Passive and reversible area regulation of supported lipid bilayers in response to fluid flow

Abstract

Biological and model membranes are frequently subjected to fluid shear stress. However, membrane mechanical responses to flow remain incompletely described. This is particularly true of membranes supported on a solid substrate, and the influences of membrane composition and substrate roughness on membrane flow responses remain poorly understood. Here, we combine microfluidics, fluorescence microscopy, and neutron reflectivity to explore how supported lipid bilayer patches respond to controlled shear stress. We demonstrate that lipid membranes undergo a significant, passive, and partially reversible increase in membrane area due to flow. We show that these fluctuations in membrane area can be constrained, but not prevented, by increasing substrate roughness. Similar flow-induced changes to membrane structure may contribute to the ability of living cells to sense and respond to flow.

Significance

Using isolated membrane patches derived from giant unilamellar vesicles, we show that supported lipid bilayer area increases quickly and significantly upon exposure to shear flow. This increase is partially reversible in membranes on smooth glass surfaces, and manifests as membrane tubulation when the membrane is supported on an oxidized polydimethylsiloxane surface instead of glass. This phenomenon may be of interest to those who use supported bilayers as an experimental platform.

Introduction

A membrane-cytoskeleton complex forms the boundary between the inside and outside of animal cells. This complex is made of a fluid lipid bilayer studded with membrane proteins, which is closely linked to a scaffolding network of cytoskeletal protein fibers. This hybrid structure has complex mechanical properties. Dynamic and structural qualities of lipid membranes, such as out-of-plane membrane fluctuations, individual lipid and protein diffusion, and lipid asymmetry are altered by the close association of the membrane with its cytoskeleton. The mechanical properties of membrane-cytoskeleton complex regulate biological function; for example, membrane lipid viscosity limits lateral diffusion of proteins (1).

Supported lipid bilayers (SLBs) are widely used in experimental biophysics because they provide an efficient way to study lipid and protein dynamics on a single, flat lipid membrane with a simplified composition. SLBs are separated from their solid substrate by a layer of water approximately 1 nm thick, which allows individual lipids to diffuse nearly as fast as those in membranes surrounded by fluid (2). However, strong coupling with the substrate can modify membrane properties; for example, dimyristoylphosphocholine (DMPC) bilayers on mica substrates exhibit distinct gel transition temperatures in their upper and lower leaflets (separated by approximately 8°C) (3). Substrate interactions limit the accuracy of solid-supported bilayers as models for the membrane-cytoskeleton complex. However, these interactions can be reduced by increasing the spacing between the membrane and the surface; for example, using polymer or oligonucleotide tethers to link the membrane to the surface increases membrane protein mobility (4,5).

SLBs are generally made by allowing small or large unilamellar vesicles to fuse on a hydrophilic surface, forming a large, continuous bilayer. Forming membranes in this way confers control over lipid and protein content, and recent techniques incorporate lipids and proteins obtained directly from cell plasma membranes (4,6). One of the practical advantages supported membranes offer to experimentalists is that their adhesion to the surface keeps them flat and stable to perturbations such as solvent exchange. This stability makes many important experiments possible. Examples include measuring virus binding to membranes (7,8), observing oligomerization and concentration waves in diffusing membrane proteins (9,10), measuring flow transport of membrane-bound proteins (11,12), and measuring interleaflet friction and interleaflet coupling (13,14,15).

Recently, experimentalists have generated supported bilayers by a new method: forming discrete patches of supported membrane by allowing giant unilamellar vesicles (GUVs) to rupture at the surface (16,17,18). GUV fusion with a hydrophilic surface results in a supported membrane with the inner leaflet of the GUV facing up and the outer leaflet proximal to the surface (19), in contrast to small unilamellar vesicles (SUVs), which fuse so that the outer leaflet faces away from the surface (5). GUV-derived membrane patches contain membrane edges and preserve any pattern of coexisting lipid domains that was present in the GUV (16,20). Discrete supported membrane patches retain the control of lipid composition, stability, and physical properties of continuous supported membranes, while providing additional spatial constraints to the experimental system.

In unsupported fluid membranes, flow in the surrounding liquid generates corresponding flows of lipids, which can be observed as motion of liquid domains in giant vesicles or the transfer of momentum across biological membranes (21,22,23,24). However, proximity to a solid substrate prevents collective lipid motion, as indicated by the immobility of micrometer-sized domains in supported bilayers formed from membranes with fluid-fluid coexistence (16,25). Experiments investigating the impact of shear stress on living endothelial cell membranes via fluorescence recovery after photobleaching (FRAP), Laurdan polarization, and a viscosity-dependent fluorescence probe all indicate a rapid drop (within tens of seconds) in cell membrane order after flow onset, followed by a partial recovery after cessation of flow (26,27,28). The change was particularly significant in areas where membrane order was initially highest. These rapid changes to membrane order are expected to modulate shear stress responses in these cells, and a specific explanation for the change remains lacking (27). In living cells, multiple mechanisms for such a change are possible, including changes to membrane lipid and cholesterol content, shifts in local concentrations of ions such as Ca²⁺ and protons, and membrane remodeling via exo- and endocytosis.

Here, we report an unexpected observation we made when applying fluid flow to single-phase GUV-derived supported membrane patches: under physiologically relevant levels of shear stress, the apparent area of membrane patches increases quickly by 15%–20% (Fig. 1), and this increase is partially reversed after flow stops. Significant alterations to lipid packing and order parameter, as well as mobility, may accompany this area expansion. We describe observations of this area change, which we have investigated using fluorescence microscopy, atomic force microscopy (AFM), and neutron reflectivity.

Figure 1.

Flow alters the area of SLBs. Lipid bilayer patches were deposited on the glass coverslip base of a rectangular PDMS channel (A and B). Membrane patches composed of 99.2 mol % DiphyPC and labeled with 0.8 mol % Texas red DPPE initially form with smooth edges (C and D, left columns). Upon exposure to low (C) or high (D) shear stress, lipid patches expand (C and D, right columns). In all images, the flow direction was from right to left. To see this figure in color, go online.

Materials and methods

Experiment

Microfluidic flow device

Rectangular polydimethylsiloxane (PDMS) microfluidic channels of dimensions 300 μm wide and 100 μm tall were made using a 10:1 mixture of Sylgard 184 to curing agent (Dow Corning, Midland, MI) and cured overnight at 65°C. Master molds of the microfluidic channels were made from SU8 50 (MicroChem, Newton, MA), which was spun to a thickness of ∼100 μm, and then exposed to ultraviolet light through a high-resolution mask. Glass coverslips were treated in a saturated solution of potassium hydroxide in ethanol to flatten them following a previously published procedure (29), to produce a smooth substrate surface for lipid patch formation. The treated glass slides were exposed to low-pressure air plasma for 10 min. The PDMS microfluidic devices were exposed to low-pressure air plasma for 30 s to allow the glass slide and channel to be bonded together. To produce PDMS substrates, a 100-μm layer was spin-coated onto a coverslip as described previously, cured, and the coated coverslip was exposed to low-pressure air plasma for 30 s just before the microfluidic device was assembled as above (30). All microfluidic devices were filled with deionized water directly after bonding to maintain the hydrophilicity of the surface. We calculated the shear stress at the lower coverslip using a simulation (COMSOL Multiphysics, COMSOL) of the rectangular channel for the flow rates used in our experiments (Fig. 2 A). In rectangular channels, shear stress varies with the distance from the channel center (Fig. 2 B); here, the values we report are the average over the entire channel width. For the elongated gasket used in the flow cell for neutron reflectometry, we report the average shear stress inside the beam footprint as indicated in Fig. 2 C.

Figure 2.

(A) COMSOL simulation of shear stress at the lower surface of a rectangular microfluidic channel. (B) Values of shear stress at the coverslip surface for a range of different flow rates. Values were measured across the dotted white line in (A). (C) COMSOL simulation of shear stress at the lower surface of the neutron flow gasket. The dotted region shows the sample area where shear stress is approximately uniform, which was used to define the footprint of the neutron beam. To see this figure in color, go online.

Bilayer formation

1,2-Diphytanoyl-sn-glycero-3-phosphocholine (DiphyPC), 1,2-dioleoyl-sn-glycero-3-phosphocholine (DOPC), 1,2-dimyristoyl-sn-glycero-3-phosphocholine (DMPC), 1,2-dilauroyl-sn-glycero-3-phosphocholine (DLPC), and 1-palmitoyl-2-oleoyl-sn-glycero-3-phosphocholine (POPC) were purchased from Avanti Polar Lipids (Alabaster, AL). Trizma base, glucose, sucrose, sodium chloride, ethylenediaminetetraacetic acid (EDTA), and calcium chloride were obtained from Sigma-Aldrich (St. Louis, MO). All lipid bilayers used in fluorescence imaging were doped with 0.8 mol % Texas red 1,2-dipalmitoyl-sn-glycero-3-phosphoethanolamine (TR-DPPE), which was obtained from Invitrogen (Eugene, OR). All lipid samples were stored in chloroform at −20°C until use. We produced GUVs by electroformation using standard protocols (14,31). Vesicles were formed in a 200 mM sucrose solution to aid vesicle sedimentation, and were used within 6 h of electroformation. Prepared GUVs were diluted into a solution of 200 mM glucose with 3.7 mM calcium chloride, then injected into a prepared microfluidic channel to form many discrete supported lipid patches via spontaneous rupture of GUVs. A buffer solution (100 nM NaCl, 10 mM Tris, and 1 mM EDTA at pH 8) was then slowly flowed through the microfluidic channel using a syringe pump (Harvard Apparatus, Holliston, MA) (resulting in average shear stress of 0.016 Pa) to flush excess vesicles from the device. The syringe pump was then used to force the buffer solution across the lipid patches at a range of higher flow rates. To make continuous supported bilayers, GUVs prepared as above were extruded through a filter with 100-nm pores (Avanti Polar Lipids) nine times to convert them into large unilamellar vesicles (LUVs). The LUV solution was diluted into the same glucose and calcium solution as above, injected into a microfluidic channel, and allowed to rupture on the surface.

Fluorescence imaging

For each experiment between 15 and 20 membrane patches were selected and imaged with a 40× objective using a spinning disk confocal microscope (3i, Denver, CO) equipped with a motorized translation stage. Fluid flow was applied for approximately 20 min at each flow rate. During this period, images were captured every 30 s using an sCMOS camera (Photometrics, Tuscon, AZ). The change in lipid patch area and the mean fluorescence intensity of lipid samples during the flow experiments were analyzed using ImageJ (32). During introduction of the tubing connected to the syringe pump, transient flows occur, which may prematurely expand the lipid patches. Therefore, in experiments to determine the extent of expansion, reference images of each lipid patch were recorded after the vesicles were allowed to settle and burst, but before tubing was connected, and the flushing step was eliminated.

AFM imaging

A Bruker Dimension V atomic force microscope was used to characterize the surface roughness of glass and PDMS substrate samples. Images were taken in contact mode using silicon nitride PPP-NCHR tips from Nanosensors (Neuchatel, Switzerland). AFM tips had a nominal spring constant of 42 Nm⁻¹. Initial roughness characterization was obtained from five 5 × 5-μm scans of substrate surface at 256 × 256 pixels and values of root-mean-square (RMS) roughness were calculated using Gwyddion.

Neutron reflectivity

Neutron reflectometry experiments were performed at the LIQREF beamline at the Spallation Neutron Source at Oak Ridge National Laboratory with a two-dimensional position-sensitive ³He detector (33). The reflected intensity was measured as a function of the momentum transfer q = 4πsin(Θ/λ), where Θ and λ are the incident angle and wavelength of the neutron beam, respectively. A 3.4 Å bandwidth, extracted from a wavelength range of 2.55–16.70 Å, was used in conjunction with four angle settings (Θ = 0.60°, 0.76°, 1.40°, and 2.59°) to attain a q range of 0.008–0.222 Å⁻¹. A fluid flow cell was previously developed and used at this instrument to measure the impact of shear stress on cell-substrate connections (34). The cell was constructed to compress two 5-cm-thick wafers on opposite sides of a gasket, creating a flow chamber. The upper quartz wafer has two holes, which are accessed by Luer fittings to serve as inlet and outlet ports for fluid, allowing buffer to be circulated through the cell using a peristaltic pump. The material of the lower wafer, where the bilayer is formed, can be chosen to optimize experimental conditions. To most closely match the glass coverslip substrates that we used for fluorescence measurements, our lipid bilayer samples were deposited on quartz wafers (El-Cat, Ridgefield Park, NJ). The neutron beam was directed up through the lower quartz wafer and reflected downward from its top surface, at the interface with the deuterated flow buffer. Our objective for neutron reflectometry experiments was to probe changes to membrane thickness. Therefore, we used identical protiated lipids to those used in fluorescence experiments. We chose a D₂O-based flow buffer to minimize incoherent background at higher q while ensuring that the entire sample was seen by neutrons. When making the deuterated flow buffer, pH meter readings were adjusted by adding 0.41 to compensate for the deuterium content (35).

We modified the flow chamber using a custom-manufactured silicon gasket (Bioptechs, Butler, PA) based on previous calculations of shear stress in parallel-plate flow chambers (36). This gasket created approximately constant shear stress over the 17 × 17-mm footprint of the neutron beam (Fig. 2 C). The footprint was kept constant through adjustments in the slit opening commensurate with the angle of incidence. We generated GUVs by electroformation, then extruded them through a membrane with 30-nm pores to generate SUVs. The quartz surface was prepared by repeated rinsing with chloroform, ethanol, and deionized water, followed by 30 min of UV-ozone treatment just before sample loading. We injected the SUV solution into the flow cell to deposit continuous bilayers of DiphyPC or DLPC on a freshly cleaned quartz wafer that formed the lower surface of the cell. After slowly flushing excess SUVs out of the chamber with a hand-held syringe, the flow cell was connected to a peristaltic pump. We collected reflectivity data on these samples in three conditions: 1) before applying flow, called “static”; 2) while applying shear stress at the quartz surface of approximately 0.18 Pa using the peristaltic pump, called "shear"; and 3) after stopping the pump and allowing at least 60 min of relaxation time before beginning acquisition again, called "recovery." The value of shear stress applied in this experiment was lower than the ones applied in fluorescence experiments because higher flow rates caused the sample chamber to leak during data acquisition, which required signal collection over periods ranging from 4 to 16 h. We confirmed using fluorescence microscopy that pulsatile flow generated by a peristaltic pump resulted in membrane patch expansion similar to that observed when continuous flow from a syringe pump was applied.

The neutron reflectometry data were collected with a fixed relative resolution, δq/q = 0.023, more than adequate for the measurements of lipid bilayers. We used LSFIT (37), a least-squares fitting routine applying the Levenberg-Marquardt algorithm to fit the data, then confirmed these models with Refl1D (38). Laterally averaged scattering length density (SLD) profiles perpendicular to the sample were finally obtained. Details of the data fitting procedure can be found in previous works (39,40).

In our experiments, incoherent background from the protiated lipids and the solvent began to take over signal at q values higher than about 0.19 Å⁻¹. This meant that we could only capture one fringe of the reflectivity curve from a single lipid bilayer. For a film with thickness typical for lipid bilayers, 40 Å, the first fringe occurs at about q = 0.185–0.2 Å⁻¹, but the second fringe would occur at 0.31 Å⁻¹. Therefore, we do not obtain membrane thickness directly from the inter-fringe distance in our curves, but instead use curve fitting to estimate membrane thickness. The fit layers included the quartz, deuterated buffer, inner and outer bilayer head group regions, and inner and outer bilayer tail regions, as illustrated in Fig. 7 A. We fixed roughness at 3 Å, used an initial guess for the thickness D of the layers, and fit to find the SLD values. Then, SLD and thickness were fit iteratively, with SLD and thickness values for each layer in the inner leaflet constrained to match those in the outer leaflet. The deuterated buffer cushion between the quartz and the bilayer was estimated to have a thickness of 1 nm. This fitting method neglects any membrane asymmetry that may be present. Our resulting fit values for SLD and D are reported in Table S1; roughness parameters did not change from the estimated 3.0 Å and are not reported.

Figure 7.

Neutron reflectivity results suggest that supported membranes become thinner during shear stress. (A) A sketch of the model corresponding to the SLD profile in (B) with z = 0 Å assigned to the center of the deuterated buffer layer between the quartz substrate and the lipid bilayer. Slabs representing the upper and lower leaflet head regions (HR_upper and HR_lower) and tail regions (TR_upper and TR_lower) were allowed to vary in thickness, but symmetry was required. (C) The reflectivity curves for a single continuous DiphyPC bilayer before, during, and after application of shear stress. Curves collected under each condition are vertically displaced for ease of comparison. Fits of the resulting SLDs (C and D) reflect a small decrease in the membrane thickness, as well as a decrease in the SLD in the hydrophobic tail region, during shear. (E and F) A similar flow response was observed in DLPC bilayers. To see this figure in color, go online.

Results

Reversible expansion of membrane area under flow

Isolated membrane patches derived from single GUVs displayed a dramatic and partially reversible area expansion (Figs. 1 C, D, and 3 A–C; Video S1). Expansion was complete within a few minutes of flow onset, and was independent of the flow direction. Surprisingly, after flow stopped, the patches slowly contracted again and the cycle of expansion and contraction could be repeated, although recovery to the original area was incomplete (Fig. 4 A). In the absence of flow, membrane area stayed constant, with less than a 2% increase in fluid membrane area after 12 h of observation.

Figure 3.

Fluorescence images of a typical DiphyPC patch before (A), during (B), and after (C) shear stress. Relative changes to the average surface area (D and E) and the fluorescence intensity (F and G) versus time show that intensity dropped and recovered commensurate with the area expansion. Blue shading indicates time points when flow was turned on. In (D)–(E), points denote the area of the single patch, and red-shaded regions indicate the estimated uncertainty in the area measurement. In (F)–(G), points denote the average intensity measured over the entire patch, and the red-shaded region is one standard deviation wide. To see this figure in color, go online.

Figure 4.

Lipid bilayer patches composed of 99.2 mol % DiphyPC and labeled with 0.8 mol % Texas red were exposed to 5 on/off flow cycles applying 3.14 Pa of shear stress. (A) Relative changes in the average surface area during a series of five flow cycles, with blue vertical shading indicating time points when flow was turned on. The points show the mean area increase from 13 different patches. (B) Fluorescence intensity dropped and partially recovered in a continuous supported membrane when flow was applied (red line), but stayed constant under static conditions (black line; blue shading indicates time points when flow was turned on). Intensity was measured in 10 different locations on the same continuous bilayer. (C) DiphyPC patches were subjected to 10, 20, or 60 min of low (0.314 Pa) shear stress; similar expansion and retraction behavior was observed in all cases. In all plots, symbols represent the average relative area, and the shaded area indicates a region that is one standard deviation wide. To see this figure in color, go online.

Video S1. Three different membrane patches made from DiphyPC were first imaged just after deposition in static conditions, then subjected to 0.314 Pa of shear stress by flowing buffer at 0.05 mL/minute through the microfluidic channel

Flow was applied for approximately 20 min, then the syringe pump was stopped and imaging continued for approximately 10 min. Expansion of each patch, followed by a partial relaxation, can be seen. All scalebars are 20 μm.

Fluorescence intensity measured in the patches dropped during flow and recovered in synchrony with the area expansion and recovery (Fig. 3 F–G). Since this drop was also partially reversible, it is likely that it resulted from a temporary dilution of the labeled lipid due to the apparent area increase. Fluorescence intensity was previously observed to increase in lipid patches as membrane area decreased due to removal of cholesterol (30).

Relaxation of the membrane area after flow stopped occurred more slowly than expansion, and the membrane appeared unlikely to fully return to its original area. This incomplete recovery may be due to adhesions or pinning sites where lipids are strongly coupled to the glass surface. We observed that part of the expansion was reversible by applying five subsequent flow cycles to a microchannel containing DiphyPC patches. Expansion and recovery were observed to occur with each cycle (Fig. 4 A). In continuous SLBs, no membrane edge is visible (unless membrane defects are present), so it is not generally possible to observe changes to membrane area. However, we observed that flow-driven drop and recovery of fluorescence intensity also occurred in continuous bilayers made by LUV fusion (Fig. 4 B), suggesting that continuous supported bilayers respond to shear stress in the same way that discrete membrane patches do. By applying low shear stress (0.314 Pa) to DiphyPC patches for 10, 20, and 60 min, we determined that the extent of area increase and recovery were not dependent on the time flow that was applied (Fig. 4 C).

The magnitude of initial expansion does not depend on shear stress or lipid chain structure

The membrane area increases reported in Figs. 3, 4 A, and 6 E are determined relative to the initial area of the membrane patch, which was recorded after slowly flushing excess vesicles out of the microchannel. Subsequently, we investigated whether the magnitude of shear stress was correlated with that of the resulting expansion. We were surprised to find that patch expansion was roughly independent of shear rate, down to the smallest value we applied (0.314 Pa, Fig. 5 A and B). This sensitivity to even very small flows meant that, to quantify changes in area, we had to adjust our procedure so that initial membrane area was recorded before applying any flow at all. This method yielded reproducible values for the extent of initial expansion of membrane patches, and it was used to obtain the data shown in Figs. 4 C and 5 A–D. This insensitivity to shear stress magnitude is the reason that we report average shear stresses instead of the individual stress applied to each patch, which varies with proximity to the edge of the microfluidic channel. Fig. 5 A shows that the rate of expansion was slower for the lowest shear stress investigated, but that expansion was complete within 1–2 min at all higher stresses. For this reason, the expansion at 0.314 Pa in Fig. 5 B was measured after 20 min of flow, whereas others were measured after 10 min.

Figure 6.

DiphyPC patches on an oxidized PDMS substrate exhibit a distinct flow response depending on substrate surface roughness, as determined by AFM. Glass roughness (A) was small (RMS roughness = 0.2 nm), consistent with previous measurements of glass prepared by this method (29). The PDMS surface undulations (B) have larger wavelength and amplitude. Average surface roughness (C) of glass and PDMS substrate measured with AFM; error bars denote standard deviation. (D) After a small expansion, bright puncta appear on the membrane surface and grow in intensity over time, suggesting that membrane tubules or protrusions extend out of the supported bilayer. (E) Apparent membrane area compared for membrane patches on PDMS and glass over the course of two flow and rest cycles, with blue shading indicating time points when flow is turned on. Plots represent the mean response from 16 different patches for the PDMS and 13 patches on the treated glass, with shading representing the standard deviation. To see this figure in color, go online.

Figure 5.

(A) Initial area expansion for patches exposed to four different shear stresses for 10 min was roughly constant. The yellow-shaded rectangle indicates the time points averaged to give the values shown in (B). In (B), area measurements from multiple microchannels were averaged. Expansion was measured after 10 min for all shear stresses except the lowest one, which was measured after 20 min. (C) Membrane patches composed of five different lipids were subjected to 3.14 Pa of shear stress; fluid-phase saturated and unsaturated lipids expanded by approximately the same amount. Patches made from DMPC were in a fluid phase at laboratory temperature, but entered a gel phase when flow was applied and showed no expansion. (D) Lipid patch expansion was independent of starting patch area. Pooled measurements from multiple experiments (n = 164) show no trend in the extent of expansion of individual lipid patches with patch size over the range we observed (patch radius between approximately 5 and 400 microns). Error bars and shading denote standard deviation for averaged values. To see this figure in color, go online.

Although the magnitude of patch expansion on glass was independent of flow rate, the final shape of the expanded membranes did depend on flow rate. Individual membrane patches generally form with smooth, round edges. At lower flow rates, ripples formed all around the membrane edge, and the patch expanded roughly isotropically (Fig. 1 C; Video S1). At higher flow rates, some of the original edge stayed intact and the expansion occurred only within distinct sections of the original boundary (Figs. 1 D and 3 A–C).

A second surprising result was that the extent of initial membrane expansion did not depend strongly on lipid acyl chain structure. To determine whether lipid acyl chain order was related to expansion extent, we decided to compare membrane patches made from lipids with different degrees of tail unsaturation. Molecular structures of the lipids we chose, as well as of the fluorescent lipid, are shown in Fig. S1. We produced lipid patches made from DiphyPC, DOPC, POPC, DMPC, and DLPC, and determined the average first expansion extent for each lipid at the same flow rate (Fig. 5 C). All experiments occurred at laboratory temperature (measured to be 27°C–28°C), which is significantly higher than the chain melting temperature for all lipids except DMPC. The extent of initial expansion was similar for DiphyPC, DOPC, POPC, and DLPC at approximately 18%. DMPC patches did not expand, but instead exhibited a flow-dependent transition to an apparent gel phase, consistent with results observed previously in continuous supported bilayers (15). Fluorescence in DMPC patches was sharply reduced after flow was turned on, and recovered over 10–15 min after flow stopped.

Surface roughness constrains but does not prevent area increase

We investigated how substrate surface chemistry affects expansion by forming membrane patches on an oxidized PDMS surface. PDMS surfaces oxidized in this way are expected to be hydrophilic, but their surface roughness is very different to that of etched glass (Fig. 6 A–C). Membrane patches formed readily on oxidized PDMS, but responded to shear stress differently than those on glass. After a small expansion, membranes appeared to form tubules that extended perpendicular to the surface during flow. Tubules coarsened over time and appeared to be partially reabsorbed after flow stopped. Only a minimal area decrease was observed during the no-flow period in patches on PDMS, in contrast to patches on glass. When flow was applied for a second time, no area increase was observed (Fig. 6 D and E).

Neutron reflectivity suggests that membrane thickness changes reversibly under shear

We collected a neutron reflectivity signal consistent with the presence of a single lipid bilayer (Fig. 7). We fit our reflectivity data using a five-layer model, including the deuterated buffer between the lipid bilayer and the quartz substrate, head, and tail for the upper and lower leaflets as shown in Fig. 7 A and B. Our reflectivity data contained incoherent background signal at the largest wavenumbers, which made it challenging to identify the first minimum in the signal (approximately 0.17 Å⁻¹). Neutron reflectivity data are useful because of sensitivity to changes in the SLD, which is related to changes in membrane density upon applied flows.

We were able to fit our reflectivity curves to estimate the thickness and SLD of each layer with a few constraints applied that neglected any membrane asymmetry possibly induced by the flow. Our results are listed in Table S1 and plotted in Fig. 8. We recorded reflectivity data for continuous bilayers made from DiphyPC and DLPC (Fig. 7 C and E). In static conditions, DiphyPC bilayers were 41.0 ± 2.1 Å thick, whereas DLPC bilayers were 36.2 ± 1.3 Å thick. These values agree well those measured previously (41). We found that bilayer thickness decreased from 41 ± 0.28 Å to 39 ± 0.17 Å for DiphyPC and from 36.2 ± 0.18 Å to 35 ± 0.16 Å for DLPC when shear was applied. After flow stopped, the DiphyPC bilayer thickness increased again, although not to the original value (Figs. 7 D and 8 A).

Figure 8.

(A) Total membrane thickness for DiphyPC bilayers (circles) and DLPC bilayers (triangles) decreased during shear flow, then partially recovered after shear flow stopped (recovery shown for DiphyPC bilayer only). (B) The SLD in the tail region of the membrane decreased during shear flow, indicating lower water content. This could be due to bilayer expansion filling holes or to increased ordering of the acyl tails. Error bars indicate fit uncertainties, which were estimated by manually adjusting the fit. To see this figure in color, go online.

If total membrane volume were conserved, the thickness changes we observed would require that the membrane area temporarily increase by ≈ 5% for DiphyPC, and ≈ 3% for DLPC during shear flow. This is significantly smaller than the ≈ 18% expansions observed by fluorescence for membrane patches. However, it is not possible to observe the same initial expansion of a membrane patch in the reflectivity experiment. Instead of discrete patches, we formed a continuous supported lipid bilayer on the quartz wafer to maximize our reflected signal. In addition, after forming the SLB for this experiment, we flushed excess vesicles and replaced the solution with deuterated buffer for contrast before starting to collect the static measurement. Therefore, the expansion we observed with shear flow was expected to be limited in extent, similar to the later expansions shown in Fig. 4 A.

During flow, the SLD values for the tail regions of both membranes decreased relative to the static value, and the value for DiphyPC partially recovered to its original value (Fig. 8 B). This reflects a decrease in water content, which could arise either from membrane expansion filling defects in the bilayer or from an increase in acyl chain order. These observations suggest that the reversible part of the membrane expansion we observe using fluorescence microscopy is accompanied by a change in membrane thickness. However, our results are not sufficiently precise to distinguish between changes in membrane thickness and changes in membrane roughness. We plan to complete additional experiments to examine these flow responses with improved resolution and repetition.

Discussion

Using fluorescence microscopy of discrete membrane patches, we observed a two-stage response to shear stress. First, within minutes after applying flow, the area of individual membrane patches increased dramatically, by between 17% and 20%. This area increase was independent of shear stress and of lipid acyl chain structure. However, the speed and the final shape of the increase was dependent on the magnitude of the flow rate (Fig. 1). Second, we observed that part of this initial expansion was reversible. After flow stopped, membrane area decreased again by approximately 5%, and we were able to repeatedly cycle the membrane area up and down by applying and stopping flow. The expansion response to flow only occurred in supported membranes that were in a fluid phase. DMPC membranes have previously been observed to enter a gel-like state when flow was applied (15), and we found that they did not expand under flow at all.

Flattening of membrane undulations

One possible explanation for the observed expansion is that membrane patches contain small vertical undulations that flatten out under shear flow. Earlier studies of the GUV rupture process suggest that membrane tension in vesicle-derived SLB patches is heterogeneous (18). Transient, micrometer-scale intensity fluctuations in GUV-derived supported membranes have been observed within the first 10–30 s after fusion onto multiple different substrates, indicating that the membrane undergoes significant reorganization even after vesicle bursting has completed (42). Measurements made with reflection interference contrast microscopy indicated no apparent membrane wrinkles greater than 10 nm in height across GUV-derived supported membranes, but the presence of transient membrane protrusions was identified. The membrane protrusions were around 20 nm in height and formed shortly after the initial fusion event, before dissipating within minutes (42). These observations support the idea that GUV-derived membrane patches contain excess area after they are formed, although the amplitude of membrane wrinkles, if present, is so small that they are not visible under confocal microscopy. These undulations might be flattened by a small shear force, then slowly re-form after the flow is stopped (Fig. 9 C and D).

Figure 9.

Two possible mechanisms for isotropic supported bilayer expansion under flow. The observed increase in membrane area could result from flattening of membrane undulations (C and D) or from an increase in the average area per lipid (E and F). To see this figure in color, go online.

This hypothesis could explain our observations that the amount of initial expansion is approximately constant regardless of the membrane lipid composition or the applied flow rate; expansion would be insensitive to these factors if it is primarily determined by the mechanism of membrane patch formation from GUVs. However, it would be surprising to observe a similar process occurring in continuous SLBs made from SUV fusion (Fig. 4 B), since these are likely to experience different forces during their formation. It is also challenging to reconcile this idea with our observations of tubule formation in membrane patches formed on textured PDMS: it requires that the membrane form very highly curved tubules at the expense of low-amplitude, large radius of curvature bends. However, flattening of membrane undulations is consistent with some of our observations, in particular the features of the initial expansion.

Alteration to area per lipid

Alternatively, shear stress could change the area per lipid in fluid membranes, perhaps by inducing lipid tilt, or by modifying head group orientation. To keep the total membrane volume constant, the membrane thickness would decrease (Fig. 9 E and F). Our neutron reflectivity results showing a small decrease in membrane thickness are consistent with this picture. A global change in the area per lipid could help explain why membrane expansion occurs symmetrically rather than in the downstream direction. Previous experiments show how lipid area is related to membrane thickness in series of saturated lipids with increasing acyl chain lengths (43). To make similar observations in our system, we could record neutron reflectivity measurements using additional solvent contrasts by using flow buffer made with different ratios of deuterated to protiated water. This would increase the accuracy of our data fitting to determine membrane thickness, and also make it possible to simultaneously determine membrane area changes.

Multiple coarse-grained simulations observe lipid tilting in the direction of parallel shear flow. This suggests that a similar area expansion could occur in unsupported membranes if sufficient shear stress were applied. Khoshnood et al. used similar simulations to observe that tilt angles of proteins embedded in bilayers increased under shear flow, but did not describe lipid tilting (44). In unsupported membranes, tilt angle is observed to be proportional to the shear rate. Benazieb et al. found that, for fluid-phase membranes, lipid tilt increased by 0.125° for each 100 kPa of applied stress (45); Shkulipa and den Otter also observed that tilt was proportional to shear stress (46). Applied to our experiments, the coefficient above would result in an undetectable area change since the largest shear stress we applied was approximately 5 Pa; however, it is not always appropriate to directly apply numerical results from coarse-grained simulations. Lipid tilt is a possible explanation for our observation of shear-induced gel phase in supported bilayers: flow-induced lipid tilt could effectively increase the transition temperature required for the membrane to enter an L_β phase.

We are not aware of any simulations of lipid bilayers in close proximity to a solid substrate. However, Blood et al. performed atomistic simulations of a two-bilayer system in shear flow with the objective of modeling flow near cell plasma membranes in 2005 (47). In this simulation, shear was effectively applied only to one leaflet of the membrane, whereas the other was in contact with stationary water. Lipids in contact with the flow tilted to an angle of approximately 15°, and increases in acyl chain length and lipid order were observed. In this simulation, tilt also increased with shear rate. Interestingly, their results were specific to the flow-side leaflet; negligible changes to lipid tilt and order occurred in the stationary-side leaflet.

In contrast to the atomistic and coarse-grained simulation results, we do not observe a strong dependence of membrane expansion on shear rate. It is also difficult to explain why lipids of different length and acyl chain saturation would expand by the same amount, if the expansion depends on creating lipid tilt. One possibility is that the constraint on lipid mobility in the lower leaflet due to the surface limits the ability of lipids in both leaflets to tilt. In the absence of tilt, it is unclear how shear stress would create lateral expansion in membrane lipids, rather than compression.

Temperature

The observed expansion under flow is unlikely to be explained by temperature changes. Flow buffer was prepared several hours before beginning flow experiments and allowed to come to equilibrium with the room, equipment, and syringe pump, so temperature changes resulting from flow are likely to be less than 1°C. The thermal expansivity constants for DiphyPC membranes were measured by Kucerka et al. in 2011 to be 0.0019 ± 0.0004 T ⁻¹ (41). Using this constant and considering a membrane that is initially at 20°C, a 40°C temperature change would be required to increase the membrane area by 8.7%. This is a smaller increase than the one we observe after applying moderate shear stress.

Modification of membrane mechanical properties by ions and buffer

In our experiment, vesicles were deposited on the coverslip in a 200 mM glucose solution containing 5 mM calcium. The solution was next exchanged for one containing 100 mM NaCl and 10 mM Tris, as well as EDTA to remove any residual calcium. The solution osmolarity stayed approximately constant during the change to the flow buffer. All subsequent flow cycles were applied using this buffer, so it is natural to consider whether the solution change might contribute to the initial phase of area expansion.

In vesicles, area per lipid remains constant as salt concentration is increased up to 1 M NaCl; when salt concentrations are increased further, an area reduction is observed, rather than an increase, as the membrane thickness increases (48, 49). Increases in bending rigidity (48) and order parameter (50) of zwitterionic lipid membranes occur when salt concentrations exceed 1 M. Finally, Wacklin found that the area per lipid for supported DOPC bilayers was identical whether the bilayers were deposited in 0 or 100 mM NaCl (51). Considering these results, we do not believe that the expansion we observe can be explained by the increase in NaCl concentration.

Calcium interacts more strongly with zwitterionic membranes. Uhrikova et al. found that low concentrations (below 3 mM) cause a drop in area per lipid (in DPPC bilayers at 60°C) of approximately 5% (52). As the concentration increases, the area increases, returning to the no-calcium value when the concentration exceeds 100 mM. If a supported membrane is relatively free of defects, then calcium ions could stay trapped below the membrane during the change from deposition solution to flow buffer. In that case, we would remove Ca²⁺ from the upper leaflet only and, as a consequence, could expect a 5% increase in the area per lipid in that leaflet. Thus, calcium removal might partially explain some of the initial expansion and explain why this initial expansion is not completely reversible.

These calcium effects were observed in multilamellar vesicles. In the specific context of solid-supported bilayers, another effect of cation concentration is likely to be important. Calcium and NaCl are frequently used to expedite formation of SLBs from SUVs, since in pure water there is an electrostatic repulsion between lipids and a glass surface (53). Thus, the presence of salt or Ca²⁺ in the flow buffer may facilitate flow-mediated area expansion by lowering substrate repulsion, which would otherwise be an obstacle to movement of the membrane. Schenk et al. observed in 2018 that continuous supported bilayers formed in high salt (0.5–1 M NaCl), washed with water, and then subsequently washed with buffer solution containing roughly physiological salt concentrations (150 mM) undergo rapid area expansion, resulting in tubulation (54). The extent of membrane expansion could not be measured since it was not possible to quantify the excess membrane area that formed as tubules. We observe that some of this tubule-forming expansion may have resulted from flow during solution exchange.

By observing thermal fluctuations in vesicles, Bouvrais et al. showed in 2014 that the bending modulus for POPC vesicles decreased significantly in the presence of various buffers. They also found that NaCl reduces bending rigidity, and the effect was even more pronounced in the presence of both Tris and NaCl (bending rigidity was reduced by almost 50%) (55). Small-angle X-ray scattering experiments from multilayer vesicles shows that buffers can alter membrane spacing (56, 57). AFM experiments indicate that buffers such as Tris can induce nanometer-scale texturing of supported membranes (58). We are not aware of any direct experimental measurements of membrane area per lipid in the presence of buffers, but, in general, decreases in bending energy are consistent with decreased membrane thickness and increased area per lipid.

In summary, the change in calcium and buffer concentrations that occur at the beginning of our experiment may explain some of the approximately 20% initial increase in membrane area that we observe. However, the repeated cycles of expansion and relaxation shown in Fig. 4 occur under constant chemical conditions and show that a part of the expansion is due solely to the mechanical effect of shear stress. In future experiments, we plan to investigate the contributions of the flow solution components individually to isolate their effects.

Surface coupling and flow responses

Supported bilayer experiments are notoriously sensitive to the methods used to prepare the membrane substrate. Blachon et al. showed that surface roughness has a strong impact on lipid diffusion constants in supported membranes, suggesting that high surface curvature creates nanoscale gel-like defects in fluid-phase membranes (59). Our observation of flow-mediated expansion and contraction required careful preparation of the glass coverslip surfaces. On flat, smooth glass, we found that the amount of initial expansion is independent of patch size within the range we observe (patches between 10 and 100 μm in diameter; Fig. 5 D). This indicates that the force causing expansion is independent of the membrane area (unlike the total shear force applied to the upper leaflet of the membrane, which is proportional to the area (14)). In addition, this suggests that, on our glass surfaces, the resistance to expansion caused by pinning or adhesion of lipids is small and unable to limit membrane movement. In contrast, patches on roughened PDMS surfaces underwent more increase in their membrane area than could be accommodated by spreading across the surface. On PDMS substrates, excess area created by flow was instead accommodated by bending the membrane and forming tubules. In this case, we did not observe contraction of the membrane area after flow stopped, or a second expansion when flow started again.

The different methods used to form GUV-derived patches and continuous bilayers formed by SUV fusion are likely to result in membranes with different numbers of sub-microscopic defects, which may have large effects on the mobility of lipids in the membranes overall. The very different responses we observed on glass and PDMS also suggest that lipid mobility at the membrane edge is an important factor in the membrane flow response. Lipid mobility and conformation are likely to be altered near membrane edges. For example, Heath et al. used bilayer stripes printed on mica surfaces with an AFM tip to obtain evidence that lipid mobility is sharply reduced within 50 nm of supported membrane edges and that membrane height is reduced suggesting possible lipid tilt in these regions (60).

In summary, we cannot exclude either membrane undulations or changes in area per lipid as an explanation for the expansion that we observe, and it is possible that both occur simultaneously. Changes to the buffer chemistry can only account for part of the expansion.

Biological impacts of shear-mediated membrane expansion

The flow-mediated area response that we observe here may affect the results of experiments that use supported bilayers as a platform to study processes such as lipid or protein mobility, membrane fusion, or membrane protein function; all of these might be affected by a temporary change in the area per lipid or the membrane tension. It is possible that this flow response is confined to single-component membranes supported on solid substrates. We plan to extend our experiments to include lipid mixtures and to explore whether tethered bilayers or those supported on soft substrates exhibit similar changes to lipid area. If so, then this passive shear response may contribute to the mechanical changes observed in the plasma membranes of living cells under shear stress.

Although we cannot yet completely explain the mechanism underlying flow-mediated expansion, our observations may help explain previously observed flow responses in biological membranes. The shear stress magnitudes discussed in this experiment encompass a range of shear stresses observed in biological processes, including shear stress from blood flow on endothelial cells, up to 2–4 Pa (61), the fluid shear stress on swimming microorganisms (62), and stresses due to interstitial fluid flow caused by deformation of bones (0.8–3 Pa) (63). We do not directly observe membrane order in our experiments. However, the timescale and partial reversibility of the dramatic change in membrane area that we observe correlate well with the changes to apparent membrane viscosity reported by experiments on living endothelial cells (26,27,28). As in our experiments, the changes observed in cells could result from flattening of membrane undulations, changes to area per lipid in the membrane, or a combination of the two. In any case, changes to global membrane properties such as tension and viscosity are likely to be more complex and inhomogeneous in cell plasma membranes than in solid-supported membranes.

The broader relevance of this finding could be limited by the fact that expansion and contraction were only observed in very specific circumstances: membrane patches formed on very flat and hydrophilic glass. However, the parallel responses observed in continuous bilayers and flow-mediated tubulation in membrane patches on rough substrates suggest that the flow response may be general, even when it is difficult to observe. It remains to be determined whether membranes on softer and more uneven substrates, including cell plasma membranes, exhibit similar flow responses. Continued investigations using solid-supported membranes with minimally complex lipid compositions and coupling to the substrate will help clarify the origins of any lipid membrane flow response.

Author contributions

E.J.M., M.D.P., and A.R.H.-S. designed the research. E.J.M., M.D.P., J.S., and A.R.H.-S. carried out the experiments and analyzed the data. E.J.M., M.D.P., and A.R.H.-S. wrote the article.

Editor: Michael F. Brown.