Genome-wide association studies of structural birth defects: A review and commentary

Abstract

Background:

While there is strong evidence that genetic risk factors play an important role in the etiologies of structural birth defects, compared to other diseases, there have been relatively few genome-wide association studies (GWAS) of these conditions. We reviewed the current landscape of GWAS conducted for birth defects, noting novel insights, and future directions.

Methods:

This article reviews the literature with regard to GWAS of structural birth defects. Key defects included in this review include oral clefts, congenital heart defects (CHDs), biliary atresia, pyloric stenosis, hypospadias, craniosynostosis, and clubfoot. Additionally, other issues related to GWAS are considered, including the assessment of polygenic risk scores and issues related to genetic ancestry, as well as utilizing genome-wide single nucleotide polymorphism array data to evaluate gene–environment interactions and Mendelian randomization.

Results:

For some birth defects, including oral clefts and CHDs, several novel susceptibility loci have been identified and replicated through GWAS, including 8q24 for oral clefts, DGKK for hypospadias, and 4p16 for CHDs. Relatively common birth defects for which there are currently no published GWAS include neural tube defects, anotia/microtia, anophthalmia/microphthalmia, gastroschisis, and omphalocele.

Conclusions:

Overall, GWAS have been successful in identifying several novel susceptibility genes and genomic regions for structural birth defects. These findings have provided new insights into the etiologies of these phenotypes. However, GWAS have been underutilized for understanding the genetic etiologies of several birth defects.

1 |. INTRODUCTION

Approximately one in 33 babies born in the United States each year is affected by a birth defect (Centers for Disease Control and Prevention (CDC), 2008); however, individual types of birth defects are relatively rare. In the United States, birth defects are the leading cause of infant death (Kochanek, Murphy, Xu, & Arias, 2017), contribute substantially to morbidity and disability, and in 2013, the estimated annual cost of birth defect-associated hospitalizations was $22.9 billion (Arth et al., 2017). Substantial stress and disruption of family life accompany this economic burden. Despite their public health significance, the causes of most birth defects remain unknown. There are several lines of evidence indicating that inherited genetic risk factors play an important role in the etiologies of these conditions. Existing evidence from human studies includes increased concordance among monozygotic twins compared to dizygotic twins, among full siblings compared to half siblings, and among first-degree relative compared to second- and third-degree relatives (reviewed in [Webber et al., 2015]). However, candidate gene studies (i.e., where genes are selected based on current understanding of the disease) have largely produced equivocal findings related to genetic susceptibility for structural birth defects (Hobbs et al., 2014; Lupo et al., 2017; Webber et al., 2015).

Advances in technology that permit affordable and reliable genotyping of millions of single nucleotide polymorphisms (SNPs) have provided the opportunity to expand beyond candidate gene association studies to genome-wide studies (GWAS) that do not require prior hypotheses regarding underlying disease biology. Although the first GWAS, for age-related macular degeneration, was not published until 2005, the NHGRI-EBI GWAS Catalog (https://www.ebi.ac.uk/gwas/) currently includes data from over 5,600 GWAS (Buniello et al., 2019). In GWAS, hundreds of thou-sands or millions of genetic variants are tested for association. Because of the large number of tests, the commonly accepted threshold for statistical significance in a GWAS of common variants (i.e., minor allele frequency > 5%) is p < 5 × 10⁻⁸ (Fadista, Manning, Florez, & Groop, 2016), and thus very large study populations are required to provide adequate statistical power. Additionally, a tiered approach is typically used in GWAS, where a subset of SNPs from the first stage (i.e., discovery set) is moved to a second stage (i.e., replication set) for confirmation. This process limits the potential for false positives. Meta-analyses performed across studies are commonly used to confirm or refute previously reported associations and can identify novel candidate loci (de Bakker et al., 2008; Willer, Li, & Abecasis, 2010). Meta-analyses also provide opportunities to identify genes with multiple significant SNPs or regions that might not be identified in a single GWAS.

In comparison to other diseases, there have been relatively few GWAS of structural birth defects (Agopian et al., 2014; Birnbaum et al., 2009; Cordell et al., 2013; Mitchell et al., 2015; van der Zanden et al., 2011). This is largely due to the difficulty of assembling the large study populations needed for GWAS (i.e., >1,000 affected individuals), especially as structural birth defects tend to be individually rare. Nonetheless, GWAS have been completed for some of the most common structural birth defects, and these studies have provided new insights regarding the genetic contribution to disease etiology. The purpose of this review is to outline the discoveries made in GWAS of selected structural birth defects and to propose strategies for future genomic studies of these conditions.

2 |. STUDY DESIGNS

The majority of GWAS have used a case–control study design. In this approach, the frequency of variants in putative disease genes is compared between cases (i.e., individuals with the condition) and controls (i.e., individuals without the condition). This study design is particularly useful when studying rare out-comes, such as structural birth defects, and is often used in genetic association studies of other conditions. For instance, the case–control study design was used to identify genes associated with neuroblastoma (a relatively rare pediatric malignancy), which has led to improved therapeutic options for these children (Bosse & Maris, 2016). However, genetic association studies using the case–control design are vulnerable to a type of confounding referred to as population stratification bias, where a false association between a genotype and disease, or the masking of a true genotypic effect, is induced by the existence of subgroups within a population (e.g., different racial or ethnic groups) that have different genotype frequencies and frequencies of disease (Campbell et al., 2005).

Another study design that emerged in the 1990s has proven to be very useful in genetic association studies of structural birth defects: the case-parent trio design (or child-parent trio). Studies of birth defects that employed this design (e.g., [Mitchell, 2008]) were initiated shortly after the advent of this method. The trio is composed of the affected child and his or her biological parents. This design is particularly useful for studies of birth defects and conditions with early disease onset, since parents of children with these conditions are generally available. Several methods for analyzing the data generated in a case-parent trio study have been developed, including the transmission disequilibrium test (Schaid & Sommer, 1993; Spielman, McGinnis, & Ewens, 1993) and approaches using log-linear models (Weinberg, Wilcox, & Lie, 1998). The child–parent trio design has the advantage (as compared to case–control studies) of being immune to population stratification bias when assessing the effects of the inherited genotype. Furthermore, this design can be used to assess maternal genetic effects (i.e., the effect of the maternal genotype on the phenotype of offspring) without incurring additional genotyping expenses (i.e., in the case–control design, evaluation of the maternal and case genotype would require genotyping cases and controls as well as the mothers of these individuals).

3 |. GWAS OF SELECTED BIRTH DEFECTS

3.1 |. Oral clefts

Oral clefts represent a group of structural birth defects where there is a gap or break in normal features of the mouth, most commonly the roof of the mouth (the palate) or the upper lip or both. The most common anatomical forms of orofacial clefts include cleft lip (CL), cleft palate (CP), and cleft lip and palate (CL/P). The overall prevalence of oral clefts is one in 1,000, and because it is one of the most common groups of birth defects, there have been a large number of genetic association studies to map genes underlying susceptibility (Beaty, Marazita, & Leslie, 2016). Not surprisingly, several novel genes and genomic regions underlying the occurrence of oral clefts have been identified through these efforts, including 8q24, IRF6, and NOG. A list of these genes and regions is presented in Table 1. Additionally, oral cleft-associated loci reported in the NHGRI-EBI GWAS Catalog (https://www.ebi.ac.uk/gwas/) are displayed in Figure 1. Of these loci (Figure 1), four (IRF6, 8q24, 17q22, and 10q25.3) appear to account for 20–25% of the estimated heritability to CL/P, a much larger proportion of the estimated heritability attributable to markers identified by GWAS than seen for many other complex disorders (Beaty et al., 2016).

TABLE 1.

A sample of genes and genomic regions identified in GWAS of oral clefts (adapted from Beaty et al., 2016)

Locus	Gene	Phenotype	p-valuea
1p36.13	PAX7	CL/P	6 × 10−13
1p36	GRHL3	CP	4 × 10−9
1q32.2	IRF6	CL/P	9 × 10−22
2p21	THADA	CL/P	9 × 10−8
2p24	FAM49A	CL/P	6 × 10−22
3p11	EPHA3	CL/P	4 × 10−8
3q12	COL8A1/FILIPIL	CL/P	4 × 10−7
8q24	Gene Desert	CL/P	8 × 10−44
10q25.3	VAX1	CL/P	7 × 10−13
13q31.2	SPRY2	CLP	8 × 10−6
15q22	TPM1	CL/P	4 × 10−7
15q24	ARID3B	CL/P	2 × 10−8
16p13	CREBBP	CL/P	9 × 10−12
17p13.1	NTN1	CL/P	8 × 10−21
17q22	NOG	CL/P	9 × 10−9
20q12	MAFB	CL/P	9 × 10−13

Abbreviations: CL/P, cleft lip with or without cleft palate; CLP, cleft lip and palate; CP, cleft palate.

^aLowest p-value for representative SNP in gene or genomic region as reported in the GWAS Catalog (https://www.ebi.ac.uk/gwas/).

FIGURE 1.

Loci identified in genome-wide association studies (GWAS) of oral clefts (obtained from the NHGRI-EBI GWAS Catalog (https://www.ebi.ac.uk/gwas/)

GWAS have confirmed the genetic contribution to the etiology of oral clefts. However, it has also demonstrated that these defects can result from variation in multiple genes. Therefore, compared to other birth defects where less is known in terms of genetic etiologies, the next challenge will be to develop strategies for characterizing the function of these genes in relation to oral cleft development and to trans-late this information into prevention strategies.

3.2 |. Congenital heart defects

With a birth prevalence of ~1%, congenital heart defects (CHDs) are the most common group of birth defects. Consequently, it is not surprising that CHDs are one of the few birth defects for which there have been several GWAS (Table 2). These studies have identified variants in several genes and regions of the genome with genome-wide significant (p < 5×10⁻⁸) or suggestive (commonly defined as 5 × 10⁻⁸ < p < 1 × 10⁻⁵ [Aminkeng et al., 2015; Kraja et al., 2017; Sung et al., 2018]) evidence of association with CHDs as a broad group, narrower subtypes of CHDs (e.g., septal or conotruncal defects), and even individual CHD phenotypes (e.g., tetralogy of Fallot). The top hits reported from these studies do not overlap, which—given the stringent threshold for declaring significance—is not uncommon for GWAS. Several of the implicated variants have, however, showed evidence of association in replication samples evaluated as part of the original GWAS (Cordell et al., 2013; Cordell, Bentham, et al., 2013; Lin et al., 2015), and a few have been independently replicated. In particular, the association of atrial septal defects (ASDs) and a locus at chromosome 4p16 (Cordell, Bentham, et al., 2013), has been independently replicated in two studies of ASDs in Chinese populations (Zhao et al., 2014; Zhao, Li, et al., 2015).

TABLE 2.

Summary of findings from GWAS of congenital heart defects

Reference	Phenotype	Population	Design	No. of cases	Chromosome/Region	Genes	SNP	p-value	OR (95% CI)
Cordell, Bentham, et al. (2013)	TOF	Europe, Australia, and Canada	CC	1,633	12q24	1.6 Mb, evolutionarily conserved, haplotype including 15 genes; strongest candidate, PTPN11	rs11065987	7.7 × 10−11	1.3 (1.1–1.4)
					13q32	GPC5 (intron)	rs7982677	3.0 × 10−11	1.3 (1.2–1.5)
Cordell, Topf, et al. (2013)	osASD	Europe, Australia, and Canada	CC	791	4p16	Intergenic, between STX18 and MSX1	rs870142	2.6 × 10−10	1.5 (NR)
Hu et al. (2013)	ASD and/or VSD	Han Chinese	CC	3,097	1p12	Intergenic, between SPAG17 and TBX15	rs2474937	8.4 × 10−10	1.4 (1.3–1.6)
					4q31.1	MAML3	rs1531070	5.0 × 10−12	1.4 (1.3–1.5)
Lin et al. (2015)	CHD	Han Chinese	CC	6,053	4q31.22	upstream of EDNRA	rs1400558	1.6 × 10−9	1.2 (1.1–1.2)
					9p24.2	close to SMARCA2	rs7863990	3.7 × 10−14	1.3 (1.2–1.4)
					12q24.13	upstream of TBX3 and TBX5	rs2433752	1.0 × 10−10	0.8 (0.8–0.9)
					20q12	PTPRT	rs490514	1.2 × 10−13	1.2 (1.1–1.2)
Agopian et al. (2014)	CTD	United States	CPT	1,108	22q13.1	KCNJ4 (intron)	rs2267386	1.4 × 10−6	3.1 (1.9–5.1)
					20p12.3	Intergenic, between BMP2 and FERMT1	rs6140038	1.0 × 10−6	5.2 (2.5–11.0)
Mitchell et al. (2015)	LSL	United States	CPT	601	16q24.2	Intergenic, between FOXL1 and c16orf95	rs8061121	4.0 × 10−9	2.6 (1.9–3.7)
					3p14.2	SYNPR (intron)	rs1975649	3.4 × 10−7	1.6 (1.4–2.0)
Agopian et al. (2017)	CTD	United States	MA (CPT + CC)	1,431	5p12	LOC648987 (noncoding)	rs6886261	1.7 × 10−7	1.8 (1.4–2.2)
	LSL		MA (CPT)	509	2q11.2	MGAT4A (intron)	rs11894932	1.5 × 10−7	0.4 (0.3–0.6)
					6p24.3	Intergenic, between OFCC1 and HULC	rs72820264	2.1 × 10−8	2.0 (1.6–2.6)
	CTD + LSL		MA (CPT + CC)	1,940	2q22.1	LRP1B (intron)	rs11895588	7.3 × 10−7	0.5 (0.4–0.7)
					6p24.3	Intergenic, between OFCC1 and HULC	rs56409046	2.7 × 10−7	1.3 (1.2–1.5)
					14q21.1	Intergenic, between SLC35B3 and OFCC1	rs1176869	9.2 × 10−7	1.3 (1.2–1.5)
Hanchard et al. (2016)	LSL	United States, Austria	CC	778	20q11.22	200 kb region, including MYH7B, miR-499A, miR-499B, GSS, and TRPC4AP	rs6088703	3.0 × 10−9	1.6 (NR)
Bjornsson et al. (2018)	CoA	Iceland	CC	120	14q11	MYH6	p.Arg721Trp	5.0 × 10−22	44.2 (20.5–95.5)

Abbreviations: ASD, atrial septal defect; CHD, congenital heart defect (broad definition); coA, coarctation of the aorta; CTD, conotruncal heart defect; LSL, left-sided lesion, osASD, ostium secundum atrial septal defect; TOF, tetralogy of Fallot; VSD, ventricular septal defect; CC, case–control study; CPT, case-parent trio study; MA, meta-analysis.

Attempts to replicate the associations uncovered by GWAS have generally been conducted using data from cases with the same CHD phenotypes that were included in the original GWAS. However, there have been attempts to determine whether associations detected in one phenotypic group (e.g., ASDs) are also observed in other CHD subtypes. These studies indicate that some associations are phenotype-specific, whereas others apply to a range of different phenotypes. For example, the association reported in the 4p16 region appears to be specific to ASDs (Cordell, Bentham, et al., 2013; Zhao et al., 2014), several of the associations reported by Lin and colleagues seem to apply to a broad range of CHD phenotypes (Lin et al., 2015), and the rare variant identified as being strongly associated with coarctation of the aorta is associated with other CHDs (e.g., bicuspid aortic valve, p = 7.3 × 10⁻⁸), as well as other cardiovascular disease phenotypes (e.g., atrial fibrillation, p = 1.1 × 10⁻¹⁴; Bjornsson et al., 2018).

In addition to the traditional GWAS of inherited variants (summarized above and in Table 2), several additional GWAS of CHDs have been conducted. These include a study of inherited compound heterozygous genotypes (Jiang et al., 2018), and a study of inherited genotypes and neurodevelopmental outcomes following cardiac surgery in infancy (Kim et al., 2012). In addition, three GWAS have focused on the maternal genotype (Agopian et al., 2014; Agopian et al., 2017; Mitchell et al., 2015). There have also been two GWAS conducted in syndromic populations: Down syndrome (Ramachandran et al., 2015) and the 22q11 deletion syndrome (Guo et al., 2017). Each of these studies focused on the inherited genotype and identified genes and genomic regions with at least suggestive evidence of association. For example, a genome-wide significant association between tetralogy of Fallot and an intronic variant in GPR98 (p = 3.0 × 10⁻⁸) was identified in the study of individuals with the 22q11 deletion syndrome. However, given the relatively unique approaches used in each of these studies, there has been little to no internal or external replication of these findings.

In summary, GWAS have provided evidence that common genetic variants are associated with CHDs and identified new candidate CHD genes and genomic regions. These studies provide support for a genetic model of CHDs that includes genes that influence specific CHD phenotypes (e.g., ASD) as well as genes that are associated with a broader spectrum of CHD phenotypes. In addition, GWAS conducted in syndromic populations indicate that common genetic variants may also contribute to variability in CHD phenotypes observed across affected individuals. As has been observed for other complex traits (e.g., autism), additional insights regarding the genetic basis of CHDs are expected to be gained by further analyses of the existing GWAS datasets (e.g., meta-analyses, analyses of specific CHD phenotypes) and through the evaluation of new GWAS samples.

3.3 |. Other defects

3.3.1 |. Biliary atresia

Biliary atresia is a birth defect characterized by inflammation and obliteration of the extrahepatic and intrahepatic bile ducts (Lee, Lewis, Schoen, Brand, & Ricketts, 2001; Sanchez-Valle et al., 2017; Sundaram, Mack, Feldman, & Sokol, 2017). While this condition is relatively rare with an estimated birth prevalence of 0.7–0.9 per 10,000 births (Caton, Druschel, & McNutt, 2004; Yoon, Bresee, Olney, James, & Khoury, 1997), biliary atresia is the most common cause of extrahepatic obstructive jaundice in the newborn and the most frequent indication for liver transplantation in children (Sundaram et al., 2017; Yoon et al., 1997). Four independent GWAS of biliary atresia among relatively small cohorts of patients (35–499 patients) have identified four novel biliary atresia susceptibility loci (Table 3): (a) an intergenic locus on 10q24.2 between ADD3 and XPNPEP1 (Garcia-Barcelo et al., 2010); (b) a deletion in 2q37.3 that included AGXT and GPC1 (Cui et al., 2013); (c) ARF6 (Ningappa et al., 2015); and (d) EFEMP1 (Chen et al., 2018).

TABLE 3.

Summary of findings from GWAS of noncardiac defects

Reference	Population	Design	No. of cases	Chromosome/Region	Genes	SNP	p-value	OR (95% CI)
Biliary atresia
Garcia-Barcelo et al. (2010)	Han Chinese	MA (CC)	324	10q25.1	Between XPNPEP1 and ADD3	rs17095355	6.9 × 10−9	1.8 (1.4–2.3)
Cui et al. (2013)	CHOP patients	CC	61	2q27.3	GPC1	CNV	4.4 × 10−10	NR
Ningappa et al. (2015)	Caucasian	CC	63	14q21.3	3′ flanking enhancer region for ARF6	rs3126184	5.9 × 10−7	2.7 (NR)
						rs10140366	5.6 × 10−7	2.7 (NR)
Chen et al. (2018)	European-American	MA (CC)	499	2p16.1	EFEMP1	rs10865291	3.7 × 10−8	NR
						rs6761893	3.4 × 10−8	NR
						rs727878	4.3 × 10−8	NR
Pyloric stenosis
Feenstra et al. (2012)	Denmark	MA (CC)	1,700	3p25.1	150 kb upstream of MBNL1	rs11712066	1.5 × 10−17	1.6 (1.4–1.8)
			1,693	3p25.2	Intergenic,1.3 Mb downstream of MBNL1	rs573872	4.3 × 10−12	1.4 (1.3–1.6)
			1,691	5q35.2	64 kb downstream of NKX2-5	rs29784	1.1 × 10−15	1.4 (1.3–1.6)
Feenstra et al. (2013)	Denmark, Sweden, United States	MA (CC)	2,656	11q23.3	301 bases downstream of APOA1	rs12721025	1.9 × 10−10	1.6 (1.4–1.8)
Fadista et al. (2019)	Denmark, Sweden, United States	MA (CC)	3,179	2p21	EML4	rs6736913	3.0 × 10−15	2.3 (2.1–2.5)
			3,178	9q22.32	1 kb downstream of BARX1	rs1933683	3.1 × 10−9	1.3 (1.2–1.4)
Hypospadias
van der Zanden et al. (2010)	Netherlands, Sweden	CC	835	X	DGKK	rs1934179	2.5 × 10−11	2.5 (1.9–3.2)
					DGKK	rs7063116	2.9 × 10−9	2.3 (1.7–3.0)
Geller et al. (2014)	European	CC	2,978	2	PKDCC (1.6 kb)	rs988958	1.6 × 10−10	1.3 (NR)
Craniosynostosis, sagittal
Justice et al. (2012)	European, non-Hispanic White	MA (CPT + CC)	302	20p12.3	345 kb downstream of BMP2	rs1884302	1.1 × 10−39	4.4 (3.5–5.5)
			302	7p14.3	167 kb region of BBS9	rs10262453	5.6 × 10−20	0.2 (0.2–0.3)

Abbreviations: CC, case–control study; CPT, case-parent trio study; MA, meta-analysis.

3.3.2 |. Pyloric stenosis

Pyloric stenosis is characterized by a narrowing of the pyloris, a sphincter muscle connecting the stomach and the duodenum (Ranells, Carver, & Kirby, 2011). It regulates the movement of food into the small intestine. The incidence of pyloric stenosis varies between two and five per 1,000 live births and there is a four- to five-fold higher risk in males than females (Peters, Oomen, Bakx, & Benninga, 2014). Three GWAS have been conducted using surgery-confirmed cases and controls from Denmark in the discovery phase (Fadista et al., 2019; Feenstra et al., 2012; Feenstra et al., 2013). Replication samples were drawn solely from the same population as the discovery phase in the earliest study (Feenstra et al., 2012) while the 2013 study (Feenstra et al., 2013) also included replication samples from the United States (mostly non-Hispanic white) and Sweden. In the most recent study, Fadista et al. (2019)) conducted a genome-wide meta-analysis, combining their previous GWAS cohort with an additional 427 surgery-confirmed cases and 2,031 controls, all of Danish descent, in their discovery phase. Genome-wide significant results were replicated in populations of European descent followed by confirmation in a Hispanic population. The meta-analysis confirmed the genome-wide significant SNPs identified in the two earlier GWAS (SNPs located close to MBNL1 and NKX2-5 (Feenstra et al., 2012), and APOA1 (Feenstra et al., 2013)) and reported two novel loci that included EML4, MTA3, and BARX1 (Table 3).

MBNL1 is a member of the muscleblind protein family that is involved in regulation of alternative splicing and NKX2–5 is crucial for the formation of pyloric sphincter muscle tissue (Ho et al., 2004; Self, Geng, & Oliver, 2009). The three genome-wide significant SNPs located close to MBNL1 and NKX2–5 collectively explain 1.8% of the variance in liability to pyloric stenosis (Feenstra et al., 2012). Due to the excess risk in males, heterogeneity of effects between the sexes was assessed for the three SNPs with no evidence of a difference between males and females. However, a variant on chromosome 19p13.2 had a strong effect in males with no effect in females, warranting further investigation. Apolipoprotein A-1 is the major protein component of HDL cholesterol in plasma (Davidson & Thompson, 2007) and pyloric stenosis is a clinical feature in patients with Smith–Lemli–Opitz syndrome, which is characterized by low cholesterol levels. The relationship between lipid levels and pyloric stenosis warrants further study. EML4, MTA3, and BARX1 are expressed in the fetal and adult stomach (Kim, Woo, Kanellopoulou, & Shivdasani, 2011; The Human Protein Atlas, 2019), establishing them as strong candidate genes for pyloric stenosis.

3.3.3 |. Hypospadias

Hypospadias is one of the most common genitourinary malformations and occurs when the urethral opening develops ventrally at varying degrees of severity rather than at the distal end of the glans penis (Carmichael, Shaw, & Lammer, 2012). The first GWAS of hypospadias identified two loci in or near DGKK on the X chromosome that were associated with risk (Table 3) (van der Zanden et al., 2011). Notably, these associations were relatively strong (odds ratios [ORs] >2.0). Additionally, there is emerging evidence that these variants have subtype specificity. Specifically, genetic variation in DGKK appears to be limited to mild forms of hypospadias as compared to moderate or severe phenotypes (Richard et al., 2019). This supports hypotheses related to etiologic heterogeneity of hypospadias by classifications of severity.

A subsequent GWAS by Geller et al. (2014)) not only confirmed the role of DGKK (represented by another locus, rs4554617) on hypospadias risk, but also identified 21 additional SNPs (17 of which reached genome-wide level significance) associated with hypospadias. When considering these 22 SNPs together, they jointly explain 9.4% of the variance in liability to hypospadias.

3.3.4 |. Craniosynostosis

Craniosynostosis (CS), the premature fusion of one or several sutures of the skull, is a common birth defect that affects ~1 in 2,250 births (Boulet, Rasmussen, & Honein, 2008; Lajeunie, Le Merrer, Bonaiti-Pellie, Marchac, & Renier, 1995). The defect presents as nonsyndromic (i.e., without unrelated, major birth defects or developmental delay) or as a component of more than 100 genetic syndromes (Boulet et al., 2008; Kimonis, Gold, Hoffman, Panchal, & Boyadjiev, 2007). These syndromes account for only 15% of all cases, leaving the etiology undetermined for most individuals with CS. Major sutures involved in CS include sagittal (40–58%), coronal (20–29%), metopic (4–10%), and lambdoid (2–4%; Kimonis et al., 2007).

To date, the only GWAS completed for CS included cases with sagittal nonsyndromic CS and their parents (Justice et al., 2012). An international consortium identified candidate loci on chromosome 7, within BBS9, and on chromosome 20, near BMP2, and successfully replicated these findings in an independent, population-based case–control sample of newborn residual blood spots. Notably, the loci had opposite effects on risk; the locus on chromosome 7 showed a strong negative association (OR = 0.2; 95% CI: 0.2–0.3) and the locus on chromosome 20 showed a strong positive association (OR = 4.4; 95% CI: 3.5–5.5) with CS risk. In addition, genotyping of the common variant located near BMP2 in CS patients with rare heterozygous SMAD6 mutations has provided the first evidence for a 2-locus disease mechanism (Timberlake et al., 2016). SMAD6 is an inhibitor of BMP-induced osteoblast differentiation and this 2-locus model was estimated to account for approximately 3.5% of all CS cases.

3.3.5 |. Clubfoot

Clubfoot is a birth defect of the lower limb with a birth prevalence of ~1 in 1,000. An important role for genetic factors in clubfoot etiology is supported by high concordance rates in identical twins compared to fraternal twins (33% vs. 3%), and an increased risk to first-degree relatives compared to the general population. We identified one published GWAS of isolated clubfoot, which included 766 cases (discovery + replication) of European ancestry (Zhang et al., 2014). In this assessment, no SNP reached genome-wide level significance. The strongest evidence for genetic association was found with an intergenic SNP on chromosome 12q24.31 between NCOR2 and ZNF664 (rs7969148, combined OR = 0.6, p = 1.9 × 10⁻⁷).

Overall, GWAS have demonstrated that common genes are involved in the etiologies of many birth defects and have identified previously unrecognized functional components of the human genome. In fact, these findings may lead to new biological insights and prevention strategies for these conditions. Further analyses of existing GWAS datasets (e.g., meta-analyses), as well as GWAS in new datasets, will continue to mitigate knowledge gaps in birth defects research. Due to the demonstrated genetic and clinical heterogeneity of these birth defects, analyses of specific phenotypes (e.g., individual types of orofacial clefts, CHDs, CS, or hypospadias; isolated defects vs. multiple defects) would be informative.

4 |. THE VALUE OF GWAS IN THE AGE OF SEQUENCING

There is a debate about the utility of GWAS for identifying the role of inherited genetic variation on disease susceptibility utilizing SNP array data in the age of sequencing (Tam et al., 2019). Whole-exome and whole-genome sequencing (WES and WGS, respectively) studies have moved to the forefront of genomic research in recent years. This is largely due to advances in technology that have led to reductions in the cost and time required to sequence DNA. However, there are still important advantages to GWAS in relation to WES/WGS. In terms of WES, the focus is strictly on genetic variants that alter protein sequences, which only constitute 1% of the human genome. While this could be important in clinical settings for highly penetrant pathogenic variants (Yang et al., 2013), these account only for a small proportion of cases. In fact, most replicated SNPs from GWAS are in noncoding regions of the genome and would, thus, likely have been excluded from analyses of WES data. While interpreting findings from noncoding regions is challenging, it has led to new insights into the underlying biology of birth defects (e.g., 8q and oral clefts).

While WGS provides more complete coverage of the genome (like GWAS), there are still advantages to performing GWAS utilizing SNP array data. First, the costs of SNP arrays are lower compared to WGS (Tam et al., 2019). In fact, WGS remains the most costly of these three options (i.e., SNP arrays, WES, and WGS). Therefore, WGS in large sample sizes is often cost prohibitive. Second, the technology underlying SNP arrays is highly accurate and more mature compared to WGS. Third, the analytic pipelines for GWAS of SNP arrays are well-established and require less computational complexity. While some of these factors may be mitigated in the future (e.g., cost), there is still a clear rationale for GWAS using SNP arrays compared to WGS (Tam et al., 2019). Further, as illustrated in the study of coarctation of the aorta (Bjornsson et al., 2018), sequencing data can be used to impute rare variants into array-based data, thereby allowing for the evaluation of both common and rare inherited variants without the need to sequence all study participants.

5 |. OTHER CONSIDERATIONS AND FUTURE DIRECTIONS

5.1 |. Defects in need of GWAS

As previously noted, while GWAS have been successful in identifying genetic susceptibility loci associated with other complex traits (Visscher et al., 2017), there have been relatively few GWAS of birth defects (Agopian et al., 2014; Birnbaum et al., 2009; Cordell, Bentham, et al., 2013; Mitchell et al., 2015; van der Zanden et al., 2011). Those studies that have been conducted are limited by small sample sizes and single ancestry populations, leaving us with much to gain from this approach for future studies of these conditions.

Most birth defects have unknown causes (Nelson & Holmes, 1989); however, there is strong evidence that genetic factors contribute to their etiologies. To date, much of what is known about the genetics of birth defects includes effects of high-risk alleles that cause rare multiple malformation syndromes (Belmont, Mohapatra, Towbin, & Ware, 2004; de Munnik et al., 2015; Lewin, Glass, & Power, 2004; Maslen, 2004; Mori & Bruneau, 2004; Yates, Turner, Firth, Berg, & Pilz, 2017). Such alleles occur at very low frequency in the general population and explain relatively little of the population burden of birth defects. Common modest-risk alleles may explain a much greater proportion of overall cases (Reich & Lander, 2001). However, aside from structural birth defects outlined above, to date, most structural birth defects have not been included as part of GWAS. The main barrier to the GWAS approach for relatively rare individual birth defects is the need for large numbers of specimens. Collaborations between researchers with access to specimens and environmental data, such as the National Birth Defects Prevention Study (NBDPS) (Reefhuis et al., 2015; Yoon et al., 2001), will provide opportunities to discover novel gene-birth defect associations, and environmental factors with which they may interact. Such discoveries will improve the accuracy of risk assessment information, provide information about the biological mechanisms underlying birth defects, and identify potential therapeutic targets. Notably, GWAS is likely to be more useful for birth defects that are relatively common and for which there is evidence of multifactorial etiologies (Agopian, Eastcott, & Mitchell, 2012; Jenkins et al., 2019). A short list of birth defects with no GWAS to date, and for which this approach would be beneficial, include but are not limited to:

1. congenital anomalies of the nervous system, including spina bifida, encephalocele, anencephaly, and hydrocephalus;

2. congenital anomalies of the eye, including anophthalmia/microphthalmia and anterior chamber segment defects;

3. congenital anomalies of the ear, including anotia/microtia; (d) gastroschisis; and (e) omphalocele. Rare birth defects, and birth defects where de novo mutations are likely to play a role, would likely be better candidates for WES/WGS studies (Veltman & Brunner, 2012).

5.2 |. Race/ethnicity in GWAS

Over 80% of subjects included in all published GWAS have been of European ancestry (Bustamante, Burchard, & De la Vega, 2011; Rosenberg et al., 2010). In part, this is to limit the impact of population stratification bias. However, this exclusive focus on a few selected ancestry groups raises a number of critical questions. For example, are findings from studies dominated by those of European ancestry transferable to other populations (Ioannidis, 2009; Ntzani, Liberopoulos, Manolio, & Ioannidis, 2012)? Can disease biology be different among populations and thus characterized by distinct risk factors (Torgerson et al., 2011)? What is the contribution of ancestry-related genetic variation to ethnic differences in birth prevalence? These issues are of particular relevance to structural birth defects, where the prevalence varies substantially by race/ethnicity. While some GWAS of structural birth defects have not been limited to those of European ancestry (Tables 1–3), it is incumbent on genetic epidemiologists to conduct GWAS of structural birth defects among multi-ethnic populations.

5.3 |. Gene–environment interactions

Evidence of gene–environment interactions in birth defect etiologies has been observed for many years (e.g., [Christensen et al., 1999; Etheredge et al., 2012; Lacasana et al., 2012; Padula et al., 2018; Shaw et al., 2005; Wu et al., 2010]). Assessing these interactions is critical to uncovering genetic and/or environmental contributions that might otherwise be undetectable (i.e., genetic variants related to the birth defect might only be expressed in the subgroup of the population that is exposed to a specific environmental factor).

Although many gene–environment interaction studies of birth defects have included small study populations and modest numbers of variants, the approach has been expanded recently using GWAS data to assess common exposures among pregnant women (e.g., maternal alcohol consumption, maternal active and passive smoking, and multivitamin supplement use) (Beaty et al., 2011; Haaland et al., 2018; Wu et al., 2014). In these studies, both increased and reduced orofacial cleft risks were observed between the exposures and genetic variants identified in the GWAS (i.e., association with the genetic variant differed as a function of the environmental exposure). The transition from assessment of candidate genes to genome wide investigations should continue to increase in the foreseeable future due to decreasing costs of GWAS.

Expanded approaches that include analyses to assess gene–environment interactions using GWAS data are referred to as gene–environment wide interaction studies (GEWIS; Khoury & Wacholder, 2009) or genome-wide environmental interactions (GWEI; Aschard et al., 2012). Methods that improve limitations inherent in these early designs have been developed and include approaches that can account for the complex correlations between individuals in admixed populations (Chen et al., 2019), assess the impact of exposure misclassification (Boonstra et al., 2016), evaluate interactions using case subjects only (vs. the traditional method using case and control subjects) (Cornelis et al., 2012; Helbig et al., 2012), and assess interactions using exposed subjects only (Zhao, Fan, et al., 2015).

Of note, these approaches are limited by the exposure data available and novel approaches are needed to measure and estimate exposure. However, it is important to conduct these studies or risk missing a key component to understanding the biological mechanisms causing birth defects, improving the accuracy of risk assessment, and identifying potential targets for prevention.

5.4 |. Polygenic risk scores

Polygenic risk scores (PRS) are quantitative measures of risk summed across multiple risk alleles identified through GWAS. More specifically, the goal of PRS is to utilize aggregated genetic information, often obtained from GWAS, to better estimate the likelihood of a specific outcome (Gibson, 2019; Sugrue & Desikan, 2019). PRS have been generated for several conditions, including coronary artery disease, atrial fibrillation, Crohn’s disease, and Type 2 diabetes, in each case identifying a threshold above which a small percentage of the population has disease risk at least threefold higher than the general population (Khera et al., 2018). In one study of breast cancer risk, a PRS combined with conventional risk factors was able to identify 16% of the population who could benefit from earlier screening and 32% who could delay screening (Maas et al., 2016). While there is promise for the use of PRS in identifying at-risk populations, these tools have not become part of routine clinical care or prevention strategies (Sugrue & Desikan, 2019). However, there have been no large-scale studies to generate PRS for birth defects, much less evaluate the clinical utility of these models.

5.5 |. Gene-level GWAS

The genotype data (e.g., array and imputed) and summary statistics (e.g., association p-values, estimates of relative risk) generated as part of SNP-level (i.e., variant by variant) GWAS can be used for genome-wide studies conducted at the level of the gene. Gene-level analyses are therefore extremely cost-effective and also have the advantage of a reduced multiple correction burden relative to SNP-level GWAS (i.e., correction for approximately 20,000 genes vs. millions of SNPs). Thus, gene-level GWAS provide a useful complement to SNP-level GWAS, providing the opportunity for additional gene discovery (Tam et al., 2019).

Although statistical methods for gene-level GWAS are not as well established as the methods for SNP-level analyses, several approaches have been described and can be implemented using publically available programs (e.g., (de Leeuw, Mooij, Heskes, & Posthuma, 2015); (Wang et al., 2017)). Nonetheless, despite the availability of both data and methods, there has been only one published gene-based GWAS for birth defects (Sewda et al., 2019). This study identified eight candidate genes for conotruncal heart defects and provided additional evidence that genes involved in chromatin-modification and in ribonucleic acid splicing are associated with CHDs.

5.6 |. Mendelian randomization

In Mendelian randomization, investigators use genetic variants to determine whether an observational association between a nongenetic risk factor and an outcome is consistent with a causal effect (Davies, Holmes, & Davey Smith, 2018; Ross et al., 2015). The underlying assumption of Mendelian randomization relies on the natural, random assortment of genetic variants. More specifically, individuals are naturally “assigned” at birth a genetic variant that is associated with certain traits (e.g., elevated body mass index [BMI]). When determining the role of BMI on the risk of a disease, it is often difficult to disentangle the confounding effects of other variables. However, genetic variants associated with BMI are not likely to be associated with the confounders in question. Therefore, by leveraging information from GWAS of BMI or other cardiometabolic traits, investigators can evaluate the impact of these factors on the risk of a given disease. As GWAS of several traits and conditions (e.g., smoking, alcohol intake, infection) continue to grow, this information can be used to more fully characterize associations between nongenetic factors and structural birth defects.

6 |. RESEARCH PRIORITIES AND CONCLUSIONS AND FUTURE DIRECTIONS

Overall, GWAS have been successful in identifying novel susceptibility loci for common structural birth defects. These findings have provided new insights into the etiologies of these phenotypes. In spite of these successes, GWAS have been under-utilized and as a result, understanding of the genetic contribution to birth defects etiologies lags behind that of other conditions. As GWAS continue to evolve to include rare and coding variants, variants optimized for multi-ethnic populations, as well as improved capability to interrogate copy number variants (CNVs), the application of this approach to all structural birth defects will continue to improve. Future assessments could expand on these findings to better ascertain the mechanisms (e.g., evaluation of phenotypic heterogeneity) underlying these associations, lever-age existing GWAS data for additional studies (e.g., gene-level analyses, Mendelian randomization), expand GWAS to individuals of non-European ancestry, and follow up with functional studies on genes that have been identified.