Cancer somatic mutations cluster in a subset of regulatory sites predicted from the ENCODE data

Abstract

Background: Transcriptional regulation of gene expression is essential for cellular differentiation and function, and defects in the process are associated with cancer. The ENCODE project has mapped potential regulatory sites across the complete genome in many cell types, and these regions have been shown to harbour many of the somatic mutations that occur in cancer cells, suggesting that their effects may drive cancer initiation and development. The ENCODE data suggests a very large number of regulatory sites, and methods are needed to identify those that are most relevant and to connect them to the genes that they control.

Methods: Predictive models of gene expression were developed by integrating the ENCODE data for regulation, including transcription factor binding and DNase1 hypersensitivity, with RNA-seq data for gene expression. A penalized regression method was used to identify the most predictive potential regulatory sites for each transcript. Known cancer somatic mutations from the COSMIC database were mapped to potential regulatory sites, and we examined differences in the mapping frequencies associated with sites chosen in regulatory models and other (rejected) sites. The effects of potential confounders, for example replication timing, were considered.

Results: Cancer somatic mutations preferentially occupy those regulatory regions chosen in our models as most predictive of gene expression.

Conclusion: Our methods have identified a significantly reduced set of regulatory sites that are enriched in cancer somatic mutations and are more predictive of gene expression. This has significance for the mechanistic interpretation of cancer mutations, and the understanding of genetic regulation.

Keywords: Cancer mutations; Cis regulation; Gene regulation; Modelling; Regulatory regions.

Fig. 1

Building an expression model for CNN3 (ENST00000370206.4). a shows the mean squared error against the log (λ) LASSO penalty parameter with numbers above the graph indicating the number of predictive variables (non-zero coefficients) in the corresponding LASSO model. Dotted lines show possible choices of λ at minimum mean-squared error (λ_min) and more conservatively at that value plus 1 standard error. This identifies models with 2 predictive variables as optimal. b shows the correlation between observed expression and predicted expression from the model. c and d show the correlation of DNaseI signal intensities and expression for the two candidate regulatory elements (CRRs) chosen by the LASSO method. e shows the correlation between DNaseI signal intensities and expression for an example rejected CRR. f shows the genomic location of the two chosen CRRs and one example rejected CRR

Fig. 2

The chosen CRRs for NAB2 (ENST00000342556.5), STAT6 (ENST00000300134.2) and LRP1 (ENST00000243077.2). Black arrows link CRRs to the transcripts for which they were chosen in expression models; note that one CRR was chosen for both STAT6 and NAB2. Details of the chosen CRRs are given red boxes, including the bound transcription factors, sizes of the CRRs and mutations mapped from the COSMIC database. CRRs are labelled as enhancers if they show positive correlation with expression and repressors if they show negative correlation. The chosen CRRs are marked as red boxes if there is at least one reported mutation in them, and black otherwise

Fig. 3

Graphical summary of the methodology employed