DNA methylation of distal regulatory sites characterizes dysregulation of cancer genes.
Aran D, Sabato S, Hellman A. · 2013
Read DNA methylation of distal regulatory sites characterizes dysregulation of cancer genes. by Aran D, Sabato S, Hellman A. (2013) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/europmc-pmc4053839.txt.
Introduction
Dvir Aran, Sivan Sabato, and Asaf Hellman show that methylation at distant enhancer sites tracks gene activity across the genome, in Genome Biology in 2013. Methylation near gene starts was the classic epigenetic signal. The control sites far from genes, the enhancers, had barely been studied in this role.
Working from methylation and expression data across dozens of cell types, the team modelled how promoter methylation relates to activity, then extended the logic genome-wide. They mapped many far-off sites whose methylation matched the activity of distant genes. These sites carry the marks of a known enhancer class and bind their factors in a methylation-sensitive way. One enhancer could set a gradient of activity across many cell types, not just an on-off state. In cancers the pattern broke hard: enhancer methylation shifts lined up with cancer gene misregulation more tightly than promoter shifts did.
Catalog record
| CREATOR | Aran D, Sabato S, Hellman A. |
|---|---|
| YEAR | 2013 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 59 KB |
| LICENSE | CC BY 4.0 |
| RIGHTS | CC BY |
| SOURCE | Europe PMC |
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
DNA methylation of distal regulatory sites characterizes dysregulation of cancer genes By Dvir Aran, Sivan Sabato, Asaf Hellman Published in: Genome Biology DOI: 10.1186/gb-2013-14-3-r21 ABSTRACT Abstract Background Abnormal epigenetic marking is well documented in gene promoters of cancer cells, but the study of distal regulatory siteshas lagged behind.We performed a systematic analysis of DNA methylation sites connected with gene expression profilesacross normal and cancerous human genomes. Results Utilizing methylation and expression data in 58 cell types, we developed a model for methylation-expression relationships in gene promoters and extrapolated it to the genome. We mapped numerous sites at which DNA methylation was associated with expression of distal genes. These sites bind transcription factors in a methylation-dependent manner, and carry the chromatin marks of a particular class of transcriptional enhancers. In contrast to the traditional model of one enhancer site per cell type, we found that single enhancer sites may define gradients of expression levels across many different cell types. Strikingly, the identified sites were drastically altered in cancers: hypomethylated enhancer sites associated with upregulation of cancer-related genes and hypermethylated sites with downregulation. Moreover, the association between enhancer methylation and gene deregulation in cancerwas significantly stronger than the association of promoter methylationwith gene deregulation. Conclusions Methylation of distal regulatory sites is closely related to gene expression levels across the genome. Single enhancers may modulate ranges of cell-specific transcription levels, from constantlyopen promoters. In contrast to the remote relationships between promoter methylation and gene dysregulation in cancer, altered methylation of enhancer sites is closely related to gene expression profiles of transformed cells. Keywords: Cancer, DNA methylation, distal control elements, enhancers, epigenomics, gene-enhancer pairing, gene regulation, machine-learning BACKGROUND DNA methylation is a key determinant of regulatory chromatin complexes transmitted through cell divisions. Relationships between DNA methylation and gene expression levels were first recognized at gene promoters and CpG islands [1], but were recently observed across the genome [2-6]. In gene promoters, DNA methylation mediates silencing by altering the binding of transcription modulators to their DNA targets [7], but the cause and function of expression-associated methylation away from gene promoters has been elusive. The relationships between aberrant DNA methylation and the altered expression profiles of cancer cells are also not well understood: the predominant pattern in gene promoters is denovo methylation of polycomb-repressed genes [8-11]. Since these genes are already inactive in the normal tissue and generally remain inactive in the cancer [12,13], it is hard to establish direct contribution to the cancer process [14]. Away from gene promoters, perturbed methylation has been associated with reduced genomic stability [15,16] and global silencing of large chromatin domains [17,18], but the effect on transcription of particular genes remains unknown. Transcriptional enhancers support tissue-specific expression profiles through physical interactions with gene promoters. Based on analyses of chromatin structures, hundreds ofthousands of transcriptional enhancers were predicted in mammalian genomes (reviewed in [19]). These sites bind chromatin-modulating factors, interact with distal promoters through DNA loops, and demonstrate a unique pattern of DNA methylation [20]. In some particular examples, enhancer sites turned out to be less methylated in cells expressing the controlled genes than in non-expressing cells [21-24]. Taking a more systematic approach, Wienchet al. recently revealed activity-dependent methylation in a group of distal enhancers in the mouse [25], and Bock et al.[26] demonstrated correlations between enhancer methylation and expression of particular developmental genes during mouse tissue differentiation. We conducted a methodical analysis of distal DNA methylation sites associated with gene expression in normal and cancerous human cells. The results suggest that enhancer methylation corresponds closely with expression profiles of cancer genes in transformed cells. RESULTS AND DISCUSSION To explore relationships between DNA methylation and gene expression levels across the genome, we took the following strategy. First, we developed a model for the relationships between methylation in gene promoters and gene expression using a machine-learning algorithm. Then, we applied the derived model to methylation sites away from gene promoters and characterized the chromatin structure and binding profile of the discovered sites. Finally, we explored how the identified sites were altered in cancer. We analyzed available DNA methylation and gene expression data for 58 human cell types (Table S1 in Additional file 1). The methylation data were produced by two different assays: reduced representation bisulfite sequencing (RRBS) andInfinium HumanMethylation450 BeadChip(Illuminainc., San Diego, CA, USA). Following verification of high agreement between the assays (Figures S1A and S2A, Bin Additional file 1), we combined the two datasets. We then analyzed variation in methylation levels across the panel of cell types (see Materials and methods for detailed description).Sites with methylation levels that did not change across the cell types were eliminated, and the remaining 670, 906 variable methylation sites (VMSs) were used in the study (Figure S1Bin Additional file 1). VMSs in gene promoterswere negatively correlated with expression whereas VMSs in gene bodies were positively correlated with expression (Figure S2 in Additional file 1), as previously reported [6]. The connection with expression levels was most evident at -500 to +2, 000 base pairs relative to transcription start sites (TSSs) (Figure S2A-Cin Additional file 1). Therefore, we used the CpG methylation sites in this range for model development. Promoter-based model of regulatory methylation sites As a first step in our study, we tested whether we could couple promoters with the appropriate genes based solely on DNA methylation and gene expression data. We identified 15, 205 VMSs in promoters of 3,434 genes. These gene-CpG pairs provided the 'true' methylation-expression sample inputs (Figure 1A). We then trained a machine-learning algorithm to discriminatethese true gene-CpG pairs out of an excess offalse (randomized)pairs, produced by matching promoter CpGs with genes from other chromosomes. Through rounds of training and test sets, the algorithm optimizes parameters of linear and monotonic correlations between methylation and expression, allowing the best discrimination of the true pairs (Figure 1B). The output was a general model for the methylation-expression relationship in gene promoters (including promoters with or without CpG islands). Based on the learned model, we further produced a score for each of the gene-CpG pairs. At score ≥0.85, the model successfully paired 87.2% ofthe genes to their actual promoters, compared with a null expectation of 50% (Figure 1C). The sensitivity of the test under these conditions was 2.63% (that is, predictions were made for 2.63% of the promoter VMSs), and the false discovery rate was 12.8%. Thus, the developed model successfully paired promoter methylation sites with their genes based on methylation and expression data alone. [Figure: Promoter-based model of methylation-expression relationships at regulatory sites. (A) A typical gene-CpG sample pair, conveying the methylation of a promoter-based CpGsite in relation to the expression levels of its linked gene across cell types.(B) A machine-learning algorithm (SVM-MAP) was trained to distinguish true gene-CpG pairs out of 50-fold excess of false (randomized) pairs. Through rounds of training and test sets, the algorithm optimized parameters of linear (Pearson coefficient) and monotonic (Spearman) correlations to provide the best discrimination between true and false pairs, producing a general model for methylation-transcription relationships in gene promoters. Based on fitting with the learned model,a score was assigned to each gene-CpG pair. (C) Rates of successful gene-promoter pairing as a function of thresholds on the scores (null expectation = 50%). At score ≥0.85, the model successfully