Alteration of DNA Methylation and Epigenetic Scores Associated With Features of Schizophrenia and Common Variant Genetic Risk

,

https://doi.org/10.1016/j.biopsych.2023.07.010Schizophrenia is a complex and heterogeneous neuropsychiatric condition that is thought to originate from interplay between genetic and environmental factors.Large-scale genetic studies have revealed that schizophrenia is highly heritable and characterized by a vast polygenic architecture across the genome (1,2), with recent genome-wide association studies identifying 287 associated common frequency loci (3).Despite this, affected individuals often possess unique combinations of risk variants ranging in frequency from common to ultrarare (2), which likely contributes to heterogeneity in clinical presentation and responses to pharmacological interventions.Indeed, clinical features such as cognitive function and treatment resistance are likely influenced by genetic heterogeneity and are consequently difficult to predict and treat (4)(5)(6).In view of this, there remains a growing need for further exploration of genetic mechanisms and causal genes associated with schizophrenia for the development of personalized intervention strategies.However, reconciling schizophrenia-associated genetic variation with clinical features of the disorder remains a major challenge, as many of these variants reside in noncoding regulatory elements with capacity to disrupt a vast array of mechanisms associated with gene expression (7,8).
There is growing interest in dissecting epigenetic mechanisms associated with the etiology of schizophrenia.In particular, DNA methylation at CpG dinucleotides may index the effect of some schizophrenia-associated genetic variants on gene expression by regulating multiple aspects of transcription.Previous work has established that DNA methylation within enhancer and promoter regions represses gene expression by impeding transcription factor binding and promoting the formation of heterochromatin, whereas methylation within gene bodies is hypothesized to support cotranscriptional splicing or silence cryptic promoters (9).DNA methylation can also be influenced by genetic variants (10) and environmental exposures, such as nutritional factors, tobacco use, and air pollutants (11).Consequently, altered methylation in schizophrenia may simultaneously index both genetic and environmental risk factors.Recent epigenome-wide association studies (EWASs) (also known as methylation-wide association studies) have uncovered dysregulation of DNA methylation in schizophrenia and treatment resistance as well as epigenetic changes associated with environmental and physiological factors (12)(13)(14)(15).However, few studies have explored epigenetic features associated with other important dimensions of schizophrenia, such as cognitive status, age of onset, and other measures of progression and severity.To address this, we comprehensively analyzed DNA methylation in 381 individuals with schizophrenia from the Australian Schizophrenia Research Bank (ASRB) (16) to examine epigenetic changes associated with several phenotypic indices and determine interplay between schizophrenia-associated genetic variation and epigenetic dysregulation.Our findings reveal several groups of differentially methylated loci, which in some cases localize to genes previously associated with schizophrenia and comorbid psychiatric conditions.Furthermore, we explore phenome-wide epigenetic overlap between schizophrenia-associated traits and other phenotypes and exposures and demonstrate that epigenetic scores (ESs) may offer utility in subcategorizing affected individuals.

Study Cohort
Genomic DNA from peripheral blood mononuclear cells was obtained from the ASRB (16) with approval from the University of Newcastle Human Research Ethics Committee and the ASRB access committee.DNA methylation was quantified for 384 individuals of European descent diagnosed with schizophrenia as per ICD-10 criteria, with 381 retained after quality control.
We first examined associations between DNA methylation and cognitive status in schizophrenia.As detailed previously (17), we used the multidimensional grade of membership clustering of 9 cognitive measures to define individuals as cognitively spared (n = 221, 58%) or cognitive deficit (n = 160, 42%) (see Supplemental Methods in Supplement 1).We additionally analyzed individuals with treatment resistance, defined by current or previous administration of clozapine (n = 67, 18%), versus individuals without previous clozapine usage (n = 314, 82%) as per Hannon et al. (13), noting that clozapine is the only drug approved for treatment-resistant schizophrenia (TRS) in Australia, North America, and Europe (18,19).Three continuous measures were also examined: age of onset (n = 381); Global Assessment of Functioning (GAF) score as defined by the American Psychiatric Association (20) (n = 367), noting that low scores correspond to higher symptom severity; and polygenic risk score for schizophrenia (PRS SZ ) (n = 315) (see Supplemental Methods in Supplement 1) (21).A modest difference was observed between the mean ages for the cognitive status (p = .03,t test) and clozapine (p = .02)groups, while a positive association between age and age of onset was detected (p , 2.2 3 10 216 ) (Table 1).There were proportionally more male participants in the cognitive deficit subgroup versus the cognitively spared subgroup (p = 7.86 3 10 23 , c 2 test, df = 1), while GAF score was significantly lower in male participants than female participants (p = 4 3 10 25 , t test) (Table 1).Age and sex were consequently included as covariates in our statistical analyses.13), with the exception of employing effect sizes from multiple regression of clozapine prescription.ESs were calculated as the average of the CpG b values weighted on the meta-analysis effect sizes as follows: where for probe i in individual j, B ij denotes the methylation level, b i is the meta-analysis effect size at each probe, and N j is the total number of probes used for calculation of the ES.ESs were constructed at 5 meta-analysis p EWAS thresholds (0.5, 0.05, 0.001, 1 3 10 25 , and 1 3 10 27 ), analogous to thresholding used in polygenic risk scoring (35).ESs were also systematically constructed for significantly correlated ASRB and EWAS Catalog trait pairings at 4 p EWAS quantiles (100%, 75%, 50%, and 25%), noting that only probes with p EWAS , 1 3 10 24 are reported in this database, and thus epigenome-wide effects could not be examined.Association between each trait-ES pairing (with covariates) was assessed via binomial logistic regression for categorical traits and linear regression for continuous traits.R 2 values were calculated via 100 iterations of 5-fold cross-validation, noting that for cognitive subtypes and clozapine usage, Nagelkerke's pseudo R 2 was converted to the liability scale (36), assuming a population prevalence of 0.29% for cognitive deficit (37) and 0.21% for clozapine usage (38,39).See Supplemental Methods in Supplement 1 for further details.

Differentially Methylated Probes Associated With Features of Schizophrenia
EWASs were conducted for the 5 features of schizophrenia controlling for age, sex, genetic principal components (for PRS SZ only), and methylation-derived cellular proportions and smoking scores (40-42); Figure 1 provides a methodological overview.No probes survived Bonferroni correction; thus, we further considered a suggestively significant threshold of p EWAS , 6.72 3 10 25 given our modest sample size; however, DMPs at this threshold should be treated with appropriate caution (Figure 2A).The most significant signal was the association between probe cg01176363 and clozapine prescription, which almost surpassed the Bonferroni cutoff (p = 7.18 3 10 28 ).At the suggestively significant threshold, DMPs

Bisulfite conversion
Generate beta values.

individuals retained.
Treatment resistance.

Functional analysis
Base gene/region annotations from EPIC manifest.

PsychEncode annotations.
Toppfun suite.processing and QC, EWASs were conducted for 5 schizophrenia-associated traits.These EWASs were repeated for 4 traits after residualizing on the effect of PRS SZ to explore changes independent of common variant risk for the disorder.A battery of downstream analyses was conducted to further explore EWAS signatures, including functional enrichment analyses, identification of DMRs, exploration of genes associated with schizophrenia and comorbid psychiatric conditions, epigenetic covariance with EWASs in the EWAS Catalog (33), and analysis of mQTL.Finally, epigenetic scores weighted on previous EWASs of schizophrenia and treatment-resistant schizophrenia were constructed and analyzed with respect to the 5 primary EWAS traits.Supplement 2), whereas enrichment was comparatively sparse for the remaining traits and DMRs (Table S8 in Supplement 2).
Genes Associated With Psychiatric Illness Observed Among DMPs and DMRs We next analyzed enrichment of suggestive DMPs within genes previously associated with 11 psychiatric conditions, as defined by PsychENCODE schizophrenia-associated gene sets (43), and DisGeNET (45) curated and inferred gene sets, which summarize gene-disorder associations from sources including Clinvar, GWAS Catalog, Clingen and UniProt (see Supplemental Methods in Supplement 1).At an FDR BH , .05,suggestive DMPs associated with GAF scores were underrepresented among PsychENCODE schizophrenia-associated genes (OR # 0.5) and overrepresented among anxietyassociated genes (OR = 3.66) (Figure 3C; Table S9 in Supplement 2).DMPs associated with PRS SZ -residualized GAF scores were also enriched among genes associated with anxiety and attention-deficit/hyperactivity disorder (OR $ 3.31), while DMPs for PRS SZ -residualized cognitive status exhibited enrichment within the autism spectrum disorderrelated gene set (OR = 5.78) (Figure 3C; Table S9 in Supplement 2).Notably, there was nominal evidence for underrepresentation of DMPs associated with age of onset (with and without PRS SZ ) and PRS SZ -residualized GAF scores within PsychENCODE schizophrenia-associated genes (Figure 3C; Table S9 in Supplement 2).We also examined DMP enrichment within promoters and enhancers of these same gene sets; however, no findings surpassed correction for multiple testing (Tables S10 and S11 in Supplement 2).
To extend these results, the methylation status of DMPs and DMRs associated with these psychiatric conditions was examined.A total of 176 suggestive DMPs were associated with at least one psychiatric condition, including 67 exclusively associated with schizophrenia and 61 associated with more than one condition (Figure 4).These genes included calcium voltage-gated channel subunits (CACNA1A, CAC-NA1C, CACNA1E, and CACNA1H), genes related to neuronal function (GRID1, GABBR2, KCNN3, NCAM1, NGFR, and NOS1, among others), interleukin 2 receptor subunits (IL2RA and IL2RB), and genes associated with chromatin regulation (DNMT3A, TET3, and TOP2B, among others) (Figure 4; Table S12 in Supplement 2).Four DMRs within genes associated with psychiatric conditions were additionally uncovered, including ERCC8 (associated with cognitive status with and without PRS SZ and schizophrenia), OLIG2 (GAF score PRS SZ -residualized and schizophrenia), PEX5 (cognitive status and impaired cognition), and PUF60 (GAF score and attention-deficit/hyperactivity disorder) (Table S13 in Supplement 2).
Epigenetic Covariance Among Schizophrenia-Associated Traits and Clinically Relevant Phenotypes Shared epigenetic signals among the schizophreniaassociated traits and other phenotypes and exposures were investigated phenome-wide by regressing ASRB EWAS estimates against that of the same CpGs for traits reported in the EWAS Catalog (33) (Figure 5A).After harmonizing ancestry and covariates, 31 phenotypes and exposures were significantly correlated with at least one ASRB trait (FDR BH , .05)(Figure 5B).Interestingly, we observed epigenetic covariance with disease traits including coronary heart disease, lung cancer, and type 2 diabetes; exposures such as maternal smoking during pregnancy and prenatal maternal stress; and quantitative traits including C-reactive protein (CRP) levels, fasting glucose, and fasting insulin (Figure 5B; see Figure 5C for representative analysis of CRP; Table S14 in Supplement 2).As anticipated, previous EWASs of schizophrenia and psychosis (12)(13)(14) were consistently represented, further suggesting that the features of schizophrenia considered in the current study exhibit epigenetic overlap with diagnostic status.We additionally examined EWASs for plasma concentrations of 4231 proteins from the EWAS Catalog, uncovering 1454 significant proteins associated with at least one ASRB trait, including CCL17, MMP10, and PRG2, among others (Figure 5D; Table S15 in Supplement 2).Collectively, these proteins were enriched for a range of pathways including the innate immune system (236 proteins, 6.96 3 10 226 ); ensemble of genes encoding extracellular matrix and extracellular matrix-associated proteins (194 proteins, 4.66 3 10 223 ); and platelet activation, signaling, and aggregation (76 proteins, 1.7 3 10 216 ) (Table S16 in Supplement 2).

Relationship Between ESs and Schizophrenia-Associated Traits
ESs were constructed for the 5 clinical traits using recently published EWASs of schizophrenia (ES SZ ) and treatmentresistant schizophrenia (ES TRS ) (12,13) (Table S17 in Supplement 2).Interestingly, ES TRS was nominally associated with clozapine usage across all 5 p EWAS thresholds, with the p EWAS , .001threshold exhibiting the best fit at a median R 2 of = reported by PsychENCODE (43).Enrichment of DMPs was examined via Fisher's exact test, with correction for multiple testing via Benjamini-Hochberg false discovery rate and Bonferroni methods.Note that promoters were defined as regions 2 kb upstream of TSSs as per PsychENCODE (43).(B) As in panel (A), except examining DMP enrichment within genomic features reported by the IlluminaMethylationEPICmanifest package (version 0.3.0)(25).(C) Enrichment of DMPs within gene sets associated with schizophrenia and comorbid psychiatric conditions as per DisGeNET (45) curated and inferred gene sets, and Psy-chENCODE (schizophrenia high confidence and risk gene sets).Note that anorexia nervosa, obsessive-compulsive disorder, posttraumatic stress disorder, and Tourette's syndrome are not depicted due to lack of overlap with Australian Schizophrenia Research Bank DMPs for most traits.See Tables S6 and S9 in Supplement 2 for full results.All analyzed datasets are publicly available at http://resource.psychencode.org.All data presented as OR 6 95% CI; *nominal p , .05;**Benjamini-Hochberg false discovery rate , .05;***Bonferroni p , .05.ADHD, attention-deficit/hyperactivity disorder; ANX, anxiety; ASD, autism spectrum disorders; BD, bipolar disorder; DMP, differentially methylated probe; GAF, Global Assessment of Functioning; IC, impaired cognition; MDD, major depressive disorder; OR, odds ratio; PEC, PsychENCODE; PRS, polygenic risk score; SZ, schizophrenia; TSS, transcription start site.
Epigenetic Disruption in Features of Schizophrenia   S18 in Supplement 2).We subsequently used this optimal ES to compare the relative effect of epigenetic risk and PRS SZ on treatment resistance indexed via clozapine administration.Using 315 individuals with matched genetic data and including genetic principal components as covariates, ES TRS significantly improved the model rather than just the covariates alone (R 2 = 0.022, p = 3.85 3 10 23 , c 2 test of residual deviance, df = 1).In contrast, PRS SZ was not associated with clozapine usage (p = .41,df = 1).We additionally tested ES TRS in a combined model with PRS SZ , in which the variance explained by ES TRS on the liability scale was very similar (R 2 = 0.023, p = 2.76 3 10 23 , df = 1).We also systematically constructed ES for all significantly correlated ASRB-EWAS Catalog trait pairings.Interestingly, 62 significant (FDR BH , .05)ES-phenotype associations involving 23 ESs for plasma protein concentrations were identified as follows: age of onset, one protein (COCH); clozapine, 19 proteins; cognitive status, 3 proteins (IAPP, KIR3DL3, and STAB1) (Figure 6D; Tables S19 and S20 in Supplement 2).Together, these proteins were enriched for a variety of pathways associated with the extracellular matrix, small molecule pathways, and blood coagulation, among others (Table S21 in Supplement 2).In contrast, the nonprotein EWAS Catalog traits exhibited no association with the ASRB traits (Table S19 in Supplement 2).

DISCUSSION
Schizophrenia is an extremely heterogeneous disorder in terms of onset, clinical presentation, and progression.Characterizing the basis of clinical features of schizophrenia that contribute to this variability would be useful to better diagnose and treat the disorder.Here, we examined DNA methylation profiles associated with specific clinical features and polygenic risk for schizophrenia to investigate epigenetic disruptions with greater precision than traditional case-control studies, which may be obscured by phenotypic heterogeneity arising from broad diagnostic criteria (46).Across all traits, we identified significant epigenetic covariance with previous EWASs of schizophrenia and psychosis (12)(13)(14), and differential methylation of recent EWAS-nominated genes such as CACNA1C (12,13) and AUTS2 (14), suggesting that some epigenetic changes associated with schizophrenia diagnostic status are related to specific features of the disorder.Strikingly, we also observed enrichment of DMPs within genes broadly associated with psychiatric illness (45), and epigenetic covariance between features of schizophrenia and known comorbidities (e.g., cardiovascular disorders, lung cancer, type 2 diabetes) and risk factors (e.g., air pollution, CRP, fasting insulin and glucose, and maternal factors) (47)(48)(49)(50)(51)(52)(53)(54)(55)(56)(57)(58).CRP is a particularly interesting example, as observational studies have shown that this protein is highly dysregulated in individuals with schizophrenia (54,55), yet genetic causal inference paradoxically suggests that CRP is protective for the disorder (59) despite exhibiting an association with region-specific thinning of the cerebral cortex (60).Epigenetic covariance with other proteins enriched for immune-and extracellular matrixrelated functions was also consistent with previous studies implicating these broad systems in psychiatric illness (61,62).Collectively, our findings support the existence of shared epigenetic signatures among features of schizophrenia and known comorbidities and risk factors, which may assist in characterizing relationships between these trait pairings.
For age of onset, cognitive status, and GAF scores, patterns of DMP enrichment were observed among gene promoters and brain-specific H3K27ac peaks, suggesting that altered DNA methylation for these traits may disproportionally affect transcriptional activity (9,43,(63)(64)(65).Strikingly, however, we observed a negative correlation between EWAS signatures and methylation-associated gene expression within the cohort (Figures S9 and S10 in Supplement 1; Tables S22-S26 in Supplement 2), suggesting that EWAS-nominated loci elicit their effects through mechanisms that cannot be broadly reconciled with cis-acting changes in gene expression (9,66).However, this analysis was relatively underpowered (40 individuals with matched gene expression data) and did not account for trans-acting effects or additional regulatory mechanisms that may alter processes such as splicing without necessarily impacting gene expression (9,67).In contrast, a strong correlation was observed among EWAS signals and schizophrenia-associated methylation quantitative trait loci mapped within the cohort, expanding on recent studies examining the interrelationship between genetic variation and epigenetics in psychiatric illness (Figures S11-S15 in Supplement 1; Tables S27-S31 in Supplement 2) (10,12,13,68).Indeed, recent evidence also suggests that schizophrenia-associated common variants are enriched within promoters and enhancers, particularly cortical H3K27ac domains (69), necessitating further exploration within larger cohorts to explore any putative relationships between common variants, DNA methylation, and gene expression related to features of schizophrenia.
We additionally constructed ESs in a manner analogous to polygenic risk scoring to investigate the utility of aggregated epigenetic risk for schizophrenia in terms of its clinical dimensions (35,70).Interestingly, ES TRS was nominally associated with clozapine administration, while PRS SZ and ES SZ exhibited no associations with the 5 clinical dimensions.The lack of association for PRS SZ and ES SZ = trait pairings on comparison of EWAS z scores for the ASRB traits (y-axis) and phenotypes/exposures from EWAS Catalog (x-axis).(C) Scatter plots comparing probe-level EWAS z scores for a representative trait, C-reactive protein (x-axis), and features of schizophrenia in the ASRB (y-axis).Pearson's correlation coefficient and associated p values are reported in the top left.(D) As in panel (B), except presenting epigenetic overlap among the ASRB and plasma protein concentrations for the top 30 (of 1454) significantly associated proteins.See Tables S14 and S15 in Supplement 2 for full results.ASRB, Australian Schizophrenia Research Bank; EWAS, epigenome-wide association study; GAF, Global Assessment of Functioning; PRS, polygenic risk score; resid, residualized.

Epigenetic Disruption in Features of Schizophrenia
Biological Psychiatry --, 2023; -:---www.sobp.org/journal11 potentially reflects the phenotypically heterogeneous nature of schizophrenia (71)(72)(73) and supports the need for genetic and epigenetic studies of specific clinical features.Furthermore, refinement of ESs to quantify epigenetic risk within clinically actionable pathways, analogous to recently developed genetic methods (21,74), may yield further utility for dissecting pathophysiological mechanisms and guide novel personalized intervention strategies that may include drug repurposing.To this end, ESs for plasma protein concentrations such as those in the current study could assist identification of seed genes for exploration of precision interventions; for instance, targets of anti-inflammatory drugs were collectively identified among proteins associated with features of schizophrenia.However, full EWAS summary statistics and high-resolution whole-genome bisulfite sequencing may be required to capture true epigenome-wide effects to enhance the predictive utility of these scores, while validation is also needed to confirm the present findings in independent, deeply phenotyped cohorts.
In conclusion, our findings reveal novel insights into the alteration of DNA methylation associated with important features of schizophrenia; however, we acknowledge several caveats to be considered when interpreting these results.First, our analyses are subject to the inherent limitations of EPIC BeadChips, which sample approximately 3% of the 28 million CpG sites throughout the genome (75); thus, some important epigenetic changes may remain undetected.Second, the reported methylation profiles are reflective of peripheral changes in blood, restricting conclusions regarding changes impacting the brain.Nevertheless, we note that using peripheral blood mononuclear cells is advantageous due to the minimally invasive nature of sample collection and ability to index the collective alteration of many tissues, given that schizophrenia is not exclusively associated with disruption of the brain (76)(77)(78).Our findings also require validation in independent, deeply phenotyped cohorts to confirm the discrete profiles of methylation reported in the current study.Finally, we acknowledge that our selection of covariates was not exhaustive; however, we addressed this by estimating E-values, which quantify the minimum association required between unmeasured confounders and each probe and trait to explain away the reported association, above and beyond measured covariates (30,79).Our analysis revealed that unmeasured confounders require a median 2.34-fold association with both DNA methylation and schizophrenia to account for the suggestive EWAS associations in the current study.We consider this scenario unlikely; however, it is plausible that factors such as environmental exposures (e.g., air pollution) may have affected a subset of probe-trait pairings to this extent (11).Similarly, it is possible that quantification of DNA methylation was affected by antipsychotic medications (80), which could not be readily accounted for due to a lack of quantitative dosage information collected within the ASRB protocol.In summary, our study presents important molecular insights into clinical dimensions of schizophrenia to further dissect phenotypic heterogeneity in the disorder and improve understanding of the molecular landscape for the development of personalized intervention strategies.

Figure 1 .
Figure 1.Investigating altered DNA methylation associated with clinical features of schizophrenia and common variant genetic risk.gDNA from 381 individuals with schizophrenia from the ASRB was subjected to bisulfite conversion and epigenome-wide analysis of DNA methylation.After initial data

Figure 3 .
Figure 3. Enrichment of DMPs within genomic features and genes associated with psychiatric conditions.(A) Forest plots depicting enrichment of suggestive DMPs for each of the 5 primary epigenome-wide association study traits and their PRS SZ -residualized counterparts within promoters and enhancers

Figure 4 .Figure 5 .
Figure 4. Differential methylation within genes associated with schizophrenia and comorbid psychiatric conditions.Forest plots of differentially methylated probes neighboring or residing in genes previously associated with schizophrenia and comorbid psychiatric conditions, as defined by curated and inferred gene sets from DisGeNET (45) and PsychENCODE (43) schizophrenia-associated gene sets.Note that DisGeNET and PsychENCODE schizophreniaassociated gene sets were merged for this analysis.Data presented as epigenome-wide association study b AE 95% CI.Labels on the y-axis depict the probe identification, associated gene, and location of the probe relative to the gene.See TableS12in Supplement 2 for full results.ADHD, attention-deficit/ hyperactivity disorder; ANX, anxiety; ASD, autism spectrum disorders; BD, bipolar disorder; GAF, Global Assessment of Functioning; IC, impaired cognition; MDD, major depressive disorder; PRS, polygenic risk score; SZ, schizophrenia.

Figure 6 .
Figure 6.Association between ESs and schizophrenia-associated traits.(A) Forest plots depicting effect sizes (6 95% CI) for regression models examining the 5 primary ASRB traits with respect to ESs constructed for SZ and TRS.ESs were constructed at 5 p EWAS value thresholds (y-axis) using summary statistics

Table 1 .
Hannon et al. in 2021 ( the 5 Clinical Dimensions AnalyzedESs for schizophrenia were derived using 1223 and 1048 schizophrenia-associated probes reported in meta-analyses from Hannon et al. in 2016(12)and in 2021(13), respectively.Scores for TRS were constructed using the same 1048 probes reported byHannon et al. in 2021 ( d t 208.38 = 21.8,p = .07t 252.68 = 24.18,p = 4 3 10 25 t 181.07 = 20.53,p = .6GAF, Global Assessment of Functioning; PRS SZ , polygenic risk score for schizophrenia.a t test comparing mean age between groups for categorical traits.b c 2 statistic comparing the proportion of males and females among groups for categorical traits.c Adjusted R 2 from a linear model comparing each continuous trait vs. reported age.d t test comparing each continuous trait between males and females.