Allosteric modulation of NF1 GAP: Differential distributions of catalytically competent populations in loss‐of‐function and gain‐of‐function mutants

Abstract

Neurofibromin (NF1), a Ras GTPase‐activating protein (GAP), catalyzes Ras‐mediated GTP hydrolysis and thereby negatively regulates the Ras/MAPK pathway. NF1 mutations can cause neurofibromatosis type 1 manifesting tumors, and neurodevelopmental disorders. Exactly how the missense mutations in the GAP‐related domain of NF1 (NF1^GRD) allosterically impact NF1 GAP to promote these distinct pathologies is unclear. Especially tantalizing is the question of how same‐domain, same‐residue NF1^GRD variants exhibit distinct clinical phenotypes. Guided by clinical data, we take up this dilemma. We sampled the conformational ensembles of NF1^GRD in complex with GTP‐bound K‐Ras4B by performing molecular dynamics simulations. Our results show that mutations in NF1^GRD retain the active conformation of K‐Ras4B but with biased propensities of the catalytically competent populations of K‐Ras4B–NF1^GRD complex. In agreement with clinical depiction and experimental tagging, compared to the wild type, NF1^GRD E1356A and E1356V mutants effectively act through loss‐of‐function and gain‐of‐function mechanisms, leading to neurofibromatosis and developmental disorders, respectively. Allosteric modulation of NF1^GRD GAP activity through biasing the conformational ensembles in the different states is further demonstrated by the diminished GAP activity by NF1^GRD isoform 2, further manifesting propensities of conformational ensembles as powerful predictors of protein function. Taken together, our work identifies a NF1^GRD hotspot that could allosterically tune GAP function, suggests targeting Ras oncogenic mutations by restoring NF1 catalytic activity, and offers a molecular mechanism for NF1 phenotypes determined by their distinct conformational propensities.

1. INTRODUCTION

Neurofibromatosis type 1 is one of the most common genetic diseases caused by inherited or de novo germline mutations in tumor suppressor gene NF1 on chromosome 17q11.2 (Yap et al. 2014). Its phenotypes are highly variable, showing a wide range of clinical manifestations (Kiuru and Busam 2017). Patients with the disease develop benign peripheral nerve tumors and plexiform neurofibromas that can form malignant peripheral nerve sheath tumors (MPNSTs) (Ratner and Miller 2015). The optic pathway glioma (OPG) is another common tumor in individuals with the disease (Ratner and Miller 2015). Children with the disease have increasing risk for neurodevelopmental disorders including learning deficits, autism symptomatology, and impairment in attention and social skills (Vogel et al. 2017). Mutations in the NF1 gene can also lead to abnormalities in the cardiovascular and musculoskeletal systems (Friedman et al. 2002; Patel and Stacy ²⁰¹²).

Tumor suppressor protein neurofibromin (NF1) is a multifunctional protein (Baez‐Flores et al. ²⁰²³; Gutmann et al. ¹⁹⁹¹). The well‐characterized GTPase‐activating‐protein‐related domain (GRD) functions as Ras GTPase‐activating protein (GAP) by accelerating Ras‐mediated GTP hydrolysis, thus negatively regulating Ras/MAPK signaling pathway (Rauen 2022). More than 5000 somatic variants have been identified in human tumors, spreading over the entire NF1 gene (Baez‐Flores et al. ²⁰²³). Mutations on the GRD of NF1 (hereafter referred to as NF1^GRD) are of particular interest as they are closely related to NF1 pathogenesis (Scheffzek and Shivalingaiah 2019). Loss‐of‐function mutation R1276P/Q significantly reduced the NF1 GAP catalytic efficiency (~8000‐fold or 500‐fold reduction compared with the wild type), with Arg1276 being the most essential catalytic residue (arginine finger) for GAP‐stimulated GTP hydrolysis reaction of Ras‐GTP (Ahmadian et al. 1997; Fahsold et al. ²⁰⁰⁰; Klose et al. ¹⁹⁹⁸). The missense mutation R1391S identified in NF1 patients displayed ~300‐fold lower activity than the wild‐type NF1^GRD (Upadhyaya et al. 1997). A subsequent crystal structure of NF1^GRD in complex with Ras revealed that Arg1391 is located in the binding interface between NF1^GRD and Ras (Rabara et al. 2019). Other patient‐derived mutations (e.g., mutations identified from patients, such as E1356G/K and S1463F) that are distant from the catalytic site and the binding interface between NF1 and Ras were also reported (Scheffzek and Shivalingaiah 2019), but the correlation between the impact of these mutations on GAP activity and their clinical significance is less clear.

Same‐gene mutations can promote both cancer and neurodevelopmental disorders (Nussinov et al. 2022; Nussinov et al. ^2023c), as shown by mutations on Src homology 2 domain‐containing protein tyrosine phosphatase 2 (SHP2) encoded by the PTPN11 gene (Liu et al. ^2023b), by multiple mutations in phosphatase and “tensin” homolog (PTEN) phosphatase, for example, H93Y, H93D, H93Q; T131I, T131N; C105F, C105F/Y (Jang et al. 2021; Jang et al. ²⁰²³), and by K‐Ras G12D (Schmidt et al. 2021). The distinct phenotypes of mutational variants on a protein can be predicted from their relative conformational propensities (Nussinov et al. ^2023a; Nussinov et al. ^2023b). K‐Ras4B harboring cancer driver mutations prefers an active conformation (state 2) in favor of interactions with effectors (Kolch et al. 2023; Nussinov et al. ^2023a). The Ras‐homology small GTPase RhoA mutants (A161P and A161V) in the GDP‐bound state favor a conformation resembling the active conformation of wild‐type RhoA, endowing effector binding activity (Lin et al. 2024). NF1^GRD seems to follow the same scenario. The E1356A mutant from the ClinVar database (Landrum et al. 2014), as well as the E1356G/K in other reports, are associated with neurofibromatosis type 1. Due to its rare occurrence, the E1356A variant was classified as “Uncertain significance,” indicating that it is still unclear whether it is pathogenic or benign. The E1356V mutant from the de novo database (https://denovo-db.gs.washington.edu/denovo-db/) shows that E1356V is related to neurodevelopmental disorders (Deciphering Developmental Disorders Study ²⁰¹⁷). How mutations of the same residue allosterically differentially modulate NF1 GAP activity is puzzling. Insight into such mechanisms is critical, as it may offer novel strategies for repressing active K‐Ras4B mutants by increasing RasGAP activity (Berta et al. 2023). Besides mutations, the alternative splicing of exon 23a of the NF1 gene inserts additional 21 residues into the NF1^GRD domain, leading to NF1 isoform 2 (Andersen et al. 1993). This NF1 variant suppresses its GAP activity (Hinman et al. 2014), dysregulates Ras/ERK signaling, and impairs learning and memory in mouse models (Cui et al. 2008; Nguyen et al. ²⁰¹⁷). However, the structural basis for the regulatory role of this insertion in NF1 GAP activity has been unclear.

Here, we performed molecular dynamics (MD) simulations to explore the conformational propensities of NF1^GRD in complex with GTP‐bound K‐Ras4B (Figure 1). Our results show that NF1 binding retains the conformation of the catalytic site of K‐Ras4B, making it insensitive to NF1^GRD mutations. The complementary electrostatic interactions between NF1^GRD and K‐Ras4B contribute to the stability of the GTPase transition state. Compared to the wild‐type NF1^GRD, E1356A and E1356V mutants populate the catalysis‐competent conformations differently, leading to a loss of function and gain of function, respectively. In contrast, the 23a insertion of NF1 isoform 2 distorts the structure of the active site of K‐Ras4B and populates the inactive conformation of K‐Ras4B. Our results suggest that the NF1 GAP activity can be allosterically modulated, offering a new avenue for the development of NF1^GRD mutants/binders that fine‐tune its GAP activity.

FIGURE 1.

Sequence and modeled structure of NF1 GRD in complex with GTP‐bound K‐Ras4B. (a) Sequence of K‐Ras4B including residues Met1‐Lys169, and NF1 GRD (NF1^GRD for short) including residues Phe1205‐His1530. The position of the arginine finger Arg1276 in NF1 GRD is indicated by an arrow. The nonpolar, basic, acidic, and polar residues are colored black, blue, red, and green, respectively. (b) The initial structure of K‐Ras4B–NF1^GRD complex. The Switch I (residues 25–40) and Switch II (residues 60–76) regions of K‐Ras4B are colored yellow and magenta, respectively. The GTP molecule is shown as stick representation and Mg²⁺ is shown as a green sphere. The arginine finger Arg1276 and Glu1356 are also shown as stick representation. Note that the initial complex is in the ground OFF‐state in which Arg1276 has not interacted with the γ‐phosphate of GTP (highlighted on the right panel).

Allostery is complex and how nature exploits it for function is riveting, as shown by the remarkable series of papers by the Melacini's lab (Akimoto et al. 2022; Byun et al. ²⁰²⁰), and the recent observations by the Hristova lab (Karl et al. 2024; Krzyscik et al. ²⁰²⁴; Wirth et al. ²⁰²⁴). Their work documented that different FGF ligands, and mutations, can trigger dramatically different biological responses of a single FGFR receptor through altered phosphorylation events, and especially, the quantification of the induced bias in EGFR phosphorylation (Wirth et al. 2023). The mechanisms explored here could similarly act via differential allosteric biasing of the conformational ensembles.

2. RESULTS

2.1. Mutations on NF1^GRD retain the active conformation of K‐Ras4B

The initial complex structure shows that K‐Ras4B‐GTP interacts with NF1^GRD in the ground OFF‐state, in which the side chain of the arginine finger (Arg1276) has not moved into the active site of K‐Ras4B (Figure 1b). In contrast, the ground ON‐state refers to the conformational state in which the side chain of Arg1276 has interacted with the γ‐phosphate of GTP. To investigate the conformational change of K‐Ras4B upon binding of wild‐type and mutant NF1^GRD, we examined the coordination environment of the active site of K‐Ras4B. Mg²⁺ is hexa‐coordinated with one oxygen from each of the β‐ and γ‐phosphates, the two side chain hydroxyl groups of Ser17 and Thr35, and two water molecules. In addition, Gly13, Lys16, and Gly60 coordinate the GTP (Figure 2a). Two stably Mg²⁺‐coordinated water molecules were observed throughout MD simulations of all systems. We calculated the corresponding distances between the protein residues and GTP/Mg²⁺ and found that our simulation results were in good agreement with the experimental values (Figures 2b–d and S1), indicating that mutations (E1356A/V and R1276P/Q) on the NF1^GRD have little impact on the conformation of the active site of K‐Ras4B. Notably, we observed that the arginine finger moved into the active site of K‐Ras4B in the complexes comprising the wild‐type and the E1356V mutant of NF1^GRD, but not always in the complex comprising E1356A mutant. The involvement of Arg1276 of NF1 in the active site of K‐Ras4B indicated that the complex switched from the ground OFF‐state to the ground ON‐state. Such a state transition seems less favorable for NF1^GRD carrying the E1356A mutation as the state transition was only observed in one of the three replica simulations (Figure 2c). In contrast, for the wild‐type and E1356V mutant, the state transition was observed in all replica simulations (Figure 2b,d). These data suggest that the long‐range mutations (E1356A and E1356V) on the NF1^GRD may affect the catalytic activity of NF1 GAP differently.

FIGURE 2.

Interactions within the active site of K‐Ras4B–NF1^GRD complexes. (a) The active conformation of the catalytic site of K‐Ras4B. Mg²⁺ is hexa‐coordinated with the β‐ and γ‐phosphates of GTP, the side chain of Ser17 and Thr35, and two water molecules. Gly13, Lys16, and Gly60 coordinate the GTP. In the ground ON‐state, Arg1276 interacts with the γ‐phosphate of GTP, and the distance between Arg1276 and GTP was calculated based on the distance between the nitrogen atom of the “amino group” of Arg1276 and the oxygen atom of the γ‐phosphate of GTP. (b) The distances corresponding to the Mg²⁺ and GTP coordination, as well as between Arg1276 and GTP, calculated for wild‐type NF1^GRD in complex with K‐Ras4B. (c) The corresponding distances for K‐Ras4B–NF1^GRD complex with E1356A mutant. (d) The corresponding distances for K‐Ras4B–NF1^GRD complex with E1356V mutant. The hollow circle, square, and triangle represent data from three independent trajectories of the same system. The black square shown in (b) represents experimental value taken from the crystal structure (PDB ID: 6OB2). The interactions in the active site of K‐Ras4B indicate that K‐Ras4B–NF1^GRD complex harboring E1356 mutant retains the active conformation of K‐Ras4B when it transitions from the ground OFF‐state to the ground ON‐state.

2.2. E1356 mutants are primed for GAP‐catalyzed GTP hydrolysis

To obtain the representative conformation of each complex, we merged three replicas of each system and performed clustering analysis using the ensemble clustering method implemented in Chimera software (Kelley et al. 1996; Pettersen et al. ²⁰⁰⁴). Conformationally related clusters were generated based on the best pairwise root‐mean‐square deviations (RMSDs) between different conformations with automatically determined cut‐off values in the clustering (Kelley et al. 1996). The representative conformations were selected from the largest cluster of each system and superimposed with each other. Here we focused on the conformational change of K‐Ras4B and compared with the crystal structure of wild‐type GTP‐bound K‐Ras4B (Figure 3a). The calculated Cα RMSD between different conformations and the wild‐type K‐Ras4B is 1.3 Å, suggesting that the binding of E1356 mutants of NF1^GRD has little effect on the active conformation of K‐Ras4B. Of particular interest are the orientations of two residues, Tyr32 and Thr35, on Switch I of K‐Ras4B. In the initial, ground OFF‐state, Tyr32 forms a “hydrogen bond” by interacting with the oxygen atom of the γ‐phosphate of GTP via its hydroxyl group, blocking the interaction between the arginine finger Arg1276 and the GTP (Figure S2). In the ground ON‐state, we observed similar rearrangement of the orientation of Tyr32 in different systems, which allows the arginine finger to interact with the γ‐phosphate of GTP. The influence of Tyr32 on GTP hydrolysis has been studied by simulations and Fourier transform infrared experiments (Li et al. 2018; Rudack et al. ²⁰¹⁵). Here, we calculated the distributions of the dihedral angle C‐Cα‐Cβ‐Cγ of Tyr32 in different systems (Figure S3). We observed that Tyr32 in the isolated K‐Ras4B populates two conformational states, corresponding to the Tyr32_OUT and Tyr32_IN substates observed earlier (Li et al. 2018). When K‐Ras4B binds to NF1^GRD, Tyr32 predominantly adopts the Tyr32_OUT conformation, which is also consistent with the previous finding that Tyr32 uses the Tyr32_OUT substate to interact with GAPs (Li et al. 2018).

FIGURE 3.

Structural characterization of K‐Ras4B–NF1^GRD complexes. (a) Structure superimposition of the representative conformations of K‐Ras4B in the wild‐type (yellow), E1356A (green), and E1356V (magentas) K‐Ras4B–NF1^GRD complexes. For comparison, the crystal structure of GTP‐bound H‐Ras determined at 100 K is also shown (white, PDB ID: 1QRA). The orientations of Tyr32 and Thr35 of Ras are highlighted in the right panel. (b) The binding affinity between GTP‐bound K‐Ras4B and NF1^GRD is evaluated in terms of the binding free energy for different systems. (c) The volume of the active site of K‐Ras4B for different systems. (d) The number of water molecules within 4 Å of the γ‐phosphate of GTP, excluding the two water molecules that stably coordinated to Mg²⁺. The hollow circle, square, and triangle represent data from three independent replicas of the same system. Compared to the wild‐type NF1^GRD, E1356 mutants are primed for GAP‐catalyzed GTP hydrolysis.

The dynamics of Thr35 are critical for the conformation of Switch I, and Thr35 mutations shifted the active conformation of K‐Ras4B to the non‐binding conformations that have a reduced affinity to Ras effector proteins such as Raf and RalGDS (Matsumoto et al. 2016; Shima et al. ²⁰¹⁰; Spoerner et al. ²⁰⁰¹). Our results show that compared to the wild‐type K‐Ras4B, Thr35 exhibits a similar orientation upon binding to E1356‐mutated NF1^GRD, suggesting that the conformations of Switch I of K‐Ras4B are preserved. Structural superposition (Figure 3a) and the calculated distance between Gly60 (in Switch II) and GTP (Figure 2) imply that the conformations of Switch II of K‐Ras4B are also maintained. These results indicate that the E1356 mutants have little effect on the active conformation of the GTP‐bound K‐Ras4B.

Here we showed that both wild‐type and E1356 mutant NF1^GRD can switch to the ground ON‐state when bound to the active state of K‐Ras4B, with Switch I stabilized by the interaction of Thr35 with the nucleotide and Mg²⁺. The GAP‐stimulated GTP hydrolysis by K‐Ras4B needs to meet other structural and enzymatic requirements, including adequate binding affinity, appropriate volume of the active site, the number of available water molecules, and the proper orientation of Gln61 in Switch II of K‐Ras4B. We characterized these properties and delineated the difference between the wild‐type and mutant NF1^GRD.

The binding free energy is applied to estimate the relative binding affinity between the K‐Ras4B and NF1^GRD (Figure 3b). Compared to the wild‐type NF1^GRD (−47.5 ± 5.3 kcal/mol), the E1356V mutant shows an increased binding affinity for K‐Ras4B (−55.1 ± 12.1 kcal/mol), whereas the E1356A mutant shows a reduced binding affinity for K‐Ras4B (−17.7 ± 16.6 kcal/mol). The high standard deviation in the binding free energy of E1356A is because the electrostatic interaction between the arginine finger Arg1276 and GTP occurred only once in three replica simulations. This result is consistent with early experimental data showing that single‐point mutant R1276A/Q/N displayed up to three‐fold reduced affinity for GMPPNP‐bound Ras (Ahmadian et al. 1997). Based on previous experimental data (Ahmadian et al. 1997), we plotted the binding affinity with respect to the NF1 GAP activity, and found that there is no direct correlation between the two measurements (Figure S4). In addition, previous experimental data also showed that the binding affinity of NF1^GRD for GTP‐bound Ras is about 50‐fold higher than that of p120GAP (also known as RASA1) (Scheffzek et al. 1998), but the increase in the rate of GTP hydrolysis by NF1^GRD is similar to that by p120GAP (Ahmadian et al. 1997; Scheffzek et al. ¹⁹⁹⁸). Thus, the binding affinity between the NF1^GRD and K‐Ras4B seems a prerequisite but not determinant for the catalytic efficiency of NF1 GAP.

The volume of the active site pocket centered on the γ‐phosphate of GTP was calculated using POVME 3.0 (Durrant et al. 2011; Durrant et al. ²⁰¹⁴). For comparison, we performed 1 μs MD simulation on the nucleotide‐free K‐Ras4B and calculated the volume of the active site pocket. A value of 575.3 ± 66.9 ų is obtained, which is comparable to the value of 538.0 ± 93.8 ų calculated for the H‐Ras using the same program in a previous theoretical study (Calixto et al. 2019). In complex with NF1^GRD (Figure 3c), the volume of the active site pocket of K‐Ras4B reduces to 272.2 ± 79.8 ų. This value is very close to the calculated volume of the active site pocket of H‐Ras in complex with p120GAP (268.0 ± 33.5 ų) (Calixto et al. 2019). For the E1356 mutants, we find a slight decrease in volume of 241.1 ± 17.5 ų for E1356V and an increase in volume of 315.6 ± 53.4 ų for E1356A due to its generally low binding affinity between NF1^GRD and K‐Ras4B. Based on the proposed solvent‐assisted mechanism for GTP hydrolysis, the rate‐limiting step involves the nucleophilic attack of a water molecule, leading to a loose dissociative transition state followed by the proton transfer from the water molecule to the γ‐phosphate oxygen of GTP (Calixto et al. 2019). To examine the impact of the E1356 mutations on the solvation of the active site, we calculated the average number of water molecules within 4 Å of the γ‐phosphate group of GTP (Figure 3d). No significant change in the number of water molecules is observed due to different E1356 mutants. The presence of 2–3 water molecules in the vicinity of the γ‐phosphate is in agreement with the crystal structure of GTP‐bound Ras determined at a low temperature (100 K) (Scheidig et al. 1999), and the empirical valence bond simulation of GTP hydrolysis catalyzed by Ras and p120GAP (Calixto et al. 2019). Thus, the binding of E1356 mutant, like wild‐type NF1^GRD, appears to be able to maintain catalytic water molecules at positions around the γ‐phosphate of GTP.

Lastly, we characterized the orientation of Gln61, an important catalytic residue on the Switch II of Ras. Gln61 was suggested to contribute to GTP hydrolysis by orientating the nucleophilic water molecule toward the γ‐phosphate of GTP (Martin‐Garcia et al. ²⁰¹²; Scheffzek et al. ¹⁹⁹⁷; Zeng et al. ²⁰²¹). But its catalytic geometry was only found in the transition state of the catalytic reaction not in the ground state (Li and Zhang 2004). To examine the difference between wild‐type and mutated NF1^GRD, we calculated the distance between Gln61 and the γ‐phosphate of GTP for different systems, and find that the distances are all around 7 Å, which cannot stably position one nucleophilic water molecule in the vicinity of the γ‐phosphate of GTP (Figure S5a). Such a result is in line with previous observations in the ground state of Ras‐like GTPase and G‐protein α subunits (Li and Zhang 2004). The conformational space sampled by Gln61 was further characterized by the distributions of its two dihedral angles (φ and ψ, Figure S5b). The binding of E1356A mutant slightly increased the flexibility of the side chain of Gln61, but still predominantly sampled the same conformations as shown in the wild‐type and E1356V mutant.

Taken together, our results imply that there may exist a higher barrier that prevents E1356A mutant from switching to the ground ON‐state. However, our structural characterization shows no significant difference in the conformational state of K‐Ras4B and its catalytic site in the presence of wild‐type and E1356‐mutated NF1^GRD. It appears that all complexes are primed for the GAP‐catalyzed GTP hydrolysis by K‐Ras4B, which in turn raises an intriguing question of how the mutations on the NF1^GRD differentially affect the catalytic efficiency of K‐Ras4B. Below, we seek clues by examination of the interaction between K‐Ras4B and NF1^GRD.

2.3. E1356 mutants differentially populate catalysis‐competent conformations

The interaction between NF1^GRD and GMPPNP‐bound K‐Ras4B has been characterized in terms of the two available crystal structures in the absence (PDB ID: 6OB2) (Rabara et al. 2019) and presence of Spred1 (PDB ID: 6V65) (Yan et al. 2020). The protein–protein binding interface is stabilized by four conserved salt bridges: Asp37^Ras–Lys1419^NF1, Glu38^Ras–Lys1423^NF1, Glu62^Ras–Lys1283^NF1, and Glu63^Ras–Arg1391^NF1. In the presence of Spred1, an additional salt bridge is formed between Asp54^Ras and Arg1325^NF1. In our simulations of K‐Ras4B–NF1^GRD complex, we also observed another salt bridge between Asp33^Ras and Lys1436^NF1 that contributed to the association of the two proteins. We calculated the populations of these six salt bridges and observed significant differences in the populations where these interactions occur at the interface between Switch II of K‐Ras4B and NF1^GRD. Specifically, in the case of wild‐type NF1, Glu63 of K‐Ras4B could interact with Lys1283, Lys1290, and Arg1391, with Lys1290 being the preferred (Figure 4a). However, in the above crystal structures, Glu63 interacts with Arg1391. Such a difference could be due to (i) the crystal structures were solved using the GTP analog GMPPNP and (ii) both crystal structures are in the ground OFF‐state. As GMPPNP is a good analog of GTP for Ras (Scheidig et al. 1999), the switch from the ground OFF‐state to the ON‐state during our simulations, that is, with the interaction between the arginine finger Arg1276 and GTP, may cause a conformational change in the Switch II region of K‐Ras4B, consequently shifting the population harboring the Glu63–Arg1391 interaction. To support this, we calculated the relative populations of the salt bridges at the binding interface between K‐Ras4B and NF1^GRD for both ground OFF and ON states (Figure S6). Binding of K‐Ras4B to the wild‐type NF1^GRD maintained the interaction between Glu62 and K1283 in both the ON and OFF states but shifted the interaction of Glu63 with Arg1391 populated in the OFF state to the interaction with K1290 (most populated) in the ON state. This population shift was more pronounced in the E1356A mutant. In contrast, the Glu63–Arg1391 interaction was well maintained in the E1356V mutant.

FIGURE 4.

Different populations of the catalysis‐ready conformations of the wild‐type and mutated K‐Ras4B–NF1^GRD complexes. (a) Populations of salt bridges at the binding interface between K‐Ras4B and wild‐type NF1^GRD (left panel). The representative conformation of the active site of K‐Ras4B–NF1^GRD complex, with the Glu63–Lys1290 salt bridge being preferred (right panel). (b) Populations of salt bridges at the binding interface between K‐Ras4B and NF1^GRD E1356A (left panel). The representative conformation of the active site of K‐Ras4B–NF1^GRD E1356A complex, with the Glu63–Lys1283 salt bridge being slightly preferred (right panel). (c) Populations of salt bridges at the binding interface between K‐Ras4B and NF1^GRD E1356V (left panel). The representative conformation of the active site of K‐Ras4B–NF1^GRD E1356V complex, with the Glu63–Arg1391 salt bridge being highly preferred (right panel). These interactions suggest that wild‐type and E1356 mutants differently populate catalysis‐prone conformations. The hollow circle, square, and triangle represent data from three independent replicas of the same system. The same ensemble clustering method used for Figure 3a was applied to obtain the largest cluster, from which the representative conformation was selected with the most populated salt bridge highlighted.

Previous experimental finding showed that free arginine, imidazole, and other small nitrogenous molecule fail to accelerate GTP hydrolysis (Patel et al. 2023). As such, interaction of the arginine finger Arg1276 with GTP has a more profound effect on the conformational state of K‐Ras4B–NF1^GRD complex. As in the wild‐type NF1, we also observed the Glu63–Arg1391 interaction, with a lower population (37%) than the Glu63–Lys1290 interaction (60%) (Figure 4a). Compared to the wild‐type NF1^GRD, the Glu63–Arg1391 interaction is significantly reduced (to 6%) in the E1356A mutant, and Glu63 seems to switch its interaction between Lys1290 (44%) and Lys1283 (48%), with Lys1283 being slightly favored (Figures 4b and S7). In contrast, for the E1356V mutant, Glu63 predominantly interacts with Arg1391 (93%), but can also interact with Lys1290 with a much lower population (34%) (Figure 4c). The stable and specific salt bridges between the residues in Switch I (Asp33, Glu37, and Asp38) of K‐Ras4B with NF1^GRD provide a structural basis for the similar conformations of Switch I as observed in different systems (Figure 3a). This led us to ask whether the dynamics of the Glu63–Arg1391 interaction could be related to the NF1 GAP activity.

We characterized the conformational properties of the ground ON‐state of the NF1^GRD in complex with K‐Ras4B‐GTP. During GTP hydrolysis, a transition state analog formed by GDP, tetrafluoroaluminate (AlF₄ ⁻, mimic of γ‐phosphate of GTP), and RasGAP (NF1^GRD or p120GAP) was experimentally validated (Mittal et al. 1996). For Ras–RasGAP complex, the transition state of H‐Ras bound to p120GAP was solved (Scheffzek et al. 1997). In addition, crystal structures of the Rho family (Cdc42 and RhoA) bound to the RhoGAP were also determined in the ground OFF‐state and transition state, respectively (Rittinger et al. ^1997a; Rittinger et al. ^1997b). Previous studies identified the essential role of the arginine finger and Gln in the GTP hydrolysis (Bourne 1997), and recent theoretical studies focused on the molecular mechanism of the GTP hydrolysis (Calixto et al. 2020). Our simulations of the ground state of NF1^GRD in complex with K‐Ras4B‐GTP, whose crystal structure in the transition state is unavailable, may provide structural insights into the conformational transition of the GAP‐stimulated GTP hydrolysis.

We superimposed the four crystal structures, including the ground OFF‐state of K‐Ras4B bound to NF1^GRD (the starting structure used in this work), the transition state of H‐Ras bound to p120GAP, the ground OFF‐state of Cdc42 bound to RhoGAP, and the transition state of RhoA bound to RhoGAP (Figure 5a). In addition to the conserved arginine finger that stabilizes the active site of the transition state, one or two conserved salt bridges between Glu/Asp of Ras and Arg/Lys of GAP are also exhibited in both ground and transition states. In the case of H‐Ras bound with p120GAP, the salt bridge is formed between Glu63 and Arg903 (Figure 5c). Sequence alignment of the core region of NF1^GRD with that of p120GAP shows that Arg903 is equivalent to Arg1391 in NF1^GRD (Figure S8). In the transition state of RhoA bound to RhoGAP, interactions of Asp65 with both Lys122 and Arg126 are found to stabilize the conformation of Switch II (Figure 5d), and such interactions are also formed in the ground state of another member of the Rho family Cdc42 in which Asp63 interacts with both Lys122 and Arg126 simultaneously (Figure 5e). Consequently, the Gln63 of RhoA (or Gln61 of Cdc42) in Switch II becomes more ordered than its equivalent Gln61 in Ras (Rittinger et al. ^1997b). Our structural analysis suggests that the Glu63–Arg1391 interaction plays a similar role in the stabilization of the conformation of Switch II of K‐Ras4B bound to NF1^GRD, which facilitates Gln61 to position the hydrolytic water molecule for GAP‐activated GTP hydrolysis (Li and Zhang 2004). This conclusion is further supported by the experimental observation that NF1 R1391M mutant blocked the formation of the transition state analog (Mittal et al. 1996). Further, NF1 R1391A mutant decreased the GTP‐hydrolysis reaction 45‐fold, but R1391K substitute only decreased the reaction 2.4‐fold (Ahmadian et al. 1997). We therefore propose that the Glu63–Arg1391 interaction is a critical structural determinant for effective GTP hydrolysis stimulated by NF1^GRD. To provide structural evidence, we calculated the root‐mean‐square‐fluctuation (RMSF) of K‐Ras4B for each system (Figure S9). Compared to the wild‐type NF1^GRD, the E1356A mutant displayed larger fluctuations of residues neighboring Gln61, whereas the E1356V mutant displayed smaller fluctuations in the Switch II region. Consequently, the catalytic efficiency of the E1356A mutant could be lower than that of the wild‐type NF1^GRD. In contrast, the catalytic efficiency of the E1356V mutant could be higher than that of the wild‐type NF1^GRD. Therefore, compared to the wild‐type NF1, the E1356A mutant prefers to populate catalysis‐deficient conformations with diminished Glu63–Arg1391 interaction and thus more flexible Switch II region of K‐Ras4B, whereas the E1356V mutant efficiently populates the catalysis‐competent conformations with enhanced Glu63–Arg1391 interaction and consequently more stable Switch II region of K‐Ras4B. In addition, we also calculated the population of the Glu63–Arg1391 salt bridge in the cases of R1276P and R1276Q (Figure S10). Without the arginine finger, R1276Q shows a 3‐fold higher probability of establishing such a salt bridge as compared to R1276P, providing a rationale for the difference in the reduction of the NF1^GRD GAP activity in R1276Q (~500‐fold) as compared to R1276P (~8000‐fold) (Ahmadian et al. 1997; Klose et al. ¹⁹⁹⁸).

FIGURE 5.

The presence of salt bridge(s) in the binding interface between GTPases and their GAPs. (a) Superposition of the conformation of K‐Ras4B–NF1^GRD complex in the ground OFF‐state (white, PDB ID: 6OB2), H‐Ras–p120GAP in the transition state (yellow, PDB ID: 1WQ1), RhoA–RhoGAP in the transition state (green, PDB ID: 1TX4), and Cdc42–RhoGAP in the ground OFF‐state (magenta, PDB ID: 1AM4). (b) The salt bridge Glu63–Arg1391 in the K‐Ras4B–NF1^GRD complex. (c) The salt bridge Glu63–Arg903 in the H‐Ras–p120GAP complex. (d) The salt bridges Asp65–Lys122 and Asp65–Arg126 in the RhoA–RhoGAP complex. (e) The salt bridges Asp63–Lys122 and Asp63–Arg126 in the Cdc42–RhoGAP complex. The presence of salt bridge(s) near the catalytic Gln stabilizes the switch II of GTPases in both ground and transition states.

2.4. E1356 mutants allosterically modulate the Glu63–Arg1391 interaction

To further characterize how E1356 mutations allosterically modulate the Glu63–Arg1391 interaction, we performed dynamic network analysis implemented in VMD (Sethi et al. 2009), and constructed the optimal and suboptimal allosteric signal paths between Glu1356/Val1356^NF1 and Glu63^Ras, or between Ala1356^NF1 and Arg1391^NF1 (Figure 6a). For the complex comprising wild‐type NF1^GRD, among the top five (one optimal and four suboptimal) allosteric signal paths propagating from Glu1356^NF1 to Glu63^Ras, we identified one suboptimal path (path 4) that involves Arg1391 and Glu63, indicating their correlated motion and consequently the formation of Glu63–Arg1391 salt bridge (Figures 6b and S11). Notably, four paths (paths 1, 2, 3, and 5) involve Glu62^Ras and Lys1283^NF1, showing that this stable salt bridge not only contributes to the active conformation of Switch II of K‐Ras4B but also transduces an allosteric signal from Glu1356^NF1 to Glu63^Ras. In the case of E1356A mutant, we could not identify any allosteric signal transmission between Ala1356^NF1 and Glu63^Ras that involve Arg1391^NF1 in all signal paths, so we constructed the signal paths between Ala1356^NF1 and Arg1391^NF1. Again, there is no signal transfer path between Ala1356 and Arg1391 that involves Glu63 (Figures 6c and S11), indicating that E1356A mutation blocks the signal transmission between Glu63^Ras and Arg1391^NF1. Consequently, the formation of Glu63–Arg1391 salt bridge is greatly inhibited. By contrast, in the case of E1356V mutant, there are one optimal and three suboptimal signal paths that involve Arg1391^NF1 and Glu63^Ras (Figures 6d and S11), implying that E1356V mutation significantly enhances their correlated motion and therefore promotes the formation of the salt bridge between Arg1391^NF1 and Glu63^Ras. These distinct signal transduction paths suggest that E1356 mutations could allosterically modulate the Glu63–Arg1391 interaction, which in turn leads to distinct effects on the active conformation of Switch II in K‐Ras4B.

FIGURE 6.

Dynamic network analysis of the signal transfer path between E/A/V 1356 of NF1^GRD and E63 of K‐Ras4B or R1391 of NF1^GRD. (a) To identify the signal propagation pathway between NF1^GRD and K‐Ras4B, we define the source residue is Glu1356/Ala1356/Val1356, and the sink residue is either Glu63 or Arg1391. (b) The signal propagation paths from Glu1356 of typeNF1^GRD (wild type) to Glu63 of K‐Ras4B. (c) The signal propagation paths from Ala1356 of NF1^GRD (E1356A) to Arg1391 NF1^GRD. (d) The signal propagation paths from Val1356 of NF1^GRD (E1356V) to Glu63 of K‐Ras4B. In all cases, the signal transfer paths include the optimal path (red, path 1) and four suboptimal paths (paths 2–5). For details, see Figure S11. The dynamic network analysis reveals that E1356 mutants allosterically modulate Glu63–Arg1391 interaction.

2.5. Insertion of exon 23a disrupts the active conformation of K‐Ras4B

We have demonstrated that mutations on E1356 of NF1^GRD can allosterically modulate its GAP activity. Another structural alteration on NF1^GRD is the insertion of 21 amino acids by exon 23a between Gln1370 and Val1371 (isoform 2, denoted as NF1^GRD23a), which inhibits NF1 GAP activity (Hinman et al. 2014; Nguyen et al. ²⁰¹⁷). To elucidate the underlying molecular mechanism, we constructed two structural models of NF1^GRD23a in complex with K‐Ras4B‐GTP, where the insertion is initially either a random coil or a helix‐turn‐helix (Figure 7a). Note that the CHARMM36m force field used here is applicable for folded as well as intrinsically disordered structures (Huang et al. 2017). These two starting second structures represent two typical conformations as predicted by AlphaFold2 (Figure S12) (Jumper et al. 2021). We superimposed the two NF1^GRD23a model structures with the GRD of the cryo‐electron microscopy structure of NF1 isoform 2 (PDB ID: 7PGT) and no steric clash with other domains were observed (Figure S13). Over a total of 3 μs MD simulations (1 μs each replica), the insertion can adopt primarily random coil structure or maintain partial helical structure in the Ala1371–Ala1380 fragment (Figure S14). The representative conformational dynamics of the two NF1^GRD23a are shown in Figure S15.

FIGURE 7.

Modeling and characterization of the effect of exon 23a on the GAP activity of NF1^GRD. (a) Sequence and initial conformations of the NF1^GRD23a with the inserted 21 residues by alternative splicing of exon 23a in NF1 gene. The starting structure of the exon 23a is either random coil (in red circle, left panel) or helix‐turn‐helix (in red circle, right panel). The starting structure of K‐Ras4B is the same as shown in Figure 1. (b) The distances corresponding to the Mg²⁺ and GTP coordination, and between Arg1276 and GTP for NF1^GRD23a in complex with K‐Ras4B. The initial structure of exon 23a is random coil. (c) The distances corresponding to the Mg²⁺ and GTP coordination, and between Arg1276 and GTP for NF1^GRD23a in complex with K‐Ras4B. The initial structure of exon 23a is helix‐turn‐helix. (d) Representative conformation of the active site of K‐Ras4B, in which Thr35 no longer binds to Mg²⁺ and Gly60 no more binds to GTP even when Arg1276 interacts with the γ‐phosphate of GTP. (e) Superposition of the inactive conformations of K‐Ras4B bound to NF1^GRD23a. The inactive conformations of K‐Ras4B with random coil and helix‐turn‐helix structures of exon 23a are colored green and magenta, respectively. For comparison, the active conformation of K‐Ras4B shown in Figure 3a (yellow), and the GDP‐bound K‐Ras4B in the inactive state 1 conformation (white, PDB ID: 6BP1) are also shown. The orientations of Tyr32 and Thr35 of K‐Ras4B are highlighted in the right panel. The insertion of exon 23a allosterically disrupts the active conformation of K‐Ras4B.

As we examined the coordination environment of the active site of K‐Ras4B bound to the NF1^GRD without the insertion (Figure 2), we calculated the corresponding distances and found that the active conformation of K‐Ras4B was disrupted, independent of the insertion conformations (Figure 7b,c). The coordination of Mg²⁺ and GTP are not maintained in all simulations, suggesting that Switch I and Switch II of K‐Ras4B become more flexible. Figure 7d shows one such collapsed conformation in which Thr35 moved away from the active site, regardless of the interaction between Arg1276 of NF1 and the γ‐phosphate of GTP. We also performed clustering analysis on the conformational ensemble of NF1^GRD23a in complex with K‐Ras4B and selected the representative conformations from the top one cluster for each system. We then superimposed these representative conformations with that of NF1^GRD. For comparison, the inactive conformation of K‐Ras4B (state 1, PDB ID: 6BP1) taken from the crystal structure was also included in the superimposition (Figure 7e). Compared to the active conformation of K‐Ras4B, the conformations of both switches are more dynamic in the presence of the insertion, but still less dynamic than the experimentally determined inactive conformation of K‐Ras4B, in which the Switch I is more open, and Tyr32 and Thr35 move further away from the active site (Figure 7e). Nevertheless, the loss of stable coordination of Mg²⁺ and GTP due to the insertion on the NF1^GRD renders K‐Ras4B unable to exert its catalytic activity properly.

The conformational dynamics shows that there exist interactions between the insertion and K‐Ras4B when the insertion adopts a helical structure (Figure S15). To characterize such interactions, we calculated the contact maps between NF1^GRD23a and K‐Ras4B, as well as between the insertion and other part of NF1^GRD23a (Figure S16). Weak to moderate interactions between the insertion and other part of NF1^GRD23a are found when the insertion adopts a disordered conformation (Figure S16a), primarily between the lysine residues in the insertion (Lys1385, Lys1386, Lys1379, and Lys1389) and Glu/Asp (Asp1490, Asp1485, and Glu1220) in other part of NF1^GRD23a. No direct interactions between the disordered insertion and K‐Ras4B were identified, thus the insertion seems to disrupt the active conformation of K‐Ras4B by an allosteric effect. However, when the insertion adopts a helix structure, in addition to the electrostatic interactions established between the lysine residues and other residues in NF1^GRD23a, it can also form interactions with K‐Ras4B (Figure S16b–d). The interactions between the insertion (e.g., Lys1385 and Lys1386) and Switch I (e.g., Asp30 and Glu31) of K‐Ras4B seem to be responsible for the disruption of the coordination of Thr35 to Mg²⁺ as shown in Figure 7c. Also, the loss of binding between Gly60 and GTP indicates that the insertion allosterically disturbs the Switch II conformation of K‐Ras4B, regardless of the secondary structure the insertion adopts (Figure 7b,c). Collectively, our results demonstrate that the insertion on NF1^GRD23a shifts K‐Ras4B population toward inactive conformations by allosteric regulation. And such regulation seems largely insensitive to mutations on NF1^GRD23a as K‐Ras4B–NF1^GRD23a complex harboring E1356V mutation persistently populates inactive conformation of K‐Ras4B (Figure S17).

3. DISCUSSION

NF1 is a multifunctional protein that regulates multiple signaling pathways including Ras/Raf/MEK/ERK and PI3K/AKT/mTOR (Baez‐Flores et al. ²⁰²³). As a RasGAP, NF1 catalyzes Ras‐mediated GTP hydrolysis, thus Ras/MAPK signaling (Bahar et al. 2023). Mutations on NF1 are associated with cutaneous neurofibromas, as well as developmental disorders (Ratner and Miller 2015). Most mutations are clearly loss‐of‐function alleles, especially missense mutations on NF1^GRD (e.g., Arg1276 and Lys1423) that diminish GAP activity, supporting NF1 as a tumor suppressor (Vigil et al. 2010). Individuals carrying mutations at codons 844 to 848 exhibit severe phenotypes, including plexiform and symptomatic spinal neurofibromas, symptomatic optic pathway gliomas, other malignant neoplasms, and bone abnormalities (Koczkowska et al. 2018). These mutants, which are outside NF1^GRD, act as dominant‐negative alleles by destabilizing the NF1 dimer (Young et al. 2023). Besides loss‐of‐function via dominant‐negative mechanisms, many pathogenic mutations also act via gain‐of‐function mechanism, with differential effects on protein structure (Gerasimavicius et al. 2022; Li et al. ²⁰¹⁹), but have not been reported for NF1 mutations. Understanding the genotype–phenotype correlation has important implications for disease diagnosis and precision treatment (Leier et al. 2020; Nussinov et al. ²⁰¹⁹; Zhou et al. ²⁰²¹).

Here, we focused on the E1356A and E1356V mutants on NF1^GRD, with the former related to neurofibromatosis and the latter associated with the neurodevelopmental disorder. Their pathogenicity predicted by AlphaMissense is considered benign, with a value of 0.118 for E1356A and 0.191 for E1356V (Cheng et al. 2023). But how these pathogenicity values relate to clinical outcomes, which can be severe, is unclear. In this work, we show that the two mutants have different effects on the conformational propensities of K‐Ras4B–NF1^GRD complexes, which determines their distinct phenotypes. We show that both mutations have negligible impact on the active conformation of K‐Ras4B (Figure 2), which is also observed for mutations at the arginine finger Arg1276 (R1276P/Q) (Figure S1). For NF1^GRD E1356A, there exists a higher energy barrier for the movement of Arg1276 into the active site of K‐Ras4B and interaction with the γ‐phosphate of GTP, suggesting that its conformational transition from the ground state to the transition state is kinetically less favorable than the wild‐type and E1356V mutant. Compared to the wild‐type NF1^GRD, both mutants display similar properties in the active site that are preorganized for the GTP hydrolysis. However, the E1356V mutant exhibits a higher binding affinity to K‐Ras4B (Figure 3). Since the binding affinity between NF1^GRD and K‐Ras4B is not directly correlated with the NF1 GAP activity (Figure S4), we analyzed the structural properties of different GTPases and their GAPs in the ground and transition states (Figure 5). We identified the Glu63–Arg1391 interaction as a critical determinant that stabilizes the transition states of NF1^GRD‐catalyzed GTP hydrolysis. The presence of such an ionic interaction near the catalytic Gln at the Switch II region of GTPases seems a general structural arrangement provided by GTPases and their corresponding GAPs. In the case of GTPases Ras and Rho, the arginine finger is provided by the RasGAP and RhoGAP, respectively, but for the heterotrimeric G‐protein G_iα1 in combination with RGS4 (a regulator of G‐protein signaling), the arginine finger is provided by G‐protein itself (Scheffzek et al. 1995). Although structurally different, such an ionic interaction at the binding interface that stabilizes the Switch II region (Gln in particular) of G_iα1 is conserved (Figure S18). For GAP‐catalyzed GTP hydrolysis without an arginine finger, such as the nuclear Ras‐related protein (Ran), whose GTP hydrolysis requires both RanGAP and Ran‐binding protein RanBP1 or RanBP2 (Seewald et al. 2002). In the crystal structure of the ternary GMPPNP‐bound Ran–RanBP1–RanGAP^yeast complex, there also exists such an ionic interaction between Lys71^Ran and Asp103^GAP at the binding interface between Ran and RanGAP (Figure S19). Sequence alignment and structural superposition showed that Lys71 of Ran is located at the Glu63 position of K‐Ras4B, and Gln69 of Ran is equivalent to Gln61 of K‐Ras4B (Scheffzek et al. 1995). Such an ionic interaction is conserved when Ran binds to human RanGAP in the presence of Ran binding protein, with Asp103 in yeast RanGAP equivalent to Asp121 of human RanGAP (Figure S20). The exact conformation of the intermediates in the transition state for GTP hydrolysis of the GTPase catalyzed by RasGAP remains unknown, and X‐ray crystal structures were resolved only for the transition state analogue in the presence of aluminum fluoride (Mittal et al. 1996; Scheffzek et al. ¹⁹⁹⁷). Capturing more detailed conformational changes that occur as the Ras–RasGAP complex moves from the ground state to the transition state is beyond the capabilities of the classical MD simulations used in this study (van Gunsteren and Oostenbrink ²⁰²⁴). Time‐resolved Fourier transform infrared (tr‐FTIR) spectroscopy provides a high temporal and spatial resolution to monitor the kinetics of GAP‐mediated GTP hydrolysis by GTPases (Kotting et al. 2008; Rudack et al. ²⁰¹²). The study of the GTPase reaction of the Ras−RasGAP by tr‐FTIR suggests that the movement of the arginine finger into the active site of Ras lowers the activation energy for the cleavage reaction (Kotting et al. 2008).

The catalytic efficiency of an enzyme reaction depends on the stabilization of the transition state in a very specific arrangement (Schramm 1998). The ground state conformations of K‐Ras4B–NF1^GRD complex in our simulations resemble the near attack conformation (NAC), which is defined as the required conformation for juxtaposed reactants to enter a transition state (Lightstone and Bruice 1996). A greater fraction of reactant conformations that are present as NACs corresponds to a greater rate constant (Bruice and Lightstone 1999). A NAC is still a ground state conformation where bond breaking and forming have not been initiated, but reactants have been properly aligned at 1–3 Å separation for the reaction (Bruice and Lightstone 1999). As we show, both wild‐type and E1356‐mutated complexes are structurally prepared for the GTP hydrolysis (Figure 3), except for Gln61 that has not orientated the nucleophilic water for reaction (Figure S5). Notably, compared to the wild‐type, the NF1^GRD E1356V mutant populates more NACs with enhanced Glu63–Arg1391 interaction that stabilize Glu61 position at the Switch II region of K‐Ras4B (Figure 4). Therefore, the NF1^GRD E1356V mutant appears more catalytically efficient than the NF1^GRD E1356A mutant. NF1 forms a high‐affinity dimer in human cells (Sherekar et al. 2020). Based on the cryo‐electron microscopy structure of NF1 dimer (PDB ID: 7R04) (Chaker‐Margot et al. ²⁰²²), E1356 has a few contacts with several residues at the C‐HEAT domain in the closed conformation of GRD. However, the contacting interface is mostly exposed, and thus these contacts may not impair the dimer stability. Meanwhile, E1356 has no contact with other domains in the open conformation of GRD (Figure S21). Therefore, the E1356 mutation could not exert a dominant‐negative effect. Due to the distinct conformational propensities, the NF1^GRD E1356V mutant is more likely to act through a gain‐of‐function mechanism, whereas the NF1^GRD E1356A mutant acts via a loss‐of‐function mechanism. Like E1356A, we also found that NF1^GRD E1356G mutant acts through a loss‐of‐function mechanism (Figure S22).

Like gain‐of‐function mutations in SHP2 that promote Noonan syndrome (NS) by enhanced activation of the Ras/MAPK pathway (Araki et al. 2009; Neel et al. ²⁰⁰³), some NF1 patients demonstrate neurofibromatosis‐Noonan syndrome (NFNS) with phenotypic features of both neurofibromatosis type I and NS (De Luca et al. 2005). Notably, there is a high prevalence of defects affecting NF1^GRD in NFNS, including E1415X (truncating), K1423E, N1430T, V1432L, and E1438del (in‐frame deletion) (De Luca et al. 2005). Among these defects, K1423E in NF1^GRD disrupts the Asp38–Lys1423 salt bridge that helps stabilize Switch I of K‐Ras4B (Figure 4), and thus acts as a loss‐of‐function mutation. This is consistent with early experimental data showing that K1423E mutant occurred in NF1 and severely impaired the GAP activity of NF1^GRD (Li et al. 1992). In addition to mutations at Arg1276 and Glu1356, we further elucidate the molecular mechanism of how the alternative splicing of NF1 exon 23a diminishes the GAP activity of NF1^GRD primarily allosterically (Figure 7). Exon 23a inserts 21 residues in NF1^GRD, including six lysine and two glutamic acid residues, significantly altering NF1^GRD total charge and its conformational dynamics, which in turn allosterically impairs the active conformation of the bound K‐Ras4B. This result is in line with the experimental results showing that the isolated NF1^GRD has a low conformational stability and is sensitive to pH changes of the solution (Neira et al. 2021). Our study thus provides a structural basis for the inhibitory effect of exon 23a on the GAP activity of NF1^GRD, and consequently a significant increase in Ras/MAPK signaling that impairs learning behaviors in mice (Hinman et al. 2014; Nguyen et al. ²⁰¹⁷). On the other hand, the increased RasGAP activity of NF1^GRD due to E1356V may attenuate Ras/MAPK signaling, which also contributes to the neurodevelopmental disorders (Iroegbu et al. 2021). Normal “central nervous system” development requires a balanced cellular Ras/MAPK signaling (Iroegbu et al. 2021).

Deregulation of NF1 and other RasGAPs has been associated with a variety of cancers (Maertens and Cichowski 2014; Mishra and Lambright ²⁰¹⁶). NF1 mutations are also highly related to developmental learning disabilities (Iroegbu et al. 2021; Shilyansky et al. ²⁰¹⁰). They were classically considered as regulatory proteins but not druggable targets (Maertens and Cichowski 2014; Sran and Bedrosian ²⁰²³; Vigil et al. ²⁰¹⁰). With pharmacological advances that however may change. Most Ras oncogenic mutations at Gly12, Gly13, or Gln61 may prevent the formation of the Ras–GAP transition state, thereby blocking RasGAP‐catalyzed GTP hydrolysis (Scheffzek et al. 1997). The K‐Ras4B G13D mutation, however, was shown to respond to NF1^GRD‐mediated GTP hydrolysis (Rabara et al. 2019). Another K‐Ras4B mutation G12C, which is insensitive to RasGAPs, can hydrolyze GTP in the presence of RGS3. In addition, RGS3 also enhanced GTP hydrolysis by K‐Ras4B G12D/V and G13C/D mutants, but less effectively than NF1 or p120GAP (Li et al. 2021). Thus, restoring the GAP activity emerges as a new, albeit indirect, strategy to target specific Ras mutations. A recent QM/MM study implied that R903E is the most promising mutant of p120GAP for decreasing the activation barrier of the G12D Ras–GAP complex (Berta et al. 2023). We note that with Arg903 of p120GAP being equivalent to Arg1391 of NF1^GRD (Figure S8), the above QM/MM result is consistent with our present results showing that formation of the Glu63–Arg1391 salt bridge promotes the catalysis‐ready conformation of NF1^GRD mutant E1356V. As R1391E of NF1^GRD would disrupt the salt bridge with Glu63 of K‐Ras4B, our results suggest that E1356V is a promising allosteric mutant for targeting K‐Ras4B G12D mutant through enhancing NF1 GAP activity.

4. CONCLUSIONS

NF1 is an essential negative regulator of Ras‐driven oncogenesis owing to its GAP function. In addition to the loss‐of‐function and dominant‐negative mutations identified in NF1 gene, we decipher the molecular mechanism of the gain‐of‐function mutant E1356V on NF1^GRD, which to date is the only mutation identified on NF1^GRD associated with neurodevelopmental disorders. Our results suggest that NF1 GAP is allosterically regulated, offering novel structural insights into the activation of the NF1^GRD to target K‐Ras4B oncogenic driver mutations. We demonstrate that the differential conformational propensities of E1356A and E1356V NF1^GRD determine their distinct clinical phenotypes (neurofibromatosis or neurodevelopmental disorder), within the framework of our hypothesis, that is, that stronger, cancer‐associated mutants harbor a higher propensity of active conformations than the weaker, neurodevelopmental‐related variants (Nussinov et al. 2022; Nussinov et al. ^2023a; Nussinov et al. ^2023b). A high population of the active state is a requirement for cell proliferation, which is a hallmark of cancer. Weaker, neurodevelopmental disorders‐related variants are associated with cell growth in differentiation. We further provide the structural basis for the allosteric modulation of the NF1^GRD GAP by exon 23a, whose loss‐of‐function alteration in NF1^GRD impairs learning behaviors in mice (Hinman et al. 2014; Nguyen et al. ²⁰¹⁷).

Taken together, we provide a genotype‐ensembles‐phenotype correlation for NF1 mutations, in line with our tenet of sequence‐conformational ensemble‐function replacing the sequence‐structure‐function paradigm (Nussinov et al. ^2023b). Our blueprint would need to be expanded for precision therapy. This approach could be envisioned as akin to experimental energetic and allosteric landscapes of mutations, as executed for several proteins, including K‐Ras (Weng et al. 2024), H‐Ras (Hidalgo et al. 2022), and Src (Hobbs et al. 2022). Propensities of conformational ensembles are powerful predictors of protein function. Coupled with computational deep high‐throughput mining and comprehensive mapping of allosteric protein landscapes (Verkhivker et al. 2023), and performed comprehensively across mutational landscapes, they can help decipher the intricate mechanisms driving pathogenesis and target them. They may also help perceiving the differential functional outcomes of multiple ligands binding to a receptor, as in the case of RTKs (Karl et al. 2024; Krzyscik et al. ²⁰²⁴; Wirth et al. ²⁰²³; Wirth et al. ²⁰²⁴), even if the change is too small to be observable experimentally. They can also couple with strategies such as the recently offered, powerful prediction of high‐throughput mutational scans (Sora et al. 2023). We hope that the concept and approach presented here contributes toward this aim.

5. MATERIALS AND METHODS

5.1. Modeling of NF1^GRD in complex with GTP‐bound K‐Ras4B

The crystal structure of wild‐type K‐Ras4B (GMPPNP‐bound) in complex with the GRD of NF1 (PDB ID: 6OB2) (Rabara et al. 2019) was used to construct the initial structural model. The nonhydrolyzable GTP analog GMPPNP was converted into the GTP by replacing the secondary amine group bridging the β‐phosphate to the γ‐phosphate with the oxygen atom. The initial GRD model (residues 1219–1462) contains the minimal central catalytic domain of NF1 (GAPc, residues 1248–1477), which exhibits full GAP activity (Ahmadian et al. 1996). In addition to GAPc, the GRD contains an extra domain (GAPex) at both ends of GAPc formed by the N‐ and C‐terminus that can bind to Spred1 (Sprouty‐related, EVH1 domain‐containing protein 1) without affecting the catalytic activity of NF1 (Dunzendorfer‐Matt et al. ²⁰¹⁶). Recent simulation study suggested that the binding of Spred1 allosterically enhanced the binding between K‐Ras4B and NF1^GRD, but impeded the conformational transformation of the catalytic center (Li et al. 2022). We adopted the GAPex conformation from the crystal structure of NF1^GRD (residues 1205–1528) in complex with Spred1 and GMPPNP‐bound K‐Ras4B (PDB ID: 6V65) (Yan et al. 2020) by superimposing 6OB2 with respect to the GRD. Other missing residues were modeled using the Rosetta loop modeling and refinement methods (Leman et al. 2020). Our initial model complex for MD simulations contains the wild‐type NF1^GRD (residues 1205–1530) and the GTP‐bound K‐Ras4B (residues 1–169) (Figure 1a). The NF1^GRD mutants (R1276P/Q and E1356A/V) were generated using the CHARMM program (version 46b1) (Brooks et al. 1983).

5.2. MD simulation protocol

All MD simulations were performed with NAMD 2.14 (Phillips et al. 2005; Phillips et al. ²⁰²⁰). The simulation protocols were similar to those used in our previous works (Jang et al. 2021; Liu et al. ^2022a; Liu et al. ^2022b; Liu et al. ^2023a; Liu et al. ²⁰²⁴; Maloney et al. ²⁰²¹; Zhang et al. ²⁰¹⁹; Zhang et al. ²⁰²¹; Zhang et al. ^2024a; Zhang et al. ^2024b). Each complex structure was represented using the updated and modified version of the CHARMM36m force field parameters (Huang et al. 2017; Klauda et al. ²⁰¹⁰) and solved in a cubic water box filled with TIP3P water molecules. The minimum distance between the protein and the edge of the water box was 12 Å. Each system was neutralized by adding 100 mM NaCl. The Nosé‐Hoover Langevin piston pressure control and Langevin temperature control were used to maintain the pressure at 1 atm and the temperature at 310 K (Braga and Travis 2005; Martyna et al. ¹⁹⁹⁴). Long‐range electrostatics were treated using the Particle Mesh Ewald (PME) method (Darden et al. 1993), with the PME grid spacing of 1.0 Å and PME interpolation order of 6. The van der Waals (vdW) interactions were calculated using a switching function with a twin cutoff of 10 Å and 12 Å, and the SHAKE algorithm was applied to constrain the motion of bonds involving hydrogen atoms (Ciccotti and Ryckaert 1986). An integration time step of 2 fs was used for all simulations. Each system was first energy minimized for 20,000 steps, and then equilibrated in the NVT ensemble (constant volume and temperature at 310 K) for 5 ns with Cα atoms of protein, GTP, and Mg²⁺ restrained using harmonic restraints. The production run was performed in the NPT ensemble (constant pressure at 1 atm and temperature at 310 K) without any restraints and lasted for 1 μs. Three independent simulations were performed for each system, and statistical data were obtained by averaging the last half of each independent simulation for each system.

5.3. Binding free energy calculation

The binding free energy between NF1^GRD and K‐Ras4B‐GTP was evaluated in terms of the molecular mechanics energies combined with generalized Born and surface area solvation (MM/GBSA) method (Genheden and Ryde 2015; Wang et al. ²⁰¹⁹). The averaged binding free energy $⟨∆G_b⟩$ is calculated as a sum of the gas‐phase molecular mechanics energy $⟨∆E_{\mathrm{MM}}⟩$, the solvation energy contribution $⟨∆G_{\mathrm{sol}}⟩$, and the entropy contribution –TΔS,

$$⟨∆G_b⟩=⟨∆E_{\mathrm{MM}}⟩+⟨∆G_{\mathrm{sol}}⟩-T∆S,$$

where angle bracket denotes an average along the MD trajectory. The gas‐phase molecular mechanics energy is a sum of the internal energy, the electrostatic energy, and the vdW interaction energy,

$$∆E_{\mathrm{MM}}=∆E_{\text{inter}}+∆E_{\text{elec}}+∆E_{\mathrm{vdW}}.$$

The solvation energy contribution can be divided into the electrostatic contribution and the nonpolar contribution,

$$\mathrm{\Delta }{G}_{\mathrm{sol}}=\mathrm{\Delta }{G}_{\mathrm{sol}}^{\text{elec}}+\mathrm{\Delta }{G}_{\mathrm{sol}}^{\text{nonpolar}},$$

where the electrostatic contribution was calculated using the generalized Born model (Onufriev et al. 2004). The nonpolar contribution is obtained using an equation

$$\mathrm{\Delta }{G}_{\mathrm{sol}}^{\text{nonpolar}}=\gamma \bullet \text{SASA}+\beta ,$$

where SASA denotes the solvent‐accessible surface area (Wang and Kollman 2000), with a set of the surface tension parameters, γ = 0.00542 and the constant β = 0.92.

The entropy contribution, TΔS, to the binding free energy was calculated by normal model analysis (Srinivasan et al. 1998). The gas‐phase, solvation energy, and entropy contributions were calculated based on a set of conformational snapshots taken from the MD simulations. A total of 2000 snapshots taken from the last half of each simulation were used to calculate the gas‐phase and solvation energy contributions, whereas a total of 50 snapshots were used to calculate the entropy contribution due to its heavy computational cost. Finally, the change in binding free energy due to the complex formation is calculated using the equation,

$$\mathrm{\Delta }{G}_b=\mathrm{\Delta }{G}_b^{\text{complex}}-\left(∆G_b^{\mathrm{NF}1}+∆G_b^{\mathrm{Ras}}\right).$$

The molecular interconversion program ParmEd was used to convert CHARMM topology files into Amber input format (Shirts et al. 2017). The binding free energy was calculated using the Perl script mm_pbsa.pl with default parameters in Amber22 program (Case et al. 2023).

AUTHOR CONTRIBUTIONS

Liang Xu: Methodology; formal analysis; conceptualization; data curation; writing – original draft; software. Hyunbum Jang: Conceptualization; methodology; writing – review and editing; software. Ruth Nussinov: Conceptualization; funding acquisition; project administration; supervision; writing – review and editing.

CONFLICT OF INTEREST STATEMENT

The authors declare no conflicts of interest.

Supporting information

Data S1. Supporting Information.

Xu L, Jang H, Nussinov R. Allosteric modulation of NF1 GAP: Differential distributions of catalytically competent populations in loss‐of‐function and gain‐of‐function mutants. Protein Science. 2025;34(2):e70042. 10.1002/pro.70042

DATA AVAILABILITY STATEMENT

The data that support the findings of this study are available from the corresponding author upon reasonable request.