In the current study, our systematic comparisons of schizophrenia EWAS meta-analyses and simulations revealed that a sex-stratified approach improves discoveries as evidenced by consistent upward and leftward deviations in the QQ plots. Consequently, the sex-stratified meta-analysis resulted in the identification of a larger number of DMPs associated with schizophrenia than the standard sex-adjusted approach. Our simulations also indicate that the gain in power from sex-stratified analysis is more pronounced when there are substantial differences in sex-specific effect estimates. This translated into a higher explained liability for PMRS derived from the sex-stratified results, which outperformed the sex-adjusted PMRS in predicting schizophrenia. Interestingly, PMRS prediction of schizophrenia appeared to have better utility in females than in males, regardless of the training dataset. The boost in the discovery of DMPs and DMRs associated with schizophrenia using a sex-stratified EWAS analytical approach shows the methodological advantage and expands our understanding of the molecular pathways involved in the pathophysiology of schizophrenia.
Our application of a sex-stratified analytical approach boosted the discovery of DNA methylation loci. Although the advantages of using sex-stratification in GWAS meta-analyses have previously been suggested,18 our study is the first to systematically examine the method in the EWAS of schizophrenia. The sex-stratified analysis improved the discovery of DMPs with concordant effect direction in both sexes but with different magnitudes.18 This observation is in agreement with the recent report from Zhou et al. who observed a larger magnitude of changes in DNA methylation associated with schizophrenia in females than in males.17 Thus, further supporting the sex-stratified meta-analysis approach in EWAS. Despite the smaller sample of females across all four cohorts included in our study, we identified three DMPs and four DMRs associated with schizophrenia. However, neither sex-stratified nor sex-adjusted meta-analyses would capture DMPs with discordant effects in males and females as the effect estimates cancel each other out.17
The identification of novel DMPs in the sex-specific approach suggests that some molecular pathways involved in the pathophysiology of schizophrenia might be present only in males or females. Studies have shown more changes in DNA methylation and gene expression associated with schizophrenia in females than in males in postmortem prefrontal cortex,16, 17 and in neuronal cell lines from schizophrenia-discordant monozygotic twins.44 These DMPs and DMRs with large effect sizes in females are more likely to be identified. Consequently, the PMRS derived from sex-specific female analysis performed better in predicting schizophrenia across all target datasets compared to the PMRS from sex-specific male analysis.
The genes mapped to the identified DMRs link to genetic, biological, and environmental factors relevant to understanding the pathophysiology of schizophrenia. GWAS of schizophrenia has identified a locus in the gene GABBR1 in East Asian population.45 GABBR1 encodes the receptor for the main inhibitory neurotransmitter, gamma-aminobutyric acid (GABA), in the brain.46 Also, candidate gene studies have reported associations between schizophrenia and THRB,47 COMT,48 DIXDC1,49 and WBP1L.50 Other genes LRRC32, KCNAB3, NNAT, TGFB1, VWC2, and KDM2B were linked to neurodevelopmental processes and neuropsychiatric phenotypes.51–56 Additionally, differential methylation involving LRRC32 and BLCAP/NNAT were previously linked to exposure to adverse perinatal factors (e.g., viral infection, iron deficiency, birth asphyxia) with potential neuropsychiatric consequences.25, 51 Furthermore, some of the genes mapped to the DMRs are involved in molecular pathways relevant to the immune system (LAIR1, TNF),57, 58 regulation of gene expression (KDM2B, ZIK1),59, 60 and energy metabolism (PRKAA2).61 The identification of these genes sheds light on the pathways involved in the pathophysiology of schizophrenia.62
DNA methylation changes in COMT have not been reported in previous EWAS of schizophrenia. COMT encodes the enzyme which plays a vital role in the methylation of catecholamines (e.g., dopamine) and is a drug target for neuropsychiatric disorders.63 The identification of a novel DMR associated with schizophrenia mapping to KCNAB3, a gene coding subunit of voltage-gated potassium ion channel,64 may also have the potential as a drug target. The gene and its product are linked to migraine and epilepsy.52, 65 A recent animal study reported that KCNAB2, a product of a gene closely related to KCNAB3, is involved in the regulation of the firing of dopaminergic neurons.66 Dysregulation of dopaminergic neurotransmission has been suggested as a final common pathway of multiple genetic and environmental factors known to be risk factors for schizophrenia.67 DNA methylation changes at COMT and KCNAB3 may be one of the pathways involved in the pathophysiology of schizophrenia at least in a subgroup of patients although this needs further research.
The neurodevelopment-related pathways enrichment for the genes annotated to the top DMPs across females and males is consistent with other reports. Some of the pathways appear overrepresented in females (immune-related) or males (cell adhesion) suggesting potential sex differences in the pathobiology of schizophrenia. These sex differences may also relate to the sex-specific effects of environmental factors on the epigenome,25 or sex-specific resilience.17 The existence of sex differences in the biology of schizophrenia was previously reported, however, the final molecular pathways leading to the symptoms may be similar.44
To progress towards the use of EWAS results in predicting schizophrenia, we have tested the potential for the different models to identify sets of methylation signals identified in a discovery sample, which could predict disease status in an independent sample. Not surprisingly, since sex-stratified analyses capture better effects that differ between sexes, PMRS prediction using sex-stratified analysis performed better than sex-adjusted PMRS and might thus be more efficient than the standard approach. Moreover, the observation that the sex-adjusted PMRS yields the highest Nagelkerke R2 values when applied to the female sample but does not explain variability in schizophrenia in the male sample suggests that this approach may be significantly influenced by the greater magnitude of DNA methylation changes in females. Accordingly, epigenetic discoveries in schizophrenia using sex-adjusted models may lead to sex disparities.
We acknowledge that our study has important limitations. The DNA methylation probes provide coverage of a small fraction of the human epigenome and therefore, it does not capture effects for the regions of the genome that are not covered. Furthermore, the DNA methylation changes in the blood are not identical to those in the brain, however, there is a significant similarity between the two.5, 68 The cross-sectional study designs of the cohorts included in our analyses do not allow causal associations to be established. Although the identified DNA methylation changes need further investigation for their functional molecular effects, our findings are still relevant for the development of biomarkers of schizophrenia especially since blood is an accessible tissue.
In conclusion, we have shown the sex differences in the epigenome and sex-stratified analysis can be leveraged to enhance epigenetic discoveries in schizophrenia. We identified novel DMPs and DMRs associated with schizophrenia, which advance our understanding of the molecular pathways involved in the pathophysiology of schizophrenia. The advantages of sex-stratified analysis extend beyond the discovery of novel loci to improved prediction using PMRS, which may have potential clinical applications. The potential downstream effects of DNA methylation changes warrant further research to advance precision psychiatry. Future investigations with larger coverage of the epigenome are needed to reveal the male and female differences contributing to the phenotypic heterogeneity. Given that a significant proportion of DMPs associated with schizophrenia are under the influence of genetic variants,5, 17 the advantage of sex-stratified analysis may also extend to genetic studies highlighted previously.18 We show that sex differences have a potential implication for the development of sex-specific biomarkers, enhance prediction scores, improve our understanding of the pathophysiology and thus perhaps progress towards precision psychiatry.