Association of Protein Phosphatase PPM1G With Alcohol Use Disorder and Brain Activity During Behavioral Control in a Genome-Wide Methylation Analysis

Twin pairs were part of the FinnTwin16 study, a population-based Finnish cohort of monozygotic and dizygotic twins. They were studied at age 16, with subsequent surveys at ages 17, 18.5, and 24). Among 104 monozygotic twin pairs, 36 were discordant for alcohol dependence or alcohol abuse. From this group, we randomly selected 18 pairs for genome-wide methylation analysis (Table 1; see also Table S1 in the data supplement that accompanies the online edition of this article). Discordant twins were selected using the Rutgers Alcohol Problem Index (23), administered at ages 18.5 and 24, with differences in DSM-III-R symptoms used as additional criteria (assessed using the Semi-Structured Assessment for the Genetics of Alcoholism). We focused on the Rutgers Alcohol Problem Index because in this age group, with a recent and short history of problems, DSM criteria are not sufficiently informative. Scores at ages 18 and 24 were significantly different between individuals with and without alcohol use disorders among the 18 twin pairs (t=−2.552, df=16, p=0.011). At interview at age 24, blood was sampled from both twins, and DNA was extracted using the Gentra Autopure LS System (Gentra Systems, Minneapolis). Monozygosity of twin pairs was confirmed by the Paternity Testing Unit at the Finnish National Public Health Institute. The study protocol was approved by local ethics committees at Helsinki and Indiana Universities.

Enrichment of unmethylated fraction of genomic DNA (gDNA) was performed by methylation-sensitive restriction of gDNA, amplification, and labeling prior to array hybridization, as previously described (24). (For more details on this and other aspects of the study’s methodology, see the data supplement). Cy3-labeled DNA from each twin was hybridized with Cy5-labeled control DNA to a NimbleGen DNA Methylation 385k Array (Roche NimbleGen, Madison, Wisc.). The array consists of 385,000 60-mer oligonucleotide probes covering promoter regions and cytosine-guanine dinucleotide (CpG) islands. Microarray data were normalized using GenePix Pro 6.0 (Molecular Devices, Sunnyvale, Calif.).

Analysis of raw intensity data was performed using Bioconductor, an open-source software suite in the R statistical software environment (25). Raw intensity data were normalized using quantile normalization implemented in the Ringo package (26). Normalized intensities were log₂ transformed, and the ratio between each twin pair was used for analysis. A one-class RankProd analysis was used to identify differentially methylated probes between discordant twins (27). RankProd is based on nonparametric statistics derived from the rank product of each gene. As this procedure is similar to false-discovery-rate estimation, we used percentage of false positives <0.05 for selecting differentially methylated probes.

Top-ranked alcohol use disorder-associated differentially methylated regions were verified in 18 discordant twin pairs (Table 1) using bisulfite polymerase chain reaction (PCR) and the Sequenom EpiTYPER system (Sequenom, Hamburg). Methylation sites contained in the following genes were assessed: PPMG1, INS-IGF2, FMN1, SEPHS2, SLC6A3, AIM1, OPRL1, and PIPOX. Since we enriched samples for unmethylated fraction of the gDNA using methylation-sensitive restriction enzymes, assays were designed to cover areas containing HpaII and Hin6I cut-sites, adjacent to the 60-mer probes in the NimbleGen DNA Methylation. DNA was bisulfite treated using the EZ-96 DNA methylation kit (Zymo Research, Irvine, Calif.). Primer sequences, PCR conditions, amplicon length, and number of CpG sites are listed in Table S2 in the online data supplement. To detect differentially methylated regions, we performed paired t tests for each single CpG site detectable within each amplicon.

The methylation profile of the 3′-protein-phosphatase-1G (PPM1G) gene locus was investigated in 499 14-year-olds (Table 1) from IMAGEN, a European multicenter imaging-genetic/epigenetic study. Participants were tested in eight assessment centers in four European countries. The Psytools software package (Delosis, London) was used for behavioral characterization. The assessment battery of questionnaires and cognitive tasks was self-administered both online in participants’ homes and at the neuroimaging facilities. The study was approved by local ethics research committees at each site. Written informed consent was obtained from all participants and from their legal guardians. A detailed description of recruitment and assessment procedures as well as inclusion and exclusion criteria has been published previously (22).

Escalation of daily drinking between ages 14 and 16 was measured by calculating the difference in daily drinking between the two ages using the European School Survey Project on Alcohol and Other Drugs. Externalizing behavioral risk factors for substance abuse or dependence were measured with the Substance Use Risk Profile Scale (28), which assesses independent risks factors, including two externalizing factors, impulsivity and sensation seeking (both the sum of five items). Pubertal development was determined using the Pubertal Development Scale (29), and a development score was obtained by calculating the mean of the scores on the five indicators. Data quality was controlled by context checks administered at the start of each task.

DNA was extracted from whole blood samples (∼10 mL) collected at age 14 using the Gentra Puregene Blood Kit (QIAGEN, Valencia, Calif.). DNA was bisulfite treated using the EZ-96 DNA methylation kit, and DNA methylation of PPM1G was assessed using bisulfite PCR and the Sequenom EpiTYPER system.

Data on an event-related stop signal task from 393 14-year-old IMAGEN participants for whom complete data were available were analyzed using the contrast “stop success versus go success” to study neural responses to successful and unsuccessful inhibitory control (30).

Structural and functional MRI (fMRI) data were acquired with 3-T scanners at eight sites using compatible scanning variables and the same scanning protocol at all sites. Full details of MRI acquisition protocols and quality checks have been described previously, including standardization across MRI scanners (22). Effect of MRI site was controlled for as a nuisance covariate in all statistical analyses. Effect of MRI site and scanner on the magnitude of the fMRI response has been formally evaluated previously (31). All acquired images were preprocessed and analyzed using SPM8 (http://www.fil.ion.ucl.ac.uk/spm) at NeuroSpin (Paris). We extracted regions of interest using the MarsBaR toolbox (http://marsbar.sourceforge.net). The right inferior frontal gyrus was defined according to a human automated anatomical labeling atlas (32), and the right subthalamic nucleus was defined by the Lucerna et al. atlas (33). All regions of interest were extracted to match the stop success versus go success contrast t maps. The center of the resulting right inferior frontal gyrus region of interest was at Montreal Neurological Institute coordinates x=49, y=22, z=16, with a volume of 25,504 mm³. The center of the right subthalamic nucleus region of interest was located at coordinates x=10, y=−15, z=−5, with a volume of 1,000 mm³. Averaged beta values based on all voxels across time series in the regions of interest were used for all analyses.

Total RNA was extracted from whole blood cells collected at age 14 using the PAXgene Blood RNA Kit (QIAGEN). Gene expression profiling was performed by hybridizing labeled cRNA on Illumina HumanHT-12 v4 Expression BeadChips (Illumina, San Diego). Normalized expression data of the probe mapping to PPM1G were extracted, log-transformed, screened for outliers, and tested for association with PPM1G methylation data.

DNA was purified from whole blood samples (∼10 mL) preserved in EDTA tubes using the Gentra Puregene Blood Kit. Genotype information was collected at 582,982 markers using the Illumina HumanHap610 Genotyping BeadChip. Genotype data from four single-nucleotide polymorphisms (SNPs), present within 50 kb of the PPM1G CpG site analyzed (rs7602534, rs11675428, rs704791, and rs1260342), were tested for association with PPM1G methylation. Data for rs2384629, previously associated with alcohol use disorders (34), were imputed, and haplotypes covering the gene were calculated using Plink (http://pngu.mgh.harvard.edu/∼purcell/plink/).

The general linear model was used to determine associations among methylation of a CpG site present in the 3′ untranslated region (3′-UTR) of PPM1G, found differentially methylated in the twins discordant for alcohol use disorders, as well as PPM1G gene expression and genotype, quantity and frequency of drinking, escalation of daily drinking, impulsiveness, and stop signal task blood-oxygen-level-dependent (BOLD) responses. Gender, puberty, and study site were controlled for in all analyses. Handedness was controlled for in association analyses of BOLD activation. Corrected p values (p_corrected) were computed using 10,000 permutations. Partial eta-squared (η²_p), a measure of effect size calculated after removing the contribution from other control variables, is reported for all associations.

Genome-wide DNA methylation was first assessed by microarray analysis in 18 monozygotic twin pairs discordant for alcohol use disorder (Table 1) of 385,000 60-mer oligonucleotide probes covering both promoter regions and CpG islands. Using a nonparametric method and a cutoff of the estimated percentage of false positive predictions <0.05 and a p value <1×10⁻⁵, we identified 77 differentially methylated regions that exhibit an alcohol use disorder-associated differential methylation (see Table S3 in the online data supplement).

The majority of the 77 differentially methylated regions (68%) were hypermethylated in the affected twin, consistent with previous findings showing an association between DNA hypermethylation and alcohol use disorder (15). The identified regions were associated with 62 genes, with 57% located in the gene body, 30% in promoters, and the rest intergenic (see Figure S1A in the data supplement). The majority of differentially methylated regions (65%) were located outside of CpG-rich regions (CpG islands), with 41% of the sites located in shores flanking the islands, which have been implicated in gene regulation (35) (see Figure S1B in the data supplement).

The most significant differentially methylated regions located within genes (body and promoter) were divided into hypermethylated and hypomethylated regions; the top 10 differentially methylated regions of each category are listed in Table 2. Among these regions, we selected those with an absolute delta (Δ) DNA methylation intensity value (the mean difference in DNA methylation intensity between the affected and the unaffected member in each twin pair) higher than 0.5 for technical replication using mass spectrometry. These included differentially methylated regions in the PPMG1, INS-IGF2, FMN1, SEPHS2, SLC6A3, AIM1, OPRL1, and PIPOX genes, which we analyzed in the 18 monozygotic twin pairs with a Sequenom EpiTYPER (see Table S2 in the data supplement). In the twins with an alcohol use disorder, we found significant increases in percent methylation of the CpG site in the 3′-UTR of PPM1G (p=0.005, mean Δ=2.4%) and in a CpG site contained in the OPRL1 probe (p=0.017, mean Δ=3.6%), confirming the findings from the microarray. We were unable to confirm differential DNA methylation in the other regions, possibly because the Sequenom assay could not interrogate the specific methylation-sensitive restriction enzyme cut-site driving the difference reported in the microarray.

As the PPM1G differentially methylated region showed the most significant association with alcohol use disorders, both in genome-wide analysis and in technical replication, we proceeded with a more detailed analysis and validation of PPM1G in an independent sample of 499 adolescents from the IMAGEN imaging-genetics study (Table 1) for whom gene expression and DNA methylation data were available. A linear regression analysis of PPM1G methylation and PPM1G gene expression revealed significant association (t=−1.93, p=0.027 [one-tailed]; η²_p=0.007), with higher methylation levels associated with lower mRNA levels (see Figure S2 in the data supplement).

To rule out possible genotype effects on methylation, we performed a linear regression of PPM1G methylation and five SNPs, including rs2384629, which was previously reported to be associated with alcohol dependence (37). The other SNPs covering the PPM1G locus were rs7602534, rs11675428, rs704791, and rs1260342. Neither a single SNP nor a haplotype analysis (±50 kb of the PPM1G CpG site analyzed) yielded significant association with methylation of PPM1G (see Table S4 in the data supplement).

To investigate whether the differentially methylated CpG site in PPM1G was associated with degree of alcohol exposure, we explored in the IMAGEN sample the association of PPM1G methylation and drinking measures at age 14. We found no association of PPM1G methylation and quantity (t=−0.45, p=0.65; η²_p =0.0004) or frequency of drinking during the 12 months prior to blood acquisition (t=−0.15, p=0.88; η²_p =0.00005) or during the month prior to blood acquisition (t=–0.86, p=0.39; η²_p=0.0017) (see Table S5 in the data supplement). This suggests that PPM1G hypermethylation is not caused by alcohol exposure at age 14. As escalation of alcohol use in adolescents is associated with increased risk for alcohol-related problems and alcohol use disorders (28), we explored a possible association of PPM1G methylation with escalation in daily use of alcohol between ages 14 and 16. We selected among the 499 adolescents of the IMAGEN study those who reported drinking at ages 14 or 16 (N=352) and found that PPM1G hypermethylation was positively associated with escalation of daily drinking (t=2.17, p=0.015 [one-tailed], p_corrected=0.024; η²_p =0.014) (Figure 1A). Next, we investigated a possible association of escalation of alcohol use with externalizing behavioral risk factors for alcohol use disorders (36). Escalation of daily drinking was associated with impulsiveness (t=2.33, p=0.020, p_corrected=0.04; η²_p=0.017) (Figure 1B) but not with sensation seeking. Impulsiveness, in turn, was associated with PPM1G methylation (t=2.56, p=0.012, p_corrected=0.032; η²_p =0.016; N=499) (Figure 1C), indicating a role of PPM1G methylation in both risky drinking and behavioral risk factors for alcohol use disorders in adolescents. When we controlled for impulsiveness, we lost the significance of the association of PPM1G methylation and escalation of daily drinking (p=0.065) and observed a decrease in effect size by 50% (with the η²_p value decreasing from 0.014 to 0.007). There was no association of the previously reported PPM1G SNP rs2384629 (34) with escalation of drinking or with trait impulsiveness (see Table S4 in the data supplement).

We examined a possible association between PPM1G methylation and brain activity during functional neuroimaging registration of the stop signal task, which assesses behavioral (motor) inhibition and is associated with alcohol use disorders (37, 38). In humans, voluntary inhibition of manual movements relies on a right-lateralized frontal–basal ganglia–thalamic pathway (39). This network includes the inferior frontal gyrus, which revokes planned movements through the subthalamic nucleus (40, 41) of the right hemisphere. We selected these regions of interest in the stop success versus go success contrast. In 14-year-olds (N=393) we found a positive association between PPM1G methylation and right subthalamic nucleus BOLD response (t=2.25, p=0.021, p_corrected=0.038; η²_p=0.013) (Figure 2A,B) but not with right inferior frontal gyrus BOLD response. There was no association of the previously reported PPM1G SNP rs2384629 (34) with activation of the right subthalamic nucleus (see Table S4 in the data supplement).