A self tuning sliding window method for detecting phenotype linked regional poly-methylation architecture in sparse wildlife methylomes

Thomas Stocker 

Despite featuring extreme physiological adaptations integration of wildlife species into the modern ‘omics’ frameworks are limited due to the sparsity in the data. To address the sparsity limitation a self-tuning sliding-window framework was developed for the identification of the regional poly-CpG methylation architecture associated with phenotypic traits. Under the framework the iteratively expanded window aligns against each chromosome for two constraints: (i) the window must increase its penalised R2 statistic derived from univariate (ADVI-EWAS) and multivariate (Horseshoe) CpG-phenotype models and (ii) both models but agree upon the direction of the methylome-phenotype association. Any windows not passing the two conditions were discarded if they lacked sufficient CpG density, otherwise were retained but ceased to grow. Application of the framework to the blood methylome of 69 California Sea Lions (Zalophus californianus) detected three ‘trusted’ windows across the entire genome. Each of the windows were uniquely mapped to a single gene, these genes were GCSH, DNMT1, LMX1A. All three genes are mechanistically linked to the phenotypic trait and are consistent with the extreme phenotype of marine mammals. The results were independent of the age of the individuals tested. The results were not predictive of phenotype when applying epigenome wide association studies (EWAS); however, application of the poly-CpG architecture provided a phenotypically linked signal that was biologically interpretable. The sliding window method is computationally efficient, isolates short, directionally stable genome regions and outperforms single-site analyses. The generalisable statistical framework provided within the study could extract phenotype-linked molecular architecture from sparse wildlife methylomes.