Discovery of Pyridopyrimidinones that Selectively Inhibit the H1047R PI3Kα Mutant Protein

Abstract

The H1047R mutation of PIK3CA is highly prevalent in breast cancers and other solid tumors. Selectively targeting PI3Kα^H1047R over PI3Kα^WT is crucial due to the role that PI3Kα^WT plays in normal cellular processes, including glucose homeostasis. Currently, only one PI3Kα^H1047R-selective inhibitor has progressed into clinical trials, while three pan mutant (H1047R, H1047L, H1047Y, E542K, and E545K) selective PI3Kα inhibitors have also reached the clinical stage. Herein, we report the design and discovery of a series of pyridopyrimidinones that inhibit PI3Kα^H1047R with high selectivity over PI3Kα^WT, resulting in the discovery of compound 17. When dosed in the HCC1954 tumor model in mice, 17 provided tumor regressions and a clear pharmacodynamic response. X-ray cocrystal structures from several PI3Kα inhibitors were obtained, revealing three distinct binding modes within PI3Kα^H1047R including a previously reported cryptic pocket in the C-terminus of the kinase domain wherein we observe a ligand-induced interaction with Arg1047.

Introduction

The PI3K (Phosphoinositide 3-Kinase) pathway regulates many important cellular processes including cell growth, survival, and proliferation.¹⁻⁶ PI3K family members are lipid kinases that phosphorylate specific species of phosphoinositides in cellular membranes and transduce signals in a tightly regulated manner. There are three classes of PI3K proteins (Classes I–III),¹⁻⁶ of which the Class I PI3K proteins catalyze the phosphorylation of PIP2 (phophoinositol-3,4-diphosphate) to PIP3 (phophoinositol-3,4,5-triphosphate), resulting in the recruitment of AKT to the plasma membrane and eventual phosphorylation of AKT. Class IA PI3Ks, like PI3Kα, are heterodimers comprised of a regulatory subunit (p85α, p85β, p55α, p55γ, or p50α) and a catalytic subunit (p110α, p110β, or p110δ). Activating mutations of the PIK3CA gene that encodes for the p110α isoform (PI3Kα) are highly prevalent in solid tumors and are present in nearly 30% of all breast cancers.⁷⁻¹³ The three most common activating point mutations of PIK3CA occur within the p110α subunit, resulting in PI3Kα^H1047R, PI3Kα^E545K, and PI3Kα^E542K mutant proteins.¹⁴ The H1047R mutation accounts for the highest occurrence of somatic PIK3CA mutations in breast cancers,¹⁵⁻¹⁷ and additional therapies are needed for these patients.^18,19

The high occurrence of these PI3Kα mutant proteins in solid tumors has led to the design of isoform-selective orthosteric inhibitors of PI3Kα,²⁰⁻²³ including the only clinically approved inhibitor of PI3Kα, alpelisib (1).^19,22,24,25 However, treatment with alpelisib commonly results in hyperglycemia within patients, a form of on-target toxicity that can result in dose interruption, reduction, or discontinuation.²⁴⁻²⁶ This on-target toxicity is a consequence of the key role that PI3Kα^WT plays in the regulation of glucose metabolism,²⁶⁻²⁸ and results in a limited therapeutic index. For this reason, recent research in this field has focused on the discovery of compounds that selectively inhibit the mutant forms of PI3Kα (i.e., H1047X, E545X, and E542X).

In 2012, Williams and co-workers reported the binding of a PI3Kβ-selective inhibitor, PIK-108³⁰ (2), within the wild-type form of PI3Kα (Figure 1).²⁹ Serendipitously, although PIK-108 displays high selectivity for PI3Kβ (PI3Kβ IC₅₀: 57 nM, PI3Kα IC₅₀: 2600 nM),³⁰ it produced superior cocrystals with PI3Kα when compared to more potent and selective inhibitors of PI3Kα. The cocrystal structure revealed that PIK-108 bound to both the ATP-binding site (Figure 1, cyan) and a second, cryptic site in close proximity to the His1047 residue (Figure 1, magenta). This allosteric binding site provided a potential druggable pocket for a selective inhibitor of PI3Kα^H1047R.

Figure 1.

Reported cocrystal structure of PIK-108 bound to two sites within PI3Kα^WT (PDB: 4A55(29)). PIK-108 in cyan is bound within the orthosteric site, and PIK-108 in magenta is binding within the allosteric site near His1047.

Years later, in 2021, Petra Pharma/Loxo/Eli Lilly published a patent application detailing selective PI3Kα^H1047R inhibitors (e.g., 3–4, Figure 2) that closely resemble the structure of PIK-108,³¹ and several additional closely related applications were published shortly thereafter.³²⁻³⁴ Simple removal of the hinge binding motif (morpholine) from the C2-substituent on the chromenone ring in 2 and installation of a carboxylic acid on the phenyl ring to provide compounds like 3 and 4, resulted in potent and selective inhibitors of PI3Kα^H1047R (13.5-fold selectivity reported for 3 against PI3Kα^WT).³¹ Within the same year, Relay Therapeutics³⁵⁻⁴⁰ published a patent application describing pan mutant selective PI3Kα inhibitors (e.g., 5),³⁵ which contain no obvious motif for targeting the Arg1047 residue. The patent application from Relay also contained no reported selectivity data against PI3Kα^WT. While Lilly currently has the only reported PI3Kα^H1047R selective inhibitor in the clinic (LOXO-783), Relay Therapeutics and Scorpion Therapeutics⁴¹⁻⁴³ are currently in clinical trials with pan mutant selective inhibitors of PI3Kα (Relay: RLY-2608³⁹ and RLY-5836; Scorpion: STX-478). Although the structures of STX-478 (6)⁴³ and RLY-2608³⁹ were both recently disclosed, the chemical structures and structural details of binding to PI3Kα for LOXO-783 and RLY-5836 have not been released. Herein, we provide crystallographic evidence for three distinct binding modes of PI3Kα inhibitors and detail the discovery of a pyridopyrimidinone series that selectively inhibits the function of PI3Kα^H1047R.

Figure 2.

Chemical Structures of Alpelisib (1), PIK-108 (2), and representative literature compounds (3–6) from Lilly, Relay, and Scorpion.

Results and Discussion

To provide more insight into the binding modes of these literature compounds, two discrete cocrystal structures of compounds 4 and 5 bound to PI3Kα^H1047R (Figure 3A,B) were solved (in the presence of the alpelisib derivative A-66,⁴⁴ PDB: 8V8H and 8V8I, respectively). Interestingly, while 4 binds near the Arg1047 residue in the same allosteric pocket as PIK-108, we observed the binding of compound 5 in a previously unreported cryptic pocket within the protein. This cryptic pocket was recently revealed as the binding site of STX-478 (6)⁴³ and RLY-2608.³⁹

Figure 3.

A. Cut-away view of the discrete cocrystal structure of compound 4 bound to PI3Kα^H1047R showing the cryptic binding pocket (PDB: 8V8H) B. Discrete cocrystal structure of compound 5 bound to PI3Kα^H1047R (PDB: 8V8I). C. Co-crystal structure of A-66, 4, and 5 simultaneously bound to PI3Kα^H1047R (PDB: 8 V8J).

In the initial cocrystal structure of compound 4 bound to PI3Kα^H1047R (Figure 3A), the benzoic acid appears to form a salt bridge with the Arg1047 residue that likely accounts for the reported selectivity³¹ against wild-type for this series of chromenone inhibitors. While the carboxylate also makes a hydrogen bonding interaction with Gln981 and the chromenone core participates in a π-stacking interaction with Phe954, no additional significant polar interactions between 4 and the protein are observed. However, the cocrystal structure also revealed an intramolecular hydrogen bond between the anilinic N–H and the carboxylate of 4 (Figure S3).

The pan mutant PI3Kα inhibitor 5 was found to bind in a second cryptic pocket under the activation loop (residues 933–958) of PI3Kα^H1047R, secluded from any of the common point mutations found in PI3Kα mutant cancers. The cocrystal structure (Figure 3B) revealed that 5 participates in hydrogen bonding interactions with three distinct backbone carbonyls: the first between the benzothiophenyl amide N–H and the backbone carbonyl of Leu911, the second between the anilinic N–H and Gly912, and the final hydrogen bond between the isoindolinone N–H and Asp1018. Additionally, the 2-methylphenyl and benzothiophene moieties share edge-to-face interactions with Phe1002, and the thiophenyl sulfur makes a mildly favorable sigma-hole interaction with the side chain of Thr813.

Based on results from our laboratory and recent reports detailing the cocrystal structures of STX-478⁴³ bound to PI3Kα^H1047R and RLY-2608³⁹ bound to WT PI3Kα, it is clear that compounds like 5, RLY-2608, and STX-478 (6) bind within the same cryptic pocket of PI3Kα. It is difficult to provide justification solely on the basis of the cocomplexed crystal structures for how the binding mode of 5 can provide selectivity against PI3K^WT since the oncogenic mutations are relatively distal to the binding pocket. Buckbinder et al. suggest selectivity is the result of altered kinetics of the mutants.⁴³ Varkaris et al. concur and further demonstrate that the altered kinetics are due to mutant perturbed conformations increasing accessibility of the cryptic pocket.³⁹

In our work, further crystallographic studies led to the discovery that under optimized crystallization conditions, a cocomplex crystal structure with A-66, 4, and 5 simultaneously bound to PI3Kα^H1047R could be obtained (Figure 3C). The interactions observed in the discrete crystal structure of 4 are present in this structure. Similarly, 5 showed no significant changes between the discrete structure and the cocrystal containing all three compounds.

Compounds 4 and 5 were further evaluated in two cellular assay formats, an AlphaLISA assay (measuring phospho-AKT (pAKT) levels) and a viability assay in both the T47D (PI3Kα^H1047R) and SKBR3 (PI3Kα^WT) cell lines (Table 1). In these cell-based assays, SKBR3 is a PIK3CA wild-type cell line used as a surrogate model to evaluate the effects of WT PI3Kα inhibition and to determine selectivity for these compounds. Compound 4 gave high potency and selectivity across both cellular assay formats (pAKT T47D IC₅₀: 3 nM, Viability T47D IC₅₀: 160 nM, Table 1) while compound 5 (pAKT T47D IC₅₀: 332 nM, Viability T47D IC₅₀: 488 nM, Table 1) was less potent and selective under the assay conditions.

Table 1. Evaluation of Published Selective PI3Kα Inhibitors and Discovery of a Novel Pyridopyrimidinone Core.

compound	pAKT AlphaLISA IC50 (nM)a		viability IC50 (nM)b		viability selectivity 1047R/WT IC50
	T47D	SKBR3	T47D	SKBR3
4	3	3185	160	9082	57
5	332	115	488	4309	9
7	78	5894	1110	>10,000	9
8	3	1465	120	>10,000	83
9	179	>10,000	6239	>10,000	2
10	2544	>10,000	>10,000	>10,000

^aMeasured inhibition of pAKT formation using alphaLISA after 24 h of treatment.

^bMeasured decrease in viability using CellTiter-Glo 2.0 (CTG) after 72 h of treatment.

Utilizing a structure-based drug design, our goal was to discover a highly selective inhibitor of PI3Kα^H1047R. The pyridopyrimidinone 7 was designed and resulted in modest potency in T47D cells (pAKT T47D IC₅₀: 78 nM, Viability T47D IC₅₀: 1110 nM, Table 1). Further installation of a C3-methyl to give 8 significantly increased the potency of this novel scaffold and provided high selectivity against the cells expressing wild-type PI3Kα. The transposition of the pyridone nitrogen to the 3-position of the 6,6-scaffold to give compounds 9 and 10 resulted in a 39-fold decrease or complete loss in potency (viability) when compared to compound 4. An X-ray cocrystal structure of 7 bound to PI3Kα^H1047R revealed that 7 shares a similar binding mode to chromenone 4 (Figure 4A) and a superposition of the cocomplex structures of compounds 4 and 7, demonstrate that the compounds overlay well (Figure 4B). Further analysis of the structure reveals that the C3-vector could potentially be used to target the His931 residue. Additionally, the C2-substituent protrudes into a large lipophilic portion of the allosteric binding site, providing a potential growth vector for new analogs.

Figure 4.

A. Discrete co-crystal structure of compound 7 (PDB: 8V8V). B. Overlay of compounds 4 (pink) and 7 (magenta) bound to PI3Kα^H1047R.

While installation of the C3-methyl in compound 8 resulted in a significant increase in potency, this change was met with a decrease in metabolic stability in human hepatocytes (Cl_int: 340 mL/min/kg, Table 2). Introduction of a fluorine in the C3-position (11) further increased the potency of the scaffold (pAKT T47D IC₅₀: 0.3 nM); however, the fluoro-analog 11 provided less stability than the C3-methyl analog 8 (Cl_int for 11: 862 mL/min/kg). In an attempt to decrease the metabolism and target the His931 residue, the C3-nitrile analogue 12 was synthesized and provided a compound that was 3-fold more potent and 4-fold more stable in human hepatocytes (Cl_int for 12: 79 mL/min/kg) than 8. Conversely, replacing the C3-nitrile in 12 with a trifluoromethyl (13) or an amide (14) resulted in a substantial loss in potency.

Table 2. Examination of C3-substitution on the Pyridopyrimidinone Core.

compound	R	pAKT AlphaLISA IC50 (nM)a		Clint human hepatocytes (mL/min/kg)b
		T47D	SKBR3
7	–H	78	5894	181
8	–Me	3	1465	340
11	–F	0.3	>10,000	862
12	–CN	1	4739	79
13	–CF3	28	>10,000	N.D.
14	–CONH2	>10,000	>10,000	N.D.

^aMeasured inhibition of pAKT formation using alphaLISA after 24 h of treatment.

^bIntrinsic clearance obtained from scaling in vitro half-lives from pooled hepatocytes.

A cocrystal structure of 12 bound to PI3Kα^H1047R was obtained and confirmed that the nitrogen of the C3-nitrile participates in a hydrogen bonding interaction with His931 (Figure 5A). As observed in the structure of 7 (Figure 4), the cocrystal structure of 12 and PI3Kα^H1047R also displays a salt bridge formed between the benzoic acid moiety and Arg1047, further supporting the hypothesis that H1047R selectivity for this chemotype is largely driven by direct interaction with the point mutation. Interestingly, the position of the Arg1047 side chain is in a qualitatively similar position relative to the cryo-EM structure of PI3Kα^H1047R (Figure 5B)⁴⁵ suggesting that minimal protein rearrangement is required for engagement of the salt bridge interaction. The binding of 12 displaces the activation loop C-terminus by ∼6 Å at Pro953 compared to that of apo PI3Kα^H1047R. This, in turn, changes the positioning of the second basic box membrane binding element (residues 948–951),⁴⁶ which we hypothesize negatively impacts membrane interaction and leads to inhibition of enzymatic activity for PI3Kα^H1047R. While the C-terminal end of kα11 is displaced ∼2 Å at Asp1045 relative to the apo protein, it is unclear if this impacts the autoinhibitory ordering of the C-terminal tail including the WIF membrane interaction element (residues 1057–1059).⁴⁷ However, both 12 and the displaced C-terminus of the activation loop sit adjacent to the catalytic HRD element (residues 915–917, Figure 5C). Dynamic changes in HRD could impact phosphoryl transfer, potentially resulting in an additional mechanism of inhibition.

Figure 5.

A. X-ray cocrystal structure of compound 12 bound to PI3Kα^H1047R (PDB: 8V8U). B. Overlay of the X-ray cocrystal structure of 12 (green protein) with the cryo-EM structure of PI3Kα^H1047R (gray protein, PDB: 8GUB)⁴⁵ showing inhibitor-induced perturbation of the activation loop and basic box 2 (blue). Inset shows kα11 helix and similar relative positions of the Arg1047 side chain between apo and liganded structures. C. X-ray cocrystal structure of compound 12 showing proximity of ligand to HRD motif (orange) and activation loop (yellow).

To further decrease the metabolism of these pyridopyrimidinones, designs were focused on finding replacements for the C2-piperidine ring (Table 3). Decreasing the ring size (15) or installing dimethyl groups at the 4-position (16) resulted in compounds with nearly the same potency and stability as parent compound 12. However, substituting the piperidine ring with two fluorines at the 4-position to give 17, decreased the intrinsic clearance in human hepatocytes by 3-fold (Cl_int for 17: 29 mL/min/kg) while retaining high cellular potency (Viability T47D IC₅₀: 48 nM) when compared to 12. Increased substitution around the 4,4-difluoropiperidine ring to give 18, preserved the potency of the scaffold while slightly decreasing the metabolic stability (Cl_int for 18: 39 mL/min/kg). Removing the nitrogen from the piperidine ring to give the 4,4-difluorocylcohexane 19 resulted in a significantly decreased intrinsic clearance in human hepatocytes (Cl_int for 19: 12 mL/min/kg), albeit with a 5-fold loss in potency within the T47D cellular viability assay. Replacing the piperidine ring in 12 with a difluoroazetidine (20) was well-tolerated and provided a similar in vitro profile to the difluoropiperidine 17. Conversely, the installation of the difluoroazaspiroheptane 21 or the thiomorpholine dioxide 22 resulted in analogs that were completely inactive in the T47D viability assay.

Table 3. Compounds Designed to Lower the Cl_int in Human Hepatocytes.

^aMeasured inhibition of pAKT formation using alphaLISA after 24 h of treatment.

^bMeasured decrease in viability using CellTiter-Glo 2.0 (CTG) after 72 h of treatment.

^cIntrinsic clearance obtained from scaling in vitro half-lives from pooled hepatocytes.

While the difluoropiperidine 17 displayed high cellular potency and moderate stability, it was found that glucuronidation of the benzoic acid was the only discernible metabolite in human hepatocytes (see Figure S1, Supporting Information). In the hopes of minimizing the glucuronidation of this acid, several benzoic acid replacements were explored (23–27, Table 4). Unfortunately, substitution around the benzene ring (23–24, Table 4) or replacement of the ring with a heteroaromatic ring (25–27, Table 4) led to a significant decrease in cellular potency. For this reason, compound 17 was progressed into pharmacokinetic studies within mice, rats, and dogs.

Table 4. Evaluating Analogs to Replace the Benzoic Acid Found in Compound 17.

^aMeasured inhibition of pAKT formation using alphaLISA after 24 h of treatment.

^bMeasured decrease in viability using CellTiter-Glo 2.0 (CTG) after 72 h of treatment.

When administered in rodents, 17 displayed moderate bioavailability (48–57%F, Table 5) with high clearance in mice and moderate clearance in rats, similar to the intrinsic clearance measured in hepatocytes for these species. IV administration of 17 in dogs also resulted in high clearance, correlating well with Cl_int found in dog hepatocytes. Upon oral dosing of 17 in dogs, the resulting bioavailability was low (7%), indicating that 17 was unsuitable for progression into development. However, the moderate bioavailability in rodents and high selectivity for PI3Kα^H1047R (see Tables SI-2–SI-4 in Supporting Information) suggested 17 as a compelling tool compound for pharmacodynamic (PD) and efficacy models in mice, while further analogs were designed to address the metabolic stability of the scaffold.

Table 5. Overall Pharmacokinetic (PK) Parameters for Compound 17 Across Species.

ADME parametersa	mouse	rat	dog
Clint hepatocytes (mL/min/kg)	304	70	104
in vivo Cl (mL/min/kg)	180	46	64
Vd,ss (L/kg)	13.6	4.8	1.5
IV t1/2 (h)	2	3.5	0.7
F (%)	48	57	7
dose: IV/PO (mg/kg)	3/100	3/100	2/10

^aSee Supporting Information for details.

Compound 17 was first evaluated in an HCC1954 (ER-/HER2+ human breast tumor cell line with endogenous PI3Kα^H1047R) mouse xenograft model to assess its pharmacodynamic properties. Modulation of pAKT was monitored after a single dose of 17 via oral or intraperitoneal (IP) administration. The results from the PD study showed that oral dosing of 17 at 100 or 300 mg/kg resulted in negligible inhibition of pAKT (Figure 6), likely due to the low concentrations of 17 at these time points (Table SI-5, Supporting Information). However, administration of 17 via IP injection at either 60 or 120 mg/kg (as a single dose, qd) resulted in a much higher concentration of 17 (Table SI-5), leading to 42 and 81% inhibition of pAKT, respectively, for up to 8 h (Figure 6). When compared to the PI3Kα^WT inhibitor alpelisib, the 120 mg/kg IP dose of 17 displayed similar inhibition of pAKT at both the 1 and 8 h time points.

Figure 6.

Single Dose Pharmacodynamic Study and a 14-Day TGI Study of Compound 17 in the HCC1954 tumor model. A. pAKT modulation at 1 h post dose B. pAKT modulation at 8 h post dose C. 14-day efficacy study.

Based on the results from the PD study, a 14-day tumor growth inhibition (TGI) study was performed in HCC1954 tumor-bearing mice (Figure 6C). Compound 17 was dosed at IP at either 100 mg/kg qd or 80 mg/kg twice daily (b.i.d.) for 14 days. While the 100 mg/kg (IP, qd) dose resulted in significant tumor growth inhibition (88%), the 80 mg/kg (IP, bid) dosing regimen provided tumor regressions (−34%). To the best of our knowledge, this is the first reported case of an H1047R-selective inhibitor of PI3Kα providing tumor regressions in an HCC1954 tumor model.

Chemistry

Cyclization of the 4H-pyrido[1,2-a]pyrimidin-4-one core (30) was accomplished through the condensation of 2-amino-3-bromo-5-methylpyridine (28) and malonyl dichloride (27, Scheme 1). Next, direct installation of 4,4-difluoropiperidine onto the core was realized using PyBOP for activation of the cyclic amide motif to provide 31.^48,49 A Heck reaction of the aryl bromide 31 using butyl vinyl ether followed by acidic hydrolysis provided aryl ketone 32, which was easily brominated using NBS to provide 33. Condensation of the chiral Ellman’s sulfinamide⁵⁰ onto ketone 33, followed by a highly diastereoselective reduction using Schwartz’s reagent (zirconocene hydrogen chloride) provided the benzylic sulfinamide 34 with >97% de. Installation of the key nitrile at the C3 position was achieved through a palladium-catalyzed cross-coupling with zinc cyanide, and acidic removal of the chiral auxiliary provided the benzylic amine 35. Finally, the synthesis was completed with the installation of the critical benzoic acid motif that provided selectivity for PI3Kα^H1047R mutant inhibition. Initial efforts were focused on Ullmann–Goldberg type cross-couplings of aryl halides bearing carboxylic acids; however, yields were generally low. A Chan-Evans-Lam coupling using 2-carboxylphenylboronic acid proved fruitful when run in amide-based solvents, and the use of the stronger base DBU in comparison to triethylamine was found to be necessary to achieve higher conversions, providing 17 in 40% isolated yield.⁵¹

Scheme 1. Gram-Scale Synthesis of 17.

Reagents and conditions: (a) DCM, 25 °C, 48 h, 61% yield; (b) PyBOP, DIPEA, DMF, 25 °C, 12 h, 38% yield; (c) butyl vinyl ether, P(t-Bu)₃Pd G2, N,N-dicyclohexylamine, dioxane, 100 °C, 1 h; (d) HCl (12M), THF, 25 °C, 1 h, 25% yield over 2 steps; (e) NBS, DMF, 25 °C, 2 h, 78% yield; (f) (R)-tert-butylsulfinamide, Ti(OEt)₄, THF, 80 °C, 12 h, 80% yield; (g) Cp₂ZrHCl, DCM, 25 °C, 1 h, 70% yield; (h) Zn(CN)₂, XantPhos Pd G3, DMA, 100 °C 1 h; (i) HCl (4 M in dioxane), dioxane, 25 °C, 1 h, 89% yield over 2 steps; (j) 2-carboxyphenylboronic acid, Cu(OAc)₂, DBU, DMF, 25 °C, 12 h, 40% yield.

Conclusions

Our efforts toward a PI3Kα^H1047R mutant selective inhibitor led to the discovery of a pyridopyrimidinone series that provided compound 17, a potent and selective inhibitor of PI3Kα^H1047R that provided tumor regressions in the HCC1954 tumor model in mice. Crystallographic studies provided insights into the binding mode of several literature inhibitors of PI3Kα. Additionally, we elucidated several cocrystal structures of the pyridopyrimidinone inhibitors bound within the PI3Kα^H1047R mutant protein, leading to suggested mechanisms of inhibition and selectivity. Encouraged by the results from the HCC1954 tumor studies, we began further optimization of our scaffold to help increase the metabolic stability of these selective PI3Kα^H1047R inhibitors. However, increasing competitive activity focused on this chemical scaffold^38,52,53 drove us to alternative designs, to be disclosed in due course.

Experimental Section

General Procedures

All final compounds were purified to ≥95% purity by either high-performance liquid chromatography (HPLC) or supercritical liquid chromatography (SFC). Purity was determined by HPLC and additional structural characterization was performed by proton NMR, carbon NMR, and high-resolution mass spectrometry, as described below. All chemicals were purchased from commercial suppliers and used as received unless otherwise indicated. Proton nuclear magnetic resonance (¹H NMR) spectra were recorded on Bruker Avance 400 MHz spectrometers. Chemical shifts are expressed in δ ppm and are calibrated to the residual solvent peak: proton (e.g., CDCl₃, 7.27 ppm). Coupling constants (J), when given, are reported in hertz. Multiplicities are reported using the following abbreviations: s = singlet, d = doublet, dd = doublet of doublets, t = triplet, q = quartet, m = multiplet (range of multiplets is given), br = broad signal, dt = doublet of triplets. Carbon nuclear magnetic resonance (¹³C NMR) spectra were recorded using a Bruker Avance HD spectrometer at 100 MHz. Chemical shifts are reported in parts per million (ppm) and are calibrated to the solvent peak: carbon (CDCl₃, 77.23 ppm). The purity of the test compounds was determined by high-performance liquid chromatography (HPLC) on an LC-20AB Shimadzu instrument. HPLC conditions were as follows: The purity for test compounds was determined by high-performance liquid chromatography (HPLC) on an LC-20AB Shimadzu instrument. HPLC conditions were as follows: HPLC conditions were as follows: Kinetex EVO C18 3.0 × 50 mm, 2.6 um, 10–80% ACN (0.0375%TFA) in water (0.01875% TFA), 3–10 min runs, flow rate 1.2 mL/min, UV detection (λ = 220, 215, 254 nm), or Kinetex C18 LC column 4.6 × 50 mm, 5 μm, 10–80% ACN (0.0375%TFA) in water (0.01875% TFA), 4–10 min runs, flow rate 1.5 mL/min, UV detection (λ = 220, 215, 254 nm), or XBridge C18, 2.1 × 50 mm, 5 μm, 10–80% ACN in water buffered with 0.025% ammonia, 4–10 min runs, flow rate 0.8 mL/min, UV detection (λ = 220, 215, 254 nm). The mass spectra were obtained using liquid chromatography–mass spectrometry (LC-MS) on an LCMS-2020 Shimadzu instrument using electrospray ionization (ESI). LCMS conditions were as follows: Kinetex EVO C18 30 × 2.1 mm, 5 μm, 5–95% ACN (0.0375% TFA) in water (0.01875% TFA), 1.5 min run, flow rate 1.5 mL/min, UV detection (λ = 220, 254 nm), or Kinetex EVO C18 2.1 × 30 mm, 5 μm, 5–95% ACN in water buffered with 0.025% ammonia, 1.5 min run, flow rate 1.5 mL/min, and UV detection (λ = 220, 254 nm). High-resolution mass measurements were carried out on an Agilent 1290LC and 6530Q-TOF series with ESI. The SFC purity was determined with a Shimadzu LC-30ADsf.

Preparation of Compound 17

9-Bromo-2-hydroxy-7-methyl-4H-pyrido[1,2-a]pyrimidin-4-one (30)

To a solution of 3-bromo-5-methylpyridin-2-amine (300 g, 1.60 mol, 1.0 equiv) in dichloromethane (2.0 L) was added malonyl dichloride (249 g, 1.76 mol, 172 mL, 1.10 equiv), and the mixture was stirred at 25 °C for 48 h. The mixture was then filtered, and the filter cake was washed with dichloromethane (500 mL × 2). The filter cake was further concentrated in vacuo to give compound 30 (269 g, 983 mmol, 61% yield, 93% purity) as a yellow solid. LCMS [M + 3]⁺= 256.9.

9-Bromo-2-(4,4-difluoropiperidin-1-yl)-7-methyl-4H-pyrido[1,2-a]pyrimidin-4-one (31)

To a solution of 30 (269 g, 983 mmol, 1.00 equiv) in DMF (2.0 L) was added PyBOP (614 g, 1.18 mol, 1.20 equiv), N,N-diisopropylethylamine (508 g, 3.93 mol, 685 mL, 4.00 equiv), and 4,4-difluoropiperidine (131 g, 1.08 mol, 1.10 equiv), and the mixture was left to stir at 25 °C for 12 h. After this time, water (4.0 L) was added to the mixture. The resultant precipitate was filtered, and the filter cake was dried under a vacuum to give 31 (172 g, 377 mmol, 38% yield, 78% purity) as a yellow solid. LCMS [M + 3]⁺= 360.0.

9-Acetyl-2-(4,4-difluoropiperidin-1-yl)-7-methyl-4H-pyrido[1,2-a]pyrimidin-4-one (32)

A mixture of 31 (167 g, 366 mmol, 1.00 equiv), butyl vinyl ether (110 g, 1.10 mol, 141 mL, 3.00 equiv), dicylcohexylamine (73.0 g, 402 mmol, 80.2 mL, 1.10 equiv), P(t-Bu)₃Pd G2 (9.38 g, 18.3 mmol, 0.05 equiv) in dioxane (1.70 L) was degassed and purged with nitrogen 3 times, and then the mixture was stirred at 100 °C for 1 h under a nitrogen atmosphere. The mixture was then cooled to 25 °C, filtered, and concentrated under reduced pressure to give 9-(1-butoxyvinyl)-2-(4,4-difluoropiperidin-1-yl)-7-methyl-4H-pyrido[1,2-a]pyrimidin-4-one (200 g, crude) as a yellow solid, which was without further purification. LCMS [M + 1]⁺= 378.2. To a solution of 9-(1-butoxyvinyl)-2-(4,4-difluoropiperidin-1-yl)-7-methyl-4H-pyrido[1,2-a]pyrimidin-4-one (200 g, 530 mmol, 1.00 equiv) in tetrahydrofuran (1.5 L) was added hydrochloric acid (12 M, 132 mL, 3.00 equiv), and the mixture was left to stir at 25 °C for 1 h. The reaction mixture was then concentrated under reduced pressure to give a residue. The residue was diluted with dichloromethane (1.0 L) and the pH of the mixture was adjusted (pH ∼8) with sodium bicarbonate (saturated aqueous solution). The mixture was then extracted with dichloromethane(500 mL), and the organic layer was washed with brine (1.0 L), dried over anhydrous sodium sulfate, filtered, and the filtrate was concentrated under reduced pressure to give a residue. The residue was purified by flash column chromatography (SiO₂, petroleum ether/ethyl acetate = 10/1 to 1/1) to give 32 (45.6 g, 134 mmol, 25% yield, 94% purity) as a yellow solid. LCMS [M + 1]⁺ = 322.1. ¹H NMR (400 MHz, CDCl₃) δ = 8.85 (d, J = 0.8 Hz, 1H), 7.84 (d, J = 2.0 Hz, 1H), 5.70 (s, 1H), 3.91–3.68 (m, 4H), 2.76 (s, 3H), 2.36 (s, 3H), 2.09–1.96 (m, 4H).

9-Acetyl-3-bromo-2-(4,4-difluoropiperidin-1-yl)-7-methyl-4H-pyrido[1,2-a]pyrimidin-4-one (33)

To a solution of 32 (45.6 g, 134 mmol, 1.00 equiv) in DMF (450 mL) was added NBS (26.1 g, 147 mmol, 1.10 equiv), and the mixture was left to stir at 25 °C for 2 h. The mixture was then diluted with water (900 mL) and filtered, and the filter cake was dried under vacuum to give 33 (47.0 g, 104 mmol, 78% yield, 89% purity) as a yellow solid. LCMS [M + 1]⁺= 402.0. ¹H NMR (400 MHz, CDCl₃) δ = 9.00–8.96 (m, 1H), 8.92 (s, 1H), 7.95 (d, J = 2.0 Hz, 1H), 3.83–3.74 (m, 4H), 2.81 (s, 3H), 2.43 (s, 3H), 2.22–2.07 (m, 4H).

(R)-N-((R)-1-(3-Bromo-2-(4,4-difluoropiperidin-1-yl)-7-methyl-4-oxo-4H-pyrido[1,2-a]pyrimidin-9-yl)ethyl)-2-methylpropane-2-sulfinamide (34)

To a solution of 33 (47.0 g, 104 mmol, 1.00 equiv), (R)-2-methyl-2-propanesulfinamide (15.2 g, 125 mmol, 1.20 equiv) in tetrahydrofuran (500 mL) was added titanium(IV) ethoxide (47.5 g, 208 mmol, 43.2 mL, 2.00 equiv) and 1,2-dimethoxyethane (9.39 g, 104 mmol, 10.8 mL, 1.00 equiv), and the mixture was stirred at 80 °C for 12 h. The reaction was then diluted with water (15.0 mL) and ethyl acetate (2.00 L), filtered, and concentrated under reduced pressure to give a residue. The residue was purified by column chromatography (SiO₂, petroleum ether/ethyl acetate = 5/1 to 1/1) to give (R)-N-(1-(3-bromo-2-(4,4-difluoropiperidin-1-yl)-7-methyl-4-oxo-4H-pyrido[1,2-a]pyrimidin-9-yl)ethylidene)-2-methylpropane-2-sulfinamide (51.7 g, 83.3 mmol, 80% yield, 81% purity) as a yellow oil. LCMS [M + 1]⁺ = 505.2. To a solution of (R)-N-(1-(3-bromo-2-(4,4-difluoropiperidin-1-yl)-7-methyl-4-oxo-4H-pyrido[1,2-a]pyrimidin-9-yl)ethylidene)-2-methylpropane-2-sulfinamide (51.7 g, 83.3 mmol, 1.00 equiv) in dichloromethane (500 mL) was added Schwartz’ Reagent (25.8 g, 100 mmol, 1.20 equiv), and the mixture was stirred at 25 °C for 1 h. The mixture was then diluted with dichloromethane (500 mL) and washed with water (500 mL × 2). The organic phase was dried over anhydrous sodium sulfate, filtered, and concentrated under reduced pressure to give a residue. The residue was purified by column chromatography (SiO₂, petroleum ether/ethyl acetate = 10/1 to 1/0) to give 34 (30.0 g, 57.9 mmol, 70% yield, 98% purity) as a yellow solid. LCMS [M + 1]⁺= 505.2. ¹H NMR (400 MHz, CDCl₃) δ = 8.74 (s, 1H), 7.60 (d, J = 1.6 Hz, 1H), 4.89 (t, J = 7.2 Hz, 1H), 4.72 (br d, J = 7.6 Hz, 1H), 3.87–3.57 (m, 4H), 2.40 (s, 3H), 2.17 (m, 4H), 1.64 (d, J = 6.8 Hz, 3H), 1.22 (s, 9H).

(R)-9-(1-Aminoethyl)-2-(4,4-difluoropiperidin-1-yl)-7-methyl-4-oxo-4H-pyrido[1,2-a]pyrimidine-3-carbonitrile (35)

A mixture of 34 (20.0 g, 38.6 mmol, 1.00 equiv), zinc cyanide (9.06 g, 77.1 mmol, 4.90 mL, 2.00 equiv), XantPhos Pd G3 (3.66 g, 3.86 mmol, 0.10 equiv) in N,N-dimethylacetamide (200 mL) was degassed and purged with nitrogen 3 times, and then, the mixture was stirred at 100 °C for 1 h under a nitrogen atmosphere. The mixture was then cooled to 25 °C and filtered, and the filter cake was washed with ethyl acetate (600 mL). The filtrate was diluted with water (600 mL) and extracted with ethyl acetate (300.0 mL × 2). The combined organic layers were washed with brine (600 mL × 2), dried over anhydrous sodium sulfate, filtered, and concentrated under reduced pressure to give (R)-N-((R)-1-(3-cyano-2-(4,4-difluoropiperidin-1-yl)-7-methyl-4-oxo-4H-pyrido[1,2-a]pyrimidin-9-yl)ethyl)-2-methylpropane-2-sulfinamide (23.0 g, crude) as a yellow solid, which was used without further purification. LCMS [M + 1]⁺ = 452.3. To a solution of (R)-N-((R)-1-(3-cyano-2-(4,4-difluoropiperidin-1-yl)-7-methyl-4-oxo-4H-pyrido[1,2-a]pyrimidin-9-yl)ethyl)-2-methylpropane-2-sulfinamide (23.0 g, 45.5 mmol, 1.00 equiv) in ethyl acetate (230 mL) was added hydrochloric acid (4 M in dioxane, 45.5 mL, 4.00 equiv), and the mixture was left to stir at 25 °C for 1 h. After this time, a precipitate formed, and the precipitate was filtered. The filter cake was dried under vacuum to give 35 (15.6 g, 40.3 mmol, 89% yield, 99% purity, hydrochloric acid salt) as a yellow solid. LCMS [M + 1]⁺ = 348.3. ¹H NMR (400 MHz, DMSO-d₆) δ = 8.64 (br s, 4H), 8.17 (d, J = 1.6 Hz, 1H), 5.04–4.87 (m, 1H), 4.07–3.93 (m, 4H), 2.38 (s, 3H), 2.26–2.05 (m, 4H), 1.59 (d, J = 6.8 Hz, 3H).

(R)-2-((1-(3-Cyano-2-(4,4-difluoropiperidin-1-yl)-7-methyl-4-oxo-4H-pyrido[1,2-a]pyrimidin-9-yl)ethyl)amino)benzoic Acid (17)

To a solution of 35 (12.0 g, 31.1 mmol, 1.00 equiv, hydrochloric acid salt) in DMF (100 mL) was added copper acetate (8.47 g, 46.6 mmol, 1.50 equiv), DBU (14.2 g, 93.2 mmol, 14.1 mL, 3.00 equiv), 4 Å molecular sieves (1.00 g), and 2-carboxyphenylboronic acid (7.74 g, 46.6 mmol, 1.50 equiv), and the mixture was left to stir at 25 °C for 12 h. The pH of the mixture was then adjusted to pH ∼5 with hydrochloric acid (2N in water). The solution was extracted with ethyl acetate (250 mL × 2), and the combined organic layers were washed with brine (500 mL), dried over anhydrous sodium sulfate, filtered, and concentrated under reduced pressure. The residue was purified by column chromatography (SiO₂, petroleum ether/ethyl acetate = 1/0 to 1/1) to give a yellow solid. The crude product was triturated with methyl tert-butyl ether (100 mL) at 25 °C for 1 h to give (R)-2-((1-(3-cyano-2-(4,4-difluoropiperidin-1-yl)-7-methyl-4-oxo-4H-pyrido[1,2-a]pyrimidin-9-yl)ethyl)amino)benzoic acid (17, 5.77 g, 12.3 mmol, 40% yield, 99% purity) as a yellow solid. ¹H NMR (400 MHz, CD₃OD) δ = 8.51 (br s, 1H), 7.91 (br d, J = 7.6 Hz, 1H), 7.76 (s, 1H), 7.20 (br t, J = 7.6 Hz, 1H), 6.56 (br t, J = 7.6 Hz, 1H), 6.37 (br d, J = 8.4 Hz, 1H), 5.28 (q, J = 6.4 Hz, 1H), 4.14 (br s, 4H), 2.25 (s, 3H), 2.22–2.10 (m, 4H), 1.64 (br d, J = 6.4 Hz, 3H). ¹³C NMR (101 MHz, DMSO-d₆) δ = 170.0, 159.6, 159.1, 149.3, 146.9, 138.5, 137.2, 134.5, 131.7, 124.4, 123.8, 117.4, 115.0, 112.1, 110.6, 70.5, 47.3, 43.8, 40.1, 33.5(t, J = 23.0 Hz), 21.5, 17.5 HRMS: Calculated for C₂₄H₂₃F₂N₅O₃ [M + H]⁺: 468.1842; found 468.1852. [α]_D²⁵ – 318.62 (c = 0.93 g, DMSO).

Glossary

Abbreviations

bid

bis in die (twice a day)

Cl

clearance

IV

intravenous

PI3Kα

phosphoinositide 3-kinase alpha

PK

pharmacokinetics

PO

per os (oral)

qd

quaque die (once daily)

SAR

structure–activity relationship

SFC

supercritical fluid chromatography

t_1/2

half-life

TGI

tumor growth inhibition

V_d,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/acs.jmedchem.4c00078.

• Molecular formula strings (CSV)

• Synthetic experimental procedures for compounds 7–27; NMR spectra and HPLC trace of final compounds; PI3Kα X-ray structures for compounds 1, 4, 5, 7, 12, and A-66; biochemical and cellular assay protocols; and in vivo PK, PD, and TGI studies (PDF)

Accession Codes

Atomic coordinates for PI3Kα cocrystal structures are available from the RCSB Protein Databank (www.rcsb.org) with accession codes: 8V8H, 8V8I, 8V8J, 8V8U, 8V8V.

Author Contributions

^‡ J.K. and S.H. are co-first authors. The manuscript was written through contributions of all authors. All authors have given approval to the final version of the manuscript. J.M.K. and S.J.H. contributed equally to the preparation of this manuscript.

The authors declare the following competing financial interest(s): All authors of this manuscript are employees of Mirati Therapeutics.