If you don't remember your password, you can reset it by entering your email address and clicking the Reset Password button. You will then receive an email that contains a secure link for resetting your password
If the address matches a valid account an email will be sent to __email__ with instructions for resetting your password
Department of Psychiatry, Yale University School of Medicine, New HavenGenetics and Neurobiology, Yale University School of Medicine, New HavenVA Connecticut Healthcare System, West Haven Campus
Human cannabinoid receptor 1 (CB1), which is encoded by the CNR1 gene, may play a role in the development of substance dependence (SD). Following initial reports of association of CNR1 with SD, we studied multiple markers at this locus in a large case–control sample.
Methods
Ten CNR1 markers and 38 ancestry-informative markers were genotyped in 451 healthy control subjects and 550 SD (AD and/or DD) patients (including European Americans [EAs] and African Americans [AAs]). Common confounding effects on association analysis of population stratification and admixture, age, and sex were controlled for using regression analysis. Disease risk and protective alleles were fine-mapped using a linkage disequilibrium measure (δ).
Results
In EAs, risk for each SD subtype significantly increased with the number of “G” alleles at rs6454674 (single nucleotide polymorphisms [SNP]3). SNP3^G+ (the genotypes containing a G allele) and SNP8^T/T genotypes had significant interaction effects (p = .0003 for comorbid DD and AD, .0002 for DD, and .007 for AD). SNP3 and SNP8 together exerted stronger genetic effects on SD than either did individually. The peak δ values among all the markers were seen for SNP3 and SNP8 (rs806368).
Conclusions
We demonstrate that CNR1 variation and interactive effects play important roles in risk for both DD and AD.
Substance dependence (SD), including drug dependence (DD) and alcohol dependence (AD), is described as a cycle of increasing dysregulation of brain reward systems (
Several studies have shown that the endocannabinoid system serves to regulate DA reward circuits, an effect that may play an important role in the reward processes involved in SD (e.g.,
). The administration of delta-9-tetrahydrocannabinol (Δ9-THC), the main psychoactive ingredient of cannabis, increases extracellular DA concentrations in the nucleus accumbens (NAc), and this Δ9-THC effect can be blocked by the μ opioid receptor antagonist naloxonazine when it is infused into the ventral tegmental area (VTA) (
D9-tetrahydrocannabinol produces naloxone-blockable enhancement of presynaptic basal dopamine efflux in nucleus accumbens of conscious, freely-moving rats as measured by intracerebral microdialysis.
). Additionally, animal studies have demonstrated that both acute alcohol- and morphine-induced DA release in the NAc and dependence-inducing properties of opiates, cocaine, and alcohol were reduced in CB1 knockout mice, and similar results were obtained after the blockade of CB1 receptors with the selective CB1 receptor antagonist, SR141716A (
So far, two subtypes of cannabinoid receptors have been identified—the brain cannabinoid receptor (CB1) and the peripheral cannabinoid receptor (CB2) (
). CB1 is a G-protein coupled receptor in presynaptic nerve terminals and is the target of Δ9-THC.
CB1 is encoded by the cannabinoid receptor 1 gene (CNR1), which maps to 6q14–q15. Alternative splicing leads to two transcript variants, that is, transcript variant 1 (NM_016083/X81120), which encodes isoform A, and transcript variant 2 (NM_033181/X81121), which encodes isoform B. Recently,
found that a CNR1 TAG haplotype consisting of rs806379, rs1535255, and rs2023293 was related to polysubstance abuse in both EAs (p = 3.0 × 10−5) and AAs (p = .007). They also found it to be significantly associated with AD in a Japanese sample (p = 8.0×10−6). CNR1 seems to be a promising candidate gene for SD, but positive associations between CNR1 and SD (including
We conducted a population-based association study to investigate the role of CNR1 in risk for DD (cocaine dependence [CD] and/or opioid dependence [OD]) and AD in EAs and AAs using a powerful study design. First, we considered confounders that are common in population-based association studies, such as population admixture, age, and sex, but that have generally not been considered in previous studies. Several studies have shown that the allele frequency distributions of many genetic variants at CNR1 differ significantly by population and sex (
). Furthermore, to varying degrees, AAs and EAs are admixed populations, that is, AA individuals often have some degree of European ancestry and EA individuals may have small proportions of African ancestry (
). Younger control subjects who are assessed as “healthy” may still develop SD when they become older. Thus, ethnicity, population admixture, sex, and age may confound gene-phenotype association analysis. Consequently, in this study, we performed structured association (SA) analysis and multivariate logistic regression analysis to control for these potential confounding effects. Second, we used a set of markers covering the full length of CNR1. Except for the studies by
, previous studies have focused on two CNR1 polymorphisms, that is, the (AAT)n polymorphism and a 1359G/A variant at codon 453. The small number of markers studied and the incomplete linkage disequilibrium (LD) across the gene would be expected to limit the detection of association. Additionally, besides the known CNR1 region shown in the current National Center for Biotechnology Information (NCBI) database, unknown functional variants in the 5′ flanking region of CNR1 could affect disease risk. On this basis, we extended our region of interest to 67.09 kb across the CNR1 locus and investigated 10 polymorphisms in this region based on the LD and fine-mapping information available from the current NCBI, ABI, and HapMap databases and the extant literature. Third, we considered multiple phenotypes that could be influenced to varying degrees by CNR1 polymorphism. We also investigated the phenotype of comorbid DD and AD (DD&AD). Finally, single-locus analysis used in most previous studies simply reflects the association between certain markers and phenotype; limited LD might limit the ability to map the disease locus. To address these issues, we applied multivariate logistic regression analysis, haplotype trend regression (HTR;
), and marker–marker interaction analysis, which preserve much more genetic information than conventional single-locus analysis.
Methods and Materials
Subjects
Included in the study were 1001 subjects, comprising 451 healthy control subjects and 550 cases with CD, OD, and/or AD. This sample included two populations: 794 self-reported European Americans (EAs) and 207 self-reported African Americans (AAs). The ages were 28.5 ± 17.0 years for the control subjects and 39.5 ± 18.0 years for the cases. The control group consisted of 187 men and 264 women; in the cases, there were 405 men and 145 women. The cases met lifetime DSM-III-R or DSM-IV criteria (
) for CD, OD, AD, or combinations of these disorders. The control subjects were screened to exclude major Axis I disorders, including SD, schizophrenia, mood disorders, and major anxiety disorders, using the Structured Clinical Interview for DSM-III-R, the Computerized Diagnostic Interview Schedule for DSM-III-R (C-DIS-R), the Schedule for Affective Disorders and Schizophrenia (SADS) (
), or an unstructured interview. The subjects were recruited at the University of Connecticut Health Center (UCHC) or the VA Connecticut Healthcare System—West Haven Campus. Genotyping data for rs806379 (single nucleotide polymorphisms [SNP]4) in an overlapping set of UCHC subjects (n = 817) were obtained independently and analyzed as part of another study (
); there was 100% concordance between duplicate genotypes in this overlapping set of subjects. All subjects gave informed consent before participating in the study, which was approved by the institutional review boards at each institution.
Marker Selection and Genotyping
Ten CNR1 SNP markers (average spacing, 7.45 kb) were selected. The 10 SNPs were designated as SNPs 1–10 in 5′ to 3′ order (Figure 1).
Figure 1CNR1 gene model and the genotyped markers. a Single nucleotide polymorphisms (SNP) numbers correspond to Table 1. bTAG Haplotype refers to that risk haplotype in the study by
: T(rs806379), A(rs1535255), G(rs 2023293). c The exon 4 region corresponds to the exonic region of CNR1 in National Center for Biotechnology Information build 36.2 and the University of California at Santa Cruz Browser. The other three exons in this figure are the potential exons proposed by
To detect the population structure of our sample, we genotyped 38 ancestry-informative markers (AIMs), including 37 short tandem repeat markers (STRs) and one Duffy antigen gene (FY) marker (rs2814778). The characteristics of this marker set have been described in detail previously (
High throughput genotyping for the detection of a single nucleotide polymorphism in NAD(P)H quinone oxidoreductase (DT diaphorase) using TaqMan probes.
). The 37 ancestry-informative STR markers were genotyped using the ABI PRISM 3100 semiautomated capillary fluorescence analyzer. Genomic DNA was extracted from peripheral blood by standard methods. Polymerase chain reaction conditions were described in detail elsewhere (
ADH4 gene variation is associated with alcohol dependence and drug dependence in European Americans: Results from family-controlled and population-structured association studies.
Estimated ancestry proportion scores for each subject and number of ancestral populations were obtained through use of the program STRUCTURE, based on a model-based clustering method (
). We set burn-in period length as 100,000, then used 100,000 Markov Chain Monte Carlo (MCMC) repetitions to obtain parameter estimates: ln Pr(X⊻K), where Pr denotes posterior probability, X denotes genotypes of the sampled individuals, and K denotes the number of populations assumed. The optimal K is considered to be the one with the highest posterior probability.
The correction for multiple tests was performed with the program SNPSpD to calculate the effective independent marker number from the nonindependent markers (
Liu K, Muse S (2004): PowerMarker: new genetic data analysis software. Version 3.0. Distributed by the author. Available at: http://www.powermarker.net. Accessed June 1, 2006.
The case–control comparisons for allele and genotype frequency distributions were performed with the exact tests implemented in the program PowerMarker.
Structured association (SA) analysis was performed using the program STRAT (
) to control for population admixture effects. We also performed multivariate logistic regression analysis in SPSS 14.0 to control for the other potential confounding effects. Because we do not know the exact genetic mode of inheritance for each marker beforehand, we performed this analysis under the genotype model (“model-free”) first. Then we performed this analysis under the additive and/or recessive genetic models, which were indicated post hoc by our results. In the regression model, phenotypes served as the dependent variable, and the independent variables included ancestry proportions, age, sex, and genotypes of each CNR1 marker.
The interaction effect was analyzed via logistic regression, using SPSS 14.0. The dependent variable was phenotype, and the independent variables included ancestry proportions, age, sex, the genotypes of the risk CNR1 markers (identified by the regression analysis described earlier) and the two-way interactions between the disease-related marker and the other CNR1 markers. We performed this regression analysis under the genotype model first to derive the most possible genetic modes of inheritance for the disease related marker. Under these derived models, we reanalyzed the data. To reduce type I error, only those markers that showed main effects were included in the interaction analyses.
The positive likelihood ratios (LRs+) for each risk genotype were calculated by dividing the genotype frequencies in cases by those in control subjects (details see
). Before running PHASE, we set the minimal threshold of the rare diplotype frequency at .00001. We performed 10,000 iterations for each run.
For HTR and DTR analyses, stepwise logistic regression implemented in SPSS 14.0 was used. Phenotypes served as the dependent variable, and the covariates included ancestry proportions, age, sex, and haplotype or diplotype probabilities.
To assess the degree of LD between the putative disease locus and the adjacent markers, we used δ = (PD − PN)/(1 − PN), where PD and PN are the frequencies of the candidate allele on the disease chromosome and the normal chromosome, respectively (
). δ represents an estimation of the proportion of disease chromosomes carrying the related allele. The proportion of disease chromosomes carrying a risk allele should be higher than that of normal chromosomes carrying this risk allele (δ > 0), and the proportion of disease chromosomes carrying a protective allele should be lower than that of normal chromosomes carrying the protective allele (δ < 0); that is, the highest positive δ value (δ+) suggests a risk allele (among a set of markers in an associated block), and the lowest negative δ value (δ−) suggests a protective allele. Thus, we calculated both δs+ to map risk alleles and δs− to map protective alleles. The pairwise correlation between the specific markers that have peak δ values among all markers was analyzed using the Spearman correlation analysis in SPSS 14.0.
Results
Two Ancestries Were Identified in All Subjects
According to the optimal K and ancestry proportions, all the subjects in this study were separated into two genetically distinct populations: “genetic” EAs (European ancestry proportion > .5) and “genetic” AAs (African ancestry proportion > .5). All of the following analyses were performed separately within the “genetic” EAs and AAs.
The concordance rates between the self-reported ethnicity and the “genetic” ethnicity were 100% in EAs and 99.0% in AAs, respectively. The degree of admixture ranged from .2% to 50.0% (mean 1.1%) in “genetic” EA individuals, and from .2% to 46.7% (mean 4.3%) in “genetic” AA individuals.
All Markers Were Consistent with HWE in Control Subjects, but Some Markers Were in HWD in Cases
In control subjects, the genotype frequency distribution of every CNR1 marker was consistent with HWE in both EAs and AAs (after correction for multiple tests). In cases, however, the genotype frequency distributions of SNPs 7 and 9 were in significant HWD in EAs (p = .001 and .0002, respectively), and the genotype frequency distributions of SNPs 4 and 9 were in significant HWD in AAs (p = .003 and .002, respectively). These HWD observations required us to choose downstream haplotype reconstruction analysis methods that do not require the assumption of HWE, such as the aforementioned PHASE (for haplotype reconstruction) and DTR (for association analysis), to ensure the validity, reliability, and power of the analyses.
The Genotype and Allele Frequency Distributions of Two Markers Were Significantly Different Between Cases and Control Subjects in EAs
In EAs, the genotype frequency distribution of SNP1 in DD and the allele frequency distribution of SNP3 in DD&AD were significantly different from those in control subjects (after correction for multiple tests, α = .007; Table 1). In AAs, none of the allele and genotype frequency distributions of the CNR1 SNPs was significantly different between cases and control subjects (after correction for multiple tests, α = .006; Table 2).
Table 1Allele and Genotype Frequencies of CNR1 Markers in European Americans
DD&AD, comorbid DD and AD; DD, drug dependence; AD, alcohol dependence; Total-SD, all the cases in the present study; n, number of chromosomes (for alleles) or individuals (for genotypes); f, frequency.
a p ≤ .05
b p ≤ .01
c p ≤ .007 (= α) for conventional case-control comparison.
d p ≤ .05
e p ≤ .01
f p ≤ .007 (= α) for structured association (SA) analysis using STRAT.
After Controlling for the Effects of the Confounders Including Population Admixture, Age and Sex, SNP3 Was Significantly Associated with DD&AD, DD, and AD in EAs
After controlling for population admixture effects using STRAT, the genotype frequency distribution of SNP1 was significantly associated with DD, the allele frequency distribution of SNP3 was suggestively associated with DD&AD, and neither the genotype nor the allele frequency distributions of any of the other SNPs were significantly associated with phenotypes after correction for multiple tests (Table 1).
The multivariate logistic regression analysis under the genotype model suggested that the mode of inheritance for SNP3 may be additive. Under an additive model, risk for SD significantly increased with the number of SNP3^G alleles increasing, as did risk for DD&AD, DD and AD (Table 3).
Table 3Regression Analysis on Marker–Phenotype Associations in European Americans
Phenotypes
Genotype Model
Additive Model
Covariates
β
p
OR
95% CI for OR
Covariates
β
p
OR
95% CI for OR
Lower
Upper
Lower
Upper
DD&AD
Constant
−6.318
7.3E-22
.002
Constant
−6.291
3.9E-22
.002
Sex
1.199
1.7E-05
3.317
1.919
5.731
Sex
1.200
1.7E-05
3.321
1.922
5.738
Age
.131
8.4E-15
1.140
1.103
1.178
Age
.131
8.5E-15
1.140
1.103
1.178
SNP3
.019
SNP3
.578
.005
1.783
1.190
2.671
SNP3^G/G
1.102
.014
3.009
1.246
7.265
SNP3^G/T
.644
.029
1.904
1.068
3.395
DD
Constant
−5.778
1.3E-24
.003
Constant
−5.777
6.5E-25
.003
Sex
1.030
3.4E-05
2.800
1.720
4.558
Sex
1.030
3.4E-05
2.801
1.721
4.558
Age
.129
3.0E-18
1.138
1.105
1.171
Age
.129
3.0E-18
1.138
1.105
1.171
SNP3
.032
SNP3
.491
.009
1.635
1.132
2.360
SNP3^G/G
.980
.018
2.665
1.184
5.996
SNP3^G/T
.495
.061
1.640
.978
2.749
AD
Constant
−7.029
1.7E-32
.001
Constant
−7.004
7.7E-33
.001
Sex
1.537
3.5E-10
4.653
2.878
7.522
Sex
1.539
3.4E-10
4.660
2.883
7.533
Age
.167
1.6E-28
1.182
1.148
1.218
Age
.167
1.5E-28
1.182
1.147
1.217
SNP3
.039
SNP3
.459
.012
1.582
1.108
2.260
SNP3^G/G
.852
.038
2.344
1.047
5.247
SNP3^G/T
.519
.039
1.680
1.025
2.752
Total SD
Constant
−6.399
1.1E-33
.002
Constant
−6.396
4.7E-34
.002
Sex
1.360
2.0E-09
3.895
2.497
6.075
Sex
1.360
2.0E-09
3.896
2.498
6.076
Age
.159
2.2E-30
1.172
1.141
1.204
Age
.159
2.2E-30
1.172
1.141
1.204
SNP3
.056
SNP3
.411
.016
1.509
1.078
2.111
SNP3^G/G
.813
.037
2.255
1.049
4.844
SNP3^G/T
.420
.075
1.521
.959
2.415
DD&AD, DD, AD, and Total-SD abbreviations same as Table 1; Phenotypes, the dependent variables; Covariates, the independent variables left in the final regression equation; Genotype model, model-free; Additive model, SNP3^T/T = 0, G/T = 1, G/G = 2; β, regression coefficients; OR, odds ratios; CI, confidence interval; E − n = ×10−n.
Because the sample size of AAs was small, we did not perform this analysis and the subsequent analyses in this population group.
The Interaction of SNP3 with SNP8 Significantly Increased Risk for DD&AD, DD, and AD in EAs
Genotype interactions between SNP3^G/G and SNP8^T/T and between SNP3^G/T and SNP8^T/T significantly increased risk for SD. The interaction effect between SNP3^G/G and SNP8^T/T on SD was more than double that between SNP3^G/T and SNP8^T/T (Table 4).
Table 4Regression Analysis on the Association Between SNP3 × SNP8 Interaction and Phenotypes in European Americans
Phenotype
Genotype Model
Mixed Model
Covariates
β
p
OR
95% CI for OR
Covariates
β
p
OR
95% CI for OR
Lower
Upper
Lower
Upper
DD&AD
Constant
−6.488
4.8E-22
.002
Constant
−6.302
1.1E-22
.002
Sex
1.265
9.4E-06
3.543
2.024
6.200
Sex
1.229
1.3E-05
3.418
1.966
5.942
Age
.133
9.3E-15
1.142
1.104
1.181
Age
.132
7.9E-15
1.141
1.104
1.179
SNP3 × SNP8
.003
SNP3 × SNP8
.763
3.0E-04
2.145
1.420
3.250
SNP3^G/G × SNP8^T/T
1.784
4.0E-04
5.952
2.216
15.985
SNP3^G/T × SNP8^T/T
.815
.013
2.259
1.184
4.312
SNP3^G/T × SNP8^T/C
.717
.085
2.049
.907
4.628
SNP3^G/G × SNP8^T/C
−.833
.488
.435
.041
4.588
DD
Constant
−5.958
7.9E-25
.003
Constant
−5.882
1.4E-25
.003
Sex
1.093
1.7E-05
2.984
1.814
4.910
Sex
1.069
2.2E-05
2.913
1.778
4.773
Age
.132
2.3E-18
1.141
1.108
1.175
Age
.132
2.0E-18
1.141
1.107
1.175
SNP3 × SNP8
.003
SNP3 × SNP8
.730
2.0E-04
2.080
1.420
3.040
SNP3^G/G × SNP8^T/T
1.644
5.0E-04
5.175
2.064
12.972
SNP3^G/T × SNP8^T/T
.714
.015
2.041
1.148
3.631
SNP3^G/T × SNP8^T/C
.449
.241
1.567
.740
3.320
SNP3^G/G × SNP8^T/C
−1.170
.327
.310
.030
3.213
AD
Constant
−7.029
1.7E-32
.001
Constant
−6.912
9.0E-34
.001
Sex
1.537
3.5E-10
4.653
2.878
7.522
Sex
1.532
3.9E-10
4.628
2.864
7.478
Age
.167
1.6E-28
1.182
1.148
1.218
Age
.167
1.4E-28
1.182
1.147
1.217
SNP3
.039
SNP3 × SNP8
.530
.007
1.700
1.160
2.490
SNP3^G/G
.852
.038
2.344
1.047
5.247
SNP3^G/T
.519
.040
1.680
1.025
2.752
SNP3 × SNP8
>.05
Total SD
Constant
−6.490
7.2E-34
.002
Constant
−6.399
6.2E-35
.002
Sex
1.386
1.4E-09
3.997
2.552
6.262
Sex
1.371
1.7E-09
3.940
2.522
6.157
Age
.160
2.0E-30
1.174
1.142
1.207
Age
.160
1.9E-30
1.173
1.141
1.205
SNP3 × SNP8
.041
SNP3 × SNP8
.550
.003
1.740
1.210
2.490
SNP3^G/G × SNP8^T/T
1.279
.007
3.594
1.428
9.041
SNP3^G/T × SNP8^T/T
.558
.035
1.747
1.041
2.931
SNP3^G/T × SNP8^T/C
.342
.328
1.408
.710
2.794
SNP3^G/G × SNP8^T/C
−.020
.977
.980
.251
3.821
DD&AD, DD, AD and Total-SD abbreviations same as Table 1; Phenotypes, Covariates, Genotype model, β, OR, CI, and E − n, explanations same as Table 3. Mixed model, the genotypes of SNP3 and SNP8 are in an additive and recessive models respectively in the regression model, that is, SNP3^T/T = 0, G/T = 1, G/G = 2, SNP8^C/C = C/T = 0, and SNP8 ^T/T = 1.
The LRs+ of SNP3^G/G genotype, SNP3^G/T genotype, and SNP8^T/T genotype were > 1. The LRs+ of SNP3^G/G×SNP8^T/T genotypes were significantly higher than those of SNP3^G/G alone, especially in DD&AD (2.50 vs. 1.70) and DD (2.03 vs. 1.47). The LRs+ of SNP3^G/T×SNP8^T/T genotypes were also significantly higher than those of SNP3^G/T alone (Figure 2).
Figure 2Positive likelihood ratios (LRs+) for the genotypes of single nucleotide polymorphism (SNP)3 and SNP8 in European Americans (x axis represents the genotypes of SNP3 or SNP8, or the interaction between SNP3 genotypes and SNP8 genotypes: 1, SNP8^C/C; 2, SNP8^T/C; 3, SNP3^T/T; 4, SNP8^T/T; 5, SNP3^G/T; 6, SNP3^G/T×SNP8^T/T; 7, SNP3^G/G; 8, SNP3^G/G×SNP8^T/T).
Both the regression analysis and the LR+ analysis suggested that SNP3 and SNP8 followed additive and recessive genetic models, respectively. Based on this hypothesis, we performed regression analysis on the marker–disease association under a mixed genetic model in which the genotypes of SNPs 3 and 8 were recoded according to an additive and a recessive genetic model, respectively. Under this mixed model, we found that SNP3^G+ (the genotypes containing the G allele) and SNP8^T/T genotypes have significant interaction effects on risk for DD&AD, as well as DD and AD.
Some Specific Haplotypes and Diplotypes at CNR1 Were Significantly Associated with DD&AD, DD, or AD in EAs
Haplotypewise association analyses using the program PHASE (which directly compared the haplotype frequency distributions between cases and control subjects) and the HTR implemented in the program Powermarker showed that DD&AD was significantly associated with the SNP3 haplotype block (i.e the haplotype block where SNP3 is located; p values for both methods = .001), but not with the SNP8 haplotype block in EAs (Figure 3).
Figure 3Linkage disequilibrium map for 10 CNR1 markers in European Americans (A) and African Americans (B). Numbers in the squares are D′ × 100. Empty red squares mean D′ = 1.0 (i.e., complete linkage disequilibrium between two single nucleotide polymorphisms [SNPs]). Blue squares represent logarithm of the odds (LOD) score < 2. The haplotype block was defined by solid spine of linkage disequilibrium (extend spine if D′ > .9 and LOD score ≥ 2).
The frequencies of common haplotypes (f > 2.0%) and diplotypes (f > 1.0%) reconstructed with all 10 SNPs are shown in Table 5. There were 12 common haplotypes and 8 common diplotypes in EAs. Because these haplotypes and diplotypes spanned a very wide region (67 kb), and the LD between the SNPs was weak in a “sample” European population (from the ABI database), the SNPs comprised numerous haplotypes and diplotypes in our sample, which resulted in the frequency of each common haplotype or diplotype being lower than .2.
Table 5Frequencies of Common Haplotypes and Diplotypes
European Americans
African Americans
Codes
Haplotype/Diplotype
f
Codes
Haplotype/Diplotype
f
EA-Haplotype1
CATTTTGTTT
.109
AA-Haplotyp1
CGTACTGTCC
.098
EA-Haplotype2
CGTTTTGTTT
.090
AA-Haplotyp2
CGTTTTGTTT
.062
EA-Haplotype3
CAGACTGTTT
.072
AA-Haplotyp3
CATTTGGCCC
.054
EA-Haplotype4
CATTTTATCC
.071
AA-Haplotyp4
CATTTTGTCC
.051
EA-Haplotype5
CAGACTGTCC
.060
AA-Haplotyp5
CAGACTGTCC
.051
EA-Haplotype6
CGTACTATCC
.044
AA-Haplotyp6
CAGACTGTTT
.046
EA-Haplotype7
CGTTTGGCCT
.031
AA-Haplotyp7
CGTACTGTTT
.044
EA-Haplotype8
CGTTTTATCC
.030
AA-Haplotyp8
CGTACTGTCT
.044
EA-Haplotype9
CAGACTATCC
.028
AA-Haplotyp9
CATTTTGTTT
.043
EA-Haplotype10
CGGACTATCC
.028
AA-Haplotyp10
CGTACGGTCC
.037
EA-Haplotype11
CATTTGGCCT
.024
AA-Haplotyp11
CGGACTGTCC
.033
EA-Haplotype12
CATACTGTTT
.022
AA-Haplotyp12
CGGTCTGTTT
.026
AA-Haplotyp13
CGGTCTGTCC
.025
AA-Haplotyp14
CGTTTTGTCC
.024
AA-Haplotyp15
CGTACTATCC
.022
EA-Diplotyp1
CAGACTGTTT/CATTTTGTTT
.026
AA-Diplotyp1
CGTTTTGTTT/CATTTTGTTT
.017
EA-Diplotyp2
CGTTTTGTTT/CATTTTGTTT
.025
AA-Diplotyp2
CAGACTGTCC/CAGACTGTTT
.015
EA-Diplotyp3
CATTTTATCC/CGTTTTGTTT
.016
EA-Diplotyp4
CAGACTGTTT/CGTTTTGTTT
.015
EA-Diplotyp5
CATTTTATCC/CATTTTATCC
.015
EA-Diplotyp6
CATTTTGTTT/CATTTTGTTT
.013
EA-Diplotyp7
CAGACTGTCC/CATTTTGTTT
.013
EA-Diplotyp8
CAGACTGTCC/CAGACTGTTT
.012
f, frequency; only the haplotypes with f > .02 and the diplotypes with f > .01 are listed. Bold italics are the disease-related haplotypes or diplotypes.
In EAs, HTR analysis showed that EA-Haplotype3 was significantly associated with DD&AD (p = .033), EA-Haplotype9 was significantly associated with AD (p = .010), and EA-Haplotype10 was significantly associated with DD&AD and DD (p = .012, .034, respectively). DTR analysis in EAs showed that EA-Diplotype7 was significantly associated with DD&AD (p = .036).
Risk and Protective Alleles for SD Phenotypes Can Be Mapped Close to Specific CNR1 Markers
The δs+ and δs− are plotted in Figure 4. δs+ reached peaks at SNPs 6 and 8 in all phenotype groups and at SNP1 in the DD&AD and DD groups. The lowest δ− is at SNP3 in all phenotype groups.
Figure 4Fine-mapping disease risk alleles (A) and protective alleles (B) in CNR1 locus in European Americans. The x axis represents the markers corresponding to Table 1, and the y axis represents the delta values (δs). Positive δ suggests that the marker allele is a risk allele; negative δ suggests that the marker allele is a protective allele.
The Spearman correlation analysis between any two of SNPs 1, 3, 6, and 8 showed that SNP3 was significantly correlated with SNP1 (p = .001, ρ = .123) but not with SNP6 or SNP8. There was significant correlation between SNP6 and SNP8 as well (p = 1.0 × 10−6, ρ = .746).
The finding that SNP8 was neither in LD nor correlated with SNP3 further suggested that the interaction effects between the two SNPs reflected two independent loci, rather than single locus effects.
Discussion
This study demonstrates that CNR1 variation is associated with SD in EAs, with the effect mostly attributable to 2 of the 10 SNPs queried, i.e., SNP3 and SNP8. In AAs, SNP3 was “suggestively” associated with DD&AD, DD, and AD.
In the initial case–control comparison analysis, SNP3 was significantly associated only with EA DD&AD. When we applied multivariate logistic regression analysis, the SNP3^G allele was significantly associated not only with DD&AD but also with DD and AD individually. Although the distributions of age and sex are asymmetric between cases and control subjects in our sample, the confounding effects from age, sex, and other covariates were controlled in a single regression model, and multiple tests were avoided; these findings reliably suggest that SNP3 is associated with SD. SNP3 is located in a haplotype block with SNP4 (rs806379) and SNP5 (rs806377). Haplotypewise association analyses showed that this haplotype block was significantly associated with DD&AD. The risk TAG haplotype reported to be strongly associated with polysubstance abuse (
) is located within this haplotype block region. Taken together, these findings suggest that a potential disease locus for SD may be in LD with SNP3. Although SNP3 is located in intron 2, it might indirectly affect risk for SD via LD with other flanking functional variants or via involvement in alternative gene splicing. It might also affect risk for SD via a direct pathway.
Our interaction analysis revealed that SNP3^G+ and SNP8^T/T genotypes had strong interaction effects on SD including DD&AD, DD, and AD in EAs. According to the estimation of LRs+ for the risk genotypes, SNP3^G+ genotypes exerted minor risk effects on each SD subtype (1 < LR+ < 2), whereas SNP3^G/G × SNP8^T/T genotypes exerted small risk effects on DD&AD and DD (2 < LR+ < 5).
SNP8 is located in exon 4. Its allele frequency distribution in EA subjects with DD&AD or DD was “suggestively” different from that in control subjects. The LRs+ of SNP8^T/T genotype were similar to those of the risk SNP3^G/T genotype and larger than those of SNP3^T/T genotype and the other two SNP8 genotypes, suggesting that SNP8^T/T genotype might exert risk effects on SD as well. SNP8 is 22.83 kb away from SNP3 and in a different haplotype block (D′ = .003). After controlling for confounding effects from SNP3, SNP8 was still associated with DD&AD (p = .010, data not shown), although no longer associated with DD. Therefore, in CNR1, there might be a risk locus for SD that is in LD with SNP8, besides the one that is in LD with SNP3. Interestingly, SNP8 was in strong LD with SNP7 (rs1049353), which is the well-known 1359G/A (Thr453Thr) variant at CNR1 that was reported to be associated with severe alcohol dependence (
). Taken together, these findings suggest that an SD-related locus might be in LD with SNP8.
Recently, two powerful and robust approaches for assessing gene–disease association, HTR and DTR, have been developed. Results from these two analyses supported CNR1 as a risk gene for SD in EAs. Both risk haplotypes and diplotypes contain the SNP3^G allele and SNP8^T allele, consistent with our findings.
Using δ to guide fine mapping, we detected, in total, four markers with peak δs, including SNPs 1, 3, 6, and 8. SNP3 and SNP8 were discussed earlier. The other two markers, SNP1 and SNP6, were significantly correlated with SNP3 and SNP8, respectively. The peak δs of these four markers are consistent with the results from marker–phenotype association analysis.
Markers being in HWD in cases might indicate a valid marker–disease association (
ADH4 gene variation is associated with alcohol dependence and drug dependence in European Americans: Results from HWD tests and case–control association studies.
). The two markers that flank SNP8—SNP7, which is 5′ to SNP8, and SNP9, which is 3′ to SNP8—were in HWD in the EA cases (p = .001, and .0002). SNP7^A and SNP9^C were significantly correlated with SNP8^T (p = 1.0 × 10−6 and .010). SNP7 was in a haplotype block with SNP8. Thus, the observed HWD for the two markers might be caused by their correlation with SNP8, shown to be an SD-related marker in this study.
Additionally, we found that both the ORs and the LRs+ for both SNP3 and the SNP3 × SNP8 interaction were in the following order: DD&AD > DD > AD. Furthermore, among these three SD subtypes, the marker–phenotype association for DD&AD was most significant (i.e had the lowest p value). These findings suggest that the phenotype structure of SD (i.e the combination of substance dependencies that are reflected by the SD diagnosis) influenced the magnitude of gene risk effects.
Alcohol dependence is one of the disorders most commonly comorbid with drug dependence (
Alcohol use disorders comorbid with anxiety, depression and drug use disorders Findings from the Australian National Survey of Mental Health and Well Being.
Specificity of genetic and environmental risk factors for use and abuse/dependence of cannabis, cocaine, hallucinogens, sedatives, stimulants, and opiates in male twins.
). CB1 is involved in the modulation of DA in mesocorticolimbic DA circuits, and these brain reward circuits are implicated in all common SD, including CD, OD, and AD (
). Therefore, it is understandable that CNR1, which encodes CB1, is a risk gene common for both DD and AD, as indicated in our study.
In conclusion, this study showed that CNR1 variation was significantly associated with SD. Further studies using denser marker sets are needed to locate the disease-influencing loci near SNP3 and SNP8 and to investigate the mechanism of the interaction between these two loci. Furthermore, we presume that the magnitude of genetic risk effects may vary among the different SD subtypes. Recent behavioral evidence from an animal study indicates that cannabinoid receptor antagonists such as SR141716A (Rimonabant) can reduce the self-administration of, and craving for, several commonly used addictive drugs (
). The demonstrations of association between CNR1 and SD in human subjects might contribute to a clearer understanding of the role of the cannabinoid system in SD, thereby making possible new approaches to the prevention and treatment of SD.
This work was supported in part NIH Grant Nos. R01-DA12849, R01-DA12690, K24-DA15105, R01-AA11330, R01-AA016015, K24-AA13736, P50-AA12870, P50-AA03510, and M01-RR06192 (University of Connecticut General Clinical Research Center); by funds from the U.S. Department of Veterans Affairs (the VA Medical Research Program and the VA Connecticut–Massachusetts Mental Illness Research, Education and Clinical Center [MIRECC], and the VA Research Enhancement Award Program [REAP] research center); and Alcoholic Beverage Medical Research Foundation (ABMRF) Grant Award R06932 (X. Luo). We thank Ann Marie Lacobelle for excellent technical assistance.
References
Alger B.E.
Retrograde signaling in the regulation of synaptic transmission: focus on endocannabinoids.
D9-tetrahydrocannabinol produces naloxone-blockable enhancement of presynaptic basal dopamine efflux in nucleus accumbens of conscious, freely-moving rats as measured by intracerebral microdialysis.
Specificity of genetic and environmental risk factors for use and abuse/dependence of cannabis, cocaine, hallucinogens, sedatives, stimulants, and opiates in male twins.
Liu K, Muse S (2004): PowerMarker: new genetic data analysis software. Version 3.0. Distributed by the author. Available at: http://www.powermarker.net. Accessed June 1, 2006.
ADH4 gene variation is associated with alcohol dependence and drug dependence in European Americans: Results from HWD tests and case–control association studies.
ADH4 gene variation is associated with alcohol dependence and drug dependence in European Americans: Results from family-controlled and population-structured association studies.
High throughput genotyping for the detection of a single nucleotide polymorphism in NAD(P)H quinone oxidoreductase (DT diaphorase) using TaqMan probes.
01549-6&id=gr1.jpg){kind=link}
01549-6&id=gr2.jpg){kind=link}
01549-6&id=gr3.jpg){kind=link}
01549-6&id=gr4.jpg){kind=link}