Additional Supporting Information may be found in the online version of this article.


Table S1A. Significant SNPs for lipids traits in BWHHS individual level data. Beta is standardised by Z score: Z = (beta – mean of beta) / SD of beta

Table S1B. Significant SNPs for ECG traits in BWHHS individual level data. Effect sizes are shown as beta estimates from linear regression models for increasing copy of the coded allele and are on the standard deviation scale

Table S2. Significant SNPs for ECG traits in meta-analysis

Table S3. Possible haplotype frequencies combination for the haplotypes simulation

Table S4. Multiplied t-test p-values for all 1000 Genomes SNPs. The Top hit means we assume the SNP as the top hit and running SSS-RAP for six BWHHS SNPs. The multiplied p value is a simple adjustment we applied which multiplied all the t-test p-values for six SNPs. The top hit with highest multiplied p-values is the best fit SNP as top hit.


Figure S1. An example of linear trend line for a single locus. The grey dots are phenotype values (QTc interval) for 2686 participations, the black dashed line is the linear trend line, the x axis is the genotypes of SNP rs4657139, y-axis is the normalised QTc interval (normalised by Z score). Assume the effect of homozygous mutant genotype (AA) is 2e. Meanwhile the effect of homozygous wild type genotype (TT) is 2b. The slope of the regression line (beta) will be e-b.

Figure S2. Relationship between r2 and beta estimate of the possible dependent SNP for a given major allele frequency of the top hit and observed beta for the top SNP. Different colors refer to different intervals of major allele frequency of the possible dependent SNP (called q1): red for 0.5< q1 ≤ 0.6, blue for 0.6< q1 ≤ 0.7, green for 0.7< q1 ≤ 0.8, yellow for 0.8< q1 ≤ 0.9 and grey for 0.9< q1 <1.

Figure S3. Comparison of model selection approaches, conditional analysis and SSRAP using BWHHS individual level data. All the methods and criteria used to select: (A) SCN5A loci most likely to be tightly associated with PR interval; (B), SCN5A loci most likely to be tightly associated with QRS duration; (C), LPL loci most likely to be tightly associated with a TG variant; (D), LPL loci most likely to be tightly associated with a HDL variant; (E) CETP loci most likely to be tightly associated with a HDL variant. For all the methods, cross (×) denotes inclusion in the best model.

Figure S4. Comparison between SSS-RAP and conditional analysis using meta-analysis data in ECG traits. Linkage disequilibrium and haplotype block structure were linked to related SNPs in A and B, SCN5A loci associated with PR interval and QTC interval, respectively. For both methods, cross (×) denotes inclusion in the best model.

Figure S5. The influence of MAF on the standard error of beta. (A) Relationship between the observed standard errors and the MAF. (B) Relationship between the simulated standard errors and the MAFs.

Figure S6. Comparison of the observed standard error to the simulated standard errors and the transformed standard errors in different r2 region.

Figure S7. The three SNPs model for the admixed population simulations. We assume rs328 was the top hit, rs327 was an independent effect SNP and rs263 was a bystander SNP. Alleles in red means they were the effect/minor alleles.

Figure S8. A special case of the three SNPs model. (A) We used the same three SNPs model as showed in Figure S7. In this model, we assumed that rs327 (MAF = 0.4) and rs328 (MAF = 0.3) were two significant SNPs associated with a trail and there is no LD between them. In (B) we calculated the haplotype frequencies between rs327 and rs328. Then in (C and D), we introduced a previous hidden SNP, rs263 (unknown MAF = X/X-bar). Since we know the haplotypes of rs263, we can calculate the haplotype frequencies between rs327/rs328 and rs263 (values at the bottom of each grids). We then assumed there were some LD between rs263 and rs327/rs328 (D and D-bar were the LD measures). After the calculation, we got X, X-bar, D and D-bar. Interestingly, both X and X-bar was 0.28. This special case proves that when MAF of rs263 is 0.28, this SNP can be in LD to both rs327 and rs328 although there is no LD between rs327 and rs328.

Figure S9. Comparison of the observed betas to the expected betas, top hits were six BWHHS direct typed SNPs and 14 untyped SNPs from 1000 Genomes. X axis is the observed beta, y axis is the expected beta, each point is the expected beta divided by the observed beta for each BWHHS significant SNP, blue line is the best fit line for the 6 points, black line is the y = x line which means expected beta is the same as the observed beta.

Figure S10. An example of one “bystander” efficiently represents two separate causal sites themselves in LD. In Model 1, bystander C shows the most signal. However, the haplotype effect scores are not consistent with C being causal because not all C bearing haplotypes show the same effect score. On the other hand, effect scores A = B = 1 would be consistent with observed haplotype effects. In Model 2, C shows the most signal (same as in Model 1). In this model, the haplotype effect score are consistent with C being causal, whereas A and B would not be.

Please note: Wiley Blackwell is not responsible for the content or functionality of any supporting information supplied by the authors. Any queries (other than missing content) should be directed to the corresponding author for the article.