Adaptive Resistance to Dual BRAF/MEK Inhibition in BRAF-Driven Tumors through Autocrine FGFR Pathway Activation

Oncogenic mutations affecting the BRAF serine/threonine kinase account for approximately 8% of all solid tumors, most commonly in melanoma but also in colorectal and lung adenocarcinomas, hairy-cell leukemia, and thyroid cancers (1–4). Among the observed mutations, the BRAF^V600Evariant occurs most frequently across tumor types. This mutation promotes oncogenesis through hyperactivation of the MAPK pathway leading to uncontrolled cellular proliferation (5–7). Targeted therapies using inhibitors directed against the mutated BRAF protein have dramatically improved survival. Clinical trials in selected patients using vemurafenib or dabrafenib (BRAF inhibitors) have yielded response rates (RR) in melanoma and non–small cell lung cancer (NSCLC) ranging from 33%–53% (8–10). Unfortunately, resistance to these targeted agents invariably occurs and limits their clinical efficacy over time. Numerous mechanisms of resistance to single-agent BRAF inhibitors have been described, including (i) activation or mutations of alternative growth/survival pathways leading to downstream ERK activation such as oncogenic mutation in NRAS, NF1, COT, or increased PAK signaling (5, 7, 11–13), (ii) mutations bypassing the original BRAF^V600Esuch as splice variants and amplifications in the BRAF gene (14–16), or (iii) paracrine activation of alternative signaling pathways such as the hepatocyte growth factor (HGF; refs. 17, 18). More recently, clinical trials utilizing dual MAPK blockade with the addition of MEK inhibitors have further extended the progression-free survival (PFS) of these patients to a median of 11.4–12.25 months and RR ranging from 67%–76% (19–22). Even in the setting of combination therapy, resistance ultimately emerges, although the mechanisms of resistance to combination MAPK blockade remain less well-characterized (23–27).

Here we attempt to understand the mechanisms underlying resistance to combination BRAF/MEK inhibitors by adopting a pharmacologic synthetic lethal screen in a panel of cancer cell lines that have been rendered resistant through long-term drug exposure. Our results revealed a novel reversible mechanism involving transcriptional feedback upregulation of the growth factor FGF1, which subsequently activates the FGFR cascade resulting in sustained ERK signaling.

Cell lines were obtained from the ATCC. All cells were cultured in RPMI1640, supplemented with 10% FBS, penicillin/streptomycin, and l-glutamine and grown at 37°C in a humidified chamber with 5% CO₂. A375 parental and resistant lines were authenticated by short tandem repeat profiling. Cells were verified to be Mycoplasma free (MycoAlert PLUS, Lonza). Chemical inhibitors were dissolved in DMSO for in vitro studies: vemurafenib (LC Labs), cobimetinib (MedChem Express), ponatinib (LC Labs), NVP-BGJ398 (MedChem Express), PD173074 (Selleckchem), dabrafenib (MedChem Express), and trametinib (LC Labs). A375 cells were treated with vemurafenib, cobimetinib, or the combination starting at IC₅₀ concentration and escalated weekly. Once resistant populations had been established, they were maintained at 2 μmol/L for vemurafenib, 0.5 μmol/L for cobimetinib, and 1.5/0.5 μmol/L for the combination, respectively.

Exon capture was performed using the Agilent SureSelect Human All Exon v5-51 Mb kit and the resulting libraries ran on HiSeq 2000. Average sequencing coverage was 106× with an average of 5.6 Gbase per sample. The raw sequences were aligned to the human reference genome hg19 using the Burrows–Wheeler software. Variant calling was performed using Genome Analysis ToolKit in Galaxy. SNPs were compared with dbSNP, version 135, and In/Dels analyses were performed using PICARD, SAMTOOLS, and GATK. Variants derived from each resistant line were compared with the A375 parental. Sequencing data are available at the European Nucleotide Archive under the accession PRJEB34013.

A total of 1,500–2,000 cells were plated in 96-well plates. Twenty-four hours later, the drug library, along with 1.5 μmol/L and 0.5 μmol/L cobimetinib, was dispensed into prespecified wells using the Agilent Bravo Liquid Handling Platform. Viability was determined 72 hours after drug treatment using CellTiter-Glo (Promega). IC₅₀ was determined using dose–response curve fit in Prism. Each condition was run in triplicates.

Cells were lysed in RIPA buffer (Thermo Fisher Scientific) containing protease inhibitors (Sigma P8340) and phosphatase inhibitor cocktails II and III (Sigma). Protein concentrations were quantitated using Bradford assay (Bio-Rad). SDS-PAGE was used to separate the proteins (Invitrogen NuPAGE 4%–12% Bis-Tris) in MOPS buffer. Resolved protein was transferred to nitrocellulose membranes (Invitrogen), blocked in 5% milk, and probed using the following primary antibodies in either 5% milk or BSA according to the manufacturer's protocol: PathScan Multiplex Western I (Cell Signaling Technology, 5301), PD-L1 (Cell Signaling Technology, 13684), pSTAT3 (Cell Signaling Technology, 9145), pAxl (Cell Signaling Technology, 5724), TAxl (Cell Signaling Technology, 8661), MITF (Cell Signaling Technology, 97800), pCRAF (Cell Signaling Technology, 9427), pPAK1/PAK2 (Cell Signaling Technology, 2606), pFGFR1 (Cell Signaling Technology, 2544), and pFRS2-α (Cell Signaling Technology, 3861).

Cells were seeded at a density of 1 to 2 × 10³cells per well in 96-well plates. Drug treatment commenced 24 hours later at 10 μmol/L with 2-fold decreasing dilutions. After 72–96 hours of drug treatment, cellular proliferation was assessed using CellTiter-Glo (Promega). IC₅₀s were determined using Prism (GraphPad). For colony formation assays, 10,000 cells were seeded in a 6-well plate and incubated with the appropriate drug for approximately two weeks. Media were changed every 4 to 6 days. Cells were then fixed with 4% formaldehyde and stained with 0.5% crystal violet, washed with distilled water, and photographed.

hFGF acidic ELISA was performed according to manufacturer's protocol (Raybiotech ELH-aFGF-1). Serum was diluted 2 fold prior to testing. For conditioned media experiment, 1 × 10⁶cells were plated on a 10-cm plate with 2% FBS in the presence and absence of the BRAF/MEK combination. After 72 hours, media were replaced, collected after an additional 24, 48 hours, and 1 week, and spun down for 10 minutes at 300 RCF. Drug-naïve cells were plated in 96-well plates at a density of 1 to 2 × 10³cells per well with conditioned media. Exogenous growth factors were added at 50 ng/mL to cells in 96-well plates. For the neutralizing assay, a polyclonal antibody (Abcam 9588) to FGF1 was added to the conditioned media at 10 μg/mL and incubated for 30 minutes to 1 hour prior to performing the cell viability assay in 96-well plates described above. A polyclonal IgG isotype control (Abcam 37415) was used in parallel. Cell viability was assayed after 72 hours utilizing CellTiter-Glo.

Xenograft studies were approved by the UCSF Animal Care and Use Committee. A375 parental and A375 VCR cells were injected subcutaneously into both flanks of 6-week-old NCr nude mice (Taconic) at a concentration of 2 × 10⁶/200 μL in Matrigel (354234). Each arm of the study included 5 mice (10 tumors per treatment group). When tumors reached approximately 100 mm³, as determined by an electronic caliper, the animals were randomized depending on the study. For the pilot study, animals were randomized into two groups, each receiving either vehicle or vemurafenib 25 mg/kg and cobimetinib 4 mg/kg dissolved in 5% DMSO, 30% PEG300, and 5% Tween 80 by oral gavage daily. A375 VCR xenografts were randomized into four groups receiving either vehicle, ponatinib 20 mg/kg dissolved in 5% DMSO, 30% PEG300, and 5% Tween 80, vemurafenib 25 mg/kg and cobimetinib 4 mg/kg, or the triple combination via oral gavage daily. Tumors and body weight were measured twice a week.

Tumor samples were harvested postmortem and fixed in formalin for 72 hours and then placed in 70% ethanol for 24 hours prior to embedding in paraffin. Samples were sectioned via microtome at a thickness of 5 μm. IHC was performed by HistoWiz Inc. on a Bond Rx autostainer (Leica) with enzyme treatment (1:1,000) using standard protocols. Antibodies used were pERK (Cell Signaling Technology, 4370 1:1,000) and Ki67 (Leica PA0230). Bond Polymer Refine Detection (Leica) was used according to manufacturer's protocol. Whole-slide scanning (40×) was performed on an Aperio AT2 (Leica). The images were quantified using Halo Image analysis software (Indica Labs) using CytoNuclear module.

Clinical data and tissue collection from patients with BRAF melanoma and NSCLC was performed with informed consent of patients at the Netherland Cancer Institute (Amsterdam, the Netherlands), Massachusetts General Hospital (Boston, MA), or Memorial Sloan Kettering (New York, NY), in accordance with the Institutional Review Boards at these institutions and the US Common Rule. RNA was isolated from FFPE samples as described previously (28, 29). For RNA sequencing, the library was prepared using TruSeq RNA sample prep kit according to the manufacturer's protocol (Illumina). Paired-end 2 × 75 bp sequencing was performed using Illumina HiSeq 2000. RNA-sequencing data are available at http://www.ncbi.nlm.nih.gov under accession number GSE50535 and the European Genome-phenome Archive (EGA S00001000992).

To model mechanisms of acquired resistance of BRAF^V600E-driven tumors to dual BRAF/MEK inhibition, cell lines were generated from A375 human melanoma cells harboring the BRAF^V600Emutation using escalating doses of vemurafenib (A375 VemR), cobimetinib (A375 CobiR), or the combination (A375 VCR; Fig. 1A). Viability without any drug added was normalized to be one-hundred percent. These cells also demonstrate resistance to other BRAF and MEK inhibitors such as dabrafenib and trametinib (Supplementary Fig. S1A). A375-VemR cells were resistant to BRAF inhibitors only but remained sensitive to MEK and ERK inhibitors, suggesting that the resistance mechanism lies at the level of RAF (Fig. 1A). In contrast, A375-CobiR cells exhibited resistance to both BRAF and MEK inhibitors but remained sensitive to ERK inhibition. Therefore, the resistance mechanism likely lies downstream of BRAF but upstream of ERK. The dual resistant A375 VCR cells showed strong resistance to all agents tested, including approximately 100-fold increase in resistance to single-agent vemurafenib or cobimetinib compared with the parental cells and a greater than 10-fold increase in resistance to the combination and to ERK inhibition, suggesting either an on-target mutation of ERK or a bypass track resulting in ERK activation. Both A375-VemR and A375-CobiR remained resistant to the drug even after a prolonged drug holiday. Upon cessation of drug treatment, however, the growth rate of A375 VCR slowed and these cells became resensitized to the drugs after three weeks of drug holiday (Supplementary Fig. S1B). Exome sequencing of the A375 VemR and A375 CobiR cells revealed previously described resistance-associated mutations in NRAS and MEK1, respectively (Supplementary Tables S1–S3; refs. 7, 23). However, no driver mutations or genomic amplifications were identified for the dual resistant A375 VCR line. Immunoblotting of the A375 VCR cells showed elevated pAKT, pSTAT3, and persistent pERK activation despite treatment with BRAF/MEK inhibitors (Fig. 1B). All of these cells exhibited robust MITF expression. pCRAF levels were increased in the resistant lines but contrary to previous reports, pPAK1/2 levels was only mildly increased in this model (12). At baseline, A375 cells express high levels of PD-L1, which is abolished upon inhibition of the MAPK pathway, consistent with prior reports that oncogenic signaling may trigger immune escape by inhibiting antitumor immunity (30). Because pSTAT3 has previously been implicated as a resistance mechanism in several different tumors, we tested whether STAT3 loss of function would resensitize the cells to the combination treatment and gain of function would confer resistance (31). Targeting STAT3 using siRNAs did not alter sensitivity to MAPK blockade in the A375 VCR–resistant cells (Supplementary Fig. S1C and S1D); neither did constitutive overexpression of STAT3 using a lentiviral construct (Supplementary Fig. S1D). Together, these data suggest that resistance of the A375 VCR cells was likely mediated through a STAT3-independent compensatory pathway resulting in sustained ERK or AKT activation.

To identify the pathway(s) mediating resistance in the A375 VCR cells, a pharmacologic screen was conducted using an 86 compound library that targeted kinases, epigenetic regulators, metabolic enzymes, and apoptosis regulators (Supplementary Table S4). Several chemotherapeutic agents and antibody–drug conjugates were also included. Each drug was tested under three conditions: A375 parental, A375 VCR, and A375 VCR in the presence of vemurafenib and cobimetinib. Targets of interest synergized with BRAF/MEK inhibition to selectively enhance killing of the A375 VCR cells while having minimal effect on the viability of parental or resistant cells in the absence of BRAF and MEK inhibitors (Fig. 1C). For each compound tested, the IC₅₀ ratio of the parental line treated with drug X versus the resistant line treated with drug X plus BRAF/MEK inhibitors (IC₅₀ 1) and the ratio of the resistant line treated with drug X versus with dual inhibitors and drug X (IC₅₀ 2) served as the activity readouts of the agent in question to promote cellular killing. Compounds that exhibited low IC₅₀ 1 but high IC₅₀ 2 ratios served as proof-of-concept controls. This class included MAPK inhibitors capable of killing parental A375 cells as a single agent, but not the A375 VCR cells. Molecules that synergized with the BRAF/MEK combination to promote selective killing of the resistant population displayed both high IC₅₀ 1 and IC₅₀ 2 ratios (Fig. 1D). Broadly defined, these agents fall into several pharmacologic categories: PI3K/AKT inhibitors, Src inhibitors, and potential novel targets including CHK1, BTK, and FGFR inhibitors (Supplementary Table S4). Antibody–drug conjugates and chemotherapy differentially enhanced killing of the resistant cells in the presence of the drug combination without affecting the parental cells. Consistent with our prior observations, STAT3 inhibitors did not synergize with dual MAPK inhibition.

Top hits from the screen were retested in a 72-hour viability assay (Fig. 2A; Supplementary Fig. S1E). Cells treated with the relevant drugs were also tested for apoptosis using the Caspase Glo assay. Staurosporine serves as a positive control, and staurosporine-treated cells (1 μmol/L) all exhibit high levels of apoptosis. As expected, vemurafenib and/or cobimetinib induced apoptosis in the A375 parental cells but not in the A375 VCR cells. Addition of ponatinib to the combination induced higher level of cell death in the A375 VCR cells (Supplementary Fig. S1F). Because PI3K/AKT and Src inhibitors have previously been reported to overcome resistance to MAPK inhibition, we focused on the novel targets. Ponatinib, an inhibitor of multiple kinases including Bcr-Abl, PDGFR, VEGFR2, FGFR1, c-Src, and c-Kit, was the highest scoring hit based on both IC₅₀ 1 and 2 that is not an exclusive PI3K/AKT or Src inhibitor. Because multiple downstream effectors, including pSTAT3 and pAKT, appeared to be activated in the A375 VCR cells, we hypothesized that the resistance mechanism is likely mediated through upstream receptor activation and we therefore tested several more specific FGFR inhibitors. Both NVJ-BGJ398 (a FGFR1/2/3 inhibitor) and PD173074 (a selective FGFR1 inhibitor) enhanced A375 VCR tumor killing with the BRAF/MEK inhibitor combination in both short-term viability assays and long-term colony formation studies while having minimal perturbation on the survival of A375 parental and A375 VCR cells in the absence of the BRAF/MEK combination (Fig. 2A and B). In comparison, the selective FGFR4 inhibitor, BLU554, did not augment tumor killing. A second, independently generated dual resistant cell line, Mel888 DR, also exhibited increased sensitivity to ponatinib compared with the parental Mel888 cells (Fig. 2A). This cell line has been previously described as being highly resistant to dabrafenib and trametinib, which was confirmed in our assay. Similar to the observation in A375 VCR cells, both ponatinib and NVJ-BGJ398 increased killing Mel888 DR cells in the presence of dabrafenib and trametinib (Fig. 2B). Taken together, we conclude that the mechanism mediating resistance of the dual resistant population most likely involves compensatory FGFR1 activation.

To elucidate the biochemical consequences of FGFR blockade on the A375 VCR cells, immunoblotting was performed after escalating doses of ponatinib, either as a single agent or in combination with fixed doses of vemurafenib and cobimetinib. With the triple combination, there was a striking decrease in the level of pERK, a finding that was also recapitulated with NVJ-BGJ398 and PD173074 treatments but not with the FGF4-selective inhibitor (Fig. 2C; Supplementary Fig. S2A). Suppression of pERK in the A375 parental and A375 VCR cells at the drug concentrations used occurred by 2 hours, whereas suppression of pFGFR occurred between 4 to 6 hours. Twenty-four hours after drug treatment, pFGFR was completely suppressed, and pERK appeared diminished in the A375 VCR cells even with single-agent ponatinib treatment (Supplementary Fig. S1G). Similarly, treatment with dabrafenib, trametinib, and ponatinib potently suppressed pERK and pFGFR in Mel888 DR cells (Supplementary Fig. S1H). Interestingly, treatment with ponatinib did not alter protein levels of other downstream effectors such as pSTAT3 and pAKT.

We hypothesized that there may be one or more survival factor(s) secreted by the tumor cells in response to drug treatment serving to activate a signaling pathway, as reported previously by other groups (17, 18, 32). To test this hypothesis, conditioned media from both A375 parental and A375 VCR cells pre- and post-drug treatment were harvested. The media were subsequently tested on drug-naïve, parental A375 cells in the presence of escalating doses of vemurafenib and cobimetinib. Post-drug–treated media conferred resistance to the BRAF/MEK inhibitor combination compared with the control, suggesting the presence of soluble factors promoting cellular survival (Fig. 3A). The ability of the conditioned media to confer resistance is “extinguishable” over time after drug withdrawal (Fig. 3A; Supplementary Fig. S2B).

To identity this factor or combination of factors, a targeted growth factor real-time PCR array was utilized to compare differential gene expression pre- and post-drug treatment. In both drug-treated A375 and Mel888 parental and dual resistant A375 VCR and Mel888 DR cells, only a few select FGFs are transcriptionally upregulated in response to dual MAPK inhibition including FGF1, FGF7, FGF14, and FGF17 (Fig. 3B). The landscape of upregulated FGF factors is similar across both cell lines. Addition of each of these factors to the drug-naïve A375 parental cells treated with the BRAF/MEK inhibitor combination showed that only FGF1 was able to confer significant resistance, by over 10-fold, whereas FGF7 and FGF17 did not confer any appreciable resistance (Fig. 3C). Furthermore, exogenous treatment with FGF1 resulted in increased level of pFGFR1 and signaling through this pathway (Fig. 3C). Consistent with this, the media showed a 3-fold increase in FGF1 levels after treatment with the drug combination compared with the pretreatment control (Fig. 3D). In contrast, there was no increase in the A375 CobiR and a less prominent increase in the A375 VemR samples (Supplementary Fig. S2C), likely due to the fact that these cells are driven by acquired mutations rather than a nongenetic, compensatory signaling mechanism. Depletion of FGF1 using a neutralizing antibody diminished the ability of the conditioned media from drug-treated cells to promote resistance in drug-naïve cells (Fig. 3E). We extended this analysis to a larger panel of cells lines harboring the BRAF^V600Emutation, including HCC364, a lung cancer cell line dependent on BRAF. In most of the cell lines where FGF1 was able to promote drug resistance (MALME-3M, LOX-IMVI, 1205-LU, HCC364, and UACC-62), FGF1 RNA was increased by at least 3-fold in response to dual vemurafenib and cobimetinib treatment for 72 hours (Fig. 4A and B). There was a concomitant increase in FGF1 levels detectable in the media within 72 hours post-drug treatment (Fig. 4C), suggesting that this phenomenon might be broadly applicable to subsets of BRAF-driven tumors. Conversely, cell lines in which resistance was not induced by addition of FGF1 in response to BRAF/MEK inhibition did not demonstrate any appreciable change in FGF1 levels in the media from drug- treated cells (SK-MEL 28 and WM-164; Fig. 4B and C). Furthermore, upregulation of FGFR is specifically observed in the dual resistant A375 VCR cells (Fig. 4D). These cells appear to rely primarily on FGFR, the insulin receptor, and to a lesser extent, PDGFR, although one cannot definitely conclude whether this is due to hyperactivation of parallel signaling pathways within the same clone or coexistence of multiple clones, each with unique RTK dependency. In contrast, the A375 parental, VemR, and CobiR exhibit elevated EGFR and HGFR/c-Met signaling as reported previously (refs. 17, 18, 33). Taken together, these results indicate that tumor cells transcriptionally upregulate prosurvival mediators, such as FGF1, after exposure to drug, and these factors subsequently enhance tumor survival in an autocrine manner through activation of downstream signaling pathways.

To establish the physiologic relevance of our findings, xenograft models were established using A375 parental and VCR lines. Pilot experiments wherein A375 parental xenografts were treated with vemurafenib 25 mg/kg and cobimetinib 4 mg/kg daily via oral gavage resulted in significant decrease in tumor volume (Supplementary Fig. S3A). Toxicity was minimal as assessed by body weight (Supplementary Fig. S3B). IHC on tumor sections verified decreased pERK expression (Supplementary Fig. S3C). In a four-arm study, the A375 VCR xenografts were treated with vehicle, ponatinib at 20 mg/kg, dual vemurafenib and cobimetinib (25 mg/kg and 4 mg/kg respectively), or the triple combination including ponatinib. The triple combination arm showed significant tumor regression compared with the other three arms (Fig. 5A) with virtually no difference in body weight alterations over time (Supplementary Fig. S3D). Similarly, Mel888DR xenografts treated with the triple combination had significantly smaller tumor volume compared with those treated with vemurafenib and cobimetinib (25 mg/kg and 4 mg/kg, respectively; Fig. 5G). Tumors in the dual combination arm grew after approximately 14 days, although interestingly, tumors in the vehicle and single-agent ponatinib-treated animals remained static, which may be related to oncogene-induced senescence upon withdrawal of the inhibitors as reported previously (28). The animals in the vemurafenib and cobimetinib arm were subsequently rerandomized to either continuing on the combination or to addition of ponatinib (Fig. 5B). Results indicated that sequential addition of ponatinib reduced tumor volume compared with those animals that continued on the dual combination therapy with a slight decrement in weight, although this observation can be attributed at least, in part, due to the rapidly enlarging tumor volume in the dual combination animals rather than treatment induced cachexia (Supplementary Fig. S3E). In addition to inducing tumor regression in the dual resistant A375 VCR cells, addition of FGFR blockade upfront in drug-naïve parental cells delayed the emergence of resistance without evidence of toxicity both in vitro and in xenograft models (Fig. 5C and D; Supplementary Fig. S3F). Consistent with our results in vitro, the triple combination inhibited pERK and exhibited decreased Ki-67 expression (a marker of cellular proliferation) in IHC studies compared with the other arms (Fig. 5E). Serum level of FGF1 harvested from A375 parental xenografts was also higher after drug treatment (Fig. 5F).

To determine whether any of the FGF genes identified by our functional approach might promote clinical resistance to MAPK inhibitors, we obtained RNA-sequencing data from paired biopsies of melanoma and patients with NSCLC, two tumor types where BRAF mutations occur at an appreciable frequency (35%–50% and 5%–8% respectively). All patients underwent a biopsy prior to treatment initiation and then either while on treatment or at the time of disease progression. Among the 4 patients with NSCLC, all of whom received the BRAF/MEK combination, two demonstrated an increase in FGF1 expression at progression, one without change, and one showed a decrease in FGF1 levels (Fig. 6A). Progression biopsy harvested from patient one, who had been treated with dabrafenib and trametinib and exhibited elevated FGF1 levels, was used to generate a patient-derived xenograft (PDX) model. These animals were subsequently treated with vemurafenib and cobimetinib or the triple combination including ponantinib. The triple combination arm exhibited reduced tumor growth compared with the vemurafenib and cobimentinib combination (Fig. 6B).

Out of 20 patients with melanoma, 9 received a single-agent BRAF inhibitor while 11 were treated with the BRAF/MEK inhibitor combination. Only 1 out of 9 patients (22%) treated with a BRAF inhibitor demonstrated an increase of FGF1 levels posttreatment by more than 2-fold. In contrast, 6 of 11 patients (55%) who received the combination had post-treatment FGF1 levels of greater than 2-fold (Fig. 6C). The mean overall survival was markedly worse among the 7 patients whose posttreatment FGF1 increased by more than 2-fold, regardless of the treatment received, compared with those who did not (15.27 months vs. 71.44 months, P = 0.037). Furthermore, all 4 patients who were alive at the time of censoring were FGF1 low (Fig. 6D).

Implementing a pharmacologic synthetic lethal screen using a reversibly resistant BRAF^V600E-driven cell line, we identified a novel mechanism underlying drug resistance to combination BRAF/MEK therapies, providing insight into a signaling network rewiring in the dual resistant population. Upon initial cytotoxic insult with the drug combination, the majority of treatment naive cells will undergo apoptosis. However, a small fraction of the tumor cells capable of transcriptionally upregulating growth factors such as FGF1 may survive the initial drug onslaught, possibly reflecting heterogeneity of chromatin states among the cellular population. This subset of therapy-induced persister cells in vitro appears to recapitulate the state of residual disease in vivo that resumes proliferation over time leading to disease recurrence either through the development of secondary mutations or, as in our model, by coopting the transcriptional activation of growth factors to ensure its long-term survival. In our system, both the parental A375 drug–treated and A375 VCR cells exhibited higher FGF1 levels relative to the parental nondrug-treated cells. After prolonged exposure to the drug combination, however, the A375 VCR cells showed lower FGF1 expression compared with the parental A375 cells treated acutely with the drug combination, suggesting a potential feedback mechanism of FGF1 regulation that remains to be identified.

Our data suggest that FGF1 activation is an important mediator of cellular survival in a subset of BRAF^V600E-driven melanomas and NSCLCs treated with the BRAF/MEK inhibitor combination. To this end, abrogation of this feedback response with the addition of FGFR-directed therapy upfront may decrease the number of persister cells and ultimately delay the emergence of resistant clones. The concept of an autocrine circuit involving a secreted factor that subsequently activates a prosurvival receptor tyrosine kinase pathway has been described in other tumor resistance models including AML, and EGFR- and ALK-driven NSCLC (32, 34). Furthermore, our findings substantiate prior reports whereby FGFR3 activation confers resistance to vemurafenib in vitro (35) and that feedback activation of FGFR1 served as an adaptive resistance mechanism in KRAS-mutated lung cancers after MEK inhibition (36), implicating the FGFR feedback loop as a broad resistance mechanism in various tumor types. Consistent with our findings, MEK inhibition increased FRS2, pERK, and pAKT levels but addition of ponatinib blunted these responses. However, FGFR inhibition is unlikely to be successful in BRAF- or KRAS-driven colorectal cancer models because these tumors appear to rely largely on EGFR activation as the prominent feedback mechanism, demonstrating important tumor-specific differences (37). A recent report has shown increased PAK signaling as an important response to MAPK inhibitors (12). While we did not observe substantial increase of pPAK in our dual resistant population (Fig. 1B), one of the cell lines that demonstrated high levels of PAK expression in response to MAPK inhibitors was WM-164 which, interestingly, did not respond to exogenous stimulation by FGF1 (Supplementary Fig. S2D). This may represent an example of tumor heterogeneity whereby tumor cells evolve parallel pathways to circumvent resistance. Understanding and predicting the feedback mechanism utilized by the tumor cells would be highly informative for precision deployment of rational drug combinations.

Even after prolonged exposure to the BRAF/MEK combination, no acquired resistance mutations were detected in the dual resistant population. In comparison, acquired resistance mutations in NRAS and MEK1 emerged relatively rapidly in each of the single treatment arms, likely because the combination treatment presented a higher fitness threshold. Because of the fitness disadvantage conferred by combination therapy, subclones with acquired mutations never propagate sufficiently to reach a detectable level within the tumor population (38). As a result, these tumor cells may preferentially hyperactivate a compensatory growth signaling pathway to surpass this fitness threshold. This observation is supported by our clinical data, because patients who received dual BRAF and MEK inhibitors exhibit a strikingly higher frequency of upregulated FGF1 expression compared with those who received single-agent therapy. Furthermore, patients who displayed increased intratumoral FGF1 mRNA levels in response to MAPK inhibition had a worse overall survival. This result, coupled with elevated serum FGF1 levels detected after dual MAPK inhibition, raises the possibility that FGF1 may serve as a prognostic marker for disease survival in BRAF-driven melanomas and as a predictive marker identifying the subset of patients who may benefit from the addition of a FGFR inhibitor. It is also interesting to note that ponatinib has a lower IC₅₀ compared with the more specific FGFR inhibitors. This may be due to the more promiscuous nature of ponatinib and its additional inhibitory activity on PDGFR and Src. Taken together, our data suggest that (i) clinical deployment of FGFR-directed therapy may be a strategy to resensitize a broad range of resistant tumor cells after MAPK blockade, and that (ii) inhibiting multiple signaling pathways using low-dose combination therapies upfront may augment the killing of tumor persister cells and prevent the emergence of a resistant population by targeting multiple subclones present at low frequencies.

Clinical deployment of the rational combination to include a FGFR inhibitor in addition to dual BRAF and MEK inhibitors in phase I clinical trials is warranted in BRAF^V600E-driven melanoma and NSCLC to test for safety and efficacy. Future work should be directed to the prediction of feedback mechanisms likely to be utilized by resistant tumors, and an understanding of why certain pathways are preferentially selected in patients over others, with the goal of improving rational drug combinations. Additional studies are also needed to understand how the autocrine secretome contributes to the establishment and maintenance of a tumorigenic niche, as well as its effect on tumor stromal cells and the immune microenvironment.

D.P. Carbone is an employee/paid consultant for Janssen, Kyowa Kirin, Loxo Oncology, Merck, Merck Sharp Dohme, Nexus Oncology, Novartis, Pfizer, Takeda Oncology and Teva, and reports receiving commercial research grants from Bristol-Myers Squibb. J. Settleman is an employee/paid consultant for Pfizer. F. McCormick is an employee/paid consultant for, reports receiving commercial research grants from and holds ownership interest (including patents) in BridgeBio. K.T. Flaherty reports receiving commercial research grants from Novartis and Sanofi, holds ownership interest (including patents) in Clovis Oncology, Strata Oncology, Vivid Biosciences, Checkmate Pharmaceuticals, X4 Pharmaceuticals, PIC Pharmaceuticals, Fount Therapeutics, Shattuck Labs, Apricity, Oncoceutics, Fog Pharma, Tvardi, xCures, and Vibliome, and is an advisory board member/unpaid consultant for Sanofi, Amgen, Asana Biosciences, Adaptimmune, Aeglea, Tolero Pharmaceuticals, Neon Therapeutics, Monopteros, Novartis, Genentech, Bristol-Myers Squibb, Merck, Takeda, Verastem, Boston Biomedical, Pierre Fabre, and Debiopharm. No potential conflicts of interest were disclosed by the other authors.

The Editor-in-Chief of Clinical Cancer Research is an author on this article. In keeping with AACR editorial policy, a senior member of the Clinical Cancer Research editorial team managed the consideration process for this submission and independently rendered the final decision concerning acceptability.

Conception and design: V.E. Wang, R. Bernards, J. Settleman, F. McCormick, K.T. Flaherty

Development of methodology: V.E. Wang

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): V.E. Wang, J.Y. Xue, D.T. Frederick, C. Wilson, A. Urisman, P. Lito

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): V.E. Wang, D.T. Frederick, Y. Cao, E. Lin, A. Urisman, D.P. Carbone, R. Bernards, P. Lito, J. Settleman, F. McCormick, K.T. Flaherty

Writing, review, and/or revision of the manuscript: V.E. Wang, Y. Cao, A. Urisman, D.P. Carbone, J. Settleman, F. McCormick, K.T. Flaherty

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): V.E. Wang, D.T. Frederick, E. Lin

Study supervision: J. Settleman, F. McCormick

This research was supported by a Damon Runyon Postdoctoral Award (to V.E. Wang, 121570), an ASCO Young Investigator Award (to V.E. Wang, P0511381), a Lung Cancer Research Foundation Fellowship (to V.E. Wang, LC170486), and the Department of Defense (W81XWH-17-1-0365, W81XWH-18-1-0551, to V.E. Wang). P. Lito is supported in part by the NIH/NCI (1R01CA23074501, 1R01CA23026701A1) and The Pew Charitable Trusts. We thank members of the McCormick laboratory for helpful discussions and input and Andrew Wolfe for critical reading of the manuscript.

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.