PMCCPMCCPMCC

Search tips
Search criteria 

Advanced

 
Logo of wtpaEurope PMCEurope PMC Funders GroupSubmit a Manuscript
 
Cancer Discov. Author manuscript; available in PMC 2017 March 6.
Published in final edited form as:
PMCID: PMC5338732
EMSID: EMS71309

High-Level Clonal FGFR Amplification and Response to FGFR Inhibition in a Translational Clinical Trial

Abstract

FGFR1 and FGFR2 are amplified in many tumor types, yet what determines response to FGFR inhibition in amplified cancers is unknown. In a translational clinical trial, we show that gastric cancers with high-level clonal FGFR2 amplification have a high response rate to the selective FGFR inhibitor AZD4547, whereas cancers with subclonal or low-level amplification did not respond. Using cell lines and patient-derived xenograft models, we show that high-level FGFR2 amplification initiates a distinct oncogene addiction phenotype, characterized by FGFR2-mediated transactivation of alternative receptor kinases, bringing PI3K/mTOR signaling under FGFR control. Signaling in low-level FGFR1-amplified cancers is more restricted to MAPK signaling, limiting sensitivity to FGFR inhibition. Finally, we show that circulating tumor DNA screening can identify high-level clonally amplified cancers. Our data provide a mechanistic understanding of the distinct pattern of oncogene addiction seen in highly amplified cancers and demonstrate the importance of clonality in predicting response to targeted therapy.

Significance

Robust single-agent response to FGFR inhibition is seen only in high-level FGFR-amplified cancers, with copy-number level dictating response to FGFR inhibition in vitro, in vivo, and in the clinic. High-level amplification of FGFR2 is relatively rare in gastric and breast cancers, and we show that screening for amplification in circulating tumor DNA may present a viable strategy to screen patients.

Introduction

Activation of FGFRs is a common oncogenic mechanism, occurring in a subset of nearly all common cancers (1, 2). Activating genetic events in carcinomas include receptor amplification, mutation, and generation of aberrant receptor fusions through translocation (1, 2). There is limited understanding of how these diverse oncogenic events differ in their signaling to downstream pathways and, consequently, of whether the same or different therapeutic approaches may be required to treat cancers with different genetic events.

Amplification of FGFR1 is found in approximately 10% of breast cancers and approximately 15% of squamous lung cancers, with amplification of FGFR2 found in approximately 8% of gastroesophageal cancers and 2% of breast cancers (1). We conducted a translational clinical trial to assess whether cancers with amplification of FGFR1 or FGFR2 respond to the selective FGFR inhibitor AZD4547 (3). Patients were screened for the presence of amplification using criteria defined for detection of HER2 amplification (4). Through the clinical trial we demonstrated that high-level copy-number amplification is required for response to selective FGFR inhibition. Through analysis of functional genomic screens (5), we showed that high-level FGFR2-amplified cell lines have a distinct oncogene addiction phenotype that is characterized by PI3K and mTOR signaling becoming dependent on FGFR signaling. We investigate the factors that predispose to oncogene addiction phenotype, and the implications for response to inhibitors targeting amplified receptor tyrosine kinases (RTK).

Results

FGFR2-Amplified Cancers Have a High Response Rate to AZD4547

We assessed the activity of the FGFR-selective inhibitor AZD4547 in FGFR1-amplified breast cancer and FGFR2-amplified gastroesophageal cancer. We screened 341 patients with advanced cancer by FISH for the presence of gene amplification, identifying FGFR1 amplification in 18% of advanced HER2-negative breast cancers and FGFR2 amplification in 9% of advanced gastroesophageal cancer. Eight patients with FGFR1-amplified breast cancer and 9 with FGFR2-amplified gastroesophageal cancer were treated with AZD4547 in a translational clinical trial (Supplementary Fig. S1A). One FGFR1-amplified (12.5% response rate) and 3 FGFR2-amplified (33% response rate) patients had confirmed responses to AZD4547 (Fig. 1A and B), reaching the predefined criteria for efficacy in the FGFR2-amplified cohort. The responses were durable, with a median progression-free survival of responding patients of 6.6 months (range, 6.2–10.5 months; Fig. 1A). Comparison of 18F-fludeoxyglucose positron emission tomography (FDG-PET) in responding patients between baseline and day 15 demonstrated a substantial reduction in glucose uptake in all responding FGFR2-amplified gastric cancers, although there was no change in the FGFR1-amplified breast cancer (Fig. 1C). The FDG-PET responses were maintained at 8 weeks (Supplementary Fig. S1B). Two additional patients with FGFR2-amplified gastric cancer had a response in day 15 FDG-PET but did not have a confirmed response by RECIST criteria (patients 135 and 99; Supplementary Fig. S1C). Serum phosphate was elevated from baseline in the majority of patients in the study (Supplementary Fig. S1D, P = 0.0002 Wilcoxon matched pairs-signed rank test), as a pharmacodynamic marker of interrupting FGF23 signaling by FGFR inhibition (6). Phosphate level was not correlated with response. Both a continuous and an intermittent schedule of AZD4547 were used in the trial (see Methods). Of the responders, patients 21 and 207 were treated on the intermittent schedule, and patients 269 and 316 received continuous treatment.

Figure 1
Clinical activity of the FGFR inhibitor AZD4547 in FGFR-amplified breast and gastric cancers.

High-Level Clonal FGFR2 Amplification Is a Therapeutic Target for Selective FGFR Inhibitors

We investigated the pathologic basis of response in FGFR-amplified cancers, the relatively high response rate of FGFR2-amplified gastric cancers compared with FGFR1-amplified breast cancers, with FDG-PET reductions only in FGFR2-amplified cancers. We assessed relative FGFR copy number by digital PCR in baseline tumor biopsies of patients treated in the clinical trial. For all patients with gastric cancer, the gastroesophageal disease was biopsied by endoscopy. The response of the site biopsied was concordant with response of other sites of disease. Multiple different tumor sites were biopsied in patients with breast cancer. In the one patient who responded, the biopsied site responded along with other nonbiopsied sites.

Cancers with high copy-number (high-level) FGFR amplification were more likely to respond to AZD4547 (P = 0.0026, Mann–Whitney U test; Fig. 2A), with response observed only in cancers with high-level amplification in this trial. Within FGFR2-amplified cancers, high-level amplification was associated with a substantially higher expression of both FGFR2 mRNA and FGFR2 protein assessed by immunohistochemistry on baseline biopsies (Fig. 2B and Supplementary Fig. S2A and S2B). Truncated isoforms of FGFR2 have been shown to be potentially important for FGFR2 oncogenic transformation (7, 8). Only cancers with high-level FGFR2 amplification had expression of the truncated FGFR2-C3 isoform, although at varying levels, suggesting that the C3 isoform was not critical to established cancers (Supplementary Fig. S2C). However, expression of FGFR2-C3 may have potential as a biomarker for enrichment of patients likely to respond to FGFR inhibition.

Figure 2
High-level clonal amplification of FGFR2 predicts for sensitivity to FGFR inhibitor.

We explored the role of clonality in response to AZD4547 in FGFR2-amplified cancers, using in situ heterogeneity mapping (Supplementary Methods). Two responding patients had clonal homogeneously amplified tumors (>99% of tumor cells FGFR2 amplified), the third patient having insufficient residual baseline tissue to make an assessment, whereas all nonresponding patients had tumors with subclonal heterogeneous amplification with the presence of nonamplified tumor cells (Fig. 2C). In particular, heterogeneity in FGFR2 amplification was noted in patient 135, who had a high-level amplified tumor that did not respond to AZD4547. There was limited evidence of heterogeneity and the presence of nonamplified tumor cells in patient 99 who also did not respond. Both patients 135 and 99 had day 15 FDG-PET responses (Supplementary Fig. S1C), suggesting that subclonality may explain the clinical pattern of FDG response that did not subsequently result in tumor shrinkage and clinical response. We compared paired gene expression between baseline and day 15 on-treatment biopsies with a custom NanoString panel. In the heterogeneously amplified tumor from patient 135, high FGFR2 mRNA expression was lost at day 15 (Fig. 2D), whereas in the homogeneously amplified cancers, high FGFR2 mRNA expression was maintained at day 15. Similarly, in patient 135, FGFR2 copy number was substantially reduced at day 15 (Supplementary Fig. S2D). These findings may reflect the inherent difficulty of sampling in a heterogeneous tumor, but could equally reflect clonal selection by AZD4547. In paired gene expression analysis, the one FGFR1-amplified estrogen receptor–positive breast cancer that responded to AZD4547 had upregulation of estrogen receptor target genes in the day 15 biopsy (Supplementary Table S1), suggesting that upregulated estrogen signaling may have limited sensitivity to the FGFR inhibitor in this tumor.

High-level FGFR2 amplification occurred in only 5% of gastric cancers (7/135, arbitrarily defined as FISH ratio > 5), a prevalence that may present a barrier to future drug development. We assessed whether screening for amplification in circulating free plasma DNA could identify high-level clonal amplified cancers (9). FGFR2 copy number was elevated in the plasma DNA of all three responding patients (Fig. 2E), and also in patient 99, who had a response in day 15 FDG-PET (Supplementary Fig. S1C). In contrast, FGFR2 copy number was not elevated in the plasma DNA of patient 135 with the subclonal amplification, and not in low-level amplified cancers, suggesting that plasma assessment has the potential to screen for high-level and clonal amplified cancers, to overcome the challenge posed by screening for amplifications on tumor biopsies.

High-Level FGFR2-Amplified Cancer Models Are Highly Sensitive to FGFR Inhibition

These data suggested that high-level clonal amplification, in particular for FGFR2-amplified cancers, may be required for response to selective FGFR inhibitors, and we investigated why high-level amplification may associate with distinct addiction to FGFR signaling. We examined FGFR2 copy number in a panel of 74 gastric cancer cell lines (n = 25) and tumors (n = 49). A distinct bimodal pattern of FGFR2 copy number was observed (bimodality index 2.74; Fig. 3A), suggesting a distinct high-level FGFR2-amplified group (10). In contrast, in a panel of 82 breast cancer cell lines (n = 22) and tumors (n = 60), there was a less pronounced bimodal distribution in FGFR1 amplification (bimodality index 1.3; Fig. 3A). This suggested that high-level amplification was seen more distinctly in FGFR2-amplified cancers.

Figure 3
High-level FGFR2-amplified cell lines are highly sensitive to FGFR inhibition.

In a panel of FGFR-driven cell lines (Supplementary Table S2), FGFR autophosphorylation (Tyr653/654 loop phosphorylation) was substantially increased in high-level FGFR2-amplified cell lines (Fig. 3B), suggesting that high-level FGFR2 amplification resulted in FGFR hyperactivation, potentially reflecting the very high level of FGFR2 copy number seen in FGFR2-amplified cell lines (Supplementary Fig. S3A). Furthermore, FGFR2 high-level amplified cell lines were more sensitive to the FGFR inhibitors AZD4547 and PD173074 in vitro than low-level FGFR1-amplified cell lines, both in terms of lower EC50 and increased maximal effect (Fig. 3C and D). Similarly, FGFR2-amplified cell lines were more sensitive to PD173074 than FGFR2-mutant cell lines (P = 0.0357, Mann–Whitney U test; Supplementary Fig. S3B), and the level of copy-number variation (CNV) correlated with sensitivity to FGFR inhibition (R = 0.8095, P = 0.0218; Supplementary Fig. S3C). In addition, AZD4547 robustly increased caspase-3/7 activation only in FGFR2-amplified cell lines (Fig. 3E).

To validate these findings, we established patient-derived xenografts (PDX) in nude mice from the baseline biopsies of two FGFR2 clonally amplified patients who responded to AZD4547 in the clinic (Fig. 1). To validate the models, we performed whole-exome sequencing on the patients’ archival diagnostic tumor biopsies, baseline tumor biopsies, and germline DNA, and on the established PDX models. There was high agreement in somatic mutations between the PDX models and the patients’ baseline tumors, suggesting an absence of substantial genetic drift and that the models were an accurate representation of the baseline tumor biopsies (FG42/pt316 R = 0.86, FG51/pt269 R = 0.85; Supplementary Fig. S3D). Both PDX models, FG51 (pt269) and FG42 (pt316), were highly sensitive to AZD4547 in vivo: in FG51 median 70% tumor shrinkage and 2 of 5 complete responses, and in FG42 median 74% tumor shrinkage and 0 of 5 complete responses (Fig. 3F). AZD4547 induced an apoptotic response as demonstrated by elevated levels of cleaved caspase-3 in PDX tumor lysates (Fig. 3G).

High-Level FGFR2 Amplification Initiates a Distinct Oncogene Addiction Phenotype Characterized by FGFR-Dependent Control of PI3K–AKT Signaling

To investigate why high-level FGFR2-amplified models demonstrated high sensitivity to FGFR inhibitors, we analyzed a functional genomic screen of a panel of FGFR-driven cell lines transfected with an siRNA screening library directed against the kinome and phosphatome (5). Using supervised hierarchical clustering, we identified the siRNAs that selectively modified sensitivity to the FGFR inhibitor PD173074 (5) in FGFR2-amplified cell lines, compared to other FGFR-driven cell lines, identifying a distinct set of siRNAs that were strongly enriched for components of the PI3K pathway and canonical NFκB signaling (Fig. 4A and Supplementary Fig. S4A). Increased sensitivity to FGFR inhibition was confirmed with multiple different siRNA for the NFκB signaling pathway components IKBKB and MAP3K7 (Supplementary Fig. S4B). PTEN siRNA reduced the sensitivity of FGFR2-amplified cell lines to PD173074 with little effect in other FGFR-driven cell lines (Fig. 4A and Supplementary Fig. S4A; P = 0.009 Mann–Whitney U test). In validation work, PTEN silencing reduced the sensitivity of the FGFR2-amplified breast cancer cell line SUM52 to PD173074 in clonogenic assays (Fig. 4B), induced AKT phosphorylation, and increased the concentration of PD173074 required to inhibit AKT phosphorylation (Fig. 4C).

Figure 4
Functional genomic screens reveal that PI3K and mTOR signaling contributes to the sensitivity of FGFR2-amplified cell lines to FGFR inhibition.

These data suggested that FGFR2-amplified cell lines may show a distinct reliance on PI3K signaling compared to other FGFR-driven cell lines. Interestingly, in gastric cancer data from The Cancer Genome Atlas, PIK3CA mutation occurred in 23% (48/206) of nonamplified cancers and in 0% (0/13, P = 0.036 Fisher exact test) of FGFR2-amplified cancers, suggesting distinct mechanisms of PI3K pathway activation (11). We studied downstream signaling in response to PD173074 across a panel of FGFR-driven cell lines (Fig. 4D and Supplementary Fig. S4C). Phosphorylation of ERK1/2 was reduced in all cell lines by PD173074 (Fig. 4D and Supplementary Fig. S4C) in keeping with the physiologic effects of FGFR signaling through the MEK–ERK pathway (1). In contrast, only in FGFR2-amplified cell lines was AKT Ser473 phosphorylation reduced by PD173074, with no reduction in FGFR1-amplified cell lines as well as minimal reduction in FGFR2- and FGFR3-mutant cell lines (Fig. 4D and Supplementary Fig. S4C). In FGFR1-amplified cell lines, PD173074 also had little effect on AKT Thr308 or PDK1 Ser241 phosphorylation (Supplementary Fig. S4D). We established in vitro primary spheroid cultures from the FGFR2-amplified PDXs FG51 (pt269) and FG42 (pt316). In these primary spheroid cultures, AZD4547 inhibited AKT phosphorylation (Fig. 4E), suggesting that a similar pattern of FGFR-dependent PI3K signaling was seen in patient-derived material.

AZD4547 inhibited 4EBP1 phosphorylation in FGFR2- amplified cell lines (Fig. 4F) and PDX-derived spheroids (Fig. 4E), suggesting that mTOR signaling was substantially inhibited by FGFR inhibition. In contrast, AZD4547 had minimal effect on 4EBP1 phosphorylation in FGFR1-amplified cell lines (Fig. 4F). AZD4547 decreased levels of phospho-S6 in both FGFR1- (2/3) and FGFR2- (2/3) amplified cell lines, in keeping with the potential regulation of S6 phosphorylation by MEK–ERK MAPK signaling (Supplementary Fig. S4E). In FGFR1-amplified cell lines, inhibition of mTOR was synergistic with FGFR inhibition (Supplementary Fig. 4F), converting a cytostatic effect of FGFR inhibition alone to a cytotoxic effect (Supplementary Fig. S4G), illustrating the importance of mTOR inhibition in initiating a cytotoxic response. Finally, in FGFR2-amplified cell lines, AZD4547 inhibited eIF4G loading onto 7methyl GTP (Fig. 4G) and inhibited nascent protein synthesis determined using incorporation of the methionine analog l-azidohomoalanine (Fig. 4H), demonstrating inhibition of mTOR signaling and cap-dependent translation.

Amplified FGFR2 Activates PI3K in Part through Transactivation of Multiple Alternative RTKs

We set out to determine how PI3K signaling becomes FGFR dependent in highly amplified cell lines. FGFRs signal to PI3K in a canonical fashion through binding of the FGFR-specific adapter protein FRS2 to GRB2 and GAB1 (12). Silencing GRB2 modestly increased AKT phosphorylation in the SUM52 FGFR2-amplified cell line, and silencing FRS2α resulted in only a minor reduction of AKT phosphorylation (Supplementary Fig. S5A and S5B). In addition, the PI3K regulatory subunit p85 (PIK3R1) did not appear to associate with FGFR2 as assayed by immunoprecipitation (Fig. 5A), suggesting that FGFR2 signaling controlled PI3K signaling through a noncanonical mechanism.

Figure 5
Hyperactive FGFR2 in high-level amplified cell lines signals to PI3K indirectly through ERBB3 and IRS1/IGF1R.

To explore how FGFR2 activated PI3K, we assessed the phosphorylation of 49 RTKs in response to FGFR inhibition, noting reduced tyrosine phosphorylation of all RTKs assayed in SUM52 cells (Fig. 5B), and FGFR3, ERBB3, and partially EGFR in the FGFR2-amplified gastric cancer SNU16 cell line (Fig. 5B). Similarly, in the PDX model FG42 (pt316), there was a similar pattern of phosphorylation of FGFR2, EGFR, and ERBB3, with substantial reduction in phosphorylation of all RTKs to FGFR inhibition in RTK arrays and by Western blot (Fig. 5C). In contrast, PD173074 treatment had little effect on RTK phosphorylation in the FGFR1-amplified cell line JMSU1 (Supplementary Fig. S5C). Phosphorylation of ERBB3 Tyr1289 and EGFR Tyr1068 was inhibited by PD173074 in FGFR2-amplified cell lines (Supplementary Fig. S5D), with ERBB3 phosphorylation unaffected by inhibition of EGFR/HER2 with gefitinib or lapatinib (Supplementary Fig. S5E).

The interaction between ERBB3 and p85 PI3K assayed by immunoprecipitation was inhibited by PD173074 (Fig. 5D), and silencing of ERBB3 with multiple different siRNAs partially reduced AKT phosphorylation in SUM52 and SNU16 cells (Fig. 5E). ERBB3 was not observed to interact with FGFR2, suggesting that FGFR2 may not phosphorylate ERBB3 directly (Fig. 5D). Among a panel of cell lines, SNU16 cells were observed to have high expression of IGF1R (Supplementary Fig. S5F). In contrast to ERBB3, IGF1R phosphorylation was not decreased by PD173074 in SNU16 cells (Fig. 5B and F), whereas inhibition of IGF1R with AEW541 also blocked AKT phosphorylation (Fig. 5G) at concentrations that had no effect on FGFR2 autophosphorylation (Supplementary Fig. S5G). Similarly, levels of AKT phosphorylation were decreased by knockdown of IGF1R using siRNA (Supplementary Fig. S5H). Therefore, both FGFR2 and IGF1R signaling were necessary for signaling to PI3K–AKT in SNU16 cells. Activation of PI3K catalytic activity requires binding of GTP bound RAS to p110α, or GTP bound CDC42/RAC1 to p110β (13), and release from the inhibitory effects of the p85 regulatory subunit (14). Inhibition of FGFR signaling with PD173074, not IGF1R signaling with AEW541, reduced RAS activation in SNU16 cells (Fig. 5H). Conversely, p85 PI3K that did not associate with FGFR2 (Fig. 5A) bound to IRS1 in an IGF1R kinase–dependent fashion (Fig. 5I). Therefore, PI3K pathway signaling is activated through the combined effects of FGFR2-dependent activation of RAS, and p85 binding to ERBB3 and IGF1R/IRS1. FGFR2 signaling transphosphorylates ERBB3 to act as a scaffold to bind p85, with IGF1R signaling promoting IRS1 binding to p85, to thereby activate PI3K signaling.

Discussion

The diversity of mechanisms through which FGFR signaling is activated in cancer presents a clinical translational challenge: How similar or dissimilar are the different mechanisms of activating FGFR signaling? With a translational clinical trial, we identify that cancers with high-level FGFR2 amplification form a distinct group characterized by a strong oncogene addiction phenotype and high sensitivity to AZD4547. High-level amplification initiates distinct signaling characterized by transactivation of alternative RTKs to bring PI3K and mTOR signaling under control of FGFR2.

Our study has implications for targeting amplified RTKs. The archetypal model for targeting amplified RTKs is that of HER2 amplification in breast cancer (15), where the level of HER2 amplification does not affect sensitivity to the HER2-targeting antibody trastuzumab (16). Our data suggest that HER2 amplification in breast cancer may be a limited model for targeting other amplified RTKs, and that criteria developed to identify cancers sensitive to HER2 targeting (a HER2 to centromere copy-number ratio of ≥2, and/or absolute HER2 copy number of ≥6) may not necessarily translate to other RTKs. HER2 has unique biology among RTKs, lacking an activating ligand and with inherent potential for constitutive activation when overexpressed, and a very tight relationship between the presence of amplification of any copy-number level and overexpression of HER2 (4). In contrast, high-level FGFR copy-number amplification is associated with sensitivity to FGFR inhibition, with high-level amplification resulting in elevated expression of FGFR2, initiating a distinct oncogene addiction phenotype. In addition, low-level FGFR1/2 amplification does not strongly associate with FGFR overexpression (4). Our data strongly suggest that clinical trials targeting FGFR, and potentially other RTKs, should consider higher thresholds for amplification. We also demonstrate the importance of assessing amplification clonality in predicting durable responses to therapy, the potential to assess the degree of heterogeneity using automated in situ heterogeneity mapping of FISH, and that it may be possible to identify patients with highly FGFR2-amplified gastric cancers by analysis of FGFR2 copy number in cell-free plasma DNA, providing a potential simple strategy to screen patients for rare amplifications.

We show that high-level FGFR2 amplification initiates an oncogene addiction phenotype in part through activation of alternative receptor kinases that brings PI3K pathway signaling under the control of FGFR signaling. FGFR2 lacks consensus-binding sites for the PI3K p85 regulatory subunit, and we show that overexpressed FGFR2 activates PI3K signaling indirectly through cooperation with other RTKs, including ERBB3 (17) and IGF1R. FGFR2 transphosphorylates ERBB3, via a mechanism not identified in these studies, to promote binding to p85 PI3K, with ERBB3 acting as a scaffold protein. In contrast, FGFR2 cooperates with IGF1R in a related but distinct fashion, where FGFR2 promotes PI3K signaling through RAS activation and IGF1R signaling promotes p85 binding via IRS1 (Supplementary Fig. S6). For lower-level FGFR1-amplified cell lines, PI3K and mTOR signaling is not blocked by FGFR inhibition, and this limits the sensitivity of these cell lines to FGFR inhibition, characterized by a lack of FGD-PET changes seen in the FGFR1-amplified responding patient in our clinical–translational trial. Both PI3K and mTOR inhibitors synergize with FGFR inhibition in these cell lines (Supplementary Fig. S4), identifying a potential combination strategy for FGFR1-amplified cancers that do not exhibit a strong oncogene addiction phenotype that can be investigated in future research.

In this study, we identify the importance of high-level clonal amplification in predicting response to FGFR-selective inhibitors. High-level amplifications are of relatively low prevalence, and we show that high-level clonal FGFR2 amplification can be detected through ctDNA screening, opening up a screening strategy to facilitate future development of drugs targeting amplified RTKs.

Methods

Clinical Trial Design

The FGFR trial (EudraCT No. 2011-003718-18) is a phase II, open-label, nonrandomized study of AZD4547 in patients with previously treated advanced FGFR-amplified cancer. Written informed consent was obtained from all patients. The study was carried out in accordance with the Declaration of Helsinki and approved by local institutional review boards. The study originally consisted of three independent tumor cohorts (FGFR1-amplified breast and squamous non–small cell lung cancers and FGFR2-amplified gastroesophageal cancers); however, due to poor accrual the lung cohort was closed to further recruitment. FGFR amplification testing was performed centrally on archival tissue, and an FGFR ratio of ≥2.0 (FGFR1:CEP8 and FGFR2:CEP10) was required for study entry. Prior to treatment, patients underwent a biopsy and PET-CT, repeated on study between days 10 and 14. Treatment consisted of AZD4547 80 mg twice daily (initially on an intermittent schedule of 2 weeks on, 1 week off, which was subsequently amended to continuous dosing). Primary endpoint was confirmed overall response rate. The study followed a Simon 2 stage, optimal design. One or more responses were required in the initial 9 patients in each cohort, to recruit a total of 17 patients. Three or more patients were required to conclude that the cohort had sufficient efficacy for further study.

Cell Lines and Materials

Bladder cell lines (RT112M, MGHU3, JMSU1, 97.7, 94.10) were obtained from the laboratory of M.A. Knowles (Leeds University, UK; ref. 18) in 2010. Ocum2M cells were obtained from the laboratory of Masakazu Yashiro (Department of Surgical Oncology, Osaka City University Graduate School of Medicine, Osaka, Japan) in 2010. All other cell lines were obtained from the laboratory of A. Ashworth (Gene Function, Institute of Cancer Research, London) in 2014. CAL120, MFE296, MFE280, Kato III, and An3CA were originally obtained from the CTS cell line service. BT20, BT483, BT549, CAMA1, DMS114, Hs578T, HCC1143, MDAMB134, MDAMB157, MDAMB175, MDAMB231, MDAMB361, MDAMB453, MCF12A, NCI-H1581, NCI-H520, T47D, and ZR-75-1 cells were originally obtained from the ATCC. MFM223, SUM44, SUM52, SUM149, and SNU16 cells were originally obtained from Asterand. All cell lines were banked in multiple aliquots on receipt to reduce risk of phenotypic drift, and identity was confirmed by STR profiling with the PowerPlex 1.2 System (Promega). Cultured cell lines were subjected to STR profiling after approximately 20 passages. In addition, cultured cell lines were tested once per month for the presence of Mycoplasma. Cell lines were maintained in phenol red free DMEM, DMEM/F12, or RPMI with 10% FBS (PAA gold) and 2 mmol/L l-glutamine (Sigma-Aldrich).

Antibodies used were phosphorylated (p) AKT Ser473 (4058), pAKT T308 (2965), AKT (4691), p4EBP1 T37/46 (2855), 4EBP1 (9452), pERK1/2-Thr202/Tyr204 (4370), ERK1, 2 (9102), EGFR (2232), pEGFR Y1068 (3777), pERBB3 Y1289 (4791), pERBB2 Y1221/1222 (2249), pFGFR Y653/654 (3471), pFRS2a (3864), pIGF1R (3918), IGF1R (9750), p85 PI3K (4292), PIK3CA (4255), PTEN (9559), phospho-ribosomal protein S6 (5364), ribosomal protein S6 (2217; all Cell Signaling Technology); β-Actin (A5441; Sigma); FGFR2 (sc-122), EGFR (sc-03), PARP1 (sc-8003), and FRS2 (sc-8318; Santa Cruz Biotechnology); ERBB3 (ab32121) and FGFR1 (ab76464; Abcam); Grb2 (610111; BD Transduction Laboratories); p85 PI3K (ABS233; Millipore); and AZD4547 and AZD8055 were provided by AstraZeneca. CI-1040, GDC0941, KU0063794, and rapamycin were purchased from Selleckchem; PD173074 was from Sigma; gefitinib was from Tocris; NVP-AEW541 was from Cayman Chemicals; siRNAs were from Dharmacon nontargeting siRNA Pool#2, PTEN (MU-003023-02) set of 4, PLK1 (siPLK1, M-003290-01), FR2Sa (MU-006440-02) set of 4, Grb2 (MU-019220-00) set of 4, ERBB3 (MU-003127-03) set of 4, and IGF1R (MU-003012-05) set of 4. siRNAs against IKBKB and MAP7K3 were obtained from Dharmacon as part of a custom validation panel.

Tumor Samples

Additional tumor samples for analysis were obtained from existing trials. Gastric cancer samples (n = 40) were from the MAGIC trial (19). Breast cancer samples (n = 42) were from tissue-collection studies approved by multicenter research ethics committees (ref. nos. 10/H0805/50 and 11/LO1595). Written informed consent was obtained from all patients. The study was carried out in accordance with the Declaration of Helsinki and approved by local institutional review boards.

Purification of DNA and RNA

DNA and RNA were extracted from fresh-frozen tumor samples using AllPrep micro DNA/RNA extraction kit (QIAgen 80284) and from FFPE tumor samples using AllPrep DNA/RNA FFPE extraction kit (QIAGEN 80234). In both cases, a section was stained with hematoxylin and eosin and the tumor marked out by a pathologist. Sections were cut and stained with nuclear fast red and the tumor macrodissected prior to nucleic acid extraction using the appropriate kit. Plasma DNA was extracted from 2 mL of plasma using QIAamp Circulating Nucleic Acid Kit (QIAGEN 55114) according to the manufacturer’s guidelines.

Purified RNA was quantified using the Qubit RNA HS Assay Kit (Life Technologies; Q32855). Purified DNA was quantified using the Qubit dsDNA HS Assay Kit (Life Technologies; Q32854). Plasma DNA was quantified by ddPCR using the RPPH1 CNV reference assay to calculate copies/well and multiplying by the c-value (3.3 pg), an estimate of the mass of a single haploid human genome.

Droplet Digital PCR

Digital PCR was performed on a QX100 droplet PCR system (Bio-Rad). PCR reactions were prepared as previously described (9, 20). Briefly, emulsified PCR reactions were run on a 96-well plate on a G-Storm GS4 thermal cycler incubating the plates at 95°C for 10 minutes followed by 40 cycles of 95°C for 15 seconds, 60°C for 60 seconds, followed by 10 minutes of incubation at 98°C. Plates were read on a Bio-Rad QX100 droplet reader using QuantaSoft v1.6.6.0320 software. CNV and gene expression for target genes were calculated as a ratio with multiplexed reference genes (Supplementary Table S3). CNV assays were performed using 1 to 3 ng genomic DNA, aiming to obtain 300 to 600 reference droplets.

For gene expression assays, cDNA was prepared using a SuperScript III First Strand Kit (Life Technologies; 18080-051) according to the manufacturer’s guidelines, using 50 to 200 ng total RNA primed with random hexamers. ddPCR gene expression reactions (Supplementary Table S3) were typically set up with 1 to 5 ng RNA equivalent of cDNA. Target expression was normalized using β-Actin and GAPDH reference assays.

In Situ Heterogeneity Mapping

Briefly, fluorescent sections were scanned into the MIRAX (Pannoramic) scanner at high magnification in the x, y, and z planes and analyzed using custom HALO software. Tumor compartments were marked by a pathologist and the entire tumor area was analyzed for amplified and nonamplified tumor cells. Amplified cells were determined as those with a ratio of FGFR2:CEP10 >2, and assessed only in cells with CEP10 signals ≥2. The percentage of amplified tumor cells and total copy number were calculated. A manuscript describing in situ heterogeneity mapping is under submission.

NanoString

Probe sequences were custom designed and manufactured by NanoString. Multiple probes were included against key genes, including FGFR2. Probe specificity was confirmed using BLAT and Array-viewer (Browser and Land software). The codeset was validated with cell lines and clinical tissue known to overexpress FGFR2. In addition to genes of interest, the codeset included a number of housekeeping genes to correct for RNA input amount and/or quality. A positive control (Universal Human Reference RNA, Stratagene, catalog #75000-41) was run routinely to ensure consistency between runs. Input total RNA amount was 100 ng or 5 μL neat RNA for more dilute/poor-quality samples (based on internal quality control criteria).

Protocol was followed according to standard nCounter instructions. GEN2 Prep Station incubation time was set at the higher sensitivity setting (3 hours), and 280 fields of view were routinely captured unless otherwise noted.

Data were normalized through an internally developed Pipeline Pilot Tool (publicly available for use on the Comprehensive R Archive Network). In brief, data were log2 transformed after being normalized in two steps: raw NanoString counts were first background adjusted with a Truncated Poisson correction using negative control spikes followed by a technical normalization using positive control spikes. Data were then corrected for input amount variation through a Sigmoid shrunken slope normalization step using the mean expression of housekeeping genes. A transcript was designated as not detected if the raw count was below the average of the internal negative control raw counts plus 2 standard deviations.

siRNA Screen

Screening was done in 384-well plates with a Dharmacon siGENOME SMART pools library targeting all known protein kinases and phosphatases, as described previously (5). To assess the effect of siRNA on growth/survival, the effect of siRNA in the vehicle plates was expressed as a Z score, with the standard deviation estimated from the median absolute deviation. To assess the effect of siRNA on sensitivity to PD173074, the log2 ratio between growth in PD173074 plates and vehicle plates was assessed and expressed as a Z score.

Western Blotting

Cell lines were grown on 35-mm plates, treated as indicated, and lysed in NP40 lysis buffer (1% v/v NP40, 10 mmol/L Tris–Cl pH8, 150 mmol/L NaCl, 1 mmol/L EDTA, 1 mmol/L DTT) supplemented with phosphatase (5 mmol/L Na4P2O7, 50 mmol/L NaF and 1 mmol/L NaVO4) and protease inhibitor cocktail (Roche; 11697498001). Cells were reverse transfected with siRNA 72 hours prior to lysis. Western blots were carried out with precast TA or Bis-Tris gels (Life Technologies).

In Vitro Cell Line Assessment

Clonogenic assays were conducted in 6-well plates, with 1,000 cells seeded per well, and 24 hours later, cells were exposed to vehicle, or the indicated treatments followed by growth in media for 2 weeks to allow colony growth. Colonies were fixed, stained with sulforhodamine B, and counted. For PTEN siRNA, clonogenic cells were treated for 1 week with PD173074 before washout. For short-term survival assays, cells were exposed to indicated drugs with survival assessed after 72 hours of exposure with a CellTiter-Glo cell viability assay (Promega). To assess the effect of siRNA on drug sensitivity, cells were reverse transfected at a final siRNA concentration of 20 nmol/L, and at 48 hours after transfection, plates were exposed to compound, with survival assessed after 72 hours of exposure to the drug. To assess synergy, cell lines were plated in 384-well plates, and the following day exposed to fixed-ratio combinations of indicated drugs for 72 hours, with combination index assessed according to Chou and Talalay (21) using Calcusyn v2.1 (BIOSOFT).

Whole Exome Sequencing

Genomic DNA (30–200 ng) was fragmented to 200 bp using a Covaris E Series, and the resultant libraries were subjected to DNA Capture using a SureSelect XT Human All Exon v4 kit (Agilent) following the manufacturer’s instructions. Final libraries were quantified using qPCR and clustered at a molarity of 14.5 pmol/L, and sequencing was performed on an Illumina HiSeq 2000 using 2 × 75 cycles of version 3 SBS chemistry. Reads were aligned to the human reference genome (GRCh37) using Burrows-Wheeler Algorithm (v0.7.5a; ref. 22). PCR duplicates were filtered out from the subsequent analysis using Picard Tools (v1.94) and variants were called using the GATK pipeline (v2.3.9) best practices (23). Somatic changes among germline, cancer samples, and the PDX samples were investigated using MuTect (v1.1.4; ref. 24) and filtered for on-target regions using bedTools (v2.17.0). Comparisons between allele frequencies of somatic mutations in the cancer and PDX samples were investigated using R (3.1.2). All sequencing data have been deposited in the National Center for Biotechnology Information Sequence Read Archive under accession SRP072158.

Apoptosis Assessment

To assess apoptosis, activated caspase-3/7 activity was assessed using the Caspase-Glo 3/7 Assay according to the manufacturer’s instructions (Promega, G8090) and adjusted for cell number as assessed by CellTiter-Glo.

Phosphorylated RTK Arrays

Cells and PDXs were treated as indicated; lysates were prepared using Lysis Buffer 17 and analyzed using Human Phospho-RTK arrays (R&D Systems, ARY001B) according to the manufacturer’s guidelines.

Immunoprecipitation

Cells were treated as indicated, and lysates were prepared using NP40 lysis buffer (without DTT). The antibodies used for immunoprecipitation were p85 PI3K (ABS233; Millipore), ERBB3 (Thermo Scientific; MS-262-P1) and FGFR2 (Santa Cruz; SC122). Control IPs were performed with normal mouse (SC2025) and rabbit IgG (SC2027, both Santa Cruz Biotechnologies). Veriblot anti-rabbit IgG (ab131366) and anti-mouse IgG (ab131368, both Abcam) HRP-conjugated secondary antibodies were used for IPs. Total cellular protein (500–1,000 mg) was incubated with the antibodies and the protein complexes precipitated using Protein G coated Dynabeads (Life Technologies; 10001D).

RAS Activation Assay

RAS activation was determined using the RAS activation assay kit (Millipore; 17-218) according to the manufacturer’s guidelines. Briefly, cell lysates were prepared using Mg2+ Lysis/wash buffer (MLB). Samples were incubated with a 10-μg RAS assay reagent (RAF1 RAS binding domain-agarose) and incubated for 45 minutes at 4°C. Agarose beads were pelleted by centrifugation and washed 3 times with MLB. Agarose beads were resuspended in NuPAGE LDS sample buffer (Life Technologies). Samples were subjected to Western blotting as previously described and blots probed using anti-RAS antibody (clone RAS10, 05-516; Millipore).

Methyl 7- GTP Pulldown

Cells were lysed in NP40 lysis buffer. Pulldown mixes were prepared with 200 μL containing 200 μg total cellular protein and 40 μL 7methyl GTP-sepharose 4B beads (GE Healthcare; 275025) to give a total volume of 240 μL. Before use beads were washed twice with NP40 lysis buffer. Pulldown mixes were incubated overnight at 4°C. Beads were washed twice with NP40 lysis buffer, bound complexes eluted in Western sample loading buffer and resolved on precast Bis–Tris gels (Life Technologies).

AHA Translation Assay

Protein synthesis was assessed using the Click IT AHA–Alexa Fluor 488 Protein Synthesis HCS Assay (Invitrogen C10289). l-azidohomoalanine (AHA) is an analogue of methionine that contains an azide moiety. AHA that has been incorporated into proteins can be detected by ligation of a fluorescently labeled alkyne.

Cells were grown on coverslips for 2 days; the media were changed and treated with 50 nmol/L AZD4547 for 24 hours. Experimental conditions were removed, and the cells were washed with methionine-free RPMI media (Sigma; R7513). Cells were treated with AHA prepared according to the manufacturer’s guidelines in methionine-free RPMI media for 30 minutes. The AHA media were removed, the cells washed twice with sterile cold PBS and then fixed with 4% paraformaldehyde. Coverslips were washed twice with 3% w/v BSA in PBS and then permeabilized in 0.5% v/v Triton-X100. Coverslips were washed before incubation with the Alexa Fluor 488 alkyne reaction mix for 30 minutes. Coverslips were washed, and cells were counterstained with Hoechst 33342. The coverslips were mounted onto glass slides with Vectastain (Vector, H1000). Fluorescently stained cells were imaged using an SP2 system Leica confocal microscope, and images were analyzed using Adobe Photoshop Extended CS5.

Patient-Derived and Cell Line Xenografts

All animal experiments were performed under a UK Home Office Project License assessed by an ethical review committee.

PDXs were established from fresh core biopsies collected into DMEM containing penicillin and streptomycin. Tissue samples were washed with three changes of DMEM containing antibiotics and implanted subcutaneously on either the flank or intrascapular of male BALB/c nude mice. Once tumors were established, they were collected, dissected, and implanted into more male BALB/c nude mice for therapeutic experiments using AZD4547. Tumors were size-matched and randomized to control (vehicle, 0.5% w/v hydroxypropylmethylcellulose, 0.1% v/v polysorbate 80 in water) or AZD4547 (5 mg/kg every day, in vehicle, unless stated otherwise). Mice were dosed daily via oral gavage.

NCI-H1581 xenografts were generated by subcutaneous injection of 5 × 106 cells in 50% v/v Growth Factor Reduced Matrigel (BD Biosciences, 356231) in 40 BALB/c nude mice. After 7 days, tumors were measured, size-matched, and the mice were randomized to 4 experimental groups, including control (vehicle), AZD4547 (5 mg/kg every day, in vehicle), AZD8055 (20 mg/kg every day, in vehicle), and AZD4547 + AZD8055 (5 mg/kg every day and 20 mg/kg every day, respectively, in vehicle). Mice were dosed daily via oral gavage. Tumor size was assessed 3 times a week and expressed relative to the size at the start of treatment. To assess the signaling effect of the combinations, mice with NCI-H1581 xenografts were dosed daily for 3 days. Mice were sacrificed 6 hours after the last treatment and tumors fixed with 4% paraformaldehyde in phosphate buffered saline.

Immunohistochemistry

Following deparafinization and rehydration, antigen retrieval was followed by blocking of endogenous peroxidize activity using H2O2 block (Dako Envision Flex Kit; K8000).

For FGFR2 IHC, antigen retrieval was done with pressure cooking in pH 9 retrieval buffer (DakoS2367). Sections were rinsed with TBS-Tween 0.05% and endogenous peroxidise blocked using H2O2 block (Dako Envision Flex Kit; K8000). Sections were incubated for 1 hour with 1/10,000 FGFR2 antibody (AZ AGG 2935-1C11, mouse monoclonal, from AstraZeneca), followed by Envision+/HRP (Dako; K4001) for 30 minutes and staining with 3,3′-diaminobenzadine (DAB). Sections were washed and counterstained in Gill’s Hematoxylin.

Supplementary Material

Note: Supplementary data for this article are available at Cancer Discovery Online (http://cancerdiscovery.aacrjournals.org/).

Supp Fig 1

Supp Fig 2

Supp Fig 3

Supp Fig 4

Supp Fig 5

Supp Fig 6

Supplementary methods and tables

Acknowledgments

Grant Support

This research has been supported by Cancer Research UK grant C30746/A16642. N.C. Turner acknowledges generous support from both Breast Cancer Now and the Mary-Jean Mitchell Green Foundation. E. Smyth, G. Brown, S. Chua, I. Chau, S. Popat, D. Cunningham, and N.C. Turner acknowledge the support of the NIHR ICR/RMH biomedical research center. The authors acknowledge the generous donation of the Birk and Katri families to the RMH Research fund.

Footnotes

Disclosure of Potential Conflicts of Interest

A. Thomas is a consultant/advisory board member for Lilly. S. Johnston is a consultant/advisory board member for AstraZeneca, Roche/Genentech, and Novartis. S. Popat is a consultant/advisory board member for AstraZeneca. D. Cunningham reports receiving a commercial research grant from AstraZeneca and other commercial research support from Merrimack, Celgene, Bayer, Amgen, Merck Serono, and Sanofi. N.C. Turner reports receiving a commercial research grant from AstraZeneca and is a consultant/advisory board member for the same. No potential conflicts of interest were disclosed by the other authors.

Contributed by

Authors’ Contributions

Conception and design: A. Pearson, N. Tarazona, E. Kilgour, C. Rooney, G. Brown, I. Chau, S. Popat, D. Cunningham, N.C. Turner

Development of methodology: A. Pearson, E. Smyth, I.S. Babina, N. Tarazona, N.R. Smith, C. Geh, C. Rooney, G. Brown, I. Chau, N.C. Turner

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): A. Pearson, E. Smyth, I.S. Babina, M.T. Herrera-Abreu, N. Tarazona, C. Peckitt, E. Kilgour, N.R. Smith, C. Geh, C. Rooney, J. Ning, K. Fenwick, A. Swain, G. Brown, S. Chua, A. Thomas, S.R.D. Johnston, M. Ajaz, K. Sumpter, D. Watkins, I. Chau, S. Popat, D. Cunningham

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): A. Pearson, E. Smyth, I.S. Babina, N. Tarazona, C. Peckitt, E. Kilgour, N.R. Smith, C. Rooney, R. Cutts, J. Campbell, G. Brown, S. Chua, I. Chau, S. Popat, D. Cunningham, N.C. Turner

Writing, review, and/or revision of the manuscript: A. Pearson, E. Smyth, I.S. Babina, N. Tarazona, C. Peckitt, E. Kilgour, N.R. Smith, C. Rooney, G. Brown, S. Chua, A. Thomas, S.R.D. Johnston, D. Watkins, I. Chau, S. Popat, D. Cunningham, N.C. Turner

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): A. Pearson, E. Smyth, N. Tarazona, N.R. Smith, J. Ning, S. Chua, A. Gillbanks

Study supervision: E. Smyth, N. Tarazona, N.R. Smith, G. Brown, A. Thomas, D. Watkins, S. Popat, D. Cunningham, N.C. Turner

References

1. Turner N, Grose R. Fibroblast growth factor signalling: from development to cancer. Nat Rev Cancer. 2010;10:116–29. [PubMed]
2. Dieci MV, Arnedos M, Andre F, Soria JC. Fibroblast growth factor receptor inhibitors as a cancer treatment: from a biologic rationale to medical perspectives. Cancer Discov. 2013;3:264–79. [PubMed]
3. Gavine PR, Mooney L, Kilgour E, Thomas AP, Al-Kadhimi K, Beck S, et al. AZD4547: an orally bioavailable, potent, and selective inhibitor of the fibroblast growth factor receptor tyrosine kinase family. Cancer Res. 2012;72:2045–56. [PubMed]
4. Wolff AC, Hammond ME, Hicks DG, Dowsett M, McShane LM, Allison KH, et al. Recommendations for human epidermal growth factor receptor 2 testing in breast cancer: American Society of Clinical Oncology/College of American Pathologists clinical practice guideline update. J Clin Oncol. 2013;31:3997–4013. [PubMed]
5. Herrera-Abreu MT, Pearson A, Campbell J, Shnyder SD, Knowles MA, Ashworth A, et al. Parallel RNA interference screens identify EGFR activation as an escape mechanism in FGFR3-mutant cancer. Cancer Discov. 2013;3:1058–71. [PMC free article] [PubMed]
6. Razzaque MS. The FGF23-Klotho axis: endocrine regulation of phosphate homeostasis. Nat Rev Endocrinol. 2009;5:611–9. [PMC free article] [PubMed]
7. Cha JY, Maddileti S, Mitin N, Harden TK, Der CJ. Aberrant receptor internalization and enhanced FRS2-dependent signaling contribute to the transforming activity of the fibroblast growth factor receptor 2 IIIb C3 isoform. J Biol Chem. 2009;284:6227–40. [PMC free article] [PubMed]
8. Ueda T, Sasaki H, Kuwahara Y, Nezu M, Shibuya T, Sakamoto H, et al. Deletion of the carboxyl-terminal exons of K-sam/FGFR2 by short homology-mediated recombination, generating preferential expression of specific messenger RNAs. Cancer Res. 1999;59:6080–6. [PubMed]
9. Gevensleben H, Garcia-Murillas I, Graeser MK, Schiavon G, Osin P, Parton M, et al. Noninvasive detection of HER2 amplification with plasma DNA digital PCR. Clin Cancer Res. 2013;19:3276–84. [PubMed]
10. Wang J, Wen S, Symmans WF, Pusztai L, Coombes KR. The bimodality index: a criterion for discovering and ranking bimodal signatures from cancer gene expression profiling data. Cancer Inform. 2009;7:199–216. [PMC free article] [PubMed]
11. Cancer Genome Atlas Research N. Comprehensive molecular characterization of gastric adenocarcinoma. Nature. 2014;513:202–9. [PMC free article] [PubMed]
12. Ong SH, Hadari YR, Gotoh N, Guy GR, Schlessinger J, Lax I. Stimulation of phosphatidylinositol 3-kinase by fibroblast growth factor receptors is mediated by coordinated recruitment of multiple docking proteins. Proc Natl Acad Sci U S A. 2001;98:6074–9. [PubMed]
13. Castellano E, Sheridan C, Thin MZ, Nye E, Spencer-Dene B, Diefenbacher ME, et al. Requirement for interaction of PI3-kinase p110alpha with RAS in lung tumor maintenance. Cancer Cell. 2013;24:617–30. [PMC free article] [PubMed]
14. Ueki K, Algenstaedt P, Mauvais-Jarvis F, Kahn CR. Positive and negative regulation of phosphoinositide 3-kinase-dependent signaling pathways by three different gene products of the p85alpha regulatory subunit. Mol Cell Biol. 2000;20:8035–46. [PMC free article] [PubMed]
15. Slamon DJ, Leyland-Jones B, Shak S, Fuchs H, Paton V, Bajamonde A, et al. Use of chemotherapy plus a monoclonal antibody against HER2 for metastatic breast cancer that overexpresses HER2. N Engl J Med. 2001;344:783–92. [PubMed]
16. Dowsett M, Procter M, McCaskill-Stevens W, de Azambuja E, Dafni U, Rueschoff J, et al. Disease-free survival according to degree of HER2 amplification for patients treated with adjuvant chemotherapy with or without 1 year of trastuzumab: the HERA trial. J Clin Oncol. 2009;27:2962–9. [PMC free article] [PubMed]
17. Kunii K, Davis L, Gorenstein J, Hatch H, Yashiro M, Di Bacco A, et al. FGFR2-amplified gastric cancer cell lines require FGFR2 and Erbb3 signaling for growth and survival. Cancer Res. 2008;68:2340–8. [PubMed]
18. Lamont FR, Tomlinson DC, Cooper PA, Shnyder SD, Chester JD, Knowles MA. Small molecule FGF receptor inhibitors block FGFR-dependent urothelial carcinoma growth in vitro and in vivo. Br J Cancer. 2011;104:75–82. [PMC free article] [PubMed]
19. Cunningham D, Allum WH, Stenning SP, Thompson JN, Van de Velde CJ, Nicolson M, et al. Perioperative chemotherapy versus surgery alone for resectable gastroesophageal cancer. N Engl J Med. 2006;355:11–20. [PubMed]
20. Garcia-Murillas I, Lambros M, Turner NC. Determination of HER2 amplification status on tumour DNA by digital PCR. PLoS One. 2013;8:e83409. [PMC free article] [PubMed]
21. Chou TC. Drug combination studies and their synergy quantification using the Chou–Talalay method. Cancer Res. 2010;70:440–6. [PubMed]
22. Li H, Durbin R. Fast and accurate short read alignment with Burrows–Wheeler transform. Bioinformatics. 2009;25:1754–60. [PMC free article] [PubMed]
23. McKenna A, Hanna M, Banks E, Sivachenko A, Cibulskis K, Kernytsky A, et al. The Genome Analysis Toolkit: a MapReduce framework for analyzing next-generation DNA sequencing data. Genome Res. 2010;20:1297–303. [PubMed]
24. Cibulskis K, Lawrence MS, Carter SL, Sivachenko A, Jaffe D, Sougnez C, et al. Sensitive detection of somatic point mutations in impure and heterogeneous cancer samples. Nat Biotechnol. 2013;31:213–9. [PMC free article] [PubMed]