View Article Online

View Journal

Lab on a Chip
Devices and applications at the micro- and nanoscale

Accepted Manuscript

This article can be cited before page numbers have been issued, to do this please use: H. Nguyen, M.
Massino, C. Keita and J. Salmon, Lab Chip, 2020, DOI: 10.1039/D0LC00279H.
Volume 17
Number 1
7 January 2017
This is an Accepted Manuscript, which has been through the
Pages 1-200
Royal Society of Chemistry peer review process and has been
accepted for publication.
Lab on a Chip
Devices and applications at the micro- and nanoscale Accepted Manuscripts are published online shortly after acceptance,
rsc.li/loc

before technical editing, formatting and proof reading. Using this free
service, authors can make their results available to the community, in
citable form, before we publish the edited article. We will replace this
Accepted Manuscript with the edited and formatted Advance Article as
soon as it is available.

You can find more information about Accepted Manuscripts in the

Information for Authors.

Please note that technical editing may introduce minor changes to the
text and/or graphics, which may alter content. The journal’s standard
ISSN 1473-0197 Terms & Conditions and the Ethical guidelines still apply. In no event
PAPER
Chengzhi Hu, Bradley J. Nelson et al.
Charactization of size-dependent mechanical properties of
shall the Royal Society of Chemistry be held responsible for any errors
tip-growing cells using a lab-on-a-chip device

or omissions in this Accepted Manuscript or any consequences arising

from the use of any information it contains.

rsc.li/loc
 Page 1 of 10 Lab on a Chip
View Article Online
DOI: 10.1039/D0LC00279H

Jour
nalName

Microuidic dialysis using photo-patterned hydrogel membranes in

PDMS chips†
a a a
Hoang-Thanh Nguyen , Morgan Massino , Camille Keita , and Jean-Baptiste Salmon
∗a
Published on 27 May 2020. Downloaded by Uppsala University on 5/28/2020 8:18:52 AM.

Lab on a Chip Accepted Manuscript

We report the fabrication of permeable membranes for microuidic dialysis applications in
poly(dimethylsiloxane) (PDMS) channels. A maskless UV projection device was used to photo-
pattern long hydrogel membranes (mmcm) with a spatial resolution of a few microns in PDMS
chips integrating also micro-valves. We show in particular that multi-layer soft lithography allows
one to deplete oxygen from the PDMS walls using nitrogen gas ow and therefore makes possible
in-situ UV-induced polymerization of hydrogels. We also report a simple surface modication of
the PDMS channels leading to strongly anchored hydrogel membranes that can withstand trans-
membrane pressure drops up to 1 bar without leakages. We then measured the Darcy permeability
of these membranes and estimated their cut-o by measuring the kinetics of diusion of macro-
molecules of dierent sizes through the membrane. Finally, we illustrate the opportunities oered
by such microuidic chips for dialysis applications by observing in real time the crystallization of a
model protein in a chamber of a few nanoliters.

1 Introduction
taining monomers (or oligomers), cross-linkers, and a photo-
In recent years, many studies have reported the possibility of initiator, directly inside a microfluidic channel. Exposure to UV
integrating membranes in microfluidic systems for various ap- light triggers free-radical polymerization leading to the formation
plications such as purification/concentration and separation of of membranes, either dense and permeable, or nanoporous when
samples for (bio-)analytical chemistry 1 , chemical engineering 2–4 , porogens are added to the formulation to induce a microscopic
soft matter 5,6 , desalination 7 , etc. In particular, nanoporous mem- phase separation. This technique leads to the integration of mem-
branes are attracting a great deal of interest due to the many ap- branes with a wide range of nanoporosities in microfluidic chips,
plications in the field of dialysis and ultra-filtration. Many works and as above, mainly for applications dealing with microdialysis
integrate a commercial dialysis membrane by sandwiching it be- steps of biological samples and biomolecules (concentration, pu-
tween two microfluidic chips, with applications as diverse as the rification, buffer exchange, etc.) 22–28 . Several groups recently fo-
study of phase diagrams of colloidal systems 8 , protein crystal- cused on poly(ethylene glycol) diacrylate oligomers (PEGDA) as
lization 9 , or to investigate osmotically-driven flows 10 and foul- these precursors lead to permeable hydrogels that can be photo-
ing in ultra-filtration processes 11 . But it is in the field of analyt- patterned using microscope projection photolithography at a high
ical biochemistry that this type of technique is of particular in- resolution, down to a few microns 29–31 . Squires et al. in par-
terest with numerous applications such as on-chip microdialysis ticular have extensively used such hydrogel membranes to study
sample cleanup, fractionation, concentration, separation, etc, see colloid diffusio-phoresis at the microfluidic scale 32,33 . More re-
e.g. 12–17 and the review 18 . cently, Decock et al. 34 also showed that highly permeable nano-
Beyond this mechanical technique of membrane integration, porous hydrogel membranes can be obtained in the same way
in-situ membrane fabrication by interfacial polymerization 19,20 but using porogens in the formulation 35 , therefore paving the
or by photo-patterning offers many advantages 21 . This tech- way for microfluidic studies of processes ranging from ultra- to
nique makes it possible in particular to integrate a membrane micro-filtration.
perpendicularly in a microfluidic channel and not as a dis- The works cited above have demonstrated the successful in-
tinct layer above or below the channel network. In-situ photo- tegration of hydrogel membranes in microfluidic chips made of
patterning consists in photo-polymerizing specific mixtures con- various materials such as PEGDA, glass, and thiolene-based re-
sists, while poly(dimethylsiloxane) (PDMS) would offer many
other possibilities, such as the integration of micro-valves and
a
CNRS, Solvay, LOF, UMR 5258, Univ. Bordeaux, F-33600 Pessac, France. E-mail: pumps 36 . The main challenge is the permeability of PDMS to
jean-baptiste.salmon-exterieur@solvay.com
oxygen which inhibits free-radical polymerization 37 , thus pre-
† Electronic Supplementary Information (ESI) available: [details of any supplemen-
tary information available should be included here]. See DOI: 00.0000/00000000.
venting the polymerization of hydrogels through the whole thick-

J
our
nal
Name,
[yea
r][
,vol
.
],110 | 1
 Lab on a Chip Page 2 of 10
View Article Online
DOI: 10.1039/D0LC00279H

ness of the channel and their anchorage on the PDMS walls 38 . (a)
Many works even take advantage of this oxygen-permeability to H
e
fabricate free-standing hydrogel microstructures using stop flow h
lithography 39 . Many other works reported the integration of hy- UV
drogels inside PDMS chips, mainly for biological applications, see glass slide silane acrylate N2
e.g. for stretching DNA 29 , biological assays 30 , or studies of stem PDMS PEGDA hydrogel silicone oil
cells in controlled chemical gradients 40 . Nevertheless, the an- (b) (c)
choring of the hydrogel to the channel walls is rarely stated, and
in-situ fabrication of pressure-resistant hydrogel membranes in
PDMS devices has never been reported to our knowledge.
In this work, we report a protocol for the fabrication of mem-
Published on 27 May 2020. Downloaded by Uppsala University on 5/28/2020 8:18:52 AM.

branes in PDMS chips for dialysis applications. More precisely,

Lab on a Chip Accepted Manuscript

we use a maskless UV projection setup to photo-pattern PEGDA-
based hydrogel membranes of any dimensions and at a high spa-
tial resolution, down to a few microns. To overcome the problems
mentioned above, we show that multi-layer soft lithography is a
simple technique to perform photo-polymerization in a oxygen-
free environment. We also describe a simple silanisation protocol
to add vinyl group to the PDMS surface leading to strongly an- Fig. 1 (a) Schematic protocol for making hydrogel membranes in a multi-
chored membranes which can withstand trans-membrane pres- level PDMS chip. H ' 85 µ m, e ' 65 µ m, and h ' 20 µ m. (b) Hydrogel
sure drops up to 1 bar. We also report in-situ measurements of microstructure photo-patterned inside a PDMS channel of height 20 µ m,
the smallest width is 10 µ m, scale bar 50 µ m. (c) Long hydrogel mem-
their Darcy permeability, as well as kinetics of diffusion of solutes
brane (9.1 mm) of width 25 µ m inside a PDMS channel (height 20 µ m).
of different molecular weights through the hydrogel membrane. Scale bars 150 µ m and 2 mm.
These measurements show that these membranes are particularly
suitable for dialysis applications with a molecular weight cut-off
in the range of ' 10–20 kDa. We finally illustrate this last point
then placed again in the oven for ' 60 min to promote silanisa-
with real-time observations of the crystallization of a model pro-
tion and further enhance the PDMS-glass bonding. The channels
tein by the dialysis method at the microfluidic scale.
are finally washed by flowing acetone for several minutes and the
2 Materials and Methods chip is placed in a vacuum bell for several hours to remove any
possible traces of acetone. Note that we have not observed that
2.1 Multi-layer PDMS chips and silanisation protocol this rinsing step with acetone irreversibly changes the shape and
Multi-layer soft lithography techniques were used to make the size of the channels, although acetone is known to slightly swell
PDMS chip shown schematically in Fig. 1a (Sylgard 184, Dow PDMS 41 .
Corning). The bottom layer contains rounded channels (height The integration of the valves as explained in this protocol
h ' 20 µm) obtained from a mold made using a positive pho- (rounded channels, alignment with actuation channels of the sec-
toresist (AZ-40XT, AZ Electronic Materials) which is melted and ond layer) is obviously not a mandatory step in the integration
reflowed by heating after the lithography step 36 . The top layer of a hydrogel membrane, but allows to combine it with Quake
is obtained from a mold made using a negative photoresist (SU- valves for new functionalities, as will be demonstrated later. On
8 3050, MicroChem) and contains rectangular channels (height the other hand, the alignment of the channel used for the nitrogen
H ' 85 µm) to close the Quake valves 36 as well as a channel flow is not a critical step in the protocol, as it can be significantly
for flowing nitrogen gas to carry out the photo-polymerization in wider than the fluidic channels, see for example the masks shown
a oxygen-free environment, see below. The fluidic channels are in the ESI†, Fig. S1, for making the chips used in this work.
separated from the gas and actuation channels by a thin PDMS
layer, e ' 65 µm. The actuation pressure to close the valves is 2.2 Photo-patterning of hydrogel membranes
' 1.6 bar and is imposed using a pressure controller (Fluigent, The aqueous formulation for the polymerization of the hydro-
MFCS-EZ). The rounded shape of the fluidic channels is crucial in gel membranes is made of PEGDA-700 (average Mn , Sigma-
order to obtain a perfect closing of the valves with the pressure Aldrich), 2-hydroxy-2-methylpropiophenone for the photo-
imposed in the actuation channels 36 . initiator (Sigma-Aldrich), hydroquinone for the photo-inhibitor
The multi-layer PDMS chip is sealed by a glass slide using a (Sigma-Aldrich), and water. More precisely, a stock solution of
plasma treatment. This step not only ensures covalent anchor- PEGDA/photo-initiator (90/10 % vol.) containing 23 g/L of photo-
ing of the PDMS stamp to the glass slide, but also activates the inhibitor was mixed with water at a volume fraction of 50%. The
glass and PDMS surfaces for the silanisation step. After plasma, same batch of PEGDA formulation was used in all the experiments
the chip is placed in a oven at 65◦ C for ' 5 min and pure 3- presented in this work.
(trimethoxysilyl)propyl acrylate (Sigma-Aldrich) is then injected For the photo-polymerization of hydrogels of any shape in the
in the fluidic channels to silanize the inner walls. The device is channels, we use the Primo device from Alvéole. This setup uses

2| J
our
nal
Name,
[yea
r][
,vol
.
],110
 Page 3 of 10 Lab on a Chip
View Article Online
DOI: 10.1039/D0LC00279H

a UV laser diode (375 nm) combined with a digital micro-mirror P (a)

device, to project almost-collimated UV patterns (illumination '
tracers 0.02-1 µm
7.7 mW/mm2 ) at the focal plane of a microscope (IX73 Olympus),
with dimensions ' 460×285 µm2 and spatial resolution of ' 1 µm water
(20X objective, N.A. 0.45). membrane
The protocol to photo-pattern membranes in a PDMS channel is lm = 500 µm
shown in Fig. 1a. First, we flow nitrogen gas into the top channel (b) 0.3 (c)
of the chip (pressure drop ' 20 mbar) causing a rapid depletion of
dissolved oxygen in the channel, as well as in the PDMS matrix in P = 0-1 bar
0.2
the vicinity of the fluid channel, due to the thinness of the PDMS

A
layer between the channels and the high diffusivity of oxygen in wm
Published on 27 May 2020. Downloaded by Uppsala University on 5/28/2020 8:18:52 AM.

the PDMS matrix (DO2 ' 3 × 10−9 m2 /s 42 , e2 /DO2 ' 1.4 s). Next, 0.1

Lab on a Chip Accepted Manuscript

the PEGDA formulation is injected to fill the fluidic channels and
UV patterns are projected with a density of energy of 10 mJ/mm2
P =0
0
and exposure time of ' 1.3 s to polymerize the formulation. Then, 1680 1720 1760
the channels are rapidly flushed with water, while the top channel ν (cm−1 )
is filled with silicon oil and closed with stoppers to minimize the
Fig. 2 (a) Experiment to test the tightness of a membrane. (b) Super-
deformations of the fluidic channels with the applied pressure, imposed uorescence and bright eld images showing the accumulation
see Fig. 1a. Finally, the membrane is extensively flushed with of uorescent colloids (diameter 1 µ m) on the permeable membrane, see
pure water using an imposed trans-membrane pressure drop of ' the movie M1 in the ESI† (membrane width wm = 25 µ m). Scale bar
250 mbar for 24 h. Performing the photo-patterning in a oxygen- 250 µ m. (c) IR spectra, absorbance A vs. wavenumber ν . Blue: drop
of pure acrylate silane, violet and red: PDMS surfaces treated with the
free environment is absolutely crucial. Indeed, we observe that liquid silane before and after the acetone rince step. The cyan spectrum
without nitrogen gas, hydrogels do not adhere to the PDMS due corresponding to a PDMS surface silanized in vapour phase is superim-
to the presence of a thin uncrosslinked layer in the vicinity of the posed with the spectrum of the bare PDMS surface (thin black line).
PDMS surface 38 .
This protocol combined with the maskless projection setup al-
the pure liquid silane acrylate leads to strongly anchored mem-
lows us to fabricate membranes with sharp interfaces and at a
branes on the PDMS and glass surfaces.
high resolution, as clearly evidenced in Fig. 1b showing a hy-
The mechanical resistance of a hydrogel membrane depends a
drogel with a complex shape and transverse dimensions down to
priori on its anchoring surface on the glass and PDMS substrates,
10 µm. As shown in Fig. 1c, hydrogel membranes with dimen-
and therefore on its width and length. The tests described above
sions much larger than the size of the pattern projected by the
were carried out with a lm = 500 µm long and wm = 25 µm wide
digital micro-mirror device can also be fabricated by successive
membrane in order to visualize the whole membrane in the same
steps of photo-polymerization over large dimensions, with a pre-
field of view of the microscope to detect any possible leaks with a
cise alignment of the hydrogel microstructures using a motorized
high precision. However, it is likely that much thinner and/or
stage (Märzhäuser). During the photo-patterning, the consecu-
much longer membranes may have different pressure rupture
tive patterns are slightly superimposed (' 10–20 µm) to avoid
thresholds. However, we will show further on that membranes of
any leakages at their junction.
the same thickness, but much longer (lm ' 9 mm), also mechan-
ically resist trans-membrane pressure drops of the order of a few
2.3 Leakage test hundred mbar. Finally, it should also be noted that the lifetime
In order to test the robustness of hydrogel membranes made ac- of a membrane as well as the threshold of mechanical resistance
cording to the previously explained protocol, we carried out the probably also depends on the history of the trans-membrane pres-
experiments illustrated in Fig. 2a, see also the movies M1 and M2 sure drops imposed. From our experience, microfluidic devices in-
in the ESI†. For these experiments, we photo-patterned a mem- corporating long and thin membranes (lm ' 9 mm, wm = 25 µm)
brane (lm = 500 µm long, wm = 25 µm wide) in a PDMS chip to withstood trans-membrane pressures of a few hundred mbar over
separate two channels of height ' 25 µm and width 250 µm. We periods of a few days (see below). However, a quantitative deter-
then injected a highly dilute aqueous dispersion of fluorescently mination of the precise rupture thresholds remains to be carried
labeled colloids in one of the channels (FluoSpheres, Thermo out according to multiple parameters: duration of the imposed
Fisher Scientific) and closed its outlet using a stopper. The pres- pressures, membrane history, dimensions, etc.
sure P imposed at the inlet of the channel then causes a flow
through the membrane, accumulating colloids on the surface of 2.4 FTIR experiments
the membrane. For P ≤ 1 bar (relatively to the ambient pressure), This covalent bonding was further confirmed by ATR-FTIR char-
we did not observe any leakages even with colloids of 20 nm in di- acterizations from several PDMS surfaces treated with different
ameter, see the movies M1 and M2 in the ESI†. At higher pressure, protocols, see Fig. 2c. More precisely, IR spectra were collected
the membrane suddenly detaches irreversibly from the channel with identical acquisition parameters with a BRUKER Vertex 70
walls. These observations show that the surface treatment with spectrometer associated to a Golden Gate ATR accessory. A DTGS

J
our
nal
Name,
[yea
r][
,vol
.
],110 | 3
 Lab on a Chip Page 4 of 10
View Article Online
DOI: 10.1039/D0LC00279H

detector was used and 25 scans were collected for each spectrum
with a spectral resolution of 4 cm−1 . Figure 2c shows in particu-
lar that the surface treatment using the liquid silane acrylate leads
to carbonyl groups (ν ' 1735 cm−1 ) related to acrylate bonds on
the PDMS surface 43 , even after the washing step by acetone. On
the contrary, these measurements also show that vapour deposi-
tion of the same silane (at 65◦ C, after plasma treatment) does not
lead to any detectable peak. We also observed that silanisation
carried out in a vapour phase does not allow efficient anchoring
of the hydrogels as leakages and detachment of the membranes
are observed at low imposed pressure (' 100–200 mbar).
Published on 27 May 2020. Downloaded by Uppsala University on 5/28/2020 8:18:52 AM.

3 Results

Lab on a Chip Accepted Manuscript

(a) P
3.1 Measurements of the Darcy permeability
The Darcy permeability κ of a membrane of thickness wm is de-
fined by the Darcy’s law:
air water
κ δP
vm = , (1)
ηw wm Xm
where δ P is the trans-membrane pressure drop, vm the trans-
membrane flow (m/s), and ηw the water viscosity. To estimate
P
κ for our hydrogel membranes, we performed the experiments
wm = 25 µm
shown schematically in Fig. 3a. The chip contains two rounded
channels (height h ' 20 µm, width 150 µm) separated by a long
and thin membrane (length lm = 9.1 mm, width wm = 25 µm). lm = 9.1 mm
One of the channels is connected to a PTFE tube (Scientific Com- 400
modities Inc., inner radius R = 190 µm, outer radius 550 µm) and (b) (c)
40

vm (nm/s)
its outlet is blocked. The inlet and outlet of the other channel 300 P = 350-0 mbar
δXm (µm)

are left opened. The tube is connected to a pressure controller 30

200
(Fluigent MFCS-EZ) to impose a precise pressure P (relatively to 20
the ambient pressure). The tube is completely filled with wa- 100 10
ter, with the exception of a small air space whose position Xm
is recorded using a stereo-microscope (SZX10, Olympus). The 0 0
imposed pressure P leads to a trans-membrane flux through the 0 2000 4000 6000 0 100 200 300
membrane and thus to a flow rate Q in the tube, displacing the
t (s) δP (mbar)
air/water interfaces. To minimize water pervaporation through Fig. 3 (a) Measurements of the permeability κ of the membrane. Im-
PDMS, possibly leading to flow rates of the same order of mag- posed pressure P drives a ow through the membrane. The chip is
nitude as Q 44 , the chip is fully immersed in a water bath several immersed in a water bath to minimize pervaporation through PDMS.
days prior to the measurements. Scale bar 1 mm. (b) Relative displacement δ Xm of the air/water
meniscus in the tube for various imposed pressure P. (c) Water ux
In a typical experiment, we imposed constant pressure from vm vs. trans-membrane pressure drop δ P. Errorbars are calculated
P = 350 mbar to P = 0, by steps of 50 mbar during about 2 from the standard deviation over 4 experiments. The linear t yields
hours, while measuring simultaneously the position Xm of the κ ' 2.6 ± 0.5 × 10−20 m2 .
air/water interface along the tube. At each change of pressure
level, the air/water interface moves instantaneously by about
200 µm (corresponding to a volume of δV ' 20 nL for pres-
sure steps of δ P ' 50 mbar), due to the elasticity of the system
(air bubble, PDMS channel, tube). As the hydrodynamic resis-
tance Rh of the whole setup (tube + PDMS channel) is of the
order of Rh ' 1014 Pa s/m3 , the elasto-hydrodynamic time scales
τ = Rh δV /δ P are of the order of only a few seconds, and elasticity
plays no role on the meniscus displacement on long time scales.
The temporal relative displacements δ Xm for each imposed pres-
sure step are shown in Fig. 3b. These data are well-fitted by linear
fits leading to flow rates Q ranging from ' 5 to ' 40 nL/h for P
ranging from 0 to 350 mbar. Note that such small flow rates can-

4| J
our
nal
Name,
[yea
r][
,vol
.
],110
 Page 5 of 10 Lab on a Chip
View Article Online
DOI: 10.1039/D0LC00279H

not be measured by any commercial flowmeter, thus justifying PDMS valves (a)
the technique described above. Strikingly, the measured flow rate
sample
does not vanish for P = 0 and is roughly Q0 ' 7 ± 2 nL/h. We have
verified that this offset is not due to the pervaporation of water reservoir
through the PDMS chip, but to the low permeability of the PTFE membrane
tube itself, 30 cm in length. Indeed, we confirmed this result by lm = 9.1 mm dead volumes
performing similar experiments but with only a tube closed by a
stopper. The measured flow rate is also in agreement with esti- (b)
mations made using literature values of the permeability of water
through PTFE (in the range 0.0045–0.3 g.mm/(m2 day) 45 ). In the
following, we therefore subtracted this constant offset in order to
Published on 27 May 2020. Downloaded by Uppsala University on 5/28/2020 8:18:52 AM.

estimate the trans-membrane flow rate Qm = Q − Q0 .

Lab on a Chip Accepted Manuscript

For the small flow rates estimated from Fig. 3b, one can fully sample
neglect the pressure drops along the tube and fluidic channels. w x
The trans-membrane pressure drop in eqn (1) is thus homoge- wm = 25 µm
nous along the membrane and given by δ P ' P. Consequently, membrane
the trans-membrane flux vm is also homogeneous and related to reservoir
the trans-membrane flow rate by Qm = (lm h)vm . Figure 3c shows
vm against δ P calculated from the mean of different experiments
performed with several membranes photo-polymerized in similar Fig. 4 (a) Schematic view of the microuidic chip for dialysis applications,
see Fig. S1 in the ESI for the masks used for the dierent layers†. (b)
conditions (errorbars show the standard deviation of these data). Bright eld image of the whole chip showing the long and thin hydrogel
These data are well-fitted by eqn (1) with κ ' 2.6±0.5×10−20 m2 . membrane inside the PDMS channel as well as the two (closed) valves.
The measured variation (± 0.5 × 10−20 m2 ) is probably the result Dead volumes are highlighted by dotted red lines. Scale bar 2 mm.
of multiple uncertainties such as the precise determination of the Inset: zooming in on the central region to better highlight the spatial
resolution of the membrane (w = 150 µ m). Molecular dyes (brillant blue
dimension of the membrane, especially its width, slightly differ-
FCF, Mw ' 790 g/mol) contained in the closed sample chamber diuse
ent illumination conditions, or the error associated to the deter- through the membrane into the reservoir channel, see the snapshots in
mination of the meniscus velocity. We observed that a higher the ESI, Fig. S2†.
exposure energy during the photo-polymerization results in sig-
nificantly less permeable membranes (data not shown), thus sug-
gesting that our conditions of polymerization lead to partially Quake valves, therefore leading to a closed chamber of volume
cross-linked hydrogels. As explained in Sec. 2.2, we made long Vc ' 38 nL (estimated using profilometry measurements), referred
membranes by successive steps with slightly superimposed pat- below to as the sample chamber. The content of this chamber can
terns to avoid leakages. These superimposed regions have thus be modulated by the diffusion of molecules through the mem-
undergone a double exposure and are a priori less permeable. brane from (or to) the other channel called below the reservoir
However, the overall size of the overlaps (' 300 µm) should have channel.
no significant effect on the permeability of the membrane due to The first experiments we performed are illustrated in Fig. 4. We
its length (lm = 9.1 mm). first prepared two aqueous solutions of fluorescein with the same
The measured permeability κ ' 2.6 ± 0.5 × 10−20 m2 of our hy- concentration (0.015 mM), but with different ethanol contents,
drogel membranes is slightly higher than the macroscopic mea- 0 and 50% vol. We then enclosed the aqueous fluorescein solu-
surement made by Ju et al., κ ' 1.35 × 10−20 m2 , on thick tion in the sample chamber, while the 50% water/ethanol mix-
polymerized film (300 µm) obtained from the same PEGDA for- ture constantly flows in the reservoir channel (imposed pressure
mulation 46 . This difference could probably be explained by the drop of 50 mbar). Ethanol and water rapidly interdiffuse through
cross-linking density of the hydrogel, as Ju et al. stated that their the hydrogel membrane, leading to a change of the fluorescence
membranes were fully polymerized. as the emissivity of the fluorescein molecules depends on the wa-
ter/ethanol content. The kinetics of the mass transfer is estimated
by recording fluorescence images with a 10X objective using an
3.2 Kinetics of solute diffusion through the membrane inverted microscope (Olympus IX83). Typical frame rate is about
To estimate the molecular weight cut-off of our hydrogel mem- 10 images per minute, with exposure time of 80 ms. We checked
branes, we performed kinetic measurements of diffusion of in separate experiments that the overall exposure (a few seconds
molecules of different weights through the membrane of the chip using a shutter) does not lead to significant photo-bleaching with
shown in Fig. 4, see also Fig. S1 in the ESI for the masks used for our illumination conditions. This experiment, as well as those
making the different layers of this chip†. In this chip designed presented below, was not carried out by immersing the chip in
for dialysis applications, a long and thin hydrogel membrane, water, as in the case of permeability measurements. The pervapo-
lm = 9.1 mm and wm = 25 µm, was photo-patterned in a PDMS ration of the water contained in the PDMS chip, and in particular
channel (height h = 20 µm) to separate two identical channels of in the sample chamber, therefore leads to a flow of water from
width w = 150 µm. One of the channels can be closed by two the reservoir channel to compensate for the loss of water mass.

J
our
nal
Name,
[yea
r][
,vol
.
],110 | 5
 Lab on a Chip Page 6 of 10
View Article Online
DOI: 10.1039/D0LC00279H

This mechanism does not change the solute concentration in the rescence intensity at x = w/2 can be again fitted by an exponen-
chamber, and the very low associated flow rates (at most a few tial decrease but with a much longer time scale, τs ' 11 ± 1 min
nL/h 44 ) do not change the measured mass transfer kinetics. (Fig. 5d).
Figure 5a shows the space-time plot of the fluorescence pro- To span a wider range of molecular weights, we performed
files In (x,t) measured along the x axis shown in Fig. 4b at the similar experiments, but with fluorescein isothiocyanate dextrans
center of the channel. These data are normalized between 0 and with molecular weights of Mw = 4, 10, and 20 kDa, referred below
1, using the initially homogeneous concentration profile to ac- to as FD4, FD10, FD20 (Sigma-Aldrich). Typical concentrations
count for the rounded shape of the channel and the fluorescence were 0.25, 0.1, and 0.05 mM for FD4, FD10, and FD20 respectively
intensity of the water/ethanol mixture to set In = 0. The fluores- therefore ensuring dilute conditions. For FD4 and FD10, we also
cence intensity in the chamber decreases on a time scale ' 20– observed a homogenous decrease of the fluorescence intensity in
30 s with noticeable concentration gradients. The decrease of the chamber as for the case of fluorescein, yet with much longer
Published on 27 May 2020. Downloaded by Uppsala University on 5/28/2020 8:18:52 AM.

the fluorescence is due to the interdiffusion of ethanol and water times scales τs ' 110 ± 10 h and τs ' 120 ± 10 days for FD4 and

Lab on a Chip Accepted Manuscript

through the hydrogel membrane leading ultimately to a homo- FD10 respectively, see Fig. S3 in the ESI†. These time scales were
geneous 50% vol. water/ethanol mixture in the chamber. Fig- again estimated by fitting the temporal decrease of the fluores-
ure 5b shows the normalized fluorescence intensity measured in cence in the middle of the chamber by In (w/2,t) = exp(−t/τs ).
the middle of the sample chamber (x = w/2 = 75 µm) against The τs values are nevertheless associated to large uncertainties,
time. These data are crudely fitted by the exponential decrease as measurements were only performed during ' 30 and ' 100 h
In (w/2,t) = exp(−t/τs ) with τs ' 23 ± 2 s. for FD4 and FD10 respectively, leading only to a slight decrease of
the intensity (In (w/2) ' 0.7 for FD4 and In (w/2) ' 0.97 for FD10,
0 0 1
(a) (c) see Fig. S3 in the ESI†). In the case of FD20, we did not measure
20 0.8 any noticeable decrease of the fluorescence even after 3 days in
20
the sample chamber (typical frame rate 10 images per day).
t (min)

40 0.6
t (s)

40 Figure 6 displays the measured time scales τs against the

60 0.4 molecular weight of the corresponding solute, see also Table 1.
60
80 These data yield a crude estimate of the molecular weight cut-
0.2
80 off (MWCO) of our membrane, in the range of 10–20 kDa. Note
100
0 that our estimate differs from a strict measurement of the MWCO
0 50 100 150 0 50 100 150 traditionally done using retention experiments in the context of
x (7m) x (7m)
1 1 membrane science 21 . Nevertheless, our kinetic measurements
(b) 0 (d) 0
unambiguously show that the chip shown in Fig. 4 can be used
0.8 0.8
log In

log In

-2
for any dialysis applications on time scales of several days with
In (w=2; t)

-5
-4
0.6 0.6 macromolecules of molecular weight ≥ 10 kDa. Note that macro-
0.4
-6
0 50 100 0.4
-10
0 50
scopic measurements performed by Ju et al. on thick cross-linked
t (s) t (min) hydrogels obtained using the same PEGDA formulation lead to a
0.2 0.2 significantly smaller MWCO, around 1.2 kDa 46 . As for the Darcy
0 0 permeability, this difference could be due to the difference of
0 100 200 0 50 100 cross-linking density, leading to different mesh sizes of the hy-
t (s) t (min) drogel network. This discrepancy could also be due to the differ-
ence between the methodology used by Ju et al. to measured the
Fig. 5 Space-time plots of the normalized uorescence proles In (x,t)
measured along the x axis shown in Fig. 4 for the diusion through the
cut-off (retention of PEG macromolecules in a dead-end filtration
membrane of ethanol/water (a) and uorescein (c). In (w/2,t) vs. t for experiment) and our kinetic measurements of a purely diffusive
the ethanol/water (b) and uorescein (d) cases. Errorbars are calculated mass transfer.
from the standard deviation of 3 experiments. The insets show the same
data but in a log-lin scale. The black lines are ts by an exponential Table 1 Values of molecular weight Mw , diusion coecient in water
decrease with time scales τs = 23 s in (b) and τs = 11 min in (d). Dw , τs , τ̃s , ℜm /ℜw , and Dm /Dw (assuming Pa = 0.5) for the dierent
solutes studied. For the ethanol/water case, the water content varies
We performed similar experiments with a dilute aqueous fluo- from 0.5 to 1 and Dw ' 412 × 10−10 m2 /s 47 . We chose the average
value Dw = 8 × 10−10 m2 /s for simplicity and we used the average molar
rescein solution (0.03 mM) to probe the mass transfer of a larger mass Mw = 32 g/mol.
solute (fluorescein molar mass Mw = 332 g/mol). In this case,
the solution is first injected in the sample channel and isolated Solutes water/ethanol fluorescein FD4 FD10
by closing the valves. Fluorescein molecules then diffuse through Mw (g/mol) 32 332 4000 10000
Dw × 1010 (m2 /s) 8 47 4.2 48 1.1 49 0.8 49
the membrane to the reservoir channel, where they are constantly τs (s) 23 660 4×105 1×107
depleted by a water flow driven by a pressure drop of 50 mbar. As τ̃s 0.8 12.3 2000 37000
shown by Fig. 5c, concentration profiles In (x,t) (normalized again ℜm /ℜw 0.45 11.6 1350 25000
Dw /Dm 1.3 35 4000 75000
to account for the rounded shape of the chamber) remain homo-
geneous in the chamber in this case and the decrease of the fluo-

6| J
our
nal
Name,
[yea
r][
,vol
.
],110
 Page 7 of 10 Lab on a Chip
View Article Online
DOI: 10.1039/D0LC00279H

10
6 fusive permeation 31 , leading to the following condition for the
108
solute flux at the membrane surface:
104

Dm =Dw
 
∂C Dm
102
Dw = Pa C(x = 0, y,t) . (3)
106 ∂ x x=0,y,t wm
100
On the other boundaries, we simply impose solute no-flux condi-
=s (s)

101 102 103 104 105

Mw (g/mol) tions:
104
Dw ∇C.n = 0 , (4)
with n the unit normal vector. Eqn (2)-(4) can be made dimen-
102 sionless, defining t˜ = tDw /w2 , x̃ = x/w, ỹ = y/w, and c = C/C0 lead-
ing finally to:
Published on 27 May 2020. Downloaded by Uppsala University on 5/28/2020 8:18:52 AM.

101 102 103 104 105

Lab on a Chip Accepted Manuscript

∂c ∂c ∂c
Mw (g/mol) = + , (5)
∂ t˜ ∂ x̃2 ∂ ỹ2

Fig. 6 τs vs. molar mass Mw of the dierent solutes studied, see also
 
∂c ℜw
Table 1. The inset shows Dw /Dm vs. Mw assuming Pa = 0.5, see text. = c(x̃ = 0, ỹ, t˜) , (6)
∂ x̃ x̃=0,ỹ,t˜ ℜm
The gray area indicates the range of molar mass for which solutes do not
cross the membrane on the time scale of the measurements (' 3 days).
∇c.n = 0 (solid boundaries) , (7)

evidencing the crucial role played by the ratio ℜm /ℜw on the

3.3 Mass transfer kinetics: role of the dead volumes overall mass transfer. These equations were solved numerically
As demonstrated above, the chip shown in Fig. 4 can be used for
dialysis applications with macromolecules of molecular weight
(a) 2
≥ 10 kDa. The same concept was previously demonstrated by
Paustian et al. 31 with a similar design integrating also hydrogel
7
membranes in a NOA chip, yet with a smaller membrane length x̃
(900 µm vs. lm = 9.1 mm) and valves located outside the chip.
28
1
Mass transfer is mainly governed in such experiments by two pro- ỹ
cesses: (i) solute diffusion in the sample channel with resistivity
ℜw = w/Dw and (ii) solute diffusion through the membrane with (b)
resistivity ℜm = wm /(Pa Dm ) 31 . In the above relations, Dw , Dm are 4
10
the diffusion coefficients of the solute in the solution and in the
hydrogel respectively and Pa is the partition coefficient defined
τ̃s

by the ratio between the equilibrium concentrations in the hydro-

gel membrane and in the solution, Pa = Cm /C. Nevertheless, the 102
sample chamber shown in Fig. 4 has also dead volumes for obvi-
2D model
ous geometrical constraints (i.e. volumes that are not in direct
eqn (8)
contact with the membrane), which can also influence the overall
100 eqn (8) with no dead volumes
kinetics of mass transfer.
In order to get more insights into the role of the dead volumes 100 102 104
and of the resistivity ℜw and ℜm , we consider theoretically the m /w
following case mimicking the experiments performed above: a
Fig. 7 (a) Computing domain of the dimensionless model for solute dif-
dilute solution at an initial solute concentration C0 enclosed in fusion inside the sample channel. The red line is the hydrogel membrane
the sample chamber and a flow of the pure solvent in the reservoir with the boundary condition eqn (6) and the dead volume is highlighted
channel, large enough to impose C = 0 at the membrane surface by a dotted red line (only half of the chip is shown). (b) τ̃s vs. ℜm /ℜw
in the reservoir. We then performed numerical resolutions of the from the 2D model (red line). The dashed black line corresponds to the
assumption of homogeneous concentration in the chamber, eqn (8). The
solute conservation equation for this experimental case, but with green line corresponds to the hypothesis of homogeneous concentration
the simplified 2D geometry shown in Fig. 7a. For the sake of without dead volumes τ̃s = ℜm /ℜw . The black dots are the experimental
simplicity, we solved the following 2D diffusion equation: data given in Table 1. The vertical dashed lines located at ℜm /ℜw = 10
  and ℜm /ℜw = 600 delineate the dierent regimes.
∂C ∂C ∂C
= Dw + , (2)
∂t ∂ x 2 ∂ y2
(partial differential equation toolbox, Matlab R2017b) on the do-
where C(x, y) is the solute concentration in the chamber, assuming main shown in Fig. 7a using the initial condition c(t = 0) = 1 for
homogeneous concentration over the channel height and omit- various ℜm /ℜw ranging from 0.1 to 105 . The dimensions of the
ting the rounded shape of the channel. After a transient of the computational domain have been chosen to correspond approxi-
order of ∼ w2m /Dm (for Pa ≤ 1), one can assume quasisteady dif- mately to those of the actual chip shown in Fig. 4.

J
our
nal
Name,
[yea
r][
,vol
.
],110 | 7
 Lab on a Chip Page 8 of 10
View Article Online
DOI: 10.1039/D0LC00279H

These numerical results help to evidence different kinetic in Table 1 for the different solutes investigated (see also the black
regimes, see the 2D maps in Fig. S4 in the ESI†. For ℜm /ℜw  dots in Fig. 7b). It could also be tempting to extract from these
600, the concentration decreases homogeneously within the data the ratio Dw /Dm assuming a constant partition coefficient
whole chamber down to c = 0. For 10  ℜm /ℜw  600, the con- (Pa ' 0.5 in our case, the hydrogel water volume fraction), valid
centration profiles decrease almost homogeneously only along for negligible interactions between the solutes and the hydrogel
the membrane, as concentration gradients persist over longer network and negligible size exclusion effects 50 .
time scales in the vicinity of the dead volumes. For smaller The inset of Fig. 6 displays Dw /Dm vs. the molecular weight
ℜm /ℜw , noticeable concentration gradients are evidenced along Mw of the investigated molecules, see also Table 1. For the
x̃, even at the center of the channel (ỹ = 0). These 2D numer- ethanol/water interdiffusion, we found Dw /Dm ∼ 1 demonstrat-
ical results also help us to reveal that the concentration in the ing that the hydrogel hardly affects the diffusivity of these small
middle of the chamber is well-approximated over the range of in- molecules. For larger molecules, Dw /Dm strongly increases with
Published on 27 May 2020. Downloaded by Uppsala University on 5/28/2020 8:18:52 AM.

vestigated ℜm /ℜw by an exponential decay c(x̃ = 1/2,ỹ = 0,t) ' Mw due to hindered diffusion through the hydrogel network 51–53 .
exp(−t˜/τ̃s ), see Fig. S4 in the ESI†.

Lab on a Chip Accepted Manuscript

It may also be tempting to compare these data with different mod-
Figure 7b shows the estimated τ̃s vs. ℜm /ℜw on a log-log scale, els of molecular transport in hydrogels, see e.g. 54,55 , but such a
evidencing the crucial role played by ℜm /ℜw on the kinetics. As- comparison would require careful measurements of the partition
suming a homogeneous concentration over the chamber, integra- coefficient Pa (assumed above constant), as Pa is expected to de-
tion of eqn (5)-(7) over the domain strictly leads to an exponen- crease significantly due to size exclusion effects, in particular for
tial decrease with the time scale: the polymers investigated above 50 .
S̃ ℜm
τ̃s = , (8)
˜lm ℜw 3.4 Application: protein crystallization
To illustrate the opportunities offered by the integration of valves
where l˜m is the dimensionless length of the membrane and S̃
and a membrane inside a PDMS microfluidic chip, we used the
the dimensionless surface of the chamber (S̃/l˜m ' 1.47 in our 2D
device shown in Fig. 4 to perform real-time observations of the
geometry). Figure 7b shows this analytical approximation, as
crystallization of a model protein by the dialysis method. More
well as the approximation assuming negligible dead volumes, i.e.
precisely, we prepared lysozyme solutions (chicken egg white,
τ̃s = ℜm /ℜw . These results help us again to evidence the different
HR7-110, Hampton Research) at 30 mg/ml in a 100 mM sodium
regimes discussed above. In particular, it demonstrates that the
acetate buffer at pH 4.0. We also prepared an aqueous solution
kinetics far from the dead volumes can be approximated by a sim-
of crystallyzing agents, NaCl (7 wt%) also in a 100 mM sodium
ple 1D diffusion problem for ℜm /ℜw  600. For larger ratios of
acetate buffer at pH 4.0.
resistivity, ℜm /ℜw  600, dead volumes also play a role and con-
centration decreases homogeneously within the whole chamber. First, the protein solution was injected in the sample channel
Note that for the actual chip shown in Fig. 4, dead volumes are and the Quake valves were closed. The crystallizing solution was
relatively small compared the volume of the chamber, Vc ' 38 nL, then injected in the reservoir channel with an imposed pressure
because our microfabrication protocol makes it possible to inte- drop of 50 mbar. Real-time observations were performed with
grate micro-valves within the PDMS chip. More precisely, the ra- bright-field microscopy of the sample channel, see Fig. 8a. The
tio S̃/l˜m in eqn (8) can be estimated in the actual geometry by lysozyme molecules (molecular weight Mw = 14.6 kDa) are not
integrating eqn (2)–(4) and assuming homogeneous concentra- expected to diffuse through the membrane over the timescale of
tion leading to the experiments, see Fig. 6. On the other hand, salt molecules
Vc ℜm do diffuse rapidly through the membrane leading to a supersat-
τ̃s = , (9)
hlm w ℜw urated solution in which protein crystals form and grow 56 , see
and thus Fig. 8b and the movie M3 in the ESI†. Crystals that nucleated in
Vc wwm the sample chamber reached their maximal size (' 25 µm) within
τs = . (10)
hlm w Dm Pa about 30 min. This time scale is much longer than the expected
The dimensions of the actual device yield Vc /(hlm w) ' 1.4, show- time scale of the diffusion of salts and the crystallization kinetics
ing that a simple 1D model (i.e. neglecting the dead volumes) is therefore not limited by the mass transfer through the mem-
should roughly approximate the global kinetics, even for a large brane, see the horizontal and vertical lines A→B and B→C in the
resistivity ratio. solubility diagram shown in Fig. 8c. This rapid mass transfer ki-
All these numerical results are in line with our experimental netics (due to the thinness of the membrane and the microfluidic
observations on the diffusion of water/ethanol mixtures and of scale) is an important advantage compared to other conventional
larger solutes (fluorescein and fluorescently labeled dextrans), methods such as vapour diffusion and it offers the possibility of
see in particular the space-time plots of Fig. 5, but also the ex- fine-tuning the trajectories explored in a phase diagram.
tended views, Fig. S2 in the ESI†, showing the transport of a To illustrate the possibility of dynamic investigations, we then
molecular dye of molar mass Mw ' 790 g/mol (brillant blue FCF) flowed a NaCl solution at a lower concentration (2 wt% in a
in the dialysis chip. One can go a step further and use the exper- 100 mM sodium acetate buffer at pH 4.0) to dissolve the crys-
imental values τs to estimate the dimensionless time scale τ̃s and tals, see the movie M3 in the ESI† and the trajectory C→D in the
therefore the ratio ℜm /ℜw using Fig. 7b. These values are listed diagram of Fig. 8c. Before reaching the complete dissolution ex-

8| J
our
nal
Name,
[yea
r][
,vol
.
],110
 Page 9 of 10 Lab on a Chip
View Article Online
DOI: 10.1039/D0LC00279H

pected for this salt concentration 57 , we again imposed a higher one measured here without porogens (κ ' 1–10 × 10−17 m2 vs.
concentration of salt (7 wt%) to make the crystals grow again and κ ' 2.6 ± 0.5 × 10−20 m2 ). We also made such highly perme-
nucleate new ones, trajectory D→E→C. This simple result shows able membranes in our PDMS chips with the protocol explained
the capability to perform complex kinetic explorations within the above and demonstrated that the latter also withstand large trans-
phase diagram of a protein confined in a nanoliter chamber. membrane pressure drops (data not shown). Our methodolo-
gies thus offer the possibilities to integrate membranes in PDMS
(a) chips to investigate processes ranging from dialysis to ultra- and
reservoir: crystallizing agents micro-filtration (with moderate trans-membrane pressure drops
< 1 bar), with the versatility of PDMS devices, such as the inte-
sample: protein solution gration of valves and pumps. We hope in the near future to use
such unique tools to probe mass transfer within complex fluids.
Published on 27 May 2020. Downloaded by Uppsala University on 5/28/2020 8:18:52 AM.

(b) Conicts of interest

Lab on a Chip Accepted Manuscript

There are no conflicts to declare.

Acknowledgements
We would like to thank the exchange program with the University
[P ] solubility (c)
of Lehigh and in particular D. Ou-Yang, for the visit of M. Massimo
A
in our laboratory. We thank J. Jolly for his help for the FTIR ex-
B
E periments, V. Studer and A. Pasturel for scientific exchanges con-
D
cerning the photo-patterning of hydrogels, Y. Hallez, M. Meireles,
C and P. Bacchin for discussions concerning the characterization of
[S] the membranes. The authors also thank ANR OSMOCHIP (ANR-
18-CE06-0021), as well as Solvay and CNRS for funding.
Fig. 8 (a) First moments of the experiment: a protein solution is trapped
in the sample chamber, while salts (crystallizing agents) diuse through
the membrane from the reservoir channel (membrane width 25 µ m, scale References
bar 300 µ m). (b) Protein crystals grow and reach their maximum size in
about 30 min. (c) Sketch of the kinetic exploration shown in the video M3 1 L. Moskvin and T. Nikitina, Journal of Analytical Chemistry,
provided in the ESI† within the solubility diagram, protein concentration 2004, 59, 2.
[P] vs. salt concentration [S]. 2 I. Ziemecka, B. Haut and B. Scheid, Lab Chip, 2015, 15, 504.
3 Y. Zhang, N. E. Benes and R. G. H. Lammertink, Chemical
Engineering Journal, 2016, 284, 1342.
4 Conclusions
4 Y. Kaufman, R. Kasher, R. G. Lammertink and V. Freger, J.
In this work, we presented a method to fabricate thin hydrogel Membrane Sci., 2012, 396, 67.
membranes with possibly large dimensions (mm-cm) in PDMS 5 J. Scrimgeour, J. K. Cho, V. Breedveld and J. Curtis, Soft Mat-
channels. The fabrication protocol is based on standard tech- ter, 2011, 7, 4762.
niques (multi-layer soft lithography, silanisation, microscope pro-
6 L. Daubersies, J. Leng and J.-B. Salmon, Lab Chip, 2013, 13,
jection photolithography) and leads to membranes which easily
910.
withstand trans-membrane pressure drops up to 1 bar. In ad-
7 B. Gumuscu, A. S. Haase, A. M. Benneker, M. A. Hempenius,
dition, we also performed a thorough investigation of the mass
A. van den Berg, R. G. Lammertink and J. C. Eijkel, Adv. Funct.
transfer through such membranes (solvent permeability and ki-
Mater., 2016, 26, 8547.
netics of diffusion of solutes of different weights), showing that
8 M. Kornreich, M. Heymann, S. Fraden and R. Beck, Lab Chip,
these membranes are particularly suitable for any dialysis appli-
2014, 14, 3700.
cation with a molecular cut-off in the 10–20 kDa range. We also
9 N. Junius, S. Jaho, Y. Sallaz-Damaz, F. Borel, J.-B. Salmon
reported a microfluidic chip combining Quake valves and a long
and M. Budayova-Spano, Lab Chip, 2020, 20, 296.
hydrogel membrane to perform dialysis in a nanoliter enclosed
chamber and illustrated its functioning using dynamic crystalliza- 10 K. H. Jensen, J. Lee, T. Bohr and H. Bruus, Lab Chip, 2009, 9,
tion experiments of a model protein. The possibility to integrate 2093.
directly the valves within the channel, in the vicinity of the mem- 11 H. Di, G. J. Martin and D. E. Dunstan, J. Membr. Sci., 2017,
brane, is also a strong asset regarding the ratio membrane sur- 532, 68–75.
face/chamber volume (negligible dead volumes). 12 N. Xu, Y. Lin, S. A. Hofstadler, D. Matson, C. J. Call and R. D.
In a related work 34 , we recently reported the fabrication of Smith, Anal. Chem., 1998, 70, 3553.
pressure-resistant nano-porous hydrogel membranes using a sim- 13 F. Xiang, Y. Lin, J. Wen, D. W. Matson and R. D. Smith, Anal.
ilar method but in PEGDA chips and with porogens added in Chem., 1999, 71, 1485.
the hydrogel formulation (PEG) 35 . The Darcy permeability of 14 R. C. Kolfschoten, A. E. Janssen and R. M. Boom, Journal of
such membranes is several orders of magnitude higher than the Separation Science, 2011, 34, 1338.

J
our
nal
Name,
[yea
r][
,vol
.
],110 | 9
 Lab on a Chip Page 10 of 10
View Article Online
DOI: 10.1039/D0LC00279H

15 M. Skou, S. Skou, T. G. Jensen, B. Vestergaard and R. E. 37 S. C. Ligon, B. Husar, H. Wutzel, R. Holman and R. Liska,
Gillilan, J. Appl. Cryst., 2014, 47, 1355. Chem. Rev., 2013, 114, 557.
16 N. To, I. Sanada, H. Ito, G. S. Prihandana, S. Morita, Y. Kanno 38 D. Dendukuri, P. Panda, R. Haghgooie, J. M. Kim, T. A. Hatton
and N. Miki, Front. Bioeng. Biotechnol., 2015, 3, 70. and P. S. Doyle, Macromolecules, 2008, 41, 8547.
17 F. Li, R. M. Guijt and M. C. Breadmore, Anal. Chem., 2016, 39 D. Dendukuri, D. C. Pregibon, J. Collins, T. A. Hatton and P. S.
88, 8257. Doyle, Nat. Mater., 2006, 5, 365.
18 P.-C. Wang, D. L. DeVoe and C. S. Lee, Electrophoresis, 2001, 40 A. Manfrin, Y. Tabata, E. R. Paquet, A. R. Vuaridel, F. R. Rivest,
22, 3857. F. Naef and M. P. Lutolf, Nat. Methods, 2019, 16, 640.
19 B. Zhao, N. O. Viernes, J. S. Moore and D. J. Beebe, 2002, 41 J. N. Lee, C. Park and G. M. Whitesides, Anal. Chem., 2003,
124, 5284–5285. 75, 6544.
20 H. Hisamoto, Y. Shimizu, K. Uchiyama, M. Tokeshi, Y. Kiku- 42 D. A. Markov, E. M. Lillie, S. P. Garbett and L. J. McCawley,
Published on 27 May 2020. Downloaded by Uppsala University on 5/28/2020 8:18:52 AM.

tani, A. Hibara and T. Kitamori, Anal. Chem., 2003, 75, 350. Biomedical microdevices, 2014, 16, 91.

Lab on a Chip Accepted Manuscript

21 J. De Jong, R. G. Lammertink and M. Wessling, Lab Chip, 43 G. Socrates, Infrared and Raman characteristic group frequen-
2006, 6, 1125. cies: tables and charts, John Wiley & Sons, 2004.
22 S. Song, A. K. Singh, T. J. Shepodd and B. J. Kirby, Anal. 44 G. C. Randall and P. S. Doyle, Proc. Natl. Acad. Sci. USA, 2005,
Chem., 2004, 76, 2367. 102, 10813.
23 S. Song, A. K. Singh and B. J. Kirby, Anal. Chem., 2004, 76, 45 L. W. McKeen, Permeability properties of plastics and elas-
4589. tomers, William Andrew, 2016.
24 M. F. Bedair and R. D. Oleschuk, Anal. Chem., 2006, 78, 1130. 46 H. Ju, B. D. McCloskey, A. C. Sagle, V. A. Kusuma and B. D.
25 H. Chun, Electrophoresis, 2018, 39, 2181. Freeman, J. Membrane Sci., 2009, 330, 180.
26 H. Kim, J. Kim, E.-G. Kim, A. J. Heinz, S. Kwon and H. Chun, 47 A. Klinov and I. Anashkin, Processes, 2019, 7, 947.
Biomicrofluidics, 2010, 4, 043014. 48 C. T. Culbertson, S. C. Jacobson and J. M. Ramsey, Talanta,
27 R. T. Davies, J. Kim, S. C. Jang, E.-J. Choi, Y. S. Gho and 2002, 56, 365.
J. Park, Lab Chip, 2013, 12, 5202. 49 M. Arrio-Dupont, S. Cribier, G. Foucault, P. F. Devaux and
28 J. Moorthy and D. J. Beebe, Lab Chip, 2003, 3, 62. A. d’Albis, Biophys. J., 1996, 70, 2327.
29 G. C. Randall, K. M. Schultz and P. S. Doyle, Lab Chip, 2006, 50 C. Kotsmar, T. Sells, N. Taylor, D. E. Liu, J. Prausnitz and
6, 516. C. Radke, Macromolecules, 2012, 45, 9177.
30 J. Liu, D. Gao, H. F. Li and J. M. Lin, Lab Chip, 2009, 9, 1301. 51 G. M. Cruise, D. S. Scharp and J. A. Hubbell, Biomaterials,
31 J. S. Paustian, R. N. Azevedo, S.-T. B. Lundin, M. J. Gilkey and 1998, 19, 1287.
T. M. Squires, Phys. Rev. X, 2013, 3, 041010. 52 A. Pluen, P. A. Netti, R. K. Jain and D. A. Berk, Biophys. J,
32 N. Shi, R. Nery-Azevedo, A. I. Abdel-Fattah and T. M. Squires, 1999, 77, 542.
Phys. Rev. Lett., 2016, 117, 258001. 53 E. W. Merrill, K. A. Dennison and C. Sung, Biomaterials, 1993,
33 R. Nery-Azevedo, A. Banerjee and T. M. Squires, Langmuir, 14, 1117.
2017, 33, 9694. 54 A. G. Ogston, B. Preston and J. Wells, Proceedings of the
34 J. Decock, M. Schlenk and J.-B. Salmon, Lab Chip, 2018, 18, Royal Society of London. A. Mathematical and Physical Sciences,
1075. 1973, 333, 297.
35 A. G. Lee, C. P. Arena, D. J. Beebe and S. P. Palecek, Biomacro- 55 H. Matsuyama, M. Teramoto and H. Urano, J. Membrane Sci.,
molecules, 2010, 11, 3316. 1997, 126, 151–160.
36 M. A. Unger, H. P. Chou, T. Thorsen, A. Scherer and S. R. 56 A. McPherson, Methods, 2004, 34, 254.
Quake, Science, 2000, 288, 113. 57 S. B. Howard, P. J. Twigg, J. K. Baird and E. J. Meehan, J.
Crystal Growth, 1988, 90, 94.

10 | J
our
nal
Name,
[yea
r][
,vol
.
],110