Phase I Basket Study of Taselisib, an Isoform-Selective PI3K Inhibitor, in Patients with PIK3CA-Mutant Cancers

The phosphatidylinositol 3-kinase (PI3K)–AKT–mTOR signaling pathway is one of the most frequently dysregulated pathways in human cancers and has been implicated in tumor cell growth and survival (1). This pathway can be activated in cancer cells due to genomic aberrations in one or more of its key components, most commonly involving the phosphatidylinositol-4,5-bisphosphate 3-kinase catalytic subunit alpha (PIK3CA) gene, encoding the PI3K alpha catalytic subunit, p110α (1–3). Mutations in PIK3CA cluster in the kinase and helical domains, but additional recurrent mutations have been described throughout the gene (4). Both helical and kinase domain mutations have been extensively characterized and lead to aberrant activation of the PI3K signaling pathway (5), while the functional consequences of less frequent mutations outside these domains are not as well understood. The frequency and distribution of mutant PIK3CA alleles differ according to tumor lineage; the therapeutic relevance of targeting these mutations in different disease settings remains to be determined.

To date, the primary focus of clinical development for many PI3K inhibitors has been in hormone receptor (HR)-positive, human epidermal growth factor receptor 2 (HER2)-negative breast cancer, due, in large part, to the high prevalence (∼40%) of PIK3CA mutations in this subtype of breast cancer (6). Specifically, randomized phase III trials of the pan-isoform PI3K inhibitor, buparlisib, the alpha-selective inhibitor, alpelisib, and the alpha, delta, and gamma inhibitor, taselisib, have all been completed in HR-positive, HER2-negative breast cancer. Studies of all three agents in combination with fulvestrant have demonstrated statistically significant prolongation of progression-free survival (PFS) in patients with estrogen receptor (ER)-positive, HER2-negative breast cancer, with a more pronounced benefit in the PIK3CA-mutant population (7–9). However, the modest prolongation in PFS with buparlisib and taselisib were not deemed clinically meaningful due to their respective toxicity profiles. In contrast, addition of alpelisib to fulvestrant led to a clinically meaningful, 5.5-month improvement in PFS, leading to the FDA approval of this combination in HR-positive, PIK3CA-mutant, HER2-negative breast cancer (10). Collectively, these data act as further clinical proof of the biologic and therapeutic importance of this pathway in human cancer.

To evaluate the therapeutic importance of PIK3CA mutations outside of HR-positive, HER2-negative breast cancer, we designed a phase I global, multi-center, multi-histology basket trial in advanced solid tumors harboring PIK3CA mutations. Patients received taselisib, a small-molecule inhibitor of class I PI3K-alpha, -delta, and -gamma isoforms, that displays increased selectivity against cell lines with mutant PI3K-alpha compared with cell lines with wild-type PI3K-alpha (11, 12). Tumor tissue was collected for detailed genomic characterization of patient tumors and correlated with clinical outcomes.

Eligible patients had PIK3CA-mutant solid tumors as identified by either local or central tumor testing, performed on either archival tumor tissue or a fresh biopsy, if the former was not available. Patients needed to have locally advanced or metastatic disease that had progressed or failed to respond to at least one prior systemic treatment and must not have been considered candidates for regimens with proven clinical benefit. Patients were enrolled into one of 10 tumor-defined cohorts: endometrial cancer, bladder cancer, head and neck squamous cell carcinoma (HNSCC), cervical cancer, gastric and gastroesophageal junction cancer, small-cell lung cancer, triple-negative breast cancer (TNBC), colorectal cancer (KRAS wild-type), squamous cell cancer of a primary site not otherwise specified, or epithelial ovarian cancer (Supplementary Fig. S1). Patients with any other PIK3CA-mutant solid tumor were eligible to enroll in a 11th “all-comers” cohort. As inhibitors of this pathway, including taselisib, have been evaluated previously in HR-positive breast cancer, as well as non–small cell lung cancer, these histologies were excluded specifically.

Complete eligibility criteria are available in the Supplementary Materials and Methods. Briefly, patients were required to have an Eastern Cooperative Oncology Group performance status (ECOG PS) of 0 or 1, a fasting plasma glucose level ≤120 mg/dL, adequate bone marrow and liver function, serum albumin concentration ≥ 2.5 g/dL, and measurable disease as defined by RECIST version 1.1 criteria (13). Patients who had prior PI3K inhibitor therapy, inflammatory bowel disease, diabetes requiring treatment, or untreated/symptomatic central nervous system metastases were excluded.

The study was approved by institutional review boards and conducted in accordance with the principles of the Declaration of Helsinki, International Council for Harmonisation Guidelines, and the laws and regulations of the countries in which it was conducted. All patients provided written informed consent before undergoing any study procedures.

This was a multi-center, multi-cohort, single-arm basket trial incorporated into the original first-in-human phase I study of taselisib (ClinicalTrials.gov identifier: NCT01296555) through protocol amendment version 8, implemented on 31 March 2015. The dose-escalation portion of this phase I study has been described previously (12). For the current study, the tablet formulation of taselisib was utilized and administered at 6 mg once daily on a continuous basis in 28-day cycles (14). A subsequent protocol amendment, version 9, permitted lowering the starting dose to 4 mg to test whether a lower dose might improve tolerability. Patients with HNSCC with gastrostomy tubes (g-tube) who were unable to swallow tablets were permitted at the physician’s discretion to receive an extemporaneously prepared taselisib suspension (crushed taselisib tablets suspended in water and administered via g-tube) at equivalent doses. Pharmacokinetic blood draws were obtained on cycle 1 day 15, and day 1 of cycles 2 and 6, except for patients receiving the suspension formulation, in which case dense pharmacokinetics sampling was obtained. Optional tumor biopsies, if deemed safe and clinically feasible, were performed pre- and on-treatment for pharmacodynamics assessment, as well as optionally upon progression to evaluate mechanisms of acquired resistance.

Dosing with taselisib could be interrupted for up to 28 days in the event of toxicity or unanticipated medical events not associated with either study drug toxicity or disease progression (Supplementary Materials and Methods).

Patients were enrolled and treated on the basis of PIK3CA mutation status by local or central Cobas® PIK3CA Mutation Test (Roche Molecular Systems; ref. 15). Retrospective central testing for PIK3CA confirmation, as well as broader molecular profiling, were performed utilizing the FoundationOne® Next-Generation Sequencing Assay (Foundation Medicine, Inc.; ref. 16) on formalin-fixed, paraffin-embedded tumor tissue samples obtained either from prior excisions or fresh biopsies, as available. Patients enrolled at Memorial Sloan Kettering Cancer Center (New York, NY) also had local next-generation sequencing performed on some pretreatment and postprogression tumor samples (when sufficient sample was available in a non-prespecified analysis) utilizing the Memorial Sloan Kettering-Integrated Mutation Profiling of Actionable Cancer Targets (MSK-IMPACT™) Assay as described previously (3, 17).

Safety was assessed by monitoring and recording protocol-defined adverse events (AE) and serious AEs (SAE), and by monitoring protocol-specified laboratory parameters and vital signs. AEs were graded according to the NCI Common Terminology Criteria for Adverse Events Version 4.0 (18).

AEs of special interest (AESI) were selected on the basis of previously established “on-target” toxicities of PI3K inhibition (defined as hyperglycemia, diarrhea, colitis, rash, stomatitis, or pneumonitis).

Measurable disease per RECIST v1.1 was documented at screening. The same radiographic procedure used at baseline was used throughout the study for each patient. Postbaseline tumor assessments were conducted at the end of cycles 2, 4, 6, and 8, and every 12 weeks thereafter. Bone scans and brain scans were performed as clinically indicated. Tumor response was determined by investigators using RECIST v1.1 criteria with confirmation required (13).

As an amendment to the original phase I study, the primary endpoint was to assess the safety and tolerability of taselisib. However, key secondary endpoints included objective response rate, duration of response (DoR), PFS, clinical benefit rate (defined as confirmed objective response or without disease progression for ≥6 months per RECIST v1.1), safety, pharmacokinetics, and pharmacodynamics. Analysis of antitumor activity by DNA alterations was an exploratory objective of the study.

The study was designed to estimate the safety and activity of taselisib, while minimizing the number of patients treated with ineffective therapy. Thus, if objective response was observed in ≥2 of the first 10 patients enrolled onto each cohort, or other clinical benefit (e.g., a majority of patients demonstrate stable disease at week 8, although there were fewer than two responders), then expansion of enrollment up to a total of 20 patients for that cohort was permitted per protocol. This design provided a maximum enrollment of approximately 250 patients across all 11 cohorts.

All safety and activity analyses were based on the safety-evaluable population, defined as all patients who received at least one dose of taselisib. Objective overall response was defined as a complete or partial response, as determined by investigator assessment using RECIST v1.1 and confirmed by one consecutive assessment ≥4 weeks after initial documentation. Ninety-five percent confidence intervals (CI) were calculated using the Clopper–Pearson method. PFS and DoR were estimated by Kaplan–Meier methodology. PFS was defined as the time from the first day of taselisib treatment until documented disease progression or death on study, whichever occurred first. DoR was defined as the time from the initial complete or partial response to the time of disease progression or death, whichever occurred first. A data cut-off of 1 September 2017 was used for all analyses.

The Cancer Genome Atlas (TCGA) Pan-Cancer Atlas (N = 10,967 samples) PIK3CA mutations were derived on cBioPortal (Memorial Sloan Kettering Cancer Center, New York, NY; v3.0, accessed in June 2019). Pathway-level analyses, RTK/Ras/Raf, cell-cycle checkpoint, DNA damage response, epigenetic, and PI3K pathway genes were selected on the basis of the overlap with the FoundationOne® 395 Cancer Gene Panel (Foundation Medicine, Inc.) and Kyoto Encyclopedia of Genes and Genomes pathway database (Supplementary Table S1). Fisher exact tests were performed to compare gene-/pathway-level associations between dichotomous clinical benefit (defined as confirmed objective response or without disease progression for ≥6 months per RECIST v1.1) groups. All figures were generated using R v3.5.1.

Qualified researchers may request access to individual patient-level data through the clinical study data request platform (https://vivli.org/). Further details on Roche’s criteria for eligible studies are available at: https://vivli.org/members/ourmembers/. For further details on Roche’s Global Policy on the Sharing of Clinical Information and how to request access to related clinical study documents, see here: https://www.roche.com/research_and_development/who_we_are_how_we_work/clinical_trials/our_commitment_ to_data_sharing.htm.

Overall, 166 patients were enrolled and treated in 11 histology-specific cohorts (19 May 2015–1 September 2017; Table 1; Supplementary Fig. S1). The median age of patients enrolled in the trial was 61 years (range, 22–88 years) and median time from primary diagnosis to enrollment was 29.7 months (range, 3.7–304.8 months). The most common tumor types were HNSCC (26; 15.7%), cervical cancer (20; 12%), TNBC (17; 10.2%), ovarian cancer (14; 8.4%), and endometrial cancer (11; 6.6%). Forty-seven unique cancer types were represented in the study. A total of 110 (66.3%) patients were female and 110 (67.1%) also had an ECOG PS of 1. The median number of prior systemic therapies was three (range, 1–10); 18.7% of patients had prior immune checkpoint inhibitor therapy, including 38.5% of patients with recurrent or metastatic HNSCC (Table 1).

Locally obtained PIK3CA mutation testing was utilized as the basis for enrollment in 97.6% (162/166) of patients. For tissues sent to the sponsor, 93.9% were classified as archival by the site, and 4.5% were classified as fresh biopsies. The duration of time between tissue collection and enrollment on study ranged from 3 to 3,399 days, with an average time of 300 days. Central broader molecular profiling utilizing the FoundationOne® next-generation sequencing platform was completed in 87.3% (145/166) of patients.

The distribution of PIK3CA mutations per central testing was similar in previously treated compared with untreated patients across different TCGA tumor types (ref. 19; Supplementary Fig. S2). Single helical domain mutations were most common (73/145; 50.3%), followed by single kinase domain mutations (26/145; 17.9%), single mutations outside of either of these domains (17/145; 11.7%), mutations in more than one codon (15/145; 10.3%; patients with mutations in more than one codon were only included in this category, even if one of the codons was in the helical or kinase domains), and finally no mutation detected (14/145; 9.7%). Overall, 80% (32/40) of all unique PIK3CA mutations observed occurred at previously established hotspots (4); 87.6% (127/145) of patients harbored at least one PIK3CA mutation at a known hotspot. More than one PIK3CA mutation was observed in 10% of study participants with a centrally confirmed PIK3CA mutation, consistent with analysis from primary tumors in TCGA (11%; ref. 19), as well as metastatic tumors in MSK-IMPACT™ (11%; ref. 4) and METABRIC (13%; ref. 20). In total, 40 unique alterations were identified, with the majority being missense variants (37/40; 92.5%) and a minority harboring indels (3/40; 7.5%).

Overall concordance between locally reported PIK3CA mutation status and central profiling among patients with both results available was 86% (117/136) for at least one overlapping PIK3CA variant (Supplementary Table S2). In 9.7% of patients with a locally reported PIK3CA mutation, no mutation was detected by central testing. There were five codon discordant cases where central testing identified a PIK3CA variant different from the locally reported variant.

Safety is summarized in Table 2. Combining across dose levels, the most common treatment-emergent AEs in >30% of patients, regardless of causality, were diarrhea (57.8%), nausea (39.8%), fatigue (32.5%), decreased appetite (31.9%), and hyperglycemia (31.3%). At least one grade ≥3 AE, regardless of causality, was observed in 66.9% of patients, with a numerically higher frequency observed in patients started at 6 mg (72.1%) compared with 4 mg (52.3%).

Grade ≥3 AESIs were seen in 28.3% of patients, most commonly hyperglycemia (10.2%), diarrhea (7.8%), rash (6.6%), and colitis (6.0%; Table 2). Numerically higher rates of both all-grade (82.8% vs. 59.1%) and grade ≥3 (32.8% vs. 15.9%) AESIs were seen with 6 mg compared with 4 mg, respectively, including higher rates of grade ≥3 diarrhea (9.8% vs. 2.3%), colitis (7.4% vs. 2.3%), rash (8.2% vs. 2.3%), stomatitis (2.5% vs. 0%), and pneumonitis (two cases; 1.6%, vs. 0%).

Dose reductions and interruptions occurred in 30.3% and 48.4% of patients treated at 6 mg and 13.6% and 36.4% of patients treated at 4 mg, respectively. Patients treated at 6 mg also had a higher frequency of related SAEs compared with patients treated at 4 mg (21.3% vs. 9.1%). There were nine fatal AEs with two (septic shock and pneumonia) identified by the investigator as related to taselisib.

When stratified by tumor type, confirmed partial responses were observed in 15 (9%) patients, including those with HNSCC (four), endometrial (two), cervical (two), gallbladder/cholangiocarcinoma (two), TNBC (one), colorectal (one), pancreatic (sarcomatoid and pseudopapillary; one each), and mucoepidermoid parotid (one) tumors (Table 3; Fig. 1A). The median duration of objective response for the 15 confirmed partial responses was 8.3 months (95% CI, 6.6–18.2). Overall, median PFS was 3.6 months (95% CI, 3.1–4.2) and varied by tumor histology. Nine patients had a PFS >12 months, with a maximum of up to 23 months (gall bladder adenocarcinoma, ongoing at time of clinical cut-off date). Activity in four tumor-specific cohorts (HNSCC, cervical, TNBC, and ovarian cancer) met the prespecified criteria for expanded enrollment. Although the TNBC cohort met expansion criteria with two responses, one of the responses was later reassessed by the site as a nonresponse. In addition, although the endometrial cohort also met expansion criteria with two responses, these two responses occurred in different histologies (one in endometrioid and one in a rare mixed Mullerian tumor); the remainder of the patients did not show tumor reduction (Fig 1), and therefore, the decision was made not to expand this cohort.

Activity signals for single-agent taselisib were observed in HNSCC, cervical cancer, ovarian cancer, and the mixed histology “other” cohort (Supplementary Figs. S3–S6, respectively). In HNSCC, tumor reductions were observed in the majority (15/26; 57.7%) of patients, although all confirmed responders (4/26; 15.4%) were patients started at the 6 mg dose. Activity was observed regardless of human papillomavirus (HPV) status, as well as in patients previously treated with anti-programmed death-1 (PD-1) therapies, which have been approved as monotherapy for second-line advanced HNSCC. In patients with ovarian cancer, activity was mainly observed in the subset with clear-cell histology, with most patients of this subtype (4/6; 66.7%) achieving tumor regressions.

When stratified by mutant PIK3CA allele, tumor regressions were observed across mutations involving both helical and kinase domains, as well as those in other less recurrently altered regions (Fig. 1B). Activity was highest in helical domain mutants, with helical domain mutations present in 11 of 15 (73.3%) of confirmed responders. Among patients harboring multiple PIK3CA mutations, there was one confirmed response of 15 patients. A confirmed partial response was observed in a patient with cervical adenocarcinoma harboring a nonhelical/kinase domain PIK3CA mutation, R38C. Tumor regressions were also observed in patients in whom central testing did not detect a PIK3CA mutation, including two confirmed responders.

Integrating activity data by both tumor and variant type further demonstrated the complex interplay between these factors in determining outcome from taselisib treatment (Fig. 2). Specifically, in tumor types that were more refractory to taselisib, such as colorectal and endometrial cancers, tumor regressions were limited to helical domain mutants. In comparison, in the relatively more sensitive tumor lineages, such as HNSCC and cervical cancers, activity was observed across a broad range of alterations.

Given the low objective response rate and to evaluate whether the broader context of alterations influenced the outcome of PI3K inhibition, co-mutation patterns at both the gene and pathway level were correlated with clinical benefit (Fig. 3). Among patients with confirmed PIK3CA mutations, TP53 and PTEN mutations were more frequently found in patients with no clinical benefit compared with those who derived benefit [TP53: 40.3% (48/119) vs. 28% (7/25); P = 0.18 and PTEN: 11.8% (14/119) vs. 4% (1/25); P = 0.22; one-sided Fisher test, both not statistically significant]. Consistent with previous reports (21), co-occurring PTEN mutations were associated with a lower PFS (1.7 vs. 3.1 months; P = 0.002). We also noted more frequent mutations in the protein kinase, DNA-activated, catalytic subunit (PRKDC) gene among nonresponders [19.3% (23/119) vs. 8% (2/25); P = 0.12; not statistically significant], which was associated with a poorer PFS (1.7 vs. 3.1 months; P = 0.1; not statistically significant). Pathway-level analyses identified that while alterations in RTK/Ras/Raf pathway members were similar between patients who derived clinical benefit versus those who did not, concurrent alterations in cell-cycle checkpoint and DNA damage response pathways were more frequently found in patients who did not derive clinical benefit compared with those who did (cell-cycle checkpoint, 57.1% vs. 40% and DNA damage response, 58.8% vs. 40%; both not statistically significant).

Analysis of whether tumor mutational burden (TMB) was correlated with response was performed, and among six tumor types with sufficient sample sizes (N > 10) tested overall, the TNBC cohort observed a greater decrease in tumor size in patients with lower TMB (P_nominal = 0.02; Supplementary Fig. S7).

Tumor biopsies from patients who had progressed on taselisib (optionally obtained, with patient consent where feasible) identified a 53-year-old male with HPV-positive, PIK3CA E542K-mutant metastatic tonsillar cancer who was treated with four prior regimens, including cetuximab and pembrolizumab, before starting on taselisib. This patient experienced a dramatic and rapid resolution of a 7-cm chest wall mass leading to an unconfirmed partial response at 8 weeks. At his 16-week scan, new small lung lesions and recurrence of the chest wall lesion were noted. Postprogression biopsies of both sites were obtained, revealing the original PIK3CA E542K mutation in each sample, and two additional acquired PTEN nonsense mutations (Q171* and T319*) in the chest wall lesion and an acquired STK11 S216F hotspot mutation in the lung lesion (Fig. 4). Consistent with these genomic findings, IHC staining showed diffuse PTEN expression pretreatment, but focal PTEN loss in the chest wall tumor.

Whole-exome sequencing was performed on pretreatment and postprogression biopsies in 4 additional patients. In a patient with HNSCC and PIK3CA E542K mutation achieving a confirmed partial response, we detected loss of PIK3CA E542K and gain of E458_K459 deletion mutation in PIK3R1 in the postprogression biopsy, the gene encoding the p85 regulatory subunit of the heterodimeric PI3K enzyme. In one additional case in a patient with a best response of progressive disease, PIK3CA mutation was maintained, but a new PIK3R1 E555Q mutation (located in the iSH2 domain) was observed. In two other cases, PIK3CA mutation was also maintained at progression, and no acquired alterations considered to explain acquired resistance were identified.

In this multi-histology, genome-driven basket study, the isoform-selective PI3K inhibitor, taselisib, as a single-agent demonstrated limited activity across PIK3CA-mutant tumor types with an overall objective response rate of 9% and with most tumor regressions being relatively short-lived. The overall safety profile of taselisib was similar to prior experience from studies utilizing equivalent monotherapy dosing (12). As has been observed with other oncogenic alterations, activity appeared to vary as a function of both tumor type and PIK3CA mutation variant. Specifically, taselisib activity was numerically higher in HNSCC, cervical cancer, and clear-cell ovarian cancer, all areas of high unmet medical need that are characterized by poor prognosis and limited therapeutic options. In particular, recurrent/metastatic HNSCC, a disease with a response rate of approximately 5% to standard therapy (22, 23), demonstrated a confirmed response rate of 15.4% to taselisib, with responses observed in multiple patient subsets, including those which were HPV positive or HPV negative, and in patients who had previously been treated with anti-PD-1 mAb therapies.

Among PIK3CA variants, activity was greatest among hotspot helical domain mutants, but evidence of activity was also observed in several low frequency hotspots whose biologic activity has been less comprehensively characterized. Analysis of the interplay between tumor histology and PIK3CA variant type suggested that the apparent greater activity of taselisib in helical domain mutants may at least partially be explained by the increased prevalence of these mutations in HNSCC and cervical cancer (24, 25). There did not appear to be an enrichment for greater response among patients harboring multiple PIK3CA mutations, which is in contrast to previous studies demonstrating an increased response to PI3K inhibitors in patients with multiple PIK3CA-mutant tumors in ER-positive, HER2-negative, PIK3CA-mutant locally advanced or metastatic breast cancer (26). However, consistent with previous experience of PI3K inhibitor monotherapy in HR-positive/HER2-negative breast cancer (12), a prototype for PI3K pathway oncogene addiction and other histologies (27), overall single-agent activity of taselisib was modest. These results contrast with other basket studies targeting other molecular alterations; for example, the high activity of TRK inhibitors across tumor types harboring NTRK gene fusions (28, 29), indicating dependence on these single oncogenic drivers.

One reason for the relatively modest activity of taselisib is that its suboptimal therapeutic index prevents maximal inhibition of the PI3K pathway. In this study, we observed some evidence of dose-dependent activity with taselisib, with a numerically higher response rate among patients who received 6 mg versus those who received 4 mg, albeit with a less favorable toxicity profile at 6 mg. The relatively narrow therapeutic index for taselisib suggests that alternative strategies to enhance mutant and PIK3CA isoform selectivity may be important to improve the benefit–risk profile when targeting this pathway, an improvement which is essential in many cell types. One such strategy is the development of alpha-specific PI3K inhibitors, including alpelisib, which is approved in HR-positive, HER2-negative, PIK3CA-mutant advanced and metastatic breast cancer in combination with fulvestrant (9, 10) and GDC-0077, which has entered clinical testing (NCT03006172 and NCT04191499). GDC-0077, a potent PI3K inhibitor with >300-fold more selectivity for PI3K alpha over the other class I PI3K isoforms, appears to have promising clinical activity and a favorable toxicity profile both as a single agent and in combination with standard-of-care therapies (30, 31). Intermittent dosing schedules have also been explored as another approach to optimize the therapeutic index of these agents (32).

Another reason for the limited single-agent activity is related to the fact that oncogenic activation of the PI3K–AKT–mTOR pathway often occurs alongside pro-tumorigenic aberrations in other signaling networks, such as ER signaling in HR-positive breast cancer, highlighting the importance of combination strategies with PI3K inhibitors (33). Downstream inhibition of AKT1 and mTOR, in combination with endocrine therapy, has shown efficacy in patients with ER-positive, HER2-negative breast cancer, with this activity not limited to tumors harboring PIK3CA mutations (34). However, off-target toxicity has been seen with these agents. Also, because the activity of this pathway is modulated by feedback inhibition, efficacy of mTORC1 and AKT inhibitors could be limited by loss of mTORC1-dependent feedback inhibition and secondary FOXO-dependent activation of receptor expression, leading to upstream reactivation of the pathway (35).

To limit off-target toxicity and further improve efficacy, isoform-selective inhibitors, such as alpelisib and taselisib (4 mg dose), were evaluated in phase III trials in combination with fulvestrant (SOLAR-1; ref. 9 and SANDPIPER; ref. 36). Each demonstrated a statistically significant improvement in PFS in patients with HR-positive, HER2-negative, PIK3CA-mutant advanced or metastatic breast cancer (8, 9). However, the improvement in PFS with taselisib was modest and associated with significant toxicity, notably diarrhea due to inhibition of the δ isoform (grade ≥3 AEs: 12% for taselisib arm vs. <1% for placebo) and hyperglycemia (grade ≥3 AEs: 11% vs. <1%); further clinical development of taselisib has therefore been halted in this patient population. In contrast, alpelisib plus fulvestrant led to a clinically meaningful improvement in PFS and consequent FDA approval in patients with HR-positive, HER2-negative, PIK3CA-mutant advanced and metastatic breast cancer (9, 10). This disparity may be related to more potent and specific inhibition of p110α by alpelisib as evidenced by higher rates of grade ≥3 hyperglycemia (37% for alpelisib arm vs. <1% for placebo) in the SOLAR-1 trial (9) compared with those in SANDPIPER. Collectively, these data suggest that PIK3CA mutations play an oncogenic role in multiple cancer types, but that the therapeutic index of PI3K pathway inhibitors may be narrow, and that differences in both isoform and mutant selectivity, disease type and context, and pathway node inhibition may have important consequences for combinability, efficacy, and safety.

Other mechanistically driven combinations addressing the adaptive upregulation of compensatory pathways, including dual vertical node inhibition of the PI3K pathway with CDK4/6–cyclin D1–RB pathway inhibition in breast cancer (37) and EGFR inhibitors in HNSCC (38), have been promising preclinically, and such efforts are ongoing clinically in breast cancer (39–41). However, other attempts at combinatorial strategies with mTOR inhibitors, IGF1-R, or HER3 have been challenging to implement clinically, due to tolerance issues (42). Given the intra- and intertumoral molecular heterogeneity underlying cancer, a uniform combinatorial strategy is unlikely to be effective in all disease types, and a more systematic mapping of cancer dependencies for a given disease type and context will likely be necessary to identify optimal combinatorial strategies (43–45).

Finally, activity could be limited because of the presence of co-occurring mutations at both the gene and pathway level. Our exploratory analysis revealed that TP53 and PTEN mutations were numerically more frequent in patients who did not derive clinical benefit from taselisib compared with those who did. In addition, concurrent alterations in cell-cycle checkpoint and DNA damage response pathways were numerically more frequently found in patients who did not derive clinical benefit compared with those who did. The identified co-occurring mutations reflect, in part, the tumor types enrolled and are hypothesis-generating, observational data; expanded studies in additional tumor types may be performed to examine whether these associations hold.

Further indications of the importance of the PI3K pathway were demonstrated by the observation at disease progression of acquisition of genetic alterations in PTEN, PIK3R1, and STK11, indicating a convergent mechanism of resistance involving PI3K pathway reactivation. Of note, acquisition of PTEN alterations that signal downstream through the PI3K beta isoform has been described previously as a mechanism of acquired resistance to the PI3K alpha inhibitor, alpelisib (46). These findings extend this observation to other isoform-selective PI3K inhibitors. Similarly, STK11 is a serine-threonine kinase tumor suppressor that negatively regulates mTOR signaling. STK11 S216F is a previously characterized loss-of-function mutation associated with decreased activity of AMPK, a negative regulator of mTOR signaling (47, 48). Furthermore, recurrent mutations in PIK3R1 K459, which is located in the intervening antiparallel coiled-coil (iSH2) domain required for binding to the PI3K p110α catalytic subunit, have been described as a potential mechanism to activate the PI3K pathway (49).

Some limitations of this study include mixed histology within tumor type cohorts, small numbers of patients with multiple “rare” histologies, heterogeneous prior treatments, and lack of data on mutation testing via cell-free DNA and from paired normal tissue, the latter of which could facilitate analysis of mutant allele sub(clonality). A low rate of discordance between local and central PIK3CA mutations in archival specimens was observed, potentially reflecting intratumoral or tissue heterogeneity, or differences between local and central testing platforms. Low purity is unlikely to be contributing to the majority of discordant cases, as at least one other non-PIK3CA somatic variant was identified in 12 of the 14 cases without a centrally detected PIK3CA mutation. Overall, this concordance rate was similar to observations from prior similarly designed studies (50). Basket studies, in general, allow for efficient evaluation of multiple tumor types in a single study for hypothesis generation and signal-seeking in tumors and/or genotypes for further investigation. However, disadvantages typically include small sample sizes that limit statistical power, lack of a control arm, and a multiple relapsed patient population in which other driver mutations may be present, all of which could impact the results of a single-agent basket study.

In summary, through the conduct of a genome-driven multi-histology basket study, we provide additional evidence that PIK3CA mutations are both oncogenic and actionable in a broader range of tumor types and genomic variants than previously recognized. While the optimal strategy for inhibiting PI3K signaling in human malignancy remains to be determined, target biology, disease type, therapeutic index of an agent, and combination strategies potentially play an important role in determining benefit durability. Although taselisib is not under further clinical development, the modest single-agent activity signals observed in this study in some histologies (for instance, HNSCC), despite a heavily pretreated patient population, indicate that inhibitors with a better therapeutic index targeting the PI3K pathway, likely in combination, may provide value in the treatment of PIK3CA-mutant tumors.

K. Jhaveri reports personal fees from Novartis, Pfizer, Genentech, AstraZeneca, Taiho Oncology, Seattle Genetics, Jounce Therapeutics, BMS, AbbVie, ADC Therapeutics, and Spectrum Pharmaceuticals, and other from Novartis, Genentech, Pfizer, Lilly Pharmaceuticals, AstraZeneca, Zymeworks, Puma Biotechnology, Clovis Oncology, Debio Pharmaceuticals, and Immunomedics outside the submitted work. M.T. Chang reports other from Genentech during the conduct of the study. D. Juric reports grants from Genentech during the conduct of the study; grants from Takeda, Amgen, Celgene, Placon Therapeutics, InventisBio, and Infinity Pharmaceuticals; grants and personal fees from Novartis, Eisai, EMD Serono, Syros, and Petra Pharma; and personal fees from Ipsen, Relay Therapeutics, MapKure, and Vibliome outside the submitted work. C. Saura reports personal fees from AstraZeneca, Celgene, Daiichi Sankyo, Eisai, F. Hoffmann - La Roche Ltd, Genomic Health, Merck, Sharp and Dohme España S.A, Novartis, Odonate Therapeutics, Pfizer, Philips HealthWorks, Pierre Fabre, prIME Oncology, Puma, Synthon, and Sanofi Aventis outside the submitted work. V. Gambardella reports research funding from Bayer, Boehringer, and Roche, and institutional funding from Genentech, Merck Serono, Roche, BeiGene, Bayer, Servier, Lilly, Novartis, Takeda, Astellas, FibroGen, Amcure, Natera, Sierra Oncology, AstraZeneca, Medimmune, BMS, and MSD. M.R. Patel reports other from Genentech during the conduct of the study, and other from Genentech, Pfizer, EMD Serono, Pharmacyclics, Janssen, and Bayer outside the submitted work. C.X. Ma reports other from Washington University in St. Louis during the conduct of the study; personal fees from Eisai, Athenex, AstraZeneca, Seattle Genetics, Novartis, and Eli Lilly; and grants and personal fees from Puma and Pfizer outside the submitted work. R. Aljumaily reports other from Roche during the conduct of the study, and grants from Alliance Foundation Trials, LLC, Boston Biomedical, Inc, Syneos Health, Array BioPharma, Bristol-Myers Squibb, Huntsman Cancer Institute, Merck Co., AstraZeneca, AbbVie Inc., Regeneron, G1 Therapeutics, Inc., F. Hoffman-LA Roche AG, Genentech, Inc., MedImmune, LLC, GlaxoSmithKline, Novartis, Peloton Therapeutics, Inc, Baxalta, Eli Lilly and Company, EMD Serono Inc., Boehringer Ingelheim, TERSARO, Inc., Pfizer Inc., Checkpoint Therapeutics, Inc., and Eli Lily outside the submitted work. P.L. Bedard reports grants from Roche/Genentech during the conduct of the study, and grants from Roche/Genentech, BMS, AstraZeneca, Lilly, Seagen, Merck, Pfizer, Zymeworks, Servier, Mersana, Novartis, PTC Therapeutics, Amgen, and Sanofi outside the submitted work, and reports advisory board membership (uncompensated) with Roche/Genentech, Merck, Seagen, Lilly, Amgen, Sanofi, and Pfizer. J.C. Sachdev reports other from Genentech during the conduct of the study, and personal fees from Pfizer, Novartis, Ipsen, and Tempus outside the submitted work. L. Dunn reports personal fees from Merck, CUE-Biopharma, Pfizer, Regeneron, and Eisai outside the submitted work. J. Bond reports personal fees and nonfinancial support from Genentech during the conduct of the study and personal fees and nonfinancial support from Genentech outside the submitted work. H.M. Savage reports other from Roche/Genentech outside the submitted work. M. Scaltriti reports grants from AstraZeneca during the conduct of the study; grants from Menarini Ricerche, Immunomedics, TargImmune, and Puma Biotechnologies; and grants and personal fees from Daiichi Sankyo outside the submitted work. T.R. Wilson reports other from Genentech, Inc and Roche during the conduct of the study, as well as a patent for methods of treating with taselisib pending. M.C. Wei reports other from Genentech and F. Hoffman La-Roche Ltd during the conduct of the study. D.M. Hyman reports other from Loxo Oncology/Lilly during the conduct of the study. No disclosures were reported by the other authors.

K. Jhaveri: Conceptualization, resources, data curation, supervision, validation, investigation, visualization, methodology, writing-original draft, writing-review and editing. M.T. Chang: Conceptualization, data curation, software, formal analysis, funding acquisition, validation, investigation, visualization, methodology, writing-original draft, project administration, writing-review and editing. D. Juric: Conceptualization, resources, validation, investigation, visualization, writing-review and editing. C. Saura: Conceptualization, resources, data curation, supervision, validation, investigation, visualization, writing-original draft, writing-review and editing. V. Gambardella: Conceptualization, resources, data curation, supervision, validation, investigation, visualization, writing-review and editing. A. Melnyk: Conceptualization, resources, data curation, validation, investigation, visualization, writing-review and editing. M.R. Patel: Conceptualization, resources, data curation, validation, investigation, visualization, writing-review and editing. V. Ribrag: Conceptualization, resources, validation, investigation, visualization, writing-review and editing. C.X. Ma: Conceptualization, resources, data curation, supervision, validation, investigation, visualization, project administration, writing-review and editing. R. Aljumaily: Conceptualization, resources, data curation, supervision, validation, investigation, visualization, writing-review and editing. P.L. Bedard: Conceptualization, resources, validation, investigation, visualization, writing-review and editing. J.C. Sachdev: Conceptualization, resources, validation, investigation, visualization, methodology, writing-review and editing. L. Dunn: Conceptualization, resources, validation, investigation, visualization, writing-review and editing. H. Won: Conceptualization, resources, formal analysis, validation, investigation, visualization, writing-review and editing. J. Bond: Conceptualization, data curation, software, formal analysis, supervision, funding acquisition, validation, investigation, visualization, methodology, writing-original draft, project administration, writing-review and editing. S. Jones: Conceptualization, data curation, software, formal analysis, funding acquisition, validation, investigation, visualization, methodology, writing-original draft, project administration, writing-review and editing. H.M. Savage: Conceptualization, data curation, software, formal analysis, funding acquisition, validation, investigation, visualization, methodology, writing-original draft, project administration, writing-review and editing. M. Scaltriti: Conceptualization, resources, validation, investigation, visualization, writing-review and editing. T.R. Wilson: Conceptualization, data curation, software, formal analysis, supervision, funding acquisition, validation, investigation, visualization, methodology, writing-original draft, project administration, writing-review and editing. M.C. Wei: Conceptualization, data curation, software, formal analysis, supervision, funding acquisition, validation, investigation, visualization, methodology, writing-original draft, project administration, writing-review and editing. D.M. Hyman: Conceptualization, resources, data curation, validation, investigation, visualization, writing-review and editing.

The authors thank the patients and their families for participation in this study and all participating study sites and staff. The authors acknowledge the Memorial Sloan Kettering Cancer Center support grant (P30 CA008748) and David Pfister (MSKCC, New York, NY), Alan Ho (MSKCC, New York, NY), Jing He (Genentech, Inc.), Jerry Hsu (Genentech, Inc.), Thomas Stout (Genentech, Inc.), Joseph Ware (Genentech, Inc.), Alison Cardenas (Genentech, Inc.), and Paul Ku (Genentech, Inc.) for their contributions. All authors received medical writing support for this manuscript from F. Hoffmann-La Roche Ltd. Support for third-party writing assistance was furnished by Islay Steele, PhD, and Stephen Salem, BSc, of Health Interactions. This study was funded by F. Hoffmann-La Roche Ltd and Memorial Sloan Kettering Cancer Center Support Grant (P30 CA008748). Additional support was received from the NIH (P30 CA008748), Cycle for Survival, and the Noona’s Garden Foundation.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.