A genome‐wide association and replication study of blood pressure in Ugandan early adolescents

Abstract Background Genetic association studies of blood pressure (BP) have mostly been conducted in non‐African populations. Using the Entebbe Mother and Baby Study (EMaBS), we aimed to identify genetic variants associated with BP among Ugandan adolescents. Methods Systolic and diastolic BP were measured among 10‐ and 11‐year olds. Whole‐genome genotype data were generated using Illumina omni 2.5M arrays and untyped variants were imputed. Genome‐wide association study (GWAS) was conducted using linear mixed model regression to account for population structure. Linear regression analysis was used to assess whether variants previously associated with BP (p < 5.0 × 10−8) in published BP GWASs were replicated in our study. Results Of the 14 million variants analyzed among 815 adolescents, none reached genome‐wide significance (p < 5.0×10−8) for association with systolic or diastolic BP. The most strongly associated variants were rs181430167 (p = 6.8 × 10−7) for systolic BP and rs12991132 (p = 4.0 × 10−7) for diastolic BP. Thirty‐three (17 single nucleotide polymorphisms (SNPs) for systolic BP, 15 SNPs for diastolic BP and one SNP for both) of 330 variants previously identified as associated with BP were replicated in this study, but none remained significant after accounting for multiple testing. Conclusion Variants showing suggestive associations are worthy of future investigation. Replication results suggest that variants influencing adolescent BP may overlap somewhat with those already established in previous studies, largely based on adults in Western settings.


| INTRODUCTION
Genome-wide association studies (GWAS) predominantly from Caucasian and Asian populations have identified several single nucleotide polymorphisms (SNPs) associated with blood pressure (BP; Cho et al., 2009;Evangelou et al., 2018;Levy et al., 2009;Qian, Lu, Tan, Liu, & Lu, 2007;Warren et al., 2017;Xi et al., 2014). Despite having a high burden of hypertension, and opportunities for improved fine-mapping of causative variants, including higher genetic diversity and lower linkage disequilibrium (LD; Addo, Smeeth, & Leon, 2007;Luoni et al., 2005;Noubiap et al., 2017;Tishkoff et al., 2009), African populations are underrepresented in published genetic studies of BP (Peprah, Xu, Tekola-Ayele, & Royal, 2015). Among 38 studies that investigated genetic polymorphisms associated with hypertension in Africa-based populations (participant numbers ranging from 65 to 1939) reviewed in (Yako et al., 2018), all adopted a candidate gene approach rather than conducting a GWAS. It remains unclear whether variants associated with BP in non-African populations also influence BP among Africans or whether the patterns of genetic susceptibility differ markedly.
Blood pressure GWASs conducted among populations of African origin in the diaspora are rare, and often report different variants associated with BP compared to those reported in non-African populations (Adeyemo et al., 2009;Fox et al., 2011;Franceschini et al., 2013;Kidambi et al., 2012;. Attempts to replicate genetic findings in independent populations have returned mixed results (Kayima et al., 2017;Kidambi et al., 2012;Li et al., 2017;Xi et al., 2014). The largest African American GWAS of BP included a metaanalysis of 29,378 individuals, and identified only one (the SOX6 locus) of the five loci that were associated with BP in a multiethnic (African American, European and East Asian) sample of 99,382 individuals (Franceschini et al., 2013). Of the 17 SNPs most strongly associated with BP (p < 1 × 10 −4 ) in African Americans, three SNPS were replicated (p < .05) in a West African sample (Adeyemo et al., 2009). Among Ugandan adults, 11 out of 27 BP related candidate SNPs (selected because of previous association with BP from BP GWASs or admixture mapping analysis) were replicated (p < .05) with eight of the 11 SNPs having the same effect direction as in the discovery sample (Kayima et al., 2017).
Twin studies report that over 30% of BP variability is heritable (Biron, Mongeau, & Bertrand, 1976;Williams et al., 1991) but established variants associated with BP account for only 2%-5% of BP variation (Ehret & Caulfield, 2013;Salfati, Morrison, Boerwinkle, & Chakravarti, 2015). This strongly suggests the existence of important undiscovered variants. This "missing heritability" could be due to rare or to common SNPs, all conferring small increases or decreases in expected BP (Maher, 2008;Manolio et al., 2009). There is a need for both GWAS and replication studies to further elucidate and improve the generalizability of BP genetic findings.
The role of environmental and of lifestyle factors in hypertension among Africans is well documented as reviewed previously (Addo et al., 2007;Noubiap et al., 2017). Previously, we described the role of environmental and of lifestyle factors in adolescents' BP and the findings were reported in Lule, Namara, Akurut, Muhangi, et al., 2019). However, the contribution of genetic variants remains unknown and understudied (Yako et al., 2018). It is not clear whether genetic loci associated with hypertension among populations of non-African origin influence susceptibility to or protection from hypertension in populations on the African continent. Independent confirmation is necessary to validate BP SNPs in different populations. We used data from the Entebbe Mother and Baby Study (EMaBS) birth cohort (Elliott et al., 2007) to conduct 1) a GWAS of systolic and diastolic BP and 2) a replication study of candidate SNPs identified in previously published BP GWASs. The GWAS aimed to identify novel BP loci unique to this population while candidate gene analysis aimed to identify variants influencing BP across different ethnic groups.
We hypothesized that genetic variants (either unique or not unique to populations in Africa) would be associated with BP among Ugandan adolescents. Individuals of African origin have different genetic makeup from individuals of non-African ancestries (Addo et al., 2007;Luoni et al., 2005;Noubiap et al., 2017;Tishkoff et al., 2009). Identifying genetic variants associated with BP enhances our understanding of BP regulation and might highlight potential drug targets for hypertension treatment and prevention. Furthermore, the identification of variants associated with hypertension in both adolescence and adulthood could offer opportunities for early risk prediction.

| Study design, population and setting
The EMaBS [trial registration ISRCTN32849447] in Uganda, was originally designed to investigate the influence of worms and their treatment in pregnancy and early childhood on vaccine response and on infections in childhood (Elliott et al., 2007). Briefly, between April 2003 andNovember 2005, 2,507 pregnant women in their second or third trimester were randomized in a 2 × 2 factorial design to receive single dose albendazole (400 mg) or matching placebo and single dose praziquantel (40 mg/kg) or matching placebo. At 15 months of age, the resulting 2,345 live-born infants were randomized to receive quarterly albendazole or matching placebo up to 5 years of age.

| Phenotyping
The offspring continued under follow-up after the trial ended in 2011. From 20 May 2014 to 16 June 2016, cohort participants who were now aged 10-11 years were enrolled in a BP sub-study, as part of which additional anthropometric and BP data were collected (Lule, Namara, Akurut, Muhangi, et al., 2019). Adolescents were included in this study if they were aged 10 or 11 years and attending their routine annual followup visit during the BP sub-study period (11-year-olds who had previously enrolled as 10-year-olds were not included twice). Where necessary, enrollment was postponed until the participant was free of malaria (fever or axillary temperature ≥37.5°C and parasitemia) and other illnesses (Lule, Namara, Akurut, Muhangi, et al., 2019). BP was measured as previously described (Lule, Namara, Akurut, Muhangi, et al., 2019). Briefly, on the BP study visit day, after 5 min rest period, trained nurses measured BP thrice 5 min apart, on the right arm supported at the heart level, with the participant seated upright all the way to the back of the chair, with legs uncrossed and feet flat on the floor. Automated Omron (M6, HEM-700) machines validated every 6 months by the Uganda National Bureau of Standards were used. Blood pressure phenotypes for this analysis were the mean of the second and third readings for systolic and diastolic BP, that is, systolic and diastolic BP were analyzed as two separate phenotypes. The second and third BP readings were, on average, lower than the first BP reading but similar to each other for both systolic and diastolic BP (Lule, Namara, Akurut, Muhangi, et al., 2019).

| Genotyping and quality control
Earlier in 2013, whole-genome genotyping of 1,391 EMaBS participants was undertaken. Genotypic data were generated from red cell pellets that had been separated and stored at −80°C until processing. Approximately 2.2 million genetic variants were generated at Wellcome Sanger Institute using the Illumina HumanOmni2.5M-8 ("octo") Beadchip arrays, version 1.1 (Illumina Inc.). Quality control (using standard pipelines) was performed at the University of Oxford using commands in PLINK (version 1.7; Purcell et al., 2007) to remove individuals and variants with high levels of missingness or deviations from expected levels of heterozygosity or Hardy-Weinberg equilibrium (p < 1 × 10 −8 ). Untyped genetic variants and the variants identified for replication analysis were imputed in the EMaBS sample using a merged panel ( Marchini, Howie, Myers, McVean, & Donnelly, 2007) were used for imputation using settings as recommended for African populations. Only SNPs with an INFO score >0.3 and a minor allele frequency >0.01 were taken forward for analysis.

| Association analysis
The analysis included EMaBS adolescents with phenotypic and genotypic data. The two outcomes (mean systolic BP and mean diastolic BP) were analyzed separately. GWAS of BP (systolic and diastolic) as quantitative traits was done using mixed linear regression methods (accounting for population substructure) assuming an additive model and controlling for age and body mass index (BMI) as covariates in genomewide complex trait analysis (GCTA) version 1.22 (Yang, Lee, Goddard, & Visscher, 2011). A p < 5 × 10 −8 was considered as the threshold to denote genome-wide significance for SNPs. Results for SNPs with p < 1 × 10 −6 are reported. Manhattan plots and Quantile-Quantile (Q-Q) plots were constructed to show the distributions of association p-values and the departure of the observed p-values from the null, respectively.
For the replication component of this study, previously published BP GWASs were searched to identify SNPs reported to be associated with systolic or diastolic BP (p < 5 × 10 −8 in the original GWAS) and these variants were considered for the replication. Replication analysis was conducted using linear regression adjusting for age and BMI in Stata version 14 (College Station, Texas, USA). Variants were tested for association with the phenotype they were associated with in the published GWAS, that is variants associated with systolic BP in a published GWAS were tested for association with systolic BP but not with diastolic BP and vice versa. P < .05 was considered the threshold for statistical significance for the replication study although results were also interpreted in light of a Bonferroni correction allowing for all tests done in the replication analysis.
The base pair position is based on the Genome Reference Consortium Human Build 37, February 2009 (GRCh37/ hg19).

| RESULTS
The discovery GWAS analysis used data on 20,074,711 SNPs from 815 adolescents. These adolescents had a mean age of 10.4 years, a mean BMI of 16.0 kg/m 2 , mean systolic BP of 106.0 mmHg, and mean diastolic BP of 65.3 mmHg. Four hundred and seventeen (51%) of the adolescents were male. Detailed characteristics of EMaBS participants included and not included in the analysis are described in Table S1. Offspring included in the genetic analysis were similar for most characteristics to those not included, except that those included were more likely to have been delivered in Entebbe hospital than elsewhere and to have been exclusively breastfeeding at 6 weeks of age.
The distributions of association p-values (Manhattan plot) for systolic and diastolic BP phenotypes are shown in Figure  1 and the Q-Q plots in Figure 2. The observed P-values show no departure from the null (Figure 2), either for systolic or diastolic BP, with lambda values of 1.006 and 0.995, respectively. The results show adequate control for population substructure in the analysis.
The SNPs most strongly associated with BP are shown in Table 1. None of the SNPs reached genome-wide level of significance (p < 5 × 10 −8 ) for association with adolescent systolic or diastolic BP. Borderline significance (5 × 10 −8 < p < 1 × 10 −6 ) for association with systolic BP was achieved for four index SNPs at four separate loci; there were four SNPs showing borderline significant associations with diastolic BP at four separate loci. There was no overlap between the SNPs most strongly associated with systolic BP and those most strongly associated with diastolic BP. None of these index SNPs have been identified as associated with BP in previously published BP GWAS. The most strongly associated SNP for systolic BP with p-value 6.8 × 10 −7 was rs181430167, located in the intron region of KLHL29 on chromosome 2. The lowest p-value (4.0 × 10 −7 ) for association with diastolic BP was for rs12991132 located between ZFP36L2, THADA, LOC1001297261 on chromosome 2.
Of the 389 SNPs (Adeyemo et al., 2009;Cho et al., 2009;Ehret et al., 2016; International Consortium for Blood Pressure Genome-Wide Association Studies et al., 2011;Evangelou et al., 2018;Fox et al., 2011;Franceschini et al., 2013;Ganesh et al., 2014Ganesh et al., , 2013Ho et al., 2011;Hoffmann et al., 2017;Johnson, Newton-Cheh, et al., 2011;Johnson, Gaunt, et al., 2011;Kato et al., 2015Kato et al., , 2011Kidambi et al., 2012;Levy et al., 2009;Li et al., 2017;Liang et al., 2017;Liu et al., 2016;Newton-Cheh, Johnson, et al., 2009;Newton-Cheh, Larson, et al., 2009;Parmar et al., 2016;Simino et al., 2014;Surendran et al., 2016;Takeuchi et al., 2010;Tragante et al., 2014;Wain et al., 2017Wain et al., , 2011Warren et al., 2017) identified for replication, 330 (85%) SNPs were included in the replication analysis. Fifty-nine SNPs that were either rare (<0.01) or poorly imputed (INFO score <0.3) in the EMaBS sample were not included in the replication analysis. Thirty SNPs of the 330 SNPs included in the replication had been previously associated with BP in populations of African origins. Forty SNPs had been previously associated with both systolic and F I G U R E 1 Manhattan plots for the association of SNPs with (a) systolic blood pressure and (b) diastolic blood pressure adjusting for age and body mass index as fixed covariates | 5 of 18 LULE Et aL. diastolic BP and were tested for association with both. Tables  2 and 3 show results from the replication analysis. Briefly, 33 SNPs (17 for systolic, 15 for diastolic and one for both systolic and diastolic BP) were associated with BP in this population, with the same effect direction as the discovery population for 14 of the SNPs (five for systolic BP and eight for diastolic BP and one for both systolic and diastolic BP).
Of the 30 SNPs previously known to be associated with BP specifically in individuals of African origin, three (ATP2B rs2681492, MDM4 rs2169137 and EVX1/HOXA rs17428471) were associated with BP in the present study.
Only the EVX1/HOXA rs17428471 had the same effect direction as in the discovery population. The BAT2/BAT5 rs805303 variant was associated with systolic BP and diastolic BP of the 40 SNPs tested for association with both traits. The G allele of the BAT2/BAT5 rs805303 variant was associated with higher systolic and higher diastolic BP among adolescents in this study: the same effect direction observed in the discovery population for both traits.
There were 370 independent tests (197 for systolic BP, 173 for diastolic BP) conducted, thus approximately 19 SNPs (370 × 0.05 = 18.5) would be expected to be associated F I G U R E 2 Quantile-Quantile (Q-Q) plots for the two phenotypes and the genomic control coefficient (lambda). (a) systolic blood pressure and (b) diastolic blood pressure T A B L E 1 Genome-wide association study results: SNPs associated with blood pressure (systolic or diastolic) at p < 1.0 × 10 −7

| DISCUSSION
To our knowledge, this is the first genetic analysis examining variants associated with BP among African adolescents. We hypothesized that common genetic variants (unique or not unique to populations in Africa) were associated with systolic and, or diastolic BP in Ugandan adolescents and that these associations may overlap with associated variants identified in previous studies of Africans. The GWAS revealed no novel or previously identified variant associated with systolic or diastolic BP in our study population. Thirty-three SNPs were associated with BP in the replication analysis, with the direction of effect consistent with the discovery population for 14 SNPs. There were no SNPs reaching a Bonferroni-adjusted significance level. None of the replicated SNPs were located in genes with monogenic effect on hypertension (Ehret & Caulfield, 2013). The SNPs most strongly associated with either systolic or diastolic BP were of borderline significance and none have been reported as associated with BP in previous BP GWASs. The most strongly associated SNPs were mostly common variants with modest effect sizes and might uniquely influence BP in African population. It is important for larger genetic studies of African population to investigate the role of these SNPs in BP regulation among Africans. These top SNPs are potential candidates for replication analysis in African populations.
The failure to identify variants strongly associated with BP presumably occurred because the study was underpowered to detect effects of rare variants or small effects of common variants. Blood pressure is most likely a polygenic trait influenced by the simultaneous presence of several gene variants each with a small effect size and contributing in an additive manner to BP expression. Thus, the large effect sizes that this study had good power to detect, may not be realistic. For example, the present study had 80% power to detect a 3.2 mmHg change in mean systolic BP for a minor allele frequency of 20% at genome-wide significance level, p < 5 × 10 −8 . Many of the variants reliably associated with BP in adults have an effect size of 0.5 mmHg or less (Evangelou et al., 2018). In addition to the limitation caused by the relatively small sample size, imputation did not allow inference for rare variants not included in the imputation SNP panel. Few GWAS "top SNPs" from non-African populations have been replicated in populations of African ancestry (Adeyemo et al., 2009;Franceschini et al., 2013;Kayima et al., 2017). Variants associated with BP in populations of African origin might be different from variants that influence BP in Caucasian populations or not in LD with the BP causing variants. Our replication study was limited to variants associated with BP from previous GWAS of BP in other populations. Thirty-three SNPs identified from previous BP GWAS were replicated, most of these were previously identified in populations of non-African origins. Of the identified loci, PAX2 is essential in the development of the renal epithelium (Dressler & Woolf, 1999) and plays a critical role in kidney development (Hou, Chen, & Wang, 2011). The kidneys are critical in BP regulation. Two of the replicated SNPs are located on ATP2B1. ATP2B1 is involved in calcium homeostasis (Hirawa, Fujiwara, & Umemura, 2013). The ATP2B1 rs2681492, MDM4 rs2169137, EVX1/HOXA rs17428471 SNPs previously associated with BP in transethnic populations (African, Caucasian and Asian) were associated with adolescent BP in the present study. These genes most likely influence BP across different ethnic groups.
Replication studies in diverse populations have returned mixed results. This current study conducted 370 tests using 330 SNPs, of which 33 SNPs (one SNP for both traits) were associated with BP. Failure to replicate most variants associated with BP in other populations could be due to differences in minor allele frequencies across populations or differences in LD patterns combined with a poor understanding of the causative variants or due to spurious initial findings. Blood pressure is likely to be influenced by the simultaneous presence of several genetic variants, each conferring a small change in BP.
Although none of the variants reached Bonferroni level of significance more associated variants were identified than expected under the null suggesting that some of these variants found to be associated could be worthy of further followup. Fourteen variants more than those expected by chance (19 variants) under the null hypothesis were associated with BP in this study.
Of the 40 SNPs tested for association with both traits, only BAT2/BAT5 rs805303 was associated with both traits in the adolescents, suggesting that not many genetic loci have influence on both systolic and diastolic BP. Similar to an earlier replication study among adult Ugandans (Kayima et al., 2017), the ATP2B1 rs2681492 was associated with BP in these Ugandan adolescents, but with an opposite effect direction to the discovery population (Hoffmann et al., 2017) and Ugandan adults (Kayima et al., 2017). The G allele of the ATP2B1 rs2681492 was associated with lower systolic BP in Ugandans adults (Kayima et al., 2017) but with higher systolic BP in the present study.
Some loci may have varying roles in BP regulation across different populations. A European study investigated SNPs associated with BP at different age epochs (using independent samples for each age group): prepuberty (age 4-7 years), pubertal (8-12 years), and postpubertal (13-20 years). The A allele of TGA11 rs1563894 was associated with lower systolic BP in prepuberty while the T allele of SMARCA2/VLDLR rs872256 was associated with higher systolic BP during puberty (Parmar et al., 2016). No SNP was associated with BP in the postpubertal period, and no SNP was consistently associated with BP across all three age groups. The TGA11 rs1563894 and SMARCA2/VLDLR rs872256 were not replicated in this present study.
The present study has several strengths. This is the first BP GWAS of an African population and the first candidate gene analysis among adolescents residing in Africa. Participants in this study were similar to nonparticipants with respect to most baseline characterizes. Rigorous quality control procedures were used during the measurement of the various variables including BP and genotyping. Data from this study can contribute to future BP GWAS meta-analyses. Key limitations of this study were that it was underpowered to detect effects of rare variants and to enable testing for effect modification by sex and other environmental variables, and the lack of a replication sample from a similar setting with which to confirm our GWAS findings.
Future work should take advantage of various African cohorts to form consortia that can enable the conduct of GWAS meta-analysis well powered to identify rare and low-frequency variants that may be associated with BP in African populations. Future candidate gene analysis using a sample from a different geographical region or ethnic background should investigate for interactions between variants, this might help our understanding of the etiology of BP. It is possible that multiple interacting variants (rare and common) are influencing BP levels in this population. Although we did not formally allow for multiple testing in the replication analysis, the current study had 33 associated SNPs 14 more than expected by chance. Polygenic scores analysis of variants associated with BP among African populations may explain the missing BP heritability.
In summary, we conducted the first genetic study of BP phenotypes among Ugandan adolescents. Although this study did not identify novel BP variants, replication of some previously identified variants suggests that some genetic variants may universally influence BP susceptibility. Large scale studies in African populations are required to identify novel and evaluate previously reported loci.