Discovery and Optimization of Allosteric Inhibitors of Mutant Isocitrate Dehydrogenase 1 (R132H IDH1) Displaying Activity in Human Acute Myeloid Leukemia Cells

A collaborative high throughput screen of 1.35 million compounds against mutant (R132H) isocitrate dehydrogenase IDH1 led to the identification of a novel series of inhibitors. Elucidation of the bound ligand crystal structure showed that the inhibitors exhibited a novel binding mode in a previously identified allosteric site of IDH1 (R132H). This information guided the optimization of the series yielding submicromolar enzyme inhibitors with promising cellular activity. Encouragingly, one compound from this series was found to induce myeloid differentiation in primary human IDH1 R132H AML cells in vitro.


■ INTRODUCTION
The enzyme isocitrate dehydrogenase 1 (IDH1) is a nicotinamide adenine dinucleotide phosphate (NADP + )dependent dehydrogenase which catalyzes the conversion of isocitrate to α-ketoglutarate (α-KG). Mutations have been found in both IDH1 (R132H) and IDH2 (R140Q, R172K, R172M) in several cancer types, including up to 70% of low grade and secondary glioma cases and up to 10% of acute myeloid leukemia cases. 1−4 IDH1 has been found to be frequently mutated in its active site at residue arginine 132 to histidine (R132H). This mutation leads to a loss of the normal enzymatic activity of IDH1 and the acquisition of a neomorphic activity, the conversion of α-ketoglutarate to 2-hydroxyglutarate (2-HG). 5−7 2-HG is considered to be an "oncometabolite" which accumulates to high levels and competitively inhibits αketoglutarate-dependent enzymes, such as histone demethylases and TET2, which mediates DNA demethylation. There is an increasing body of research to investigate the effect of inhibition of IDH1 R132H on the cancer disease state. 8−13 A number of inhibitors of IDH1 R132H have been reported 14−24 (Figure 1). The crystal structures of IDH1 R132H containing isocitrate (ICT) have provided some understanding of the change of function of the enzyme. 25 Of the reported IDH1 R132H inhibitors, a number have associated crystallographic data, and these may be separated into two distinct classesthose that occupy the active site of the protein, 16,21,22 such as 2 and 4 (Figure 1), and those that occupy a remote allosteric site which then renders the protein inactive through conformational change, such as 3 and 6. 20,23 As part of the AstraZeneca (AZ) Open Innovation initiative, the Cancer Research UK Manchester Institute (CRUK-MI) and AZ entered into a collaboration to screen IDH1 (R132H) against the AZ compound collection. A high-throughput screen (HTS) of approximately 1.35 million compounds was undertaken against the R132H mutant form of IDH1 in order to identify novel chemical matter for development as inhibitors.
The results of HTS screening (see Figure 2 for screening cascade) are described below. The primary HTS performed well, with a mean Z′ of 0.58, and generated an initial hit rate of 2.1% using a cut off of 50% inhibition at 30 μM compound concentration in the IDH1 R132H biochemical assay.
Compounds were then retested at a lower concentration (10 μM), resulting in 61.3% of hits at 30 μM reconfirming (17126 compounds). False positives (e.g., those that interfered with the assay detection system) were identified using an orthogonal mass spectrometry assay, leaving 50.5% confirmed as active. Prior to IC 50 determination, confirmed actives were subjected to cluster analysis and virtual chemical filtering (to remove potential reactive and assay interference compounds), resulting in the selection of 5976 compounds alongside 2748 "near neighbors" for 10 point dose response determination. In order to further improve the quality of the hit output, compounds were tested in a ratio assay, i.e. testing at 1× and 10× enzyme concentration to identify any additional assay interference compounds. Compounds showing a greater than 0.5 log10 shift in IC 50 were removed from further testing, with the majority of these showing a decrease in potency under the 10× standard enzyme concentration conditions, potentially due to redox cycling or colloidal aggregation.
After these measures, further clustering by substructural analysis and Tanimoto similarity gave 585 compounds of interest, with IC 50 values ranging from 0.3 to 30 μM. These compounds were tested against wild-type IDH1 in a biochemical assay to determine whether the compounds were selective for the mutant isoform. None of the compounds tested showed significant activity against wild-type IDH1.
In parallel, compounds were also examined in a surface plasmon resonance (SPR) assay to confirm direct binding to the IDH R132H enzyme. Compounds that bound with a 1:1 stoichiometry were prioritized for further exploration (data not shown).
The overall result from the described screening cascade and triaging process was that compounds belonging to three distinct chemical series were identified for further evaluation. Previously we had found that the IDH1 (R132H) biochemical assay was sensitive to a number of metal contaminants including palladium, potentially giving rise to false positives. Several other hits, which had used palladium in a late stage of their synthesis, were resynthesized with particular care taken to eliminate any potential contamination of the products with residual palladium by the use of a palladium scavenger (Quadrapure MPA) and purification of reaction products by both normal and reverse phase chromatography. However, one hit series did not use palladium in the synthetic route, and examples were resynthesized in-house without these additional precautions.
On testing the in-house prepared compounds in the IDH1 R132H biochemical assay, only this palladium-free series reconfirmed strongly, suggesting that contamination of the original HTS samples (potentially by palladium) was responsible for the false positive response. The results for the one remaining series are shown in Table 1.
Compounds from this chemical series have previously been reported by AZ as inhibitors of 11β-hydroxysteroid dehydrogenase Type 1 (11β-HSD1), 26 offering us the opportunity to  rapidly examine additional near neighbors of the original hits, sourced from the AZ compound collection.
The preliminary SAR obtained in this manner suggested that the molecule could be simplified. Thus, the hydroxyadamantylamide group was replaced with cyclohexyl-amide and the sulfur linkage replaced by an ether linkage with no loss of enzyme potency, e.g. compound 12 ( Figure 3).
As an aid to guide optimization, attempts were made early in the project to determine the binding mode of the initial hits using X-ray crystallography. The crystal structure of compound 9 complexed with IDH1 (R132H) homodimer in the presence of NADPH (PDB submission code 5L58) was obtained at 3.0 Å resolution ( Figure 4). Although several loop regions proved to be highly disordered at this resolution, the crystal structure revealed that the unit cell contained a symmetrical dimer of (R132H) IDH1 in an "open" conformation, 5 with electron density for compound 9 and the NADP cofactor found in both protomer subunits.
The inhibitor occupies a binding site at the dimer interface in a similar region to that described previously for other allosteric inhibitors. 20,23 This binding site is distinct from the substrate and cofactor binding sites, with the closest distance between 9 and NADP + being approximately 14 Å. Notably, the packing of the two protomers differs significantly between our crystal structure and the structure of (R132H) IDH bound to αketoglutarate (PDB 3INM 5 ), with considerable disruption of the helical dimer interface, in particular with the helix Asn271-Tyr285 being completely disordered.
The binding interactions around 9 are mainly lipophilic in nature, with the only direct protein hydrogen bonds arising from an ionic interaction between the carboxylic acid of compound 9 and Arg119. The central pyridine ring is π-stacked with the side chain of Tyr285. The pyridyl 2-cyclohexylthio group occupies a discrete lipophilic pocket bounded by the residues Trp124 and Trp 267 of chain A and Val255 and Met259 of chains A and B. The -NH-of the pyridine 3carboxamide potentially forms a hydrogen bond with the 2pyridyl sulfur atom, which may serve to constrain the ligand conformation. While the side chain of R132H is located close to the substrate binding site in the α-ketoglutarate-bound crystal structure 3INM (where it presumably modulates the neomorphic catalytic activity), in our structure it has rotated approximately 5 Å away from the substrate site and toward compound 9. Although the 3-carboxamide carbonyl group of 9 points toward R132H, it appears too distant for direct hydrogen bonding. However, the recruitment of R132H to form part of the binding cavity, away from the catalytic site, offers some explanation as to the loss of neomorphic activity in the presence of these compounds. The pendant hydroxy adamantyl group is sited in an area which is largely unresolved (the residues Tyr272−Val281 are missing in the electron density), giving only partial structural information. The adamantyl moiety appears loosely bound, although it makes several lipophilic contacts with Ile251 A , Ile251 B , Val255 B , and Trp267 A . The hydroxyl group is positioned close to the side chain of Asn271 but does not appear to make any direct hydrogen bonds.
Examination of the X-ray crystal structure suggested a number of avenues for the rapid exploration of the SAR ( Figure  5). Thus, for Region 1 (R1), the potential for hydrogen bonding between the pyridyl carbonyl and the protein could be explored, alongside attempts to increase potency by finding additional contacts in the large pocket currently occupied by the hydroxy adamantyl group. The crystal structure shows that the pocket occupied by the -S-cyclohexyl moiety, Region 2 (R2), is a discrete lipophilic pocket and is likely to respond to the precise nature and shape of any "small" substituents. The effects of changes to the linkage could also be explored. Region 3 (R3) is dominated by the charged interaction between the carboxylic acid and Arg119. However, the linkage between the carboxylic acid and the pyridyl ring offers scope for exploration. It was imagined that modification of each of the areas in the molecule in turn would give an understanding of the component SAR, which may allow a rationale for the design of molecules with acceptable in vitro and in vivo DMPK properties, alongside improvements in selectivity against 11β-HSD1.

■ CHEMISTRY
A series of compounds allowing simple substitution at the 2position of the central pyridine (R2) were prepared with the 3carboxamide function held as cyclohexyl and the 6-carboxylic acid function maintained as (S)-2-(piperidin-3-yl)acetic acid using the methods shown in Scheme 1.
2,6-Dichloronicotinoyl chloride was condensed with cyclohexylamine in dichloromethane (DCM) in the presence of diisopropylethylamine (DIPEA) to give 2,6-dichloro-N-cyclohexylnicotinamide 13 as a key intermediate. Displacement of the 2-chloro group was accomplished using the appropriate alcohols and either sodium hydride in DMF or potassium tertbutoxide in THF to give the 2-O-substituted 6-chloro-Ncyclohexyl nicotinamides 14a−az. The regiochemistry of this reaction was confirmed for R = -O-isopentyl by removal of the remaining chlorine atom by hydrogenation and examination of the 1 H NMR aryl region for characteristic coupling patterns. The 6-chloro group was then displaced by ethyl (S)-2-(piperidin-3-yl)acetate using potassium carbonate in butyronitrile at 110°C to give the intermediate esters 15a−az, which  were hydrolyzed to the acid using aqueous sodium hydroxide in methanol to give the carboxylic acids 16a−az. The final ester hydrolysis step was frequently carried out on unpurified isolated esters, allowing expedient production of final compounds.
Compounds designed to explore the SAR around the 3carboxamide group, R1, were prepared either by the sequential route shown in Scheme 1 (although steps were "telescoped" by taking crude isolated products into the next reaction and using preparative HPLC to isolate the final product in a suitable purity) or by using an alternative route which was developed to be more efficient, as shown in Scheme 2.

Journal of Medicinal Chemistry
Article 6-Chloro-2-(isopentyloxy)nicotinic acid 17 was prepared by reaction of 2,6-dichloronicotinic acid with isopentanol (confirmation of the regiochemistry of the alkoxide addition was again obtained by removal of the remaining chlorine atom by hydrogenolysis and examination of the 1 H NMR aryl region for characteristic coupling patterns). Condensation of the aminoester ethyl (1R,5S,6R)-3-azabicyclo[3.1.0]hexane-6-carboxylate (an alternative amino-acid function that had been identified in the original hit set) using conditions previously described followed by hydrolysis of the ester with sodium hydroxide gave the target compounds 20a−ae. The latter two steps were combined without purification of the intermediate ester product.
Compounds bearing alternative carboxylic acid groups were prepared either using similar methods as described above (21 − 23) or by using Suzuki or Buchwald−Hartwig chemistry (24− 28). (For these examples, particular care was taken to remove residual palladium from final compounds to avoid apparent activity due to metal contamination.) Reverse carboxamide compounds 39a−39c were prepared from the appropriately substituted aminopyridine compound 37 as shown in Scheme 3 using cyclohexane carboxylic acid chloride, cyclohexylisocyanate, and cyclohexane sulfonyl chloride as electrophiles.

Journal of Medicinal Chemistry
Article carboxylate followed by addition of the hydroxyadamantyl amine by reductive amination and then hydrolysis of the ester to the carboxylic acid to give compound 47, albeit with poor recovery after preparative HPLC purification.
■ RESULTS Region R2. Compounds 16a−16az with differing substituents in the core pyridine 2-position, R2, were intended to explore the topology and the nature of the small lipophilic pocket identified in the crystal structure of compound 3. These compounds were screened in the IDH1 (R132H) biochemical assay, the results of which are shown in Table 2.
The introduction of a heteroatom (i.e., N or O) to the substituent was poorly tolerated, as exemplified by the methyl ether compound 16s, the pyranyl compound 16t, and the dimethylamine compound 16w, which were all inactive (IC 50 > 50 μM). These results were consistent with the X-ray crystallographic data from the original hit, which showed that the alkyl group at the 2-position occupies a lipophilic pocket devoid of any potential polar interactions. Reducing the length of the O-linked alkyl chain at the 2-position to an isobutyl 16n, ethyl 16u, or methyl group 16v all led to a 3-fold to >7-fold drop off in potency. Given the potencies observed for similarly sized moieties at this position, these data (and the data for 16x and 16y) are rather surprising and remain unexplained.
Attempts to reduce the conformational flexibility of the 2substituent by introducing sp 2 character or a small ring, compounds 16a, 16d, 16j, 16p, and 16q, all led to a decrease in the potency compared with compound 12. The introduction of a 1,1,1-trifluorobutoxy group 16h resulted in a 2-fold decrease in the activity toward IDH1 (R132H) compared with an isopentyl group, roughly equipotent with the butoxyl compound 16g. All this provided confirmation that the binding pocket of the IDH1 (R132H) enzyme preferred lipophilic, nonpolar groups at the 2-position of the pyridine ring and that, from the examples prepared, the isopentyl moiety was optimal.
The introduction of a variety of different substituted Olinked benzyl groups at this position bearing electron-donating or electron-withdrawing substituents is shown in Table 3. Unsubstituted benzyl compound 16z gave an IC 50 of 8.6 μM, representing a marginal decrease in the potency compared with the isopentyl group at this position, 12.
The addition of an electron-donating methyl group to the phenyl ring of the benzyl group had a significant impact on the activity of these compounds. For example, the ortho-methyl (16aa) and meta-substituted benzyl (16ab), compounds led to a 2-fold increase in the potency compared with the unsubstituted system compound 9z. However, the corresponding para-substituted compound 16ac was inactive against IDH1 (R132H). Methoxy-substituted benzyl systems, however, gave a marginal loss in potency for ortho-and meta-substituted benzyl groups, compounds 16ad and 16ae, while the para-methoxy substituted benzyl compound 16af was inactive. Interestingly, 16ad and 16ae contrast strongly with 16s and 16t, where an

Journal of Medicinal Chemistry
Article oxygen atom in a similar spatial location is clearly not tolerated, perhaps highlighting a requirement for additional steric bulk or lipophilic interactions. The introduction of electron-withdrawing substituents onto the phenyl ring of the benzyl group was also investigated. In contrast to the results observed for the electron-donating substituents, the introduction of a fluoro or chloro substituent was tolerated in all positions of the phenyl ring, showing only a marginal increase in potency for the ortho fluoro compound 16ag and equivalent potency for fluorine in the meta-and para-positions and chlorine in all positions.
The introduction of a methyl group at the α-benzylic position of the unsubstituted benzyl was investigated, with the (R)-enantiomer, compound 16an, showing a small increase in potency compared with the unsubstituted system; conversely, the (S)-enantiomer, compound 16ao showed a small loss in potency. Extending from benzyl to phenethyl (16am) gave a significant decrease in the potency, consistent with the small size of the binding pocket.
Taken together, these data show that the introduction of an ortho-methyl substituted benzyl system at the 2-position of the pyridine ring was the optimal benzyl moiety, albeit delivering only a marginal increase in potency at best, compared with the isopentyl substituent (4.5 vs 6.4 μM). Combination of the ortho-methyl group and an α-methyl group, compound 16ap, showed no advantage over the individual substitutions, while combination of the ortho-methyl group and a para-chloro group, compound 16aq, was detrimental.
In order to understand the influence of linkers other than oxygen on the SAR, a small set of matched pairs were prepared in a similar manner to that shown in Scheme 1, using the appropriate starting materials with O, N, and S linker atoms. Full details of the synthesis are given in the Supporting Information.
Comparison of the IDH1 (R132H) activities of the Oisopentyl compound 12 and S-isopentyl compound 16ar showed only a small difference in potencies (Table 4).
However, it was apparent that, in all other examples, the Slinkage was inferior to the O-linkage and that the N-linkage was not tolerated at all, perhaps as a result of the change from an Hbond acceptor to an H-bond donor.
Summarizing SAR in this region, having investigated a wide variety of small lipophilic groups in the R2 position, -Oisopentyl and -O-ortho-methylbenzyl were found to be the preferred substituents for this vector.
Region R1. Compounds 20a−20af, with differing substituents on the pyridine 3-carboxamide, R1, were intended to explore topology and the nature of the more open and less welldefined pocket identified in the crystal structure of compound 9. The results of the investigation can be seen in Table 5. All new compounds were compared against the R1 cyclohexyl carboxamide compound 20ab, enzyme IC 50 7.9 μM.
Aliphatic and cycloaliphatic groups, with or without the addition of heteroatoms or single bridge structures, failed to offer any increase in potency against IDH1 (R132H), except for 20l. Direct aryl attachments gave a range of inhibitory activities from unsubstituted phenyl (>50 μM) (not shown in Table 5) to 20f (2.9 μM), with compounds with a hydroxyl pendent group (direct or indirect) being the most potent, suggesting the

Journal of Medicinal Chemistry
Article possibility of a positive interaction between the hydroxyl group and the protein structure.
Benzyl carboxamides gave a range of activities. The baseline benzyl substituent group 20y was marginally improved upon by simple monosubstitutions, and modest (up to 2-fold) improvements were obtained from the R-α-methyl compound 20t and S-α-methyl compound 20k. Extending the aryl group further with R1 as phenethyl, 20i, gave an additional small increase in enzyme potency.
The most significant improvements in IDH1 (R132H) inhibition came from the use of bulkier structures. Although the adamantyl carboxamide 20b was essentially inactive (>50 μM), substitution of the adamantyl group in the 1-position with hydroxyl 20a or methoxy 20c gave compounds with submicromolar activity. Extending the hydroxyadamantyl group further from the carboxamide link with a methyl substituted methylene 20d gave a similar submicromolar level of activity, while the addition of a difluoromethoxy group 20e gave a noticeable decrease in activity, albeit still at a submicromolar level.
Region R3. Simple replacement of the carboxylic acid group of 20b by ethyl ester (19a) dramatically reduced activity, as might be expected by elimination of the key interaction with Arg119. Although a number of alternative carboxylic acid scaffolds were tolerated, it was noted that the scaffolds in 20b, 21, and 22 were the most effective (Table 6). One exception to this trend was the thiazole carboxylic acid 26, which displayed submicromolar potency, although this is still some three-fold less active than the most potent compound 20a.
With this information in hand, a set of compounds were then synthesized combining the more potent R2 and R3 components with the hydroxyadamantyl carboxamide R1 group to examine the additive nature of the group SAR.
It is evident from the results in Table 7 that the component SAR is not truly additive and that the overall SAR is dominated by the R3 carboxylic acid group. It is logical to suggest that the rigid nature of the aza-bicyclohexane acid, which gave rise to the most potent inhibitors with a small subset of R1 and R2 groups, would be less tolerant of alternative R2 substitutions than the more flexible (and accommodating) piperidine acetic acid.
The X-ray crystal structure obtained for compound 9 indicated that there may be some potential for hydrogen bonding between the pyridine carboxamide -NH-with Tyr285

Journal of Medicinal Chemistry
Article and -CO-with His132, in this latter case via water mediation. In order to understand further the importance of the role of the 3carboxamido group, compounds were prepared by modifying the nature of the 3-carboxamide group by substitution, by reversal of the carboxamide, or by replacement with other groups.
A comparison of benzyl carboxamide compound 20y and cyclohexyl carboxamide compound 20ab with the N-methyl analogues, 20aa and 20ac, shows either no change or a small loss of potency on replacement of the carboxamide -NH-by -NMe-, suggesting that either any potential internal hydrogen bond to the pyridyl 2-oxygen atom added little to the potency of the compounds or the increased rotational freedom of the amide bond is beneficial.
Replacement of the pyridine carboxamide with the reversed carboxamide 39a or urea 39b resulted in complete loss of activity, as did direct replacement of the carbonyl group with a sulfonamide 43 ( Figure 6). Further, reduction of the carbonyl group completely to yield 47 resulted in a complete loss of activity, suggesting that the carbonyl group has a role in hydrogen bonding and/or rotational stability. Only the reverse sulfonamide 39c (8.6 μM) was found to have a similar potency when compared to 20ab. Here, perhaps the differing geometry allows a strong hydrogen bond with His132 or reorientates the R1 lipophilic group improving contact with the enzyme, presenting an opportunity for further study.
Having established that compound 20a was a potent compound against IDH1 (R132H) in this series, we chose to re-examine selectivity against other related enzymes (Table 8).
Selected compounds with a range of enzyme inhibitions (including a related but inactive analogue 16v) were screened in a cellular assay (using mutated R132H isocitrate dehydrogenase 1 transfected HEK293T cellssee Experimental section) measuring the inhibition of production of 2-HG directly by mass spectrometry (Table 9).
Compounds 20a, 21, and 16ag show modest cellular activity for the inhibition of production of 2-HG while displaying no general cytotoxicity against HeLa cells (72-h proliferation assay). There was, however, a 3−10× fall off from the enzyme activity to the cell activity, which may be explained by the poor permeability and high efflux ratio found for compound 20a (Caco-2: P app 2.56 × 10 −6 cm s −1 , Efflux ratio 16.1).
A series of surface entropy reduction mutations were made to R132H to enhance the repeatability of the crystallogenesis. In all five, additional surface residues were mutated (E80A, K81A, E240A, Q242A, and K243A) to alanine residues. These mutations produced a construct, R132H*, that behaved favorably during crystallogenesis and was subsequently used throughout this study.

Article
The crystal structure of compound 20a complexed with IDH1 (R132H) homodimer in the presence of NADPH (PDB accession code 5L57) was obtained at 2.7 Å (Figure 7). The electron density for compound 20a and for the cofactor were found in both dimer subunits. This crystal structure is very similar to that obtained with compound 9, with the inhibitor binding at the interface of the two components of the protein dimer. The ionic interaction between the carboxylic acid of compound 20a and Arg119 is unchanged from the crystal structure of compound 9, as is the π-stacking of the central pyridine ring with the side chain of Tyr285. The pyridyl 2-isopentoxide group occupies the same lipophilic pocket bounded by the residues Leu120, Val121, Trp124, Val255, and Met259 but perhaps makes better contacts. There is an internal hydrogen bond between the -NH-of the pyridine 3-carboxamide and 2-O linker, but no hydrogen bond with the oxygen atom of Tyr 285. The 3carboxamide carbonyl group points in the general direction of His132 but again is too distant for direct hydrogen bonding. The pendent hydroxy adamantyl group is situated in the same area as in the structure of compound 9, making several lipophilic contacts but no direct hydrogen bonds from the hydroxyl group. Note that, as in the previous structure, solvent molecules were not resolvable at this resolution.
In this crystal structure there is sufficient electron density to assign several previously missing loops, including Tyr272−V-al281 and Lys212−Lys218. Of these newly resolved residues, only Val281 interacts directly with the ligand, providing an additional lipophilic contact with the pyridyl ring.
It is informative to compare this crystal structure with those of previously published allosteric inhibitors 3 and 6 ( Figure 8). The binding mode of 20a can be summarized in terms of four key pharmacophore features: hydrogen-bonding between the acidic moiety and Arg119; aromatic stacking between the pyridyl and Tyr285; lipophilic contacts between the isopentyl moiety and the enclosed dimeric pocket around Val255 and

Journal of Medicinal Chemistry
Article Met 259; and looser lipophilic contacts around the adamantyl moiety. Of these interactions, only lipophilic filling of the Val255/Met259 pocket is a consistent feature. In the Sanofi bisimidazole 3 20 structure, this pocket is enlarged due to Trp267 adopting an alternative rotamer, allowing a bulky branched octyl moiety to be accommodated. Overall, the ligand occupies a similar binding location to 20a: when the binding sites are overlaid, the central phenol is observed to be somewhat displaced from the pyridyl of 20a, with the two imidazole moieties aligning loosely with the adamantyl and azabicyclohexane moieties of 20a. A notable difference between the structures is the significant repositioning of the Asn271−Gly286 loop, with Ser280−Gly284 now adopting a short helical structure. As a result, Tyr285 is significantly displaced and hence not available to π-stack to the phenol ring; instead, the side chain of Val281 now occupies a similar position. A further consequence is that Asp279 moves closer to the ligand and forms a direct hydrogen bond with the phenolic hydroxyl. The side chain of Arg119, which formed an ion pair with the acidic moiety of 20a, adopts a different rotamer and does not interact with 3.
Inhibitor 6 again binds in the dimer interface but displays some strikingly different interactions. Although Ser280-Gly284 forms a short helix similar to that in the Sanofi structure, most of this loop (Asn271−Asp279) is unresolved. The pyrazolopiperidine ring system of 6 overlays with the aza-bicyclohexane of 20a, with the fluorobenzyl moiety of 6 filling the Val255/ Met259 lipophilic pocket. The pyrrole ring of 6 extends beyond the acidic moiety of 20a, in effect displacing the side chain of Arg119. However, the most notable difference is the location of the hydroxyethylanilino substituent, which extends between the Ser280−Gly284 helix and the Ile112−Pro127 loop into a region not occupied by either 3 or 20a. Direct protein:ligand hydrogen bonds, unique to 6, are formed between the carboxamide linker and the backbone atoms of Leu120 and Val281, and between the pyrrolo carbonyl and Ile128.

Journal of Medicinal Chemistry
Article binding site at the dimer interface which is broadly similar to that seen for the above compounds.
Although the structural basis of inhibitor selectivity for mutant over wild type IDH1 is not fully understood, it is apparent that these inhibitors bind to a conformation of the protein in which the location of several catalytically important residues is significantly disrupted, including, for example, His132 and Tyr139 (involved in substrate binding) and Asp275 and Asp279 (involved in binding the catalytic magnesium ion). Hence, it is likely that allosteric inhibitors bind preferentially to an inactive conformation that is more accessible to the mutant protein than to wild type. For example, Deng et al. 20 highlight differences in binding affinity for the catalytic magnesium ion between mutant and wild type IDH1.

Article
In summary, comparison of our crystal structures with those of other allosteric inhibitors reveals that the allosteric binding site is accessible to lipophilic ligands of diverse chemotypes, which share some common binding interactions, notably to the Val255/Met259 lipophilic pocket, but also achieve a number of unique polar and lipophilic interactions. The allosteric binding site displays varying degrees of induced fit, as well as differences in the overall packing of the dimers, to accommodate these diverse ligands. Taken together, the crystal structures may guide the future optimization of allosteric inhibitors and facilitate the identification of additional allosteric chemotypes through focused screening campaigns.
We next tested compound 20a in acute myeloid leukemia (AML) cells. Approximately 10% of patients carry IDH1 mutations, and inhibitors of mutant IDH1 and IDH2 are currently being evaluated in early phase trials. 28−30 As no immortalized human AML cell lines carrying native IDH1 mutations are available, we generated an AML cell line system using THP1 AML cells, whereby Myc-tagged IDH1 R132H could be inducibly expressed following the addition of doxycycline ( Figure 9A). As expected, induced expression of mutant IDH1 resulted in production of 2-hydroxyglutarate, and this was inhibited in a dose-dependent manner by compound 20a, with IC 50 1.9 μM and complete reduction to baseline levels at ≥10 μM. Interestingly, treatment of cells with compound 20a had no effect on proliferation or differentiation (data not shown), suggesting limited off-target or nonspecific cellular toxicity.
We next tested the activity of compound 20a versus primary AML blast cells (cultured in semisolid medium) from three IDH1 R132H-mutated patients (BB171; BB340; BB539; BB = Biobank number) and four IDH1 wild type patients. Three doses of 20a reflecting the cellular IC 50 , 4×IC 50 , and 8×IC 50 , or DMSO vehicle control, were used. After 14 days in culture, primary cells exposed to doses commensurate with near-total (9.2 μM) or total (18.4 μM) 2-HG reduction in THP1 cells exhibited significant myeloid differentiation compared with vehicle-treated controls, as evidenced by upregulation of the myeloid differentiation markers CD11b and CD15, downregulation of the stem cell marker CD117 ( Figure 10A), and induction of mature cellular morphology ( Figure 10B). Interestingly, treatment of IDH1 R132H mutant primary AML cells with 20a led to an increase in clonogenic activity ( Figure 10D). This "proliferative burst" is associated with release of the differentiation block and has previously been described following treatment of IDH2 mutant primary AML cells with the mutant IDH2 inhibitor AGI-67801. 28 By contrast, and in keeping with 20a being selective for mutant IDH1, no differences in clonogenic potential or myeloid differentiation were observed in similarly treated IDH1 wild type AML blasts ( Figure 10A, 10D).

■ CONCLUSIONS
As part of a collaborative program, a high throughput screen of 1.35 million compounds against isocitrate dehydrogenase IDH1 (R132H) led to the identification of three series of inhibitors, which, when resynthesized, reduced to a single series (having discarded several positives potentially arising as a result of assay interference by metal contamination). The remaining hit series was rapidly optimized to a moderately potent, cell active compound.
Elucidation of the crystal structure of 9 showed that this series of inhibitors exhibited a novel binding mode in a previously identified allosteric site of IDH1 (R132H). Subsequent elucidation of the crystal structure of the complex between IDH1 (R132H) and inhibitor 20a allowed the position of the previously unresolved protein loop to be observed, and it is hoped that this more complete structure will aid the design of future inhibitors of IDH1 (R132H).
When tested using in vitro assays, the most encouraging compound, 20a, promoted differentiation of patient derived human IDH1 R132H AML cells, but not in IDH1 wild type AML cells, confirming on-target activity in a primary cell setting.

■ EXPERIMENTAL SECTION
Chemical Experimental. All reagents obtained from commercial sources were used without further purification. Anhydrous solvents were obtained from Sigma-Aldrich Chemical Co. Ltd. or Fisher Chemicals Ltd. and used without further drying. Solutions containing products were either passed through a hydrophobic frit or dried over anhydrous MgSO 4 or Na 2 SO 4 , and filtered prior to evaporation of the solvent under reduced pressure. Thin layer chromatography (TLC) was conducted with 5 cm × 10 cm plates coated with Merck type 60 F254 silica gel to a thickness of 0.25 mm. Chromatography was performed on Biotage SNAP HP-Sil cartridges using a CombiFlash Companion machine. Proton (1 H) NMR spectra were recorded on a 300 MHz Bruker spectrometer at ambient temperature. Solutions were typically prepared in either deuterochloroform (CDCl 3 ) or deuterated dimethyl sulfoxide (DMSO-d 6 ) with chemical shifts referenced to deuterated solvent as an internal standard. 1 H NMR data are reported indicating the chemical shift (d), the multiplicity (s, singlet; d, doublet; t, triplet; q, quartet; m, multiplet; br, broad; dd, doublet of doublets; etc.), the coupling constant (J) in Hz, and the integration (e.g., 1H). Deuterated solvents were obtained from the Sigma-Aldrich Chemical Co., Goss, or Fluorochem. LCMS spectra with UV detection were recorded on a Waters Acquity UPLC. Mass spectrometry was performed on a Waters Acquity SQD quadrupole spectrometer running in dual ES + and ES − modes. High pH runs were conducted at pH 10, and low pH runs were conducted at pH 4, with a run time of 2 min. The column temperature was 40°C, and the flow rate was 0.6 mL/min. Further details, including solvent gradients, are given in the Supporting Information. Details of the preparative HPLC instrument and the solvent gradient used to purify compounds are also given in the Supporting Information. All compounds were 95% purity, as determined by examination of both the LCMS and 1 H NMR spectra unless otherwise indicated. When Cl or Br was present, expected isotopic distribution patterns were observed.
Preparation of Compound 20a. 6-Chloro-2-(isopentyloxy)-3pyridine Carboxylic Acid (17). 3-Methylbutan-1-ol (9.93 mL, 91.2 mmol) was added to a stirred suspension of sodium hydride (60% dispersion in mineral oil) (4.17 g, 104.2 mmol) in a mixture of DCM (200 mL) and THF (100 mL) at 0°C. After 20 min of stirring, 2,6dichloropyridine-3-carboxylic acid (5.0 g, 26.0 mmol) was added portionwise over 10 min and the resultant mixture allowed to stir at room temperature for 96 h. To the mixture was added water (300 mL), and the aqueous layer was collected. The DCM solution was extracted once more with water (150 mL), and the combined aqueous layers were then acidified to pH 1 using conc HCl. The acidic aqueous was then extracted with DCM 3(150 mL) and EtOAc (100 mL). The combined organic solution was dried over MgSO 4 and concentrated in vacuo to afford a residue that was purified on a 100g SNAP column using a 0−5% MeOH:DCM eluent to afford 6-chloro-2-isopentyloxypyridine-3-carboxylic acid 17 and triethylamine (40 μL, 0.29 mmol) in DMF (2 mL), and the resulting mixture was stirred for 30 min at room temperature. 4-Aminoadamantan-1-ol (0.10 mmol) was then added, and the mixture stirred for 2 h. The reaction mixture was partitioned between EtOAc (10 mL) and water (10 mL). The organic layer was collected and the aqueous solution extracted with a further portion of EtOAc (10 mL). The combined organic solvents were washed with water (10 mL) and brine (10 mL), dried over Na 2 SO 4 , and concentrated to an oil. The crude intermediate was dissolved in 1,4-dioxane (2 mL), and a solution of sodium hydroxide (77.3 mg, 1.93 mmol) in water (4 mL) was added, and the mixture was stirred for 2 h at 50°C before being acidified to pH 1 with 2 N HCl. The aqueous mixture was then extracted with EtOAc (2 × 50 mL) and the combined organic solution dried over MgSO 4 and concentrated in vacuo. Biochemical Assays. All proteins used in the biochemical assays were produced by the Protein Expression Facility at the Manchester Institute of Biotechnology. The activity of the various dehydrogenase enzymes was measured by coupling NADPH consumption/ production to a diaphorase/resazurin-based Amplite detection system, measuring resorufin production.
All assays were performed in 384-well black plates (Corning #3575) in a reaction volume of 25 μL. IDH1 R132H was incubated at a final concentration of 32 nM with compounds in the presence of 20 mM Tris pH 7.5, 10 mM MgCl 2 , 150 mM NaCl, 0.03% BSA (w/v) buffer, and 1 mM α-ketoglutarate and 50 μM NADPH substrates. IDH1 R132C was incubated at a final concentration of 100 nM with compounds in the presence of 20 mM Tris pH 7.5, 10 mM MgCl 2 , 150 mM NaCl, 0.03% BSA (w/v) buffer with 0.5 mM α-ketoglutarate, and 50 μM NADPH substrates. IDH1 was incubated at a final concentration of 3 nM with compounds in the presence of 20 mM Tris pH 7.5, 1 mM MgCl 2 , 150 mM NaCl, 0.03% BSA (w/v) buffer with 0.2 mM isocitrate, and 50 μM NADP substrates. IDH2 was incubated at a final concentration of 9 nM with compounds in the presence of 20 mM Tris pH 8.0, 1 mM MgCl 2 , 150 mM NaCl, 0.03% BSA (w/v) buffer with 0.2 mM isocitrate, and 50 μM NADP substrates. G6PD was incubated at a final concentration of 7 nM with compounds in the presence of 20 mM Tris pH 8.0, 10 mM MgCl 2 , 0.03% BSA (w/v) buffer with 0.5 mM glucose-6-phosphate, and 50 μM NADP substrates.
All reactions were initiated by the addition of NADPH/NADP. The assays were allowed to proceed at 26°C for 30 min (IDH1 R132C, IDH1, and G6PD) or 60 min (IDH1, R132H, and IDH2) before terminating with the addition of 5 μL of Amplite detection system (The G6PD assay was stopped using 2.5 μL of 0.5% SDS (w/v) prior to the addition of Amplite). Following incubation at room temperature for 30 min, the fluorescence signal at Ex544/Em590 nm was measured using the BioTek Synergy 2 multimode plate reader.
11-β-HSD1 Biochemical Assay. 11-β-Hydroxysteroid dehydrogenase (11β-HSD1) activity was measured in a coupled HTRF immuno-competitive assay. The assay was performed in 384-well black plates (Corning #3575) in a reaction volume of 10 μL in a buffer of 100 mM potassium phosphate pH 7.6 supplemented with 160 nM cortisone (Sigma # C2755), 1.5 μg/mL 11β-HSD1 (Cayman Chem # 10007815), test inhibitor compound, 1 mM EDTA, 1 mM glucose-6phosphate, 12.5 μg/mL glucose-6-phosphate dehydrogenase, and, to start the reaction, 0.1 mM NADPH. The assay was allowed to proceed at 37°C for 30 min before terminating with the addition of 5 μL of 0.5 mM 11β-glycyrrhetinic acid (Sigma # G10105) prepared in cortisol-d 2 solution (Cisbio # 7TRF30D). Following addition of the inhibitor/ acceptor solution, the assay signal was developed by the addition of 5 μL of the donor antibody α-cortisol-cryptate (Cisbio # 7TRF30C). The plate was then incubated at room temperature for 2 h before measuring the TR-FRET signal on a Pherastar FS Microplate Reader. mutIDH1 R132H Cellular Mode of Action Assay. 20 μL of Lipofectamine 2000 (Thermofisher #11668019) was added to 980 μL of OptiMEM reduced serum media (Thermofisher #31985062). Separately 16 μg of pcDNA3.1 mammalian expression vector (Thermo scientific #V79020) carrying isocitrate dehydrogenase 1 mutated at R132H was added to 1000 μL of OptiMEM reduced serum media. Following 5 min incubation, the solutions were mixed and incubated for 30 min at RT, generating a transfection mix. Meanwhile, HEK293T cells cultured in assay media (DMEM (Sigma #D6546), 10%FBS (Sigma #F7524), 2 mM L-glutamine (Sigma #G7513), and 1% Penicillin Streptomycin (Sigma #P0721)) were detached and placed into plating media (DMEM, 1%FBS, and 2 mM L-glutamine). 10 mL of the HEK293T cells (180,000 cells/ml) were then mixed with 2 mL of transfection mix and subsequently seeded into a 96 well poly-Dlysine plate at 100 μL/well (SLS #356461). Plates were then incubated at 37°C with 5% CO 2 for 6 h. Plating media was carefully removed and 100 μL of assay media added to the wells. Compounds were prepared in 100% DMSO, diluted in media, and dosed, making the final assay media volume 150 μL with compounds at a concentration of 0.15% DMSO. After 48 h of incubation with compound at 37°C, 150 μL of media was removed from each well and frozen; 2HG concentrations were determined subsequently by LC-MS/MS. Cells were fixed in 3% formaldehyde and stained with Hoechst, allowing a nuclear count to be performed on the cell insight; 2HG concentrations were normalized to this value.
LC-MS/MS Measurement of 2-HG. Media 2-HG was measured by hydrophilic interaction liquid chromatography (HILIC) LC-MS/ MS. Cryopreserved media samples were defrosted and diluted 1:10 with water containing deuterated (D3)-2-HG as an internal standard (IS). Solid phase extractions were performed using an Oasis MAX 96well-plate (Waters, Milford, MA) containing 10 mg of Oasis MAX sorbent per well. Plates were conditioned with 750 μL of methanol and 750 μL of water before buffered test samples were loaded and pulled through under vacuum. Plates were washed with 750 μL of water and 750 μL of 50% methanol before extracts were eluted with 2 × 250 μL of 1% formic acid in methanol. Eluates were dried under nitrogen gas, reconstituted in initial conditions (88:2:10 acetonitrile:water:50 mM ammonium acetate; pH5.8), and injected into the LC-MS.
Chromatographic separation of 2-HG from coextracted interferences was performed using high pressure liquid chromatography (Agilent 1200, Agilent) with a Luna HILIC 5uM column (150 × 3.0 mm) (Phenomenex, Torrance, CA). The flow rate was 750 μL/min using a gradient starting at 95% mobile phase A, moving to 80% over 180 s before rising back to 95% over 60 s. This was followed by a hold to equilibrate at 95% phase A for a further 150 s. The column temperature was ambient, and the sample compartment was chilled to 10°C. Mobile phase A was acetonitrile:50 mM ammonium acetate
Mass spectrometry detection was made using an API4000 QTrap (AB Sciex, Framingham, MA) operating in negative ion mode with multiple reaction monitoring mass ion transitions 147/85 (2-HG) and 150/88 (D3-2-HG IS). Data was processed using Analyst Software (Applied Biosystems) and signal intensities obtained by standard peak integration methods. Quantitations were performed by comparison against a standard curve generated from multiple dilutions of growth medium spiked with 1 mM 2-HG stock in water to known concentrations. Final 2-HG quantification was derived from a mean of two biological replicates. The lower limit of accurate quantification (LLQ) was set at a level twice the peak area measured in conditioned media, to ensure robust and confident discrimination of 2-HG from low-level background peaks which were observed to coelute during development of the method.
Generation of Stable THP1-IDH1-WT/R132H Clones. Ectopic human IDH1 wild type and IDH1 R132H were stably overexpressed in THP1 AML cells using standard site-directed mutagenesis and lentiviral transduction techniques. Briefly, the coding sequence of human IDH1 (transcript NCBI accession NM_001282387) was PCR amplified from THP1 cDNA with addition of sequence encoding a Myc epitope tag, and ligated into the pGEMT-Easy vector (Promega). Mutant forms were generated by overlap extension PCR and expression construct s ligat ed int o t he pTRE2 min-CMV_MCS_SV40_Blast lentiviral vector. Lentivirus was generated in 293FT cells using standard techniques, and was used to spinoculate THP1 cells previously stably transduced with EF1α_rtTAadv_IR-ES_Neo (i.e., the tetracycline transactivator component). Successfully transduced cells were selected by culture for at least 5 days in 6 μg/mL blasticidin. Ectopic construct expression was confirmed in lysates by Western blotting for the Myc epitope tag and full length IDH1, using standard techniques. Vinculin was used as a loading control. Primary antibodies used were: anti-IDH1 (DH21/rabbit; Cell Signaling, Beverley MA); anti-MYC-TAG (9B11/mouse; Cell Signaling); anti-VINCULIN (hVIN-1/mouse; Sigma-Aldrich).
Primary AML Patient Samples. Primary material was obtained from bone marrow and peripheral blood samples donated to the Manchester Cancer Research Centre (MCRC) Tissue Biobank, following informed written consent and with approval of the South Manchester Research Ethics Committee. Samples were collected, and IDH mutation status was determined as previously described. 31,32 Samples selected for use contained ≥80% blasts.
Flow Cytometry Immunophenotyping. Cell were suspended in staining medium (SM) buffer (5. LC-MS/MS measurement of 2-HG. Extracellular and intracellular 2-HG were measured by hydrophilic interaction liquid chromatography (HILIC) LC-MS/MS, on media supernatant removed from cell cultures or cell lysates, respectively. Cryopreserved media samples were defrosted and diluted 1:10 with water containing deuterated (D3)-2-HG as an internal standard (IS). Solid phase extractions were performed using an Oasis MAX 96-well-plate (Waters) containing 10 mg of Oasis MAX sorbent per well. Plates were conditioned with 750 μL of methanol and 750 μL of water before buffered test samples were loaded and pulled through under vacuum. Plates were washed with 750 μL of water, 750 μL of 50% methanol, and 100% methanol before extracts were eluted with 2 × 250 μL 1% formic acid in methanol. For cell lysates, pelleted cells were washed twice with ice-cold PBS before sequential quenching with −80°C 80% methanol:20% water containing 0.5 μg/mL D3-2-HG IS; transfer to dry ice bed; sonication on a Biorupter (Diagenode) at high power for 5 min (at 30 s on/30 s off); and subsequent incubation at −80°C for ≥40 min. Debris was removed by centrifugation at 14,000g, and supernatant was used for LC-MS/MS. Eluates were dried under nitrogen gas, reconstituted in initial conditions (88:2:10 acetonitrile:water:50 mM ammonium acetate; pH5.8), and injected into the LC-MS.
Chromatographic separation of 2-HG from coextracted interferences was performed using high pressure liquid chromatography (Agilent 1200, Agilent) with a Luna HILIC 5uM column (150 × 3.0 mm) (Phenomenex). The flow rate was 750 μL/min using a gradient starting at 95% mobile phase A, moving to 80% over 180 s before rising back to 95% over 60 s. This was followed by a hold to equilibrate at 95% phase A for a further 150 s. The column temperature was ambient, and the sample compartment was chilled to 10°C. Mobile phase A was acetonitrile:50 mM ammonium acetate (90:10v/v) pH 5.8, and mobile phase B was acetonitrile:water:50 mM ammonium acetate (50:40:10v/v/v) pH 5.8.
Mass spectrometry detection was made using an API4000 QTrap (AB Sciex) operating in negative ion mode with multiple reaction monitoring mass ion transitions 147/85 (2-HG) and 150/88 (D3-2-HG IS). Data was processed using Analyst Software (Applied Biosystems) and signal intensities obtained by standard peak integration methods. Quantitations were performed by comparison against a standard curve generated from multiple dilutions of growth medium spiked with 1 mM 2-HG stock in water to known concentrations. Final 2-HG quantification was derived from a mean of two biological replicates. The lower limit of accurate quantification (LLQ) was set at a level twice the peak area measured in conditioned media, to ensure robust and confident discrimination of 2-HG from low-level background peaks which were observed to coelute during development of the method.

Journal of Medicinal Chemistry
Article (Qiagen), and constructs were sequence confirmed (Stopford Sequencing Facility, Manchester University).
Protein Expression and Purification for Crystallography. pET22b: IDH1 R132H, E80A, K81A, E240A, Q242A, K243A mutation construct was freshly transformed every time into JM109 (DE3) E. coli cells and selected on an LB ampicillin agar plate (100 μg/mL). Culture was grown to an OD600 nm of ∼0.6 and induced by 0.2 mM IPTG with shaking at 30°C for 20 h. Pelleted cells were lysed in 50 mM Tris-HCL, pH 8.0, 300 mM NaCl, and 2% glycerol buffer containing protease inhibitor cocktail (Sigma, EDTA free), 5 mM imidazole, and 1% Triton X-100 (TX100) and then sonicated for 7 × 20 s on/off on ice (setting 20%). The lysate was clarified at 17,000 rpm, 30 min @ 4°C, and bound in batch to Nickel NTA beads (Qiagen) and washed to 20 mM imidazole stringency in lysis buffer lacking TX100. Bound protein was eluted in 250 mM imidazole, 50 mM Tris-HCl pH 8.0, 100 mM NaCl, 2% glycerol buffer. The protein sample was concentrated down to at least 1.6 mL using a Vivaspin20 MWCO 5,000 column and injected onto a HiLoad 16/60 Superdex 200 pg gel filtration (GF) column (GE Healthcare), and the separation was run at 4°C against 50 mM Tris-HCl pH 8.0, 100 mM NaCl without glycerol. For substrate binding, NADP substrate to [0.5 mM] (Sigma) was added to the dilute GF purified protein for 30 min on ice, clarified by spinning at 4.5 Krpm for 10 min at 4°C before concentrating using a Vivaspin 6 or 20 MWCO 3,000 (Generon) to >10 mg/mL. For ligand complexing, the dilute GF purified protein was incubated overnight at 4°C on a roller with an excess of dry compound. The protein was clarified by spinning at 4.5 Krpm for 10 min at 4°C and concentrated to >10 mg/mL. All protein concentration estimations were carried out by Bradford analysis using BSA standards (Biorad).
Crystallization and Structure Determination. A sample of apo protein was prepared 24 h prior to the ligand complexes and used to obtain crystals for seeding purposes. All crystals were obtained by the sitting drop vapor diffusion method. Apo crystals were grown by mixing 200 nL of protein with 200 nL of a reservoir solution of 0.06 M Divalents, 0.1 M Tris/Bicine pH 8.5, 50% (40% v/v PEG 500 MME, 20% w/v PEG 20K) [Morpheus E9]. Cocrystallization trials with compounds 9 and 20a were setup with 180 nL of protein, 20 nL of apo crystal seed stock, and 200 nL of reservoir solution. All plates were incubated overnight at 4°C prior to manual inspection. Single crystals of compound 9 were obtained from the same reservoir solution as described above for the apo crystals [Morpheus E9]. Single crystals of compound 20a were obtained from a reservoir solution of 0.12 M ethylene glycols, 0.1 M Bicine pH 8.5, 50% (40% v/v PEG 500 MME, 20% w/v PEG 20K) [Morpheus A9].
Single cocrystals of IDH1 R132H 9/20a were flash cooled by immersion in liquid nitrogen prior to X-ray data collection at the Diamond Light Source, Oxford, UK. All data were subsequently scaled and merged with XDS. Structures were determined by molecular replacement with phenix.phaser 33 and models rebuilt and refined through iterative cycles of COOT 34 and phenix.refine. Validations with MolProbity 35 and PDB_REDO 36 were integrated into the iterative rebuild and refinement procedure. Data collection and refinement statistics are presented in Supporting Information Data Table S2 ■ ASSOCIATED CONTENT

* S Supporting Information
The Supporting Information is available free of charge on the ACS Publications website at DOI: 10.1021/acs.jmedchem.6b01320.
Summary of LC−MS methods and solvent gradients; preparative HPLC instrument and solvent gradients; purity data; detailed chemical experimental information for all compounds described; selected 1 H and 13 C NMR spectra; selected LCMS and HRMS spectra; protein purification and crystallographic methods (PDF) Molecular Formula Strings (CSV)

Accession Codes
Atomic coordinates and experimental data for the cocrystal structures of 20a (PDB 5L57) and 9 (PDB 5L58) in complex with mtIDH will be released upon article publication.

Journal of Medicinal Chemistry
Article