Emerging applications of genome-editing technology to examine functionality of GWAS-associated variants for complex traits

Abstract

Over the last decade, genome-wide association studies (GWAS) have propelled the discovery of thousands of loci associated with complex diseases. The focus is now turning toward the function of these association signals, determining the causal variant(s) among those in strong linkage disequilibrium, and identifying their underlying mechanisms, such as long-range gene regulation. Genome-editing techniques utilizing zinc-finger nucleases (ZFN), transcription activator-like effector nucleases (TALENs), and clustered regularly-interspaced short palindromic repeats with Cas9 nuclease (CRISPR-Cas9) are becoming the tools of choice to establish functionality for these variants, due to the ability to assess effects of single variants in vivo. This review will discuss examples of how these technologies have begun to aid functional analysis of GWAS loci for complex traits such as cardiovascular disease, Type 2 diabetes, cancer, obesity, and autoimmune disease. We focus on analysis of variants occurring within noncoding genomic regions, as these comprise the majority of GWAS variants, providing the greatest challenges to determining functionality, and compare editing strategies that provide different levels of evidence for variant functionality. The review describes molecular insights into some of these potentially causal variants and how these may relate to the pathology of the trait and look toward future directions for these technologies in post-GWAS analysis, such as base-editing.

INTRODUCTION

Genome-wide association studies (GWAS) have transformed our ability to detect genetic loci for many complex diseases and traits. Fine-mapping arrays (9a, 53), exome-sequencing, and whole-genome sequencing (24, 71a) have helped to define regions of genetic association and identify 99% credible sets of variants (75). Despite the success of these methodologies to detect associations in large, well-phenotyped cohorts, the identity of the causal variant(s) often remains hidden (61). The primary cause for this can be attributed to the strong linkage disequilibrium (LD) present in much of the genome, where the sentinel variant providing the strongest association signal at a locus e.g., a single nucleotide polymorphism (SNP), may be tagging (r² > 0.8) hundreds of other co-inherited variants. An additional level of complexity for determining the functionality of trait-associated variants occurs since the majority of associated SNPs from GWAS occur within noncoding regions of the genome (61): thus an even bigger challenge compared with coding variants to predict function using in silico tools, as well as to verify experimentally. Occasionally signals may appear close to strong candidate genes for the trait, but often, there are no genes that have shown a prior link to the trait, or they are located within gene deserts without obvious mechanisms for functionality.

To steer GWAS variant discoveries toward functionality in order to identify new mechanisms of disease, several complementary public resources are available. Global efforts to annotate the coding and noncoding functional elements of the genome, such as the Encyclopedia of DNA Elements (ENCODE) Project (21) and Roadmap Epigenomics (3), have uncovered a wealth of data to enable researchers to speculate the potential role of variants, such as their location within transcription factor binding sites, histone modifications, or DNA methylation sites for a number of transformed and primary cell types. Summary data from these resources can be readily accessed from a number of databases, such as HaploReg v4.1 (74) and RegulomeDB (7). The predicted effects of variants on transcription factor binding sites can be examined through databases such as TRANSFAC (50) and JASPAR (49), although such predictions in transcription factor binding affinity do not accurately predict functional outcomes as with coding variants, and laboratory verification with the suitable cell type and conditions is required. Analysis of expression quantitative trait loci (eQTL) can provide valuable data regarding regulation of gene targets possibly affected by GWAS variants, and resources such as the Genotype-Tissue Expression (GTEx) project are facilitating these efforts by expanding the tissues and numbers of samples available for such analyses (30a). Tools such as Combined Annotation Dependent Depletion (CADD) attempt to predict pathogenicity by highly scoring variants that are not stabilized by evolutionary selection (40).

Studies into allele-specific effects on chromatin accessibility can aid the localization of potentially functional variants (17, 62), and examination of the three-dimensional organization of chromatin, based on chromosome conformation capture (3C) techniques can give clues toward the target genes for variants located within distal enhancers (60). Other emerging techniques include the massively parallel reporter assay (MPRA), which can screen thousands of potential functional variants in a single assay to determine effects on gene expression (51).

Despite these complementary tools, there are very few methods available to reliably test functionality of single variants in vivo, due to either the nature of LD in the examined genomic region or the necessity to perform the experiments outside of the context of the cell’s chromatin organization. New genome-editing techniques, however, are beginning to revolutionize the methods of studying functionality of variants from GWAS. This review will discuss the current progress and future directions for these technologies.

GENOME-EDITING TECHNOLOGIES

Until recently, efforts to modify genetic material of an organism or cell have been hindered by a lack of specificity and inefficiency, relying on site-directed mutagenesis or recombination-based methods (48, 65). Since the development of engineered nucleases (25), genome-editing has fast become an efficient and highly targeted approach to modify the genetic architecture of a cell and a promising tool to investigate functionality of GWAS loci. Three genome-editing techniques have been widely used to date: zinc-finger nucleases (ZFNs) (38), transcription activator-like effector nucleases (TALENs) (75a), and clustered regularly interspaced short palindromic repeats (CRISPR) with Cas9 nuclease (CRISPR/Cas9) (33, 47). These methods rely on the ability to create synthetic vectors to target specific oligonucleotide sequences and to induce a targeted double-strand break in the genome (Fig. 1). These breaks can then be repaired by the cell’s imperfect repair mechanisms: nonhomologous end joining (NHEJ) or by homology-directed repair (HDR), to create alterations in the genome at the site of the break. Several comprehensive articles are available detailing genome-editing technologies (12, 20, 33, 36, 57, 59, 72, 76): this review will only briefly outline the principles of the three major techniques, followed by an examination of where these have been, or have potential to be applied in the post-GWAS era to characterize causal variants or pathways.

Fig. 1.

Genome-editing technologies employing different methods to create a targeted double-strand break. A: zinc-finger nucleases (ZFNs) target the genome with zinc-finger modules that each identify nucleotide triplets. This series of targeting modules is attached to a FokI nuclease, which when dimerized can form a double-strand break in the DNA. B: transcription activator-like effector nucleases (TALENs) target the genome with transcription activator-like effectors (TALEs), which target single nucleotides, as with ZFNs, these are attached to FokI to form the double-strand break. C: clustered regularly interspaced short palindromic repeats with Cas9 nuclease (CRISPR/Cas9) uses a guide RNA to base-pair with the target DNA, adjacent to the protospacer adjacent motif (PAM), and to recruit Cas9 nuclease to create the double-strand break.

ZFNs

The earliest targeted genome-editing tools to realistically allow for follow-up of GWAS targets were the ZFNs, pioneered by Kim et al. (39), who created a chimeric restriction endonuclease that demonstrated in vitro activity. The zinc-finger domain comprises ~30 amino acids in a ββα configuration, where the α-helix can contact 3 bp of DNA in a sequence-specific manner (56). Zinc-finger modules have now been designed that identify almost all 64 potential nucleotide triplets; by engineering synthetic zinc-finger domains between a conserved linker sequence, zinc-finger proteins can be created to distinguish DNA sequences up to 18 bp in length, allowing for sequence specificity within the human genome. Genome-editing relies on the creation of a double-strand break in the target DNA, and this can be achieved by engineering a FokI cleavage domain to the zinc-finger modules. The FokI cleavage domain must dimerize to cleave DNA (4), and this requirement of two targeted ZFNs for a single genomic locus results in strong sequence specificity and low likelihood for off-target effects (Fig. 1A).

Despite the remarkable advances made by ZNFs to genome-editing, the technique does have drawbacks, primarily the difficulty in constructing the zinc-finger domain.

TALENs

Following from the success of ZFNs to target and cleave specific genomic regions, TALEN technology was developed to facilitate construction of targeting nucleases. This technique uses transcription activator-like effectors (TALEs): proteins originating from Xanthomonas spp. proteobacteria that can alter gene transcription in host cells (5). An individual TALE repeat recognizes and binds to a single nucleotide, where the hypervariable residues located at positions 12 and 13 of the TALE are key to specificity, and bind in the major groove of the DNA. These hypervariable residues can either be NN or NK (recognizing guanine), NI (adenine), HD (cytosine), or NG (thymine), and the construction of TALE repeats can be engineered to recognize almost any genomic sequence (19). The only restriction for TALE repeats is the necessity for a 5′-thymine preceding the first base of the TALE sequence. As with ZFNs, engineering a nuclease such as FokI results in a synthetic protein capable of targeted genomic cleavage (Fig. 1B).

Although TALENs provide a similar efficiency to ZNFs and also have a low chance of off-target effects due to their requirement for dimerization of FokI constructs to enable cleavage, they are significantly easier to assemble, making this a much more viable option for targeted genome-editing. One drawback is their sensitivity to cytosine methylation, requiring the methylation state of the target site to be known before design.

CRISPR-CaS9

More recently, Cas9, a nuclease from Streptococcus pyogenes, has been exploited to facilitate the targeting of genomic sequences: a protein that relies on association with RNA to target the nuclease toward the DNA target (35). The CRISPR system has evolved as a protection mechanism for bacteria against foreign nucleic acids such as plasmids and viruses, where these sequences can be incorporated between CRISPR repeat sequences encoded within the host genome. When these sequences are processed into CRISPR RNA (crRNA), hybridization with a transactivating CRISPR RNA (tracrRNA) is facilitated, forming a complex with Cas9 nuclease. Cas9 is directed toward the DNA sequence to be cleaved by the crRNA, where it requires a protospacer adjacent motif (PAM) to be present (5′-NGG). This CRISPR system has been adapted for use in genome-editing, whereby a guide RNA (gRNA), formed from a fusion between tracrRNA and crRNA, and Cas9 are introduced in a cell or organism. Cas9 is guided toward a specific site by RNA-DNA base pairing from the sequence of 20 nucleotides at the 5′-end of the gRNA. It has been shown that double-strand breaks created by Cas9 are able to result in indels from NHEJ and HDR from either single-stranded oligonucleotides or double-stranded plasmid DNA (Fig. 1C).

The use of CRISPR-Cas9 for genome-editing has many advantages over ZNFs and TALENs for the majority of situations, primarily the ease of creating targeting plasmids based directly on DNA sequence and the ability to multiplex. Unlike TALENs, CRISPR-Cas9 is not methylation sensitive. The key disadvantage with this system, however, occurs with the increased likelihood for off-target effects. There are several ways to reduce this possibility, such as dimeric CRISPR RNA-guided FokI nuclease (RFN) technology (70). RFN relies on the expression of two gRNAs targeting opposing strands flanking the target site and FokI-Cas9 fusion proteins, resulting in similar specificity to ZFNs and TALENs. Use of paired single-strand “nickase” mutants has a similar effect (46). The CRISPR system also relies on Cas9 to recognize the PAM: in situations where simply knocking out a gene is required, this is not normally a concern. When attempting to alter a particular nucleotide with HDR or using RFNs or nicking Cas9, this can provide difficulties, particularly with the spacing constraints required for RFNs and nickases. Different PAM sequences are available, such as an engineered Cas9 from Francisella novicida recognizing 5′-YG-3′ (32), or the Cpf1 nuclease from this bacteria that recognizes 5′-TTN-3′ (81) or 5′-YTN-3′ (23).

CHARACTERIZATION OF LOCI IDENTIFIED BY GWAS

There are multiple ways genome-editing is set to transform the functional analysis of GWAS loci, and we will discuss some early examples of the two main approaches in relation to noncoding variants: that of examining the effects of the variant on downstream pathways and that of identifying the causal variant, and these are summarized in Table 1. There is opportunity to use genome-editing to study the majority of traits examined by GWAS, and here we will focus on a subset of these.

Table 1.

Use of genome-editing techniques to examine functionality of noncoding variants

Genome-editing Tool	Trait	Locus	Lead SNPs	Investigated SNP	Functionality Demonstrated through Genome-editing	Reference
Zinc-finger nuclease	breast cancer	FGFR2	rs2981578 rs1219648 rs2981582 rs2981579	rs2981578	effects on cell proliferation, cell cycle progression, and transcription factor binding examined, showing no allele-specific effects	(43)
TALEN	prostate cancer	RFX6	rs339331	rs339331	prostate cancer cell lines homozygous for risk alleles demonstrated increased RFX6 gene expression and increased HOXB13 transcription factor occupancy	(44)
CRISPR-Cas9	Type 2 diabetes	TCF7L2	rs7903146	rs7903146	allele-specific effects demonstrated in a human colorectal carcinoma cell line on ACSL5 gene expression and long-range chromatin contacts	(31)
	Type 2 diabetes	PPARG2	n/a	rs4684847	preadipocytes with nonrisk allele introduced showed increased PPARG2 transcript levels, mediated by PRRX1 transcription factor	(32)
	lipids	CPNE1	rs2277862	rs2277862	allele-specific effects on gene CPNE1 gene expression observed upon differentiation of hPSCs into adipocytes	(39)
	coronary artery disease	PHACTR1	rs9349379	rs9349379	deletion surrounding rs9349379 resulted in decrease of PHCTR1 gene expression	(40)
	obesity	IRX3/IRX5	rs1558902	rs1421085	allele-specific gene expression observed upon adipocyte differentiation in presence of the repressor, ARID5B	(41)
	renal cancer	MYC/PVT1	rs35252396	rs35252396	mutations introduced to an HIF-binding site surrounding rs35252396 led to reduction in MYC and PVT1 gene expression	(46)
	Parkinson's disease	SNCA	rs356220 rs356219 rs356182	rs356168	iPSCs differentiated into neuronal precursors showed higher levels of SNCA gene expression for carriers of the risk allele	(47)
	ankylosing spondylitis	PTGER4	n/a	rs9283753	using lymphoblast cell lines, allele-specific effects on PTGER4 gene expression were observed	(48)

CRISPR-Cas9, clustered regularly interspaced short palindromic repeats with Cas9 nuclease; iPSC, induced pluripotent stem cell; SNP, single-nucleotide polymorphism; TALEN, transcription activator-like effector nucleases.

TYPE 2 DIABETES

A study by Claussnizter et al. (10) sought to examine the functionality of cis-regulatory variants causing predispositions to Type 2 diabetes (T2D) with integrative computational analysis of phylogenetic conservation. They examined enrichment for transcription factor binding within conserved transcription factor binding site modules at GWAS loci, and their analysis identified clustering of distinct homeobox transcription factor binding sites, specifically, identifying the paired-related homeobox 1 (PRRX1) factor as a repressor of peroxisome proliferator activated receptor gamma gene (PPARG2) expression in adipose cells. The authors demonstrated that PRRX1 showed adverse effect on lipid metabolism and systemic insulin sensitivity, according to rs4684847 genotype, where the risk allele triggers PRRX1 binding. The authors employed CRISPR-Cas9 homology-directed repair genome editing to introduce the rs4684847 nonrisk allele into human Simpson-Golabi-Behmel syndrome preadipocytes. Introduction of this allele led to a 5.4-fold increase in PPARG2 transcript levels, which was in accordance with experiments performing PRRX1 knockdown in human adipose stromal cells, where the risk allele-driven suppression of PPARG2 expression was reversed by silencing PRRX1. PRRX1 silencing did not, however, affect expression of PPARG2 in cells harboring the nonrisk allele.

The TCF7L2 locus has been associated with T2D with a relative-risk of 1.36 per allele: the strongest effect size observed for T2D for a common variant (30). An intronic variant in TCF7L2, rs7903146 t > C, has long been considered to be functional not only due to its consistent appearance as the lead SNP at the locus (53), but also from allele-specific effects on reporter gene expression and the presence of open chromatin at this locus in pancreatic islets (27). Other tissues implicated in the role of TCF7L2 in T2D include the liver (6, 37), adipose tissue (37), and intestinal cells (79). To ascertain a potential effect of this variant on gene regulation in intestinal cells, Xia et al. (77) used CRISPR-Cas9 tools, which generated a 1.4 kb deletion surrounding the variant in HCT116 cells, a human colorectal carcinoma cell line, followed by an examination of global gene expression and chromosome conformation capture-based techniques to assess chromatin structure in the mutant vs. wild-type cell line. The authors identified 99 genes that showed differential expression in the mutant line carrying the deletion, compared with the wild type, but only one gene whose promoter formed intrachromosomal contacts in proximity to the variant location in the wild-type cell line: Acyl-CoA synthetase long-chain family member 5 (ACSL5). To further define the regulatory effects surrounding rs7903146, further deletions of 66 and 104 bp surrounding the variant were performed, leading to similar reductions in ACSL5 protein levels. The authors speculated that this revealed a role for rs7903146 in creating a colon-specific enhancer associated with expression of ACSL5, producing a protein with known roles in fatty acid metabolism and involvement in T2D. The authors also note, however, that studies in different tissues need to be performed to demonstrate tissue-specific effects, particularly as chromatin accessibility surrounding this variant has only been detected in pancreatic cells (27). The authors did not examine the effects of rs7903146 alleles on ACSL5 expression or protein levels outside the context of large deletions, and so these initial findings do not characterize SNP functionality at this locus specifically for T2D.

Genome-editing not only provides means to explore functionality of genes implicated in disease through GWAS but also offers the potential to screen potential therapeutics. Using isogenic human embryonic stem cells (hESCs), Zeng et al. (80) created insertion or deletion mutations with CRISPR/Cas9 to create early frame shifts in three genes identified from GWAS for T2D: CDKAL1 (CDK5 regulatory subunit associated protein 1 like 1), KCNQ1 (potassium voltage-gated channel subfamily Q member 1), and KCNJ11 (potassium voltage-gated channel subfamily J member 11). These mutant hESCs were differentiated into functional islet cells that did not affect insulin production, although they showed effects on impaired glucose secretion and glucose homeostasis. Using a high-content chemical screen on these cells, the authors identified a candidate drug, T5224, which was able to rescue the CDKAL1-specific defects by inhibition of the FOS/JUN pathway.

CARDIOVASCULAR TRAITS

A number of continuous traits such as blood pressure and circulating lipid levels are well-established risk factors for cardiovascular disease. The MTHFR-NPPB locus has been associated with blood pressure in a number of GWASs (34a, 73). Since the locus contains many candidate genes for hypertension, a study by Flister et al. (22) employed genome-editing to establish which genes at this locus could be implicated with the trait. Using ZFN tools, the authors introduced deleterious mutations into rodent models of genetic hypertension, resulting in frame shifts or disruption to functional protein domains into each of the six genes spanning this locus: Agtrap (angiotensin II receptor associated protein), Mthfr (methylenetetrahydrofolate reductase), Clcn6 (chloride voltage-gated channel 6), Nppa/Nppb (natriuretic peptides A and B), and Plod1 (procollagen-lysine, 2-oxoglutarate 5-dioxygenase 1). The study revealed that five of the six genes could affect hypertension by modifying blood pressure or renal phenotypes, with mutations in Nppa, Plod1, and Mthfr increasing hypertension risk and mutations in Agtrap and Clcn6 decreasing disease susceptibility.

The transcription factor nuclear receptor 2 family 2 gene (NR2F2) has been implicated in essential hypertension through GWAS (8). To demonstrate this, Kumarasamy et al. (44) created a ZFN-based 15 bp deletion in the Nr2f2 gene, leading to a deletion within the hinge region of the protein. The authors demonstrated that the mutant rats had significantly lower blood pressure than control rats (systolic BP 179 ± 3 vs. 197 ± 5 mmHg), due to the interaction of the hinge region of Nr2f2 and zinc finger protein, FOG family member 2 (Fog2), which was increased following the mutation, and that the interaction of these two transcription factors was an important regulator of blood pressure. Such studies where specific deletions are created within proteins cannot only give answers to which pathways may be involved from GWAS, but also specific molecular mechanisms leading to the trait.

The use of large-scale GWAS has confirmed and revealed new loci for lipid traits (67, 68, 75a). A study by Pashos et al. (55) employed multiple approaches involving induced pluripotent stem cells (iPSCs) and hepatocyte-like cells to isolate functional variants, including genome-wide mapping of eQTLs, allele-specific expression, and an MPRA. Following characterization of likely functional variants, the authors focused on several loci with genome-editing. First, they used CRISPR/Cas9 in a human pluripotent stem cell (hPSC) line to knock in one minor allele at the CPNE1 locus into the rs2277862 CC homozygous cells, using single-strand DNA oligonucleotide as a template. This, however, led to low efficiency, resulting in only a single clone with the heterozygous allele, but did demonstrate decreased CPNE1 expression in both knock-in undifferentiated hPSCs and differentiated hepatocyte-like cells. Second, the authors deleted ~38 bp surrounding the SNP, allowing for higher efficiency. This demonstrated 31% decreased expression of CPNE1 and 20% decrease in ERIGC3 expression. The authors also used CRISPR/Cas9 to generate a knock-in mouse with a minor allele in the homozygous C57BL/6J CC homozygous strain. Again, a 37% decrease of Cpne1 was observed in the liver of TT mice compared with CC wild-type mice. In the same study, the authors also examined the ANGPTL3/DOCK7 locus, associated with a 4.9 mg/dl change in TG levels (68). Using single-strand template was not successful for HDR using CRISPR/Cas9, and the authors used a targeting vector with 500 bp homology arms on a transposon that was able to undergo scarless removal from a TTA site with piggyBac. Puromycin selection allowed for knock-in of rs10889356 minor alleles to both chromosomes resulting in decreased DOCK7 expression (36%) and increased ANGPTL3 expression (60%). Similar findings were seen when the authors multiplexed CRISPR/Cas9 to create a 36–39 bp deletion surrounding the variant.

A consistent finding in GWASs for coronary artery disease (CAD) is seen with variants at the phosphatase and actin regulator 1 gene locus (PHACTR1) (9a). To examine functionality of this locus, Beaudoin et al. (2) performed fine-mapping and DNA resequencing to prioritize rs9349379 as a causal variant, also determining the variant as an eQTL for PHACTR1 expression in the coronary artery. Using an endothelial cell extract, the authors demonstrated that rs9349379 differentially bound the transcription factor myocyte enhancer factor-2 (MEF2). Using siRNA to knockdown MEF2 in human umbilical vein endothelial cells was unsuccessful, and so the authors used CRISPR/Cas9 to introduce a heterozygous deletion surrounding rs9349379 and the MEF2-binding site in hESCs. Following differentiation into endothelial cells, a decrease of 35% PHACTR1 expression was observed validating the potentially causal role of rs9349379 in CAD.

OBESITY

The fat mass and obesity associated gene (FTO) locus has shown the strongest association with obesity in GWAS, although it was established that the variants in the FTO locus might be acting through long-range interactions with the IRX3 (iroquois homeobox 3) locus (60). A recent study performed phylogenetic module complexity analysis at the locus to determine the most likely functional variant, with rs1421085 achieving the highest score, and in perfect LD with the lead GWAS SNP, rs1558902 (11). Reporter assays and electrophoretic mobility shift assays indicated the loss of a repressor protein, AT-rich interaction domain 5B (ARID5B), binding to the risk allele in adipocytes, in line with effects that the authors observed on IRX3 and IRX5 gene expression. As with similar examples outlined previously, CRISPR/Cas9 genome-editing was performed in primary preadipocytes to create cells that differed only by rs1421085 genotype. It was found that conversion from the risk allele to the nonrisk allele restored low expression levels of IRX3 and IRX5 in the presence of ARID5B. Further examination of the differences between edited and nonedited preadipocytes during differentiation into white and beige adipocytes showed that peak expression occurred in days 0–2 of differentiation in the unedited risk alleles compared with those edited to wild type, which maintained low levels of expression throughout differentiation, indicating that rs1421085 plays a causal role in developmental gene expression.

CANCERS

Variants in the fibroblast growth factor receptor 2 (FGFR2) have shown a strong association with development of breast cancer (34). To examine one of the putative functional variants at the locus, an estrogen receptor alpha-positive breast cancer cell line, MCF7, was edited with ZFN technology to modify the lead GWAS variant: rs2981578, from wild type to heterozygous status (58). The resulting mutant cell line showed no effects on cell proliferation, cell cycle progression, or indeed, binding of the transcription factor runt-related transcription factor 2 (RUNX2), as previously indicated. In this case genome-editing was used to refute the role of a putative functional SNP as a single causal variant at the locus.

Spisak et al. (64) examined a GWAS locus on 6q22.1 for pancreatic cancer, which also acts as an eQTL for regulatory factor X6 (RFX6) gene expression. Following approaches to reduce the number of candidate causal variants that included fine-mapping and analyses of chromatin annotations in the prostate cancer cell line LNCaP, the authors confirmed a previous finding that the risk allele of rs339331 was able to create a binding site for the prostate lineage-specific homeobox B13 (HOXB13) transcription factor. In an effort to characterize the locus further, genome-editing using TALE-based tools were performed. Initially, the authors used a fusion protein, where the TALE element was fused to LSD1, a histone lysine-specific demethylase, previously shown to remove H34K methylation marks at the site of DNA-targeting (52), thereby decreasing their regulatory activity. By targeting the HOXB13 site at rs339331, the authors observed a threefold decrease in RFX6 expression levels. A similar experiment was performed replacing LSD1 with VP64, a transcriptional activator, where a twofold increase in RFX6 expression was observed, confirming the locus as a key regulator of RFX6 expression. To determine whether rs339331 was responsible for this effect, TALEN-mediated HDR was employed to create isogenic 22Rv1 prostate cancer cell lines containing each of the three rs339331 genotypes. The cells homozygous for the protective alleles showed decreased RFX6 expression and those homozygous for the risk alleles showed increased expression compared with the heterozygous cell lines. In concordance with these findings, the authors demonstrated increased HOXB13 occupancy in the risk allele carriers to account for the effects on gene expression. When performing a global analysis of gene expression with RNA-Seq in the edited cell lines, the authors observed enrichment in genes affected by androgenic compounds and androgen receptor.

Renal cancer susceptibility has been examined with GWAS, where a study identified a single intergenic variant, rs35252396, in a region located between the oncogenes v-myc avian myelocytomatosis viral oncogene homolog (MYC) and pvt1 oncogene (PVT1) (31). A study by Grampp et al. (29) examined ENCODE annotations for chromatin accessibility and chromatin immunoprecipitation (ChIP)-seq data for hypoxia inducible factor (HIF), a transcription factor implicated in MYC and PVT1 regulation. The authors identified a HIF-binding signal that coincided with the renal cancer susceptibility SNP rs35252396 and showed by chromosome conformation capture techniques that this formed physical connections to both the MYC and PVT1 promoters. To confirm whether the HIF binding site surrounding rs35252396 or another HIF site was responsible for the effects on regulation, the authors targeted the HIF-response element in 786-O renal cancer cells with CRISPR/Cas9 technology. Using seven clones of cells with mutations that affected the HIF-binding site, the authors confirmed reduced binding of HIF to the locus and observed a reduction in MYC and PVT1 RNA expression by 40 and 32%, respectively. The results indicated that the enhancer site affected by rs35252396 interacts with MYC and PVT1 promoters and is necessary for HIF-mediated transactivation of both genes to influence downstream effects of MYC in disease.

NEURODEGENERATIVE DISORDERS

While patient-specific iPSCs can be useful for examining effects of pathogenic variants in vitro, biological heterogeneity hinders the identification of causal variants, particularly for more complex traits. Genome-editing using iPSCs is becoming an increasingly attractive option for studies into GWAS variants as the technology to create lineage-specific cell lines continues to improve. A study by Soldner et al. (63) employed CRISPR/Cas9 genome editing to explore two risk variants, rs356168 and rs3756054, for Parkinson’s disease in a noncoding distal enhancer element that regulates synuclein alpha (SNCA), an important gene in the pathogenesis of the disease, by inserting all genotype combinations into the iPSCs and measuring alteration in gene expression. The cells were differentiated into neuronal precursors or mixed neuronal cultures and expression examined by qRT-PCR. The authors observed that cells carrying the rs356168 G allele showed significantly higher expression, independent of other variants, in line with the GWAS findings.

AUTOIMMUNE DISORDERS

Tewhey et al. (69) used MPRA to identify a number of potentially functional variants at GWAS locations in lymphoblast cell lines. Looking at 32,373 variants, the study identified 842 whose alleles showed differential gene expression. The authors focused on one variant, rs9283753, associated with both risk for ankylosing spondylitis and allele-specific reporter expression. Using a CRISPR/Cas9 system to switch only this allele in their cellular system, the study demonstrated that rs9283753 affected enhancer activity on Prostaglandin E Receptor 4 (PTGER4) gene expression, providing strong indications of causality at this GWAS locus.

CONCLUSIONS AND FUTURE DIRECTIONS

The examples described here provide insight into the multiple ways genome-editing promises to transform studies into the functionality of GWAS variants, primarily within regulatory genomic regions, and these processes are summarized in Fig. 2. The approaches used to examine these variants also highlight some of the advantages and disadvantages of study design when using genome-editing technology, where focus should be directed toward examination of an individual variant rather than deletion of large regions and use of appropriate tissues. Tools such as TALENs or CRISPR/Cas9 are still relatively recent, and there is substantial scope to expand and further develop genome-editing technology to facilitate these goals. Recent advances have been made with base-editing, a form of genetic engineering that facilitates irreversible conversion of base pairs without requiring double-strand DNA breaks or donor DNA templates (26, 41, 42, 54). These new tools are increasing their capabilities to allow for different PAM compatibilities, enhanced DNA specificity, improved editing efficiency, and narrowed editing windows and have the potential to overtake CRISPR/Cas9 in future genome-editing strategies.

Fig. 2.

Experimental methodologies to build functional evidence for genome-wide association study (GWAS) variants using genome-editing technologies.

Due to the lengthy process required to isolate edited cell lines with the desired variant, studies have so far focused on loci where there are only a few candidate functional SNPs either due to low LD in the region or through fine-mapping and analysis of chromatin marks before commencing genome-editing. The latter requires that “reduction” methodologies do not overlook potentially functional variants, which will always be a concern, particularly if there may be more than one causal variant at a locus. Future directions for improving genome-editing to streamline GWAS variant analysis will focus on increasing the efficiency and targeting of genome-editing, facilitating the investigation of multiple variants, and perform unbiased functional analysis in the same way that GWAS performs unbiased association analyses.

Genome-editing applied to model pathogenic variants is becoming increasingly reliant on hPSCs and their differentiation to cell types that closely resemble those relevant to the traits being examined. Such cells are readily edited, and differentiation to multiple cell types allows for potential effects in multiple tissues to be examined, an important issue when a GWAS variant shows no clear target tissue. Continued development of hPSC differentiation protocols will increase the number of cell types available for meaningful genome-editing research. In parallel to hPSCs, the use of genome-editing to create specific animal models of disease by targeting protein-coding regions is becoming well established (20, 71). One of the next challenges for examination of GWAS variants will involve expanding gene-editing techniques to study noncoding variations in the same manner in animal models to gain a fuller understanding of the physiological role of risk variants. Such studies have begun, for example, the deletion of the mouse Myc-355, a putative regulatory element in mouse, which in humans, contains rs6983267, a variant that accounts for more cancer-related morbidity than any other variant (66). These mice showed reduced expression of Myc and increased resistance to tumorigenesis induced by APC mutations. The use of technologies such as CRISPR/Cas9 is likely to facilitate such studies. However, since noncoding elements are less conserved between species, and these enhancers often differ in critical components (9), it will be a greater challenge, likely to involve insertion of human regulatory sequences at the orthologous location in the animal model to examine allele-specific phenotypic differences.

In addition to single nucleotide variants, future studies may also examine structural variations in more detail, where differences in genomic DNA can range from kilobase to chromosomal magnitude. Such variations are often associated with disease and can contribute large portions of genomic variability. Structural variants have been created using ZFN (45) and TALENs (1) and CRISPR-based systems (43, 78). Many of these structural variants encompass noncoding regions of the genome, and studies in the relevant human cellular models are needed to evaluate their impact.

The last 10 yr of genetic association studies was transformed by the development of low-cost, high-throughput genotyping arrays. The next 10 yr and beyond is likely to be focused on analyzing of this wealth of association signals, with genome-editing one of the major tools to facilitate this endeavor. As with genotyping and sequencing technologies, the challenge for genome editing will be to develop cheaper, highly parallel methodologies that will allow us to test every variant in the genome for function.

GRANTS

This work is supported by the National Institute for Health Research Biomedical Research Centre at Barts. A. J. P. Smith is a British Heart Foundation (BHF) Intermediate Fellow (FS/13/6/29977). P. Deloukas is supported by BHF Grant RG/14/5/30893.

DISCLOSURES

No conflicts of interest, financial or otherwise, are declared by the authors.

AUTHOR CONTRIBUTIONS

A.J.P.S. prepared figures; A.J.P.S., P.D., and P.B.M. drafted manuscript; A.J.P.S., P.D., and P.B.M. edited and revised manuscript; A.J.P.S., P.D., and P.B.M. approved final version of manuscript.

APPENDIX: USEFUL URLs:

GWAS Catalog, https://www.ebi.ac.uk/gwas/

Encyclopedia of DNA Elements - ENCODE, http://genome.ucsc.edu/ENCODE/

Roadmap Epigenomics Project, http://www.roadmapepigenomics.org

HaploReg, https://www.broadinstitute.org/mammals/haploreg

RegulomeDB, http://www.regulomedb.org/

Addgene guide to CRISPR/Cas9, https://www.addgene.org/crispr/guide/

Addgene guide to TALENs, https://www.addgene.org/talen/guide/