Behavioral Risk Exposure and Host Genetics of Susceptibility to HIV-1 Infection

Abstract

BackgroundSome individuals are readily infected with low human immunodeficiency virus type 1 (HIV-1) exposure, whereas others appear less susceptible, suggesting that host genetics plays a role in the viral entry pathway. The matched case-control study design with measured risk exposures provides an avenue for discovering genes involved in susceptibility to infection

MethodsWe conducted a nested case-control study of African Americans (266 HIV-1 seroconverter cases and 532 seronegative controls from the AIDS Link to Intravenous Experience cohort), to examine the association between 50 single-nucleotide polymorphisms (SNPs) in 9 candidate genes (CCR5, CCR2, RANTES, MIP1A, MCP2, IL10, IFNG, MCSF and IL2) and susceptibility to HIV-1 infection. To account for differential exposure propensities, risk behavior self-reported during semiannual visits was used to estimate a standardized cumulative risk exposure (SCRE). Individual SNPs were evaluated using conditional logistic-regression models, and the inferred haplotypes were assessed in the haplotype trend regression analyses after adjusting for age and SCRE

ResultsFour SNPs (CCR2−V64I, CCR5−2459, MIP1A+954,and IL2+3896) and specific haplotypes in the IL2 and CCR2/CCR5 regions were significantly associated with HIV-1 infection susceptibility in different genetic models

ConclusionsOur results suggest that genetic variants in associated host genes may play an important role in susceptibility to HIV-1 infection

There were 4.9 million new cases of HIV-1 infection worldwide during 2004 alone, and ∼39.4 million people are currently living with HIV [1]. The primary risk factors for HIV-1 infection are unprotected sexual intercourse, sharing of syringes, and being an infant born to an infected mother. In most cases, behavioral modification remains a foremost priority with regard to prevention of infection. The importance of biological and genetic differences between individuals in explaining differential susceptibility to HIV-1 infection is largely unknown. The course of HIV-1 susceptibility varies widely even among individuals with similar risk exposure levels [2–4]. For example, some sex workers and homosexual men have remained uninfected despite repeatedly engaging in unprotected sexual intercourse with HIV-1–infected partners or constantly engaging in high-risk behavior [4–6]

A possible role of host genetics in determining susceptibility to HIV-1 exposure is also suggested by the differential immunological responses that individuals have during the course of infection. Population-based genetic studies of HIV-1 infection susceptibility have been limited, since, in many cases, the seroconversion time is not known and the nongenetic risk exposure data are seldom available to evaluate disease infectivity related to host genetics. Most studies have focused on high-risk exposed uninfected or highly exposed but persistently seronegative individuals, such as discordant couples who have unprotected sex [2, 7, 8] and commercial sex workers [4, 9–12]. Immunologic and genetic studies of exposed yet uninfected individuals have helped to elucidate protective mechanisms for HIV-1 infection [13–15]

Previous studies have identified 14 genetic polymorphisms that show associations with disease progression [16]. However, apart from the HLA class I and II genes, only 4 other genes (CCR5, IL10, RANTES and the MCP1-MCP3-Eotaxin gene clusters) show associations with infection in European Americans [16–21]. The most significant polymorphism associated with HIV-1 infection is a 32-bp deletion in the coding region of the CCR5 gene [16, 22], where homozygotes (CCR5-Δ32/Δ32) show protection [14, 15, 23] and nearly resistance [4, 23, 24]. The frequency of CCR5-Δ32/Δ32 is essentially zero in African Americans, and no distinct genetic polymorphisms have been associated with HIV-1 infection in this ethnic group

We investigated 9 genes involved in the complex pathway of HIV-1 entry and replication, focusing on the CCR5 coreceptor, which is predominantly used during primary infection [25–27]. Specifically, we examined polymorphisms in CCR2, CCR5, RANTES, MIP1A, MCP2, IL10, IFNG, MCSF and IL2 The natural ligands of the CCR5 coreceptor—RANTES, macrophage inflammatory protein 1α (MIP1-α), and monocyte chemotactic protein 2 (MCP2)—inhibit HIV-1 entry and down-regulate CCR5 expression [28–32]. The cytokines interleukin (IL)–10 [33, 34] and IL-2 [35, 36] up-regulate the expression of CCR5 in vitro and induce other cytokines that are potentially involved in the CCR5 viral entry mechanism [37]. Macrophage colony-stimulating factor (MCSF) [38, 39] up-regulates CCR5 expression and also enhances HIV-1 replication. Interferon (IFN)–γ [40, 41] has been shown to enhance HIV-1 transcription but also to inhibit viral entry. Although not exhaustive, this list of genes includes those among the major cellular participants in primary HIV-1 infection

We examined polymorphisms in these host genes, using a nested case-control study within a cohort of African American injection drug users (IDUs), to investigate the genes’ potential roles in susceptibility to HIV-1 infection. We assessed both individual single-nucleotide polymorphism (SNP) and haplotype associations with HIV-1 seroconversion in the CCR2/CCR5 region and 7 other candidate gene regions, using a conditional logistic model framework accounting for differential risk exposure. To our knowledge, this is the first study to quantify HIV-1 risk exposure as a measurement and to explore its confounding effects in a controlled genetic analysis setting

Subjects, Materials, and Methods

SubjectsAIDS Link to Intravenous Experience (ALIVE) is a prospective cohort study of primarily African American IDUs in Baltimore. Participants have been studied for natural history and risk factors of HIV-1 infection and progression to AIDS. Details of the cohort, along with materials and methods for the genetic study design, have been described elsewhere [42–44]. Only African American participants were included in the present study, to avoid the confounding effects associated with racial differences and the lack of statistical power for other groups in addressing HIV infection within the ALIVE cohort. This study was approved by the Committee on Human Research at the Johns Hopkins Bloomberg School of Public Health and the National Cancer Institute

Cases and controlsA nested case-control study within the ALIVE cohort was used to investigate genetic variants of specific candidate genes associated with HIV-1 seroconversion, after adjusting for measured risk exposure. Cases were participants who underwent seroconversion during clinical follow-up. The date of HIV-1 seroconversion was estimated to be the midpoint between the last HIV-1–seronegative test and the first documented HIV-1–seropositive test. Two controls per case (cases, n=266; controls, n=532) were randomly chosen from among HIV-1–seronegative individuals and were matched according to the duration of follow-up. Incidence density sampling [45, 46] was used to select the controls from the pool of HIV-1–seronegative IDUs in active follow-up within a 6-month period (±3-month window) of the seroconversion date for cases. Under this scenario, 5 subjects were initially controls but later became cases, resulting in a total sample size of 793

SNP selection and genotypingFifty SNPs in the candidate genes were identified and assessed for association with HIV-1 susceptibility (table A1 in the appendix, which is not available in the print editiontable A1 in the appendix, which is available only in the electronic edition of the Journal). SNPs were genotyped by Taqman assay [47], using an Applied Biosystems 7900 genetic analyzer. CCR5-Δ32 was the only non-SNP polymorphism, and the variant was determined by an agarose gel sizing assay [48]

Risk exposure assessmentThe matched follow-up time between cases and controls partially adjusts for unknown confounding factors; however, in addition, we quantified the cumulative risk exposure on the basis of self-reported risk behavior and subsequently adjusted for its effect. Previously, Nelson et al. [43] defined independent risk behavior associated with HIV-1 seroconversion within the ALIVE cohort [43]. On the basis of these documented risk factors, a measure of standardized cumulative risk exposure (SCRE) for each individual was estimated. The log of the estimated relative incidence values for each risk factor in the model were used as weights in the summary risk measure for all visits up to the seroconversion point for each individual, accounting for the person-time exposure. CRE_i is the cumulative risk exposure for person i and is calculated as follows:

where β_k is the log relative incidence associated with risk variable k in the seroconversion model, R_kj is the indicator for risk behavior k at visit j (1 = yes; 0 = no), and t_j is the number of days of follow-up at visit j

Although the CRE values within the matched trios we examined were comparable, a measure that has a uniform interpretation among all individuals regardless of the differential follow-up time—an SCRE—was deemed appropriate for the analyses and was calculated as follows:

where CRE_i is the cumulative risk exposure (eq. [1]) and TFT_i is the total follow-up time for individual i First, the proportions of cases were explored in different quantile groups of the SCRE distribution. Since the values of SCRE were based on risk factors for HIV-1 seroconversion, the proportion of cases in the risk groups was expected to increase with incremental values of SCRE. High, medium, and low risk levels were categorized on the basis of the proportion of seroconverters in the SCRE quantile groups

Statistical analysesCases and controls were first compared in terms of demographic and risk behavior data. χ² tests were used to compare categorical variables (e.g., sex), whereas the Mann-Whitney U test was used to compare continuous variables (e.g., SCRE values and age). A relatively stringent threshold of r²>0.85 was used to identify redundant SNPs. Analyses were conducted using the statistical packages SAS (version 9.0; SAS Institute), STATA (version 7.0; STATA), and S-Plus (version 6.0; MathSoft)

Conformity of the genotype proportions to Hardy-Weinberg equilibrium (HWE) was examined for each polymorphism in cases and controls. Pairwise linkage disequilibrium (LD) between SNPs at each gene was measured and viewed in the Graphical Overview of Linkage Disequilibrium program [49], using Lewontin’s D^′ [50] and the square of the correlation coefficient, r² [51]. Initial exploratory analyses for differences in allele frequencies and genotype distribution between the cases and controls were performed using Fisher’s exact test. Analyses comparing genotypes at each individual SNP between cases and controls were performed using conditional logistic regression. In our analyses, P values ⩽.05 were considered to be statistically significant, and P values between .05 and .1 were deemed to indicate trends. Indicator variables for all genotypes were created by using the most common genotype as the reference category. In the case of the dominant model, heterozygotes and less common homozygotes were combined. Age and the SCRE risk levels were modeled as covariates

The expectation/maximization (EM) algorithm [52], modeled after SNPHAP software (available at: http://www-gene.cimr.cam.ac.uk/clayton/software/) using an efficient progressive-insertion algorithm, was applied to estimate haplotype frequencies in a combined pool of cases and controls. On the basis of all possible haplotypes given the genotype, a haplotype matrix of posterior probabilities for each individual was estimated, and these probabilities were used in a haplotype trend regression (HTR) model as independent variables [53, 54]. The HTR model has been expanded to the conditional logistic-regression framework [46], allowing adjustment for matched case-control status, age, and risk levels. All rare haplotypes with frequencies <1% were collapsed into 1 haplotype group, and the most frequent haplotypes were considered the references in the analyses. Global tests were conducted to assess the significance for the whole effect of haplotypes and individual tests for each haplotype. Further, stepwise regression was used to obtain the most parsimonious model. Haplotype-pair regression analysis was further performed for the significant haplotypes. When haplotypes were significantly associated with HIV-1 infection, the odds ratios (ORs) and 95% confidence intervals (CIs) were calculated. The haplotype analyses were implemented using the S-PLUS version (MathSoft) of the HaploStats statistical package (available at: http://mayoresearch.mayo.edu/mayo/research/biostat/schaid.cfm)

Results

The case and control groups did not differ by sex, but the cases were slightly younger than the controls (median age, 36.1 vs. 38.5 years; P<.0001) (table 1). As expected, the median SCRE value was higher among cases than controls (0.62 vs. 0.49; P < .0001). SCRE values ranged from 0 to 1.86 (figure 1), and the distribution of values for all subjects was stratified into quintiles. The proportions of HIV-1 seroconverters (cases) in each quantile were calculated and are informative on a relative basis. The first and second quantiles (24% and 26% cases, respectively) were categorized as low risk, the third and fourth quantiles (30% and 34% cases, respectively) were categorized as medium risk, and the fifth quantile (49% cases) was categorized as high risk

Table 1

Demographic characteristics of the study participants

Figure 1

Risk level stratification based on distribution of standardized cumulative risk exposure (SCRE) values. Nelson et al. [43] showed that injection cocaine use (relative risk [RR], 1.61), homosexual activity (RR, 7.03), visiting “shooting galleries” (RR, 1.47), having any sexually transmitted diseases (RR, 1.81), and an interaction between drug injection and heterosexual behaviors (heterosexual sex and no drug injection: RR, 1.59; heterosexual sex and drug injection: RR, 1.77; heterosexual sex and drug injection more than once daily: RR, 3.14; no sex and drug injection less than once daily: RR, 4.28; no sex and drug injection more than once daily: RR, 5.06) were independently associated with seroconversion in a multivariate analysis. These estimates and measures were used to calculate SCRE values according to equations (1) and (2), from semiannual visits that queried the risk behaviors of all participants. The length of follow-up for exposure at each visit was determined as the no. of days between each visit and the previous one (maximum, 365 days). The first visit was considered to represent 180 days. The frequencies of SCRE values among subjects (n=798) is labeled on the left vertical axis. The SCRE distribution was divided into quintiles, and the proportion of cases (seroconverters) in each quantile was calculated (right vertical axis) Quantiles I and II have 24% and 26% cases, respectively (low risk), quantiles III and IV have 30% and 34% cases, respectively (medium risk), and quantile V has 49% cases (high risk)

A graphical representation of pairwise LD for 50 SNPs in 8 gene regions among controls is shown in figure 2. The distribution of alleles and genotypes in cases and controls at 41 SNPs (9 were excluded because of low frequency or high LD) was explored (table 2). Six (CCR2−V64I, CCR5−2459, MIP1A+954, IL2+161, IL2+3896 and IFNG+2112) trended toward statistical significance (P<.10) in both the allelic and genotypic frequency analyses. Matched case-control analysis using conditional logistic regression showed the effects of particular genotypes, both with and without (data not shown) adjustment for age and risk measured as SCRE; 4 of these 6 SNPs showed modest effects (figure 3). For the CCR2−V64I SNP, the adjusted OR (OR_adj) for heterozygotes was 0.69 (95% CI, 0.48–0.98), and that for A/A homozygotes was 0.39 (95% CI, 0.10–1.50), relative to the most common homozygotes. The rare minor-allele homozygote at CCR2−V64I was not statistically significant, but its effect was similar to that of the heterozygote comparison, suggesting a dominant model (combining G/A and A/A: OR_adj, 0.66 [95% CI, 0.46–0.94]). Likewise, at CCR5−2459, the OR_adj values for heterozygotes and minor-allele homozygotes, compared with G/G homozygotes, were 0.71 (95% CI, 0.51–0.97) and 0.62 (95% CI, 0.38–1.00), respectively, whereas, when combined, a 0.69 (95% CI, 0.50–0.95) infection-protective effect was seen

Figure 2

Linkage disequilibrium (LD) between single-nucleotide polymorphisms (SNPs) examined in 8 candidate gene regions. The physical positions of introns, coding regions (CDS), and untranslated regions (UTRs) are illustrated for each gene region. The SNPs examined are shown in the same order as in table 2 and the disequilibrium plots. D′ (top left triangle) and r²(bottom right triangle) in controls are shown for very little (blue) to modest (green) to very strong (red) LD, ranging from 0 to 1

Table 2

Single-locus allele and genotype frequency distributions between cases and controls

Figure 3

Genotypic frequencies and HIV infection adjusted odds ratios (OR_adj) at single-nucleotide polymorphisms (SNPs) CCR2−V64I (A), MIP1A+954 (B), CCR5−2459 (C) and IL2+3896 (D) Adjustments in the conditional logistic regression models were made for risk levels (also see figure 2) and age. Common homozygotes at each SNP were used as the reference group (frequencies not shown)

Other candidate genes within the CCR5 infection pathway also showed some signals of association with HIV infection. Effects seen for the MIP1A+954 T allele suggest a recessive model (OR_adj for T/A heterozygotes, 1.10 [95% CI, 0.79–1.56]; OR_adj for T/T homozygotes, 2.25 [95% CI, 1.03–12.29]) (figure 3). The support for the T/T homozygote effect is drawn from 11 individuals, and similar results are seen at the 2 highly linked loci (r²⩾0.85; data not shown). None of the SNPs in RANTES or MCP2 showed any significant association with susceptibility to HIV-1 infection. Since the genes encoding the CCR5 coreceptor ligands RANTES, MIP1-α, and MCP2 are localized to a 1.8-Mb region on 17q11-12, haplotype analysis was considered for all of the SNPs in the 3 ligand genes, but no significant LD was seen in this extended region (analysis not shown). For IL2+3896 the genotypes A/− and A/A did not separately indicate a significant association, compared with the common homozygous deletion genotype, but they showed a significant protective effect together (unadjusted OR, 0.74 [95% CI, 0.55–0.99]; OR_adj, 0.71 [95% CI, 0.52–0.97]) (figure 3)

Analysis of haplotypes (table 3) indicated that haplotypes in the CCR2/CCR5 (P=.03) and IL2 (P=.02) regions were significantly associated with HIV-1 infection. In the HTR analysis, when an additive approach was used, the CAAGAAC+ haplotype in the CCR2/CCR5 region showed significant association with HIV-1 infection in the overall model (OR, 0.71 [95% CI, 0.57–0.85]), as well as in a more parsimonious model comparing it with all other haplotypes (OR, 0.77 [95% CI, 0.61–0.98]). In the analyses of the IL2 haplotypes, TGCAA (OR, 0.79 [95% CI, 0.69–0.89]) and TGT-A (OR, 0.53 [95% CI, 0.25–0.81]) showed significant associations in the overall model. A more parsimonious model contrasting these 2 haplotypes with all others showed similar associations (ORs, 0.78 [95% CI, 0.71–0.86] and 0.52 [95% CI, 0.39–0.68], respectively). The results of the HTR analysis were consistent with the single-SNP results, in which the SNPs in LD with the significant haplotypes were also associated with HIV-1 infection (CCR5−2459A in LD with the CAAGAAC+ haplotype and IL2+161A in LD with the TGCAA haplotype)

Table 3

Global and individual haplotype tests, based on regression analysis, and the frequencies of haplotypes, estimated using the expectation/maximization algorithm

Discussion

Our study suggests that some host genes may play a role in susceptibility to HIV-1 infection. The independent associations of specific SNPs and haplotypes were modest but statistically significant after age and differential risk exposure were adjusted for (figure 3 and table 3). We observed associations between susceptibility to HIV-1 infection and variants in the gene regions of CCR2 and CCR5 that encode the coreceptors; MIP1-α, a natural ligand of the coreceptor CCR5; and IL-2, a cytokine involved in immune regulation that has widely been used in therapy and vaccine trials. In contrast to previous studies that focused on only highly exposed uninfected individuals [14, 60–62], our study included a nested case-control design in which we focused on seroconverters and adjusted for measured risk behaviors in genetic analyses of HIV-1 susceptibility

Exposure levels were measured in our study as SCRE, on the basis of self-reported risk behavior. Humans significantly influence their level of exposure to infectious agents like HIV-1 through their behavior, and such behavior is generally difficult to quantify or assess. The reliability and validity of behavior self-reported by IDUs in numerous publications [63–65], coupled with relatively low rates of abstinence from risk behavior reported at ALIVE clinical visits, suggests that these self-reports are robust indicators of true behavior. The SCRE we estimated from ALIVE self-reports of both injection and sexual behavior captured varying IDU behavior over several years. In contrast, most cross-sectional studies of IDUs [66, 67] represent a less reliable snapshot of varying risk behavior. Given the high seroprevalence of HIV-1 among IDUs in Baltimore, our study subjects are at greater risk for HIV-1 infection than are other populations. The SCRE we developed quantifies the level of risk for each individual and estimates the differential propensity of risk exposure for genetic analyses

Correcting for multiple comparisons, using methods such as Bonferroni or false discovery rate, did not result in significant associations with any of the SNPs or haplotypes. Given the biological plausibility of the importance of CCR5 pathway genes in primary HIV infection, some exploration of these results with P<.05 seems justified to guide later research on African Americans. We did not observe any CCR5-Δ32/Δ32 homozygotes in our study sample, which is not surprising, given the rarity of the Δ32 allele in African Americans. We found that individuals with the 64I allele in CCR2 a gene encoding a β chemokine receptor, were less susceptible (dominant allele A model) to HIV-1 infection (figure 3). The G→A polymorphism in the coding region of the HIV-1 coreceptor CCR2 (−64I A allele frequency, 0.098 in white individuals and 0.151 in African Americans) causes a single amino acid change from Val to Ile in the first transmembrane domain but does not appear to alter the mechanism of the coreceptor [60]. This A variant has been found to be associated with delayed HIV-1 disease progression in several studies [60, 68–70], but no effect on HIV-1 transmission has yet been reported. In vitro, an isoform of CCR2 (CCR2A) binds to CCR5 in the cytoplasm and sequesters CCR5 receptor before it reaches the cell surface [71], providing a possible cellular mechanism underlying delayed progression to AIDS in individuals with the 64I polymorphisms. The consistency of CCR2−64I decreasing surface availability of CCR5 for viral entry supports our association in which individuals with these polymorphisms are less susceptible to HIV-1 infection. We also observed that another variant in the promoter region of CCR5, −2459A appeared to be protective in a dominant model. A protective effect of the −2459A allele against HIV infection, however, seems unlikely, because it has previously been shown to accelerate progression to AIDS [57, 72]. The SNP showing association with infection in IL2 is in a noncoding region, and the one in MIP1A is a synonymous SNP. Very little is known regarding the relationship between these genes and HIV infection, which requires further examination—for example, the association between structural variation of MIP1A (CCL3L1) in the number of copies and HIV-1 susceptibility should be investigated [73]

We developed a regression-based approach to study the association between inferred haplotype probabilities and susceptibility to HIV-1 infection in a nested case-control analysis. The primary assumptions of the method are (1) that sampling of cases and controls is random and (2) that HWE conditions are met when posterior probabilities of all possible haplotype pairs are estimated [53, 54]. Unlike the methods used in previous studies with inferred haplotypes based on the likelihood-ratio test [74, 75], our method has several advantages. First, haplotype-specific association can be easily computed, along with evaluation of individual haplotypes; second, nongenetic covariates can be adjusted; and third, computation time is much shorter than with the likelihood-based approach, especially when heterozygotes are frequent

Applying the new analysis method to the 8 gene regions, we found associations between IL2 and CCR2/CCR5 haplotypes and HIV-1 infection. Beyond studies implicating CCR5-Δ32 homozygosity, there have been no reports indicating associations between haplotypes in the CCR2/CCR5 region and susceptibility to infection. Some studies have shown that, among European Americans, +.P1.+ haplotype homozygotes (table 3, footnote “a”) have accelerated AIDS progression [57, 58]. Among African Americans, a weak association with AIDS progression (using a set of samples that included 164 in our analyses) was reported in one study [72], and no association was reported in another [58]. One might expect concordance between AIDS progression and HIV susceptibility genes, but we observed a protective effect for HIV-1 infection (OR, 0.71 [95% CI, 0.57–0.85]) with the AIDS progression–accelerating +.P1.+-containing haplotype, CAAGAAC+ (table 3). Haplotypic analysis also indicates that variants in the IL2 gene could be involved in protection against infection. Single-SNP analyses indicated an association with IL2+3896A that is also seen with the TGCAA haplotype it defines; however, it missed the haplotype TGT-A, which has a relatively low frequency. The functions of these 2 haplotypes are not known, but the importance of IL2 is clear. IL2 has been successfully used in therapy for HIV/AIDS and currently has been extended to phase 3 trials [76, 77]. A deficiency in IL-2 production is one of the first immunologic defects to be described in HIV-1–infected individuals. The consistency of these modest associations with both the allelic and haplotype analysis of the CCR2/CCR5 and IL2 gene regions awaits replication in other cohorts and studies

Discovering the role of host genetics in susceptibility to HIV-1 infection can provide important insights into the growing pandemic. Comprehensive scans [78–80] of genes involved in innate immunity and, indeed, the entire genome are possible with the study design and samples we have developed. Human genetic variation is known to play a role in susceptibility to many infectious diseases, most notably malaria, schistosomiasis, and tuberculosis [81–83]. Most previous studies have focused on twin, adoptee, and family designs, to control for differential exposure and the environment. However, studies of such related individuals are difficult, since exposure to infectious agents such as HIV-1 is frequently not similar among family members. To address these limitations, we developed a promising case-control study that explicitly takes into account measured exposure levels, to study the association of host genetics with HIV-1 infection. On examination of polymorphic markers in 9 CCR5 pathway genes, SNPs in 4 of these genes (CCR2, CCR5, MIP1A and IL2) showed moderate associations with HIV-1 infection in single-SNP and haplotype analyses. One of these SNPs (CCR2−V64I) has a plausible biological role, limiting the availability of HIV-1 coreceptor (CCR5) in the host cells [84, 85]. Like studies of the genetics of progression to AIDS [16], our study suggests a complex association of multiple genes with small contributory effects to HIV-1 infection

Acknowledgments

First and foremost, we thank the AIDS Link to Intravenous Experience cohort participants and staff. We thank Cheryl Winkler, Michael Dean, and Terri H. Beaty, for helpful insights in developing the ideas for this article, and David Thomas, for buffy coat samples. We are also grateful to Kai Zhao, Ann Truelove, Bailey Kessing, Michael Malasky, Mary Thompson, Mahboobeh Safaeian, and Joseph Bareta for their assistance. The content of this publication does not necessarily reflect the views or policies of the Department of Health and Human Services, nor does mention of trade names, commercial products, or organizations imply endorsement by the US government

Presented in part: 53rd Annual Meeting of the American Society of Human Genetics (poster 1259)

Potential conflicts of interest: none reported

Financial support: This research was supported by the Intramural Research Program of the National Institutes of Health (NIH), National Cancer Institute (NCI), Center for Cancer Research. This publication has been funded, in part, with federal funds from the NCI, NIH (contract N01-CO-12400), and National Institute on Drug Abuse (grants DA09225, DA8009, and DA12568)

(See the editorial commentary by Telenti and Ioannidis, on pages 4–6.)

Present affiliation: Department of Epidemiology, School of Public Health, University of Alabama at Birmingham, Birmingham