In medical research, we often observe that two things happen together—for instance, people with gum disease often have liver issues, or people who sleep poorly often have metabolic disorders. However, observational studies struggle to answer the "chicken-and-egg" question: does A cause B, does B cause A, or are they both caused by a hidden third factor like socioeconomic status or diet?
To solve this, researchers are increasingly using Mendelian Randomization (MR). This technique uses genetic variants as "natural experiments" to determine whether an exposure actually causes a disease. By leveraging the random assortment of genes at conception, MR mimics the structure of a Randomized Controlled Trial (RCT), providing a way to infer causality where traditional clinical trials might be unethical or impractical.
The Genetic Randomized Controlled Trial
The intuition behind MR lies in Mendel’s Law of Independent Assortment. When you were conceived, the alleles you inherited from your parents were essentially randomized. If a specific genetic variant is known to increase a person's circulating levels of branched-chain amino acids (BCAAs), and we find that people with this variant also have higher blood pressure, we can infer that the BCAAs are likely causing the high blood pressure.
Because your DNA sequence is fixed at birth and generally not influenced by your lifestyle or the onset of disease, this approach bypasses the two biggest headaches in epidemiology: confounding (hidden factors) and reverse causation (the disease causing the exposure).
Researchers like Lang Wu, Chong Wu, and Jingjing Zhu have used this "genetic RCT" approach to show that a genetic predisposition to coronary artery disease (CAD) causally increases the risk of contracting COVID-19, as detailed in Mendelian randomization analysis to characterize causal association between coronary artery disease and COVID-19.
The Challenge of Horizontal Pleiotropy
The biggest threat to the validity of an MR study is horizontal pleiotropy. For MR to work, the genetic variant (the "instrument") must affect the outcome only through the exposure being studied.
If a gene that increases BCAA levels also happens to directly affect kidney function through a completely different biological pathway, the "instrument" is "leaky." This leakiness can create a false signal of causality. To combat this, the field has developed a suite of "pleiotropy-robust" statistical tools:
- MR-Egger Regression: This method can detect and adjust for directional pleiotropy by looking at the intercept of the regression line.
- Weighted Median Estimators: These provide a valid causal estimate even if up to 50% of the genetic instruments are pleiotropic.
- MR-PRESSO: This technique identifies and removes specific "outlier" variants that seem to be influencing the outcome through alternative pathways.
As yaoxian Xiang and colleagues demonstrated in their study on colorectal cancer and the gut microbiome, robust pleiotropy filtering is essential to ensure that identified bacterial taxa, like Porphyromonadaceae, are true drivers of disease rather than statistical artifacts.
Disentangling Complex Relationships
MR is particularly powerful when relationships are bidirectional. For example, does depression cause gut issues, or do gut issues cause depression? In Depression and 24 Gastrointestinal Diseases: A Mendelian Randomization Study, Shuai Yuan and colleagues used MR to investigate whether mental health is a causal driver of digestive health across two dozen conditions.
Similarly, Xingang Sun and Liangrong Zheng used bidirectional MR in Obstructive sleep apnea and atrial fibrillation: insights from a bidirectional Mendelian randomization study to show that while sleep apnea causes heart rhythm disorders, the reverse is not true.
A recent advancement in this area is the HVP model, introduced by Lamessa Dube Amente and Sang Lee in Disentangling horizontal and vertical Pleiotropy in genetic correlation estimation: introducing the HVP model. This model helps distinguish between "vertical" pleiotropy (where one trait causes another) and "horizontal" pleiotropy (where they share a common genetic root), solving a long-standing "identity crisis" in genomic data.
From Risk Factors to Drug Targets
Beyond understanding disease, MR is being used to screen for new treatments. By treating genetic variants that mimic the effect of a drug (like those affecting the IL-6 receptor) as instruments, researchers can predict whether a drug will be effective before a single patient is enrolled in a trial.
For instance, Guangyang Ou and Kunpeng Yao used this to explore repurposing IL-6 receptor blockers for cardiovascular disease. By aggregating effects across multiple genetic variants and applying sensitivity analyses like "leave-one-out" testing, they can ensure that the predicted therapeutic effect is robust and not driven by a single idiosyncratic gene.
By turning the "noise" of genetic variation into a tool for causal inference, Mendelian Randomization is transforming how we identify the metabolic, microbial, and lifestyle drivers of human health.