Specificity, length and luck drive gene rankings in association studies

GWAS
burden tests
complex traits
genetic architecture
  • Systematic comparison of GWAS and rare variant burden tests across 209 UK Biobank traits revealing they prioritize different genes through distinct mechanisms
  • Burden tests favor trait-specific genes while GWAS capture both trait-specific genes and context-specific variants on pleiotropic genes
  • Gene length and genetic drift are major confounders affecting rankings in burden tests and GWAS respectively
Published

23 January 2026

PubMed: 41193809
DOI: 10.1038/s41586-025-09703-7
Overview generated by: Claude Sonnet 4.5, 25/11/2025

Key Findings

This study provides a systematic comparison of genome-wide association studies (GWAS) and rare variant loss-of-function (LoF) burden tests, revealing fundamental differences in how these methods prioritize genes.

Main Discoveries

  1. Different Gene Rankings: GWAS and LoF burden tests systematically prioritize different genes, even when accounting for power differences and gene-mapping issues

  2. Trait Specificity vs. Importance:

    • Burden tests prioritize trait-specific genes (genes primarily affecting the studied trait)
    • GWAS prioritize trait-specific variants (which can act on pleiotropic genes through context-specific effects)

  3. Three Key Factors:

    • Specificity: How specific a gene/variant’s effects are to the trait under study
    • Length: Longer genes are more likely to be discovered by burden tests due to more LoF sites
    • Luck: Random genetic drift causes variant frequencies to vary, making GWAS rankings partially stochastic

Study Design

Data Source

  • 209 quantitative traits from UK Biobank
  • Analyzed both GWAS and LoF burden test results
  • ~360,000 individuals for GWAS
  • ~450,000 individuals for burden tests

Analytical Framework

The authors proposed two criteria for gene prioritization:

  1. Trait Importance: How much a gene quantitatively affects a trait (effect size)
  2. Trait Specificity: The importance of a gene for the studied trait relative to all other traits

Major Results

1. Discordant Gene Rankings

  • Only 26% of burden test hits fall within top GWAS loci
  • 74.6% of burden hits are within any GWAS locus, but often ranked much lower
  • Example: NPR2 is the 2nd most significant burden test gene for height but contained in the 243rd most significant GWAS locus

2. Why Burden Tests Prioritize Trait-Specific Genes

Burden tests aggregate LoF variants within genes. The expected association strength is proportional to:

\[\frac{\gamma_1^2}{\sum_t \gamma_t^2}\]

where \(\gamma_1^2\) is the gene’s effect on the studied trait and \(\sum_t \gamma_t^2\) is its total effect across all traits.

Key mechanisms: - Natural selection acts more strongly on genes with larger total effects across traits - This keeps LoF frequencies lower for pleiotropic genes - Result: Trait-specific genes have more power in burden tests despite potentially smaller effects

3. Why GWAS Capture Pleiotropic Genes

GWAS prioritize trait-specific variants, which can arise in two ways:

  1. Variants affecting trait-specific genes
  2. Context-specific variants on pleiotropic genes (e.g., tissue-specific regulatory variants)

The study found: - Variants in tissue-specific ATAC peaks show higher heritability enrichment - Coding variants in specifically expressed genes contribute more to heritability - This explains why GWAS can identify highly pleiotropic genes missed by burden tests

4. Gene Length Bias in Burden Tests

  • Longer genes have more potential LoF sites
  • This increases LoF carrier frequency, boosting statistical power
  • Effect: Longer genes appear more significant and more pleiotropic, independent of their true biological importance

5. The Role of Genetic Drift in GWAS

  • Random drift causes variant frequencies to vary widely around their expected values
  • For sufficiently important variants, GWAS rankings become largely random with respect to true effect size
  • High-frequency variants have more power, creating an apparent “pleiotropy” of top GWAS hits (statistical artifact)

Estimating Trait Importance

Neither method directly ranks genes by trait importance:

  • Burden tests: Flattening effect - most important genes are most constrained, leading to smallest frequencies and largest standard errors
  • GWAS: Individual variant rankings dominated by random frequency variation

Solution: Aggregate signals across multiple variant types - Methods like AMM that combine evidence across many variants per gene - Better correlates with selection coefficients (proxy for importance) - Overcomes the flattening problem

Biological Implications

Context-Specific Variants

The finding that many GWAS loci lack burden signals suggests context-specific variants acting on pleiotropic genes are major drivers of complex traits. The authors hypothesize these may include: - Developmental genes - Variants that perturb developmental trajectories in trait-specific ways - Tissue-specific regulatory elements

Drug Target Discovery

  • Trait-specific genes (identified by burden tests) may be better drug targets due to fewer side effects
  • Explains why LoF burden evidence is more predictive of drug trial success than GWAS evidence
  • However, if pleiotropic genes can be targeted context-specifically, they may have greater clinical impact

Methodological Insights

S-LDSC Analysis

The study used stratified LD score regression to show: - Heritability enrichment in tissue-specific ATAC peaks - Coding variants in specifically expressed genes contribute more to heritability - Both axes (gene specificity and variant context-specificity) independently contribute to GWAS signals

Population Genetics Modeling

Used stabilizing selection models to predict: - LoF frequencies inversely proportional to selection coefficient (\(s_{het}\)) - Selection coefficient proportional to total trait effects across all traits - Explains observed relationship between constraint and LoF frequency

Limitations and Considerations

  1. Complexity of gene effects: The simplified model (\(\alpha = \beta \gamma\)) may not capture non-linear relationships
  2. Incomplete pleiotropy landscape: Only 27 traits analyzed, actual pleiotropy may be higher
  3. Context-specificity: Not all pleiotropic genes can be therapeutically targeted in context-specific ways

Practical Recommendations

For Association Study Interpretation

  1. Use both methods: GWAS and burden tests reveal complementary aspects of trait biology
  2. Consider gene length: Longer genes in burden test results may be artifacts
  3. Aggregate variants: For trait importance, use methods that combine signals across variants (AMM, MAGMA)
  4. Mind the drift: Top GWAS hits are partially determined by random frequency variation

For Drug Development

  1. Burden test hits may indicate better targets for minimizing side effects
  2. GWAS hits may reveal pleiotropic genes with larger phenotypic impact
  3. Consider whether context-specific targeting is feasible

Conclusions

GWAS and LoF burden tests systematically prioritize different genes because:

  • Burden tests rank by gene-level trait specificity, favoring long, trait-specific genes
  • GWAS rank by variant-level trait specificity, capturing both trait-specific genes and context-specific variants on pleiotropic genes

Neither method directly ranks by trait importance due to: - Burden tests: Flattening from natural selection - GWAS: Random genetic drift

Both methods are valuable and reveal distinct aspects of trait biology. The choice of method depends on the research question and application, with burden tests better for identifying specific biology and GWAS better for comprehensive discovery including pleiotropic mechanisms.