Small GTPases regulate many aspects of cell logistics by alternating between an inactive, GDP-bound form and an active, GTP-bound form. This nucleotide switch is coupled to a cytosol/membrane cycle, such that GTP-bound small GTPases carry out their functions at the periphery of endomembranes. A global understanding of the molecular determinants of the interaction of small GTPases with membranes and of the resulting supramolecular organization is beginning to emerge from studies of model systems. Recent studies highlighted that small GTPases establish multiple interactions with membranes involving their lipid anchor, their lipididated hypervariable region and elements in their GTPase domain, which combine to determine the strength, specificity and orientation of their association with lipids. Thereby, membrane association potentiates small GTPase interactions with GEFs, GAPs and effectors through colocalization and positional matching. Furthermore, it leads to small GTPase nanoclustering and to lipid demixing, which drives the assembly of molecular platforms in which proteins and lipids co-operate in producing high-fidelity signals through feedback and feedforward loops. Although still fragmentary, these observations point to an integrated model of signaling by membrane-attached small GTPases that involves a diversity of direct and indirect interactions, which can inspire new therapeutic strategies to block their activities in diseases.
Cytoplasmic small GTPases constitute a large family of proteins with pivotal functions in most aspects of cellular logistics, and failures in their regulation are therefore associated with many diseases. The small GTPase family is divided into several subfamilies, including the Ras GTPases, which have essential functions in signal transduction, Rho GTPases, which regulate cell shape and motility, Rab GTPase, which determine membrane compartment identities and regulate traffic, and Arf/Arf-like GTPases, which regulate membrane traffic at the crossroads with signaling and cytoskeleton dynamics (reviewed in ref. ). A hallmark of small GTPases is their ability to switch between an inactive form bound to GDP, which is poorly competent for intermolecular interactions, and an active form bound to GTP, which binds to effectors to actuate cellular responses (reviewed in ref. ). Alternation between the inactive and active forms is precisely regulated by guanine nucleotide exchange factors (GEFs), which activate small GTPases by stimulating GDP/GTP exchange, and by GTPase-activating proteins (GAPs), which inactivate them by stimulating the hydrolysis of GTP (reviewed in ref. ). Another hallmark feature of all small GTPases, except the nuclear GTPase Ran, is their association with membranes, at the periphery of which they are activated by GEFs and inactivated by GAPs and they recruit effectors to organize signaling modules (Figure 1A). Here, we review our current knowledge on the association of small GTPases with lipids and membranes, and on how membrane-attached GTPases drive supramolecular organizations that shape signaling in time and space.
The GDP/GTP cycle of small GTPases takes place at the membrane periphery.
Determinants of small GTPase interactions with membrane lipids
Small GTPases of the Ras, Rho, Rab and Arf families comprise a conserved G domain, which carries the guanine nucleotide-binding site, and of an N- or C-terminal extension of variable size and sequence [hypervariable region (HVR) hereafter] which is post-translationally modified by one or several lipids and represents their primary site of interaction with membranes (reviewed in ref. ) (Figure 2A,B). The G domain features two regions, Switch 1 and Switch 2, which recognize the nature of the bound nucleotide and select protein partners when GTP is loaded (reviewed in refs [2,5]). Structures of close to a hundred small GTPases, often in multiple functional states, have elucidated the structural features and variability of the G domain and the modalities of its interactions with other proteins (reviewed in ref. ). In contrast, much less is known on the structures and interactions of the lipidated extensions, which were truncated in most crystal structures to facilitate crystallization. In a few complexes of Ras, Rho and Rab proteins with solubilizing factors (RabGDI, RhoGDI and PDEδ), the lipidated HVR could be observed at high resolution, depicting stretched peptides that extent away from the G domain (e.g. , reviewed in ref. ). In the context of membranes, HVRs may thus function as spacers to define the position of small GTPases with respect to the membrane, at distances that depend on their flexibilities and lengths (Figure 2C). For example, the structure of Rab6 bound to a distal site of the golgin GCC185 predicts that it is separated from the membrane from about the length of the HVR , which, conversely, implies that the length of the HVR of Rab6 contributes to its specific recognition of this golgin. Small GTPases of the Arf family represent a case apart, in which the myristoylated N-terminal extension folds as a well-ordered α-helix rather than an extended peptide (reviewed in ref. ).
Determinants of the interaction of small GTPases with lipids.
The major driving force of small GTPase association with membranes stems from their lipid anchor (Figure 2A) (reviewed in refs [4,10,11]). The nature of the lipid is determined by specific sequences in the HVRs, which are recognized by a diverse set of lipidating enzymes and lipid chaperones, which we do not discuss here. The most frequent lipid modification is the prenylation of cysteines located in C-terminal HVRs, including farnesylation in Ras family and a few Rho proteins, and single or double geranylgeranylation in Rab GTPases and most Rho GTPases. In addition, several Ras GTPases isoforms are reversibly modified by one or two palmitoylations (reviewed in ref. ). Farnesylated small GTPases are cleaved beyond the lipidated cysteine by a carboxypeptidase followed by carboxymethylation, as was originally shown for K-Ras (, reviewed in ref. ). Arf and Arf-like GTPases differ from Ras, Rab and Rho family GTPases in that they are modified by myristoylation on their N-terminal glycine, when present (reviewed in ref. ). Lipids attached to small GTPases wedge between membrane lipids, which enables them, in principle, to recognize specific lipids as well as bulk features of the lipid bilayer, such as packing, curvature and fluidity (reviewed in ref. ) (Figure 2B).
Post-translational lipids attached to small GTPases bind to lipid membranes with different strengths, with myristate having the weakest affinity [15,16]. Thus, in most cases, more than one lipid and/or additional interactions with the membrane are needed for stable association (Figure 2B). In addition to their lipid modifications, HVRs expose residues that can form hydrophobic or electrostatic interactions with lipids. NMR analysis showed that the N-terminal fragment of myristoylated Arf1 forms a helix that lies parallel to the membrane surface to insert aromatic and hydrophobic side chains into the lipid bilayer . Likewise, solid-state NMR indicated that the farnesylated C-terminus of N-Ras inserts two hydrophobic residues in the lipid bilayer . In isoprenylated HVRs, hydrophobic interactions are reinforced by C-terminal methylation, which also removes repulsive electrostatic interactions of the C-terminal carboxylate with anionic lipids (reviewed in ref. ). Although mostly unspecific, hydrophobic interactions may, however, contribute some recognition of collective membrane features, such as lipid packing or curvature. An interesting example in that regard is that of Arf GTPases, in which the amphipathic N-terminal helix has been associated with the recognition, and possibly induction, of positive membrane curvature [20–22]. It should be noted, however, that curvature sensitivity was modest compared with bona fide curvature sensors such as the ALPS motif found in ArfGAPs . Likewise, the N-terminal amphipathic helix of Sar1p, which is related to Arf GTPases but is not myristoylated, was shown to generate curved membranes, which were pivotal for coat assembly and cargo capture . Electrostatic interactions between anionic lipid headgroups and basic residues in the HVRs of many small GTPases play an important role in assisting their association to negatively charged membranes (Figure 2B). This was first shown for K-Ras, which lacks the second lipidation site found in other Ras isoforms and is directed to the plasma membrane by a polybasic stretch in its HVR . In a large-scale study of 125 small GTPases, 37 out of the 48 small GTPases that localized to the plasma membrane were found to feature four or more Lys or Arg in their HVR, and this association was dependent on PI(4,5)P2 and PI(3,4,5)P3 phosphoinositides . These observations predict that the presence of a polybasic tract in the HVR of a small GTPase is a signature for its recruitment to membranes enriched in negatively charged phosphoinositides. It should be noted that differences in the net positive charge of such HVRs can have a strong impact on their actual membrane-binding properties. This is highlighted by the comparison of the closely related Rac1 and Rac2 GTPases, whose HVRs have estimated charges of +7 and +3, which suffices to drive their preferential binding to the plasma membrane and internal membranes, respectively . In vitro reconstitutions showed that Rab proteins, which target membrane compartments with diverse physicochemical characteristics, use both their lipid anchors and the sequence of their HVRs to detect lipid packing defects, curvature or electrostatics differentially . As for the Arf proteins, the sensitivity of Rab GTPases to membrane curvature was, however, modest compared with the ALPS motif . The structural basis for the interaction of polybasic tracts with membrane lipids has remained an experimental ‘gray zone’, notably because phosphoinositides diffuse rapidly and interact transiently with basic amino acids of the HVR. This has been compensated for, in part, by molecular dynamic simulations of lipidated Ras GTPases in model lipid bilayers, which predict that HVRs sample preferred conformations, which, in turn, may define specificities for defined lipid compositions ([29,30], reviewed in ref. ).
The proximity of the lipidated HVR to the lipid bilayer makes it likely that it brings the G domain sufficiently close to the membrane to form direct interactions. Such interactions have been consistently detected by molecular dynamics and biophysical approaches, but discrepancies remain as to the modes of binding. Direct interactions of the G domain of H-Ras with a neutral lipid bilayer were first predicted by molecular dynamics simulations . These simulations predicted that the membrane-binding region spans β2–β3 loop, helix α4 and helix α5, which are located opposite to the effector-binding site (Figure 2B), and that this region has two different binding modes whose relative populations are sensitive to the bound guanine nucleotide. Molecular simulations predicted that the same region is involved in binding of K-Ras to membranes . Mutations of H-Ras and K-Ras in this region resulted in altered signaling in cells [29,32,33] and increased tumorigenicity , probably reflecting disruption of its interaction with the membrane. The direct interaction of the G domain of K-Ras with membranes has been detected experimentally, for example, by a shift in the α-helical subband in FTIR spectra  and by NMR using a K-Ras4 construct chemically tethered to anionic nanodiscs . An intriguing observation from the latter study is that, whereas GDP-bound K-Ras contacts the membrane opposite to the nucleotide-binding site as predicted by simulations, GTP-bound K-Ras contacts the membrane by the effector-binding region, suggesting that the membrane exerts autoinhibitory interactions . A binding mode involving the effector-binding region could be reproduced in simulations, but only when K-Ras carried an oncogenic mutation . In Arf GTPases, the G domain is close to the N-terminal lipidic anchor, which may facilitate contacts with the membrane (reviewed in ref. ). In addition, Arf and Arf-like GTPases have a unique switch element, the β2–β3 loop, whereby the nucleotide-binding site is allosterically coupled to the lipidic anchor (reviewed in ref. ). Accordingly, molecular dynamics simulations carried out in the presence of a GEF predicted that the β2–β3 loop of myristoylated Arf1 forms direct interactions with an anionic lipid bilayer . The NMR analysis of myristoylated Arf1–GTP in neutral lipid bicelles did not identify direct contacts of its G domain with lipids , which may be due to the lack of anionic lipids in the NMR setup.
Together, these observations, although still fragmentary and sometimes conflicting, point to an association of small GTPases with membrane that involves contributions of the post-translational lipid, the HVR and the G domain in various combinations. They also underline the importance of quantitative measurements in assessing the membrane-binding properties of small GTPases and of their comparisons to those of their regulators and effectors when inferring their functional relevance.
Membrane association shapes small GTPase signaling
There are many ways in which the association of small GTPases with membranes affects their functional activities. Association of small GTPases to subcellular membranes restricts their activities to specific locations in the cell, and it is therefore considered as a key feature of organelle identities (reviewed in ref. ). Whether such subcellular localization is entirely encoded within the small GTPase alone has remained an open question. Early studies of hybrid Rab GTPases with switched HVRs  and of Ras isoforms  suggested that the lipidated HVR is sufficient to define their subcellular localization. However, when switched across a larger group of Rab GTPases, the HVR alone was not always sufficient to specify localization or to support function, indicating that other elements were needed . A recent study, based on semisynthetic Rab GTPases, suggested that their HVRs have, in fact, complex contributions to their association with membranes, which, depending on the Rab GTPase, range from simple tethering to defining the specificity of their membrane targeting . At another extreme, the Arf1 GTPase localizes to membranes as diverse as Golgi, vesicular and plasma membranes, where it activates distinct effectors and pathways (reviewed in ref. ), suggesting that it contains determinants that allow it to recognize membranes of various physicochemical characteristics. For this family of GTPases and for Rabs at least, the current assumption is that GEFs, and possibly other factors, drive recruitment to and activation at specific sites in the cell (reviewed in refs [45,46]). Interestingly, the lipidated HVR of K-Ras alone was shown to be unable to sustain stable association with the plasma membrane, which was achieved dynamically by combined extraction by solubilizing factors, delivery to endomembranes and unidirectional vesicular transport (reviewed in ref. ). Together, these observations suggest that the subcellular localization of small GTPases is ultimately defined by determinants encoded in the GTPase itself and by their interactions with binding partners, which combine to enhance or decrease their association with specific membranes.
By associating with membranes, small GTPases increase their local concentration and hence the likelihood of their encounters with regulators and effectors. Membranes also provide an environment that can modify the activity of small GTPases, regulators and effectors directly, notably by inducing regulatory conformational changes. Together, entropic effects due to reduction in dimensionality and conformational effects combine to determine the efficiency and specificity of signaling activities originating from membrane-attached GTPases (reviewed in ref. ). In the last decade, quantitative studies of GEFs and GAPs reconstituted in artificial membranes uncovered a variety of regulatory mechanisms that require a membrane context (reviewed in ref. ). Notably, membranes play a key role in displacing autoinhibitory elements, either directly as exemplified by the bacterial ArfGEF RalF , or indirectly, for example, in cytohesin ArfGEFs in which the autoinhibitory PH domain is displaced by membrane-attached Arf-GTP to implement a positive feedback loop [50,51]. Bulk properties of membranes can also influence activities, as illustrated by activation of ArfGAP1 through recognition of curved membranes by its ALPS motif . At the other extreme, regulators that establish strong and stable interactions with specific lipids can only activate membrane-attached GTPases located in their vicinity, which decreases their overall efficiency while providing highly localized signaling [53,54]. Enhancement of GEF efficiencies towards membrane-associated GTPases compared with their activities in solution can also result in a broader specificity, as shown by reconstitution of the RhoGEF Trio in artificial membranes, which points to the same GEF-activating several small GTPases with different efficiencies in cells . A combination of these large and more subtle effects is thus anticipated to contribute to the overall efficiency and specificity of signaling, in diverse ways that remain to be fully elucidated.
Potentiation of GEF efficiencies by membranes, as originally shown for the RhoGEF Tiam , can reach three orders of magnitude even in the absence of conformational change , which cannot be accounted for by entropic effects only. An increasing number of recent studies, both theoretical and experimental, highlighted that the orientation of small GTPases and their protein partners with respect to the membrane constitutes an important signaling determinant. A recent survey of a large number of crystal structures of small GTPases in complex with effectors and regulators showed that the flat side located opposite to the nucleotide-binding site (see Figure 2C) is generally devoid of protein–protein interactions; this suggested that this region is available for direct interaction with membrane lipids, thereby defining the orientation in which the membrane-attached GTPase binds to regulators and effectors . The oriented apposition of small GTPases with respect to membranes implies that their productive interactions with regulators and effectors require that their protein–membrane and protein–protein interactions are precisely and simultaneously matched. Simultaneous optimization of protein–protein and protein–membrane interactions through multiple interactions with lipids could thus explain the extremely high efficiency of the ArfGEF BRAG2 towards myristoylated Arf [38,56] (Figure 3A). Optimization by coincidence detection of lipids and small GTPases is also important for their effectors, as exemplified by the Arf effector FAPP, which recognizes both Arf-GTP and phosphatidylinositol-4-phosphate by its PH domain . At the extreme, Arf-GTP was shown to be unable to interact with the GRAB domain of the golgin GMAP-210 in solution, and interaction requires the simultaneous binding of Arf and the GRAB domain to the membrane through their respective amphipathic helices . A measure of the importance of oriented association of small GTPases with the membrane is their functional susceptibility to small molecules that induce misorientation. This was recently shown for a small molecule that inhibits K-Ras functions by inducing the occlusion of its effector-binding site by the membrane , or for an inhibitor of the ArfGEF BRAG2 that induces misorientation of its membrane- and Arf-binding domain with respect to the membrane (our unpublished results).
Membrane association shapes small GTPases signaling.
It is generally assumed that signaling by small GTPases requires that a sufficient number of small GTPases are activated simultaneously to generate a signaling platform that produces downstream effects. Evidence is mounting, mostly focused on Ras proteins, that membrane association provides a means to segregate activated small GTPases in restricted membrane areas (reviewed in refs [31,61,62,63]). Quantitative imaging methods, notably immunogold electron microscopy of plasma membrane sheets coupled with pattern analysis, identified nanometer-sized domains containing 5–6 Ras proteins each, termed nanoclusters [64,65]. Rac1 nanoclusters of similar size were also recently observed by immunogold electron microscopy . Larger nanoclusters containing 50–100 immobilized Rac1 molecules were identified by single-molecule imaging and super-resolution microscopy . Nanoclusters have been proposed to facilitate the assembly of signaling particles with high fidelity (reviewed in ref. ) and/or to define signaling gradients .
Currently, whether the formation of nanoclusters involves small GTPase oligomerization has remained unclear. Several studies suggested that dimerization may play a role in their stabilization. Dimerization of membrane-associated Ras isoforms has been reported (e.g. [68,69,70]), and oncogenic effects in K-Ras were decreased by mutations predicted to impair its dimerization . Imaging of isoprenylated Ras isoforms reconstituted in supported lipid bilayers, however, questioned their ability to assemble into dimers [72,73]. Dimerization of Arf-GTP mediated by helix α1 has also been reported, and a mutation that disrupted dimerization also impaired vesicle formation . It should be noted that dimers associated by 2-fold symmetry, such as those proposed for Ras and Arf protein, are ‘closed’ structures that cannot assemble into larger oligomers in the absence of other components, which leaves open the question of their ability to function as the basic component in signaling assemblies and nanoclusters. Structural biology of the distribution of Arf GTPases in their complexes with effectors in informative in this regard. Notably, although many Arf effectors are dimers that can bind two Arf molecules, thereby assembling complexes with diverse membrane-binding expanses, curvature, hydrophobicity and electrostatics, none is compatible with an Arf dimer . Furthermore, recent cryo-electron tomography of the COPI coat assembled on membranes in vitro found that Arf1 occupies contrasting molecular environments within the coat, including a monomer and a trimer , which could not be inferred from studies of the elementary interactions of the Arf GTPase with itself or with the membrane. Thus, the role of dimers or higher order oligomers in constituting the seeds of nanoclusters remains to be fully elucidated.
Exciting developments with regard to the association of small GTPases with the membrane have emerged from the identification of supramolecular effects that impart to the membrane itself. A remarkable feature of K-Ras that has been extensively investigated is its ability to induce demixing of a specific phospholipid, phosphatidylserine. The HVR of K-Ras carries a stretch of positively charged residues that targets it to the plasma membrane through interactions with phosphatidylserine, resulting in its co-clustering with this lipid (reviewed in ref. ). Analysis of co-clustering effects by molecular dynamics and immunogold electron microscopy uncovered that the sequence of the HVR is necessary for the recognition of both the head groups and the saturation and length of the acyl chains of phosphatidylserine . Conversely, phosphatidylserine clustering was shown to be highly sensitive to conservative mutations (Lys/Arg) in the HVR , possibly due to the multivalent nature of the guanidinium group of Arg that engages lipids in a different manner than Lys . Rac1 co-clustered with different sets of lipid (phosphatidic acid and PI(3,4,5)P3), and this was also highly sensitive to mutations of basic residues in its HVR . Co-clustering of these small GTPases and specific lipids by the HVR has been proposed to define a ‘lipid code’, possibly specific to each small GTPase (reviewed in ref. ).
Co-ordinated protein nanoclustering and lipid demixing has the potential to drive feedback and feedforward effects, which have been best described for anionic phospholipids (reviewed in ref. ). In the case of small GTPases, positively charged HVRs bind to anionic phospholipids and induce their local enrichment, which, in turn, can be perceived by other phosphoinositide-binding proteins located in their neighborhood. This can lead, for example, to the recruitment of GEFs, resulting in amplification of small GTPase activation and nanoclustering, and/or of effectors through direct interactions with nanoclustered small GTPases or indirectly by their detection of lipid demixing effects (Figure 3B). In that schema, small GTPases, their regulators and effectors can locally accumulate into platforms to orchestrate signaling, in a manner where direct and indirect interactions reinforce each other.
Recent advances have highlighted that the interaction of small GTPases with membranes involves the post-translational lipid, the HVR and the G domain, which together determine the orientation of membrane-attached GTPases and how strongly they bind to membranes with specific lipid composition and bulk properties. These molecular determinants affect the subcellular localization, the efficiency and the specificity of signaling mediated by small GTPases through their regulators and effectors, and they participate in the clustering of lipids and proteins that result in high-fidelity signaling platforms that are finely controlled in time and space. While a general model integrating these molecular and supramolecular determinants is beginning to emerge, our general understanding of the underlying molecular modalities remains rudimentary. It can be anticipated that each individual small GTPase harnesses all contributions offered by the membrane context in a ‘weighed’ manner, such that different effects may dominate and/or combine to define unique signaling patterns. Conversely, dissolution of signaling platforms by GAPs implies that GAPs also perceive the membrane context in a supramolecular manner. Unraveling this diversity by integrated structural, biophysical and functional approaches will be key to understanding signal transduction originating from membrane-attached small GTPases, which may ultimately reveal new Achille's heels for therapeutic intervention (reviewed in refs [78,79]).
The present study was supported by grants from the Fondation pour la Recherche Médicale and the Institut National du Cancer to J.C. and F.P. was supported by a PhD grant from the ENS Paris-Saclay. We apologize to all authors whose work could not be cited in this review due to lack of space.
The Authors declare that there are no competing interests associated with the manuscript.