Introduction

The WBC count, a classic marker of immune or inflammatory response, varies substantially among healthy individuals. The counts of constituent cell subtypes comprising the WBC count measure are assayed as part of a standard clinical WBC differential test. While the WBC count and WBC differential count are often obtained to assess for evidence of infection or underlying inflammation, prospective epidemiologic studies have consistently linked higher WBC counts, within the clinically designated normal range, along with other inflammatory markers, to increased risk of coronary artery disease, cancer, and total mortality [1], [2], [3], [4], [5], [6], [7], [8], [9], [10]. Studies are often not consistent on the specific WBC subpopulations involved, but granulocytes, in general, and neutrophils, in particular, are most often implicated in these observations [11], [12]. In the general population, the total WBC count is also directly associated with many cardiovascular disease risk factors, such as higher blood pressure, cigarette smoking, adiposity, lower socioeconomic status, and higher levels of plasma inflammatory markers [13].

WBC counts are also moderately heritable [14], with heritability estimates varing from approximately 0.14 to 0.4 across total leukocyte count and cell subtypes, as assessed in a Sardinian population, with the highest heritability estimates for monocyte counts [14]. In addition, the substantially lower neutrophil count and total WBC count in African Americans compared to European-ancestry individuals seems to be at least partially explained by a regulatory variant in the Duffy Antigen Receptor for Chemokine (DARC) gene, which accounts for ∼20% of total variation in the measures [15], [16]. Recent studies have sought to investigate the common genetic variants associated with several blood count traits in European-ancestry and Japanese individuals, but have not focused specifically on the multiple cell types comprising the total WBC count measurement [17], [18], [19], [20].

In the current study, we sought to identify and replicate common genetic variants that influence normal variation in six WBC phenotypes commonly measured in the clinical setting and in population studies that comprise the CHARGE Consortium [21], the HaemGen Consortium [18] and independent collaborative studies. We utilized genome-wide association study (GWAS) data and meta-analytic techniques to identify 10 genome-wide significant loci in a study of over 31,000 individuals (Table 1), examining variants possibly associated with total WBC count, three granulocyte phenotypes (neutrophil, basophil and eosinophil counts), and two non-granulocyte phenotypes (lymphocytes and monocytes). We also investigated the shared functional connectivity of identified loci across phenotypes with regards to both known biological pathways and nearby gene expression effects. As previous research has shown strong selective pressures at the locus identified to affect WBC counts in African American populaftions, we examined the possibility of recent selective effects at significantly associated loci in European ancestry individuals [15], [16]. Additional efforts were made in collaboration with RIKEN and COGENT investigators to identify homogenous associations across populations of diverse ancestral backgrounds.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. Descriptive statistics of contributing cohorts.

https://doi.org/10.1371/journal.pgen.1002113.t001

Results

In the discovery meta-analysis of 19,509 subjects from seven cohorts, we identified 11 genome-wide significant associations with six white cell phenotypes (total WBC, neutrophil, basophil, eosinophil, lymphocyte and monocyte counts, see Table 1, Table S1, Figure 1). Further, we found strong evidence for replication of 10 of the 11 trait-locus associations in 11,823 independent samples from 10 GWAS cohorts who contributed summary statistics for SNPs of interest. The discovery analysis results (Figure 1 and Figures S1, S2, S3, S4, S5, S6, S7, S8, S9, S10 for details), and the results of replication testing for the 10 replicated trait-loci are summarized in Table 2. These results are presented in greater detail for all genome-wide significant SNPs in Table S2.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 1. Results of discovery phase meta-analyses for white blood cell phenotypes.

Red regions denote loci containing SNPs reaching genome-wide significance at p-values<5.00E-08. Blue regions denote suggestive loci containing SNPs with p-values between 5.00E-08 and 1.00E-07.

https://doi.org/10.1371/journal.pgen.1002113.g001

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 2. Summary results from discovery and replication phase analyses for all genome-wide significant and successfully replicated loci.

https://doi.org/10.1371/journal.pgen.1002113.t002

Total WBC count was associated with two independent loci in the discovery phase of analyses; the first locus was on chromosome 6p21 encompassing a region from 31,131,127 bp to 31,161,846 bp near HLA and PSORS1 gene families, and the second locus on chromosome 17q21 from 35,345,186 bp to 35,470,048 bp near candidate genes ORMDL3 and CSF3. Both loci showed independent replication (Table 2). Neutrophil count was associated with a 196,381 bp region on chromosme 17q21 containing 46 genome-wide significant SNPs from the discovery phase. This region overlaps the locus on chromosome 17q21 identified for the total WBC count phenotype and showed positive association in replication testing at all but 2 SNPs. Basophil count was associated with one SNP, rs4328821 on chromosome 3q21 near RPN1 and C3orf27. This SNP showed a significant positive association between basophil count and minor allele dosage (minor allele frequency 0.110, p-value in discovery phase 2.58E-08, p-value in replication phase 8.40E-06). No regions showed genome-wide significance for association with eosinophil count.

Lymphocyte count was associated with two loci, with one locus on chromosome 6p21 overlapping with the chromosome 6p21 total WBC count locus. The second locus associated with lymphocyte counts is on chromosome 19p13, from 16,32,871 bp to 16,429,197 bp at EPS15L1 and was successfully replicated. Monocyte count was associated with the largest number of independent hits for any of the traits, with five loci identified in the discovery analysis, four of which showed significant associations in replication testing. These four loci include two intergenic regions (>100 kb to nearest known genes) on chromosome 8q24 (from 130,672,817 to 130,693,287) and chromosome 9q31 (112,917,232 to 113,073,157). We also identified a novel association on chromosome 2q31 (182,027,546 to 182,036,459) near ITGA4, and a single genome-wide significant SNP on chromosome 3q21, which is located 18,866 bp from the SNP significantly associated with basophil count (r² = 0.076, D′ = 0.841). The monocyte- associated locus on chromosome 1q22 contained only one genome-wide significant SNP which failed to replicate and is not included in Table 2 (denoted by rs17131683, which exhibited a replication p-value of 0.770 but a consistent negative effect associated with the A allele).

Many of the loci described in this report contain genome-wide significant SNPs spread throughout much of each locus suggestive of either extensive LD or multiple association signals at each locus. For loci described in Table S2 (except those containing less than 3 genome-wide significant SNPs), conditional analyses were conducted using the allele dosage of the most significant SNP per locus as a covariate in a subset of discovery cohorts (AGES, ARIC, BLSA, Health ABC and InChianti). Statistical models were identical to those used in the discovery analyses except for the additional SNP covariates. No SNPs analyzed remained significant after correcting for 147 tests, showing that only one signal of association exists per locus. The complete results for these analyses are evident in Table S3. As each locus accounts for only one unique signal per trait, adjusted r² estimates were calculated for each trait across loci within this subset of cohorts, and may be found in Table 2.

To assess the independence of associated SNPs from the total WBC count measurement, all genome-wide significant SNP associations for white cell subtypes (Table S2) were re-analyzed as per the discovery phase analysis methods adjusting for total WBC count as an additional covariate in a subset of discovery cohorts (AGES, ARIC, BLSA, Health ABC and InChianti). After Bonferroni correction for 97 independent tests at least one SNP per locus remained significant, suggesting some independence from the total WBC measure in the SNP associations. The extended results of this analysis and a table of r² values for the phenotypes of interest based on the same subset of discovery cohorts may be found in Table S4 and Table S5.

These results demonstrate a high degree of relatedness across traits, with individual loci affecting multiple WBC traits, that may be pleiotropic to some degree or due to the biological relatedness of the traits. Neutrophils are the most abundant WBC subtype, and the locus on chromosome 17q21 associated with both total WBC count and neutrophil count independently based on conditional analyses described above. In this region, 38 SNPs were common to both traits in the discovery analysis, and these SNPs showed identical directions of effects across both traits. In the chromosome 3q21 between C3orf27 and RPN1, rs9880192 was associated with monocyte count and rs4328821 was associated with basophil count, suggesting pleiotropy in this region. The region on chromosome 6p21 near the PSORS1 family of genes as well as HLA-C and HLA-B contains associated SNPs with both total WBC count and lymphocyte count, although, interestingly, spatially overlapping SNPs failed to replicate in the other trait, further suggesting independence of effects seen in conditional analyses of all loci.

To further examine the functional connectivity of these loci across white blood cell phenotypes, the Gene Relationships Among Implicated Loci package (GRAIL, http://www.broad.mit.edu/mpg/grail/) was utilized to mine PubMed archives using textual analyses of known functional associations to identify concurrent effects and relationships across phenotypes [22]. In brief, GRAIL incorporates functional annotations from text mining related to specific genomic loci, usually genome-wide association study results, to assess the functional inter-relatedness of genes in linkage disequilibrium with the regions of interest and construct networks of related genes sharing biological function. In our analyses, we utilized GRAIL to survey the functional relatedness of all regions containing significant results passing Bonferroni correction in both the discovery and replication phases. We identified four clusters of related genes with false-discovery rate adjusted p-values<0.05 out of the 49 gene clusters generated by the GRAIL package, which are described in Figure 2 and Table S6. All four clusters show significant interconnectivity between genes proximal to loci on chromosome 17q21 associated with total WBC and neutrophil counts and the chromosome 19p13 locus associated with lymphocyte count. The two most significant clusters also show relationships between genes proximal to the previously mentioned loci and candidate genes at the chromosome 8q24 region associated with monocyte counts. Genes at the chromosome 17q21 locus associated with both total WBC and neutrophil counts appeared in all significant pathways identified in the GRAIL analyses, suggesting biological connectivity across both granulocyte and non-granulocyte cell lineages. Candidate genes from the gasdermin (GSDML) and mediator complex subunit (MED) families were highly enriched in the significant gene clusters and comprised 37.5% of genes in these clusters. These results suggest shared biological pathways between these genes and cell proliferation among WBC subtypes associated with these genomic regions, although as emphasized in conditional analyses, these effects at these loci remain to some degree independent of the total WBC measure.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 2. GRAIL gene clusters represented graphically for all clusters reaching a FDR adjusted p-value <0.05.

The relative thickness of each line connecting nodes is indicative of a p-value closer to zero. Gene symbols are designated on each node, with the cytogenetic position following the colon. Individual gene clusters are color-coded.

https://doi.org/10.1371/journal.pgen.1002113.g002

We also examined possible functional consequences of individual SNP associations by analyzing whole blood genome-wide gene expression data from the InChianti study to identify associations between SNPs found to be significant in the meta-analysis and cis changes in gene expression. All SNPs significant in both the discovery and replication phases for all phenotypes were used in the expression analysis. Each SNP in this dataset which was within 500 kb of an expression probe was treated as a possible expression quantitative trait locus (eQTL). For 85 SNPs in our subset of significant meta-analysis results, we tested at total 741 candidate eQTL associations using multivariate linear regression. This analysis tested each SNP in the subset for an association with each expression probe within 500 kb. After Bonferroni correction for the 741 tests conducted, 36 SNPs in the chromosome 17q21 locus associated with total WBC count and neutrophil count in the meta-analysis were also significantly associated with cis-expression levels. In total, these 36 SNPs were associated with three expression probes, 2 probes tagging transcripts in the GSDML gene (probes ILMN_2347193 and ILMN_1666206) and another probe tagging a single probe in ORMD3L (probe ILMN_1662174), for a total of 103 signficicant eQTLs (Figure 3). Both probes in GSDML were highly correlated (r² = 0.582), although neither GSDML probe was strongly corellated with the probe of interest in the ORMD3L gene (r²<0.200). With each SNP's minor allele as the point of reference, all effect directions for significant meta-analysis and eQTL associations were concordant, showing a strong correlation between the effect sizes in the meta-analysis and eQTL analysis, suggesting that the effect of the identified SNPs on the corresponding WBC trait may be transcriptionally mediated. For example, the correlation of effect estimates between the eQTL and meta-analysis for neutrophil associated SNPs also associated with the ILMN_1666206 probe highly significant, with an r² of 0.898. For details of all eQTLs tested, please refer to Table S7 and the Methods and Materials section.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 3. The directionality and magnitude of effects for SNPs significant in both the meta-analysis and eQTL analysis are highly correlated (r² = 0.795).

Effect estimates in both the eQTL analysis and meta-analysis were standardized to effects based on dosages of the minor allele for each SNP. Meta-analysis effects are based on beta coefficients from regression, while eQTL effects are based on standardized betas. This figure only includes significant SNPs to improve clarity.

https://doi.org/10.1371/journal.pgen.1002113.g003

Previous studies of WBC genetic variation in African American populations have shown evidence of WBC associated loci in a region of high selective pressures, with the putative functional SNP rs2814778 being almost fixed in frequency in sub-Saharan African populations where malaria is endemic [15]. We therefore undertook an investigation of recent selective pressures acting upon SNPs identified associated with WBC phenotypes in European-ancestry populations. Evidence for selection for all 10 significant trait-loci from the meta-analysis was evaluated by mining HapMap2 CEU data. Integrated Haplotype Scores (iHS) were used to quantify selection at each locus based on homozygosity decay in extended haplotypes and are availible for download from the Haplotter website (http://hg-wen.uchicago.edu/selection) [23]. Selection was quantified by an absolute value of iHS>2 indicating strong selective pressures, and absolute iHS values ≤2 and ≥1.635 were interpreted as indicating recent-moderate selection, positive or negative iHS values indicated the direction of selection [24]. All replicated SNPs were evaluated for signatures of selection. One locus showed evidence of selection, and this was the lymphocyte associated locus on chromosome 19p13, from 16,336,388–16,429,197 bp. This locus on chromosome 19p13 showed evidence of selection for all SNPs analyzed in the replication phase. All SNPs in this locus appear to be under some degree of negative selection with 2 of the 11 SNPs being under strong negative selection and the rest under recent moderate selection. However, when testing the correlation between effect estimates and iHS at this locus using the ancestral allele as a reference for effect direction, no clear correlation between iHS and effect size was detected (r² = 0.226, p-value = 0.140). Full iHS annotation for replicated SNPs are shown in in Table S8.

As part of collaborative efforts with the COGENT and RIKEN groups, a coordinated exchange of summary statistics for SNP identified in Table 2 was organized from the parallel studies conducted by these groups. Random-effects meta-analysis techniques were utilized to identify effects that were consistent across studies of diverese ethnic backgrounds. While all joint effect estimates remained in consistent directions with those described in Table 2, heterogeneity of effects across the 3 ancestral populatons severely attenuated the strength of the associations for all but 2 of the genome-wide significant associations identified in this report. The associations at rs4794822 (WBC count) and rs11878602 (lymphocyte count) remained genome-wide significant. Consistent robust effects across ancestral populations at rs11878602 may lend some support to the recent-moderate negative selection at this SNP (iHS = −1.924) being associated with an increase in frequency of the derived A allele associated with decreasing monocyte counts. These results suggest that these SNPs may be very close to the functional variants associated with these effects, as well as exhibiting relatively consistent effects across multiple population ancestries with differing LD structures. The results of these analyses are detailed in Table 3.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 3. Results of random-effects meta-analyses incorperating summary statistics from COGENT, RIKEN, and CHARGE meta-analyses for loci of interest.

https://doi.org/10.1371/journal.pgen.1002113.t003

Discussion

This meta-analysis has identified ten genome-wide significant trait-locus associations with WBC phenotypes spread across seven genomic loci. Of these trait-locus associations, seven associations represent novel findings, and three associations, across two genomic loci, confirm previously identified associations. The association of chromosome 17q21 near ORMDL3 and CSF3 associated in our study with total WBC count and neutrophil count, and the 9q31 locus associated with monocyte count have been previously demonstrated in European-ancestry and Japanese populations [25], [26]. The chromosome 3q21 locus near RPN1 and C3orf27 has been previously shown to be associated with eosinophil count, and is instead associated with related granulocyte cell measures of monocyte and basophil counts in this study, although we do identify suggestive p-values at ∼1.00E-4 to 1.00E-8 and the same direction of effect for the additional loci identified in Gubjartsson et al., 2010 [19], [27]. In addition, a number of previous GWAS have implicated the monocyte count associated locus at chromosome 8q24.21 as affecting height, renal function, serum protein levels, multiple sclerosis, glioma, leukemia and a number of other cancers [28]–[59]. Through conditional analyses and an analysis of functional relatedness, we have shown correlation among related traits and possible pleiotropic connectivity of these loci across phenotypes that represent measures of biologically related cellular lineages, as well as identified loci showing a direct association between allelic gene expression differences and variation in phenotypic measures.

The associations identified in this study are robust, and several have been previously identified in GWAS studies of immunologically relevant phenotypes, such as the association of celiac disease with the chromosome 2q31.1 locus containing ITGA4. ITGA4 encodes an alpha integrin subunit present on monocytes, lymphocytes, endothelial cells and erythrocytes that serves as an adhesion molecular receptor for VCAM-1, fibronectin. VCAM-1 is expressed at high levels on the vasculature of the bone marrow, and therefore alpha4integrin receptors may play a role in homing and recruitment of certain cell types during hematopoiesis [60], [61].

The overlapping loci on chromosome 17q21 associated with both total WBC and neutrophil counts constitute a single plausible candidate locus contributing to general variability in total WBC count and neutrophil count via hematopoietic mechanisms. These measures are highly correlated, and the estimated SNP effects are also likely correlated for this reason. From a functional perspective, the role of G-CSF, the CSF3 gene product, has been well-described in the biology of myeloid progenitor production and differentiation. This locus previously reached genome-wide significance in a joint analysis of discovery and replication cohorts of total WBC count in individuals of European ancestry [18]. The same locus, containing the genes PSMD3 and CSF3, was associated with neutrophil count in a cohort of Japanese participants [27]. This study of Japanese subjects also reported a significant genome-wide association with neutrophil count for three SNPs in a locus at chromosome 20, containing the gene PLCB4. Due to the lower minor allele frequency in European ancestry individuals of these three correlated SNPs (minor allele frequency = 0.076 in HapMap CEU samples versus 0.289 in HapMap JPT samples for rs2072910) and the marginal effect size detected in the original report, statistical power to detect this association was limited in our analysis [27].

Our data suggest that the locus on chromosome 17q21 has functional connectivity across white cell subtypes. Multiple genes at this locus appeared in all significant pathways identified in the GRAIL analyses showing a functional connectivity across both granulocyte and non-granulocyte cell lineages. Gene clusters detailed in the GRAIL analyses show significant functional clusters including genomic regions that are separately associated with granulocyte and non-granulocyte traits within the same cluster. However, the results of the GRAIL analyses may be influenced by both funding avenues and publication bias, as the classifications are based only on PubMed searchable published results.

Cis-eQTL effects on transcripts in the ORMDL3 and GSDML were shown to be highly correlated with allelic effects associated with neutrophil and total WBC count measures quantified in the meta-analysis. This suggests that variation in total WBC count and neutrophil counts are at least in part due to polymorphism-based regulation of gene expression. This regulation of gene expression by SNPs in this region may be related to some form of systemic immunological function, as variants in this region were also shown to be associated with expression of ORMDL3 in childhood asthma [62]. Interestingly, Okada et al. also showed significant eQTL associations between SNPs and transcripts at this locus, although their associations implicate regulation of PSMD3 as affecting neutrophil variation, rather than the exact transcript associations identified in this report [27]. This slight difference may be attributable to the larger sample size in our study, differences in population ancestry of the two studies, and the use of expression data derived from lymphoblastoid cell lines instead of whole blood. The CSF3 gene at this locus is a possible candidate contributing to effects across multiple cell subtypes at this locus, as functional studies have demonstrated that this gene creates a protein integral to the differentiation and functionality of granulocyte cells [63], [64], [65], [66]. One important caveat to our eQTL analysis is the lack of representation of the CSF3 gene on the expression array used.

The loci on chromosome 6p21 associated with total WBC and lymphocyte counts appear to be independent of each other as the lymphocyte association persists after adjustment for total WBC. While the closest replicated SNP associations are roughly 200 kb from each other, and a shared functional connectivity between these regions was not elucidated in the GRAIL analysis. At this locus, rs2524079 associated with lymphocyte count is in moderate LD with a number of SNPs in the periphery of the total WBC count locus, including rs2844619, a SNP significant in the total WBC count discovery phase analysis (D′ = 0.762, r² = 0.305, from HapMap Phase II CEU samples). The finding of multiple independent effects at a single locus has occurred in prior studies and include examples such as the finding of two independent signals within the PLAG1 locus associated with human height, suggesting a locus specific effect, in the current example affecting leucopoiesis or leukocyte homeostasis [48]. Our chromosome 6p21 WBC and lymphocyte loci, both harbor candidate genes that have been previously implicated as associated with phenotypes closely related to immunological function. The locus associated with total WBC count on chromosome 6p21 contains genes associated with follicular lymphoma (CHCG22), progression of HIV-1 infection (CDSN, PSORS1C1 and PSORS1C2) and psoriasis (CCHCR1, PSORS1C1 and PSORS1C2) [67], [68], [69]. This gene rich region includes HLA family genes, particularly HLA-B and HLA-C that are candidates within the lymphocyte associated locus, and actually overlap with the psoriasis candidate locus identified previously via linkage mapping studies and includes PSORS1C1 and PSORS1C2, showing the relatedness of these two loci on chromosome 6p21 [69], although conditional analyses adjusting for total white blood cell count validate these as primarily independent effects. The lymphocyte associated regions containing HLA-B and HLA-C, harbors two genes that have been implicated in multiple GWAS as modifiers of immunological responses, associated with IL-18 levels, HIV-1 control, vitiligo, multiple clerosis, and psoriasis [49], [68], [70], [71], [72], [73], [74], [75], [76].

The region associated with basophil and monocyte counts on chromosome 3q21 proximal to the GATA2 gene was previously described as associated with variation in eosinophil count in European and Asian ancestry populations [19]. This association with eosinophil count was not identified as genome-wide significant in our analyses, with multiple SNPs in the GATA2 region (GATA2+/−250 kb) approaching genome-wide significance, exhibiting p-values in the discovery meta-analysis including a regional minimum of 6.73E-07 (rs4328821 ). This is likely due to our decreased sample size for analyses of eosinophil count compared to the original report. However, our data show a significant association between this genomic region and basophil count, and basophils and eosinophils are both granulocytic cells with common lineage in WBC differentiation. GATA2 is a well-known transcription factor involved in maintenance of early hematopoietic cell pools and proximal hematopoietic pathways. In addition, this region proximal to GATA2 is also significantly associated with monocyte counts, showing overlapping associations across both granulocyte and non-granulocyte cell lineages and supporting the previously described functional role of GATA2 more proximally in the WBC differentiation process [77], [78]. The independence across granulocyte and non-granulocyte lineages is evident as both associations showed independent signals of association after adjustment for total WBC.

Our analyses have identified genomic loci associated with total WBC and constituent white cell subtypes in European ancestry cohorts. Our findings differ from the results of similar studies of African American and Asian ancestry populations. Population variation at previously identified total WBC count associated loci of DARC in African American cohorts and PLCB4 in Asian ancestry samples motivated our investigation of possible selection at the loci identified in this report, as allele frequencies of SNPs in DARC and PLCB4 differ across populations. This may be suggestive of recent selection. Our analyses of iHS statistics for genome-wide significant SNPs yielded only one locus under selection, with all SNPs investigated in this region being under negative selection in European ancestry populations. These SNPs on chromosome 19p13 associated with lymphocyte counts are proximal to candidate genes such as CHREP, which function in calcium homeostasis in lymphocytes, and mutations in the coding region of CALR3 are associated with familial hypertrophic cardiomyopathy [79], [80]. A thorough search of literature did not reveal any known selective factors associated with this locus. In addition, the fact that this locus remained genome-wide significant in random-effects modeling across diverse ancestral populations suggests a highly generalizable effect at this locus that may or may not be related to selective factors.

In conclusion, we have identified and replicated a set of 10 independent trait-locus associations influencing multiple related WBC traits, of which seven are novel associations. Integrative analyses of our association data and gene expression analyses support pleiotropic effects that will require further functional testing to clarify.

Materials and Methods

Supporting Information

Acknowledgments

The authors acknowledge the essential role of the Cohorts for Heart and Aging Research inGenome Epidemiology (CHARGE) Consortium in development and support of this manuscript. CHARGE members include The Rotterdam Study (RS), Framingham Heart Study (FHS), Cardiovascular Health Study (CHS), the NHLBI's Atherosclerosis Risk in Communities (ARIC) Study, and the Age, Gene/Environment Susceptibility – Reykjavik Study (AGES) Study. The participation of studies representing the HeamGen Consortium aided in verifying the results from CHARGE, as well as further developing this manuscript, and include the Sorbs, the Twins UK cohort (TwinsUK), Kooperative Gesundheitsforschung in der Region Augsburg (KORAF3 & KORAF4), and UK Blood Services Donor Panel 1 (UKBS1) studies. Three of the Italian Network on Genetic Isolates (INGI) studies (Carlantino, Val Borbera, and Friuli Venezia Giulia) made valuable contributions to this research. The collaboration of studies such as the Health Aging and Body Composition Study (Health ABC), the Baltimore Longitudinal Study of Aging (BLSA), the Invecchiare in Chianti Study (InChianti), the Rotterdam Study II (RSII), and the Heart and Vascular Health Study (HVH) also played a vital role. Collaboration with the COGENT and RIKEN consortia facilitated additional research and collaboration enriching this project. We thank Dr. Michael Moorhouse, Pascal Arp, Mila Jhamai, Marijn Verkerk, and Sander Bervoets for their help in creating the genetic database. We thank the laboratory technicians Jeannette M. Vergeer-Drop, Bernadette H. M. van Ast-Copier, Andy A. L. J. van Oosterhout, Sue Ellen Mauricia, Andrea J. M. Vermeij-Verdoold, Els Halbmeijer-van der Plas, Debby M. S. Lont, and Hasna Kariouh for their help in phenotype assessment. We thank the staff from the Genotyping Facilities at the Wellcome Trust Sanger Institute for sample preparation, quality control, and genotyping led by Leena Peltonen and Panos Deloukas; Le Centre National de Geçnotypage, France, led by Mark Lathrop, for genotyping; Duke University, North Carolina, USA, led by David Goldstein, for genotyping; and the Finnish Institute of Molecular Medicine, Finnish Genome Center, University of Helsinki, led by Aarno Palotie. We would like to thank Knut Krohn (Microarray Core Facility of the Interdisciplinary Centre for Clinical Research, University of Leipzig) for the genotyping/analytical support and Joachim Thiery (Institute of Laboratory Medicine, Clinical Chemistry and Molecular Diagnostics, University of Leipzig) for clinical chemistry services. We thank Nigel W. Rayner (WTCHG, University of Oxford, UK) for the excellent bioinformatics support. The authors thank the WHI investigators and staff for their dedication, and the study participants for making the program possible. A listing of WHI investigators can be found at http://www.whiscience.org/publications/WHI_investigators_shortlist.pdf. The authors would also like to acknowledge Dr. Mark R. Cookson and Dr. Bryan J. Traynor for assistance manuscript formatting.