Gene-environment interactions in birth defect etiology: Challenges and opportunities

Abstract

Birth defects are relatively common congenital outcomes that significantly impact affected individuals, their families, and communities. Effective development and deployment of prevention and therapeutic strategies for these conditions requires sufficient understanding of etiology, including underlying genetic and environmental causes. Tremendous progress has been made in defining the genetic basis of familial and syndromic forms of birth defects. However, the majority of birth defect cases are considered nonsyndromic and thought to result from multifactorial gene-environment interactions. While substantial advances have been made in elucidating the genetic landscape of these etiologically complex conditions, significant biological and technical constraints have stymied progress toward a refined knowledge of environmental risk factors. Defining specific gene-environment interactions in birth defect etiology is even more challenging. However, progress has been made, including demonstration of critical proofs of concept and development of new conceptual and technical approaches for resolving complex gene-environment interactions. In this review, we discuss current views of multifactorial birth defect etiology, comparing them with other diseases that also involve gene-environment interactions, including primary immunodeficiency and cancer. We describe how various model systems have illuminated mechanisms of multifactorial etiology and these models’ individual strengths and weaknesses. Finally, suggestions for areas of future emphasis are proposed.

1. Introduction

Birth defects occur in approximately 1 in 33 babies born in the United States and affect around 8 million newborns worldwide each year (Christianson, Howson, & Modell, 2006; Prevention, 2008). Defined broadly as any congenital structural or functional anomaly with a measurable impact on physical, intellectual, or social well-being, birth defects are a major cause of morbidity and mortality (CDC, 1992; Prevention, 2017). Defining the causes of birth defects is a considerable challenge. Despite the notable progress summarized below, up to 80% of birth defect cases still cannot be attributed to a specific cause (Feldkamp, Carey, Byrne, Krikov, & Botto, 2017; Toufaily, Westgate, Lin, & Holmes, 2018). Reducing the incidence and impact of birth defects through development of prevention and therapeutic strategies will require a more complete understanding of these underlying causes.

It has long been recognized that the inborn errors of development that manifest as birth defects have both genetic and environmental origins (Lancaster, 2011). A community of researchers, including pediatricians, geneticists, developmental biologists, and epidemiologists, have made substantial progress in elucidating the etiology of a subset of birth defects that are caused by individual genetic or environmental factors. However, it is now widely recognized that the vast majority of birth defects are multifactorial in origin and likely result from complex gene-environment interactions (Beames & Lipinski, 2020; Krauss & Hong, 2016; Lovely, Rampersad, Fernandes, & Eberhart, 2017; Nishimura & Kurosawa, 2022). Over the last several decades, our understanding of the genetic landscape of birth defect etiology has steadily progressed, with even greater resolution on the immediate horizon as emergent technological advancements are deployed. On the other hand, major gaps in our understanding of the environmental contribution to birth defect etiology remain. Moving forward, there is also increasing need—and opportunity—for integration across disciplines (Leslie, 2020; Sparrow, 2022). While a daunting challenge, defining specific gene-environment interactions and elucidating the nature of their mechanisms of action holds great promise for identifying high-risk individuals and advancing prevention strategies based upon communication of risk.

2. Defining the genetic landscape of birth defect etiology

A significant genetic component to birth defect etiology has long been recognized, but our current knowledge has largely been driven by an ever-expanding and increasingly powerful toolset developed over the last 50 years. Application of these tools has defined the genetic landscape of birth defect etiology to a remarkable degree, including the effective solving of most rare congenital disorders and the identification of dozens of genetic risk factors that contribute to the complex etiology of the most common human birth defects.

2.1. Syndromic birth defects

Syndromic birth defects are typically recognized by multiple anomalies affecting more than one organ system, and often segregate as autosomal dominant, autosomal recessive, or X-linked traits. Advances made in defining the genetic causes of these conditions were recently summarized by Innes and Lynch (Innes & Lynch, 2021), who framed their analyses around the classic textbook, Smith’s Recognizable Patterns of Human Malformation, now in its 8^th edition (Jones, Jones, & del Campo, 2021). The authors note that of the 147 conditions described in the seminal edition of Smith’s published in 1970, the majority are now “solved.” This is a true success story that can largely be credited to the individual and concerted efforts of clinicians and medical geneticists. Solving these conditions typically began with careful physical examination by the astute pediatrician to identify specific major and minor dysmorphologies. Subsequent ascertainment of a cohort of individuals sharing similar patterns of dysmorphology facilitated investigation of genetic underpinnings through cytogenetic assays, linkage analysis, array CGH, and SNP homozygosity mapping (Innes & Lynch, 2021). Over the last decade, next generation sequencing (NGS) approaches have become the main driver of discovery in this arena. With increasing power and decreasing cost, NGS promises to further elucidate the genetic architecture of syndromes, including mosaic disorders and ultra-rare conditions.

2.2. Nonsyndromic birth defects

The remarkable progress that has been made in delineating the etiology of syndromes and rare disorders should be celebrated, but it must also be recognized that the vast majority of human birth defects are considered non-syndromic and cannot be attributed to specific known causes (Feldkamp et al., 2017). Nonsyndromic or isolated birth defects (i.e. those not associated with multiorgan syndromes) account for up to 75% of all birth defect cases and encompass the majority of the most prevalent malformations, including heart and neural tube defects, orofacial clefts, and limb defects (Feldkamp et al., 2017; Parker et al., 2010). Nonsyndromic birth defects are etiologically complex traits generally thought to result when multiple genetic and environmental influences reach a critical threshold of insult (Dixon, Marazita, Beaty, & Murray, 2011; Krauss & Hong, 2016; Lovely et al., 2017; Murray, 2002). The relatively high prevalence of these conditions has enabled large population-based genetic epidemiology studies comparing cases and controls (Hobbs et al., 2014). These studies have been dominated by candidate gene approaches, like SNP genotyping, that are typically directed at genes linked to particular outcomes through animal model investigation or knowledge of the signaling pathways that regulate specific developmental processes. Genome-wide association studies (GWAS) on the other hand do not require pre-existing knowledge and have been applied to discover new candidate genes and novel risk loci for several non-syndromic birth defects including congenital heart defects, orofacial clefts, and hypospadias (Lupo, Mitchell, & Jenkins, 2019; Webber et al., 2015). Increasing application of NGS to nonsyndromic birth defects has already proven useful, particularly in identifying the contribution of relatively rare genetic variants.

Collectively, these efforts have been fruitful in discovering susceptibility loci associated with complex traits, thereby informing genetic counseling and family planning strategies; this should be counted among the major successes in birth defects research. At the same time, nonsyndromic birth defects are complex disorders that in most cases will not be accounted for by a single gene or chromosomal abnormality. Moreover, considering the multifactorial nature of nonsyndromic birth defects, it could be argued that even a complete understanding of the genetic architecture of these conditions is unlikely to be sufficient to drive prevention or correction strategies. While the ultimate goal of birth defect prevention is supported by elucidation of their genetic landscape to highlight high-risk populations and individuals, success will also require a refined understanding of modulable environmental factors that contribute to birth defect risk.

3. Elucidating the environmental landscape of birth defect etiology

Formal investigation into how the environment influences development and contributes to birth defect etiology began in earnest in the 1950s, following recognition that the uterus is not impervious to such influences. In the context of birth defect etiology, the “environment” broadly encompasses non-genetic factors, including drugs, medications, contaminants, pollution, and maternal disease, infection, and nutritional status. Relative to the genome, which provides a permanent record that can be retrospectively examined and defined to the individual base pair, our current knowledge of the prenatal environment and ability to assess individual and combinatorial environmental risk factors during critical periods of development is extremely limited.

3.1. Human teratogens

Perhaps the most well-known human teratogen is thalidomide, a drug marketed to pregnant women that caused limb malformations in thousands of children in the 1950s and 1960s. Early concern over thalidomide was raised by clinicians who noticed a dramatic increase in cases of an otherwise rare malformation involving shortened or absent limbs (Lancaster, 2011). Subsequently, alcohol and several other major human teratogens were identified by the astute clinician approach. As described by Jones and Carey, this approach relies upon observation, documentation, and delineation of cases in which a specific phenotype follows a particularly rare and known exposure (Jones & Carey, 2011). Prospective cohort studies follow a related path, in which tracking of prenatal exposures is paired with careful examination of children by a dysmorphologist to identify major and minor malformations. Facilitated by organizations like the Organization of Teratogen Information Services (OTIS), this approach has been particularly effective in evaluating the potential teratogenicity of drugs that have recently been marketed, as well as those that are commonly used by pregnant women but have not been adequately evaluated. While effective in identifying several major human teratogens, including the Zika virus, these approaches are best suited to examination of clearly defined and relatively rare prenatal exposures (Jones & Carey, 2011). While teratogen exposure can account for a small percentage of birth defects cases, the majority are thought to be multifactorial. For these etiologically complex outcomes, it is important to consider environmental factors that, while not individually causative, contribute to overall risk (Feldkamp, Botto, & Carey, 2015).

3.2. Environmental risk factors

Traditional epidemiological studies are equipped to identify common environmental influences that alter the risk for specific birth defect outcomes (Hernandez-Diaz & Oberg, 2015). Designed to assess the correlation between risk factors and specific outcomes, epidemiological studies have been central in establishing several key environmental risk factors, most notably folic acid. Epidemiological studies have proven particularly useful in examining intentional exposure to general classes of risk factors, like multivitamin use and smoking, as discussed in detail in the chapters on orofacial clefting and craniofacial malformations within this volume. Many epidemiological studies, however, rely on the recall of study participants of events that may have occurred years previously. Most structural birth defects result from insult during the 3rd to the 8th week of embryogenesis and have specific and narrow critical periods of susceptibility that may span only days (Finnell, Waes, Eudy, & Rosenquist, 2002; Shenefelt, 1972). Ascribing exposures to these windows of susceptibility is difficult, particularly for non-persistent chemicals (Kamai, McElrath, & Ferguson, 2019), as many pregnancies are not even recognized at these early stages. Moreover, the capacity of traditional epidemiological approaches is overwhelmed by the sheer number of chemicals used in global commerce, recently estimated in the tens or even hundreds of thousands (Bond & Garny, 2019; Wang, Walker, Muir, & Nagatani-Yoshida, 2020). Efforts to define the prenatal exposome (i.e. the totality of the environment) are in the very early stages and mostly limited to individual studies examining particular classes of chemicals (e.g. flame retardants, pesticides, metals etc.) from placenta or cord blood (Rager et al., 2020). These studies have produced evidence supporting relationships between prenatal exposure to select chemicals and adverse developmental outcomes. However, analyses to-date have generally been limited to a priori hypotheses, account for only a small fraction of the exposome, and were not equipped to examine chemical interactions. Moving forward there is a clear need to more comprehensively assess the prenatal exposome and to develop sophisticated approaches for assessing the risk of environmental factors in isolation and in combination.

3.3. High-throughput predictive approaches

The sheer number of chemicals in the environment outmatches the capacity of animal-based testing, which has traditionally been relied upon for developmental and reproductive toxicity assessment. The limited throughput and resource demands of these approaches, along with the desire to reduce animal usage, has spurred the development of alternative approaches. At the forefront of these efforts is Tox21, a collaboration between the National Toxicology Program, the Environmental Protection Agency National Center for Computational Toxicology, the National Institutes of Health Chemical Genomics Center, the National Center for Advancing Translational Sciences, and the Food and Drug Administration (Attene-Ramos et al., 2013). The Tox21 approach integrates high throughput robotic platforms, chemical profiling strategies, and hundreds of individual in vitro biological assays to build a foundation for mechanism-based toxicity prediction (Jeong, Kim, & Choi, 2022). While not specifically targeted toward developmental toxicity, assays examining endocrine disruption and other developmentally important signaling pathways and mechanisms are represented within the Tox21 repertoire. Meanwhile, the same conceptual and experimental approaches are being applied with specific focus on assessing teratogenic potential, via rodent whole embryo culture, zebrafish embryo, and stem cell assays (Zhang, Ball, Panzica-Kelly, & Augustine-Rauch, 2016). While the throughput nature of these assays addresses a central need for toxicity evaluation, key challenges and limitations remain to be fully addressed including ability to account for chemical metabolism, multiple exposures, and gene-environment interactions. These are particularly important challenges to address with respect to assessment of developmental toxicity given the dynamic nature of embryogenesis, multiple molecular and cellular targets of teratogens, and the unique profile of chemical metabolism during pregnancy. As such, high throughput screens are ideally positioned as one approach to developmental toxicity assessment that serves as a complement to traditional animal-based testing and human epidemiological approaches.

4. Gene-environment interactions

As described above, the challenges to understanding gene-environment interactions are many and include the historical siloing of scientific disciplines and the overwhelming number of potential permutations of genetic and environmental factors. But addressing these challenges promises great reward, including potential development of prevention strategies based upon identification of high-risk individuals and communication of culpable environmental risk factors. Toward that goal, it is instructive to consider the progress made in understanding other maladies, such as infectious diseases and cancer, that are also influenced by gene-environment interactions.

4.1. Gene-environment interactions in infectious diseases

Humans display enormous variability in the clinical outcome to primary infections with specific microbes (Casanova & Abel, 2013, 2020). Multiple variables are at play here, including route of infection, size of inoculum, and microbial variability. However, for most microbes, most people survive infections, with lethal disease a rare outcome (Casanova & Abel, 2013, 2020). In many cases, such severe reactions to specific microbial infections arise in individuals with germline mutations that cause specific primary immunodeficiencies (Casanova & Abel, 2013, 2020). Therefore, as is likely to be true for many birth defects, genetic variations predispose individuals to extreme outcomes in response to an environmental exposure (in this case, a specific microbial infection).

Almost 300 inborn errors of immunity have been identified that link monogenic mutations to rare phenotypes arising after specific infections, and these mutations may or may not result in phenotypes in the absence of infection (Casanova, 2015). In the simplest cases, the severity of the genetic defect is directly related to the tendency for clinical manifestation. A classic example is Mendelian susceptibility to mycobacterial disease (MSMD), in which susceptibility to environmental mycobacteria is associated with genetic deficiency in interferon (IFN)-γ production, signaling, and/or responses (Bustamante, 2020; Noma, Mizoguchi, Tsumura, & Okada, 2022). Eighteen different genes have been identified in different forms of MSMD, with various inheritance patterns, including autosomal recessive, autosomal dominant, and X-linked recessive (Bustamante, 2020; Noma et al., 2022). A spectrum of sensitivity exists in MSMD. Patients with autosomal recessive null mutations invariably develop environmental mycobacterial infections by 5 years of age, whereas partial (often autosomal dominant) genetic deficiencies result in a range of responses, from a longer period without symptoms, to mild or no disease at all (Bustamante, 2020; Gruber & Bogunovic, 2020; Noma et al., 2022). Therefore, MSMD shows incomplete penetrance and variable expressivity, dependent on the nature of the genetic alteration. Additionally, MSMD can be divided into “isolated” and “syndromic” forms, the former displaying selective susceptibility to mycobacterial and related infections and the latter also having additional phenotypes (Bustamante, 2020; Noma et al., 2022). Multiple other Mendelian and monogenic susceptibilities to infection display similar patterns (Casanova & Abel, 2020).

Primary immunodeficiencies therefore hold several features in common with birth defects associated with gene-environment interactions, including genetic heterogeneity, a spectrum of phenotypic severity, and isolated vs. syndromic disease. Furthermore, there are additional points of overlap between birth defects and infectious disease in less-well understood aspects of gene-environment interactions. In scenarios with partial penetrance, such as familial cases of the birth defect holoprosencephaly (HPE) and common variable immunodeficiency, respectively, cases may be digenic or polygenic in nature (Gruber & Bogunovic, 2020; Kim et al., 2019). Direct evidence for polygenic disease has been found in only a small number of cases of each, however, and the variants implicated tend to be rare ones (Hong et al., 2017). The role of common variants as modifier genes is still largely unknown for both birth defects and infectious diseases, but they may play significant roles when paired with specific mutations or environmental exposures (Gruber & Bogunovic, 2020; Timberlake et al., 2016). Finally, it has recently been appreciated that somatic mutation, i.e., genetic mosaicism, may underlie cases of both primary immunodeficiency and birth defects, and this phenomenon is a potential explanation for variable penetrance and severity of these defects (Gruber & Bogunovic, 2020; Latorre-Pellicer et al., 2021; Rodin & Walsh, 2018).

Despite these similarities, the study of gene-environment interactions in infectious disease holds certain advantages to such studies in birth defects. Most significantly, it is easier to unambiguously identify a specific microbial infection in a patient than it is to identify a potential teratogenic exposure occurring early in gestation, especially when the clinical outcome—which triggers initial interest in the question—is many months later. Additionally, discovery of such potential teratogens lags far behind our knowledge of the specific common microbes that are risk factors in primary immunodeficiencies. Finally, medicinal intervention may, in some cases, be simpler in primary immunodeficiencies. For example, some forms of MSMD result in suboptimal production of IFN-γ (e.g., those carrying mutations in IL-12Rb2 or IL-23R) and these patients may benefit by IFN-γ therapy (Bustamante, 2020; Noma et al., 2022). Nevertheless, these successes offer potential ways forward for study of developmental disorders.

4.2. Gene-environment interactions in cancer

Birth defects and cancer can both arise from defects in major signaling pathways, including the Hedgehog (HH), WNT, and receptor tyrosine kinase pathways. Frequently, loss of function of a pathway is associated with birth defects, while gain of function of the same pathway is associated with cancer (e.g., HH signaling deficiency results in HPE, while unregulated HH signaling underlies both sporadic and inherited predisposition to basal cell carcinoma and medulloblastoma (Ingham, 2022; Jiang & Hui, 2008)). There are clear exceptions to this generality, however, including activating mutations in the RAS-MAP kinase pathway, which are frequent occurrences in sporadic adult cancers and also responsible for the syndromes collectively designated “RASopathies” (we note that, although gain-of-function in nature, the specific mutations identified in these two scenarios are overlapping but not identical) (Hebron, Hernandez, & Yohe, 2022).

There are more than 100 distinct tumor predisposition syndromes (TPSs) (Carbone et al., 2020). The classical mechanism—seen with such tumor suppressor genes as RB1—involves inheritance of one mutant copy of a gene, with somatic mutation of the other copy invariably found in tumor cells (Dimaras et al., 2012). Therefore, inheritance is autosomal dominant, but the tumor cells lack any wild type protein. Most cancers are caused by accumulation of specific somatic mutations, and many known carcinogens are DNA-damaging agents (Totsuka, Watanabe, & Lin, 2021; Wu, Powers, Zhu, & Hannun, 2016; Wu, Zhu, Thompson, & Hannun, 2018). It is not surprising then, that a subset of TPSs are caused by germline mutations in genes involved with DNA repair and maintenance of genomic integrity (Carbone et al., 2020). For example, Lynch syndrome, the most common autosomal dominant TPS, is caused by heterozygous mutations in genes that control DNA mismatch repair (Carbone et al., 2020; Lynch, Snyder, Shaw, Heinen, & Hitchins, 2015). Consequently, Lynch syndrome patients have high mutation rates and high incidences of GI tract and other tumors. Interestingly, the prevalence of tumor types that develop in Lynch syndrome varies with geography; those in Asian countries develop stomach cancer more frequently than those in Western countries, suggesting an environmental impact in tumor tropism (Park et al., 2016; Wei et al., 2010).

TP53 (encoding the tumor suppressor, p53) is the most frequently mutated gene in human cancer (Levine, 2020). Heterozygous germline mutations in TP53 cause Li-Fraumeni syndrome, a highly penetrant TPS (Guha & Malkin, 2017). Humans and mice carrying TP53 mutations are highly susceptible to ionizing and UV radiation, and to tobacco-related carcinogens and cigarette smoke (Heymann et al., 2010; Hwang et al., 2003; Jiang, Ananthaswamy, Muller, & Kripke, 1999; Kilroe-Smith, 1976). Furthermore, p53-deficient mice exposed to the tobacco-related carcinogen, benzo[a]pyrene had higher levels of carcinogen-DNA adducts than control mice, demonstrating a gene-environment interaction at a molecular level that is relevant to disease causation (Lorente et al., 2004).

Finally, a striking gene-environment interaction in cancer was observed in villages in Cappadocia, Turkey, where >50% of the population died of mesothelioma (Carbone et al., 2020). Mesothelioma is associated with exposure to asbestos, but only a relatively small fraction of workers with high occupational exposure develop the disease (Miller, 1991). Erionite, a potent fiber similar to asbestos, was used in housing and roads in the Turkish villages, but cases and deaths clustered in families. These families carried heterozygous germline mutations in BAP1 (encoding a protein which binds to BRCA1, another tumor suppressor); furthermore, tumor cells had somatic inactivation of the wild type copy, consistent with a classical tumor suppressor mechanism (Testa et al., 2011). Additional BAP1 families in the United States and elsewhere show a similar susceptibility to mesothelioma and additional cancers (Abdel-Rahman et al., 2011). While mesothelioma can occur in BAP1 mutation carriers not exposed to asbestos, and asbestos can cause mesothelioma in individuals without such mutations, the combination of the two results in much higher incidence than either alone.

Gene-environment interactions are therefore important in both birth defects and cancer. In each case, identifying the most significant genes has been more straightforward than identifying rate-limiting environmental exposures. For both, one complicating issue is that clinical manifestation is observed sometime after exposure. For birth defects and developmental disorders, many key exposures occur early in the first trimester, with anomalies found later in gestation, shortly after birth or, sometimes in the cases of behavioral phenotypes, years later. Latent periods in cancer are longer, and decades may pass before disease is detected. Nevertheless, epidemiological and population studies in cancer have been highly successful (Clapp, Jacobs, & Loechler, 2008; Doll, Peto, Wheatley, Gray, & Sutherland, 1994), and they offer hope that similar studies will illuminate birth defect etiology. Many carcinogens were identified via cohort studies of occupational exposure and lifestyle factors (Irigaray et al., 2007; Ward et al., 2003). As discussed above, this is also true of some known human teratogens, but fewer have been discovered to date.

Taken together, successes in studying gene-environment interactions in infectious diseases and cancer offer some lessons for birth defects. In particular, identification of individuals or populations with severe outcomes to specific environmental exposures can be extremely informative. Additionally, identification of genetic variants that clearly predispose to specific environmental insults will lead to recommendations for behavior modification and risk avoidance.

5. Model systems for the study of gene-environment interactions in birth defects

Combined genetic and epidemiological analyses have broken important ground in methodology and understanding of gene-environment interactions in human birth defects (for example, see (Finnell et al., 2021; Tai et al., 2015)). However, these studies have been limited in what they can conclude about mechanistic interactions between candidate genes and non-genetic risk factors. Consequently, use of model organisms has been important to explore concepts underlying gene-environment interactions, as well as identifying new potential risk factors (Krauss & Hong, 2016). Both invertebrate and vertebrate models have shed light on important questions and have therefore helped in design of human studies.

5.1. Invertebrate models

Invertebrate model organisms such as the roundworm Caenorhabditis elegans and the fruit fly Drosophila melanogaster offer some major advantages for any studies of developmental biology. Among these are short generation time, sophisticated genetic tools, very well-characterized developmental stages, and the ability to screen large numbers of organisms, thereby permitting genome-wide and forward genetic screens. These species have been used successfully to study a wide-range of issues, such as sensitivity to hypoxia, response to ethanol, and a variety of complex phenomena, including trans-generational inheritance and life span (Bellier, Chen, Kao, Cinar, & Aroian, 2009; Gao et al., 2018; Hong, Choi, & Lee, 2008; Jha et al., 2016; Mabon, Scott, & Crowder, 2009; Morozova, Shankar, MacPherson, Mackay, & Anholt, 2022; Ojelade et al., 2015; Weinhouse, Truong, Meyer, & Allard, 2018). They have also been used to model difficult, multifactorial medical conditions like Parkinson disease (Kim, Perentis, Caldwell, & Caldwell, 2018; Sarkar & Feany, 2021).

Invertebrate body plans are obviously different in important ways from humans. Nevertheless, fundamental signaling pathways that drive embryonic patterning are evolutionarily conserved, and they are often perturbed in human developmental disorders. Among these are the Notch, WNT, HH, BMP, and receptor tyrosine kinase pathways. While it is not possible to use invertebrates to study the specific structural patterning defects seen in common human birth defects, the affected pathways can be genetically perturbed in worms and flies to study mechanisms (Dorsett & Krantz, 2009; Link & Bellen, 2020). Furthermore, genetic variants identified in patients can be verified for loss- or gain-of-function in these organisms, taking advantage of the wealth of knowledge about their development (Humphries, Narang, & Mlodzik, 2020; Marcogliese et al., 2022). Some teratogens directly target components of such pathways, and it may be possible to combine these systems to study interactions between specific gene variants and non-genetic insults within animal models amenable to high throughput analyses.

5.2. Zebrafish

The zebrafish is an attractive model organism for studying birth defects. They are vertebrates, with organ systems closer to those of humans than are those of worms or flies. Furthermore, zebrafish embryos are transparent and develop rapidly and externally, allowing live imaging approaches. They have a long history of forward genetic screens for developmental phenotypes (Driever et al., 1996; Haffter et al., 1996), and are now also an efficient model organism for chemical screens (Wiley, Redfield, & Zon, 2017). These properties make them accessible to analyses of gene-environment interactions.

Zebrafish have been particularly useful in addressing genetic susceptibility to teratogenesis associated with prenatal alcohol exposure (PAE). This topic is covered by another chapter in this volume, so we will highlight only a few significant studies here. First, Eberhart and colleagues identified an interaction between pdgfra mutation and PAE in development of craniofacial defects (McCarthy et al., 2013). A small genome-wide association study provided evidence that this interaction may be of significance to specific individuals with fetal alcohol spectrum disorders (McCarthy et al., 2013). Zebrafish mutants in several other genes associated with craniofacial defects in humans are also sensitized to ethanol-induced phenotypes (McCarthy et al., 2013; Yoon et al., 2022). Most significantly, Swartz et al. took advantage of the tractability of zebrafish to perform the first vertebrate forward genetic screen to identify new mutants with enhanced susceptibility to ethanol teratogenesis (Swartz, Lovely, McCarthy, Kuka, & Eberhart, 2020). This and similar follow-up studies may permit a much more complete understanding of the genetic basis of ethanol’s actions as a teratogen, yielding insight into a major clinical and social health issue.

Only a few other gene-environment interaction screens have been performed in zebrafish to date. Ding et al. screened for genetic modifiers of doxorubicin-induced cardiomyopathy (CM) (Ding et al., 2017). Four modifiers were identified, three of which had been linked to CM in humans or mice. The fourth, dnajb6b(L), led to discovery of human CM-associated variants in the orthologous gene, DNAJB6 (Ding et al., 2017). These results offer proof of principle that such studies are both practicable and likely to illuminate disorders with complex etiology.

Zebrafish have been used more frequently to address specific candidate gene-environment interactions. Long QT syndrome (LQTS, a heart ventricle disorder that can lead to fatal arrhythmias) is associated with heterozygous mutations in KCNH2 (encoding a potassium channel). Zebrafish with a heterozygous mutation in the orthologous kcnh2 gene were sensitized to treatment with the potassium channel blocking drug, terfenadine; this mimics observations in humans (Arnaout et al., 2007). Everson et al. demonstrated that piperonyl butoxide (PBO), a component of commonly used pesticides that also inhibits HH pathway signaling, induced craniofacial defects in heterozygous Sonic HH (shha) mutants (Everson, Batchu, & Eberhart, 2020). Finally, 2,3,7,8-tetrachlorodibenzo-p-dioxin (TCDD) impairs craniofacial development in zebrafish and other species and heterozygosity for the pro-chondrogenic transcription factor, sox9b, sensitized to fish to TCDD-induced jaw malformations (Xiong, Peterson, & Heideman, 2008).

5.3. Mice

Mice have been the animal model of choice for most studies on gene-environment interactions in birth defects. They are genetically tractable, including the ability to knock out genes in a cell type-specific and temporal manner. As mammals, they are anatomically closer to humans than are zebrafish. Forward genetic screens are possible, but challenging to the point that we are unaware of any that involved inclusion of a teratogen. Nevertheless, studies with candidate genes and teratogens have been highly illuminating, discovering mechanisms of gene-environment interaction and identifying potential non-genetic risk factors for further study in humans. Here, we briefly focus on two birth defects not covered elsewhere in this volume, HPE and congenital heart defects (CHD).

5.3.1. Gene-environment interactions in mouse models of HPE

HPE has served as a model birth defect for exploring gene-environment interactions in mice (Grinblat & Lipinski, 2019; Hong & Krauss, 2018; Lo, Hong, & Krauss, 2021). It occurs at very high frequency, conservatively estimated at 1 in 250 conceptions, with >97% succumbing in utero (Matsunaga & Shiota, 1977; Shiota, 2021). HPE is caused by a failure to define the midline of the forebrain and/or midface and characterized by a broad spectrum of phenotypic severity (Grinblat & Lipinski, 2019; Lo et al., 2021). HPE is associated with heterozygous mutations in the HH, Nodal, and FGF pathways that display incomplete penetrance and variable expressivity; in the great majority of cases, these mutations are considered necessary but not sufficient to produce the associated clinical phenotypes (Lo et al., 2021). Additional “hits” are therefore invoked, and these may be genetic, environmental, or both. In 1996, SHH mutations were first identified in HPE patients, and it is the most frequently mutated gene in HPE (Roessler et al., 1996; Roessler, Hu, & Muenke, 2018). Shh knockout mice were also first reported in 1996, and they have severe HPE (Chiang et al., 1996). Subsequently, cyclopamine, an HPE-inducing teratogen, was shown to block HH signaling by directly binding and inhibiting Smoothened (SMO), a key HH signal transducer (Chen, Taipale, Cooper, & Beachy, 2002; Cooper, Porter, Young, & Beachy, 1998). Cyclopamine therefore became a prototype for clinically useful SMO inhibitors, such as Vismodegib, which also induces HPE in mice (Heyne et al., 2015). Taken together, disruption of HH signaling by either genetic or teratogenic mechanisms can produce HPE, and such risk factors may interact, helping explain the complex etiology of human HPE.

This notion has received support in mouse studies of HPE, which offer a set of guideposts on how gene-environment interactions may be systematically pursued. A good initial step in pursuing gene-environment interactions in mice is to assess whether partially penetrant mutations in a given signaling pathway can synergize with subthreshold doses of chemical inhibitors of the same pathway, to produce a specific developmental defect. Gli2 encodes a HH pathway-responsive transcription factor. Heyne et al. showed that mice heterozygous for Gli2, which resemble wild type animals, were sensitized to HPE induced by ineffective, or weakly effective, doses of Vismodegib (Heyne et al., 2016). Therefore, genetic and non-genetic insults each targeting the HH pathway, but insufficient on their own, synergized to produce HPE. These types of observations validate mouse models for gene-environment studies.

A logical next step is to test environmental risk factors implicated in HPE but for which a clear teratogenic mechanism may be lacking (as opposed to Vismodegib, whose teratogenic mechanism is clear). A good example here is PAE, which is associated with HPE in some but not all epidemiological studies (see Lo et al., 2021). Cdon encodes a HH coreceptor mutated in some HPE patients (Bae et al., 2011). When placed on a genetically resistant background (129S6 mice), homozygous mutation of Cdon has very mild phenotypic effects due to redundancy with other HH coreceptors; PAE is also non-teratogenic to these mice (Hong & Krauss, 2012). A combination of Cdon mutation and transient in utero ethanol exposure, however, resulted in a complete spectrum of HPE phenotypes in ~75% of mice (Hong & Krauss, 2012). Similarly, heterozygosity for Shh or Gli2 on a C57BL/6J genetic background also sensitized mice to PAE-induced HPE (Kietzman, Everson, Sulik, & Lipinski, 2014). Therefore, PAE synergizes with subthreshold genetic deficits in HH signaling to induce HPE. It may be that some of the inconsistent results in epidemiological studies arose because PAE requires a predisposing genetic background to exert its ability to induce HPE.

These types of studies with mice also allow for mechanistic insight. Gestational windows of sensitivity to teratogens can be identified, shedding light on key developmental processes occurring during peak sensitivity. PAE is most effective when delivered at E7.0, whereas direct SMO inhibitors are most effective at E7.5 (Heyne et al., 2015; Hong, Christ, Christa, Willnow, & Krauss, 2020). These and additional data argue that PAE exerts its HPE-inducing effects upstream of HH signaling, inhibiting HH as an indirect outcome of Nodal pathway inhibition, which is a proximal, perhaps direct, target of PAE (Hong et al., 2020). PAE inhibits HH signaling indirectly in zebrafish also (Sidik et al., 2021). When PAE in mice occurs later, at E9.0, HH signaling is disrupted as well. This may also occur indirectly by affecting expression of factors that regulate primary cilia, organelles which are the subcellular sites of HH signal transduction (Boschen, Fish, & Parnell, 2021).

Finally, some well-studied compounds with known pharmacological targets outside the HH pathway, were recently discovered to target HH signaling as well. Such compounds are therefore potential risks for HPE teratogenicity. Among these are PBO and Δ9-tetrahydrocannabinol (THC), which were shown to inhibit SMO many years after discovery of their primary targets (Khaliullina, Bilgin, Sampaio, Shevchenko, & Eaton, 2015; Wang et al., 2012). PBO and THC were then each demonstrated to synergize with HH pathway mutations in mice to produce HPE (Everson et al., 2019; Lo et al., 2021). Such studies in mice have potential public health value, as they suggest PBO and THC should be further assessed as potential risk factors for human HPE.

5.3.2. Gene-environment interactions in mouse models of CHD

Like HPE, most cases of CHD (of which there are many subtypes) have a complex etiology thought to be best explained by gene-gene and/or gene-environment interactions (Kalisch-Smith, Ved, & Sparrow, 2020; Kodo, Uchida, & Yamagishi, 2021; Majumdar, Yasuhara, & Garg, 2021). Variants in many genes are associated with CHD, and many non-genetic risk factors have also been implicated. Studies with mice have been instrumental in identifying how these classes of risk factors may interact. As one example, we discuss gestational hypoxia, an environmental stressor that promotes multiple types of birth defects.

Maternal exposure to hypoxia is a potential risk factor for CHD (Hutter, Kingdom, & Jaeggi, 2010; Kalisch-Smith et al., 2020). Mice carrying several mutations found in human CHD, including heterozygosity for Nkx2-5, Tbx1, Tbx5, and Notch1, are sensitized to CHD induced by short-term gestational hypoxia (Chapman et al., 2020; Moreau et al., 2019). Percentages of mouse embryos with CHD correlate with degree of hypoxia in these scenarios, suggesting that low oxygen may be one cause of variable penetrance and expressivity in human CHD. Additional work has provided insight into mechanisms underlying mutation + hypoxia-induced phenotypes. Gestational hypoxia led to a prolonged hypoxia inducible factor 1α (HIF1α) response in the developing myocardium, which in turn depressed levels of Nkx2-5 expression (Moreau et al., 2019). Furthermore, hypoxia rapidly induced an unfolded protein response (UPR) in cardiac progenitors, which reduced levels of FGFR1, a key signaling receptor for cardiac development (Shi et al., 2016). Therefore, gestational hypoxia may exert its effects via multiple mechanisms that are capable of cooperating with predisposing genetic variants. Unsurprisingly, therefore, mechanisms underlying the interaction between even a single mutation and single environmental insult may be multiple and complex.

An important question raised by these mouse studies is the identity of potential clinical conditions or teratogens that may cause gestational hypoxia. Mice are again a useful system for assessing candidate factors. Chapman et al. found that the anti-arrhythmic drug dofetilde, known to cause hypoxia in rodent embryos, induced a higher incidence of CHD in Notch1 heterozygotes (Chapman et al., 2020). Anti-arrhythmic drugs that cause brachycardia, leading to hypoxia, may therefore be contraindicated during pregnancy. It is critical, of course, to consider the relative doses required for a therapeutic effect vs. a teratogenic effect.

Another possible cause of hypoxia could be maternal anemia, which is a worldwide health problem. To test the possibility that anemia might be a teratogenic cause of CHD, Kalisch-Smith et al. assessed the effects of maternal iron deficiency in mice. Mature female mice maintained on a low iron diet were anemic; when mated, embryos of low-iron diet females developed multiple types of CHD and displayed embryonic lethality (Kalisch-Smith et al., 2021). The effects of an iron-deficient diet were linked to increased retinoic acid (RA) signaling (Kalisch-Smith et al., 2021); RA is a known cardiac teratogen (D’Aniello & Waxman, 2015; Kalisch-Smith et al., 2020). Surprisingly, short-term gestational hypoxia leads to reduced RA signaling (Walton et al., 2018). Therefore, iron deficiency is a condition that may exert teratogenic effects independent of its ability to produce hypoxia. These results show that, even in the absence of a cooperating genetic variant, mechanisms underlying teratogenesis may be complex, requiring rigorous analysis at the level of signaling pathways and transcriptomics.

5.4. Human systems

Animal models have been critical to addressing questions of gene-environment interactions in birth defect etiology, and they will remain so for the foreseeable future. Nevertheless, there are some significant differences in human vs. e.g., mouse development (Rayon et al., 2020; Shahbazi & Zernicka-Goetz, 2018). Furthermore, the high number of combinatory genome-by-exposome permutations makes screening a challenge, even for the most tractable model organisms (Beames & Lipinski, 2020). The ability to generate complex organoids from human pluripotent cells is likely to help bridge this gap. Use of organoid systems is increasingly common, and it is possible to perform genome-wide CRISPR screens with them (Kampmann, 2020). Furthermore, it is possible to generate organoids from patient-derived cells. Nevertheless, it is still early days with this technology, and very few gene-environment interaction studies have been performed. In one interesting example, Modaferri et al. used CRISPR to generate an iPSC line with a heterozygous mutation in CHD8, similar to a mutation found in some autism patients (Modafferi et al., 2021). Brain organoids derived from CHD8^+/+ and CHD8^+/− lines were compared for their response to Chlorpyrifos, a pesticide with adverse effects on brain development, including pathways affected by autism (Grandjean & Landrigan, 2014). The CHD8^+/− line was sensitized to several negative effects of chlorpyrifos, revealing an interaction between a defined genetic predisposition and a putative environmental risk factor.

It is possible to imagine that organoids of various types (e.g., brain, heart, palate) could be used for automated, high-throughput assays involving genome-wide CRISPR screens and multiple teratogens for a given organ system. iPSC-derived differentiated cells and organoids tend to be “immature” (i.e., at embryonic or fetal stages of development), something that is a hindrance for studies of adult-onset diseases. In contrast, this may be a benefit for studies of birth defects and developmental disorders. An important, but surmountable, caveat with use of organoids for such studies is that protocols are often not fully standardized, and there is much lab-to-lab variability. It is important that attention be paid to this issue in general, but it will be critical for complex studies assessing multiple variables, so as to ensure reproducibility. Once standardized, it will be possible to design studies that take advantage of the enormous amount of information available on genetic variants in birth defects, combined with relevant exposures to environmental conditions.

6. Future directions: Potential mechanisms of gene-environment interaction

Understanding potential mechanisms of gene-environment interaction is important as it may open prevention strategies. Among the likely molecular targets are critical morphogenetic signaling pathways and their associated gene regulatory networks, and cellular structural elements (e.g., components and regulators of cell polarity and the cytoskeleton) (Anvarian, Mykytyn, Mukhopadhyay, Pedersen, & Christensen, 2019; Krauss & Hong, 2016; Nikolopoulou, Galea, Rolo, Greene, & Copp, 2017). Myriad possible mechanisms exist and these have recently been reviewed (Krauss & Hong, 2016) and are presented in subsequent chapters in this volume. One area of likely importance in this realm is epigenetic regulation of gene expression.

6.1. Epigenetics and DNA methylation

Epigenetic mechanisms that interface the environment and the genome have emerged as a relatively underexplored but promising domain of birth defects research. DNA methylation is a particularly intriguing focus for investigation into etiologically complex birth defects as this epigenetic mechanism is inherently sensitive to environmental cues and a practical target for prevention and therapeutic strategies. DNA methylation is influenced by diverse environmental influences including maternal diet and stress, and exposure to drugs, toxins, and environmental pollutants (Breton-Larrivee, Elder, & McGraw, 2019; Lee, 2015; Martin & Fry, 2018) Several lines of investigation suggest that DNA methylation may influence birth defect risk. For example, periconceptional intake of the dietary methyl donor, folic acid, is well established to reduce the risk for neural tube defects and appears to lessen the risk for nonsyndromic orofacial clefts as well (Millacura, Pardo, Cifuentes, & Suazo, 2017; Zhou et al., 2020). Studies have identified differential methylation profiles in tissue from individuals with and without clefts (Gonseth et al., 2019; Howe et al., 2019; Xu, Lie, Wilcox, Saugstad, & Taylor, 2019; Young, Slifer, Hecht, & Blanton, 2021), and differential DNA methylation has been suggested to underlie discordance among monozygotic twins for orofacial clefts and several forms of congenital malformations (Lyu et al., 2018; Weksberg et al., 2002; Young et al., 2021; Zhang et al., 2020). Animal studies have shown that DNA methylation is required for embryonic development and that environmentally mediated DNA methylation changes can influence development and drive phenotypic changes, even among littermates (Dolinoy, Huang, & Jirtle, 2007). Additional animal studies have demonstrated that pharmacologic inhibition of DNA methyltransferase activity can cause structural birth defects, including orofacial clefts (Branch, Chernoff, Brownie, & Francis, 1999; Branch, Francis, Brownie, & Chernoff, 1996; Bulut, Ozdemir, Başimoglu-Koca, Korkmaz, & Atalay, 1999; Rogers et al., 1994).

While these findings provide evidence linking epigenetic modification to birth defect risk, substantial knowledge gaps remain. A fundamental question not yet fully resolved is which developmental processes are most sensitive to epigenetic influences. For individual birth defect outcomes, it will be critical to determine pathogenic epigenetic events, and specific environmental triggers. Most studies have been limited to analysis of blood or other surrogate tissue, and whether DNA methylation changes in these tissues reflect pathogenic changes remains unclear. A foundation for filling these knowledge gaps is being built through initiatives like the Toxicant Exposures and Responses by Genomic and Epigenomic Regulators of Transcription (TaRGET) program funded by the National Institute of Environmental Health Sciences, as well as independent efforts to develop epigenetic-focused cell based assays that could be leveraged for high content chemical screening (Stelzer, Shivalila, Soldner, Markoulaki, & Jaenisch, 2015; Wang et al., 2018, 2021).

7. Concluding remarks

In this article we have described challenges to understanding the complex etiology of commonly occurring, non-syndromic birth defects. While identifying the specific combinations of genetic and environmental insults that underlie individual cases remains difficult, there is reason to be hopeful about opportunities for progress. Lessons can be drawn by comparison with knowledge of other diseases with multifactorial causation. Multiple animal and in vitro models for birth defects have been successfully generated and display ever-increasing accuracy. Together, these approaches offer promise for identification of new risk factors and fundamental mechanisms. Such discoveries are vital to design of epidemiological studies and to development of preventive and risk management strategies.

Abbreviations

CHD

congenital heart defects

CM

cardiomyopathy

HPE

holoprosencephaly

MSMD

Mendelian susceptibility to mycobacterial disease

NGS

next generation sequencing

PAE

prenatal alcohol exposure

RA

retinoic acid

TPS

tumor predisposition syndrome