Role of expression quantitative trait loci (eQTL) in understanding genetic mechanisms underlying common complex diseases

Abstract

Attaining a complete understanding of the genetic architecture underlying common complex traits is challenging due to the substantial contributions of nongenetic factors and the involvement of numerous influencing genes. Genome-wide association studies (GWAS) have identified novel variants associated with such traits, but our understanding of the molecular genetic mechanisms underlying those associations remains limited. Additionally, variants without significant associations from GWAS can influence gene expression, contributing to individual-level variation in traits. This review summarizes the evolution, advancements in, and practical applications of expression quantitative trait loci analysis. Recent large-scale expression quantitative trait loci studies, often at the single-cell level, provide an opportunity to explain how at least some GWAS variants behave and to elucidate the mechanisms underlying individual-level variations. This approach can further be utilized to identify novel drug targets that are tailored to individuals harboring specific genotypes.

INTRODUCTION

It is challenging to fully explain the genetic architecture of human complex traits. Indeed, it is daunting to construct a complete picture of monogenic Mendelian disorders, wherein a single mutation accounts for a majority of symptoms, let alone for common complex diseases. This is due to 2 major factors. First, the nature of complex traits means they are influenced by environmental as well as genetic factors. For example, fatty liver disease is more common in individuals with obesity, a western diet, and a sedentary lifestyle than in those without said risk factors (Younossi et al., 2018, Eslam et al., 2020). Second, the molecular activities underpinning complex traits are usually regulated by multiple genes with varying effect strengths (Bellenguez et al., 2022, Mahajan et al., 2022, Yengo et al., 2022). That is, complex traits typically require complex interactions among many different biological pathways across multiple tissues and organs, the effects of which become more severe as our life span gets longer. This is aggravated by the population expansion of the human species since the out-of-Africa era, in which a lack of purifying selection excluded opportunities to eradicate variants with only moderate functionality (Lohmueller et al., 2008, Tennessen et al., 2012).

Genome-wide association studies (GWAS), first conducted in 2006, have become the major methodology by which geneticists uncover associations between genetic variants and traits. The first round of such studies was instrumental in revealing the genetic architecture of important chronic diseases (Klein et al., 2005, Duerr et al., 2006, Wellcome Trust Case Control, 2007); however, these initial studies also provoked disappointment in the field as the discovered variants were difficult to validate and as a set were able to explain only a fraction of observed phenotypic variations in the population. Over time, the scale of GWAS has grown; recently, we have witnessed million-participant-scale studies such as a GWAS of height (Keaton et al., 2024), and one of weight and obesity using whole genome sequencing data that included rare and common variants, which successfully established a complete picture of the investigated traits (Wainschtein et al., 2022). Still, there remains much room for further development. First, a functional interpretation is lacking for the majority of statistically associated variants, limiting our understanding of the molecular genetic pathways that culminate in the investigated traits. Second, variants published to date were mostly identified from individuals of European background, creating a critical requirement to illuminate variants among non-Europeans, including East Asians. Lastly, research into GWAS-derived polygenic risk scores has demonstrated that individuals with similar scores still tend to show variable trait susceptibility and severity.

The expression quantitative trait loci (eQTL) approach has the potential to offset the limitations of GWAS. Its concept is derived from quantitative trait locus (QTL), an analysis method used to locate genetic loci that regulate quantitative traits (eg, height, weight, crop yield, etc.) in crops, animals, and humans (Osborn et al., 1987, de Vries and Veerkamp, 2000, Schadt et al., 2003). The eQTL approach extends QTL analysis by assuming that gene expression in a given tissue is also genetically regulated (ie, the expression of a gene is itself regarded as a quantitative trait) (Fig. 1A). By identifying genetic variants (eQTL single nucleotide polymorphisms (eSNPs)) that alter the expression of nearby genes (eGenes), the eQTL approach can instantly elucidate a mechanism by which GWAS variants modulate phenotype in an individual-specific manner (Ko et al., 2024).

Fig. 1.

Overview of eQTL concepts and approaches. (A) Schematic illustration of a genetic variant associated with the expression of Gene X, which could be discovered by simultaneously sequencing the DNA and RNA obtained from multiple individuals. (B) Diverse eQTL approaches based on studying design. Conventional eQTLs (top left) are typically identified in adult, steady-state tissues using bulk RNA sequencing. Context-specific eQTLs (top right) exhibit variable effect sizes depending on the biological or environmental context. Single-cell eQTLs (bottom left) leverage single-cell RNA sequencing to detect regulatory associations at the level of individual cell types or cell states. Trans-eQTLs (bottom right) represent associations between a genetic variant and a gene located on a different chromosome or distant genomic region, often mediated by an intermediate cis-eQTL. CNS, central nervous system.

In this review, we will introduce notable eQTL studies showcasing the evolution of the field evolved over time and present emerging techniques. We further discuss the relationship between eQTL and GWAS more precisely to show that these 2 entities are more independent than previously thought. Finally, we introduce the practical utility of the eQTL approach, hopefully providing illuminating insights on this emerging and important technique in genetics.

MAIN

Evolution of eQTL Studies

Large-scale eQTL Projects

The first eQTL studies were conducted on animal or plant models. One notable example of the early but powerful use of eQTL was the identification of the genetic origin of phenotypic differences between inbred mouse lines (Schadt et al., 2005), an approach that has remained useful into recent years (Hsiao et al., 2020). The first landmark eQTL study on human tissues was the Genotype-Tissue Expression (GTEx; https://gtexportal.org/) project (Lonsdale et al., 2013, Chiang et al., 2017, Consortium et al., 2017, Li et al., 2017, Yang et al., 2017, Consortium, 2020, Kim-Hellmuth et al., 2020), funded by the National Institute of Health, which aimed to document gene expression from 54 nondiseased tissues across more than 1,000 individuals. All data are publicly accessible, with the portal providing comprehensive documentation of isoform-level gene expression in many human tissues alongside corresponding genotype data. Accordingly, GTEx has served as a gold standard for many subsequent studies. In addition, the project established important features of eQTLs that other studies should refer to, such as tissue specificity, enrichment in regulatory regions, and phenotype-associated variants. Another notable large-scale eQTL project is the eQTLGen consortium (Vosa et al., 2021), which focuses exclusively on blood tissue but includes a total of 31,684 individuals, and hence offers a comprehensive catalog of both cis- and trans-eQTLs in blood. Finally, the Metabrain resource is a large-scale eQTL meta-analysis of previously published human brain eQTL data (8,613 RNA-seq samples) that aims to identify cis- and trans-eQTLs in multiple brain region– and ancestry-specific datasets (de Klein et al., 2023) (Fig. 1B).

Utility of Context-specific eQTLs

Regulatory genetic effects are known to be context-specific. With regard to tissue specificity, the GTEx study revealed that eQTL tissue detection number follows a U-shaped curve, wherein eQTLs tend to be either highly specific to certain tissues or broadly shared across many tissues (GTEx Consortium, 2020). While the molecular mechanisms underlying this pattern are not well-understood, genetic variations in transcribed regions or within chromatin domains whose states are shared between tissues are more likely to have shared effects. Beyond tissue specificity, many studies have explored other context-specific eQTLs or dynamic eQTLs (eQTLs whose effects change in varying contexts), such as those active during development (Strober et al., 2019, Walker et al., 2020, Young et al., 2021) or in response to immune stimuli (Gutierrez-Arcelus et al., 2020), drug treatment (Zhong et al., 2023), cell stress (Ward et al., 2021), or disease (Ota et al., 2021, Yoo et al., 2021) (Fig. 1B). For instance, a study on 28 immune cell types from 416 donors with immune-mediated diseases identified cell-type- and disease-specific eQTLs, which were enriched in immune diseases (Ota et al., 2021). Additionally, a study on liver tissue from 293 donors with or without metabolic dysfunction–associated steatotic liver disease (MASLD) identified eQTLs that are exclusively active in patients, suggesting their potential as drug targets (Yoo et al., 2021). These findings suggest that many eQTLs cannot be detected by conventional assays using postmortem tissues (Strober et al., 2019), highlighting the importance of studying eQTLs in varied tissue and life course contexts.

eQTLs Resolved at the Single-cell Level

With advancements in high-throughput single-cell RNA sequencing (scRNA-seq), efforts to map eQTLs at the single-cell level are gaining traction (Jung and Lee, 2023). Bulk RNA-seq, which averages gene expression across tissues, inherently overlooks cellular heterogeneity within a tissue; in addition, identifying dynamic eQTLs using bulk methods is difficult due to limited disease-associated samples and challenges in modeling continuous cell states and contexts. scRNA-seq overcomes these limitations by enabling unbiased quantification of gene expression and cell states while preserving intercellular variability. Correspondingly, an increasing number of studies have been and are being published on single-cell eQTLs (sc-eQTLs) in a variety of contexts, including blood (van der Wijst et al., 2018; Nathan et al., 2022, Yazar et al., 2022), brain (Bryois et al., 2022, Fujita et al., 2024), lung (Natri et al., 2024), and induced pluripotent stem cells (Cuomo et al., 2020, Jerber et al., 2021) (Fig. 1B). One particularly notable study is the OneK1k project (Yazar et al., 2022), which analyzed scRNA-seq data from 1.27 million peripheral blood mononuclear cells collected from 982 donors and identified thousands of cell-type-specific and dynamic eQTLs. Among these cis-eQTLs, 19% were found to share the same causal locus as a GWAS risk association. Recent research from our group includes an sc-eQTL study on blood samples from patients with COVID-19 of varying severity, which identified regulatory variants and mechanisms linked to disease severity (Lee et al., 2024) and one on liver tissues from MASLD patients and identified genotype- and cell-state-specific sc-eQTLs that may offer prospective therapeutic targets (Hong et al., 2025). Finally, novel statistical methods are being developed to model the scRNA-seq data structure along with the nonlinear and dynamic genetic effects (Cuomo et al., 2022, Nathan et al., 2022, Kumasaka et al., 2023).

Value of Trans-eQTLs

Most studies have focused on variants that alter the expression of a nearby gene (ie, cis-eQTLs, conventionally defined as an eQTL less than 1 Mb from a gene’s transcription start site [TSS]), but some identified variants that regulate distant genes, known as trans-eQTLs (ie, eQTLs located more than 1 Mb from a TSS, including on a different chromosome). Trans-eQTLs typically have smaller effect sizes, are more context-specific, and are often enriched for complex trait relevance (GTEx Consortium, 2020) (Fig. 1B). At the same time, detecting trans-eQTLs is more challenging because they can theoretically affect any gene across the genome; this greatly expands the search space and increases the multiple testing burden, which in turn raises the risk of false positives and makes statistical significance harder to achieve (Vosa et al., 2021). Larger sample sizes are therefore required to detect robust trans signals (Mostafavi et al., 2023). The majority of trans-eQTLs are mediated through cis-regulatory effects on nearby genes (Pierce et al., 2014, Yang et al., 2017, Consortium, 2020), and mechanisms for those without cis-regulatory effects remain poorly understood (Umans et al., 2021).

Recent Technical Advancements

Conventional statistical models for bulk-eQTL mapping rely on linear regression models, testing the linear additive effect of allele dosage on gene expression. In these models, covariates, including genotype principal components, probabilistic estimation of expression residuals, and other known factors such as age, sex, and batch, are commonly included to account for confounding effects. In addition to regression models, some studies utilize linear mixed models, which incorporate random-effect terms that could account for population structure or cryptic relatedness. For single-cell studies, the gene expression pattern is typically much sparser than in bulk studies and does not follow a normal distribution, characteristics that hinder the use of linear models. Many studies have addressed these issues through the pseudo bulk approach, in which the aggregated gene expression of cells from the same donor and cell type serves as a single bulk sample, thereby allowing methodologies for bulk-eQTL mapping to be applied. However, recent studies have begun to explore ways to utilize the additional information conferred by single-cell profiles and model individual single cells. Such approaches include the Poisson mixed-effects model (Nathan et al., 2022), the linear mixed model in CellRegMap (Cuomo et al., 2022), the Gaussian process latent-variable model in GASPACHO (Kumasaka et al., 2023), and scalable and efficient implementations with set-based tests for rare variants in SAIGE-QTL (Zhou et al., 2024). Refer to this paper for detailed descriptions of various tools for each step of eQTL mapping (Ko et al., 2024).

Application of the eQTL Approach

Interpretation of GWAS Loci

As described above, GWAS studies have identified numerous genetic loci associated with various human traits; however, the majority of this variation lies in noncoding regions, making it difficult to understand the associated molecular mechanisms (GTEx Consortium, 2020). Given the enrichment of GWAS signals in regulatory elements, connecting these signals with eQTLs provides an easy avenue for explaining their mechanisms. Especially, using gene expression as a bridge to better understand the functional mechanisms connecting genetic variants to organismal phenotypes helps fine-map a GWAS association signal to a specific target gene.

Several approaches to combining eQTL and GWAS data are available. One approach, called colocalization, tests whether the same variant causally affects both gene expression (eQTL) and trait (GWAS) (Wallace, 2021). Another approach, broadly termed molecular associations, focuses on the associated gene (eGene) rather than the specific variant (eSNP). For example, a transcriptome-wide association study integrates eQTL effect sizes and GWAS genotypes to assess the genetic influence on gene expression levels in relation to a trait (Gamazon et al., 2015). A third method with a similar approach is Mendelian randomization, which uses eQTL data as an instrumental variable to assess the effect of gene expression on trait outcomes (Porcu et al., 2019). While none of these methods can definitively establish the causality of a gene expression difference, colocalization helps eliminate false overlapping signals caused by linkage disequilibrium, and molecular association approaches provide more precise measurements of the correlated effect (Umans et al., 2021).

Leveraging eQTLs for Understanding Diseases and Developing Personalized Drug Targets

Another utility of eQTL analysis is to illuminate disease mechanisms; in particular, the tissue-specific patterns of eQTLs indicate which biological processes or cell types are affected by the regulatory effects of the corresponding variants. For example, while BIN1 was expressed across various cell types, the eQTL signal in BIN1 that colocalized with Alzheimer’s disease (AD) GWAS was observed exclusively in microglia (Fujita et al., 2024). Given the essential role of microglia in AD, this suggests that genotype-specific regulation of BIN1 may contribute to disease progression. Outside of the disease context, eQTLs can provide insight into the mechanisms underlying gene expression regulation generally. Perhaps, the most parsimonious explanation is via differential binding of transcription factors to eSNPs, which regulates the activity of eQTLs and explains their context specificity (Flynn et al., 2022). A representative study focused on an eQTL involving SORT1, wherein a substitution variant alters a C/EBP transcription factor–binding site with the effect of reducing eGene expression and elevating serum cholesterol level (Musunuru et al., 2010). Lastly, eQTLs allow the documentation of interactions between genotype and environment. That is, environmental factors modify disease risk and cellular states, which in turn could be detected by changes in eQTL signals. For example, a study on induced pluripotent stem cell (iPSC)-derived cardiomyocytes applied dynamic-eQTL analysis to identify genetic determinants of the response to oxygen deprivation (Ward et al., 2021). Recent studies have further identified interaction eQTLs, where eQTL effect size varied across disease and cellular states, and demonstrated their regulation by upstream transcription factors, highlighting the role of interaction eQTLs in conferring differential disease susceptibility (Lee et al., 2024, Hong et al., 2025). Beyond the mechanistic level, integration of eQTL data with other data could offer new drug targets by pinpointing the most likely dysregulated gene in a disease (Bryois et al., 2022, Yazar et al., 2022, Fujita et al., 2024 Natri et al., 2024). For example, an sc-eQTL study on AD brains utilized transcriptome-wide association study to identify 21 eGenes that colocalized with AD risk loci and 24 novel microglial genes (Fujita et al., 2024), which could be potential drug targets for AD. Disease-critical genes or pathways can also be prioritized through sophisticated analysis methods, such as network-based methods (Vitali et al., 2019; van der Wijst et al., 2020; Sadler et al., 2023). A third powerful approach to identifying drug targets is the combination of eQTL analysis with experimental perturbation and functional studies. Two previous studies by our team identified a disease-specific response eQTL in AGXT2 (Yoo et al., 2021) and a metabolic state–interacting sc-eQTL in EFHD1 (Hong et al., 2025), for which functional validation supported their roles as genotype-dependent protective genes with potential therapeutic applications against MASLD.

Limitations

Low Colocalization Rate

Transcriptome-level traits alone pose challenges for linking eQTLs to phenotypes, particularly in the absence of supporting GWAS evidence. Despite active efforts to integrate eQTL and GWAS signals, a large fraction of GWAS signals remain unexplained by eQTLs. Several explanations for the low colocalization rate have been proposed. One study systematically compared eQTL and GWAS signals and suggested the approaches to systematically prioritize different loci and genes (Mostafavi et al., 2023), with eQTL signals being concentrated closer to TSSs, whereas GWAS signals are skewed more toward enhancers. A second study proposed that conventional eQTLs, discovered in healthy, adult, steady-state tissues, do not necessarily align with the patterns observed in disease-associated contexts (Umans et al., 2021). These eQTLs often lack tissue-specific effects and typically involve genes under weak selection, which suggests changes in their expression to have minimal functional impact and impose low fitness costs. Notably, GWAS-associated genes are enriched with functional annotations, are under strong selective pressure, and exhibit more complex regulatory landscapes than eQTL genes. The study proposed that mapping dynamic eQTLs during cellular processes could reveal hidden regulatory variations linked to diseases. In addition to these explanations, other studies have indicated that genetic variants might influence cellular status through mechanisms beyond gene expression, such as chromatin accessibility or splicing (Li et al., 2016; Aracena et al., 2024), or that redundant enhancers could buffer the impact of trait-associated variants on gene expression, making the eQTLs more difficult to detect (Wang and Goldstein, 2020).

Difficulty in Causal Variant Definition

Any given genetic variant is tightly tangled with other nearby variants due to linkage disequilibrium structure, leading to them being inherited together as a haplotype. Because of this genetic structure, canonical eQTL analyses calculate each single nucleotide polymorphism (SNP) separately and often identify multiple SNPs in a region as marginally associated with a trait. Disentangling causal variants from noncausal ones therefore presents a key challenge. Correspondingly, another challenge in fine-mapping is estimating the number of causal variants within a region (Zou et al., 2019). Several statistical approaches have been developed to address these issues. One simplifying method is forward stepwise conditional regression, which involves conditioning on the lead SNP by including it as a covariant and testing the remaining SNPs within the region of interest. Notably, as the number of conditioning steps increases, this method may pose increased risk of false-positive calls and reduced power to detect secondary SNPs.

In recent years, many studies have applied Bayesian methods to search for the multiple set of causal variants by computing the posterior probabilities of models. Bayesian models for determining the causality of variants rely on a probability of a specific model, and leverage the prior probability of a model and the likelihood of the data to calculate the posterior probabilities from which the probability of a SNP to be causal (posterior inclusion probability: PIP) and the set of credible causal SNPs are generated. Each available tool incorporates different prior assumptions and uses a different strategy for calculation. The CAVIAR model (Hormozdiari et al., 2014) applies joint modeling of multiple causal variants based on uniform priors over SNPs and limits maximum causal variants to lower the computational burden, but uses an exhaustive searching strategy for joint modeling that constrains its capacity for scaling. FINEMAP (Benner et al., 2016) overcame the scaling problem by leveraging a shotgun stochastic searching strategy and allows configuration of prior assumptions through providing effect sizes. A more recent model, SuSiE (Wang et al., 2020), applies sequential and refinement modeling by iterating single-effect regression and updating the PIP. This model supports a more scalable approach, faster speed, and customization of the number of causal variants, which aspects were limited in CAVIAR and FINEMAP (Table 1, Supplementary Table 1).

Table 1.

Tools for fine-mapping causal variants

Software	Model type	Searching	Joint inference	No. of causal variants	Output	Reference
PLINK/GCTA/custom R	Frequentist	Stepwise forward	No	1 per step	P-value, beta	.
CAVIAR	Baysian	Exhaustive	Yes	Limited	CLPP, credible sets	Hormozdiari et al. (2014)
FINEMAP	Baysian	Shotgun stochastic search	Yes	Limited	PIP, credible sets	Benner et al. (2016)
SuSiE	Baysian	Sequential and variational update	Yes	Flexible	PIP, credible sets, and alpha matrix	Wang et al. (2020)

Abbreviations: CLPP, colocalization posterior probability.

Difficulty in Functional Validation

Despite advancements in statistical approaches for identifying putative trait-associated eQTLs, validation and experimental replication of these signals remain challenging. Incorporating animal models may allow for the validation of putative trait-associated eQTLs, but eQTLs located in noncoding regions are difficult to investigate in model organisms due to weak conservation with human noncoding regions. In addition, given that eQTLs result from complex genomic regulatory mechanisms that are highly context-specific, replicating eQTLs in cellular models is challenging. Nonetheless, several efforts have been made to validate eQTL signals and interpret the underlying biological mechanisms. Leveraging well-established in vitro models can sometimes be an effective approach to observe eQTLs that represent specific molecular mechanisms. Cuomo et al. (2020) performed sc-eQTL analysis on differentiating iPS cell lines from 125 donors and elucidated dynamic eQTLs across developmental stages and molecular markers that varied among individuals. Organoid models provide opportunities to better replicate the physiological context; in one study, brain organoids were differentiated and subjected to oxygen manipulation experiments to investigate the mechanisms connecting gene regulation to environmental stress (Umans and Gilad, 2025).

Perspectives: Future of the eQTL Approach

In Vitro Validation Experiments

There have been numerous experimental efforts aiming to pinpoint causal eQTLs or validate eQTL disease associations (Lee et al., 2024), such as through massively parallel reporter assays (MPRA) and clustered regularly interspaced short palindromic repeats (CRISPR)-based genetic screening approaches (Fig. 2A). MPRA is based on the well-established reporter gene assay, in which the functional effects of SNPs are validated by incorporating them into a promoter and measuring changes in reporter gene expression (Tewhey et al., 2016). Due to its high-throughput nature and straightforward interpretation, MPRA has been widely used to fine-map functional variants (Ulirsch et al., 2016, Abell et al., 2022). However, the approach has distinct limitations, including its reliance on reporter genes and the introduction of exogenous constructs, which hinder the ability of MPRA to link variants to their putative target genes and to replicate the endogenous genomic landscape on which those variants depend (McAfee et al., 2022). These limitations can be overcome by leveraging CRISPR-based screening. By characterizing the transcriptomes of phenotypes of cells that have been genetically or epigenetically modified at a specific genomic locus, this approach enables the functional characterization of noncoding elements and validation of whether disease-associated eQTLs contribute to disease phenotypes (Gasperini et al., 2019; Li et al., 2024; Yao et al., 2024). Recently, CRISPR screen methods have been combined with scRNA-seq to achieve higher throughput and resolution in expression profiling (Adamson et al., 2016, Dixit et al., 2016). Recent studies further expanded this approach by integrating pooled CRISPR perturbations with single-cell multimodal readouts—simultaneously profiling gene expression, cell surface proteins, or chromatin accessibility—to functionally characterize genetic variants and facilitate the interpretation of disease-associated loci (Rubin et al., 2019, Morris et al., 2023).

Fig. 2.

Various efforts to overcome the limitations of eQTLs. (A) Overview of the high-throughput experimental validation methods MPRA and CRISPR screening. MPRA evaluates the regulatory activity of thousands of DNA sequences simultaneously by linking each sequence to a unique barcode and measuring reporter gene expression. CRISPR-based screens assess whether targeted mutations or epigenetic modifications introduced at specific genomic loci influence gene expression. (B) Leveraging of multiomics datasets in QTL studies, in which different layers of molecular phenotypes are used as quantitative traits. Integrating multilayered omics data facilitates the functional interpretation of regulatory variants and their underlying mechanisms. (C) Incorporation of individuals from diverse ancestries in eQTL studies (top panel) and stratification of eQTL effects by cell types and context (bottom panel). Expanding population increases the discovery of ancestry-specific or shared regulatory variants, while expanding context diversity reveals context-dependent regulation and enhances phenotype relevance. AMR, admixed Americans; AFR, Africans; caQTL, chromatin accessibility QTL; EAS, East Asians; EUR, Europeans; mQTL, metabolomics QTL; meQTL, methylation QTL; pQTL, protein QTL; SAS, South Asians.

Multiomics Approach

Many types of QTL analyses have been developed to link genotypes with molecular traits, such as meQTL (DNA methylation), caQTL (chromatin accessibility), and pQTL (protein expression) analyses (Ye et al., 2020). As a consequence, joint analyses that integrate multiple traits have become increasingly prominent. Studies utilizing such multiomics datasets harbor advantages in annotating comprehensive functional features of genetic variants, elucidating the mechanisms underlying traits, and reducing false-positive findings by focusing on the shared signals across multiple trait layers (Fig. 2B). For instance, Zhao et al. (2023) identified genetic variants influencing inflammation-related proteins through integrating eQTL and pQTL signals. With further incorporation of GWAS data, they proposed a pathogenic model in which a multiple sclerosis risk allele leads to increased circulating levels of LTA (trans-pQTL) mediated by reduced levels of LTBR (cis-eQTL). A recent study utilized blood cytokine level and HLA allele information along with single-cell gene expression data to explore and expand our understanding of complex mechanisms regulating cytokine production and disease severity (Lee et al., 2024). In another study, prioritization of key regulatory variants was made feasible through the colocalization of eQTLs, GWAS, and caQTL data, which helped refine the identification of variants and genes with functional effects on GWAS traits. Specifically, the set of putative variants was narrowed from 9,013 eQTLs to 60, which were then validated using MPRA (Broadaway et al., 2024).

Diverse Ancestry and Physiological Contexts

As the context-dependent nature of eQTLs is increasingly emphasized, so too is the importance of studying eQTLs in specific cells or tissues relevant to the model of interest (Fig. 2C). However, currently available samples remain biased toward particular ancestries and tissues; in particular, most large eQTL cohorts comprised whole blood samples originated from individuals of European ancestry. However, recent large-scale eQTL projects have deliberately included individuals from diverse ancestries to address the Eurocentric bias. To overcome this Eurocentric bias, a recent study defined eQTLs using lymphoblastoid cell lines from individuals spanning 5 continental groups and identified population-specific eQTLs not listed in GTEx (Taylor et al., 2024). Similarly, the Asian Immune Diversity Atlas recruited immune cells from 5 countries in Asia and constructed a specifically Asian single-cell reference atlas. They also found sc-eQTLs whose allele frequencies are rare or absent in the European super-population (Kock et al., 2025). In addition, emerging statistical methods for multiancestry eQTL meta-analysis enhance the utility of diverse ancestry datasets by improving both discovery power and fine-mapping resolution (Zeng et al., 2022; Akamatsu et al., 2024). More studies are also being conducted on multiple tissues, such as the lung, brain, and retina, highlighting the role of context-specific eQTLs in regulating tissue-specific mechanisms and diseases (Ratnapriya et al., 2019; de Klein et al., 2023; Natri et al., 2024). As mentioned in “Evolution of eQTL studies,” statistical models have been developed that capture the nature of single cells and reflect context-dependent heterogeneity. Finally, many studies are applying integrative analyses for multiomics and multiethnic datasets. For multiomics datasets, it is common to colocalize signals from diverse molecular measurements associated with traits and link genetic variants to molecular traits through Mendelian randomization (Wu et al., 2018, Soliai et al., 2021, Storm et al., 2021, Assum et al., 2022). For multiancestry eQTL analyses, meta-analysis is employed to maximize power and improve fine-mapping resolution (Zeng et al., 2022).

Closing Remarks

Over the past 2 decades, eQTL studies have become a cornerstone of functional genomics, offering valuable insights into how genetic variations influence gene expression and contribute to complex traits. From the foundational works using model organisms to large-scale projects like GTEx and eQTLGen, developments in the field have progressively deepened our understanding of tissue-specific and context-specific regulatory mechanisms. Advances in scRNA-seq and statistical modeling have further enhanced the resolution and utility of eQTL studies, enabling the discovery of dynamic and cell-state-dependent regulatory variants. These breakthroughs have laid a strong foundation for valuable applications, such as interpreting GWAS loci and therapeutic targets and elucidating disease mechanisms.

Despite these advances, challenges persist in linking eQTLs to disease phenotypes, fine-mapping causal variants, and validating their biological effects. Emerging tools like CRISPR-based screening, MPRA, and organoid models, combined with efforts to diversify ancestry and tissue contexts, are addressing these gaps and enhancing the translational impact of eQTL research. As computational and experimental methods continue to evolve, eQTL studies will remain vital for decoding gene regulation, advancing precision medicine, and improving insights into human biology.

Author Contributions

Sung Eun Hong: Writing – review & editing, Writing – original draft, Visualization. Murim Choi: Writing – review & editing. Jeongha Lee: Writing – review & editing, Writing – original draft, Visualization.

Declaration of Competing Interests

The authors declare that they have no competing interests.

Glossary

Effect size

estimated magnitude and direction of the association between a SNP genotype and the expression level of a gene.

Summary z statistics

standardized test statistics, dividing effect size by standard error.

Linear additive effect

assumption that the effect of each additional copy of the minor allele adds a constant amount to the gene expression level.

Compounding effect

Systematic and nongenetic factors that influence gene expression and can create spurious associations between genotype and expression if not accounted for. There are known confounding effects, including population structure, age, sex, and batch effects, hidden confounding effects that can be accounted with PEER factor, and SVA components.

Linear mixed model

extension of linear model with random effects that account for correlation among samples.

Poisson mixed model

a type of generalized linear mixed model that assumes the gene expression follows a Poisson distribution. It is usually relevant for single-cell eQTL.

Gaussian process latent-variable model

a probabilistic and nonlinear dimensionality reduction technique that infers hidden factors. It is more flexible than linear methods such as PCA.

Posterior probability

the probability that the certain configuration of SNPs is truly causal for the gene expression, given the data and the model.

Posterior inclusion probability

the probability that a particular SNP is causal in any configuration, which is calculated by the sum of posterior probabilities of all configurations that include the SNP.

Bayesian model

a statistical framework that combines prior belief with the observed data (summary statistics) to compute the posterior probability that each SNP is causal.

Credible set

the set of SNPs such that the sum of posterior probabilities equals or larger than a confidence level.

Appendix A. Supplemental material

Supplementary material.