MAJOR ARTICLE

Immunogenetics of CD4 Lymphocyte Count Recovery during Antiretroviral Therapy: An AIDS Clinical Trials Group Study Downloaded from https://academic.oup.com/jid/article/194/8/1098/870426 by guest on 28 September 2021 David W. Haas,1 Daniel E. Geraghty,2 Janet Andersen,3,4 Jessica Mar,3,4 Alison A. Motsinger,1 Richard T. D’Aquila,1 Derya Unutmaz,1 Constance A. Benson,5 Marylyn D. Ritchie,1 and Alan Landay6 1Vanderbilt University, Nashville, Tennessee; 2Fred Hutchinson Cancer Research Center, Seattle, Washington; 3Harvard School of Public Health and 4Statistical Data and Analysis Center, Boston, Massachusetts; 5University of California, San Diego, San Diego; 6Rush University Medical Center, Chicago, Illinois

During antiretroviral therapy, CD4 lymphocyte count increases are modest in some patients despite virologic control. We explored whether polymorphisms in genes important for T cell expansion, survival, and apoptosis are associated with the magnitude of CD4 lymphocyte count recovery during antiretroviral therapy. We studied treatment-naive individuals who achieved sustained control of plasma viremia (!400 HIV-1 RNA copies/mL) for at least 48 weeks after initiation of antiretroviral therapy and compared genotypes among individuals who had an increase of either !200 or у200 CD4 cells/mm3 from baseline. A total of 137 single- poly- morphisms across 17 genes were characterized in 873 study participants. In multivariate analyses that controlled for clinical variables, polymorphisms in genes encoding tumor necrosis factor (TNF)–related apoptosis-in- ducing ligand (TRAIL), TNF-a, Bcl-2–interacting molecule (Bim), interleukin (IL)–15, and IL-15 receptor a chain (IL-15Ra) were associated with the magnitude of the increase in CD4 lymphocyte count, as were haplotypes in genes encoding interferon-a, IL-2, and IL-15Ra (P ! .05 , for each). Multifactor dimensionality reduction identifie a gene-gene interaction between IL-2/IL-15 receptor common b chain and IL-2/IL-7/IL- 15 receptor common g chain. Immune recovery during antiretroviral therapy is a complex phenotype that is influence by multiple genetic variants. Future studies should validate these tentative associations and defin underlying mechanisms.

HIV-1 infection causes decreased production and in- several cytokines (interleukin [IL]–2, IL-7, and IL-15) creased destruction of CD4 lymphocytes [1–3]. Anti- and their receptors (IL-2R, IL-7R, and IL-15R) [10, 11]. retroviral therapy typically brings about substantial im- Both IL-7 and IL-15 are required for naive T cell sur- mune reconstitution [4, 5], but there is considerable vival [12], and both promote memory T cell prolifer- interindividual variability in peripheral-blood CD4 lym- ation [13, 14]. In addition, both IL-7 and IL-15 may phocyte count recovery during HIV therapy. Predictors enhance memory T cell survival by increasing the ex- of lesser CD4 lymphocyte count increases include in- pression of antiapoptotic proteins [15, 16]. Growth of creasing age [6] and lower pretreatment plasma HIV- 1 RNA concentrations [6–9]. Proliferation and survival of T cells are regulated by Presented in part: 3rd International AIDS Society Meeting, Rio de Janeiro, 24– 27 July 2005 (abstract MoOa0303); 13th Conference on Retroviruses and Op- portunistic Infections, Denver, 22–25 February 2006 (abstract 443). Potential conflicts of interest: D.W.H. has received research grants from Bristol- Myers Squibb (BMS), Boehringer Ingelheim (BI), Schering Plough, Bavarian Nordic, Received 24 March 2006; accepted 16 May 2006; electronically published 6 and Gilead Sciences. J.A. is a consultant for Tibotec. R.T.D. is a consultant for September 2006. and/or has received honoraria or research grant support from BMS, Gilead Reprints or correspondence: Dr. David W. Haas, Center for Human Sciences, GlaxoSmithKline (GSK), Abbott Laboratories, BI, and Tibotec. M.D.R. is Research, Vanderbilt University School of Medicine, 345 24th Ave. N., Suite 105, a consultant for BI. C.A.B. is a consultant for and/or has received research grant Nashville, TN 37203 ([email protected]). support from Gilead Sciences, Merck, BI, GSK, Tibotec, and Johnson and Johnson The Journal of Infectious Diseases 2006;194:1098–1107 Research. A.L. has received a research grant from Abbott Laboratories. A.A.M., 2006 by the Infectious Diseases Society of America. All rights reserved. J.M., D.E.G., and D.U. report no potential conflicts of interest. 0022-1899/2006/19408-0010$15.00 Financial-support information is given in the Acknowledgments.

1098 • JID 2006:194 (15 October) • Haas et al. CD4 lymphocytes in vitro requires IL-2, and, after T cell ac- ation, survival, and apoptosis. This is the firs study to system- tivation, IL-2R expression is transiently up-regulated [17], al- atically address the immunogenetics of CD4 lymphocyte count though the importance of IL-2 for T cell expansion in vivo is recovery during antiretroviral therapy. less certain [18]. The effects of IL-2, IL-7, and IL-15 may be regulated through the differential expression of their receptors. PARTICIPANTS, MATERIALS, AND METHODS All 3 receptors share an identical g chain (IL-2Rg). In addition, IL-2R and IL-15R share a common b chain (IL-2Rb) [19], Study participants. This retrospective analysis included par- ticipants from AIDS Clinical Trails Group (ACTG) protocol whereas each receptor has a distinct a chain (IL-2Ra, IL-7Ra, A5001, a prospective, longitudinal cohort study that coenrolls and IL-15Ra). Tumor necrosis factor (TNF)–a also affects T participants from randomized ACTG antiretroviral treatment cell proliferation, whereas interferon (IFN)–a and IFN-b in- trials in the United States and Puerto Rico to evaluate long- fluenc activated T cell survival. term responses to antiretroviral therapies. The participants in-

Normal homeostasis requires a balance between cellular pro- Downloaded from https://academic.oup.com/jid/article/194/8/1098/870426 by guest on 28 September 2021 cluded in the present analyses (1) were antiretroviral naive and liferation and programmed death (apoptosis). During the ac- had a CD4 lymphocyte count of !800 cells/mm3 at baseline quired immune response to foreign , T cell expansion and (2) achieved a plasma viremia of !400 HIV-1 RNA copies/ is followed by the apoptosis of activated antigen-specifi cells, mL during therapy and maintained this level of suppression at with relatively few cells surviving to provide memory. The ex- every determination (generally at 8–16-week intervals) for at trinsic apoptosis pathway is initiated by the binding of several least 48 weeks. As a convenience sample, we included every TNF family ligands to death receptors. Specificall , Fas ligand eligible A5001 participant with sustained control of plasma binds to the death receptor Fas [20], TNF-related apoptosis- viremia for 48 weeks as of June 2004 who contributed speci- inducing ligand (TRAIL) binds to 2 death-inducing receptors mens to the ACTG Human DNA Repository (Center for Hu- (DR4 and DR5) [21], and TNF-a binds to TNF receptor man Genetics Research, Vanderbilt University) under protocol (TNFR)–I and TNFR-II. Signaling through TNFR-I can both A5128 [30]. We compared participants with a CD4 lymphocyte abrogate and potentiate apoptosis [22]. Death receptor binding count increase of !200 cells/mm3 from baseline to 48 weeks of triggers activation of the cysteine protease, caspase 8, and cul- virologic suppression with those with an increase of у200 cells/ minates in cell death [23, 24]. mm3. Although this value was chosen a priori, it happened to The intrinsic pathway to apoptosis is controlled by Bcl-2 approximate the median CD4 lymphocyte count increase. Ad- family proteins [25, 26]. In vivo, many activated T cells die of ditional post hoc analyses excluded individuals with a CD4 lym- the effects of the proapoptotic Bcl-2 family member, Bcl-2– 3 phocyte count of 1350 cells/mm at baseline and with a CD4 interacting molecule (Bim) [27]. Normally, Bcl-2 protects T lymphocyte count increase between 100 and 200 cells/mm3. cells from the apoptotic actions of Bim [27]. Cytokine signaling The present study complied with the Helsinki Declaration and through g chain receptors (IL-2, IL-7, and IL-15) increases Bcl- was approved by institutional review boards at each site, and 2 expression, which protects against Bim. Conversely, cytokine participants provided written, informed consent. withdrawal allows Bim to trigger apoptosis [27]. The death of Identificatio of genetic variants. Primer pairs amplifie naive T cells may also involve Bim [28]. intervals from 800 bp to 1.2 kbp; the human sequence There is extensive cross talk between proliferation and ap- consensus was used as template, and software that incorporated optosis pathways. Proliferation after T cell receptor engagement fundamentals of Primer3 was used [31]. Repetitive regions were is enhanced by TRAIL, IL-2 primes activated T cells for apo- avoided by use of RepeatMasker [32]. Polynucleotide and mul- ptosis, and both TNF-a and IFN-b can stimulate apoptosis. tisatellite sequences were avoided or directed toward the center Furthermore, IFN-a and IFN-b produced by HIV-1–activat- of amplicons. SNPs were chosen on the basis of 3 criteria: (1) ed plasmacytoid dendritic cells can regulate the expression of known nonsynonymous coding variants, (2) variants in or near TRAIL by CD4 lymphocytes [29]. In addition, TNF-a can stim- promoter sequences, and (3) variants of unknown function that ulate HIV-1 replication, whereas IFN-a and IFN-b inhibit HIV- marked genes, including haplotype-tagging polymorphisms [33, 1 replication. 34]. A balance among these criteria was achieved within the The influenc that human genetics has on CD4 lymphocyte funding available for this project. Primer pairs were firs tested count recovery during antiretroviral therapy is not known. We on from 4 cell lines, to assure the ability to generate hypothesized that increased expression or activity of genes that reliable amplicons and derivative sequence data. Primer se- are important for T cell proliferation will increase the size of quences are presented in table 1. the peripheral T cell pool, as will decreased signaling through Our resequencing strategy did not require preexisting knowl- extrinsic or intrinsic apopototic pathways. To explore this edge of polymorphisms. However, because a subset of exons possibility, we analyzed 137 single-nucleotide polymorphisms or promoter regions for each targeted gene was sampled, regions (SNPs) across 17 genes that are important for T cell prolifer- with SNPs in the dbSNP database (available at: http://www.ncbi

Immunogenetics of CD4 Recovery • JID 2006:194 (15 October) • 1099 Table 1. Primer sequences. Haplotypes reconstructed with a posterior probability of !0.05 were discarded, as were unique haplotype pairs with a frequency The table is available in its entirety in the online of !5%. For each gene, score tests were run to test for asso- edition of the Journal of Infectious Diseases. ciations between haplotype variants and CD4 lymphocyte re- sponse category. The haplo.score function from the haplo.stats package was used to output permutation P values. .nlm.nih.gov/SNP/) were chosen, and haplotype-tagged poly- Multifactor dimensionality reduction (MDR). Gene-gene morphisms were given firs priority. Resequencing was done interactions were identifie using MDR [41–43]. First, data using Beckman Fx robots for polymerase chain reaction (PCR) were divided into a training set and an independent testing set, assembly, sequence reaction assembly was done using Genetics for cross-validation. Five-fold cross-validation was used, with Management System software [35], and ABI 3730xl sequencers four-fifth of the data used for training and one-fift used for were used for analysis [36]. The PCR failure rate was 1%, whereas

testing. A set of n genetic factors was then selected. These factors Downloaded from https://academic.oup.com/jid/article/194/8/1098/870426 by guest on 28 September 2021 ! the genotyping error rate was 0.2%. Cycle sequencing reactions and their multiple classes were divided in n-dimensional space. were performed in 96-well thermocyclers (PerkinElmer 9700 or Ratios of cases to controls were then calculated within each 2700), by use of kits and protocols developed by PE-Applied multifactor class. Each multifactor cell class was labeled as high Biosystems (BigDye; version 3.1). Reaction products were filte ed risk if the ratio was 11.0 and as low risk if it was !1.0, thus through sephadex columns, and trace data were collected on an reducing the n-dimensional space to 1 dimension with 2 levels. ABI 3700 or ABI 3730xl sequencer. The collection of these multifactor classes comprised the MDR Homozygous and heterozygous traces from each amplicon model. For each possible model size (1 locus, 2 loci, …, n loci), were firs combined and analyzed for sequence content by use the single MDR model that misclassifie the fewest participants of Phred and Phrap [37], and the data were viewed by use of was chosen. Prediction error was calculated using 5-fold cross- Consed [38]. A program for heterozygous sequence analysis, validation. The result was a set of models, one for each model Heterozygous Trace Resolution, was used to perform most anal- size considered. The fina model was chosen that minimized ysis of sequence-based polymorphism discovery and typing for prediction error while maximizing cross-validation consistency. heterozygous DNAs [39]. This software is similar to PolyPhred The statistical significanc of the final best model was deter- [40] and Surveyor (Soft Genetics) and is adapted for mined through permutation testing, which involved the crea- high throughput. We tested the program on a subset of 26 tion of 1000 permuted data sets by randomization of the dis- amplicons encompassing 389 polymorphic positions. Hetero- ease-status labels. The entire procedure was repeated for each, zygous trace resolution reduced genotype errors and eliminated generating a distribution of 1000 prediction errors and cross- false-negative calls. validation consistencies that could be expected by chance alone. Statistical analysis. The primary outcome was absolute The significanc of the fina model was determined by com- CD4 lymphocyte count increase from study entry to 48 weeks paring the prediction error and cross-validation consistency of of sustained control of plasma viremia. Contingency tables were the fina model with the distribution. A P value was extracted constructed to measure associations between clinical variables for the model on the basis of its theoretical location in the and CD4 lymphocyte response category. Univariate logistic re- permutation distribution. gressions for associations between individual polymorphisms and CD4 lymphocyte response category, adjusted for race/eth- RESULTS nicity, were performed with the Benjamini-Hochberg pro- cedure. Logistic regression models were applied to the clinical Demographics. A total of 873 participants from ACTG pro- data to explore potential associations with CD4 lymphocyte tocol A5001 were included in the present analyses. From base- response category, with step-up and step-down logistic regres- line to 48 weeks of sustained control of plasma viremia after sion used to converge to a fina model. Each stepwise model the initiation of antiretroviral therapy, increases in CD4 lym- was run without adjustment for race/ethnicity, forcing this var- phocyte count were у200 cells/mm3 in 463 (53%) of the par- iable to be retained in the fina model. Two-sided Jonkheere- ticipants and were !200 cells/mm3 in 410 (47%). Clinical Terpstra tests were used to assess CD4 lymphocyte count in- characteristics at baseline for all study participants and by sub- crease as a continuous variable. Haplotype reconstruction was sequent CD4 lymphocyte response category are presented in performed using the haplo.em function in the haplo.stats pack- table 2. By multivariate logistic regression analysis, baseline age (Sinnwell and Schaid, authors; R-unix 1.2.2; Mayo Foun- characteristics that were significantl associated with a subse- dation for Medical Education and Research, Rochester, MN; quent increase in CD4 lymphocyte count of у200 cells/mm3 available at: http://mayoresearch.mayo.edu/mayo/research/ were plasma HIV-1 RNA concentration of 1100,000 copies/mL biostat/schaid.cfm). Consensus haplotypes with the highest (P p .0002 ), male sex (P p .0036 ), CD4 lymphocyte count of probability were chosen for inclusion in the present analyses. 1200 cells/mm3 (P p .0049 ), positive result for hepatitis B sur-

1100 • JID 2006:194 (15 October) • Haas et al. Table 2. Baseline characteristics of the study participants.

CD4 lymphocyte count increase All participants у200 cells/mm3 !200 cells/mm3 Characteristic, parameter (n p 873) (n p 463) (n p 410) Age, mean SD, years 38.1 9.0 37.5 9.1 38.8 8.8 Sex Male 721 (83) 368 (79) 353 (86) Female 152 (17) 95 (21) 57 (14) Race/ethnicity White 455 (52) 236 (51) 219 (53) Black 225 (26) 122 (26) 103 (25)

Hispanic 175 (20) 96 (21) 79 (19) Downloaded from https://academic.oup.com/jid/article/194/8/1098/870426 by guest on 28 September 2021 Other 18 (1) 9 (2) 9 (2) History of injection drug use 56 (6) 29 (6) 27 (7) CD4 lymphocyte count, mean SD, cells/mm3 243 184 249 186 236 181 Plasma HIV-1 RNA concentration, mean SD, log10 copies/mL 4.96 0.76 5.03 0.77 4.88 0.74 Hepatitis B surface antigen positivea 27 (3) 8 (2) 19 (5) Hepatitis C positivea 70 (8) 35 (8) 35 (9)

NOTE. Data are no. (%) of participants, unless otherwise indicated. a Baseline hepatitis B surface antigen and hepatitis C antibody test results were unavailable for 25 subjects. face antigen (P p .0144 ), and older age (P p .0149 ). Hepatitis ulations. Univariate logistic regression stratifie by race/ethnicity C antibody status and history of injection drug use were not was performed to identify associations between each polymor- significantl associated with CD4 lymphocyte response category. phism and CD4 lymphocyte response category. When the Ben- Genetic variants. We characterized 137 polymorphisms jamini-Hochberg procedure was used to correct for multiple across 17 genes by sequencing specimens from all 873 partic- comparisons, none were significant As exploratory analyses, ipants. Of these polymorphisms, 86 had been previously iden- the remainder of this analysis usesP ! .05 as the criterion for tified and 51 were discovered in this study. Genes encoding significance Because of the large number of SNPs, the results 10 proteins (accounting for 67 polymorphisms) important for are interpreted cautiously. There were significan associations T cell proliferation and/or survival were those for IL-2 (4 poly- with CD4 lymphocyte response category (P ! .05 ) (table 4) for morphisms), IL-2Rg (3 polymorphisms), IL-2Rb (10 polymor- 7 polymorphisms in genes encoding TRAIL (Ϫ1674TrG), TNF- phisms), IL-7 (5 polymorphisms), IL-7Ra (11 polymorphisms), a (Ϫ488GrA), Bim (26368CrT), IL-15Ra (14287ArG), IL- IL-15 (8 polymorphisms), IL-15Ra (10 polymorphisms), IFN-a 2Rb (Ϫ5585TrC), IL-15 (Ϫ156TrC), and IL-7Ra (4072CrG). (7 polymorphisms), IFN-b (5 polymorphisms), and TNF-a (4 There were trends toward associations (P ! .08 ) for poly- polymorphisms). Genes encoding 7 proteins (accounting for 70 morphisms in genes encoding IL-15Ra (13994ArG), IL-15 polymorphisms) important for T cell apoptosis were those for (Ϫ83125GrA), and TNFR-I (Ϫ1276GrC). Bim (9 polymorphisms), caspase 8 (14 polymorphisms), Fas (8 Multivariate logistic regression was performed controlling for polymorphisms), Fas ligand (8 polymorphisms), TNFR-I (9 poly- race/ethnicity as well as for age, sex, baseline CD4 lymphocyte morphisms), TNFR-II (14 polymorphisms), and TRAIL (8 poly- count, baseline plasma HIV-1 RNA concentration, and hepatitis morphisms). Of the 137 polymorphisms, 51 were in promoters, B antigen status. In these adjusted analyses (table 4), 3 of the 43 were in introns, 35 were in exons, and 8 were in 3 untranslated polymorphisms noted above remained significantl associated regions. Allelic frequencies exceeded 5% at 67 positions, exceeded with CD4 lymphocyte response category atP ! .05 —TRAIL 20% at 40 positions, and differed among white, black, and/or Hispanic participants by at least 10% at 54 positions. Information Table 3. Single-nucleotide polymorphisms in- on all polymorphism positions and allelic frequencies are pre- cluded in analyses of CD4 lymphocyte count sented in table 3. recovery during antiretroviral therapy. Associations between individual polymorphisms and CD4 lymphocyte count recovery. The frequencies of many poly- The table is available in its entirety in the online edition of the Journal of Infectious Diseases. morphisms differed considerably among the racial/ethnic pop-

Immunogenetics of CD4 Recovery • JID 2006:194 (15 October) • 1101 Table 4. Associations between single-nucleotide polymorphisms and CD4 lymphocyte response category.

CD4 lymphocyte count increase, no. (%)a of subjects P Gene Gene location Allele Encoded protein у200 cells/mm3 !200 cells/mm3 Unadjustedb Adjustedc TRAIL Ϫ1674 GG TRAIL 0 (0) 4 (1) .0016 .0019 GT 14 (3) 28 (7) TT 449 (97) 378 (91) TNFA Ϫ488 AA TNF-a 14 (3) 14 (3) .0049 .0052 AG 85 (18) 113 (27) GG 360 (78) 279 (67) BIM 26368 TT Bim 68 (15) 41 (10) .0378 .0216

CT 192 (41) 169 (41) Downloaded from https://academic.oup.com/jid/article/194/8/1098/870426 by guest on 28 September 2021 CC 202 (44) 198 (48) IL15RA 13994 AG IL-15 receptor a chain 15 (3) 5 (1) .0630 .0305 AA 446 (96) 403 (97) IL15 Ϫ83125 AA IL-15 2 (0) 0 (0) .0726 .0480 AG 23 (5) 13 (3) GG 436 (94) 396 (95) IL15RA 14287 AG IL-15 receptor a chain 5 (1) 13 (3) .0408 .0517 AA 456 (98) 394 (95) TNFRSF1A Ϫ1276 CC TNF receptor I 4 (1) 0 (0) .0525 .0538 GC 23 (5) 14 (3) GG 430 (93) 393 (94) IL2RB Ϫ5585 CC IL-2/IL-15 receptor common b chain 2 (0) 3 (1) .015 1.05 CT 23 (5) 35 (8) TT 433 (93) 366 (88) IL15d Ϫ156 CC IL-15 61 (13) 73 (18) .0212 1.05 CT 211 (45) 196 (47) TT 179 (39) 133 (32) IL7R 4072 GG IL-7 receptor a chain 40 (9) 20 (5) .0269 1.05 GC 171 (37) 146 (35) CC 249 (54) 244 (69)

NOTE. Bim, Bcl-2–interacting molecule; IL, interleukin; TNF, tumor necrosis factor; TRAIL, TNF-related apoptosis-inducing ligand. a Percentages may not add up to 100 because a small proportion of assays failed for some specimens. b Unadjusted analyses were not controlled for age, sex, baseline CD4 lymphocyte count, baseline plasma HIV-1 RNA concentration, or hepatitis B antigen status but were controlled for race/ethnicity. c Adjusted analyses were controlled for race/ethnicity, age, sex, baseline CD4 lymphocyte count, baseline plasma HIV-1 RNA concentration, and hepatitis B antigen status. d This IL15 polymorphism (Ϫ156TrC) is in near complete linkage disequilibrium with an adjacent Ϫ155TrC polymorphism.

Ϫ1674TrG, TNF-a Ϫ488GrA, and Bim 26368CrT—and 2 (70%) and was у200 cells/mm3 in 14 (30%) (P p .0016 ). How- additional polymorphisms became significantl associated, IL- ever, only 5% of the 873 participants had G alleles (table 4). 15Ra 13994ArG and IL-15 Ϫ83125GrA. There were trends Associations between haplotypes and CD4 lymphocyte count toward associations (P ! .06 ) for IL-15Ra 14287ArGand recovery. Haplotypes were reconstructed for each of the 17 TNFR-I Ϫ1276GrC. No other polymorphisms were signifi genes, and associations between haplotype variants and CD4 cantly associated with CD4 lymphocyte response category. lymphocyte response category were assessed. Haplotypes in 3 Although a number of polymorphisms were significantl as- genes were significantl associated with CD4 lymphocyte re- sociated with CD4 lymphocyte response category, the ability of sponse category by unadjusted analysis and/or analysis adjusted any individual polymorphism to predict the response category for race/ethnicity, age, sex, baseline CD4 lymphocyte count, base- for the total population was modest. Statistical significanc was line plasma HIV-1 RNA concentration, and hepatitis B antigen enhanced by the large sample size. For example, of 46 partic- status (figu e 1). For IFN-a, 4 haplotypes accounted for 87.9% ipants with at least 1 G allele at TRAIL Ϫ1674TrG (associated of all haplotypes. A haplotype (Ϫ2489C, Ϫ1823G, 1944G, 2005G, with less favorable CD4 lymphocyte count recovery), the in- 2167A, 2218A, and 2451G) present in 7.7% of participants was crease in CD4 lymphocyte count was !200 cells/mm3 in 32 significantl associated with an increase in CD4 lymphocyte

1102 • JID 2006:194 (15 October) • Haas et al. Downloaded from https://academic.oup.com/jid/article/194/8/1098/870426 by guest on 28 September 2021

Figure 1. Haplotypes associated with CD4 lymphocyte response category. Each panel shows the most frequently reconstructed haplotypes for interferon (IFN)–a (A), interleukin (IL)–2 (B), and IL-15 receptor a chain (IL-15Ra) (C). The indicated positions are relative to the ATG start site. Variant polymorphisms that define haplotypes are indicated by shaded boxes. Proportions of all haplotypes accounted for by these haplotypes are shown. Separate P values are given for unadjusted analyses (controlled for race/ethnicity only) and for analyses adjusted for race/ethnicity, age, sex, baseline CD4 lymphocyte count, baseline plasma HIV-1 RNA concentration, and hepatitis B antigen status. Asterisks identify haplotypes that are significantly associated with CD4 lymphocyte response category atP ! .05 . count of у200 cells/mm3 by unadjusted analysis (P p .006 ) and sence of main gene effects. A comprehensive analysis that in- by adjusted analysis (P p .021 ). Similarly, for IL-2, 4 haplotypes cluded all 137 polymorphisms across the 17 genes did not reveal accounted for 87.9% of all haplotypes. A haplotype (Ϫ476A, significan gene-gene interactions that were associated with CD4 Ϫ385T, 4303T, and 4462A) present in 7.2% of participants was lymphocyte response category. However, in an analysis involving associated with an increase in CD4 lymphocyte count of у200 5 genes encoding proteins important for IL-2/IL-15 signaling cells/mm3 by unadjusted analysis (P p .018 ) and by adjusted (IL-2, IL-15, IL-15Ra, IL-2Rb, and IL-2Rg), CD4 lymphocyte analysis (P p .014 ). For IL-15Ra, 4 haplotypes accounted for response category was predicted by an interaction between poly- 86.5% of all haplotypes. A haplotype (Ϫ2094G, Ϫ2092A, morphisms in genes encoding IL-2Rb (16491CrG) and IL-2Rg Ϫ1867A, Ϫ1609C, 13772C, 13781C, 13945T, 13994C, 14287G, (4735TrC) with 57% accuracy (odds ratio, 1.91 [95% confidenc and 14327T) present in 10.7% of participants was associated interval, 1.46–2.49];P ! .001 ) (figu e 2). with an increase in CD4 lymphocyte count of у200 cells/mm3 Influenc of CD4 lymphocyte response definition All anal- by unadjusted analysis (P p .060 ) and by adjusted analysis yses described above included all 873 participants with CD4 ().P p .049 lymphocyte counts of !800 cells/mm3 at baseline, with CD4 Exploring gene-gene interactions with MDR. We per- lymphocyte response categories dichotomized at 200 cells/mm3. formed MDR to assess whether gene-gene interactions were However, small fluctuation around the cutoff of 200 cells/mm3 associated with CD4 lymphocyte response category. This com- could change a participant’s response classificatio and con- putation-intensive method collapses multidimensional space found our ability to detect associations. Furthermore, some into low-risk and high-risk groups, allowing gene-gene inter- participants in these analyses enrolled in ACTG clinical trials actions for predicting outcome to be identifie even in the ab- when recommendations were to begin therapy at higher CD4

Immunogenetics of CD4 Recovery • JID 2006:194 (15 October) • 1103 (Ϫ1774GrA [P p .0310 ] and 21536GrA [P p .0443 ]). We did not identify strong new associations that were not apparent in the analyses involving all 873 participants. Additional post hoc analyses explored associations between genetic variants and CD4 lymphocyte response category as a continuous variable, analyzing white, black, and Hispanic sub- groups separately. By this analysis, no association was consis- tently seen in all 3 subgroups atP ! .05 . Four polymorphisms noted in previous sections remained associated with CD4 lym- phocyte response category in at least 1 subgroup (TRAIL Ϫ1674TrG in white participants, TNF-a Ϫ488GrA in black r

and Hispanic participants, and IL-2Rg 4735T C and Bim Downloaded from https://academic.oup.com/jid/article/194/8/1098/870426 by guest on 28 September 2021 26368CrT in Hispanic participants). Four additional poly- morphisms were associated with CD4 lymphocyte response cat- egory by this analysis (IL-15 Ϫ246CrT in white participants, 16892GrA and caspase 8 19903CrT in black participants, and IL-2Rb Ϫ5293TrA in Hispanic participants).

DISCUSSION

There is considerable interindividual variability in the magni- Figure 2. Multifactor dimensionality reduction models for predicting tude of CD4 lymphocyte count recovery during antiretroviral CD4 lymphocyte response category. Light grey cells indicate genotypes therapy. This is the firs study to assess whether allelic variants that are more likely to predict an increase in CD4 lymphocyte count of in genes that are important for T cell proliferation, survival, !200 cells/mm3, and dark grey cells indicate genotypes that are more likely to predict an increase in CD4 lymphocyte count of у200 cells/ and apoptosis are associated with CD4 lymphocyte count re- mm3. In each cell, the left bar represents participants with an increase covery. We performed extensive genetic analyses in 873 partic- of у200 cells/mm3, and the right bar represents participants with an ipants who received randomized antiretroviral regimens in pro- 3 increase of !200 cells/mm ; the nos. of participants are given above spective clinical trials and explored potential associations with each bar. This analysis involved 5 genes encoding proteins important for respect to individual polymorphisms, reconstructed haplotypes, interleukin (IL)–2/IL-15 signaling (IL-2, IL-15, IL-15 receptor [R] a chain, IL-2R b chain [IL-2Rb; common to IL-2R and IL-15R], and IL-2R g chain and gene-gene interactions. Because incomplete pharmacologic [IL-2Rg; common to IL-2R, IL-7R, and IL-15R]) (P ! .001 , best 2-loci model). control of HIV replication, whether due to regimen potency Because IL-2Rg is on the X and because women transcribe or adherence, may portend less robust increases in CD4 lym- only 1 of the 2 X , homozygotes are represented by single phocyte count, we included only individuals with sustained alleles. control of plasma viremia. We found that a number of poly- morphisms, alone or in combination, and several haplotypes lymphocyte counts, but current guidelines discourage the in- were associated with the magnitude of CD4 lymphocyte count itiation of antiretroviral therapy when CD4 lymphocyte counts recovery during antiretroviral therapy. Genetic variants were are 1350 cells/mm3 [4]. (Although increases in CD4 lymphocyte implicated in genes encoding IL-2, IL-2Rb, IL-2Rg, IL-15, IL- count were substantial even with relatively high baseline values; 15Ra, TRAIL, Bim, TNF-a, and IFN-g. Although numerous the 39 participants with counts of 600–800 cells/mm3 at baseline associations were identified the strengths of the associations experienced a median increase of 210 cells/mm3.) We therefore were relatively modest, with no profound effect for any par- repeated all analyses while excluding participants who either ticular variant and no variant consistently identifie across the (1) had increases in CD4 lymphocyte counts between 100 and different statistical methods that were used. These results sug- 200 cells/mm3 or (2) had counts of 1350 cells/mm3 at baseline. gest that immune reconstitution during antiretroviral therapy These post hoc analyses included 429 participants. Although is a complex phenotype that is influence by multiple genetic the smaller sample size reduced the power to detect associa- or other factors. tions, the results were generally consistent with the finding in Mechanisms underlying the genotype-phenotype associations the full data set. By univariate logistic regression stratifie by identifie in this study are speculative at present, because the race/ethnicity, 5 of the polymorphisms shown in table 4 were functional effects of these polymorphisms are not known. For associated atP ! .05 (TRAIL Ϫ1674TrG, TNF-a Ϫ488GrA, example, if TRAIL Ϫ1674TrG and/or TNF-a Ϫ488GrA were Bim 26368CrT, IL-2Rb Ϫ5585TrC, IL-15 Ϫ156TrC, and associated with increased signaling through extrinsic apoptosis IL-7Ra 4072CrG), as were 2 polymorphisms in TNFR-II pathways, this could decrease CD4 lymphocyte count recovery.

1104 • JID 2006:194 (15 October) • Haas et al. Conversely, if Bim 26368CrT was associated with decreased Acknowledgments signaling through the intrinsic apoptosis pathway, this could The present work was supported in part by the AIDS Clinical Trials lead to more favorable responses. More favorable responses Group (ACTG), funded by the National Institute of Allergy and Infectious associated with IL-15Ra 13994ArG, IL-15 Ϫ83125GrA, IL- Diseases (grant AI38858); investigator support included grants AI46339 (to r D.W.H.), AI38855 (to J.A., J.M., and C.A.B.), AI54999 (to D.W.H.), 15Rb 16491C G, and/or the IL-2 haplotype could be explained AI27670 (to C.A.B.), and AI29193 (to R.T.D.), and clinical trials unit grants by enhanced cytokine signaling by these variants, as could the were AI25915, AI25924, AI25868, CFAR AI50410, AI27670, AI34832, gene-gene interaction identifie between IL-15Rb 16491CrG AI25897, AI46383, AI27661, AI27661-19S2, AI13656, 303-0804, AI27659, r AI27659-18S2, AI25859-19, AI25903, AI27658, AI27665, AI27675,AI25879, and IL-2Rg 4735T C. AI32770, AI27660, AI46386, AI27666, AI46376-05, AI34853, AI32782, The strategy for selecting genetic variants for analysis focused AI46370, AI27664, AI27668, AI46381, AI38858-09S1, and ISL 204IC005. largely on tagging SNPs. These are polymorphisms that are The present work was also supported in part by the General Clinical Re- search Center Units, funded by the National Center for Research Resources known to be in linkage disequilibrium with numerous other (grants RR00046, RR00044, RR00096, RR00051, RR00047, RR00052, and polymorphisms that were not included in these analyses. This RR00095). Downloaded from https://academic.oup.com/jid/article/194/8/1098/870426 by guest on 28 September 2021 approach captures much of the population diversity of these We are grateful to the many persons with HIV infection who volunteered for this study. In addition, we acknowledge the contributions of the many genes with a feasible number of assays, potentially giving a more ACTG personnel involved in protocol A5001. comprehensive picture of genotype-phenotype relationships. Other A5001 team members included Ann Collier, Jeanne Conley, and This strategy differs from more functional approaches that em- Joan Dragavon (University of Washington, Seattle); Ron Bosch, Marlene Smurzynski, and Kunling Wu (Harvard School of Public Health, Boston, phasize polymorphisms that have been shown to alter gene MA); Barbara Bastow (Social and Scientifi Systems, Inc., Silver Spring, expression or protein function. It is possible that further char- MD); Mike Klebert (Washington University, St. Louis, MO); and Bernadette acterization of additional polymorphisms in these genes would Jarocki and Nancy Webb (Frontier Science and Technology Research Foun- dation, Amherst, NY). identify stronger associations with CD4 lymphocyte count The following persons and institutions participated in the conduct of recovery. A5001: Donna McGregor and Margarita Aguilar (Northwestern University, The present study had several limitations. Associations be- Chicago, IL); Harold A. Kessler and Beverly E. Sha (Rush University, Chi- cago, IL); Oluwatoyin Adeyemi and Joanne Despotes (CORE Center, Chi- tween genetic polymorphisms and clinical phenotypes do not cago, IL); Michael F. Para and Susan L. Koletar (Ohio State University, establish causation. Other polymorphisms linked to those char- Columbus); David Ragan and Susan Pedersen (University of North Car- olina, Chapel Hill); Timothy Lane and Kim Epperson (Moses Cone Health acterized here, or in other genes, may be very important. Of System, Greensboro, NC); Linda Meixner and Julie Hoffman (University particular interest would be genes involved in T cell homing of California, San Diego); Santiago Marreo and Jorge L. Santana (University and trafficking because much of the increase in CD4 lympho- of Puerto Rico, San Juan); Diane Daria and Judith Feinberg (University of Cincinnati, Cincinnati, OH); Deborah McMahon and Barbara Rutecki cyte count during the firs 8–12 weeks of antiretroviral therapy (University of Pittsburgh, Pittsburgh, PA); Princy N. Kumar and Joseph may reflec redistribution from lymphoid tissues into peripheral Timpone (Georgetown University Hospital, Washington, DC); Winston blood [44]. Associations we identifie may not translate to all Cavert and Robyn Schacherer (University of Minnesota, Minneapolis); Jeff Meier and Ann Wiley (University of Iowa, Iowa City); Susan Swindells racial/ethnic groups, because haplotype structures may differ and Frances Van Meter (University of Nebraska, Omaha); Fred R. Sattler between populations. Importantly, because associations sug- and Connie A. Funk (University of Southern California, Los Angeles); J. gested by a firs genetic study may not be validated in subse- Michael Kilby and Karen Savage (University of Alabama, Birmingham); Molly Eaton and Clifford Gunthel (Emory University, Atlanta, GA); Martha quent analyses [45], it is critical to confi m—or refute—the Silberman and Deitra Wade (Duke University Medical Center, Durham, relationships between these variants and increases in CD4 lym- NC); Amy Sbrola (Harvard University/Massachusetts General Hospital, phocyte count during antiretroviral therapy in other studies. Boston); Neah Kim (Beth Israel Deaconess, Boston, MA); Charlene Gaca and Paul R. Skolnik (Boston Medical Center, Boston, MA); Mitchell Gold- Because we characterized numerous polymorphisms, it is al- man and Carol Schnizlein Bick (Indiana University, Indianapolis); David most certain that at least some genotype-phenotype associations B. Clifford, Ge-Youl Kim, and Michael K. Klebert (Washington University, suggested by this study occurred by chance and are, therefore, St. Louis, MO); Michael Morgan and Vicki Bailey (Vanderbilt University, Nashville, TN); Christine Hurley and Mary Shoemaker (University of Roch- not valid [46, 47]. ester, Rochester, NY); Gene D. Morse and Tamara O’Hara (State University Major initiatives are providing antiretroviral medications to of New York, Buffalo); Nayef El-daher and Mary Shoemaker (St. Mary’s HIV-infected people worldwide. For patients to benefi from the Hospital, Rochester, NY); Susan Holland and Karen Cavanagh (New York University/Bellevue, New York, NY); Margaret A. Fischl and Jose G. Castro revolution in human , it is critical to thoroughly char- (University of Miami, Miami, FL); Trisha Walton and Barbara Philpotts acterize associations between genetic variants and outcomes. (Case Western Reserve University, Cleveland, OH); Kim Whitely and Mary Large databases from prospective clinical trials, linked to robust Wild (MetroHealth, Cleveland, OH); M. Graham Ray, Beverly Putnam, and Cathi Basler (University of Colorado, Denver); Judith Currier (Uni- DNA repositories, greatly facilitate such work. The present study versity of California, Los Angeles); Sadia Shaik and Mario Guerrero (Harbor should encourage continued investigations into the importance General/University of California, Los Angeles); Joe Quinn and Harvey of human genetics for immune reconstitution during HIV ther- Friedman (University of Pennsylvania, Philadelphia); Jolene Noel-Connor and Madeline Torres (Columbia University, New York, NY); Valery Hughes apy. Emerging knowledge in this area should ultimately be used and Glenn Sturge (Cornell Clinical Trials Unit, New York, NY); Todd to test strategies to improve HIV treatment outcomes. Stroberg and Natacha Joseph (Cornell Chelsea Clinic, New York, NY); Pat

Immunogenetics of CD4 Recovery • JID 2006:194 (15 October) • 1105 Cain and Sandy Valle (Stanford University, Stanford, CA); Philip Keiser survival of activated T cells in vitro and in vivo. Proc Natl Acad Sci and Tianna Petersen (University of Texas, Southwestern Medical Center, USA 1998; 95:3810–5. Dallas); Nancy Hanks and Loran Nagamine (University of Hawaii, Manoa); 17. Smith KA. Interleukin-2: inception, impact, and implications. Science William A. O’Brien and Gerianne Casey (University of Texas, Galveston); 1988; 240:1169–76. Donna Mildvan and Gwendolyn Costantini (Beth Israel Medical Center, 18. Leung DT, Morefiel S, Willerford DM. Regulation of lymphoid ho- New York, NY); Ann C. Collier and N. Jeanne Conley (University of Wash- meostasis by IL-2 receptor signals in vivo. J Immunol 2000; 164:3527–34. ington, Seattle); Ilene Wiggins and Robin Worrell-Thorne (Johns Hopkins 19. Lin JX, Migone TS, Tsang M, et al. The role of shared receptor motifs University, Baltimore, MD); Diane Havlir and Brad Hare (San Francisco and common Stat proteins in the generation of cytokine pleiotropy General Hospital, San Francisco, CA); Charles E. Davis, Jr., and Onyinye and redundancy by IL-2, IL-4, IL-7, IL-13, and IL-15. 1995;2: Erondu (University of Maryland, Baltimore); Pamela Poethke and Kath- 331–9. erine Wright (Miriam Hospital/Brown University, Providence, RI); David 20. Nagata S. Fas ligand-induced apoptosis. Annu Rev Genet 1999; 33:29–55. Asmuth and Melissa Schreiber (University of California, Davis). 21. Miura Y, Misawa N, Maeda N, et al. Critical contribution of tumor Luc van Kaer, Jacek Hawiger, and Mark Boothby (Department of Mi- necrosis factor-related apoptosis-inducing ligand (TRAIL) to apoptosis crobiology and , Vanderbilt University, Nashville, TN) pro- of human CD4+ T cells in HIV-1-infected hu-PBL-NOD-SCID mice. vided helpful suggestions regarding candidate genes to include in this study. J Exp Med 2001; 193:651–60.

22. Hsu H, Xiong J, Goeddel DV. The TNF receptor 1-associated protein Downloaded from https://academic.oup.com/jid/article/194/8/1098/870426 by guest on 28 September 2021 TRADD signals cell death and NF-kappa B activation. Cell 1995; 81: References 495–504. 23. Gupta S. Molecular steps of death receptor and mitochondrialpathways 1. McCune JM, Hanley MB, Cesar D, et al. Factors influencin T-cell of apoptosis. Life Sci 2001; 69:2957–64. turnover in HIV-1-seropositive patients. J Clin Invest 2000; 105:R1–8. 24. Shi Y. Mechanisms of caspase activation and inhibition during apo- 2. Stellbrink HJ, van Lunzen J. Lymph nodes during antiretroviral therapy. ptosis. Mol Cell 2002; 9:459–70. Curr Opin Infect Dis 2001; 14:17–22. 25. Strasser A, Harris AW, Huang DC, Krammer PH, Cory S. Bcl-2 and 3. Deeks SG, Hoh R, Grant RM, et al. CD4+ T cell kinetics and activation Fas/APO-1 regulate distinct pathways to lymphocyte apoptosis. EMBO in human immunodeficienc –infected patients who remain vi- J 1995; 14:6136–47. remic despite long-term treatment with protease inhibitor–based ther- 26. Van PL, Biuckians A, Abbas AK. Functional roles of Fas and Bcl-2- apy. J Infect Dis 2002; 185:315–23. regulated apoptosis of T lymphocytes. J Immunol 1998; 160:2065–71. 4. Panel on Clinical Practices for Treatment of HIV Infection. Guidelines 27. Hildeman DA, Zhu Y, Mitchell TC, Kappler J, Marrack P. Molecular for the use of antiretroviral agents in HIV-1-infected adults and adoles- mechanisms of activated T cell death in vivo. Curr Opin Immunol cents (updated 6 October 2005). Available at: http://aidsinfo.nih.gov/ 2002; 14:354–9. guidelines/GuidelineDetail.aspx?MenuItempGuidelines&SearchpOff& 28. Bouillet P, Metcalf D, Huang DC, et al. Proapoptotic Bcl-2 relative GuidelineIDp7&ClassIDp1. Accessed 20 March 2006. Bim required for certain apoptotic responses, leukocyte homeostasis, 5. Staszewski S, Morales-Ramirez J, Tashima KT, et al. Efavirenz plus and to preclude . Science 1999; 286:1735–8. zidovudine and lamivudine, efavirenz plus indinavir, and indinavir plus 29. Herbeuval JP, Hardy AW, Boasso A, et al. Regulation of TNF-related zidovudine and lamivudine in the treatment of HIV-1 infection in apoptosis-inducing ligand on primary CD4+ T cells by HIV-1: role of adults. Study 006 Team. N Engl J Med 1999; 341:1865–73. type I IFN-producing plasmacytoid dendritic cells. Proc Natl Acad Sci 6. Lederman MM, McKinnis R, Kelleher D, et al. Cellular restoration in USA 2005; 102:13974–9. HIV infected persons treated with abacavir and a protease inhibitor: 30. Haas DW, Wilkinson GR, Kuritzkes DR, et al. A multi-investigator/ age inversely predicts naive CD4 cell count increase. AIDS 2000; 14: institutional DNA bank for AIDS-related human genetic studies: AACTG 2635–42. protocol A5128. HIV Clin Trials 2003; 4:287–300. 7. Lederman MM, Connick E, Landay A, et al. Immunologic responses 31. Rozen S, Skaletsky H. Primer3 on the WWW for general users and associated with 12 weeks of combination antiretroviral therapy con- for biologist programmers. Methods Mol Biol 2000; 132:365–86. sisting of zidovudine, lamivudine, and ritonavir: results of AIDS Clin- 32. Smit AF. Interspersed repeats and other mementos of transposable ele- ical Trials Group Protocol 315. J Infect Dis 1998; 178:70–9. ments in mammalian . Curr Opin Genet Dev 1999; 9:657–63. 8. Keita-Perse O, Roger PM, Pradier C, Pugliese P, Cottalorda J, Della- 33. Murphy WJ, Bourque G, Tesler G, Pevzner P, O’Brien SJ. Recon- monica P. Do viral load and CD8 cell count at initiation of tritherapy structing the genomic architecture of mammalian ancestors using mul- influenc the increase of CD4 T-cell count? AIDS 1998; 12:F175–9. tispecies comparative maps. Hum Genomics 2003; 1:30–40. 9. Renaud M, Katlama C, Mallet A, et al. Determinants of paradoxical 34. Margulies EH, Green ED. Detecting highly conserved regions of the CD4 cell reconstitution after protease inhibitor-containing antiretro- human genome by multispecies sequence comparisons. Cold Spring viral regimen. AIDS 1999; 13:669–76. Harb Symp Quant Biol 2003; 68:255–63. 10. Schluns KS, Lefrancois L. Cytokine control of memory T-cell devel- 35. Geraghty DE, Fortelny S, Guthrie B, et al. Data acquisition, data storage, opment and survival. Nat Rev Immunol 2003; 3:269–79. and data presentation in a modern genetics laboratory. Rev Immu- 11. Khaled AR, Durum SK. Lymphocide: cytokines and the control of nogenet 2000; 2:532–40. lymphoid homeostasis. Nat Rev Immunol 2002; 2:817–30. 36. Geraghty DE, Daza R, Williams LM, Vu Q, Ishitani A. Genetics of the 12. Lodolce JP, Boone DL, Chai S, et al. IL-15 receptor maintains lymphoid immune response: identifying immune variation within the MHC and homeostasis by supporting lymphocyte homing and proliferation. Im- throughout the genome. Immunol Rev 2002; 190:69–85. munity 1998; 9:669–76. 37. Ewing B, Hillier L, Wendl MC, Green P. Base-calling of automated se- 13. Schluns KS, Kieper WC, Jameson SC, Lefrancois L. Interleukin-7 me- quencer traces using phred. I. Accuracy assessment. Genome Res 1998; diates the homeostasis of naive and memory CD8 T cells in vivo. Nat 8:175–85. Immunol 2000; 1:426–32. 38. Gordon D, Abajian C, Green P. Consed: a graphical tool for sequence 14. Kanegane H, Tosato G. Activation of naive and memory T cells by finishing Genome Res 1998; 8:195–202. interleukin-15. Blood 1996; 88:230–5. 39. Smith WP, Vu Q, Li SS, Hansen JA, Zhao LP, Geraghty DE. Toward 15. Maraskovsky E, O’Reilly LA, Teepe M, Corcoran LM, Peschon JJ, Stras- understanding MHC disease associations: partial resequencing of 46 ser A. Bcl-2 can rescue T lymphocyte development in interleukin-7 distinct HLA haplotypes. Genomics 2006; 87:561–71. receptor-deficien mice but not in mutant rag-1Ϫ/Ϫ mice. Cell 1997; 40. Nickerson DA, Tobe VO, Taylor SL. PolyPhred: automating the de- 89:1011–9. tection and genotyping of single nucleotide substitutions using fluo 16. Vella AT, Dow S, Potter TA, Kappler J, Marrack P. Cytokine-induced rescence-based resequencing. Nucleic Acids Res 1997; 25:2745–51.

1106 • JID 2006:194 (15 October) • Haas et al. 41. Ritchie MD, Hahn LW, Roodi N, et al. Multifactor-dimensionality 44. Bucy RP, Hockett RD, Derdeyn CA, et al. Initial increase in blood reduction reveals high-order interactions among estrogen-metabolism CD4+ lymphocytes after HIV antiretroviral therapy reflect redistri- genes in sporadic breast cancer. Am J Hum Genet 2001; 69:138–47. bution from lymphoid tissues. J Clin Invest 1999; 103:1391–8. 42. Ritchie MD, Hahn LW, Moore JH. Power of multifactor dimensionality 45. Ioannidis JP, Ntzani EE, Trikalinos TA, Contopoulos-Ioannidis DG. reduction for detecting gene-gene interactions in the presence of ge- Replication validity of genetic association studies. Nat Genet 2001; 29: notyping error, missing data, phenocopy, and genetic heterogeneity. 306–9. Genet Epidemiol 2003; 24:150–7. 46. Colhoun HM, McKeigue PM, Davey SG. Problems of reporting genetic 43. Hahn LW, Ritchie MD, Moore JH. Multifactor dimensionality reduc- associations with complex outcomes. Lancet 2003; 361:865–72. tion software for detecting gene-gene and gene-environment interac- 47. Freimer NB, Sabatti C. Guidelines for association studies in human tions. Bioinformatics 2003; 19:376–82. . Hum Mol Genet 2005; 14:2481–3. Downloaded from https://academic.oup.com/jid/article/194/8/1098/870426 by guest on 28 September 2021

Immunogenetics of CD4 Recovery • JID 2006:194 (15 October) • 1107