Prefrontal cortex eQTLs/mQTLs enriched in genetic variants associated with alcohol use disorder and other diseases

Abstract

Aim:

This study aimed to investigate the function of genome-wide association study (GWAS)-identified variants associated with alcohol use disorder (AUD)/comorbid psychiatric disorders.

Materials & methods:

Genome-wide genotype, transcriptome and DNA methylome data were obtained from postmortem prefrontal cortex (PFC) of 48 Caucasians (24 AUD cases/24 controls). Expression/methylation quantitative trait loci (eQTL/mQTL) were identified and their enrichment in GWAS signals for the above disorders were analyzed.

Results:

PFC cis-eQTLs (923 from cases+controls, 27 from cases and 98 from controls) and cis-mQTLs (9,932 from cases+controls, 264 from cases and 695 from controls) were enriched in GWAS-identified genetic variants for the above disorders. Cis-eQTLs from AUD cases were mapped to morphine addiction-related genes.

Conclusion:

PFC cis-eQTLs/cis-mQTLs influence gene expression/DNA methylation patterns, thus increasing the disease risk.

Alcohol use disorder (AUD), including alcohol abuse and dependence, is a psychiatric condition caused by the interaction of genetic and environmental factors. It is characterized by compulsive alcohol use, loss of control over drinking and alcohol withdrawal symptoms. About 6.2% (or 15.1 million) American adults aged 18 and above have AUD [1]. Quantitative genetic studies suggest a heritability of approximately 50% for AUD [2]. To understand the genetic mechanisms of AUD, a number of genome-wide association studies (GWAS) have been conducted [3–9]. AUD-associated variants or SNPs (identified by GWAS) are mainly located in noncoding regions, although several AUD-associated coding variants have been found in alcohol metabolizing genes including ADH1 and ALDH2 genes. Similarly, GWAS has also identified a large number of noncoding variants associated with other psychiatric disorders. For protein-coding variants, it is relatively easier to understand their function as they may change the encoded protein sequence or influence mRNA stability. It is a challenge to interpret the function of disease-associated noncoding variants. Additionally, several studies explored the epigenetic mechanisms of AUD by examining AUD-associated DNA methylomic changes [10–12].

Disease-associated noncoding variants potentially affect gene transcription, alternative splicing and allele-specific expression, and thus contribute to an increased/decreased risk of the disease. Functional study approaches, such as luciferase reporter gene assay and electrophoretic mobility shift assay, have been applied to validate the function of individual noncoding variants. For example, we used luciferase reporter gene assay and electrophoretic mobility shift assay to examine the function of an opioid dependence-associated variant (or SNP rs569356) located in the promoter region of OPRD1 [13]. Moreover, the genome editing approach has been used to study the function of noncoding variants. For example, CRISPR/Cas9-mediated genome editing demonstrated that two noncoding SNPs (rs1345417 and rs12651896) affect the transcription of SOX2 and FOXD1 genes that participate in hair development [14]. However, functional study experiments are costly and time-consuming. It is unrealistic to use the above approaches to study the function of every disease-associated noncoding variant. A more efficient way is to prioritize disease-associated noncoding variants and then validate their regulatory functions by functional study approaches.

Functional genomics studies can pinpoint potentially regulatory variants from a large number of noncoding variants by investigating the relationship between genotypes and gene expression or epigenetic events (e.g., DNA methylation) at the genome-wide level. Genetic variants that influence gene expression are referred to as expression quantitative trait loci (eQTLs). Similarly, genetic variants that impact DNA methylation are called methylation quantitative trait loci (mQTLs). These genetic variants are further categorized as cis- or trans-acting, depending on the physical distance between variants and the transcript (or the CpG site) they regulate. Conventionally, variants that are within 1 Mb (megabase) from the transcription start site (or a CpG site) are considered to be cis-acting, while those that are over 1 Mb away from the transcription start site (or a CpG site) or on a different chromosome are considered to be trans-acting. There is an enrichment of cis- versus trans-eQTLs (or mQTLs) [15] because genetic variants are more likely to regulate gene transcription (or DNA methylation) locally. Additionally, the power to identify trans-eQTLs (or trans-mQTLs) is low, due to the need to control millions of potential false positives.

Mapping eQTLs and mQTLs using genome-wide genotype, transcriptome and DNA methylome data can provide an unbiased approach to identify novel functional variants. One critical issue of the eQTL/mQTL analysis is that the role of genetic variants in regulating gene expression (or DNA methylation) can be tissue- or cell-specific. For instance, eQTLs (or mQTLs) identified in the peripheral blood may have no regulatory function in the brain, given that the expression (or DNA methylation) of many genes is tissue- or cell-specific [16,17]. The Genotype-Tissue Expression (GTEx) database is a valuable resource for studying tissue-specific gene expression and regulation as well as mapping tissue-specific eQTLs [18]. It includes genome-wide genotype and gene expression data from 53 tissue types (including tissues from 13 different brain regions) across nearly 1000 subjects. The eQTLs identified by GTEx have been widely used to interpret GWAS-identified noncoding variants [19–21]. Nevertheless, these eQTLs may not be disease-specific, even though they may regulate the expression of genes involved in maintaining the function of specific tissues or cells. To unravel causal variants from disease-associated loci, it would be necessary to map eQTLs and mQTLs in both patients and controls. Those eQTLs and mQTLs appearing in patients are more likely to be causal variants for the disease. Identification of disease- and tissue-specific eQTLs and mQTLs will help disentangle the role of the specific tissue in disease development as well.

The prefrontal cortex (PFC) carries out a number of important executive functions such as working memory, attentional control and cognitive inhibition and flexibility. There is evidence that dysfunction of the PFC leads to impulsive behavior and lack of control over drinking, the two major characteristics of AUD [22]. Additionally, PFC-related impairments in cognitive control or working memory have been observed in patients with schizophrenia (SCZ) [23–25], bipolar disorder (BIP) [26–28], major depressive disorder (MDD) [29,30] and attention-deficit/hyperactivity disorder (ADHD) [27,31] as well as a number of other central nervous system disorders. Therefore, in the present study, we mapped eQTLs and mQTLs in the PFC of AUD patients and matched control subjects by integrative analyses of genome-wide SNP genotype, mRNA transcriptome and DNA methylome data. As AUD is comorbid with SCZ [32,33], BIP [34–37], MDD [38,39] and ADHD [40–43], we further examined the co-localization of PFC eQTLs/mQTLs and GWAS-identified SNPs (mainly noncoding SNPs) that were associated with AUD [4] as well as comorbid SCZ [44], BIP [45], MDD [46] and ADHD [47]. Our study demonstrated the enrichment of PFC eQTLs/mQTLs in the above psychiatric disorder-associated genetic variants identified by GWAS.

Materials & methods

Postmortem PFC tissues

Postmortem brain tissue samples (n = 48) were obtained from the New South Wales Tissue Resource Centre at the University of Sydney. The New South Wales Tissue Resource Centre is partially sponsored by the National Institute on Alcohol Abuse and Alcoholism for collecting human brain tissues for alcohol-related research. It has ethics approval from the Sydney Local Health Network and The University of Sydney. These brain tissue samples were dissected from Brodmann area 9 (BA9, mainly the dorsolateral PFC) of postmortem brains of 48 European Australians. Half of the subjects (16 males and 8 females) were diagnosed with alcohol abuse (n = 15) or dependence (n = 9), but they did not use illegal drugs of abuse or were not diagnosed with major psychotic disorders (such as SCZ and BIP) according to the criteria in the Diagnostic and Statistical Manual of Mental Disorders 4th Edition (DSM-IV) [48]. The other half of the subjects were sex- and age-matched healthy controls, who were free of alcohol or drug abuse or dependence or major psychotic disorders.

Genotyping

Genomic DNA was extracted from the above 48 postmortem PFC tissue samples using the QIAamp DNA Micro Kit (Qiagen, CA, USA), and was checked for quality by 0.7% agarose gel and quantified using a NanoDrop™ 8000 spectrophotometer (Thermo Fisher Scientific, MA, USA). These DNA samples (500 ng/sample) were genotyped using the Illumina HumanCoreExome-12 v1.0 BeadChip (Illumina, CA, USA) following the Illumina Infinium HD Array Super Manual Protocol. The BeadChip contains two sets of SNPs (538,448 SNPs in total): over 240,000 tagSNPs from the Illumina HumanCore BeadChip and over 240,000 exome-focused SNPs from the Illumina HumanExome BeadChip. The genotype data were processed using the Illumina GenomeStudio software V2011.1 (Genotyping Module V1.9.4) and verified by the zCall program [49]. SNPs were excluded from eQTL and mQTL analyses if they met any of the following criteria: GenCall score ≤0.15 at any sample; minor allele frequency <0.05; p-value of Hardy–Weinberg equilibrium test <0.001; missing genotyping rate >10%. After the above data quality control process, 254,374 SNPs remained for eQTL and mQTL analyses. The genotype data have been deposited in the National Center for Biotechnology Information (NCBI) dbGaP database (Accession Number: phs001981.v1.p1).

DNA methylome profiling

The above 48 genomic DNA samples (500 ng/sample) were treated with bisulfite reagents included in the EZ-96 DNA methylation kit (Zymo Research, CA, USA) according to the manufacturer’s protocol. The genome-wide DNA methylation (or DNA methylome) was profiled using the Illumina Infinium HumanMethylation450 BeadChip (Illumina), following the Illumina Infinium HD Methylation protocol [50]. This BeadChip interrogates 485,577 CpGs across the genome [50]. The raw DNA methylome data were processed using the Illumina GenomeStudio software V2011.1 (Methylation Module V1.9.0) to generate β values for CpG sites, with β values ranging from 0 (0% methylation) to 1.0 (100% methylation). After the data quality control process as described in our previous study [51], 430,407 CpGs remained for the mQTL analysis. The DNA methylome data have been deposited in the NCBI Gene Expression Omnibus (GEO) database (Accession Number: GSE49393).

mRNA expression microarray assays

Total RNAs were isolated from about 10 mg of the above 48 frozen postmortem PFC tissues using the miRNeasy Mini Kit (Qiagen). RNA concentration was measured using a NanoDrop™ 8000 spectrophotometer (Thermo Fisher Scientific) and the RNA Integrity Number (RIN) was measured on an Agilent 2100 Bioanalyzer (Agilent Technologies, CA, USA). PFC mRNA transcriptomes were profiled using the Illumina HumanHT-12 v4 Expression BeadChip (Illumina) (one female AUD and one female control subjects were not included in the assay). The Illumina protocol (Illumina) was described in our previous study [52]. This array provides coverage of the transcribed human genome using 47,231 probe sets and thus captures the expression of more than 31,000 annotated genes. Probe intensity and gene expression data were analyzed using Illumina GenomeStudio software V2011.1 (Gene Expression Module V1.9.0). After the data quality control process as previously described [52], there were 21,521 transcripts remained for the eQTL analysis. The transcriptome data have been deposited in the NCBI GEO database (Accession Number: GSE49376).

eQTL & mQTL analyses

The correlation of SNP genotypes with gene expression (or DNA methylation) levels was analyzed by QTLtools [53]. The genotype of a SNP was coded as 0, 1 or 2 under an additive model. Linear regression models were used to assess the effect of SNPs on expression levels of transcripts (or methylation levels of CpG sites), adjusted for sex, age, postmortem interval, RIN and AUD status. SNPs showing a significant association with gene expression were referred to as eQTLs, while SNPs showing a significant association with DNA methylation were called mQTLs. Due to the potential linkage disequilibrium between genetic variants, we used an empirical false discovery rate (FDR) [54,55] to account for multiple testing. At each permutation, we randomly assigned the sample labels of gene expression data (or DNA methylation data) and then tested their association with genetic variants. The empirical FDR was estimated to be the ratio of the average number of associations from the permuted samples to the actual number of associations in the real data. 1000 permutations were performed to obtain a stable FDR estimation [54,55]. The estimation of FDR was performed for eQTLs and mQTLs, separately. Significant eQTLs and mQTLs were defined as those with the FDR <5%. Given the small number of the brain tissue samples, trans-eQTL and trans-mQTL analyses were not performed. The function of genes with expression or DNA methylation potentially impacted by PFC eQTLs/mQTLs was annotated by Enrichr [56]. Additionally, the overlap of PFC eQTLs with eQTLs identified in GTEx tissues [18] was analyzed.

Analysis of the enrichment of PFC eQTLs/mQTLs in disease-associated variants

We examined the co-localization of PFC eQTLs/mQTLs with GWAS signals for AUD and related psychiatric disorders. GWAS summary statistics for AUD [4], SCZ [44], BIP [45], MDD [46] and ADHD [47] were downloaded from the Psychiatric Genomics Consortium database (www.med.unc.edu/pgc/results-and-downloads/). All these GWAS were performed among participants of European descent. We examined the association of PFC cis-eQTLs (empirical FDR <0.05) and cis-mQTLs (empirical FDR <0.05) with each disorder, and compared them with a randomly selected SNP set with the same number of SNPs by the Kolmogorov–Smirnov test. Significant enrichment was claimed if the two-sided p-value was less than 0.05. In addition, we analyzed the enrichment of PFC eQTLs/mQTLs in 15,831 unique SNPs (p < 5 × 10^-8) that were associated with 1,564 other complex disorders or traits that were available at the GWAS catalog (www.ebi.ac.uk/gwas/; accessed on 10 May 2018) by the Fisher’s exact test.

Results

PFC tissue sample descriptive statistics

As shown in Table 1, postmortem PFC tissues from 24 AUD patients (16 males and 8 females) and 24 control subjects (16 males and 8 females) were included in this study. There were no significant differences between AUD patients and control subjects in terms of age (55 ± 11 vs 56 ± 12 years), postmortem intervals (39.5 ± 14.7 vs 32.6 ± 13.1 h) and RIN (7.2 ± 1.2 vs 7.4 ± 0.8). The average alcohol consumption per day among AUD patients was 191.2 ± 152.0 g, which was significantly higher than that of control subjects (11.1 ± 8.9 g per day) (p = 2.8 × 10⁻⁹).

Table 1. . Characteristics of study samples.

Variable	Subjects with alcohol use disorder (n = 24)†	Controls (n = 24)	p-value‡
Males, n (%)	16 (66.7%)	16 (66.7%)	1.000
Age, years	56 ± 12	55 ± 11	0.606
PMI, h	33 ± 13	40 ± 15	0.114
Alcohol consumption per day, g	191 ± 152	11 ± 9	2.8 × 10−9
RIN	7.2 ± 1.2	7.4 ± 0.8	0.449

Data are means (±SD) or frequencies (percentages).

^†15 subjects with alcohol abuse and nine subjects with alcohol dependence.

^‡p-values were calculated by Wilcoxon rank-sum tests for continuous variables or Fisher’s exact tests for categorical variables.

PMI: Post-mortem interval; RIN: RNA integrity number; SD: Standard deviation.

Cis-eQTLs identified in the PFC

As summarized in Table 2, 923 SNPs (or cis-eQTLs) were significantly associated with expression levels of 325 genes in the combined dataset (cases + controls) (FDR <0.05). These genes are enriched in pathways including Glutathione Metabolism, Chemical Carcinogenesis, Metabolism of Xenobiotics by Cytochrome P450, Antigen Processing and Presentation, RNA Polymerase, DNA Replication, Fluid Shear Stress and Atherosclerosis, Homologous Recombination and Drug Metabolism. When the eQTL analysis was performed in cases only, 27 SNPs (or cis-eQTLs) were significantly associated with expression levels of 15 genes (FDR <0.05; enriched in pathways including Protein Export and Morphine Addiction). When the eQTL analysis was performed in controls only, 98 SNPs (or cis-eQTLs) were significantly associated with expression levels of 50 genes (FDR <0.05; enriched in pathways including Glutathione Metabolism, Metabolism of Xenobiotics by Cytochrome P450, Chemical Carcinogenesis, TGF-β Signaling Pathway, Drug Metabolism and Fluid Shear Stress and Atherosclerosis). In addition, 12 cis-eQTLs potentially influenced expression levels of six genes (ATP5G2, IPO8, PADI4, RIPK5, RPS26 and USP36; enriched in the Ribosome pathway) in both cases and controls.

Table 2. . Significant cis-expression quantitative trait loci and cis-methylation quantitative trait loci (FDR < 0.05) identified in 48 human postmortem prefrontal cortex tissues.

	cis-eQTLs	cis-eQTL-associated genes	cis-mQTLs	cis-mQTL-associated CpGs
Cases + Controls	923	325	9,832	5,723 (3,966 CpGs mapped to 2,538 genes)
Cases	27	15	264	192 (119 CpGs mapped to 106 genes)
Controls	98	50	695	483 (319 CpGs mapped to 255 genes)

Cases: Subjects with alcohol use disorder; Controls: Healthy control subjects.

eQTL: Expression quantitative trait locus; FDR: False discovery rate; mQTL: Methylation quantitative trait locus.

Cis-mQTLs identified in the PFC

As presented in Table 2, 9,382 SNPs (or cis-mQTLs) were significantly associated with methylation levels of 5,723 CpGs (including 3,966 CpGs mapped to 2,538 genes) in the combined dataset (cases + controls) (FDR <0.05). Among them, 295 were also cis-eQTLs, accounting for 32.0% of the above 923 cis-eQTLs identified in the PFC. When the mQTL analysis was performed in cases only, 264 SNPs (or cis-mQTLs) were significantly associated with methylation levels of 192 CpGs (including 119 CpGs mapped to 106 genes) (FDR <0.05). These 106 genes are enriched in pathways including Type I Diabetes Mellitus, Antigen Processing and Presentation, Human Papillomavirus Infection, Phagosome, Allograft Rejection, Graft-versus-host Disease, Viral Carcinogenesis, Human Cytomegalovirus Infection, Autoimmune Thyroid Disease, Pathogenic E. coli Infection, Viral Myocarditis, Endocytosis and Cell Adhesion Molecules. When the mQTL analysis was performed in controls only, 695 SNPs (or cis-mQTLs) were significantly associated with methylation levels of 483 CpGs (including 319 CpGs mapped to 255 genes) (FDR <0.05). These 255 genes are enriched in pathways including Type I Diabetes Mellitus, Allograft Rejection, Graft-versus-host Disease, Antigen Processing and Presentation, Autoimmune Thyroid Disease and Asthma Viral Myocarditis. In addition, 61 cis-mQTLs potentially influenced methylation levels of 66 CpGs (including 36 CpGs mapped to 30 genes) in both cases and controls. These 30 genes are enriched in pathways including Type I diabetes Mellitus, Antigen Processing and Presentation, Epstein-Barr Virus Infection, Viral Carcinogenesis, Human Immunodeficiency Virus 1 Infection, Human Cytomegalovirus Infection and Thermogenesis.

Overlap of PFC cis-eQTLs with eQTLs identified in GTEx tissues

Figure 1 shows the overlap of PFC cis-eQTLs obtained from the present study with those reported in GTEx, which includes eQTLs from 48 different tissues, each one with more than 70 samples. An average of 26.2% of PFC cis-eQTLs from our study appeared in at least one tissue in GTEx. Among all types of tissues, the tibial nerve shared the highest proportion (39.3%) of cis-eQTLs identified in the PFC. Altogether, 71.4% of PFC cis-eQTLs identified in the present study overlapped with eQTLs identified in at least one type of tissue in GTEx.

Figure 1. . Proportions of expression quantitative trait loci shared between the prefrontal cortex and 48 different tissues included in the Genotype-Tissue Expression database.

EBV: Epstein–Barr virus; eQTL: Expression quantitative trait locus; PFC: Prefrontal cortex.

Enrichment of PFC cis-eQTLs/cis-mQTLs in GWAS-identified SNPs associated with AUD & related psychiatric disorders as well as other diseases or traits

We examined the co-localization of PFC cis-eQTLs with GWAS signals for AUD and comorbid psychiatric disorders (SCZ, BIP, MDD and ADHD). The cis-eQTL enrichment analysis results are summarized in Table 3. PFC cis-eQTLs identified in all samples (cases and controls together) were significantly enriched in GWAS-identified SNPs that were associated with AUD (p = 0.011), SCZ (p = 2.2 × 10^-16), BIP (p = 2.0 × 10^-6), and MDD (p = 1.2 × 10^-13). PFC cis-eQTLs identified in cases only were significantly enriched in GWAS-identified SNPs that were associated with SCZ (p = 0.002). PFC cis-eQTLs identified in controls only were significantly enriched in GWAS-identified SNPs that were associated with AUD (p = 0.012), SCZ (p = 1.8 × 10⁻⁵), BIP (p = 3.5 × 10⁻⁵), and MDD (p = 0.012).

Table 3. . Enrichment of prefrontal cortex cis-expression quantitative trait loci in genome-wide association study-identified variants associated with psychiatric disorders.

Psychiatric	PFC cis-eQTLs (cases + controls)		PFC cis-eQTLs (cases only)		PFC cis-eQTLs (controls only)
disorders	Overlapped with GWAS SNPs	p-value	Overlapped with GWAS SNPs	p-value	Overlapped with GWAS SNPs	p-value
AUD	915	0.011	27	0.223	96	0.012
SCZ	849	2.2 × 10-16	27	0.002	97	1.8 × 10-5
BIP	849	2.0 × 10-6	27	0.178	97	3.5 × 10-5
MDD	836	1.2 × 10-13	27	0.298	97	0.012
ADHD	909	0.918	27	0.056	96	0.189

ADHD: Attention-deficit/hyperactivity disorder; AUD: Alcohol use disorder; BIP: Bipolar disorder; eQTL: Expression quantitative trait locus; GWAS: Genome-wide association study; MDD: Major depressive disorder; PFC: Prefrontal cortex; SCZ: Schizophrenia.

The cis-mQTL enrichment analysis results are summarized in Table 4. PFC cis-mQTLs identified in all samples (cases and controls together) were significantly enriched in GWAS-identified SNPs that were associated with AUD (p = 1.2 × 10^-10), SCZ (p = 2.2 × 10^-16), BIP (p = 2.2 × 10^-16), and MDD (p = 2.2 × 10^-16). PFC cis-mQTLs identified in cases only were significantly enriched in GWAS-identified SNPs that were associated with SCZ (p = 2.1 × 10^-5) and BIP (p = 8.7 × 10^-6). PFC cis-mQTLs identified in controls only were significantly enriched in GWAS-identified SNPs that were associated with AUD (p = 0.004), SCZ (p = 1.6 × 10^-13), BIP (p = 0.008), MDD (p = 0.006) and ADHD (p = 0.028).

Table 4. . Enrichment of prefrontal cortex cis-methylation quantitative trait loci in genome-wide association study-identified variants associated with psychiatric disorders.

Psychiatric	PFC cis-mQTLs (cases + controls)		PFC cis-mQTLs (cases)		PFC cis-mQTLs (controls)
disorders	Overlapped with GWAS SNPs	p-value	Overlapped with GWAS SNPs	p-value	Overlapped with GWAS SNPs	p-value
AUD	9277	1.2 × 10-10	262	0.093	685	0.004
SCZ	8345	2.2 × 10-16	244	2.1 × 10-5	606	1.6 × 10-13
BIP	8340	2.2 × 10-16	244	8.7 × 10-6	606	0.008
MDD	8132	2.2 × 10-16	240	0.123	580	0.006
ADHD	9045	0.006	257	0.556	668	0.028

AUD: Alcohol use disorder; ADHD: Attention-deficit/hyperactivity disorder; BIP: Bipolar disorder; GWAS: Genome-wide association study; MDD: Major depressive disorder; mQTL: Methylation quantitative trait locus; PFC: Prefrontal cortex; SCZ: Schizophrenia.

We also analyzed the overlap of PFC eQTLs and mQTLs with 15,831 unique SNPs (p < 5 × 10^-8) associated with 1,564 other complex diseases or traits that were available at the GWAS catalog (accessed on 10 May 2018). Among these 15,831 unique SNPs, 39 were also eQTLs in PFC, representing a 14.7-fold enrichment (p < 2.2 × 10^-16). Similarly, 379-disease or trait-associated SNPs were also mQTLs in PFC, representing 14.1-fold enrichment (p < 2.2 × 10^-16). Thus, some PFC eQTLs and mQTLs were associated with other complex diseases or traits as well.

Discussion

GWAS research has identified thousands of genetic variants that were associated with hundreds of complex traits or common diseases [57]. A major focus in the post-GWAS era is to unravel causal variants and their biological functions. Genetic variants associated with AUD and related psychiatric disorders may impact gene expression in specific brain regions such as the PFC. We utilized genome-wide genotype, gene expression and DNA methylation data from human postmortem PFC tissue samples to map cis-eQTLs and cis-mQTLs. We then analyzed the co-localization of PFC cis-eQTLs/cis-mQTLs with GWAS loci for AUD and related psychiatric disorders as well as other diseases or traits. There were three major findings from the present study.

First, we uncovered cis-eQTLs and cis-mQTLs in the PFC, an important brain region mainly implicated in executive functions (working memory, decision-making and cognitive behaviors) (Table 2). The identification of cis-eQTLs/cis-mQTLs in the PFC suggests that these genetic variants may modulate the expression/DNA methylation levels of genes in the PFC, thus altering an individual’s vulnerability to PFC-related disorders or leading to PFC functional variation among individuals. Previous studies have demonstrated that variations in genes, such as NPY [58], VMAT1 [59], CRHR1 [60] and DRD2 [61], influenced PFC functions in humans. Some of these genes also harbor PFC cis-eQTLs/cis-mQTLs. For example, SNP rs8078967, a PFC cis-mQTL (FDR = 0.018), was mapped to CRHR1, a critical gene involved in the hypothalamic–pituitary–adrenal axis. To distinguish cis-eQTLs/cis-mQTLs in the PFC of AUD patients from those in the PFC of control subjects, we performed cis-eQTL/cis-mQTL analyses in cases and controls separately. Of interest, genes mapped by cis-eQTLs identified in the PFC of AUD patients participate in pathways such as Protein Export and Morphine Addiction.

Second, we found that PFC cis-eQTLs/cis-mQTLs co-localized with GWAS loci for AUD and comorbid psychiatric disorders (Tables 3 and 4). Mounting evidence suggests that PFC-dependent cognitive dysfunction (or compromised executive functions) can increase the risk of psychiatric disorders including AUD [62], SCZ [63], BIP [64], MDD [65] and ADHD [66]. The enrichment of PFC cis-eQTLs/cis-mQTLs in GWAS-identified SNPs associated with these disorders implies that some PFC cis-eQTLs/cis-mQTLs might be causal variants for them. PFC cis-eQTLs/cis-mQTLs may alter the expression of cognition-related genes in the PFC, leading to an increased risk of these disorders. Additionally, the PFC makes up over 10% of the volume of the brain, and there is a tight interconnection between the PFC and other brain regions [67]. Thus, the PFC participates in numerous other brain functions (e.g., memory, movement, sensory perception, intelligence and emotional responses) besides executive functions. Presumably, PFC cis-QTLs/cis-mQTLs may confer susceptibility to many other complex diseases or traits by influencing the function of the PFC. Our study did show a significant enrichment of PFC cis-eQTLs/cis-mQTLs in GWAS-identified SNPs associated with many other diseases or traits.

Third, we noted that PFC cis-eQTLs/cis-mQTLs might exert pleiotropic effects on multiple diseases or traits (Tables 3 and 4). That is to say, a PFC cis-eQTL or cis-mQTL may contribute to the risk of AUD and related disorders as well as other diseases or traits. The pleiotropy is caused by genetic variants that are associated with diseases or traits with shared biological pathways. For example, PFC cognitive dysfunction is the common base of AUD and related disorders (SCZ, BIP, MDD and ADHD). Polygenic risk score analyses have demonstrated significant genetic correlations between AUD and SCZ [68,69], SCZ and BIP [70] and AUD and MDD [71]. The present study further supported the genetic overlap between AUD and related psychiatric disorders. This is consistent with the findings that about 37% AUD subjects had comorbid psychiatric disorders including SCZ, BIP, MDD and ADHD [72]. In addition, there is also evidence that a single genetic variant may have pleiotropic effects on multiple diseases or traits via independent biological pathways [73]. PFC cis-eQTLs/cis-mQTLs may modulate the expression or function of PFC proteins that interact with a number of proteins involved in other biological pathways or expressed in other tissues or organs. In other words, a PFC cis-eQTL or cis-mQTL carried by a subject may influence the expression of genes in other tissues or organs, leading to an increased risk of other diseases or traits. The present study demonstrated the enrichment of PFC cis-eQTLs/cis-mQTLs in GWAS-identified SNPs that were associated with PFC-related as well as PFC-unrelated disorders or traits.

The most obvious limitation in this research was that the sample size was small, thus trans-eQTLs/trans-mQTLs and those cis-eQTLs/cis-mQTLs with a minor effect were not uncovered in the PFC. Nevertheless, generating genome-wide genotype, gene expression and DNA methylation data in a large sample is costly. The smaller sample size and the exclusion of less common SNPs in our eQTL analysis may lead to a relatively low overlap of PFC eQTLs from the present study with those from the GTEx study (48 for the present study vs 129 for the GTEx study) (Figure 1). The low overlap of PFC eQTLs may also result from the differences in sample constitutions (one half of the subjects included in the preset study were AUD patients, while all samples for the GTEx study were collected from nondisease tissue sites) and transcriptome profiling platforms (the present study used the microarray approach while the GTEx study used the RNA-seq technology). Second, the Illumina Infinium Human Methylation 450 K BeadChip used for this study can only quantify methylation levels of around 450 K methylated CpG sites, and it cannot detect the hydroxymethylation of CpGs (or 5 hmC). In our future studies, we could use the Illumina Infinium MethylationEPIC BeadChip, which targets >850 K methylated CpGs and is also useful for 5 hmC detection. We can also use the reduced representation bisulfite sequencing method to analyze genome-wide DNA methylation at the single nucleotide level [74]. Third, the analysis of eQTLs and mQTLs was conducted only in the PFC. Since a number of other brain regions are also involved in the development of AUD and its comorbid disorders, future eQTL/mQTL analyses should be performed in other brain regions as well. Finally, the present study was conducted only in participants of Caucasian ancestry. It is unknown if the eQTLs and mQTLs identified in our study are applicable to other populations. In addition, due to the limited sample size and insufficient genome coverage of the exome array used in the current study, we did not consider the potential bias caused by population stratification. Follow-up studies should use postmortem tissues from multiple brain regions of a larger number of subjects from different ethnic groups.

Conclusion

In conclusion, we performed a post-GWAS study to interpret the function of GWAS-identified SNPs (particularly those noncoding variants) associated with AUD and comorbid disorders as well as other diseases or traits. We observed a co-localization of PFC cis-eQTLs/cis-mQTLs in GWAS loci for the above diseases or traits. Our study provides further evidence that the PFC plays an important role in many diseases or traits. Future studies should consider increasing the sample size for PFC eQTL/mQTL analysis. In addition, we should also identify eQTLs/mQTLs in other brain regions that are involved in AUD and comorbid disorders.

Future perspective

Since eQTLs/mQTLs are likely the molecular link between genetic variation and phenotypic differences, the integration of GWAS findings with brain eQTLs/mQTLs is a promising approach for identifying genes that are causally related to neuropsychiatric disorders. Recent large-scale meta-analyses of GWAS have reliably identified genetic variants that are associated with neuropsychiatric disorders, such as AUD, SCZ, BIP, MDD, Alzheimer’s disease and ADHD. However, it is necessary to develop reference datasets for eQTLs/mQTLs to determine whether a gene implicated in disease risk by GWAS is directly related to disease pathogenesis. To achieve this goal, we should identify eQTLs/mQTLs in multiple brain regions in tissues obtained from a larger sample of patients and healthy subjects representing diverse ethnic populations. Such a resource would be essential to the future discovery of novel biomarkers and therapeutic approaches for neuropsychiatric disorders.

Summary points.

• It is challenging to study the function of genome-wide association study (GWAS)-identified genetic variants (particularly noncoding variants) associated with diseases or traits. Causal variants for diseases or traits can be expression quantitative trait loci (eQTLs) and/or methylation quantitative trait loci (mQTLs) in specific tissues or organs.

• The present study aimed to identify eQTLs and mQTLs in the prefrontal cortex (PFC) for interpreting the function of GWAS-identified single nucleotide polymorphisms associated with alcohol use disorder (AUD) and comorbid disorders as well as other diseases or traits.

• The present study demonstrated that PFC cis-eQTLs/cis-mQTLs co-localized with GWAS signals for AUD and comorbid disorders as well as other diseases or traits.

• These findings suggest that PFC cis-eQTLs/cis-mQTLs may regulate the expression of genes in the PFC, leading to an increased risk of AUD and comorbid disorders as well as other diseases or traits.