The clinical development of risovalisib indicates that PIK3CA alteration is best regarded as an entry biomarker rather than a complete predictor of response. In the first-in-human phase Ia study, 36 of 51 patients had PIK3CA-mutant tumors. The confirmed objective response rate was 11.9% among all 42 response-evaluable patients and 14.3% among the 28 evaluable patients with PIK3CA-mutant tumors [5]. Thus, PIK3CA mutation enriched only modestly for response across an unselected mixture of solid-tumor lineages. This result does not diminish the value of target genotyping; instead, it shows that genomic activation of PI3Kα must be interpreted together with the biological context that determines whether pathway inhibition is propagated to a tumor-controlling phenotype.
One such context is the coupling between receptor tyrosine kinase (RTK) signaling and ERK output. In breast cancer models, activating PIK3CA mutations and HER2 amplification were associated with CYH33 sensitivity, whereas the treatment-induced decrease in phosphorylated ERK (p-ERK) tracked antitumor activity more closely than AKT inhibition alone [6]. A related principle emerged in head and neck squamous cell carcinoma (HNSCC): PI3Kα-dependent recruitment and phosphorylation of GAB1 permitted CYH33 to suppress ERK, whereas EGFR-dependent maintenance of GAB1–ERK signaling weakened the antiproliferative effect [23]. The translational implication is not that p-ERK is a universal biomarker, but that its early change may reveal whether PI3Kα blockade has interrupted the lineage-relevant RTK–MAPK circuitry. Baseline RTK status and an on-treatment p-ERK readout could therefore complement PIK3CA genotyping in diseases in which PI3K and ERK signaling are functionally coupled.
Cell-cycle competence provides a second layer of discrimination. In esophageal squamous cell carcinoma (ESCC), PIK3CA mutation or amplification did not fully account for the activity of CYH33 across 16 patient-derived xenograft models. By contrast, CCND1 copy number and the ability to engage p21–Rb–E2F1–SKP2 control were more closely associated with response [7]. This observation shifts biomarker development from measuring target inhibition to measuring its phenotypic consequence: a tumor may show reduced p-AKT yet remain insensitive if Rb phosphorylation and E2F-dependent proliferation persist. CCND1 and the integrity of the G1/S checkpoint should therefore be evaluated as context-dependent modifiers rather than as stand-alone pan-tumor biomarkers.
These findings support a layered selection framework. PIK3CA alteration can define the target-positive population; tumor lineage and co-alterations such as HER2/RTK activation or CCND1 amplification can refine the prior probability of response; and early pharmacodynamic markers, including p-AKT, p-ERK and Rb/E2F activity, can establish whether pathway inhibition has reached the decisive proliferative program. ARID1A and PTEN are biologically plausible modifiers in selected disease contexts, but neither should yet be presented as a clinically validated biomarker for risovalisib. Their value will require prospective testing within defined lineages and in conjunction with dynamic pathway measurements. The practical objective is therefore not to replace PIK3CA testing, but to embed it in a biomarker hierarchy that separates target presence from functional PI3Kα dependency.
Resistance studies across breast cancer and ESCC converge on a common distinction between proximal target engagement and distal growth control. CYH33 can inhibit PI3K–AKT signaling in both sensitive and resistant models, but this biochemical effect does not necessarily produce durable cell-cycle arrest or tumor regression. Resistance arises when lineage-specific survival programs, bypass signaling or checkpoint defects uncouple PI3Kα inhibition from its downstream phenotypic consequence.
In breast cancer, prolonged CYH33 exposure generated resistant MCF7R and T47DR cells with increased mTORC1, E2F, KRAS and estrogen-response programs. An unbiased combination screen identified exportin 1 (XPO1) inhibition as a means of restoring sensitivity. The XPO1 inhibitor selinexor promoted nuclear accumulation of p53 and enhanced the activity of CYH33 in resistant models [13]. This mechanism illustrates an acquired state in which resistance is not explained by failure to inhibit PI3Kα itself, but by reinforcement of nuclear survival and proliferative programs downstream of, or parallel to, the inhibited pathway. Its clinical relevance is likely to depend on p53 competence and will require biomarker-defined validation.
ESCC provides complementary models of both intrinsic and acquired resistance. Gain-of-function screening identified EZH2 as a determinant of intrinsic resistance: EZH2-dependent repression of CDKN1A limited p21 accumulation, maintained Rb phosphorylation and prevented CYH33-induced G1 arrest. Pharmacological EZH2 inhibition restored checkpoint engagement and enhanced CYH33 activity in cell and patient-derived xenograft models [12]. In independently generated ESCC lines with acquired resistance after approximately six months of escalating drug exposure, no new PIK3CA mutation was detected. Instead, one resistant model acquired an HRAS G12S mutation, while resistant states showed persistent MAPK, mTORC1 and MYC-associated programs; pathway-matched inhibition of MEK, mTORC1 or BET proteins partially restored sensitivity in the corresponding models [11]. These results argue against a single universal resistance mechanism even within one histological lineage.
The shared principle is a decoupling of proximal PI3K inhibition from distal control of proliferation. Consequently, reduced p-AKT should not be used alone as a surrogate for effective treatment. Resistance-oriented sampling should also assess ERK and mTORC1 activity, MYC output, the p21–Rb–E2F checkpoint and lineage-specific survival programs. This broader pharmacodynamic panel can distinguish insufficient target suppression from biological bypass and, importantly, can nominate a combination partner on the basis of the escape route actually operating in the tumor.
Combination development should be organized by the failure mechanism being corrected and by the maturity of the supporting evidence. Within the current risovalisib program, endocrine therapy has the strongest clinical basis. In a phase Ib study of risovalisib plus fulvestrant in 52 patients with HR-positive/HER2-negative advanced breast cancer, 39 patients had PIK3CA-mutant disease and had not previously received fulvestrant. In this subgroup, the confirmed objective response rate was 17.9%, the clinical benefit rate was 41.0% and the median progression-free survival was 10.91 months. The recommended phase II combination dose was risovalisib 30 mg once daily with fulvestrant 500 mg; among the 21 patients treated at this dose, median progression-free survival was 10.97 months [19]. Although these conference-reported results require confirmation in larger controlled studies, they support the clinically established logic of suppressing estrogen receptor signaling and PI3Kα in parallel in PIK3CA-mutant HR-positive/HER2-negative breast cancer.
Preclinical combinations address two related forms of signaling escape. The first is failure to translate upstream inhibition into G1 arrest. CDK4/6 inhibition strengthened Rb dephosphorylation and G1 blockade and synergized with CYH33 in KRAS-mutant non-small cell lung cancer models [15], whereas EZH2 inhibition restored p21-dependent checkpoint competence in ESCC [12]. The second is maintenance of bypass signaling. In acquired CYH33-resistant ESCC models, MEK, mTORC1 or BET inhibition was effective only when matched to the dominant MAPK, mTORC1 or MYC-associated resistance program [11]. In HNSCC, EGFR inhibition has a mechanistic rationale when EGFR-dependent GAB1–ERK signaling persists after PI3Kα blockade [23]. Conversely, in KRAS G12C-mutant models, CYH33 combined with sotorasib suppressed EGFR/IGF1R-driven PI3K feedback and inhibited both parental and sotorasib-resistant tumors [14]. Collectively, these data favor pathway-matched combinations over empiric drug pairing.
Strategies directed at tumor dissemination and microenvironmental support require equally careful interpretation. In lung squamous cell carcinoma, CYH33 itself reduced tumor-cell migration, invasion and metastasis while attenuating cancer-associated fibroblast conversion and HGF secretion [9]. This study establishes an effect of CYH33 on dissemination and stromal support, but it did not test a separate anti-migratory agent in combination with CYH33. More direct combination evidence comes from hepatocellular carcinoma models, in which CYH33 plus the multi-target RTK inhibitor AL3810 suppressed complementary AKT, ERK and angiogenic outputs and produced greater antitumor activity than either agent alone [18]. In immunocompetent breast cancer models, fatty acid synthase inhibition enhanced the in vivo activity of CYH33 by altering fatty-acid availability, increasing CD8+ T-cell function and reducing immunosuppressive macrophage states [10]. The stronger interaction in vivo than in vitro suggests that this is an immunometabolic, microenvironment-dependent strategy rather than a purely tumor-cell-autonomous combination.
Glucose-lowering therapy belongs to a different category. Hyperglycemia is a frequent on-target toxicity of systemic PI3Kα inhibition, and its management through dose interruption or reduction and antidiabetic medication can preserve treatment continuity and dose intensity [5]. However, there is currently no direct evidence that metformin or another glucose-lowering drug produces antitumor synergy with risovalisib. Metabolic supportive care should therefore not be ranked alongside fulvestrant, CDK4/6, MEK, RTK or other mechanism-tested partners as an anticancer combination. This distinction is important because a regimen may be essential for safe drug delivery without constituting a tumor-directed therapeutic strategy.
The next indications for risovalisib should be prioritized according to context-specific PI3Kα dependency rather than the prevalence of PIK3CA mutation alone. A practical development framework can integrate five dimensions: lineage, co-alterations, cell-cycle competence, pathway coupling and the tumor microenvironment. These dimensions are not interchangeable biomarkers; together, they estimate whether pharmacological PI3Kα inhibition is likely to reach a disease-defining phenotype and whether a rational combination can prevent or reverse escape.
Lineage establishes the initial biological and therapeutic context. HR-positive/HER2-negative breast cancer offers a clear endocrine combination backbone, whereas HER2-amplified breast cancers may show stronger RTK–PI3Kα coupling. ESCC appears more dependent on CCND1 status and the integrity of G1/S checkpoint control. Ovarian clear cell carcinoma provides the strongest current example of aligning a PIK3CA-enriched lineage with a defined unmet clinical need and validated risovalisib activity. Co-alterations should then refine, rather than replace, this lineage-based prior. Factors with direct CYH33 evidence, including HER2/RTK activation and CCND1 amplification, should be distinguished from plausible but unvalidated modifiers such as ARID1A or PTEN.
Functional pathway coupling provides the next decision point. If ERK remains active through EGFR, HER2, GAB1, HRAS or another RTK after AKT inhibition, the tumor is unlikely to be fully dependent on PI3Kα monotherapy. By contrast, coordinated suppression of AKT and ERK together with Rb dephosphorylation and reduced E2F output identifies a pharmacodynamic state more compatible with deep and sustained response. This model suggests that short-window on-treatment biopsies or blood-based pharmacodynamic assays may be more informative than increasingly complex baseline genomic panels alone. They can determine whether the predicted dependency is realized in the treated tumor and reveal the bypass pathway before overt clinical progression.
The microenvironment should also influence indication and combination selection. CAF–HGF signaling in lung squamous cell carcinoma, angiogenic and ERK outputs in hepatocellular carcinoma, and lipid-metabolic control of antitumor immunity in breast cancer all show that PI3Kα dependency is not solely encoded by the cancer-cell genome [9,10,18]. Candidate indications should therefore be compared not only by PIK3CA mutation frequency, but also by the extent to which lineage-specific stromal, vascular or immune programs are disrupted by PI3Kα inhibition or can be targeted through a tolerable partner drug.
Non-malignant PIK3CA-driven disorders offer a complementary test of this framework. PIK3CA-related overgrowth spectrum and PIK3CA-driven complex vascular malformations may have a more direct driver–phenotype relationship than genomically complex cancers. In an ongoing phase I/II study of risovalisib in patients with PIK3CA-related overgrowth spectrum or PIK3CA-related vascular malformations, a 2026 conference abstract reported reductions in target-lesion volume in all 31 treated adults and adolescents; responses defined as at least 20% volume reduction occurred in 82.6% of adults and 50% of adolescents, and 10 mg once daily was selected as the recommended phase II dose [26]. These findings are preliminary, derive from a meeting abstract and should not yet be treated as confirmatory clinical evidence. Nevertheless, they illustrate how development may extend from histology-defined cancer indications toward diseases defined more directly by a mosaic oncogenic driver.
Taken together, identical PIK3CA alterations may confer different responses to CYH33 because PI3Kα dependency is jointly determined by lineage, co-alterations, cell-cycle competence, pathway coupling and the tumor microenvironment. Translational development should therefore use PIK3CA status to identify a target-positive population, but rely on context-specific biology and dynamic pharmacodynamic measurements to select patients, explain resistance and design combination therapy.
References [5–23] follow the numbering already used in the current manuscript source. Add the following conference abstract as reference [26]: