Linkage analysis followed by association show NRG1 associated with cannabis dependence in African-Americans

Abstract

Background

A genetic contribution to cannabis dependence (CaD) has been established, but susceptibility genes for CaD remain largely unknown.

Methods

We employed a multi-stage design to identify genetic variants underlying CaD. We first performed a genomewide linkage scan for CaD in 384 African-American (AA) and 354 European-American (EA) families ascertained for genetic studies of cocaine and opioid dependence. We then conducted association analysis under the linkage peak, first using data from a genomewide association study from the Study of Addiction: Genetics and Environment (SAGE), followed by replication studies of prioritized single nucleotide polymorphisms (SNPs) in independent samples.

Results

We identified the strongest linkage evidence with CaD (lod=2.9) on chromosome 8p21.1 in AAs. In the association analysis of the SAGE sample under the linkage peak, we identified one SNP (rs17664708) associated with CaD in both AAs (minor allele frequency (MAF) = 0.02, OR=2.93, 95% CI=1.47–5.85, P=0.0022) and EAs (MAF=0.096, OR=1.38, 95% CI=1.05–1.81, P=0.02). This SNP, located at NRG1, a susceptibility gene for schizophrenia, was prioritized for further study. We replicated the association of rs17664708 with CaD in an independent sample of AAs (MAF=0.013, OR=2.81, 95% CI=1.23–6.45, P=0.0068). The joint analysis of the two AA samples demonstrated highly significant association between rs17664708 and CaD with adjustment for either global (OR=2.34, 95% CI=1.42–3.85, P=0.00044) or local ancestry (OR=2.33, 95% CI=1.39–3.91, P=0.00075).

Conclusions

Our study shows that NRG1 is probably a susceptibility gene for CaD, based on convergent evidence of linkage and replicated associations in two independent AA samples.

Introduction

Cannabis is the most commonly used illicit substance in the world with 143–190 million people in 2007 having used the drug at least once worldwide (1). An estimated 14.4 million Americans aged 12 or older reported cannabis use over the past month (2). Of these individuals, ~7% develop cannabis dependence (CaD) defined by DSM-IV criteria (3). With the expansion of legalization in the United States (generally for “medical” use), availability and use of cannabis is rising. Cannabis use is often accompanied by dependence on alcohol and other drugs (4), and is associated with serious consequences, including cognitive and psychomotor impairments (5, 6). The use of cannabis is associated with roughly twofold increased risk of schizophrenia; there is interindividual variability in susceptibility to cannabis-induced psychosis that could be, in part, genetic in origin (7, 8). Thus, it is important to identify factors that influence individual vulnerability to the development of CaD.

Family and twin studies have shown that CaD has an important genetic component. The heritability of cannabis abuse or dependence was estimated to be 45–78% (9). Genomewide linkage studies (GWLS) and candidate gene association studies have identified a list of possible chromosomal risk regions and candidate genes for cannabis use disorders (9–16). For example, linkage studies have identified genomic regions harboring candidate genes with biological relevance, such as the monoacylglycerol lipase gene (MGLL) on chromosome 3 and the gamma-aminobutyric acid receptor subunit alpha-2 gene (GABRA2) on chromosome 4 (14). The cannabinoid receptor gene (CNR1) and several other genes (CRN2, FAAH, and MGLL), which are specific to the endogenous cannabinoid system, have been selected for candidate gene association studies, although the results were largely inconclusive (14). Recently, a genomewide association study (GWAS) for CaD was conducted in 708 individuals with DSM-IV CaD cases and 2346 cannabis-exposed non-dependent controls, using a GWAS dataset from the Study of Addiction: Genetics and Environment (SAGE) (17). However, no results achieved genomewide significance in this study. Despite the effort that has been made in gene mapping for CaD, genetic factors underlying CaD susceptibility remain largely unknown.

With the growing evidence for the role of rare variants and copy number variation (CNV) in psychiatric disorders (18–20), linkage analysis remains a useful approach to gene discovery. An adequately powered linkage study can detect diverse kinds of genetic polymorphism that segregate in families, including common variants, multiple rare variants within one locus, and inherited CNVs. The apparent failure to identify association under linkage peaks could in part be attributable to the fact that often, only common variants are examined under the linkage peak, whereas the linkage signal could be caused by multiple rare variants with higher penetrance (21–23).

The current study employed a multi-stage design using a linkage scan, a GWAS dataset, and replication in independent samples to identify genetic variants associated with CaD. Specifically, the objectives of the current study were to: 1) conduct a genomewide linkage scan to detect genetic loci influencing CaD risk in African-Americans (AAs) and European-Americans (EAs); 2) assess the genetic association between CaD and single nucleotide polymorphisms (SNPs) under the strongest linkage peak using the GWAS dataset from SAGE; and 3) replicate prioritized SNPs in independent AAs and EAs from our samples.

Subjects and Methods

Study samples

The basic demographic information for the three samples involved in each stage of the study is summarized in Table 1. The following section provides details of the sample recruitment and characteristics for each sample set. Written informed consent was obtained from all subjects; the IRB at each recruitment site approved the study, and NIAAA and NIDA issued certificates of confidentiality for the work.

TABLE I.

Basic demographic information for the three samples included in the current study

	African-Americans	European-Americans
Linkage sample
No. of families	384	355
No. of genotyped individuals	1022	874
No. of pedigrees with 1 CaD	134	136
No. of pedigrees with ≥ 2 CaD	40	54
Age (years ±SD) (CaD)	38.6±6.5	33.5±9.6
Males (%) (CaD)	57.8	60.4
SAGE sample
CaD	275	422
Males (%)	66.5	67.3
Age (years ±SD)	38.7±7.8	34.8±8.5
Control	401	1049
Males (%)	38.2	33.1
Age (years ±SD)	39.8±7.5	39.1±9.7
Replication Sample
CaD	758	568
Males (%)	67.7	70.4
Age (years ±SD)	40.1±8.7	34.7 ±10.8
Control	280	318
Males (%)	26.4	40.3
Age (years± SD)	37.3±13.2	38.7±13.9

CaD, cannabis dependence

Linkage scan sample

Subjects were originally ascertained for genetic studies of cocaine dependence (CD) and opioid dependence (OD) using the affected sibling pair (ASP) linkage approach (24, 25). The recruitment procedure has been previously described in detail (26–28). Briefly, there were four recruitment sites: University of Connecticut Health Center, Yale University School of Medicine, Medical University of South Carolina, and McLean Hospital. Families were selected on the basis of having at least two siblings affected with either cocaine and/or opioid dependence. The distribution of family numbers recruited at each site is presented in Supplementary Table 1. Cannabis use played no role in proband selection or pedigree extension. We evaluated these subjects with the Semi-Structured Assessment for Drug Dependence and Alcoholism (SSADDA) (29, 30), a polydiagnostic instrument that assesses a range of psychiatric diagnoses, including DSM-IV CaD. Probands with an axis I clinical diagnosis of a major psychotic disorder such as schizophrenia or schizoaffective disorder were excluded from participation. Subjects were classified as AA or EA on the basis of a Bayesian model-based clustering method with ancestry informative genetic markers using STRUCTURE (31, 32), as described previously (28).

Study of Addiction: Genetics and Environment (SAGE)

We obtained the individual level SAGE GWAS dataset from the database of Genotypes and Phenotypes (dbGaP). SAGE aims to identify genetic risk factors and the interplay of genes and environmental factors for addiction. Cases and controls were selected from three large, complementary cohorts: the Collaborative Study on the Genetics of Alcoholism (COGA), the Family Study of Cocaine Dependence (FSCD), and the Collaborative Genetic Study of Nicotine Dependence (COGEND), all of which have been previously described (33–36). The current study included 4036 unrelated self-reported AA (1297 in total including 275 CaD and 422 healthy controls) or EA (2740 in total including 401 CaD and 1049 healthy controls) subjects. Lifetime CaD was defined in accordance with the DSM-IV diagnosis. Controls used for association analysis in the current study were defined as subjects without dependence on any substances, including cannabis, alcohol, cocaine, opioid, nicotine, and other substances.

Replication sample

Subjects were recruited for participation in studies of the genetics of CD, OD, and alcohol dependence (AD) from the communities around four sites listed above for linkage scan sample plus the University of Pennsylvania. The number of samples recruited at each site for the current study is presented in Supplementary Table 1. Part of the set of samples genotyped at the replication stage overlapped with the samples used for linkage analysis. To obtain an independent sample set for replication study, the overlapping samples (72 CaD cases) were excluded from analysis in the replication stage. Subjects were interviewed with the SSADDA and the diagnosis of CaD was derived based on DSM-IV diagnostic criteria. Most controls were recruited at the same recruitment sites (excluding McLean) and were screened to exclude those with a diagnosis of DSM-IV substance abuse or dependence, and a major Axis I psychiatric disorders. All subjects included in the current study are self-reported AA or EA. Subjects were re-classified as AA or EA on the basis of 41 ancestry informative genetic markers (AIMs) using STRUCTURE, as described previously (20, 37). Among the subjects included in current study, 3% of the subjects reporting to be of AA descent clustered in the EA group, and 2% of subjects reporting to be EA clustered in the AA group.

Genotyping and Quality Control

Linkage analysis

Genotyping and quality control (QC) for linkage analysis have been previously described in detail (28). Briefly, 1630 subjects were genotyped at the Center for Inherited Disease Research (CIDR) for the 6,008 SNPs Illumina Human Linkage IVb Marker Panel. An additional 266 subjects were genotyped at the Yale Keck Center with the 6,090 SNP Illumina Infinium-12 Human Linkage Marker Panel. We limited our analyses to 4,518 autosomal SNPs available in both panels. After QC (genotyping rate ≥ 95%, minor allele frequency (MAF) ≥0.1, and HWE P ≥0.01), 4,133 and 4,395 autosomal SNPs were retained for analysis in AAs and EAs, respectively. Mendelian inconsistencies and potential genotyping errors were identified and set as missing data using PedCheck (38) and Merlin (39) programs. We used the Pedigree RElationship Statistical Test (PREST) (40) to verify family relationships, which showed pedigree errors in two AA families and five EA families. Of these, the relationships in one AA family and five EA families were corrected based on the shared identical by decent (IBD) patterns and the re-assigned family relationships were verified by PREST. One AA family relationship could not be resolved and the family was excluded from further analysis.

SAGE dataset

SAGE samples were genotyped on the ILLUMINA Human 1 M platform at CIDR. We included 4036 unrelated self-reported AA (1297) or EA (2740) subjects (60 duplicate genotype samples were excluded from analysis). We used PLINK software to perform basic data cleaning steps before analysis (41). After QC(sample call rate ≥ 97%; SNP call rate ≥ 95%; MAF ≥ 0.005 in controls; HWE P ≥ 0.00001 in controls), a total of 1297 (2740) unrelated subjects and 953,258 (888,092) autosomal SNPs for AAs (EAs) were available for further analysis. To obtain a more genetically homogeneous sample and correct for population stratification in the association analysis, we computed principal components (PC) using the EIGENSOFT package (42). Specifically, 172,891 pruned SNPs common to AA and EA samples and in low linkage disequilibrium (LD) (genotypic correlation < 0.5) with one another, were fed into EIGENSOFT. The top two PCs of AA and EA samples and with the Phase II Hapmap CEU and YRI samples are shown in Supplementary Figure 1. Outliers were defined as subjects whose ancestry was at least 3 standard deviations from the mean of the two largest PCs. This step removed 33 AAs and 127 EAs, retaining 1264 AAs and 2613 EAs in the final cleaned dataset.

Replication study

The prioritized SNP, rs17664708, was genotyped for a replication study in 2543 AAs and 2042 EAs (ethnicity verified or re-classified by STRUCTURE using 41 mostly short tandem repeat (STR) AIMs (20, 37). Genotyping was performed with a fluorogenic 5′ nuclease assay method (TaqMan technique), using the ABI 7900HT real time PCR system (ABI, Foster City, CA). The missing rate for genotyping was 0.03 in the replication sample. For genotyping quality control, 8% of samples were re-genotyped with 100% concordance.

A subset of the replication sample (n=931, including 672 cases and 259 controls) of AAs were also genotyped by the Illumina Omni1-Quad platform at CIDR (690 subjects) or the Yale Keck center (241 subjects) for our ongoing GWAS study of alcohol, cocaine, opioid, and nicotine dependence. Genotype information from these subjects were used to control for the potential population stratification in association test in AAs (The prioritized SNP rs17664708 was not included on the Illumina Omni1-Quad chip). After QC (SNP call rate ≥ 95%, sample call rate ≥ 97%, MAF ≥ 0.005 in controls, HWE P ≥ value 0.00001 in controls), 2745 AIMs across the whole genome and 188 SNPs located at NRG1 were extracted to estimate the global and local ancestry in this subset of AA samples, respectively.

Statistical analysis

Linkage analysis

We used Merlin (39) to perform the linkage scan using a nonparametric allele-sharing model. Allele frequencies were estimated by counting all genotyped individuals. The Kong and Cox linear allele-sharing model (43) was used to estimate the lod score. To minimize the inflation of linkage signals caused by marker-marker LD, we grouped SNPs by LD into clusters using the Merlin “--rsq” option (44). Analyses were repeated with r² thresholds of 0.05, 0.2, and 0.3 to evaluate the robustness of the linkage results. We assessed the thresholds for autosomal genomewide suggestive and significant linkages, and the autosomal genomewide empirical significance of an observed lod score, by 1000 computer simulations, as described previously (28).

Global ancestry estimation in AAs

Spurious association between a marker and a phenotype can arise from population stratification, especially in admixed populations such as AAs. To account for the effect of population stratification in AAs, we estimated the individual global ancestry using STRUCTURE program and included it as a covariate in the association analysis. To obtain a more consistent estimate of individual global ancestry for SAGE and our replication AA samples, we selected 2475 AIMs that were genotyped in both the SAGE AA sample and a subset of our replication sample (931 subjects). These AIMs were common to a reported AIMs panel for AAs based on a subset of SNPs on the ILLUMINA Human 1 M platform (45). The log likelihood of each analysis at varying numbers of assumed population groups (k) was estimated from the average of 3 independent runs (5,000 burn in and 5,000 iterations). As expected, the results favored a two-ancestry population model. The average proportion of European ancestry was 0.186 in the SAGE AA sample and 0.166 in our replication AA sample.

Local ancestry estimation in AAs

Because the global ancestry information obtained across the whole genome may not reflect the variation of ancestry at the tested genomic locus, methods that adjust the global ancestry to control population stratification may be insufficient. However, methods that conditioned on local ancestry at the tested locus more fully account for the confounding effect of hidden population structure (46). Therefore, we estimated the local ancestry at the NRG1 locus using the HAPMIX program (47), and the overall ancestry across SNPs at the NRG1 locus was included as a covariate in the association analysis to control for the local population stratification, (i.e., the estimated ancestry of the specific genomic region under consideration). Briefly, we downloaded the phased YRI and CEU data from Hapmap Phase II as the parental reference haplotype input for HAPMIX. After QC and filtering the monomorphic SNPs in the phased YRI and CEU data, 188 SNPs that were located at the NRG1 gene and were genotyped in both the SAGE AA sample and in the subset of the replication AA sample (931), were used to estimate local ancestry at the NRG1 locus. The estimated average European ancestries at the NRG1 locus were 0.194 and 0.171 for the SAGE and the replication AA samples, respectively, which approximated the values of the global European ancestry proportion in each sample.

Association analysis

The association between each SNP and the binary trait was estimated in a multivariate logistic regression framework under a log-additive genetic model. We used PLINK for the SNP-trait association test in the region of the linkage peak in the SAGE sample, with sex, age and the top 10 PCs as covariates. For the replication analysis and joint data analysis stages for rs17664708, we included sex, age, and global or local ancestry estimates as covariates where appropriate, and analysis was performed using the R package “SNPassoc” (48).

Results

Genomewide linkage analysis for CaD

Empirical genomewide thresholds for suggestive and significant linkage for non-parametric linkage analyses in our family dataset were determined based on 1000 simulations. The thresholds for genomewide suggestive and significant linkage in AAs (EAs) were 1.79 (1.76) and 3.23 (3.22), respectively.

The genomewide non-parametric linkage results for AAs and EAs are presented in Figure 1. The strongest linkage signal was identified on chromosome 8p21.1 at 54.9 cM (lod=2.9, pointwise P = 0.00013, empirical genomewide P = 0.097) in AAs, and weak linkage evidence was detected at the same location in EAs (lod=0.62, pointwise P = 0.05). We identified a second genomewide suggestive linkage peak in AAs on chromosome 14 with a peak lod of 2.26 at 89.9 cM, though no linkage evidence was detected at this peak region in EAs. In EAs, only one region showed genomewide suggestive linkage on chromosome 7 at 107 cM (lod=1.85), where weak linkage signal was observed in AAs (lod=0.19). The locations and values of these suggestive linkage peaks did not change when analyses were repeated using different r² values to group SNPs into clusters.

Figure 1.

Genomewide non-parametric linkage results in AAs and EAs. The black and red virtue lines represent the empirical genomewide suggestive (significant) linkage thresholds for AAs and EAs, respectively.

Association analysis under the linkage peak using SAGE GWAS dataset

Utilizing the SAGE GWAS dataset, we examined the genetic association between CaD and 4853 SNPs under the strongest linkage peak from our family sample (lod > 2, from 48.9 cM to 65.2 cM) on chromosome 8. All of these SNPs passed QC and were tested for genetic association with adjustment for sex, age and the top 10 PCs in AAs and EAs separately. Eleven SNPs were nominally significantly associated with CaD in both AAs and EAs (Table 2). One SNP (rs17664708), which is relatively rare in AA controls (MAF = 0.02) but common in EA controls (MAF=0.096), showed consistent evidence for association in both AAs (OR=2.93, 95% CI=1.47–5.85, P=0.0022) and EAs (OR=1.38, 95% CI=1.05–1.81, P=0.02). This SNP is located at NRG1, which has been previously implicated in the risk for schizophrenia. Considering the high degree of commorbidity between cannabis use and schizophrenia (7, 8, 49–53), this SNP (rs17664708) was prioritized for further study.

TABLE 2.

Summary of association analysis results under the linkage peak for SNPs nominally significantly associated with CaD in both AAs and EAs using SAGE GWAS dataset

SNP	Position1	Function	Gene	African-Americans (275 cases/401 controls)						European-Americans (422 cases/1049 controls)
				MA	MAF (Cases)	MAF (Conts)	OR	95% CI	P	MA	MAF (Cases)	MAF (Conts)	OR	95% CI	P
rs931874	25301315	Intronic	DOCK5	T	0.19	0.24	0.74	0.56–0.99	0.043	T	0.38	0.36	1.20	1.00–1.43	0.05
rs6558049	28148960	Intergenic	-	A	0.23	0.18	1.41	1.05–1.89	0.021	A	0.026	0.045	0.54	0.33–0.90	0.017
rs7015363	28159947	Intergenic	-	G	0.23	0.18	1.36	1.02–1.81	0.037	G	0.026	0.046	0.53	0.32–0.88	0.013
rs7011660	30405716	Intronic	RBPMS	A	0.19	0.23	0.74	0.56–0.98	0.038	A	0.19	0.22	0.77	0.62–0.95	0.016
rs7833229	30542430	Intronic	RBPMS	A	0.024	0.042	0.45	0.22–0.90	0.025	A	0.11	0.14	0.74	0.57–0.97	0.031
rs17664708	32556559	Intronic	NRG1	T	0.049	0.020	2.93	1.47–5.85	0.0022	T	0.13	0.096	1.38	1.05–1.81	0.020
rs12155970	34064572	Intergenic	-	C	0.065	0.034	2.20	1.26–3.86	0.0057	C	0.093	0.062	1.51	1.09–2.09	0.012
rs2593095	38587762	Intergenic	-	T	0.54	0.48	1.27	1.00–1.61	0.049	T	0.33	0.39	0.77	0.64–0.93	0.0056
rs6989042	40301460	Intergenic	-	G	0.30	0.23	1.32	1.01–1.73	0.041	G	0.23	0.20	1.25	1.01–1.55	0.037
rs7826645	40924370	Intergenic	-	A	0.13	0.16	0.68	0.49–0.96	0.026	A	0.16	0.11	1.50	1.17–1.93	0.0016
rs12547609	41091390	Intergenic	-	C	0.15	0.21	0.68	0.50–0.93	0.015	C	0.21	0.17	1.31	1.05–1.63	0.018

CaD, cannabis dependence; SNP, single nucleotide polymorphism; MA, minor allele; MAF, minor allele frequency; OR, odds ratio; CI, confidence interval; Conts, Controls;

¹Coordinate based on Genome Build 36.3. Association tests are adjusted by sex, age and the top 10 principal components. Prioritized SNP (rs17664708) for follow up study is shown in bold.

Replication of the prioritized SNP in independent AAs and EAs

We tested the association between rs17664708 and CaD in independent AAs (758 CaD and 280 healthy controls) and EAs (568 CaD and 318 healthy controls) from our own sample. The genotype distribution of this SNP was in HWE in both cases and controls for AAs and EAs (In AAs, P=1 for both cases and controls; In EAs, P=0.17 and P=0.80 for cases and controls, respectively). We replicated the association of rs17664708 with CaD in AAs after adjusting for sex and age (OR=2.81, 95% CI=1.23–6.45, P=0.0068), but not in EAs (OR=1.04, 95% CI=0.77–1.40, P=0.82). The MAF of rs17664708 in the replication AA sample was 0.013 and 0.038 in controls and cases, respectively, which approximated the values from the SAGE AA sample.

Joint analysis in AAs

Considering that the strongest linkage signal on chromosome 8 was observed only in AAs, we performed a joint analysis for the association of rs17664708 to CaD in AAs including sex, age, and global or local ancestry as covariates (Table 3). To permit comparison, analysis results with adjustment for global or local ancestry in each AA sample are also listed in Table 3. The association results remained significant after adjusting for either global (P=0.00044) or local ancestry (P=0.00075), which argues against the possibility that the significant association was caused by population stratification in AAs.

TABLE 3.

Association analysis for rs17664708 with CaD in the joint samples of AAs

Datasets (Cases/Controls)	MAF (case)	MAF (cont)	Association test adjusted by
			Sex and age			Sex, age and global ancestry			Sex, age and Local ancestry
			OR	95% CI	P	OR	95% CI	P	OR	95% CI	P
SAGE (275/401)	0.049	0.020	2.60	1.37±4.92	0.0028	2.71	1.38±5.32	0.0031	2.73	1.35±5.51	0.0043
Replication (672/259)1	0.038	0.013	2.70	1.17±6.24	0.010	2.15	0.92±5.03	0.059	2.17	0.91±5.16	0.063
Joint (947/660)	0.042	0.018	2.45	1.49±4.02	0.00017	2.34	1.42±3.85	0.00044	2.33	1.39–3.91	0.00075

CaD, cannabis dependence;

¹To estimate the local ancestry at NRG1 locus and the global ancestry in the replication AA sample, only a subset of individuals (931) who were genotyped by Illumina Omni1-Quad platform were included.

Discussion

The current study demonstrated how linkage analysis could inform genetic association studies and lead to discovery of a rare variant in NRG1 associated with CaD in AAs. We first conducted a dense GWLS for CaD in small nuclear families recruited on the basis of ASPs with either CD and/or OD. The strongest evidence for linkage in our family sample was observed on 8p21.1 (lod =2.9) in AAs. To our knowledge, this region has not been previously reported by linkage studies of any substance dependence disorder. However, chromosome 8p22-p21, which overlaps our linkage signal, has been repeatedly implicated in several neuropsychiatric disorders including schizophrenia, bipolar affective disorder, and major depression (54). Using the SAGE GWAS dataset, we found that rs17664708 located at NRG1 under the linkage peak was associated with CaD in both AAs and EAs. The association was further replicated in an independent AA sample. Our rigorous QC and statistical analysis adjusting for both global and local ancestry argues against the possibility that the significant association between the rs17664708 and CaD is from the effect of population stratification in AAs.

None of the SNPs that were nominally significant (P ≤ 0.05) in both AAs and EAs can survive multiple testing correction in the discovery stage of association analysis under the linkage peak using the SAGE dataset. However, our prioritization strategy, designed to order SNPs for follow up studies, relies on both statistical evidence and prior knowledge. Accordingly, the SNP rs17664708 located at NRG1 was prioritized for further studies. NRG1 is a well established susceptibility gene for schizophrenia in many populations supported by genetic linkage, association studies, and meta-analysis (55–57). There are at least three lines of prior knowledge which prompt us to consider NRG1 as a candidate gene for CaD as well. First, elevated rates of cannabis use have repeatedly been reported among individuals with schizophrenia (49, 50), and epidemiological studies suggest that frequent cannabis use is associated with about 2-fold increased risk for developing schizophrenia and related disorders (7,8,51–53). The significant commorbidity between cannabis use and schizophrenia may be attributable in part to a shared underlying genetic component for CaD and schizophrenia. Second, the neurobiology of CaD and schizophrenia overlap; for example, the mesolimbic pathway is heavily implicated in the neurobiology of CaD (and other substance dependence) and schizophrenia (58, 59). Third, animal studies have provided evidence supporting the role of NRG1 in CaD. NRG1 heterozygous mice have increased sensitivity to the acute neurobehavioral effects of cannabinoids (60, 61); NRG1 modulates the development of tolerance to cannabinoids in mice (62).

The gene NRG1 encodes neuregulin 1, a pleiotropic growth factor that is important in nervous system development and function. It has been implicated in the modulation of many processes of neural development, including neuronal migration, synapse formation, synaptic plasticity and neuronal survival (63). The SNP rs17664708 we reported here is located at the intron region of NRG1. We did not find any evidence supporting the functional role of this SNP involved in CaD. As it is the case for most genetic association studies, the associated SNP may not be the causal SNP but could represent a tag SNP which is in linkage disequilibrium with the surrounding causal variants. Further studies using a higher density SNP panel and deep sequencing technology are needed to fully characterize the genetic architecture of NRG1 and pinpoint the functional variants that could be involved in CaD.

Cannabinoids have been shown to produce greater behavioral effects in female than in male rats (64, 65). There is also evidence showing that male Nrg1 HET mice being more sensitive to the acute effects of the psychotropic cannabis constitute Δ9-tetrahydrocannabinol (THC) which is not observed in females (66). We investigated whether there is a sex-specific effect for the rs17664708 in CaD susceptibility by sex-stratified analysis. We observed a stronger effect for rs17664708 in females (OR=3.08, 95% CI=1.56–6.07, P=0.0007) than in males (OR=1.88, 95% CI=0.95–3.74, P=0.055). The stronger effect of rs17664708 in females provides further evidence for the interactions between NRG1 and sex in CaD susceptibility.

Samples included in the current study were ascertained for DSM-IV cocaine or opioid dependence (linkage sample), and AD (SAGE), and CD, OD or AD (in the replication sample). Therefore, the linkage and association signals detected for CaD in AAs could be attributable to other substance dependence (SD). We examined this alternate explanation for the findings in several ways. First, we examined the linkage signals across other SD traits (AD, CD, OD, and nicotine dependence (ND)) in the same family dataset (Supplementary Figure 2). The only linkage signal that we found with a lod greater than 1 in the chromosome 8 region reported above for CaD was a modest signal (lod = 1.4) for ND at 41 cM (i.e., ~14 cM away from the CaD linkage peak). Second, we tested the association of rs17664708 with other SD in each AA sample and the joint datasets (Supplementary Table 2). In the joint analysis, we observed highly significant associations between rs17664708 and CD (P=0.0023), AD (P=0.0076), ND (P=0.0026) and OD (P=0.00018). However, significant associations disappeared when subjects affected with CaD were excluded from AD (P=0.3), ND (P=0.069) and CD (P=0.053). Nonetheless, the association remained highly significant for OD when the CaD cases were excluded (P=0.0013), which suggests that rs17664708 could also be a risk variant for OD in AAs independent of CaD. Finally, to remove the confounding effect of OD, we tested the association between rs17664708 and CaD by either controlling for OD status in the regress model or excluding participants who are OD. The results remain significant in the joint datasets by OD status adjusted analysis (OR=2.03, 95% CI=1.17–3.55, P=0.011) or after excluding OD participants (OR=2.47, 95% CI=1.48–4.11, P=0.0003) arguing against the possibility that the significant associations between rs17664708 and CaD arise solely from OD. The association results from other SD might indicate that the association is not specific for CaD and could reflect a shared liability between different substances. The role of NRG1 in other SD remains to be further investigated.

In summary, our study shows that NRG1 is probably a susceptibility gene for CaD, based on convergent evidence of linkage and replicated associations in two independent samples of AAs. Further studies using a high density SNP panel or deep sequencing are necessary to confirm the role of NRG1 in CaD.