A powerful approach to estimating annotation-stratified genetic covariance using GWAS summary statistics

Qiongshi Lu, Boyang Li, Derek Ou, Margret Erlendsdottir, Ryan L. Powles, Tony T. Jiang, Yiming Hu, David Chang, Chentian Jin, Wei Dai, Qidu He, Zefeng Liu, Shubhabrata Mukherjee, Paul K. Crane, Hongyu ZhaobioRxiv 2017 · DOI

TL;DR

This paper introduces GNOVA, a new method to estimate genetic covariance between complex traits using readily available GWAS summary statistics. GNOVA goes beyond previous methods by allowing for "annotation-stratified" analysis, meaning it can pinpoint which parts of the genome contribute to the shared genetic basis of traits, offering deeper biological insights and improved statistical power.

What Problem Does This Paper Address?

Genome-wide association studies (GWASs) have been incredibly successful in identifying genetic variants associated with complex traits and diseases. However, understanding the shared genetic architecture between different traits (i.e., how genes influence multiple traits) has been more challenging. Existing methods for estimating genetic covariance often require individual-level data, which is not always accessible, or they lack the power to detect subtle genetic relationships. Furthermore, these methods typically treat the entire genome uniformly, missing opportunities to understand how specific genomic regions or functional elements contribute to shared genetic effects. This limits our ability to dissect the complex interplay of genes in disease etiology.

Methods

The core of GNOVA is a principled framework based on the method of moments to estimate genetic covariance using only GWAS summary statistics. This is a significant advantage as summary statistics are widely available and easier to share than raw individual-level data.

The key innovation is annotation-stratified genetic covariance. GNOVA allows researchers to define functional annotations of the genome (e.g., regions with predicted regulatory function, protein-coding genes) and estimate genetic covariance within these specific annotations, as well as across the entire genome. This means GNOVA can tell us not just if two traits share genetic influences, but where in the genome those influences are concentrated.

The method is robust to issues like sample overlap between GWASs and uses a block-wise jackknife approach for variance estimation, providing reliable statistical inference.

Datasets

The authors utilized publicly available GWAS summary statistics for 48 complex traits, with a total sample size of approximately 4.5 million individuals. They also used functional annotation data from GenoCanyon and GenoSkyline-Plus, which integrate various epigenomic and transcriptomic data to predict functional regions in the human genome. For LD (linkage disequilibrium) estimation, they used data from the 1000 Genomes Project.

Key Insights and Ideas

  • Annotation-stratified analysis is crucial: Simply looking at overall genetic covariance can mask important biological signals. By stratifying analysis by functional annotations, GNOVA can reveal that shared genetic effects are enriched in specific genomic regions (e.g., predicted functional regions).
  • High MAF SNPs drive covariance for some traits: For certain trait pairs, like Alzheimer's disease (LOAD) and amyotrophic lateral sclerosis (ALS), the genetic covariance is predominantly driven by common variants (high minor allele frequency). This is a surprising finding given that ALS heritability is often enriched in lower MAF variants.
  • GNOVA offers improved power: Simulations show GNOVA is more powerful than existing methods like LDSC, especially when genetic correlations are moderate or when analyzing annotation-stratified covariance.
  • Neurodegenerative diseases share genetic architecture: The study found significant genetic covariance between LOAD and ALS, suggesting shared underlying genetic mechanisms.

Key Results

  • GNOVA identified over 170 pairs of traits with statistically significant genetic covariance among 50 complex traits analyzed.
  • The method demonstrated higher statistical power compared to LDSC in simulations, particularly for detecting moderate genetic correlations.
  • For the LOAD and ALS pair, GNOVA found a significant positive genetic correlation (0.175, p = 2.0x10⁻⁴). This correlation was largely concentrated in predicted functional genomic regions (82.6% of covariance in 33% of the genome) and driven by high MAF SNPs (54.6% of covariance in the highest MAF quartile).
  • The analysis of 48 complex traits revealed broad patterns of disease relatedness through hierarchical clustering, with distinct genetic covariance profiles for LOAD and ALS when compared to other traits. LOAD showed negative correlations with cognitive and hormonal traits, while ALS showed positive correlations with immune-related traits.

Limitations and Future Work

  • Interpretation of covariance: While GNOVA can identify where genetic covariance is concentrated, interpreting the biological meaning of these shared genetic effects remains a challenge.
  • Generalized metrics: The current method assumes proportional effect sizes for shared genetic components. Future work could extend the method to estimate more generalized metrics, such as consistency in effect directions.
  • SNP-level pleiotropy: GNOVA identifies broad genomic regions with shared genetic influences but does not pinpoint specific SNPs with pleiotropic effects. Future research could integrate SNP-based methods to complement GNOVA's findings.
  • Environmental factors and rare variants: The study acknowledges that common SNPs explain only a portion of phenotypic covariance, and rare variants, copy number variations, and environmental factors likely play significant roles that are not captured by this method.
  • Sample size growth: The authors anticipate that as GWAS sample sizes continue to grow, GNOVA will become even more powerful in identifying shared genetic components and uncovering novel insights into disease etiology.

Summary updated

Abstract

Abstract Despite the success of large-scale genome-wide association studies (GWASs) on complex traits, our understanding of their genetic architecture is far from complete. Jointly modeling multiple traits’ genetic profiles has provided insights into the shared genetic basis of many complex traits. However, large-scale inference sets a high bar for both statistical power and biological interpretability. Here we introduce a principled framework to estimate annotation-stratified genetic covariance between traits using GWAS summary statistics. Through theoretical and numerical analyses we demonstrate that our method provides accurate covariance estimates, thus enabling researchers to dissect both the shared and distinct genetic architecture across traits to better understand their etiologies. Among 50 complex traits with publicly accessible GWAS summary statistics (N total ≈ 4.5 million), we identified more than 170 pairs with statistically significant genetic covariance. In particular, we found strong genetic covariance between late-onset Alzheimer’s disease (LOAD) and amyotrophic lateral sclerosis (ALS), two major neurodegenerative diseases, in single-nucleotide polymorphisms (SNPs) with high minor allele frequencies and in SNPs located in the predicted functional genome. Joint analysis of LOAD, ALS, and other traits highlights LOAD’s correlation with cognitive traits and hints at an autoimmune component for ALS.
BibTeX
@misc{lu2017powerful,
  author = {Qiongshi Lu and Boyang Li and Derek Ou and Margret Erlendsdottir and Ryan L. Powles and Tony T. Jiang and Yiming Hu and David Chang and Chentian Jin and Wei Dai and Qidu He and Zefeng Liu and Shubhabrata Mukherjee and Paul K. Crane and Hongyu Zhao},
  title = {{A powerful approach to estimating annotation-stratified genetic covariance using GWAS summary statistics}},
  year = {2017},
  doi = {10.1101/114561},
  howpublished = {bioRxiv},
}
Figure