Identification of Novel Loci and Cross-Disorder Pleiotropy Through Multi-Ancestry Genome-Wide Analysis of Alcohol Use Disorder in Over One Million Individuals

Alcohol use disorder (AUD) is highly heritable and burdensome worldwide. Genome-wide association studies (GWASs) can provide new evidence regarding the aetiology of AUD. We report a multi-ancestry GWASs across diverse ancestries focusing on a narrow AUD phenotype, using novel statistical tools in a total sample of 1,041,450 individuals [102,079 cases; European, 75,583; African, 20,689 (mostly African-American); Hispanic American, 3,449; East Asian, 2,254; South Asian, 104; descent]. Cross-ancestry functional analyses were performed with European and African samples. Thirty-seven genome-wide significant loci were identified, of which seven were novel for AUD and six for other alcohol phenotypes. Loci were mapped to genes enriched for brain regions relevant for AUD (striatum, hypothalamus, and prefrontal cortex) and potential drug targets (GABAergic, dopaminergic and serotonergic neurons). African-specific analysis yielded a unique pattern of immune-related gene sets. Polygenic overlap and positive genetic correlations showed extensive shared genetic architecture between AUD and both mental and general medical phenotypes, suggesting they are not only complications of alcohol use but also share genetic liability with AUD. Leveraging a cross-ancestry approach allowed identification of novel genetic loci for AUD and underscores the value of multi-ancestry genetic studies. These findings advance our understanding of AUD risk and clinically-relevant comorbidities.


Introduction
Severe alcohol use disorder (AUD) is a chronic and devastating illness characterized by maladaptive patterns of alcohol use.AUD is common, reaching up to 14% lifetime prevalence in the U.S. (1) and up to 30% in Southern and Eastern African sub-Saharan countries (2).Current treatment options for AUD show modest and variable e cacy (3).Further, AUD leads to several complications, including impaired lipid metabolism, liver and cardiovascular dysfunction, severe psychiatric comorbidity and cognitive impairment, all contributing to the high mortality (4).
AUD is a complex, heritable disorder, with twin heritability estimated at ~ 50% (5).Genome-wide association studies (GWASs) of AUD have begun to elucidate its genetic underpinnings and characterize its polygenic architecture.Most available GWAS ndings consistently suggest distinct patterns of genetic correlations for AUD with other mental phenotypes as compared to drinking frequency (6, 7).One of the challenges for genetic studies of alcohol-related traits is genetic heterogeneity (7)(8)(9), which makes it critical to map the genetic architecture of clinically-relevant phenotypes with clear de nitions.Several previous studies have combined DSM-based AUD case de nition with quantitative and/or screening measures from tools such as the AUDIT screening questionnaire.A recent study reported 110 risk loci in a GWAS of problematic alcohol use (PAU) -a relevant proxy for the genetic study of AUD (genetic correlation 85%) (10).
These broad de nitions have enabled large sample sizes to maximise power for genetic discovery, but capture heterogenous phenotypes.For example, the AUDIT has a two-factor structure -consumption and problems -that is remarkably consistent with its underlying two-factors genetic architecture (11)(12)(13).Despite the clear relevance of the problem subscore for genetic studies mixing positively-screened and AUD cases (8, 9), the AUDIT remains a screening tool designed to avoid false-negatives, so that a substantial number of false-positive cases are expected (14).Thus, substantial gaps remain in the knowledge of the the genetic underpinnings of AUD, and how this differs from previously used alcohol-related phenotypes.Yet, AUD represents the most burdensome alcohol-related clinical phenotype for patients and caregivers [AUD is, in part, de ned by the distress/burden induced by the disorder (15)].
Additionally, cross-disorder post-GWAS analyses have been made standard to explore the shared vs. unique genetic liability to several disorders and traits.This has been particularly relevant for alcohol-related traits, showing how much alcohol (excessive) consumption differs from AUD and PAU in terms of genetic correlation with other mental traits, although they tend to be highly correlated between them.Genetic correlation (r g ) has been useful in revealing the relationship between AUD and other phenotypes in EUR samples [see, e.g.(6,11,13)], with scarce recent ndings in AFR samples using recently-developed analytical methods such as POPCORN [see, e.g.(10)].However, r g remains unable to capture scenarios of a mixture of positive and negative correlations.The MiXeR method is able to characterize overlapping genetic architectures beyond r g in polygenic disorders (19).
The genes mapped to genome-wide signi cant (GWS) loci in AUD/PAU implicated alcohol metabolism (three alcohol dehydrogenase genes), response to stress (corticotropin releasing hormone and broblast growth factor genes), opioid signalling (OPRM1) and metal transport (SLC39A8).These ndings encourage further functional annotations for druggable targets [see, e.g., (20) and (10)] to prioritize subsequent preclinical and clinical research to discover new drugs for AUD.In silico functional genomic tools that link loci to genes to expression patterns across speci c tissues and cell types can improve the discovery of biological pathways involved in AUD with potential for clinical translation.
We aimed to boost discovery of genetic loci associated with AUD, leveraging novel GWAS data across multiple ancestries and relying on diagnostic criteria for de ning AUD cases.We applied novel analytical tools to the multi-ancestry and to ancestry-speci c samples to better characterize (i) the genetic architecture of AUD, (ii) its genetic overlap with clinically-relevant mental and general medical traits, disorders and risk factors -to disentangle genetic risk beyond the direct effect of alcohol consumption and (iii) to investigate molecular pathways of AUD including potential druggable targets.

GWAS samples
Alcohol Use Disorder.We extracted summary statistics with p-values and Z-scores from recent GWASs (Table 1) relevant to the AUD phenotype de ned according to the DSM-5 or the International Classi cation of Diseases (ICD) 9/10 (abuse and/or dependence).We included AUD GWASs showing global Linkage Disequilibrium Score Regression (LDSC) genetic correlation (r g ) > 0.8 with each other (21) (Supplementary Fig. 1).Based on these criteria, we selected the following GWAS results for inclusion: Million Veteran Program (MVP): ICD 9/10 alcohol abuse/dependence (AUD) -and severe acute intoxication (either one inpatient or two outpatient diagnoses, ICD9 codes 303 to 303.03) (9,22).All diagnoses were based on validated electronic health records from clinical encounters at settings a liated with the U.S. Veterans Affairs system.We analyzed MVP AUD data downloaded from the dbGaP website (accession phs001672.v9.p1).
Although we did not plan to include alcohol-related traits other than AUD, the summary statistics from the MVP GWAS were not available without intoxication cases, which only represented 0.3% of the sample (N = 226); FINNGEN [https://www.nngen./en/access_results, R6 public release (23)]: ICD-9/10 abuse/dependence (AUD) based on validated electronic health records from in-or outpatient care settings in Finland; UK Biobank (UKB): ICD 10 abuse/dependence (AUD, see above) based on validated electronic health records from inpatient and primary care settings in United Kingdom (individual level genotypes under accession number 27412); Psychiatric Genomics Consortium (PGC) [https://pgc.unc.edu/for-researchers/download-results/(7)]: DSM-IV alcohol dependence (considered equivalent to severe DSM-5 AUD) diagnosed by trained clinicians' ratings or semi-structured interviews.These data are not publicly available without the meetaanalysis with the FinnGen sample, and require request to the PGC workgroup.Combining the samples from all ancestries yielded a total multi-ancestry sample of 1,041,450 individuals, including 102,079 AUD cases (Effective sample size -Neff = 321,343 -see Supplementary Methods for calculation).Ancestry-speci c sample sizes enabled ancestry-speci c analyses for EUR and AFR samples only, as supported by visual examination of the QQ plots (Supplementary Fig. 2).
GWAS samples for comorbid disorders and traits.To investigate the genetic architecture of AUD overlapping with mental traits and disorders and with general medical conditions and risk factors which represent frequent comorbidities and complications of alcohol use and AUD (24), we performed cross-disorder analyses using GWAS data downloaded between June 15 and July 1st 2022 (Table 2 & Supplementary Methods).

Statistical analysis
The AUD GWASs included in the meta-analysis were adjusted for sex and the rst 6-10 principal components of ancestry.We applied the same procedure to perform our own AUD GWAS in the UK Biobank sample using Regenie (25), which allowed us to keep related individuals for this sample, increasing EUR sample size by 1,408 cases and 70,225 controls (~ 15%).GWAS sample size weighted meta-analysis was performed with METAL (26), with p-values < 5 x 10 − 8 considered genome-wide signi cant.LD score intercept was calculated using linkage disequilibrium score regression (LDSC) (27).To estimate statistical power and population strati cation, QQ plots were produced and genetic in ation factors lambdaGC and lambda1000 were estimated using custom scripts (Supplementary Fig. 2, Supplementary Table 1).We also investigated the concordance of ndings across ancestries by (i) sign tests assessing the similarity of effect and (ii) testing the Bonferroni-corrected signi cance of GWS associated loci from the EUR sample in the AFR sample (i.e.(p < 0.05/n EUR signi cant loci actually found in the AFR sample).

De nition of genomic loci
Genomic loci were de ned using the standard procedure applied in Functional Mapping and Annotation of Genome-Wide Association Studies (FUMA GWAS, https://fuma.ctglab.nl/).Novel loci were identi ed using a particularly conservative procedure in-house, considering genomic loci +/-1 kb compared with the GWAScatalog, a curated list of relevant publications, the MRC IEU PheWas tool, and the potential more recent GWASs published since their last updates (Supplementary Methods).We report separate novelty assessments for AUD vs. other alcohol-related phenotypes.Of note, for the current study, the novelty checking procedure encompassed GWS loci from Zhou et al.'s PAU (10), and Saunders et al.'s drinking frequency (17).
Functional annotation GWS loci were annotated with FUMA (version 1.5.0) by mapping loci to lead SNPs and lead SNPs to credibly mapped genes, de ned by at least two converging signals among positional, gene expression and chromatin interaction mapping (FUMA SNP2GENE).These genes were linked to cell types using the ad hoc FUMA function (28) within 15 available human tissue types.We included the datasets reporting the largest number of cell types available, refered to as "level 3" in FUMA (217 cell types).We also report the functional impact of candidate SNPs on protein structure, chromatin conformation and tissue-speci c gene expression using ad hoc in silico databases, as well as the presence of credibly mapped mapped genes as Drugbank hits (FUMA GENE2FUNCTION).The full parameters provided to FUMA are available as plain text at the end of the Supplementary data, listed as FUMA parameters 1, 2 and 3 for multi-ancestry, African and European samples, respectively.

Quanti cation of polygenic overlap
We applied univariate MiXeR ( 29) to estimate SNP-based heritability, discoverability (i.e. the average magnitude of additive genetic effects among traitin uencing variants), and polygenicity (i.e. the number of trait-in uencing variants expected to explain 90% of heritability) of AUD.We used bivariate MiXeR to quantify total polygenic overlap between AUD in EUR and other phenotypes of interest at the genome-wide level (30).To evaluate MiXeR reliability, we reported analyses with Akaike Information Criterion (AIC) differences > 0, as previously reported (19).Due to model t and use of a European reference genome, MiXeR could only be applied to the EUR sample.

Cross-disorder genetic correlations across European and African ancestries
Cross-ancestry genetic correlations (r g ) were estimated between AUD and relevant mental traits and disorders and general medical risk factors and disorders (described above) using Popcorn (31).
We applied Bonferonni correction for multiple testing to GWS loci and genetic correlation and False Discovery Rate (FDR, Benjamini-Hochberg) for tissue enrichment.Cell type speci city analyses included both Bonferroni (step one) and FDR (steps two and three).

GWAS meta-analysis
The meta-analysis of AUD with the multi-ancestry sample identi ed 105 genome-wide signi cant (GWS) risk variants from 37 loci (Fig. 3).We established that seven novel loci for AUD and six for other alcohol-related phenotypes (e.g.alcohol consumption, lifetime alcohol use).SNP-based heritability was 0.075 (se = 0.004, p < 1E-17) for the EUR GWAS and 0.053 (se = 0.017, p = 1.4E-3) for the AFR GWAS.Eight loci were unique to the multi-ancestry sample, seven to the EUR sample, which elicited 88 GWS risk variants from 35 loci.In the AFR sample we identi ed eight GWS risk variants from one locus represented by the lead variant rs1229987, mapped to RP11-696N14.1.This variant was in the putative regulatory region of ADH1B and in high LD with rs2066702, a functional locus (32) (Fig. 3).
Table 3 shows the location and functional signi cance of GWS lead variants for AUD for each loci, by ancestry.Credible gene mapping, based on both variant position and in silico effect on gene expression or chromatin conformation (see methods section) revealed 55 unique genes mapped to the GWS loci: 41 for multi-ancestry, 40 for EUR and one for AFR.The six loci that were novel for any alcohol-related phenotype, were mapped to ERI3, BARHL2, SRFBP1 or LOX, RP11-756H20, CNTLN.Sixteen locus boundaries were unique to one of the meta-analyses: 9/41 (22%) for the multi-ancestry sample and 7/40 (18%) for EUR sample, leaving 28 locus boundaries that overlapped across multi-ancestry and EUR samples.Nine loci out of the 73 associated with AUD in the three samples had signi cant heterogeneity in the corresponding meta-analyses (four shared by the multi-ancestry and EUR samples, one in EUR only).
Table 3 Genome-wide signi cant loci in the multi-ancestry, European (EUR), and African (AFR) samples (genome build GRCh37.p13).uniqID is chromosome:positi pairs):alternate allele:reference allele.rsID represents the SNP ID according to the reference database dbSNP.CADD, Combined Annotation Dependent Depletion variants; RDB, regulomeDB score (not all variants could be found in this database, as indicated by blank cells).POLYPHEN estimates the tolerability/deleterio exonic variants only.Credibly mapped genes are considered when designated by two out of positional/expression/chromatin interaction mapping analyses.a N AUD, b Novel loci for other alcohol-related phenotypes, C Locus unique to the sample.Ten genes were identi ed in multi-ancestry only (RP5-947P14, ERI3, SLC4 From 27 lead SNPs common to both the EUR and AFR samples, eighteen had a concordant direction of effect in both samples (sign test p-value not signi cant at 0.052, although relatively inconclusive regarding its closeness to reaching signi cance).Three from the highly signi cant ADH1B locus were replicated in the AFR sample after Bonferroni correction.All these replicated variants had concordant directions of effect compared to the EUR sample.The nine variants with discordant effects in the AFR vs. EUR sample all had p-values in the original GWAS > 0.215, making them far from any association with AUD.
Functional analyses: cells, tissues and gene sets Among 217 cell types tested, ten were signifcantly enriched for genes mapped to lead SNPs in the multi-ancestry sample vs. ve in the EUR and one in the AFR sample.(Fig. 1).These cell types mainly included cortical GABAergic neurons, but the multi-ancestry sample also elicited serotonergic and dopaminergic neurons from the adult midbrain and excitatory and inhibitory neurons from the prefrontal cortex.The AFR sample elicited dopaminergic hippocampus neurons.All these cell types remained independently associated with AUD in the multi-ancestry vs. two in EUR (both developmental GABAergic neurons) and none in AFR (see Supplementary Table 2).These ndings were consistent with tissue enrichment analysis (Fig. 2), where signi cantly enriched tissues were mostly brain-related (six in the multi-ancestry, three in EUR sample), including the cortex, hippocampus and nucleus accumbens.No tissue was signi cantly enriched in the AFR sample.
Figure 1.Independent cell types associated with the GWAS meta-analysis results in the African (AFR), European (EUR) and multi-ancestry (MA) samples.
Results from FUMA step 3 analysis obtained with 217 Human cell types.woFetal, dataset considered without developing cells; GW, gestation week; PFC, prefrontal cortex; exCA1, hippocampal cornu ammonis excitatory neurons; The complete datasets description is available at https://fuma.ctglab.nl/tutorial#celltype.Using MAGMA, there were four signi cant gene sets in the multi-ancestry analysis, three in EUR, and 14 in AFR (Table 3).All samples were enriched for the alcohol dehydrogenase activity geneset.Compared to EUR, the multi-ancestry sample yielded additional 'response to alkaloids' and 'maintenance_of_presynaptic_active_zone_structure' gene sets.Interestingly, the AFR analysis elicited nine gene sets related to immunity and in ammation, two to cancer risk, and one to cell aging.Investigating gene enrichment patterns for Drugbank associations (Supplementary table 3A-C) across ancestries, we found 195 unique Drugbank hits in the multi-ancestry analysis, 167 in the EUR analysis and 26 in the AFR analysis.All associations in the AFR and EUR analyses were also found in the multiancestry analysis.Twenty-eight ndings were unique to the multi-ancestry analysis, including loci related to CDK5R1 (Cyclin-dependent kinase 5 activator 1 gene), POR (cytochrome p450 oxidoreductase) and DRD2 (Dopamine receptor D2) (Supplementary Table 3).Drugbank hits were similar across EUR and AFR ancestries, except for Blood and blood forming organs (Fisher extact test p = 0.037) (Supplementary Table 4).

Genetic overlap (MiXeR)
We obtained reliable MiXeR estimates of polygenicity for age at smoking initiation, drinks / week, neuroticism, CUD, OUD, ADHD, bipolar disorder, major depression, and schizophrenia (mental traits and disorders) and systolic blood pressure (general medical condition) (Fig. 4).Univariate MiXeR showed that AUD was moderately polygenic (7.8-7.9k'causal' variants), and discoverability was 0.0025 (SD = 0.0002).Bivariate MiXeR showed large overlap between AUD and other substance use phenotypes, sharing 62% of its 'causal' variants with CUD, 76% with drinks/week, 95% with OUD and 74% with age at smoking initiation (Fig. 4A).Overlap was substantial, albeit smaller between AUD and non-substance related mental disorders: AUD shared more than half its loci with ADHD (52%), bipolar disorder (53%), major depression (60%) and schizophrenia (62%, Fig. 4B), and mental traits, with a particularly large genetic overlap with neuroticism (96%, Fig. 4C).There was also a large overlap between AUD and systolic blood pressure (37%, Fig. 4C).Genetic correlation (r g ) The genetic correlation of AUD was moderate between EUR and AFR ancestries (r g =0.65, FDR = 9.3 x 10 − 7 ).Genetic correlation patterns were overall similar in signi cance and magnitude across mental traits and disorders (Fig. 5).The strongest correlations were found for ADHD (EUR/AFR r g =0.47/0.30),age at smoking initiation (0.54/0.37) and OUD (0.85/0.81); amongst which the highest p corrected was 0.047 for ADHD in the AFR sample).The r g between AUD and CUD, major depression, schizophrenia, bipolar disorder and PTSD were only signi cant in the EUR sample (highest p corrected =0.0054 for PTSD).Consistent patterns were also found for cognitive traits (r g =-0.23/-0.4)and educational attainment (-0.32/-0.40);highest p corrected =3.3 x 10 − 4 for education in the AFR sample.Signi cant genetic correlation was also observed between AUD and heart failure (r g =-0.22,p corrected =7x10 − 6 ), liver age (r g =0.16, p corrected =0.011) and abdominal age (r g =-0.17, p corrected =5.4x10 − 3 ) in the EUR samples.The complete r g results are presented in Supplementary Table 5.

Discussion
This multi-ancestry meta-analysis identi ed 105 genome-wide signi cant AUD risk variants from 37 independent genomic loci, including seven novel loci for AUD and six novel loci for other alcohol-related traits.Compared to the EUR sample, multi-ancestry meta-analysis resulted in a slight increase in loci discoverability, but a stronger increase in biological diversity, as advocated by, e.g., Saunders et al. (17).AUD loci implicated genes enriched in key brain regions and cells.We con rmed the strong association between ADH1B SNPs and AUD in both EUR and AFR samples.We con rmed extensive shared genetic liability between AUD and other substance use phenotypes, mental traits and disorders (especially ADHD and neuroticism) and other medical conditions, consistent with and extending previous work with AUD and PAU phenotypes (6, 7, 9, 13).
To the best of our knowledge, the current study represents the largest GWAS meta-analysis of DSM/ICD-de ned AUD, with 1,041,450 participants; identifying ~ 50% more loci than recently published AUD GWASs (6, 9).A multi-ancestry meta-analysis of PAU GWAS data (AUD + AUDIT-problem subscale; N = 1,079,947) identi ed a total of 110 genetic risk variants across diverse ancestries (10), without providing the speci c AUD results.Our study identi es almost as many risk variants, yet focusing on a narrow AUD phenotype.As with the current study, Zhou et al. (10) also provided evidence for the role of the dopamine receptor type 2 gene (DRD2), in line with previous studies (6, 33).
Several genes mapped to our novel AUD loci had previously been associated with alcohol-related phenotypes (e.g., drinking quantity or frequency, lifetime alcohol use).This suggests some molecular mechanisms are shared between such alcohol-related phenotypes and AUD.It is noticeable that about half of the AUD loci were related to regulatory regions and not to altered protein structure or function.This supports the role of regulatory mechanisms in the pathophysiology of AUD, as in many other complex phenotypes (34).The single AFR GWS locus was mapped to a long ncRNA, but it likely re ects the strong LD with the functional ADH1B variant rs2066702 (0.78-0.98 across 1000 genomes AFR populations).
A key nding in the current study was that ancestral diversity improved the functional mapping, and allowed the discovery of cell types, tissue types and gene sets with potential relevance to the neurobiology of AUD.This was more than expected from the modest increase in the number of GWS loci in the multiancestry vs. EUR samples.Since most of the cell types enriched in the multi-ancestry meta-analysis also represent the EUR sample top signal (still not signi cant), we believe the multi-ancestry meta-analysis shows an actual power increase.First, functional analyses doubled the number and diversity of signi cantly enriched cell types in multi-ancestry vs. EUR samples.These cell types included GABAergic, serotoninergic (Sert + ) and dopaminergic (labelled DA1 in EUR) neurons from the midbrain, dopaminergic neurons (DA0 in AFR) from the hippocampus, and prefrontal neurons.This is consistent with previous evidence supporting the involvement of this brain region and these cell types in AUD, particularly GABAergic neurons in the midbrain in a previous GWAS of maximum alcohol consumption (18) and -more generally -in emotion processing (35) and the action of benzodiazepine medication against alcohol withdrawal syndrome.Secondly, eight tissues were associated with GWS AUD loci in multi-ancestry meta-analysis, 38% more than the EUR sample (N = 5), and mostly included striatal brain regions.Interestingly, the multi-ancestry meta-analysis revealed pathways of relevance to the neuroimaging ndings in AUD (36) and to our cell type analysis, including the substantia nigra, frontal cortex and nucleus accumbens.The association with the hypothalamus, a region that regulates liquid intake, could be related to the consumption component of AUD, in line with our previous ndings associating hypothalamus with both alcohol consumption and AUD loci (33) and with experimental data regarding altered hypothalamic-pituitary-adrenal axis after chronic alcohol exposure (37).The current ndings further illustrate the complementarity of the tissue-level and cell-level bioinformatic approaches to open therapeutic avenues in AUD (38, 39).
However, actual experimental biological work is needed to con rm the mechanistic understanding, as in, e.g., (40).
The alcohol dehydrogenase genes were enriched in all samples, but the AFR sample showed additional enrichment in several immunity/in ammation-related pathways (14 sets vs. four in multi-ancestry and two in EUR), in line with a recent meta-analysis showing associations between cytokine levels and AUD (41).
Although such a difference in the number of gene sets in AFR compared to EUR samples may represent a degree of noise in the analysis, MAGMA has shown high detection power with little type I error in ation (42).
The current sample size was large enough to apply MiXeR to a wider range of traits and disorders than in our previous work on AUD (33).We provide novel evidence that substance use phenotypes are highly polygenic, and estimated the polygenicity of AUD to 7.2k − 8.5k causal variants.We recently showed that other mental disorders have similar polygenicities, ranging between 5.6k (ADHD) and 14.5k (major depression) causal variants (19).This high polygenicity could partly explain the high level of comorbidity in AUD given the extensive genetic overlap between AUD and other substance use and mental disorders (notably ADHD, bipolar disorder, major depression and schizophrenia; Fig. 5).The particularly large overlap between AUD and neuroticism (92%) -even in the absence of signi cant genetic correlation -further supports the hypothesis that the shared genetic component of mental disorders also includes AUD (19) and partly relies on the shared liability to neuroticism.Overlapping variants seem to exert bidirectional effects, both increasing or decreasing the risk across phenotypes.More GWAS data, especially from non-EUR samples, are needed to reliably estimate the overlap between AUD and general medical conditions.
Still, we report substantial shared polygenicity between AUD and systolic blood pressure (36%) in the absence of signi cant genetic correlation.
There were mostly consistent patterns of genetic correlations across ancestries, especially regarding AUD and mental traits and disorders, especially OUDsigni cantly extending recent work in the AFR population (10).Of particular interest were the correlations between AUD and conditions that are usually attributed to the toxic effects of alcohol (e.g., MRI-predicted abdominal age and heart failure).This suggests a more mixed picture of highly probable toxic alcohol effects and shared molecular underpinnings.We plan to compare these correlation patterns by using individual-level genetic and phenotypic data regarding alcohol consumption vs. AUD in the near future.
Our study has limitations.There was insu cient statistical power in several cross-disorder analyses involving AUD, calling for urgent action to gather more genetic data in non-EUR populations.The effect direction across EUR and AFR ancestries was concordant in 2/3 of GWS loci, which may limit the reliability of some aspects of the multi-ancestry meta-analysis.However, this particular analysis had strong statistical power.Some in silico data used for downstream analyses are sourced from EUR samples only.Additionally, although the reference SNP and LD maps from the 1000 genomes project are sourced from relatively diverse non-EUR samples, AUD cases of AFR ancestry were recruited in Western countries (US, notably) and no data was available in terms of genetic heterogeneity for these particular samples.Same applies to FUMA analyses, for which the detail regarding LD structure is higher for EUR than for other ancestries.Overall, caution is thus advised when interpreting ndings for functional or cross-ancestry analyses.Finally, the almost exclusive use of summary statistics, preventing us from performing subgroup analyses that could help to identify clinical subgroups and control for potential mediating factors.
However, this remains the only way to leverage very large GWAS samples to date, especially for case-control analyses.
The current study leveraged multi-ancestry samples to discover several novel AUD risk loci and improve the biological diversity of associated molecular pathways, cell types and brain regions implicated in AUD.

Figure 2 :
Figure 2: Tissue-speci c gene expression enrichment from the multi-ancestry (top panel).African (AFR, middle panel) and European (EUR, bottom panel) analyses.Signi cant enrichment is represented in pink (p < 0.05 after False discovery rate correction).

DisclosuresDr.
Dale is a founder of and holds equity in Cortechs.aiand serves on its scienti c advisory board; he is a member of the scienti c advisory boards of HealthLytix and the Mohn Medical Imaging and Visualization Center (Bergen, Norway); and he receives funding through a research agreement between General Electric Healthcare and UCSD.Prof. Andreassen has received speaking honoraria from Lundbeck and has served as a consultant for HealthLytix.Dr. Kranzler has served on scienti c advisory boards for Dicerna and Sophrosyne Pharmaceuticals, as a consultant for Sobrera Pharmaceuti-cals, and as a member of the American Society of Clinical Psychophar-macology's Alcohol Clinical Trials Initiative, which during the past 3 years was supported by AbbVie, Alkermes, Amygdala Neurosciences, Arbor Pharmaceuticals, Dicerna, Eli Lilly, Ethypharm, Indivior, Lundbeck, Otsuka, and P zer, and he is named as an inventor on PCT patent appli-cation 15/878,640, "Genotype-guided dosing of opioid agonists."The other authors report no nancial relationships with commercial interests.

Figures
Figures

Figure 1 Independent
Figure 1Independent cell types associated with the GWAS meta-analysis results in the African (AFR), European (EUR) and multi-ancestry (MA) samples.Results from FUMA step 3 analysis obtained with 217 Human cell types.woFetal, dataset considered without developing cells; GW, gestation week; PFC, prefrontal cortex; exCA1, hippocampal cornu ammonis excitatory neurons; The complete datasets description is available at https://fuma.ctglab.nl/tutorial#celltype.

Figure 2 Tissue
Figure 2 Tissue-speci c gene expression enrichment from the multi-ancestry (top panel).African (AFR, middle panel) and European (EUR, bottom panel) analyses.Signi cant enrichment is represented in pink.(p <0.05 after False discovery rate correction).

Figure 3 Ancestry
Figure 3 Ancestry Speci c Genetic Architecture of AUD in the multi-ancestry analysis (top, red, AUD multi-ancestry) and for the European (top, blue, AUD EUR) and African (bottom, yellow, AUD AFR) samples.-log10 (p-values) obtained by meta-analysis (METAL) are shown on the y-axis while the x-axis represents increasing chromosome numbers from 1 to 22 and positions in K-base pairs.Y-axis is truncated to -log10(P) = 32.

Table 1
GWAS data, including ancestry breakdown.EUR, European; AFR, African; HA, Hispanic American; EAS, East Asian; SAS, South Asian; AUD, alcohol Use Disorder; ICD, International Classi cation of Diseases; DSM, Diagnostic and Statistical Manual of Mental Disorders; GWAS, Genome-wide association studies; MVP, Million Veteran Program; PGC, Psychiatric Genomic Consortium; SNP, Single nucleotide polymorphism.Neff, effective sample size.*Individuals suffering from AUD are expected to be more often males; median age is less relevant for the FINNGEN cohort compared to the others since all individuals are followed-up from birth.**Pooled mean age was obtained using the cohorts mean ages, except for the PGC sample, where it was extrapolated from the mean age of the Pale-Yenn and SAGE cohorts, which represent ~ 30% of the total PGC sample -yielding pooled mean age = 34.7 based on PMID 27028160.

Table 4
Signi cant gene sets associated with AUD loci according to MAGMA analysis in the GWASs with multi-ancestry, European (EUR) and African (AFR) samples.FDR, false discovery rate after Bejamini-Hochberg correction.