GTP-Dependent K-Ras Dimerization

SUMMARY

Ras proteins recruit and activate effectors, including Raf, that transmit receptor-initiated signals. Monomeric Ras can bind Raf; however, activation of Raf requires its dimerization. It has been suspected that dimeric Ras may promote dimerization and activation of Raf. Here we show that the GTP-bound catalytic domain of K-Ras4B, a highly oncogenic splice variant of the K-Ras isoform, forms stable homodimers. We observe two major dimer interfaces. The first, highly populated β-sheet dimer interface is at the Switch I and effector binding regions, overlapping Raf’s, PI3K’s, RalGDS’ and additional effectors’ binding surfaces. This interface has to be inhibitory to such effectors. The second, helical interface also overlaps some effectors’ binding sites. This interface may promote Raf‘s activation. Our data reveal how Ras self-association can regulate effector binding and activity, and suggest that disruption of the helical dimer interface by drugs may abate Raf’s signaling in cancer.

INTRODUCTION

Ras-family proteins are small membrane-associated GTPases. Acting as molecular switches in response to receptor-mediated extracellular signals, they regulate cell survival, proliferation, motility and cytoskeletal organization (Karnoub and Weinberg, 2008; Stephen et al., 2014). Ras-GTP activates downstream effectors, including Raf kinase (Brennan et al., 2011; Rajakulendran et al., 2009) and phosphatidylinositol 3-kinase (PI3K) (Castellano and Downward, 2011; Castellano et al., 2013; Mitin et al., 2005; Repasky et al., 2004). Signal-activated Ras recruits Raf to the cell membrane and activates its kinase domain. Activation requires Ras binding, which opens the closed Raf conformation, thereby allosterically altering the kinase domain and promoting its dimerization (Lavoie et al., 2013). Raf side-to-side dimer formation (Rajakulendran et al., 2009) is necessary for normal Ras-dependent Raf kinase activation (Freeman et al., 2013). Dimerization also takes place in disease-associated constitutively active mutant Raf proteins and in inhibitor-induced Raf activation (Lavoie et al., 2013; Tsai and Nussinov, 2014). Through phosphorylation, active Raf triggers the MEK and ERK protein kinases, resulting in cell proliferation and survival (Nussinov, 2013; Rajakulendran et al., 2009). The crystal structure of Ras in complex with Raf’s Ras binding domain (RBD, PDB code: 4G0N) indicates that the high affinity Raf-Ras interaction involves extension of the Ras β-sheet at the Ras effector binding region (Fetics et al., 2015). Early studies suggested that Raf dimerization is induced downstream by the dimeric 14-3-3 cofactor protein (Tzivion et al., 1998). However, a recent observation that C-Raf forms dimers, trimers, and tetramers in the presence of Ras-GTP argues that Ras plays a significant role in Raf dimerization (Nan et al., 2013). How Ras promotes dimerization of Raf is unknown. A plausible hypothesis that Ras dimerization assists in self-association of Raf (Inouye et al., 2000) is gaining significant attention, leading to an overarching community goal to validate and understand Ras dimerization and how it relates to Raf’s regulation- activation and inhibition (Santos, 2014; Thompson, 2013). All three Ras isoforms, including H-Ras (Harvey sarcoma viral oncogene), N-Ras (neuroblastoma oncogene), and K-Ras (Kirsten sarcoma viral oncogene), with splice variants K-Ras4A and K-Ras4B, activate Raf to varying degrees. K-Ras, which is frequently mutated in human cancer (Lawrence et al., 2014), is the most potent activator of Raf among Ras isoforms (Prior et al., 2012). The reason for the differential activity of Ras proteins towards Raf is unclear. One possible reason may relate to differences in the dimeric structures of Ras isoforms and the differential ways through which they associate with raft/non-raft regions in the cell membrane (Jang et al., 2015). The sequences and structures of the catalytic domains (G-domains) of Ras isoforms are almost identical, but differ significantly in the 22 residue long C-terminal hypervariable region (HVR), which is lysine-rich in K-Ras4B (Hancock et al., 1990) (Figure 1). There are over 130 crystal structures of the Ras catalytic domain (Rose et al., 2015); the majority present Ras as a functional monomer. The HVR is invariably missing due to its high flexibility. Recent studies suggest that N-Ras-GDP can form dimers in a model membrane (Guldenhaupt et al., 2012). Native H-Ras can dimerize on membrane surfaces and the Switch II region (residues 60–76) plays a role in dimer formation (Lin et al., 2014). While mechanistic details of H-Ras and N-Ras dimerization are emerging, how K-Ras dimerizes and how Ras dimerization relates to activation of its effectors are still unclear.

Figure 1. K-Ras4B Sequence and Structure.

(A) The sequence of the K-Ras4B protein. In the sequence, hydrophobic, polar/glycine, positively charged, and negatively charged residues are colored black, green, blue, and red, respectively. Underlined residues denote hypervariable region (HVR). (B) Connection of each domain in K-Ras4B protein. (C) A crystal structure of K-Ras4B protein in the GTP-bound state (PDB code: 3GFT). The structure was generated for wild-type sequence modified from the crystalized mutant conformation. Functional regions are marked on the structure.

Here, we show that GTP-bound, but not GDP-bound K-Ras4B catalytic domain is capable of forming stable homodimers. Modeling of the GTP-bound structure reveals two major dimer interfaces. Using dynamic light scattering (DLS), isothermal calorimetry (ITC), microscale thermophoresis (MST), fluorescence spectroscopy, Förster resonance energy transfer (FRET) and NMR, we test and confirm the modeled structures. Importantly, the first highly populated dimer interface spans the Switch I and effector binding regions at the effector lobe. It involves β-sheet extension and is very similar to Raf’s interaction as well as to other effectors binding at this surface. In the second major dimerization mode, the interaction involves helical interfaces at the C-terminal allosteric lobe of the G-domain. These results not only validate the ability of Ras catalytic domain to form dimers, but also suggest that dimer formation provides a previously unknown mechanism for how possible Ras dimers may regulate Raf’s dimerization and signaling outputs. They suggest that under certain conditions Ras-binding domains of Ras effectors and regulators compete with the β-sheet dimer interface of K-Ras4B. At the same time, Raf’s dimerization can be promoted by Ras dimerization through the helical interface. Our results point to the complex mechanisms of signaling regulation through dynamic Ras oligomerization. In addition, they argue that targeting the helical interfaces via orthosteric or allosteric inhibitors may provide therapeutic intervention targeting Raf’s dimerization.

RESULTS

The GTP-bound, but Not GDP-bound K-Ras Catalytic Domain Forms Stable Dimers

To evaluate the oligomerization state of K-Ras G-domain, we performed dynamic light scattering (DLS) experiments. DLS of GDP-bound K-Ras4B_1–166 showed a broad size distribution of protein particles (Figure 2A). The predominant species has a radius that corresponds to a globular protein of 18 kDa molecular mass (see numerical data for K-Ras4B in Table 1), which is similar to the mass of a G-domain monomer (18.8 kDa for K-Ras4B_1–166). In contrast, K-Ras4B_1–166 freshly loaded with GTP-γ-S exhibits a narrow particle size distribution (with 9.9% polydispersity) and a radius that corresponds to a globular protein with a molecular mass of 41 kDa. Although the experiment was performed in the presence of high concentration of reducing agent, we analyzed the samples by high-resolution mass spectrometry to further confirm the absence of a covalent dimer that could be formed through oxidation of cysteine residues. The data suggest that K-Ras4B_1–166-GTP-γ-S forms a stable non-covalent dimer, while K-Ras4B_1–166-GDP may have a reduced tendency to self-associate. It is remarkable that the G-domain loaded with the most commonly used GTP analog, guanosine 5′-[β, γ-imido] triphosphate (GppNHp) is a predominantly monomeric 21 kDa protein as observed from available crystal structures. Full-length K-Ras4B demonstrated dimer formation in the GDP-bound form, while the GTP-γ-S protein had a radius corresponding to a tetramer with large polydispersity (Figure 2B; Table 1).

Figure 2. Dynamic Light Scattering Demonstrates Nucleotide-dependent Oligomerization of K-Ras4B.

(A) Dynamic light scattering (DLS) shows formation of a stable dimer (41 kDa) by an “active” GTP-γ-S-loaded K-Ras4B G-domain, but not by “inactive” GDP-loaded (18 kDa) or GppNHp-loaded (21 kDa) protein. (B) DLS suggests formation of a loose dimer by the full-length K-Ras4B_1–188-GDP (38 kDa) and a tetramer by K-Ras4B_1–188-GTP-γ-S (74 kDa).

Table 1.

Numerical data for different forms of K-Ras4B.

Sample	Radius (nm)	%Pd*	Mw-R (kDa)	%Intensity
K-Ras4B1–166-GTP-γ-S (25μM)	2.9	9.9	41	98.1
K-Ras4B1–166-GDP (83 μM)	2.0	37.5	18	99.9
K-Ras4B1–166-GppNHp (83 μM)	2.17	9.5	21	88
K-Ras4B1–188-GTP-γ-S (9.8μM)	3.7	33.0	74	99.8
K-Ras4B1–188-GDP (21 μM)	2.8	9.7	38	99.7

^*Percent polydispersity

Predictions of K-Ras4B Dimer Structures

To evaluate the structural basis for dimer formation by Ras proteins, we exploited a powerful template-based protein-protein complex structure prediction algorithm (PRISM) (Acuner Ozbabacan et al., 2012; Aytuna et al., 2005; Kar et al., 2012; Ogmen et al., 2005; Tuncbag et al., 2011; Tuncbag et al., 2012). The list of target PDB codes used for the predictions is provided in Table S1. Our modeling results suggest that in both GTP- and GDP-bound states, K-Ras4B_1–180 can form homodimers through the same interface (Figure 3A; Figure S1). In both cases, the interface regions include α1, β2, and β3. However, the K-Ras4B-GTP homodimer is more stable than its GDP counterpart. The enhanced stability is associated with a larger interface and more hot spots (see Methods) in the GTP-bound homodimer and thus greater favored binding energy score (BES). Analysis of atomic interactions showed that the Ras-GTP homodimer interface contains more H-bonds than the GDP-bound (Tables S2–S5). To corroborate the predicted interface, we reconstructed all possible crystal dimer interfaces of Ras structures using the Evolutionary Protein-Protein Interface Classifier (EPPIC) (Duarte et al., 2012) Server. One of the crystal interfaces of the Ras dimer is similar to our predicted K-Ras4B-GTP homodimer and even H-Ras-GTP dimer interfaces (Figure S2A–S2C). To evaluate the structural similarity between the crystal and predicted dimer interfaces, we aligned the interface regions and calculated the interface RMSD using the backbone atoms. The superimposed structures are shown in Figure S2D–S2G. The interface RMSD of crystal-H-Ras-GTP and crystal-K-Ras4B-GTP aligned structures are 5.62Å and 5.06Å, respectively. Most of the residues that are predicted to participate in the dimerization interface also reside at the crystal interface. Residues like Q25 and Y40 are shared by both crystal and predicted K-Ras4B-GTP interactions and are part of the critical region of the interface site. The first major predicted K-Ras4B-GTP dimer interface involves a (shifted) β-sheet extension, forming intermolecular H-bonds between β2 strands (hereafter referred to as β-Homodimer). Computations also predict a helical dimer interface involving helices α3 and α4 (hereafter referred as α-Homodimer) (Figure 3B; Table S6). We also observed two minor interfaces: α (helices α4 and α5) and β (exact alignment of the β2 strands) (Figure S3). We further predicted that GTP-bound Ras can bind GDP-bound Ras through these interface regions with differing stabilities (Figure S4; Table S7); however, the most favorable interface (Heterodimer 1) remarkably overlaps with the major one predicted for the GTP- and GDP-bound homodimers, involving a shifted β-sheet extension.

Figure 3. Predicted Major Homodimer Structures of Active Ras Isomer.

(A) Predicted homodimer structure of GTP-bound K-Ras4B_1–180 with a shifted β-sheet extension (β-Homodimer; PDB code: 4DSO). Binding energy score (BES) for the prediction is -44.53 (Template interface: 2erxAB). Interface residues are I21, I24, Q25, H27, V29, E31, D33, I36, E37, D38, S39, Y40, R41, K42, Q43, and L52. A minor β-sheet homodimer structure of K-Ras4B_1–180-GTP with an exact sheet-alignment is shown in Figure S3A. (B) Predicted homodimer structure of GTP-bound K-Ras4B_1–167 with the interface involving helices α3 and α4 (α-Homodimer; PDB code: 3GFT). Dimer interface is found to be at the allosteric lobe with binding energy score (BES) of −17.92 (Template interface: 4a9eAB). Interface residues are E91, H94, R97, E98, K101, R102, D105, S106, E107, Q129, D132, L133, R135, and S136. A minor helical homodimer structure of K-Ras4B_1–167-GTP with the interface involving helices α4 and α5 is shown in Figure S3B.

Biophysics of K-Ras4B Dimers

Because of the structural complexity and heterogeneity of Ras oligomerization, we limited our initial biophysical studies to the Ras catalytic domain. The observed tetramer formation by full-length K-Ras even in the absence of the membrane anchoring is in good agreement with the previously observed self-association of K-Ras-4B HVR domain (Jang et al., 2015), suggesting that the HVR can contribute further to stabilization of Ras oligomeric structures. The mechanisms of the catalytic domain dimerization are important due to their involvement in regulation of Ras signaling. To characterize the stability of the G-domain dimer, we performed isothermal calorimetry (ITC) studies on K-Ras4B_1–166-GTP-γ-S. 30 μM solution of the freshly GTP-γ-S-loaded protein was titrated into the instrument’s cell that contained the buffer used for gel-filtration of the protein sample. Dissociation of the dimer was accompanied by an enthalpy change allowing the determination of the dimer’s apparent dissociation constant, K_D = 770 ± 200 nM (Figure 4A) that was calculated assuming that dimers were the only oligomerized species. To further confirm the high affinity for dimerization of K-Ras4B_1–166-GTP, we employed microscale thermophoresis (MST) (Jerabek-Willemsen et al., 2011). The G-domain was labeled with fluorescein maleimide on Cys118. Substitution of this residue with Ser had no influence on Ras activity (Mittar et al., 2004). Thus, we assume that addition of fluorescein to this position might have an insignificant effect on dimerization. We titrated the fluorescent protein with unlabeled GDP- or GTP-γ-S-loaded K-Ras4B_1–166. For K-Ras4B_1–166-GTP-γ-S, the K_D is 1180 ± 92 nM (Figure 4B). Although the addition of K-Ras4B_1–166-GDP is also accompanied by a change in thermophoretic mobility of K-Ras4B_1–166-GTP-γ-S, the K_D could not be determined because the saturation could not be achieved due to the limited solubility of the protein. The data suggest that a ‘heterodimer’ is significantly less stable than a ‘homodimer’ formed by K-Ras4B_1–166-GTP-γ-S. We detected the interaction at concentrations of K-Ras4B_1–166-GDP as low as 1 μM, which could be physiological because membrane anchoring can result in significant local concentrations of the Ras protein in cells. While performing MST titration, we found that addition of non-fluorescent K-Ras to the fluorescent protein caused dramatic changes in the fluorescence of the fluorescein moiety due to the changes in the dye environment. This change in fluorescence allowed the determination of the binding K_D by an additional method (Figure 4C). The data further confirmed high affinity dimer formation. The discrepancy in K_D values obtained from MST and fluorescence studies (1180 ± 92 nM vs. 2530 ± 200 nM) may be due to the structural heterogeneity of the dimers since different association modes will have different impacts on the measured values resulting in different combined effects and thus different apparent K_D. To evaluate the dimer configuration, we determined the distance between fluorophores positioned on C118 residues of truncated K-Ras4B_1–166-GTP-γ-S in the homodimer complex using Förster resonance energy transfer (FRET). Fluorescein- and tetramethylrhodamine were used for labeling K-Ras4B_1–166-GTP-γ-S because this donor and acceptor FRET pair has been extensively characterized. Addition of tetramethylrhodamine-labeled truncated G-domain caused almost two-fold decrease in the fluorescence of fluorescein-labeled GTP-γ-S-loaded K-Ras4B_1–166. The energy transfer efficiency was calculated after corrections for inner filter effect and changes in fluorescence outputs due to changes in the fluorophore’s environment during the dimerization. The determined efficiency was very high, 95.4 ± 5%. The calculations detailed in the Methods section indicate that the distance between the fluorophores in the dimer is less than 39 Å. This distance is in agreement with the interfaces predicted by the computational studies, but does not allow differentiating between them.

Figure 4. Biophysics of K-Ras4B Dimers.

(A) Isothermal titration calorimetry (ITC) studies of K-Ras4B_1–166-GTP-γ-S dimer dissociation. 30 μM protein solution was titrated into calorimeter cell containing the buffer. (B) Dissociation constant, K_D, was determined by microscale thermophoresis (MST) using 30 nM K-Ras4B_1–166-GTP-γ-S labeled with fluorescein maleimide on Cys118, which was mixed with increasing concentrations of non-labeled protein. (C) Addition of non-labeled K-Ras4B_1–166-GTP-γ-S to fluorescein-labeled K-Ras4B1-166-GTP-γ-S results in significant increase in fluorescence, which was used for determination of K_D. Concentration of labeled protein was kept constant at 30 nM.

NMR Identification of K-Ras4B Dimerization Interfaces

To map the interfaces involved in the association of K-Ras4B molecules in solution, we analyzed NMR chemical shift perturbations (CSPs) induced by dilution of K-Ras4B samples. Sample dilution can cause dissociation of the protein-protein complexes and reveal the interfaces by changes in the chemical shift values. Earlier we observed that the HVR associates with the catalytic domain of K-Ras4B-GDP primarily via the high affinity β-sheet interface (data not shown), raising the possibility that this intramolecular interaction influences K-Ras4B dimerization. To address this possibility, we studied the dimer interface of K-Ras4B_1–166. We compared the ¹⁵N HSQC spectra of 245 μM and 30 μM K-Ras4B_1–166-GTP-γ-S (Figure 5A). These spectra show less line-broadening and were amenable to a more complete analysis. We had to use concentrations significantly higher than K_D during spectra collection to obtain sufficiently strong signals. However, the estimated concentration of the monomer at 30 μM total K-Ras concentration is 4 μM, which is capable of generating detectable signals. Residues that exhibited dilution-dependent CSPs include K16 and I24 in α1, D30, E31, and E37 in the Switch I region, S39 and V44 in β2 (Figure 5B; Figure S5A). In addition, we observed significant changes in N86 in α3, D119 in the G4 loop, Q131 and R135 in α4, and K147 in G5 loop. Next, we measured the changes in ¹⁵N HSQC spectra of high 200 μM and low 18 μM concentrations of K-Ras4B_1–188-GTP-γ-S. Spectral overlays demonstrated CSPs caused by sample dilution (Figure 5C); however, significant line-broadening due to conformational fluctuations and protein-protein interactions prevented observation of many resonances in the GTP-γ-S-bound protein. Residues exhibiting significant CSPs were scattered: A18 in α1, D38 in the Switch I, I84 in β4, E91 in α3, V114 in β5, V125 in G4 loop, A130 in α4, E143 in β6, and Y157 in α5 regions (Figure 5D; Figure S5B). Although the dilution-induced CSPs in GTP-γ-S-loaded full-length and truncated K-Ras4B are not identical, the involvement of α1, Switch I, α3, G4 loop, α4, and G5 loop can be observed in both. The HVR might also engage β4, β5, β6 and the N-terminal portion of α5 in K-Ras4B-GTP-γ-S dimerization.

Figure 5. NMR Chemical Shift Perturbations of Residues in the GTP-bound K-Ras4B Dimer.

(A) An overlay of high (blue) and low (red) concentrations of ¹⁵N-HSQC NMR spectra of truncated K-Ras4B_1–166-GTP-γ-S (245 μM and 30 μM). Examples of chemical shift changes are marked with arrows. (B) The perturbed residues of K-Ras4B_1–166-GTP-γ-S are mapped on the catalytic domain structure of GTP-bound K-Ras4B. (C) An overlay of high (blue) and low (red) concentrations of ¹⁵N-HSQC NMR spectra of full-length K-Ras4B_1–188-GTP-γ-S (200 μM and 18 μM). Examples of chemical shift changes are marked with arrows. (D) The perturbed residues of K-Ras4B_1–188-GTP-γ-S are mapped on the catalytic domain structure of GTP-bound K-Ras4B.

Residues with large chemical shift changes are mainly located at the N-terminal effector lobe of the G-domain (Figure 5A, 5B; Figure S5A), in good agreement with the predictions (Figure 6). Of the residues showing large chemical shift changes, 57% are identical with those predicted by computations and 29% are located adjacent to the predicted interface residues. Those residues located at the C-terminal allosteric lobe can serve as the second major binding interface, as predicted by computations (Figure 3B; Table S6). Two of the five perturbed residues, Q131 and R135 in α4, correlate with our prediction. Taken together, these results indicate that the chemical shift perturbation data agree well with the predicted binding sites. For the GDP-bound state, we used high (275 μM) and low (30 μM) concentrations of K-Ras4B_1–166-GDP. The spectral overlays demonstrate significant structural differences (Figure S6A). Most CSPs concentrate in only five residues, E3 in β1, F28 and D38 in the Switch I region, H94 in α3, and Q129 in α4 (Figure S6B). Spectral overlays of ¹⁵N HSQC spectra of 780 μM and 50μM K-Ras4B_1–188-GDP indicated dilution-associated chemical shift changes (Figure S7A). Most CSPs concentrate in the β2 (K42, I46, E49, C51-L53), α2 (M72 and G75), β4 (V81 and N85), and α3 (K88, I93, and I100) regions (Figure S7B). The CSPs also indicate potential involvement of E3 in β1, D38 in the Switch I region, and Q150 in the G5 loop regions. More extensive interfaces were found for K-Ras4B_1–188-GDP suggesting a role for HVR in K-Ras4B-GDP dimerization.

Figure 6. Comparison of NMR Chemical Shift Data with Computational Predictions.

Mappings of residues from NMR chemical shift perturbations (CSPs) results for K-Ras4B-GTP-γ-S (Figure S5A) (red and orange sticks) and the predicted residues in the dimer interface by computations for K-Ras4B-GTP β-Homodimer (as in Figure 3A) (dark blue sticks) and α-Homodimer (as in Figure 3B) (light blue sticks). Yellow sticks enclosed by transparent surface represent the common residues from both NMR and computations.

Ras-Effector Binding and Tetramerization

Comparison of crystal structures of Ras with its effectors (Raf, PI3K and RalGDS) show that the predicted major β-sheet extension dimer interface (Figure 7A) overlaps the effectors’ binding site. Structural analysis revealed that the binding surfaces and interaction modes of Raf (Figure 7B and 7C; Table S8) and other Ras effectors, important in cell transformation and proliferation such as PI3K (Figure 7D), PLCe, Byr2 and Nore1A (crystal structures of the complexes of these effectors bound to Ras are available), and Cdc42 and Ftase (crystal structures of these Ras regulators are not available, but the complexes are predicted by PRISM), overlap Ras’s β-sheet dimer interface; RalGDS (crystal structure, Figure 7E), IMPA1 (predicted) and RIN1 and RassF1 (structures modeled and interfaces predicted), partially overlap. Overlapping (or partially overlapping) interfaces argue for an effective local effector concentration threshold as a regulatory mechanism in activation and signaling. Computations also predict a major helical interface at 3α/α4 (Figure 3B), supported by NMR chemical shift experiments for homo K-Ras4B-GTP (Figure S5) and hetero K-Ras4B-GTP/K-Ras4B-GDP (Figure S8) dimers. Ras regulators, PKCα (crystal), RAIN, RGS12, AFAD (structure modeled and interface predicted) overlap these interfaces, and RGL1 (modeled and predicted) partially overlaps. Finally, DLS data suggest higher order tetramer organization in the full-length protein (Figure 2B). We do not detect tetramer formation by the G-domain. This suggests that any pairs of non-overlapping interfaces are unlikely to allow for tetramer formation in the absence of the HVR. However, the HVR facilitates tetramerization even in the absence of the membrane. Three types of dimers and the higher-order tetrameric organizations further affirm nanocluster formation. Our interface predictions supported by NMR data suggest that the major nanocluster species is Ras-GTP, in good agreement with experimental data (Plowman et al., 2005). Since effector binding overlaps the dimeric interfaces, dynamic cluster reorganization can be expected. For dimeric Raf, nanoclusters can also be networked via Raf’s catalytic domain.

Figure 7. Ras-Effector Binding.

Side by side comparisons of (A) the predicted β-sheet interface of K-Ras4B-GTP β-Homodimer (as in Figure 3A) with interfaces observed in crystal structures of Ras-effectors. Predicted dimer structures of the GTP-bound K-Ras4B with (B) Ras-binding-domain (RBD) of Raf1 (Target PDB code: 1C1Y, Template interface: 1c1yAB) and (C) RBD-B-Raf (Target PDB code: 3NY5, Template interface: 1c1yAB). (D) Crystal dimer structures of H-Ras(G12V) with PI3Kγ (PDB code: 1LHE8) and (E) H-Ras(E31K) with Ras-interacting-domain (RID) of RalGDS (PDB code:1LFD). Arrows indicate the same β-sheet extension motif in the homodimer and in Ras-effector (Raf, PI3K and RalGDS) crystal structure complexes.

DISCUSSION

Ras dimerization and higher oligomerization states were long believed to increase signaling output and facilitate interactions with effectors, such as Raf (Guldenhaupt et al., 2012; Inouye et al., 2000; Kolch, 2000). However, dimerization was not observed in soluble Ras leading to an assumption that interfaces observed in crystals were due to crystal packing interactions. Dynamic light scattering studies have unexpectedly revealed GTP-dependent dimerization of K-Ras G-domain, which we further confirmed by microscale thermophoresis, isothermal titration calorimetry, fluorescence spectroscopy, FRET, NMR spectroscopy and computational analysis. Application of a powerful structural prediction algorithm confirmed high affinity homodimer formation for K-Ras4B-GTP, but only low affinity dimerization for K-Ras4B-GDP. A survey of Ras crystal structures showed that while the functional state of Ras has been identified as a monomer, its predicted membrane-anchored dimeric interface is among its crystal packing interactions (Figure S2). Dilution induced NMR CSPs of K-Ras4B_1–188 and K-Ras4B_1–166 dimers further identify dimer interfaces. Both suggest a highly dynamic K-Ras self-association that is likely to involve more than a single interface. Remarkably, the structural elements involved in the dimer interactions predicted by the computations are in good agreement with those detected by NMR (Figure 3A); the predictions cover 50% of all the interface residues identified by chemical shift changes. The most stable dimer interface is located at α1, β2, and β3, including the Switch I and effector binding regions. Although the helical dimer interface appears less populated for the G-domain, the relative populations of the interfaces may change in full-length membrane-anchored K-Ras4B.

The dimerization of N-Ras and H-Ras underscore the importance of the plasma membrane for Ras association (Guldenhaupt et al., 2012; Lin et al., 2014). Farnesylated and methylated synthetic HVR of K-Ras4B tends to aggregate at concentrations as low as 5–10 μM (Chavan et al., 2013) and thus can significantly enhance dimerization. Our results demonstrate that K-Ras dimers can also form in solution; however, the measured relatively high dimer dissociation constants suggest that in order to dimerize, the concentration of K-Ras has to be high. The plasma membrane may act to increase the effective local concentration of K-Ras. This is in agreement with previous observations which indicate that the requirement of the plasma membrane for Ras dimerization and ability to activate Raf can be overcome by a GST fusion tag that increases the affinity of Ras monomers towards each other (Inouye et al., 2000). Expression levels and the partners’ activation state can also affect dimerization. The range of crystal dimer interfaces observed in different crystal forms capture snapshots of heterogeneous dimer conformations. Interfaces, including those of Ras dimers, are dynamic, interconverting between substates (Aramini et al., 2014). Along these lines, a large-amplitude rocking motion has been detected by ¹⁹F NMR in the dimer interface of influenza A virus nonstructural protein 1 (NS1A) (Aramini et al., 2014). We propose that both α-helical and β-sheet dimers are populated in membrane nanoclusters; a single interface type is unlikely to support nanocluster avidity.

It is remarkable that K-Ras4B_1–166 loaded with non-hydrolysable GTP analog GppNHp (guanosine 5′-[β,γ-imido] triphosphate) showed a greatly reduced tendency to dimerize compared with the protein loaded with GTP-γ-S (guanosine 5′-O-[γ-thio] triphosphate). This high sensitivity to the nucleotide analog structure may explain why Ras dimerization was not previously observed in solution (Guldenhaupt et al., 2012; Lin et al., 2014): GppNHp is the most commonly used GTP analog, due to its chemical stability, and it was used in all earlier studies. However, recent data suggest that it is a poorer mimetic of GTP compared to GTP-γ-S (Spoerner et al., 2010). GTP-γ-S, on the other hand, undergoes not only catalyzed, but also spontaneous hydrolysis, which makes using it in traditional, but ‘slow’ methods of oligomerization analysis such as size-exclusion chromatography and ultracentrifugation impractical. The ability of GppNHp to induce a GDP-like state has also been detected in other GTPases (Paleskava et al., 2012). The observed difference in the ability to induce Ras dimerization between two GTP analogs, GTP-γ-S and GppNHp, along with reported data on their effects on Ras conformation and dynamics correlate well with the predicted mechanisms of dimer formation. The affinity towards Ras varies significantly among the effectors’ Ras-binding domains (RBDs), and some have K_D comparable to that of Ras dimerization. We do not observe a significant presence of trimers and tetramers of the G-domain in solution. However, we observe computationally, supported by experimental data, that K-Ras4B can form K-Ras4B-GTP/K-Ras4B-GDP heterodimers at the same interface (Figure S4A). This is expected: this region is populated by effectors via β-sheet extension, also a structural property of K-Ras4B-GDP. Further, the NMR CSP cluster between T20 - M67 (Figure S8) fits our main predicted interface (Heterodimer 1 with BES = −45.66), and the cluster between V125 - D154 fits the second interface prediction (Heterodimer 2 with BES = −33.25).

Oligomerization enhances cooperativity, and the dimerization interfaces are also involved in binding the effectors. Thus, the oligomerization’s major raison d’être could be higher specificity among Ras isoforms by restricting binding events and averting redundant signaling. The Ras tetramer landscape is heterogeneous, can be nucleotide-dependent, and shift with the effector concentrations, isoform sequence and membrane environment, resulting in preferred orientation with respect to the membrane (Abankwa et al., 2010; Jang et al., 2015). In a similar vein, EphA isoforms assemble into different crystal organizations for distinct functions (Himanen et al., 2010; Seiradake et al., 2013), with a single mutation converting a circular architecture to an extended one with a concomitant functional change. Along these lines, the varied structures of the allosteric lobe across Ras isoforms suggest that they may populate different dimerization (and higher oligomerization) states; thus, our minor (α and β) dimeric interface species which are sparsely populated in K-Ras4B may be highly populated in N- or H-Ras. Together with the different HVR sequences and preferred membrane interaction domains, they may result in isoform-specific signaling from the membrane.

The dimer interfaces described here suggest a paradigm shift in Ras effector activation. Common view holds that Ras dimerization is involved in Raf’s activation, with each Ras monomer able to simultaneously dimerize and interact with the Raf’s RBD; however, the Ras homo/hetero-dimer β-interface is almost identical to that involved in Ras binding to Raf, PLCε, PI3K, Ftase and CDC42. The helical interfaces are involved in Ras binding to PKCα, TIAM, and possibly other effectors. The emerging scenario paints a dynamic landscape populated by β-sheet- and helical-interface-mediated dimers that associate further into nanoclusters. Shared interfaces imply Ras oligomerization dependence on the effectors. This is supported by the observed enhancement of K- and N-Ras nanoclustering by BRaf inhibition (Cho et al., 2012). Raf, with nanomolar affinity for Ras, outcompetes the β-sheet mediated dimerization. Raf dimerization may be enhanced via self-association of membrane-anchored Ras through the helical interfaces (Figure 8). Dimerization can prevent external stimuli from causing repetitive cellular firing. We reason that tetramerization driven by farnesylated K-Ras4B on the inner plasma membrane via dynamic, concentration- and actin-dependent (Plowman et al., 2005) associations, through galectin and possibly IQGAP1 (Nussinov et al., 2013; Osman et al., 2013), confers higher specificity and cellular control. Dimerization may further induce higher selectivity across Ras isoforms through preferred orientation of interaction with the membrane.

Figure 8. Ras dimerization and Raf activation.

Cartoons represent the (A) Ras dimerization and (B) Ras-Raf binding in the cell membrane.

To conclude, a powerful combination of four interdisciplinary techniques allowed us to elucidate the structural basis of Ras dimerization. The results reveal a complex mechanism of downstream signaling regulation by Ras oligomerization that involves cooperativity and selectivity through shared binding interfaces. Our results argue that therapeutically abolishing the major β-sheet mediated dimerization may not deter Raf’s activation; however, the helical dimer interface may be targeted by synthetic compounds to abate Raf signaling in cancer. Ras shape and stability as a monomer argue that it can function and activate the effectors as a monomer. The emergence of multiple dimer interfaces affirms the significance of nanoclustering in regulation of Ras effectors’ activation and signaling (Nussinov, 2013).

EXPERIMENTAL PROCEDURES

Computational Prediction of the Ras Dimer Structures

In this study, we used PRISM (Aytuna et al., 2005; Ogmen et al., 2005; Tuncbag et al., 2011), a template-based protein-protein complex structure prediction algorithm, to predict models for the dimeric structures of the Ras protein. PRISM is based on our observation that protein-protein interface motifs are conserved in nature, similar to single chain architectures (Keskin et al., 2008; Keskin et al., 2004; Tsai et al., 1996). PRISM uses the tertiary structure of proteins of interest to build models. The solutions are optimized by a docking refinement protocol and ranked based on the binding energy scores (BES) computed by FiberDock (Mashiach et al., 2010), which has been implemented in PRISM. PRISM has been tested extensively for a range of proteins and pathways (Tuncbag et al., 2012). The structural data for the dimer predictions (target proteins) have been obtained from the PDB. We used the G-domains (corresponding to Ras residues 4–166) of both H- and K-Ras as target proteins. K- and H-Ras catalytic domains have 10 and 115 structures in the PDB, respectively. We included all K-Ras structures in the prediction (Table S1). For H-Ras, only X-ray structures with resolution of 2.00 Å or lower were considered. To reduce the redundancy due to very similar interface architectures, we calculated the root mean squared deviation (RMSD) for each pair. For structures with RMSD below 0.25 Å, we chose a representative with the highest resolution. In this way, the number of structures for H-Ras was reduced to 46 (Table S1). Then, we identified the interface regions of the putative dimers using the HotPoint web server (Tuncbag et al., 2010), which accounts for conservation, solvent accessibility and the total contact potential of the interface residues. We compared our results with crystal structures to further affirm our predictions. We used the Evolutionary Protein-Protein Interface Classifier (EPPIC) (Duarte et al., 2012) Server to classify the crystal interfaces in Ras structures. We further built unit cells of Ras crystals using Chimera to check whether the interfaces match.

Protein Preparation

Protein preparation was carried out according to our previously published protocol (Abraham et al., 2010). K-Ras4B cDNA (full length and catalytic domain) from Invitrogen was cloned into the pET42a vector (Novagen). Full details are provided in the Supplemental Experimental Procedures.

Dynamic Light Scattering

DLS studies were performed on a DynaPro Tytan instrument equipped with a Temperature-Controlled MicroSampler (Wyatt Technology Corp., Santa Barbara, CA) at a laser wavelength of 830 nm, scattering angle of 90° in a 12-μl quartz cuvette at 25°C. Each measurement consisted of sixty 10-second acquisitions. All samples were centrifuged at 15000 g for 10 min before measurements. To obtain the hydrodynamic radii (R_h) and percentage polydispersity, the intensity autocorrelation functions were fitted with a non-negative least squares algorithm by Dynamics 7.1.1.3 software (Wyatt Technology Corp., Santa Barbara, CA.).

Isothermal Titration Calorimetry

The dissociation process was studied using an isothermal titration microcalorimeter iTC₂₀₀ (GE Healthcare/Microcal, Northampton, MA) at 25°C. The typical experiment included injection of 10 aliquots (3.0 μL each) of 30–70 μM K-Ras protein solution into buffer in the ITC cell (volume 200 μL). The buffer used was collected during the size exclusion chromatography purification of the protein and therefore its composition exactly matched the buffer of the protein solution. After filling the syringe with the solution, the level at the tip of the syringe was thoroughly adjusted to eliminate an air bubble which could affect the precision of the first injection. The experiments were run at “High feedback mode/gain” setting. The stirring speed was 1000 RPM. The duration of the injection in seconds was usually twice the value of injection volume. The integrated heat values were fit using the “Origin 7.0” - based ITC data analysis software provided by GE Healthcare/MicroCal. If necessary, prior to the integration procedures the baseline was manually adjusted to minimize the background noise. The model “Dissociation” was used as the basic option, yielding the dissociation constant K_D and other thermodynamic parameters.

Microscale Thermophoresis

For labeling the Ras protein with fluorescein, 1.6 mg fluoresceine maleimide (Santa Cruz Biotechnology Inc.) was dissolved in 25 μl DMSO. 5 μl were diluted in 10 μl buffer. The resulting solution was slowly added to 400 μl of 75 μM GTP-γ-S-loaded K-Ras solution in 100 mM HEPES buffer pH 6.5 containing 5 mM MgCl_2, 50 mM NaCl and 0.5 M BCEP. The mixture was incubated overnight at 4°C and filtered through NP-5 column (GE) equilibrated with the reaction buffer. LC/MS showed no traces of the maleimide and complete labeling of intact K-Ras. For MST studies we have prepared 16 2-fold serial dilution of GDP or GTP-γ-S-loaded K-Ras_1–166 starting from 60 μM. Titration series were prepared that contained 15 μl of 60 nM fluorescein-labeled K-Ras and 15 μl of non-labeled K-Ras. The final buffer composition included 25 mM HEPES pH 7.2, 50 mM NaCl, 2.5 mM MgCl_2. MST measurements were taken in standard treated capillaries on Monolith NT.115 instrument (NanoTemper Technologies GmbH, Germany) using 50% IR-laser power and LED excitation source with λ= 470 nm. NanoTemper Analysis 1.2.20 software was used to fit the data and to determine the K_D values.

Förster Resonance Energy Transfer (FRET)

Full details of the method are provided in the Supplemental Experimental Procedures.

NMR Experiments

All ¹H-¹⁵N HSQC NMR spectra were collected on a 900 Mhz Bruker Avance Spectrometer at 25°C. The buffer used for dissolving the proteins contained 50 mM Tris-citrate, pH 6.5, 50 mM NaCl, 5 mM MgCl₂, 10 mM β-mercaptoethanol and 10 mM CaCl₂ and 10% deuterated water. Data processing and analysis were carried out using NMRPipe. The BMRB database (http://www.bmrb.wisc.edu) was used for chemical shift assignments (BMRB ID 17785 and 18529). HNCA, HNCB and CBCACOHN experiments were used to make assignments for the hypervariable region domain. For probing the chemical shift perturbations, the mean chemical shifts were calculated by

$$\mathrm{\Delta }{\delta }_{NH}=\sqrt{\frac{{(\mathrm{\Delta }{\delta }^{N}H)}^2+{(\mathrm{\Delta }\delta N)}^2/25}{2}}$$

As per convention, chemical shift perturbations higher than the sum of the average and one standard deviation were considered to be statistically significant. The figure legends note the concentrations of the proteins used in each experiment.

Highlights.

• The GTP-bound K-Ras4B catalytic domain can form stable homodimers

• The K-Ras4B predicted β-sheet dimer interface overlaps the effectors’ binding site

• The K-Ras4B α3/α4 helical dimer interface may promote Raf‘s dimerization

• The two interfaces may illuminate K-Ras4B nanocluster formation