Structural basis of the atypical activation mechanism of KRASV14I

RAS regulation and signaling are largely accomplished by direct protein-protein interactions, making RAS protein dynamics a critical determinant of RAS function. Here, we report a crystal structure of GDP-bound KRASV14I, a mutated KRAS variant associated with the developmental RASopathy disorder Noonan syndrome (NS), at 1.5–1.6 Å resolution. The structure is notable for revealing a marked extension of switch 1 away from the G-domain and nucleotide-binding site of the KRAS protein. We found that this extension is associated with a loss of the magnesium ion and a tilt in the position of the guanine base because of the additional carbon introduced by the isoleucine substitution. Hydrogen-deuterium exchange MS analysis confirmed that this conformation occurs in solution, but also disclosed a difference in kinetics when compared with KRASA146T, another RAS mutant that displays a nearly identical conformation in previously reported crystal structures. This conformational change contributed to a high rate of guanine nucleotide-exchange factor (GEF)-dependent and -independent nucleotide exchange and to an increase in affinity for SOS Ras/Rac GEF 1 (SOS1), which appears to be the major mode of activation for this RAS variant. These results highlight a mechanistic connection between KRASA146T and KRASV14I that may have implications for the regulation of these variants and for the development of therapeutic strategies to manage KRAS variant-associated disorders.

RAS mutations are common in disease, including cancer (1) and developmental disorders (2). RAS mutations vary in their properties including their impact on global signal transduction (3) and transforming potency (4,5). These higher-order characteristics likely relate to differences in fundamental properties such as enzymatic activity and protein dynamics. Moreover, the recent development of covalent inhibitors of KRAS G12C , a mutation common in lung cancer, suggests that successful therapeutic interventions against RAS will require a mutationspecific approach (6).
RASopathy syndromes are a collection of developmental disorders characterized by hyperactivity of canonical RAS pathways and hyperproliferation of selected tissues (7). One such syndrome is Noonan syndrome (NS), 2 an autosomal dominant genetic disorder characterized by variable expression of short stature, craniofacial dysmorphism, congenital heart defects, and myeloproliferative disorders, such as juvenile myelomonocytic leukemia (8). Many NS patients are highly functional, despite requiring specialty care to manage aspects of their condition. However, available therapies do not address the root cause of NS directly. Mutations in multiple RAS-associated signaling pathway genes can cause NS (9 -11) with approximately 5% occurring in KRAS (12). KRAS V14I and T58I substitutions are particularly associated with NS (12). V14I is also seen in a limited number of large bowel tumors (14) and lung tumors (11), although at a much lower level than the classical KRAS G12V oncogene. KRAS V14I has been extensively characterized from a biological standpoint, given the development of a genetically engineered mouse model that phenocopies key aspects of Noonan syndrome (9 -11). Nevertheless, the structural basis for the reported phenotypes has not been investigated.
RAS proteins are GTPases that function as molecular switches, cycling between active, GTP-bound, and inactive, GDP-bound, states, and regulating multiple pathways such as the MAPK pathway (1). RAS proteins contain two mobile structural elements, switch 1 (SW1) and switch 2 (SW2), which, based on their conformation, regulate many biological functions of RAS including its enzymatic activity and other proteinprotein interactions (15)(16)(17). SW1 primarily interacts with downstream effectors such as RAF and PI3K, whereas SW2 is more involved in GTP hydrolysis, although it also participates in direct interactions with some RAS effectors (16,18,19). Guanine exchange factors (GEFs) and GTPase activating proteins (GAPs) engage both SW1 and SW2 during catalysis of nucleotide exchange (20) and GTP hydrolysis (21), respectively. Differences in switch dynamics between disease-associated RAS mutants translate into variable biochemical behaviors such as nucleotide-binding affinity, GTP hydrolysis, and protein-protein interactions (22,23). Structural evaluation of these RAS variants is crucial to understanding the mechanisms that drive biochemical and disease phenotypes (21,24,25).
Many activating RAS mutations, such as the common KRAS mutation G12D, function by impairing GTPase activity. However, KRAS V14I displays an intermediate level of intrinsic and GAP-mediated GTPase activity when compared with other oncogenic isoforms (12). Moreover, KRAS V14I shows increased nucleotide-exchange activity, which is likely responsible for the accumulation of its active, GTP-bound state (9). Rapid nucleotide exchange is a poorly understood mechanism for RAS activation, but has been hypothesized to be caused by structural changes in RAS itself, particularly in the switch regions (2).
We solved the crystal structure of KRAS V14I :GDP and found that nucleotide exchange is facilitated by destabilization and opening of the GTP-binding site, similar to other classes of rapid nucleotide-exchanging RAS mutants such as KRAS A146T . Hydrogen-deuterium exchange (HDX)-mass spectrometry (MS) confirmed the open conformation occurs in solution, but with different kinetics than KRAS A146T . This open conformation enhances the affinity of KRAS V14I for the GEF SOS1. This atypical mechanism of RAS activation may provide new opportunities for targeting open conformations of RAS.

Crystal form 1: extended SW1 conformation
We solved crystal structures of KRAS V14I in complex with GDP in two different space groups. The data collection and refinement statistics are in Table 1. The P32 1 form, form 1, diffracted to ϳ1.5 Å, showed a single molecule in the asymmetric unit and demonstrated a remarkable extended SW1 and ␤2 conformation with respect to KRAS WT (Fig. 1, A and B). Electron density is clear and defined for all SW1 residues (Fig. 1C). The SW1 conformation is constrained by few crystal contacts and the refined model demonstrates a low B-factor, suggesting this conformation is not an artifact of crystallization (Fig. 2). In most monomeric RAS structures, SW1 starts just after ␣1 and extends to ␤2, participating in nucleotide binding and forming interactions with the base, sugar, and phosphate groups (22,26,27). In addition, previous structures showed an antiparallel interaction between ␤2 and ␤3. However, in the KRAS V14I structure, the C-terminal region of ␣1 undergoes a large conformational change, forming a new anti-parallel ␤ interaction with what was originally ␤2 (now ␤3 new ). In this conformation, the usual ␤2-␤3 interaction is disrupted and SW1 no longer participates in GDP binding. This conformation is essentially the same as KRAS A146T in complex with GDP (28).
The conformation of SW2 is also altered; an unsurprising finding given that SW1 and SW2 are immediately adjacent and demonstrate concerted movements in other structures (23). We also note a lack of density for the magnesium ion from the active site (Fig. 1D). In the absence of magnesium and ␤2-␤3 interactions, greater flexibility is observed in SW2, leading to decreased helicity, causing SW2 residues to extend further into the solvent relative to the KRAS WT (Fig.  1, A and B). As a result of these changes, the catalytic residue Gln 61 and p-loop residues 10 GAGGVGKS 17 are highly exposed to solvent, with a solvent-accessible area for the protein and GDP of 2621 and 57 Å 2 , respectively. Both values where F obs is from a test set of reflections that are not used in structural refinement.

Structural basis of KRAS V14I activation
are higher than in the KRAS WT structure (PDB code 4obe) at 2290 and 26 Å 2 , respectively (29). As a consequence of SW1 and SW2 movements and the loss of the active-site magnesium, crystal form 1 displays alterations in nucleotide interactions found in prior RAS structures that may explain the high rate of nucleotide exchange observed for KRAS V14I (30). In KRAS WT :GDP (PDB code 4obe), the main chain carbonyl of Asp 30 hydrogen bonds with ribose, and Ser 17 bonds to the ␤-phosphate oxygen (22). In the V14I mutant both interactions are lost. Also, the guanine base is typically sandwiched between Phe 28 and Lys 117 side chains (31), but this interaction is also lost in KRAS V14I (Fig. S1).
Finally, we noted electron density consistent with oxidation of the thiol at Cys 118 . This modification has previously been shown to increase the nucleotide-exchange rate in RAS (13), suggesting that this may contribute to the observed biochemical properties of this mutant. However, it also suggests that Cys 118 is exposed in solution, similar to how Cys 118 becomes exposed when RAS is bound to SOS1. This may hold significance for potential therapeutic efforts, given that Cys 118 has been targeted by covalent small molecules in RAS:SOS complexes (32).

Crystal form 2: disorder in the switches
KRAS V14I also formed crystals in the C2 space group that diffracted to ϳ1.6 Å, and included three molecules in the asymmetric unit (data collection and refinement statistics in Table  1). Unlike crystal form 1, the switch regions are generally disordered, with the exception of one molecule where density for SW2 is observable and shows a closed, helical (58 -62) conformation (Fig. 1E). The side chain of catalytic residue Gln 61 is oriented toward the nucleotide phosphate group and interacts via solvent molecules. These interactions would be predicted to allow transitions between the inactive and active state (33,34). No electron density was observed for 6 -9 residues in SW1, depending on the molecule, but the observable ␤2-␤3 motif formed an antiparallel interaction, similar to closed KRAS structures in the PDB. This feature suggests that GDP-bound KRAS V14I may also exist in solution in a more typical closed conformation. The core G-domains of both forms 1 and 2 are similar, with exceptions in Tyr 32 , a residue previously implicated in rapid nucleotide-exchange mechanisms (35,36). The position of Tyr 32 is disordered in all three monomers. p-Loop residues 10 GAGGVGKS 17 are also highly exposed to solvent,

Structural basis of KRAS V14I activation
similar to the form 1 structure, and no magnesium ion density is observed.

Hydrogen-deuterium exchange in solution is consistent with form 1
Protein crystallization requires that relatively stable protein conformations exist in solution for crystals to form. Nevertheless, it is theoretically possible that protein conformations could be driven by crystal contacts themselves. Nevertheless, in this particular case, the atomic displacement parameter, Wilson B-value, of the diffraction data are 13.68 Å 2 and the overall protein isotropic B-factor is below 18 Å 2 suggesting that the protein conformation is highly stable within the crystalline lattice. Nevertheless, to confirm that the protein conformation observed in crystal form 1 exists in solution, we used HDX-MS to evaluate switch dynamics in RAS, given that HDX-MS has been used previously to measure alterations in RAS switch dynamics (17,37,38). Here we directly compared HDX of KRAS V14I :GDP versus KRAS WT : GDP. We also compared these data with a previously published dataset KRAS A146T :GDP (28). The HDX-MS results support the validity of the crystal structures that these structures exist in solution, and that they are not the result of crystal contacts. We observed a faster deuterium exchange in peptides that cover both the p-loop and the N terminus of ␣1 (12)(13)(14)(15)(16)(17)(18)(19) as well as areas beneath the p-loop (115-124) when compared with KRAS WT (Fig. 3 and Fig. S3). At residues 23-39, the deuterium exchange rate was initially slower than in KRAS A146T , but similar to the exchange rate in KRAS WT (Fig. S3). We also observed higher exchange in loops beneath the nucleotide ring (132-144), likely related to rapid nucleotide exchange. Although the estimated time that KRAS V14I spends in the open state, relative to KRAS A146T , cannot be fully estimated by these data, they do support that KRAS A146T spends more time overall in the open state. This finding could explain, in part the ability of A146T mutant to form tumors, whereas V14I-associated tumors are less common (39).

GDP displacement leads to SW1 instability in V14I
The mechanism of SW1 destabilization is not immediately apparent, given that residue 14 is not located in the switches. Residue 14 is in the p-loop, opposite the nucleotide relative to SW1. We therefore explored whether the mechanism of SW1 destabilization was due to displacement of the nucleotide because of the substitution at residue 14, which incorporates an extra buried carbon. Previous work has shown that the nucleotide binds to RAS with picomolar affinity, which can be rationalized by multiple, specific interactions between the nucleotide ligand and the protein-binding pocket that highly constrain its position (40,41). In particular, the position of the guanine base is stabilized by four key residues, namely Phe 28 , Lys 117 , Asp 119 , and Lys 147 (Fig. 4A), including a sandwich interaction between Phe 28 and Lys 117 side chains and the guanine base.
We posited that addition of a carbon in the side chain at position 14, as seen with the V14I substitution, has the potential to displace the nucleotide. The KRAS V14I structures themselves provide evidence of small rearrangements surrounding position 14 that could influence nucleotide position. For example, in V14I we see loss of an interaction between the carbonyl oxygen of Ala 11 and the main chain nitrogen of Ile 14 relative to WT. To further demonstrate V14I-related changes, we aligned KRAS V14I , using the C␣ residues excluding the switch regions, with other X-ray structures of KRAS-GDP and compared the distances between structural elements adjacent to the p-loop as well as relative distances between those elements and the nucleotide. We expected that an additional carbon in Ile 14 would push the main chain of the p-loop away from structural elements below the p-loop. We compared the change in the distance between C ␣ for each structure relative to KRAS WT (PDB 4obe) and saw an average difference of 0.79 Å (S.D. 0.02),  Fig. 4 and Table S1. B, regions of KRAS V14I (highlighted in red) demonstrate increased deuterium exchange relative to KRAS WT .

Structural basis of KRAS V14I activation
whereas other KRAS structures, including A146T, showed an average difference of 0.20 Å (S.D. 0.02) from WT (Fig. 4, B and C, red bars). The differences were statistically significant (p Ͻ 1.2 E-8). The intramolecular distance between C ␣ atoms of residues Ile 14 and Phe 82 for each structure also show similar differences from V14I (Fig. 4, B and C, blue bars). The net result of these changes is that the nucleotide base is pushed up from below, altering the pitch of the nucleotide by ϳ14 degrees (Fig.  4D). Of note, a similar change in the pitch of the nucleotide base also occurs for KRAS A146T , but for different reasons. In KRAS A146T , formation of a new hydrogen bond between N7 of the purine base and the side chain O of the threonine substitution at position 146, which sits on the opposite side of the nucleotide relative to position 14, pulls the guanine ring toward Thr 146 and SW1 (28). As a result, the purine base is tilted by ϳ14 degrees. Altogether, these observations suggest that the remarkable similarities in the extended open conformation of SW1 between KRAS A146T and KRAS V14I structures, together with the shared rapid nucleotide-exchange phenotype, are evidence of a common underlying atypical mechanism of RAS activation.

Open SW1 leads to enhanced interactions with SOS
The open SW1 conformation exposes the binding interface accessed by the GEF SOS1 (Fig. 5A), suggesting that even though intrinsic nucleotide exchange is elevated for this class of exchange mutants, this protein conformation may further enable interactions with GEFs, leading to additional enhancements in nucleotide exchange. We performed protein-protein affinity measurements to evaluate the interaction between V14I and SOS1. Evaluation of protein affinity between RAS and SOS1 by microscale thermophoresis (MST) demonstrated ϳ40-fold increase in binding affinity with SOS1:KRAS V14I affinity at 220 nM, whereas SOS1:WT was 8300 nM (Fig. 5B). This demonstrates that although the intrinsic rate of GDP dissociation in KRAS V14I is high, the open conformation enhances GEF interactions, further accelerating nucleotide exchange.

Discussion
KRAS V14I , together with KRAS A146T , constitute a unique class of RAS mutations that are activated primarily by rapid intrinsic nucleotide cycling, driven by altered protein dynamics. We provided the structural basis for rapid nucleotide exchange, specifically that the nucleotide-binding pocket is disrupted by changes in nucleotide positioning due to V14I substitutions, resulting in a major structural reorganization centered around SW1. These changes are accompanied by a loss of the active-site magnesium ion. Both aspects are reminiscent of structures of HRAS bound to the SOS catalytic pocket, suggesting that these structures could provide snapshots of intermediate steps involved in SOS-mediated nucleotide exchange (20).
The biological effects of the altered dynamics of SW1 in V14I, and the functionally similar A146T, also appear to extend beyond nucleotide exchange, to other processes essential for

Structural basis of KRAS V14I activation
KRAS signaling. It is tempting to speculate from the open KRAS V14I :GDP and KRAS A146T :GDP structures is that GTPbound proteins might also present an open SW1, which is incompatible with RAS-RAF interactions (34). However, this does not appear to be the case based on studies of KRAS V14I and KRAS A146T in biological systems. For example, interrogation of RAS signaling in a genetic mouse model of KRAS V14I demonstrated up-regulation of GTP-bound KRAS V14I as measured by pull downs using the Ras-binding domain of the c-RAF. These findings are incompatible with an extended SW1, suggesting that KRAS V14I adopts a closed conformation when GTPbound. Nevertheless, these mice showed little up-regulation of pERK and pAKT relative to WT throughout their development (9). KRAS A146T :GTP also retains the ability to bind RAF, but KRAS A146T mouse models showed modest up-regulation of pERK and pAKT relative to WT, although significantly less than KRAS G12D (28). Together, these data demonstrate that nucleotide-exchange mutants such as KRAS V14I remain competent to bind RAF, and signal through the MAPK pathway, but are attenuated in signaling, likely explaining the less aggressive clinical phenotypes relative to mutations characterized primarily by GAP-insensitivity such as KRAS G12D .
These findings may be explainable by our observations that active site destabilization in KRAS V14I and KRAS A146T results in increased interactions with GEF (30). Presumably, this is brought about by the GEF-binding surface of KRAS V14I being exposed in the open SW1 conformation. Whether this interaction leads to a dependence on GEFs in vivo is unknown. However, recent data on KRAS A146T is suggestive (28). In that study, tissues from mice expressing KRAS A146T or KRAS G12D were analyzed for alterations in specific kinase pathways. Up-regulation of p90RSK activity in KRAS A146T was noted. This is potentially significant given that p90RSK is a negative regulator of SOS1 (42). One possibility is that down-regulation of GEF activity via p90RSK is able to suppress nucleotide exchange to some degree, explaining the relatively mild phenotypes seen with nucleotide-exchange mutants. However, it also raises the possibility of discovering synthetic lethal interactions within these regulatory frameworks that could be exploited for therapeutic purposes. Whether other regulators of GEFs play a role in modulating the activity of nucleotide exchangers is another critical question.
Although this study places KRAS V14I and KRAS A146T in the same functional category, it is unclear if other RAS mutants with rapid exchange phenotypes share similar mechanisms and regulation. For example, KRAS G13D has been identified as a rapid exchanger, but the mechanism of exchange is thought to relate to altered electrostatics of the nucleotide-binding pocket, not protein dynamics (22). Given this fundamental difference, it seems likely that regulatory mechanisms of G13D activation, if any, will be different for V14I or A146T. On the other hand, KRAS K117N , another nucleotide-exchange mutation found mostly in colorectal cancer (39), may be subject to similar regulatory mechanisms given that the mutation occurs near the guanine base, similar to A146T.
Finally, the open conformation of SW1 in this mutant class raises the possibility of directly and selectively targeting this class of mutants using covalent inhibitors. As noted, we identified that Cys 118 is oxidized in our crystal structures, suggesting that this residue is exposed to solvent. Interestingly, this residue has been targeted in the context of the RAS:SOS complex using covalent ligands (32). Selectively targeting this cysteine in uncomplexed exchange mutants due to the open SW1 conformation is an intriguing possibility.

Protein expression and purification
A bacterial expression construct encoding human KRAS V14I lacking the C terminus residues, 170 MSKDGKKKKKKSKTKC-VIM 188 (expressed residues 1-169), that includes an N-terminal His 6 tag and tobacco etch virus cleavage sequence was made by site-directed mutagenesis starting from a WT KRAS expression construct. Escherichia coli BL21 (DE3) cells were transformed and grown in Luria-Bertani (LB) medium at 37°C to an A 600 of 0.8 -1.0. Protein expression was induced by the addition of 0.5 mM isopropyl ␤-D-thiogalactopyranoside. Cells were collected by centrifugation after 16 h at 16°C and lysed in lysis buffer (50 mM Tris, pH 8.0, 500 mM NaCl, 2 mM MgCl 2 , 5 mM 2-mercaptoethanol, 1 mM benzamidine, and 1 mM PMSF).

Structural basis of KRAS V14I activation
Lysate was clarified by centrifugation at 51,000 ϫ g for 60 min at 4°C. The supernatant was applied to an IMAC cartridge (Bio-Rad Laboratories, Inc.) and protein was desalted into crystallization buffer (20 mM Tris, pH 8.0, 150 mM NaCl, 5 mM MgCl 2 , 1 mM DTT). His-tagged tobacco etch virus protease (1:5, w/w) was added to the elute, and protein was incubated at 4°C overnight. Finally, the cleaved protein was passed through an IMAC cartridge to isolate from the tag. Further fractionation was done using a Superdex 75 size exclusion column. The peak fractions were analyzed by SDS-PAGE and the fractions containing KRAS V14I were pooled and concentrated using an Amicon Ultra Centrifugal Filter with a molecular mass cutoff of 10,000 Da to 30 mg/ml and stored at Ϫ80°C.

Crystallization and data collection
Concentrated KRAS V14I (30 mg/ml) in a buffer composed of 20 mM HEPES and 20 mM NaCl, pH 8.0, was used for crystallization trials. Initial crystallization trials were performed with a Mosquito (TTP LabTech) crystallization robot using the sitting-drop method (100 nl of protein plus 100 nl of crystallization solution). Optimized single protein crystals were grown using hanging drop vapor diffusion at 20°C in a reservoir solution containing 1.8 M sodium phosphate monobasic monohydrate and potassium phosphate dibasic, pH 8.2. Three-weekold crystals were mounted with 15% glycerol in mother liquor as cryoprotectant and flash cooled in liquid nitrogen. X-ray diffraction data were collected at the Advanced Photon Source beamline 19-BM at 100 K. The crystal belonged to the P32 1 space group and diffracted to 1.49 Å. The second crystal was grown in the same conditions and 2-week-old crystals were harvested and flash cooled in liquid nitrogen. The crystal belonged to the C2 space group and diffracted to 1.6 Å at the Advanced Photon Source beamline 19-ID at 100 K. The initial diffraction data were indexed, integrated, and scaled using HKL2000/3000 (43).

Structure determination and refinement
Crystal structures were determined by molecular replacement using WT KRAS (PDB 4obe) as an initial search model with Phaser in the CCP4 Suite (44). The models were further built and refined using Phenix (45) and COOT (46) by manual model correction. Solvent molecules were added, and the final refinement performed by phenix.refine. Stereochemical parameters of the final model were analyzed with the assistance of Molprobity (47). Data collection and refinement statistics are listed in Table S1.

HDX-MS of KRAS V14I
HDX-MS experiments were performed similarly to previous reports (17,48). Purified proteins were diluted into sample buffer (10 mM HEPES, pH 8.0, 5 mM MgCl 2 , 5 M GDP, 150 mM NaCl) to a concentration of 10 M prior to labeling. Samples were deuterated for the times described using labeling buffer, and labeling was stopped with an equal volume of quench buffer (Table S1). Quenched samples were immediately analyzed. Deuterium measurement with MS was performed as previously described (17, 48) using a Synapt G2si HDMSE, online digestion Enzymatic column, 1 ϫ 50, 1.8-m HSS T3 separation, 1.7-m BEH trap. Deuterium incorporation graphs were generated using DynamX 3.0 software (Waters) by subtracting the centroid of the isotopic distribution at each labeling time point from the centroid of the isotopic distribution of the undeuterated reference species. Because the data were not corrected for back-exchange, each data point represents the relative deuterium level at each time point for each peptide (49).

MST with KRAS V14I and SOS1
KRAS V14I (100 M) was incubated with 200 M GDP in 1ϫ PBS at room temperature for 1 h. The solution was then desalted into 1ϫ PBS with 2 mM MgCl 2 and 1% Tween 20 via a Zeba TM (Thermo Fisher) spin desalting column. 16-Point serial dilutions of KRAS V14I were prepared and mixed 1:1 with GFP-556 SOS 1049 to a final volume of 20 l. The reaction mixtures were loaded into premium-treated capillaries and analyzed by a Monolith NT.115 (Nanotemper Technologies) at 60% MST power and 40% LED power with a laser on-time of 30 s. The K D was calculated by taking the average of triplicate F norm measurements at each concentration and fitting the data by using Palmist (50) and plotting in Prism 7. Raw data have been normalized and converted to percent response (fluorescence signal of lowest and highest concentration of KRAS V14I are set as 0 and 100%, respectively).