Additional file 2 of Sex and statin-related genetic associations at the PCSK9 gene locus: results of genome-wide association meta-analysis

Additional file 2: Figure S1. Flowchart of the stratified genome-wide analyses. We included the data of six studies of European descent. All participating studies provided GWAS summary statistics stratified for both statin-treatment and sex. In the first round of meta-analyses, we combined the study-wise data for the double-stratified subgroups. In the second round of meta-analyses, we combined pairwise strata to estimate the single-strata SNP effects, for example statin-free and statin-treated females combined to estimate SNP effects in females. Associated loci (p<1 x 10-6) were then tested for sex- and statin-interactions. All loci were annotated with candidate genes and tested for colocalization with gene expression and lipid data. For the PCSK9 gene locus, we performed fine-mapping using GCTA COJO. Finally, we compared the causal effects of PCSK9 on LDL-C using the subgroup-specific effect estimates of four SNPs at the PCSK9 gene locus. Figure S2. Manhattan Plot of all eight subgroups (min. p-value per SNP). The y-axis was limited to 20, and all SNPs with higher values set to 20 (max. original log10(p) = 143.8). Color indicates the subgroup with the lowest p-value for each SNP with log10(p) > 6. The 11 loci with sufficient support (3 or more associated SNPs) are labeled. The red dashed horizontal line indicates genome-wide significance (p<5 x 10−8), while the blue dotted line indicates suggestive significance (p<1x10−6). Figure S3. Regional association plots at PCSK9 gene. For each subgroup, an RA plot is given. In all plots, the lead SNP rs11591147 is plotted in blue. SNPs in LD with this variant are plotted in yellow (LD r2 ranging between 0.1 and 0.5). Independent variants as identified by GCTA COJO select are encircled in red. Figure S4. LD-Matrix plot generated by LDlink. We included all seven SNPs that were selected as independent signals in one of the subgroups to test their pairwise LD using the European reference set. The lower triangle in red indicates LD r2, while the upper triangle in blue indicates D‘. There are four LD-clusters visible, and their best-associated SNPs per cluster are rs2495491, rs11591147, rs11583680, and rs693668. LD between the clusters is low (r2 < 0.05), and LD within the cluster is high (r2 > 0.7). Figure S5. Forest Plots of the four independent SNPs over the eight subgroups. Each SNP and subgroup are plotted using the GWAS (unconditional) beta estimates and 95% confidence intervals (CI). Subgroups are sorted by increasing beta estimates per SNP (different sorting per SNP). A) rs11591147 (lead SNP) with significant statin-interaction B) rs693668 with sex-interaction C) rs11583680 with statin-interaction (males treated vs males free) D) rs2495491 without interaction. Figure S6. Regional Association Plot for novel loci. We detected five novel loci: NOS1/KSR2 (12q24.22) in males, KHDRBS2 (6q11.1) in statin-treated subjects, ALOX5 (10q11.21) in statin-free males, SLCO1B1 (12p12.2) in statin-free females, and PRKAG2 (7q36.1) in statin-free males. Figure S7. Forest Plots for the novel loci over the eight subgroups. Each SNP and subgroup are plotted using the GWAS beta estimates and 95% confidence intervals (CI). Subgroups are sorted by increasing beta estimates per SNP (different sorting per SNP). A) rs4767549 (NOS1/KSR2) with significant sex-interaction B) rs3076276 (KHDRBS2) with significant statin-interaction C) rs76849715 (ALOX5) with significant sex- and statin-interaction D) rs4763806 (SLCO1B1) with significant sex-interaction E) rs34924001 (PRKAG2) with significant sex- and statin-interaction. Figure S8. Regional Association Plot for known loci. We replicated five known PCSK9 loci in the statin-free subgroup. These are APOB (2p24.1), TM6SF2 (19p13.11), FADS1/2 (11q12.2), HP/HPR (16q22.2), and JMJD1C (10q21.3). Figure S9. Forest Plots for the known loci over the eight subgroups. Each SNP and subgroup are plotted using the GWAS beta estimates and 95% confidence intervals (CI). Subgroups are sorted by increasing beta estimates per SNP (different sorting per SNP). A) rs1367117 (APOB) with significant statin-interaction B) rs8107974 (TM6SF2) C) rs174535 (FADS1/2) D) rs34042070 (HP/HPR) E) rs10740131 (JMJD1C) with statin-interaction. Figure S10. Directed acyclic graph (DAG) for Mendelian Randomization (MR). We analyzed the causal effect of PCSK9 on LDL-C, stratified by sex and statin treatment. Statin treatment induces indirectly gene expression of LDLR and PCSK9. PCSK9 increases the degradation of LDLR and hence increases LDL-C plasma levels. Biological sex is a known risk factor for both PCSK9 and LDL-C. Figure S11. Forest Plot of the causal estimates per subgroup. In each subgroup and SNP, we estimated the Wald ration and used the first term of the delta method for the standard error. Then we combined the single SNP estimates in an inverse-variance-weighted (IVW) meta-analysis (fixed effect), and tested for heterogeneity leaving one SNP out (“w/o SNP x”). Throughout all subgroups, the causal estimate of rs11583680 is weaker than the other three introducing heterogeneity in the IVW analysis. A) Statin-free subjects B) Statin-treated subjects C) Males D) Females E) Statin-free males F) Statin-free females G) Statin-treated males H) Statin-treated females.