DNA methylation differences in cortical grey and white matter in schizophrenia

Abstract

Aim:

Identify grey- and white-matter-specific DNA-methylation differences between schizophrenia (SCZ) patients and controls in postmortem brain cortical tissue.

Materials & methods:

Grey and white matter were separated from postmortem brain tissue of the superior temporal and medial frontal gyrus from SCZ (n = 10) and control (n = 11) cases. Genome-wide DNA-methylation analysis was performed using the Infinium EPIC Methylation Array (Illumina, CA, USA).

Results:

Four differentially methylated regions associated with SCZ status and tissue type (grey vs white matter) were identified within or near KLF9, SFXN1, SPRED2 and ALS2CL genes. Gene-expression analysis showed differential expression of KLF9 and SFXN1 in SCZ.

Conclusion:

Our data show distinct differences in DNA methylation between grey and white matter that are unique to SCZ, providing new leads to unravel the pathogenesis of SCZ.

Lay abstract

This study investigated the way gene activity is regulated in brain cells of patients with schizophrenia (SCZ; a severe mental illness characterized by psychosis) compared with unaffected controls. The study focuses on the differences between parts of the brain with many cell bodies (grey matter) in contrast to those parts with mainly conducting fibers (white matter). For that purpose, grey and white matter were separated from brain tissue of ten individuals with SCZ and 11 without. All brains were obtained after the patients died and donated their brains to science. Array technology was used to analyze 800,000 sections of the DNA at once. The study identified regions on four genes that can turn the genes on and off differently in schizophrenic patients compared with controls, these genes were also turned on or off depending on their location either in grey or white matter. Two of these genes showed different activation in schizophrenic patients compared with controls. Overall this study identified distinct differences between grey and white matter that are unique to SCZ, providing new leads to unravel the biology of SCZ.

Schizophrenia (SCZ) is a complex chronic psychiatric disorder characterized by a heterogeneous spectrum of symptoms like hallucinations, lack of motivation and cognitive impairment. Neuro-imaging and postmortem brain studies have identified multiple cortical abnormalities in SCZ patients, such as a decrease in grey and white matter volume and density, disruption of white matter connectivity, and changes in gyrification and cortical thickness [1–10].

Epigenetic regulation that acts as a mediator of both genetic differences [11] and environmental risk factors that are involved in SCZ pathology [12–16], is hypothesized to underlie these cortical differences in SCZ. One of the best characterized epigenetic modifications is DNA methylation, the process of adding a methyl group (CH3) to a CpG dinucleotide. Methylated cytosines impact gene transcription with varying directionality [17].

Much remains unknown about the possible role of DNA methylation in SCZ. There have been a large number of epigenetic studies that analyzed DNA methylation differences in peripheral tissue in SCZ (recently reviewed by Smigielski et al.) [18]. However, the number of studies on brain tissue of SCZ patients and control donors using an epigenome-wide approach remains limited [19–29]. Recent studies have explored the role of tissue-specific DNA methylation. Although compelling evidence suggests that individual differences exceed differences in tissue type, substantial epigenetic variation has been detected between functionally distinct regions of the brain and between different brain cell types [30,31]. Therefore, an effort is made to identify epigenetic differences specific to the primary tissue type that is relevant for SCZ in the brain. In addition, significant differences have been detected between cortical grey and white matter in function, representation of cell populations, developmental trajectories and epigenetic marks, including DNA methylation [32–35].

This study aims to identify tissue-specific DNA methylation differences between SCZ and healthy control (CTR) postmortem brain tissue from two cortical areas (gyrus frontalis medialis [GFM] and gyrus temporalis superior [GTS]). To this end, we separated grey and white matter from the brain specimens, followed by epigenome-wide methylation analysis using the Infinium Methylation EPIC Bead Chip (Illumina). The study focuses on differences in grey and white matter between SCZ and CTR donors. Differences between patients and controls, independent from the tissue type (grey vs white matter), have previously been identified. Therefore, in this study, we aimed to identify differences between SCZ and CTR that are distinct between grey and white matter tissues. The strongest DNA methylation findings were validated by gene expression analyses. The resultant data contribute to our understanding of grey–white matter (GW) DNA methylation differences in SCZ and provide valuable targets for follow-up research of the pathogenesis of SCZ.

Materials & methods

Donors

Postmortem brain tissue of SCZ patients and CTR donors was obtained from the Netherlands Brain Bank (NBB; Amsterdam, The Netherlands) and the Edinburgh Brain and Tissue Bank (Edinburgh, UK). We obtained freshly frozen (SCZ: n = 10; CTR: n = 11) tissue from the rostral side of the GFM (area 1 and 2), and the middle of the GTS (area 2 –4; Figure 1A). A detailed description of the anatomical dissection boundaries of these regions used during autopsy can be found on the website of the NBB (www.brainbank.nl/brain-tissue/autopsy/). These brain regions have previously been associated with the pathogenesis of SCZ [36–38]. At the NBB, exclusively Dutch, donors are registered ante-mortem, giving permission for brain autopsy and the use of these materials and medical records for research purposes. The NBB-psychiatry program focuses especially on the establishment of a resource of brain tissue of major psychiatric disorders, together with five Dutch university medical centers. They actively approach extensively phenotyped participants of 15 large psychiatric research cohorts to register as prospective brain donor at the NBB [39]. At the Edinburgh Brain and Tissue Bank, permission for brain autopsy is obtained (verbally and written) from the next of kin upon sudden deaths, occurring in the general population, reported by police [40]. The Edinburgh Brain and Tissue Bank gains access to general practice and hospital records, including psychiatric records, and confirms the absence of illness with the next of kin [40]. CTR donors did not have a history of major neurological or psychiatric disorders and SCZ donors were diagnosed according to the DSM-III/IV criteria. At the Edinburgh Brain and Tissue Bank, two psychiatrists assess the clinical information to confirm a psychiatric diagnosis [40]. At the NBB, the diagnoses of participants were confirmed by trained research assistants using standardized instruments (like the Structured Clinical Interview for DSM-IV [SCID-I], Mini International Neuropsychiatric Interview [M.I.N.I.] plus, Comprehensive Assessment of Symptoms and History [CASH] or Composite International Diagnostic Interview [CIDI]) [39]. Furthermore, the clinicopathological information from all donors of the NBB was reviewed by two individual psychiatrists in our group (LdW and GS) to confirm a SCZ diagnosis. SCZ and CTR donors were matched by age and sex. The tissues from all donors were checked for neuropathology by a pathologist and excluded in case of neurodegenerative pathology in the cortex. For each donor, we obtained information on postmortem delay (PMD), pH, age, sex, onset of disease, cause of death, Braak- and amyloid score, substance dependency (smoking, alcohol and drugs), somatic and DSM comorbidities, when available (Table 1 & Supplementary Table 1). Four main possible confounding factors (sex, age, pH and PMD) were assessed for difference on group level with Mann–Whitney and Chi-square tests where appropriate (significance level p < 0.05). Brain samples from the Edinburgh Brain and Tissue Bank were only used for the analysis of RNA expression differences at the identified regions with DNA methylation differences and as a consequence comparisons between the two sample sources are not informative.

Figure 1. . Schematic overview of the study approach.

(A) Tissue from the GFM area 1 and 2 and GTS area 2–4 were obtained from SCZ (n = 12) and CTR (n = 14) donors, as indicated by the red circles. The Nissl staining shows grey and white matter, which were separated by hand and processed for DNA isolation and RNA isolation for DNA methylation analysis by the Infinium EPIC Methylation Array (Illumina, CA, USA) and gene expression analyses by qPCR, respectively. (B) Results from the DNA methylation array were analyzed in two steps. Step 1: the differences of CpG methylation between grey and white matter were assessed for SCZ and CTR donors separately and then compared. CPGs that differently regulated between grey and white matter in SCZ compared with CTR donors were selected. Examples are shown in the right panel. This analysis resulted in 24 DMRs. Step 2: for further gene expression analysis we selected DMRs that were significantly different when comparing grey and white matter and SCZ and CTR donors. Examples are shown in the right panel. There were four DMRs that met these criteria.

CTR: Control; DMR: Differentially methylated region; GFM: Gyrus frontalis medialis; GTS: Gyrus temporalis superior; qPCR: Quantitative PCR; SCZ: Schizophrenia.

Table 1. . Summary of characteristics of donors from the Netherlands Brain Bank that were used in the study.

Methylation	SCZ (n = 9)	CTR (n = 11)
Sex (% male)	45	55
Age (years)	74 (11)	78 (15)
PMD (hours)	24 (31)	8 (4)
Brain pH	6.4 (0.4)	6.6 (0.3)

Gene expression	SCZ (n = 11)	CTR (n = 14)
Sex (% male)	30	45
Age (years)	67 (13)	74 (17)
PMD (hours)	34 (31)	24 (33)
Brain pH	6.4 (0.3)	6.5 (0.4)

Summary of main clinical and postmortem variables of SCZ and CTR donors within this study (sex, age, PMD and pH). Values represent mean and standard deviation. Sex is represented as the % of male donors. None of these possible confounders differed between SCZ patients and CTR donors on group level as analyzed with Mann–Whitney and Chi-square test where appropriate (p > 0.05).

CTR: Control; PMD: Postmortem delay; SCZ: Schizophrenia.

Ethics statement

The Netherlands Brain Bank and Edinburgh Brain and Tissue Bank received permission to perform autopsies, and to use donor tissue and medical records, from the VU medical center Ethical Committee (VUMC, Amsterdam, The Netherlands) and The East of Scotland Research Ethics Service REC1 (Edinburgh, UK) respectively. Written informed consent was obtained from brain donors antemortem (The Netherlands Brain Bank) or from relatives postmortem (Edinburgh Brain and Tissue Bank).

DNA isolation & bisulfite conversion

For DNA isolation grey and white matter from the GFM and the GTS from nine SCZ and 11 CTR donors were included. Fresh frozen tissue was sectioned (50 μm) using a cryostat (Leica Biosystems, Nussloch, Germany) after which grey and white matter were dissected manually. This resulted in the inclusion of 18 grey mater GFM samples (nine SCZ and nine CTR), 18 white matter GFM samples (nine SCZ and nine CTR), 17 grey matter GTS samples (seven SCZ and ten CTR) and 14 white matter GTS samples (six SCZ and eight CTR). The grey and white matter tissues were dissociated separately using the Ultra Turrax Homogenizer (IKA, Staufen, Germany) and stored at -80°C. DNA was isolated using the QIAamp DNA mini kit (Qiagen, Hilden, Germany) following manufacturer instructions and the DNA concentration measured with the Nanodrop Spectrometer (Thermo Fisher Scientific, MA, USA). Subsequently, DNA was sodium bisulfite converted using the EZ DNA Methylation kit (Zymo Research, CA, USA) according to manufacturer instructions.

Methylation determination & quantification

Genome-wide DNA methylation was quantified using the Infinium MethylationEPIC BeadChip Kit (Illumina, Inc., CA, USA). Data were pre-processed and analyzed using the R program for statistical computation (version 3.3.1, R core team). The level (proportion) of methylation is expressed as a β-value, ranging from 0 (unmethylated cytosine) to 1 (completely methylated cytosine), the subsequent analyses were performed using M-values (log² of beta values) for a better statistical validity [41]. The dataset was pre-processed using the meffil package [42] and functional normalization [43] on the entire dataset (n = 68), including DNA samples from both brain regions. One sample was excluded because the reported sex did not match the methylation-predicted sex. The absence of sample mismatch was verified using the genotype concordance estimate function in meffil. No samples had to be excluded due to high proportion of undetected probes (>10% of the probes in the sample had a detection p > 0.01) or for a low number of beads (>10% of the probes in the sample had a bead count less than three). Ultimately, 67 out of 68 DNA samples survived quality control. Quality control of probes resulted in the removal of 4610 probes (0.54 of total probes) because >10% of the samples had a detection p-value > 0.01. A total of 68 probes had to be removed due to a low number of beads per CpG (<3) in a high proportion (>10%) of the samples. SNP-related probes were removed [44,45] as were probes with potential cross-hybridization [45,46]. Independent surrogate variables were derived to capture technical variation and potential confounding variables [47] as implemented in meffil.

Epigenome-wide association study

To identify distinct methylation patterns associated with SCZ, a linear mixed-effects model was used. The model was run with CpG methylation as outcome, group (SCZ/CTR) and GW as main determinants, gender, age, genetic principal components as measure of ancestry (PC), and independent surrogate variables as covariates, and the brain region (GTS/GFM) as random factor. The analysis included an interaction term of grey–white and group. The full model was: Meth ∼group + GW + GW*group + age + gender + PCs + ISVA, random = brain region. Multiple testing was accounted for by applying a false discovery rate at 5% [48].

Identification of differentially methylated regions

Based on the p-value for the interaction coefficient of the CpGs, differentially methylated regions (DMRs) were determined. A DMR consists of highly significant CpGs (p < 0.0001) within proximity of 1000 base pairs. DMRs were defined as implemented DMRcate [49]. DMRcate uses a Stouffer procedure to derive a type I error percentage (p-value) for each DMR [50]. The results were inspected by rerunning the same linear model with mean methylation across each DMR using the rlmer package in R (Koller). To investigate whether there were regional differences in GW methylation within a DMR, an additional linear regression was conducted, with DMR methylation as outcome and brain region (GTS, GFM) as determinant. These models were run for each DMR individually. To identify DMRs with the strongest association to SCZ pathology, we prioritized DMRs that showed a significant main effect for DNA methylation between SCZ and CTR, grey and white matter, as well as a significant interaction effect. The significance threshold was set at p < 0.05. Figure 1 presents a schematic overview of the study approach.

Quantification of gene expression using qPCR

For RNA isolation, fresh frozen tissue was sectioned (50 μm) using a cryostat after which grey and white matter was dissected manually. Tissue was dissociated in TRIzol reagent (Thermo Life Technologies, CA, USA) using the Ultra Turrax Homogenizer (IKA) and stored at -80°C. RNA extraction, cDNA preparation and quantitative PCR (qPCR) were performed as described before [51], using the chloroform/isopropanol method, the Quantitect Reverse Transcription Kit (Qiagen) and the QuantStudio™ 6 Flex Real-Time PCR System (Life Technologies Corporation, NY, USA). Primers for KLF9, SFXN1, and SPRED2 were designed to be intron spanning, using the NCBI primer-BLAST tool (NCBI, MD, USA; Supplementary Table 2). It was not possible to develop a primer pair to detect ALS2CL gene expression in our postmortem brain tissue (previous studies have also been unsuccessful in this regard [52,53]). KLF9, SFXN1 and SPRED2 gene expression levels were normalized to the geometric mean of the three stable reference genes βactin, GAPDH and SDHA [54]. The geometric means of the raw computed tomography values from these three reference genes are included for each donor and each tissue in Supplementary Table 3. Statistical analyses and figure design were performed with Graphpad software (version 8.4; Graphpad, CA, USA). Outliers were defined when values were ± three-times the standard deviation from the mean for each gene, leading to the exclusion of three values (GFM-white 99/131 SPRED2, GFM-grey 12/001 SFXN1 and GTS white SD019-15 SPRED2). Unpaired Mann–Whitney U tests were used to compare gene expression between SCZ and CTR donors in GTS and GFM tissues, because data were not normally distributed. We assessed the effect of five main potential confounding factors, in other words, age, sex, brain pH, PMD and Braak-score for each datapoint with Spearman Rho correlation or the Kruskal–Wallis test as appropriate (Supplementary Table 4). When a significant association was found (Bonferroni corrected p < 0.001), the variable was entered as a covariate in post hoc analyses using logistic regression.

Methylation – gene expression correlation analyses

Spearman Rho correlation analyses were applied to determine the main correlation effect between methylation- and gene expression level.

Confounder analysis

To assess potential confounding, a Pearson correlation was computed to assess the relationship of DMR methylation and brain pH, and Braak-score. No association was detected between these measures. Age and sex were included in the mixed linear regression model and, therefore, were not assessed separately for association with DNA methylation. Also, Spearman correlation and Kruskal–Wallis tests with the preselected potential confounding variables (sex, age, PMD, pH and Braak-score) revealed no significant correlations with gene expression (Supplementary Table 5).

Data availability statement

Relevant data supporting the discussed findings are included in the paper and its Supplementary Material. Data are available at reasonable request pending data sharing agreements.

Results

Discovery

Epigenome-wide association analysis revealed 104 CpGs with coefficients below our significant threshold for interaction between GW differences and SCZ-CTR status (p < 0.0001). Inspection of the Q–Q plots showed no indication of inflation with a lambda of 0.91 (Supplementary Figure 1). Of the significant CpGs, 69 were within a 1000 base pair range of each other, resulting in the identification of 24 DMRs (Figure 1 & Supplementary Table 5).

DMR analysis

Among the 24 DMRs, we identified four that showed a significant main effect for DNA methylation between SCZ and CTR, grey and white matter, as well as a significant interaction effect (Table 2). All DMRs were located either within or close to the annotated gene. The first DMR (chr9: 73034238–73034460) is located 4.6 kb outside the KLF9 gene, within a CpG island of a promoter flanking region. The KLF9 DMR showed higher methylation in grey versus white matter and in SCZ versus CTR subjects. The second DMR (chr5: 174944792–174944864) was found in the 8th intron of the SFXN1 gene, associated with epigenetic regulation. The third DMR (chr2: 65540924–65541060) overlaps with a CpG island in the last protein-coding exon of the SPRED2 gene. Both the SFXN1 and the SPRED2 DMRs showed higher methylation in white versus grey matter and in SCZ versus CTR subjects. The last DMR (chr3: 46734995–46735009) is located at a CpG island in the first exon and promoter region of the ALS2CL gene. This DMR showed higher methylation in grey versus white matter and in CTR versus SCZ subjects (Supplementary Figure 2 & Supplementary Table 6).

Table 2. . Four differentially methylated regions showed a significant main effect of DNA methylation between schizophrenia and control as well as between grey and white matter.

DMR	DNAm – SCZ/control					DNAm – grey/white
	DNAm SCZ	DNAm CTR	ß (Beta)	t-value	p-value	DNAm grey	DNAm white	ß (Beta)	t-value	p-value
KLF9	0.048	0.044	0.009	3.26	0.001	0.053	0.038	0.006	2.02	0.042
SFXN1	0.811	0.806	0.014	2.05	0.04	0.732	0.891	0.025	3.59	0.0003
ALS2CL	0.035	0.0465	0.037	-3.32	0.0009	0.06	0.021	-0.026	-2.07	0.038
SPRED2	0.942	0.929	0.02	4.03	0.00005	0.927	0.944	0.01	1.96	0.05

The table shows the methylation difference of the DMRs between SCZ and CTR and between grey and white matter. The β-value represents the log² of the methylation percentage (0 = unmethylated, 1 = completely methylated). The significance level was set to p < 0.05.

CTR: Control; DMR: Differentially methylated region; SCZ: Schizophrenia.

Gene expression analyses

Following the DMR analyses, we assessed gene expression differences between SCZ and CTR donors for the three of the identified genes (KLF9, SFXN1, and SPRED2) using qPCR. For this analysis, we separated grey and white matter for the two different brain regions (GTS and GFM) in SCZ and CTR subjects. Outliers were removed, and significance corrected for multiple testing using Bonferroni correction (p < 0.0042). KLF9 gene expression was downregulated in SCZ in GFM grey matter (SCZ: median = 391.8; interquartile range [IQR] = 438.6; CTR: median = 977.6; IQR = 421.2; p = 0.0021; Figure 2A). For SFXN1, we also found a downregulation in GFM grey matter in SCZ (SCZ: median = 139.8; IQR = 162.72; CTR: median = 488.9; IQR = 208.5; p = 0.0022; Figure 2B). The SPRED2 gene was downregulated in GFM white matter in SCZ (SCZ: median = 98.6; IQR = 44.9; CTR: median = 193.8; IQR = 70.3; p = 0.0023; Figure 2C). An overview of the outcome values and the n for each test can be found in Supplementary Table 7. We did not find a correlation between gene expression and methylation level for these genes (Supplementary Figure 3).

Figure 2. . Gene expression in grey and white matter of the GFM and GTS in schizophrenia and control.

mRNA expression of genes was determined by qPCR in the GFM and GTS of SCZ patients (n = 11) and CTR donors (n = 14). Gene expression was normalized against reference genes βactin, SDHA, and GAPDH. Graphs show boxplots (white for CTR and grey for SCZ) containing individual data points for CTR (circles) and SCZ (squares) indicating the median, 25th and 75th percentile, with the whiskers showing the maximum and minimum values. Significance level was Bonferroni corrected and set to *p < 0.0042.

CTR: Control; GFM: Gyrus frontalis medialis; GTS: Gyrus temporalis superior; qPCR: Quantitative PCR; SCZ: Schizophrenia.

Discussion

This study identified grey versus white matter DNA methylation differences between SCZ and CTR postmortem cortical tissue. Four DMRs (within or close to KLF9, SFXN1, SPRED2 and ALS2CL) were differentially methylated both in grey versus white matter and in SCZ versus CTR. The observation of differential gene expression of these genes supported their role in SCZ. These findings provide novel targets for follow-up research and contribute to our understanding of the importance of the cortical grey and white matter differences in general, and in studying SCZ pathogenesis in particular.

The KLF9 DMR, is located in an active promoter of the KLF9 gene, which is involved in neuronal maturation [55], axon-growth suppression [56,57] and part of the thyroid hormone 3 (T3) signaling cascade [58,59]. In the central nervous system T3 is associated with oligodendrocyte differentiation [60], and therefore the development and regeneration of myelin. In mice, increased T3 levels are associated with decreased white matter volume [61]. The KLF9 gene, a transcriptional target of T3, transduces these promyelinating effects of T3 [58,59].

The SFXN1 gene is a mitochondrial serine transporter, important for one-carbon metabolism [61,62]. In general, mitochondrial dynamics are important for neuronal development and synaptic transmission, and impaired mitochondrial function has been associated with an increased risk for developing SCZ [63–65]. Interestingly, mice lacking the SFXN1 gene show striking phenotypic similarities to mice deprived of pyridoxine (a form of vitamin B6) [66], although pyridoxine does not seem to be a substrate of SFXN1 [62]. Low levels of vitamin B6 are associated to SCZ pathology [67] and vitamin B6 is often added to antipsychotic medication, reducing symptoms [68].

The third DMR is located in the SPRED2 gene, which is expressed in the nervous system and functions as an inhibitor of the RAS/ERK-MAPK pathway [69,70]. The family of SPRED proteins is involved in neurogenesis, cell proliferation, migration and synaptic vesicle transport [69,71]. The loss of SPRED2 in a knock-out mouse model has been associated to obsessive compulsive disorder like behavior [69].

At last, ALS2CL is a homolog of ALS2, a guanine-nucleotide exchange factor for Rab5 involved in endosome dynamics, can alter ALS2 function [72,73]. Rab5 mediated endocytosis is essential for proper cortical neurodevelopment, neuronal migration and synaptic functioning [74,75]. Previously, mutations in this gene have been associated to juvenile recessive amyotrophic lateral sclerosis [76,77], however, a nonsense de novo mutation of ALS2CL has also been associated with SCZ pathology [78,79].

Our results show that the methylation rate of these four DMRs has a different relationship between SCZ and CTR subjects in grey matter compared with white matter. When we assessed the direction of these effects between SCZ and CTR subjects, we found that the DMRs show an opposite direction in methylation rate in grey matter compared with white matter, except for SPRED2. Comparing SCZ to CTR subjects, both KLF9 and SFXN1 show hypomethylation in white matter, and hypermethylation in grey matter, while ALSCL2 shows hypermethylation in white matter, and hypomethylation in grey matter. SPRED2 is hypermethylated in SCZ in both grey and white matter.

To assess the biological relevance of the four identified DMRs, we performed gene expression analyses and comparisons to previously published RNA-sequencing, genome-wide association studies and methylation studies. Our analysis showed a downregulation of KLF9 in grey matter of the GFM, SFXN1 in GFM grey matter and SPRED2 in GTS white matter.

These results reflect the differences in grey/white methylation between SCZ and CTR subjects for the GFM, but not for the GTS. In line with our data, SFXN1 gene expression was also downregulated in SCZ patients in the largest psychiatric RNA-sequencing study to date [80]. In this context, it is noteworthy that the genes identified in our study were not previously identified in genetic studies for SCZ [80–82], but some were highlighted in the other genome-wide methylation studies (Supplementary Table 8) [18].

Interestingly, all four identified genes are involved in functions related to brain connectivity (oligodendrocyte function and synaptic transmission) either directly or through their downstream interactions. Large numbers of imaging and postmortem brain studies have provided evidence for altered brain connectivity in SCZ (the dysconnectivity hypothesis) [83–89]. Several pathological mechanisms have been proposed to be the source of this dysconnectivity, affecting the wiring (white matter) and/or synaptic function (grey matter) in SCZ [85,87,90–95]. Our study might provide new leads to understanding this phenotype and points to a contribution of altered DNA methylation of the four identified genes.

Not much is known about grey and white matter DNA methylation differences in the cortex [32,96]. And these differences have not been considered in SCZ methylation studies until now. Considerable amount of the variance in our data is explained by GW differences, as is shown by the large correlation between the first principal component of the DNA methylation data and the tissue-type (Supplementary Figure 4). This emphasizes the advantages to investigate grey and white matter separately to pinpoint relatively small tissue-specific alterations associated with SCZ pathology.

A limitation of this study is that SCZ is one of the most complex psychiatric disorders, with patients experiencing a pleiotropy of symptoms, differences in disease severity, progression and treatment. Also, the nature of DNA methylation is very dynamic, which makes epigenetic studies susceptible to biological, clinical and lifestyle confounding factors [97]. In the current study, we have adjusted for five main confounding factors (age, PMD, sex, pH and Braak-score). However, other potential confounders have not been included, such as smoking, antipsychotic use and disease severity and some information are missing. Therefore residual confounding may have remained. Sample size is a central limitation of the current study. Given the scarcity of brain samples and the ability to separate grey and white matter from them, the current sample remains unique and despite sample size limitation we find differences in four genes on the epigenetic as well as on the transcription level. Also, additional pyrosequencing may have provided a more dense understanding of the regional methylation patterns. The controls were much older and despite adjusting for age it limits the generalization of the findings. Overall, the heterogeneity of the disease and epigenetic modifications, together with the limited availability of brain tissue, pose a challenge for this type of postmortem brain studies. Therefore, replication studies are imperative.

Conclusion

Overall, in a methodically unique way, dissecting grey and white matter before running epigenome-wide association study analysis, this study informs about epigenetic variation in the cortical grey and white matter specific to SCZ pathology, resulting in the identification of four genes that are likely involved in SCZ: KLF9, SFXN1, SPRED2 and ALS2CL. Furthermore, the dataset we generated provides a valuable resource for the identification of GW DNA methylation differences in the cortex in general. In the future similar approaches could be utilized to identify more grey and white matter-specific changes, thus contributing to the elucidation of SCZ pathogenesis.

Future perspective

As insight in the complexity of transcriptional regulation grows, it is becoming increasingly clear that there are vast differences in epigenetic regulation between tissue, cells and cell population. The current study demonstrates that such differences are very relevant for understanding the underlying biology of SCZ but at the same time reflects the difficulties that more sophisticated analysis (by tissue) encounter. Given the accelerated advances in single cell and single nucleus technology, such approaches will soon become mainstream but the extra level of complexity also increases sample size requirements. Therefore technologies that depend on physical separation of tissues such as in this study will remain of great value to pilot relevant tissue-specific hypothesis.

Summary points.

• DNA methylation could mediate biological mechanisms underlying cortical differences between grey and white matter of schizophrenia (SCZ) patients.

• Previous genome-wide DNA methylation studies in SCZ did not distinguish between grey and white matter.

• Our study contained postmortem brain tissue from the superior temporal gyrus and the medial frontal gyrus of ten SCZ patients and 11 control donors.

• We identified four differential methylated regions between grey and white matter and SCZ and control samples (KLF9, SFXN1, SPRED2 and ALS2CL).

• Gene expression analysis showed differential expression of KLF9 and SFXN1 in SCZ cases.

• Our study shows that there are differences in grey and white matter methylation that are unique to SCZ pathology, but confirmation is required in further studies with increased sample sizes.