Diminishing GSH-Adduct Formation of Tricyclic Diazepine-based Mutant IDH1 Inhibitors

Abstract

Mutant isocitrate dehydrogenase 1 (IDH1) has been identified as an attractive oncology target for which >70% of grade II and III gliomas and ∼10% of acute myeloid leukemia (AML) harbor somatic IDH1 mutations. These mutations confer a neomorphic gain of function, leading to the production of the oncometabolite (R)-2-hydroxyglutarate (2-HG). We identified and developed a potent, selective, and orally bioavailable brain-penetrant tricyclic diazepine scaffold that inhibits mutant IDH1. During the course of in vitro metabolism studies, GSH-adduct metabolites were observed. The hypothesis for GSH-adduct formation was driven by the electron-rich nature of the tricyclic core. Herein, we describe our efforts to reduce the electron-rich nature of the core. Ultimately, a strategy focused on core modifications to block metabolic hot spots coupled with substitution pattern changes (C8 N → C linked) led to the identification of new tricyclic analogues with minimal GSH-adduct formation across species while maintaining an overall balanced profile.

Mutant isocitrate dehydrogenase 1 (IDH1) has been identified as an attractive oncology target prevalent in >70% of low-grade gliomas¹ and secondary glioblastomas (GBMs),² as well as approximately 10% of acute myeloid leukemia (AML).^3,4 The paralogue IDH2 mutation is, however, more prevalent in AML than in glioma.¹⁻⁴ In its mutant form of IDH1, a single amino acid change at Arg132 (e.g., R132H) in the catalytic domain, promotes reduction of α-ketoglutarate (α-KG) to (R)-2-hydroxyglutarate (2-HG).^5,6 2-HG has been characterized as an oncometabolite. It competes α-KG off dioxygenases that use it as a cofactor, thereby disrupting multiple regulatory cellular signaling pathways and ultimately contributing to aberrant cell growth and tumorigenesis.⁷ Inhibiting mutant IDH1 has been clinically validated, with ivosidenib (AG-120)⁸ approved by the FDA for the treatment of relapsed/refractory AML.⁹ A handful of other small molecule mutant IDH1 inhibitors, including IDH1/2 dual-inhibitor AG-881,¹⁰ BAY-1436032,¹¹ GSK321,¹² IDH305,¹³ FT-2102,¹⁴ and others,^15,16 have also shown efficacy in various preclinical and clinical studies, potentially offering new treatment options for different tumor types carrying IDH1 mutations.

Previously, we reported an optimized tool compound MRK-A with high potency, mutant selectivity, and brain-penetrant profile to support target validation studies.¹⁷ It demonstrated robust in vivo tumor PK/PD, efficacy, and dose-dependent survival benefit in a patient-derived orthotropic glioma xenograft model.¹⁷ However, because of its poor pharmacokinetic properties (e.g., high intrinsic clearance (Cl_int), along with pregnane X receptor (PXR) activation and time-dependent inhibition (TDI) liability), co-dosing with the CYP450 inhibitor 1-aminobenzotriazole (ABT)¹⁸ was required in order to achieve high exposure and target engagement in vivo. Leveraging structural information and a customized multiparameter optimization (MPO) model, with focus on improvements in Cl_int, PXR, and TDI, led to the identification of multiple lead compounds within this tricyclic diazepine series.¹⁹ These molecules (represented by 1) demonstrated reduced liability for Cl_int, PXR, and TDI, improved ligand-lipophilicity efficiency (LLE, based on cellular potency in MOG, which expressed exogenous mutant IDH1 R132H protein in the wild-type IDH1 glioma cell line, MOG-G-UVW, through lentiviral transduction), tractable CNS properties (non-P-gp substrates), and robust tumor 2-HG reduction in the BT142 orthotopic glioma model¹⁷ expressing mutant IDH1^R132H without ABT co-dosing (Figure 1),¹⁹ a significant improvement compared to MRK-A.

Figure 1.

Representative molecules of the tricyclic diazepine series. MOG cell-based assay was run using engineered wild-type IDH1 glioma cell line MOG-G-UVW to express mutant IDH1 R132H protein.¹⁷ *Time-dependent inhibition (TDI) of CYP3A4, wherein an IC₅₀ shift >1 (the ratio between IC₅₀ values in the absence vs presence of NADPH) is considered a TDI risk.

Although the overall lipophilicity of MRK-A has been reduced to optimize its pharmaceutical properties, the electron-rich nature of the tricyclic pyridobenzodiazepine core in 1 presented a concern due to the potential for CYP-mediated oxidative metabolism. In addition, multiple N-substitutions on the aromatic rings pose the risk of generating reactive species such as quinones, which can be linked to potential drug-induced liver injury (DILI) findings.²⁰ Indeed, during the course of in vitro metabolism studies with 1, GSH-adducts were observed (Table 1, entry 1), thus raising a potential concern for this series. It was hypothesized that the GSH-adduct formed post CYP-mediated oxidation of the electron-rich tricyclic system. To that end, we examined different ways to reduce overall oxidative metabolism while monitoring the effects that had on the propensity to form GSH-adducts. Herein, we describe our medicinal chemistry efforts aiming to alter the electronics of the core via core modification as well as substitution pattern change (N → C linked), which ultimately led to the identification of new tricyclic analogues with diminished GSH-adduct formation across species while maintaining an overall balanced profile.

Table 1. Nucleophilic Trapping with GSH and ^–CN in both Rat and Human Liver Microsomes^a.

^a*Substrate concentration 10 μM, protein concentration 1 mg/mL, 1 h, temperature 37 °C. Percentages are based on peaks integrated in the extracted ion chromatogram and do not account for likely differences in ionization efficiencies.

As part of the early optimization efforts to improve metabolic stability,²¹ MetID was conducted on 2 in rat and human hepatocytes to understand the metabolite profile of the molecule (Scheme 1, putative structures shown). The major findings after 2 h incubation of 2 at 5 μM concentration with rat hepatocytes included oxidative phase I metabolites on the C8 morpholine (2-i) and peripheral dealkylation products on the N6 substituent (2-ii). In addition, appreciable amounts of GSH-adducts (2-iii) were identified, presumably occurring on the tricyclic core, an observation that was consistent across different species for similar analogues in the series. Although GSH-adduct formation alone is not always associated with DILI, a recent survey of 54 marketed drugs uncovered a high tendency for elevated DILI risk for compounds with GSH-adducts when coupled with a high daily clinical dose (>100 mg).²² Therefore, we focused our efforts toward investigating the relationship between structure and GSH-adduct formation to help mitigate the potential risk for the tricyclic series.

Scheme 1. Major Metabolites of 2 in Rat Hepatocytes.

Before deploying resources on synthetically challenging core modifications, we sought to leverage our previous SAR¹⁹ (selected examples shown in Table 1) at R¹ (C8 N-linked) and R² positions for a deeper assessment of GSH-adduct formation in this series. As shown in Table 1, various morpholine isosteres were tolerated at R¹. For example, ethylene-bridged morpholines (1 and 3) slightly improved cell potency by ∼4-fold compared with 2. Monocarbon bridged morpholine (4), 4,6-spiro azetidine (5), and bridged piperidine (6) showed equal or slightly inferior cell potency. Modifications at R² with tetrahydropyran (1) or trifluoroethylpiperidine (6) were also compatible. Of note, all compounds described throughout this paper are single isomers separated by chiral SFC. Only the most potent isomers were discussed. The absolute configuration was not assigned for ethylene-bridged morpholine R¹ and tetrahydropyran R². Despite the flexibility at R¹ (C8 N-linked) and R² for tuning various properties, modifications at these positions were not effective at changing the GSH-adduct profile. The GSH-adduct formation in rat liver microsomes (RLM) represented up to 27% of total metabolism and was notably higher in human liver microsomes (HLM) for nearly all tested compounds (Table 1). In addition, nucleophilic trapping was also confirmed by cyanide adduct formation with a slightly lower, but appreciable percentage (Table 1). Moreover, we also observed cysteine adducts in vivo after pronase digestion of liver protein pellets (data not shown), further confirming the formation of reactive metabolites.²³

After ruling out the possibility of addressing the GSH-adduct propensity with C8 N-linked R¹ and R² substitution modifications, we envisioned that a change to the core was needed to reduce formation of oxidative metabolites. To prioritize synthesis of compounds with likely the highest impact on GSH-adduct formation, we used MetaSite²⁴ for metabolic hotspot prediction. Consistent with the MetID findings, in silico prediction of metabolic hotspots for 1 with MetaSite suggested C8 morpholine oxidation and N6 dealkylation (Figure 2, top red bars). It also suggested the C3 core position as the next most metabolically labile site (Figure 2, cyan bar). Therefore, we sought to explore C3 substitution on the core to block oxidative metabolism. To enable the C3 SAR, we envisioned that Hartwig’s metaselective late-stage C–H borylation²⁵ of the pyridine could be achieved to set the stage for further C3 modification. Indeed, C–H borylation of 3 exclusively occurred at the C3 position to afford the pinacol boronic ester, which was further converted into small substituents, including halogens, methyl, and sulfone under various conditions (Table 2). Among them, only a fluorine substitution showed comparable cellular potency (18 nM for 7vs 17 nM for 3), which suggested limited space is available at the C3 trajectory within the binding pocket. Consistent with our goal of identifying brain-penetrant inhibitors, we tracked P-gp liability closely throughout our optimization efforts. Notably, C3 F-substituted analogues (including 7) demonstrated consistently lower BA/AB ratios compared to nonsubstituted analogues (Figure 3). However, this same substitution at C3 failed to attenuate formation of GSH-adducts (Table 3, compounds 11 and 12). Similarly, chlorination at C3 afforded no change in GSH-adduct formation (Table 2, compound 8). We also explored effects of substitution at other positions on the tricyclic core, however, no changes were effective at blocking formation of GSH-adducts (Table 3, compounds 13–15). C9 F-substituted analogue (13) was comparable to its nonfluorinated analogue 3 in terms of both potency and amount of GSH-adduct formation. However, fluorination at C7 (14) resulted in reduced potency, increased Cl_int, and increased amounts of GSH-adduct compared with 3. The methyl group of compound 15 was designed to block the potential for iminium formation but was still not effective at changing the GSH-adduct profile. It is worth mentioning that while fluorination on the core generally led to a favorable P-gp profile, modifications at C5 (15) and C7 (14) increased P-gp efflux ratios up to 3.4 (Table 3).

Figure 2.

Site of metabolism prediction of 1 by MetaSite.

Table 2. C3 Structure–Activity Relationship of Tricyclic Inhibitors^a.

compd	R =	IDH1 R132H IC50 (nM)	MOG cell IC50 (nM)	% GSH-adducts (r)b	P-gp BA/AB ratio @ 0.1 μM (h)c
3	H	11.0	17.2	17	3.6
7	F	21.2	18.2	NA	2.9
8	Cl	28.7	95.9	10%	4.6d
9	Me	112.5	456.3	NA	2.2
10	SO2Me	11210	>10000	NA	4.5

^aReagents and conditions: (1) [Ir(OMe)(1,5-COD)]₂, B₂pin₂, 3,4,7,8-tetramethyl-1,10-phenanthroline, 2-methyl-THF, 100 °C microwave irradiation; (2) see Supporting Information for details.

^bSubstrate concentration 10 μM, protein concentration 1 mg/mL, 1 h, temperature 37 °C. Percentages are based on peaks integrated in the extracted ion chromatogram and do not account for likely differences in ionization efficiencies.

^cHuman LLC-MDR1 BA/AB ratio.

^dRat Mdr1 BA/AB ratio.

Figure 3.

P-gp BA/AB ratio analysis of 15 matched pairs with structure difference at C3 only (with or without fluorine substituent). C3-F substituted compounds are shown in green cycles, and C3 unsubstituted analogues are shown in red cycles.

Table 3. Fluorination and Methylation around the Core to Assess the Impact on GSH-Adduct Formation^a.

compd	IDH1 R132H IC50 (nM)	MOG cell IC50 (nM)	Mic Clint,u, mL/min/kg (h)	% GSH-adducts (r)b	P-gp BA/AB ratio @ 0.1 μM (h)
11	11.0	8.1	22	9	1.4
12	8.7	42.4	33	11	2.2
13	35.2	36.2	124	16	2.5
14	57.0	204.7	210	33	3.1
15	25.3	88.6	143	27	3.4

^aFluorine was installed either through de novo synthesis of fluorinated core (11 and 12) or through late-stage direct fluorination of the core and separation (13 and 14).

^bSubstrate concentration 10 μM, protein concentration 1 mg/mL, 1 h, temperature 37 °C. Percentages are based on peaks integrated in the extracted ion chromatogram and do not account for likely differences in ionization efficiencies.

Next, we explored aza-core modifications to block reactive metabolite formation. However, in many cases, this modification resulted in a drastic loss of cellular potency and/or significant increase in clearance (Table 4). Compounds 16 and 17 with submicromolar cellular potency and reasonable clearance profile were followed-up with a metabolism assessment. Unfortunately, both failed to diminish GSH-adduct formation. Of note, aza-core modification resulted in substantially increased P-gp liability (Table 4, P-gp BA/AB ratio 8.1 for 16 and 6.6 for 17, respectively) compared with 4 (P-gp BA/AB ratio 3.7), which could be attributed to an increased number of HBA and an elevated PSA (87 Ų for 16, 85 Ų for 17, and 73 Ų for 4, respectively).

Table 4. Aza-core Modifications.

compd	IDH1 R132H IC50 (nM)	MOG cell IC50 (nM)	Mic Clint,u, mL/min/kg (h)	% GSH-adducts (r)a	P-gp BA/AB ratio @ 0.1 μM (h)
4	15.9	60.6	34	10	3.7
16	196.7	810.4	177	17	8.1
17	116.0	261.3	64	26	6.6
18	1387	5015	n.d.b	n.d.	n.d.

^aSubstrate concentration 10 μM, protein concentration 1 mg/mL, 1 h, temperature 37 °C. Percentages are based on peaks integrated in the extracted ion chromatogram and do not account for likely differences in ionization efficiencies.

^bNot determined.

To further focus our exploration of the relationship between structure and GSH-adduct formation, we investigated various truncated analogues. Interestingly, common intermediate C8–Br 19 showed only 4% of GSH-adduct formation in RLM (Figure 4), suggesting that a non-N-linked C8 modification could be a strategy to reduce GSH-adduct formation. Leveraging judicious SAR at C8 to maintain potency and balance physicochemical properties, we designed and synthesized several C8 carbon-substituted analogues that to our delight showed improved stability with diminished GSH-adduct formation in RLM (Table 5). Specifically, a C-linked pyrazole (20) was well tolerated in terms of potency and effective at reducing GSH-adduct formation (1.9% in RLM). Remarkably, Csp³-linked tetrahydropyran (21) and dioxane (22 and 23) substitutions at C8 were also very effective at blocking GSH-adduct formation (<1% in RLM). Importantly, dioxane analogue 23 showed a combination of excellent cellular potency (89 nM) and low unbound Cl_int(h) (34 mL/min/kg). Collectively, these data suggest that utilizing a C8 carbon substituent may reduce GSH-adduct formation by blocking formation of putative para-quinone reactive metabolite.

Figure 4.

Ar–Br intermediate 19 showed low GSH-adduct formation.

Table 5. C8 C-Linked Analogues with Diminished GSH-Adducts.

compd	IDH1 R132H IC50 (nM)	MOG cell IC50 (nM)	Mic Clint,u, mL/min/kg (h)	% GSH-adducts (r/d/h)a	P-gp BA/AB ratio @ 0.1 μM (h)
20	4.9	18.3	51	1.9/NA/NA	1.2
21	22.4	146.2	92	<1/NA/NA	1.4
22b	180.8	580.9	20	<1/NA/NA	3.7
23b	41.8	88.5	34	<1/7/NA	1.1
24b	24.4	56.9	50	2/12/7	0.9
25b	32.5	53.6	58	<1/1/1	1.2

^aSubstrate concentration 10 μM, protein concentration 1 mg/mL, 1 h, temperature 37 °C. Percentages are based on peaks integrated in the extracted ion chromatogram and do not account for likely differences in ionization efficiencies.

^bA diastereomeric mixture of these compounds was generated by coupling an enantiomerically pure N6 acid to a racemic core followed by chiral SFC separation. Both diastereomers were tested and only the more potent isomer is shown.

Encouraged by the improvement in GSH-adduct formation in RLM with C8 C-linked analogues, coupled with the overall balanced profile of 23 in terms of potency, intrinsic clearance, and P-gp efflux (Table 5), we decided to evaluate 23 for metabolism in higher species. As we learned from previous in vitro profiling of C8-morpholine analogues, the GSH-adduct levels are often more significant in higher species (Table 1). To that end, 23 was tested in dog liver microsome (DLM) and found to have a GSH-adduct percentage of 7%. While F-substitution at C3 or C9 was unable to reduce the GSH-adduct formation in the C8 N-linked analogues, it was generally tolerated in terms of potency and offered a favorable P-gp profile. We therefore wanted to explore this SAR in the context of C8 C-linked analogues. Thus, both C3 (24) and C9 (25) fluorine-substituted analogues were tested across species. Although GSH-adduct formation of C3-F analogue 24 remained low (2%) in RLM, no improvement was found in higher species (12% in DLM and 7% in HLM). On the other hand, the C9-F analogue 25 is much more stable in vitro with a remarkably low (≤1%) GSH-adduct in all species (Table 5, entry 6). Presumably, the left-hand aromatic ring of the core is responsible for the formation of reactive metabolite and therefore the C9-F is more effective than the more remote C3-F substituent.

To further validate the correlation between GSH-adduct formation and oxidative metabolism for this series, we measured their oxidation half-wave potential (E_1/2) with a coulometric electrochemistry²⁶ assay. This technique has been used to probe oxidative drug transformations to complement traditional metabolic profiling and chemical trapping experiments. It has the advantage of ultrahigh sensitivity and high throughput. On the basis of an assessment of known DILI positive marketed compounds, those with lower half-wave potential tended to have higher oxidation liabilities and risk toward adduct formation (for example, lapatanib, acetominophen, and trogliazone).²⁷ In this study of IDH1 tricyclic inhibitors, we found an excellent correlation between the measured half-wave potential (E_1/2) and the level of GSH trapping products. As shown in Figure 5, the E_1/2 for C8 C-linked analogues (green circles) is on average about two times higher than that of C8 N-linked analogues (red rhombus). This implies that the C-linked analogues have a lower propensity to form oxidative reactive intermediates (precursors to form GSH-adducts). These results are consistent with the low throughput GSH-trapping data from rat liver microsomes. It also explains why core fluorination on C8 N-substituted analogues had little impact on the level of GSH-adducts because their oxidative potential remained in a similar range as nonfluorinated analogues (E_1/2 ranging from 230 to 350 mV). This technique provided an important avenue to assess whether compounds could easily oxidize and thus contribute to the formation of reactive species.

Figure 5.

Relationship between oxidation half-wave potential (E_1/2) measured by coulometric electrochemistry and %GSH-adducts in RLM. Red squares, N-linked compounds including core modifications (F-substitutions and aza-cores) [higher risk]; green circles, C-linked analogues [lower risk]. See Supporting Information, Table S2 for details.

Finally, in vivo pharmacokinetic profiling of 25 was conducted in both rat and dog following IV or PO administration (Table 6). Following IV administration, 25 exhibited low to moderate clearance (69 mL/min/kg in rat and 26 mL/min/kg in dog) with moderate terminal half-life (1.1 h for rat and 6 h for dog, respectively). The volume of distribution was 3.7 L/kg in rat and 5.1 L/kg in dog, respectively. Excellent oral bioavailability of 25 in both rat (100%) and dog (77%) reflected a positive impact of balanced physicochemical properties on absorption and first-pass extraction. With HPLC-LogD_7.4 of 2.67, compound 25 exhibited high solubility across a range of pH values (143, 154, and 123 μM @ pHs 2, 6.5 and 7, respectively). In addition, high passive permeability (mean P_app 38.7 × 10^–6 cm/s) coupled with low P-gp efflux ratio (human LLC-MDR1 BA/AB 1.2) suggests excellent CNS properties.

Table 6. In Vivo PK Parameters of Compound 25.

	rat	dog
IV/PO dose (mg/kg)	0.5/1	2/10
t1/2 (h)	1.1	6.0
Vd (L/kg)	3.7	5.1
Cl (mL/min/kg)	69	26
F (%)	100	77

In conclusion, GSH-adduct evaluation and de-risking is a critical component of drug discovery, especially in lead optimization while progressing molecules to preclinical models or beyond. Tricyclic diazepine IDH1^R132H inhibitors showed a high level of GSH-adduct formation in their metabolism profile. By focusing on strategies to reduce the electron richness of the tricyclic core, we demonstrated that GSH-adduct formation can be significantly diminished without compromising potency, PK, and/or P-gp profile. In addition, good correlation between oxidation half-wave potential and the level of GSH-adduct formation provides a valuable risk assessment and de-risking approach for future compound designs. Finally, the strategic introduction of a F-substituent and C8 alkyl group further reduced the levels of GSH-adduct in higher species and led to the discovery of the new tricyclic lead molecule 25 with negligible GSH-adduct formation in all species (r/d/h) while maintaining an overall balanced profile.

Supporting Information Available

The Supporting Information is available free of charge at https://pubs.acs.org/doi/10.1021/acsmedchemlett.2c00089.

• Synthesis, characterization, and general experimental procedures (PDF)

Author Contributions

All authors have given approval to the final version of the manuscript.

The authors declare no competing financial interest.