Multi-omic characterization of genome-wide abnormal DNA methylation reveals diagnostic and prognostic markers for esophageal squamous-cell carcinoma

Abstract

This study investigates aberrant DNA methylations as potential diagnosis and prognosis markers for esophageal squamous-cell carcinoma (ESCC), which if diagnosed at advanced stages has <30% five-year survival rate. Comparing genome-wide methylation sites of 91 ESCC and matched adjacent normal tissues, we identified 35,577 differentially methylated CpG sites (DMCs) and characterized their distribution patterns. Integrating whole-genome DNA and RNA-sequencing data of the same samples, we found multiple dysregulated transcription factors and ESCC-specific genomic correlates of identified DMCs. Using featured DMCs, we developed a 12-marker diagnostic panel with high accuracy in our dataset and the TCGA ESCC dataset, and a 4-marker prognostic panel distinguishing high-risk patients. In-vitro experiments validated the functions of 4 marker host genes. Together these results provide additional evidence for the important roles of aberrant DNA methylations in ESCC development and progression. Our DMC-based diagnostic and prognostic panels have potential values for clinical care of ESCC, laying foundations for developing targeted methylation assays for future non-invasive cancer detection methods.

Fig. 1. Characteristics of differentially methylated probes in ESCC.

a The proportion of all filtrated CpG sites that are differentially methylated or not methylated. b The proportion of differentially hyper-methylated (red) or hypo-methylated (blue) CpG islands. c, d The proportion (c) and odds ratio (d) of hyper-methylated (red) or hypo-methylated (blue) CpG sites in different chromosomes. e, f The category (e) and odds ratio (f) of genomic locations for hyper-methylated (red) or hypo-methylated (blue) sites. g, h The proportions (g) and odds ratio (h) of methylated CpG sites correlated with the expression levels of genes in different chromosomes. i, j The category (i) and odds ratio (j) of genomic locations for CpG sites correlated with genes expression. Odds ratio was computed against the general distribution and P value was computed by Hypergeometric test. Island, CpG island; shore, 0–2 kb from CpG island; shelf, 2–4 kb from CpG island; open sea, other genomic regions; TSS1500, 200–1500 bases upstream of the transcriptional start site (TSS); TSS200, 0–200 bases upstream of the TSS; 5’UTR, within the 5’ untranslated region and between the TSS and the ATG start site; body, between the ATG and stop codon regardless the presence of introns, exons, TSS or promoters; 3’UTR, between the stop codon and poly A signal

Fig. 2. Integrative analysis of whole-genome DNA and RNA-sequencing data uncovered methylation-mediated dysregulation of multiple TFs in ESCC.

a, b The association between promoter or gene-body methylation and host gene expression were identified. There are four clusters: genes (n = 90) that are hyper-methylated in promoter with low expression in ESCC; genes (n = 44) that are hypo-methylated in promoter with high expression; genes (n = 70): that are hyper-methylated in gene-body with high expression; genes (n = 274) that are hypo-methylated in gene-body with low expression. Number of known TFs are shown in each cluster. P value was computed by Hypergeometric test. c Differential expression (top) and promoter methylation (bottom) levels of ZNF382 in ESCC and normal samples. d, e Differential expression (top) and gene-body methylation (bottom) levels of HOXB13 (d) and DLX1 (e) in ESCC and normal samples. f The correlation between mRNA expression and promoter DNA methylation levels of ZNF382. g, h The correlation between mRNA expression and gene-body DNA methylation levels of HOXB13 (g) and DLX1 (h). P of Student’s t test for gene expression and Wilcoxon signed-rank test for methylation. Genes mRNA expression level (RSEM) was added by 1 and then log2 transformed. Dotted short line indicates mean expression level of each gene

Fig. 3. Hyper- and hypo-methylation events across ESCC and integrated profiling of ESCC driver genes.

a Map overview of genetic and epigenetic alterations in 20 ESCC driver genes previously identified. ach column denotes an individual patient and each row represents the status of one gene including somatic mutations (black squares), copy number amplifications (red bars), copy number deletions (blue bars), hyper- (pink bars) and hypo-methylated events (azure bars). Wild-type cases are in gray. Right, percentage of alterations for each gene in 91 ESCC patients while the X axis represents total percentage of alterations for each gene. b, c We tested recurrent genetic alterations in ESCC for their associations with frequency of hyper- (b) or hypo-methylated event (c). Significant associations (Wilcoxon P < 0.05) were shown in above and labeled by gene symbol for somatic mutations or cytoband for amplifications and deletions. Each column denotes an individual patient and each row is one genetic alteration including somatic mutations (black bars), copy number amplifications (red bars) and copy number deletions (blue bars). Wild-type cases are in gray. Top color bars represent the frequency of DNA methylation

Fig. 4. Diagnosis of ESCC with a DNA methylation panel.

a–c The confusion tables of binary results of diagnostic prediction model in the training (a), validation (b) and TCGA ESCC (c) datasets. d–f The receiver operating characteristic curve (ROC) of the diagnostic prediction model in the training (d), validation (e) and TCGA ESCC (f) datasets. g−i Unsupervised hierarchical clustering and heatmap of 12 methylation markers screened for constructing the diagnostic prediction model in the training (g), validation (h) and TCGA ESCC (i) datasets

Fig. 5. The correlation of the methylation signature and survival time in patients with ESCC.

a, b Kaplan–Meier survival curves for all our patient sample (a) and all TCGA ESCC patient sample (b). c, d Kaplan–Meier survival curves for patients with early stage ESCC in our sample (c) and in TCGA ESCC sample (d). e, f Kaplan–Meier survival curves for patients with advanced stage ESCC in our sample (e) and in TCGA ESCC sample (f). High- or low-risk group was defined by the weighted hazard ratios of the 4 methylation sites in patients. The P value was calculated by log rank test. HR and 95% CI was computed with Cox hazard proportion model

Fig. 6. Effects of silencing some genes in diagnostic and prognostic panels on ESCC cell phenotypes.

a–d Silencing the expression of MMP13 (a), YEATS2 (b), HOXC10 (c) and NECAB2 (d) significantly suppressed KYSE30 and KYSE150 cell proliferation. e–h Silencing the expression of MMP13 (e), YEATS2 (f), HOXC10 (g) and NECAB2 (h) significantly suppressed KYSE30 and KYSE150 cell migration and invasion. Left panel shows representative cell migration and invasion images and right panel shows quantification statistics. Data represent mean ± SEM from 3 independent experiments. *P < 0.05; **P < 0.01; ***P < 1.00e-3; ****P < 1.00e-4 and ns not significant of Student’s t test compared with corresponding control