Acquired FGFR and FGF Alterations Confer Resistance to Estrogen Receptor (ER) Targeted Therapy in ER⁺ Metastatic Breast Cancer

Approximately 70% of breast cancers express the estrogen receptor (ER), and estrogen signaling drives breast cancer cell growth and progression (1). Although endocrine therapies, including tamoxifen, aromatase inhibitors (AI), and the selective ER degrader (SERD) fulvestrant have improved survival for patients with ER⁺ breast cancer, within the metastatic setting resistance to endocrine therapies is nearly universal (2).

Although various resistance mechanisms have been proposed for tamoxifen and AI resistance, including loss or modification in ER expression (ESR1-activating mutations and ESR1 fusions; refs. 3–7), and regulation of alternative signal transduction pathways (PI3K/AKT/mTOR and EGFR/ERBB2/MAPK; refs. 8–10), mechanisms of resistance to SERDs remain understudied. Mechanisms of endocrine resistance identified in patients include acquired mutations in the ER itself (4–7), acquired activating mutations in ERBB2 (HER2; refs. 11, 12), loss of function of NF1 (13), and other alterations in MAPK pathway genes (14). Additional clinically relevant mechanisms remain to be identified.

Gain-of-function screens have played a pivotal role in identification of resistance mechanisms to targeted therapies in various cancer types (15–17). In breast cancer, several functional screen studies identified IGF1R, KRAS, and ESR1 as mechanisms of resistance to tamoxifen and/or estrogen deprivation (18–20). However, genome-scale functional screens for SERD resistance have not been reported.

We conducted a genome-scale gain-of-function screen in ER⁺ breast cancer cells spanning 17,255 overexpressed lentiviral open reading frames (ORF) to investigate genes whose overexpression was sufficient to confer resistance to the SERDs fulvestrant and GDC-0810 (21). In parallel, we sought to identify endocrine resistance mechanisms of clinical significance through genomic profiling of paired pretreatment and posttreatment tumor samples from 60 patients with ER⁺ metastatic breast cancer (MBC) who developed resistance to endocrine therapy.

293T, T47D, and MCF7 cells were obtained directly from ATCC as frozen vials and were cultured as described in the Supplementary Methods. MCF7 and T47D were validated by Western blotting for ER and HER2 (based on known genotype). We used cell lines of passage number ranging from 4 to 15 in the described experiments. All cell lines tested negative for Mycoplasma contamination by ATCC Universal Mycoplasma Detection Kit. The last test was performed in November 2019.

The pooled lentiviral ORF library hORFeome (22) consists of 17,255 barcoded human ORFs, corresponding to 10,135 distinct human genes with at least 99% nucleotide and protein match. These ORFs were cloned into pLX317 vector and pooled together for transfection into 293T cells to make pooled lentivirus (with 2nd generation packaging plasmids). In 6-well plates, pooled lentivirus was infected in cells to achieve approximately 50% infection rate and ensure approximately 1,000 infected cells per ORF for 17,255 ORFs. Media was supplemented with 4 μg/mL polybrene (Thermo Fisher Scientific # TR1003G) to boost transfection efficiency. After infection, cells were pooled and selected with 1.5 μg/mL puromycin for 5 days. Upon completion of selection, cells were plated for three different drug conditions: DMSO, 100 nmol/L fulvestrant, 1 μmol/L GDC-0810. There were three replicates for each condition screened. A subset of cells was saved for sequencing as early time point (ETP) samples to confirm ORF representation. Infected cells were passaged upon confluency and maintained in DMSO or drugs for 21 days to allow sufficient time for cells carrying resistance to be enriched from the population. At the end of the time course, cells were harvested for isolating genomic DNA as late time point samples (LTP). All genomic DNA samples were amplified with PCR primers flanking the ORF region and sequenced. The ORF representation at the final harvesting (LTP) is compared to the representation of ORFs in cells collected before drug addition (ETP). Cells carrying ORFs that are driving resistance will grow and gradually enrich the population and therefore, will be over-represented in the sequencing data for the final passage compared to the early time point. An ORF with significant enrichment (a Z-score >3) is defined as a resistance candidate gene. A secondary validation screen was performed as described in the Supplementary Methods.

Prior to any study procedures, all patients provided written informed consent for research biopsies and whole-exome sequencing (WES) of tumor and normal DNA, as approved by the Dana-Farber/Harvard Cancer Center Institutional Review Board (DF/HCC Protocol 05-246). Metastatic core biopsies were obtained from patients and samples were immediately snap frozen in OCT and stored in −80°C. Archival FFPE blocks of primary tumor samples were also obtained. A blood sample was obtained during the course of treatment, and whole blood was stored at −80°C until DNA extraction from peripheral blood mononuclear cells (for germline DNA) was performed. In a few instances, cell free DNA was obtained from plasma for circulating tumor DNA analysis, as described previously (23). The studies were conducted in accordance with U.S. Common Rule for ethical guidelines.

DNA was extracted from primary tumors, metastatic tumors, plasma, and peripheral blood mononuclear cells (for germline DNA) from all patients and WES was performed, as detailed in the Supplementary Methods. Sequencing data were analyzed using tools to identify somatic point mutations and small insertions/deletions (indels), and copy number changes using established algorithms (see Supplementary Methods).

To better measure segment-specific copy number, we subtracted the genome ploidy for each sample to compute copy number above ploidy (CNAP). CNAP of at least 3 are considered as amplifications (AMP); CNAP below 3 are considered low amplification and ignored in our analysis. CNAP of at least 6 are considered high amplifications (HighAMP), and CNAP of at least 9 and fewer than 100 genes (24) is considered very high focal amplification (FocalAMP).

The evolutionary classification of amplifications accounts for the magnitude of the observed copy-number difference between the pretreatment and the posttreatment samples. We used the same method as previously described to make the evolutionary classifications (25). If the difference between the CNAP of the posttreatment and the CNAP of the pretreatment is smaller than 50%, the amplification is defined as “Shared”. If the CNAP of the post-treatment is larger than the CNAP by more than 50% and the lower pretreatment CNAP is not at “FocalAMP” level, the evolutionary classification is “Acquired”. If CNAP of the posttreatment is smaller by at least 50%, comparing with the pretreatment sample and the lower posttreatment CNAP is not at “FocalAMP” level, the evolutionary classification is “Loss”. Otherwise, the evolutionary classification of amplifications is defined as “Indeterminate”.

We performed RNA sequencing (RNA-Seq) on T47D cells perturbed to overexpress FGFR pathway activation including FGFR1, FGFR2 (WT, K660N, M538I, and N550K), and FGF3, as well as GFP and parental as a control, as described above. Cells were plated in 96-well plates, and then treated with DMSO, fulvestrant (100 nmol/L), palbociclib (1 μmol/L), FIIN-3 (100 nmol/L), and trametinib (500 nmol/L) as single agent and in combinations for 24 hours. FGFR1/2 cell lines were treated with or without FGF2 (10 ng/mL) in various conditions. RNA was extracted from cells and sequencing libraries prepared as described in Supplementary Methods. For each specific construct and treatment combination, we performed at least 6 replicates, for a total of 672 RNA-Seq profiles.

T47D or MCF7 cells were infected with lentivirus to derive stable cell lines overexpressing wild type (WT) or mutant ORFs. All WT ORFs were obtained from the Broad Institute. Mutant ORFs (FGFR2 M538I, N550K and K660N) were made using QuickChange II Site-Directed Mutagenesis Kit (Agilent Technologies #200523). Most stable cells lines express ORFs in pLX317 vector and were selected with puromycin (Life Technologies #A1113803). Stable cell lines expressing CCND1 and PIM1 in pLX304 vector were selected with blasticidin (Life Technologies #A1113903).

Cells were plated in 96-well tissue culture ViewPlate (PerkinElmer # 6005181) on day 1 and treated with drug on day 2. Media with or without drugs was refreshed on day 5. On day 8, cells were equilibrated to room temperature, media was removed, and cells were lysed in a mixture of 50 μL media and 50 μL CellTiter-Glo 2.0 reagent (Promega # G9243) per well. Plates were then incubated on an orbital shaker for 2 minutes. Following another 10 minutes of incubation at room temperature to stabilize signal, luminescence was recorded to measure cell viability on Infinite M200 Pro microplate reader (Tecan).

Western blotting was performed as described in Supplementary Methods.

Tumor and germline WES data generated and analyzed for this study have been deposited in the access-controlled public repository dbGAP with accession code phs001285 (https://www.ncbi.nlm.nih.gov/gap). RNA-Seq data are available through GEO under accession GSE153509. Additional data generated in this study including tumor exome analysis and RNA-Seq data are available within the paper and in the supplementary information files.

Statistical analyses related to drug response curve were performed with two-tailed Student t test in GraphPad Prism. Fisher exact test was used to calculate OR and q-value for volcano plots in the RNA-Seq analysis. Cohen D test with Hedges correction and Welch t test were used to estimate the effect size and significance for signature strength of gene sets.

To identify the spectrum of genes whose overexpression confers resistance to SERDs in vitro, we expressed a pooled lentiviral library of 17,255 human ORFs, corresponding to 10,135 distinct human genes, in ER⁺ T47D breast cancer cells in the presence of fulvestrant or GDC-0810 (22). Genes that confer drug resistance will be enriched under drug selection for 21 days, indicated by a positive log fold change (LFC) for ORF representation before and after DMSO/drug selection.

Using a Z-score >3 as a criterion to identify resistance candidates, we identified 64 genes (93 ORFs) that conferred resistance to fulvestrant and 57 genes (83 ORFs) that conferred resistance to GDC-0810 (Fig. 1A; Supplementary Table S1). Thirty-seven genes (55 ORFs) conferred resistance to both drugs, a degree of overlap which was anticipated given the mechanistic similarities between fulvestrant and GDC-0810. The LFC and corresponding Z-score for each ORF in fulvestrant and GDC-0810 treatment arms were highly correlated, with a correlation coefficient of 0.77 (Fig. 1A).

To confirm these results, we conducted a secondary screen using a smaller pooled library consisting of 570 ORFs in T47D and MCF7 cells to validate candidates nominated by the primary screen. Both cell lines are commonly used and well-characterized ER⁺ breast cancer cell and both are sensitive to SERDs, although they have different levels of ER signaling and harbor some genomic differences (https://portals.broadinstitute.org/ccle; refs. 26, 27). Top resistance genes found in the primary screen were again enriched in the secondary screen for T47D cells, including FGF genes, FOXR1, AKT genes, PIM genes and several GPCR genes (Supplementary Fig. S1A). Many top ranked resistance genes (CSF1R, FGF3, FGF6, FOXR1, and PIM2) were shared between T47D and MCF7 cells (Supplementary Fig. S1B). However, distinct resistance genes were also observed in each cell line, suggesting some resistance mechanisms may be cell context-dependent.

Functional categories of candidate resistance genes include serine/threonine kinases (PIK3CA, AKT1/2/3, PIM1/2/3), receptor tyrosine kinases (EGFR, ERBB2, PDGFRB), growth factors (FGF3/6/10/22), cell-cycle regulatory proteins (CCND1, CCND2, CCND3, CDK6), and G-protein–coupled receptors (GPCR; Fig. 1B). As further validation, we overexpressed 13 ORFs belonging to these categories individually in T47D cells and they conferred resistance to fulvestrant and GDC-0810 to various degrees (Fig. 1C; Supplementary Fig. S2A and S2B).

Gene set enrichment analysis (GSEA) of the candidate resistance genes demonstrated enrichment in 4 functional pathways: FGFR signaling, ERBB signaling, insulin receptor signaling, and the MAPK pathway (Fig. 1D; Supplementary Fig. S3; Supplementary Table S2). Consistent with this, we and others recently demonstrated that ERBB2-activating mutations (11) and alterations in MAPK pathway genes can cause endocrine resistance in patients with ER⁺ MBC (13, 14, 25). We sought to further examine the role of FGFR and FGF genes in resistance to SERDs in MBC.

To examine the potential role of FGFR and FGF alterations in the development of endocrine resistance clinically, we analyzed WES data from paired pretreatment and posttreatment metastatic tumor biopsies or cell free DNA from 60 patients with ER⁺ MBC who had received at least one endocrine therapy (tamoxifen, AI, SERDs) for more than 120 days between the two biopsies (28).

Amongst the 60 posttreatment samples, we found FGFR1 amplifications in 15% (9/60), FGFR2 amplifications in 5% (3/60), FGFR2-activating mutations in 3.3% (2/60), and FGF3 amplifications in 28.3% (17/60), for a total of 40% (24/60) of the cohort with at least one alteration in one of these three genes (Fig. 2A). Overall, the prevalence of FGFR1, FGFR2, and FGF3 alterations in the resistant metastatic setting seen here is increased compared to what was observed in previously published cohorts of primary ER⁺ breast cancer, such as The Cancer Genome Atlas (TCGA; ref. 29), and comparable to other datasets of metastatic ER⁺ breast cancer (Supplementary Table S3). The incidence of FGFR2 alterations (6.7%), in particular, is markedly increased compared with primary treatment-naive breast cancer, in which the incidence is less than 2% in TCGA (Supplementary Table S3).

To determine whether this enrichment of FGFR/FGF alterations in the metastatic setting was due to acquisition/selection under the selective pressure of endocrine therapy, we performed an evolutionary analysis to evaluate clonal structure and dynamics, including changes in mutations and copy number for the 24 patients harboring FGFR/FGF alterations. For this analysis, we define “acquired” alterations as alterations with higher representation in the posttreatment sample as compared with the pretreatment sample (see Materials and Methods). Although we use the term “acquired”, we recognize that when the mutation is not detected in the pretreatment sample, we cannot distinguish between preexisting alteration that was selected for and clonally enriched versus de novo alterations that developed during the treatment.

In 12 of the 24 patients with FGFR or FGF alterations (50%), the alterations were acquired in the posttreatment sample as compared with the pretreatment sample (Fig. 2A, marked in red). Five of nine FGFR1 amplifications were acquired (55.6%), while all four FGFR2 alterations were acquired (100%), including one patient (Pt 0300350) with acquisition of both an FGFR2 mutation and amplification. FGF3 amplifications were acquired in 4 of 17 tumors (23.5%), including one case in which an FGFR1 amplification was coacquired. The concurrent acquisition may suggest that the evolutionary selection of both the ligand and receptor provided additional fitness in this tumor. Among the other 12 patients, the alterations in 8 patients were shared in both pretreatment and posttreatment samples (Fig. 2A, marked in black), and evolutionary status of alterations in the remaining 4 patients was inconclusive (Fig. 2A, marked in gray). The increase in copy number (corrected for tumor purity and ploidy, Supplementary Table S4) from pretreatment to posttreatment for FGFR1, FGFR2, and the FGF3 amplicon in all 12 patients is depicted in Fig. 2B.

Two of the acquired alterations found in these 12 patients were single-nucleotide variants (SNV) in the FGFR2 gene: M538I and N550K. N550K is the most common FGFR2 mutation in breast cancer while M538I was previously identified in lung cancer but has not yet been characterized in breast cancer (30). Figure 2C illustrates the change in the estimated fraction of tumor cells harboring each genomic alteration (CCF) from the pretreatment biopsy to the resistant biopsy. In both patients, the FGFR2 mutations were either not detected in the primary tumor (N550K in Pt 0300350) or detected in a small fraction (CCF of 2%) of the pretreatment tumor (M538I in Pt 0300348). In both patients, the activating FGFR2 mutations in the posttreatment biopsies were clonally acquired (CCF of 100%; Fig. 2B).

Notably, the acquired alterations in FGFR1, FGFR2, and FGF3 were largely mutually exclusive with acquired ESR1 mutations and HER2 mutations (Fig. 2A). ESR1 mutations are the most common mechanism described for acquired endocrine resistance (31). Although the overall rate of acquired ESR1 mutation in this cohort is 22% (13/60), among the 12 cases of acquired FGFR and FGF alterations, only one patient also has an acquired ESR1 mutation (Fig. 2A). Similarly, only one of these 12 patients had acquired a HER2 mutation ERBB2 I628M (which was an alteration of unknown significance), suggesting that these are also mutually exclusive mechanisms of resistance.

Although we highlight FGF3 as the key gene in the amplicon (given the results of the gain-of-function screen), FGF3 resides in genomic proximity to FGF4, FGF19, and CCND1 and these four genes are often coamplified. Here, in 3 of the 4 cases with acquired FGF3 amplification, FGF3 copies were gained without coacquisition of CCND1 amplification, suggesting that this acquisition can occur as an independent genomic event (Supplementary Fig. S4). Similarly, 2 of the 4 cases with acquired FGF3 did not have coacquisition of FGF19 amplification. In all 4 of the cases with FGF3 amplification, FGF4 was also co-amplified. The other concurrent genetic alterations for the 12 patients with acquired FGFR/FGF alterations are shown in Supplementary Fig. S5 and Supplementary Tables S5 and S6.

Figure 3 depicts clinical vignettes for 6 of the patients with acquired FGFR1, FGFR2, and/or FGF3 alterations in their posttreatment biopsies. All patients were treated with ER-directed therapy before acquiring FGF or FGFR alterations, including tamoxifen (3 patients), AIs (6 patients), and fulvestrant (3 patients). Vignettes for the other 6 patients with acquired FGFR1, FGFR2, and/or FGF3 alterations in their posttreatment biopsies are shown in Supplementary Fig. S6A. Detailed clinicopathologic features and therapy details for all 12 patients are found in Supplementary Table S7.

In addition to these 12 patients in our cohort, we identified several additional patients with acquired activating mutations in FGFR1(N546K) and FGFR2 (N550K, K660N) following the development of resistance to endocrine therapy (Supplementary Fig. S6B; Supplementary Table S8).

In summary, we observed acquired alterations in FGFR1, FGFR2, or FGF3 in 20% (12/60) of patients with endocrine resistant ER⁺ MBC, comparable with the known frequency of acquired mutations in ESR1, highlighting the important role of the FGFR pathway in acquired resistance to ER-directed therapies.

To further investigate how FGFR/FGF genes may confer resistance to ER-directed therapy, we treated T47D cells with FGF3, FGF6, FGF10, or FGF22 ligand. Each of these ligands resulted in resistance to fulvestrant (Fig. 4A). This effect was reversed by PD173074, a pan-FGFR inhibitor (Fig. 4A). The addition of these FGF ligands enhanced phosphorylation of ERK and AKT, which was reversed by PD173074 (Supplementary Fig. S7A). FGF3, FGF6, FGF10, and FGF22 also reduced fulvestrant sensitivity in MCF7 cells (Supplementary Fig. S7B and S7C).

We next overexpressed FGFR1, FGFR2, or GFP in T47D cells through lentiviral transduction and examined the impact on susceptibility to SERDs. Overexpression of FGFR1 or FGFR2 alone did not affect sensitivity to fulvestrant. However, with the addition of FGF2 ligand, both FGFR1 and FGFR2 rendered cells highly resistant to fulvestrant (Fig. 4B). A similar resistance phenotype was also observed when T47D cells expressing FGFR1 or FGFR2 were treated with GDC-0810 or tamoxifen (Supplementary Fig. S8A–S8C). In comparison, FGF2 ligand alone reduced sensitivity to SERDs in control cells expressing GFP to a much lesser extent than in the FGFR1 or FGFR2-expressing cells, suggesting the potent resistance phenotype requires both FGF ligand and receptor. This requirement for the presence of both FGF ligand and receptor for maximal resistance phenotype may also explain why only FGFs but not FGFR1 or FGFR2 scored in the resistance screen (Fig. 1A and B). The resistance phenotype resulting from FGFR1 and FGFR2 overexpression was completely reversed by the addition of PD173074 (Fig. 4B). Similar results were obtained in MCF7 cells (Supplementary Fig. S8E and S8F).

FGFR1 and FGFR2 overexpression (in the presence of FGF2 ligand) induced more potent phosphorylation of AKT and ERK than the GFP control, which was reversed by PD173074 (Fig. 4C; Supplementary Fig. S8D and S8G). These results are consistent with previous findings that FGFR1 activation led to MAPK activation and fulvestrant resistance (32). Collectively, these findings suggest that FGFR1 and FGFR2 cause SERD resistance through the activation of MAPK and/or PI3K/AKT pathways.

We examined the sensitivity of cells overexpressing FGFR1 or FGFR2 to several inhibitors of downstream effectors: the MEK inhibitor trametinib, the AKT inhibitor AZD5363, and the mTOR inhibitor everolimus. FGFR1 or FGFR2 overexpression in the presence of FGF2 led to hypersensitivity to trametinib (Supplementary Fig. S9A), but reduced sensitivity to AKT and mTOR inhibitors (Supplementary Fig. S9A).

We attempted to reverse FGFR-induced resistance to fulvestrant by inhibiting the MAPK pathway. Treatment of FGFR1-overexpressing cells with trametinib partially resensitized cells to fulvestrant, and treatment of FGFR2 overexpressing cells with trametinib fully resensitized the cells to fulvestrant (Fig. 4D). In contrast, trametinib did not reverse an increase in viability induced by FGF2 in the GFP control cells. This is likely because FGF2 alone may not sufficiently increase MAPK signaling to induce a dependency, which is consistent with much lower sensitivity to trametinib in GFP cells as compared to FGFR1 and FGFR2 cells in the presence of FGF2 (Supplementary Fig. S9A and S9B). Colony formation assays produced similar results (Supplementary Fig. S9C). Treatment with the mTOR inhibitor everolimus also partially reversed resistance conferred by FGFR1 or FGFR2 overexpression (Supplementary Fig. S9D). Together, these results suggest that the MAPK pathway is the primary downstream effector of FGFR activation resulting in endocrine resistance.

Because the combination of endocrine therapy and CDK4/6 inhibitors is now a standard-of-care treatment for patients with ER⁺ MBC, we also examined the effect of FGFR signaling on sensitivity to the combination of fulvestrant and the CDK4/6 inhibitor palbociclib. In T47D cells, FGFR1 and FGFR2 overexpression in the presence of FGF2 also conferred resistance to combination treatment of fulvestrant and palbociclib. Resistance to fulvestrant and palbociclib was abrogated by PD173074 and also partially reversed by trametinib (Fig. 4D; Supplementary Fig. S9E), further providing the support for the role of MAPK pathway activation in FGFR-mediated drug resistance. The reversal of resistance by trametinib was accompanied by reduced ERK phosphorylation (Fig. 4E). Similar results were achieved in MCF7 cells, although everolimus was more effective than trametinib in reversing the resistance phenotype by FGFR1 or FGFR2 overexpression in this cell line (Supplementary Fig. S10). In the presence of fulvestrant and palbociclib, FGFR1 or FGFR2 overexpression was accompanied by increased p-Rb and CCND1 levels, both of which were partially reversed by trametinib (Fig. 4E). This is consistent with prior results suggesting that CCND1 was involved in FGF2-mediated drug resistance (33).

Clinical evidence also supports the finding that FGFR alterations can cause resistance to CDK4/6 inhibitors. Following the acquisition of FGFR2 N550K (along with FGFR2 amplification), Pt 0300350 did not respond to the combination of letrozole and palbociclib (Fig. 3), suggesting that FGFR2 alterations may lead to intrinsic resistance to the combination of endocrine therapy and CDK4/6 inhibitors. Another patient with an FGFR2 N550K mutation (FM Patient 2) also did not respond to the combination of fulvestrant and palbociclib (Supplementary Fig. S6B). Collectively, this suggests targeting the FGFR pathway may also be a viable strategy to overcome FGFR/FGF–mediated resistance to SERDs and CDK4/6 inhibitors.

We identified 3 acquired mutations in the kinase domain of FGFR2 in patients who developed resistance to endocrine therapy. Two of these, FGFR2 N550K and K660N, are known activating FGFR2 mutations that have been previously identified in breast cancer (30, 34). The third mutation, FGFR2 M538I, has not been previously reported in breast cancer, but was identified through in vitro screening as a reversible pan FGFR inhibitor resistance mutation and confirmed to increase kinase activity in vitro (Fig. 5A; ref. 35).

We expressed all three FGFR2 kinase domain mutants in T47D cells through lentiviral transduction, as well as wildtype (WT) FGFR2 and GFP as negative controls. All three mutants elicited higher kinase activity than WT FGFR2 constitutively, demonstrated by levels of p-FRS2 (a direct substrate for FGFR2), p-ERK, and p-AKT (Fig. 5B). The addition of FGF2 ligand further enhanced downstream signaling for all FGFR2 mutants, and the enhanced signaling was blocked by PD173074 for FGFR2 M538I and K660N, but not for N550K (Fig. 5B). FGFR2 mutants were also expressed under a tetracycline responsive promoter in T47D cells grown in low doses of doxycycline to determine the functionality at lower expression levels. All three FGFR2 mutations acquired in patients with breast cancer are functionally active; FGFR2 N550K is constitutively active while FGFR2 M538I and K660N may be more ligand-dependent at low levels of expression (Supplementary Fig. S11A).

All 3 FGFR2 mutants led to modest resistance to fulvestrant or the combination of fulvestrant and palbociclib (Fig. 5C), which was enhanced in the presence of FGF2 ligand. PD173074 reversed the resistance phenotype for cells overexpressing FGFR2 M538I and FGFR2 K660N as well as WT FGFR2 to fulvestrant, but not cells overexpressing FGFR2 N550K (Fig. 5C). Similar results were obtained in MCF7 cells (Supplementary Fig. S12).

Because of the differential responses of these FGFR2 mutants to PD173074, we tested the ability of additional FGFR inhibitors to resensitize cells expressing these mutants to fulvestrant. FIIN-2 and FIIN-3 are two irreversible covalent pan-FGFR inhibitors that target a cysteine conserved in FGFR1-4 and have exquisite selectivity for some FGFR2 mutations including M538I and K659N (36). In addition, AZD4547 is a selective FGFR inhibitor that was previously shown to inhibit FGFR2 N550K (37).

Both FIIN-2 and FIIN-3 were more effective in inhibiting the downstream signaling (p-FRS2, p-ERK, and p-AKT) induced by FGFR2 N550K as compared with PD173074 and AZD4547 (Fig. 5D), with FIIN-3 being more potent than FIIN-2 (Supplementary Fig. S13). T47D cells stably overexpressing FGFR2 mutant were exquisitely sensitive to FIIN-2 and FIIN-3 as compared with cells expressing GFP or FGFR2 WT (Supplementary Fig. S11B). Although resistance to WT FGFR1/2 and FGFR2 M538I and FGFR2 K660N can be reversed by multiple FGFR inhibitors, for some mutants like FGFR2 N550K, only the irreversible pan-FGFR inhibitors FIIN-2 and FIIN-3 successfully resensitized cells to fulvestrant (Fig. 5E), highlighting the fact that specific resistance mutations might require different strategies to overcome or preempt endocrine resistance.

To examine transcriptional changes associated with FGFR pathway activation, we performed RNA-Seq on T47D cells overexpressing FGFR1, FGFR2 (wild type), FGFR2-activating mutants (M538I, N550K, K660N), or FGF3, as well as on GFP and parental lines as controls (Supplementary Table S9). We combined these profiles with transcriptional profiles we previously generated (11) from of T47D cells overexpressing wild-type and kinase-dead HER2, HER2-activating mutants (S653C, L755S, V777L, and L869R), and ESR1 Y537S, for a total of 15 genetic perturbations examined.

Linear discriminant analysis of these profiles indicated that samples from cells overexpressing FGFR1, wild-type FGFR2, FGFR2-activating mutants, FGF3, wild-type HER2, or HER2-activating mutants are separated from controls (parental cells, GFP expressing, or kinase-dead HER2 D845A) along the first LD component (LD1), as well as from a separate group of cells overexpressing mutant ESR1 along the second LD component (LD2), indicating a common receptor tyrosine kinase (RTK)/growth factor–driven cell state that is distinct from the mutant ER state (Fig. 6A). Along LD1, the cells overexpressing FGF3 have distinct scores from those overexpressing FGFR (P = 2.12 × 10⁻¹⁰, Welch t test), suggesting that in our model system, FGF ligand overexpression is different from FGFR overexpression (Fig. 6A). The overall signature strength of RTKs and growth factor signaling genes (defined in Supplementary Methods and shown in Supplementary Table S10) were highly correlated with scores on LD1 (Spearman ρ = 0.926), suggesting that LD1 broadly represents RTK and growth factor pathway activation.

Using differential expression analysis to compare activating constructs with GFP and parental controls, we defined transcriptional signatures for all FGFR1/2 constructs (FGFR1/2), FGF3, and all HER2-activating mutants (HER2-MUT). Comparison of the top 200 highly expressed genes in each of the FGFR1/2, FGF3, and HER2-MUT signatures (Supplementary Table S11) highlights the high degree of similarity between these groups, with 32 genes common to the FGFR1/2 and FGF3 signatures (OR = 39.63 P = 4.7 × 10⁻³⁷, two-sided Fisher exact test), 84 genes common to the FGFR1/2 and HER2-MUT signatures (OR = 219.4, P = 1.025 × 10⁻¹⁴⁰, two-sided Fisher exact test), and 18 upregulated genes common to all 3 groups (Supplementary Fig. S14A).

We next defined an FGF/R-ACT transcriptional state based on 377 genes that are upregulated in FGF3 and FGFR1/2 cell lines compared with controls (Fig. 6B; Supplementary Table S10). GSEA (38) of the FGF/R-ACT state versus 5,150 previously characterized gene sets (Supplementary Table S12) demonstrated that the shared FGF/R-ACT state is enriched for upregulation of RTK/growth factor receptors signaling, RAS/MAPK signaling, and growth factor–induced ER target genes (Fig. 6C), similar to what we have previously shown for HER2-activating mutants (11). This is consistent with a model in which FGFR/FGF activation leads to ER reprogramming via MAPK pathway activation, potentially shifting the transcriptional spectrum from genes activated by transcription factor AF2 to those activated by AF1. Of note, the genes associated with mTOR signaling were variably upregulated and downregulated in FGF/R-ACT state (Fig. 6C), suggesting only moderate mTOR activation as compared with the robust MAPK activation signal observed. Similar transcriptional signatures were seen when the FGF/R-ACT signature was characterized under fulvestrant treatment, demonstrating that the FGF/R-ACT state is present even when ER signaling is inhibited, consistent with the phenotype of fulvestrant resistance (Supplementary Tables S11 and S12).

Next, we evaluated the effect of 10 different drug combinations on the transcriptional signatures in FGFR/FGF expressing cell lines and control cells profiled by RNA-Seq. The activated RAS/MAPK signature, ER signaling driven by growth factors signature, and FGF/R-ACT signature present in cells overexpressing FGFR/FGF constructs persisted under treatment with fulvestrant, palbociclib, or their combination (Fig. 6D; Supplementary Fig. S14B and S14C). Treatment with FIIN-3, alone or in combination with fulvestrant and/or palbociclib largely reversed the activation of all three signatures (Supplementary Table S13). Effective suppression of all three signatures was also achieved by trametinib plus fulvestrant, with or without palbociclib (Fig. 6D; Supplementary Fig. S14B and S14C; Supplementary Table S13).

Data from the gain-of-function ORF screen, the transcriptomic analysis, and our individual in vitro experiments together suggest that MAPK pathway may represents a common node for drug resistance in ER⁺ MBC. FGFR2 mutants, in particular, rendered cells more sensitive to trametinib than did GFP or WT FGFR2 (Supplementary Fig. S14D), further supporting the finding that FGFR signaling requires the MAPK pathway in this context. Src homology phosphotyrosyl phosphatase 2 (SHP2) is protein downstream of RTKs that is required for RAS activation. Several recent studies have demonstrated that cotargeting SHP2 prevented adaptive resistance to MEK inhibition in multiple RAS-driven cancer models (39–41). Consistent with this, cell lines expressing FGFR2 mutants were hypersensitive to the SHP2 inhibitor SHP099 (Supplementary Fig. S14E). We next examined the ability of combinations of SHP099, trametinib, fulvestrant, and/or palbociclib to overcome resistance in FGFR-expressing cell lines. Similar to trametinib, SHP099 as a single agent partially rescued resistance to fulvestrant and/or palbociclib conferred by FGFR1, and completely rescued resistance conferred by FGFR2. Cells expressing the FGFR2 constructs (wild type or activating mutations) were particularly sensitive to the treatment regimen of trametinib and SHP099 in addition to fulvestrant and palbociclib (Fig. 6E and F), in comparison with minimal inhibitory effect of MEK and SHP2 inhibition in GFP control cells. These results suggest that targeting MEK and SHP2 may serve as an even more effective strategy to overcome multiple forms of FGFR pathway mediated resistance to endocrine therapy and CDK4/6 inhibitors in MBC, as well as, potentially, other RTK-induced mechanisms of resistance.

Taken together, our results from genome-scale gain-of-function screen and genomic profiling of patient samples suggest that activating FGFR pathway alterations are a distinct mechanism of acquired resistance to multiple forms of ER-directed therapy in MBC that can be overcome by FGFR and/or MAPK pathway inhibitors.

Our findings are consistent with two recently published studies. Formisano and colleagues identified FGFR1 amplification as a mechanism of resistance to CDK4/6 inhibitors (42). In this study, FGFR1 knockdown sensitized FGFR1-amplified CAMA-1 cells to fulvestrant/palbociclib, and the addition of FGFR inhibitor erdafitinib to fulvestrant/palbociclib further induced tumor regression in patient-derived xenografts. In another study, Drago and colleagues demonstrated that FGFR1 amplification confers resistance to ER, PI3K, and CDK4/6 inhibitors while retaining TORC sensitivity (43). Our results extend our understanding of endocrine resistance with multiple novel findings, identifying novel pathways and genes associated with endocrine resistance by functional screen, demonstrating the acquisition of multiple FGFR pathway alterations after the development of resistance to endocrine therapy and/or CDK4/6 inhibitors in matched tumor samples, and highlighting therapeutic agents that can overcome this resistance.

Our genome-scale screen provided a comprehensive view into the resistance mechanisms to SERDs. Similar resistance genes were nominated for fulvestrant and GDC-0810, thereby confirming the two drugs have similar mechanism of action. Of note, two ESR1 ORFs conferred resistance specifically to GDC-0810 but not fulvestrant, possibly due to GDC-0810 having a less potent effect on ER degradation than fulvestrant (21, 44). Among the resistance mechanisms shared by fulvestrant and GDC-0810, many are frequently altered in ER⁺ MBC, such as CCNDs/CDK6, KRAS/MAPK, EGFR/ERBB2, and PIK3CA/AKTs/PIMs, and agents targeting those alterations are under clinical development to be combined with endocrine therapy (14, 45). We also identified potential resistance mechanisms that are not characterized to the same extent, such as G protein–coupled receptors, Wnt pathway (FZD10, RSPO1, RSPO3) and Src family kinases (YES1, FYN, FGR), providing clues as to the potential crosstalk between these pathways and ER signaling (46–48) and suggesting that patients with breast cancer harboring functional alterations in these pathways may develop resistance to SERDs. We recognize that those screens were performed using cell lines as model system, and the results will need to be further validated in the clinical setting. We have provided the full genome-scale screen data as a resource to the community of researchers interested in resistance to ER-directed therapies as well as the biology of ER dependencies in ER⁺ breast cancer.

Our ultimate goal is to identify resistance mechanisms that are clinically relevant and can be therapeutically targeted. By comparing paired pretreatment and posttreatment tumors, our evolutionary analyses identified acquired FGFR1 and FGFR2, and FGF3 alterations in 12 of 60 posttreatment samples, further highlighting a potential role for the FGFR pathway in driving drug resistance and disease progression. Most notably, all four alterations in FGFR2 in our cohort were found to be acquired after the development of resistance to endocrine therapy. Our overall findings are consistent with other recent studies which noted some patients with acquired FGFR1 and FGFR2 alterations following treatment of endocrine therapy (42, 43, 49, 50), and provides a mechanistic explanation for these acquisitions.

This analysis was enabled by a novel method we developed to compare the magnitude of amplification in matched pre- and posttreatment samples while considering key confounders to allow for more reliable assignment of copy gain or loss. Because matched tumor samples of the same patient are highly variable in the cancer cell fraction (purity) and often variable in ploidy, we computed the purity-corrected copy number above ploidy and set a relatively stringent threshold of changes in CNAP to define acquired amplification (see Materials and Methods), as cancer clones bearing amplifications with high focality and magnitude in FGFR/FGF genes are more likely to induce dependency on FGFR pathway and result in endocrine resistance.

Our genomic analysis has some limitations and caveats. The observed alterations may not exclusively result from endocrine therapy as some patients received other therapies between the two collected biopsies. Moreover, tumors with FGFR/FGF alterations also harbor alterations in other cancer genes, which may contribute to drug resistance as well (Supplementary Fig. S5; Supplementary Table S5). Despite these caveats, with the evidence from unbiased screens, genomic evidence in relevant patient samples, and confirmatory experimental models, the FGFR pathway clearly emerges as a clinically important resistance mechanism for SERDs and CDK4/6 inhibitors.

Strategies to target the FGFR pathway in breast cancer patients with FGFR alterations are currently being assessed in clinical trials (51–56). The combination of FGFR inhibitors and endocrine therapy is also being clinically investigated (51). As FGFR pathway activation also results in resistance to CDK4/6 inhibitors, a triple combination with the addition of CDK4/6 inhibitors may also be considered. One challenge for the use of FGFR inhibitors is to identify reliable biomarkers. Our results suggest focal and high-level amplifications, clonal activating mutations or high expression levels of FGFR and FGF genes, particularly in the metastatic setting, may be used to guide the clinical use of FGFR inhibitors. Activating alterations in FGFR2, which are rare in primary treatment naïve breast cancer but appear to be clonally acquired in a subset of patients with resistant ER⁺ MBC, may be a particularly good biomarker for the development of FGFR inhibitors.

Our work also highlights that the effective clinical use of FGFR inhibitors needs to consider the variable drug sensitivity of different FGFR2 mutations, which were acquired in some patients following endocrine therapy. The two irreversible pan-FGFR kinase inhibitors, FIIN-2 and FIIN-3, had superior efficacy in targeting all FGFR2 mutants including N550K when compared with other FGFR inhibitors. Both FIIN compounds exhibits good overall kinase selectivity at the concentration of 1 μmol/L, although FIIN-3 can also target wild-type and several mutant EGFR (36). The in vivo efficacy, other off-target effects and toxicity of FIIN compounds still warrant further investigation.

Alterations in FGFR1 and FGFR2 activated the MAPK pathway, and MEK inhibition was able to overcome the resistance conferred by FGFR pathway to some degree. We previously demonstrated acquired ERBB2 mutations resulted in a reprogrammed ER signature and an elevated MAPK transcriptional signature (11). In this study, the transcriptional analysis results suggest that MAPK activation resulting from FGFR/FGF overexpression is more pronounced than mTOR activation. This is consistent with our experimental findings, which demonstrated that MAPK inhibition is particularly effective in cells expressing FGFR2-activating alterations. Furthermore, increased frequency of alterations in MAPK pathway genes have been found in tumors posthormonal therapy, including EGFR, ERBB2, and NF1 (14). The fact that multiple mechanisms of resistance to ER-directed therapies and/or CDK4/6 inhibitors activate the MAPK pathway suggests that this may be an important node of resistance in ER⁺ MBC. Thus, combining endocrine therapy and CDK4/6 inhibitors with agents that target MAPK pathway, such as MEK inhibitors and/or SHP2 inhibitors (57, 58), may be a unifying strategy to overcome or prevent resistance resulting from multiple genetic aberrations that lead to resistance in ER⁺ MBC. Further in vivo validation is needed to establish the translational implication of the combinational strategies proposed herein.

In summary, the integration of a functional genomic screen and genomic analysis of pre- and posttreatment biopsies revealed the FGFR pathway as an important resistance mechanism for endocrine therapy and CDK4/6 inhibitors in ER⁺ breast cancer. With the increasing use of SERDs and CDK4/6 inhibitors in the clinic, we anticipate that the prevalence of FGFR/FGF alterations might increase in the future. Targeting the FGFR pathway with FGFR inhibitors or agents that target downstream MAPK signaling may improve clinical outcomes in patients with aberrations in FGFR/FGF genes. Furthermore, our study highlights the need to sequence metastatic biopsy or blood biopsies at the time of resistance to identify patients with these alterations who may benefit from targeting the FGFR pathway.

P. Mao reports grants from AACR, Stand Up to Cancer, and The V Foundation during the conduct of the study. S.A. Wander reports personal fees from Foundation Medicine (consulting), InfiniteMD (consulting), and Puma Biotechnology (ad board) outside the submitted work. A.G. Waks reports other from Genentech (institutional research funding) and other from MacroGenics (institutional research funding) outside the submitted work. J.H. Chung reports personal fees from Foundation Medicine/Roche (employment, stock) during the conduct of the study. S.S. Freeman reports a patent for methods for determining tumor fraction of cell-free DNA (cfDNA) as well as cfDNA whole exome sequencing (WO2017161175A1), assigned to the Broad Institute and not licensed. O. Rozenblatt-Rosen reports other from Broad Institute (funding to the KCO from the Klarman Family Foundation) during the conduct of the study. V.A. Miller reports other from EQRX (employee) and other from Revolution Medicines (board of directors, stockholder) during the conduct of the study; other from ROCHE/FMI (stockholder) and other from Mirati Therapeutics (stockholder) outside the submitted work; as well as a patent, (Methods and compositions for detecting a drug resistant EGFR mutant) US8501413B2 licensed to MolecularMD, assignee Sloan Kettering Institute for Cancer Research. D.E. Root reports grants from Abbvie, Janssen, Merck, and Vir outside the submitted work. A. Regev reports personal fees from Syros, ThermoFisher Scientific, Celsius Therapeutics, Immunitas, Asimov, and Neogene Therapeutics during the conduct of the study; A. Regev also reports employment with Genentech that began on August 1, 2020. E.P. Winer reports personal fees from Carrick Therapeutics, G1 Therapeutics, Genentech/Roche, Genomic Health, GSK, Jounce, Leap, Lilly, Novartis, Seattle Genetics, and Syros outside the submitted work. N.U. Lin reports grants from Genentech, Merck, and Pfizer, grants and personal fees from Seattle Genetics, personal fees from Daichii Sankyo, Denali, California Institute for Regenerative Medicine, and Puma outside the submitted work. N. Wagle reports grants from Department of Defense, American Association for Cancer Research (AACR), NCI, Susan G. Komen, The V Foundation, The Breast Cancer Alliance, The Cancer Couch Foundation, Twisted Pink, Hope Scarves, Stand Up to Cancer, and National Science foundation (NSF) during the conduct of the study; grants and personal fees from Novartis, other from Foundation Medicine, personal fees from Eli Lilly, grants from Puma Biotechnology, and personal fees and other from Relay Therapeutics outside the submitted work. No potential conflicts of interest were disclosed by the other authors.

P. Mao: Conceptualization, formal analysis, validation, investigation, methodology, writing-original draft, writing-review and editing. O. Cohen: Conceptualization, data curation, software, formal analysis, methodology, writing-original draft, writing-review and editing. K.J. Kowalski: Validation, investigation, methodology. J.G. Kusiel: Validation, investigation, methodology. J.E. Buendia-Buendia: Data curation, formal analysis, methodology. M.S. Cuoco: Methodology. P. Exman: Data curation. S.A. Wander: Data curation, writing-review and editing. A.G. Waks: Data curation. U. Nayar: Resources. J. Chung: Resources, data curation. S. Freeman: Methodology. O. Rozenblatt-Rosen: Resources. V.A. Miller: Resources, data curation. F. Piccioni: Resources, methodology. D.E. Root: Resources, methodology. A. Regev: Formal analysis. E.P. Winer: Supervision. N.U. Lin: Resources, supervision. N. Wagle: Conceptualization, formal analysis, supervision, funding acquisition, writing-original draft, writing-review and editing.

This work was supported by the Department of Defense W81XWH-13-1-0032 (N. Wagle), NCI Breast Cancer SPORE at DF/HCC #P50CA168504 (N. Wagle, N.U. Lin, and E.P. Winer), Susan G. Komen CCR15333343 (N. Wagle), The V Foundation (N. Wagle), The Breast Cancer Alliance (N. Wagle), The Cancer Couch Foundation (N. Wagle), Twisted Pink (N. Wagle), Hope Scarves (N. Wagle), Breast Cancer Research Foundation (N.U. Lin and E.P. Winer), ACT NOW (to Dana-Farber Cancer Institute Breast Oncology Program), Fashion Footwear Association of New York (to Dana-Farber Cancer Institute Breast Oncology Program), Friends of Dana-Farber Cancer Institute (to N.U. Lin), Stand Up to Cancer (N. Wagle), National Science Foundation (N. Wagle), and SU2C-TVF Convergence Scholarship (P. Mao). Research supported by the 2013 Landon Foundation-AACR INNOVATOR Award for Research in Personalized Cancer Medicine, Grant Number 13-60-27-WAGL (N. Wagle). Research supported by the 2017 AACR Basic Cancer Research Fellowship, Grant Number 17-40-01-MAP (P. Mao). We thank Qaren Quartey, Christian Kapstad and Gabriela Johnson for technical assistance; Karla Helvie, Laura Dellostritto, Lori Marini, Nelly Oliver, Shreevidya Periyasamy, Colin Mackichan, Max Lloyd, and Mahmoud Charif for assistance with patient sample collection and annotation; and Flora Luo, Tinghu Zhang and Nathanael Gray for providing reagents. We thank Jorge Gómez Tejeda Zañudo for helpful discussions and comments on the manuscript. We are grateful to all the patients who volunteered for our tumor biopsy protocol and generously provided the tissue analyzed in this study.

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.