The ISME Journal (2014) 8, 187211

& 2014 International Society for Microbial Ecology All rights reserved 1751-7362/14
www.nature.com/ismej

ORIGINAL ARTICLE
Metagenomic analysis of size-fractionated
picoplankton in a marine oxygen minimum zone
Sangita Ganesh1, Darren J Parris1, Edward F DeLong2,3 and Frank J Stewart1
1
School of Biology, Georgia Institute of Technology, Atlanta, GA, USA; 2Department of Civil and
Environmental Engineering, Massachusetts Institute of Technology, Parsons Laboratory 48, Cambridge,
MA, USA and 3Center for Microbial Ecology: Research and Education, Honolulu, Hawaii, USA

Marine oxygen minimum zones (OMZs) support diverse microbial communities with roles in major
elemental cycles. It is unclear how the taxonomic composition and metabolism of OMZ
microorganisms vary between particle-associated and free-living size fractions. We used amplicon
(16S rRNA gene) and shotgun metagenome sequencing to compare microbial communities from
large (41.6 lm) and small (0.21.6 lm) filter size fractions along a depth gradient in the OMZ off
Chile. Despite steep vertical redox gradients, size fraction was a significantly stronger predictor of
community composition compared to depth. Phylogenetic diversity showed contrasting patterns,
decreasing towards the anoxic OMZ core in the small size fraction, but exhibiting maximal values
at these depths within the larger size fraction. Fraction-specific distributions were evident for key
OMZ taxa, including anammox planctomycetes, whose coding sequences were enriched up to
threefold in the 0.21.6 lm community. Functional gene composition also differed between
fractions, with the 41.6 lm community significantly enriched in genes mediating social
interactions, including motility, adhesion, cell-to-cell transfer, antibiotic resistance and mobile
element activity. Prokaryotic transposase genes were three to six fold more abundant in this
fraction, comprising up to 2% of protein-coding sequences, suggesting that particle surfaces may
act as hotbeds for transposition-based genome changes in marine microbes. Genes for nitric and
nitrous oxide reduction were also more abundant (three to seven fold) in the larger size fraction,
suggesting microniche partitioning of key denitrification steps. These results highlight an
important role for surface attachment in shaping community metabolic potential and genome
content in OMZ microorganisms.
The ISME Journal (2014) 8, 187211; doi:10.1038/ismej.2013.144; published online 12 September 2013
Subject Category: Microbial ecology and functional diversity of natural habitats
Keywords: Archaea; bacteria; microbial diversity; OMZ; oxycline; particle

Introduction gradients, microbial taxa and microbial processes

associated with particles, including aggregates of
Marine oxygen minimum zones (OMZs) contain decaying organic matter, as well as live phytoplank-
diverse communities of Bacteria and Archaea whose ton or zooplankton cells (Karl et al., 1984;
metabolisms control key steps in marine biogeo- DeLong et al., 1993; Fenchel, 2002; Stocker, 2012).
chemical cycling. Metagenome and single-gene These particles support complex surface-attached
surveys have identified marked transitions in microbial communities whose composition and life
microbial community composition and metabolism history strategies differ substantially from those of
from oxygenated surface waters to the suboxic OMZ free-living microbes. Although taxonomic surveys
core (Stevens and Ulloa, 2008; Zaikova et al., 2010; have compared free-living and particle-attached
Bryant et al., 2012; Stewart et al., 2012b). These communities in a variety of marine ecosystems, the
shifts have been linked to meter-scale vertical differences in functional gene composition that
gradients in dissolved oxygen and organic and distinguish free-living from surface-attached life
inorganic energy substrate availability. However, histories have been explored only sparingly. For
heterogeneity in the marine water column also OMZs in particular, the microscale partitioning of
potentially exists in the form of microscale chemical microbial communities and metabolisms has not
been explored, despite a potentially significant role
for particle-associated microhabitats in these zones
Correspondence: F Stewart, School of Biology, Georgia Institute of (Whitmire et al., 2009).
Technology, Ford ES and T Building, Room 1242, 311 Ferst Drive, OMZs occur where the aerobic respiration
Atlanta, GA 30332, USA.
E-mail: frank.stewart@biology.gatech.edu of organic matter combines with water column
Received 19 April 2013; revised 21 July 2013; accepted 22 July stabilization to form a persistent, low-oxygen layer.
2013; published online 12 September 2013 The largest and most oxygen-depleted OMZs are
 Size-fractionated metagenomics
S Ganesh et al
188
found in regions of nutrient upwelling, as in the 2012). Relative to free-living bacteria, particle-
Eastern Tropical South Pacific off the coasts of Chile associated bacteria are typically larger (Caron
and Peru (ETSP OMZ; Karstensen et al., 2008; Ulloa et al., 1982; Lapoussiere et al., 2011), occur at higher
and Pantoja, 2009; Ulloa et al., 2012). In the ETSP, local densities (Simon et al., 2002) and exhibit
dissolved oxygen declines from near saturation higher rates of substrate acquisition, enzymatic
levels (4250 mM) at the surface to below the activity, protein production and respiration
level of detection (o10 nM) at the OMZ core (Kirchman and Mitchell, 1982; Karner and Herndl,
(B200300 m; Thamdrup et al., 2012; Ulloa et al., 1992; Grossart et al., 2003, 2007). Not surprisingly,
2012). This steep drawdown drives changes in the free-living and particle-associated communities can
water column microbial community. Notably, com- differ significantly in composition at multiple levels
munities along the oxycline are phylogenetically of phylogenetic resolution (DeLong et al., 1993;
and metabolically diverse relative to other depths Ploug et al., 1999; Grossart et al., 2006; Hunt et al.,
(Bryant et al., 2012), containing both microaerophi- 2008; Kellogg and Deming, 2009). Notably,
lic assemblages, which include ammonia- and phytoplankton-derived particle communities tend
nitrite-oxidizing members, as well as microbes to be relatively enriched in members of the Bacter-
capable of anaerobic metabolism (Stewart et al., oidetes (notably, Cytophaga and Flavobacteria),
2012b). In contrast, community metabolism at the Planctomycetes and Deltaproteobacteria (DeLong
anoxic OMZ core is dominated by anaerobic auto- et al., 1993; Crump et al., 1999; Smith et al., 2013),
trophic and heterotrophic processes that primarily whereas planktonic communities are enriched
utilize oxidized nitrogen compounds as terminal in taxa adapted for oligotrophic or autotrophic
oxidants (Ulloa and Pantoja, 2009; Ulloa et al., 2012; free-living lifestyles (for example, Pelagibacter,
Wright et al., 2012). Up to half of oceanic nitrogen picocyanobacteria; Lauro et al., 2009).
loss occurs in OMZs through the anaerobic pro- Analyses of prokaryotic microbial diversity in
cesses of denitrification and anaerobic ammonium OMZs have focused primarily on the microfilter size
oxidation (anammox; Thamdrup et al., 2006; Lam fraction (B0.23 mm; Stevens and Ulloa, 2008;
et al., 2009; Zehr and Kudela, 2011). Recent studies Canfield et al., 2010; Stewart et al., 2012b), with
also have identified an important role in OMZs for larger or particle-associated cells being excluded.
chemoautotrophic bacteria that oxidize reduced However, certain taxonomic groups with important
sulfur compounds with nitrate, as well as for functional roles in OMZ elemental cycling may
sulfate-reducing heterotrophs (Stevens and Ulloa, differ in abundance and activity between free-
2008; Lavik et al., 2009; Walsh et al., 2009; Canfield living versus particle-associated communities.
et al., 2010; Stewart et al., 2012b). It is unclear, Recent diversity surveys based on 16S rRNA gene
however, whether these key biochemical processes sequences confirm distinct taxon compositions
are differentially partitioned between free-living among size-fractionated samples (30 mm cutoff) from
versus particle-associated microbial communities the suboxic water column of the Black Sea
in suboxic water columns. (Fuchsman et al., 2011, 2012). Notably, members of
In studies of non-OMZ systems, particle-associa- the Alphaproteobacteria (for example, SAR11-like
tion has been shown to be a significant component sequences) and the marine anammox planctomycete
of microbial distributions, community composition Candidatus Scalindua were significantly enriched
and activity (Delong et al., 1993; Hollibaugh et al., in the smaller size fraction, while key anaerobic
2000; LaMontagne and Holden, 2003; Eloe et al., lineages, including members of the sulfate-reducing
2010). Most analyses impose a prefiltering step to Deltaproteobacteria were more abundant in the
separate microbial communities according to parti- larger, particle-associated fraction (430 mm). A recent
cle size, with typical prefilter and microfilter study that demonstrated sulfate-reduction activity in
(collection filter) pore sizes of 0.830 and 0.2 mm, the ETSP OMZ also described 16S rRNA and sulfur
respectively. Although the microfilter fraction (typi- metabolism gene (for example, aprA, encoding ade-
cally cells between 0.2 and 1.63 mm) is presumably nosine-50 -phosphosulfate (APS) reductase) sequences
dominated by non-surface-attached (free-living) matching sulfate-reducing clades of the Deltaproteo-
prokaryotes (Cho and Azam, 1988), the prefilter bacteria, although these sequences were at relatively
may retain a range of organisms, including larger low abundance (Canfield et al., 2010). However,
free-living prokaryotes (for example, filamentous metagenomic sequencing in this study focused only
cyanobacteria), microbial eukaryotes and zooplank- on the free-living community (0.21.6 mm cell size),
ton, but presumably also captures particulate aggre- and it remains unclear to what extent OMZ processes
gates composed of organic debris and surface- such as sulfate reduction may instead be mediated by
attached microbial cells (marine snow). Compared particle communities.
with the open water column, these particles con- Shotgun metagenomics provides a snapshot of the
stitute unique microhabitats that are relatively metabolic functions available in a mixed commu-
enriched in nutrients and contain potentially steep nity, while simultaneously allowing for taxonomic
microscale (microns) gradients in pH and redox identification of community members (Delong et al.,
substrates, including organic carbon (Alldredge and 2006). Surprisingly, metagenomics has been used
Cohen, 1987; Alldredge and Silver, 1988; Stocker, only sparingly to identify differences between

The ISME Journal

 Size-fractionated metagenomics
S Ganesh et al
189
marine microbial communities separated by size DNA extraction
fraction. Seminal metagenomic studies by Venter Genomic DNA was extracted from prefilter disc and
et al. (2004) and Rusch et al. (2007) sampled Sterivex cartridge filters using a phenol:chloroform
different filter size fractions from diverse ocean protocol modified from Frias-Lopez et al. (2008).
sites visited on the Global Ocean Survey expedi- Briefly, cells were lysed by adding lysozyme (2 mg in
tions. This work identified important linkages 40 ml of lysis buffer per filter) directly to the prefilter-
between gene content and environmental variables, containing microcentrifuge tube or to the Sterivex
as well as patterns of metagenomic sequence cartridge, sealing the caps/ends and incubating for
similarity representing distinct community types. 45 min at 37 1C. Proteinase K (1 mg in 100 ml lysis
However, a direct comparison of protein-coding buffer, with 100 ml 20% SDS) was added, and the tubes
gene content between fractions was not presented, and cartridges were resealed and further incubated for
and high-resolution taxonomic surveys from these 2 h at 55 1C. Lysate was removed from each filter, and
analyses are not available. Although such comparisons nucleic acids were extracted once with phenol:chlor-
identified functional differences in size-fractionated oform:isoamyl alcohol (25:24:1) and once with chlor-
picoeukaryote (Not et al., 2009) and viral (Williamson oform:isoamyl alcohol (24:1). The purified aqueous
et al., 2012) communities, to the best of our knowledge phase was concentrated by spin dialysis using
only two other studies (Allen et al., 2012; Smith et al. Amicon Ultra-4w/100 kDa MWCO centrifugal filters
2013), focusing on temperate sites in the North Pacific, (Millipore). Aliquots of purified DNA from each depth
have directly compared microbial (Bacteria and and filter size fraction were used for PCR. Additional
Archaea) functional gene content between filter size aliquots (5 mg) were used to prepare libraries for
fractions. Here, we use metagenome and 16S rRNA shotgun pyrosequencing of microbial metagenomes.
gene amplicon sequencing to compare microbial
communities from two size classes in the ETSP
OMZ. The results identify surface attachment as a 16S rRNA gene PCR
major driver of community composition and genome Pyrosequencing of PCR amplicons encompassing
diversity, and highlight the potential for key physiolo- hypervariable regions of the 16S rRNA gene was
gical processes to be partitioned between free-living used to assess bacterial community composition in
and particle-associated OMZ microbes. both filter types from all water column depths.
Archaeal 16S rRNA gene diversity was not
evaluated via amplicon analysis. PCR amplicons
were synthesized according to established protocols.
Materials and methods Briefly, a B480-bp fragment of the bacterial 16S
rRNA gene was amplified using barcoded universal
Sample collection primers targeting the V1V3 region, as described in
Microbial community samples were collected from the the protocol established for the Human Microbiome
ETSP OMZ as part of the Center for Microbial Ecology: Project by the Broad Institute (Jumpstart Consortium
Research and Education (C-MORE) BiG RAPA Human Microbiome Project Data Generation Working
(Biogeochemical Gradients: Role in Arranging Plank- Group, 2012). Briefly, thermal cycling conditions
tonic Assemblages) cruise aboard the R/V Melville (18 were: initial denaturation at 95 1C (2 min), followed
November14 December 2010). Seawater was sampled by 30 cycles of denaturation at 95 1C (20 s), primer
from 7 depths (5, 32, 70, 110, 200, 320 and 1000 m) at annealing at 50 1C (30 s) and primer extension at 72 1C
Station 1 (201 04.999 S, 701 48.001 W) off the coast of (5 min). 16S rRNA gene PCR was replicated using a
Iquique, Chile on November 19th (5 m), 20th (32 m), second set of barcoded primers to assess potential
21st (70, 110, 200 and 320 m) and 23rd (1000 m). variation introduced during PCR. Amplicons were
Collections were made using Niskin bottles deployed analyzed by agarose gel electrophoresis to verify size,
on a rosette containing a Conductivity-Temperature- purified using the QIAQuick PCR Clean-Up Kit,
Depth profiler (Sea-Bird SBE 911plus) equipped with pooled (B200 ng per sample), and used as template
a dissolved Oxygen Sensor, fluorometer and trans- for multiplex amplicon pyrosequencing.
missometer (see Supplementary Figure S1). Microbial
biomass was collected by sequential in-line filtration
of seawater samples (10 l) through a prefilter (GF/A, Pyrosequencing
1.6 mm pore-size, 47 mm diameter, Whatman, GE Shotgun pyrosequencing (454 Life Sciences, Roche
Healthcare, Piscataway, NJ, USA) and a primary Applied Science, Branford, CT, USA) was used to
collection filter (Sterivex, 0.22 mm pore-size, Millipore, characterize the community DNA (metagenome)
Billerica, MA, USA) using a peristaltic pump. from two microbial size fractions (prefilter and
Prefilters were transferred to microcentrifuge tubes Sterivex) from four depths (70, 110, 200 and
containing lysis buffer (B1.8 ml; 50 mM Tris-HCl, 1000 m), as well as the multiplexed 16S amplicon
40 mM EDTA and 0.73 M sucrose). Sterivex filters were samples (seven depths, two filter types per depth,
filled with lysis buffer (B1.8 ml), and capped at both two PCR replicates per filter). DNA templates were
ends. Both filter types were stored at  80 1C until used to prepare single-stranded libraries for emul-
DNA extraction. sion PCR using established protocols (454 Life

The ISME Journal

 Size-fractionated metagenomics
S Ganesh et al
190
Sciences, Roche Applied Science). Each metagen- low quality sequences using default parameters
ome sample was sequenced with a half-plate run (minimum quality score 25, minimum sequence
on a Roche Genome Sequencer FLX Instrument length 200, no ambiguous bases allowed).
using Titanium chemistry. The multiplexed De-multiplexed sequences were clustered into
amplicon sample was sequenced using a single full- operational taxonomic units (OTUs) at 97%
plate run. sequence similarity, with taxonomy assigned to
representative OTUs from each cluster using the
Ribosomal Database Project classifier in QIIME,
Sequence analysis16S rRNA gene amplicons trained on the Greengenes database. OTU counts
Amplicons were analyzed using the software were rarefied (10 iterations) and alpha diversity was
pipeline QIIME (Caporaso et al., 2010), according quantified at a uniform sequencing depth across
to standard protocols. Briefly, barcoded 16S data samples using the phylogenetic diversity (PD)
sets were de-multiplexed and filtered to remove metric as described by Faith (1992); (Figure 1a).

Figure 1 OMZ bacterial community diversity revealed by 16S rRNA gene pyrosequencing. (a) PD as a function of water column depth
and dissolved oxygen concentration (left). Data points are mean values based on rarefaction of OTU counts at a standardized sequence
count (n 4996) per sample, with bars indicating 95% confidence intervals for the rarefied measurements. Data from both PCR
duplicates are combined for averaging. (b) Principle component analysis of community taxonomic relatedness, as quantified by the
weighted Unifrac metric. OMZ depths are circled. (c) Relative abundance of major bacterial divisions within the Ribosomal Database
Project (RDP) classification, as a percentage of total identifiable bacterial sequences. Colors and ordering of taxa match those in D.
Samples are labeled by depth and filter type, where p prefilter (41.6 mm), s Sterivex (0.21.6 mm). Duplicates in B and C reflect
duplicate PCR reactions. (d) Variation in the relative abundance of bacterial divisions between filter size fractions. Values are the base-10
logarithm of the odds ratio: the ratio of the odds a taxon occurs on the prefilter to the odds it occurs on the Sterivex. Positive values
indicate taxa that are more likely to occur on the prefilter. Values are based on counts pooled from all depths, with corrections for
differences in data set size. Panels c and d exclude divisions present in only one filter type and occupying less than 0.01% of total
bacterial sequences. Stars mark taxa whose relative abundance differed significantly between size fractions (Po0.05).

The ISME Journal

 Size-fractionated metagenomics
S Ganesh et al
191
To compare community composition between sample clusters was evaluated via multiscale boot-
samples, sequences were aligned using the PyNAST strapping (1000 replicates) based on the approxi-
aligner in QIIME and beta diversity was calculated mately unbiased method, implemented using the
using the weighted Unifrac metric. This metric program pvclust in the R language. BrayCurtis
compares samples based on the phylogenetic related- dissimilarities were also calculated from transformed
ness (branch lengths) of OTUs in a community, count data to assess dissimilarity in functional gene
while taking into account relative OTU abundance composition (Supplementary Table S3).
(Lozupone and Knight, 2005). Values range from 0 to To further evaluate genes or gene categories not
1, with 1 indicating the maximum distance between represented in SEED, BLASTX results (4bit score
samples. Sample relatednesss based on Unifrac was 50) were manually parsed via keyword searches
visualized using a two-dimensional Principal Coor- based on NCBI-nr annotations, as in Canfield et al.
dinate Analysis (Figure 1b). For all pairs of samples, (2010). NCBI-nr genes representing top BLASTX
a Monte Carlo permutation test (1000 permutations) matches were recovered from GenBank, and each
was used to determine if the Unifrac distance database entry was examined manually to confirm
between samples was greater than expected by gene identity. Entries with ambiguous annotations
chance, with a false discovery rate correction were further verified by BLASTX. Manual searches
(a 0.05) imposed for multiple tests according to focused on key enzymes of dissimilatory nitrogen and
Benjamini and Hochberg (1995) (see Supplementary sulfur metabolism (Lam et al., 2009; Canfield et al.,
Table S2). 2010; Figure 3), including: ammonia monooxygenase
(amoC), nitrite oxidoreductase (nxrB), hydrazine
oxidoreductase (hzo), nitrate reductase (narG), nitrite
Sequence analysismetagenomes reductase, nitric oxide reductase (norB/norZ) and
Analysis of protein-coding metagenome sequences nitrous oxide reductase (nosZ). Nitrite reductase genes
followed that of Canfield et al. (2010) and Stewart were further characterized as either nirK or nirS,
et al. (2012b). Duplicate reads sharing 100% encoding the functionally equivalent copper-contain-
nucleotide similarity and identical lengths, which ing and cytochrome cd1-containing nitrite reductases,
may represent artifacts of pyrosequencing, were respectively, or as nrfA, encoding the cytochrome
identified by clustering in the program CD-HIT c nitrite reductase. Sulfur metabolism enzymes
(Li and Godzik, 2006) and removed from each data included dissimilatory sulfite reductase (dsrA), APS
set as in Stewart et al. (2010). Metagenome reductase (aprA) and sulfate thiol esterase (soxB). We
sequences were compared using BLASTX against also present results for key genes involved with
the NCBI-nr database of nonredundant protein mobile element activity, notably transposases and
sequences (as of January 2012). BLASTX matches integrases, which were highly represented in the data
to prokaryote genes (Bacteria and Archaea) above a sets. Genes encoding transposases and integrases from
bit score of 50 were retained and classified accord- diverse families were combined into single categories
ing to functional category based on the SEED for presentation.
classification of functional roles and subsystems When possible (Figure 3), gene abundances were
(Overbeek et al., 2005), using the program normalized based on best approximate gene length
MEtaGenome ANalyzer 4 (MEGAN Version 4; (kb), estimated based on full-length open reading
Huson et al., 2011). The relative abundance of a frames from sequenced genomes: amoC (750 bp);
SEED subsystem was calculated for each sample as nxrB (1500 bp); hzo (1650 bp); narG (3600 bp); nirK
the number of sequences per subsystem normalized (1140 bp); nirS (1620 bp); nrfA (1440 bp); norB
by the total number of sequences matching (1410 bp); nosZ (1950 bp); dsrA (1200 bp); aprA
subsystems; these values were then averaged across (1860 bp); soxB (1680 bp). The length of genes
the four depths to obtain the SEED abundances encoding transposases and integrases varies among
shown in Figure 2. Normalized SEED counts for gene family type. Here, averages of 900 and 1150 bp
individual depths are available in Supplementary were used for transposases and integrases, respec-
Table S4. The taxonomic composition of protein- tively, based on averaging randomly selected full-
coding sequences was determined based on the length transposase and integrase genes (100 each)
taxonomic annotation of each gene, according to identified in our data sets. Sequence counts per
the NCBI-nr taxonomy in MEGAN4. kilobase of target gene were then normalized to
Samples were clustered based on SEED subsystem counts of sequences matching the universal, puta-
profiles (Figure 2, inset). For each sample, hit counts tively single-copy gene encoding RNA polymerase
per subsystem were normalized to the proportion of subunit B (rpoB, 4020 bp). A value of 1 (Figure 3)
total prokaryote reads matching SEED. An arcsine indicates abundance in the metagenome equivalent
square root transformation was applied to propor- to that of rpoB, assuming the gene lengths used in
tions to stabilize variance relative to the mean. our calculations.
Pearsons correlation coefficients were calculated for The proportion of differentially abundant genes
each pair of transformed data sets and used as (in metagenomes) or taxa (in 16S amplicon data)
similarity metrics for hierarchical clustering using between Sterivex and prefilter samples was esti-
the complete-linkage method. The probability of mated via an empirical Bayesian approach in the

The ISME Journal

 Size-fractionated metagenomics
S Ganesh et al
192

The ISME Journal

 Size-fractionated metagenomics
S Ganesh et al
193
R program baySeq (Hardcastle and Kelly, 2010), as in microbes in general. Microbial biomass from envir-
the study of Stewart et al. (2012a). As true replicates onmentally distinct depth zones spanning the
at each depth were not available, statistical validation permanent OMZ off Chile was separated by filtra-
of depth-specific differences between filter fractions tion into a large size fraction (41.6 mm) and a small
was not possible. Consequently, all samples (depths) size fraction (0.21.6 mm), which for convenience are
belonging to each filter type were modeled as referred to here as prefilter and Sterivex, respec-
biological replicates. The baySeq method assumes a tively. Although the prefilter fraction is presumably
negative binomial distribution of the data with prior enriched in particle-associated microorganisms and
distributions derived empirically from the data cellcell aggregates, it may also contain larger free-
(100 000 iterations). Dispersion was estimated via a living cells (for example, bacterial filaments and
quasi-likelihood method, with the sequence count protists). Similarly, the small size fraction is likely
data normalized by data set size (that is, total number dominated by free-living Bacteria and Archaea, but
of prokaryotic protein-coding genes and total number may also contain surface-associated microorganisms
of bacterial 16S rRNA sequences for the metagenome dislodged from particles during sampling (Hunt
and amplicon analyses, respectively). Posterior like- et al., 2008). Consequently, filter size fraction is a
lihoods per gene category or taxon were calculated potentially uncertain indicator of microbial life
for models (sample groupings) in which genes/taxa history strategy (that is, particle-attached versus
were either predicted to be equivalently abundant in free-living). Nonetheless, in comparing the taxo-
both prefilter and Sterivex samples or differentially nomic and functional gene compositions between
abundant between filter types. A false discovery rate fractions, the following sections highlight a signifi-
threshold of 0.05 was used for detecting differentially cant role of size fraction in structuring microbial
abundant categories. communities and identify physiological and geno-
All sequence data generated in this study are mic properties suggestive of a partitioning between
publicly available in the NCBI database under surface-associated and free-living microbial strate-
BioSample numbers SAMN02317187-SAMN02317 gies, as well as key OMZ processes of dissimilatory
194 (metagenomes) and SAMN02339399-SAMN023 nitrogen and sulfur metabolism.
39426 (amplicons).

Oxygen conditions
Results and discussion The ETSP OMZ sample site near Iquique, Chile was
characterized by steep vertical gradients in dissolved
Attachment to suspended or sinking particles is a oxygen (Figure 1a), similar to what has been reported
major life history strategy for marine microorganisms previously for this region (Dalsgaard et al., 2012;
(Smith et al., 1992; Simon et al., 2002; Grossart, 2010) Stewart et al., 2012b; Ulloa et al., 2012).
and consequently has an important role in structur- The base of the photic zone (1% surface PAR)
ing community taxonomic composition and bio- occurred at B40 m, within the oxycline (B3070 m).
chemical activity (DeLong et al., 1993; Crump et al., Dissolved oxygen conditions at the time of sampling
1999). However, the patterns by which microbial decreased from B250 mM at the surface to below 5 mM
physiological traits are distributed between particle- through the OMZ core (B100400 m), before
associated and free-living communities are not well gradually increasing below 400 m to B60 mM at
understood for many ocean regions. Characterizing 1000 m. The oxygen sensor used here has resolution
particle-associated microbes may be especially in the micromolar oxygen range. However, recent
important in OMZs, where local particle maxima measurements with high-resolution (10 nM) switchable
have been positively related to both oxygen depletion trace oxygen sensors indicated that the ETSP OMZ
(Garfield et al., 1983; Whitmire et al., 2009) and core is anoxic, with oxygen below detection through-
microbial metabolic activity (Naqvi et al., 1993). out the OMZ core (Thamdrup et al., 2012). Our
This study presents the first metagenomic com- amplicon data sets therefore span the oxygenated
parison of microbial size fractions in a marine OMZ, photic zone and oxycline (5 and 32 m samples), the
and one of the first to examine community suboxic (o10 mM) upper OMZ (70 m), the anoxic OMZ
metabolic traits between size-fractionated marine core (110, 200, 320 m), and the oxic zone beneath the

Figure 2 Differences in the relative abundance of functional gene categories between microbial size fractions (filter type), summarized
across depths. Categories on the left are subsystems in the SEED classification, with the figure showing only subsystems comprising
40.1% of the total sequences matching SEED. Higher level classifications of each subsystem are listed on the right. Filled and unfilled
black bars reflect the relative abundance of prokaryotic sequence reads matching each category, normalized to the total number of
prokaryotic sequences matching SEED. Light gray bars reflect the base-10 logarithm of the odds ratio: the ratio of the odds a gene category
occurs on the prefilter to the odds it occurs on the Sterivex. Positive values indicate categories that are more likely to occur on the
prefilter. Values are based on counts pooled from all depths sampled for metagenomics (70, 110, 200 and 1000 m), with corrections for
differences in data set size. Categories whose relative abundance differed significantly between size fractions (Po0.05; baySeq) are
starred for analyses based on all four depths (filled stars) or only the OMZ depths (70, 110 and 200 m; open stars). Dendrogram (inset)
shows relatedness of individual samples based on SEED subsystem profiles, with samples labeled by depth and filter type (p prefilter,
s Sterivex). Numbers at nodes are probabilities based on multiscale bootstrap resampling (1000 replicates). X axis correlation
coefficients.

The ISME Journal

 194

The ISME Journal

S Ganesh et al
Size-fractionated metagenomics

Figure 3 Relative abundance (a) and taxonomic representation (b) of sequences matching genes of key dissimilatory nitrogen and sulfur pathways. Abundance is calculated as read
count per gene per kilobase of gene length, and shown as a proportion of the abundance of the universal single-copy gene encoding RNA polymerase subunit B (rpoB). A value of 1
indicates gene abundance equal to that of rpoB. Taxonomic identifications are based on annotations of NCBI reference sequences identified as top matches (above bit score 50) in
BLASTX searches. See Methods for gene identifications. Stars mark genes whose abundance differed significantly between size fractions (Po0.05; baySeq) in an analysis of only the
OMZ depths (70, 110 and 200 m). Samples are labeled by depth and filter type, where p prefilter (41.6 mm), s Sterivex (0.21.6 mm). Taxonomic group Proteobacteria, gamma_S
indicates Gammaproteobacteria of the sulfur-oxidizing SUP05 clade (Walsh et al., 2009). Inset shows the base-10 logarithm of the odds ratio for each gene category: the ratio of the odds a
gene occurs on the prefilter to the odds it occurs on the Sterivex. Values are based on counts pooled across depths, with corrections for differences in data set size.
 Size-fractionated metagenomics
S Ganesh et al
195
OMZ (1000 m). The metagenome samples focus on a study of the 0.22.7 mm fraction from a seasonal
subset of depths in the upper OMZ (70 m), OMZ core OMZ off British Columbia (Zaikova et al., 2010). In
(110, 200 m) and beneath the OMZ (1000 m). contrast, Stevens and Ulloa (2008), based on
These data sets likely also span a gradient in bulk libraries of cloned 16S sequences, identified a
particle load. Consistent with prior studies reporting peak in OTU diversity (multiple indices) at the
elevated particle concentrations within the ETSP ETSP OMZ within the 0.23 mm fraction, a pattern
OMZ (Pak et al., 1980; Whitmire et al., 2009), consistent with that observed for the prefilters in
particulate load, inferred indirectly here from beam our study. Similarly, elevated OTU richness in the
attenuation measurements, exhibited local maxima 0.21.6 mm fraction has been shown to coincide
within the upper photic zone (B15 m) and then with the zone of minimum oxygen concentration at
again within the OMZ core (B140 m), before tropical non-OMZ sites (Brown et al., 2009;
declining to a consistent minimum below the OMZ Kembel et al., 2011).
(Supplementary Figure S1). However, the size These studies consistently highlight a shift in
distribution and composition of particles contribut- microbial community complexity associated with
ing to the beam attenuation signal are not character- zones of low oxygen. For the ETSP OMZ, where
ized here. It therefore remains to be determined how oxygen declines to the nanomolar range, increasing
changes in bulk particle load relate to changes in the diversity in the OMZ core has been hypothesized to
abundance of the size-fractionated communities be linked to the use of a wider range of terminal
discussed in detail below. oxidants, compared with non-OMZ depths where
oxygen is the dominant electron acceptor (Stevens
and Ulloa, 2008). Conversely, Bryant et al. (2012)
Taxonomic diversity16S rRNA gene amplicons argue that niche diversity declines within the
Pyrosequencing of 16S rRNA gene amplicons revealed anoxic OMZ as niches linked with light and labile
a species-rich OMZ bacterial community whose organic matter utilization, which are more preva-
composition varied over depth and between size lent at the surface and oxycline, are lost. Our data
fractions. A total of 17 014 bacterial OTUs confirm that taxonomic diversity varies between
(97% similarity clusters) were recovered across all size fractions, with diversity elevated in larger size
samples, with per sample OTU counts ranging from fractions within the OMZ. Similarly, elevated OTU
658 to 2484 based on data set size (Supplementary richness in particle-associated compared with free-
Table S1). Despite relatively high numbers of living communities has been reported for other
sequences per sample (mean: 14 527), this analysis marine habitats (for example, Eloe et al., 2010),
did not capture the total OTU richness in each sample suggesting that higher diversity may be linked to an
(that is, no rarefaction curves approached saturation; increase in niche richness associated with micro-
Supplementary Figure S2), as anticipated for marine gradients in substrate availability and composition
bacterioplankton assemblages (Huber et al., 2007). on particles. However, this pattern is not observed
across all depths or ocean sites (for example,
OTU diversity with depth. OTU diversity patterns Figure 1a; Moeseneder et al., 2001; Ghiglione
varied with both depth and filter size fraction et al., 2007), suggesting a need for quantifying
(Figure 1, Supplementary Figures S2S4). For both niche availability in response to diverse para-
size fractions, PD, the total branch length connecting meters, notably the organic composition, abun-
all OTUs in the 16S rRNA gene phylogeny dance, and size distribution of particles,
(Faith, 1992), was shortest at the surface (5 m) combined with physical and chemical gradients
and increased within the oxycline (B3070 m) of the bulk water column.
(Figure 1a). However, PD of the two size fractions
differed within the OMZ. PD of Sterivex commu- Taxonomic composition. Bacterial taxonomic com-
nities decreased from the oxycline to the anoxic OMZ position varied markedly among samples. Vertical
core at 200 m, whereas PD of prefilter communities trends in the community composition of free-living
increased within the core (Figure 1a). PD trends bacteria in the ETSP OMZ have been reported
closely paralleled those of other alpha diversity previously (Stevens and Ulloa, 2008; Bryant et al.,
indices, including counts of observed OTUs and 2012) and agree broadly with those observed here.
Chao1 estimates (Supplementary Figure S3). We instead focus primarily on comparisons between
Vertical patterns of OMZ microbial diversity are size fractions. Figure 1d shows the odds of a given
not consistent among studies. A decline in PD from taxonomic division occurring in the prefilter frac-
the oxycline to anoxic depths, based on 16S rRNA tion relative to the Sterivex fraction, based on OTU
gene fragments from metagenomes, was observed counts pooled across depths. Of the 25 major
for the 0.21.6 mm size fraction across years and bacterial divisions identified in the amplicon ana-
seasons in the ETSP OMZ off Chile (Bryant et al., lysis, 15 were significantly over- or underrepre-
2012), suggesting temporal stability and a consis- sented in the prefilter fraction (Po0.05, baySeq).
tent decline in diversity within the OMZ free- A subset of these trends is discussed below.
living community. Low diversity associated with Alphaproteobacteria sequences were abundant in
suboxia was also reported in a gene fingerprinting both filter fractions but were consistently enriched

The ISME Journal

 Size-fractionated metagenomics
S Ganesh et al
196
in the free-living community, where they consti- the Spirochetes and Mollicutes, though at negligible
tuted an average of 32% of all sequences (versus overall abundance in our data sets (o0.1%), were
13% in the prefilters) (Figure 1c). Enrichment was also significantly enriched in the large size fraction.
driven primarily by the SAR11 clade (Pelagibacter Spirochetes, which are traditionally associated with
sp.), which represented 5097% (mean: 84%) of marine sediments or microbial mats, were only
Alphaproteobacteria sequences from Sterivex filters, detected at the core OMZ depths (110, 200 m),
and 2572% (mean: 58%) of those from prefilters. consistent with this group being dominated by
SAR11 enrichment in the Sterivex fraction is strictly or facultatively anaerobic members (Munn,
consistent with these bacteria being free-living 2011). Marine spirochetes have also been found in
oligotrophs adapted for the efficient use of dissolved association with eukaryotes (Ruehland et al., 2008;
substrates (Giovannoni et al., 2005). In contrast, Demiri et al., 2009), and may therefore be enriched
SAR11 sequences in the prefilter fraction may in the particle fraction via attachment to larger
represent unique surface-associated ecotypes, as organisms or sinking fecal matter. Similarly,
proposed for SAR11 detected in larger size fractions sequences matching Mollicutes, which here were
(0.83.0 and 3.0200.0 mm) from an oxic upwelling affiliated exclusively with the Mycoplasma (data not
zone (Allen et al., 2012). Diversity surveys that do shown), may have originated from eukaryotic mate-
not examine the prefilter fraction may be excluding rial, as mycoplasmas have been found in the larval
important components of the SAR11 community. stages of marine invertebrates and in the intestinal
Here, SAR11 were abundant at both oxic and anoxic microflora of several fish species (Zimmer and
depths, as shown previously for the ETSP OMZ Woollacott, 1983; Bano et al., 2007).
(Stevens and Ulloa, 2008; Stewart et al., 2012a,b). Sequences matching eukaryotic chloroplasts or
However, the metabolic adaptations that enable cyanobacteria were also more abundant on average
these putatively aerobic bacteria to grow under low on the prefilters. This sequence group was a
or no oxygen remain uncharacterized (Wright et al., substantial component (2038%) of both filter
2012). Future analyses at finer levels of taxonomic fractions at 5 m (Figure 1c), but was confined
resolution may identify SAR11 subclades primarily to the prefilter beginning at 32 m. This
unique to both particle-associated and low-oxygen change in size fraction was accompanied by a shift
environments of the ETSP. in the structure of the eukaryotic phototroph
Sequences matching high GC gram-positive Acti- community, from dominance by Chlorophyta and
nobacteria (Actinomycetes) were a relatively minor Cryptomonadaceae at 5 m to Bacillariophyta
component of the total amplicon pool (mean: 2% (diatoms) at 32 m and below (data not shown).
across all samples), but were significantly more Throughout the depth range, cyanobacterial
abundant in Sterivex filters (excluding the 5 m sequences primarily matched clade GpIIa (for
sample where Actinobacteria were negligible in example, Prochlorococcus and Synechococcus),
both size fractions; Figure 1c). A similar enrichment with abundance peaking in the prefilters at 32
was observed recently in the 0.10.8 mm bacterio- and 70 m within the photic zone. The presence of
plankton fraction from temperate coastal commu- cyanobacterial-like sequences below the photic
nities (Smith et al., 2013). Actinobacteria are most zone (Figure 1c) has been reported previously for
commonly associated with terrestrial soil habitats the ETSP OMZ (Bryant et al., 2012) and in other
but are also regularly cultivated from marine sedi- deep-water habitats (Smith et al., 2013) and may
ments and, less commonly, from suspended organic be due to aggregation onto or release from sinking
aggregates (Grossart et al., 2004) and pelagic envir- particles, including fecal pellets. Here, the relative
onments (Rappe et al., 1999; Bull et al., 2005), abundance of these sequences (namely chloroplasts)
including OMZs (Fuchs et al., 2005). Here, the increased in the 1000 m prefilter, which could be
majority of Actinobacterial sequences (76%) were due to changes in the turnover rates of different
unclassified (data not shown), suggesting the particle-associated cell fractions (that is, choroplasts
possibility of novel planktonic lineages, potentially embedded in fecal particles increase in relative
distinct from those associated with particles abundance as bacterial activity and cell numbers on
(Jensen and Lauro, 2008; Prieto-Davo et al., 2008). particles decrease).
Twelve major bacterial divisions were Consistent with several prior studies of particle-
significantly enriched in prefilter communities associated bacteria (DeLong et al., 1993; Fuchsman
(Figure 1d). Diverse clades of the Bacteroidetes, et al., 2012), Deltaproteobacteria were significantly
including the Flavobacteria and Sphingobacteria, enriched on prefilters. Although this group was a
were among the most overrepresented groups in this negligible component of both size fractions at the
fraction. As suggested in prior reports from non- surface (o0.1% of total sequences at 5 m), deltapro-
OMZ systems (Crump et al., 1999; Simon et al., teobacterial sequences increased in relative abun-
2002; Allen et al., 2012), elevated numbers of dance in both fractions by 70 m, and were 8- to
Bacteroidetes on particles may be linked to the 28-fold more abundant in prefilters from 70 m down
enhanced capacity of these bacteria to degrade high to 1000 m, representing up to 3% of total sequences
molecular weight biopolymers, such as chitin or in the larger size fraction (Supplementary Figure S8).
proteins (Cottrell and Kirchman, 2000). Members of Notably, the Myxococcales, a widely distributed

The ISME Journal

 Size-fractionated metagenomics
S Ganesh et al
197
Order with both terrestrial and marine members the presence of anoxia, which may be scarce on newly
that exist primarily in surface-attached swarms formed particles in the oxic depths, but relatively
(Shimkets et al., 2006; Jiang et al., 2010), were up common in older, deeper particles where microbial
to 95-fold more abundant on prefilters (relative to respiration has created local pockets of oxygen
Sterivex) from depths below the oxycline. Marine depletion. It is also possible that the sinking of
myxobacteria have been found in anoxic sediments, particles into the suboxic OMZ facilitates particle-
but are associated primarily with oxic habitats associated anaerobic metabolisms.
(Brinkhoff et al., 2012), and the Order as a whole is Amplicons matching Planctomycetes provided
dominated by strictly aerobic heterotrophs (Shimkets limited phylogenetic information, with 63% of all
et al., 2006). Myxobacteria have also been found in planctomycete amplicons (both filter fractions)
open ocean picoplankton (DeLong et al., 2006; Pham identified only to the Family Planctomycetaceae.
et al., 2008). It therefore is possible that OMZ This pattern was most pronounced at 1000 m, where
myxobacteria, as well as other particle-associated taxa, 84% of Planctomycetaceae sequences were unclas-
have been transported to anoxic depths after attach- sified. Of the sequences assignable to a Genus, 54%
ment to sinking particles in the oxic zone, but are not matched the Genus Planctomyces, with the vast
metabolically active in the OMZ. majority of these being detected only in the
Deltaproteobacteria clades known to contain prefilters where Planctomyces were enriched on
sulfate-reducing members (for example, Desulfobac- average 75-fold compared with the Sterivex fraction.
terales) were generally more abundant at core OMZ This pattern agrees with genetic and isolation-based
depths (110, 200, and 320 m; Supplementary Figure S8), studies identifying surface attachment as a key life
consistent with their low-oxygen requirements. history state for diverse Planctomycete genera,
However, the relative abundance of these groups including Planctomyces (Bauld and Staley, 1976;
did not differ appreciably between prefilter and Morris et al., 2006; Bengtsson and vreas, 2010) and
Sterivex fractions, except within the 320 and 1000 m with a general Planctomycetes enrichment in parti-
samples where these sequences were barely (or not) cle-associated microbial cell fractions (DeLong et al.,
detectable in the free-living fraction (Supplementary 1993; Fuchsman et al., 2011, 2012). Although the
Figure S8). Sulfate reduction in oxic water columns 16S amplicon pool provided limited phylogenetic
presumably is localized to reduced microzones resolution for some taxonomic groups discussed
on particles (Shanks and Reeder, 1993). Our data raise here, additional insight into the composition of key
the possibility that sulfate reduction in the OMZ, OMZ clades can be provided by analyzing the
which has been demonstrated recently using radi- taxonomic identification of protein-coding genes
olabeling of bulk water samples (Canfield et al., 2010), (see Metagenome data below).
may not be confined to particle-associated microhabi-
tats. However, the vast majority (mean: 72%) of Community relatedness. Sample relatedness based
deltaproteobacterial 16S sequences across both filter on community phylogenetic composition (Unifrac
fractions were unclassified (Supplementary Figure metric) varied with depth. For both the prefilter and
S8). Classification at higher levels of phylogenetic Sterivex sample sets, communities on the periphery
resolution will be necessary to clarify how particle- (70 m) and within the OMZ (110, 200 and 320 m)
attachment in the OMZ affects the distribution of clustered to the exclusion of those from the surface (5
deltaproteobacterial and 32 m) and beneath the OMZ (1000 m), although
subclades, including those with sulfate reducers. this pattern was most pronounced for the prefilter
rRNA amplicons matching members of the super- communities (Figure 1b and Supplementary Figure
phylum comprising the Planctomycetes, Verrucomi- S4B,C). Notably, the communities at 5 m (prefilter
crobia, Lentisphaerae and Chlamydiae (Wagner and and Sterivex) were highly distinct from those at
Horn, 2006) were significantly overrepresented deeper depths, due primarily to the high abundance
on prefilters (Figure 1d). Notably, on average across of cyanobacteria and eukaryotic chloroplasts at the
the depths, the relative abundance of Planctomy- surface, as well as a shift in the structure of the
cetes was 15-fold higher in prefilter compared with proteobacterial community (Figures 1b and c). Of the
Sterivex samples. However, this enrichment was not OMZ depths, samples from 70, 110 and 200 m were
uniform throughout the water column. Planctomycete most closely related (Figure 1b and Supplementary
sequences were either not detectable or a very Figure S4B,C). The 320 m sample was a relative
minor percentage (00.3%) of total amplicons in outlier. Specifically, the 320 m Sterivex community
the oxic 5 and 32 m samples, even within the prefilter was enriched in Flavobacteria (primarily unclassi-
fraction (Supplementary Figure S7). In contrast, fied Flavobacteria) compared with the other OMZ
Planctomycetes represented 12% of total sequences depths (Figure 1c). This shift highlights the potential
in the prefilter at 70 m and increased to a peak of for community variation throughout the OMZ core,
B5% at 1000 m (Supplementary Figure S7). A similar despite apparent uniformity in some environmental
depth-specific increase in relative enrichment was conditions, notably oxygen, across these depths
observed in the Verrucomicrobia and Lentisphaerae. (Figure 1a, Thamdrup et al., 2012).
Presumably, the distribution of these groups, which Filter size fraction also had a major role in
predominantly contain anaerobic members, is tied to determining community relatedness. Sterivex

The ISME Journal

 Size-fractionated metagenomics
S Ganesh et al
198
communities clustered to the exclusion of the spanning the oxycline and OMZ core clustered
corresponding prefilter communities from the same by size fraction despite strong vertical stratification
depth (Figure 1b). Ninety-three percent (182/196) of in environmental parameters such as oxygen
the pairwise comparisons between filter types (p (Figure 1).
versus s in Supplementary Table S2, top) revealed These trends suggest that life history mode
significant differences in taxonomic composition associated with size fraction (free-living versus
based on phylogenetic relatedness (Po0.05; mean particle-associated) has a greater role in structuring
Unifrac: 0.53). At only one depth, 320 m, did OMZ communities than water column oxygen
prefilter and Sterivex communities not differ levels. Although O2 and nutrient concentrations at
significantly (P 0.110.13; note similarity in PC1 the anoxic OMZ core are dramatically different from
coordinates in Figure 1b), although even in this those of the overlying oxic layer, the taxonomic
sample clear differences between filters were evident composition of OMZ particles broadly reflects that
(Figure 1c). In contrast, in comparisons involving of particles from oxic marine habitats (Delong et al.,
the same filter type, 54 (49/91) and 26% (24/91) 1993; Rath et al., 1998), with a relative enrichment
of prefilter and Sterivex comparisons showed of the Bacteroidetes, Firmicutes, Planctomycetes
significant differences (mean Unifrac: 0.39 and and Deltaproteobacteria, and an underrepresenta-
0.35 for prefilters and Sterivex, respectively). Of tion of bacteria typically associated with oligo-
the 14 comparisons involving data from duplicate trophic conditions (for example, SAR11). Many of
PCR reactions, two (the 32 m prefilter and 1000 m the clades enriched on prefilters also contain
Sterivex samples) showed significant compositional aerobic members (for example, Flavobacteria and
differences, indicating a potential for PCR-induced Myxobacteria), raising the question of whether
variation to influence diversity comparisons. particle-associated bacteria are metabolically active
Even when the outlier surface sample (5 m) was within the OMZ, or are quiescent, having been
excluded from the analysis, clustering patterns transported to OMZ depths on particles originating
indicated that microbial size fraction was a stronger in the oxic zone. Indeed, size fraction-specific
predictor of community relatedness than was clustering of samples suggests that passage through
vertical position in the water column (Figure 1b, the anoxic OMZ may have a relatively minor effect
Supplementary Figure S4A). Many prior studies on the composition of the particle-associated micro-
have confirmed fundamental differences in commu- bial communities. However, release of microbes
nity composition between size fractions (DeLong from sinking particles may represent a valuable
et al., 1993; Acinas et al., 1999; Crump et al., 1999; conduit of anaerobic bacteria to OMZ depths.
Ghiglione et al., 2007; Parveen et al., 2011).
However, others show relative similarity between Taxonomic and functional variation among size
fractions (Hollibaugh et al., 2000). For example, fractionsMetagenomics
communities from three size fractions (3.020, Shotgun metagenomics was used to examine differ-
0.83.0 and 0.10.8 mm) from a surface ocean sample ences in taxonomic composition and metabolic
grouped together based on shared metagenome function between free-living and particle-associated
sequence, distinct from communities at distant size fractions. Pyrosequencing of eight metagenome
oceanic sites (Rusch et al., 2007). A similar pattern, samples (four depths, two filters per depth) gener-
whereby bacterial communities from distinct size ated 1 660 922 reads (range: 163 987275 575 per
fractions of the same sample are more similar than sample; mean length: 305 bp; Table 1). Of these, 48%
communities from other samples, has been shown were designated as protein-coding based on
for the deep ocean (Eloe et al., 2010) and the anoxic BLASTX matches (bit score 450) to proteins in the
Black Sea (Fuchsman et al., 2011) based on 16S gene NCBI-nr database, with 10% of these matching
sequences, and for a coastal hypoxic layer based on prokaryotic genes in SEED Subsystem categories.
the taxonomic annotations of coding genes from The percentage of identifiable protein-coding
metagenomes (Smith et al., 2013). Conflicting sequences was consistently higher in the Sterivex
patterns of community relatedness are potentially fraction (mean: 66% of total sequences matching
due to differences in size-fractionation and taxo- NCBI-nr, compared with 24% on prefilters). This
nomic identification methods across studies, as well discrepancy was likely due in part to the enrichment
as variation in water column conditions among (816-fold) of eukaryotic genes on prefilters
samples. Indeed, zonation in parameters such as (Table 1), which would have increased the propor-
light, temperature or nutrient availability signifi- tion of non-coding DNA per metagenome. In addi-
cantly influences taxonomic diversity and metabolic tion, prefilters were enriched threefold in sequences
function across diverse marine habitats (DeLong matching genes annotated as viral in origin. Pre-
et al., 2006; Qian et al., 2011), including the ETSP sumably, most free-living marine viruses are too
OMZ (Stevens and Ulloa, 2008; Bryant et al., 2012; small (o0.2 mm) to have been retained in either filter
Stewart et al., 2012b), and is therefore a critical fraction. The viral reads in the data therefore likely
driver of niche differentiation in ocean microbes originate either from extracellular viruses attached
(Rocap et al., 2003; Johnson et al., 2006). None- to the surfaces of cells or particles (for example,
theless, in our study, samples from depths within a biofilm), or from prophage. As we did not

The ISME Journal

 Size-fractionated metagenomics
S Ganesh et al
199
Table 1 Metagenome sequence statistics and taxonomic (Domain) identities

Sample Reads Coding % Of % Of % Of unidentified % Of % Of % Of SEEDb

readsa Bacteria Archaea prokaryotes eukaryotes viruses other

70 m pf 163 987 37 841 79.4 5.3 1.6 10.5 2.1 1.1 2823
70 m st 275 575 188 452 87.3 6.4 3.2 1.3 0.7 1.1 20 268
110 m pf 142 140 38 957 84.6 3.0 1.1 8.6 1.7 1.0 3190
110 m st 238 182 151 560 90.5 2.6 4.3 1.1 0.7 0.9 18 104
200 m pf 204 644 65 128 85.5 2.1 1.3 8.3 1.9 1.0 6081
200 m st 253 507 170 856 92.5 1.6 3.8 0.9 0.5 0.7 22 837
1000 m pf 209 778 31 332 72.1 3.4 1.1 21.3 1.4 0.7 2205
1000 m st 173 089 113 693 73.5 20.2 3.7 1.3 0.5 0.9 10 132

Abbreviations: pf, prefilter; st, sterivex.

a
No. of reads matching coding genes in the NCBI-nr database (above bit score 50) % percentage of coding reads matching reference genes from
bacteria, archaea, unidentified prokaryotes, eukaryotes, viruses or other (genes lacking a taxonomic annotation, or annotated as unknown).
b
No. of prokaryote reads assigned to SEED subsystem categories.

distinguish between prokaryote and eukaryote- this discrepancy between data types is unclear, but
derived viruses, it is possible that the prefilter may involve differences in the representation of
viruses originated from eukaryotes. In the following taxonomic groups between the Ribosomal Database
sections, we focus on protein-coding sequences of Project and NCBI-nr databases, in rRNA operon
prokaryotes. We first describe taxonomic patterns copy number among taxa, and in the phylogenetic
inferred from coding gene annotations in contrast to resolution between protein-coding and rRNA gene
those based on amplicon data, and then highlight fragments. Indeed, both the 16S and coding gene
functional categories involved with key OMZ data sets contained large numbers of sequences that
biogeochemical processes, as well as categories that could not be assigned to a bacterial phylum
differed notably between size fractions. (Supplementary Figure S5). Additional reference
sequences, including whole genomes of OMZ
Taxonomic compositionprotein-coding genes. microorganisms, will likely increase the probability
The taxonomic identities of bacterial protein-coding and accuracy of read assignment, and potentially
genes broadly reflected those inferred from 16S resolve differences between rRNA and coding gene-
rRNA gene amplicons, with important caveats. For based identifications. Until then, however, studies
this comparison, certain phylogenetic groups should account for the chance that community
detected in the amplicon data sets (Figure 1) composition shifts can be underestimated or mis-
were collapsed to higher taxonomic levels to interpreted when based on metagenome data.
match groupings based on protein-coding genes Protein-coding gene annotations can nonetheless
(Supplementary Figure S5). Analysis of 16S ampli- provide taxonomic insight by identifying groups not
cons from the four depths where metagenomes were well represented in 16S databases. For example, in
sampled (70, 110, 200 and 1000 m; all depths contrast to the amplicon data, which revealed
combined) revealed 10 major bacterial divisions an overall enrichment of Deltaproteobacteria on
(out of 20) that were at higher relative abundance on prefilters, coding sequences matching the ubiqui-
prefilters. Of these, nine also were enriched in tous deltaproteobacterial SAR324 cluster were 2 to
prefilters based on protein-coding data sets 12-fold more abundant on Sterivex filters compared
(Supplementary Figure S5). However, the magni- with prefilters (Supplementary Figure S8). Notably,
tude of this enrichment was markedly higher in SAR324-like reads peaked at 44% of total bacterial
comparisons using amplicon data (Supplementary coding reads in the Sterivex fraction at 1000 m,
Figure S5). For example, in the amplicon analysis, consistent with reports of the mesopelagic distribu-
the odds of detecting a deltaproteobacterial tion of this group (Wright et al., 1997). SAR324
sequence were 11-fold greater in the prefilter enrichment in the free-living fraction contrasts with
relative to the Sterivex fraction. In contrast, when recent genomic evidence indicating that this group
inferred from protein-coding sequences, these odds is adapted for a particle-associated lifestyle (Swan
were effectively equal between filter fractions (odds et al., 2011). However, this lineage also contains
ratio: 1.1). Similar patterns were evident for the chemoautotrophic members (Swan et al., 2011),
Planctomycetes, Spirochetes, Tenericutes and Epsi- which presumably would have less of a need to
lon- and Betaproteobacteria (Supplementary Figure attach to organic-rich particles.
S5C, F). On average, the major bacterial divisions Coding genes suggested that the archaeal commu-
were more evenly represented in the metagenome nity also differs markedly between size fractions
data (Simpsons E: 0.45 and 0.36 for prefilter and (Supplementary Figure S6). The Sterivex fraction
Sterivex, respectively) compared with the amplicon was enriched up to sixfold (mean 3.3) in reads
data (Simpsons E: 0.32 and 0.21 for prefilter and matching aerobic ammonia-oxidizing autotrophs of
Sterivex) (Supplementary Figure S5). The cause of the Thaumarchaeota, with the highest representation

The ISME Journal

 Size-fractionated metagenomics
S Ganesh et al
200
of Thaumarchaea in the suboxic 70 m sample at Protein-coding sequences provided additional
the top of the OMZ. A similar enrichment in the insight into the taxonomic composition of the
small (0.10.8 mm) size fraction was reported for OMZ Planctomycete community. In contrast to the
low-oxygen (B20 mM) waters in the North Pacific amplicon data, sequences matching Planctomycete
(Smith et al., 2013), highlighting the free-living genes were a large component of metagenomes from
lifestyle of this group, as well as adaptation to the free-living fraction, peaking at a high of 9% of
low-oxygen conditions. Similarly, reads matching identifiable coding reads at the 110 m OMZ depth
Crenarchaeota were 2 to 11-fold (mean 4.2) more before declining to 2% beneath the OMZ
abundant in the free-living size fraction in the ETSP (Supplementary Figure S7). This pattern agrees with
OMZ, peaking at 15% of total coding reads at 1000 m. prior metagenomic data from the ETSP OMZ
The increase in Crenarchaeota with depth agrees (Stewart et al., 2012b). The taxonomic identities of
with patterns from other subtropical and tropical Planctomycete-like coding genes differed signifi-
sites (DeLong, 2003). Indeed, the overwhelming cantly between size fractions and between OMZ
majority of Crenarchaeota sequences (91%) from and non-OMZ depths. Notably, Planctomycete
1000 m matched fosmid clones representing Group I sequences from Sterivex filters predominantly
Crenarchaeota collected at 4000 m in the Pacific matched the marine anammox genus Candidatus
Ocean (Konstantinidis and DeLong, 2008). Func- Scalindua, whose sequences constituted 5981%
tional genes on these fosmids indicate a role in of Planctomycete reads from OMZ depths
aerobic ammonia oxidation, suggesting the potential (Supplementary Figure S7). In comparison, Candi-
for re-classifying these sequences as Thaumarchaeota datus Scalindua represented 1420% of Planctomy-
(Brochier-Armanet et al., 2008). In contrast, cete reads in the prefilters at these depths, which
sequences matching Euryarchaeota, which peaked were instead enriched in the non-anammox genus
in abundance in the upper OMZ samples (70 m), Planctomyces, as also indicated by the amplicon
were on average 50% more abundant on prefilters. Of data. Analyses with genus-specific FISH probes
these sequences, those matching uncultured marine previously showed that a minor fraction of total
group II (MG-II) constituted the single largest fraction Candidatus Scalindua cells in the Namibian and
(mean: 40% of total Euryarchaeota sequences; ETSP OMZs associates directly with particles
Supplementary Figure S6). Recently, sequencing of (Woebken et al., 2007), consistent with our data. In
a MG-II euryarchaeote genome indicated a motile contrast, marker gene surveys from the suboxic
heterotrophic lifestyle, with genes for proteins Black Sea detected this genus in the 0.230 mm
mediating adhesion, fatty acid metabolism and fraction but not in the larger particle-associated
protein degradation suggesting adaptations to growth fraction above 30 mm (Fuchsman et al., 2012).
on marine particles (Iverson et al., 2012). Consistent As Candidatus Scalindua is presumed to be the
with this prediction, MG-II-like sequences were at 26 primary lineage responsible for anammox in OMZs
to 82% higher relative abundance in the prefilter (Woebken et al., 2008; Galan et al., 2012), these
fraction in the OMZ and oxycline depths (70, 110 and patterns suggest that the bulk of anammox-capable
200 m). Other Euryarchaeota sequences matched cells in OMZs may be spatially separated at the
diverse clades, including those with methanogenic microscale from potentially linked metabolic trans-
members, which were marginally enriched (1544%) formations on particles, for example, nitrite produc-
on prefilters (Supplementary Figure S6). Methano- tion and ammonia remineralization by heterotrophic
genic and potentially hydrogenotrophic Euryarch- denitrifiers. A shift to a higher proportion of free-
aeota have previously been detected on marine living Candidatus Scalindua may be facilitated in
particles (van der Maarel et al., 1999; Ditchfield OMZs where suboxia extends beyond particle
et al., 2012), suggesting that anoxic microzones on microniches and also by the autotrophic metabolism
particles create conditions favorable for these anae- of this organism, which may eliminate pressure to
robic taxa. It also has been hypothesized that attach to carbon-rich particles.
methanogens in OMZs may occur in symbiotic
associations with anaerobic protists (Orsi et al., Trends in SEED subsystems. Classification of
2012), as observed in other diverse reducing habitats sequences into SEED subsystems highlighted varia-
(Nowack and Melkonian, 2010; Edgcomb et al., tion in functional content between filter fractions.
2011). Overall, however, data directly comparing Subsystem abundances were correlated (R40.94)
marine Archaea between free-living and particle- between samples (Figure 2, inset), and BrayCurtis
associated niches remain limited, and conflicting. distances between prefilter and Sterivex SEED
Moderate differences in archaeal community compo- profiles (mean: 0.11) did not differ appreciably from
sition have been observed between size fractions at those between samples of the same filter type (mean:
coastal sites (Crump and Baross, 2000; Galand et al., 0.10 and 0.07 for Prefilter-only and Sterivex-only
2008, Smith et al., 2013), but not at an open ocean comparisons, respectively; Supplementary Table S3).
site (Galand et al., 2008). Our data indicate size This similarity reflects redundancy in house-
fraction-specific variation, suggesting the need for keeping gene categories, notably protein biosynthesis,
more targeted studies exploring potential archaeal DNA repair and central carbohydrate metabolism
genotype and ecotype variation between microniches. (Figure 2), which are ubiquitous and abundant across

The ISME Journal

 Size-fractionated metagenomics
S Ganesh et al
201
even highly divergent taxa (Burke et al., 2011). significantly higher proportions, with enrichment
However, despite broad similarity in subsystem driven by genes for branched chain amino-acid
profiles, filter fractions clustered independently transporters (Figure 2), which accounted for 65%
of one another based on SEED content (Figure 2, (average) of all sequences in this category and were
inset), suggesting functional differences separating two to threefold more abundant in the Sterivex
free-living from surface-attached life history fraction across depths (Supplementary Table S4).
modes. Consistent with this pattern, Sterivex metagenomes
The prefilter fraction was enriched in genes for contained higher fractions of genes encoding the
navigating and persisting within a spatially and guanosine 50 ,30 bispyrophosphate (ppGpp)-controlled
chemically heterogeneous environment. Figure 2 stringent response. Induced under diverse stress
(main) shows the odds of a SEED subsystem conditions, notably amino-acid starvation, the strin-
occurring in the prefilter relative to the Sterivex gent response regulates a shift from growth-supporting
fraction. These trends are based on data pooled functions (for example, stable RNA synthesis, transla-
across depths (that is, the four depths are treated as tion and cell division) to those enabling survival
replicates) and therefore only identify categories under growth limitation, such as amino-acid
that consistently differed between filter fractions. biosynthesis and DNA replication (Cashel et al.,
Genes mediating motility, chemotaxis and adhesion 1996; Durfee et al., 2008; Traxler et al., 2008). The
were among the most overrepresented on prefilters free-living fraction was also overrepresented in genes
(Figure 2, Supplementary Table S4). These functions for electron transport and energy generation, notably
are presumably critical for detecting local patches of components of fermentation and respiration path-
nutrients and energy substrates (Fenchel, 2002; ways. These included formate dehydrogenase, which
Stocker et al., 2008), colonizing surfaces (Fenchel, in some bacteria is used for respiratory nitrate
2001), and potentially also for navigating substrate reduction (Sawers, 1994), as well as genes for
gradients on particles themselves. Notably, prefilters assimilatory and respiratory nitrogen and sulfur
contained significantly higher counts of genes metabolism (for example, ammonia assimilation,
involved in bacterial secretion. Of these, 97% nitrate and nitrite ammonification; discussed in more
encoded elements of Type IV secretion systems, detail below). Genes for biosynthesis and cell division
notably Type IV pili (84%) and mannose-sensitive were also at higher abundance, as were genes for CO2
hemagglutinin Type IV pili (8%). These cell surface fixation, suggesting a propensity for autotrophic cells
structures mediate diverse functions, including gene to be decoupled from organic-rich particles.
exchange, transfer of effector proteins between cells, Together, these patterns indicate distinct micro-
twitching motility and adherence (Christie et al., bial life history strategies. Free-living bacteria
2005; Burrows, 2012), and have been shown to exhibit an overall greater investment in genes
promote attachment to algal surfaces by marine mediating core cellular functions and growth. These
bacteria (Dalisay et al., 2006). include adaptations for metabolic regulation under
Prefilters also were enriched in genes encoding potential substrate limitation and a significant
virulence and antibiotic resistance functions. Life in investment in mechanisms for the uptake of
particle-associated biofilms would presumably low-molecular weight compounds (that is, dissolved
increase the frequency of cell-to-cell contact, and organic carbon), a pattern consistent with metagen-
therefore the likelihood of antagonistic interactions. omes from free-living bacteria at other ocean sites,
Indeed, the production of antibacterial compounds including members of the SAR11 clade (Kirchman,
is more common in surface-attached bacteria rela- 2003; Malmstrom et al., 2005; Poretsky et al., 2010),
tive to planktonic cells (Long and Azam, 2001; Long which were well represented in our OMZ samples.
et al., 2005; Gram et al., 2010) and has been shown In contrast, prefilter-associated cells are more likely to
to affect the colonization dynamics of marine encode functions for signal recognition and cell-to-cell
particles (Grossart et al., 2003). Prefilter metagen- interactions, presumably key adaptations for detecting
omes also contained high abundances of genes and colonizing organic-rich particles and for life in
mediating signaling via the universal secondary close proximity to neighbors. Several of these trends
messenger cAMP, which in prokaryotes regulates are broadly consistent with functional genomic
functions ranging from virulence, stress response and differences separating copiotrophic and oligotrophic
energy and carbon metabolism. A high abundance of life history strategies (Lauro et al., 2009). These trends
cAMP signaling genes was shown recently for soil are detected here at the community-level across
metagenomes (Delmont et al., 2012) and may be a multiple depth zones. These patterns indicate that
general feature of bacterial communities on surfaces metagenome-based inferences about the relative
with fluctuating (spatially, temporally) substrate importance of microbial traits in marine environments
conditions and potentially high cell densities. will vary depending on the ratio of particle-associated
Compared with prefilter metagenomes, Sterivex to free-living bacteria in a sample.
metagenomes were proportionally enriched in
genes for substrate acquisition, energy and nutrient Marker genes of nitrogen and sulfur metabolism.
metabolism, and cell growth. Notably, genes Analysis of marker genes suggests that key OMZ
encoding ATP-binding cassette transporters were at metabolic processes may be partitioned between

The ISME Journal

 Size-fractionated metagenomics
S Ganesh et al
202
particle-associated and free-living microbial com- Denitrification genes were also differentially
munities. Results of BLASTX against the NCBI-nr partitioned between fractions. Sequences encoding
database were manually queried to determine the NarG, catalyzing nitrate reduction to nitrite, were
relative abundances of target genes of nitrogen enriched in both fractions at OMZ depths compared
and sulfur energy metabolism, some of which are with the oxic 1000 m sample, but were consistently,
not well represented in the SEED hierarchy. albeit marginally, overrepresented in the free-living
Abundances are shown in Figure 3, normalized to fraction (Figure 3). This pattern was reversed at the
gene length and to the abundance of a universal, oxic 1000 m depth, where narG was 13-fold
single-copy gene (rpoB). more abundant in the prefilter compared with the
Genes of dissimilatory nitrogen oxidation (ammo- Sterivex. A similar pattern was observed in meta-
nia and nitrite oxidation) exhibited variable abun- genomes from the coastal North Pacific, where narG
dance but were generally overrepresented in the at a high oxygen site was enriched in larger size
small size fraction (Figure 3, inset). Both hzo and fractions (0.8300 mm) compared with a 0.10.8 mm
amoC, markers for anammox and aerobic ammonia fraction, but occurred at relatively uniform abun-
oxidation respectively, were enriched on average dance across fractions at low-oxygen (B20 mM) sites
approximately fourfold in Sterivex metagenomes. (Smith et al., 2013). These studies suggest that
Hzo sequences were detected only in the OMZ nitrate respiration in oxic water columns is confined
depths and were closely related to hzo of Candida- primarily to suboxic microniches on particles, but
tus Scalindua, consistent with the anaerobic nature under low-oxygen conditions is utilized by both
of anammox and the overall distribution of Scalin- free-living and surface-attached cells. In both filter
dua-matching protein-coding reads (Figure 3 and fractions within the ETSP OMZ, a substantial
Supplementary Figure S7). Sequences matching proportion (4250%) of the narG pool matched
amoC were affiliated exclusively with the Thau- sequences from an uncultivated bacterium in candi-
marchaeota and Crenarcheaota, supporting recent date division OP1 isolated from a subsurface
evidence that nitrification in the OMZ is mediated thermophilic microbial mat community (Takami
primarily by Archaea (Stewart et al., 2012b). et al., 2012). Although relatives of the OP1 division
In contrast, the relative abundance of nxrB, a have been detected in OMZs (Stevens and Ulloa,
marker for aerobic nitrite oxidation, did not vary 2008; Wright et al., 2012), their contributions to
substantially between size fractions (Figure 3, inset). OMZ biogeochemistry remain uncharacterized.
However, fraction-specific patterns were evident Nitrite reductase genes (nirK, nirS and nrfA),
when the nxrB pool was evaluated according to involved in both denitrification and anammox,
taxonomic affiliation. The majority (55%) of all nxrB exhibited contrasting distributions between size
sequences were most closely related to nxrB of Ca. fractions. In the OMZ depths, nirK genes encoding
Nitrospira defluvii, a nitrite oxidizer isolated from the copper-containing enzyme were distributed
activated sludge (Spieck et al., 2006). Nitrospira relatively evenly among filter fractions at all depths,
nxrB abundance peaked at the oxycline base (70 m) excluding the 1000 m sample (Figure 3), whereas
in both filter fractions. At the lower depths, nirS genes for the cytochrome cd1-containing nitrite
including at the anoxic OMZ core, Nitrospira nxrB reductase were most abundant on prefilters from the
was found exclusively in the free-living fraction OMZ core (110, 200 m). In contrast, nrfA sequences,
(Figure 3 bottom). Subsequent to our analysis, the indicators of dissimilatory nitrate reduction to
genome of the nitrite oxidizer Nitrospina gracilis ammonium (Simon, 2002; Jensen et al., 2011), were
was published (Lucker et al., 2013). Nitrospina is confined almost exclusively to the free-living size
the dominant nitrite-oxidizer genus in the oceans, fraction and were affiliated predominantly with
and the N. gracilis genome is closely related members of the Chlamydiae and Deltaproteobacteria.
evolutionarily to that of Ca. Nitrospira defluvii Genes involved in the two terminal steps
(Lucker et al., 2013), raising the possibility that re- of denitrification, the reduction of nitric oxide
analysis of our data may instead classify OMZ to nitrous oxide (norB/norZ) and nitrous oxide to
Nitrospira-like sequences as belonging to Nitrospina. dinitrogen (nosZ), exhibited among the strongest
Both Nitrospina and Ca. Nitrospira genomes size fraction-specific patterns, being on average
show adaptations to low-oxygen environments fourfold more abundant on prefilters. Prefilter
(Lucker et al., 2010, 2013). This is consistent with enrichment of these genes is consistent with recent
the distribution of related sequences in the ETSP metagenome data from an estuarine site (Smith
and other low-oxygen zones (Labrenz et al., 2007; et al., 2013) and with studies linking N2O and N2
Jorgensen et al., 2012), and with the recent detection production or nor/nos expression with diverse
of nitrite oxidation (by Nitrospina or Nitrococcus surface-attached environments (for example,
bacteria) under low oxygen (O2 o1 mM) in the OMZ sediments, soils, algal epibiont communities; Scala
off Namibia (Fussel et al., 2012). Together, these and Kerkhof, 1998; Rosch et al., 2002; Long et al.,
studies indicate a role for nitrite oxidation in the 2013), including suspended particles and the epi-
suboxic zones of the upper OMZ, potentially by biotic communities of marine algae (Michotey and
diverse bacteria with distributions varying vertically Bonin, 1997; Wyman et al., 2013). Here, taxonomic
and at the microscale. partitioning between filter fractions was evident,

The ISME Journal

 Size-fractionated metagenomics
S Ganesh et al
203
notably for the norB/Z pool. The majority (8393%) of contributions to the aprA and dsrA pools (14% and
norB/Z sequences on Sterivex filters were most 20%, respectively). These patterns suggest a complex
closely related to nitric oxide reductase from the sulfur-oxidizing community, with distinct pathways
anammox planctomycete Candidatus Scalindua of sulfur oxidation mediated by different taxa.
profunda (scal02135). It has been hypothesized that Together, indicator gene distributions highlight
in this bacterium NorB may act to relieve oxidative the potential that key metabolic processes are
stress (van de Vossenberg et al., 2012), as opposed to partitioned spatially between free-living and sur-
functioning in energy metabolism. In contrast, face-associated microbial communities. We antici-
large proportions (4771%) of particle-associated pated that genes traditionally associated with
norB/Z sequences at the OMZ core (110 and 200 m) autotrophic processes (for example, anammox,
were most closely related to nitric oxide reductases aerobic nitrification and sulfur oxidation) would be
of Ca. Methylomirabilis oxyfera (CBE69496.1 and more prevalent in the free-living fraction, presuming
CBE69502.1), a member of the NC10 candidate that other metabolic substrates are not limiting in
division originally enriched from sediment the open water column. In contrast, surface attach-
(Figure 3). Ca. M. oxyfera oxidizes methane under ment would potentially benefit heterotrophic taxa
anaerobic conditions using O2 generated intracellu- (for example, heterotrophic denitrifiers) by provid-
larly through an alternative denitrification pathway ing a localized source of organic substrates. A subset
involving the dismutation of nitric oxide into of the data support this prediction (for example, hzo,
dinitrogen and O2, potentially via a Nor enzyme amoC, aprA, norB and nosZ), whereas the distribu-
acting as a dismutase (Raghoebarsing et al., 2006; tion of other genes is less uniform. In one of the only
Ettwig et al., 2010). Here, sequences matching Ca. studies to compare bacterioplankton metagenomes
M. oxyfera genes were recovered across all across size fractions (Smith et al., 2013), genes of
four depths (31212 distinct genes per sample; anaerobic metabolism, including nar, nir, nor and
0.110.23% of total prokaryotic coding reads), the reductive variant of dsr, were enriched in
raising the possibility that Methylomirabilis-like particle-associated cell fractions (relative to free-
organisms may contribute to methane cycling in living fractions) at an oxic site, but not at hypoxic
these waters. sites. These data, interpreted alongside our results,
Genes of dissimilatory sulfur metabolism also had confirm that the distribution of metabolic functions
variable distribution patterns. The gene (aprA) between particle-associated and free-living niches
encoding APS reductase, which controls the AMP- varies among sites and is likely driven by the oxygen
dependent oxidation of sulfite to APS but also acts and substrate conditions of the surrounding water
reversibly during sulfate reduction, was consistently column, combined with the composition of the
enriched (twofold) in the free-living fraction. Other particles themselves.
genes of sulfur metabolism did not show strong
fraction-specific trends. These included the sulfur Mobile element genes. Prefilter communities
oxidation gene soxB, an indicator of thiosulfate were significantly enriched in genes mediating
oxidation, as well as the dissimilatory sulfite mobile element activity via transposition and phage
reductase gene dsrA, which is present in diverse integration. Many of these genes were not recovered
sulfate reducers (notably Deltaproteobacteria), as during functional analysis of BLASTX results using
well as in chemolithotrophic sulfide oxidizers MEGAN, presumably due to limited representation
(Dhillon et al., 2005), in which Dsr operates in the in the SEED classification. However, manual parsing
reverse direction. Here, aprA and dsrA sequences of significant BLASTX matches revealed that genes
matching known sulfate-reducing taxa (determined encoding transposases, which catalyze the move-
based on groupings by Canfield et al. (2010) ment of DNA segments (transposons) within gen-
constituted minor fractions of the total aprA and omes, were three to sixfold more abundant in
dsrA pools (3% and 0%, respectively) and did not prefilter compared with Sterivex metagenomes, with
exhibit clear size fraction-specific distributions, peak abundances at the OMZ core (110 and 200 m;
although such patterns could be obscured by the Figure 3). At these depths, transposase genes were
low representation of these sequences. The majority among the most abundant in the data sets, being
of aprA and dsrA sequences instead matched either up to 50-fold more abundant than the reference
known sulfur-oxidizing taxa, or groups for which rpoB and constituting up to 2% of identifiable
the functional role of these genes is uncertain coding genes. By 1000 m, transposase abundance
(for example, aprA in the Alphaproteobacterium had declined by B80% of peak values. A similar
Pelagibacter). However, taxonomic composition enrichment on prefilters and at OMZ depths was
differed between sulfur metabolism genes. For exam- observed for genes encoding integrases that mediate
ple, sequences matching the SUP05 clade, containing site-specific recombination of phage DNA, typically
free-living sulfur oxidizers from low-oxygen pelagic via an RNA intermediate. This pattern is consistent
environments as well as thiotrophic deep-sea vent with the observed prefilter enrichment of viral
symbionts (see Gamma_S in Figure 3; Walsh et al., DNA (Table 1), as well as a 2.5-fold enrichment of
2009), were relatively abundant among soxB restriction modification systems and restriction
sequences (49% of total), but made relatively minor enzymes in prefilters at OMZ depths (70, 110 and

The ISME Journal

 Size-fractionated metagenomics
S Ganesh et al
204
200 m; Supplementary Table S4), suggesting an This raises the possibility that transposase enrich-
increased need for protection against foreign DNA. ment on prefilters is independent of taxonomic
The transposase and integrase gene pools matched identity and may instead be a consequence of the
database genes belonging to taxonomically diverse microenvironment itself. (This is similar to the
Bacteria and Archaea (Figure 3, bottom), with enrichment of transposes as a function of depth in
compositions resembling those inferred from all the water column, where deep-sea conditions seem
coding sequences and from 16S gene amplicons to favor the prevalence of transposonssee below
(Figure 1 and Supplementary Figure S5), notably (DeLong et al., 2006; Konstantinidis et al., 2009)).
with large contributions from groups abundant in Presumably, the spread of transposons and other
both filter fractions (for example, Alpha- and mobile elements to new host genomes could be
Gammaproteobacteria, Bacteroidetes). The majority facilitated by growth in particle-associated biofilms
of prefilter transposable element genes therefore where cells, relative to those in a planktonic form,
appear to originate from the bulk community and occur in close spatial proximity and have an
not disproportionately from those groups that increased likelihood of encountering foreign DNA
appear most strongly overrepresented in this frac- (for example, via conjugation or via phage or naked
tion (for example, Planctomycetes), suggesting that DNA caught in the biofilm matrix; Madsen et al.,
the observed enrichment is a community-wide 2012). In this scenario, the ubiquity of transposases
phenomenon. in prefilter communities would not preclude the
These results reinforce emerging evidence for the possibility that transposition is selectively deleter-
global significance of transposition in diverse ious, and may instead simply reflect an increased
microorganisms. In a survey of over 10 million rate of mobile element transfer among and within
annotated genes or gene fragments, transposases genomes (Werren, 2011).
were determined to be the single most abundant Alternatively, transposition may be beneficial for
protein-coding genes in sequenced genomes and particle-associated microbes if it generates genetic
environmental metagenomes, and to be ubiquitous variants that enhance host fitness, or provides raw
across all domains of life (Aziz et al., 2010). Notably, genetic material for evolving new functions (Aertsen
transposases constituted up to 8% of all sequences and Michiels, 2005; Chou et al., 2009). Marine
in a deep-sea hydrothermal vent biofilm (Brazelton particles can contain steep gradients in nutrient or
and Baross, 2009) and were overrepresented in an redox substrate availability, locally high concentra-
algal-associated community compared with plank- tions of bacteriocidal chemicals produced by neigh-
tonic bacteria (Burke et al., 2011). In addition, boring microbes, and increased incidences of
transposase gene transcription in oral spirochete antagonism (Alldredge and Cohen, 1987; Long and
bacteria was upregulated during growth as a biofilm Azam, 2001; Long et al., 2005), consistent with the
(Mitchell et al., 2010). These patterns, interpreted enrichment of virulence, motility and chemotaxis
alongside our data, indicate a significant potential genes observed in OMZ prefilter metagenomes
for transposase-mediated gene transfer in the (Figure 2). Unlike free-living bacterial cells, parti-
genomes of surface-attached microbial communities. cles may sink through the water column, exposing
It is unclear why transposases are enriched in attached microbes to changing physical and biolo-
surface-attached communities. Transposable ele- gical conditions along the depth gradient. Zooplank-
ment activity is often considered harmful for ton feeding on particles also has the potential to
genomes, as transposition can inactivate genes or rapidly shift the local environment for particle-
alter chromosome structure (Mahillon et al., 1999; associated microbes. Such spatial and temporal
Kidwell and Lisch, 2001). Under this assumption, variability may select for genomes with an elevated
prefilter enrichment of transposases suggests a potential for acquiring new functions via lateral
reduced capacity of particle-associated microbes to gene movement. Indeed, transposition in bacteria
selectively purge mobile elements. This could occur has been shown to be induced by environmental
in taxa with reduced effective population sizes, as change (Ohtsubo et al., 2005; Twiss et al., 2005) and
has been hypothesized to explain an increase of to facilitate adaptive responses to repeated cycles of
transposase abundance with depth in free-living growth and stress (Sleight et al., 2008).
marine microbes (Konstantinidis et al., 2009) and to The vertical distribution of transposase genes in
partly explain mobile element accumulation in the OMZ, while limited to only four depths,
symbiont genomes (McCutcheon and Moran, 2011). suggests interesting contrasts to patterns observed
As of yet, however, there is not strong evidence that in other oceanic regions. Metagenome analysis of
particle-associated microorganisms have reduced the free-living microbial fraction revealed a steady
population sizes compared with free-living taxa, increase in the relative abundance of transposase
although the prefilter fraction presumably is more genes with depth down to 4000 m in the
likely to contain symbionts (for example, of marine oligotrophic North Pacific off Hawaii (DeLong
protists). Here, the taxonomic composition of the et al., 2006; Konstantinidis et al., 2009). An increase
transposase pool was broadly similar between filter in both relative transposase abundance and tran-
fractions (Figure 3) and included groups both with scription, similar in magnitude to that recorded off
and without strong size fraction-specific tendencies. Hawaii over a 4000 m gradient, has been observed

The ISME Journal

 Size-fractionated metagenomics
S Ganesh et al
205
previously in the free-living microbial community of structure than depth-specific environmental varia-
the ETSP OMZ (Stewart et al., 2012b), but over a tion (Figure 1). This was unexpected given studies
much narrower depth range. Indeed, the latter study showing the strong influence of vertical environ-
sampled to a depth of only 200 m (OMZ core). In the mental gradients (for example, oxygen, nutrient
current study, transposase abundance in both filter availability) in shaping OMZ community structure
fractions spiked within the OMZ, but then declined (Bryant et al., 2012), but agrees with taxon-specific
substantially by 1000 m. Given the patterns we (for example, Vibrio) analyses identifying particle-
observe in our size fraction analysis, transposase association as a major driver of ecological niche
abundance in the free-living fraction may be diversification in marine bacteria (Hunt et al., 2008).
influenced by the relative load of particulate However, in other ocean regions, microbial biomass
material in the water column (and therefore the per unit volume has been shown to be substantially
proportion of the total community that is surface- higher for the smallest size fraction (Rosso and
attached). This may be possible if sampling causes Azam, 1987; Cho and Azam, 1988; Karl et al., 1988),
the detachment of particle-associated cells, which suggesting that analysis of the bulk OMZ commu-
are then sampled as part of the free-living fraction. nity (all size fractions combined) may show patterns
Indeed, the vertical distribution of transposase of community clustering resembling those observed
abundance in ETSP metagenomes roughly parallels for the small size fraction. Measurements of live cell
that of particulate backscattering, with peaks in the abundance and activity among different size frac-
upper oxycline and suboxic core and a decline with tions in the OMZ are clearly needed.
depth out of the OMZ (Van Mooy et al., 2002; Particle-associated microbial communities also
Whitmire et al., 2009; Supplementary Figure S1). were markedly enriched in functions controlling
The enrichment of transposable element genes genetic exchange. This is consistent with experi-
within OMZ metagenomes may coincide with mental analyses of cultured bacteria showing that
distributions of ETSP OMZ viruses (Table 1). life on surfaces (in biofilms) is an important route
Cassman et al. (2012) described atypically low ratios for diversification (Boles et al., 2004; Boles and
of free-living (non-surface-attached) viruses to Singh, 2008). Although the adaptive value for
microbes in the Chilean OMZ, compared with biofilm-mediated diversification is unclear, our data
surface waters and to other marine regions. The suggest that particle-associated communities may be
authors hypothesized a transition from lytic to hotbeds for genome reshuffling, with a strong role
lysogenic viral lifestyles within the anoxic zone, for horizontal gene transfer via transposition. Future
potentially linked to low microbial division rates. studies should test the hypothesis that rates of
An enhanced role for lysogeny in the OMZ may be genetic exchange differ between particle-associated
consistent with the vertical distribution of transpo- and free-living marine microbes.
sases and integrases in the particle fraction, notably Key metabolic pathways mediating OMZ elemental
if these genes originate from prophage. Alternatively, cycling also were differentially partitioned between
it is possible that viral and mobile element genes size fractions. This implies that particle abundance,
detected in the prefilter metagenomes originated from and by extension the percentage of the total
lysed viral particles entrained within surface-asso- microbial community attached to surfaces, may
ciated biofilms. Collectively, our data indicate an have an important impact on community biogeo-
enhanced potential for mobile element activity in chemical transformations. For example, genes facil-
surface-associated OMZ bacterioplanktonit remains itating the last two steps of denitrification were
to be clarified how this potential is linked to phage markedly enriched on particles, raising the possibi-
activity on particles. lity that bulk denitrification rates could be corre-
lated with particle load. However, the relative
abundance of genes and taxa is an imperfect
Conclusions predictor of functional significance. For example,
Deltaproteobacteria of the genus Desulfosporosinus
Metagenomic comparisons of different bacterio- constituted 0.006% of microbial cell counts in a
plankton size fractions are (surprisingly) rare but peatland ecosystem but were responsible for the
are necessary for clarifying the distribution of majority of sulfate reduction in the community
functional traits in marine microbes. OMZ microbial (Pester et al., 2010). Such results highlight the need
communities from the particle-associated size frac- to directly couple genetic and process rate char-
tion were consistently overrepresented in functions acterizations across multiple size fractions to deter-
suggesting adaptation to surface attachment and mine how the activity of particle-associated
growth in a biofilm, notably genes mediating communities contributes to OMZ biochemical cycling.
colonization and cellcell interactions. This pattern Resolving questions about the role of particle-
highlights fundamental divisions between particle- associated microbes in OMZs should be a research
associated and free-living life history modes and priority, but will require standardizations of sam-
correlates with a strong partitioning of taxonomic pling design and methodology. As our study focused
groups between fractions. Indeed, association with on a single site, conclusions about partitioning
prefilters was a stronger predictor of community between fractions are most robust for genes and taxa

The ISME Journal

 Size-fractionated metagenomics
S Ganesh et al
206
with consistent patterns across depths. However, of genes encoding 16S rRNA. Appl Environ Microb 65:
these data, and data from other recent metagenome 514522.
studies (Smith et al., 2013), also suggest that the Aertsen A, Michiels CW. (2005). Diversify or die:
level of partitioning varies with depth and environ- generation of diversity in response to stress. Crit Rev
mental (for example, oxygen) conditions Microbiol 31: 6978.
additional vertical profiles from diverse low-oxygen Alldredge AL, Cohen Y. (1987). Can microscale chemical
patches persist in the sea? Microelectrode study
sites will be required to statistically identify such
of marine snow, fecal pellets. Science 235: 689691.
trends. The potential for sampling to physically Alldredge AL, Silver MW. (1988). Characteristics,
disrupt particles should also be considered. It is dynamics and significance of marine snow. Prog
possible that some of the taxa identified in the free- Oceanogr 20: 4182.
living fraction originate from particles that are Allen LZ, Allen EE, Badger JH, McCrow JP, Paulsen IT,
broken apart during water collection (for example, Elbourne LDH et al. (2012). Influence of nutrients
during rosette bottle-firing, or pumping during the and currents on the genomic composition of
filtration step). If this occurs, inferences of commu- microbes across an upwelling mosaic. ISME J 6:
nity metabolism in the free-living fraction may be 14031414.
especially prone to bias in regions of high particle Aziz RK, Breitbart M, Edwards RA. (2010). Transposases
load. There is also a critical need to standardize are the most abundant, most ubiquitous genes in
sample treatment between process rate measure- nature. Nucleic Acids Res 38: 42074217.
Bano N, DeRae Smith A, Bennett W, Vasquez L,
ments and genetic characterizations. Rate measure- Hollibaugh JT. (2007). Dominance of mycoplasma in
ments are often based on incubations (for example, the guts of the long-jawed mudsucker, Gillichthys
in exetainers) of the bulk water (for example, mirabilis, from five California salt marshes. Environ
Dalsgaard et al., 2012), whereas the majority of Microbiol 9: 26362641.
OMZ molecular studies have focused on microbes Bauld J, Staley JT. (1976). Planctomyces maris sp. nov.,
retained after prefiltration (for example, Stevens and marine isolate of Planctomyces blastocaulis group of
Ulloa, 2008; Stewart et al., 2012a,b). Furthermore, budding bacteria. J Gen Microbiol 97: 4555.
different studies use different filter pore-size cutoffs, Bengtsson MM, vreas L. (2010). Planctomycetes domi-
making direct comparisons challenging. As marine nate biofilms on surfaces of the kelp Laminaria
particles differ significantly in age, size, organic hyperborea. BMC Microbiol 10: 261.
substrate contents and redox state, conclusions about Benjamini Y, Hochberg Y. (1995). Controlling the false
particle-associated communities should involve uni- discovery rate: a practical and powerful approach to
multiple testing. J R Stat Soc Ser B 57: 289300.
form comparisons across multiple particle size Boles BB, Singh PK. (2008). Endogenous oxidative
fractions. These comparisons should be directly stress produces diversity and adaptability in biofilm
coupled to measurements of bulk particle load and communities. Proc Natl Acad Sci USA 105:
to gene expression and process rate characterizations 1250312508.
in order to fully understand how particle-associated Boles BR, Thoendel M, Singh PK. (2004). Self-generated
communities contribute to OMZ biochemical cycling. diversity produces insurance effects in biofilm
communities. Proc Natl Acad Sci USA 101:
1663016635.
Conflict of Interest Brazelton WJ, Baross JA. (2009). Abundant transposases
encoded by the metagenome of a hydrothermal
The authors declare no conflict of interest. chimney biofilm. ISME J 3: 14201424.
Brinkhoff T, Fisher D, Vollmers J, Voget S, Beardsley C,
Acknowledgements Thole S et al. (2012). Biogeography and phylogenetic
diversity of a cluster of exclusively marine myxo-
We thank Jess Bryant, Adrian Sharma, Jaime Becker and the bacteria. ISME J 6: 12601272.
captain and crew of the R/V Melville for help in sample Brochier-Armanet C, Boussau B, Gribaldo S, Forterre P.
collection, and Tsultrim Palden and Raghav Sharma for (2008). Mesophilic Crenarchaeota: proposal for a third
help in DNA library preparation and pyrosequencing. This archaeal phylum, the Thaumarchaeota. Nat Rev
work is a contribution of the Center for Microbial Microbiol 6: 245252.
Oceanography: Research and Education (C-MORE) and Brown MV, Philip GK, Bunge JA, Smith MC, Bissett A,
was made possible by generous support from the National Lauro FM et al. (2009). Microbial community structure
Science Foundation (1151698 to FJS and EF0424599 to in the North Pacific Ocean. ISME J 3: 13741386.
EFD), the Gordon and Betty Moore Foundation (EFD), and Bryant JA, Stewart FJ, Eppley JM, DeLong EF. (2012).
the Agouron Institute (EFD). Microbial community phylogenetic and trait diversity
decline steeply with depth in a marine oxygen
minimum zone. Ecology 93: 16591673.
Bull AT, Stach JEM, Ward AC, Goodfellow M. (2005).
Marine actinobacteria: perspectives, challenges, future
References directions. Anton Van Lee J M S 87: 6579.
Burke C, Steinberg P, Rusch D, Kjelleberg S, Thomas T.
Acinas SG, Anton J, Rodriguez-Valera F. (1999). Diversity (2011). Bacterial community assembly based on
of free- living and attached bacteria in offshore functional genes rather than species. Proc Natl Acad
western Mediterranean waters as depicted by analysis Sci USA 108: 1428814293.

The ISME Journal

 Size-fractionated metagenomics
S Ganesh et al
207
Burrows LL. (2012). Pseudomonas aeruginosa twitching among stratified microbial assemblages in the oceans
motility: type IV pili in action. Annu Rev Microbiol 66: interior. Science 311: 496503.
493520. DeLong EF. (2003). Oceans of archaea. ASM News 69:
Canfield DE, Stewart FJ, Thamdrup B, De Brabandere L, 503511.
Dalsgaard T, Delong EF et al. (2010). A cryptic sulfur Demiri A, Meziti A, Papaspyrou S, Thessalou-Legaki M,
cycle in oxygen-minimum-zone waters off the Chilean Kormas K. (2009). Abdominal setae and midgut
Coast. Science 330: 13751378. Bacteria of the mudshrimp Pestarella tyrrhena. Cent
Caporaso JG, Kuczynski J, Stombaugh J, Bittinger K, Eur J Biol 4: 558566.
Bushman FD, Costello EK et al. (2010). QIIME allows Dhillon A, Goswami S, Riley M, Teske A, Sogin M. (2005).
analysis of high-throughput community sequencing Domain evolution and functional diversification of
data. Nat Methods 7: 335336. sulfite reductases. Astrobiology 5: 1829.
Caron DA, Davis PG, Madin LP, Sieburth JM. (1982). Ditchfield AK, Wilson ST, Hart MC, Purdy KJ, Green DH,
Heterotrophic bacteria and bacterivorous protozoa in Hatton AD. (2012). Identification of putative
oceanic macroaggregates. Science 218: 795797. methylotrophic and hydrogenotrophic methanogens
Cashel M, Gentry D, Hernandez VJ, Vinella D. (1996). The within sedimenting material and copepod faecal
stringent response. In: Neidhardt FC (eds). Escherichia pellets. Aquat Microb Ecol 67: 151160.
coli and Salmonella: cellular and molecular biology. Durfee T, Hansen AM, Zhi H, Blattner FR, Jin DJ. (2008).
American Society for Microbiology Press: Washington, Transcription profiling of the stringent response in
DC, USA, pp 14581496. Escherichia coli. J Bacteriol 190: 10841096.
Cassman N, Prieto-Davo A, Walsh K, Silva GG, Angly F, Edgcomb V, Leadbetter ER, Bourland W, Beaudoin D,
Akhter S et al. (2012). Oxygen minimum zones Bernhard JM. (2011). Structured multiple endosym-
harbour novel viral communities with low diversity. biosis of bacteria and archaea in a ciliate from
Environ Microbiol 14: 30433065. marine sulfidic sediments: a survival mechanism
Cho BC, Azam F. (1988). Major role of bacteria in in low oxygen, sulfidic sediments. Front Microbiol
biogeochemical fluxes in the oceans interior. Nature 2: 55.
332: 441443. Eloe EA, Shulse CN, Fadrosh DW, Williamson SJ,
Chou HH, Berthet J, Marx CJ. (2009). Fast growth increases Allen EE, Bartlett DH. (2010). Compositional
the selective advantage of a mutation arising recur- differences in particle-associated and free-living
rently during evolution under metal limitation. PLoS microbial assemblages from an extreme deep-ocean
Genet 5: e1000652.
environment. Environ Microbiol Rep 3: 449458.
Christie PJ, Atmakuri K, Krishnamoorthy V, Jakubowski S,
Ettwig KF, Butler MK, Le Paslier D, Pelletier E, Mangenot
Cascales E. (2005). Biogenesis, architecture, and
S, Kuypers MM et al. (2010). Nitrite-driven anaerobic
function of bacterial type IV secretion systems. Annu
methane oxidation by oxygenic bacteria. Nature 464:
Rev Microbiol 59: 451485.
543548.
Cottrell MT, Kirchman DL. (2000). Natural assemblages of
Faith DP. (1992). Conservation evaluation and phylo-
marine proteobacteria and members of the Cytophaga-
genetic diversity. Biol Conserv 61: 110.
Flavobacter cluster consuming low- and high-mole-
Fenchel T. (2001). Eppu si muove: many water column
cular-weight dissolved organic matter. Appl Environ
Microb 66: 16921697. bacteria are motile. Aquat Microb Ecol 24: 197201.
Crump BC, Armbrust EV, Baross JA. (1999). Phylogenetic Fenchel T. (2002). Microbial behavior in a heterogenous
analysis of particle-attached and free-living bacterial world. Science 296: 10681071.
communities in the Columbia river, its estuary, and Frias-Lopez J, Shi Y, Tyson GW, Coleman ML, Schuster SC,
the adjacent coastal ocean. Appl Environ Microb 65: Chisholm SW et al. (2008). Microbial community gene
31923204. expression in ocean surface waters. Proc Natl Acad Sci
Crump BC, Baross JA. (2000). Archaeaplankton in the USA 105: 38053810.
Columbia River, its estuary and the adjacent coastal Fuchs BM, Woebken D, Zubkov MV, Burkill P, Amann R.
ocean, USA. FEMS Microbiol Ecol 31: 231239. (2005). Molecular identification of picoplankton
Dalisay DS, Webb JS, Scheffel A, Svenson C, James S, populations in contrasting waters of the Arabian Sea.
Holmstrom C et al. (2006). A mannose-sensitive Aquat Microb Ecol 39: 145157.
haemagglutinin (MSHA)-like pilus promotes Fuchsman CA, Kirkpatrick JB, Brazelton WJ, Murray JW,
attachment of Pseudoalteromonas tunicata cells Staley JT. (2011). Metabolic strategies of free-living and
to the surface of the green alga Ulva australis. aggregate-associated bacterial communities inferred
Microbiology 152: 28752883. from biologic and chemical profiles in the Black Sea
Dalsgaard T, Thamdrup B, Farias L, Revsbech NP. (2012). suboxic zone. FEMS Microbiol Ecol 78: 586603.
Anammox and denitrification in the oxygen minimum Fuchsman CA, Staley JT, Oakley BB, Kirkpatrick JB,
zone of the eastern South Pacific. Limnol Oceanogr 57: Murray JW. (2012). Free-living and aggregate-
13311346. associated Planctomycetes in the Black Sea. FEMS
Delmont TO, Prestat E, Keegan KP, Faubladier M, Robe P, Microbiol Ecol 80: 402416.
Clark IM et al. (2012). Structure, fluctuation and Fussel J, Lam P, Lavik G, Jensen MM, Holtappels M,
magnitude of a natural grassland soil metagenome. Gunter M et al. (2012). Nitrite oxidation in
ISME J 6: 16771687. the Namibian oxygen minimum zone. ISME J 6:
DeLong EF, Franks DG, Alldredge AL. (1993). Phyloge- 12001209.
netic diversity of aggregate-attached vs free-living Galan A, Molina V, Belmar L, Ulloa O. (2012). Temporal
marine bacterial assemblages. Limnol Oceanogr 38: variability and phylogenetic characterization of plank-
924934. tonic anammox bacteria in the coastal upwelling
DeLong EF, Preston CM, Mincer T, Rich V, Hallam SJ, ecosystem off central Chile. Prog Oceanogr 92-95:
Frigaard NU et al. (2006). Community genomics 110120.

The ISME Journal

 Size-fractionated metagenomics
S Ganesh et al
208
Galand PE, Lovejoy C, Pouliot J, Vincent WF. (2008). metagenomes: revealing an uncultured class of marine
Heterogenous archaeal communities in the particle- Euryarchaeota. Science 335: 587590.
rich environment of an arctic shelf ecosystem. Jensen PR, Lauro FM. (2008). An assessment of actino-
J Marine Syst 74: 774782. bacterial diversity in the marine environment. Anton
Garfield PC, Packard TT, Friedrich GE, Codispoti LA. Van Lee J M S 94: 5162.
(1983). A subsurface particle maximum layer and Jensen MM, Lam P, Revsbech NP, Nagel B, Gaye B, Jetten MS
enhanced microbial activity in the secondary nitrite et al. (2011). Intensive nitrogen loss over the Omani
maximum of the northeastern tropical Pacific Ocean. Shelf due to anammox coupled with dissimilatory
J Mar Res 41: 747768. nitrite reduction to ammonium. ISME J 5: 16601670.
Ghiglione JF, Mevel G, Pujo-Pay M, Mousseau L, Lebaron P, Jiang DM, Kato C, Zhou XW, Wu ZH, Sato T, Li YZ. (2010).
Goutx M. (2007). Diel and seasonal variation Phylogeographic separation of marine and soil myx-
in abundance, activity, and community structure obacteria at high levels of classification. ISME J 4:
of particle-attached and free-living bacteria 15201530.
in NW Mediterranean Sea. Microb Ecol 54: Johnson ZI, Zinser ER, Coe A, McNulty NP, Woodward
217231. EMS, Chisholm SW. (2006). Niche partitioning
Giovannoni SJ, Tripp HJ, Givan S, Podar M, Vergin KL, among Prochlorococcus ecotypes along ocean-scale
Baptista D et al. (2005). Genome streamlining in environmental gradients. Science 311: 17371740.
a cosmopolitan oceanic bacterium. Science 309: Jorgensen SL, Hannisdal B, Lanzen A, Baumberger T,
12421245. Flesland K, Fonseca R et al. (2012). Correlating
Gram L, Melchiorsen J, Bruhn JB. (2010). Antibacterial microbial community profiles with geochemical data
activity of marine culturable bacteria collected from in highly stratified sediments from the Arctic
a global sampling of ocean surface waters and Mid-Ocean Ridge. Proc Natl Acad Sci USA 109:
surface swabs of marine organisms. Mari Biotechnol E2846E2855.
12: 439451. Jumpstart Consortium Human Microbiome Project Data
Grossart HP, Hietanen S, Ploug H. (2003). Microbial Generation Working Group (2012). Evaluation of
dynamics on diatom aggregates in resund, Denmark. 16S rDNA-based community profiling for human
Mar Ecol Prog Ser 249: 6978. microbiome research. PLoS ONE 7: e39315.
Grossart HP, Kirboe T, Tang KW, Allgaier M, Yam EM, Karl DM, Knauer GA, Martin JH, Ward BB. (1984).
Ploug H. (2006). Interactions between marine snow Bacterial chemolithotrophy in the ocean is associated
with sinking particles. Nature 309: 5456.
and heterotrophic bacteria: aggregate formation,
Karl DM, Knauer GA, Martin JH. (1988). Downward flux of
bacterial activities and phylogenetic composition.
particulate organic matter in the ocean: a particle
Aquat Microb Ecol 42: 1926.
decomposition paradox. Nature 332: 438441.
Grossart HP, Schlingloff A, Bernhard M, Simon M,
Karner M, Herndl G. (1992). Extracellular enzymatic
Brinkhoff T. (2004). Antagonistic activity of bacteria
activity and secondary production in free-living
isolated from organic aggregates of the German
and marine-snow-associated bacteria. Mar Biol 113:
Wadden Sea. FEMS Microbiol Ecol 47: 387396. 341347.
Grossart HP, Tang KW, Kirboe T, Ploug H. (2007). Karstensen J, Stramma L, Visbeck M. (2008). Oxygen
Comparison of cell-specific activity between minimum zones in the eastern tropical Atlantic and
free-living and attached bacteria using isolates and Pacific oceans. Prog Oceanogr 77: 331350.
natural assemblages. FEMS Microbiol Lett 266: Kellogg CTE, Deming JW. (2009). Comparison of free-
194200. living, suspended particle, and aggregate-associated
Grossart HP. (2010). Ecological consequences of bacterial and archaeal communities in the Laptev Sea.
bacterioplankton lifestyles: changes in concepts are Aquat Microb Ecol 57: 118.
needed. Environ Microbiol Rep 2: 706714. Kembel SW, Eisen JA, Pollard KS, Green JL. (2011). The
Hardcastle TJ, Kelly KA. (2010). BaySeq: Empirical phylogenetic diversity of metagenomes. PloS ONE 6:
Bayesian methods for identifying differential expres- e23214.
sion in sequence count data. BMC Bioinformatics Kidwell MG, Lisch DR. (2001). Perspective: transposable
11: 422. elements, parasitic DNA, and genome evolution.
Hollibaugh T, Wong PS, Murrell MC. (2000). Similarity of Evolution 55: 124.
particle-associated and free-living bacterial commu- Kirchman DL. (2003). The contribution of monomers and
nities in northern San Francisco Bay, California. other low-molecular weight compounds to the flux
Aquat Microb Ecol 21: 102114. of dissolved organic material in aquatic ecosystems.
Huber JA, Mark Welch DB, Morrison HG, Huse SM, In: Findlay SEG, Sinsabaugh RL (eds). Aquatic
Neal PR, Butterfield DA et al. (2007). Microbial Ecosystems: Interactivity of Dissolved Organic Matter.
population structures in the deep marine biosphere. Academic Press: San Diego, CA, USA, pp 218237.
Science 318: 97100. Kirchman D, Mitchell R. (1982). Contribution of
Hunt DE, David LA, Gevers D, Preheim SP, Alm EJ, particle-bound bacteria to total microheterotrophic
Polz MF. (2008). Resource partitioning and sympatric activity in five ponds and two marshes. Appl Environ
differentiation among closely related bacterioplankton. Microb 43: 200209.
Science 320: 10811085. Konstantinidis KT, DeLong EF. (2008). Genomic patterns
Huson DH, Mitra S, Ruscheweyh HJ, Weber N, of recombination, clonal divergence and environ-
Schuster SC. (2011). Integrative analysis of environ- ment in marine microbial populations. ISME J 2:
mental sequences using MEGAN4. Genome Res 21: 10521065.
15521560. Konstantinidis KT, Braff J, Karl DM, DeLong EF. (2009).
Iverson V, Morris RM, Frazar CD, Berthiaume CT, Morales RL, Comparative metagenomic analysis of a microbial
Armbrust EV. (2012). Untangling genomes from community residing at a depth of 4,000 meters at

The ISME Journal

 Size-fractionated metagenomics
S Ganesh et al
209
station ALOHA in the North Pacific subtropical gyre. McCutcheon JP, Moran NA. (2011). Extreme genome
Appl Environ Microb 75: 53455355. reduction in symbiotic bacteria. Nat Rev Microbiol
Labrenz M, Jost G, Jurgens K. (2007). Distribution of 10: 1326.
abundant prokaryotic organisms in the water column Michotey V, Bonin P. (1997). Evidence for anaerobic
of the central Baltic Sea with an oxic-anoxic interface. bacterial processes in the water column: denitrifica-
Aquat Microb Ecol 46: 177190. tion and dissimilatory nitrate ammonification in the
Lam P, Lavik G, Jensen MM, van de Vossenberg J, Schmid M, northwestern Mediterranean Sea. Mar Ecol Prog Ser
Woebken D et al. (2009). Revising the nitrogen cycle in 160: 4756.
the Peruvian oxygen minimum zone. Proc Natl Acad Sci Mitchell HL, Dashper SG, Catmull DV, Paolini RA, Cleal SM,
USA 106: 47524757. Slakeski N et al. (2010). Treponema denticola
LaMontagne MG, Holden PA. (2003). Comparison biofilm-induced expression of a bacteriophage,
of free-living and particle-associated bacterial toxin-antitoxin systems and transposases. Microbiology
communities in a coastal lagoon. Microb Ecol 46: 156: 774788.
228237. Moeseneder MM, Winter C, Herndl GJ. (2001). Horizontal
Lapoussiere A, Michel C, Starr M, Gosselin M, Poulin M. and vertical complexity of attached and free-living
(2011). Role of free-living and particle-attached bacteria of the eastern Mediterranean Sea, determined
bacteria in the recycling and export of organic by 16S rDNA and 16S rRNA fingerprints. Limnol
materal in the Hudson Bay system. J Marine Syst 88: Oceanogr 46: 95107.
434445. Morris RM, Longnecker K, Giovannoni SJ. (2006). Pirellula
Lauro FM, McDougald D, Thomas T, Williams TJ, Egan S, and OM43 are among the dominant lineages identified
Rice S et al. (2009). The genomic basis of trophic in an Oregon coast diatom bloom. Environ Microbiol 8:
strategy in marine bacteria. Proc Natl Acad Sci USA 13611370.
106: 1552715533. Munn C. (2011). Marine Microbiology: Ecology and
Lavik G, Stuhrmann T, Bruchert V, Van der Plas A, Applications, 2nd edn. Garland Science: New York,
Mohrholz V, Lam P et al. (2009). Detoxification NY.
of sulphidic African shelf waters by blooming Naqvi SWA, Kumar MD, Narvekar PV, De Sousa SN,
chemolithotrophs. Nature 457: 581586. George MD, Silva CD. (1993). An intermediate
Li WZ, Godzik A. (2006). Cd-hit: a fast program for nepheloid layer associated with high microbial rates
clustering and comparing large sets of protein or and denitrification in the northwest Indian Ocean.
J Geophysical Res 98: 1646916479.
nucleotide sequences. Bioinformatics 22: 16581659.
Not F, del Campo J, Balague V, de Vargas C, Massana R.
Long A, Heitman J, Tobias C, Philips R, Song B. (2013).
(2009). New insights into the diversity of marine
Co-occurring anammox, denitrification, and codeni-
picoeukaryotes. PLoS One 4: e7143.
trification in agricultural soils. Appl Environ Microb
Nowack ECM, Melkonian M. (2010). Endosymbiotic
79: 168176.
associations within protists. Philos T Roy Soc B 365:
Long R, Azam F. (2001). Antagonistic interactions among
699712.
marine pelagic bacteria. Appl Environ Microbiol 67:
Ohtsubo Y, Genka H, Komatsu H, Nagata Y, Tsuda M.
49754983. (2005). High-temperature-induced transposition of
Long RA, Rowley DC, Zamora E, Liu JY, Bartlett DH, Azam F.
insertion elements in Burkholderia multivorans ATCC
(2005). Antagonistic interactions among marine bacteria 17616. Appl Environ Microbiol 71: 18221828.
impede the proliferation of Vibrio cholerae. Appl Environ Orsi W, Song YC, Hallam S, Edgcomb V. (2012). Effect
Microb 71: 85318536. of oxygen minimum zone formation on communities
Lozupone C, Knight R. (2005). UniFrac: a new phylogenetic of marine protists. ISME J 6: 15861601.
method for comparing microbial communities. Appl Overbeek R, Begley T, Butler RM, Choudhuri JV, Chuang HY,
Environ Microb 71: 82288235. Cohoon M et al. (2005). The subsystems approach
Lucker S, Nowka B, Rattei T, Spieck E, Daims H. (2013). to genome annotation and its use in the project
The genome of Nitrospina gracilis illuminates the to annotate 1000 genomes. Nucleic Acids Res 33:
metabolism and evolution of the major marine nitrite 56915702.
oxidizer. Front Microbiol 4: 27. Pak H, Codispoti LA, Zaneveld JRV. (1980). On
Lucker S, Wagner M, Maixner F, Pelletier E, Koch H, the intermediate particle maxima associated with
Vacherie B et al. (2010). A Nitrospira metagenome oxygen-poor water off Western South America. Deep
illuminates the physiology and evolution of globally Sea Res 27: 783797.
important nitrite-oxidizing bacteria. Proc Natl Acad Parveen B, Reveilliez JP, Mary I, Ravet V, Bronner G,
Sci USA 107: 1247913484. Mangot JF et al. (2011). Diversity and dynamics of
Madsen JS, Burmlle M, Hansen LH, Srensen SJ. (2012). free-living and particle-associated Betaproteobacteria
The interconnection between biofilm formation and and Actinobacteria in relation to phytoplankton and
horizontal gene transfer. FEMS Immunol Med Mic 65: zooplankton communities. FEMS Microbiol Ecol 77:
183195. 461476.
Mahillon J, Leonard C, Chandler M. (1999). IS elements as Pester M, Bittner N, Deevong P, Wagner M, Loy A. (2010).
constituents of bacterial genomes. Res Microbiol 150: A rare biosphere microorganism contributes to
675687. sulfate reduction in a peatland. ISME J 4: 15911602.
Malmstrom RR, Cottrell MT, Elifantz H, Kirchman DL. Pham VD, Konstantinidis KT, Palden T, DeLong EF.
(2005). Biomass production and assimilation of (2008). Phylogenomic assessment of bacterioplankton
dissolved organic matter by SAR11 bacteria in the distribution in a 4000 metre vertical profile of the
Northwest Atlantic Ocean. Appl Environ Microb 71: North Pacific Subtropical Gyre. Environ Microbiol 10:
29792986. 23132330.

The ISME Journal

 Size-fractionated metagenomics
S Ganesh et al
210
Ploug H, Grossart HP, Azam F, Jorgensen BB. (1999). Sleight SC, Orlic C, Schneider D, Lenski RE. (2008).
Photosynthesis, respiration, and carbon turnover in Genetic basis of evolutionary adaptation by
sinking marine snow from surface waters of Southern Escherichia coli to stressful cycles of freezing, thawing
California Bight: implications for the carbon cycle and growth. Genetics 180: 431443.
in the ocean. Mar Ecol Prog Ser 179: 111. Smith DC, Simon M, Alldredge AL, Azam F. (1992).
Poretsky RS, Sun S, Mou X, Moran MA. (2010). Transpor- Intense hydrolytic enzyme activity on marine
ter genes expressed by coastal bacterio- aggregates and implications for rapid particle
plankton in response to dissolved organic carbon. dissolution. Nature 359: 139142.
Environ Microbiol 12: 616627. Smith MW, Allen LZ, Allen AE, Herfort L, Simon HM.
Prieto-Davo A, Fenical W, Jensen PR. (2008). Comparative (2013). Contrasting genomic properties of free-living
actinomycete diversity in marine sediments. Aquat and particle-attached microbial assemblages within
Microb Ecol 52: 111. a coastal ecosystem. Front Microbiol 4: 120.
Qian PY, Wang Y, Lee OO, Lau SCK, Yang J, Lafi FF et al. Spieck E, Hartwig C, McCormack I, Maixner F, Wagner M,
(2011). Vertical stratification of microbial commu- Lipski A et al. (2006). Selective enrichment and
nities in the Red Sea revealed by 16S rDNA molecular characterization of a previously uncultured
pyrosequencing. ISME J 5: 507518. Nitrospira-like bacterium from activated sludge.
Raghoebarsing AA, Pol A, van de Pas-Schoonen KT, Environ Microbiol 8: 405415.
Smolders AJ, Ettwig KF, Rijpstra WI et al. (2006). Stevens H, Ulloa O. (2008). Bacterial diversity in the
A microbial consortium couples anaerobic methane oxygen minimum zone of the eastern tropical South
oxidation to denitrification. Nature 440: 918921. Pacific. Environ Microbiol 10: 12441259.
Rappe MS, Gordon DA, Vergin K, Giovannoni SJ. (1999). Stewart FJ, Dalsgaard T, Thamdrup B, Revsbech NP,
Phylogeny of Actinobacteria small subunit rRNA (SSU Ulloa O, Canfield DE et al. (2012a). Experimental
rRNA) gene clones recovered from diverse sea water perturbation and oxygen addition elicit profound
samples. Syst Appl Microbiol 22: 106112. changes in community transcription in OMZ
Rath J, Wu KY, Herndl GJ, DeLong EF. (1998). High bacterioplankton. PLoS One 7: e37118.
phylogenetic diversity in a marine-snow-associated Stewart FJ, Ottesen EA, DeLong EF. (2010). Development
bacterial assemblage. Aquat Microb Ecol 14: 261269. and quantitative analyses of a universal rRNA-sub-
Rocap G, Larimer FW, Lamerdin J, Malfatti S, Chain P, traction protocol for microbial metatranscriptomics.
Ahlgren NA et al. (2003). Genome divergence in two ISME J 4: 896907.
Prochlorococcus ecotypes reflects oceanic niche Stewart FJ, Ulloa O, DeLong EF. (2012b). Microbial
differentiation. Nature 424: 10421047.
metatranscriptomics in a permanent marine oxygen
Rosch C, Mergel A, Bothe H. (2002). Biodiversity of
minimum zone. Environ Microbiol 14: 2340.
denitrifying and dinitrogen-fixing bacteria in an acid
Stocker R. (2012). Marine microbes see a sea of gradients.
forest soil. Appl Environ Microb 68: 38183829.
Science 338: 628633.
Rosso AL, Azam F. (1987). Proteolytic activity in coastal
Stocker R, Seymour JR, Samadani A, Hunt DE, Polz MF.
oceanic waters: depth distribution and relationship to
(2008). Rapid chemotactic response enables marine
bacterial populations. Mar Ecol Prog Ser 41: 231240.
Ruehland C, Blazejak A, Lott C, Loy A, Erseus C, Dubilier bacteria to exploit ephemeral microscale nutrient
N. (2008). Multiple bacterial symbionts in two species patches. Proc Natl Acad Sci USA 105: 42094214.
of co-occurring gutless oligochaete worms from Swan BK, Martinez-Garcia M, Preston CM, Sczyrba A,
Mediterranean sea grass sediments. Environ Microbiol Woyke T, Lamy D et al. (2011). Potential for
10: 34043416. chemolithoautotrophy among ubiquitous bacteria
Rusch DB, Halpern AL, Sutton G, Heidelberg KB, lineages in the dark ocean. Science 333: 12961300.
Williamson S, Yooseph S et al. (2007). The sorcerer Takami H, Noguchi H, Takaki Y, Uchiyama I, Toyoda A,
II global ocean sampling expedition: Northwest Nishi S et al. (2012). A deeply branching thermophilic
Atlantic through eastern tropical pacific. PLoS Biol bacterium with an ancient acetyl-CoA pathway
5: e77. dominates a subsurface ecosystem. PLoS One 7:
Sawers G. (1994). The hydrogenases and formate e30559.
dehydrogenases of Escherichia coli. Anton Van Lee Thamdrup B, Dalsgaard T, Jensen MM, Ulloa O, Farias L,
J M S 66: 5788. Escribano R. (2006). Anaerobic ammonium oxidation
Scala DJ, Kerkhof LJ. (1998). Nitrous oxide reductase in the oxygen-deficient waters off northern Chile.
(nosZ) gene-specific PCR primers for detection Limnol Oceanogr 51: 21452156.
of denitrifiers and three nosZ genes from marine Thamdrup B, Dalsgaard T, Revsbech NP. (2012). Wide-
sediments. FEMS Microbiol Lett 162: 6168. spread functional anoxia in the oxygen minimum zone
Shanks AL, Reeder ML. (1993). Reducing microzones and of the eastern South Pacific. Deep Sea Res I 65: 3645.
sulfide production in marine snow. Mar Ecol Prog Ser Traxler MF, Summers SM, Nguyen HT, Zacharia VM,
96: 4347. Hightower GA, Smith JT et al. (2008). The global,
Shimkets LJ, Dworkin M, Reichenbach H. (2006). The ppGpp-mediated stringent response to amino acid
myxobacteria. In: Dworkin M, Falkow S, Rosenberg E, starvation in Escherichia coli. Mol Microbiol 68:
Schleifer KH, Stackebrandt E (eds). The Prokaryotes, 11281148.
3rd edn. Springer: Heidelberg, Germany, pp 31115. Twiss E, Coros AM, Tavakoli NP, Derbyshire KM. (2005).
Simon J. (2002). Enzymology and bioenergetics or respira- Transposition is modulated by a diverse set
tory nitrate ammonification. FEMS Microbiol Rev 26: of host factors in Escherichia coli and is
285309. stimulated by environmental stress. Mol Microbiol
Simon M, Grossart HP, Schweitzer B, Ploug H. (2002). 57: 15931607.
Microbial ecology of organic aggregates in aquatic Ulloa O, Canfield DE, DeLong EF, Letelier RM, Stewart FJ.
ecosystems. Aquat Microb Ecol 28: 175211. (2012). Perspective: microbial oceanography of anoxic

The ISME Journal

 Size-fractionated metagenomics
S Ganesh et al
211
oxygen minimum zones. Proc Natl Acad Sci USA 109: Williamson SJ, Allen LZ, Lorenzi HA, Fadrosh DW,
1599616003. Brami D, Thiagarajan M et al. (2012). Metagenomic
Ulloa O, Pantoja S. (2009). The oxygen minimum zone exploration of viruses throughout the Indian Ocean.
of the eastern South Pacific. Deep ea Res Pt II56: PLoS ONE 7: e42047.
987991. Woebken D, Fuchs BM, Kuypers MM, Amann R. (2007).
van de Vossenberg J, Woebken D, Maalcke WJ, Wessels HJ, Potential interactions of particle-associated anammox
Dutilh BE, Kartal B et al. (2012). The metagenome bacteria with bacterial and archaeal partners in the
of the marine anammox bacterium Candidatus Namibian upwelling system. Appl Environ Microb 73:
Scalindua profunda illustrates the versatility of this
46484657.
globally important nitrogen cycle bacterium. Environ
Woebken D, Lam P, Kuypers MM, Naqvi SW, Kartal B,
Microbiol 15: 12751289.
van der Maarel MJ, Sprenger W, Haanstra R, Forney LJ. Strous M et al. (2008). A microdiversity study
(1999). Detection of methanogenic archaea in seawater of anammox bacteria reveals a novel Candidatus
particles and the digestive tract of a marine fish Scalindua phylotype in marine oxygen minimum
species. FEMS Microbiol Lett 173: 189194. zones. Environ Microbiol 10: 31063119.
Van Mooy BAS, Keil RG, Devol AH. (2002). Impact of Wright JJ, Konwar KM, Hallam SJ. (2012). Microbial
suboxia on sinking particulate organic carbon: ecology of expanding oxygen minimum zones.
enhanced carbon flux and preferential degradation of Nat Rev Microbiol 10: 381394.
amino acids via denitrification. Geochim Cosmochim Wright TD, Vergin KL, Boyd PW, Giovannoni SJ. (1997).
Ac 66: 457465. A novel delta-subdivision proteobacterial lineage from
Venter JC, Remington K, Heidelberg JF, Halpern AL, Rusch the lower ocean surface layer. Appl Environ Microb 63:
D, Eisen JA et al. (2004). Environmental genome 14411448.
shotgun sequencing of the Sargasso Sea. Science 304: Wyman M, Hodgson S, Bird C. (2013). Denitrifying
6674. Alphaproteobacteria from the Arabian Sea that
Wagner M, Horn M. (2006). The Planctomycetes, express the gene (nosZ) encoding nitrous oxide
Verrucomicrobia, Chlamydiae and sister phyla reductase in oxic and sub-oxic waters. Appl Environ
comprise a superphylum with biotechnological and Microbiol 79: 26702681.
medical relevance. Curr Opin Biotechnol 17: 241249. Zaikova E, Walsh DA, Stilwell CP, Mohn WW,
Walsh DA, Zaikova E, Howes CG, Song YC, Wright JJ,
Tortell PD, Hallam SJ. (2010). Microbial
Tringe SG et al. (2009). Metagenome of a versatile
chemolithoautotroph from expanding oceanic dead community dynamics in a seasonally anoxic fjord:
zones. Science 326: 578582. Saanich Inlet, British Columbia. Environ Microbiol 12:
Werren JH. (2011). Selfish genetic elements, genetic 172191.
conflict, and evolutionary innovation. Proc Natl Acad Zehr JP, Kudela RM. (2011). Nitrogen cycle of the open
Sci USA 108: 1086310870. ocean: from genes to ecosystems. Ann Rev Mar Sci 3:
Whitmire AL, Letelier RM, Villagran V, Ulloa O. (2009). 197225.
Autonomous observations of in vivo fluorescence and Zimmer RL, Woollacott RM. (1983). Mycoplasma-like
particle backscattering in an oceanic oxygen minimum organisms: occurrence with the larvae and adults of
zone. Optics Exp 17: 2199222004. a marine bryozoan. Science 220: 208210.

Supplementary Information accompanies this paper on The ISME Journal website (http://www.nature.com/ismej)

The ISME Journal