Raf-1 Cysteine-Rich Domain Increases the Affinity of K-Ras/Raf at the Membrane, Promoting MAPK Signaling

SUMMARY

K-Ras4B preferentially activates Raf-1. The high-affinity interaction of Ras-binding domain (RBD) of Raf with Ras was solved, but the relative position of Raf’s cysteine-rich domain (CRD) in the Ras/Raf complex at the membrane and key question of exactly how it affects Raf signaling are daunting. We show that CRD stably binds anionic membranes inserting a positively charged loop into the amphipathic interface. Importantly, when in complex with Ras/RBD, covalently connected CRD presents the same membrane interaction mechanism, with CRD locating at the space between the RBD and membrane. To date, CRD’s role was viewed in terms of stabilizing Raf-membrane interaction. Our observations argue for a key role in reducing Ras/RBD fluctuations at the membrane, thereby increasing Ras/RBD affinity. Even without K-Ras, via CRD, Raf-1 can recruit to the membrane; however, by reducing the Ras/RBD fluctuations and enhancing Ras/RBD affinity at the membrane, CRD promotes Raf’s activation and MAPK signaling over other pathways.

INTRODUCTION

As a small p21 guanosine triphosphatase (GTPase), Ras plays a central role in cellular signal transduction pathways (Lacal et al., 1986; Lu et al., 2016a; Malumbres and Barbacid, 2003), including proliferation, differentiation, apoptosis, and senescence (Quinlan and Settleman, 2009; van Hattum and Waldmann, 2014), by binding effectors and regulators, such as Raf and phosphatidylinositol 3-kinase (PI3K) (Yan et al., 1998). Raf leads to cancer via major signaling cascades, primarily the Raf/MEK/ERK (MAPK) pathway. Raf activation requires active Ras that anchors to the plasma membrane (PM). Membrane-anchored, active Ras binds Raf, promoting homodimerization of Raf’s catalytic kinase domain and trans-autophosphorylation (Jambrina et al., 2016; Tse and Verkhivker, 2016). Active Raf dimer phosphorylates and activates MEK1/2, which induces ERK1/2 activation, with the phosphorylation signal cascading downstream. Although all Ras isoforms, H-Ras, N-Ras, and K-Ras can activate MAPK signaling, K-Ras is a more potent Raf activator than H-Ras (Castellano and Santos, 2011; Plowman et al., 2008; Yan et al., 1998). Proximal Ras molecules, dimers and nanoclusters, promote Raf dimerization and activation, thus MAPK signaling (Chen et al., 2016; Jang et al., 2016b; Muratcioglu et al., 2015; Nan et al., 2015).

There are three Raf kinase families: A-Raf, B-Raf, and C-Raf (or Raf-1) (Lavoie and Therrien, 2015; Leicht et al., 2007). Raf typically consists of the Ras-binding domain (RBD) and the cysteine-rich domain (CRD or C1 domain, C-kinase homologous domain 1) at the N-terminal region, and the kinase domain at the C-terminal region (Figure 1A). The chain between the CRD and kinase domain is disordered (Nussinov et al., 2017a) and accommodates the highly flexible hinge region. The kinase domain can be divided into the N-terminal and C-terminal lobes. Crystal structures of Raf separately solved the kinase and RBD domains. No crystal structure for the whole Raf molecule is currently available. Even for the CRD, to date there is no crystal structure; however, the solution structure of Raf-1 CRD was determined by nuclear magnetic resonance (NMR) (Mott et al., 1996) (Figure 1B). The secondary structures of CRDs from protein kinase C (PKC) α and δ (Hommel et al., 1994; Zhang et al., 1995) resemble that of Raf-1, with conserved sequences (Mott et al., 1996). The solution structure of CRD from the kinase suppressor of Ras (KSR) (Zhou et al., 2002) and other CRD family members (Houssa and van Blitterswijk, 1998) also exhibit a similar topology, implicating homologous roles at the membrane.

Figure 1. Sequence and Domain Structures of Raf-1.

(A) Raf-1 is composed of RBD (residues 56–131), CRD (138–184), and kinase domain (349–609). A long flexible linker connecting the CRD and kinase domain is largely disordered. In the sequence, hydrophobic, polar/glycine, positively charged, and negatively charged residues are colored black, green, blue, and red, respectively. Gray denotes the unstructured region.

(B) Crystal structure of Raf-1 RBD (PDB: 4G0N) and solution structure of Raf-1 CRD (PDB: 1FAR). The structure of 1FAR is a minimized average. Similarly, in the ribbon representation for the secondary structures, the same colors were used, except for the hydrophobic residues, which are colored white.

Ras interaction with Raf is of profound interest, given its key role in the MAPK pathway. Ras interacts with Raf RBD through the effector binding site with high affinity (Chuang et al., 1994; Herrmann et al., 1995). Several structural studies targeted the Ras/Raf-RBD complex exploiting the RBD crystal structures in complex with Ras (Aramini et al., 2015; Fetics et al., 2015; Kauke et al., 2017; Walker et al., 2014; Wan et al., 2004; Zeng et al., 1999). NMR and X-ray crystal structure studies illustrated that upon Ras binding, B-Raf RBD undergoes conformational change with the allosteric signal propagating from the GppNHp-bound H-Ras to the core region of the RBD (Aramini et al., 2015). For Raf-1 RBD in complex with the GppNHp-bound H-Ras, X-ray crystal studies combined with molecular dynamics (MD) simulations demonstrated that allosteric signals propagate through pathways connecting the Ras allosteric lobe to Raf-1 RBD (Fetics et al., 2015). Compared with Raf RBD, fewer studies involved Raf CRD due to lack of structural information. Although RBD is essential for Raf activation, CRD can unveil an additional facet in Raf activation (Drugan et al., 1996). NMR spectroscopy of H-Ras demonstrated that Raf-1 CRD could attract the farnesyl group in the H-Ras/CRD complex, confirming a second Ras-binding site (Thapar et al., 2004). ELISA demonstrated that Raf-1 CRD interacts selectively with phosphatidylserine (PS) (Ghosh et al., 1994). The authors have subsequently defined the cluster of basic amino acids, Arg-143, Lys-144, and Lys-148, in Raf-1 CRD as critical for the interaction with PS (Improta-Brears et al., 1999).

To activate Raf, Ras needs to localize in the PM. The hypervariable region (HVR), which differs among Ras isoforms, modulates Ras microdomains selectivity and thus membrane localization (Banerjee et al., 2016; Muratcioglu et al., 2015). Prenylated K-Ras4B HVR preferentially interacts with the liquid phase anionic membrane and spontaneously inserts the farnesyl group into the phospholipid bilayers (Jang et al., 2015). The 22–23 amino acid long HVRs of Ras isoforms vary in their sequences and properties, resulting in distinct interactions (Laude and Prior, 2008). The HVR of K-Ras4B is highly positively charged, targeting anionic lipid bilayers. The HVR of K-Ras4A is also positively charged, but to a lesser extent (Chakrabarti et al., 2016). Depalmitoylation of K-Ras4A renders a behavior resembling K-Ras4B (Nussinov et al., 2016, 2017a). In solution, K-Ras4B HVR may directly interact with the catalytic domain, yielding an autoinhibited form of K-Ras4B (Chavan et al., 2015). HVR autoinhibition due to blockage of the effector binding site can be released in the guanosine triphosphate (GTP)-bound and oncogenic mutant states (Lu et al., 2015, 2016b). HVR-induced autoinhibition may persist in the K-Ras4B inactive state at the anionic lipid bilayer, while the catalytic domain with membrane-anchored farnesyl and unrestrained HVR fluctuates reinlessly in the active state, exposing its effector binding site (Jang et al., 2016a). Effector binding site accessibility influences the active and inactive Ras states at the membrane. Although the GDP-bound state populates the inactive Ras conformation at the membrane, if the effector binding site is occluded, GTP-bound K-Ras4B can also be inactive. In the active-state K-Ras4B membrane orientation, the catalytic domain orients its allosteric helices perpendicular to the membrane surface. In this conformation, K-Ras4B can dimerize through its α3 and α4 helices. The allosteric helices align parallel to the membrane normal, a robust organization for dimer interface formation and efficient Raf activation. Ras dimerization through the allosteric lobe interface facilitates Raf dimerization and activation (Jang et al., 2016b; Muratcioglu et al., 2015; Nan et al., 2015).

Here, we employ all-atom MD simulations to investigate membrane-anchored Raf-1 CRD and suggest a model of K-Ras4B/Raf-1 complex at the anionic lipid bilayer composed of 1,2-dioleoyl-sn-glycero-3-phosphocholine (DOPC) and 1,2-dioleoyl-sn-glycero-3-phosphoserine (DOPS) in mole ratio 4:1. Active Ras modulates Raf activation, with K-Ras4B preferentially activating Raf-1 (Carey et al., 2003; Jaumot et al., 2002; Plowman et al., 2008; Yan et al., 1998). The HVRs of active Ras isoforms, which display selective membrane localization, influence the association with Raf’s CRD. Our studies on isolated Raf-1 CRD show a specific preferred binding mode with the membrane, with electrostatic interactions being the driving force. With the linker connecting RBD to CRD only six amino acids long, Raf-1 CRD abuts the RBD (Figure 1). The crystal structures of the complex of K-Ras4B with RBD provide its location (Aramini et al., 2015; Fetics et al., 2015); but the relative position of RBD with respect to CRD is unknown. Our simulations indicate that Raf-1 CRD uniquely interacts with the lipid bilayer in the K-Ras4B/RBD-CRD complex. Notably, the interaction patterns for both isolated and complexed CRDs with the membrane are practically identical, suggesting that CRD is an intrinsic membrane binding domain of Raf kinase. Raf-1 CRD not only offers an additional anchor for the K-Ras4B/Raf-1 complex; by reducing the fluctuations of Ras/RBD it increases the population of Ras/RBD at the membrane and enhances the already high affinity between Ras and RBD. This promotes MAPK signaling, the key Ras proliferative pathway, above other pathways. Thus, even though in the absence of CRD the Raf-1 RBD/Ras interaction can elicit MAPK signaling, the enhanced stability of the complex at the membrane rendered by CRD can considerably intensify it by helping in Raf’s dimerization. Taken together, our data reveal the roles and mechanism of CRD at the membrane in Raf activation.

RESULTS

CRD Is a Membrane Binding Domain of Raf-1

The solution structure of Raf-1 CRD from NMR (PDB: 1FAR) reveals two Zn²⁺ coordination sites, essential for preserving the globular conformation (Figure 1). The first is held by Cys-152, Cys-155, His-173, and Cys-176, and the second by His-139, Cys-165, Cys-168, and Cys-184. MD simulations of CRD in an aqueous environment verified that the whole structure becomes unstable in the absence of Zn²⁺; the Zn²⁺ binding sites started to unfold due to large fluctuations (Figure S2). Thus, hereafter all our simulation studies include Zn²⁺ coordination. To sample the membrane-associated CRD conformations and investigate the specific binding mode of Raf-1 CRD with the membrane, we implemented the ZDOCK docking program (Pierce et al., 2014). We defined seven basic residues in CRD (Arg-143, Lys-144, Lys-148, Lys-157, Arg-164, Lys-171, and Lys-179), which we considered as key in the event of membrane attachment. From the predicted decoys, we selected the best initial configurations based on docking energy scores, ensuring that at least three basic residues faced the surface of the anionic membrane. Eight different initial configurations of CRD were subjected to explicit MD simulations with anionic lipid bilayer containing DOPC:DOPS (4:1, mole ratio) (Figure S3).

Basic Residues of Raf-1 CRD Are Responsible for Membrane Binding

We observed significant residue fluctuations in membrane-bound Raf-1 CRD, with large RMSD values compared with those in solution (Figure S4A). Electrostatic interactions of the basic CRD residues with the anionic lipids dynamically stimulated conformational evolution into the lipid-anchored conformations (Figure S4B). The Zn²⁺ coordinates were preserved, indicating that CRD retained its overall globular conformation. However, in some configurations, CRD significantly altered its membrane orientation during the simulations. The measured membrane orientation of CRD was calculated for the angle, θ_Zn, between the vector connecting two zinc ions and the direction of membrane normal. For configurations 1, 6, 7, and 8, wide probability distributions over θ_Zn indicate that the initial CRD alignment with respect to the membrane surface is less favorable (Figure 2A). In this case, CRD weltered on the membrane surface until the basic residues stably anchored to the amphipathic interface of the lipid bilayer. For configurations 2–5, the CRDs exhibit unique orientations, yielding prominent peaks in the probability distribution curve above θ_Zn. Different locations of peaks indicate that CRD employs diverse combinations of the basic residues to interact with the membrane (Figure 2B). We observed that except for configuration 3, at least three basic residues diffused into the amphipathic interface of the lipid bilayer. For configuration 3, instead of the basic residues, both termini are buried in the membrane, indicating an unsuitable model for the CRD-membrane interaction. For configurations 1, 5, and 8, all basic residues, except Lys-179, are commonly localized at the amphipathic interface, suggesting that the basic residues of CRD play a key role in the membrane anchoring mechanism.

Figure 2. Membrane Interaction of Raf-1 CRD.

(A) Probability distribution functions of the angle, θ_Zn, between the vector connecting two zinc ions in Raf-1 CRD and the direction of membrane normal for eight different CRD configurations (Config. 1–8) interacting with the anionic bilayer composed of DOPC:DOPS (4:1, mole ratio).

(B) Snapshots of Raf-1 CRD interacting with the anionic bilayer depicted from the final conformation at 200 ns. In the bilayer structure, DOPC and DOPS are shown as white and gray surfaces, respectively. In the CRD structure, blues spheres denote zinc ions, and red arrows represent the vector. Key basic residues are marked.

To determine the most significant residue that is responsible for Raf-1 CRD-membrane attachment, we calculated the probability of lipid contact for each basic residue during the simulations (Figure 3A). For configurations 1, 5, and 8, the CRD generates high contact probability of the basic residues with the lipids. The cumulative occurrence of lipid contacts for the key basic residues from all configurations suggests that Lys-148 is the most important residue responsible for CRD-membrane binding, followed by Lys-143, Lys-144, and Lys-157, which also play important roles in the anchoring process (Figure 3B), which is consistent with earlier experimental observations (Improta-Brears et al., 1999). To grade the membrane-associated CRD conformations, we calculated the binding energy of CRD with the membrane using the molecular mechanics generalized Born surface area (MM-GBSA) method. The calculations show that configurations 1, 5, and 8 have low values of binding energy of CRD with the membrane (Figure 3C). These may represent models for CRD-membrane interaction with high binding affinity. Among these configurations, based on our observations, configuration 5 can be designated as the best representation for the membrane-bound CRD conformation. The “stand-up” motif with cos θ_Zn ~ 1 provided the most stable CRD conformation at the membrane.

Figure 3. Key Basic Residues of Raf-1 CRD.

(A) The probability of lipid contact for each basic residue in Raf-1 CRD at the anionic bilayer composed of DOPC:DOPS (4:1, mole ratio).

(B) Cumulative frequency of the lipid contact for the key basic residues from all configurations.

(C) Binding energy of Raf-1 CRD with the anionic bilayer calculated by the MM-GBSA method for eight different CRD configurations (1–8).

Raf-1 CRD Can Restrain the K-Ras4B/RBD Fluctuations at the Membrane

Raf CRD is known as a putative membrane binding domain, recruiting Raf to the membrane. It is in close proximity to RBD, with a short six amino acids linker between RBD and CRD (Figure 1). Because in the crystal structure Raf-1 RBD interacts with Ras in the absence of a membrane-mimicking environment, the relative positions of RBD and CRD at the membrane and the consequences are still unknown. Studies indicated that when its HVR anchored to anionic membrane, an effector-free active K-Ras4B exhibits multiple orientations resulting in unimpeded fluctuations of the catalytic domain (Jang et al., 2016a). However, the fluctuations can be restrained by effector attachment to the binding site. To observe the impact of effector association on membrane localization and orientation of Ras, we modeled seven K-Ras4B/RBD-CRD complexes on the anionic DOPC:DOPS (4:1) membrane. These consist of two different conformations of the active, GTP-bound K-Ras4B complexed with RBD-CRD; DCs 1–5 with model 1, and DCs 6 and 7 using model 2 of K-Ras4B/RBD at the membrane (Figure S1). Thus, in the initial configurations, the membrane orientation and location of both K-Ras4B and RBD across the bilayer were the same for each derived K-Ras4B/RBD model; however, CRD had different orientations. During the simulations, we observed that the K-Ras4B/RBD-CRD complex altered its membrane orientation and location, moderately drifting away from its initial setting (Figure 4). No immediate dissociation of the Raf-1 RBD from K-Ras4B was observed. The K-Ras4B catalytic domain retained a strong β-sheet interface with the Raf-1 RBD, verifying that the β-sheet interaction possessed high binding affinity. In contrast, large conformational fluctuations in both HVR and CRD could be observed. The dynamic interactions of the HVR and CRD with the lipids induced fluctuations leading to diverse membrane orientations and locations across the bilayer for the complex.

Figure 4. Membrane Interaction of Raf-1 RBD-CRD in Complex with K-Ras4B.

Snapshots of GTP-bound K-Ras4B interacting with Raf-1 RBD-CRD at the anionic bilayer composed of DOPC:DOPS (4:1, mole ratio) depicted from the final conformation at 300 ns. In each dimeric configuration (DC), the K-Ras4B-GTP is shown on the left-hand side. In the K-Ras4B cartoon, thick deep-blue tube, yellow, red sticks, and green sphere represent the HVR, farnesyl, GTP, and Mg²⁺, respectively. The Raf-1 is shown on the right-hand side. The pink and green cartoons are RBD and CRD, respectively. Blue spheres in CRD denote Zn²⁺. In the bilayer structure, DOPC and DOPS are shown as white and gray surfaces, respectively.

To determine the location of each component of the K-Ras4B/RBD-CRD complex across the membrane, we calculated the average positions of each domain of the proteins and lipid groups over the simulation trajectories. Position probability distribution functions for two different component groups, phosphate (PO₄) and terminal methyl (CH₃) groups of both DOPC and DOPS lipids, for the catalytic domain, HVR and farnesyl group of K-Ras4B, and for RBD and CRD of Raf-1 were calculated as a function of distance from the bilayer center (Figure 5). For lipids, the terminal CH₃ group is located at the center of the bilayer and the PO₄ group is located at both leaflets of the lipid bilayer. The bilayer surfaces can be defined by ±5 Å from the locations of the peaks for the PO₄ group. For K-Ras4B, the average positions of the catalytic domain and the HVR across the membrane are ~30 Å and ~5 Å above the bilayer surface, respectively, while the farnesyl group is located at ~11 Å below the bilayer surface. For Raf-1, the average position of RBD across the membrane varies among the dimeric configurations (DCs), but in general for K-Ras4B, the RBD is located at similar positions as the catalytic domain. The location of CRD across the membrane also varies, but it can be found at the same position where the K-Ras4B HVR is located.

Figure 5. Probability Distribution Across the Lipid Bilayer.

Probability distribution functions for different component groups of lipids (CH₃, methyl group, red; PO₄, phosphate group, orange), different domains of K-Ras4B (catalytic domain, yellow; HVR, green; farnesyl, blue), and different domains of Raf-1 (RBD, dark blue; CRD, purple) as a function of distance from the bilayer center for seven different DCs (1–7) of K-Ras4B/RBD-CRD complex at the anionic bilayer composed of DOPC:DOPS (4:1, mole ratio).

To quantify how the K-Ras4B/RBD-CRD complex localizes on the membrane, we measured deviations of selected residues/atoms locations from the bilayer surface for the catalytic domain and the HVR of K-Ras4B, and RBD and CRD of Raf-1. For K-Ras4B catalytic domain, the calculated deviations of selected residues for DCs 1–5 show that Gly-48 in L3 loop (between β2 and β3 strands) has lower deviation from the bilayer surface than Gln-61 at the Switch II region (Figure S5A), whereas for DCs 6 and 7 the deviations show that Gln-61 is located more closely to the bilayer surface than Gly-48. Their relative locations vary among the configurations, suggesting that the catalytic domain may have a seesaw movement supported by a pivot point near Ser-39. Probability distribution functions of membrane contacts for the residues verify that, for DCs 1, 4, and 5, residues nearby Gly-48 yield higher contact probability, and the Switch II region and the loop region around Lys-104 exhibit higher contact probability for DC 6 (Figure 6). For the K-Ras4B HVR, all configurations show a similar pattern with the anchor portion of the HVR residing on the membrane surface and the farnesyl group inserting into the hydrophobic core of the lipid bilayer with negative deviations (Figure S5B). For Raf-1 RBD, DCs 1–5 show that the RBD is reluctant to interact with the membrane (Figure S5C). However, for DCs 6 and 7 the positively charged residue Lys-106, which is one of eight positive charges on the protein surface, has a lower deviation from the bilayer surface, leading to higher contact probability for residues in the loop containing Lys-106 (Figure 6). For Raf-1 CRD, all configurations except DCs 3 and 5 show that the positively charged residues, Lys-144, Lys-148, and Lys-157, have lower deviations from the bilayer surface (Figure S5D). For DC 3, we observed that CRD transiently lost contact with the membrane during the simulation. Our simulations verify that, for the K-Ras4B/RBD-CRD complex, the key CRD residues interacting with the membrane are those observed in the isolated CRD-membrane simulations. High membrane contact probabilities of residues nearby the key residues (Figure 6) suggest that CRD acts as a membrane binding domain of Raf-1.

Figure 6. Lipid Contact Probability.

The probability of lipid contact for the residues of K-Ras4B (upper), Raf-1 RBD (middle), and Raf-1 CRD (lower) for seven different DCs (1–7) of K-Ras4B/RBD-CRD complex at the anionic bilayer composed of DOPC:DOPS (4:1, mole ratio). Inset in the upper panel highlights the probability for the residues of K-Ras4B in the range of 40–110.

Raf-1 CRD Supports K-Ras4B Functional States at the Membrane

To determine the orientations of K-Ras4B in the presence of Raf-1 RBD and CRD at the lipid bilayer, we calculated the helix angle, θ, with respect to the bilayer normal (Figure 7). Ras proteins contain five α helices; two effector lobe helices, α1 and α2, are almost perpendicular to the other three allosteric lobe helices, α3 to α5. These helices can be oriented with θ ~ 90° when the helix axis lies on a plain of the bilayer surface; however, they can be θ ~ 0° or θ ~ 180° when the helix axis is parallel or antiparallel to the bilayer normal. Probability distributions of the helix tilt angles indicate that K-Ras4B exhibits active-state helix orientations as observed in the case of the Raf-free membrane interaction (Jang et al., 2016a). The active-state helix orientations show that the effector lobe helices generally orient with θ → 90°, while the allosteric lobe helices favor orienting with θ → 180°. To corroborate the active-state helix orientations, we calculated the orientational order parameter of the helix (Berneche et al., 1998; Lague et al., 2005) using an equation defined by:

$$S_{\mathsf{C}=\mathsf{O}}=\frac{1}{N}{\sum }_{k=1}^N\left(\frac{3{\cos }^2{\theta }_{\mathsf{S}}-1}{2}\right),$$ (Equation 1)

where θ_S is the angle between the vector of the backbone carbonyl bond, C=O, in the helix residues and the bilayer normal, and N is the total number of the vectors. The helix orientation in the bilayer has been measured experimentally by polarized attenuated total reflectance Fourier transform infrared (ATR-FTIR) spectroscopy (Axelsen et al., 1995; Han and Tamm, 2000). When the carbonyl bond vectors lie on a plane of the bilayer surface but randomly orient in the plane, the helix order parameter is S_C=O ~ −0.5. In this case, the helix is partially unfolded and the average orientation of the vector is not well defined (Yang and Wu, 2014). When all carbonyl bond vectors in a folded helix lie on the plane of the bilayer surface, the helix order parameter is S_C=O ~ 0. For a helix aligning its axis perpendicular to the bilayer surface, the helix order parameter is S_C=O ~ 1. For all configurations, the probability distributions of the helix order parameter are generally correlated with the probability distributions of the helix tilt angle (Figure S6). Our simulations demonstrate that CRD supports the active-state orientations of K-Ras4B at the membrane, which indicate that the allosteric lobe helices orient their axes parallel to the bilayer normal and the effector lobe helices lie with their axes on the bilayer surface.

Figure 7. Helix Tilt Angle of K-Ras4B.

Probability distribution functions of the helix tilt, θ, with respect to the bilayer normal for two effector lobe helices (α1 and α2) and for three allosteric lobe helices (α3, α4, and α5) in the catalytic domain of K-Ras4B in complex with Raf-1 RBD-CRD for seven different DCs (1–7) at the anionic bilayer composed of DOPC:DOPS (4:1, mole ratio). The helix tilt was calculated for the angle between two vectors formed by the helix axis and normal axis of the bilayer.

To better quantify Ras orientation at the membrane, we measured the angle between the vector connecting two selected atoms in GTP and the bilayer normal. In GTP, we considered two vectors connecting the atom pairs $\text{C6}\to \text{O6}(\overrightarrow{\text{CO}})$ and $\text{PA}\to \text{O3A}(\overrightarrow{\text{PO}})$. These vectors are almost perpendicular to each other. The reason for the selection was that GTP was highly stable at the catalytic site and conserved a rigid conformation during the simulations. We sampled vector orientations over the trajectories for all configurations and monitored the probability based on occurrence frequency as a function of the angles of these vectors with respect to the bilayer normal. The two-dimensional contour surface of the probability map provides highly populated states enclosed by many contour lines (Figure 8). While the orientation of $\overrightarrow{\text{CO}}$ varied in the range of 40° < θ_CO < 100°, the orientation of $\overrightarrow{\text{PO}}$ is highly populated in the narrow range of 60° < θ_PO < 90°. Orientation III denotes the highly populated K-Ras4B state at the membrane, representing the active-state Ras orientation. Membrane-anchored CRD restricts the orientation of K-Ras4B to directions where the α4 and α5 helices or the plane of β sheets formed by β1, β2, and β3 strands face the membrane surface. This allows K-Ras4B to fluctuate between states where the α2 helix faces the membrane (state I) and the membrane contacting β2-β3 loop (VI), with the path proceeding via states I to II to III to V to VI. For DC 1, the highly populated K-Ras4B orientations are in states V and VI. DC2 highly populates the K-Ras4B orientation in states II and III. For DC3, the low population of orientation state IV suggests that K-Ras4B fluctuations are toward the CRD-restricted direction, when CRD loses contact with the membrane. Later in the simulation, when CRD attaches to the membrane, DC 3 populates K-Ras4B orientation state III. DC 4 populates the K-Ras4B orientation in states III, V, and VI, and DC 5 populates states III, IV, and V. In DC 5, occasional CRD detachments from the membrane yield low state IV population. In contrast, DC 6 shows a trapped K-Ras4B orientation, highly populated in state I, while DC 7 reorients K-Ras4B from state I to state III. Based on our observations, DC 2 can be designated as the best representation for the K-Ras4B/RBD-CRD complex with the active-state K-Ras4B orientation at the membrane.

Figure 8. K-Ras4B Orientation at the Membrane.

Two-dimensional probability distributions of orientation angles for two vectors, $\overrightarrow{\text{CO}}$ and $\overrightarrow{\text{PO}}$, in GTP with respect to the bilayer normal sampled from all DCs of K-Ras4B/RBD-CRD complex at the anionic bilayer composed of DOPC:DOPS (4:1, mole ratio). $\overrightarrow{\text{CO}}$ and $\overrightarrow{\text{PO}}$ denote the vectors connecting the atom pairs C6 → O6 and PA → O3A in GTP, respectively. The probability surface can be calculated from the occupancy frequency of visiting each grid point on the plane of $\overrightarrow{\text{CO}}$ and $\overrightarrow{\text{PO}}$ orientation angles, θ_CO and θ_PO, respectively, with respect to the bilayer normal (right). Six K-Ras4B conformations are depicted from the population map, representing the populated K-Ras4B membrane orientations in complex with Raf-1 RBD-CRD. For the protein structures on the membrane surface, the curved arrows on the top of the thick gray arrows measure the molecular inclination to the left- or right-hand sides. The curved arrows on the left measure the molecular rotation at the axis perpendicular to the thick gray arrows representing the molecular inclination to the forward or backward directions.

CRD Increases the Population of Functional K-Ras4B/RBD at the Membrane

We carried out simulations on two K-Ras4B/RBD model systems to monitor the effect on the K-Ras4B membrane orientations in the absence of CRD. In the initial configurations, these model complexes exhibited that the catalytic domain of K-Ras4B was aligned with extremely different membrane orientations. In model 1, the catalytic domain of K-Ras4B faced the membrane with the β2-β3 loop, while its α2 helix contacted with the membrane in model 2 (Figure S1). Model 1 showed Raf-1 RBD slipping away from the bilayer surface, and in model 2 the RBD partially interacts with the lipids. Without CRD, these simulations reflect two extreme cases: (1) K-Ras4B does not stably anchor to the membrane, and (2) K-Ras4B persists in a steric clash of its catalytic site with the lipid bilayer due to unexpected RBD-lipid interaction (Figure S7). For model 1, the probability distribution of the location of RBD on the bilayer surface is as widely dispersed as the K-Ras4B catalytic domain, suggesting that K-Ras4B/RBD is liberated from the membrane interaction. In this case, the helix order parameter is not well defined due to the fluctuations. The HVR marginally holds the complex, preventing departure from the membrane. In contrast, for model 2 the RBD is very close to the bilayer surface, like the HVR. The helix order parameters for the allosteric helices are S_C=O ~ 0, indicating that they lie on the bilayer surface. The two-dimensional contour surface of the probability map of $\overrightarrow{\text{CO}}$ and $\overrightarrow{\text{PO}}$ orientations in GTP clearly defines The CRD-absent K-Ras4B membrane orientation (Figure S8). Without CRD, the orientations of $\overrightarrow{\text{PO}}$ and $\overrightarrow{\text{CO}}$ are populated at (θ_PO, θ_CO) ~ (110°, 40°) and (70°, 110°) for model 1 and 2, respectively, which are beyond both ends of the CRD-bound ranges. We conclude that CRD facilitates K-Ras4B convergence toward attaining an active state at the membrane.

DISCUSSION

Raf interacts with Ras at the membrane; however, the role of the CRD domain in Raf’s membrane-indispensable productive activation has been enigmatic. Ras interaction with Raf’s RBD has been studied by crystallography and biochemical characterization. It is well established that Ras molecules need to be anchored in the membrane and be in spatial proximity. The rationale is understood: Ras nanocluster formation (or dimerization) increases the effective local concentration, which serves to promote Raf’s catalytic domain dimerization and auto-crossphosphorylation. However, the exact role of the CRD and how it operates to help in Raf’s activation have been elusive. Even though it was already postulated over two decades ago that CRD-membrane anchoring acts in stabilizing the Ras/Raf interaction (Chuang et al., 1994), its mode of interaction with the membrane and with Ras-RBD and its exact mechanism are still unresolved. Here we asked whether Ras/RBD would be stable at the membrane in the absence of CRD, and how CRD would affect the dynamics of the K-Ras4B/Raf complex, Raf’s activation and, critically, thus Ras-driven MAPK signaling at the membrane (Nussinov et al., 2017b). These questions are challenging to address experimentally. We focused on the role of CRD in the interaction of Raf-1 with K-Ras4B, the most oncogenic among the Ras isoforms, at the membrane.

In experiments carried out in the absence of a membrane-mimicking environment, the catalytic domain of active Ras directly binds to Raf RBD with nanomolar affinity via an extended intermolecular β sheet (Chuang et al., 1994; Herrmann et al., 1995). Raf CRD is in close proximity to RBD; the linker between these two domains is only six amino acids long. However, the relative position between RBD and CRD at the membrane is unknown. We observed that isolated Raf-1 CRD stably anchors to the anionic lipid bilayer by translocating its basic residues into the amphipathic interface. Two Zn²⁺ coordinates in the CRD are essential to retain its globular conformation. We further observed that three basic residues, Lys-144, Lys-148, and Lys-157, at the loop region (residues 144–161) are responsible for the membrane attachment. In line with our modeling, point mutations of the key basic residues, Arg-143, Lys-144, and Lys-148, at the loop region resulted in the loss of Raf-1’s capability to translocate to the PM (Improta-Brears et al., 1999). Notably, the loop region contains not only basic but also hydrophobic residues, suggesting that its intrinsic features pave the way for the membrane interaction. Thus, we designate this loop region as a membrane-insert loop. Our simulations demonstrate that this loop has a high probability of interacting with lipids, particularly PS (also in agreement with experiment; Improta-Brears et al., 1999), resulting in strong binding energy. Importantly, for the membrane-attached K-Ras4B/RBD-CRD complex, with Raf-1 CRD covalently connected to the RBD, the loop reenacts the innate lipid preference constituting the same membrane anchoring mechanism as the isolated CRD.

Our simulations show that Raf-1 CRD uniquely interacts with the lipid bilayer. The conformation and orientation of the CRD in the best representative model for the membrane-bound isolated CRD (configuration 5) are very similar to that of the CRD in the best representative model for the K-Ras4B/RBD-CRD complex (DC 2). The populated orientation and location across the bilayer indicate that the CRD in the complex is highly restricted in the space underneath the RBD due to the short linker. Notably, unlike the observation that the farnesyl group of H-Ras provided an additional binding site for the Raf-1 CRD (Thapar et al., 2004), the farnesyl group of K-Ras4B remains in the lipid bilayer without any interaction with the CRD. This suggests that the two additional palmitoyl groups of H-Ras are sufficient for membrane attachment without the help of the farnesyl. By contrast, for K-Ras4B, there is no replacement for the single prenyl group, which is essential in membrane association. Based on docking, the catalytic domain of K-Ras4B is averse to an interaction with the Raf-1 CRD, suggesting that Raf-1 CRD solely engages in membrane anchoring, and emphasizing the role of the short RBD-CRD linker. Acting as a footstone on the membrane, CRD’s role resembles that of the HVR, bolstering the arch-like morphology of the K-Ras4B/RBD-CRD complex. In that architecture, Raf-1 CRD locates in the space between the RBD and the membrane. The two anchor points in the membrane, the HVR and the CRD, can act to reduce the fluctuations of the Ras catalytic domain/RBD complex. We anticipate that this reduction in the fluctuations increases the affinity of the K-Ras4B/RBD complex at the membrane, thereby promoting Raf dimerization, its auto-crossphosphorylation and MAPK signaling, making it the first and most populated proliferative Ras signaling pathway at the membrane (Nussinov et al., 2017b).

Ras orientation at the PM is still a subject of debate. Earlier computational studies showed that H-Ras^G12V-GTP oriented its α4 and α5 helices parallel to the surface of the DMPC bilayer, while H-Ras^G12V−GDP oriented its β1-3 strands perpendicular to the bilayer surface, with the β2-β3 loop interacting with lipids (Gorfe et al., 2007). In contrast to H-Ras^G12V, which showed nucleotide-dependent membrane orientations, the catalytic domain of K-Ras4B^G12V exhibited similar membrane orientations, with helix α4 deviating from the membrane interaction in both GTP- and GDP-bound states (Abankwa et al., 2010). Solution NMR studies of K-Ras4B tethered to nanodiscs provided strikingly different results, showing that the catalytic domain orientation in the exposed GDP-bound state favored helices α4 and α5 facing toward the membrane, which is similar to the H-Ras^G12V−GTP orientation, whereas the catalytic domain orientation in the GTP-bound occluded state favored a β-sheet plane formed by β1-3 strands facing the membrane (Mazhab-Jafari et al., 2015). Recent computational studies of K-Ras4B^G12D-GTP at the membrane reported two populated orientations of the catalytic domain; state 1 sampled orientations with the α3 and α4 helices directly interacting with membrane, and state 2 represented the orientation with helix α2 and β1-3 strands forming direct contact with the membrane (Prakash et al., 2016). For GTP-bound K-Ras4A^G12D, five distinct membrane orientations of the catalytic domain were reported, suggesting that dominant orientation states at the membrane depend on the lipid composition, such as PS and phosphatidylinositol 4,5-bisphosphate (PIP₂) (Li and Buck, 2017), mutations, and nucleotide-binding states. Computational studies mostly predicted Ras orientations through sampling of catalytic domain conformations when bound to the membrane. However, in addition, the simulations show significant residence times for ensembles of conformation of the catalytic domain when not at the membrane, which are also populated (Jang et al., 2016a; Li and Buck, 2017). The lateral diffusion of Ras in the PM is as fast as lipid probes and significantly faster than a typical membrane protein (Kenworthy et al., 2004; Niv et al., 2002; Nussinov et al., 2017b). Ras can obtain faster lateral diffusion when only the lipidated hydrophobic HVR is engaged in membrane association, because membrane-bound catalytic domains can constrain the lateral dynamics and mobility of Ras in the PM. Higher mobility of Ras monomers can result in faster cluster formation, promoting Raf’s dimerization and the MAPK signaling. However, whether, or to what extent, these differences in the lateral diffusion speeds of Ras monomers in dimer/nanocluster formation are functionally significant is still unclear. Ras can be confined transiently in membrane microdomains depending on the lipid composition (Goodwin et al., 2005; Murakoshi et al., 2004). We suggest that, under physiological conditions, the most populated ensembles of membrane-anchored Ras are membrane-unbound catalytic domain conformations with only the HVR anchors to the membrane. Membrane-anchored Ras with the catalytic domain interacting with the membrane is a less populated transient state. Further, only active, membrane-anchored Ras monomers with membrane-unbound catalytic domain are likely to exist in clusters, with the monomers in equilibrium, diffusing into and out of nanoclustered assemblies. In the oncogenic state, the active conformation of K-Ras4B can be extracted from the membrane by calmodulin (Jang et al., 2017; Sperlich et al., 2016). Of note, phosphorylation of Ser-181 weakens membrane attachment, although may not abolish it (Zhang et al., 2017). Previously we suggested that, regardless of the nucleotide-binding states, active K-Ras4B may liberate the HVR from the catalytic domain exposing the effector binding site, while inactive K-Ras4B may sequester the membrane-interacting HVR burying the effector binding site (Jang et al., 2016a). In line with this, GTP-bound Ras with membrane-unbound catalytic domain can be an active form, because its effector binding site is exposed; however, to activate Raf (and RASSF5 [NORE1A], which leads to MST1/2 dimerization; Liao et al., 2016), active, spatially clustered, anchored Ras molecules are still essential.

For membrane-attached Ras conformations, the accessibility of the effector binding site may influence the functional orientations, and this holds particularly for low-affinity effectors. Occasionally, Raf RBD was modeled on membrane-bound Ras to define the functional state (Mazhab-Jafari et al., 2015; Prakash et al., 2016). This aimed to detect steric clash of the RBD with the lipid bilayer, to ensure that Ras was in an active conformation. Some predicted catalytic domain orientations, such as with allosteric helices, α3/α4 or α4/α5, facing toward the membrane, and with the RBD fitting into the effector binding site without any steric clash with the bilayers, were deemed to be active Ras conformations. These orientations allow the effector binding site to face opposite to the bilayer surface and the RBD to sit on top of the Ras catalytic domain. However, these models with the allosteric helices parallel to the membrane surface do not allow the CRD to effectively contact with the bilayer. Thus, Ras active-state orientation is supported by the interaction of CRD with the membrane, suggesting that determining Ras active orientation by modeling Ras with only Raf RBD may not be enough. Membrane-anchored CRD promotes the perpendicular orientations of the allosteric helices to the membrane surface, thereby removing occluded membrane orientations of the Ras catalytic domain. In the active state, Ras exposes the allosteric helices with increasing accessibility to another Ras, which facilitates Ras and Raf dimerization, thus activation. Importantly, here we show that a complex with high-affinity interaction measured in solution, such as Ras/RBD, may fluctuate when at the membrane. We speculate that reducing its fluctuation, which can be accomplished by CRD, increases the Ras/RBD affinity, thus its population, and, most importantly, boosting the signaling output of the MAPK pathway. We believe that, in the case of the complex of Raf-1 with K-Ras4B, this is the primary role of this Raf CRD domain.

To conclude, Raf-1 CRD plays critical roles in enhancing the stability of the interaction of the RBD with K-Ras4B at the membrane and in supporting the active-state orientation at the membrane, functions that to date have not been appreciated. The membrane insertion loop containing the positively charged and hydrophobic residues serves as a membrane binding domain of Raf-1 CRD. Raf-1 is preferentially activated by K-Ras4B, because its CRD selectively targets an anionic membrane as K-Ras4B does (Carey et al., 2003; Ghosh et al., 1994; Jaumot et al., 2002; Plowman et al., 2008; Yan et al., 1998). Comparison of the Raf-1 CRD sequence with other Raf isoforms illustrates that the membrane insertion loop from all isoforms conserves the lipid-interacting residues Lys-144 and Lys-157, but both A-Raf and B-Raf do not preserve Lys-148 as Raf-1 does, which is the residue with the high probability of membrane interaction. The highly polybasic nature of the membrane anchoring K-Ras4B HVR coincides with the membrane insertion loop of Raf-1 CRD involving additional basic residues. Even without K-Ras4B, via CRD, Raf-1 can recruit to the membrane; however, by reducing the K-Ras4B/Raf-1 RBD fluctuations, CRD enhances the K-Ras4B/RBD affinity at the membrane, increasing the population of the active signaling complex, and, most importantly, boosting MAPK signaling at the membrane.

STAR★METHODS

CONTACT FOR REAGENT AND RESOURCE SHARING

Further information and requests for resources and reagents should be directed to and will be fulfilled by the Lead Contact, Ruth Nussinov (NussinoR@mail.nih.gov).

METHOD DETAILS

Preparing the Isolated Raf-1 CRD Interacting with the Membrane

To model the isolated Raf-1 CRD interacting with the membrane, the solution structure of Raf-1 CRD (PDB code: 1 FAR) was obtained from the Protein Data Bank (PDB). The structure of 1FAR is a minimized average. CRD is the Raf-1’s membrane binding domain (residues 136 – 187) and contains two Zn²⁺ coordinates (Figure 1). The ZDOCK docking program (Pierce et al., 2014) was used to generate initial modes of the membrane-bound CRD configurations. A total of 8 different initial configurations were prepared for explicit molecular dynamics (MD) simulations with anionic lipid bilayer containing 1,2-dioleoyl-sn-glycero-3-phosphocholine (DOPC) and 1,2-dioleoyl-sn-glycero-3-phosphoserine (DOPS) (DOPC:DOPS = 4:1, mole ratio). To observe the Zn²⁺ coordinates effect on the CRD conformation, we prepared two additional simulations of CRD w/o Zn²⁺ in solution.

Generating Initial Configurations of K-Ras4B/RBD-CRD Complex at the Membrane

We employed the crystal structure (PDB code: 4G0N) of Raf-1 RBD interacting with Ras as a reference. Because the crystal structure was defined in solution, there was no information how Ras orients and localizes on the membrane. Recently, the relative membrane orientation of Ras was computed for the post-translationally modified K-Ras4B at the anionic lipid bilayer composed of DOPC:DOPS (4:1, mole ratio). To model K-Ras4B interacting with Raf-1 RBD at the membrane, we took two final conformations of the GTP-bound K-Ras4B from previous studies (Jang et al., 2016a) and superimposed the RBD-bound Ras crystal structure onto the membrane-bound K-Ras4B-GTP structures, generating two K-Ras4B/RBD conformations (model 1 & 2) at the membrane (Figure S1). Model 1 orients the catalytic domain with β2-β3 loop and α5 helix facing the bilayer surface, and model 2 presents the catalytic domain orientation with α2 and α3 helices interacting with the lipids. These models allow us to construct K-Ras4B/RBD-CRD complex at the membrane, since among (five) possible Ras membrane orientations reported in the literature (Abankwa et al., 2010; Gorfe et al., 2007; Jang et al., 2016a; Li and Buck, 2017; Mazhab-Jafari et al., 2015; Prakash et al., 2016), catalytic domain orientations with α3 and α4 helices interacting with the membrane or α4 and α5 helices facing toward the membrane, prevent Raf-1 CRD from contacting the membrane. Additionally, catalytic domain orientation with helix α2 and β1-3 strands facing toward the membrane induces steric clash with both RBD and CRD. Thus, modeling K-Ras4B with Raf-1 RBD-CRD intrinsically restricts the membrane orientation of K-Ras4B. In our simulations, models 1 and 2 represent K-Ras4B orientations at both ends of the spectrum, along the seesaw-like fluctuations when in complex with Raf-1 RBD-CRD. To model K-Ras4B interacting with Raf-1, the solution structure of Raf-1 CRD (PDB code: 1FAR (Mott et al., 1996)) was docked to the membrane-bound K-Ras4B/RBD conformations by using the ZDOCK docking program (Pierce et al., 2014). Owing to the short linker connecting CRD to RBD and the presence of the membrane, the location of CRD was limited to the space encompassed by Ras, RBD, and the membrane. For model 1 of K-Ras4B/RBD, we could obtain five dimeric configurations (DCs) of K-Ras4B/RBD-CRD, DCs 1 – 5. However, for model 2 we could only obtain two K-Ras4B/RBD-CRD complexes, DCs 6 and 7, due to steric clash of CRD with the membrane. A total of 7 K-Ras4B/RBD-CRD complexes were subjected to explicit all-atom MD simulations on the anionic DOPC:DOPS (4:1, mole ratio) membrane. For a comparison, two models of K-Ras4B/RBD without CRD were also simulated at the same membrane.

Atomistic Molecular Dynamics Simulations

We performed the MD simulations using the AMBER 11 package (Case et al., 2005) with the AMBER ff03 force field (Damjanović et al., 2009). Simulations were carried out following our previously published protocols (Lu et al., 2015, 2014, 2016b). For all systems, the initial configurations were subjected to 2000 steps of the steepest descent energy minimization, followed by 3000 steps of the conjugate gradient energy minimization with a positional restraint of 500 kcal/mol/Ų imposed on the heavy atoms of the proteins, ensuring that bad contacts in the solvated systems are removed. Subsequently, the entire system was minimized without any restraints. After minimization, each system was heated gradually from 0 K to 310 K within 300 ps. This was followed by constant temperature equilibration at 310 K for 700 ps, with a positional restraint of 10 kcal mol⁻¹ Å⁻² in the complex in a canonical NVT ensemble. Simulations were performed with periodic boundary conditions using the NPT ensemble. For the isolated CRD systems in water and lipid environments, production runs of 200 ns were performed. For both K-Ras4B/RBD and K-Ras4B/RBD-CRD systems, production runs of 300 ns were performed in the lipid bilayers. Langevin dynamics was used to maintain the temperature at 310 K with a collision frequency of 1 ps⁻¹ (Wu and Brooks, 2003). Langevin piston pressure control was used to sustain the pressre at 1 atm. Long-range electrostatic interactions were incorporated by using the particle mesh Ewald method (Darden et al., 1993). A local interaction distance of 10 Å was used for short-range electrostatics and vdW interactions. The SHAKE method was used to constrain the motion of bonds involving hydrogen atoms (Ryckaert et al., 1977). All production runs were performed using the NAMD parallel-computing code (Phillips et al., 2005) on a Biowulf cluster at the National Institutes of Health (Bethesda, MD).

QUANTIFICATION AND STATISTICAL ANALYSIS

In the analysis, the first 50-ns trajectories were removed, and thus averages were taken afterward. The simulated trajectories were analyzed using the Jupyter Notebook with Python package 3.3. The Jupyter Notebook application was used in the calculations for the probability distribution functions of the angle θ_Zn (Figure 2), the probability of lipid contacts for each basic residue in Raf-1 CRD (Figure 3), and the time series of RMSD, residue RMSF, and two dimensional DCCM of the residues motion across Raf-1 CRD (Figure S2 and S4). In Figure 3C, the binding energies of CRD with the membrane were calculated by molecular mechanics/generalized Born surface area (MM-GBSA) method using the AMBER package (Case et al., 2005). For dimeric K-Ras4B/Raf-1 systems, a series of coordinates in the pdb format from the Amber trajectory files were extracted and combined into the binary format trajectory files using the CHARMM programming package (Brooks et al., 2009) in the Biowulf PC/Linux cluster at the National Institutes of Health, Bethesda, MD (biowulf.nih.gov). In-house python codes combined with the CHARMM programming package were used to calculate the probability distribution functions across the lipid bilayer (Figure 5), the deviation from the bilayer surface for the selected backbone atoms of K-Ras4B and Raf-1 (Figure S5), the probability of lipid contact for the residues of K-Ras4B and Raf-1 (Figure 6), the probability distribution functions of the peptide order parameter (Figure S6), the probability distribution functions of the helix tilt (Figure 7), and the two-dimensional probability distributions of orientation angles for two vectors, $\overrightarrow{\text{CO}}$ and $\overrightarrow{\text{PO}}$, in GTP with respect to the bilayer (Figures 8 and S8). The contour maps in Figures 8 and S8 were generated by calculating the probability for the orientation angles of two vectors at each grid point in 5° interval.

KEY RESOURCES TABLE

REAGENT or RESOURCE	SOURCE	IDENTIFIER
Software and Algorithms
ZDOCK docking program	Pierce et al., 2014	http://zdock.umassmed.edu/
AMBER 11 package	Case et al., 2005	http://ambermd.org/
Jupyter Notebook	Open-source web application	http://jupyter.org/
Python 3.3	Python package	https://www.python.org/download/releases/3.3.0/
CHARMM programming package	Brooks et al., 2009	https://www.charmm.org/
NAMD 2.12	Phillips et al., 2005	http://www.ks.uiuc.edu/Research/namd/