Haplotypic Variants in DRD2, ANKK1, TTC12 and NCAM1 Are Associated With Comorbid Alcohol and Drug Dependence

Abstract

Background

Each gene in the chromosome 11q23 cluster of NCAM1, TTC12, ANKK1 and DRD2 is functionally linked to dopamine in brain. Many association studies of DRD2 and substance dependence (SD), including alcohol dependence (AD) and drug dependence (DD) have been reported; the results have been inconsistent. Recent association studies have considered this cluster more comprehensively, examining the association of SD with several risk variants mapped to the other genes in the cluster. Because comorbid AD with DD (AD+DD) is common, we hypothesized that heterogeneity of the SD diagnoses studied might have contributed to the inconsistency of prior results.

Methods

We conducted two separate association studies of AD+DD and AD without DD (AD-only) in 1,090 European-Americans using family-based and case-control designs and 43 single nucleotide polymorphisms mapped to this cluster. We used a generalized linear model and haplotype score tests for the case-control sample, and the family-based association test for the family sample.

Results

For AD+DD, the risk regions centered on TTC12 exon 3 (optimal individual haplotype simulated p (p_oihs) = 0.000015), and another extended from ANKK1 exon 8 to DRD2^C957T (p_oihs = 0.0028), in both samples. NCAM1 exon 12 markers showed global significance in both designs, but were significant for specific haplotypes (p_oihs = 0.0029) only for the family sample. For AD-only, NCAM1 intron 14–18 and the junction of ANKK1 and DRD2 were associated globally. Population stratification was excluded as the basis for these results. Linkage disequilibrium contrast tests supported selection at TTC12 exon 3 and ANKK1 exon 2.

Conclusions

We conclude that variants in TTC12 exon 3, NCAM1 exon 12, and the two 3′ ends of ANKK1 and DRD2 co-regulate risk for comorbid AD and DD.

Introduction

Substance dependence (SD) has a complex etiology, with constituents of risk that include genetic, neurobiological, and environmental factors. The D₂ dopamine receptor gene (DRD2), the protein product of which links to the brain mesocortical dopamine reward system, has been the focus of much attention in addiction genetics, and in psychiatric genetics in general. There is a large volume of inconsistent research on the possible association of SD with the putative DRD2 TaqI “A” polymorphism, which has been shown to be located in ANKK1, an adjacent gene (Neville et al., 2004).

The long history of inconsistent reports could reflect more complicated genetic substrates than have heretofore been recognized. In two recent studies, we (Gelernter et al., 2006; Yang et al., 2007) found a broader genomic region on chromosome 11q23, including a gene cluster of NCAM1, TTC12, ANKK1 and DRD2 (“NTAD” cluster), spanning ~542 kb, to be associated with the genetic risk of nicotine dependence (ND) and alcohol dependence (AD), respectively.

We speculated previously that these genes could be functionally related. Within this gene cluster, a haplotype spanning TTC12 and ANKK1 was strongly associated with ND in European- and African-American (EA and AA, respectively) family samples (Gelernter et al., 2006). Haplotypes around TTC12 exon 3, NCAM1 exon 12, and exons 2 and 5 of ANKK1 were also associated with AD in two other independent EA samples, one population-based and one family-based (Yang et al., 2007). These associations were not attributable to linkage disequilibrium (LD) with DRD2 and DRD2 itself showed only very limited association. Among these identified susceptibility regions, the association signals from TTC12 were consistently prominent. The present study considers the latter samples (i.e., from Yang et al, 2007) in the context of additional diagnostic information, with the goal of adding to our understanding of the nature of the associated phenotypes.

While the functions of TTC12 and ANKK1 are not well established, present knowledge suggests that they may be related in part to the development or modulation of dopaminergic neurotransmission. TTC12 encodes a protein called TPARM, which includes one tetratricopeptide repeat domain (TPR) and three armadillo repeat domains (ARMs) (Katoh and Katoh, 2003). Its ARM, included in a family of proteins containing several 42-amino acid repeat domains, is thought to play a variety of roles at the cellular level, including regulation of desmosome assembly, cell adhesion, neurogenesis, signal transduction, and post developmental synaptogenesis. β-catenin, a prototypical ARM-containing protein, was recently identified as a risk locus for bipolar affective disorder (Baum et al., 2007). These findings, and others, suggest that the ARMs encoded by TTC12 might also be involved in the cascade of events in the Wnt signal transduction pathway that occurs in early neuronal development (Castelo-Branco and Arenas, 2006; Castelo-Branco et al., 2004), and could potentially influence the level of dopamine D₂ receptors and thereby modulate SD risk.

ANKK1 was also found to associate with the executive attention network in the anterior cingulate gyrus, a dopamine-rich region of the brain (Fossella et al., 2006). A recent study of association between ND and a region spanning ANKK1 and DRD2 reported a putatively functional polymorphism, rs2734849, in an African American sample (Huang et al., 2008); this variant alters the expression level of NF-κB regulated genes (Huang et al., 2008). Further, NF-κB1 gene encoding the transcription factor NF-κB was associated with AD (Edenberg et al, 2008). Another recent extensive analysis of association between AD and a smaller region spanning only ANKK1 and DRD2 from the Collaborative Study on the Genetics of Alcoholism (COGA) showed that the strongest evidence of association is in a linkage disequilibrium (LD) block in the 5′ end of ANKK1 (Dick et al., 2007b). This LD block extends from an upstream intergenic single nucleotide polymorphism (SNP) to exon 2 of ANKK1, which to some degree replicated our finding of an association of AD to ANKK1 (Yang et al., 2007).

We previously reported association of SNPs in this region to two different SD disorders: viz. AD and ND. These results led us to ask whether one of these phenotypes (i.e., AD), or another that is highly comorbid with it (i.e., drug dependence (DD)), is driving the association. Further, some or all of the associations within this cluster that we reported could be attributable to a common risk factor for multiple forms of SD. Since dopamine is a major constituent of reward pathways and is strongly believed to be involved in many forms of SD (Connor et al., 2007; Sasabe et al., 2007; Shahmoradgoli Najafabadi et al., 2005; Volkow et al., 2006; Xu et al., 2004), this seems a likely possibility. One way to answer this question is to subdivide samples characterized by AD, based on the presence or absence of DD comorbidity. We performed this study with the intention to determine whether some, or all, of the risk loci we identified in this gene cluster are specific to AD or comorbid AD+DD.

Presumably, complex biological mechanisms are responsible for SD, and the heterogeneity of SD phenotypes reflects the existence of a spectrum of diseases with different mechanisms. The co-existence of many different SD disorders in an individual is common clinically; in the United States, the odds ratio of comorbid DD with AD, adjusted for demographic characteristics, is as high as 18.7 (Compton et al., 2007). Narrowing down the shared and specific genetic risk factors for different SD disorders would help to understand the biology of this multifaceted set of disorders. To advance such an understanding, in this study, we investigated the association of the NTAD gene cluster with comorbid of AD+DD, or AD alone.

Materials and methods

Subjects

We implemented case control and family study designs. All relevant institutional review boards approved the study protocol and related materials, and all subjects, who were paid for their participation, provided written informed consent. All subjects included in this study were also included in our previous study (Yang et al., 2007) but this sample is completely distinct and nonoverlapping from our previous study of this gene cluster in ND (cf. Gelernter et al 2006, page 3504, Subject recruitment and assessment).

Case-control Sample

A total of 716 EAs were recruited at either the University of Connecticut Health Center or the VA Connecticut Healthcare System-West Haven campus. Among them, 136 were affected with comorbid AD and DD (“AD+DD”), 166 were affected with AD and no comorbid drug dependence (“AD-only”) and 414 were healthy controls. Details of sex and age distributions are shown in Table 1. Each affected subgroup was examined separately in relation to the 414 control subjects for association analyses. AD+DD subjects met lifetime DSM-III-R [APA, 1987] or DSM-IV [APA, 1994] diagnostic criteria for AD and also met diagnostic criteria for cocaine dependence (CD) and/or opioid dependence (OD) (see Table 1). Healthy controls, recruited by advertisement from the general population, were screened to exclude subjects with AD, DD, major Axis I psychiatric disorders such as mood disorders, major anxiety disorders, and psychotic disorders (i.e., schizophrenia or schizophrenia-like disorders). The Structured Clinical Interview for DSM-III-R (SCID-III-R) or DSM-IV (SCID-IV), the Computerized Diagnostic Interview Schedule for DSM-III-R (C-DIS-R), the Schedule for Affective Disorders and Schizophrenia (SADS), or an unstructured interview were used for screening. Many of the subjects were not characterized with respect to ND, so this diagnosis could not be considered.

Table 1.

Demographic and clinical characteristics of the study sample.

Case-control sample
AD+DDa (total number)	136
Number of males (%)	93 (68.4)
Age (years)b	36.6 ± 7.8
AD-onlyc (total number)	166
Number of males (%)	135 (81.3)
Age (years)b	44.7 ± 9.8
Control (total number)	414
Number of males (%)	167 (40.3)
Age (years)b	27.9 ± 8.5
Family sample
Number of subjects (% male)	374 (50)
Number of nuclear families	112
Number of informative familiesb,d	42.9 ± 10.8
Family sizeb	3.3 ± 0.9
Number of parents	200
Number of siblings (% male)	174 (53.4)
Age (years)b
Parents	61.8 ± 10.1
All offspring	34.2 ± 8.1
Number with AD (%)	163 (43.6)
Number with CD (%)	99 (26.5)
Number with OD (%)	78 (20.9)

^aAD+DD indicates alcohol dependence (AD) comorbid with drug dependence (DD).

^bmean± standard deviation.

^cAD-only indicates AD without other DD.

^dnumber of informative families, defined as those with at least one heterozygous parent, for the 43 studied SNPs.

Family Sample

Four hundred eighty-eight EAs in 143 nuclear families were recruited at the same centers and were assessed using the SCID or the Semi-Structured Assessment for Drug Dependence and Alcoholism (SSADDA), as described elsewhere (Gelernter et al., 2005; Pierucci-Lagha et al., 2007; Pierucci-Lagha et al., 2005). Each member of the family sample was evaluated for comorbidity status. A family classified as AD+DD was one in which not all affected family members were AD-only. Thirty-one and 112 families were identified as AD-only and AD+DD families, respectively. The 31 AD-only families, which included a total of 114 subjects, provided insufficient statistical power for analysis, so association analyses were conducted only for the 374 subjects in the AD+DD families. Details of the demographic characteristic for the samples included here are presented in Table 1. The study sample included all subjects presented elsewhere (Yang et al., 2007), but 16 affected subjects from that analysis who had missing DD status were excluded from the present study.

Marker Selection and Genotyping

Forty-three SNPs in the NTAD gene cluster were selected (cf. Gelernter et al., 2006). These 43 SNPs (Figure 1(A)) were designated in sequence order (from the 5′ end of NCAM1 to the 5′ end of DRD2) as N1-N17 (mapped to NCAM1, 5′ to 3′), N18-N19 and T1-T2 (intergenic between NCAM1 and TTC12), T3-T9 (mapped to TTC12, 5′ to 3′), T10 (intergenic between TTC12 and ANKK1), A1-A7 (mapped to ANKK1, 5′ to 3′), D1-D4 (mapped to DRD2, 3′ to 5′), and D5-D7 (intergenic, 5′ to DRD2; for details refer to online supplemental Table 1). SNPs were genotyped as described previously (Gelernter et al., 2006). In addition, 37 ancestry informative markers (AIMs), as discussed in Yang et al. (2005a, 2005b), including 36 short tandem repeat (STR) markers and one SNP (rs2814778) in the Duffy antigen (FY) gene were genotyped in all cases and controls to infer ancestry. Two markers did not pass the HWE test. They are SNP N4 (p-value = 7.6×10⁻⁵) only in the control sample and SNP A4 (p-value = 2.3×10⁻⁶) only in the AD-only sample. Since causal variants in the affected sample could cause a failure of the HWE test, SNP A4 in the AD-only sample could be a risk-influencing variant. As for SNP N4, it fails the HWE test only in the control sample, and we kept this marker in the analysis for comparison purpose. Genotyping errors were detected and resolved (details described in Yang et al. (2007).

Figure 1.

Map of the NCAM1, TTC12, ANKK1 and DRD2 gene region. (A) Relative gene locations, and the distribution of 43 SNPs studied. (B) Associated regions identified in analyzing phenotypes of AD-all, AD+DD, AD-only and ND-all. Solid black or grey indicates a finding in the case-control (“c”) sample, black or grey with white stripes indicates a finding in the family (“f”). A finding in black color indicates significance both in global and individual specific haplotypic association; a finding in grey color indicates significance in global haplotypic association only. Each diamond above the block indicates the position of the corresponding SNP in the associated haplotype. The number underneath each block is annotated for the corresponding genetic region.

Statistical Analysis

Inferences of Linkage Disequilibrium and Haplotype Blocks

LD, haplotype blocks, and allele frequencies were estimated or inferred using the Haploview 3.32 program (Barrett et al., 2005). The LD plots encapsulate the standardized LD coefficients, D’, LD patterns and haplotype blocks (online supplemental Figures 1–4). Haplotype blocks in the LD plots were defined according to the criteria of Gabriel et al (2002).

Association Analysis for the Case-Control Design

A generalized linear model (GLM implemented in the R Project for Statistical Computing) was used to test single marker genotypic additive effects for the relevant traits (AD+DD or AD-only), adjusting for the covariates sex and age. Haplotype-based association analysis was also completed in a GLM framework using the Haplo.stats Package in R (Schaid et al., 2002). The permutation test on global haplotypic effect was implemented by randomly permuting the case vs. control status (null hypothesis of no association), computing and comparing the test statistic with the observed one, replicating the simulation an appropriate number of times, and summarizing with the p-value, which is the probability of the permutated test statistic being greater than the observed one. Only haplotypes with frequency greater than 5.0% were examined for the haplotype-specific tests. The critical level of significance for the haplotype-specific tests within each haplotype set was adjusted by Bonferroni correction, i.e., the conventional α = 0.05 was divided by the number of major haplotypes observed in each case. The LD contrast test (Zaykin et al., 2006) was employed to investigate differences in LD between cases and controls (cf. Yang et al., 2007, for more details), and to infer from the differences evidence of selection of risk variants in the disease group. Population stratification for the case-control sample was tested using the program Structure 2.1. More details regarding the analyses were reported elsewhere (Yang et al., 2007).

Association Analysis for the Family Design

The family-based association test (FBAT)(Horvath et al., 2001; Horvath et al., 2004) was used to test the association of the single SNPs and haplotypes of various SNPs to AD+DD. The FBAT model and parameters used included an additive genetic model under the null hypothesis of no linkage and no association, biallelic mode, minimum number of informative families of 20 for each analysis, and offset of zero. The “hbat” command in FBAT was issued to analyze haplotypic association with an assumption of a biallelic mode of inheritance for haplotype-specific association and a multi-allelic mode for testing all haplotypes as a whole for the global association test. Permutation tests (with 100,000 iterations) were used to calculate all p-values of global associations. The significance level of haplotype-specific tests was set up the same as that in the case-control design.

Results

For AD+DD in the case-control sample, seven nominally significant individual SNPs emerged: T6, T7, T8, A5, A6, D1 and D2 (p = 0.049 to 0.0025), though none was significant after Bonferroni correction, as shown in Table 2. The global haplotype analyses revealed three risk regions: almost the entirety of TTC12, the two merging 3′ regions of DRD2 and ANKK1, and around exon 12 of NCAM1.

Table 2.

Association analyses for AD-only and AD+DD for the case-control sample.

Starting SNPc	MAFa			p-valuesb				p-valuesb
	Control	AD-Only	AD+DD	AD-onlyd (166 cases) & control				AD+DDe (136 cases) & control
				1SNP	3SNP-hapf	4SNP-hap	5SNP-hap	1SNP	3SNP-hap	4SNP-hap	5SNP-hap
N1	0.295	0.330	0.335	0.96	0.97	0.91	0.67	0.14	0.42	0.94	0.94
N2	0.499	0.531	0.481	0.72	0.75	0.39	0.33	0.43	0.67	0.81	0.90
N3	0.352	0.314	0.366	0.9	0.13	0.23	0.38	0.20	0.42	0.74	0.79
N4	0.408	0.354	0.392	0.39	0.35	0.65	0.36	0.64	0.80	0.91	0.57
N5	0.221	0.203	0.225	0.82	0.54	0.21	0.37	0.71	0.76	0.37	0.40
N6	0.424	0.411	0.394	0.34	0.32	0.46	0.68	0.63	0.25	0.21	0.48
N7	0.426	0.417	0.366	0.32	0.52	0.83	0.71	0.24	0.28	0.59	0.86
N8	0.455	0.491	0.508	0.59	0.97	0.88	0.84	0.075	0.015	0.78	0.87
N9	0.198	0.220	0.174	0.54	0.92	0.91	0.84	0.39	0.70	0.90	0.95
N10	0.170	0.165	0.145	0.77	0.66	0.55	0.74	0.65	0.92	0.98	0.85
N11	0.413	0.439	0.402	0.4	0.37	0.57	0.51	0.75	1.00	0.62	0.43
N12	0.242	0.299	0.227	0.12	0.49	0.58	0.012	0.93	0.35	0.21	0.26
N13	0.143	0.146	0.132	0.97	0.39	0.24	0.72	0.73	0.49	0.42	0.97
N14	0.212	0.180	0.225	0.9	0.29	0.47	0.64	0.18	0.58	0.38	0.76
N15	0.110	0.142	0.105	0.068	0.41	0.58	0.28	0.70	0.91	0.89	0.72
N16	0.245	0.276	0.246	0.28	0.64	0.62	0.77	0.91	0.67	0.51	0.62
N17	0.387	0.394	0.376	0.65	0.95	0.64	0.56	0.87	0.32	0.57	0.10
N18	0.412	0.410	0.433	0.66	0.39	0.24	0.34	0.52	0.45	0.079	0.069
N19	0.286	0.279	0.315	0.9	0.79	0.40	0.28	0.19	0.066	0.011	0.057
T1	0.317	0.326	0.298	0.56	0.71	0.41	0.086	0.64	0.20	0.23	0.0041
T2	0.335	0.372	0.417	0.55	0.34	0.18	0.21	0.055	0.18	0.021	0.019
T3	0.221	0.244	0.260	0.41	0.12	0.26	0.36	0.59	0.022	0.017	0.013
T4	0.350	0.369	0.420	0.7	0.27	0.37	0.53	0.080	0.00055	0.0039	0.0011
T5	0.494	0.479	0.463	0.35	0.71	0.92	0.32	0.14	0.011	0.0032	0.0051
T6	0.326	0.346	0.419	0.45	0.77	0.51	0.41	0.037	0.0032	0.019	0.011
T7	0.141	0.161	0.184	0.29	0.83	0.22	0.29	0.0086	0.091	0.0090	0.020
T8	0.142	0.154	0.176	0.77	0.32	0.33	0.31	0.038	0.017	0.030	0.032
T9	0.223	0.233	0.272	0.43	0.14	0.11	0.14	0.91	0.016	0.033	0.013
T10	0.292	0.309	0.368	0.44	0.060	0.0730	0.054	0.088	0.023	0.010	0.011
A1	0.306	0.321	0.375	0.98	0.24	0.14	0.18	0.17	0.27	0.054	0.056
A2	0.309	0.299	0.365	0.52	0.22	0.23	0.34	0.52	0.39	0.11	0.054
A3	0.315	0.306	0.351	0.39	0.031	0.18	0.11	0.76	0.018	0.020	0.0016
A4	0.192	0.185	0.240	0.94	0.040	0.13	0.19	0.37	0.022	0.015	0.020
A5	0.479	0.478	0.602	0.97	0.23	0.67	0.75	0.0025	0.039	0.024	0.062
A6	0.283	0.332	0.357	0.41	0.79	0.61	0.091	0.049	0.11	0.29	0.15
A7	0.203	0.168	0.252	0.63	0.36	0.010	0.43	0.22	0.17	0.057	0.058
D1	0.438	0.478	0.530	0.42	0.0037	0.17	0.50	0.013	0.065	0.13	0.21
D2	0.359	0.412	0.472	0.39	0.58	0.72	0.36	0.0074	0.080	0.13	0.18
D3	0.149	0.140	0.176	0.79	0.43	0.57	0.54	0.53	0.17	0.26	0.017
D4	0.092	0.130	0.125	0.38	0.37	0.55		0.080	0.16	0.017
D5	0.369	0.325	0.309	0.2	0.47			0.057	0.019
D6	0.446	0.395	0.416	0.42				0.16
D7	0.369	0.319	0.342	0.26				0.33

^aMAF: minor allele frequency.

^bp-values for the haplotype tests were simulated global statistics and the significance level was set at 0.05, while p-values for the single allelic association were from the χ² test. The grey-marked p-values are significant.

^cStarting SNP for the haplotype analyses.

^dAD-only indicates alcohol dependence (AD) without other drug dependence (DD).

^eAD+DD indicates AD comorbid with DD.

^f“hap” abbreviates ”haplotype.”

In contrast, the analyses for AD-only (case-control) gave rise to significant global haplotypic association only in A3-A6, D1-D3, A7-D3, and N12-N16 (p = 0.04 to 0.0037; Table 2); however, no specific individual haplotype was significant. The association analysis for AD+DD in the family sample (Table 3) showed that global haplotypes were significantly associated in the three risk or protective regions, NCAM1 exon 12, TTC12 exon 3 and the junction of ANKK1 and DRD2, corroborating those identified in the global association analysis of AD+DD in the case-control sample. There were small differences for these AD+DD association findings between the family and the case-control samples in terms of the statistical significance level, the number of significant haplotypes, and the extent to which the haplotypes extended along the chromosome in the three regions (compare Table 2 with Table 3). These differences could be due to lower statistical power associated with the family study design, a relatively small sample size in that study, and sampling variation, among other possible explanations.

Table 3.

Association analyses for alcohol dependence comorbid with drug dependence (AD+DD) for the family sample.

Starting SNPa	Minor Allele	MAFb	p-valuec
			Single SNP	3SNP-hapd	4SNP-hap	5SNP-hap
N1	T	0.25	0.24	0.024	0.082	0.026
N2	G	0.46	0.52	0.79	0.60	0.29
N3	C	0.32	0.73	0.51	0.32	0.26
N4	C	0.33	0.88	0.31	0.32	0.15
N5	C	0.22	0.75	0.52	0.26	0.050
N6	G	0.44	0.37	0.068	0.026	0.0074
N7	C	0.44	0.37	0.039	0.011	0.0014
N8	T	0.47	0.12	0.016	0.0056	0.027
N9	A	0.18	0.010	0.0040	0.0049	0.0035
N10	C	0.16	0.95	0.48	0.15	0.29
N11	C	0.37	0.34	0.34	0.49	0.65
N12	C	0.21	0.46	0.96	0.99	0.96
N13	G	0.12	0.53	0.95	1.00	0.87
N14	A	0.25	0.97	1.00	0.75	0.66
N15	G	0.13	0.77	0.94	0.56	0.53
N16	T	0.21	0.89	0.59	0.90	0.62
N17	A	0.35	0.70	0.62	0.36	0.67
N18	C	0.40	0.85	0.17	0.74	0.89
N19	A	0.33	0.68	0.42	0.60	0.080
T1	A	0.26	0.037	0.28	0.019	0.012
T2	G	0.39	0.15	0.088	0.016	0.15
T3	G	0.27	0.34	0.024	0.18	0.17
T4	G	0.40	0.36	0.11	0.11	0.079
T5	C	0.47	0.073	0.27	0.18	0.39
T6	G	0.34	0.34	0.22	0.28	0.28
T7	G	0.15	0.72	0.57	0.53	0.35
T8	T	0.15	0.40	0.70	0.34	0.17
T9	G	0.23	0.51	0.70	0.54	0.51
T10	A	0.32	0.40	0.82	0.81	0.98
A1	C	0.33	0.38	0.71	0.83	0.83
A2	A	0.32	0.94	0.76	0.24	0.066
A3	T	0.32	0.69	0.23	0.078	0.12
A4	A	0.16	1.00	0.17	0.15	0.25
A5	C	0.47	0.11	0.22	0.053	0.053
A6	G	0.31	0.073	0.12	0.027	0.037
A7	A	0.16	0.60	0.042	0.015	0.026
D1	C	0.44	0.16	0.048	0.083	0.29
D2	C	0.39	0.081	0.028	0.19	0.16
D3	A	0.14	0.28	0.069	0.11	0.0093
D4	A	0.09	0.033	0.42	0.11
D5	G	0.40	0.19	0.17
D6	T	0.45	0.21
D7	T	0.40	0.28

^aStarting SNP for the haplotype analyses.

^bMAF: minor allele frequency.

^cp-values were simulated and the significance level was set at 0.05. The grey-marked p-values are significant.

^d”hap” abbreviates ”haplotype.”

As we moved our focus in the association analysis from the significant global haplotypes to the specific individual haplotypic associations, additional interesting results were revealed. Overall in TTC12, the risk haplotypes extended from T1 to T9 and covered the whole gene (individual haplotype simulated p-values (p_ihs) = 0.0094 ~ 0.000015 for the case-control sample; 0.0094 ~ 0.0034 for the family sample; Table 4). In the family sample, all the risk haplotypes shared the same polymorphic nucleotides at the same risk locus, and no risk haplotypes contained markers downstream of SNP T5, which is in TTC12 exon 3. In the case-control sample, more risk haplotypes were identified, and those with >10% difference in haplotype frequency between cases and controls centered around T4-T5, i.e., TTC12 exon 3 as well, and contained two “yin-yang” haplotypes “G-A” or “T-C”. The difference in haplotype frequency (between cases and controls) decreased as the risk haplotype was more distant from TTC12 exon 3.

Table 4.

Summary results of significant individual haplotypes associated with alcohol dependence comorbid with drug dependence (AD+DD) in the TTC12 region.

SNP#		T1	T2	T3	T4	T5	T6	T7	T8	T9	# of famb	Hap-freqc				t-stat/Zd	I-hap pe	Global p
Allelesa		G/A	A/G	A/G	T/G	A/C	A/G	A/G	G/T	A/G		pooled	control	case	diff
Case-Control	3-SNP				G	A	A				---	0.18	0.16	0.26	0.10	3.43	0.00065	0.00055
						A	A	G			---	0.12	0.11	0.15	0.04	3.56	0.00040	0.011
							A	G	T		---	0.15	0.14	0.18	0.04	3.10	0.0020	0.0032
					T	C	G				---	0.14	0.12	0.23	0.11	3.26	0.0012	0.00055
						C	G	A			---	0.18	0.16	0.25	0.09	3.51	0.00048	0.011
							G	A	G		---	0.35	0.33	0.41	0.08	2.72	0.0068	0.0032
	4-SNP		G	A	G	A					---	0.11	0.09	0.16	0.07	2.83	0.0049	0.021
				A	G	A	A				---	0.11	0.08	0.17	0.09	4.27	0.000023	0.017
				G	G	A	G				---	0.11	0.09	0.17	0.08	3.08	0.0022	0.017
					G	A	A	G			---	0.10	0.08	0.14	0.06	4.24	0.000027	0.0039
						A	A	G	T		---	0.12	0.11	0.15	0.04	3.65	0.00028	0.0032
							A	G	T	A	---	0.14	0.13	0.16	0.03	2.68	0.0076	0.019
				A	T	C	G				---	0.15	0.12	0.23	0.11	3.29	0.0011	0.017
					T	C	G	A			---	0.15	0.12	0.23	0.11	3.80	0.00016	0.0039
						C	G	A	G		---	0.18	0.16	0.25	0.09	3.81	0.00016	0.0032
	5-SNP		G	A	G	A	A				---	0.09	0.07	0.15	0.08	4.25	0.000025	0.019
			G	G	G	A	G				---	0.11	0.09	0.16	0.07	3.11	0.0019	0.019
				A	G	A	A	G			---	0.07	0.05	0.12	0.07	4.37	0.000015	0.013
				G	G	A	G	A			---	0.11	0.09	0.16	0.07	3.58	0.00038	0.013
					G	A	A	A	G		---	0.10	0.09	0.12	0.03	2.61	0.0094	0.0011
					G	A	A	G	T		---	0.10	0.08	0.14	0.06	3.95	0.000087	0.0011
						A	A	G	T	A	---	0.11	0.11	0.14	0.03	3.00	0.0028	0.0051
			A	A	T	C	G				---	0.14	0.12	0.23	0.11	3.24	0.0013	0.019
				A	T	C	G	A			---	0.15	0.12	0.23	0.11	3.56	0.00040	0.013
					T	C	G	A	G		---	0.15	0.12	0.23	0.11	3.71	0.00023	0.0011
						C	G	A	G	A	---	0.10	0.08	0.16	0.08	3.81	0.00015	0.0051
Family				G	G	A					33	0.26	---	---	---	2.80	0.0051	0.024
		G	G	G	G						32	0.24	---	---	---	2.60	0.0094	0.019
			G	G	G	A					32	0.25	---	---	---	2.68	0.0075	0.016
		G	G	G	G	A					30	0.24	---	---	---	2.93	0.0034	0.012

^aThe allele listed on top is the common allele. Black square with white allele denotes risk haplotypes.

^bNumber of informative families.

^cHaplotype frequency for the pooled (combined the case and control samples for the case-control design, but representing overall estimates of haplotype frequency for the family design), control, case, and the difference between cases and controls.

^d“t-stat” is the t-statistic calculated for the case-control sample, and “Z” for the family sample.

^eIndividual haplotype p-value.

Another region important for AD+DD was observed where the 3′ regions of DRD2 and ANKK1 meet (Table 5). The risk variants were the significant individual risk haplotype G-G-G-G-G in A6-D3 (p_ihs = 0.0093) identified in the family sample, or G-C-G-G-G in A4-D1 (p_ihs = 0.0087) in the case-control sample. All 11 risk haplotypes (p_ihs = 0.017 ~ 0.0028) share the same polymorphic nucleotides at the corresponding SNP loci, consistent with all of them identifying the same risk locus, i.e., the locus was tagged with the same set of markers. The various risk haplotypes, between samples from the two designs, overlapped for 3 SNPs (A6-D1). This common region includes the region of DRD2^C957T (rs6277) in exon 7 of DRD2, and SNP-A6 and SNP-A7 (rs4938016 and rs1800497) in exon 8 of ANKK1.

Table 5.

Summary results of significant individual haplotypes associated with alcohol dependence comorbid with drug dependence (AD+DD) in the ANKK1 and DRD2 regions.

SNP#	A3	A4	A5	A6	A7	D1	D2	D3	# of fam b	Hap-freqc				t-stat/Zd	I-hap pe	Global p
Allelesa	C/T	G/A	C/T	C/G	G/A	A/G	A/G	G/A		pooled	control	case	diff
Case-control		G	C	G					---	0.30	0.28	0.35	0.07	3.01	0.0028	0.022
			C	G	G				---	0.29	0.27	0.34	0.07	2.99	0.0029	0.039
	C	G	C	G					---	0.13	0.12	0.16	0.04	2.81	0.0051	0.020
		G	C	G	G				---	0.28	0.27	0.33	0.06	2.76	0.0060	0.015
			C	G	G	G			---	0.28	0.27	0.34	0.07	2.77	0.0058	0.024
		G	C	G	G	G				0.28	0.27	0.33	0.06	2.63	0.0087	0.020
Family					G	G	G		34	0.24	---	---	---	2.40	0.017	0.042
						G	G	G	34	0.24	---	---	---	2.76	0.0059	0.048
				G	G	G	G		30	0.24	---	---	---	2.73	0.0064	0.027
					G	G	G	G	33	0.24	---	---	---	2.59	0.0096	0.015
				G	G	G	G	G	29	0.24	---	---	---	2.60	0.0093	0.037

^aThe allele listed on top is the common allele. Black square with white alleles in a row denotes risk haplotypes.

^bNumber of informative families.

^cHaplotype frequency for the pooled (combined the case and control samples for the case-control design, but representing overall estimates of haplotype frequency for the family design), control, case, and the difference between cases and controls.

^d“t-stat” is the t-statistic calculated for the case-control sample, and “Z” for the family sample.

^eIndividual haplotype p-value.

The third susceptibility region for AD+DD occurred at exon 12 of NCAM1 (Table 6); this association was only observed in the family sample. In the case-control sample, we detected a globally significant 3-SNP haplotype (p = 0.015, Table 2), but no specific individual associated haplotype was found. In the family sample, there was evidence for risk (G-A-A) or protective (T-C-A) alleles in the haplotype N8-N10 (Table 6, p = 0.016 ~ 0.0029), located around exon 12 of NCAM1. All of the protective haplotypes contained the same nucleotides at the corresponding SNP loci, consistent with tagging of the same marker locus, and all had the “C” nucleotide in SNP-N9. The risk haplotype N8-N10 of the “G-A-A” allele (p = 0.0036) had a haplotype frequency (12%) that was lower than those of several protective haplotypes (22% ~ 42%) in this family sample.

Table 6.

Summary results of significant individual haplotypes associated with AD+DD in the NCAM1 region.

SNP#		N6	N7	N8	N9	N10	N11	N12	N13	# of famb	Hap-freqc	t-stat/Zd	I-hap pe	Global p
Allelesa		A/G	T/C	G/T	C/A	A/C	T/C	T/C	T/G
Family	3-, 4- or 5-SNP haplotypes			G	A	A				25.1	0.12	2.91	0.0036	0.016
				T	C	A				52	0.40	−2.45	0.014	0.016
					C	A	T			37	0.42	−2.40	0.016	0.0040
			T	T	C	A				45.4	0.38	−2.82	0.0048	0.011
				T	C	A	T			30.8	0.24	−2.58	0.0098	0.0056
					C	A	T	T		37.8	0.37	−2.43	0.015	0.0049
		A	T	T	C	A				44.4	0.38	−2.98	0.0029	0.0074
			T	T	C	A	T			27.2	0.22	−2.90	0.0037	0.0014
				T	C	A	T	T		28.3	0.23	−2.55	0.011	0.027
					C	A	T	T	T	36.8	0.36	−2.90	0.0038	0.0035

^aThe allele listed on top is the common allele. Black square with white alleles in a row denotes risk haplotypes, while white square with black alleles in a row denotes protective haplotypes.

^bNumber of informative families.

^cHaplotype frequency for the pooled (combined case and control samples for the case-control design, but representing overall estimates of haplotype frequency for the family design), control, case, and the difference between cases and controls.

^d“t-stat” is the t-statistic calculated for the case-control sample, and “Z” for the family sample.

^eIndividual haplotype p-value.

LD comparisons between cases and controls for AD+DD and AD-only were performed using LD Contrast Test software. Two regions, one around TTC12 exon 3 (SNPs T2 to T5) and another around ANKK1 exon 2 (SNPs T10 to A3), were identified as differing in the standardized composite LD for AD+DD (p = 0.018 and 0.0012, respectively) and AD-only (p = 0.00005 and 0.00013, respectively). The LD difference gradually disappeared as the number of SNPs that were included increased. The differences in pairwise LD matrices are displayed using ellipse plots (Figure 3) in which the shape of an ellipse indicates the magnitude of LD and the direction shows the sign of the disequilibrium. The more circular ellipse reflects a low degree of LD, and the ellipse oriented at 45° suggests a positive value of LD. The two regions, identified by the LD contrast test, are composed of markers in stronger LD among cases than controls. This phenomenon occurs when selection causes transmission of more disease alleles and/or haplotypes in common among cases than controls. The two identified regions of differential LD between the two groups further corroborate the results of haplotype association analysis and provide evidence of harboring disease variants in the regions.

Figure 3.

Ellipse plots for the results of the LD contrast tests. For each ellipse plot, the pairwise LD matrices are displayed for the contrast of cases and controls, respectively, above and below the diagonal line. The shape of an ellipse indicates the magnitude of LD and the direction shows the sign of the disequilibrium. The more circular shape of an ellipse reflects a low degree of LD, and 45o-oriented ellipses suggest a positive value of LD.

Population stratification originating from cryptic genetic patterns was examined using a panel of 37 short tandem repeat (STR) markers that we designed to distinguish among major US populations (Yang et al., 2005a; Yang et al., 2005b). There was no evidence of population stratification in these study samples. We conclude that the significant association findings in the case-control sample could not be attributed to population stratification.

In summary, we found evidence that (1) three regions in EAs harbor variants predisposing to AD comorbid with DD: exon 3 in TTC12 (the region for which evidence for association is the strongest and most consistent), the two adjacent 3′ regions of DRD2 and ANKK1 (i.e., between exon 7 of DRD2 and exon 8 of ANKK1), and risk or protective variants close to exon 12 to intron 13 of NCAM1; (2) For AD without DD comorbidity, NCAM1 intron 14 to 18 and the junction of ANKK1 and DRD2 were associated globally; no individual haplotype was significantly associated. We showed in our previous study that overall, risk for AD, irrespective of comorbidity of DD, was attributable in part to variants in four regions: exon 3 of TTC12, exon 12 of NCAM1, and exons 2 and 5 of ANKK1, but not the junction of DRD2 and ANKK1.

Discussion

Findings from this study support the hypothesis that heterogeneity of the SD disorders studied and their high rates of comorbidity are reflected in genetic association findings. The contrast among the findings for the four phenotypes – AD+DD, AD-only, AD irrespective of DD comorbidity (AD-all), and ND irrespective of AD or DD (ND-all) from the current and prior studies, provides important insight not evident when the previous studies are considered individually, as summarized in Table 7 and Figure 1(B). In a previous article (Gelernter et al., 2006), we reported very strong evidence of association of ND (ND-all) to a single haplotype spanning TTC12 and ANKK1. In another recent article (Yang et al., 2007), we reported that risk for AD (AD-all) is attributable in part to variants in four regions within the NTAD gene cluster: exon 3 of TTC12, exon 12/intron13 of NCAM1, and exons 2 and 5 of ANKK1. LD contrasts between cases and controls supported historical selection at TTC12 exon 3 and ANKK1 exon 2, and corroborated the identified risk variants from the association analysis. Further, the risk variants in TTC12 suggested a fundamental neurobiological basis for SD that involves dopaminergic neurotransmission, since the armadillo repeat domain encoded by TTC12 interacts with dopamine through the Wnt signal transduction pathway and may have effects on dopamine neuronal development in the ventral midbrain (Castelo-Branco and Arenas, 2006; Castelo-Branco et al., 2006). In the present article, we examined the effects of DD comorbidity on the observed association, with the goal of evaluating its specificity. To that end, we undertook analyses comparing AD with or without DD comorbidity (AD+DD vs AD-only) using the same 43 SNP markers in the NTAD gene cluster. By doing so, we hoped to learn whether AD or DD was the primary driver of the observed association.

Table 7.

Comparison of the three studies using the same 43 SNPs spanning NCAM1-TTC12-ANKK1-DRD2 gene cluster.

Study	Gelernter et al, 2006	Yang et al, 2007	Current study
Phenotypes	ND-alla	AD-allb	AD+DDc AD-onlyd
Study Design	Family	Family, Case-control	Family, Case-control
Sample Size	1615 subjects	1220 subjects	1090 subjects (subset of the sample in Yang et al, 2007)
Ethnicity	EAe and AAf	EA only	EA only
Sample allocation	319 AA families 313 EA families (independent of the other two studies)	143 families 318 affecteds 414 controls	112 families 136 AD+DDs 166 AD-onlys 414 controls
Haplotypic association analysis	SNPg selection based on the single SNP association findings and the LDh block structure. Four SNPs were selected for haplotype analysis.	Exhaustive search for haplotypic association by sliding window	Same as Yang et al, 2007
Results: regions with evidence of association	Multiple SNPs at TTC12 and ANKK1 in both populations. A single haplotype spanning TTC12 and ANKK1.	TTC12 exon 3, NCAM1 exon 12, ANKK1 exon 2 & exon 5	For AD+DD, TTC12 exon 3, the two adjacent 3′ regions of DRD2 and ANKK1, NCAM1 exon 12 to intron 13 For AD-only, NCAM1 intron 14 to 18 and the junction of ANKK1 and DRD2 were associated globally.

^aND-all denotes nicotine dependence (ND) irrespective of other substance dependence (SD).

^bAD-all denotes alcohol dependence (AD) irrespective of other SD.

^cAD+DD denotes AD comorbid with DD.

^dAD-only denotes AD without other DD.

^e“EA” abbreviates “European American.”

^f“AA” abbreviates “African American.”

^g“SNP” abbreviates “single nucleotide polymorphism.”

^h“LD” abbreviates “linkage disequilibrium.”

The results of these analyses suggest that partitioning the total AD sample reduced genetic heterogeneity. Figure 1 summarizes the location of the four genes in the NTAD gene cluster, the distribution of the 43 SNPs, and the identified susceptibility regions. For AD-all, risk regions were consistent in both samples, with the exception of NCAM1 exon 12 (which was identified as associated only in the family sample). In particular, TTC12 exon 3 was significant in both global and individual haplotypic associations and in both samples. A broader region from ANKK1 exon 2 to DRD2 intron 1 was significant in the global association analyses in both samples, and was narrowed to ANKK1 exons 2 and 5 in the individual specific haplotypes for the case-control sample.

The junction of the 3′ ends of ANKK1 and DRD2 was globally associated with all three phenotypes involving AD (Figure 1), but harbored specific risk variants only for AD+DD. The global significance might be due to a few weak or moderate variants in this region but the individual effect was not strong enough to meet the criterion for statistical significance. A larger sample size for AD-only would be helpful to examine this hypothesis and to detect smaller specific effects from this region. On the other hand, the heterogeneity of AD-all statistically may well have diluted the association signal in this region. Thus, it appears that those strong effects depend on the presence of comorbidity, which might reflect a different neurobiological mechanism or more severe form of substance dependence. This result underscores the importance of the phenotype definition, and possible effects of comorbidity, in studies of complex traits such as substance dependence.

The results for variants mapped to TTC12 for AD+DD were similar to our previously published results in the AD-all analyses, but yielded much greater statistical significance (e.g., with the p-value being reduced from 0.00067 to 0.000023 for the A-G-A-A haplotype at markers T3-T6; see Table 5), and additional specific haplotypes being significant, whereas in the earlier analysis they were not. On the other hand, TTC12 was not observed to harbor risk variants for AD-only, a finding that was probably not due to lack of power, as the sample size was greater for AD-only than for AD+DD.

Interestingly, the LD contrast test supports selection at TTC12 exon 3 for AD-only (p-value = 0.00005), and the LD contrast plot shows the difference between cases and controls (see Figure 3C). “One situation when the LD contrast test is particularly powerful is when dissimilar haplotypes (‘yin yang haplotypes’) carry susceptibility. This could either indicate a true heterogeneity of the mechanism, or that they are ‘proxies’ for an unobserved causative mutation or haplotype.” (personal communication from Dr. Dmitri Zaykin, January 12^th, 2007). The haplotypes in TTC12 exon 3 in this sample are exactly “yin-yang haplotypes.” Therefore, it is unknown why there was selection at TTC12 exon 3 for AD-only but no evidence for association emerged. Additional studies with large samples are necessary to validate these findings. It should be noted, however, that in studies from COGA, evidence for an association of GABRA2 and CHRM2 previously observed with AD came only from individuals with comorbid DD, with no evidence of association among individuals with AD but no DD (Agrawal et al., 2006; Dick et al., 2007a). We conclude that TTC12 exon 3 is a risk region for AD+DD, with evidence of association with AD-all due only to the contribution by the comorbid AD+DD part of the sample. This suggests that the analysis of samples with drug dependence in the absence of alcohol dependence is warranted, to ascertain whether the locus predisposes to drug dependence specifically or, as suggested here, to the comorbid phenotype.

The major risk variants identified in the ND study delineate a haplotype spanning TTC12 and ANKK1. This haplotype is composed of four SNPs, T6-A1-A3-A5 (Figure 1), where T6 is very close to the identified risk region in TTC12 exon 3 from the other two studies, and the other 3 SNPs are located in the small ANKK1 region extending from intron 1 to exon 6. This study was analyzed by a different approach, without the “sliding window” exhaustive search for risk variants employed in the other two studies. The identified risk variants were very strongly supported in terms of their statistical significance and were close to the risk regions from the other two studies. In addition, since the ND sample was ascertained for a sib-pair linkage analysis for the primary traits of cocaine and opioid dependence, there is especially high comorbidity with DD in this sample.

We conclude that exon 3 of TTC12, and the region close to exon 12 of NCAM1, both regulate risk for AD+DD. Variants at the two 3′ ends of ANKK1 and DRD2 also regulate risk for AD, with effects depending on comorbidity with DD. The complexity of these relationships, many of which were replicated in our independent samples, may explain prior inconsistent results. Further replication in other samples and functional analysis of this allelic variation are warranted.

Supplementary Material

SupMat 1

Online Supplemental Figures:

LD map structure and locations of 43 SNPs in the NTAD gene cluster. The LD maps for the family (AD+DD) and case-control samples (AD+DD, AD-only and control) are shown. The measure of LD (D′) among all possible pairs of SNPs is shown graphically according to the shade of red where white represents very low D′ and dark red represents very high D′.

Supplemental Figure 1. Cases: AD+DD

SupMat 2

Supplemental Figure 2. Cases: AD-only

SupMat 3

Supplemental Figure 3. Controls

SupMat 4

Supplemental Figure 4. AD+DD Family sample

Figure 2.

Results of single SNP and haplotype association analyses for the case-control samples of AD+DD and AD-only, and AD+DD of family samples. Global simulated p-values are shown. For haplotype analyses, the value at each SNP location represents the haplotype with SNPs starting from the current location downward from the NCAM1 direction to DRD2. The horizontal dashed lines correspond to type I error rate of 0.05. The three vertical lines from left to right indicate the positions of SNPs N19, T10 and A7, respectively.