Identification of Potent Reverse Indazole Inhibitors for HPK1

Abstract

Hematopoietic progenitor kinase (HPK1), a negative regulator of TCR-mediated T-cell activation, has been recognized as a novel antitumor immunotherapy target. Structural optimization of kinase inhibitor 4 through a systematic two-dimensional diversity screen of pyrazolopyridines led to the identification of potent and selective compounds. Crystallographic studies with HPK1 revealed a favorable water-mediated interaction with Asp155 and a salt bridge to Asp101 with optimized heterocyclic solvent fronts that were critical for enhanced potency and selectivity. Computational studies of model systems revealed differences in torsional profiles that allowed for these beneficial protein–ligand interactions. Further optimization of molecular properties led to identification of potent and selective reverse indazole inhibitor 36 that inhibited phosphorylation of adaptor protein SLP76 in human PBMC and exhibited low clearance with notable bioavailability in in vivo rat studies.

Hematopoietic progenitor kinase 1 (HPK1) is a member of a family of mitogen-activated protein kinase kinase kinase kinase (MAP4K) of Ste20 serine/threonine kinases and is also known as MAP4K1.¹ HPK1 is exclusively expressed in the hematopoietic compartment, with the implication that HPK1 is involved in the regulation of signaling in hematopoietic lineages.² In addition to negative T cell regulation via the T cell receptor (TCR), HPK1 also mediates T-cell suppression through PGE2 signaling.^3,4 As a regulator of MAPK activity in T cells, HPK1 works by destabilization of the SLP76 adapter protein signaling complex via a proteasomal degradation pathway to transmit TCR signals to elicit downstream cellular events.⁵ At the same time, HPK1 may be triggered by various stimuli, such as growth factors, stress, inflammation, and differentiation cues, to transmit signals in different cell types including B-cells, dendritic cells (DC), and natural killer (NK) cells.⁶ Phosphorylation of SLP76 SH2 domain on S376 by HPK1 leads to dissociation of the LAT complex and proteasomal degradation of SLP76.^7,8 This dissociation prevents downstream T-cell activation and proliferation. A recent study has noted that HPK1 kinase activity suppresses the immune functions of CD4⁺ T cells, CD8⁺ T cells, and DCs and inactivation of the HPK1 domain is capable of eliciting antitumor immune responses.^9,10 Mouse genetic experiments suggest a kinase-activity-dependent combination benefit with anti-PD1 in a murine colon adenocarcinoma model. HPK1 kinase dead mice demonstrated improved antitumor response to anti-PD1 therapy, compared with HPK1 wild-type (WT) mice. While certain mouse models showed improved responses with triggered protective immunity of anti-PD-L1, the HPK1 knockdown phenotype alone was able to suppress tumor formation in a select mouse model.^7,11 These experimental findings suggest a small molecule HPK1 inhibitor may be able to promote antitumor immunity as a single agent or combination therapy with anti-PD1 (Tables 1 and 2).^10,12−14

Table 1. Potency and In Vivo PK Profiles of HTS Hits.

^aMeasured at [Tracer] ≈ K_d[Tracer].

^bMeasured at [ATP] ≈ K_m[ATP].

^cSelectivity expressed as the ratio of JAK1 or JAK2 IC₅₀ over HPK1 IC₅₀.

^dMale rat Wistar Han at intravenous dose of 0.1 mg/kg as a solution in DMSO/PEG400/H₂O.

^eMale rat Wistar Han at intravenous dose of 2 mg/kg as a solution in 30% captisol/70% water. CL_p = total plasma clearance; Vd = volume of distribution; MRT = mean residence time.

Table 2. SAR of C(6)-Aryl Analogues.

^aMeasured at [Tracer] ≈ K_d[Tracer].

^bMeasured at [ATP] ≈ K_m[ATP].

^cSelectivity expressed as the ratio of JAK1, JAK2, or CDK2 IC₅₀ over HPK1 IC₅₀.

Despite its discovery two decades ago, there are few selective small molecule inhibitors of HPK1 reported.⁷ Thus, a high-throughput screening (HTS) campaign was initiated to identify potent kinase inhibitor leads. Among the hits, numerous potent indazoles were identified (e.g., 1–2); however, the pharmacokinetic (PK) profiles of indazole 1 and azaindazole 2 were poor. The high rat plasma clearances greater than hepatic clearance suggested potential issues of glucuronidation of the exposed hydrogen bond donor.^15,16 Among the hits, we also found reverse indazoles (RI) 3 and 4, that while greater than 75-fold less potent, showed appreciably lower plasma clearance and represented a better starting point to achieve a desirable DMPK profile.

Based on the promising PK of reverse indazole hits 3 and 4 and the absence of SAR from the screen, we generated an enriched two-dimensional diversity set to develop the N(1)- and C(6)- substitution SAR off the pyrazolopyridine core,¹⁸ from which we discovered para-piperazine phenyl-containing RI 5 that contains the para-methoxy pyridine ring from hit 1. We counterscreened against a small panel of kinases, including JAK1 and JAK2, as they were commonly observed off-targets of the HTS hits. CDK2 was incorporated because of its essential role in cell proliferation. In addition to similarity sequence analyses, linear-regression plot analyses provided clarity as to which kinases could act as surrogates for other off-target kinases that could pose significant toxicity or T-cell function issues. JAK1 correlated well with TYK2; JAK2 correlated well with JAK3; and CDK2 was a good surrogate for both CDK5 and GSK3A.

Reverse indazole 5 showed an excellent improvement in potency by over a log unit versus 4 and, additionally, increased selectivity against JAK1 and JAK2 (83×, 83×). This C(6) pyridyl piece, though, proved to be unselective against CDK2 (4×). We then further developed the C(6) SAR with compounds with a static para-piperazine phenyl N(1) group to improve potency and selectivity. An unsubstituted phenyl group (6) led to a significant loss in potency (820 nM). o-Tolyl group replacement (7) further supported that ortho substituents are tolerated, but nonetheless, there was a 4-fold potency loss from 5 (410 nM). 2-Fluoro-6-methylphenyl (9) yielded a similar level of potency (121 nM) as the initial molecule 5 in a more lipophilic space (cLogD, 2.77). Moreover, this group showed better selectivity against CDK2, suggesting that a 2,6-disubstituted aryl could be a replacement for the methoxypyridine moiety. Given the boost in potency with the fluorine substituent, we kept the fluorine constant while we explored the size and polarity at the other ortho position. Bicyclic methylamine group 10 exhibited weakened potency at 193 nM. Incorporating a methoxy group (11) showed an enhanced potency of 89 nM. Extending this group to an ethoxy (12), however, led to a significant loss in potency (557 nM). Thus, we opted to continue SAR exploration at N(1) with this 2-fluoro-6-methoxy C(6) group.

We next investigated the SAR of the N(1) substituent. Without the piperazine attached 13, a potency loss of greater than 10-fold was observed (1.0 μM, Table 3). A similar effect was observed when the phenyl group was exchanged for a smaller methyl pyrazole 14(19) (1.0 μM). We attributed this loss in potency to reduced nonspecific binding in a lipophilic pocket. Morpholine replacement 15 showed similar activity in vitro as the piperazine (101 nM). The replacement of the basic amine with a polar ether without any potency loss is consistent with a nonspecific van der Waal contact. Altering the trajectory of the piperazine (16) further supported this contact with a 4-fold loss in potency (408 nM).

Table 3. SAR of N(1)-Aryl Fragments.

^aMeasured at [Tracer] ≈ K_d[Tracer].

^bMeasured at [ATP] ≈ K_m[ATP].

^cSelectivity expressed as the ratio of JAK1, JAK2, or CDK2 IC₅₀ over HPK1 IC₅₀.

^dSee ref (19).

Continuing exploration in this region, we focused on the aryl piperazine motif. We replaced the phenyl linker with an ortho-pyridine (17) and found an unexpected 3-fold boost in potency (31 nM, Table 3) versus phenyl piperazine 5. Compound 17 showed encouraging selectivity with JAK1, JAK2 and CDK2 (95×, 61×, 77×, respectively). We were then pleasantly surprised that a meta-piperazine-substituted pyridine (18) further dropped the potency to 2.6 nM. While JAK1 and JAK2 selectivity improved significantly, selectivity against CDK2 was only 44-fold. Moving the nitrogen atom to the 4-position (19) lost potency (60 nM) and subsequently kinase selectivity (JAK2, 61×). Pyrimidine linker 20 regained a bit of potency (6.7 nM) and showed encouraging selectivities against the JAK family (JAK1 sel, 1483×; JAK2 sel, 125×). However, the diminished selectivity against CDK2 (10×) led to continued exploration with the 2-pyridine linker. Removal of the methyl substituent (21) afforded a compound with similar activity to 18. Addition of a chloro unit at the 3-position (22) offered enhancements in both potency (1.8 nM) and selectivities, especially CDK2 selectivity (597×). While the chlorine substituent taught us a valuable lesson in improving kinome selectivity by growth in this vector, the lipophilicity increased significantly (cLogD 3.73).

The unexpected potency boost with inhibitor 18 led us to investigate what protein–ligand interactions were made. Co-crystallization of para-piperazine 11 and meta-piperazine 18 each with the HPK1 protein provided insight into the binding mode of the reverse indazole core. Both ligands show two-point hinge contacts, one of which is the C(3)–H and the other being N(2). Each hinge is about 3.1 Å from the binding pocket (Glu92 and Phe93, Figure 1a,b). We observed the 2-fluoro-6-methoxyphenyl group sits in the lysine region, twisted out of plane with the more lipophilic fluoro group pointing toward the P-loop of the protein. The N(1) aryl substituents of the ligands protrude into the solvent front. One of the main differences between the two ligands is that the piperazine ligand 11 extends farther into the solvent front, while the piperazine of 18 makes an interaction to Asp101 from the altered trajectory.

Figure 1.

Co-crystal of HPK1 enzyme with different RI ligands show two-point hinge contacts with Cys94 and Glu92. (a) Co-crystal of HPK1 with 11 (PDB: 7L24, resolution 2.7 Å). (b) Co-crystal of HPK1 with 18. Aza of core generates water network with Asp155 (PDB: 7L25, resolution 1.9 Å). (c) Co-crystal of HPK1 with 38. Displacement of unfavorable water molecule and Asp155 moves closer to ligand (PDB: 7L26, resolution 2.3 Å).

Upon comprehensive analysis of different meta-substituted aryl linkers through crystallography and modeling, large differences in dihedral angles from the indazole core are observed. On the basis of the crystal structures, the phenyl linker of 11 shows a dihedral angle of 39°, while the 2-pyridyl linker of 18 is only out of plane by 22°. We modeled des-piperazine analogues of 16, 18, and 19 to better understand the effects of different heteroaryl groups (Figure 2). Analysis of the lowest energy conformation of these compounds show that the phenyl linker still had the greatest dihedral angle of 35° (Figure 2a). Replacement with 4-piperidyl has a comparable dihedral angle of 29° (Figure 2c). Moving the nitrogen to the 2-position yields the flattest landscape with no out-of-plane twist (Figure 2b). While the dihedral angle of the N(1)-phenyl relative to the core showed only ∼4° difference between the crystal structure and the calculated lowest energy conformer, a bigger difference was observed when we replaced the phenyl with the 2-pyridyl group. Analysis of the torsional profile of rotation of the 2-pyridyl at the N(1) juncture revealed that the barrier to rotation was only ∼0.5 kcal to reach a dihedral angle of 20°, indicating that while the bioactive conformation is not at the ground state, the energy penalty is low. The caveat of comparing calculations to protein crystallography data is the two different media in which each are achieved. Calculations were performed in vacuum and thus most likely simulate an environment divergent of the active site comprising of the protein and water molecule network.

Figure 2.

(a) Lowest energy conformer of phenyl analogue of 16. Dihedral angles of RI core to phenyl group calculated to be 35°. (b) Lowest energy conformer of 2-pyridyl analogue 18. Dihedral angles of RI core to 2-pyridyl group calculated to be 0°. (c) Lowest energy conformer of 4-pyridyl analogue of 19. Dihedral angles of RI core to 4-pyridyl group calculated to be 29°. Calculated with M06-2X/6-31G**.

Incorporating the piperazines at the 3-position, with respect to N(1), shows that the piperazine could be twisted out of plane more due to the adjacent heteroatoms. However, because of the greater flexibility and number of possible conformations of the piperazine, more impact was accredited to the heteroaryl linking group. Such flexibility may be what allows for the additional interaction to Asp101 that may outweigh any energetic barriers in twisting the pyridine out of plane. In the case of the phenyl analogue, the greater rotation may lead to clashing with the protein or unfavorable torsional angles to create additional salt bridge contacts, highlighting the 2-pyridyl linker as a key inflection point.

With the success of piperazine analogue 18, we focused our attention to exploring substituents at the 3-position with diverse functionalities to achieve high kinome selectivity while maintaining potency (Table 4). Substituting the pyridine with smaller groups such as OMe (23, 103 nM) or NH₂ (24, 518 nM) demonstrated modest inhibition and thus the need for larger substituents to achieve desirable levels of potency. Unlike the para-substituent scenario, morpholine replacement (25) in this setting yielded substandard potency compared with the piperazine (34 vs 2.6 nM). While a nonspecific van der Waals contact may still be present, the defined salt bridge with the N-Me of the piperazine proved to be advantageous for the binding of the ligand. Azetidine substituted RIs 26 and 27, while both bear the second nitrogen for such an interaction, are likely too distant from the Asp101 residue, leading to lower levels of potency (32 and 13 nM, respectively). Larger unsubstituted heterocycles like pyrrolidine (28, 109 nM) and piperidine (31, 40 nM) likewise had poorer potency. Integrating a dimethyl amine group off the pyrrolidine (30) appeared to provide the necessary contact to the protein, leading to an improved biochemical potency of 1.2 nM. The compound proved to be exceptionally selective against the selected off-targets as well (JAK1 3067×, JAK2 393×, CDK2 281×). Attachment of a methylamine group off the piperidine heterocycle 32 provided minimal enhancements in potency (17 nM) and also lost selectivity by greater than 2× across the off-targets, contrasting with its five-membered analogue 30. Carbonyl capping groups off of piperazine groups demonstrated interesting divergency as well. Methyl carbonate 33 illustrated a 3-fold loss in potency compared to methyl urea 34 (32 vs 11 nM, respectively). The selectivity profile of urea 34 displayed up to 5-fold improvement when assessed against the methyl carbonate analogue. However, JAK2 and CDK2 were still <100× selective over HPK1. Transforming the heterocycle to a heterobicycle produced significant advancements. Integration of greater sp3 character to the inspiration point of pyrrolidine 30 led us to discover diazabicycloheptane 35 which showed exquisite potency (1.9 nM) along with high levels of selectivity against the JAK family (JAK1 5187×; JAK2, 599×). Similarly, larger diazabicyclooctane 36 was slightly more potent (1.1 nM) and selective in that regard (JAK1 8961×; JAK2 1453×). Both bicycles had similar levels of selectivity against CDK2 (61×).

Table 4. SAR of Substitution of N(1)-Pyridyl Solvent Front.

^aMeasured at [Tracer] ≈ K_d[Tracer].

^bMeasured at [ATP] ≈ K_m[ATP].

^cSelectivity expressed as the ratio of JAK1, JAK2, or CDK2 IC₅₀ over HPK1 IC₅₀.

While meta-substitution off the aza-RI core led to potent analogues, we continued to search for a diverse set of kinome selective compounds. We returned to studying the crystal structures for further guidance. Upon closer inspection of the crystal structures of these reverse indazole cores, we observed that one of the most noticeable variations in the protein lay in the DFG loop (Figure 1).¹⁷ Though these structures have different packings in their lattice structures, the differences in the DFG loop are attributed to the compound itself. Around this region lies a water molecule that can interact favorably with the 5-aza of the pyrazolopyridine core. With meta-substituted RI 18, Asp155 is within 3.1 Å of this water molecule to generate a water network with the core (Figure 1b). Omitting this nitrogen at the 5-position leads to a significant drop of more than a log unit in potency (37, 1470 nM, Table 5) compared with the aza-analogue (9, 121 nM), underscoring the significance of interacting with Asp155. We leveraged this opportunity by displacing the water molecule with a nitrile at the 5-position (38). The displacement proved to be valuable for potency by 4-fold (28 nM) compared with the parent aza-core (9, 121 nM). Co-crystallization of the enzyme with compound 38 showed that indeed the water molecule is displaced but also that the side-chain of Asp155 has now reoriented itself toward the ligand to create an interaction with the nitrile that is 3.1 Å away (Figure 1c). Hybridization with the solvent front piece of 18 yielded further improvements that landed nitrile analogue (39, 2.0 nM) with similar potency as its aza-counterpart. Most attractive of all, kinome selectivity was enhanced across the board for both compounds. Incorporation of the nitrile provided selectivity against not only the JAK kinases but also CDK2 with both 5-nitrile compounds (CDK2 sel, 38, 357×; 39, 115×; Table 5).

Table 5. SAR of 5-Substitution of Reverse Indazoles.

^aMeasured at [Tracer] ≈ K_d[Tracer].

^bMeasured at [ATP] ≈ K_m[ATP].

^cSelectivity expressed as the ratio of JAK1, JAK2, or CDK2 IC₅₀ over HPK1 IC₅₀.

The leads were further profiled side-by-side in pharmacokinetic studies Wister-Han rats. Similar to our initial HTS hit, aza-reverse indazole 18 had a low total clearance of 9.4 mL/min/kg with a moderate unbound clearance (2171 mL/min/kg) and long half-life (2.3 h) (Table 6, entry 1). However, the 5-nitrile RIs demonstrated high clearance in rats. Nitrile analogue 38 exhibited a high total plasma clearance of 151 mL/min/kg that is most likely due to elimination pathways including but not limited to renal or hepatic pathways (entry 2). Despite having a much lower Cl_p (12 mL/min/kg) and a long t_1/2 (3.1 h), meta-pyridine derivative 39 still suffered from high unbound clearance (19 913 mL/min/kg, entry 3). Moreover, these compounds were significantly more lipophilic compared to its aza-analogues (>3.5 vs average of 2.6) and had unmeasurably low permeability in both cases (<14 × 10^–6 cm/s). To fully capitalize on the selectivity impact of the C(5)-nitrile substitution, a new optimization plan to rebalance the overall polarity would be required.²⁰

Table 6. In Vivo PK Profiles^a and In Vitro Cell Activity.

entry	no.	CLp (mL/min/kg)	Vd (L/kg)	t1/2 (h)	%F	pSLP76 IC50,u (pSLP76 IC50) (μM)d	Papp (10–6 cm/s)	cLogD
1	18	9.4	1.1	2.3	-	0.6 (8.0)	28	2.75
2	38	151	18	2.2	-	nd (>10)	<8	3.51
3	39	12	2.3	3.1	-	nd (>10)	<14	3.78
4	30	4.8	0.3	1.2	-	nd (>10)	30	3.18
5	35b	4.2	0.5	1.8	>100	1.1 (4.3)	34	2.21
6	36b	19	2.3	2.1	67	0.3 (4.0)	26	2.64

^aMale rat Wistar Han at intravenous dose of 0.1 mg/kg as a solution in DMSO/PEG400/H₂O.

^bMale rat Wistar Han at intravenous dose of 0.5 mg/kg and oral dose 0.5 mg/kg as a solution in DMSO/PEG400/H₂O.

^cHuman PBMC were used;²¹ unbound values correct for binding to 10% FBS. Total cell potencies displayed in parentheses.

^dMDCKII cell line used. nd = not determined.

Profiling of leads 35 and 36in vivo led us to discover reasonable terminal half-lives of both compounds (1.8 h, entry 5 and 2.1 h, entry 6, respectively, Table 6). Compound 35 has a favorably low plasma clearance of 4.2 mL/min/kg but is offset by high plasma protein binding, leading to a much higher unbound clearance (2962 mL/min/kg). Compound 36, while higher, still has an attractive plasma clearance of 19 mL/min/kg and a slightly decreased unbound clearance of 1197 mL/min/kg. The unbound volumes of distribution differed by approximately 2-fold less for lead 36 (143 L/kg). While both compounds had adequate half-lives in rat, the bioavailability of 35 was measured to be greater than 100%, which is attributed to rate-limiting absorption. This caveat is reflected in the unanticipated high oral exposure and the low volume of distribution. Diazabicyclooctane-substituted RI 36, with lower exposure in the oral dosing, possessed a more balanced pharmacokinetic profile with a favorable bioavailability of 67%.

We investigated the more potent lead RIs in vitro for effective response in inhibiting phosphorylation of SLP76. HPK1 regulates T cell signaling through phosphorylating SLP76 on serine 376, an adaptor protein that is critical for the signaling pathway of T cell activation. To evaluate this in vitro, we initially created Jurkat WT and three HPK1 KO clones through CRISPR technology to measure HPK1 target engagement.²¹ Because of the mutations and inconsistencies observed with this cell line,²² we opted to employ human PBMC for better translation of T cell responses in the clinic. Application of ELISA technology delivered a bigger and consistent window for the readout of the target engagement assay.²¹

We profiled a set of potent and selective compounds in the SLP76 PBMC assay. meta-Piperazine RI 18 was potent in the cell assay with an unbound IC₅₀ of 0.6 μM (entry 1, Table 6). Profiling our most selective compounds, nitriles 38 and 39, proved to be discouraging with a complete lack of cell activity. To our disappointment, exquisitely selective dimethyl amine substituted pyrrolidine 30 also showed no cell activity (entry 4). The initial hypothesis was that these compounds may be too lipophilic, with cLogD values greater than 3.1 in all three cases, to enter the cell. Given the good permeability of ligand 30 (30 × 10^–6 cm/s), we hypothesized that higher lipophilicity gave rise to nonspecific media binding (media binding of 92.87% in 10% FBS). Bicyclic substituted pyridines 35 and 36 both showed respectable in vitro unbound cell activity at 1.0 and 0.3 μM, respectively. Analogous to the two aforementioned compounds, both 35 and 36 showed adequate levels of permeability in the MDCKII cell line (34 × 10^–6 cm/s, entry 5 and 26 × 10^–6 cm/s, entry 6, respectively). The large cell shift from the biochemical assay is attributed to the low apparent K_m for ATP measured for HPK1 (ATP K_m, 8.9 μM).²¹ Such a low ATP K_m places considerable demand on intrinsic potency to achieve suitable target engagement in cells as well as good selectivity against off-targets kinases having higher ATP K_m values. Considering both cell activity and DMPK properties, compound 36 displayed the best overall profile with sufficient levels of kinome selectivity.

We have described our efforts toward the development of a potent HPK1 inhibitor, starting from leads with appreciable DMPK profiles from our HTS campaign RIs 3 and 4. Comprehensive SAR studies along with guidance from X-ray crystallography helped us to identify potent reverse indazole 36 with significantly improved kinome selectivity and a promising in vivo PK profile. The efforts in this lead identification exercise reconfigured the indazole series to generate more tractable leads with regards to selectivity and PK, for further optimization toward drugging HPK1, which will be the subject of a future communication.

Glossary

ABBREVIATIONS

HPK1

hematopoietic progenitor kinase 1

MAP4K1

mitogen-activated protein kinase kinase kinase kinase 1

TCR

T-cell receptor

SLP76

SH2-domain-containing leukocyte protein of 76 kDa

LAT

linker for activation of T cells

PD-1

programmed cell death protein 1

PD-L1

programmed death-ligand-1

RI

reverse indazole

PBMC

peripheral blood mononuclear cell

DC

dendritic cells

NK

natural killer cells

HTS

high throughput screening

ATP

adenosine triphosphate

DFG

Asp-Phe-Gly

CRISPR

clustered regularly interspaced short palindromic repeats

ELISA

enzyme-linked immunosorbent assay

SAR

structure–activity relationship

PK

pharmacokinetics

DMPK

drug metabolism and pharmacokinetics.

Supporting Information Available

The Supporting Information is available free of charge at https://pubs.acs.org/doi/10.1021/acsmedchemlett.0c00672.

• Biochemical and cell assay protocols; general synthetic procedures; crystallography data with compounds 11, 18, and 38; and kinome wide selectivity data for compounds 35 and 36 (PDF)

Accession Codes

The crystal structure of HPK1 in complex with compound 11, 18, and 38 has been deposited at the with accession code 7L24, 7L25, 7L26, respectively.

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.