Comprehensive and Integrative Genomic Characterization of Hepatocellular Carcinoma

Abstract

Liver cancer has the second highest worldwide cancer mortality rate and has limited therapeutic options. We analyzed 363 hepatocellular carcinoma (HCC) cases by whole-exome sequencing and DNA copy number analyses, and we analyzed 196 HCC cases by DNA methylation, RNA, miRNA, and proteomic expression also. DNA sequencing and mutation analysis identified significantly mutated genes, including LZTR1, EEF1A1, SF3B1, and SMARCA4. Significant alterations by mutation or downregulation by hypermethylation in genes likely to result in HCC metabolic reprogramming (ALB, APOB, and CPS1) were observed. Integrative molecular HCC subtyping incorporating unsupervised clustering of five data platforms identified three subtypes, one of which was associated with poorer prognosis in three HCC cohorts. Integrated analyses enabled development of a p53 target gene expression signature correlating with poor survival. Potential therapeutic targets for which inhibitors exist include WNT signaling, MDM4, MET, VEGFA, MCL1, IDH1, TERT, and immune checkpoint proteins CTLA-4, PD-1, and PD-L1.

Keywords: IDH1/2; TP53; cancer subtyping; expression profile; hepatocellular carcinoma; metabolic reprogramming; promoter hypermethylation; significantly mutated genes; sonic hedgehog signaling; stem cell phenotype.

Conflict of interest statement

There were no identified conflicts of interest by any of the authors.

Copyright © 2017 Elsevier Inc. All rights reserved.

Figures

Figure 1. The genomic landscape of liver hepatocellular carcinoma and mutational signatures

Top panel shows individual tumor mutation rates while the middle panel details ethnicity, tumor grade, age, gender, hepatitis C virus (HCV) and hepatitis B virus (HBV) infection status, and cirrhosis for 363 HCC. Bottom panel shows genes with statistically significant levels of mutation (MutSig suite, false discovery rate, 0.1) and mutation types are indicated in the legend at the bottom. The bottom six rows display significant DNA copy number alterations in likely cancer driver genes.

Figure 2. Liver cancers show distinct gene hypermethylation patterns

(A) Unsupervised clustering analysis of gene hypermethylation in HCC relative to normal tissue reveals four distinct subgroups. Roughly 15,000 CpG sites showing significant hypermethylation in 196 HCC were analyzed and are shown in heat map format with normal tissues and tumors organized in columns according to cluster designation. Intensity of methylation for each CpG site is indicated by row. Above the heat map the four distinct hypermethylation clusters are shown, and below are bars indicating the distribution of clinical and molecular attributes of the individual tumors by cluster. To the right, P values indicate significant non-random distributions for each attribute. (B–I) Scatter plots of representative CpG sites in gene promoters shown to be frequently hypermethylated in HCC, where gene RNA expression (y axis) is plotted against relative promoter site hypermethylation (x axis). Gray dots are results from tumor samples, blue dots normal tissues, and red dots tumors with mutations in the gene. (B) CDKN2A (cg13601799). (C) HHIP (cg23109129). (D) PTGR1 (cg13831329). (E) TMEM106A (cg21211480). (F) MT1M (cg15134649). (G) MT1E (cg02512505). (H) CPS1 (cg21967368).

Figure 3. Multiplatform clustering analysis identified three integrated molecular subtypes of liver cancer

(A) Heat maps organized by iCluster groupings for DNA copy number, DNA methylation status, mRNA expression, and miRNA expression, and correlated with selected molecular features (top tracks). Tumors are in columns, grouped by the iCluster membership. (B) Relative proportions in each iCluster of Hoshida et al. (2009) subtypes defined by RNA expression profiling of a separate HCC cohort. (C) Patient survival outcome fitting three external clinically annotated HCC patient cohort sets of RNA expresson data to the TCGA iClusters (NCI, Lee et al. 2006; Fudan, Roessler et al. 2010; MDACC, Sohn et al. 2016). See also Figure S3.

Figure 4. HCC with IDH1/2 mutations and with IDH-like gene expression share miRNA and RNA expression profiles and worse clinical outcomes

(A) Integrated analysis of IDH1/2 mutations (bottom), mRNA and miRNA expression data (middle), and iCluster and molecular subtypes of HCC (top). HS, hepatic stem cell subtype; CCL, cholangiocarcinoma-like subtype; Hippo, Hippo pathway subtype; RS65, 65-gene risk score subtype; NCIP, National Cancer Institute proliferation subtype; SNUR, Seoul National University recurrence subtype; HB16, 16-gene hepatoblastoma subtype; Hoshida, HCC RNA expression subtype profiling category. (B) Comparison of mRNA expression profiles of two TCGA HCC cohorts and four other HCC cohorts showing subsets of tumors with IDH-like gene expression. (C) Clinical significance of IDH-like subtype in HCC. Patients in three external cohorts were stratified according to IDH-like gene expression signature. See Figure S4.

Figure 5. P53-induced gene target expression signature for improved clustering of HCC molecular and biological attributes

(A) Clustering of 191 HCC by composite expression of known p53 target genes. An expression heat map of 20 p53-induced target genes is shown above that of 10 p53-repressed target genes. To the right are shown mean expression ratios of top quartile p53 target genes relative to bottom quartile. Asterisks indicate level of significance. ***PTP53 mutation. Top bars show molecular and clinical attributes and correlation (p values) with high and low p53 target gene expression. MDM4 copy number and expression are significantly increased in those HCC with wildtype TP53 and with low p53 target expression relative to all other HCC (p values with asterisks). (B) Frequent copy number amplification of MDM4 gene in HCC. A segment of chromosome 1 centered on the MDM4 locus (in black box) is shown. The intensity of red bars corresponds to degree of copy number gain. Each horizontal line corresponds to a single tumor. (C) MDM4 copy number gain and amplification correlates with increased RNA expression. RNA expression for each tumor is represented by a red dot (mutant TP53) or blue dot (WT TP53) according to MDM4 copy number (−1 = deletion, 0 = diploid, 1 = copy number gain, 2 = amplification).

Figure 6. Integrated molecular comparison of somatic alterations in signaling pathways across iCluster groups

Each gene box includes 3 percentages representing the frequency of activation or inactivation in iCluster 1, 2 and 3 based on the core 196 sample HCC dataset. All somatic changes are tallied together in calculating the percentages of altered cases within each of the iCluster sample groups. Somatic alterations include mutations and copy-number changes (homozygous deletion and high-level amplifications), as well as epigenetic silencing of CDKN2A. Missense mutations are only counted if they have known oncogenic function, have been reported in COSMIC, or occur at known mutational hotspots. Genes are grouped by signaling pathways, with edges showing pairwise molecular interactions.

Figure 7. Characterization of LIHC immune microenvironment using RNA-seq data

(A) Unsupervised hierarchical clustering of gene expression identifies immune profiles within HCC patients. Sixty-six manually curated immune cell markers were used for clustering. (B) The CIBERSORT-inferred relative fractions of different immune cell types varied across tumor and tumor adjacent normal samples and were not associated with virus status. (C) CIBERSORT cellular composition analysis revealed striking differences in relative compositions of immune cell populations between tumor and tumor-adjacent normal tissues. P values were calculated by Wilcoxon rank-sum test and adjusted for multiple testing (q value). The red dotted lines on the y axis indicate q value of 0.01. The red dotted lines on × axis indicates Z score of 0. The analysis was performed for all CIBERSORT immune cell types but only the significant ones are labeled on the plot.