Kidney cancer is characterized by aberrant methylation of tissue-specific enhancers that are prognostic for overall survival

Abstract

Purpose: Even though recent studies have shown that genetic changes at enhancers can influence carcinogenesis, most methylomic studies have focused on changes at promoters. We used renal cell carcinoma (RCC), an incurable malignancy associated with mutations in epigenetic regulators, as a model to study genome-wide patterns of DNA methylation at a high resolution.

Experimental design: Analysis of cytosine methylation status of 1.3 million CpGs was determined by the HELP assay in RCC and healthy microdissected renal tubular controls.

Results: We observed that the RCC samples were characterized by widespread hypermethylation that preferentially affected gene bodies. Aberrant methylation was particularly enriched in kidney-specific enhancer regions associated with H3K4Me1 marks. Various important underexpressed genes, such as SMAD6, were associated with aberrantly methylated, intronic enhancers, and these changes were validated in an independent cohort. MOTIF analysis of aberrantly hypermethylated regions revealed enrichment for binding sites of AP2a, AHR, HAIRY, ARNT, and HIF1 transcription factors, reflecting contributions of dysregulated hypoxia signaling pathways in RCC. The functional importance of this aberrant hypermethylation was demonstrated by selective sensitivity of RCC cells to low levels of decitabine. Most importantly, methylation of enhancers was predictive of adverse prognosis in 405 cases of RCC in multivariate analysis. In addition, parallel copy-number analysis from MspI representations demonstrated novel copy-number variations that were validated in an independent cohort of patients.

Conclusions: Our study is the first high-resolution methylome analysis of RCC, demonstrates that many kidney-specific enhancers are targeted by aberrant hypermethylation, and reveals the prognostic importance of these epigenetic changes in an independent cohort.

Figure 1. Methylation profiling separates RCC from normals

Methylation profiles generated by the HELP assay were used to cluster 13 RCC and 13 control samples by unsupervised hierarchical clustering (1-Pearson correlation distance and Ward's agglomeration method). Heatmap shows hypermethylation (in green) and hypomethylation (in red) based on HpaII/MapI ratios. The controls formed a cluster that was distinct from RCC samples. No correlation with clinical characteristics was seen. Female gender, age 50-70 years old, diabetes, hypertension, and VHL mutation were represented in gray. Male gender, age <50 years old, no diabetes, no hypertension, no VHL deletion represented by light grey boxes. Age >70 years of age were represented by black boxes. (A). Unsupervised clustering based on gene expression profiles reveals transcriptional differences between controls and RCC samples (B).

Figure 2. The majority of differentially methylated loci are hypermethylated in RCC and reside outside of CpG islands

A volcano plot is shown demonstrating the difference in mean methylation between all RCC samples and controls on the x axis and the log of the p values between the means on the y axis. A two tailed T test was used to calculate the p values. Significantly methylated loci (FDR < 0.1 by multiple testing) with a log fold change in mean methylation are labeled in red and significantly hypomethylated loci are labeled in red (A). Volcano plots for gene body, promoter, and CpG shore loci also reveal mostly hypermethylated loci with highest percentages in gene bodies and CpG shores. (B, C, D). Percentage of differentially methylation loci reveals significantly more hypermethylation genome wide (TTest, P value < 0.001) (E). Difference of mean gene expression between RCC and controls is shown as box plots for hypermethylated and hypomethylated loci for all genomic locations. Hypermethylation is significantly associated with decrease in gene expression (P values<0.01). (F).

Figure 3. Differential methylation in RCC preferentially occurs at enhancer regions associated with H3K4Me1 marks

Intragenic, differentially methylated regions (DMRs) were significantly located in intronic regions (P value<0.05 when compared to location of the whole array, Test of Proportions) (A). Overlap with adult kidney ChIP-seq data shows significant overlap of H3K4me1 marks with the DMRs (P value<0.05, Test of Proportions when compared to overlap with whole array).(B). Aberrantly methylated regions in the SMAD6 (C) locus overlap with the kidney H3K4me1 peaks. Methylation values generated by the HELP assay (log(HpaII/MspI) are shown as peaks (positive values corresponding with less methylation are shown in black and negative values corresponding to increased methylation are shown in grey). Histone modification peaks are shown for adult kidney (BI27). Chromatin occupancy for other tissue types are shown at bottom. Legend shows color coding for different regions with yellow regions representing enhancers. GM12878: Lymphoblastoid cell H1-hESC: Embryonic stem cells, K562: Leukemic cells, HepG2: Hepatic cells, HUVEC: Umbilical vein endothelial cells, HMEC: Mammary epithelial cells, HSMM: Skeletal muscle myoblasts, NHEK: Epidermal keratinocytes, NHLF: Lung fibroblasts.

Figure 4. Aberrant methylation of intronic regions and reduced expression of SMAD6 is seen in independent cohort of RCC samples and is predictive of adverse prognosis

Methylation analysis of 4 intronic probes (arrows) from 405 RCC samples and controls reveals increased methylation in RCC samples (A). SMAD6 expression was assessed in CAKI cells after 5 days of exposure to 0.5uM 5-azacytidine. Mean fold change over control is shown for two independent experiments, TTest, P Value =0.02 (B). SMAD6 expression in this cohort is also significantly decreased in RCC samples when evaluated by RNA-seq (C). Kaplan-Meier survival curves of overall survival of 405 RCC patients were plotted. Blue solid lines represent OS of patients with a higher expression of SMAD6 (top 20%tile), while red dotted lines represent OS of patients with a lower expression (lower 20%tile) (Log rank P value 0.0002) (D).

Figure 5. Aberrant enhancer hypermethylation is prognostic in large independent cohort of patients and RCC is sensitive to low doses of DNMT inhibitors

Heat map of the respective patients (horizontal order) and the top 500 enhancer loci (vertical order). Patients are ranked in descending order based on the methylation. (A) Kaplan-Meier survival curves of overall survival of 405 RCC patients were plotted. Blue solid lines represent OS of patients with a higher methylation (top 20%tile) of kidney specific enhancers, while red dotted lines represent OS of patients with a lower methylation (lower 20%tile) (Log rank P value 0.03 for univariate and 0.03 for multivariate analysis) (B). RCC cell lines (769-P, 786-O) and healthy kidney cells (HK-2, HKC-8) were grown for 5 days in the presence and absence of .5uM Decitabine and BrdU incorporation was measured and normalized. Significant growth inhibition was seen in RCC cells (Two tailed TTest, p value< 0.05) (C)

Figure 6. MspI representations can detect copy number variations in RCC that are validated in an independent cohort

Copy-number variation (CNV) information can be detected through the MspI representation in the HELP assay. An ideogram with deletions represented in green and amplifications in red of all 13 RCC patients is shown (A). An example of deletion of LATS (B) and amplification of PDGF-α (B) is shown for control and RCC samples using MspI representations. Validation by qPCR shows amplification of PDGF-α (D) and deletion of LATS gene (E) is seen in RCC when compared to pooled normal controls. PDGF-α is found to be amplified in 151/405 cases and LATS is found to be deleted in 121/405 cases in the TCGA dataset (F). RNA-seq expression data reveals overexpression of PDGF-a and reduced expression of LATS in TCGA RCC samples when compared to non-tumor controls (n=68) as shown by heatmaps (G).