Whole-genome mutational landscape of liver cancers displaying biliary phenotype reveals hepatitis impact and molecular diversity

Abstract

Intrahepatic cholangiocarcinoma and combined hepatocellular cholangiocarcinoma show varying degrees of biliary epithelial differentiation, which can be defined as liver cancer displaying biliary phenotype (LCB). LCB is second in the incidence for liver cancers with and without chronic hepatitis background and more aggressive than hepatocellular carcinoma (HCC). To gain insight into its molecular alterations, we performed whole-genome sequencing analysis on 30 LCBs. Here we show, the genome-wide substitution patterns of LCBs developed in chronic hepatitis livers overlapped with those of 60 HCCs, whereas those of hepatitis-negative LCBs diverged. The subsequent validation study on 68 LCBs identified recurrent mutations in TERT promoter, chromatin regulators (BAP1, PBRM1 and ARID2), a synapse organization gene (PCLO), IDH genes and KRAS. The frequencies of KRAS and IDHs mutations, which are associated with poor disease-free survival, were significantly higher in hepatitis-negative LCBs. This study reveals the strong impact of chronic hepatitis on the mutational landscape in liver cancer and the genetic diversity among LCBs.

ARTICLE

Received 30 May 2014 | Accepted 17 Dec 2015 | Published 30 Jan 2015

Akihiro Fujimoto1,2,*, Mayuko Furuta1,*, Yuichi Shiraishi3, Kunihito Gotoh4, Yoshiiku Kawakami5, Koji Arihiro6, Toru Nakamura7, Masaki Ueno8, Shun-ichi Ariizumi9, Ha Hai Nguyen1,10, Daichi Shigemizu2, Tetsuo Abe2, Keith A. Boroevich2, Kaoru Nakano1, Aya Sasaki1, Rina Kitada1, Kazihiro Maejima1, Yujiro Yamamoto1,Hiroko Tanaka11, Tetsuo Shibuya11, Tatsuhiro Shibata12, Hidenori Ojima13, Kazuaki Shimada14, Shinya Hayami8, Yoshinobu Shigekawa8, Hiroshi Aikata5, Hideki Ohdan15, Shigeru Marubashi4, Terumasa Yamada4,Michiaki Kubo16, Satoshi Hirano7, Osamu Ishikawa4, Masakazu Yamamoto9, Hiroki Yamaue8,Kazuaki Chayama5,17, Satoru Miyano3,11, Tatsuhiko Tsunoda2 & Hidewaki Nakagawa1

Intrahepatic cholangiocarcinoma and combined hepatocellular cholangiocarcinoma show varying degrees of biliary epithelial differentiation, which can be dened as liver cancer displaying biliary phenotype (LCB). LCB is second in the incidence for liver cancers with and without chronic hepatitis background and more aggressive than hepatocellular carcinoma (HCC). To gain insight into its molecular alterations, we performed whole-genome sequencing analysis on 30 LCBs. Here we show, the genome-wide substitution patterns of LCBs developed in chronic hepatitis livers overlapped with those of 60 HCCs, whereas those of hepatitis-negative LCBs diverged. The subsequent validation study on 68 LCBs identied recurrent mutations in TERT promoter, chromatin regulators (BAP1, PBRM1 and ARID2), a synapse organization gene (PCLO), IDH genes and KRAS. The frequencies of KRAS and IDHs mutations, which are associated with poor disease-free survival, were signicantly higher in hepatitis-negative LCBs. This study reveals the strong impact of chronic hepatitis on the mutational landscape in liver cancer and the genetic diversity among LCBs.

1 Laboratory for Genome Sequencing Analysis, RIKEN Center for Integrative Medical Sciences, Tokyo 108-8639, Japan. 2 Laboratory for Medical Science Mathematics, RIKEN Center for Integrative Medical Sciences, Yokohama 230-0045, Japan. 3 Laboratory of DNA Information Analysis, Human Genome Center, The Institute of Medical Science, The University of Tokyo, Tokyo 108-8639, Japan. 4 Department of Surgery, Osaka Medical Center for Cancer and Cardiovascular Diseases, Osaka 537-8511, Japan. 5 Department of Medicine & Molecular Science, Hiroshima University School of Medicine, Hiroshima 734-8551, Japan. 6 Department of Anatomical Pathology, Hiroshima University School of Medicine, Hiroshima 734-8551, Japan. 7 Department of Gastroenterological Surgery II, Hokkaido University Graduate School of Medicine, Sapporo 060-8638, Japan. 8 Second Department of Surgery, Wakayama Medical University, Wakayama 641-8510, Japan. 9 Department of Gastroenterological Surgery, Tokyo Womens Medical University, Tokyo 162-8666, Japan.

10 Genome Analysis Laboratory, Institute of Genome Research, Vietnam Academy of Science and Technology, Hanoi Vietnam. 11 Laboratory of Sequence Analysis, Human Genome Center, The Institute of Medical Science, The University of Tokyo, Tokyo 108-8639, Japan. 12 Division of Cancer Genomics, National Cancer Center, Chuo-ku, Tokyo 104-0045, Japan. 13 Division of Molecular Pathology, National Cancer Center, Chuo-ku, Tokyo 104-0045, Japan.

14 Hepatobiliary and Pancreatic Surgery Division, National Cancer Center, Chuo-ku, Tokyo 104-0045, Japan. 15 Department of Gastroenterological Surgery, Hiroshima University School of Medicine, Hiroshima 734-8551, Japan. 16 Laboratory for Genotyping Development, RIKEN Center for IntegrativeMedical Sciences, Yokohama 230-0045, Japan. 17 Laboratory for Digestive Diseases, RIKEN Center for Integrative Medical Sciences, Hiroshima 734-8551, Japan. * These authors contributed equally to this work. Correspondence and requests for materials should be addressed to H.N.(email: mailto:[email protected]

Web End [email protected] ) or to T.T. (email: mailto:[email protected]

Web End [email protected] ).

NATURE COMMUNICATIONS | 6:6120 | DOI: 10.1038/ncomms7120 | http://www.nature.com/naturecommunications

Web End =www.nature.com/naturecommunications 1

& 2015 Macmillan Publishers Limited. All rights reserved.

DOI: 10.1038/ncomms7120

Whole-genome mutational landscape of liver cancers displaying biliary phenotype reveals hepatitis impact and molecular diversity

ARTICLE NATURE COMMUNICATIONS | DOI: 10.1038/ncomms7120

Primary liver cancer is the fth most common cancer and the third leading cause of cancer death worldwide. Virus infection is the most common and strongest aetiological

factor for liver cancer development. Pathologically, primary liver cancer can be classied into B90% hepatocellular carcinoma (HCC), and 5B10% intrahepatic cholangiocarcinoma (ICC), and the combined hepatocellular cholangiocarcinoma (cHCC/CC), representing only a small portion13. Clinically, ICC and cHCC/ CC show much more aggressive behaviour with poorer prognosis than HCC, and no standard treatment currently exists, other than surgical resection1.

One of the major risk factors for the development of ICC is chronic inammation of the bile ducts, including chronic infections caused by biliary ukes, primary sclerosing cholangitis and hepatolithiasis2,4. Furthermore, recent epidemiological studies recognized that chronic hepatitis associated with viral infection (hepatitis B virus (HBV) and hepatitis C virus (HCV)) is also an important aetiologic factor of ICC, as well as HCC, in Asia2, and indicated that hepatitis-associated ICC and HCC share a common disease process for carcinogenesis5. HCC and ICC have been reported to develop simultaneously in both human and mouse models57 and a combined or mixed phenotype (cHCC/CC) displays intimately mixed components of both hepatocellular and biliary epithelial differentiation8, as shown in Fig. 1a. The presence of these phenotypes indicates the possibility that some of liver cancers can arise from liver progenitor or liver stem cell, although exact cell origins of ICC and cHCC/CC are still controversial and remain to be elucidated6,9. We here dene ICC and cHCC/CC, both of which contain varying degrees of biliary epithelial differentiated cells, as liver cancer displaying biliary phenotype (LCB), distinguishing from HCC phenotype (Fig. 1a). Although several genome analyses of HCC and exome studies of ICC have been recently reported1020, the whole-genomic aberration signature of LCB and its comparison with HCC has yet to be comprehensively explored. In addition, the inuence of aetiological factors, such as chronic hepatitis and virus infection type, on the mutational landscape of primary liver cancer remains unknown and has just begun to be analyzed21.

To elucidate the molecular features of these liver cancer phenotypes, we compared whole-genome sequencing (WGS) data of 30 LCBs and 60 HCCs, and RNA sequencing data for 25 LCBs and 60 HCCs. We examine the WGS to elucidate the substitution pattern and identify driver genes. Finally, we investigate the tumour heterogeneity to nd the genes with clonal mutations by target deep-sequencing analysis. This study demonstrates the rst genome-wide comparison of LCBs with and without chronic hepatitis and characterizes their molecular features.

ResultsSamples and WGS. In this study, frozen tumour and matched normal tissues were collected from 30 patients with LCB (Table 1) and 60 patients with HCC (Supplementary Table 1). All samples were used for WGS. RNA from 25 LCB and 44 HCC samples, from which high-quality RNA was obtained, was also sequenced. Of the LCB samples, 21 were pathologically classied as typical ICC, 2 as rare type of ICC, cholangiocellular carcinoma (CoCC) and 7 as cHCC/CC. Epidemiologically, 9 were from a HCV-related hepatitis background, 7 from HBV-related hepatitis and 14 were not infected with HBV nor HCV. Among them, 10 cases had normal livers with no pathological feature of chronic hepatitis and brosis. RK204 (ICC) and RK209 (HCC) developed metachronously in a single individual as multicentric tumours. Genomic DNA was extracted from the tumour and matched lymphocyte or non-tumour liver samples, and WGS was performed at 40.0x average coverage for tumour samples and

33.0x for matched normal tissue, after removing polymerase chain reaction (PCR) duplicates (Supplementary Table 2). For comparison, we used the somatic mutation data of the 60 HCCs that have been whole-genome sequenced by RIKEN and deposited to the ICGC dataset version 8 released on 2012 March (https://dcc.icgc.org/

Web End =https://dcc.icgc.org/).

Whole-genome mutational landscape of LCBs. We identied point mutations, short indels, copy number alternations, HBV integration sites and somatic rearrangements using custom algorithms (see Methods). We detected between 345 and 180,117 point mutations per tumour. RK308 had an exceptionally large number of somatic mutations (180,117 point mutations), exhibiting a DNA mismatch-repair deciency with a homozygous deletion in the MLH1 gene and a missense mutation (C199R) in the MSH2 gene, which was previously found in Lynch syndrome patient with DNA mismatch-repair deciency and proved to disrupt its function22. Excluding RK308, the average number of nonsynonymous mutations in the 29 LCBs was 26.2, larger than previously reported for leukaemia, but lower than for HCC, lung cancer and melanoma23 (Supplementary Fig. 1, and Supplementary Table 3). The number of detected somatic rearrangements varied greatly among samples (0260) (Supplementary Fig. 2 and Supplementary Dataset 1). Chromothripsis was observed in RK142 and RK316 (Supplementary Fig. 2j and 2ac). HBV integration sites were identied in three LCBs (RK069:cHCC/CC, RK166:cHCC/CC, RK208:ICC) using read-pair information12,14,15 (Supplementary Table 4 and Supplementary Methods), indicating that HBV-infection and its genomic integration can be involved with the carcinogenesis of LCBs24,25.

Gene expression patterns of LCBs and HCCs. To gain molecular insight into the differences between LCBs and HCCs, we examined their gene expression proles (Fig. 1b). RNA-seq analysis on25 LCBs and 44 HCCs, whose high-quality RNAs were available among the 30 LCBs and 60 HCCs, clustered most LCBs into one group, along with some poorly differentiated HCCs, suggesting that LCBs may be similar to poorly differentiated HCCs (Fig. 1b) and that LCBs may have some progenitor feature similar to poorly differentiated HCCs7. Some cHCC/CCs clustered in the LCB group, whereas others were in the main HCC group, which is consistent with their histological combined features. Interestingly, three ICCs with HCC metachronous multicentric tumours were classied within the HCC group (RK204, RK073 and RK137). One CoCC was clustered in the HCC group, which suggests that this particular liver tumour may have an origin similar to that of cHCC/CC26.

Somatic substitution pattern of LCB and HCC. Next we examined somatic substitution patterns of LCBs and compared with those of 60 HCCs (Fig. 2ad). The distribution of genome-wide somatic substitution patterns is signicantly different from random expectation (w2-test; P-valueo10 16). In the LCB genomes, the most predominant substitution was C:G to T:A (odds ratio 2.2, comparison from the assumption of the uniform

mutation rate), followed by T:A to C:G (odds ratio 1.9) and C:G

to A:T (odds ratio 1.3) (Fig. 2b). The distribution of the sub

stitution pattern for LCBs and HCCs was similar (Fig. 2a,b), but the proportion of C:G to T:A was signicantly higher in LCBs (Supplementary Fig. 3). To examine the differences between each sample, principal component analysis (PCA) was applied to the somatic substitution patterns. Although most LCB somatic substitution patterns overlapped the 60 HCC cluster, those of the eight LCBs, all of which developed in livers with no evidence of

2 NATURE COMMUNICATIONS | 6:6120 | DOI: 10.1038/ncomms7120 | http://www.nature.com/naturecommunications

Web End =www.nature.com/naturecommunications

& 2015 Macmillan Publishers Limited. All rights reserved.

NATURE COMMUNICATIONS | DOI: 10.1038/ncomms7120 ARTICLE

RK184

Hepatocellular phenotype

RK194

RK010

Biliary tubular phenotype

ICC 510% cHCC/CC few% HCC 90%

LCB

RK142C

RK146C

RK208C

RK312C

RK310C

RK272C

RK226C

RK184C

RK269C

RK316C

RK194C

RK309C

RK307C

RK308C

RK041C

RK067C

RK084C

RK182C

RK138C

RK159C

RK069C

RK204C

RK012C

RK055C

RK073C

RK075C

RK026C

RK036C

RK014C

RK137C

RK054C

RK002C

RK109C

RK092C

RK050C

RK083C

RK034C

RK166C

RK147C

RK106C

RK031C

RK023C

RK098C

RK003C

RK051C

RK130C

RK141C

RK099C

RK061C

RK100C

RK126C

RK079C

RK019C

RK107C

RK037C

RK108C

RK010C

RK004C

RK209C

RK025C

RK103C

RK112C

RK063C

RK056C

RK020C

RK001C

RK007C

RK058C

RK032C

RK052C

ICC cHCC/CC Diff. HCC

Poorly diff. HCC

CoCC

Figure 1 | LCB phenotype and analysis of the transcriptome pattern on the 30 LCBs and 60 HCCs. (a) Representative pathological images of ICC, cHCC/CC and HCC. RK194 is intrahepatic well-differentiated cholangiocarcinoma. RK184 is cHCC/CC where some sections show moderately or poorly differentiated cholangiocarcinoma (lower) and some show moderately differentiated HCC (upper). RK010 is moderately differentiated HCC. We dene ICC and cHCC/CC, both of which contain varying degrees of biliary tubular-differentiated cells, as liver cancer displaying biliary phenotype (LCB), distinguishing from HCC phenotype. (b) Clustering by the transcriptome of 25 LCBs and 44 HCCs. Hepatitis-negative LCBs are indicated by black dots. ICC, cHCC/CC, CoCC (cholangiolocellular carcinoma), poorly differentiated HCC and differentiated HCC are indicated by coloured rectangles. Three ICCs with metachronous MCTs of HCC are indicated by arrows.

chronic hepatitis, diverged (Fig. 2e and Supplementary Fig. 4). To compare the difference between the HCCs, the hepatitis-positive LCBs and the hepatitis-negative LCBs in the PCA, we performed a permutation test. The difference between the hepatitis-positive and -negative LCBs, and between the hepatitis-negative LCBs and the HCCs were signicantly larger than those from randomly selected samples after the Bonferroni correction (hepatitis-positive and negative LCBs; P-value 0.00116, and the hepatitis-

negative LCBs and the HCCs; P-valueo0.00001).

To compare the impact of chronic hepatitis and inammation on the somatic substitution pattern, we performed PCA with several types of cancer21. In the PCA plot, cancers strongly inuenced by specic mutagens, such as melanoma (UV-exposure) and lung cancer (smoking), were tightly clustered, suggesting that a strong impact of these mutagen exposures on substitution patterns causes a reduction in the divergence among the samples (Fig. 2f). The HCCs, most of which were associated with chronic hepatitis, and the hepatitis-positive LCBs tightly

clustered together, whereas the hepatitis-negative LCBs were more spread out (Fig. 2f and Supplementary Fig. 5). This result indicates that chronic inammation involved with hepatitis strongly inuences the somatic substitution pattern (Fig. 2e, f). In addition, the substitution pattern of the hepatitis-negative LCBs was more similar to the recently reported ICCs18.

To identify the mutational signatures of the hepatitis-positive or -negative LCBs, we used EMu software27 for the 30 LCBs, and ve mutational signatures were detected (Supplementary Fig. 6 and 7). In these signatures, the inuence of signature E, which consists of C4T mutations in CpG sites, differed signicantly between the hepatitis-positive and -negative LCBs, indicating a potential role for methylated cytosines in carcinogenesis related with chronic hepatitis because C4T transitions preferentially occur in methylated CpG sites (Supplementary Fig. 8).

These ndings suggest that the pattern of expressed genes is mainly inuenced by cancer type and reects histological or

NATURE COMMUNICATIONS | 6:6120 | DOI: 10.1038/ncomms7120 | http://www.nature.com/naturecommunications

Web End =www.nature.com/naturecommunications 3

& 2015 Macmillan Publishers Limited. All rights reserved.

ARTICLE NATURE COMMUNICATIONS | DOI: 10.1038/ncomms7120

Table 1 | Clinical and pathological features of 30 LCBs analyzed by whole-genome sequencing.

ID Age Gender Viral infection Histology* TNMw Tumour size (mm) vpz vvy b8 Liver brosisz Note RK067 89 M HCV cHCC/CC T2N0M0 25 3

RK069 85 F ( ) cHCC/CC T3N0M0 80 3

RK073 62 M HBV ICC T1N0M0 10 4 HCC MCT #

RK084 67 F HCV cHCC/CC T3N0M0 30 4

RK108 74 M HCV cHCC/CC T1N0M0 12 2

RK109 84 M HCV CoCC T2N0M0 16 1

RK112 83 M HBV cHCC/CC T3N0M0 25 1

RK137 74 F HCV ICC T3N0M0 24 3 HCC MCT

RK138 75 F ( ) ICC T3N0M0 120 0 Hepatitis-negative**

RK142 57 M ( ) ICC T1N0M0 15 3

RK146 57 F HBV ICC T2N0M0 23 3

RK166 67 M HBV cHCC/CC T3N0M0 25 4

RK182 65 M HCV ICC T3N0M0 28 3

RK184 64 M ( ) cHCC/CC T3N0M0 110 1

RK194 67 M ( ) ICC T3N0M0 35 0 Hepatitis-negative

RK204 83 M HCV ICC T1N0M0 12 4 HCC MCT (RK209)

RK208 60 M ( ) ICC T3N0M0 60 2

RK226 59 M HBV ICC T3N0M0 45 2

RK269 74 M ( ) ICC T1N0M0 12 0 Hepatitis-negative

RK272 78 F ( ) ICC T3N0M0 45 0 Hepatitis-negative

RK279 69 M HCV ICC T3N0M0 35 3

RK298 68 M HCV ICC T3N0M0 40 1

RK303 76 M ( ) CoCC T2N0M0 20 2

RK307 61 F ( ) ICC T3N1M0 75 0 Hepatitis-negative

RK308 70 F ( ) ICC T1N0M0 14 0 Hepatitis-negative

RK309 56 M ( ) ICC T2N1M0 36 0 Hepatitis-negative

RK310 62 F ( ) ICC T3N0M0 90 0 Hepatitis-negative

RK312 66 M HBV ICC T3N1M0 48 0 Hepatitis-negative

RK316 54 F ( ) ICC T3N3M0 54 0 Hepatitis-negative

RK317 73 M HBV ICC T3N0M0 45 2

*ICC (intrahepatic cholangiocellular carcinoma), CoCC (cholangiolocellular carcinoma), cHCC/CC (combined hepatocellular cholangiocellular carcinoma). wTNM staging in UICC.

zvp; portal vein invasion. yvv; hepatic vein invasion.

8b; bile duct invasion.zFibrosis in non-cancerous liver tissue is determined according the New Inuyama Classication.

#MCT; multicentric tumour.**Liver brosis 0 indicates hepatitis-negative LCB.

morphological classication, whereas the somatic substitution pattern in the LCBs is strongly determined by the aetiological background of chronic hepatitis.

Recurrently mutated genes and pathway analysis. We examined recurrently mutated genes in our LCB samples. RK308 had an exceptionally large number of mutations and was excluded from the subsequent analyses. Across the 29 LCB genomes, we detected 892 protein-altering mutations, including 760 nonsynonymous, 108 short coding indels and 24 splice-site mutations (Supplementary Table 3). Thirty-two genes were recurrently mutated (Fig. 3 and Supplementary Table 5): cytoskeleton genes (XIRP2, KIF2B and MYO10), a cell adhesion molecule (CDH2), known tumour suppressors (TP53, PTEN and BAP1), known oncogenes (KRAS and PIK3CA), chromatin regulators (PBRM1, ARID1A and ARID2), which are highly mutated in HCC and other cancers10,12, neuron growth genes (ODZ1, EPHA2 and PLCO) and tyrosine kinase receptors (ERBB4 and EPHA2). To validate the frequency of the mutations, the protein-coding exonic regions of recurrently mutated genes (BAP1, CDH2, EPHA2, KIF2B, MGAT4C, ODZ1, PBRM1, PCLO, SYT1, ARID2 and XIRP2) were amplied in an additional 68 LCB samples (Supplementary Table 6) and sequenced by the Illumina HiSeq2000 sequencer. In addition, as KRAS, IDH1, IDH2 and TERT promoter mutations were frequently observed in ICCs28, uke-related ICCs18 and HCCs29, we sequenced exons 2 and 3 of

KRAS, exon 4 of IDH1 (codon 132), exon 4 of IDH2 (codon 172) and TERT promoter hotspots (chr5:1,295,228 and 1,295,250) in the additional 68 samples by Sanger sequencing. TERT promoter hotspots were also examined in the 30 WGS samples by Sanger sequencing owing to a low depth of coverage in the WGS. This validation experiment revealed that one TERT promoter hotspot (chr5:1,295,228) was mutated in 14 samples (15.2%), KRAS and PBRM1 in 7 samples (10.3%), ARID2 in 5 (7.4%), BAP1, PCLO and IDH1 in 4 (5.9%), ODZ1 in three (4.4%) and EPHA2, SYT2, CDH2, XIRP2 and IDH2 in two samples (2.9%) (Fig. 3 and Supplementary Table 7). In the ARID2, PBRM1 and BAP1 genes, which encode chromatin regulators, an accumulation of loss-of-function mutations was observed, suggesting that they are likely to function as tumour suppressors in LCBs as well as HCCs10,12. As observed in previous HCC studies10,12,13, more than half of LCBs had somatic mutations and rearrangements accumulated in chromatin regulators (Supplementary Fig. 8).

We then examined the frequency of gene mutation and its association with clinical information in the WGS and validation samples (Table 2 and Supplementary Table 8). The frequency of the TERT promoter hotspot mutations was signicantly lower in LCBs than in HCCs (Fishers exact test P-value 1.2 10 9)

(Table 2). Furthermore, the frequency was signicantly higher in cHCC/CCs and HCCs than in the ICCs (Fishers exact test ICCs versus cHCC/CC; P-value 6.5 10 5 and ICCs versus

HCC; P-value 2.1 10 11), but no signicant difference was

observed between cHCC/CCs and HCCs (Table 2). The frequency

4 NATURE COMMUNICATIONS | 6:6120 | DOI: 10.1038/ncomms7120 | http://www.nature.com/naturecommunications

Web End =www.nature.com/naturecommunications

& 2015 Macmillan Publishers Limited. All rights reserved.

NATURE COMMUNICATIONS | DOI: 10.1038/ncomms7120 ARTICLE

HCC (c60) All LCB (n=30)

Hepatitis LCB (n=10)

Hepatitis + LCB (n=10)

T:AtoA:T T:AtoG:C T:AtoC:G C:GtoA:T C:GtoG:C

C:GtoT:A

T:AtoA:T

T:AtoG:C T:AtoC:G C:GtoA:T C:GtoG:C

C:GtoT:A

T:AtoA:T T:AtoG:C T:AtoC:G C:GtoA:T C:GtoG:C

C:GtoT:A

T:AtoA:T T:AtoG:C T:AtoC:G C:GtoA:T C:GtoG:C

C:GtoT:A

0.0 10.0 20.0 30.0 40.0

0.0 10.0 20.0 30.0 40.0 0.0 10.0 20.0 30.0 40.0 0.0 10.0 20.0 30.0 40.0

Proportion (%)

Hepatitis + LCB

Hepatitis+ LCB and HCC

Hepatitis LCB

Melanoma

Lung cancer

6

4

4

2

2

PC2

PC2

0

0

2

Hepatitis + LCB

Hepatitis LCB

HCC

2

4

6 4

2

0

2

4

6

PC1

0 PC1

4

2

2

4

Figure 2 | Genome-wide substitution pattern on the 30 LCBs and 60 HCCs. Average proportion of somatic nucleotide substitutions for (a) the 60 HCCs, (b) the 30 LCBs, (c) the 10 hepatitis-negative LCBs and (d) the 20 hepatitis-positive LCBs. (e) Principal component analysis (PCA) of the whole-genome substitution patterns of the 30 LCBs and 60 HCCs. Hepatitis-positive LCBs (black dots) overlap the HCC cluster (gray). LCBs developed in livers without hepatitis (blue dots) diverged from others. (f) PCA of the whole-genome substitution patterns of the 30 LCBs, the 60 HCCs and other types of cancers21.

Hepatitis-positive liver cancers (HCC and LCBs) are tightly clustered, as are melanomas, indicating that chronic hepatitis or inammation can strongly impact the somatic mutation signature.

TERT promoter KRAS IDH1 or IDH2

PBRM1

ARID2

BAP1 PCLO ODZ1 XIRP2 Liver fibrosis

Virus infection

Histology

Sample ID

Gene WGS (n=29) Validation (n=68)

3 3 3 3 3 3

3 3 3 3 3 3 3 3 3 3 3

2 2 2

2 2 2 2 2 2 2 2 2 2

2

4 4 4 4 4 4

NBNC

1 1 1 1 1 1 1

1

2

0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0

0 0 0

0

0

1 1 1 1 1 1 1 1 1 1 1 1 1 1 1

1

HCV

HCV

NBNC

HCV

HCV

HBV

HCV

NBNC

HCV

NBNC

NBNC

NBNC

HBV

NBNC

HCV

HBV

NBNC

NBNC

NBNC

HBV

NBNC

NBNC

NBNC

HCV

NBNC

HCV

HBV

HBV

HBV

NBNC

NBNC

HCV

HCV

HCV

NBNC

NBNC

NBNC

NBNC

NBNC

NBNC

NBNC

NBNC

NBNC

NBNC

NBNC

NBNC

NBNC

HBV

HBV

HBV

HBV

NBNC

NBNC

HCV

NBNC

HBV

HCV

NBNC

NBNC

NBNC

NBNC

NBNC

HBV

NBNC

NBNC

NBNC

NBNC

NBNC

NBNC

NBNC

NBNC

NBNC

NBNC

NBNC

NBNC

HCV

NBNC

HCV

NBNC

NBNC

NBNC

HBV

NBNC

NBNC

NBNC

NBNC

NBNC

NBNC

NBNC

NBNC

HBV

NBNC

NBNC

NBNC

NBNC

HCV

cHCC/CC

cHCC/CC

cHCC/CC

cHCC/CC

ICC

ICC

ICC

ICC

ICC

CoCC

ICC

ICC

ICC

ICC

ICC

cHCC/CC

ICC

ICC

ICC

ICC

CoCC

cHCC/CC

ICC

ICC

ICC

ICC

ICC

cHCC/CC

cHCC/CC

cHCC/CC

cHCC/CC

ICC

CoCC

cHCC/CC

ICC

ICC

ICC

ICC

ICC

ICC

ICC

ICC

ICC

ICC

ICC

ICC

ICC

ICC

ICC

ICC

ICC

ICC

ICC

ICC

ICC

ICC

ICC

ICC

ICC

ICC

ICC

ICC

ICC

ICC

ICC

ICC

ICC

ICC

ICC

cHCC/CC

ICC

ICC

ICC

ICC

ICC

ICC

ICC

ICC

ICC

ICC

ICC

CoCC

ICC

ICC

ICC

ICC

ICC

cHCC/CC

ICC

ICC

ICC

ICC

ICC

ICC

ICC

cHCC/CC

cHCC/CC

RK067

RK084

RK184

RK108

RK137

RK182

RK317

RK316

RK109

RK307

RK194

RK310

RK312

RK269

RK279

RK166

RK309

RK138

RK226

RK272

RK303

RK069

RK142

RK204

RK208

RK298

RK073

RK112

RK146

OS09

OS02

OS22

OS19

RK347

HK16

OS27

OS01

OS04

OS21

HK98

HK10

HK14

HK22

OS05

OS24

NCC2

OS_NBNC37

OS10

2900

2371

NCC1

HK15

HK08

OS12

OS28

OS29

NCC3

HK101

HK02

HK03

HK04

HK05

HK06

HK07

HK09

HK11

HK12

HK13

OS03

HK17

HK18

HK21

HK24

HK25

OS06

OS07

OS08

OS11

OS13

OS14

OS15

OS16

OS17

OS18

OS20

OS25

OS26

OS23

OS_NBNC21

RK348

RK349

HK99

HK102

2256

HK103

2745

HK100

Missense Nonsense/indel Copy number loss Splice site ND

Figure 3 | Analysis of recurrently mutated genes in LCBs. Recurrently mutated genes in the 30 LCB WGS set and the 68 validation LCB set. Histological subtype of LCB: ICC; intrahepatic cholangiocarcinoma, cHCC/CC; combined hepatocellular-cholangiocarcinoma and CoCC; cholangiolocellular carcinoma. Liver brosis was determined according to the New Inuyama Classication (0B4). Liver brosis 0 indicates hepatitis-negative LCB.

Virus infection status is classied as HBV, HCV or negative, which was determined by serological study.

of mutations in the KRAS and IDH gene hotspots and PBRM1 was higher in the LCBs than in the HCCs (not signicant after the Bonferroni correction) and no mutations were observed in the60 HCCs.

When taking into account the presence of hepatitis, we found signicant difference in the frequency of the TERT promoter hotspot mutations among the hepatitis-positive LCBs, the hepatitis-negative LCBs and the HCCs. The HCCs and

hepatitis-positive LCBs had a higher frequency of the TERT promoter mutations than the hepatitis-negative LCBs (Table 2). In contrast, HCCs had signicantly lower frequency of KRAS and IDH gene hotspot mutations than the hepatitis-negative LCBs, all of which are ICCs (Fishers exact test KRAS; P-value 0.0006,

IDH genes; P-value 0.0006). Hepatitis-positive LCBs shared

more mutated genes with HCCs, which is consistent with the substitution pattern shown in Fig. 2f. However, the frequency of

NATURE COMMUNICATIONS | 6:6120 | DOI: 10.1038/ncomms7120 | http://www.nature.com/naturecommunications

Web End =www.nature.com/naturecommunications 5

& 2015 Macmillan Publishers Limited. All rights reserved.

ARTICLE NATURE COMMUNICATIONS | DOI: 10.1038/ncomms7120

Table 2 | Summary of mutations in WGS and validation set of LCBs.

LCBs HCCs Hepatitis Histology Unadjusted P-value for comparison

WGS (n 29)

Validation (n 68)

WGS (n 60)

Hepatitis-positive

LCBs (n 62)

Hepatitis-negative

LCBs (n 35)

ICC (n 82)

cHCC/ CC(n 15)

LCB versus HCC

Hepatitis-positive LCB versus hepatitis-negative LCB

Hepatitis-negative LCB versus HCC

Hepatitis-positive LCB versus HCC

ICC versus cHCC/CC

ICC versus HCC

cHCC/ CC versus

HCC

12* NS

KRAS 2 (7%) 7 (10%) 0 (0%) 2 (3%) 7 (20%) 9 (11%) 0 (0%) 0.013 0.010 0.00061* NS NS 0.010 NS PBRM1 3 (10%) 7 (10%) 0 (0%) 8 (13%) 2 (6%) 7 (9%) 3 (20%) 0.014 NS NS 0.0062 NS 0.021 0.0067 IDH1/

IDH2

TERT Promoter

9 (30%) 5 (10%) 38 (63%) 12 (20%) 2 (6%) 4 (8%) 8 (53%) 1.2 10

9* NS 1.8 10

8* 2.8 10

6* 6.5 10

5* 2.1 10

2 (7%) 6 (9%) 0 (0%) 1 (2%) 7 (20%) 8 (10%) 0 (0%) 0.024 0.0030* 0.00061* NS NS 0.021 NS

ARID2 2 (7%) 5 (7%) 3 (5%) 4 (6%) 3 (9%) 3 (4%) 4 (27%) NS NS NS NS 0.010 NS 0.026 BAP1 2 (7%) 4 (6%) 1 (2%) 3 (5%) 3 (9%) 6 (7%) 0 (0%) NS NS NS NS NS NS NS PCLO 2 (7%) 4 (6%) 4 (7%) 4 (6%) 2 (6%) 3 (4%) 3 (20%) NS NS NS NS NS NS NS ODZ1 2 (7%) 3 (4%) 0 (0%) 3 (5%) 2 (6%) 4 (5%) 1 (7%) NS NS NS NS NS NS NS XIRP2 5 (17%) 1 (1%) 5 (8%) 4 (6%) 2 (6%) 4 (5%) 2 (13%) NS NS NS NS NS NS NS

*Signicant after the Bonferroni correction.

RK308 had an exceptionally large number of mutations and was excluded from this analysis. Number and frequency (%) of mutations are shown. For TERT promoter, KRAS and IDH genes, hotspot mutations were counted. TERT promoter hotspots (chr5:1,295,228 and 1,295,250) were examined by Sanger sequencing method. CoCCs (n 4) were included in ICCs. P-values were obtained by the

Fishers exact test.

mutations in the PBRM1 gene was different between hepatitis-positive LCBs and HCCs, which may be related to cell differentiation in the liver cancer development30 (marginal signicance after the Bonferroni correction).

Mutations in the KRAS gene were signicantly enriched in patients with lymph node metastasis (Supplementary Table 8), and mutations in the BAP1 gene were signicantly enriched in patients with vascular or bile duct invasion (Supplementary Table 8). Mutations in IDH genes were associated with poor disease-free survival after adjustment for age, which is consistent with the previous study20 (Supplementary Fig. 10).

To identify gene sets and pathways related to the LCB development, we carried out gene set enrichment analysis for all nonsynonymous mutations, short indels and rearrangements31. After adjustment for the multiple testing, 36 categories including synapse organization and cytoskeleton were signicantly overrepresented (Supplementary Table 9). These results suggest that genes in these categories have an important role in the carcinogenesis or cancer development in the LCBs. As axon guidance genes were signicantly mutated in pancreatic cancer32, genes related to neuron growth may have an important role in the carcinogenesis and development of aggressive hepatobiliary-pancreatic cancers, including LCBs.

To determine any possible biological activity of these mutated genes in LCBs, we examined four genes (PCLO, EPHA2, ODZ1 and XIRP2), which are involved in synapse organization and/or cytoskeleton structure. We knocked down the expression of each candidate gene by short interfering RNA in liver cancer cell lines, and examined their proliferation, migration and invasion abilities. These experiments conrmed that silencing of PCLO promoted cell invasion in liver cancer cell lines (Supplementary Figs 11 and 12).

Genetic heterogeneity within liver tumour. A tumour is a population of heterogeneous cancer cells, and the analysis of this heterogeneity should provide us with deeper insights into tumorigenesis12,3336. To examine intratumour heterogeneity, the clonal proportion of the 1,085 nonsynonymous point mutations and short indels, detected in randomly selected 15 LCBs and10 HCCs, were sequenced to an average depth of 56,462x by ultra-deep sequencing (Supplementary Methods). Copy number alternations were adjusted for mutant-allele frequencies using allelic imbalance ratio, and proportion of mutated allele (PMA) was obtained. The distribution of PMA in the ICCs and the cHCC/CCs signicantly differed (Wilcoxons test P-value

0.0047), and ICC genomes had a larger number of mutations

with higher PMA (Supplementary Fig. 13). One possible explanation is that the pattern of genetic heterogeneity is different between cHCC/CC and ICC, and cHCC/CC had larger number of shared mutations in the tumour population than ICC, which is consistent with the histological diversity of cHCC/CC, showing mixed components of both hepatocellular and biliary epithelial differentiation. We then examined the clonal proportion for each gene (Supplementary Dataset 2). Genes in 15 categories, including replicative senescence and negative regulation of DNA replication had a higher PMAs after adjustment for multiple testing (Supplementary Table 10). All categories contained the TP53 gene, indicating that the TP53 gene conferred clonal advantage to cancer cells. This result is consistent with a breast cancer study33. Various genes showed high frequency of mutations in each tumour and would be candidates for tumour initiators (Supplementary Dataset 2).

DiscussionIn the present study, we comprehensively analyzed 30 LCBs by WGS and RNA-seq, and compared their genomic landscapes with those of 60 HCCs. To our knowledge, this is the rst study that demonstrates the impact of chronic hepatitis and inammation on the mutational landscape of the cancer genome and the rst whole-genome comparison between LCBs and HCCs. In our analysis, gene expression patterns are consistent with the histological classications; HCCs and LCBs were differentially clustered and hepatitis-positive and -negative LCBs were clustered together. In contrast, hepatitis-positive and -negative LCBs differentiated in their genome-wide somatic substitution pattern. The hepatitis-positive LCBs clustered more tightly to hepatitis-related HCCs, whereas hepatitis-negative LCBs were more spread out. These results suggest that gene expression depends on the histological phenotype, whereas the somatic substitution pattern is strongly inuenced by aetiological background like the occurrence of chronic hepatitis. Previous studies suggested that the expression pattern is consistent with pathological phenotype, but does not reect tumour origin3739. A mouse study on pancreatic ductal adenocarcinoma suggested that inammation can promote neoplasia by altering cell differentiation38, and a comparison between virus-associated ICCs and HCCs suggested that they can arise from the hepatic progenitor cells5. Considering these studies, the similarities between the hepatitis-positive LCBs and the HCCs in the somatic substitution pattern may indicate their same cellular origin, such as liver progenitor cell. In contrast, hepatitis-negative LCB may arise from different origins, such as cholangiocytes40.

6 NATURE COMMUNICATIONS | 6:6120 | DOI: 10.1038/ncomms7120 | http://www.nature.com/naturecommunications

Web End =www.nature.com/naturecommunications

& 2015 Macmillan Publishers Limited. All rights reserved.

NATURE COMMUNICATIONS | DOI: 10.1038/ncomms7120 ARTICLE

The frequency of some driver mutations, such as hotspot mutations in KRAS, IDH1/2 and TERT promoter, differed among cancer types, hepatitis-positive and -negative LCBs. Mutations of KRAS and IDH genes were more frequent in the hepatitis-negative LCBs, and the TERT promoter mutation was more frequent in the cHCC/CCs and HCCs. As almost all cHCC/CC and HCCs were hepatitis-positive, it is difcult to differentiate the impact of hepatitis from that of the histology. In general, HCC and cHCC/CC, which mainly developed under a hepatitis background, had a larger frequency of TERT promoter mutations and a lower frequency of KRAS and IDH1/2 mutations.

In the current study, we found that the occurrence of chronic hepatitis impacted the mutational landscape, discovered new driver gene and examined intratumour heterogeneity in the LCBs. Our analysis indicates that the WGS can reveal the impact of aetiological background on the genome-wide substitution pattern, and suggest that the WGS can contribute to molecular classication based on their aetiology. However, we did not nd mutations in the driver gene candidates in about a half of the samples, suggesting that LCB is a highly heterogeneous cancer. Analysis of larger number of samples would be necessary for deeper understanding of LCB.

Methods

Clinical samples. The clinical and pathological features of 30 LCBs that were used in WGS analysis are in Table 1. Our pathologists evaluated hematoxylin and eosin-stained slides and diagnosed HCC, ICC and cHCC/CC according to the 2010 WHO Classication of Tumors of the Digestive System41. We dened ICC and cHCC/CC, both of which contain varying degrees of epithelial tubular-differentiated cells (Fig. 1a), as liver cancer displaying biliary phenotype (LCB), distinguishing them from the hepatocellular phenotype (HCC). Viral infection was dened by the presence of HB surface antigen in patients serum, or by the presence of antibody to HCV in patients serum. Hepatitis-negative LCB was dened as a tumour showing no sign of chronic inammation and liver brosis, which was determined according the New Inuyama Classication. All subjects had undergone partial hepatectomy, and pathologists estimated the ratio of viable tumour cells in each sample. High molecular weight genomic DNA was extracted from fresh-frozen tumour specimens and blood. Non-cancerous liver tissues were used as the normal tissue for RK182, RK307, RK308, RK309 and RK310. All subjects agreed with informed consent to participate in the study following ICGC guidelines42. IRBs at RIKEN and all groups participating in this study approved this work.

Whole-genome sequencing. DNA was extracted from tumours and non-cancer frozen tissues, and 500 bp insert libraries were prepared according to the protocol provided by Illumina. The libraries were sequenced on HiSeq2000 platforms with paired reads of 101 bp. The mutation data for the 60 HCCs have been generated in the same way by RIKEN and deposited to the ICGC dataset version 8 released at 2012 March (http://icgc.org/

Web End =http://icgc.org/).

Somatic mutation and short indel calling. Point mutations and somatic indels were identied using our in-house methods12. In brief, read pairs were mapped by BWA43, and the result les were converted to pileup le by samtools44. After PCR duplications were removed and comparing between cancer genome sequences and non-cancer genome sequences, somatic point mutations and indels were identied by our in-house mutation caller12. False-negative and false-positive rates of our analysis pipeline were described previously12. Information for all point mutations and indels in the 30 LCBs and the 60 HCCs was deposited to the ICGC web site (http://www.icgc.org/

Web End =http://www.icgc.org/).

Identication of rearrangements. Inconsistent read pairs which occurred within 500 bp of each other were considered to support the same rearrangement. We identied candidate rearrangements in both tumour (support read pairs Z4) and normal tissue (support read pairs Z1) samples, and tumour-specic rearrangement candidates were identied. To exclude mapping errors, we performed a blast search of read pairs that support rearrangements against the reference genome. If a read pair mapped with correct orientation and distance (r500 bp) with an E-value o10 7, we excluded that read pair. Reads mapped with more than two mismatches were also discarded. After ltering, candidates supported by Z4 read pairs and at least one perfect match pair were considered as somatic rearrangements. The candidates that the same rearrangement was found in other normal samples were ltered out. False discovery rate of this method was estimated to be 2.3% (4/176).

Statistical analysis. The random distribution was calculated by multiplying (proportion of nucleotide in the reference genome sequence) and (total number of mutations) as done in the previous study12. Tests for signicantly mutated genes and PCA of the substitution pattern were carried out as described previously12.

Survival analyses were done using the survival package for the R programming environment (http://www.r-project.org

Web End =http://www.r-project.org). A Cox proportional hazards model was used to test association between disease-free survival and mutations in the genes (TERT promoter, KRAS, XIRP2, ARID2, BAP1, PBRM1, PCLO, ODZ1 and IDH genes) and clinical factors (age, gender, virus type and liver brosis). Model selection was done by the stepAIC function, and the model with age and mutations in IDH genes was selected.

Estimation of PMA was described in the Supplementary Methods.

To test the difference of the clonal proportion of mutations among ICCs, cHCC/CCs and HCCs, we calculated PMA for each mutation, which was standardized by the maximum PMA in each sample. Then we compared the median of the distribution of PMA between ICCs, cHCC/CCs and HCCs by Wilcoxons test.

To identify gene sets with high clonal proportion, we used biological process terms with depth 5 in the Gene Ontology (GO) database (http://www.geneontology.org

Web End =http://www.

http://www.geneontology.org

Web End =geneontology.org ). The clonal proportions of the genes within and outside the gene category were compared by Wilcoxons test as a previous study33. Note that we used unadjusted clonal proportions (not PMAs) for this analysis to consider the inuence of copy number changes.

Sanger sequencing and ultra-deep amplicon sequencing. Sanger sequencing of PCR products was performed on ABI 3770x. For ultra-deep sequencing of mutations, each of the 100 bp target regions was amplied and the amplicons were directly ligated with Illumina TruSeq adaptors and sequenced on HiSeq2000 platform. Mapping was done by BWA to the target region, and uniquely mapped read pairs with proper distance and orientation were selected. More than 98% of the exonic target regions were covered with a depth Z100. We ltered out reads with a mapping quality o10 and base calls with base quality o10. Base calls with a depth Z100 were used for the analysis. We identied variants with frequency Z0.05. Variants found in more than one individual in the 1000 Genome database45 were discarded. We performed Sanger sequencing verication for the predicted candidates in the both cancer and matched normal tissues.

RNA sequencing. RNA-seq was carried out for 25 LCBs and 44 HCCs for which high-quality RNA was available among the 30 LCBs and the 60 HCCs. Total RNA was extracted by Trizol from the frozen liver cancer tissues and the corresponding non-cancerous liver tissues and quality and quantity were evaluated by Bioanalyzer (Agilent). The high-quality RNA was subjected to polyA selection and

chemical fragmentation, and 100200 base RNA fraction was used to construct complementary DNA libraries according to Illuminas protocol. RNA-seq was performed on HiSeq2000 using the standard paired-end 101 bp protocol.

Analysis of RNA sequencing data. First, all sequencing reads were aligned to the known transcript sequences of UCSC known gene database (http://hgdownload.cse.ucsc.edu/goldenPath/hg19/database/knownGene.txt.gz

Web End =http://hgdownload. http://hgdownload.cse.ucsc.edu/goldenPath/hg19/database/knownGene.txt.gz

Web End =cse.ucsc.edu/goldenPath/hg19/database/knownGene.txt.gz ) using Bowtie46, with-a --best --strata -m 20 -v 3 options, and the coordinates of the aligned reads were converted to the human reference sequence (hg19). Then, reads unaligned in the above step were aligned to the human reference sequence (hg19) and as well as HBV sequence (AP011098) using Blat47, with -stepSize 5 -repMatch 2253, and

aligned reads by Bowtie or Blat were combined together. For each short read, the alignment positions with the maximum number of matched bases were adopted, and mapping quality for each read was assigned to as follows: for a location a, let B (a) denote the number of matched bases and let abest denote the best location selected arbitrarily from those with the maximum number of matched bases.

min 100 10 log10 1

1

0

@

0

@

P

a 0:02

Ba Ba

1

A

1

A

Finally, sorting and PCR duplicate removal of short reads were performed by using Picard (http://picard.sourceforge.net/

Web End =http://picard.sourceforge.net/). For quantication of expression values, we used a slightly modied version of RKPM (reads per kilobase of exon per million mapped reads) measures48. After removing improperly aligned or low-quality (mapping quality o60) sequencing reads, the number of bases on each exonic region for each refSeq genes (http://hgdownload.cse.ucsc.edu/goldenPath/hg19/database/refGene.txt.gz

Web End =http://hgdownload.cse.ucsc.edu/goldenPath/hg19/ http://hgdownload.cse.ucsc.edu/goldenPath/hg19/database/refGene.txt.gz

Web End =database/refGene.txt.gz ) were counted. Then, the numbers of bases were normalized as per kilobase of exon and per 100 million of aligned bases. Finally, the expression value of each gene was determined by choosing the maximum of multiple refSeq genes, if any, corresponding to the gene symbol.

References

1. Blechacz, B. & Gores, G. J. Cholangiocarcinoma: advances in pathogenesis, diagnosis, and treatment. Hepatology 48, 308321 (2008).

2. Shin, H. R. et al. Epidemiology of cholangiocarcinoma: an update focusing on risk factors. Cancer Sci. 101, 579585 (2010).

NATURE COMMUNICATIONS | 6:6120 | DOI: 10.1038/ncomms7120 | http://www.nature.com/naturecommunications

Web End =www.nature.com/naturecommunications 7

& 2015 Macmillan Publishers Limited. All rights reserved.

ARTICLE NATURE COMMUNICATIONS | DOI: 10.1038/ncomms7120

3. Sia, D., Tovar, V., Moeini, A. & Llovet, J. M. Intrahepatic cholangiocarcinoma: pathogenesis and rationale for molecular therapies. Oncogene 32, 48614870 (2013).

4. Sripa, B. et al. Liver uke induces cholangiocarcinoma. PLoS Med. 4, e201 (2007).5. Lee, C. H. et al. Viral hepatitis-associated intrahepatic cholangiocarcinoma shares common disease processes with hepatocellular carcinoma. Br. J. Cancer 100, 17651770 (2009).

6. Sekiya, S. & Suzuki, A. Intrahepatic cholangiocarcinoma can arise from Notch-mediated conversion of hepatocytes. J. Clin. Invest. 122, 39143918 (2012).

7. Xu, J. et al. Intrahepatic cholangiocarcinoma arising in chronic advanced liver disease and the cholangiocarcinomatous component of hepatocellular cholangiocarcinoma share common phenotypes and cholangiocarcinogenesis. Histopathology 59, 10901099 (2011).

8. Coulouarn, C. et al. Combined hepatocellular-cholangiocarcinomas exhibit progenitor features and activation of Wnt and TGFbeta signaling pathways. Carcinogenesis 33, 17911796 (2012).

9. Guest, R. V. et al. Cell lineage tracing reveals a biliary origin of intrahepatic cholangiocarcinoma. Cancer Res. 74, 10051010 (2013).

10. Li, M. et al. Inactivating mutations of the chromatin remodeling gene ARID2 in hepatocellular carcinoma. Nat. Genet. 43, 828829 (2011).

11. Totoki, Y. et al. High-resolution characterization of a hepatocellular carcinoma genome. Nat. Genet. 43, 464469 (2011).

12. Fujimoto, A. et al. Whole-genome sequencing of liver cancers identies etiological inuences on mutation patterns and recurrent mutations in chromatin regulators. Nat. Genet. 44, 760764 (2012).

13. Guichard, C. et al. Integrated analysis of somatic mutations and focal copy-number changes identies key genes and pathways in hepatocellular carcinoma. Nat. Genet. 44, 694698 (2012).

14. Jiang, Z. et al. The effects of hepatitis B virus integration into the genomes of hepatocellular carcinoma patients. Genome Res. 22, 593601 (2012).15. Sung, W. K. et al. Genome-wide survey of recurrent HBV integration in hepatocellular carcinoma. Nat. Genet. 44, 765769 (2012).

16. Huang, J. et al. Exome sequencing of hepatitis B virus-associated hepatocellular carcinoma. Nat. Genet. 44, 11171121 (2012).

17. Kan, Z. et al. Whole-genome sequencing identies recurrent mutations in hepatocellular carcinoma. Genome Res. 23, 14221433 (2013).

18. Ong, C. K. et al. Exome sequencing of liver uke-associated cholangiocarcinoma. Nat. Genet. 44, 690693 (2012).

19. Chan-On, W. et al. Exome sequencing identies distinct mutational patterns in liver uke-related and non-infection-related bile duct cancers. Nat. Genet. 45, 14741478 (2013).

20. Jiao, Y. et al. Exome sequencing identies frequent inactivating mutations in BAP1, ARID1A and PBRM1 in intrahepatic cholangiocarcinomas. Nat. Genet. 45, 14701473 (2013).

21. Alexandrov, L. B. et al. Signatures of mutational processes in human cancer. Nature 500, 415421 (2013).

22. Gammie, A. E. et al. Functional characterization of pathogenic human MSH2 missense mutations in Saccharomyces cerevisiae. Genetics 177, 707721 (2007).

23. Vogelstein, B. et al. Cancer genome landscapes. Science 339, 15461558 (2013).24. Raimondo, G., Pollicino, T., Cacciola, I. & Squadrito, G. Occult hepatitis B virus infection. J. Hepatol. 46, 160170 (2007).

25. Tsai, W. L. & Chung, R. T. Viral hepatocarcinogenesis. Oncogene 29, 23092324 (2010).

26. Komuta, M. et al. Clinicopathological study on cholagiolocellular carcinoma suggesting hepatic progenitor cell origin. Hepatology 47, 15441566 (2008).

27. Fischer, A., Illingworth, C. J., Campbell, P. J. & Mustonen, V. EMu: probabilistic inference of mutational processes and their localization in the cancer genome. Genome Biol. 14, R39 (2013).

28. Borger, D. R. et al. Frequent mutation of isocitrate dehydrogenase (IDH)1 and IDH2 in cholangiocarcinoma identied through broad-based tumor genotyping. Oncologist 17, 7279 (2011).

29. Nault, J. C. et al. High frequency of telomerase reverse-transcriptase promoter somatic mutations in hepatocellular carcinoma and preneoplastic lesions. Nat. Commun. 4, 2218 (2013).

30. Romero, O. A. & Sanchez-Cespedes, M. The SWI/SNF genetic blockade: effects in cell differentiation, cancer and developmental diseases. Oncogene 33, 26812689 (2014).

31. Huang da, W., Sherman, B. T. & Lempicki, R. A. Bioinformatics enrichment tools: paths toward the comprehensive functional analysis of large gene lists. Nucleic Acids Res. 37, 113 (2009).

32. Biankin, A. V. et al. Pancreatic cancer genomes reveal aberrations in axon guidance pathway genes. Nature 491, 399405 (2012).

33. Shah, S. P. et al. The clonal and mutational evolution spectrum of primary triple-negative breast cancers. Nature 486, 395399 (2012).

34. Nik-Zainal, S. et al. The life history of 21 breast cancers. Cell 149, 9941007 (2012).

35. Ding, L. et al. Clonal evolution in relapsed acute myeloid leukaemia revealed by whole-genome sequencing. Nature 481, 506510 (2012).

36. Landau, D. A. et al. Evolution and impact of subclonal mutations in chronic lymphocytic leukemia. Cell 152, 714726 (2013).

37. Lim, E. et al. Aberrant luminal progenitors as the candidate target population for basal tumor development in BRCA1 mutation carriers. Nat. Med. 15, 907913 (2009).

38. Gidekel Friedlander, S. Y. et al. Context-dependent transformation of adult pancreatic cells by oncogenic K-Ras. Cancer Cell 16, 379389 (2009).

39. Goldstein, A. S., Huang, J., Guo, C., Garraway, I. P. & Witte, O. N. Identication of a cell of origin for human prostate cancer. Science 329, 568571 (2010).

40. Komuta, M. et al. Histological diversity in cholangiocellular carcinoma reects the different cholangiocyte phenotypes. Hepatology 55, 18761888 (2012).

41. Bosman, F. T. et al. WHO Classication of Tumors of the Digestive System 4th edn, 225227 (IARC, 2010).

42. Hudson, T. J. et al. International network of cancer genome projects. Nature 464, 993998 (2010).

43. Li, H. & Durbin, R. Fast and accurate short read alignment with Burrows-Wheeler transform. Bioinformatics 25, 17541760 (2009).

44. Li, H. et al. The Sequence Alignment/Map format and SAMtools. Bioinformatics 25, 20782079 (2009).

45. Abecasis, G. R. et al. A map of human genome variation from population-scale sequencing. Nature 467, 10611073 (2010).

46. Langmead, B., Trapnell, C., Pop, M. & Salzberg, S. L. Ultrafast and memory-efcient alignment of short DNA sequences to the human genome. Genome Biol. 10, R25 (2009).

47. Kent, W. J. BLAT--the BLAST-like alignment tool. Genome Res. 12, 656664 (2002).

48. Mortazavi, A., Williams, B. A., McCue, K., Schaeffer, L. & Wold, B. Mapping and quantifying mammalian transcriptomes by RNA-Seq. Nat. Methods 5, 621628 (2008).

Acknowledgements

The super-computing resource SHIROKANE was provided by Human Genome Center, The University of Tokyo (http://sc.hgc.jp/shirokane.html

Web End =http://sc.hgc.jp/shirokane.html). This work was supported partially by RIKEN Presidents Fund 2011, the Princess Takamatsu Cancer Research Fund and Takeda Science Foundation.

Author contributions

A.F., Y.Shiraishi, M.F., D.S., T.A., K.A.B., T.T. and H.N. performed data analyses. M.F., K.N., A.S., R.K. and H.N. performed WGS and validation sequencing study. M.F. A.F. and H.H.N. performed functional experiments. H.T., T.Shibuya and S.Miyano operated the super-computer system. Y.K., M.U., K.G., S.A., T.N., T.Shibata, K.A., H.O., K.S., Y.Shigekawa, S.Maruhashi, T.Y., O.I., H.A., H.O., S.H., M.Y., H.Y. and K.C. collected clinical samples and cell lines. A.F., M.F., Y.Shiraishi, K.A.B. and H.N. wrote the manuscript. H.N. conceived the study and led the design of the experiments. A.F., M.K. and H.N. contributed to the funding for this study.

Additional information

Accession codes: WGS data have been deposited in the EGA under the accession codes: EGAN00001187542, EGAN00001187543, EGAN00001187546, EGAN00001187547, EGAN00001187552, EGAN00001187553, EGAN00001187568, EGAN00001187569, EGAN00001187612, EGAN00001187613, EGAN00001187614, EGAN00001187615, EGAN00001187618, EGAN00001187619, EGAN00001187650, EGAN00001187651, EGAN00001187652, EGAN00001187653, EGAN00001187658, EGAN00001187659, EGAN00001187664, EGAN00001187665, EGAN00001187690, EGAN00001187691, EGAN00001187707, EGAN00001187708, EGAN00001187709, EGAN00001187710, EGAN00001187712, EGAN00001187713, EGAN00001187714, EGAN00001187715, EGAN00001187716, EGAN00001187717, EGAN00001187719, EGAN00001187720, EGAN00001187721, EGAN00001187722, EGAN00001187723, EGAN00001187724, EGAN00001187725, EGAN00001187726, EGAN00001187727, EGAN00001187728, EGAN00001187729, EGAN00001187730, EGAN00001187731, EGAN00001187732, EGAN00001187733, EGAN00001187734, EGAN00001187735, EGAN00001187736, EGAN00001187737, EGAN00001187738, EGAN00001187739, EGAN00001187740, EGAN00001187741, EGAN00001187742, EGAN00001187744 and EGAN00001187743 (summarized in Supplementary Table 2).

Supplementary Information accompanies this paper at http://www.nature.com/naturecommunications

Web End =http://www.nature.com/ http://www.nature.com/naturecommunications

Web End =naturecommunications

Competing nancial interests: The authors declare no competing nancial interest.

Reprints and permission information is available online at http://npg.nature.com/reprintsandpermissions/

Web End =http://npg.nature.com/ http://npg.nature.com/reprintsandpermissions/

Web End =reprintsandpermissions/

How to cite this article: Fujimoto, A. et al. Whole-genome mutational landscape of liver cancers displaying biliary phenotype reveals hepatitis impact and molecular diversity. Nat. Commun. 6:6120 doi: 10.1038/ncomms7120 (2015).

8 NATURE COMMUNICATIONS | 6:6120 | DOI: 10.1038/ncomms7120 | http://www.nature.com/naturecommunications

Web End =www.nature.com/naturecommunications

& 2015 Macmillan Publishers Limited. All rights reserved.