Introduction

IDH1 is a highly conserved, homodimeric enzyme that reversibly converts isocitrate (ICT) to α-ketoglutarate (αKG) through NADP+-dependent oxidative decarboxylation. Tumor-driving IDH1 mutants catalyze a NADPH-dependent conversion of αKG to the oncometabolite D-2-hydroxyglutarate (D2HG), while typically ablating the conventional reaction1,2,3. D2HG competitively inhibits αKG-dependent enzymes like TET2 and JmjC lysine demethylases, causing DNA and histone hypermethylation and cellular de-differentiation4,5. Mutations at R132 drive >85% of lower grade and secondary gliomas6 and ~40% of cartilaginous tumors7, with R132H typically the most common8,9. Mutated IDH1 has been successfully therapeutically targeted, with several FDA-approved selective inhibitors in use and more in clinical trials (reviewed in refs. 10,11,12).

While early kinetic characterization of IDH focused on bacterial forms, recent efforts have illuminated details of human IDH1. As wild type (WT) IDH1 binds its substrates, a conformational change occurs where the large domain (residues 1–103, 286–414) and small domain (residues 104–136, 186–285) move towards each other owing to a hinge (residues 134–141) within the clasp domain (residues 137–185)13. This movement closes the active site cleft with the concomitant opening of a back cleft13. In the absence of bound substrates, the α10 helix (residues 271–285) helps stabilize IDH1 in its open, inactive conformation13. This critical regulatory element undergoes a conformational change to help properly orient the active site residues upon substrate binding-driven closure13. These structural features are generally preserved in IDH1 R132H1,3,14, but inherent catalytic deficiencies coupled with improved NADPH binding result in this mutant catalyzing D2HG production, albeit inefficiently though at great benefit to the tumor environment.

To better understand how D2HG production occurs, there is tremendous value in studying a variant of IDH1 with more robust neomorphic reaction activity. IDH1 R132C/S/L/G/Q mutations have been reported in patients at lower frequencies15,16,17,18,19 and support distinct tumor D2HG levels20. We have demonstrated that these mutants display distinct kinetic profiles for both neomorphic and conventional reactions21,22, suggesting that their kinetic features may drive some of the variability of patients’ D2HG levels22. We identified one mutant, R132Q, that maintained weak conventional catalytic activity, drove robust D2HG production21, and was resistant to mutant IDH1 inhibitors via a mechanism not yet understood22. Additionally, IDH1 R132Q drove enchondroma tumor formation in mouse models23. By identifying distinct features of R132Q and R132H, we can uncover selectivity handles for improved mutant IDH1 inhibitors, as an H-to-Q mutation requires only a single base change. Investigating the atomic-level mechanisms that drive diverse kinetic activity and inhibition among tumor-relevant IDH1 mutants can also inform chemical features that guide the field of enzyme design24.

Here, we report the static and dynamic structural features that drive the distinct kinetic properties among tumor-relevant IDH1 mutants, capitalizing on the unusual active site attributes that allow R132Q to maintain conventional and enhance neomorphic activities. We observe by X-ray crystallography that the neomorphic substrate αKG, but not the conventional substrate ICT, binds via multiple conformations to R132Q. Solution-based kinetics and structural experiments demonstrate that ability of R132Q to explore multiple conformations and substrate binding modes depends upon a relatively immobile, solvent-inaccessible enzyme that is better optimized for substrate binding, hydride transfer, and mutant IDH1 inhibitor resistance compared to R132H.

Results

R132Q is optimized for substrate binding and catalysis

We previously demonstrated that IDH1 R132Q maintains weak catalytic efficiency for the conventional reaction (ICT to αKG), while also displaying higher catalytic efficiency for the neomorphic reaction (αKG to D2HG) relative to R132H21,22. Steady-state kinetics analysis (Supplementary Fig. 1) revealed a 5.9-fold increase in catalytic efficiency for the conventional reaction in R132Q versus R132H, driven primarily by an increase in kcat. R132Q catalyzed the neomorphic reaction 9-fold more efficiently than R132H via optimization of both kcat and Km. This suggested that R132Q exhibits a more stable transition state and provides more optimized on/off paths of the reactants and products compared to R132H.

Pre-steady-state kinetics experiments indicated that hydride transfer, or a step preceding it, was rate-limiting for the conventional reaction catalyzed by WT and R132Q, and for the neomorphic reaction catalyzed by R132Q and R132H (Fig. 1). NADPH consumption by R132H showed an initial lag that was eliminated when using higher concentrations of αKG (Supplementary Fig. 2). A lag has been reported previously with IDH1 WT, which was eliminated via pre-incubation of both ICT and metal25,26,27,28. Interestingly, we did not observe a lag in the neomorphic reaction catalyzed by R132Q, despite using a concentration of αKG that was 10-fold lower than the concentration associated with a lag in R132H. This suggested that αKG is more proficient at driving R132Q from an inactive to an active state compared to R132H, though it was not apparent through these experiments whether this was achieved by a more catalytically primed ground state or a faster conformational change.

Fig. 1: Pre-steady-state single-turnover kinetic features of IDH1 WT, R132H, and R132Q catalysis.
figure 1

NADPH formation in the conventional reaction and consumption in the neomorphic reaction was monitored over the course of a single turnover (top plot) and compared with a control experiment lacking enzyme (bottom plot, in green). Traces represent an average of four technical replicates. Residuals (middle plot) were obtained to assess the goodness of a single exponential equation fit in the top plots. Kinetic parameters were calculated and reported as +/−SEM resulting from deviation of the mathematical fit. A IDH1 WT, conventional reaction. B IDH1 R132H, neomorphic reaction. C IDH1 R132Q, conventional reaction. D IDH1 R132Q, neomorphic reaction.

Full size image

We were unable to capture rates of conformational change when monitoring intrinsic protein fluorescence. However, we measured rates of NADPH binding to IDH1 WT, R132H, and R132Q using enzyme that was stripped of cofactor14 (Supplementary Fig. 3). We found that all three IDH1 proteins displayed single-step binding events, with an NADPH binding on rate (kon) that was ~2-fold faster for WT than R132Q, while kon rates for R132H were profoundly slower. We also used isothermal titration calorimetry (ITC) to measure equilibrium binding affinity of NADPH for IDH1 (Supplementary Fig. 4). We found that both mutants exhibited a decrease in Kd compared to WT, suggesting that a slower koff rate drove the improved affinity for NADPH observed for R132H despite the slow kon rate. Taken together, these kinetic data further supported the finding that when compared to R132H, IDH1 R132Q has a lower barrier to adopting the closed, active conformation that is driven by substrate and metal binding.

R132Q has a less solvent-accessible active site

To illuminate possible mechanisms behind the time-resolved changes exhibited by IDH1 R132Q versus those in WT and R132H, we first used hydrogen/deuterium exchange-mass spectrometry (HDX-MS) analysis. We probed solvent accessibility as indicated by deuterium uptake in the binary IDH1:NADP(H) form, as WT and mutant IDH1 are known to copurify bound to NADP(H)26,27. We also measured deuterium uptake upon the addition of substrate (ternary complex, IDH1:NADP(H):ICT/αKG), or upon the addition of substrate and Ca2+ (quaternary complex, IDH1:NADP(H):ICT/αKG:Ca2+). By far the most substantial change in deuterium uptake for WT, R132H, and R132Q occurred in the quaternary form, indicative of closed, catalytically competent conformations among all enzyme species (Supplementary Fig. 5). This is consistent with previous findings that both substrate (ICT, but also presumably αKG in the neomorphic reaction) and divalent metal binding are required to drive IDH1 into its fully closed, active conformation25,26,27,28. Deuterium uptake generally showed the following trend: R132H:NADPH:αKG:Ca2+ ≫ WT:NADP+:ICT:Ca2+ > R132Q:NADPH:ICT:Ca2+ > R132Q:NADPH:αKG:Ca2+ (Fig. 2, Supplementary Figs. 5 and 6), with R132Q overall appearing to have a less structurally dynamic, more closed conformation compared to R132H.

Fig. 2: IDH1 R132Q has lower deuterium uptake than R132H in both binary and quaternary complexes.
figure 2

A Plots of deuterium uptake encompassing residues 86–120, 168–182, and 269–291 (left) are shown with the structural features of these residues shown in cartoon (right) for IDH1 R132Q, WT13, and R132H14. B Plots of deuterium uptake for residues 168–191, 217–227, 257–267, and 305–354 (left) are shown, with the structural features of IDH1 R132Q, WT13, and R132H30 encompassing these regions indicated in cartoon (right). Each point represents the mean of three technical replicates.

Full size image

Since our kinetic studies suggested IDH1 R132Q had a lower barrier to achieve the closed conformation compared to R132H, we hypothesized that binary R132Q:NADP(H) would be in a more quaternary-like state. To test this, we compared deuterium uptake among the binary states, predicting that the R132Q:NADP(H) complex would experience less deuterium uptake than R132H:NADP(H). Unsurprisingly, in general the IDH1:NADP(H) form of all three proteins had high deuterium uptake, particularly in the substrate binding pocket, clasp, and dimer interface (Fig. 2, Supplementary Figs. 5–7). As predicted, R132Q:NADP(H) and WT:NADP(H) had the least deuterium uptake overall, while R132H:NADP(H) exhibited, by far, the most uptake. As this suggested that NADP(H)-bound R132Q had a more closed/less mobile conformation compared to R132H, we wondered if the temporal features of our HDX-MS data suggested a faster closing upon substrate binding for R132Q. This would provide one mechanism of the improved catalytic efficiency shown by IDH1 R132Q relative to R132H in the conventional and neomorphic reactions. To address this, we inspected peptides that included residues within 4 Å of bound NADP(H) and ICT/αKG to compare deuterium exchange rates, as uptake plots represent combined exchanged rates for all amides in the peptide, to compare the composition of exchange rates in R132Q versus R132H. We expect that fewer amides would be exchanging at slower exchange rates for R132Q if this enzyme had a primed ground state that reached a closed conformation more easily29. Indeed, many active-site peptides had fewer amides with slower/intermediate exchange rates for IDH1 R132Q and WT compared to R132H (Supplementary Fig. 7). Specifically, peptides 210–216 (including catalytic residue K212), 240–253, and 257–267 all showed contributions of amides exchanging at faster rates for R132Q versus R132H. This favors a model where the ground state of R132Q is a more closed conformation that follows a simpler path to a catalytically competent state compared to R132H.

Seeking to pair the dynamic, intermediate-resolution HDX-MS data with static, high-resolution X-ray crystal structures, we report here six crystallographic models representing the structures of IDH1 R132Q: binary IDH1 R132Q bound to NADP(H) (R132Q:NADP(H), PDB 8VHC, PDB 8VH9; R132Q bound to conventional reaction substrates (R132Q:NADP(H):ICT:Ca2+, PDB 8VHD); R132Q bound to neomorphic reaction substrates (R132Q:NADP(H):αKG:Ca2+, PDB 8VHB), PDB 8VHA, and R132Q bound to a NADP-TCEP adduct (R132Q:NADP-TCEP:Ca2+, PDB 8VHE). These structures facilitated comparisons with previously solved IDH1 WT13 and R132H structures14,30, including among binary and ICT- and αKG-bound models.

Binary structures of IDH1 R132Q (Fig. 3A) were valuable to help us understand differences among the mutant active sites. While R132Q:NADP(H) showed no major global structural alterations upon alignment with previously solved structures of WT:NADP(H)13 and R132H:NADP(H)30, local shifts were observed (Fig. 3B–D). Unsurprisingly, NADP(H)-bound R132Q had the typical open, inactive conformation seen in WT and R132H, with a larger active site cleft and smaller back cleft relative to the quaternary complexes (Supplementary Table 1). These distances in the binary R132Q structure more closely resembled binary WT than R132H, supportive of a more closed, catalytically competent ground state for R132Q. However, R132Q exhibited notable differences compared to WT and R132H binary complexes. In particular, the clasp domain and helices proximal to the substrate and cofactor binding site were shifted, with the α1, α2, α4, α5, and α11 helices adjusted upwards and inwards in R132Q versus WT and R132H, resulting in a similar shift of the NADP(H) molecule itself in dimer-based alignments (Fig. 3B). Importantly, this inward shifting of the α1 helix is a feature of closed, catalytically competent IDH1 conformations. R132Q also contained longer, more intact β strands in the clasp domain, which plays a major role in maintaining the dimer, compared to WT and R132H (Fig. 3C). The fully intact β7 and β8 strands in R132Q were reminiscent of quaternary, fully substrate-bound forms of IDH1 WT and R132Q (vide infra). Consistent with such stable secondary structure, peptides in the β8 strand of R132Q:NADP(H) had lower deuterium uptake than WT:NADP(H) and R132H:NADP(H) (Fig. 2, Supplementary Fig. 6). IDH1 R132Q also maintained an extensive hydrogen bonding network envelo** the NADP(H) molecule; this network was far less robust in R132H (Supplementary Fig. 8). Together, dynamic and static structural data suggested that the IDH1 R132Q active site pocket and surrounding features have greater rigidity and more defined structural features typical of fully-substrate-bound forms of IDH1, suggesting a more catalytically primed state for R132Q:NADP(H) compared to R132H:NADP(H).

Fig. 3: Crystal structure of NADP(H)-bound IDH1 R132Q shows a typical open conformation.
figure 3

A The binary R132Q:NADP(H) complex is shown with each monomer highlighted using a slight color change. B Dimer-based alignments of R132Q:NADP(H) (red), WT:NADP(H)13 (black), and R132H:NADP(H) (light green)30. C Monomer-based alignments of the structures in (B). D The view show in (C) was simplified to highlight catalytic residues Y139 and K212 (though the latter residue drives catalysis in the monomer not shown as this is a monomer-based alignment), residue R132(H/Q), and the cofactor.

Full size image

ICT-bound R132Q is in a closed conformation

Here, we also report an ICT-bound quaternary structure of IDH1 R132Q (R132Q:NADP(H):ICT:Ca2+, Fig. 4A). Upon alignment with WT:NADP(H):ICT:Ca2+ 13 (Fig. 4B, C, Supplementary Fig. 9B), there was obvious overlap in both global features and active site details. ICT-bound IDH1 R132Q also aligned well with R132H bound to its preferred substrate, αKG (R132H:NADP(H):αKG:Ca2+)14 (Fig. 4B, C, Supplementary Fig. 9D). Like ICT-bound WT and αKG-bound R132H structures, ICT-bound R132Q adopted a catalytically competent, closed conformation, with ICT maintaining many of the same polar interactions with the protein and divalent ion as observed with WT. This is supportive of our kinetic data showing R132Q’s preservation of the conventional activity.

Fig. 4: IDH1 R132Q bound to ICT, NADP(H) and Ca2+ is in a closed, catalytically competent conformation.
figure 4

A The R132Q:NADP(H):ICT:Ca2+ complex is shown with each monomer highlighted using a slight color change. B Monomer-based alignments of R132Q:NADP(H):ICT:Ca2+/R132Q:NADP(H):Ca2+ monomers (dark and light cyan) with WT:NADP(H):ICT:Ca2+13 (dark green); R132H:NADP(H):ICT30 (wheat); and R132H:NADP(H):αKG:Ca2+14 (dark purple). C For clarity, only the catalytic residues, residue R132X, cofactor, substrates, Ca2+ and hinge are shown in the same orientation for the structures shown in (B).

Full size image

Though alignment of ICT-bound WT and R132Q was strikingly similar (Supplementary Fig. 9B), the 220-fold decrease in catalytic efficiency suggested that maintaining hydrogen bonding features and active site structuring was not sufficient for robust conventional activity in R132Q. Interestingly, ICT was observed only in one monomer of the R132Q quaternary complex, resulting in a shift of the α11 helix and the NADP(H) molecule upward and outward in the ICT-absent R132Q monomer (Fig. 4C), reminiscent of the WT:NADP(H) binary structure (Fig. 3). This lack of active site saturation suggested a lower affinity toward ICT for R132Q versus WT. Though Km values are not affinity measurements, it is noteworthy that there was a 32-fold increase in Km when comparing R132Q to WT (Supplementary Fig. 1). To address differences in binding affinity, we again turned to ITC experiments. ICT binding affinity for IDH1 R132H was too poor to be detected, while R132Q exhibited ~170-fold worse affinity for ICT compared to WT (Supplementary Fig. 4). Structural studies provided a possible mechanism for ICT’s poor binding to R132H versus R132Q; in contrast to the closed, catalytically competent conformation of ICT-bound R132Q, a previously solved ternary R132H:NADP(H):ICT30 structure revealed quasi-open monomers that had α4 and α11 helices shifted upwards and outwards from the dimer interface and an unraveled α10 helix (Fig. 4B, Supplementary Fig. 9C), regions we and others have shown to be highly flexible13,30,31,32. Notably, ICT was found in a posited pre-binding site that was shifted to the left of its catalytically-competent position (Fig. 4C)30. This resulted in limited polar interactions by ICT to R132H30 in contrast to ICT’s extensive polar contacts to R132Q, including hydrogen bonding to catalytic residue Y139 in R132Q that indicated a catalytically-ready binding conformation (Supplementary Fig. 8). As further evidence that ICT-bound R132H was ill-prepared for catalysis, its catalytic residues were swung away from the active site (Fig. 4C), akin to the positioning found in binary, catalytically incompetent IDH1 structures. Though this R132H structure did not include a divalent metal that may be required for full closure30, it is nonetheless unsurprising that R132H, in contrast to R132Q, is essentially unable to convert ICT to αKG.

αKG-bound R132Q has a shifted binding pocket

Since IDH1 R132Q distinctly maintains both conventional and neomorphic catalytic abilities, we asked how the binding conformations for ICT, the conventional reaction substrate, and αKG, the neomorphic reaction substrate, compared. Here, we report two αKG-containing R132Q quaternary structures (R132Q:NADP(H):αKG:Ca2+). These co-crystallization experiments led to a variety of complexes, with monomer asymmetry observed (Fig. 5). One structure contained a dimer that had αKG and a covalent NADP-αKG adduct bound in its monomers (Fig. 5A). Cleft measurements in both monomers indicated a slightly more open conformation when compared to the closed quaternary R132Q (ICT-bound), WT (ICT-bound) and R132H (αKG-bound) structures, with the α11 helix shifted out away slightly from the substrate binding pocket (Fig. 5, Supplementary Fig. 10). As a result, the NADP(H) itself shifted outwards compared to the ICT-bound R132Q structure, resulting in a semi-closed conformation (Fig. 5D, Supplementary Table 1).

Fig. 5: Crystal structure of IDH1 R132Q bound to αKG and NADP-adducts are in closed and semi-closed conformations.
figure 5

In (A–C) and (E), a description of the ligands present is listed below each monomer. A R132Q:NADP(H):αKG:Ca2+/R132Q:NADP-αKG:Ca2+ dimer. Each R132Q monomer is highlighted using a slight change in color. B R132Q:NADP-αKG:Ca2+/ R132Q:NADP(H):αKG:Ca2+ dimer 1 (yellow) aligned with the dimer shown in (A) (magenta). C R132Q:NADP-αKG:Ca2+/R132Q:NADP(H):Ca2+ dimer 2 (orange) aligned with the dimer shown in (A) (magenta). D Monomer-based alignment of ICT- and αKG-containing R132Q monomers. E R132Q:NADP-TCEP:Ca2+/R132Q:NADP-TCEP:Ca2+ dimer. F Monomer-based alignment of adduct-containing R132Q monomers.

Full size image

A second αKG-bound structure had distinct features among two dimers in the crystallographic asymmetric unit. One catalytic dimer contained one NADP-αKG adduct and one αKG molecule (Fig. 5B), and again appeared as an intermediate between the R132Q:NADP(H) and the R132Q:NADP(H):ICT:Ca2+ structures (Supplementary Table 1, Supplementary Fig. 10). A second dimer contained an NADP-αKG adduct in one monomer, and no αKG-containing molecule in the other monomer (Fig. 5C). This dimer was in a more closed, catalytically competent conformation, reminiscent of the fully closed WT quaternary structure (Supplementary Table 1, Supplementary Fig. 10). The Ca2+ ion clearly led to extensive restructuring, as the R132Q:NADP(H):Ca2+ monomer aligned relatively poorly with the R132Q:NADP(H) complex despite the only difference being the metal ion (Supplementary Fig. 10G). Thus, closing of R132Q to the αKG-bound form may be driven just as much by metal binding as by substrate binding. This finding was recapitulated by the overall decrease seen in deuterium uptake upon treatment of substrate-bound R132Q with Ca2+ (Supplementary Fig. 5). Overall, we were able to capture snapshots of stable conformations of αKG binding ranging from semi-closed (αKG-bound) to essentially fully closed (NADP-αKG adduct-bound).

Closed conformations are seen for WT13 and R132H14 when bound with their preferred substrates (ICT and αKG, respectively). As αKG-bound R132Q was often not as fully closed as the ICT-bound form, we wondered how αKG-bound R132Q compared to these WT and R132H closed conformations. In alignments of R132Q:NADP(H):αKG:Ca2+ with quaternary WT and R132H structures, the catalytic residue Y139 in R132Q was shifted away from the αKG molecule (Supplementary Fig. 10), with this molecule making fewer hydrogen bond contacts within the R132Q active site compared to R132H (Supplementary Fig. 8). In R132Q, the αKG binding site was shifted upwards towards NADP(H) and away from the substrate binding sites seen in the ICT-bound WT and αKG-bound R132H structures. This shift might be facilitated by one surprising feature of all non-αKG-containing R132Q monomers -- the nicotinamide ring could not be reliably modeled due to missing electron density. This suggests that when αKG was absent (such as in the R132Q:NADPH:Ca2+ monomer in Fig. 5C) or, more unexpectedly, even when αKG was bound (R132Q:NADPH:αKG:Ca2+ monomers), this portion of NADP(H) was more dynamic. Since the αKG-containing R132Q structures did not appear in a catalytically-ready form, it is possible that the enzymatic mechanism may rely on different amino acids used in the conventional reaction, or, since αKG serves as a substrate and product for R132Q, we may have a view into a product-bound conformation.

ICT-bound and αKG-bound R132Q are structurally distinct

We found that the α10 regulatory segment underwent the expected notable restructuring upon substrate binding, with this segment forming a helix in both the ICT- and αKG-bound quaternary forms of R132Q (Fig. 5D), just like in ICT-bound WT and αKG-bound R132H. However, our HDX-MS experiments captured more subtle differences in R132Q that depended on which substrate was bound. The α10 regulatory segment and nearby α9 helix were more protected from deuterium exchange in both αKG and αKG + Ca2+ conditions in R132Q than in the ICT and ICT + Ca2+ conditions (Figs. 2 and 6). Beyond its proximity to the regulatory segment, the α9 helix has an additional role in active site remodeling in that it helps form a “seatbelt” envelo** the NADP(H) cofactor (reviewed in ref. 33). This seatbelt was observed in the WT:NADP(H):ICT:Ca2+ quaternary structure13, with residue R314 in the α11 helix shifted inward to form polar contacts with D253’ and Q256’ in α9 of the adjacent monomer and with a water molecule (Fig. 7). The absence of the seatbelt was not limited to binary R132Q, R132H, and WT structures; no seatbelt was observed in the ternary ICT-bound or, more surprisingly, in the closed, quaternary αKG-bound R132H structures14,30. As no αKG-bound WT structure is available at this time, we compared a structure of a non-R132 mutant, G97D, which generates D2HG but exhibits a high degree of structural similarities with IDH1 WT14. The αKG-bound form of G97D also did not show a seatbelt conformation, suggesting this is a distinct feature of ICT-bound, fully closed structures.

Fig. 6: Deuterium uptake by IDH1 WT, R132Q, and R132H in helices bounding the substrate binding pocket.
figure 6

Deuterium uptake is shown as a gradient from red (high uptake) to blue (low uptake). A Deuterium uptake by IDH1 WT, R132Q, and R132H upon no ligand treatment. These HDX-MS data were overlaid on NADP(H)-only bound forms of WT13 in all three cases, as the αKG helix was disordered in the NADP(H)-only bound forms of IDH1 R132Q and R132H30. B Deuterium uptake by WT and R132Q upon treatment with NADP+ and ICT, and by IDH1 R132Q and R132H upon treatment with NADPH and αKG. These HDX-MS data were overlaid on WT:NADP(H):ICT:Ca2+ 13, R132Q:NADP(H):ICT:Ca2+ and R132Q:NADP(H):αKG:Ca2+, or R132H:NADP(H): αKG:Ca2+ 14. C Deuterium uptake by IDH1 WT and R132Q upon treatment with NADP+, ICT, and Ca2+, and by IDH1 R132Q and R132H upon treatment with NADPH, αKG, and Ca2+. These HDX-MS data were overlaid on the structures described in (B).

Full size image
Fig. 7: Hydrogen bond network facilitates a “seatbelt” that overlays NADP(H) in only some quaternary structures of IDH1.
figure 7

A Unlike the binary structure of IDH1 WT13 and quaternary structure of G97D:NADP(H):αKG:Ca2+14, the quaternary IDH1 WT complex13 forms a seatbelt over the NADP(H). B Binary R132Q:NADP(H) and quaternary R132Q:NADP(H):αKG:Ca2+ structures do not form a seatbelt, while R132Q:NADP(H):ICT:Ca2+ and the most closed conformation of R132Q:NADP-αKG:Ca2+ form a seatbelt. C No seatbelt is formed in the binary R132H:NADP(H), ternary R132H:NADP(H):ICT, or quaternary R132H:NADP(H):αKG:Ca2+ structures of IDH1 R132H14,30.

Full size image

IDH1 R132Q behaved like WT (Fig. 7A) when binding the conventional reaction substrate (ICT), with a seatbelt forming over the cofactor since residue R314 was in position to contact Q256’, D253’, and, distinct in this protein, E247’ in β11 of the adjacent monomer, as well as a water molecule (Fig. 7B). However, R132Q behaved more like R132H (Fig. 7C) when binding the neomorphic substrate, with αKG-bound monomers showing residue R314 swung away from the α9’ helix, precluding the necessary polar contacts (Fig. 7B). Interestingly, the closed R132Q:NADP-αKG:Ca2+/R132Q:NADP(H):Ca2+ dimer (Fig. 5C) had an intact seatbelt over the NADP-αKG adduct (Fig. 7B), suggesting that a fully closed conformation of αKG-bound R132Q is possible if the nicotinamide ring of NADP(H) is stabilized in some way, such as via adduct formation. Interestingly, HDX-MS dynamics showed seatbelt formation was associated with an increase in deuterium uptake, with the α11 helix, which contains the seatbelt-forming R314 residue, being more protected in the αKG-bound R132Q and R132H (seatbelt-lacking) complexes relative to the ICT-bound WT and R132Q (seatbelt-forming) complexes (Fig. 6). We note that all of these mutant structures (both R132H and R132Q) describe mutant:mutant homodimers; as these mutations are found heterozygously in patients, a possibility exists for WT:mutant heterodimers, which could result in still different structural features and conformations. Overall, multiple conformations were possible with αKG-containing R132Q structures, including those associated with fully closed forms.

R132Q accommodates multiple NADP-containing adducts

In addition to the NADP-αKG adduct, we encountered an NADP-tris(2-carboxyethyl)phosphine (NADP-TCEP) adduct when attempting to crystallize ICT-bound R132Q (Fig. 5E, Supplementary Fig. 11). There may be catalytic relevance to these adducts since the TCEP and αKG carboxylates helped coordinate Ca2+ and maintained many hydrogen bonds in their respective active sites, though the metal ion was slightly shifted to accommodate these adducts (Supplementary Figs. 10, 12). All TCEP and αKG adducts appeared as hybrids between the semi-closed, αKG-bound and fully closed, ICT-bound R132Q complexes (Supplementary Table 1). In general, one NADP-αKG adduct-containing monomer (Fig. 5C) aligned well to the fully closed ICT-bound R132Q structure in all regions except the clasp domain, where the adducted monomer was shifted towards the dimer interface and the β9 strand was more intact (Fig. 5, Supplementary Fig. 10). As further evidence of its fully closed conformation, this NADP-αKG adduct-containing monomer also had an intact seatbelt (Fig. 7B).

To better understand how these adducts were forming, we performed density functional theory (DFT) calculations for model NADP-TCEP and NADP-αKG adducts (Supplementary Tables 2 and 3), which suggested that adduct formation would not occur if not for the constraining environment of the crystal structure. We considered an alternative possibility that the R132Q active site favored adduct formation and binding. If the NADP-TCEP adduct could form in the R132Q active site, it would act as a competitive inhibitor. Thus, we treated R132Q with varying concentrations of three reducing agents (TCEP, dithiothreitol (DTT), and β-mercaptoethanol (BME)) to determine the effects of conventional reaction catalysis (Supplementary Fig. 13, Supplementary Table 4). Dose-dependent inhibition of R132Q catalysis was profound with TCEP, while DTT and BME had minimal effects. More modest, though notable effects on catalysis were also observed when challenging WT with the highest concentration of TCEP tested (10 mM, Supplementary Fig. 14). Together, these results support the hypothesis that adduct formation occurred outside of the non-physiologically-relevant crystal packing environment, with the adducts mimicking αKG binding, ICT binding, or transition between the two.

As these adduct-containing structures showed hybrid binding features of αKG and ICT, we wondered if transition state features could be extrapolated. Here, the nicotinamide ring of the adduct lent an interesting clue. Calculations suggested that the nicotinamide ring is likely planar in the oxidized form34,35. During NADP+ activation for hydride transfer, the enzyme is predicted to distort the nicotinamide ring to form a puckered transition state as a partial positive charge on C4N develops34,35,36 (Supplementary Fig. 15). NAD(P)-adducts with reducing agents have been reported previously, including with TCEP37 and DTT38, and were found to have a more puckered nicotinamide ring, reminiscent of a transition state. Here, unlike the planar ring observed in our non-adducted forms of NADP(H) (R132Q:NADP(H):ICT:Ca2+), both the αKG- and TCEP-containing NADP-adducts showed a more puckered nicotinamide ring (Supplementary Fig. 12, Supplementary Table 3), suggestive of a transition-state-like conformation.

In summary, we highlight discrete catalytic and structural features among two tumor-relevant IDH1 mutants, with the R132Q mutant serving as an invaluable tool to probe the journey through substrate turnover of two reactions that typically cannot be performed by the same enzyme. Together, our kinetics experiments and static and dynamic structural data suggested that substrate binding and conformational changes associated with the conventional and the neomorphic reactions have distinct paths through turnover that can be described in terms of differences in substrate affinity, substrate binding site location, solvent accessibility, and propensity for conformational activation and active site remodeling (summarized in Fig. 8). IDH1 R132Q’s accommodation of catalytically-relevant adducts, perhaps due to its active site appearing better optimized for catalysis compared to R132H, illuminate snapshots of substrate and substrate analogs in varying degrees of catalytic readiness.

Fig. 8: Conformations and solvent accessibility of IDH1 WT, R132Q, and R132H upon substrate binding.
figure 8

Helices displaying profound differences in alignment of the three forms of IDH1 are highlighted. The seatbelt feature is indicated on the α11 and α9 helices. A Binary WT:NADP(H)13 collapses to a closed conformation upon ICT binding, though moderate levels of deuterium exchange are still permitted. B Binary R132Q:NADP(H) collapses to a closed conformation upon ICT binding, showing improved catalytic efficiency for the conventional reaction and lower deuterium uptake compared to R132H. However, catalytic activity is much lower compared to WT. C Binary R132H:NADP(H)30 collapses to a fully closed conformation only upon αKG binding14, but a seatbelt is not formed and deuterium uptake remains high. D Binary R132Q:NADP(H) forms semi-closed and closed conformations upon binding αKG and NADP-αKG, respectively, with a seatbelt successfully formed in the closed state in some of our crystallographic snapshots. The αKG binding site was shifted away from the α9 helix, though catalytic activity was much higher than that seen in R132H.

Full size image

Discussion

Kinetic, HDX-MS, and crystallography experiments revealed fundamental differences in the catalytic mechanisms of WT and tumor-relevant IDH1 mutants (Fig. 8). We were surprised to identify adducts binding to and, in the case of the TCEP adduct, inhibiting R132Q. While we do not expect this adduct to form under physiologically conditions, our experiments measuring reducing agent inhibition were supportive of the possibility of these adducts representing catalytically meaningful conformations. Determining that the NADP-TCEP adduct was competitive with ICT was unsurprising as the TCEP portion of the adduct mimicked features of ICT binding to R132Q (Supplementary Fig. 12). While this experiment itself does not differentiate whether the adduct forms outside the protein and then binds, or whether the active site pocket drives adduct formation, our DTT experiments may support a model where the enzyme drives adduct formation. NADP-DTT adducts have been previously reported with yeast xylose reductase38. While DTT has some similar structural features compared to ICT, it does not recapitulate the carboxylates that the TCEP and αKG adducts contain. These adducts preserved many polar contacts that non-adducted NADP(H) and ICT form with Ca2+ and active site residues. Further supportive of this adduct being distinct to R132Q, and thus perhaps formed with the help of this enzyme, TCEP was much less effective at inhibiting WT (Supplementary Fig. 14). This work also highlights a liability for using TCEP as a reducing agent in kinetic and structural studies on dehydrogenases, as adduct formation with NAD(P)+ may complicate structure/function analysis.

We reported previously that IDH1 R132Q has distinct catalytic profiles for the conventional and neomorphic reactions compared to more common tumor-driving IDH1 mutants (R132H, R132C)21,22. Though both mutations retain a polar amino acid, a mutation to a glutamine versus a histidine would be expected to be less disruptive due to a more similar size and shape relative to arginine, though it is unsurprising that WT is far more efficient at catalyzing the conventional reaction than the mutants since R132 coordinates the C3 carboxylate of ICT3,13. As neither mutant can directly participate in this coordination with ICT, we asked why the conventional reaction was more efficient in R132Q than R132H. We found that R132Q employed an active site water that mitigated the loss of hydrogen bonding to ICT resulting from the R to Q mutation by imperfectly mimicking the polar interactions with the substrate normally afforded by R132 (Supplementary Fig. 16). Despite the shifting of the αKG binding site, we noticed a similar compensatory mechanism in our αKG-bound R132Q structure, with a water molecule again recapitulating these polar interactions. Here, however, the water molecule did not appear to hydrogen bond with the substrate. Instead, a second water molecule was found at the same location as the Ca2+ ion in the quaternary ICT-bound WT IDH1 structure (Supplementary Fig. 16A), which presumably helped stabilize the αKG substrate in R132Q. We previously reported the importance of water molecules in facilitating mutant IDH1 inhibition31, and this current work highlights the importance of water in substrate binding by providing a possible mechanism by which R132Q is more catalytically efficient compared to R132H.

In addition to affecting catalysis, the α10 regulatory segment may serve as a selectivity filter for mutant IDH1 inhibitor binding64,65. This treatment of solvation effectively models the species as though they were in solution rather than crystalline form. Harmonic frequency analysis was carried out to obtain the vibrational corrections needed to calculate the free energies. Finally, because basis set superposition error can be substantial relative to intermolecular bond energies, the counterpoise correction was applied to our final energies of reaction66,67. The transition state (TS) for the TCEP + NADP+ binding was identified and confirmed by analysis of the single imaginary vibrational frequency. The DFT calculations for the model NADP-TCEP adduct predicted values of 25° for \(\Delta {\theta }_{{{{{{\rm{C}}}}}}}\) and −11° for \(\Delta {\theta }_{{{{{{\rm{N}}}}}}}\), where the experimental values in the X-ray structure were \(\Delta {\theta }_{{{{{{\rm{C}}}}}}}\) = 29.2° and \(\Delta {\theta }_{{{{{{\rm{N}}}}}}}\) = −1.1° (Supplementary Table 2). For the NADP-αKG adduct, agreement was similar, with DFT predicting \(\Delta {\theta }_{{{{{{\rm{C}}}}}}}\) = 29° and \(\Delta {\theta }_{{{{{{\rm{N}}}}}}}\) = −14° as compared to \(\Delta {\theta }_{{{{{{\rm{C}}}}}}}\) = 25° and \(\Delta {\theta }_{{{{{{\rm{N}}}}}}}\) = −25° in the X-ray structure (Supplementary Table 2). The binding was energetically favored, and appeared to occur without barrier when vibrational effects were included, with a calculated binding energy of 9.4 kcal mol−1 at 298 K. However, the calculated free energies indicated that in solution, the entropy decrease would preclude spontaneous binding. Quenching the translational entropy of the species in the crystal may be what allowed the process to occur. We noted that the counterpoise corrections to the transition state and adduct energies were essential, having magnitudes of 7–8 kcal mol−1 and comparable to the uncorrected energy differences.

For the dihedral angles, the deviation from planarity \(\Delta \theta\) of the NADP pyridine ring in the adduct was reported using the average of two dihedral angles. Numbering the carbon atoms in the ring by convention as shown in Supplementary Fig. 15, the C-P bond in NADP-TCEP formed at atom 4. The positions of the N atom 1 and the opposite C atom 4 are referenced to the plane defined by the roughly coplanar atoms 2, 3, 5, and 6. The average of the dihedral angles 2-3-5-4 and 6-3-5-4 (Supplementary Fig. 15) was subtracted from 180° to yield \(\Delta {\theta }_{{{{{{\rm{C}}}}}}}\) as a metric for the deviation from planarity of C4, while the average of 3-2-6-1 and 5-2-6-1 subtracted from 180° is used to calculate \(\Delta {\theta }_{{{{{{\rm{N}}}}}}}\) for N1. A sign convention was applied such that if \(\Delta {\theta }_{{{{{{\rm{C}}}}}}}\) and \(\Delta {\theta }_{{{{{{\rm{N}}}}}}}\) had the same sign, the two corners of the ring bend away each other in chair fashion, whereas opposite signs indicate a boat-like conformation. Comparison of the results from the different functionals and basis sets showed very little difference in the geometry. Optimized geometries obtained with the pc-2 basis set on a smaller geometry (omitting sugar and explicit waters) were not significantly different from those obtained with pc-1, so we chose to report the B3LYP/pc-1 results here, with the sugar and explicit waters included (Supplementary Table 2). An additional geometry optimization was run on the NADP-αKG adduct with two explicit waters and a -2 charge, employing the aug-pc-1 basis set59,68 to obtain the diffuse functions necessary to adequately model anions.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.