White blood and red blood indices of all participants in the DDS cohort were determined using a SYSMEX XT-2000i machine. For the UGR cohort, an 8.5-ml vacutainer was used to collect the venous blood while a 6-ml EDTA bottle was used to collect whole blood. The collected blood samples were stored at a temperature of 4–8°C. For hematological analysis, a swing bucket centrifuge was used to centrifuge the samples at 1,000–13,000 RCF (g) for 10 min.

Genotyping and quality control techniques have been described for each population previously (Gurdasani et al., 2019). Briefly, the DDS samples were genotyped on Illumina HumanOmni Multi-Ethnic GWAS/Exome Array employing the Infinium Assay. Illumina GenCall algorithm was used for genotype calling. The 5,000 GPC samples were genotyped on the Illumina HumanOmni 2.5M BeadChip array (Table 1). Quality control of the DDS cohort took into consideration the following criteria: exclusion of SNPs with heterozygosity > 4 SD from the mean, called proportion <97%, and sex check fails (F statistic >0.2 for women and <0.8 for men). Likewise, SNP QC ensured that called proportion was <−97%, relatedness (IBD >0.90), and Hardy–Weinberg disequilibrium (p < 10⁻⁶). Imputation was done on pre-phased data with IMPUTE2 using a merged reference label of the whole genome sequence data from the African Genome Variation Project. Affymetrix Axiom PANAFR SNP array was used to genotype the AADM data as described previously (Adeyemo et al., 2015). Genotyping, imputation, and quality control of the African ancestry cohort of the BCX2 has been described somewhere else (Chen et al., 2020a).

Prior to metanalysis, all summary statistics data were manually checked for integrity and accuracy (i.e., summary statistics downloaded have the required variables and are appropriately labeled). Some quality control measures were applied to the summary statistics, SNPs in each cohort having MAF >0.05 were selected. The SNP association p values from the summary statistics were meta-analyzed with the aid of GWAMA (Genome-Wide Association Meta-Analysis) (Morris, 2010). We further applied genomic control, and Manhattan plots and quantile–quantile plots were plotted for the meta-analyzed result.

Following our result output from meta-analysis, we used fine-mapping analysis to pick out possible causal SNPs for the locus ±250 kb of all the lead SNPs, using a Bayesian approach (Maller et al., 2012). The Z score was used to calculate the Bayes factor for each SNP denoted as B F i , given by B F i = e [ Z ∗ Z − log ( K ) 2 ] Where K is the number of studies. The posterior probability of driving the association for each SNP was computed by P o s t e r i o r   p r o b a b i l i t y     =           B F i ∑ j B F j Where the summation in the denominator is over all SNPs at the locus.

Ninety-nine percent credible set sizes were derived by sorting all the SNPs according to their posterior probability B F i at the locus from the highest to the lowest, and then counting the number of SNPs needed to attain a cumulative posterior probability that is greater than or equal to 0.99. Index SNPs accounting for more than 50% posterior probability of driving the WBC association at a given signal were defined as high confidence.

Chen et al. (2020a) had previously performed a trans-ethnic and ancestry-specific GWAS of blood cell traits using 746,667 individuals (Chen et al., 2020b). To determine if the variants derived from our study are novel, and perhaps identify novel variants, we checked if any of our loci was reported or fall within ±250 kb window of those identified by Chen et al. A locus is further defined as ±250 kb around the significant SNPs (p < 5 × 10⁻⁰⁹).

We performed a two-sample Mendelian randomization analysis using the R-based MR package Mendelian Randomization (Yavorska and Burgess, 2017). To identify independent genetic instruments, we used significant SNPs (p < 5 × 10⁻⁰⁹) in the summary statistics of Lym, Mon, Neu, and WBCc derived from the metanalysis of APCDR and BCX2 summary statistics. We ensured that the instruments selected were not in LD with each other so that their impact on the exposure and outcome are uncorrelated. This was achieved by using r ² < 0.001 and a 250-kb clumping upstream and downstream of the lead SNPs. We used asthma as our outcome phenotype. These data were selected from the Consortium on Asthma among African Ancestry Populations (CAAPA; 7,009 cases and 7,645 controls) (Daya et al., 2019). Causal estimates were calculated based on IVW and sensitivity analysis was carried out using MR-Egger and median-weighted methods. For each trait, we used Q-statistics to account for heterogeneity in the instruments and also excluded SNPs that may show pleiotropy. Proxy SNP for each missing SNP was obtained from LDProxy (Machiela and Chanock, 2015).

Functional analysis of SNPs identified by the metanalysis was carried out using FUMA (Watanabe et al., 2017). Independent SNPs in linkage disequilibrium or within the same genomic location with the sentinel SNPs were separated. A p-value cutoff of p < 5 × 10⁻⁰⁹ and 1,000 G Phase3 AFR reference panel were used. We carried out other functional analysis such as eQTL tissue expression, pathway enrichment analysis, and biological process to get more insights into the functionality of the loci identified. eQTL mapping was performed by mapping SNPs to genes up to 1 Mb and using the Blood eQTL. We used GeneCards (Stelzer et al., 2016) to determine the functions of the gene. GWAS catalogue and Open Targets were used to identify any previously associated phenotypes of the lead SNPs. Annotation of all the genes identified in this study was done using NCBI’s Genome data viewer. Pathway analysis of the identified loci was performed using Enrichr; Enrichr is an integrative web-based server that facilitates the visualization of the functional characteristics of a gene set. Enrichr is available online at http://amp.pharm.mssm.edu/Enrichr.

The total samples analyzed for each WBC trait are shown in Table 1. WBCc is the trait with the highest number of individuals and Baso has the lowest samples analyzed.

For each of the WBC trait subtypes: Lym: 13 SNPs, Mon: 680 SNPs, Neu: 5,308 SNPs, and WBCc: 4,462 SNPs attained genome-wide significance (p < 5 × 10⁻⁰⁹) (Figure 1). No SNPs in the Baso and Eos were significant at p < 5 × 10⁻⁰⁹. Using a genomic distance of ±250 kb, we identified 4, 34, 124, and 108 lead SNPs within different loci in Lym, Mon, Neu, and WBCc, respectively, at p < 5 × 10⁻⁰⁹ [Supplementary Table S1 (ST1–ST4)]. We examined if the sentinel SNPs were within ±250 kb of SNPs reported in the Chen et al. study. When compared to this, all the SNPs in Lym fall with ±250 kb of those previously reported, while only eight SNPs in Mon, 38 SNPs in Neu, and 13 SNPs in WBCc were unique. However, 4 SNPs out of the 8 SNPs unique in Mon, 3 out of 38 in Neu, and 1 out of 13 in WBCc have been previously identified to be associated with WBC traits when queried on the GWAS catalogue (Buniello et al., 2019) and Open Targets (Koscielny et al., 2017). Two SNPs (rs1103700 and rs6693634 mapped near RPS10P8/CD1A and UHMK1/UQCRBP2) are common in Mon and Neu.

Fine mapping seeks to analyze a trait-associated region to determine variants that are causal to a trait of interest. Fine mapping of loci was carried out within 250 kb upstream and downstream genomic distance of the sentinel variant identified via metanalysis. For each locus fine-mapped, we established the 99% credible set of SNPs that jointly make 99% of the posterior probability of driving the association (Figure 2). The fine-mapping analysis revealed that some of the lead variants such as rs369124352 (MAGI3), rs10918211 (LRRC52), and rs10800292 (LINCO1363/POU2F1) accounted for more than 50% of the posterior probability driving the association with these SNPs as the only variant within the 99% credible set. Several other variants were identified as the causal SNPs other than the sentinel SNP driving the association in WBC trait subtypes. Summary of the 99% credible set of variants driving the WBC trait can be found in Supplementary Table S2. We further went ahead and checked if these causal SNPs and their corresponding genes have been previously reported using the GWAS catalogue and Open Target; only five variants—rs11184898 (LOC126987, MTCO3P14) (Figure 3), rs907662 (LINC01525), rs7553527 (GAPDHP32, HSD3BP3), rs1923504 (FLG-AS1, HMGN3P1), and rs10733045 (TRK-CTT13-1, MGST3)—have not been reported to be associated with any WBC trait.

We set out to evaluate the causal relationship between WBC traits (Lym, Mon, Neu, and WBCc) and asthma. Our MR analysis identified strong positive association of Mon (IVW estimate = 0.38, CI: 0.221, 0.539, p < 0.001), Neu (IVW estimate = 0.189, CI: 0.133, 0.245, p < 0.001), and WBCc (IVW estimate = 0.185, CI: 0.108, 0.262, p < 0.001) with increased risk of asthma. However, there were no evidence of causal relationship between Lym and asthma risk (IVW estimate = −0.079, CI: −0.779, 0.621, p = 0.825) (Figure 4, Supplementary Table S3), though there was an inverse relationship between Lym and asthma risk. The causal estimates of the weighted median and the MR-egger methods of Mon showed a similar effect to the IVW method (Supplementary Table S3).

MAGMA analysis using GTEx v8: 54 tissue types and GTEx v8: 54 general tissue types showed significant expression in whole blood (Figure 5E). The loci were also expressed in other tissues such as stomach and the colon but not at a significant level. Lipid and atherosclerosis, nitrogen metabolism, and FoxO signaling pathway are some of the pathways enriched by these loci [Figure 5A, Supplementary Table S4 (ST1)]. These loci were mapped at a significant level to thrombocytopenia-absent radius syndrome, autoimmune lymphoproliferative syndrome, lactose intolerance, central core myopathy etc. [Figure 5B, Supplementary Table S4 (ST2)]. They are also involved in the regulation of cellular response to transforming growth factor beta stimuli, regulation of transmembrane receptor protein serine/threonine kinase signaling pathway, regulation of cellular biosynthetic process, etc. (Figure 5C), Supplementary Table S4 (ST3)] and also in exogenous lipid antigen binding molecular functions (Figure 5D).