Resistance to Selective FGFR Inhibitors in FGFR-Driven Urothelial Cancer

Molecular alterations in the fibroblast growth factor receptor (FGFR) gene family are frequent events in urothelial cancer. FGFR3 mutations, a hallmark of low-grade, non–muscle invasive bladder cancer, are present in 12% to 22% of high-grade, muscle-invasive tumors (1–4). Upper tract urothelial carcinomas, representing less than 10% of urothelial cancers, harbor FGFR3 mutations in up to 60% of high-grade cases (1, 4, 5). FGFR3 activating mutations occur mainly in the extracellular domain of FGFR3 (S249C, R248C, and Y373C hotspots; refs. 3–5). Oncogenic FGFR3 fusions (with TACC3 as the most frequent partner gene) are found in up to 2% to 3% of bladder cancer and detected in upper tract urothelial carcinoma as well (2, 4–6). Activating alterations in other FGFR genes, although rare, are also reported in urothelial tumors (7, 8).

Oncogenic activating FGFR alterations in urothelial carcino­mas represent an opportunity for targeted therapeutic intervention with selective FGFR inhibitors. In a phase II trial leading to FDA approval of erdafitinib, 101 patients with FGFR3 mutations or FGFR2/3 rearrangements treated with erdafitinib at the starting dose of 8 mg daily, the objective response rate (ORR) and median progression-free survival (PFS) were 40% and 5.5 months, respectively (9, 10). Two additional reversible inhibitors, infigratinib and pemigatinib, and the irreversible inhibitor futibatinib led to slightly inferior outcomes in FGFR-selected populations of patients with urothelial cancer; however, the futibatinib data were obtained mainly from a population pretreated with FGFR inhibitors (11–13). Recently, in a phase II trial, the reversible inhibitor rogaratinib did not show superiority to chemotherapy in pretreated patients selected for FGFR1/3 mRNA overexpression. Among the 21 patients with FGFR3 mutations or fusions randomized to receive rogaratinib, responses were observed in 11 cases (ORR 52.4%); four out of 15 patients with FGFR3 molecular alterations treated with chemotherapy achieved disease response (ORR 26.7%; ref. 14).

In line with other oncogene-driven diseases, both primary and acquired resistance limit the efficacy of FGFR inhibitors in urothelial carcinomas. Evidence regarding mechanisms of resistance to FGFR inhibitors is limited to the analysis of 22 circulating tumor DNA (ctDNA) samples at progression on infigratinib. In four patients, acquired mutations in the FGFR3 tyrosine kinase domain were detected (FGFR3 V555L, V555M, and L608V, according to the NM_000142.4 transcript; ref. 11).

Here we present a molecular and functional study of resistance to selective FGFR inhibitors in patients with FGFR-driven urothelial cancer. FGFR tyrosine kinase domain mutations and off-target molecular activations were frequently observed. According to the resistance mechanisms recorded, we evaluated adaptive (i.e., alternative FGFR inhibitors) and combinatorial treatment strategies in preclinical models, providing potential future guidance for clinical application.

We collected clinical and molecular data for a total of 56 patients with advanced/metastatic urothelial cancer (n = 39 bladder, n = 15 upper urothelial tract, n = 2 with both localizations). Thirty-eight patients had both whole-exome sequencing (WES) and RNA sequencing (RNA-seq) performed on tissue biopsy, nine had WES only, and one had RNA-seq only; eight patients had a targeted next-generation sequencing (NGS) panel on ctDNA and/or tissue biopsy.

The assessment of the mutational features has been the main objective of seminal studies shedding light on the frequency of FGFR3 alterations and other molecular features in urothelial carcinomas, focusing on localized diseases (2, 4, 5). We therefore performed a global mutational landscape analysis using tumor samples from the 47 urothelial cancer patients with advanced/metastatic disease and WES available (Fig. 1). The most frequently altered genes were TP53 (51%), FGFR3 (34%), and KMT2D (34%). Taking into account the limitation of evaluating expression profiles from bulk RNA-seq analyses of tissue biopsies with variable tumor/stroma cellularity, we investigated differential expression of genes involved in the main signaling pathways in the 39 samples with RNA-seq available. No clear pattern was highlighted between FGFR3-driven and non–FGFR3-driven patient samples with the exception of FGFR3 mRNA (Supplementary Fig. S1A). Analyzing FGFR family member expression levels specifically, we did indeed reveal a strong significant overexpression of FGFR3 mRNA in FGFR3-driven samples compared with FGFR3 wild-type (WT) ones (Supplementary Fig. S1B). By contrast, only a slight decrease was observed for FGFR1 and FGFR4 mRNA and no difference was observed for FGFR2 expression.

We then focused on the 21 patients treated with selective FGFR inhibitors; 12 and seven had bladder and upper tract tumors, respectively, and in two additional cases both localizations were present. Eleven cases harbored an FGFR3 S249C mutation, five had an FGFR3::TACC3 fusion, and three displayed an FGFR3 Y373C mutation. One patient (MR904) harbored an FGFR2::FAM83H-AS1 rearrangement and one (MR103) an FGFR4 D276N mutation (Fig. 2; Supplementary Table S1).

Postprogression samples were collected on erdafitinib (n = 14), futibatinib (n = 4), or pemigatinib (n = 3). As best objective responses, we observed one patient with a complete response (CR), 12 patients with partial responses (PR), six patients with stable disease (SD), and two patients with progressive disease (PD). Of note, all patients were previously treated for advanced disease, and the majority received the FGFR inhibitor as the second treatment line (range, second–sixth), but no clear correlation between the type of cytotoxic agents received by patients prior to FGFR inhibitors and molecular alterations could be evidenced (Supplementary Table S2).

Twelve patients had both postprogression tissue biopsy and ctDNA, six had ctDNA only, and three had tissue only. Four patients received futibatinib after having progressed to either erdafitinib or pemigatinib. Longitudinal analysis was possible for 5 patients thanks to additional ctDNA samples, including one patient with tissue molecular profiles obtained both at erdafitinib and futibatinib progression. Overall, we analyzed 39 postprogression samples: 16 postprogression tissue analysis and 23 ctDNA samples.

In the cohort of 19 FGFR3-driven urothelial cancers that progressed on erdafitinib, pemigatinib, or futibatinib, FGFR3 kinase domain mutations were detected in seven patients (37%; Fig. 3A). Sequencing data of pretreatment tissue biopsy and/or ctDNA were available for five out of seven patients (71%), and no preexisting FGFR3 kinase domain mutation was detected in those baseline samples. Of note, the activating FGFR3 alterations were detected in all ctDNA and tissue samples available at progression. In six patients, we detected single mutations in the FGFR3 kinase domain in either ctDNA or tissue biopsy: N540K, V553L, V553M, V555L, V555M, and E587Q (Fig. 3A). Three concomitant FGFR3 kinase domain mutations (N540K, V555L, and L608V), in addition to the activating Y373C, were identified in the ctDNA of patient CTC1845 (Fig. 3A and B), who progressed on futibatinib after having received erdafitinib (PR, 67%; PFS, 8.6 months). The allelic status (cis/trans configuration) of these four FGFR3 mutations could not be assessed due to ctDNA amplicon size limitation.

Patient MR410 had an FGFR3 E587Q mutation detected in ctDNA only at progression on erdafitinib, a TSC1 S561fs alteration detected both in tissue and in liquid biopsies, whereas FGFR3 amplification and PIK3CA E726K mutation were found in the tissue biopsy only (Fig. 3B).

In the erdafitinib “baseline” ctDNA sample of patient M322, we observed an FGFR3::TACC3 fusion, later detected at higher variant allele frequency (VAF) in the postprogression liquid biopsy, together with an FGFR3 V555M mutation and a TSC2 frameshift event (P28Lfs; Fig. 3C).

In line with limited ctDNA shedding in cases of intrathoracic-only disease, the FGFR3 V553L mutation was detected only in the tissue biopsy and not in the ctDNA of patient MR406, who experienced mild, thoracic-only progression on pemigatinib (Fig. 3D). Given the availability of mRNA from the tissue biopsy, we performed TOPO-TA cloning and observed FGFR3 S249C and V553L mutations on the same allele, confirming that the mutational events arose in cis with the driver mutation (Supplementary Fig. S2).

The VAF increase of both FGFR3 S249C and V555L reflected the progression during erdafitinib treatment in patient MR560 (Fig. 3E).

For patient MR86, an FGFR3 N540K mutation was detected in tissue biopsy in addition to the driver mutation FGFR3 S249C and a PIK3CA E545K mutation already existing in the pre-erdafitinib sample (Fig. 3F).

Additionally, upon detection of an FGFR2 rearrangement in one patient with urothelial cancer (MR904), third-line pemigatinib resulted in a CR (Fig. 3G). Upon progression (after 3 years and 4 months), ctDNA analysis revealed the fusion partner (FGFR2::FAM83H-AS1), as well as an FGFR2 L551F mutation (according to NM _022970.3 transcript).

In 11 of 19 patients (58%) with FGFR3-driven urothelial carcinoma, we detected an alteration in the PI3K–mTOR pathway in postprogression and/or in baseline samples (Table 1).

In five samples, a PIK3CA activating mutation was detected at progression, but in three cases the PIK3CA alteration was present before the treatment with FGFR inhibitors (patients MR15, MR86, and ST2267). Patients MR15 and MR86 achieved clinical responses on erdafitinib before progression. Patient MR86 acquired an FGFR3 N540K mutation, whereas an EGFR hyperphosphorylation [determined by a phospho–receptor tyrosine kinase (RTK) array] potentially explained the resistance for patient MR15 (see the section “Nonmutational Resistance Mechanism Identified Using Patient-Derived Models”). By contrast, patient ST2267, who had a baseline PIK3CA H1047R tumor mutation, experienced primary resistance to futibatinib. For patient MR336, the PIK3CA E545K mutation was acquired during treatment with erdafitinib, which potentially conferred the primary resistance to futibatinib (Supplementary Fig. S3A).

In addition to PIK3CA mutations, TSC1/2 inactivating mutations were also frequently detected (5/19) in patients progressing on FGFR inhibitors (Table 1). Patient MR410 had a TSC1 S561fs mutation detected in both postprogression (erdafitinib) tissue and ctDNA (VAF 69.96%), concurrent with PIK3CA E726K and FGFR3 E589Q (VAF 31.12%) mutations.

A TSC2 P28Lfs mutation in patient M322, a TSC1 Q865* mutation in patient M521, and a TSC1 Q830* mutation in patient ST701 emerged at progression on erdafitinib, and a TSC1 A186fs mutation in patient MR1105 emerged at futibatinib progression (Supplementary Fig. S3B).

In patient MR246, who had an FGFR3 S249C-driven bladder cancer that ultimately progressed on erdafitinib, an NF2 frame­shift event was detected by ctDNA together with an FGFR2 V517M mutation.

Finally, a PTEN C136fs mutation was detected in a patient with primary resistance to erdafitinib (M2057).

The frequency of these mutations arising in tumor suppressor genes suggests they should be considered as potential targets for overcoming treatment resistance.

In order to evaluate the functional impact of FGFR3 kinase domain mutations on the activity of FGFR inhibitors, we generated Ba/F3 cells with an FGFR3::TACC3 fusion harboring either N540K, V553L, V553M, V555L, V555M, or L608V mutations. The expression levels of the FGFR3::TACC3 exogenous fusion protein were homogeneous across cell lines (Supplementary Fig. S4). Ba/F3 cells were exposed to increasing doses of the FGFR inhibitors erdafitinib, infigratinib, pemigatinib, rogaratinib, futibatinib, derazantinib, AZD4547, and zoligratinib to determine their corresponding IC₅₀ values (Fig. 4A; Supplementary Table S3). LY2874455 had an IC₅₀ of 87.5 nmol/L against parental Ba/F3 cells (without FGFR3::TACC3 fusion), suggesting the drug exerted off-target effects on proliferation. In light of this observation, and considering that LY2874455 is no longer in clinical development, we did not include it in the pharmacologic experiments.

The eight selective FGFR inhibitors had IC₅₀ values in the subnanomolar/nanomolar range against FGFR3::TACC3 WT-driven Ba/F3 cell lines. Each FGFR3 kinase domain mutation conferred a variable increase in the IC₅₀ values to FGFR inhibitors. The IC₅₀ against the frequent N540K mutant was >100 nmol/L for all the agents with the exception of erdafitinib and futibatinib (5 times lower). Similarly, erdafitinib and futibatinib retained low activity against the gatekeeper V555L/V555M, V553M, and L608V mutations. The V553L mutant appeared to confer a moderate degree of resistance to FGFR inhibitors, as IC₅₀ values were close to 10 nmol/L for five inhibitors but at subnanomolar/nanomolar concentrations for erdafitinib, infigratinib, and futibatinib. Of note, this mutation emerged in a patient progressing on pemigatinib (MR406; Fig. 3A and D).

To corroborate our observations, we performed immunoblot analyses with increasing concentrations of erdafitinib in Ba/F3 cells harboring a WT or a mutated FGFR3::TACC3 construct, in the presence of stimulation with heparin and acidic FGF (aFGF). Analysis of intracellular signaling across models confirmed the viability assays. FGFR3, AKT, and ERK phosphorylation was abrogated after treatment with low concentrations of erdafitinib in the WT model, whereas higher doses were required in the N540K, V553M, V555L, and L608V mutants (Fig. 4B). FGFR3 phosphorylation and intracellular signaling were maintained in the presence of high doses of FGFR inhibitors with the V555M mutation. We confirmed that compared with pemigatinib lower concentrations of erdafitinib were sufficient to abrogate intracellular signaling in the V553L mutant (Fig. 4C). This sustains the hypothesis of an efficient sequential treatment with erdafitinib in case of the emergence of a V553L mutation, which would also be the case with infigratinib and futibatinib according to the viability assays (Fig. 4A).

When Ba/F3 cells having FGFR3::TACC3 WT, FGFR3::TACC3 N540K, and FGFR3::TACC3 V555L constructs were exposed to 30 nmol/L of erdafitinib, the inhibitor was able to abolish the signaling only in the WT model (Fig. 4D). Of note, in the untreated condition, we noticed increased signaling in both mutants, which was especially marked in the N540K mutant in terms of basal ERK phosphorylation, in line with the “molecular brake” nature of the N540K mutation (15, 16).

We also aimed to validate the role of the FGFR2 L551F mutation in conferring resistance to pemigatinib in patient MR904, and to seek alternative agents retaining activity against the mutant. We generated Ba/F3 cells harboring FGFR2::BICC1 with the L551F mutation, and exposed them to the eight FGFR inhibitors (Fig. 4E and F). In addition to confirming pemigatinib resistance induced by the L551F mutation, our data revealed that erdafitinib and futibatinib could overcome this mutation.

To elucidate the potential role of PIK3CA E545K in resistance to FGFR inhibitors, we established patient-derived models as previously described (17). In the corresponding MR86 patient-derived xenograft (PDX) model, we evaluated the tumor growth using erdafitinib and pictilisib (PI3K inhibitor) as single agents or in combination. As depicted in Fig. 5A, only the combination treatment was efficient in inhibiting the tumor growth, suggesting that, whereas responses in patients with a PIK3CA preexisting mutation are possible, the PIK3CA E545K mutation contributes to the oncogenic signaling and its targeting seems relevant to maximize tumor response.

Patient MR15 suffered from an upper urothelial tumor driven by FGFR3 S249C. At progression on erdafitinib, the tissue biopsy did not reveal any additional genomic alterations compared with the baseline sample. A PDX and patient-derived cell line were established. After confirming that the eight FGFR inhibitors could not restore sensitivity in the patient-derived cell line, we performed a phospho-RTK array and detected hyperphosphorylation of EGFR (Fig. 5B). Although erdafitinib and the EGFR inhibitor gefitinib did not have a significant effect on cell growth inhibition as single agents, the combinatorial treatment showed a synergistic effect in the MR15 cell line (Fig. 5C; Supplementary Fig. S5). Immunoblot analyses confirmed that the double inhibition was required to inhibit intracellular signaling (Fig. 5D). The in vivo pharmacologic study confirmed the synergistic effect of combining erdafitinib and gefitinib (not active as single agents) in inhibiting tumor growth in the MR15 PDX (Fig. 5E).

The positive clinical data of FGFR inhibitors in FGFR-driven bladder cancer provide the first proof of concept for precision medicine in urothelial cancer. Nevertheless, the ORR of 40% and the median PFS of 5.5 months observed with erdafitinib indicate that there is space for improvement in the field of FGFR-targeted therapies for urothelial malignancies (10). Understanding the mechanisms of resistance to FGFR inhibitors represents a key step in developing new treatment strategies to improve the overall response and the durability of clinical benefit for patients in the setting of molecular-guided therapies.

In this study, we identified both on-target and off-target mechanisms of resistance to FGFR inhibitors in FGFR-driven urothelial carcinoma. Our data sustain the complementarity of the information derived from both tissue biopsies and ctDNA. If ctDNA is able to catch the heterogeneity of molecular modifications arising across tumor lesions, tissue biopsy is informative in the case of low ctDNA shedding, especially when disease progression is limited to the thorax, with a mild tumor burden (18, 19). In addition, tissue biopsy allows the establishment of patient-derived models, which are essential tools to further understand resistance and to test new treatments and combination strategies (20). We were not able to identify the resistance mechanism in five out of 21 patients (24%) included in our cohort. Although this proportion of “unknown resistance mechanisms” can be considered low, likely thanks to the availability of multiple sources of molecular analysis at progression on FGFR inhibitors (blood, tissue, and patient-derived models), it remains challenging to systematically characterize the cause of resistance when no explanatory mutational event is detected or when tumor cell–extrinsic factors within the tumor microenvironment contribute to the resistance.

Across our cohort of 19 FGFR3-driven urothelial cancers, we identified FGFR3 kinase domain mutations in seven patients at progression on FGFR inhibitors (37%). Of note, the characteristics of the FGFR3 kinase domain at resistance are distinctly different from what has been observed in FGFR2-driven cholangiocarcinoma. In patients suffering from cholangiocarcinoma, the occurrence of polyclonal (i.e., multiple) FGFR2 kinase domain mutations is frequently reported at progression on reversible FGFR inhibitors (21–23). In our study, multiple FGFR3 kinase domain mutations were detected only in patient CTC1845 (N540K, V555L, and L608V). The sole availability of ctDNA (lack of tissue biopsy) at the moment of progression on futibatinib in this patient previously treated with erdafitinib precludes the assessment of the potential serial acquisition of the mutations, and their allelic status could not be determined due to ctDNA amplicon size limitation. Pal and colleagues have previously reported the detection of V555L/M and L608V in four patients at progression on erdafitinib, but only one case had a polyclonal presentation (V555M and L608V; ref. 11). In addition to these two mutations, we reported for the first time additional mutations occurring in the FGFR3 kinase domain mutation at progression on selective inhibitors: N540K, V553L, V553M, and E587Q. With the exception of the latter, all the other mutational events have been reported in their corresponding positions of FGFR2 (N550K, V563L, V565I/L/F, L618V, according to the NM_001144913.1 transcript), suggesting that these kinase domain conserved residues are resistance hotspots in the FGFR3 kinase domain as well (21–23). Whereas it is difficult to derive the incidence of precise mutations from our study with a limited sample size, the gatekeeper residue FGFR3 V555 and the molecular brake N540 could represent relevant targets to consider for the development of novel inhibitors due to their higher frequency and resistance pattern.

Although confirming the role of FGFR3 kinase domain mutations in conferring resistance to the inhibitors, our functional studies were not informative in recognizing agents able to overcome on-target resistance. The only exception was represented by the potential role of erdafitinib, infigratinib, and futibatinib against the V553L mutation acquired in the post-pemigatinib sample. Of note, among reversible FGFR inhibitors, erdafitinib yielded the lowest IC₅₀ values against both WT and mutated FGFR3:TACC3 Ba/F3 cell lines, potentially explaining its better outcomes observed in clinical trials.

The immunoblots and cell viability analyses highlighted the differential contribution of the molecular brake (N540K) and gatekeeper (V555L) mutation in conferring resistance. V555L, thought to interfere with drug binding (24), boosts intracellular signaling as well, mainly represented by ERK phosphorylation. This latter was even more marked in the presence of N540K, in line with the “molecular brake” function of the N540 residue (15, 16). From this point of view, novel inhibitors will need to harbor higher potency, and their design should allow them to circumvent the steric clash induced by gatekeeper (and other) mutations.

The functional evidence generated regarding the FGFR2 L551F mutation also suggested it could be therapeutically targeted in FGFR2-driven tumors. In our patient MR904, the mutation emerged at progression on pemigatinib, and Varghese and colleagues reported a similar finding at progression on infigratinib in a patient affected by an FGFR2-driven cholangiocarcinoma (one additional patient developed L551F in the context of polyclonal FGFR kinase domain mutations; ref. 21). The FGFR2 L551F mutation was not previously characterized, and our data suggest that L551F, while engendering resistance to pemigatinib and infigratinib, could be overcome by futibatinib and erdafitinib.

In addition to on-target alterations responsible for resistance, we identified recurrent events affecting the PI3K–mTOR pathway at resistance to FGFR inhibitors in urothelial tumors, concomitantly or not with FGFR3 kinase domain mutations. In our population and in molecular landscape studies of urothelial tumors, PIK3CA and TSC1/2 alterations are not mutually exclusive with FGFR3 in baseline samples (1–5, 25). Nevertheless, their enrichment in postprogression samples strongly implicates their involvement in resistance. As supported by models derived from patient MR86, a baseline activating mutation in PIK3CA may not be fully capable of conferring resistance but may represent a suitable target to improve responses. The observed alterations in the PI3K–mTOR pathway in clinical samples from this study have been shown to be responsible for progression on targeted therapies in other oncogene-driven diseases and suitable targets for specific inhibition (26–31).

The establishment of the patient-derived cell line and PDX of patient MR15 illustrates the relevance of such models in providing critical elements in addressing off-target, nongenomic mechanisms of resistance and enables the development of new strategies for treatment paradigms. Of note, EGFR bypass activation has been reported as an early event in the setting of FGFR inhibition in FGFR3-altered urothelial cancer and in FGFR2-driven cholangiocarcinoma, suggesting the potential of EGFR targeting in progressive FGFR-driven diseases (32, 33).

This study has several limitations. Although being the largest study on the subject thus far, the limited sample population precludes the generalization of our findings. Despite the availability for the majority of the patients, pretreatment sequencing analyses could not be performed systematically. The lack of a matched collection of postprogression tissue samples and ctDNA in some patients limits the depth of the molecular insights we could obtain. Despite our systematic attempt in establishing patient-derived models, the relatively low success rate limited the identification of nongenetic-driven resistance mechanisms. Finally, we did not validate functionally the role of TSC1/2 mutations or their allelic status.

Altogether, our study represents the first systematic approach toward understanding resistance to FGFR inhibitors in FGFR-driven urothelial cancer. The evidence generated on both on-target and off-target molecular events responsible for resistance will serve as the backbone to address the development of new therapeutic strategies to improve patient outcomes.

To be included in this study, patients had to satisfy the following criteria: (i) diagnosis of an advanced urothelial cancer; (ii) treatment with a selective FGFR inhibitor due to the detection of a molecular alteration in one of the FGFR family members; and (iii) availability of postprogression tissue and/or ctDNA available for molecular analyses.

Patients started FGFR inhibitor treatment between 2013 and 2021. Patients were treated in the setting of clinical trials or compassionate use programs allowing treatment with FGFR inhibitors on the basis of molecular selection. Disease response was measured according to RECIST, and PFS was calculated from the date of the targeted inhibitor start to the day of radiologic evidence of progression.

The molecular analyses were performed within institutional studies ongoing at Gustave Roussy whose aim is the molecular characterization of tumors: MATCH-R (NCT02517892; ref. 17), MOSCATO (NCT01566019; ref. 34), STING-UNLOCK (NCT04932525), and CTC (NCT02666612).

All patients participating in the mentioned studies were fully informed and signed written informed consent. The studies have been approved by the ethics committee at Institut Gustave Roussy and the French National Agency for Medicines and Health Products Safety (ANSM), and are being conducted in accordance with the Declaration of Helsinki.

The objectives of the MOSCATO and MATCH-R studies are to address patients to targeted treatments according to molecular alterations and to assess molecular mechanisms of resistance, respectively. These studies initially focused on molecular analysis performed on tissue biopsy, as ctDNA technologies were not commonly used and blood/plasma samples were not collected (see MOSCATO patient M521, and MATCH-R patients MR15 and MR86). On the other hand, CTC and STING-UNLOCK were developed to study molecular alterations from blood samples (exclusively for CTC, including tissue biopsy for STING-UNLOCK). The availability of postprogression samples relied on the precise study in which individual patients were enrolled.

Postprogression tumor, when possible, underwent WES with or without concomitant RNA-seq. The main limitation of WES/RNA-seq performance was the proportion of tumor cells ≥30% in the tissue sample. In cases in which the proportion of tumor cells was between 10% and 30%, molecular analyses with targeted NGS panels (Mosc-3, Oncomine v3) were performed. For WES, the mean coverage was 140×.

ctDNA analyses were performed with Illumina or Foundation Medicine liquid biopsy panels. Importantly, all ctDNA sequencing had complete coverage of the kinase domains of FGFR genes. Briefly, extracted circulating free DNA from patient samples was processed through a library preparation that includes the addition of unique molecular identifiers, followed by a hybrid capture–based workflow and high-depth sequencing. For each patient with longitudinal ctDNA assessment, only analyses performed with the same platform were reported.

Lentiviral vectors expressing FGFR3::TACC3 fusions were created using the pLenti6/V5 directional TOPO Cloning Kit (#K495510, Thermo Fisher Scientific) according to the manufacturer's instructions. Point mutations in the FGFR3 kinase domain of the FGFR3::TACC3 fusion were introduced using the QuickChange XL Site-Directed Mutagenesis Kit (#200516, Agilent) according to the manufacturer's protocol and using the following primers:

• FGFR3 N540K forward (F) CAAAAACATCATCAAGCTGCTGGGCGCCTGC

• FGFR3 N540K reverse (R) GCAGGCGCCCAGCAGCTTGATGATGTTTTTG

• FGFR3 V553L F CGCAGGGCGGGCCCCTGTACTTGCTGGTGGAGTACGCGGCC

• FGFR3 V553L R GGCCGCGTACTCCACCAGCAAGTACAGGGGCCCGCCCTGCG

• FGFR3 V553M F CGCAGGGCGGGCCCCTGTACATGCTGGTGGAGTACGCGGCC

• FGFR3 V553M R GGCCGCGTACTCCACCAGCATGTACAGGGGCCCGCCCTGCG

• FGFR3 V555L F GCCCCTGTACGTGCTGCTGGAGTACGCGGCCAA

• FGFR3 V555L R TTGGCCGCGTACTCCAGCAGCACGTACAGGGGC

• FGFR3 V555M F GCCCCTGTACGTGCTGATGGAGTACGCGGCCAA

• FGFR3 V555M R TTGGCCGCGTACTCCATCAGCACGTACAGGGGC

• FGFR3 L608V F AGGTGGCCCGGGGCATGGAGTACGTGGCCTCCCAGAAGTGCATCCAC

• FGFR3 L608V R GTGGATGCACTTCTGGGAGGCCACGTACTCCATGCCCCGGGCCACCT

The same procedure was used to create lentiviral vectors expressing FGFR2::BICC1, and the primers used for creating L551F mutations were as follows:

• FGFR2 L551F F GGGAAACACAAGAATATCATAAATTTTCTTGGAGCCTGCACACAGGATG

• FGFR2 L551F R CATCCTGTGTGCAGGCTCCAAGAAAATTTATGATATTCTTGTGTTTCCC

To assess the allelic status of FGFR3 S249C and V553L in patient MR406, mRNA obtained from the tissue biopsy underwent retro-transcription, and a PCR spanning the two mutated residues was performed on cDNA. Amplicons were subcloned into the pCR2.1-TOPO vector (Invitrogen) according to the manufacturer's protocol. Individual cDNA was sequenced by Sanger sequencing to determine the cis/trans status of mutations.

All animal procedures and studies have been approved by the French Ministry of “Education nationale, de l'Enseignement supérieur et de la Recherche” (APAFIS#2790-2015112015055793 and APAFIS#2328-2015101914074846). Fresh tumor fragments were implanted in the subrenal capsule of 6-week-old female NOD/SCID gamma (NSG) mice obtained from Charles River Laboratories. PDX-bearing NSG mice were treated with gefitinib [100 mg/kg once daily (qd) in 0.5% methylcellose, 0.5% Tween 80] or erdafitinib [15 mg/kg qd in 10% HP-beta-CD (hydroxypropyl beta cyclodextrin)] or pictilisib (150 mg/kg qd in 0.5% methylcellose, 0.5% Tween 80) alone or in combination by oral gavage. Eight mice per group were treated for up to 40 days, and tumor volume and mouse weight were measured twice per week.

Parental Ba/F3 cells were purchased from DSMZ (ref. ACC 300) and cultured in DMEM 10% FBS in the presence of IL3 (0.5 ng/mL). Ba/F3 cells were infected with lentiviral constructs, as reported previously (35), to express the FGFR3::TACC3 or the FGFR2::BICC1 fusions with a WT kinase domain or with the mentioned kinase domain mutations. Infected Ba/F3 cells harboring FGFR fusions were selected in the presence of blasticidin (14 mg/mL) and IL3 (0.5 ng/mL) until recovery, and a second selection by culturing the cells in the absence of IL3 was then performed. FGFR2/3 fusions and corresponding kinase domain mutations were confirmed on the established cell lines by Sanger sequencing.

After several attempts, we were not able to establish IL3-independent Ba/F3 cells driven by FGFR3 S249C or FGFR3 Y373C mutations. Previous experience showed that FGFR2 cannot drive IL3-independent proliferation in the same way as FGFR1 (16), and we envisage that FGFR3 mutations occurring in the extracellular domain, oncogenic drivers in human tumors, are not sufficient to fulfill this function in Ba/F3 models in isolation.

Patient-derived cell lines (MR15) were developed from PDX samples by enzymatic digestion with a tumor dissociation kit (ref. 130-095-929, Miltenyi Biotec) and mechanical degradation with the gentleMACS dissociator. Cells were cultured with DMEM/F-12 GlutaMAX 10% FBS and 10% enriched with hydrocortisone 0.4 mg/mL, cholera toxin 8.4 ng/mL, adenine 24 mg/mL, and ROCK inhibitor 5 mmol/L (Y-27632, S1049 Selleckchem) until a stable proliferation of tumor cells was observed, as described previously (36). Culture media were then transitioned to DMEM and cultured in the presence of erdafitinib 300 nmol/L to 1 mmol/L.

As for cell verification, we confirmed that engineered Ba/F3 as well as MR15 patient-derived cell line harbored the corresponding FGFR fusion and mutations by RT-PCR and Sanger sequencing. The cells were not tested for Mycoplasma contamination, but cells were not maintained in culture for more than 2 months after establishment or thawing.

Erdafitinib, infigratinib, pemigatinib, futibatinib, derazantinib, AZD4547, zoligratinib, and LY2874455 were purchased from Selleck Chemicals. Rogaratinib was purchased from MedChemExpress.

Cell viability assays were performed in 96-well plates using the CellTiter-Glo Luminescent Cell Viability Assay (G7570, Promega). For Ba/F3, we seeded 4,000 cells/well and we treated cells for 48 hours. For the MR15 patient-derived cell line, we seeded 3,000 cells/well and treated cells for 5 days.

Ba/F3 cells with FGFR3::TACC3 fusion with a WT or a mutated kinase domain were treated for 24 hours with the corresponding doses of FGFR inhibitors in the presence of heparin 25 μg/mL (Sigma-Aldrich, ref. H3149) and aFGF 50 ng/mL (R&D Systems, ref. 232-FA-025/CF), as previously reported (37).

For Western blot assays, the following antibodies were purchased from Cell Signaling Technology: pFGFR (3471S), pAKT (4060S), AKT (4691S), pERK (9101S), ERK (9102S), pS6 (4858S), S6 (2217S), and BIM (2933S). Anti-FGFR3 (ab133644) and anti-pEGFR (ab5644) antibodies were purchased from Abcam, anti-EGFR (sc-03) antibody was purchased from Santa Cruz Biotechnology, and anti–β-actin (A1978) antibody was purchased from Sigma-Aldrich.

WES/RNA-seq raw data files from this study are deposited at the European Genome–phenome Archive (EGA) using the accession code EGAS00001007335. Access to this shared dataset is controlled by the institutional Data Access Committee, and requests for access can be sent to the corresponding author. Further information about EGA can be found at https://ega-archive.org/. Any additional information required to reanalyze the data reported in this article is available upon request from the corresponding author.

F. Facchinetti reports personal fees from BeiGene outside the submitted work. A. Hollebecque reports grants from Incyte and nonfinancial support from Taiho during the conduct of the study, as well as personal fees from Basilea, Debiopharm, Eisai, Incyte, QED Therapeutics, Relay Therapeutics, and Servier, grants from AstraZeneca, nonfinancial support from AbbVie, Adaptimmune, AstraZeneca, Celgene/Bristol Myers Squibb, Loxo/Lilly, Merus, Mirati, Pfizer, Relay Therapeutics, Roche, Sanofi, and Taiho, and personal fees and nonfinancial support from Seattle Genetics outside the submitted work. G. Guaitoli reports personal fees from MSD and other support from AstraZeneca outside the submitted work. S. Michiels reports personal fees (scientific study member) from Roche and personal fees (data and safety monitoring member for clinical trials) from Sensorion, Biophytis, Servier, IQVIA, Yuhan, and Kedrion outside the submitted work. B. Besse reports grants from AbbVie, Amgen, AstraZeneca, Chugai Pharmaceutical, Daiichi Sankyo, Ellipse Pharma, Eisai, Genmab, Genzyme Corporation, Hedera Dx, Inivata, Ipsen, Janssen, MSD, PharmaMar, Roche/Genentech, Sanofi, Socar Research, Taiho Oncology, and Turning Point Therapeutics during the conduct of the study. Y. Loriot reports grants, personal fees, and nonfinancial support from Janssen during the conduct of the study, as well as personal fees and nonfinancial support from MSD, Pfizer, Merck KGaA, Astellas, Gilead, and Bristol Myers Squibb, grants, personal fees, and nonfinancial support from Roche, and nonfinancial support from Incyte and Exelixis outside the submitted work. L. Friboulet reports grants from Debiopharm and Incyte, and nonfinancial support from Illumina during the conduct of the study, as well as grants from Relay Therapeutics, Sanofi, and Nuvalent, and nonfinancial support from Guardant Health outside the submitted work. No disclosures were reported by the other authors.

F. Facchinetti: Conceptualization, resources, data curation, software, formal analysis, investigation, methodology, writing–original draft. A. Hollebecque: Resources, supervision, validation, investigation, writing–review and editing. F. Braye: Data curation, formal analysis. D. Vasseur: Resources, data curation, software, writing–review and editing. Y. Pradat: Resources, data curation, software, formal analysis, writing–review and editing. R. Bahleda: Resources, formal analysis, investigation. C. Pobel: Resources, formal analysis. L. Bigot: Resources, data curation, formal analysis, investigation. O. Déas: Resources, formal analysis. J.D. Florez Arango: Resources, data curation. G. Guaitoli: Resources, formal analysis, writing–review and editing. H. Mizuta: Resources, data curation, writing–review and editing. D. Combarel: Supervision, validation, writing–review and editing. L. Tselikas: Resources, investigation, writing–review and editing. S. Michiels: Data curation, validation. S.I. Nikolaev: Resources, data curation, formal analysis, supervision. J.-Y. Scoazec: Resources, data curation, writing–review and editing. S. Ponce-Aix: Resources, investigation. B. Besse: Conceptualization, investigation. K.A. Olaussen: Supervision, validation, writing–review and editing. Y. Loriot: Conceptualization, resources, investigation. L. Friboulet: Conceptualization, resources, data curation, software, supervision, funding acquisition, validation, methodology, project administration, writing–review and editing.

We thank Illumina for performing ctDNA sequencing analysis in the setting of an academic collaboration. The work of F. Facchinetti is supported by a grant from Philanthropia–Lombard Odier Foundation. The work of G. Guaitoli is supported by MIUR Dipartimenti Eccellenti 2017. The work of S.I. Nikolaev is supported by a grant from Foundation ARC 2017, Foundation Gustave Roussy, and the Swiss Cancer Research Foundation (KFC3985-08-2016). The work of L. Friboulet is supported by an ERC Consolidator Grant (agreement number 101044047). Financial support was provided by Incyte and Debiopharm.

The publication costs of this article were defrayed in part by the payment of publication fees. Therefore, and solely to indicate this fact, this article is hereby marked “advertisement” in accordance with 18 USC section 1734.