Uncertainty quantification of reference-based cellular deconvolution algorithms.

The majority of epigenetic epidemiology studies to date have generated genome-wide profiles from bulk tissues (e.g., whole blood) however these are vulnerable to confounding from variation in cellular composition. Proxies for cellular composition can be mathematically derived from the bulk tissue profiles using a deconvolution algorithm; however, there is no method to assess the validity of these estimates for a dataset where the true cellular proportions are unknown. In this study, we describe, validate and characterize a sample level accuracy metric for derived cellular heterogeneity variables. The CETYGO score captures the deviation between a sample's DNA methylation profile and its expected profile given the estimated cellular proportions and cell type reference profiles. We demonstrate that the CETYGO score consistently distinguishes inaccurate and incomplete deconvolutions when applied to reconstructed whole blood profiles. By applying our novel metric to >6,300 empirical whole blood profiles, we find that estimating accurate cellular composition is influenced by both technical and biological variation. In particular, we show that when using a common reference panel for whole blood, less accurate estimates are generated for females, neonates, older individuals and smokers. Our results highlight the utility of a metric to assess the accuracy of cellular deconvolution, and describe how it can enhance studies of DNA methylation that are reliant on statistical proxies for cellular heterogeneity. To facilitate incorporating our methodology into existing pipelines, we have made it freely available as an R package (https://github.com/ds420/CETYGO).

DNA methylation meta-analysis reveals cellular alterations in psychosis and markers of treatment-resistant schizophrenia.

We performed a systematic analysis of blood DNA methylation profiles from 4483 participants from seven independent cohorts identifying differentially methylated positions (DMPs) associated with psychosis, schizophrenia, and treatment-resistant schizophrenia. Psychosis cases were characterized by significant differences in measures of blood cell proportions and elevated smoking exposure derived from the DNA methylation data, with the largest differences seen in treatment-resistant schizophrenia patients. We implemented a stringent pipeline to meta-analyze epigenome-wide association study (EWAS) results across datasets, identifying 95 DMPs associated with psychosis and 1048 DMPs associated with schizophrenia, with evidence of colocalization to regions nominated by genetic association studies of disease. Many schizophrenia-associated DNA methylation differences were only present in patients with treatment-resistant schizophrenia, potentially reflecting exposure to the atypical antipsychotic clozapine. Our results highlight how DNA methylation data can be leveraged to identify physiological (e.g., differential cell counts) and environmental (e.g., smoking) factors associated with psychosis and molecular biomarkers of treatment-resistant schizophrenia.

Guidance for DNA methylation studies: statistical insights from the Illumina EPIC array.

BACKGROUND: There has been a steady increase in the number of studies aiming to identify DNA methylation differences associated with complex phenotypes. Many of the challenges of epigenetic epidemiology regarding study design and interpretation have been discussed in detail, however there are analytical concerns that are outstanding and require further exploration. In this study we seek to address three analytical issues. First, we quantify the multiple testing burden and propose a standard statistical significance threshold for identifying DNA methylation sites that are associated with an outcome. Second, we establish whether linear regression, the chosen statistical tool for the majority of studies, is appropriate and whether it is biased by the underlying distribution of DNA methylation data. Finally, we assess the sample size required for adequately powered DNA methylation association studies. RESULTS: We quantified DNA methylation in the Understanding Society cohort (n = 1175), a large population based study, using the Illumina EPIC array to assess the statistical properties of DNA methylation association analyses. By simulating null DNA methylation studies, we generated the distribution of p-values expected by chance and calculated the 5% family-wise error for EPIC array studies to be 9 × 10- 8. Next, we tested whether the assumptions of linear regression are violated by DNA methylation data and found that the majority of sites do not satisfy the assumption of normal residuals. Nevertheless, we found no evidence that this bias influences analyses by increasing the likelihood of affected sites to be false positives. Finally, we performed power calculations for EPIC based DNA methylation studies, demonstrating that existing studies with data on ~ 1000 samples are adequately powered to detect small differences at the majority of sites. CONCLUSION: We propose that a significance threshold of P < 9 × 10- 8 adequately controls the false positive rate for EPIC array DNA methylation studies. Moreover, our results indicate that linear regression is a valid statistical methodology for DNA methylation studies, despite the fact that the data do not always satisfy the assumptions of this test. These findings have implications for epidemiological-based studies of DNA methylation and provide a framework for the interpretation of findings from current and future studies.

Genetic risk variants for brain disorders are enriched in cortical H3K27ac domains.

Most variants associated with complex phenotypes in genome-wide association studies (GWAS) do not directly index coding changes affecting protein structure. Instead they are hypothesized to influence gene regulation, with common variants associated with disease being enriched in regulatory domains including enhancers and regions of open chromatin. There is interest, therefore, in using epigenomic annotation data to identify the specific regulatory mechanisms involved and prioritize risk variants. We quantified lysine H3K27 acetylation (H3K27ac) - a robust mark of active enhancers and promoters that is strongly correlated with gene expression and transcription factor binding - across the genome in entorhinal cortex samples using chromatin immunoprecipitation followed by highly parallel sequencing (ChIP-seq). H3K27ac peaks were called using high quality reads combined across all samples and formed the basis of partitioned heritability analysis using LD score regression along with publicly-available GWAS results for seven psychiatric and neurodegenerative traits. Heritability for all seven brain traits was significantly enriched in these H3K27ac peaks (enrichment ranging from 1.09-2.13) compared to regions of the genome containing other active regulatory and functional elements across multiple cell types and tissues. The strongest enrichments were for amyotrophic lateral sclerosis (ALS) (enrichment = 2.19; 95% CI = 2.12-2.27), autism (enrichment = 2.11; 95% CI = 2.05-2.16) and major depressive disorder (enrichment = 2.04; 95% CI = 1.92-2.16). Much lower enrichments were observed for 14 non-brain disorders, although we identified enrichment in cortical H3K27ac domains for body mass index (enrichment = 1.16; 95% CI = 1.13-1.19), ever smoked (enrichment = 2.07; 95% CI = 2.04-2.10), HDL (enrichment = 1.53; 95% CI = 1.45-1.62) and trigylcerides (enrichment = 1.33; 95% CI = 1.24-1.42). These results indicate that risk alleles for brain disorders are preferentially located in regions of regulatory/enhancer function in the cortex, further supporting the hypothesis that genetic variants for these phenotypes influence gene regulation in the brain.

The correlation between reading and mathematics ability at age twelve has a substantial genetic component.

Dissecting how genetic and environmental influences impact on learning is helpful for maximizing numeracy and literacy. Here we show, using twin and genome-wide analysis, that there is a substantial genetic component to children's ability in reading and mathematics, and estimate that around one half of the observed correlation in these traits is due to shared genetic effects (so-called Generalist Genes). Thus, our results highlight the potential role of the learning environment in contributing to differences in a child's cognitive abilities at age twelve.