TL;DR
This paper introduces PESCA, a novel method to disentangle population-specific and shared genetic influences on complex traits using only GWAS summary statistics. By doing so, it sheds light on why genetic risk prediction doesn't transfer perfectly across populations and identifies shared genetic underpinnings of diseases, even in regions previously thought to be population-specific.
What Problem Does This Paper Address?
Despite strong genetic correlations observed across human populations for many complex traits, polygenic risk scores (PRS) often perform poorly when applied to individuals from different ancestral backgrounds. This suggests that the genetic architecture of these traits isn't uniform across populations. While some genetic variants might have similar effects globally, others might be specific to certain ancestries, or their effects might be modulated by population-specific linkage disequilibrium (LD) patterns. Existing methods struggle to precisely pinpoint these population-specific and shared genetic components using readily available GWAS summary data, hindering our ability to understand disease etiology and improve cross-population genetic predictions.
Methods
The core of this work is the development of PESCA (Population-spEcific/Shared Causal vAriants). PESCA models GWAS summary statistics from two populations (East Asian and European in this study) to estimate the genome-wide proportions of population-specific and shared causal single nucleotide polymorphisms (SNPs). It uses a multivariate Bernoulli model to represent the causal status of SNPs in two populations and a Bayesian framework to infer these proportions. A key innovation is its ability to work with summary statistics and LD information, making it broadly applicable. PESCA then uses these estimated proportions as priors for fine-mapping, allowing it to identify SNPs with a high posterior probability of being causal in a population-specific or shared manner. The method also defines LD blocks that are approximately independent within each population to improve computational efficiency.
Datasets
The study utilized publicly available GWAS summary statistics for 9 complex traits and diseases from East Asian (EAS) and European (EUR) ancestry individuals. These traits included:
- Body Mass Index (BMI)
- Mean Corpuscular Hemoglobin (MCH)
- Mean Corpuscular Volume (MCV)
- High-Density Lipoprotein (HDL)
- Low-Density Lipoprotein (LDL)
- Total Cholesterol (TC)
- Triglycerides (TG)
- Major Depressive Disorder (MDD)
- Rheumatoid Arthritis (RA)
LD information was primarily derived from the 1000 Genomes Project reference panel.
Key Insights and Ideas
- Most common causal SNPs are shared: Contrary to the idea that population-specific genetic architectures are driven by entirely different causal variants, the study found that a large majority of common causal SNPs are shared between East Asian and European populations for most complex traits.
- Population-specific LD masks shared signals: Population-specific LD patterns can lead to GWAS associations appearing as population-specific, even when the underlying causal variant is shared. This means that regions identified as population-specific risk regions might actually harbor shared causal SNPs that are simply not well-tagged in the other population's GWAS due to LD differences.
- Tissue-specific expression links to shared genetics: Enrichments of shared causal SNPs are found in genes expressed in trait-relevant tissues, suggesting that these shared genetic influences are biologically meaningful and related to specific biological functions.
- SNP-heritability is driven by low-effect SNPs: The study provides evidence that SNP-heritability enrichments in functional categories are largely driven by many low-effect SNPs rather than a few high-effect SNPs.
Key Results
- Unbiased estimates with in-sample LD: PESCA provides approximately unbiased estimates of population-specific and shared causal SNPs when using in-sample LD from the GWAS data.
- Slight upward bias with external LD: When using external LD reference panels (like 1000 Genomes), estimates show a slight upward bias, which is more pronounced when effective sample sizes between populations differ significantly.
- High-posterior SNPs are correlated: The effect sizes of SNPs with high posterior probabilities of being causal are highly correlated between EAS and EUR, regardless of whether they are population-specific or shared.
- 2.8x enrichment in population-specific regions: Population-specific GWAS risk regions show a significant 2.8x enrichment of shared high-posterior SNPs compared to the genome-wide background.
- Broad applicability: PESCA was applied to 9 complex traits, demonstrating its utility across diverse phenotypes.
Limitations and Future Work
- Gene-environment interactions: The interpretation of population-specific causal SNPs can be influenced by gene-environment interactions that are specific to one ancestry group. PESCA might misinterpret these as purely genetic effects.
- External LD limitations: Performance of PESCA decreases with external LD when effective sample sizes differ by more than a factor of 2x. The method also cannot be directly applied to admixed populations without in-sample LD. Future work could focus on improving robustness with noisy external LD estimates.
- Non-OLS association statistics: PESCA is defined for GWAS marginal effects from OLS regression. Its sensitivity to association statistics from fixed-effect meta-analyses or linear mixed models needs further investigation.
- Low-frequency/rare variants: The current analysis is restricted to common SNPs (MAF > 5%) to reduce noise in LD matrices. Future work could explore using lower MAF thresholds if in-sample LD is available to capture effects of low-frequency or rare variants.
- Model misspecification: PESCA relies on defining approximately LD-independent regions. LD leakage between regions can bias estimates. Additionally, it does not explicitly model cross-population correlations of effect sizes at shared causal variants, which could potentially improve performance.
Summary updated