We sought to fine map and extend in African Americans the association of variation at FTO
with BMI. In this study, different fine-mapping approaches were taken in parallel in the Multiethnic Cohort (MEC; n
= 3482) and the Jackson Heart Study (JHS; n
= 4217) samples (see Materials and Methods; baseline characteristics of the samples are described in Table ), so the sets of SNPs genotyped in each sample are slightly different. In the JHS, we chose a set of SNPs designed to capture variation in African Americans across a 245 kb region (chr16: 52,181,324 to 52,406,062) which includes the 51 kb LD block in intron 1 of FTO
(HapMap CEU) containing the index signal (chr16: 52,355,066 to 52,406,062) as well as the neighboring RPGRIP1L
gene (Fig. ). In JHS, we focused on variants that are in LD with the associated SNPs in European ancestry reference panels (HapMap CEU) but not in LD with each other in an African-ancestry panel (HapMap YRI). In the MEC, fine mapping was limited to the defined LD block containing the index signal and common alleles in the YRI HapMap panel that had previously been genotyped in an African-American population and had not been associated with BMI (7
Characteristics of the African-derived populations
Figure 1. Linkage disequilibrium and haplotype block structure in intron 1 of FTO. (A) and (B) represent the CEU, and YRI populations from HapMap phase 2 data. Pairwise correlation among SNPs (r2) are shown. The black box highlights the region targeted for fine (more ...)
We did not replicate the initial observed association from studies conducted in European populations between BMI and the lead SNPs rs9939609 and rs8050136 for JHS and MEC, respectively (Table ). In the JHS sample, we observed strong evidence of association at rs3751812, and its proxy rs1558902 (r2
= 0.96 with rs3751812 in JHS; Table ). Slightly weaker associations were observed at two other proxies of rs3751812 (Table ): rs1421085 and rs17817964 (r2
= 0.95 and r2
= 0.94 in JHS with rs3751812, respectively). We also observed a strong association at rs9941349, a SNP with a higher minor allele frequency (MAF) but only in moderate LD with rs3751812 (r2
= 0.50 in JHS), and at rs9931494, a nearly perfect proxy for rs9941349 (r2
= 0.97 in JHS; Table ). The rs3751812 and rs9941349 SNPs are both strongly correlated with the lead SNPs in the HapMap CEU sample (r2
= 0.90–1.0) but not in the HapMap YRI sample (r2
= 0.02–0.07) or in JHS (r2
= 0.03–0.11) (19
). In the MEC, none of the SNPs initially genotyped were found to be associated with BMI after correction for multiple comparisons, but additional genotyping confirmed the association at rs3751812 as well as at rs9941349 (Table ).
Further testing in additional cohorts of largely African ancestry provided additional evidence for association with these SNPs (Table and Supplementary Material, Table S1
). A meta-analysis of the four cohorts in which BMI was analyzed as a continuous trait (JHS, MEC, Spanish Town, and Maywood) revealed a signal of association for rs3751812 (P
= 2.58 × 10−6
; Table ). A comparable association was seen for rs9941349 (P
= 3.61 × 10−6
; Table ). The estimated effect size per allele at rs3751812 was 0.0121 log BMI units and 0.0093 log BMI units for rs9941349. I2
values were 0% (P
= 0.70) for rs3751812 and 22.7% (P
= 0.27) for rs9941349 indicating low heterogeneity between the cohorts. Although the Jamaican Gene × Environment Panel (G×E) was not included in the overall meta-analysis because BMI was analyzed as a dichotomous trait in this case–control cohort, the results for rs3751812 and rs9941349 trended in the same directions in this sample (Table ).
Meta-analysis across studies with continuous traits
To help assess the effects of possible confounding by ancestry in these admixed populations, we estimated global African and European ancestry in both the JHS and a subset of the MEC, and global and local ancestry in JHS. The association evidence in the subset of individuals with ancestry estimates was not diminished by adjusting for ancestry, suggesting that the results were not false positives due to admixture (Supplementary Material, Table S2
). We also observed nominally significant association when using family-based tests of association in the subset of related samples in JHS for both rs3751812 and rs9941349 and in the family-based Maywood cohort for rs9941349 (rs3751812 approached significance in this sample) (Table and Supplementary Material, Table S1
). Furthermore, the effect in the African Americans is unlikely to be due to the presence of European chromosomes in this population, because in this case we would expect to see equal association among all of the variants that are strongly correlated in European populations, whereas in our data we saw strong differences in association among this group of variants. Thus, these data provide the strongest evidence to date that variation in the FTO
locus influences BMI in African-ancestry populations.
We next attempted to discern which of the two associated SNPs, rs3751812 and rs9941349, were more closely correlated with the biologically relevant variant at the locus (assuming that there is one such variant responsible for the observed association signal). We focused on these two SNPs because (i) they had the strong signals of association (ii), their associations with BMI remained significant when conditioning on other individual SNPs (except for rs3751812 or rs9941349 or their proxies) and (iii) other SNPs in the region were not strongly associated after conditioning on either of these two SNPs (Supplementary Material, Tables S3 and S4
). We performed conditional analyses of rs3751812 and rs9941349 in the JHS, MEC, Spanish Town (SPT), G×E and the Maywood cohort, testing for the association of each SNP when using the other SNP as a covariate (Supplementary Material, Table S3
). A meta-analysis of these conditional analyses in the samples with continuous traits (JHS, MEC, SPT and Maywood) showed that the association signal at SNP rs3751812 remained nominally significant after conditioning on the effect of SNP rs9941349 (overall P
= 0.0354; Table ). However, the association signal at rs9941349 was nearly significant when conditioning on rs3751812 (overall P
= 0.0678; Table ). We therefore do not feel that this analysis can demonstrate with confidence which of rs3751812 and rs9941349 are more tightly correlated with a true causal variant, and may even suggest that neither of these two variants are themselves causal.
Conditional analysis on rs3751812 and rs9941349
We also performed a haplotype analysis to try to assess which of these two SNPs might be more tightly correlated with a causal variant at the locus (see Materials and Methods). Because LD is strong between these two SNPs (D′ = 91–100), there are three major haplotypes. We designated the three major haplotypes as H–H, H–L and L–L, corresponding to the BMI-increasing (H) or decreasing (L) alleles at rs9941349 and rs3751812, respectively. The H–H, H–L and L–L haplotypes (T–T, T–G, and C–G) accounted for 9.6, 8.0 and 82.3% of haplotypes in the JHS population, 11.5, 7.7 and 80.5% of haplotypes in the MEC population, 7.7, 9.5 and 82.8% of haplotypes in the SPT population, 8.3, 9.0 and 82.7% of haplotypes in the G×E population and 7.2, 7.9 and 84.4% of haplotypes in the Maywood cohort. In this analysis, the H–L haplotype was associated with a higher BMI than the L–L haplotype (P = 0.009), suggestive of an effect of the rs9941349 SNP (or a variant in LD with this SNP) independent of rs3751812; the H–H and H–L haplotypes were not significantly different, although the data were directionally consistent with the H–H haplotype having a stronger effect on increasing BMI than the H–L haplotype (Table ). Thus, the conditional analysis showed a nominally significant effect of rs3751812, and the haplotype analysis showed a nominally significant effect of rs9941349; together these results suggest that neither variant has completely captured the signal of association at FTO, but are consistent with both variants being strongly correlated with a causal variant at this locus.
Test for differences between haplotypes