Positive cooperativity between RAS-binding and cysteine-rich domains regulates RAF membrane binding kinetics via lateral rebinding

Abstract

RAF activation requires coordinated interactions with both RAS and membrane lipids, yet the molecular basis of this process remains unclear. Using a bottom-up reconstitution approach, we show how coordinated protein–protein and protein–lipid interactions regulate membrane binding dynamics of RAF to drive its multistep activation. Within membrane environments, the RAS-binding domain (RBD) and cysteine-rich domain (CRD) exhibit cooperativity, with CRD-mediated phosphatidylserine binding stabilizing the RBD:RAS complex. Importantly, RAF remains membrane-bound through lateral rebinding to RAS, where a weak CRD–lipid interaction plays an essential role. The lateral rebinding extends RAF’s membrane dwell time under high RAS density conditions. This prolonged membrane residence may facilitate completion of RAF’s multistep activation. Given the high abundance of weak multivalent membrane interactions, lateral rebinding may be a common mechanism for regulating the activity of signaling proteins through sustained membrane retention.

The authors investigate RAF– membrane interactions using a quantitative bottom-up reconstitution approach. The study reveals that positive cooperativity between lipid and protein domains modulates membrane binding kinetics via lateral rebinding to RAS, providing mechanistic insight into RAF activation.

Introduction

The RAF kinases are critical regulators of the mitogen-activated protein kinase (MAPK) pathway, controlling diverse cellular processes¹. RAF isoforms (A-, B-, and CRAF) share a conserved domain organization: an N-terminal regulatory region composed of the RAS-binding domain (RBD) and the cysteine-rich domain (CRD), and a C-terminal kinase domain (Fig. 1a). In their basal state, RAF kinases reside in the cytosol as autoinhibited monomers, and they are activated when the RBD binds to RAS⋅GTP at the membrane surface. Once activated, RAF initiates the three-tiered RAF-MEK-ERK kinase cascade, serving as a central node that modulates pathway dynamics. Mutations in RAF can lead to constitutive signaling in cancers and developmental disorders^2,3. Paradoxical activation has been observed when inhibitors targeting oncogenic BRAF^V600E are used under high RAS·GTP conditions, inadvertently promoting RAF dimerization and unintended ERK pathway activation^4–11.

Fig. 1. RBD-dependent lipid engagement of CRD induces synergistic binding on RAS-functionalized supported lipid bilayers.

a Domain diagrams of RAF kinase and CRAF N-terminal regulatory domain constructs used in this study. b Structure of the fully autoinhibited CRAF/MEK1/14-3-3 complex (PDB:9MMP). c Schematic illustration of RAF binding assays on RAS-functionalized supported lipid bilayers using TIRF microscopy. Fluorescent labels are represented as a gray circle (outside the TIRF volume) and a red circle (within the TIRF volume). d Representative binding kinetics of different CRAF constructs on RAS-functionalized membranes containing 2% or 20% PS. RAF concentration: 20 nM. RAS surface density: ~600 μm⁻². Representative RAF titration binding curves. RAS surface densities are indicated with dashed lines. Solid and open circles represent equilibrium RAF surface densities. The average apparent dissociation constant, K_d, and standard deviation were obtained by fitting the data (solid lines) to a one-site specific binding model. Values are reported as mean ± SD. K_d for RAF binding to RAS-functionalized membranes was 270 ± 90 nM (2% PS) and 240 ± 20 nM (20% PS) for RBD (e), and 25 ± 10 nM (2% PS) and 6 ± 3 nM (20% PS) for RBD-CRD (f). K_d values were estimated from three independent experiments (N = 3) shown in Supplementary Fig. 9. For all other conditions, K_d was too high to be determined using TIRF microscopy. For CRD (g), similar results were reproduced in two independent experiments, including results shown in Supplementary Fig. 10. Source data are provided as a Source Data file.

RAF activation proceeds through multiple intermediates governed by complex molecular interactions on membrane surfaces^1,12–14. However, the precise mechanisms underlying the transition from inactive monomer to active dimer remain to be fully determined. In the active state, RAF associates with the plasma membrane via interaction of its RBD with active RAS⋅GTP^15,16. The CRD of RAF further stabilizes its membrane association by engaging both RAS and negatively charged lipids, such as phosphatidylserine (PS)^17–19. Thus, anchored at the membrane, RAF becomes dephosphorylated by the MRAS-SHOC2-PP1C phosphatase complex and dimerizes via its kinase domain—an essential step for allosteric activation^20–24. The dimeric 14-3-3 protein reinforces an active dimeric assembly by binding a phosphoserine-containing motif on each RAF kinase domain^25,26.

Recent cryo-electron microscopy (EM) structures of RAF provide important insight into how RAF activity is regulated^27–30. The inactive monomeric CRAF:MEK1:14-3-3 complex can adopt multiple conformations, including a fully autoinhibited confirmation (Fig. 1b) that closely resembles that of BRAF, as well as more open conformations in which the CRD is structurally liberated³⁰. In the fully autoinhibited conformation of both BRAF and CRAF, the 14-3-3 dimer binds two phosphoserine residues within the CR2 and CR3 regions of the same RAF molecule, forming a cradle-like structure that sequesters the CRD^27,28. The CRD is a critical component that stabilizes the overall autoinhibited complex by making multiple contacts with the kinase domain and 14-3-3. For BRAF, RBD–RAS binding initiates the disassembly of the autoinhibited state, and CRD–lipid interaction is required for full activity³¹. Interestingly, in CRAF, the open monomer states expose the kinase dimerization interface, yet they do not spontaneously progress to form active dimers in solution³⁰. These observations highlight the engagement of active RAS at the membrane surface as a critical step in RAF activation.

A key aspect of RAF activation is that RAS nanoclusters serve as primary sites for RAF recruitment and activation by presenting high RAS density and locally enriched lipids that facilitate activation^32,33. The RBD and CRD—responsible for binding protein and lipids, respectively—are arranged in tandem and separated by a short, six-amino-acid-long linker. X-ray crystal structures show that the RBD and CRD together form an extended structural unit when bound to RAS^34,35. Prior studies reported cooperativity between RBD and CRD in which RAS–RBD binding induces a transient interaction between RAS and CRD, and mapped CRD–lipid interfaces¹⁶. However, how these interactions control membrane kinetics has remained unclear. A quantitative understanding of membrane-binding kinetics and the underlying mechanisms is crucial because RAF is exclusively activated on cellular membranes¹⁴. RAF is also among the least abundant signaling proteins in many cancer cell types³⁶. Upon receptor activation, only a small number of RAF molecules—estimated to be as few as 100–200 per cell—are recruited to the membrane, where they serve as a stoichiometric bottleneck that determines the overall signaling outcome³⁷. Therefore, understanding RAF membrane dynamics is fundamental for elucidating how normal and oncogenic RAS activities are transduced into ERK signaling³⁸.

Here, we employ a bottom-up reconstitution approach with supported lipid bilayers (SLBs) to systematically investigate how RAS proteins and lipids cooperatively regulate RAF membrane binding dynamics. By quantifying the kinetics, affinities, and diffusion of various CRAF regulatory domain constructs on RAS-functionalized membranes, we demonstrate membrane-specific positive cooperativity in which the RBD and CRD mutually modify their respective binding interactions. Specifically, RBD binding to RAS facilitates engagement of the RAF CRD with negatively charged PS lipids, resulting in two kinetically distinct steps during the binding of protein and lipid partners. This sequential association enables the CRD to bind abundant PS lipids in a RAS·GTP-dependent manner, ensuring that RAF membrane binding is strictly initiated by RAS activation. Moreover, CRD–lipid engagement reinforces RBD–RAS and CRD–RAS association, effectively reducing the membrane dissociation rate. Importantly, we show that RAF maintains its membrane-bound state through lateral rebinding to RAS, with a weak CRD–lipid interaction playing an essential role. This lateral rebinding mechanism significantly extends RAF’s membrane dwell time under conditions of high active RAS density and elevated anionic lipid content. This prolonged membrane residence may facilitate kinetic proofreading of RAF activation by increasing the likelihood of completing the multistep activation sequence. Overall, our study offers a mechanistic insight into how kinetic regulation by coordinated protein and lipid inputs modulate RAF signaling, advancing our understanding of signaling activation and specificity.

Results

RBD binding to RAS promotes lipid engagement of CRD

To investigate the mechanisms underlying RAF membrane binding, we reconstituted CRAF and RAS interaction on a SLB (Fig. 1a,c). A 20% PS composition was used to closely approximate the physiological fraction (~26%) of PS in the inner leaflet of plasma membranes while maintaining high-quality bilayers³⁹. HRAS (1-184, C118S, hereafter referred to as RAS), which retains its native cysteine palmitoylation sites (181 and 184) in the hypervariable region, was linked to lipids via maleimide chemistry. This widely used lipidation method preserves the biological function and structure of RAS and has been applied to all three RAS isoforms (H-, N-, and KRAS) and other membrane-anchored small GTPases^40–47. Maleimide-conjugated HRAS is expected to undergo rapid free rotational diffusion on membranes, allowing ready access for effector protein binding, as previously observed in fully processed NRAS and KRAS^48,49. RAS was then activated by exchanging nucleotides with GTP using the catalytic domain of Son of Sevenless (SOS) (Supplementary Fig. 1)⁴¹. The membrane binding kinetics of CRAF constructs fluorescently labeled with mNeonGreen (mNG) or AZ647 were measured using total internal reflection fluorescence (TIRF) microscopy (Fig. 1c). The surface density of membrane-bound CRAF was determined from TIRF intensity, calibrated by fluorescence correlation spectroscopy (FCS) (Supplementary Fig. 2)⁵⁰. Data demonstrating protein quality—SEC and SDS–PAGE, and single-molecule intensity analyses—are provided in Supplementary Figs. 4–6. The single-molecule intensity distribution suggests the presence of a minor dimeric fraction of the CRD-mNG construct, with no evidence of larger oligomers, at 150 mM NaCl, while all other proteins were purely monomeric.

RAF is a low-abundance protein, existing in the cytoplasm at a concentration ranging from high picomolar to low nanomolar⁵¹. Accordingly, we characterized membrane binding kinetics of truncated CRAF regulatory domains within this physiologically relevant range. At 20 nM, the RBD exhibited low-level, fast equilibrium binding (~30 molecules per µm²), independent of PS content (2% or 20%) (blue, Fig. 1d). Despite the known affinity of the CRD for PS, 20 nM of CRD displayed insignificant binding (approximately a few molecules per µm²) on both 2% and 20% PS membranes functionalized with RAS (gray, Fig. 1d). Notably, the tandem RBD-CRD exhibited a marked increase in membrane binding (~600 molecules per µm²) that depended on PS and had slower kinetics (red, Fig. 1d). This finding is intriguing because the CRD alone shows negligible binding, yet it effectively engages PS lipids when arranged in tandem with RBD. The enhanced lipid sensing exhibited by CRD requires a specific interaction between RAS·GTP and the RBD, as negligible binding was observed on RAS·GDP-functionalized membranes (Supplementary Fig. 7). Collectively, these results demonstrate that the RBD and CRD are functionally coupled, such that RBD binding to RAS promotes the CRD for lipid engagement.

To better understand how RAF integrates protein and lipid inputs for membrane binding, we sought to disentangle the contributions of protein interaction and lipid interaction to the enhanced affinity of RBD-CRD on RAS-functionalized membranes. To quantify the affinity arising solely from protein–protein interactions, the apparent dissociation constants (K_d) of RBD and RBD-CRD for RAS were determined on PEGylated (non-membrane) surfaces. An additional protein contact between RAS and the CRD increases the overall affinity of RBD-CRD by approximately fivefold—220 ± 80 nM for RBD and 40 ± 10 nM for RBD-CRD (Supplementary Fig. 8)—a trend similar to the twofold enhancement previously observed with KRAS³⁵. For membrane-tethered RAS, the RBD exhibited K_d that depended little upon PS content, measuring 270 ± 90 nM and 240 ± 20 nM at 2% and 20% PS, respectively (Fig. 1e and Supplementary Fig. 9a,b). These values are consistent with our membrane-free measurements and previously reported affinities of the CRAF RBD for HRAS^35,52. In contrast, RBD-CRD exhibited an affinity for membrane-tethered RAS more than an order of magnitude higher than that of the RBD alone (K_d = 25 ± 10 nM and 6 ± 3 nM at 2% and 20% PS, respectively; Fig. 1f and Supplementary Fig. 9c,d). Removing RAS essentially abolished this high-affinity binding (Fig. 1f). The progressive increase in affinity to RAS—from K_d = 220 nM for RBD (PEG surface), to 40 nM for RBD-CRD (PEG surface), to 6 nM for RBD-CRD (20% PS membrane)—demonstrates that low-nanomolar membrane affinity arises from the cooperative contributions of RAS–RBD binding, the additional RAS–CRD contact, and CRD–PS engagement. The CRD binds RAS with micromolar affinity via an interface distinct from the RAS:RBD contact¹⁷. However, this weak interaction between RAS and the isolated CRD does not independently promote lipid engagement, as the isolated CRD showed similarly low binding on both RAS-free and RAS-functionalized membranes (Fig. 1g and Supplementary Fig. 10). Together, these results demonstrate that both the protein (RAS) and lipid (PS) binding modalities of the CRAF CRD depend on the RBD being bound to RAS, particularly under the low nanomolar physiological RAF concentration.

CRD and lipid interaction selectively modulates the membrane dissociation rate

Characterizing the kinetic parameters of membrane binding is important for understanding the mechanisms that govern RAF activation. Given that the RBD and CRD have distinct primary binding partners—RAS and lipids, respectively—we first examined how these interactions interplay to modulate the association (k_on) and dissociation (k_off) rates. The k_on values were quantified by counting the single-molecule recruitment events of AZ647-labeled CRAF to the membrane (Supplementary Fig. 11). In line with our ensemble binding measurements, clearly spatially resolved membrane recruitment events required the interaction with RAS (Fig. 2a). Interestingly, we observed highly transient interactions of RBD-CRD (appearing as faint, diffusive traces) on RAS-free 20% PS membranes. Although those particles could not be localized, manual intensity analysis estimated their dwell times on the order of several milliseconds (Supplementary Fig. 12). Importantly, these short-lived interactions between RBD-CRD and lipids do not result in productive RAS engagement, as k_on values for both RBD and RBD-CRD were unaffected by PS concentration (Fig. 2b). This suggests that the initial recruitment is primarily driven by the protein–protein interaction between RBD and RAS, without involving CRD and lipid interactions. The CRD exerts a modest inhibitory effect on RBD–RAS binding, reducing the k_on values for RBD-CRD (see statistical test summary in Supplementary Table 1). Because the RBD-CRD exhibits a lower k_on than the RBD, its higher binding affinity must stem from a reduced membrane dissociation rate. We quantified the apparent k_off on RAS-functionalized 20% PS membranes using an ensemble free-dissociation assay under continuous buffer flow. The RBD exhibited highly dynamic and transient interactions with a half-life of ~1 s, consistent with a previous study using proteins pulled down from cell lysates⁵³. Indeed, mNG-fused RBD-CRD displayed an extended membrane residence time, with a half-life of several tens of seconds (Fig. 2c). Photobleaching had a minimal impact on the estimation of k_off (Supplementary Fig. 14). Such a long dwell time cannot be achieved solely from protein–protein interactions. The RBD-CRD dissociates from RAS on a lipid-free PEGylated surface with a half-life of a few seconds (Supplementary Fig. 15a). This kinetic parameter analysis reveals that a crucial role for the CRD–PS interaction in selectively regulating membrane dissociation kinetics without affecting the association rate.

Fig. 2. Kinetic analysis reveals that CRD selectively regulates RAF’s membrane dissociation rate.

a Representative images showing single-molecule association events of RAF constructs on RAS-functionalized and RAS-free membranes containing 20% PS. Scale bar: 3 µm. b Association rates, K_on, of various RAF-AZ647 constructs on RAS-functionalized membranes containing 2% or 20% PS. The average (horizontal bar) and standard deviation (error bars) were calculated using mean values (outlined symbols) obtained from multiple independent experiments. Non-outlined symbols represent individual data points. Each independent experiment is color-coded. Statistical comparisons were performed using one-way ANOVA followed by Tukey’s multiple-comparisons test ($\alpha$ = 0.05). P values are summarized in Supplementary Table 1. Sample sizes: RBD 2% PS (n = 10, N = 4); RBD 20% PS (n = 8, N = 3); RBD–CRD 2% PS (n = 12, N = 3); RBD-CRD 20% PS (n = 12, N = 3); CRD 2% PS (n = 5, N = 2); CRD 20% PS (n = 9, N = 2); (n = total SLBs, N = independent experiments). Protein concentration: 50 pM. c Ensemble free dissociation curves for RAF-mNG constructs on RAS-functionalized membranes containing 2% or 20% PS. Solid and open circles represent normalized time-lapse intensities. The curves were fit to a single-exponential decay to estimate half-lives, while the underlying dissociation mechanism is more complex as discussed later in this study. Similar delayed dissociation kinetics for RBD-CRD were observed in two independent experiments, including results shown in Supplementary Fig. 13. Protein concentration: 50 nM for RBD; 10 nM for RBD-CRD. d Schematic illustration of two-step membrane association mechanism in which RBD–RAS binding promotes CRD–lipid engagement. e Step size distribution analysis of RAS, RBD and RBD-CRD on RAS-functionalized membranes with 20% PS. The average (sold lines) and standard deviation (shaded regions) of step size distributions were obtained from multiple measurements (n = 5–8 SLBs) conducted in two independent experiments (N = 2). D₁ and D₂ represent diffusion coefficients of fast and slow species, and α denotes the population of the fast species. Apparent diffusion coefficients (μm²/s), calculated by the weighted average of D₁ and D₂, are 0.99 for RAS, 0.96 for RBD, and 0.35 for RBD–CRD. Individual step size distributions, curve fits, standard deviations, and parameter estimates are shown in Supplementary Fig. 17. Protein concentration: 1–10 pM for RBD-CRD; 10–50 pM for RBD. The time interval between frames was 20 ms for (b) and 10 ms (e). RAS surface density (μm⁻²): ~400 for (a); ~200–500 for (b); ~500 for (c); ~200–900 for (e). Membrane composition: 77% DOPC, 20% DOPS and 3% MCC-DOPE. Source data are provided as a Source Data file.

These results suggest that CRAF localizes to membranes through two kinetically distinct steps (Fig. 2d). In the first step, the RBD binds active RAS. In the second step, lipid engagement by the CRD—particularly a productive interaction—occurs only after the RBD has bound to RAS. This two-step mechanism ensures RAS·GTP-specific membrane localization of RAF while leveraging abundant PS lipids to prolong membrane dwell time and achieve high-affinity binding.

CRD persistently engages PS lipids upon RBD and RAS binding

To investigate how the RAS:RAF complex interacts with lipids, we performed diffusion analysis using single-particle tracking. Membranes containing 20% DOPS was functionalized with RAS at a density ranging from 200–900 per µm² (see detailed values in Supplementary Fig. 17). For single particle tracking of RAS, a small fraction of RAS was labeled with Alexa647-GppNp (a fluorescent, nonhydrolyzable GTP analog), ensuring well-dispersed single molecules (Supplementary Fig. 16). RAS alone exhibited two distinct diffusion states (gray, Fig. 2e and Supplementary Fig. 17a,b). The majority (~90% occupancy, α) had a fast diffusion coefficient of 1.05 µm²/s (D₁). The secondary, slower state exhibited a diffusion coefficient of 0.29 µm²/s (D₂). This minor slow state may result from microscale heterogeneity of PS membranes or strong coupling with the solid support, a behavior also observed in fluorescent lipids (Supplementary Fig. 18). RBD-AZ647 bound to unlabeled RAS-functionalized membranes mirrored RAS diffusion, indicating minimal lipid interaction by RBD (blue, Fig. 2 and Supplementary Fig. 17c,d). By contrast, the RBD-CRD-AZ647 diffused significantly more slowly due to lipid engagement by CRD (red, Fig. 2e and Supplementary Fig. 17e,f). Two diffusional states were identified at 0.39 µm²/s (85% occupancy) and 0.14 µm²/s. Even the faster state is substantially slower than that of RBD with minimal lipid interaction, suggesting that once recruited by RAS, the RBD-CRD persistently engages PS lipids. Because RAS-RAF complexes were tracked at the single-molecule level and separated by several micrometers, the observed decrease in diffusion coefficient reflects lipid interactions with individual RAS:RBD-CRD complexes, rather than collective behaviors such as clustering. In support of this, the RBD-CRD at saturation density displayed diffusion behavior identical to that at single-molecule density, confirming the absence of clustering in our system (Supplementary Fig. 19). Previous studies have shown that RAS:RAF clustering requires more complex lipid compositions^54,55. The profound reduction in diffusion coefficient in our study could be explained by multivalent lipid interactions, facilitated by the membrane insertion of interdigitated hydrophobic and positively charged residues within the CRD loop 1 (residues 143–149) and 2 (residues 157–164)^2,55.

Engagement of the CRD with lipids appears to be a critical step involved in the structural rearrangement of the autoinhibited RAF complex. To evaluate the kinetics and efficiency of CRD–lipid interaction following RBD binding, we compared the step-size distributions of RBD-CRD at the initial moment of contact (first-step sizes) versus the entire observation period (all-step sizes). The majority of RBD-CRD rapidly engage lipids within the 20 ms time resolution of the experiment, as evidenced by the similarity between these two distributions (Supplementary Fig. 20). This rapid lipid engagement (forward step 2 in Fig. 2d) effectively renders the two-step association mechanism unidirectional, with the first step serving as the rate-determining step. However, in the full-length protein, particularly in BRAF, CRD–lipid binding may be significantly slower, because the lipid-interacting residues of the CRD are occluded by 14-3-3 dimers. The rate of this step likely plays an important role in RAF activation and can be modulated by the degree of autoinhibition, which varies among isoforms and is alleviated by pathogenic mutations and inhibitors^2,30,56.

CRD–lipid engagement enhances the affinity of protein–protein interaction in RAS:RAF complex at the membrane

A previous computational study predicted that CRD–lipid interactions reduce fluctuations between RAS and the RBD due to the short linker, thereby potentially increasing their overall association⁵⁷. To test this prediction experimentally, we developed a FRET-based competition assay to examine whether CRD–lipid interactions influence the protein–protein interaction between RAS and CRAF (Fig. 3a). In this assay, RAS is labeled with the Atto488-GppNp donor, and CRAF (either RBD or RBD-CRD) is tagged with the AZ647 acceptor. The assay begins by forming the CRAF:RAS complex on the membrane, which generates a FRET signal. A high concentration of unlabeled RBD-CRD (3 µM) is introduced to compete with the labeled CRAF (see Supplementary Fig. 21 for the concentration optimization). Upon dissociation of AZ647-CRAF from RAS, unlabeled RBD-CRD occupies RAS, preventing the rebinding of labeled CRAF. The time-dependent change in FRET signal thus specifically tracks protein–protein dissociation (CRAF dissociation from RAS), while excluding dissociation of membrane-bound CRAF that is not associated with RAS. Using this assay, we observed that RBD-CRD dissociates from RAS significantly more slowly than RBD alone (Fig. 3b). The slower dissociation for RBD-CRD at 2% PS is primarily due to an additional interaction between the CRD and RAS, as its FRET dissociation curve closely follows the ensemble dissociation curve of RBD-CRD measured on lipid-free PEG surfaces (Supplementary Fig. 15b). Importantly, 20% PS lipids, which promote a greater degree of CRD–lipid interaction, further accentuates this slowed dissociation for RBD-CRD (Fig. 3b). These findings indicate that CRD–lipid interactions enhance the binding strength between protein domains (RAS:RBD and RAS:CRD), thereby stabilizing the complex and slowing its dissociation. Consequently, the protein–protein and protein–lipid interactions within the RAF regulatory domains mutually reinforce each other: RBD binding to RAS facilitates CRD–lipid engagement, which in turn strengthens RAS:RBD and RAS:CRD association (Fig. 3c).

Fig. 3. Lipid–CRD engagement enhances the affinity of protein–protein interaction in the RAS:RBD-CRD complex.

a Overview of the FRET-based competitive assay used to measure RAS-RAF dissociation kinetics. b FRET dissociation curves for RBD and RBD-CRD on RAS-functionalized membranes containing 2% or 20% PS (n = 6 SLBs, N = 3 independent experiments). The solid line and error bars represent the average values and standard deviations, respectively. RAS surface density: ~500 µm⁻². Protein concentration: 200 nM for RBD-AZ647 and 20 nM for RBD-CRD-AZ647; 3 µM for unlabeled RBD-CRD based on the protein concentration determined by absorbance at 280 nM. c Schematic illustration of the positive cooperativity between protein–lipid and protein–protein interactions in RAF at the membrane surface. RBD binding to RAS promotes lipid engagement by the CRD (i), which in turn reinforces the association between RAS and RBD–CRD (ii). Source data are provided as a Source Data file.

A linker connecting RBD and CRD is crucial for positive cooperativity

We aimed to identify key structural features that mediate the positive cooperativity between RBD and CRD on membranes. X-ray crystallography and solution NMR studies reported that the RBD and CRD simultaneously bind to KRAS through distinct interfaces, forming a trimeric complex^16,35. In this arrangement, the six-amino-acid-long linker connecting the RBD and CRD plays an essential role in forming the extended structure for KRAS binding by positioning these two tandem domains within close proximity (Fig. 4a). The residues involved in trimer formation are highly conserved across all RAS isoforms³⁵. Guided by the KRAS:RBD-CRD(CRAF) structure, we investigated how the trimer formation mediates the positive cooperativity between the two domains, focusing on regulating membrane binding kinetics. We introduced double alanine mutations at L136 and T178 in the RBD-CRD construct (RBD-CRD^L136A/T178A) (Fig. 4a). L136, located within the linker region, is buried within a hydrophobic pocket formed by F130 in the RBD and M183 in the CRD, bringing the two domains into close proximity³⁵. The CRD residue T178 mediates contact with RAS through hydrogen bonding³⁵. The L136A and T178A mutations are expected to disrupt the linker–RBD and CRD–RAS interactions, respectively, which would destabilize the RAS:RBD-CRD trimer complex. A previous study reported that L136A reduced KRAS binding affinity fourfold and T178A decreased RAF activity in cells³⁵. Yet, we observed only marginal binding differences between the wild-type and mutant constructs on RAS-functionalized membranes containing 20% PS (red and blue, Fig. 4b), indicating that the CRD may have sufficient mobility to change its binding interface with RAS, as previously observed in molecular dynamic (MD) simulations⁵⁸. Next, we replaced the entire linker sequence with a flexible GGGGGS linker of the same length. This flexible linker is expected to eliminate residue-specific interactions while increasing overall flexibility, which disrupts the proper orientation of the CRD. The replacement caused a modest decrease in membrane binding (yellow, Fig. 4b). To introduce a greater perturbation by increasing the physical separation and flexibility between the RBD and CRD, we characterized a construct containing an extended linker of four GGGGGS repeats (4×GGGGGS; 24 amino acids total). The extended linker constructs showed a greater reduction in membrane binding (cyan, Fig. 4b). Protein quality analyses (Supplementary Fig. 3–5) and circular dichroism (Supplementary Fig. 24) confirmed that all RBD-CRD linker variants were properly purified and folded. The reduced membrane binding observed for these constructs is therefore attributed to perturbations in the relative positioning between the RBD and CRD caused by the increased linker flexibility and length. Notably, the association rate remained largely unchanged across all four constructs (Fig. 4c), while differences in membrane binding affinity were mainly due to changes in the membrane dissociation rate (Fig. 4d). Thus, the CRD-mediated interactions (CRD–RAS and CRD–lipid) primarily contribute to membrane stabilization rather than to the initial association between RBD and RAS, consistent with the K_on and K_off analyses shown in Fig. 2.

Fig. 4. The short linker connecting RBD and CRD is critical for the positive cooperativity.

a Structure of KRAS:RBD-CRD(CRAF) trimer complex (PDB:6XI7). Inset (i) shows residues interacting with L136, which is part of the interdomain linker, and inset (ii) shows residues interacting with T178 at the CRD and RAS interface. b Binding kinetics of mNG-fused WT RBD-CRD, an L136A/T178A mutant that disrupt CRD and RAS interaction, and linker mutants with various linker lengths. The average values (solid line) and standard deviations (shaded regions) are calculated from three technical replicates (n = 3 SLBs). c The k_on values for the mNG-fused WT RBD-CRD, L136A/T178A mutant, and linker variants determined by single-molecule recruitment assays. The means (bars) and standard deviations (error bars) are obtained from three technical replicates (solid circles, n = 3 SLBs). d Ensemble free dissociation measurements of WT and mutant RBD-CRD constructs. The average values (solid line) and standard deviations (shaded regions) are calculated from three technical replicates (n = 3 SLBs). Similar results of ensemble binding and dissociation were observed in two independent experiments, including those shown in Supplementary Fig. 22. Ensemble binding (b) and dissociation (d) measurements were performed at 10 nM RAF-mNG (diluted according to the protein concentration determined by absorbance at 280 nm) on 20% PS membranes functionalized with RAS at a density of ~500 µm^-2. Single-molecule recruitment assays (c) were performed at 50 pM RAF-mNG (diluted according to the mNG concentration) with RAS at a density of ~200 µm^-2. Source data are provided as a Source Data file.

RAS density modulates RAF membrane residence through lateral rebinding

We found that the dissociation kinetics of RBD-CRD differ markedly between free dissociation assays and competitive FRET assays. In free dissociation assays (red, Fig. 5a), the half-life was several tens of seconds—an order of magnitude longer than the half-life observed in competitive FRET assays (blue, Fig. 5a). The key difference between these two assays resides in the allowance of RAS rebinding pathways. In FRET assays, rebinding from both solution and the membrane is blocked by excess unlabeled RBD-CRD competing with the labeled CRAF. In contrast, free dissociation assays only prevent solution-based rebinding via continuous buffer flow, leaving membrane-based rebinding possible. This observation suggests that prolonged RAF membrane residence involves a lateral rebinding mechanism, where RAF briefly remains on the membrane via its lipid interactions and can subsequently rebind to another RAS molecule. In cases where RAF fails to rebind to RAS, it rapidly dissociates from the membrane. Based on these results, we propose a kinetic model for RAF membrane dissociation (Fig. 5b). The process begins with RAS dissociation from the RAS:RBD:CRD complex, resulting in a transient RBD-CRD:PS intermediate. This intermediate then unbinds from lipids, leading to complete dissociation of RAF into solution. Alternatively, it can rebind laterally to RAS, restoring the RAS:RBD-CRD:PS complex and extending its membrane residence time. We exclude the scenario where RBD-CRD first loses lipid contact yet remains bound to RAS, as a RAS:RBD-CRD complex with the CRD disengaged from lipids was not observed in single-particle tracking experiments (Fig. 2e). The slow diffusion coefficient of RBD-CRD is comparable to that of PKC regulatory domains known to penetrate into the hydrocarbon core of the bilayer⁵⁹, indicating persistent lipid engagement of the CRD through its basic and hydrophobic residues in loop1 and loop 2^16,55. Even if the CRD momentarily dissociates from lipids, it is likely to re-engage rapidly while still bound to RAS, given the fast lipid engagement timescale observed in our single-molecule tracking analysis (Supplementary Fig. 20).

Fig. 5. RAS surface density modulates membrane residence time of RAF via lateral rebinding.

a Comparison between free dissociation (Fig. 2c) and competitive FRET dissociation (Fig. 3b) curves for RBD-CRD on RAS-functionalized membranes containing 20% PS. b Schematic illustration of RAF RBD-CRD dissociation mechanism involving lateral rebinding to RAS. c Simulations for RAF RBD-CRD membrane dissociation kinetics at various RAS densities, with (+) and without a rebinding pathway (−). d Experimental free dissociation kinetics of RBD-CRD under different RAS densities on 20% PS membranes. Solid circles represent normalized time-lapse intensities. The dissociation data are fit by a lateral rebinding model (solid lines). RBD-CRD concentration: 20 nM. Similar results were observed in two independent experiments, including those shown in Supplementary Fig. 23. See Supplementary Information for detailed methods for kinetic simulation and the estimated kinetic constants. Source data are provided as a Source Data file.

Kinetic simulations further illuminate how elevated RAS·GTP densities, which can be achieved under receptor-triggered signaling, regulate RAF membrane dissociation kinetics. The lateral rebinding of a transient RBD-CRD:PS intermediate to RAS is a second-order process whose rate is linearly proportional to RAS surface density. Consequently, higher RAS densities increase RAF dwell times on membranes (left, Fig. 5c). In contrast, without lateral rebinding, dissociation follows a two-step, irreversible first-order pathway, removing any dependence on RAS density (right, Fig. 5c). To validate the lateral rebinding mechanism, we measured RAF dissociation kinetics at various RAS densities, revealing a strong modulation of membrane dissociation by RAS density. In particular, RBD-CRD remained on the membrane significantly longer at higher RAS densities (Fig. 5d). Although more complex models could be constructed, this two-step dissociation model successfully and quantitatively accounts for TIRF desorption data. The calculations and estimated rate constants for RAS dissociation from RAS:RBD-CRD:PS (k₃), RAS rebinding (k_–3), and lipid unbinding (k₄) are presented in Supplementary Table 3 and its associated text.

Positively charged and hydrophobic residues in loops 1 and 2 of the CRD penetrate the bilayer and mediate membrane binding⁵⁵. To test how these specific CRD–lipid interactions contribute to lateral rebinding, we characterized an RBD-CRD⁴ᴬ construct containing four alanine mutations (K148A, L149A, K157A, and F158A; hereafter referred to as 4 A). The 4 A construct exhibited a diffusion behavior similar to that of the isolated RBD, and its membrane binding was substantially reduced compared to WT RBD-CRD (Supplementary Fig. 25a,b). Notably, the extension of membrane dwell time at higher RAS densities was significantly attenuated, confirming the critical role of CRD–lipid interactions in lateral rebinding and in the formation of the RBD-CRD:PS intermediate (Supplementary Fig. 25c). The rebinding mechanism may play a pivotal role in RAF activation under conditions of elevated RAS·GTP density. Together, these findings provide a kinetic framework for understanding how coordinated RAS and lipid inputs regulate RAF membrane residence time and activity.

Discussion

The present study provides a quantitative kinetic model of membrane interaction for RAF activation. The kinetic analysis revealed that CRAF membrane association occurs in two kinetically distinct steps: first, RBD binding to RAS, followed by CRD engagement with lipids (Fig. 2d). The RBD and RAS interaction strictly dictates the membrane association rate, k_on (Fig. 2b). In contrast, CRD-mediated interactions (CRD–RAS and CRD–lipids) selectively modulate membrane dissociation rate, k_off (Fig. 2c), synergizing with the RBD–RAS interaction to yield a low-nanomolar membrane binding affinity (Fig. 1f, solid red). This two-step association mechanism is compatible with the fully autoinhibited structure observed in BRAF and CRAF, in which the RBD remains exposed for RAS binding^27,28,30, as well as with inactive open monomeric CRAF conformations, in which the CRD is structurally liberated and accessible to PS lipids³⁰. Such an association mechanism will enable RAF to exploit abundant PS lipids as a specific cue in response to RAS activation.

Our study demonstrates that the enhanced affinity arises solely from an extended membrane dwell time (Fig. 2b,c), driven by membrane-specific positive cooperativity between the RBD and CRD. We identified three interwoven mechanisms contributing to this behavior. First, RBD binding to RAS promotes the lipid-sensing ability of the CRD, thereby enhancing overall membrane binding (Fig. 1d). This cooperative interaction provides the mechanistic basis for two-step association, with the flexibility and length of the short linker acting as critical determinants (Fig. 4). Second, membrane stabilization by the CRD arises from a synergistic effect, in which CRD–lipid interaction, enabled by RBD–RAS binding, in turn reinforces protein-protein (RAS–RBD and RAS–CRD) association (Fig. 3). This finding aligns with a computational study, which predict that CRD–membrane interactions reduce fluctuations in the RAS:RBD complex⁵⁷. Lastly, RAF dissociation involves a lateral rebinding mechanism (Fig. 5). Following RAS dissociation from the RAS:RBD-CRD:PS complex, a transient RBD-CRD:PS intermediate can remain on the membrane and laterally rebind to RAS, significantly extending membrane residence at high RAS density. In this mechanism, a weak CRD–lipid interaction plays an essential role in momentarily holding the intermediate on the membrane (Supplementary Fig. 25). Previous computational and experimental studies have shown that RBD induces local anionic lipid enrichment independently of RAS, thereby enhancing the overall membrane affinity of RBD-CRD⁶⁰. The RBD-induced local enrichment of anionic lipids may support lateral rebinding by slowing the dissociation of the transient RAS-free RBD-CRD:PS intermediate. Moreover, the orientational flexibility of the RBD-CRD, observed in MD simulations and NMR studies, likely facilitates lateral rebinding by enabling efficient sampling of conformational ensembles compatible with RAS binding^16,60. Overall, synergistic enhancement of the RBD and CRD affinities for their respective binding partners, combined with lateral rebinding, regulates RAF membrane-binding kinetics. Coexistence of protein and lipid binding is broadly observed in many signaling proteins⁶¹. Lateral rebinding may be a common mechanism to extend membrane residence via the multivalency of reversible weak interactions.

The lateral rebinding has a critical functional outcome: it prolongs RAF membrane residence at elevated RAS·GTP densities (Fig. 5d). This raises two important questions: how does membrane dwell time modulation facilitate RAF activation, and what are the functional benefits of this mechanism? We hypothesize that RAF activation is regulated by a type of kinetic proofreading mechanism^62–64, whereby molecules that remain on the membrane for extended periods are disproportionately more likely to become activated. RAF possesses two necessary features for a cytosolic enzyme to achieve kinetic proofreading on membranes: (i) elongation of dwell time, and (ii) a multistep activation process⁶³. Under this mechanism, enzymes with slow dissociation kinetics that dwell on membranes for sufficiently long periods can complete the multistep activation process, whereas short-dwelling species dissociate prematurely before activation is complete, as previously demonstrated in the guanine nucleotide exchange factor SOS⁶⁴. At elevated RAS·GTP densities under receptor-triggered signaling, the extended membrane residence of RAF may increase the activation probability of individual kinase molecules by allowing them to complete their multistep activation process. Based on previous studies and our quantitative kinetic analysis, we propose a model for RAF activation on membranes, as depicted in Fig. 6. Given that dwell time modulation is a recurring theme in multistep-activation enzymes^64–66, kinetic proofreading may serve as a regulatory mechanism in many membrane-proximal signaling reactions.

Fig. 6. A model for membrane-dependent activation of RAF.

RBD binding to RAS initiates RAF membrane recruitment (i). RAS engagement within the membrane environment displaces 14-3-3 dimers, exposing the CRD, which then engages lipids and transitions into an open conformation (ii). Successful engagement of both RBD–RAS and CRD–lipid interactions mutually enhances their affinities (depicted in the inset), further stabilizing RAF on the membrane. High-density RAS·GTP prolongs RAF membrane residence through lateral rebinding (iii). Failure to rebind leads to RAF dissociation before full activation (iv). During the extended membrane residence, the SHOC2-MRAS-PP1C complex dephosphorylates the phosphoserine residue in the CR2 region (pS365 for BRAF, pS259 for CRAF), preventing reversion to a closed conformation. Ultimately, RAF completes multistep activation by forming an active dimer (v).

It is worth mentioning that the average density of RAS on the plasma membrane is in the range of 30–150 RAS molecules per μm^2,67. RAS is not uniformly distributed; instead, it forms dynamically exchanging nanoclusters that serve as the primary sites for recruitment and activation of RAF³². The radius of a nanocluster is estimated to be 6–12 nm, with a local surface density of ~4000 to ~16,000 RAS per μm^2,32,33,67. Such high local RAS densities within nanoclusters may facilitate RAF activation by extending its membrane residence time through lateral rebinding, although this hypothesis remains to be experimentally validated.

Our in vitro reconstitution system using supported lipid bilayers has successfully quantified the membrane-specific cooperativity between the CRAF RBD and CRD, providing a promising foundation for further study. To investigate how membrane-binding kinetics influence activation outcomes under more physiologically relevant conditions, it will be necessary to use the autoinhibited full-length RAF:14-3-3:MEK complex³¹. These advancements will enable us to rigorously recapitulate the sequential steps of RAF activation and elucidate the underlying molecular mechanisms and kinetics with high resolution.

Methods

Plasmid cloning

CRAF constructs were cloned into modified 2CT pET plasmids containing an N-terminal TwinStepII tag, maltose-binding protein (MBP), and a TEV protease cleavage site, followed by the protein of interest (2CT pET-TwinStrep-MBP). For mNG labeling, three repeats of the GGGGS amino acid linker were appended to the C-terminus, followed by the mNG tag. For sortase-mediated labeling with Gly-Gly-Gly-AZ647 (Vector Laboratories), a C-terminal LPETGG sequence was introduced. CRAF domains (51-131 for RBD, 51-188 for RBD-CRD, and 138-188 for CRD) were PCR-amplified and assembled using Gibson Assembly to yield TwinStrep-MBP-TEV-CRAF-mNG. Additionally, HRAS(1-184, C118S) and the SOS1 catalytic domain (SOS_cat, 565-1049) were cloned into a pProEXHTB plasmid to construct His₆-TEV-HRAS and His₆-TEV-SOS_cat, respectively.

Protein purification

CRAF constructs and mNG were expressed in Rosetta(DE3) cells. An overnight preculture was grown, then diluted into 1 L, and further grown at 37 °C. Protein expression was induced with 30 µM IPTG at an O.D. of 0.6 and continued for an additional five hours at 32 °C. Cells were harvested by centrifugation after five hours of induction, flesh-frozen in liquid nitrogen, and stored at −80 °C until purification. For purification, cell pellets were resuspended in lysis buffer consisting of Buffer A (20 mM HEPES [pH 7.3], 500 mM NaCl, 1 mM tris(2-carboxyethyl) phosphine [TCEP], and 10% glycerol) supplemented with 15 µg⋅mL⁻¹ DNase and 1 mM phenylmethylsulfonyl fluoride (PMSF). Cells were lysed, and the lysate was clarified by centrifugation before purification using fast protein liquid chromatography on an NGC system (Bio-Rad). An initial capture step for CRAF constructs was performed using a Strep-Tactin affinity column (Strep-TactinXT 4Flow, IBA Lifesciences), followed by elution with buffer A supplemented with 50 mM biotin. HiTrap chelating HP column (Cytiva Life Sciences™ HiTrap™ Chelating HP Column) was used for the initial capturing of (His)₁₀-tagged mNG and eluted with buffer A containing 500 mM imidazole. Selected fractions were dialyzed overnight with TEV protease at 4 °C. The cleaved tag and uncleaved protein were removed by reintroducing the cleaved sample onto a Strep-Tactin column for CRAF or HiTrap chelating HP column for mNG. Further purification was performed using size-exclusion chromatography (Superdex 75 Increase HiScale, Cytiva) with buffer A. Collected fractions were analyzed by SDS-PAGE, pooled as appropriate, aliquoted, snap-frozen in liquid nitrogen, and stored at −80 °C. HRAS and SOS were expressed and purified as described elsewhere⁴¹.

SLB preparation

Glass coverslips (D263 Schott glass, Ibidi) were first cleaned by sonication in a 1:1 isopropanol/water mixture for 15 minutes, followed by incubation in 2% warm Hellmanex III for 30 min. The substrates were then etched for 10 minutes in a piranha solution (3:1 H₂SO₄/H₂O₂). The coverslips were assembled on six-channel flow chambers (sticky-Slide VI 0.4, Ibidi). SLBs were formed by fusion of small unilamellar vesicles as described elsewhere⁴¹. For SLBs containing 2% PS, the lipids were mixed at 97 mol% 1,2-dioleoyl-sn-glycero-3-phosphocholine (DOPC), 2 mol% 1,2-dioleoyl-sn-glycero-3-phospho-L-serine (DOPS), and 1 mol% 1,2-dioleoyl-sn-glycero-3-phosphoethanolamine-N-[4-(p-maleimidomethyl) cyclohexane-carboxamide] (MCC-DOPE). For SLBs containing 20% PS, the lipids were mixed at 77 mol% DOPC, 20 mol% DOPS, and 3 mol% MCC-DOPE. All lipids were purchased from Avanti Research. RAS was functionalized via a covalent reaction between the terminal cysteine residues and the maleimide group of MCC lipids as described elsewhere⁴¹. Typically, a concentration of RAS in PBS (Corning, Cat no. 46-013-CM) was incubated for 2.5 h. The reaction was terminated by incubation of 10 mM beta-mercaptoethanol (BME) for 10 minutes. Nucleotides bound to RAS were stripped with 10 mM EDTA in HBS (40 mM HEPES, 150 mM NaCl, pH 7.4). Finally, RAS was loaded with the desired nucleotide by overnight incubation of the sample with either 1 µM Atto488-GDP (Jena Bioscience) or 10 µM GppNp in HBS (40 mM HEPES, 150 mM NaCl, pH 7.4) supplemented with 5 mM MgCl₂. For mNG FCS calibration measurements, SLB containing 96 mol% DOPC and 4 mol% 1,2-dioleoyl-sn-glycero-3-[(N-(5-amino-1-carboxypentyl)iminodiacetic acid)succinyl] (DOGS-NTA) were prepared by vesicle fusion method identical to SLBs containing PS lipids. A concentration ranging from 1 to 10 nM of His₁₀-mNG was incubated in PBS for 10 min to yield various surface densities. The unbound mNG was rinsed with PBS.

PEGylated surface preparation

Glass coverslips (D263 Schott glass, Ibidi) were first cleaned by sonication in a 1:1 isopropanol/water mixture for 15 min, followed by incubation in 2% warm Hellmanex III for 30 min. The substrates were then etched for 10 min in a piranha solution (3:1 H₂SO₄/H₂O₂). The coverslips were assembled on six-channel flow chambers (sticky-Slide VI 0.4, Ibidi). A 2:1 ratio of Poly-L-Lysine mPEG:Poly-L-Lysine Biotin (SuSoS Surface Technology) at a total concentration of 0.1 mg/mL in DIW incubated in each flow chamber for 30 min, then washed out with twice with 1 mL of 20 mM TBS (Corning, Cat no. 46-012-CM). 0.1 mg/mL Streptavidin in TBS (MilliPore Sigma, Cat no. 18-973-010MG) was then incubated for 30 min and washed similarly. HRas181-AviTag was then conjugated to the bound Streptavidin at 45 nM for 30 min before another wash with TBS followed by buffer exchange to HBS (40 mM HEPES, 150 mM NaCl, pH 7.4). Any nucleotides bound to RAS were stripped with 10 mM EDTA in HBS for 15 min before addition of 1 μM Atto488-GDP or non-fluorescent GppNp (Jena Bioscience) in HBS (40 mM HEPES, 150 mM NaCl, pH 7.4) supplemented with 5 mM MgCl₂ and overnight incubation at 4 °C. RAS labeled with Atto488-GDP and non-fluorescent GppNp were prepared in parallel. Fluorescent samples were used to determine RAS density and nonfluorescent samples were used for binding assays. The samples were rinsed with HBS (40 mM HEPES, 150 mM NaCl, pH 7.4) supplemented with 5 mM MgCl₂ immediately before use.

Microscopy

Imaging was performed using a Nikon Ti-2 microscope equipped with an Apo TIRF 100$\times$ oil immersion objective and an EMCCD camera (iXon Life 897, Oxford Instruments). For RAS density and RAF-mNG measurements, a 488 nm laser (Coherent), filtered through a laser cleanup filter (ZET488/10x, Chroma Inc.), was directed onto the sample using a dichroic mirror (zt488/640rpc, Chroma Inc.). The fluorescence was collected through two stacked emission filters (ZET488/640 m and ET535/70 m, Chroma Inc.). RAF-AZ647 was excited with a 640 nm laser (Coherent), filtered through a laser cleanup filter (LD01-640/8, Semrock) and directed by the same dichroic mirror (zt488/640rpc). The fluorescence was collected through two stacked emission filters (ZET488/640 m and ET706/95m, Chroma Inc.).

Ensemble adsorption and desorption

In ensemble experiments, SLBs were initially prepared with RAS loaded with the fluorescent nucleotide Atto488-GDP to determine surface density. Nucleotide exchange to GTP was facilitated by incubation of 100 nM catalytic domain of SOS and 100 µM GTP in HBS (40 mM HEPES, 150 mM NaCl, pH 7.4) supplemented with 5 mM MgCl₂ and 0.1 mg·mL⁻¹ casein at 23 °C for 6 min. Active RAS·GTP-functionalized membranes are nonfluorescent and suitable for subsequent binding assays with RAF fused with mNG. RAF-mNG was diluted as specified in the figure captions according to the protein concentration determined by absorbance at 280 nm. The binding measurements were performed in HBS (40 mM HEPES, 150 mM NaCl, pH 7.4) supplemented with 5 mM MgCl₂, 0.1 mg·mL⁻¹ casein, 1 µM GTP, and 1 mM BME at 23 °C. Upon reaching a plateau in the binding curve (typically in 5 min), desorption measurements were conducted in the same complete buffer including all supplemented components. Desorption was initiated by injecting 200 µL of the buffer, followed by continuous buffer flow at a rate of 1 mL·min⁻¹ using a syringe pump. K_d measurements and desorption assays on PEGylated surfaces were performed at 23 °C identically to SLB samples, except that nucleotide exchange by SOS was omitted; RAS was incubated overnight with GppNp. For K_d measurements, RBD-mNG and RBD-CRD-mNG were incubated at concentrations ranging from 5 to 750 nM. For desorption measurements, 200 nM was used. Background signals, $I_{bg}$, were obtained from the average intensities of the frames prior to RAF injection and were subtracted from the raw binding intensities. The intensity values were then divided by the maturation efficiency, $M_{Eff}$, of RAF-mNG proteins. The solution contribution, $I_{Soln}$, of mNG was then subtracted to calculate the binding/desorption net intensities, $I$, for analysis.

$$I=\left(\frac{I_{raw}-I_{bg}}{M_{Eff}}\right)-\left(I_{Soln}\right)$$ 1

The net intensity was converted to the surface density using the TIRF-FCS calibration curve.

Competitive FRET dissociation

Competitive desorption assays were performed using Förster Resonance Energy Transfer (FRET) to quantitatively assess the stability of the RAS:RAF complex. The donor Atto488-GppNp (Jena Bioscience) was loaded on RAS (RAS-Atto488-GppNp) and the acceptor AZ647(Vector Laboratories) was conjugated to either RBD or RBD-CRD of CRAF (RAF-AZ647). The assays were performed in HBS (40 mM HEPES, 150 mM NaCl, pH 7.4) supplemented with 5 mM MgCl₂, 0.1 mg·mL⁻¹ casein, 1 µM GTP, and 1 mM BME at 23 °C. To mitigate photobleaching, the buffer was further supplemented with enzymatic oxygen scavenging reagents: 0.32 mg·mL⁻¹ glucose oxidase from Aspergillus niger (Serva), 0.05 mg·mL⁻¹ catalase from bovine (Sigma-Aldrich), 2 mM Trolox (Sigma-Aldrich), and 20 mM glucose. The RAS:RAF complex was formed by incubating 200 nM of RBD-AZ647 or 20 nM RBD-CRD-AZ647 (according to the protein concentration determined by absorbance at 280 nm) with RAS tethered on SLBs for 5 min. 3 μM unlabeled RBD-CRD was subsequently introduced. FRET measurements were acquired and analyzed as described elsewhere⁶⁸. Three fluorescence signals were recorded: donor/acceptor (DA), acceptor/acceptor (AA), and donor/donor (DD). DD and AA are regular fluorescence signal acquired with the optical configurations described in the “Microscopy” section. The DA signal was acquired with a 488 nm laser excitation and emission configured for 640 nm channel. The measured DA signal, representing total acceptor fluorescence under donor excitation, is described by the following equation:

$${FL}_{DA}={Em}_{FRET}+\alpha \left({FL}_{DD}\right)+\beta \left({FL}_{AA}\right)$$ 2

Here, ${FL}_{DA}$ is the total acceptor emission under donor excitation, ${Em}_{FRET}$ is the sensitized emission due to FRET, $\alpha \left({FL}_{DD}\right)$ accounts for donor emission bleed-through into the acceptor channel, and $\beta \left({FL}_{AA}\right)$ accounts for direct acceptor excitation at the donor excitation wavelength. The FRET signal was calculated by subtracting the contribution of $\alpha \left({FL}_{DD}\right)$ and $\beta \left({FL}_{AA}\right)$ from ${FL}_{DA}$.

Single-molecule recruitment

Single-particle recruitment measurements of RAF-AZ647 were conducted on RAS · GTP-functionalized membranes in HBS (40 mM HEPES, 150 mM NaCl, pH 7.4) supplemented with 5 mM MgCl₂, 0.1 mg·mL⁻¹ casein, 1 µM GTP, and 1 mM BME at 23 °C RAF-AZ647 was diluted as specified in figure captions according to the AZ647 concentration. TIRF measurements were acquired using a 20 ms exposure time in stream mode, with the 640 nm laser power set to 5 mW. Single-molecule trajectories were detected and built using the TrackMate plugin in ImageJ⁶⁹. Cumulative binding events were then counted and plotted over the duration of the measurement using custom MATLAB scripts. The slope of this plot, normalized by RAS density, was used to determine k_on⁴¹.

Single-molecule tracking

Single-molecule diffusion experiments were conducted at 23 °C in HBS (40 mM HEPES, 150 mM NaCl, pH 7.4) supplemented with 5 mM MgCl₂, 0.1 mg·mL⁻¹ casein, 1 µM GTP, and 1 mM BME, as well as oxygen scavenging reagents including 0.32 mg·mL⁻¹ glucose oxidase from Aspergillus niger (Serva), 0.05 mg·mL-1 catalase from bovine (Sigma-Aldrich), 2 mM Trolox (Sigma-Aldrich), and 20 mM glucose. RAF-AZ647 was diluted as specified in figure captions according to the AZ647 concentration. RAF-AZ647 and RAS-Alexa647-GppNp diffusion measurements were performed on a RAS · GTP-functionalized SLBs with 10 ms time resolution. Single-molecule trajectories were detected and built using the TrackMate plugin in ImageJ⁶⁹. Trajectories with a mean speed of 0.5 pixels per 10 ms were accepted to exclude a minor fraction of immobile particles. The step size distribution was calculated using custom MATLAB scripts⁷⁰. A Brownian diffusion model, described below, was used to fit the step size distribution of each measurement:

$$p\left(r,t,D\right)=\frac{ar}{{2D}_{1}t}expexp\left(-\frac{r^2}{4D_{1}t}\right)+\frac{\left(1-\alpha \right)r}{2D_{2}t}exp\left(-\frac{r^2}{4D_{2}t}\right)$$ 3

Diffusion coefficients, D₁ and D₂, and relative population, for the fast species, α were calculated from the corresponding fitting. Step size is represented by r, at an experimental time interval, t.

Fluorescent correlation spectroscopy

Measurements were performed on a custom-built FCS setup integrated with a Nikon Ti-2 microscope. A SuperK Evo HP white-light laser source (NKT Photonics) was reflected by a dichroic mirror (zt488rdc, Chroma Inc.) and filtered through a cleanup filter (LL01-488, Semrock) to select a 488 nm wavelength. The beam was coupled into a polarization-maintaining single-mode fiber (PM-S405-XP, Thorlabs) via a fiber coupler (PAF2-A4A, Thorlabs). The beam collimated via a reflective collimator (RC08FC-P01, Thorlabs), directed to the back port of the microscope and then to the objective by a quad-band dichroic filter cube (ZT405/488/561/640rpc, Chroma Inc.). The excitation beam was focused by Apo TIRF 100$\times$ oil immersion objective. Fluorescence emission from the sample was spatially filtered through a 50 µm pinhole, collimated by an achromatic doublet lens (AC254-125-A, Thorlabs), reflected by a dichroic mirror (550lpxr, Chroma Inc.), and further filtered through a bandpass filter (ET520/40 m, Chroma Inc.). It was then refocused by another achromatic doublet lens (AC254-75-A, Thorlabs) onto an avalanche photodiode (SPCM-AQRH-16, Excelitas Technologies) connected to a time-correlated single-photon counting module (PicoHarp 300, PicoQuant). Data acquisition was performed using PicoHarp software (PicoQuant) and analyzed in Prism (GraphPad). The autocorrelation function, $G\left(\tau \right)$, was fitted to a two-dimensional Gaussian diffusion equation to calculate surface density of RAF and RAS.

$$G\left(\tau \right)=\frac{1}{N}\left(\frac{1}{1+\tau /{\tau }_d}\right)$$ 4

where N is the number of particles within the focal area, $\tau$ is the time delay, and ${\tau }_d$ is the correlation time. The focal radius was determined to be 0.169 µm and was used to calculate the protein density.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.

Author contributions

A.J.S. expressed and purified all proteins, with assistance from K.T. and A.C. A.J.S., K.T., and Y.K.L. performed and analyzed ensemble and single-molecule binding assays, with assistance from J.G., A.C., and A.M. J.G. performed FCS experiments. Y.K.L. designed the experiments and directed the project. A.J.S. and Y.K.L. drafted the manuscript with input from all authors.

Peer review

Peer review information

Nature Communications thanks Joseph Falke and the other, anonymous, reviewer(s) for their contribution to the peer review of this work. A peer review file is available.

Data availability

The authors declare that the data supporting the findings of this study are available within the paper and its Supplementary Information files. The raw image data are available from the corresponding author upon request. These image data are not deposited in a public database because of their large file sizes. Source Data are provided with this paper. The structures of the CRAF/MEK1/14-3-3 complex in the autoinhibited conformation and the CRAF RBD-CRD/KRAS complex were obtained from the Protein Data Bank under accession codes 9MMP [10.2210/pdb9mmp/pdb] and 6XI7 [10.2210/pdb6xi7/pdb], respectively. Source data are provided with this paper.

Code availability

The custom MATLAB script used for TrackMate data processing, association kinetics, and step size distribution analysis is available on GitHub at https://github.com/youngkwanglee-mem/SPT_Kon_StepSize. The specific version of the code used for the analysis in this study has been deposited in Zenodo with the 10.5281/zenodo.18126517⁷⁰.

Competing interests

The authors declare no competing interests.

Supplementary information

The online version contains supplementary material available at 10.1038/s41467-026-69437-6.