Microﬂuidic Microdialysis: Spatiotemporal Control over Solution Microenvironments Using Integrated Hydrogel Membrane Microwindows

We present a powerful and versatile technique that enables exquisite spatial and temporal control over local solution chemistry in microﬂuidic devices. Using a microscope and a UV lamp, we use projection lithography to photopolymerize thin ( 10 – 25 (cid:1) m ) hydrogel membrane ‘‘microwindows’’ (HMMs) into standard microﬂuidic devices. These microwindows are permeable to solute and solvent diffusion and to electric ﬁelds, yet act as rigid walls from the standpoint of ﬂuid ﬂow. Reservoirs of solution may thus be rapidly imposed, switched, and maintained on one side of a HMM using standard microﬂuidic techniques, provoking changes in solution conditions on the other side without active mixing, stirring, or diluting. We highlight three paradigmatic experimental capabilities enabled by HMMs: (1) rapid dialysis and swapping of solute and/or solvent, (2) stable and convection-free localized concentration gradients, and (3) local electric permeability. The functional versatility of hydrogel microwindow membranes, coupled with the ease and speed of their fabrication and integration into simple microchannels or multilayer devices, will open a variety of novel applications and studies in a broad range of ﬁelds.


I. INTRODUCTION
Provoking the response of cells, organisms, solutions, and materials to environmental changes is a nearly ubiquitous demand in biology, chemistry, physics, and materials science. Microfluidic devices enable such studies to be performed on micron scales [1][2][3][4][5], with precise geometric and fluidic control over small sample volumes [6,7]. Techniques for microfluidic logic [8] have enabled large-scale integration of microfluidic devices capable of complex experiments and high-throughput parameter sweeps [9]. Robust and versatile methods for sculpting local chemical environments in space and time, however, have remained elusive. The vertebrate circulatory system provides an ideal model for the rapid and precise chemical delivery to soft targets, by using a stepwise method of convection followed by diffusion. Gas, nutrients, and other solutes are convected rapidly over large distances through arteries and vessels until they reach smaller capillaries, which are distributed densely enough that solute can diffuse rapidly across membranes or pores and into tissues and cells.
Here, we describe a powerful, versatile, yet simple element for microfluidic systems-hydrogel microwindow membranes (HMMs)-that enable an analogous stepwise convective-diffusive chemical delivery in microfluidic systems. HMMs provide an unparalleled combination of rapid delivery, spatial control, optical accessibility, simple and rapid device fabrication, and integration with standard devices. HMMs are compatible with existing microfluidic materials and fabrication techniques, and they enable the rapid imposition and switching of local microchemical environments, without introducing convective flows near the sample that might disrupt or wash away fragile or unbound materials. Moreover, HMMs can operate scalably within complex microfluidic devices, as each HMM functions independently, without interfering with others in the network.
The core HMM strategy is illustrated in Figs. 1 and 2. A ''reservoir channel'' (which will contain the solution to be delivered) is made to run alongside a ''sample'' channel (to which the solution will be delivered), with a gap in the wall separating them. A thin (10-25-m-wide) hydrogel membrane microwindow placed in this gap then allows local microdialysis. Its pores are small enough that it acts like an impermeable wall from the standpoint of fluid flow, yet is permeable to electric fields, solute, and solvent. A solution flowing through the reservoir channel establishes and maintains a steady reservoir of solute, which subsequently diffuses through the membrane and into the sample channel. The thinness of the membrane, and the small dimensions characteristic of microfluidic devices, renders a short time scale for diffusive delivery ( D $ L 2 =D). Dissolved salts (D $ 10 3 m 2 =s), for example, require D $ 100 ms to diffuse 10 m, and D $ 10 s to diffuse 100 m.
Several essential features are readily apparent. (1) HMMs are exceptionally thin, which enables fast diffusive delivery. (2) Microdevices can be designed with multiple microwindow membranes, each with a distinct, independently controlled reservoir channel. Almost arbitrarily complex solution microenvironments can thus be designed into microfluidic systems, e.g., with multiple components, introduced at multiple locations and/or times, as well as sharp concentration gradients (e.g., several molar over tens of m). (3) HMMs are permeable to electric fields, enabling localized electrokinetic effects (e.g., selective electroporation, electrophoresis, sorting, and separations) to be incorporated into complex microfluidic devices. (4) Integration with microfluidic plumbing and logic (including elastomeric valves in polydimethylsiloxane (PDMS) devices [10]) enables rapid switching of reservoir solutions and, consequently, of the local solution chemistry within the sample chamber. (5) HMMs hydrodynamically isolate reservoir and sample channels in complex-device geometries, without requiring pressures to balance as in laminar-flow strategies. Each HMM can thus operate independently, without interfering with upstream or downstream HMMs. (6) The fabrication strategy is relatively simple, requiring little more equipment than is standard in most research laboratories. (7) HMMs do not interfere with optical access to the sample, whereas, e.g., track-etched membranes used to introduce solute to a channel from above or below [11] interfere with the optical path. Instead, sample and reservoir channels can lie entirely within a single focal plane, allowing high-resolution gradient visualization within microscale regions and improved synchronization with reservoir channel flow switching. Hydrogel microwindow membranes can be integrated into conventional microfluidic channel materials, and hydrogel chemistries can be readily adapted as needed for particular chemical demands, e.g., to be permeable to a range of solutes and solvents. We therefore anticipate that HMMs will enable a broad range of capabilities for microfluidic studies and applications that demand precise spatial and temporal control over solute and solvent distribution.
We begin with a brief discussion of existing techniques for microfluidic solution manipulation in Sec. II, then continue with a brief description of a HMM fabrication procedure in Sec. III. Section IV presents the basic operation and properties of HMMs, with Sec. IVA showing HMM functions for the flow-free introduction of solute (pH buffers, evidenced with a fluorescent pH indicator; Fig. 2) and solvent (triggering local antisolvent crystallization; Fig. 3). Section IV B details methods to measure Here, we use the three-channel device shown in Fig. 1(b) and specifically show the same close-up view as Fig. 1(c). All channels contain a fluorescent pH indicator at the same concentration, and each channel is buffered at a different pH: 8.1 (left, bright), 6.3 (center, intermediate), and 4.7 (right, dark), with ionic strengths balanced. Fluorescent tracer particles (bright lines) travel in straight lines past the center channel, indicating no convective inflow through either the left or right HMM. Diffusion of pH buffers through the membrane is clearly evident. Insets: Contrast-enhanced and enlarged images showing tracer particle paths. Scale bar: 20 m. The microfluidic device is placed on the microscope stage, and a photomask is inserted into the microscope field aperture. A computercontrolled mechanical shutter exposes the PEG-DA solution to UV irradiation for the desired exposure time. Inset: Natural fluorescence of NOA-81 optical glue enables the patterned UV light to be aligned. and control the diffusive and convective permeabilities of the microwindows. We then present paradigmatic experiments that highlight the versatility and range of capabilities afforded by HMMs. Section V shows rapid dialysis of solute and solvent, Sec. VI shows local electric permeability and field sculpting, and Sec. VII shows rapid imposition of localized concentration gradients, which enables the first direct visualization of colloidal solvophoresis [ Fig. 7(c)]. Along with these demonstrations, we describe several novel methods yielded by HMMs to manipulate and visualize colloids and solutions at the microscale.

II. PREVIOUS WORK
Steady chemical gradients have been created convectively, by using T junctions [12] to bring two solutions into coflowing contact. These can be used to expose surfacefixed samples to chemical gradients or dynamic environments [13,14]. However, resulting shear may disrupt or perturb soft samples (e.g., cells or tissues, soft materials, or growing crystals), and flow itself may displace unbound samples (including material precursors, existing solutions, particles, or cells) or disrupt gradients. Although shear may be reduced using channel networks for gradient generation [15], all flow-based strategies require precise pressure balances to properly align flows [16], as well as relatively large device areas [17]. Therefore, they become increasingly impractical as devices become more complex. Coflowing techniques are well suited for certain experiments but do not represent a versatile, robust technique to rapidly introduce, controllably vary, or precisely sculpt local chemical microenvironments. Convective solution modification has also been achieved by adding ingredients sequentially, via a rotary mixer [18], enabling automated batch preparation of solutions [19], yet each added aliquot necessarily flushes out an equal volume of the existing sample. In some cases, this flushing is advantageouse.g., in maintaining a stable microbial population in a microchemostat [20]-but continuous dilution is undesirable in other experiments.
Introducing solute or solvent diffusively can alleviate some of the issues associated with flow, but it poses separate challenges. Diffusive transport becomes prohibitively slow over long distances: The time D for a species with diffusivity D to diffuse a distance L scales like D $ L 2 =D, and any flow U-stray or otherwise-overwhelms this diffusive delivery beyond a critical length scale L c $ D=U. Batch diffusive mixing has been enabled through free-interface diffusion, initiated by opening PDMS valves between microchambers [21][22][23][24] or sliding lubricated substrates to bring loaded microchambers on opposite sides into contact [25][26][27]. Such strategies, however, cannot maintain steady gradients. Steady gradients have been produced by controllably contacting flowing source and sink solutions with large (L $ 1 mm, D * 1000 s) flow-free channels [28] or chambers [29], but reductions in chamber size (and D ) would result in increased stray convection.

III. MICROWINDOW FABRICATION
To make HMMs, we employ microscope projecton photolithography (MPP) [50], in which a standard microscope provides high-resolution photopatterning with UV illumination. In MPP, a photomask is placed within the field conjugate plane, usually in the microscope's field diaphragm, yielding patterned UV illumination in the focal plane. The photomask pattern is also magnified by the objective, resulting in enhanced resolution. The technique has recently become common for flow lithography, wherein hydrogel particles are synthesized within flowing solutions of UV-curable molecules through PDMS channels [51][52][53][54][55][56]. The polymerization reaction is inhibited by oxygen that permeates through the PDMS walls, resulting in freely flowing hydrogel particles [57]. MPP has also been used to pattern 150-m-wide gel barriers [58] and hydrogel posts [59] on glass surfaces in glass/PDMS channels.
Using MPP, we pattern the photopolymerization of poly (ethylene-glycol)-diacrylate (PEG-DA) precursor solutions to form hydrogel membranes that are much thinner than would be possible using standard methods such as contact lithography. (See Fig. 1 and Appendix A 1 for experimental details.) Owing to its versatile chemistry, PEG-DA has been widely used in continuous-flow lithography (e.g., [52]) and in gel-based microfluidic devices (e.g., [47]). We focus on the fabrication, operation, and capabilities of HMMs polymerized into devices made from microfluidic ''stickers'' (Norland Optical Adhesives, NOA-81, thiol-ene-based photopolymer) [60], which we found to be straightforward because of the fast templating, oxygen impermeability, and reactivity [61] of these microchannels. We have also developed techniques for fabricating HMMs in PDMS channels through surface modifications, which will be described in a later work. Either device type can be integrated with PDMS microvalves, if desired, to achieve precise flow control and rapid switching (e.g., Fig. 5).

A. Solution sculpting
For HMMs to function appropriately, they must be permeable to the diffusive transport of solute and solvent, yet concurrently act as rigid, impermeable walls from the standpoint of fluid flow. These twin demands require a hydrogel that is chemically inert, with pores that are large enough to admit molecular diffusion, yet small enough to prevent appreciable flow. Figure 2 and related Video 1 demonstrate the successful operation of HMMs. Each of the three channels contains 25 M fluorescein, which functions as a pH indicator because of its increasing fluorescence intensity between pH 5 and 9. Solutions buffered at different pH are driven through the three channels: The left (reservoir) channel is maintained at pH 8.1 by a 10-mM tris buffer and appears bright; the right (reservoir) is held at pH 4.7 by a 10-mM acetate buffer and appears dark; and the center (sample) channel is held at pH 6.3 by a 1-mM 2-(N-morpholino)ethanesulfonic acid (MES) buffer along with 9-mM KCl to balance ionic strength, and it appears grey. Fluorescent polystyrene beads are also present in the center solution and serve as flow tracers (Bangs Labs FS03F, 500nm, 0.02wt%). All channel outlets were connected to a waste reservoir, the two reservoir channel inlets were pressurized at 150 mbar using compressed air, and the sample channel inlet was held at 2 mbar with hydrostatic pressure, giving a transmembrane pressure ÁP m ¼ 75 mbar and flow velocities that are 75 times faster in reservoir channels than in the sample channel. Buffers in the reservoir channels visibly diffuse into the sample channel, as indicated by the change in fluorescence intensity, and are convected downstream by the sample channel flow. Tracer particles in the sample channel follow straight trajectories past the HMMs, indicating the absence of detectable hydrodynamic flow through the membrane, despite the trans-HMM pressure difference. Transport of buffer through the HMM from the reservoir to the channel is thus predominantly diffusive.
HMMs are also permeable to solvents that are miscible with water, allowing solvent, as well as solute, to be introduced from reservoirs. Ethanol, for example, reduces the solubility of KCl in water and can be delivered through HMMs to trigger antisolvent crystallization.

B. Permeability properties
To examine HMMs with different polymer network densities (Fig. 4), we photopolymerized HMMs from different precursor solutions within a single, three-channel device: the left HMM using a 50% v/v PEG-DA solution and the right HMM from the standard 20% v/v solution. , the amount of convection admitted by HMMs at large transmembrane pressure drops, here ÁP m ¼ 1:5 bar, is lower for a highly cross-linked HMM (50% v/v, left) than for a lightly crosslinked HMM (20%, right). Tracer particles are focused towards the center of the channel by transmembrane flow from the side channels. (b) The diffusive flux through a lightly cross-linked membrane (20% v/v, right) is nearly identical to that through a highly cross-linked (50% v/v, left) membrane. (c) After diffusing through each HMM, the pH buffer is convected downstream, forming a boundary layer whose thickness decreases with Peclet number (flow velocity) like Pe 1=2 , as expected for 2D convection-diffusion. Scale bars: 20 m.
We first look at permeability to trans-HMM flow, which is undetectable (Fig. 2) under typical pressures (e.g., ÁP m ¼ 75 mbar). Twentyfold higher trans-HMM pressures (ÁP m ¼ 1:5 bar), however, drive trans-HMM flows strong enough to displace tracer particles by some distance f from each wall [ Fig. 4(a)]. A mass balance relates the fluid velocity u L;R m through the left and right HMMs to the downstream particle-free widths L;M f , and the maximum downstream tracer velocity u bot , according to under the assumption that the downstream flow profile has developed to the standard Poiseuille flow profile [see the derivation in the Supplemental Material (SM) [62]; see also [63]]. Trans-HMM flows obey Darcy's law [64], where is the fluid viscosity and w m is the membrane thickness. The Darcy permeability can be used to estimate the typical pore size l p $ ffiffiffi ffi p [65], which is, in turn, set by the cross-link density and entanglement spacing (typically 3-8 nm for many polymers [66]). We found L;R for the left and right HMMs to be 4.4 and 11 nm 2 , giving l L;R p $ 2 and 3 nm for the 50% and 20% HMMs, respectively. We note that the pore size would change for different exposure times; the exposure times were selected as the minimum required for the solution to gel, which does not occur immediately, possibly because of an oxygen depletion time [57]. Larger exposure times result in wider, more diffuse gel membranes and smaller pores.
Despite the lower Darcy permeability of the 50% HMM, its diffusive flux is indistinguishable from that of the 20% HMM. In Fig. 4(b), both reservoir channels are buffered at pH 9.8 with 10-mM Na 2 CO 3 and pressurized to 150 mbar, as done previously, whereas the sample channel is held at pH 6.3 by a 1-mM MES buffer. We set the flow velocity in the sample channel by controlling the hydrostatic pressure [67]. All solutions contain 25-M fluorescein as a pH indicator, revealing a bright boundary layer alongside each HMM [ Fig. 4(b)], with width defined when the intensity drops to 20% of its maximum value. Boundary layer widths for 20% and 50% HMMs are nearly identical at all channel velocities measured [ Fig. 4(c)]. This indicates either identical effective diffusivity or solution-dominant mass transfer resistance for Na 2 CO 3 in PEG-DA, since w m is nearly identical for the two HMMs. Larger solutes, however, have been shown to have decreased diffusivity in high-concentration PEG-DA gels [47].
The shape of the boundary layer reflects the competing effects of diffusion and convection and thus depends upon the Peclet number Pe ¼ Uw s =D, where U is the average velocity (calculated from the pressure head and hydraulic resistance [63]), w s is the sample channel width, and D is the solute diffusivity (e.g., D¼10 3 m 2 =s for Na 2 CO 3 [68]).
Depth-averaged ''Hele-Shaw'' flows in low-aspect ratio channels (here, 10 m tall by 150 m wide) have approximately uniform profiles. Solute diffusing perpendicular to such uniform flows at high Pe establishes boundary layers with thickness =w s $ Pe À1=2 , as observed here [ Fig. 4(c)]. The detailed, three-dimensional concentration profile, however, will reflect the inhomogeneous (parabolic) flow across the channel [69].

V. RAPID DIALYSIS
We now describe HMM-rapid dialysis (HMM-RaD) experiments, in which HMMs are used to rapidly change a chemical microenvironment. We expect that HMM-RaD will enable a broad range of new studies in a variety of disciplines, with examples including provoking the responses of cells, embryos, organisms, and microbes to morphogens, chemokines, signaling proteins, drugs, or toxins [31]; modifying colloids and surfaces (e.g., for sample sorting, binding assays, or fundamental surface studies) [70]; and provoking the transient response of complex and soft materials [71].
We start with reversible HMM-RaD in a 150-m-wide sample channel [ Fig. 5 details appear in Appendix A 2. NaCl was chosen as solute because of its relatively high diffusivity (D ¼ 1:6 Â 10 À5 cm 2 =s [68]) among salts with an available indicator dye (CoroNa Green, Invitrogen). Solutions were prepared consisting of 40-M CoroNa Green, a 10-mM 3-(N-morpholino)propanesulfonic acid buffer (MOPS, pH 7), and 0-mM NaCl (dark) or 250-mM NaCl (bright). The fluorescence intensity of CoroNa was found to increase linearly with NaCl concentration in this range. Three 300-m HMMs were polymerized to form a 900-m-long HMM. The reservoir solutions were periodically switched every 40 seconds, and the relative fluorescence intensity ðIðtÞ ÀI f Þ=ðI 0 À I f Þ was averaged over the sample channel region adjacent to the middle 300-m HMM. The fluorescence intensity changes exponentially with a time scale meas % 20 s, which is slower than the D ¼ 7 s predicted from simple diffusion through an inert HMM,

(b) and related Video 4]. Fabrication
(or, if solute was introduced from both sides and reservoirs synchronized, a factor of 4 faster). This result suggests that the HMM introduces a mass transport resistance comparable to that of the sample channel, discussed below. By decreasing the width of the sample channel, even faster solution swapping could be performed (see Video 5). A HMM-RaD device with a sample channel of width w s ¼ 23 m and a HMM of width w m ¼ 17 m [for which Eq. (3) predicts D ¼ 0:4 s] was fabricated for this purpose. Figure 5(c) shows the relative intensity averaged over the entire sample channel width and 1 m of HMM at the sample interface. The intensity in the sample chamber closely follows that in the HMM, which changes exponentially with time constant m ¼ 2:5 s. Swapping between 250 mM and 0 mM NaCl results in 90% NaCl removal in 5.6 seconds. Given the much faster D from Eq. (3), the HMM-RaD is limited by the HMM itself. Figure 5(d) shows excellent agreement between the experimental concentration profile cðx; tÞ and analytical theory within the sample chamber, imposing the experimentally observed m as a boundary condition, but introducing no other free parameters (see the SM [62] for theoretical profile derivation; see also [72]). We anticipate that faster switching (of order D ) will be possible by optimizing HMM chemistry.
Several mass transport processes may lengthen the time required for HMM-RaD beyond Eq. (3): (i) diffusion across the boundary layer between the reservoir and the HMM (with resistivity < R $ R =D, where R $ Pe À1=2 R L m ), (ii) diffusion across the sample chamber (< s $ w s =D), and (iii) diffusion through the HMM itself (< m $ w m = ðD m PÞ). Here, P is the partition coefficient, which expresses the ratio between equilibrium concentrations in the HMM and in the solution. The diffusivity D m may be slower within the HMM than in solution, with corrections due to lower porosity or mesh size that are often minor, or corrections that are more significant when the species adsorbs and desorbs from the gel. In addition to these steady-state mass transport limitations, highly partitioning solutes P ) 1 require a finite time C $ Pw m C s =j D to ''fill'' the HMM, before even quasisteady diffusive permeation occurs.

VI. LOCAL ELECTRIC PERMEABILITY
Most microfluidic walls are electrically insulating, including NOA-81. The electric permeability of hydrogels, by contrast, suggests that HMMs may be used to sculpt electric fields in microfluidic systems, using electrodes in solutions that are physically distinct from the solution of interest.
To demonstrate the electrical behavior near HMMs, we electrophoretically oscillated 500-nm fluorescent polystyrene colloids [73] (Bangs Labs, FS03F) in a 1-mM NaCl solution in the sample channel of our standard threechannel device by imposing a 5-Hz, 400-V pp , potential difference between stainless-steel pins at the inlets of the two reservoir channels. A 0.4-s exposure (Fig. 6) shows colloids tracing the local electric-field lines, much like iron filings in a magnetic field, confirming that the electric field passes through HMMs but not microchannel walls.
Integrating HMMs into electrode systems alleviates many issues that arise in conventional electrokinetic microfluidic systems. First, HMMs require no additional cleanroom processes, unlike photolithographically patterned electrodes, which require cleanroom metal deposition steps and separate masks. Second, electrochemical (Faradaic) reactions change solution composition near electrodes, e.g., generating gas bubbles or introducing gradients in pH or metal ions, potentially fouling nearby samples of interest. Instead, placing metal-pin electrodes in flowing electrolyte reservoirs behind HMMs allows undesirable electrochemical-reaction products or bubbles to be continually flushed away, thus maintaining steady electrolyte solutions and unfouled sample environments. No cleanroom time is required for the electrokinetic systems in Fig. 6, other than initial silicon master fabrication, although photopatterned electrodes could obviously be deposited onto channel floors using standard cleanroom techniques, if desired. Figures 6(b) and 6(c) and related Video 6 show a continuous-flow particle separation enabled by HMMs. A dc field is applied across the HMMs, transverse to a pressuredriven flow, causing the flowing particles to move electrophoretically to one side of the channel. The direction of colloidal motion and downstream width-fraction containing colloids can be controlled by changing the voltage.
This ability to sculpt electric fields naturally suggests novel capabilities for a variety of experiments, including electrophoretic separations [74], local electrophoretic delivery [75], microfluidic salt bridges (e.g., for maintaining separate reference and working electrode solutions [76]), and localized or selective electroporation of cells, organisms, or vesicles [77]. While electrically permeable gels could also be made with contact lithography for these applications (and have been, in some cases [78][79][80][81]), HMMs enable closer electrode spacing, allowing higher currents and smaller dimensions to be obtained. With further development, HMM electrophoresis could be used for detection of changes in electrophoretic mobility (e.g., from binding of analyte) or cell sorting (cell separation based on an upstream measurement).
The locally integratable nature of HMMs allows novel design capabilities for concentration fields, e.g., gradient shaping and integration into more complex chips without stringent demands on overall pressure balances. Multiple HMMs, each with its own distinct and individually addressable reservoir channel, enable the creation of almost arbitrarily complex chemical microenvironments composed of multiple species. The speed, ease, and small chip area of the HMMs makes them particularly appealing for producing on-chip concentration gradients, compared to other microfluidic methods.
Using our standard three-channel geometry, we apply HMMs to suddenly impose a fluorophore concentration gradient [ Fig. 7(a)]. All three channels contain a 10-mM MES buffer, but each carries a different concentration of the fluorophore Oregon Green 488 (Invitrogen)-0 mM, 0.5 mM, and 1 mM, respectively-so that flowing channels appear dark, medium, and bright. Within several minutes of injecting these solutions, flow in the sample channel is quickly stopped using PDMS valves (0 s). A linear gradient is quickly established (20 s) and maintained at a steady state. The predicted diffusion time [ D ¼ ð2w m þ w s Þ 2 = ð 2 DÞ ¼ 9 s, based on D ¼ 4 Â 10-10 m 2 =s [100]] agrees well with the transient time scale (i.e., t % 2 D for the transient to decay by 90%).
Solute gradients drive colloids into diffusiophoretic motion [Figs. 7(b) and 7(c)]. Nonetheless, it has been difficult to directly visualize colloidal diffusiophoresis, as concentration gradients strong enough to drive diffusiophoresis generally give rise to hydrodynamic instabilities in macroscopic experiments. By contrast, the small dimensions of microfluidic systems enable even strong gradients to remain hydrodynamically stable. Recent microfluidic experiments have visualized diffusio-electrophoresis in salt gradients [48,49,91,92]. Complementing these studies, here we demonstrate localized diffusio-electrophoresis along the sculpted gradients enabled by HMMs. Figure 7 both inlet tubes in a small beaker [101], and the following 8.3 s of fluorescence images were superposed. Beads clearly migrate diffusiophoretically up the gradient in NaCl, even tracking the two-dimensional (2D) ''fringing'' shape once the gradient is established. HMMs can also be used to establish solvent gradients, which can drive colloidal diffusio-chemiophoresis [87,90] or ''solvophoresis'' [102]. While solvophoresis has been seen in macroscopic aggregation experiments [102], it has never been visualized experimentally. Exploiting the solvent compatibility of HMMs and NOA-81 devices, we show the first direct visualization of colloids moving solvophoretically [ Fig. 7(c)].
To generate solvophoresis using HMMs, we start with flows of (left to right) 50%, 25%, and 0% v/v ethanol solutions in DI water ( D ¼ 2:9 s), with 10 mM KCl in each to balance ionic strength, and we superpose 10.4 s of images once the central flow is stopped. Polystyrene colloids migrate down the ethanol gradient solvophoretically [ Fig. 7(c) and Video 8].

VIII. CONCLUSION
We have demonstrated a simple method for making thin (10-25-m) hydrogel membranes that can be integrated into specified locations inside standard microchannels. HMMs enable exquisite control, in space and time, over local chemical microenvironments in microfluidic systems. Because of their high-diffusive and low-hydraulic permeability, reservoirs of constant composition can be established on one side of a HMM, driving diffusion of solute and solvent to the other side. Established techniques for microfluidic logic and flow control allow reservoirs to be switched quickly, resulting in rapid diffusive delivery to samples due to the short length scales typical in microfluidic systems. HMMs are permeable to a broad range of solutes and solvents, and their chemical or physical properties can be tuned for many applications. Integrating multiple HMMs, each with its own independently controlled reservoir channel, will enable strong gradients of multiple species to be imposed and, more broadly, spatiotemporal sculpting of local chemical microenvironments. Permeability of HMMs to electric fields opens avenues for electrokinetic studies and electrophoretic (rather than diffusive) delivery of solute.
We anticipate that this basic HMM strategy will enable a broad range of novel experimental capabilities in the science and engineering of physical, biological, chemical, medical, and material systems. To aid in adapting the HMMs for new applications, we note that the chemical and mechanical properties of the microwindow membranes themselves can be modified simply by adjusting the precursor solution composition, taking advantage of the versatile acrylate chemistry. One such method was described here, wherein the hydraulic permeability was decreased by increasing the PEG-DA precursor concentration. Other methods to tune hydrogel properties have appeared in the flow lithography and UV-curable hydrogel literature, including enabling permeability to larger solutes (e.g., polymers, proteins, quantum dots), e.g., by introducing a porogen in the precursor solution [103] or novel photolithography techniques [104]; modifying hydrogel chemistry by polymerizing in the presence of other reactive molecules (e.g., acrylate-functionalized charged groups or receptors for specific ligands [105,106]), magnetic nanoparticles [107], or degradable components for controlled release [108]. Photopolymerizing within parallel laminar streams of distinct precursors [52] or in the presence of gradients [109,110] would generate anisotropic HMMs. Electrical conductivity of HMMs can be enhanced by photopolymerization of highly conductive polymers [79][80][81]. More generally, nearly any UV-curable solution can be patterned using this method, as long as the curing time is fast compared to the diffusion time for reactive species in the precursor solution (i.e., Dämkohler number Da ) 1). Incorporating reactive species with lower diffusivities within the precursor solution may also decrease the HMM width.
New HMM applications and experiments will likely require new channel and membrane designs. Various photomask geometries could be used to make other HMM shapes, as long as uncured precursors can be flushed away or removed. The ultimate resolution of HMMs is set by the optical properties (e.g., depth of focus) of the microscope, as well as the optical wavelength used for photocuring. In particular, sharply defined HMM features can be attained within channels whose maximum height is no taller than the objective's depth of focus at a desired magnification [50,52]. To cross-link HMMs uniformly throughout taller channels (e.g., to accommodate larger samples), an increased depth of focus is required. This depth of focus can be attained with a lower magnification objective, but at the expense of HMM resolution in the sample plane. More complex, multiheight silicon masters could be fabricated to preserve this high resolution.
To conclude, hydrogel membrane microwindows provide exquisite control over microscale chemical transport, at low expense and with little fabrication time required. Few barriers exist for incorporating the technique in a wide variety of experiments, as they can be integrated into standard microfluidic devices, equipment, and techniques. The combination of simple fabrication, facile integration, versatile chemistry, independent control of multiple, noninterfering HMMs, and tight spatiotemporal control suggests broad impact across physics, chemistry, biology, and materials studies.

ACKNOWLEDGMENTS
We gratefully acknowledge support from the Arnold and Mabel Beckman Foundation and the Institute for Collaborative Biotechnologies through Grant No. W911NF-09-0001 from the U.S. Army Research Office. The content of the information does not necessarily reflect the position or the policy of the Government, and no official endorsement should be inferred. A portion of this work was performed in the UCSB Nanofabrication Facility, part of the NSF-funded NNIN network. The authors wish to thank Dave Walker and Carl Hansen for aiding our multilayer-soft-lithography foundry setup, as well as Nathaniel Lynd for early discussions.
APPENDIX A: FABRICATION DETAILS

Standard HMM fabrication
Our standard HMM fabrication procedure ( Fig. 1; see the SM [62] for details, with Refs. [9,10,60,67,[111][112][113] therein) builds on recent techniques involving in situ photopolymerization of UV-curable hydrogels [52][53][54][55][56]. An aqueous hydrogel precursor solution was prepared using 20% v/v polyethylene glycol diacrylate (PEG-DA, n ¼ 400, Polysciences Inc.), 2% v/v photoinitiator (2-hydroxy-2-methylpropiophenone, Sigma Aldrich), and silica tracer beads (1 mg/mL, Bangs Labs, 500 nm diameter). Just prior to use, this mixture was vortexed for 3 min; then, compressed air was bubbled through for 10 min to increase dissolved O 2 and prevent excess cross-linking. A photomask (a 60-m-wide transparent slit in 1.5-cm-diameter transparency, CAD/Art Services) was inserted into the field stop of an inverted microscope (Nikon TE-2000U). Patterned UV light was magnified through a 20x extra-long workingdistance (NA ¼ 0:45) objective and aligned to the device and fluorescence camera (Andor iXon 885). Then, 10-m-tall microchannels made of NOA-81 [60] were filled using a syringe connected to the central channel. The focus was adjusted to the center of the channel height. The flow was slowed to & 10 m=s by adjusting the syringe pressure to balance any stray flows and viewing the resulting tracerbead motion under phase contrast. Cross-linking was achieved using 550 ms of UV illumination from a mercury lamp (X-Cite 200DC, Lumen Dynamics) sent through a UVfilter cube (Nikon UV-2A, UV intensity after filtering: 30 mW=cm 2 ). Leak-free HMMs were verified by gently pressurizing the syringe and checking for bead motion on the other side of the membrane. If no leaks were found, the channels were promptly flushed with more precursor to prevent excess cross-linking. The cross-linking was repeated for each additional HMM. Precursor solution was then removed by flushing channels with DI water for several minutes.
To fabricate the higher-density HMM described in Sec. IV B, the same procedure was used, except the precursor solution contained 50% v/v PEG-DA and 5% v/v photoinitiator, and 110-ms exposure time was used.

HMM-rapid dialysis device fabrication
To investigate the temporal response of HMM-RaD experiments, we designed and fabricated trilayer devices [ Fig. 5(a); see the SM [62] for the fabrication diagram]. The fluidic layer, cast in NOA-81, consisted of a sample channel (width w s either 150 m or 23 m), separated from a reservoir channel (300 m wide) by a HMM of width w m of 17 m. Solutions were delivered to the sample and reservoir channels through channels in a PDMS layer bonded atop the NOA-81 layer [112], and flows within the PDMS delivery channels were controlled with multilayer soft lithography ''pushup'' valves [10] at intersections of the control and injection channels. In order to rapidly change between two reservoir solutions, two valved injection channels met over the NOA-81 reservoir inlet; all other inlets had a single valved injection channel for eliminating flow. The single-layer NOA-81 device and two-layer PDMS device were fabricated separately according to standard ''stickers'' [60] and multilayer-soft-lithography [9] protocols, respectively, then ozone-bonded together. To achieve precise valve timing, PDMS control channels were filled from sealed fluid containers, which were switched between 0 and 750 mbar using computer-controlled solenoid valves (Pneumadyne S10MM-30-12-3) [9], while the two reservoir solutions were pressurized to 250 mbar.