This study represents the first comprehensive analysis of disease-associated DNA methylation differences in MZ twins discordant for SZ and BD, using a genome-wide approach. We found no alterations in global DNA methylation between affected and unaffected twins, but considerable disease-associated between-twin differences at specific loci across the genome. Some differences were consistently altered across a combined discordant psychosis group, whereas others appear to be specific to either SZ or BD. Furthermore, although many differences were identified across all discordant twin pairs for each diagnostic category, others were specific to one or a couple of pairs. Our hypothesis-free experimental design allowed us to identify disease-associated DNA methylation differences at loci not previously implicated in psychiatric disorders, but we also found evidence for DNA methylation differences at genes previously implicated in psychosis, such as GPR24 and CTNNA2 in BD. Pathway analysis of our top loci highlighted a significant enrichment of epigenetic disruption to biological networks and pathways relevant to psychiatric disease and neurodevelopment. Overall, our data provide further evidence to support a role for DNA methylation differences in the etiology of both SZ and BD.
The overall top-ranked psychosis-associated methylation difference identified in this study was at a CpG site in the promoter of ST6GALNAC1
, located at 17q25.1, which we found was hypomethylated in affected twins. Interestingly, we also found that >13% of postmortem brain samples from SZ and BD patients tested showed marked hypomethylation over an extended region encompassing the nominated CpG site, suggesting that epigenetic changes at this region may be present in a subset of psychosis patients. The psychosis-associated CpG site does not reside in a CpG island, reflecting the observation that our top disease-associated, differentially methylated loci are significantly under-represented for classical CpG-rich promoters and concurring with data from recent epigenomic profiling studies showing that phenotypically relevant variation in DNA methylation often occurs outside of such regions (29
is a member of the sialyltransferease family of molecules, involved in protein glycosylation, with a key role in mediating cell–cell interactions. ST6GALNAC1
is expressed in the brain and other CNS tissue and is differentially regulated during key periods of neurodevelopment (30
). We performed in silico
analysis of the specific CpG site (cg13015534) nominated in our study, using publicly available ENCODE ChIP-seq data in the UCSC genome browser (http://genome.ucsc.edu/
) (Supplementary Material, Fig. S7
). The site lies directly proximal to a peak of Pol2 occupancy and within a region able to bind neuron restrictive silencer factor/RE1-silencing transcription factor (NSRF/REST), a key regulator of neuronal differentiation that acts to silence neuronal genes in non-neuronal tissue (31
). ENCODE DNA methylation data generated using the Illumina 27K array shows that DNA methylation at cg13015534 is inversely correlated with Pol2 binding and ST6GALNAC1
expression, suggesting that variation in methylation at this locus is directly associated with gene transcription and function. Interestingly, a rare inherited genomic duplication spanning the gene, including the disease-associated CpG identified in this study, was reported recently in a single case of SZ (32
), suggesting that over-expression of the gene may be implicated in pathogenesis. Given that promoter hypomethylation is typically associated with increased gene expression, a disease-associated duplication of ST6GALNAC1
is consistent with our data. Hypermethylation at this locus has also been associated with estrogen and progesterone receptor-positive breast cancers (33
), highlighting that epigenetic variation in the ST6GALNAC1
promoter has important phenotypic consequences that may well be pleiotropic. Finally, another member of the sialyltransferease family of molecules, sialyltransferease 8B (SIAT8B
), has been previously associated with SZ (34
) and animal studies indicate that ablation of this gene leads to incomplete polysialylation of the neuronal cell adhesion molecule causing severe defects in the anatomical organization of the forebrain (35
), which potentially provides a direct mechanism for the role of this gene in the predisposition to SZ.
A number of the other top-ranked genes nominated by our study have been previously implicated in SZ and BD. For example, animal models and human genetic studies suggest that GPR24
(22q13.2), encoding G protein-coupled receptor 24, is involved in the susceptibility to BD. GPR24
knock-out animals show dysregulation of the mesolimbic dopamine system (36
), a key system in psychosis. Furthermore, GPR24
antagonists have antidepressant and anxiolytic effects, implicating this gene in the regulation of mood (37
). There is also genetic evidence to suggest a direct involvement of GPR24
in psychosis; it lies within one of the strongest linkage peaks nominated by a comprehensive meta-analysis (39
), and polymorphisms in the gene have been associated with both BD and SZ (40
). Interestingly, our significant CpG site is located only 21 bp from rs133073 (although not overlapping the probe), one of the SNPs strongly associated with psychosis.
The alpha catenin gene (CTNNA2
) is located in an SZ linkage region 2p11 (41
), and encodes a neuronal cadherin-associated protein that plays a major role in folding and lamination of the cerebral cortex (42
). This gene has been previously identified as a susceptibility gene in a genome-wide association study and meta-analysis of BD comprising 3683 cases and 14 507 controls (43
). Furthermore, polymorphisms in LRRTM1
, a gene residing in an intron of CTNNA2
, have been associated with psychosis in a parental-origin-specific manner (44
). Schalkwyk et al
) found evidence for allele-specific methylation in this region supporting the theory that this region could be imprinted and mediating risk via an interaction between genetic and epigenetic factors.
Previous studies investigating MZ-discordant pairs for psychosis have been limited to one or a few twin pairs. One strength of our study is that by using DNA samples from 22 discordant MZ twin pairs, representing one of the largest twin studies performed for any complex disease phenotype to date, we were also able to identify DNA methylation differences that are consistently changed across multiple discordant MZ twin pairs in addition to identifying family-specific disease-associated epigenetic changes. The only other large-scale genome-wide study of DNA methylation changes associated with psychosis investigated postmortem brains from patients and controls and found a number of highly significant alterations in DNA methylation (18
). Interestingly five of our top 100 ranked genes overlap with those found to be significantly associated with psychosis in that study—GGN
(Supplementary Material, Table S10
)—suggesting that epigenetic variation at these loci deserves further investigation. Individual twin pair analysis in our study also highlighted considerable familial heterogeneity, with some pairs showing much greater disparity in DNA methylation at certain loci. Furthermore, we observed some large methylation changes that were specific to one or a few twin pairs (Supplementary Material, Table S5
), suggesting that some rare epigenetic alterations of large effect may be associated with psychosis.
Our study has a number of limitations that should be considered when interpreting the data presented here. First, although this represents the largest disease-discordant MZ twin study to date for any complex psychiatric disorder, our analysis was limited to only 22 pairs of twins. A power calculation based on an estimate of the standard deviation of the whole data set from the Illumina array data shows that, at a stringent Bonferroni-corrected P
-value cut-off (1.86E − 06), our sample of 22 discordant MZ pairs gives good power (>80%) to detect a Δβ
of 0.06, although the power to detect smaller differences at this level of significance is more limited, as are analyses in the diagnostic subgroups. It is noteworthy that using our weighted t
-test based on the magnitude of change observed between affected and unaffected twins at each locus, many of our top-ranked loci are highly significant, with FDR values (q
-values) <0.05. The significant over-enrichment of biologically relevant pathways in our IPA network analyses of top-ranked loci adds further weight to the power of our study to uncover valid differences. Because the discordant MZ-twin design is such a powerful tool for identifying epigenetic variation associated with phenotype (19
), a notion affirmed by our success in identifying disease-associated epigenetic changes, future efforts should focus on collecting biological material from additional twin pairs for use in methylomic analyses.
Second, by necessity, this study was performed on DNA samples extracted from peripheral blood rather than from the brain. To our knowledge, there is no archived collection of postmortem MZ twin brains discordant for psychosis. Although the epigenome clearly shows tissue-specific patterns, there is mounting evidence from other disorders that disease-associated epimutations may be detectable across tissues (46
). Our own methylomic profiling across the brain and blood from multiple individuals shows that although tissue-specific variation far outweighs between-individual variation, many between-individual differences are observed across tissues (unpublished data, see http://epigenetics.iop.kcl.ac.uk/brain
). In this regard, it is reassuring that we find evidence for disease-associated hypomethylation of the ST6GALNAC1
promoter in postmortem brain tissue from affected patients, and that some of the genes nominated from a previous study on brain samples were also in our list of top-ranked loci (Supplementary Material, Table S10
). Furthermore, our IPA data demonstrates that we can clearly identify epigenetic changes to disease-relevant pathways from peripheral samples. A recent study by Kaminsky et al
) found BD-associated methylation changes in the promoter of HCG9
in both brain and peripheral samples, and other reports have also successfully used peripheral tissues to identify disease-associated epigenetic changes in psychosis (24
). Finally, the use of peripheral samples may have some advantages over the use of the brain; sample collection is more standardized between twins and is not subject to potential confounding issues such as postmortem delay (8
Third, it is hard to draw conclusions about causality for any of the differentially methylated regions identified in our study, in part because we do not have DNA samples from the twins taken before they became discordant for disease. It is plausible that many of the changes we identify have occurred downstream of the disease, resulting from exposure to antipsychotic medication, for example. There is mounting evidence that many of the drugs used to treat both SZ and BD induce epigenetic changes (50
). Such changes are interesting, however, and understanding the pathways via which these drugs work may provide information about the neurobiological processes involved in disease. The ideal study design would be to longitudinally assess DNA methylation changes in the brain during an individual's' transition into disease, although such a study is clearly unfeasible at present.
Finally, although the microarray platform utilized in this study (the Illumina 27K Methylation Array) is reliable and has been previously used to identify quantitative changes in DNA methylation (51
), it is relatively limited in the density of probe coverage, interrogating only a couple of CpG sites for the majority of genes with a bias to loci implicated in cancer. This actually makes the results of our pathway analysis more striking, given the clear over-representation of DMRs related to neurobiological processes. Future work should build on the recent technological advances in methylomic profiling and examine DNA methylation differences between discordant twins at much higher resolution.
To conclude, we have undertaken the first large-scale analysis of disease-associated epigenetic changes in MZ twins discordant for major psychosis identifying numerous DNA methylation differences associated with both SZ and BD. These findings further support the notion that epigenetic processes probably play an important role in neuropsychiatric disease and highlight the power of the discordant MZ twin design for epigenomic studies of complex disease.