Water Networks and Correlated Motions in Mutant Isocitrate Dehydrogenase 1 (IDH1) Are Critical for Allosteric Inhibitor Binding and Activity

Abstract

Point mutations in human isocitrate dehydrogenase 1 (IDH1) can drive malignancies, including lower-grade gliomas and secondary glioblastomas, chondrosarcomas, and acute myeloid leukemias. These mutations, which usually affect residue R132, ablate the normal activity of catalyzing the NADP⁺-dependent oxidation of isocitrate to α-ketoglutarate (αKG) while also acquiring a neomorphic activity of reducing αKG to d-2-hydroxyglutarate (D2HG). Mutant IDH1 can be selectively therapeutically targeted due to structural differences that occur in the wild type (WT) versus mutant form of the enzyme, though the full mechanisms of this selectivity are still under investigation. Here we probe the mechanistic features of the neomorphic activity and selective small molecule inhibition through a new lens, employing WaterMap and molecular dynamics simulations. These tools identified a high-energy path of water molecules connecting the inhibitor binding site with the αKG and NADP⁺ binding sites in mutant IDH1. This water path aligns spatially with the α10 helix from WT IDH1 crystal structures. Mutating residues at the termini of this water path specifically disrupted inhibitor binding and/or D2HG production, revealing additional key residues to consider in optimizing druglike molecules against mutant IDH1. Taken together, our findings from molecular simulations and mutant enzyme kinetic assays provide insight into how disrupting water paths through enzyme active sites can impact not only inhibitor potency but also substrate recognition and activity.

Graphical Abstract

Mutations in isocitrate dehydrogenase 1 (IDH1) drive the majority of lower-grade gliomas and secondary glioblastomas, as well as a significant fraction of chondrosarcomas and acute myeloid leukemias.^1–5 Most tumor-driving IDH1 mutations confer two properties: deficiency in the normal NADP⁺-dependent oxidation of isocitrate to α-ketoglutarate (αKG) and, more critically, the gain of a neomorphic NADPH-dependent reduction of αKG to form the oncometabolite d-2-hydroxyglutarate (D2HG), which drives tumorigenesis (Figure 1). Importantly, mutant IDH1 has proven to be an effective anticancer therapeutic target, with the first selective R132H mutant IDH1 inhibitor recently approved for use in the clinic against liquid tumors.^7,8

Figure 1.

Catalytic reactions catalyzed by WT and mutant IDH.⁶

Interestingly, there is vast structural and chemical variation among the IDH1 mutations that drive tumors; while R132H and R132C IDH1 are by far the most common, R132L, R132S, R132Q and R132G IDH1 have also been identified.^9,10 We have recently characterized the catalytic efficiency of the normal and neomorphic reactions catalyzed by tumor-relevant IDH1 mutants and showed that most mutants are unable to catalyze αKG production while there is wide catalytic variation in the neomorphic reaction.^11,12 Notably, R132Q IDH1 maintained the ability to catalyze the normal reaction while also having a high efficiency for D2HG production,¹¹ and this unique catalytic profile appeared to confer resistance to binding mutant IDH1 inhibitors.¹² All-atom molecular dynamics (MD) simulations of the IDH1 homodimers showed that near the NADP⁺/NADPH binding site in R132QIDH1, a pair of α helices switch between conformations that are more WT-like or more mutant-like, and that this switch occurs in concert with hydrogen bond shifts between NADP⁺ and IDH1.¹²

The α10 regulatory domain (Figure 2A), consisting of one of the helices that undergoes such conformational changes, is an important structural element that is posited to have a regulatory role in the protein.^13,14 In the holo [isocitrate and NADP(H)-bound] crystal structures of WT IDH1 [e.g., Protein Data Bank (PDB) entry 1T0L¹³], these residues are in a helical from, while they are a loop in apo [NADP(H)-bound only] crystal structures (e.g., PDB entry 1T09¹³). However, this domain is highly disordered in the apo form of R132H IDH1 (Figure 2) (e.g., PDB entry 3MAR¹⁵) in contrast to the well-defined loop structure in WT IDH1.¹³ An array of mutant IDH1 crystal structures bound to various inhibitors (such as PDB entries 5TQH,¹⁶ 6B0Z,¹⁷ 5K11,¹⁸ and 5DE1¹⁹) reveals that the inhibitor binding site is dramatically dynamic and accompanied by unfolding of these α10 helix residues, which neighbors the inhibitor binding site. It has been proposed that the flexibility of these residues may serve as a selectivity gate for mutant IDH1 inhibitors, enabling preferential inhibitor binding to the mutant form.¹⁴

Figure 2.

(A) WT IDH1 holo dimer, with each monomer coordinated with isocitrate, NADP⁺, and Ca²⁺ colored magenta and light pink (PDB entry 1T0L¹³). This holo structure is overlaid with that of WT IDH1 in the apo form, shown as cyan and dark blue monomers (NADP⁺-bound only, PDB entry 1T09¹³). The α10 regulatory domains are colored green in the holo WT IDH1 structure, which is in the helical form, and in orange for apo WT IDH1, which is unwound. (B) Zoomed-out view of panel A. (C) The NTD (yellow), NADP⁺ binding site (purple), αKG binding site (blue), dimer interface (cyan), and CTD (orange) are highlighted in a crystal structure of R132H IDH1 complexed with NADP⁺ and GSK321A (PDB entry 5DE1¹⁹). The zoomed-in view is shown in panel C, with the full view in panel D.

In drug discovery research, finding handles that improve selectivity and potency is a critical component of successfully generating targeted small molecule antitumor therapies. Here we expand on our previous work on understanding the molecular characteristics of mutant IDH1 catalysis¹¹ and inhibition¹² by examining the role of water dynamics and protein residue-correlated motions in these processes. By using all-atom explicit solvent MD simulations including WaterMap analyses,²⁰ we identify a series of high-energy waters that lie along a path that connects dimer interface residues with active site residues and NADP⁺ binding site residues. This path is reminiscent of allosteric pathways that have been characterized through computational methods,^21–23 except that this allosteric pathway passes through the first water shell of IDH1 and not through protein residues. Protein residues at the termini of these water paths in the IDH1 dimer interface were mutated to establish their role in inhibitor binding and catalysis. In particular, site-directed mutagenesis of these sites indicated that the dynamics of residue M259 plays an important role in catalyzing D2HG production and binding mutant IDH1 inhibitors, while conservative L120 mutations can impact inhibitor binding to varying degrees without altering the water energetics, thus maintaining enzyme activity.

METHODS

Preparation of Proteins and Ligands for Molecular Modeling.

System preparation and modeling for all publicly available IDH1 protein structures were performed using the Schrödinger Suite (2018-4, 2019-1, 2019-2).²⁴ Protein Preparation Wizard was used to prepare all protein structures at pH 7.4. Prime was used to fill in missing side chains or loops when necessary (Table 1).

Table 1.

Comparisons of Binding Features from Structures of Commercially Available Mutant IDH1 Inhibitors Bound to R132H IDH1

PDB ID	Resolution (Å)	Missing loops/modeled buried cavity volume	# Ligands in binding pocket	Chemical structure	Inhibitor binding site symmetry	Potency range of series
5TQH16	2.2	135-137, 272-278	Two		not symmetrical	10 nM-2 μM
6B0Z17	2.33	135-140, 272-276	Two		not symmetrical	4 nM-100 nM
5DE119	2.25	134-139, 272-278 648 Å	Two		symmetrical	4 nM-11μM
5K1118	3.8	135-139, 272-280 397 Å	Unknown		symmetrical	44 nM-10 μM

Modeling the ML309-Bound Dimeric Structure.

Phase, a part of the 2019-1 Schrödinger suite,²⁵ was used to generate structure-based pharmacophore models of the ML309 binding site in the prepared R132H IDH1 structure (PDB entry 5K11¹⁸). Half of the pharmacophore models produced demonstrated symmetry in the features along the two monomer faces of the binding site. ML309 was docked into all of these models as a monomer and then translated into the adjacent monomer’s binding site. Of these dockings, only one resulted in an ML309 dimeric model without steric clashes between the two ML309 molecules. This structure was further refined with minimization and MD before testing with free energy perturbation.

Free Energy Perturbations.

FEP calculations were performed using version 2019-1 of the Schrödinger suite.²⁵ The OPLS3e force field, GPU-enabled parallel MD engine Desmond, REST-enhanced sampling technique, and cycle-closure correction algorithm were used to generate the free energy estimate. The REST region included only the heavy atoms of the ligand. Missing force field torsion parameters were added by the Force Field Builder panel that performs additional QM calculations for ligand torsions. We report theoretical error estimates compared with experiment via the correlation coefficient and the mean unsigned error.

WaterMap Simulations.

WaterMap simulations were run for all prepared protein structures using Schrödinger Suite release 2019-1.²⁶ The binding site was defined for each structure by residues D273, K212, W267, L120, and Q283 of IDH1 in both of the chains of the protein. The simulation was run for 2.0 ns using the OPLS3e force field, and waters within 10 Å of the binding site were analyzed. For structures with a cocrystallized ligand, the calculations were re-run with the apo form of the structure.

MD Simulations.

MD simulations were run for prepared protein structures of PDB entries 5K11,¹⁸ 5K11 (C132H IDH1 modified from ref 18), 4UMX,²⁷ 4UMY,²⁷ 4KZO,²⁸ and 6BL0²⁹ using Schrödinger Suite 2019-2 and 2019-3.²⁶ For PDB entries 5K11, 4UMX, 4UMY, and 4KZO structures, systems were prepared using an orthorhombic water (SPC) box and neutralized with Na⁺ and with 150 mM NaCl. Systems were then relaxed and simulated for 100 or 1000 ns using an NPT ensemble at 300 K and the OPLS3e force field. Different random seeds were used for the 100 ns replicates with subsequent trajectories concatenated for analysis. For C132H IDH1 (PDB entry 5K11¹⁸) and R132H IDH1 (PDB entry 6BL0³⁰), the previously prepared PDB 5K11 and 6BL0 structures were mutated to histidines at position 132 on both chains and prepared using the Protein Preparation Wizard (Schrödinger Suites 2019-2 and 2019-3²⁶) with default settings except for H bond optimization at pH 7.4. Systems were prepared using an orthorhombic water (SPC) box, neutralized with Na⁺, and with 150 mM NaCl, and then minimized for 100 ps. The MD simulation was performed using an orthorhombic water (SPC) box, neutralized with Na⁺, and with 150 mM NaCl. Systems were then relaxed and simulated for 100 ns with 10 different random seeds with subsequent trajectories concatenated for analysis.

Chemicals and Reagents.

Tris-hydrochloride, sodium chloride, magnesium chloride hexahydrate, αKG (sodium salt, adjusted to a pH of 7.5 with 5 M NaOH before use), and resazurin were purchased from Fisher (Hampton, NH). Diaphorase from Clostridium kluyveri, DMSO, DTT, and ML309 were purchased from Sigma-Aldrich (St. Louis, MO), though the manufacturer could not confirm if ML309 was stereospecific or racemic. NADPH (tetrasodium salt) was purchased from Calbiochem (San Diego, CA). GSK864 was purchased from Cayman Chemical Co. (Ann Arbor, MI).

cDNA Construct Design, Protein Expression, and Purification.

All IDH1 constructs are in a pET-28b vector with an N-terminal hexahistidine tag. Site-directed mutagenesis was used to generate single-point mutations in R132H IDH1 according to the manufacturer’s guidelines (Kappa Bioscience, Oslo, Norway) using the following primers: L120V, 5′-cattccgcgtgtggttagcggtt (forward) and 5′-tttttgcaaataattgcttcacgaaaaacg (reverse); L120A, 5′-attatttgcaaaaacattccgcgtgcggttagcggttggg (forward) and 5′-cccaaccgctaaccgcacgcggaatgtttttgcaaataat (reverse); M259R, 5′-attgatgatatggttgcacaggcacgcaaaagcgaaggtggttttatttgg (forward) and 5′-ccaaataaaaccaccttcgcttttgcgtgcctgtgcaaccatatcatcaat (reverse); M259E, 5′-gatgatatggttgcacaggcagagaaaagcgaaggtggttttat (forward) and 5′-ataaaaccaccttcgcttttctctgcctgtgcaaccatatcatc (reverse); and M259A, 5′-gatgatatggttgcacaggcagcgaaaagcgaaggtggttttat (forward) and 5′-ataaaaccaccttcgcttttcgctgcctgtgcaaccatatcatc (reverse). R132H, R132H/L120A, R132H/L120V, R132H/M259R, R132H/M259E, and R132H/R259A IDH1 were expressed in BL21 gold (DE3) Escherichia coli cells and purified using nickel-nitrilotriacetic acid column chromatography (Quiagen, Valencia, CA) as previously described.¹¹ All purified protein was used within 1 month of purification. For ITC measurements, following nickel-nitrilotriacetic acid column chromatography, R132H, M259E/R132H, and L120V/R132H IDH1 were further purified via size-exclusion chromatography. Proteins were eluted in a buffer containing 50 mM Tris-HCl (pH 7.5 at 4 °C), 100 mM sodium chloride, and 1 mM DTT. Fractions were combined and dialyzed against 50 mM Tris-HCl (pH 7.5 at 4 °C), 100 mM sodium chloride, and 2 mM β-mercaptoethanol. Protein was concentrated to 30–50 μM and used immediately for ITC measurements. At least two different protein preparations were used when testing catalytic activity to ensure batch-to-batch consistency.

Steady-State Kinetic and Inhibition Assays.

The rate of NADPH-dependent conversion of αKG to D2HG was measured as previously described.¹¹ Briefly, 1–30 μM protein in 50 mM Tris (pH 7.5 at 37 °C), 150 mM NaCl, and 0.1 mM DTT was preincubated at 37 °C, and the reaction was initiated with saturating concentrations of NADPH (200 μM) and αKG (0.5–30 mM). ΔA₃₄₀ was monitored to measure the rate of NADPH consumption as described previously,¹¹ which were reported as k_obs values and plotted against substrate concentration. Results were fit to the hyperbolic Michaelis–Menten equation using GraphPad Prism (GraphPad Software, La Jolla, CA) to estimate k_cat and K_m mean values ± the standard error (SE). Concentrations were performed as replicates (typically duplicates), and the average values are shown. In the case of M259R/R132H IDH1, no measurable activity was observed despite using a wide range of enzyme to increase signal and high substrate concentrations in the case of very poor substrate binding. Eventually, high concentrations of the substrate compounded signal-to-noise issues. Biochemical IC₅₀ measurements were obtained as previously described,^7,12,29 though for M259A/R132H and M259E/R132H IDH1, the enzyme concentration was increased to 40 nM and the ML309 concentrations ranged from 50 nM to 100 μM. Because K_m values could not be obtained for M259R/R132H IDH1 due to its inactivity, IC₅₀ measurements could not be taken for this mutant.

CD and ITC Measurements.

A subset of mutants was selected to test protein secondary structure and stability using CD and affinity for ML309 using ITC. In CD measurements, IDH1 (5 μM) was incubated with 100 mM KCl and 10 mM potassium phosphate monobasic (pH 7.5) and wavelengths from 190 and 280 nm were scanned on an AVIV (Lakewood, NJ) model 420 circular dichroism spectrometer. The melting point was measured at 222 nm by scanning samples at increasing temperatures (from 19 to 70 °C). Data were fit using established methods.³¹

ITC experiments were performed at 23 °C using an Affinity ITC instrument (TA Instruments, New Castle, DE). Twenty 3–4.3 μL injections of 150 μM ML309 were made at 150 s intervals into the ITC cell containing 38–45 μM IDH1. Assay buffer also contained 20 or 50 mM Tris (pH 7.5) at 4 °C, 100 mM NaCl, 2 mM β-mercaptoethanol, 5% DMSO, and 133–266 μM NADPH. ITC data were analyzed using a NanoAnalyze software package provided by the ITC manufacturer.

RESULTS

Free Energy Simulations: Aligning Theory with Experiment.

Inhibitors at the protein–protein interface present a unique challenge in drug discovery, especially if accompanied by conformational changes upon binding. Such is the case with allosteric inhibitor binding to mutant IDH1. The impact of water thermodynamics and residue dynamics on inhibitor potency and neomorphic activity were unknown for tumor-relevant IDH1 mutants. To confidently analyze the atomic-level dynamics of R132H IDH1 through in silico methods, it is important to verify the accuracy of the starting crystal structures and the loops and residues that were modeled due to poor resolution. To do this, we turned to FEP calculations³² to attempt to recapitulate the inhibitor affinities for a ligand series (Table 1).

Only a few of the many published structures of R132H IDH1 with inhibitors bound were amenable to FEP.³³ We identified four mutant IDH1 structures bound to commercially available inhibitors derived from a series of published compounds with a suitably wide biochemical potency range, thus making them appropriate for comparison with FEP calculations (Table 1). All of these structures have two ligands bound to the dimer interface. Of these, two structures (PDB entries 5TQH¹⁶ and 6B0Z¹⁷) have different conformational states of the α10 helix in each monomer, resulting in a different chemical environment for each ligand. Given this difference, we chose not to pursue FEP calculations on these series because each ligand monomer had a different starting conformation that would complicate calculations tremendously. The GSK (PDB entry 5DE1¹⁹) and ML309 series (PDB entry 5K11¹⁸) both have symmetrical structures for the α10 region neighboring the ligand binding site for each monomer and an FEP-amenable range of inhibitor affinities. There was a significant number of loops missing in the crystal structures of the GSK-bound (PDB entry 5DE1¹⁹) and ML309-bound (PDB entry 5K11¹⁸) crystal structures, including the loop consisting of the unwound α10 helix that directly neighbors the inhibitor binding site. Missing loops and solvent accessible regions in protein crystal structures adversely affect the ability to model the binding affinity of compound series bound to those structures.³⁴ This is why we chose to use FEP calculations to validate our starting structures before pursuing analyses of the atomic-level dynamics of waters and residues.

In the structures of R132H IDH1 bound to GSK321A (PDB entry 5DE1¹⁹), the α10 loops were built into the structure and refined using Prime³⁵ and energy minimization protocols (Figure S1A). Initial models of the GSK series bound to R132H IDH1 were generated using the MCS docking protocol within Schrödinger Suites²⁴ starting from the structure of PDB entry 5DE1.¹⁹ We also generated models of the ML309-bound structure based on the cryoelectron microscopy structure of ML309-bound R132H IDH1¹⁸ (Figure S1B and Figure 3). Models of the ML309 series bound to R132H IDH1 were more difficult to generate because the starting structure of the ML309 series was obtained using cryoelectron microscopy¹⁸ and the electron density could not be refined to adequately model the ML309 ligand.

Figure 3.

Comparison between the (A) GSK321A¹⁹ and (B) modeled ML309^18,36 binding poses. Inhibitors are shown as sticks.

The series of compounds, including ML309, consists of peptide derivatives that bind between the dimer interface. The IDH1 conformation is different when ML309 is bound compared to the GSK compound bound. Notably, in the ML309-bound cryo-EM structure,¹⁸ the α10 helices are completely unwound as loops, with electron density able to resolve the C-terminal portion of each monomer’s α10 loop (residues 281–284). These loops occupy space similar to that of the α10 helices that were resolved for the GSK series-bound crystal structures¹⁹ (Figure S1C). Additionally, the α11 helices of the ML309 structure align well with the α11 helices of an AG-881-bound series (PDB entry 6ADG,³⁷ Figure S2). Taken together, this crystallographic evidence suggests that the ML309 series binds to R132H IDH1 as a dimer, just as the GSK¹⁹ and AG-881³⁷ series do.

After loop building and refinement with Prime,³⁵ the volume of the ML309-bound R132H IDH1 cavity at the dimer interface was measured to be 397 ų, compared to the volumes of the GSK- and AG-881-bound structures of ~600 ų. On the basis of the symmetry seen in the electron density from the crystal structure, we find that a dimer of ML309 is a better fit of the electron density than the monomer. The large volume difference between the ML309-bound model and the GSK/AG-881-bound crystal structures is highlighted in Figure 3. A structure-based pharmacophore model was built from the ML309-bound cryoEM structure, which exhibited notable symmetry in the binding site (Figure S3). Docking ML309 into this pharmacophore resulted in a reasonable model with extensive interactions between the ML309 dimers at the R132H IDH1 dimer interface (Figure 3B). Subsequent FEP calculations of the ML309 series did not find a correlation between experimental and theoretical IC₅₀ values despite the ML309 compound fitting the electron density well. It is likely that a novel FEP methodology will be needed to adequately model the transition between the twin ML309 series ligands at the dimer interface.

With the GSK series, a strong correlation was seen between the experimental and theoretical binding affinities, with an R² of 0.93 and a root-mean-square error (RMSE) of 0.58. To the best of our knowledge, this is the first example in the literature of correct IC₅₀ predictions for ligands bound as dimers at a protein dimer interface. FEP calculations confirmed this improvement in affinity upon the addition of magic methyls³⁸ to the phenyl ring in the transition from GSK009 to GSK849 (Figure S4) and to the piperidine ring from GSK849 to GSK303 (Figure S5A,B). Thus, the modeled α10 loops were correctly able to predict the affinities of the GSK series and were able to correctly model the displacement of high-energy waters through magic methyl additions in the series.

Water Paths Connecting the Inhibitor Binding Site with Substrate Binding Sites.

We sought to investigate the role of these high-energy waters, and water thermodynamics in general, in water displacement and replacement in the apo structures of the GSK and ML309 series. All-atom MD simulations of the R132H IDH1 dimer were performed to investigate the dynamics of the ML309 apo structure (PDB entry 5K11¹⁸) around the inhibitor and substrate binding sites using WaterMap.^20,26 A network of high-energy waters was identified in many of the crystal structures of R132H IDH1, including 5K11,¹⁸ suggesting that water dynamics may play a role in the correlation of motions between the dimer interface residues and the NADPH binding site. Water dynamics in enzyme active sites have previously been hypothesized to contribute to the rates of enzyme catalysis, including increasing catalytic rates in the presence of higher-salt environments.³⁹

The GSK-bound (PDB entry 5DE1¹⁹) and ML309-bound (PDB entry 5K11¹⁸) structures both have a path of high-energy waters from residues at the dimer interface to the NADP⁺ binding pocket. Given our previous findings¹² about the possible correlation of motions between the two monomers in the NADP⁺ binding sites, we hypothesized that these high-energy waters were important for maintaining activity. Thus, we sought to identify mutations that could impact the thermodynamics of the waters in this path. In GSK-bound structures (PDB entry 5DE1¹⁹), a high-energy water path around the α10 helix along the top of the αKG binding cavity connects the dimer interface with the nicotinamide side of NADP⁺ through interactions with the backbone of residues S287 and L288 (Figure 4) in mutant IDH1. In contrast, the ML309-bound structures (PDB entry 5K11¹⁸) show a high-energy path of waters that connects the dimer interface with the adenosine side of NADP⁺ (Figure 4). In both structures, the remaining waters in the surrounding NADP⁺ and αKG binding cavity are low-energy, stable waters (Figure 4).

Figure 4.

Analysis of water thermodynamics reveals a network of waters connecting the α10 regulatory domain with the NADP⁺ binding site in R132H IDH1. High-energy waters are shown as red spheres connecting residue M259 to the NADP⁺ binding site for the ML309-bound cryoEM structure 5K11¹⁸ for the in silico mutants (A) R132H, (B) R132H/L120V, and (C) R132H/M259E. Stable waters (small green spheres) were omitted from panels A–C for the sake of clarity but included in D to show how many low-energy waters in the GSK-bound structure occupy the same space as the high-energy water path seen in the ML309 structures (A–C). These same analyses on the (D) GSK321A-bound crystal structure (5DE1¹⁹) reveal a different water path around the folded C-terminus of the α10 domain. Key residues are labeled. (E) Average ΔG (blue), ΔH (red), and TΔS (yellow) values and distances between waters (green) were measured for each of the water paths in the IDH1 structures. While these values were within error of each other, it is worth noting that the greatest difference was seen for the average enthalpy of the waters along these two paths. The significance of this is that in the GSK-bound structures, the waters are more enthalpically stable and can participate in inhibitor binding, which is important for driving future structure-based drug design approaches on the IDH1 dimer interface. Protein regions are colored as in Figure 2. In panels A–D, the water pathways were cut and overlaid onto a transparent version of the original image to highlight the pathways.

The corresponding water path in WT IDH1¹³ follows the same path as seen with the GSK321-bound mutant structure (Figure 4D) but contains far fewer waters (Figure 5). Interestingly, in WT IDH1, α10 helix sits in the same space as the high-energy water path identified in the GSK321-bound mutant structure (Figure 5). The α10 helix in the WT IDH1 structure also has high B-factors and is in a different conformation in each of the monomers. Taken together, these observations suggest that in WT IDH1, the dynamics that result in correlated motions between the NADP⁺ binding site and L120/M259 IDH1 are mediated by the α10 helix, while in the mutant inhibitor-bound IDH1 dimer, they are mediated by high-energy waters.

Figure 5.

Analysis of water thermodynamics of WT IDH1 (PDB entry 1T09¹³). The unraveled α10 helix (white) and monomers A (cyan) and B (green) are shown with NADP⁺ bound. Key residues in the α10 helix that bridge the high-energy waters are L288, Y285, and S280 of IDH1. Distances between waters and key residues are colored pink.

The ends of both of these paths of high-energy water molecules in the inhibitor-bound structures (Figure 4) terminate at the dimer interface near residues M259 and L120 in R132H IDH1. In silico mutations were made to these residues to see if the water dynamics could be tuned by these residues alone. The M259E mutation stabilized the most proximal neighboring water, resulting in a reduction in the number of waters in the high-energy water network connecting the binding interface with the NADP⁺ binding site and altering the water thermodynamics in the inhibitor binding site (Figure 4A,C). The L120V mutation maintained the water thermodynamics of the high-energy water network while altering the relative positions and energetics of waters in the inhibitor binding site (Figure 4A,B). Thus, we hypothesized that mutations at the dimer interface could be made either to affect inhibitor binding alone or to alter inhibitor binding and activity.

Disrupting Correlated Motions with Dimer Interface Mutants.

All-atom explicit solvent MD simulations were used to uncover the correlation of residue motions in residues in the N-terminal domain (NTD) (residues 1–79) and C-terminal domain (CTD) (residues 329–415) that bind NADP⁺, and in the dimer interface that binds isocitrate, αKG, and inhibitors. Analysis of the R132H-correlated motions and WaterMap results reveal that M259 and L120 represent key positions in allosteric networks between dimers, in the interaction with residues in the α10 regulatory domain, and with the NADP⁺ binding site (Figure 4). Given this finding, MD simulations of in silico mutants of M259 and L120 were made to assess their effects on these correlated motions (Figure 6). There are three observations from the correlated motion heat maps (Figure 6) of the R132H mutant and the R132H/M259/L120 mutants that provide key insights into how these mutations could affect IDH1 dimer motions. First, for all L120/M259 mutants, the positive correlation of motions among the NADP⁺ (light blue), αKG/isocitrate binding sites (dark blue), and dimer interface residues (cyan) across the two monomers is significantly reduced compared to that of the R132H background. This difference in correlations across dimers and between each monomer’s active site could impact the conformation of the dimer interface as well as the stability of the quaternary structure. Second, for the M259R/R132H IDH1 mutant, there is an increase in negatively correlated motions between the two monomers, and an increase in positively correlated motions within residues within each monomer. Third, for all mutants but especially L120V/R132H IDH1, there is a reduction in the negatively correlated motions within each monomer between the residues in the NTD and the CTD and residues in the NADP⁺ (light blue), αKG/isocitrate binding sites (dark blue), and dimer interface residues (cyan) as compared with WT IDH1.

Figure 6.

Heat map showing the correlated motions altered by mutations in L120 and M259 of the 5K11 structure¹⁸ for (A) R132H IDH1, (B) R132H/L120V IDH1, (C) R132H/L120A IDH1, (D) R132H/M259E IDH1, (E) R132H/M259A IDH1, and (F) R132H/M259R IDH1. Residues with motions that are positively correlated are colored red, whereas residues with motions that are negatively correlated are colored blue. Each of the subdomains of the IDH1 dimer is indicated on the x-axis and y-axis of the heat map, including the NTD, the NADP⁺ binding site, the αKG/isocitrate binding site, the IDH1 dimer interface, and the CTD.

Experimental Validation of MD Simulations.

To confirm the impact of these residues on correlated motions necessary for activity and inhibitor binding, site-directed mutagenesis was used to generate M259A, M259E, M259R, L120A, or L120V within the R132H IDH1 background, and these double mutants were heterologously expressed and purified from E. coli. R132H IDH1 was selected to serve as a representative tumor-relevant mutant due to its very high frequency among mutant IDH1-driven cancers.⁹ To ensure mutations at M259 and L120 did not significantly affect protein folding, we analyzed M259E/R132H and L120V/R132H using CD spectroscopy (Figure S6). Minimal changes in these proteins were observed compared to R132H IDH1, indicating that proteins were stable and overall folding was intact. Interestingly, M259E/R132H IDH1 had a slightly lower T_m (45.6 ± 0.1 °C) compared to those of L120V/R132H IDH1 and R132H IDH1 (49.79 ± 0.09 and 49.76 ± 0.05 °C, respectively), though the difference was not significant (Figure S6).

In general, mutations at residue L120 had negligible effects on the production of D2HG by R132H IDH1. There was a 3.3-fold decrease in the catalytic efficiency of the production of D2HG by L120A/R132H, while L120V/R132H showed a 1.9-fold increase (Figure 7 and Table 2). However, there were measurable effects of this mutation on binding affinity for the selective mutant IDH1 inhibitor ML309³⁶ (Figure 8A and Table 2). L120A/R132H IDH1 had a 10-fold increase in IC₅₀. L120V/R132H IDH1 had a smaller impact on ML309 binding affinity, with a 2.3-fold increase in IC₅₀ and a 5.2-fold increase in K_d. A more notable impact was observed for the less selective GSK864 inhibitor,¹⁹ where IC₅₀ values increased 60–120-fold (Figure 8B and Table 2).

Figure 7.

Concentration dependence of αKG on the observed rate of NADPH depletion by IDH1 mutants. Values for k_obs were determined from the linear portion of plots of NADPH concentration vs time at varying concentrations of αKG normalized for enzyme concentration. Plots shown here were fit with a hyperbolic equation to generate k_cat and K_m values. At least two enzyme preparations were used, and each point represents at least two replicates. Results from R132H IDH1 have been previously reported.¹²

Table 2.

Kinetic Parameters for IDH1 Double Mutants for the Conversion of αKG to D2HG in the Presence and Absence of Mutant IDH1 Inhibitors^a

mutant	kcat (s−1)	Km (mM)	kcat/Km (mM−1 s−1)	ML309 IC50 (μM) (confidence interval)	ML309 Kd (μM)	GSK864 IC50 (μM) (confidence interval)
R132H	1.44 ± 0.0512	1.5 ± 0.212	1.0 ± 0.112	0.04 (0.03–0.05)12	0.48 ± 0.0512	0.005 (0.004–0.008)12
L120A/R132H	2.0 ± 0.2	6 ± 2	0.3 ± 0.1	0.4 (0.2–0.8)	–b	0.6 (0.4–0.9)
L120V/R132H	1.44 ± 0.03	0.75 ± 0.07	1.9 ± 0.2	0.09 (0.06–0.1)	2.5 ± 0.4	0.3 (0.2–0.6)
M259A/R132H	1.0 ± 0.2	26 ± 9	0.04 ± 0.02	0.7 (0.2–2)	–b	0.9 (0.5–2)
M259R/R132H	≤0.07c	NAc	NAc	NAc	NAc	NAc
M259E/R132H	2.1 ± 0.4	21 ± 10	0.10 ± 0.05	>100	11 ± 6	>50

^aFor k_cat, K_m, k_cat/K_m, and K_d values, deviations from equation-based fitting are shown as standard errors, and confidence intervals are listed for IC₅₀ values.

^bNot measured; only a subset of mutants were selected for K_d measurements to identify trends.

^cNo measurable activity was detected despite testing a range of enzyme and substrate concentrations. This lack of activity (and subsequent inability to obtain a K_m measurement) precluded making IC₅₀ measurements.

Figure 8.

Biochemical IC₅₀ measurements for (A) the selective mutant IDH1 inhibitor ML309³⁶ and (B) the pan-inhibitor GSK864.¹⁹ The loss of the ability of mutant IDH1 to convert αKG to D2HG was monitored with increasing inhibitor concentrations. An IC₅₀ value was not reported for M259E/R132H IDH1, as no inhibition was observed at concentrations where the inhibitor was soluble.

Interestingly, the catalytic efficiency of the production of D2HG by R132H IDH1 was significantly decreased upon the introduction of mutations at M259. The most deficient mutant was M259R/R132H IDH1, which had no measurable catalytic activity even at high enzyme concentrations (0.1–10 μM) and a wide range of αKG concentrations (0.5–50 mM) (Figure 7 and Table 2). It is possible that the deficiency seen in this mutant is due to the increase in negatively correlated motions between the two monomers as measured from the MD simulations. This precluded the ability to obtain IC₅₀ values of the inhibitor binding to this mutant. M259E/R132H and M259A/R132H IDH1 also had significant catalytic defects, though not as severe as M259R/R132H IDH1. Catalytic deficiencies decrease 10- and 25-fold for M259E/R132H and M259A/R132H IDH1, respectively, which were driven primarily by increases in K_m (Figure 7 and Table 2). The loss of affinity for both ML309 and GSK864 was striking for these two mutants. Notably, M259E/R132H IDH1 had a 23-fold increase in K_d for ML309 binding compared to R132H IDH1, and inhibition was undetectable in IC₅₀ measurements for both inhibitors (Figure 8 and Table 2). A decreased level of inhibitor binding was also observed for M259A/R132H IDH1, though not as severe (Figure 8 and Table 2).

DISCUSSION

The conservative mutations L120V and L120A in the R132H IDH1 background had negligible effects on catalytic activity, with modest changes in K_m mitigated through corresponding changes in k_cat in the case of L120A. Despite minimal changes in catalytic efficiency, small but notable decreases in binding affinity were observed for ML309, with more drastic decreases seen with GSK865 binding. Modeling of the ML309-bound structure provides one possible explanation of this observed discrepancy. First, ML309 (PDB entry 5K11¹⁸) was predicted to bind as a dimer of ligands at the dimer interface of IDH1 mutants in a much more compact structure than that seen for GSK865 binding (PDB entry 5DE1¹⁹). Second, GSK865 displaces high-energy waters that reside proximal to L120, and these high-energy waters are not predicted to be as prominent in the L120V or L120A mutant. Finally, while these waters are not prominent, the high-energy water path from between the dimer interface residues L120 and M259 to the αKG and NADP⁺ binding sites is conserved in the L120V and L120A mutants. This suggests that the water dynamics that guide catalytic efficiency is maintained in these mutants while altering the binding efficiency of inhibitors due to slight shifts in water thermodynamics.

The impact of mutations on M259 was far more striking. M259R/R132H IDH1, which introduces a bulkier and more polar residue, essentially ablated catalytic activity, possibly by increasing the amount of negatively correlated motions between residues in each monomer. M259E/R132H IDH1 had a significant impact on catalytic efficiency due to an increase in K_m and had a drastic influence on inhibitor affinity. We note that biochemical IC₅₀ measurements have limitations as values can vary significantly depending on experimental conditions. While protein steric hindrance may play a role in M259R/R132H IDH1 deficiencies, this is less likely to drive the changes observed in M259E/R132H IDH1. This suggests that the change from a nonpolar to a polar residue at this position can have a major impact on inhibitor binding and catalysis. These data support our hypothesis that correlated motions facilitated by destabilized waters are important for substrate, cofactor, and inhibitor binding, and that inserting polar residues to stabilize these waters disrupts the ability of this water channel to accommodate the residue dynamics required for neomorphic activity. However, it is worth noting that a thorough understanding of how activity and subsequent changes in water dynamics at the atomic level are affected by these mutations will require the use of high-level quantum mechanical simulations.⁴⁰ This will be an interesting direction to explore in the future.

Mutations that hinder water channeling but preserve D2HG production activity, like L120V, may represent possible resistance mechanisms against mutant IDH1 inhibitors. While IDH1 mutations conferring resistance to mutant IDH1 inhibitors have not yet been reported in patients, resistance mutations to the mutant IDH2 inhibitor enasidenib⁴¹ have.⁴² These mutations, Q316E and I319M IDH2, occur at the dimer interface, are found in the domain homologous to the IDH1 α10 helix,⁴² and are not homologous to L120 IDH1, which is conserved. Interestingly, M259 is not conserved in IDH2. However, this domain likely has a very different role in substrate binding in IDH2 as compared to IDH1; for example, this domain is in a fully helical form in both mutant IDH2 holo⁴¹ and apo¹⁴ structures and is predicted to play a minor or negligible role in inhibitor selectivity.¹⁴ Mutations at M259 and L120 IDH1 have not been reported in TCGA,⁹ which is expected as mutations at these sites are unlikely to confer D2HG production ability on their own.

Here, we demonstrate how an in-depth look at water thermodynamics and residue conformational changes associated with IDH1 neomorphic activity can uncover handles in the inhibitor binding site. This same approach could be used in a wide array of drug discovery projects on targets that have large solvent accessible binding sites and in projects targeting a dimer or protein–protein interface. Additionally, understanding the role of the dynamics inherent in water shells around the protein active sites has the potential to provide insight into designing better inhibitors for these enzymes and could even be used to design and engineer new enzymes to perform specific chemistry through a rational approach.⁴³

ABBREVIATIONS

CD

circular dichroism

CTD

C-terminal domain

D2HG

d-2-hydroxyglutarate

DMSO

dimethyl sulfoxide

DTT

dithiothreitol

FEP

free energy perturbation

IDH1

isocitrate dehydrogenase 1

ITC

isothermal titration calorimetry

αKG

α-ketoglutarate

MD

molecular dynamics

NADP⁺

β-nicotinamide adenine dinucleotide phosphate

NADPH

reduced β-nicotinamide adenine dinucleotide phosphate

NTD

N-terminal domain

TCGA

The Cancer Genome Atlas

T_m

melting temperature

WT

wild type