Lab on a Biomembrane: Rapid prototyping and manipulation of 2D fluidic lipid bilayers circuits

Lipid bilayer membranes are among the most ubiquitous structures in the living world, with intricate structural features and a multitude of biological functions. It is attractive to recreate these structures in the laboratory, as this allows mimicking and studying the properties of biomembranes and their constituents, and to specifically exploit the intrinsic two-dimensional fluidity. Even though diverse strategies for membrane fabrication have been reported, the development of related applications and technologies has been hindered by the unavailability of both versatile and simple methods. Here we report a rapid prototyping technology for two-dimensional fluidic devices, based on in-situ generated circuits of phospholipid films. In this “lab on a molecularly thin membrane”, various chemical and physical operations, such as writing, erasing, functionalization, and molecular transport, can be applied to user-defined regions of a membrane circuit. This concept is an enabling technology for research on molecular membranes and their technological use.


S1. Preparation of surfaces
Glass surfaces. Circular microscope cover glasses #1.5 (Menzel-Gläser, 47mm diameter) were obtained from Thermo Scientific (Sweden). Before use, the glass surfaces were cleaned in the MC2 process laboratory at Chalmers University of Technology (class 100 semiconductor clean room), using the processing facilities, chemicals and materials provided there. First, the slides were immersed in freshly prepared piranha solution (3:1 v/v mixture of concentrated H 2 SO 4 and 30% H 2 O 2 , heated to 100-110 °C) for 10 min, followed by rinsing with deionised water and blow drying with nitrogen. Thereafter the glass slices were mounted to a Willco Wells® dish frame using a dedicated double sided tape and assembly kit (Willco Wells B. V., Amsterdam, Netherlands) and stored in a sealed plastic bag until use. Critical: smooth and homogeneous surfaces are essential for the function of the 2D fluidic network. For example, sputtered silicon dioxide surfaces can be readily covered with lipids, but the forming films are not diffusively conductive and do not support transport of material.
SU-8 surfaces. The cleaned cover glasses were coated with ~2 µm high SU-8 patterns using the procedure provided by Microchem Corporation. SU-8 2002 (Microchem Corp, Massachusetts, USA) was spin-coated at 3000 rpm for 30 s, followed by soft baking for 2 min at 95 °C on a hot-plate. The SU-8 film was exposed with a dose of 120 mJ/cm 2 on a Karl-Süss contact mask aligner MA6 (G-line, 5-6 mW/cm 2 ), using the "Low-Vac" mode with a bright-field chromium mask. The substrates were then post-exposure-baked for 2 min at 95 °C on a hot-plate. Thereafter, the SU-8 was developed in SU-8 Developer (Microchem) for 1 min using two sequential bathes, rinsed by spraying with clean developer, and blow dried with nitrogen, yielding a SU-8 coated cover glass where the channels are formed by the exposed glass. The surfaces were plasma cleaned briefly in a Plasma Therm BatchTop RIE (50 W, 250 mTorr, 1 min) plasma chamber, and hard baked for 10 min at 200 °C on a hot plate with slow heating and cooling to prevent crack formation. The so prepared glass slides were mounted to dish frames like the plain glass slides described in the previous section.

Vesicle preparation
The compositions of the six lipid mixtures used in the experiment are summarized in the table S1.

S3. Multifunctional pipette
Fig S1. Layout of the multifunctional pipette. Schematics and control circuitry: (a) multifunctional pipette layout and external connections, and (b) magnification of the flow switch and pipette tip. Pressures used for operation: V recirc =V switch =-100 mbar, P standby =20 mbar, P on =200 mbar. Table S2. Essential parameters of the four-solution pipette, and of its operation.

S5. Control algorithm
The algorithm used to control the writing and subsequent processing of 2D fluidic networks is illustrated in figure S3. The translation of the motorized stage, i.e., the substrate surface, occurred step-wise from one point to the next with step length , where deposition time was allowed in each point to process the surface. In order to achieve a diffusively connected lane, the step length is chosen to be smaller than the spot size ( < ), where is the diameter of the hydrodynamically confined flow (HCF) volume. If the step length is decreased, the edges of the lane become accordingly smoother. To move between unconnected parts of a network, i.e., to translate without writing, the outflow is switched off ( = 2 -5 s). Different parts of a network can thus be created or processed with different solutions. There are up to four different solutions possible with the current system, but the number of individual solutions can be greater, depending on the pipette design.

S6. Protocols
The protocols which were used in the experiments are listed in table S3. Table S3. Experimental protocols.

Protocol
Experimental description P1 Deposition of a POPC spot. The pipette was positioned and the flow of lipid vesicle suspension (POPC-488) was switched-on. Density and size evolution of the spot were analysed.

P2
Deposition kinetics and density of the POPC spot. The pipette was positioned and the flow of lipid vesicle suspension (POPC-655) was switched-on. Kinetics of the density change was analyzed.

P3
Deposition of a DOTAP spot. The pipette was positioned and the flow of lipid vesicle suspension (DOTAP-655) was switched on. Density and size evolution of the deposition spot were analysed.

P4
On-surface dilution series of POPC with different labels. The pipette was loaded with two different types of lipid vesicles (POPC-488 and POPC-655). Dilutions were created using the pulse width modulation (PWM) principle on the surface. For this, the deposition was multiplexed between the two sources, where the PWM period was 10 s. Each spot was deposited for 30 s, thus each spot deposition contained three periods. In total 11 spots were created with the duty cycle changing in 1 second steps (e.g. 10:0, 9:1, 8:2, 0:10 seconds). The spots with a diameter of 100 µm were created as a 3 x 4 grid with a pitch of 300 µm. The stage movement between the spots was controlled by the protocol software.

P5
Fusion of DOTAP into a POPC spot. The pipette was loaded with two types of lipid vesicles (POPC-488 and DOTAP-655). Thereafter the pipette was positioned, and the flow of POPC-488 vesicle suspension was switched on for 60s to deposit a POPC spot. Thereafter, the flow was switched off and the pressure conditions were set such that the outflow rate would be about half the previous rate, to ensure that the size of the hydrodynamically confined flow (HCF) volume is reduced and the DOTAP is deposited within the boundaries of the already existing POPC film. Thereafter DOTAP-655 deposition was switched-on.

P6
Overlapping DOTAP-on-POPC spot. This experiment was performed in a manner similar to the previous one, except that the flow-conditions were not adjusted between the different lipid spots. However, the pipette position was moved slightly by ~30 µm to create two partly overlapping films.

P7
Deposition of POPC onto a DOTAP spot. This experiment was performed similarly to P5, but the deposition sequence was exchanged. First DOTAP-655 was deposited for 60s, followed by POPC-488 for 90s. Since the area of the initially deposited DOTAP-655 film grows due to spreading, there was no need to reduce the size of the HCF volume between different lipid types.

P8
Overlapping POPC-on-DOTAP spot. This experiment was performed similarly to P6, but the deposition sequence was exchanged, such that a DOTAP-655 spot was deposited first. Subsequently, after the pipette was positioned to the edge of the spot, POPC-488 was deposited. Total writing time was ~15 min, followed by 15 min time for diffusion.

P12
Writing a two-component network 2 (cross). This experiment was performed similarly to P11, using a different cross geometry, composed of two lines, a vertical (POPC-488) and a horizontal (POPC-655) one.

P13
Diffusion of two dyes in a POPC film 1. The pipette was loaded with two types of lipid vesicles (POPC-488 and POPC-655) and positioned. Two 300µm long and 100 µm wide lanes were written ( = 25 μ and = 10 ).
First POPC-488 was written, followed by POPC-655. The lanes were offset by 50 µm and were overlapping partially. After writing, the diffusion was monitored for 15 min.

P14
Diffusion of two dyes in a POPC film 2. This experiment was performed similarly to P13, but the offset distance between the lanes was increased to 120 µm. The diffusion was monitored for 8 min.

P15
Writing a gradient. This experiment was performed similarly to P4, but instead of depositing individual spots, the lipid vesicles were deposited gradually on a single lane in 80 µm steps, creating a smooth gradient.

P16
Erasing. The pipette was loaded with three solutions: two types of lipid vesicles (POPC-488 and POPC-655) and a solution of dilute Triton-X. Two parallel lanes of 150 µm length were written, using POPC-655. Thereafter the outflow was switched off and the supply pressure was increased from 200 mbar to 250 mbar, since Triton-X is more viscous. Thereafter Triton-X solution was switched on. The HCF volume is easily visible in the transmission channel of the microscope, due to a higher refractive index (Triton-X has a refractive index of n = 1.49 vs. n = 1.33 for water). Triton-X was scanned perpendicularly over the lower lane. Thereafter the outflow was switched off again, was set to its initial value, and the pipette was Functionalization Sequence 2. This experiment was performed similarly to P18, but the 10 min diffusion time after functionalization with goat anti-biotin was omitted, and anti-goat was applied directly after goat anti-biotin. P20 Trap. This experiment was started as described in P7, by depositing POPC vesicles onto a DOTAP film. After POPC vesicles were deposited onto a DOTAP film, the supply pressures were adjusted (V recirc =0 mbar, V switch =-350 mbar, P standby = P on =20 mbar) for maximal inflow ( ≈ 33 / ) through the middle channel. Trap formation and collection of vesicles under the middle channel were monitored.

P21
Flow in patterned 2D channels. The pipette was loaded with two types of lipid vesicles (POPC-488 and DOTAP-655) and positioned onto the patterned area. First, the surface of the patterned channel was covered with POPC-488, thereafter DOTAP-655 was deposited onto the circular supply area, and transport of fluorescently labelled lipid from the supply area into the channel was monitored for about 25 min.
In each experiment the response was recorded as a series of fluorescent confocal microscopy images, collected in up to three different fluorescence emission channels corresponding to the fluorophores used (Table S4), A brightfield transmission image was also recorded.

Supplementary Data
All microscopy images presented in following section are false coloured. The actual collection ranges of emitted light, controlled by the confocal microscope, are given in Table S4 above.  (Table S1). Simulations were performed with COMSOL Multiphysics 4.1, combining models of laminar flow (spf) and transport of dilute species (chds) in stationary 3D models. The viscosity of water was 1 mP*s. The pipette was positioned 10 µm above the surface at an angle of 30°. Due to a sharp chemical boundary at high Pé, an about 10x finer mesh was used for the diffusion calculation compared to the flow calculation. Intensity profiles before and after injection of DOTAP (a). Integrals of total film area and fluorescence in two channels (b). Microscopy images are provided in Fig. 2 of the main text. Already a small scale injection of DOTAP-655 (area increase only about ~15%) causes significant quenching of green (POPC-488) fluorescence (about -60%). However, a local fluorescence increase of the green fluorescence can be observed outside the injection site directly after injection. This can be due to an initial compression of the film. The net decrease of the fluorescence is due to energy transfer from the green (488) to the red (655) dye. We speculate that the dyes are forming a molecular assembly, where energy from many green flurophores is efficiently funnelled to a single red flurophore, since small amounts of Atto-655 dye quench the fluorescence of many Atto-488 dye molecules. Alternatively, it could be possible that the film remains compressed and the amount of actually incorporated DOTAP is higher than predicted by the total area increase of the lipid film. Fig S7. Overlapping DOTAP spot on a POPC spot (Protocol P6). Confocal microscopy images (a), time evolution of the fluorescence profile across the membrane patches (b-d), as well as the integrals of the total film area and fluorescence in both channels (e). DOTAP is readily depositing on the glass surface and fusing into a POPC film. Immediately after DOTAP-655 deposition has started (t=135 s), the red (655) fluorescence increases more rapidly on top of the POPC-488 film near its boundary, while the green (488) fluorescence is decreasing there. This supports the hypothesis of energy transfer between these dyes. The mixing is most efficient directly on the edge, where a sharp spike can be seen. Further evidence for energy transfer is provided by the integrals. After applying DOTAP, mixing between DOTAP-655 and POPC-488 occurs, which results in rapid decrease in green and increase in red florescence. Even after switching off the deposition, the membrane area increases slightly along with increasing red florescence. This can be due to on-going spreading and diffusion. Evaluating the areas, the membrane should be composed of 60% of POPC-488 (green) and 40% of DOTAP-655 (red) lipids. However, mixing with the red dye results in quenching of about 70% of the green fluorescence. The red fluorescence of ATTO-655, collected between 650 -800 nm, is represented by purple false colour. The green fluorescence of ATTO-488, collected between 496 -586 nm, is represented by green false colour.

Fig S8. Deposition of POPC on the DOTAP spot (Protocol P7).
Confocal microscopy images (a), integrals of total film area and fluorescence in the two channels (b), and time evolution of the profile of fluorescence intensities (c-d). These experiments indicate that POPC vesicles are adhering to the DOTAP film, but do not fuse readily, which should result in an area increase, or at least in diffusion and energy transfer. Instead, deposition of POPC results in a very small area increase (<20%), which could be also due to ongoing spreading of DOTAP. There is almost no energy transfer discernible. Note that the step in the red channel between 50 and 150 s during the application of POPC-488 is an artefact caused by mechanical interaction of the pipette and the dish, as well as slight mechanical motion caused by switching. Minor fusion is distinguishable by a slight area increase, as well as an increase in red, and decrease in green fluorescence (<10%). The red fluorescence of ATTO-655, collected between 650 -800 nm, is represented by purple false colour. The green fluorescence of ATTO-488, collected between 496 -586 nm, is represented by green false colour.  profile (b-c). Directly after the deposition of POPC-488 was started (t=0 s), a sharp increase in the green (488) fluorescence channel can be observed on the glass side of the boundary of the lipid films, as POPC vesicles immediately adhere and fuse there. This forming POPC film is diffusively connected to the DOTAP patch. This results in energy transfer and a sharp increase in red (655) fluorescence on the edge of the DOTAP patch (t=13 s). As diffusion progresses, the green florescence is reduced (t=60 s), and eventually both parts of the film are homogeneously mixed (t=150 s). In contrast, a darker rim is visible in the green channel on the lipid side of the boundary, indicating a loss of vesicles after deposition. POPC vesicles do not fuse with the DOTAP membrane, but maintain mobility after adhering. They are rapidly affected by the pipette inflow and removed from the patch perimeter. The accumulation of green vesicles on both sides under the inlet channels are indicators for a hydrodynamic trap, a phenomenon which was further investigated in another experiment (see P20, S12). The red fluorescence of ATTO-655, collected between 650 -800 nm, is represented by purple false colour. The green fluorescence of ATTO-488, collected between 496 -586 nm, is represented by green false colour.  Confocal microscopy image after diffusion (a) and fluorescent intensity profiles across the lanes at different time points (b). The diffusive mixing of two lipid-anchored dyes (488 and 655) can be monitored. However, a closer look reveals that a narrow (about 5 µm) stripe at the edge of the patch is not diffusively connected (inset of panel a, and lower left and right corners in the graph in b). This is explained by the diffusive boundary of the HCF volume, where a 5 µm distance would correspond to about one order of magnitude in concentration drop (Fig. S5). While this lower vesicle concentration is sufficient to create a visible coverage with nanoscopic lipid patches, the density is too low to form a connected and cohesive film which would support diffusive transport. The red fluorescence of ATTO-655, collected between 650 -800 nm, is represented by purple false colour. The green fluorescence of ATTO-488, collected between 496 -586 nm, is represented by green false colour.

Fig S12.
Integrated fluorescence intensities and normalized lipid film areas. (a) 50 µm offset between two lanes, in which case the film is composed of 80% of green lipid and 20% of red lipid. (b) 120 µm offset between two lanes, in which case the film is composed of 55% of green lipid and 45% of red lipid. During mixing there is no further deposition of lipids and the area is preserved (no spreading). However, the integrated intensities are changing such that the intensity of the green fluorescence drops while that of the red increases. This can be explained by energy transfer from the green absorbing (488 nm) to the red absorbing (655 nm) dye. The red fluorescence of ATTO-655, collected between 650 -800 nm, is represented by purple false colour. The green fluorescence of ATTO-488, collected between 496 -586 nm, is represented by green false colour. Correlation of time required in simulation and experiment for the same dye distribution (e). In case of a single dye, where there is no possibility for energy transfer, the integral of fluorescence intensity is well preserved with small drifts of <2.5% (std), which are caused by a combination of the stability of the microscope focus, laser intensity, detector gain and photobleaching, which is most significant in case of the low wavelength (488 nm) dye. This stability for a single dye clearly suggests that the observed large changes in other cases are due to the interactions between fluorophores. This allows us to assume that fluorescence intensity is directly proportional to fluorophore concentration and can be used to characterize diffusion. The diffusion constant was estimated using a numerical analysis. The lane was considered as a one dimensional system and a respective model was constructed in COMSOL. The initial concentration distribution was taken from the experiment and a transient simulation was used with An abrupt change of the fluorescent intensity is due to cross-talk between channels. If the channel cross-talk is subtracted, the fluorescence intensity ratio was found to increase ~41%. The average intensity ratio of the secondary antibody remains constant, indicating no diffusion. Even though diffusing, the intensity ratio of the lipid remains constant due to the symmetrical deposition of the lane. (b) In a second experiment, the diffusion time between primary and secondary antibody was omitted. In this case the primary antibody was bound immediately, and no significant diffusion was observed. During this time, the intensity ratio decreased only by 16%.

S12. Trapping
Fig S16. Hydrodynamic trap simulation to estimate the trapping force. The simulation cell (a) and the hydrodynamically generated stress profile on the surface (b). In this simulation, the pipette was positioned 10 µm above the surface at a 30° angle. A flow rate of ~ 33 nl/s was applied. For 20 to 50 nm vesicles, the estimated trapping force is in the range of 0.5 to 5 fN, at a maximum shear stress of ~0.5 Pa. The trapping force scales ∝ and ∝ , where Q is the flow rate and d is the vesicle/particle size.

S13. Auxiliary diffusion analysis of the lipid films using FRAP
In order to determine the transport properties of different membranes and their dependence on basic fabrication parameters, and deposition time, a diffusion analysis using fluorescence recovery after photobleaching (FRAP) was employed. FRAP experiments were performed on the Leica confocal microscope described in the S4, using the FRAP module in the Leica confocal software, a 40x 1.25 oil objective and 488 nm excitation light. In all experiments, ATTO-488 labelled lipids were used. The imaging frame rate was 0.51 fps over 187.5 µm x 187.5 µm area. Firstly, 10 pre-bleach frames were collected, followed by 4 bleaching frames, during which the scanning area was reduced to 20 µm x 20 µm and laser power was increased 5-folds using the AOTF, which results in about a 440x increase in light intensity during bleaching compared to imaging. Directly after bleaching the fluorescence intensity had dropped to about 10% of its initial value. The power was then reduced, the observation area increased again, and 50 frames were collected in order to follow the fluorescence intensity recovery. In order to determine the diffusion properties, the intensity development was compared with a model (Fig S17). The normalized experimental recovery was calculated using the following formula It is zero at = 0, and becomes 1, after complete recovery. However, the lipid film can also contain none-diffusive parts. The diffusive part or mobile fraction is denoted with Fig S17. FRAP model. The model to fit the collected FRAP data was developed empirically from a finite element model in COMSOL, using 2D time dependent diffusion. The lipid patch was modelled as a 90 µm x 90 µm square area with closed boundaries, where the bleached area (20 µm x 20 µm) was placed in the canter. Initial concentrations inside and outside the bleached area were 0 and 1, and the diffusion constant was assumed to be = 1 μ / . The resulting time series of the concentration profiles is shown in (a). In order to accommodate the reduced effect of bleaching and other sources of general intensity variations during the recovery time, the normalized intensity (intensity in the bleached area relative to the intensity in the defined reference area) was used instead of the directly measured intensity (b). The data from this model was fitted to a twocomponent exponential decay, giving an excellent fit (c):

S14. Membrane deposition dependence on operation parameters of the pipette
The following studies show, how the deposition process and spot geometry are depending on the operation parameters of the multifunctional pipette. Increase and decrease of the flow rates about 2 times had only a weak effect on the shape and deposition kinetics of the spot. On the other hand, we observed that changing the ouflow/inflow (OI) ratio changes the spot geometry from circular to elliptical ( Figure S19) and reduces the achievable feature size quite considerably. Similarly, we studied the dependence on the height of the pipette ( Figure S20). Both studies suggest that the deposition process is generally not very sensitive to small deviations in operation parameters, which indicated the overall robustness of liquid handing and dispensing of lipid material. The OI ratio dependence can potentially be utilized to purposefully achieve a reduced feature size of the deposits. OI is recommended to be 0.5 or less, to ensure contamination free operation. Initially the spot geometry is only weakly dependent on the OI, while later (OI<0.2) the spot is narrowing into an ellipse with its short axis perpendicular to the channel outlet array. The long axis of the elliptical spot (along the channel outlet array) is only weakly affected.

Fig S20.
Deposition quality dependence on the pipette height from the surface. The series of micrographs shows lipid patches deposited at different pipette heights (a), a chart of the deposition kinetics for each height (b), and the spot geometry dependence on height (c). In all cases the pipette is operated in its regular mode (OI = 0.3). The series is starting from physical contact of the pipette tip with the substrate (0 µm) and continues in 25 µm steps to the final height of 100 µm. Vertical positioning has been carried out by a water-hydraulic micromanipulator. The deposition is initially only weakly dependent on the pipette height, which indicates good mechanical stability and robustness of the deposition process.