2.1. Data Resource
Since type 1 and type 2 diabetes cover the majority of diabetic patients, type 1 and type 2 diabetes were selected as diabetes phenotypes in this study for further MR analysis1.
Ankylosing spondylitis (n = 273,824; 2,860 cases and 270,964 controls), T1DM (n = 264,137; 8,671 cases and 255,466 controls), and T2DM (n=49,303 cases and 255,466 controls) were obtained from the FinnGen Research Project.
Sample overlap between AS and DM did not exceed 1% from the FinnGen Research Project’s June 1, 2022 release (0.44% overlap between T1DM and AS samples and 0.69% overlap between T2DM and AS samples) 13, therefore, the samples included in this study were considered independent sample.
Table 1 Description of contributing studies.
Trait
|
Ancestry
|
Release Date
|
SNPs
|
Sample size
|
Consortium
|
AS
|
European
|
May 11 2023
|
20,166,920
|
273,824
|
FinnGen
|
T1DM
|
European
|
June 1 2022
|
16,383,236
|
264,137
|
FinnGen
|
T2DM
|
European
|
June 1 2022
|
16,383,311
|
304,769
|
FinnGen
|
AS, ankylosing spondylitis; SNPs: single nucleotide polymorphisms. T1DM, type 1 diabetes mellitus; T2DM, type 2 diabetes mellitus.
2.2. Selection of instrumental variables.
Three conditions need to be met for plausible MR results: (1) the selected IVs are strongly correlated with exposure. (2) the extracted IVs are not correlated with the outcome. (3) the IVs are not correlated with the outcome related confounders.
We selected the IVs according to the following order and criteria: (1) First, we set the GWAS correlation p-value < 5 × 10-8 ,linkage disequilibrium r2 <0.001 and 1-Mb distance from other IVs with a stronger association;(2) Then we performed the calculation of the F-value of individual SNPs to avoid weakly correlated SNPs affecting the final results14,and sensitive SNPs were extracted for further analysis based on F ≥10 (F value calculation formula and results are presented in the Supplementary Material) 15. SNPs meeting the above two conditions were identified as strongly correlated with exposure; (3) Coordinate the extracted SNPs with SNPs with the same target ending for base orientation, and remove SNPs with inconsistent base orientation and those whose orientation cannot be determined;(4)Remove SNPs that are strongly associated with the outcome (pval.outcome<5e-8); (5)The remaining SNPs were screened at the PhenoScanner website (www.phenoscanner.medschl.cam.ac.uk, a database of human genotype-phenotype associations) to see if these SNPs were associated with potential risk of outcome related factors and remove SNPs associated with these potential confounders16. potential risk factors associated with AS include inflammatory bowel disease17, atopic dermatitis18, smoking. Risk factors associated with T1DM include atopic dermatitis18, life stress19,obesity20. Risk factors associated with type 2 diabetes include sedentary lifestyle, physical inactivity, smoking, alcohol consumption, obstructive sleep apnea21–25.
After the above five sessions we obtained the final SNPs for subsequent MR analysis (Final SNPs as well as the removed SNPs are presented in the Supplementary Material).
2.3. Mendelian randomization analyses and statics analysis.
We used inverse variance weighted (IVW) as the primary outcome, assuming that all information from genetic instruments was valid to provide the most accurate results, and when there was heterogeneity in the results (p<0.05), we used multiplicative random effects26. weighted median, MR Egger as additional supplements to the IVW results to verify the validity and robustness of the results. Weight median can provide accurate estimates when the valid instrumental variable exceeds 50 percent27. MR Egger was used to detect the presence of pleiotropy based on its intercept term, and also can provide support when the direction of the results is consistent with IVW28.
For the heterogeneity test, the Cochrane's Q test was used to assess the heterogeneity of individual causal effects and when p<0.05 suggested the presence of heterogeneity29. MR presso detects horizontal pleiotropy among the MR instruments and the outlier test corrects for horizontal pleiotropy via outlier removal30; Leave-one-out sensitivity analysis assesses the impact of individual instrumental variables on the direction of the results. Briefly, when there was no pleiotropy, the directional agreement between IVW and weight median with p<0.05 was identified as a positive result, and if there were outliers, the result of MR presso replaced the result of IVW.
All analyses were performed using TwoSampleMR(version 4.3.0) in the R package(version 0.5.7).