Porcine brain lipids (PC, PE, PS, and SM) and ovine cholesterol were purchased from Avanti Polar Lipids (Alabaster, AL, USA) and bovine MBP from Sigma-Aldrich (St. Louis, MO, USA).

Porcine brain lipids (PC, PE, PS, and SM) with chain compositions given on the Avanti webpage¹ and ovine cholesterol have been separately dissolved in chloroform, or bought in chloroform. The lipids are mixed in ratios found in native and EAE modified myelin (Table 1; Krugmann et al., 2020b) and afterwards the chloroform is evaporated using a gentle nitrogen stream followed by vacuum annealing at 50°C overnight. The lipids are then dissolved in D₂O-buffer [150 mM NaCl, 10 mM 3-(N-morpholino) propanesulfonic acid (MOPS)]. The solution is shaken until the lipid cake is no longer stuck at the glass surface. If necessary remaining lipid cake is removed by pipetting. In the following the solution is sonicated for 30 min at 40°C and freeze-thawed 5 consecutive times. To remove remaining large aggregates or giant vesicles the solution is centrifuge-filtered through a 0.45 μm for 10 min at 10,000 g-force. Finally, the solution is extruded through a 100 nm membrane 21 × at 50°C.

Polished silicon blocks were cleaned by subsequent immersion in chloroform, acetone, ethanol and water under sonication for 20 min each, followed by UV/ozone irradiation for 30 min and rinsing in H₂O to verify hydrophilicity. All silicon substrates were stored in water until utilization to prevent contamination. For the neutron reflectometry measurements, each substrate was sealed in a solid/liquid cell. In detail, the bottom part of the cell is equipped with a PEEK through, which represents the liquid reservoir, while the upper part is an aluminum/Teflon lid. The silicon block is sandwiched between the two parts and the trough (reservoir) filled with Milli-Q water by injection through Teflon tubes (0.5 mm inner diameter), closed by valves. Prior to liposomes solution injection, the cell was pre-heated at 50°C to promote vesicle fusion. When the liposomes were in contact with the hydrophilic SiO surface, a supported membrane bilayer (SMB) was formed and residual liposomes were removed by rinsing with buffer solution.

Neutron reflectivity measurements were performed on the neutron reflectometer D17 at the Institut Laue-Langevin (ILL, Grenoble, France) (Saerbeck et al., 2018; Krugmann et al., 2020a) operated in time-of-flight mode. The instrumental resolution ΔQ/Q was varying between 2 and 10% along the full q-range (0.006 0.35 A⁻¹) with accessible wavelengths from 2 to 30 Å and two angles of incidence (0.8 and 3.2°).

The footprint length (in the beam direction) and width (in the perpendicular direction) were L = 60 mm and W = 35 mm, respectively.

The samples were deposited at the solid/liquid interface of a silicon substrate sealed inside a solid/liquid cells with 1 mL liquid reservoir which was kept at 50°C to promote vesicle fusion; during the rest of the experiment a constant temperature of 37°C was maintained via a thermostatic bath. Contrast variation was carried out via an automatized pumping system at a flow rate of 1 mL/min for a full volume exchange of 20 mL. Firstly, the silicon blocks are characterized in three contrasts (D₂O-buffer, H₂O-buffer, silicon matched water (SMW)-buffer (38% D₂O). Afterwards the respective SMB is formed using the protocol described above. Now, the SMBs are characterized in the same three contrasts. We inject MBP in the cells to coat the SMB as described in Krugmann et al. (2020b). To remove remaining MBP solution the cells are flushed. Again, the membrane is measured at the three contrasts. Finally, the vesicles are injected and the adsorption kinetics is monitored (in D₂O). To enable very fast acquisition the beam divergence is increased and a time resolution of 1 min can be achieved (Cubitt et al., 2015).

For the specular NR, the data first has been reduced using the COSMOS tool in the LAMP software of the ILL. The fitting of those reduced datasets was done using the Anaklasis package based on python (Koutsioubas, 2021). This software allows co-refining multiple datasets as e.g., the same sample at different contrasts or the same sample in D₂O at different steps of the kinetic process. The model assumed is a stratified layer model, while an additional adjustable multiplicative parameter is used for the proper scaling of the reflectivity curves. During data reduction a normalization factor is set accounting for neutron beam attenuation as defined by a direct beam measurement. In the pristine membrane samples only a small deviation to unity is observed which could be explained by a not perfect sample alignment. In the later steps the deviation from unity increases especially in the samples with attached vesicles. This might be accounted to diffuse scattering in these samples. Therefore, less signal is scattered specular and a value smaller than unity needs to be chosen as scaling factor. For the bare membrane we assume the layer order head|chain|head|. This layer sequence will just be called membrane in the following. When adding MBP an additional layer is added and the order is changed to membrane|MBP. After injection of the vesicles the order is changed to membrane|MBP|membrane|diffuse vesicle layer. The diffuse vesicle layer accounts for the 3D structure of the attached vesicles and has a SLD of 0.7 · 10⁻⁶ Å⁻² between the one of the chain and the head section. As front end we use in all cases Si|SiO|buffer and as back end buffer. Obtained membrane thicknesses are compiled in Table 2.

The RSA simulation is modified from Erban and Chapman (2007) for our system. As starting point we take a test volume (x-y-z = 1 μm – 1 μm – 1 mm) with a constant concentration of vesicles in z-direction. In practice a test area (x-y) is defined. The x-y-surface at z = 0 is the adhesive surface. Then we define vesicles in z-direction. Since the concentration should be uniform in x-y-direction we only have to calculate the diffusion of vesicles in z-direction and only initialize the x-y-position in the case that the vesicles touch the surface.

Now we start the diffusion simulation. Therefore, we calculate the diffusion of each vesicle in time steps Δt by

Here, D is the diffusion constant and χ is a normal distributed random number.

In the case that the vesicles diffuse to the adhesive surface and no other adhered vesicle is overlapped in x-y-position they get adsorbed by a certain probability and are removed from the simulation. This is done in the simulation in the case that one of two conditions are fulfilled. The first condition is that the end z-position is negative. If this condition is fulfilled the vesicle adsorbs with the probability:

In Erban and Chapman (2007) P describes a positive constant which can be related to the rate constant of the chemical reaction between the virus surface and diffusing polymers. As the geometry of our system is similar in our case P is related to the interaction of vesicles and the MBP coated membrane surface. The second condition checks if the vesicles z-position was negative during the step. The adsorption probability is given by:

At the beginning only, the vesicle concentration close to the surface gets perturbed by the surface adhesion. Therefore, only vesicle diffusion close to the surface is calculated in the beginning. The affected z-range L(t) increases is given by:

In this model we assumed the vesicles with diameter d = 100 ± 30 nm as spherical hard shells which exclude the area Aex=π4d2 when attached to the membrane. The simulation has been run 10 consecutive times and the results have been averaged. The time step was set to Δt = 1 s. We estimated a value P = 0.015·1/t leading to a quite convincing agreement with the measured data. The maximum coverage of such a system has been determined before (Wang, 1994; Cieśla and Nowak, 2016; Cieśla and Ziff, 2018) to be 55.47% for monodisperse disks—which is a good description of the 2D-projection of the vesicles on the membrane surface. In case of polydispersity this value increases as vesicles with smaller diameter can fit in smaller holes (Adamczyk et al., 1997). We simulated the effect for a polydispersity of 0.3 (Supplementary Figure 1) which we measured for vesicles of the same compositions and almost identical extrusion procedure via SANS and SAXS (Krugmann et al., 2020b). The coverage seems to be almost completely stable over time—at a time range relevant for the incubation time of the vesicles until the static measurements were started (native ~20 h, EAE modified ~10 h). After 1 day we simulated a steric vesicle coverage of 0.60 ± 0.05 (the error was estimated from the incubation time range of the samples) which is used in the further calculations as explained in the following chapter. Here, we used a time step Δt = 10 s in the simulation. The concentration of vesicles was chosen to be 1 mg/ml. From this value the number concentration was calculated by estimating the weight of one vesicle.

The system of interest is a supported membrane bilayer (SMB) coated with MBP which is kept in a liquid cell. Into this cell large unilamellar vesicles (LUV) are injected which can diffuse toward the SMB surface and adhere to it. The main purpose of our study was to investigate the mechanism of that adhesion process. Prior to that we validated the structural properties of the membrane components—an essential prerequisite for the interpretation of the kinetic NR experiments. The different stages of the preparation process are schematically depicted in Figure 2 alongside with the average thicknesses of the membrane components.

As a first step the SMB was created by vesicle fusion on a hydrophilic silicon substrate (Figure 2A; Stroumpoulis et al., 2006; Rondelli et al., 2017). The neutron reflectometry curves were measured on the neutron reflectometer D17 at the ILL (Saerbeck et al., 2018) at three different contrasts having buffer compositions of 100% D₂O, 100% H₂O and a mixture (38%/62% v/v) D₂O/H₂O which matches the SLD of silicon being abbreviated as silicon matched water (SMW) in the following. The NR data are shown for a native and EAE modified SMB in Figures 3A, 4A, respectively. The fit of a simple head—chain—head layer sequence can reproduce the measured NR data nicely (see Figures 3A, 4A). A SMB with native-like lipid composition and two SMBs with EAE modified lipid composition referred to as Native01, Modified01 and Modified1 SMBs were characterized. The obtained thicknesses of the head (d_head) and chain sections (d_chain) of those three SMBs are reported in Table 2 and the respective water penetration values are given in Supplementary Table 1. The respective SLD profiles are shown in Figure 3B of the Native01 membrane and in Figure 4B of the Modified01 SMB.

As a next step MBP was injected in the liquid cell at either a concentration of c_MBP = 0.1 mg/ml for the Native01 and Modified01 SMBs or at 1 mg/ml for the Modified1 SMB. Due to electrostatic interaction, it interacts with the negatively charged membrane and forms a dense protein layer on top of the membrane (see Figure 2B). The data of the Native01 membrane with 0.1 mg/ml MBP in combination with a fit is shown in Figure 3C. The respective data of EAE modified membrane Modified01 with 0.1 mg/ml MBP including theoretical fit is shown in Figure 4C. Those datasets have been fitted as a layer sequence head—chain—head and an additional dense MBP layer on top of the membrane with thickness d_MBP,SBL and concentration c_MBP,SBL. During these fits the fitted values of d_head and d_chain, water penetration and roughness between the layers from the SMB without MBP were used and constrained in ±5% of the best value around those values, while the thickness of the MBP layer d_MBP,SBL (constrained for Modified01: 60–100 Å, otherwise: 70–100 Å) and its water penetration were fitted. The water penetration of the chain section was left free between 0 and 5%. The SLD of MBP was estimated with the primary structure of bovine MBP using the Biomolecular Scattering Length Density Calculator² considering 90% exchange of labile hydrogen atoms. These values are in D₂O ρD2O = 3.45· 10−6A−2, in SMW ρSMW=2.55· 10−6 A−2, and in H₂O ρH2O=2.00·10−6 A−2. The respective SLD profiles of the Native01 and Modified01 SMB are depicted in Figures 3D, 4D. The fit results are reported in Table 2, the corresponding buffer penetration values are given in Supplementary Table 1. We got a convincing fit with a MBP layer with thickness between d_MBP,SBL = 60–80 Å (the determined parameters are characterized by large error bars, due to the dilute nature of the MBP layer) and a protein concentration in the range of c_MBP,SBL = 3–7% vol/vol (corresponding to MBP concentrations of around 22 and 51 mg/ml, respectively, when a protein specific volume of 0.73 ml/mg is considered) for both membrane types and both injected MBP concentrations of 0.1 and 1 mg/ml. The MBP layer thicknesses and concentrations as reported in this study are in agreement with previous observations (Krugmann et al., 2020b).

Vesicles were injected onto the MBP/SMB systems and the interaction process was studied with kinetic NR acquisition. The kinetic data will be discussed in the next section, but beforehand we take a look at the steady state specular NR measurement after an almost constant state is reached after 1–2 h. A schematic visualization of that steady-state membrane system in shown in Figure 2C. Experimental NR data of native and EAE modified conditions under steady-state conditions are shown in Figures 3E, 4E including theoretical fits. We can see a strong change in the reflectivity profiles in comparison to the membranes without adhered vesicles. As a fit model we assume a double-bilayer system because the vesicles diffuse to the membrane and adhere to it. In this case, the specular NR profile should be a double bilayer as specular NR only is sensitive for the z-structure close to the Si-surface since the penetration depth of the neutrons under such small angle is quite low. However, to account for the part of unfused adsorbed vesicles that extends into the solution, we add in our model a long 800 Å diffuse vesicle layer as last layer with a SLD of 0.7 · 10⁻⁶Å⁻² and high roughness and buffer penetration. The thickness of the MBP layer in the double bilayer model has been fitted to d_MBP,DBL ≈ 30 Å which is close to what we have measured before in between 2 bilayers in small-angle X-ray scattering experiments (Krugmann et al., 2020b). The fitted values of d_head and d_chain, water penetration and roughness between the layers from the SMB without MBP were again used and constrained in ±5% of the best value around those values. The thicknesses of the second membrane layer (d_chain, d_head,inner, and d_head,outer) are fixed to the values d_chain and d_head,outer of the first layer. They are fixed to d_head,outer as they are both not attached to the silicon substrate and therefore less ordered. Fitted parameters are the buffer penetration value corresponding to the MBP concentration between the bilayer c_MBP,DBL. The fitted buffer penetration values are given in Supplementary Table 1 and the MBP concentrations in c_MBP,DBL in Table 2. The SLD-profiles we attain form those fits are shown in Figures 3F, 4F.

The fitted structural models of the three stages of the experiment are summarized visually in Figure 2 including the averaged thickness values of the membrane components. Overall, our characterization by NR of the assembly process under steady-state conditions yields a coherent picture of the assembly process. MBP that is injected as a rather dilute concentration interacts strongly with the first SMB and forms a concentrated fluid phase with concentration in the range of 3–7% and a thickness of 60–80 Å. Upon formation of the second membrane layer the MBP layer thickness is reduced to around half of its initial thickness (~30 Å) and a concentration of 0–1% vol/vol for the 0.1 mg/ml samples and 11% for 1 mg/ml sample. This MBP layer is, hence, regulating the assembly and fusion process of the LUV with the SMB. However, due to the complexity of the system in its final state and the dilute nature of the MBP layer, it appears that our modeling is sensitive only to the thickness of the MBP layer, leading to relative underestimation on protein concentration.

Furthermore, from the buffer penetration value h_chain,2 of the second chain section (given in Supplementary Table 1) it is possible to calculate the coverage η of the second bilayer as the hydrophobic chain section is nearly water free in both cases. Therefore, it is possible to calculate the coverage as η = 1 − h_chain,2. The definition of the coverage η can be expressed mathematically as the area of the second bilayer A_dbl divided by the total area A_tot of the first bilayer of the sample

In Table 2 the coverage values of η are given for the investigated membrane systems.

In Figure 5A the 2D off-specular map of an EAE-modified sample in the final steady-state of LUVs bound to the MBP-coated SMB measured at the MARIA reflectometer is given. For technical details of the instrument MARIA, we refer to the instrument paper (Mattauch et al., 2018). In Figure 5B the 2D image of an EAE modified sample of LUVs bound to the MBP-coated SMB measured using the D17 reflectometer is shown. Clearly, an off-specular signal is visible in both cases. This off-specular signal is proof that the samples have strong surface roughness which is an indication for vesicles adhered to its surface (Ott and Kozhevnikov, 2011).

Time-resolved NR experiments were carried out to understand the process of adhesion. In Figure 6A an exemplary kinetic NR measurement of a modified membrane is shown. To maximize the NR signal and to reduce the incoherent background, time-resolved NR experiments have been performed only at one contrast in 100% D₂O buffer. Clearly, in the time range of one to a few hours the adhesion process is happening. The off-specular data is growing in this time as well. The specular curves in Figure 6A can be fitted by the earlier explained double bilayer model. Here, only the coverage (η = 1 − h_chain,2), water penetration of the diffuse vesicle layer its roughness and the scaling factor are fitted, while the rest of the remaining parameter values are fixed to the ones obtained from the static measurements of the membrane with adhered vesicles. The influence of the head section is not that strong as the contrast to D₂O is higher to the chain section. In Figure 6B the respective SLD profiles are depicted. In Figure 6C the normalized coverage (scaled to the RSA simulation curve) from the specular fit, the normalized off-specular signal (scaled to the RSA simulation curve with subtracted background) and the behavior represented by the RSA simulation curve are plotted together. The important observation here is that the normalized experimental coverage values obtained from both specular and off-specular NR data are identical within the statistical uncertainty and their time-dependence can be reproduced by the theoretical model of an RSA simulation.

RSA is a model in which particles diffuse and adhere to a surface and cannot detach from it. The details of the RSA simulation are explained in the Materials and Methods section. Strikingly, the kinetics of the specular data—representing the formation of a double bilayer—and the off-specular data—representing the adhesion of vesicles to the surface—follow the same trend and can both be described by the same RSA simulation. We can conclude here that both signals are governed by the same kinetic time constant of the diffusion driven adhesion of the vesicles. The deformation of the vesicles which is responsible for the double bilayer formation is, therefore, quasi-instantaneous in comparison to the adhesion process. In RSA the vesicles that adsorb to the membrane are described as disks that—when adsorbed—exclude a certain area A_ex. After waiting for an infinite amount of time (t = inf) the coverage of vesicles ρ(t)–which is different from η as not the whole excluded diameter is in contact with the membrane—a coverage ρ(inf) is reached. To verify if RSA is the correct kinetic process to describe our data we plot the logarithm of ρ(inf)-ρ(t) over the logarithm of time in Figure 6D. This should give a slope of −1/2 for an RSA process of disks on a surface, which is indeed the case of our NR data when we exclude the first initial time points (Hinrichsen et al., 1986).