Quantitative trait locus (xQTL) approaches identify risk genes and drug targets from human non-coding genomes

Abstract

Advances and reduction of costs in various sequencing technologies allow for a closer look at variations present in the non-coding regions of the human genome. Correlating non-coding variants with large-scale multi-omic data holds the promise not only of a better understanding of likely causal connections between non-coding DNA and expression of traits but also identifying potential disease-modifying medicines. Genome–phenome association studies have created large datasets of DNA variants that are associated with multiple traits or diseases, such as Alzheimer’s disease; yet, the functional consequences of variants, in particular of non-coding variants, remain largely unknown. Recent advances in functional genomics and computational approaches have led to the identification of potential roles of DNA variants, such as various quantitative trait locus (xQTL) techniques. Multi-omic assays and analytic approaches toward xQTL have identified links between genetic loci and human transcriptomic, epigenomic, proteomic and metabolomic data. In this review, we first discuss the recent development of xQTL from multi-omic findings. We then highlight multimodal analysis of xQTL and genetic data for identification of risk genes and drug targets using Alzheimer’s disease as an example. We finally discuss challenges and future research directions (e.g. artificial intelligence) for annotation of non-coding variants in complex diseases.

Introduction

Estimates of the percentage of the human genome that is ultimately translated into protein vary but are generally believed to be within the range of 1–2% of the entire human genome (1). The rest of the human genome is classified into a wide variety of categories, such as regulatory elements, promoters (2), non-coding genes (3), introns (4) and sections relating to chromosomal structure, activity or segregation (5). Most previous research on DNA variants associated with diseases and traits have focused on the exome part of the genome due to cost, speed and the belief that the coding regions hold the majority of functional variation. However, falling costs of next-generation sequencing technologies allows for broader-scope studies of the whole genome and with more participants than was previously feasible. Apart from teasing out disease associations with genes that do have known functions, a non-insignificant portion of disease-associated genetic risk loci that have been identified are located in non-coding sections of the genome, whose functions are poorly understood.

The influence that non-coding DNA variants identified in whole-genome sequencing (WGS) and genome-wide association studies (GWAS) have on gene expression, post-transcriptional modification and regulation of protein activity suggests that study of those regions can be valuable in finding new associations between DNA variants and disease phenotypes (6–8). Recent advances in functional genomics and computational approaches have led to the identification of potential roles of DNA variants, such as various quantitative trait loci (xQTL), for disease risk gene and drug target identification. However, the predisposition to complex diseases [including Alzheimer’s disease (AD)] involves a complex, polygenic and pleiotropic genetic architecture. Systematic investigation of DNA variant effects on xQTL by unique integration of human genome, transcriptome, epigenome, proteome and metabolome are highly valuable for research and healthcare applications.

Characteristics of xQTL

A QTL is a relationship between a genomic locus whose variation is associated with a quantitative phenotype. xQTL refers to a range of techniques that correlate genetic variants with multi-omic data, for the purposes of connecting genomic variants with phenotypic traits. While expression quantitative trait loci (eQTL) is one of the more common xQTL analyses performed, epigenomic, transcriptomic, proteomic, metabolomics and other QTL analyses present the need for analyzing multi-omic data types and drawing correlations to phenotypes. We briefly highlight several xQTLs below, including eQTL, protein expression QTL (pQTL), histone-QTL (hQTL), splicing QTL (sQTL), methylation QTL (meQTL) and metabolite QTL (mQTL). Figure 1 illustrates a diagram of several types of xQTL.

Figure 1.

Diagrammatic illustration of various quantitative trait loci (xQTL). The discussed xQTL include gene expression quantitative trait loci (eQTL), protein expression QTL (pQTL), splicing QTL (sQTL), histone QTL (hQTL), methylation QTL (meQTL) and other types of QTL [such as metabolite QTL (mQTL)]. These xQTL can be identified from various transcriptomic (RNA-seq), epigenomic (DNA methylation-seq and ChiP-seq, ATAC-seq) and other types of multi-omic assays.

Expression quantitative trait loci

eQTL refers to identifying genetic loci that are connected to levels of expression of DNA products (9). eQTL relies on the quality of multiple techniques: DNA sequencing, quantification of mRNA expression (10,11), and the accuracy of data analysis methods (12,13). The cheapening of DNA sequencing technology over time, as well as advances of mRNA-sequencing technologies, have contributed to the feasibility of conducting large-scale eQTL studies with input from thousands of people. eQTL’s potential is not restricted to comparison of traits in populations, but is also highly useful in analyzing transcriptome data from various tissues and cell types (14,15). An example is the Genotype-Tissue Expression (GTEx) project (16,17), which aims to catalog the tissue transcriptome in connection with genomic variants throughout the body. eQTL also has potential in researching transcriptional changes in response to drug treatment (18,19).

Protein expression quantitative trait loci

While quantitation of mRNA expression is very often used to indirectly measure the expression of proteins overall within cells or tissues, eQTL data may not accurately reflect protein levels due to post-transcriptional regulation and degradation (11,20). Unlike pQTL, eQTL may not capture tissue-specific differences, complex pathways or protein interactions (21–23). pQTL has the potential advantage of being better able to link specific proteins to expression of a disease or trait than eQTL alone (24), and thus, is highly advantageous in the process of functional annotation and target identification during drug discovery and development (25). pQTL also is not restricted to cellular analysis but can test a variety of biological samples and tissues, such as blood plasma and cerebrospinal fluid (CSF) (24,26).

Histone-QTL

Transcription of a DNA segment is generally negatively correlated to the amount of histones bound to that segment (27). Histone–DNA affinity is mainly mediated by electrostatic attraction between lysine and arginine on histones and the DNA (28). The activity and binding of histones can be modified through several mechanisms that can add/remove charges on histone proteins or alter their conformation (29). Besides altering histone binding itself, modification of histone proteins can also affect transcription by presenting/hiding epitopes to transcription factors (30,31). Histone modifications are heavily involved in regulation of aging, responses to environmental stressors and diseases such as cancer (32–34). Polymorphisms in histone acetylation and methylation regulating enzymes have been implicated in a broad range of neurodevelopmental disorders (37). Histone binding/modifications are one of the two major branches of epigenetics, the other being DNA methylation (35,36). Because of histone’s influence on DNA transcription, hQTL studies can research interactions between histones and DNA variants in two ways: (i) researching genomic regions and variants associated with variations in histone modifications/binding themselves (37) and (ii) analyzing locations of heavy histone binding/modification as loci themselves, and correlating the impact of histones with eQTL and phenotypic data (38,39).

Splicing QTL

sQTL refers to techniques that correlate genomic variants associated with alternative splicing and that look for causal connections between alternative splicing and trait expression. Alternative splicing of pre-mRNAs to produce different variants of the same protein is extremely common, with 92–94% of all genes in humans being subject to alternative splicing, at least occasionally (40). Patterns of alternative splicing is highly tissue-specific (41) and are important for regulation of cell differentiation and embryonic development (42). Within the brain, splicing is highly regulated, especially during embryonic development (43). Alternative splicing can directly contribute towards disease, such as alternative splicing of tau protein isoforms in AD and other tauopathies (44,45), or by impairment of the spliceosomal complex itself (46,47). In Takata et al. (46), a high percentage of splicing loci was shown to be associated with previously identified genetic loci associated with neurological disorders, and dysregulation of splicing is heavily associated with AD, schizophrenia and autism spectrum disorder in multiple studies (47–50).

DNA methylation QTL

An additional dimension of DNA sequencing is methyl sequencing, or methyl-seq, which identifies the presence of methyl groups on adenine and cytosine bases on DNA. DNA methylation is an important pathway for modifying gene expression, displaying a variety of effects on up/downregulating transcriptional activity and alternative splicing dependent on the region(s) of the genome that are methylated. DNA methylation’s role in human disease has mainly been investigated in relation to cancer (51,52), but meQTL studies have identified methylation loci that are involved in immune system functioning (53) and AD (54). As with hQTL, researchers can take the same two approaches discussed previously and apply them to studies on meQTL (55,56).

Metabolite QTL

Metabolites are chemical compounds that are produced by a cell or organism as it goes through metabolic processes. The metabolome, like other qualities of an organism, can be profiled with a variety of technologies (57). Metabolites, unlike RNAs or proteins, do not have a direct relationship with DNA variants, being the result of the interactions of many cellular processes. However, metabolites can and have been correlated with genomic variants in mQTL studies (58,59). mQTL can also be coupled with investigations into the microbiome (microbiome QTL), in order to analyze relationships between the microbial composition of a tissue, metabolites produced by resident microbes, genomic variations within the host and resulting effects on the phenotype of the host (60–62). Thus, mQTL offers a powerful tool to investigate gene–environment interactions for non-coding variants.

Resources of human genome-sequencing and multi-omic data

High-throughput DNA/RNA sequencing technologies have generated a large amount of genomic data in multiple national/international genome sequencing projects, such as UK Biobank (63,64), TOPMed (65), All of US (66), Biobank Japan (67,68) and the GWAS Catalog (69). All of these databases tend to focus on diverse populations or different diseases, but share the similarities of being data repositories for genomic and health information for large number of participants. A list of these databases can be located in Table 1 of this article.

Table 1.

List of data resources for human genome sequencing and functional genomic studies (including xQTL). Resources of human genome-sequencing and multi-omic data

Name	Description	Website	Ref.
Human genome-sequencing
Alzheimer’s Disease Sequencing Project (ADSP)	Identifying genomic variants that affect the risk of developing AD	https://www.niagads.org/adsp/content/home	(73)
Alzheimer’s Disease Neuroimaging Initiative (ADNI)	Use of brain MRI and PET scans, SNP sequencing, and biospecimens to diagnose and intervene in AD development	https://adni.loni.usc.edu/	(74)
The Encyclopedia of DNA Elements (ENCODE)	Functional annotation of elements in the human genome	https://www.encodeproject.org/	(79)
NIAGADS	Repository of genomic sequencing data to study late-onset AD	https://www.niagads.org/adsp/	(75)
Functional Annotation Of The Mammalian Genome (FANTOM5)	Transcriptomic analysis of various cell types to understand how transcriptional control and gene expression occur	https://fantom.gsc.riken.jp/5/	(83)
UK BioBank	Collecting DNA variants and health information from participants in the UK	https://www.ukbiobank.ac.uk/	(64)
Trans-Omics for Precision Medicine (TOPMed) Program	Collecting genomics heart, lung, blood and sleep disorders	https://topmed.nhlbi.nih.gov/	(65)
GWAS Catalog	To serve as a database for GWAS data in general	https://www.ebi.ac.uk/gwas/	(69)
All of Us	Collecting diverse health data from more than 1 million participants within the US	https://allofus.nih.gov/	(66)
BioBank Japan	Collecting genomic, biospecimen and health data from Japanese participants	https://biobankjp.org/en/	(68)
Multi-omic resources
The Genotype-Tissue Expression (GTEx) project	To collate together gene expression data across the various tissue types and organs present in the human body	https://gtexportal.org/home/	(16)
AD knowledge portal	Mapping of human proteins within the body	https://adknowledgeportal.synapse.org/	(132)
alzGPS	Integrating multi-omics, clinical trials and drug target data for AD drug discovery/repurposing	https://alzgps.lerner.ccf.org/	(77)
The Alzheimer’s Cell Atlas (TACA)	Correlation of AD pathobiology with single-cell RNA-Seq from AD patients and model mice	https://taca.lerner.ccf.org/	(78)

For example, UK Biobank, which follows approximately 500 000 residents of the UK, has generated the whole-genome sequences of 150 119 individuals and identified more than 600 million sequence variants (70). TOPMed, which is sponsored by NIH, focuses on multi-omic studies on heart, lung, blood and sleep disorders, and has sequenced more than 53 000 whole genomes with 410 million sequence variants identified (65). All of US, also operated by NIH, focuses on collecting electronic health records, surveys and biospecimens to reflect the diverse population of the US, and has gathered data from 370 000 participants (accessed July 24, 2022) (71, 72). BioBank Japan collected medical records and tissue and DNA samples from approximately 200 000 participants in Japan enrolled from 2003 to 2008 (68). The GWAS Catalog, run cooperatively by the National Human Genome Research Institute (NHGRI) and the European Bioinformatics Institute (EBI), does not run data collection by themselves, but does collect and organize GWAS data from a variety of other sources (69).

For AD, there are several specific genome-sequencing and multi-omic resources, including the Alzheimer’s Disease Sequencing Project (ADSP) (73), the Alzheimer’s Disease Neuroimaging Initiative (ADNI) (74) and the Genetics of Alzheimer’s Disease Data Storage Site (NIAGADS) (75). NIAGADS is a national genetic and genomic data repository for AD research that to date has collected 90 datasets and 144 242 samples (accessed July 19, 2022). The ADSP, launched in 2012, has sequenced and analyzed genomes to identify a wide range of both risk and protective genetic variants in AD, including whole-exome sequencing (WES) data from nearly 20 000 samples, as well as WGS data from nearly 17 000 samples (73). The AD Knowledge Portal (76) contains upwards of 100 000 data files from over 80 studies of people with AD and related model organisms. This resource provides rich xQTL, genomic, epigenomic, transcriptomic, proteomic and metabolomic data from thousands of human brains. A few projects play an outsized role in being sources for xQTL data involving the brain or cognitive functions: the ROSMAP project, the Mount Sinai Brain Bank (MSBB) and the Mayo RNA-Seq project (77–79). The ROSMAP study, which is the combined data from the Religious Orders Study and the Memory and Aging Project, recruits religious members (ROS) and people from the Greater Chicago Area (MAP), collect data from cognitive tests and correlate them to genomic variants and other –omics data. MSBB combines results from cognitive tests and correlates them to Braak-Staged brains and –omics data from participants. Mayo RNA-Seq focuses mainly on transcriptomic data from banked brains with AD and other neurodegenerative disorders. Several recent multi-omic databases, such as alzGPS (77) and The Alzheimer’s Cell Atlas (78) have leveraged large-scale multi-omic data available for target identification.

Several commonly used functional genomic data resources include ENCODE (79,80), GTEx (16,81) and FANTOM5 (82,83). ENCODE aims to functionally annotate the entire human and mouse genomes, and thus it is focused on data analysis of other studies, with over 5000 datasets (79). GTEx specifically aims to create an atlas of expression data based on tissue. In its most recent official data release (GTEx Analysis V8), the GTEx project has obtained 17 382 total samples and 15 201 eQTL analyzed samples, from 948 and 838 donors, respectively. FANTOM5, which is a project of the FANTOM consortium, collects transcriptome, lncRNA and miRNA data from a wide variety of human and mouse tissues and cell lines to study transcriptional regulation (84), with over 1000 human and mouse samples.

xQTL-based annotation of non-coding variants in AD

While non-coding DNA, by its very definition, is not transcribed into proteins, non-coding DNA serves a variety of other functions essential to functioning and regulation of gene expression (85,86). The functions of these non-coding genes include: the transcription of regulatory RNAs from non-coding genes (87–89), promoters and regulatory elements that control gene transcription (90,91), introns and structural functions, such as centromere activity (5,92). With this in mind, the effects of non-coding DNA variants fall generally under either affecting levels of gene expression, affecting post-transcriptional modification of mRNA transcripts or changing the sequence of non-coding RNAs (6,7,93). For analyzing these effects, eQTL, pQTL and splice-QTL are of great value.

AD is one of the most common forms of dementia diagnosed in the elderly (94), characterized by amyloid-beta plaques and phosphorylated tau tangles. AD can generally be classified into two types, based on disease etiology: Familial Alzheimer’s Disease (FAD), which is characterized by mutations in key AD-related genes (PSEN1, PSEN2 and APP) and an early age of onset (95) and Late-Onset Alzheimer’s Disease (LOAD), which is dependent on complex interactions between genetic factors, health conditions and lifestyle factors. While many WGS/WES association studies have been conducted on patients with AD and other cognitive impairments, fewer genes have been conclusively linked to an increased risk of developing LOAD, with the notable exception of the APOE4 variant of APOE (96). Variants in proximity to APOE, which is itself a well-known AD risk gene, have been found in Blue et al. (97) and Zhou et al. (98), from data obtained from two different groups of participants. Other non-coding variants have found variants in proximity to AD risk genes observed in other studies (99), such as BIN1 (100) and CD33 (101).

In Raj et al. (48), which studied alterations in splicing in the prefrontal cortex of 450 subjects, used LeafCutter software to estimate excised introns in mRNAs (102). In Raj et al., the RNA-Seq results were also compared against data from MSBB and iPSC lines in a replication analysis. The data were compared with genomic data from the participants to analyze associations between sQTLs and AD-related genes. Yang et al. (24) performed pQTL studies on CSF, brain tissue and blood from participants with or without AD. The researchers used an aptamer platform, and discovered that cis-QTLs (within close proximity of a protein-coding gene), tend to affect all sample tissues from a subject, while trans-QTLs tend to be more tissue-specific. The researchers also replicated their findings against multiple pre-existing datasets. Zhao et al. (103) conducted a meta-analysis on five GWAS datasets, three eQTL datasets and three mQTL datasets, obtained mostly from brain tissue, to analyze AD-related genes. The researchers used Summary data-based Mendelian randomization to compare the different datasets, and found significant concordance between all three types of datasets in terms of AD-related genes identified. In Wang et al. (104), the researchers focused on DNA methylation in peripheral blood in the context of MCI and AD, and found a reduction in promoter methylation in the PM20D1 locus in MCI and AD patients. Similarly, Hillary et al. (105) also focused on DNA methylation, as well as pQTL, correlating the blood levels of 18 AD-associated proteins to variations in 19 CpG sites, and pQTL associated with increased expression of TREM2. Ng et al. (106) take a broader approach, eQTL, hQTL and meQTL techniques to analyze the dorsolateral prefrontal cortex of 411 participants to map the epigenome and transcriptome, and identify new susceptibility genes for schizophrenia and bipolar disorder. Finally, Novikova et al. (107) studied myeloid cell and microglial enhancers with hQTL. In all of these studies, certain commonalities regarding the cell types or genes affected in AD or MCI can be noted, regardless of the tissue type or xQTL studied in the particular article.

There are several potential limitations that can occur with xQTL and GWAS studies. Variants may be present at very low allele frequencies within the population or may be de novo mutations present only in the patient or their family, which might mean that statistical analyses might not have enough power to make a meaningful correlation (108,109). Synergistic effects among different variants is also a factor that needs to be considered during analysis (110,111). SNP/variant frequencies can differ markedly between demographic groups being analyzed (112,113), which can create inaccuracies when applying GWAS results to populations from ancestries other than the population sampled in the study (113–115). An additional challenge when analyzing and annotating variants is linkage disequilibrium due to co-inheritance of sections of the genome during homologous recombination. As a result, techniques such as massively parallel reporter assays are often need to be used to discern causal variants (116).

Discussion and Future Directions

Investigating the polygenic influences on the majority of human illnesses and traits requires the accurate collection and analysis of increasingly larger sets of data generated from patients. While the advances in xQTL technologies have been beneficial in shedding light on biological pathways, sifting through the massive amounts of data to identify meaningful results requires taking care to avoid bias in participant selection, as well as the aid of a variety of computational algorithms. xQTL analysis is a powerful tool for discovering biological pathways, annotation of non-coding variants and potential therapeutic treatments. The main advantage of xQTL studies is the ability to collect and correlate multi-omics together in order to identify likely causal risk genes and drug targets. While examining genomic, transcriptomic or other data alone in comparison with patient data is valuable, integrative analysis allows for information to be collected about the effects of genomic and other variation not just on the expressed phenotype, but also on biological pathways in-between (117,118), which aids significantly in functional annotation. Many of the computational techniques behind integrative analysis today rely upon advances in machine learning. The ability of machine learning to make predictions (supervised learning), and find correlations (unsupervised learning), can exceed that of traditional statistical analysis if correctly performed (119).

Most xQTL studies concentrate on correlating sequence variations with trait outcomes. Recent advances in genomic editing technologies, such as Cre-LoxP (120) and CRISPR-Cas9 (121), allow for testing the effects of genomic variations in an experimental setting. Based on xQTL data, a researcher can choose to replicate candidate variants in a model animal, tissue chips or in primary/iPSC cells. An additional approach is to use CRISPR interference (CRISPRi) (122,123), or CRISPR activation (CRISPRa) (124), to induce sequence-specific repression or expression of particular sections of the genome to better our understanding of those gene’s functions (125). CRISPRi and CRISPRa are also ideal for examining how gene expression can affect the development of organs and tissues themselves (126).

At the root of this so-called ‘precision medicine’ is the idea of tailoring treatment to a patient’s individual biology and variants. Recent studies have showed the elevated successful rates of drug target discovery and drug development from genetic and genomic discoveries (127). Artificial intelligence (AI) and network-based methodologies offer a powerful framework to identify risk genes and drug targets from genetic, genomic and xQTL discoveries (128–131). We believe that a genome-wide, integrative AI framework (Fig. 2) to identify novel risk genes and networks from massive genetic and xQTL data will enable a more complete mechanistic understanding of the disease and the development of targeted therapeutic approaches for human complex diseases if broadly applied.

Figure 2.

Diagram illustrating a genome-wide, integrative AI framework to identify risk genes and drug targets from genetic, genomic and xQTL (pQTL, eQTL, sQTL, hQTL and meQTL) findings.

Conflict of Interest statement. The authors have declared no competing interests.

Funding

National Institute of Aging (NIA) of the National Institutes of Health (NIH) (under Award Number U01AG073323, R01AG076448, R01AG066707 and R56AG074001 to F.C.); Sondra J. and Stephen R. Hardis Endowed Chair of Cancer Genomic Medicine at the Cleveland Clinic (to C.E.).

Authors’ contributions

M.B., Y.H., F.C. and C.E. wrote the draft and critically revised the manuscript. All authors critically revised the manuscript and gave final approval.