Anionic Lipids Impact RAS-Binding Site Accessibility and Membrane Binding Affinity of CRAF RBD-CRD

Abstract

CRAF activation requires binding to membrane-anchored and active GTP-bound RAS. Whereas its RAS-binding domain (RBD) contains the main binding interface to the RAS G domain, its cysteine-rich domain (CRD) is responsible for association to anionic lipid-rich membranes. Both RAF domains are connected by a short linker, and it remains unclear if the two domains act independently or if one domain can impact the function of the other. Here, we used a combination of coarse-grained and all-atom molecular dynamics simulations of a CRAF RBD-CRD construct to investigate the dynamics of the RBD when it is tethered to CRD that is anchored to a POPC:POPS model membrane. First, we show that the RBD positioning is very dynamic with a preferential localization near the membrane surface. Next, we show that membrane-localized RBD has its RAS-binding interface mostly inaccessible because of its proximity to the membrane. Several positively charged residues in this interface were identified from simulations as important for driving RBD association to the membrane. Surface plasmon resonance (SPR) measurements confirmed that mutations of these RBD residues reduced the liposome partitioning of RBD-CRD. Last, simulations indicated that the presence of RBD near the membrane led to a local enrichment of anionic lipids that could potentially enhance the membrane affinity of the entire RBD-CRD construct. This was supported by SPR measurements that showed stronger liposome partitioning of RBD-CRD relative to CRD alone. These findings thus suggest that the RBD and CRD have synergistic effects on their membrane dynamics, with CRD bringing RBD closer to the membrane that impacts its accessibility to RAS and with RBD causing local anionic lipid enrichment that enhances the overall affinity between the membrane and RBD-CRD. These mechanisms have potential implications on the order of events of the interactions between RAS and CRAF at the membrane.

Significance

The MAPK signaling pathway transmits external cues detected by membrane receptor proteins into the interior of the cell to promote its growth, proliferation, and survival. Although significant progress has been made in understanding the molecular processes underlying this signal transduction pathway, many details remain to be elucidated such as how RAS and RAF encounter one another at the membrane and if their interactions are driven by particular lipid types. We used a combined computational and experimental approach to investigate the synergistic membrane dynamics of CRAF RBD-CRD. We show that membrane-anchored CRD causes the RBD to preferentially localize near the membrane surface, and in turn, membrane-associated RBD locally enriches the concentration of anionic lipids, leading to enhanced membrane affinity of RBD-CRD.

Introduction

Activation of the MAPK (RAS/RAF/MEK/ERK) pathway involves the association of RAF kinase with active RAS at the membrane, which then triggers a cascade of phosphorylation events, leading to enhanced expression of genes controlling cellular growth, proliferation, and survival (1,2). In humans, there are three RAF isoforms (ARAF, BRAF, and CRAF) (3, 4, 5) that all comprise the following domains (Fig. 1 A): 1) a variable-sized and disordered N-terminal region that has been implicated in the formation of the RAS/RAF complex (6), 2) a RAS-binding domain (RBD) that binds with high (nanomolar) affinity to RAS (7,8), 3) a cysteine-rich domain (CRD) that binds with weaker (micromolar) affinity to RAS (9, 10, 11), 4) a long disordered linker called the hinge region containing multiple phosphorylation sites that bind the RAF-modulating 14-3-3 proteins (12), and 5) a kinase domain that initiates the MAPK phosphorylation cascade (13,14). Although significant progress has been made in understanding the roles of the different domains of RAF during signaling, many molecular details still remain unclear, such as those underlying the membrane dynamics of RAS and RAF that lead to their complex formation. The elucidation of these mechanisms could contribute to the development of more effective therapies against cancers that are driven by mutations in proteins of the MAPK pathway (15, 16, 17).

Figure 1.

Modeling of membrane-anchored RBD-CRD. (A) Domain structure of CRAF. (B) One of the sampled conformations of CRAF RBD-CRD in the simulations. All simulations were started with the CRD (orange cartoons) anchored to a membrane patch but with the RBD (green cartoons) initially positioned away from the the membrane. The membrane patch is shown here using atom-based coloring (blue for carbon, red for oxygen, orange for phosphorus, darker blue for nitrogen, and light gray for hydrogen). The conformation shown here corresponds to pose GH5 in (24), in which the RBD has approached the membrane surface while keeping the main RAS-interacting β-strand (red cartoons) accessible for binding to the RAS G domain. The two zinc ions coordinated with the CRD are shown as gray spheres.

The membrane association of RAF is primarily driven by the CRD, which adopts a similar structural fold (18,19) as the protein kinase C C1 domains that anchor into membranes (20, 21, 22). Liposome-binding assays (23) and, recently, molecular dynamics (MD) simulations and NMR measurements (24, 25, 26) have shown that the CRD can also anchor into membranes, with a particular preference for phosphatidylserine-containing (i.e., anionic) membranes. The CRD also plays a minor role in binding to RAS (9, 10, 11); however, the main RAS-binding site (RBS) of RAF is found on the RBD, which in the CRAF isoform comprises a β-strand at K65-N71 and α-helix at K84-R89 (27,28). The RBD adopts a ubiquitin α/β roll superfold (29), and to date, around 120 human proteins are annotated in the SMART (Simple Modular Architecture Research Tool) database (30,31) as containing an RBD or the closely related RAS association (RA) domain. Recent data indicate that the ARAF RBD can directly associate with nanodisks containing a mixture of zwitterionic and anionic lipids when in complex with the G domain of KRAS4b (32), thereby suggesting that the RBD may play dual roles like the CRD. However, the biological significance of the observed RBD-membrane association is not clear for two reasons. First, this association has only been observed when the RBD is in complex with KRAS4b, and so this may simply be due to distinct orientations adopted by the G domain at the membrane (33, 34, 35, 36, 37, 38). Second, the CRD was not used in these measurements, and the short linker between the RBD and CRD raises the issue of whether or not both RAF domains can be treated as completely independent from each other.

A recent high-resolution cryo-electron microscopy structure of the autoinhibited complex of BRAF with MEK1 and 14-3-3 (39) showed that the two hydrophobic and membrane-anchoring loops of the CRD are covered by 14-3-3 in this complex while it remains in solution. BRAF RBD was not resolved in this complex; the potentially unrestrained orientation of the RBD in the autoinhibited complex could potentially enable its interaction with KRAS4b and lead to membrane localization of RAF. However, there is evidence suggesting that this membrane localization can be RAF isoform dependent. First, small interfering RNA (siRNA) knockdowns of the membrane proteins prohibitin or CNK1 were found to block the interactions of CRAF with active RAS (40, 41, 42, 43), indicating that the presence of “free-floating” and accessible CRAF RBD along with active RAS is not sufficient to drive the membrane localization of CRAF. Consistent with this, reduced activation of MEK and ERK was observed in cells with siRNA knockdown of prohibitin (44). Second, these results are CRAF specific as the loss of prohibitin only impacts the phosphorylation and activation of CRAF but not of BRAF and ARAF (44). As the RBD and CRD both play a role in the membrane localization of RAF (45), consideration of the two linked domains together should serve as a useful construct to explore possible functional distinctions among the different RAF isoforms (46).

Here, we investigate the dynamics of the RAF RBD-CRD construct in a membrane environment. We focus on RBD-CRD from CRAF because this isoform is recognized for its importance as a hub in RAS-mediated MAPK signaling (47,48). In particular, can the RBD localize close to the membrane when the nearby CRD is already membrane embedded, and if so, what orientations relative to the membrane surface does the RBD adopt? This has important implications as the accessibility of the RBS can be modulated by proximity or direct interaction with the membrane surface. In turn, does membrane-localized RBD have an impact on the membrane affinity of RBD-CRD, and do specific lipid types play a role here? This also has important implications as an enhanced membrane affinity would allow for an increased encounter rate between active RAS and CRAF at the membrane. We address these questions through insights gained from coarse-grained (CG) and all-atom (AA) MD simulations followed by validation with surface plasmon resonance (SPR) measurements. Using this combined computational and experimental approach, we were able to explore in more detail the synergistic membrane dynamics of CRAF RBD-CRD and the role of anionic lipids in these dynamics.

Materials and Methods

Models and system setup for membrane-anchored CRAF CRD

A representative snapshot from the single basin observed from previous AA simulations (24) of membrane-anchored BRAF CRD was used here to build a corresponding model of membrane-anchored CRAF CRD. This was done via optimal structural alignment of backbone atoms between the BRAF CRD snapshot with the solution NMR structure of CRAF CRD (Protein Data Bank, PDB: 1FAQ) (18) using PyMOL (49). The CRD is anchored to the membrane in this snapshot via its two hydrophobic loops (Fig. 1 B). The bilayer from the snapshot was replaced with a membrane patch that was generated using the Membrane Builder module of CHARMM-GUI (50,51) and that contains 100 lipids per leaflet in a 70:30 ratio of 1-palmitoyl-2-oleoyl-sn-glycero-3-phosphocholine (POPC) to 1-palmitoyl-2-oleoyl-sn-glycero-3-phospho-L-serine (POPS). An AA representation of this system was generated using the CHARMM36 force field (52,53). TIP3P water molecules (54) were added to fill the simulation box. Monovalent K+ and Cl− ions were added to neutralize the system charge and to reach a physiological ionic strength of 150 mM. This system was used for running ten 2-μs AA simulations (see AA Simulations below).

This AA model was transformed into CG beads using parameters based on the MARTINI 2.2 force field (55). An elastic network was applied to the entire protein backbone to preserve its secondary structure (56). To be more consistent with physical properties computed from atomistic CHARMM36 runs, the CG MARTINI parameters for POPC and POPS were reparametrized as described in (24). The system length along the membrane-normal z coordinate was set at 24 nm to assure that the minimal image convention is always satisfied for the protein, and it does not see its periodic image. CG water beads were used to solvate the system, along with Na⁺ and Cl− ions to neutralize the excess charges and to mimic the physiological ionic strength of 150 mM. Another CG system for membrane-anchored CRAF CRD was also constructed with a membrane patch containing 1600 lipids per leaflet in a 70:30 ratio of POPC:POPS. This larger membrane patch was built using the insane script (57). Ten 25-μs CG simulations were run for both systems (see CG Simulations below).

Models and system setup for membrane-anchored CRAF RBD-CRD

The initial structure for CRAF RBD was taken from its cocrystal structure with HRAS G domain (PDB: 4G0N) (28). This RBD structure was added to the earlier membrane-anchored CRD system such that it was initially positioned with its center of mass (COM) around 3 nm away along the z axis (or membrane-normal axis) from the membrane surface and such that its C-terminus can be connected by the adjacent short peptide linker to the CRD N-terminus. The z coordinate of the membrane surface was calculated as the average z coordinate for the N atoms from the POPC and POPS headgroups in the leaflet where CRD is embedded. Three starting conformations were generated by rotating the linker backbone such that the RBD “twist” relative to membrane-anchored CRD differs by around 120° between different starting conformations. The rest of the AA system setup for membrane-anchored RBD-CRD was similar as described earlier for membrane-anchored CRD. Ten 2-μs AA runs were performed for each starting conformation, yielding a total of 30 AA simulations (see AA Simulations below).

Conversion of this AA model of membrane-anchored RBD-CRD to a CG representation and the rest of the CG system setup were similar as described earlier for membrane-anchored CRD. Two sets of 25-μs CG simulations were then run; one set used 100 lipids per leaflet, whereas the other set used 1600 lipids per leaflet (see CG Simulations below).

Models and system setup for solvated complex between CRAF RBD and KRAS

The cocrystal structure of CRAF RBD with HRAS G domain (PDB: 4G0N) (28) was used to construct a homology model of CRAF RBD bound to the KRAS4b G domain. In silico mutagenesis was done using PyMOL (49) to change from the HRAS sequence to that for KRAS4b. An AA representation of this system was generated using the CHARMM36 force field (52,53). TIP3P water molecules (54) were added to fill a rhombic dodecahedral box around the protein complex, with a minimal distance of 1.2 nm from any protein atom to any edge of the simulation unit cell. Monovalent K+ and Cl− ions were added to neutralize the system charge and to reach a physiological ionic strength of 150 mM. Ten 1-μs AA simulations were the performed for this system (see AA Simulations below).

CG simulations

The GROMACS MD engine (version 5.1.3) (58) was used in combination with the MARTINI 2.2 force field (55) for running all CG simulations. We followed a recent update in CG parameters setup for performing these runs (59). No force field constraints were added to keep the CRD membrane anchored. Reaction-field electrostatics (60) were used with a Coulomb cutoff of 1.1 nm and dielectric constants of 15 or 0 within or beyond this cutoff, respectively. A cutoff of 1.1 nm was also used for calculating Lennard Jones interactions, using a scheme that shifts the van der Waals (vdW) potential to zero at this cutoff. Constant temperature was maintained at 310 K via separate coupling of the solvent and membrane/protein components to a velocity rescaling thermostat (61) with a relaxation time of 1.0 ps. Semi-isotropic pressure coupling was set for each system at 1 bar using a Berendsen barostat (62) with a relaxation time of 12.0 ps. Simulations used a 30-fs time step.

AA simulations

The AMBER MD engine (version 18) that has been GPU-optimized for simulating explicit solvent systems (63,64) was used for running all atomistic simulations. No force field constraints were added to keep the CRD membrane-anchored. Particle mesh Ewald (PME) electrostatics (65) were used along with Coulomb and Lennard Jones cutoffs of 1.2 nm and potential switching at 1.0 nm. Constant temperature was maintained at 310 K via Langevin dynamics (66) with a collision frequency of 1.0 ps⁻¹. Semi-isotropic pressure coupling was set for each system at 1 bar using a Monte Carlo barostat (67) with a relaxation time of 4.0 ps. Bonds containing hydrogen atoms were constrained using the SHAKE algorithm (68). A hydrogen mass repartitioning approach (69) allowed the use of a 4-fs time step.

Simulation analyses

All analyses of the CG and AA simulations were performed using a combination of GROMACS tools (58), CPPTRAJ (70), VMD (71), and in-house scripts. To measure the RBS accessibility to KRAS4b in the CG simulations, the accessible surface area (ASA) of the RBS was first calculated for each frame in the presence of the membrane; the ratio of this ASA value to that obtained with the membrane removed (to mimic in solution conditions) gives the relative ASA of the RBS for that frame. This ratio provides an estimate of how much the membrane covers the RBS binding surface at each frame. ASA terms were calculated using a spherical probe with a radius equal to the smallest component of the radius of gyration for solvated KRAS4b G domain (residues 1–166), which is ∼1 nm. To measure the RBS accessibility to KRAS4b in the AA simulations, a homology model of the complex between KRAS4b G domain and CRAF RBD was structurally aligned at the RBD Cα atoms to RBD from each AA simulation frame. The G domain atoms that overlap with the membrane after this fitting were identified, and the vdW volume of these membrane-overlapping atoms from the G domain was computed. The ratio of this volume with the total G domain vdW volume then gives the percentage of potential volume overlap between the G domain and the membrane for each AA frame. Large volume overlap values signify that the RBS is covered by the membrane to occlude RAS binding, whereas small values indicate that the RBS is accessible. We chose a cutoff value of 5% as the upper limit for classifying volume overlaps as small because our tests showed that energy minimization can remove steric clashes between the G domain and membrane up to this value. Calculation of the vdW volume for different groups of atoms was done using ProteinVolume (72).

Cloning, protein expression, and protein purification

Gateway Entry clones for CRAF constructs were synthesized using optimization for Escherichia coli or insect cells (ATUM, Newark, CA) that incorporate an upstream tobacco etch virus protease cleavage site (ENLYFQ/G) followed by the appropriate CRAF (human) sequences: CRAF RBD-CRD (52–192) wild-type (WT) and mutants (mts), CRAF RBD (52–131), and CRAF CRD (136–188). Sequence-validated Entry clones were subcloned into E. coli vector pDest-566 (73) or baculovirus vector pDest-636 (74). The Gateway Destination vectors incorporate His6 and maltose binding protein (MBP) tags to produce final expression clones of the form His6-MBP-tev-POI, where “tev” is the recognition site for tobacco etch virus protease and POI is the protein of interest. CRAF CRD was expressed and purified using protocols described previously (24). CRAF RBD-CRD constructs were expressed and purified as described in (45). CRAF RBD was expressed and purified following the protocols outlined in (75) for proteins of the format His6-MBP-tev-POI.

Liposome preparation

POPC and POPS were purchased as stock solutions from Avanti Polar Lipids (Alabaster, AL). Desired amounts of the POPC and POPS were aliquoted to make 70:30 and 50:50 POPC/POPS liposomes. After evaporation, the dried lipid films were reconstituted in buffer composed of 20 mM HEPES (pH 7.2) and 100 mM NaCl. The reconstituted samples were then subjected to several freeze/thaw cycles and sonicated for 1 h in a Branscon Ultrasonics Bath (Thermo Fisher Scientific, Waltham, MA). The samples were then extruded using a 0.1-μM filter membrane. Liposomes were then spun for 30 min at 20,000 × g to remove any large aggregates.

SPR measurements and data analysis

SPR binding experiments were collected on a Biacore T200 Instrument (GE Healthcare, Chicago, IL). Binding of RBD, CRD, RBD-CRD (WT), and RBD-CRD (mts) to both 70:30 and 50:50 POPC:POPS were measured as follows: the Series 5 sensor chip L1 (GE Healthcare) surface was first activated with two injections of 20 mM CHAPS (Sigma, St. Louis, MO). Five millimolar of either 70:30 or 50:50 liposomes were captured on flow cells 2 and 4, respectively. Flow cells 1 and 3 were used for referencing purposes. After capturing the liposomes onto the L1 Chip, a series of buffer injections were performed in running buffer composed of 20 mM HEPES (pH 7.2) and 250 mM NaCl to achieve a stable baseline. RBD, CRD, RBD-CRD (WT), and RBD-CRD (mts) were diluted in running buffer from 50 to 0.1 μM as a 2:1 dilution series and consecutively injected onto the captured liposomes. The L1 was regenerated using two injections of 20 mM CHAPS. The SPR data were analyzed following methodologies developed in (45,76). Specifically, equilibrium partition coefficients (K_p) were extracted for RBD-CRD (WT) and RBD-CRD (mts) using Eq. 1:

$$\frac{R{U}_S}{R{U}_L}=\frac{{\gamma }_L{K}_p\frac{M_S}{M_L}{\left[S\right]}_w}{1+\sigma {\gamma }_L{K}_p{\left[S\right]}_w},$$ (1)

where RU_S gives the solute membrane association response units, RU_L is the total lipid deposition response, γ_L is the lipid molar volume, M_S and M_L give the molecular mass of, respectively, the solute and lipid, [S]_w is the solute concentration of the continuously injected aqueous phase, and K_p is the equilibrium partition coefficient. Under conditions of low membrane saturation, such that σγ_LK_p[S]_w ≪ 1, Eq. 1 can be simplified to the following Eq. 2 for a linear partition relationship:

$$\frac{R{U}_S}{R{U}_L}={\gamma }_L{K}_p\frac{M_S}{M_L}{\left[S\right]}_w,$$ (2)

which was used here for extracting the equilibrium partition coefficients (K_p) for the RBD and CRD constructs. The Gibbs free energy of the protein transfer from the aqueous phase to the lipid phase can then be calculated from Eq. 3:

$$\Delta G_p=-RT\ln K_p,$$ (3)

where R is the universal gas constant, and T is the absolute temperature at which the SPR measurements were performed (298 K). Last, the fraction of membrane-associated protein per liposome for each of the protein constructs can be calculated using Eq. 4:

$$f_b=\frac{K_p\left[L\right]}{55.6+K_p\left[L\right]},$$ (4)

where [L] is half of the total lipid concentration at the surface of the chip because the protein constructs here can only partition to the outer leaflet of the liposomes, and 55.6 gives the molar concentration of water.

Results and Discussion

RBD alternates between two conformational basins when tethered to membrane-anchored CRD

To observe the long timescale membrane dynamics of the RBD imposed by its short linker to the CRD, we performed ten 25-μs CG simulations using the MARTINI 2.2 force field (55) for ten starting conformations in which the CRD is anchored to the membrane while the RBD is initially positioned away from the membrane (i.e., in solution). The membrane patch in these CG simulations contained 1120 POPC and 480 POPS lipids per leaflet to attain a 70:30 ratio (see Materials and Methods for a detailed description of modeling and simulation setup). We first monitored the distance along the z axis between the RBD COM and membrane surface to observe changes in RBD localization relative to the membrane. The membrane surface was defined here as the average z coordinate of the headgroup beads for the POPC and POPS lipids in the proximal leaflet (i.e., the leaflet where the CRD was anchored). We observed in these CG simulations that large fluctuations occurred in the distance between the RBD COM and the membrane surface (Fig. S1). The RBD exhibited multiple association/dissociation events with the membrane surface and, in several instances, remained in close proximity to this surface for an extended duration of time (Fig. S1, red arrow). It is interesting that the RBD was observed to alternate between membrane-associated and membrane-dissociated conformations as the RBS is more likely to be accessible when the RBD is away from the membrane.

To determine the preferred conformations of the RBD from the CG ensemble, we defined two reaction coordinates for describing RBD orientations relative to the membrane based on the RAS-interacting β-strand (residues K65–N71) that must be exposed to bind RAS: 1) the distance along the z axis between the backbone COM of this β-strand and the membrane COM and 2) the angle between a vector along this β-strand and the z axis. These coordinates provide a measure of the distance between the RBS and the membrane and a measure of the tilting of the RBS relative to the membrane, respectively. A free energy surface map for both of these coordinates showed two dominant basins (Fig. 2 A). The first basin has a COM-COM distance between 2.5 and 4 nm and an orientation angle between 50 and 110° such that the RAS-interacting β-strand is localized near and oriented almost parallel to the membrane surface (Fig. 2 B, left panel). The second basin has an average COM-COM distance of around 5 nm and does not bring the RBS in close proximity to the membrane surface (Fig. 2 B, right panel). The broad distributions along both axes of this energy landscape clearly suggest a very dynamic behavior of the RBD when it is connected to membrane-anchored CRD.

Figure 2.

Membrane orientations of the RBD from CG simulations of CRAF RBD-CRD. (A) Free energy surface map for RBD orientations relative to the membrane surface from the CG RBD-CRD simulations. The reaction coordinates along the two axes are described in the text. Two major basins can be seen from this map. (B) Representative snapshots of CG RBD-CRD configurations for the two basins identified in (A). Both CG frames were backmapped to show the secondary structure of RBD-CRD. The RBS of the RBD, comprising a β-strand at residues K65–N71 and the adjacent α-helix at residues K84–R89, are colored red. The left and right panels show the RBS located near the membrane surface or away from it, respectively. Also highlighted in the right panel is a CRAF loop homologous to an ARAF loop that was previously shown to associate with membranes in the presence of KRAS (32). (C) The relative ASA of the RBS in each CG frame is projected onto the surface map of both reaction coordinates from (A). (D) Percentage of total CG simulations frames that fall under particular ranges of relative ASA for the RBS.

To assess how the membrane orientations of the RBD from the CG simulations impact RBS accessibility for forming the initial encounter complex between RBD-CRD and KRAS4b at the membrane, we calculated the relative ASA of the RBS in all the configurations from the CG simulations (see Simulation Analyses in Materials and Methods for calculation details). The RBS relative ASA over the whole CG ensemble of 250 μs is shown in Fig. 2 C after projecting onto the map for the two RBS reaction coordinates. Around a third of the conformations (∼35%) exhibited relative ASA values greater than 60% (Fig. 2 D), which is indicative of sufficient RBS exposure for binding RAS. These conformations (yellow- to light green-colored region in Fig. 2 C) comprise the second basin in Fig. 2 A. We note that because all CG frames were considered in this analysis, the second basin includes frames in which the RBS is accessible because of the RBD being away from the membrane. The remaining conformations (∼65%) exhibited lower relative ASA values for the RBS (dark green- to violet-colored region in Fig. 2 C) and comprise the first basin in Fig. 2 A. Notably, conformations with COM-COM distances (x axis in Fig. 2 C) less than 3.5 nm almost exclusively belong to the first basin where the membrane can reduce RBS accessibility and thus interfere with RAS-RAF association.

Membrane-localized RBD prefers orientations that provide limited access to RAS binding

To get a finer molecular recognition of the RBD dynamics in membrane-localized RBD-CRD, we performed AA simulations of the membrane-anchored RBD-CRD construct using the CHARMM36 force field (52,53). A smaller membrane patch was used with 100 lipids per leaflet while still maintaining a 70:30 ratio of POPC/POPS to reduce the computational cost of these AA simulations. The RBD in all these simulations approached the membrane surface with an average approach time of around 0.6 μs; all the simulations except one showed the RBD remaining localized near the membrane for the remainder of the trajectory after approaching the membrane. We then filtered for only those AA frames in which the RBD was near the membrane (i.e., z distance ≤2 nm between RBD COM and membrane surface) as we were interested here in further analyzing the orientations of the RBD when it is membrane localized. Because the average approach time was around 0.6 μs over all the 2-μs AA trajectories, this means that around 30% (100% × 0.6/2.0) of the total AA frames were filtered out for the subsequent analyses. A free energy surface map of the postfiltered frames showed a clear and dominant basin with an average COM-COM distance of around 2.5 nm and an average angle of around 90° (Fig. 3 A). The RAS-interacting β-strand was localized near the membrane and oriented mostly parallel to the membrane surface in this basin. A second less-populated basin was also observed that had the RAS-interacting β-strand positioned farther from the membrane with an average COM-COM distance of around 4.5 nm. These results are consistent with the CG simulations and indicate that the RBD adopts preferred orientations when it is near the membrane.

Figure 3.

Limited accessibility of the RBS on the RBD from AA simulations of CRAF RBD-CRD. (A) Free energy surface map for RBD orientations relative to the membrane surface from the AA RBD-CRD simulations, using the same two reaction coordinates as in Fig. 2A. The AA frames were filtered to include only those showing a z distance of 2 nm or less between the RBD COM and the membrane surface. Black vertical dashed line at x = 3.5 nm separates the two observed basins. (B) Normalized probability distribution for percentage volume overlap between KRAS4b G domain and the membrane, based on structural alignment of a homology model of the KRAS4b G domain-CRAF RBD complex and the AA RBD-CRD simulations. The x axis is binned with widths of 5%. A majority of the frames show volume overlaps of greater than 5% (orange columns); however, a separate peak occurs for small volume overlaps between 0 and 5% (blue column). (C) Normalized probability distributions of the z distance between the RAS-interacting β-strand COM and the membrane COM for AA frames showing low (<5%) volume overlap of RAS with the membrane (blue columns) and for AA frames showing high (>5%) volume overlap (orange columns). Black dashed line indicates that a z distance cutoff value of 3.5 nm can effectively distinguish between conformations that have the RBS accessible (low overlap) or inaccessible (high overlap). Error bars in (B) and (C) give SEM over 30 AA simulations.

We next evaluated how the membrane orientations of the RBD from the AA simulations impact the accessibility of the RBS for forming the initial encounter complex between RBD-CRD and KRAS4b at the membrane through calculations of the potential volume overlap between the membrane and RBD-bound KRAS4b G domain (see Simulation Analyses in Materials and Methods for calculation details). A majority of the AA frames with RBD near the membrane showed large volume overlaps between the RAS G domain and membrane (orange columns in Fig. 3 B). However, a distinct peak occurred for small volume overlaps, with around 7.4% of the frames comprising this peak (blue column in Fig. 3 B). This indicates that even when localized near the membrane, the RBD can adopt orientations that are conducive to RAS binding. Because around 70% of the total AA frames were not filtered out from earlier, we note that the frames with membrane-associated RBD and accessible RBS actually comprised around 5.2% (0.70 × 7.4%) of the total AA frames. Adding this to the percentage of filtered out frames (which have RBD away from the membrane and thus accessible RBS) from earlier indicates that around 35% of the total AA frames have accessible RBS. This is consistent with the value obtained earlier using the ASA-based analysis over all CG frames (Fig. 2 D). Despite this similarity, we note that both basins are wider and deeper in the CG free energy surface map (Fig. 2 A vs. Fig. 3 A) because of the more extensive conformational sampling in the CG simulations. In addition, the 25-μs CG runs showed multiple association/dissociation events by the RBD with the membrane surface per trajectory (Fig. S1), whereas the 2-μs AA runs were less dynamic in terms of showing only a single membrane association event per trajectory.

We then examined if the two reaction coordinates for defining RBD orientations (x and y axes in Figs. 2 A and 3 A) can effectively distinguish between AA orientations showing small or large volume overlaps. For the reaction coordinate based on the z distance between the RAS-interacting β-strand COM and the membrane COM, we found that a cutoff value of 3.5 nm can effectively separate frames with small volume overlaps from those with large volume overlaps (Fig. 3 C). Applying this cutoff to the free energy surface map in Fig. 3 A indicated that the dominant AA basin contains only frames that would lead to large volume overlaps between the G domain and membrane, whereas the less-populated basin comprises orientations that allow the binding of RAS to membrane-localized RBD. In contrast, the reaction coordinate based on the angle between a vector along the RAS-interacting β-strand and the z axis did not indicate an angle value that could effectively separate between small overlap or large overlap frames (Fig. S2). Although a volume overlap cutoff of 5% was used here to separate AA frames with small and large volume overlaps (see Simulation Analyses in Materials and Methods), similar results were obtained for volume overlap thresholds up to 20% (Fig. S3), indicating that this analysis is applicable to membranes that are more dynamic than the relatively small (100 lipids per leaflet) membrane patches used here in the AA simulations.

Basic residues from the RBS of the RBD interact with anionic lipids

We next wanted to determine what drives the RBD toward the membrane surface in these simulations. A closer look at the conformations falling into either of the two basins observed in the CG simulations provides a clearer picture of the dynamics of the RBD as related to the RBS. Conformations from the first basin (Fig. 2 B, left panel) highlight a lysine (K65) and two arginines (R59 and R67) from the RBS that are contacting POPS lipids, leading to occlusion of the RBS by the membrane. Conformations from the second basin (Fig. 2 B, right panel), on the other hand, show POPS contact by two lysines (K106 and K109) and one arginine (R111). These latter residues are part of a CRAF RBD loop whose homologous loop in ARAF was previously shown to contact the membrane in the presence of KRAS (32), which is consistent with the observed exposure of the RBS in this basin. The conformations from both basins therefore involve contacts with the anionic POPS headgroups by positively charged residues from two separate surfaces of the RBD (Fig. S4).

We then looked at what residues from the RBD are contacting POPS lipids in the AA simulations. The residues showing the largest proportion of POPS interactions involved four arginines (R59, R67, R73, and R89) from the RBS (Fig. 4 A). Note that two of these arginines (R59 and R67) were also identified in the CG simulations for the RBS-inaccessible basin (Fig. 2 B, left panel); in lieu of K65 from the CG runs though, the AA simulations showed more frequent POPS contacts for two other arginines (R73 and R89). Per-residue interaction energies (Coulombic plus Lennard Jones) with POPS lipids favored these four arginines (Fig. S5). Interestingly, in the RAS/RBD co-crystal structure (PDB: 4G0N) (28) as well as in AA simulations based on this structure, these four arginines were also involved in key interactions at the protein-protein binding interface (Fig. S6). It is worth noting that the RBS and its adjacent regions comprise the most electrostatically positive surface of the RBD (Fig. S7).

Figure 4.

Arginines from the RBS on the RBD are important for membrane association. (A) RBD per-residue profile for the proportion of AA frames in which any heavy atom from that residue is within 0.45 nm of a POPS headgroup. Data were taken from the AA RBD-CRD simulations. Yellow stars indicate four arginines (R59, R67, R73, and R89) that have the highest preference for being close to the membrane. Error bars give SEM over 30 AA simulations. (B) Representative SPR measurements of 70:30 POPC:POPS liposome association for WT (blue), R59A/R73A (green), R67A/R89A (pink), and R59A/R67A/R73A/R89A RBD-CRD (yellow) constructs. Protein concentrations between 50 and 1 μM were tested for each of the constructs. The solute membrane association response units (RU_S) from these measurements have been normalized by the total lipid deposition response (RU_L). (C) Steady-state binding isotherms calculated from the SPR sensorgrams in (B) as a function of protein concentration. (D) Comparison of the equilibrium partition coefficients (K_p) for 70:30 POPC/POPS liposome association. Error bars give the SD over two experiments. Asterisks indicate p < 0.05 based on two-sided Student’s t-test when comparing with WT levels (blue) or quadruple mutant levels (yellow).

To test if these four RBS arginines are critical for membrane association of RBD-CRD, SPR assays were performed to measure binding to POPC:POPS liposomes. As these four arginines are located on the same face of the RBD surface, we were concerned that single R-to-A mutations may be compensated by any of the other RBS arginines. We thus tested two double R-to-A mutants and a quadruple R-to-A mutant covering the four arginines. To select which double mutants to test, we took every possible pair of these four arginines and calculated the fraction of AA frames in which both arginines are simultaneously in contact with POPS headgroups. The R67-R89 pair showed the highest such fraction (Fig. S8), so we chose this pair as one of the double R-to-A mutants to reduce the likelihood of compensation taking place. The other two arginines, R59 and R73, were then paired together as the other double R-to-A mutant to prevent redundancy in both double mutants, even though this pair has the lowest probability of both arginines being close to the membrane (Fig. S8).

SPR measurements were then used to assess the binding of WT, the two double mutants (R67A/R89A and R59A/R73A), and the quadruple mutant (R59A/R67A/R73A/R89A) constructs of RBD-CRD to liposomes composed of 70:30 POPC:POPS. Comparison of the SPR sensorgrams clearly showed lower levels of binding by the two double mutants (R67A/R89A and R59A/R73A) to the liposomes (Fig. 4 B) relative to WT. As indicated by the steady-state binding isotherms, there is no evident difference between both double mutants for liposome association (Fig. 4 C). To provide a quantitative assessment of the impact of these mutations on liposome association, we calculated the equilibrium partition coefficient (K_p) as described previously (45,76). A K_p of 2815 was obtained here for WT RBD-CRD (Fig. 4 D; Table S1), which is consistent with previous measurements (45), although it should be noted that the study in (45) used a different liposome composition and buffer conditions than here. Both RBD-CRD double mutants showed a decrease in the free energy for transfer from the aqueous phase to the membrane (ΔG_p) by 1.6–2.1 kJ/mol compared to WT RBD-CRD as well as a decrease in the fraction of protein bound to membrane (f_b) by 16.6–18.8% (Table S1). The quadruple mutant (R59A/R67A/R73A/R89A) showed even weaker liposome association compared to the two double mutants; however, the effects of both double mutants were not additive in the quadruple mutant (Fig. 4, B–D). In particular, the quadruple mutant showed a decrease in ΔG_p of 2.8 kJ/mol compared to WT RBD-CRD and a decrease in f_b of 24.8% (Table S1). The nonadditive decrease for the quadruple mutant could be due to several lysines near the RBS that were observed to make weaker contacts with POPS headgroups in the simulations (Fig. 4 A; Fig. S4) and that could compensate for the loss of the arginines. Nevertheless, the SPR results do show that the four RBS arginines are indeed involved in the membrane association of RBD-CRD.

To further test the roles of the four RBS arginines in membrane association, SPR measurements of the RBD-CRD constructs to liposomes with higher POPS content (50:50 POPC:POPS) were also performed (Fig. S9). The two double mutants again showed a significant decrease in membrane association compared to WT RBD-CRD based on the measured K_p and the two quantities, ΔG_p and f_b, calculated from the K_p (Fig. S9 C; Table S1). When comparing between liposomes with 50% POPS over those with 30% POPS, WT RBD-CRD showed an increased K_p of 3× to liposomes with 50% POPS (Table S1). Both double mutants also showed an increased K_p to 50% POPS liposomes, although the observed increase for the double mutants (1.6–2×) was less than that of the WT construct. These differences in the relative K_p translate to a ΔG_p difference for the two liposomes types of −2.8 kJ/mol for WT while just −1.2 to −1.8 kJ/mol for both double mutants. Because the four RBS arginines are intact in the WT construct, these would allow for an increased amount of cumulative electrostatic interactions with the membrane compared with the double mutants. The quadruple mutant also showed decreased association to liposomes with 50% POPS relative to WT and both double mutants (Fig. S9 C; Table S1). Interestingly, however, there was almost no change in the association of the quadruple mutant to liposomes with either 30% POPS or 50% POPS (Table S1). These results suggest that the bulk of the RBD makes nonspecific interactions with the membrane (as seen for the quadruple mutant that showed similar binding to both liposome compositions), whereas the four RBS arginines facilitate stronger interactions between RBD-CRD and the membrane, possibly through the conformational basin seen in the earlier simulations that has the RBS hidden (conformation 1 in Fig. 2 B). Consistent with these observations, a recent study showed through NMR paramagnetic relaxation enhancement (PRE) measurements that although the membrane orientation of the RBD is variable (likely through nonspecific interactions), the RBS was still the major region from the RBD involved in interactions between CRAF RBD-CRD and the membrane (77).

Enhanced membrane affinity of RBD-CRD construct by local enrichment of anionic lipids

Past evidence clearly suggests that membrane association of the CRD is increased by the local concentration of anionic lipids (23, 24, 25, 26). As the RBD also makes contacts with anionic POPS, we wanted to determine if the presence of the RBD can enhance the local anionic lipid concentration that can in turn enhance the membrane affinity of RBD-CRD. We first ran ten AA simulations of CRAF CRD anchored in a membrane containing 100 lipids per leaflet with a 70:30 ratio of POPC:POPS. We compared these CRD simulations with the earlier AA RBD-CRD simulations by calculating the distribution of POPS lipids within a threshold distance of either 0.5 or 3 nm from the CRD (Fig. S10 A). No evident difference was observed for the distribution of POPS lipids when the RBD is either absent or present. We next performed CG simulations with either CRAF CRD or CRAF RBD-CRD in membrane patches containing the same number of lipids as in the AA simulations (100 lipids per leaflet) and still observed no striking difference for the POPS distribution in the absence or presence of the RBD (Fig. S10 B).

To evaluate whether these observations might be artifacts because of the limited size of the model membrane patches, we performed CG simulations of CRAF CRD anchored in a membrane with 1600 lipids per leaflet (1120 POPC and 480 POPS) and compared these with the earlier CG membrane-anchored RBD-CRD simulations with the same membrane patch size. We found that the larger membrane patches were able to dissipate the effect of the narrow confinement of the protein in the smaller membrane patches with a clear indication of anionic lipid colocalization within 3 nm of the CRD in the presence of the RBD (Fig. S10 C). Calculation of the density of POPS lipids in the CG simulations with larger membrane patches indeed showed that the presence of RBD enhances the local POPS density (Fig. 5 A). This POPS enrichment propagates up to distances 4 nm away from the CRD COM, although the majority of the enrichment occurs within the first and second lipid solvation shells of the CRD. As expected, a concomitant displacement of POPC lipids from these two shells (Fig. 5 B) occurs along with the RBD-induced POPS enrichment.

Figure 5.

Local enrichment of anionic lipids can enhance the membrane affinity of RBD-CRD. (A) Densities of POPS from the CG RBD-CRD simulations using larger membrane patches (1600 lipids per leaflet). Shown are the densities of an 8 nm × 8 nm portion of this membrane patch centered on the CRD. These densities were calculated considering only the serine and phosphate groups of POPS, either in the absence (−RBD) or presence (+RBD) of the RBD. (B) Same as in (A) but for POPC densities, considering only the choline and phosphate groups of POPC. (C) Representative SPR measurements of 70:30 POPC:POPS liposome association for RBD alone (black), CRD alone (red), and RBD-CRD (blue) constructs. Protein concentrations between 50 and 1 μM were tested for each of the constructs. The RU_S from these measurements have been normalized by the RU_L. (D) Steady-state binding isotherms calculated from the SPR sensorgrams in (C) as a function of protein concentration. (E) Comparison of the equilibrium partition coefficients (K_p) for 70:30 POPC:POPS liposome association. Error bars give the SD over two experiments. Blue asterisk indicates p < 0.05 based on two-sided Student’s t-test when comparing with RBD-CRD levels.

We next wanted to determine if the observed local enrichment of anionic POPS when RBD is present in the simulations translates to enhanced membrane association of RBD-CRD. SPR measurements for association to liposomes composed of 70:30 POPC:POPS indeed showed stronger association of RBD-CRD compared to CRD alone (blue versus red curves in Fig. 5, C–E), with RBD-CRD exhibiting an almost 16-fold increase in K_p and a −6.7 kJ/mol change in ΔG_p compared with CRD alone (Fig. 5, D and E; Table S1). Similar trends were also observed for the association of RBD-CRD and CRD alone to liposomes with higher POPS content (50:50 POPC:POPS) (Fig. S11). However, the CRD showed only a modest increase in its association with these liposomes (ΔG_p change of −0.6 kJ/mol between 50% POPS and 30% POPS liposomes), whereas the RBD-CRD showed a larger increase in its association between the two liposome types (corresponding ΔG_p change of −2.8 kJ/mol) (Table S1), further highlighting the role of RBD-induced anionic lipid enrichment around RBD-CRD. Note that the RBD alone showed no detectable association to 30% POPS liposomes in the SPR measurements (black curves in Fig. 5, C–E), whereas weak binding of RBD alone to the 50% POPS liposomes was detected (Fig. S11; Table S1).

Because the liposome association of RBD alone was found to be negligible (for 70:30 POPC:POPS liposomes; Fig. 5, C–E) or very weak (for 50:50 POPC:POPS liposomes; Fig. S11), the earlier observations on the role of various RBS positively charged residues for bringing the RBD closer to the membrane likely arise because of the connection of the RBD to an already membrane-anchored CRD via a short peptide linker. We note that the percentage of POPS in the plasma membrane is typically lower (78) than the amount considered in our simulations and experiments. In addition, strong charge screening mechanisms can strongly impact the nature of the interactions observed in our simulations. Regardless, it is intriguing that the presence of the RBD can enhance the membrane affinity of RBD-CRD, possibly through the local enrichment of surrounding POPS lipids, which could account for the increased K_p to POPC:POPS liposomes observed for RBD-CRD compared to CRD alone. The measured K_p for CRD alone was around 3.6- to 5-fold lower than that for the quadruple mutant of RBD-CRD (Fig. 5 E vs. Fig. 4 D for 70:30 POPC:POPS; Fig. S11 C vs. Fig. S9 C for 50:50 POPC:POPS; Table S1), which may be due to the nonspecific membrane interactions by the RBD mentioned earlier. This may also be due to RBD-induced allosteric effects in the CRD that could contribute to enhancing the membrane affinity of RBD-CRD (such as by increasing the exposed surface area of the two CRD membrane-anchoring hydrophobic loops, leading to deeper membrane embedding), which our results here neither address nor discount.

Two studies were recently published that highlighted the membrane association of the linked RBD-CRD domains. In one study, NMR PRE measurements were used to study the interactions between CRAF RBD-CRD and the membrane surface of anionic lipid-containing nanodisks (77). They found that the strongest PRE signals from the RBD came from residues found in the RBS, which is consistent with basin 1 from our simulations (representative conformation 1 in Fig. 2 B), which provides limited access for KRAS4b binding. Despite RBS showing the strongest PRE signals, their PRE-driven HADDOCK models suggested that membrane-associated RBD-CRD can adopt variable orientations to promote productive encounters with the KRAS4b G domain. This is consistent with basin 2 from our simulations (representative conformation 2 in Fig. 2 B), which has an accessible RBS. As mentioned earlier, these variable orientations may arise from nonspecific membrane interactions of non-RBS residues from the RBD and would account for why the RBD-CRD quadruple mutant here showed similar binding to 30% POPS and 50% POPS liposomes (Table S1). Of note, they attributed the enhanced membrane association of RBD-CRD relative to either domain alone as due to CRD being able to recruit RBD to the membrane through a multivalent effect from the short peptide linker between both domains. Our results here, particularly the larger increase in membrane partitioning of WT RBD-CRD (compared with CRD alone) to 50% POPS over 30% POPS liposomes (Table S1), provide support to the notion that the membrane recruitment of RBD by the linked CRD leads to a local enrichment of anionic lipids around RBD-CRD that can further enhance its membrane association. The simulation results in Fig. S10 indicate that anionic lipid enrichment is impacted by the size of the membrane patch, with only larger membrane patches having enough available anionic lipids to observe this enrichment. Similarly, we anticipate that anionic lipid enrichment around RBD-CRD is more pronounced in cell membranes (that additionally contain multiple anionic lipid types) and even in liposomes compared to smaller membrane patches such as those found in nanodisks.

In another study, time-lapse tapping mode atomic force microscopy (AFM) was used to characterize the organization of RBD-CRD from the three RAF isoforms on raft-like anionic lipid-containing model biomembranes (79). RAS-independent nanoclusters of RBD-CRD were found on both liquid-disordered ($l_d$) and liquid-ordered ($l_o$) phases of the heterogeneous membrane, whereas, consistent with earlier reports (80, 81, 82, 83), the formation of KRAS4b nanoclusters was localized to the $l_d$ phase that is anionic lipid rich (84,85). A “two-dimensional” mechanism of engagement at the membrane surface was then suggested in which the formation of RBD-CRD nanoclusters around the KRAS4b nanoclusters can potentially enhance the formation of RAS-RAF complexes that allow RAF KD dimerization and activation. The RBD-induced enrichment of anionic lipids around RBD-CRD that we propose here could contribute to this mechanism by increasing the encounter rate between the KRAS4b and RBD-CRD nanoclusters at the membrane. One question that remains unanswered, however, is how RAS-independent membrane recruitment of RAF would take place given that the recent cryo-electron microscopy structure of the autoinhibited BRAF complex shows the membrane-anchoring loops of the CRD buried within the complex (39). Indeed, a “three-dimensional” mechanism for RAS-RAF engagement was proposed in (39), which involves active RAS recruiting the autoinhibited complex to the membrane via interactions with the RBD followed by “extraction” of the CRD for membrane anchoring, although the molecular details of how the CRD gets freed from the complex are unknown. Several studies indicate that the mechanism for membrane localization of RAF can be isoform dependent; siRNA knockdowns of the membrane proteins prohibitin or CNK1 specifically blocked the interactions of active RAS with CRAF but not BRAF or ARAF (40, 41, 42, 43, 44), suggesting that the membrane recruitment of CRAF is mediated by other membrane proteins and with direct interactions between CRAF and active RAS occurring later. The molecular details on the release from the autoinhibited complex are also unknown here, and more investigations are needed to understand the potential RAF isoform-specific mechanisms linking membrane recruitment and the transition away from the autoinhibited state after RAS activation.

Conclusions

A combination of CG and AA simulations revealed the synergistic dynamics of the two domains of CRAF RBD-CRD in the context of a POPC:POPS membrane model. These simulations predicted that membrane-anchored CRD can bring the RBD near the membrane surface in a dominant conformational basin that has the RBS occluded because of interactions of several positively charged residues from the RBS with anionic lipids. SPR measurements validated the role of these arginines in the membrane association of RBD-CRD. In turn, simulations also predicted that the RBD enriches the local distribution of anionic lipids that could lead to increased membrane association of RBD-CRD. This represents an interesting multivalency effect in which the attached domain enhances multivalent binding by recruiting specific lipids. SPR measurements indeed showed that RBD-CRD has a 16-fold increase in membrane partitioning compared to CRD alone. The RBD-induced local concentrations effects on anionic lipids may be of little functional relevance to BRAF whose membrane localization is thought to be initiated by the binding of RBD to membrane-anchored and activated RAS. However, these local concentration effects on anionic lipids may become very relevant for CRAF in enhancing the membrane anchoring of the CRD after the release of its hydrophobic loops from 14-3-3 in the autoinhibited complex.

Author Contributions

T.T. and J.J.H. performed the AA simulations, which were conceived with S.G. C.A.L. performed the CG simulations. T.T., C.A.L., and J.J.H. analyzed all the simulations. C.A. and A.G.S. performed and interpreted the SPR measurements. S.M. developed protein purification methods. T.T. wrote the article; all authors participated in discussing the results and commented on the manuscript.

Editor: Roland Winter.

Supporting Citations

Reference (86) appears in the Supporting Material.