich and Zhang, 2016; Incontro et al., 2014; Straub at E12. At P14–P28, we found that a sparse subset of layer 2/3
et al., 2014; Swiech et al., 2015), immunohistochemistry with neurons exhibited mEGFP fluorescence fused to endogenous
anti-MeCP2 antibody showed that MeCP2 was absent in CaMKIIa and CaMKIIb (Figures 6B and 6D). The efficiency of
48.8% of mEGFP-positive CA1 pyramidal neurons at P28. This mEGFP knockin was lower than that of HA knockin (mEGFP-pos-
reduced efficiency was probably due to the dilution of the itive/DsRed2-positive, <1%), consistent with previous studies
knockout constructs during cell divisions (Chen and LoTurco, showing that a long sequence insertion through HDR is less effi-
2012; Yusa et al., 2011). As expected, both MeCP2-negative cient (Cox et al., 2015). The specificity of mEGFP knockin was
and positive neurons were observed in HA-b-Actin-positive confirmed by control experiments using correct and incorrect
CA1 neurons in the same slices (Figure 5D). Notably, the template-sgRNA pairs and DNA sequencing following PCR
knockout efficiency in HA-b-Actin-positive neurons (24.1%) amplification of the targeted locus (Figure S6).
was substantially lower than that in whole population (48.8%). To further verify the benefit of imaging endogenous proteins,
This is presumably because the knockout constructs are diluted we compared the distribution pattern of endogenous and over-
in a population of cells that divides extensively, and this popula- expressed mEGFP-tagged CaMKII in layer 2/3 cortical neurons
tion of cells likely provides more efficient HDR. A loss of function at P14. Since CaMKIIb has an actin-binding domain (Lisman
mutation in MeCP2, which leads to Rett syndrome in humans, et al., 2012), overexpressed CaMKIIb tends to accumulate in
has been reported to impair maturation of dendritic spines spines to a greater extent than overexpressed CaMKIIa. Notably,
(Chen et al., 2001; Swiech et al., 2015). Therefore, we compared the degree of enrichment of endogenous CaMKIIa and CaMKIIb
the density of dendritic spines between MeCP2-negative and in spines, as measured by the spine/dendrite ratio of the peak in-
-positive neurons (Figures 5D and 5E). Spine density in the apical tensity, was significantly higher and lower than that of overex-
dendrite of CA1 neurons was comparable in MeCP2-positive pressed CaMKIIa and CaMKIIb, respectively (Figures 6E and
neurons to that reported in these neurons in a previous electron 6F; CaMKIIa, p < 0.001; CaMKIIb, p = 0.004, Student’s t test).
microscopic study (Harris et al., 1992). Importantly, spine density This is perhaps because overexpressed subunits tend to form
was significantly lower in MeCP2-negative neurons, consistent homomeric enzymes, while endogenous CaMKIIa and CaMKIIb
with previous studies (Figure 5E; p = 0.046, Student’s t test) tend to form heteromers (Lisman et al., 2012). Thus, the localiza-
(Swiech et al., 2015). tion of endogenous (heteromeric) subunits would be enriched in
Although high-quality antibodies against endogenous proteins spines to an extent that is between homomeric CaMKIIa and
are required to distinguish knockout and control cells, our strat- CaMKIIb.
egy combining SLENDR with NHEJ-mediated gene knockout We next performed live imaging to monitor the synapse-spe-
provides a useful means to investigate cell-autonomous effects cific translocation of CaMKII after the stimulation of a single den-
of a gene of interest on endogenous proteins. dritic spine (Bosch et al., 2014; Lee et al., 2009; Lisman et al.,
2012; Nishiyama and Yasuda, 2015). We prepared organotypic
Live Imaging of Endogenous Proteins by SLENDR cortical slice cultures from mice in which endogenous CaMKIIa
Labeling endogenous proteins with fluorescent proteins would or CaMKIIb was labeled with mEGFP by SLENDR. When we
allow us to image protein dynamics in live cells without overex- applied a train of two-photon glutamate uncaging pulses (1 Hz,
pression artifacts, further expanding the applicability of SLENDR. 30 s) at single spines, mEGFP-tagged CaMKIIa and CaMKIIb
Thus, we tested if SLENDR can be used to insert a long sequence were rapidly accumulated in the stimulated spines and remained
encoding a fluorescent protein into a gene of interest. For this, we for more than 30 min. The translocation was highly restricted to
targeted mEGFP to the sequence just upstream of the stop the stimulated spine, with no significant increase in fluorescence
codon of CaMKIIa or downstream of the start codon of CaMKIIb intensity in the surrounding spines (Figures 6G–6L; CaMKIIa,
in the genome (Figures 6A and 6C). We used a plasmid-based p < 0.001; CaMKIIb, p < 0.001; Dunnett’s test). These results
template (2.3–2.5 kb) containing the 0.7 kb sequence of demonstrate that SLENDR enables monitoring of the dynamics
mEGFP (see Supplemental Information). The SLENDR con- of endogenous proteins in living tissue.
Figure 6. Localization and Dynamics of Endogenous Proteins Labeled with mEGFP by SLENDR
(A and C) Graphical representation of the mouse genomic loci of CaMKIIa (A) and CaMKIIb (C) showing the targeting sites for Cas9, sgRNA, and HDR donor
plasmid. The sgRNA targeting sequences are labeled in magenta. The PAM sequences are labeled in green. The stop and start codons of CaMKIIa (A) and
CaMKIIb (C) are marked in orange.
(B and D) Confocal microscopic images of the somatosensory cortex showing the fluorescence of DsRed2 (magenta) and mEGFP (green) fused to CaMKIIa/b.
(E) Images of apical secondary dendrites in layer 2/3 fixed at P14 showing mEGFP-tagged endogenous (knockin) or overexpressed (OE) CaMKIIa/b.
(F) The spine/dendrite ratio of the peak intensities of mEGFP-tagged CaMKIIa/b. CaMKIIa, knockin, n = 51/6 (spines/neurons); OE, n = 40/5. CaMKIIb, knockin,
n = 58/4; OE, n = 83/5.
(G–L) Two-photon microscopic images before and 30 min after glutamate uncaging showing mEGFP-tagged endogenous CaMKIIa/b in layer 2/3 neurons (G and
J). Red circles, uncaging spots; red arrowheads, stimulated spines. Averaged time courses (H and K) and sustained values (I and L; averaged over 20–30 min) of
CaMKIIa/b intensity change in the stimulated (red; H and I, n = 11/8; K and L, n = 9/7), nearby (2–5 mm from the stimulated spines; blue; H and I, n = 31/8; K and L,
n = 31/7) and control spines with no stimulation (gray; H and I, n = 23/5; K and L, n = 13/2). ***p < 0.001, Student’s t test (F) and Dunnett’s test (I and L).
Data are represented as mean ± SEM. Scale bars, 50 mm (B and D); 5 mm (E); 2 mm (G and J).
See also Figure S6 and Table S3.
Cell 165, 1803–1817, June 16, 2016 1813

DISCUSSION cell to cell. In addition, although off-target cleavage by Cas9
may be induced in the genome, HDR is unlikely to occur at the
In this study, we have developed SLENDR, which allows in vivo off-target sites, because homologous recombination is known
genome editing in the mammalian brain for single-cell labeling of to be a highly sequence-specific event. The specificity of
endogenous proteins. We demonstrated that SLENDR is a sim- SLENDR could be further enhanced by using recently reported
ple and efficient technique to rapidly determine the subcellular Cas9 variants with minimum or no off-target effects while retain-
localization of endogenous proteins with the resolution of micro- ing comparable on-target cleavage activity (Kleinstiver et al.,
to nanometers in brain tissue. Importantly, the technique is 2016; Slaymaker et al., 2016).
generalizable to a broad spectrum of proteins and various cell One of the merits of SLENDR is that high-quality antibodies
types in widespread brain regions. SLENDR also can be used can be used for the detection of tags. In addition, once staining
for multiplex labeling of different proteins or for mosaic analysis conditions have been optimized for a tag, SLENDR can be
by combining labeling with single-cell knockout. Furthermore, applied to image various tagged proteins without extensive opti-
SLENDR is capable of inserting a long sequence such as that en- mization. Furthermore, since SLENDR allows protein labeling in
coding mEGFP in vivo and thus enables live imaging of endoge- a sparse subset of cells in the tissue, the specificity of immuno-
nous proteins during biological processes in the brain. staining can be easily validated by examining surrounding nega-
HDR-mediated genome editing has been a challenge in the tive control cells in the same specimen. These features are
brain due to the lack of HDR activity in postmitotic neurons particularly advantageous for immunoelectron microscopy im-
and the inefficient delivery of HDR machinery into the brain. aging, which often requires extensive optimization of staining
This has limited its application in the field of neuroscience (Hei- conditions and good control samples (e.g., knockout mice).
denreich and Zhang, 2016). We circumvented these problems We have demonstrated that introducing a single epitope tag
by targeting mitotic progenitors in the embryonic brain using by SLENDR is sufficient to detect relatively low abundant pro-
IUE. In our experiments, the knockin efficiency for the HA tag teins such as CaV1.2 and Arc (Figures 2J and 2P). The sensitivity
insertion was sufficient to image the subcellular localization of could be further increased by inserting multiple copies of
a protein of interest in single cells in brain tissue. We also showed epitopes or more antigenic probes such as Spaghetti-monster
that SLENDR can be used to knockin a long mEGFP sequence, (Viswanathan et al., 2015). Overall, because many studies are
albeit at lower efficiency. To extend the SLENDR technique to constrained by the lack of high-quality antibodies against pro-
broader applications for multiplexed labeling and fluorescent teins of interest, SLENDR will provide a generalizable and reli-
protein fusion, it may be necessary to increase the efficiency able platform for exploring localization of uncharacterized
of HDR. endogenous proteins using light and electron microscopy.
It is known that the efficiency of HDR highly depends on the Multiplexing CRISPR-Cas9-mediated processes provided
cell cycle and is limited by the competition with NHEJ (Chu some insight into how HDR and NHEJ occur following double
et al., 2015; Lin et al., 2014; Maruyama et al., 2015). Consistent strand breaks. We found that the cells labeled with a tag with
with this, we found that SLENDR was efficient in a limited time SLENDR, or in other words the cells that underwent HDR, showed
window when neuronal progenitors were still dividing. CRISPR- less NHEJ-mediated knockout (Figures 5C–5E). This result sug-
Cas9-mediated HDR occurred within a few days after IUE (Fig- gests that HDR occurs more frequently in a population of highly
ure 2F) and the HDR efficiency was significantly reduced when dividing cells. Consistent with this, the efficiency of double-label-
IUE was performed near the stage of the final cell divisions of ing via multiple HDRs was higher than that obtained by the simple
target cells (Figures 1B and 1D). In this regard, the direct delivery multiplication of each labeling efficiency (Figures 5A and 5B). The
of pre-assembled Cas9 protein-guide RNA ribonucleoprotein high efficiency of double-labeling by SLENDR opens the possibil-
complexes (RNPs), rather than expressing these components ity of co-localization assay for a pair of endogenous proteins with
from plasmids, might increase the time window for SLENDR, high resolution and contrast. Given relatively consistent knockin
as RNPs have recently been reported to provide rapid nuclease efficiency (Table S1), the SLENDR-based double labeling tech-
action with high efficiency and low off-target effects for genome nique would be applicable to many pairs of proteins.
editing (Kim et al., 2014; Lin et al., 2014). Alternatively, genetic or There are a few limitations in SLENDR. First, the technique
pharmacological inhibition of the NHEJ pathway may increase might lead to unintended sequence changes through insertion/
the HDR efficiency (Chu et al., 2015; Maruyama et al., 2015), deletion mutations mediated by NHEJ in the transfected cells.
although potential side effects must be addressed. This problem may cause a large impact particularly in the case
We validated the specificity of the SLENDR-mediated of N-terminal tagging, since altered sequence near the N-termi-
sequence insertion in the genome in several ways. First, we de- nal region could change the transcription and translation of the
signed sgRNAs by unbiased genome-wide analysis to minimize target gene. Therefore, throughout this study, we attempted to
the potential off-target cleavages by Cas9 (Ran et al., 2013). Sec- minimize the effect of on-target NHEJ by choosing target se-
ond, we performed control experiments by using incorrect quences at 50 -UTR or near the stop codon for N- or C-terminal
sgRNAs and found no cells expressing the HA tag. Third, we tagging, respectively. As expected from this design, the expres-
used distinct sgRNAs targeting the identical gene and observed sion of DCX was detected in 98.7% of the mEGFP-positive
similar HA staining patterns. Fourth, the localization of HA neurons following SLENDR for the N-terminal tagging of DCX
staining was consistent with that previously reported based on (Figure S5A). In addition, the expression level of MeCP2 was
immunohistochemical, biochemical or electron microscopic not significantly affected for the N-terminal tagging of MeCP2
studies. Fifth, the localization of HA staining was similar from (Figure S4D). However, potential effects of NHEJ have to be
1814 Cell 165, 1803–1817, June 16, 2016

carefully evaluated for each gene depending on the purpose of 2001). The final concentration of each plasmid (pX330-derivatives,
experiments. For example, when the target is a secreted or a pPB-CAG-mEGFP, pPB-CAG-DsRed2, pPB-CAG-tdTomato, and pCAG-
hyPBase), the ssODNs for HDR and the double-stranded DNA template for
type I membrane protein with a signal sequence, one should
mEGFP insertion were 1 mg/ml, 20 mM, and 1 mg/ml, respectively (Table S3).
select a target sequence for CRISPR-Cas9-mediated cleavage The position and angle of the electrode was set as previously described with
to minimize the possibility of deletion or mislocalization of the some modifications (see Figure 4A). Other details are described in the Supple-
gene products. Second, immunodetection of a fused epitope mental Experimental Procedures.
tag may be difficult for some proteins, because the accessibility
of antibodies may be sterically limited. For example, we failed to Histology
detect the HA signal with immunostaining in tissue in which the Mice were fixed by transcardial perfusion with 4% paraformaldehyde in 0.1 M
HA tag was inserted to the C terminus of PSD-95 with SLENDR, phosphate buffer and coronal sections (50 or 100 mm in thickness) were
whereas PCR detected the HA knockin allele at the DNA level prepared using a vibratome (VT1200, Leica). Immunohistochemistry was
performed as previously described (Fukaya and Watanabe, 2000; Nishiyama
(data not shown). This is likely due to high protein density of
et al., 2012). Other details are described in the Supplemental Experimental
PSD, which is known to often prevent antibodies to access to Procedures.
targeted proteins in the structure (Fukaya and Watanabe,
2000). This limitation of immunostaining can be overcome by Pre-embedding Immunoelectron Microscopy
fusing a fluorescent protein tag with SLENDR and directly An HA-CaMKIIb knockin mouse was transcardially perfused with 4% para-
observing the fluorescence (Fortin et al., 2014). These limitations formaldehyde and 0.1% glutaraldehyde in 0.1 M phosphate buffer. The
would not detract the impact of SLENDR, as the technique fixed brain was sectioned into 50-mm thick slices with a vibratome (VT
should be applicable to most proteins with little optimization as 1200, Leica), and sections with positive DsRed2 labeling were chosen.
Tissue processing for pre-embedding immunoelectron microscopy was
shown in this study.
performed as previously described (Parajuli et al., 2012). After processing,
Since the 50 -NGG protospacer-adjacent motif (PAM) of a region containing HA-positive neurons was trimmed out, and 50-nm thick
SpCas9 abundantly exists in the genome and other CRISPR en- sections were cut and collected onto a Kapton tape by ATUMtome (RMC/
donucleases with different PAMs are also available to target Boeckeler) (Kasthuri et al., 2015; Schalek et al., 2011). The Kapton tape
different sites in the genome (Cong et al., 2013; Hsu et al., was placed on a silicon wafer, and a layer of 5-nm-thick carbon was coated
2014; Zetsche et al., 2015), most proteins should be suitable on the wafer surface with a high vacuum sputter coater (ACE600, Leica).
The sections on the wafer were imaged under a scanning electron micro-
for SLENDR. Furthermore, its throughput and cost-effectiveness
scope (SEM, Merlin VP Compact, Zeiss) assisted with Atlas 5 AT software
are much higher than previous single-cell protein labeling (Zeiss). Other details are described in the Supplemental Experimental
methods (Fortin et al., 2014; Gross et al., 2013). Thus, SLENDR Procedures.
should allow large-scale, potentially genome-wide, determina-
tion of precise subcellular localization of endogenous proteins Two-Photon Glutamate Uncaging
in various cell types and ages, providing a new level of under- Organotypic cortical slice cultures were prepared from mEGFP-CaMKIIa/b
standing of protein and cellular function in the brain. knockin mice as described previously (Stoppini et al., 1991; Yamamoto
et al., 1989). Secondary apical dendritic branches of layer 2/3 neurons were
imaged at 15–17 days in vitro under a custom-built two-photon microscope
EXPERIMENTAL PROCEDURES
with two Ti:Sapphire lasers (Chameleon, Coherent) (Lee et al., 2009). MNI-
caged L-glutamate (4-methoxy-7-nitroindolinyl-caged L-glutamate, Tocris)
Full experimental details can be found in the Supplemental Experimental
was uncaged with a train of 6-ms laser pulses (3.5–4 mW under the objective,
Procedures.
30 times at 1 Hz) near a spine of interest. Experiments were performed at 32 C.
Images were analyzed with MATLAB (MathWorks) and ImageJ. Other details
Animals are described in the Supplemental Experimental Procedures.
All experimental procedures were approved by the Max Planck Florida Insti-
tute for Neuroscience Institutional Animal Care and Use Committee and
were performed in accordance with guidelines from the US NIH. Swiss Statistical Analysis
Webster mice were obtained from Charles River. The day on which the vaginal All statistical values were presented as mean ± SEM. The Student’s t test was
plug was detected was designated as E0). The first 24 hr after birth was used when two independent samples were compared. Dunnett’s test was
referred to as P0). Both of male and female mice were used. used for multiple comparisons. Statistical analysis was performed with
GraphPad Prism 6. Differences between datasets were judged to be signifi-
cant at p < 0.05.
DNA Constructs
The human codon-optimized SpCas9 and sgRNA expression plasmid was a
gift from F. Zhang (pX330, Addgene plasmid #42230) (Cong et al., 2013). To SUPPLEMENTAL INFORMATION
generate sgRNA-expressing plasmids, a pair of annealed oligos (20 bp)
was ligated into the sgRNA scaffold of pX330 (Ran et al., 2013). We purchased Supplemental Information includes Supplemental Experimental Procedures,
single-stranded oligodeoxynucleotides (ssODNs) for HDR from Integrated six figures, and three tables and can be found with this article online at
DNA Technologies. To fuse mEGFP (monomeric EGFP, A206K) to the C or N http://dx.doi.org/10.1016/j.cell.2016.04.044.
terminus of CaMKIIa or CaMKIIb, respectively, plasmid-based donor tem-
plates for HDR were prepared. Other details are described in the Supplemental AUTHOR CONTRIBUTIONS
Experimental Procedures.
T.M. and J.N. conceived the SLENDR idea. T.M., J.N., and R.Y. designed the
In Utero Electroporation experiments. T.M. and J.N. performed most experiments and data analysis.
In utero electroporation was performed as previously described (Borrell et al., Y.S. and N.K. performed electron microscopy imaging and data analysis.
2005; Chen and LoTurco, 2012; Imamura and Greer, 2015; Kitazawa et al., T.M., J.N., and R.Y. wrote the paper. All the authors discussed the results
2014; Nishiyama et al., 2012; Soma et al., 2009; Tabata and Nakajima, and commented on the manuscript.
Cell 165, 1803–1817, June 16, 2016 1815

ACKNOWLEDGMENTS and adult ages: implications for the maturation of synaptic physiology and
long-term potentiation. J. Neurosci. 12, 2685–2705.
The authors thank Sanger Institute for the gift of the piggyBac transposon Heidenreich, M., and Zhang, F. (2016). Applications of CRISPR-Cas systems in
vector plasmids (pPB-CAG.EBNXN and pCMV-hyPBase) and Y. Hayashi for neuroscience. Nat. Rev. Neurosci. 17, 36–44.
CaMKIIa and CaMKIIb cDNA; L. Colgan, M. Yuzaki, S. Soderling, and
Hell, J.W., Westenbroek, R.E., Warner, C., Ahlijanian, M.K., Prystay, W.,
M. Kano for critical reading of the manuscript; the R.Y. lab members for discus-
Gilbert, M.M., Snutch, T.P., and Catterall, W.A. (1993). Identification and differ-
sion; J. Richards, K. Liu, and J. Chen for technical assistance; and D. Kloetzer
ential subcellular localization of the neuronal class C and class D L-type cal-
for laboratory management. This work was supported by the NIH
cium channel alpha 1 subunits. J. Cell Biol. 123, 949–962.
(R01MH080047 and DP1NS096787 to R.Y.) and the Human Frontier Science
Program (long-term fellowship to T.M.). Hsu, P.D., Lander, E.S., and Zhang, F. (2014). Development and applications
of CRISPR-Cas9 for genome engineering. Cell 157, 1262–1278.
Received: February 3, 2016 Hutagalung, A.H., and Novick, P.J. (2011). Role of Rab GTPases in membrane
Revised: March 14, 2016 traffic and cell physiology. Physiol. Rev. 91, 119–149.
Accepted: April 13, 2016
Imamura, F., and Greer, C.A. (2015). Segregated labeling of olfactory bulb pro-
Published: May 12, 2016
jection neurons based on their birthdates. Eur. J. Neurosci. 41, 147–156.
Incontro, S., Asensio, C.S., Edwards, R.H., and Nicoll, R.A. (2014). Efficient,
REFERENCES
complete deletion of synaptic proteins using CRISPR. Neuron 83, 1051–1057.
Borrell, V., Yoshimura, Y., and Callaway, E.M. (2005). Targeted gene delivery to Jinek, M., Chylinski, K., Fonfara, I., Hauer, M., Doudna, J.A., and Charpentier,
telencephalic inhibitory neurons by directional in utero electroporation. E. (2012). A programmable dual-RNA-guided DNA endonuclease in adaptive
J. Neurosci. Methods 143, 151–158. bacterial immunity. Science 337, 816–821.
Bosch, M., Castro, J., Saneyoshi, T., Matsuno, H., Sur, M., and Hayashi, Y. Kasthuri, N., Hayworth, K.J., Berger, D.R., Schalek, R.L., Conchello, J.A.,
(2014). Structural and molecular remodeling of dendritic spine substructures Knowles-Barley, S., Lee, D., Vázquez-Reina, A., Kaynig, V., Jones, T.R.,
during long-term potentiation. Neuron 82, 444–459. et al. (2015). Saturated reconstruction of a volume of neocortex. Cell 162,
648–661.
Chen, F., and LoTurco, J. (2012). A method for stable transgenesis of radial glia
lineage in rat neocortex by piggyBac mediated transposition. J. Neurosci. Kim, S., Kim, D., Cho, S.W., Kim, J., and Kim, J.S. (2014). Highly efficient
Methods 207, 172–180. RNA-guided genome editing in human cells via delivery of purified Cas9 ribo-
nucleoproteins. Genome Res. 24, 1012–1019.
Chen, R.Z., Akbarian, S., Tudor, M., and Jaenisch, R. (2001). Deficiency of
methyl-CpG binding protein-2 in CNS neurons results in a Rett-like phenotype Kim, K., Lakhanpal, G., Lu, H.E., Khan, M., Suzuki, A., Hayashi, M.K., Nar-
in mice. Nat. Genet. 27, 327–331. ayanan, R., Luyben, T.T., Matsuda, T., Nagai, T., et al. (2015). A temporary
Chu, V.T., Weber, T., Wefers, B., Wurst, W., Sander, S., Rajewsky, K., and gating of actin remodeling during synaptic plasticity consists of the interplay
Kühn, R. (2015). Increasing the efficiency of homology-directed repair for between the kinase and structural functions of CaMKII. Neuron 87, 813–826.
CRISPR-Cas9-induced precise gene editing in mammalian cells. Nat. Bio- Kitazawa, A., Kubo, K., Hayashi, K., Matsunaga, Y., Ishii, K., and Nakajima, K.
technol. 33, 543–548. (2014). Hippocampal pyramidal neurons switch from a multipolar migration
Cong, L., Ran, F.A., Cox, D., Lin, S., Barretto, R., Habib, N., Hsu, P.D., Wu, X., mode to a novel ‘‘climbing’’ migration mode during development.
Jiang, W., Marraffini, L.A., and Zhang, F. (2013). Multiplex genome engineering J. Neurosci. 34, 1115–1126.
using CRISPR/Cas systems. Science 339, 819–823. Kleinstiver, B.P., Pattanayak, V., Prew, M.S., Tsai, S.Q., Nguyen, N.T., Zheng,
Contractor, A., Klyachko, V.A., and Portera-Cailliau, C. (2015). Altered Z., and Joung, J.K. (2016). High-fidelity CRISPR-Cas9 nucleases with no
neuronal and circuit excitability in Fragile X syndrome. Neuron 87, 699–715. detectable genome-wide off-target effects. Nature 529, 490–495.
Cox, D.B., Platt, R.J., and Zhang, F. (2015). Therapeutic genome editing: pros- Korb, E., Wilkinson, C.L., Delgado, R.N., Lovero, K.L., and Finkbeiner, S.
pects and challenges. Nat. Med. 21, 121–131. (2013). Arc in the nucleus regulates PML-dependent GluA1 transcription and
homeostatic plasticity. Nat. Neurosci. 16, 874–883.
Doudna, J.A., and Charpentier, E. (2014). Genome editing. The new frontier of
genome engineering with CRISPR-Cas9. Science 346, 1258096. Lee, S.J., Escobedo-Lozoya, Y., Szatmari, E.M., and Yasuda, R. (2009).
Activation of CaMKII in single dendritic spines during long-term potentiation.
Fortin, D.A., Tillo, S.E., Yang, G., Rah, J.C., Melander, J.B., Bai, S., Soler-Ce-
Nature 458, 299–304.
deño, O., Qin, M., Zemelman, B.V., Guo, C., et al. (2014). Live imaging of
endogenous PSD-95 using ENABLED: a conditional strategy to fluorescently Lin, S., Staahl, B.T., Alla, R.K., and Doudna, J.A. (2014). Enhanced homology-
label endogenous proteins. J. Neurosci. 34, 16698–16712. directed human genome engineering by controlled timing of CRISPR/Cas9
delivery. eLife 3, e04766.
Fukaya, M., and Watanabe, M. (2000). Improved immunohistochemical detec-
tion of postsynaptically located PSD-95/SAP90 protein family by protease Lisman, J., Yasuda, R., and Raghavachari, S. (2012). Mechanisms of CaMKII
section pretreatment: a study in the adult mouse brain. J. Comp. Neurol. action in long-term potentiation. Nat. Rev. Neurosci. 13, 169–182.
426, 572–586. Maruyama, T., Dougan, S.K., Truttmann, M.C., Bilate, A.M., Ingram, J.R., and
Gleeson, J.G., Lin, P.T., Flanagan, L.A., and Walsh, C.A. (1999). Doublecortin Ploegh, H.L. (2015). Increasing the efficiency of precise genome editing with
is a microtubule-associated protein and is expressed widely by migrating neu- CRISPR-Cas9 by inhibition of nonhomologous end joining. Nat. Biotechnol.
rons. Neuron 23, 257–271. 33, 538–542.
Gomez-Ospina, N., Tsuruta, F., Barreto-Chang, O., Hu, L., and Dolmetsch, R. Nishiyama, J., and Yasuda, R. (2015). Biochemical computation for spine
(2006). The C terminus of the L-type voltage-gated calcium channel Ca(V)1.2 structural plasticity. Neuron 87, 63–75.
encodes a transcription factor. Cell 127, 591–606. Nishiyama, J., Hayashi, Y., Nomura, T., Miura, E., Kakegawa, W., and Yuzaki,
Gross, G.G., Junge, J.A., Mora, R.J., Kwon, H.B., Olson, C.A., Takahashi, T.T., M. (2012). Selective and regulated gene expression in murine Purkinje cells by
Liman, E.R., Ellis-Davies, G.C., McGee, A.W., Sabatini, B.L., et al. (2013). in utero electroporation. Eur. J. Neurosci. 36, 2867–2876.
Recombinant probes for visualizing endogenous synaptic proteins in living Parajuli, L.K., Nakajima, C., Kulik, A., Matsui, K., Schneider, T., Shigemoto, R.,
neurons. Neuron 78, 971–985. and Fukazawa, Y. (2012). Quantitative regional and ultrastructural localization
Harris, K.M., Jensen, F.E., and Tsao, B. (1992). Three-dimensional structure of of the Ca(v)2.3 subunit of R-type calcium channel in mouse brain. J. Neurosci.
dendritic spines and synapses in rat hippocampus (CA1) at postnatal day 15 32, 13555–13567.
1816 Cell 165, 1803–1817, June 16, 2016

Platt, R.J., Chen, S., Zhou, Y., Yim, M.J., Swiech, L., Kempton, H.R., Dahlman, Toyo-oka, K., Wachi, T., Hunt, R.F., Baraban, S.C., Taya, S., Ramshaw, H.,
J.E., Parnas, O., Eisenhaure, T.M., Jovanovic, M., et al. (2014). CRISPR-Cas9 Kaibuchi, K., Schwarz, Q.P., Lopez, A.F., and Wynshaw-Boris, A. (2014).
knockin mice for genome editing and cancer modeling. Cell 159, 440–455. 14-3-3ε and z regulate neurogenesis and differentiation of neuronal progenitor
Ran, F.A., Hsu, P.D., Wright, J., Agarwala, V., Scott, D.A., and Zhang, F. (2013). cells in the developing brain. J. Neurosci. 34, 12168–12181.
Genome engineering using the CRISPR-Cas9 system. Nat. Protoc. 8, 2281– Viswanathan, S., Williams, M.E., Bloss, E.B., Stasevich, T.J., Speer, C.M.,
2308. Nern, A., Pfeiffer, B.D., Hooks, B.M., Li, W.P., English, B.P., et al. (2015).
Riedl, J., Crevenna, A.H., Kessenbrock, K., Yu, J.H., Neukirchen, D., Bista, M., High-performance probes for light and electron microscopy. Nat. Methods
Bradke, F., Jenne, D., Holak, T.A., Werb, Z., et al. (2008). Lifeact: a versatile 12, 568–576.
marker to visualize F-actin. Nat. Methods 5, 605–607.
Wang, H., Yang, H., Shivalila, C.S., Dawlaty, M.M., Cheng, A.W., Zhang, F.,
Saleh-Gohari, N., and Helleday, T. (2004). Conservative homologous recombi- and Jaenisch, R. (2013). One-step generation of mice carrying mutations in
nation preferentially repairs DNA double-strand breaks in the S phase of the multiple genes by CRISPR/Cas-mediated genome engineering. Cell 153,
cell cycle in human cells. Nucleic Acids Res. 32, 3683–3688. 910–918.
Sander, J.D., and Joung, J.K. (2014). CRISPR-Cas systems for editing, regu-
Xue, W., Chen, S., Yin, H., Tammela, T., Papagiannakopoulos, T., Joshi, N.S.,
lating and targeting genomes. Nat. Biotechnol. 32, 347–355.
Cai, W., Yang, G., Bronson, R., Crowley, D.G., et al. (2014). CRISPR-mediated
Schalek, R., Kasthuri, N., Hayworth, K., Berger, D., Tapia, J., Morgan, J., direct mutation of cancer genes in the mouse liver. Nature 514, 380–384.
Turaga, S., Fagerholm, E., Seung, H., and Lichtman, J. (2011). Development
of high-throughput, high-resolution 3D reconstruction of large-volume biolog- Yamamoto, N., Kurotani, T., and Toyama, K. (1989). Neural connections
ical tissue using automated tape collection ultramicrotomy and scanning elec- between the lateral geniculate nucleus and visual cortex in vitro. Science
tron microscopy. Microsc. Microanal. 17, 966–967. 245, 192–194.
Shepherd, J.D., and Bear, M.F. (2011). New views of Arc, a master regulator of Yang, J., Siao, C.J., Nagappan, G., Marinic, T., Jing, D., McGrath, K., Chen,
synaptic plasticity. Nat. Neurosci. 14, 279–284. Z.Y., Mark, W., Tessarollo, L., Lee, F.S., et al. (2009). Neuronal release of
Slaymaker, I.M., Gao, L., Zetsche, B., Scott, D.A., Yan, W.X., and Zhang, F. proBDNF. Nat. Neurosci. 12, 113–115.
(2016). Rationally engineered Cas9 nucleases with improved specificity. Sci- Yang, H., Wang, H., Shivalila, C.S., Cheng, A.W., Shi, L., and Jaenisch, R.
ence 351, 84–88. (2013). One-step generation of mice carrying reporter and conditional alleles
Soma, M., Aizawa, H., Ito, Y., Maekawa, M., Osumi, N., Nakahira, E., Okamoto, by CRISPR/Cas-mediated genome engineering. Cell 154, 1370–1379.
H., Tanaka, K., and Yuasa, S. (2009). Development of the mouse amygdala as Yin, H., Xue, W., Chen, S., Bogorad, R.L., Benedetti, E., Grompe, M., Kotelian-
revealed by enhanced green fluorescent protein gene transfer by means of in sky, V., Sharp, P.A., Jacks, T., and Anderson, D.G. (2014). Genome editing
utero electroporation. J. Comp. Neurol. 513, 113–128. with Cas9 in adult mice corrects a disease mutation and phenotype. Nat.
Steinberg, S.F. (2008). Structural basis of protein kinase C isoform function. Biotechnol. 32, 551–553.
Physiol. Rev. 88, 1341–1378.
Yusa, K., Zhou, L., Li, M.A., Bradley, A., and Craig, N.L. (2011). A hyperactive
Stoppini, L., Buchs, P.A., and Muller, D. (1991). A simple method for organo- piggyBac transposase for mammalian applications. Proc. Natl. Acad. Sci. USA
typic cultures of nervous tissue. J. Neurosci. Methods 37, 173–182. 108, 1531–1536.
Straub, C., Granger, A.J., Saulnier, J.L., and Sabatini, B.L. (2014). CRISPR/
Yuste, R., and Bonhoeffer, T. (2004). Genesis of dendritic spines: insights from
Cas9-mediated gene knock-down in post-mitotic neurons. PLoS ONE 9,
ultrastructural and imaging studies. Nat. Rev. Neurosci. 5, 24–34.
e105584.
Swiech, L., Heidenreich, M., Banerjee, A., Habib, N., Li, Y., Trombetta, J., Sur, Zetsche, B., Gootenberg, J.S., Abudayyeh, O.O., Slaymaker, I.M., Makarova,
M., and Zhang, F. (2015). In vivo interrogation of gene function in the mamma- K.S., Essletzbichler, P., Volz, S.E., Joung, J., van der Oost, J., Regev, A., et al.
lian brain using CRISPR-Cas9. Nat. Biotechnol. 33, 102–106. (2015). Cpf1 is a single RNA-guided endonuclease of a class 2 CRISPR-Cas
system. Cell 163, 759–771.
Tabata, H., and Nakajima, K. (2001). Efficient in utero gene transfer system to
the developing mouse brain using electroporation: visualization of neuronal Zong, H., Espinosa, J.S., Su, H.H., Muzumdar, M.D., and Luo, L. (2005).
migration in the developing cortex. Neuroscience 103, 865–872. Mosaic analysis with double markers in mice. Cell 121, 479–492.
Cell 165, 1803–1817, June 16, 2016 1817

Resource
Ultra-High Resolution 3D Imaging of Whole Cells

Fang Huang,1,2,14 George Sirinakis,1,3,14 Edward S. Allgeyer,1,3 Lena K. Schroeder,1 Whitney C. Duim,1,4
Emil B. Kromann,1,5 Thomy Phan,1 Felix E. Rivera-Molina,1 Jordan R. Myers,1 Irnov Irnov,6,7 Mark Lessard,8
Yongdeng Zhang,1 Mary Ann Handel,8 Christine Jacobs-Wagner,6,7,9,10 C. Patrick Lusk,1 James E. Rothman,1,11
Derek Toomre,1,11 Martin J. Booth,12,13 and Joerg Bewersdorf1,5,11,*
1Department of Cell Biology, School of Medicine, Yale University, New Haven, CT 06520, USA
2Weldon School of Biomedical Engineering, Purdue University, West Lafayette, IN 47907, USA
3The Gurdon Institute, University of Cambridge, Cambridge CB2 1QN, UK
4Department of Chemistry, Harvey Mudd College, Claremont, CA 91711, USA
5Department of Biomedical Engineering, Yale University, CT 06520, USA
6Microbial Sciences Institute, Yale University, West Haven, CT 06516, USA
7Department of Molecular, Cellular and Developmental Biology, Yale University, New Haven, CT 06520, USA
8The Jackson Laboratory, Bar Harbor, ME 04609, USA
9Howard Hughes Medical Institute, Yale University, New Haven, CT 06520, USA
10Department of Microbial Pathogenesis, Yale School of Medicine, New Haven, CT 06520, USA
11Nanobiology Institute, Yale University, West Haven, CT 06516, USA
12Department of Engineering Science, University of Oxford, Oxford OX1 3PJ, UK
13Centre for Neural Circuits and Behaviour, University of Oxford, Oxford OX1 3SR, UK
14Co-first author
*Correspondence: joerg.bewersdorf@yale.edu
SUMMARY ing nanoscopy (SMSN) (Betzig et al., 2006; Hess et al., 2006;
Rust et al., 2006) has extended the application range of fluores-
Fluorescence nanoscopy, or super-resolution micro- cence microscopy beyond the diffraction limit, achieving up to
scopy, has become an important tool in cell biolog- 10-fold improvement in resolution (Gould et al., 2012a). These
ical research. However, because of its usually inferior methods are now maturing and offering the opportunity to
resolution in the depth direction (50–80 nm) and observe biological phenomena never before seen (Chojnacki
rapidly deteriorating resolution in thick samples, its et al., 2012; Kanchanawong et al., 2010; Liu et al., 2011; Xu
et al., 2013). Nanoscopy techniques share a common principle:
practical biological application has been effectively
they spatially separate unresolvable fluorescent molecules by
limited to two dimensions and thin samples. Here, independently switching their emission ‘‘on’’ and ‘‘off’’ (Hell,
we present the development of whole-cell 4Pi sin- 2007). In particular, SMSN methods such as photoactivated
gle-molecule switching nanoscopy (W-4PiSMSN), localization microscopy (PALM), fluorescence photoactivation
an optical nanoscope that allows imaging of three- localization microscopy (FPALM), and stochastic optical recon-
dimensional (3D) structures at 10- to 20-nm resolu- struction microscopy (STORM) use a stochastic approach where
tion throughout entire mammalian cells. We demon- only a small subset of fluorescent molecules is switched on at
strate the wide applicability of W-4PiSMSN across any particular moment in time while the majority remains in a
diverse research fields by imaging complex molecu- non-fluorescent ‘‘dark’’ or ‘‘off’’ state (Gould et al., 2012a). Su-
lar architectures ranging from bacteriophages to nu- per-resolved images are reconstructed from the positions of
clear pores, cilia, and synaptonemal complexes in thousands to millions of single molecules that have been re-
corded in thousands of camera frames.
large 3D cellular volumes.
This imaging strategy was initially applied to single-objective
microscopes in two dimensions (2D) (Betzig et al., 2006; Hess
INTRODUCTION et al., 2006; Rust et al., 2006) and later extended to three dimen-
sions (3D) (Huang et al., 2008; Juette et al., 2008; Pavani et al.,
Major advances in cell biology are tightly linked to innovations in 2009). While these instruments achieve 20- to 40-nm resolution
microscopy. The development of fluorescence microscopy, for in the focal plane (lateral, x-y), the resolution in the depth direc-
example, enabled sub-cellular localization of specifically labeled tion (axial, z) is typically limited to only 50–80 nm. The resolution
proteins of interest (Lichtman and Conchello, 2005). However, can, however, be further improved by using a dual-objective
the wave nature of light restricts the resolution of conventional ‘‘4Pi’’ detection geometry (Bewersdorf et al., 2006).
light microscopy to 200 nm, making details of subcellular struc- Using two objectives doubles the detection efficiency (Xu
tures and protein assemblies unresolvable (Hell, 2007). The et al., 2012) and thus improves the localization precision
advent of super-resolution fluorescence microscopy, or nano- 1.4-fold in all three dimensions. Additionally, employing two
scopy, techniques such as stimulated emission depletion objectives in a 4Pi geometry allows the creation of a single-mole-
(STED) (Hell and Wichmann, 1994) and single-molecule switch- cule emission interference pattern at the detector leading to an
1028 Cell 166, 1028–1040, August 11, 2016 ª 2016 The Authors. Published by Elsevier Inc.
This is an open access article under the CC BY license (http://creativecommons.org/licenses/by/4.0/).
7-fold improvement in axial localization precision over single- poor localization density because of the highly selective compu-
objective approaches as demonstrated using interferometric tational processes focusing on subtle features of PSFs. The
PALM (iPALM) (Shtengel et al., 2009) and 4Pi single marker methods are also susceptible to sample-induced optical aberra-
switching nanoscopy (4Pi-SMSN) (Aquino et al., 2011). This tions, which change the PSF shape when imaging biological
improved resolution enabled, for example, the generation of structures deeper in the sample (Burke et al., 2015; von Die-
anatomical maps of focal adhesions at 10-nm axial resolution zmann et al., 2015). As a result, applications have been restricted
(Case et al., 2015; Kanchanawong et al., 2010). However, this to thin structures close to the coverslip (Case et al., 2015; Van
method was initially restricted to samples of 250 nm in thick- Engelenburg et al., 2014; Kanchanawong et al., 2010).
ness (Shtengel et al., 2009) and more recently to 700–1,000 nm To enable 4Pi-SMSN to probe deeper into the cell and extend
(Aquino et al., 2011; Brown et al., 2011). As the typical thickness the application of this technology to larger cellular features, we
of a mammalian cell is 5–10 mm, this has limited optical micro- have developed W-4PiSMSN. First, expanding on the optical
scopy at the 10-nm isotropic resolution scale to thin sub-vol- design by Aquino et al. (2011), we included deformable mirrors
umes of cells, thus precluding the ability to image organelles in both arms of the 4Pi-interferometric cavity (Figure 1A; Fig-
that can extend over several microns throughout the whole cell. ure S1). We use these mirrors to correct for imperfections in
Here, we present a new implementation of iPALM/4Pi-SMSN, the instrument beam path and optimize the PSF quality for sam-
termed whole-cell 4Pi single-molecule switching nanoscopy ples with various thicknesses (Supplemental Information; Fig-
(W-4PiSMSN), which extends the imaging capabilities of this ure S2). Deformable mirrors also allow us to compensate for
technology to whole cells without compromising resolution. sample-induced aberration modes, such as spherical aberra-
W-4PiSMSN allows volumetric reconstruction with 10- to tions (Burke et al., 2015; Gould et al., 2012b), which vary from
20-nm isotropic resolution of 10-mm-thick samples, a 10- to sample to sample and with depth. Additionally, we can use these
40-fold improvement in sample thickness over previous mirrors to introduce astigmatism in both interference arms
iPALM/4Pi-SMSN implementations (Aquino et al., 2011; Brown without adding further complexity to the system (Supplemental
et al., 2011; Van Engelenburg et al., 2014; Shtengel et al., Information). Thus, the deformable mirrors enable compro-
2009). Our approach permits ultra-high resolution 3D imaging mise-free, reproducible PSFs in a depth- and sample-indepen-
of virtually any subcellular structure. To demonstrate this, we dent manner.
image the endoplasmic reticulum (ER), bacteriophages, mito- Second, building on an earlier approach by Brown et al. (2011),
chondria, nuclear pore complexes, primary cilia, Golgi-appa- we developed an analysis method that combines information
ratus-associated COPI vesicles, and mouse spermatocyte from (1) the 4Pi-PSF’s interference phase, which allows for pre-
synaptonemal complexes. By these examples, we show that cise axial localization but does not distinguish between different
W-4PiSMSN opens the door to address cell biological questions interference peaks, and (2) the eccentricity of the astigmatic 4Pi-
that were previously unanswerable. PSF, which narrows axial localizations down to individual inter-
ference peaks but in itself does not offer the precision of 4Pi
RESULTS interference. Our new analysis algorithm interprets the large
number of molecules imaged in each time and z-depth window
Development of W-4PiSMSN as an ensemble measurement of the concurrent state of the
To realize a system that achieves 10- to 20-nm 3D resolution W-4PiSMSN system (Supplemental Information; Figure S3A)
across the thickness of entire mammalian cells, we expanded and determines the relationship between the eccentricity of the
on previous iPALM and 4Pi-SMSN developments (Aquino astigmatic PSF and the interference phase of the 4Pi-PSF.
et al., 2011; Shtengel et al., 2009). In these systems, fluores- Then the axial positions of all corresponding molecules can be
cence emission is collected by two opposing objective lenses assigned with high precision and unambiguously using a mono-
and combined to interfere (Aquino et al., 2011; von Middendorff tonic metric designed to describe the overall shape of the PSF
et al., 2008; Shtengel et al., 2009). Depending on the axial posi- and maintain its monotonicity in the presence of moderate
tion of a molecule, the light will interfere constructively or amounts of aberrations (Supplemental Information; Figure S3B).
destructively, as indicated by the brightness of the molecule’s Since this analysis is performed for well-defined temporal and
image on the detector. However, molecules at axial positions axial data subsets, we have generalized it to identify and correct
that differ by multiples of half the wavelength of light lead to for drift (from both the system and the sample) over the course of
the same interference pattern and cause ambiguity in deter- imaging. Our method is robust against aberrations and improves
mining their axial positions. This localization ambiguity leads to the reliability and efficiency of axial position assignment as it
scrambled images that contain axially shifted image artifacts, automatically adapts to changes in the shape and interference
known as ghost images, in samples thicker than 250 nm. pattern of the 4Pi-PSF.
This can be avoided by using not only the brightness but also
the z-position-dependent shape of the single-molecule images Ultra-High Resolution Imaging with W-4PiSMSN
to determine a molecule’s axial position. To address this, a To demonstrate the resolution capabilities of our new system,
higher-moment based analysis (Aquino et al., 2011) and analysis we first imaged the ER. ER membranes were stained using
of the point-spread function (PSF) eccentricity in a hyperbolic anti-GFP antibodies to the overexpressed transmembrane pro-
mirrors-modified system (Brown et al., 2011) were developed, tein, mEmerald-Sec61b, in COS-7 cells. We visualized the ER
which extended the image volume thickness to 700–1,000 nm. as a connected network of hollow tubes with 60–100 nm diame-
However, these methods pose significant drawbacks such as ters (Figures 1B and 1C; Movie S1). Both horizontal and vertical
Cell 166, 1028–1040, August 11, 2016 1029

Figure 1. W-4PiSMSN Design and Resolu-
tion Demonstrations with ER, Microtubules,
and Bacteriophages
(A) Simplified optical diagram of W-4PiSMSN.
(B) Overview and cross-sections of the ER network
in an immunolabeled COS-7 cell. Cross-sections
of the W-4PiSMSN reconstruction show clearly
separated membranes of the tubular structures,
which cannot be resolved with conventional
astigmatism-based nanoscopy (light blue frame).
(C) x-y slice through the mid-section of the ER
network shown in (B) highlights the distinct mem-
brane contour of ER tubules (arrowheads).
(D) Overview of immunolabeled microtubules in a
COS-7 cell.
(E and F) 20-nm-thin x-y slices of the red (E) and
green (F) segments shown in (D) demonstrate that
microtubules can be easily resolved as hollow
cylinders in W-4PiSMSN.
(G) A look through a 120-nm-long segment of the
microtubule of (F).
(H) A histogram showing the number of localiza-
tions in a cross-section of the microtubule, white
dotted box in (G).
(I) A bacteriophage reconstructed from 115 aver-
aged viral particles rendered in 3D.
(J and K) 5-nm-thin vertical (J) and horizontal (K)
slices through the averaged dataset correspond-
ing to the planes shown in (I).
(L) The internal angle measurements of the hexa-
gon shape identified from the viral capsid shown
in (J).
OBJ, objective; QWP, quarter-wave plate; DM,
dichroic mirror; QBF, quad-band band-pass filter;
Def. Mirror, deformable mirror; Cam, camera;
50/50, beam splitter cube.
mary and secondary antibodies adds un-

certainty to the actual position of the
imaged dye molecules, we conclude
that the 3D resolution of our instrument
is well below 20 nm (FWHM).
To demonstrate our approach on
another challenging target, we imaged T7
bacteriophages. They feature an icosahe-
dral-shaped capsid of 60-nm diameter,
cross-sections reveal the 3D membrane contour that was previ- which has only been visible by cryo-electron microscopy (cryo-
ously resolvable only with electron tomography (Frey and Man- EM) techniques before (Hu et al., 2013). We non-specifically
nella, 2000). This high 3D resolution is quantitatively supported labeled proteins on the surface of purified T7 phages using an
by a Fourier Shell Correlation value of 22 nm (Figure S4) (Nieu- Alexa Fluor 647 NHS Ester, which reacts with primary amines,
wenhuizen et al., 2013). To test our approach on even smaller and mounted the phages on a coverslip (Figure S5). Image slices
structures, we imaged antibody-labeled microtubules in COS-7 of a single phage in the x-y, y-z, and x-z directions show a hollow
cells, a gold standard in SMSN (Figures 1D–1H). Without any center in all orientations. To further refine the details of the
detectable imaging artifacts, W-4PiSMSN resolves this 25-nm detected phage structures, we adapted the tomogram-aver-
microtubule filament, which appears in all orientations as a hol- aging approach originally developed for cryo-EM (Briggs, 2013;
low core coated with antibodies (Figures 1E–1H). In addition, Broeken et al., 2015). By combining 115 T7 images, our averaged
the dataset features a high localization density of 5.5 localiza- reconstruction reveals the icosahedral shape of the T7 phages
tion events per 10 3 10 nm2 of surface area. Displaying the local- (Figures 1I–1L; Figure S4). As presented in Figures 1J–1L, a slice
ization events by their radial distance from the tubule axis perpendicular to the major axis shows the expected pentagonal
shows a Gaussian peak with a full width at half maximum shape while a slice parallel to the major axis reveals a hexagonal
(FWHM) of 16–24 nm (Figure S4). Considering that the use of pri- shape. Our approach, however, has yet to clearly resolve the
1030 Cell 166, 1028–1040, August 11, 2016

Figure 2. Two-Color Reconstruction of Mitochondria and Microtubules
(A and B) W-4PiSMSN reconstruction of microtubules (A) and mitochondria (TOM20) (B) in a COS-7 cell immunolabeled with Alexa Fluor 647 and Cy3B,
respectively. An x-y overview and x-z and y-z slices (yellow and magenta lines, respectively) are shown.
(C) The combined two-color image reveals microtubules running adjacent to the mitochondria surface.
23-nm tail and fiber structures of the T7 phage (Hu et al., 2013). ple-induced aberrations or drift can lead to significant deteriora-
This is likely due to either incomplete labeling of the surface tion of the resolution and distortions of the super-resolved vol-
proteins or the flexibility of these structures. Nonetheless, our ume (Huang et al., 2008; Mlodzianoski et al., 2011).
W-4PiSMSN system has enabled the visualization of the ultra- We designed our system to minimize drift: our instrument
structure of bacteriophages using light microscopy. design takes advantage of a horizontal symmetry plane coin-
We tested the two-color imaging capability of W-4PiSMSN ciding with the common focal planes of the objectives and the
by imaging microtubules and mitochondria in a COS-7 cell im- beam splitter cube of the interference cavity. This symmetric
munolabeled with Alexa Fluor 647 and Cy3B, respectively. Our design desensitizes the interferometric cavity of the microscope
reconstructions show microtubules running in close proximity to temperature changes leading to approximately equal thermal
10–20 nm from the mitochondria top and bottom surfaces (Fig- expansion in both arms of the interference cavity. To compen-
ures 2A–2C; Movie S2; Figure S5). Our system decouples axial sate for any remaining instrument and sample drift caused by
localization from the PSF shape, the latter being susceptible to mechanical and thermal fluctuations, we developed a set of
depth-dependent distortions caused by sample-induced optical hardware and software tools (Figure S6). The objectives are sta-
aberrations (Liu et al., 2013; McGorty et al., 2014). While single- bilized in 3D relative to each other by focusing a near-infrared
objective systems rely on the PSF shape, the W-4PiSMSN laser beam by one objective and detecting the focus with the
approach uses the relative interference amplitudes to determine other objective in a ‘‘biplane’’ configuration (Figure S6) (Juette
the axial location of individual molecules. However, multicolor et al., 2008; Ram et al., 2008). This allows the detection of relative
imaging is challenging because the spatial interference modula- objective movement in 3D, which can then be compensated for
tion frequency is wavelength dependent and differs between via a feedback loop. Furthermore, we cross-correlate 3D volume
color channels. To address this, we derived the modulation fre- data segments of 1- to 2-min windows using a redundancy-
quency using a pupil-function based approach (Supplemental based drift correction method (Li et al., 2013; Wang et al.,
Information). Our simulation results were verified experimentally 2014) in an extended correlation volume. Within each of these
by registering two color channels from an affine transformation short data segments, an independent relationship between
matrix, which was calibrated using two-color-labeled mitochon- astigmatism and interference phase is established. Any discrep-
dria in fixed cells (Supplemental Information). ancies between these relationships for different segments
are treated as drift (Supplemental Information). The above-
Whole-Cell 3D Imaging with W-4PiSMSN described methods enable us to fully compensate for sample
Imaging volumes thicker than 1.2 mm requires axial sample and instrument drift and changes in the optical path between
scanning, because molecules more than 600 nm out of focus the two arms of the interferometric 4Pi cavity due to the axial
cannot be identified and localized efficiently (Huang et al., scanning nature of the measurements.
2008; Juette et al., 2008). Thus, optical sections must be re- To demonstrate the whole-cell imaging capabilities of the
corded at different axial sample positions and subsequently W-4PiSMSN system, we imaged mitochondria using antibodies
merged to obtain the complete cellular volume. Compared to against the outer membrane protein TOM20 over the whole
conventional 3D nanoscopes, the superior localization precision thickness of a COS-7 cell. Figure 3 reveals the outer membrane
of W-4PiSMSN puts high demands on the localization accuracy contour and the remarkably interconnected mitochondrial
in each volume section (i.e., avoiding volume distortions) and network over a depth of 4.3 mm (Figures 3B–3D; Movie S3). We
the merging process. In the section-merging process, small were not able to detect any significant ghost images within the
misalignments of neighboring optical sections caused by sam- volume (Figures 3A–3D).
Cell 166, 1028–1040, August 11, 2016 1031

Figure 3. W-4PiSMSN Reconstruction of
TOM20 on Mitochondria in COS-7 Cell
(A) Overview of the mitochondria network visual-
ized by immunolabeling TOM20 with Alexa Flour
647. The dataset is assembled from 11 optical
sections with 500-nm step sizes.
(B) x-z cross-section of the purple plane in (A)
showing the distribution of TOM20 on the outer
mitochondrial membrane. Ghost images are
completely negligible.
(C and D) Top (C) and side (D) views of the orange
box in (A) show the 3D arrangement of the
organelle.
Moreover, as the Golgi complex is located

close to the middle of the cell, recording
high-quality data in a central z-plane is a
challenging test of the instrument’s 3D
resolution capabilities. Figure 5 shows
the b0 COP, a protein in the outer COPI
complex, immunostained using Alexa
Fluor 647 in BSC-1 cells. Strikingly, we
visualized distinct hollow COPI-coated
spheres within cells (Figures 5B, 5C, 5E,
5F, and 5G–5N). Our 3D images resolve
individual COPI vesicles with 100-nm
diameter, consistent with previous mea-
surements (Pellett et al., 2013). Addition-
ally, a 300-nm-thick section shows that
COPI-coated structures are packed
around a 500- to 1,000-nm (x and y) by
500-nm (z) area devoid of COPI labeling,
presumably containing a Golgi stack (Fig-
ures 5D–5F).
Revealing Ciliary Membrane GPCR

Organization
Most high-resolution studies of the pri-
mary cilium, a solitary microtubule-based
organelle that protrudes from the cell sur-
face and acts as a cellular antenna, have
relied on EM (Wood and Rosenbaum,
To further demonstrate that image quality is maintained 2015). A transmission EM image typically shows only a small
throughout the thickness of whole cells, we imaged nuclear subsection of a cilium as the sample is a random oblique 70-
pore complexes (NPCs) on the nuclear envelope. By immunolab- to 100-nm-thick section through the structure, which can be
eling with an antibody that recognizes a component of the cyto- up to 10 mm long and 250 nm wide. Scanning EM images
plasmic filaments (Nup358) of NPCs (von Appen et al., 2015), we can easily show an entire cilium with high resolution; however,
can reconstruct NPCs on the top, side, and bottom of the nu- these images completely lack information about specific protein
cleus (Figure 4; Movie S4). As with mitochondria, our approach localization. Previous nanoscopy studies on cilia relied on infer-
reveals the contours of almost the entire nuclear surface, where ring the 3D organization from 2D datasets (Yang et al., 2013,
both prominent invaginations and subtle undulations (typically 2015). Here, we used W-4PiSMSN to image the G-protein-
visualized only by electron microscopy [EM]) are apparent (Fig- coupled receptor Smoothened (SMO) on whole primary cilia in
ures 4A and 4B). hTERT-RPE1 cells with high 3D resolution (Figure 6; Movie S5).
SMO was tagged with a pH-sensitive GFP (pH-SMO), which
Revealing Golgi-Apparatus-Associated COPI Vesicles was used as an epitope for antibody labeling with Alexa Fluor
We next imaged COPI vesicles, which have traditionally been 647 (Figure 6; Supplemental Information). We observe that over-
resolved only by EM as they have 100-nm diameters and are expressed pH-SMO localizes to the membranes of cilia, which
densely packed around the Golgi cisternae (Orci et al., 1997). form hollow cylinders 3–10 mm long (Figure 6) and vary in
1032 Cell 166, 1028–1040, August 11, 2016

Figure 4. W-4PiSMSN Imaging of Nuclear
Pore Complexes over the Thickness of a
Cell Nucleus
Nup358 was Immunolabeled with Alexa Fluor 647
in hTERT-RPE1 cells.
(A) Overview of a region of the nucleus. The axial
location of the nuclear pores is color coded.
(B) Side view of (A).
(C) A subregion indicated by the dashed box in (A)
shows a zoomed in view of multiple nuclear pores.
(D) Overview of a 3D reconstruction of the nucleus
obtained by combining nine optical sections.
(E) A section of the reconstruction in (D) confirms
that the labeling is largely limited to the nuclear
envelope.
(F) Different view of the same section.
(G and H) Bottom (G) and top (H) half of the nucleus
shown in (D). The images reveal ring-like nuclear
pores on the top and the bottom nuclear envelope
as well as at the sides of the nucleus (arrowheads).
sions may be vesicles (diameter 150–

200 nm) budding from the cilium (Figures
6F–6H), as ectosomes have been re-
ported to bud from some cilia (Wood
and Rosenbaum, 2015).
Resolving Synaptonemal
Complexes in Whole-Mouse
Spermatocytes
As a final demonstration of the capacity of
our instrument to image deep into cells as
thick as 10 mm, we stained synaptonemal
complexes in mouse spermatocyte nuclei
in the pachytene phase of meiotic pro-
phase (Figure 7; Movies S6). While synap-
tonemal complexes have been imaged
using structured illumination (Carlton,
2008; Qiao et al., 2012) and 4Pi micro-
diameter from 160 to 280 nm (Figures 6A–6E). Our W-4PiSMSN scopy at 100- to 200-nm resolution (Fritsche et al., 2012), higher
images of the ciliary membrane allow us to precisely measure the resolution optical images have been limited to chromosome
cilium’s diameter along its entire length. Interestingly, we find spreads of <1-mm thickness (Schücker et al., 2015). Here,
that cilia diameters are not always constant. Rather, one we examined spermatocytes harvested from testes of 17- to
example cilium shows an abrupt contraction of 50 nm midway 18-day-old mice with W-4PiSMSN and imaged the twisting
along its length (Figures 6C–6E; Movie S5). We speculate this band of the paired lateral elements of autosomal synaptonemal
change in diameter may correspond to the thinning of the 9+0 complexes, highlighted by immunolabeling synaptonemal
microtubule axoneme, which is known to transition from triplet complex protein 3 (SYCP3), a constituent component of the
microtubules, to doublets, and eventually singlets. The ciliary lateral elements (Page and Hawley, 2004). Reconstructed from
tip is not narrow but has a bulbous shape, consistent with struc- a total of 126 optical sections (21 depth positions imaged in six
tures observed in EM (He et al., 2014; Wang et al., 2013). The repetition cycles), the entire 3D image spanned nearly 9 mm in
high-resolution reconstruction of the ciliary membrane also al- depth and resolved SYCP3 substructure of the individual auto-
lowed us to ‘‘unwrap’’ the membrane tube into a flat surface (Fig- somal synaptonemal complexes with unprecedented clarity in-
ure 6H; Supplemental Information) permitting data quantification dependent of their orientation or depth (Figures 7A–7E; Movie
such as cluster analysis and co-localization measurement in a S6). Furthermore, 19 synaptonemal complexes representing
complex geometry. Next, we examined the local density of mol- pairs of individual autosomal homologs could be isolated using
ecules along the ciliary membrane to identify regions with higher a Euclidian metric-based clustering algorithm on the individual
concentrations of pH-SMO. Higher local density is present single-molecule localizations (Supplemental Information). Thus,
around the base, on small bulbous protrusions, and on stripes our approach promises the capacity to visualize the nanoscale
along the cilium length (Figures 6F–6H; Figure S7). These protru- spatial organization of chromosomal scaffolds in the context of
Cell 166, 1028–1040, August 11, 2016 1033

Figure 5. W-4PiSMSN Resolves Individual COPI-Coated Vesicles
COPI complexes were immunolabeled with an antibody against b0 COP and imaged with Alexa Fluor 647 in BSC-1 cells.
(A) Overview of a region of the field of view, with axial location of molecules color coded.
(B and C) Top (B) and side (C) view of the blue-boxed subregion indicated in (A) showing that COPI often forms round and hollow sphere-like structures. Dark gray
and light gray arrowheads indicate the same COPI structures.
(D) x-y view of the area devoid of COPI as indicated by the yellow box in (A).
(E and F) x-z and y-z view of the orange (E) and magenta (F) boxed regions shown in (D) show that COPI surrounds an area presumably containing the Golgi
cisternae.
(G–N) COPI vesicle structures as indicated by the respective labels in (B)–(F) shown at the same enlarged scale reveal circular structures.
1034 Cell 166, 1028–1040, August 11, 2016

Figure 6. GPCR Smoothened on a Primary
Cilium
(A and B) Top (A) and side (B) views of a primary
cilium on an hTERT-RPE1 cell expressing pH-
SMO, which was immunolabeled with Alexa Fluor
647.
(C and E) Views of sections close to the tip (C) and
the base (E) as indicated by the light green and
orange boxes in (A) show the localization of pH-
SMO to the cilium membrane.
(D) Radius of different sections of the cilium as a
function of their distance from the base.
(F) Overview of a cilium in another region of the
sample, showing vesicle-like buds on the ciliary
membrane surface (arrowheads). The inset shows
the local density of the boxed region, which sug-
gests a helical stripe organization of pH-SMO (ar-
rowheads in inset).
(G and H) A bud-like profile shown in (F) can be
unwrapped as depicted in (G), showing the height
of the vesicle above the cilium membrane and the
high molecular density of pH-SMO at the bud (H).
depth improved 10-fold over state-of-

the-art iPALM and 4Pi-SMSN. This devel-
opment extends the application range of
4Pi-based SMSN dramatically: imaging
is no longer limited to features within small
sub-volumes of cells. Instead, we are
capable of imaging organelles that span
large volumes, exemplified by the mito-
chondrial network, the nuclear envelope,
and synaptonemal complexes, which
we capture in virtual entirety. Thus,
W-4PiSMSN is a versatile and powerful
tool that promises a new perspective on
how proteins distribute across entire
organelles throughout whole cells, a key
unmet challenge in cell biology.
Is there room to further increase the
spatial resolution of SMS nanoscopy?
First, SMS resolution depends on the
precision with which one can localize
blinking molecules. The precision is
approximately proportional to the sharp-
ness of the PSF and, for negligible back-
ground noise, is inversely proportional to
the square root of the number of de-
tected photons. Our approach has
architectural elements of the nucleus, many of which are lost in focused mainly on creating the sharpest PSF and detecting
commonly used spread chromatin preparations. as much of the emitted fluorescence light as possible.
Recently, there have been promising developments that in-
DISCUSSION crease the number of emitted photons per molecule (Klehs
et al., 2014; Ong et al., 2015; Vaughan et al., 2012), which we
Through a confluence of several technological innovations, we have not exploited here. Unfortunately, these advances have
have demonstrated that W-4PiSMSN provides unprecedented so far come at the expense of an increase in recording time.
access to the ultrastructure of cells with 10- to 20-nm isotropic We anticipate, however, that with new generations of fluoro-
resolution throughout their entire volume. This resolution is phores or refined imaging buffers, these approaches can
20–50 times higher than conventional microscopy with imaging further improve image quality.
Cell 166, 1028–1040, August 11, 2016 1035

Figure 7. W-4PiSMSN of the Synaptonemal Complexes in a Whole-Mouse Spermatocyte
(A) Overview reconstructed from 21 optical sections. Lateral elements of the synaptonemal complex, spaced 200 nm apart, are resolved throughout the 9-mm
depth of the spermatocyte at uniform resolution.
(B and C) Different views from locations inside the spermatocyte centered on top (B) and bottom (C) regions of the dataset.
(D) x-z view of (A).
(E) The 19 synaptonemal complexes from an entire mouse spermatocyte haploid genome were computationally isolated using a Euclidian distance-based
clustering algorithm.
(F) A conventional image of the 19th synaptonemal complex in x-z view.
Scale bars in (E-19) and (F), 2 mm.
Second, image quality, or spatial resolution, of SMSN images a tubule by mentally extrapolating from the expected tubular
depends on the density of localized molecules (Patterson et al., structure. We utilized several computational image processing
2010; Shroff et al., 2008). The application examples we pre- techniques, including particle averaging of the bacteriophage
sented demonstrate that small features such as cylinder-shaped data (Figure 1), clustering of the synaptonemal complexes
immunolabeled microtubules (40-nm diameter) or ER tubules (Figure 7; Movie S6), and ‘‘unwrapping’’ of cilia (Figure 6), to
(60-nm diameter) can now be easily resolved in 3D using light reconstruct complex structures. These approaches, which add
microscopy. This image interpretation is aided by the fact that constraints to data interpretation (e.g., the fact that cilia are
the observer fills the gaps in the distribution of molecules along tubular) and can be tailored to the cell biological question at
1036 Cell 166, 1028–1040, August 11, 2016

hand, allow us to extract structural details, which are more In short, 25-mm-diameter coverslips were cleaned by sonication in 1 M KOH
subtle than the labeling density suggests. Ultimately, labeling for 15 min before use. Fluorescent 100-nm-diameter crimson beads were
attached to the coverslip surface using poly-L-lysine. Cultured mammalian
density is limited by the density of probe targets, usually pro-
cells were grown on coverslips for 24–48 hr before fixation. COS-7 cells
teins. The development of new labeling approaches that allow were used for microtubule, ER, and mitochondria samples. RPE-hTERT cells
membrane targeting (Erdmann et al., 2014) overcome this re- were used for nuclear pore complex and cilia samples. BSC1 cells were
striction and additionally offer the possibility of revealing the used for COPI samples. For synaptonemal complex samples, spermatocytes
membrane boundaries of individual organelles. Complementary were isolated from the testes of mice and settled on coverslips before being
approaches utilizing stochastic transient chemical binding fixed. T7 bacteriophages were isolated from E. coli cultures and labeled with
can further address the limited pool of fluorescent labels and Alexa Fluor 647 NHS Ester before being placed on coverslips.
Microtubule, mitochondria, and ER samples were fixed with 3% paraformal-
theoretically allow unlimited numbers of localized molecules
dehyde (PFA) + 0.1% glutaraldehyde before antibody labeling. A saponin pre-
(Giannone et al., 2010; Jungmann et al., 2014; Lew et al., 2011; extraction step preceded fixation when only microtubules were labeled.
Sharonov and Hochstrasser, 2006). However, even with the Nuclear Pore Complex samples were fixed in 20 C methanol. COPI and syn-
use of conventional labeling techniques, W-4PiSMSN is capable aptonemal samples were fixed in 4% PFA.
of visualizing otherwise inaccessible structures in a multitude of Antibodies against endogenous proteins were used to label microtubules
settings as demonstrated by the large range of presented (anti-a-tubulin), mitochondria (anti-TOM20), nuclear pore complexes (anti-
Nup358), COPI (anti-b0 COP), and synaptonemal complexes (anti-SYCP3).
applications.
Overexpressed proteins were labeled with antibodies in ER samples (mEmer-
In conclusion, we believe that the development of W-4PiSMSN ald-Sec61b using anti-GFP) and Cilia samples (pHlourin-mSmoothened using
represents the culmination of more than a decade’s research on anti-GFP).
high-resolution fluorescence imaging techniques and establishes All one-color samples were labeled with Alexa Fluor 647, either using a com-
3D biological imaging with molecular specificity and resolution in mercial secondary antibody or an NHS ester. Two-color microtubule and mito-
the 10-nm range as a general imaging technique. chondria samples were labeled with Alexa Flour 647 and Cy3B, respectively.
Cy3B-labeled secondary antibodies were made by conjugating reactive
Cy3B with unlabeled antibodies. After secondary antibody labeling, COS-7
EXPERIMENTAL PROCEDURES and BSC1 cell samples were post-fixed with 3% PFA + 0.1% GA.
Labeled biological samples were placed in an aluminum sample frame and
Microscope Setup covered with a second cleaned coverslip. A thin spacer and imaging buffer
A detailed description is provided as Supplemental Information. In brief, the was placed between the two coverslips. The coverslips were held in the sam-
4Pi cavity of the system was set up vertically around two opposing high-NA ple frame using an addition-curing silicone. The imaging buffer was either con-
objective lenses (Movie S7). Excitation light from three laser lines at wave- ventional or COT containing thiol buffer (Supplemental Information). Samples
lengths of 642, 561, and 405 nm was coupled into the upper objective for were imaged immediately after the silicone solidified.
wide-field illumination. Following the concept by Aquino et al. (2011), fluores-
cence was collected by both objectives and passed through quarter wave
W-4PiSMSN Data Acquisition
plates, which enforced equal fractions of s and p-polarized light independent
Four phase images are arranged along the splitting line of the sCMOS cam-
of the dipole emitter orientation. A custom Babinet-Soleil compensator
era’s upper and lower readout region. 50,000 to 320,000 camera frames
corrected for dispersion and allowed adjusting the phase delay between the
were recorded at 50 or 100 fps, resulting in 10 min to 1.5 hr total acquisition
upper and lower interferometer arm independently for the two polarization
time per dataset. For sample volumes thinner than 1.2 mm, the sample stage
components before the light was combined at a 50/50 beam splitter cube.
was not translated during data acquisition. For thicker samples, the stage
We added deformable mirrors (Boston Micromachines, Multi-5.5) in planes
was moved axially in 500-nm steps every 1,000–3,000 frames until the whole
conjugate to the back pupils of the objectives, which allowed for aberration
targeted imaging volume was covered, resulting in up to 21 z-steps and imag-
correction, optimization of the PSF and introduction of astigmatism for arti-
ing volume depths of up to 9 mm. This axial scan was automatically repeated
fact-free 3D localization. s and p-polarized fluorescence exiting the beam
six to 19 times, and the data from the scans were combined. The laser intensity
splitter cube at two sides was split with a polarizing beam splitter cube into
was manually adjusted during each experiment to optimize the emitter density
four images featuring different interference phases. The four images were re-
per frame and to maximize detectable emissions starting at intensities as high
corded simultaneously by a scientific complementary metal-oxide semicon-
as 35 kW/cm2 to transfer emitters efficiently into dark states and decreasing to
ductor (sCMOS) camera (Hamamatsu, ORCA-Flash 4.0v2). The entire system typically 5 kW/cm2 near the end of the acquisition when the pool of blinking
was controlled by a custom-written program in LabVIEW. molecules had declined (Lin et al., 2015). Additionally, the laser intensity of
the 405-nm laser was manually controlled over the course of imaging to opti-
Aberration Correction Using Deformable Mirrors mize the active emitter density.
Two deformable mirrors were independently adjusted as follows. For each
interfering arm, starting from the flat voltage map (provided by the manufac- Single-Molecule Interference Phase Estimation
turer), 28 theoretically generated membrane modes were applied sequentially Raw camera frames were first isolated into four different phase images. The
with ten different voltage amplitudes per mode. The detected peak signals (0th four phase images were then merged into a single image using a transforma-
moment Gaussian weighted sum) of a single fluorescent bead in focus were tion matrix obtained from a combination of algorithms using log-polar and
extracted for all amplitudes of the applied modes. The optimal amplitude for affine transformations (Supplemental Information). Single-molecule candi-
each mode was determined as the value that gave the maximum signal level dates from the merged frames were isolated and fitted with an elliptical
by fitting a quadratic function to the measurements. These newly obtained am- Gaussian model using a maximum likelihood estimator accounting for the
plitudes were added to the flat voltage map and serve as a starting point of camera-specific noise associated with sCMOS cameras (Huang et al., 2013)
another iteration. We used five iterations to achieve optimal system aberration (Supplemental Information). Estimates of single-molecule positions, width, to-
correction. Details are provided the Supplemental Information. tal number of detected photons, background, and log-likelihood ratio were ob-
tained. Intensities of the 0th moment Gaussian (Supplemental Information)
Biological Sample Preparation were calculated by a weighted least-square fitting of a Gaussian with the
A complete description of cell culture, fluorescence labeling, coverslip, and amplitude being the only fitting parameter and weighted to take the sCMOS
buffer preparation is included in sections 15 to 26 of the Supplemental noise into account (Huang et al., 2013). Subsequently, the reduced moments
Information. of each polarization were extracted using a previously described approach
Cell 166, 1028–1040, August 11, 2016 1037

(Aquino et al., 2011). For two-color imaging, phase shifts between s and p code. F.H., G.S., E.S.A., L.K.S., W.C.D., E.B.K., Y.Z., and J.B. optimized and
polarization for the two color channels differ by 0.3 radians. These phase shift tested the microscope. L.K.S., W.C.D., F.E.R.-M., J.R.M., I.I., M.L., M.A.H.,
differences were measured independently from images of fluorescent beads C.J.-W., C.P.L., J.E.R., D.T., and J.B. designed biological experiments. F.H.,
recorded in two detection channels. With the obtained phase differences be- G.S., L.K.S., W.C.D., F.E.R.-M., J.R.M., I.I., and M.L. optimized sample prep-
tween the s and p polarization, the phase shifts between the four different aration protocols and prepared samples. F.H., G.S., E.S.A., L.K.S., E.B.K.,
phase images are known. Subsequently, the interference phases of the de- T.P., and J.B. visualized the data. All authors contributed to writing the
tected molecules were obtained by solving a set of equations describing the manuscript.
0th moment intensities in s and p polarization channels as functions of an un-
known offset and their relative phase shifts as characterized above (Supple-
ACKNOWLEDGMENTS
mental Information).
We would like to thank Travis Gould (Bates College) for initial discussion of the
Axial Position Analysis with Ridge-Finding Algorithm instrument design, Hideo Takakura, Hoong Chuin Lim, Stephanie Baguley
We developed a metric m: ðs3x =sy Þ ðs3y =sx Þ, where sx and sy are the esti- (Yale University), and Yasuhiro Fujiwara (The Jackson Laboratory) for help in
mated SDs of the 2D Gaussian for a single-molecule emission pattern. The sample preparation, and Wen Jiang (Purdue University) for providing purified
metric preserves its monotonicity in the presence of aberrations, and we T7 phage samples. We thank Elias Coutavas and Günter Blobel (Rockefeller
used it to estimate an unambiguous (but still rough) position of each single University) for the Nup358 antibody. We thank Mike Mosbacher and Christo-
emitter, before the phase estimate was used to pinpoint the exact axial posi- pher Jost for help on early CAD design and thermal simulations of the instru-
tion of a molecule. ment, and Markus Schwarzhuber, Florian Schroeder, Stefan Politzka, Tobias
For every 3,000–5,000 camera frames, we generated a 2D histogram image Wagner, and Robert Muth for help on mechanical part design, custom elec-
from phase estimations and the metric m from all detected single molecules. tronics, sample chamber optimization, and data visualization. We thank
As m is monotonic and the single-molecule phase is periodic with a period George Takahashi from the Purdue Envision Center for help on data visualiza-
of 2p, the resulting 2D histogram looks like a pattern of tilted repeating stripes tion. We thank Michael Mlodzianoski for discussions on the manuscript. This
(Figure S3). We developed a ridge-finding algorithm to determine a series of work was primarily supported by a grant from the Wellcome Trust (095927/
connected vectors defining the correspondence between the determined A/11/Z). F.H. received funding from a James Hudson Brown–Alexander Brown
phase values and the values of m. This allowed us to unwrap the periodic Coxe Postdoctoral Fellowship. E.B.K. was supported by The Denmark-Amer-
phase for unambiguous axial localization. A detailed description of the algo- ica Foundation (Coloplast), Civilingeniør Frants Allings Legat, Knud Højgaards
rithm is provided in the Supplemental Information. Fond, Reinholdt W. Jorcks Fond, Berg Nielsens Legat, and Ingeniør Alexandre
Haynman og Hustru Nina Haynmans Fond. G.S. and E.S.A. were supported by
Drift Correction the Wellcome Trust (092096) and Cancer Research UK (C6946/A14492).
3D drift correction was performed by first calculating the distance pairs be- F.E.R-M. and D.T. were supported by a grant from the NIH (R21 HD078851-
tween image segments (3,000–5,000 frames), and subsequently forming three 01). I.I. was supported by a grant from the NIH (R01 GM065835) to C.J.-W.
sets of equations for x, y, and z, respectively (Li et al., 2013; Wang et al., 2014). C.J.-W. is an investigator of the Howard Hughes Medical Institute. M.A.H
A least-square solution that minimizes the overall error of the set of equations was supported by a grant from NIH (P01 GM99640). J.R.M. and C.P.L. were
was obtained and back-substituted into all equations. Errors can be calculated supported by grants from the NIH (R01 GM105672 and R01 HL124402). J.B.
from each of these equations and a specific equation within the set is removed declares significant financial interest in Bruker, which manufactures SMSN in-
from the stack if its error is larger than 7 nm. This process was repeated until no struments. F.H. and J.B. have filed a patent application for sCMOS-related
error was larger than 7 nm or the matrix was no longer at its full rank. This al- data analysis (WO 2014144443 A2). The invention and software is licensed
lowed us to correct system and sample induced drift in 3D with short segments to Hamamatsu Photonics K.K.
of data (Supplemental Information).
Received: February 29, 2016
Optical Segment Alignment Revised: May 2, 2016
To image thick samples, optical sections were recorded at different axial posi- Accepted: June 3, 2016
tions of the sample by axially translating the z-piezo holding the sample stage. Published: July 7, 2016
The localization data contain x, y, and z position estimates of different optical
sections and must be aligned/stitched seamlessly to support the high precision REFERENCES
obtained in W-4PiSMSN. Previous methods (Huang et al., 2008) that shifted
each optical section by a constant in the axial direction have been prone to intro- Aquino, D., Schönle, A., Geisler, C., Middendorff, C.V., Wurm, C.A., Okamura,
duce misalignment of the optical sections and subsequently deteriorate the res- Y., Lang, T., Hell, S.W., and Egner, A. (2011). Two-color nanoscopy of three-
olution achievable in thick samples. Here, we developed an optical alignment dimensional volumes by 4Pi detection of stochastically switched fluorophores.
method based on 3D cross-correlation. In the W-4PiSMSN system, optical sec- Nat. Methods 8, 353–359.
tions are 1.2 mm thick. Whole-cell samples were scanned in the axial direction
Betzig, E., Patterson, G.H., Sougrat, R., Lindwasser, O.W., Olenych, S., Boni-
with 500-nm step sizes. This allowed for abundant overlapping regions between
facino, J.S., Davidson, M.W., Lippincott-Schwartz, J., and Hess, H.F. (2006).
adjacent optical sections and provided critical information for precise optical
Imaging intracellular fluorescent proteins at nanometer resolution. Science
section alignment using the developed 3D cross-correlation methods as
313, 1642–1645.
described in Supplemental Information.
Bewersdorf, J., Schmidt, R., and Hell, S.W. (2006). Comparison of I5M and
4Pi-microscopy. J. Microsc. 222, 105–117.
Briggs, J.A.G. (2013). Structural biology in situ–the potential of subtomogram
Supplemental Information includes Supplemental Experimental Procedures, averaging. Curr. Opin. Struct. Biol. 23, 261–267.
seven figures, and seven movies and can be found with this article online at Broeken, J., Johnson, H., Lidke, D.S., Liu, S., Nieuwenhuizen, R.P.J., Stallinga,
http://dx.doi.org/10.1016/j.cell.2016.06.016. S., Lidke, K.A., and Rieger, B. (2015). Resolution improvement by 3D particle
averaging in localization microscopy. Methods Appl. Fluoresc. 3, 014003.
AUTHOR CONTRIBUTIONS Brown, T.A., Tkachuk, A.N., Shtengel, G., Kopek, B.G., Bogenhagen, D.F.,
Hess, H.F., and Clayton, D.A. (2011). Superresolution fluorescence imaging
F.H., G.S., E.S.A., M.J.B., and J.B. designed the microscope. F.H., G.S., of mitochondrial nucleoids reveals their spatial range, limits, and membrane
E.S.A., T.P., M.J.B., and J.B. built the instrument and developed the computer interaction. Mol. Cell. Biol. 31, 4994–5010.
1038 Cell 166, 1028–1040, August 11, 2016

Burke, D., Patton, B., Huang, F., Bewersdorf, J., and Booth, M.J. (2015). Adap- Klehs, K., Spahn, C., Endesfelder, U., Lee, S.F., Fürstenberg, A., and Heile-
tive optics correction of specimen-induced aberrations in single-molecule mann, M. (2014). Increasing the brightness of cyanine fluorophores for sin-
switching microscopy. Optica 2, 177. gle-molecule and superresolution imaging. ChemPhysChem 15, 637–641.
Carlton, P.M. (2008). Three-dimensional structured illumination microscopy Lew, M.D., Lee, S.F., Ptacin, J.L., Lee, M.K., Twieg, R.J., Shapiro, L., and
and its application to chromosome structure. Chromosome Res. 16, 351–365. Moerner, W.E. (2011). Three-dimensional superresolution colocalization of
Case, L.B., Baird, M.A., Shtengel, G., Campbell, S.L., Hess, H.F., Davidson, intracellular protein superstructures and the cell surface in live Caulobacter
M.W., and Waterman, C.M. (2015). Molecular mechanism of vinculin activation crescentus. Proc. Natl. Acad. Sci. USA 108, E1102–E1110.
and nanoscale spatial organization in focal adhesions. Nat. Cell Biol. 17, Li, X., Mooney, P., Zheng, S., Booth, C.R., Braunfeld, M.B., Gubbens, S.,
880–892. Agard, D.A., and Cheng, Y. (2013). Electron counting and beam-induced mo-
Chojnacki, J., Staudt, T., Glass, B., Bingen, P., Engelhardt, J., Anders, M., tion correction enable near-atomic-resolution single-particle cryo-EM. Nat.
Schneider, J., Müller, B., Hell, S.W., and Kräusslich, H.-G. (2012). Matura- Methods 10, 584–590.
tion-dependent HIV-1 surface protein redistribution revealed by fluorescence Lichtman, J.W., and Conchello, J.-A. (2005). Fluorescence microscopy. Nat.
nanoscopy. Science 338, 524–528. Methods 2, 910–919.
Erdmann, R.S., Takakura, H., Thompson, A.D., Rivera-Molina, F., Allgeyer, Lin, Y., Long, J.J., Huang, F., Duim, W.C., Kirschbaum, S., Zhang, Y.,
E.S., Bewersdorf, J., Toomre, D., and Schepartz, A. (2014). Super-resolution Schroeder, L.K., Rebane, A.A., Velasco, M.G.M., Virrueta, A., et al. (2015).
imaging of the Golgi in live cells with a bioorthogonal ceramide probe. Angew. Quantifying and optimizing single-molecule switching nanoscopy at high
Chem. Int. Ed. Engl. 53, 10242–10246. speeds. PLoS ONE 10, e0128135.
Frey, T.G., and Mannella, C.A. (2000). The internal structure of mitochondria. Liu, K.S.Y., Siebert, M., Mertel, S., Knoche, E., Wegener, S., Wichmann, C.,
Trends Biochem. Sci. 25, 319–324. Matkovic, T., Muhammad, K., Depner, H., Mettke, C., et al. (2011). RIM-bind-
Fritsche, M., Reinholdt, L.G., Lessard, M., Handel, M.A., Bewersdorf, J., and ing protein, a central part of the active zone, is essential for neurotransmitter
Heermann, D.W. (2012). The impact of entropy on the spatial organization of release. Science 334, 1565–1569.
synaptonemal complexes within the cell nucleus. PLoS ONE 7, e36282.
Liu, S., Kromann, E.B., Krueger, W.D., Bewersdorf, J., and Lidke, K.A. (2013).
Giannone, G., Hosy, E., Levet, F., Constals, A., Schulze, K., Sobolevsky, A.I., Three dimensional single molecule localization using a phase retrieved pupil
Rosconi, M.P., Gouaux, E., Tampé, R., Choquet, D., and Cognet, L. (2010). Dy- function. Opt. Express 21, 29462–29487.
namic superresolution imaging of endogenous proteins on living cells at ultra-
McGorty, R., Schnitzbauer, J., Zhang, W., and Huang, B. (2014). Correction of
high density. Biophys. J. 99, 1303–1310.
depth-dependent aberrations in 3D single-molecule localization and super-
Gould, T.J., Hess, S.T., and Bewersdorf, J. (2012a). Optical nanoscopy: from resolution microscopy. Opt. Lett. 39, 275–278.
acquisition to analysis. Annu. Rev. Biomed. Eng. 14, 231–254.
Mlodzianoski, M.J., Schreiner, J.M., Callahan, S.P., Smolková, K., Dlasková,
Gould, T.J., Burke, D., Bewersdorf, J., and Booth, M.J. (2012b). Adaptive op- A., Santorová, J., Jezek, P., and Bewersdorf, J. (2011). Sample drift correction
tics enables 3D STED microscopy in aberrating specimens. Opt. Express 20, in 3D fluorescence photoactivation localization microscopy. Opt. Express 19,
20998–21009. 15009–15019.
He, M., Subramanian, R., Bangs, F., Omelchenko, T., Liem, K.F., Jr., Kapoor, Nieuwenhuizen, R.P.J., Lidke, K.A., Bates, M., Puig, D.L., Grünwald, D., Stal-
T.M., and Anderson, K.V. (2014). The kinesin-4 protein Kif7 regulates mamma- linga, S., and Rieger, B. (2013). Measuring image resolution in optical nano-
lian Hedgehog signalling by organizing the cilium tip compartment. Nat. Cell scopy. Nat. Methods 10, 557–562.
Biol. 16, 663–672.
Ong, W.Q., Citron, Y.R., Schnitzbauer, J., Kamiyama, D., and Huang, B. (2015).
Hell, S.W. (2007). Far-field optical nanoscopy. Science 316, 1153–1158.
Heavy water: a simple solution to increasing the brightness of fluorescent pro-
Hell, S.W., and Wichmann, J. (1994). Breaking the diffraction resolution limit teins in super-resolution imaging. Chem. Commun. (Camb.) 51, 13451–13453.
by stimulated emission: stimulated-emission-depletion fluorescence micro-
Orci, L., Stamnes, M., Ravazzola, M., Amherdt, M., Perrelet, A., Söllner, T.H.,
scopy. Opt. Lett. 19, 780–782.
and Rothman, J.E. (1997). Bidirectional transport by distinct populations of
Hess, S.T., Girirajan, T.P.K., and Mason, M.D. (2006). Ultra-high resolution im- COPI-coated vesicles. Cell 90, 335–349.
aging by fluorescence photoactivation localization microscopy. Biophys. J.
Page, S.L., and Hawley, R.S. (2004). The genetics and molecular biology of the
91, 4258–4272.
synaptonemal complex. Annu. Rev. Cell Dev. Biol. 20, 525–558.
Hu, B., Margolin, W., Molineux, I.J., and Liu, J. (2013). The bacteriophage t7
virion undergoes extensive structural remodeling during infection. Science Patterson, G., Davidson, M., Manley, S., and Lippincott-Schwartz, J. (2010).
339, 576–579. Superresolution imaging using single-molecule localization. Annu. Rev.
Phys. Chem. 61, 345–367.
Huang, B., Jones, S.A., Brandenburg, B., and Zhuang, X. (2008). Whole-cell 3D
STORM reveals interactions between cellular structures with nanometer-scale Pavani, S.R.P., Thompson, M.A., Biteen, J.S., Lord, S.J., Liu, N., Twieg, R.J.,
resolution. Nat. Methods 5, 1047–1052. Piestun, R., and Moerner, W.E. (2009). Three-dimensional, single-molecule
fluorescence imaging beyond the diffraction limit by using a double-helix point
Huang, F., Hartwich, T.M.P., Rivera-Molina, F.E., Lin, Y., Duim, W.C., Long,
spread function. Proc. Natl. Acad. Sci. USA 106, 2995–2999.
J.J., Uchil, P.D., Myers, J.R., Baird, M.A., Mothes, W., et al. (2013). Video-
rate nanoscopy using sCMOS camera-specific single-molecule localization Pellett, P.A., Dietrich, F., Bewersdorf, J., Rothman, J.E., and Lavieu, G. (2013).
algorithms. Nat. Methods 10, 653–658. Inter-Golgi transport mediated by COPI-containing vesicles carrying small
cargoes. eLife 2, e01296.
Juette, M.F., Gould, T.J., Lessard, M.D., Mlodzianoski, M.J., Nagpure, B.S.,
Bennett, B.T., Hess, S.T., and Bewersdorf, J. (2008). Three-dimensional Qiao, H., Chen, J.K., Reynolds, A., Höög, C., Paddy, M., and Hunter, N. (2012).
sub-100 nm resolution fluorescence microscopy of thick samples. Nat. Interplay between synaptonemal complex, homologous recombination, and
Methods 5, 527–529. centromeres during mammalian meiosis. PLoS Genet. 8, e1002790.
Jungmann, R., Avendaño, M.S., Woehrstein, J.B., Dai, M., Shih, W.M., and Yin, Ram, S., Prabhat, P., Chao, J., Ward, E.S., and Ober, R.J. (2008). High accu-
P. (2014). Multiplexed 3D cellular super-resolution imaging with DNA-PAINT racy 3D quantum dot tracking with multifocal plane microscopy for the study of
and Exchange-PAINT. Nat. Methods 11, 313–318. fast intracellular dynamics in live cells. Biophys. J. 95, 6025–6043.
Kanchanawong, P., Shtengel, G., Pasapera, A.M., Ramko, E.B., Davidson, Rust, M.J., Bates, M., and Zhuang, X. (2006). Sub-diffraction-limit imaging by
M.W., Hess, H.F., and Waterman, C.M. (2010). Nanoscale architecture of in- stochastic optical reconstruction microscopy (STORM). Nat. Methods 3,
tegrin-based cell adhesions. Nature 468, 580–584. 793–795.
Cell 166, 1028–1040, August 11, 2016 1039

Schücker, K., Holm, T., Franke, C., Sauer, M., and Benavente, R. (2015). Eluci- von Middendorff, C., Egner, A., Geisler, C., Hell, S.W., and Schönle, A. (2008).
dation of synaptonemal complex organization by super-resolution imaging Isotropic 3D Nanoscopy based on single emitter switching. Opt. Express 16,
with isotropic resolution. Proc. Natl. Acad. Sci. USA 112, 2029–2033. 20774–20788.
Sharonov, A., and Hochstrasser, R.M. (2006). Wide-field subdiffraction imag- Wang, W.-J., Tay, H.G., Soni, R., Perumal, G.S., Goll, M.G., Macaluso, F.P.,
ing by accumulated binding of diffusing probes. Proc. Natl. Acad. Sci. USA Asara, J.M., Amack, J.D., and Tsou, M.-F.B. (2013). CEP162 is an axoneme-
103, 18911–18916. recognition protein promoting ciliary transition zone assembly at the cilia
Shroff, H., Galbraith, C.G., Galbraith, J.A., and Betzig, E. (2008). Live-cell pho- base. Nat. Cell Biol. 15, 591–601.
toactivated localization microscopy of nanoscale adhesion dynamics. Nat. Wang, Y., Schnitzbauer, J., Hu, Z., Li, X., Cheng, Y., Huang, Z.-L., and Huang,
Methods 5, 417–423. B. (2014). Localization events-based sample drift correction for localization mi-
Shtengel, G., Galbraith, J.A., Galbraith, C.G., Lippincott-Schwartz, J., Gillette, croscopy with redundant cross-correlation algorithm. Opt. Express 22,
J.M., Manley, S., Sougrat, R., Waterman, C.M., Kanchanawong, P., Davidson, 15982–15991.
M.W., et al. (2009). Interferometric fluorescent super-resolution microscopy
Wood, C.R., and Rosenbaum, J.L. (2015). Ciliary ectosomes: transmissions
resolves 3D cellular ultrastructure. Proc. Natl. Acad. Sci. USA 106, 3125–3130.
from the cell’s antenna. Trends Cell Biol. 25, 276–285.
Van Engelenburg, S.B., Shtengel, G., Sengupta, P., Waki, K., Jarnik, M., Ablan,
Xu, K., Babcock, H.P., and Zhuang, X. (2012). Dual-objective STORM reveals
S.D., Freed, E.O., Hess, H.F., and Lippincott-Schwartz, J. (2014). Distribution
three-dimensional filament organization in the actin cytoskeleton. Nat.
of ESCRT machinery at HIV assembly sites reveals virus scaffolding of ESCRT
Methods 9, 185–188.
subunits. Science 343, 653–656.
Vaughan, J.C., Jia, S., and Zhuang, X. (2012). Ultrabright photoactivatable flu- Xu, K., Zhong, G., and Zhuang, X. (2013). Actin, spectrin, and associated pro-
orophores created by reductive caging. Nat. Methods 9, 1181–1184. teins form a periodic cytoskeletal structure in axons. Science 339, 452–456.
von Appen, A., Kosinski, J., Sparks, L., Ori, A., DiGuilio, A.L., Vollmer, B., Yang, T.T., Hampilos, P.J., Nathwani, B., Miller, C.H., Sutaria, N.D., and Liao,
Mackmull, M.-T., Banterle, N., Parca, L., Kastritis, P., et al. (2015). In situ struc- J.-C. (2013). Superresolution STED microscopy reveals differential localization
tural analysis of the human nuclear pore complex. Nature 526, 140–143. in primary cilia. Cytoskeleton 70, 54–65.
von Diezmann, A., Lee, M.Y., Lew, M.D., and Moerner, W.E. (2015). Correcting Yang, T.T., Su, J., Wang, W.-J., Craige, B., Witman, G.B., Tsou, M.F., and Liao,
field-dependent aberrations with nanoscale accuracy in three-dimensional J.-C. (2015). Superresolution pattern recognition reveals the architectural map
single-molecule localization microscopy. Optica 2, 985–993. of the ciliary transition zone. Sci. Rep. 5, 14096.
1040 Cell 166, 1028–1040, August 11, 2016

Cr eat ing a Selec t ion t o Fi t Your Needs
Scissors - Retractors - Magnifiers - Probes & Hooks - Bone Instruments - Animal Identification
Hemostats - Forceps - Surgical & Laboratory Equipment - Feeding Needles - Spatulae & Spoons
Wound Closure - Surgical Plates - Instrument Care & Sterilization - Rongeurs - Scalpels & Knives
Clamps - Pins & Holders - Needles & Needle Holders - Student Quality Instruments & Much More
F I N E S U R G I C A L I N S T R U M E N T S F O R R E S E A R C H TM
Visit us at finescience.com or call 800 521 2109
Cell_Ad_FullPage_BRONZE_8.375x10.875_final.indd 1 11/22/16 12:47 PM

CAN’T FIND THE MOUSE MODEL YOU NEED?
BUILD IT WITH
CRISPR!
From simple indel KOs to more complex conditional KOs and reporter insertions,
we’ve created a wide range of models using CRISPR/ Cas9 technology.
If we aren’t able to generate your CRISPR model, you don’t pay.
TO LEARN MORE,
VISIT JAX.ORG/CRISPR.
1.800.422.6423
1.207.288.5845

Publication

Uploaded by

Copyright:

Available Formats

Publication

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Publication

Uploaded by

Copyright:

Available Formats

CAN’T FIND THE MOUSE MODEL YOU NEED?

A MAJOR LEAP FORWARD IN PHENOTYPING

XF Cell Energy Phenotype Profile

LEARN MORE at www.agilent.com Quiescent Glycolytic

Extracellular Acidification Rate (ECAR)

BioLegend is ISO 9001:2008 and ISO 13485:2003 Certified

World-Class Quality | Superior Customer Support | Outstanding Value

from Bone Marrow Cultures the frequency of MSCs in BM is 0.001–

following gating strategy. Since P1

50 Sloan Kettering Cancer Center) for their generous

within FL1 dim/neg populations.

1,500 events per second (eps) using CD73+

16-8119_CBD_S3e_MSC_Advertorial_Elsevier_Cell_ASCB_FINAL.indd All Pages 11/17/16 1:34 PM

from Bone Marrow Cultures the frequency of MSCs in BM is 0.001–

following gating strategy. Since P1

50 Sloan Kettering Cancer Center) for their generous

within FL1 dim/neg populations.

1,500 events per second (eps) using CD73+

16-8119_CBD_S3e_MSC_Advertorial_Elsevier_Cell_ASCB_FINAL.indd All Pages 11/17/16 1:34 PM

052.A1.0101.D © 2017 Eppendorf AG.

Full page journal ad Cell - CEL

Expedite Your Research

• Outsource troubleshooting and

• Deliver your project on time

Setting the standard in quality

Learn more | rndsystems.com/customservices

Global info@bio-techne.com bio-techne.com/ﬁnd-us/distributors TEL +1 612 379 2956

For information for the Best of Series, please contact:

Your Discovery Awaits.

For research use only. Not for use in diagnostic procedures.

THE NGS EXPERTS™

 Standard Amplification Curves for

16GIP054_PIPETMAXqPCR_ad_Cell_asm1.indd 2 6/20/16 1:36 PM

See More Than Before

There’s so much more! For more information, contact your local

NIK-5959 Ti2 Full Page Ad_R5.indd 1 11/18/16 1:50 PM

More Than an IncuCyte®

Learn more. Visit www.biotek.com/biospasystem

IncuCyte® ZOOM is a registered trademark of Essen BioScience www.biotek.com

BioSpa System ad-Cell Press.indd 1 10/17/16 2:44 PM

Since 1998 Baker Ruskinn have

Nano Culture Plate

Do You Know Where Your Antibodies Are Coming From?

Protect your experiments with

Expanding the Ubiquitin Code Krishnaraj Rajalingam and Ivan Dikic

Epigenomic Assays Martin Krzywinski and Martin Hirst

Signaling in Symbiosis Sharon R. Long

Neuronal Regulation of Aging Heather J. Weir and William B. Mair

Microglia in Disease Simon Beggs and Michael W. Salter

Types of Ubiquitin Ligases Francesca Ester Morreale and Helen Walden

The Basis of Oncoimmunology A. Karolina Palucka and Lisa M. Coussens

Modeling Development and Disease with Organoids Hans Clevers

Articles and Resources

Precision Medicine Series: EACR-AACR-SIC Special Conference 2017:

AACR International Conference Third CRI-CIMT-EATI-AACR International

Learn more and register at

1608100_UpcomingConf_Dec16_Cell_8.375x10.875.indd 1 10/25/16 8:43 AM

Pol II pause, release, elongation rate, and processivity factors

Paused Pol II Super Elongation Complex (SEC) Protein Function

Pol II CTD phosphorylation Histone modifications involved in elongation