A New Wave of PI3Kα Inhibitors

Over the past decades, PI3K inhibitor (PI3Ki) drug develop­ment has made slow but steady progress in overcoming many intertwined preclinical, clinical, and translational challenges. This has included the improvement of kinase target selectivity, optimization of the therapeutic window to balance efficacy with toxicity, and discovery of biomarkers of response and resistance across cancer types. The “first wave” of PI3Ki (e.g., wortmannin, LY294002) targeted all classes of PI3K and also PI3K-like kinases, including mTOR. These molecules had little efficacy with significant toxicity or solubility issues when tested in animal models (1). The “second wave” of PI3Ki (e.g., buparlisib) targeted all class I PI3K isoforms (α, β, γ, δ). Buparlisib demonstrated a small progression-free survival (PFS) improvement in combination with fulvestrant in estrogen receptor–positive metastatic breast cancer (ER⁺ MBC), which was outweighed by significant toxicities, but also displayed improved differential efficacy in patients with PIK3CA-mutant tumors (2). The “third wave” of PI3Ki (e.g., alpelisib) selectively targets the α isoform and demonstrated a moderate PFS improvement in combination with fulvestrant in ER⁺ MBC with significant but manageable toxicities (3). Given its favorable benefit-to-risk ratio, alpelisib was recently approved by the FDA for the treatment of ER⁺ MBC bearing PIK3CA mutations (4).

To date, almost all PI3Ki are orthosteric, binding at the active site and inhibiting both mutant PI3Kα in the cancer cell but also wild-type (WT) PI3Kα in healthy tissues, including metabolic tissues where PI3Kα is essential for insulin-dependent glucose uptake (Fig. 1). This latter mechanism, which is on-target with respect to PI3Kα but off-target with respect to the tumor, most notably causes hyperglycemia and rash. At best, these side effects can be managed with antidiabetic medications, antihistamines, and dietary modifications or may require dose reduction, but at worst they cause significant morbidity or necessitate discontinuation (5). As a result, treating patients with alpelisib and other PI3Kα inhibitors (PI3Kαi) in the real world is far from straightforward. Drugs that inhibit mutant PI3Kα while sparing WT PI3Kα would be predicted to maximize both efficacy and tolerability in PIK3CA-mutant tumors (6).

In this issue of Cancer Discovery, Buckbinder and colleagues report the development and preclinical characterization of STX-478, an allosteric, mutant-selective PI3Kαi (7). Promising preclinical and clinical data for other mutant-selective PI3Kαi, including LOXO-783 and RLY-2608, have also been previously presented at international meetings; however, this is the first peer-reviewed article of a mutant-selective PI3Kαi.

The authors report that STX-478 had kinome-wide selectivity in vitro and was selective for a variety of PI3Kα kinase domain mutants. Strikingly, the IC₅₀ for the most common PI3Kα variant found in cancer, H1047R, was 14-fold lower than for WT PI3Kα, but STX-478 showed only extremely minor (1.2- to 1.8-fold) selectivity for common PI3Kα helical domain mutants E542K and E545K. This mutant selectivity was not due to altered drug binding conformation between WT and H1047R PI3Kα, with STX-478 binding PI3Kα identically in a new allosteric site arising after major conformational rearrangement as shown through cocrystal structure. Rather, STX-478 had 15-fold lower binding affinity for WT compared with H1047R PI3Kα. It thus appears that STX-478 is selective for PI3Kα kinase domain mutants such as H1047R in vitro.

Aligning with these biochemical data, STX-478 selectively lowered phosphorylated AKT levels in nontransformed isogenic MCF10A cells expressing H1047R PI3Kα, compared with WT or E545K PI3Kα. However, when these analyses were extended to investigate cell viability in a panel of 900 tumor cell lines, surprisingly STX-478 inhibited the proliferation of both kinase (e.g., H1047R) and helical (e.g., E545K) domain PI3Kα-mutant cell lines over WT PI3Kα cell lines, as has previously been shown with alpelisib (8). The varied selectivity between mutations that is observed for STX-478 in different preclinical models deserves further mechanistic exploration.

Most importantly, Buckbinder and colleagues presented a series of compelling and rigorous experiments showing that the major functional difference between STX-478 and alpelisib is their effects on metabolism. In primary human subcutaneous adipocytes, it took an approximately 10 times greater concentration of STX-478 than alpelisib to achieve 50% inhibition of glucose uptake. In mice, alpelisib but not STX-478 was associated with insulin resistance as measured by insulin and glucose tolerance tests. Corroborating these effects in normal cells, in an H1047R PI3Kα squamous cancer cell line–derived xenograft (CDX) mouse model, both alpel­isib and STX-478 similarly reduced tumor volume, yet only alpel­isib raised serum insulin and glucose levels. Mechanistically, while alpelisib and STX-478 both reduced phosphorylated Akt and glucose oxidation in the tumor, only alpelisib had this effect in skeletal muscle. Along with similar data from other CDX and patient-derived xenograft (PDX) models, the authors convincingly establish that unlike alpelisib, STX-478 selectively inhibits mutant PI3Kα in the tumor but spares the WT enzyme found in healthy tissues. This is a remarkable result and is promising for reducing the adverse metabolic effects of PI3Kαi in patients. Encouragingly, preliminary data that have been presented for other mutant-selective PI3Kαi, RLY-2608 and LOXO-783, also show little to no metabolic dysfunction in mice.

The authors then showed that STX-478 monotherapy causes tumor regression in some but not all CDX and PDX models bearing helical domain, kinase domain, or double mutations in PI3Kα across multiple cancer types. Of note, models that were resistant to STX-478 were also resistant to alpelisib. PI3Kαi monotherapy has limited clinical efficacy in cancer (9) but is effective when part of a combination therapy, as with fulvestrant in ER⁺ MBC. Thus, the authors investigated the efficacy of STX-478 combination therapies with fulvestrant and the CDK4/6 inhibitor palbociclib in ER⁺ breast cancer xenograft models. Most strikingly, in an H1047R PI3Kα ER⁺ breast cancer PDX model, the combination of STX-478 and fulvestrant led to tumor regression for almost 100 days. When dosing ended, modest tumor growth was observed only a month after stopping treatment. Notably, the addition of palbociclib did not improve the effects of STX-478 alone or STX-478 and fulvestrant doublet therapy. Given the FDA approval of alpelisib in combination with fulvestrant for PIK3CA-mutant ER⁺ breast cancer (4), these data indicate that STX-478 in combination with fulvestrant could be a more desirable clinical strategy, as it should spare metabolic dysfunction.

These results are paradigm-shifting; however, important preclinical, clinical, and translational questions remain. Preclinically, the complete mechanism of action by which this compound inhibits mutant PI3Kα remains elusive. While the authors have comprehensively proven that this compound is selective for H1047R over WT PI3Kα, there is discordance between biochemical and cellular data as to whether helical domain PI3Kα mutants such as E545K are sensitive to the compound. Clinically, it will be critical to determine if this improvement in toxicity will translate into similar clinical efficacy in ER⁺ MBC as compared with PI3Kα isoform–selective inhibitors or even superior efficacy due to patients being able to stay on drug longer. Moreover, mutant-selective PI3Kαi may be optimal to start combining with existing targeted therapies, chemotherapies, and immunotherapies to expand the scope of treatment. Translationally, hundreds of other PIK3CA mutations across the entire gene have been identified in cancer (10), and so which of these are also sensitive to the compound remains unknown. On the basis of the authors’ analysis of 900 tumor cell lines, not all PIK3CA-mutant cell lines were sensitive to STX-478, so the repertoire of patients who could respond to STX-478 would greatly expand with a more in-depth analysis of mutant–response relationships.

Overall, the authors have presented a detailed preclinical characterization of a new mutant-selective PI3Kαi, STX-478, that reduces tumor growth but displays a superior metabolic safety profile relative to alpelisib in mouse models. This study sets a benchmark for the development and preclinical characterization of other mutant-selective PI3Kαi. Together with the current study and limited available data for other mutant-selective PI3Kαi, such as RLY-2608 and LOXO-783, the outlook for this “fourth wave” of PI3Kαi is promising. These compounds have potential to greatly impact treatment regimens for PIK3CA-mutated ER⁺ breast cancer, as well as other PIK3CA-mutated cancers and diseases such as PIK3CA-related overgrowth syndrome. We eagerly await clinical data for STX-478 and are excited to ride this new wave of PI3Kαi.