The Co-mutational Spectrum Determines the Therapeutic Response in Murine FGFR2 Fusion-Driven Cholangiocarcinoma

Animal Experiments

All animal experiments were approved by local authorities (the Lower Saxony State Office for Consumer Protection and Food Safety). The B6.12954 Krastm4Tyj (Kras^lslG12D/wt) strain was a gift from Albrecht Neesse (Georg-August University, Göttingen, Germany) and maintained at the local animal facility of Hannover Medical School. NOD.Cg-Prkdcscid Il2rgtm1Wjl/SzJ (NSG) mice were bred at and purchased from the local animal facility. Seven-week to 8-week-old male and female animals were used for all experiments. Unless otherwise stated, experimental animals were fed a standard diet (1324 M; Altromin, Lage, Germany) and maintained in individually ventilated cages with access to food and water ad libitum. Interventions were done during the 14-hour day cycle.

To generate primary tumors, liver electroporations were performed as previously described⁽¹¹⁾ (and detailed in the Supporting Methods). For in vivo drug treatments, animals bearing subcutaneous tumors were randomized, and treatments were started when tumor volumes reached approximately 100 mm³. Tumor growth was followed by digital caliper measurements and calculated as 0.5 × length × width² (length > width). Experimental animals were euthanized when tumor volumes reached 1,200 mm³.

The following compounds and concentrations were used for in vivo experiments: BGJ398 (30 mg/kg, once a day, by oral gavage, resuspended in polyethylene gycol 400/0.5% Tween80/5% propylene glycol), gemcitabine (100 mg/kg, twice per week, intraperitoneal delivery, diluted in NaCl 0.9%), and trametinib (1 mg/kg, once a day, by oral gavage). Doxycycline (Dox)–containing food (625 mg/kg) was purchased from Envigo Teklad (Horst, Netherlands).

Short Hairpin RNA Transduction and Depletion Assays

To enable tet-regulatable gene expression, the FPK (FGFR2-PPHLN1-Kras^G12D) and FAK (FGFR2-AHCYL1-Kras^G12D) cell lines were transduced with reverse tetracycline-transactivator (rtTA3)–encoding lentiviral particles (pRRL.SFFV-rtTA3-IRES-EcoR-PGK-Puro), kindly provided by Johannes Zuber, IMP, Vienna). Puromycin-selected cells were then subjected to a second round of transduction with short hairpin RNA (shRNA)–encoding retroviruses: 48 hours after polyethylenimine transfection with 7 µg of vector DNA (TRIN_miRE_shKras.247, TRIN_miRE_shKras.368, or TRIN_miRE_shRenilla.713; 3 µg polyethylenimine per µg DNA), viral supernatant was collected from Platinum E cells (Cell Biolabs, San Diego, CA) and supplemented with 4 µg/mL polybrene (Sigma-Aldrich, Milwaukee, WI). Cells were neomycin-selected to 90%-95% Venus-positive cells. For depletion assays, Dox (Sigma-Aldrich, St. Louis, MO) was added to the cell culture medium at a final concentration of 1 µg/mL, and cells were cultured in the absence or presence of the FGFR inhibitor BGJ398 (2 µM). The shRNA-expressing population (Venus+/dsRed2+) was followed longitudinally by flow cytometry.

FGFR2 Fusions Drive Malignant Transformation In Vitro and In Vivo

We used a soft-agar assay to test the colony formation capacity of six different FGFR2 fusions (Fig. 1A). All fusions enhanced anchorage-independent growth compared with empty vector control (Fig. 1B,C). To further evaluate whether the increased colony formation potential observed in vitro translates into accelerated tumor development in vivo, we used an orthotopic ICC liver electroporation model.⁽¹¹⁾ Based on an early study that reported a frequent co-occurrence of FGFR2 fusions and KRAS (Kirsten rat sarcoma oncogene) mutations in ICC,⁽⁸⁾ we electroporated transposon constructs encoding for the respective fusion complementary DNAs (cDNAs), a positive control cDNA (Yes1 associated transcriptional regulator [Yap1]) or a negative control cDNA (green fluorescent protein), under the control of a phosphoglycerate kinase (PGK) promoter into the livers of transgenic mice harboring the lox-stop-lox-KrasG12D allele (Kras^lslG12D/wt).⁽¹²⁾ In contrast to overexpressed mutant KRAS, which led to rapid tumor development in the original liver electroporation ICC model,⁽¹¹⁾ the transgenic allele that is transcribed from the endogenous locus achieves more “physiologic” levels of oncogene expression. The latent allele was activated by the co-electroporated and transiently expressed Cre recombinase, and Trp53 (tumor protein p53) was disrupted by CRISPR/Cas9 gene editing⁽¹³⁾ (Fig. 2A,B).

Mice that received the plasmid mix containing a fusion cDNA developed liver tumors with full penetrance and had to be sacrificed with a mOS of 3 to 4 months following electroporation (Fig. 2C). PCR-based amplification of a region spanning the respective fusion “breakpoints” on genomic DNA extracted from the tumors confirmed stable integration of the fusion cDNA in all tested samples (Supporting Fig. S1A).

Although mice that received the positive control cDNA (Yap1) also rapidly developed tumors and had to be sacrificed between 9 and 10 weeks after electroporation (Supporting Fig. S1B), no tumor growth was detected in any of the 15 control-electroporated mice within the observational timeframe of up to 6.5 months (Supporting Fig. S1B). The unexpected observation that the activation of the latent mutant Kras^G12D allele in combination with single-guide RNA (sgRNA)–mediated disruption of Trp53 is not sufficient to induce tumors in the adult murine liver indicates that the FGFR2 fusions function as the relevant oncogenic drivers in this model. Histologically, tumors predominantly presented as poorly differentiated adenocarcinomas with regions of moderate differentiation that displayed cytokeratin 19–positive ductal structures surrounded by a dense tumor stroma, overall reminiscent of human ICC (Fig. 2D). Three of 22 evaluated tumors showed features of poorly differentiated HCC.

For further in vitro studies, we derived cell lines from a subset of FGFR2-driven tumors (FGFR2–periphilin 1 [PPHLN1] and FGFR2–adenosylhomocysteinase like 1 [AHCYL1], denoted here as FPK and FAK cells). Efficient CRISPR/Cas9-mediated editing of Trp53 was confirmed by surveyor assay and immunoblotting (Supporting Fig. S1C). Despite extensive efforts to validate expression of the fusion proteins by western blot or immunohistochemistry, we were not able to conclusively detect the chimeric fusion proteins using different commercially available antibodies designed to recognize an epitope in the 5’ region of FGFR2 (data not shown). Therefore, to provide proof of active transcription, we determined FGFR2 levels by semiquantitative PCR on reverse-transcribed cDNA from fusion versus nonfusion cell lines. For nonfusion controls, we used a cell line that we derived from a Yap1-driven control tumor, in addition to a previously published murine KRAS-driven cholangiocarcinoma cell line.⁽¹⁴⁾ Primers were designed to detect both the fusion and the endogenous murine wild-type (WT) FGFR2 cDNA (Fig. 3A). The increased band intensity indicated elevated mRNA expression in the fusion lines compared with endogenous FGFR2 levels in the nonfusion controls. In addition, liquid chromatography–mass spectrometry analysis confirmed the presence of the fusions on the protein level in crude cell lysates from FPK and FAK cell lines (Supporting Table S1). Further functional proof of FGFR2 activity was provided by detection of enhanced levels of canonical FGFR2 downstream targets, such as phosphorylated fibroblast growth factor receptor substrate 2 (pFRS2) and phosphorylated SH2 domain–containing protein tyrosine phosphatase 2 (pSHP2) in fusion-positive versus fusion-negative cell lines (Fig. 3B).

FGFR2 Fusion-Positive Tumor Cell Lines Harboring a Mutant Kras Allele Are Resistant to FGFR Inhibitors Despite Efficient Interruption of Downstream Signaling

Based on the observation that all mice injected with the FGFR2 fusion plasmid developed tumors, whereas no tumor development was observed in the negative control mice, we assumed that FGFR2 acts as the oncogenic driver in our model. Therefore, the respective tumor-derived cell lines should be genuinely sensitive to treatment with FGFR inhibitors.

As expected, treatment with BGJ398 (infigratinib), a clinically advanced FGFR inhibitor with positive phase 2 data for patients with ICC,⁽¹⁵⁾ resulted in decreased levels of phosphorylated extracellular signal-regulated kinase (pERK), phosphorylated mitogen-activated protein kinase (pMEK), pSHP2, and pFRS2, specifically in the fusion-positive cell lines. The minor impact of the drug on phosphorylated AKT serine/threonine kinase 1 is in line with previous reports, that phosphatidylinositol 3-kinase (PI3K) signaling is only marginally affected by FGFR inhibitors (Fig. 3B).⁽¹⁶⁾ Surprisingly, although pharmacologic FGFR inhibition efficiently disrupted FGFR-downstream signaling in the presence of mutant Kras, half-maximal inhibitory concentrations (IC₅₀) of four different FGFR inhibitors were unexpectedly high in the fusion cell lines, and without striking or consistent differences between the fusion and nonfusion cells (Fig. 3C and Supporting Table S2). In contrast, an FGFR2-PPHLN1-driven cell line derived from a tumor that was engineered in a Kras WT littermate control (termed FP; Supporting Fig. S2A,B) was decisively more responsive to BGJ398 (Fig. 3D) and TAS120 (Supporting Table S2), but was rendered resistant through the introduction of alternative RAS signaling activators (Supporting Fig. S2C).

Therefore, despite the observation that Kras^G12D was not sufficient to drive malignant transformation in our model, we reasoned that the co-mutant Kras allele was the cause of primary resistance to the FGFR inhibitors.

KRAS Inhibition Re-sensitizes FGFR2 Fusion–Positive, KRAS Mutant Cells to FGFR Inhibition

Thus far, no clinically developed direct KRAS^G12D inhibitor exists. KRAS requires the correct subcellular localization to exert its oncogenic activity, which is in part regulated by the prenyl-binding protein PDEδ. Therefore, to address whether targeting KRAS signaling together with FGFR inhibition acts synergistically in our FGFR2 fusion model, we performed an in vitro two-drug combination assay using BGJ398 together with the preclinical small molecule KRAS-PDEδ inhibitor deltarasin.⁽¹⁷⁾ Cells were cotreated with BGJ398 and deltarasin at different concentrations, and CompuSyn was used to calculate the combination index (CI) and fraction affected (Fa) for each combination drug value. Most of the CI values were within the synergistic range (Fig. 4A and Supporting Fig. S3A).

As a small molecule inhibitor, deltarasin does not exclusively affect KRAS localization, but also binds to other G-protein-coupled receptors, ion channels, and transporters. Off-target effects can be promoted by the high drug concentrations that are necessary to suppress residual PDEδ activity, which can be sufficient to sustain oncogenic RAS signaling.⁽¹⁸⁾ To rule out that the synergistic effects observed in the drug assay were due to off-target activity, we chose to recapitulate our findings by using a genetic approach. We cloned shRNAs targeting Kras, or a neutral control (shRenilla), into the TRE-dsRed2-miRE/shRNA-PGK-Venus-IRES-NeoR vector system.⁽¹⁹⁾ This neomycin-selectable tet-on vector system contains fluorescent reporters that enable longitudinal tracking of cells with potent target knockdown: Transduced cells constitutively express the fluorescent marker Venus, and Dox induces activation of the tetracycline-responsive element that drives shRNA expression, linked to the red fluorescent marker dsRed. Following insertion of an rtTA cassette, we stably transduced the FPK and the FAK cell line with the inducible shRNA constructs. In addition, to control for the functional efficacy of KRAS knockdown, we included a previously characterized murine cancer cell line (KRPC) that depends on mutant Kras^G12D as an oncogenic driver, and was rendered tet-on competent by integration of an rtTA cassette.⁽²⁰⁾ To rule out RNA interference (RNAi)–induced off-target effects, two separate shRNAs targeting Kras were used throughout the experiment, and efficient KRAS knockdown was confirmed by western blot on lysates from Dox-treated cells after neomycin selection (Supporting Fig. S3B).

The modified cell lines were cultured in the presence of either Dox only to induce KRAS knockdown, or Dox plus BGJ398 for KRAS knockdown in the presence of an FGFR inhibitor. The relative percentage of the Venus/dsRed double-positive, shRNA-expressing population was followed by flow cytometry (Supporting Fig. S3C) for up to 12 days.

In all cell lines transduced with the neutral control shRNA, the proportion of hairpin-expressing cells remained constant over the course of the experiment, independent of the presence or absence of the FGFR inhibitor. Notably, FPK_shKras and FAK_shKras cells were not depleted following induction of KRAS knockdown alone, demonstrating that proliferation of these cells does not critically depend on signaling through mutant KRAS (Fig. 4B and Supporting Fig. S3D, green curve). KRAS-driven KRPC cells (without FGFR2 fusion) were rapidly depleted after Dox addition, thus providing proof for the functional activity of the Kras hairpins (Supporting Fig. S3E).

In contrast, when the shKras-transduced FPK and FAK cells were cultured in the presence of the FGFR inhibitor after induction of KRAS knockdown, the hairpin-expressing population rapidly decreased (Fig. 4B and Supporting Fig. S3D, red curves), while the addition of the FGFR inhibitor did not influence the depletion kinetics observed in KRPC cells (Supporting Fig. S3E).

The superior activity of the FGFR inhibitor in the presence of KRAS knockdown in FPK_shKras and FPK_shRenilla cells was further corroborated in a clonogenic growth assay. Of note, the results of this assay likely underestimate the consequence of KRAS knockdown in the BGJ-treated cells, as the quantification of “bulk” cell densities does not control for the approximate 30% of cells that fail to efficiently express the shRNA (Fig. 4C).

Together, these data support the view that the FGFR2-driven, Kras mutant cell lines are not addicted to the Kras co-mutation, and that the superior depletion in the presence of KRAS knockdown is caused by synergistic effects.

In Vivo RNAi Confirms That KRAS Knockdown in FGFR2 Fusion-Positive, KRAS Mutant Tumors Delays Tumor Growth Only in the Presence of an FGFR Inhibitor

Next, we set out to confirm these results in an in vivo environment. First, we addressed the therapeutic consequences of either monotherapy with gemcitabine as part of the standard chemotherapy regiment in patients with ICC, or treatment with BGJ398. Compared with vehicle-treated controls, gemcitabine treatment significantly reduced tumor growth in mice injected subcutaneously with FPK cells. Treatment with BGJ398 resulted in a further delay in tumor growth, suggestive of a superior performance in vivo compared with its in vitro efficacy (further discussed subsequently) (Fig. 5A). In a second experiment, we injected mice with the FPK_shKras.247 cell line. Following tumor growth, mice were randomized to receive vehicle, vehicle plus Dox (to induce KRAS knockdown), BGJ398, or BGJ398 plus Dox. Recapitulating our in vitro observations, KRAS knockdown alone did not affect tumor growth compared with vehicle treatment, whereas KRAS knockdown in the BGJ-treated cohort led to superior inhibition of tumor growth compared with mice that received only BGJ398 (Fig. 5B). Histologically, the superior response to BGJ398 in the presence of KRAS knockdown was accompanied by a significant reduction in Ki67 expression (Fig. 5C).

Resistance to BGJ398 in FPK Cells Is Mediated Through the MEK/ERK Axis

Compounds targeting crucial KRAS-signaling hubs have largely failed as monotherapy, in part due to the activation of strong feedback mechanisms that are capable of rapidly overruling the effects of the respective inhibitors. Two downstream pathways with pivotal importance for KRAS oncogenic signaling are the MEK/extracellular signal-regulated kinase (ERK) cascade and the PI3K cascade. To assess whether pharmacologic blockage of one or the other pathway could enhance the effect of FGFR inhibition, we treated our original FPK cell line with the PI3K inhibitor buparlisib, or the MEK inhibitor trametinib, at their respective approximate IC₂₀-IC₃₀ concentrations (Supporting Fig. S2D) and with or without concurrent BGJ398 treatment. Clonogenic growth assays on single-treated versus combination-treated cells showed that cotreatment with BGJ398 and the MEK inhibitor outperformed the effect of the FGFR inhibitor plus the PI3K inhibitor (Fig. 6A). Consistently, administration of BGJ398 in combination with trametinib in vivo resulted in a dramatic therapy response in mice bearing subcutaneous FPK tumors (Fig. 6B). This provides evidence that the MEK/ERK axis acts as the crucial downstream effector of FGFR-inhibitor resistance in our KRAS mutant cells.

Rapid activation of feedback loops can severely compromise the efficacy of targeted therapies.⁽²⁰⁾ The MAPK pathway is activated by KRAS, which is also a downstream target of FGFR2, resulting in reduced pERK levels in FPK and FAK cells treated with BGJ398 (Fig. 3B). To address the mechanistic underpinnings of the FGFR-inhibitor resistance, and its relief following KRAS knockdown or MEK inhibitor treatment, we longitudinally monitored pERK expression levels in FPK cells treated with either BGJ398 or trametinib. Both BGJ398 and trametinib as monotherapies led to pERK rebound by 12 hours after addition of the drug. In contrast, in cells that were treated with both the FGFR inhibitor and the MEK inhibitor, pERK re-expression was decisively delayed (Fig. 6C).

Together, despite the genuine lack of sensitivity of FPK and FAK cell lines toward KRAS knockdown, pERK rebound following FGFR inhibition can be suppressed by MEK inhibition. This observation suggests that targeted inhibition of FGFR signaling permits de-repression or activation of alternative pathways and creates a “secondary” dependency on KRAS signaling, which is overcome by combinatorial treatment. Of note, despite the high in vitro IC₅₀, we observed a significant delay in tumor growth of FPK cells treated with an FGFR inhibitor in vivo, although further inhibition could be achieved by concomitant KRAS knockdown. Under standard cell culture conditions, high concentrations of growth factors are present in fetal calf serum-supplemented medium. Therefore, pro-proliferative cues that are transmitted from the microenvironment to the tumor cells, such as after relief of negative feedback loops, are likely most effective in the presence of intense external stimuli.

In summary, these results indicate that the co-mutational spectrum of the FGFR2 fusion–driven ICC cell lines decisively influences response to pharmacologic FGFR inhibition, and that the sensitivity to FGFR inhibitors can be restored by implementing co-targeting strategies.

To determine whether human FGFR2 fusion patients are a molecularly “homogeneous” or rather “heterogeneous” cohort, we performed unsupervised clustering of bulk transcriptome data from local patients with ICC (n = 162). FGFR2 fusion–positive cases were broadly distributed across the clusters, which indicated that the presence of the FGFR2 fusion is not the overarching determinant of the molecular profile (Supporting Fig. S4A). Second, we asked whether a subset of the FGFR2 patients displayed a profile similar to the KRAS mutant cases. By comparing 15 KRAS codon 12 or codon 13 mutant ICCs to 29 FGFR2 fusion–positive cases, we determined 258 differentially regulated genes (Supporting Table S3). Unsupervised clustering on the basis of these genes showed that 4 of the 29 FGFR2 fusion–positive ICCs, as well as a subset of non-KRAS/nonfusion cases, display a molecular pattern reminiscent of KRAS mutant ICC (Supporting Fig. S4B). Mutational analysis of transcriptome data from these four FGFR2 fusion–positive patients revealed that these tumors bore genetic alterations in genes associated with MAPK activation (Supporting Fig. S5A). This is in line with the previously described frequent MAPK signaling activation in biliary tract cancers⁽²¹⁾ (Supporting Fig. S5B). Within a cohort of 457 FGFR2 fusion–positive patients with ICC, genetic alterations in NF1, NF2, EGFR, ERBB2, ERBB3, MET, PDGFRA, NRAS, and BRAF, were found in 7.4% of cases.

Overall, human ICC transcriptome analysis suggests that FGFR2 fusion–positive patients are a molecularly heterogeneous subgroup, but that individual cases exhibit expression patterns reminiscent of mutant KRAS ICC.