Association of Regions on Chromosomes 6 and 7 With Blood Pressure in Nigerian Families

Abstract

Background

Elevated blood pressure (BP) shares a level of heritability similar to many other traits related to cardiovascular risk; however, specific susceptibility loci have been difficult to localize. We conducted a multistage study of BP as a continuous trait in a low-risk West African population in which it was anticipated that environmental exposures would be reduced in complexity and intensity. In our earlier genome-wide linkage study for BP in this population, strong linkage evidence was noted on chromosomes 6 and 7.

Methods and Results

We subsequently genotyped a total of 3431 tag single-nucleotide polymorphisms (SNPs) in 3 regions (viz, 152.68 to 165.99 Mb on chromosome 6, 0.29 to 20.67 Mb, and 104.09 to 123.06 Mb on chromosome 7) in 713 individuals from 199 families. We conducted a family-based association analysis using individual SNPs and associated haplotypes. After correction for multiple comparisons, 6 intronic and 1 intergenic SNPs achieved nominal statistical significance (P< 0.05) for the association with BP. The associated intronic SNPs include 2 in the PARK2 gene on chromosome 6; 2 in the KCND2 gene, and 1 each in the C7orf58 and HDAC9 genes on chromosome 7. The intergenic SNP is located between the RPA3 and GLCCI1 genes on chromosome 7. The haplotypes on which these SNPs resided were more strongly associated with BP than their respective single SNPs. The frequency of the “at-risk” haplotypes ranged from 14% to 48%.

Conclusions

These data provide preliminary evidence that regions on chromosomes 6 and 7 may influence susceptibility to elevations in BP.

The genetic epidemiology of complex traits has evolved rapidly, leading to a flood of important new observations within the last year.^1–8 Despite enormous effort, however, relatively little progress has been made in unraveling the genetics of hypertension. Studies of physiological candidate genes have been uniformly unrewarding, with the potential exception of angiotensinogen.^9,10 The family-based design, including studies with sample sizes over 10 000, has yielded similarly disappointing results.^11–13 Even the initial set of genome-wide association studies, which have been remarkably productive for a range of other conditions, has failed to identify reproducible markers for hypertension risk.¹⁴ On the other hand, there is no doubt that blood pressure (BP) is a moderately heritable trait, and it is unlikely to be influenced by alleles of a smaller effect than those associated with height, for which more than 20 loci have been identified.^4,7,15,16

Three well-recognized factors are likely to complicate the search for genes that influence BP. First, the trait is much harder to measure reliably than are many other phenotypes. Second, susceptibility to hypertension is nearly universal, with over half of the US population expressing the trait by age 60 and the lifetime risk approaching 85%.^17–19 Third, the set of causal environmental exposures is complex and will modify the impact of a given locus. Given the complexity, which these factors impose over and above the underlying genetics of BP, additional persistence will be required to identify the relevant susceptibility loci.

We have undertaken a long-term study of hypertension in the African diaspora which includes characterization of both environmental and genetic factors. At the Nigerian site, heritability of BP and associated traits is high among families in the Yoruba-speaking community, suggesting that genetic effects may be more prominent, and variability of 2 key exposures—sodium intake and obesity—are at low levels.^20,21 Robust linkage evidence was previously obtained on chromosomes 6 and 7 in a set of these Nigerian families (logarithm of odds >3.0),²² providing regions suitable for fine mapping. We report here the result of the fine mapping

Methods

Participant Recruitment

The sampling frame for this study was provided by the International Collaborative Study on Hypertension in Blacks (ICSHIB), as described in detail elsewhere.^23–25 The protocol was approved by the Institutional Review Board of Loyola University and the Joint Ethical Committee of the University of Ibadan/University College Hospital, Ibadan. The consent process was presented in Yoruba or English and written informed consent was obtained from all participants. A screening examination was completed by trained research staff using a standardized protocol.²³ Trained local interviewers obtained a medical history and a family pedigree in the participant’s native language. BP observers were trained and certified by a previously described procedure.^23,24 An oscillometric device, previously evaluated in our field settings, was used for all BP measurements.²⁴ Three measurements were taken 3 minutes apart, and the average of the final 2 was used in the analysis. Participants with hypertension were offered treatment after detection at the screening examination. However, none of the participants for this particular fine mapping study was receiving treatment at enrollment. All study participants were recruited from the Igbo-Ora community in southwest Nigeria and were part of samples described in previous reports.²²

Genotyping

DNA samples were extracted and submitted to Affymetrix, Inc. (South San Francisco, Calif) for SNPs genotyping using a custom 3K Panel. Tag SNPs were identified from the 2 linkage regions using the HapMap Yoruba in Ibadan, Nigeria (YRI) data, release 16c.1 (study participants all belonged to the Yoruba ethnic group that the HapMap YRI sample came from). Selection of tag SNPs was done using the pairwise tagging algorithm²⁶ with a minimum coefficient of determination (r²) of 0.8 and minor allele frequency (MAF) cut off of 5.0%. Three thousands four hundred thirty-one tag SNPs were typed from a 13.31 Mb region on chromosome 6 (152.68 to 165.99 Mb; SNPs=1203) and 2 discrete regions totaling 39.35 Mb on chromosome 7 (0.29 to 20.67 Mb and 104.09 to 123.06 Mb; SNPs=2228). Of the 911 unique samples, 843 (93%) yielded successful genotypes, whereas the rest failed because of concentrations below the Affymetrix specification of 150 ng/μL (n=57), contamination (n=5), and unknown cause (n=6). Genotype data completeness was 99.32% and repeatability based on positive controls was 99.97%. Trio accuracy based on positive controls was 99.95%.

Statistical Analysis

Quality control and Hardy-Weinberg equilibrium tests were performed for each SNP using the software Haploview.²⁷ SNPs with greater than 5% missing genotypes (n=38) or with minor allele frequencies less than 1% (n=56) were excluded from all subsequent analyses. A total of 60 SNPs had Hardy-Weinberg equilibrium probability values <0.001, but were not excluded from association analysis since the lack of Hardy-Weinberg equilibrium could be evidence of association. The total number of SNPs analyzed was 1169 on chromosome 6 and 2168 on chromosome 7. There were no Mendelian inheritance errors detected in the resulting cleaned data. Descriptive statistical analysis was performed using the SAS software,²⁸ whereas distribution of familial relationship types and familial trait correlations were determined using the PEDINFO and FCOR procedures implemented in the software SAGE.²⁹

SNP Association Analysis

In this study BP was used as quantitative trait because none of the participants was receiving consistent treatment at the baseline examination when biological samples were collected for DNA analysis. Tests for association of each SNP separately for systolic blood pressure (SBP) and diastolic blood pressure (DBP) adjusting for sex, age, age², and body mass index were performed by using the variance-components-based total association procedure implemented in the QTDT software.³⁰ In QTDT, this approach evaluates the total evidence for association by simultaneously modeling the means and the variances,³⁰ as

$$y_{ij}=u+{\beta }_a{g}_{ij}+{\beta }_x{X}_{ij}$$

where y_ij is the phenotype of the jth member of the ith family, u is the population mean, β_a is the additive effect of each SNP, g_ij is the genotype score of the jth member of the ith family for the SNP being tested with score equal to 0, 1, or 2 depending on the number of reference allele in the individual’s genotype for that SNP and β_x is the vector of covariate effects corresponding to the covariates X_ij. The additive, polygenic and environmental variances are estimated in the variance-covariance matrix Ω_i for each ith family as

$${\mathrm{\Omega }}_{\mathit{ijk}}=\{\begin{array}{ll}{\sigma }_a^2+{\sigma }_g^2+{\sigma }_e^2\hfill & \text{if}j=k\hfill \\ {\pi }_{\mathit{ijk}}{\sigma }_a^2+2{\varphi }_{\mathit{ijk}}{\sigma }_g^2\hfill & \text{if}j\ne k\hfill \end{array}$$

Where ${\sigma }_a^2,{\sigma }_g^2$, and ${\sigma }_e^2$ are the additive, polygenic, and environmental variances, respectively; and π_ijk represents identity-by-descent sharing between individuals j and k in ith family, and ϕ_ijk is the kinship coefficient between the 2 individuals.^31–34

For each SNP, evidence for association was evaluated by likelihood ratio test between the null model in which β_a was constrained to zero and the alternative model where β_a was estimated. The significance level of each association was empirically derived by permutation. Rather than using the permutation procedure implemented in QTDT, which is not applicable to total association, we adopted a permutation approach based on permuting trait values from which dependence or resemblance among relatives was first filtered out. Residuals from covariate-adjusted (sex, age, age², and body mass index) polygenic models fitted separately for SBP and DBP were used as traits in the permutation procedure to avoid the simultaneous permutation of traits with covariates. Individuals without trait values but who were retained for the purpose of pedigree connections were assigned the trait mean value. This helped to insure computational efficiency in the generation of the null data for permutations without change in the effective sample size because the individuals without trait values also did not have genotype data, and hence do not enter into the association analysis. The next step involved the computation of the singular value decomposition of the matrix of coefficients of relationship for each family. Let R be an m×m symmetrical matrix of coefficient of relationship (which is 2 times coefficient of kinship) for a given family, then singular value decomposition of R is of the form

$$R=U\mathrm{\sum }{U}^T$$

where U is an m×m orthogonal matrix containing the eigenvectors of R, Σ is an m×m diagonal matrix of the singular values (s_i) of R with s_ij=0 if i≠j and s_ij=s≥0.³⁵ Also, let Y be an m×1 matrix of the trait values for the m members of the family, then the new trait values without dependence among relatives is obtained as follow:

$$\mathrm{\Upsilon }=U{\mathrm{\sum }}^{-\frac{1}{2}}Y$$

where ϒis the m×1 matrix of new trait values which are independent among related individuals. The new trait values are subsequently permuted among those corresponding s_ii=s_i≥0 across families; this procedure satisfies the exchangeability required by permutation test because of the absence of phenotypic dependence among relatives in the new trait values. After permutation, phenotypic dependence among relatives or family members is then re-established as follow:

$$\text{P}={\mathrm{\sum }}^{\frac{1}{2}}{U}^T\mathrm{\Upsilon }$$

where P is the m×1 matrix of the resulting permuted trait values with dependence among family members.

This permutation technique provides the advantage of keeping intact the linkage disequilibrium patterns between the genetic markers, and also mitigates any possible normality problem with continuous traits. Ten thousands permuted data sets were generated for SBP and DBP and association analysis carried out the same way as with the real data. For each permutation, the test statistic corresponding to the most significant SNP of all the SNPs tested was recorded, and the empirical significance level of each SNP was then calculated as the proportion of the 10 000 permutation test statistics that were at least as extreme as that observed for the SNP in the real data. These empirical probability values also control for multiple comparisons because they were based on the most extreme permutation test statistics across all loci tested.

Haplotype Analysis

Markers with significant association with BP were selected and used in haplotype analysis with other markers within the 200 kb flanking region of the significant markers (100 kb both to the right and left of each significant marker). The software Haploview²⁷ was used to compute estimates of linkage disequilibrium for each pair of SNPs by the standard D-prime method³⁶ and to determine the haplotype blocks—regions with no evidence of a historical recombination event, but significant level of linkage disequilibrium. The haplotype blocks were defined by the method of four-gamete test^37,38 as implemented in Haploview. Analysis was restricted to haplotypes with frequencies greater than 1% and on which any of the BP-associated SNPs resided. Haplotype assignment of family members was performed using the software Merlin.³⁹ Haplotypes were coded as the number of copies carried by each individual and tested for association with BP in a similar approach to the SNP association described earlier. Empirical significance of the test statistics for the haplotype association was evaluated by a permutation method similar to that described for the SNP association analysis above. The authors had full access to and take full responsibility for the integrity of the data. All authors have read and agree to the manuscript as written.

Results

The descriptive characteristics of the study participants are presented in Table 1. The final analytic study sample consisted of 713 adult individuals from 199 nuclear and extended families with both phenotype and genotype data. The distribution of gender shows that females constituted 49% of the total participants and were on average 1 year older than the males (44 versus 43 years). Females also had higher body mass indexs and BPs compared with the males; however, the difference was only significant for body mass index. Table 2 shows the distribution of the relative pair types and their corresponding familial correlations for BP. There were 381 parent/offspring pairs, 272 sib pairs, and 90 half-sib pairs in the first-degree relative category; whereas in the second-degree relative category there were 6 grandparent/grandchild pairs, 120 avuncular pairs, and 55 cousin pairs. All familial correlations for both SBP and DBP were positive and ranged between 0.22 and 0.36 for first-degree relatives (Table 2).

Table 1.

Descriptive Characteristics of Phenotypes

	Females	Males	All
N (%)	350 (49.1)	363 (50.9)	713
Age, years	43.96±15.7	43.31±16.1	43.62±15.9
Body mass index, kg/m2†	22.78±4.2	21.13±3.4	21.94±3.9
Systolic BP, mm Hg	128.19±32.4	125.25±23.5	126.70±28.2
Diastolic BP, mm Hg	78.84±17.8	77.73±15.3	78.28±16.6

Values are expressed as mean±SD.

^†Significantly different between females and males.

Table 2.

Distribution of Relative Pair Types and Familial Correlations for BP

		Correlation Coefficient for BP
	N	Systolic	Diastolic
Families	199
Parent/offspring	381	0.36	0.22
Sib pair	272	0.33	0.26
Half-sib pair	90	0.18	0.09
Avuncular	120	0.30	0.34
Cousin	55	0.23	0.07

SNP Association Analysis

A total of 3337 SNPs distributed across the 13.31 Mb region on chromosome 6 and 2 discrete regions, 20.38 and 18.97 Mb, respectively, on chromosome 7 were included in the covariates-adjusted quantitative family-based total association tests for SBP and DBP, as described earlier. A total of 42 (1.3%) of the SNPs were significantly (P<0.01) associated with DBP and 36 (1.1%) were associated with SBP prior to adjustment for multiple comparisons. The 20 most significant BP associated-SNPs are presented in Table 3, along with their corresponding empirical probability values corrected for multiple comparisons within each BP phenotype. None of the 60 SNPs with Hardy-Weinberg equilibrium probability values <0.001 was associated with BP. After accounting for multiple comparisons, 2 SNPs each on chromosomes 6 and 7 remained significantly associated with SBP. The 2 SBP-associated SNPs on chromosome 6 are rs2315314 (MAF 12.8%) and rs16892620 (MAF 8.6%), both of which are located in intron 9 of the PARK2 (Parkinson disease (autosomal recessive, juvenile 2) gene on 6q25.2-q27. On chromosome 7, the 2 SBP-associated SNPs are rs13237260 (MAF 45.9%) and rs2068637 (MAF 37.5%), both of which are intergenic SNPs, with the closest genes being RPA3 and GLCCI1.

Table 3.

List of Top 20 Most Significant BP-Associated SNPs

Chromosome	dbSNP	Position, Mb	Gene	Alleles*	MAF, %	χ2	Observed P	Empirical P†
Systolic BP
6	rs2315314	161.822	PARK2	T:C	12.80	15.23	0.0006	0.0089
6	rs16892620	161.831	PARK2	C:A	8.60	14.65	0.0005	0.0089
7	rs13237260	7.746	GLCCI1	G:T	45.90	12.41	0.0006	0.0276
7	rs2068637	7.747	GLCCI1	G:A	37.50	11.65	0.0007	0.0381
7	rs10085563	120.483	C7orf58	A:G	40.50	10.53	0.0009	0.0717
7	rs12706309	120.483	C7orf58	T:G	28.40	10.17	0.0016	0.0851
7	rs2160010	120.476	C7orf58	C:T	29.00	10.05	0.0017	0.0905
7	rs17136953	7.742	GLCCI1	G:A	48.40	9.52	0.0029	0.1178
7	rs6953454	120.252	TSPAN12	T:C	28.30	9.46	0.0018	0.1219
7	rs6466759	120.257	TSPAN12	A:T	28.10	9.29	0.0016	0.1338
7	rs6973457	120.444	C7orf58	T:C	27.90	9.25	0.0028	0.1396
7	rs2215786	120.448	C7orf58	G:A	27.90	9.25	0.0028	0.1396
7	rs12673992	120.160	KCND2	G:A	24.90	9.09	0.0017	0.1505
6	rs12664209	162.566	PARK2	T:C	16.30	8.95	0.0023	0.1618
6	rs10945805	162.574	PARK2	G:A	16.30	8.95	0.0023	0.1618
7	rs12056299	15.663	MEOX2	C:T	6.70	8.92	0.0059	0.1618
7	rs637766	4.177	SDK1	C:T	46.10	8.62	0.0038	0.1858
7	rs7804315	120.058	KCND2	C:T	33.40	8.5	0.0043	0.1955
6	rs9356092	163.358	PACRG	T:G	37.90	8.42	0.0028	0.2043
6	rs9285542	154.422	OPRM1	C:T	15.40	8.26	0.0048	0.2196
Diastolic BP
7	rs2160010	120.476	C7orf58	C:T	29.00	17.19	0.00003	0.0018
7	rs12706309	120.483	C7orf58	T:G	28.40	16.36	0.00005	0.0031
6	rs2315314	161.822	PARK2	T:C	12.80	15.92	0.00007	0.0053
6	rs16892620	161.831	PARK2	C:A	8.60	14.96	0.0001	0.0095
7	rs12673992	120.160	KCND2	G:A	24.90	13.38	0.0003	0.0209
6	rs7744171	162.273	PARK2	G:A	17.00	12.36	0.0004	0.0263
7	rs11505418	18.960	HDAC9	T:C	16.70	12.51	0.0004	0.0263
7	rs7804315	120.058	KCND2	C:T	33.40	11.9	0.0006	0.0405
7	rs2248890	120.385	ING3	G:A	23.70	10.97	0.0009	0.0581
7	rs10085563	120.483	C7orf58	A:G	40.50	11.04	0.0009	0.0581
7	rs11768780	18.893	HDAC9	C:T	9.40	10.58	0.0011	0.0690
7	rs4730972	120.094	KCND2	T:C	33.60	10.54	0.0012	0.0754
7	rs6953454	120.252	TSPAN12	T:C	28.30	10.21	0.0014	0.0858
7	rs10265031	4.189	SDK1	T:G	25.10	10.04	0.0015	0.0909
7	rs2286209	7.612	FLJ20323	A:G	36.10	10.07	0.0015	0.0909
7	rs6466759	120.257	TSPAN12	A:T	28.10	9.94	0.0016	0.0976
7	rs10251692	120.502	C7orf58	C:T	42.60	9.94	0.0016	0.0976
7	rs10275526	120.504	C7orf58	T:C	42.60	9.94	0.0016	0.0976
6	rs2156745	162.357	PARK2	G:T	11.70	9.73	0.0018	0.1070
6	rs9458849	163.863	QKI	A:G	10.70	9.79	0.0018	0.1070

^*First allele is the minor.

^†Based on 10 000 permutations, this accounts for multiple comparisons within each phenotype.

There are 8 SNPs with sustained significant association with DBP after correcting for multiple comparisons. The DBP-associated SNPs include rs2315314, rs16892620, and rs7744171 (MAF 17.0%), all located on the PARK2 gene on chromosome 6; rs2160010 (MAF 29.0%) and rs12706309 (MAF 28.4%) both intronic SNPS in FLJ21986 (C7orf58) on 7q31.31. Others are rs12673992 (MAF 24.9%) and rs7804315 (MAF 33.4%) both intronic SNPs in the KCND2 gene on 7q31; and rs11505418 (MAF 16.7%) an intronic SNP in the HDAC9 gene on 7p21.1 (Table 3).

To be sure that none of the observed associations was due to population stratification, we tested for population stratification at each marker locus by decomposing the genotype score into orthogonal between- and within-family components according to the method described by Fulker et al⁴⁰ and implemented in QTDT but observed no significant (P>0.05) evidence of population stratification for the SNPs.

Further analyses to investigate whether the observed significant association at any of the loci is accounted for by significant association at a nearby locus were carried out. This was done by including the genotype score of nearby significant SNPs one at a time as a covariate in the analysis model involving each significant SNP. The results revealed that the association of rs16892620 with both SBP and DBP on chromosome 6 could be fully accounted for by the association of rs2315314 with both traits. Likewise, the association of rs13237260 on chromosome 7 with SBP was found to sufficiently explain that of rs2068637 with SBP. Also on chromosome 7, rs2160010 was found to fully explain the association of rs12706309 with DBP. On the basis of these results, SNPs independently associated with DBP are rs7744171, rs2160010, rs12673992, rs11505418, and rs7804315. rs2315314 is independently associated with both SBP and DBP; whereas rs13237260 is with SBP. The covariates-adjusted effect sizes for the SBP-associated SNPs are 7.84 and 5.24 for rs2315314 and rs13237260, respectively. For the DBP-associated SNPs, the covariates-adjusted effect sizes are −3.68 (rs2160010), 4.74 (rs2315314), −3.56 (rs12673992), −4.02 (rs7744171), 3.85 (rs11505418), and −3.00 (rs7804315).

Haplotype Analysis

To explore possible haplotype association with BP, SNPs within a 200 kb flanking region of the 7 independently associated SNPs (ie, 100 kb both upstream and downstream of each marker) were selected and included in a haplotype-based analysis. Haplotypes with frequencies less than 1% were excluded from analysis. With the exception of rs11505418 for which there was no haplotype, there were 25 different haplotypes involving the other 6 SNPs. The distribution and frequencies of the haplotypes are presented in Table 4.

Table 4.

List of Haplotypes Bearing the Associated SNPs

Haplotype Serial	Haplotype No.	Frequency, %
(rs13209191–rs9458278 –rs6937182– rs9456676 –rs2315314*–rs766313)
Hap1-1	G-T-G-C-T-A	23.8
Hap1-2	G-T-G-C-T-G	22.5
Hap1-3	A-A-T-T-T-G	16.8
Hap1-4	G-T-G-C-C-G	13.6
Hap1-5	G-T-T-T-T-G	11.5
Hap1-6	G-A-T-T-T-G	10.6
(rs7744171*–rs7746164–rs7759273)
Hap2-1	G-T-A	36.1
Hap2-2	G-C-C	33.2
Hap2-3	A-C-A	16.8
Hap2-4	G-C-A	12.8
(rs17136953–rs17137001– rs13237260*)
Hap3-1	A-A-G	48.4
Hap3-2	G-A-T	37.1
Hap3-3	G-G-T	11.8
Hap3-4	G-A-G	2.5
(rs7804315*–rs4730972–rs7795646– rs11971312)
Hap4-1	C-T-G-A	54.3
Hap4-2	T-C-A-A	33.5
Hap4-3	C-T-A-G	11.1
(rs12673992*–rs7807447– rs10256493)
Hap5-1	G-C-T	33.2
Hap5-2	G-C-C	27.1
Hap5-3	A-C-T	24.3
Hap5-4	G-T-C	15.4
(rs2160010*–rs2538488– rs12706309–rs10085563)
Hap6-1	C-C-T-A	56.1
Hap6-2	T-C-G-G	27.2
Hap6-3	C-C-T-G	12.2
Hap6-4	C-T-T-A	4.5

^*Associated SNP.

Results of haplotypes with significant association with BP are presented in Table 5. Two haplotypes, one carrying rs2315314 on chromosome 6 and the other carrying rs2160010 on chromosome 7, showed significant association with both SBP and DBP after adjusting for multiple comparisons. In addition, 3 haplotypes carrying rs7744171, rs7804315, or rs12673992 remained significantly associated with DBP; and 2 of the rs13237260-residing haplotypes retained significant association with SBP based on adjusted empirical significance levels.

Table 5.

Association of Selected Haplotypes With BP

				Systolic BP			Diastolic BP
Haplotype Serial No.	* SNP	Haplotype	Frequency%,	χ2	Observed P	Empirical P†	χ2	Observed P	Empirical P†
Hap1-4	rs2315314	G-T-G-C-C-G	13.6	14.17	0.00020	0.00820	15.82	0.00007	0.00240
Hap2-3	rs7744171	A-C-A	16.8	2.89	0.08930	0.84543	12.12	0.00050	0.01460
Hap3-1	rs13237260	A-A-G	48.4	10.97	0.00090	0.02260	7.54	0.00600	0.13137
Hap3-2	rs13237260	G-A-T	37.1	11.94	0.00050	0.01420	7.34	0.00680	0.14537
Hap4-2	rs7804315	T-C-A-A	33.5	8.23	0.00410	0.10458	11.93	0.00060	0.01880
Hap5-3	rs12673992	A-C-T	24.3	7.99	0.00470	0.10098	11.96	0.00050	0.01460
Hap6-2	rs2160010	T-C-G-G	27.2	10.22	0.00140	0.03579	17.22	0.00003	0.00160

^*BP-associated SNP residing on haplotype.

^†Based on 5000 permutations, this accounts for multiple comparisons within each phenotype.

Discussion

We report here findings from a family-based total association analysis of BP and SNPs selected from regions with prior evidence of linkage to BP in this sample of an African population from southwest Nigeria. The SNPs used in this report were selected to provide higher coverage density across the linked regions than was achieved with the set of sparsely distributed microsatellite markers with which the linkage evidence was documented. We note that the resolution provided by the number of SNPs in this study and the opportunity of treating BP as continuous phenotype in a family-based total association helped to conserve statistical power that would otherwise be impaired when BP is treated as categorical than as continuous phenotype.

The evidence presented here suggests that at least 7 loci lying on chromosomes 6 and 7 are likely to contribute to the linkage results observed previously in this set of families. One of the BP-associated SNPs—rs2315314 (which also accounts for the observed association at rs16892620)—has an “at-risk” allele frequency of almost 13% and is an intronic SNP in the PARK2 gene on 6q25.2-q27, linked with autosomal recessive juvenile Parkinsonism.⁴¹ This SNP is also located in the region detected by admixture mapping for hypertension in an African-American population,⁴² whether they reflect the same evidence requires further studies. Follow-up haplotype analysis revealed stronger association between the haplotypes on which these SNPs reside and BP. Other loci that also demonstrated significant association with DBP included rs7804315 and rs12673992 on 7q31. These 2 loci map to the KCND2 gene (potassium voltage-gated channel, Shal-related subfamily, member 2) whose functions include, among others, regulating epithelial electrolyte transport and heart rate.⁴³ Loci rs2160010 and rs12706309 that also demonstrated significant association with DBP are both located on 7q31.31, which maps to chromosome 7 open reading frame 58 (C7orf58), which codes for a hypothetical protein.⁴⁴ Two other genes to which associated SNPs map to are the glucocorticoid induced transcript 1 (GLCCI1) gene (rs13237260 and rs2068637) and histone deacetylase 9 (HDAC9) gene (rs11505418). After control for multiple comparisons, a number of the hyplotypes on which these SNPs reside remained significantly association with BP for the above individual SNPs and their related haplotypes (P<0.04 to 0.0016).

On the basis of the observed association in these regions where prior linkage evidence occurred, we infer that these regions may harbor BP-linked loci. Of course, further independent investigation of these regions would be required, and demonstration of a functional role of the SNPs located in these regions before a possible mechanistic hypothesis could be advanced. It is relevant to note that majority of the associated SNPs reported in this study are intronic and their minor allele frequencies are similar to those of HapMap YRI samples.⁴⁵

We note that the prior likelihood of a “true” association is lower in our study than would be the case in typical candidate gene studies, because we were following up strong linkage evidence. Application of a novel permutation procedure to determine significance levels of the quantitative trait family-based total associations which controlled for relatedness reported in the study provided strong statistical support for the specific association findings that were identified.

Identification of susceptibility loci through family-based linkage analysis has been a difficult challenge. For fully complex traits no clear-cut successes have yet emerged. On the basis of the experience to date with genome-wide association studies (GWAS) the fundamental problem is likely to be the small effect size.^1–8 From a statistical perspective single common variants with weak effects should not be reliably detected with linkage. Therefore, either the linkage signals that we detected reflect multiple different mutations in the same locus or, of course, they could be false-positives. Given these uncertainties our subsequent association finding must necessarily be interpreted with some degree of caution. Some technical limitations of this study must also be recognized. The marker set used for this analysis was somewhat sparse, tagging ≈60% of the variation defined by the YRI samples in Phase II HapMap.⁴⁵ Although it has generally been assumed that the HapMap provided a reasonable guide to coverage with tagging SNPs in this Yoruba population a recent analysis based on extensive resequencing suggests this assumption may not be correct.⁴⁶ Stronger evidence at these loci, or identification of entirely different loci, might have emerged with denser markers.

The immediate challenge for these findings is the identification of an appropriate replication sample. With the notable exception of the fat mass and obesity-associated locus for obesity, most loci associated with complex traits have similar effects across populations, although allele frequencies may vary.⁴⁷ Furthermore, although gene×environment effects are widely assumed to play a role, their specific relevance for SNPs associated with complex diseases has not been convincingly documented.^48–51 Thus, while replication in the same population would be ideal, the putative loci we have reported should be detectable in other populations. The large data resources currently being assembled as part of population-based GWAS will be the most efficient source of replication.

Perspective

Progress toward an understanding of the genetics of hypertension remains very limited. Rapid shifts in genotyping technology have led sequentially to an emphasis on candidate genes, family-based linkage analysis, and genome-wide association studies. In the end it is likely that information from all approaches will need to be combined. We report here association with several markers on chromosomes 6 and 7 in a Nigerian family set which seem to confirm prior linkage evidence. These results form part of the knowledge base that can be pooled to determine replication and consistency of the association with elevated BP.