Carbon Nanotubes Mediate Fusion of Lipid Vesicles - ACS Nano

Jan 19, 2017 - The fusion of lipid membranes is opposed by high energetic barriers. In living organisms, complex protein machineries carry out this bi...
2 downloads 13 Views 5MB Size
Carbon Nanotubes Mediate Fusion of Lipid Vesicles Ramachandra M. Bhaskara,†,§ Stephanie M. Linker,†,§ Martin Vögele,† Jürgen Köfinger,† and Gerhard Hummer*,†,‡ †

Department of Theoretical Biophysics, Max Planck Institute of Biophysics, Max-von-Laue Straße 3, 60438 Frankfurt am Main, Germany ‡ Institute for Biophysics, Goethe University Frankfurt, 60438 Frankfurt am Main, Germany S Supporting Information *

ABSTRACT: The fusion of lipid membranes is opposed by high energetic barriers. In living organisms, complex protein machineries carry out this biologically essential process. Here we show that membrane-spanning carbon nanotubes (CNTs) can trigger spontaneous fusion of small lipid vesicles. In coarse-grained molecular dynamics simulations, we find that a CNT bridging between two vesicles locally perturbs their lipid structure. Their outer leaflets merge as the CNT pulls lipids out of the membranes, creating an hourglass-shaped fusion intermediate with still intact inner leaflets. As the CNT moves away from the symmetry axis connecting the vesicle centers, the inner leaflets merge, forming a pore that completes fusion. The distinct mechanism of CNT-mediated membrane fusion may be transferable, providing guidance in the development of fusion agents, e.g., for the targeted delivery of drugs or nucleic acids. KEYWORDS: membrane fusion, vesicle, carbon nanotube, membrane staples, coarse-grained simulations

I

compounds are tethered to CNTs for targeted delivery and localized actions.29 Several experiments and all-atom molecular dynamics (MD) simulations have shown that CNTs can interact favorably30−32 with lipid bilayers, penetrate into them,32−34 and embed into them spontaneously.35,36 Furthermore, mammalian cells actively take up CNTs either by endocytosis or passive diffusion.33 CNTs can also mimic membrane-protein channels and contribute to passive transport,35 high-efficiency proton conduction,37,38 and sensing39 across membranes. Recently, Geng et al.35 were able to reconstitute CNTs in lipid vesicles of ∼200 nm diameter. Using cryo-transmission electron microscopy (cryo-TEM), they showed that CNTs are embedded into lipid vesicles, spanning both membrane leaflets, with their axes approximately perpendicular to the membrane. In their micrographs, several images show CNT-stapled vesicles, with single CNTs bridging two vesicles. Furthermore, the images contain unusual, elongated lipid vesicle structures that could have resulted from fusion.40,41 Putting these two observations together, we hypothesize that bridging CNTs could mediate vesicle fusion, which would result in larger, elongated vesicles.

n living organisms, the fusion of lipid membranes is essential for cellular transport and communication.1,2 To overcome its multiple free energy barriers,3 complex protein machineries have evolved that carry out this challenging task in a controlled and efficient manner.1,2,4−8 Several nonbiological factors induce fusion, such as osmotic stress,9 dehydration by addition of polymers,10 cations,11,12 amphiphiles,12−14 electric fields,15 surface tension,16 and irradiating membrane-attached gold nanoparticles.17 They act by bringing vesicles together and by destabilizing the vesicle surfaces, thereby promoting fusion-stalk formation. By contrast, specifically designed molecules for initiating and regulating vesicle-fusion reactions are few.12,18−20 Here we show with unbiased computer simulations that carbon nanotubes (CNTs) can induce spontaneous vesicle fusion. Previously, membrane fusion has been studied theoretically using a range of methods, from membrane elastic models21,22 over field theoretic descriptions23,24 to computer simulations. Inducing fusion in coarse-grained7,25−27 or atomistic simulation models21 required the imposition of constraints or the addition of bridging lipids with tails anchored in opposing membranes. By contrast, CNTs may offer a route to fusion realizable to both experiment and simulation. CNTs have already been used as delivery system for specific and localized targeting of anticancer drugs.28 Exploiting the chemistry of end-group functionalization, multiple chemical © 2017 American Chemical Society

Received: August 12, 2016 Accepted: January 19, 2017 Published: January 19, 2017 1273

DOI: 10.1021/acsnano.6b05434 ACS Nano 2017, 11, 1273−1280

Article

www.acsnano.org

Article

ACS Nano

we simulated self-fusion between a vesicle and its periodic image. In the simulations, we observed more than 400 events of spontaneous CNT-mediated vesicle fusion (Figure 1b; Supplementary Movies S1 and S2; Table 1). These events share a common sequence of steps as detailed below (Figure 2). In a first phase, the CNT-stapled vesicles rearrange to minimize water exposure of the hydrophobic CNT surface in the gap between the two vesicles (Figure 2a). To minimize this hydrophobic mismatch, the CNT pulls the two vesicles together. Lipids of the outer leaflets slide along the CNT and coat its entire outer surface (Figure 2a, bottom). To quantify these interactions, we count the number of lipid tail-beads within 1 nm of each CNT bead and color the CNT accordingly (Supplementary Text S1 and Figure S3). In a second phase, tilting of the CNT promotes the formation of a fusion stalk. By tilting away from the symmetry axis connecting the centers of the lipid vesicles, the CNT buries even more of its hydrophobic surface in the lipid environment (Figure 2b). As the entire CNT surface becomes lipid-coated, tails of lipids in the two outer leaflets are brought into contact (Figure 2b; bottom). This causes the outer leaflets to merge, creating the prefusion stalk with a continuous outer leaflet enveloping both vesicles (Figure 2b; Supplementary Movies S1 and S2). Lipids of the fused outer leaflet (Figure 2b, blue and green tails) coat a continuous ring around the center of the CNT, flanked by lipids of the still intact inner leaflets of the two vesicles (Figure 2b, red and yellow tails). This arrangement maximizes interactions between the hydrophobic exterior of the CNT and the lipid tails, while maintaining interactions of the functionalized ends of the CNT with polar head-groups of lipids in the inner leaflets (red and yellow beads in Figure 2) and with water inside the vesicles. This stalk structure constitutes a metastable intermediate along the fusion pathway. Further CNT-lipid rearrangements promote the transition from the fusion stalk to an hourglass-shaped structure (Figure 2c). In the fusion stalk, the CNT ends (at ±4−6 nm from the center) are predominantly coated by lipids from inner leaflets (Figure 2b). In the transition to the hourglass shape, the two continuous rings of outer leaflet tails around the CNT center open up (green and blue tails in Figure 2b,c). Driven by hydrophobic interactions, the acyl chains of lipids in the inner leaflets now protrude toward the center of the CNT (Figure 2c bottom; Supplementary Movie S1). These interactions (red and yellow lipid tails) on opposite sides of the CNT bring the inner leaflets into close contact and cause deformations of the inner water-filled cavities from spherical into teardrop-shaped structures (Figure 2c, bottom). At this stage, the tilt angle of the CNT relative to the symmetry axis connecting the vesicle centers has risen to nearly 45 deg (Figure S4). The delicately balanced hourglass-shaped metastable intermediate has formed. The subsequent opening of the fusion pore is associated with further CNT movement (Figure 2d). As the CNT moves away from the symmetry axis, it pulls lipid tails of the two inner leaflets together. A continuous line of inner leaflet-CNT contacts develops along the length of the CNT (Figure 2d, red and yellow tails; Figure S3d center, right). These interactions are spirally arranged around the CNT axis. In Figure 2d, the CNT is coated by lipids of the upper vesicle (red tails; top and front), which merge with lipids of the lower vesicle (yellow tails; bottom and back). The teardrop-shaped protrusions of the inner cavities are pulled together and the two inner leaflets merge. As a result, the seal between the two vesicles breaks.

In the present study, we use MD simulations to explore this hypothesis of vesicle fusion induced by bridging CNTs. We base our initial simulation setup on the structures seen in the EM images by Geng et al.35 Starting from states with CNTstapled vesicles, we performed extensive MD simulations of hundreds of fusion events to characterize the molecular details of the CNT-mediated fusion process. A comparison of CNTmediated fusion with the current picture of free fusion identifies key differences that are likely responsible for the speed-up of fusion. In particular, we highlight the role of CNTs in lowering the distinct barriers for merging first outer and then inner leaflets. We also show that the rate of fusion decreases with increasing vesicle size.

RESULTS To explore the possibility of CNT-mediated membrane fusion, we simulated vesicles stapled together by bridging CNTs (Figure 1). We considered vesicles of three different sizes (11.5,

Figure 1. CNT-stapled vesicles fuse spontaneously. Coarse-grained simulation setup A showing the initial (a) and final configuration (b) of fusion simulations. (a) A 11.8 nm long CNT (black) stapling two POPC vesicles (green, ∼15 nm diameter) modeled to mimic selected cryo-TEM images (Figure S1a,b) was simulated in explicit water (blue).

15, and 28 nm diameter) composed of three different lipids (POPC, DOPC, and DSPC) with varying degrees of unsaturation. CNT-stapled vesicles of similar size and shape can be seen in the cryo-TEM images35 (Figure S1). We used the coarse-grained MARTINI model, which has been instrumental in modeling vesicles and fusion processes,42 and a flexible CNT model with modified parameters for polar end groups to mimic end group functionalization (see Methods). We performed multiple simulation runs in two different system setups with 40 000 to 330 000 particles, accumulating more than 370 μs of total simulation time. In setup A, we simulated CNT-mediated fusion of two vesicles in coarse-grained explicit water (Figure 1a). In the reduced setup B (Figure S2), we exploited periodic boundary conditions for fusion.43,44 In a box containing only one vesicle, a CNT, and coarse-grained water, 1274

DOI: 10.1021/acsnano.6b05434 ACS Nano 2017, 11, 1273−1280

Article

ACS Nano Table 1. Simulation Setups in Studies of CNT-Mediated Vesicle Fusiona setup

lipid type

lipid molecules

total beads

simulation time (ns)

vesicle diameter (nm)

simulations

fusion events

Setup A

POPC DOPC DSPC POPC POPC POPC POPC DSPC DOPC DOPCb POPC POPC

4031 4032 4035 2438 4031 1183 1980 1983 2016 6666 2500 4096

202480 202587 231039 124042 219603 39632 85810 75844 115152 329881 102451 215234

1000 1000 1000 500 1000 500 500 220 500 5000 10000 5000

15 15 15 11.5 15 11.5 15 15 15 28 11.5 15

10 1 1 100 100 100 100 1 1 10 2 2

10 1 1 100 99 97 95 1 1 0 0 0

Setup B

Control a b

In the control simulations, POPC vesicles were simulated at close proximity with and without harmonic restraints on their center of mass-distance. Simulations at 350 K.

compensated in part by the CNT. Owing to the large desolvation cost of the functionalized CNT ends and the strong hydrophobic interactions of the inner leaflet and the CNT, the CNT sticks to the inside of the fused vesicles for extended times in our simulations. On a much longer time scale, we expect that the CNT will reestablish a membranespanning orientation to bring its polar rims into better contact with water, to reduce the disturbance of the bilayer structure, and to establish a more favorable lipid coat of the CNT surface. This process is likely associated with the flipping of lipids between the two leaflets. Major fusion steps are associated with distinct transitions in lipid coating patterns of the CNT (Figure 2). These are quantified as CNT-lipid interactions and mapped onto the cylindrical surface of the CNT to build interaction maps (Figure S3). During fusion, the character of the interaction map changes (Supplementary Movie S1), as old symmetries break and new ones emerge (Supplementary Text S1). All simulations with small and medium-sized CNT-stapled vesicles showed fusion events (Table 1). Fusion events occurred in all three different lipid environments, making CNT-mediated fusion lipid-type independent (Table 1). As negative control, we performed long simulations of systems with two vesicles without CNTs. In the absence of CNTs, vesicles simulated at near-contact for long times (10 μs) did not fuse, showing that the CNT clearly lowered the barrier for fusion. The kinetics of the process and the rate of CNT-mediated fusion is dictated by the system sizes and the ease of transiting through key intermediate stages. We find that the overall fusion process of 15 nm CNT-stapled vesicles can take anywhere from hundreds of nanoseconds to 1−2 μs. To characterize the timescales associated with transitions into these intermediate states, we determined waiting times to arrive at the fusion stalk, the hourglass shape, and the fusion pore from repeated fusion simulations (Figure S5). To estimate the characteristic times associated with the three distinct transitions along the fusion pathway, we fit convolutions of single exponential functions, with added lag times, to the waiting time distributions (see Methods). The lag times represent the drift of the system toward a particular intermediate state before the next stochastic event. We find three distinct characteristic times along the fusion pathway corresponding to the formation of the fusion stalk, the hourglass shape, and the fusion pore, starting from CNT-stapled vesicles (Figure S5). The formation of the fusion

Figure 2. CNT-mediated fusion of vesicles. (Top) Snapshots of a simulation trajectory showing CNT-mediated vesicle fusion in setup A. Lipid phosphate groups are colored to distinguish the two vesicles, their outer leaflets (blue/green in top/bottom vesicle), and their inner leaflets (red/yellow in top/bottom vesicle; drawn larger for clarity). (Bottom) Zoom-up of the boxes (clipped plane with CNT) showing the coating of the CNT (gray) by lipid tails (colored by their respective leaflets). Panels (a−e) show the progression of fusion. (a) CNT-stapled vesicles. (b) Fusion-stalk formation. (c) Hourglass shape. (d) Fusion-pore formation. (e) Postfusion state.

Opening of the fusion pore allows water to pass through (Figure 2d). The two vesicles have thus fused into one. After the pore has formed, the interior volumes of the two vesicles merge and their water contents mix (Figure 2e). The fusion pore expands, allowing lipids from each vesicle to diffuse freely within the membrane of a single large vesicle. Note that during fusion, the two outer leaflets merge first (fusion stalk; Figure 2b, mixing of blue and green tails), and the two inner leaflets merge later (pore formation; Figure 2d, mixing of red and yellow tails). There is no exchange of lipids between outer and inner leaflets. The lipid surface area and volume of the postfusion vesicle are roughly those of the two initially spherical vesicles combined, A12 = A1 + A2 and V12 = V1 + V2. The reduced area of a12 = A12/(V12)2/3 thus exceeds that of a sphere by up to a factor of 21/3. The fused vesicle therefore adopts a prolate axisymmetric structure (Figures 1b; 2e and S1c,d).40 Its membrane is perturbed further by an imbalance of the numbers of lipids in postfusion inner and outer leaflets as a result of the finite membrane thickness. A lack of lipids in the inner leaflet is 1275

DOI: 10.1021/acsnano.6b05434 ACS Nano 2017, 11, 1273−1280

Article

ACS Nano stalk with a tilted CNT is a fast process, requiring only ∼10−70 ns in our simulations (characteristic time, τStalk = 18 ns). The next process involves the transition into an hourglass shaped intermediate with teardrop shaped inner cavities. In 90% of our simulations, this transition occurred within the first 200 ns (τHourglass = 23 ns). Finally, pore formation takes ∼80−800 ns (τFusion = 116 ns). Small vesicles stapled by CNTs fuse faster (Figure S6). The high curvature of their membranes causes lipid packing defects, which facilitate the formation of the fusion stalk. The finite thickness of the membrane, and the difference in curvature of its two leaflets in both magnitude and sign cause further perturbations to the lipid organization. These perturbations ease CNT tilting and fusion-stalk formation. By contrast, larger vesicles face higher barriers to fusion.45 Their bilayers are relatively more ordered. As a result, CNT-mediated fusion of large vesicles exhibits longer pauses in the hourglass intermediate. Indeed, for the largest vesicles studied (28 nm diameter; Table 1), fusion did not complete on the MD time scale even at elevated temperature (50 μs of aggregate simulation time). This corresponds to a >50-fold drop in the estimated rate of fusion upon doubling the vesicle size. The disruption of inner leaflet packing and reorganization into teardrop shaped structures eases formation of hourglass shapes in small vesicles. In larger vesicles the disruption of the inner leaflets by CNTs is small and hence delays the transition toward hourglass shapes. We explored the effect of CNT length on fusion kinetics with additional simulations in setup A (Supplementary Text S2 and Table S1). We found that vesicles stapled by CNTs longer than 10 nm fuse readily (Figure S8; Table S1). By contrast, CNTs shorter than the combined thickness of the two bilayers are unable to staple two vesicles completely. 7−10 nm long CNTs form fusion stalks, but the vesicles do not transition to hourglass structures within the simulation time (Figure S8; Table S1). The distortions of the inner leaflets that can be induced by the long CNTs, but not the shortest ones, appear to aid fusion. From the MD simulation results, we can also make at least a rough estimate on how much a bridging CNT lowers the free energy barrier to fusion of two vesicles at near-contact. With a CNT bridging two 15 nm vesicles, we found fusion to complete in the MD simulations after about 200 ns on average. By contrast, two 15 nm vesicles without CNT and restrained to near contact did not fuse during 2 × 5 μs of MD. From the ratio of these two times we can estimate a lower bound on the drop in the activation free energy for fusion with CNTs, ΔΔG‡ > kBT ln(10 000/200) ≈ 3 kBT. We obtain an even sharper bound of ΔΔG‡ > 6 kBT by noting that free fusion did not even proceed to the fusion-stalk state, which occurred with bridging CNTs on a 25 ns time scale. In Supplementary Text S3 and Figure S9, we also provide a rough estimate of the free energy landscape of fusion, extracted from the MD simulations with the help of a 1D diffusion model46 (Supplementary Text S3). On this free energy profile, the fusion-stalk and hourglass states appear as local minima, and the fused state as the global minimum (Figure S9).

bilayers33,47,48 and vesicles49 using a range of computational methods. The effect of CNT length on orientations of functionalized CNTs has also been studied.50 In Supplementary Text S4 and Figure S10, we examine two possible routes of CNT insertion into vesicles. We show that CNTs can insert both concomitantly with vesicle formation (Figure S10a) and subsequently into an already formed vesicle (Figure S10b). Coarse-grained MD and dissipative particle dynamics simulations have already been used to study CNT coating51 and transmembrane insertion52 mechanisms. However, in the small simulation systems studied over limited times, compared to experiment, we did not observe simultaneous CNT embedding into multiple vesicles, which is likely a rare process favored by longer tubes (Text S4). The cryo-TEM images by Geng et al.35 are suggestive of CNT-mediated fusion, as observed in our simulations. Most notably, several of their images show vesicles with axisymmetric prolate shapes (Figure S1; and Extended Data Figures 4 and 5 of ref 35), as expected for postfusion states from our simulations (Figures 1b and 2e) and from elastic theory.40 Indeed, the occurrence of prolate-shaped vesicles in EM micrographs has previously been used as a reporter of vesicle fusion.41 We note that CNTs are visible in some of the prolate vesicles (Figure S1c,d), reinforcing the possibility that they resulted from CNT-mediated fusion. The elongated shapes of some structures indicates multiple fusion events. Alternatively, prolate vesicles in EM images could also be explained by (a) volume changes through water transport, (b) solution extrusion through filters, and (c) crowding effects in thin films promoting free fusion during EM sample preparations. (a) Osmotically driven water transport out of the vesicle through membrane embedded CNTs could have resulted in vesicle volume shrinkage. This would be possible only under conditions of severe osmotic shock and hypertonic solution exterior of vesicles, which seems unlikely because the hydration process used deionized water.35 (b) Extrusion of hydrated lipid solutions through 200 nm polycarbonate filters was reported to have initially formed large, multilamellar vesicles.35 After repeated passes through the polycarbonate filters, a unimodal distribution of unilamellar spherical vesicles was obtained. The EM micrographs in Geng et al.35 were obtained after 10 cycles of extrusion, which would decrease the chances of obtaining long prolate shaped vesicles after extrusion.35 (c) Finally, EM grid preparation could have induced fusion. The samples to be imaged were trapped in thin-films on Cu grids and dried by hand with filter paper. The film thickness is therefore variable and depends on the extent of excess water removal. This could have resulted in artificially high concentrations of lipid vesicles, which would have favored both CNT-mediated and direct fusion and could have resulted in prolate vesicles. However, a recent study of the effect of αsynuclein on vesicle fusion also showed prolate vesicles in the presence of the protein and almost spherical vesicles in the absence of protein, using similar techniques for EM sample preparation, i.e., film formation by blotting for 2.5 s using filter paper.41 The large spherical vesicles stapled by CNTs (Figure S1a,b) in the cryo-TEM images are indicative of trapped intermediates during the fusion of large vesicles. Our finding that fusion slows down with increasing vesicle size (Table 1) provides a possible explanation. We tested this further by performing simulations of a flat bilayer (effectively an infinite-size vesicle) stapled to a 15 nm vesicle by a single CNT (Figure S11). Formation of the

DISCUSSION Our simulations started from a single CNT bridging two vesicles, as a model of the structures seen in the experimental images by Geng et al.35 The formation of these structures requires membrane insertion of CNTs, as studied for 1276

DOI: 10.1021/acsnano.6b05434 ACS Nano 2017, 11, 1273−1280

Article

ACS Nano

Figure 3. Comparison of fusion pathways. Schematic comparison of (top) CNT-mediated vesicle fusion and (bottom) free fusion. The topology and interactions of inner (red and yellow) and outer leaflets (blue and green) display characteristic patterns that distinguish the two fusion pathways. The top panels show snapshots taken from MD fusion simulations. The middle panels show corresponding cartoon representation. The bottom panels for free fusion are based on a schematic by Chernomordik and Kozlov.4

hourglass-shaped intermediate was slow (≈1.5 μs) by comparison to small vesicles, and fusion did not complete on the MD time scales (>6 μs), consistent with our findings for large vesicles. Alternatively, CNT-stapled but unfused vesicles in the cryo-TEM images could have been caused by defects in the CNT that prevented lipids from sliding along the hydrophobic CNT surface to form the fusion stalk. Bending, buckling, and breaking defects often arise on CNT surfaces due to mechanical shearing during sonication cutting. Sidewall oxidation defects53 could attract water and prevent efficient lipid coating essential for fusion.53 To test these different hypotheses will require additional experimentation, e.g., by studying the effects of CNT processing. Moreover, to decouple the contributions of the CNT fusogenic activity and of vesicle curvature stress would require the calculation of accurate maps of the underlying free energy surfaces from extensive biased simulations. In summary, we reason that the CNT-induced fusion is a parsimonious but certainly not a definitive explanation for the prolate shaped vesicles observed in the cryo-TEM images.35 The images are consistent with our findings, but do not provide direct validation. Direct and quantitative studies of the various factors of vesicle size, CNT length, CNT processing and defect content, grid preparation, etc. and controls without CNTs will be required for a conclusive assessment of the fusogenic activity of CNTs in experiment. To better understand the fusion process, we juxtapose in Figure 3 the conventional free fusion-mechanism with the CNT-mediated fusion mechanism. In free-fusion, two lipid membranes come into close proximity, allowing their proximal outer leaflets to interact.27,54 Interaction also requires local dewetting to destabilize the surface bilayer interactions at a

single point.22 This defect then grows with time and allows the proximal leaflets to merge and form the fusion stalk, while the distal leaflets remain intact. This so-called hemifusion intermediate adopts an hourglass shape.4 Subsequently, the stalk grows and forms the hemifusion diaphragm where the distal leaflets come in close contact. The hemifusion diaphragm facilitates the formation of a fusion pore. Finally, the fusion pore opens and expands, connecting the two vesicle compartments. We note mechanistic differences between CNT-mediated and free fusion. First, stalk formation is rapid, in stark contrast to protein-free fusion. The lipid molecules slide along the hydrophobic surface of the CNT to form a continuous coat that eases stalk formation. We find that CNT-stapled vesicles are able to spontaneously transit to the fusion-stalk state in all simulation runs, indicating that CNTs greatly reduce the barrier for fusion-stalk formation. By contrast, in conventional free fusion, the stalk-formation step is very slow, thought to be associated with the expansion of small surface defects on the proximal leaflets.3 Second, no hemifusion is observed. Free fusion can occur with or without a hemifusion diaphragm intermediate.3,55−57 The size and stability of the hemifusion diaphragm is dependent on several factors such as the size of the fusion stalk and the ability of lipids to spontaneously splay.58 In the hourglass-shaped intermediate of CNT-mediated fusion, the bridging CNT itself blocks the passage of solvent between the two cavities, instead of either a disc-shaped bilayer diaphragm or a small bilayer-shaped prepore rim in the freefusion pathways. Pore formation is a result of the CNT moving away from the central axis of the delicately balanced hourglass shape, whereas in free fusion, pore formation is dictated by bilayer diaphragm stability. 1277

DOI: 10.1021/acsnano.6b05434 ACS Nano 2017, 11, 1273−1280

Article

ACS Nano

× 60 nm3) and solvated using the coarse-grained water model. The systems were first energy minimized and equilibrated at 295 K for 10 ns. The x- and y-axis of the boxes were fixed and pressure was only coupled to the z-axis at 1 bar, to keep the bilayer patches from merging with its periodic images. The systems were then simulated for ∼1 μs using a 20 fs time step. During these simulations, the bilayer patches spontaneously curved to form spherical vesicles (11.5, 15, and 28 nm diameter). For POPC lipids, the 11.5 nm vesicle contained 1250 lipids (395 in the inner leaflet and 855 in the outer leaflet) and the 15 nm vesicle contained 2048 lipids (726 in the inner leaflet and 1322 in the outer leaflet). Larger vesicles of 28 nm diameter were made only with DOPC lipids and contained 6728 lipid molecules (2815 in the inner leaflet and 3913 in the outer leaflet). These large vesicles were used for fusion simulations at higher temperatures (350 K) only in setup B. Carbon Nanotubes. Coarse-grained MARTINI models generally use a 4:1 mapping for interaction sites (beads). For ring structures, special beads with a 2:1 mapping are used.65 Here, all carbon-only interaction sites of CNTs were mapped to CNP interaction sites used previously for fullerenes.68 A similar mapping has also been employed previously for capped carbon nanotubes.50 We modeled open CNTs with a polar end-group functionalization (typically −OH groups) by employing a more polar interaction site (SNda beads) for the rims (Figure S7).69 Bonded interactions between neighboring beads were modeled with a bond length of a = 0.47 nm and a force constant of 5000 kJ mol−1 nm−2, similar to the model for capped nanotubes.50 The CNTs consist of 30 rings with 10 beads each, resulting in a total length of 11.8 nm and a diameter of 1.5 nm. Angle potentials along the rings were modeled with a force constant of 350 kJ mol−1 rad−2. The equilibrium angle α = π(N − 2)/N depends on the diameter of the CNT, as determined by the number N of beads per ring (here: N = 10). Previous CNT models had additional long-range bonds between distal rings to account for the stiffness of capped CNTs. We found that in the absence of capped structures, they failed to maintain the equilibrium shapes and were bent in lipid simulations. Here, stiffness along the tube was maintained by introducing improper dihedral angle potentials that maintain the angle between two adjacent triangles near the target value of β = 2 cos−1[tan(π /2N )/ 3 ] with a force constant of 350 kJ mol−1 rad−2 (Figure S7).69 Kinetics of CNT-Mediated Fusion. For the small vesicles studied here, the whole fusion process was completed in a few hundreds of nanoseconds to a couple of microseconds (see Supplementary Movie S1 and S2). Owing to the smoother energy landscape of the MARTINI model, diffusion and other dynamic processes tend to speed up by factors of four to eight when compared with experiments and atomic simulations.42 Here, we always refer to simulation (“MARTINI”) time, without rescaling, when discussing time scales. We record the times of forming the fusion-stalk (t1), the hourglass shape (t2), and the fusion pore (t3). Since the transit through these phases is sequential and directional (i.e., t1 < t2 < t3), we modeled the kinetics of each of the three sequential transitions between the states along the fusion pathway as independent Poisson processes (see, e.g., ref 70) with added lag times. Stalks form at times t1 = τ1 + t′1; hourglass shapes at times t2 = t1 + τ2 + t2′ ; and fusion pores at t3 = t2 + τ3 + t3′ . The constant lag times τi (i = 1, 2, 3) capture initial relaxation phases. The times ti′ (i = 1, 2, 3) for the three Poisson processes are distributed exponentially with rates ki. The resulting distributions of the times ti are thus p1(t1) = k1e−k1(t1−τ1), p2(t2) = k1k2[e−k1(t2−τ12) − e−k2(t2−τ12)]/(k2 − k1), and

The relative ease with which CNTs can induce fusion, at least of small vesicles, also provides guidance for the design of fusogenic molecules. Our results show that hydrophobic rods penetrating into fusing membranes can dramatically enhance fusion efficiency. For large vesicles, these hydrophobic spikes should be combined with factors that enhance the local membrane curvature to facilitate point-like contacts between local membrane protrusions. Fusion constructs that exploit the mechanisms highlighted here could result in more facile and controlled fusion of vesicles, addressing two major challenges in the targeted delivery of drugs or nucleic acids. CNTs themselves serve as potential tools to direct and regulate fusion of small vesicles. It is also conceivable that biology exploits at least aspects of the powerful downhill CNT fusion mechanism, e.g., by using the hydrophobic surface of rodshaped viral fusion proteins exposed in their extended state59 to connect host and viral membranes.60 Aspects of CNT-mediated fusion may prove relevant for problems ranging from protein reconstitution into membranes to the inhibition of membrane fusion in viral infections.

METHODS System Setup and Simulations. All molecular dynamics simulations were performed using Gromacs 4.5.661−64 with the MARTINI force field (version 2.3).65 Two kinds of systems were set up to study CNT-mediated vesicle fusion. Setup A comprises two vesicles stapled by a CNT (Figure 1a). The vesicles were initially placed at a distance of 16.5 nm from each other, so that the line connecting their centers was parallel to the z-axis of the box. The CNT was then added to connect the interior cavities of the two vesicles, leaving a gap of 1−1.5 nm between the head groups of the outer leaflets of the two vesicles. Lipids overlapping with the CNT envelope were removed. The smaller setup B contains only a single vesicle, with a CNT embedded in such a way that it staples the vesicle with its periodic image along the z-axis (Figure S2). These systems were then solvated with coarse-grained water particles, energy minimized, and equilibrated with position restraints on the CNT under NPT conditions for 5 ns, followed by ∼1 μs production runs. Simulations were performed using a 20 fs time step. Coordinates were saved every 200 ps. The system pressure was held at 1 bar using the Berendsen coupling scheme64 with a coupling constant of 3 ps during the equilibration phase, and the Parrinello−Rahman barostat66 for the production phase with a coupling constant of 12 ps. The compressibility parameter was set to 3 × 10−4 bar−1 during both phases.63 In setup A and setup B, the CNT was placed initially along the z-axis and the pressure was coupled differently along the xy plane and the z-axis using a semi-isotropic barostat. Test simulations with isotropic barostats also showed vesicle fusion events. Temperature was maintained at 295 K throughout the simulation by velocity rescaling with an added noise term to obtain a canonical ensemble.67 The center-of-mass motion of the entire system was removed. The longrange cutoff distance was 1.2 nm with a switching distance of 1 nm and a pair-list distance of 1.4 nm. All stiff bonds and ring systems were constrained using the LINCS algorithm.62 Vesicle Generation. To obtain vesicles with properly balanced numbers of lipids in the inner and outer leaflets, we simulated spontaneous bilayer-to-vesicle transitions using three different types of lipids (POPC, 1-palmitoyl-2-oleoyl-sn-glycero-3-phosphocholine, 16:0−18:1 PC; DOPC, 1,2-dioleoyl-sn-glycero-3-phosphocholine, 18:1 PC; and DSPC, 1,2-dioctadecanoyl-sn-glycero-3-phosphocholine, 18:0 PC) of varying degree of unsaturation. POPC and DOPC are unsaturated lipids with one and two double bonds in their tails, respectively, while DSPC is a completely saturated lipid with no double bonds. Discontinuous square lipid bilayer patches (23 × 23 nm2, 25 × 25 nm2, and 50 × 50 nm2) of lipids were generated using the script insane.py (MARTINI version 2.3),65 immersed in much larger cubic boxes (30 × 30 × 30 nm3, 32 × 32 × 32 nm3, and 60 × 60

p3 (t3) =

k1k 2k 3e−k1(t3− τ123) k1k 2k 3e−k 2(t3− τ123) + (k1 − k 2)(k1 − k 3) (k 2 − k1)(k 2 − k 3) +

k1k 2k 3e−k3(t3− τ123) (k 3 − k1)(k 3 − k 2)

where t1 ≥ τ1, t2 ≥ τ12, and t3 ≥ τ123 with τ12 = τ1 + τ2 and τ123 = τ1 + τ2 + τ3. We determined the characteristic times of forming fusion stalks (τStalk = 1/k1), hourglass shapes (τhourglass = 1/k2), and fusion pores (τFusion = 1/k3) as averages of t1, t2, and t3, and the associated lag times 1278

DOI: 10.1021/acsnano.6b05434 ACS Nano 2017, 11, 1273−1280

Article

ACS Nano (τ1, τ2, and τ3), by sequential fits of the cumulative distribution functions for the above probability densities to the observed waiting time distributions.

(12) Richard, A.; Marchi-Artzner, V.; Lalloz, M.-N.; Brienne, M.-J.; Artzner, F.; Gulik-Krzywicki, T.; Guedeau-Boudeville, M.-A.; Lehn, J.M. Fusogenic Supramolecular Vesicle Systems Induced by Metal Ion Binding to Amphiphilic Ligands. Proc. Natl. Acad. Sci. U. S. A. 2004, 101, 15279−84. (13) Caschera, F.; Sunami, T.; Matsuura, T.; Suzuki, H.; Hanczyc, M. M.; Yomo, T. Programmed Vesicle Fusion Triggers Gene Expression. Langmuir 2011, 27, 13082−13090. (14) Vandenburg, Y. R.; Smith, B. D.; Pérez-Payán, M. N.; Davis, A. P. Non-Leaky Vesicle Fusion and Enhanced Cell Transfection Using a Cationic Facial Amphiphile. J. Am. Chem. Soc. 2000, 122, 3252−3253. (15) Dimova, R.; Riske, K. A.; Aranda, S.; Bezlyepkina, N.; Knorr, R. L.; Lipowsky, R. Giant Vesicles in Electric Fields. Soft Matter 2007, 3, 817. (16) Gao, L.; Lipowsky, R.; Shillcock, J. Tension-Induced Vesicle Fusion: Pathways and Pore Dynamics. Soft Matter 2008, 4, 1208. (17) Rørvig-Lund, A.; Bahadori, A.; Semsey, S.; Bendix, P. M.; Oddershede, L. B. Vesicle Fusion Triggered by Optically Heated Gold Nanoparticles. Nano Lett. 2015, 15, 4183−8. (18) Sommerdijk, N. A. J. M.; Hoeks, T. H. L.; Synak, M.; Feiters, M. C.; Nolte, R. J. M.; Zwanenburg, B. Stereodependent Fusion and Fission of Vesicles: Calcium Binding of Synthetic Gemini Phospholipids Containing Two Phosphate Groups. J. Am. Chem. Soc. 1997, 119, 4338−4344. (19) Kashiwada, A.; Tsuboi, M.; Matsuda, K. Target-Selective Vesicle Fusion Induced by Molecular Recognition on Lipid Bilayers. Chem. Commun. (Cambridge, U. K.) 2009, 695−697. (20) Papahadjopoulos, D.; Allen, T. M.; Gabizon, A.; Mayhew, E.; Matthay, K.; Huang, S. K.; Lee, K. D.; Woodle, M. C.; Lasic, D. D.; Redemann, C. Sterically Stabilized Liposomes: Improvements in Pharmacokinetics and Antitumor Therapeutic Efficacy. Proc. Natl. Acad. Sci. U. S. A. 1991, 88, 11460−4. (21) Shillcock, J. C.; Lipowsky, R. The Computational Route from Bilayer Membranes to Vesicle Fusion. J. Phys.: Condens. Matter 2006, 18, S1191−S1219. (22) Smirnova, Y. G.; Marrink, S. J.; Lipowsky, R.; Knecht, V. Solvent-Exposed Tails as Prestalk Transition States for Membrane Fusion at Low Hydration. J. Am. Chem. Soc. 2010, 132, 6710−6718. (23) Katsov, K.; Müller, M.; Schick, M. Field Theoretic Study of Bilayer Membrane Fusion. I. Hemifusion Mechanism. Biophys. J. 2004, 87, 3277−3290. (24) Katsov, K.; Müller, M.; Schick, M. Field Theoretic Study of Bilayer Membrane Fusion: II. Mechanism of a Stalk-Hole Complex. Biophys. J. 2006, 90, 915−26. (25) Marrink, S. J.; Mark, A. E. The Mechanism of Vesicle Fusion as Revealed by Molecular Dynamics Simulations. J. Am. Chem. Soc. 2003, 125, 11144−11145. (26) Smeijers, A. F.; Pieterse, K.; Markvoort, A. J.; Hilbers, P. A. J. Coarse-Grained Transmembrane Proteins: Hydrophobic Matching, Aggregation, and their Effect on Fusion. J. Phys. Chem. B 2006, 110, 13614−13623. (27) Grafmüller, A.; Shillcock, J.; Lipowsky, R. The Fusion of Membranes and Vesicles: Pathway and Energy Barriers from Dissipative Particle Dynamics. Biophys. J. 2009, 96, 2658−75. (28) Zakaria, A. B.; Picaud, F.; Rattier, T.; Pudlo, M.; Saviot, L.; Chassagnon, R.; Lherminier, J.; Gharbi, T.; Micheau, O.; Herlem, G. Nanovectorization of TRAIL with Single Wall Carbon Nanotubes Enhances Tumor Cell Killing. Nano Lett. 2015, 15, 891−895. (29) Kam, N. W. S.; O’Connell, M.; Wisdom, J. a.; Dai, H. Carbon Nanotubes as Multifunctional Biological Transporters and NearInfrared Agents for Selective Cancer Cell Destruction. Proc. Natl. Acad. Sci. U. S. A. 2005, 102, 11600−5. (30) Lacerda, L.; Ali-Boucetta, H.; Kraszewski, S.; Tarek, M.; Prato, M.; Ramseyer, C.; Kostarelos, K.; Bianco, A. How Do Functionalized Carbon Nanotubes Land on, Bind to and Pierce through Model and Plasma Membranes. Nanoscale 2013, 5, 10242−10250. (31) Zhou, X.; Moran-Mirabal, J. M.; Craighead, H. G.; McEuen, P. L. Supported Lipid Bilayer/Carbon Nanotube Hybrids. Nat. Nanotechnol. 2007, 2, 185−190.

ASSOCIATED CONTENT S Supporting Information *

The Supporting Information is available free of charge on the ACS Publications website at DOI: 10.1021/acsnano.6b05434. Movie S1 (MPG) Movie S2 (MPG) Supplementary Figures S1−S11; Supplementary Texts S1−S4 (PDF)

AUTHOR INFORMATION Corresponding Author

*E-mail: [email protected]. ORCID

Gerhard Hummer: 0000-0001-7768-746X Author Contributions §

R.M.B. and S.M.L. contributed equally to this work. S.L., R.M.B., J.K., and G.H. designed the study. S.L. and R.M.B. performed the simulations and made figures. M.V. designed the coarse-grained CNT model. R.M.B., J.K. and G.H. wrote the manuscript. Notes

The authors declare no competing financial interest.

ACKNOWLEDGMENTS We thank Prof. A. Noy, Dr. A. Bahrami, and Dr. R. Covino for inspiring and insightful discussions. This work was supported by the Max Planck Society. REFERENCES (1) Martens, S.; McMahon, H. T. Mechanisms of Membrane Fusion: Disparate Players and Common Principles. Nat. Rev. Mol. Cell Biol. 2008, 9, 543−556. (2) Bonifacino, J. S.; Glick, B. S. The Mechanisms of Vesicle Budding and Fusion. Cell 2004, 116, 153−66. (3) Kozlov, M. M.; Leikin, S. L.; Chernomordik, L. V.; Markin, V. S.; Chizmadzhev, Y. A. Stalk Mechanism of Vesicle Fusion - Intermixing of Aqueous Contents. Eur. Biophys. J. 1989, 17, 121−129. (4) Chernomordik, L. V.; Kozlov, M. M. Mechanics of Membrane Fusion. Nat. Struct. Mol. Biol. 2008, 15, 675−83. (5) Risselada, H. J.; Kutzner, C.; Grubmüller, H. Caught in the Act: Visualization of SNARE-Mediated Fusion Events in Molecular Detail. ChemBioChem 2011, 12, 1049−55. (6) Vrljic, M.; Strop, P.; Ernst, J. a.; Sutton, R. B.; Chu, S.; Brunger, A. T. Molecular Mechanism of the Synaptotagmin-SNARE Interaction in Ca2+-Triggered Vesicle Fusion. Nat. Struct. Mol. Biol. 2010, 17, 325−331. (7) Baoukina, S.; Tieleman, D. P. Direct Simulation of ProteinMediated Vesicle Fusion: Lung Surfactant Protein B. Biophys. J. 2010, 99, 2134−2142. (8) Earp, L. J.; Delos, S. E.; Park, H. E.; White, J. M. Membrane Trafficking in Viral Replication. Curr. Top. Microbiol. Immunol. 2005, 285, 25−66. (9) Malinin, V. S.; Lentz, B. R. Energetics of Vesicle Fusion Intermediates: Comparison of Calculations with Observed Effects of Osmotic and Curvature Stresses. Biophys. J. 2004, 86, 2951−2964. (10) Malinin, V. S.; Frederik, P.; Lentz, B. R. Osmotic and Curvature Stress Affect PEG-Induced Fusion of Lipid Vesicles but not Mixing of their Lipids. Biophys. J. 2002, 82, 2090−100. (11) Mondal Roy, S.; Sarkar, M. Membrane Fusion Induced by Small Molecules and Ions. J. Lipids 2011, 2011, 528784. 1279

DOI: 10.1021/acsnano.6b05434 ACS Nano 2017, 11, 1273−1280

Article

ACS Nano (32) Gangupomu, V. K.; Capaldi, F. M. Interactions of Carbon Nanotube with Lipid Bilayer Membranes. J. Nanomater. 2011, 2011, 1−6. (33) Kraszewski, S.; Bianco, A.; Tarek, M.; Ramseyer, C. Insertion of Short Amino-Functionalized Single-Walled Carbon Nanotubes into Phospholipid Bilayer Occurs by Passive Diffusion. PLoS One 2012, 7, e40703. (34) Pogodin, S.; Baulin, V. A. Can a Carbon Nanotube Pierce through a Phospholipid Bilayer? ACS Nano 2010, 4, 5293−5300. (35) Geng, J.; Kim, K.; Zhang, J.; Escalada, A.; Tunuguntla, R.; Comolli, L. R.; Allen, F. I.; Shnyrova, A. V.; Cho, K. R.; Munoz, D.; Wang, Y.; Grigoropoulos, C.; Ajo-Franklin, C.; Frolov, V.; Noy, A. Stochastic Transport through Carbon Nanotubes in Lipid Bilayers and Live Cell Membranes. Nature 2014, 514, 612−615. (36) Tran, I. C.; Tunuguntla, R. H.; Kim, K.; Lee, J. R. I.; Willey, T. M.; Weiss, T. M.; Noy, A.; van Buuren, T. Structure of Carbon Nanotube Porins in Lipid Bilayers: An in Situ Small-Angle X-ray Scattering (SAXS) Study. Nano Lett. 2016, 16, 4019−4024. (37) Dellago, C.; Hummer, G. Kinetics and Mechanism of Proton Transport across Membrane Nanopores. Phys. Rev. Lett. 2006, 97, 245901. (38) Tunuguntla, R. H.; Allen, F. I.; Kim, K.; Belliveau, A.; Noy, A. Ultrafast Proton Transport in Sub-1-nm Diameter Carbon Nanotube Porins. Nat. Nanotechnol. 2016, 11, 1−12. (39) Liu, L.; Yang, C.; Zhao, K.; Li, J.; Wu, H.-C. Ultrashort SingleWalled Carbon Nanotubes in a Lipid Bilayer as a New Nanopore Sensor. Nat. Commun. 2013, 4, 2989. (40) Seifert, U.; Berndl, K.; Lipowsky, R. Shape Transformations of Vesicles: Phase Diagram for Spontaneous-Curvature and BilayerCoupling Models. Phys. Rev. A: At., Mol., Opt. Phys. 1991, 44, 1182− 1202. (41) Fusco, G.; Pape, T.; Stephens, A. D.; Mahou, P.; Costa, A. R.; Kaminski, C. F.; Kaminski Schierle, G. S.; Vendruscolo, M.; Veglia, G.; Dobson, C. M.; DeSimone, A. Structural Basis of Synaptic Vesicle Assembly Promoted by α-Synuclein. Nat. Commun. 2016, 7, 12563. (42) Marrink, S. J.; Tieleman, D. P. Perspective on the Martini Model. Chem. Soc. Rev. 2013, 42, 6801−6822. (43) Knecht, V.; Marrink, S.-J. Molecular Dynamics Simulations of Lipid Vesicle Fusion in Atomic Detail. Biophys. J. 2007, 92, 4254− 4261. (44) Risselada, H. J.; Bubnis, G.; Grubmüller, H. Expansion of the Fusion Stalk and its Implication for Biological Membrane Fusion. Proc. Natl. Acad. Sci. U. S. A. 2014, 111, 1−6. (45) Kawamoto, S.; Klein, M. L.; Shinoda, W. Coarse-Grained Molecular Dynamics Study of Membrane Fusion: Curvature Effects on Free Energy Barriers along the Stalk Mechanism. J. Chem. Phys. 2015, 143, 243112. (46) Hummer, G. Position-Dependent Diffusion Coefficients and Free Energies from Bayesian Analysis of Equilibrium and Replica Molecular Dynamics Simulations. New J. Phys. 2005, 7, 34. (47) Höfinger, S.; Melle-Franco, M.; Gallo, T.; Cantelli, A.; Calvaresi, M.; Gomes, J. A.; Zerbetto, F. A Computational Analysis of the Insertion of Carbon Nanotubes into Cellular Membranes. Biomaterials 2011, 32, 7079−7085. (48) Sarukhanyan, E.; De Nicola, A.; Roccatano, D.; Kawakatsu, T.; Milano, G. Spontaneous Insertion of Carbon Nanotube Bundles Inside Biomembranes: A Hybrid Particle-Field Coarse-Grained Molecular Dynamics Study. Chem. Phys. Lett. 2014, 595, 156−166. (49) Dutt, M.; Kuksenok, O.; Little, S. R.; Balazs, A. C. Forming Transmembrane Channels using End-Functionalized Nanotubes. Nanoscale 2011, 3, 240−250. (50) Baoukina, S.; Monticelli, L.; Tieleman, D. P. Interaction of Pristine and Functionalized Carbon Nanotubes with Lipid Membranes. J. Phys. Chem. B 2013, 117, 12113−12123. (51) Wallace, E. J.; Sansom, M. S. P. Carbon Nanotube SelfAssembly with Lipids and Detergent: a Molecular Dynamics Study. Nanotechnology 2009, 20, 045101. (52) Dutt, M.; Kuksenok, O.; Nayhouse, M. J.; Little, S. R.; Balazs, A. C. Modeling the Self-Assembly of Lipids and Nanotubes in Solution:

Forming Vesicles and Bicelles with Transmembrane Nanotube Channels. ACS Nano 2011, 5, 4769−4782. (53) Collins, P. G. In Oxford Handbook of Nanoscience and Technology; Narlikar, A. V., Fu, Y. Y., Eds.; Oxford University Press, 2010; Vol. 2, Chapter 2, pp 31−93. (54) Kinnunen, P. K. J.; Holopainen, J. M. Mechanisms of Initiation of Membrane Fusion: Role of Lipids. Biosci. Rep. 2000, 20, 465−482. (55) Kuzmin, P. I.; Zimmerberg, J.; Chizmadzhev, Y. A.; Cohen, F. S. A Quantitative Model for Membrane Fusion Based on Low-Energy Intermediates. Proc. Natl. Acad. Sci. U. S. A. 2001, 98, 7235−7240. (56) Markin, V. S.; Kozlov, M. M.; Borovjagin, V. L. On the Theory ofMembrane Fusion. The Stalk Mechanism. Gen. Physiol. Biophys. 1984, 3, 361−377. (57) Siegel, D. P. Energetics of Intermediates in Membrane Fusion: Comparison of Stalk and Inverted Micellar Intermediate Mechanisms. Biophys. J. 1993, 65, 2124−2140. (58) Kozlovsky, Y.; Chernomordik, L. V.; Kozlov, M. M. Lipid Intermediates in Membrane Fusion: Formation, Structure, and Decay of Hemifusion Diaphragm. Biophys. J. 2002, 83, 2634−51. (59) Siebert, X.; Hummer, G. Hydrophobicity Maps of the N-peptide Coiled Coil of HIV-1 gp41. Biochemistry 2002, 41, 2956−2961. (60) Harrison, S. C. Viral Membrane Fusion. Virology 2015, 479− 480, 498−507. (61) Pronk, S.; Páll, S.; Schulz, R.; Larsson, P.; Bjelkmar, P.; Apostolov, R.; Shirts, M. R.; Smith, J. C.; Kasson, P. M.; Van Der Spoel, D.; Hess, B.; Lindahl, E. GROMACS 4.5: A High-Throughput and Highly Parallel Open Source Molecular Simulation Toolkit. Bioinformatics 2013, 29, 845−854. (62) Hess, B.; Kutzner, C.; Van Der Spoel, D.; Lindahl, E. GROMACS 4: Algorithms for Highly Efficient, Load-Balanced, and Scalable Molecular Simulation. J. Chem. Theory Comput. 2008, 4, 435− 447. (63) Van Der Spoel, D.; Lindahl, E.; Hess, B.; Groenhof, G.; Mark, A. E.; Berendsen, H. J. C. GROMACS: Fast, Flexible, and Free. J. Comput. Chem. 2005, 26, 1701−1718. (64) Berendsen, H. J. C.; van der Spoel, D.; van Drunen, R. GROMACS: A Message-Passing Parallel Molecular Dynamics Implementation. Comput. Phys. Commun. 1995, 91, 43−56. (65) Marrink, S. J.; Risselada, H. J.; Yefimov, S.; Tieleman, D. P.; De Vries, A. H. The MARTINI Force Field: Coarse Grained Model for Biomolecular Simulations. J. Phys. Chem. B 2007, 111, 7812−7824. (66) Parrinello, M.; Rahman, A. Polymorphic Transitions in Single Crystals: A New Molecular Dynamics Method. J. Appl. Phys. 1981, 52, 7182−7190. (67) Bussi, G.; Donadio, D.; Parrinello, M. Canonical Sampling through Velocity Rescaling. J. Chem. Phys. 2007, 126, 014101. (68) Wong-Ekkabut, J.; Baoukina, S.; Triampo, W.; Tang, I.-M.; Tieleman, D. P.; Monticelli, L. Computer Simulation Study of Fullerene Translocation through Lipid Membranes. Nat. Nanotechnol. 2008, 3, 363−368. (69) Vögele, M.; Hummer, G. Divergent Diffusion Coefficients in Simulations of Fluids and Lipid Membranes. J. Phys. Chem. B 2016, 120, 8722−8732. PMID: 27385207. (70) Zhou, Y.; Zhuang, X. Kinetic Analysis of Sequential Multistep Reactions. J. Phys. Chem. B 2007, 111, 13600−13610.

1280

DOI: 10.1021/acsnano.6b05434 ACS Nano 2017, 11, 1273−1280