Organization, dynamics, and segregation of Ras nanoclusters in membrane domains

Abstract

Recent experiments have shown that membrane-bound Ras proteins form transient, nanoscale signaling platforms that play a crucial role in high-fidelity signal transmission. However, a detailed characterization of these dynamic proteolipid substructures by high-resolution experimental techniques remains elusive. Here we use extensive semiatomic simulations to reveal the molecular basis for the formation and domain-specific distribution of Ras nanoclusters. As model systems, we chose the triply lipidated membrane targeting motif of H-ras (tH) and a large bilayer made up of di16∶0-PC (DPPC), di18∶2-PC (DLiPC), and cholesterol. We found that 4–10 tH molecules assemble into clusters that undergo molecular exchange in the sub-μs to μs time scale, depending on the simulation temperature and hence the stability of lipid domains. Driven by the opposite preference of tH palmitoyls and farnesyl for ordered and disordered membrane domains, clustered tH molecules segregate to the boundary of lipid domains. Additionally, a systematic analysis of depalmitoylated and defarnesylated tH variants allowed us to decipher the role of individual lipid modifications in domain-specific nanocluster localization and thereby explain why homologous Ras isoforms form nonoverlapping nanoclusters. Moreover, the localization of tH nanoclusters at domain boundaries resulted in a significantly lower line tension and increased membrane curvature. Taken together, these results provide a unique mechanistic insight into how protein assembly promoted by lipid-modification modulates bilayer shape to generate functional signaling platforms.

Electron microscopy (EM) and fluorescence resonance energy transfer (FRET) measurements have shown that lipid-anchored signaling proteins assemble into dynamic substructures on the plasma membrane (PM) (1–3). Stimulus dependent assembly and disassembly of these nano-sized signaling platforms, or nanoclusters, is emerging as a crucial mechanism by which cells achieve high-fidelity signal transmission (4, 5). Well characterized examples include clusters of Ras proteins (6) at the cytosolic side and glycosylphosphatidylinositol-anchored receptors (GPI-ARs) (3) at the periplasmic side of the PM, where 30–40% of Ras and 20–40% of GPI-AR proteins form clusters at all expression levels (1, 3). Our goal here is to examine the physical basis for the formation and distribution of H-ras nanoclusters.

H-ras and its close homologues N- and K-ras belong to the Ras family of monomeric guanosine triphosphatases (GTPases). Ras proteins cycle between GDP-bound off and GTP-bound on conformational states to regulate signaling pathways involved in cell differentiation, proliferation, and survival. Aberrant Ras function accounts for approximately 30% of all human tumors and a variety of developmental disorders (7–9). The primary site of Ras action is the plasma membrane, where all Ras proteins are anchored via their lipid-modified C terminus known as the lipid-anchor. The lipid-anchor of H-ras is a seven-residue motif containing a polyunsaturated farnesyl and two saturated palmitoyls (Fig. 1). Both N- and K-ras are also farnesylated, but N-ras lacks the second palmitoyl while the polybasic lipid-anchor of K-ras has none. Due primarily to these differences, Ras proteins form distinct and nonoverlapping nanoclusters (2). Because nanoclusters are the exclusive site of effector recruitment and activation (10) differential PM clustering may partially explain the observed functional differences among Ras proteins (11). Moreover, the number, but not the size, of Ras nanoclusters is a linear function of the stimulating ligand concentration which led to the suggestion that nanoclusters may digitize and thereby prevent signal degradation (4). The stability of the inactive GDP-bound, but not the active GTP-bound, H-ras nanoclusters exhibit sensitivity to cholesterol (CHOL) (2, 12). The isolated lipid-anchor of H-ras (residues 180–186, referred to as tH) also forms clusters in a CHOL-dependent manner, suggesting a critical role for the lipid-modified moiety in the formation and distribution of Ras nanoclusters (2, 12). Therefore, tH represents a highly suitable model system to probe the atomic details of Ras nanoclustering.

Fig. 1.

Molecular models. (A) CG models for (from left to right): DPPC, DLiPC, CHOL, and tH. For the lipids the beads represent NCH₃ (blue), PO₄ (red), and glycerol (gray) groups. Cyan and pink represent the saturated and unsaturated portions of the lipid tails, respectively. For CHOL, the ring is shown in yellow, the tail in cyan and the OH group in red. The heptapeptide tH is comprised of Gly180, Pa181, Met182, Ser183, Pa184, Lys185, and Fa186; Pa and Fa represent palmitoylation and farnesylation, respectively. Note that, despite the different color schemes used here, the hydrocarbon tail of Pa is the same as that of DPPC. (B) Top view of the initial structure (left) and after 24 μs MD (right) of the 5∶3∶2 DPPC∶DLiPC∶CHOL ternary bilayer in the absence of tH. The initial structure shown here is after minimization and preequilibration (heating) of the system. DPPC, DLiPC and CHOL are shown as tan, blue and white dots, respectively. (C) Top view of the initial structure (left) and after 24 μs (right) of a system with 64 tH molecules bound to one side of the ternary bilayer. The 64 tH molecules were initially distributed on a grid with random orientations. Similarly to the free bilayer, the initial configuration is after minimization and heating of the system with the tH restrained to their original grid positions. The tH beads are represented in red balls, with palmitoyl and farnesyl tails colored in green and yellow, respectively. Images rendered with VMD (44).

Recent in vitro studies provided additional insights that complement the cell-based data (13, 14). Using atomic force microscopy and biochemical techniques, these studies have shown that K- and N-ras form clusters in synthetic membranes, and that K-ras clusters reside at the liquid disordered (L_d) domain (14) while those of N-ras prefer the boundary between L_d and liquid ordered (L_o) domains (13).

Despite these intriguing biological and biophysical observations at the macrolevel, the molecular details and physical basis for the formation, stability, and distribution of nanoclusters are still missing. A major difficulty is that most high-resolution experimental techniques are not suitable to study such highly dynamic substructures. Modern computational techniques, such as coarse-grained (CG) molecular dynamics (MD), offer an appealing alternative. It is well established that these techniques can provide key insights into the molecular details of bilayers and protein-lipid complexes that are difficult to access by other methods (see for instance refs. 15, 16). Most importantly, recent advances in these techniques now allow a detailed characterization of membrane subdomains in much larger spatial and temporal scales (e.g., refs. 17, 18). In this work, we used CG-MD at semiatomic resolution to study spontaneous clustering and distribution of tH in a bilayer that forms coexisting L_o/L_d domains: a 5∶3∶2 mixture of dipalmitoylphosphatidylcholine (DPPC), dilinoleoylphosphatidylcholine (DLiPC) and CHOL.

Results

The stability of tH nanoclusters varies with the CHOL content of the host membrane, suggesting a preference for CHOL-enriched L_o (12) or coexisting L_o/L_d domains. Spontaneous formation of such domains is a long time scale process (μs or longer). Therefore, the use of fully atomistic methods is impractical. A number of CG models have been proposed to overcome this challenge (19–23). The MARTINI CG model (19, 20) used here (Fig. 1A) had a remarkable success in studying nanoscale lipid domains (17, 18), vesicles (24), liposomes (17, 25, 26), and aggregation of transmembrane helices (27–30). After testing various lipid compositions and molar ratios for their ability to form coexisting L_o/L_d domains at ambient temperatures, we determined that a 5∶3∶2 mixture of di-16∶0-PC (DPPC), di-18∶2-PC (DLiPC), and CHOL is suitable to study tH clustering (see Methods; SI Text). We note that DLiPC is not representative of a majority of eukaryotic lipids but was chosen to facilitate domain formation.

Domain Formation in a Ternary Lipid Mixture With and Without H-ras Lipid-Anchor.

Two nanoscale domains are apparent in the tH-free DPPC∶DLiPC∶CHOL bilayer shown in Fig. 1B: a DLiPC-dominated L_d and DPPC/CHOL-dominated L_o domains. Except at 48 °C where domains are small and dynamic, the system evolved into two distinct regions at 8, 18, 28, and 38 °C. At the room and physiologic temperatures of 28 °C and 38 °C, these striped regions exhibit equilibrium properties typical of L_d and L_o domains (Fig. 1B). For instance, at 28 °C, the L_o domain is thicker by 0.94 nm than the L_d, with the lateral diffusion coefficient of the DPPC lipids being fourfold smaller and the chain order parameters twofold larger in the L_o than in the L_d region (Fig. S1).

A similar behavior was found for simulations in which 64 randomly oriented tH molecules (equivalent to approximately 3.2% of the total number of nonsolvent molecules) were arranged grid-like on one side of the bilayer (Fig. 1C). (The asymmetric tH distribution mimics the attachment of Ras only to the cytosolic side of the PM.) tH insertion had minor effect on some of the bilayer equilibrium properties (e.g., a slightly smaller L_o/L_d thickness difference at 28 °C), but the effect on bilayer mechanical properties is significant (discussed later).

H-ras Lipid-Anchors Cluster in Model Membranes.

The lateral dimension of the simulation box was large enough to permit the allocation of approximately 8 nm² area per tH so that no inter-tH contact exists at the start of the simulations. After equilibration, tH molecules formed nonrandom assemblies at all simulation temperatures (Fig. 1C). Another simulation with a different initial condition (48 randomly oriented tHs) shows a similar clustering behavior (Fig. S2). To quantitatively characterize the clusters in the 64 tH system, we used the criterion that two tH molecules are part of the same cluster if any of their beads are within 0.75 nm (Fig. S3). The evolution of the total number of tH species (monomers or clusters of any size) and the weight-averaged aggregation numbers (N_W, calculated by a single-linkage algorithm, see Fig. S3) indicate that fast-forming random assemblies of tH quickly grow into larger clusters (Fig. S4). Clustering precedes lipid domain formation (Fig. S5A), suggesting that tH, which has three tails and a large head (backbone plus nonlipidated side chains), is less compatible with either DPPC or DLiPC than are the two lipid types with each other. In fact, we found similar clustering profiles in pure DPPC or DLiPC bilayers (Fig. S5B). Thus, tH clustering involves the same fundamental forces that segregate lipids, and does not require preexisting membrane domains. However, the stability and size of the clusters varies with the simulation temperature and hence the extent of lipid segregation (see below).

A single-linkage (Fig. S3) analysis of the cluster size distributions after equilibration (last 16 μs) indicate that on average 17, 22, and 29% of the tH molecules are monomers in the simulations at 18, 28, and 38 °C (Fig. 2A). Approximately 10–20% form dimers or trimers whereas the remaining 62, 43, and 34% form clusters of size 4 or larger (s≥4). The right hand-side inset of Fig. 2A shows that 80% of idealized noninteracting particles (characterized by complete spatial randomness) remain monomers with only 20% existing as dimers or trimers; no cluster of s≥4 exists. Clearly, the formation of tH clusters of size s≥4 is interaction-driven; i.e., not random. The fraction of tHs belonging to these larger clusters is 43 and 34% at 28 and 38 °C, respectively, in close agreement with the 36–40% clustered fraction measured by EM as well as single fluorescence video tracking experiments (12, 31). For a more direct comparison with experimental data, we reanalyzed the EM-derived spatial distribution of tH clusters using both single-linkage and the widely used Ripley’s K-function cluster analysis techniques (see SI Text and Fig. S6). There is an excellent agreement between the calculated and EM-derived (1) cluster size distributions (Fig. 2A).

Fig. 2.

Size and localization of tH nanoclusters. (A) Comparison of cluster size distributions derived from EM (gray bars) and MD simulations at 18 °C (turquoise), 28 °C (black) and 38 °C (red). Insets: left, results from single-linkage (empty squares) and Ripley’s K-function (filled circles) analysis of the EM data; right, results from a simulated 2D Poisson distribution of 64 noninteracting particles in the same box size as our system. (B) Lipid composition across the bilayer with the origin set to be the middle of the L_d domain. Dotted lines demarcate the center of the L_o/L_d boundary, defined as the location at which the average DPPC and DLiPC compositions are equal. (C) Probability distribution of tH across the bilayer (black line) and its decomposition into clusters of different sizes (s) shown in purple (monomers), cyan (dimers and trimers), and red (four or larger). All data in this and subsequent figures are from simulations performed at 28 °C.

To study the dynamics of the tH clusters, we estimated the characteristic lifetimes of molecular exchange among clusters or between clusters and the pool of monomers (SI Text). The results suggest that two main mechanisms underlie tH cluster dynamics: dissociation/association of monomers from/into clusters at the sub-μs time scale and splitting/merging of clusters/subclusters occurring at the μs time scale. The latter is dominant at all temperatures, accounting for approximately 84% of the total dynamics at 28 °C. Both processes exhibit strong dependence on temperature, but the fact that each process is fast even at low temperatures (Fig. S7), and that clustering occurs also in the pure bilayers, underline the transient nature of Ras nanoclusters (12). Note, however, that the current time scales should not be directly compared with the cell-based experimental estimates of cluster lifetimes (< 0.1 s) as the simulations lack many other PM components and the actin cytoskeleton (1).

tH Nanoclusters Localize at Domain Boundaries.

To determine the lateral organization of the tH clusters, we plotted the lipid composition (Fig. 2B) and the tH fraction (Fig. 2C) across the bilayer with the origin chosen to be the center of the DLiPC-dominated L_d domain. It is clear that tH molecules overwhelmingly prefer the domain boundaries. Deconvolution of the distribution into monomers, dimers, trimers, and larger clusters (s≥4) (Fig. 2C) shows that monomers exist across the bilayer while dimers and trimers have some preference for the interfacial regions. In contrast, clusters of four or more tH molecules segregate almost exclusively to the interface. The same conclusion could be reached from analysis of the simulation at the near-physiologic temperature of 38 °C: medium-to-large tH clusters are restricted to the interfacial regions while monomers and smaller clusters are not. In subsequent sections, we will use data from the 28 °C trajectory (unless stated otherwise) to assess the physical basis and structural consequence of tH clusters’ interfacial localization.

Antagonistic Action of Farnesyl and Pamitoyl Tails Segregate tH Nanoclusters to Domain Boundaries.

Individual lipid modifications of H-ras have distinct roles in trafficking and plasma membrane binding (32) and their contribution to membrane affinity is nonadditive (33). To delineate the contribution of individual lipidations to Ras clustering and segregation, we generated four tH variants representing various levels of de-palmitoylation (de-Pa) and de-farnesylation (de-Fa) by systematically replacing the lipid tails with the parent Cys. Simulations of each variant yielded nanoclusters of similar sizes as those of the wild-type tH (Fig. 3A). In particular, removal of Pa181 and Pa184 separately or together has very little effect on cluster size. Removal of the farnesyl tail (i.e., de-Fa186) resulted in a comparatively more significant reduction in the number of the smaller (s < 4) and increase in the larger species (s = 10 - 16).

Fig. 3.

Size and localization of mutant tH nanoclusters. (A) Comparison of wild-type tH cluster sizes (gray bars) with those of de-palmitoylated (de-Pa181: cyan, de-Pa184: blue, de-Pa181/184: green) and de-farnesylated (de-Fa186: magenta) tH variants. (B) and (C) The probability distribution of the four tH variants across the bilayer. The wild-type tH distribution is shown for reference (dashed line).

Dramatic changes were found when comparing the tH variants in terms of cluster localization (Fig. 3 B and C). While removal of the palmitoyls individually (de-Pa181 and de-Pa184) has little impact on interfacial clustering, their simultaneous removal (de-Pa181/184) led to complete segregation of the clusters to the L_d domain. In contrast, removal of the farnesyl tail (de-Fa186) redistributes the clusters to the L_o domain, with some preference for the peripheral region to the core. In each case monomers and smaller clusters (s < 4) do not exhibit significant domain preference (see Fig. S8 for deconvolution of the curves in Fig. 3 B and C). We conclude that clusters of lipidated proteins containing both Pa and Fa segregate to domain boundaries, those with only Pa prefer the L_o domain while those with only Fa prefer L_d. In other words, the antagonistic action of farnesyl and pamitoyl modifications is responsible for the segregation of tH nanoclusters to domain boundaries. Moreover, the minimum requirement for interfacial localization is one palmitoyl and one farnesyl, consistent with the preference of N-ras for domain boundaries (13) and K-ras for the L_d domain (14).

Interface-Localized Nanoclusters Stabilize Membrane Curvature.

The hydrocarbon tails of the DPPC and DLiPC lipids have different saturation levels and chain order parameters (Fig. S1A) that lead to a large L_o/L_d thickness difference (Δd = 0.94 nm in the tH-free system). This phenomenon, often referred to as hydrophobic mismatch, is associated with lateral pressure and partial exposure of the DPPC hydrocarbon tails to solvent, costing energy in the form of interfacial line tension (F_l). Recent studies have shown that hydrophobic mismatch (28, 30), as well as membrane curvature and the shape or orientation of the protein (30), affect the aggregation behavior of transmembrane proteins. Though the surface-bound tH differs from transmembrane proteins in many respects, we found that decreasing the simulation temperature (i.e., increasing the hydrophobic mismatch) stabilizes tH clusters (Fig. S7). It is therefore important to ask if tH aggregation may also modulate the bilayer’s physical properties, such as line tension, hydrophobic mismatch and spontaneous curvature. For a bilayer with coexisting L_o and L_d domains, theory (34) and experiment (35) have shown that these properties exhibit a complicated interdependence. Here we computed line tension, lipid tilt angles, and pressure profiles to highlight some of the key effects of tH clustering on the bilayer.

Comparison of F_l’s (see SI Text) derived from the trajectories with and without tH yielded striking results. For each monolayer of the tH-free bilayer, we obtained F_l = 2.63 ± 0.41 pN and 1.93 ± 0.30 pN for the L_o and L_d sides of the interface, respectively. (Because the bilayer is symmetric, the line tension is approximately the same for the two layers.) The presence of tH clusters at the domain boundaries reduced both these values by 40% at the tH-containing monolayer while the effect on the tH-free monolayer was negligible. Clearly, tH clustering has a thermodynamic consequence on the bilayer structure. The link between line tension and membrane curvature (34, 36–38) implies that the tH-induced reduction of F_l may affect the curvature behavior of our model bilayer. To explore this issue, we analyzed the average orientation of the bilayer lipids at each monolayer by computing their tilt angle, ϕ_lipid, defined as the angle between a vector along the principal axis of lipids and the bilayer normal (Fig. 4, Fig. S9). The dramatically different ϕ_lipid values derived from the tH-bound and tH-free simulations indicates a major difference in the mechanical properties of the two bilayers, especially at the domain boundary (compare Fig. 4 A, B and C, D). The larger lipid tilt in the tH-bound bilayer is coupled with significant swaying of palmitoyl tails from clustered tHs toward the L_o and farnesyl toward the L_d domains, respectively (the corresponding effect of the backbone is described in the next section). Of note is the opposite sign of ϕ_lipid for lipids facing each other in the two monolayers of the tH-free but not the tH-bound bilayer, indicating that cluster-induced curvature is much greater than the comparatively small curvature arising from the L_o/L_d coexistence (Fig. 4B). To further examine the mechanical properties of the bilayer, we calculated pressure profiles along the bilayer normal using the method described in ref. 39 (Fig. S10). The profiles are asymmetric, with the magnitude of the tensors for the lower leaflet, where tH is bound, being significantly smaller than those for the upper leaflet, consistent with curvature of the bilayer visualized in Fig. 4D. Taken together, these data demonstrate that accumulation of tH at the domain interface in the lower leaflet leads to overall bilayer curvature. Results from simulations at other temperatures led to the same general conclusion: increased concentration of tH clusters at domain boundaries leads to higher curvature. These results strongly suggest that the attachment of H-ras on the cytoplasmic side of the PM induces membrane curvature.

Fig. 4.

Nanocluster-induced bilayer shape change. (A) Average lipid tail tilt angle from the bilayer normal in the tH-free bilayer. (B) Cross-section of a snapshot from the tH-free simulation showing the opposite curvature of the two monolayers. (C) Same as in (A) but for the tH-bound bilayer, with the tH-containing monolayer in solid line and the tH-free monolayer in a dashed line. (D) Same as in (B) but for the tH-bound bilayer, showing the same overall curvature of the two monolayers and a larger nanocluster-induced curvature at the L_o/L_d interface.

Organization of H-ras Lipid-Anchors within Clusters.

Orientation analyses based on time-averaged tH lipid tilt angles, , calculated as the angle between the bilayer normal and a vector along the principal axis of a given tH lipid tail x, where x is either Pa181, Pa184, or Fa186, indicate that tH lipids tilt in a similar manner as those of the bilayer lipids at the interface (see Fig. 5A). In fact, at domain boundaries, , indicating a correlated behavior between tH palmitoyl and DPPC hydrocarbon chains at the L_o-side of the interface. The magnitude of is smaller because the polyunsaturated and polymethylated farnesyl does not pack with the DLiPC lipids as efficiently as does palmitoyl with DPPC.

Fig. 5.

Organization of tH components in the bilayer. (A) Average tilt angle of tH lipids from the bilayer normal. (B) DPPC/(DPPC + DLiPC) fraction around the backbone of each tH residue. A 0.7 nm cutoff was used. Since CHOL represents only approximately 0.5 % of the molecules at the interface, the interfacial line is assumed to be located at a DPPC/(DPPC + DLiPC) ratio of 0.5.

An interface-resident tH molecule has to straddle between two domains of different thickness. Considering the short length of the backbone, such a task requires a specific orientation with respect to the membrane plane and the interfacial line. Indeed, the end-to-end vector of the backbone has an average angle of 70 ± 0.7° from the global bilayer normal. To further investigate this issue, we calculated the mole fraction of lipids “solvating” each backbone bead of the tH (Fig. 5B). The data at 8 °C shows that the backbone beads of Ser183 and Pa184 are at or near the center of the interface, which is defined as the region where the DPPC mole fraction is close to 50%. Fa186 is mostly surrounded by DLiPC lipids while Pa181 is surrounded by DPPC. To achieve this distribution, the tH backbone adopts an “S” shape that lies perpendicular to the interfacial line (see SI Text and Figs. S11 and S12 for details). The intrinsic length difference between the palmitoyl and the farnesyl tails allows for the N-terminal arc of the “S” to lie at the “taller” L_o domain and the C-terminal arc at the “shorter” L_d domain. Clustering minimizes the resultant solvent exposure of the hydrophobic palmitoyl tails, and the clustered molecules mostly align in parallel with their C terminus pointing to the L_d domain. At higher temperatures, a significant fraction of the clustered tH molecules maintain the same parallel orientation, but ordering is predictably less pronounced due to the higher entropy and smaller interdomain thickness difference. Nonetheless, the predominant organization of interfacial tH molecules within clusters at 28 °C can be described as follows (Fig. 6). The backbone aligns parallel to the curved surface of the bilayer so that Pa181 is in the L_o domain, Fa186 in the L_d, and Ser183 at the center of the interface. A twist of the backbone ensures that Pa184 is also near the center of the interface, and puts the C-terminal arc of the “S” at a smaller angle with the L_o/L_d interfacial line.

Fig. 6.

Schematic representation of the tH anchor configuration in the curved bilayer with the L_o domain in tan and L_d in blue. Backbone and phosphate beads in ball representation and lipid tails in lines, with straighter lines representing more ordered tails. tH backbone color-coding is the same as in Fig. 1A. Cholesterol is shown as gray ellipsoid within the L_o domain. For clarity, the distance between different tH backbone beads is exaggerated. The figure highlights the parallel alignment of the protein backbone, the L_o- and L_d-oriented palmitoyl and farnesyl tails, and membrane curvature (especially at the tH-containing monolayer).

Discussion

Cell biological experiments have shown that Ras proteins form transient protein-lipid nanoclusters on the cytosolic side of the plasma membrane (reviewed in refs. 2, 12). The size and protein content of these clusters is independent of Ras expression level but their number is proportional to the concentration of an external stimulant, such as a growth factor (4). These observations led to the suggestion that nanoclusters may serve as analogue-digital-analogue converters for high-fidelity signal transmission (4). Recent reports based on biophysical techniques established the ability of Ras proteins to form nanoclusters in model membranes as well (13, 14). Despite the intriguing implication of these insights for a unique mechanism of signal regulation at the plasma membrane, and the opportunity they present for a deeper understanding of membrane biophysics, very little has been known about the physical basis of cluster formation. As a result, it remained unclear why the highly homologues H-, N-, and K-ras proteins form distinct clusters (2).

The goal of the current work was to determine the physical principles underlying H-ras clustering in specific lipid domains, and to assess whether clustering modulates membrane mechanical properties. Using simulations of a model bilayer that forms nanoscale L_o and L_d domains both in the absence and presence of the minimal membrane anchor of H-ras (tH), we have shown that approximately 40% of the tH molecules spontaneously assemble into clusters of the same sizes as those from EM (1). The clusters are in dynamic equilibrium amongst each other and with the nonclustered pool of monomers. The fast (sub-μs time scale) dynamics of these proteolipid assemblies supports the view that nanoclusters are spontaneously forming dynamic structures that are stabilized by other membrane components in the cell (1, 12).

That clusters were found to form in both a domain-forming mixed bilayer and pure bilayers suggests clustering is an intrinsic property of tH, but the fact that the clusters are larger and more stable in the former highlights the importance of lipid domain stability (see Results). Note that cells may actively generate and maintain coexisting lipid domains to provide a stable platform for an even more efficient and stable clustering (40). The segregation of tH clusters to the L_o/L_d domain interface, whose formation is facilitated by CHOL, explains why the stability of tH nanoclusters was found to be sensitive to the CHOL-content of membranes. This finding also suggests an intriguing possibility that tH clusters in the PM may preferentially trace domain interfaces in a quasi-1-dimensional line rather than reside in a given domain. Monomers have less preference for interfaces, clearly indicating that the interaction of a single tH with interfacial bilayer lipids is not sufficient for confinement. Clusters of four or more tH cannot escape the interface, suggesting that interfacial localization is driven by a mechanism that involves supramolecular interaction. This interaction is dominated by the lipid tails of tH, as demonstrated by the complete redistribution of de-palmitoylated and de-farnesylated tH variants to the L_d and L_o region, respectively; farnesylated tH with at least one palmitoyl invariably clusters at domain boundaries (Fig. 3, Fig. S8). This finding implies that the same entropy-dominated mechanism that drives lipid phase separation underlies clustering and domain-specific distribution of Ras. Thus, in addition to revealing the hitherto elusive molecular basis of Ras clustering, our semiatomic data allowed us to probe the structural and thermodynamic impact of Ras clustering on the host membrane.

Fig. S1A shows an almost 1 nm L_o/L_d thickness difference in the tH-free bilayer, which resulted in a 4.5 pN line tension at each monolayer. This value is dramatically reduced in the presence of tH, where the interface-localized clusters reversed the curvature sign of the host monolayer and increased the overall bilayer curvature. Our finding is in line with a previous report that protein-induced membrane curvature results in a net attraction and aggregation of the bound proteins (41). We propose that in cells, too, clusters of palmitoylated/farnesylated proteins behave as lineactants that cooperatively reduce the line tension between membrane subdomains; the backbone of their lipidated motif will lie perpendicular to the interfacial line with its midpoint being at the center of the interface (see the location of Ser183 and the orientation of the tH backbone in Fig. 6). We further speculate that this organization will be facilitated by raft-like domains and fine tuned by the rest of the protein. In fact, the catalytic domain and its nucleotide ligand have been shown to modulate the clustering behavior of cellular H-ras (42). How these factors might affect H-ras clustering in our model membrane awaits exploration.

In sum, our simulations and model systems were able to capture the key aspects of Ras nanoclustering in the plasma membrane, and provided otherwise inaccessible insights into the complex relationship between membrane geometry and protein assembly (see Fig. 6). Our data further show that posttranslational lipid modification not only provides a hydrophobic membrane anchor but also plays a key role in the formation and spatiotemporal distribution of proteolipid nanodomains. These findings pave the way for a better understanding of signal transduction events mediated by protein clustering at the plasma membrane.

Methods

The simulations were carried out with the GROMACS program (43) and the MARTINI CG model (19, 20). The C-terminal minimal membrane anchor of H-ras (residues 180-186, tH, see Fig. 1A), a hetpapeptide containing two palmitoyls (Pa181 and Pa184) and a farnesyl (Fa186), was chosen to model lipid-modified proteins owing to its ability to form CHOL-dependent nanoclusters (42). Such a model allows for a detailed characterization of the role of lipidation in clustering without the potential interference of the rest of the protein. Thus, 64 tH molecules were embedded in one layer of a bilayer containing 960 DPPC, 576 DLiPC and 384 CHOL molecules (5∶3∶2). This mixture has been previously shown to form nanoscale domains at a slightly different ratio of 4.2∶2.8∶3 (17).

Two sets of simulations were performed in a box of 24 nm × 24 nm lateral dimensions. With the first set we studied the temperature dependence of tH clustering by performing simulations at 8, 18, 28, 38, and 48 °C in the absence and presence of tH (Table S1). The second set compared wild-type tH with its de-palmitoylated and de-farnesylated variants, where individual lipid modifications were systematically replaced by the parent Cys residue at 28 °C. Each simulation was run for an effective duration of 24–40 μs and the best equilibrated last 16 μs of each was used for analysis. Further details about the systems, simulation setups and data analyses is provided in SI Text.