DRD2 methylation is associated with executive control network connectivity and severity of alcohol problems among a sample of polysubstance users

Abstract

Chronic exposure to alcohol and other drugs of abuse has been associated with deleterious consequences, including functional connectivity deficits within neural networks associated with executive control. Altered functional connectivity within the executive control network (ECN) might underlie the progressive inability to control consumption of alcohol and other drugs as substance use disorders progress. Genetic and epigenetic factors have been associated with substance use disorders (SUDs). For example, dopamine receptor 2 (DRD2) functioning has been associated with alcohol use disorder (AUD) and related phenotypes, including correlates of executive functioning. The present study aims to explore the relationship between a continuous measure of alcohol-related problems, epigenetic markers (methylation) within the DRD2 gene, and functional connectivity within the ECN among a sample of polysubstance users. A community sample of 658 subjects, whose consumption of alcohol, nicotine, and cannabis span across a spectrum of quantity and frequency of use, were obtained across previous studies in polysubstance using populations. Resting state functional magnetic resonance imaging was analyzed to identify intrinsic connectivity networks using a priori regions of interest. Methylation measurement of functionally relevant sites within the DRD2 gene was achieved via pyrosequencing. Regression-based models, including mediation and moderation models, tested the association between DRD2 methylation, functional connectivity within intrinsic neural networks (including the ECN), and severity of alcohol problems. Results suggest that average DRD2 methylation was negatively associated with right ECN (RECN) and left ECN (LECN) connectivity, but not associated with other networks tested, and DRD2 methylation was significantly associated with alcohol problems severity. Mediation models were not supported, although moderation models suggested that connectivity between edges within the RECN moderated the relationship between DRD2 methylation and AUD severity. Results support a theoretical model in which epigenetic factors are associated with neurobiological correlates of alcohol consumption among a sample of polysubstance users.

1 |. INTRODUCTION

An extensive body of literature suggests that chronic exposure to alcohol and other drugs of abuse is associated with deleterious consequences, both in terms of brain structure and function (see Oscar-Berman and Marinković¹). Researchers have developed a model of alcohol use disorder (AUD) that emphasizes neuroadaptations within the reward and control neural networks of the brain, and a resulting imbalance between these networks.^2–4 Such models seek to explain the behavioral manifestations of the disease (ie, compulsive alcohol use). Epigenetic modifications are one set of mechanisms that might underlie the neural changes associated with chronic alcohol use.

The extant literature implicates dopamine receptor levels and binding in AUD and related phenotypes. Basic animal and human literature has tested the association between dopamine receptor 2 (D2) functioning and cognitive processes related to executive functioning and addictive behaviors using experimental and task-based paradigms. Executive functioning is a psychological construct that has been defined by experts as “the ability to maintain an appropriate problem solving set for attainment of a future goal.”⁵ As outlined in Talpos and Shoaib,⁶ executive functioning is operationalized by a variety of task-based measures that fall into three broad categories, namely, rule generation and selection (eg, attentional set shifting), adapting behavioral responses to incoming stimuli (eg, stop signal, go/no-go), and overcoming a habit (eg, reversal learning). Results suggest that drug-addicted rats with ventral striatal D2 dysfunction fail to learn to inhibit responding to drug-related cues even when they are followed by punishment.⁷ In the human literature, D2 dysfunction has also been associated with decrements in response inhibition to drug cues,⁸ and degree of dopamine signaling has been associated with sensitivity of prefrontal activation during risk taking.⁹ Results from a recent positron emission tomography (PET) imaging study, which allowed researchers to get an approximate measure of D2 binding via the use of radioactive ligands, suggests that D2 binding was negatively associated with modulation of prefrontal cortical activation during risk taking.¹⁰

Functional neuroimaging paradigms are able to quantify differences in the connections within neural networks based on temporal synchronicity.¹¹ Such connectivity analyses have allowed researchers to investigate neural networks underlying executive control processes within the context of substance use disorders. Research in this area has shown that chronic heavy alcohol use is associated with altered strength of connectivity, or temporal correlation of activity, between brain regions, namely, the frontal and limbic areas. For example, recently abstinent alcoholic patients compared with healthy controls showed reduced connectivity in the fronto-cerebellar network taken from a motor task,¹² and reduced fronto-striatal connectivity during a response inhibition task paradigm.¹³ Reduced connectivity among regions associated with executive control (eg, reduced prefrontal cortical connectivity) has been associated with bias towards reward-seeking behavior in drug users.⁹ It follows that chronic heavy drinking has been shown to confer risk for disrupted cognitive processes, namely, executive functioning.^14–17 Throughout the paper, we refer to the executive control network (ECN). This network includes dorsolateral frontal and parietal regions. As described in Seeley and colleagues,¹⁸ the ECN is thought to comprise the dorsolateral prefrontal cortex (DLPFC), ventrolateral PFC, dorsomedial PFC, and lateral parietal cortices.¹⁸

Furthermore, some results suggest that pattern and strength of neural connectivity is associated with AUD status and severity. For example, long-term abstinent alcoholic subjects showed a pattern of connectivity whereby there was reduced connectivity between appetitive drive and decision-making regions, and increased connectivity within inhibitory control networks.¹⁹ Across multiple networks, functional connectivity was lower in currently relapsed alcoholics compared with controls.^20,21 Additionally, individuals with problematic alcohol use showed reduced connectivity strength across several networks, including the left executive control network (LECN), in particular between the DLPFC and parietal temporal regions. This reduced connectivity was inversely correlated with measures of AUD severity and failure to control drinking.²² Collectively, these results suggest that patterns and strength of neural connectivity covary with measures of AUD status and severity.

Taken together, literature suggests that D2 receptor function, ECN connectivity, and executive functioning are associated with AUD and related phenotypes. Despite these advancements, little is known about the mechanisms underlying D2 receptor functioning and correlates of executive functioning. The dopamine receptor 2 gene (DRD2) has been shown to regulate D2 receptor functioning (eg, Hirvonen et al²³; thus, mechanisms that alter DRD2 gene expression could possibly underlie D2 functioning and executive functioning. Epigenetic modifications are one set of processes that might underlie D2 receptor functioning and neural decrements associated with chronic alcohol use. Epigenetics refers to chemical processes that modulate the surrounding structures of DNA as opposed to changing the sequence itself.²⁴ Methylation, one type of epigenetic modification, involves the addition of a methyl group to the fifth carbon of cytosine residues.²⁵ Epigenetic modifications, including methylation, have been associated with alterations in gene expression.²⁶ Although DRD2 methylation has yet to be investigated extensively in the context of AUD phenotypes, Hillemacher and colleagues²⁷ demonstrated that lower D2 methylation was associated with abstinence among individuals with a history of pathologic gambling (Hillemacher et al, 2015),²⁷ which shares common neural mechanisms with substance use disorders.²⁸ Furthermore, among a sample of alcohol and nicotine dependent patients, methylation of the DRD2 gene was associated with alcohol and nicotine codependence, along with methylation of additional addiction-related genes.²⁹ Finally, Bidwell et al³⁰ found an association between DRD2 methylation and neural correlates of reward among problem alcohol users.

The current study aimed to add to the literature by further characterizing the mechanistic relationships between AUD, DRD2 methylation, and ECN connectivity among a polysubstance using population. Towards this end, the current study leveraged DRD2 methylation and ECN functional connectivity data in a large sample of polysubstance users. Specifically, we were interested in whether DRD2 methylation is inversely related to connectivity in the ECN and whether DRD2 methylation is associated with alcohol problem severity. The current paper is a secondary set of analyses of previously reported results. A subset of the connectivity data included in the present set of analyses were included in previously published works from our group.^22,31 A subset of the DRD2 methylation data included in the present paper were included in the Bidwell and colleagues’³⁰ paper, which investigated functional neural correlates of reward. Importantly, the aims of the current study (ie, testing the relationship between DRD2 methylation and ECN connectivity) are unique, as no functional connectivity analyses using these data have investigated the association between functional connectivity and epigenetic markers (ie, methylation) in any previously published works.

2 |. MATERIALS AND METHODS

2.1 |. Study samples

Resting state connectivity and epigenomic data were obtained across previous studies in polysubstance using populations.^32–34 The current analyses included data from three studies. The main differences across the three studies include the following: One study enrolled regular binge drinkers, another study enrolled treatment seekers who met criteria for AUD, and the third study enrolled treatment seeking cigarette smokers. See Table 1 for descriptive statistics on substance use by source study. Two of the three studies were clinical trials involving pharmacotherapies (olanzapine and varenicline); however, the imaging data included in the present set of analyses were collected during baseline magnetic resonance imaging (MRI) scans. Therefore, the imaging data were collected prior to the initiation of the medication regimens. Briefly, participants were recruited from the Albuquerque or the Boulder/Denver metropolitan regions through advertisements in local print, online media, and radio advertisements. Exclusionary criteria across studies included previous brain injury or loss of consciousness for more than 5 minutes, history of bipolar disorder or a psychotic disorder, or a positive pregnancy test. Additionally, across the three studies, subjects were excluded if they tested positive for illicit drugs of abuse (other than marijuana). Subjects abstained from smoking for 2 hours before scanning, from alcohol consumption for 24 hours (per patient report), and were required to pass a breathalyzer test immediately prior to participation. The resting state neuroimaging and methylation analysis procedures were identical across all studies that were included in these analyses. Written informed consent, approved by the University of New Mexico Human Research Committee, was obtained from all participants. Data collection for all studies included in this manuscript has concluded. While data from these studies have been previously published (eg, Bidwell et al³⁰ and Claus et al³²), the aims of the current investigation are distinct from aims of any of the original studies and resulting manuscripts. A total of 658 participants were included on the basis of having complete epigenetic, and behavioral data and a subsample of 621 subjects were included in all neuroimaging analyses on the basis of having complete neuroimaging, epigenetic, and behavioral data after excluding subjects with excessive motion during the MRI scan (see below). The average age of the sample was 31.4 (SD = 9.51), and the gender breakdown was as follows: 242 females and 416 males. On average, study samples had an AUDIT score of 12.11 (SD = 8.12) at the time of the study appointment. On average, subjects within the sample reported 1.57 drinks per drinking day (SD = 6.09), 10.46 cigarettes per smoking day (SD = 8.39). According to the 30-day Timeline Follow Back,³⁵ subjects used alcohol 12.131 days/month on average (SD = 8.947), used cannabis 4.562 days/month on average (SD = 8.872), and smoked nicotine cigarettes 17.79 days/month on average (SD = 14.21).

TABLE 1.

Mean (SD) demographics and substance use variables represented by study.

Source study	Age	AUDIT	Drinks/ Drinking Day	Drinking Days/30	Cigarettes/ Smoking Day	Cigarette Days/30	Cannabis Days/30	AUDIT Skewness	AUDIT Kurtosis
Claus et al.32	28.80 (7.94)	13.73 (6.36)	4.17 (10.15)	15.31 (6.71)	4.723 (5.76)	13.30 (9.98)	5.37 (9.25)	0.788	3.347
Littlewood et al.33	38.93 (9.23)	22.60 (7.19)	0.153 (.23)	19.77 (7.92)	0.23 (.22)	12.66 (14.41)	1.52 (4.44)	0.0255	2.293
Littlewood et al.34	33.76 (9.65)	5.47 (5.16)	2.023 (5.14)	4.55 (6.86)	15.79 (6.23)	29.78 (.962)	4.30 (9.59)	1.345	4.579

Mean (SD) demographics and substance use variables represented by study. AUDIT: The nine-item Alcohol Use Disorders Identification Test; Drinks/Drinking Day: average number of standard drinks per reported drinking day within the past 30 days as measured by the TLFB; Drinking Days/30: number of reported drinking days within the past 30 days as measured by the TLFB; Cigarettes/Smoking Day: average number of cigarettes used per reported smoking day within the past 30 days as measured by the TLFB; Cigarette Days/30: number of reported days when cigarettes were smoked within the past 30 days as measured by the TLFB; Cannabis Days/30: number of reported days when cannabis was used within the past 30 days as measured by the TLFB; AUDIT Skewness: The degree of skewness on the AUDIT measure; AUDIT kurtosis: The degree of kurtosis on the AUDIT measure.

We conducted a power analysis for the primary models (ie, DRD2 methylation predicting RECN/LECN connectivity). The power analysis was specified with the following parameters: testing a single regression coefficient in linear multiple regression, based on a sample size of 621 subjects with three predictors (DRD2 methylation, age, and sex), with a two-tailed alpha significance threshold of 0.05 and 80% power. With this sample size, the primary analyses were fully powered to detect small effect size f ² = 0.01 (G*Power 3.1; Faul et al, 2009).³⁶

2.2 |. Measures

In addition to providing basic demographic information, participants responded to a battery of mood and substance use questionnaires. Relevant to the current study, participants completed the Alcohol Use Disorders Identification Test,³⁷ which assesses symptoms of alcohol abuse and dependence over the past 6 months. Given that the sample included aggregated data from three different subsamples, we felt it important to examine the distribution of the AUDIT measure, both within the full sample and within each of the subsamples. Within the full sample, the skewness of the AUDIT measure is.658, and the kurtosis of the AUDIT measure within the first sample is 2.908. See Table 1 for skewness and kurtosis of the AUDIT measure across each of the samples that comprise the full aggregate sample (ie, Littlewood et al³³; Littlewood et al³⁴; Claus et al³²). Skewness < 2 and kurtosis < 7 are commonly used cutoffs on these parameters.³⁸ Thus, this suggests that the distribution of the AUDIT measure for the full sample and for each of the subsamples that comprise the full sample is considered normal, which lends justification for combining the subsamples with regard to the AUDIT measure.

2.3 |. Image acquisition and preprocessing

Whole brain resting state functional magnetic resonance imaging (fMRI) was performed on a 3-T Siemens Trio scanner with a 12-channel radio frequency coil. In the scanner, tape was placed across the participants’ forehead to serve as feedback for movement reduction. T2*-weighted functional images were acquired using a gradient-echo echo-planer imaging (EPI) sequence: TE = 29 milliseconds, TR = 2 seconds, flip angle = 75°, slice thickness = 3.5 mm, slice gap = 1.05 mm, field of view = 240 mm, 64× 64matrix, voxel size = 3.75 mm × 3.75 mm × 3.5 mm. Resting-state scans were 5 minutes in duration. Subjects were instructed to keep their eyes open and fixate on a cross. High-resolution T1-weighted structural image were acquired with a 5-echo MPRAGE sequence with the following parameters: TE = 1.64, 3.5, 5.36, 7.22, and 9.08 milliseconds, TR = 2.53 seconds, TI = 1.2 seconds, flip angle = 7°, excitations = 1, slice thickness = 1 mm, field of view = 256 mm, resolution = 256 × 256 × 176, voxel size = 1 × 1 × 1 mm, pixel bandwidth = 650 Hz. Images were preprocessed with Statistical Parametric Mapping Version 5 (SPM5) software (WellcomeTrust Centre for Neuroimaging, London, UK) for slice-time and motion correction, and then normalization to Montreal Neurological Institute (MNI) space. Additionally, participants with excessive head motion (>2 mm translational or 0.035 radians rotational movement) were excluded from all analyses for additional quality control, consistent with the motion exclusion criterion used in other published studies from our group.²² A total of 65 subjects were excluded for excessive motion based on these parameters. The decision to exclude a given participant on the basis of excessive motion absolutely (rather than frame to frame) was based on findings suggesting that functional connectivity analyses are particularly vulnerable to excessive motion, and head motion is a relatively stable trait within individuals.³⁹

2.4 |. Network functional connectivity analysis

In order to investigate connectivity, we used methods consistent with previous studies conducted within our group.²² Briefly, we used the functionally defined intrinsic connectivity networks (ICNs), which have been defined previously.⁴⁰ These networks are derived from 90 distinct network nodes that represent regions of interest (ROIs) that span across the majority of cortical and subcortical gray matter. Using in house scripts,²² functional connections between nodes (ie, edges) were calculated by creating a correlation matrix between the time series of all nodes within each network for each subject. A Fisher r- to-z transformation was applied, yielding z-scores for use in subsequent analyses. As noted in Lynall et al,⁴¹ connectivity strength is a global measure of connectivity calculated as the mean of all pairwise correlations between nodes within each network. Given past findings that implicate RECN and LECN connectivity in AUD,^22,42 and previous literature suggesting an association between D2 functioning and measures of executive functioning,^8,43 we were particularly interested in the association between DRD2 methylation and connectivity within the LECN and RECN. In order to assess whether the association between DRD2 methylation and connectivity is unique to the ECN, we a priori selected a subset of additional networks for follow up analyses, including the default mode network (DMN), anterior salience (AS) network, posterior salience (PS) network, and the ventral default mode network (VDMN).

2.5 |. DNA preparation and methylation assay

Genomic DNA was isolated from saliva samples using Qiagen’s Puregene DNA prep kit (Qiagen, Venlo, Limburg). CpG sites within the DRD2 gene were selected a priori based on prior research in the lab. To determine the methylation status of CpG sites near of the DRD2 gene, pyrosequencing was performed at EpigenDX (Worcester, MA). Pyrosequencing quantitatively monitors the real-time incorporation of nucleotides through the enzymatic conversion of released pyrophosphate into a proportional light signal. The DRD2 methylation assay covered six CG dinucleotides across the DRD2 gene. Methylation assays were designed by EpigenDx. Site analysis was based on the ability to generate primers located around CpG islands and that meet the requirements for pyrosequencing of each site accurately. All primer sequences are owned by EpigenDx. Because methylation values across the six sites of interest were highly correlated, we created an average methylation variable (eg, average methylation across the six CpG sites assayed), which was used in all analyses outlined below.

2.6 |. Data analysis

R version 3.2.1 (2015-06-18) was used to perform all general linear models (GLMs), where we tested the association between DRD2 methylation and network connectivity, controlling for age and sex covariates. We planned to include age and sex as covariates in all analyses given that the literature suggests an association between age and methylation (see Jung and Pfeifer⁴⁴ for review), between age and functional connectivity,⁴⁵ and between sex and functional connectivity.⁴⁶ Based on findings from prior studies implicating the ECN connectivity with problem drinking²² as well as D2 function with substance abuse, we were particularly interested in testing the association between DRD2 methylation and the ECN. To start, we tested the relationship between DRD2 methylation and both the average connectivity within the RECN and LECN. We then chose four additional networks to test individually in GLMs, where we tested the association between DRD2 methylation and network connectivity for each of the four networks, controlling for age and sex covariates. The additional networks tested included DMN, AS, PS, and VDMN. These four additional networks served as comparison networks, which we used to determine whether the association between DRD2 methylation and connectivity was unique to the ECN in this sample of substance users.

As the ECN was the primary focus of our analyses, we tested the association between DRD2 methylation and “edges” (ie, connectivity between individual nodes within the LECN and RECN). Using a series of GLMs, we tested the relationship between DRD2 methylation and edge connectivity for each of the 15 edges within the RECN and LECN. We conducted these post hoc GLMs for the purpose of determining which edges within the networks were driving the association between RECN and LECN connectivity and DRD2 methylation. Based on these GLMs, edges were selected based on a test statistic threshold of P < 0.05 for the correlation between edge connectivity and DRD2 methylation. Subsequently, the edges that demonstrated a significant association between methylation and edge connectivity were used in mediation and moderation models. Based on these criteria, the top 4 edges within the RECN (ie, the top 4 edges that showed the most significant association) and the 3 and only significant edges within the LECN were included in follow-up models, where connectivity of these edges were tested as mediators and moderators between DRD2 methylation and AUDIT scores (see below for details).

Per our theoretical model, mediation analyses were executed using SPSS 19.0 (IBM Corp, 2010) via the products of the coefficients method, which estimates the following: a (independent variable [IV]-mediator [M]), b (M-independent variable [DV]), c (total effect of IV on DV), and ab (indirect effect of IV on DV via M) paths for each mediation model.^47,48 Because normality of the sampling distribution of total indirect effects cannot be assumed, bias-corrected 95% confidence intervals of the indirect effect were estimated by using bootstrapping methods.^47,48 We tested the edges that were significantly associated with DRD2 methylation within the RECN and LECN as mediators of the association between DRD2 methylation and our clinical variable of interest, AUDIT. This was tested using a multiple mediation model, where the significant edges were used as the mediators (Ms), AUDIT score was used as the DV, and DRD2 methylation was used as the IV.

In the event that the mediation model was not supported, we planned to follow up the mediation analyses by testing a series of exploratory interaction models. These models tested whether edge connectivity moderated the relationship between DRD2 methylation and AUDIT, again using R version 3.2.1 (2015-06-18) for the interaction model GLMs. For these models, we planned to test the interaction between each of the edges within the RECN and LECN that demonstrated a significant relationship between DRD2 methylation edge connectivity and DRD2 methylation on AUDIT (for each edge within the RECN and LECN, AUDIT score was regressed on edge connectivity, DRD2 methylation, and the edge connectivity*DRD2 methylation interaction), using separate models for each edge within the RECN and LECN).

Given that the study tests multiple models, we conceptualize the issue of multiple comparisons as follows. We consider the two models in which we test the relationship between DRD2 methylation and average RECN/LECN to be the primary models. Given that we used the other networks (ie, DMN, PS, and AS) as control networks, we used an uncorrected P < 0.05 significance threshold for the primary models. Next, we consider the follow up analysis of individual edges within both the RECN and LECN as a way of determining which edges within the overall networks are driving the association between LECN/RECN connectivity and DRD2 methylation. Given that the purpose of these follow up tests is to determine what is driving the overall effect, these tests are considered significant at the P < 0.05 threshold. This logic is akin to the argument made by Stevens⁴⁹ regarding multivariate analyses of variance (MANOVAs) and follow up univariate tests. Briefly, Stevens argues that if the overall MANOVA is significant, then the individual univariate tests are protected from multiple comparisons. In order to use this logic, we also tested the relationship between RECN edge connectivity and DRD2 methylation using 15 degree of freedom MANOVAs, which included all 15 RECN edges and all 15 LECN edges in two separate MANOVA models. Finally, the moderation and mediation analyses are considered exploratory post hoc analyses and are therefore presented as uncorrected P values.

3 |. RESULTS

3.1 |. Relations of DRD2 methylation and ECN connectivity

Controlling for age and sex, DRD2 methylation was significantly associated with average connectivity in the RECN (β = −0.0134; t₆₁₈ = −2.892; P = 0.00397), and the LECN (β = −0.0103; t₆₁₈ = −2.202; P = 0.028), such that greater methylation was associated with a lesser degree of right and left ECN connectivity. Results from these primary models remain consistent when including cigarette smoker status, marijuana use status, and parent study as covariates, as well as without the age and sex covariates. Results from these primary models also remain consistent when tested in two subsets of the full data set at the P < 0.05 threshold. Subset 1 excluded all subjects from the sample first reported in Littlewood et al,³⁴ which was characterized by heavy cigarette smoking, and thus only includes participants recruited on the basis of heavy drinking. Subset 2 was composed of subjects who reported AUDIT scores that were indicative of hazardous drinking/dependence.³⁷ Note that we calculated average network connectivity (eg, RECN and LECN) as the average connectivity between all nodes within each of the networks. Given that we use a MANOVA framework to justify the follow-up edge connectivity tests below, we also tested the association between RECN and LECN connectivity with 15 degree of freedom models in which each of the 15 RECN/LECN edges predicted DRD2 methylation. The 15 degree of freedom MANOVA models suggest that the set of 15 RECN edges are associated with DRD2 methylation (F _{15, 606} = 2.0094, P = 0.013), and the set of 15 LECN edges are associated with DRD2 methylation (F _{15, 606} = 2.173, P = 0.0063).

3.2 |. Relations of DRD2 methylation and connectivity within comparison networks

In order to assess whether the association between DRD2 methylation and connectivity was unique to the ECN in our data, we tested the association between DRD2 methylation and average connectivity within other major networks, including the DMN, AS network, PS network, and the VDMN network. Specifically, the AS and PS networks were chosen due to the role of dopamine in attention processes,⁵⁰ so these networks act as an active control for the possible unique effect of alcohol. In contrast, the DMN/VDMN were chosen as general comparison networks, given that activation of these networks are not largely driven by dopaminergic activation. Results from our data suggest that controlling for sex and age, there were no significant associations between DRD2 methylation and the DMN (β = −0.00453; t₆₁₈ = −1.050; P = 0.294), VDMN (β = 0.00094; t₆₁₈ = 0.243; P = 0.808), AS (β = −0.00653; t₆₁₈ = −1.599; P = 0.119), and PS (β = −5.903 × 10⁻⁵; t₆₁₈ = −0.023; P = 0.98). These results suggest that DRD2 methylation is associated with RECN and LECN connectivity and is not associated with a more global reduction in connectivity across widespread networks in our data.

3.3 |. Relations of DRD2 methylation and edge connectivity within the RECN and LECN

Given that the RECN and LECN were associated with DRD2 methylation among networks tested, we set out to identify which edges within the RECN and LECN drove these associations. In order to determine which specific connections between nodes in the right and left ECN (ie, edges) were associated with DRD2 methylation, we conducted a series of GLMs, controlling for age and sex. Within the RECN, five edges demonstrated an association with DRD2 methylation with an associated test statistic of P < 0.05: the edge including the left cerebellum and right caudate (LCere-RCaudate; β = −0.0152; t₆₁₈ = −2.102; P = 0.0359), right middle frontal gyrus and the right inferior parietal gyrus (R DLPFC-R Parietal; β = −0.0228; t₆₁₈ = −2.536; P = 0.0115), the right middle frontal gyrus and the right middle orbitofrontal gyrus (R DLPFC-R MFG; β = −0.0216; t₆₁₈ = −2.179; P = 0.0297), the right middle orbitofrontal gyrus and the right inferior parietal gyrus (R MFG-R Parietal; β = −0.0233; t₆₁₈ = −2.464; P = 0.0140), and the right inferior parietal and the right superior medial frontal gyrus (R Parietal-R mSFG; β = −0.0215; t₆₁₈ = −2.548; P = 0.0111). Within the LECN, three edges demonstrated an association with DRD2 methylation with an associated test statistic of P < 0.05: the edge including the left middle frontal gyrus and the left angular lateral occipital (L DLPFC-L Parietal; β = −0.0188; t₆₁₈ = −2.117; P = 0.0347), the left frontal pole and left middle temporal gyrus (L MFG-L Temporal; β = −0.0204; t₆₁₈ = −2.090, P = 0.0370), and left middle frontal gyrus and left middle temporal gyrus (L DLPFC-L Temporal; β = −0.0208; t₆₁₈ = −2.180; P = 0.0296). See Figure 1 for representation of networks within the RECN and LECN that drive the association between DRD2 methylation and right and left ECN connectivity. Each of the models that indicated a significant association between RECN/LECN edge connectivity and DRD2 methylation were tested again using subsets 1 and 2 in order to determine whether results were consistent across the full sample and each of the subsets. The direction of the association between DRD2 methylation and edge connectivity was consistent across all five edges. Results replicated across all three samples (full aggregate sample, subset 1, and subset 2) at P < 0.05 significance threshold for three of the eight edges (R DLPFC-R Parietal; R DLPFC-R MFG; R MFG-R Parietal). Results replicate across the full aggregate sample and subset 1 at P < 0.05 for six of the eight edges (R DLPFC-R Parietal; R DLPFC-R MFG; R MFG-R Parietal; R Parietal-R mSFG; L DLPFC-L Parietal; L MFG-L Temporal).

FIGURE 1.

Representation of edges within the right executive control network (RECN) (blue) and left executive control network (LECN) (green) that demonstrate significant association with DRD2 methylation. Degree of synchronous activation of two regions of interest (ROIs) (edge) is significantly associated with methylation. Five edges within the RECN and three edges within the LECN show this significant association. The top 4 RECN significant edges and each of the significant LECN edges were advanced for testing in interaction models

3.4 |. Relations of DRD2 methylation and clinical outcomes of AUD severity

Given previous findings in the literature, we were interested in whether DRD2 methylation was also associated with drinking problems severity among a sample of polysubstance users. In our sample, controlling for age and sex, average methylation of sites within the DRD2 gene were significantly associated with AUDIT score (β = 1.139; t₆₄₈ = 4.289; P = 2.07 × 10⁻⁵). Such that greater DRD2 methylation was associated with higher AUDIT scores. This suggests that greater DRD2 methylation, on average across the sites of interest, is associated with greater severity of problem drinking, which is consistent with the Bidwell and colleagues’³⁰ findings.

3.5 |. Mediational analyses

Given that connectivity within the ECN has been associated with clinical status of problem drinking,²² and DRD2 methylation was associated with clinical outcomes of AUD in our data, we were interested in whether RECN and LECN connectivity mediated the relationship between DRD2 methylation and AUDIT, the most reliable and widely used clinical measure of AUD severity. Included as mediators in these models were the edges that drove the association between DRD2 methylation and RECN/LECN connectivity (see above for description of edge selection).

DRD2 methylation (IV) was associated with decreased connectivity within each of the four RECN edges tested in a multiple mediation model (M; a_1–4, RECN pathways; see Figure 2B). However, the association between connectivity within all four RECN edges tested (Ms) and total AUDIT score (DV) were not significant (b_1–4, RECN pathways; see Figure 2B). Additionally, the indirect effect of DRD2 methylation (IV) on AUDIT total score (DV) through connectivity within each of the four RECN edges tested (M) was not significant (ab_1–4, RECN path). See Figure 2B for a summary of RECN multiple mediation results.

FIGURE 2.

A-C, Mediation models testing connectivity of edges within the left executive control network (LECN) and right executive control network (RECN) (that were significantly associated with DRD2 methylation) as mediators between DRD2 methylation and AUDIT. Standardized coefficients are shown in the model pathways, represented as follows: a (independent variable [IV]-mediator [M]), b (M-independent variable [DV]), c (total effect of IV on DV), and ab (indirect effect of IV on DV via M) paths for each mediation model. Mediation was tested via the products of the coefficients method⁴⁷ with DRD2 methylation as the IV, three LECN edges (m_1–3, LECN) as simultaneous mediators, and AUDIT, a continuous measure of alcohol use disorder (AUD) severity (DV). A parallel model was run substituting four RECN edges (m_1–4, RECN) as the simultaneous mediators. Gender and age were included as covariates in both of the models. The results indicate that the mediation model was not supported, as none of the edges demonstrated significant indirect effects

With regard to the three edges from the LECN advanced for mediation analysis, DRD2 methylation (IV) was associated with decreased connectivity within the L DLPFC-L Temporal and L MFG-L Temporal edges (M; a_2–3,LECN pathways), but not the L DLPFC-L Parietal (M; a_1,LECN pathway). The association between connectivity within each of these LECN edges (M) and total AUDIT score was not significant (DV; b_1–3, LECN path; see Figure 2C). The three LECN edges tested in the multiple mediation model did not mediate the relationship between DRD2 methylation and AUDIT (ab_{1–3, LECN} path; see Figure 2C). See Figure 2C for a summary of LECN multiple mediation results. These findings suggest that DRD2 methylation impacts ECN via a direct pathway, but an indirect pathway from DRD2 methylation to AUD severity via ECN connectivity is not supported.

3.6 |. Moderation analyses

Given that the mediation models whereby ECN connectivity mediates the relationship between DRD2 methylation and AUDIT were not supported, we were interested in whether ECN connectivity moderated the relationship between DRD2 methylation and AUDIT. We planned to test the same edges within the LECN and RECN that drove the association between DRD2 methylation and network connectivity (consistent with the mediation analyses; see above for edge selection) in exploratory interaction models. The three edges within the LECN that were selected based on their association with DRD2 methylation did not significantly moderate the relationship between DRD2 methylation and AUDIT, when tested using three separate multiple regression interaction models, controlling for age and sex—L DLPFC-L Parietal (β = −1.79; t₆₁₀ = −1.743, P = 0.0819); L MFG-L Temporal (β = −445; t₆₁₀ = −0.467, P = 0.641); L DLPFC-L Temporal (β = 0.664; t₆₁₀ = 0.724, P = 0.469). Three edges within the RECN, however, significantly moderated the relationship between DRD2 methylation and AUDIT at a P < 0.05 significance threshold, again using separate multiple regression interaction models for each edge, controlling for age and sex—R DLPFC-R Parietal (β = −2.558; t₆₁₀ = −2.324, P = 0.0204); R DLPFC-R MFG (β = −2.228; t₆₁₀ = −2.216, P = 0.027); R MFG-R Parietal (β = −2.405; t₆₁₀ = −2.285, P = 0.0227); R Parietal-R mSFG (β = 0.0239; t₆₁₀ = 0.021, P = 0.9833). See Figure 3A–C for a visual representation of simple slopes of DRD2 methylation at different levels of RECN edge connectivity for the three edges in the RECN that demonstrated a significant interaction. As seen in the figures, results from our interaction models suggest that at lower levels of RECN edge connectivity, the relationship between DRD2 hypermethylation and disorder severity becomes stronger. Moderation results are presented uncorrected and did not meet significance using a corrected P value. None of the edge connectivity*DRD2 methylation moderation models were significant using subset 1 or 2. Additionally, interaction models testing the average RECN*DRD2 methylation interaction and average LECN*DRD2 methylation interaction effects on AUDIT controlling for age and sex were not supported.

FIGURE 3.

A-C, Graphical representations of the significant interactions between average DRD2 methylation and edge connectivity on AUDIT. Edges were selected based on their significant association with DRD2 average methylation. Across the three significant interaction models, the relationship between AUDIT and DRD2 methylation depends on connectivity within this edge, whereby the relationship between methylation and AUDIT becomes stronger at lower levels of connectivity

4 |. DISCUSSION

4.1 |. Summary of the current investigation

The present study set out to evaluate the relationships among methylation of CpG sites within a regulatory region of the DRD2 gene, ECN connectivity, and severity of drinking problems among an ecologically valid sample of polysubstance users, whose consumption of alcohol, nicotine, and cannabis span across a spectrum of quantity and frequency of use. We hypothesized that DRD2 methylation would be associated with reduced connectivity within the ECN and measures of drinking problems severity. Results suggest that methylation of the DRD2 gene is indeed associated with reduced connectivity within the right and left, ECN. Additionally, DRD2 methylation demonstrates clinical relevance, as average DRD2 methylation across sites within a regulatory region of the gene is associated with measures of drinking problems severity in our sample of polysubstance users. This suggests that greater DRD2 methylation, on average across the sites of interest, is associated with greater severity of problem drinking, which is consistent with the Bidwell and colleagues’³⁰ findings. We also hypothesized that ECN connectivity would mediate the relationship between DRD2 methylation and AUDIT, thereby serving as a mechanism linking epigenetic changes to behavioral measures, but this model was not supported. Connectivity of certain edges connecting the frontal and parietal regions within the RECN did, however, moderate the relationship between DRD2 methylation and AUDIT, suggesting the possibility that ECN connectivity and DRD2 methylation have a synergistic effect on disorder severity. Importantly, due to the cross-sectional nature of the data, it is not clear whether the ECN connectivity and DRD2 methylation contribute to the development of AUD, if connectivity decrements within the ECN and DRD2 hypermethylation are consequences of the disorder, or both. Additionally, while three edges within the RECN demonstrated an edge*DRD2 methylation interaction on AUDIT, these results are significant at an uncorrected P value threshold.

4.2 |. Results in the context of prior literature

Although the present investigation is the first study to examine the association between DRD2 methylation and ECN connectivity, our work builds upon past research that has linked D2 functioning to dysregulation in executive control neurocircuitry and behavioral outcomes characteristic of AUD. For example, a review by Volkow and Morales discusses a series of studies suggesting that reduced D2 signaling in the striatum is associated with reduced neural activity in regions within the control network (eg, PFC).⁵¹ These findings provide evidence that D2 signaling might be associated with neural correlates of ECN functioning. The present analyses add to this literature by suggesting an association between D2 methylation and another neural correlate of ECN functioning (ie, connectivity). Additionally, one particular study found that lower extrastriatal D2 binding is associated with variability in executive functioning.⁵² Similarly, literature suggests that downregulation of D2 availability has been associated with and poor reversal learning.^53,54 Additionally, researchers have proposed that changes in D2 functioning is a mechanism by which frontocorticostriatal circuitry becomes dysregulated, which in turn contributes to poor reversal learning within states of addiction.^55,56 Furthermore, D2 functioning has been associated with functional measures of executive functioning. The “Go-No-Go” paradigm, a frequently used in-scanner task that taps executive functioning, has demonstrated differences between AUD populations and controls.^57,58 A nice review article summarizes how dopamine modulation, including D2 receptor disinhibition during negative prediction errors, is associated with “NoGo” learning.⁵⁹

Specific edges that were found to be significant in the current set of analyses have been implicated in specific executive functions that have well-documented relationships with AUD. Of particular import, three of these regions, the DLPFC, OFC, and inferior parietal gyrus, also demonstrate network edge sensitivity to D2 receptor function. For instance, impaired attentional control and response inhibition has been associated with reduced connectivity between prefrontal regions and inferior parietal regions.⁶⁰ Such fronto-parietal connectivity, particularly in the right hemisphere, plays a major role in maintaining attention towards current goals, and responding flexibly to new and relevant environmental information.⁶¹ Indeed, the orbitofrontal cortex and its relationship with DLPFC and inferior parietal regions has been shown to matter in several cognitive functions associated with AUD.^62–64 Its connectivity with these regions, in concert with its input from regions associated with emotional memory and sensory cortices, renders it central to set shifting, reversal learning, and cognitive flexibility.^65–67

Furthermore, previous work in impaired control and AUD has implicated suboptimal microstructural connectivity (the physical connection among network edges) between frontal and parietal regions.⁶⁸ Fronto-parietal connectivity appears to be central to highly adaptive control processes, where proper functioning might prevent onset or worsening of AUD symptoms.⁶⁵ D2 receptor function has been implicated in disruption of nodes within these ECN edges, which are in turn linked to behavioral outcomes involved in AUD.⁶⁹ Prefrontal regions appear to be particularly affected⁷⁰ and have downstream impact on the functional integrity of edges that they are a part of, such as the DLPFC-inferior parietal edge.⁶⁰ Methylation-induced change in D2 receptor expression may also influence ECN through altered ratio with D1 receptors in the prefrontal cortex, with functional implications for components of executive function such as working memory and behavioral flexibility, and information processing during times of executive demand.^71–73 Optimal balance between D1 and D2 receptors in prefrontal cortical regions is credited with maintaining a high signal-to-noise ratio of incoming information.⁷⁴ Thus, DRD2 methylation-related disruptions in receptor expression may impair the neuromodulatory role of dopamine during ECN functions linked to AUD, such as inhibitory control, in part by changing individuals’ dopamine dose-response curves.^75,76

Current theoretical models of AUD suggest that the disorder is driven by an imbalance between the reward and control neural networks in response to heavy, chronic drinking, whereby the reward network becomes strengthened and the control network becomes weaker. In turn, this network imbalance contributes to continued heavy drinking despite adverse consequences. These models also acknowledge the potential role of epigenetic modifications that contribute to dysregulation within reward and control neurocircuitry. While mounting research supports various pathways of the model, few efforts have been made to elucidate the mechanistic and causal relationships between epigenetic modifications, integrity of neurocircuitry, and behavioral outcomes.

While our cross-sectional data cannot directly address questions of causality, our findings lend support to these theoretical models. Specifically, our results suggest that DRD2 epigenetic modifications are associated with aberrant neurocircuitry (reduced ECN connectivity) as well as outcomes of drinking related problems (AUDIT score), consistent with Bidwell et al.³⁰ Results from our mediation analyses suggest that there is no direct causal effect of methylation on clinical outcomes through connectivity. This suggests that connectivity might still be an important factor in neuruocircuitry imbalance in the development of AUD, but our data suggest that it is not part of the mechanistic pathway between epigenetic modifications and AUD severity. However, our moderation results leave room for the role of neurocircuitry dysregulation on disorder severity, suggesting that greater ECN connectivity attenuates the strength of the association of DRD2 hypermethylation on AUD severity. Our results generally support the idea that DRD2 epigenetic modifications and control network dysregulation are involved in the development of AUD. Future studies should address these causal questions by utilizing a longitudinal study design.

4.3 |. Methodological considerations

It is important to highlight several methodological considerations and limitations of the present investigation. This is an analysis of resting state connectivity and methylation quantified based on peripheral tissues samples. A PET study, which would allow us to test the of D2 availability as a mechanism of reduced ECN connectivity with this sample size, is not feasible, nor is obtaining brain tissue for quantification of DRD2 methylation. However, quantifying methylation using peripheral tissue types, such as saliva samples, is feasible on a large scale. Therefore, our methods allow for a larger sample size than would be feasible with a PET study design, and with this increased power, we are able to flesh out the mediating vs moderating role of executive network connectivity in the relationship among DRD2 methlyation and AUD severity. Additionally, an important limitation of the present study is the cross-sectional design. A longitudinal design will be required to determine the precise causal relationship between DRD2 methylation, reduced connectivity in the ECN, and disorder severity. Furthermore, the data used in the present set of analyses were aggregated across studies, each of which had different aims and inclusion criteria. However, we sought to mitigate concerns about this point by attempting to replicate the significant findings throughout the paper with subsets of the full dataset (ie, a subset that excluded the heavy smoker sample and a subset that included only subjects who reported AUDIT scores >8). Results from these follow-up models using the two subsets indicate that the primary models replicate across all sample subsets, thus bolstering our confidence in the results from the full dataset.

The sample is characterized by polysubstance use given that subjects across the three studies used alcohol, cannabis, and nicotine (see Table 1), which can be considered both a methodological strength and limitation. While the current analyses are focused on the association between alcohol problem severity, DRD2 methylation, and functional connectivity, the wide range of polysubstance use represented within the sample is broadly representative of the spectrum of polysubstance use that typically characterizes AUD patients. For example, empirical work suggests that alcohol use and nicotine use commonly co-occur.⁷⁷ Therefore, the sample benefits from ecological validity. Despite the ecological validity of the current sample, future studies should consider testing the association between methylation, functional connectivity, and alcohol problem severity among a sample with less polysubstance use in order to test whether the findings replicate in a sample of AUD patients who do not use other substances. Additionally, future studies should consider collecting information on severity of problem drug use for substances other than alcohol (eg, cannabis use disorder) in order to determine whether severity of polysubstance use is associated with markers of gene regulation or functional connectivity.

Furthermore, there are a number of unmeasured variables that could be associated with the predictors and/or outcomes of interest. For example, educational attainment has been associated with resting state connectivity.⁷⁸ Therefore, future connectivity studies should consider collecting data on educational attainment to add as a potential covariate. Finally, the results from the moderation analyses were post hoc and are presented as uncorrected P values. While the moderation results reported in this paper are exploratory, we encourage other research groups to attempt to replicate our findings.

4.4 |. Conclusions and future directions

Despite these limitations, the current investigation represents a distinct and novel contribution to the literature. These analyses are among the first to suggest a link between peripheral epigenetic markers and neural circuitry. In particular, the results from the a priori primary models are novel, and an important contribution to clinical science at the intersection of epigenetics and neural functioning. Some recent studies have found an association between functional connectivity and methylation.⁷⁹ To our knowledge, however, no published work has shown an association between functional connectivity and epigenetic modifications in the context of alcohol use/polysubstance use. Furthermore, our sample size is significantly larger than other studies that have investigated the association between functional connectivity and epigenetic markers.

Our results lend evidence to the potential role of epigenetic regulation of DRD2 in neural network dysregulation and behavioral outcomes of AUD among a sample of polysubstance users. Our work suggests an association between DRD2 methylation and reduced ECN connectivity. Furthermore, results from our moderation models suggest that DRD2 methylation and reduced connectivity within the ECN might interact with one another to subserve disorder severity. Given the limitations discussed above, future investigations should continue to investigate epigenetic regulation of genetic loci and networks that have been associated with AUD and related disorders.