Multigene mutational profiling of cholangiocarcinomas identifies actionable molecular subgroups.

One-hundred-fifty-three biliary cancers, including 70 intrahepatic cholangiocarcinomas (ICC), 57 extrahepatic cholangiocarcinomas (ECC) and 26 gallbladder carcinomas (GBC) were assessed for mutations in 56 genes using multigene next-generation sequencing. Expression of EGFR and mTOR pathway genes was investigated by immunohistochemistry. At least one mutated gene was observed in 118/153 (77%) cancers. The genes most frequently involved were KRAS (28%), TP53 (18%), ARID1A (12%), IDH1/2 (9%), PBRM1 (9%), BAP1 (7%), and PIK3CA (7%). IDH1/2 (p=0.0005) and BAP1 (p=0.0097) mutations were characteristic of ICC, while KRAS (p=0.0019) and TP53 (p=0.0019) were more frequent in ECC and GBC. Multivariate analysis identified tumour stage and TP53 mutations as independent predictors of survival. Alterations in chromatin remodeling genes (ARID1A, BAP1, PBRM1, SMARCB1) were seen in 31% of cases. Potentially actionable mutations were seen in 104/153 (68%) cancers: i) KRAS/NRAS/BRAF mutations were found in 34% of cancers; ii) mTOR pathway activation was documented by immunohistochemistry in 51% of cases and by mutations in mTOR pathway genes in 19% of cancers; iii) TGF-ß/Smad signaling was altered in 10.5% cancers; iv) mutations in tyrosine kinase receptors were found in 9% cases. Our study identified molecular subgroups of cholangiocarcinomas that can be explored for specific drug targeting in clinical trials.


INTRODUCTION
Cholangiocarcinoma is a phenotypical and clinical heterogeneous collection of biliary tract malignancies, classified according to the World Health Organization (WHO) as intrahepatic (ICC) or extrahepatic cholangiocarcinomas (ECC) [1,2]. The former arise in the substance of the liver, the latter in large extrahepatic ducts, i.e. hepatic ducts and common bile duct. Gallbladder carcinomas (GBC) also have biliary epithelial Oncotarget 2840 www.impactjournals.com/oncotarget differentiation. Clinically, both cholangiocarcinomas and GBC have very poor prognosis. Surgical resection is the only potentially curative therapy, but most cases are inoperable [3][4][5][6][7]. In contrast to other solid tumours, no effective molecular targeted agent has been approved for biliary tract cancers, and patients have limited access to clinical trials [8][9][10][11].
The validation of whole exome studies by sequencing analysis of hotspot mutations in larger and characterized series has been a fruitful approach in identifying potential targets for personalized therapy for several malignancies [22]. Next-generation sequencing (NGS) has been recently introduced and is the most sensitive approach to simultaneously characterize multiple genes starting from a limited amount of DNA, also DNA derived from formalin-fixed paraffin-embedded (FFPE) samples [13,[23][24][25].
In the present study, we assayed the mutational status of 56 cancer-related genes in 153 biliary tract cancers, using a targeted next-generation sequencing methodology, with the aim of identifying molecular subgroups driving the development of personalized therapy approaches for patients affected by these neoplasms.

Clinico-pathological characteristics of the series
Patients' demographic and clinico-pathological data are summarized in Table 1. Mean tumour size was 4.8±3.4 cm (median=6.5; range=0.5-20.0), and was significantly higher in ICC than ECC and GBC (p=9.77 E -12 ). Synchronous multinodular lesions were found in 31 cases (20.3%). Tumour grading was G1 in 20, G2 in 94, and G3 in 38 cases, while the remaining case was undifferentiated.
The pathologic stage of the 153 neoplasms was  Stage I in 20, II in 51, III in 43, and IV in 39. ICC  presented with more advanced stages at surgery compared  to ECC and GBC (Table 1; p=0.034).
Next-generation sequencing of 56 genes dissects cholangiocarcinoma molecular heterogeneity DNA from all samples was successfully amplified in multiplex PCR for the 56 genes and an adequate library for deep sequencing was obtained. The mean read length was 78 base pairs and a mean coverage of 1800x was achieved, with 87.1% target bases covered more than 100x. A minimum coverage of 20x was obtained in all cases.
In ECC, KRAS was the most commonly mutated gene (47.4%), with codons 12, 13, 61 and 146 affected; one mutation was observed in NRAS, and none in BRAF. TP53 was the second most mutated gene (17.5%).

mTOR pathway is dysregulated in all cholangiocarcinoma subtypes and Egfr is significantly overexpressed in intrahepatic cholangiocarcinomas
The results of immunohistochemistry are summarized in Table 3. We investigated mTOR pathway and Egfr expression in 113 neoplastic and 18 control cases. EGFR gene copy number was analyzed by FISH.
A significant over-expression of the activated forms of mTOR and its effectors p70S6K and 4EBP1 was seen in most cancers with no significant differences among subtypes, but for p70S6K (Table 3). Of interest, the expression of phosphorylated ph-mTOR was significantly associated to the expression of the activated downstream effectors ph-4EBP1 and ph-p70S6K (p=0.05 and p=0.00012, respectively). Pten was significantly down-regulated in the whole series, particularly in ECC of common bile duct and in GBC: a weak cytoplasmic/ nuclear immunolabeling was observed in most cases.
There was no significant association between both EGFR and mTOR pathway immunophenotype and mutational status.

TP53 mutation is an independent prognostic factor in cholangiocarcinoma
Survival data were available in 125 cases (ICC=51; ECC=50; GBC=24). Median survival was 31 months and 79 (63.2%) subjects were followed to their deaths from disease.

DISCUSSION
The results of our next-generation mutational survey of 56 cancer genes in 153 biliary tree carcinomas can be summarized as follows: i) the vast majority (77.1%) of cancers harbours a driver-gene mutation; ii) the diverse sites of origin in the biliary tree have significantly different molecular profiles, and each site of origin also shows molecular heterogeneity; iii) targetable pathway alterations are present in 68% (104/153) of cancers, defining molecular cancer subclasses; iv) specific alterations are eligible for the investigational development of prognostic and non-invasive follow-up markers.
The vast majority of cancers (118/153, 77.1%) harboured at least one driver-gene mutation, and 39.2% (60/153 cases) showed concurrent mutations in two or more genes. KRAS was the most frequently mutated gene (28.1%), followed by TP53 (18.3%), as reported in prior studies [9,14,[35][36][37][38]. The recently described frequent involvement of the chromatin remodeling genes ARID1A, PBRM1 and BAP1 [9,13] was also confirmed in our series, being found in 11.8%, 9.2% and 7.2% of cases, respectively. Univariate survival analysis showed that KRAS mutations were associated to a worse patients' prognosis confirming the univariate analysis previously described by Andersen [42]. Multivariate survival analysis identified tumour stage and TP53 gene mutations as independent predictors of poor survival.
The different sites of origin showed significantly diverse molecular characteristics. IDH1/IDH2 mutations were restricted to ICC (p=0.0005) and, with the exception of one GBC, BAP1 mutations were all found in ICC (p=0.0097). ECC and GBC were characterized by a high prevalence of KRAS (p=0.0019) and TP53 (p=0.0019) mutations, respectively.
The standard of care of biliary tract cancer is based on the combination of cisplatin and gemcitabine [39]. To date, clinical trials with targeted therapies for advanced biliary tract cancers have failed to produce significant benefits [11], and ongoing studies are exploring the combination of chemotherapy with novel MAPK/ERK Kinase (MEK) and mTOR inhibitors [40]. However, neither previous nor ongoing studies have considered evaluating tumour response against genetic alterations. Our identification of molecular subclasses with specific drug actionable pathway alterations in 104/153 (68.0%) tumours may tailor the design of trials based on the molecular selection of patients, irrespective of the site of origin, where actionable signaling pathways include tyrosine-kinase receptors (TKR), RAS/RAF/MAPK/ERK, mTOR, and TGF-ß.
MTOR pathway relevance in biliary tract cancers is suggested by our immunohistochemical detection of activated forms of mTor and its downstream effectors in 51.3% of the cancers. The molecular basis of this activation in a proportion of cases is the mutation in one of the genes involved in this pathway (i. e., AKT, FBXW7,  PIK3CA, PIK3C2A, PIK3C2G, PTEN). This suggests that mTOR inhibitors might play a role in this molecular subgroup of patients.
TGF-ß/Smad signaling was altered in 16/153 cases (10.5%), 9 of which were ECC (9/57, 15.8%). Our finding supports previous studies demonstrating the involvement of this pathway in cholangiocarcinomas [45,46]. Our study may have underestimated the involvement of this pathway, as only the mutational status of TGFBR2 and SMAD4 genes was investigated, and the latter is frequently inactivated by mechanisms different from intragenic mutations, such as homozygous deletions, which would not be detected by the techniques used in this study.
Alterations in chromatin remodeling genes (ARID1A, BAP1, IDH1, IDH2, PBRM1, SMARCB1) were found in 30.7% (47/153) of cancers in our series, confirming recent reports on the significant involvement of these genes in cholangiocarcinoma [9,13]. Mutations in these genes appear to be either specific to ICC, as is the case of IDH1 and IDH2 (p=0.0005), or cluster within this cancer type as is the case for ARID1A, BAP1, and PBRM1 that were found in 34.3% (24/70) of ICC. The open challenge is now to translate knowledge of the targeting of these genes to improved patient care through either the development of new disease specific markers or of therapy targets. Of interest, IDH1/2 mutated cancers accumulate 2-hydroxyglutarate in tumour tissue and release the molecule in blood, and the measurement of 2-hydroxyglutarate might be used as both a surrogate biomarker for IDH1/2 mutational status and a noninvasive test for the assessment of tumour burden in ICC [47].
Our study is limited by the number of genes analyzed; 35 cancers (22.9%) showed no alterations in the 56 genes assayed, including 50 genes from a commercial panel and a custom panel exploring 6 recently discovered cholangiocarcinoma genes [9,13]. Mechanisms different from intragenic mutations, such as amplifications, deletions, translocations, and epigenetic anomalies should also be addressed.
In conclusion, we demonstrate that specific molecular alterations are associated to different cholangiocarcinomas categories and that potentially drug actionable pathways are evident in 68% of cases. These data further support the pathological and molecular heterogeneity characterizing biliary tree carcinomas. In currently designed clinical trials, cholangiocarcinomas are grouped together irrespective of their intrahepatic or extrahepatic site of origin. Our study shows that ICC and ECC should be considered separately, since they have different molecular characteristics. However, in the advent of molecular designed clinical trials, it would be appropriate to consider them together but only for the proportion of ICC and ECC sharing common molecular alterations.
We also show that a high-throughput next-generation sequencing analysis can be successfully applied using low amounts of DNA from routinely processed paraffin tissues. Such a time-and cost-effective analysis is the basis to significantly improve the development of personalized treatments for cholangiocarcinoma patients, and their early access to innovative drugs.

MATERIALS AND METHODS Cases
A retrospective series (1990-2011) of 153 surgicallyresected primary biliary cancers were retrieved from the FFPE archives of the ARC-Net Biobank at Verona University Hospital under the local ethics committee approval (n. prog. 1959). All cases were re-classified according to WHO 2010 [1], and included 70 ICC, 57 ECC and 26 GBC. Staging was according to AJCC/ UICC 7th edition [26]. Matched normal liver was used to determine the somatic/germline nature of mutations.
In 113 cases (57 ICC, 40 ECC, 16 GBC), sufficient material for the construction of 1-mm cores tissue microarrays (TMAs) was available. Eighteen nonneoplastic controls (8 normal biliary duct and 10 chronic cholecystitis) were included in the TMAs. Three tissue cores per case were analyzed.

DNA extraction and qualification
DNA was prepared after enrichment for neoplastic cellularity to at least 70% using manual microdissection of 10 consecutive 4-µm FFPE sections, purified using the QIAamp DNA FFPE Tissue Kit (Qiagen), and qualified as reported elsewhere [24,27].
Pten expression was considered positive if more than 50% of neoplastic cells showed a moderate/strong nuclear and cytoplasmic immunoreaction [29].
Staining for phosphorylated markers (ph-mTOR, ph-4EBP1, and ph-p70S6K) was considered positive when tumour cells showed cytoplasmic and/or nuclear staining with equal to stronger intensity compared with that of endothelial cells [30].

Fluorescent in situ hybridization (FISH)
The EGFR gene copy number status was assessed applying the Vysis EGFR/CEP7 Probe Kit (Vysis/Abbott Molecular). At least 50 representative nuclei per specimen were scored and EGFR-amplification was defined as described [31].

Statistical analysis
Kruskal-Wallis test, Chi-squared test with Monte Carlo simulation, and Fisher's exact test corrected for multiple comparisons were used as appropriate. For comparison of Kaplan-Meier survival curves, Mantel-Cox log-rank test was used; for multivariate survival analysis, stepwise Cox proportional hazards regression was used; selection of the best model was performed using the "backward elimination" algorithm. For all the analyses a p-value below 0.05 was considered significant. Graphs and univariate analyses were performed using GraphPad Prism® version 5.00 for Mac (GraphPad Software, San Diego California USA), multivariate Cox regression was done with R v. 3.0.2, using survival library v.2.37-4.

ACKNOWLEDGMENTS
The study has been supported by: an unrestricted grant of GlaxoSmithKline (WEUSKOP5847), AIRC grants n. 12182, 11930 and 6421, National Cancer Institutes (USA) grant CA62924, FP7 EU project CAM-PaC no. 602783, the Italian Cancer Genome Project grant from the Italian Ministry of Research (FIRB -RBAP10AHJB) and FIMP-Ministry of Health (CUP_ J33G13000210001).