Study design

In this research, we conducted a two-sample MR investigation, utilizing genome-wide association study (GWAS) data sourced from publicly accessible repositories. MR is an epidemiological approach that relies on instrumental variable analysis. It involves the use of genetic variants, commonly single-nucleotide polymorphisms (SNPs), linked to a modifiable risk factor such as high blood pressure to make inferences about causality. By leveraging these genetic markers as proxy indicators, MR helps minimize bias arising from confounding factors since alleles are randomly inherited at conception. Moreover, it also avoids bias due to reverse causation, as the occurrence of a disease cannot influence an individual’s genotype [6].

Data sources

Exposure: We extracted SNPs as instrumental variables (IVs) associated with Lp(a) concentration from GWAS data downloaded from Neale lab (UK Biobank). Because the measured Lp(a) concentration was positively skewed, inverse rank-normalized data were used. European ancestry female and male individuals were included in this research (N = 361,194) [4].

Outcome

Summary-level data for T2D were obtained from the Finnish FinnGen consortium [7, 8]. We used data from the eighth version of the database, which included 17,268 cases and 184,778 controls, with an analysis covering 16,380,418 variants.

MR analyses

P < 5 × 10^–8 was accepted as a genome-wide significance threshold. To reduce the risk of any potential weak instrument bias, F-statistics were calculated based on the formula F = (β/SE)² [6], where ‘β’ and ‘SE’ refer to the genetic association of SNPs with the exposure and its standard error, respectively. Only the SNPs with F-statistics > 10 were considered potential IVs. To ensure the independence of IVs, SNP in the linkage of disequilibrium (LD) were excluded. The TwoSampleMR R package was used to clamp data with a threshold r² < 0.001 [9]. If selected SNPs were unavailable in the outcome dataset, they were replaced with proxies in LD of r² > 0.8 or excluded from further MR analysis. Later, for selected SNPs, the potential association with confounding or risk factors for T2D were evaluated using PhenoScanner V2 [10]. Next, variant harmonization was conducted using the TwoSampleMR package between datasets to confirm that the association between SNPs and exposure and between SNPs and the outcome reflected the same allele. To further validate our results, aside from the aforementioned method of SNP selection, we also analyzed variants listed by ESC guidelines concerning Lp(a) including a subset of loss of function variants [11].

As the main analysis for evaluating the causal effect estimates in our study, we used the inverse variance weighted (IVW) method [12]. Sensitivity analyses were performed using MR-Egger, weighted median, and weighted mode approaches [13–15].

Additionally, to detect the existence of heterogeneity, horizontal pleiotropy, and outlier SNPs we performed several statistical tests. Cochran’s Q test was used to measure the heterogeneity between variant-specific causal estimates (IVW and MR-Egger regression) [16]. The MR-Egger intercept was calculated to measure the presence of horizontal pleiotropy [17]. To detect the potential presence of pleiotropic outlier SNPs, the MR-pleiotropy residual sum and outlier (MR-PRESSO) tests were applied [17]. Finally, leave-one-out analyses were performed to check whether any of the analyzed SNPs were strongly associated with the exposure, which may dominate the estimate of the causal effect.

Statistical analysis

All statistical data analyses were conducted with R software (version 4.1.1) using the TwoSampleMR (version 0.5.6) and MR-PRESSO (version 1.0) packages. P-values < 0.05 were considered statistically significant for all tests.

Ethics statement

Ethical approval for this study was not required as our analyses were based on summary statistics from published GWAS, or the data were publicly accessible and no individual-level data were used.