Association Between Brain Structure and Alcohol Use Behaviors in Adults

Key Points Question Are there directional associations between cortical or subcortical macrostructure and alcohol use? Findings This mendelian randomization study including 763 874 participants in UK Biobank, Enhancing NeuroImaging Genetics through Meta Analysis (ENIGMA), Psychiatric Genomics Consortium (PGC), and GWAS & Sequencing Consortium of Alcohol and Nicotine use (GSCAN) studies identified a significant negative association between genetically predicted global cortical thickness and alcohol consumption and binge drinking. Downstream multiomic analyses indicate that 17q21.31 genes and glutamatergic cortical neurons contribute to this association. Meaning The results from this study support emerging literature suggesting that cortical structure is associated with alcohol use and identify transcriptomic and cellular associations between these phenotypes that warrant further investigation.

A lcohol misuse causes large health and economic burdens globally 1,2 and is a leading risk factor for premature death and disability in individuals aged 15 to 49 years. 1 Heavy alcohol consumption impairs the nervous system, [3][4][5][6] which may lead to neurological, cognitive, and psychiatric health ramifications. 7 Altered macroscale brain structure is associated with psychopathology 8 and could represent a mechanistic link between alcohol-associated neurotoxicity and health outcomes. Studies have consistently associated greater alcohol use and alcohol misuse with lower cortical and subcortical volumes. [9][10][11][12][13][14][15][16] However, the directionality of these associations remains unclear, with some studies suggesting a predispositional impact of brain anatomy on alcohol use, 9,11,13,17 challenging the notion that brain structure changes as a result of alcohol exposure. 14, [18][19][20][21] Robert et al 13 analyzed longitudinal adolescent brain imaging data among 726 individuals and concluded that a greater rate of gray matter atrophy in frontal and temporal regions may lead to greater frequency of drunkenness. Similarly, a 2021 latent causal variable analysis 11 suggested that greater pars opercularis volume, greater cuneus thickness, and lower basal forebrain volume were associated with increased alcohol misuse. By contrast, other studies continue to suggest that alcohol use alters neuroanatomy. For example, a 2021 co-twin study 18 among 436 individuals found that alcohol exposure and genetic predisposition to alcohol use decreased thickness in multiple cortical regions. These studies highlight the ongoing debate regarding the directionality of associations between brain structure and alcohol use. Randomized clinical trials conducted to infer causality cannot be ethically or practically applied to study these associations and tens of thousands of participants may be required to identify replicable associations between brain magnetic resonance imaging (MRI) measures and behavioral traits. 22 Alternative approaches using large data sets are needed to characterize associations between brain structure and alcohol use.
Recently developed genomics methods, including latent causal variable analysis and mendelian randomization (MR), facilitate the identification of directional associations between genetically influenced variables from populationbased observational data and have been underapplied to questions regarding alcohol use and brain structure. 18,23-25 Latent causal variable analysis only evaluates 2 phenotypes 26 and does not explicitly test bidirectional associations. 27 By contrast, MR is frequently used to evaluate directionality in neuropsychiatry. 28 The multivariable extension of MR (MVMR) enables the assessment of multiple exposures to identify the direct association of each exposure with an outcome, 29 which could help clarify the associations between brain structure and alcohol consumption accounting for potential mediating or confounding phenotypes.
In this study, we investigated associations between brain anatomy and alcohol use using summary-level genome-wide association study (GWAS) data for brain MRI measures and alcohol-related phenotypes. Our primary MR associated genetically predicted global cortical thickness (GCT) with alcohol use. We investigated whether GCT broadly associates with substance use by evaluating its association with smoking. Given differences in alcohol use patterns between men and women, 30 we examined sex-specific associations between GCT and alcohol use. Next, we used MVMR accounting for confounding or mediating phenotypes to test the robustness of our GCT findings and performed multiomic analyses, including transcriptomic imputation 31 and cell-type enrichment analysis, 32 to describe the biological underpinnings of GCT-alcohol use associations. Figure 1 presents a study overview. This study is reported in accordance with the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) reporting guideline (eTable 1 in Supplement 1). 33 This study uses deidentified publicly available data, so no ethical approval from an institutional review board was required. The study protocol was not preregistered.

Data Sources
Summary-level data were obtained from GWAS. Included GWAS have existing ethical permissions from their respective institutional review boards and include participant informed consent with rigorous quality control. Participants with missing phenotypic data were excluded from source GWAS. Exact data for sex and ethnicity were not available from all sources and are reported here as approximate percentages.

Cortical and Subcortical Structure
Our primary analyses evaluated 19 measures of cortical and subcortical brain structure sampled with MRI. We analyzed measures of global cortical thickness (GCT) and global cortical surface area (GCSA) from a recent GWAS of T1-weighted MR images from 1.5-3 T scans (n = 33 709). 34 We performed secondary analyses on 34 regional cortical measures (eMethods in Supplement 1), but emphasized global averages rather than regional phenotypes because global measures may be less impacted by interindividual neuroanatomical variability, 35 the limited functional relevance of gyral-based atlases like the Desikan-Killiany atlas used by Grasby et al, 34,36 and the multiple-testing burden associated with a hypothesis-free regional analysis. GCSA was measured at the gray-white matter boundary and GCT was defined as the average distance between white matter and pial surfaces across both cortical hemispheres. 34 Images were processed using FreeSurfer. 34,37 Additionally, because a previous mega-analysis identified left-right hemispheric asymmetry in associations between subcortical structure and alcohol use, 12 we focused our subcortical analysis on left-right volumes (n = 19 629). 38 We evaluated 17 left-right volumes derived from T1-weighted MR images from 1.5-3 T scans: amygdala, hippocampus, accumbens, putamen, pallidum, thalamus, insula, caudate, and brainstem (combined volume). Mindboggle 101 atlases were used to label subcortical structures. 38 MRI data were processed using advanced normalization tools. 38 In exploratory bidirectional analyses, we analyzed phenotypes from a GWAS of overall subcortical volumes. 39 We also investigated the association of genetically predicted alcohol consumption with longitudinal changes in brain structure (eMethods in Supplement 1).

Alcohol Use Behaviors
We used 3 GWAS of alcohol use behaviors in predominantly EA samples: a meta-analysis GWAS 40 of alcoholic drinks consumed per week (DPW) (n = 537 349), a GWAS 41 of binge drinking frequency among participants in the UK Biobank (n = 143 685), and a case-control GWAS 42 of alcohol use disorder (AUD) (8845 individuals with AUD and 20 657 control individuals; total n = 29 502). Further description of these studies can be found in the eMethods in Supplement 1.

MR MR Instruments
The eMethods in Supplement 1 provides detailed methodology for instrument clumping, evaluations of instrument strength, sample overlap, procedures for missing instrument data, and testing of MR assumptions (eFigure 1 in Supplement 1). Our DPW and global cortical structure instruments included all genome-wide significant (GWS) single-nucleotide variants (SNVs) at a threshold of P < 5 × 10 −8 . Like previous neuropsychiatric MR studies evaluating exposures with few GWS SNVs, 43,44 we used a P value threshold of 5 × 10 −6 to select AUD, binge drinking, subcortical structure, and regional thickness instruments (eTable 3 in Supplement 1 and eTables 4-6 in Supplement 2). We performed MR using SNVs within the ADH1B (alcohol dehydrogenase 1B) locus, a primary enzyme in alcohol metabolism, 45 as sensitivity analyses further assessing relationships of alcohol use on brain structure (eMethods in Supplement 1).
We evaluated our main GCT findings with additional MR. First, we tested the association of 34 regional cortical thickness exposures 34 with alcohol use. Next, we evaluated the associations of GCT and smoking behaviors and examined potential sex-specific associations of GCT and alcohol use using sex-specific alcohol use GWAS data from the UK Biobank.  Finally, we performed 11 MVMR analyses (eFigures 2 and 3 in Supplement 1) accounting for neuropsychiatric phenotypes, substance use, trauma, and neurodegeneration. We concatenated, extracted, and harmonized the independent instrument sets for GCT and the controlled-for exposure with each alcohol use behavior using standard MVMR methods (eMethods in Supplement 1 and eTables 7-17 in Supplement 2). 46 We also performed leave-1-out MR 47 and investigated the biological function of the GCT instrument with a gene-set enrichment analysis (eMethods in Supplement 1).

MR Statistical Analysis
We used the conventional inverse-variance weighted estimator (IVW) as our primary MR method. We supplemented IVW MR with MR-Egger, weighted median, weighted mode, and simple mode estimators, which rely on different assumptions than IVW. 48 55 We conservatively defined significance at a false discovery rate (FDR) of 0.05 for each MR analysis (eMethods in Supplement 1). We also discuss nominally significant results (P < .05). We report MR estimates as β values representing a change in outcome units per change in exposure unit. The unit for DPW was log-transformed, 40 AUD was a binary measure, and binge drinking frequency was a categorical measure quantified as (0) never (1) less than monthly (2) monthly (3) weekly (4) daily/almost daily. Brain structure phenotypes were quantified as cortical thickness (mm); cortical surface area (mm 2 ), and subcortical volumes (cm 3 ) (eMethods in Supplement 1).

Transcriptome-Wide Association Studies (TWAS)
We used the FUSION method 31 to identify gene transcriptlevel associations with the alcohol use and brain structure phenotypes. To perform TWAS, we integrated alcohol use and GCT GWAS summary statistics with cortical RNA sequence reference panels from the CommonMind Consortium 56 and the Genotype-Tissue Expression Consortium 57 (eMethods in Supplement 1).

Cell-Type Enrichment Analyses
We used Cell-Type Expression-Specific Integration for Complex Traits 32 with default parameters to perform cell-type enrichment analyses using the alcohol-associated GWAS data and single-cell RNA sequencing data of 120 cortical cell types (56 excitatory neurons, 54 inhibitory neurons, and 10 nonneuronal cells) from the Allen Brain Map Human Multiple Cortical Areas SMART-sequence data set 58 (eMethods in Supplement 1).

Bidirectional MR Reveals Negative Association Between Genetically Predicted Global Cortical Thickness and Alcohol Use Behaviors
The main bidirectional MR analyses included 763 874 individuals who were predominantly of European ancestry (more than 94%). Cohorts had mean ages between 40 and 63 years, and 52% to 54% of included individuals were female (eTable 2 in Supplement 1). Analyses revealed significant associations of GCT with alcohol use at FDR = 0.05. These associations were unidirectional. The MR analyses failed to find any nominally significant associations between genetically predicted alcohol use and GCT (eTables 18 and 19 in Supplement 2). Conversely, we found that genetically predicted GCT has a negative association with DPW and binge drinking frequency (DPW β, −0.88; 95% CI −1.36 to −0.40; P = 3.58 × 10 −4 ; binge drinking β, −2.52, CI; −4.13 to −0.91; P = .002) ( Figure 2; IVW estimators are presented unless otherwise specified). The associations between GCT, DPW, and binge drinking remained significant using weighted median and MR-Lasso estimators, supporting the validity of the IVW estimate (Table 1). Regarding associations between global cortical surface area (GCSA) and alcohol phenotypes, one finding suggested genetically predisposed GCSA was positively associated with DPW (β, 3.87 × 10 −6 ; 95% CI, 1.16 × 10 −6 to 6.59 × 10 −6 ; P = .005); however, other MR methods did not corroborate this association. Additionally, unlike GCT, GCSA was not associated with binge drinking.
MR identified nominally significant unidirectional associations between binge drinking and right amygdala volume (β, −0.19; 95% CI, −0.35 to −0.04; P = .01) and between AUD and right putamen volume (β, −0.04; 95% CI, 0.00 to 0.08; P = .04) (Figure 2; eTable 21 in Supplement 2). However, exploratory bidirectional results using overall subcortical volumes were null (eTables 22 and 23 in Supplement 2). MR estimates using cis-ADH1B instruments were also null for both cortical and subcortical structures (eTables 24-26 in Supplement 2), as were exploratory estimates of the associations of alcohol use with age-independent and age-dependent longitudinal changes in brain structure (eTables 27-30 in Supplement 2). Ultimately, our most robust finding was an association between genetically predicted GCT and alcohol use, motivating the focus of our downstream analyses.

MR Testing the Robustness of GCT-Alcohol Consumption Associations
To investigate whether specific cortical regions underlie the association between GCT and alcohol use, we performed MR using regional thickness phenotypes as exposures and DPW, binge drinking, and AUD as outcomes. No results approached FDR significance (eResults in Supplement 1 and eTables 31, 32, and 33 in Supplement 2). Additionally, we failed to find an association between genetically predicted GCT and smoking, and exploratory MR identified no sex differences in GCT-alcohol consumption associations (eResults in Supplement 1 and eTables 34 and 35 in Supplement 2). Leave-1-out analyses found

Cell-Type Enrichment Analysis-Associated Glutamatergic Cortical Neurons With Alcohol Consumption
eTable 43 in Supplement 2 contains full results from our celltype enrichment analysis. We found a total of 31 nominally significant associations between a cell type and a alcohol use behavior representing 30 distinct cell types, 27 of which are excitatory glutamatergic cells and 3 of which are inhibitory GABAergic cells. Twelve excitatory cell types remained significant at FDR = 0.05, including 10 cells associated with DPW and 2 associated with binge drinking (Figure 3).

Discussion
This MR study used large population-based data on the genetic architecture of cortical and subcortical structure, 34,38 MR, and novel multiomic methods to identify directional and biological associations between human brain structure and alcohol use. Our large sample sizes (between 19 629 and 537 349 participants 34,38,40-42 ) increased statistical power relative to previous brain structure-alcohol consumption studies. 13,18 Our findings suggest that a predisposition toward lower GCT may be associated with greater alcohol consumption and binge drinking. Conversely, we failed to find strong evidence that a genetic predisposition for alcohol use was associated with brain structure or its longitudinal plasticity. More modestly, our study suggests genetically predicted right pallidum volume was positively associated with alcohol consumption. This finding was not replicated in either our MR of overall subcortical volumes or a recent MR by Logtenberg et al 59 investigating substance use and overall subcortical volumes. Additionally, while Logtenberg et al 59 associated alcohol dependence with reduced overall amygdala and hippocampal volumes, after multiple testing corrections, we failed to associate a genetic liability for binge drinking, DPW, AUD, or an ADH1B instrument with these regions using hemispheric, overall, and longitudinal subcortical outcomes. Discrepancies between our studies may have resulted from differences in statistical methodology, power, or the specific phenotypes evaluated (eDiscussion in Supplement 1).
Our consistent identification of an association between genetically predicted GCT and alcohol use behaviors across MR methods and sex-specific analyses implicates the cortex as a potential driver of vulnerability to alcohol consumption and binge drinking. Interestingly, GCT had no association with smoking, suggesting its association with alcohol consumption may not reflect a broader association with substance use. Our failure to identify specific cortical regions associated with alcohol use may mean larger data sets are needed to characterize such associations. Importantly, GCT estimates from MVMR analyses remained significant when accounting for 11 potential mediators or confounders. The reduction in GCT effect estimates in MVMR models accounting for cognition suggests mediation of the GCT-alcohol use associations, especially given the significant MVMR estimates for cognition on alcohol use (eTable 37 in Supplement 2). Additionally, while we failed to find evidence for alcohol-associated cortical thinning in a population of adults with a mean (SD) age of 40 (8) years, alcohol use could cause cortical thinning in younger adults and adolescents due to increased cortical plasticity during these developmental stages. 19 For instance, recent work analyzing young adults showed that alcohol use predisposition leads to decreased thickness of cortical control and salience networks. 18 While participant age and other methodological particularities may influence the results of studies investigating alcohol-brain structure interactions, we found that for a middle-aged population, alcohol use primarily followed cortical anatomy. Our investigation of the transcriptomic relationship between GCT and alcohol use identified 5 protein coding genes oppositely associated with GCT and alcohol use behavior: PLEKHM1, LRRC37A2, CRHR1, ARHGAP27, and LRRC37A. These 5 genes could contribute to the negative association between GCT and alcohol use. All 5 are located at 17q21.31. This locus, characterized by extensive linkage disequilibrium, 60 is the site of 2 haplotypes: the inverted H2 haplotype (found in approximately 20% of individuals of European ancestry), and the H1 haplotype. 61 Comparing our imputed transcriptomes with limited cortical RNA-sequence data and past association studies suggests that lower GCT and greater alcohol use may be associated with the H1 haplotype. 62, Notably, CRHR1 encodes a G-protein coupled receptor that binds corticotropin-releasing hormone and its agonists. In line with our findings, CRHR1 upregulation in the amygdala 63,64 and cortex 63 have been associated with greater alcohol consumption and dependence. Additionally, previous studies have suggested that CRHR1 modulates the behavioral and cognitive outcomes associated with stress. 65,66 CRHR1 may also affect cortical macrostructure, as previous studies indicate CRHR1 overexpression may be associated with early life stress-induced neuroanatomical changes and dendritic spine loss, 67,68 suggesting a potential mechanism whereby early life stress interacts with CRHR1 to impact cortical structure, leading to behavioral adaptations and harmful alcohol use. We present this hypothesis cautiously due to CRHR1's location in a linkage disequilibrium block containing genes like MAPT, which may be involved in neurodegenerative diseases and cortical anatomy. 69 Our cell-type analysis also found that excitatory neurons may underlie GCT's association with alcohol use. These data support the notion that glutamatergic transmission plays an important role in alcohol misuse. 70,71 Interestingly, CRHR1 is expressed in glutamatergic, but not GABAergic, cortical neurons. 72 Activation of CRHR1 in the forebrain is associated with alteration in glutamatergic neurotransmission and increased behavioral susceptibility to stress in mice. 72 Therefore, our single-cell findings support our hypothesis associating cortical CRHR1 expression with increased stress susceptibility, cortical thinning, and alcohol misuse.
A total of 120 cortical cell types were analyzed from the Allen Brain Map Human Multiple Cortical Areas SMART-sequence data set. Cell types are organized by broad class: excitatory (56 cell types), inhibitory (54 cell types), and nonneuronal (10 cell types). Thirty distinct cell types were nominally significant (P < .05); 27 were excitatory, and 3 were inhibitory. See eTable 43 in Supplement 1 for full results and eMethods in Supplement 1 for a full explanation of cell type nomenclature. AUD indicates alcohol use disorder; LDSC, linkage disequilibrium score regression.

Limitations
This study has several limitations. First, MR instrumentation in neuropsychiatry remains challenging due to the complexity of the phenotypes and frequent uncertainty of genetic variants' biological functions. 28 For several of our alcohol use phenotypes, we used a relaxed P value threshold due to the limited number of variants at GWS, in line with previous psychiatric MR studies. 43,44 While these relaxed thresholds could introduce weak instrument bias or increase the possibility of horizontal pleiotropy, all instrument SNVs had F statistics exceeding 10, the conventional cutoff for designating strong instruments. 73 To protect our MR estimates from the influence of invalid instruments and violations of MR's core assumptions, we used sensitivity analyses (eg, Steiger directionality test, ADH1B instrument, and leave-1-out), MR estimators with relaxed assumptions (eg, weighted median and post-Lasso IVW), and MVMR accounting for possible confounding or mediating traits, which yielded largely consistent results and suggested minimal violations of MR's assumptions. 51 However, causal inference requires triangulating evidence, 74 and we emphasize that our results should be interpreted in the context of other studies investigating similar questions with different methodologies. 9,11,13,17 Furthermore, we recognize the neuroanatomical phenotypes we analyzed may not fully encapsulate brain damage and caution that our null findings do not imply that alcohol does not affect brain health.

Conclusions
The results of this study provide evidence that genetically predicted GCT was associated with alcoholic drinks consumed per week and binge drinking frequency after accounting for neuropsychiatric phenotypes, substance use, trauma, and neurodegeneration. We also found that several genes located at 17q21.31 and glutamatergic cortical neurons may be biological mechanisms associating GCT with alcohol consumption. These findings should be replicated in larger samples to better understand the interactions between brain structure and alcohol use.