Identification of Novel, Potent, and Orally Available GCN2 Inhibitors with Type I Half Binding Mode

Abstract

General control nonderepressible 2 (GCN2) is a master regulator kinase of amino acid homeostasis and important for cancer survival in the tumor microenvironment under amino acid depletion. We initiated studies aiming at the discovery of novel GCN2 inhibitors as first-in-class antitumor agents and conducted modification of the substructure of sulfonamide derivatives with expected type I half binding on GCN2. Our synthetic strategy mainly corresponding to the αC-helix allosteric pocket of GCN2 led to significant enhancement in potency and a good pharmacokinetic profile in mice. In addition, compound 6d, which showed slow dissociation in binding on GCN2, demonstrated antiproliferative activity in combination with the asparagine-depleting agent asparaginase in an acute lymphoblastic leukemia (ALL) cell line, and it also displayed suppression of GCN2 pathway activation with asparaginase treatment in the ALL cell line and mouse xenograft model.

Adaptation to cellular stress is one of the most fundamental features of cancer cells.¹ Dysregulation of the integrated stress response (ISR) has been implicated with various types of diseases including cancer.² Pharmacological modulation of the ISR is expected to be a novel approach to treat cancer.³⁻⁵ General control nonderepressible 2 (GCN2) is a serine/threonine-protein kinase, one of eukaryotic initiation factor 2α kinase that is a master regulator in the ISR.⁶ GCN2 is activated by binding to uncharged tRNA (tRNA) and regulates amino acid homeostasis.^7,8 Several reports indicate that GCN2 is important for cancer survival in the tumor microenvironment with amino acid depletion.⁹⁻¹¹ However, development of potent, selective, and orally available inhibitors targeting the GCN2 pathway has been limited, and the potential of GCN2 inhibitors as cancer therapeutic agents has not been reported. Therefore, we initiated an investigation to devise a series of novel, potent, and orally available GCN2 inhibitors endowed with high kinase selectivity as first-in-class antitumor agents. During an in-house high-throughput screening campaign, there were no promising compounds and most of the compounds showed weak to moderate GCN2 inhibitory activity with promiscuity in terms of kinase selectivity. Thus, we surveyed reported compounds possessing inhibitory activity or binding effects on GCN2 for embarking our medicinal chemistry program. As shown in Figure 1, four distinct chemical series have been disclosed to date. Triazolopyrimidine derivatives represented by 1a are filed in the patent as inhibitors of GCN2.¹² Similar to this series, we have reported 1b with potent enzymatic and cellular GCN2 inhibitory activities, and the in vitro biology of GCN2 inhibition was explored using this agent.¹³ However, further animal testing was suspended due to its poor pharmacokinetic (PK) profile. Compound 2 (RAF-265) is a B-RAF inhibitor and advanced into clinical trials for the treatment of metastatic melanoma.¹⁴ This compound was reported to bind to GCN2¹⁵ with the expected DFG-out binding mode (type II) that is observed in binding to B-Raf.¹⁴ In 2015, a series of pyrazolones including 3 as protein kinase R-like endoplasmic reticulum kinase (PERK) inhibitors has been disclosed, and some of these derivatives showed GCN2 inhibitory activities as an off-target effect.¹⁶ The representative compound was reported to bind to the inactive DFG-out conformation of PERK with packing an allosteric binding pocket close to αC-helix by its 2-phenyl substituent of the pyrazolone (type II half). Most recently, sulfonamides including 4 have been reported, which demonstrated a binding effect not only on B-Raf as a primary target but on GCN2 and assumed to bind to B-Raf with the activation loop folded in (DFG-in) occupying a cavity adjacent to αC-helix in their computational analyses (type I half).¹⁷ Our in-house GCN2 enzymatic assay¹⁸ revealed that these compounds 1–4 showed moderate to potent inhibitory activity (IC₅₀ values are shown in Figure 1), and among these with characteristic differences in both chemical scaffolds and binding modes, the expected type I half inhibitor 4 would be highly attractive as our footholds. As disclosed in the Supporting Information, this agent possessed high selectivity for B-Raf, GCN2, and ZAK.¹⁷ On the other hand, other chemical series were deprioritized on our program due to insufficient target specificity and/or undesirable kinase inhibition regarding toxicity concerns in in-house assay. As far as we know, kinase inhibitors with type I half binding mode are rare compared to typical type I and II binders, which led us to one possible assumption that the induced conformational change around the αC-helix with DFG-in folding is available only in the limited kinases, and type I half inhibitor would realize high target specificity. Therefore, we initiated our medicinal chemistry efforts from compound 4 aiming type I half GCN2 inhibitor with enhanced potency and excellent selectivity over off-target kinases, especially in B-Raf. In addition, improvement of PK profile and aqueous solubility of this series to conduct in vivo validation study is mentioned in this report.

Figure 1.

Reported compounds possessing GCN2 inhibitory activities.

There have been no cocrystal structures of type I half inhibitor with GCN2 reported. In order to develop our synthetic strategy by a structure-based drug design (SBDD) approach, we constructed a GCN2 homology model by combining the reported cocrystal structure of vemurafenib (type I half B-Raf inhibitor)¹⁹/B-Raf with those of type II GCN2 inhibitor 5(20)/GCN2 that was obtained and analyzed by Takeda (PDB: 6N3L). Utilizing these ligand–protein complex structures, a docking model of 4 with GCN2 was successfully developed (left image of Figure 2A), which indicated a type I half binding mode of 4 to GCN2 and was informative to design new analogues. In this model, the pyridin-3-yl sulfonamide moiety of 4 is located in a lipophilic cavity around αC-helix of GCN2, and relatively larger space exists around this moiety, compared to vemurafenib/B-Raf (purple vs gray in the right image of Figure 2A). Therefore, incorporation of a substituent on this right-hand aromatic portion fleshing out the allosteric pocket was put in a main strategic focus of our synthetic efforts to enhance GCN2 inhibitory activity through gaining additional interaction with the enzyme and selectivity against overall kinases, especially in B-Raf. In addition, optimization of the p-fluorophenyl 1H-pyrazolo[3,4-b]pyridine portion as a hinge binding part, which would extend to a solvent region, was planned with the expectation that this could contribute to improvement of PK and/or physicochemical properties of a molecule with maintaining potencies (Figure 2B).

Figure 2.

Docking study of 4 with GCN2 homology model. (A) Left: overview of modeling of 4/GCN2 homology model. Right: overlay of a solvent accessible protein surface for the binding site around αC-helix of 4/GCN2 (purple mesh) and vemurafenib/B-Raf (gray, PDB: 3OG7). (B) Illustration of the drug design which led to compounds 6a–g described in this article.

Scheme 1 outlined the general synthetic procedure for sulfonamides 6. Condensation of substituted aromatic sulfonyl chlorides 8 with 3-ethynyl-2,4-difluoroaniline (7)¹⁷ yielded intermediates 9. Then Sonogashira coupling reaction with halogenated heteroaromatic reagents (10) and following treatments such as reduction, deprotection, and substitution reactions were conducted to provide the target compounds 6. As an alternative, the intermediate 11 was synthesized by Sonogashira coupling reaction of 7 and 10, and compounds 6 were obtained by sulfonamidation of 11 with 8. The detailed procedures are described in the Supporting Information.

Scheme 1. General Synthetic Procedure.

Reagents and conditions: (a) 8, pyridine; (b) 10, Pd(II) catalyst, Cs₂CO₃, DMSO; (c) additional reactions such as reduction and substitution.

The biological activities and aqueous solubilities of sulfonamide derivatives are summarized in Table 1. The reported compound 4 showed moderate GCN2 inhibitory activities confirmed by in-house enzymatic and cellular assays (IC₅₀ = 720 and 3300 nM, respectively),¹⁸ which should be enhanced to render a chemical tool to probe in vivo pharmacological effects of GCN2 inhibition. In addition, the aqueous solubility of 4 is low (<0.14 μg/mL at pH6.8), which would result from the large number of the aromatic rings composing the molecule. According to the synthetic strategy mentioned in Figure 2, we quickly checked that removal of the para-fluorophenyl portion from the hinge-binding pyrazolopyridine (ring A) does not affect the GCN2 inhibitory activity (data not shown), and the introduction of the meta-bromo atom into the right-hand pyridine ring (6a) demonstrated a >16-fold increase of potency in the biochemical assay (IC₅₀ = 43 nM) from 4, which suggested that the substituent on the pyridine (ring B) would contribute an additional lipophilic interaction with GCN2. According to this result, the contribution of substituents on ring B was exhaustively investigated, which revealed that installment of a small and lipophilic substituent such as methyl, methoxy, and halogen at the ortho or meta-position resulted in enhancement of potency (data not shown). Particularly, compound 6b possessing an ortho, meta-dichloro phenyl ring exhibited substantial enhancement of both enzymatic and cellular GCN2 inhibitory activities (IC₅₀ = 3.7 and 190 nM, respectively) probably due to gaining further lipophilic interaction with GCN2. The aqueous solubilities of these analogues were still insufficient; however, the introduction of a hydroxymethyl group into the meta-position of ring B (6c) was found effective to make a molecule more water-soluble. Notably, 6c showed a slight increase of potencies from 6b, which suggested potential hydrogen bonding between the hydroxymethyl group and GCN2. Regarding the hinge-binding moiety, replacement of pyrazolopyridine with similar bivalent 2-aminopyrimidine (6d) was investigated and resulted in potent cellular activity with a nanomolar level of IC₅₀ value (9.3 nM) with expected improvement of aqueous solubility through decrease of the number of aromatic rings. Instead of an ortho, meta-disubstituted phenyl group of the ring B, introduction of an ortho, meta-disubstituted pyridine was permitted regarding potency (6e). Finally, optimization of the substituent on the pyrimidine-2-amine moiety, which is assumed to extend to the solvent region in our docking model, was conducted for adjustment of physicochemical properties, which led to 6f and 6g with maintaining potent GCN2 inhibitory activities and improved aqueous solubilities.

Table 1. Biological Activities and Solubilities of 4 and 6a–g.

^aValues in parentheses indicate 95% confidence interval.

^bInhibition of human GCN2 enzymatic activity.

^cSuppression of ATF4 induction in U2OS cells with amino acid-free medium.

^dKinetic solubility was measured at pH 6.8.

To assess whether our compounds suppress GCN2-mediated eIF2α phosphorylation, the effects of 4 and 6c–e on eIF2α phosphorylation were evaluated in U2OS cells during amino acid starvation (Table S1). Consistent with the cellular activities (suppression of ATF4) shown in Table 1, 6c–e showed inhibition of eIF2α phosphorylation. Our further analysis revealed that GCN2 inhibitory activities in the enzymatic assay of our sulfonamide derivatives showed a good correlation with those in the cellular assays whereas a set of PERK selective inhibitors prepared in house did not result in a clear correlation (Figure S3 and S4). These results proved the validity of our GCN2 cellular assays.

The PK and binding kinetics of 6d–g were further investigated (Table 2). Compared to 4, favorable PK profiles of 6d–f in mice were observed. Since the microsomal stabilities and cellular permeabilities of 4 and 6d–f are high (data not shown), the improvement of aqueous solubility in 6d–f (Table 1) would be projected to their higher bioavailabilities than 4. In addition, profiling of these compounds regarding target residence time was conducted, and it was noteworthy that 6d and 6g possessing a hydroxymethyl group at the meta-position of the right-hand phenyl ring (ring B) exhibited much slower dissociation properties (T_1/2 = >1440 and >720 min, respectively) compared to 6e and 6f, which possesses pyridine without a hydroxymethyl group as the right-hand ring (T_1/2 = 381 and 236 min, respectively).

Table 2. Pharmacokinetic Parameters and Kinetics Data of 4 and 6d–g.

	PKa
compd.	Cmaxb(ng/mL)	CLc(mg/h/kg)	AUCpod(ng/mL*h)	MRTpoe(h)	Ff(%)	fupg	GCN2 dilution assayhT1/2 (min.)
4	45	801	93	1.8	7.0	NTi	NTi
6d	1201	313	2588	1.7	79.9	0.03	>1440
6e	2469	48	13389	3.2	63.5	0.04	381
6f	1372	150	3578	1.9	51.9	<0.01	236
6g	124	835	163	1.6	13.5	<0.01	>720

^aCassette dosing test in C57BL/6J mice (1.0 mg/kg, po and 0.1 mg/kg, iv, 0–8 h).

^bC_max: maximum concentration.

^cCL: clearance.

^dAUCpo: area under the curve (po).

^eMRTpo: mean residence time (po).

^fBioavailability.

^gFraction unbound in plasma.

^hAssay protocol is shown in the Supporting Information.

ⁱNot tested.

With respect to target specificity, 6e and 6f were screened against a panel of 468 kinases in a binding assay to determine its selectivity (KINOMEscan from DiscoveRx Corp.).²¹ At a concentration of 1000 nM, both did not bind to any of the 468 kinases tested with >99.5% binding affinity (<0.5% control), except GCN2 (Figure S5). Unlike 4, these compounds did not show any binding affinities to B-Raf at the same concentration (100% control). The exquisite selectivity profile verified our drug design aiming type I half GCN2 inhibitor.

In order to further examine selectivity, 6d and 6e were investigated regarding inhibitory activities against eIF2α kinase PERK, which was not included in the above panel. As illustrated in Table S2, both compounds showed potent PERK inhibitory activity in enzymatic assay (6d: IC₅₀ = 0.26 nM and 6e: IC₅₀ = 3.8 nM). The ATP concentration used in the PERK enzyme assay (4 μM; 4 × Km concentration) was extremely lower than those in GCN2 enzyme assay (190 μM; Km concentration) and cellular ATP levels (a few mM). Kinase inhibitory activities of ATP competitive compounds decrease in an ATP concentration-dependent manner,²² which would result in difficulty in the interpretation of selectivity. Therefore, a cellular PERK assay was conducted by induction of CHOP expression in U2OS cells during thapsigargin-mediated ER stress in order to examine selectivity against PERK under physiological conditions. As a result, 6d and 6e showed moderate to weak PERK inhibitory activities in cells (IC₅₀ = 230 and 9000 nM, respectively, Table S2). Selectivity for GCN2 against PERK under physiological conditions was evaluated from inhibitory activities of PERK in cells and that of GCN2 in cellular ATF4 assay and was calculated to be 25-fold (6d) and 450-fold (6e), respectively (Table S2). We also confirmed that the cellular PERK inhibition of both compounds was weak in comparison with the reference PERK inhibitor 12(23) (Figure S6). Based on these results, we conclude that 6d possesses unique slow-binding profile in GCN2 inhibition with some cellular PERK inhibitory activity and 6e is a potent GCN2 inhibitor endowed with high kinase selectivity. Both compounds were selected for further pharmacological studies. The studies using 6e have been reported,¹³ and the biological tests for 6d are described in the latter part of this article.

X-ray cocrystal structures of 6d and 6e with GCN2 were successfully resolved, which validated our SBDD and revealed that both bind to GCN2 as DFG-in binding mode while burying the allosteric pocket adjacent to αC-helix (Figure 3). The compounds sit in the ATP-binding site of the kinase domain, and two-pairs of hydrogen bonds were formed by the 2-aminopyrimidine portion of ligands with the backbone of hinge residue Cys805. The oxygen atom of the sulfonamide moiety forms a hydrogen bond with the backbone amide Phe867 from the conserved DFG-motif. In addition, the interaction between the deprotonated nitrogen atom and Asp866 was assumed, which has been reported in the case of other type I half B-Raf inhibitors including vemurafenib.²⁴ The right-hand rings of compounds pack between gatekeeper residue Met802 and other hydrophobic residues (Leu640 and Val637) around αC-helix. Both cocrystal structures showed almost identical alignment, but it is noteworthy that the dichloro hydroxymethyl phenyl ring of 6d is stabilized via the hydrogen bond formed with the backbone carbonyl oxygen of Leu640 of αC-helix, which is the clear contrast between the binding feature of 6d and 6e. As mentioned in Table 2, compound 6d demonstrated slower dissociation profile, and this additional hydrogen bond with Leu640 of GCN2 could be a key factor resulting in its characteristic binding kinetics.

Figure 3.

Superposition of X-ray cocrystal structures of 6d (purple) and 6e (green) with GCN2. Key amino acid residues are depicted as sticks (gray), and hydrogen bonds are represented as dotted yellow lines.

To examine the impact of GCN2 inhibition on cancer cell proliferation, acute lymphoblastic leukemia (ALL) CCRF-CEM cells were treated with 6d in the presence of asparagine-depleting agent asparaginase. CCRF-CEM cells are insensitive to asparaginase, because asparaginase treatment induces asparagine synthetase through GCN2/ATF4 pathway activation.¹³ As shown in Figure S7, treatment with 6d greatly sensitized CCRF-CEM cells to asparaginase. We also examined whether the antiproliferative activity of 6d was mediated by GCN2 inhibition. Importantly, the moderate antiproliferative effects achieved by combining asparaginase and 6d treatment were observed in GCN2-wildtype (WT) mouse embryonic fibroblast (MEF) cells but not in GCN2-knockout (KO) MEF (Figure S8). These results suggest that our GCN2 inhibitors could be useful to treat cancer in combination with asparaginase.

The GCN2 inhibition of 6d in asparaginase-treated CCRF-CEM cells was further investigated. CCRF-CEM cells were treated with 6d and asparaginase, and phosphorylated GCN2 (p-GCN2), phosphorylated eIF2α (p-eIF2α), and ATF4 were measured by Western blot analysis. As shown in Figure S9, 6d demonstrated suppression on p-GCN2, p-eIF2α, and ATF4 activated by asparaginase, which is consistent with our previous result using 6e.¹³ We also confirmed that ATF4 induction observed in asparaginase-treated GCN2-WT MEF cells is not occurred in GCN2-KO MEF cells and clearly GCN2-dependent.¹³ Other groups convincingly demonstrated in mammalian cells that amino acid starvation-mediated eIF2α phosphorylation and ATF4/CHOP expression depend on GCN2.²⁵ These results support that suppression on GCN2 pathway activation observed in Figure S9 results from GCN2 inhibition by 6d.

Finally, 6d was evaluated in the CCRF-CEM xenograft model. In the previous report, 6e suppresses GCN2 pathway activation to the basal level following pretreatment with asparaginase when orally dosed at 10 mg/kg to the same mice model.¹³ The xenograft mice were pretreated with 1,000 U/kg asparaginase for 24 h, and then 6d was orally administered at doses of 0.3, 1, and 3 mg/kg, respectively. The effects of 6d on the GCN2 pathway were assessed by autophosphorylation of GCN2 and ATF4 levels 8 h after treatment of the inhibitor. As shown in Figure 4, treatment with 6d at 3 mg/kg suppressed both self-phosphorylation of GCN2 and the downstream effector ATF4 to the basal level following pretreatment with asparaginase, and this phenomenon is also observed with highly selective compound 6e.¹³ This result suggested that 6d functionally inhibits GCN2 and the downstream effector in vivo, even though the involvement of PERK inhibition with 6d is not completely excluded. Interestingly, the minimum effective dose of 6d (3 mg/kg) was lower than 6e (10 mg/kg)¹³ despite its shorter duration of drug concentration than 6e (Table 2). We hypothesize that the slower dissociation binding profile of 6d might contribute to the long-lasting pharmacological effect.

Figure 4.

Suppression of GCN2 pathway activation in CCRF-CEM xenograft tumor at 8 h after oral administration of 6d (0.3, 1, and 3 mg/kg, qd) with pretreatment of asparaginase (1000 U/kg, qd, ip).

In summary, potent and orally available GCN2 inhibitors as potentially first-in-class antitumor agents were successfully identified through our strategy aiming at type I half inhibitors. Detailed biological evaluation of 6e endowed with high kinase selectivity has been separately reported by our team.¹³ Compound 6d with a slow dissociation binding profile demonstrated suppression of GCN2 pathway activation with asparaginase treatment in the CCRF-CEM cells and mouse xenograft model. Notably, 6d displayed more potent in vivo target inhibition than 6e. Further chemical optimization of this compound will be warranted to achieve exquisite target specificity like 6e. Our findings validate the potential of GCN2 inhibition as a new approach for treating cancer.

Glossary

Abbreviations

GCN2

general control nonderepressible 2

ALL

acute lymphocytic leukemia

ISR

integrated stress response

tRNA

tRNA

PERK

protein kinase R-like endoplasmic reticulum kinase

SBDD

structure-based drug design

PK

pharmacokinetics

ATF4

activating transcription factor 4

p-GCN2

phosphorylated GCN2

PD

pharmacodynamics

Cs₂CO₃

cesium carbonate

DMSO

dimethyl sulfoxide

Supporting Information Available

The Supporting Information is available free of charge on the ACS Publications website at DOI: 10.1021/acsmedchemlett.9b00400.

• Data regarding characterization of compounds in this letter, detailed experimental information on the biological activity assays, and other crystallographic experiments (PDF)

Accession Codes

X-ray cocrystal structural data have been deposited in the Protein Data Bank with accession codes 6N3L, 6N3N, 6N3O.

Author Present Address

^§ (O.K.) Axcelead Drug Discovery Partners, Inc., 26-1, Muraoka-Higashi 2-chome, Fujisawa, Kanagawa, Japan.

Author Present Address

^∥ (D.R.C.) PeptiDream, Inc., 3-25-23, Tonomachi, Kawasaki-ku, Kawasaki, Kanagawa, Japan.

Author Contributions

The manuscript was written through contributions of all authors. All authors have given approval to the final version of the manuscript.

The authors declare no competing financial interest.