A large-scale transcriptome-wide association study (TWAS) of ten blood cell phenotypes reveals complexities of TWAS fine-mapping

TL;DR: In this paper, the authors performed a transcriptome-wide association study (TWAS) using PrediXcan to systematically investigate the association between genetically-predicted gene expression and hematological measures in 54,542 individuals of European ancestry from the Genetic Epidemiology Research on Adult Health and Aging cohort.

Abstract: Hematological measures are important intermediate clinical phenotypes for many acute and chronic diseases. Hematological measures are highly heritable, and although genome-wide association studies (GWAS) have identified thousands of loci containing trait-associated variants, the causal genes underlying these associations are often uncertain. To better understand the underlying genetic regulatory mechanisms, we performed a transcriptome-wide association study (TWAS) using PrediXcan to systematically investigate the association between genetically-predicted gene expression and hematological measures in 54,542 individuals of European ancestry from the Genetic Epidemiology Research on Adult Health and Aging (GERA) cohort. We found 239 significant gene-trait associations with hematological measures. Among this set of 239 associations, we replicated 71 at p < 0.05 with same direction of effect for the blood cell trait in a meta-analysis of TWAS results consisting of up to 35,900 European ancestry individuals from the Womens Health Initiative (WHI), the Atherosclerosis Risk in Communities Study (ARIC), and BioMe Biobank. We further attempted to refine this list of candidate genes by performing conditional analyses, adjusting for individual variants previously associated with these hematological measures, and performed further fine-mapping of TWAS loci. To assist with the interpretation of TWAS findings, we designed an R Shiny application to interactively visualize TWAS results, one genomic locus at a time, by integrating our TWAS results with additional genetic data sources (GWAS, TWAS from other gene expression reference panels, conditional analyses, known GWAS variants, etc.). Our results and R Shiny application highlight frequently overlooked challenges with TWAS and illustrate the complexity of TWAS fine-mapping efforts. Author SummaryTranscriptome-wide association studies (TWAS) have shown great promise in furthering our understanding of the genetic regulatory mechanisms underlying complex trait variation. However, interpreting TWAS results can be incredibly complex, especially in large-scale analyses where hundreds of signals appear throughout the genome, with multiple genes often identified in a single chromosomal region. Our research demonstrates this complexity through real data examples from our analysis of hematological traits, and we provide a useful web application to visualize TWAS results in a broadly approachable format. Together, our results and web application illustrate the importance of interpreting TWAS studies in context and highlight the need to carefully examine results in a region-wide context to draw reasonable conclusions and formulate mechanistic hypotheses.

Summary

Introduction

• Hematological measures (red cell, white cell, and platelet traits) have a critical role in oxygen transport, immunity, infection, thrombosis, and hemostasis and are associated with many acute and chronic diseases, including autoimmunity, asthma, cardiovascular disease, and COVID-19 [1-5].
• Unfortunately, these individual SNP-based GWAS make it difficult to identify regulatory variants with small effect sizes which in aggregate impact the same gene, even in very large sample sizes, and they identify regions of associated variants whose biological function is often not clear [6].
• Hematological phenotypes are particularly good candidates for TWAS analysis due to the availability of large RNA-sequencing datasets in a relevant tissue type, high heritability across traits, and the large number of known genetic associations, most with poorly understood mechanisms and target genes.

Results

• The authors applied the PrediXcan method to identify expression-trait associations using individual level genotype and phenotype data from the GERA non-Hispanic white ethnic group.
• CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity.
• The copyright holder for this preprintthis version posted February 23, 2021.
• Marginal TWAS result displayed in (A), with Black colored genes and variants denoting those previously reported by GWAS, blue variants denote those not previously reported as GWAS sentinel variants.
• (B) is a mirrored-Manhattan locus-zoom plot displaying genes connected to their predictive model variants with TWAS results (top panel) and GWAS results .

Discussion

• The authors performed a large-scale TWAS using PrediXcan on 54,542 GERA individuals of European ancestry and present compelling evidence that results from marginal TWAS analyses alone cannot illuminate causal genes at loci for complex traits.
• The copyright holder for this preprintthis version posted February 23, 2021.
• CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity.
• Additionally, results for these two genes differ slightly by reference panel.
• While the authors show that TWAS may help in some cases to pinpoint likely causal genes, they emphasize the need for investigators to carefully interpret TWAS results alone, out of context.

Materials and Methods

• CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity.
• Genotyping was performed using the Illumina GSA array (~640.000 variants) and genotype data were imputed using the “1000G Phase 3 v5” reference panel.
• In total, 8,455 European ancestry participants with hematological phenotypes were included in the analysis.
• In order to replicate the conditionally significant gene-trait association, the authors tested each association via a meta-analysis of the ARIC, WHI, and BioMe cohorts.

FOCUS:

• The authors used the Fine-mapping Of CaUsal gene Sets [15] software to fine-map TWAS statistics at genomic risk regions.
• CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity.
• BioRxiv preprint 28 has not yet been assigned to a locus and continue in this fashion until all statistically significant TWAS genes have been assigned to a locus.
• The authors excluded those without a valid date of blood cell count measurement, with age < 18 years, or with discordant genotypic and phenotypic sex, as well as .
• Specifically, the authors computed the eigenvalue decomposition on the GERA sample outside of the cpgen script (for each phenotype), and then subsequently loaded the appropriate eigenvectors and eigenvalues into the program, modifying the script so that it could take these eigenvectors and eigenvalues as input.

Full text

1
Full title: A large-scale transcriptome-wide association study (TWAS) of ten blood cell
phenotypes reveals complexities of TWAS fine-mapping
Short title: TWAS fine-mapping of blood cell phenotypes
Authors
Amanda L Tapia MS
1
, Bryce T Rowland BS
1
, Jonathan D Rosen MS
1
, Michael Preuss PhD
2
, Kris
Young PhD
3
, Misa Graff PhD
3
, Hélène Choquet PhD
4
, David J Couper PhD
1
, Steve Buyske PhD
5
,
Stephanie A Bien PhD
6
, Eric Jorgenson PhD
4
, Charles Kooperberg PhD
6
, Ruth J.F. Loos PhD
2
,
Alanna C Morrison PhD
7
, Kari E North PhD
3
, Bing Yu PhD
7
, Alexander P Reiner MD
8
, Yun Li
PhD
9,1,10*
, Laura M Raffield PhD
9*
*Contributed equally to this work
1
Department of Biostatistics, University of North Carolina, Chapel Hill, NC, USA,
2
The Charles
Bronfman Institute for Personalized Medicine, Icahn School of Medicine at Mount Sinai, New
York, NY, USA,
3
Department of Epidemiology, University of North Carolina, Chapel Hill, NC, USA,
4
Division of Research, Kaiser Permanente Northern California, Oakland, CA, USA,
5
Department
of Statistics, Rutgers University, Piscataway, NJ, USA,
6
Division of Public Health Sciences, Fred
Hutchinson Cancer Research Center, Seattle, WA, USA,
7
Human Genetics Center, Department of
Epidemiology, Human Genetics, and Environmental Sciences, School of Public Health, The
University of Texas Health Science Center at Houston, Houston, TX, USA,
8
Department of
Epidemiology, University of Washington, Seattle, WA, USA,
9
Department of Genetics, University
of North Carolina, Chapel Hill, NC, USA,
10
Department of Computer Science, University of North
Carolina at Chapel Hill, Chapel Hill, NC, USA
Correspondence: Laura M. Raffield, PhD
Assistant Professor, Department of Genetics
University of North Carolina at Chapel Hill, Chapel Hill, NC, 27599
laura_raffield@unc.edu
.CC-BY 4.0 International licenseavailable under a
(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made
The copyright holder for this preprintthis version posted February 23, 2021. ; https://doi.org/10.1101/2021.02.23.432444doi: bioRxiv preprint

2
Abstract
Hematological measures are important intermediate clinical phenotypes for many acute
and chronic diseases. Hematological measures are highly heritable, and although genome-wide
association studies (GWAS) have identified thousands of loci containing trait-associated
variants, the causal genes underlying these associations are often uncertain. To better
understand the underlying genetic regulatory mechanisms, we performed a transcriptome-
wide association study (TWAS) using PrediXcan to systematically investigate the association
between genetically-predicted gene expression and hematological measures in 54,542
individuals of European ancestry from the Genetic Epidemiology Research on Adult Health and
Aging (GERA) cohort. We found 239 significant gene-trait associations with hematological
measures. Among this set of 239 associations, we replicated 71 at p < 0.05 with same direction
of effect for the blood cell trait in a meta-analysis of TWAS results consisting of up to 35,900
European ancestry individuals from the Women’s Health Initiative (WHI), the Atherosclerosis
Risk in Communities Study (ARIC), and BioMe Biobank. We further attempted to refine this list
of candidate genes by performing conditional analyses, adjusting for individual variants
previously associated with these hematological measures, and performed further fine-mapping
of TWAS loci. To assist with the interpretation of TWAS findings, we designed an R Shiny
application to interactively visualize TWAS results, one genomic locus at a time, by integrating
our TWAS results with additional genetic data sources (GWAS, TWAS from other gene
expression reference panels, conditional analyses, known GWAS variants, etc.). Our results and
R Shiny application highlight frequently overlooked challenges with TWAS and illustrate the
complexity of TWAS fine-mapping efforts.
.CC-BY 4.0 International licenseavailable under a
(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made
The copyright holder for this preprintthis version posted February 23, 2021. ; https://doi.org/10.1101/2021.02.23.432444doi: bioRxiv preprint

3
Author Summary
Transcriptome-wide association studies (TWAS) have shown great promise in furthering our
understanding of the genetic regulatory mechanisms underlying complex trait variation.
However, interpreting TWAS results can be incredibly complex, especially in large-scale
analyses where hundreds of signals appear throughout the genome, with multiple genes often
identified in a single chromosomal region. Our research demonstrates this complexity through
real data examples from our analysis of hematological traits, and we provide a useful web
application to visualize TWAS results in a broadly approachable format. Together, our results
and web application illustrate the importance of interpreting TWAS studies in context and
highlight the need to carefully examine results in a region-wide context to draw reasonable
conclusions and formulate mechanistic hypotheses.
Introduction
Hematological measures (red cell, white cell, and platelet traits) have a critical role in
oxygen transport, immunity, infection, thrombosis, and hemostasis and are associated with
many acute and chronic diseases, including autoimmunity, asthma, cardiovascular disease, and
COVID-19 [1-5]. Genome-wide association studies (GWAS) have identified thousands of loci
containing such trait-associated variants, and previous Mendelian randomization and
phenome-wide association study analyses have highlighted the likely causal role of blood cell
trait-associated genetic variants in a variety of disorders, including autoimmune conditions and
.CC-BY 4.0 International licenseavailable under a
(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made
The copyright holder for this preprintthis version posted February 23, 2021. ; https://doi.org/10.1101/2021.02.23.432444doi: bioRxiv preprint

4
coronary heart disease [1-3]. Unfortunately, these individual SNP-based GWAS make it difficult
to identify regulatory variants with small effect sizes which in aggregate impact the same gene,
even in very large sample sizes, and they identify regions of associated variants whose
biological function is often not clear [6]. Thus, utilizing a gene-based method to aggregate the
effect of multiple regulatory variants may increase the study power to identify novel trait-
associated loci and elucidate mechanisms of biological function.
A transcriptome-wide association study (TWAS) is one gene-based method which
systematically investigates the association between genetically predicted gene expression and
phenotypes of interest [6-9]. Here, we report results from a large TWAS of hematological
measures using the PrediXcan method [6] to analyze data from 54,542 individuals of European
ancestry from the Genetic Epidemiology Research on Adult Health and Aging (GERA) cohort
(our discovery data set) [10] [11]. Hematological phenotypes are particularly good candidates
for TWAS analysis due to the availability of large RNA-sequencing datasets in a relevant tissue
type, high heritability across traits, and the large number of known genetic associations, most
with poorly understood mechanisms and target genes. We perform this analysis using whole
blood RNA-sequencing in 922 individuals from the Depression Genes and Networks (DGN) [12]
study as our primary reference panel. After association analysis of imputed gene transcript
levels with hematological indices in GERA, we performed conditional analyses, adjusting for
variants previously identified to affect hematological measures, to evaluate if TWAS-identified
genes represented novel statistical signals or were primarily driven by variants known from
GWAS [3]. These direct conditional analyses represent a major advantage of the use of
individual level data for our TWAS analyses, since these conditional tests could not be
.CC-BY 4.0 International licenseavailable under a
(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made
The copyright holder for this preprintthis version posted February 23, 2021. ; https://doi.org/10.1101/2021.02.23.432444doi: bioRxiv preprint

5
performed as easily or accurately using summary statistic-based methods. We replicated our
significant set of gene-trait associations in a meta-analyzed sample of TWAS results containing
18,100 individuals from the Women’s Health Initiative (WHI), 9,345 individuals from the
Atherosclerosis Risk in Communities Study (ARIC), and 8,455 individuals from Mount Sinai
BioMe Biobank (BioMe), all of European ancestry (Supplementary Table 1). We also compared
the TWAS results from the primary DGN reference panel to three additional reference panels
(whole blood and Epstein-Barr virus (EBV) transformed lymphocytes from the Genotype-Tissue
Expression (GTEx) Project [13], and monocytes from the Multi-Ethnic Study of Atherosclerosis
(MESA)[14]); these are considered secondary reference panels due to their smaller sample
sizes. These additional analyses helped us to determine if relevant tissues with smaller sample
sizes support our primary TWAS findings with DGN.
We employ several strategies to improve our understanding and interpretation of
complex genomic regions containing multiple TWAS-identified genes. First, we used FOCUS
(fine-mapping of causal gene sets [15]) to seek to identify a set of causal genes within genomic
loci containing multiple significant TWAS gene-trait associations. FOCUS is a software used to
fine-map TWAS statistics at genomic risk regions, while accounting for linkage disequilibrium
(LD) among variants and predicted expression correlation among genes at those risk regions.
Second, we present a novel web-based tool for integrating and visualizing TWAS and GWAS
results, as well as results from multiple expression reference datasets. Additionally, we discuss
frequently overlooked challenges of TWAS interpretation, such as failure to consider the
number of proximal genes which cannot be accurately imputed with a given reference panel,
but which may still be influenced by variants identified in GWAS studies. Our results illustrate
.CC-BY 4.0 International licenseavailable under a
(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made
The copyright holder for this preprintthis version posted February 23, 2021. ; https://doi.org/10.1101/2021.02.23.432444doi: bioRxiv preprint

Frequently asked questions

Q1. What are the contributions in "Full title: a large-scale transcriptome-wide association study (twas) of ten blood cell phenotypes reveals complexities of twas fine-mapping short title: twas fine-mapping of blood cell phenotypes authors" ?

Authors Amanda L Tapia MS1, Bryce T Rowland BS1, Jonathan D Rosen MS1, Michael Preuss PhD2, Kris Young PhD3, Misa Graff PhD3, Hélène Choquet PhD4, David J Couper PhD1, Steve Buyske PhD5, Stephanie A Bien PhD6, Eric Jorgenson PhD4, Charles Kooperberg PhD6, Ruth J. F. Loos PhD2, Alanna C Morrison PhD7, Kari E North PhD3, Bing Yu PhD7, Alexander P Reiner MD8, Yun Li PhD9,1,10 *, Laura M Raffield PhD9 * * Contributed equally to this work 

Q2. What are the future works mentioned in the paper "Full title: a large-scale transcriptome-wide association study (twas) of ten blood cell phenotypes reveals complexities of twas fine-mapping short title: twas fine-mapping of blood cell phenotypes authors" ?

Joint/multiple tissue TWAS approaches such as UTMOST [ 9 ] and MR-JTI [ 8 ] could be employed in the future to assess the relevance of other tissues at blood-cell related loci.