Exploring the phenotypic consequences of tissue specific gene expression variation inferred from GWAS summary statistics

Alvaro Barbeira, Scott Dickinson, Rodrigo Bonazzola, Jiamao Zheng, Heather E. Wheeler, Jason Torres, Eric S. Torstenson, Kaanan P. Shah, Tzintzuni Garcia, Todd L. Edwards, Eli A. Stahl, Laura M. Huckins, François Aguet, Kristin Ardlie, Beryl B. Cummings, Ellen Gelfand, Gad Getz, Kane Hadley, Robert E. Handsaker, Katherine Huang, Seva Kashin, Konrad J. Karczewski, Monkol Lek, Xiao Li, Daniel G. MacArthur, Jared L. Nedzel, Duyen T. Nguyen, Michael S. Noble, Ayellet V. Segrè, Casandra A. Trowbridge, Taru Tukiainen, Nathan S. Abell, Brunilda Balliu, Ruth Barshir, Omer Basha, Alexis Battle, Gireesh K. Bogu, Andrew Brown, Christopher Brown, Stephane E. Castel, Lin Chen, Colby Chiang, Donald F. Conrad, Farhan N. Damani, Joe R. Davis, Olivier Delaneau, Emmanouil T. Dermitzakis, Barbara E. Engelhardt, Eleazar Eskin, Pedro G. Ferreira, Laure Frésard, Eric R. Gamazon, Diego Garrido-Martín, Ariel Dh Gewirtz, Genna Gliner, Michael J. Gloudemans, Roderic Guigó, Ira M. Hall, Buhm Han, Yuan He, Farhad Hormozdiari, Cédric Howald, Brian Jo, Eun Yong Kang, Yungil Kim, Sarah Kim-Hellmuth, Tuuli Lappalainen, Gen Li, Xin Li, Boxiang Liu, Serghei Mangul, Mark I. McCarthy, Ian C. McDowell, Pejman Mohammadi, Jean Monlong, Stephen B. Montgomery, Manuel Muñoz-Aguirre, Anne W. Ndungu, Andrew B. Nobel, Meritxell Oliva, Halit Ongen, John Palowitch, Nikolaos Panousis, Panagiotis Papasaikas, YoSon Park, Princy Parsana, A. J. Payne, Christine B. Peterson, Jie Quan, Ferrán Reverter, Chiara Sabatti, Ashis Saha, Michael Sammeth, Alexandra J. Scott, Andrey A. Shabalin, Reza Sodaei, Matthew Stephens, Barbara E. Stranger, Benjamin J. Strober, Jae Hoon SulNature Communications 2018 · DOI

Abstract

Scalable, integrative methods to understand mechanisms that link genetic variants with phenotypes are needed. Here we derive a mathematical expression to compute PrediXcan (a gene mapping approach) results using summary data (S-PrediXcan) and show its accuracy and general robustness to misspecified reference sets. We apply this framework to 44 GTEx tissues and 100+ phenotypes from GWAS and meta-analysis studies, creating a growing public catalog of associations that seeks to capture the effects of gene expression variation on human phenotypes. Replication in an independent cohort is shown. Most of the associations are tissue specific, suggesting context specificity of the trait etiology. Colocalized significant associations in unexpected tissues underscore the need for an agnostic scanning of multiple contexts to improve our ability to detect causal regulatory mechanisms. Monogenic disease genes are enriched among significant associations for related traits, suggesting that smaller alterations of these genes may cause a spectrum of milder phenotypes.

Introduction

Genome-Wide Association Studies (GWAS) have been remarkably successful in identifying genetic loci associated with complex human traits and diseases. However, a significant "mechanistic gap" remains: most associated variants are located in non-coding regions, making it difficult to identify the causal genes and biological pathways involved. Methods like PrediXcan were developed to bridge this gap by imputing gene expression levels from individual-level genotypes and testing their association with traits. However, the requirement for individual-level data often limits the scalability and accessibility of such methods due to privacy and computational constraints.

Methods

To overcome these limitations, we developed S-PrediXcan (Summary PrediXcan), a mathematical framework that computes gene-level association results using only GWAS summary statistics. This approach allows researchers to leverage the vast amount of publicly available summary data from large-scale meta-analyses.

Figure 1: Conceptual workflow comparing standard GWAS, individual-level PrediXcan, and the summary-based S-PrediXcan approach.

The S-PrediXcan method calculates a gene-level Z-score as a weighted sum of SNP-level Z-scores, adjusted by the covariance (linkage disequilibrium structure) of those SNPs derived from a reference population, such as the 1000 Genomes Project. We implemented this within the MetaXcan framework, which utilizes expression prediction models trained on datasets like GTEx (44 tissues) and DGN (whole blood). To ensure the robustness of our findings, we incorporated a colocalization (COLOC) step to distinguish between true causal associations and those driven by linkage disequilibrium (LD) contamination.

Results

We validated S-PrediXcan by comparing its results with those obtained from individual-level PrediXcan. The results were nearly identical, with correlation coefficients ($r^2$) exceeding 0.99 when the reference population was well-matched to the GWAS study population.

Figure 3: High concordance ($r^2 > 0.99$) between PrediXcan and S-PrediXcan association results for Bipolar disorder and Type 1 diabetes.

Applying S-PrediXcan to over 100 phenotypes across 44 tissues, we identified thousands of significant gene-trait associations. For instance, we found 1,686 unique genes associated with height. We also demonstrated high replication rates in an independent cohort (GERA), with 78.5% of LDL cholesterol-associated genes replicating.

Our analysis revealed that most gene-trait associations are tissue-specific, emphasizing the importance of context in disease etiology. Interestingly, we found that an "agnostic" scanning of all tissues often identified significant associations in unexpected contexts, suggesting that regulatory mechanisms may be shared across tissues or act in non-obvious ways. Furthermore, we observed an enrichment of monogenic disease genes among our results, suggesting that common complex traits may be driven by subtle expression variations in the same genes that cause severe rare diseases.

Discussion

S-PrediXcan provides a scalable and robust method for integrating transcriptomic data with GWAS results. While the method is generally robust to reference set misspecification, optimal performance is achieved when the reference population closely matches the GWAS cohort. Future work will focus on improving colocalization methods to account for multiple causal variants at a single locus, further refining our ability to identify the precise regulatory mechanisms underlying human phenotypes.

Summary updated

Abstract

Scalable, integrative methods to understand mechanisms that link genetic variants with phenotypes are needed. Here we derive a mathematical expression to compute PrediXcan (a gene mapping approach) results using summary data (S-PrediXcan) and show its accuracy and general robustness to misspecified reference sets. We apply this framework to 44 GTEx tissues and 100+ phenotypes from GWAS and meta-analysis studies, creating a growing public catalog of associations that seeks to capture the effects of gene expression variation on human phenotypes. Replication in an independent cohort is shown. Most of the associations are tissue specific, suggesting context specificity of the trait etiology. Colocalized significant associations in unexpected tissues underscore the need for an agnostic scanning of multiple contexts to improve our ability to detect causal regulatory mechanisms. Monogenic disease genes are enriched among significant associations for related traits, suggesting that smaller alterations of these genes may cause a spectrum of milder phenotypes.
BibTeX
@inproceedings{barbeira2018exploring,
  author = {Alvaro Barbeira and Scott Dickinson and Rodrigo Bonazzola and Jiamao Zheng and Heather E. Wheeler and Jason Torres and Eric S. Torstenson and Kaanan P. Shah and Tzintzuni Garcia and Todd L. Edwards and Eli A. Stahl and Laura M. Huckins and François Aguet and Kristin Ardlie and Beryl B. Cummings and Ellen Gelfand and Gad Getz and Kane Hadley and Robert E. Handsaker and Katherine Huang and Seva Kashin and Konrad J. Karczewski and Monkol Lek and Xiao Li and Daniel G. MacArthur and Jared L. Nedzel and Duyen T. Nguyen and Michael S. Noble and Ayellet V. Segrè and Casandra A. Trowbridge and Taru Tukiainen and Nathan S. Abell and Brunilda Balliu and Ruth Barshir and Omer Basha and Alexis Battle and Gireesh K. Bogu and Andrew Brown and Christopher Brown and Stephane E. Castel and Lin Chen and Colby Chiang and Donald F. Conrad and Farhan N. Damani and Joe R. Davis and Olivier Delaneau and Emmanouil T. Dermitzakis and Barbara E. Engelhardt and Eleazar Eskin and Pedro G. Ferreira and Laure Frésard and Eric R. Gamazon and Diego Garrido-Martín and Ariel DH Gewirtz and Genna Gliner and Michael J. Gloudemans and Roderic Guigó and Ira M. Hall and Buhm Han and Yuan He and Farhad Hormozdiari and Cédric Howald and Brian Jo and Eun Yong Kang and Yungil Kim and Sarah Kim-Hellmuth and Tuuli Lappalainen and Gen Li and Xin Li and Boxiang Liu and Serghei Mangul and Mark I. McCarthy and Ian C. McDowell and Pejman Mohammadi and Jean Monlong and Stephen B. Montgomery and Manuel Muñoz-Aguirre and Anne W. Ndungu and Andrew B. Nobel and Meritxell Oliva and Halit Ongen and John Palowitch and Nikolaos Panousis and Panagiotis Papasaikas and YoSon Park and Princy Parsana and A. J. Payne and Christine B. Peterson and Jie Quan and Ferrán Reverter and Chiara Sabatti and Ashis Saha and Michael Sammeth and Alexandra J. Scott and Andrey A. Shabalin and Reza Sodaei and Matthew Stephens and Barbara E. Stranger and Benjamin J. Strober and Jae Hoon Sul},
  title = {{Exploring the phenotypic consequences of tissue specific gene expression variation inferred from GWAS summary statistics}},
  booktitle = {Nature Communications},
  year = {2018},
  doi = {10.1038/s41467-018-03621-1},
}
Figure