Immune system-wide Mendelian randomization and triangulation analyses support autoimmunity as a modifiable component in dementia-causing diseases

Abstract

Immune system and blood-brain barrier dysfunction are implicated in the development of Alzheimer's and other dementia-causing diseases, but their causal role remains unknown. We performed Mendelian randomization for 1,827 immune system- and blood-brain barrier-related biomarkers and identified 127 potential causal risk factors for dementia-causing diseases. Pathway analyses linked these biomarkers to amyloid-β, tau and α-synuclein pathways and to autoimmunity-related processes. A phenome-wide analysis using Mendelian randomization-based polygenic risk score in the FinnGen study (n = 339,233) for the biomarkers indicated shared genetic background for dementias and autoimmune diseases. This association was further supported by human leukocyte antigen analyses. In inverse-probability-weighted analyses that simulate randomized controlled drug trials in observational data, anti-inflammatory methotrexate treatment reduced the incidence of Alzheimer's disease in high-risk individuals (hazard ratio compared with no treatment, 0.64, 95% confidence interval 0.49-0.88, P = 0.005). These converging results from different lines of human research suggest that autoimmunity is a modifiable component in dementia-causing diseases.

Fig. 1. Design and rationale of six complementary studies.

To study BBB- and immune system-related biology, biomarkers and drug targets for dementia-causing diseases, we conducted six separate studies. Study 1 used MR and MR-Base database to explore how BBB and immune system-related biomarkers associate with dementia-causing diseases . This hypothesis-generating study identified 127 biomarkers associated with dementia-causing diseases, many related to BBB, inflammation and self-tolerance, suggesting that inflammatory and autoimmune processes may play a role in these diseases. Study 2 is a pathway analysis on the associations of study 1. Providing additional support for the autoimmune hypothesis, the analysis showed that the biomarkers are enriched in several autoimmune-related biological processes and share pathways with amyloid-β, tau and α-synuclein proteins that characterize dementia-causing diseases. Study 3 examined the eight proteins that have protein quantitative loci near the APOE gene. In line with the autoimmune hypothesis, this study showed that IFIT2, an anti-inflammatory protein, decreases risk for dementia-causing diseases independent of APOE. Study 4 examined which diseases are associated with a polygenic risk score constructed from SNPs associated with the 127 biomarkers. Using phenome-wide analysis, this study showed that several autoimmune diseases, especially type 1 diabetes and rheumatic arthritis, share a genetic background with dementia-causing diseases. Study 5 provided further support for the autoimmune hypothesis by identifying nine HLA alleles associated with dementia-causing diseases. Study 6 used IPW analyses to simulate randomized control trials in observational data. It examined whether the autoimmune component is modifiable with anti-inflammatory medication. These analyses showed that methotrexate and TNF-α inhibitors may be preventative medications for dementia-causing diseases.

Fig. 2. Biomarkers associated with general Alzheimer´s and Parkinson´s disease in Mendelian randomization analyses.

a,b, Odds ratios (ORs) and 95% confidence intervals (CIs) for an increase of 1 s.d. in biomarkers associated with general Alzheimer’s disease outcome (a) and Parkinson’s disease (b) in MR after FDR correction of 5% (P < 0.00052). ORs were derived from Wald ratios when only one SNP was available, and from IVW estimates when two or more SNPs were available. All tests are two-sided.

Fig. 3. Biomarkers associated with late-onset Alzheimer´s disease in Mendelian randomization analyses.

ORs and 95% CIs for an increase of 1 s.d. in biomarkers associated with late-onset Alzheimer’s disease in MR after FDR correction of 5% (P < 0.00052). ORs were derived from Wald ratios when only one SNP was available, and from IVW estimates when two or more SNPs were available. All tests are two-sided.

Fig. 4. Association between plasma proteins that had pQTLs within 500 kb from APOE gene and dementia.

Hazard ratios and 95% CIs for association between an increment of 1 s.d. in plasma protein levels and dementia in the Whitehall II cohort. The analyses included eight proteins with pQTLss clustered around APOE and that were associated with at least three dementia subtypes in MR analyses. Analyses were first adjusted for age and sex and then additionally for APOE status. This analysis was not corrected for multiple testing, and all the tests are two-sided.

Fig. 5. Phenome-wide association analyses for Mendelian randomization-based Alzheimer´s diseases risk polygenic risk score.

Phenome-wide association analyses for MR–PRS constructed from SNPs associated with levels of causal Alzheimer’s disease biomarkers in MR Wald ratio or IVW analyses. ORs and –log₁₀ P values are presented. Upward- and downward-pointing triangles denote increasing and decreasing risk, while larger triangles indicate larger effect size. This analysis was not corrected for multiple testing, and all tests are two-sided.

Extended Data Fig. 1. Odds ratios and 95% confidence intervals per one standard deviation increase in biomarker level derived from Wald ratios when only one SNPs was available and from inverse variance weighted Mendelian randomization when two or more SNPs were available.

All biomarkers passed false discovery rate correction of 5% (p-value < 0.00052) and all the tests were two-sided. (A) atypical or mixed Alzheimer’s disease, (B) early onset Alzheimer’s disease, (C) vascular dementia, (D) frontotemporal dementia.

Extended Data Fig. 2. Odds ratios and 95% confidence intervals per one standard deviation increase in biomarker level derived from Wald ratios when only one SNPs was available and from inverse variance weighted Mendelian randomization when two or more SNPs were available.

All biomarkers passed false discovery rate correction of 5% (p-value < 0.00052) and all the tests were two-sided. (A) continuous cognitive performance, (B) general dementia outcome, (C) dementia in Alzheimer’s disease.

Extended Data Fig. 3. Odds ratios and 95% confidence intervals between one standard deviation change in biomarker levels and late onset Alzheimer’s disease.

Results are from Mendelian randomization sensitivity analyses when at least 3 SNPs were available. All biomarkers passed false discovery rate correction of 5% (p-value < 0.00052) in inverse variance weighted Mendelian randomization and all the tests were two-sided. The source of outcome and MR-base outcome identifier is described below the Biomarker. Black = inverse variance weighted, grey = weighted median, blue = weighted mode, red = Egger Mendelian randomization derived estimate.

Extended Data Fig. 4. Odds ratios and 95% confidence intervals between one standard deviation change in biomarker levels and (A) Alzheimer´s diseases, (B) Parkinson’s disease, and (C) cognitive performance.

Results are from Mendelian randomization sensitivity analyses when at least 3 SNPs were available. All biomarkers passed false discovery rate correction of 5% (p-value < 0.00052) in inverse variance weighted Mendelian randomization and all the tests were two-sided. The source of outcome and MR-base outcome identifier is described below the Biomarker. Black = inverse variance weighted, grey = weighted median, blue = weighted mode, red = Egger Mendelian randomization derived estimate. Egger Mendelian randomization estimate for CD11c on monocyte is omitted because it did not converge.

Extended Data Fig. 5. Phenome-wide Mendelian randomization analyses for the 127 biomarkers that associated with dementia causing diseases.

Betas are derived from Wald ratios when only one SNP was available and from inverse variance weighted Mendelian randomization when two or more SNPs were available. Results are presented for the 63 of the 127 biomarkers that passed false discovery rate correction of 5% (p-value < 0.00029). Most biomarkers associated with only few outcomes and the grey boxes indicate no association after false discovery rate correction of 5% (p-value < 0.00029) or lack of common SNPs. All the tests were two-sided.

Extended Data Fig. 6. ConsensusPathDB shortest interaction path analyses for the first 8 of the 26 proteins that were associated with Alzheimer’s diseases in Mendelian randomization analyses.

The figure describes shortest interaction path between biomarkers and amyloid and tau.

Extended Data Fig. 7. ConsensusPathDB shortest interaction path analyses for additional 8 of the 26 proteins that were associated with Alzheimer’s diseases in Mendelian randomization analyses.

The figure describes shortest interaction path between biomarkers and amyloid and tau.

Extended Data Fig. 8. ConsensusPathDB shortest interaction path analyses for the last 10 of the 26 proteins that were associated with Alzheimer’s diseases in Mendelian randomization analyses.

The figure describes shortest interaction path between biomarkers and amyloid and tau.

Extended Data Fig. 9. ConsensusPathDB shortest interaction path analyses for the 14 proteins that were associated with Parkinson’s diseases in Mendelian randomization analyses.

The figure describes shortest interaction path between biomarkers and α-synuclein.

Extended Data Fig. 10. ConsensusPathDB shortest interaction path analyses for the 2 proteins that were associated with frontotemporal dementia in Mendelian randomization analyses.

The figure describes shortest interaction path between biomarkers and α-synuclein.