Genomewide association study identifies a novel locus for cannabis dependence

Abstract

Despite moderate heritability, only one study has identified genomewide significant loci for cannabis-related phenotypes. We conducted meta-analyses of genomewide association study (GWAS) data on 2,080 DSM-IV cannabis dependent cases and 6,435 cannabis exposed controls of European descent. A cluster of correlated single nucleotide polymorphisms (SNPs) in a novel region on chromosome 10 was genomewide significant (lowest p = 1.3E-8). Among the SNPs, rs1409568 showed enrichment for H3K4me1 and H3K427ac marks, suggesting its role as an enhancer in addiction-relevant brain regions, such as the dorsolateral prefrontal cortex and the angular and cingulate gyri. This SNP is also predicted to modify binding scores for several transcription factors. We found modest evidence for replication for rs1409568 in an independent cohort of African-American (896 cases and 1591 controls; p=0.03) but not European-American (781 cases and 1905 controls) participants. The combined meta-analysis (3,757 cases and 9,931 controls) indicated trend-level significance for rs1409568 (p=2.85E-7). No genomewide significant loci emerged for cannabis dependence criterion count (n=8,050). There was also evidence that the minor allele of rs1409568 was associated with a 2.1% increase in right hippocampal volume in an independent sample of 430 European-American college students (fwe-p=.007). The identification and characterization of genomewide significant loci for cannabis dependence is amongst the first steps towards understanding the biological contributions to the etiology of this psychiatric disorder, which appears to be rising in some developed nations.

INTRODUCTION

Cannabis is amongst the most commonly used illicit psychoactive substances in developed nations (1;2). Ten percent of individuals who ever use cannabis meet criteria for lifetime cannabis dependence, which is associated with significant comorbid adverse mental health outcomes (3–5). A recent survey of U.S. adults showed that the past year prevalence of cannabis use disorders has increased from 1.5% to 2.9% in the decade spanning 2002–2012, an increase apparently attributable to a corresponding increase in use during that period of time (6).

About 50–60% of the variance in cannabis use disorders, including DSM-IV dependence, is attributable to the additive effects of genes (i.e., narrow sense heritability)(7). Despite this, only one study to date has successfully identified genomewide significant loci for any cannabis related trait (8). Table 1 provides an overview of six genomewide association studies (GWASs) of cannabis-related phenotypes (9–12), the largest being a recent meta-analysis of GWASs of ever using cannabis, even once during the lifetime (N > 32,000)(13). However, only the recent study by Sherva and colleagues (8) identified genomewide significant loci (three independent regions) for DSM-IV cannabis dependence criterion counts in a sample of European-American (EA) and African-American (AA) descent.

Table 1.

Summary of existing genomewide association studies of cannabis-related phenotypes

Author, Date	Phenotype	N	Genomewide significant SNPs	Gene-based significance	Heritability
Agrawal, 2011; PMC3117436	DSM-IV cannabis dependent	708 cases, 2346 controls (exposed)	none	-	-
Verweij, 2013; PMC3548058	Cannabis Use	10,091	none	none	6% (p=.28)
Agrawal, 2014; PMC3943464	Factor score of DSM5 criteria	3,053	none	C17orf58, BPTF and PPM1D	21% (p=.13)
Minica, 2015; PMC4561059	Cannabis use Age at initiation	6,774	none	none	25% (p=.002)
Sherva, 2016; PMID27028160	Cannabis dependence symptom count	14,754	rs143244591 (chr3:149296148; RP11-206M11.7; p=4.3E-10); rs146091982 (chr10:93900201; SLC35G; p=1.3E-9) rs77378271 (chr8:3215967; CSMD1;p=2.2E-8);	-	-
Stringer, 2016; PMID27023175	Cannabis use	32,330 + 5,627	none	NCAM1, CADM2, SCOC and KCNT2.	13–20% (p<.001)

We conducted a meta-analysis of GWAS data on individuals of European descent from five cohorts to identify loci associated with DSM-IV cannabis dependence (N=2,080). We compared individuals who met criteria for DSM-IV cannabis dependence (N=2,080) to controls who did not meet criteria for cannabis dependence but reported having used cannabis, at least once, during their lives (N=6,435). In addition to comprehensive locus (including epigenetic) annotation, we examined whether genomewide significant SNPs were associated with variability in gray matter volume within brain regions (bilateral amygdala, ventral striatum and hippocampus) previously associated with chronic cannabis use and misuse (14;15) among an independent cohort of 430 EA college students. Some prior studies have reported lower gray matter volume in these brain regions, although results are inconclusive. While a majority of studies have attributed such volumetric changes to the effects of chronic cannabis exposure (e.g.,(16)), at least one study has implicated common predisposing influences, such as genetic liability, as the major contributor to the association between casual cannabis use and variability in amygdala volume (17). As this sample of college students included <10 individuals who met criteria for cannabis dependence, we were principally interested in examining whether the top loci that emerged from the GWAS were associated with volumetric differences, whether regional brain volume varied across cannabis users and nonusers and further, whether the effects of top loci on cannabis involvement could be partly attributed to variability in brain volume.

MATERIALS AND METHODS

Samples

Data were drawn from 5 cohorts: (a) a case-control (18) and (b) family GWAS (19;20) component of the Collaborative Study on the Genetics of Alcoholism (COGA; COGA-cc and COGA-f), (c) the Study of Addictions: Genes and Environment (SAGE)(21), (d) the Australian Alcohol (22), Nicotine Addiction Genetics (23), and Childhood Trauma (24) studies (OZALC+) and (e) the Comorbidity and Trauma Study (CATS)(25). Individual studies have been described in detail in related publications and in Supplemental Text. An outline of the samples used in this study is available in Table 2. As the overwhelming majority of the data were on individuals of European-Australian and European-American descent, discovery analyses were restricted to individuals of European descent. All subjects provided informed consent and protocols were approved by the institutional review boards overseeing the individual studies (see Supplemental text).

Table 2.

Sample characteristics of discovery cohorts of European-American and European-Australian (EA) individuals included in meta-analysis (CATS, COGA-cc, COGA-f, OZALC+, SAGE), replication cohort (Yale-Penn) and neuroimaging extension (DNS) samples. Only individuals with a lifetime history of ever using cannabis are included.

Study	Ncase	Ncontrols	Median age	% Male	% Alcohol dependent	% Nicotine dependent	% Cocaine dependent	% Opioid dependent
Discovery Samples
CATS	799	813	36	57.5 %	38.8 %	60.2 %	24.8 %	76.1 %
COGA-cc	311	593	40	60.1 %	79.4 %	49.0 %	34.4 %	13.3 %
COGA-f	368	894	36	50.6 %	47.0 %	40.3 %	13.9 %	5.7 %
OZALC	357	3094	43	51.7 %	40.2 %	50.7 %	0.4 %	0.5 %
SAGE	245	1041	38	46.4 %	55.0 %	53.4 %	25.2 %	9.3 %
Replication Samples
Yale-Penn EA	781	1591	38	57.7 %	74.2%	77.3 %	78.9 %	62.0 %
Yale-Penn AA	896	1905	42	54.6 %	59.9 %	57.5 %	75.6 %	21.2 %
Neuroimaging Sample
DNS	-	-	19	46.7%	6.3%	0%	0%	0%

CATS: Comorbidity and Trauma Study; COGA-cc: Case-control component of the Collaborative Study of the Genetics of Alcoholism, COGA-f: family-based component of the Collaborative Study of the Genetics of Alcoholism; OZALC+: Australian alcohol, nicotine and trauma studies; SAGE: Study of Addictions: Genes and Environment; DNS: Duke Neurogenetics Study.

Summary statistics from European ancestry subjects in CATS, COGA-ccGWAS, COGA-fGWAS, OZALC+ and SAGE were combined to form the discovery analysis. Replication analyses were conducted in the Yale-Penn (8) sample which was the major dataset contributing to the prior study by Sherva et al (8). Yale-Penn includes a large number of AA participants; thus, results from both EA and AA subjects were separately examined. Sherva et al also included SAGE data in their discovery cohort and used CATS as a replication sample. In our analyses, only the Yale-Penn component of Sherva et al (8) was used for replication, while SAGE and CATS were part of the discovery cohort.

Genotyping

A variety of Illumina platforms were used to genotype the cohorts (Table S1). Quality control and imputation metrics (26–29) for the individual samples are provided in referenced publications (18;19;21;22;25) and in Table S1.

Phenotype

Cases met criteria for DSM-IV cannabis dependence(30), which included withdrawal (i.e., 3 or more of 7 criteria) in COGA and SAGE but not in CATS or OZALC+. Controls did not meet criteria for cannabis dependence but reported a lifetime history of ever having used cannabis, even once. Follow-up analyses of top loci examined whether excluding those with DSM-IV cannabis abuse or 1–2 dependence criteria modified the results. A natural log-transformed (to account for skewed data) count of DSM-IV dependence criteria (0–6, excluding withdrawal; adding “1” for 0 values) was also analyzed (n=8,050). Finally, the effect of comorbid DSM-IV alcohol, nicotine and opioid dependence was investigated by examining their association with top loci in post hoc analyses.

Statistical analysis

Each sample was analyzed separately using specific analytic protocols that have been validated for that sample (18;22;25;31;32). Prior to meta-analysis, SNPs that did not satisfy quality control standards imposed for the current study were excluded (see Supplemental Text); only SNPs that survived quality control in all 5 samples were included in the meta-analysis. PLINK (v1.07)(33) was used to analyze allele dosage data for SAGE, CATS and COGA-cc. GWAF-GEE (34) was used to analyze the family data for DSM-IV cannabis dependence from COGA-f and OZALC+. Linear mixed models and Merlin-offline(35) were used to analyze criterion counts in COGA-f and OZALC+ respectively. Logistic and linear regressions were used for the diagnosis and count definitions, respectively (see Table S1 for covariates used for each sample). Results were meta-analyzed in METAL (36) using inverse variance weighting procedures and genomic control correction. Gene-based association analyses were conducted using MAGMA (37) with the 1000 Genomes European data (release version 3, May 22, 2014) as the reference panel. Gene boundaries were extended to include a 10kb window at the 3′ and 5′ ends.

Annotation

Top SNPs (p<5E-8) were annotated using a variety of resources that are described in Supplemental Text.

Replication

Replication analyses were conducted in the Yale-Penn study (described in Supplemental Text, related publications (13;38) and Table S1). Cases met criteria for DSM-IV cannabis dependence (N_EA=781, N_AA=896) and controls (N_EA=1,591, N_AA=1,905) reported a lifetime history of cannabis use.

Neuroimaging extension

Data on 430 EA college students aged 18–22 years were drawn from the Duke Neurogenetics Study (DNS (39); Supplemental Text). First, we examined the association between genotype (rs1409568, modeled as C-allele carriers vs. T allele homozygotes) and (a) cannabis use (ever used and frequency of use in ever users) and (b) regional brain volume. A Generalized Linear Model in SPM8 was used to test whether genotype predicted regional volume within 6 brain regions (i.e., left and right amygdala, hippocampus, striatum) previously associated with cannabis use and misuse (14;15). Familywise error correction (FWE p < .05) with a 10-voxel extent cluster threshold was applied to each of these 6 anatomical regions of interest (ROIs) derived from the Automated Anatomical Labeling atlas (40) within Wake Forest University Pick Atlas software (41). Additional methodological details are presented in Supplemental Text. Second, we tested whether cannabis use (ever used; frequency of use in ever users) was associated with regional gray matter volume in any of these regions. Third, we examined whether associations between genotype and regional brain volume persisted after controlling for cannabis use. All DNS analyses controlled for sex and age; analyses on regional brain volume also controlled for total intracranial volume (ICV), while analyses including genotype additionally included the first three principal components of ancestry. All non-imaging analyses and group comparisons were conducted using the R (3.1.2) ‘Stats’ package.

RESULTS

Sample characteristics

Samples were relatively similar in age and gender distribution. By ascertainment design, there was considerable overrepresentation of all forms of substance use disorder across the samples (Table 2), with the exception of OZALC+ which included families that were ascertained based on family size rather than substance-related problems.

GWAS results

DSM-IV cannabis dependence

Lambdas for individual studies and meta-analyses were close to 1.0 (Supplemental Table S1; Figure S1A). Genomewide significant loci did not emerge in any individual study. Meta-analysis of summary statistics from the 5 discovery samples (CATS, COGA-cc, SAGE, COGA-f, OZALC) revealed a cluster of genomewide significant SNPs in a region on chromosome 10 (Table 3 for loci at p-value < 10^−6; Supplemental S2A for Manhattan plot; full results available upon request), with genomewide significant loci representing a single signal (Figure 1A for regional association plot (42)). The lowest p-value was associated with rs77300175 (p-value = 1.3E-8; Table 3), with stronger contributions from the 3 case-control cohorts (SAGE, CATS, COGA-cc; Supplemental Table S2) than the family-based cohorts (COGA-f and OZALC+).

Table 3.

Association results for SNPs at p-value ≤ 1 × 10⁻⁶ in 2,080 cannabis dependent cases and 6,435 cannabis-exposed controls of European-American descent.

SNP	Chr: position	Function	Effect Allele	Alternate Allele	Meta-analysis			Direction of effects**
					Effect size (β)	SE	P-value
rs112825709	10:120622014	intergenic	A*	C	0.50	0.09	8.04E-08	+++++
rs151284751	10:120622746	intergenic	A*	C	0.50	0.09	8.06E-08	+++++
rs145575521	10:120622907	intergenic	T	C	−0.51	0.09	4.39E-08	− − − − −
rs79516280	10:120624808	intergenic	A	G	−0.51	0.09	4.34E-08	−−−−−
rs75312482	10:120626227	intergenic	T*	C	0.51	0.09	4.66E-08	+++++
rs1409568g	10:120630785	intergenic	T	C	−0.50	0.09	3.95E-08	−−−−−
rs77300175	10:120633376	intergenic	T*	C	0.53	0.09	1.30E-08	+++++
rs7098706g	10:120639977	intergenic	T	C	−0.52	0.09	2.44E-08	−−−−−
rs118006754	10:120641184	intergenic	T*	G	0.51	0.09	4.12E-08	+++++
rs7074123	10:120643763	intergenic	A	C	−0.52	0.09	1.79E-08	−−−−−
rs7920901	10:120648450	intergenic	T*	C	0.52	0.09	1.74E-08	+++++
rs57602752	10:120649972	intergenic	A	C	−0.52	0.09	1.88E-08	−−−−−
rs115048844	10:120651442	intergenic	C*	G	0.52	0.09	1.86E-08	+++++
rs1961317	10:120654022	intergenic	T*	C	0.52	0.09	2.08E-08	+++++
rs147702664	10:120658617	intergenic	A	G	−0.57	0.10	4.07E-08	−−−−−
rs149791363	10:120658646	intergenic	A	C	−0.58	0.11	3.76E-08	−−−−−
rs150525973	10:120659352	intergenic	T*	C	0.54	0.10	3.27E-08	+++++
rs79277226	10:120660716	intergenic	A*	G	0.49	0.09	5.23E-08	+++++
rs113036365	10:120663067	intergenic	T*	G	0.48	0.09	6.37E-08	+++++
rs60120125	10:120663137	intergenic	T	C	−0.48	0.09	6.40E-08	−−−−−
rs61538293	10:120663338	intergenic	C	G	−0.49	0.09	7.34E-08	−−−−−
rs111332403	10:120666212	intergenic	A	G	−0.49	0.09	2.15E-08	−−−−−
rs12771281	10:120675667	intergenic	C	G	−0.41	0.08	7.11E-07	−−−−−
rs12413263	10:120675738	intergenic	A	C	−0.40	0.08	5.67E-07	−−−−−
rs35728709	10:120706542	intergenic	T*	C	0.41	0.08	7.30E-07	+++++

^*indicates that the effect allele is also the minor allele in individuals of European descent.

^**Order of effect sizes from studies is CATS, COGA-cc, COGA-fGWAS, OZALC+, and SAGE

^gSNP genotyped in at least one sample. All other SNPs were imputed across samples.

Figure 1.

Regional association plot of chromosome 10 SNPs (centered at rs1409568 ± 500kb) associated with cannabis dependence cases-status (N=2,080) compared with cannabis exposed controls (N=6,435).

Cannabis dependence criterion count

There was no evidence for genomewide significant loci associated with cannabis dependence symptom counts (Supplemental Table S3; Supplemental Figures S1B and S2B). The most promising association was noted for a cluster of SNPs in chromosome 2 (e.g. rs2287641, p=9E-7). The chromosome 10 SNPs were similarly associated, but not at genomewide significant levels (e.g. rs150525973 P = 1.2E-6).

Replication

For the DSM-IV dependence diagnosis, findings were not replicated in Yale-Penn EA participants (Supplemental Table S4); effect sizes were consistently in the same direction, but smaller (e.g. rs1409568: β = −0.072, p=0.6). Consistent with our finding, the T allele of rs1409568 was associated with a reduced likelihood of cannabis dependence among the AA participants from Yale-Penn (β = −0.18, p = 0.028). When results from all datasets, discovery and replication (EA and AA), were meta-analyzed together (N_case=3,757, N_control=9,931), rs1409568 remained associated with DSM-IV cannabis dependence at a trend level (β = −0.28; p = 2.9E-7). In addition, there was no evidence from our meta-analysis for association between cannabis dependence diagnosis or symptom counts with previously identified loci for cannabis use (i.e., top 10 signals from Stringer et al(13)) and top EA locus from Sherva et al (8) – Supplemental Table S5).

Gene-based association

There was no evidence for enrichment of association within genes for cannabis dependence diagnosis. (Supplemental Table S6); However, for symptom count, MEI1, on chromosome22, was associated at a gene-level (p=2.55E-6; Supplemental Table S7). Several other genes with SNPs of nominal significance clustered in this chromosomal region (Figure S3 for chromosome 22 regional association plot).

Genomic and Epigenomic Annotation

Genomewide significant SNPs on chromosome 10 were not in linkage disequilibrium (r² ≥ 0.6) with non-synonymous variants in neighboring genes. No significant cis-eQTLs were identified for any chromosome 10 variant in any tissue in GTEx (43) as well as dorsolateral prefrontal cortex (dlPFC) tissue from the Common Mind Consortium data(44). However, there was preliminary evidence that rs1409568 (RegulomeDB score 3a), but not other variants in the region (scores ≥ 5) may have regulatory effects (45). Closer inspection in the Epigenome Browser (46) showed that rs1409568 was accompanied by enhancer-enriched active histone modifications (H3K4me1 and H3K27ac) in a variety of brain tissues (Figure 2). Evidence of an active enhancer was particularly prominent in the dorsolateral prefrontal cortex (dlPFC), angular gyrus, cingulate gyrus and the inferior temporal lobe. There were also enriched H3K4me1 and H3K27ac signals in the middle hippocampus and the substantia nigra, however these signals were not detected at corrected thresholds defined by MACS (q-value cutoff 0.05). All of these regions are strongly implicated in the etiology of addiction (47).

Figure 2.

Epigenetic annotation of rs1409568 on chromosome 10 depicting preliminary in silico evidence for an active enhancer mark.

We determined that rs1409568 was within a chromosome 10 regulatory domain spanning 120,300,000bp – 120,790,000bp that encompassed all of the genome-wide significant SNPs. The regulatory domain included 12 genes, including 3 protein coding genes (PRLHR, CACUL1, NANOS1), 4 pseudogenes (SLC25A18P1, TOMM22P5, RP11-215A21.2, LDHAP5), and 5 non-coding genes (AL356865.1, AL356865.2, U3, RP11-498J9.2, AL15778.1). Seven of 12 genes were expressed in several brain-derived tissues (Supplementary Figure S4). RP11-215A21.1 is the gene closest to rs1409568 (1.3kb from the transcription start site), however, there is no evidence that rs1409568 regulates the expression of any gene within the regulatory domain. The T allele of rs1409568 is conserved within primates, but not between primates and rodents (Supplementary Figure S5).

There was also evidence that rs1409568 altered the binding motif for several transcription factors that are critical during embryogenesis, including those encoded by genes that include homeodomains (e.g., HOXD8, VAX1) and those from the Pit-Oct-Unc (POU) family (e.g., POU4F1, POU4F3, POU6F2: for full list, see Supplemental Table S8). Although predictions were based on common tissue sources, several transcription factors showed brain-related expression (e.g., POU6F2).

We identified 26 CpG probes that corresponded to genes with transcription start sites (TSS) within 1 Mb of rs1409568. Differences in CpG methylation were examined in CT (n=34) and TT (n=313) individuals in tissue from the frontal cortex and cerebellum(48). Only one probe (cg23182539), corresponding to TIA-1 related protein isoform 1 (TIAL1) showed nominal support for change in methylation scores as a function of genotype (Supplemental Table S9), with lower methylation scores in C allele carriers (β = −0.56, p = 0.0017; Wilcoxon p = 0.005). However, methylation change in this gene was not significant after Bonferroni correction (26 probes x 2 regions; p_corrected = 0.00096).

Sensitivity to definition of controls

The chromosome 10 SNPs represented a single signal (Supplemental Figure S6), so follow-up analyses used a representative locus. Controls (N=6,435) included individuals who did not meet criteria for DSM-IV cannabis dependence but may have met criteria for a lifetime history of DSM-IV cannabis abuse or endorsed 1–2 dependence criteria. Exclusion of individuals with abuse (N=1,590) from among the controls yielded similar effect sizes but diminished statistical significance, likely due to the reduced statistical power (rs7098706: b = −0.53, p = 5.90E-7; rs1409568: b = −0.50, p = 1.21E-6). Excluding control individuals with abuse or 1–2 dependence criteria (N=2,152) had a similar effect (e.g., rs7098706 b = −.50, p = 1.85 E-6; rs1409568: b = −0.48, p = 3.95E-6). Thus, heterogeneity within the control population is not responsible for the observed association.

Comorbidity with other substance use disorders

Only nicotine dependence was associated with rs1409568, and in CATS alone (p = .003) – adding nicotine dependence as a covariate to the CATS analysis did not greatly alter the significance of rs1409568 (p = 5.51E-8; Supplemental Table S10). Alcohol dependence was not associated with rs1409568 in any individual study, although the meta-analytic p-value was less than 0.05.

Genotype and brain volumetric variation

In the DNS, 51% of the sample reported lifetime cannabis use, with 12% (n=52), 15% (n=66), 8.8% (n=38) and 15% (n=65) using cannabis 1–2, 3–10, 11–20 and >21 times during their lifetime respectively. Ever using cannabis and the frequency of use within lifetime users were not associated with rs1409568 genotype (C-allele carrier vs TT; only 2 individuals with CC genotype). However, the C allele, which was associated with increased likelihood of cannabis dependence in the meta-analysis, was associated with increased gray matter volume in the right hippocampus (2.13% greater than TT individuals; Cohen’s d = 0.62, maximal voxel p-fwe = 0.007; Bonferroni p-value for 6 a priori regions = .008; Supplemental Figure S7A). This association remained unchanged when cannabis use was included as a covariate in the analysis (Cohen’s d = 0.62, maximal voxel p-fwe = 0.008). Other regions previously associated with cannabis use (i.e., left hippocampus and bilateral amygdala and ventral striatum) showed no relationship with the SNP. Finally, ever having used cannabis was associated with increased volume in a cluster in the left hippocampus (3.18% greater in ever versus never users; Cohen’s d = 0.39, maximal voxel p-fwe = 0.002; Supplemental Figure S7B). No significant volumetric differences were observed for the right hippocampus, where the SNP exerted main effects, nor was the cluster in the left hippocampus in the same region as the cluster in the right hippocampus to which rs1409568 was associated. Lastly, rs1409568 was not associated with hippocampal volume in an independent large meta-analysis (p=0.33; N=12, 516)(49).

DISCUSSION

This study identified a genomewide significant locus on chromosome 10 for cannabis dependence diagnosis in subjects of European descent. To date, only one other (Table 1) study (8) identified genomewide significant loci for cannabis dependence criterion count. The novel locus identified in the present study included a representative SNP, rs1409568, which showed modest evidence for replication in the AA, but not EA, participants from the independent Yale-Penn sample that was part of the only other study with genomewide significant SNPs. The lack of replication in the EA component of Yale-Penn may reflect lower power (i.e., fewer cases than the AA component, or higher minor allele frequency in AA than EA) or ascertainment differences. It is also noteworthy that patterns of LD for the SNPs in Table 3 differ across CEU and ASW populations (based on 1000 Genomes data; Supplemental Figure S8)(50); replication that was noted in the AAs was present in spite of these differences. Nonetheless, associations in the Yale-Penn EA participants were in the same direction as the current meta-analysis.

The genomewide significant chromosome 10 SNPs represent a single LD signal, and are located in a region that is primarily intergenic. However, based on GENCODEv19 (51) annotation, there are multiple genes within the regulatory domain spanning these SNPs. While 5 of these 12 genes are expressed in brain-derived tissues (Supplemental Figure S4), none of the genomewide significant SNPs served as eQTLs for expression of these genes in GTEx, which includes modestly sized samples for a variety of brain tissue, nor in the larger Common Mind Consortium data, which includes 279 dlPFC samples. We found no evidence in the literature for the role of the genes within the regulatory domain in the etiology of addiction-related or other behavioral phenotypes.

One genomewide significant SNP, rs1409568, appears to be located within an active enhancer (52). This finding is consistent with a recent study that reported modest enrichment of H3K27ac marks for a variety of complex traits (e.g., Crohn’s disease)(53). Importantly, there is growing evidence that intergenic genomewide significant loci are disproportionately overrepresented in regulatory regions, such as enhancers (54–56). For example, functional partitioning of SNP-attributable heritability for 11 complex traits found that DNase1 hypersensitivity sites were 1.6- and 5.1-fold enriched in genotyped and imputed data respectively, with enhancers being the most common subcategory, representing 31.7% of total SNP heritability and 9.8-fold enrichment (54).

Importantly, rs1409568 is predicted to bear active enhancer marks in several brain-derived tissues that are critical to addiction, most notably the dlPFC and the cingulate and angular gyri, which play a major role in the development of addictive behaviors, particularly in the regulation of executive control and attentional bias (57). These in-silico findings imply that the C allele is associated with reduced or no binding of several homeodomain-containing (58) developmentally relevant transcription factors, with some difference scores (e.g., POU6F2) being substantial (>8.0). These genes and their products have been variously implicated in embryogenesis and in cell-type specific pathways of differentiation, particularly in visual systems (59–61), but have not been related to behavioral traits thus far.

There was also nominal evidence that rs1409568 genotype was associated with changes in CpG methylation of TIAL1. C allele carriers, on average, had lower methylation scores than T homozygotes. There is no published evidence for a role of the RNA-binding protein encoded by this gene in addictive processes.

In an independent sample, the C allele of rs1409568 was also associated with a modest increase in right hippocampal volume (2.13%) but not with cannabis use itself. The hippocampus has been implicated in addiction (47), including volumetric differences that have been observed in chronic cannabis users (14;15). This, in addition to tentative evidence for the role of rs1409568 as a potential enhancer in the middle hippocampus (Figure 2), indicates that this SNP may regulate neural effects that are central to the development of addictions. The lack of association between cannabis use and genotype is not surprising given the vanishingly low number of problem users (e.g., 12 individuals with cannabis abuse) in the DNS sample.

Cannabis use itself was associated with a modest increase (3.18%) in left (but not right) hippocampal volume. This finding contradicts prior studies that have linked chronic, but not occasional, cannabis use to decreases (not increases) in hippocampal volume. We speculate that the association between cannabis use and increased hippocampal volume may be due to the nature of DNS, which includes casual, non-problem users who are also likely enriched for other factors that might protect against progression to problem use (and against hippocampal deficits). In support of this, we found that cannabis users in DNS were more likely to represent higher socioeconomic status (t=3.70, p<0.001) and even showed modest increases in digit-span performance (t=2.50, p =.013), an index of working memory suggesting that cannabis users in DNS may be characterized by adaptive factors that protect them from progression to problem use. Therefore, if previously documented associations between cannabis use and smaller hippocampal volumes are a consequence of chronic exposure to cannabis, then we would not expect to see these reductions in the DNS.

The minor allele of rs1409568, which was more common in cannabis dependent cases in the meta-analysis, was associated with increased hippocampal volume. This finding is also inconsistent with the hypothesis that liability to heavy cannabis use should relate to decreased hippocampal volume. There are, at least, two plausible explanations for our observation of the opposite association. First, it is possible that the association between rs1409568 and hippocampal volume is independent of its association with cannabis dependence in the meta-analysis. While such a pleiotropic effect adds encouraging evidence favoring a role of rs1409568 in neural regions typically associated with addiction, and augments its functional plausibility, it does not help reconcile the mechanism by which rs1409568 might influence liability to cannabis dependence. Second, the association between rs1409568 and hippocampal volume did not replicate in the large ENIGMA meta-analysis. This raises the possibility that the association is a false positive in DNS and suggests that caution is warranted in its interpretation.

It is also noteworthy that the current study did not replicate previously noted associations for cannabis use (13) or dependence (8). These are not unexpected. For cannabis use, our sample excluded individuals who had never used cannabis, thus limiting our ability to detect loci associated with initiation of cannabis involvement. Our lack of replication of one prior locus identified for cannabis dependence in EAs (rs77378271) might further underscore differences between our European samples and those comprising Yale-Penn. A full meta-analysis of these datasets might yield additional novel loci.

While no single SNP was genomewide significant for the count of DSM criteria, gene-level testing identified MEI1 (meiotic double-stranded break formation protein 1). Relative to other tissues, MEI1 is more robustly expressed in the testes and variants in the gene have been associated with azoospermia due to early and complete meiotic arrest (62). In parallel, there is compelling epidemiological and biological support for the relationship between prolonged/heavy cannabis use and male reproductive health, including fertility. Weekly cannabis use has been associated with a 28–29% reduction in sperm concentration and count (63). The endocannabinoid system actively participates in the regulation of male fertility (64), including by promoting meiosis via CB2 activation (65). Therefore, the possibility of shared genetic pathways to male fertility and heavy cannabis use might provide a plausible alternative to more causal explanations. However, we are not aware of any prior studies that link MEI1 to cannabis use or addiction.

Some limitations are noteworthy. First, despite aggregating across several large datasets, our meta-analytic sample was relatively underpowered to detect small effects and also, for analyses that would allow us to estimate genetic correlations between cannabis dependence and other traits (e.g. cigarettes per day (68) for which genomewide summary statistics are available. Such calculations typically rely on unrelated cases and controls and our study included two samples with complex pedigree structures. Second, we did not have adequate numbers of AA participants for a full examination of loci identified in Sherva et al (8). In EAs, the only SNP associated at genomewide significant levels in Sherva et al was rs77378271 (CSMD1). In the current study, rs77378271 shows some evidence for independent association with cannabis dependence in COGA-cc (p=5.3E-3); however, the meta-analytic p-value was not significant, with indication of heterogeneity across the samples included in the present meta-analysis. We anticipate that additional data on cannabis dependence in both EA and AA participants will be available in the future. Finally, the minor allele frequency for rs1409568 (and related genomewide significant SNPs) was <10% across cases and controls from each sample.

We identified a new genomewide significant locus on chromosome 10 that was associated with vulnerability to cannabis dependence in European ancestry individuals. One of the representative SNPs, rs1409568, showed promising epigenetic evidence and might also contribute to variation in hippocampal volume, which has been related to risk for and resilience to psychiatric disorders, including addictions. Replication, however, was limited to a subset of AA, but not EA, individuals and analyses in the DNS contradicted prior findings for hippocampal volume and did not extend to a broader meta-analysis of hippocampal volume. Therefore, the identification of this chromosome 10 locus should be viewed as preliminary. Future work that aggregates additional cannabis dependent cases and controls, would allow for the detection of smaller effect sizes and a more thorough investigation of comparability of loci across population groups. This is critical, as genomic research into cannabis involvement has lagged behind that of other drugs, despite the pressing public health significance of the problem. Continuing to identify risk factors, both genetic and environmental, that are associated with cannabis dependence is a public health priority, as understanding the genetic etiology of cannabis use disorders can ultimately help to identify individuals who are at greatest risk of the disorders and enhance efforts aimed at prevention and personalizing pharmacotherapy among affected individuals.