Triplet Therapy with Palbociclib, Taselisib, and Fulvestrant in PIK3CA-Mutant Breast Cancer and Doublet Palbociclib and Taselisib in Pathway-Mutant Solid Cancers

Cyclin-dependent kinase 4/6 (CDK4/6) inhibitors in combination with endocrine therapy are widely approved for the treatment of advanced hormone receptor–positive (HR⁺), HER2-negative (HER2⁻) metastatic breast cancer, for patients relapsing both on and off adjuvant endocrine therapy (1, 2). Yet despite high initial response rates, the majority of patients develop disease progression on treatment, and a further subset of cancers are intrinsically resistant to CDK4/6 inhibitor and endocrine therapy combinations. Furthermore, translating the efficacy of CDK4/6 inhibitors to other solid tumors has been challenging. Identifying new treatment approaches to enhance the efficacy of CDK4/6 inhibitors is therefore a major priority.

In parallel to the development of CDK4/6 inhibitors, multiple agents against the PI3K–AKT–mTOR pathway have entered clinical practice, with the mTOR inhibitor everolimus being the first approved in combination with exemestane (3). The pan-PI3K inhibitor buparlisib modestly improved progression-free survival (PFS), albeit with a significant increase in toxicity (4). The more selective beta-sparing PI3K inhibitor taselisib improved PFS when combined with fulvestrant (5), and recently the alpha-selective PI3K inhibitor alpelisib in combination with fulvestrant significantly improved PFS for patients with PIK3CA-mutant (PIK3CA-mt) HR⁺/HER2⁻ cancers, was approved by the FDA, and has become a standard of care (6).

Multiple preclinical studies have demonstrated synergy between PI3K and CDK4/6 inhibition in estrogen receptor– positive (ER⁺) breast cancer (7, 8). PI3K inhibition increases ER transcriptional activity (9), which may be partly overcome by CDK4/6 inhibition. Conversely, CDK4/6 inhibition alone results in incomplete cell-cycle arrest and early adaptation to CDK4/6 inhibition, and so addition of PI3K inhibition leads to more profound cell-cycle arrest (7, 8), also with induction of apoptosis observed (8). Combinations of CDK4/6 and PI3K inhibitors have demonstrated substantial efficacy in vivo (10–12). Furthermore, the triplet combination of CDK4/6 inhibitor, PI3K inhibition, and ER degradation with fulvestrant ablates clonal growth when compared with the doublet combinations (8), also with activity confirmed in vivo (10). Combinations of PI3K and CDK4/6 inhibitors have also demonstrated antitumor potential in other cancer types, with activity observed in preclinical models of PIK3CA-mt triple-negative breast cancer (TNBC; ref. 11) and other solid tumors (13, 14).

This phase Ib trial investigated the activity of triplet therapy, with the combination of CDK4/6 inhibitor palbociclib plus PI3K inhibitor taselisib and fulvestrant in PIK3CA-mt ER⁺/HER2⁻ advanced breast cancer, and doublet therapy of palbociclib plus taselisib in advanced solid tumors with pathway mutations. Detailed translational studies were conducted to assess pharmacodynamic (PD) markers of PI3K inhibition, with potential biomarkers of sensitivity and resistance tested in both tissue and plasma circulating tumor DNA (ctDNA).

From March 2015 to November 2018, a total of 78 patients were enrolled between part A (escalation phase) and part B (expansion phase; Fig. 1A), with the main demographic characteristics detailed in Table 1. Patients had received a median of 3 prior lines of therapy for advanced disease, with a median of 2 prior chemotherapy lines for advanced disease. In part B1, all 25 patients had received prior aromatase inhibitor (AI), 6 of 25 (24%) patients had received prior fulvestrant, and 10 of 25 (40%) patients had received prior everolimus. A summary of mutation distribution as per enrollment can be also found in Table 1. Patients consented to provide both tissue and blood samples for pharmacokinetics (PK), PD, and biomarker analyses, detailed in Fig. 1B.

Part A enrolled 20 patients with solid tumors in a 3+3 dose-escalation trial design (Supplementary Fig. S1A and S1B). No dose-limiting toxicities (DLT) were observed at dose level 1 (n = 4). At dose level 2 (taselisib 4 mg with palbociclib 100 mg, n = 3), 2 patients experienced 3 DLTs of G3 mucositis, G3 hyperglycemia, and G3 fatigue, and the combination dose was deemed intolerable. There were no DLTs at dose level 3 (taselisib 2 mg 3 week-on/1 week-off with palbociclib 125 mg, n = 7) and dose level 4 (taselisib 2 mg continuous dosing with palbociclib 125 mg, n = 6). The recommended phase II dose (RP2D) was established at taselisib 2 mg orally once daily continuous dosing, plus palbociclib 100 mg orally once daily 3-weeks-on, 1-week-off for cycle 1, escalating to 125 mg orally once daily 3-weeks-on, 1-week-off in the absence of grade 3/4 myelosuppression.

Combination palbociclib and taselisib PK was consistent with single-agent PK, suggesting no drug–drug interaction (Supplementary Fig. S2A–S2C). In part A, platelet-rich plasma (PRP) was analyzed from all 20 patients at baseline and on-treatment for AKT and GSK3β. At the 2 mg taselisib dose (at dose levels 3 and 4), there was decreased AKT and GSK3β phosphorylation at cycle 1 day 1 (C1D1) 4 hours, C1D15 6 hours and 8 hours (Fig. 1C). We noted an increase in phosphorylation of AKT and GSK3β at C1D8 predose, which may indicate compensatory upregulation often seen in PI3K–AKT pathway inhibition, and may potentially suggest incomplete PI3K pathway inhibition. In part A, 5 patients had their tumor tissue samples analyzed. Grouping all patients, there is a significant approximately 25% decrease in the ratio of phosphoSer780-Rb to total Rb at C1D15 predose when compared with C1D1 predose, indicating CDK4/6 signaling was modulated as predicted for palbociclib treatment (Fig. 1D).

The most frequently reported all-grade adverse events (AE) across the 58 patients enrolled in the expansion cohorts (all treated with RP2D) were neutropenia (88%), thrombocytopenia (57%), anemia (48%), fatigue (45%), leukopenia (36%), and diarrhea (34%). The most common grade 3/4 AEs in the 58 expansion-phase patients were neutropenia (53%) and leukopenia (21%), while the most common grade 3/4 treatment-related AEs were neutropenia (53%) and leukopenia (19%). A complete record of all adverse events reported in >10% of all patients across the trial can be found in Table 2. Overall, 30.8% of patients (24/78) had 35 serious adverse events (SAE) of which 25.7% (9/35) were considered as definitely, probably, or possibly related to treatment. There were no treatment-related deaths. Dose interruptions were required for taselisib in 45 of 78 patients (58%) and for palbociclib in 39 of 78 patients (50%). In part B1, 1 of 25 patients (4%) required interruption for fulvestrant, and in part B3 4/7 (57.1%) for letrozole. Dose reductions for taselisib were required in 13 of 78 (17%) patients enrolled, and 15 of 78 (19%) for palbociclib. There were delays in 28 of 78 patients (36%) for taselisib and in 41 of 78 (53%) for palbociclib. Treatment was discontinued in 54 of 77 patients (70%) following radiographic progression, 13 of 77 (17%) following clinical progression, 3 of 77 (4%) due to adverse events, and 7 of 77 (9%) due to other factors.

In part B1, 25 patients with PIK3CA-mt ER⁺ HER2⁻ advanced breast cancer were treated with the triplet of palbociclib plus taselisib plus fulvestrant; one patient was not assessable for the primary endpoint as the patient had not completed first cycle of treatment. The confirmed objective response rate (ORR) was 37.5% [9/24, 95% confidence interval (CI), 18.8–59.4, Fig. 2A]. Clinical benefit rate (CBR) was 58.3% (14/24; 95% CI, 36.6–77.9), with median PFS of 7.2 months (95% CI, 3.9–9.9; Fig. 2B).

Part B2 recruited a total of 26 patients. Twelve patients with PIK3CA-mt ER⁻ advanced breast cancer (9 TNBC and 3 HER2⁺) were treated with taselisib plus palbociclib, and 10 of them were evaluable for efficacy (8 TNBC and 2 HER2⁺). The ORR was 10% (1/10, 95% CI, 0.2–26.5), CBR 30% (3/10, 95% CI, 6.7–65.2), and median PFS was 3.6 months (95% CI, 1.7–5.6; Fig. 2C). There were 14 patients with solid tumors of various subtypes treated in part B2: 4 PIK3CA-mt colorectal adenocarcinoma, 3 PIK3CA-mt ER⁺/HER2⁻ breast cancer (included as part of the solid tumors subcohort), 1 PIK3CA-mt high-grade serous ovarian cancer, 1 PIK3CA-mt clear cell ovarian carcinoma, 2 PIK3CA-mt non–small cell lung cancer (NSCLC), 1 PIK3CA-mt endometrial cancer, 1 PIK3CA-mt cervical cancer (adenosquamous carcinoma), and 1 PIK3CG-mt anaplastic oligodendroglioma. For the 12 evaluable solid tumors in part B2, ORR was 0% (0/12, 95% CI, 0–26.5) and CBR 16.7% (2/12, 95% CI, 2.1–48.4), with median PFS of 3.8 months (95% CI, 1.2–5.0; Supplementary Fig. S3A).

Part B3 recruited 7 patients, 6 of whom were evaluable. CBR was 50% (3/6, 95% CI, 11.8–88.1) and median PFS 8.6 months (95% CI, 1.6–24.2); ORR was not assessable (ORR 0% (0/6, 95% CI, 0–45.9) as patients did not require measurable disease for B3.

Overall, 57 archival formalin-fixed paraffin-embedded (FFPE) blocks (54 baseline + 3 on-progression) were retrieved and had tumor content for translational analysis (Fig. 1B). These consisted in 5 samples from the patients with breast cancer enrolled in part A (2 ER⁺/HER2⁻, 1 ER⁺/HER2⁺, 1 TNBC baseline + progression), 23 samples from patients in part B1 (all ER⁺/HER2⁻), 26 samples from patients coming from part B2 (8 baseline + 1 progression TNBC, 3 baseline + 1 progression ER⁻/HER2⁺ breast cancer, 3 ER⁺/HER2⁻ breast cancer, 4 colorectal cancer, 1 NSCLC, 1 endometrial carcinoma, 1 cervical adenosquamous carcinoma, 1 high-grade serous ovarian carcinoma, 1 clear cell ovarian carcinoma, and 1 anaplastic oligodendroglioma), and 3 samples from patients in part B3. Biopsies had been taken a median of 4.67 years before first dosing of trial medication, 54% coming from metastatic and 46% from primary tumor specimens.

To assess tumor copy number, ultra-low whole genome sequencing (ulWGS) was conducted on baseline FFPE samples, with a mean coverage of 0.49× and mean 1.91 ploidy for 43 evaluable patients, 2 of them with paired on-progression samples (Fig. 1B). For patients with PIK3CA-mt ER⁺/HER2⁻ breast cancer treated with the triplet combination in part B1 and tumor-sequencing data, 6 of 17 patients (35%) had CCND1 amplification, 9 of 17 patients (53%) had FGFR1 amplification, and 2 of 17 patients (12%) had CCNE1 amplification (Fig. 3A). There was no evident association found between copy-number aberrations and efficacy, supported by further analysis of cyclin D1 amplification by FISH assessment (Supplementary Fig. S4A and S4B). No association with efficacy was observed with copy-number aberrations for patients with PIK3CA-mt ER⁻ breast cancer or other solid tumors (Supplementary Fig. S4C). Paired baseline/progression copy-number variation (CNV) analysis for the two available patients showed mostly chromosomal stability, with evidence of homozygous deletion of RB1 in a patient with ER⁻/HER2⁺ breast cancer (Supplementary Fig. S4D).

Baseline tissue samples were analyzed for cyclin D1 and cyclin E1 expression by IHC. In the 42 archival FFPE tissue samples from patients with breast cancer analyzed for cyclin D1, the median H-score was 140 (0–240), and there was no association with PFS in part B1 or in patients with ER⁻ breast cancer in part B2 (Supplementary Fig. S4E and S4F). In the 42 breast cancer samples with cyclin E1 immunostaining, the median percentage of tumor cells stained was 30% (0%–90%; Fig. 3B). In B1, patients with high cyclin E1 expression had worse PFS than patients with low cyclin E1 expression (median PFS 6.4 vs. 10.4 months, respectively; HR = 4.2; 95% CI, 1.3–13.1; P = 0.02, log-rank test, n = 21; Fig. 3C).

Within patients with PIK3CA-mt TNBC, 8 tumors were analyzed by androgen receptor (AR) immunostaining. Median H-score was 140 (0–270) and median percentage of stained cells was 70% (0%–90%). Overall, 6 of 7 evaluable patients (86%) had positivity for AR. The two patients with long-term clinical benefit, and a third additional patient achieving the only RECIST 1.1 confirmed partial response (PR) in the TNBC subset, had high AR expression in their tumors, with median 270 versus 90 H-score for patients achieving CBR versus non-CBR, respectively (P = 0.20, Mann–Whitney test; Fig. 3D).

A total of 40 C1D1–C2D1 ctDNA pairs were analyzed for early PIK3CA ctDNA dynamics, with 10 pairs excluded for failing quality control (see Methods). Comparing allele frequencies (AF) between C2D1 and C1D1, the median ctDNA ratio-day 28 (CDR₂₈) was 0.74 (range 0–3.46), decreased from baseline (P < 0.01, Wilcoxon signed-rank test; Fig. 4A). In the 16 evaluable patients with PIK3CA-mt ER⁺/HER2⁻ breast cancer in part B1 receiving triplet therapy, on-treatment ctDNA was suppressed (P < 0.01, Wilcoxon signed-rank test; Fig. 4B), with median CDR₂₈ AF 0.31 (0–3.46). Patients without supressed CDR₂₈ (>0.1) had worse PFS than patients with suppressed (<0.1) CDR₂₈ (6.9 vs. 15.2 months, respectively; HR = 5.2; 95% CI, 1.4–19.4; P = 0.04, log-rank test; Fig. 4C). Early changes in ctDNA by mutant copies/ml were consistent with analysis by AF (Supplementary Fig. S5A–S5C). Only two patients were found to derive complete ctDNA suppression (C1D1⁺ and C2D1 under limit of detection), both in part B1. Patients with incomplete versus complete ctDNA suppression had worse PFS (6.9 vs. 15.2 months respectively; HR = 5.1; 95% CI, 1.3–20.6; P = 0.02, log-rank test; Supplementary Fig. S5D).

To investigate genomic evolution through therapy, we performed paired error-corrected ctDNA sequencing with a 200 cancer driver gene panel (detailed in Methods) at C1D1–end of treatment (EOT) in 23 patients, including 16 patients from part B1, 4 patients from B2 (1 PIK3CA-mt endometrium, 2 PIK3CA-mt ER⁺/HER2⁻ breast, and 1 PIK3CA-mt ER⁻/HER2⁺ breast), and 3 from B3 (2 PIK3CA-mt and 1 PIK3CA nonmutant ER⁺/HER2⁻ breast). From these, 20/23 (87%) had confirmed progression on their EOT sample. Median family coverage for all sequenced samples was 696X (11–1880).

Overall, 5 of 23 (22%) patients were found to have multiple PIK3CA mutations in any of the timepoints sequenced. Dynamics for PIK3CA mutations in the 4 patients with at least 1 read in each timepoint clearly suggested the second PIK3CA mutation was subclonal in 2 patients (Fig. 5A). In part B1, 3 of 16 (19%) evaluable sequenced patients had multiple PIK3CA mutations. PFS was shorter for multiple versus single PIK3CA mutants (median PFS 2.7 vs. 8.5 months, respectively; HR = 663.5; 95% CI, 28.8–15286; P < 0.0001, log-rank test; Fig. 5B).

In total, 13 of 23 (57%) patients had an ESR1 mutation at any sequenced timepoint, with 3 of 13 (23%) patients harboring polyclonal mutations. Acquired Y537S (c.1610A>C) and Y537C (c.1610A>G) mutations were identified at EOT in 2 patients, with evidence of deselection of ESR1 mutations in two other patients at EOT (Fig. 5C). In evaluable sequenced part B1 patients, 11 of 16 (69%) had an ESR1 mutation. PFS was not significantly different for patients with ESR1-mutant tumors compared with wild-type (HR = 0.3; 95% CI, 0.1–1.4, P = 0.14, log-rank test), although analysis was underpowered (Supplementary Fig. S6A).

Overall, 5 of 23 (22%) patients were found to have FAT1 mutations (2 with stop-gains, 3 with nonsynonymous) in any timepoint (Fig. 5D), with one patient acquiring a FAT1 mutation at EOT that was not present at C1D1 (Fig. 5E). Of the two patients harboring FAT1 stop-gain mutations, one was a patient with ER⁺/HER2⁻ breast cancer treated with the triplet that was censored after 5.5 m for compliance, and the second one was a patient with endometrial cancer treated with doublet that had early progression after only 1.1 m of treatment. We did not find strong evidence for other potential genetic mechanisms of resistance in the paired sequencing.

We provide the first study report of the safety and efficacy of the triplet combination therapy of CDK4/6, PI3K, and ER signalling inhibitors in PIK3CA-mutant ER⁺ breast cancer, as well as the first data on combination palbociclib plus taselisib therapy in a range of different PIK3CA-mutant solid tumors.

We observed a 37.5% response rate for the triplet combination in PIK3CA mutant ER⁺ HER2⁻ breast cancer. In a single-arm study of a triplet combination, it is very hard to confidently differentiate the contributions of the components, although this response rate compares favorably with prior data on palbociclib and fulvestrant doublet. The PALOMA-3 registration trial reported a response rate of 15% overall, and 23% in patients with measurable disease (15). PALOMA-3 allowed for 1 line of chemotherapy in the advanced disease, and only 33% of patients treated had received prior chemotherapy. In the comparable part B1 of our study, 79% (19/24) of evaluable patients had prior chemotherapy for advanced disease, with 50% (12/24) receiving 2 or more advanced chemotherapy lines. Also, PALOMA-3 excluded prior fulvestrant and/or everolimus, whereas B1 allowed both and 46% (11/24) evaluable patients had prior fulvestrant, everolimus, or both. There is broad evidence that prior lines of chemotherapy treatment can heavily decrease response rates to subsequent lines of therapy for patients with metastatic breast cancer, with response rates reducing approximately by half in every subsequent line (16–18). Although it is possible that all efficacy in the triplet originated from palbociclib and fulvestrant, our response rate does suggest triplet efficacy. Current standard of care for patients with PIK3CA-mutant ER⁺ HER2⁻ breast cancer who do not relapse on AI is sequential doublets of AI plus CDK4/6 followed by fulvestrant plus PI3K inhibition with alpelisib at progression. In this setting, a triplet combination will need to be assessed against the sequential therapy approach. Therefore, we suggest that a triplet therapy is likely more relevant in the context of relapse on AI, where fulvestrant plus CDK4/6 inhibition is standard, and there is no clear role for subsequent alpelisib.

The safety profile observed in patients receiving palbociclib 125 mg once daily 3-weeks-on, 1-week-off in combination with taselisib 2 mg daily is consistent with prior studies investigating palbociclib or taselisib in combination with fulvestrant. This triplet combination required a dose reduction in taselisib from the 4 mg dose with fulvestrant, to 2 mg in the triplet, and this may have compromised efficacy. It is possible that treatment delays may reduce efficacy of triplet combinations, and triplet combinations should likely be designed to limit the incidence of such delays. Regardless, there was still evidence of AKT pathway modulation in PRP taken at 4 hours postdose on C1D1 and at 6 to 8 hours postdose on C1D15, providing evidence of PI3K inhibition by taselisib (Fig. 1C). Interestingly, there was evidence of increased AKT phosphorylation in PRP at 24 hours postdose (C1D8 predose), suggesting a rebound of PI3K pathway activation. Our study provides clinical evidence for the activity of triplet therapy, but also suggests that further development of triplet combinations with alternative PI3K or AKT inhibitors may maximize the potential of such triplet therapy by optimising PI3K–AKT pathway suppression. In this study, we show that biomarkers of early disease progression on endocrine–CDK4/6 combinations also predicted for shorter PFS on triplet therapy. High cyclin E expression levels confer resistance to palbociclib in preclinical models (8), and high mRNA levels of cyclin E1 were found to confer relative resistance to palbociclib for patients enrolled in PALOMA-3 (19). Similarly high cyclin E1 associated with short PFS in this study (Fig. 3C).

Early PIK3CA dynamics in ctDNA have been shown to predict patient outcomes in both the PALOMA-3 study of fulvestrant and palbociclib (20) and the BEECH study of capivasertib and paclitaxel (21), both in the ER⁺/HER2⁻ population. Here, we further validated the findings that early ctDNA dynamics could replace other assessments. In addition, patients with ER⁺/HER2⁻ breast cancer with poor ctDNA suppression had worse PFS. Overall ctDNA suppression was less pronounced than in previous analyses, with only 2 patients having absent ctDNA at C2D1. We hypothesize that this might reflect time of sampling, with the C2D1 samples in this study taken after at least 1 week off palbociclib, whereas in our prior work on PALOMA-3 samples were taken at C1D15. It is likely that ctDNA rebound occurred during the week off palbociclib, and that some element of rebound in ctDNA levels was not fully suppressed by taselisib. In this study patients with double PIK3CA mutations had shorter PFS on triplet (Fig. 5B). This contrasts with prior data suggesting that patients with multiple PIK3CA mutations had higher response rates on taselisib at a higher dose (22). Potentially this may reflect the lower dose of taselisib in the triplet, resulting in incomplete PI3 kinase inhibition in tumors with hyperactive multiple mutations.

ctDNA sequencing demonstrated ongoing cancer evolution during triplet combination therapy, with acquisition of a FAT1 mutation in one patient, selection and loss of a second double PIK3CA mutation, and ongoing evolution of ESR1 mutations. Our findings are consistent with prior observations that loss of the FAT1 tumor suppressor promotes CDK4/6 resistance through CDK6 overexpression via the Hippo pathway (23) and that ESR1 Y537S and Y537C are acquired through fulvestrant therapy (24). In this study, only 1 of 23 (4.3%) patients was found to have an RB1 mutation, and this was already present at C1D1, matching previous observation in the PALOMA-3 trial that RB1 acquisition on palbociclib is a rather uncommon mechanism of resistance (24). We did not identify any further mechanisms of resistance with plasma sequencing, including no evidence of acquired PTEN mutations to suggest bypass PI3Kβ activation (25). This potentially suggests that biomarkers of CDK4/6 inhibitor resistance may also predict relative resistance to triplet combinations, identifying an important priority for future triplet studies. However, we do note that more potent PI3K inhibition may generate different results.

We acknowledge a number of limitations in our study. First, our patient population is too small to make any robust conclusions in antitumor efficacy for the palbociclib and taselisib doublet combination. However, although only one confirmed RECIST 1.1 response was observed in a patient with PIK3CA-mutant TNBC, a subset of patients with TNBC potentially characterized by high AR expression had durable disease control, which correlates with in vitro studies showing higher sensitivity to palbociclib in luminal androgen receptor (LAR) TNBC (11). Interestingly, 6 of 7 patients with PIK3CA-mutant TNBC in our study had some degree of IHC expression for AR, a finding that adds evidence to TNBC LAR subtypes being enriched for PIK3CA mutations (26, 27). Previous studies have demonstrated the potential of nonchemotherapy approaches for selected TNBC subtypes, with the potential for antiandrogen therapy in TNBC with AR expression (28, 29). Our study suggests that targeted nonchemotherapy combinations could have activity in subtypes of biologically distinct subgroups, and further clinical studies are warranted. We also note that our biomarker analyses are limited by small numbers and are hypothesis-generating, although observations on cyclin E1 IHC (19), AR IHC in TNBC (26, 27), and ctDNA dynamics (20, 21) are all supported by prior evidence.

In conclusion, our study demonstrates safe and manageable toxicity in solid tumors for the combination of palbociclib and taselisib with or without the addition of endocrine therapy. Our data suggest promising efficacy for the fulvestrant triplet combination in previously treated patients with PIK3CA-mutant ER⁺/HER2⁻ breast cancer. Our study further supports high expression of cyclin E1 as a poor prognostic marker on palbociclib combinations and reinforces the evidence of early ctDNA changes predicting treatment efficacy. These findings provide preliminary proof-of-concept that supports future combination approaches involving both PI3K and cyclin-dependent pathway inhibitors [NCT03065062, NCT01872260, NCT02088684, NCT04191499].

The phase Ib study to assess the safety, tolerability, and activity of the PI3K inhibitor taselisib (GDC-0032) in combination with palbociclib with the subsequent addition of fulvestrant in PIK3CA-mutant breast cancers or letrozole in advanced breast cancers (PIPA) trial (NCT02389842) was an open-label nonrandomized trial sponsored by the Royal Marsden NHS Foundation Trust (RMH) and The Institute of Cancer Research (ICR). Primary objectives were to identify an RP2D by establishing a combination maximum tolerated dose (MTD) for palbociclib plus taselisib, define the adverse event profile for the doublet combination with or without fulvestrant or letrozole, and to assess the efficacy of the triplet palbociclib, taselisib, and fulvestrant in a cohort of patients with PIK3CA-mt advanced ER⁺/HER2⁻ breast cancer. Secondary objectives were to determine preliminary efficacy in patients with other solid tumors including a cohort of patients with PIK3CA-mt ER⁻ advanced breast cancer, PK, PD characterization in tumor biopsies and PRP, and mechanisms of antitumor response and resistance by analysing tumor biopsy and ctDNA samples.

Patients with previously treated advanced solid tumors enriched for, but not requiring, PIK3CA mutations were included in part A (escalation phase) and treated at different dose levels in a 3+3 dose-escalation trial design (Supplementary Fig. S1A and S1B). Subsequent patients were recruited into part B expansion cohorts (Fig. 1A); part B1 (triplet therapy) included patients with PIK3CA-mt, ER⁺/HER2⁻ advanced breast cancer who received at least one prior line of endocrine therapy and up to two lines of chemotherapy (initially open to any number of prior lines of chemotherapy, but this was amended after 15 patients were recruited). Prior exposure to fulvestrant and/or everolimus was permitted but prior exposure to CDK4/6 inhibitors was not. Part B2 (doublet therapy) was composed of patients with PIK3CA-mt, ER⁻ (whether HER2⁺ or HER2⁻) advanced breast cancer or any advanced solid tumor with mutations leading to a hyperactivated PI3K–AKT–mTOR pathway. Patients with ER⁻/HER2⁺ cancer were eligible if they had exposure to at least two prior lines of anti-HER2 therapy (one prior line permitted if no further HER2-directed therapy was available locally), whereas patients with TNBC and solid tumors were eligible with a minimum of one prior line of chemotherapy exposure. Part B3 included patients with ER⁺/HER2⁻ advanced breast cancer without regard for PIK3CA status to assess the safety of palbociclib plus taselisib plus letrozole.

PIK3CA mutations testing was performed by an accredited laboratory on archival or fresh tumor samples or ctDNA in plasma. Patients in parts A, B2, and B3 required measurable disease assessed by RECIST 1.1 or evaluable disease. For patients in part B1, measurable disease was mandatory. Written informed consent was obtained from the patients. The study was approved by an institutional review board and conducted in accordance with the Declaration of Helsinki.

For dose expansions, palbociclib was administered at 100 mg orally once daily 3-weeks-on, 1-week-off for cycle 1, escalating to 125 mg orally once daily 3-weeks-on, 1-week-off from cycle 2 in absence of grade 3/4 myelosuppression, plus taselisib 2 mg orally once daily, plus fulvestrant 500 mg IM or letrozole as per standard schedule. Treatment was given until patient asked to be withdrawn from the trial, evidence of disease progression, or unacceptable toxicity. In parts B1 and B3, pre-/perimenopausal women had ovarian suppression with the luteinizing hormone-releasing hormone agonist goserelin.

Collection of blood samples for PK analysis in part A and part B is illustrated in Fig. 1B and plasma analysed with fully validated analytic methods. Pharmacokinetic parameters were derived from noncompartmental analysis (Phoenix 64). PD analysis (pSer473 and total AKT, and pSer9 and total GSK3β) was performed in PRP collected in part A on Cycle 1 Day 1 (predose and 4 hours postdose), Cycle 1 Day 8 (predose), Cycle 1 Day 15 (6 and 8 hours postdose), and Cycle 2 Day 15 (2 hours postdose), as reported previously (30). Ratios of phosphorylated biomarker as percentage of C1D1 predose normalized to total biomarker as percentage of C1D1 predose and platelet counts were performed. Optional snap-frozen tumor tissue samples for PD were lysed and analysed for phosphorylated Ser780 and total Rb on C1D1 (predose/baseline), C1D15 (predose), and at disease progression using ICR validated assays on the MesoScale Discovery (MSD) technology platform.

All patients provided an archived FFPE sample from either the primary or a metastatic location (Fig. 1B). Cyclin D1 IHC staining used specific rabbit mAb SP4 clone (Thermo Scientific, catalog number RM-9104-S1) and reported as H-score. Cyclin D1 FISH staining used Vysis CCND1/CEP11 FISH Probe Kit (Abbott Molecular, catalog number 03N88020) and reported as counted ratio CCND1/CEP11. Cyclin E1 IHC used specific mouse mAb HE12 (Abcam, catalog number ab3927) and scored as percentage of invasive nuclei stained. For patients with TNBC, AR IHC staining was performed using the specific mouse mAb AR441 clone and scored as percentage of invasive nuclei stained.

DNA extraction from FFPE was performed using QIAGEN AllPrep DNA/RNA FFPE Kit using MiniColumns (Qiagen catalog number: 80234) and quantified using an RNAseP assay for droplet digital PCR (ddPCR) as described previously (31).

Blood samples were collected in EDTA or STRECK tubes before each cycle of treatment (∼28 days) from C1D1 to the end of treatment (Fig. 1B). Plasma and buffy coat were separated by centrifugation for 20 minutes at 1,600 × g at room temperature at the ICR, and stored immediately at −80°C until nucleic acid extraction. EDTA tubes were processed within 2 hours following venipuncture and STRECK tubes within 48–72 hours. Plasma DNA was extracted from 4 mL of sample (for C1D1-EOT pairs) or 2 mL (for C2D1) using the MagMAX Cell-Free DNA Isolation Kit (Thermo Scientific, catalog number A29319) on a KingFisher Flex Purification System as per the manufacturer's instructions.

For ultra-low passage whole-genome sequencing (ulWGS) on tissue DNA (Fig. 1B), libraries were prepared from 5–50 ng input using Illumina's KAPA HyperPlus Kit with Library Amplification (catalog no. 07962428001) and quantified with Illumina's Library Quantification Kit ABI Prism qPCR Master Mix (catalog no. 07960204001) as per the manufacturer's instructions, and sequenced on a NovaSeq 6000, using SP 300 cycles reagents to a target depth of 0.8X. Data were analyzed for CNVs by alignment with reference hg19 genome using BWA (RRID:SCR_010910), duplicate removal with Picard (RRID:SCR_006525, http://broadinstitute.github.io/picard/); purity assessment and copy-number calling used ichorCNA (32) with settings for high tumor fraction to get a thresholded adjusted copy-number call (amplification/gain/neutral/heterozygous deletion/homozygous deletion) with adjusted values plotted for each sample. Copy-number evaluations were limited to tumors with at least 20% tumor purity.

For paired C1D1 and EOT plasma sequencing (Fig. 1B), an in-house targeted error-corrected capture panel targeting 200 cancer driver genes (RMH200, 1.2 Mb size) was used (Supplementary Table S1). Briefly, IDT xGen Dual Index UMI adapters were ligated with KAPA HyperPlus Kit, and sequenced to target depth 20,000X on a NovaSeq 6000 Sequencing System (Illumina). Samples were aligned using BWA (33) then annotated and combined into UMI consensus families using fgbio (http://fulcrumgenomics.github.io/fgbio/). Sixteen plasma samples from healthy donors were sequenced to generate an error model using a modified pileup pipeline (34). Variant calls were generated using VarDict (35) and pileup (35). Four alternative families of size 3 or more reads were required to call a mutation (5 alternative families for an indel) removing mutations with significant strand bias. All calls were manually curated with Integrative Genomics Viewer (36), reported as AF and annotated using annovar (37). All sequencing was performed on the NovaSeq 6000 (Illumina). Functional predictions of FAT1 mutations were assessed using SIFT (RRID:SCR_012813; ref. 38), MutationAssessor (RRID:SCR_005762; ref. 39), and PROVEAN (RRID:SCR_002182; ref. 40).

A ctDNA ratio (CDR) for PIK3CA mutations between C1D1 and C2D1 (CDR₂₈ – day 28) was performed by ddPCR (Fig. 1B). Singleplex PIK3CA assays were used for common mutations E542K (c.1624 G>A), E545K (c.1633 G>A), H1047R (c.3140 A>G), and H1047L (c.3140 A>T) located in exons 9 and 20, as described previously (20). For patients with multiple PIK3CA mutations, the mutations in exon 9 or 20 were used for CDR analysis. The CDR₂₈ was calculated as both the PIK3CA-mutant copies per ml and the allele frequency at C2D1 relative to C1D1.

Part B1 followed a two-stage minimax Simon design, with an unacceptable response rate of 0.1 and an acceptable response rate of 0.3 (with a one-sided α = 0.05 and β = 0.2). Fifteen patients were to be treated in stage one and if two or more responses were observed 25 patients in total were recruited with 6 responses in total required to infer efficacy. Part B2 was exploratory and intended to allocate up to 38 patients. Data for this part is presented separate for patients with ER⁻ breast cancer and other solid tumors. Part B3 planned to recruit 6 patients reflecting a 3+3 design. Safety analyses were performed in all patients receiving any study medication at any dose level. All efficacy and biomarker studies were conducted in the response-evaluable population, which includes any enrolled patient meeting all eligibility criteria, receiving at least one cycle of trial medication and having baseline assessment of disease. Primary efficacy endpoint was confirmed ORR by RECIST 1.1. CBR was defined as complete response, PR, or stable disease lasting ≥24 weeks. PFS was defined as the time from first dose (C1D1) until progression by RECIST 1.1 or death from any cause. For ORR, CBR, and PFS analysis, 95% CIs were estimated. Log-rank test was used to test associations with PFS, with HRs calculated using the Mantel-Haenszel method. For CDR₂₈, nonparametric Wilcoxon matched-pairs signed-rank test was used to detect differences between C2D1 and C1D1 concentration. A P value <0.05 was considered for statistical significance in all analysis. Statistics were performed using GraphPad PRISM version 8.2.1 and Stata version 15.

J. Pascual reports grants from Spanish Society of Medical Oncology (SEOM) and NIHR Biomedical Research Centre during the conduct of the study. J.S.J. Lim reports grants from National Medical Research Council Singapore (research training fellowship; NMRC/Fellowship/0024/2015) during the conduct of the study; personal fees and nonfinancial support from Pfizer, Astrazeneca; personal fees from Novartis, and grants from CTI biopharma outside the submitted work. I.R. Macpherson reports personal fees from Novartis, Pfizer, Daiichi Sankyo, Genomic Health, Pierre Fabre, MSD; personal fees and nonfinancial support from Roche, Eli Lilly, and Eisai outside the submitted work. A.C. Armstrong reports other from Astrazeneca (spousal shares) outside the submitted work. A. Ring reports personal fees from Lilly, personal fees from Pfizer, personal fees from Roche, MSD, and Novartis outside the submitted work. A.F.C. Okines reports personal fees from Roche; grants from Pfizer; nonfinancial support from Leo pharmaceuticals; and personal fees from Seagen outside the submitted work. M. Hubank reports personal fees from Astrazeneca (advisory board) and Roche Diagnostics (advisory board) during the conduct of the study; personal fees from Bayer (advisory board), Guardant (advisory board), Boehringer Ingelheim (advisory board), Eli Lilley (honorarium), and Bristol Myers Squibb (advisory board) outside the submitted work. L. Finneran reports other from Roche (funding provided to my institution for databse and statistical support for this project) during the conduct of the study; personal fees from AstraZenca (funding provided to my institution to suport my salary) outside the submitted work. J. Malia reports grants from CRUK [CRUK (grant number C309/A11566) support of the Clinical PD Biomarker Group laboratory equipment servicing and maintenance] during the conduct of the study; other from Boehringer Ingelheim Ltd (research funding), AstraZeneca (research funding), Verastem Inc (research funding), and Sierra Oncology Inc (research funding) outside the submitted work; and is an employee of The Institute of Cancer Research, which is involved in the development of PI3K, HSP90, HDAC, AKT, ROCK, RAF, CHK1 and HSF1 inhibitors. K.E. Swales reports grants from CRUK [CRUK (grant number C347/A18077) funds Karen Swales post and CRUK (grant number C309/A11566) support of the Clinical PD Biomarker Group laboratory equipment servicing and maintenance] during the conduct of the study; other from AstraZeneca (research funding), Bayer (research funding), Boston Pharmaceuticals Inc (research funding), Sierra Oncology (research funding), Menarini Ricerche (research funding), and other from Merck (research funding) outside the submitted work; and is an employee of The Institute of Cancer Research, which is involved in the development of PI3K, HSP90, HDAC, AKT, ROCK, RAF, CHK1 and HSF1 inhibitors. A. Turner reports grants and non-financial support from Pfizer “Palbociclib” and grants and nonfinancial support from Hoffman La Roche (taselisib) during the conduct of the study. E. Hall reports nonfinancial support from AstraZeneca (study drug supplied by AstraZeneca; grants and nonfinancial support from Janssen-Cilag (grant received by institution as contributionto support trial health economic evaluation in an academic trial in prostate cancer), Merck Sharpe & Dohme (educational grant to host institution (ICR) to support central trial costs at ICR-CTSU and IMP supplies for an investigator initiated prostate cancer study), Bayer Heathcare Pharmaceuticals Inc (educational grant to ICR to support central trial costs at ICR-CTSU for an investigator initiated cancer study), AstraZeneca (educational grant to ICR to support central trial costs at ICR-CTSU for investigator initiated prostate cancer studies), Accuray Inc (grant received by Institution as contribution to the central trial management of the PACE clinical trial of SBRT in prostate cancer), Aventis Pharma Limited (Sanofi; educational grant to ICR to support central trial costs at ICR-CTSU for an investigator initiated prostate and penile cancer studies); grant funding to support central trial management and running costs at ICR-CTSU) during the conduct of the study; Kyowo Hakko UK (educational grant to ICR to support central trial costs at ICR-CTSU for the BOXIT trial in intermediate/high risk non-muscle invasive bladder cancer), Alliance Pharma (was Cambridge Laboratories; educational grant to ICR to support central trial costs at ICR-CTSU for the BOXIT trial in intermediate/high risk non-muscle invasive bladder cancer); Varian Medical Systems Inc Grant received by institution as contribution to support central trial costs at ICR-CTSU for academic prostate radiotherapy trial), and grants from Roche Products Ltd (grant received by institution as contribution to support central trial costs at ICR-CTSU for an academic bladder cancer trial) outside the submitted work. T.A. Yap grants and personal fees from AstraZeneca (research support to institution; consultant), Bayer (research support to institution; consultant), Clovis (research support to institution; consultant), Constellation (research support to institution; consultant), EMD Serono (research support to institution; consultant), F-Star (research support to institution; consultant), Merck (research support to institution; consultant), Pfizer (research support to institution; consultant); Repare (research support to institution; consultant), Seattle Genetics (research support to institution; consultant); reports grants from Artios (research support to institution), Cyteir (research support to institution), Eli Lilly (research support to institution); Forbius (research support to institution; consultant) (research support to institution), Genentech (research support to institution), ImmuneSensor (research support to institution), Ipsen (research support to institution), Jounce (research support to institution), Karyopharm (research support to institution), Kyowa (research support to institution); Novartis (research support to institution), Ribon Therapeutics (research support to institution), Regeneron (research support to institution), grants from Sanofi (research support to institution), Scholar Rock (research support to institution), Tesaro (research support to institution), Vertex Pharmaceuticals (research support to institution); personal fees from Almac (consultant), Aduro (consultant), Atrin (consultant), Axiom (consultant), Bristol Myers Squibb (consultant), Calithera (consultant), Cybrexa (consultant), Guidepoint (consultant), Ignyta (consultant), I-Mab (consultant), Jansen (consultant), Roche (consultant), Rubius (consultant), Schrodinger (consultant), Varian (consultant), and personal fees from Zai Labs (consultant) outside the submitted work. J.S. Lopez reports grants from Roche (grant funding for this study to the institution), Pfizer (grant funding for this study to the institution) during the conduct of the study; grants and personal fees from Basilea, Genmab; and grants from Roche-Genentech outside the submitted work. N. Turner has received advisory board honoraria from Astra Zeneca, Bristol-Myers Squibb, Lilly, Merck Sharpe and Dohme, Novartis, Pfizer, Roche/Genentech, Bicycle Therapeutics, Taiho, Zeno pharmaceuticals, Repare therapeutics and research funding from Astra Zeneca, BioRad, Pfizer, Roche/Genentech, Clovis, Merck Sharpe and Dohme, and Guardant Health. No disclosures were reported by the other authors.

J. Pascual: Data curation, formal analysis, investigation, visualization, writing-original draft, project administration, writing-review and editing. J.S.J. Lim: Investigation, writing-review and editing. I.R. Macpherson: Investigation, writing-review and editing. A.C. Armstrong: Investigation, writing-review and editing. A. Ring: Investigation, writing-review and editing. A.F.C. Okines: Investigation, writing-review and editing. R.J. Cutts: Data curation, software, investigation, writing-review and editing. M.T. Herrera-Abreu: Investigation, writing-review and editing. I. Garcia-Murillas: Investigation, writing-review and editing. A. Pearson: Investigation, writing-review and editing. S. Hrebien: Investigation, writing-review and editing. H. Gevensleben: Investigation, writing-review and editing. P.Z. Proszek: Data curation, investigation, writing-review and editing. M. Hubank: Data curation, investigation, writing-review and editing. M. Hills: Investigation, writing-review and editing. J. King: Project administration, writing-review and editing. M. Parmar: Data curation, investigation, writing-review and editing. T. Prout: Data curation, investigation, writing-review and editing. L. Finneran: Data curation, investigation, writing-review and editing. J. Malia: Data curation, investigation, writing-review and editing. K.E. Swales: Data curation, investigation, writing-review and editing. R. Ruddle: Data curation, investigation, writing-review and editing. F.I. Raynaud: Data curation, investigation, writing-review and editing. A. Turner: Project administration, writing-review and editing. E. Hall: Data curation, writing-review and editing. T.A. Yap: Conceptualization, supervision, methodology, visualization, investigation, writing-review and editing. J.S. Lopez: Investigation, writing-review and editing. N.C. Turner: Conceptualization, formal analysis, supervision, funding acquisition, investigation, visualization, project administration, writing-review and editing.

These studies were supported by Breast Cancer Now funding to the Breast Cancer Now Research Centre at The Institute of Cancer Research, NIHR funding to the Royal Marsden and Institute of Cancer Research, Pfizer, and Roche. J. Pascual was partially funded by a Spanish Society for Medical Oncology (SEOM) translational research grant and through an NIHR grant. J.S.J. Lim was funded by the Research Training Fellowship grant (NMRC/Fellowship/0024/2015) from the National Research Medical Council, Singapore. This project represents independent research supported by the National Institute for Health Research (NIHR) Biomedical Research Centre at The Royal Marsden NHS Foundation Trust and the Institute of Cancer Research, London. The views expressed are those of the author(s) and not necessarily those of the NIHR or the Department of Health and Social Care. We also acknowledge NIHR/Cancer Research UK Christie Clinical Research Facility. We thank patients and their families for participating in this study.