Transcriptomics and Transposon Mutagenesis Identify Multiple Mechanisms of Resistance to the FGFR Inhibitor AZD4547

FGFRs are members of the receptor tyrosine kinase (RTK) family that bind to different FGF family members and are upstream of both the MAPK–ERK and PI3K–AKT signaling pathways. FGFRs dimerize upon FGF ligand binding, which results in cross-phosphorylation of the receptors cognate kinase domains and allows the binding of the adaptor protein FGFR substrate 2α (FRS2α), a key transducer of FGFR signaling (1). Once bound, subsequent phosphorylation of FRS2α induces the recruitment of growth factor receptor-bound 2 (GRB2) and son of sevenless (SOS), resulting in activation of the MAPK–ERK signaling pathway. In contrast, activation of the PI3K–AKT signaling pathway is mediated by interactions between the FRS2α complex and GRB2-associated binding protein 1 (GAB1; ref. 1).

In human cancers, FGFR signaling is frequently hyperactivated by deregulation of FGF ligands or by activating mutations in the receptors, which predominantly consist of gene amplifications, point mutations, and gene fusions (2, 3). As such, FGFR inhibitors are considered to be an attractive therapeutic strategy for patients with mutations in FGFR family members. Currently, no FGFR-targeted therapies are approved for the treatment of human cancer, but multiple therapeutics targeting FGFR signaling are under investigation in several phase I/II clinical trials in different types of cancer (2, 3). These encompass several different approaches for inhibiting FGFR, including nonselective and selective FGFR small-molecule tyrosine kinase inhibitors, monoclonal antibodies against FGFRs and FGF ligand traps.

Although these initial trials have shown promising results concerning tolerability and antitumor activity of several FGFR inhibitors in a subset of patients (2, 4–9), more research is required to determine the right criteria for patient selection and to tackle potential resistance mechanisms to improve FGFR-targeted therapies. Several studies have already identified resistance mechanisms to FGFR-targeting agents, including polyclonal secondary FGFR mutations (including gatekeeper mutations; refs. 10–12), activation of alternative RTKs (13–16), and paracrine signaling of the tumor stroma (17, 18). However, because most of these mechanisms are identified in in vitro studies, they may not encompass the complete spectrum of resistance mechanisms to FGFR inhibitors.

In a previous study, we identified Fgfr2 as a key driver of invasive lobular breast carcinoma (ILC) using a Sleeping Beauty (SB)-based transposon insertional mutagenesis screen in mice (19). In this work, we explore how mouse ILCs (mILC) with SB transposon insertions in Fgfr2 respond to treatment with the selective FGFR inhibitor AZD4547, and by which mechanisms they acquire resistance to FGFR inhibition. Our results show that the tumors exhibit increased FGFR signaling and initially regress upon treatment with AZD4547, but eventually develop treatment resistance. By performing a multiomics analysis focusing on the resistant tumors, we identify several known and novel mechanisms by which tumors become resistant to AZD4547 treatment. Two of these mechanisms were uniquely identified from de novo transposon insertions that were acquired during treatment, demonstrating that insertional mutagenesis in mice is an effective tool for identifying resistance mechanisms to targeted cancer therapies.

Orthotopic transplantations of small tumor fragments were performed as previously described by Doornebal and colleagues (20). For the WESB-Fgfr2 tumor-derived cells, 200,000 cells were injected orthotopically into the right fat pad of 8- to 15-week-old wild-type syngeneic recipient FVB females in 20 μL Matrigel (Corning) and PBS (1:1). For the WESB-Fgfr2-EV and WESB-Fgfr2-ABCG2 cells, 200,000 cells were injected orthotopically into the right fat pad of 8-week-old NMRI-nude females (Janvier Labs) in 20 μL Matrigel (Corning) and PBS (1:1). All the drug interventions were initiated as soon as the mammary tumors reached a size of 5 × 5 mm (62.5 mm³; tumor volume: length × width² × 0.5). The treatments were performed daily by oral gavage for the indicated time with either the vehicle (1%-Tween-80 in demineralized water) or AZD4547 (AstraZeneca) at a dose of 12.5 mg/kg/day. The experimental cohort was monitored and mice were sacrificed (overall survival) when the (total) mammary tumor burden reached a size of approximately 1500 mm³ (tumor volume: length × width² × 0.5) or suffered from clinical signs of distress (respiratory distress, ascites, distended abdomen, rapid weight loss, and severe anemia) caused by mammary tumor burden or metastatic disease. One hour after the last dosing, mice were sacrificed and the tumor, lungs, liver, spleen, and tumor-draining lymph nodes were collected for further analysis. The mouse technicians were blinded to the sample groups for the treatments of WESB-Fgfr2 and WESB-Fgfr2-ABCG2 established tumors in mice. All animal experiments were approved by the Animal Ethics Committee of the Netherlands Cancer Institute and performed in accordance with institutional, national and European guidelines for Animal Care and Use.

The isolation of primary tumor cells of the SB-induced mILCs (referred to as WESB cells) was performed as previously described by Kas and colleagues (19). WESB cells were cultured in DMEM-F12 medium containing 10% FBS, 100 IU/mL penicillin, 100 μg/mL streptomycin (all from Life Technologies). MEF3.8 cells were cultured in DMEM-F12 medium containing 10% FBS, 100 IU/mL penicillin, 100 μg/mL streptomycin (all from Life Technologies). Phoenix packaging cells were cultured in Iscove's medium (Life Technologies) containing 10% FBS, 100 IU/mL penicillin, and 100 μg/mL streptomycin. WESB-Fgfr2 cells were transduced with LZRS-IRES-GFP or LZRS-Bcrp1-IRES-GFP as previously described by Allen and colleagues (21). Single GFP⁺ cells were sorted and allowed to recover before they were used in the experiments. Cell authentication was not conducted. All cell lines were kept at low passage and routinely tested for Mycoplasma contamination using the MycoAlert Mycoplasma Detection Kit (Lonza).

Additional experimental details regarding the cell viability, clonogenic, and competition assays are described in the Supplementary Data.

Vesicular transport assays were performed using the rapid filtration method as previously described (22, 23). Additional experimental details are described in the Supplementary Data.

DNA and RNA were isolated from whole tumor pieces using the Allprep DNA/RNA Mini Kit (Qiagen) according to the manufacturer's protocol.

Additional experimental details regarding the detection of the endogenous Fgfr2–Tbc1d1 fusion and the Met qPCR copy number analysis are described in the Supplementary Data.

Transposon insertions were amplified and mapped following a previously described tagmentation-based DNA sequencing protocol (24). Additional experimental details and the analysis of the insertions sites are described in the Supplementary Data.

The primary antibodies to the following proteins were used: FGFR2 (1:1,000, GeneTex 10648), phospho-FGFR (1:1,000, CST 3471), FRS2 (1:1,000, ProteinTech 11503-1-AP), phospho-FRS2 (Tyr436) (1:1,000, Abcam 193363), AKT1 (1:1000, CST 2938), phospho-AKT(Ser473; 1:1,000, CST 4060), p44/42 MAP kinase (1:1,000, CST 4695), phospho-p44/42 MAPK ERK1/ERK2 (Thr202/Tyr204; 1:1,000 CST 9101), ABCG2 (1:400, Abcam 24115) and β-actin (1:50,000, Sigma A5441).

Additional details regarding immunoblotting and IHC are described in the Supplementary Data.

A composite complex of AZD4547 bound to mutated FGFR2 was built using the crystal structure of the FGFR1 kinase domain in complex with AZD4547 (PDB code V405) as a template. A structural alignment of the FGFR2 kinase domain crystal structure (PDB code 2PVF) was performed and the positions of the resistance mutations were mapped onto this alignment. Molecular graphic images were prepared using the CHIMERA package (25).

For the mouse studies, no statistical tests were performed to determine the appropriate sample size. Survival probabilities were estimated using the Kaplan–Meier method and compared using the Mantel–Cox test. The effect of AZD4547 treatment on tumor growth of WESB-Fgfr2-EV and WESB-Fgfr2-ABCG2 established tumors was tested using mixed linear models (details in Supplementary Data). To test for differential expression of Abcg2, Rasa1, and Pcdh15 over the insertion sites across all SB-induced tumors, we used the group-wise differential expression test implemented in IM-Fusion (26). The investigators were not blinded to the sample groups for all experiments. Graphs and error bars represent mean ± SD. Python 3.5, R 3.3.1 and GraphPad Prism 7.03 were used for the statistical analyses. P values < 0.05 were considered significant.

Raw tagmentation and RNA-sequencing data are available in ENA under accession number PRJEB25507.

In a previous study, we performed a SB insertional mutagenesis screen in mice with mammary-specific inactivation of Cdh1 (encoding E-cadherin) to identify genes and pathways driving the development of ILC (19). Analysis of common insertion sites (CIS) in the SB-induced mILCs showed that a majority (56 of 99) of these tumors had SB insertions in Fgfr2, providing strong evidence that Fgfr2 is a driver of ILC.

To confirm active FGFR signaling in mILCs with SB insertions in Fgfr2, we established cell lines from two SB-induced tumors, one with an insertion upstream of Fgfr2 (WESB-Fgfr2) and one without an insertion in or near Fgfr2 (WESB). We next compared the expression of FGFR2 and downstream signaling proteins between the two tumor cell lines. Although immunoblot analysis with an anti-FGFR2 antibody revealed no expression of native FGFR2 in either of the cell lines, we observed a doublet of approximately 240 kDa that was only expressed in the WESB-Fgfr2 cell line (Fig. 1a). This size coincided with the predicted protein size of a gene fusion between Fgfr2 and Tbc1d1 (Supplementary Fig. S1A and S1B), which we previously identified in RNAseq data from this SB-induced mILC (26). Similar FGFR2 gene fusions were previously identified in several other studies, which demonstrated that these fusions result in increased FGFR signaling (27). In line with this, comparison of signaling proteins downstream of FGFR2 showed increased expression of phosphorylated FRS2α in WESB-Fgfr2 tumor-derived cells compared with WESB cells.

These results demonstrate that WESB-Fgfr2 cells show increased expression of an Fgfr2–Tbc1d1 fusion gene, which is driven by an SB insertion upstream of Fgfr2. The increased expression of this fusion gene results in activation of FGFR signaling, suggesting that FGFR inhibition could be an interesting therapeutic strategy in these tumors.

To determine if WESB-Fgfr2 cells were indeed sensitive to FGFR inhibition, we treated these cells with the selective FGFR inhibitor AZD4547, which is currently being evaluated in several early-phase clinical trials (2). After treatment with 100 nmol/L AZD4547, WESB-Fgfr2 cells showed a decrease in expression of phosphorylated FGFR, FRS2α, and ERK1/2 (Fig. 1b), confirming inhibition of the FGFR signaling pathway. Compared with WESB cells, the WESB-Fgfr2 cells also showed reduced viability upon exposure to increasing concentrations of AZD4547 (Supplementary Fig. S1C), indicating that WESB-Fgfr2 cells are dependent on FGFR signaling for their survival in vitro.

To determine the efficacy of AZD4547 in vivo, we orthotopically transplanted WESB-Fgfr2 tumor fragments into multiple wild-type syngeneic recipient FVB animals and treated these animals with vehicle or AZD4547 (12.5 mg/kg/day) daily via oral gavage (Fig. 1C). To reduce the potential toxicity of prolonged treatment and to test the effect of a “drug holiday” on tumor growth, the animals were treated using either a continuous or an intermittent dosing schedule (Supplementary Fig. S1D). In both dosing schedules, tumors treated with AZD4547 showed decreased expression of downstream FGFR signaling and increased expression of cleaved caspase-3 (Supplementary Fig. S1E–S1G), which resulted in tumor regression within 10 to 20 days after start of the treatment (Fig. 1D and E). Furthermore, the majority of the AZD4547-treated tumors (9 of 10 continuous-treated and 13 of 14 intermittent-treated tumors) did not show any regrowth within the first treatment cycle of 24 days.

In the majority of mice, continuous treatment with AZD4547 resulted in tumor control for at least 40 days after start of the treatment, resulting in an increased overall survival compared with the vehicle-treated animals (Fig. 1F). Notably, 2 of 10 sacrificed animals did not show any remaining tumor cells. In contrast, all intermittently treated mice showed tumor regrowth after the first treatment cycle of 24 days. However, these tumors remained sensitive to multiple additional cycles of AZD4547 treatment (Supplementary Fig. S2A), resulting in an increased overall survival compared with the vehicle-treated animals (Fig. 1G). Similar results were obtained with continuous AZD4547 treatment of mice after the orthotopic injection of WESB-Fgfr2 tumor-derived cells (Supplementary Fig. S2B and S2C).

Although the overall survival of AZD4547-treated mice was increased compared with vehicle-treated animals, there was no significant difference in survival between the continuous dosing (107 days) and the intermittent dosing (126 days) groups. However, in the continuous dosing group an increased number of animals succumbed due to clinical signs of distress (respiratory distress, ascites, distended abdomen, rapid weight loss, and severe anemia), suggesting that the intermittent treatment schedule is less toxic for the animals (Supplementary Fig. S2D). In spite of the potent anticancer activity of AZD4547, both treatment schedules failed to deliver long-term tumor control, most likely due to the emergence of acquired therapy resistance. This reflects the major problem observed in patients with cancer treated with targeted anticancer therapies.

To explore potential resistance mechanisms to FGFR inhibition, we performed RNA-sequencing of AZD4547-sensitive and -resistant tumors obtained from vehicle-treated and AZD4547-treated animals (Supplementary Table S1), respectively, and compared their mutational spectra to identify mutations that were acquired during AZD4547 treatment. In this approach, we initially focused on known resistance mechanisms to FGFR-targeting therapeutics (10–16), which include upregulation of alternative RTKs and secondary FGFR mutations.

To assess if upregulation of other RTKs could explain the resistance of these tumors, we used RNA-sequencing data to determine changes in gene expression for Kit, Met, and all FGFR-, EGFR-, and IGF-related RTKs. For this purpose, we used DIDS, an algorithm that is specifically designed to identify differentially expressed genes in heterogeneous populations (28). Although this analysis did not identify any RTKs that were significantly differentially expressed across multiple samples, it did identify two AZD4547-resistant tumors with increased expression of Met [also known as hepatocyte growth factor receptor (HGFR)] and one AZD4547-resistant tumor with increased expression of insulin-like growth factor 1 receptor (Igf1r), compared with vehicle-treated tumors (Fig. 2A; Supplementary Fig. S3A). Subsequent analysis of IGF1R and MET protein expression by IHC did not support IGF1R as a potential resistance mechanism, because no correlation was observed between expression levels of Igf1r mRNA and IGF1R protein (Fig. 2B). In contrast, both AZD4547-resistant tumors with high Met mRNA expression showed amplification of Met (Supplementary Fig. S3B) and overexpression of MET protein (Fig. 2C), whereas vehicle-, AZD4547-treated, and other AZD4547-resistant tumors were negative for MET expression (Supplementary Fig. S3C). These in vivo results are in line with previous in vitro studies showing that upregulation of MET attenuates the efficacy of FGFR inhibition in tumor cells (14, 15), indicating that upregulation of MET may also counteract the therapeutic efficacy of AZD4547 in vivo. In contrast to previous in vitro studies (13, 16), we did not observe an obvious increase in mRNA expression of the Egfr or other EGFR-family members in any of the AZD4547-resistant tumors.

Next, to determine if any mutations in RTKs or members of the MAPK–ERK pathway could explain the resistance of these tumors, we used the RNA-sequencing data to identify mutations in the above-mentioned RTKs and genes involved in the MAPK–ERK signaling pathway (Supplementary Table S2). Using this approach, we identified 12 missense mutations in Fgfr2 affecting 11 different amino acids, of which four were located in the third immunoglobulin (Ig)-like domain (IgIII) and seven were located in the tyrosine kinase domain (Fig. 2D).

To predict the effects of the mutations in the FGFR2 kinase domain on AZD4547 binding, we mapped the seven missense mutations onto the FGFR2 protein structure and observed that residues I567, N568, V581, E584, S587 reside in the ATP-binding pocket of FGFR2 (Fig. 2E). As a consequence, these mutated residues directly perturb the binding site of AZD4547. In addition, the E584G mutation is located in the kinase hinge and introduces flexibility at the key recognition motif for AZD4547. The K660R mutation is adjacent to the binding site, which suggests that the binding of AZD4547 is indirectly perturbed. Finally, the K678M mutation is located in the kinase activation loop, suggesting that this mutation alters the dynamics of the activation loop and favors the active conformation.

Interestingly, four out of the seven missense mutations in the tyrosine kinase domain reflected recurrent point mutations that were previously reported in patients with FGFR2 fusion–positive cholangiocarcinoma, whose tumors acquired resistance to the selective FGFR inhibitor NVP-BGJ398 (10). In this previous work, structural characterization combined with functional in vitro studies showed that these FGFR2 kinase mutations either induce a steric clash with NVP-BGJ398 in the ATP-binding pocket or destabilize the inactive conformation of the kinase. Together, these data indicate that the seven mutations in the ATP-binding pocket of FGFR2 disrupt the binding of selective FGFR tyrosine kinase inhibitors and therefore hamper their therapeutic efficacy.

Because of the presence of a constitutively active SB transposase, the SB-induced mILCs could be capable of developing resistance by acquiring de novo transposon insertions in or near resistance genes during AZD4547 treatment. To determine if SB-mediated mutagenesis might indeed be driving resistance in some of these tumors, we performed an insertion analysis of 27 AZD4547-resistant SB-induced tumors and compared the identified SB insertions to those found in the donor tumor and 15 vehicle-treated tumors. Globally, this analysis showed that the majority of the clonal insertions in the donor tumor (e.g., Fgfr2, Ppp1r12a, Slc16a9, and Trps1) were maintained after orthotopic transplantation and long-term treatment of the tumor-bearing mice (Supplementary Fig. S4A; Supplementary Tables S1 and S3). Interestingly, additional SB insertions were observed in both vehicle- and AZD4547-treated tumors, indicating that mobilization of transposons still occurs after transplantation of SB-induced tumors.

To specifically identify de novo insertions that might be driving resistance to AZD4547, we filtered for genes that contained SB insertions in the AZD4547-resistant tumors but not in the donor or vehicle-treated tumors (Supplementary Fig. S4B). This analysis revealed three candidate resistance genes (Abcg2, Rasa1, and Pcdh15) with insertions in at least three AZD4547-resistant tumors (Fig. 3A). Of these three genes, Abcg2 contained several independent insertions that were mainly in the sense orientation and located upstream of the transcription start site (Fig. 3B), indicating that these insertions result in increased Abcg2 expression. In support of this, these insertions coincided with increased mRNA and protein expression of ABCG2 (Fig. 3C; Supplementary Fig. S5A–S5C). Variable ABCG2 expression highlights intratumor heterogeneity in mechanisms of AZD4547 resistance. In contrast to Abcg2, Rasa1, and Pcdh15 contained either a mix of sense/antisense insertions or purely antisense insertions, suggesting that these genes are inactivated (Fig. 3D–G). Further analysis revealed decreased expression of exons downstream of the insertion sites in Rasa1, supporting inactivation of Rasa1 via truncation of the gene, whereas expression of Pcdh15 was not markedly affected.

To investigate whether insertions from the donor tumor might contribute to intrinsic treatment resistance, we compared the relative support scores of insertions between untreated tumors (vehicle-treated tumors and the donor tumor) and AZD4547-resistant tumors to determine if insertions in specific genes were enriched after AZD4547 treatment. This analysis identified six genes (Arid1a, Myh9, Fbxw7, Matr3, Slc16a9, and Map4k4) with increased support scores in AZD4547-resistant tumors, indicating that subclones with insertions in these genes are selected for during treatment (Supplementary Fig. S5D). These genes might therefore be involved in intrinsic resistance to AZD4547. Interestingly, the top three genes (Arid1a, Myh9, and Fbxw7) were previously identified as candidate driver genes in ILC formation (19).

Collectively, these results show that persistent mobilization of transposons in SB-induced mILCs allows them to acquire new insertions during treatment and that this approach can be used to identify novel resistance mechanisms. Our analysis implicates upregulation of Abcg2 and inactivation of Rasa1 as additional resistance mechanisms to AZD4547, which were not previously identified with our mutational analyses. This demonstrates that combining insertional mutagenesis with drug treatments poses an effective strategy for identifying resistance mechanisms to targeted therapies in mice.

To test whether inactivation of Rasa1 induces resistance to AZD4547 treatment, we transfected WESB-Fgfr2 tumor-derived cells with modified pX330 vectors containing single guide RNAs (sgRNA) targeting three different genomic regions of Rasa1 or a nontargeting sgRNA (sgNT). All the Rasa1 targeting sgRNAs induced efficient modification of the Rasa1 target sites in the transfected cell populations (Supplementary Fig. S5E–S5G), as determined by tracking of insertions or deletions [indels] by decomposition (TIDE) analysis (29). To test for drug sensitivity, we performed an in vitro competition assay with a mixture of WESB-Fgfr2-sgNT and WESB-Fgfr2-sgRasa1 cells (1:1 ratio) in the presence or absence of 2 μmol/L AZD4547 and subsequently quantified the allele distribution of the polyclonal population using the frequency of frameshift mutations in Rasa1 (Fig. 3H). After prolonged AZD4547 treatment, the polyclonal population was enriched for Rasa1 frameshift mutations for all three Rasa1-targeted regions (Fig. 3I), indicating that Rasa1-depleted cells were less sensitive to AZD4547 treatment compared with control cells. In contrast, the allele distributions were not affected when cells were cultured without AZD4547, demonstrating that the observed effect was not due to a difference in proliferation between WESB-Fgfr2-sgNT and WESB-Fgfr2-sgRasa1 cells. Altogether, these data show that inactivation of Rasa1 reduces the sensitivity of WESB-Fgfr2 cells to AZD4547 treatment.

Abcg2 is an ATP-binding cassette (ABC) efflux transporter, suggesting that overexpression of this gene may induce resistance through increased extrusion of AZD4547 from the tumor cells. To determine if this is indeed the case, we first sought to confirm that AZD4547 is a substrate for the ABCG2 transporter. To this end, we performed a vesicular transport assay (Fig. 4a), in which we measured the uptake of tritium-labeled methotrexate ([³H]-MTX) in inside-out Sf9-membrane vesicles expressing ABCG2 (Sf9-ABCG2), both in the presence and absence of increasing concentrations of AZD4547, and compared the results to the uptake of [³H]-MTX in control Sf9-membrane vesicles (Sf9-control). This showed that ATP-dependent uptake of [³H]-MTX by ABCG2 was inhibited by AZD4547 (Fig. 4b), indicating that AZD4547 is indeed a substrate of ABCG2.

To further explore whether increased expression of Abcg2 reduces the sensitivity of cells to AZD4547, we used mouse embryonic fibroblasts (MEF) derived from Abcb1a^−/−;Abcb1b^−/−;Abcc1^−/− mice (hereafter referred to as MEF3.8), which have very low background expression of endogenous ABCG2 (21). Furthermore, these MEFs lack both P-glycoprotein (P-gp, encoded by Abcb1a and Abcb1b), and the multidrug resistance-associated protein 1 (MRP1, encoded by Abcc1), allowing us to exclude confounding influences of these other drug efflux transporters. To test for drug sensitivity, MEF3.8 cells were transduced with an empty retroviral expression vector (MEF3.8-EV) or a vector containing Abcg2 (MEF3.8-ABCG2) and subsequently exposed to increasing concentrations of AZD4547 in long-term clonogenic assays (Fig. 5A). Compared with MEF3.8 cells, MEF3.8-ABCG2 cells were able to survive higher concentrations of AZD4547 (Fig. 5B and C), indicating that increased ABCG2 expression reduces the efficacy of AZD4547.

To confirm that increased expression of ABCG2 also reduces the sensitivity of treatment-naive WESB-Fgfr2 cells to AZD4547, we transduced these cells with an empty retroviral expression vector (WESB-Fgfr2-EV) or a vector containing Abcg2 (WESB-Fgfr2-ABCG2) and treated the transduced cells with AZD4547. Short-term treatment of WESB-Fgfr2-EV cells with AZD4547 resulted in decreased phosphorylation of FGFR, FRS2α, and ERK1/2, whereas the phosphorylation levels of these proteins were less affected in AZD4547-treated WESB-Fgfr2-ABCG2 cells (Fig. 5D).

To test the effect of ABCG2 overexpression on the responsiveness of established tumors to AZD4547, we injected WESB-Fgfr2-EV and WESB-Fgfr2-ABCG2 cells into the mammary glands of immunocompromised NMRI-nude mice and these animals were treated with either vehicle or AZD4547 (12.5 mg/kg/day) daily for 30 days when the tumors reached the size of 62.5 mm³. Interestingly, the NMRI-nude mice did not show tumor regression upon treatment with AZD4547, in contrast to the previously used FVB syngeneic animals (Supplementary Fig. S2B), suggesting that an intact immune system might enhance the therapeutic efficacy of AZD4547. Nonetheless, mice with WESB-Fgfr2-EV tumors did show stable disease, whereas WESB-Fgfr2-ABCG2 tumors progressed during treatment (Fig. 5E–G). These results show that increased ABCG2 expression also reduces the sensitivity of FGFR2-activated tumors to AZD4547 in vivo, confirming that upregulation of this drug efflux transporter can drive resistance to AZD4547.

In this work, we performed an SB-based insertional mutagenesis screen in a mouse model of ILC to identify genes that are involved in the development of resistance to FGFR-targeting therapies. As a starting point for this screen, we used SB-induced mILCs, in which we previously identified Fgfr2 as the most frequently mutated candidate gene (19). By orthotopically transplanting an SB-induced mILC with activated FGFR signaling into multiple recipient mice, we showed that treatment with the FGFR inhibitor AZD4547 initially results in tumor regression and provides long-term tumor control, but eventually results in acquired treatment resistance. Our mutational analysis of the AZD4547-resistant tumors identified several potential resistance mechanisms, including secondary mutations in FGFR2, inactivation of RASA1, a negative regulator of RAS signaling, and overexpression of MET and the drug-efflux transporter ABCG2. Together, these mechanisms explain acquired resistance to AZD4547 in (21 of 27) tumors (Fig. 6A). Resistance mechanisms in the remaining six tumors remain to be identified.

In line with previous studies in a mouse model of melanoma (30) and Arf^−/− mice (31), we show that transposon mutagenesis in mice cannot only be used to identify candidate cancer genes, but is also an effective strategy to identify genes involved in in vivo drug resistance. In our mutational analysis of the AZD4547-resistant tumors, we exploited the constitutive activity of the SB-mediated insertional mutagenesis system in SB-induced mILCs to identify potential resistance mechanisms in an unbiased, genome-wide fashion. This allowed us to identify two resistance mechanisms (activation of ABCG2 and inactivation of RASA1), which might not have been identified without SB mutagenesis. However, resistance mechanisms that involve specific amino acid substitutions may not be uncovered by transposon mutagenesis, but only arise from spontaneous mutations. A comprehensive characterization of the various mechanisms of resistance to targeted anticancer therapeutics may therefore require a multipronged approach, combining transposon mutagenesis with other sequencing modalities to identify spontaneous mutations and/or transcriptional changes that may be driving resistance. Given enough sequencing depth, RNA-sequencing based approaches for identifying transposon insertions may be able to provide the most comprehensive analysis from a single dataset (26, 32), by allowing detection of transposon insertions, mutations, gene-fusions, and transcriptional changes in RNA-sequencing data. However, targeted DNA-sequencing approaches (as we have used here to detect SB transposon insertions) are likely to yield more detailed detection of insertions and/or mutations with a low frequency, by effectively providing deeper sequencing at a lower cost.

The diverse spectrum of identified resistance mechanisms illustrates the major challenge that (intra-) tumor heterogeneity poses for the prevention of therapy resistance, as we observe multiple resistance mechanisms arising from and within a single (donor) tumor. All of the identified mechanisms center on reactivation of the canonical MAPK–ERK signaling pathway, suggesting that this is a dominant mechanism for overcoming vulnerability to FGFR inhibition (Fig. 6B). Reactivation of MAPK–ERK signaling has also been identified as a predominant resistance mechanism to EGFR inhibitors (33). In our analysis of AZD4547-resistant tumors, we observed recurrent alterations in several components of the MAP–ERK pathway, including secondary mutations in FGFR2, overexpression of the MET receptor and inactivation of RASA1. MET overexpression can induce resistance by driving reactivation of signaling pathways downstream of FGFR2, as has previously been shown in the context of FGFR and other RTK inhibitors (2, 14, 15, 34). Also loss of RASA1, which is a negative regulator of RAS, may cause resistance to FGFR inhibition via reactivation of the MAPK–ERK pathway (35).

Our analysis of the secondary mutations in FGFR2 showed that the majority of these mutations occurred in the tyrosine kinase domain, suggesting that they mainly provide resistance by preventing the inhibitor from binding to the ATP-binding pocket and thereby reactivating the FGFR signaling pathway. This finding agrees with previous studies with other FGFR inhibitors, which identified polyclonal secondary FGFR mutations (including gatekeeper mutations) as a main resistance mechanism to FGFR-targeting treatments (10–12). Our observations are further supported by studies with other RTK inhibitors, which also describe secondary mutations in the receptor as one of the main resistance mechanisms to tyrosine kinase inhibitors (34).

Our validation of the drug efflux transporter Abcg2 showed that increased ABCG2 expression can induce resistance by reducing the concentration of AZD4547 within tumor cells, which results in decreased inhibition of FGFR and reactivation of the FGFR signaling pathway. In patients, overexpression of the drug efflux pump MDR1 (encoded by ABCB1) has been observed in chemotherapy-resistant ovarian cancer (36). Our results suggest that drug efflux transporters such as ABCG2 cannot only drive therapy resistance in hematologic malignancies (37), but may also have similar effects on therapy efficacy in solid tumors.

Recent approaches have aimed to overcome resistance to FGFR-targeting therapies either by combining multiple existing RTK inhibitors (14, 15), or by designing irreversibly binding inhibitors such as FIIN-2, FIIN-3, and PRN1371 (38, 39), which cannot be disrupted as easily by secondary mutations in the receptor. However, our results suggest that combining FGFR and MEK/ERK inhibitors might be a more effective strategy, as this prevents reactivation of MAPK–ERK signaling. In addition, to avoid resistance resulting from drug efflux transporters, novel inhibitors should be specifically designed to be poor substrates for common transporters. Alternatively, CRISPR/Cas9 genetic screens could be used to identify synthetic lethal interactions with FGFR inhibitors in the context of FGFR inhibitor-resistant tumors to design rational and more effective combination therapies to overcome drug resistance (40).

In summary, SB insertional mutagenesis in mice is an effective tool to identify mechanisms of drug resistance. A comprehensive analysis of AZD4547-resistant mILCs, in which SB-based mutagenesis is combined with targeted DNA- and RNA-sequencing, allowed us to explain the mechanism of resistance in 78% of the resistant tumors, of which all converged to the reactivation of the canonical MAPK–ERK signaling cascade. Altogether, our findings suggest that FGFR-targeting drugs might be improved by designing FGFR inhibitors that are poor substrates of drug efflux transporters and irreversibly bind to the ATP-binding pocket of the receptor to prevent secondary mutations in the tyrosine kinase domain. In addition, combining these novel FGFR inhibitors with MEK/ERK inhibitors might be an even more effective strategy for preventing resistance to FGFR-targeted therapies.

C. Phillips and P.D. Smith have ownership interest (including patents) in AstraZeneca PLC. L.F.A. Wessels reports receiving a commercial research grant from Genmab. No potential conflicts of interest were disclosed by the other authors.

Conception and design: S.M. Kas, J.R. de Ruiter, J. Jonkers

Development of methodology: S.M. Kas, J.R. de Ruiter, K. Schipper, L. Bombardelli, A. Berns, J. Jonkers

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): S.M. Kas, K. Schipper, A.P. Drenth, R. de Korte-Grimmerink, S. Mahakena, K. van de Wetering,

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): S.M. Kas, J.R. de Ruiter, K. Schipper, C. Phillips, P.D. Smith, S. Klarenbeek, K. van de Wetering, L.F.A. Wessels, J. Jonkers

Writing, review, and/or revision of the manuscript: S.M. Kas, J.R. de Ruiter, C. Phillips, P.D. Smith, K. van de Wetering, L.F.A. Wessels, J. Jonkers

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): E. Schut, L. Bombardelli, E. Wientjens, A.P. Drenth, R. de Korte-Grimmerink, A. Berns,

Study supervision: J. Jonkers

We are grateful to Els Wagenaar, Sander Canisius, Chris Doornebal, Eline van der Burg, and Ute Boon for providing technical suggestions and/or assistance with the experiments. We thank Alfred Schinkel (The Netherlands Cancer Institute, Amsterdam) for the MEF3.8 and MEF3.8-ABCG2 cells. We thank Elaine Kilgour, Lindsay Bridgett, and Paul Smith (Oncology IMED, IMED Biotech Unit, AstraZeneca) for providing the AZD4547. We thank the NKI animal facility, RHPC computing facility, the animal pathology facility, the mouse clinic transgenic and intervention unit, the core facility molecular pathology and biobanking (CFMPB), and the genomics core facility for their expert technical support. Financial support was provided by the Netherlands Organization for Scientific Research [NWO: Cancer Genomics Netherlands (CGCNL), Cancer Systems Biology Center (CSBC), Netherlands Genomics Initiative (NGI) grant Zenith 93512009 (J. Jonkers), infrastructural grant from the National Roadmap for Large-Scale Research Facilities], the EU Seventh Framework Program (Infrafrontier-I3 project 312325 to J. Jonkers), and the European Research Council (ERC Synergy project CombatCancer to J. Jonkers). This work was carried out on the Dutch national e-infrastructure with the support of SURF Cooperative (e-infra160136 to J. Jonkers). This work is part of the Oncode Institute, which is partly financed by the Dutch Cancer Society (KWF; to J. Jonkers).

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.