Identification of 5mC within heterogenous tissue using de-novo somatic mutations

Tissues represent a fundamental evolutionary interface at the junction of genotype and phenotype. Indeed, gene regulation often occurs at the tissue level and manifests itself through tissue-specific epigenetic modifications. Studies investigating tissue epigenetics are limited by access to pure tissues. Tissues not only differ epigenetically, they are also subject to genetic differentiation through somatic mutations. As somatic mutations follow predictable patterns of inheritance, the application of population genomic approaches to inter- and intra-tissue variation could allow for the efficient detection of epigenetic modifications, even when tissue samples are convoluted. Here, we present an approach that uses de-novo somatic mutations to deconvolute 5mC methylation patterns through analysis of shifts in tissue-specific allele frequencies. We use simulations and bisulfite sequencing data to show that somatic mutations are common and detectable in next-generation sequencing data. We then use changes in mutation frequencies to accurately derive the proportional tissue of origin along a gradient of in silico subsamples of mixed-tissue bisulfite reads. We confirm that mixed tissues bias estimates of methylation levels and prevent detection of methylation differences at high levels of mixture. Our derived estimates of tissue contamination allow for unbiased and accurate deconvolution of mixed-tissue methylations in CpG and non-CpG context. We are ultimately able to recover 15-30% of differentially-methylated sites, and approximately 40-90% of differentially-methylated CpG islands and gene bodies in any cytosine context at contamination levels up to 90%. Our findings highlight the utility of population genomic approaches across scales, and expand the accessibility of epigenetics studies within evolutionary biology.