RAS nanoclusters are cell surface transducers that convert extracellular stimuli to intracellular signalling

Abstract

Mutations of rat sarcoma virus (RAS) oncogenes (HRAS, KRAS and NRAS) can contribute to the development of cancers and genetic disorders (RASopathies). The spatiotemporal organization of RAS is an important property that warrants further investigation. In order to function, wild-type or oncogenic mutants of RAS must be localized to the inner leaflet of the plasma membrane (PM), which is driven by interactions between their C-terminal membrane-anchoring domains and PM lipids. The isoform-specific RAS–lipid interactions promote the formation of nanoclusters on the PM. As main sites for effector recruitment, these nanoclusters are biologically important. Since the spatial distribution of lipids is sensitive to changing environments, such as mechanical and electrical perturbations, RAS nanoclusters act as transducers to convert external stimuli to intracellular mitogenic signalling. As such, effective inhibition of RAS oncogenesis requires consideration of the complex interplay between RAS nanoclusters and various cell surface and extracellular stimuli. In this review, we discuss in detail how, by sorting specific lipids in the PM, RAS nanoclusters act as transducers to convert external stimuli into intracellular signalling.

RAS small GTPases (encoded by HRAS, KRAS and NRAS) are key upstream regulators of mitogen-activated protein kinases (MAPKs), participating in cell growth, division, proliferation and migration [1,2]. As molecular switches, RAS toggles between the GDP-bound inactive state and the GTP-bound active state [3]. Under normal cell function, the guanine nucleotide exchange between GDP and GTP on RAS is tightly regulated [4]. However, in pathological conditions (including RASopathies and cancer), mutant RAS (particularly at glysine 12 (Gly12), Gly13 and glutamine 61 (Gln61)) remains constitutively active, contributing to disease [3]. In particular, a mutant isoform, KRAS4B, is found in 98% of pancreatic, 45% of colorectal and 31% of lung adenocarcinoma, as well as 23% of multiple myeloma [3,5,6]. Mutant NRAS is found in 28% of cutaneous melanoma, and mutant HRAS is found in 5% of head and neck squamous cell carcinoma and 6% of bladder carcinoma [3,5,6]. Although KRAS4A has been previously thought to be a minor variant, the latest findings strongly suggest that it is efficiently expressed in various cancers, such as colon cancer [7]. Together, oncogenic mutations of RAS occur in ~20% all human tumours [3,5,6]. Over the past 40 years, intense investigations have focused on discovering treatment options against oncogenic mutant RAS. Unfortunately, because their highly dynamic globular enzymatic G-domains are difficult for small-molecule inhibitors to bind, it has been challenging to find effective ways to inhibit these proteins [8]. Furthermore, in the clinic, patients treated with inhibitors against specific oncogenic mutants of RAS quickly develop adaptive resistance [9-13]. Thus, RAS oncogenesis may be multifaceted, involving more than just enzymatic activation and signalling.

An important aspect of RAS signalling in both normal biological and pathological activities involves their proper localization to the inner leaflet of the plasma membrane (PM) [5,14,15]. Typically, membrane-associating proteins utilize extensive protein–membrane contacts, such as transmembrane domains, amphipathic helices, lipid-binding domains and/or bin-amphiphysin/Rvs (BAR) domains, to localize to the PM [16]. However, RAS isoforms, including HRAS, NRAS and splice variants KRAS4A and KRAS4B, rely on their post-translationally modified C-terminal lipid anchors for PM binding (Fig. 1). The post-translational modification of RAS has been described in detail in previous reviews [14,15,17,18]. Briefly, all RAS isoforms share a poly-unsaturated and highly branched 15-carbon farnesyl anchor conjugated to the C-terminal cysteine residue (Cys186 for HRAS, NRAS and KRAS4A; Cys185 for KRAS4B), which is also carboxymethylated (Fig. 1). This farnesyl chain is one of the main drivers of membrane anchoring for RAS [5,19-21]. Isoforms HRAS, NRAS and KRAS4A are additionally palmitoylated (fully saturated and unbranched 16-carbon), with HRAS dually palmitoylated on Cys181 and Cys184, respectively, NRAS singly palmitoylated on Cys181 and KRAS4A singly palmitoylated on Cys180 (Fig. 1).

Fig. 1.

RAS isoforms possess distinct C-terminal membrane-anchoring domains. RAS isoforms, including HRAS, NRAS, splice variants KRAS4A and KRAS4B, share near identical G-domains (sequence homologies >95%) and distinct C-terminal hypervariable regions (sequence homologies <20%). The C-terminal cysteines of all RAS isoforms are farnesylated and methylated. HRAS is additionally dual-palmitoylated and NRAS is mono-palmitoylated. KRAS4A is palmitoylated and contains a short stretch of positively charged lysine residues, whereas KRAS4B possesses a long polybasic domain comprising 8 lysine residues with not additional fatty acid chains.

Another membrane-interacting feature of RAS is the polybasic domains, with KRAS4A containing a short polybasic sequence (KIKK) between the palmitoyl chain and the C-terminal farnesyl chain, and KRAS4B possessing contiguous polybasic residues (Lys175-180, Fig. 1). Distinct interactions of various lipids with these combinations of different lipid anchors and/or polybasic domains of RAS isoforms give rise to the formation of nanometer-sized domains, termed nanoclusters [19-21]. These nanoclusters are the main sites at which effector recruitment occurs [22-27]. As such, even the constitutively active oncogenic mutants of RAS still must segregate to appropriate nanoclusters on the PM for efficient recruitment of their effectors and activation of MAPK signalling [28,29].

Initially, perturbation of the PM localization of RAS was considered as a strategy to inhibit oncogenesis of mutant RAS. In particular, since the farnesyl anchor is instrumental to the PM localization of all RAS isoforms, farnesyltransferase inhibitors (FTIs) were used to prevent the covalent conjugation of the fatty anchor [19,30,31]. Although this strategy was successful in interfering with oncogenic activities of mutant HRAS [32], it did not work for the most prevalent isoform, KRAS4B [19,30,31]. When treated with FTIs, KRAS4B undergoes alternative geranylgeranylation to covalently link a polyunsaturated and highly branched 20-carbon geranylgeranyl chain to Cys185 [19,30,31]; the geranylgeranylated KRAS4B still efficiently localized to the PM and remained transforming [19,30,31]. As such, the community became less enthusiastic about the strategy of targeting the membrane interactions of mutant RAS. However, the latest advances in quantitative super-resolution imaging, in vitro biophysical assays, bioengineering fabrication and theoretical predictions together provide intriguing new evidence that the C-terminal anchors of RAS proteins possess intricate codes for highly selective interactions with distinct PM lipids, which allow RAS to laterally compartmentalize, to recruit specific effectors and to respond to different extracellular perturbations in an isoform-specific manner [14,15]. These new findings yield remarkable insight into the spatiotemporal regulation of RAS oncogenesis, which may inspire alternative strategies for treating RAS-dependent diseases.

In this review, we will discuss in detail how RAS isoforms undergo layers of compartmentalization on the PM. The more refined nanoclustering depends on highly specific RAS–lipid interactions, thus allowing RAS nanoclusters to act as transducers to convert external stimuli into intracellular signalling. Tumour cells use these signalling platforms as sensors for the tumour microenvironment, which is a key influencer of tumour aetiology, maturation and adaptive resistance. Future treatment strategies may be developed to target the multifaceted oncogenesis of RAS. Although this review focuses on RAS, other lipid-anchored small GTPases may undergo similar select lipid-sorting mechanisms that will contribute to their biological and pathological activities.

Phospholipids laterally segregate to distinct domains in membranes

The main driving force of RAS nanoclustering on the PM is the compartmentalization of constituents in the PM. The most prevalent description of native cell membranes is the fluid mosaic model, first developed by Singer and Nicolson [33]. In this model, membrane proteins distribute in an ocean of phospholipids. As such, interactions with membranes, which are commonly thought as homogeneous fluids, have been largely considered as lacking any specificity worthy of pharmaceutical targeting. However, recent in silico simulations and in vitro biophysical assays, as well as imaging and signalling studies in cells, revealed exciting new evidence that, instead of being homogeneously distributed in a two-dimensional ocean, lipids laterally segregate to co-existing domains in bilayers and membranes [34]. Sizes, lifetimes and morphologies of these domains vary widely and are determined by the composition of membrane bilayers [35]. Focusing on the most abundant phospholipids in cells, the zwitterionic phosphatidylcholine (PC) species undergo de-mixing in simple bilayers comprising two or three PC species [36,37]. At physiological temperature, the PC species with long and saturated acyl chains prefer to associate with cholesterol (CHOL) in the formation of the liquid-ordered (L_o) domains, whereas the PC species with unsaturated acyl chains segregate to form the liquid-disordered (L_d) domains [36,37]. Lipids in the L_o domains are more tightly packed and diffuse slower, whereas lipids in the L_d domains are more loosely packed and diffuse faster [36]. A similar phase separation also occurs in native cell membranes [34]. The CHOL-enriched domains in cell membranes are called lipid rafts, while the CHOL-poor domains in cell membranes are referred to as non-rafts [34]. Because cell membranes are composed of >4000 lipid species [34], along with hundreds of thousands of membrane proteins and intracellular cytoskeletal structures and extracellular matrix, lipid domains in native cell membranes are more dynamic and much smaller, with diameters of around 20–50 nanometers [14]. Membrane-associating proteins, depending on their membrane-interacting structures, prefer different physical/electrical environments in the coexisting L_o or L_d domains in bilayers, and rafts or non-rafts in native membranes, thus spatially segregating in the biomembranes [34,38]. Here, we will discuss how shifting lipid distributions among various domains impacts the structural integrity of the proteolipid nano-assemblies of RAS nanoclusters.

Although most of the biophysical studies on phase separation have focused on the zwitterionic PC species, how charged phospholipids behave in membranes remains less clear. In particular, anionic phospholipids, such as phosphatidylserine (PS), phosphatidic acid (PA) and phosphoinositols (PI), play important biological roles in cells. Because of their charged headgroups, most anionic phospholipids are considered for their electrostatic interactions. Furthermore, many proteins possess specific lipid-binding domains, such as C2, Pleckstrin homology (PH) and phagocytic oxidase homology (PX) domains, which target the headgroups of various anionic phospholipids with high affinity [39]. For these reasons, anionic phospholipids are commonly considered for their headgroup interactions. However, the latest studies are revealing that different species of an anionic phospholipid type with distinct acyl chains also undergo phase separation and distribute to different regions of membranes. For instance, PS is the most abundant anionic phospholipid in mammalian cells [40]. Non-overlapping PS domains have been shown to exist in the cell PM [23,40,41]. PS has been shown to associate with the CHOL-independent KRAS4B nanoclusters in the PM (will be elaborated below, [23-26,42]). On the other hand, acute depletion of CHOL via methyl-β-cyclodextrin (MβCD) mislocalized PS on the PM [40]. Together, these data suggest that separate CHOL-enriched and CHOL-poor PS pools coexist in the PM of cells.

Indeed, this was illustrated in an electron microscopy (EM)-bivariate co-clustering analysis between the CHOL and PS probes on intact PM sheets [26]. Briefly, phosphatidylserine auxotroph (PSA3) cells co-expressing a PS-binding domain [green fluorescence protein (GFP)-LactC2] and a CHOL probe (red fluorescence protein [RFP]-D4H) were depleted of endogenous PS and acutely supplemented with synthetic PS species. After immunolabeling with 2-nm gold nanoparticles linked to anti-RFP antibody and 6-nm gold coupled to anti-GFP antibody, respectively, distribution of gold particles on the intact PM sheets was imaged using transmission EM (TEM) at a magnification of 100 000×. Co-clustering between the 2-nm and 6-nm gold populations was calculated using the Ripley’s bivariate K-function analysis [18,26]. The EM-bivariate co-clustering analysis showed that the two lipid-binding domains only efficiently co-clustered when supplemented with the fully saturated 1,2-distearoyl-sn-glucero-3-phospho-L-serine (di18:0) PS, but not with other PS species with unsaturated acyl chains [26]. These data strongly suggest the existence of separate PS pools. The compartmentalization of the cell membranes has become more refined and the nanodomains in the biomembranes are now viewed to possess distinct compositions of lipids and proteins. This suggests that the signalling nanodomains can be specifically targeted to perturb cell function.

In synthetic bilayers, the L_o/L_d phase separation is sensitive to various physical conditions of the surrounding environments, such as temperature, electrolyte concentrations, osmotic stress, and mechanical and electrostatic perturbations [43-48]. For instance, Veatch and Keller demonstrated that a two-component bilayer is in a homogeneous gel phase at low temperatures, undergoes effective solid/liquid phase separation at higher temperature and coalesces to a homogeneous liquid phase at even higher temperature [43]. The actual threshold temperatures at which the lipids transition between phases depend on the acyl chain structures of the lipids in bilayers [43]. Similar phase separation also occurs in native biomembranes. Because of the complex composition of biomembranes, the coexisting domains in biomembranes are much smaller and more dynamic. This is nicely illustrated in the native biomembranes of the giant plasma membrane vesicles (GPMVs) isolated from live cells. Levental and colleagues showed that, on a micrometer level, GPMVs displayed homogeneous distribution at physiological and room temperatures [49,50]. The temperatures, at which half of the GPMVs underwent L_o/L_d phase separation were between 15 and 20 °C [43].

In micropipette aspiration experiments, generating mechanical strains induced coalescence of L_o/L_d co-existing domains in giant unilamellar vesicles (GUVs) [44]. Intercalation of amphipathic molecules, such as nonsteroidal anti-inflammatory drugs (NSAIDs) and bile salts, promoted further phase separation and stabilized L_o/L_d coexisting domains [51-54]. Electrical potential applied either across the bilayer norm or along the surface of a bilayer also induced phase separation [45-47]. Interestingly, with both hypo-osmotic-stress-induced swelling and hyper-osmotic-stress-induced ruffling, GUVs underwent dynamic oscillation between L_o/L_d phase separation and homogeneous coalescence of domains [48]. Taken together, these results show that the lateral distribution and phase separation of lipids are highly sensitive to various external perturbations and stimuli. As discussed below, RAS isoforms possess specific preferences for the L_o or L_d domains in synthetic bilayers, as well as rafts or non-rafts in the native PM. These isoform-specific domain preferences allow the RAS–lipid nanoclusters to act as transducers to convert environmental perturbations to intracellular signalling.

RAS isoforms display distinct preferences for CHOL in model bilayers and native PM

Because lipids undergo phase separation and laterally segregate to spatially distinct domains, RAS molecules interact with different lipids in the formation of nanoclusters on the inner leaflet of the PM. The first evidence for such lipid specificity came from a variation of the super-resolution EM-spatial analysis [55]. In this methodology, intact PM sheets of cells expressing a GFP-tagged RAS were attached to EM grids and immunolabeled with 4.5-nm gold nanoparticles conjugated with anti-GFP antibody. Following EM imaging, the spatial distribution of gold particles was calculated using the Ripley’s K-function analysis [18]. All RAS isoforms were found to form nanoclusters on the PM, with an average diameter of ~20 nm (Fig. 2). RAS nanoclusters are the main sites for effector recruitment because all RAS effectors possess their own specific lipid-binding domains, and synergistically bind to RAS (using their RAS-binding domains) and specific PM lipids (using their specific membrane-anchoring domains) for stable PM localization, activation and signal propagation. Disruption of RAS nanoclusters not only decreases the number of RAS molecules that can be targeted by effectors, but also dilutes the local concentration of specific lipids required for effector recruitment to the PM. This is nicely illustrated in the case of KRAS4B and its main effector, CRAF. As discussed below, KRAS4B nanoclusters are selectively enriched with PS and PA lipids [23-26,42,56-62]. Effector CRAF has a PS-binding domain in its cysteine-rich domain and a separate PA-binding domain in its C-terminus [63-65]. Disruption of KRAS4B nanoclustering diluted PS/PA contents near KRAS4B, compromised recruitment of effector CRAF, and inhibited RAS-dependent MAPK signalling and MAPK-dependent cell proliferation [23-25], as well as oncogenic activities of mutant RAS, in tumour cells and xenografts [56-62]. Taken together, the PM nanoclusters are the main signalling platforms of RAS on the cell surface.

Fig. 2.

RAS proteins form nanoclusters in separate domains in the PM in isoform-specific and guanine-nucleotide-binding-specific manners. EM-spatial analysis, FLIM, AFM and MD simulations consistently illustrate that RAS isoforms segregate to distinct domains in the PM and model bilayers. Inactive HRAS.GDP and active NRAS.GTP distribute to the domain boundary between CHOL-dependent and -independent domains. Active HRAS.GTP, inactive NRAS.GDP, active KRAS4Aand active/inactive KRAS4B all distribute to the CHOL-independent domains. Interestingly, all these nanoclusters are spatially segregated from each other on the PM of unperturbed cells. Various environmental perturbations will induce coalescence of these nanodomains. CHOL, cholesterol.

The main driver of RAS lateral nanoclustering on the PM is the interactions between their C-terminal lipid-anchoring domains and PM lipids. Nanoclustering of GFP-tagged inactive HRAS.GDP was disrupted by the acute depletion of PM CHOL via MβCD [22], suggesting that nanoclustering of inactive HRAS is CHOL dependent (Fig. 2). On the other hand, nanoclustering of active GFP-HRAS.GTP was independent of CHOL depletion (Fig. 2) [22]. This is consistent with further bivariate EM-co-clustering analysis, where GFP-HRAS.GDP and RFP-HRAS.GTP did not co-cluster [22], suggesting that inactive and active HRAS spatially segregate to different regions of the PM. This was consistent with the molecular dynamic (MD) simulations and fluorescence lifetime imaging microscopy–fluorescence resonance energy transfer (FLIM-FRET), which showed that, when GDP-bound, the G-domain of HRAS is farther away from the lipid bilayer [66,67]. This configuration allows the fully saturated palmitoyl chains to fully insert into the bilayer and extensively interact with a tightly packed CHOL-enriched domain [68-70]. GTP binding caused re-orientations of the β2-β3 loops and helix α5, which led to enhanced interactions between the charged lipids and polar resides of helix α4 of the G-domain of HRAS [66,67]. This re-orientation of the G-domain caused the palmitoyl anchors to be partially pulled out of the bilayer and interact with the more loosely packed CHOL-poor domains. The simulated predictions provided mechanisms for how inactive and active HRAS segregate to separate CHOL-enriched and CHOL-poor domains in the cell PM.

Further coarse-grained MD (CG-MD) simulations predicted that the minimal HRAS anchoring domain preferred to localize to the domain boundaries between L_o and L_d domains, and, intriguingly, induced bending of the bilayer [71]. The dual palmitoyl chains on Cys181 and Cys184 of HRAS differentially contribute to the domain preferences of HRAS (Fig. 3). Indeed, comparing the CHOL dependence of the mono-palmitoylated mutants HRAS.C181S and HRAS.C184S in EM-nanoclustering analysis revealed that conjugation of the palmitoyl chain to Cys181 mainly contributes to a CHOL preference, whereas the palmitoyl chain attached to Cys184 guides HRAS to the Golgi apparatus (Fig. 3) [72].

Fig. 3.

Lipid anchors of RAS isoforms differentially contribute to membrane localization. Although the C-terminal farnesyl chain is important for general localization to membranes, the palmitoyl chain attached to Cys181 guides HRAS to the PM and the palmitoyl chain linked to Cys184 is key for Golgi localization. While on the PM, the palmitoyl chain on Cys181 is more responsible for the recognition of the L_o domains. Similar to HRAS, the palmitoyl chain conjugated to Cys181 of NRAS guides the isoform to the CHOL-dependent domains on the PM. While the farnesyl chain guides the protein to all membranes, the hexa-lysine polybasic domain provides further fine-tuning and guides the protein to the PM. Once on the PM, lysine residues play different roles in sorting distinct PM lipids. Lys177 and Lys178 are particularly important in the sorting of PS lipids in the PM. PS, phosphatidylserine.

In addition to the lateral nanoclustering, palmitoyl chains of HRAS facilitate the intracellular transport of HRAS. Specifically, Roy et al. [72] used confocal imaging and showed that HRAS-G12V mono-palmitoylated on Cys181 localized to multiple compartments, including the Golgi, endoplasmic reticulum (ER) and the PM, whereas HRAS-G12V mono-palmitoylated on Cys184 had a strong preference for the Golgi over the PM.

Via probing photoactivatable GFP, Rocks et al. [73] further found that a single palmitoyl chain, either on Cys181 or Cys184, samples all membrane compartments in cells. Strikingly, the Bastiaens group further illustrated that NRAS conjugated to a palmitate analogue with a non-cleavable thioester bond rapidly sampled non-specifically to all membrane compartments [73]. Since palmitoylation is reversible, de-palmitoylation and re-palmitoylation allow HRAS to rapidly exchange between the Golgi and the PM, thus resulting in precise sampling of intracellular compartments [74,75]. These data nicely show that de-palmitoylation and re-palmitoylation dynamically regulate membrane trapping. Thus, the dual palmitoyl chains of HRAS contribute to its preferences for lateral PM nanoclustering and intracellular transport.

NRAS is mono-palmitoylated on Cys181 and farnesylated on Cys186, similar to the mutant HRAS-C184S described above. EM-nanoclustering analysis showed that acute CHOL depletion using MβCD effectively disrupted nanoclustering of active GFP-NRAS.GTP, without affecting nanoclustering of inactive GFP-NRAS.GDP on the intact PM sheets (Fig. 3) [22]. These spatial data, again, suggest a guanine-nucleotide-dependent spatial segregation of a RAS isoform (Fig. 2).

On supported bilayers composed of co-existing L_o/L_d domains in atomic force microscopy (AFM) experiments, NRAS preferentially localized to the boundaries between L_o and L_d domains [76]. EM-bivariate analysis further confirmed that the active GFP-NRAS.GDP and RFP-NRAS.GTP spatially segregated from each other on the PM [22]. Intracellularly, the mono-palmitoylated NRAS efficiently localizes to the Golgi, similar to the mono-palmitoylated HRAS [73,74].

KRAS4B is a unique RAS isoform because its only lipid anchor is the commonly shared farnesyl chain linked to the C-terminal Cys185 (Fig. 1). KRAS4B has a distinct contiguous hexa-lysine polybasic domain (Lys175-180, Fig. 3). EM-nanoclustering analysis showed that CHOL depletion via MβCD had no effect on nanoclustering of either inactive GFP-KRAS4B.GDP or active GFP-KRAS4B.GTP (Fig. 2) [22,77]. This was further confirmed in AFM experiments using purified KRAS4B on supported bilayers, where KRAS4B molecules distributed to the L_d domains but not the L_o domains [78]. Interestingly, bivariate EM-co-clustering analysis further showed that the inactive GFP-KRAS4B.GDP spatially segregated away from the active RFP-KRAS4B.GTP [22,77]. Taken together, RAS isoforms undergo guanine-nucleotide-dependent spatial segregation on the PM.

KRAS4A is the least studied splice variant of RAS because it has been previously thought to be a minor variant [7,79,80]. However, recent studies revealed that KRAS4A was widely expressed in various tumour cells [7,79,80]. EM-bivariate co-clustering analysis showed that the GFP-tagged CHOL probe (D4H) did not colocalize with RFP-KRAS4A.GTP [81], suggesting that the PM nanoclusters of active KRAS4A were depleted of CHOL. This was consistent with the finding that acute addback of exogenous CHOL did not affect the nanoclustering of KRAS4A [58]. Taking together, KRAS4A potentially favours distributing to the CHOL-poor L_d domains (Fig. 2). Further experiments will be needed to gain a better understanding of the local lipid environment of KRAS4A.

Electron microscopy-spatial analysis, in vitro biophysical assays and MD simulations consistently showed an interesting observation: that active HRAS.GTP, inactive NRAS.GDP, inactive KRAS4B.GDP and active KRAS4B.GTP all distribute to the CHOL-poor domains. Yet, these RAS nanoclusters are all spatially segregated from each other. Similarly, HRAS.GDP and NRAS.GTP both localize to the CHOL-enriched domains, but still segregate from each other. These data strongly suggest that further compartmentalization occurs within the non-raft, as well as raft, domains. These observations also suggest that other lipids, in addition to CHOL and PC, contribute to the observed further compartmentalization of these RAS isoforms.

PM nanoclusters of RAS isoforms selectively sort distinct anionic phospholipids

As suggested above, select enrichment of other lipids, in addition to PC and CHOL, may contribute to further spatial segregation of RAS isoforms within a raft or a non-raft domain in the PM. To explore lipid compositions of RAS nanoclusters, EM-bivariate analysis was used to calculate co-clustering between a GFP-tagged lipid-binding domain and an RFP-tagged RAS on intact PM sheets. These analyses focused on anionic phospholipids because a major portion of the PM inner leaflet comprises anionic phospholipids, which have been traditionally thought to primarily contribute to the negative charges on the PM inner leaflet [82]. Many anionic phospholipids, such as PA, phosphoinositol 4,5-bisphosphate (PIP₂) and phosphoinositol 3,4,5-triphosphate (PIP₃) also act as messenger molecules and directly participate in cell signalling. The EM-bivariate analysis showed that, while the inactive HRAS.GDP nanoclusters were enriched with PIP₂ and CHOL, active HRAS.GTP nanoclusters were enriched with PIP₃ [23,24]. The nanoclusters of the active KRAS4B.GTP were also enriched with PS and PA [23,24]. Interestingly, the spatially segregated nanoclusters of inactive HRAS.GDP, active HRAS.GTP and active KRAS4B.GTP all contained high levels of PS lipids. However, depleting endogenous PS in PSA3 cells, via growing cells in dialyzed fetal bovine serum (DFBS), disrupted the nanoclustering of KRAS4B and mislocalized KRAS4B from the PM, without affecting the PM distribution of inactive and active HRAS [23]. Acute addback of only natural extract of mouse brain PS lipids, but not other lipids tested (PC, phosphatidylethanolamine [PE], PIP₂ and CHOL), effectively restored the PM distribution of KRAS4B in cells depleted of endogenous PS [23-26,42,58]. These data suggest that PS is a structural component of KRAS4B nanoclusters, but is not for HRAS.

Proportions of PC and CHOL components in phase diagrams nicely illustrate their distinct roles in phase separation and spatial segregation in bilayers [36]. To explore the effects of anionic lipids on phase separation, EM-bivariate analysis calculated co-clustering between active KRAS4B.GTP and HRAS.GTP, which spatially segregate on the PM of unperturbed cells. Depleting endogenous PS or elevating the PS levels above normal levels promoted coalescence of KRAS4B.GTP and HRAS.GTP, which only spatially segregate within a narrow range of PS levels found in unperturbed cells [23]. These data suggest that spatial segregation and phase separation in cell membranes are highly dynamic events. Furthermore, PS lipids behave in a similar manner as PC lipids in their ability to mediate phase separation in biomembranes.

KRAS4B polybasic domain possesses intricate coding for sorting lipid headgroups and acyl chains in the PM

The polybasic domain of KRAS4B has been commonly considered as a charge sensor that interacts with anionic lipid headgroups in the PM inner leaflet. The observation that the PM nanoclusters of KRAS4B are selectively enriched in the monovalent PS but not the highly charged and multivalent PIP₂ is intriguing [23-26,42]. Its selectivity towards the less charged PS suggests non-electrostatic contribution to the interactions between the KRAS4B polybasic domain and the PM. Indeed, a combination of EM-spatial analysis and FLIM-FRET revealed that the equivalently charged single-point mutants of the hexa-lysine polybasic domain of KRAS4B sorted distinct PM lipids (Table 1) [25]. Especially interestingly, mutants KRAS4B-K177Q and KRAS4B-K178Q switched their PS preferences to PIP₂ (Fig. 4 and Table 1) [25]. Since PIP₂ level in the PM is much lower than that of PS, this switched PIP₂ preference resulted in a decrease in PM localization and a disrupted nanoclustering of KRAS4B-K177Q and KRAS4B-K178Q [25]. Concordantly, KRAS4B-K177Q and KRAS4B-K178Q also preferred to activate the PIP₂-dependent PI3K/Akt pathway, instead of the normal PS-associating RAF/MEK/ERK pathway [25]. All-atom MD and meta-dynamics simulations predicted that the farnesylated polybasic domain, an intrinsically disordered structure traditionally thought to assume random structures, sampled distinct and well-defined conformational states on lipid bilayers [25,26]. Among the ordered, intermediate and disordered conformational states, the original farnesylated hexa-lysine polybasic domain favoured the disordered states (~60% of the simulated structures), whereas KRAS4B-K177Q and KRAS4B-K178Q favoured ordered states [25]. Further analysis showed that the polybasic domain in the disordered states established more hydrogen bonds with PS headgroups than the polybasic domain in the ordered states [25]. Thus, unlike traditionally thought, the polybasic domain of KRAS4B undergoes the well-defined conformational sampling that provides the structural basis for its select lipid sorting.

Table 1.

Lysine residues in the KRAS4B polybasic domain play distinct roles in sorting distinct PM lipids. EM-spatial analysis, FLIM-FRET and MD simulations consistently show that each lysine residue differently contributes to the select sorting of various PM lipids.

Single-point mutations of KRAS4B polybasic domain	Altered lipid enrichment when compared with the original KRAS4B polybasic domain
K175Q	Enhanced PIP3
K176Q	No change observed
K177Q	Enhanced PIP2; Depleted PS
K178Q	Enhanced PA and PIP2; Depleted PS
K179Q	Enhanced PIP3
K180Q	No change observed

Fig. 4.

Equivalently charged KRAS4B polybasic-domain constructs sort distinct lipids in the PM. EM-spatial analysis, FLIM and all-atom MD simulations systematically examined lipid-sorting capabilities of four equally charged polybasic-domain constructs of KRAS4B, including the original farnesylated hexa-lysine (6K), farnesylated hexa-arginine (6R), geranylgeranylated hexa-lysine (C20) and geranylgeranylated hexa-arginine (6R-C20). Although these constructs possess identical charges, they sort distinct lipids via sampling different conformational states on membranes.

As a result of the conformational sampling, each lysine residue within the KRAS4B polybasic domain plays a distinct role in interacting with different PM lipids (Table 1). MD simulations predicted that the KRAS4B polybasic domain assumes a pseudo-helical hairpin structure, with its centre portion inserting into the bilayer [25,26,83]. This structure explains why mutating the lysine residues in the middle part of the polybasic domain (Lys177 and Lys178) had the biggest impact on the lipid sorting of the polybasic domain [25]. Furthermore, this favoured hairpin structure suggests that the KRAS4B polybasic domain possesses the capability to interact with lipid acyl chains. EM-spatial analysis showed that depletion of endogenous PS effectively mislocalized KRAS4B from the PM and disrupted the nanoclustering of KRAS4B left on the PM [23-25,56,58-60]. In the PS-depleted cells, subsequent acute addback of the synthetic PS species with unsaturated acyl chains effectively restored the PM localization of KRAS4B, whereas the saturated PS had no effect on the PM transport of KRAS4B [25,26,42]. Laterally, only the mixed-chain PS species, such as 1-palmitoyl-2-oleoyl-sn-glycero-3-phospho-L-serine (16:0/18:1 PS) and 18:0/18:1 PS, effectively recovered the nanoclustering of KRAS4B, whereas other symmetric PS species had no effect on the nanoclustering of KRAS4B [25,26,42]. A similar selectivity for different PS species with distinct acyl chain structures was also demonstrated in vitro using surface plasmon resonance (SPR) to measure binding of purified KRAS4B to large unilamellar vesicles (LUVs) composed of different PS species [84]. These data strongly suggest that the KRAS4B polybasic domain selectively sorts mixed-chain PS species, displaying specificities for lipid headgroups and acyl chains.

RAS nanoclusters act as transducers on the cell surface

A transducer converts one form of energy to another. Such events occur on the cell surface constantly. Cell function is sensitive to changes of surrounding temperature, electrolyte concentration, osmotic stress, mechanical stress, shear flow, etc. Tumour cells are also very sensitive to extracellular environments. Solid tumours possess distinct mechanical environments compared with normal tissues. During metastasis, tumour cells must alter their morphologies drastically and experience various distinct environments in different tissues and the circulation. How these thermal, mechanical and electric stimuli are converted to intracellular signalling events is not clear, although many factors, such as membranes, ion channels, surface receptors, cytoskeletal structures and the extracellular matrix, have been suggested to contribute.

As described above, shifting lateral distribution of lipids, such as phase separation, is sensitive to changes in extracellular environments. Also described above, RAS isoforms prefer to distribute to different L_o/L_d domains in synthetic bilayers or raft/non-raft regions in native cell membranes. Thus, the structural integrity and signalling capabilities of RAS nanoclusters may respond to changing external environments via selectively sorting distinct lipids in the PM. This was, indeed, demonstrated in recent studies. In unperturbed cell PM, the active and inactive HRAS spatially segregate to L_d and L_o domains, respectively. Acute depletion of CHOL in the PM via MβCD induced coalescence of the inactive HRAS.GDP and active HRAS.GTP in the PM, which inhibited the signalling of HRAS [85].

Remote crosstalk with caveolae

Caveolae are well-established CHOL-dependent lipid raft structures on the PM. In a series of EM-spatial analyses, disruption of caveolae on the PM, via knocking out caveolin 1 (CAV1), enhanced the nanoclustering of KRAS4B but disrupted the nanoclustering of HRAS [85]. Interestingly, knocking out CAV1 effectively abolished the spatial segregation between the inactive HRAS.GDP and the active HRAS.GTP, and induced the coalescence between the HRAS molecules bound with different guanosine nucleotides [85]. Intriguingly, RAS isoforms are not incorporated in caveolae [23,85,86]. Caveolae are lipid reservoirs containing high levels of CHOL and PS. It was found that abolishing caveolae, via knocking out CAV1 or flattening of the PM, elevated the clustering of PS in the PM and, in turn, further promoted the nanoclustering of KRAS4B [85]. On the other hand, knocking out CAV1 depleted CHOL from the cells [87], and acute depletion of CHOL in the PM disrupted the nanoclustering of the inactive HRAS.GDP [85]. More interestingly, acute depletion of CHOL, similar to knocking out CAV1, effectively induced coalescence of the inactive HRAS.GDP and the active HRAS.GTP on the PM, which, in turn, compromised HRAS signalling [85]. Thus, the PM nanoclusters of RAS isoforms selectively enrich lipids with distinct preferences for CHOL-dependent or CHOL-independent domains, allowing these signalling platforms to respond to perturbation of CHOL and membrane structures in distinct manners.

Sensing surface electrostatics

Various cell functions depend on the transmembrane potential on the cell surface [88]. For instance, long-term potentiation in the process of memory involves tight correlation between transmembrane potential on the PM and intracellular MAPK signalling [89,90]. In cancer cells, proliferation has long been observed to depend on the surface voltages, with many depolarizing potassium channels participating in the stimulation of tumour cell survival and proliferation [91-94]. Indeed, targeting these depolarizing potassium channels has been suggested as a strategy to treat cancer [91,93,95-98]. Despite the well-observed connection between cell surface voltages and mitogenic signalling in normal and pathologic cell activities, how cells convert the changing transmembrane potential to intracellular signalling is not very clear [94].

As described above, applying electric potential along the surface, or across the norm, of lipid bilayers induced phase separation of lipids, especially the charged lipids [45-47]. EM-spatial analysis and FLIM-FRET consistently demonstrated that depolarization of the PM, either by increasing extracellular potassium concentration or using glutamate treatment to activate glutamate-receptor-mediated PM depolarization, promoted the clustering of PS in the PM and, in turn, enhanced the nanoclustering and signalling of KRAS4B [24]. The ability of PS to mediate responses of KRAS4B to PM depolarization was also confirmed in intact fly embryos, where expression of mutant PS flippase ATP81 abolished the ability of the KRAS4B-regulated MAPK signalling to respond to PM depolarization [24].

The mediating roles of PS was further validated in later EM-spatial analysis. PM depolarization had no effect on the nanoclustering of the geranylgeranylated RAC1, which did not interact with PS [26]. When RAC1 was mutated to be farnesylated (RAC1-C15) and was enriched with PS, PM depolarization effectively enhanced the nanoclustering of RAC1-C15 [26]. Thus, PS lipids mediate the responses of KRAS4B nanoclustering and signalling to changing transmembrane potential. This electro-to-mitogen conversion allows cells to coordinate their intracellular MAPK signalling with surface voltages, and provides a molecular mechanism for long-term potentiation and memory establishment in the central nervous system, as well as for how mitogenic signalling communicates with potassium channels in tumour cells.

Membrane curvature sensing

Cell surface curvature defines cell morphology. Biofabrication established that cell growth, division, proliferation and migration were promoted in flatter and rounder epithelial-shaped cells [99,100]. On the other hand, these cell activities were compromised in the same cells confined to a more elongated morphology [99,100]. Tumour cells expressing oncogenic mutant KRAS4B.G12D adopted a more epithelial-like morphology, whereas inhibition of KRAS4B signalling (via co-expressing a dominant-negative mutant of effector BRAF-D594A) converted the cells to the elongated fibroblast morphology [101]. Expressing active HRAS resulted in an elongated morphology of tumour cells, which converted to a round morphology when expressing a dominant-negative mutant HRAS-T17N [102]. A recent study used RNA interference (RNAi) technology to screen a large cohort of pancreatic cancer cell lines further identified mutant KRAS4B as a determining factor for the epithelial morphology of cancer cells [103]. These data strongly suggest that RAS isoforms possess distinct membrane curvature sensing capabilities. Indeed, MD simulations first predicted that the lipid anchor of HRAS preferentially intercalated at the domain boundaries between L_o and L_d domains, and induced bilayer bending [71]. In vitro vesicle-binding assays further showed that the mono-farnesylated NRAS anchor favoured intercalatation into the L_o domains on smaller vesicles (<50 nm in diameters) with higher curvatures, but preferred the L_d domains on larger vesicles (>200 nm) with flat vesicles [104,105]. These in silico and in vitro data suggest that interactions between RAS and lipids efficiently mediate the ability of RAS isoforms to detect different bilayer curvatures.

The preferences of the palmitoylated RAS isoforms, such as HRAS and NRAS, for the curved biomembranes in cells were further shown in a series of EM-spatial analyses, bio-fabrication of nanobars, confocal imaging and isolated GPMVs [42,106,107]. Initial evidence for membrane curvature sensing of KRAS4B showed that, on intact cell PM sheets and live cells, the nanoclustering and PM binding of KRAS4B favoured flatter membranes with a lower curvature [26,42], which was confirmed in experiments where positive PM curvature, induced by transforming growth factor (TGFβ), mislocalized KRAS4B from the PM [107]. The preference of KRAS4B was confirmed in vitro, where purified KRAS4B favoured binding to larger vesicles with flatter bilayers in SPR [42]. Depletion of endogenous PS abolished the responses of KRAS4B to changing PM curvature, which was effectively restored by the acute addback of the mixed-chain 16:0/18:1 PS, but not the other PS species tested [26,42]. In an SPR-binding assay, binding of purified KRAS4B was elevated with increasing diameters of LUVs; 16:0/18:1 PC and 16:0/18:1 PS (80 : 20) [42]. Interestingly, the binding of KRAS4B was no longer dependent on the sizes of LUVs composed of di18:1 PC and di18:1 PS (80 : 20) [42]. These in vitro binding assays were consistent with the EM-spatial analysis in cells, and illustrated that the membrane curvature sensing of KRAS4B is mediated selectively by the mixed-chain PS species. Indeed, although the original farnesylated hexa-lysine (6K-C15) polybasic domain of KRAS4B, farnesylated hexa-arginine (6R-C15), geranylgeranylated hexa-lysine (6K-C15) and geranylgeranylated hexa-arginine (6R-C20) were equally charged, they sorted different PS species (Fig. 4) [25,26]. As summarized in Table 2, EM-bivariate co-clustering analysis showed that the PM nanoclusters of the original 6K-C15 and 6R-C20 both enriched the mixed-chain 16:0/18:1 PS [26]. Nanoclusters of 6R-C15 associated with the saturated di18:0 PS, whereas 6K-C20 sorted both di18:0 PS and di18:1 PS [26]. Using the BAR domain of amphiphysin II (BAR_amph2; inducing positive curvature) and planar intestinal- and kidney-specific BAR (pinkBAR; inducing flat membranes) to manipulate PM curvature, EM-spatial analysis showed that those KRAS4B polybasic domain constructs sorting the mixed-chain PS (6K-C15 and 6R-C20) favoured the flatter PM with low curvature [26]. On the other hand, the constructs associating with the symmetric PS species (6R-C15 and 6K-C20) preferred a more curved PM, the opposite of the other two [26]. These findings were concordant with earlier studies that found that increasing membrane curvature disrupted the PM distribution of the mixed-chain PS, while enhancing the PM interactions of the symmetric PS species [42]. Taking together, select sorting of lipid headgroups and acyl chains mediate the membrane curvature sensing of KRAS4B (Fig. 5).

Table 2.

Equivalently charged KRAS4B polybasic domain constructs favour different domains in the PM. These constructs favourably associate with different PS species with distinct acyl chain structures, which contribute to their preferences for different domains in the PM.

Polybasic domain mutants	PS species	CHOL domains
6K + C15	Mixed-chain PS	CHOL-independent
6R + C15	Saturated PS	CHOL dependent
6K + C20	Saturated PS Mono-unsaturated PS	CHOL dependent (possible domain boundary)
6R + C20	Mixed-chain PS	CHOL independent

Fig. 5.

KRAS4B polybasic-domain constructs favour different membrane curvatures. Despite packing identical charges, the spatial distribution of four KRAS4B polybasic-domain constructs favour different membrane curvatures, via sorting distinct PS species. CHOL, cholesterol.

RAS nanoclusters act as scaffolds to communicate with the extracellular environment

A key characteristic of solid tumours is the stiffening of the extracellular matrix, which increases cell tension and flattens the cancer cell morphology. It is well established that stiffening of the extracellular matrix correlates with elevated cell growth, survival and proliferation [108]. In addition to lipids, RAS nanoclusters on the cell surface incorporate extracellular matrix components, such as galectins, thus performing as scaffolds in communication of the cell with extracellular stimuli. In particular, EM-bivariate analysis and FLIM-FRET showed that GFP-HRAS-G12V co-clustered extensively with RFP–galectin-1 (Gal-1) on the PM of mammalian cells [109]. Increasing expression of Gal-1 in cells further promoted nanoclustering and facilitated epidermal growth factor (EGF)-stimulated signalling of HRASG12V in mammalian cells [109]. Mechanistically, it has been revealed that Gal-1 dimers formed high-affinity complexes with the RAS-binding domains of RAS effectors, thus indirectly forming complexes with HRAS [110]. Concordantly, rapamycin and its analogues inhibited FK506-binding protein 12 (FKBP12), which upregulated Gal-1 expression, enhancing both the nanoclustering of HRASG12V and the HRAS-dependent stemness of cancer cells [111]. Galectin-3 (Gal-3), on the other hand, was shown to colocalize with KRAS-G12V on the PM in EM and FLIM-FRET analyses, and promoted the nanoclustering and signalling of KRAS-G12V [112]. Although the previous studies did not distinguish the intracellular galectins and the secreted galectins, it is possible that the observed effects of galectins may be caused by the secreted galectins as components of the extracellular matrix. Although RAS anchors to the PM inner leaflet, communication between extracellular constituents and RAS on the PM inner leaflet has been demonstrated before. Specifically, antibody-induced crosslinking of glycophosphatidylinositol (GPI)-anchored protein on the PM outer leaflet effectively elevated the nanoclustering of GFP-tH (the truncated minimal anchor of HRAS) on the PM inner leaflet [22]. This remote communication between leaflets of the PM, or interleaflet coupling, has also been observed in synthetic model bilayers, where the diffusion of sphingomyelin in the outer leaflet coupled with the diffusion of PC lipids in the inner leaflet of giant vesicles [113]. Taken together, RAS nanoclusters may act as platforms that translate extracellular stimuli to intracellular mitogenic signalling.

Other lipid-anchored small GTPases undergo similar lipid-sorting mechanisms

The lipid-sorting capabilities of RAS isoforms largely originate from the distinct interactions between their lipid anchors and various lipids in the membranes. As such, other lipid-anchored small GTPases may possess similar select lipid-sorting abilities. Indeed, single-particle tracking photoactivated localization microscopy (SPT-PALM) effectively captured nanoclusters of RAC1, which contains a geranylgeranylated and a palmitoylated polybasic domain, on the PM of live cells [114]. The nanoclustering of RAC1 was more extensive at the front of migrating cells [114]. Further colocalization assays showed that, at the front of moving cells, RAC1 co-clustered efficiently with PIP₃ [114]. Later EM-spatial analysis revealed that RAC1 nanoclusters were selectively enriched in PIP₃ and PA [81]. The select sorting of PIP₃ and PA contributes to the ability of the constitutively active RAC1G12V to promote macropinocytosis [81]. Comparing a large cohort of single-point mutations of the polybasic domain of RAC1G12V further showed that each lysine or arginine residue plays a distinct role in sorting different lipids in the PM. In particular, an equivalently charged mutant RAC1-G12V-R185K showed significant depletion of PA in its nanoclusters and less overall nanoclustering on the PM [81]. Intriguingly, when the geranylgeranyl chain was changed to a farnesyl chain in a mutant RAC1-G12V (with its C-terminal tetrapeptide sequence composed of a cysteine, two aliphatic amino acids and an additional amino acid, or a CAAX motif, mutated to CVLS), the farneylated RAC1-G12V became significantly more enriched with PS, while being depleted of PIP₃ and PA [81]. Furthermore, whereas the originally geranylgeranylated RAC1-G12V was insensitive to PM depolarization, the nanoclustering of the farnesylated RAC1-G12V was elevated by PM depolarization [26]. Taken together, select lipid sorting may be a general mechanism for the lipid-anchored small GTPases to form nanoclusters on membranes, which may contribute to their abilities to localize to distinct intracellular organelles.

Conclusion

RAS proteins, especially the isoform KRAS4B, are major pharmacological targets for cancer. The majority of research effort has been focused on inhibiting the enzymatic G-domains of RAS; however, targeting the G-domains alone may not be enough to inhibit the multifaceted oncogenesis of this oncoprotein. Another important aspect of RAS signalling/pathology is its spatiotemporal organization on the PM. In this review, we described the latest findings on how select lipid sorting allows RAS nanoclusters to act as transducers to convert external stimuli into intracellular signalling. Thus, altering RAS–lipid specificities may be an alternative strategy for inhibiting RAS oncogenesis. Other small GTPases with distinct lipid anchors may possess similar select lipid-sorting mechanisms that may contribute to their spatial distribution and function. Furthermore, our understanding of biomembranes has matured from a homogeneous mixture, which uniformly respond to perturbations, to a heterogeneous environment containing co-existing L_o/L_d domains. While PC lipids and CHOL have been primarily focused for the raft/non-raft phase separation in membranes, the latest findings suggest that spatially non-overlapping raft domains coexist in the PM. Similar spatially segregated non-raft domains also co-exist in the PM. Anionic phospholipids, such as PS, contribute to the further fine-tuning of these domains. These new findings pinpoint the important biological roles of anionic lipids, as well as lipid acyl chain structures, in cell function and pathology.

Abbreviations

16:0/18:1 PS

1-palmitoyl-2-oleoyl-sn-glycero-3-phospho-L-serine

Akt

Stock A Strain k AKR mouse thymoma

BAR

bin-amphiphysin/Rvs

C15

farnesyl

C20

geranylgeranyl

CAV1

caveolin 1

CHOL

cholesterol

Di18:0 PS

1,2-distearoyl-sn-glycero-3-phospho-L-serine

Di18:1 PS

1,2-dioleoyl-sn-glycero-3-phospho-L-serine

EGF

epidermal growth factor

EM

electron microscopy

ER

endoplasmic reticulum

ERK

extracellular signal-regulated kinases

FLIM-FRET

fluorescence lifetime imaging microscopy–fluorescence resonance energy transfer

Gal1

galectin 1

Gal3

galectin 3

GFP

green fluorescence protein

GPMV

giant plasma membrane vesicles

GUVs

giant unilamellar vesicles

HRAS

Harvey Rat sarcoma virus

KRAS

Kirsten Rat sarcoma virus

L_d

liquid-disordered

L_o

liquid-ordered

LUVs

large unilamellar vesicles

MAPK

mitogen-activated protein kinase

MD

molecular dynamic

MEK

mitogen-activated protein kinase

MβCD

methyl–cyclodextrin

NRAS

neuroblastoma Rat sarcoma virus

NSAIDs

nonsteroidal anti-inflammatory drugs

PA

phosphatidic acid

PC

phosphatidylcholine

PE

phosphatidylethanolamine

PH

pleckstrin homology

PI3K

Phosphoinositide 3-kinases

PIP₂

phosphoinositol 4,5-bisphosphate

PIP₃

phosphoinositol 3,4,5-trisphosphate

PM

plasma membrane

PS

phosphatidylserine

PX

phagocytic oxidase homology

RAF

Rapidly Accelerated Fibrosarcoma

RAS

Rat sarcoma virus

RFP

red fluorescence protein

RNAi

RNA interference

SPR

surface plasmon resonance

SPT-PALM

single-particle tracking photoactivated localization microscopy

TGFβ

transforming growth factor β