Optimization of Versatile Oxindoles as Selective PI3Kδ Inhibitors

Abstract

The 3,3-disubstituted oxindole moiety is a versatile and rigid three-dimensionally shaped scaffold. When engineered with a purine hinge-binding core, exceptionally selective PI3Kδ kinase inhibitors were discovered by exploiting small differences in isoform selectivity pockets. Crystal structures of early lead 2f bound to PI3Kδ and PI3Kα helped rationalize the high selectivity observed with 2f. By attenuating the lypophilicity and metabolic liabilities of an oxindole moiety, we improved the preclinical species PK and solubility and reduced adenosine uptake activity. The excellent potency and kinome selectivity of 7-azaoxindole 4d and spirooxindole 5d, together with a low plasma clearance and good half-life in rat and dog, supported a low once-daily predicted human dose.

Phosphoinositide 3-kinase delta (PI3Kδ) phosphorylates phosphatidylinositol, a messenger which serves as a membrane anchor for several protein kinases (e.g., AKT) to regulate a broad range of cellular activities. PI3Kδ knockout or kinase dead mice are viable and fertile and have resistance to induced inflammatory disease as well as tumor growth in syngeneic cancer models. Hence, inhibitors of PI3Kδ have been proposed for the treatment of asthma, rheumatoid arthritis, COPD, activated PI3Kδ syndrome as well as cancer.^1,2 Two structurally related inhibitors, idelalisib and duvelisib, have gained approval for the treatment of hematologic malignancies driven by AKT.^3,4 Unfortunately, their use requires monitoring for possible colitis and pneumonitis,^5,6 which are believed to be on-target effects from decreased regulatory and increased cytotoxic T-lymphocyte activities.^7,8 There are numerous other inhibitors and clinical candidates that have been reviewed in the literature.⁹⁻¹¹

As part of our PI3Kδ inhibitor program, we characterized potent and selective PI3Kδ inhibitor 1a (Table 1) which features a 6,8,9-trisubstituted purine hinge binding motif, PI3Kδ IC₅₀ of 6.3 nM, and 71× selectivity versus PI3Kα.¹² PI3Kα is a ubiquitously expressed off-target kinase with key roles in cell proliferation and metabolism, and genetic knockouts are embryonically lethal. Phosphorylation of AKT is mediated by PI3Kδ in Ramos Burkitt′s lymphoma cells, with pAKT IC₅₀ of 63 nM for 1a. Unfortunately, we identified inhibition of adenosine uptake (AdU) in HeLa cells with similar potency (AdU IC₅₀ of 0.072 μM). Inhibition of adenosine uptake can be associated with hemodynamic changes and a potential liability for development.¹³⁻¹⁹ Hence, we sought chemical matter with higher selectivity for PI3Kδ versus PI3Kα and higher cell selectivity for pAKT versus AdU. Oxindole inhibitors were inactive versus isoforms PI3Kβ and PI3Kγ.

Table 1. Oxindole Selectivity Motif.

We employed computationally empowered structure-based scaffold-hopping strategies to create novel classes of PI3Kδ inhibitors. From successive virtual screens coupled with the generation of focused libraries of modular scaffolds, we discovered and disclosed a versatile oxindole selectivity motif that was successfully paired with multiple hinge-binding cores to provide potent leads with high isoform selectivity.²⁰ Two previously described examples from this effort are pyrazolopyrimidines 1b and 1c, in which an oxindole serves a scaffolding function, orienting the benzyl or cyclopropylmethyl motifs toward Trp760 and Met752.^21,22 As preliminary leads, compounds 1b and 1c had promising PI3Kδ potency and selectivity versus PI3Kα and were inactive versus PI3Kβ and PI3Kγ; however, they lacked selectivity versus adenosine uptake in HeLa cells. We explored the oxindole selectivity motif in combination with numerous hinge-binding cores. Herein, we describe the coupling of an oxindole with the 8-aryl-9-alkyl purine hinge-binding motif of 1a (see Table 2).

Table 2. Early Oxindole Purines.

Even without an affinity motif attached, purines bearing the benzyl and cyclopropylmethyl oxindoles (2a and 2b) were reasonably potent leads with good ligand efficiency. With a 1-ethyl-5-methylpyrazole attached at purine-C(8) (e.g., compound 2c), PI3Kδ potency readily matched that observed for 1a. However, adenosine uptake (AdU) activity was again an issue with HeLa AdU IC₅₀ 0.017 μM, within range of the pAKT cell potency. While the 1-ethyl-5-methylpyrazole ring at purine-C(8) provided good potency, this group was typically prone to metabolism by N-dealkylation resulting in inferior PK profiles. A 2-methylpyrimidine affinity motif was more metabolically stable across several series; however, pyrimidines 2d and 2e were less potent as PI3Kδ inhibitors. We noted improved selectivity versus PI3Kα, while AdU inhibition (AdU IC₅₀ 0.20, 0.40 μM) was still within range of pAKT cell potency. As with all oxindoles, these compounds were inactive versus PI3Kβ and PI3Kγ.

Benzyl oxindole 2f was more potent than 2e, and significantly more selective versus PI3Kα (540×), suggesting that improved interactions with Trp760 and Met752 are important for PI3K family selectivity. Consistent with this, groups smaller that cyclopropylmethyl gave weaker potency and selectivity. For 2f we noted a 10× selectivity in pAKT cell activity versus AdU cell activity. Interestingly, benzyl isoindolinone 2g was only 2-fold less potent than oxindole 2f; representing new future opportunities. Truncation of 2f to a phenyl oxindole 2h deteriorated the cell selectivity over adenosine uptake. Over the course of exploring the oxindole C(3) substitution SAR we noted that introducing polarity can lead to decreased adenosine uptake activity. For example, oxetane 2i had only modest AdU activity (IC₅₀ 2.9 μM), although it was less potent on target as well. However, isomeric oxetane 2j was equipotent to benzyl oxindole 2f (pAKT IC₅₀ 110 nM) while maintaining a 10× selectivity in pAKT versus adenosine uptake (AdU IC₅₀ 1.3 μM). While 2f and 2j represented improvements in selectivity versus adenosine uptake, we also noted moderate–high plasma clearance and short half-lives in rat and dog (profile of 2j in Wistar-Han Rat: Cl/Cl_int 27/99 mL/min/kg; Vd/Vd_u 2.4/6.0 L/kg; t_1/2 1.0 h; in Beagle Dog: Cl/Cl_int 24/250 mL/min/kg; Vd/Vd_u 1.7/3.5 L/kg; t_1/2 1.0 h).

To confirm the predicted bioactive S enantiomer and to aid in the design of new inhibitors, we obtained a cocrystal structure of 2f bound to PI3Kδ, which was resolved to 2.42 Å (Figure 1A and 1B). The inhibitor was bound to the catalytic kinase domain and anchored by hinge interactions between the purine N(1) and C(2)-H to Val828-NH and CO of PI3Kδ, respectively, and filled pockets typical of many type-I PI3K inhibitors targeting the active form.^23,24 The purine N(9)-ethyl group partially occupied a hydrophobic pocket lined by the gatekeeper Ile825 as well as Tyr813 of PI3Kδ. The positioning of the gatekeeper Ile825 (conserved in the PI3K family) above the plane of the inhibitor, part of the roof of the active site and creating space for Tyr813, is a distinct feature of the lipid kinases versus protein kinases.²⁵

Figure 1.

(A and B) Crystal structure of 2f bound to the PI3Kδ kinase domain (2.42 Å). (C) Crystal structure of 2f bound to the PI3Kα kinase domain (2.12 Å). These structures have been deposited in the protein data bank (PDB ID 7JIS for 2f bound to PI3Kδ, 7JIU for 2f bound to PI3Kα).

The purine C(8)-pyrimidine heterocycle occupied the polar pocket lined by Lys779, Asp787, Asp911, and Ile910 of PI3Kδ with a possible direct or water-mediated contact to Tyr813, a pocket commonly referred to as the affinity pocket for PI3K inhibitors. The purine core was also sandwiched by an edge-to-face interaction with the indole of Trp760 and a hydrophobic interaction with Met900 of PI3Kδ.

The oxindole stereochemistry was assigned S and largely served as scaffolding to orient the benzyl group toward the specificity, or selectivity, pocket lined by residues Trp760 and Met752. The oxindole lactam also contacts a water molecule. The potency preference between enantiomers was at least 100× throughout the SAR campaign. This unique binding mode was distinct from that observed for the idelalisib structural class, in which a larger specificity pocket is created between Trp760 and Met752 as the putative basis for isoform selectivity.²⁶ It is proposed that this interaction is critical for isoform selectivity, and inhibitors in this class exploit differences in the dynamics of the G-loop-equivalent between isoforms.

Next, we obtained a structure of 2f bound to the kinase domain of PI3Kα (Figure 1C) to better understand the 540-fold selectivity for PI3Kδ. The active sites are remarkably similar between PI3Kδ and PI3Kα, with one key difference being the replacement of Thr750 for the larger Arg770. Clashing with the larger Arg770 residue appears to be contributing to the observed selectivity.²⁷

Encouraged by the reduced AdU activity with the increased polarity of 2i and 2j, we then explored N-linked oxindole purines (Table 3). We began with 3a, an N-linked analog of 2e. Compound 3a offered good potency (PI3Kδ IC₅₀ 2.1 nM) and PI3Kα selectivity and modest AdU activity (AdU IC₅₀ 1.7 μM) but had high intrinsic clearance in rat²⁸ leading to a short half-life (t_1/2 0.6 h), suggesting that additional optimization was required. Given the impact of oxetanes on AdU activity with 2i and 2j, we continued SAR with cyclic ethers. Aminotetrahydrofuran 3b, for example, was promising with good potency (pAKT IC₅₀ 70 nM) and devoid of AdU activity (AdU IC₅₀ > 10 μM). Plasma clearance was moderate in rat and high in dog, leading to moderate half-lives of 1.6 h.

Table 3. 3-Amino Oxindole Purines.

^aLLC-PK1 cell permeability model.

^bDosed 0.5 mg/kg iv, 1 mg/kg po in 20/60/20-DMSO/PEG400/H₂O.

Overall, we did not observe a good correlation between microsome or hepatocyte turnover versus intrinsic plasma clearance in rat, nor did we observe any significant parent excretion in rat. We speculated that perhaps some of the metabolism may not be well captured by our in vitro systems. Nevertheless, MetID experiments with early leads in this series generally indicated oxindole C(3)-substituent metabolism as a key metabolic liability, and so we focused on the C(3) amine group to address the plasma clearance.

We prepared fluoroalkylamine oxindoles 3c and 3d, to slow putative metabolism of the pendant amine. Selectivity versus PI3Kα was excellent, and cell selectivity versus AdU good (3c pAKT IC₅₀ 72 nM, AdU IC₅₀ 2.3 μM). Intrinsic clearances in rat were low (Cl_int 89, 79 mL/min/kg) and coupled to reasonable unbound volumes gave good half-lives in rat (t_1/2 1.6, 1.9 h). Compound 3c had good bioavailability in rat (F 58%). Unfortunately, the dog PK profiles were less favorable for 3c and 3d, with high plasma clearances, moderate half-lives (t_1/2 1.4, 1.1 h), and lower bioavailability (F 25% for 3c). We explored cyclic amine-substituted oxindoles, with four-membered azetidine oxindole purines being more potent. For example, 3e offered good potency (pAKT IC₅₀ 100 nM) and good selectivity (AdU IC₅₀ 5.7 μM). However, cyclic amine oxindoles generally had higher intrinsic clearances in rat and dog (260, 350 mL/min/kg for 3e) than acyclic amine analogs, leading to short-moderate half-lives (t_1/2 0.6, 1.8 h for 3e). The SAR that we developed with this collection of 3-amino oxindole purines provided new avenues to achieve potency with improved selectivity over PI3Kα and especially versus adenosine uptake. The PK profiles were promising (e.g., 3c) but still suboptimal.

To improve the PK profiles of leads such as 3c, we sought additional means to lower lipophilicity as a means to improve metabolic stability while maintaining the unbound volume. We discovered that the 7-azaoxindole selectivity motif was well-tolerated and the increased polarity further improves selectivity versus adenosine uptake (Table 4). Cyclobutyl amine azaoxindole 4a was nearly 3-fold less potent than parent analog 3a; however, adenosine uptake activity was significantly less (AdU IC₅₀ 8.2 μM), as was the clearance in rat (Cl/Cl_int 9.3/16 mL/min/kg) and greatly improved rat half-life (t_1/2 3.0 h). The 3-aminotetrahydrofuran 7-azaoxindoles 4b and 4c were more potent and had exceptional selectivity over PI3Kα (1600×, 860×), and the margin between cell pAKT potency and AdU inhibition was good (25×, >50×). The rat intrinsic clearance for 4c was low (Cl/Cl_int 38, 90 mL/min/kg) together with a good half-life (t_1/2 3 h). In dog, both 4b and 4c had low clearance; however, the larger unbound volume for 4c contributed to an improved half-life (t_1/2 3.4 h). The bioavailability for 4c (F 33 and 91% in rat and dog, respectively) was good despite the moderate Papp (5.7 × 10^–6 cm/s). We also noted that azaoxindoles were stable to microsomes and hepatocytes.

Table 4. 3-Amino 7-Azaoxindole Purines.

^aLLC-PK1 cell permeability model.

^bDosed 0.5 mg/kg iv, 1 mg/kg po in 20/60/20-DMSO/PEG400/H₂O.

Following the promising potency but suboptimal PK profile of fluoroisobutylamine oxindole 3c, we prepared the 7-azaoxindole analog 4d. Analog 4d retained potency of 3c, but with diminished adenosine uptake activity (AdU IC₅₀ 5.4 μM). The azaoxindole analog also offered lower rat plasma clearance leading to a better rat half-life (t_1/2 2.5 h) and good bioavailability (F 61%). In dog, we observed a dramatic improvement. Dog plasma clearance was low for 4d (Cl 2.4 mL/min/kg) leading to an excellent half-life (t_1/2 9.3 h) and bioavailability (F 100%) in dog. Similarly, the 7-azaoxindole azetidine analog 4e also offered diminished adenosine uptake activity (AdU IC₅₀ > 10 μM). A lower plasma clearance in rat gave a good rat half-life (t_1/2 2.8 h) for 4e. In dog, the PK profile again improved in the azaoxindole series, with a 3-fold drop in plasma clearance. However, the unbound volume was lower than that for 3e, resulting in a suboptimal half-life in dog (t_1/2 2.0 h).

Over the course of our studies we considered cyclizing the 3,3′-substituents of the oxindole, forming a spiro oxindole. Generally, direct spirocyclic versions of previous lead oxindoles were potent and had good selectivity versus PI3Kα and were inactive versus PI3Kβ and PI3Kγ (Table 5). However, they also shared many of the same liabilities. For example, spiro oxindole 5a had potency comparable to close analog 3a, but it too had modest adenosine uptake activity (AdU IC₅₀ 1.5 μM), as well as high clearance in rat and dog (Cl 76, 32 mL/min/kg).

Table 5. 3,3′-Spiro Oxindole Purines.

^aLLC-PK1 cell permeability model.

^bDosed 0.5 mg/kg iv, 1 mg/kg po in 20/60/20-DMSO/PEG400/H₂O.

To address the poor AdU selectivity and PK with the spiro oxindole purines, we again resorted to increased polarity. Over the course of SAR studies, we identified a novel series of spiro oxindole amides such as 5b and 5c that have exceptional potency (PI3Kδ IC₅₀ 0.98, 0.90 nM), selectivity versus PI3Kα (1700×, 850×), and low adenosine uptake activity. Based on our models, the carbonyl of the pendant cyclopropyl amide is directed toward and likely contacts bulk solvent, while also orienting the cyclopropyl group toward Trp760 and Met752. However, we noted poor permeability for 5b and 5c (Papp 9, 10 × 10^–6 cm/s), and poor rat PK for 5b and 5c, giving short half-lives (t_1/2 0.6, 0.5 h) and low bioavailability.

To address the permeability issue, we capped the spiro oxindole with an N-methyl group, giving compounds 5d and 5e in which the permeability increased 3-fold (Papp 31, 32 × 10^–6 cm/s). The potency was retained with this spiro oxindole N-methyl cap, and the selectivity versus PI3Kα (1400×, 430×) and AdU (IC₅₀ 9.1, 8.5 μM) was outstanding. We observed improved PK in rat with lower clearances (22, 10 mL/min/kg), better half-lives versus earlier spiro oxindoles (t_1/2 1.4, 1.1 h), and excellent bioavailability (F 100, 57%). The trend continued in dog, where lower clearances (6.1, 3.3 mL/min/kg) gave better half-lives (t_1/2 2.0, 3.0 h) and bioavailability (82, 99%) for 5d and 5e.

Potency evaluation in human whole blood was used for clinical human dose prediction, in which we targeted the B-cell surface biomarker CD69 IC₅₀ at trough concentration. CD69 is expressed in several hemopoietic cells as an early activation marker in chronic lymphocytic leukemia and is correlated with poor clinical prognosis.²⁹⁻³¹ The CD69 biomarker whole blood activity for selected inhibitors is shown in Table 6, along with the predicted human doses for each. Due to a lack of turnover of optimized leads in microsomes or hepatocytes, we used the allometry method from rat and dog PK profiles for dose predictions.^32,33

Table 6. Human Dose Predictions Based on Human Whole Blood CD69 Biomarker Activity.

		Human Dose Predictionb
	Human WB CD69 IC50 (nM; unbound IC50 in parentheses)a	Dose (mg)	t1/2 (h)	P/Tc
4c	84 (50)	54, qd	7	10
4d	80 (40)	8, qd	20	2
5d	38 (13)	17, qd	10	8
5e	71 (23)	34, qd	6	13

^aDose prediction target C_trough.

^bAllometry approach using Tang method, based on rat and dog PK data.

^cRatio of predicted C_max/C_trough.

Lead compounds 4c and 4d have comparable activities in the CD69 whole blood assay. However, the predicted human PK for 4c (Cl 2.9 mL/min/kg; Vd 1.8 L/kg; F 72%) is inferior to the predicted human PK for 4d (Cl 1.2 mL/min/kg; Vd 2.1 L/kg; F 91%). Consequently, the predicted human dose for 4d is better (8 mg qd) with a longer human half-life (20 h) and low peak/trough ratio.

While the predicted human clearance and volume values for 4c and 4d were within 2× based on allometry from either rat or dog, we found a larger 4× spread in these predicted parameters for 5c and 5d based on allometry from either rat or dog. The predicted human PK for spiro oxindole 5d (Cl 2.7 mL/min/kg; Vd 2.2 L/kg; F 82%) is comparable to what was predicted for 4c; however, the better potency led to a more favorable dose prediction (17 mg qd with t_1/2 10 h). Predicted human PK for ethyl analog 5e is characterized by a lower volume (Cl 1.2 mL/min/kg; Vd 0.6 L/kg; F 65%) and shorter human half-life.

In addition to having a low once-daily predicted human dose, compounds 4d and 5d exhibit high kinome selectivity (>1000-fold over PI3Kδ for 265 kinases; see Supporting Information) and no ion channel activity (HERG, Nav1.5, Cav1.2 IC₅₀ > 30,000 nM) or CYP activity (CYP-2C8/2C9/2D6/3A4 IC₅₀ > 50,000 nM). Solubility in high-throughput solubility assays was high, particularly for optimized compounds. Given the number of aromatic rings embedded in our lead structures, we carried out high-dose rat PK studies to ensure good dose proportionality. Compounds 4d and 5d exhibited good dose proportionality to 100 and 200 mg/kg, reaching AUC values of 390 and 490 μM·h, respectively, as summarized in the Supporting Information. Additional oxindole PI3Kδ SAR and synthetic procedures are available in the patent literature.³⁴

The oxindole moiety is a versatile three-dimensional scaffold that directs benzylic substitution in a rigid fashion. When linked to a purine hinge-binding core, exquisitely selective inhibitors of PI3Kδ kinase were discovered by exploiting small differences in isoform selectivity pockets. By attenuating the lypophilicity of the oxindole moiety, we improved selectivity versus adenosine uptake, preclinical species PK, and physical properties. The excellent potency and kinome selectivity of 7-azaoxindole 4d, together with a low plasma clearance and good half-life in rat and dog, supported a low once-daily predicted human dose. Spiro oxindole 5d also offered outstanding potency and good PK to support low a predicted human dose. Together, these oxindole purines offer a novel highly selective chemotype with structurally distinct features from approved PI3Kδ inhibitors.

Glossary

Abbreviations

AdU

adenosine uptake

AKT

protein kinase B

CD69

cell surface protein expressed by cluster of differentiation 69 gene

PI3Kα

phosphoinositide 3-kinase alpha

PI3Kβ

phosphoinositide 3-kinase beta

PI3Kγ

phosphoinositide 3-kinase gamma

PI3Kδ

phosphoinositide 3-kinase delta

Supporting Information Available

The Supporting Information is available free of charge at https://pubs.acs.org/doi/10.1021/acsmedchemlett.0c00441.

• Synthetic procedures and compound characterization data. Biochemical, cell and whole blood assay protocols. Kinome selectivity data tables for compounds 4c, 4d, 5d, and 5e. (PDF)

The authors declare no competing financial interest.