This study includes two unrelated admixed African American ancestry SCD cohorts. Study protocols of both cohorts were approved by the Institutional Review Board (IRB) of Johns Hopkins University and Boston Medical Center. Additionally, IRB approval was acquired from all of the participating sites for subject enrollment and conducted in accordance with institutional guidelines.

The Silent Infarct Transfusion (SIT) Trial is an international, multi-center clinical study funded by the National Institute of Neurological Disorders and Stroke (NINDS) (http://sitstudy.wustl.edu/) [24]. All the participants included in our study are of African American ancestry and written informed consent was obtained from parents of the SCD-affected individuals. For each patient, DNA was collected from Epstein-Barr virus (EBV) transformed lymphoblasts using Puregene Genomic DNA Purification kits (Gentra Systems, Inc). Demographic and phenotypic information were collected for each participant and the inclusion criteria for the recruitment were age (5-15 years) and hemoglobinopathy diagnosis (either Hb SS or Hb SB⁰-thalassemia). Details of the SIT Trial study design are given elsewhere [24].

The Cooperative Study of Sickle Cell Disease (CSSCD) was a multi-institutional prospective longitudinal study of SCD funded by the National Heart, Lung, and Blood Institute (NHLBI) of the National Institutes of Health (NIH) [25]. In our study, we only included CSSCD participants who are of African American ancestry, and to match the samples with the SIT Trial cohort, age inclusion criteria of <15 years was used. Details of the CSSCD study design are given elsewhere [25].

Genotyping of the SIT Trial cohort was performed in two stages. For stage 1, a subset of 573 samples, along with 24 International HapMap Consortium [26] controls and 13 known duplicates, were genotyped at the Center for Inherited Disease Research (CIDR) at Johns Hopkins University using the Illumina HumanHap650Y SNP array (Illumina Inc., San Diego CA, USA). This array contains approximately 661,000 SNPs, of which ~100,000 were selected as tags for populations with African ancestry [27]. The Beadstudio software (Illumina Inc.) was used to cluster the data and samples with <96.5% call rate were re-genotyped. The reproducibility, calculated from duplicate pairs was 99.98% and genotype concordance with HapMap data was 99.76%. In stage 2, 509 samples were genotyped at the Center for Disease Control (CDC) at Washington University using the Illumina Infinium HumanOmni1-Quad SNP array (Illumina Inc.) and achieved the call rate of ≥96%.

For quality control (QC), we performed several rounds of data cleaning and a 96.5% cutoff was used for the sample call rate and SNP coverage in the combined SIT Trial data (n=1082), and resulted the exclusion of 6 individuals and 1,260 SNPs from the study. Cryptic relatedness was determined by examining pair-wise identity-by-descent (IBD), and 77 samples were identified as first-degree relatives and dropped from the study. Given the admixed nature of the study participants, we used principal component analysis (PCA) as implemented in EIGENSTRAT [28] to both identify genetic outliers (>6 standard deviations on any of the top ten principal components) and correct for any potential residual population substructure. In the SIT Trial, twenty-six individuals were identified as genetic outliers and further excluded from the analysis. Additionally, 48 samples, due to incomplete phenotype data, were also dropped from the study, leaving 925 samples (males: 51.9% & females: 48.1%) for the subsequent GWAS analysis. Among them, 89% of the samples (n=826/924) had a confirmed SCI status and 98 individuals (11%) were not classified. In total, 251 SCI positive and 575 SCI negative samples were assigned as cases and controls, respectively.

In the CSSCD cohort, DNA samples were genotyped at Boston University using Illumina Human610-Quad SNP arrays (Illumina, San Diego, CA, USA) with approximately ~600,000 SNPs. All samples were processed according the manufacturer’s protocol and the BeadStudio software was used to make genotype calls utilizing the Illumina pre-defined clusters. Samples with <95% call rate were removed and SNPs with a call rate <97.5% were re-clustered. After re-clustering, SNPs with call rates >97.5%, cluster separation score >0.25, and excess heterozygosity between -0.10 and 0.10 were retained in the analysis. The pair-wise IBD was used to identify cryptic relatedness and PCA was applied to detect genetic outliers. After excluding these samples and following the SIT Trial age inclusion criteria (<15 years), analysis was restricted to a dataset of 692 samples (males: 52% & females: 48%).

To infer un-genotyped SNPs and fill-in missing data in the SIT Trial and CSSCD cohort, HumanHap650Y, HumanOmni1-Quad and Human610-Quad SNP array datasets were merged and subsequently imputation was performed for autosomes using a Hidden Markov model, as implemented in the MaCH software [29] (version 1.16) (http://www.sph.umich.edu/csg/abecasis/MaCH/), with 50 rounds and 200 states. QC was performed both before and after imputation and poorly imputed SNPs (RSq <0.5, squared correlation between imputed and true genotypes) were excluded and total 1,019,297 SNPs were analyzed.

Due to the lack of any published studies that report genetic determinants for SBP in children, and/or more specifically in African American children at genome-wide significance, we used the ICBP identified SBP SNPs for validation. To validate results from the ICBP study in SCD patients, we examined the 28 SNPs that were reported associated with SBP [23]. For SNPs that were not available on our genotyping array, a close proxy for the index SNP with criteria of r² ≥ 0.6 from HapMap Phase III (release 2, ASW panel) [30] or 1000 Genomes project (Pilot 1, YRI panel) [31] was used.

All quality control measures in both cohorts were performed using the PLINK software package [32], version 1.06 (http://pngu.mgh.harvard.edu/purcell/plink/). In the SIT Trial cohort, to account for the uncertainty of the imputed data, the estimated allele dosage was analyzed using ProABEL [33] under a multivariate linear regression framework. Association for each SNP was assessed after adjusting for age, sex, height and the 1^st principal component, and assuming an additive effect of allele dosage on SBP. In the CSSCD cohort, SBP measurements were available for longitudinal time points and data was analyzed using a linear mixed effect model using the lme4 package (http://lme4.r-forge.r-project.org/) in R (http://www.r-project.org/) (version 2.14.1). In the mixed effect model, age, sex, height and the 1^st principal component were used as fixed effect covariates, while multiple SBP measurements within each individual were treated as random effect. GWAS results from both the cohorts were meta-analyzed using inverse-variance weighted fixed-effect models as implemented in METAL (http://www.sph.umich.edu/csg/abecasis/metal) [34]. The variance inflation factor for genomic control (λ_GC), as described by Devlin and Roeder [35], was evaluated in each cohort prior to meta-analysis and a total of 1,019,297 SNPs were meta-analyzed. To explore the previously published SBP associated SNPs [23], a one-sided test of significance was used. To estimate the effect of these SNPs on SCI (data available only from the SIT Trial cohort), multivariate logistic regression was used after adjustments for age, sex and 1^st principal component. The genetic risk scores for SBP and SCI were constructed (based on previously published SBP variants and weighted according to their effect sizes) using an R package Genetics ToolboX (http://cran.r-project.org/web/packages/gtx/index.html). To construct the genetic risk scores, this R package uses the same underlying statistics which was used by the ICBP [23] and can be defined as follows: Assuming a set of m SNPs from a discovery panel, for the i-th SNP in the j-th individual denotes x_ij as the coded genotype (for directly genotyped SNPs) or estimated allele dosage (in case of imputation). If the set of regression coefficients of the reported SNPs are w₁, w₂. . w_m, then the risk score for individual j is defined as: s_{j =} s_o + w₁ x_{1j +} w₂ x_{2j + … +} w_m x_mj, where s_o is the intercept. In our analyses, we specify the coefficient w₁, w₂. . w_m, to be the effect sizes (in mmHg per coded allele). Further, to identify known functional regulatory variants within or in proximity to the loci of interest, the GTEx (Genotype-Tissue Expression) expression quantitative trait loci (eQTL) database was queried (http://www.ncbi.nlm.nih.gov/gtex/GTEX2/gtex.cgi) [36].

To increase power by combining independent associations within a gene into a single, stronger aggregated signal, gene-based association tests were performed using GWiS [37]. GWiS uses greedy Bayesian model selection (selecting a minimal subset of associated SNPs within a gene) to identify independent effects and estimates overall significance through permutation. For each test statistics, using meta-analysis summary data, the gene P-values were computed using 1,000,000 permutations and utilizing the 1000 Genomes Project ASW panel as a reference population to account for linkage disequilibrium (LD) between SNPs.

Genome-wide association and meta-analysis was performed for SBP in 1,617 subjects (843 males, 774 females) from the SIT Trial and CSSCD cohorts. The average ages of studied samples from the SIT Trial and CSSCD cohorts were 8.96 and 9.57 years, respectively. Detailed demographic and clinical characteristics for the study subjects are described in Table 1. The observed P-values show no early departure from the null (Figure S1), indicating minimal inflation (λ_GC=0.998) in test statistics due to potential population stratification and/or cryptic relatedness. None of the SNPs reached genome-wide significance (P-value <5.0x10^-8) (Figure 1). However, a number of suggestive candidate loci (P-value <5.0x10^-5) were identified that approached genome-wide significance (Table 2). The most significant signal was observed for rs7952106 (P-value: SIT Trial=6.40x10^-3; CSSCD=3.94x10^-5; Meta-analysis=8.57x10^-7). This SNP is located ~78 kb 5’ upstream of the dopamine receptor D2 subtype (DRD2) gene on chromosome 11. rs7952106 is a common SNP with a minor allele frequency (MAF) of 23%, and directly genotyped in both the cohorts. The direction effect of the minor allele (G) is consistent across both GWAS (SIT Trial: Effect size=1.65 mmHg/allele; CSSCD: Effect size=1.50 mmHg/allele) and associated with increase in SBP. A second DRD2 intronic SNP (rs17529477; minor allele frequency [A] =12%) showed the same direction effect (Effect size=1.76 mmHg/minor allele; P-value=1.93x10^-5) and was in low LD with rs7952106 (r²: SIT Trial=0.25 and CSSCD=0.24).

Given the suggestion of multiple independent signals in DRD2, we performed a gene-based test combining independent associations within a gene (using 20kb flanking region) and obtained a P-value for each gene using permutation. In total, 32,155 autosomal genes were tested, and results are shown in Table 3. Among all the tested genes, none met the genome-wide significant criteria (P-value < 2.0x10^-6). The most significant gene was MIR4301 (micro-RNA4301, Gene ID: 100422855) with a P-value=5.2x10^-5 (Table 3). MIR4301 is a 65 base-pair long non-coding RNA at chromosome 11 and contained within the DRD2 intronic region, and shares the same set of associated SNPs as observed for the DRD2 gene. To examine micro-RNA target binding predictions, we used the software RNA22 (version 1.0) (http://cbcsrv.watson.ibm.com/rna22.html) [38]. MIR4301 showed a predicted target site in the 3’ UTR region of the DRD2 transcript (ENSEMBL: ENST00000355319). In our study, the observed suggestive genetic signals of DRD2 region SNPs and the prediction of MIR4301 binding to DRD2 as a potential target, suggests the plausible involvement of DRD2 region in the regulation of SBP in SCD cohorts. Examining the GTEx database, which queries lymphoblastoid, liver, brain cerebellum, frontal cortex, and temporal cortex tissue, we found no known eQTLs within or in close proximity of this locus.

A total of 29 independent chromosomal loci associated with blood pressure have been reported from the ICBP meta-analysis, of which 28 show strong evidence for association with SBP. We attempted to validate these loci in the combined SCD cohorts. To ensure uniform comparison of the genetic effect and its direction, the SNPs were analyzed according to the reported coded alleles (under an additive genetic model). We were able to test 27 of these SNPs directly or with a proxy SNP (r²≥0.6), and 15/27 SNPs showed the same direction affect on SBP as reported in the ICBP study (Table 4, Figure S2). Of the directly genotyped/imputed or proxy SNPs, 4 were nominally significant in the combined SCD cohorts, and 6 were significant at P-value <0.10 (Table 4), demonstrating a clear enrichment of signal (P-value =0.0045). However, none of the 27 tested SNPs was significant after Bonferroni correction (α=0.05/27=1.85x10^-3). Given the limited power to detect significance for individual SNPs in the relatively smaller SCD samples (Figure S3), we constructed a genetic risk score for SBP incorporating the 27 previously reported SNPs; weighted according to effect sizes observed in the ICBP meta-analysis. The risk score derived from these 27 directly genotyped or proxy SNPs was nominally associated with SBP (P-value= 0.05), demonstrating the role of these SNPs in aggregate in the modulation of SBP in the SCD cohort.

Given that higher SBP is associated with increased risk of an SCI in SCD patients, we determined whether any of the SNPs (or their proxies) associated with SBP in ICBP study was associated with SCI. Three SNPs were nominally associated with SCI (Table 4), with only the CACNB2 (5’) locus (rs2357790) consistent between the SBP (P-value=0.05) and SCI (P-value=0.04) analyses. None of these SNPs was significant after Bonferroni correction. Further, to estimate the aggregated effect of these SNPs on increased risk of SCI, we constructed the genetic risk score for SCI (weighted according to effect sizes of published SBP SNPs) and no significant association was observed (P-value= 0.95).