End points

The primary end points were ORR (percentage of patients with complete responses or partial responses) confirmed by independent review committee (IRC) per Response Evaluation in Solid Tumors (RECIST) v.1.1 criteria or Response Assessment in Neuro-Oncology (RANO) in cohorts A and B. Secondary end points were duration of response (DOR), IRC-assessed progression-free survival (PFS), overall survival (OS) and safety and tolerability as assessed by the incidence, type, and severity of adverse events (AEs) in cohorts A and B. Selected exploratory end points were ORR, DOR, PFS and OS in cohort C and genomic analysis of baseline and on-treatment tumor and plasma samples for markers of response and pemigatinib resistance. IRC-assessed clinical benefit rate (CBR) in all cohorts was conducted as a post hoc analysis.

Patients

Between 17 October 2019 and 12 July 2021, 111 patients enrolled. Of these, 107 patients were divided into three cohorts: A (FGFR1–FGFR3 fusions/rearrangements; n = 49), B (activating FGFR1–FGFR3 non-kinase domain single-nucleotide variants (SNVs); n = 32) or C (FGFR1–FGFR3 kinase domain mutations or variants of unknown significance (VUS) with potential pathogenicity; n = 26; Fig. 1a). Four remaining patients were included in the safety analysis but were excluded from the efficacy analysis per protocol because their FGFR alterations were not centrally confirmed (Supplementary Table 1). All patients received pemigatinib 13.5 mg orally once daily (QD) continuously. Of the patients in the efficacy-evaluable cohorts, 89 had ctDNA analysis for plasma collected at baseline and, among these, 73 had both baseline and progression samples (Fig. 1b).

Fig. 1: Patient disposition and samples for genomic analysis.

a, Patient disposition. b, Samples for genomic analysis. The primary reason for treatment discontinuation is shown for each patient. *FoundationOne, FMI. †The four patients originally misassigned to cohort C based on local test uncertainty were analyzed here with the relevant set of gene alterations in cohorts A and B. EOT, end of treatment; FMI, Foundation Medicine, Inc.

Median age among efficacy-evaluable patients was 62 (range, 25–84) years. Overall, 57% of patients were women, 69% were white and 23% were Asian (Table 1). Cholangiocarcinoma (16%), urothelial tract/bladder cancer (11%) and glioblastoma (9.3%) were the most common tumors. Duration of treatment was longest in cohort A (median [range], 4.1 months [0.3–20.2]), followed by cohort B (3.2 months [0.2–15.4]) and cohort C (2.1 months [0.2–18.6]). The most common primary reason for treatment discontinuation was disease progression (77%) and the least common primary reason was AEs (5.4%).

Table 1 Patient demographics and baseline clinical characteristicsEfficacy

The primary end points were ORRs in cohorts A and B. ORR (95% confidence interval (CI)) in cohort A was 27% (15%, 41%; n = 13) and 9.4% (2%, 25%; n = 3) in cohort B. ORR (95% CI) in cohort C, which was an exploratory end point, was 3.8% (0.1%, 20%; n = 1; Fig. 2 and Table 2). One patient in cohort A had a complete response. Secondary end points were DOR, PFS and OS in cohorts A and B. Median DOR was 7.8 months in cohort A and 6.9 months in cohort B. Median PFS and OS in cohort A were 4.5 and 17.5 months, respectively, and 3.7 and 11.4 months in cohort B, respectively. Efficacy outcomes are summarized in Table 2 and Extended Data Fig. 1.

Fig. 2: Best percent change from baseline by FGFR co-alteration subgroup.

Best percent change from baseline by RECIST or RANO for all evaluable patients with tissue NGS report and reported best change in lesion size: FGFR fusions/rearrangements (n = 48); FGFR actionable SNVs (n = 32); FGFR kinase domain mutations or VUS (n = 20). Best OR and PFS by IRC indicated where evaluable. Patients are arranged by FGFR alteration type. Bars are colored by major tumor histologies. Dashed lines indicate a criterion for partial response (change from baseline in target lesion size ≥30%). Tumors are grouped into the following histologies based on ≥5 patients: Cholangiocarcinoma, gynecologic cancers (cervical, endometrial and uterine), CNS (glioblastoma, low-grade pediatric glioma and astrocytoma), pancreatic cancer, breast cancer, urothelial tract/bladder cancer, non-small cell lung cancer and other (adrenal cancer, anal cancer, cancer of unknown primary origin, colorectal cancer, gastric/gastroesophageal cancer, gallbladder cancer, giant cell bone tumor, head and neck cancer, lung neuroendocrine cancer, nasopharyngeal cancer, ovarian cancer, prostate cancer, renal cell cancer, sarcoma and solitary fibrous tumor). Genomic analysis is included for all reportable samples and included NGS analysis of tumor tissues and ctDNA at baseline, and of ctDNA at time of progression (gray boxes indicate no report).

Table 2 Efficacy outcomes

Objective responses were observed in multiple tumor types, including histologies for which no FGFR inhibitors are approved (Fig. 3 and Supplementary Table 2). Histologies of particular note included central nervous system (CNS) tumors, pancreatic tumors (all KRAS wild-type), cervical tumors and urothelial carcinomas harboring FGFR fusions or mutations.

Fig. 3: Best percent change from baseline by tumor type.

Best percent change from baseline by RECIST or RANO (denoted by +) for all evaluable patients with tissue NGS report and reported best change in lesion size; BOR and PFS by IRC indicated where evaluable. Patients are arranged by major tumor histologies as previously described. Bars are colored by FGFR alteration type. Dashed lines indicate a criterion for partial response (change from baseline in target lesion size ≥30%; top) and clinical benefit (PFS ≥ 6 months; bottom). Tumors are grouped into the following histologies based on ≥5 patients: Cholangiocarcinoma, gynecologic cancers (cervical, endometrial and uterine), CNS (glioblastoma, low-grade pediatric glioma and astrocytoma), pancreatic cancer, breast cancer, urothelial tract/bladder cancer, non-small cell lung cancer and other (adrenal cancer, anal cancer, cancer of unknown primary origin, colorectal cancer, gastric/gastroesophageal cancer, gallbladder cancer, giant cell bone tumor, head and neck cancer, lung neuroendocrine cancer, nasopharyngeal cancer, ovarian cancer, prostate cancer, renal cell cancer, sarcoma and solitary fibrous tumor).

Safety

Among 111 patients who received ≥1 dose of pemigatinib, no new safety signals were seen. A full list of treatment-emergent AEs (TEAEs) is provided in Supplementary Table 3. The rate of grade ≥3 TEAEs was 68% (Extended Data Table 1). Fatal TEAEs occurred in six patients and included general physical health deterioration (n = 3; 2.7%), acute respiratory failure (n = 1; 0.9%), confusional state (n = 1; 0.9%) and sepsis (n = 1; 0.9%). None of the fatal TEAEs was considered by investigators to be related to pemigatinib. TEAEs leading to dose interruption and reduction occurred in 79 (71%) and 48 (43%) patients, respectively. Eight (7.2%) patients discontinued pemigatinib due to TEAEs. The most common any-grade TEAEs were hyperphosphatemia (84%) and stomatitis (53%). Nail toxicities and serous retinal detachment occurred in 45% and 14% of patients.

Genomic analysis of putative primary driver FGFR alterations

Clinical genomic analysis was performed on tissue and plasma samples collected from patients in cohorts A, B and C. Four patients from cohort C, initially determined with local testing to have VUS, were reassigned for this translational analysis to the other cohorts based on central review and reconsideration of their gene alterations. DMBT1-FGFR2 (patient 16) and FGFR1 rearrangements with indeterminate partner (patient 26 and patient 48) were assigned to cohort A and FGFR3 G370C (patient 57) was assigned to cohort B.

Among the FGFR gene alterations, fusions were most sensitive to FGFR inhibition (Fig. 2). The majority of patients in this cohort had type II FGFR fusions (n = 49; 94%), wherein FGFR was the 5′ fusion gene and the breakpoint occurred after the kinase domain in the region spanning intron 17 to exon 18 (ref. 18). Three additional rearrangements (BAG4-FGFR1, RGS12-FGFR3 and DMBT1-FGFR2) were considered putative type I fusions, a less-common oncogenic FGFR rearrangement observed primarily in MLNs, wherein a 5′ partner gene fuses with FGFR at a breakpoint after the transmembrane domain18. Both type I and II fusions are typically oncogenic and can be sensitive to FGFR inhibition. Although FGFR fusions and rearrangements were the most responsive gene alterations across tumor histologies, response was not uniform across histologies; differential rates of objective response and clinical benefit may indicate differential dependencies on FGFR across histologies with common gene alterations subgroups; however, given the relatively small populations evaluated for each histology, analysis of larger populations will likely be required for a more definitive assessment of FGFR pathway dependencies.

FGFR non-kinase domain SNVs that were considered actionable based on publicly available alterations databases or clinical study data (cohort B) were localized in extracellular and transmembrane domains. Among these FGFR SNVs, clinical benefit was observed for patients with urothelial carcinoma (n = 4), cholangiocarcinoma (n = 3) and squamous cell carcinoma (n = 1). Among five patients with intrahepatic cholangiocarcinoma that had FGFR2 SNVs, two (C382R (patient 79) and extracellular domain in-frame deletion I291_Y308D del (patient 78)) experienced partial response and two (W290C (patient 75) and Y375C (patient 77)) had stable disease with PFS of 10.5 and 3.7 months, respectively. While cholangiocarcinomas harboring these actionable mutations are less prevalent than FGFR2 rearrangements, they seem to represent an additional population that may benefit from FGFR inhibition.

FGFR kinase domain mutations (cohort C) were considered to be of uncertain actionability given that some kinase domain mutations demonstrate reduced sensitivity to FGFR inhibitors, including pemigatinib in preclinical models19. Notably, 2 of 12 patients with FGFR kinase domain mutations experienced clinical benefit. One patient with FGFR1 K656E grade II diffuse astrocytoma had a partial response (patient 100) and one patient with an FGFR1 N546K low-grade pediatric type glioma had stable disease and a 6.2-month PFS. Notably, activating mutations in K656 in the FGFR1 activation loop and N546, a controlling residue in the ‘molecular brake’ function, represent the two most common sites of activating FGFR1 SNVs in gliomas and other CNS tumors; however, among the remaining ten patients with kinase domain mutations without clinical benefit, eight had mutations in molecular brake residues (Extended Data Table 2; FGFR1 N546K/D (n = 5); FGFR2 N549K (n = 3)). Four additional patients in cohort C had mutations downstream of the FGFR2 kinase domain (patients 82, 89, 98 and 99). These mutations produce truncations before exon 18 and were recently described to be potentially pathogenic17. Among these, two patients (Q774* (patient 99) and E769fs (patient 98)) had stable disease ≥6 months, suggesting a modest but real clinical benefit.

Tissue next-generation sequencing (NGS) analysis also identified instances of FGFR amplification (defined as FGFR copy number ≥6). Concurrent FGFR gene amplifications were detected in nine patients (Supplementary Table 4), including concurrent amplifications with the corresponding FGFR mutation (n = 4) or FGFR fusion/rearrangement (n = 1) as well as FGFR amplifications occurring in an alternative FGFR to the enrollable FGFR gene alteration (n = 4). There were not enough patients in FIGHT-207 with concurrent FGFR gene amplification to conclude whether it had a meaningful impact on response to pemigatinib.

Correlation of co-alterations with patient outcomes

This FIGHT-207 basket study provided the opportunity to assess possible patterns of intrinsic resistance associated with co-alterations across multiple histologies and multiple FGFR alterations using combined genomic analysis of tumor tissue and ctDNA. Among patients with FGFR fusions/rearrangements and actionable SNVs (cohorts A and B, respectively), 79 evaluable patients had baseline tissue sequencing and 55 of these additionally had baseline ctDNA sequencing. Baseline ctDNA analysis had limited concordance with tissue NGS analysis for detection of FGFR variants and some co-alterations across all study samples (Supplementary Fig. 1), likely explained by multiple technical (for example, assay sensitivity, analytical thresholds for variant reporting and variable variant annotations) and biological (for example, age of samples and variable ctDNA shedding) factors. This correlation analysis is therefore focused on the complementary value of combining the gene alterations detectable by the two methods. Tumors were categorized as having a specific co-mutation if this mutation was seen by tissue or ctDNA analysis or both. Based on baseline tissue NGS analysis alone, patterns seen in patients with FGFR2 fusion-positive cholangiocarcinoma in FIGHT-202 were recapitulated here across multiple histologies harboring a variety of FGFR1–FGFR3 fusions and mutations. Specifically, none of 27 patients with tumors harboring alterations in TP53 had an objective response. Moreover, patients with tumors with TP53 alterations or one of several other tumor-suppressor genes had a lower PFS than those with wild-type copies of these genes (Extended Data Table 3). New correlations seen in FIGHT-207 included the associations with oncogenic alterations in the MAPK pathway or inactivating alterations in ARID1A with low PFS and between alterations in BAP1 and high clinical benefit. Notably, by baseline ctDNA analysis alone, these associations with ARID1A, MAPK pathway and BAP1 alterations held, but the association seen with TP53 and tumor-suppressor gene alterations did not (Extended Data Tables 4–6).

Acquired resistance in multiple histologies

All 73 patients who had post-progression ctDNA samples with matched baseline ctDNA also had baseline tumor biopsy molecular profiling. Fourteen (19%) patients acquired one or more secondary FGFR mutation in the kinase domain, in residues known or likely to confer resistance (Extended Data Table 7)20,21,22,23,24,25. For patients with cholangiocarcinoma, kinase domain mutations emerged exclusively in patients with clinical benefit from pemigatinib, supporting the case for acquired-resistance mechanisms. While diverse FGFR1–FGFR3 alterations and multiple tumor types were represented, the common pattern across histologies was the emergence of mutations in the gatekeeper residues (FGFR2 V564F/I/L; FGFR3 V555L/M) or closely neighboring residues (FGFR1 V559L/M) and molecular brake residues (FGFR1 N546K; FGFR2 N549D/H/K, E565A and K641R). Other emergent FGFR2 mutations included M537I, L617V and K659M. Ten of 14 (71%) patients developed polyclonal FGFR resistance mutations, with most patients developing concurrent gatekeeper and molecular brake residue mutations and many developing co-occurring mutations at the same codon (N549K and N549D). No mutations in an FGFR gene other than the originally altered FGFR gene were detected in post-progression plasma samples (for example, FGFR2 mutations were not detected in FGFR1-altered tumors).

In addition to secondary FGFR variants, new mutations in co-altered genes emerged in end-of-treatment but not baseline plasma ctDNA samples that may be associated with resistance as they involved TP53, PIK3CA and/or RAS (Extended Data Fig. 2)26,27. A larger set of additional emergent variants is presented in Extended Data Fig. 3.

Pooled co-alteration data from pemigatinib studies

To increase the power of our analysis, we investigated pooling the FIGHT-207 data with datasets from previous pemigatinib clinical studies, including FIGHT-101 (ref. 9) (phase 1/2; multiple histologies), FIGHT-201 (ref. 28) (phase 2; urothelial tract/bladder cancer) and FIGHT-202 (ref. 26) (phase 2; cholangiocarcinoma) in which co-alteration analysis has been previously reported. This analysis included patients with available tissue NGS analysis, FGFR fusions/rearrangements or actionable FGFR SNVs, centrally determined best overall response and treatment with pemigatinib at or above the recommended dose. Combined FIGHT-101 (n = 20) and FIGHT-207 (n = 72) data increased the power of the analysis for various solid tumors, but did not result in any change to the identification of co-altered genes significantly correlated with best overall response to pemigatinib. The tumor suppressors BAP1 and TP53 remained the genes whose alteration correlated significantly with objective response (Supplementary Table 5). Similarly, analysis of combined FIGHT-202 (n = 104) and FIGHT-207 (n = 11) data for patients with cholangiocarcinoma (Supplementary Table 6) did not result in any change to the identification of co-altered genes significantly correlated with best overall response to pemigatinib, and only TP53 was found to be nominally significant (significance was not maintained following stringency correction for multiple testing). Combined FIGHT-201 (n = 149) and FIGHT-207 (n = 13) data for patients with urothelial carcinoma (Supplementary Table 7) identified TSC1, which was reported in earlier analysis and CDKN1A, which was now found to be correlated nominally significantly with objective response. Notably, a combined analysis including samples from all four studies was not considered to be valid due to skewing resulting from the inclusion of larger sample sets for cholangiocarcinoma and urothelial carcinoma. This imbalance precludes inference of global correlations of co-alterations with response to pemigatinib.