Genetic contribution of SCARB1 variants to lipid traits in African Blacks: a candidate gene association study

The present study was carried out on 788 African Black subjects from Benin City, Nigeria, who were recruited as part of a population-based epidemiological study on CHD risk factors. Detailed information on the study design and population description is provided elsewhere [41]. In brief, 788 recruited subjects were healthy civil servants (37.18 % females) from three government ministries of the Edo state in Benin City, Nigeria, aged between 19 and 70 years, including 464 junior staff (non-professional staff with salary grades 1–6), and 324 senior staff (professional and administrative staff with salary grades 7–16). The summary features, including biometric and quantitative data of the entire sample of 788 subjects are given in Table 1 and Additional file 1: Table S1.

For resequencing, 95 individuals with extreme HDL-C levels (within the upper and lower 10th percentiles of HDL-C distribution) were chosen from the entire sample of 788 African Blacks. Resequencing sample comprised of 48 individuals with high HDL-C levels (≥90th percentile, range 68.30–99.00 mg/dL; Table 1) and 47 individuals with low HDL-C levels (≤10th percentile, range 10.30–35.00 mg/dL; Table 1). The University of Pittsburgh Institutional Review Board approved the study protocol. All participants gave their informed consent.

At least 8-hour fasting blood samples were collected from all participants. Serum specimens were separated by centrifugation of blood samples and then stored at −70 °C for 6–12 months until ready for testing. Lipid and apolipoprotein measurements included total cholesterol, HDL-C, TG, ApoA-I, and ApoB and were done with standard assays at the Heinz Nutrition Laboratory, University of Pittsburgh under the Centers for Disease Control Lipid Standardization Program [41]. LDL-C was calculated with the Friedewald equation [42] when TG levels were less than 400 mg/dL.

Genomic DNA was isolated from clotted blood using the standard DNA extraction procedure. All 13 SCARB1 exons (isoform 1, NM_005505), exon-intron boundaries, and 1 kb of each of 5′ and 3′ flanking regions on chromosome 12 (hg19, chr12: 125,262,175-125,348,519) were polymerase chain reaction (PCR) amplified and sequenced. Specific primers were designed using the Primer3 software (Whitehead Institute for Biomedical Research, http://​bioinfo.​ut.​ee/​primer3-0.​4.​0/​) to cover 13 target regions, resulting in 14 amplicons, including two overlapping amplicons for the largest last exon 13. PCR reaction and cycling conditions are available upon request. The primer sequences and amplicon sizes are given in Additional file 2: Table S2.

Automated DNA sequencing of PCR products was performed in a commercial lab (Beckman Coulter Genomics, Danvers, MA, USA) using Sanger method and ABI 3730XL DNA Analyzers (Applied Biosystems, Waltham, MA, USA). Variant analysis was performed using Variant Reporter (version 1.0, Applied Biosystems, Waltham, MA, USA) and Sequencher (version 4.8, Gene Codes Corporation, Ann Arbor, MI, USA) software in our laboratory.

Of 83 variants identified in the discovery step (see Additional file 3: Table S3, Additional file 4: Table S4, Additional file 5: Figure S1, and Additional file 6: Figure S2), 78 (28 with MAF ≥5 % and 50 with MAF <5 %) were selected based on the pairwise linkage disequilibrium (LD) and Tagger analysis using an r ² threshold of 0.90 (5 were excluded due to high LD) in Haploview (Broad Institute of MIT and Harvard, https://​www.​broadinstitute.​org/​scientific-community/​science/​programs/​medical-and-population-genetics/​haploview/​haploview) [43] for follow-up genotyping in the entire sample (n = 788). Since our sequencing was focused primarily on coding regions, in addition we selected 69 HapMap tag single nucleotide polymorphisms [SNPs] (out of total 108 HapMap tagSNPs; see Additional file 7: Table S5 and Additional file 8: Figure S3) based on Tagger analysis (MAF ≥5 % and r ²  ≥ 0.80) of HapMap data (Release #27) from the Yoruba people of Ibadan, Nigeria (YRI), in order to cover the entire gene for common genetic variation information. Moreover, we selected two SCARB1 variants previously reported to be significantly associated with lipid traits in the literature (Additional file 9: Table S6). Conclusively, a total of 149 variants, comprising of 78 sequence variants, 69 common HapMap-YRI tagSNPs, and two relevant associated variants, were selected for follow-up genotyping.

Genotyping of selected variants in the total sample of 788 individuals was performed by using either iPLEX Gold (Sequenom, Inc., San Diego, CA, USA) or TaqMan (Applied Biosystems, Waltham, MA, USA) methods and following the manufacturers’ protocols.

Out of 149 selected variants, two failed assay designs and nine failed genotyping runs (see details in Additional file 3: Table S3, Additional file 7: Table S5, and Additional file 9: Table S6). Quality control (QC) measures for successfully genotyped variants were as follow: a genotype call rate of ≥90 %, a discrepancy rate of <1 in 10 % replicates, and no deviation from Hardy-Weinberg equilibrium [HWE] (P >3.62 × 10⁻⁴ after Bonferroni correction). Ultimately, a total of 137 QC-passed genotyped variants were included in genetic association analyses (see Additional file 9: Table S6, Additional file 10: Table S7, Additional file 11: Figure S4, and Additional file 12: Figure S5).

We used the Haploview program to determine allele frequencies, to test HWE for genotype distribution, and to evaluate the LD and pairwise correlations (r²) between variants [43].

The values of each lipid phenotype outside the mean ± 3.5 standard deviation (SD) were excluded from downstream gene-based, single-site, and haplotype analyses. However, the extreme phenotypic values associated with rare variants (MAF ≤1 %) were maintained during rare variant analysis, as was the case for the p70201/chr12:125279319 variant (see study workflow in Fig. 1). Values of the five lipid and apolipoprotein traits—HDL-C, LDL-C, TG, ApoA-I, and ApoB—were transformed using the Box-Cox transformation. For each trait, we used stepwise regression method to select the most parsimonious set of covariates from the following list: sex, age, body mass index, waist, current smoking (yes/no), minutes of walking or biking to work each day (jobmin), and occupational status (staff: junior [non-professional staff]/senior [professional and administrative staff]). Genetic association analyses, including gene-based, single-site, LoF/rare variant, and haplotype association tests, were performed using linear regression models that included significant covariates for each variable (Additional file 13: Table S8).

The gene-based association analysis was conducted under linear additive model for the combined evaluation of common and LoF/rare variants (n = 136, excluding p70201/chr12:125279319; see details above in paragraph two of this section) for five major lipid traits using the versatile gene-based association study [VEGAS] (http://​gump.​qimr.​edu.​au/​VEGAS/​) software [44]. The significance threshold for the gene-based test was set at P-value of 0.05.

Following gene-based analysis, which primarily implicated SCARB1 in regulation of HDL-C and ApoA-I levels, we further elucidated the association of SCARB1 variants with these two traits using additional tests. In single-site association analysis, P-values for each trait were adjusted for multiple testing using Benjamini-Hochberg procedure [45] to determine the false discovery rate [FDR] (q-value). For common variants (MAF ≥5 %), a nominal P-value of <0.05 was considered to be suggestive evidence of association, and an FDR cut-off of 0.20 was used to define statistical significance. For LoF/rare variants (MAF <5 %), the single-site association results were interpreted separately because of inadequate power of our study to detect individual statistical significance for these variants.

We conducted an optimal sequence kernel association test (SKAT-O) [46] to evaluate the association between a total of 43 LoF/rare variants (MAF <5 %) and the two lipid traits (HDL-C and ApoA-I) by using three different MAF thresholds: <5 % (n = 43), ≤2 % (n = 26), and ≤1 % (n = 23). A significant SKAT-O test was set at a P-value of <0.05.

Haplotype association analysis was performed using the generalized linear model. We applied a fixed sliding window approach that included four variants per window and sliding for one variant at a time. For each window, a global P-value was used to assess the association between the haplotypes with frequency >1 % and a given trait. A global P-value threshold of 0.05 was used to define significant haplotype association.

All analyses, except for VEGAS, were performed using the R statistical software (http://​www.​r-project.​org/​) and relevant R packages (i.e., Haplo.Stats for haplotype analysis and SKAT for SKAT-O analysis).

Resequencing of SCARB1 exons and exon-intron boundaries plus flanking regions in 95 African Blacks with extreme HDL-C levels identified 83 variants, of which 51 had MAF <5 % (Additional file 3: Table S3 and Additional file 5: Figure S1). The majority of 83 variants (n = 73) were previously identified (dbSNP build 139: GRCh37.p10). Most variants (n = 80) were singlenucleotide variations [SNVs] (67 transitions and 13 transversions); the rest (n = 3) were short insertion and deletion variations (indels).

Tagger analysis using an r ² cutoff of 0.9 identified 28 bins for 32 common variants (MAF ≥5 %), of which three included more than one variant (r² ranging from 0.95 to 1.0) (Additional file 6: Figure S2). One of these three bins contained two variants (rs204901986 and rs34339961) in complete LD (r² = 1.0). Of 51 LoF/rare variants (MAF between 1 and 5 %, n = 31; MAF ≤1 %, n = 20), 17 were present only in the high HDL-C group (MAF ranging between 0.010 and 0.042) and eight were observed only in the low HDL-C group (MAF ranging between 0.011 and 0.033). In the high HDL-C group, 29 of 48 (~60 %) individuals cumulatively carried at least one LoF/rare variant, ranging from 1 to 7 variants. Similarly, in the low HDL-C group, 27 of 47 (~57 %) individuals carried at least one LoF/rare variant, ranging from 1 to 9 variants.

Of the total eight coding variants observed, four were common variants (rs2070242 [p.Ser4Ser], rs10396208 [p.Cys21Cys], and rs5888 [p.Ala350Ala], and rs701103 [p.Gly499Arg]—3′ untranslated region [UTR] in isoform 1 and exon 13 in isoform 2), and the remaining four were LoF/rare variants (rs4238001 [p.Gly2Ser], rs5891 [p.Val135Ile], rs5892 [p.Phe301Phe], and rs141545424 [p.Gly501Gly]). Of note, two LoF/rare coding variants, (rs5891 [p.Val135Ile] and rs141545424 [p.Gly501Gly]), were found only in the high HDL-C group.

Fifteen variants were located in either UTRs (n = 5) or flanking regions (n = 10). One 3′ UTR variant (rs150512235, MAF = 0.006) was very close to a predicted microRNA-145 (miR-145) target site (TargetScanHuman version 6.2, http://​www.​targetscan.​org/​). One 5′ flanking variant (rs181338950, MAF = 0.048) was located in the putative promoter region [47].

All 10 novel variants (9 SNVs and 1 insertion) identified in this study have been submitted to dbSNP database ([batch ID: SCARB1_AB]:

http://​www.​ncbi.​nlm.​nih.​gov/​SNP/​snp_​viewTable.​cgi?​handle=​KAMBOH) and were non-coding with MAF <5 % (ranging between 0.005 and 0.011; Additional file 4: Table S4). Of these novel variants, six and four were present only in the high and low HDL-C groups, respectively.

Since our sequencing was focused primarily on coding regions, we selected additional HapMap tagSNPs from the HapMap-YRI data in order to cover the entire SCARB1 gene for common genetic variation in SCARB1. Altogether we selected 149 variants for genotyping in our entire African Black sample as follows: 78 variants (28 common variants and 50 LoF/rare variants) discovered in the sequencing step (Additional file 3: Table S3, Additional file 5: Figure S1, and Additional file 6: Figure S2), 69 common HapMap-YRI tagSNPs (Additional file 7: Table S5), and two additional variants with reported association in the literature (Additional file 9: Table S6).

Of these 149 variants, 11 (10 from sequencing, including one promoter [rs181338950], one coding (rs4238001 [p.Gly2Ser]), and one novel [p87459/chr12:125262061], and 1 from HapMap tagSNPs [rs4765180]) failed genotyping, and one (rs866793 from HapMap tagSNPs) failed QC measures. Thus, a total of 137 variants (Additional file 9: Table S6 and Additional file 11: Figure S4) that passed QC were advanced into association analyses with five lipoprotein-lipid traits.

The majority of 137 genotyped variants (n = 120) were located in introns, 11 were in exons, and six were in 3′ flanking region (Table 2 and Additional file 12: Figure S5). Ninety-four of 137 variants had MAF ≥5 %, including four coding variants, one UTR variant, two deletions, and one splice site variant. The remaining 43 variants had MAF <5 % (MAF between 1 and 5 %, n = 20; MAF ≤1 %, n = 23), including three coding variants, three UTR variants, one insertion, and one splice variant.

Of the 10 novel variants discovered in the sequencing step, nine (8 SNVs and 1 insertion) with MAF <1 % were successfully genotyped (Additional file 4: Table S4). There was one individual with plasma HDL-C levels above the mean + 3.5 SD carrying one novel variant—p70201/chr12:125279319 (MAF = 0.0010). Although this extreme HDL-C value was excluded as outlier from the gene-based, single-site, and haplotype analyses, it was included in the SKAT-O rare variant analysis considering a possible large effect size of this variant (Fig. 1).

Gene-based tests revealed a nominally significant association (P = 0.0421; Table 3) of SCARB1 variants with HDL-C levels (best SNP: rs141545424 [p.Gly501Gly], exon 12, MAF = 0.0007, P = 0.0016). Additionally, a trend for association (P = 0.1016) was also observed for ApoA-I levels (best SNP: rs7134858, intron 6, MAF = 0.1560, P = 0.0052).

Of 94 common SCARB1 variants with MAF ≥5 %, 10 showed nominal associations (P < 0.05) with HDL-C and/or ApoA-I (Table 4; see results for each trait in Additional file 14: Table S9 and Additional file 15: Table S10), of which three (rs11057851, rs4765615, and rs838895) exhibited associations with both HDL-C and ApoA-I.

The most significant association was found between rs11057851 and HDL-C (β = −0.5924, P = 0.0043, FDR = 0.1465). The second best association was between rs7134858 and ApoA-I (β = 1.7537, P = 0.0052, FDR = 0.2918), followed by the association of rs5888 (p.Ala350Ala) with ApoA-I (β = 2.0962, P = 0.0080, FDR = 0.2918).

Of 10 variants that showed nominal associations, high LD (r² > 0.80) was observed for two pairs of variants (Fig. 2), between rs8388912 and rs5888 (p.Ala350Ala; r² = 0.86), and between rs838896 and rs838895 (r² = 0.84).

The LoF/rare variants (n = 43) were categorized into three groups based on their frequencies for association analysis with HDL-C and ApoA-I using SKAT-O: MAF <5 % (n = 43), MAF ≤2 % (n = 26), and MAF ≤1 % (n = 23). Although no association between LoF/rare variants and ApoA-I was detected, the group of 23 variants with MAF ≤1 % yielded nominal association with HDL-C levels (P = 0.0478; Table 5).

We then individually examined the association of 23 variants with MAF ≤1 % with HDL-C and ApoA-I. Six of these rare variants showed association with either HDL-C levels or both HDL-C and ApoA-I levels (Table 6). While three of them are known variants (rs115604379, rs377124254, and rs141545424 [p.Gly501Gly]), the other three are novel (p52919/chr12:125296601, p54611/chr12:125294909, and p54856/chr12:125294664). Moreover, four of these six rare variants (rs377124254, rs141545424 [p.Gly501Gly], p54611/chr12:125294909, and p54856/chr12:125294664) were present in individuals with extreme phenotypic values (above or below the 3rd percentile). Two of these variants (rs377124254: β = 11.5518, P = 0.0016; rs141545424 [p.Gly501Gly]: β = 11.585, P = 0.0016) were found in a single subject who had very high HDL-C level. Whereas the other two were observed in one individual each, who had extremely low HDL-C levels (p54611/chr12:125294909: β = −9.5243, P = 0.0097; p54856/chr12:125294664: β = −8.4305, P = 0.0215) and ApoA-I levels (p54611/chr12:125294909: β = −19.3821, P = 0.0344; p54856/chr12:125294664: β = −24.0757, P = 0.0082). This rare variant group also included a novel variant (p70201/chr12:125279319) that was observed in one individual with an unusually high plasma HDL-C level (above the mean + 3.5 SD).

The 4-SNP sliding window haplotype analyses revealed associations of 32 haplotype windows with HDL-C and/or ApoA-I (global P < 0.05; Table 7; see results for each trait in Additional file 16: Table S11), of which five (windows #47, #72, #111, #112, and #123) were associated with both.