RAS G-domains allosterically contribute to the recognition of lipid headgroups and acyl chains

Membrane anchors of RAS play major roles in select lipid enrichment of RAS on the plasma membrane. Arora et al. show that the cytoplasmic RAS G-domains fine-tune selectivity of lipids. Their findings provide a mechanism for the allele-specific oncogenesis of KRAS4B.

Abstract

Mutant RAS are major contributors to cancer and signal primarily from nanoclusters on the plasma membrane (PM). Their C-terminal membrane anchors are main features of membrane association. However, the same RAS isoform bound to different guanine nucleotides spatially segregate. Different RAS nanoclusters all enrich a phospholipid, phosphatidylserine (PS). These findings suggest more complex membrane interactions. Our electron microscopy-spatial analysis shows that wild-types, G12V mutants, and membrane anchors of isoforms HRAS, KRAS4A, and KRAS4B prefer distinct PS species. Mechanistically, reorientation of KRAS4B G-domain exposes distinct residues, such as Arg 135 in orientation state 1 (OS1) and Arg 73/Arg 102 in OS2, to the PM and differentially facilitates the recognition of PS acyl chains. Allele-specific oncogenic mutations of KRAS4B also shift G-domain reorientation equilibrium. Indeed, KRAS4B^G12V, KRAS4B^G12D, KRAS4B^G12C, KRAS4B^G13D, and KRAS4B^Q61H associate with PM lipids with headgroup and acyl chain specificities. Distribution of these KRAS4B oncogenic mutants favors different nanoscale membrane topography. Thus, RAS G-domains allosterically facilitate membrane lateral distribution.

Introduction

Rat sarcoma virus (RAS) small GTPases, including isoforms Harvey RAS (HRAS), neuroblastoma RAS (NRAS), splice variants KRAS4A and KRAS4B, toggle between the active GTP-bound and inactive GDP-bound states and participate in cell growth, division, survival, proliferation, and migration (Cox et al., 2015; Nissley and McCormick, 2022; Prior et al., 2020). Mutant RAS at key residues, such as glycine 12 (Gly 12), Gly 13, and glutamine 61 (Gln 61), remain constitutively active and are main drivers of cancer (Cox et al., 2015; Nissley and McCormick, 2022; Prior et al., 2020). Signaling of wild-type and oncogenic mutants of RAS proteins is mostly compartmentalized to the spatially segregated nanodomains, termed nanoclusters, on the plasma membrane (PM) (Cox et al., 2015; Morstein et al., 2023; Ozdemir et al., 2022; Plowman et al., 2005; Prior et al., 2003; Zhou et al., 2015). Their isoform-specific C-terminal membrane-anchoring domains have been attributed to mainly facilitate association with lipids and the formation of nanoclusters on the PM (Cox et al., 2015; Hancock et al., 1991a, 1991b; Hancock et al., 1990; Koester et al., 2022; Lee et al., 2019; Ozdemir et al., 2022; Zhou and Hancock, 2018a, 2023). Interestingly, it has long been observed that, with the identical membrane anchor, the same RAS isoforms bound with GDP and GTP are spatially segregated on the PM (Lee et al., 2019; Plowman et al., 2005; Prior et al., 2003; Zhou et al., 2014). These findings strongly suggest that the largely cytoplasmic RAS G-domains contribute to the sensing of lipids. Since most oncogenic mutations of RAS occur in their G-domains, do RAS oncogenic mutants differ in their lipid preferences? Further, despite being spatially segregated, all RAS nanoclusters tested enrich the same type of anionic phospholipid, phosphatidylserine (PS) (Koester et al., 2022; Lee et al., 2019; Liang et al., 2019; Ozdemir et al., 2022; Zhou et al., 2014, 2015, 2017, 2021b). This finding implies that different PS pools co-exist among various RAS nanoclusters on the PM. Indeed, we previously showed that the nanoclusters of oncogenic mutant KRAS4B^G12V selectively enrich the mixed-chain PS species, but not other symmetric PS species (Liang et al., 2019; Zhou et al., 2017, 2021b). We further showed the efficient recruitment of effector CRAF by the constitutively active mutant KRAS4B^G12V only occurs in the presence of the mixed-chain PS species, but not other symmetric PS species tested (Zhou et al., 2017). Intact caveolae selectively enrich symmetric PS species, but not the mixed-chain PS (Zhou et al., 2021a). Parallel cholesterol-dependent and cholesterol-independent PS pools also coexist in the PM (Kay et al., 2012; Zhou et al., 2021b). Do RAS G-domains, which do not have direct access to the bilayer core, participate in the sensing of lipid acyl chains? Our current study aims to systematically examine whether and how RAS G-domains contribute to lipid binding in cells.

Via super-resolution electron microscopy (EM)-spatial analysis, we here tested how full-length wild-types, oncogenic G12V mutants, and the truncated minimal membrane-anchoring domains of HRAS, KRAS4A, and KRAS4B bind distinct PS species on the PM. Typical mammalian cells contain 30–40 PS species. We chose to compare three PS species: the fully saturated 1,2-distearoyl-sn-glycero-3-phospho-L-serine (DSPS), the mono-unsaturated 1,2-dioleoyl-sn-glycero-3-phospho-L-serine (DOPS), and the mixed-chain 1-stearoyl-2-oleoyl-sn-glycero-3-phospho-L-serine (SOPS). These species represent major species in their respective saturated, mono-unsaturated, and mixed-chain species in cells. Further, despite sharing similar chemical structures (Fig. S1), these PS species possess markedly different biophysical properties (listed in Table S1) and laterally segregate to distinct regions in the PM (Skotland and Sandvig, 2019; Zhou et al., 2021b). These PS species were used as probes for distinct regions of the PM with diverse diffusion and packing characteristics. We, here, show that wild-types, G12V mutants, and minimal anchors of RAS isoforms prefer to bind distinct sets of PS species. The G-domain of KRAS4B has been predicted to sample various orientation states (OS) on the PM, each of which presents distinct membrane-interacting interfaces. To explore mechanistically how RAS G-domains facilitate sensing of lipid acyl chains, we mutated key residues in the membrane-interacting interfaces of the G-domain of KRAS4B, including Arginine 73 (Arg 73), Arg 102, and Arg 135. We show that between wild-type KRAS4B and KRAS4B^G12V, R73Q, R102Q, and R135Q mutations differentially alter the preferences of PS species. Oncogenic mutations of KRAS4B also shift the reorientation equilibrium of its G-domain (Lu and Martí, 2020, 2022; Mazhab-Jafari et al., 2015; Pantsar, 2020; Pantsar et al., 2018; Vatansever et al., 2020). We found that oncogenic mutants G12V, G12C, G12D, G13D, and Q61H of KRAS4B possess distinct preferences for different PS species and other lipids with different headgroups, as well as distinct affinities for membrane curvature. Thus, RAS G-domains contribute to the recognition of lipid headgroups and acyl chains, and fine-tune the lateral segregation of RAS isoforms on the PM. KRAS4B oncogenic mutants bind distinct sets of lipids, which may contribute to their allele-specific pathological activities.

Figure S1.

Acyl chains of PS species contribute to distinct biophysical properties and diverse responses to external perturbations. We compared three PS species, including DSPS, DOPS, and SOPS. These PS species are major species in their respective categories of saturated, mono-unsaturated, and the mixed-chain species.

Results

RAS G-domains contribute to the recognition of PS acyl chains

To examine the potential roles of RAS G-domains in facilitation of lipid association, we first compared how full-length wild-types, oncogenic G12V mutants, and the minimal membrane anchors of isoforms HRAS, KRAS4A, and KRAS4B co-clustered with different PS species. To modulate PS contents in cells, PS auxotroph (PSA3) cells were grown in medium containing 10% dialyzed fetal bovine serum (DFBS) for 72 h to deplete endogenous PS (Kay et al., 2012; Lee et al., 2012; Liang et al., 2019; Zhou et al., 2021a; Zhou et al., 2014; Zhou et al., 2017; Zhou et al., 2021b; Zhou et al., 2015). To restore the normal endogenous PS levels, PSA3 cells were supplemented with 10 μM ethanolamine (Etn) for 72 h (Kay et al., 2012; Lee et al., 2012; Liang et al., 2019; Zhou et al., 2021a; Zhou et al., 2014; Zhou et al., 2017; Zhou et al., 2021b; Zhou et al., 2015). To compare effects of individual PS species, we acutely added back synthetic PS species to the PSA3 cells depleted of endogenous PS via 1 h incubation with synthetic PS species (Liang et al., 2019; Zhou et al., 2017, 2021b). EM-bivariate co-clustering analysis quantified the co-localization between GFP-LactC2 (a PS-specific binding domain) and an RFP-tagged RAS construct on intact PM sheets of PSA3 cells (Liang et al., 2019; Zhou et al., 2021a; Zhou et al., 2014; Zhou et al., 2017; Zhou et al., 2021b; Zhou et al., 2015). Briefly, the apical PM sheets of PSA3 cells were attached to copper EM grids. Following fixation, GFP-LactC2 bound to PS lipids in the PM was labeled with 6-nm gold nanoparticles conjugated to anti-GFP antibody, while RFP-RAS anchored to the PM was labeled with 2-nm gold nanoparticles coupled to anti-RFP antibody. Both gold populations were imaged via transmission EM (TEM) at 100,000× magnification (Fig. S2, A–D). Ripley’s bivariate K-function analysis calculated the extent of co-clustering between 6- and 2-nm gold particles within a 1 μm² PM area. L_biv(r) – r represents the extent of co-clustering and was plotted against distance r in nanometers (Fig. S2 E). The bivariate L_biv(r) – r curves were integrated between the r values of 10 and 110 nm and termed L-function bivariate integrated (LBI). The LBI values above the 95% confidence interval (CI) of 100 indicate statistically significant co-clustering, with larger LBI values corresponding to more extensive co-clustering and higher enrichment.

Figure S2.

EM-bivariate co-clustering analysis quantifies the co-localization between PS and a RAS construct on the plasma membrane. (A and C) Intact PM sheets of PSA3 cells co-expressing GFP-LactC2 and an RFP-KRAS4B^G12V (A) or RFP-KRAS4B^G12VR73Q (C) were attached to copper EM grids. GFP and RFP anchored to the PM sheets were immunolabeled with 6-nm gold nanoparticles conjugated to anti-GFP antibody and 2-nm gold coupled to anti-RFP antibody, respectively. (B and D) Digitized images of A and C are shown in B and D, respectively (black dots indicate 2-nm gold and red dots indicate the 6-nm gold). Ripley’s bivariate K-function calculated the co-clustering between the two populations of gold particles within a PM area of 1 μm². (E) The extent of co-clustering, L_biv(r) – r, was plotted as a function of length scale, r in nanometer (nm). The L_biv(r) – r curves for A and C are shown. The dashed line indicates the 95% CI. (F) GFP-tagged KRAS4B mutants anchored to the intact PM sheets were labeled with 4.5-nm gold nanoparticles conjugated to anti-GFP-antibody. Following TEM imaging, the univariate distribution of gold particles within a PM area of 1 μm² was calculated using Ripley’s univariate K-function. The extent of nanoclustering, L(r) – r, was plotted against r. The peak L(r) – r values serve as a statistical summary for the extent of nanoclustering. The dashed line indicates the 99% CI.

Fig. 1 A shows that the LBI value between GFP-LactC2 and RFP-KRAS4B^G12V was ∼178 in PSA3 cells supplemented with Etn, illustrating extensive co-clustering between KRAS4B^G12V and PS in the PM of PSA3 cells containing the normal level of endogenous PS. Depletion of endogenous PS (DFBS) significantly decreased the LBI value below 95% CI, suggesting efficient spatial segregation of KRAS4B^G12V away from the remaining endogenous PS after PS depletion. In the PS-depleted PSA3 cells, acute addback of the mixed-chain SOPS, but not other PS species tested, effectively restored the co-clustering between GFP-LactC2 and RFP-KRAS4B^G12V. This set of data was entirely consistent with our previous findings that the PM nanoclusters of KRAS4B^G12V selectively enrich the mixed-chain PS species (Liang et al., 2019; Zhou et al., 2017, 2021b). Similar to RFP-KRAS4B^G12V, the wild-type RFP-KRAS4B and its minimal anchor RFP-tK(4B) also co-clustered extensively with the endogenous PS (Etn, Fig. 1, B and C). Depletion of endogenous PS (DFBS) segregated GFP-LactC2 from RFP-KRAS4B or RFP-tK(4B). Interestingly, while both RFP-KRAS4B and RFP-tK(4B) still preferred SOPS the most, they co-localized with other PS species, such as DOPS for RFP-KRAS4B and DSPS for RFP-tK(4B) (Fig. 1, B and C). Thus, while all three KRAS4B constructs preferentially associate with the mixed-chain PS, KRAS4B^G12V possesses higher specificity.

Figure 1.

Wild-types, G12V mutants, and minimal anchors of RAS isoforms bind distinct PS species. To manipulate PS profiles in cells, we used a PS auxotroph PSA3 cell line. Endogenous PS was depleted by growing the PSA3 cells in 10% DFBS for 72 h. The normal endogenous PS level was restored via supplementation of 10 μM Etn for 72 h. To compare individual PS species, exogenous PS species were acutely added back to the PS-depleted PSA3 cells. Specifically, the PS-depleted PSA3 cells grown in DFBS were incubated in the medium containing 10 μM synthetic PS species for 1 h before harvesting. These PS species included the fully saturated DSPS, the mono-unsaturated DOPS, and the mixed-chain SOPS. (A–I) PSA3 cells ectopically co-expressed GFP-LactC2 (a PS probe) and an RFP-tagged RAS construct: KRAS4B constructs (A–C), KRAS4A (D–F), and HRAS constructs (G–I). For each isoform, the G12V oncogenic mutant (G12V), wild-type (WT), and the truncated C-terminal membrane-anchoring domain in the absence of G-domain (tK or tH) were compared. Intact apical PM sheets of PSA3 cells were attached to copper EM grids. GFP and RFP anchored to the PM inner leaflet were immunolabeled with 6-nm gold nanoparticles conjugated to anti-GFP antibody and 2-nm gold coupled to anti-RFP antibody, respectively. EM-bivariate co-clustering analysis quantified the co-clustering between the 6- and 2-nm gold populations within a PM area of 1 μm², with LBI values indicating the extent of co-clustering or enrichment. The LBI value of 100 is the 95% CI (the dashed lines), the values above which indicate statistically significant co-clustering. (J) A heat map summarizes the LBI values in all panels. For each condition, between 6 and 34 apical PM sheets from individual cells were imaged, analyzed, and pooled together. Data are shown as mean ± SEM. Statistical significance was evaluated using non-parametric bootstrap tests, with * indicating P < 0.05.

The splice variants KRAS4A and KRAS4B differ significantly in their C-terminal membrane-anchoring domains, with KRAS4A possessing an additional palmitoyl chain without the polybasic domain. We previously showed that KRAS4A^G12V also co-clustered with the endogenous PS (Maxwell et al., 2018). Consistently, Fig. 1, D–F, shows that KRAS4A^G12V, wild-type KRAS4A, and the minimal anchor tK(4A) all enriched endogenous PS extensively (Etn). PS depletion (DFBS) effectively separated the PS probe from KRAS4A constructs (Fig. 1, D–F). For the PS acyl chain selectivity, RFP-KRAS4A^G12V co-clustered with all three PS species tested (Fig. 1 D). RFP-KRAS4A excluded SOPS (Fig. 1 E). RFP-tK(4A) co-clustered with all three PS species, but most extensively co-clustering with DSPS (Fig. 1 F). Taken together, KRAS4A constructs recognize PS acyl chains in distinct manners.

We next compared the PS acyl chain selectivity of HRAS. Consistent with previous findings (Zhou et al., 2014, 2015), HRAS^G12V, wild-type HRAS, and the minimal anchor tH all enriched with the endogenous PS (Etn) and dissociated from PS upon PS depletion (DFBS) (Fig. 1, G–I). When comparing individual PS species, RFP-HRAS^G12V co-clustered equivalently with all three PS species tested (Fig. 1 G). Interestingly, LBI values between GFP-LactC2 and RFP-HRAS were near 95% CI for all three PS species tested, suggesting that the wild-type HRAS does not significantly colocalize with the three PS species tested. The minimal anchor RFP-tH more preferentially associated with DSPS and DOPS, but not SOPS, consistent with molecular dynamic (MD) simulations that tH preferentially localizes to the boundaries between the liquid-ordered domains (enriching saturated lipids) and the liquid-disordered domains (containing unsaturated lipids) (Janosi et al., 2012). Taken together and summarized in a heatmap of LBI values (Fig. 1 J), RAS isoforms selectively bind to different PS species in G-domain– and mutation-specific manners, strongly suggesting that G-domains participate in the selectivity of lipid acyl chains.

KRAS4B G-domain residues contribute to the recognition of PS acyl chains

To examine mechanistically how these largely cytosolic RAS G-domains might recognize lipid acyl chains, we focused on KRAS4B since the structures and conformational orientation of its G-domain on membranes have been extensively studied. As illustrated in a schematic inspired by Prakash et al. (2019) in Fig. 2, KRAS4B G-domain mainly switches between two distinct orientation states, OS1 and OS2, as well as an intermediate OS0 state (Prakash et al., 2016, 2019). KRAS4B^G12V has been predicted to disfavor the OS2 state (Prakash et al., 2019). Key G-domain residues, such as Arg 135 in OS1 and Arg 73/Arg 102 in OS2, form close contacts with lipid bilayers with high probability (Neale and García, 2020; Prakash et al., 2019; Prakash et al., 2016). In consequence, the C-terminal membrane anchor of KRAS4B alters conformations in coordination with the reorientation of its G-domain (Prakash et al., 2016, 2019) (schematic in Fig. 2). Intercalation of KRAS4B membrane anchor into membranes may coordinate with the reorientation of its G-domain, thus altering the binding of PS acyl chains. To examine how G-domain reorientation participated in the recognition of PS acyl chains, we mutated Arg 73, Arg 102, or Arg 135 of wild-type KRAS4B and KRAS4B^G12V and generated a cohort of OS mutants: RFP-KRAS4B^R73Q, RFP-KRAS4B^R102Q, RFP-KRAS4B^R135Q, RFP-KRAS4B^G12V.R73Q, RFP-KRAS4B^G12V.R102Q, and RFP-KRAS4B^G12V.R135Q. We performed EM-bivariate co-clustering analysis following acute addback of various PS species to PSA3 cells depleted of endogenous PS. Unlike the original G-domain of RFP-KRAS4B^G12V (Fig. 2 A), RFP-KRAS4B^G12V.R73Q no longer co-clustered with the endogenous PS (Etn) (Fig. 2 B). Interestingly, RFP-KRAS4B^G12V.R73Q co-clustered extensively with fully saturated DSPS (Fig. 2 B). RFP-KRAS4B^G12V.R102Q behaved similarly to the original RFP-KRAS4B^G12V and preferentially bound with SOPS (Fig. 2 C). RFP-KRAS4B^G12V.R135Q preferred DOPS and then DSPS, but not SOPS (Fig. 2 D).

Figure 2.

Key KRAS4B G-domain residues at the membrane interface contribute to the recognition of PS acyl chains. (A–H) Key membrane-associating residues, such as Arg 73, Arg 102, and Arg 135, were mutated in the oncogenic mutant KRAS4B^G12V (A–D, KG12V) and the wild-type KRAS4B (E–H, KWT) to generate a cohort of OS mutants. PSA3 cells co-expressing GFP-LactC2 and an RFP-tagged OS mutant were supplemented with Etn (normal endogenous PS), DFBS (PS depletion), and/or individual exogenous PS species (DSPS = di18:0 PS, DOSP = di18:1 PS, SOPS = 18:0/18:1 PS) as described in Fig. 1. EM-bivariate analysis quantified the co-clustering of PS with OS mutants of KRAS4B. Dashed lines indicate the 95% CI. (I) All LBI values shown in the previous panels were combined in a heat map to illustrate enrichment or depletion of different PS species. For each condition, between 12 and 15 PM sheets from individual cells were imaged, analyzed, and pooled. LBI values are shown as mean ± SEM. The non-parametric bootstrap tests evaluated the statistical significance between data sets, with * indicating P < 0.05.

The OS mutants of wild-type RFP-KRAS4B displayed different patterns of PS acyl chain selectivity. RFP-KRAS4B^R73Q and RFP-KRAS4B^R102Q associated with all three PS species tested, suggesting that mutating Arg 73 and Arg 102 in OS2 of the wild-type KRAS4B lost selectivity for acyl chains of PS, while still maintaining high affinities for the PS headgroup (Fig. 2, E–G). For RFP-KRAS4B^R135Q, acute addback of any of the three PS species did not effectively restore its co-clustering with GFP-LactC2 when compared with the PS-depleted condition (DFBS) (Fig. 2 H). This suggests that KRAS4B^R135Q does not associate with any of the PS species tested. The significant differences between the OS mutants of KRAS4B^G12V and those of wild-type KRAS4B strongly suggest that reorientation of KRAS4B G-domain contributes to the selectivity of PS acyl chains (summarized in a heatmap of LBI values in Fig. 2 I).

KRAS4B oncogenic mutants prefer distinct PS species

KRAS4B oncogenic mutations mostly occur at residues Gly 12, Gly 13, and Gln 61, such as G12D, G12V, G12C, G13D, and Q61H (Cox et al., 2015; Nissley and McCormick, 2022; Prior et al., 2020). While these sites are not proximal to membranes, their mutations have been predicted to alter the structures and conformational dynamics of their G-domains (Lu and Martí, 2020, 2022; Mazhab-Jafari et al., 2015; Pantsar, 2020; Pantsar et al., 2018; Vatansever et al., 2020). Thus, it is possible that these oncogenic mutants may prefer different PS species. To test this, we performed a similar EM-bivariate co-clustering analysis and compared co-clustering between GFP-LactC2 and an RFP-tagged KRAS4B oncogenic mutant in PSA3 cells. Shown in Fig. 3, B and D, KRAS4B^G12D and KRAS4B^Q61H behaved similarly with KRAS4B^G12V (Figs. 1 and 2) and preferentially enriched with SOPS but not other PS species tested. Interestingly, KRAS4B^G12C associated with all three PS species tested (Fig. 3 A), while KRAS4B^G13D more preferentially associated with the DSPS and DOPS, but not SOPS (Fig. 3 C). Taken together, oncogenic mutants of KRAS4B possess distinct capabilities to recognize different PS species (summarized in a heatmap of LBI values in Fig. 3 E).

Figure 3.

Oncogenic mutants of KRAS4B selectively enrich distinct profiles of PS species. (A–D) PS contents in the PSA3 cells co-expressing GFP-LactC2 and an RFP-tagged oncogenic mutant of KRAS4B, such as KRAS4B^G12C (A), KRAS4B^G12D (B), KRAS4B^G13D (C), and KRAS4B^Q61H (D), were modulated as described above. Etn = normal endogenous PS level, DFBS = endogenous PS depletion, DSPS = di18:0 PS, DOSP = di18:1 PS, SOPS = 18:0/18:1 PS. EM-bivariate co-clustering analysis calculated the co-clustering LBI values and quantified the enrichment of PS species within the nanoclusters of KRAS4B oncogenic mutants. The LBI value of 100 is the 95% CI (dashed lines). (E) All LBI values were combined in a heat map to illustrate enrichment and depletion of different PS species. (F) A schematic summarizes the distinct locations or local lipid environments that various KRAS4B oncogenic mutants favorably distribute to. For each condition, between 14 and 15 apical PM sheets were imaged, analyzed, and pooled. LBI values are shown as mean ± SEM. The non-parametric bootstrap tests evaluated the statistical significance between data sets, with * indicating P < 0.05.

KRAS4B oncogenic mutants bind to different lipid types with distinct headgroups

The distinct PS acyl chain preferences of KRAS4B oncogenic mutants suggest that these oncogenic mutants may distribute to distinct regions of the PM enriched with different sets of lipids. We next examined how various oncogenic mutants of KRAS4B associated with various lipid types. Using the EM-bivariate co-clustering analysis, we quantified the co-localization between the GFP-tagged lipid binding domains and the RFP-tagged KRAS4B mutants ectopically expressed in baby hamster kidney (BHK) cells. The lipid-binding domains included GFP-LactC2 (for PS, Fig. 4 A), GFP-PH-PLCδ (for phosphoinositol 4,5-bisphosphate [PIP₂], Fig. 4 B), GFP-PH-Akt (for phosphoinositol 3,4,5-triphosphate [PIP₃], Fig. 4 C), GFP-PASS (for phosphatidic acid [PA], Fig. 4 D), and GFP-D4H (for cholesterol, Fig. 4 E). KRAS4B oncogenic mutants included RFP-KRAS4B^G12C, RFP-KRAS4B^G12D, RFP-KRAS4B^G12V, RFP-KRAS4B^G13D, RFP-KRAS4B^Q61H. Interestingly, although all KRAS4B mutants co-clustered with endogenous PS on the PM, the PS enrichment of KRAS4B^G13D was markedly lower than other mutants (Fig. 4 A). While RFP-KRAS4B^G12V, RFP-KRAS4B^G12D, RFP-KRAS4B^G13D, and RFP-KRAS4B^Q61H segregated from PIP₂, RFP-KRAS4B^G12C significantly enriched PIP₂ (Fig. 4 B). RFP-KRAS4B^G12C, RFP-KRAS4B^G13D, and RFP-KRAS4B^Q61H also co-clustered with PIP₃ more extensively than RFP-KRAS4B^G12D and RFP-KRAS4B^G12V (Fig. 4 C). RFP-KRAS4B^G12C and RFP-KRAS4B^G13D extensively co-clustered with cholesterol, while other mutants did not (Fig. 4 E). All KRAS4B mutants associated with PA with similar extent (Fig. 4 D). A heat map of LBI values, summarized in Fig. 4 F, illustrates the diverse lipid profiles of different KRAS4B oncogenic mutants. Taken together, oncogenic mutants of KRAS4B display allele-specific recognition of PM lipids.

Figure 4.

KRAS4B oncogenic mutants display allele-specific lipid-binding capability. (A–E) EM-bivariate analysis quantified the extent of co-clustering between a GFP-tagged lipid-binding domain and an RFP-tagged oncogenic mutant of KRAS4B co-expressed in BHK cells. Lipids, including PS (A), PIP₂ (B), PIP₃ (C), PA (D) and cholesterol (E), were tested. KRAS4B mutants included KRAS4B^G12C, KRAS4B^G12D, KRAS4B^G12V, KRAS4B^G13D, and KRAS4B^Q61H. GFP and RFP anchored to the PM sheets were immunolabeled with 6-nm gold nanoparticles conjugated to anti-GFP antibody and 2-nm gold coupled to anti-RFP antibody, respectively. The co-clustering LBI values are shown as mean ± SEM and used as summary statistics for the co-clustering. Dashed lines indicate the 95% CI. (F) A heat map of LBI values between KRAS4B mutants and various lipids is shown. To quantify the cholesterol dependence of KRAS4B oncogenic mutants, BHK cells ectopically expressing a GFP-tagged KRAS4B oncogenic mutant were treated without/with 2% MβCD for 30 min. GFP anchored to the PM was labeled with 4.5-nm gold coupled to anti-GFP antibody. Following EM imaging, univariate K-function analysis calculated the extent of nanoclustering of gold particles within a PM area of 1 μm². L_max is a statistical summary for the extent of nanoclustering. (G and H) The nanoclustering (G) and PM localization (H) of GFP-tagged KRAS4B oncogenic mutants (G12C, G12D, and G13D) are shown as mean ± SEM, with the dashed line indicating the 99% CI. (I) EM-bivariate co-clustering analysis calculated the colocalization between a GFP-tagged KRAS4B mutant and RFP-CRAF in BHK cells. LBI values are shown as mean ± SEM. For each bivariate and univariate condition, between 10 and 22 apical PM sheets were imaged, analyzed and pooled. The non-parametric bootstrap tests evaluated the statistical significance between data sets, with * indicating P < 0.05.

To further validate our finding that KRAS4B^G12C and KRAS4B^G13D associate with cholesterol, we performed EM-univariate nanoclustering analysis to measure how cholesterol depletion impacted the extent of nanoclustering of KRAS4B mutants. Briefly, BHK cells ectopically expressing a GFP-tagged KRAS4B mutant were treated without/with 2% methyl β-cyclodextrin (MβCD) for 30 min before the apical PM was attached to copper EM grids. GFP-KRAS4B mutants anchored to the PM were labeled with 4.5-nm gold nanoparticles conjugated to anti-GFP antibody. The univariate K-function analysis calculated the extent of nanoclustering of the gold particles within a 1 μm² PM area and the peak L(r) – r value (L_max) was used as a statistical summary for nanoclustering (Fig. S2 F). Cholesterol depletion effectively disrupted the nanoclustering (Fig. 4 G) and PM localization (Fig. 4 H) of GFP-KRAS4B^G12C, without affecting those of GFP-KRAS4B^G12D. Interestingly, cholesterol depletion mislocalized GFP-KRAS4B^G13D from the PM (Fig. 4 H) without affecting its nanoclustering (Fig. 4 G). Taken together, oncogenic mutations G12C and G13D shift KRAS4B to cholesterol-dependent nanoclusters on the PM.

Lipids enriched in KRAS4B nanoclusters contribute to efficient effector recruitment since most of its effectors possess their own lipid-binding motifs. Recruitment and activation of effector CRAF favor cholesterol-poor domains enriched with the mixed-chain PS (Inder et al., 2008; Zhou et al., 2017). Figs. 3 and 4 suggest that KRAS4B oncogenic mutants may recruit effectors with different efficacies. We performed EM-bivariate co-clustering experiments using BHK cells co-expressing a GFP-tagged KRAS4B oncogenic mutant and RFP-CRAF. Fig. 4 I shows that GFP-KRAS4B^G12C and GFP-KRAS4B^G13D associated with RFP-CRAF less efficiently than GFP-KRAS4B^G12D, GFP-KRAS4B^G12V, and GFP-KRAS4B^Q61H. Taken together, select lipid association of KRAS4B oncogenic mutants contributes to their distinct effector recruitment.

Allele-specific KRAS4B oncogenic mutants exhibit distinct sensitivity to nanoscale membrane topography

Lipid preferences of RAS correlate with their distinct responses to changing membrane properties (Cox et al., 2015; Inder et al., 2008; Zhou et al., 2015). HRAS mediates the adoption of fibroblast morphology of prostate cancer cells, while expression of mutant KRAS4B induces epithelial morphology of tumor cells (Ghosh et al., 1999; Sanchez-Laorden et al., 2014). A recent siRNA screen of a large cohort of human cancer cells further identifies KRAS4B oncogenic mutants as a major determining factor for the epithelial morphology of cancer cells (Yuan et al., 2018). These data suggest a potential correlation between RAS and membrane curvature. We and others recently showed that HRAS and KRAS4B prefer different membrane curvatures, mediated by their distinct preferences for lipid headgroups and acyl chains (Damalas et al., 2022; Liang et al., 2019; Mu et al., 2022). Figs. 3 and 4 imply that KRAS4B oncogenic mutants may respond to changing membrane curvature in different manners. To test this, we quantified how KRAS4B oncogenic mutants responded to changing membrane topography defined by local curvatures, an important mechanosensing property that contributes to intracellular transport, cell morphology, macropinocytosis, migration, and invasion. To generate membrane curvature of cells, vertically aligned nanobars were fabricated in array form on flat substrates (Fig. 5 A) using electron beam lithography (EBL) (Li et al., 2019; Liang et al., 2019; Mu et al., 2022). The pitch size between the adjacent nanobars was 5 μm. Because of the high biocompatibility of these nanobars, the cells cultured on it can wrap their PM around the nanobars and, thus, the cell membrane follows the nanobar curvature closely. As illustrated in Fig. 5 B, the nanobars were designed 2 μm long and 250 nm wide, creating two positively curved surfaces with 250-nm diameter at the bar ends and a flat region with zero curvature at the bar center. Therefore, when wrapped by cell PM, it allows for the local comparison of the protein behaviors upon the membrane curvature alteration (Fig. 5 C) (Li et al., 2019; Liang et al., 2019; Mu et al., 2022). GFP-KRAS4B^G12C (KG12C), GFP-KRAS4B^G12D (KG12D), GFP-KRAS4B^G12V (KG12V), GFP-KRAS4B^G13D (KG13D), or GFP-KRAS4B^Q61H (KQ61H) was expressed in human bone osteosarcoma epithelial U2OS cells. The distribution patterns of KG12C, KG12D, KG12V, KG13D, and KQ61H were visualized via confocal microscopy (Fig. 5 D). Fluorescence intensity ratio of bar end to bar center is termed the enrichment index (EI) and represents the relative distribution of proteins at the curved membrane to the flat membrane. Larger EI values indicate higher membrane curvature preferences. The EI values for five KRAS4B oncogenic mutants differ significantly, in an order of KQ61H > KG12C > KG12D > KG12V > KG13D (Fig. 5 E). Shifts of the EI distribution curves of the five mutants were also in the same order (Fig. 5 F). To correlate the curvature sensing of KRAS4B oncogenic mutants with their lipid preferences, we next compared the EI values of the specific lipid-binding domains of PS, PA, PIP₂, PIP₃, and cholesterol. In Fig. S3, the EI values of PIP₃, PA, and cholesterol were higher than those of PS and PIP₂. Enrichment of PIP₃, PA, and/or cholesterol may contribute to the higher curvature preferences of KRAS4B^Q61H and KRAS4B^G12C. However, association of other lipids and additional PM components, such as actin and extracellular matrix, may contribute to the distinct curvature sensing of KRAS4B oncogenic mutants. Taken together, the distinct responses of KRAS4B oncogenic mutants to nanobar-curved membrane sites highlight the interconnection between G-domain–mediated membrane association of KRAS4B.

Figure 5.

KRAS4B oncogenic mutants sense the nanotopography-induced membrane curvature differently. (A) A scanning electron microscopy image (tilted at 45°) of the 250-nm-wide nanobar arrays. Scale bar, 2 μm. (B) An individual nanobar with 300 nm in height, 2 μm in length, and 250 nm in width in scanning electron microscopy and its schematics. Scale bar, 1 μm. (C) Schematic illustration of nanobar-induced plasma membrane wrapping and the curved membrane sensing of Ras proteins. Ras preferentially accumulates at the curved ends than the flat membrane along the side wall of the nanobar. (D) Sample fluorescent images of U2OS cells expressing GFP-tagged KRAS4B^G12C (G12C), KRAS4B^G12D (G12D), KRAS4B^G12V (G12V) KRAS4B^G13D (G13D), and KRAS4B^Q61H (Q61H) cultured on nanobars. Scale bar, 10 μm for cell overview, 3 μm for inset. (E) Bar plot of EI of KRAS4B^G12C (G12C), KRAS4B^G12D (G12D), KRAS4B^G12V (G12V) KRAS4B^G13D (G13D), and KRAS4B^Q61H (Q61H) shown in median with 95% CI. More than three independent experiments with ≥5 cells per batch were analyzed. Kolmogorov–Smirnov tests (unpaired, nonparametric) were performed for statistical analysis. * for P < 0.05, ** for P < 0.01, *** for P < 0.001, **** for P < 0.0001. (F) Relative frequency distributions of EI of KRAS4B^G12C (G12C), KRAS4B^G12D (G12D), KRAS4B^G12V (G12V), KRAS4B^G13D (G13D), and KRAS4B^Q61H (Q61H) with Lognormal fitting curves.

Figure S3.

Nanotopography-induced membrane curvature regulates lipid distribution differently. (A) Fluorescent images of U2OS cells expressing GFP-tagged lipid probes of PS, PIP₂, PIP₃, PA, and cholesterol compared to cell mask staining cultured on nanobars. Scale bar, 10 μm for cell overview, 3 μm for inset. (B) Bar plot of the EI of cell mask and lipid probes of PS, PIP₂, PIP₃, PA, and cholesterol is shown as median with 95% CI. Kolmogorov–Smirnov tests (unpaired, nonparametric) were performed for statistical analysis. **** for P < 0.0001. (C) Relative frequency distributions of EI of lipid probes of PS, PIP₂, PIP₃, PA, and cholesterol compared with cell mask with Lognormal fitting curves.

Discussion

RAS signaling is mostly compartmentalized to the cell PM (Cox et al., 2015). The C-terminal hypervariable regions (HVRs) of RAS proteins contain membrane anchors and have been primarily attributed to their membrane association. It has long been observed that the same RAS isoforms bound to different guanine nucleotides occupy non-overlapping spaces on the PM (Plowman et al., 2005; Prior et al., 2003), suggesting that RAS G-domains play roles in lipid binding. Indeed, GTP- and GDP-bound G-domain of HRAS possess distinct conformations, which alters the insertion depth of its dual palmitoyl chains in the membranes and results in the preferential partitioning of the GTP-bound HRAS to the cholesterol-independent domains and the GDP-bound HRAS to the cholesterol-enriched domains (Abankwa et al., 2008, 2010; Li and Gorfe, 2013). KRAS4B G-domain is more dynamic, but also adopts distinct conformational orientations and dynamically interacts with lipids at the membrane surface (Neale and García, 2020; Prakash et al., 2019; Prakash et al., 2016; Sarkar-Banerjee et al., 2017). KRAS4A G-domain also samples distinct orientation equilibrium when anchored to bilayers composed of pure phosphatidylcholine (PC), PC/PS, and PC/PIP₂ (Li and Buck, 2017). By comparing binding to three PS species as probes for distinct PM regions, we, here, show that RAS G-domains contribute to the recognition of lipids.

We propose that reorientation of RAS G-domains mechanistically contributes to their intricate sensing of local lipid environments on the PM. KRAS4B G-domain mainly samples among three orientation states, OS0, OS1, and OS2 (Cao et al., 2019; Neale and García, 2020; Packer et al., 2021; Prakash et al., 2019; Prakash et al., 2016; Sarkar-Banerjee et al., 2017). The α3-5 helices in OS1 and the β1-3 loops in OS2 contact membranes, with an intermediate OS0 state (Fig. 2). A cohort of charged residues interact with the charged lipids: Lys 16, Asp 47, Glu 49, Lys 88, Asp 92, Arg 97, Glu 98, Lys 101, Lys 104, Glu 107, Asp 108, Lys 128, Asp 132, and Arg 135 in OS1, and Arg 73 and Arg 102 in OS2 (Cao et al., 2019; Neale and García, 2020; Prakash et al., 2019; Prakash et al., 2016; Sarkar-Banerjee et al., 2017). Tyr 64, Ser 65, Arg 68, Tyr 71, and Arg 73, also associate with the zwitterionic PC lipids (Neale and García, 2020; Prakash et al., 2016). As illustrated in the schematic in Fig. 2 adopted from Prakash et al. (2019), the HVR of KRAS4B alters conformations in different OS states, with the HVR facing lobe 2 (amino acids 87–166) in OS 1 and facing lobe 1 (amino acids 1–86) in OS 2. Thus, shifting reorientation equilibrium among different OS states of the G-domain may influence the conformation of the farnesylated HVR of KRAS4B and contribute to the sensing of lipid headgroups and acyl chains. This view is supported in Fig. 2, where mutating residues at the membrane-interacting interfaces, such as Arg 73, Arg 102, and Arg 135, differentially modify the binding of PS species by the wild-type KRAS4B and KRAS4B^G12V (Fig. 2). It has been proposed that KRAS4B^G12V disfavors OS2 (Prakash et al., 2019). This is consistent with our finding that mutating Arg 135, which associates with PS in OS1 (Neale and García, 2020; Prakash et al., 2019; Prakash et al., 2016; Sarkar-Banerjee et al., 2017), most profoundly alters the binding of PS species of KRAS4B^G12V (Fig. 2, A and D), while having less effects on recognizing PS acyl chains by the wild-type KRAS4B (Fig. 2, E and H). Mutating Arg 102 did not impact the PS acyl chain recognition of KRAS4B^G12V (Fig. 2 C), but significantly altered PS acyl chain selectivity of the wild-type KRAS4B (Fig. 2 G). Since Arg 102 associates with PS in OS2 (Neale and García, 2020; Prakash et al., 2019; Prakash et al., 2016; Sarkar-Banerjee et al., 2017) and KRAS4B^G12V disfavors OS2 (Prakash et al., 2019), these data support our view that reorientation of G-domain contributes to the recognition of lipid acyl chains.

While there is a consensus that KRAS4B has high affinities for anionic phospholipids, it has been debated which anionic lipid KRAS4B prefers to bind. Super-resolution imaging of live/fixed cells or intact cell PM suggests that KRAS4B extensively associates with PS, especially the mixed-chain PS species (Koester et al., 2022; Lee et al., 2019, 2021; Zhou et al., 2014, 2015, 2017, 2021b). MD simulations and in vitro assays reveal preferential binding of KRAS4B to PIP₂ (Cao et al., 2019; Gregory et al., 2017; Ingólfsson et al., 2022; McLean et al., 2019; Shrestha et al., 2021). Similarly, KRAS4A has been shown to prefer PS on intact cell PM (Maxwell et al., 2018), but PIP₂ in MD simulations (Li and Buck, 2017). In vitro binding and imaging show that bilayer binding of the purified KRAS4B and recruitment of effector CRAF fragments favor the unsaturated PS species over the saturated PS, PA, and PIP₂ (Lakshman et al., 2019). The acyl chain-dependent co-existing PS pools may provide an explanation for the different observations. In native membranes, the saturated lipids are commonly found to be enriched with PIP₂, cholesterol, and/or actin (Rosenhouse-Dantsker et al., 2023). Thus, the affinities for the saturated PS may be consistent with the strong PIP₂ binding observed in previous studies. Indeed, in our previous studies, the PIP₂-enriched RAS constructs, such as HRAS anchor (tH), the phosphorylated KRAS4B and KRAS4B with a geranylgeranyl anchor, always bind to the saturated PS species (Liang et al., 2019; Zhou et al., 2014, 2015, 2017, 2021b). We now show that wild-types and G12V mutants of KRAS4A and KRAS4B associate with the saturated and unsaturated PS species with varying affinities (Fig. 1). As discussed above, the reorientation of KRAS4B G-domain contributes to the shift between the saturated and unsaturated PS species. Indeed, the reorientation of KRAS4B G-domain has been proposed to contribute to the PIP₂ affinity (Cao et al., 2019; Gregory et al., 2017; Ingólfsson et al., 2022; McLean et al., 2019; Shrestha et al., 2021). Thus, it is possible G-domains of KRAS4A and KRAS4B may experience distinct reorientation equilibria under distinct experimental and theoretical conditions in various model systems, thus altering their preferences between domains enriched with the saturated PS/PIP₂ and domains containing unsaturated PS.

Additional mechanisms can also contribute to how RAS G-domains detect lipid acyl chain structures. RAS G-domains specifically interact with complex constituents, such as RAF/PI3K effectors, actin, galectins, and chaperones such as phosphodiesterase and/or G protein-coupled receptor GPR31 (Belanis et al., 2008; Chandra et al., 2011; Elad-Sfadia et al., 2004; Fehrenbacher et al., 2017; Plowman et al., 2005; Posada et al., 2017; Schmick et al., 2014; Shalom-Feuerstein et al., 2008). Indeed, the formation of signalosomes composed of KRAS4B, effectors BRAF/CRAF, downstream kinase MEK, galectin-3, and 14-3-3 exposes G-domain α4 and α5 helices to membranes (Mysore et al., 2021). Posttranslational modifications, such as acetylation of lysine 104 (K104), alter allosteric dynamics of switches I and II of KRAS4B, which attenuates the guanine nucleotide exchange factor–induced nucleotide exchange (Yang et al., 2012). K104 acetylation compromises the oncogenic activities of KRAS4B mutants (Yang et al., 2012, 2023). As such, the G-domain–mediated formation of signaling platforms, such as signalosomes, on the PM may impact the select lipid association of RAS. As listed above, K104 has also been predicted to bind to charged lipid headgroups (Cao et al., 2019). Acetylation of K104 may also directly alter lipid binding. Taken together, the complex nanoclusters and signaling platforms of RAS may integrate a series of protein/protein and protein/membrane interactions and allow synergistic regulation of lipid preferences observed in different systems.

Although all oncogenic mutations allow RAS to remain active for prolonged periods, their oncogenic activities differ significantly (Burge and Hobbs, 2022; McCormick, 2019; Prior et al., 2020; Yang et al., 2023). KRAS4B^G12V is more efficient than KRAS4B^G12D in the activation of the mitogen-activated protein kinases (Li et al., 2018). The mechanisms for these allele-specific pathological activities are still poorly understood. We here show that KRAS4B^G12D, KRAS4B^G12V, and KRAS4B^Q61H prefer the mixed-chain PS species, whereas KRAS4B^G12C and KRAS4B^G13D favor the saturated PS, PIP₂, and/or cholesterol (Figs. 3 and 4). The observed allele-specific lipid association has important biological and pathological implications. For instance, effector CRAF must be recruited to cholesterol-poor regions of the PM to propagate signaling downstream (Inder et al., 2008). CRAF only associates with KRAS4B^G12V in the presence of the mixed-chain PS species, but not other PS species tested (Zhou et al., 2017). Indeed, we show that KRAS4B^G12C and KRAS4B^G13D recruit CRAF less efficiently than KRAS4B^G12D, KRAS4B^G12V, and KRAS4B^Q61H (Fig. 4 I). Concordantly, cholesterol depletion attenuated the PM association of KRAS4B^G12C and KRAS4B^G13D (Fig. 4, G and H). Interestingly, cholesterol depletion disrupted the lateral nanoclustering and PM binding of KRAS4B^G12C but only mislocalizes KRAS4B^G13D from the PM. This suggests distinct patterns of PM association for KRAS4B^G12C and KRAS4B^G13D although both associate with cholesterol, which is suggested by distinct curvature sensing abilities of the two mutants (Fig. 5). The distinct lipid preferences of these KRAS4B oncogenic mutants also suggest that they sense and respond to changing membrane properties in distinct manners. Indeed, we now show that KRAS4B oncogenic mutants display allele-specific responses to changing membrane curvature, suggesting potential allele-specific effects on intracellular transport and morphology.

In accordance with the reorientation equilibrium model discussed above, we propose that G-domains of KRAS4B oncogenic mutants undergo the allele-specific reorientation and contribute to the allele-specific recognition of lipid headgroups and acyl chains. Indeed, recent all-atom MD simulations predicted that the local residue conformations, especially the time-averaged distance between Switch II and α3 helix, of G12C, G12V, G13D, and Q61H mutants of KRAS4B differ significantly, with more profound differences between G12X mutations and G13D/Q61H (Vatansever et al., 2020). Pantsar et al. (2018) also compared a large cohort of G12X mutations of KRAS4B, including G12A, G12C, G12D, G12R, G12S, and G12V, in their all-atom MD simulations and revealed marked allele-specific conformational differences among G12X mutations. Thus, the distinct reorientation equilibria of G-domains of KRAS4B oncogenic mutants may contribute to their different lipid preferences. Indeed, a recent single-molecule imaging reveals that KRAS4B^G12D possesses less monomer population than KRAS4B^Q61H on native membranes (Walker et al., 2024). Further, binding of G12C-specific inhibitors, such as MRTX849 (Adagrasib) or AMG510 (Sotorasib), modulates the conformational dynamics of KRAS4B^G12C (Issahaku et al., 2023; Pantsar, 2020; Vasta et al., 2022). Since the G12 position is away from the membranes, binding of these inhibitors does not directly impact membrane association of KRAS4B^G12C. Interestingly, MRTX849 treatment mislocalized KRAS4B^G12C from the PM and disrupts the PM nanoclustering of KRAS4B^G12C, while having no effect on KRAS4B^G12D (Morstein et al., 2023). Taken together, the distinct reorientation and conformational dynamics of KRAS4B oncogenic mutants may contribute to their allele-specific lipid association.

Conclusion

Mutant RAS proteins primarily signal from their spatially distinct nanoclusters on the PM, largely mediated by associations of their membrane anchors and PM lipids. Here, we show evidence that the allosteric reorientation of RAS G-domains participates in the intricate sensing of lipid headgroups and acyl chains. This mechanism contributes to the allele-specific pathological activities of KRAS4B oncogenic mutants. Peripheral juxtamembrane domains are commonly found in many membrane proteins. Similar to the G-domains of RAS, these domains are largely cytosolic and intrinsically disordered and have been ignored for their roles in membrane interactions. We, here, propose that allosteric dynamics of these membrane peripheral domains may present distinct membrane-associating surfaces and participate in intricate lipid binding.

Materials and methods

Cell lines

Cell lines are listed in Table 1.

Table 1.

Cell lines

Reagents	Source	Catalog number
BHK-21 (Cellosaurus accession number: CVCL_1914)	Gift of Dr. Robert G. Parton (University of Queensland, Brisbane, Australia)	N/A
Chinese ovarian CHO-K1 PSA-3 cells (Cellosaurus accession number: CVCL_A5KQ)	Gift of Dr. Tomohiko Taguchi (University of Tokyo, Tokyo, Japan)	N/A
U2OS (Cellosaurus accession number: CVCL_0042)	Gift of Dr. David G. Drubin (University of California, Berkeley, Berkeley, CA, USA)	N/A

Phospholipids

Phospholipids are listed in Table 2.

Table 2.

Phospholipids

Reagents	Source	Catalog number
DSPS	Avanti Polar Lipid, Inc.	840029P
DOPS	Avanti Polar Lipid, Inc.	840035C
SOPS	Avanti Polar Lipid, Inc.	840039C

Antibodies

Antibodies are listed in Table 3.

Table 3.

Antibodies

Reagents	Source	Catalog number
GFP (rabbit monoclonal) for EM-immunolabeling	Gift of Dr. John F. Hancock (University of Texas Health Science Center, Houston, TX, USA)	N/A
RFP (rabbit monoclonal) for EM-immunolabeling	Gift of Dr. John F. Hancock (University of Texas Health Science Center, Houston, TX, USA)	N/A

Chemicals, peptides, and recombinant proteins

Chemicals, peptides, and recombinant proteins are listed in Table 4.

Table 4.

Chemicals, peptides, and recombinant proteins

Reagents	Source	Catalog number
DMEM	Hyclone	SH30243.01
F-12K medium	Corning	10-025-CV
FBS	Gibco	16000-044
Dialyzed FBS	Gibco	26400-036
BCS	Hyclone	SH30072.03
Trypsin-EDTA (0.5%)	Gibco	15400-054
Trypsin-EDTA (0.25%)	Gibco	25200-072
Pioloform	Ted Pella	19244
Chloroform	Electron Microscopy Sciences (EMS)	12550
Poly-L-lysine	Sigma-Aldrich	P8920-100ML
EM copper grids	EMS	G200-CU
Trisodium citrate	Sigma-Aldrich	C3674-100G
Tannic acid	EMS	21710
Potassium carbonate	Sigma-Aldrich	P5833-500G
Potassium acetate	Sigma-Aldrich	P1190-1KG
Glycerol	Sigma-Aldrich	G55-1L
Paraformaldehyde	EMS	1570
Glutaraldehyde	EMS	16220
Fish gelatin	Sigma-Aldrich	G7765-250ML
BSA	Sigma-Aldrich	A6003-100G
Gold chloride	Sigma-Aldrich	254169-5G
Lipofectamine	Invitrogen	50470
Opti-MEM	Gibco	31985-070
PMMA	MicroChemicals	950A-2
Protective coating electra 92	Allresist	AR-PC 5090.02
Isopropanol	Thermo Fisher Scientific	327270010
MIBK	MicroChemicals	2016
Chromium	Lee & Lim International	N/A
CF4 and CHF3, Oxford plasmalab 80	Oxford Instruments	N/A
Chromium etchant	Sigma-Aldrich	651826

Recombinant DNA

Recombinant DNA are listed in Table 5.

Table 5.

Recombinant DNA

Reagents	Source	Catalog number
RFP-KRAS4A_WT	This paper	N/A
RFP-KRAS4AG12V	Maxwell et al. (2018)	N/A
RFP-tK(4A)	This paper	N/A
RFP-KRAS4B_WT	Zhou et al. (2017)	N/A
RFP-KRAS4BG12V	Zhou et al. (2017)	N/A
RFP-tK(4B)	Zhou et al. (2017)	N/A
RFP-HRAS_WT	Zhou et al. (2014)	N/A
RFP-HRASG12V	Zhou et al. (2014)	N/A
RFP-tH	Zhou et al. (2014)	N/A
RFP-KRAS4BR73Q	This paper	N/A
RFP-KRAS4BR102Q	This paper	N/A
RFP-KRAS4BR135Q	This paper	N/A
RFP-KRAS4BG12VR73Q	This paper	N/A
RFP-KRAS4BG12VR102Q	This paper	N/A
RFP-KRAS4BG12VR135Q	This paper	N/A
RFP-KRAS4BG12C	This paper	N/A
RFP-KRAS4BG12D	This paper	N/A
RFP-KRAS4BG13D	This paper	N/A
RFP-KRAS4BQ61H	This paper	N/A
GFP-KRAS4BG12C	This paper	N/A
GFP-KRAS4BG12D	This paper	N/A
GFP-KRAS4BG12V	Zhou et al. (2017)	N/A
GFP-KRAS4BG13D	This paper	N/A
GFP-KRAS4BQ61H	This paper	N/A
GFP-LactC2	Gift of Dr. Sergio Grinstein (The Hospital for Sick Children, Toronto, Canada)	N/A
GFP-PH-PLCδ	Gift of Dr. Guangwei Du (University of Texas Health Science Center, Houston, TX, USA)	N/A
GFP-PH-Akt	Gift of Dr. Guangwei Du (University of Texas Health Science Center, Houston, TX, USA)	N/A
GFP-PASS	Gift of Dr. Guangwei Du (University of Texas Health Science Center, Houston, TX, USA)	N/A
GFP-D4H	Gift of Dr. Gregory D. Fairn (Dalhousie University, Halifax, Canada)	N/A

Cell culture and transfection

BHK cells and PSA3 cells were grown in DMEM (10% bovine calf serum [BCS]) and F-12K medium (10% DFBS without/with 10 μM ethanolamine), respectively. For transient transfection, 1 μg of cDNA encoding a GFP-tagged protein and 1 μg of cDNA encoding an RFP-tagged protein were added in 100 μl of opti-MEM containing 5 μl of lipofectamine. Following a 20-min incubation at room temperature, the mix was added to cells (washed twice with serum-free medium) and incubated at 37°C for 5 h. Cells were then incubated in DMEM (10% BCS) or F-12K (10% DFBS) for 18 h before experiments.

For U2OS cells, after removing the chromium (Cr), the nanostructured chips were cleaned in piranha solution for 2–3 h and air plasma (Harrick Plasma) for 1 h. Then the chips were attached to 40-mm culture dishes and ready for use. UV treatment was done for 30 min before cell culture each time to prevent contamination. Before plating the cells, fibronectin of 5 μg/ml was coated onto the chips and incubated for 30 min at room temperature. The Homo sapiens bone osteosarcoma U2OS cells were cultured for experiments. DMEM with GlutaMAX was used for cell growth supplemented with 10% FBS and 1% Penicillin-Streptomycin. To perform transfection for the transient expression of Ras variants, a cell confluency of around 80% was reached on the day of transfection with lipofectamine 3000 transfection kit. The total transfection volume was 125 μl and the amount of plasmid added was optimized, mostly to be 1 μg. Starvation was conducted on the cells in Opti-MEM for 4 h followed by overnight protein expression in FBS-starved DMEM in a standard incubator. The same protocol was applied for the transfection of mEGFP-tagged lipid probes, LactC2 for PS, PLCd-PH for PIP₂, Akt-PH for PIP₃, PASS for PA, and D4H for cholesterol into the cells cultured on nanostructures. In addition, to stain the PM of the cells grown on nanobars, DeepRed CellMask (Invitrogen) with 1,000-time dilution was used and incubated for 2 min at 37°C. The dye was subsequently quickly removed to prevent its intake into the cell.

PS depletion and acute addback

PSA3 cells were grown in F-12K medium containing 10% DFBS for 72 h to deplete endogenous PS. Supplementation of 10 μM Etn in F-12K medium (10% DFBS) for 72 h restored normal endogenous PS in PSA3 cells. For acute addback, all synthetic PS species were dissolved in chloroform at 1 mg/ml, purged with nitrogen, and stored in glass vials with Teflon-lined caps at −20°C. On the day prior to the addback experiments, an appropriate amount of lipid/chloroform solution (for a final working concentration of 10 μM) was spotted on the bottom of a glass vial. Chloroform was evaporated under vacuum overnight. On the day of the experiment, the dried lipid film was rehydrated with F-12K medium (10% DFBS) and sonicated for 20 min before being applied to PSA3 cells. PSA3 cells were incubated with F-12K medium containing the sonicated synthetic PS species for 1 h before the EM procedures.

EM-spatial analysis

EM-univariate nanoclustering

The K-function univariate analysis quantifies the extent of nanoclustering of 4.5-nm gold nanoparticles conjugated to primary antibodies against a targeted protein on intact PM sheets (Zhou et al., 2015, 2017). To probe nanoclustering of a GFP-tagged RAS, the apical PM of BHK cells ectopically expressing GFP-tagged RAS was attached to copper EM grids. Following fixation with 4% paraformaldehyde (PFA) and 0.1% glutaraldehyde, GFP on the PM sheets was immunolabeled with 4.5-nm gold nanoparticles conjugated to anti-GFP antibody and embedded in methylcellulose containing 0.3% uranyl acetate. Gold distribution on the intact PM sheets was imaged using TEM (JEM-1400; JEOL USA) at 100,000× magnification at room temperature. Images were acquired via Gatan Orius SC1000A 1 digital camera (Gatan) and Gatan DigitalMicrograph software (License ID: 165642147, Architecture: 32-bit, Version 2.02.800.0). The coordinates of every gold particle were assigned via ImageJ (National Institutes of Health and Laboratory for Optical and Computational Instrumentation). Nanoclustering of gold particles within a selected 1 μm² PM area was quantified using Ripley’s K-function under a null hypothesis that the point pattern in a selected area is distributed randomly:

$$K\left(r\right)=\mathrm{A}{n}^{-2}\sum _{i\ne j}{w}_{ij}1\left({\Vert x}_i-x_j\Vert \le r\right)$$ (1)

$$L\left(r\right)-r=\sqrt{\frac{K\left(r\right)}{\pi }}-r,$$ (2)

where K(r) indicates the univariate K-function for n gold nanoparticles in an intact PM area of A; r is the length scale between 1 and 240 nm with an increment of 1 nm; and || ^. || is Euclidean distance, where the indicator function of 1(^.) = 1 if ||x_i-x_j|| ≤ r and 1(^.) = 0 if ||x_i-x_j|| > r. To achieve an unbiased edge correction, a parameter of w_ij⁻¹ was used to describe the proportion of the circumference of a circle that has the center at x_i and radius ||x_i-x_j||. K(r) is then linearly transformed into L(r) – r, which is normalized against the 99% CI estimated from Monte Carlo simulations. An L(r) – r value of 0 for all values of r indicates a complete random distribution of gold. An L(r) – r value above the 99% CI of 1 at the corresponding value of r indicates statistical clustering at a certain length scale. Approximately 15–20 PM sheets were imaged, analyzed, and pooled for each condition in the current study. Statistical significance was evaluated by comparing our calculated point patterns against 1,000 bootstrap samples in non-parametric bootstrap tests (Zhou et al., 2015, 2017).

EM-bivariate co-clustering analysis

The K-function bivariate co-clustering analysis quantifies the co-clustering between two differently sized gold nanoparticles tagging two different constituents on the intact PM sheets (Zhou et al., 2015, 2017). Similar to the univariate nanoclustering protocol described above, intact apical PM sheets of PS auxotroph PSA3 cells co-expressing GFP-LactC2 (probing PS lipids) and an RFP-tagged RAS construct were attached to EM grids and fixed with 4% PFA and 0.1% glutaraldehyde. The PM sheets were incubated with 6-nm gold nanoparticles linked to anti-GFP antibody, blocked with 0.2% bovine serum albumin (BSA) and 0.2% fish skin gelatin, and then incubated with 2 nm gold conjugated to anti-RFP antibody. Gold distribution on the intact PM sheets was imaged using TEM (JEM-1400; JEOL USA) at 100,000× magnification at room temperature. Images were acquired via Gatan Orius SC1000A 1 digital camera (Gatan) and Gatan DigitalMicrograph software (License ID: 165642147, Architecture: 32-bit, Version 2.02.800.0). ImageJ was used to assign coordinates to the gold nanoparticle. A bivariate K-function analysis tested the null hypothesis that the two populations of gold particles spatially separate from each other (Eqs. 3, 4, 5, and 6):

$$K_{biv}\left(r\right)={\left(n_b+n_s\right)}^{-1}\left[n_b{K}_{sb}\left(r\right)+n_s{K}_{bs}\left(r\right)\right]$$ (3)

$$K_{bs}\left(r\right)=\frac{A}{n_b{n}_s}\sum _{i=1}^{n_b}\sum _{j=1}^{n_s}{w}_{ij}1\left(x_i-x_j\le r\right)$$ (4)

$$K_{sb}\left(r\right)=\frac{A}{n_b{n}_s}\sum _{i=1}^{n_s}\sum _{j=1}^{n_b}{w}_{ij}1\left(x_i-x_j\le r\right)$$ (5)

$$L_{biv}\left(r\right)-r=\sqrt{\frac{K_{biv}\left(r\right)}{\pi }}-r$$ (6)

where K_biv(r) denotes a bivariate estimator and contains two individual bivariate K-functions: K_bs(r) quantifies how the big 6-nm gold particles (b = big gold) distribute around each 2-nm small gold particle (s = small gold); K_sb(r) describes how small gold particles distribute around each big gold particle. The value of n_b indicates the number of 6-nm big gold and n_s indicates the number of 2-nm small gold within a PM area of A. Other parameters denote the same definitions as defined in the univariate calculations in Eqs. 1 and 2. L_biv(r) − r is a linear transformation of K_biv(r) and is normalized against the 95% CI). An L_biv(r) − r value of 0 indicates spatial segregation between the two populations of gold particles, whereas an L_biv(r) − r value above the 95% CI of 1 at the corresponding distance of r indicates yields statistically significant co-localization at certain distance yields. Area-under-the-curve for each L_biv(r) − r curve was calculated within a fixed range 10 < r < 110 nm and was termed bivariate L_biv(r) − r integrated (or LBI):

$$LBI=\underset{10}{\overset{110}{\int }}Std{L}_{biv}\left(r\right)-r.dr\text{.}$$ (7)

For each condition, ∼15–20 PM sheets were imaged, analyzed, and pooled, shown as mean of LBI values ± SEM. Statistical significance between conditions was evaluated by comparing against 1,000 bootstrap samples in non-parametric bootstrap tests as described (Zhou et al., 2015, 2017).

Limitations of EM-spatial analyses have been recently discussed in detail (Zhou and Hancock, 2018b, 2021). Binding of fluorescently tagged lipid-binding domains to lipids in the PM may impact the fluidity and diffusion of these lipids. Highly specific binding of these lipid-binding domains may also quench the availability of these lipids to other binders. We and others have extensively validated that the expression of these lipid-binding domains effectively probe distinct PS species and do not interfere with the association of non-specific binders such as small GTPases (Del Vecchio and Stahelin, 2018; Garcia et al., 1995; Kay et al., 2012; Koester et al., 2022; Lee et al., 2019; Zhou et al., 2017, 2021b). Fluorescence protein conjugated to the N-terminal of the small GTPases may impact the orientation and conformational changes of the GTPases. Experimental results extracted from the GFP/RFP-tagged RAS isoforms in cells have been in agreement with theoretical predictions and in vitro assays using untagged proteins (Abankwa et al., 2008, 2010; Cao et al., 2019; Fang et al., 2020; Janosi et al., 2012; Mazhab-Jafari et al., 2015; Prakash et al., 2016, 2019).

Fabrication of quartz nanostructures

The nanobar structure utilized here was fabricated on the quartz chip (1.5 × 1.5 cm) using the EBL. The quartz chip was initially spin-coated with a 300-nm positive electron-beam resist poly(methyl methacrylate) (PMMA) (MicroChem), followed by a conductive polymer layer, AR-PC 5090.02 (Allresist). Customized nanostructure patterns were then exposed by EBL (FEI Helios NanoLab). After that, a mixture of isopropanol:methylisobutylketone (MIBK) = 3:1 solution was used to eliminate the exposed PMMA. Subsequently, a 30-nm Cr mask was formed via thermal evaporation (UNIVEX 250 Benchtop) and lift-off in acetone. Nanostructures were finally etched down through reactive ion etching with a combination of CF4 and CHF3 (Oxford Plasmalab 80). Scanning electron microscopy (FEI Helios NanoLab) imaging was conducted after 10-nm Cr coating to assess the dimensional properties of nanostructure arrays. Before use, the nanochips were treated with Cr Etchant to get rid of any residual metal layer.

Live cell imaging and image analysis

The nanostructured chip was attached to the culture dish for the purpose of live cell imaging. A Nikon Ti2-E inverted microscope with a total internal reflection fluorescence (TIRF) module (iLasV2 Ring TIRF; GATACA Systems) and an ORCA-Fusion sCMOS camera (Hamamatsu Photonics) was used for cell imaging on a Plan-Apochromat 100×/1.45 oil objective. A laser of 488 nm/150 mW (Vortran) excited the fluorescence-tagged Ras mutants and the images were taken with 50% laser power and 2,000-ms exposure time. During imaging, the cells grown on the chip were maintained in FBS-starved DMEM with an on-stage incubator providing an incubation condition of 5% CO₂ and 37°C (Live Cell Instrument). The image acquisition was done using MetaMorph software (Molecular Device).

EI, referring to the fluorescence intensity ratio at the nanobar ends to the nanobar center, was calculated to quantify the curvature-guided Ras signal accumulation. The raw fluorescence image of Ras was firstly preprocessed by subtracting the background signals using a rolling ball algorithm (radius = 10 pixels for 250-nm-diameter nanobars) in ImageJ. Then the positions of the nanobars which were covered by the cell of interest were identified by a custom-written MATLAB code. By specifying the nanobar length, the two nanobar ends and the nanobar center can be anchored for the intensity value extraction. The intensity values were validated with ImageJ and those non-valid were removed. The remaining EIs were ready for plotting and analysis using Prism9. The relative frequency distribution curves were constructed using Prism9 as well.

Online supplemental material

Fig. S1 shows the chemical structures of the three synthetic PS species used in the acute addback experiments. Fig. S2 shows sample electron micrographs of the bivariate spatial analysis, as well as the sample curves for the K-function spatial calculations. Fig. S3 shows the sample confocal microscopic images and calculations of the curvature preferences of various lipids, including PS, PIP2, PIP3, PA, and cholesterol. Table S1 lists the biophysical properties of the three synthetic PS species used in the acute addback experiments.

Supplementary Material

Table S1

shows distinct biophysical properties of PS species.