A Novel Application of Mixed Effects Models for Reconciling Base-Pair Resolution 5-Methylcytosine and 5-Hydroxymethylcytosine Data in Neuroepigenetics

Abstract

Epigenetic marks operate at multiple chromosomal levels to regulate gene expression, from direct covalent modification of DNA to three-dimensional chromosomal structure. Research has shown that 5-methylcytosine (5-mC) and its oxidized form, 5-hydroxymethylcytosine (5-hmC), are stable epigenetic marks with distinct genomic distributions and separate regulatory functions. In addition, recent data indicate that 5-hmC plays a critical regulatory role in the mammalian brain, emphasizing the importance of considering this alternative DNA modification in the context of neuroepigenetics. Traditional bisulfite (BS) treatment-based methods to measure the methylome are not able to distinguish between 5-mC and 5-hmC, meaning much of the existing literature does not differentiate these two DNA modifications. Recently developed methods, including Tet-assisted bisulfite treatment and oxidative bisulfite treatment, allow for differentiation of 5-hmC and/or 5-mC levels at base-pair resolution when combined with next-generation sequencing or methylation arrays. Despite these technological advances, there remains a lack of clarity regarding the appropriate statistical methods for integration of 5-mC and 5-hmC data. As a result, it can be difficult to determine the effects of an experimental treatment on 5-mC and 5-hmC dynamics. Here, we propose a statistical approach involving mixed effects to simultaneously model paired 5-mC and 5-hmC data as repeated measures. We tested this approach using publicly available BS/oxidative bisulfite-450K array data and showed that our new approach detected far more CpG probes with paired changes in 5-mC and 5-hmC by Alzheimer’s disease status (n = 14,183 probes) compared with the overlapping differential probes generated from separate models for each epigenetic mark (n = 68). Of note, all 68 of the overlapping probe IDs from the separate models were also significant in our new modeling approach, supporting the sensitivity of our new analysis method. Using the proposed approach, it will be possible to determine the effects of an experimental treatment on both 5-mC and 5-hmC at the base-pair level.

Prenatal Lead (Pb) Exposure and Peripheral Blood DNA Methylation (5mC) and Hydroxymethylation (5hmC) in Mexican Adolescents from the ELEMENT Birth Cohort

“…A mixed-effects model was run, treating 5mC and 5hmC values as repeated measures of a single outcome variable—DNA methylation, given that 5mC and 5hmC are both measured at each CpG site, and the values for these two marks are dependent on each other both biologically and statistically ( Kochmanski et al. 2019 ).…”

Parkinson’s disease-associated shifts between DNA methylation and DNA hydroxymethylation in human brain

“…44 Collectively, these data suggest that there are significant PD-associated shifts between 5mC and 5hmC at iDMCs that are not captured by analyzing BS-based data alone or by analyzing each mark as distinct datasets (Figure 1C). 44,45 Because many of these iDMCs are located within genes previously implicated in PD and in genes within pathways thought to play a role in PD, these data suggests that shifts in the balance between DNA modifications may play an important but unrecognized role in PD etiology in both known and novel PD-related genes.…”

“…IDAT files were imported into R and processed using an in-house bioinformatics pipeline that utilizes minfi (version 1.48.0), ChAMP (version 2.32.0), posibatch (version 1.0), and ENmix (version 1.38.01), and CETS (version 3.03) packages in R, as previously described (Supplementary File 1). 44,45,49,[52][53][54][55][56][57][58][59] After QC, eleven female samples and six male samples were removed due to a high level (>10%) of failed probes, leaving 57 male (28 PD, 29 control) and 26 female (13 PD, 13 control) samples. Failed probes (53,305) were removed from remaining samples when detection p-value was > 0.01 in more than 5% of samples.…”

“…Here, we performed an integrated genome-wide analysis of 5mC and 5hmC using our novel application of mixed effects modeling in enriched neuronal nuclei from PD post-mortem parietal cortex samples. [45][46][47] These PD-associated iDMCs were largely unique from DMCs identified in our previous BS-based EWAS: 43 genes were identified in both studies, 652 only in the current analysis, and 480 genes identified only in the previous analysis (Supplementary Files 5,6). 44 Collectively, these data suggest that there are significant PD-associated shifts between 5mC and 5hmC at iDMCs that are not captured by analyzing BS-based data alone or by analyzing each mark as distinct datasets (Figure 1C).…”

“…Here, we report the results of an epigenome-wide analysis of 5hmC and 5mC in enriched neurons from PD brain using our recently proposed method for reconciling base-pair resolution 5mC and 5hmC data. [45][46][47] This method considers 5mC and 5hmC as paired data since these marks are biologically and statistically dependent on each other. We utilized additional DNA isolated from the same neuron-enriched samples and performed oxidative BS (oxBS) conversion paired with the Illumina EPIC BeadChip array to specifically measure 5mC.…”