Discovery of Spiro-azaindoline Inhibitors of Hematopoietic Progenitor Kinase 1 (HPK1)

Abstract

Hematopoietic progenitor kinase 1 (HPK1) is implicated as a negative regulator of T-cell receptor-induced T-cell activation. Studies using HPK1 kinase-dead knock-in animals have demonstrated the loss of HPK1 kinase activity resulted in an increase in T-cell function and tumor growth inhibition in glioma models. Herein, we describe the discovery of a series of small molecule inhibitors of HPK1. Using a structure-based drug design approach, the kinase selectivity of the molecules was significantly improved by inducing and stabilizing an unusual P-loop folded binding mode. The metabolic liabilities of the initial 7-azaindole high-throughput screening hit were mitigated by addressing a key metabolic soft spot along with physicochemical property-based optimization. The resulting spiro-azaindoline HPK1 inhibitors demonstrated improved in vitro ADME properties and the ability to induce cytokine production in primary human T-cells.

In contrast to traditional anticancer therapies where the goal is to target pathways that drive tumor proliferation or survival, cancer immunotherapies aim to leverage the patient’s own immune system to combat tumors.¹ Over the past several years, a number of antibody inhibitors of immune checkpoints have found considerable clinical success by enhancing cancer patients’ immune T-cell activity. The survival benefits engendered by those approved antibody inhibitors have been impressive; however, the response rate to these agents remains modest. Hence, there is tremendous interest in the identification of additional agents to complement the approved immunomodulators.²

Hematopoietic progenitor kinase 1 (HPK1) is a member of the Ste20 serine/threonine family of kinases that serves as a negative regulator of T-cell receptor (TCR) induced T-cell activation.³ Studies using HPK1 kinase-dead knock-in mouse models have demonstrated that HPK1 kinase activity limits TCR signaling and cytokine production.⁴ In preclinical syngeneic models, loss of HPK1 kinase function was found to suppress tumor growth. Given the compelling evidence, HPK1 has been proposed to be a promising cancer immunotherapy target.⁵ As such, the goal of our program was to identify a selective small molecule inhibitor of HPK1.

A high-throughput screen using the kinase domain of HPK1 yielded compound 1, a literature Abl inhibitor⁶ (Abl IC₅₀ < 0.51 nM), as a promising hit with respect to potency for HPK1 (K_i = 0.4 nM) and lipophilic ligand efficiency⁷ (LLE = 6.3). In addition to activity against Abl, 1 also showed inhibitory activity against LCK (IC₅₀ = 24 nM, 55-fold selectivity over HPK1). Since Abl⁸ and LCK⁹ are essential components of cell survival and TCR signaling, respectively, improving selectivity against these two off-targets was a major focus of our optimization effort.

To improve selectivity against Abl, we first explored the difference in the ATP binding site gatekeeper residue between HPK1 (Met) and Abl (Thr).¹⁰ Due to the shorter Thr side chain in Abl, modeling suggested that Abl would suffer from a greater loss of favorable van der Waals interactions if the size of the C3 substituent of the azaindole motif were reduced (R₁ in Table 1). Gratifyingly, conversion of the anisole substituent (1, Abl IC₅₀ < 0.51 nM) to the smaller ethyl group (2, Abl IC₅₀ = 22 nM) resulted in considerable improvement in selectivity against Abl while improving the LLE against HPK1. The increase in LLE suggested that the small alkyl substituent was better suited for HPK1 binding. While compound 2 exhibited only modest Abl selectivity, this promising result suggested that optimizing other vectors of the ligand using a structure-based approach may further enhance the selectivity against Abl.

Table 1. Optimization of 7-Azaindoles.

compd	R1	R2	cLogP	HPK1 Ki (nM)a	LLEb	Abl selectivityc	LCK selectivityd
1	2-MeOPh	H	3.1	0.4	6.3	<1×	55×
2	Et	H	2.1	2.1	6.6	10×	237×
3	H	H	1.5	130	5.4		66×
4	Et	Me	2.4	0.4	7.0	58×	808×
5	Et	Et	2.8	0.2	6.9		2556×
6	Et	Cl	2.6	0.5	6.7		501×

^aAll apparent K_i values represent arithmetic means of at least two determinations using the HPK1 Lantha binding biochemical assay.

^bLLE, lipophilic ligand efficiency based on HPK1 K_i.

^cSelectivity expressed as the ratio of Abl K_i over HPK1 K_i.

^dSelectivity expressed as the ratio of LCK K_i over HPK1 K_i.

An X-ray structure of 3 bound to HPK1 revealed a binding mode that was atypical among kinase structures (Figure 1A). While the 7-azaindole motif engaged the hinge region of the kinase in a canonical manner, the P-loop of the kinase adopted an unusual “folded” conformation. In the “folded P-loop” conformation, the side chain of Tyr28 was positioned to engage the ligand through hydrophobic interactions.¹¹ Furthermore, the main-chain peptide bond N–H of Gly24 in the P-loop is directed toward the ligand and engaged the amide carbonyl of the ligand via a hydrogen bond.

Figure 1.

(A) X-ray structure of 3 bound in HPK1 (PDB: 7R9L) and (B) X-ray structure of G1858¹² bound in HPK1 (PDB: 7R9N). Predicted hydrogen bond interactions are shown as dashed lines. Distances from N/O to hydrogen are shown.

In contrast, in the more usual “extended P-loop” conformation (as exemplified by G1858;¹²Figure 1B), both structural features (Gly24 N–H and Tyr28 side chain) would not be available to interact with the azaindole ligands. In the “extended” conformation, the N–H of Gly24 forms a hydrogen bond with Val31 within the P-loop β-sheet, while the side chain of Tyr28 is situated in the DFG region of the binding pocket, away from the hinge region.¹²

The “folded” P-loop conformation is not commonly observed among kinases that have been studied.¹¹ The ability of HPK1 to adopt the folded conformation was attributed to the unusually high number of glycine residues within its P-loop resulting in a high degree of flexibility, a feature that is absent in the LCK sequence. Literature examples suggested that stabilizing or engaging the “folded” P-loop conformation may have the potential to improve selectivity.^11,13 Hence, we hypothesized that further engagement of Gly24 and Tyr28 in the folded P-loop conformation of HPK1 may result in improvement in both potency and selectivity against our key off-target LCK and the wider kinome.

To further optimize the interactions between the 7-azaindole ligands and Tyr28 in the folded P-loop, small lipophilic substituents at the C4 position (R₂ in Table 1) were evaluated. Compared to the parent molecule 2 (R₂ = H), installation of a methyl (4), ethyl (5), or chloro (6) group afforded gains in potency for HPK1 and improved selectivity against LCK and Abl (compound 4). It is also noteworthy that the alkyl substitutions at C4 also improved the LLE of the molecules, suggesting that productive interactions were established. Encouraged by the potency and selectivity of the 3-ethyl-7-azaindole series, our attention turned to improvement of the ADME-PK properties.

As shown in Table 2, compound 2 showed high turnover in human liver microsomes and time-dependent inhibitory activity against CYP3A4. Attempts to attenuate oxidative metabolism, such as deletion of the amino group (7) or installation of an electron-withdrawing group (8 and 9), did not improve in vitro metabolic stability (as measured by lowering of human liver microsome clearance) or address time-dependent inhibition of CYP3A4. Interestingly, introduction of a fluoro substituent improved both the potency for HPK1 and selectivity against LCK. Based on modeling, we hypothesized that the fluorine atom occupies a lipophilic pocket and stabilizes the nonplanar conformation of the ligand such that its aniline N–H could effectively hydrogen-bond with Asp101 (see Figure 1A). It is also worthy to note that, compared to analogues with primary or secondary amides, analogues with a tertiary amide (such as 2) demonstrated superior potency and selectivity against LCK (data not shown). We hypothesized that the tertiary amide preferred to reside in an out-of-plane conformation relative to the aniline such that the carbonyl group was well-positioned to hydrogen bond with the backbone N–H of Gly24. Given the lack of impact on metabolism with modifications within the aniline portion of the molecule, the focus of our optimization efforts then shifted to the 7-azaindole motif.

Table 2. Optimization of Solvent-Exposed Region.

^aAll apparent K_i values represent arithmetic means of at least two determinations using the HPK1 Lantha binding biochemical assay.

^bSelectivity expressed as the ratio of LCK K_i over HPK1 K_i.

^cLM = Liver microsomes clearance measured in mL/min/kg, H = human, R = rat, M = mouse.

^dCYP3A4 time-dependent inhibition (TDI) assay performed using testosterone as a probe substrate.

Despite the presence of the 7-aza substitution, the electron-rich 5-membered ring within the 7-azaindole motif was implicated as the primary cause of oxidative metabolism and time-dependent inhibitory activity against CYP3A4.¹⁴ Examination of the X-ray cocrystal structure of 3 bound to HPK1 suggested that further substitutions at the 2- and 3-positions of the 7-azaindole should be tolerated and could provide opportunities to address the metabolic issues. Installation of a chlorine atom (10) at the 2-position to sterically and inductively deactivate the azaindole toward oxidation failed to improve human liver microsome (HLM) stability and time-dependent inhibition of CYP3A4 (Table 3).

Table 3. Optimization of Hinge-Binding Azaindole Group.

^aAll apparent K_i values represent arithmetic means of at least two determinations using the HPK1 Lantha binding biochemical assay.

^bLLE, lipophilic ligand efficiency based on HPK1 K_i.

^cSelectivity expressed as the ratio of LCK K_i over HPK1 K_i.

^dHLM = Human liver microsomes predicted clearance (mL/min/kg).

^eCYP3A4 time-dependent inhibition (TDI) assay performed using testosterone as a probe substrate.

Further modeling suggested saturation of the azaindole to the corresponding azaindoline system should be tolerated and would also provide alternate exit vectors for structure-based design. 3-Spiro-cyclopropyl-indoline 11 demonstrated encouraging potency against HPK1 while reducing human liver microsome turnover and CYP inhibition. Expansion of the spiro-cyclopropyl ring to the cyclobutyl (12) or cyclopentyl groups (13) were also well-tolerated and offered alternative design exit vectors. In general, the spiro-indoline analogues showed a promising combination of good metabolic stability (as measured by HLM), minimal time-dependent inhibition of CYP3A4, acceptable potency, and selectivity that warranted further optimization.

As predicted by the SAR from the 7-azaindole series, installation of a small lipophilic group at the 4-position of the spiro-azaindoline (14–16) to enhance the van der Waals interactions with Tyr28 led to improved selectivity against LCK (14–16) and potency for HPK1 (15, 16). However, possibly due to the increase in lipophilicity, the C4 substitution resulted in the erosion of metabolic stability and undesired CYP3A4 inhibition. This observation suggested that a reduction in general lipophilicity would be required to maintain a balance between potency and metabolic stability.

An X-ray structure of 14 bound to HPK1 (Figure 2) revealed a binding mode that was similar to that of the 7-azaindole HPK1 inhibitors, where the indoline motif formed two hydrogen bonding interactions with the hinge residues of HPK1. The carboxyl-aniline group extended toward solvent where the aniline N–H formed a hydrogen bonding interaction with Asp101. The phenolic side chain of Tyr28 engaged the inhibitor via hydrophobic interactions. The spirocyclic cyclopropyl ring occupied a pocket flanked by the Met90 gatekeeper residue, Leu144, Val31, and a water network in the direction of the catalytic Lys46. A close examination of this region of the binding pocket suggested that installation of polar groups on the spirocyclic system should not only reduce the overall lipophilicity of the molecules, but also engage polar functionalities in the vicinity of Lys46.

Figure 2.

X-ray structures of (A) 14 (PDB: 7R9P) and (B) 17 (PDB: 7R9T) bound to HPK1. Red circles represent water molecules. Predicted hydrogen bonds: Interactions are shown as dashed lines; distances from N/O to hydrogen are shown. Pink arrow represents the distance between the carbons labeled with asterisks.

While the cyclobutyl analogue 15 was superior with respect to LLE, modeling suggested that the larger cyclopentyl (16) or cyclohexyl templates should provide vectors and distances that are more suited to target Lys46. To test this hypothesis, a series of alcohol- and nitrile-functionalized analogues of 16 were designed to indirectly interact with Lys46 via the water network. Hydroxylated analogues 17–19 demonstrated HPK1 activity comparable to the parent 16 while improving in vitro metabolic stability significantly (Table 4). The nitrile analogues 20 and 21 maintained potency without improving HLM stability. An X-ray structure of 17 bound to HPK1 (Figure 2B) showed that the binding pose was identical to that of 14 with the hydroxyl group forming several hydrogen bonding interactions with the water network adjacent to Lys46. This observation was consistent with the increase in LLE and suggested that productive interactions were formed between the polar functionalities and the binding site.

Table 4. Optimization of Spiro-indolines.

^aRelative stereochemistry as drawn. Absolute stereochemistry as noted.

^bAll apparent K_i values represent arithmetic means of at least two determinations using the HPK1 Lantha binding biochemical assay.

^cLLE, lipophilic ligand efficiency based on HPK1 K_i.

^dSelectivity expressed as the ratio of LCK K_i over HPK1 K_i.

^eCellular assay that measures the direct phosphorylation of SLP76 by HPK1 in Jurkat cells.

^fHLM = Human liver microsome predicted clearance (mL/min/kg).

^gMDCK permeability assay (apical-to-basolateral).

^hApparent HPK1 K_i values for compounds 28 and 29 represent arithmetic means of two determinations using the HPK1 HTRF enzymatic assay.

To further improve HPK1 potency, we opted to design extended substituents that may directly engage Lys46 or its neighboring polar functionalities. Introduction of a primary amide (22) resulted in a gain in LLE, which suggested productive hydrogen bonding interactions were made. While demonstrating good HLM stability (HLM = 7.4 mL/min/kg), 22 only showed modest permeability (MDCK apical-to-basolateral P_app = 0.2 × 10^–6 cm s^–1), possibly due to the high number of hydrogen bond donors (five) and high topological polar surface area (114 Ų). We anticipated that this might limit oral bioavailability; thus, we turned our attention to improving the permeability of 22.

Our strategy to improve permeability involved (a) reducing the number of hydrogen bond donors and (b) sterically masking the polar functionalities. Conversion of the primary amide to a secondary amide (compound 23) failed to maintain potency. Installing a methyl group adjacent to the amide (24) maintained potency and metabolic stability but did not improve permeability. Replacing the amide with other bioisosteres such as a triazole (27) resulted in loss of potency. Interestingly, expansion to the cyclic lactams (28 and 29) improved permeability while maintaining potency. When compared to 23 (where the N-methyl-amide likely existed preferentially in the s-trans conformation), the amide moiety in 28 and 29 was restricted to the s-cis conformation that may have maintained productive H-bonding interactions with the protein. We hypothesized that the increase in lipophilicity and steric shielding of the polar amide drove the improvement in permeability. However, the increase in lipophilicity also eroded metabolic stability.

Despite its low permeability as measured by the MDCK assay, we were encouraged by the potency and metabolic stability in HLMs of 24. Hence, its single enantiomers were isolated to give compounds 25 and 26. The two single enantiomers were tested in a cellular assay that measures the direct phosphorylation of SLP76 by HPK1 in Jurkat cells. Gratifyingly, compound 25 showed strong inhibitory effects in the cellular assay despite its modest MDCK permeability. Compound 25 also showed excellent selectivity against key off-targets such as LCK as well as in a broader kinase panel (5/47 kinases inhibited >70% at 0.1 μM; Abl inhibited at <35% at 0.1 μM). To evaluate its stimulatory effect on immune response, human pan T-cells were treated with 25 for 4 h followed by stimulation with αCD3 and αCD28. The resulting increase in cytokine production (as measured by IL2 concentration) was then determined. As shown in Figure 3, treatment with 25 resulted in dose-dependent augmentation of IL2 production as one would expect from inhibition of HPK1 kinase activity. Further ADME-PK profiling showed that 25 addressed the two major liabilities of the original 7-azaindole HPK1 inhibitors: 25 demonstrated good metabolic stability in human hepatocytes (human hepatocyte clearance for 25 = 9.2 mL/min/kg; human hepatocyte clearance for 2 = 15 mL/min/kg) and minimal inhibitory activity against CYP3A4 (TDI IC₅₀ > 10 μM).

Figure 3.

Dose–response curve measuring IL2 levels in human pan T-cell cultures after treatment with 25 followed by stimulation with αCD3 and αCD28 (calculated EC₅₀ = 1.56 ± 0.10 μM). Activity displayed as percent increase in IL2 concentration relative to stimulated/untreated controls. Data points represent results from two replicates.

After an iv bolus dose in mouse, 25 showed a blood clearance of 57 mL/min/kg (Table 5). The good in vitro–in vivo correlation for mouse suggested that human liver microsomes and hepatocytes could be a reasonable predictor of metabolic stability in human. Despite its modest MDCK permeability, 25 showed 13% oral bioavailability in mouse after oral dosing using a suspension formulation (25 mg/kg; MCT suspension).

Table 5. Pharmacokinetic Profile of 25.

compd	LM H/R/Ma,d	Hep H/R/Mb,d	mouse iv CL, Vssc	mouse F%c
25	6.9/8.7/38	9.5/18/33	57, 1.9	13%

^aLM = Liver microsome predicted clearance (mL/min/kg), H = human, R = rat, M = mouse.

^bHep = Hepatocyte clearance measured in mL/min/kg, H = human, R = rat, M = mouse.

^cMouse PK: C57BL/6, 1 mg/kg iv dose or 25 mg/kg po dose, blood clearance measured in mL/min/kg, V_ss = volume of distribution (L/kg).

^dHLM and Hep clearance values represent arithmetic means of two determinations.

Herein, we described the lead optimization of a series of HPK1 inhibitors where improvement in selectivity was achieved by optimizing interactions with the induced, noncanonical P-loop conformation in HPK1. The key transformation in the lead optimization process to address metabolic liabilities included the conversion of the 7-azaindole to a series of spiro-indoline motifs as the hinge-binding element. Results from optimization of 25 to further improve its ADME properties will be disclosed in due course.

Chemistry

As outlined in Scheme 1, the synthesis of 25 began with SEM protection of the commercially available 5-bromo-4-chloro-7-azaindole 30. The azaindole motif was then converted to the 3,3-dibromoxindole 31 via oxidative bromination with NBS. Zinc reduction then yielded oxindole 32. The spirocyclic cyclopentenyl ring was then installed via bis-alkylation using allyl bromide, followed by a ring-closing metathesis to give cyclopentene 34. The oxindole was then reduced and N-protected to give indoline 36. Hydroboration/oxidation then yielded a diastereomeric mixture of 37 and 38. The alcohol 37 was then advanced to the corresponding nitrile 39 by activating the alcohol via mesylation and cyanide displacement. α-Alkylation of the nitrile using methyl iodide then afforded a diastereomeric mixture of 40a and 40b (dr 4:1). Separation of the diastereomers was achieved via silica gel flash chromatography to yield 40a (racemic). Deprotection of the PMB group with concomitant hydrolysis using TFA yielded both the nitrile 41 (result of incomplete hydrolysis) and primary amide 42. Compound 42 was then coupled with the boronate 43(15) to give (±)-25. The single enantiomers were separated using chiral supercritical fluid chromatography (SFC) to give 25 and 26. Other analogues described herein were prepared using a similar synthetic scheme. (Experimental procedures can be found in the Supporting Information.)

Scheme 1. Synthesis of Compound 25.

Glossary

Abbreviations

A

apical

Abl

Abl kinase

ADME

absorption, distribution, metabolism, elimination

B

basolateral

CL

clearance

CYP

cytochrome P450

F

oral bioavailability

Hep

hepatocytes

HLM

human liver microsomes

HPK1

Hematopoietic progenitor kinase 1

HTS

high-throughput screening

IL2

interleukin-2

iv

intravenous

LCK

lymphocyte-specific protein tyrosine kinase

LLE

ligand lipophilic efficiency

Liver Mic

liver microsomes assay

MCT

0.5% methylcellulose with 0.2% Tween in water

MH

mouse hepatocytes

MLM

mouse liver microsomes

NBS

N-bromosuccinimide

P_app

apparent permeability

PK

pharmacokinetics

PMB

para-methoxybenzyl

po

oral administration

SEM

trimethylsilylethoxymethyl

SFC

supercritical fluid chromatography

SLP76

SH2-domain-containing leukocyte protein of 76 kDa

T

testosterone, TCR, T-cell receptor

TDI

time-dependent inhibition

TFA

trifluoroacetic acid

V_ss

volume of distribution at steady state

Supporting Information Available

The Supporting Information is available free of charge at https://pubs.acs.org/doi/10.1021/acsmedchemlett.1c00473.

• Experimental procedures, compound characterization, detailed kinase selectivity panel data for compound 25, crystallographic statistics for protein cocrystal structures of 3, G1858, 14 and 17 (PDF)

The authors declare no competing financial interest.