Molecular dynamics simulations of anchored viral peptide interactions Tyrone J. YacoubIgal Szleifer
Citation: Biointerphases 10, 029513 (2015); doi: 10.1116/1.4919408 View online: http://dx.doi.org/10.1116/1.4919408 View Table of Contents: http://avs.scitation.org/toc/bip/10/2 Published by the American Vacuum Society
Molecular dynamics simulations of anchored viral peptide interactions Tyrone J. Yacoub Department of Chemical and Biological Engineering, Northwestern University, Evanston, Illinois 60208
Igal Szleifera) Department of Biomedical Engineering, Department of Chemistry, Department of Chemical and Biological Engineering, and Chemistry of Life Processes Institute, Northwestern University, Evanston, Illinois 60208
(Received 15 March 2015; accepted 20 April 2015; published 1 May 2015) The authors use molecular dynamics simulations to investigate viral peptide interactions as the cause of pH-dependent fusion in liposomal drug delivery. Viral peptides (LEFN) are composed of a linker peptide (LELELELE) connected to a synthetic viral peptide (DRGWGNGCGLFGKGSI). Rather than being anchored in a lipid bilayer, the viral peptides are anchored to a neutral surface by the amino termini of the linker peptide (anchor atoms are mobile in the xy-plane). Atomistic-level peptide pair arrangement on a surface depends on pH; however, the overall propensity to cluster is independent of pH, indicating that pH-sensitive liposome fusion is not due to peptide clustering. To further investigate a molecular cause of pH-sensitive fusion, the authors treat the linker peptides as ectodomains, with the assumption that the viral peptides are already inserted into a target membrane. In these simulations, the linker peptides are elongated to encourage them to bundle. At both high and low pH, the peptides readily bundle. At high pH, however, bundling was constrained by long-range order induced by sodium ions bridging negatively charged glutamic acid residues on neighboring peptides. The authors hypothesize that this constraint hinders the ability of the linker peptides to support viral peptide insertion, resulting in decreased levels of fusion observed experiC 2015 American Vacuum Society. [http://dx.doi.org/10.1116/1.4919408] mentally. V
I. INTRODUCTION Liposomes have shown great promise in the field of drug delivery, due in large part to the ease of adding surface modifications.1–6 A wide variety of molecules can be anchored to a membrane by the attachment of lipid tails. Popular examples include poly(ethylene glycol),7–9 which increases circulation time, and folate, which targets folate receptors on cancer cells.10–12 In this work, we investigate the molecular interactions between anchored viral peptides, which are used to fuse the liposome with an endosomal membrane and release therapeutics into the cytoplasm.13–15 Understanding the effect of pH on peptide interactions will prove useful in the design of viral liposomes, as the pH drops significantly from physiological (7.5) to endosomal (4.5). Here, we use molecular dynamics simulations to study synthetic viral peptides coupled to pH-sensitive linker domains, which are chosen based on experimental evidence16 that they induce pH-sensitive liposome fusion, a promising indicator of their ability to fuse with a tumor endosome and allow therapeutics to diffuse readily into the cytoplasm. It is known that clustering of flavivirus fusion peptides is crucial in the development of fusion pores.17 Experimentally, it has been observed that peptide clustering occurs during, rather than before, the fusion process.18 Atomistic molecular dynamics simulations have shown the propensity of short (4 amino acid) lipopeptides to cluster in a lipid membrane,19 but it is unknown how longer peptides,
a)
Author to whom correspondence should be addressed; electronic mail: [email protected]
029513-1 Biointerphases 10(2), June 2015
and specifically viral peptides, will interact with each other at the molecular level. On a fundamental level, we are interested in characterizing surface interactions between molecules with many degrees of freedom. The peptides (denoted “LEFN”) studied in this work are anchored to a surface by a linker peptide (LELELELE, “LE”) connected to a synthetic flavivirus native peptide (DRGWGNGCGLFGKGSI, “FN”), as depicted in Fig. 1. The LE peptide was chosen based on the work of Cavalli et al., who showed that lipopeptides with the LE sequence form antiparallel b-sheets in a monolayer at low pH.20 This mechanism may then be exploited to cause pH-sensitive peptide clustering, purportedly leading to increased levels of fusion. Here, we discuss whether this phenomenon indeed explains pH-sensitive fusion. As we are concerned with the effect of pH on peptide interactions, the main difference in our simulations between “high” and “low” pH is that the glutamic acid residues of the linker peptide are either fully protonated (at endosomal pH) or fully deprotonated (at physiological pH), according to a pKa of 4.3 (despite endosomal pH of closer to 4.5).21 Although local pH is realistically different from these idealized protonation states, we are interested in observing the most dramatic possible changes in peptide–peptide interaction based on pH. To further investigate a cause for pH-sensitive liposome fusion, we study the organization of the pH-sensitive linker peptides in the ectodomain, as if they are facilitating fusion by supporting the viral peptides. Once the viral peptides insert into a target membrane, their activity is dependent not only on peptide clustering, but on their insertion angle, penetration depth, etc., which are influenced by the organization
1934-8630/2013/10(2)/029513/8/$30.00
C 2015 American Vacuum Society V
029513-1
029513-2 T. J. Yacoub and I. Szleifer: Molecular dynamics simulations
029513-2
FIG. 1. Atomistic representation of the LEFN peptide. Carbon is shown in teal, oxygen in red, hydrogen in white, nitrogen in blue, and sulfur in yellow. Peptides are anchored to a surface by the nitrogen terminus of first leucine residue. pH was simulated by fully (de)protonating glutamic acid (E) residues, where they are neutral at endosomal pH and negatively charged at physiological pH. Snapshot to right shows two LEFN peptides, with anchor atoms (blue) enlarged. The surface, shown enlarged in transparent black, is an 8 nm 8 nm grid of neutral dummy particles, with spacing 0.2 nm. Peptide anchors are 0.4 nm above the surface.
of the supporting ectodomains.22–24 We simulate the linker peptides as ectodomains, as if the viral peptides are already inserted into a target membrane, and observe their tendency to bundle and interact in a manner much different from the surface-bound peptides. By elongating the linker peptides to encourage bundling, charged groups are exposed and sensitive to the influence of counterions, yielding interesting pHsensitive interactions. II. METHODS A. Molecular dynamics simulations
Atomistic MD simulations of peptides on a surface consist of LEFN peptides anchored (fixed) in the z-dimension by the amino terminus (see Fig. 1), 0.4 nm above a surface, but unconstrained in the xy-plane. The snapshot on the right of Fig. 1 shows the surface as a square lattice of lattice size 0.2 nm, made of neutral dummy atoms fixed in all dimensions with a purely repulsive interaction, serving only as a physical barrier preventing peptide penetration. The reasons for not explicitly simulating the lipid bilayer include: (1) there is no force field optimized for both lipids and proteins (simultaneously), (2) we can access much longer time scales due to increased peptide diffusion, and (3) we are focusing on peptide–peptide interactions. Atomistic molecular dynamics simulations were performed using GROMACS 4.5.5,25 and the OPLS force field26 optimized for proteins. The simple point charge model27 was used for water. Velocity-rescaling temperature-coupling with st ¼ 0.4 ps and T ¼ 323 K, and semi-isotropic Parrinello-Rahman pressure-coupling at 1 atm, with sp ¼ 6 ps were used. This temperature was chosen for consistency with simulations of Biointerphases, Vol. 10, No. 2, June 2015
lipid bilayers in the liquid-disordered state. A time step of 1 fs was used, along with nonbonded interaction cutoffs of 1.6 nm. All systems were equilibrated with steepest descent before molecular dynamics runs. LEFN peptides were created using the Biomer (B) program.28 The amino acid sequence of the viral portion of the peptide, denoted FN, is DRGWGNGCGLFGKGSI. The FN peptide was directly attached to an LELELELE linker domain, denoted LE, as depicted in Fig. 1. LEFN is anchored to the surface via the amino end-terminus in the first leucine (L) residue of the linker peptide. pH is simulated by fully (de)protonating the oxygens of the carboxylic acids on glutamic acid (E) residues, which at “high pH” have partial charge 0.8, and at “low pH” have partial charge 0.3, where partial charge is the elementary charge unit. Realistically, the partial charges on these residues would be dependent on the local environment, but since we are interested in observing the effect of pH on peptide organization, we use these idealized (most extreme) values for partial charge. Arginine and Lysine are charged in the simulations, as the pKa values of their side chains are 12.5 and 10.5, respectively, which are well above the pH being considered.21 Simulations with two or three peptides had dimensions of 8 8 3.8 nm3. Those with 16 peptides had dimensions of 16 16 3.8 nm3. Peptide pair simulations began with peptides completely separated (no contact), and were run for at least 700 ns at both high and low pH. The three-peptide simulation consisted of a third peptide added to two peptides equilibrated from the previously mentioned simulation, and was run for 200 ns at low pH. Sixteen peptides were initially completely separated, and run for 110 ns at low pH.
029513-3 T. J. Yacoub and I. Szleifer: Molecular dynamics simulations
B. Ectodomain bundling
Ectodomain bundling was simulated by assuming that the LE peptides are elongated between the viral membrane and the target membrane, as if the viral peptides were already inserted into the target membrane. Therefore, we only included the LE peptides and solvent. Figure 2 shows a schematic of the simulations, with the inset to the right showing a side view of the high pH system with counterions, and the inset to the left showing a fusion domain, with the ectodomain bundle between (and not directly interacting with) the viral and target membranes. Simulations with 64 LE domains were run at high and low pH for 170 ns, the only differences being the partial charge of glutamic acid residues (see above), and the inclusion of 256 sodium counterions for charge neutrality at high pH (sodium ions are not used in the low pH system). The domains are elongated by fixing the first and last atoms in the z-dimension, and oriented vertically so that the two atoms have the same coordinates in the xy-plane, meaning the bulk of peptide movement is twodimensional. Bundling simulations are run with the same molecular dynamics settings described above for peptide clustering. The simulation box is 12 12 4.3 nm3, and contains around 17 000 water molecules. III. RESULTS A. Peptide pair-interactions
Anchored peptide pairs were run for over 700 ns at both high and low pH, where pH is defined by the protonation state of glutamic acid residues. At low pH, LEFN peptides
029513-3
immediately cluster and reach an anchor distance of r ¼ 1.5 nm [shown in black in Fig. 3]. At t ¼ 40 ns, the snapshot to the right shows the peptides aligning, where the backbone hydrogen–oxygen bond distance between the second “L” residues (shown in green in the figure) being approximately 0.25 nm. After 50 ns, the peptides transition into a local free energy minimum with an anchor distance of r ¼ 1 nm. This configuration is short-lived, however, as the peptides enter a side-by-side arrangement (stacked, not like a b-sheet) at t ¼ 100 ns. The peptides continue to transition to different conformations, as depicted in the snapshots in Fig. 3. At t ¼ 200 ns and beyond, the linker peptides do not exhibit any apparent order. This observation demonstrates that while the LE peptides appear to form distinct structures at low pH, these are just a few of many potential structures corresponding to local free energy minima. According to Fig. 3, peptides at low pH appear to make sharp transitions from one conformation to the next. At high pH (Fig. 4), by contrast, anchors do not make sharp transitions and peptides do not exhibit transient structure, as evidenced by the high variability of anchor distance as compared to low pH. There is the potential for a stabilizing disulfide bond between Cys atoms on neighboring LEFN peptides, which is represented by the distance between sulfur atoms in the low pH system in Fig. S4.29 Although there appear to be periods of nearly constant distance (on the order of 10 s of ns), the distance between sulfur atoms does not approach the disulfide bond ˚. distance of 2.05 A Figure 5 represents the data from Figs. 3 and 4 as histograms of anchor distance. In red, anchors at low pH do not
FIG. 2. Schematic of peptide bundling simulations. LE linker peptides are elongated upright (in the z-direction) by fixing the first and last atoms in the z-dimension. Viral peptides are assumed to be inserted into the target cell (i.e., endosome). Inset to the left adapted from Ref. 22. Snapshot to the right shows a side view of a simulation at high pH, where atom colors correspond to Fig. 1 and blue spheres are sodium counterions. Biointerphases, Vol. 10, No. 2, June 2015
029513-4 T. J. Yacoub and I. Szleifer: Molecular dynamics simulations
FIG. 3. Low pH. Distance between a pair of LEFN anchor atoms (black) and a potential hydrogen bond between backbone atoms of the second leucine residues: nitrogen of one peptide and oxygen of the other (green) over more than 700 ns. Snapshots show linker peptide orientation at various time points (viral peptide hidden for clarity). Note distinct structures at 40 and 100 ns. Large spheres represent anchors, and small spheres represent the potential NH–O bond described above. Peptides began fully separated.
separate by more than 2.5 nm. The maximum (corresponding to a free energy minimum) is clearly located around r ¼ 1 nm with a smaller maximum around r ¼ 1.4 nm. At high pH, maxima of comparable magnitude exist at r ¼ 1 nm and r ¼ 2 nm, which correspond to free energy minima of equal magnitude. Thus, both high and low pH peptides show a maximum at an anchor distance near 1 nm, while the
029513-4
FIG. 5. Histogram of anchor distance between peptide pairs for over 700 ns at both high and low pH, representing the data from Figs. 3 and 4. Maxima correspond to most likely distance between anchor atoms.
location of the second maximum is different, as is the relative magnitude of the maxima. Figures 3–5 show that while the specific interactions between the pH-sensitive linker peptides (locations of most likely anchor distance) are clearly different between the high and low pH cases, charge does not have a qualitative effect on clustering, as peptides at either pH do not separate beyond r ¼ 3.5 nm. Repulsion between negatively charged glutamic acid residues at high pH is not strong enough to completely separate the peptides. B. Multipeptide interactions 1. Adding a third peptide
FIG. 4. High pH. Distance between a pair of LEFN anchor atoms (black) and a potential hydrogen bond between backbone atoms of the second leucine residues: nitrogen of one peptide and oxygen of the other (green) over 700 nsþ. Unlike peptides at low pH (Fig. 3), those at high pH did not exhibit clear structural transitions. Peptides began fully separated. Biointerphases, Vol. 10, No. 2, June 2015
The energetics of peptide clustering were further studied by adding a third peptide to two peptides (at low pH), which had reached an anchor distance of r ¼ 1 nm, corresponding to the maximum in Fig. 5. Clustering was measured with a histogram of peptide anchor distances (Fig. 6). Adding a third peptide did not significantly disrupt the strong interaction between the two original peptides (shown in gray in Fig. 6, over the course of 300 ns). Interestingly, the most probable anchor distance between the original peptides actually decreases, evidenced by a shift inward in the peak from r ¼ 1 nm to r ¼ 0.8 nm, and the emergence of a peak around r ¼ 0.5 nm. In the blue and red plots of Fig. 6, it is evident that the third peptide remained strongly attached over the 200 ns simulation. The third peptide only had 1.5 nm of mobility with respect to its anchor distance with the other peptides (the approximate “spread” of the blue and red plots), and was never fully released from the cluster as it did not separate from the first peptide by more than 2.5 nm. Although the third peptide is strongly bound, the interaction strength between it and the original peptides is significantly different. As an estimation of this difference, we can compare the maxima of the
029513-5 T. J. Yacoub and I. Szleifer: Molecular dynamics simulations
FIG. 6. Histogram of anchor distance within three peptides. A third peptide (pep3) was added to two peptides (pep1 and pep2) with an anchor distance of 1 nm. pep1–pep2 is the distance between the original peptide anchors. pep1–pep3 and pep2–pep3 are distances between each original peptide and the added third peptide.
histograms in Fig. 6. For the original two peptides, the maximum is approximately three times higher than that of the third peptide and either of the original peptides. By treating these values as probabilities, we can estimate the free energy difference as DG ¼ kTln(3) or approximately 1 kT. Given sufficient time, these interactions may change. For example the first peptide anchor may approach the third anchor as it weakens its interaction with the second peptide. However, Fig. 6 shows obvious asymmetry in peptide cluster interactions. It is clear that creating a simple model of peptide interaction, for example, by representing the peptides as spheres with an effective pair interaction potential, will be qualitatively inaccurate. Therefore, to accurately model multibody peptide interactions, one would need to construct a multibody interaction potential.
029513-5
FIG. 7. Representative snapshot of 16-peptide cluster, after 60 ns. Anchor atoms shown enlarged. Beginning with 16 separated peptides, this cluster began to form immediately, and remained intact for the duration of the 110 ns simulation.
peptides remained clustered over the course of the 110 ns simulation. C. Ectodomain bundling
LEFN is a viral peptide attached to a linker peptide and is used specifically to cause fusion in liposomal drug delivery, as we have observed in unpublished experiments.16 The liposome fusion experiments show a distinct sensitivity of fusion to pH, favoring higher levels of fusion at lower pH. Here, we aim to understand the physical cause. Since the viral peptide is not significantly pH sensitive (considering the pKa of its side chains), we focus on the pH-sensitive linker peptide. Based on several studies, the domains supporting viral peptide insertion, which here are assumed to be the linker peptides, are elongated in bundles.22–24 Therefore, we studied
2. Cluster of 16 peptides
To observe peptide interactions within a larger cluster of peptides, we attached 16 LEFN peptides at low pH to a 16 nm 16 nm surface. The peptides began to aggregate within a few nanoseconds, as expected from the two- and three-peptide simulations. Figure 7 shows a snapshot of a 16-peptide cluster after 60 ns, displaying further evidence of the strong clustering propensity of the peptides. Figure 8 shows a histogram of the distance between each anchor and its nearest neighboring anchor in the cluster. The first 60 ns was excluded to allow time for peptides to form a single cluster (see Fig. S1).29 Like the pair peptide simulations (Fig. 5), a maximum around r ¼ 1 nm and a smaller maximum around r ¼ 1.5 nm are observed. Significantly, a clear third maximum around r ¼ 0.5 nm is observed, and is more pronounced than in the three-peptide simulations (Fig. 6), indicating again that additional peptides tend to decrease the distance between anchors and tighten peptide clusters. The Biointerphases, Vol. 10, No. 2, June 2015
FIG. 8. Histogram of distances between nearest-neighbor anchors in a 16peptide cluster over the final 50 ns of a 110 ns simulation (Fig. 7). The maxi˚ is more pronounced than in the cluster of three peptides mum at r ¼ 5 A (Fig. 6).
029513-6 T. J. Yacoub and I. Szleifer: Molecular dynamics simulations
029513-6
FIG. 9. Sixty four linker peptides at high pH (a) and low pH (b) in the xy-plane. Inset shows side view of (a). Peptides are elongated in the z-dimension. Sodium counterions at high pH are blue. Note the long-range order induced by sodium ions bridging negatively charged linker peptides at high pH.
the interaction between linker peptides in water. To encourage the bundle structure, the first and last atoms of the peptides were fixed in the z-direction, with backbones fully elongated and upright, all at the same height. The only movement (beyond vibrational and slight torsional motion) is in the xy-plane. We find that at low pH, the peptides bundle occasionally forming what appear as b-sheets (hydrogen bonds between backbone atoms). Interestingly, at high pH, Fig. 9(a) shows sodium ions, in blue, bridging the negatively charged glutamic acid residues (see Fig. 1), inducing long-range domains resembling ion channels. These domains remained structurally unchanged following their creation, for the remainder of the 170 ns simulation. Based on our observations, we expect the bundles at high pH to maintain some form of long-range order indefinitely [Fig. 9(a)]. At low pH [Fig. 9(b)], on the other hand, the bundles slowly aggregate over time, and we expect them to fully aggregate (into a single large bundle) after another 100–200 ns. Ectodomain bundling is therefore qualitatively dependent on pH, suggesting a physical mechanism to explain pH-sensitive liposome fusion. Figure 10 shows the radial distribution functions for the first atoms (fixed in z) of each elongated linker peptide, representing peptide organization in the xy-plane. There is
FIG. 10. Radial distribution function for base atoms of the ectodomains at high and low pH. Note the third peak around r ¼ 1.2 nm at high pH, reflecting a long-range order not present at low pH. Biointerphases, Vol. 10, No. 2, June 2015
clearly long-range (1.2 nm) ordering present at high pH and not low pH. Without knowing the particular ectodomain structure needed to support LEFN peptide insertion, it is conceivable that LE peptides at high pH are physically constrained by sodium counterions, potentially preventing them from achieving this structure. IV. DISCUSSION We have likely observed a few of many possible peptide cluster conformations. In fact, as represented in Fig. 3, at low pH, the peptides continuously engage in complex binding/unbinding events, signified by clear transitions in peptide structure. However, based on the extensive simulations presented, we strongly believe that the observed peptide conformations are representative of the numerous possible conformations, meaning that peptide clustering will occur regardless of pH (and is therefore not the cause of pH-sensitive liposome fusion). This expectation is due to the fact that, after several hundred nanoseconds of simulation time, the peptides have shown no sign of separation. There are two main goals of this work: (1) to characterize the interactions between surface-bound molecules with many degrees of freedom, and (2) to discuss the relationship between molecular interactions and the pH-sensitive liposome fusion observed in our experimental work.16 We observe definitively that LEFN peptides consistently and quickly cluster, as evidenced in the pair peptide simulations (Figs. 3–5), simulations with a third peptide (Fig. 6), and simulations with 16 peptides (Fig. 7). Adding a third peptide to a pair resulted in a decrease in the anchor distance between the original pair, as evidenced by the maximum in Fig. 6 at r ¼ 0.5 nm, which is much greater than that of the pair simulations. Simulations of 16 peptides made this even more evident, as the maximum at r ¼ 0.5 nm was more pronounced, and represented a free energy minimum as deep as the one at r ¼ 1 nm. These multipeptide simulations indicate that additional peptides tighten peptide clusters in a cooperative effect (more peptides translates to tighter clusters). The effect may be inversely proportional to the size of the cluster, as a very large cluster should not be as sensitive to the influence of additional peptides.
029513-7 T. J. Yacoub and I. Szleifer: Molecular dynamics simulations
Importantly, these cooperative effects imply that simple pair interaction potentials cannot sufficiently describe or predict the behavior of peptide clusters. This became more clear as we attempted to calculate a potential of mean force between two LEFN peptides at low pH, using two methods. The first was to place the peptide anchors at 0.1 nm intervals and anneal them to ensure complete separation and attempt a relatively path-independent calculation. The free energy minimum was found to be at an anchor distance of 2 nm, in qualitative disagreement with the free peptide simulations (Fig. 3). The second method was to pull the peptides from an equilibrated position to obtain conformations, also at 0.1 nm intervals. This clearly path-dependent calculation yielded a free energy minimum near 1 nm (in agreement with Fig. 3); however, the free energy barrier vanished as the peptides relaxed. Without the ability to freely rearrange, the peptides were limited in conformational freedom, inevitably resulting in a biased representation of entanglement. These observations emphasize the difficulty in studying molecules with many degrees of freedom, reinforcing the conclusion that pair interactions cannot be represented by a simple potential and are thus insufficient to describe peptide clustering. It is useful to discuss our approximations and their potential influence on the results. In addition to the issues addressed above and the general limitations associated with molecular dynamics in biology,30 our main approximation is that of a lipid bilayer as a neutral, noninteracting surface. We use this approximation because simulations can be run for longer times, peptides can diffuse faster, and we are focusing on peptide–peptide interactions, which are optimized by the OPLS force field. Although the OPLS force field has also been optimized for lipids, it has not been optimized for both lipids and proteins simultaneously. Whereas a lipid bilayer would surely be needed to study peptide insertion and fusion domains, we are interested in peptide interaction on the surface of the liposome. Peptide cluster mobility may depend on the interplay between peptide cluster size and lipid membrane rigidity, and lateral peptide mobility will depend on lipid–peptide interactions; however, simulations of larger clusters in the full membrane environment would be necessary to elucidate these effects. An additional limitation is that lipids may encourage peptides to interact or they may screen peptide interactions. From preliminary (unpublished) results, coarse-grained simulations of the full membrane-lipopeptide system show peptides clustering on the membrane surface and not inserting into the membrane after 1 ls, somewhat justifying our use of a flat surface and supporting the conclusion that the peptides tend to cluster. Indeed, Pecheur et al. have shown experimentally that while there is a quantitative dependence of fusion on lipid membrane composition, the qualitative dependence is weak. Our results are in agreement with Pecheur et al., who have shown that fusion peptide clustering and oligomerization is not a prerequisite for pH-sensitive fusion;18 therefore, fusion is likely the cause rather than the effect of peptide clustering. While this is in line with our conclusion that clustering is not responsible for pH-sensitive fusion, we expect LEFN Biointerphases, Vol. 10, No. 2, June 2015
029513-7
peptides to cluster readily, whereas Pecheur et al. found that their peptides (which had 11 amino acids and were amphipathic and negatively charged) were driven to cluster by the fusion process.18 To further search for a molecular-level explanation for the pH-sensitive fusion found experimentally,16 we look to the role of the ectodomain in support of viral peptide insertion in the target membrane. The most important assumption is that the LE linker peptides are able to serve this function. In simulations of LE ectodomains, linker peptides are elongated in the z-dimension to simulate their extension between the liposomal membrane and the target membrane (see Fig. 2). Figure 9(b) shows that at low pH, linker peptides form more compact bundles. Interestingly, we find that at high pH [Fig. 9(a)], sodium counterions induce a bridging effect between negatively charged glutamic acid residues in vertically oriented linker peptides. Due to the significant pH-sensitivity of this phenomenon, we believe this long-range order is at least partially responsible for the pH-sensitivity of liposome fusion. At high pH, the linker ectodomains are constrained in their ability to form a tight bundle, translating into greater difficulty in providing proper support to the viral peptide, which is crucial in the specific interactions involved with insertion and orientation of viral peptides in a target membrane.22–24 Often only a small number of peptides (i.e., three in the inset of Fig. 2) may be needed to support viral peptide insertion, but by including many peptides we achieve better sampling to characterize possible effects of pH. Although peptide clustering appears to be independent of pH, bundling of the LE ectodomains is clearly pH-dependent in a qualitative manner. This constraint is a possible explanation of the fact that the viral peptides cause less fusion at higher pH. V. CONCLUSION From a fundamental standpoint, we have characterized the interactions between two molecules with many degrees of freedom anchored to a surface. Specifically, we have studied LEFN peptides composed of a pH-sensitive linker peptide attached to a synthetic flavivirus-type viral peptide. The tendency of the peptides to cluster is independent of pH. At both high and low pH, peptide pairs do not separate on a time scale of 700 ns. We conclude that pH-sensitive membrane fusion is not the result of peptide clustering. This result agrees with experimental findings in that clustering was not found to cause fusion, but disagrees with the same findings as we predict peptides to cluster readily whereas the experiments showed that the fusion itself process causes clustering. However, the peptides used experimentally were significantly different, and less than half the length of LEFN.18 Interestingly, although all simulations showed that the LEFN peptides tend to cluster, we find that the distance between peptide anchors in a cluster cannot be predicted by studying a peptide pair alone. Adding a third peptide decreases the anchor distance between two previously interacting peptides. The anchors in a cluster of 16 peptides are
029513-8 T. J. Yacoub and I. Szleifer: Molecular dynamics simulations
also more closely bound than those in a pair. Additionally, the role of the linker peptides in the ectodomain was investigated. When elongated, linker peptides formed bundles at both high and low pH. However, at high pH, sodium counterions bridge negatively charged glutamic acid residues, inducing long-range order, creating the appearance of an ion channel [Fig. 9(a)] and constraining their ability to achieve the necessary organization to support viral peptide insertion. We hypothesize that this constraint on the organization of the linker peptide ectodomains is a significant contribution to the decreased liposome fusion observed at higher pH.16 We have shed light on the molecular-level clustering of surface-bound molecules, a poorly understood yet ubiquitous occurrence in liposomal drug delivery. This molecular organization is much more complex than the homogeneous distribution of molecules normally depicted, as clustering and multibody interactions have shown. Despite their significant approximations, the methods developed in this work can be used to study virtually any anchored molecules or ectodomains. ACKNOWLEDGMENTS This work was supported by NSF Grant No. CBET1403058. This research was supported in part through the computational resources and staff contributions provided for the Quest high performance computing facility at Northwestern University which is jointly supported by the Office of the Provost, the Office for Research, and Northwestern University Information Technology. 1
A. Samad, Y. Sultana, and M. Aqil, Curr. Drug Delivery 4, 297 (2007). A. Gabizon, R. Catane, B. Uziely, B. Kaufman, T. Safra, R. Cohen, F. Martin, A. Huang, and Y. Barenholz, Cancer Res. 54, 987 (1994). 3 A. Sharma and U. S. Sharma, Int. J. Pharm. 154, 123 (1997). 4 S. Ganta, H. Devalapally, A. Shahiwala, and M. Amiji, J. Controlled Release 126, 187 (2008). 2
Biointerphases, Vol. 10, No. 2, June 2015
029513-8 5
T. M. Allen and P. R. Cullis, Adv. Drug Delivery Rev. 65, 36 (2013). K. Maruyama, Adv. Drug Delivery Rev. 63, 161 (2011). 7 K. Knop, R. Hoogenboom, D. Fischer, and U. S. Schubert, Angew. Chem. Int. Ed. 49, 6288 (2010). 8 M. L. Immordino, F. Dosio, and L. Cattel, Int. J. Nanomed. 1, 297 (2006). 9 S. M. Moghimi and J. Szebeni, Prog. Lipid Res. 42, 463 (2003). 10 Y. Lu and P. S. Low, Adv. Drug Delivery Rev. 64, 342 (2012). 11 A. Gabizon, H. Shmeeda, A. Horowitz, and S. Zalipsky, Adv. Drug Delivery Rev. 56, 1177 (2004). 12 J. Sudimack and R. J. Lee, Adv. Drug Delivery Rev. 41, 147 (2000). 13 T. Kakudo, S. Chaki, S. Futaki, I. Nakase, K. Akaji, T. Kawakami, K. Maruyama, H. Kamiya, and H. Harashima, Biochemistry 43, 5618 (2004). 14 A. Bolhassani, Biochim. Biophys. Acta, Rev. Cancer 1816, 232 (2011). 15 E. Mastrobattista, G. A. Koning, L. van Bloois, A. Filipe, W. Jiskoot, and G. Storm, J. Biol. Chem. 277, 27135 (2002). 16 T. J. Yacoub, “Molecular interactions in liposomal drug delivery,” Ph.D. thesis (Northwestern University, 2014). 17 Y. Mendes et al., PLoS One 7, e47596 (2012). 18 E. I. Pecheur, J. Sainte-Marie, A. Bienven€ ue, and D. Hoekstra, Biochem. J. 38, 364 (1999). 19 J. N. Horn, T. D. Romo, and A. Grossfield, Biochem. J. 52, 5604 (2013). 20 S. Cavalli et al., J. Am. Chem. Soc. 128, 13959 (2006). 21 R. H. Garrett and C. M. Grisham, Principles of Biochemistry: With a Human Focus (Thomson Brooks/Cole, Toronto, 2001). 22 J. L€ uneberg, I. Martin, F. N€ ußler, J.-M. Ruysschaert, and A. Herrmann, J. Biol. Chem. 270, 27606 (1995). 23 R. M. Epand, Y. Shai, J. P. Segrest, and G. M. Anantharamiah, Biopolymers 37, 319 (1995). 24 J. E. Donald, Y. Zhang, G. Fiorin, V. Carnevale, D. R. Slochower, F. Gai, M. L. Klein, and W. F. DeGrado, PNAS 108, 3958 (2011). 25 B. Hess, C. Kutzner, D. van der Spoel, and E. Lindahl, J. Chem. Theory Comput. 4, 435 (2008). 26 W. D. Cornell, J. Am. Chem. Soc. 117, 5179 (1995). 27 H. J. C. Berendsen, J. P. M. Postma, W. F. van Gunsteren, and J. Hermans, Intermol. Forces 11, 331 (1981). 28 N. B. White, “B home page” http://casegroup.rutgers.edu/Biomer/. 29 See supplementary material at http://dx.doi.org/10.1116/1.4919408 for four figures, (1) anchor distance in the 16-peptide cluster, (2) more interatomic distances in the clustered linker peptides at low pH, (3) sodium ion radial distribution functions over time in the ectodomain simulation at high pH, and (4) potential di-sulfide bond distance. 30 J. D. Durrant and J. A. McCammon, BMC Biol. 9, 71 (2011). 6
Citation: Biointerphases 10, 029513 (2015); doi: 10.1116/1.4919408 View online: http://dx.doi.org/10.1116/1.4919408 View Table of Contents: http://avs.scitation.org/toc/bip/10/2 Published by the American Vacuum Society
Molecular dynamics simulations of anchored viral peptide interactions Tyrone J. Yacoub Department of Chemical and Biological Engineering, Northwestern University, Evanston, Illinois 60208
Igal Szleifera) Department of Biomedical Engineering, Department of Chemistry, Department of Chemical and Biological Engineering, and Chemistry of Life Processes Institute, Northwestern University, Evanston, Illinois 60208
(Received 15 March 2015; accepted 20 April 2015; published 1 May 2015) The authors use molecular dynamics simulations to investigate viral peptide interactions as the cause of pH-dependent fusion in liposomal drug delivery. Viral peptides (LEFN) are composed of a linker peptide (LELELELE) connected to a synthetic viral peptide (DRGWGNGCGLFGKGSI). Rather than being anchored in a lipid bilayer, the viral peptides are anchored to a neutral surface by the amino termini of the linker peptide (anchor atoms are mobile in the xy-plane). Atomistic-level peptide pair arrangement on a surface depends on pH; however, the overall propensity to cluster is independent of pH, indicating that pH-sensitive liposome fusion is not due to peptide clustering. To further investigate a molecular cause of pH-sensitive fusion, the authors treat the linker peptides as ectodomains, with the assumption that the viral peptides are already inserted into a target membrane. In these simulations, the linker peptides are elongated to encourage them to bundle. At both high and low pH, the peptides readily bundle. At high pH, however, bundling was constrained by long-range order induced by sodium ions bridging negatively charged glutamic acid residues on neighboring peptides. The authors hypothesize that this constraint hinders the ability of the linker peptides to support viral peptide insertion, resulting in decreased levels of fusion observed experiC 2015 American Vacuum Society. [http://dx.doi.org/10.1116/1.4919408] mentally. V
I. INTRODUCTION Liposomes have shown great promise in the field of drug delivery, due in large part to the ease of adding surface modifications.1–6 A wide variety of molecules can be anchored to a membrane by the attachment of lipid tails. Popular examples include poly(ethylene glycol),7–9 which increases circulation time, and folate, which targets folate receptors on cancer cells.10–12 In this work, we investigate the molecular interactions between anchored viral peptides, which are used to fuse the liposome with an endosomal membrane and release therapeutics into the cytoplasm.13–15 Understanding the effect of pH on peptide interactions will prove useful in the design of viral liposomes, as the pH drops significantly from physiological (7.5) to endosomal (4.5). Here, we use molecular dynamics simulations to study synthetic viral peptides coupled to pH-sensitive linker domains, which are chosen based on experimental evidence16 that they induce pH-sensitive liposome fusion, a promising indicator of their ability to fuse with a tumor endosome and allow therapeutics to diffuse readily into the cytoplasm. It is known that clustering of flavivirus fusion peptides is crucial in the development of fusion pores.17 Experimentally, it has been observed that peptide clustering occurs during, rather than before, the fusion process.18 Atomistic molecular dynamics simulations have shown the propensity of short (4 amino acid) lipopeptides to cluster in a lipid membrane,19 but it is unknown how longer peptides,
a)
Author to whom correspondence should be addressed; electronic mail: [email protected]
029513-1 Biointerphases 10(2), June 2015
and specifically viral peptides, will interact with each other at the molecular level. On a fundamental level, we are interested in characterizing surface interactions between molecules with many degrees of freedom. The peptides (denoted “LEFN”) studied in this work are anchored to a surface by a linker peptide (LELELELE, “LE”) connected to a synthetic flavivirus native peptide (DRGWGNGCGLFGKGSI, “FN”), as depicted in Fig. 1. The LE peptide was chosen based on the work of Cavalli et al., who showed that lipopeptides with the LE sequence form antiparallel b-sheets in a monolayer at low pH.20 This mechanism may then be exploited to cause pH-sensitive peptide clustering, purportedly leading to increased levels of fusion. Here, we discuss whether this phenomenon indeed explains pH-sensitive fusion. As we are concerned with the effect of pH on peptide interactions, the main difference in our simulations between “high” and “low” pH is that the glutamic acid residues of the linker peptide are either fully protonated (at endosomal pH) or fully deprotonated (at physiological pH), according to a pKa of 4.3 (despite endosomal pH of closer to 4.5).21 Although local pH is realistically different from these idealized protonation states, we are interested in observing the most dramatic possible changes in peptide–peptide interaction based on pH. To further investigate a cause for pH-sensitive liposome fusion, we study the organization of the pH-sensitive linker peptides in the ectodomain, as if they are facilitating fusion by supporting the viral peptides. Once the viral peptides insert into a target membrane, their activity is dependent not only on peptide clustering, but on their insertion angle, penetration depth, etc., which are influenced by the organization
1934-8630/2013/10(2)/029513/8/$30.00
C 2015 American Vacuum Society V
029513-1
029513-2 T. J. Yacoub and I. Szleifer: Molecular dynamics simulations
029513-2
FIG. 1. Atomistic representation of the LEFN peptide. Carbon is shown in teal, oxygen in red, hydrogen in white, nitrogen in blue, and sulfur in yellow. Peptides are anchored to a surface by the nitrogen terminus of first leucine residue. pH was simulated by fully (de)protonating glutamic acid (E) residues, where they are neutral at endosomal pH and negatively charged at physiological pH. Snapshot to right shows two LEFN peptides, with anchor atoms (blue) enlarged. The surface, shown enlarged in transparent black, is an 8 nm 8 nm grid of neutral dummy particles, with spacing 0.2 nm. Peptide anchors are 0.4 nm above the surface.
of the supporting ectodomains.22–24 We simulate the linker peptides as ectodomains, as if the viral peptides are already inserted into a target membrane, and observe their tendency to bundle and interact in a manner much different from the surface-bound peptides. By elongating the linker peptides to encourage bundling, charged groups are exposed and sensitive to the influence of counterions, yielding interesting pHsensitive interactions. II. METHODS A. Molecular dynamics simulations
Atomistic MD simulations of peptides on a surface consist of LEFN peptides anchored (fixed) in the z-dimension by the amino terminus (see Fig. 1), 0.4 nm above a surface, but unconstrained in the xy-plane. The snapshot on the right of Fig. 1 shows the surface as a square lattice of lattice size 0.2 nm, made of neutral dummy atoms fixed in all dimensions with a purely repulsive interaction, serving only as a physical barrier preventing peptide penetration. The reasons for not explicitly simulating the lipid bilayer include: (1) there is no force field optimized for both lipids and proteins (simultaneously), (2) we can access much longer time scales due to increased peptide diffusion, and (3) we are focusing on peptide–peptide interactions. Atomistic molecular dynamics simulations were performed using GROMACS 4.5.5,25 and the OPLS force field26 optimized for proteins. The simple point charge model27 was used for water. Velocity-rescaling temperature-coupling with st ¼ 0.4 ps and T ¼ 323 K, and semi-isotropic Parrinello-Rahman pressure-coupling at 1 atm, with sp ¼ 6 ps were used. This temperature was chosen for consistency with simulations of Biointerphases, Vol. 10, No. 2, June 2015
lipid bilayers in the liquid-disordered state. A time step of 1 fs was used, along with nonbonded interaction cutoffs of 1.6 nm. All systems were equilibrated with steepest descent before molecular dynamics runs. LEFN peptides were created using the Biomer (B) program.28 The amino acid sequence of the viral portion of the peptide, denoted FN, is DRGWGNGCGLFGKGSI. The FN peptide was directly attached to an LELELELE linker domain, denoted LE, as depicted in Fig. 1. LEFN is anchored to the surface via the amino end-terminus in the first leucine (L) residue of the linker peptide. pH is simulated by fully (de)protonating the oxygens of the carboxylic acids on glutamic acid (E) residues, which at “high pH” have partial charge 0.8, and at “low pH” have partial charge 0.3, where partial charge is the elementary charge unit. Realistically, the partial charges on these residues would be dependent on the local environment, but since we are interested in observing the effect of pH on peptide organization, we use these idealized (most extreme) values for partial charge. Arginine and Lysine are charged in the simulations, as the pKa values of their side chains are 12.5 and 10.5, respectively, which are well above the pH being considered.21 Simulations with two or three peptides had dimensions of 8 8 3.8 nm3. Those with 16 peptides had dimensions of 16 16 3.8 nm3. Peptide pair simulations began with peptides completely separated (no contact), and were run for at least 700 ns at both high and low pH. The three-peptide simulation consisted of a third peptide added to two peptides equilibrated from the previously mentioned simulation, and was run for 200 ns at low pH. Sixteen peptides were initially completely separated, and run for 110 ns at low pH.
029513-3 T. J. Yacoub and I. Szleifer: Molecular dynamics simulations
B. Ectodomain bundling
Ectodomain bundling was simulated by assuming that the LE peptides are elongated between the viral membrane and the target membrane, as if the viral peptides were already inserted into the target membrane. Therefore, we only included the LE peptides and solvent. Figure 2 shows a schematic of the simulations, with the inset to the right showing a side view of the high pH system with counterions, and the inset to the left showing a fusion domain, with the ectodomain bundle between (and not directly interacting with) the viral and target membranes. Simulations with 64 LE domains were run at high and low pH for 170 ns, the only differences being the partial charge of glutamic acid residues (see above), and the inclusion of 256 sodium counterions for charge neutrality at high pH (sodium ions are not used in the low pH system). The domains are elongated by fixing the first and last atoms in the z-dimension, and oriented vertically so that the two atoms have the same coordinates in the xy-plane, meaning the bulk of peptide movement is twodimensional. Bundling simulations are run with the same molecular dynamics settings described above for peptide clustering. The simulation box is 12 12 4.3 nm3, and contains around 17 000 water molecules. III. RESULTS A. Peptide pair-interactions
Anchored peptide pairs were run for over 700 ns at both high and low pH, where pH is defined by the protonation state of glutamic acid residues. At low pH, LEFN peptides
029513-3
immediately cluster and reach an anchor distance of r ¼ 1.5 nm [shown in black in Fig. 3]. At t ¼ 40 ns, the snapshot to the right shows the peptides aligning, where the backbone hydrogen–oxygen bond distance between the second “L” residues (shown in green in the figure) being approximately 0.25 nm. After 50 ns, the peptides transition into a local free energy minimum with an anchor distance of r ¼ 1 nm. This configuration is short-lived, however, as the peptides enter a side-by-side arrangement (stacked, not like a b-sheet) at t ¼ 100 ns. The peptides continue to transition to different conformations, as depicted in the snapshots in Fig. 3. At t ¼ 200 ns and beyond, the linker peptides do not exhibit any apparent order. This observation demonstrates that while the LE peptides appear to form distinct structures at low pH, these are just a few of many potential structures corresponding to local free energy minima. According to Fig. 3, peptides at low pH appear to make sharp transitions from one conformation to the next. At high pH (Fig. 4), by contrast, anchors do not make sharp transitions and peptides do not exhibit transient structure, as evidenced by the high variability of anchor distance as compared to low pH. There is the potential for a stabilizing disulfide bond between Cys atoms on neighboring LEFN peptides, which is represented by the distance between sulfur atoms in the low pH system in Fig. S4.29 Although there appear to be periods of nearly constant distance (on the order of 10 s of ns), the distance between sulfur atoms does not approach the disulfide bond ˚. distance of 2.05 A Figure 5 represents the data from Figs. 3 and 4 as histograms of anchor distance. In red, anchors at low pH do not
FIG. 2. Schematic of peptide bundling simulations. LE linker peptides are elongated upright (in the z-direction) by fixing the first and last atoms in the z-dimension. Viral peptides are assumed to be inserted into the target cell (i.e., endosome). Inset to the left adapted from Ref. 22. Snapshot to the right shows a side view of a simulation at high pH, where atom colors correspond to Fig. 1 and blue spheres are sodium counterions. Biointerphases, Vol. 10, No. 2, June 2015
029513-4 T. J. Yacoub and I. Szleifer: Molecular dynamics simulations
FIG. 3. Low pH. Distance between a pair of LEFN anchor atoms (black) and a potential hydrogen bond between backbone atoms of the second leucine residues: nitrogen of one peptide and oxygen of the other (green) over more than 700 ns. Snapshots show linker peptide orientation at various time points (viral peptide hidden for clarity). Note distinct structures at 40 and 100 ns. Large spheres represent anchors, and small spheres represent the potential NH–O bond described above. Peptides began fully separated.
separate by more than 2.5 nm. The maximum (corresponding to a free energy minimum) is clearly located around r ¼ 1 nm with a smaller maximum around r ¼ 1.4 nm. At high pH, maxima of comparable magnitude exist at r ¼ 1 nm and r ¼ 2 nm, which correspond to free energy minima of equal magnitude. Thus, both high and low pH peptides show a maximum at an anchor distance near 1 nm, while the
029513-4
FIG. 5. Histogram of anchor distance between peptide pairs for over 700 ns at both high and low pH, representing the data from Figs. 3 and 4. Maxima correspond to most likely distance between anchor atoms.
location of the second maximum is different, as is the relative magnitude of the maxima. Figures 3–5 show that while the specific interactions between the pH-sensitive linker peptides (locations of most likely anchor distance) are clearly different between the high and low pH cases, charge does not have a qualitative effect on clustering, as peptides at either pH do not separate beyond r ¼ 3.5 nm. Repulsion between negatively charged glutamic acid residues at high pH is not strong enough to completely separate the peptides. B. Multipeptide interactions 1. Adding a third peptide
FIG. 4. High pH. Distance between a pair of LEFN anchor atoms (black) and a potential hydrogen bond between backbone atoms of the second leucine residues: nitrogen of one peptide and oxygen of the other (green) over 700 nsþ. Unlike peptides at low pH (Fig. 3), those at high pH did not exhibit clear structural transitions. Peptides began fully separated. Biointerphases, Vol. 10, No. 2, June 2015
The energetics of peptide clustering were further studied by adding a third peptide to two peptides (at low pH), which had reached an anchor distance of r ¼ 1 nm, corresponding to the maximum in Fig. 5. Clustering was measured with a histogram of peptide anchor distances (Fig. 6). Adding a third peptide did not significantly disrupt the strong interaction between the two original peptides (shown in gray in Fig. 6, over the course of 300 ns). Interestingly, the most probable anchor distance between the original peptides actually decreases, evidenced by a shift inward in the peak from r ¼ 1 nm to r ¼ 0.8 nm, and the emergence of a peak around r ¼ 0.5 nm. In the blue and red plots of Fig. 6, it is evident that the third peptide remained strongly attached over the 200 ns simulation. The third peptide only had 1.5 nm of mobility with respect to its anchor distance with the other peptides (the approximate “spread” of the blue and red plots), and was never fully released from the cluster as it did not separate from the first peptide by more than 2.5 nm. Although the third peptide is strongly bound, the interaction strength between it and the original peptides is significantly different. As an estimation of this difference, we can compare the maxima of the
029513-5 T. J. Yacoub and I. Szleifer: Molecular dynamics simulations
FIG. 6. Histogram of anchor distance within three peptides. A third peptide (pep3) was added to two peptides (pep1 and pep2) with an anchor distance of 1 nm. pep1–pep2 is the distance between the original peptide anchors. pep1–pep3 and pep2–pep3 are distances between each original peptide and the added third peptide.
histograms in Fig. 6. For the original two peptides, the maximum is approximately three times higher than that of the third peptide and either of the original peptides. By treating these values as probabilities, we can estimate the free energy difference as DG ¼ kTln(3) or approximately 1 kT. Given sufficient time, these interactions may change. For example the first peptide anchor may approach the third anchor as it weakens its interaction with the second peptide. However, Fig. 6 shows obvious asymmetry in peptide cluster interactions. It is clear that creating a simple model of peptide interaction, for example, by representing the peptides as spheres with an effective pair interaction potential, will be qualitatively inaccurate. Therefore, to accurately model multibody peptide interactions, one would need to construct a multibody interaction potential.
029513-5
FIG. 7. Representative snapshot of 16-peptide cluster, after 60 ns. Anchor atoms shown enlarged. Beginning with 16 separated peptides, this cluster began to form immediately, and remained intact for the duration of the 110 ns simulation.
peptides remained clustered over the course of the 110 ns simulation. C. Ectodomain bundling
LEFN is a viral peptide attached to a linker peptide and is used specifically to cause fusion in liposomal drug delivery, as we have observed in unpublished experiments.16 The liposome fusion experiments show a distinct sensitivity of fusion to pH, favoring higher levels of fusion at lower pH. Here, we aim to understand the physical cause. Since the viral peptide is not significantly pH sensitive (considering the pKa of its side chains), we focus on the pH-sensitive linker peptide. Based on several studies, the domains supporting viral peptide insertion, which here are assumed to be the linker peptides, are elongated in bundles.22–24 Therefore, we studied
2. Cluster of 16 peptides
To observe peptide interactions within a larger cluster of peptides, we attached 16 LEFN peptides at low pH to a 16 nm 16 nm surface. The peptides began to aggregate within a few nanoseconds, as expected from the two- and three-peptide simulations. Figure 7 shows a snapshot of a 16-peptide cluster after 60 ns, displaying further evidence of the strong clustering propensity of the peptides. Figure 8 shows a histogram of the distance between each anchor and its nearest neighboring anchor in the cluster. The first 60 ns was excluded to allow time for peptides to form a single cluster (see Fig. S1).29 Like the pair peptide simulations (Fig. 5), a maximum around r ¼ 1 nm and a smaller maximum around r ¼ 1.5 nm are observed. Significantly, a clear third maximum around r ¼ 0.5 nm is observed, and is more pronounced than in the three-peptide simulations (Fig. 6), indicating again that additional peptides tend to decrease the distance between anchors and tighten peptide clusters. The Biointerphases, Vol. 10, No. 2, June 2015
FIG. 8. Histogram of distances between nearest-neighbor anchors in a 16peptide cluster over the final 50 ns of a 110 ns simulation (Fig. 7). The maxi˚ is more pronounced than in the cluster of three peptides mum at r ¼ 5 A (Fig. 6).
029513-6 T. J. Yacoub and I. Szleifer: Molecular dynamics simulations
029513-6
FIG. 9. Sixty four linker peptides at high pH (a) and low pH (b) in the xy-plane. Inset shows side view of (a). Peptides are elongated in the z-dimension. Sodium counterions at high pH are blue. Note the long-range order induced by sodium ions bridging negatively charged linker peptides at high pH.
the interaction between linker peptides in water. To encourage the bundle structure, the first and last atoms of the peptides were fixed in the z-direction, with backbones fully elongated and upright, all at the same height. The only movement (beyond vibrational and slight torsional motion) is in the xy-plane. We find that at low pH, the peptides bundle occasionally forming what appear as b-sheets (hydrogen bonds between backbone atoms). Interestingly, at high pH, Fig. 9(a) shows sodium ions, in blue, bridging the negatively charged glutamic acid residues (see Fig. 1), inducing long-range domains resembling ion channels. These domains remained structurally unchanged following their creation, for the remainder of the 170 ns simulation. Based on our observations, we expect the bundles at high pH to maintain some form of long-range order indefinitely [Fig. 9(a)]. At low pH [Fig. 9(b)], on the other hand, the bundles slowly aggregate over time, and we expect them to fully aggregate (into a single large bundle) after another 100–200 ns. Ectodomain bundling is therefore qualitatively dependent on pH, suggesting a physical mechanism to explain pH-sensitive liposome fusion. Figure 10 shows the radial distribution functions for the first atoms (fixed in z) of each elongated linker peptide, representing peptide organization in the xy-plane. There is
FIG. 10. Radial distribution function for base atoms of the ectodomains at high and low pH. Note the third peak around r ¼ 1.2 nm at high pH, reflecting a long-range order not present at low pH. Biointerphases, Vol. 10, No. 2, June 2015
clearly long-range (1.2 nm) ordering present at high pH and not low pH. Without knowing the particular ectodomain structure needed to support LEFN peptide insertion, it is conceivable that LE peptides at high pH are physically constrained by sodium counterions, potentially preventing them from achieving this structure. IV. DISCUSSION We have likely observed a few of many possible peptide cluster conformations. In fact, as represented in Fig. 3, at low pH, the peptides continuously engage in complex binding/unbinding events, signified by clear transitions in peptide structure. However, based on the extensive simulations presented, we strongly believe that the observed peptide conformations are representative of the numerous possible conformations, meaning that peptide clustering will occur regardless of pH (and is therefore not the cause of pH-sensitive liposome fusion). This expectation is due to the fact that, after several hundred nanoseconds of simulation time, the peptides have shown no sign of separation. There are two main goals of this work: (1) to characterize the interactions between surface-bound molecules with many degrees of freedom, and (2) to discuss the relationship between molecular interactions and the pH-sensitive liposome fusion observed in our experimental work.16 We observe definitively that LEFN peptides consistently and quickly cluster, as evidenced in the pair peptide simulations (Figs. 3–5), simulations with a third peptide (Fig. 6), and simulations with 16 peptides (Fig. 7). Adding a third peptide to a pair resulted in a decrease in the anchor distance between the original pair, as evidenced by the maximum in Fig. 6 at r ¼ 0.5 nm, which is much greater than that of the pair simulations. Simulations of 16 peptides made this even more evident, as the maximum at r ¼ 0.5 nm was more pronounced, and represented a free energy minimum as deep as the one at r ¼ 1 nm. These multipeptide simulations indicate that additional peptides tighten peptide clusters in a cooperative effect (more peptides translates to tighter clusters). The effect may be inversely proportional to the size of the cluster, as a very large cluster should not be as sensitive to the influence of additional peptides.
029513-7 T. J. Yacoub and I. Szleifer: Molecular dynamics simulations
Importantly, these cooperative effects imply that simple pair interaction potentials cannot sufficiently describe or predict the behavior of peptide clusters. This became more clear as we attempted to calculate a potential of mean force between two LEFN peptides at low pH, using two methods. The first was to place the peptide anchors at 0.1 nm intervals and anneal them to ensure complete separation and attempt a relatively path-independent calculation. The free energy minimum was found to be at an anchor distance of 2 nm, in qualitative disagreement with the free peptide simulations (Fig. 3). The second method was to pull the peptides from an equilibrated position to obtain conformations, also at 0.1 nm intervals. This clearly path-dependent calculation yielded a free energy minimum near 1 nm (in agreement with Fig. 3); however, the free energy barrier vanished as the peptides relaxed. Without the ability to freely rearrange, the peptides were limited in conformational freedom, inevitably resulting in a biased representation of entanglement. These observations emphasize the difficulty in studying molecules with many degrees of freedom, reinforcing the conclusion that pair interactions cannot be represented by a simple potential and are thus insufficient to describe peptide clustering. It is useful to discuss our approximations and their potential influence on the results. In addition to the issues addressed above and the general limitations associated with molecular dynamics in biology,30 our main approximation is that of a lipid bilayer as a neutral, noninteracting surface. We use this approximation because simulations can be run for longer times, peptides can diffuse faster, and we are focusing on peptide–peptide interactions, which are optimized by the OPLS force field. Although the OPLS force field has also been optimized for lipids, it has not been optimized for both lipids and proteins simultaneously. Whereas a lipid bilayer would surely be needed to study peptide insertion and fusion domains, we are interested in peptide interaction on the surface of the liposome. Peptide cluster mobility may depend on the interplay between peptide cluster size and lipid membrane rigidity, and lateral peptide mobility will depend on lipid–peptide interactions; however, simulations of larger clusters in the full membrane environment would be necessary to elucidate these effects. An additional limitation is that lipids may encourage peptides to interact or they may screen peptide interactions. From preliminary (unpublished) results, coarse-grained simulations of the full membrane-lipopeptide system show peptides clustering on the membrane surface and not inserting into the membrane after 1 ls, somewhat justifying our use of a flat surface and supporting the conclusion that the peptides tend to cluster. Indeed, Pecheur et al. have shown experimentally that while there is a quantitative dependence of fusion on lipid membrane composition, the qualitative dependence is weak. Our results are in agreement with Pecheur et al., who have shown that fusion peptide clustering and oligomerization is not a prerequisite for pH-sensitive fusion;18 therefore, fusion is likely the cause rather than the effect of peptide clustering. While this is in line with our conclusion that clustering is not responsible for pH-sensitive fusion, we expect LEFN Biointerphases, Vol. 10, No. 2, June 2015
029513-7
peptides to cluster readily, whereas Pecheur et al. found that their peptides (which had 11 amino acids and were amphipathic and negatively charged) were driven to cluster by the fusion process.18 To further search for a molecular-level explanation for the pH-sensitive fusion found experimentally,16 we look to the role of the ectodomain in support of viral peptide insertion in the target membrane. The most important assumption is that the LE linker peptides are able to serve this function. In simulations of LE ectodomains, linker peptides are elongated in the z-dimension to simulate their extension between the liposomal membrane and the target membrane (see Fig. 2). Figure 9(b) shows that at low pH, linker peptides form more compact bundles. Interestingly, we find that at high pH [Fig. 9(a)], sodium counterions induce a bridging effect between negatively charged glutamic acid residues in vertically oriented linker peptides. Due to the significant pH-sensitivity of this phenomenon, we believe this long-range order is at least partially responsible for the pH-sensitivity of liposome fusion. At high pH, the linker ectodomains are constrained in their ability to form a tight bundle, translating into greater difficulty in providing proper support to the viral peptide, which is crucial in the specific interactions involved with insertion and orientation of viral peptides in a target membrane.22–24 Often only a small number of peptides (i.e., three in the inset of Fig. 2) may be needed to support viral peptide insertion, but by including many peptides we achieve better sampling to characterize possible effects of pH. Although peptide clustering appears to be independent of pH, bundling of the LE ectodomains is clearly pH-dependent in a qualitative manner. This constraint is a possible explanation of the fact that the viral peptides cause less fusion at higher pH. V. CONCLUSION From a fundamental standpoint, we have characterized the interactions between two molecules with many degrees of freedom anchored to a surface. Specifically, we have studied LEFN peptides composed of a pH-sensitive linker peptide attached to a synthetic flavivirus-type viral peptide. The tendency of the peptides to cluster is independent of pH. At both high and low pH, peptide pairs do not separate on a time scale of 700 ns. We conclude that pH-sensitive membrane fusion is not the result of peptide clustering. This result agrees with experimental findings in that clustering was not found to cause fusion, but disagrees with the same findings as we predict peptides to cluster readily whereas the experiments showed that the fusion itself process causes clustering. However, the peptides used experimentally were significantly different, and less than half the length of LEFN.18 Interestingly, although all simulations showed that the LEFN peptides tend to cluster, we find that the distance between peptide anchors in a cluster cannot be predicted by studying a peptide pair alone. Adding a third peptide decreases the anchor distance between two previously interacting peptides. The anchors in a cluster of 16 peptides are
029513-8 T. J. Yacoub and I. Szleifer: Molecular dynamics simulations
also more closely bound than those in a pair. Additionally, the role of the linker peptides in the ectodomain was investigated. When elongated, linker peptides formed bundles at both high and low pH. However, at high pH, sodium counterions bridge negatively charged glutamic acid residues, inducing long-range order, creating the appearance of an ion channel [Fig. 9(a)] and constraining their ability to achieve the necessary organization to support viral peptide insertion. We hypothesize that this constraint on the organization of the linker peptide ectodomains is a significant contribution to the decreased liposome fusion observed at higher pH.16 We have shed light on the molecular-level clustering of surface-bound molecules, a poorly understood yet ubiquitous occurrence in liposomal drug delivery. This molecular organization is much more complex than the homogeneous distribution of molecules normally depicted, as clustering and multibody interactions have shown. Despite their significant approximations, the methods developed in this work can be used to study virtually any anchored molecules or ectodomains. ACKNOWLEDGMENTS This work was supported by NSF Grant No. CBET1403058. This research was supported in part through the computational resources and staff contributions provided for the Quest high performance computing facility at Northwestern University which is jointly supported by the Office of the Provost, the Office for Research, and Northwestern University Information Technology. 1
A. Samad, Y. Sultana, and M. Aqil, Curr. Drug Delivery 4, 297 (2007). A. Gabizon, R. Catane, B. Uziely, B. Kaufman, T. Safra, R. Cohen, F. Martin, A. Huang, and Y. Barenholz, Cancer Res. 54, 987 (1994). 3 A. Sharma and U. S. Sharma, Int. J. Pharm. 154, 123 (1997). 4 S. Ganta, H. Devalapally, A. Shahiwala, and M. Amiji, J. Controlled Release 126, 187 (2008). 2
Biointerphases, Vol. 10, No. 2, June 2015
029513-8 5
T. M. Allen and P. R. Cullis, Adv. Drug Delivery Rev. 65, 36 (2013). K. Maruyama, Adv. Drug Delivery Rev. 63, 161 (2011). 7 K. Knop, R. Hoogenboom, D. Fischer, and U. S. Schubert, Angew. Chem. Int. Ed. 49, 6288 (2010). 8 M. L. Immordino, F. Dosio, and L. Cattel, Int. J. Nanomed. 1, 297 (2006). 9 S. M. Moghimi and J. Szebeni, Prog. Lipid Res. 42, 463 (2003). 10 Y. Lu and P. S. Low, Adv. Drug Delivery Rev. 64, 342 (2012). 11 A. Gabizon, H. Shmeeda, A. Horowitz, and S. Zalipsky, Adv. Drug Delivery Rev. 56, 1177 (2004). 12 J. Sudimack and R. J. Lee, Adv. Drug Delivery Rev. 41, 147 (2000). 13 T. Kakudo, S. Chaki, S. Futaki, I. Nakase, K. Akaji, T. Kawakami, K. Maruyama, H. Kamiya, and H. Harashima, Biochemistry 43, 5618 (2004). 14 A. Bolhassani, Biochim. Biophys. Acta, Rev. Cancer 1816, 232 (2011). 15 E. Mastrobattista, G. A. Koning, L. van Bloois, A. Filipe, W. Jiskoot, and G. Storm, J. Biol. Chem. 277, 27135 (2002). 16 T. J. Yacoub, “Molecular interactions in liposomal drug delivery,” Ph.D. thesis (Northwestern University, 2014). 17 Y. Mendes et al., PLoS One 7, e47596 (2012). 18 E. I. Pecheur, J. Sainte-Marie, A. Bienven€ ue, and D. Hoekstra, Biochem. J. 38, 364 (1999). 19 J. N. Horn, T. D. Romo, and A. Grossfield, Biochem. J. 52, 5604 (2013). 20 S. Cavalli et al., J. Am. Chem. Soc. 128, 13959 (2006). 21 R. H. Garrett and C. M. Grisham, Principles of Biochemistry: With a Human Focus (Thomson Brooks/Cole, Toronto, 2001). 22 J. L€ uneberg, I. Martin, F. N€ ußler, J.-M. Ruysschaert, and A. Herrmann, J. Biol. Chem. 270, 27606 (1995). 23 R. M. Epand, Y. Shai, J. P. Segrest, and G. M. Anantharamiah, Biopolymers 37, 319 (1995). 24 J. E. Donald, Y. Zhang, G. Fiorin, V. Carnevale, D. R. Slochower, F. Gai, M. L. Klein, and W. F. DeGrado, PNAS 108, 3958 (2011). 25 B. Hess, C. Kutzner, D. van der Spoel, and E. Lindahl, J. Chem. Theory Comput. 4, 435 (2008). 26 W. D. Cornell, J. Am. Chem. Soc. 117, 5179 (1995). 27 H. J. C. Berendsen, J. P. M. Postma, W. F. van Gunsteren, and J. Hermans, Intermol. Forces 11, 331 (1981). 28 N. B. White, “B home page” http://casegroup.rutgers.edu/Biomer/. 29 See supplementary material at http://dx.doi.org/10.1116/1.4919408 for four figures, (1) anchor distance in the 16-peptide cluster, (2) more interatomic distances in the clustered linker peptides at low pH, (3) sodium ion radial distribution functions over time in the ectodomain simulation at high pH, and (4) potential di-sulfide bond distance. 30 J. D. Durrant and J. A. McCammon, BMC Biol. 9, 71 (2011). 6
