FIVEx: an interactive eQTL browser across public datasets

Abstract

Summary

Expression quantitative trait loci (eQTLs) characterize the associations between genetic variation and gene expression to provide insights into tissue-specific gene regulation. Interactive visualization of tissue-specific eQTLs or splice QTLs (sQTLs) can facilitate our understanding of functional variants relevant to disease-related traits. However, combining the multi-dimensional nature of eQTLs/sQTLs into a concise and informative visualization is challenging. Existing QTL visualization tools provide useful ways to summarize the unprecedented scale of transcriptomic data but are not necessarily tailored to answer questions about the functional interpretations of trait-associated variants or other variants of interest. We developed FIVEx, an interactive eQTL/sQTL browser with an intuitive interface tailored to the functional interpretation of associated variants. It features the ability to navigate seamlessly between different data views while providing relevant tissue- and locus-specific information to offer users a better understanding of population-scale multi-tissue transcriptomic profiles. Our implementation of the FIVEx browser on the EBI eQTL catalogue, encompassing 16 publicly available RNA-seq studies, provides important insights for understanding potential tissue-specific regulatory mechanisms underlying trait-associated signals.

Availability and implementation

A FIVEx instance visualizing EBI eQTL catalogue data can be found at https://fivex.sph.umich.edu. Its source code is open source under an MIT license at https://github.com/statgen/fivex.

Supplementary information

Supplementary data are available at Bioinformatics online.

1 Introduction

Expression quantitative trait loci (eQTLs) are an important piece of the puzzle to understand the regulatory mechanisms underlying genetic associations (Gallagher and Chen-Plotkin, 2018). Continuing advances in genomic technology have allowed researchers to generate enormous amounts of molecular profiles across many individuals and tissues. For example, the Genotype-Tissue Expression (GTEx) Consortium analyzed transcriptomic profiles of 49 different tissues across 838 samples and identified >4 million eQTLs (Aguet et al., 2020). Recently, the EBI eQTL catalogue uniformly processed 16 RNA-seq datasets over 95 tissues, including GTEx, to systematically identify and fine-map eQTLs and transcript-usage QTLs (Kerimov et al., 2021). The sheer number of eQTLs produced in such datasets require scalable, custom-designed visualization tools as aids for interpretation and analysis which will allow the exploration of a wide range of clinically relevant hypotheses, such as interpreting potential regulatory mechanisms in individual genome-wide association study (GWAS) signals (Roselli et al., 2018; Yengo et al., 2018), the understanding tissue-specific epigenetic architecture of complex traits (Ehrlich et al., 2019), and pinpointing likely causal variants by colocalizing GWAS and eQTL signals (Liu et al., 2018; Wu et al., 2019).

Interactive web applications, such as the GTEx Portal (https://gtexportal.org) facilitate the functional interpretation of disease-associated regulatory variants. However, existing tools mainly focus on providing regional summaries of cis-eQTLs rather than tailored information relevant to functional interpretation. These tools also do not provide connections to other relevant online resources, such as PheWeb (Gagliano Taliun et al., 2020), BRAVO (http://bravo.sph.umich.edu), gnomAD (Karczewski et al., 2020), or the OpenTargets Platform (Carvalho-Silva et al., 2019), making it challenging for users to gain a holistic understanding of the functional mechanisms underlying gene regulation.

To facilitate functional interpretation of regulatory variants from population-scale transcriptomic resources, we developed FIVEx (Functional Interpretation and Visualization of Expression), an eQTL-focused web application that leverages the widely used tools LocusZoom.js (Boughton et al., 2021) and LD server (https://github.com/statgen/LDServer). FIVEx visualizes the genomic landscape of cis-eQTLs and cis-sQTLs across multiple tissues, focusing on a variant, gene, or genomic region. FIVEx is designed to aid the interpretation of the regulatory functions of genetic variants by providing answers to functionally relevant questions, for example, (1) how likely is a specific genetic variant to be causal for a cis-eQTL/sQTL; (2) is a cis-eQTL/sQTL tissue-specific or shared across tissues; (3) what is the linkage disequilibrium (LD) structure around the variant or gene; (4) which nearby genes are likely co-regulated by the variant and in which tissues; and (5) is there additional information from other resources, such as biobank-based PheWAS results or regulatory genomic resources, that corroborates functional interpretation. FIVEx provides interactive visualizations of cis-eQTLs/sQTLs, capitalizing on functional interpretation and connecting to relevant external resources (Supplementary Figs. S1–S20). We expect FIVEx to serve as a useful community resource of public eQTL/sQTL datasets, complementing the GTEx Portal and other widely used web tools.

2 Key features

The primary goal of FIVEx’s visualizations is to aid exploratory analysis to help investigators interpret the regulatory functions of the variants identified from GWAS or other genetic studies. FIVEx allows investigators to query an eQTL database for a variant, gene, or region to interactively visualize multi-tissue cis-eQTLs/sQTLs from various viewpoints, either in a single-variant view of all associations, or in multiple LocusZoom.js cis-eQTL/sQTL plots for selected genes, tissues, and datasets. When a variant is queried, FIVEx visualizes the landscape of cis-eQTLs associated with the variant across all nearby genes and all tissues. FIVEx also highlights the variants most likely to have causal regulatory effects. When a gene or a region is queried, FIVEx offers multi-tissue and/or multi-gene LocusZoom.js visualizations of strongly associated cis-eQTLs, along with a summary list of variants that most likely regulate genes across the tissues. By effectively visualizing GTEx data, FIVEx can help unravel potential tissue-specific regulatory effects underlying the associations.

FIVEx’s variant-centric visualization helps users interpret the potentially causal role of a variant in tissue-specific regulation. For example, if a user queries for rs12740374, the chromosome 1 variant most strongly associated with cardiovascular diseases (coronary atherosclerosis, ischemic heart disease, and myocardial infarction) in UK Biobank (Supplementary Fig. S1), FIVEx provides a comprehensive view of all cis-eQTLs of nearby genes, including SORT1, CELSR2, and PSRC1 across 16 public eQTL datasets uniformly processed by EBI eQTL catalogue (Fig. 1A). FIVEx provides posterior inclusion probabilities (PIPs) from SuSiE (Wang et al., 2020) to inform whether the variant is likely a causal variant (Fig. 1B). For example, the three most significant eQTL P-values at rs12740374 are found in SORT1 in GTEx liver (P = 5.0 × 10⁻³⁹), PSRC1 in Estonian Biobank blood (Lepik et al., 2017) (P = 6.0 × 10⁻²⁷), and PSRC1 in ROSMAP brain (Ng et al., 2017) (P = 2.5 × 10⁻²⁶). However, the PIP values for the second (0.032) and the third (0.11) are relatively low, suggesting that the variant may not be causal (Supplementary Fig. S2). However, the next two strongest signals, PSRC1 (P = 3.5 × 10⁻²⁶, PIP = 0.97) and CELSR2 (P = 9.1 × 10⁻²⁴, PIP = 0.69) are both found in GTEx liver. Indeed, the liver is the only tissue that has PIP > 0.5 across SORT1, CELSR2, and PSRC1, suggesting liver-specific coregulation of these genes by rs12740374, consistent with previous findings (Musunuru et al., 2010; Schadt et al., 2008; Wang et al., 2018).

Fig. 1.

Examples of FIVEx views. (A) A variant-centric FIVEx view of eQTL P-values for rs12740374, the variant most strongly associated with self-reported high cholesterol level in UK Biobank near the SORT1 gene. This variant is a strong cis-eQTL regulating multiple genes (SORT1, PSRC1, and CELSR2), particularly in liver tissue. There is also a strong signal in blood. (B) The same view showing PIPs generated using SuSiE (Wang et al., 2020), highlighting the strong signals in liver tissue across the three genes, along with similar signals for CELSR2 in muscle. (C) A locus-centric FIVEx view of eQTL P-values in the region around CELSR2. Both liver and muscle share the same top variant in rs12740374, highlighted in purple, while the top signal in the blood is a different downstream variant (rs34293021) in low LD with rs12740374, providing strong evidence that the liver and blood signals are independent of each other

FIVEx’s region view can help us explore the cis-eQTL landscape around the variant of interest in more detail. Users can interactively add multiple panels of eQTL LocusZoom plots, with each panel showing eQTLs for one gene in one tissue. In the SORT1-PSRC1-CELSR2 locus, FIVEx helps visualize the three genes in GTEx liver to assess whether they indeed share the same peak cis-eQTL (Supplementary Figs. S3 and S4). It can also visualize multiple tissues and datasets for the same gene to examine the tissue-specific nature of cis-eQTLs. For example, CELSR2 shares a cis-eQTL peak (rs12740374) between liver and skeletal muscle, but not with other tissues, such as blood (Fig. 1C;  Supplementary Fig. S5), suggesting that skeletal muscle may be another key tissue relevant for the association rs12740374 and cardiovascular traits in addition to liver.

In addition to visualizing cis-eQTLs across multiple datasets, FIVEx can also show splice and transcript usage QTLs from the EBI eQTL catalogue generated with txrevise (Alasoo et al., 2019). The cis-sQTLs aid interpretation of transcript-specific genetic regulation associated with a GWAS variant. For example, rs12740374 alters the expression of SORT1 exon and splice junctions unique to a specific transcript ENST00000538502 in GTEx liver (Supplementary Fig. S6). No strong evidence of an sQTL was observed in other tissues or genes. This suggests that the liver-specific coregulation of SORT1-CELSR1-PSRC1 by rs12740374 may be mediated by a specific transcript of SORT1.

By leveraging externally linked tools, FIVEx provides useful ways to colocalize eQTLs and GWAS signals. Users can query a specific gene (e.g. SORT1) and select a specific top variant (e.g. rs12740374) (Supplementary Figs. S7 and S8). Navigating externally linked PheWAS, such as that for the UKB (Gagliano Taliun et al., 2020) highlights the traits associated with the variant (Supplementary Fig. S1). A more comprehensive colocalization can be performed through Open Targets Genetics. When a lead eQTL variant (e.g. rs12740374) was queried, it highlights all known GWAS lead signals in high LD (r² > 0.5) with the variant (Supplementary Fig. S9). Fine-mapped GWAS signals (e.g. FinnGen) can also be used to quantify the degree of certainty that the variant of interest is causal (Supplementary Fig. S10). Through interactive navigation across multiple web apps, FIVEx provides important insights to colocalize GWAS signals with publicly available eQTLs and guide hypotheses for underlying regulatory mechanisms.

A more detailed description of these example usages can be found in the Supplementary Text and in the online tutorial at https://fivex.sph.umich.edu/tutorial.

3 Discussion

Understanding the functions of trait-associated non-coding variants is becoming increasingly important as more genomes, transcriptomes, and epigenomes are sequenced. Gene regulation is believed to be involved in a large fraction of such associations, but there are limited resources which investigators can use to generate hypotheses for explaining regulatory mechanisms underlying association signals. FIVEx offers new interactive ways to visualize and summarize eQTLs and sQTLs in a tissue-specific manner by combining key features from LocusZoom and PheWeb, focusing on putative causal QTLs through PIPs (Supplementary Figs. S4 and S5). We expect FIVEx to aid in translating GWAS associations into underlying regulatory mechanisms by enabling the exploration of plausible hypotheses through our practical, intuitive user interface. As more online resources like FIVEx become available to address tailored scientific questions on functional variants, we expect that precise and integrative translation of genomic findings will be more accessible to the broader scientific community.