A Genetic Approach to Causal Inference in Parkinson’s Disease
Can we use genetic data to uncover cause-and-effect in Parkinson’s research?
That’s the goal of Mendelian Randomization (MR): a method that uses natural genetic variation to test whether a biomarker, exposure, or behavior is likely to cause a disease or symptom.
In this blog post, we’ll cover:
-
What is MR
-
The three key assumptions behind it
-
Best practices for conducting an MR study
-
A real example: Telomere length and PD risk from a recent MR analysis
What Is Mendelian Randomization?
Mendelian Randomization (MR) is a method that uses genetic variants as *instrumental variables (*proxies) to estimate whether an exposure (like cholesterol, inflammation, or telomere length) causes a particular outcome (like Parkinson’s disease, motor severity, or age at onset). This is useful when conducting an experimental/randomized clincal trial is not feasible or ethical, or when there is an absence of robust observational data that includes the exposure, outcome, and likely confounders.
The foundation of MR is based on:
-
Genetic variation is randomly inherited at conception, similar to the random assignment of an intervention in a randomized controlled trial (RCT).
-
If people with a genetic predisposition toward higher levels of a biomarker (e.g., longer telomeres) also have different rates of a disease (e.g., lower PD risk), that suggests a causal relationship.
MR Depends on 3 Core Assumptions
To make valid causal inferences, three conditions must be met:
-
Relevance
The genetic variants used (the “instruments”) must be strongly associated with the exposure.
E.g., SNPs that predict telomere length. (In practice SNPs with associations p<5x10-8 are considered to mean this assumption) -
Independence
The variants must not be associated with confounders that affect both the exposure and the outcome. -
Exclusion Restriction
The variants should affect the outcome only through the exposure, not via other pathways.
E.g., the telomere SNPs shouldn’t also influence PD risk through a completely separate biological mechanism,but only through the fact they influence telomere length.
If these assumptions hold, MR helps untangle cause and effect in observational data that is prone to reverse causation and residual confounding.
Real Example: Does Telomere Length Cause Parkinson’s Disease?
In our recent study (Misicka, Iyengar & Briggs, 2025) used MR to test whether telomere length (TL), a marker of cellular aging, causally affects Parkinson’s disease risk or clinical features.
We did the following 2-sample MR approach which leverage GWAS summary statistics for the proposed exposure (TL) and the several PD outcomes**:**
-
Used UK Biobank GWAS data from 472,000 individuals to identify a genetic instrument for TL.
-
Evaluated whether genetically longer telomeres influenced:
-
PD risk
-
Age at onset (AAO)
-
Motor subtype (tremor-dominant vs PIGD)
-
A continuous motor severity score
-
-
Also examined the bidirectional relationship to test if PD could causally influence TL.
Findings:
-
No evidence that genetically longer or shorter telomeres affected any PD traits.
-
No reverse effect (i.e., PD did not influence TL).
-
No indication of bias from pleiotropy or invalid instruments.
Implication: Although TL and PD may be associated in observational studies, TL does not appear to causally drive PD risk or severity - at least not as measured in leukocytes, which is what was used in the exposure UKB GWAS.
Best Practices When Using MR
If you’re considering MR for your own Parkinson’s research, keep these in mind:
-
Use large, well-powered GWAS datasets
Both for the exposure and outcome. Weak instruments → misleading results. Also GWAS datasets should ideally by independent, but there are sensitivity analyses that can be done if they are not. -
Assess instrument strength
Use F-statistics or variance explained (R²) to confirm you’re not underpowered. This is based on the sample size of the GWAS, the allele frequency, and effect size. -
Test for pleiotropy
Use sensitivity analyses like MR-Egger, MR-PRESSO, and weighted median models to detect bias. -
Bidirectional analysis
Don’t just ask whether X causes Y, check if Y might influence X too. This helps confirm the directionality of the relationship. There are also additional MR tests that does this too. -
Know your tissues
Telomere length in blood ≠ telomere length in the brain. Consider tissue relevance when interpreting null or surprising results.
MR isn’t perfect, it depends on genetics, power, and assumptions, but it’s one of the best tools we have to bring causal thinking into the era of big data, allowing us to prioritize relationships for follow-up studies, confirm observational findings, and explore relationships that are not feasible/ethical.