Rapid Resistance of FGFR-driven Gastric Cancers to Regorafenib and Targeted FGFR Inhibitors can be Overcome by Parallel Inhibition of MEK

Gastric cancer is the third leading cause of cancer-related mortality worldwide, with almost a million new diagnoses and more than 700,000 people dying from this disease each year (1). Gastric cancer is often diagnosed in the advanced, metastatic stage, where chemotherapy has some efficacy, but overall survival (OS) is typically less than 12 months (2).

Regorafenib (Stivarga, Bayer) is an orally administered multikinase inhibitor of oncogenic (FGFR1, RET, PDGFRB, KIT, CRAF, and BRAF) and angiogenic (VEGFR1, VEGFR2, VEGFR3, and TIE2) kinases, and is approved for the treatment of several cancers, including refractory metastatic colorectal cancer (3, 4), hepatocellular carcinomas previously treated with sorafenib (5), and metastatic gastrointestinal stromal tumors previously treated with imatinib and sunitinib (6). Regorafenib also recently demonstrated clinical activity in chemo refractory advanced gastroesophageal cancer in the phase II INTEGRATE trial conducted by the Australasian Gastro-Intestinal Trials Group in collaboration with the National Health and Medical Research Council Clinical Trials Centre (7). This study randomized 152 patients to receive placebo or regorafenib, and demonstrated a significant improvement in progression-free survival (PFS) in patients treated with regorafenib (2.6 months) compared with placebo (0.9 months; P < 0.001; ref. 7).

Despite intense investigation of tumor tissue–based (8) and circulating angiogenic biomarkers (9, 10), there are currently no validated biomarkers which can reliably predict regorafenib benefit. One of the receptor tyrosine kinases (RTK) targeted by regorafenib is FGFR1, and a recent study in gastric cancer and colorectal cancer cell lines demonstrated FGFR2-amplified cell lines were highly sensitive to regorafenib (11). However, whether alterations of other FGFR family members similarly confer sensitivity to regorafenib, and whether these findings are clinically applicable has not been tested.

FGFR2 amplifications occur in 2%–11% (12, 13) of gastric cancers depending on the clinical characteristics of the cohort, with a lower prevalence in Asian populations (14–16). Comparatively, higher rates of FGFR2 amplification (9%–11.5%) have been consistently reported in metastatic cohorts (17–19), suggesting FGFR2-amplified cases represent a more clinically aggressive subtype. FGFR2 amplification is a marker of inferior OS in gastric cancer (14, 16, 17). FGFR2-amplified tumors also tend to have poorly differentiated histology (20), and are detected more frequently in gastric cancers with diffuse type compared with intestinal type histology (14–16). Mechanistically, FGFR2 amplification results in overexpression (21, 22) and ligand-independent receptor dimerization (23). Constitutive activation is further augmented by deletion of the FGFR2-amplified C-terminal exon, which interferes with receptor internalization (24). Clinical trials of FGFR2-targeted small-molecule inhibitors have shown limited efficacy (18, 19), however, more recently, a mAb targeting FGFR2b has shown promise in an early-phase trial (25).

Amplification of FGFR1 also occurs in approximately 2% of gastric cancers, and similar to FGFR2-amplified cases, is associated with inferior survival and distant metastasis (26). Comparatively, amplifications in FGFR3 and FGFR4 are rare, as are activating mutations in FGFR1, 3, and 4 (27).

In addition to amplification, overexpression of FGFR2 occurs in 40%–60% of gastric cancers, and is also associated with more aggressive clinical features, such as tumor depth of invasion (T stage), lymph node and distant metastases (28), and inferior OS (28, 29). The natural ligand for FGFR2, FGF7, may play a role in gastric cancer progression in this context (30), as paracrine secretion of FGF7 by fibroblasts has been reported to contribute to proliferation (31), migration, and invasion in FGFR2-expressing cells (32). FGFR1 and FGFR4 overexpression is also associated with more advanced tumor stage and it has been suggested that co-overexpression of two or more FGFRs confers an exceptionally worse prognosis (29). The frequency and clinical impact of FGFR3 and FGFR4 overexpression are less well studied, although one study reported overexpression of FGFR3 in 64% of gastric cancers, however, this was not associated with differences in OS (29).

The aim of this study was to identify tumor-based biomarkers of regorafenib response by screening an extensive panel of gastric cancer cell lines with this agent in vitro, and validation of these markers in patients treated with regorafenib in the INTEGRATE trial. We demonstrate that gastric cancer cell lines with either FGFR2 amplification, or overexpression of FGFRs 1, 2, 3, or 4, are highly sensitive to regorafenib. Notably, these lines were also highly sensitive to FGFR-specific inhibitors. However, correlation of FGFR2 amplification or FGFR 1–4 expression failed to identify a relationship with regorafenib response in the INTEGRATE trial. Exploration of the mechanisms underpinning this effect revealed FGFR-driven gastric cancers rapidly reactivate MAPK/ERK signaling in response to FGFR inhibition. Importantly, we demonstrate that this can be overcome by combinatorial FGFR and MEK inhibition. These findings suggest that upfront treatment with FGFR and MEK inhibitors may represent a more effective strategy for treating FGFR-driven gastric cancers, which warrants clinical investigation.

Cancer cell lines were obtained from the following sources: 5637, AGS, HS746T, Kato III, NCI-H716, NCI-N87, RT4, SNU-16, SNU-1, SNU-5, and SW780 (ATCC), FU97, IM95, MKN1, MKN45, MKN74, NUGC-3, OCUM-1, and RERF-GC-1B (HSRB-Japan Health Sciences Foundation), ECC10, GCIY, GSS, GSU, KE39, LMSU, MKN7, NCC-STC-K140, NUGC-4, and SH10TC (RIKEN Bioresources Centre), and MFM-223 and HT-115 (European Collection of Authenticated Cell Cultures). All cell lines were maintained in DMEM/F12 or RPMI supplemented with 10% FBS and 1% HEPES buffer, at 37°C with 5% CO₂. All cell lines were frozen down as large batches of master stocks within five passages of their purchase from these commercial manufacturers, and confirmed to be Mycoplasma negative at the point of freezing. Experiments were then performed using these master stocks for up to 20 passages. Mycoplasma testing was performed every 3 to 6 months as part of routine monitoring in our laboratory.

Regorafenib (3), BGJ398 (33), TAS-120 (34), erdafitinib (35), and trametinib (36) were obtained from Selleck Chemicals. All compounds were dissolved in DMSO (Sigma-Aldrich).

Growth inhibition was assessed by seeding 2,500 to 7,500 cells in 96-flat bottom well plates. Cell seeding density was optimized previously to achieve <80%–90% confluence at the conclusion of the experiment. Cell viability was determined using the MTS Assay (Promega), by incubating cells for 90 minutes at 37°C, and measuring absorbance at 490 (MTS) and 630 nm (background absorbance) on a SPECTROstar Nano Spectrophotometer (BMG LabTech).

Cell-cycle distribution and apoptosis were assessed by propidium iodide (PI) staining and FACS analysis. Drug-treated cells were harvested and stained overnight in PI buffer comprised of 0.1% sodium citrate (w/v), 0.1% Triton X-100 (volume/volume), and 25 mg/L PI in distilled water (37). ModFit Lt v2.0 (Verity Software House Inc.) and FlowJo v8.0 (FlowJo LLC) software were used to model the FACS data and determine the distribution of cells in various phases of the cell cycle and to determine the percentage of apoptotic cells, respectively.

Cells were lysed in RIPA Buffer (Sigma) supplemented with cOmplete EDTA-free Protease Inhibitor Cocktail (Roche/Sigma-Aldrich) and PhosSTOP Phosphatase Inhibitor Cocktail (Roche/Sigma-Aldrich). Total protein (15–30 μg) per sample was electrophoresed on NuPAGE 4%–12% Bis-Tris gels and transferred onto iBlot Nitrocellulose Membranes (Invitrogen/Thermo Fisher Scientific). Antibodies used in Western blot analysis were phospho-ERK Thr202/Tyr204 (catalog No. 4370), ERK1/2 (9107), phospho-FGFR Tyr635/654 (3476), FGFR2 (11835), FGFR1 (9740), and phospho-FRS2α (3864) from Cell Signaling Technology, β-tubulin (ab6046, Abcam), beta-actin (A5316, Sigma), FGFR3 (sc-123), and FGFR4 (sc-136988) from Santa Cruz Biotechnology. Secondary antibodies used were fluorescent-labeled goat anti-mouse (IRdye680CW, Li-Cor) and goat anti-rabbit (IRdye800CW, Li-Cor). Signal was detected using an Odyssey Imaging System (Li-Cor).

Genomic DNA (gDNA) from gastric cancer cell lines was extracted with the NucleoSpin Tissue Kit (Machery-Nagel). Quantitative PCR analysis of FGFR2 copy number was performed on 10 to 30 ng of gDNA using a TaqMan copy-number assay with a predesigned FGFR2 FAM-labeled MGB probe (Hs05182482_cn) and TERT VIC-labeled TAMRA probe (4403316), using the ViiA7 Real Time PCR System (Applied Biosystems).

FGFR2 ISH was performed on formalin-fixed, paraffin-embedded (FFPE) tissue sections using the Zytodot 2C SPEC FGFR2/CEN10 Probe (Zytovision) as per the manufacturer's instructions. A minimum of 20 nuclei were counted under a light microscope. The FGFR2/CEN10 ratio for a tumor was calculated by taking the average of the signals across 20 cells. Tumors demonstrating a clustering pattern in the FGFR2 signal were considered as potentially amplified cases.

RNA was extracted from cell pellets using the RNA Pure Kit (Roche) and cDNA was synthesized using the Transcriptor High Fidelity cDNA Synthesis Kit (Roche) using anchored-oligo (dT) primers. Gene expression was quantified using the 2^–ΔΔC_t method (38) and normalized to GAPDH. The primer sequences were FGFR1 F: CCCGTAGCTCCATATTGGACA and R: TTTGCCATTTTTCAACCAGCG; FGFR2 F: GGAAAGTGTGGTCCCATCTGA and R: TCCAGGTGGTACGTGTGATTG; FGFR3 F: CCCAAATGGGAGCTGTCTCG and R: CCCGGTCCTTGTCAATGCC; FGFR4 F: TCCTCGGGAGATGACGAA and R: CAGCAGCTTCTTGTCCATCC; and GAPDH F: ATGGAAATCCCATCACCATCTT and R: CGCCCCACTTGATTTTGG.

Knockdown experiments were performed in 24- and 96-well plates using Lipofectamine 2000 (Invitrogen/Thermo Fisher Scientific) and final siRNA concentrations of 10 to 20 nmol/L. Adherent cell lines, which were 70% to 90% confluent, were incubated with siRNA-lipid complexes for 24 to 72 hours. For the semi-adherent ECC10 cell line, 5 × 10⁵ cells were pelleted and resuspended in 200-μL Opti-Mem medium containing FGFR1 siRNA (25–100 nmol/L). Cells were subsequently electroporated at 380 V, 100 μF (exponential wave) on a Gene Pulser II (Bio-Rad) in 4-mm cuvettes at 4°C. Nontargeting (D-001810-10-20), FGFR3-targeting (L-003133-00), and FGFR4-targeting (L-003134-00) ON-TARGET plus SMARTpool siRNAs were obtained from Dharmacon. FGFR1-targeting siRNAs were obtained from Bioneer and comprised the following sequences: FGFR1 No. 1: sense, 5′-CGGCUGCCAAGACAGUGAA-3′ and antisense, 5′-UUCACUGUCUUGGCAGCCG-3′; and FGFR1 No. 2: sense, 5′-CUCACUGUGGAGUAUCCAU-3′ and antisense, 5′-AUGGAUACUCCACAGUGAG-3′.

FFPE tumor tissue was available for immunofluorescence analysis from 35 patients with gastric cancer who participated in the phase II INTEGRATE trial. All study participants gave informed written consent for tissue collection, although this was not a mandatory requirement for enrollment (INTEGRATE I HREC approval, RPAH HREC/13/RPAH/258 July 19, 2013). One section from each tumor was stained with hematoxylin and eosin and the region of the section containing tumor tissue was marked by a qualified anatomic pathologist (D.S. Williams). Because of the limited number of sections available for immunofluorescence analyses on each tumor, we optimized methods for multispectral immunofluorescence staining to simultaneously detect expression of FGFR1–4 on a single section. The method was optimized on gastric cancer cell line blocks with known FGFR expression status, which confirmed specific staining with no cross-reactivity between FGFR receptors. Epitope retrieval was performed by heating samples in sodium citrate antigen retrieval buffer (10 mmol/L sodium citrate, and 0.05% Tween 20, pH 6) in a microwave oven for 15 minutes at 10% power. Multiplexed staining was performed using Opal Fluorescence Chemistry (PerkinElmer). Primary antibodies were incubated at room temperature for 1 hour, washed three times for 2 minutes in TBST, and incubated with EnVision+ System-HRP (Dako) secondary antibody for 30 minutes. Sections were then incubated with the Opal fluorophore substrate for 10 minutes and washed with distilled water for 3 minutes. The process was then repeated for a second primary antibody by stripping the antibody complexes and repeating the antigen retrieval step. The primary antibodies used were anti-FGFR1 (9740) and anti-FGFR2 (11835) from Cell Signaling Technology and anti-FGFR3 (sc-13121) and anti-FGFR4 (sc-136988) from Santa Cruz Biotechnology, and counterstained with DAPI. Slides were imaged on a Vectra 3.0 imaging system, and expression was analyzed using InForm 2.1 (PerkinElmer). Expression of FGFRs was scored by an investigator blinded to the patient outcomes. Expression scoring was based on the strongest cytoplasmic or membranous staining intensity in each tumor section, scored as 0 (absent), 1+ (weak), 2+ (moderate), or 3+ (strong). For associations with outcomes, expression was categorized into low (0 or 1+) and high expression (2+ or 3+). Clinical correlation with PFS and OS using the Kaplan–Meier method was performed with SAS Software (version 9.3; SAS Institute).

Total RNA was extracted from cell pellets using the High Pure RNA Isolation Kit (Roche) as per the manufacturer's protocol, which included a DNase step. cDNA synthesis, library preparation of six indexed samples, and RNA sequencing (RNA-seq) analysis were performed at the AGRF on an Illumina HiSeq2000 to a minimum depth of >20 × 10⁶ 100-bp single-end reads. Raw reads were assessed for good quality using the FASTQC software. Alignment of transcript sequences to the human reference genome (build hg19) was performed using the TopHat software (39) with default parameters. Normalized gene expression values were calculated by counting aligning reads per kilobase per million reads mapped and counts per million. Gene set overlap analysis was performed using the Broad Institute molecular signatures database v7.2 (https://www.gsea-msigdb.org/gsea/msigdb/index.jsp).

Animal studies were performed with the approval of the Austin Health Animal Ethics Committee (A2018/05584). Kato III cells (2 × 10⁶) were subcutaneously injected into the left and right flanks of 8-week-old female NOD-SCID Gamma mice (five mice/treatment group). Once tumors were palpable, mice were treated via oral gavage with trametinib (0.3 mg/kg), BGJ398 (10 mg/kg), trametinib (0.3 mg/kg) + BGJ 398 (10 mg/kg), or vehicle control (60% phosal PG, 27.5% PEG400, 10% EtOH, and 2.5% DMSO). Mice were treated with drug for 5 consecutive days followed by no treatment for 2 days, and received a total of 12 treatments. Tumor growth was measured by caliper every second day. The experiment was ended when tumors in vehicle-treated mice reached 1 cm³ in size, at which point mice in all treatment groups were euthanized and tumors excised and weighed.

All data presented are mean ± SE, unless stated otherwise. Groups with continuous data were compared using unpaired Student t tests. Xenograft experiments were analyzed by two-way ANOVA. Analyses were performed on Prism v5.04 (GraphPad Software). PFS and OS were analyzed by the Kaplan–Meier method. The effect of drug combinations on cell growth in vitro was assessed using the highest single agent (HSA) synergy model in the Combenefit software (40).

To identify biomarkers predictive of regorafenib response, 25 gastric cancer cell lines were screened for regorafenib sensitivity by MTS assay (Fig. 1A). The screen identified five exquisitely sensitive lines with a GI₅₀ value of <1 μmol/L, the clinically relevant steady-state plasma concentration of regorafenib (Fig. 1A, red bars; ref. 6). The majority of the remaining cell lines showed modest sensitivity (GI₅₀ value, 2–6 μmol/L), while the GI₅₀ value of the three most resistant cell lines exceeded the maximum concentration tested of 10 μmol/L. Shown in Fig. 1B are the dose–response curves of the five most sensitive and resistant cell lines.

To determine whether regorafenib induced cytostatic or cytotoxic effects, drug-induced changes on cell-cycle kinetics and apoptosis were assessed in the five most sensitive cell lines pre- and 24 to 72 hours posttreatment. Regorafenib significantly decreased the percentage of cells in S-phase, and increased the percentage of cells in G₀–G₁-phase in all five cell lines (Fig. 1C). Minimal induction of apoptosis was observed in all cases, demonstrating the effect of regorafenib is primarily cytostatic.

Having identified gastric cancer cell lines that are exquisitely sensitive to regorafenib, we sought to identify molecular determinants of regorafenib response. Two of the five sensitive cell lines, Kato III (GI₅₀ value, 0.15 μmol/L) and SNU-16 (GI₅₀ value, 0.49 μmol/L), have been reported previously to harbor FGFR2 gene amplifications (41–43), which we independently confirmed by copy-number analysis and chromogenic ISH (CISH; Supplementary Fig. S1A and S1B). As expected, these lines also expressed high levels of FGFR2 mRNA and protein (Fig. 1D; Supplementary Fig. S1C). As none of the other regorafenib-sensitive cell lines harbored FGFR2 amplification or overexpression, we examined expression of FGFR1, 3, and 4 in the cell line panel. Remarkably, of the remaining regorafenib-sensitive cell lines, ECC10 cells (GI₅₀ value, 0.89 μmol/L) expressed high levels of FGFR1 mRNA and protein and FU97 cells (GI₅₀ value, 0.72 μmol/L) expressed high levels of FGFR3 and FGFR4 mRNA and protein (Fig. 1D; Supplementary Fig. S1E and S1F). These findings established a strong association between high expression of FGFR family members and response to regorafenib, with four of the five regorafenib-responsive cell lines harboring amplification or overexpression of an FGFR family member. Interestingly, one sensitive cell line, GSU, did not harbor overexpression of an FGFR family member, indicating additional determinants of regorafenib response also exist.

We next examined the effect of regorafenib on FGFR signaling in the four FGFR-overexpressing gastric cancer cell lines. In the two FGFR2-amplified cell lines, regorafenib markedly reduced levels of p-FGFR, the FGFR adaptor protein p-FRS2, and p-ERK within 4 hours, indicating attenuation of FGFR and downstream MAPK/ERK signaling (Fig. 2A). Similarly, in FGFR1-overexpressing ECC10 cells and in FGFR3/4-overexpressing FU97 cells, regorafenib also decreased p-FGFR and p-ERK levels, albeit at higher concentrations (Fig. 2B and C). In contrast, regorafenib had no effect on p-FRS2 or p-ERK in the GSU cell line, even at higher concentrations, indicating regorafenib inhibits proliferation of this line independent of FGFR and inhibition of MAPK signaling (Supplementary Fig. S2).

Given the relatively broad specificity of regorafenib, we next tested the sensitivity of the five most regorafenib-sensitive and -resistant cell lines to the more specific FGFR inhibitors, BGJ398, TAS-120, and erdafitinib. The four regorafenib-sensitive cell lines, which overexpressed an FGFR family member, also demonstrated marked growth inhibition in response to BGJ398, erdafitinib, and TAS-120. The similar outcome of this orthogonal drug screen strongly suggests these cell lines are dependent on FGFR-driven signaling for their proliferation (Fig. 2D–F). In comparison, regorafenib-sensitive GSU cells were refractory to the FGFR-specific inhibitors, confirming their sensitivity to regorafenib is FGFR independent (Supplementary Fig. S2).

Finally, we used RNA-seq analysis to compare the transcriptional response induced by regorafenib in FGFR2-amplified Kato III cells with that reported previously in FGFR2-amplified cells treated with the FGFR inhibitor, AZD4547 (44). The vast majority of transcripts (87%) altered in expression by AZD4547 were changed in the same orientation by regorafenib (r² = 0.78; P < 0.001), further confirming regorafenib inhibits FGFR signaling in these cells (Supplementary Fig. S3; Supplementary Table S1). To extend these findings, we also performed an unbiased exploration of published gene sets which overlapped with the 100 most up- and downregulated genes following 4-hour regorafenib treatment, using the Broad Institute molecular signatures database (45). The five gene sets with the strongest overlap: HRAS oncogenic signature (46), metastasis by ERBB2 isoform (47), EGF signaling_up (48), nuclear events (kinase and transcription factor activation), and NRG1 signaling Up (48), all related to altered MAPK signaling (Supplementary Table S2). Specific genes within these categories included classical targets and negative feedback regulators of MAPK signaling (FOSL1, FOS, EGR1, DUSP4, DUSP6, and SPRY4), indicating MAPK signaling is the primary pathway altered in FGFR2-amplified Kato III cells by regorafenib treatment.

Although the oncogenic role of FGFR2 overexpression in gastric cancer is established (23), the contribution of FGFR1, 3, and 4 in gastric cancer cell proliferation is less clear. To investigate this, we first determined the impact of FGFR1, FGFR3, and FGFR4 knockdown in gastric cancer cell lines overexpressing these receptors. FGFR1 knockdown in FGFR1-overexpressing ECC10 cells using two independent siRNAs inhibited proliferation by 20%–35%, which reflected the extent of FGFR1 mRNA knockdown conferred by these siRNAs (Fig. 3A and B). Similarly, knockdown of FGFR3 or FGFR4 in FGFR3/4-overexpressing FU97 cells attenuated proliferation by 27% and 50%, respectively (Fig. 3C–E), demonstrating that FGFR1- and FGFR3/4-overexpressing gastric cancer cell lines are dependent on these RTKs for proliferation.

To further test the link between FGFR overexpression and regorafenib sensitivity, we extended the study to cell lines derived from other cancer types with hyperactive FGFR signaling. The FGFR1/FGFR2-amplified breast cancer line, MFM-223, displayed markedly enhanced sensitivity to regorafenib compared with nonamplified MDA-MB-231 cells, which served as a control, with parallel inhibition of p-FGFR, p-FRS2, and p-ERK levels (Supplementary Fig. S4A and S4B). Similarly, FGFR2-amplified NCI-H716 colorectal cancer cells were significantly more sensitive to regorafenib compared with FGFR wild-type colorectal cancer cells (HCT-15 and HT-115; Supplementary Fig. S4C). Finally, to determine whether this extended to other FGFR alterations, we examined regorafenib response in the FGFR3-fusion harboring bladder cancer lines, SW780 and RT4, which displayed significantly greater sensitivity to regorafenib compared with the FGFR wild-type bladder cancer line, 5637 (Supplementary Fig. S4D).

On the basis of our in vitro findings, we hypothesized that FGFR gene amplification or overexpression may be predictive of clinical efficacy of regorafenib treatment. To test this in patients with gastric cancer treated with regorafenib, we analyzed the amplification status of FGFR2 by CISH and the expression status of FGFR1–4 by multiplexed IHC in tumor tissue from patients who participated in the phase II INTEGRATE trial (7), where patients were randomized to receive regorafenib or placebo.

Tumor tissue was available from 35 of the 147 (24%) patients from the clinical trial, of whom nine were randomized to placebo and 26 to regorafenib (Supplementary Fig. S5). We identified three patients (3/35, 8.3%) with FGFR2-amplified cancers (CN ratio >10; Fig. 4A and B), all of whom also had FGFR2 overexpression assessed by multispectral IHC staining (Fig. 4E). Multispectral IHC revealed that 12 of 35 patients overexpressed one or more FGFR receptor: FGFR1 (n = 5), FGFR2 (n = 7), FGFR3 (n = 6), and FGFR4 (n = 5). Strikingly, FGFR4 overexpression always cooccurred with overexpression of one other FGFR, and eight of 35 tumors harbored overexpression of two or more FGFRs (Fig. 4C–J).

The baseline clinical characteristics of FGFR-overexpressing and nonoverexpressing cases were well matched, aside from liver metastases, which were more prevalent in the FGFR-overexpressing group compared with the non-FGFR–overexpressing group (83.5% vs. 43.5%, respectively, Supplementary Table S3).

All 3 patients with FGFR2 gene amplification were randomized to receive regorafenib, however, none of these patients achieved an objective response. Furthermore, contrary to our expectation, in patients who received regorafenib (n = 26, 74.3%), median PFS was significantly shorter in FGFR-overexpressing compared with non-overexpressing cases [0.99 months, 95% confidence interval (CI), 0.79–2.04 compared with 2.79 months, 95% CI, 0.82–4.44; log-rank P = 0.02; Fig. 4K]. There was no difference in median PFS between FGFR-overexpressing and nonoverexpressing cases in the placebo group (1.38 vs. 1.25 months; HR, 1.36; 95% CI, 0.25–7.52; P = 0.72), and there was no evidence of an effect modification (P = 0.35), although sample sizes were small (n = 3 and n = 6, respectively; Fig. 4L).

Comparatively, in patients randomized to regorafenib, median OS was slightly longer in FGFR-overexpressing compared and non-overexpressing cases (5.75 months; 95% CI, 3.4–15.84 vs. 4.44 months; 95% CI, 2.14–9.40), although this was not statistically significant (HR, 0.78; 95% CI, 0.31–1.97; P = 0.60; Fig. 4M). Conversely, median OS was shorter in FGFR-overexpressing compared with non-overexpressing cases in the placebo group (4.17 months; 95% CI, 2.63–12.56 vs. 6.57 months; 95% CI, 0.92–9.43), however, this was not statistically significant (HR, 0.63; 95% CI, 0.12–3.17; P = 0.57), and there was no evidence of effect modification (P = 0.74), although sample sizes were small (Fig. 4N).

Given the lack of obvious clinical benefit of regorafenib in patients with gastric cancer with FGFR overexpression, we undertook further preclinical studies to investigate the basis for the disparity between the in vitro and in vivo findings. As the median PFS of FGFR-overexpressing gastric cancers treated with regorafenib was <1 month, we considered the possibility that these tumors may have the capacity to rapidly overcome drug-induced inhibition of cell signaling. To address this, we first examined changes in MAPK/ERK signaling over 72 hours in Kato III cells treated with regorafenib. Regorafenib rapidly inhibited p-ERK levels in FGFR2-amplified Kato III cells within 15 minutes, however, levels strongly rebounded after 24 to 72 hours. A similar effect was observed when cells were treated with the specific FGFR inhibitor, BGJ398, demonstrating FGFR2-amplified Kato III cells can rapidly adapt to FGFR inhibition and reactivate MAPK signaling (Fig. 5A and B).

We next investigated whether the reactivation of MAPK/ERK signaling could be delayed or overcome by combination treatment with the MEK inhibitor, trametinib. Combination treatment of Kato III cells with regorafenib and trametinib, or BGJ398 and trametinib, markedly attenuated the rebound of p-ERK observed in cells treated with regorafenib or BGJ398 alone (Fig. 5C and D). Importantly, combination treatment with regorafenib and trametinib or BGJ398 and trametinib further suppressed cell proliferation, as assessed by MTS assay (Fig. 5E and F), or the percentage of cells in S-phase as assessed by PI staining and FACS analysis (Supplementary Fig. S6A and S6B).

We next assessed the efficacy of combining regorafenib and trametinib, as well as BGJ398 and trametinib, on the growth of Kato III cells grown as xenografts in immunocompromised mice. Tumor growth was significantly reduced in mice treated with the regorafenib/trametinib or BGJ398/trametinib combinations compared with control or either agent alone (Fig. 5G and H). Both combinations were relatively well-tolerated at the concentrations tested with no observation of distress, although a modest reduction in body weight was observed in mice treated with regorafenib and trametinib (Supplementary Fig. S7).

Finally, we determined whether the other FGFR-overexpressing cell lines also had the capacity to rapidly reactivate MAPK signaling following FGFR inhibition. As observed for Kato III cells, initial suppression of p-ERK at 4 hours, followed by a strong rebound after 24 hours was also observed in FGFR2-amplified SNU-16 cells treated with regorafenib (Fig. 6A) or BGJ389 (Fig. 6G). Similar effects were observed in FGFR1-overexpressing ECC10 cells and FGFR3/4-overexpressing FU97 cells, although the rebound occurred with more delayed kinetics at 72 and 120 hours, respectively (Fig. 6B, C, H, and I). Combination treatment of FGFR-altered SNU-16, ECC10, and FU97 cells with regorafenib and trametinib (Fig. 6A–C) or BGJ398 and trametinib also attenuated the magnitude of p-ERK reactivation (Fig. 6G–I), with a corresponding increase in the magnitude of suppression of cell proliferation, as assessed by MTS assay (Fig. 6D–F and J–L).

The multikinase inhibitor, regorafenib, is approved for the treatment of refractory metastatic colorectal cancer (4), hepatocellular carcinoma (5), and gastrointestinal stromal tumors (6), and has demonstrated promising clinical activity in chemotherapy-refractory advanced gastroesophageal cancer (7), however, there are currently no biomarkers which can predict response to this agent. Here, by screening a large panel of gastric cancer cell lines, we demonstrate that cell lines harboring amplification or overexpression of FGFR family members are exquisitely sensitive to regorafenib, and demonstrated a similar sensitivity to three current-generation FGFR-specific inhibitors, BGJ398, erdafitinib, and TAS-120.

Cha and colleagues recently reported that FGFR2-amplified gastric and colorectal cancer cell lines are highly sensitive to regorafenib (11). Our study confirmed this finding, and extends it to gastric cancer cell lines harboring overexpression of FGFR1 and FGFR3/4. Importantly, through RNAi-mediated knockdown, we also directly demonstrate the dependence of FGFR1-, 3-, and 4-overexpressing gastric cancer cell lines on these RTKs for their proliferation, establishing them as robust therapeutic targets in gastric cancer. We also observed that colon, breast, and bladder cancer cell lines harboring amplifications or fusions in FGFR1 or FGFR3 are sensitive to regorafenib, broadening the subset of tumors that may potentially benefit from FGFR inhibition and demonstrating the efficacy of regorafenib in FGFR-aberrant cells lines is independent of tumor type.

Of note, we also identified an additional cell line, GSU, which was highly sensitive to regorafenib in vitro, but did not overexpress an FGFR family member, or any of the other known regorafenib targets. Although the basis of the sensitivity of this cell line to regorafenib is unknown, it represents a potentially useful model to identify additional targets of regorafenib.

Our cell line screen identified a strong link between FGFR amplification/overexpression and regorafenib response, however, we were unable to confirm these findings in the phase II INTEGRATE trial. Specifically, none of the three patients harboring FGFR2 amplifications who were treated with regorafenib experienced objective responses. However, it should be noted that objective responses were rare in the trial overall, occurring in three (3%) patients receiving regorafenib and one patient receiving placebo, likely reflecting the mostly cytostatic and antiangiogenic mechanism of action of regorafenib (7). Furthermore, among the patients treated with regorafenib, those with FGFR-overexpressing tumors had an inferior PFS compared with patients with no FGFR overexpression. These seemingly contrary finding to our cell line findings may have been confounded by the known poor prognosis of FGFR1–4-overexpressing gastric cancers (29). Although the placebo arm in the INTEGRATE trial had the potential to reveal potential prognostic differences, sample sizes in this arm were small, and there were no FGFR-amplified cases in this arm. We also note that median OS was longer in FGFR-overexpressing patients treated with regorafenib, although this was not statistically significant. The possibility of a treatment effect in FGFR2-amplified cases, therefore, cannot be completely ruled out. Finally, it is also important to point out that most patients in the INTEGRATE trial were chemo refractory. As the tumor tissue analyzed was mostly from early in their disease, prior to treatment, the possibility that chemotherapy may have altered FGFR expression cannot be eliminated.

Nevertheless, the failure to observe a clear link between FGFR amplification or overexpression and clinical benefit from regorafenib treatment, prompted us to investigate possible mechanisms of resistance. These studies revealed that MAPK/ERK signaling is rapidly reactivated in all FGFR-altered gastric cancer cells following FGFR inhibition, providing a potential explanation for the limited clinical efficacy of regorafenib, as well as FGFR-specific inhibitors in FGFR2-amplified gastric cancer observed to date (18). Notably, rapid reactivation of MAPK/ERK signaling following FGFR inhibition has been reported in FGFR3 fusion and FGFR3-mutant bladder cancer cell lines (49), and in FGFR1-amplified lung cancer cells lines (50), suggesting this may be an inherent feature of FGFR-driven cancers and, therefore, an important consideration in designing treatments for these tumors.

Given the potential for reactivation of MAPK/ERK signaling to be initiated by multiple mechanisms, including activation or upregulation of a large number of upstream RTKs, we focused on preventing MAPK/ERK pathway reactivation by inhibiting MEK. Combination treatment of FGFR-driven gastric cancers with an FGFR inhibitor and the clinically approved MEK inhibitor, trametinib, induced more sustained suppression of MAPK/ERK signaling, greater inhibition of cell proliferation in vitro, and more robust suppression of tumor growth in vivo.

In summary, we demonstrate that in addition to FGFR2 amplification, overexpression of FGFR1, FGFR3, and FGFR4 is also an important driver of gastric cancer proliferation. We also demonstrate that FGFR1–4 represent potential therapeutic targets of regorafenib and more targeted FGFR inhibitors in gastric cancer, however, these tumors have a strong capacity to rapidly reactivate MAPK/ERK signaling following FGFR inhibition. Importantly, we demonstrate that upfront combination treatment of these tumors with an FGFR inhibitor and an MEK inhibitor may represent a more effective strategy for treating FGFR-driven gastric cancers, and warrants clinical investigation.

D.K. Lau reports support from La Trobe University (Australian Postgraduate Award) and scholarship funding from the RMA/Pancare Foundation during the conduct of the study. K. Sjoquist reports grants from Bayer during the conduct of the study and personal fees from Ipsen, BMS, Servier, Amgen, Merck, and Competitive Drug Development outside the submitted work. A.J. Weickhardt reports personal fees from Merck and grants from Merck outside the submitted work. N. Pavlakis reports grants from Pfizer, Roche, and Bayer and personal fees from Pfizer, Roche, MSD, BMS, Amgen, Boehringer Ingelheim, Takeda, Merck KgA, Ipsen outside the submitted work. N.C. Tebbutt reports personal fees from Bayer during the conduct of the study and from BMS, Pierre Fabre, and AstraZeneca outside the submitted work. No potential conflicts of interest were disclosed by the other authors.

J.M. Mariadason: Conceptualization, resources, formal analysis, supervision, funding acquisition, investigation, writing–original draft, project administration. K. Sjoquist: Formal analysis. A. Boussioutas: Investigation, writing–review and editing. S.A. Hayes: Funding acquisition, investigation. M. Ernst: Methodology, writing–review and editing. A.J. Weickhardt: Conceptualization, supervision, writing–review and editing. N. Pavlakis: Conceptualization, resources, funding acquisition, writing–review and editing. N.C. Tebbutt: Conceptualization, supervision, funding acquisition, investigation, writing–review and editing. D.K. Lau: Conceptualization, formal analysis, funding acquisition, investigation, methodology, writing–original draft. I.Y. Luk: Conceptualization, data curation, formal analysis, supervision, investigation, methodology, writing–review and editing. L.J. Jenkins: Investigation, methodology, writing–review and editing. A. Martin: Formal analysis. D.S. Williams: Formal analysis, investigation. K.L. Schoffer: Investigation. F. Chionh: Investigation, writing–review and editing. M. Buchert: Funding acquisition, writing–review and editing.

This project was supported by NHMRC project grant (1126094), an NHMRC Senior Research Fellowship (1046092 to J.M. Mariadason), an Innovation grant from the Australasian Gastrointestinal trial group, and the Operational Infrastructure Support Program, Victorian Government, Australia. D.K. Lau and L.J. Jenkins were supported by La Trobe University Australian Postgraduate Awards, and D.K. Lau received a scholarship from the RMA/Pancare Foundation. F. Chionh was supported by NHMRC Medical Postgraduate Scholarship (1017737). The INTEGRATE clinical trial was conducted by the Australasian Gastro-Intestinal Trials Group in collaboration with the National Health and Medical Research Council Clinical Trials Centre, University of Sydney.

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