Supplementary Figure 1: Number of hydrophobic lipid tail-water contacts. Contacts were time averaged over 20 ns as a function of the distance of the

Transcription

1 Supplementary Figure 1: Number of hydrophobic lipid tail-water contacts. Contacts were time averaged over 20 ns as a function of the distance of the lipid head group from the bilayer COM. Lipid head groups are colored by the average number of lipid tail-water contacts based on the values in the plot.

2 Supplementary Figure 2: Insertion of larger 1:1 MUS:OT AuNP with a 3.0 nm core diameter. The plot shows both x and H, defined identically as for the 2.0 nm AuNPs, and showing c similar trends. Insertion is initiated upon some small number of hydrophobic contacts triggering the rapid decrease in x. Snapshots again show snorkeling and hydrophobic ligand extension into the bilayer as expected

3 Supplementary Figure 3: Full 500 ns unbiased trajectories for AuNP insertion. The trajectories are colored according to the same 3 regimes of behavior discussed in the main text and drawn in Fig. 3. (a) x, the change in the distance between the AuNP and bilayer center of mass, as a function of time for both 1:1 MUS:OT and all-mus AuNPs. (b) H, the number of atoms in lipid tails that are within 0.5 nm of alkane atoms in the ligand shell, as a function of time for both AuNPs. c

4 Supplementary Figure 4: Additional simulation snapshots of unbiased insertion. Snapshots were taken for the same times as Fig. 3 of the main text for both AuNPs. Three views are included: a top view, a side view with lipid tails removed to illustrate the locations of AuNP ligands, and a side view with the AuNP removed to illustrate lipid tail deformation. Lipid tails are slightly enlarged for the last column of images to make their positions more clear. Water and ions are removed for clarity.

5 Supplementary Figure 5: Characterization of the 2:1 MUS:OT particle batch used in Fig. 4 of the main text. The synthesis of this batch is described in section of Supplementary Methods. (a) Representative TEM image. (b) Analysis of the gold core size distribution from TEM measurements. (c) UV-Vis spectra of the particles acquired in Milli-Q water: the surface plasmon resonance is at 516 nm. (d) NMR of the same particles acquired in CD 3 OD after decomposition with KCN.

6 Supplementary Figure 6: Control experiments on a bare substrate. 1 mg ml -1 of a 2:1 MUS:OT AuNP solution was flushed in the QCM-D chamber after standard PBS (see arrow in (a)) and the frequency shift and dissipation signals were recorded. Subsequently the standard PBS was again flushed in the chamber to remove an air bubble (shaded gray region). No significant absorption is detected. (b) and (c) are AFM images of the silicon wafer and mica surfaces respectively acquired in the presence of 1 mg ml -1 2:1 MUS:OT AuNPs. Images are shown in their full scan size that for both images is 5µm.

7 Supplementary Figure 7: Characterization of the 2:1 MUS:OT particle batch used in Fig. 5 of the main text. The synthesis of this batch is described in section of Supplementary Methods. (a) Electron micrographs of the gold nanoparticle cores. (b) Particle size distribution obtained from TEM analysis: the average core diameter was 5 nm with a 20% relative standard deviation. (c) UV-Vis spectra of the particles at 0.05 mg ml -1 in water and PBS. (d) 1 H-NMR of particles etched with Iodine in MeOD-d4 reveal approximately 33% OT-content, i.e. a 2:1 MUS:OT stoichiometric ratio. (e) 1 H-NMR of nanoparticle solution in MeOD-d4 with 5% D 2 O indicating absence of free ligands: broad peaks around 3 and 1 ppm correspond to nanoparticle-bound ligands (the sharp peaks at 2.0 ppm and 1.3 ppm are impurities in the MeOD-d4 used). (f) Results from dynamic light scattering measurements: representative hydrodynamic size distribution of NPs dispersed in water and PBS.

8 Supplementary Figure 8: Time-sequence of AuNP insertion visualized with AFM. Incompletely formed, defect-rich lipid bilayers (a, b, c) and completely formed, defect-free lipid bilayers (d, e, f) were imaged with the AFM directly in PBS after the addition of 2:1 MUS:OT AuNPs (batch 1 in section of Supplementary Methods). Images (a) and (d) were obtained immediately after the injection of the AuNPs, (b) and (e) after minutes and (c) and (f) after ~90 minutes. All images are shown at the same magnification. The scale bar is 5 µm.

9 Supplementary Figure 9: Interactions of AuNPs with SLBs on a mica substrate in the presence a large defect imaged by AFM. The addition of 2:1 MUS:OT AuNPs to the solution leads to the formation of a topographically higher area surrounding the defect edge (white arrows). This change in topography is consistent with AuNP insertion at the bilayer edge followed by diffusion into the bulk (see also Supplementary Fig. 10). Occasionally, as in this case, it was possible to notice the appearance of AuNP-lipid aggregates within the previous holes of the SLB (blue arrows). These structures were normally quite delicate as simply scanning the same area with the AFM tip several times normally caused the detachment of these agglomerates. Both images are shown at the same magnification. The scale bar is 2.5 µm.

10 Supplementary Figure 10: Diffusion of the 2:1 MUS:OT AuNPs from the edge of the SLB into the bulk. The three images were acquired with the AFM in sequence 105 minutes, 120 minutes, and 180 minutes after the injection of AuNPs. The scan size of the three images is 5 µm. The white lines indicate the edge of the bilayer; the black lines are drawn to help the eyes to define the higher area of the bilayer from where we believe the AuNPs are spreading. The gray lines correspond to the area drawn in the previous image and are meant to help visualize the spreading process. The scale bar is 2.5 µm.

11 Supplementary Figure 11: Interactions of AuNPs with SLBs in the presence of large defects imaged by AFM. This experiment is a replica of the one presented in Fig. 5 of the main paper. The same batch of 2:1 MUS:OT AuNPs (batch 2 in section of Supplementary Methods) used in Fig. 5 was tested on a different DOPC lipid bilayer and with a different AFM. Initially we observe lipid islands in the absence of added AuNPs (a). With the addition of 2:1 MUS:OT AuNPs to the solution we detect the formation of higher areas surrounding the defect edge, more clearly observed in the higher magnification image (b). The scale bars are 4 µm for the left two images and 400 nm for the higher magnification image.

12 Supplementary Figure 12: Schematic illustration of computing the committor. The idealized free energy landscape is represented by the colored contours as a function of two collective variables with the thick black line denoting a transition path across this landscape connecting two stable basins A and B. The value of the committor is determined as the percentage of short trajectories launched from various points along the path that commit to basin B as illustrated by the short arrows. In this system, basin A is identified as a particle at the bilayer surface and basin B is identified as a particle inserted into the bilayer, illustrated by the two snapshots. The bottom plot illustrates that the committor does not have to be monotonic in time by projecting the transition path onto a 1D generalized reaction coordinate that connects the two basins. The unbiased transition path may not pass over the energy barrier in its first attempt, and thus the committor may rise and fall accordingly as a function of time.

13 Supplementary Figure 13: Identification of lipids in hydrophobic contact with either AuNP. Contact is shown as a function of time during the trajectories shown in Fig. 3 and Fig. 6 of the main text. The transition state is identified as a red vertical line. For both particles, the transition state coincides with the onset of the first long-lived hydrophobic contact between the AuNP and a single lipid, which then persists for the duration of the simulation. This lipid is marked in red and is referred to in the text as the anchor lipid. Lipid indices are ordered by time of first contact.

14 Supplementary Figure 14: Additional snapshots of initial lipid contacts with AuNP monolayers. Snapshots are identified as possible members of the ensemble of transition states. Each snapshot was taken from a simulation trajectory generated during committor analysis that proceeded to commit to insertion.

15 Supplementary Figure 15: Hydrophobic solvent-accessible surface area (SASA) of system during insertion. The SASA is decomposed into contributions from AuNP ligands (a) and bilayer (b). Only the AuNP SASA changes appreciably. The SASA for both the 1:1 MUS:OT (c) and all-mus (d) AuNPs decreases significantly after passing the transition state (red dashed line) as the number of hydrophobic contacts simultaneously rises. This gives rise to a hydrophobic driving force for insertion.

16 Supplementary Methods 1. MD simulations 1.1 AuNP parameterization Monolayer-protected gold nanoparticles were modeled using a recently developed parameterization [1] which will be summarized here. The GROMOS 54a7 united atom force field [2] was used to model the two ligand species, MUS and OT (see Fig. 2 of the main text). The topology of the MUS sulfonate end groups was adapted from a previous sulfonate parameterization by Hinner et al [3], with partial charges adapted from ab initio simulations of sulfonate ionic liquids [4]. Following a typical GROMOS building-block approach, standard Lennard-Jones parameters were used for sulfonate atoms with only the charges re-parameterized [5]. The GROMOS 54a7 force field lacks parameters for gold, so gold-hydrocarbon Lennard-Jones parameters were adapted from the Hautman-Klein model for alkanethiol monolayers [6, 7, 8]. All other Lennard-Jones interactions were obtained using the values for gold presented in the UFF force field [9], although in practice primarily gold-hydrocarbon interactions were relevant due to the steric barrier posed by the protecting monolayer. The gold-sulfur-carbon bond angle and gold-sulfur-carbon-carbon dihedral angle at the AuNP interface were left unrestricted as in the Hautman-Klein model, matching assumptions made in similar studies [10, 11]. The gold core itself was represented as a rigid, hollow, perfectly spherical shell maintained by constraints between neighboring gold atoms. The mass of the missing interior gold atoms was redistributed uniformly to the atoms in the hollow shell. The sulfur head groups of the bound ligands were distributed uniformly across the gold surface with a grafting density of 4.62 ligands nm -2, yielding 58 ligands for the 2.0 nm core diameter AuNPs studied here and matching previous studies [12, 13]. The gold and grafted sulfur atoms were not assigned partial charges since previous simulations have shown that the cumulative charge density approaches zero at the gold surface [13]. The sulfur atoms were rigidly bound to the surface, inhibiting any possible diffusion of the grafting points along the gold surface. To form the mixed ligand 1:1 MUS:OT monolayers, the two different ligand species were distributed in a checkerboard like-pattern to create an isotropic surface coating within the constraints of the spherical topology. Each AuNP was initially equilibrated in solution for 120 ns. This equilibration was performed in a 150 mm NaCl salt concentration with a minimum distance of 1.2 nm between any AuNP atom and the box walls. The temperature was set to 310 K and the reference pressure to 1 bar to mimic biological conditions. The temperature was controlled by a velocity rescale thermostat with a coupling time constant of 0.1 ps. The pressure was first equilibrated using a Berendsen barostat for 20 ns then controlled with a Parrinello-Rahman barostat with a time constant of 2.0 ps and a compressibility of bar -1 for the remaining 100 ns. One potential weakness of the AuNP model is the assumption that the polarizability of the gold

17 core is negligible by replacing the gold core with a hollow shell. In principle, treating the gold core as a true conductor could lead to image charge effects that might influence the behavior of the system. While there has been a recent force field parameterized for polarizable gold surfaces [14], to our knowledge no model for polarizable gold nanoparticles exists despite several other simulations of gold nanoparticles in solution [13, 15, 16, 17]. We believe that it is safe to neglect image charge effects for this system for several reasons. First, if the distribution of external charges is spherically symmetric on average, then the induced image charge distribution will also be spherically symmetric. Since the gold core must be charge neutral, the net external field will also be zero following Gauss law using a spherical dividing surface. Second, even if fluctuations in the charge distribution break this spherical symmetry, previous simulations have shown that many-body interactions between metal surfaces and adsorbates can lead to a cancellation of image charge interactions. For example, simulations of water or other small molecules adsorbed on flat gold surfaces have shown a net cancellation of image charge effects even for a single monolayer of water [18, 19, 20, 21]. As the system we study is primarily water, it might be expected that similar cancellations may occur in regions of enhanced charge density. Finally, a brief back-of-the-envelope of the calculation can show that it is unlikely that image charge effects would have an observable effect even if we ignore symmetry arguments entirely and consider the interactions of a region of charge in the system with the gold core. A key feature of the AuNP system is that the charge of the ligand end groups is compensated by both nearby counterions and by water molecules that are oriented by the end group charge. As has been shown in multiple previous simulations, the orientation of these bound water molecules leads to a positive peak in the charge density and potential of the system closer to the surface than the negative peak in the charge density and potential corresponding to the location of the end groups [1, 13]. Thus, we can approximate any particular region of charge as being dipolar in nature with a corresponding image dipole. The work done by the image dipole (and another image charge to maintain charge neutrality) is 3 2 a P given as W = where a is the radius of the gold core, P is the dipole moment, and r is the 6 0r distance of the dipole from the gold center [22]. Assuming a 1.0 nm, r 2.0 nm for the 21 position of charged end groups, and P C nm for the magnitude of the dipole given typical charge densities in a 1 nm 3 volume [1, 13], then W kt, well below thermal energy and safely negligible. Considering all of these arguments, we believe it is reasonable to ignore image charge effects in this model. Another potential weakness of the model is the parameterization of the sulfonate end groups. A recent re-parameterization of the carboxylates in the CHARMM force field has demonstrated that determining partial charges from only ab initio calculations may lead to artificial ion or solvent interactions in solution [23]. Despite these potential issues, the results from this model are reasonable. In particular, previously reported radial distribution functions of system components from this model are very similar to those calculated with other AuNP compositions, including AuNPs with amines or carboxyl functionalized ligands [13]. This structural similarity suggests that the exact nature of the end group parameterization may not induce significant changes to the system if the end group charge is preserved. Furthermore, the

18 GROMOS force field was parameterized with the SPC water model which does not reproduce certain important water properties, such as the dielectric constant [24] which may be a source of error in excess of the error from the sulfonate parameterization. As the main manuscript primarily reports qualitative behavior, and moreover the chief finding of lipid tail protrusion-mediated insertion depends on hydrophobic as opposed to electrostatic effects, this parameterization is sufficient for the mechanistic understanding uncovered in this work. 1.2 Bilayer ribbon and AuNP assembly Bilayer ribbons were prepared following a similar methodology to West et al [25]. First, 200 DOPC lipid molecules were assembled into a bilayer in aqueous salt solution and equilibrated for 100 ns at 310 K and 1 bar pressure. The temperature was maintained with a velocity rescale thermostat and the pressure was maintained with a semi-isotropic Berendsen barostat with a time constant of 2.0 ps and a compressibility of bar -1 following typical methods [26]. Lipids were hydrated to a level of 45 water molecules per lipid [26]. After equilibration, the 200 lipid bilayer was duplicated along the x-axis and the box size was increased to add an additional 10 nm of water between the edges of the bilayer in the x-direction. The ribbon system was then further equilibrated for another 100 ns with an anisotropic Berendsen barostat with a time constant of 5.0 ps. The reference pressure and compressibility were set to 0 in the y-dimension to only allow box motion in the x/z dimensions with a reference pressure of 1 bar and a compressibility of bar -1. To minimize ribbon tilting in the z-dimension and prevent periodic images from interacting with each other, the phosphorus atoms in two lipids of the bilayer were constrained to move only in the x/y dimensions. These constrained lipids were selected to be in the same monolayer, with one lipid in the center of the ribbon and the other closer to the edge far from where the AuNP was eventually placed. This methodology successfully limited ribbon tilting while still permitting full lipid fluctuations and not interfering with AuNP insertion. After equilibration the system size was x 8.20 x 7.40 nm in the x/y/z dimensions respectively. Two different initial starting conditions for the AuNP were considered as shown in Fig. 2 of the main text. First, the AuNP was inserted above the middle of the ribbon to test whether spontaneous insertion could occur through a planar bilayer interface. For these simulations, additional water was first added to increase the z-axis to 12.5 nm, allowing the AuNP to be located above the ribbon while still having approximately 3 nm of aqueous solution between the top of the AuNP (in its starting position) and the bottom the ribbon s periodic image. After the addition of the extra water the system was briefly equilibrated for another 5 ns. The AuNP was then inserted into the simulation box 0.5 nm above the surface of the ribbon, overlapping atoms and any atoms within the hollow gold shell were removed, and additional counterions were added to neutralize the system. In total, this system contained 211,964 atoms including 63,026 water molecules, 400 lipids, 208 sodium ions, and 179 chlorine ions, maintaining a salt concentration of 150 mm. The second configuration consisted of inserting the AuNP near the side of the bilayer edge. Two box possibilities were tested. First, the AuNP was inserted near the bilayer edge in the same box just described. Second, a smaller box with some water removed to resize the z-dimension

19 to 9.0 nm was created, as the side geometry does not require as much distance between periodic images. Simulations in both boxes showed similar behavior, so the smaller box was used in the simulations described in the main text. After removing the water along the z-dimension, the total system size was 148,341 atoms including 41,203 water molecules, 400 lipids, 146 sodium ions, and 117 chlorine ions. A picture of the side configuration in the reduced simulation box is shown in Fig Calculation of committor from unbiased transition path As discussed in the main text, committor analysis [27, 28, 29] was performed to identify the transition state associated with the insertion of an AuNP into the bilayer from an initial surface-adsorbed position. The committor, p, is defined as the probability that a trajectory started from a particular system configuration will reach basin B before basin A, where A and B correspond to free energy minima of interest in some multidimensional free energy landscape. The committor can yield information about transitions in the system without defining a particular reaction coordinate for the dynamic trajectory. If an order parameter is able to uniquely define the two basins A and B and a path through phase space that transitions between one basin to the other is identified, then the transition state for this path can be identified as the configuration that yields p 0.5 where there is an equal probability of transitioning to either state [27, 28, 29]. Supplementary Figure 12 schematically illustrates the principle of committor analysis and defines the basins A and B for this study. The contour plot represents a generalized free energy landscape projected onto two generalized collective variables. The free energy basins are represented by the lowest-energy blue contours and are a function of both collective variables. The thick black line represents a trajectory through phase space obtained from an unbiased simulation that connects both basins. That is, the path is found from the natural dynamics of the simulation rather than chosen a priori and by connecting both basins it is a transition path. At several points along the trajectory, system configurations are extracted and used to launch several short unbiased trajectories after randomizing initial particle velocities. These short trajectories are represented as arrows pointed toward one of the two basins in Supplementary Fig. 12. The value of the committor then represents the fraction of the trajectories that commits to basin B over a short time scale. The transition state for the original transition path is identified by the red point and corresponding red arrows where an equal number of trajectories commits to either basin. In general, there may be many possible trajectories between basin A and B, each of which will have (at least one) transition state associated with it; the collection of transition states identified from a series of different trajectories is thus referred to as the transition state ensemble [28, 29]. For this system, simulations were characterized as committed to basin A and B as judged by two order parameters: the change in the distance between the AuNP and bilayer center of mass, x, and the number of hydrophobic contacts, H, between lipid tail atoms and hydrophobic atoms in the ligand monolayer. A hydrophobic contact was counted if two atoms were within 0.5 nm of each other. These parameters were selected on the basis of observations during the c

20 unbiased trajectories. A trajectory was marked as committed to B if x < 1.0 nm and if H > 50 at the end of a commitment trajectory. A trajectory was marked as committed to A if c x > 0.5 nm or if H < 5. Finally, simulations that did not meet either criterion at the end of c the run were marked as uncommitted and did not contribute to the calculation of p. This typically occurred when the distance had decreased below the threshold but the number of hydrophobic contacts, while possibly greater than zero, was still less than the threshold and fluctuated during the trajectory. Supplementary Figure 12 indicates snapshots of end configurations that would be marked as committed to basin A and B respectively. It must be emphasized that the committor is calculated from the initial unbiased trajectory, which means that the value of the committor does not need to increase monotonically as a function of time. Fig. 6a in the main text indeed shows that the committor increases at a time t =13 ns before decreasing and eventually increasing again. This behavior corresponds to the system reaching a particular configuration where it is more likely to commit to insertion, but does not actually proceed to insertion from that configuration, instead committing at a later time after reaching another configuration that favors insertion. Supplementary Figure 12 clarifies this behavior by showing a schematic illustration of the free energy of the system projected onto a one-dimensional generalized reaction coordinate which represents the path described previously. The schematic illustrates the time evolution of the system drawn along this path. The transient increase in the committor at t = 13 ns indicates that the system enters a state with a higher probability of inserting into the bilayer during the unbiased trajectory, but it does not actually commit to insertion until a few ns later. If a reaction coordinate for this path were known, the simulations could be biased to force the system to evolve along the reaction coordinate in which case the committor would be a monotonic function of time, but that was not done here due to the uncertainty in the reaction coordinate. 2. Experimental results 2.1 NP characterization :1 MUS:OT batch number 1 (Fig. 4 of the main paper) g (0.25 mmol) of chloro(triphenylphosphine)gold(i) was introduced to a 250 ml round bottom flask ml of dimethylformamide (DMF) and 9 ml of deionized water were then added mg (0.16 mmol) of MUS and µl (0.08 mmol) of OT were prepared in a separate vial and 10 ml water and 10 ml DMF were added. Dissolution was completed by sonication mg of borane tert-butylamine complex, the reducing agent, was prepared in a separate vial and 5 ml of DMF and 5 ml of water were added. The vial containing MUS and OT was poured into the 250 ml reaction flask in which the gold precursor was previously dissolved and was then vigorously stirred for 5 minutes using a magnetic stirrer. The vial containing the reducing agent was added to the reaction flask which was directly put into a heating bath at 120 C. After adding the condenser the magnetic stirrer was increased to 1000 rpm and the reaction took place for 1.5h. The characterization of this batch is presented in Supplementary Fig. 5.

21 :1 MUS:OT batch number 2 (Fig. 5 of the main paper) A mixture of 125 ml dimethylformamide (DMF) and 14 ml of deionized water was prepared in a 250 ml round-bottom flask. Three different 20 ml aliquots of this mixture were used to completely dissolve the reagents in separate vials: (i) mg (0.56 mmol) of chloro(triphenylphosphine)gold(i); (ii) 346 mg (1.2 mmol) of MUS with 56 μl (0.3 mmol) of OT and (iii) mg (1.5 mmol) of borane tert-butylamine complex. Dissolution was completed by sonication for 15 minutes at room temperature. The gold-salt solution was added to the round-bottom flask, followed by addition of the MUS:OT (4:1) mixture. The solution was vigorously (800 rpm) stirred for 10 minutes at room temperature and became turbid. At this point (10 min of addition) the reducing agent solution was added and the flask was connected to a condenser. The flask containing all the reactants was directly put into an oil heating bath at 125 C and let stirred (800 rpm) for 1.5 h. After this, the reaction flask was taken out of the heating bath and cooled under stirring (800 rpm) for another 1.5 h. The flask was then kept at 4 C overnight to precipitate the product. After removing the supernatant the nanoparticles were washed by centrifugation using different solvents to remove molecules not bound to the nanoparticles, as seen in the crude NMR in MeOD-d4 with 5% D 2 O (SI). The washings steps were: 3 re-suspensions followed by centrifugation pelleting (5500 rpm) steps using Sigma ACS grade acetone and ethanol, followed by 5 washes with DI-water using Amicon Ultra-15 Centrifugal Filter Devides (10k NMWL). The particles were then suspended in a minimal amount of water and freeze-dried to yield 80mg of a black powder. After etching the particles in a solution of 15 mg Iodine (Acros) in 0.6 ml of MeOD-d4 (Sigma) for 30 min under sonication, the spectra revealed a 33% OT content, i.e. a 2:1 MUS:OT stoichiometry. The characterization of this batch is presented in Supplementary Fig. 7. Dynamic light scattering (DLS) and zeta potential. A Zeta Nanosizer unit, Malvern, was used to perform dynamic light scattering and zeta potential measurements. The wavelength of the laser is 633 nm. The instrument was operated in back scattering mode (BSM) at an angle of 173. Measurements were run for a minimum of 5 min (5 measurements for each sample, 25 runs, 5 seconds each) at 25 C. Before the measurement, the nanoparticle solution in water and PBS was sonicated for 15 minutes at room temperature and filtered 3 times with a filter membrane of 0.22 m pore. A high concentration zeta cell kit was used to perform zeta potential measurements. Hydrodynamic diameters in water and PBS are shown in Supplementary Fig. 7. The zeta potential was measured as -48 ± 5 mv in water and -31 ± 5 mv in PBS. 2.2 Cleaning procedures of the silicon substrates The cleaning procedure for the silicon substrates (for both the AFM and QCM-D experiments) was the following: the substrates were immersed for 5 minutes in Milli-Q water, 5 minutes in ethanol, 5 minutes in water again and then dried with nitrogen. Afterwards, they were put in the UV-ozone cleaner for 10 minutes and immediately immersed for 30 minutes in 2% sodium dodecyl sulfate solution purchased from Sigma-Aldrich. Finally, the substrates were rinsed with

22 Milli-Q water, immersed in it for 5 minutes, and dried under a nitrogen flow. Immediately before the SLB deposition the substrates were put in the UV-ozone cleaner for 30 minutes. To clean the QCM-D tubing system, O-rings, and gaskets, we used the instructions of the machine s provider. 2.3 AFM experiments combined with QCM-D on silicon wafers AFM is a local technique that normally monitors a restricted area of a sample at a specific moment. The general concern with AFM experiments is that such local monitoring may not be representative of a system. For this reason, we preferred to combine AFM with QCM-D to test the generality of our hypothesis regarding the preference of AuNP for high curvature defect edges (see Fig. 4 of the main paper). To be as consistent as possible we chose the same kind of substrate (silicon wafer) for both the AFM and QCM-D experiments and we used the exact same preparation method for the lipid bilayer. With these experiments we showed that in the case of a perfectly formed SLB no detectable signs of interactions are measured with two different techniques. At the same time, using the same batch of nanoparticles (with the same properties like solubility and concentration) we noticed a clear sign of AuNP-SLB interaction in the presence of a defective bilayer with both the techniques used Completely formed, defect-free DOPC lipid bilayers The formation of a lipid bilayer on a silicon-oxide sensor was monitored with QCM-D using frequency shift and dissipation signals (blue and orange curves respectively in the insets of Fig. 4). The typical values of frequency shift and dissipation reported in the literature for a perfectly formed lipid bilayer are -25 Hz and less than respectively [30, 31]. The inset of Fig. 4a shows the QCM-D signals of frequency shifts and dissipation for the different overtones that we normally attribute to a completely formed DOPC bilayer. The values of frequency shift after fluxing the bilayer with PBS is -28 Hz and the dissipation value is for the third overtone (the one normally presented in QCM-D experiments). The typical drift in the frequency recorded before the addition of the lipid vesicles is 0.2 Hz min -1 and cannot explain the much lower value of frequency shift and slightly higher value of dissipation. The values of these signals can indicate the presence of a more viscous layer normally associated with the presence of a small amount of unfused vesicles [32]. The possible presence of these vesicles did not significantly affect the response of the system upon addition of AuNPs as discussed below. For this reason, we consider the signal recorded in this case as that of a defect-free bilayer. Fig. 4a shows a topographic AFM image obtained on a DOPC bilayer on a silicon wafer in PBS. In this case the bilayer was completely formed. The samples used in the AFM experiments were prepared separately from those used with QCM-D but followed the same procedure Incompletely formed, defect-rich DOPC lipid bilayer In the same experimental session, but on a different sensor of the QCM-D machine, the signal of an incompletely formed, defect-rich bilayer was recorded (inset in Fig. 4c). The different types of lipid bilayers (defect-free or defect-rich) were obtained by chance. In general it is known that the presence of divalent ions in solution (Ca 2+ ) and of specific molecules like EDTA can favor bilayer formation on silicon oxide surfaces [30, 33] and increase the reproducibility of

23 the bilayer preparation. We preferred to not introduce any extra complications to the system since it is know that divalent ions can affect the physical properties of lipid bilayers [34]. Moreover, the presence of incompletely formed bilayers was functional to our purposes and easy to distinguish from the recorded QCM-D signals. In the case of an incompletely formed bilayer we measured a lower value of the frequency shift (around -39 Hz for the third overtone) with a split of the overtones and a higher value of dissipation ( for the third overtone). Although we cannot extract specific information from the QCM-D signal regarding the kind of defects present in the bilayer, we can assume that it is not perfectly formed and thus defect-rich. The lower value of the frequency shift and higher value of dissipation (much higher than the one recorded in the previous case) suggest that the system contains a partial bilayer with some unfused vesicles as the signals lie in between the two configurations classically reported in the literature for a completely formed bilayer and unfused vesicles adsorption [35]. Fig. 4c again shows a topographic AFM image obtained from a DOPC bilayer on a silicon wafer in PBS. In this case the bilayer is not perfectly formed and it is possible to observe lipid patches with a height of roughly 5 nm via AFM. In general we cannot completely exclude the presence of a few vesicles in solution even if we carefully clean the samples before the AFM imaging procedure. To obtain an incompletely formed bilayer for AFM experiments, the surface on which a perfect bilayer was previously created was de-wetted on purpose. This is know to destabilize the bilayer and, especially in the case of a fluid lipid like DOPC, enables the creation of lipid patches as previously reported by S. J. Attwood et al. [36] Defect-free DOPC lipid bilayers upon MUS:OT addition A solution of 1 mg ml -1 of 2:1 MUS:OT AuNPs (batch number 1 in section of Supplementary Methods) in PBS was inserted into the QCM-D chambers and the signals of frequency shifts and dissipation were recorded for the different overtones. Before the insertion, we decided to offset all the signals (both frequency shifts and dissipation for all the overtones) to zero for a better visualization of the results. No significant change in either the dissipation or frequency signal was detected when the particles were added to the QCM-D chamber where the completely formed bilayer was previously obtained. The signal was comparable to the standard drift always detected in the QCM-D experiments with a value of roughly 0.1 Hz min -1. When a completely formed DOPC bilayer was imaged with the AFM (see Fig. 4a and 4b) no detectable changes in the topography were observed. Images d, e and f of Supplementary Fig. 8 present the same area of the bilayer imaged immediately, minutes, and 90 minutes after the 2:1 MUS:OT AuNP injection, respectively. Aside from some very high protrusions present prior to the injection of the particles that we can attribute to residual unfused vesicles, no changes are detected. Image f appears different because the tip of the AFM was contaminated by the presence of particles and extra lipids at the end of the imaging section. We decided to publish this image anyway for the sake of consistency Defect-rich DOPC lipid bilayer upon MUS:OT addition When the same 2:1 MUS:OT AuNP solution used in the previous experiment was injected into the QCM-D chamber where the signal of an incompletely formed, defect-rich bilayer was previously obtained, a fast decrease in the frequency shifts and an increase in the dissipation were measured. These effects are normally associated with an irreversible increase of mass on

24 the sensor, especially because these values of frequency and dissipation were maintained even after the removal of the extra particles from the QCM-D chamber and their substitution with clean PBS (see inset of Fig. 4d). The confirmation of this defect-mediated interaction was obtained with the AFM experiments when the same batch of AuNPs used in the QCM-D experiment was added to the PBS solution in which the defect-rich bilayer composed of isolated islands was imaged (images Fig. 4c and 4d and a, b, c in Supplementary Fig. 8). In the case of the patchy, defective bilayer, it was possible to notice the appearance of higher topographies starting from the edges of the membrane patches only minutes after AuNP insertion. 2.4 AFM experiments on mica substrates To ensure that the choice of substrate did not have an effect, DOPC lipid bilayers with large defects were obtained on mica substrates and imaged in PBS with AM-AFM (Supplementary Fig. 9). After the several hours necessary to acquire the images and to reduce the thermal drift, 100 µg ml -1 of 2:1 MUS:OT AuNPs were added to roughly 2 ml of PBS in which the two defect-rich bilayers were imaged previously. We followed the time evolution of the same area imaged before AuNPs insertion and one representative image is presented in Supplementary Fig. 9. The general effect detected is that the particles start to interact at the membrane edges and then diffuse into the bilayers. In Supplementary Fig. 10 the same area around a previously existing hole is imaged at three different times and the higher topography around the defect that we attribute to AuNP insertion is tracked. It is possible to follow the diffusion of the AuNPs from the edge to the bulk of the bilayer. Occasionally, it was possible to notice the appearance of potential AuNP-lipid aggregates in the previous holes of the SLB (see lighter regions in Supplementary Fig. 9 indicated by blue arrows). These structures were normally quite fragile and could be detached by simply scanning the same area with the AFM tip several times. With these experiments we show that, even though the presence of the substrate affects the properties of the lipid bilayer, the conclusion of a preferential interaction at defect edges is preserved. With both mica and silicon wafer substrates the presence of bilayer defects is necessary for the interaction to take place. 2.5 Control experiments on bare silicon wafer and on different lipid bilayers We performed control experiments to exclude non-specific adsorption of the AuNPs to either the silicon wafer or mica substrates (Supplementary Fig. 6). In the case of the silicon wafer we tested the 2:1 MUS:OT AuNP interaction with a bare, freshly cleaned silicon wafer substrate with both AFM and QCM-D. No significant interaction was found in either case. Generally, unless there is a specific interaction between the adsorbate and the substrate, it is more likely that the AFM tip removes the precipitate or the weakly bound NPs cluster while scanning. The result is a noisy image of the surface (see Supplementary Fig. 6). To emphasize the reproducibility of the experiment we decided to test the interaction of the same batch of 2:1 MUS:OT AuNPs used in Fig. 5 of the main paper (characterization presented in section of Supplementary Methods) with a different DOPC bilayer. To perform this

25 measurement we used even a different AFM (Veeco Multimode and MFP3D Asylum Research). The results presented in Supplementary Fig. 11 confirm the previous results described in the experimental section of the main manuscript. 3 Supplementary Discussion 3.1 Enhanced lipid tail protrusions at ribbon edge In the main text, the unbiased simulation trajectories shown in Fig. 2 indicate that no insertion is observed for AuNPs started in the middle configuration while insertion happens swiftly for AuNPs placed in the side configuration. An explanation for this distinction can be related to the necessity for lipid tail protrusions as shown in Fig. 6. The tail protrusions that lead to insertion are associated with a large increase in the number of atoms in water molecules within 0.5 nm of hydrophobic atoms in the lipid tail, leading to unfavorable lipid-solvent contacts. Supplementary Figure 1 shows the average number of hydrophobic lipid-solvent contacts as a function of the distance in the x-y plane between the lipid center of mass and the bilayer center of mass. The shaded region indicates the standard deviation after time-averaging over 20 ns. The plot shows that the number of water contacts is much higher closer to the edge of the bilayer ribbon versus near the middle, indicating that lipid tail protrusions leading to AuNP contact are more likely near the edge. The system snapshot is identical to the one in Fig. 2 and colors lipid head groups by the average number of contacts. The prevalence of water-lipid tail contact indicates the increased likelihood of an AuNP encountering a tail protrusion near the ribbon edge. 3.2 Calculation of diffusion coefficients In Fig. 2 of the main text, it is shown that a 1:1 MUS:OT AuNP placed above the middle of a bilayer rather than near the side is unable to insert into the bilayer. Instead, the AuNP diffuses along the ribbon surface, attracted to the surface by electrostatic interactions. To quantify the ability of the AuNP to travel along the surface, the diffusion coefficient was calculated from the mean-squared displacement (MSD) of the AuNP. The MSD was first recorded using the Gromacs tool g_msd with a restart time of 1 ns for time intervals between 0 and 20 ns to yield good statistics. The MSD was only calculated in the x-y plane to match the symmetry of the system. The diffusion coefficient was obtained by first fitting a line to the MSD then using the relation 2 r( t) r0 4D t where r is the position projected onto the x-y plane. Two diffusion coefficients were calculated: D, representing the diffusion of an AuNP above the bilayer surface, and surf D bulk, representing the diffusion coefficient of an AuNP embedded in the bilayer (see section on embedding below). D surf was found to be cm 2 s -1 and bulk D was 7 found to be cm 2 s -1. For comparison, previous simulations of the diffusion 6 coefficient for similarly sized AuNPs in aqueous solution have estimated D 1 10 cm 2 s -1 [11]. The diffusion of AuNPs is thus much slower in both the surface adsorbed and embedded state as expected. However, it is still clear that D is still sufficiently large that a surface-adsorbed AuNP, such as the AuNPs in the middle configuration in Fig. 2, could migrate to defects over reasonable experimental time scales. surf solv

26 3.3 Larger particle insertion The 2.0 nm core diameter AuNPs tested previously were chosen on the basis of computational efficiency and relevance to experimental systems. However, the approximate size of these AuNPs, including the grafting monolayer, was comparable to the thickness of the bilayer. To test whether insertion could occur spontaneously even for AuNPs larger than the bilayer thickness, a 3.0 nm core diameter 1:1 MUS:OT AuNP was assembled and inserted in an unbiased simulation using the same methodology as with the 2.0 nm counterpart. Supplementary Fig. 2 shows a plot of x and H vs. time for a 100 ns trajectory and corresponding snapshots, again mimicking the approach taken with the 2.0 nm AuNPs. The behavior is qualitatively the same. We thus believe that the insertion pathway is not significantly altered by small shifts in AuNP sizes, which supports experimental observations for polydisperse AuNP samples. 3.4 Full unbiased trajectories Supplementary Figure 3 shows full 500 ns trajectories of the unbiased insertion of both 1:1 MUS:OT and all-mus AuNPs into the DOPC bilayer ribbon. The first 100 ns of these trajectories are shown in Fig. 3 of the main text. Supplementary Figure 3a shows the trajectories as a function of x, the distance between the center of mass of the AuNP and the ribbon projected onto the x-axis. It is apparent from these trajectories that the all-mus particle inserts to a smaller degree than the 1:1 MUS:OT particle as the value of x plateaus at a larger value. Supplementary Figure 3b shows similar plots but with the number of hydrophobic contacts, H, as a function of time for both AuNPs. These plots confirm similar trends, with the number c of hydrophobic contacts rising as the AuNPs push farther into the ribbon. Supplementary Figure 4 shows additional simulation snapshots for both the MUS:OT and all-mus particles. Snapshots are shown at both t = 5 ns, when the AuNPs are in the initial fluctuations regime, and t = 500 ns, when the AuNPs have inserted into the bilayer. The top views better show the extent of insertion as quantified by x, with the MUS:OT particle clearly inserting farther by the end of 500 ns than the all-mus particle. The two side views are shown with either the lipid tails or AuNP removed to show system deformations. With the lipids removed, the snorkeling of hydrophilic end groups toward water becomes more evident for both particle types and the extension of the pink hydrophobic ligands into the bilayer core is apparent for the MUS:OT particle. With the AuNP removed, the extent of ligand deformations become more obvious. In the case of the MUS:OT particle, the final snapshot at 500 ns shows that lipid positions are effectively unperturbed relative to their initial positions prior to insertion. 3.5 Identifying anchor lipid In the main text, the transition state for AuNP insertion was found using committor analysis. To determine what AuNP-lipid interactions were occurring at the transition state, the number of lipids in hydrophobic contact with the AuNP was tracked as a function of time. A lipid was recorded as being in hydrophobic contact with the AuNP if at least one hydrophobic atom in c

27 the lipid was within 0.5 nm of at least 1 hydrophobic atom in the AuNP. Supplementary Figure 13 shows a plot of the lipids in contact with the AuNP between ns of the unbiased trajectories corresponding to the time period in Fig. 6. The bars indicate which lipids were in contact, with the lipid indices assigned based on the time of first contact. The vertical dashed lines indicate the transition state as identified from committor analysis. It is observed that while many lipids make contact with the AuNP surface during this time period, the transition state corresponds to the creation of the first stable AuNP-lipid contact that persists from the transition state until the end of the simulation. This anchor lipid is thus identified as the lipid that is responsible for initiating the onset of insertion. In the case of the all-mus particle, the anchor lipid s contact with the AuNP does not actually become stable until a second lipid disengages from the AuNP surface at the transition state. The anchor lipids for both trajectories are shown in Fig Additional transition state snapshots In Fig. 6 of the main text, the transition states for the 500 ns unbiased simulations of both the 1:1 MUS:OT and all-mus particles are obtained from committor analysis. It is shown that the transition state for each particle occurs when a single lipid makes contact with the AuNP surface, at which time the committor obtains a value of 0.5. Having recognized that the critical feature of the two transition states is a protruding lipid tail, we can look for a similar state in other transition paths b identifying first stable contact between a protruding lipid and the AuNP monolayer following a procedure similar to that outlined for Supplementary Fig. 13. If a similar state is found for many such paths, then there is a strong likelihood that this is a general feature of AuNP insertion via the high curvature edge - i.e. it is a feature shared by many states in the transition state ensemble. The other unbiased 40 ns trajectories provide several more examples of transition paths for the MUS:OT particles. Moreover, because many short, independent simulations are launched from configurations where the committor is greater than 0 but less than 0.5, some of these paths represent new transition paths that insert via contact with a different anchor lipid than the original unbiased trajectories. Therefore, we have a large set of potential transition paths, and indeed we find that contact with a single lipid can be identified as a presumed transition state in these paths. Supplementary Figure 14 shows snapshots of additional putative transition states identified from the committor trajectories by finding the onset of AuNP-lipid hydrophobic contacts. These additional snapshots show that a wide variety of lipid protrusions can lead to insertion, with the chief similarity being the necessity of lipid solvent exposure as explained in the main text. 3.7 Solvent-accessible surface area The hydrophobic solvent-accessible surface area (SASA) is correlated with the magnitude of the free energy penalty for the exposure of hydrophobic material to water [37, 38]. The magnitude of the hydrophobic effect, or the driving force for association between hydrophobic molecules in water, can be approximated by calculating the reduction in the SASA. For this system, the SASA was calculated with the Gromacs tool g_sas using the method of Eisenhaber et al [39]. The SASA was measured using a probe radius of 0.14 nm to correspond to the radius of water [40]. Supplementary Figure 15a shows the hydrophobic SASA of the AuNP for both 1:1 MUS:OT and all-mus particles while Supplementary Fig. 15b shows the hydrophobic SASA of the bilayer