Novel alcohol-related genes suggest shared genetic mechanisms with neuropsychiatric disorders

Abstract

Excessive alcohol consumption is one of the main causes of death and disability worldwide. Alcohol consumption is a heritable complex trait. We conducted a meta-analysis of genome-wide association studies (GWAS) of gram/day (g/d) alcohol consumption in UK-Biobank, AlcGen and CHARGE+ consortia accumulating 480,842 people of European descent to decipher the genetic architecture of alcohol intake. We identified 46 novel, common loci, and investigated their potential functional significance using magnetic resonance imaging data and gene expression studies. Our results identify genetic pathways associated with alcohol consumption and suggest shared genetic mechanisms with neuropsychiatric disorders including schizophrenia.

Excessive alcohol consumption is a major public health problem that is responsible for 2.2% and 6.8% age-standardized deaths for women and men respectively¹. Most genetic studies of alcohol use focus on alcohol dependency, although the population burden of alcohol-related disease mainly reflects a broader range of alcohol consumption behaviors². Small reductions in alcohol consumption could have major public health benefits; even moderate amounts of alcohol/day may have significant impact on mortality³.

Alcohol consumption is a heritable complex trait⁴, but genetic studies to date have robustly identified only a small number of associated genetic variants ^5–8. These include variants in the aldehyde dehydrogenase (ADH) gene family, a group of enzymes that catalyze the oxidation of aldehydes⁹, including a cluster of genes on chromosome 4q23 (ADH1B, ADH1C, ADH5, ADH6, ADH7)⁶.

Here, we report a GWAS meta-analysis of alcohol intake (log transformed g/day) among people of European ancestry drawn from UK Biobank (UKB)¹⁰, the Alcohol Genome-Wide Consortium (AlcGen) and the Cohorts for Heart and Aging Research in Genomic Epidemiology Plus (CHARGE+) consortia. Briefly, UKB is a prospective cohort study of ~500,000 individuals recruited between the ages of 40 and 69 years. Participants were asked to report their average weekly and monthly alcohol consumption through a self-completed touchscreen questionnaire¹⁰. Based on these reports, we calculated the g/d alcohol intake (Methods). Participants were genotyped using a customized array with imputation from the Haplotype Reference Consortium (HRC) panel¹¹, yielding ~7 million common single nucleotide polymorphisms (SNPs) with minor allele frequency (MAF) ≥ 1% and imputation quality score [INFO] ≥ 0.1. After quality control (QC) and exclusions (Methods) we performed GWAS of alcohol consumption using data from 404,731 UKB participants of European descent under an additive genetic model (Methods and Supplementary Table 1). We found that genomic inflation in the UKB analysis was λ_GC=1.45, but did not adjust for inflation as the LD score regression intercept was 1.05, indicating that this was due to polygenicity rather than to population stratification¹². The estimated SNP-wide heritability of alcohol consumption in the UKB data was 0.09.

We also carried out GWAS in 25 independent studies from the AlcGen and CHARGE+ consortia including 76,111 participants of European descent for which alcohol g/d could be calculated (Supplementary Table 2). Various arrays were used for genotyping, with imputations performed using either the 1,000 Genomes Reference Panel or the HRC platforms (Supplementary Table 3). After QC, we applied genomic control at the individual study level and obtained summary results for ~7 million SNPs with imputation quality score ≥ 0.3 (Methods).

We combined the UKB, AlcGen and CHARGE+ results using a fixed effects inverse variance weighted approach for a total of 480,842 individuals¹³. To maximize power, we performed a single-stage analysis to test common SNPs with MAF ≥ 1%. We set a stringent P-value threshold of P < 5 × 10⁻⁹ to denote significance in the combined meta-analysis¹⁴, and required signals to be at P < 5 × 10⁻⁷ in UKB, with same direction of effect in UKB and AlcGen plus CHARGE+, to minimize false positive findings. We excluded SNPs within 500kb of variants reported as genome-wide significant in previous GWAS of alcohol consumption^5,6, identified novel loci by requiring SNPs to be independent of each other (LD r² < 0.1), and selected the sentinel SNP within each locus according to lowest P-value (Methods).

We then tested for correlations of alcohol-associated SNPs with Magnetic Resonance Imaging (MRI) phenotypes of brain, heart and liver, and gene expression. We tested the sentinel SNPs for association with other traits/diseases and Drosophila mutant models were used to investigate functional effects on ethanol-induced behavior.

RESULTS

Our meta-analysis identified 46 novel loci associated with alcohol consumption (log transformed g/day) (Fig. 1 and Table 1). All inferential statistics for the novel loci are reported in Table 1 whereas heterogeneity metrics are presented in Supplementary Table 4. In addition, we discovered a further eight variants in the combined analysis at nominal genome-wide significance (P < 1 × 10⁻⁸) that may also be associated with alcohol intake (Supplementary Table 5). The most significantly associated variant, rs1991556 (P = 4.5 × 10⁻²³), is an intronic variant in MAPT gene that encodes the microtubule-associated protein tau, and was found through Phenoscanner not only to be associated with dementia¹⁵ and Parkinson’s disease^16,17, but also with neuroticism, schizophrenia¹⁸and other traits^19–21 (Methods, Fig. 2 and Supplementary Table 6). The second most significantly associated variant is rs1004787 (P = 6.7 × 10⁻¹⁷), near SIX3 gene, which encodes a member of the sine oculis homeobox transcription factor family involved in eye development²². The third SNP is rs13107325 (P = 1.3 × 10⁻¹⁵), a missense SNP in SLC39A8 (https://www.ncbi.nlm.nih.gov/gene/64116), a gene that encodes a member of the SLC39 family of metal ion transporters, which has been associated with schizophrenia²³ as well as inflammatory bowel disease, cardiovascular and metabolic phenotypes ^24,25–27 in previous GWAS (Fig. 2 and Supplementary Table 6).

Figure 1. Manhattan plot showing P-values from discovery genome-wide association meta-analysis with alcohol intake (log g/d) among 480,842 individuals across UK Biobank, AlcGen and CHARGE+, excluding known variants.

The P-value was computed using inverse variance fixed effects models. The y axis shows the –log₁₀ P values and the x axis shows their chromosomal positions. Horizontal blue line represents the threshold of P = 5 × 10⁻⁹.

Table 1:

Association results of 46 novel alcohol variants identified through the meta-analysis of UK Biobank and AlcGen and CHARGE+. Results are ordered by P-value of combined analysis.

leadSNP						Combined			UKB			AlcGen and CHARGE+
Nearest_Gene	Annotated Gene	rsID_LEAD_SNP	CP	EA	EAF	BETA	SE	P	BETA	SE	P	BETA	SE	P
MAPT	STH	rs1991556	17:44083402	A	0.22	−0.012	0.001	4.5E-23	−0.013	0.001	2.4E-21	−0.011	0.004	4.0E-03
RP11–89K21.1	SIX3	rs1004787	2:45159091	A	0.54	0.009	0.001	6.7E-17	0.009	0.001	1.1E-15	0.007	0.003	1.4E-02
SLC39A8	SLC39A8	rs13107325	4:103188709	T	0.07	−0.016	0.002	1.3E-15	−0.017	0.002	4.8E-16	−0.006	0.006	3.6E-01
IZUMO1, RASIP1, FUT1	IZUMO1	rs838145	19:49248730	A	0.55	−0.008	0.001	3.2E-15	−0.009	0.001	2.4E-15	−0.004	0.003	1.7E-01
Na	PSMD7	rs1104608	16:73912588	C	0.43	−0.008	0.001	1.2E-14	−0.009	0.001	4.9E-15	−0.003	0.003	2.5E-01
MYBPC3	MYBPC3	rs2071305	11:47370957	A	0.69	0.009	0.001	4.5E-14	0.009	0.001	3.9E-13	0.007	0.003	3.1E-02
Na	DRD2	rs7121986	11:113355444	T	0.37	−0.008	0.001	6.2E-14	−0.008	0.001	1.3E-13	−0.005	0.003	1.1E-01
Na	DPP6	rs6969458	7:153489725	A	0.47	0.008	0.001	6.4E-14	0.008	0.001	1.3E-12	0.007	0.003	1.5E-02
RP11–308N19.1	ZNF462	rs74424378	9:109331094	T	0.76	0.009	0.001	1.7E-13	0.009	0.001	4.5E-13	0.006	0.003	8.4E-02
ARHGAP15, AC096558.1, RP11–570L15.2	ARHGAP15	rs13024996	2:144225215	A	0.37	−0.008	0.001	4.4E-13	−0.008	0.001	6.6E-13	−0.004	0.003	1.4E-01
MLXIPL	MLXIPL	rs34060476	7:73037956	A	0.87	−0.011	0.002	5.0E-13	−0.012	0.002	1.4E-13	−0.004	0.004	4.1E-01
Na	FAM178A	rs61873510	10:102626510	T	0.33	−0.008	0.001	5.1E-13	−0.008	0.001	9.8E-12	−0.008	0.003	1.7E-02
FTO	FTO	rs1421085	16:53800954	T	0.60	0.008	0.001	9.2E-13	0.007	0.001	1.7E-10	0.010	0.003	9.2E-04
Na	PMFBP1	rs11648570	16:72356964	T	0.89	−0.012	0.002	2.1E-12	−0.011	0.002	1.5E-10	−0.013	0.005	3.4E-03
OTX2, RP11–1085N6.6	OTX2	rs2277499	14:57271127	T	0.34	−0.008	0.001	2.2E-12	−0.007	0.001	2.4E-09	−0.012	0.003	9.1E-05
PDE4B	PDE4B	rs2310752	1:66392405	A	0.43	−0.007	0.001	2.8E-12	−0.008	0.001	1.8E-11	−0.006	0.003	4.2E-02
SERPINA1	SERPINA1	rs112635299	14:94838142	T	0.02	−0.025	0.004	3.7E-12	−0.027	0.004	9.8E-12	−0.017	0.010	9.9E-02
Na	AJAP1	rs780569	1:4569436	A	0.71	−0.008	0.001	5.2E-12	−0.008	0.001	1.1E-11	−0.005	0.003	1.2E-01
Na	VRK2	rs10496076	2:57942987	T	0.37	−0.007	0.001	9.7E-12	−0.007	0.001	1.3E-09	−0.009	0.003	1.6E-03
ACTR10, C14orf37	ACTR10	rs71414193	14:58685301	A	0.19	−0.009	0.001	1.8E-11	−0.008	0.001	5.8E-09	−0.013	0.004	4.5E-04
BEND4	BEND4	rs16854020	4:42117559	A	0.13	0.010	0.002	2.9E-11	0.010	0.002	5.8E-09	0.016	0.005	6.4E-04
Na	SORL1	rs485425	11:121544984	C	0.45	−0.007	0.001	6.1E-11	−0.007	0.001	7.3E-11	−0.004	0.003	1.9E-01
SEZ6L2	SEZ6L2	rs113443718	16:29892184	A	0.31	−0.007	0.001	7.4E-11	−0.008	0.001	4.5E-11	−0.003	0.003	2.9E-01
CBX5, RP11–968A15.2	CBX5	rs57281063	12:54660427	A	0.41	0.007	0.001	7.9E-11	0.007	0.001	1.8E-09	0.007	0.003	1.2E-02
Na	TNRC6A	rs72768626	16:24693048	A	0.94	0.014	0.002	9.7E-11	0.015	0.002	1.7E-09	0.014	0.006	1.8E-02
SYT14	SYT14	rs227179	1:210216731	A	0.59	−0.007	0.001	1.1E-10	−0.007	0.001	1.4E-09	−0.006	0.003	2.8E-02
TCF4	TCF4	rs9320010	18:53053897	A	0.60	0.007	0.001	1.1E-10	0.007	0.001	1.6E-09	0.007	0.003	2.2E-02
SBK1	NPIPB6	rs2726034	16:28336882	T	0.68	0.007	0.001	1.4E-10	0.007	0.001	1.1E-09	0.006	0.003	4.7E-02
ANKRD36	ANKRD36	rs13390019	2:97797680	T	0.87	0.010	0.002	1.6E-10	0.011	0.002	7.0E-11	0.004	0.005	4.5E-01
Na	ELAVL4	rs7517344	1:50711961	A	0.17	0.009	0.001	1.9E-10	0.008	0.001	2.5E-07	0.016	0.004	2.1E-05
LINC00461	MEF2C	rs4916723	5:87854395	A	0.58	0.007	0.001	2.1E-10	0.007	0.001	5.1E-10	0.005	0.003	1.1E-01
ARPC1B, ARPC1A	ARPC1B	rs10249167	7:98980879	A	0.87	0.010	0.002	2.9E-10	0.009	0.002	8.1E-08	0.015	0.004	3.8E-04
EFNB3, WRAP53	EFNB3	rs7640	17:7606722	C	0.80	0.008	0.001	4.3E-10	0.009	0.001	1.3E-09	0.006	0.004	9.9E-02
RP11–501C14.5	IGF2BP1	rs4794015	17:47067826	A	0.41	0.007	0.001	4.3E-10	0.006	0.001	5.4E-08	0.009	0.003	1.2E-03
TCAP, PNMT, STARD3	TCAP	rs1053651	17:37822311	A	0.27	−0.007	0.001	1.1E-09	−0.008	0.001	8.4E-10	−0.003	0.003	2.8E-01
Na	AADAT	rs7698119	4:171070910	A	0.49	−0.006	0.001	1.3E-09	−0.006	0.001	1.6E-07	−0.009	0.003	1.6E-03
STAT6, AC023237.1	STAT6	rs12312693	12:57511734	T	0.55	−0.006	0.001	1.5E-09	−0.006	0.001	9.5E-09	−0.005	0.003	5.6E-02
SCN8A	SCN8A	rs7958704	12:51984349	T	0.41	−0.006	0.001	1.6E-09	−0.006	0.001	1.7E-08	−0.006	0.003	3.5E-02
ACSS3	ACSS3	rs11114787	12:81595700	T	0.27	0.007	0.001	2.0E-09	0.007	0.001	2.7E-08	0.007	0.003	2.4E-02
RP11–32K4.1	BHLHE22	rs2356369	8:64956882	T	0.52	−0.006	0.001	2.0E-09	−0.006	0.001	4.1E-08	−0.007	0.003	1.6E-02
ZRANB2-AS2	ZRANB2	rs12031875	1:71585097	A	0.82	−0.008	0.001	2.2E-09	−0.008	0.001	7.6E-08	−0.010	0.004	8.7E-03
MSANTD1, HTT	MSANTD1	rs12646808	4:3249828	T	0.66	0.007	0.001	2.4E-09	0.007	0.001	1.1E-09	0.002	0.003	4.7E-01
TENM2	TENM2	rs10078588	5:166816176	A	0.52	0.006	0.001	2.5E-09	0.006	0.001	4.3E-08	0.007	0.003	1.9E-02
IGSF9B	IGSF9B	rs748919	11:133783232	T	0.79	0.008	0.001	3.3E-09	0.008	0.001	1.0E-08	0.005	0.003	1.1E-01
AC010967.2	GPR75-ASB3	rs785293	2:53023304	A	0.57	−0.006	0.001	3.3E-09	−0.006	0.001	3.2E-08	−0.006	0.003	3.8E-02
BDNF, RP11–587D21.4	BDNF	rs988748	11:27724745	C	0.21	−0.008	0.001	4.4E-09	−0.007	0.001	1.2E-07	−0.010	0.004	8.3E-03

SNP: Single Nucleotide polymorphism; LocusName: Nearest Gene; rsID_LEAD_SNP: Rs ID number of the lead SNP; CP: Chromosome/Position (build hg19/37); EA: Effect allele of the discovered SNP; EAF: Frequency of the effect allele; BETA_comb: Effect size in meta-analysis; SE_comb; Standard Error of the effect in meta-analysis; P_comb: Meta-analysis P-value; BETA_UKB: Effect size in UK Biobank analysis; SE_UKB: Standard Error of the effect in the UK Biobank analysis; P_UKB: UK Biobank analysis P-value;BETA_AlcGenCHARGE+: Effect size in the AlcGen meta-analysis; SE_AlcGenCHARGE+: Standard Error of the effect in the AlcGen meta-analysis; P_AlcGenCHARGE+: AlcGen meta-analysis P-value

Figure 2. Association of alcohol intake loci with other traits.

Plot shows results from associations with other traits which were extracted from the PhenoScanner database for the 46 novel sentinel SNPs including proxies in Linkage Disequilibrium (r² ≥ 0.8) with genome-wide significant associations. Each colored line connects a specific variant with the associated traits and diseases.

Another of our most significant variants, an intronic SNP rs7121986 (P = 6.2 × 10⁻¹⁴) in DRD2 (https://www.ncbi.nlm.nih.gov/gene/1813), encodes the dopamine receptor D2 that has been associated with cocaine addiction, neuroticism and schizophrenia¹⁸. We also found significant associations with SNP rs988748 (P = 4.4 × 10⁻⁹) in the BDNF gene (https://www.ncbi.nlm.nih.gov/gene/627, that encodes a member of the nerve growth factor family of proteins and rs7517344, which is near ELAVL4 (https://www.ncbi.nlm.nih.gov/gene/1996) (P = 2.0 × 10⁻¹⁰), the gene product of which is involved in BDNF regulation²⁸. Previous studies have suggested that a variant in BDNF is associated with alcohol consumption and that alcohol consumption modulates BDNF expression²⁹.

Additionally, we found association of alcohol consumption with SNP rs838145 (P = 3.2 × 10⁻¹⁵), which has been associated with macronutrient intake in a previous GWAS³⁰. This variant is nearest IZUMO (https://www.ncbi.nlm.nih.gov/gene/284359) in a locus of around 50kb that spans a number of genes including FGF21 (https://www.ncbi.nlm.nih.gov/gene/26291), whose gene product FGF21 is a liver hormone involved in the regulation of alcohol preference, glucose and lipid metabolism³¹. We previously reported significant association of alcohol intake with SNP rs11940694 in KLB (https://www.ncbi.nlm.nih.gov/gene/152831), an obligate receptor of FGF21 in the brain⁵, and we strongly replicated that finding here (P = 3.3 × 10⁻⁶⁸).

As well as variants in KLB and in the alcohol dehydrogenase locus (smallest P = 1.2 × 10⁻¹²⁵), we found support (P = 1 × 10⁻⁵) for association of common variants in the three other alcohol intake-related loci previously reported in GWAS (Supplementary Table 7), including SNP rs6943555 in AUTS2 (https://www.ncbi.nlm.nih.gov/gene/26053) (P = 2.9 × 10⁻⁶). In addition, we found a novel alcohol intake-related SNP rs1421085 in FTO (https://www.ncbi.nlm.nih.gov/gene/79068) in high LD (r² = 0.92) with a variant reported previously as genome-wide significant for association with alcohol dependence³².

Conditional analysis using Genome-wide Complex Trait Analysis (GCTA) did not reveal any independent secondary signals related to alcohol consumption. Among ~14,000 individuals in the independent Airwave cohort³³ (Methods), 7% of the variance in alcohol consumption was explained by the novel and known common variants. Using weights from our analysis, we constructed an unbiased weighted genetic risk score (GRS) in Airwave (Methods) and found a strong association of the novel and known variants on alcohol consumption levels (P = 2.75 × 10⁻¹⁴), with mean difference in sex-adjusted alcohol intake of 2.6 g/d comparing the top vs the bottom quintile of the GRS (Supplementary Table 8).

Associations with MRI imaging phenotypes

We functionally characterized novel variants by carrying out single-SNP analyses of the imaging phenotypes in UKB (Methods), focusing on brain (N=9,702), heart (N=10,706) and liver (N=8,479).

With Bonferroni correction (corrected P-value 6.6 × 10⁻⁶, corresponding to 0.05/46 SNPs*164 imaging phenotypes), we found significant positive associations between SNP rs13107325 in SLC39A8 and the volumes of multiple brain regions; All inferential statistics for these associations are reported in Supplementary Table 9. The strongest associations were with putamen (left: P = 2.5 × 10⁻⁴⁵, right: P = 2.8 × 10⁻⁴⁷), ventral striatum (left: P = 9.5 × 10⁻⁵³, right: P = 9.6 × 10⁻⁵¹) and cerebellum (strongest association for left I-IV volume; P = 1.2 × 10⁻⁹) (Supplementary Table 9); similar findings were recently reported in a GWAS on brain imaging in UKB³⁴. The other significant association was for rs1991556 with the parahippocampal gyrus (P = 1.2 × 10⁻⁶).

We then tested these brain regions for association with alcohol consumption and found a significant effect for the left (t₈₆₀₁ = −3.7; beta ± SE = −0.0019 ± 0.0005; P = 2.0 × 10⁻⁴) and right (t₈₆₀₁ = −3.65; beta ± SE = −0.0070 ± 0.0005; P = 2.6 × 10⁻⁴) putamen. Finally, we used data from N= 8,610 individuals and performed a mediation analysis using a standard three-variable path model, bootstrapping 10,000 times to calculate the significance of the mediation effect of putamen volume for genetic influences on alcohol consumption (Methods). We found evidence that the effect of SNP rs13107325 in SLC39A8 on alcohol intake is partially mediated via its association with left (t₈₆₀₁ = −3.03; beta ± SE = −0.27 ± 0.09; P = 1.9 × 10⁻³) and right (t₈₆₀₁ = −2.82; beta ± SE = −0.27 ± 0.09; P = 1.7 × 10⁻³) putamen volume (Fig. 3 and Supplementary Table 10). To exclude the possibility of an inverse causal pathway we performed additional analyses in UKB non-drinkers (N =589). With 10,000 random permutations, associations of rs13107325 with both left and right putamen remained significant (left putamen: t₅₄₁=1.06; P = 0.02; right putamen: t₅₄₁=0.38; P = 0.04) indicating that the association between rs13107325 and putamen regions is not mediated by alcohol intake.

Figure 3. Mediation effect of the grey matter volume of bilateral putamen on the relationship between SNP rs13107325 and alcohol intake.

The green is for left putamen, and, the red is for the right one. We use ‘a’ for the relationship between rs13107325 and putamen, ‘b’ for the relationship between putamen and alcohol consumption, ‘c’ for the relationship between rs13107325 and alcohol consumption, ‘c” for the relationship between rs13107325 and alcohol consumption after excluding the effect of putamen, and ‘ab’ as the mediation effect. The significance tests are based on the bootstrapping method (10,000 times). Z- statistics and the corresponding P values are provided in parentheses. The brain icon was created using Mango software, version 4.1 (http://ric.uthscsa.edu/mango/).

We did not find any significant associations of novel SNPs with either cardiac (left ventricular mass or end diastolic volume or right ventricular end diastolic volume) (Supplementary Table 11) or liver fat measures on MRI (Supplementary Table 12), after adjustment for multiple testing.

Effects of SNPs on gene expression

We carried out expression quantitative trait loci eQTL analyses using the Genotype-Tissue Expression (GTEx) and the UK Brain Expression Consortium (UKBEC) datasets; 34 of the 53 novel and known SNPs associated with alcohol consumption have a significant effect on gene expression in at least one tissue, including 33 SNPs that affect gene expression in the brain (Supplementary Tables 13 and 14, and Supplementary Figures 1–3). We found that the most significant eQTLs often do not involve the nearest gene and that several of the SNPs affect expression of different genes in different tissues. For example, SNP rs1991556 in the MAPT gene (https://www.ncbi.nlm.nih.gov/gene/4137) affects expression of 33 genes overall, with most significant effects on the expression of the non-protein coding genes CRHR1-IT1 (also known as C17orf69 or LINC02210) (https://www.ncbi.nlm.nih.gov/gene/147081) and LRRC37A4P (https://www.ncbi.nlm.nih.gov/gene/?term=LRRC37A4P), near MAPT, across a wide range of tissues including brain, adipose tissue and skin (P = 7.2 × 10⁻¹²⁶ to P = 2.5 × 10⁻⁶) (Supplementary Figure 2). Similarly, the A-allele at SNP rs2071305 within MYBPC3 (https://www.ncbi.nlm.nih.gov/gene/4607) affects the expression of several genes and is most significantly associated with increased expression of C1QTNF4 (https://www.ncbi.nlm.nih.gov/gene/114900) across several tissues (P = 1.9 × 10⁻²⁵ to P = 8.4 × 10⁻⁵).

Several of these eQTLs were found to affect expression of genes known to be involved in reward and addiction. SNP rs1053651 in the TCAP-PNMT-STARD3 gene cluster affects expression of the PPP1R1B gene (also known as DARPP-32) (https://www.ncbi.nlm.nih.gov/gene/84152) which encodes a protein that mediates the effects of dopamine in the mesolimbic reward pathway³⁵. Other known addiction-related genes include ANKK1 (https://www.ncbi.nlm.nih.gov/gene/255239) and DRD2 (expression affected by SNP rs7121986) implicated in alcohol and nicotine dependence^36,37, CRHR1 (https://www.ncbi.nlm.nih.gov/gene/1394) (affected by SNP rs1991556) involved in stress-mediated alcohol dependence^38,39 and PPM1G (SNP rs1260326) (https://www.ncbi.nlm.nih.gov/gene/5496) whose epigenetic modification was reported to be associated with alcohol abuse⁴⁰.

Over-representation enrichment analyses based on functional annotations and disease-related terms indicated that genes whose expressions are affected by the identified eQTLs are most significantly enriched for terms related to abdominal (n=91) and other malignant cancers, motor function (n= 5) and cellular homeostasis (n= 22) (Supplementary Figure 4). We performed a gene-based analysis and repeated the over-representation enrichment analysis adding the new set of identified genes (Supplementary Table 15). The results were similar supporting an enrichment for abdominal (n=100) and other cancers, as well as motor function (n=5) and cellular homeostasis (n=24) (Supplementary Figure 5).

Other traits and diseases

Using LD score regression¹², we assessed genetic correlations between alcohol consumption and 235 complex traits and diseases from publicly available summary GWAS statistics (Methods). All results including their statistics (i.e. r_g, standard errors, z value and P value) are included in Supplementary Table 16. The strongest positive genetic correlations based on false discovery rate P < 0.02 were found for smoking (r_g= 0.42, P = 1.0 × 10⁻²³) and HDL cholesterol levels (r_g= 0.26, P = 5.1 × 10⁻¹³). We also found negative correlations for sleep duration (r_g= −0.14, P = 3.8 × 10⁻⁷) and fasting insulin levels (r_g= −0.25, P = 4.5 × 10⁻⁶). A significant genetic correlation was also found with schizophrenia (r_g= 0.07, P = 3.9 × 10⁻³) and bipolar disorder (r_g= 0.15, P = 5.0 × 10⁻⁴) (Supplementary Table 16). Over-representation enrichment analysis using WebGestalt⁴¹ (http://www.webgestalt.org) showed that our list of novel and known variants is significantly enriched for several diseases and traits including developmental disorder in children (P = 7.3 × 10⁻⁵), epilepsy (P = 1.4 × 10⁻⁴), heroin dependence (P = 5.7 ×10⁻⁴) and schizophrenia (P = 8.4 × 10⁻⁴) (Supplementary Figure 6). The result of the Mendelian randomization analysis (Methods) to assess a potential causal effect of alcohol on schizophrenia risk, using the inverse variance weighted approach, was not significant (P = 0.089), with large heterogeneity of the estimates of the tested variants.

Functional studies in Drosophila

Based on our GWAS and brain imaging findings we took forward SNP rs13107325 in SLC39A8 (alias Zip8 gene) for additional testing in Drosophila, which employ conserved mechanisms to modulate ethanol-induced behaviors^42,43. First, we overexpressed human Zip8 using a Gal4-driver that included expression in neurons involved in multiple ethanol-induced behaviors⁴³. Flies carrying ics^Gal4/+ UAS-hZip8/+ showed a slight, but significant, resistance to ethanol-induced sedation compared to control flies (t₃₀ = 2.3; Hedge’s g = 0.80; 95% CI: 0.08 – 1.53; P = 0.026; N = 16 per genotype). Ethanol tolerance, induced with repeat exposures spaced by a 4-hour recovery, was unchanged in these flies (t = 1.0; P = 0.33; Fig. 4a). We next used the same Gal4-driver to knock down the endogenous Drosophila ortholog of hZip8, namely dZip71B. This caused the flies to display naïve sensitivity to ethanol-induced sedation (t₁₄ = 3.98; Hedge’s g = −1.84; 95% CI: −0.67 – −3.01; P = 0.0014; N = 8 per genotype), and in addition, these flies developed greater tolerance to ethanol upon repeat exposure (t₁₄ = 4.80; Hedge’s g = 2.29; 95% CI: 1.03 – 3.55; P = 0.0003; Fig. 4b). To corroborate this phenotype, we then tested flies transheterozygous for two independent transposon-insertions in the middle of the dZip71B gene (Supplementary Figure 7) and found that these dZip71B^{Mi / MB} flies also displayed naïve sensitivity (t₁₄ = 3.23; Hedge’s g = −1.54; 95% CI: −0.42 – −2.65; P = 0.006) and increased ethanol-induced tolerance (t₁₄ = 2.39; Hedge’s g = 1.13; 95% CI: 0.07 – 2.18; P = 0.032) compared to controls (N = 8 each) (Fig. 4c).

Figure 4. Comparison of Zip8 alcohol phenotypes in Drosophila.

Flies were exposed to 100/50 Ethanol/Air vapor for 30 min for exposure 1, and the time to 50% loss of righting was determined (ST-50, sedation time). After recovery on food for 4 hours, flies were re-exposed to the same vapors, and the second ST-50 recorded (left side). The resulting increase in ST-50, i.e. tolerance, is shown on the right. In a) overexpressed human hZIP8 in ics-expressing cells flies are compared against controls whereas in b) knockdown of the fly ortholog dZip71B is compared against controls. In c) flies carrying two transposon insertions in the endogenous dZip71B gene are compared against controls. Significance levels: ***P <0.001, **P <0.01, *P <0.05. Exact P-values are presented in the text.

DISCUSSION

Our discovery utilizing data on common variants from over 480,000 people of European descent extends our knowledge of the genetic architecture of alcohol intake, increasing the number of identified loci to 46. We found loci involved in neuropsychiatric conditions such as schizophrenia, Parkinson’s disease and dementia, as well as BDNF where gene expression is affected by alcohol abuse. Our findings illustrate that large-scale studies of genetic associations with alcohol intake in the general population, rather than on alcohol dependency alone, can provide additional insights into genetic mechanisms regulating alcohol consumption.

We highlight the role of the highly pleiotropic MAPT and SLC39A8 genes in the genetics of alcohol consumption. MAPT plays a key role in tau-associated dementia⁴⁴ and both genes are also implicated in other neuropsychiatric conditions including neuroticism, schizophrenia and Parkinson’s disease^16–18. The SLC39A8 gene encodes a member of the SLC39 family of metal ion transporters. The encoded protein is glycosylated and found in plasma membrane and mitochondria, and is involved in the cellular transport of zinc, modulation of which could affect microglial inflammatory responses⁴⁵. Our gain- and loss-of-function studies in Drosophila indicate a potential causal role of SLC39A8 in alcohol drinking behavior, even though results should be interpreted with caution due to small sample size in our experiment. The MRI brain imaging demonstrates a significant association of SNP rs13107325 in the SLC39A8 gene and putamen volume differences, and these structural differences appear to partially mediate associations of rs13107325 with alcohol consumption. The putamen has been associated with alcohol consumption and the withdrawal syndrome after chronic administration to rodents and non-human primates⁴⁶. Our mediation analysis is suggestive of a plausible causal pathway linking rs13107325 in SLC39A8 with alcohol intake via an effect on putamen volume, but follow-up work is needed to conclusively demonstrate causal links. Putamen volume differences have also been associated with both schizophrenia and psychosis^47,48 and robust association between SNP rs13107325 in SLC39A8 and schizophrenia was reported in a previous GWAS²³.

We also report SNP rs7121986 near DRD2 as a novel alcohol intake variant in GWAS. The gene product of DRD2, D2 dopamine receptor, is a G protein-coupled receptor on post-synaptic dopaminergic neurons that has long been implicated in alcoholism⁴⁹. In addition, we identify SNP rs988748 in BDNF as a novel alcohol intake variant; BDNF expression is differentially affected by alcohol exposure in animal models^50,51. Both genes (along with PPP1R1P) are centrally involved in reward-mediating mesocortico-limbic pathways and both are implicated in the development of schizophrenia. For example, there is a robust GWAS association between schizophrenia and SNP rs4938021 in DRD2 (in perfect LD with our novel alcohol intake-related variant rs7121986) and DRD2 appears to be pivotal in network analyses of genes involved in schizophrenia⁵². Taken together, our results suggest that there are shared genetic mechanisms between the regulation of alcohol intake and susceptibility to schizophrenia, as well as other neuropsychiatric disorders. In this regard, large prospective epidemiological studies report a three-fold risk of schizophrenia in relation to alcohol abuse⁵³.

We previously reported genome-wide significant associations of alcohol intake with KLB, and identified a liver-brain axis linking the liver hormone FGF21 with central regulation of alcohol intake involving β-Klotho receptor (the gene product of KLB) in the brain⁵. Here, we identify a significant variant near FGF21 gene and strongly replicate the previously reported KLB gene variant, strengthening the genetic evidence for the importance of this pathway in regulating alcohol consumption.

The LD score regression analysis showed a positive genetic correlation between alcohol consumption, smoking and HDL cholesterol levels. This confirms previous findings that reported an almost identical genetic correlation of alcohol consumption with number of cigarettes per day⁵⁴. Furthermore, the observed genetic correlation with HDL levels is consistent with previous observations of an association between alcohol consumption and HDL^55,56, including results of a Mendelian randomization study that suggested a possible causal role linking alcohol intake with increased HDL levels⁵⁷. Furthermore, we found a genetic correlation (inverse) between sleep duration and alcohol consumption, an association previously reported only in a few small epidemiological studies⁵⁸. We also found a significant genetic correlation with schizophrenia and bipolar disorder, a result that is supported by a recently published trans-ethnic meta-analysis of case-control studies on alcohol dependence⁵⁹. We could not test for a genetic association between alcohol and risk of alcohol-related cancers⁶⁰ because of limited availability of summary data. However, our gene-set enrichment analysis showed a significant enrichment for genes related to abdominal as well as other cancers.

Strengths of our study include its size, detailed attention to the alcohol phenotype, dense coverage of the genome through imputation, and incorporation of brain and other imaging data to explore potential mechanisms. Over 80% of the data came from UKB, which combines high-quality phenotypic data and imputed genome-wide genetic data with strict attention to quality control⁶¹. We adopted a stringent approach to claim novel variants involving a conservative P-value threshold, internal replication in UKB and consistent direction of effect with the other studies, to minimize the reporting of false positive signals.

However, since alcohol intake is socio-culturally as well as genetically determined, it is influenced by other lifestyle and environmental factors which may modify or dilute the genetic signal. A key limitation is that assessment of alcohol intake relies on self-report, which is prone to errors and biases including recall bias and systematic under-reporting by heavy drinkers^62,63. Furthermore, questionnaires on alcohol intake covered a short duration (e.g. day or week) at a single period, which may not be representative of broader drinking patterns of cohort participants. We harmonized data across cohorts by converting alcohol intake into a common metric of g/d, with imputation as necessary in UKB for participants reporting consumption of small amounts of alcohol. Taking this approach, we were able to detect strong genetic associations with alcohol intake that explained 7% of the variance in alcohol in an independent cohort, while our GRS analysis indicates that individuals in the lower fifth of the GRS distribution were consuming daily approximately one third of a standard drink (2.6 g/d alcohol) less compared with those in the upper fifth.

We should also point out that our eQTL analyses are a first step in the identification of causal genes. Yet, as the most significant eQTLs affected expression of many genes, not necessarily the nearest, there is a need to further prioritize potential causal genes. Unbiased strategies that leverage information from multiple data sets including extensive genomic annotations and high-throughput functional screening in a broad range of tissues will be essential for effective prioritization of genes and uncovering of underlying causal mechanisms⁶⁴. Establishing confidence in the prioritized genes in such a way is a prerequisite for performing functional follow-up studies in appropriate model systems, as demonstrated by the identification of the causal genes and potential disease mechanisms at the obesity- associated FTO locus⁶⁵.

In summary, in this large study of genetic associations with alcohol consumption, we identified common variants in 46 novel loci, with several of the genes expressed in the brain as well as other tissues. Our findings suggest that there may be shared genetic mechanisms underpinning regulation of alcohol intake and development of a neuropsychiatric disorders including schizophrenia. This may form the basis for greater understanding of observed associations between alcohol consumption, schizophrenia⁶⁶ and other disorders.

METHODS

UK Biobank data

We conducted a Genome Wide Association Study (GWAS) analysis among 458,577 UKB participants of European descent, identified from a combination of self-reported and genetic data. The details of the selection of the participants has been described elsewhere¹⁴. These comprise 408,951 individuals from UKB genotyped at 825,927 variants with a custom Affymetrix UK Biobank Axiom Array chip and 49,626 individuals genotyped at 807,411 variants with a custom Affymetrix UK BiLEVE Axiom Array chip from the UK BiLEVE study, which is a subset of UKB. For our analyses, we used SNPs imputed centrally by UKB using the Haplotype Reference Consortium (HRC) panel.

Alcohol intake

We calculated the alcohol intake as grams of alcohol per day (g/d) based on self-reported alcohol drinking from the touch-screen questionnaire. The quantity of each type of drink (red wine, white wine, beer/cider, fortified wine, spirits) was multiplied by its standard drink size and reference alcohol content. Drink-specific intake during the reported drinking period (a week for frequent drinkers defined as: daily or almost daily/once or twice a week/three or four times a week; or a month for occasional drinkers defined as: one to three times a month/special occasions only) was summed up and converted to g/d alcohol intake for all participants with complete response to the quantitative drinking questions. The alcohol intake for participants with incomplete response was imputed by bootstrap resampling from the complete responses, stratified by drinking frequency (occasional or frequent) and sex.

Participants were defined as life-time non-drinkers if they reported ‘never’ on the question on alcohol drinking frequency (UKB field 1558) and ‘no’ for the question on former drinker (UKB field 3731); they were excluded from further analysis. We considered participants with alcohol consumption > 500 g/d as outliers and they were dropped from the analyses. We also excluded participants with missing covariates, leaving data on 404,732 individuals. We log₁₀ transformed g/d alcohol and sex-specific residuals were derived from the regression of log₁₀ transformed g/d alcohol on age, age², genotyping chip and weight.

UKB genetic analysis

We performed linear mixed modeling using BOLT-LMM software⁶⁷, under an additive genetic model, for associations of measured and imputed SNPs with alcohol consumption (sex-specific residuals of the log₁₀ transformed g/d variable). Model building was based on SNPs with MAF > 5%, call rate > 98.5% and HWE P > 1 × 10⁻⁶. SNPs were imputed using the HRC panel with imputation quality INFO score > 0.1. We estimated the LD score regression (LDSR) intercept to assess the degree of genomic inflation beyond polygenicity as well as the lambda inflation factor λ_GC⁶⁸.

The Alcohol Genome-Wide Consortium (AlcGen) and the Cohorts for Heart and Aging Research in Genomic Epidemiology Plus (CHARGE+) consortia

We analyzed available GWAS data from 25 independent studies (N=76,111) from the AlcGen and the CHARGE+ consortia. All study participants were of reported European ancestry and data were imputed to either the 1000 Genome Project or the HRC panel. Alcohol intake in g/d was computed and the log₁₀ transformed residuals were analyzed as described above. Study names, cohort information and general study methods are included in Supplementary Table 2 and 3.

All studies were centrally quality-controlled using easyQC⁶⁹ including filtering for MAF. Finally, we analyzed data on ~7.1 M SNPs at MAF >1% and imputation quality score (Impute [Info score] or Mach [r²]) > 0.3. Genomic control (GC) was applied at study level. We synthesized the available GWAS using a fixed effects inverse variance weighted meta-analysis and summary estimates were derived for AlcGen and CHARGE+.

One-stage meta-analysis

We performed a one-stage meta-analysis applying a fixed-effects inverse variance weighted meta-analysis using METAL⁷⁰ to obtain summary results from the UKB and and the AlcGen plus CHARGE+ GWAS, for up to N=480,842 participants and ~7.1 M SNPs with MAF ≥ 1% for variants present in both the UKB data and AlcGen and CHARGE+ meta-analysis. We assessed the observed heterogeneity using Cochran’s Q and we quantified this using the I² metric. We considered a Cochran’s Q P < 1 $\times$ 10⁻⁴ as significant. The LDSR intercept (standard error), in the discovery meta-analysis was 1.05 and no further correction was applied. QQ plots of the combined meta-analysis summary results, UK Biobank only as well as AlcGen and CHARGE+ only, are presented in Supplementary Figure 8.

Previously reported (known) SNPs

We looked up in the GWAS catalog (http://www.ebi.ac.uk/gwas/) and identified 17 SNPs associated with alcohol consumption at genome-wide significance level (P < 5 $\times$ 10⁻⁸). We enhanced the list by reference to a recent GWAS by Clarke et al⁶ that was not covered by the GWAS catalog at the time of the analysis, reporting 14 additional rare and common SNPs. Together with a SNP in RASGRF2 shown to be associated with alcohol-induced reinforcement⁷¹, we found 31 previously reported alcohol consumption related SNPs.

Novel loci

According to locus definition of i) SNPs within ±500kb distance of each other; ii) SNPs in linkage disequilibrium LD (r² > 0.1) calculated with PLINK, we augmented the list of known SNPs with all SNPs present within our data, not contained within the previously published loci. We further excluded SNPs in the HLA region (chromosome 6, 25–34Mb) due to its complex LD structure. We performed LD clumping in PLINK on 4,515 unknown SNPs with P < 1 ×10⁻⁸ using an r² > 0.1 and distance threshold of 500kb. We further grouped the lead SNPs within 500kb from each other into the same loci and selected the SNP with smallest P-value from the locus as sentinel SNP.

To report a SNP as novel signal of association with alcohol consumption:

1. the sentinel SNP has P < 5 × 10⁻⁹ in the one-stage meta-analysis;

2. the sentinel SNP is strongly associated (P < 5 × 10⁻⁷) in the UKB GWAS alone;

3. the sentinel SNP has concordant direction of effect between UKB and AlcGen and CHARGE+ datasets;

4. The sentinel SNP is not located within any of the previously reported loci

We selected the above criteria i) to iii) to minimize false positive findings including use of a conservative one-stage P-value threshold that is an order of magnitude more stringent than a genome-wide significance P-value. (The threshold of P < 5 × 10⁻⁹ has been proposed e.g. for whole-genome sequencing-based studies.) This approach led us to the identification of 46 sentinel SNPs in total. Regional plots for all 46 sentinel SNPs are presented in Supplementary Figure 9.

Conditional analysis

We conducted locus-specific conditional analysis using the GCTA (Genome-wide Complex Trait Analysis) software (http://cnsgenomics.com/software/gcta). For each of the 46 novel sentinel SNPs, we obtained conditional analysis results for the SNPs with MAF>1% and within 500kb from the sentinel SNP after conditioning on the sentinel SNP. The meta-analysis results of the GWAS in UKB, AlcGen and CHARGE+ were used as input summary statistics and the individual-level genetic data from UKB were used as the reference sample. Results for a SNP were considered conditionally significant if the difference between the conditional P-value and the original P-value is greater than 1.5-fold (-log₁₀P/-log₁₀(P_conditional) >1.5) and the conditional P-value is smaller than 5 × 10⁻⁸.

Gene-based analysis

We performed a gene-based analysis using fastBAT, a method that performs a set-based association analysis using summary-level data from GWAS. We used the UKB dataset as a reference set for the LD calculation⁷². Gene-based associations with P < 5 × 10⁻⁹ were considered significant.

Gene expression analyses

To analyze the impact of genetic variants on expression of neighboring genes and identify expression quantitative trait loci (cis-eQTLs; i.e., SNPs associated with differences in local gene expression), we used two publicly available databases, the Genotype-Tissue Expression (GTEx) database⁷³ (www.gtexportal.org) and the UK Brain Expression Consortium (UKBEC) dataset⁷⁴ (http://www.braineac.org). We searched these databases for significant variant-transcripts pairs for genes within 1Mb of each input SNP.

With the GTEx database, we tested for cis-eQTL effects in 48 tissues from 620 donors. The data described herein were obtained from the GTEx Portal, Release: V7 and used FastQTL⁷⁵, to map SNPs to gene-level expression data and calculate q-values based on beta distribution-adjusted empirical P-values⁷⁶. A false discovery rate (FDR) threshold of ≤0.05 was applied to identify genes with a significant eQTL. The effect size, defined as the slope of the linear regression, was computed in a normalized space (normalized effect size (NES)), where magnitude has no direct biological interpretation. Here, NES reflects the effects of our GWAS A1 alleles (that are not necessarily the alternative alleles relative to the reference alleles, as reported in the GTEx database). Supplementary Table 13 lists transcripts-SNPs associations with significant eQTL effects.

With the UKBEC dataset that comprises 134 brains (http://www.braineac.org/), we searched for cis-eQTLs in 10 brain regions, including the cerebellar cortex (CRBL), frontal cortex (FCTX), hippocampus (HIPP), medulla (specifically inferior olivary nucleus, MEDU), occipital cortex (specifically primary visual cortex, OCTX), putamen (PUTM), substantia nigra (SNIG), thalamus (THAL), temporal cortex (TCTX) and intralobular white matter (WHMT), as well as across all brain tissues (aveALL). MatrixEQTL⁷⁷ generated P-values for each expression profile (either exon-level or gene-level) against the respective SNP were obtained for the 10 different tissues and overall (aveALL). Supplementary Table 14 lists transcripts-SNPs associations with a eQTL P-value < 0.0045 in at least one brain tissue. Subsequent data analysis was performed in R (http://www.R-project.org/).

We carried out over-representation enrichment analysis using a list of 146 GTEx eQTL genes that were derived from the single-variant analysis and a list of 160 eQTL genes that were derived from both single-variant and gene-based analysis. Ingenuity pathway analysis (IPA®, QIAGEN Inc.) was performed on these lists using ontology annotations from all available databases except those derived from low-confidence computational predictions.

Magnetic Resonance Imaging Data

We used the most recent release of magnetic resonance imaging (MRI) data on brain, heart and liver for UKB participants to investigate genetic associations with the 46 novel SNPs for alcohol consumption.

Brain imaging

Brain MRI acquisition and pre-processing

We used the T1 data from UKB to elucidate volumetric brain structures, including the cortical and the sub-cortical areas. The T1 data were acquired and pre-processed centrally by UKB. The brain regions were defined by combining the Harvard-Oxford cortical and subcortical atlases⁷⁸ (https://fsl.fmrib.ox.ac.uk/fsl/fslwiki/Atlases) and the Diedrichsen cerebellar atlas⁷⁹ (http://www.diedrichsenlab.org/imaging/propatlas.htm). FAST (FMRIB’s Automated Segmentation Tool)⁸⁰ was then used to estimate the grey matter partial volume within each brain region. Subcortical region volumes were also modelled by using FIRST (FMRIB’s Integrated Registration and Segmentation Tool). More details about the MRI scanning protocol and pre-processing has been provided in UKB documentation (https://biobank.ctsu.ox.ac.uk/crystal/docs/brain_mri.pdf).

Association Analyses

We performed association analyses on N = 9,702 individuals between all novel SNPs and the grey matter volume of brain regions using Pearson correlation, adjusting for age, age², sex, age × sex, age² × sex, and head size. All, brain volume features, log transformed alcohol intake data (g/d), and the confounders were firstly transformed by using a rank-based inverse Gaussian transformation. Significance levels were set at P < 0.05 adjusted using the false-discovery rate method for multiple comparisons.

Mediation analysis

To assess if the effect of a SNP on alcohol consumption is mediated through a brain region, we performed a single-level mediation analysis based on a standard three-variable path model (SNP-brain region-alcohol consumption) with corrected and accelerated percentile bootstrapping 10,000 times to calculate the significance of the mediation effect. We considered as mediator variable the grey matter volume of brain regions that had a significant association on alcohol consumption. We calculated the significance of path a, path b and a*b mediation (SNP-brain region-alcohol consumption) using a multilevel mediation and moderation (M3) toolbox^81,82. To exclude the possibility of an inverse causal pathway we performed additional analyses in UKB non-drinkers (N =589). performing 10,000 random permutations, associations of rs13107325 with both left and right putamen.

Cardiac Imaging

Cardiac MRI acquisition and pre-processing

Details of the cardiac image acquisition in UKB are reported previously⁸³. Cardiac MRI was acquired using a clinical wide bore 1.5T scanner (MAGNETOM Aera, Syngo Platform VD13A, Siemens Healthcare, Erlangen, Germany) with 48 receiver channels, a 45 mT/m and 200 T/m/s gradient system, an 18-channel anterior body surface coil used in combination with 12 elements of an integrated 32 element spine coil and electrocardiogram gating for cardiac synchronization. A two-dimensional short-axis cardiac MRI was obtained using a balanced steady state free precession to cover the entire left and right ventricle (echo time, 1.10msec; repetition time, 2.6msec; flip angle, 80°; slice thickness, 8mm with 2mm gap; typical field of view, 380×252mm; matrix size, 208×187, acquisition of 1 slice per breath-hold).

The cardiac images were segmented to provide left ventricular mass (LVM), left end-diastolic (LVEDV), left end-systolic volume (LVESV), and right end-diastolic (RVEDV) and right end-systolic volume (RVESV) using a fully convolutional network as described previously⁸⁴. Left (LVEF) and right ventricular ejection fraction (RVEF) were derived from (LVEDV–LVESV)/LVEDV×100 and (RVEDV–RVESV)/RVEDV×100, respectively.

Association Analyses

To test associations between cardiac MRI measures and alcohol consumption-related SNPs, we carried out a regression of LVM, LVEDV, LVEF, RVEDV, and RVEF onto each of the 46 SNPs adjusting for age, sex, height, weight, hypertension (defined as systolic blood pressure >140mmHg and or diastolic blood pressure >90mmHg or under antihypertensive treatment), diabetes, and smoking history on N=10,706 participants. Significance levels were set at P < 0.05 adjusted using the false-discovery rate method for multiple comparisons.

Liver Imaging

Liver MRI acquisition and pre-processing

Details of the liver image acquisition protocol have been reported previously⁸⁵. Briefly, all participants were scanned in a Siemens MAGNETOM Aera 1.5-T MRI scanner (Siemens Healthineers, Erlangen, Germany) using a 6-minute dual-echo Dixon Vibe protocol, providing a water and fat separated volumetric data set for fat and muscle covering neck to knees. For liver proton density fat fraction (PDFF) quantification, an additional single multi-echo gradient slice was acquired over the liver. Liver images were analysed by computing specific ROI for water, fat and T2* by magnitude-based chemical shift technique with a 6-peak lipid model, correcting for T1 and T2*.

Association Analyses

We performed association analyses between 46 alcohol consumption-related SNPs and liver PDFF (%), from 8,479 samples, using a linear regression model adjusting for age, age², sex, T2D, BMI, genotyping chip and first three PCs. Liver PDDF was firstly transformed by using a rank-based inverse transformation. Significance levels were set at P < 0.05 adjusted using the false-discovery rate method for multiple comparisons.

Drosophila experiments

Flies were kept on standard cornmeal/molasses fly food in a 12:12hr light:dark cycle at 25°C. Transgenc flies were obtained from the Bloomington Drosophila Stock Center: UAS-hZip8 BL#66125, UAS-dZIP71B-TRiP-RNAi^HMC04064 BL#55376, dZip71B^MI13940 BL#59234, and dZip71B^MB11703 BL#29928. For behavioral experiments, crosses were set up such that experimental and control flies were sibling progeny from a cross, and both were therefore in the same hybrid genetic background (w Berlin / unknown). Flies aged 1–5 days of adult age were collected, exposed to 100/50 (flowrates) ethanol/air vapor in the Booze-o-Mat 2 days later, and their loss of righting determined by slight tapping, as described⁸⁶. For tolerance, flies were put back onto regular food after a 30-min initial exposure and were then re-exposed to the same vapor 4 hours later. Note that tolerance is not connected to initial sensitivity, and flies naively sensitive to ethanol-induced sedation can have no, or a reduced tolerance phenotype. Flies overexpressing hZip8 (and their sibling controls) were placed at 28°C for two days to increase the expression levels of the transgene, as we did not detect a phenotype when they were kept at 25°C (data not shown). Data from experimental and control flies were compared by two-sided Student’s t-tests. Data were normally distributed according to Shapiro-Wilk testing with Bonferroni adjustment for each of the three experiments.

Effects on other traits and diseases

We queried SNPs against GWAS results included in PhenoScanner (http://www.phenoscanner.medschl.cam.ac.uk), to investigate cross-trait effects, extracting all association results with genome-wide significance at P < 5 × 10⁻⁸ for all SNPs in high LD (r² ≥ 0.8) with the 46 sentinel novel SNPs, to highlight the loci with strongest evidence of association with other traits. At the gene level, overrepresentation enrichment analysis (ORA) with WebGestalt⁴¹ on the nearest genes to all alcohol consumption loci was carried out.

The genetic correlations between alcohol consumption and 235 other traits and diseases were obtained in the online software LD Hub. LD hub is a centralized database of summary-level GWAS results and a web interface for LD score regression analysis

To estimate the potential causal effect of alcohol consumption-related variants on schizophrenia, we performed a Mendelian randomization analysis utilizing publicly available GWAS data on schizophrenia and the Mendelian randomization package in R. The effect was estimated using the inverse-variance weighted (IVM) method. Pleiotropy was tested by applying the MR-Egger regression method and heterogeneity statistics were obtained. In presence of heterogeneity the random effects inverse-variance method was applied⁸⁷.

Genetic risk scores and percentage of variance explained

We calculated an unbiased weighted GRS in 14,004 unrelated participants in Airwave, an independent cohort with high quality HRC imputed genetic data³³. All previously reported and novel variants were used for the construction of the GRS. We weighted the alcohol-increasing alleles by the beta coefficients of the meta-analysis. We assessed the association of the GRS with alcohol intake and calculated the alcohol consumption levels for individuals in the top vs the bottom quintiles of the distribution. To calculate the percent of variance of alcohol consumption explained by genetic variants, we generated the residuals from a regression of alcohol consumption in Airwave. We then fit a second linear model for the trait residuals with all novel and known variants plus the top 10 principal components and estimated the percentage variance of the dependent variable explained by the variants.

Statistical analysis

All inferential statistics for the analyses described above are provided in the text or in tables and figures. All performed tests were two-sided.

Data availability statement

The UKB GWAS data can be assessed from the UK Biobank data repository (http://biota.osc.ox.ac.uk/). The genetic and phenotypic UKB data are available upon application to the UK Biobank (https://www.ukbiobank.ac.uk). Summary GWAS data data can be assessed by request to the corresponding authors and will be available via LDHub (http://ldsc.broadinstitute.org/ldhub/).