Re-analysis of publicly available methylomes using signal detection yields new information

Abstract

Cytosine methylation is an epigenetic mark that participates in regulation of gene expression and chromatin stability in plants. Advancements in whole genome sequencing technologies have enabled investigation of methylome dynamics under different conditions. However, the computational methods for analyzing bisulfite sequence data have not been unified. Contention remains in the correlation of differentially methylated positions with the investigated treatment and exclusion of noise, inherent to these stochastic datasets. The prevalent approaches apply Fisher’s exact test, logistic, or beta regression, followed by an arbitrary cut-off for differences in methylation levels. A different strategy, the MethylIT pipeline, utilizes signal detection to determine cut-off based on a fitted generalized gamma probability distribution of methylation divergence. Re-analysis of publicly available BS-seq data from two epigenetic studies in Arabidopsis and applying MethylIT revealed additional, previously unreported results. Methylome repatterning in response to phosphate starvation was confirmed to be tissue-specific and included phosphate assimilation genes in addition to sulfate metabolism genes not implicated in the original study. During seed germination plants undergo major methylome reprogramming and use of MethylIT allowed us to identify stage-specific gene networks. We surmise from these comparative studies that robust methylome experiments must account for data stochasticity to achieve meaningful functional analyses.

Phenotypic plasticity for improved light harvesting, in tandem with methylome repatterning in reef‐building corals

“…However, the approach is especially valuable in non‐model systems where changes cannot be confirmed with targeted mutation(s). The association between phenotype change and treatment‐associated methylome modification is informative in understanding the underlying molecular features of the phenotypic change by directly identifying responsive gene networks (Hafner & Mackenzie, 2023; Kundariya et al., 2022; Sanchez et al., 2019; Sanchez & Mackenzie, 2016a, 2016b).…”

“…Under experimental conditions, spontaneous natural methylation or “background methylation” may occur. This stochasticity has complicated the discrimination of treatment‐associated signal and the understanding of phenotypic adjustments with methylome modifications (Hafner & Mackenzie, 2023; Sanchez & Mackenzie, 2023; Yang & Mackenzie, 2020). As the significance of individual cytosine variations in phenotypic responses becomes more evident, innovative methods, such as MethyIT (Sanchez et al., 2019), have incorporated these features into machine learning to distinguish treatment‐associated differential methylation.…”

“…The repatterning is generally associated with other epigenetic effects such as histone modifications and changes in noncoding RNA. These methylome modifications can be assayed at single nucleotide resolution, providing the robust datasets required for identifying responsive underlying gene networks that could explain phenotypic adjustments (Hafner & Mackenzie, 2023; Kundariya et al., 2022; Sanchez et al., 2019; Sanchez & Mackenzie, 2016a, 2020; Yang & Mackenzie, 2020).…”

“…Several studies have associated coral DNA methylation with plasticity (Dimond et al., 2017; Dimond & Roberts, 2016; Dixon et al., 2018; Durante et al., 2019; Hackerott et al., 2023; Liew et al., 2018; Putnam et al., 2016; Roberts & Gavery, 2012; Rodríguez‐Casariego et al., 2020), with whole‐genome bisulphite sequencing (WGBS) contributing to single base‐pair resolution (Liew et al., 2018). However, WGBS data analysis can be challenging due to the highly dynamic features of methylome datasets, where spontaneous natural methylation “background methylation” masks treatment‐associated signal (Hafner & Mackenzie, 2023; Sanchez & Mackenzie, 2023; Yang & Mackenzie, 2020). This stochasticity has been incorporated in a novel machine learning approach (Sanchez et al., 2019) that permits validation of treatment association of 98% confidence, confirmed with targeted mutation(s) (Kundariya et al., 2022).…”