Association of Specific Haplotypes of D2 Dopamine ReceptorGene With Vulnerability to Heroin Dependence in 2 Distinct Populations FREE

Although epidemiologic studies have shown that heroin dependence isstrongly influenced by genetic factors (h² = 0.54),¹ thenumber and identity of susceptibility genes remain unknown. Animal and humanstudies of addiction indicate that the D₂ dopamine receptor (DRD2)plays a critical role in the mechanism of reward and reinforcement behavior.Opiate rewarding effects were absent in mice lacking D₂ receptors,^2- 3 while DRD2 overexpressionin transgenic mice led to reduced self-administration of alcohol.⁴ A positron emission tomography study of human brainshowed that D₂ receptor density in the brain decreased significantlyin alcoholic compared with control subjects.⁵ Thesefindings suggest that genetically determined variation in DRD2 expression and function can alter reward responses to a varietyof substances and may contribute to vulnerability to heroin dependence inhumans.

DRD2 is located on 11q22-23 and is composedof 8 exons spanning 65.8 kilobases (kb) of genomic DNA.⁶ Thefirst DRD2 genetic marker characterized was a singlenucleotide polymorphism (SNP) originally detected as a restriction fragmentlength polymorphism (TaqIA) located 10 kb distalto the 3′ end of the gene.⁷ This markerwas extensively used in genetic association and linkage investigations ofaddiction, with controversial results.^8- 17 Linkageof the TaqIA restriction fragment length polymorphismwas also evaluated in many other psychiatric disorders, also with varyingresults.¹¹ Studies using known functional DRD2 SNPs (−141ins/delC and 311 Ser>Cys) in alcohol dependence and a mixture of othersubstance dependencies, detected no association with risk.¹⁸

Among the causes of controversial findings in population-based studiesare small sample size with reduced power to detect effect, linkage disequilibrium(LD) of associated markers with other unknown functional loci, and populationstructure (admixture). To detect association with moderately abundant alleles,the LD (also known as allele-based linkage) paradigmwith functional alleles or highly informative haplotypes offers substantiallygreater power for mapping complex disease or trait genes than does the locus-basedlinkage approach.^19- 20 Linkagedisequilibrium detects the physical correlation between the genetic markersthat define a group of alleles or haplotype. A haplotype block defines a regionof the genome showing little historical recombination. Thus, a panel of 5to 6 moderately informative SNPs contained within a haplotype block capturesthe effect of any relatively abundant, but unknown, functional allele withinthe haplotype block.²¹ Haplotype associationalso has the advantage of narrowing the location of disease loci and reducingor clarifying discrepancies in results between studies using different populations,allowing disparate data to be reconciled or at least better understood. Thus,haplotype-based association becomes an important approach to investigate therelationship of DRD2 and addictive behavior, nowthat a detailed SNP map from public and private databases (ie, Celera DiscoverySystem, Rockville, Md) is available for this gene.

As mentioned earlier, population structure has been thought of as oneof the reasons to explain unreplicated results from population-based associationstudies.²² When case and control samples arecollected from different subpopulations, allele frequencies will tend to differfor most randomly chosen loci. Admixed populations can be detected by genotypinga number of markers and detecting systematic differences in allele frequencywithin the study population. Simulated analyses have suggested that 30 SNPmarkers should have reasonable power to detect stratification in subpopulations.²³ With this approach, population admixture in AfricanAmerican subpopulations has been detected.²⁴

To better understand whether DRD2 is associatedwith substance abuse, our strategy was to use a combined haplotype–functionallocus approach in 2 large, ethnically well-defined heroin-dependent case-controlsamples derived from Chinese Han and German populations. To control for samplestratification, we genotyped 32 SNPs in our case and control groups for theChinese and German populations. In addition, we genotyped an admixed AfricanAmerican population with the use of the same SNP panel as a positive referencesample set. To our knowledge, this is the first large-scale haplotype analysisof DRD2 in heroin dependence that controls for populationadmixture.

A total of 799 subjects, composed of 486 heroin-dependent cases and313 unrelated and unaffected controls, were recruited in 3 waves during 1996,1997, and 1999 from southwestern China, including Sichuan Province and ChongqingCity, a federal district that is geographically adjacent to Sichuan Province.The Chinese Han population and data collection were described in more detailelsewhere.²⁵ Patients were interviewed withthe Structured Clinical Interview for DSM-III-R AxisI disorders, and diagnosed as opiate dependent by 2 psychiatrists using DSM-IV criteria. Other substance abuse, such as cocaineand cannabis, was uncommon in this area. Control subjects were recruited fromstudents and staff at a local medical university. Control subjects were askedonly if they had had a mental disorder, had been prescribed medication fora mental illness, or used a drug for a nonmedical purpose. The mean ±SD age of heroin-dependent subjects was 27.3 ± 5.80 years, and thatof controls was 28.0 ± 10.0 years. Informed consent was obtained undera human research protocol approved by ethics committees at the 3 local hospitalsand 1 local medical school.

A total of 663 individuals were recruited, including 471 heroin-dependentsubjects from 2 western German cities, Mainz and Bonn, and 192 unrelated controlsfrom Bonn. Both cities are situated along the Rhine River, where they areseparated by 150 km and have similar population structure. In Germany, citizensare obliged to register births and relocations with local authorities. Samplecollection took place between 1993 and 2001 as part of a study on geneticand psychosocial risk factors in alcohol and heroin dependence. Cases wereconsecutive inpatients of the university hospital detoxification units atMainz (1993-1995) and Bonn (1996-2001). Subjects were interviewed by seniorpsychiatrists using the Semi-structured Assessment for the Genetics of Alcoholismfor psychiatric disorders and diagnosed as opiate dependent by DSM-III-R . Unrelated controls were randomly ascertained from the populationregistries of Bonn and represent the local population. The mean ± SDage of cases was 30.2 ± 6.8 years and that of controls was 31.8 ±7.0 years. Control subjects were contacted by mail or telephone by the sameresearch staff who recruited case subjects. Evaluation was the same as forthe case group, including Semi-structured Assessment for the Genetics of Alcoholismdiagnostic interview for psychiatric disorders. Informed consent was obtainedunder a human research protocol approved by the ethics committees at the Universityof Mainz and the University of Bonn.

Genotyping was performed by 5′-exonuclease fluorescence assay.²⁶ We developed 10 SNP assays for DRD2 genotyping. From 5′ end to 3′ end, these 10 SNPs wereas follows: −241 A>G, −141ins/delC, TaqIB A>G,TaqID G>A, intron 4 T>C, intron 6 ins/del G, 311 Ser>Cys, 20236 C>T, exon8 22640 C>G, and TaqIA G>A. Their correspondingNational Center for Biotechnology Information SNP identification, Celera DiscoverySystem identification, and primer-probe sequences are available from the correspondingauthor on request. The SNP locations are shown in Figure 1.

For each SNP, genotyping error rates were determined by duplicate genotypingof an additional 10% of the samples randomly selected from each reaction plate.

To detect population structure, an additional 32 SNPs were chosen fromthe National Center for Biotechnology Information public database. The SNPswere distributed on 17 chromosomes in the genome. The SNP rs identificationsand physical locations are available from the corresponding author on request.Most markers showed large differences in allele frequencies across 8 differentpopulations (K.X., unpublished data, 2003). Because DNA was available in limitedamounts in Chinese samples, we were not able to genotype the entire sampleset for all 32 markers. Therefore, we randomly selected 106 individuals fromcontrol (46) and case (60) groups for genotyping with this SNP set. We genotyped194 control individuals and 286 case subjects in the Germans. In addition,we genotyped 174 African American individuals who were known to representan admixed population for use as a reference by using the same 32-marker SNPset.

For individual SNP association analyses, genotype and allele frequenciesin cases and controls were compared by χ² tests on 2 ×3 and 2 × 2 categorical tables constructed for each population. To excludefalse-positive results due to multiple testing, Bonferroni correction wasused. The P values were multiplied by the total numberof loci genotyped (10). This was recognized to be a conservative correctionbecause of extensive LD across DRD2.

For LD analysis, D = P_AB − (P_A × P_B), where D is a parameter of LD, P_AB is the expected haplotype frequency, and P_A and P_B are observedfrequencies for alleles at loci A and B, respectively. D′ is D normalized against the maximumvalue of D possible, given allele frequencies P_A and P_B. D′ for each DRD2 SNP pairwas computed with the help of PAIRWISE software (Jeffery C. Long, PhD, Universityof Michigan, Ann Arbor).

Ten-SNP DRD2 haplotype frequencies were inferredseparately for cases and controls in each population by means of an expectation-maximizationalgorithm implemented in MLOCUS.²⁷ A likelihoodratio test for global haplotype effects (G) was performedwith the following equation: G = 2x[Ln total − (Ln case+ Ln control)], where Ln indicates the natural log.²⁷ Specific haplotype frequencies were compared betweencases and controls by χ² test or by Fisher exact test whenexpected frequencies were less than 5 in more than 20% of total categories.

We performed a contingency χ² test for comparing allelefrequencies for each marker and all markers between case and control groupsin each population. Under the null hypothesis that the populations have thesame allele frequencies, the sum of the statistics for all of the markershas a χ² distribution with degrees of freedom equal to 1 lessthe total numbers of SNPs. We also compared overall the allele frequency among3 control populations: Chinese, Germans, and African Americans.

We used the computer program Structure²⁸ inan attempt to identify clusters of genetically similar individuals from multilocusgenotype data.

Genotypes determined by 5′-exonuclease assay for the 10 DRD2 SNPs were highly accurate. Genotype discrepancy ratesacross the 10 loci were only 0.021 ± 0.018 in the Chinese and 0.010± 0.017 in the Germans. No significant deviation from Hardy-Weinbergexpectations occurred in Chinese controls, German controls, or German cases.In Chinese cases, SNP8 showed a slight departurefrom Hardy-Weinberg equilibrium that remained marginally significant (P = .05 after correction for multiple testing).

Cases and controls were compared for genotype and allele frequenciesacross the 10 DRD2 markers (Table 1). In the Chinese population, genotype frequencies at 4 sitesdiffered significantly between cases and controls. The significant sites were SNP1 (−214 A>G; P = .042); SNP3 (TaqIB A>G; P = 1.71 × 10 ^{− 4}), SNP4 (TaqID G>A; P = .01), and SNP5 (intron4 T>C; P = .024). Only the SNP3 genotype frequency remained significant after applying a conservativeBonferroni adjustment for multiple testing (P = 1.71× 10⁻³). Allele frequency comparisons between casesand controls were significant at 4 sites: SNP2 (−141ins/delC; P = .002), SNP3 (TaqIB A>G; P = 3.8 × 10⁻⁵), SNP4 (TaqID G>A; P = .006), and SNP5 (intron 4 T>C; P = .005). Only 2 SNPs remained significant after Bonferroni correction: SNP2 (−141ins/delC; P = .02) and SNP3 (TaqIB; P = 3.8 × 10⁻⁴).In the German population, genotype and allele frequency comparisons were notsignificantly different between case and control groups across 10 markers.

Pairwise LD for the 10 SNPs at DRD2 is presentedseparately for cases and controls from each population in Figure 2. Values on the abscissa and ordinate are physical distances(logarithmic scale). Levels of D′ are color coded. The LD was extensiveand was increased from 5′ to 3′ in both populations. However,overall levels and patterns of D′ differedbetween populations and between clinical diagnoses. In Chinese controls, 24%of SNP pairs were in complete LD (D′>0.99),while in Chinese cases only 4% of pairs were in complete LD. In German controls,13% of SNP pairs were in complete LD, while in German cases 42% of pairs werein complete LD. In both populations, 2 SNPs within the promoter region (SNP1 and SNP2) presented weakLD with the other 8 SNPs (SNP3 to SNP10) in the rest of DRD2 region. The 8 SNPs(SNP3 to SNP10) spanned25.8 kb, with high LD levels displayed in both Chinese and Germans (D′ = 0.804 ± 0.196 in Chinese; D′ = 0.801 ± 0.228 in Germans). A core conservative LDblock included 6 SNPs (from SNP4 to SNP9) spanning 10.8 kb with strong D′ (0.934± 0.069 in the Chinese; 0.897 ± 0.174 in the Germans). The strengthof LD provided a justification to divide the entire region into discrete windowsfor subsequent haplotype-based association analyses (Figure 2).

We used 3 SNP sets for haplotype-based association analyses, groupingSNPs on the basis of the level of LD strength. We used 6-SNP (SNP set 4-9),8-SNP (SNP set 3-10), and 10-SNP (SNP set 1-10) sets to create the windowsfor performing each of 3 separate analyses.

In 6-SNP core haplotype block, there were 2 configurations (A: 111121;and B: 112112) that accounted for 89% of all chromosomes in Chinese subjects(highlighted in yellow and blue, respectively in Figure 3). The global haplotype pattern differed significantly betweencontrols and heroin-dependent subjects (G₁₀ = 129.7, P<1.771 × 10⁻¹⁰ after multiple test correction).However, specific haplotypes in the 6-SNP block did not differ significantly(Figure 3).

Adding 2 flanking loci, SNP3 (TaqIB A>G) and SNP10 (TaqIAG>A), that were in strong LD with the markers in the 6-SNP block increasedthe information content within the 25.8-kb region defined by this window.With the use of 8 SNPs, 6 haplotypes were generated and grouped as 4 majorhaplotype clusters: A, B, C, and D (Figure3). Each was defined by the core 6-SNP haplotype and by 1 alleleof each of the 2 flanking SNPs: TaqIB, at the 5′end, and TaqIA, at the 3′ end of the haplotypeblock. Tests for global haplotype association with heroin dependence weresignificant for the 8-SNP haplotype (G₉ =322.3, P<4.720 × 10⁻¹⁰,after multiple test correction). Among 4 haplotype clusters, 2 haplotype clusters,8S-A and 8S-C, were observed in cases but not in controls (frequency, 0.149vs 0.000; Fisher P = 1.425 × 10⁻²²; odds ratio [OR], 52.80; 95% confidence interval [CI], 7.290-382.5for cluster A; 0.063 vs 0.000, Fisher P = 3.471 ×10⁻⁹; OR, 40.19; 95% CI, 5.550-291.1 for cluster C). In contrast,haplotype cluster 8S-B, corresponding to 10S-B (see below), was at higherfrequency in controls than cases (0.460 vs 0.347; P =1.140 × 10⁻⁵; OR, 0.667; 95% CI, 0.456-0.857). Withthis approach, it became apparent that both adjacent SNPs (Taq1B and Taq1A) at opposite ends of the blockadded critical information, thus localizing the effective locus to this 25.8-kbregion. For example, allele 1 (shown in red in Figure 3) of the TaqIB locus combined with111121 (coded yellow in Figure 3)defined a high-risk haplotype for heroin dependence. Without the added informationfrom TaqIB allele 1, haplotype 2111211 appeared tobe low-risk (coded as green and yellow in Figure 3). On the basis of frequencies of alleles and haplotypes,the TaqIB appeared to add more predictive informationthan TaqIA.

Finally, using all 10 available SNPs, we simultaneously evaluated theentire 75.8-kb region. The global 10-SNP haplotype test for association wassignificantly different between control subjects and heroin addicts (G₁₄ = 237.2, P<1.916× 10⁻¹⁰ after multiple test correction). Two of theclusters, 10S-A and 10S-C, corresponding to 8S-A and 8S-C, were observed onlyin cases (cluster 10S-A: frequency, 0.119 in cases vs 0.000 in controls, Fisher P = 2.499 × 10⁻⁹; OR, 77.79; 95%CI, 4.793-1268; and cluster 10S-C: 0.036 in cases vs 0.000 in controls, Fisher P = .001; OR, 22.43; 95% CI, 1.340-375.3), while 10S-Bwas significantly more abundant in controls (0.422 in controls vs 0.326 incases; P = .013; OR, 0.678; 95% CI, 0.499-0.922).These data suggested that haplotype clusters 10S-A and 10S-C represented high-riskcopies of DRD2, while haplotype cluster 10S-B mayrepresent low-risk copies of DRD2 with heroin dependence.

Applying the same strategy used with the Chinese dataset, 3 SNP haplotypesets were analyzed for association with heroin dependence in Germans (Figure 4). Overall haplotype tests showedthat DRD2 was significantly associated with heroindependence in 3 SNP haplotype set analyses (G₆ = 105.0, P<1.617 × 10⁻¹⁰ for 6-locus after Bonferronicorrection; G₇ = 134.0 P<1.000 ×10⁻¹⁰ for 8-locus after Bonferroni correction; G₁₀ =138.4, P<4.570 × 10⁻¹⁰ for10-locus after Bonferroni correction).

As seen previously in the Chinese, a 6-SNP core haplotype block wasobserved in both German cases and controls. Within the core haplotype block,3 major 6-SNP haplotypes (6S-H1, 6S-H2, and 6S-H3) accounted for 79% of thechromosomes in controls and 91% in cases. Two major haplotypes (6S-H2: 111121;and 6L-H3: 112112) were identical to haplotypes in the Chinese (yellow andblue blocks, Figure 3) but the mostfrequent core haplotype (6S-H1: 221112; Figure4) in Germans differed from that in the Chinese, accounting for47% of Germans but representing only 7% of Chinese. We observed a possiblerecombination event in the German population between SNP5 and SNP6 produced from the 2 abundant haplotypes,6S-H1 and 6S-H3, resulting in 2 daughter haplotypes (6S-H5 and 6S-H6) thatwere not seen in Chinese subjects. These 2 daughter haplotypes, which accountedfor 10.2% of all haplotypes, were represented only in the control group (Fisher P = 1.614 × 10⁻¹¹). This differencein frequency strongly suggested that haplotypes 6S-H5 and 6S-H6 were associatedwith lower risk of heroin dependence in the German population. In fact, the6-SNP region covering 10.8 kb even more narrowly defined the affected regionthan in the Chinese population.

The 8-SNP analysis showed a different pattern and predictive outcomeamong haplotypes in Germans as compared with Chinese (Figure 4). Similar to the 6-SNP analysis, we also observed 2 commonhaplotypes (8S-H1 and 8S-H3) whose recombination near SNP 6 resulted in 2daughter haplotypes (8S-H5 and 8S-H6) that predicted low risk of heroin dependencein German populations (Fisher P = 1.940 × 10⁻¹¹). The SNPs TaqIA and TaqIB were in the LD block but did not contribute additional informationhere, a result that differed from the Chinese population.

With the 10-SNP window used for analysis, 1 haplotype (10S-H2) was morefrequent in the cases than in the controls in Germans (0.100 in the controls,0.148 in the cases), and was modestly significant (P =.020; OR, 1.595; 95% CI, 1.089-2.338) (Figure4). We also found evidence of recombination by means of the 10S-locushaplotype set. However, only haplotype 10S-H5 was represented at significantlyhigher frequency in controls (0.050 vs 0.000; Fisher P =1.100 × 10⁻⁵).

Although 32 SNPs were initially selected for analysis, 2 SNPs from Chineseand 4 SNPs from Germans were removed from the test because of high genotypingfailure rate or deviation from Hardy-Weinberg equilibrium or for being monomorphicin a population. Thus, a total of 30 SNPs for the Chinese and 28 SNPs forthe Germans were used for these analyses. Within each population, a comparisonof allele and genotype frequencies between case and control groups for eachmarker failed to show any significant difference (data not shown). In addition,overall allele frequencies for SNP loci did not differ between case and controlgroups in either the Chinese (P = .744) or the Germans(P = .183), as expected. Between the populations,allele frequencies for all markers showed significant differences among Chinese,Germans, and African Americans (P<.001 for eachcomparison). By this approach, there was no evidence of population admixturebetween case and control groups in each of the study populations.

Using the Structure 2.0 program (available at: http://pritch.bsd.uchicago.edu; Jonathan Pritchard, PhD, The University of Chicago, Chicago, Ill)for detecting population admixture in either Chinese or Germans produced only1 cluster when applied to the combined population or to separate case andcontrol groups (K = 1, postprobability = 0.999 for each test). However, inAfrican Americans, there was evidence of population admixture (K = 2, postprobability= 0.999). These data indicated that the markers selected were able to detectpopulation structure in an admixed African American population and providedsupport that the Chinese and German populations used in the present studywere homogeneous.

In this study, we found that specific DRD2 haplotypeswere highly associated with heroin dependence in both Chinese and German populations.In addition, single-marker association with heroin dependence in Chinese wassignificant. Global tests of haplotype association were significant at thelevel of 3 SNP sets in both populations. A 25.8-kb region defined by 8 SNPswas implicated more strongly over any individual SNP analyzed in the Chinese,while a 10.8-kb region containing 6 SNPs supported a low-risk region for heroindependence in Germans. Moreover, our data showed that there was no evidenceof population admixture in either Chinese or Germans by testing additionalgenetic markers.

Previous studies using the known functional alleles have been contradictoryor nonsupportive of DRD2 association with alcoholismand other addictions. Therefore, it would be advantageous to use markers spanningthe entire DRD2 region and incorporate into the analysisany new in vitro functional variants available. In this study, we found thatthe −141delC allele at SNP2 was slightly more abundant in Chinese heroin addicts, yet genotype-basedcomparison between cases and controls did not share this difference. Also,the significance level for −141ins/delC wasless than for the TaqIB at SNP3, even in Chinese (P = .021 for −141ins/delC vs P = 3.8 × 10 ^{− 5} for TaqIB). The −141ins/delC is outside the implicated haplotype block, suggestingthat −141delC plays a minor role in heroindependence in Chinese. Our study also showed that TaqIB was strongly associated with heroin dependence in Chinese, which wasconsistent with previous studies.²⁹TaqIB is located within intron 1, but it may be in LD with an unknownfunctional SNP within the LD block. It should be noted that this SNP was alsoin strong LD with other SNPs within the 25.8-kb block. Furthermore, haplotypedata substantially increased the significance level of the association. Asdiscussed in the introduction, the TaqIA marker waspreviously implicated in alcoholism and substance dependence, but not in heroindependence. Although our data did not support a particular role for the Taq1A polymorphism in heroin dependence, this SNP did addinformation to the 8-SNP haplotype, increasing the strength of linkage inChinese but not in Germans. These results supported the idea that associationof haplotypes rather than any individual SNP points to an unknown effectivevariant or variants within the 25.8-kb region.

Because no single functional variant of DRD2 haspreviously been associated with heroin dependence, and because it is unknownwhether the known variants that alter function in vitro also alter in vivodopamine biology, LD analysis is an important step in detecting the actionof an effective variant or variants somewhere in DRD2. Forthe pairwise LD matrix using DRD2 gene SNPs, we determinedthat a strong LD block extended to 25.8 kb in the DRD2 gene,across both populations. Similar to a report by Kidd et al,³⁰ weobserved similar LD patterns across the DRD2 genein our study populations. Three SNPs (TaqIB, TaqID,and TaqIA) used in this study were the same as thoseused by Kidd and colleagues, but we applied these markers to much larger samplesizes in this study. The mean D′ for these 3 SNPs was 0.883 ±0.084 in our study compared with 1.000 ± 0.000, determined by Kiddand coworkers' study³⁰ for the Chinese Hanpopulation, while mean D′ was 0.903 ± 0.144 in our German populationcompared with 0.700 ± 0.111 in a Finnish population. In addition, theancestral haplotype defined by the Kidd et al study corresponded to the sameancestral haplotype found in both Chinese and Caucasian populations. The mostfrequent haplotype in Chinese, B1D2A1, also the ancestral haplotype, had afrequency of 0.37 compared with 0.36 from Kidd and coworkers' study.³⁰ Another haplotype B2D1A2 was the most abundant (0.450)for the German population in this study and had a frequency that comparedwith 0.417 for a Finnish population.³⁰ Moreinterestingly, we also found that the strength of LD in Chinese was greaterthan in the German population, where approximately 10% recombination has occurredin this genomic region in the German population. This accounted for the differentpattern of haplotype diversity between heroin addicts and controls in the2 populations. This interpretation may explain why different haplotypes wereassociated with heroin addiction in the 2 populations.

It is well known that allele-based LD analysis is a powerful tool foridentifying effective loci, assuming that a sufficiently large sample sizeis used and that stratification-produced results can be minimized or eliminated.^31- 32 In the German and Han Chinese case-controlpopulations we studied, individuals were recruited from the same geographicareas and represented relatively well-defined populations. Neither Germansnor Han Chinese are isolated or semi-isolated populations. However, our resultsfor detecting sample stratification indicated no evidence of subpopulation(admixture) in either case or control group in the 2 populations. Therefore,the association of DRD2 with heroin dependence wasunlikely false positive owing to stratified samples.

By mapping haplotype blocks in different populations, the evolutionaryhistory of genes can help us to put in order apparently disparate linkageand association data on disease-associated genes. Furthermore, differencesin block structure between populations may be identified so that populationswith smaller block sizes can be chosen for the purpose of homing in on effectiveSNPs or microsatellite polymorphisms.^33- 34 However,in both of the populations we studied, the implicated DRD2 haplotype block is large, so additional markers in the region increaseinformation content and improve the strength of association. Use of otheranalytical strategies (for example, sequence variant detection and establishmentof function) or linkage disequilibrium studies in other populations will thereforebe required to identify the effective variants in the near future.

Corresponding author and reprints: Ke Xu, MD, PhD, Laboratory ofNeurogenetics, National Institute on Alcohol Abuse and Alcoholism, 12420 ParklawnDr, Park Building, Room 451, Rockville, MD 20852 (e-mail: ke@mail.nih.gov).

Submitted for publication September 9, 2003; final revision receivedJanuary 5, 2004; accepted January 21, 2004.

This study was supported by the German Federal Ministry for Educationand Research (BMBF), Bonn, in part by grants 01EB9418/5 and 01EB9802/0, andby grant 01EB0133 within the framework of the Nordrhine-Westfalian InterdisciplinaryNetwork on Addiction Research (Dr Maier at the University of Bonn). This studyalso was partly funded by the Chinese National Nature and Science Foundation,Beijing (Dr Liu at the Sichuan University).

We thank D. J. Yuan, MD, and Z. H. Zhu, MD, of the Department of Psychiatry,Medical Center of Sichuan University, Sichuan, China, and J. C. Long, PhD,Department of Human Genetics, University of Michigan, Ann Arbor.