Recurrently deregulated lncRNAs in hepatocellular carcinoma

Abstract

Hepatocellular carcinoma (HCC) cells often invade the portal venous system and subsequently develop into portal vein tumour thrombosis (PVTT). Long noncoding RNAs (lncRNAs) have been associated with HCC, but a comprehensive analysis of their specific association with HCC metastasis has not been conducted. Here, by analysing 60 clinical samples' RNA-seq data from 20 HCC patients, we have identified and characterized 8,603 candidate lncRNAs. The expression patterns of 917 recurrently deregulated lncRNAs are correlated with clinical data in a TCGA cohort and published liver cancer data. Matched array data from the 60 samples show that copy number variations (CNVs) and alterations in DNA methylation contribute to the observed recurrent deregulation of 235 lncRNAs. Many recurrently deregulated lncRNAs are enriched in co-expressed clusters of genes related to cell adhesion, immune response and metabolic processes. Candidate lncRNAs related to metastasis, such as HAND2-AS1, were further validated using RNAi-based loss-of-function assays. Thus, we provide a valuable resource of functional lncRNAs and biomarkers associated with HCC tumorigenesis and metastasis.

Figure 1. Identification of candidate lncRNAs in 60 HCC samples.

(a) Overview of the comprehensive experimental and computational scheme for the systematic identification of lncRNAs in samples from HCC patients. (b) Venn diagram showing the overlap between newly assembled lncRNAs and MiTranscriptome lncRNAs. Characterization of GENCODE lncRNAs, newly assembled lncRNAs and protein-coding genes: (c) transcript length distribution; (d) cumulative distribution curve of maximum gene expression level (RPKM); (e) conservation of exons and introns; (f) enrichment of GWAS SNPs (circle) over randomly selected SNPs (triangle).

Figure 2. Recurrently deregulated lncRNAs in primary tumours and PVTTs.

Identification of recurrently deregulated lncRNAs: recurrently downregulated and upregulated lncRNAs that were predicted by three statistical methods to be associated with (a) tumorigenesis and (b) metastasis. Fold-change of expression in each individual patient for (c) recurrently deregulated tumorigenesis-associated lncRNAs; (d) recurrently deregulated metastasis-associated lncRNAs (tumour versus PVTT). Patient I was not included in the analyses related to metastasis because the PVTT sample of patient I was contaminated. Stacked bar charts showing examples of recurrently deregulated lncRNAs, including tumorigenesis-associated (e) and metastasis-associated (f) lncRNAs. The number on the y axis is the number of patients with differential expression of each lncRNA.

Figure 3. Association of recurrently deregulated lncRNAs with TCGA clinical data and other published data.

(a) Gene set enrichment analysis (GSEA) of recurrently deregulated tumorigenesis-associated lncRNAs based on TCGA LIHC data and published liver cancer data. lncRNAs were rank-ordered by differential expression between adjacent normal tissue and primary tumour samples. (b) GSEA of recurrently deregulated metastasis-associated lncRNAs. lncRNAs were rank-ordered by differential expression between primary tumours with and without vascular invasion in the TCGA LIHC data, as well as by differential expression between primary tumours and PVTTs in published liver cancer data. (c) Kaplan-Meier analysis of overall survival in the TCGA LIHC cohort. Subjects were stratified according to the expression of lncRNA RP11-166D19.1. The P value for Kaplan-Meier analysis was determined using log-rank test. (d) Multivariate analysis using additional clinical information. Forest plot depicting correlations between the indicated clinical criteria and the expression level of RP11-166D19.1. (e) Expression levels (TCGA data) of RP11-166D19.1 in three HCC subclasses (S1, S2 and S3). ***P value<0.001, Wilcoxon rank-sum test.

Figure 4. Regulatory mechanisms for recurrently deregulated lncRNAs.

(a) Summary of the regulatory mechanisms of recurrently deregulated lncRNAs. The numbers of lncRNAs associated with CNV and/or DNA methylation data are listed. Bimorphic lncRNAs were those that were recurrently upregulated in some patients and recurrently downregulated in other patients. The total number of lncRNAs is not equal to the sum of each type because of overlapping sub-types. (b) Chromosomal view of amplification and deletion peaks between primary tumours and normal tissue. The G-scores (top) and FDR q-values (bottom) of peaks were calculated using GISTIC2.0. The G-score considered the amplitude of the aberration and its frequency of occurrence across all samples. The q-value was calculated for the observed gain/loss at each locus using randomly permuted events as a control. Examples of recurrently deregulated lncRNAs located in the peaks (only found in deletions) are labelled. (c) Scatterplots showing recurrently deregulated lncRNAs (colour labelled) that were putatively affected by alterations in DNA methylation. Recurrently deregulated lncRNAs driven by DNA methylation had expression levels that were inversely correlated with DNA methylation levels at their promoter regions (PCC, Pearson correlation coefficients<−0.3, x axis). (d) Example of a recurrently deregulated lncRNA driven by DNA methylation; the HAND2-AS1 expression level (FPKM) was inversely correlated with its promoter methylation level (beta value). Boxplot showing that the beta values of primary tumour samples were significantly higher than those of normal tissue samples, but slightly lower than those of PVTT samples.

Figure 5. Inference of potential functions for recurrently deregulated lncRNAs using a co-expression network.

(a) Network representation of 18 selected inter-connected clusters in the coding-non-coding co-expression network. (b) GO and KEGG pathway enrichment for four selected clusters (4, 9, 18 and 25). Heatmap showing enrichment scores (−log₁₀(P value)) for GO terms and KEGG pathways in four selected clusters. The most significantly enriched GO terms and KEGG pathways are displayed. (c) Sub-network showing important genes/lncRNAs in cluster 25. The subnetwork depicts the relationships among four lncRNAs and liver cancer-related driver genes.

Figure 6. Loss-of-function assay of candidate lncRNAs regulating cell migration.

Transwell migration assays were conducted to test the effects of siRNA-mediated RNAi of candidate lncRNAs in three liver cancer cell lines: HepG2, SMMC-7721 and HCCLM9. (a) The value in the heatmap is the fold-change (P value<0.05) of the transwell cell numbers for knockdown cells over those of control cells. All results are expressed as the mean derived from three independent experiments. The unpaired Student's t-test (two-tailed) was used for comparisons of two groups. Seven of the ten candidate lncRNAs were metastasis-associated lncRNAs that were recurrently deregulated in PVTTs (in bold font). *P value<0.05, **P value<0.01, ***P value<0.001, t-test, n=3. (b) RNAi was validated by qRT-PCR. Examples of migration phenotype: transwell cells (DAPI staining) (c) (Scale bar, 500 μm) and their counts (d). Error bars represent the s.d. of three experiment replicates.