Membrane-associated Ras dimers are isoform-specific: K-Ras dimers differ from H-Ras dimers

Abstract

Are the dimer structures of active Ras isoforms similar? This question is significant since Ras can activate its effectors as a monomer; however, as a dimer, it promotes Raf’s activation and MAPK (mitogen-activated protein kinase) cell signalling. In the present study, we model possible catalytic domain dimer interfaces of membrane-anchored GTP-bound K-Ras4B and H-Ras, and compare their conformations. The active helical dimers formed by the allosteric lobe are isoform-specific: K-Ras4B-GTP favours the α3 and α4 interface; H-Ras-GTP favours α4 and α5. Both isoforms also populate a stable β-sheet dimer interface formed by the effector lobe; a less stable β-sandwich interface is sustained by salt bridges of the β-sheet side chains. Raf’s high-affinity β-sheet interaction is promoted by the active helical interface. Collectively, Ras isoforms’ dimer conformations are not uniform; instead, the isoform-specific dimers reflect the favoured interactions of the HVRs (hypervariable regions) with cell membrane microdomains, biasing the effector-binding site orientations, thus isoform binding selectivity.

INTRODUCTION

The structures (and sequences) of the catalytic domains of the Ras isoforms are highly similar. There are over 170 crystal structures of the catalytic domains of H-Ras, N-Ras and K-Ras. Their RMSDs are under 0.70 Å (1 Å = 0.1 nm). Does this imply that so are the major states of the dimers that they form on the membrane? Even though never stated, this has largely been the underlying assumption in the community. Resolving this question has important functional ramifications. Evidence mounts that small membrane-associated Ras GTPase forms dimers [1–7]. Ras oligomerization and nanoclustering on the inner plasma membrane stimulates major cell signalling pathways [8–10]. Ras oligomerization promotes Raf’s dimerization, a prerequisite for its activation and initiation of downstream MAPK (mitogen-activated protein kinase) pathway signalling [11–15]. Structurally, it is intuitively understandable that the Raf side-to-side dimer, the active form of Raf [14], can competently and co-operatively interact with the Ras side-to-side dimer, with both proteins attached to the membrane. Only the GTP-bound, but not the GDP-bound, Ras is capable of forming dimers and oligomers [5,16]. Ras-GDP dimer states are less populated, and quench Raf’s activation [6]. The GTP-bound Ras dimers promote Raf association [1,17]; at the same time, Raf’s dimerization may also conceivably enhance Ras nanoclustering [18] through direct or mediated Ras interaction [19]. Ras activates signalling pathways through stimulation of effectors including Raf, PI3K (phosphoinositide 3-kinase), RalGDS (Ras-like guanine-nucleotide-dissociation stimulator), Ras and RIN1 (Rab interactor 1). It also regulates the RASSF (Ras association domain family) [20], which can link Ras activation and Hippo pathway signalling. The three major Ras pathways include Raf/MEK (MAPK kinase)/ERK (extracellular-signal-regulated kinase) pathway [21,22], PI3K/Akt/mTOR (mammalian target of rapamycin) pathway [23–25], and RalGDS/RalGEF (Ras-like guanine-nucleotide-exchange factor)/Ral pathway [26,27]. Among them, only the Raf/MEK/ERK pathway is regulated through Ras dimerization [5,6]; other Ras signalling pathways are activated by monomeric Ras [28]. This is due to Raf being the only effector which acts in a dimeric form.

There are three major Ras isoforms, H-Ras, N-Ras and K-Ras. K-Ras, which is the most frequently mutated Ras isoform in human cancer [29], has two splice variants, K-Ras4A and K-Ras4B. The protein structures and sequences of the catalytic domains of the isoforms are almost identical; however, their C-terminal membrane-associated HVRs (hypervariable regions) differ. Active oncogenic Ras anchors its HVR to the plasma membrane and interacts with its membrane-associated effectors to promote key cellular processes, including proliferation, survival and growth [30,31]. Where Ras associates with the distinct microdomains in the plasma membrane can determine the specific downstream signalling [32,33]. K-Ras favours association with non-raft regions [8,33,34], whereas H-Ras can be found in both lipid raft and non-raft regions [35–38]. The different affinities of the Ras isoforms to the membrane are mainly due to their post-translationally modified HVRs [39], which influence the orientations of the Ras–membrane associations [40]. Since the HVRs mainly act in membrane attachment, Ras proteins can form a dimer through the interactions between their catalytic domains. Importantly, in H-Ras and N-Ras, the HVRs may further directly mediate dimerization or nanoclustering [2,4,41]; by contrast, the high charge concentration of the HVR of K-Ras (+ 9 for K-Ras4B and + 6 for K-Ras4A) precludes direct electrostatically repulsed HVR–HVR interaction. Recently, we have unravelled the functional role of the HVR of K-Ras4B [42,43]. We discovered that it can be sequestered by the effector-binding region of the catalytic domain of GDP-bound K-Ras4B, a conformation referred to as the ‘autoinhibited state’. The autoinhibition is released when the catalytic domain is GTP-bound. At the membrane, the HVR autoinhibition persists in the inactive state, blocking the effector-binding site, whereas, in the active state, the catalytic domain is released and fluctuates reinlessly, exposing the effector-binding site [40]. MD simulations supported by NMR data suggested that oncogenic GDP- and GTP-bound K-Ras4B shift the equilibrium towards such an effector-binding-site-exposed state [43].

In the present study, we question whether the Ras dimer conformations are universal or isoform-specific. This has important potential functional as well as pharmacological consequences. PRISM, a powerful knowledge-based structural prediction algorithm provided four possible dimer interfaces for the catalytic domains of K-Ras4B-GTP and similar ones for H-Ras. Through comprehensive explicit solvent MD simulations of the dimers in different states [on the membrane surface with post-translationally modified HVRs, with parameterized farnesyl (for K-Ras4B and H-Ras), and palmitoyl (H-Ras), as well as in solution], we evaluate the interfaces for K-Ras4B-GTP and H-Ras-GTP. The results show similarities and, importantly, differences between the isoforms. The K-Ras4B-GTP dimer favours the allosteric lobe dimer interface involving α3 and α4 helices, whereas the H-Ras-GTP dimer relatively stabilizes the helical interface through the α4 and α5 helices interaction. These allosteric lobe dimer interfaces reflect active Ras dimer formation, since the effector-binding sites are accessible for the Raf associations, promoting Raf dimerization. Both K-Ras4B-GTP and H-Ras-GTP stabilize the effector lobe dimer interface which has a shifted β-sheet extension. This most stable Ras dimer structure is exactly shared with the Raf–Ras association, thus it constitutes an inactive dimer interface. We observe further that both K-Ras4B-GTP and H-Ras-GTP form a β-sandwich involving β1, β2 and β3 strands. The inactive Ras dimer interface at the effector lobe also overlaps other effector-binding sites, thus deterring their conjugation. However, Ras effectors such as Raf and PI3K, and Hippo regulator RASSF, can easily outcompete Ras for the effector lobe interface, suggesting that Ras dimer formation with multiple interfaces is dynamic in membrane nanoclusters. Taken together, the varied HVR sequences, prenylated states and environments, lead to Ras populating different dimerization states with altered preferred membrane interactions, suggesting that Ras dimerization is highly isoform-specific. We conclude that there is no generic preferred dimer organization for all Ras isoforms. The modes of associations are isoform-dependent, pointing to distinct dimer interface pharmacology.

MATERIALS AND METHODS

Atomistic MD simulations

Farnesylated and non-farnesylated wild-type K-Ras4B_1–185 proteins were previously constructed using a crystal structure (PDB code 3GFT) [40,42,43]. Ensembles of monomeric structure of GTP-bound K-Ras4B were extracted from the trajectories of K-Ras4B-GTP monomer simulations in solution [40]. These monomer conformations were used to construct the dimer structures of K-Ras4B-GTP. Four different types of K-Ras4B-GTP dimers based on earlier predictions for the K-Ras4B catalytic domain dimers [5] contain two allosteric lobe and two effector lobe dimer interfaces. For the GTP-bound H-Ras dimers, we followed the protocol of the K-Ras4B-GTP dimer simulations [40,42,43]. The crystal structure of H-Ras-GTP (PBD code 5P21) served as the catalytic domain structure. Preliminary simulations of H-Ras-GTP with the HVR containing the palmitoyl and farnesyl modifications generated ensembles of monomeric H-Ras structures in solution. Four H-Ras-GTP dimers with different dimer interfaces were also constructed on the basis of the predictions [5]. The initial configurations of both K-Ras4B-GTP and H-Ras-GTP dimers were subject to simulations in water and membrane environments. For the water simulations, the initially pre-assembled dimer configurations were solvated by the modified TIP3P water model [44] and gradually relaxed with the proteins held rigid. The unit cell box of 120 ų contains almost 180000 atoms, 30 Na⁺, 3 Mg²⁺ and 38 Cl⁻ for the K-Ras4B-GTP dimers and 40 Na⁺, 4 Mg²⁺ and 34 Cl⁻ for the H-Ras-GTP dimers. For the membrane simulations, the same initial configurations of both dimers for the water simulations were translated on to the surface of an anionic lipid bilayer containing DOPC (1,2-dioleoyl-sn-glycero-3-phosphocholine) and DOPS (1,2-dioleoyl-sn-glycero-3-phosphoserine) (molar ratio 4:1). In the initial construction of the bilayer system, no portion of the catalytic domain and the HVR including the farnesyl and palmitoyl groups was inserted into the lipid bilayer before the start of the simulations. Instead, the HVR backbone marginally touched the surface of the anionic bilayer at the starting points. The lipid bilayers were generated using the bilayer-building protocol involving the interactions of pseudospheres through the vdW (van der Waals) force field [34,40,45–51]. A unit cell containing a total of 400 lipids (320 DOPC and 80 DOPS) constitutes the bilayer with TIP3P waters, added at both sides with a lipid/water ratio of ∼1:122. The updated CHARMM [52] all-atom additive force field for lipids (C36) [53] was used to construct the set of starting points and to relax the systems to a production-ready stage. To neutralize the bilayer system and to also obtain a salt concentration near 100 mM, 80 Na⁺,8 Mg²⁺ and 18 Cl⁻ for the K-Ras4B-GTP dimer and 90 Na⁺,9 Mg²⁺ and 14 Cl⁻ for the H-Ras-GTP dimer were added to the bilayer systems.

A total of 3.2 μs simulations were performed for the 16 systems; each has 200 ns simulation with an integration step of 2 fs and with a constant temperature of 310 K. Our water simulations employed the NPT (constant number of atoms, pressure and temperature) ensemble. For the membrane simulations to 30 ns, we employed the NPAT (constant number of atoms, pressure, surface and temperature) ensemble with a constant normal pressure applied in the direction perpendicular to the membrane. In the production runs from 30 ns to 200 ns, the simulations employed the NPT ensemble. We used the NAMD parallel computing code [54] for the simulations on a Biowulf cluster at the NIH. Averages were taken after 30 ns, discarding initial transients trajectories, with the CHARMM programming package [52].

CHARMM parameters for the farnesylated and palmitoylated cysteine residues

To simulate Ras with the PTMs (post-translational modifications) at the HVR, the CHARMM parameters for the prenyls groups are required. The PTMs include the farnesylation at the Cys¹⁸⁵ side chain for K-Ras4B, and the farnesylation at the Cys¹⁸⁶ side chain and two palmitoylations at the Cys¹⁸¹ and Cys¹⁸⁴ side chains for H-Ras (Figure 1). Both Ras proteins have a methylated C-terminus at the carboxy group. A molecular topology was created for the farnesyl and palmitoyl using the Avogadro software [55]. Parameters including partial charges, bond lengths, angles and torsional angles for the atoms in the farnesyl and palmitoyl groups were calculated using the Gaussian09 program on a Biowulf cluster at the NIH. The calculated parameters can be directly adopted in the CHARMM [52] program. In the topology file, we generated two new amino acid residues, CYF and CYP, representing the farnesylated and palmitoylated cysteine residues respectively. The vdW parameters for aliphatic carbon (CAPL) were adopted from the retinol with protonated Schiff base [47,56,57]. For the saturated carbon chain in the palmitoyl group, the standard CHARMM parameters for the palmitate residue were adopted. The CHARMM stream files containing molecular topology and parameter for the CYF (see Appendix I of the Supplementary Online Data) and the CYP residue (see Appendix II of the Supplementary Online Data) were employed in the dimer simulations in the water and membrane environments.

Figure 1. Sequences and structures of K-Ras4B and H-Ras.

(A) Sequences of the K-Ras4B and H-Ras proteins. In the sequence, hydrophobic, polar/glycine, positively charged and negatively charged residues are coloured black, green, blue and red respectively. Underlined residues denote differences between the isoforms. HVRs are residues 167–188 for K-Ras4B and 167–189 for H-Ras. The cartoons represent the structures of GTP-bound K-Ras4B_1–185 (left) and H-Ras_1–186 (right) with the PTMs. The catalytic domain structures were adopted from the crystal structures of the GTP-bound K-Ras4B (PDB code 3GFT) and H-Ras (PBD code 5P21). The different catalytic domain residues between the isoforms are marked on the structures. The HVRs are covalently connected to their catalytic domains. (B and C) Topology diagrams of the (B) farnesylated and (C) palmitoylated cysteine residues.

Binding free energy calculation for the Ras dimers

To calculate the binding free energy of Ras dimers, we used the molecular mechanics energies combined with the Poisson–Boltzmann surface area continuum solvation (MM-PBSA) method [58,59]. The average binding free energy is calculated as a sum of the gas phase contribution, the solvation energy contribution and the entropic contribution:

$$\langle \mathrm{\Delta }{G}_{\text{b}}\rangle =\langle \mathrm{\Delta }{G}_{\text{gas}}\rangle +\langle \mathrm{\Delta }{G}_{\text{sol}}\rangle -T\mathrm{\Delta }S$$ (1)

where angle bracket denotes an average along the MD trajectory. The gas phase contribution to the binding free energy is a sum of the internal energy, the vdW interaction, and the electrostatic energy, ΔG_gas = ΔH_intra + ΔH_vdW + ΔH_elec. The solvation energy contribution can be divided into the electrostatic contribution and the non-polar contribution, $\mathrm{\Delta }{G}_{\text{sol}}=\mathrm{\Delta }{G}_{\text{sol}}^{\text{elec}}+\mathrm{\Delta }{G}_{\text{sol}}^{\text{non–polar}}$. The electrostatic contribution to solvation is calculated by solving the linear Poisson Boltzmann equation within the CHARMM program [52]. The non-polar contribution is obtained using an equation, $G_{\text{sol}}^{\text{non–polar}}=\gamma SASA+\beta$, where SASA denotes the solvent-accessible surface area with a set of the surface tension parameters γ = 0.00542 and the constant β = 0.92 [60]. The entropy term is divided into the translational, rotational, and vibrational contributions, TΔS = TΔS_trans + TΔS_rot + TΔS_vib. The translational and rotational entropies are evaluated from the calculation of principal moment of inertia, and the vibrational entropy is obtained from the quasiharmonic mode calculation in the VIBRAN module of the CHARMM program [52]. The number of vectors (mode) to calculate the vibrational analysis is set to NATOM×3, where NATOM denotes the number of atoms. Finally, the change in binding free energy due to the dimerization is calculated using the equation:

$$\mathrm{\Delta }{G}_{\text{b}}=G_{\text{b}}^{\text{dimer}}-\left(G_{\text{b}}^{\text{monomer1}}+G_{\text{b}}^{\text{monomer2}}\right)$$ (2)

Born radii for the atoms in the farnesyl and palmitoyl groups

In the calculation of the electrostatic contribution to solvation, we solved the linear Poisson–Boltzmann equation using a set of atomic Born radii defining the dielectric boundary in continuum electrostatic calculations. However, a set of atomic Born radii only provided for the 20 standard amino acids [61] and nucleic acid atoms [62]. To obtain Born radii for the atoms in the prenyl groups, we performed preliminary simulations for the farnesylated and palmitoylated cysteines in water. Then we calculated averaged radial solvent charge distribution function as described in [61] for all atoms except hydrogen in the prenyl groups. The calculated Born radii for the atoms are summarized in Supplementary Tables S1 and S2.

RESULTS

Predictions of K-Ras4B dimer structures reveal four possible dimer interfaces

We have recently predicted K-Ras4B dimer structures [5] using a powerful template-based protein–protein complex structure prediction algorithm (PRISM) [63,64]. We observed that K-Ras4B_1–180 in the GTP-bound state can form homodimers through the allosteric and effector lobe interfaces (Figure 2). We reported two major dimer interfaces, including allosteric helices α3 and α4, which we hereinafter denote the allosteric lobe dimer interface 1 (ALDI-1), and β2 strands aligned in a shifted β-sheet extension, which we refer to as effector lobe dimer interface 1 (ELDI-1). In addition, we examined two minor interfaces involving allosteric helices α4 and α5 (ALDI-2) and exact alignment of β2 strands (ELDI-2). Our predictions suggested that the GTP-bound, but not GDP-bound, K-Ras4B catalytic domains form stable dimers. Further investigation of the crystal interfaces of the Ras dimers revealed that H-Ras in the GTP-bound state has similar dimer interfaces to those predicted for K-Ras4B-GTP (Supplementary Figure S1). Note that all predicted Ras dimers were catalytic-domain-only structures; the HVR was absent from those dimer structures. The predicted dimer structures were supported by NMR chemical shift perturbations [5].

Figure 2. Predicted structures of K-Ras4B dimer.

PRISM predicted four possible GTP-bound K-Ras4B dimers for the Ras catalytic domains. Two dimer interfaces were found at the allosteric lobe (ALDI-1 and ALDI-2), and two others were located at the effector lobe (ELDI-1 and ELDI-2). The protein domains involved in the interfaces are highlighted by different colours and the remaining domains are shown as white.

Relative stability of Ras dimers at the phospholipid bilayer

To validate the Ras dimer structures in the lipid environment in which the HVR plays a critical functional role, we performed MD simulations on Ras dimers at an anionic lipid bilayer composed of DOPC/DOPS (4:1 molar ratio). We studied full-length post-translationally modified K-Ras4B and H-Ras in the GTP-bound state. The PTMs involve both prenylation and methylation that occur on cysteine residues at the C-terminal end of the HVR. In the K-Ras4B HVR, the Cys¹⁸⁵ side chain is modified with the farnesyl group. In the H-Ras HVR, in addition to the farnesyl modification on Cys¹⁸⁶, Cys¹⁸¹ and Cys¹⁸⁴ are modified with a palmitoyl group. These prenyl modifications are intrinsic features of the membrane-associated Ras isoforms. In the initial modelling, the Ras dimers were pre-assembled through the predicted dimer interfaces and oriented at the membrane in a way that the HVR backbone marginally touched the surface of the anionic lipid bilayer. No portion of the protein including the HVR and the prenyl groups was inserted into the lipid bilayer before the start of the simulations. During the course of the simulations, we observed that for both K-Ras4B-GTP and H-Ras-GTP dimers, the HVR is a major driver for the interaction with the membrane, whereas the catalytic domain mainly participates in the protein–protein interactions (Figure 3). The centre of mass of each catalytic domain in the dimer is located ∼3.0 nm from the bilayer surface, and for the HVR it is located ∼1.0 nm from the surface (Supplementary Figure S2). These locations are consistent with those obtained for monomeric Ras in the active state at the membrane [40]. Both the farnesyl and palmitoyl groups spontaneously insert into the hydrophobic core of the lipid bilayer, suggesting that Ras proteins owe their membrane-anchoring ability to the prenyl modifications. We note that a farnesyl group from one monomer of the H-Ras dimer with ELDI-1 failed to insert into the membrane (Supplementary Figure S2), indicating that, unlike the palmitoyl group, which is a saturated fatty acid chain, farnesyl insertion is reversible due to the intrinsic cis conformation in the unsaturated carbon chain [65].

Figure 3. Membrane orientation and anchoring of Ras isoform dimers.

Snapshots representing relaxed structures of GTP-bound (A) K-Ras4B and (B) H-Ras with the PTMs on the anionic lipid bilayer composed of DOPC/DOPS (molar ratio 4:1). Both allosteric (ALDI) and effector (ELDI) dimer interfaces are shown. In the catalytic domain, the red sticks and green spheres represent GTP and Mg²⁺ respectively. The HVRs in tube representation are coloured deep blue for K-Ras4B and light green for H-Ras. The farnesyl and palmitoyl as sticks are coloured yellow and orange respectively. For the lipid bilayer, the white surface denotes DOPC and the grey surface represents DOPS. The Figure illustrates that in both K-Ras4B-GTP and H-Ras-GTP dimers, the HVR is a major driver for the interaction with the membrane, whereas the catalytic domain mainly participates in the protein–protein interactions.

To corroborate the Ras dimer interfaces, we closely examined the simulated Ras dimer structures at the lipid bilayer. The K-Ras4B-GTP dimers with ALDI-1 and ELDI-1 preserve their initial dimer interfaces, which is not the case for the initial dimer interfaces in the ALDI-2 and ELDI-2 dimers, whose conformations drift (Figure 4 and Supplementary Figure S3). Although a symmetric α3-α4/α3-α4 interface in the ALDI-1 dimer converts during the simulations into an asymmetric α3-α4/α3 interface, the salt bridge interactions of Agr⁷³, Glu⁹¹, Arg⁹⁷, Glu⁹⁸, Lys¹⁰¹ Arg¹⁰², Asp¹⁰⁵, Glu¹⁰⁷ and Arg¹³⁵ and the π -stacking by His⁹⁴, His⁹⁵ and Tyr¹³⁷ strongly retain the dimer interface (Table 1). In contrast, the K-Ras4B-GTP dimer with ALDI-2 involving the α4 and α5 allosteric lobe helices is not stable; each monomer significantly shifts away from the other. A salt bridge between Asp¹⁵³ and Arg¹⁶¹ marginally holds the interface. For the β-sheet interfaces in the effector lobe, we found that the K-Ras4B-GTP dimer with ELDI-1 containing a shifted β-sheet extension holds the most stable dimeric interface. In addition to the intermolecular backbone hydrogen bonds formed by Glu³⁷, Ser³⁹ and Arg⁴¹, the dimer interface is further stabilized by the side-chain interactions, the salt-bridge interactions of Asp³³, Lys⁴², Glu³⁷ and Arg⁴¹, the π -stacking by His²⁷ and Tyr⁴⁰, and the hydrophilic interaction of Gln²⁵. Interestingly, the exact β-sheet alignment in the ELDI-2 dimer evolves into the β-sandwich motif formed by β1, β2 and β3 strands. The intermolecular hydrogen bonds formed by the Asp³⁸ and Gln⁴³ side chains and the salt bridge interactions of Glu³, Lys⁵, Arg⁴¹ and Asp⁵⁴ sustain the β-sandwich dimeric interface.

Figure 4. Relaxed K-Ras4B dimers at the membrane.

Snapshots representing average conformations of GTP-bound K-Ras4B with the PTMs on the anionic lipid bilayer composed of DOPC/DOPS (molar ratio 4:1). The K-Ras4B-GTP dimers with different dimer interfaces are shown (left). Lipids and water are removed for clarity. In the catalytic domain, the red sticks and green spheres represent GTP and Mg²⁺ respectively. The HVRs in tube representation are coloured deep blue, and the farnesyl as a stick is coloured yellow. Highlighted dimer interfaces are shown (right). In the highlight, yellow dotted lines denote salt bridges and red dotted lines indicate the intermolecular backbone hydrogen bonds. Residues involving the hydrophilic and hydrophobic interactions are also marked. The Figure illustrates that K-Ras4B-GTP dimers with ALDI-1 and ELDI-1 retain these dimer interfaces, unlike ALDI-2 and ELDI-2 dimers, whose conformations drift. Even though the symmetric α3-α4/α3-α4 interface in the ALDI-1 dimer shifts to an asymmetric α3-α4/α3 interface, the salt-bridge interactions and the π-stacking shown on the right still strongly sustain the dimer interface during the simulations, which is not the case for ALDI-2 involving the α4 and α5 allosteric lobe helices where the monomers drift away. The ELDI-1-shifted β-sheet interface is stable, whereas the exact β-sheet alignment in the ELDI-2 dimer evolves into the β-sandwich motif.

Table 1. Types of atomic interactions in the GTP-bound K-Ras4B homodimer at the membrane.

The atomic pair indicates the monomer 1/monomer 2 pair in the interface, represented in the order of ‘atom:residue:domain’. The number in parentheses denotes the percentage of the atomic pair interaction based on the distance between the paired atoms.

	Salt bridge, M1/M2 pair	Hydrophilic interaction, M1/M2 pair	Hydrogen bond, M1/M2 pair
ALDI-1 dimer	NH1:Arg97:α3/OE1:Glu98:α3 (100)	OH:Tyr137:α4/ND1:His94:α3 (50)
	OE2:Glu98:α3/NH1:Arg97:α3 (50)	OH:Tyr137:α4/ND1:His95:α3 (60)
	OE1:Glu98:α3/NZ:Lys101:α3 (100)
	NZ:Lys101:α3/OE2:Glu98:α3 (90)
	NH1:Arg102:α3/OD2:Asp105:L7 (90)
	NH1:Arg102:α3/OE1:Glu107:L7 (30)
	OE1:Glu107:L7/NH1:Arg102:α3 (80)
	OD2:Asp105:L7/NH1:Arg73:SII (40)
	OD2:Asp105:L7/NH2:Arg102:α3 (50)
	NH1:Arg135:α4/OE1:Glu91:α3 (90)
ALDI-2 dimer	NH1:Arg161 :α5/OD2:Asp153:a5 (60)	NE2:Gln150:α5/OG1:Thr50:β3 (40)
ELDI-1 dimer	OD1:Asp33:SI/NZ:Lys42:β 2(90)	NE2:Gln25:α2/OE1:Gln25:a2 (90)	O:Glu37:SI/N:Arg41:β2 (100)
	OE2:Glu37:SI/NH1:Arg41:β2 (100)	CE1:His27:α2/CE1:His27:a2 (80)	N:Ser39:β2/O:Ser39:β2 (100)
	NZ:Lys42:β2/OD1:Asp33:SI (100)	OH:Tyr40:β2/NE2:Gln25:a2 (60)	O:Ser39:β2/N:Ser39:β2 (100)
		CE2:Tyr40:β2/OH:Tyr40:β 2(100)	N:Arg41:β2/O:Glu37:SI(60)
ELDI-2 dimer	OE1:Glu3:β1/NZ:Lys5:β1 (20)		OD1:Asp38:β2/NE2:Gln43:β 2(60)
	NZ:Lys5:β1/OE1:Glu3:β1 (30)		NE2:Gln43:β2/OD1:Asp38:β 2(40)
	NH2:Arg41:β2/OD2:Asp54:β3 (70)
	OD1:Asp54:β3/NH2:Arg41:β2 (70)

The relative strengths of the dimer interfaces at the lipid bilayer of the H-Ras-GTP dimers differ from those of the K-Ras4B-GTP dimers. These differences can be found at the allosteric lobe helical interfaces (Figure 5 and Supplementary Figure S4). Unlike K-Ras4B-GTP, the H-Ras-GTP dimer with ALDI-1 is not stable; the H-Ras monomers in the dimer fall apart during the simulations. Complete separation of the dimer is prevented by the transient side-chain interactions involving the salt-bridge interactions of Glu⁶², Glu⁹¹, Arg¹²⁸ and Arg¹³⁵ and the cation–π interaction between Tyr⁶⁴ and Arg₁₃₅ (Table 2). A glimpse of the Tyr⁶⁴ interaction in the interface indicates that a new dimerization interface involving Tyr⁶⁴ can be possible in the H-Ras-GTP dimers at the H-Ras-specific lipid environments [4]. In contrast, the H-Ras-GTP dimer with ALDI-2 preserves the predicted initial dimer interface. ALDI-2 is stabilized by the salt-bridge interactions of Asp¹⁵⁴ and Arg¹⁶¹ in the α5 helix and the hydrophilic interaction of Gln¹³¹ and Gln¹⁶⁵ in the α4 and α5 helices. Furthermore, the HVR residue Lys¹⁶⁷ forms a salt bridge with Asp¹³² in α4. For the β-sheet interfaces, we observed that the ELDI-1 dimer marginally preserves the shifted β-sheet extension through the intermolecular backbone hydrogen bonds formed by Ser³⁹ in the β2 strands and the hydrophobic interactions of Ile³⁶ and Leu⁵². However, there are fewer residues interacting in the dimer interface of the ELDI-1 H-Ras-GTP dimer compared with the K-Ras4B-GTP dimer. With less supportive interactions, the shifted β-sheet extension seems to be less stable in the H-Ras-GTP dimer. Consistent with the ELDI-2 K-Ras4B-GTP dimer, the exact β-sheet alignment in the ELDI-2 H-Ras-GTP dimer is also converted into a β-sandwich, and the interface is stabilized by the salt-bridge interactions of Glu³, Lys⁵, Glu³⁷, Asp³⁸, Arg⁴¹, Lys⁴² and Asp⁵⁴.

Figure 5. Relaxed H-Ras dimers at the membrane.

Snapshots representing average conformations of GTP-bound H-Ras with the PTMs on the anionic lipid bilayer composed of DOPC/DOPS (molar ratio 4:1). The H-Ras-GTP dimers with different dimer interfaces are shown (left). Lipids and water are removed for clarity. In the catalytic domain, the red sticks and green spheres represent GTP and Mg²⁺ respectively. The HVRs in tube representation are coloured light green, and the farnesyl and palmitoyl as sticks are coloured yellow and orange respectively. Highlighted dimer interfaces are shown (right). In the highlight, yellow dotted lines denote salt bridges and red dotted lines indicate the intermolecular backbone hydrogen bonds. Residues involving the hydrophilic and hydrophobic interactions are also marked. The Figure illustrates that the behaviour of the dimer interfaces at the lipid bilayer of the H-Ras-GTP dimers differ from those of the K-Ras4B-GTP dimers. Unlike K-Ras4B-GTP, the H-Ras-GTP dimer with ALDI-1 is not stable; the H-Ras monomers in the dimer fall apart during the simulations.

Table 2. Types of atomic interactions in the GTP-bound H-Ras homodimer at the membrane.

The atomic pair indicates the monomer 1/monomer 2 pair in the interface, represented in the order of ‘atom:residue:domain’. The number in parentheses denotes the percentage of the atomic pair interaction based on the distance between the paired atoms.

	Salt bridge, M1/M2 pair	Hydrogen bond, M1/M2 pair	Cation-π interaction, M1/M2 pair	Hydrophilic interaction, M1/M2 pair	Hydrophobic interaction, M1/M2 pair
ALDI-1 dimer	OE2:Glu91:α3/NH1:Arg128:α4 (40)		OH:Tyr64:SII/NH2:Arg135:α4 (20)
	OE2:Glu62:SII/NH2:Arg135:α4 (30)
ALDI-2 dimer	OD2:Asp154:α5/NH1:Arg161 :α5 (90)			NE2:Gln131:α4/OE1:Gln165:α5 (80)
	NH1:Arg161:α5/OD2:Asp154:α5 (90)			OE1:Gln165:α5/NE2:Gln131:α4 (90)
	NZ:Lys167:HVR/OD2:Asp132:α4 (90)
ELDI-1 dimer		N:Ser39:β2/O:Ser39:β2 (90)			CD:IIe36:SI/CD1:Leu52:β3 (60)
		O:Ser39:β2/N:Ser39:β2 (80)			CD1:Leu52:β3/CD:Ile36:SI (90)
ELDI-2 dimer	OE1:Glu3:β1/NZ:Lys5:β1 (70)
	OD2:Asp38:β 2/NH1:Arg41:β 2(80)
	NH1:Arg41:β2/OE2:Glu3:β1 (20)
	NH1:Arg41:β2/OD1:Asp54:β 3(80)
	NZ:Lys42:β2/OE2:Glu37:SI (70)
	OD2:Asp54:β 3/NH2:Arg41:β 2(40)

Binding free energy for Ras dimers at the phospholipid bilayer

To quantify the Ras dimer interfaces, we calculated the binding free energy of the dimerization of both K-Ras4B and H-Ras in the GTP-bound state using molecular mechanics energies combined with the Poisson–Boltzmann surface area continuum solvation (MM-PBSA) method [58,59]. Since Ras is extramembranous and only the prenyl groups are embedded in the lipid bilayer, the membrane contribution to the solvation free energy was neglected in the binding free energy calculation. Table 3 summarizes the binding free energy, ΔG_b, which is a sum of the gas phase contribution, ΔG_gas, the solvation free energy, ΔG_sol, and the entropic contribution, − TΔS. The details of the decomposition of each term are also summarized in Supplementary Tables S3–S10. For the K-Ras4B-GTP dimers, we observed that the ALDI-1 and ELDI-1 dimers yield low values of the binding free energy: −3.67±41.23 and −8.89±68.87 kcal/mol (1 kcal = 4.184 kJ) respectively. These dimer interfaces were defined as the major interfaces in our previous predictions for the catalytic domain of the K-Ras4B-GTP dimers [5]. It can be seen that the interaction between catalytic domains largely contributes to the binding free energy of the dimerization (Supplementary Figure S5A). The HVR–catalytic domain and HVR–HVR interactions also contribute to the binding free energy, but the HVR interactions are transient, because the K-Ras4B HVR is basically involved in the interaction with the lipids. High values of the binding free energy suggest that the ALDI-2 and ELDI-2 K-Ras4B-GTP dimers are less stable. For the H-Ras-GTP dimers, we found that the ELDI-1 dimer shows low binding free energy of 0.63±82.31 kcal/mol, although the interaction between the catalytic domains is not so strong (Supplementary Figure S5B). The shifted β-sheet extension of the H-Ras-GTP dimer gradually decays due to the interaction of the released farnesyl (of monomer 2) from the bilayer with the catalytic domain (of monomer 1), which hampers the stability of the β-sheet dimer interface (Figure 5; Supplementary Figure S2B). In contrast, the exact β-sheet alignment in the ELDI-2 dimer is immediately destroyed and the β-sandwich motif emerges in less than 10 ns (Supplementary Figure S4). The vdW attraction between HVR backbones largely contributes to the binding free energy for the dimerization. For the allosteric lobe interfaces, the high binding free energy for the dimerization suggests that the ALDI-1 dimer is not stable, whereas, with slightly low binding free energy, H-Ras-GTP may form a dimer through the α4 and α5 helical interface.

Table 3. Binding free energy for the Ras dimerization at the membrane.

Ras dimer	Interface	(ΔGgas) (kcal/mol)	(ΔGsol) (kcal/mol)	−TΔS(kcal/mol)	(ΔGb) (kcal/mol)
K-Ras4B-GTP at DOPC/DOPS (4:1)	ALDI-1	−328.65±99.16	253.78±71.48	71.20	−3.67±41.23
	ALDI-2	−132.71±30.95	129.19±36.53	80.53	77.01±20.39
	ELDI-1	−1424.06±157.98	1340.33±138.94	74.84	−8.89±68.87
	ELDI-2	−1089.09±144.13	1076.43±139.42	78.92	66.26±21.45
H-Ras-GTP at DOPC/DOPS (4:1)	ALDI-1	238.63±36.46	−250.52±87.11	83.96	72.07±35.71
	ALDI-2	−45.80±50.60	−15.39±72.24	75.46	14.27±58.0
	ELDI-1	103.15±79.46	−178.27±102.11	75.75	0.63±82.31
	ELDI-2	−145.50±115.22	76.95±100.93	82.99	14.44±23.97

Ras dimer formation in solution

It is hard to imagine that Ras can form a stable dimer in an aqueous environment, since the highly flexible HVRs can interfere with the interaction between the catalytic domains, causing a large free energy barrier for the dimerization. To investigate the effects of the HVR and membrane absence on the dimerization, we simulated both pre-assembled K-Ras4B-GTP and H-Ras-GTP dimers in a bulk water environment using the same dimer interfaces from the membrane simulations (Figure 6). For the K-Ras4B-GTP, we observed that the dimers, which are detached from the membrane, evolve slightly different interfaces. The ALDI-1 dimer seems to align the initial interface, but it fails to tighten it. The ALDI-2 dimer is significantly distorted from the initial structure as observed in the bilayer simulation. The high binding free energy values suggest that both allosteric lobe interfaces are not stable (Supplementary Table S11). For the effector lobe interfaces, the ELDI-1 dimer preserves the shifted β-sheet interface, and the ELDI-2 dimer produces the β-sandwich interface, both similar to those dimers observed at the membrane. However, due to the absence of the membrane, the HVR rapidly interacts with each binding partner. In the ELDI-2 dimer, the farnesyl groups form a hydrophobic cluster (Supplementary Figure S6A), favourably contributing to the binding free energy. In addition, strong electrostatic interactions due to the polybasic K-Ras4B HVR can be observed in the HVR interactions with the binding partner (Supplementary Figure S7A), suggesting that the HVR may be an important factor for the K-Ras4B-GTP dimerization in solution. For the H-Ras-GTP dimers, there are apparent hydrophobic clusters formed by both farnesyl and palmitoyl groups in both ELDI-1 and ELDI-2 dimers (Supplementary Figure S6B). The collapsed HVR backbones also participate in the hydrophobic clusters. The interaction between the catalytic domains is very weak (Supplementary Figure S7B), indicating that the HVR hydrophobic interactions mainly reflect the binding free energy for the dimerization. Similar to the K-Ras4B-GTP dimers, both helical interfaces, ALDI-1 and ALDI-2 are not stable in the H-Ras-GTP dimer, suggesting that the helical interfaces are not populated for both K-Ras4B-GTP and H-Ras-GTP dimers in solution. We conclude that Raf’s dimerization is unlikely to be promoted in the absence of the membrane.

Figure 6. Relaxed Ras isoform dimers in solution.

Snapshots representing average conformations of GTP-bound (A) K-Ras4B and (B) H-Ras with the PTMs in solution. In the catalytic domain, the red sticks and green spheres represent GTP and Mg²⁺ respectively. The HVRs in tube representation are coloured deep blue for K-Ras4B and light green for H-Ras. The farnesyl and palmitoyl as sticks are coloured yellow and orange respectively. Interface domains are marked. The Figure illustrates that the highly flexible HVRs can hamper the interaction between the catalytic domains. The resulting large free energy barrier for the dimerization argues that it is unlikely that Ras will form a stable dimer in an aqueous environment.

DISCUSSION

In the present study, we evaluate four possible Ras dimer interfaces [5] using explicit MD simulations on both GTP-bound K-Ras4B and H-Ras dimers at the anionic lipid bilayer and in solution. We observe that the stability of the Ras dimers is isoform-specific, controlled by their distinct HVRs. The K-Ras4B-GTP dimer can be stabilized through the interaction between their catalytic domains, whereas the dimerization of H-Ras-GTP is supported by HVR–HVR and HVR–catalytic domain interactions. To activate Raf, two adjacent Ras proteins have to align their allosteric lobes, which allows exposure of their effector-binding sites, recruitment of Raf and its dimerization. Ras association through the allosteric lobe dimer interface is a prerequisite for Raf activation. We observed two possible allosteric lobe dimer interfaces whose relative strength differs among Ras isoforms.

With the allosteric lobe dimer interface involving α3 and α4 helices (ALDI-1), K-Ras4B-GTP is able to form a dimer in both lipid and water environments. This K-Ras4B-GTP dimer favours an asymmetric α3-α4/α3 helical interface at the membrane. In solution, although the interface is not very tight, the symmetric helix alignment of α3-α4/α3-α4 is preserved. The K-Ras4B-GTP dimer can promote Raf dimerization, since the effector-binding sites for Raf are exposed. The transient Ras alignment through the allosteric lobe dimer interface can be stabilized further by Raf dimerization and its interaction with the membrane (Figure 7). In contrast, the H-Ras-GTP dimer fails to align the α3-α4/α3-α4 helical interface at the membrane. Fluctuations in the interaction of the H-Ras-GTP HVR with the anionic lipids may interfere with the association of the catalytic domains at the interface. However, in solution, the H-Ras-GTP dimer marginally holds the symmetric alignment of α3 and α4 helices, and the HVR collapses on itself forming a hydrophobic cluster by the prenyl groups. We speculate that the H-Ras-GTP dimer can be populated when the dimer is translated into the H-Ras-specific lipid environment, such as lipid raft containing saturated lipid chains. An HVR engaged in membrane attachment would reduce the fluctuations and thus promote the interaction between the catalytic domains and the interface.

Figure 7. The two types of Ras dimers and the outcome for Raf dimerization and signalling.

A schematic diagram conceptualizing the results of the present study. The diagram illustrates the formation of Ras dimers via the allosteric lobe (AL) and effector lobe (EL) interfaces and how they can relate to Raf’s dimerization and MAPK signalling. The allosteric lobe interface is helical; the effector lobe interface consists of β-structures (mostly β-sheet extension). The multiple dimer states point to highly dynamic associations. The allosteric lobe interfaces are isoform-specific (α3/4 in K-Ras4B; α4/5 in H-Ras). The different interactions lead to altered orientations with respect to the membrane and thus altered exposure of the effector-binding site (not shown). The allosteric lobe Ras dimer promotes Raf dimerization, thus it is the active interface. Formation of the Raf dimer further optimizes the Ras dimer interface. The Ras–Raf dimeric complex can initiate the MAPK pathway. The effector lobe interface overlaps Raf’s binding site; however, Raf’s nanomolar affinity outcompetes Ras dimerization via this interface. RBD is Raf’s Ras-binding domain; CRD is Raf’s cysteine-rich domain. Ras is in blue; Raf is in pink. The major signalling mechanism is depicted by the thick arrow at the bottom.

For the other allosteric lobe interface, we show that K-Ras4B-GTP is unable to form a dimer with the helical interface involving α4 and α5 helices (ALDI-2), in either lipid or water environments. The dimer interface is significantly distorted, producing unfavourable binding free energy, suggesting that K-Ras4B does not favour the α4 and α5 helical interface. However, the H-Ras-GTP dimer preserves the symmetrical helix alignment of α4-α5/α4-α5 in both lipid and water environments. The dimer can be active, since the effector-binding sites are exposed, able to recruit Raf and promote its dimerization. In the α4 and α5 helical interface, there is a key difference in the sequence between K-Ras4B and H-Ras, which takes place at residue 165 in the α5 helix. Replacing Gln¹⁶⁵ of H-Ras by Lys¹⁶⁵ of K-Ras4B provides the additional hydrophilic interaction with Gln¹³¹ (Figure 5), preventing the dimer interface from distortion as observed in the case of the K-Ras4B-GTP dimer. Thus H-Ras-GTP intrinsically favours the α4 and α5 helical interface.

Both K-Ras4B-GTP and H-Ras-GTP dimers favour the shifted β-sheet interface (ELDI-1). The low values of the binding free energy indicate that the shifted β-sheet might be a populated interface for both Ras isoforms at the membrane. Ras dimerization through this interface implicates an inactive dimer, since the dimer interface covers exactly the Raf-binding site. To activate the kinase signalling cascade, Raf needs to compete with Ras for the β-sheet extension (Figure 7). We note that Raf binding is not comparable with Ras dimerization via the shifted β-sheet interface, since the binding free energy for Ras–Raf was reported to be −15.0 kcal/mol [66], which is lower than −8.89 kcal/mol for the shifted β-sheet K-Ras4B-GTP dimer. In solution, K-Ras4B-GTP still persists in aligning the shifted β-sheet interface, whereas the interface is slightly loosened in the H-Ras-GTP dimer. Ras can form a dimer in solution, albeit with very low population, since stable dimer formation through the interaction between the catalytic domains is interrupted by the unexpected HVR interactions. We observe that the HVR interactions with the binding partner’s catalytic domain and the HVR significantly contribute to H-Ras dimerization (Supplementary Figure S6).

Finally, we observe that the exact β-sheet interface (ELDI-2) containing the intermolecular backbone hydrogen bonds does not serve as a Ras dimer interface. Both K-Ras4B-GTP and H-Ras-GTP dimers, in lipid and water environments, convert the β-sheet interface into a β-sandwich that is stabilized by the salt bridges from the side chains of β1, β2 and β3 strands. We speculate that the β-sandwich can serve as a less populated minor Ras dimeric interface, since the interface is commonly observed for both Ras isoforms in different environments. The β-sandwich dimers tend to cause the HVR–HVR interaction, especially forming a hydrophobic core by the prenyl groups. In solution, the hydrophobic cluster significantly contributes to the binding free energy with lower values for both dimers compared with those at the membrane, suggesting that the HVR can mediate Ras dimerization.

In the present study, we further extend the functional role of the HVR to Ras dimerization, pointing out that the HVR significantly contributes to the dimeric Ras–Ras interaction. At the membrane, the HVRs of the K-Ras4B-GTP dimer solidly anchor to the membrane, facilitating the dimeric association between the catalytic domains. In solution, marginal contacts between the catalytic domains are partly sustained by the HVR interaction with the binding partner’s catalytic domain and HVR. The hydrophobic cluster formed by both farnesyl groups supplements the HVR–HVR interaction. In the H-Ras-GTP dimerization, the prenyl groups including farnesyl and palmitoyl tend to form a hydrophobic cluster in both lipid and water environments. With lesser charge, the HVRs in the dimer interact weakly with the anionic bilayer and appear to abandon their membrane anchoring task. In solution, the HVRs in the dimer interact strongly with each other, and the collapsed HVR backbones embrace the hydrophobic cluster formed by the prenyl groups. Membrane anchoring through the prenyl lipid moieties is not sufficient for productive and stable dimers [67]. Attachment to the phospholipids provides further stabilization, biases the monomer [40] and dimer organization and effector-binding site orientation, and constrains fluctuations, which promotes dimerization.

To conclude, Ras can form dimers, but the dimer interface appears highly dynamic. Dynamic Ras self-association can involve multiple interfaces; however, the interfaces are highly selective, reflecting the isoforms’ composition and environment. Oncogenic mutations in Ras may also shift the ensemble of Ras dimers and result in altered populated interfaces at the membrane. Ras preferences for anchoring in different membrane microdomains can drive formation of isoform-specific Ras dimers. Is Ras dimerization necessary? Data support Ras dimerization and higher oligomerization increasing signalling output and promoting Raf’s activation [1,2,68,69]. This is backed by structural considerations. However, proximal monomeric Ras proteins can also initiate downstream signalling with Raf, albeit, because of the higher degrees of freedom, less efficiently, thus it is likely to be a less populated mechanism.

The overall picture is complex. Oncogenic Ras isoforms are not uniformly expressed in different cancer types [65]. How the nature of the HVR sequence and its prenylation status play a key role in isoform-specific cell/tissue signalling is still among the questions facing our community. Finally, Ras is not unique in being a membrane-attached protein via an isoform-specific flexible stretch. Indeed, Raf provides another example. How this affects the organization of other domains and their interactions and whether, or how, this influences function are further open questions.

Abbreviations:

ALDI

allosteric lobe dimer interface

DOPC

1,2-dioleoyl-sn-glycero-3-phosphocholine

DOPS

1,2-dioleoyl-sn-glycero-3-phosphoserine

ELDI

effector lobe dimer interface

ERK

extracellular-signal-regulated kinase

HVR

hypervariable region

MAPK

mitogen-activated protein kinase

MEK

MAPK kinase

PI3K

phosphoinositide 3-kinase

PTM

post-translational modification

Ral

Ras-like

RalGDS

Ral guanine-nucleotide-dissociation stimulator

RASSF

Ras association domain family

vdW

van der Waals