A Molecularly Complete Planar Bacterial Outer Membrane Platform

The bacterial outer membrane (OM) is a barrier containing membrane proteins and liposaccharides that fulfill crucial functions for Gram-negative bacteria. With the advent of drug-resistant bacteria, it is necessary to understand the functional role of this membrane and its constituents to enable novel drug designs. Here we report a simple method to form an OM-like supported bilayer (OM-SB), which incorporates native lipids and membrane proteins of gram-negative bacteria from outer membrane vesicles (OMVs). We characterize the formation of OM-SBs using quartz crystal microbalance with dissipation (QCM-D) and fluorescence microscopy. We show that the orientation of proteins in the OM-SB matches the native bacterial membrane, preserving the characteristic asymmetry of these membranes. As a demonstration of the utility of the OM-SB platform, we quantitatively measure antibiotic interactions between OM-SBs and polymyxin B, a cationic peptide used to treat Gram-negative infections. This data enriches understanding of the antibacterial mechanism of polymyxin B, including disruption kinetics and changes in membrane mechanical properties. Combining OM-SBs with microfluidics will enable higher throughput screening of antibiotics. With a broader view, we envision that a molecularly complete membrane-scaffold could be useful for cell-free applications employing engineered membrane proteins in bacterial membranes for myriad technological purposes.


QCM-D modeling to detect adsorbed mass on the quartz sensor. QCM-D is a well-known
technique to detect the adsorbed mass, ∆ , on the crystal surface 1 . Several models have been developed to investigate mass and viscoelastic properties of the adsorbed film by fitting resonant frequency and dissipation signals: If the adsorbed film is rigid enough, which indicates that ΔD is smaller than 1x10 -6 Hz, the Sauerbrey equation can be applied to obtain the adhered mass, ∆ : ∆ is the adsorbed mass on the crystal surface, is a constant ( -17.7 2 • for crystal with f = 5 MHz), and ∆ is the shift of frequency at z overtone (z = 1, 3, 5 , 7, 9, 11, 13). The Sauerbrey equation describes the linear relationship between the adsorbed mass and the change of resonant frequency for rigid adhered layers. Where , are the density of the crystal and film, ℎ , ℎ 1 are the thickness of the crystal and film, is the ratio of the storage modulus and the loss modulus, and is the penetration depth. tan( ) represents the ratio between the viscosity and shear modulus, which reflects the viscoelasticity of material. The smaller value of tan( ) indicates a more rigid material attached on the crystal, and vice versa.
We used the Voigt-Voinova model built in the commercial software, Q-tool, to obtain the adsorbed mass on the sensor since most of the data shown in this study contained the shift of dissipations (ΔD) much greater than 1 × 10 −6 . Two overtones of frequency shifts and dissipation changes, (∆ 3 , ∆ 3 , ∆ 5 , ∆ 5 ), were fitted to the Voigt-Voinova model to generate the information of thickness (ℎ 1 ), shear elasticity and viscosity. The parameters used in the software were: film density ( ): 1100 kg/m 3 , fluid density: 1000 kg/m 3 , fluid viscosity: 0.001 kg m -1 s -1 .
The mass of the adsorbed film is the product of the film thickness and density: • Two-layer Voigt-Voinova model 2 If the adhered layer is predominately heterogeneous, a two-layer Voigt-Voinova model can be applied to capture the detailed viscoelastic properties of different layered films. For two thin viscoelastic adhered layers in a bulk fluid, the changes in frequency and dissipation at various overtones are: By fitting frequency shifts and dissipation changes at various overtones (∆ 3 , ∆ 3 … . ∆ 13 , ∆ 13 ) to two-layer Voigt-Voinova model, viscoelastic properties ( 1 , 2 , 1 , 2 ) and film thickness (ℎ 1 , ℎ 2 ) of both top and bottom layers can be extracted. The fitting process was performed using an optimization tool of a commercial software package (MATLAB 8.3, The MathWorks Inc., Natick, MA, 2014a). Matlab's fmincon function was performed to find the minimum of the following constrained nonlinear multivariable function: Due to the nonlinear nature of the F function, various sets of fitted parameters were found, indicating the existence of multiple local minimums. To determine the most appropriate solution, we constrained the range of the viscoelastic parameters and film thickness to conform to physical constraints: 8.9 × 10 −4 • ≤ ≤ 1 × 10 −1 • , 10 4 ≤ ≤ 5 × 10 6 , 1 −9 ≤ ≤ 1 −7 . Within the range specified, one specific set of fitted parameter was determined by the fitting tool to reach a global minimum. The density for both of the films was assumed to be 1100 kg/m 3 .
Preparation of lipid vesicles. The lipids used in this study were DOPC (1,2-dioleoyl-sn-glycero- of 45 ml 50% wt H2O2 (Sigma) and 105 ml H2SO4 (BDH chemicals). After cleaning, glass coverslips were rinsed with copious amounts of deionized water for 30 min. Deionized water was generated by an Ultrapure water system (Siemens Purelab). Clean glass coverslips were stored in deionized water and dried with nitrogen gas before each use.
Polydimethylsiloxane (PDMS) well fabrication. PDMS monomer and crosslinker were mixed in a ratio of 10:1. After being stirred and degased, the mixture was poured in a Petri dish and baked at 85°C overnight. The thin sheet of PDMS was then cut into small pieces such that they fit over the glass coverslips. Each piece had a hole punched at the center of diameter ~ 1 cm. The PDMS piece was attached on a clean glass coverslip to form a well to hold various solutions used here.
Formation of SLB from pure liposomes. SLBs self-assemble on clean glass by the vesicle fusion method 5 . Liposome solutions were added to a PDMS well and incubated for 10-15 min. PBS buffer was used to rinse the samples after incubation to remove excess lipid vesicles. In order to visualize the SLB formation and measure the mobility of lipids, liposomes (or OMVs) were labeled with R18 prior to bilayer formation, using the labeling procedure described above. R18 was the fluorescence probe for all FRAP measurements conducted in this study. The SLB was photobleached with a ~10 μm diameter spot under 40x objective by the laser for 3 seconds.
The fluorescence intensity of the bleached spot as it recovers with time was recorded for 15 minutes. To reduce artifacts resulting from background photobleaching, the fluorescence intensity of the spot was determined after background subtraction and normalization. The recovery data was then fit by following the method of Soumpasis 6 . The equation used to calculate the diffusivity is, , where w is the full width at half-maximum of the Gaussian profile of the focused laser D = w 2 4t 1/ 2 beam and t1/2 is the characteristic diffusion time.

Proteinase K susceptibility assays for the determination of ClyA-GFP orientation in OM-LB
100 μg/mL of proteinase K (Ambion) for probing accessibility to the GFP domain of the ClyA-GFP protein) was added to ~120 μL samples of OM-SB (created from pure DOPC liposomes).
Images of several regions of the surface were recorded before and after incubating proteinase K with OM-SB for 30 minutes. Particles of GFP were counted and the density of the spots were calculated, described in the Results and Discussion section.

S2. OMV size and surface charge characterization
The size and zeta potential of OMVs and pegylated liposomes in PBS buffer were measured by dynamic light scattering and electrophoresis. The size distributions of OMVs and DOPC with 0.5 mol% PEG (5k)-PE liposomes are plotted in Figure S1a, which shows that both vesicles are similar sizes with average hydrodynamic diameters around 70-100 nm. The zeta potential results ( Fig. S1b) show that OMVs are more negatively charged than DOPC with 0.5 mol% PEG (5K)-PE liposomes in PBS. The negative charge of OMVs results from negatively-charged lipopolysaccharides, which makes up approximately 30 wt% of bacterial outer membrane, as well as protein content. We also used a ZetaSizerNano (Malvern) to determine concentration of the native OMV samples. Duplicate runs (n>3) were performed on each sample and three samples were measured. We found the correlation of OMV protein content (BCA assay) and particle number to be 1.21 × 10 9 ± 2.6 × 10 8 particles/ mg proteins. Figure S1c shows an example of concentration measurement using an OMV sample with 1.6 mg/mL protein concentration (BCA assay), with the average particle concentration to be 1.58 × 10 9 ± 1.27 × 10 7 particles /mL.

S3. OMV adsorption measurement using QCM-D
To estimate the surface coverage of OMVs, we formed a saturated monolayer of intact OMVs on the sensor and monitored the adsorption process using QCM-D. The QCM-D measurement and the corresponding mass curve are shown in Figure S2. Figure S2. Typical result of a QCMD OMV adsorption experiment. While flowing, OMVs gradually adsorbed on the sensor, indicated by declined frequency and increased dissipation, and formed an unruptured OMV monolayer. The signals of frequency and dissipation were converted to mass curve (b) by fitting with one-film Voigt-Voinova model.

S4. OMV surface coverage discussion
The surface coverage calculation performed using eq. (1) in the main text was based on one simple assumption that the acoustic response obtained from QCM-D is proportional to the biomolecule mass. Note that acoustic response from QCM-D ( ) is the combination of both dry biomolecule mass ( ) and the coupled water mass ( ): By assuming is proportional to , we are stating that the hydration level, H, remains constant regardless of molecular level of OMV: level at all times, we will overestimate the biomolecular mass at low surface coverage, which will effectively distort the OMV rupture percentage calculation.
To further correct the artifacts caused by the assumption (eq (10)), we applied the theoretical model developed by Bingen et al (2008) 8 to describe the solvation of OMV at different surface coverage. Bingen and colleagues developed pyramid models to simulate the adsorption of vesicles on a QCM sensor. By assuming the space occupied by a vesicle with the solvent coupled to it to be a truncated pyramid, they were able to calculate the hydration level (H) as a function of surface coverage and parameters that describe the adsorbed vesicles: is the molecular weight of the vesicle,

I. Geometry
In the QCM-D system, the SiO2 sensor (inner diameter 11 mm) was mounted in the liquid cell chamber (height 0.4 mm), and the chamber volume is around 40 . The geometry of the system is shown in Figure S3.

II. Physical Models
Three physical models were chosen in this study: Laminar flow (3D), transport of diluted species (3D), and surface reaction (2D). First, laminar flow physics was solved with stationary study, and the resulting 3D velocity profiles were input to transport of diluted species and surface reaction physics. The transport and surface reaction physics were solved in a time-dependent manner.

Physics 1: Laminar Fluid Flow
The flow in the chamber is laminar flow (Re ~ O(1)) and governed by Navier-Stokes equation: is solution velocity � �.

Physics 2: Mass Transport in the stream
The governing equation of the transport of free OMVs (V) in the solution is described as below: where is the diffusivity ( Other impenetrable surfaces: • (− ∇ + ) = 0. The initial condition was = 0, which means that the concentration of the bulk at the beginning of the process was set to zero.

Physics 3: Surface Reaction
The governing equation for OMVs adsorbed on the surface can be expressed as below, including OMV surface diffusion and the adsorption reaction: where is the concentration of the adsorbed OMVs, is the surface diffusivity and is OMV adsorption rate. If the surface diffusion was assumed to be zero, the governing equation for the surface concentration can be written as: Free OMVs (V) in the solution could adsorb on the surface site (S) on the QCM-D sensor irreversibly: Where VS represents the OMVs adsorbed on the surface. The rate of adsorption can further be defined as: � 3 � is the concentration of free OMVs in the solution on the surface, Γ s is the total active binding site for OMVs (mg/m 2 ), and is the surface fraction of . Since both Γ are unknowns in this study, we further combine these two parameters to one unknown constant A: = Γ By integrating eq. (16) and (20), the material balance for the adsorbed OMVs on the surface can be written as:

S4.1.2 COMSOL Simulation results and discussion
The constraint for the simulation model is that simulated surface coverage should saturate at the same time point as the experimental surface coverage, which is around t = 1600 sec under the given flow rate and solution concentration. By tuning the unknown parameter A in eq. 12, we are able to determine the surface coverage profile to reach saturation at the desired time. As shown in Figure S5, the experimental surface coverage data calculated based on the assumption (eq.(2)), θH, is higher than the simulated result, θCOMSOL. This finding corresponds well with the proposed mechanism from the literature: the biomolecular mass will be overestimated at low surface coverage if hydration is considered to be constant. Fig. S4 also depicts the deviation of these two coverage profiles over the time course.
By expressing eq. 11 as eq. (22): we can then fit the relationship of θH and θCOMSOL to the "uniform adsorption" pyramid model (model 2 in Bingen et al) to determine the parameters r and l. The best fits are shown in Figure   S5a and r and l are determined as 35 nm and 24.6 nm respectively. Finally, we fitted r and l to simulate the adsorption of OMVs by a random sequential adsorption (RSA) following a Monte Carlo algorithm 9 . The relationship between surface coverage θmodel, RSA and θH can then be computed (Fig. S5b) and used to correct the artifacts caused by the assumption (eq. (10)).

S5. QCM-D responses of all the overtones for OM-SB formation and the following polymyxin B interaction
We performed QCM-D experiments to monitor the interaction of polymyxin B towards OM-SB, ranging from the 3 rd to 13 th overtones. The QCM-D measurements including OM-SB formation and the following polymyxin B interaction are shown in Figure S6.

S6. The QCM-D control experiment of adding polymyxin B to PEG-SLB
To exclude the possibility that non-specific artifact of polymyxin B to PEG-SLB are present, we performed QCM-D experiments to confirm that no signal was detected with the addition of PMB to PEG-SLB. Figure S7. The QCM-D measurement showing PMB has no effect on PEG-SLB at the giving PMB concentrations. PEG-SLB was first formed (1) and PMB solution was then added into the system (2). The PMB concentrations range from: 0.001mg/ml, 0.01mg/ml, 0.05mg/ml, and 0.1mg/ml; each batch of solution was followed for 30 minutes. (a) Normalized frequency response and (b) the corresponding dissipation shifts.