The public available GWAS summary statistics used in this study were summarized in Table 1. Because all data came from published GWASs, no additional ethical approval from an institutional review board was needed for this study.
Table 1
Data sources used in this study
Phenotype | Sample size (total or cases/controls) | Ancestry | Sources |
Exposures | | | |
Cardiac troponin I concentrations | 48,115 | European | HUNT and GS:SFHS[13] |
Outcomes | | | |
Any stroke | 40,585 cases/406,111 controls | European | MEGASTROKE Consortium[16] |
Ischemic stroke | 34,217 cases/406,111 controls | European | MEGASTROKE Consortium[16] |
Cardioembolic stroke | 7,193 cases/406,111 controls | European | MEGASTROKE Consortium[16] |
Large artery stroke | 4,373 cases/406,111 controls | European | MEGASTROKE Consortium[16] |
Small vessel stroke | 5,386 cases/406,111 controls | European | MEGASTROKE Consortium[16] |
Intracerebral hemorrhage | 1,545 cases/1,484 controls | European | ISGC meta-analysis[18] |
Atrial fibrillation | 60,620 cases/970,216 controls | European | GWAS meta-analysis of 6 studies[19] |
GWAS, genome-wide association study; HUNT, Trøndelag Health Study; GS:SFHS, Generation Scotland Scottish Family Health Study; ISGC: International Stroke Genetics Consortium. |
Genetic Instrument For Circulating Cardiac Troponin I Concentrations
Instrumental variables for circulating cTnI concentrations were selected from a GWAS meta-analysis of the Trøndelag Health Study ((HUNT, n = 29,839) and the Generation Scotland Scottish Family Health Study (GS:SFHS, n = 18,276) with 48,115 European individuals.[17] This GWAS identified 12 genome-wide significant (P < 5×10− 8) SNPs independently associated with cTnI concentrations (Supplementary Table 1). These 12 SNPs explained 1.2% phenotypic variation of circulating cTnI concentrations.[17]
Outcome Data Sources
Genetic association estimates for stroke, IS and IS subtypes were derived from the MEGASTROKE consortium.[14] Briefly, we only used European GWAS summary statistics for stroke (40,585 cases and 406,111 controls), IS (34,217 cases and 406,111 controls) and IS subtypes (7,193 CES cases, 4,373 LAS cases, and 5,386 SVS cases). The IS subtypes were based on the the Trial of Org 10172 in Acute Stroke Treatment criteria (TOAST).[1]
For ICH, we used the summary statistics from the International Stroke Genetics Consortium (ISGC) meta-analysis including 1,545 cases and 1,481 controls.[24]
For AF, we used a GWAS meta-analysis of 6 studies (The Nord-Trøndelag Health Study, deCODE, the Michigan Genomics Initiative, DiscovEHR, UK Biobank, and the AFGen Consortium) including 60,620 cases and 970,216 controls of European ancestry.[19]
For the SNPs were not available in the outcome dataset, the proxy SNPs (r2 > 0.9) were used to replace them by using the LDlink (http://ldlink.nci.nih.gov)[13] according to the CEU population data from 1,000 Genomes Project.
Statistical analysis
The primary MR analysis was performed by using inverse-variance weighted (IVW) method,[6] which assumes that all genetic variants are valid genetic instruments (e.g., no directional pleiotropy). Given that the IVW estimate may be biased by the directional pleiotropy, weighted median[4] and MR-Egger[3] methods were also used to test the robustness of the IVW estimate. The weighted median method can provide a consistent estimate if at least 50% of the instrumental variables are valid. The MR-Egger approach can detect and adjust for the bias due to directional pleiotropy.
Several methods were conducted to evaluate the potential pleiotropy. First, we used Mendelian Randomization Pleiotropy Residual Sum and Outlier (MR-PRESSO)[23] method to detect potential pleiotropic outlier (P < 0.1), and the IVW estimate was re-calculated if the pleiotropic outlier was identified. Second, the heterogeneity across different SNPs was assessed by using the Cochran’s Q statistics (P < 0.05 indicates significant). Third, we performed MR-Egger intercept test to examine the directional pleiotropy, and the significant MR-Egger intercept test (P < 0.05) indicates the presence of directional pleiotropy. Finally, leave-one-out (LOO) analysis was used to test whether a single SNP drive the association.
Odds ratios (ORs) with 95% confidence intervals (95% CIs) were scaled to 1- standard deviation (SD) increment of inverse rank transformed circulating cTnI concentrations. A Bonferroni corrected P value < 0.007 (correcting for 7 outcomes) was considered as statistically significant. All statistical analyses were performed using the Mendelian randomization,[25] TwoSampleMR,[10] and MR-PRESSO packages[23] for R version 4.0.2.