Integrated Genomic Characterization Reveals Novel, Therapeutically Relevant Drug Targets in FGFR and EGFR Pathways in Sporadic Intrahepatic Cholangiocarcinoma

1Biliary tract cancers (BTC) comprise malignant tumors of the intrahepatic and extrahepatic bile ducts. Known risk factors for BTC are the liver flukes O. viverrini and C. sinensis in high prevalence endemic regions in southeast Asia –, as well as primary sclerosing cholangitis –, Caroli's disease, hepatitis B and hepatitis C –, obesity, hepatolithiasis, and thorotrast contrast exposure,. Surgical approaches such as resection and liver transplantation represent the only curative treatment approaches for BTC. Unfortunately, most patients present with surgically unresectable and/or metastatic disease at diagnosis. Systemic therapy with gemcitabine and cisplatin has been established as the standard of care for patients with advanced disease, but is only palliative, emphasizing the imminent need for novel therapies.

2Multiple studies have reported the presence of mutations/allelic loss of known cancer genes in BTC – and recently, a prevalence set of 46 patients was used to validate 15 of these genes including: TP53, KRAS, CDKN2A and SMAD4 as well as MLL3, ROBO2, RNF43, GNAS, PEG3, XIRP2, PTEN, RADIL, NCD80, LAMA2 and PCDHA13. Recent studies have also identified recurrent mutations in IDH1 (codon 132) and IDH2 (codons 140 and 172) with a prevalence of 22–23% associated with clear cell/poorly differentiated histology and intrahepatic primary,. Fusions with oncogenic potential involving the kinase gene ROS1 have been identified in patients with BTC with a prevalence of 8.7% in a recent study. Less frequently, mutations in sporadic BTC have been reported in EGFR,, BRAF, NRAS,, PIK3CA,,, APC, CTNNB1, AKT1, PTEN, ABCB4, ABCB11,, and CDH1 as well as amplifications in ERRB2.

3Recently, two independent studies reported the presence of FGFR fusions in cholangiocarcinoma; a single case with FGFR2-AHCYL1 as well as several cases identifying FGFR2-BICC1 fusions,. Arai et al. evaluated the presence of FGFR2 fusions in a cohort of 102 cholangiocarcinoma patients observing that the fusions occurred exclusively in the intrahepatic cases with a prevalence of 13.6%. Due to the presence of known dimerization motifs in the fusion partners, Wu et al. conducted mechanistic studies that demonstrated the in vitro interaction of FGFR2-BICC1 and other fusions that was not observed in the presence of wildtype FGFR2. Furthermore, overexpression of the FGFR2-BICC1 and other selected fusions resulted in altered cell morphology and increased cell proliferation. These data led to the conclusion that the fusion partners are facilitating oligomerization, resulting in FGFR kinase activation in tumors possessing FGFR fusions. In addition, in vitro and in vivo assessment of the sensitivity of cell lines containing an FGFR2 fusion to an FGFR inhibitor demonstrated sensitivity to treatment only in the fusion containing cells,, suggesting the presence of FGFR fusions may be a useful predictor of tumor response to FGFR inhibitors.

4To comprehensively explore the genetic basis of sporadic intrahepatic cholangiocarcinoma (SIC), with emphasis on elucidation of therapeutically relevant targets, we performed integrated whole genome and whole transcriptome analyses on tumors from 6 patients with advanced, sporadic intrahepatic cholangiocarcinoma (SIC). Notably, recurrent fusions involving the oncogene FGFR2 (n = 3) were identified. A patient whose tumor presented with an FGFR2-MGEA5 fusion has demonstrated preliminary evidence of anti-tumor activity manifest as stable disease accompanied by CA19-9 reduction and tumor necrosis to ponatinib, a pan-FGFR inhibitor (in vitro FGFR1 IC50≈24 nM, FGFR2 IC50≈8 nM, FGFR3 IC50≈8 nM and FGFR4 IC50≈34 nM). In another patient whose tumor possessed an FGFR2-TACC3 fusion, preliminary anti-tumor activity of pazopanib (in vitro FGFR2 IC50≈350 nM) was also noted. After progression on pazopanib, the same patient also responded to ponatinib and again demonstrated tumor shrinkage. Additionally, a non-FGFR fusion patient was found to have allele-specific preferential expression of a loss of function mutation in ERRFI1, a direct negative regulator of EGFR activation. Similarly, rapid and robust disease regression was noted in the patient with an ERRFI1 mutant tumor when treated with erlotinib, an EGFR kinase inhibitor. Results suggest that these novel targets in the EGFR and FGFR pathways may be therapeutically relevant in patients with sporadic cholangiocarcinoma.

5We identified 327 somatic coding mutations, with an average of 55 mutations/tumor (range 34–112), within our cohort ( Table 1, Figure 1 ). Nonsynonymous single nucleotide variations were the predominant class in all of the patients. Patients 1 and 2 accumulated a high number of synonymous mutations in comparison to the other patients. Patient 5 carried the most stops gained likely contributing to a higher number of pseudogenes in comparison to the others and was also the only patient to carry several predicted high impact mutations affecting splice site acceptor regions ( Figure 1, light green, percentage <5%). In addition, patient 6 also carried a codon change plus insertion variation. Sequencing statistics are provided in Table 2. Genes with mutations in more than one case included CSPG4 (n = 2), GRIN3A (n = 2) and PLXBN3 (n = 2) (Table S1); with half of these predicted to be potentially damaging by SIFT, Polyphen, Mutation Assessor and Mutation Taster. While there was overlap in the somatic landscape of SIC with liver-fluke associated cholangiocarcinoma, hepatocellular cancer and pancreatic cancer, most of the aberrations detected in our study were distinct ( Table 3 ). More importantly, using previously published methods, we identified molecular fusions involving FGFR2 that were felt to be therapeutically relevant in 3 patients. Additionally, these fusions were validated with a break apart Fluorescent In situ Hybridization (FISH) assay ( Figure 2 ). Notably, the patients who did not harbor the FGFR2 fusions were negative using the same assay. Two of the three patients with FGFR2 fusions (Patients 4 and 6) were treated with FGFR inhibitors while the third patient (Patient 5), experienced clinical decline prior to the availability of results and as such did not receive any further therapy. Furthermore, overexpression of an SNV in ERRFI1 (E384X), a negative regulator of EGFR, was detected in a non-FGFR2 translocation patient's tumor. Taken together, our results constitute important therapeutically actionable alterations in patients with advanced SIC (Text S1).

6Comparative pathway analysis of genes carrying small scale nucleotide variations (SsNVs) has implicated several major pathways, possibly interacting as a network, that are predicted to underlie disease in all of our studied biliary carcinoma patients. These shared pathways include EGFR, EPHB, PDGFR-beta, Netrin-mediated and Beta1 integrin mediated signaling pathways ( Figure 3 and Tables S2 and S3). Interestingly, most of these pathways have known roles in mediating epithelial-to-mesenchymal cell transitions, which occur frequently during development as well as tumorigenesis. Cell growth and motility is inherent to the successful progression of both biological processes. Studies of the nervous system and lung development have shown that Netrins act to inhibit FGF7 and FGF10 mediated growth or cell guidance. In addition, Netrin-1 has a known role in mediating cell migration during pancreatic organogenesis. Furthermore, Netrin-1 acts as a ligand for α3β1 and α6β4 integrins, both of which are involved in supporting adhesion of developing pancreatic epithelial cells with Netrin-1 although it is thought that α6β4 plays the principle role during this process. Interestingly, α3β1 has been hypothesized to play a role during the process of angiogenesis, when chemoattractants and chemorepellents act to guide filopodia during migration. The α3β1 integrin receptor may act together with additional pathways proposed to play a role during angiogenesis such as VEGF, PDGFR-beta, and EphrinB as well as tumorigenesis. Patients 3 and 4 also shared several genes acting in cadherin signaling pathways (Tables S3, S4), which are important for maintaining cell-cell adhesion and are known to be intimately integrated with EGFR and FGFR signaling pathways.

7In addition to the variations identified in genes acting in EGFR and/or FGFR signaling pathways, we also report multiple sSNVs and copy number variations (CNVs) ( Figure 4 ) in genes such as HDAC1, TP53, MDM2 and AKT1, acting in interaction networks or regulatory pathways involving the fusion partner genes in patients 5 (BICC1), and 6 (TACC3) ( Table 4 ). Known mutations in BICC1 have been shown to disrupt canonical Wnt signaling and genes, such as BCL9, involved in this pathway are known to regulate a range of biological processes such as transcription and cell proliferation and carry variations in patient 5 ( Table 4 ). CSPG4, a target that is being investigated for antibody-based immunotherapy in preclinical studies of triple negative breast cancer, is involved in the Wnt signaling pathway, and carries variations in both patients 1 and 2, however, it is not mutated in patient 5. TACC3 is known to mediate central spindle assembly and multiple genes including CDCA8, BUB1, and TACC1, belonging to the TACC3 interaction network exhibit aberrant copy number in patient 6 ( Table 4 ). A recent study has also implicated TACC3 in EGF-mediated EMT when overexpressed, and we find that the PLCG1, MAP2K1, and MAPK8 genes, which act in both FGFR and EGFR regulatory pathways, exhibit CNV in patient 6. We also note that the DNAH5 gene encoding a dynein protein which is part of the microtubule-associated motor protein complex carries two G→C missense mutations in patient 6 (Table S1). Several genes carrying more than one variation in either the same patient or different patients also included genes with known roles similar to genes in FGFR/EGFR pathways including axon guidance, invasive growth, or cell differentiation (NAV3, LAMC3, PLXNB3, and PTPRK) (Table S1). In the case of patient 4, our studies suggest that the primary effect of the FGFR2-MGEA5 fusion is on FGFR2 related signaling, since changes in expression were observed in FGF8 (p<0.05) and the genome of this patient also carries a 4-bp insertion (∧GTGT) in the FGFR4 gene (Table S1).

8Patient 4 is a 62 year-old white female found to have a left-sided intrahepatic mass with satellite lesions, with metastasis to regional lymph nodes ( Table 5 ). A biopsy of the liver mass revealed the presence of a poorly differentiated adenocarcinoma that was consistent with intrahepatic cholangiocarcinoma (CK7+, CEA+, CK20+, Hep-par 1−, TTF-1−) ( Table 6 ). She received gemcitabine and cisplatin and obtained clinical benefit in the form of stable disease for 6 months, followed by disease progression. She was re-treated with gemcitabine and capecitabine systemic therapy and attained stable disease for 6 months, followed by disease progression. A clinical trial of pegylated hyaluronidase (PEGPH20) produced only stable disease for 4 months, followed again by disease progression. At this juncture, she underwent a liver biopsy to obtain tissue for whole genome characterization of her tumor. She was found to have an FGFR2-MGEA5 fusion ( Table 7, Figure 2 ) and ponatinib monotherapy was pursued as salvage treatment. Evaluation of pre-treatment immunohistochemistry demonstrated increased expression of FGFR2 and FGFR3 ( Figure 5 ) and Clinical Laboratory Improvement Amendments (CLIA) validation by quantitative PCR revealed increased expression of FGFR3 (Table S5). In order to further validate the activation of the receptor, we conducted immunohistochemistry (IHC) of pFRS2 Y436 and pERK(MAPK) that revealed strong expression of pFRS2 Y436 and pERK ( Figure 6 ), thus confirming activation of the receptor.

9Ponatinib was initiated at 45 mg given orally on a daily schedule. Approximately 6 weeks after initiation of therapy she was noted to have anti-tumor activity that was characterized by central necrosis of a caudate liver lobe mass, shrinkage of metastatic lymph nodes involving the right cardiophrenic angle, central necrosis and shrinkage of a metastatic supraceliac axis lymph node ( Figure 7 ) and reduction in CA 19-9 from 1408 U/ml to 142 U/ml. Per RECIST criteria, she exhibited stable disease with a 14% decrease in the sum of largest diameters but with tumor necrosis and reduction in the CA19-9 tumor marker (89.8%). While the evidence is preliminary in nature, it was felt that the combination of tumor shrinkage not meeting the RECIST criteria definition of partial response, tumor necrosis and reduction in CA19-9 constituted preliminary evidence of anti-tumor activity. She has experienced no treatment related toxicities thus far and remains on therapy of approximately 3.5 months duration thus far.

10The FGFR2 fusion partner observed in this patient, MGEA5, is an enzyme responsible for the removal of O-GlcNAc from proteins. Interestingly, soft tissue tumors myxoinflammatory fibroblastic sarcoma (MIFS) and hemosiderotic fibrolipomatous tumor (HFLT) both share a translocation event resulting in rearrangements in TGFBR3 and MGEA5,. Associated with this translocation event is the upregulation of NPM3 and FGF8, of which both genes are upregulated in this patient (fold change: NPM3 = 6.17865, FGF8 = 1.79769e+308). In breast cancer, grade III tumors had significantly lower MGEA5 expression than grade I tumors with a trend of decreasing expression observed with increasing tumor grade. In summary, MGEA5 may play an important role in carcinogenesis as an FGFR fusion partner.

11Patient 6 is a 43 year-old white female who underwent a right salpingo-oophorectomy and endometrial ablation in the context of a ruptured ovarian cyst ( Table 5 ). Postoperatively she developed dyspnea and was found to have pulmonary nodules as well as a 5 cm left sided liver mass. Pathological evaluation of the liver mass was consistent with a moderately differentiated intrahepatic cholangiocarcinoma (CK7+, CK20−, TTF-1−) in the absence of any known risk factors ( Table 6 ). She was treated systemically with gemcitabine and cisplatin and had stable disease for approximately 6 months, but was subsequently found to have disease progression. She was treated with FOLFOX for 7 months and again attained stable disease as best response to therapy but eventually experienced disease progression. Upon disease progression, she was enrolled on a clinical study with the multi-kinase inhibitor pazopanib that is FDA-approved for the treatment of advanced renal cancer or sarcoma – and fortuitously has nanomolar activity against FGFR2 (in vitro IC50 to FGFR2≈350 nM). Transcriptome analysis revealed the presence of an FGFR2-TACC3 fusion ( Table 7, Figure 2 ). Evaluation of post-pazopanib tissue by immunohistochemistry revealed increased expression of FGFR2 and FGFR3 ( Figure 5 ) Further evaluation of phosphorylation of downstream targets FRS2 Y436, and ERK(MAPK) revealed strong expression of pERK and moderate expression of pFRS2 Y436 ( Figure 6 ), confirming activation of the receptor. She had been treated with pazopanib 800 mg orally daily for 4 months and demonstrated tumor shrinkage, which in retrospect, was postulated to be secondary to the FGFR2 inhibitory activity of pazopanib ( Figure 8A ). By RECIST criteria v1.1, the patient had a partial response to therapy as evidenced by a 71% decrease in the sum of diameters. Subsequently, the same patient was treated with a dedicated pan-FGFR inhibitor, ponatinib, (45 mg daily orally; in vitro IC50: FGFR1≈24 nM, FGFR2≈8 nM, FGFR3≈8 nM and FGFR4≈34 nM). She again attained minor tumor shrinkage (stable disease by RECIST criteria v1.1, decrease of 4% in sum of largest diameters) in multiple lesions after 2 months of therapy, despite undergoing a 50% dose reduction for abdominal pain felt to be related to drug ( Figure 8B ). She remains on therapy approximately 4 months since the initiation of ponatinib. As such, anti-tumor activity was obtained with two distinct FGFR inhibitors in the same patient.

12The FGFR2 fusion partner observed in this patient's tumor, TACC3, is overexpressed in many tumor types with enhanced cell proliferation, migration, and transformation observed in cells overexpressing TACC3. Furthermore regulation of ERK and PI3K/AKT by TACC3 may contribute in part to epithelial-mesenchymal transition (EMT) in cancer, a significant contributor to carcinogenesis. Interestingly, TACC3 has been identified as a fusion partner to FGFR3 in bladder cancer, squamous cell lung cancer, oral cancer, head and neck cancer and glioblastoma multiforme.

13Patient 3 was a 50 year-old white male who presented with fevers and night sweats ( Table 5 ). He was found to have a 4 cm tumor in his liver determined to be a poorly differentiated intrahepatic cholangiocarcinoma (CK7+, CK20−, TTF1−, CD56−, synatophysin−, Hep-par 1−) with sclerotic features ( Table 6 ). No overt risk factors for cholangiocarcinoma were identified. A left hepatectomy was undertaken three months later. In addition to the primary tumor in segment 4, limited resections of segments 6 and 8 were undertaken to remove two tumor nodules. He was soon noted to have increased hypermetabolism in the left lower cervical, upper mediastinal and abdomino-retroperitoneal lymph nodes related to metastatic disease from his cholangiocarcinoma. He was treated with gemcitabine and cisplatin for 9 months and obtained stable disease as his best response, followed by eventual progression. He received treatment with pegylated hyaluronidase (PEGPH20) in the setting of an investigational study for one month and had no response to therapy. A biopsy of a left supraclavicular lymph node was obtained two months prior to the initiation of PEGPH20 in the context of a clinical study employing whole genome analysis for putative therapeutic target selection.

14Since our study goal was to identify potential therapeutically relevant events, the novel loss of function mutation in ERRFI1 (E384X) detected in Patient 3's metastatic, recurrent/refractory SIC (Table S1) warranted additional examination. Specifically, the allelic fraction of the DNA mutation constituted only 11% of the sequencing reads, is consistent with tissue heterogeneity, and constituted 78% of the sequencing reads within the RNASeq data. Such allele specific expression of the mutated allele from the same tissue specimen suggests nearly complete loss of function of ERRFI1 in this patient's tumor. Notably, the patient's tumor did not harbor any mutations or amplifications in other EGFR signaling members such as EGFR and BRAF.

15Upon availability of CLIA validated sequencing data (Table S5), the patient was treated with erlotinib 150 mg orally/daily. After 3 months, RECIST v1.1 partial response evidenced by a decrease of 58% in the sum of largest diameters was observed ( Figure 9 ). Evaluation of pretreatment tumor tissue by immunohistochemistry revealed increased expression of EGFR pathway members ( Figure 10 ).

16Integrated analysis of sporadic intrahepatic cholangiocarcinoma (SIC) genomic and transcriptomic data led to the discovery of FGFR2 fusion products in three of six assessed patients ( Table 7, Figures 4 and 11 ). Members of the FGFR family (FGFR1-4) have been associated with mutations, amplifications and translocation events with oncogenic potential. FGFR fusions with oncogenic activity have been previously identified in bladder cancer (FGFR3), lymphoma (FGFR1 and FGFR3),, acute myeloid leukemia (FGFR1), multiple myeloma, myeloproliferative neoplasms, and most recently glioblastoma multiforme (FGFR1 and FGFR3). FGFR2, FGFR3 and FGFR4 have been found to be overexpressed in IDH1/IDH2 mutant biliary cancers, a context seen within Patient 1 in our study (Tables S1 and S6, Figure 5 ); although, no fusion events were depicted in these studies or in Patient 1.

17Although the gene partner fused to FGFR2 was different for each patient (MGEA5, BICC1 and TACC3), the breakpoints in FGFR2 all occurred within the last intron distal to the last coding exon and terminal protein tyrosine kinase domain ( Figure 11 ). All three fusions were validated at the DNA and/or RNA level ( Table 8 ). Amongst these fusions, the FGFR2-BICC1 fusion has recently been independently identified in SIC,. For this particular fusion product we observed, and validated, the presence of two fusion isoforms (FGFR2-BICC1 and BICC1-FGFR2). Interestingly, BICC1 is a negative regulator of Wnt signaling and when comparing expression of tumor and normal tissue we observed differentially expressed Wnt signaling genes, APC (fold change -4.75027), GSK3B (fold change -3.35309), and CTNNB1 (fold change -1.73148), yet when the expression was compared to other cholangiocarincomas, no difference was observed.

18The FGFR genes encode multiple structural variants through alternative splicing. Notably, RNASeq data revealed that the FGFR2-IIIb isoform was present in all fusions detected in our study and has been shown to have selectivity for epithelial cells as opposed to the FGFR2-IIIc isoform, which is found selectively in mesenchymal cells. Paradoxically, wildtype FGFR2-IIIb has been described as a tumor suppressor in pre-clinical systems of bladder cancer and prostate cancer,. As such, FGFR signaling appears context-dependent and exhibits variability in disparate tumor types.

19Importantly, one critical study has shown that FGFR2 carboxy-terminal deletion mutants induce ligand-independent transformation and clonogenic growth. This is important because all of the fusion events within our study would lead to loss of the carboxy-terminus of FGFR2. Furthermore, a very recent study that described FGFR fusions in solid tumors illustrated that FGFR fusion partners in SIC resulted in dimerization domains, and suggested that activation occurred through ligand independent dimerization and oligomerization. It is likely that both loss of the carboxy terminus and the addition of dimerization domains leads to oncogenic FGFR2 activity in these tumors.

20Comparative pathway analysis of genes carrying mutations/aberrant in copy number identified additional potential therapeutic targets belonging to, or intimately integrated with, the EGFR and FGFR signaling pathways ( Figure 3, Tables S2, S3, S4). Interestingly, most of these pathways also have known roles in mediating epithelial-to-mesenchymal cell transitions, which occur frequently during development as well as during tumorigenesis. Patients 3 and 4 harbored aberrations in several genes acting in cadherin signaling pathways (Tables S3, S4), which are important for maintaining cell-cell adhesion.

21The preliminary anti-tumor activity noted in a patient with FGFR2-MGEA5 (Patient 4) and FGFR2-TACC3 fusion (Patient 6) represent the first reports of application of FGFR inhibitors to the treatment of patients with cholangiocarcinoma harboring these alterations. These results suggest that oncogenic activation of FGFR2 represent a potential therapeutically actionable event. The FGFR tyrosine kinase inhibitors (TKI) dovitinib and NVP-BGJ398 are currently in clinical development and the FGFR TKI ponatinib, was recently approved by the FDA for use in treating T315I mutant chronic myelogenous leukemia. FGF7 (keratinocyte growth factor) has been previously linked to poor prognosis in patients with biliary tract cancer and a small molecule FGFR kinase inhibitor, Ki23057, has demonstrated efficacy in preclinical models. It should be recognized that small molecule tyrosine inhibitors are almost universally promiscuous with regards to specificity and typically significant off-target effects are resultant. Off target efficacy resulting from inhibition of angiogenic kinases in addition to FGFR2 inhibition could explain the anti-tumor activity exhibited in patient 6, as pazopanib has been shown to have nanomolar range potency towards VEGFR1-3, PDGFRA/B and CKIT as well.

22Larger trials, preferably of a randomized nature with a control arm, need to be conducted to truly define the role of FGFR inhibitors in the treatment of patients with cholangiocarcinoma, particularly those harboring FGFR2 fusions. While our results provide impetus and enthusiasm towards this end, at this stage they should be considered preliminary in nature.

23The preliminary anti-tumor activity observed in patient 6 with both pazopanib, and subsequently ponatinib, is particularly intriguing, but also raises important questions. There was an initial response to pazopanib, followed by disease progression. This is a phenomenon observed with the clinical application of most targeted therapeutic approaches. Potential explanations include tumor heterogeneity resulting from clonal selection, transcriptional up-regulation of escape pathways, epigenetic mechanisms and other yet undefined mechanisms of resistance to therapy. The patient did not have additional known alterations in key oncogenic pathways in genes such as BRAF, KRAS, EGFR and PIK3CA, which if present, could provide a putative basis for eventual escape from FGFR pathway inhibition. It is unclear why patient 6 initially responded to pazopanib followed by resistance and subsequently responded to ponatinib, another FGFR inhibitor. Putative explanations include the higher potency of ponatinib observed in vitro to FGFR2 (IC50≈8 nM for ponatinib vs. 350 nM for pazopanib) and resistance being defined as >20% increase in sum of largest diameters per RECIST v1.1 standard criteria that triggered a discontinuation from pazopanib and recapturing of anti-tumor activity by subsequent inhibition of the FGFR pathway which still maintained therapeutic relevance in that patient at a later time point.

24ERRFI1 has a role as a negative regulator of EGFR dependent skin morphogenesis,, uterine steroid hormone responsiveness and as a tumor suppressor gene,,. ERRFI1 is an endogenous inhibitor of EGFR, ERRB2, ERRB3 and ERRB4 through direct interaction with the kinase domains of these proteins, and endocytosis/lysosomal degradation of ERBB receptors. ERRFI1 deletions have been found in glioblastoma multiforme and breast cancer –. Other mechanisms of ERRFI1 loss include methylation, acetylation and loss of function mutations,,. Consistent with a driver role of this mutation, previously germline homozygous disruption of ERRFI1 in mice induces hyperplasia and adenoma formation in the epithelium and development of spontaneous adenocarcinomas of the lung, gallbladder and biliary tract. The tyrosine kinase inhibitor gefitinib has demonstrated anti-tumor activity in mice in spontaneous tumors driven by ERRFI1 germline loss.

25Our results suggest immediate and actionable implications for SIC patients with tumors harboring ERFFI1 loss of function mutations or FGFR fusions, given the clinical availability of FDA-approved EGFR and FGFR tyrosine kinase inhibitors. Antibodies specific to FGFR2-IIIb have also shown preclinical efficacy and may serve as an additional platform for therapeutic development in this context. Additional studies to characterize the prevalence of these aberrations in both sporadic and liver fluke associated BTC will need to be conducted. Nevertheless, our results suggest that prospective clinical studies designed to treat patient's tumors harboring these novel genomic aberrations utilizing targeted agents on an individualized basis should be pursued more fully through larger clinical studies in order to explore the precise extent of clinical benefit that this tailored approach may have in patients with primary or advanced BTC.

26Additionally, post-treatment biopsies to assess pathway down-regulation in patients 4 and 6 (treated with FGFR inhibitors) and patient 3 (treated with EGFR inhibitor) are not available, as the treatment was not conducted in the setting of a protocol that would allow for the collection of additional research biopsies. Incorporation of post-treatment biopsies in carefully designed prospective studies will be critical towards defining the association between the use of FGFR and EGFR inhibitors in appropriately selected patients with relevant genomic aberrations.

27Clinical information was assimilated from patient records from the Mayo Clinic. Informed consent was obtained for each patient on two ongoing research protocols approved by the Mayo Clinic Institutional Review Board (10-006180 and 10-002879). Clinicopathological features collected included age, gender, stage, histological grade, sites of metastasis, tumor sample assessment for overall cellularity/necrosis and percent tumor cellularity, prior therapies and risk factors (hepatitis B and C, Caroli's disease, obesity, hepatolithiasis and cholelithiasis, primary sclerosing cholangitis, thorotrast exposure and H. pylori, H. bilis, S. typhi and S. paratyphi infections). All patients were known to not have had prior exposure to liver flukes that have been implicated in biliary carcinogenesis (O. viverrini and C. sinensis). Tissue specimens were collected fresh frozen and maintained below −80°C until nucleic acid extraction. A board certified pathologist who is experienced in biospecimen studies, evaluated a portion of each specimen to confirm the presence of tumor, the degree of necrosis and the percent cellularity.

281.1 µg genomic DNA was used to generate separate long insert whole genome libraries for each sample using Illumina's (San Diego, CA) TruSeq DNA Sample Prep Kit (catalog# FC-121-2001). In summary, genomic DNAs are fragmented to a target size of 900–1000 bp on the Covaris E210. 100 ng of the sample was run on a 1% TAE gel to verify fragmentation. Samples were end repaired and purified with Ampure XP beads using a 1∶1 bead volume to sample volume ratio, and ligated with indexed adapters. Samples are size selected at approximately 1000 bp by running samples on a 1.5% TAE gel and purified using Bio-Rad Freeze ‘n Squeeze columns and Ampure XP beads. Size selected products are then amplified using PCR and products were cleaned using Ampure XP beads.

29300 ng genomic tumor and normal DNA was used to create whole genome libraries. Samples were fragmented on the Covaris E210 to a target size of 200–300 bp and 50 ng of the fragmented product was run on a 2% TAE gel to verify fragmentation. Whole genome libraries were prepared using Illumina's TruSeq DNA Sample Prep Kit.

301.1 µg genomic DNA for each sample was fragmented to a target size of 150–200 bp on the Covaris E210. 100 ng of fragmented product was run on TAE gel to verify fragmentation. The remaining 1 µg of fragmented DNA was prepared using Agilent's SureSelectXT and SureSelectXT Human All Exon 50 Mb kit (catalog# G7544C).

3150 ng genomic tumor and normal DNA was used to create exome libraries using Illumina's Nextera Exome Enrichment kit (catalog# FC-121-1204) following the manufacturer's protocol.

321 µg of each tumor and germline DNA sample was used to generate separate exome libraries. Libraries were prepared using Illumina's TruSeq DNA Sample Prep Kit and Exome Enrichment Kit (catalog# FC-121-1008) following the manufacturer's protocols.

333 µg of genomic tumor and normal DNA was fragmented on the Covaris E210 to a target size of 150–200 bp. Exome libraries were prepared with Agilent's (Santa Clara, CA) SureSelectXT Human All Exon V4 library preparation kit (catalog# 5190-4632) and SureSelectXT Human All Exon V4+UTRs (catalog# 5190-4637) following the manufacturer's protocols.

3450 ng total RNA was used to generate whole transcriptome libraries for RNA sequencing. Using the Nugen Ovation RNA-Seq System v2 (catalog# 7102), total RNA was used to generate double stranded cDNA, which was subsequently amplified using Nugen's SPIA linear amplification process. Amplified products were cleaned using Qiagen's QIAquick PCR Purification Kit and quantitated using Invitrogen's Quant-iT Picogreen. 1 µg of amplified cDNA was fragmented on the Covaris E210 to a target size of 300 bp. Illumina's TruSeq DNA Sample Preparation Kit was used to prepare libraries from 1 µg amplified cDNA.

351 µg of total RNA for each sample was used to generate RNA sequencing libraries using Illumina's TruSeq RNA Sample Prep Kit V2 (catalog# RS-122-2001) following the manufacturer's protocol.

36Libraries with a 1% phiX spike-in were used to generate clusters on HiSeq Paired End v3 flowcells on the Illumina cBot using Illumina's TruSeq PE Cluster Kit v3 (catalog# PE-401-3001). Clustered flowcells were sequenced by synthesis on the Illumina HiSeq 2000 using paired-end technology and Illumina's TruSeq SBS Kit.

37For whole genome and exome sequencing fastq files were aligned with BWA 0.6.2 to GRCh37.62 and the SAM output were converted to a sorted BAM file using SAMtools 0.1.18. BAM files were then processed through indel realignment, mark duplicates, and recalibration steps in this order with GATK 1.5 where dpsnp135 was used for known SNPs and 1000 Genomes' ALL.wgs.low_coverage_vqsr.20101123 was used for known indels. Lane level sample BAMs were then merged with Picard 1.65 if they were sequenced across multiple lanes. Comparative variant calling for exome data was conducted with Seurat.

38Previously described copy number and translocation detection were applied to the whole genome long insert sequencing data and these are made available through https://github.com/davcraig75/tgen_somaticSV. Copy number detection was based on a log2 comparison of normalized physical coverage (or clonal coverage) across tumor and normal whole genome long-insert sequencing data, where physical coverage was calculated by considering the entire region a paired-end fragment spans on the genome, then the coverage at 100 bp intervals was kept. Normal and tumor physical coverage was then normalized, smoothed and filtered for highly repetitive regions prior to calculating the log2 comparison. Translocation detection was based on discordant read evidence in the tumor whole genome sequencing data compared to its corresponding normal data. In order for the structural variant to be called there needs to be greater than 7 read pairs mapping to both sides of the breakpoint. The unique feature of the long-insert whole-genome sequencing was the long overall fragment size (∼1 kb), where by two 100 bp reads flank a region of ∼800 bp. The separation of forward and reverse reads increases the overall probability that the read pairs do not cross the breakpoint and confound mapping.

39For RNA sequencing, lane level fastq files were appended together if they were across multiple lanes. These fastq files were then aligned with TopHat 2.0.6 to GRCh37.62 using ensembl.63.genes.gtf as GTF file. Changes in transcript expression were calculated with Cuffdiff 2.0.2. For novel fusion discovery reads were aligned with TopHat-Fusion 2.0.6 (patients 2, 3, 4 and 6). In addition, Chimerascan 0.4.5 was used to detect fusions in patient 1, deFuse 5.0 used in patients 2, 3 and 5 and SnowShoes for patients 2 and 5.

40Mutations of potential clinical relevance were confirmed in a Clinical Laboratory Improvement Amendments (CLIA) laboratory with Sanger sequencing or quantitative PCR.

41The immunohistochemistry was performed following the procedures described previously. Briefly, slides were dewaxed, rehydrated and antigen retrieved on-line on the BondMax autostainer (Leica Microsystems, INC Bannockburn, IL). Slides were then subjected to heat-induced epitope retrieval using a proprietary EDTA-based retrieval solution. Endogenous peroxidase was then blocked and slides were incubated with the following antibodies: FGFR2 (BEK, Santa Cruz, catalog# sc-20735), FGFR3 (C-15, Santa Cruz, catalog# sc-123), panAKT (Cell Signaling Technology, catalog# 4685, pAKT (Cell Signaling Technology, catalog# 4060), EGFR (Cell Signaling Technology, catalog# 4267, pEGFR (Cell Signaling Technology, catalog#2234), MAPK/ERK1/2 (Cell Signaling Technology, catalog# 4695), pMAPK/pERK (Cell Signaling Technology, catalog# 4376) and pFRS2 Y436 (Abcam, catalog# ab78195). Sections were visualized using the Polymer Refine Detection kit (Leica) using diaminobenzidine chromogen as substrate.

42FISH was performed on formalin-fixed paraffin-embedded (FFPE) specimens using standard protocols and dual-color break-apart rearrangement probes specific to the FGFR2 gene (Abbott Molecular, Inc. Des Plaines, IL) located at 10q26. The 5′ FGFR2 signal was labeled with Spectrum Orange (orange) and the 3′ FGFR2 signal was labeled with Spectrum Green (green).