Jean-François Zagury

A linear complexity phasing method for thousands of genomes

Human-disease etiology can be better understood with phase information about diploid sequences. We present a method for estimating haplotypes, using genotype data from unrelated samples or small nuclear families, that leads to improved accuracy and speed compared to several widely used methods. The method, segmented haplotype estimation and imputation tool (SHAPEIT), scales linearly with the number of haplotypes used in each iteration and can be run efficiently on whole chromosomes.

Improved whole-chromosome phasing for disease and population genetic studies

SHAPEIT2 uses multithreading so that multiple cores can be used to phase whole chromosomes, allowing users to make the best use of their computational resources. We tested SHAPEIT2 on several large-sample, whole-chromosome data sets from a range of SNP genotyping chips (Supplementary Note 1). SHAPEIT2 outperforms other methods (Fig. 1a–c) in terms of switch error rate (SER) and the mean distance between switch errors (Supplementary Figs. 1 and 2). As compared to SHAPEIT1, SHAPEIT2 reduced SER by as much as 45% on these data sets. For example, on 1,229 Vietnamese samples assayed on the Illumina 660K chip on chromosome 22, the SERs of SHAPEIT2 (K = 100, W = 2 Mb), SHAPEIT1 (K = 100) (ref. 2), HAPI-UR (v1.01) (ref. 4), Beagle (v3.3) (ref. 5), Impute2 v2.1.2 (K = 100) (ref. 3), MaCH v1.0.18 (K = 100) (ref. 6) and fastPHASE (v1.4) (ref. 7) were 2.87%, 4.64%, 4.75%, 5.14%, 5.57%, 6.05% and 6.34%, respectively. In general, SHAPEIT2 with low values of K outperformed SHAPEIT1 with high values of K (Fig. 1a–c). As the number of samples increased (up to ~9,000 samples in our tests), we found that SHAPEIT2 outperformed other methods and had the property that SER decreases as sample size increases (Fig. 1d). We assessed accuracy on sequence data by phasing 381 European samples from the 1000 Genomes Project (TGP) together with genotypes from two trio parents sequenced at high coverage. We found that SHAPEIT2 (K = 100, W = 0.3 Mb) reduced SER by 38% compared to Beagle (Supplementary Table 1 and Supplementary Note 2), illustrating that the SHAPEIT2 model can adapt to data sets with very high SNP density. The computational performance of SHAPEIT2 is competitive compared to other methods. Figure 1e shows the computational Improved whole-chromosome phasing for disease and population genetic studies

Effects of CCR5-Delta32, CCR2-64I, and SDF-1 3'A alleles on HIV-1 disease progression: An international meta-analysis of individual-patient data

The burgeoning information on the human genome creates opportunities and challenges for studies of disease associations. Because genetic differences often produce modest effects, many patients must be studied to reach definitive conclusions. In the absence of a single large study, meta-analysis of individual-patient data (13) from smaller studies offers a way to assemble an adequate sample size. This approach is based on a unifying protocol that has standardized analytic definitions. When the protocol is applied to data contributed by most investigators working in a field, this method can provide more convincing results than a simple pooling of data or a meta-analysis of published reports (3). A meta-analysis of individual-patient data is also superior to a meta-analysis of published reports for examining differences in reported results. Host genetic variability affects the risk for AIDS after infection with HIV-1, but the effect of specific alleles (that is, alternative forms of a gene that exist at a specific chromosomal location [locus]) has been inconsistent. C-C chemokine receptor 5 (CCR5) is a major co-receptor for HIV-1, but CCR5- 32, an allele that contains a 32base pair deletion, codes for a nonfunctional co-receptor (46). CCR5-32 homozygotes (people who inherited the CCR5-32 allele from both parents) are highly resistant to HIV-1 infection (47). CCR5-32 heterozygotes (people who inherited the CCR5-32 allele from one parent and a functional CCR5 allele from the other parent) are susceptible to HIV-1 infection; however, according to some reports (6, 820), they progress from HIV-1 infection to AIDS more slowly than persons with two normal CCR5 alleles (called wild-type individuals). CCR2b is a minor HIV-1 co-receptor. The gene that codes for this chemokine receptor has a variant allele (CCR2-64I) that may affect the risk for AIDS (21). Epidemiologic studies of CCR2-64I carriers have been inconsistent (1826), and in vitro studies have identified no functional differences between cells from CCR2-64I carriers and those from wild-type patients (27, 28). Stromal cellderived factor-1 (SDF-1) is the chemokine ligand of CXCR4, an important co-receptor for HIV-1 late in the disease course (29, 30). Homozygosity for the SDF-1 3A allele has been reported to slow disease progression (31), but not in all studies (18, 19, 3235). To address these inconsistencies, we conducted an international meta-analysis of individual-patient data on the CCR5, CCR2, and SDF-1 alleles; data were contributed by 19 teams of investigators. Methods Organization of the Meta-Analysis All research teams investigating associations of genetic alleles with the course of HIV-1 disease progression were invited to contribute individual-patient data to the International Meta-Analysis of HIV Host Genetics. Collaborating teams were identified through MEDLINE searches, cited references of retrieved papers, abstracts of major HIV-related meetings, and communication with investigators working in the field. The meta-analysis was also announced in Nature Medicine (36), on the Web site of the International Cochrane Collaboration, and at HIV scientific meetings. A common protocol was developed in collaboration with research teams identified through these efforts. The meta-analysis database remained open until 12 February 1999 for the collection of CCR5- 32 and CCR2-64I data; because most of the participating investigator teams evaluated SDF-1 3A after they studied CCR5-32 and CCR2-64I, we collected data on SDF-1 3A until 30 November 1999. Selection of Databases Prospective cohort studies of patients with HIV-1 infection were included in the analysis if they had collected information on the pertinent genotypes, as well as on the time from seroconversion or study entry to the development of AIDS and to death. We excluded studies if genetic data or time-to-event data were unavailable or if the participants were enrolled after 1 January 1996. We included casecontrol studies if they compared patients with rapid versus slower rates of progression. Casecontrol studies were analyzed separately from prospective cohort studies. Definitions and End Points The prospective cohort studies in our meta-analysis typically had follow-up visits every 6 months. Within each study, we divided the participants according to enrollment before (seroconverters) or after (seroprevalent patients) HIV-1 infection. For seroconverters, a negative result and a subsequent positive result on enzyme-linked immunosorbent assay (ELISA) and Western blot test were obtained after enrollment. We analyzed the data on a time scale that originated at the date of study entry for seroprevalent participants and at the estimated date of seroconversion (calculated as the midpoint between the last study visit at which the patient tested negative for HIV-1 and the first visit at which the patient tested positive for HIV-1) for the seroconverters. The cohorts differed little in the precision of the estimated date of seroconversion because semiannual data were typically available. Data for patients of European descent and data for patients of African descent (37) were considered separately. Our analysis examined four major outcomes: 1) time from seroconversion (or study entry) to the development of AIDS, according to 1987 criteria by the U.S. Centers for Disease Control and Prevention [38]; 2) time from seroconversion [or study entry] to death; 3) time from development of AIDS to death; and 4) serum or plasma HIV-1 RNA level, which was measured by using a consistent method within each study. For seroconverters, we used the first measurement of HIV-1 RNA level recorded since onset of chronic HIV-1 infection (range, 6 to 42 months after the estimated date of seroconversion); for seroprevalent patients, we used the first study measurement of HIV-1 RNA level. We censored data on follow-up after 1 January 1996 to minimize the effects of potent antiretroviral therapy. The average follow-up to AIDS development or to the point of censoring in the CCR5- 32 and CCR2-64I analyses was 6.73 years for seroconverters and 6.37 years for seroprevalent patients; for the SDF-1 analysis, the average follow-up was 7.14 years for the seroconverters and 6.51 years for seroprevalent patients. We specified our outcome variables a priori and asked all investigators to contribute data in a format consistent with the protocol. Investigators at the coordinating center, which was located at the National Cancer Institute in Rockville, Maryland, communicated with the contributing investigators to verify that the data from each study adhered to the common definitions of the meta-analysis. The contributed data sets also underwent logical tests to identify internal inconsistencies or incompatibilities. Any missing information or errors in logic that were identified at the coordinating center were referred to the contributing investigators; all identified errors were resolved. Statistical Analysis We used Cox regression to determine hazard ratios (relative hazards) for the times to events for all study cohorts and subgroups (39). The hazard ratio approximates the relative risk or incidence risk ratio. A log10 transformation was used for all analyses of HIV-1 RNA level. Differences in HIV-1 RNA level within studies were analyzed as differences for independent samples of continuous variables. Pooled summary estimates of hazard ratios and differences of means were obtained by weighting estimates from each study by the inverse of its variance (1, 40). We estimated fixed effects and random effects (1, 40). Fixed-effects models assume that any differences in results among studies are simply due to chance, whereas random-effects models assume that there may be true differences in the results of different studies. We report random-effects estimates because these provide more conservative confidence intervals when the results are highly heterogeneous across cohorts. (In the absence of heterogeneity, fixed-effects and random-effects estimates coincide.) We assessed heterogeneity by using the Q statistic, which we considered to be significant if the P value was less than 0.10 (1). However, some cohorts had few SDF-1 3A homozygotes with clinical events. Therefore, we also calculated an efficient score test for heterogeneity on the basis of the appropriate interaction terms between genotype and cohort in a cohort-stratified proportional hazards model. Inferences were similar with both tests. To model the effect of rare genotypes (CCR2-64I homozygotes or patients who were heterozygous for CCR5- 32 and CCR2-64I), we fit Cox models to the pooled data from all cohort studies with stratification by study. For cohorts with no events among SDF-1 3A homozygotes, we estimated log relative hazards values on the basis of a penalized likelihood with a penalty term of 0.5 (log[1 + exp ()] ). Penalty terms shrink extreme values (which also have large variances) toward zero. Because these extreme estimates have small weights, they contribute little to the overall results. Their exclusion yielded results similar to those of the main analysis. All calculations were performed by using the MATLAB software package, version 5.3 (The MathWorks, Inc., Natick, Massachusetts). Results We restricted the main analysis of CCR5- 32 and CCR2-64I to patients of European or African descent with genotype data for both alleles (Table 1). Because CCR5- 32 and CCR2-64I are in complete linkage of disequilibrium (nonrandom association of alleles that lie close together on a chromosome) (21), the CCR5- 32 and CCR2-64I alleles are never found on the same paternal or maternal chromosome. Therefore, we compared patients with a variant allele to patients who were wild-type homozygotes for both CCR5 and CCR2. Because the CCR5- 32 allele is almost exclusively found in persons of European descent (46), the analysis of CCR5-32 was limited to such persons. The analysis of SDF-1 3A was limited to persons

Genomewide association study of an AIDS-nonprogression cohort emphasizes the role played by HLA genes (ANRS genomewide association study 02)

To elucidate the genetic factors predisposing to AIDS progression, we analyzed a unique cohort of 275 human immunodeficiency virus (HIV) type 1-seropositive nonprogressor patients in relation to a control group of 1352 seronegative individuals in a genomewide association study (GWAS). The strongest association was obtained for HCP5 rs2395029 (P=6.79x10(-10); odds ratio, 3.47) and was possibly linked to an effect of sex. Interestingly, this single-nucleotide polymorphism (SNP) was in high linkage disequilibrium with HLA-B, MICB, TNF, and several other HLA locus SNPs and haplotypes. A meta-analysis of our genomic data combined with data from the previously conducted Euro-CHAVI (Center for HIV/AIDS Vaccine Immunology) GWAS confirmed the HCP5 signal (P=3.02x10(-19)) and identified several new associations, all of them involving HLA genes: MICB, TNF, RDBP, BAT1-5, PSORS1C1, and HLA-C. Finally, stratification by HCP5 rs2395029 genotypes emphasized an independent role for ZNRD1, also in the HLA locus, and this finding was confirmed by experimental data. The present study, the first GWAS of HIV-1 nonprogressors, underscores the potential for some HLA genes to control disease progression soon after infection.

A General Approach for Haplotype Phasing across the Full Spectrum of Relatedness

Many existing cohorts contain a range of relatedness between genotyped individuals, either by design or by chance. Haplotype estimation in such cohorts is a central step in many downstream analyses. Using genotypes from six cohorts from isolated populations and two cohorts from non-isolated populations, we have investigated the performance of different phasing methods designed for nominally ‘unrelated’ individuals. We find that SHAPEIT2 produces much lower switch error rates in all cohorts compared to other methods, including those designed specifically for isolated populations. In particular, when large amounts of IBD sharing is present, SHAPEIT2 infers close to perfect haplotypes. Based on these results we have developed a general strategy for phasing cohorts with any level of implicit or explicit relatedness between individuals. First SHAPEIT2 is run ignoring all explicit family information. We then apply a novel HMM method (duoHMM) to combine the SHAPEIT2 haplotypes with any family information to infer the inheritance pattern of each meiosis at all sites across each chromosome. This allows the correction of switch errors, detection of recombination events and genotyping errors. We show that the method detects numbers of recombination events that align very well with expectations based on genetic maps, and that it infers far fewer spurious recombination events than Merlin. The method can also detect genotyping errors and infer recombination events in otherwise uninformative families, such as trios and duos. The detected recombination events can be used in association scans for recombination phenotypes. The method provides a simple and unified approach to haplotype estimation, that will be of interest to researchers in the fields of human, animal and plant genetics.

Associations of MHC Ancestral Haplotypes with Resistance/Susceptibility to AIDS Disease Development

We tested the association of MHC ancestral haplotypes with rapid or slow progression to AIDS by comparing their frequencies in the French genetics of resistance/susceptibility to immunodeficiency virus cohort with that reported in a control French population. Seven ancestral haplotypes were identified in the genetics of resistance/susceptibility to immunodeficiency virus cohort with a frequency >1%. The 8.1 (odds ratio (OR) = 3, p = 0.006), 35.1 (OR = 5.7, p = 0.001), and 44.2 (OR = 3.4, p = 0.007) ancestral haplotypes were associated with rapid progression, whereas the 35.2 (OR = 3.6, p = 0.001), 44.1 (OR = 5.4, p < 10−4), and 57.1 (OR = 5.8, p < 10−4) ancestral haplotypes were associated with slow progression to AIDS. Although the frequency of each ancestral haplotype is low in the population, the OR were quite higher than those previously obtained for single HLA allele associations, with some p values as low as 10−4. The analysis of the recombinant fragments of these haplotypes allowed the identification of the MHC regions in the 35.1, 35.2, and 44.2 haplotypes associated with rapid progression to AIDS and the MHC regions of the 44.1 and 57.1 haplotypes associated with slow progression to AIDS. Previous studies have identified single HLA alleles associated with disease progression. Our results on recombinant fragments confirm the direct role of HLA-B35 in rapid progression. Associations with HLA-A29 and -B57 might be due to linkage disequilibrium with other causative genes within the MHC region.

Genomewide Association Study of a Rapid Progression Cohort Identifies New Susceptibility Alleles for AIDS (ANRS Genomewide Association Study 03)

BACKGROUND Previous genomewide association studies (GWASs) of AIDS have targeted end points based on the control of viral load and disease nonprogression. The discovery of genetic factors that predispose individuals to rapid progression to AIDS should also reveal new insights into the molecular etiology of the pathology. METHODS We undertook a case-control GWAS of a unique cohort of 85 human immunodeficiency virus type 1 (HIV-1)-infected patients who experienced rapid disease progression, using Illumina HumanHap300 BeadChips. The case group was compared with a control group of 1352 individuals for the 291,119 autosomal single-nucleotide polymorphisms (SNPs) passing the quality control tests, using the false-discovery rate (FDR) statistical method for multitest correction. RESULTS Novel associations with rapid progression (FDR, < or = 25%) were identified for PRMT6 (P = 6.1 x 10(-7); odds ratio [OR], 0.24), SOX5 (P = 1.8 x 10(-6); OR, 0.45), RXRG (P = 3.9 x 10(-6); OR, 3.29), and TGFBRAP1 (P = 7 x 10(-6); OR, 0.34). The haplotype analysis identified exonic and promoter SNPs potentially important for PRMT6 and TGFBRAP1 function. CONCLUSIONS The statistical and biological relevance of these associations and their high ORs underscore the power of extreme phenotypes for GWASs, even with a modest sample size. These genetic results emphasize the role of the transforming growth factor beta pathway in the pathogenesis of HIV-1 disease. Finally, the wealth of information provided by this study should help unravel new diagnostic and therapeutic targets.

32 bp CCR-5 gene deletion and resistance to fast progression in HIV-1 infected heterozygotes

there are known racial differences in the frequencies of the PM phenotype: about 3% of whites and in 13–23% of orientals. 3 Poor metabolism results from a defect in the gene associated with the cytochrome P450 isoenzyme, CYP2C19. Two genetic defects, m1 and m2, have been identified: the former accounts for 75–83% of the defective alleles in both white and Japanese PMs, while the latter was found only in Japanese. 4 We determined the distribution of the two CYP2C19 mutations in two Vanuatu islands. In March, 1996, malariometric surveys were conducted on Tanna and Malakula islands. The survey included finger prick sampling of blood for PCR from a capillary tube (75 µL) on to a filter paper. Dried filter-paper samples were collected from 493 people. DNA was extracted from a quarter of a dried blood spot. PCR was conducted as described by de Morais et al, 4 with PCR amplification of exon 5 followed by Sma1 digestion (CYP2C19m1) and amplification of exon 4 followed by BamH1 digestion (CYP2C19m2). The genotypes of the 493 villagers are shown in the table. Remarkably high frequencies of the two mutations were found. The overall frequency of the m1 alleles was 0·708 (698/986), and that of the m2 alleles was 0·133 (131/986). Only 145 individuals had at least one wild-type allele (wt). The observed genotype distribution corresponded well with those estimated from the allele frequencies of CYP2C19m1 and m2 in the study group, in accordance with a Hardy-Weinberg equilibrium (␹ 2-test, p>0·5, power >99%). The population of Tanna Island showed higher frequency of m1 and lower frequency of m2 than that of Malakula Island (p<0·05). In a separate study we correlated proguanil and cycloguanil concentration profiles in whole blood with genotypes in patients with malaria from the same area (unpublished). The results confirm that the genotyping predicted the phenotypes in all 20 patients tested. Thus, the data in the table suggest that 348 (70·6%) individuals have PM phenotype, which may have major implications for the efficacy of proguanil in this population. CYP2C19 is also involved in the metabolism of other drugs such as imipramine, omeprazole, and diazepam. 3 Anthropological evidences suggest that Melanesians are of Mongoloid origin, and the ancestors of the people in Vanuatu may have migrated from Papua New Guinea about 4000 years ago. 5 Therefore, the finding of m2 mutation in Vanuatu is not surprising. However the reasons for the high frequency …

Association Study of Common Genetic Variants and HIV-1 Acquisition in 6,300 Infected Cases and 7,200 Controls

Multiple genome-wide association studies (GWAS) have been performed in HIV-1 infected individuals, identifying common genetic influences on viral control and disease course. Similarly, common genetic correlates of acquisition of HIV-1 after exposure have been interrogated using GWAS, although in generally small samples. Under the auspices of the International Collaboration for the Genomics of HIV, we have combined the genome-wide single nucleotide polymorphism (SNP) data collected by 25 cohorts, studies, or institutions on HIV-1 infected individuals and compared them to carefully matched population-level data sets (a list of all collaborators appears in Note S1 in Text S1). After imputation using the 1,000 Genomes Project reference panel, we tested approximately 8 million common DNA variants (SNPs and indels) for association with HIV-1 acquisition in 6,334 infected patients and 7,247 population samples of European ancestry. Initial association testing identified the SNP rs4418214, the C allele of which is known to tag the HLA-B*57:01 and B*27:05 alleles, as genome-wide significant (p = 3.6×10−11). However, restricting analysis to individuals with a known date of seroconversion suggested that this association was due to the frailty bias in studies of lethal diseases. Further analyses including testing recessive genetic models, testing for bulk effects of non-genome-wide significant variants, stratifying by sexual or parenteral transmission risk and testing previously reported associations showed no evidence for genetic influence on HIV-1 acquisition (with the exception of CCR5Δ32 homozygosity). Thus, these data suggest that genetic influences on HIV acquisition are either rare or have smaller effects than can be detected by this sample size.

Outcome of Simian-Human Immunodeficiency Virus Strain 89.6p Challenge following Vaccination of Rhesus Macaques with Human Immunodeficiency Virus Tat Protein

ABSTRACT The regulatory proteins Nef, Rev, and Tat of human immunodeficiency virus type 1 (HIV-1) are attractive targets for vaccine development, since induction of effective immune responses targeting these early proteins may best control virus replication. Here we investigated whether vaccination with biologically active Tat or inactive Tat toxoid derived from HIV-1IIIB and simian-human immunodeficiency virus (SHIV) strain 89.6p would induce protective immunity in rhesus macaques. Vaccination induced high titers of anti-Tat immunoglobulin G in all immunized animals by week 7, but titers were somewhat lower in the 89.6p Tat group. Dominant B-cell epitopes mapped to the amino terminus, the basic domain, and the carboxy-terminal region. Tat-specific T-helper responses were detected in 50% of immunized animals. T-cell epitopes appeared to map within amino acids (aa) 1 to 24 and aa 37 to 66. In addition, Tat-specific gamma interferon responses were detected in CD4+ and/or CD8+ T lymphocytes in 11 of 16 immunized animals on the day of challenge. However, all animals became infected upon intravenous challenge with 30 50% minimal infective doses of SHIV 89.6p, and there were no significant differences in viral loads or CD4+ T-cell counts between immunized and control animals. Thus, vaccination with HIV-1IIIB or SHIV 89.6p Tat or with Tat toxoid preparations failed to confer protection against SHIV 89.6p infection despite robust Tat-specific humoral and cellular immune responses in some animals. Given its apparent immunogenicity, Tat may be more effective as a component of a cocktail vaccine in combination with other regulatory and/or structural proteins of HIV-1.