PI3Kα Inhibitor and Degrader Inavolisib Can Co-opt FGFR2 to Enhance Responses in Patients with PIK3CA-Mutated Solid Tumors and in Preclinical Models

Abstract

Purpose:

PIK3CA mutations frequently drive solid tumors, particularly hormone receptor–positive breast cancer. Inavolisib, an ATP-competitive p110α inhibitor, also promotes the degradation of mutated p110α. PI3K inhibitors have generally shown modest single-agent activity and have safety concerns.

Patients and Methods:

A first-in-human phase 1 study (NCT03006172) evaluated oral inavolisib in patients with PIK3CA-mutated solid tumors to determine the maximum tolerated dose and safety. Correlative analyses included ctDNA. Preclinical studies in cell lines and xenografts elucidated the role of FGFR2.

Results:

The maximum tolerated dose was 9 mg daily, with a manageable safety profile (e.g., hyperglycemia and diarrhea). Inavolisib showed linear pharmacokinetics, consistent pharmacodynamic modulation, and antitumor activity in hormone receptor–positive PIK3CA-mutated breast cancer (26% objective response rate and 45% clinical benefit rate). FGFR2 hotspot mutations in ctDNA were strongly associated with clinical benefit. Preclinically, oncogenic FGFR2 signaling enhanced inavolisib sensitivity by engaging HER3, RAS, and p85β, that facilitated mutated p110α degradation, surpassing nondegrading inhibitors. Combination therapy with FGFR2 inhibitors showed synergy and delayed resistance.

Conclusions:

These findings highlight a novel cooperativity between FGFR2 and p110α that boosts the effectiveness of inavolisib. The data support advancing precision oncology beyond single biomarkers to complex algorithms utilizing co-occurring alterations, suggesting that combining inavolisib with FGFR2 inhibitors may offer enhanced and more durable responses in PIK3CA/FGFR2-altered tumors.

Translational Relevance.

By linking ctDNA-identified mutations with reported clinical benefit from inavolisib, we found that FGFR2 hotspot mutations may drive enhanced dependency through the PI3K pathway. The elucidated cooperativity between FGFR2 and p110ɑ may further contribute to a shift in precision oncology paradigms from a single predictive biomarker to a more complex algorithm based on co-occurring alterations.

Introduction

PI3K is a multidomain lipid kinase that, upon activation by integrins and growth factor receptors, regulates cell proliferation, survival, and migration through the PI3K/AKT/mTOR signaling pathway. There are three classes of PI3K. Class I is the most responsive to external stimuli and comprises two subunits: a regulatory adapter subunit (p85) and a catalytic subunit (p110) inclusive of four isoforms—α, β, δ, and γ. The p110 α isoform (p110α) and p85 subunit make up the PI3Kα complex. The PIK3CA gene, which encodes p110α, can be affected by oncogenic hotspot mutations present in 13% of all solid tumors (1). PIK3CA driver mutations are particularly pronounced in breast cancer, with 35% to 40% of hormone receptor–positive (HR+) and 25% to 30% of human EGFR2-positive (HER2+) breast cancers exhibiting PIK3CA mutations (2, 3). More recently, tumors with two or more variants in PIK3CA have been identified (4–7). When occurring in cis, multi-mutant PIK3CA mutations are associated with increased PI3K pathway dependency, and thus, enhanced sensitivity to pharmacologic inhibition of p110α (8–10). Several PI3K inhibitors are in clinical development. Some have been discontinued because of safety concerns (11), including hyperglycemia, diarrhea, and other gastrointestinal adverse events (AE), skin toxicities, and liver toxicities (12). Moreover, PI3K inhibitors have shown modest activity as single agents, prompting the investigation of combination therapy regimens.

We previously reported that inavolisib, a potent, selective, and ATP-competitive inhibitor of p110α, promotes the degradation of mutated p110α, particularly in HER2+ breast cancer cell line models (13, 14). These data suggested that receptor tyrosine kinase (RTK)-dependent inducible degradation of mutated p110α drives sensitivity to inavolisib in HER2+ tumors (14). Herein, we report clinical results from the first-in-human phase 1 trial of inavolisib in patients with cancers harboring one or more PIK3CA hotspot mutations (PIK3CAmut) and demonstrate that the additional presence of activating mutations in the FGFR2 gene in patients with HR+/HER2− advanced breast cancer is associated with clinical benefit during inavolisib monotherapy. FGFR2 and its three paralogs, FGFR1, FGFR3, and FGFR4, encode the FGFR family of transmembrane RTKs (15). FGF ligands bind these receptors to exert their pleiotropic effects, and oncogenic FGFR signaling mediated by somatic alterations has been implicated in numerous tumor types (16–18). FGFR2-amplified tumors with FGFR2 overexpression promote oncogene addiction through partial cross-talk between FGFR2 and other receptors such as HER3 (19).

We demonstrate that high FGFR2 expression in cell lines is associated with FGFR2 dependence and susceptibility to inavolisib through multiple signaling nodes, including human EGFR3 (HER3), rat sarcoma (RAS), as well as direct binding of FGFR2 to the p85β PI3K regulatory subunit. FGFR2-activated HER3 potentiates the inavolisib degradation mechanism of action, leading to a stronger growth-inhibitory effect as compared with a nondegrading p110α inhibitor. In cells expressing high levels of FGFR2, inavolisib was synergistic with an FGFR2-selective inhibitor and delayed resistance to either single agent. Our findings suggest that FGFR2-dependent tumors may benefit from a combination of inavolisib and an FGFR2-selective inhibitor to further improve clinical outcomes.

Patients and Methods

Clinical study design

This was an open-label, multicenter, multiarm, phase I study (ClinicalTrials.gov identifier: NCT03006172) that included a single-agent (arm A), 3 + 3 dose-escalation design to evaluate the safety, tolerability, and pharmacokinetics (PK) of inavolisib (GDC-0077; RO7113755) administered orally to patients with locally advanced or metastatic PIK3CAmut solid tumors. The primary objective was to determine the maximum tolerated dose (MTD) and/or recommended phase II dose and assess the safety of inavolisib in patients with PIK3CAmut solid tumors. The starting dose was 6 mg, which was below the maximum recommended starting dose calculated from relevant toxicity data from preclinical studies. Patients received their first dose on day 1 in cycle 1 (C1D1), followed by a 7-day washout period and several PK samplings. Patients started once-daily dosing on C1D8; the first cycle was 35 days, and subsequent cycles were 28 days.

The study was conducted in full conformance with the International Council for Harmonisation (ICH) E6 guideline for Good Clinical Practice and the principles of the Declaration of Helsinki, or the applicable laws and regulations of the country in which the research was conducted, whichever afforded greater protection to the individual. The study complied with the requirements of the ICH E2A guideline (Clinical Safety Data Management: Definitions and Standards for Expedited Reporting). The protocol was approved by institutional review boards at participating institutions, and patients provided written informed consent before any study procedures.

Patients

The study enrolled male and female patients with histologically documented and locally advanced, recurrent, or metastatic, PIK3CAmut, incurable solid tumor malignancies, including breast cancer. Patients presented with disease that had progressed after all available standard therapy or for which standard therapy had proven to be ineffective or intolerable, or considered inappropriate. PIK3CAmut tumor status was determined from sponsor central testing of fresh or archival tumor tissue with the PCR-based cobas PIK3CA Mutation Test (Roche) or site-directed local testing of tumor tissue or plasma-derived ctDNA. Study-eligible mutations included H1047R/Y/L, E542K, E545K/D/G/A, Q546K/R/E/L, N345K, C420R, G1049R, R88Q, and M1043I.

Patients were ≥18 years old with evaluable or measurable disease per RECIST v1.1. They had an Eastern Cooperative Oncology Group performance status of 0 or 1; life expectancy of ≥12 weeks; adequate hematologic and organ function within 14 days before study treatment; fasting glucose ≤140 mg/dL and glycosylated hemoglobin (HbA1c) <7%; adequate liver function [bilirubin ≤1.5 × upper limit of normal (ULN), serum albumin ≥2.5 g/dL, aspartate aminotransferase and alanine aminotransferase ≤2.5 × ULN except patients with documented liver metastases]; adequate kidney function (serum creatinine ≤1.5 × ULN or creatine clearance ≥50 mL/minute); International Normalized Ratio (INR) <1.5 × ULN, and aPTT <1.5 × ULN.

Patients were allowed to join the dose-escalation cohorts (arm A) even if they had prior treatment with a PI3K inhibitor. Exclusion criteria were the presence of metaplastic breast cancer, history of leptomeningeal disease, and type 1 or 2 diabetes requiring antihyperglycemic medication. Other exclusion criteria were known and untreated active central nervous system metastases, uncontrolled pleural effusion or ascites requiring recurrent drainage procedures ≥ biweekly, and some eye conditions (concurrent ocular or intraocular condition, inflammatory or infectious conditions, and history of idiopathic or autoimmune-associated uveitis).

Safety

All patients who received any amount of study treatment were included in the safety analyses. AE occurring on or after treatment on day 1 were summarized by mapped term and NCI Common Terminology Criteria for Adverse Events v4.0 toxicity grade.

The dose-limiting toxicity (DLT) assessment window was C1D1 to C1D35. DLT definition included fasting hyperglycemia grade ≥4; fasting hyperglycemia grade ≥3 lasting >7 days despite appropriate oral treatment; fasting hypercholesterolemia grade ≥4; triglyceridemia lasting >14 days despite appropriate intervention; elevated serum hepatic transaminase (alanine aminotransferase or aspartate aminotransferase) grade ≥3 lasting >7 days; elevated serum bilirubin grade ≥3; neutropenia grade ≥4 lasting >7 days; febrile neutropenia grade ≥3; thrombocytopenia grade ≥4; and thrombocytopenia grade 3 associated with clinically significant bleeding or requiring platelet transfusion in the absence of anticoagulation therapy or a condition that presents a bleeding risk. Any nonhematologic, nonhepatic, nonmetabolic AE of grade ≥3 was also considered a DLT with the following exceptions: alopecia; vomiting or nausea grade ≥3 with improvement to grade ≤2 in ≤3 days with treatment; diarrhea grade ≥3 with improvement to grade ≤1 in ≤7 days with treatment; fatigue grade 3 lasting ≤3 days or less than two grade changes from baseline; and asymptomatic grade 3 laboratory abnormalities. During the DLT assessment window, patients who discontinued the study or missed >3 doses of inavolisib were replaced in the absence of a DLT.

PK

The PK-evaluable population consisted of all patients who received at least one dose of the study drug and had ≥1 evaluable postdose PK sample. PK parameters were derived from the plasma concentration–time profile of inavolisib following single-dose and multiple-dose administration using noncompartmental analysis.

Efficacy

The best overall response rate was defined as the percentage of patients who achieved a best overall response of complete response (CR) or partial response (PR) according to RECIST v1.1. The confirmed overall response rate required consecutive assessment of PR or CR >4 weeks after the initial documentation of response. Patients missing response data were classified as nonresponders. Clinical benefit rate was defined as the percentage of patients with confirmed CR, PR, and those with ≥24 weeks' duration of stable disease or non-CR/nonprogressive disease since the first study treatment.

Statistical considerations

The sample size for arm A was based on dose-escalation rules. The sample sizes did not reflect any explicit power and type I error considerations. This study was intended to obtain preliminary safety, PK, pharmacodynamics (PD), and activity information in the safety-evaluable population. For each cohort and dose level, 95% confidence intervals (CI) were calculated for the activity endpoints. Continuous variables were summarized using means, SD, median, and ranges; categorical variables were summarized using counts and percentages. Safety was assessed through summaries of AE, changes in laboratory test results, and changes in vital signs.

2[18F]Fluoro-2-deoxy-D-glucose -PET imaging

2[18F]Fluoro-2-deoxy-D-glucose PET (FDG-PET) scans were acquired before treatment (baseline) and after 2 weeks of once-daily treatment (C1D22-28, or C1D15-22 for backfill patients). On-treatment scans were acquired 1 to 4 hours after inavolisib dosing, with a blood glucose requirement of <180 mg/dL before all scans. Quantitative assessment of scans was performed centrally by one radiologist. Up to five target lesions with a tumor-to-background ratio of at least 2:1 and a longest diameter of at least 15 mm were selected at baseline. Partial metabolic response (PMR) was defined as a decrease of >25% in the mean percentage change in the maximum standardized uptake value (lean body mass–corrected) of the target lesions and no new FDG-avid lesions.

IHC analysis of paired tumor tissue samples

Paired formalin-fixed and paraffin-embedded tumor tissue samples collected from patients before treatment (baseline) and during treatment (C1D15) underwent IHC analysis with antibodies against Ki67 (rabbit mAb, clone 30-9, Ventana/Roche Diagnostics, cat. No. 790-4286, RRID: AB_2631262), phosphorylated AKT (pAKT; Ser473; D9E XP rabbit mAb, Cell Signaling Technology, cat. No. 4060, RRID: AB_2315049), and S6 [phospho-S6 ribosomal protein (Ser235/236; D57.2.2E) XP rabbit mAb, Cell Signaling Technology, cat. No. 4858, RRID: AB_916156]. Ki67 was scored as the total percentage of cells with nuclear staining in each sample. Cytoplasmic staining of pAKT and phosphorylated S6 (pS6) was scored by H-score methodology, in which intensity is considered “0” for absent expression, 1+ for weak staining, 2+ for moderate staining, and 3+ for strong staining and H-score = (percentage of cells with absent staining × 0) + (percentage of 1+ cells × 1) + (percentage of 2+ cells × 2) + (percentage of 3+ cells × 3). The percent change in each score within a patient was computed as 100× a patient’s (on treatment − pretreatment)/pretreatment score and displayed as a bar plot to illustrate the change in staining/expression for each marker and sample pair.

Analysis of blood plasma–derived ctDNA from study patients

ctDNA was isolated as previously described (20) from plasma collected immediately before treatment (baseline) and on C1D15 of treatment from 19 study participants (C1D15 only available for 14 participants) for sequencing with the FoundationACT comprehensive genomic profiling (CGP) assay—a laboratory-developed test (LDT) predecessor to the FDA-approved FoundationOne Liquid CDx—in a Clinical Laboratory Improvement Amendments–certified, College of American Pathologists–accredited reference laboratory (Foundation Medicine, Inc.). Comprehensive details on these assay platforms, sequencing, and mutation calling methodologies were previously described (20, 21).

For mutation allele frequency (MAF) dynamics analyses, MAF for PIK3CA is defined as the percentage of mapped reads supporting the variant. For patients with >1 PIK3CA single-nucleotide variant (includes both known pathogenic and unknown PIK3CA mutations), the sum of the MAF of each single-nucleotide variant is used. At baseline, all patients with ctDNA available contained at least one pathogenic PIK3CA mutation that met enrollment criteria.

For the comprehensive analysis to understand genomic features associated with response to inavolisib, ctDNA derived from plasma samples collected at different time points, including C1D1 (baseline, n = 20), on treatment at C1D15 (n = 16), and at study completion (end of treatment, n = 15), was sequenced using the FoundationACT or the updated and similar LDT FoundationOne Liquid assay as described (20, 21). Only samples with at least one detected alteration were included in the analysis. Cell-free DNA genomic profiling of plasma samples using the FoundationOne Liquid assay were performed in a Clinical Laboratory Improvement Amendments–certified, College of American Pathologists–accredited laboratory (Foundation Medicine, Inc.) using hybrid-capture, adapter ligation–based libraries to identify genomic alterations [base substitutions, small insertions and deletions, copy-number alterations (CNA), and rearrangements (RE)/fusion events] for 70 cancer-related genes. Processing of the sequence data and identification of different classes of genomic alterations were performed as previously described (21). Unless otherwise indicated, the analysis focused on alterations predicted to be pathogenic, defined as of known or likely oncogenic significance. Description of analysis of genomic features is described in the “Patients and Methods” subsection “Statistical analyses for translational research.”

Real-world clinical genomics database analyses

Genomic alterations (substitutions, insertions and deletions, CNA, and RE) were identified via next-generation sequencing–based CGP for >300 cancer-related genes (FDA-approved FoundationOne CDx and the LDT FoundationOne; ref. 21). For the analyses herein, we queried the CGP data from 49,834 breast tumors (including 2,626 ER+ breast tumors) sequenced up to June 30, 2023. The analysis of prevalence and co-occurrence of FGFR2 and PIK3CA alterations was performed on the CGP data from 45,847 HER2− breast tumors and 476,781 solid tumor biopsies sequenced up to June 30, 2023.

Statistical analysis for translational research

Statistical analysis, computation, and plotting were performed using R v4.2.0. All statistical tests were two-sided. Pairwise comparisons were corrected for multiple testing using the Benjamini–Hochberg method. Regularized Cox logistic regression was used to discover cell-free DNA genomic alterations associated with clinical benefit to inavolisib. Two-way ANOVA (type III) analysis was performed on the unbalanced drug sensitivity AUC data after Box–Cox transformation in R using BoxCox function with lambda set to “auto” from the forecast R package (v8.17.0) to satisfy the assumption of homogeneity of variances verified by the Levene test. All possible pairwise comparisons were performed using a Bonferroni adjustment with the emmeans R package (v1.8.3). Raw counts or normalized data were used for all other statistical tests.

Chemical reagents

Tool compounds and inhibitors were made according to known methods/procedures in the literature or acquired from commercially available sources.

Antibody reagents

Antibodies to p110α (cat. No. 4249, RRID: AB_2165248), pAKT Ser473 (cat. No. 4060, RRID: AB_2315049), pS6 S235/236 (cat. No. 2211, RRID: AB_331679), HER3 (cat. No. 12708, RRID: AB_2721919), HER2 (cat. No. 2242, RRID: AB_331015), pHER2 Y1221/Y1222 (cat. No. 2243, RRID: AB_490899), pHER3 Y128 (cat. No. 4791, RRID: AB_2099709), pHER3 Y1328 (cat. No. 14525, RRID: AB_2798501), pPLCγ Y783 (cat. No. 14008, RRID: AB_2728690), pERK T202/T204 (cat. No. 9101, RRID: AB_331646), p4EBP T37/46 (cat. No. 9459, RRID: AB_330985), FGFR1 (cat. No. 9740, RRID: AB_11178519), FGFR2 (cat. No. 11835, RRID: AB_2797742), FGFR3 (cat. No. 4574, RRID: AB_2246903), FGFR4 (cat. No. 8562, RRID: AB_10891199), pFGFR Y653/654 (cat. No. 3476, RRID: AB_331369), and pFRS2A Y196 (cat. No. 3864, RRID: AB_2106222) were obtained from Cell Signaling Technology. Antibody to β-actin (cat. No. A5441, RRID: AB_476744) was from Sigma-Aldrich. Antibodies to Ras (cat. No. ab52939, RRID: AB_2121042), p85α (cat. No. ab133595, RRID: AB_3662733), and p85β (cat. No. ab28356, RRID: AB_777259) were obtained from Abcam. Peroxidase goat anti-rabbit IgG (cat. No. 111-035-144, RRID: AB_2307391) and peroxidase goat anti-mouse IgG (cat. No. 115-035-146, RRID: AB_2307392) were obtained from Jackson ImmunoResearch. Ubiquitin reagent TUBE1 (UM101) was obtained from LifeSensors, Inc.

PRISM assay

In brief, inavolisib was treated at eight doses in triplicate across 900 PRISM-barcoded cancer cell lines for 5 days (22). All cell lines were cultured in RPMI 1640 with 10% FBS. Cell lines were lysed with Qiagen TCL buffer to isolate mRNA. mRNA was then reverse-transcribed into cDNA, and then the sequence containing the unique PRISM barcode was amplified using PCR. Finally, Luminex beads that recognize the specific barcode sequences were hybridized to the PCR products and then detected using a Luminex scanner, which reports signal as a median fluorescent intensity. The data were then processed to calculate the AUC and IC₅₀ value and create dose–response curves. The details of the data processing steps can be found at https://github.com/cmap/dockerized_mts.

Cell lines and cell culture

All cell lines used in this article were from the Genentech tissue culture cell line bank (Supplementary Table S1). All cell lines underwent authentication by short tandem repeat (STR) profiling, SNP fingerprinting, and Mycoplasma testing. STR profiles were determined for each line using the Promega PowerPlex 16 System. This is performed once and compared with external STR profiles of cell lines (when available) to determine cell line ancestry. SNP profiles are performed each time new stocks are expanded for cryopreservation. Cell line identity is verified by high-throughput SNP profiling using Fluidigm multiplexed assays. SNP were selected based on minor allele frequency and presence on commercial genotyping platforms. SNP profiles are compared with SNP calls from available internal and external data (when available) to determine or confirm ancestry. All stocks are tested for Mycoplasma before and after cells are cryopreserved. Two methods are used to avoid false-positive and false-negative results: Lonza MycoAlert and Stratagene MycoSensor.

Cell lines (Supplementary Table S1) were cultured in RPMI media prepared by the Genentech media preparation group. Powdered RPMI (Gibco, cat. No. 31800-105) was mixed with distilled water, 2-g/L sodium bicarbonate (Sigma, cat. No. S5761-1KG) was added next, and the final volume was adjusted with distilled water. The pH was adjusted to fall within the range of 7.00 ± 0.01. The prepared RPMI was supplemented with 10% heat-inactivated FBS (Corning, cat. No. 35-016-cm), 2-mmol/L Glutamax (Gibco, cat. No. 35050-061), and 100-U/mL penicillin–streptomycin (Gibco, cat. No. 15140-122). The medium was immediately processed into sterile containers by membrane filtration with a 0.2-μm filter using a positive-pressure system.

All cell lines for signaling experiments were cultured in the same growth media described above. Once we received the cells, they were expanded two to three times using 0.5% trypsin for dissociation, following the subculturing procedures described in Supplementary Table S1.

Viability assay CellTiter-Glo

Cells were seeded (1,000–2,000 cells/well) in 384-well plates for 16 hours. Compounds were added to the media to each well in order to achieve the desired concentration. Cell viability was assessed by quantifying ATP using CellTiter-Glo (Promega) after 5 days of incubation. After 5 days, relative numbers of viable cells were measured by luminescence using CellTiter-Glo (Promega) according to the manufacturer’s instructions and read on a Wallac Multilabel Reader (PerkinElmer). The EC₅₀ concentration calculations were carried out using Prism 6.0 software (GraphPad). GR values (23) were calculated using the gDR package in R (24) and reference division time (23, 24). All metrics are reported in Supplementary Table S2. Synergy was assessed by excess over single agent and excess over Bliss independence (25).

siRNA transfection

Transfection of siRNA was carried out using Lipofectamine RNAiMAX reagent (Thermo Fisher Scientific, cat. No. 13778150) 72 hours in advance of drug treatment. siRNA details are as follows:

• • FGFR1 siRNA: Synthesized and chemically modified by Dharmacon, Inc.

ON-TARGETplus Human FGFR1 (2260) siRNA-SMARTpool, cat ID: L-003131-00-0005.

• FGFR2 siRNA: Synthesized and chemically modified by Dharmacon, Inc.

ON-TARGETplus Human FGFR2 (2263) siRNA-SMARTpool, cat ID: L-003132-00-0005.

• FGFR3 siRNA: Synthesized and chemically modified by Dharmacon, Inc.

ON-TARGETplus Human FGFR3 (2261) siRNA-SMARTpool, cat ID: L-003133-00-0005.

• FGFR4 siRNA: Synthesized and chemically modified by Dharmacon, Inc.

ON-TARGETplus Human FGFR4 (2264) siRNA-SMARTpool, cat ID: L-003134-00-0005.

• ERBB3 siRNA: Synthesized and chemically modified by Dharmacon, Inc.

ON-TARGETplus Human ERBB3 (2065) siRNA-SMARTpool, cat ID: L-003127-00-0005.

To generate the T47D_mutant line bearing p110a H1047R as a homozygous mutant, two CRISPR-Cas9 constructs were designed. One was designed to specifically target the wild-type (WT) allele in exon 21 (guide RNA H1047R-2, ATG​AAT​GAT​GCA​CAT​CAT​GG), and the second was designed to target the intron of both WT and mutant alleles (guide RNA H1047R-7, ACA​TTT​GAG​CAA​AGA​CCT​GA).

Plasmids for each targeting pair were cotransfected using TurboFectin (Thermo Fisher Scientific). After 48 hours, cells were put under selection with 1-μg/mL puromycin. Puromycin-resistant cells were further selected by collecting GFP-expressing cells by flow cytometry, and clones were expanded in standard cell culture conditions to create stable lines. Targeting efficiency of the CRISPR-induced allelic knockouts was assessed by PCR flanking the target sites (forward: TGCTGTGAAGGAAAATGGAA; reverse: TGCAGTGTGGAATCCAGAGTGAGC) and clones were further validated with qRT-PCR.

RNA isolation and PIK3CA allele–specific qRT-PCR

Total RNA was isolated from cells using the RNeasy Plus Mini Kit (Qiagen) following the manufacturer’s instructions. First-strand cDNA synthesis and RT-qPCR were carried out using the One-Step qRT-PCR reagent (Applied Biosystems). The resulting signal was detected on an Applied Biosystems Real-Time PCR System. Primers and allele-specific probes were as follows:

• PIK3CA H1047R-forward: GGCTTTGGAGTATTTCATGAAACA

• PIK3CA H1047R-reverse: GAAGATCCAATCCATTTTTGTTGTC

• PIK3CA H1047R WT-probe: ATGATGCACATCATGGT

• PIK3CA H1047R mut-probe: TGATGCACGTCATGGT

Western blots

Protein concentration was determined using the Pierce BCA Protein Assay Kit (Thermo Fisher Scientific, cat. No. 23225). For immunoblots, equal protein amounts were loaded and then separated by electrophoresis through NuPAGE Novex Bis-Tris 4% to 12% gradient gels (Invitrogen) and MOPS SDS buffer, with the SeeBlue Plus2 Pre-stained Protein Standard (Thermo Fisher Scientific, cat. No. LC5925) serving as the molecular weight marker. However, the detection of ubiquitinated p110α is an exception. To achieve proper separation of this modified protein, we used a 3% to 8% Tris Acetate gel run in Tris Acetate SDS buffer. In these specific instances, the HiMark Pre-stained Protein Standard (Thermo Fisher Scientific, cat. No. LC5699) was used as the molecular weight marker. Proteins were transferred onto nitrocellulose membranes using the iBlot system and protocol from InVitrogen. Membranes were blocked for 1 hour in 5% milk in TBST (20 mmol/L Tris-HCL pH 7.5, 135-mmol/L NaCl, and 0.1% Tween 20) at room temperature (RT) and incubated overnight with the primary antibody at a 1:1,000 dilution in 5% milk TBST at 4°C with agitation. The membranes were washed three times, 5 minutes each, with TBST. After 1 hour of incubation at RT with the 1:10,000 diluted anti-mouse (Jackson ImmunoResearch) or anti-rabbit (Jackson Immuno Research) horseradish peroxidase–conjugated secondary antibodies, the membranes were washed three times in TBST, and bands were detected using ECL (Bio-Rad) and autoradiography film (Amersham Biosciences, GE Healthcare) exposure.

Ubiquitin pulldown assay

Cells were lysed in 20-mmol/L TrisHCL pH 7.5, 137-mmol/L NaCl, 1-mmol/L EDTA, 1% NP40, and 10% glycerol plus protease and phosphatase inhibitors. For the ubiquitinated protein pulldown experiment, cells were lysed in a lysis buffer containing 200-μg/mL TUBE1 (LifeSensors UM101). Lysates were isolated, and 50 μL of glutathione agarose beads were added to them (Sigma, G4705). The samples were incubated overnight, and the captured ubiquitinated protein was eluted in SDS-reducing sample buffer.

Subcellular fractionation

Cells were washed once with PBS before scraping into a 0.8-mL/dish with hypotonic lysis buffer (25-mmol/L Tris–HCl pH 7.5, 10-mmol/L Nacl, 1-mmol/L EDTA, and protease and phosphatase inhibitors). The cells were lysed by 30 strokes in a Dounce homogenizer, subjected to centrifugation at 1,500 g (3,000 rpm) for 5 minutes, to pellet nuclei and unbroken cells, followed by centrifugation of the supernatant at 100,000 g (44,000 RPM) in TLA55 rotor for 40 minutes. The supernatant (800 μL) was collected (S100 fraction), and the pellet was resuspended in 200-μL hypotonic lysis buffer plus 1% NP40 (P100 fraction). The resuspended pellet was centrifuged for 5 minutes at high speed in microfuge, and the supernatant was collected.

Immunoprecipitation and pulldown

Cells were lysed in 20-mmol/L TrisHCL pH 7.5, 137-mmol/L NaCl, 1-mmol/L EDTA, 1% NP40, 10% glycerol, and protease and phosphatase inhibitors. For the immunoprecipitation (IP) of p85β, 1 mg of prepared lysates were incubated with antibodies (Abcam, cat. No. ab28346, RRID: AB_777251) at a dilution of 1:100 at final volume of 1 mL overnight. Next, 50 μL of protein A agarose beads were added to each sample and incubated an additional 2 hours. For HER3 and FGFR2 IP, lysates were incubated overnight with a biotin-conjugated HER3 or FGFR2 antibody, respectively (Invitrogen, cat. No. MA5-13037, RRID: AB_10983790; and OriGene, cat. No. TA502917AM), which was followed by the addition of 50 μL of streptavidin agarose beads to each sample and an additional 2 hours of incubation. For RAS IP, cells were lysed in IP lysis buffer (25-mmol/L Tris HCL pH 7.5, 150-mmol/L NaCl, 1-mmol/L EDTA, 1% NP40, 5-mmol/L MgCl₂, 10% glycerol, and protease and phosphatase inhibitors) and collected on ice. Lysates were spun at 14,000 rpm for 10 minutes at 4°C, and the supernatant was collected. Lysates were quantified using the Pierce BCA Protein Assay Kit (Thermo Fisher Scientific, cat. No. 23225). Soluble proteins (1 mg) were subjected to IP with either 12 μL (6 μL of packed beads) of washed anti-RAS magnetic beads (Millipore cat. No. 16-321) or mouse anti-IgG magnetic bead conjugate (Cell Signaling Technology, cat. No. 5873S) as a control and incubated at 4°C for 5 hours. The complex was washed 3× with 1 mL of cold IP lysis buffer. Washed beads were resuspended in 50 μL of sample buffer.

FGFR2 stable cell line generation

T47D breast cancer cells were engineered using CRISPR to knock out the PIK3CA WT allele to create the T47D PIK3CA mutant cell line using the method previously described (14). FGFR2 WT and FGFR2 N550K cDNA were synthesized and cloned into doxycycline (dox)-inducible PiggyBac transposase expression vector with constitutive mCherry expression. Cells (∼2 × 10⁵) were transfected with 750 ng of transposon vector (FGFR2 WT or FGFR2 N550K) and 250 ng of piggyBac transposase in six-well culture plates. After 48 hours, cells were detached and replated in T175 flasks. Once confluent, mCherry expression was selected by flow cytometry, and clones were expanded to create stable lines. Expression of FGFR2 was validated by qRT-PCR using FGFR2 primers and probes (Applied Biosystems).

RTK array

Total protein was extracted using the lysis buffer provided by Human Phospho-RTK Array Kit (R&D Systems). Protein concentrations were determined, and 300 μg of total protein was incubated overnight at 4°C with array membranes. After washing, array membranes were incubated with Anti-Phospho-Tyrosine-HRP Detection Antibody (provided in the kit) for 2 hours at RT with agitation. pRTK was detected with Chemi Reagent Mix and autoradiography film.

Live-cell imaging by IncuCyte

Cells were cultured and plated into 96-well plates at a density of 2,000 cells per well. The next day, inhibitors were added at the indicated concentrations. Every 7 days, the media were aspirated and replaced with fresh media with or without inhibitors. Cells were imaged every 4 hours in IncuCyte Zoom (Essen Bioscience). Phase-contrast images were collected and analyzed for cell proliferation. The cell confluence of six replicate wells was determined for the duration of the experiment.

In vivo xenograft studies

Female NCR nude mice were obtained from Taconic Biosciences and housed at Genentech according to standards established by the Institutional Animal Care and Use Committee. Human MFM223x2.2 mammary carcinoma cells were cultured in vitro, harvested in log-phase growth, and resuspended in Hank’s Balanced Salt Solution containing Matrigel (BD Biosciences) at a 1:1 ratio by volume for in vivo inoculation. Mice were implanted with 17β-estradiol pellets (0.36 mg, 60-day release, Innovative Research of America, SE121) subcutaneously in the left flank 1 to 3 days before inoculation. Mice were then inoculated in the right second mammary fat pad with 10 × 10⁶ MFM223x2.2 cells in a total volume of 100 μL. Tumors were allowed to grow to a volume in an initial range before the mice were randomized to treatment groups at the start of dosing to create closely matched baseline average tumor sizes across regimens.

For the PD study, mice (n = 4–5 per group) were given an FGFR2 small-molecule inhibitor (FGFR2i; 1, 10, or 20 mg/kg) by oral gavage daily for 4 days; tumors were collected 6 hours after the last dose. For the efficacy study, mice (n = 8 per group) were given the FGFR2i (10 or 20 mg/kg) and/or inavolisib (25 mg/kg) by oral gavage daily for 22 days. The FGFR2i is formulated in 60% polyethylene glycol 400 (PEG400), 1 eq. hydrochloric acid and/or inavolisib are formulated in 0.5% methylcellulose and 0.2% Tween 80.

Tumor sizes and mouse body weights were recorded twice weekly over the course of the study. Tumor volumes were measured in two perpendicular dimensions (length and width) using Ultra Cal-IV calipers (model 54 − 10 − 111; Fred V. Fowler, Co.) and calculated according to the following formula: Tumor size (mm³) = (longer measurement × shorter measurement²) × 0.5. Animal body weights were measured using an Adventura Pro AV812 scale (Ohaus Corporation). Percent animal weight change was calculated using the following formula: Body weight change (%) = [(current body weight/initial body weight) − 1) × 100]. Analyses and comparisons of tumor growth were performed using a package of customized functions in R (v3.6.2); R Foundation for Statistical Computing, which integrates software from open-source packages as described in Forrest and colleagues (26). Growth contrast represents the difference in AUC-based growth rates (endpoint gain integrated in time) between the treatment and reference groups (26). Contrast values < 1 indicate an antitumor effect; the smaller the value below 1, the greater the magnitude of the antitumor effect. The 95% CI are based on the fitted model and variability measures of the data.

Results

Clinical profile of inavolisib monotherapy

Between December 2016 and January 2019, 19 female patients with HR+/HER2− breast cancer and one male patient with colorectal cancer were enrolled in a 3 + 3 design dose-escalation study using 6 mg (n = 7), 9 mg (n = 9), and 12 mg (n = 4) once-daily oral inavolisib dose cohorts (Fig. 1A; Supplementary Table S3). Key eligibility criteria included the presence of at least one PIK3CAmut in tumor tissue or ctDNA, fasting glucose ≤140 mg/dL, and HbA1c <7%. The median age was 65 years (range, 41–77), and 11 patients (55%) had an Eastern Cooperative Oncology Group performance status score of 0 at baseline. Five patients (25%) were obese (body mass index ≥30 kg/m²) overall, including two patients (50%) in the 12-mg cohort. Racial and ethnic diversity was limited in this small study (Table 1). The median number of prior cancer therapies in the metastatic setting was three (range, 1–10). Most patients were previously treated with chemotherapy (n = 19, 95%) as well as a cyclin-dependent kinase 4/6 inhibitor (n = 18, 90%), aromatase inhibitor (n = 19, 95%), and/or the selective estrogen receptor degrader fulvestrant (n = 13, 65%). Notably, none of the patients were previously treated with PI3K inhibitors. The median inavolisib treatment duration was 3.8 months (range, 1.1−17.6), and median cumulative dose intensity was 97% (range, 46%−101%). Five (25%) patients received treatment beyond 200 days.

Figure 1.

Clinical profile of inavolisib as a single agent. A, Twenty patients were enrolled in an open-label, phase 1, dose-escalation study to investigate once-daily inavolisib dosing. B, The mean plasma concentration vs. time profiles for single and multiple dose levels of oral inavolisib demonstrated a linear profile and supported daily dosing. C, Proliferative and PI3K pathway activity were decreased in the majority of paired tumor samples as assessed by IHC for Ki67, pAKT, and pS6. D, FDG-PET scans at baseline and after 2 weeks of daily inavolisib showed metabolic responses at all dose levels evaluated. E, Decreased PIK3CA MAF was observed over time in the majority of paired specimens from patients who experienced a PR. F, Twenty patients were evaluated for efficacy, of whom 19 had measurable disease; the best overall response rate was 26%, the confirmed overall response rate was 21% among the 19 patients with measurable disease, and the clinical benefit rate was 45% among all 20 patients. %CFB, percent change from baseline; CDK, cyclin-dependent kinase; cPR, confirmed PR; HEL, helical; KIN, kinase; MUL, multiple; PD, progressive disease; SD, stable disease; SLD, sum of the longest diameters; SUV, standardized uptake value; SUV-max, maximum standardized uptake value.

Table 1.

Demographics and baseline characteristics.

​	Inavolisib dose cohort			All cohorts
	6 mg	9 mg	12 mg	N = 20
	N = 7	N = 9	N = 4
Age (years)	​	​	​	​
Median	64	69	61	65
Range	56–73	55–73	41–77	41–77
Sex, n  (%)	​	​	​	​
Male	0	0	1 (25)	1 (5)
Female	7 (100)	9 (100)	3 (75)	19 (95)
Race, n (%)	​	​	​	​
Black or African American	0	0	1 (25)	1 (5)
White	4 (57)	8 (89)	2 (50)	14 (70)
Unknown	3 (43)	1 (11)	1 (25)	5 (25)
Eastern Cooperative Oncology Group performance status at baseline, n (%)	​	​	​	​
0	3 (43)	5 (56)	3 (75)	11 (55)
1	4 (57)	4 (44)	1 (25)	9 (45)
Weight at baseline (kg)	​	​	​	​
Median	72	61	84	65
Range	57–94	55–85	68–111	55–111
Body mass index at baseline, n (%)	​	​	​	​
≥30 kg/m2	3 (43)	0	2 (50)	5 (25)
Prior lines of therapy in metastatic setting	​	​	​	​
Median	3	3	4	3
Range	2–10	1–10	2–5	1–10
Prior therapies, n (%)	​	​	​	​
Chemotherapy	7 (100)	8 (89)	4 (100)	19 (95)
Cyclin-dependent kinase 4/6 inhibitor	6 (86)	9 (100)	3 (75)	18 (90)
Aromatase inhibitor	7 (100)	9 (100)	3 (75)	19 (95)
Selective estrogen receptor degrader	5 (71)	6 (67)	2 (50)	13 (65)
Fulvestrant	5 (71)	6 (67)	2 (50)	13 (65)

The MTD of inavolisib was 9 mg once daily; two of four DLT-evaluable patients in the 12-mg dose group experienced DLTs, including grade 4 hyperglycemia and grade 3 fatigue that lasted 5 days; AE were graded according to the NCI Common Terminology Criteria for Adverse Events v4. All patients discontinued study treatment because of disease progression (RECIST v1.1, n = 16, 80%) and clinical progression (n = 4, 20%). One death on study was due to disease progression during the safety follow-up.

Overall, 19 (95%) patients had at least one AE related to inavolisib (Table 2), the most common (≥15%) of which were hyperglycemia (70%), diarrhea (45%), decreased appetite, stomatitis (grouped term: stomatitis, mucosal inflammation, and palatal ulcer), vomiting (20% each), fatigue, nausea, weight decrease, and alopecia (15% each). Eight (40%) patients had at least one grade 3 to 5 AE related to inavolisib, including hyperglycemia (20%), fatigue, nausea, weight decrease, asthenia, and lymphopenia (5% each). One patient experienced a serious AE related to inavolisib (grade 4 hyperglycemia) in the 12-mg cohort, which was downgraded to grade 3 the same day and resolved within 6 days following treatment with oral antihyperglycemic medication. Treatment-related rash occurred in one (5%) patient (grade 1). Treatment-related stomatitis (grouped term) occurred in four (20%) patients (all grade 1) and generally responded to oral corticosteroid mouthwash. Eleven (55%) patients had at least one AE related to inavolisib leading to dose modification in the 6 mg (n = 4), 9 mg (n = 3), and 12 mg (n = 4) dose cohorts, with 13 events that included seven interruptions and six reductions (no discontinuations). No treatment-related grade ≥3 gastrointestinal toxicities were reported, and treatment-related diarrhea events were of grades from 1 to 2. Colitis was not observed with inavolisib monotherapy, including in the nine patients on treatment for ≥5 months. No grade 5 AE were reported.

Table 2.

AE related to inavolisib.

MedDRA preferred term	All grades Related AE (N = 20)	Grades 3–5 Related AE (N = 20)
Total number of patients with at least one AE, n (%)	19 (95)	8 (40)
Hyperglycemia	14 (70)	4 (20)
Diarrhea	9 (45)	0
Decreased appetite	4 (20)	0
Stomatitis (grouped term)a	4 (20)	0
Vomiting	4 (20)	0
Fatigue	3 (15)	1 (5)
Nausea	3 (15)	1 (5)
Weight decreased	3 (15)	1 (5)
Alopecia	3 (15)	0
Asthenia	2 (10)	1 (5)
Lymphopenia	2 (10)	1 (5)
Constipation	2 (10)	0
Dysgeusia	2 (10)	0
Flatulence	2 (10)	0
Headache	2 (10)	0
Hyponatremia	2 (10)	0
Taste disorder	2 (10)	0
Thrombocytopenia	2 (10)	0
Vision blurred	2 (10)	0

^aStomatitis (grouped terms) = stomatitis, mucosal inflammation, and palatal ulcer.

Twenty patients were included in the PK analysis, and the data were summarized according to treatment dose level. Patients received their first dose on C1D1, followed by a 7-day washout period and frequent PK sampling. Once-daily dosing was resumed on C1D8. Plasma exposure of inavolisib increased in a dose-proportional manner following single and multiple doses (Fig. 1B). The mean half-life (t_1/2) following a single dose was 16 hours. With continuous once-daily dosing, there was a ∼twofold accumulation, consistent with the observed t_1/2 and the dosing frequency. At the MTD (9 mg), inavolisib intersubject PK variability was low for C_max and AUC_0–24 after a single dose as well as at steady state (%CV ∼20%–40%).

Matched pretreatment (baseline) and on-treatment (C1D15) biopsy samples from nine patients were evaluable for PD response to study treatment by IHC. With some exceptions, this analysis illustrated a consistent decrease in PI3K pathway activity based on reduced H-scores for pAKT and pS6 as well as reduced proliferation as assessed by Ki67 nuclear staining (Fig. 1C).

Given that PI3K inhibition modulates cellular glucose uptake, PD response was also evaluated by FDG-PET. All 20 patients had FDG-avid disease on pretreatment (baseline) scans, and 16 (75%) achieved a PMR on scans acquired after 2 weeks of once-daily inavolisib (Fig. 1D). Metabolic responses were seen at all dose levels with a trend toward more frequent and deeper responses at higher doses, that is a PMR in 4/7 (57%) and mean of 25% reduction in uptake at 6 mg versus a PMR in 12/13 (92%) and mean of 43% reduction in uptake with >6 mg (Fig. 1D).

Reflective of the eligible PIK3CA hotspot mutations for this trial (see “Patients and Methods,”), PIK3CAmut primarily occurred in the helical (n = 8, 40%) and kinase (n = 6, 30%) domains of p110α; one patient (5%) had a missense mutation resulting in a N345K substitution in the C2 domain, and five (25%) patients’ tumors exhibited more than one PIK3CAmut. From analysis of ctDNA, decreased PIK3CA MAF was observed over time between pretreatment and on-treatment time points in the majority of paired specimens, particularly in those from patients who experienced a PMR (Fig. 1E).

Twenty patients were evaluated for response, including 19 patients with measurable disease at baseline, and were assessed for radiographic response using RECIST v1.1. Among patients with measurable disease (n = 19), the best overall response was PR in five (26%) patients, and the confirmed objective response rate was 21% (Fig. 1F). The duration of response was 3.7 months (95% CI, 3.0–not evaluable) among the responders with a confirmed response. The clinical benefit rate [defined as the percentage of patients with confirmed CR, PR, and those with ≥24 weeks' duration of stable disease or non-CR/nonprogressive disease)] since the first study treatment for single-agent inavolisib (N = 20) was 45% (n = 9). The median progression-free survival (PFS) was 3.7 months (95% CI, 3.5–7.3). The landmark PFS rate at 6 months was 37% (95% CI, 15–59).

FGFR2 mutations are associated with clinical benefit

To better understand the genomic alterations associated with response to inavolisib, we performed targeted sequencing of ctDNA from available longitudinal plasma samples from the 20 patients. We found FGFR2 short variant (SV) mutations in pretreatment liquid biopsy samples from two of the nine patients who experienced clinical benefit during the study as defined above (Fig. 2A; Supplementary Fig. S1; see “Patients and Methods”). These missense mutations translated into a P253R substitution in the FGFR2 extracellular domain and a N549K substitution in the FGFR2 kinase domain, both of which are known to result in constitutive activation of the receptor (Fig. 2A; Supplementary Table S4; refs. 18, 27). FGFR2 alterations were not observed in patients who did not receive clinical benefit from single-agent inavolisib (Supplementary Fig. S1). PTEN loss-of-function mutations were present in samples from two of the nine patients who received clinical benefit as well as in samples from two of the 11 patients who did not receive clinical benefit from single-agent inavolisib. Additionally, we found activating AKT1 E17K mutations in three patients who did not receive clinical benefit (Fig. 2A; Supplementary Fig. S1; Supplementary Table S4).

Figure 2.

FGFR2 expression is associated with inavolisib sensitivity in patients with PIK3CA-mutated breast cancer. A,FGFR2 mutations were identified in baseline liquid biopsies of patients who experienced clinical benefit from inavolisib. FGFR2 N549K mutations are highlighted with a cross, and identified mutations in PTEN and AKT1 are also noted. B, Lollipop plot of the landscape of known and likely FGFR2 alterations across 49,834 breast tumors that underwent CGP. C, Volcano plot illustrating the co-occurrence (purple) or mutual exclusivity (gray) of FGFR2 SV with alterations in other CGP-baited genes across the cohort of breast tumor samples shown in B. D, Inavolisib area under the Sigmoid-fit concentration–response curve (AUC) values across 662 PRISM cancer cell lines. Inavolisib AUC values of PIK3CA hotspot-mutant vs. PIK3CA non–hotspot-mutant (other) or WT and FGFR2 hotspot-altered vs. FGFR2 non–hotspot-altered (other) or WT were compared. PIK3CA WT/other and FGFR2 WT/other, n = 581; PIK3CA hotspot and FGFR2 WT/other, n = 78; PIK3CA WT/other and FGFR2 hotspot, n = 7; PIK3CA hotspot and FGFR2 hotspot, n = 7. E, Correlation of FGFR2 expression vs. inavolisib AUC values across PRISM cell lines (n = 662). F, Comparison of inavolisib AUC values of PIK3CA hotspot vs. PIK3CA WT/other and top 33.3% vs. bottom 33.3% FGFR2-expressing PRISM cell lines. PIK3CA WT/other and FGFR2 bottom 33.3%, n = 213; PIK3CA hotspot and FGFR2 bottom 33.3%, n = 8; PIK3CA WT/other and FGFR2 top 33.3%, n = 173; PIK3CA hotspot and FGFR2 top 33.3%, n = 48. G, Western blot of cell lysates (n = 16) with FGFR2 alterations blotted for FGFR1/2/3/4 and actin. Blue labels indicate cell lines with visually detectable FGFR2 expression, and red labels indicate cell lines with non–visually detectable FGFR2 protein expression by Western blot. Representative results from experiments (n = 2). H, Correlations of FGFR1–4 expression with relative cell viability were evaluated after specific targeting of FGFR1, 2, 3, or 4 using siRNA. Cell viability was assessed by ATP quantification 5 days post-transfection across FGFR2-high (n = 6) and FGFR2-low (n = 9) cell lines. Representative results from experiments (n = 2). I, Comparison of relative cell viability values in FGFR2-high (n = 6) vs. FGFR2-low (n = 9) cell lines following 5-day FGFR2 silencing. Representative results from experiments (n = 2). J, Top, biochemical enzyme activity of the FGFR2i. Bottom, comparison of GR₅₀ values for a tool FGFRi in FGFR2-high (n = 6) vs. FGFR2-low (n = 9) cell lines. Data in (D, F, I, and J) are represented as the median (center line) ± IQR (25th to 75th percentile, box) and ± full range (minimum to maximum, whiskers). P values were calculated with one-tailed one-way ANOVA and FDR multiple testing corrections using the two-stage step-up method from Benjamini, Krieger, and Yekutieli (D and F), two-tailed t-transformations of Pearson R correlation coefficients (E and H), two-tailed unpaired Student t tests (I), or Wilcoxon rank-sum test (J). SM, single mutation; VUS, variants of unknown significance.

To validate the possible cooperation between PIK3CA and FGFR2 alterations within a large oncogenomic dataset, we first identified FGFR2 N549 (D/H/K/T) hotspot mutations in tumor biopsies from nearly 50,000 patients with breast cancer that underwent CGP during the course of routine clinical care (Fig. 2B; ref. 28). FGFR2 N549 SV alterations occurred in 0.37% of breast cancer samples. We then examined driver alterations, including PIK3CA hotspot mutations and their incidence in FGFR2-mutant breast cancers. Interestingly, PIK3CA mutations showed the strongest co-occurrence with FGFR2 SV (Fisher exact test; OR, 4.0, adjusted P < 0.0001; Fig. 2C; Supplementary Fig. S2A), in accordance with previous reports (18). This co-occurrence was further strengthened when FGFR2 mutations affecting the tyrosine kinase domains were subselected (Fisher exact test; OR, 6.7, adjusted P < 0.0001; Supplementary Fig. S2B). Consistent with other literature, PIK3CAmut primarily occurred in the helical and kinase domains (Supplementary Fig. S2C and S2D; refs. 29–31); both single and multiple PIK3CAmut SV co-occurred with FGFR2 alterations (Supplementary Fig. S2C).

Cells harboring multiple PIK3CAmut are associated with increased sensitivity to p110ɑ inhibition as compared with cells harboring only a single PIK3CAmut (10). Therefore, using the CGP dataset (28), we evaluated the association of different PIK3CAmut categories (i.e., single and multiple PIK3CAmut) with FGFR2 SV and other types of alterations, that is, RE, CNA, and variants of unknown significance (Supplementary Fig. S3A). In tumors categorized as HER2−, we observed a statistically significant relationship between FGFR2 alterations and various PIK3CAmut categories (Supplementary Fig. S3B); in particular, we observed a significant co-occurrence between single PIK3CAmut and known or likely oncogenic FGFR2 SV (OR, 3.6; FDR <0.05) as well as known or likely oncogenic structural FGFR2 alterations (CNA or RE; OR, 1.5; FDR <0.05). Multiple PIK3CAmut SV were significantly concurrent with FGFR2 SV (OR, 2.0; FDR <0.05) but were mutually exclusive of FGFR2 CNA (OR, 0.4; FDR <0.05). The same analysis across samples representative of all solid tumor types demonstrated that both single PIK3CAmut and multiple PIK3CAmut significantly co-occur with FGFR2 alterations of any type (SV; OR, 1.8, FDR <0.05; CNA and/or RE; OR, 3.1, FDR <0.05; Supplementary Fig. S3C).

FGFR2 alterations and sensitivity to inavolisib

Based on the findings in the clinical trial samples and the CGP dataset (Figs. 1 and 2A–C), we functionally examined whether FGFR2 mutation status and PIK3CA driver mutations were associated with increased sensitivity to inavolisib. To this end, we correlated oncogenomic and RNA sequencing–derived transcriptomic profiles and inavolisib drug–response data across 736 PRISM cell lines from all tumor indications with different presentations of the two genes (hotspot alterations, other alterations, and WT; ref. 22). Cell lines containing PIK3CA hotspot mutations were significantly more sensitive to inavolisib as compared with cell lines that contained non-hotspot PIK3CA mutations or were WT for PIK3CA (WT + other; Fig. 2D). However, the presence of FGFR2 hotspot alterations, that is, SV, focal amplifications, and/or exon 18 (E18)-truncating RE (18), in PIK3CAmut cell lines did not further increase sensitivity to inavolisib versus cell lines without FGFR2 hotspot aberrations (Fig. 2D). Apart from FGFR2 mutations, high expression drives FGFR2 oncogenicity and is predictive of response to FGFR-targeted therapies (18, 19, 32). FGFR2 gene expression in FGFR2 hotspot mutant cell lines versus other cell lines was increased (Supplementary Fig. S4A), as previously observed (18). Across the PRISM cell line panel, FGFR2 expression also correlated with PI3K–AKT signaling pathway activation, as revealed by phospho-AKT reverse-phase protein array profiling (R = 0.23; P < 0.0001; Supplementary Fig. S4B). Upon analyzing FGFR2 expression and inavolisib drug–response values, we observed a moderate association between FGFR2 expression and inavolisib sensitivity (R = −0.21; P < 0.0001; Fig. 2E). Cell lines expressing higher levels of FGFR2 based on transcript levels are significantly more sensitive to inavolisib as compared with lower expressing cell lines (Supplementary Fig. S4C). Notably, PIK3CAmut cell lines with co-occurring high FGFR2 expression were most sensitive to inavolisib (Fig. 2E). The comparison of inavolisib dose–response values in PIK3CAmut versus remaining cell lines and in the top versus bottom tertile of FGFR2-expressing lines revealed that the concomitant presence of a PIK3CA hotspot mutation and high FGFR2 expression conferred particular sensitivity to inavolisib (Fig. 2F). Taken together, these data show that PIK3CA hotspot mutant tumor cells with high FGFR2 expression levels—potentially promoted by FGFR2-activating alterations–are especially sensitive to inavolisib.

High FGFR2 expression confers dependence on the PI3K pathway

To better understand how FGFR2 mutation status and expression influence sensitivity to inavolisib, we first selected 16 cell lines that capture the genomic heterogeneity of FGFR2 observed in cancer, including FGFR2 hotspot missense mutations, focal amplifications, and RE resulting in E18-truncations (18). Western blotting showed a differential abundance and substantial overexpression of FGFR2 protein in six of the 16 cell lines. The six FGFR2 top-expressing cell lines were designated FGFR2-high, whereas the remaining 10 lines were designated FGFR2-low (Fig. 2G). High FGFR2 expression was present amidst concurrent expression of other FGFR paralogs (Fig. 2G). To identify differential dependencies on the four FGFR paralogs across the FGFR2-altered cell lines, we depleted FGFR1, 2, 3, or 4 transcripts using silencing RNAs (siRNA; Supplementary Fig. S4D). FGFR2 expression levels correlated with sensitivity to FGFR2 silencing, with the FGFR2-high cell lines being the most FGFR2 knockdown–sensitive (Fig. 2H). In contrast, the cell lines were less dependent on the other FGFR paralogs, and sensitivity to siRNA targeting of the FGFRs did not correlate to the expression levels of the respective FGFR transcripts (Fig. 2H). Consequently, FGFR2-high cell lines were significantly more sensitive to FGFR2 siRNA treatment as compared with the FGFR2-low lines (Fig. 2I), suggesting that, apart from FGFR2 alteration status, the FGFR2 expression level determines the magnitude of dependency on this RTK.

We made use of a highly selective FGFR2i (33). This tool compound specifically blocks FGFR2 kinase activity, but not the activity of FGFR1, 3, or 4, as demonstrated by enzymatic assays (Fig. 2J, top). We compared the activity of the tool FGFR2i to infigratinib, a highly potent but nonselective pan-FGFR inhibitor (34). In dose–response proliferation assays, the tool FGFR2i was significantly more potent in FGFR2-high versus FGFR2-low cell lines as measured by GR₅₀ values (P = 0.0074, Wilcoxon rank-sum test; Fig. 2J; Supplementary Fig. S5A). In particular, SUM52PE cells, which harbor focal FGFR2 amplification that results in high expression of E18-truncated, oncogenic FGFR2 (18), were highly sensitive to the FGFR2i (Supplementary Fig. S5A). In contrast, all three FGFR2-high cell lines (SUM52PE, MFE296, and AN3CA) as well as the FGFR2-low cell line MFE280 were sensitive to infigratinib (Supplementary Fig. S5A). Next, we assessed the effects of both inhibitors on FGFR signaling. In SUM52PE, MFE296, and AN3CA cells, the FGFR2i and infigratinib showed equal potency in reducing phosphorylation of FGFR and its downstream signaling effectors FRS2α, PLCɣ, AKT, and ERK. Conversely, in MFE280 cells, these targets were inhibited to a lesser extent by the tool FGFR2i as compared with infigratinib (Supplementary Fig. S5B). These results supported the observation that FGFR2-high cell lines primarily depend on FGFR2 and to a negligible extent on other FGFR and suggested that FGFR2-low lines are either FGFR-independent or depend on FGFR paralogs other than FGFR2.

Next, we ranked PIK3CAmut cell lines according to FGFR2 mRNA expression using the “Dependency Map” data (35). We selected 29 cell lines that represent the spectrum of FGFR2 expression found across different cancer types (Supplementary Fig. S6A). We verified protein expression levels of the four FGFR paralogs as well as HER2 and HER3 that are known upstream regulators of the PI3K pathway (14). FGFR2 mRNA was detected in all 29 cell lines (Supplementary Fig. S6A), yet FGFR2 protein was detectable via Western blotting only in the top 12 FGFR2 mRNA-expressing cell lines (Supplementary Fig. S6B). We subjected the 29 cell lines to inavolisib dose–response assays (Supplementary Fig. S6C). Grouping the cell lines into FGFR2-high and FGFR2-low categories based on FGFR2 protein expression levels (Supplementary Fig. S6B) showed that the majority of the FGFR2-high cell lines were sensitive to inavolisib. A few FGFR2-low cell lines were insensitive to inavolisib; however, a larger fraction of FGFR2-low cell lines showed high sensitivity to inavolisib that was comparable with that observed in FGFR2-high lines (Supplementary Fig. S6C). We analyzed the oncogenomic profiles of the 29 PIK3CAmut cell lines and identified secondary mutations in PIK3CA, focal amplifications of PIK3CA, ERBB2 (encodes HER2), or KRAS, as well as RASA1 loss-of-function alterations (Fig. 3A). ERBB2 amplifications observed in seven FGFR2-low cell lines correlated with high p-HER2 levels in five of six cell lines as identified by Western blotting (Fig. 3A; Supplementary Fig. S6B). These oncogenic alterations represented drivers that may cooperate with mutated p110α to potentiate PI3K signaling and sensitivity to PI3K targeting (8–10, 14, 34, 36) and seemed to be mutually exclusive of high levels of FGFR2 (Fig. 3A). FGFR2-high cell lines were similarly sensitive to inavolisib as compared with the FGFR2-low cell lines that express high pHER2. However, the presence of secondary PIK3CA or RAS-activating alterations did not correlate with increased inavolisib sensitivity. Notably, the remaining FGFR2-low cell lines devoid of PI3K signaling-augmenting alterations displayed weak responses to inavolisib. (P = 0.044; Wilcoxon rank-sum test; Fig. 3A and B).

Figure 3.

High FGFR2 expression is associated with inavolisib sensitivity. A, Oncoplot showing FGFR2 alterations and transcript expression [log₂ transcripts per million (TPM)] in PIK3CA hotspot mutant (E542K, E545K, and H1047R/L) cell lines (n = 29). FGFR2 protein expression categories (high and low), as assessed by Western blot in Supplementary Fig. S6B, concurrent alterations affecting PIK3CA, ERBB2, and/or RAS, and inavolisib GR₅₀ values as assessed by quantifying ATP at 5 days after treatment are also indicated. B, Comparison of inavolisib GR₅₀ values in PIK3CA mutant FGFR2-high (n = 12) vs. FGFR2-low cell lines that contain secondary PIK3CA alterations (SV, n = 2; amplification, n = 1), high p-HER2 levels (n = 6), or RAS alterations [KRAS amplification, n = 2; RASA1 loss-of-function (LOF), n = 1] vs. FGFR2-low cell lines that contain none of these alterations (n = 5). Data are represented as the median (center line) ± IQR (25th–75th percentile, box) and ± full range (minimum to maximum, whiskers). P value was calculated using a Wilcoxon rank-sum test. Representative results from experiments (n = 2).

FGFR2 activation of PI3K signaling

Based on the oncogenomic correlates identified in PIK3CAmut cell lines, we set out to mechanistically decipher the FGFR2–PI3K signaling network. HER3 is a receptor that is known to activate the PI3K regulatory subunit p85β by direct binding (37). HER3 lacks intrinsic tyrosine kinase activity, and therefore, typically forms heterodimers with other RTK, such as HER2, to mediate signaling transduction through the PI3K–AKT pathway (38–40). We explored whether FGFR2 might interact with HER3 to stimulate PI3K signaling in a PIK3CA-mutated context. We focused on the two FGFR2-high cell lines SUM52PE and MFM223, both of which express E18-truncated, oncogenic FGFR2 (18), and the FGFR2-low but HER2-dependent cell line MFE280, all of which are highly sensitive to inavolisib (Fig. 3B; Supplementary Fig. S6D). The three cell lines had a similar pattern of basal RTK phosphorylation as evaluated by phospho-RTK profiling (Supplementary Fig. S6E). Upon treatment with the FGFR2i tool compound, in SUM52PE and MFM223 FGFR2-high cell lines, we observed a substantial reduction of the phosphorylation of FGFR2 and HER3. In contrast, the FGFR2i had no effect on HER3 phosphorylation in MFE280 FGFR2-low cells (Fig. 4A and B; Supplementary Fig. S6E). Conversely, treatment with the HER2 inhibitor lapatinib resulted in decreased phosphorylation of HER3 and AKT solely in the FGFR2-low, HER2-dependent MFE280 cells (Fig. 4B). Whereas 1-hour inavolisib treatment showed no feedback-induced pHER3 expression, 24-hour treatment moderately induced pHER3 Y1289 and Y1328 expression in FGFR2-high cell lines (SUM52PE and MFM223). FGFR2i treatment blocked feedback-induced pHER3 Y1289 in the SUM52PE cell line but less in the MFM223 cell line. In contrast, treatment with the FGFR2i blocked pHER3 Y1328 in FGFR2-high cell lines at all time points studied (Supplementary Fig. S7A). These results confirmed that in FGFR2-high cancer cells, FGFR2 activates PI3K signaling through HER3 to sustain PI3K activity.

Figure 4.

Inavolisib sensitivity depends on high FGFR2 expression. A, Legend related to panels in B–I. B, Cell lines were treated with 0.03 μmol/L of the FGFR2i or 2-μmol/L lapatinib for 1 hour followed by immunoblotting with the antibodies indicated at left. Representative results from experiments (n = 2). C, Cell lines were treated with inavolisib or the FGFR2i at various concentrations for 1 hour. Cell lysates were immunoprecipitated with an antibody against p85β, followed by immunoblotting with the antibodies indicated at left. Representative results from experiments (n = 2). IB, immunoblotting. D, Following RAS-GTP pulldown, cell lines were treated with the FGFR2i or lapatinib for different durations and immunoblotted with the antibodies indicated at left. Representative results from experiments (n = 2). E, Cell lysates from cells treated with inavolisib alone or in combination with the FGFR2i or lapatinib for 4 hours were immunoprecipitated with RAS antibody and blotted with p110α antibody. F, Mechanistic model of the effects of FGFR2 and HER2 inhibition on HER3 and RAS activity. FGFR2-high–expressing cell lines induced PI3K signaling through both HER3 and WT RAS activity (top) compared with HER2-induced PI3K signaling through HER3 but not RAS activity (bottom). G, FGFR2-high–expressing cell lines, MFM223 and SUM52PE, were treated with inavolisib, FGFR2i, or lapatinib for 1 hour. Membrane fractions were analyzed by reciprocal co-IP with one another using HER3 or FGFR2 antibody and Western blotting with FGFR2, HER3, RAS, and p85β antibody. Representative results from experiments (n = 2). H, SUM52PE, MFM223, and MFE280 cells were treated with inavolisib single-agent or in combination with the FGFR2i or lapatinib for 6 hours. Ubiquitinated proteins were pulled down from the membrane fraction with TUBE1 reagent and blotted with p110α antibody. Representative results from experiments (n = 2). I, Western blots of the inhibitor response in PI3K signaling (pHER3 and pAKT) in PIK3CA mutant MFM223 and PIK3CA WT SUM52PE; cell lines were treated with 0.5-μmol/L inavolisib or 1-μmol/L alpelisib for different durations. Representative results from experiments (n = 2). J, Ratio of inavolisib and alpelisib GR₅₀ values in FGFR2-high (n = 12) vs. FGFR2-low (n = 9) expressing cell lines harboring PIK3CA mutations, as assessed in a 5-day viability assay. Data are represented as median (center line) ± IQR (25th to 75th percentile, box) and ± full range (minimum to maximum, whiskers). P value was calculated using Wilcoxon rank-sum test. Representative results from experiments (n = 2). WB, Western blotting.

We next tested whether the inhibition of FGFR2 prevents HER3–PI3K interaction and pathway activation. We performed co-IP of the PI3K regulatory subunit p85β in SUM52PE, MFM223, and MFE280 cells and detected a strong interaction between p85β and HER3 (Fig. 4C). Treatment with inavolisib did not affect binding of p85β to HER3, whereas the tool FGFR2i blocked this interaction specifically in FGFR2-high SUM52PE and MFM223 cells, resulting in reduced phosphorylation of AKT (Fig. 4C). Co-IP of HER3 confirmed loss of the HER3–p85β interaction upon treatment with the FGFR2i (Supplementary Fig. S7B). Notably, the FGFR2i did not affect the HER3–p85β interaction in MFE280 FGFR2-low cells (Fig. 4C; Supplementary Fig. S7B). However, in these cells, lapatinib suppressed binding of p85β to HER3 (Supplementary Fig. S7B), consistent with the high HER2 levels and the HER2-dependent HER3–PI3K activation (14).

To further evaluate whether FGFR2 depends on HER3 to signal through PI3K, we silenced HER3 using RNAi. Interestingly, loss of HER3 only partially reduced AKT phosphorylation in SUM52PE and MFM223 cells (Supplementary Fig. S7C), suggesting that FGFR2 does not wholly rely on HER3 to activate PI3K signaling. RTK such as FGFR2 can also induce the PI3K signaling cascade through RAS (41). Pulldown of GTP-bound RAS showed that the FGFR2i causes a decline in RAS-GTP levels in the MFM223 FGFR2-high cells, whereas RAS-GTP levels remained unchanged in the MFE280 FGFR2-low cells (Fig. 4D). RAS co-IP further showed that inhibition of FGFR2 effectively impaired the interaction of RAS with the p110α PI3K catalytic subunit in MFM223 cells (Fig. 4E). However, in MFE280 cells, FGFR2 blockade did not interfere with the RAS–p110α interaction (Fig. 4E). Notably, lapatinib treatment affected neither RAS-GTP levels nor the RAS–p110α interaction in the MFM223 or MFE280 cells (Fig. 4D and E). This suggests that HER2 primarily relied on HER3 to activate PI3K, whereas FGFR2 depends on both HER3 and RAS to mediate PI3K signaling (Fig. 4F).

Next, we examined whether FGFR2 directly interacts with HER3. SUM52PE and MFM223 FGFR2-high cells were treated with inavolisib, the FGFR2i, or lapatinib, and subjected to co-IP. HER3 confirmed to interact with p85β, as expected (Fig. 4G). FGFR2 also interacted with p85β but did not interact with HER3 or p85α (Fig. 4G; Supplementary Fig. S7D). Blockade of FGFR2 activity using the FGFR2i perturbed the interaction between p85β and FGFR2 (Fig. 4G), whereas neither inavolisib nor lapatinib treatment interfered with the FGFR2–p85β or HER3–p85β interactions (Fig. 4G; Supplementary Fig. S7B). These findings propose that the PI3K regulatory subunit acts as a direct effector protein of FGFR2 and suggested that FGFR2 engages RAS as well as HER3 to mediate PI3K signaling.

Effects of FGFR2 signaling on sensitivity to inavolisib

In HER2+ PIK3CA-mutated cancer, inavolisib seems to promote the degradation of mutated p110α in an RTK-dependent manner; that is, HER2-HER3 recruits mutated PI3K to the cell membrane to promote p110α ubiquitination and degradation, resulting in sustained PI3K pathway inhibition (14). We hypothesized that similar to HER2, FGFR2 activity engaging HER3 may lead to inavolisib-induced ubiquitination of mutated p110α. Treatment with inavolisib induced ubiquitination of p110α in the PIK3CAmut cell lines MFM223 and MFE280, whereas p110α showed no ubiquitination in the PIK3CA WT cell line SUM52PE (Fig. 4H). The FGFR2i blocked p110α ubiquitination specifically in PIK3CAmut and FGFR2-high MFM223 cells. Notably, in FGFR2-low but HER2+ MFE280 cells, lapatinib prevented p110α ubiquitination (Fig. 4H). To validate these findings and measure mutated p110α degradation independently of a WT allele, we engineered the HR+/HER2− and p110α H1047R heterozygously mutated T47D breast cancer cell line to express only the H1047R allele via CRISPR/Cas9 deletion of the WT allele. Specific loss of the PIK3CA WT allele was confirmed at the RNA level (Supplementary Fig. S7E). We then introduced dox-inducible FGFR2 WT or FGFR2 N550K hotspot mutant constructs into the PIK3CA H1047R isogenic line. Overexpression of both WT and N550K FGFR2 led to comparable pathway activation, as demonstrated by phosphorylation of FRS2, HER3, PLCγ, and AKT (Supplementary Fig. S8A, left). Inavolisib treatment promoted mutated p110α ubiquitination in cells overexpressing either FGFR2 variant, whereas control vector–expressing cells showed no p110α ubiquitination (Supplementary Fig. S8A, right). Downstream pathway signaling and HER3 phosphorylation were reduced and increased, respectively, in FGFR2 WT- and N550K-expressing cells upon inavolisib treatment, although FGFR2 N550K-expressing cells seemed to be more affected by inavolisib than the control or FGFR2 WT-expressing cells (Supplementary Fig. S8A, left). FGFR2i treatment blocked inavolisib-mediated ubiquitination of H1047R-mutated p110α in FGFR2 N550K-expressing cells (Supplementary Fig. S8B). Extended inavolisib treatment resulted in the degradation of mutated p110α protein in FGFR2 N550K-overexpressing cells but not in the matched control cells (Supplementary Fig. S8C). The proteasome inhibitor MG132 rescued inavolisib-induced mutated p110α degradation (Supplementary Fig. S8D). Notably, ectopic expression of FGFR2 N550K resulted in moderately increased sensitivity to either the FGFR2i or inavolisib as compared with the respective controls (no dox vs. plus dox; Supplementary Fig. S8E).

In PIK3CAmut MFM223 cells, after treatment with the non–degradation-inducing p110α inhibitor alpelisib, pAKT rebounded after 30 minutes and continued to accumulate for 24 hours. In contrast, pAKT reactivation was markedly delayed with inavolisib treatment and observed only 24 hours after treatment start (Fig. 4I). In PIK3CA WT SUM52PE cells, a similar PI3K pathway feedback response was observed after treatment with either alpelisib or inavolisib (Fig. 4I). Consistently, the difference in potency between the mutated p110α-degrading inavolisib and the nondegrading inhibitor alpelisib was significantly larger in PIK3CAmut cell lines expressing high levels of FGFR2 compared with FGFR2-low cell lines, as measured by the ratio of GR₅₀ values (P = 0.035, Wilcoxon rank-sum test; Fig. 4J). The strength of inhibitor-induced pathway feedback seems dependent on the amount of RTK expression. As predicted, T47D-engineered cells overexpressing FGFR2, but not the matched control cells, displayed pAKT rebound only during alpelisib treatment, whereas treatment with inavolisib led to sustained pathway inhibition (Supplementary Fig. S8F). Collectively, these data provided evidence that high expression of FGFR2, and especially mutated FGFR2, dictates the cellular response to inavolisib by facilitating mutated p110α degradation.

Combined inhibition with inavolisib and the FGFR2i

We next assessed whether the sensitivity to inavolisib in FGFR2-high cell lines could be potentiated by co-targeting FGFR2. Silencing of FGFR2 with RNAi increased sensitivity to inavolisib in five of the six lines tested (Fig. 5A; Supplementary Fig. S9A). In contrast, the depletion of other FGFR paralogs in FGFR2-high cell lines as well as the silencing of FGFR2 or other FGFR in FGFR2-low cell lines had no effect on sensitivity to inavolisib (Fig. 5A; Supplementary Fig. S9B). We co-treated FGFR2-high and -low cell lines with inavolisib and the tool FGFR2i. Consistent with the FGFR2 knockdown results, we found that the FGFR2i synergizes with inavolisib particularly in FGFR2-high cell lines, as determined by Bliss scores and a fixed dose ratio method (Fig. 5B; Supplementary Fig. S10A and S10B; Supplementary Table S5). The nonselective pan-FGFR inhibitors infigratinib, AZD4547 (42), and LY2874455 (43) synergized with inavolisib in a similar manner (Supplementary Fig. S11).

Figure 5.

Inavolisib and FGFR2i combination produces a synergistic effect in FGFR2-high cell lines. A, Representative dose–response curves of SUM52PE, AN3CA2, and HEC151 cell lines transfected with FGFR1–4-selective siRNA and treated at different concentrations of inavolisib for 18 hours. Error bars represent the SD of triplicates. Representative results from experiments (n = 2). B, Drug combination effect on cell lines (n = 13) treated with inavolisib and the FGFR2i as a two-way 10 × 8 serial dilution matrix to calculate the Bliss excess scores for each cell line. Representative results from experiments (n = 2). C, SUM52PE and AN3CA cells were transfected with FGFR1–4-selective siRNA or vehicle control and 0.4-μmol/L and 2-μmol/L inavolisib for 18 hours. Cells were harvested and immunoblotted with antibodies indicated on the left. D, Fitted tumor volumes of MFM223 X2.2 (breast, p110α H1047R, FGFR2-amplified) xenograft model treated daily for 22 days with inavolisib, FGFR2i, or combination treatment. E, Cells treated with the FGFR2i or inavolisib alone or in combination with both inhibitors imaged over time using IncuCyte. PO, orally.

Finally, we evaluated how silencing different FGFR paralogs affected the PI3K signaling network activity during inavolisib treatment. In the FGFR2-high cell lines SUM52PE and ANC3A, FGFR2 depletion using siRNA led to a particularly profound suppression of the phosphorylation of the PI3K downstream effectors AKT, 4E-BP1, and ERK, as compared with inavolisib alone (Fig. 5C). These results suggested that patients with cancers that harbor PIK3CA mutations and FGFR2 overexpression may benefit from combined therapy with inavolisib and an FGFR2i.

Combination treatment regression and resistance delay

Prompted by the in vitro data, we next assessed the activity of the tool FGFR2i in combination with inavolisib using the MFM223 X2.2 tumor xenograft model. This in vivo model is derived from the MFM223 cell line that contains the p110α H1047R hotspot mutation and focal FGFR2 amplification resulting in the expression of E18-truncated FGFR2 (18). We treated immunocompromised mice bearing xenografted MFM223 X2.2 with either vehicle control or the tool FGFR2i at 1, 10, or 30 mg/kg once daily for 4 days. Tumor tissue analysis using Western blotting showed a dramatic suppression of the phosphorylation of FGFR2 downstream effectors in samples derived from mice treated with the FGFR2i at 10 or 30 mg/kg and collected 6 hours after the last dose (Supplementary Fig. S12A). Next, we conducted an efficacy study to evaluate whether inavolisib in combination with the tool FGFR2i inhibitor would improve the therapeutic response in the MFM223 X2.2 xenograft model. Our data showed that single-agent treatment with either inavolisib (25 mg/kg) or the FGFR2i (10 and 30 mg/kg) daily for 22 days did not result in tumor growth inhibition. In great contrast, treatment with the combination of inavolisib with the FGFR2i triggered a robust tumor regression (Fig. 5D; Supplementary Fig. S12A–S12D). Of note, the treatment regimens were well tolerated, and no treatment-associated body weight losses were observed (Supplementary Fig. S12E).

To explore the FGFR2i–inavolisib combination therapy further, we studied whether the emergence of resistance to FGFR2 inhibition could be delayed by a combination treatment with inavolisib. During initial FGFR2i treatment, MFM223 cell line proliferation was strongly inhibited, but eventually, these cells became resistant to the growth-inhibitory effect of the FGFR2i around day 34 after treatment initiation, as evidenced by a rebound in proliferation at this time point (Fig. 5E). Similarly, proliferation of MFM223 cells treated with inavolisib was initially inhibited, but this proliferation rebounded after just 11 days (Fig. 5E). In contrast, however, co-treatment with inavolisib and the tool FGFR2i led to sustained inhibition of proliferation in this MFM223 cell line with no apparent rebound up to day 34 after treatment start (Fig. 5E). Taken together, these findings suggest that the combination of inavolisib and an FGFR2i in PIK3CA-mutated and FGFR2-overexpressed cancer could potentially improve response rates to inavolisib and impede acquired resistance to either PI3K- or FGFR2-targeted single-agent therapies.

Discussion

The single-agent dose-escalation study of the oral p110ɑ-selective inhibitor, inavolisib (GDC-0077), demonstrated a safety profile consistent with on-target p110α inhibition. The MTD was identified as 9 mg once daily. A linear PK profile supported daily dosing. Antitumor activity showed promising preliminary results with an overall response rate of 26% in patients with tumors harboring PIK3CA mutations and measurable disease at baseline. Inavolisib treatment led to PD modulation of tumor avidity by FDG-PET and of PI3K/AKT pathway effectors and proliferation in paired tumor biopsies by IHC and led to decreased PIK3CA MAF in paired ctDNA samples. Intriguingly, we found FGFR2 mutations in the liquid biopsies of two patients who received clinical benefit on single-agent inavolisib. Based on this finding, we explored FGFR2 mutations in a large breast cancer database and conducted preclinical experiments to decipher the role of FGFR2 mutations in conjunction with PIK3CA mutations.

Based on our findings, we proposed a model (Supplementary Fig. S13) to explain the mechanism whereby tumors that express high levels of FGFR2 (FGFR2-high), in contrast to tumors with low expression of FGFR2, exhibit increased sensitivity to inavolisib (GDC-0077) that is further enhanced when combined with an FGFR2i. In FGFR2-dependent tumor models harboring PIK3CA hotspot mutations, we demonstrated that FGFR2 is a key regulator of the PI3K pathway by acting on multiple nodes that influence the PI3K signaling cascade. These include HER3, RAS, and with direct binding of FGFR2 to the PI3K regulatory subunit, p85β. Consequently, FGFR2-high cells were highly dependent on PI3K signaling, which in turn increased their susceptibility to PI3K inhibition. We also found that FGFR2 expression potentiates the degradation of mutated p110α upon inavolisib binding, ultimately leading to an increase in sensitivity to inavolisib.

Previously, we demonstrated that in ERBB2-amplified breast cancer cells, HER2 RTK activity plays a key role in regulating mutated p110α degradation upon treatment with inavolisib by recruiting p110α to the cell membrane (14, 38). In this study, we established that, in HR+/HER2− breast tumor cells, high FGFR2 RTK activity driven through FGFR2 alteration (Supplementary Figs. S3 and S8) potentiates inavolisib-induced degradation of mutated p110α. Based on the degradation-inducing effects of inavolisib and its observed synergy with FGFR2is in FGFR2-altered models, we propose that patients with tumors harboring concurrent alterations in PIK3CA and FGFR2 may achieve more complete and durable responses from combined inhibition of the PI3K and other cross-signaling pathways. These results contribute to advancing the conventional precision oncology paradigm based on a single predictive biomarker to more complex matching algorithms based on multiple alterations including in cooperating signaling nodes and their impact on sensitivity or resistance to specific therapeutic interventions (10, 44–47).

Dual blockade of the FGFR2/PI3K signaling axis using nonselective FGFR inhibitors in combination with PI3K inhibitors has previously been proposed based on preclinical cancer models (48, 49). However, toxicities associated with these drugs have been a major hurdle for their clinical development, both as single agents and especially as combination therapies. Infigratinib and alpelisib treatment combinations were tested in a phase 1b clinical trial, NCT01928459, but there was a high rate (71%) of treatment interruptions and dose reductions (49). Hyperphosphatemia-mediated tissue calcification (50) and grade 3/4 hyperphosphatemia observed in patients treated with infigratinib (49, 51, 52), pemigratinib (53, 54), and futibatinib (55) have limited the efficacy of FGFR nonselective inhibitors in the clinic. In comparison to other p110α-selective inhibitors (56), the more manageable safety profile of inavolisib may enable combinations with selective RTK inhibitors, including newer and more tolerable FGFR2-selective inhibitors such as RLY-4008 and bemarituzumab (57). Development of additional, mutant-selective PI3K inhibitors can potentially improve the therapeutic index of these combinations further; however, it remains to be seen whether lack of inhibition of WT p110α and/or lack of the mutant degradation mechanism represent any potential limitations of these agents.

Our preclinical in vitro and in vivo experiments suggested that the combination of inavolisib and FGFR2-selective inhibitors prolonged the duration of response and delayed resistance to single-agent therapy. Whereas inavolisib has demonstrated the benefit of a p110α-selective inhibitor as a single agent in patients with PIK3CA-mutated, HR+HER2− breast cancer, with an overall response rate of 26% in this population, a clinical benefit rate of 45%, and a PFS rate at 6 months of 37% (95% CI, 15–59), there may be an opportunity to further improve upon that with combination therapy. In summary, the single-agent dose-escalation study of the oral p110ɑ-selective inhibitor, inavolisib, demonstrated a safety profile consistent with on-target PI3K inhibition, evidence of PI3K pathway inhibition, and antitumor activity in a heavily pretreated advanced cancer patient population. Moreover, it illustrated the importance of reverse translational studies from phase I clinical trial data and refined existing drug development paradigms for cell signaling inhibitors.

Supplementary Material

Supplementary Table S1

Cell lines used in this study

Supplementary Table S2

Growth rate inhibition metrics

Supplementary Table S3

Representativeness of study participants

Supplementary Table S4

Study NCT03006172 Arm A participants whose liquid biopsies harbored an FGFR2 mutation

Supplementary Table S5

Drug combination effect of inavolisib and the FGFR2i as measured by a two-way 10 x 8 serial dilution matrix in a 5-day viability assay

Supplementary Figure S1

Foundation Medicine (FMI) FoundationACT (FACT) panel of NGS features selected by model as associated with response

Supplementary Figure S2

FGFR2 alterations co-occur with PIK3CAmut in breast tumors

Supplementary Figure S3

Prevalence of FGFR2 alterations in conjunction with single and multiple PIK3CAmut breast and pan-solid tumor tissue biopsies

Supplementary Figure S4

Relationship between FGFR2 expression and inavolisib sensitivity and pAKT S473 baseline expression

Supplementary Figure S5

Effect of FGFR2-selective inhibitor and non-selective inhibitor infigratinib on normalized growth and downstream FGFR signaling

Supplementary Figure S6

Effects of inavolisib and FGFR2 inhibition on downstream signaling in FGFR2-high and FGFR2-low expressing cell lines

Supplementary Figure S7

FGFR2 controls HER3 to induce p85β binding and signaling in FGFR2 high-expressing cell lines

Supplementary Figure S8

FGFR2 ectopic expression potentiates inavolisib mediated mutant p110α degradation and enhance PI3K activity more than matched control cells

Supplementary Figure S9

Effects of FGFR1/2/3/4 knockdown on inavolisib sensitivity

Supplementary Figure S10

Combination of inavolisib and FGFR2 inhibitor synergy in FGFR2-expressing cell lines

Supplementary Figure S11

Combination effect of inavolisib and FGFR inhibitors in FGFR2-expressing cell lines

Supplementary Figure S12

Evaluation of the anti-tumor activity of FGFR2 inhibitor in combination with inavolisib in MFM223x2.2 xenograft model

Supplementary Figure S13

Proposed mechanistic model of FGFR2-mediated inavolisib sensitivity

Data Availability

To request access to clinical data that support the findings of this study, see https://vivli.org/members/enquiries-about-studies-not-listed-on-the-vivli-platform/, and for up-to-date details on Roche’s Global Policy on the Sharing of Clinical Information, see https://www.roche.com/innovation/process/clinical-trials/data-sharing/. The data supporting the findings of this study, which were originated by Foundation Medicine, Inc. are deidentified data that may be made available upon request and are subject to a license agreement with Flatiron Health and Foundation Medicine; interested researchers should contact cgdb-fmi@flatiron.com to determine licensing terms.

Authors’ Disclosures

D. Juric reports grants from Genentech during the conduct of the study; grants and personal fees from AstraZeneca, Pfizer, Blueprint Medicines, Novartis, Roche, and Eli Lilly; personal fees from Relay Therapeutics and PIC Therapeutics; and grants from Arvinas outside the submitted work. K. Song reports personal fees from Genentech outside the submitted work. R.M. Johnson reports personal fees and other support from Gilead Sciences and Tempus AI outside the submitted work and is a Data Safety and Monitoring Board member for Myeloma Cancer Research Network, Lilly, and Relay Therapeutics. M.K. Accordino reports other support from Genentech during the conduct of the study and personal fees from Disney outside the submitted work. P.L. Bedard reports grants and nonfinancial support from Genentech/Roche during the conduct of the study; grants and personal fees from Bayer, Daiichi Sankyo/UCB Japan, and Boehringer Ingelheim; and grants from Bristol Myers Squibb, Sanofi, AstraZeneca, GlaxoSmithKline, Novartis, Merck, Seattle Genetics, Lilly, Amgen, Bicara, Zymeworks, Medicenna, Takeda, Gilead Sciences, and ProteinQure outside the submitted work. A. Cervantes reports grants from Genentech during the conduct of the study; grants from Merck Serono, Bristol Myers Squibb, Merck Sharp & Dohme, Roche, BeiGene, Bayer, Servier, Ely Lilly, Natera, Novartis, Takeda, and Astelas; and grants and personal fees from AbbVie outside the submitted work. V. Gambardella reports other support from institutional funding from Genentech, Merck Serono, Roche, Beigene, Bayer, Servier, Lilly, Novartis, Takeda, Astelas, Fibrogen, Amcure, Natera, Sierra Oncology, AstraZeneca, Medimmune, Bristol Myers Squibb, and Merck Sharp & Dohme outside the submitted work. E. Hamilton reports grants from Roche during the conduct of the study; grants and other support from Arvinas, AstraZeneca, Daiichi Sankyo, Gilead Sciences, Jazz Pharmaceuticals, Lilly, Mersana, Novartis, Pfizer, Roche/Genentech, and Stemline Therapeutics; grants from AbbVie, Accutar Biotechnology, Artios, AtlasMedx, BeiGene, Bicycle Therapeutics, Biohaven Pharmaceuticals, BioNTech, Compugen, Cullinan, Dantari, Day One Biopharmaceuticals, Duality Biologics, Ellipses Pharma, Elucida Oncology, Exelixis, FujiFilm, Genmab, H3 Biomedicine, Iambic Therapeutics, Immunogen, Inspirna, Inventis Bio, Jacobio, K-Group Beta, Kind Pharmaceuticals, Loxo Oncology, Mabspace Biosciences, Mabwell Bioscience, Marengo Therapeutics, MediLink Therapeutics, Merck, Olema, Orinove, Orum Therapeutics, Pionyr Immunotherapeutics, Prelude Therapeutics, Profound Bio, Regeneron, Relay Therapeutics, Rgenix, Seagen, Shattuck Labs, Simcha Therapeutics, Sutro, Systimmune, Taiho, Tesaro, TheRas, Treadwell Therapeutics, Verastem, Xadcera Biopharmaceutical, and Zymeworks; and other support from BeOne Medicines, Boehringer Ingelheim, Bristol Myers Squibb, Circle Pharma, Halda Therapeutics, Incyclix Bio, IQVIA, Janssen, Jefferies, Johnson and Johnson, Pyxis Oncology, Samsung Bioepis, Shorla Pharma, and Tempus Labs outside the submitted work. A. Italiano reports grants from Roche, Bristol Myers Squibb, Merck Sharp & Dohme, Merck, AstraZeneca, Novartis, and Amgen outside the submitted work. K. Kalinsky reports other support from Genentech/Roche, Gilead, Seattle Genetics, AstraZeneca, Daiichi Sankyo, Puma Biotechnology, Mersana, Menarini Silicon Biosystems, Myovant Sciences, Merck, Eli Lilly, Pfizer, Novartis, Mersana, ProteinQure, Biotheranostics, Regor, and Relay Therapeutics during the conduct of the study and personal fees from ADC Therapeutics outside the submitted work. I.E. Krop reports other support from Genentech during the conduct of the study and personal fees from AstraZeneca, Daiichi/Sankyo, Genentech/Roche, Gilead, Pfizer, Novartis, EMD Serono, Bristol Myers Squibb, and Janssen outside the submitted work. M. Oliveira reports grants, personal fees, and nonfinancial support from Roche during the conduct of the study; grants, personal fees, and nonfinancial support from AstraZeneca and Gilead; personal fees from Daiichi Sankyo, Lilly, Merck Sharp & Dohme, Relay Therapeutics, ProteinQure, Curio Science, iOne; and Medscape, grants and personal fees from Pfizer; personal fees and nonfinancial support from Eisai; and grants from Immutep outside the submitted work. C. Saura reports personal fees from AstraZeneca, Daiichi Sankyo, Eisai Europe, Gilead, Novartis/Pharmalex, Pfizer, Philips Health Works, Pierre Fabre, Puma Biotechnology, F. Hoffmann-La Roche, Seagen, Synthon Biopharmaceuticals, Zymeworks, Relay Therapeutics, Lilly, Menarini, Sanofi, and Merck Sharp & Dohme España; nonfinancial support from SOLTI, outside the submitted work; and other roles including member of the Consensus Panellist Group for the International Multidisciplinary Consensus on the Integration of Radiotherapy With New Systemic Treatments for Breast Cancer (European Society for Radiotherapy and Oncology-endorsed recommendations, Karger Publishers) and Associate Editor and peer reviewer, Breast Care Journal. P. Schmid reports grants and personal fees from AstraZeneca, Merck, Novartis, and Roche; personal fees from Bayer, Boehringer Ingleheim, Pfizer, Puma, Eisai, and Celgene; and grants from Astellas, Genentech, Oncogenex, and Medivation outside the submitted work. N.C. Turner reports advisory board honoraria from AstraZeneca, Bristol Myers Squibb, Lilly, Merck Sharp & Dohme, Novartis, Pfizer, Roche/Genentech, Tesaro, Bicycle Therapeutics, and Taiho and research funding from AstraZeneca, BioRad, Pfizer, Roche/Genentech, Clovis, Merck Sharp & Dohme, and Guardant Health. A. Varga reports other support from AstraZeneca outside the submitted work. S. Gendreau reports other support from Genentech/Roche during the conduct of the study. D. Maddalo is a Genentech employee. S. Martin reports other support from Genentech outside the submitted work. N.M. Sodir reports other support from Revolution Medicines outside the submitted work. E.S. Sokol reports other support from Foundation Medicine and Roche during the conduct of the study. R.L. Yauch reports other support from Genentech outside the submitted work. S. Cheeti is an employee and shareholder of Genentech/Roche. J. Fredrickson reports personal fees from Genentech/Roche outside the submitted work. S. Hilz reports personal fees, nonfinancial support, and other support from Roche/Genentech during the conduct of the study and outside the submitted work. M. Hafner reports other support from Genentech/Roche during the conduct of the study. K.E. Hutchinson is employed by and has received stock options from Roche/Genentech. Y. Jin reports personal fees from Roche during the conduct of the study. U. Peters reports other support from Genentech, a member of the Roche Group, outside the submitted work. D. Zingg reports other support from Genentech/Roche outside the submitted work. S. Royer-Joo reports other support from Genentech/Roche (employer) during the conduct of the study. J.L. Schutzman reports other support from Genentech during the conduct of the study. K.L. Jhaveri reports personal fees from Novartis, Pfizer, Genentech, Eisai, AstraZeneca, Blueprint Medicines, Daiichi Sankyo, Menarini/Stemline, Gilead, Scorpion Therapeutics, Bicycle Therapeutics, Olema Pharmaceuticals, Lilly/Loxo Oncology, Merck Pharmaceuticals, Zymeworks, Halda Therapeutics, Arivinas, and Rayzebio and other support from Novartis, Genentech, AstraZeneca, Pfizer, Lilly/Loxo Oncology, Zymeworks, Gilead, PUMA Biotechnology, Merck Pharmaceuticals, Scorpion Therapeutics, Rayzebio, Eisai, Bicycle Therapeutics, Bridge Bio Oncology Therapeutics, and Blueprint Medicines outside the submitted work. A. Dey is an employee of Genentech and shareholder at Roche. No disclosures were reported by the other authors.

Authors’ Contributions

D. Juric: Data curation, investigation, writing–review and editing. K. Song: Conceptualization, data curation, formal analysis, investigation, writing–original draft, writing–review and editing. R.M. Johnson: Conceptualization, data curation, formal analysis, investigation, writing–original draft, writing–review and editing. M.K. Accordino: Data curation, investigation, writing–review and editing. P.L. Bedard: Data curation, investigation, writing–review and editing. A. Cervantes: Data curation, investigation, writing–review and editing. V. Gambardella: Data curation, investigation, writing–review and editing. E. Hamilton: Data curation, investigation, writing–review and editing. A. Italiano: Data curation, investigation, writing–review and editing. K. Kalinsky: Conceptualization, data curation, supervision, investigation, writing–review and editing. I.E. Krop: Data curation, investigation, writing–review and editing. M. Oliveira: Data curation, visualization, writing–review and editing. C. Saura: Supervision, investigation, writing–review and editing. P. Schmid: Data curation, investigation, writing–review and editing. N.C. Turner: Data curation, investigation, writing–review and editing. A. Varga: Data curation, investigation, writing–review and editing. S. Gendreau: Supervision, investigation, writing–review and editing. M.S. Hwang: Conceptualization, investigation, writing–review and editing. Z. Kuang: Formal analysis, investigation, writing–review and editing. J.T. Lau: Data curation, investigation, writing–review and editing. E. Lin: Data curation, investigation, writing–review and editing. T. Pham: Conceptualization, supervision, investigation, writing–original draft, writing–review and editing. D. Maddalo: Conceptualization, data curation, supervision, investigation, writing–review and editing. M.G. Rees: Conceptualization, data curation, formal analysis, supervision, investigation, writing–review and editing. M.M. Ronan: Data curation, formal analysis, investigation, writing–review and editing. J.A. Roth: Data curation, formal analysis, investigation, writing–review and editing. S. Martin: Conceptualization, data curation, investigation, writing–review and editing. N.M. Sodir: Conceptualization, supervision, investigation, writing–review and editing. E.S. Sokol: Conceptualization, data curation, supervision, investigation, writing–review and editing. Z.J. Whitfield: Data curation, formal analysis, investigation, writing–review and editing. A. Wong: Conceptualization, data curation, supervision, investigation, writing–original draft, writing–review and editing. R.L. Yauch: Formal analysis, supervision, investigation, writing–review and editing. J. Aimi: Conceptualization, data curation, formal analysis, supervision, investigation, writing–original draft, writing–review and editing. S. Cheeti: Formal analysis, investigation, writing–review and editing. J. Fredrickson: Conceptualization, data curation, supervision, investigation, writing–original draft, writing–review and editing. S. Hilz: Formal analysis, investigation, writing–review and editing. M. Hafner: Conceptualization, data curation, formal analysis, supervision, investigation, writing–review and editing. K.E. Hutchinson: Conceptualization, data curation, formal analysis, supervision, investigation, writing–original draft, writing–review and editing. Y. Jin: Conceptualization, formal analysis, supervision, investigation, writing–review and editing. U. Peters: Data curation, supervision, investigation, writing–original draft, writing–review and editing. D. Zingg: Conceptualization, formal analysis, supervision, investigation, writing–original draft, writing–review and editing. S. Royer-Joo: Conceptualization, supervision, investigation, writing–original draft, writing–review and editing. N. Shankar: Data curation, formal analysis, supervision, investigation, writing–review and editing. J.L. Schutzman: Conceptualization, data curation, formal analysis, supervision, investigation, writing–review and editing. K.L. Jhaveri: Data curation, formal analysis, investigation, writing–review and editing. A. Dey: Conceptualization, data curation, formal analysis, supervision, investigation, writing–original draft, writing–review and editing.