Limited overlap between genetic effects on disease susceptibility and disease survival

Zhiyu Yang, Fanny‐Dhelia Pajuste, Kristina Zguro, Yipeng Cheng, Danielle E. Kurant, Andrea Eoli, Julian Wanner, Bradley Jermy, Stavroula Kanoni, David A. van Heel, Caroline Hayward, Riccardo E. Marioni, Daniel L. McCartney, Alessandra Renieri, Simone Furini, Reedik Mägi, Alexander Gusev, Petros Drineas, Peristera Paschou, Henrike Heyne, Samuli Ripatti, Nina Mars, Andrea GannamedRxiv 2023 · DOI

TL;DR

A massive study across seven biobanks reveals that the genes making you susceptible to a disease have almost no overlap with the genes determining whether you survive it. While we have identified hundreds of genetic markers for disease risk, this research found only one significant locus for disease-specific mortality, suggesting that survival is driven more by general longevity factors than by the specific disease’s genetic architecture.

What Problem Does This Paper Address?

In the world of genomics, we are very good at answering the question: "Who is likely to get this disease?" Thanks to Genome-Wide Association Studies (GWAS), we have identified thousands of loci associated with disease susceptibility. However, we are much worse at answering: "Once someone is sick, how will the disease progress?"

Understanding the genetics of disease progression is critical for drug discovery—most medicines are designed to treat or cure an existing condition, not prevent it from occurring. The challenge is that studying progression is statistically messy. Researchers often face "index-event bias," where conditioning a study only on people who already have a disease can create false genetic correlations. This paper seeks to systematically determine if the genetic "map" for getting a disease is the same as the map for surviving it.

Methods

The authors conducted a large-scale, multi-biobank study focusing on ten common diseases (including various cancers, Type 2 diabetes, and heart failure).

  1. Survival GWAS: Instead of a standard case-control setup, they used Cox proportional hazard models to perform a GWAS of "disease-specific mortality" among affected individuals.
  2. Meta-Analysis: They aggregated data across seven major longitudinal studies (including FinnGen, UK Biobank, and the Estonian Biobank).
  3. Polygenic Scores (PGS): They tested whether genetic risk scores for contracting a disease could predict dying from it.
  4. Longevity Comparison: They compared disease-specific risk scores against a general "longevity" PGS derived from the general population.
  5. Theoretical Simulation: To ensure their results weren't just artifacts of statistical bias (like index-event bias), they developed a mathematical framework to simulate how susceptibility and progression genetics interact.

Datasets

The study utilized data from seven large-scale biobanks:

  • FinnGen (Finland)
  • UK Biobank (UK)
  • Estonian Biobank (Estonia)
  • Generation Scotland (UK)
  • Genomics England (UK)
  • Genes & Health (UK)
  • Dana-Farber Cancer Institute / BioMe (USA)

The analysis covered ten diseases with case numbers ranging from 17,152 (Alzheimer’s) to 106,405 (Type 2 diabetes).

Key Insights and Ideas

  • The "Two-Map" Reality: The genetic factors that put you at risk for a disease (susceptibility) are largely decoupled from the factors that determine your survival (progression).
  • Longevity Trumps Disease Risk: Surprisingly, a polygenic score for general lifespan was a better predictor of survival for five out of the ten diseases than the disease-specific susceptibility score. This suggests that "survival" might be a general biological trait rather than one specific to the pathology of a single disease.
  • The Power Paradox: Even when the researchers down-sampled their susceptibility data to match the smaller sample sizes of the mortality data, the susceptibility GWAS still found many more significant hits. This indicates that the lack of findings in mortality isn't just a "small sample size" problem—it’s because the heritability of mortality is fundamentally lower.
  • Index-Event Bias isn't the Boogeyman: While often blamed for poor results in progression studies, the authors' simulations showed that this bias alone couldn't explain the lack of overlap they observed.

Key Results

  • Only One Hit: Out of all ten diseases, only one genetic locus (rs7360523 on chromosome 20, near the SULF2 gene) was significantly associated with disease-specific mortality (specifically for heart failure).
  • Weak Prediction: Susceptibility Polygenic Scores (PGS) were strong predictors of diagnosis (Hazard Ratios up to 1.90) but were nearly useless at predicting mortality (Hazard Ratios near 1.0).
  • Directional Coin Flip: While 888 variants were known to affect disease susceptibility, none were significantly associated with mortality after correcting for multiple testing. Only about 54% showed the same direction of effect, which is only marginally better than a coin toss.

Limitations and Future Work

The authors acknowledge that "disease-specific mortality" is a blunt instrument for measuring progression. Mortality is influenced by many non-genetic factors, including socioeconomic status, access to care, and treatment response (which varies wildly between patients).

Future research may need to move beyond mortality to "finer" phenotypes—like changes in biomarkers, imaging results, or specific clinical milestones—to capture the subtler genetic signals of disease progression. Additionally, as biobanks grow, even larger sample sizes may eventually reveal the "missing" heritability of how diseases evolve within the human body.

Summary updated

Abstract

Abstract Understanding disease progression is of a high biological and clinical interest. Unlike disease susceptibility whose genetic basis has been abundantly studied, less is known about the genetics of disease progression and its overlap with disease susceptibility. Considering ten common diseases (N cases ranging from 17,152 to 99,666) across seven biobanks, we systematically compared the genetic architecture of susceptibility and progression, defined as disease-specific mortality. We identified only one locus significantly associated with disease-specific mortality and show that, at a similar sample size, more genome-wide significant loci can be identified in a GWAS of disease susceptibility. Variants that were significantly affecting disease susceptibility were weakly or not associated with disease-specific mortality. Moreover, susceptibility polygenic scores (PGSs) were weak predictor of disease-specific mortality while a PGS for general lifespan was significantly associated with disease-specific mortality for five out of ten diseases. We used theoretical derivation and simulation to propose plausible explanations for our empirical observations and account for potential index-event bias. Overall, our findings point to little similarity in genetic effects between disease susceptibility and disease-specific mortality and suggest that either larger sample sizes or different measures of progression are needed to identify the genetic underpinning of disease progression.
BibTeX
@misc{yang2023limited,
  author = {Zhiyu Yang and Fanny‐Dhelia Pajuste and Kristina Zguro and Yipeng Cheng and Danielle E. Kurant and Andrea Eoli and Julian Wanner and Bradley Jermy and Stavroula Kanoni and David A. van Heel and Caroline Hayward and Riccardo E. Marioni and Daniel L. McCartney and Alessandra Renieri and Simone Furini and Reedik Mägi and Alexander Gusev and Petros Drineas and Peristera Paschou and Henrike Heyne and Samuli Ripatti and Nina Mars and Andrea Ganna and Andrea Ganna},
  title = {{Limited overlap between genetic effects on disease susceptibility and disease survival}},
  year = {2023},
  doi = {10.1101/2023.10.10.23296544},
  howpublished = {medRxiv},
}
Figure