The NBDPS was a population-based case–control study of risk factors for birth defects conducted by 10 Centers for Birth Defects Research and Prevention (CBDRP; Reefhuis et al., 2015; Yoon et al., 2001). The study included women with pregnancies affected by one of more than 30 major structural birth defects and mothers of liveborn children without major structural birth defects. A woman was eligible for inclusion if her pregnancy ended on or after October 1, 1997, and if her estimated delivery date was on or before December 31, 2011. Cases were ascertained from existing birth defect surveillance systems in each of 10 states (Arkansas, California, Georgia, Iowa, Massachusetts, New Jersey, New York, North Carolina, Texas, or Utah) and could be liveborn, stillborn, or pregnancy terminations. Liveborn control children were ascertained from vital records or birth hospital logs in the same geographic areas as cases. Clinical geneticists reviewed medical records using standardized case definitions to determine eligibility (Rasmussen et al., 2003). Cases with known syndromes, chromosomal or single-gene disorders were excluded.

After obtaining verbal consent, computer-assisted telephone interviews were conducted in English or Spanish with over 40,000 eligible women between 6 weeks and 24 months after their estimated dates of delivery. During the maternal interview, data were collected on pregnancy history, family history, and sociodemographic factors and on the following exposures from 1 year before conception through the end of pregnancy: maternal chronic diseases; infections; fever; use of medications, illicit drugs, tobacco, caffeine, alcohol, and vitamins; diet and nutrition; weight; physical activity; injuries; stress; occupations; drinking water use; occupational and environmental exposures; and other factors.

Following the completed interview, each woman was mailed cytobrushes (two per participant) to collect buccal cell specimens from herself, her child (if living), and her child’s biological father (if available). DNA specimens from more than 23,000 families were collected to evaluate genetic risk factors and to examine the role that genetic variants might play in modifying risks associated with environmental factors (Rasmussen et al., 2002; Reefhuis et al., 2015). Between 1997 and 2003, cytobrushes were mailed back to Centers in closed plastic tubes that did not allow air drying (hereafter referred to as “wet brushes” [Cyto-Pak Cytosoft Brushes CP-5B, Medical Packaging Corporation, Camarillo, CA]). In 2003, materials and procedures were modified. Specifically, after sampling, cytobrushes were instead packaged in open paper-backed peel pouches (hereafter referred to as “dry brushes” [Cytology Brush Pack CYB-1, Medical Packaging Corporation]; Gallagher et al., 2011). This new packaging mitigated bacterial and fungal growth by allowing specimens to dry during transport. Among the cytobrushes returned during the course of the study, 22% were wet brushes and 78% were dry brushes.

Institutional review boards at the Centers for Disease Control and Prevention (CDC) and each CBDRP approved the NBDPS protocol.

With the rapid advances in technologies used to perform high-throughput DNA sequencing and the accompanying reduction in sequencing costs, NBDPS investigators convened a working group in 2014 to determine the best ways to leverage NBDPS data and specimens with sequencing technologies so as to improve the efficiency and likelihood of discovery of genetic risk factors and gene–environment interaction effects for birth defects. The working group was tasked with making recommendations about the technologies and study designs and how to prioritize birth defects for sequencing.

The collaborative working group recommended exome sequencing (ES) instead of whole genome sequencing (WGS), because although both ES and WGS have comparable sensitivity to detect single nucleotide variants in coding regions, the substantially lower cost of ES at present allows more specimens to be sequenced and hence, increases statistical power for analyses. Full trios (mother, father, and child) were prioritized, a study design that enables identification of de novo variants causing disease (Chesi et al., 2013; Hunt et al., 2014). The trio study design also reduces the impact of population admixture in the assessment of inherited variants and increases flexibility of analyses, allowing investigators to conduct hybrid log-linear analyses (Weinberg, Wilcox, & Lie, 1998); transmission disequilibrium tests (Spielman, McGinnis, & Ewens, 1993); analyses of maternal genetic effects, maternal-fetal gene interaction effects, and parent-of-origin effects (Ainsworth, Unwin, Jamison, & Cordell, 2011); analyses of gene–environment interaction effects (Tai et al., 2015; Umbach & Weinberg, 2000); and other study designs. A case-only study design was chosen to maximize the number of specimens that could be included with limited funds and because use of publicly available controls and unaffected parents as controls were options for association testing. Family trios reporting use of egg, embryo, or sperm donors were excluded.

By design, NBDPS included some very rare birth defects, which were prioritized for the ES projects for several reasons. First, the specimens had not been used previously as power was low for candidate gene-based association studies. Thus, DNA in quantities sufficient for ES and follow-up studies were available. Second, methods used to analyze these trios would be similar to those used to analyze families with Mendelian traits, and ES has been used successfully to uncover the molecular basis of multiple Mendelian disorders (Bamshad et al., 2011). Third, this provided an opportunity to have an impact on families with less common phenotypes.

Birth defects that were prioritized for this initial round of ES included the following: colonic atresia or stenosis, anterior segment dysgenesis eye defects, primary congenital glaucoma, transverse limb reduction defects, split hand/foot malformation, cloacal exstrophy, bladder exstrophy, anophthalmos or microphthalmos, sacral agenesis, biliary atresia, tricuspid atresia, Ebstein anomaly, hypoplastic left heart syndrome, and heterotaxy (Table 1). Each case child’s medical records were reviewed by clinical geneticists and classified as having (a) one major structural birth defect or a complex sequence of birth defects that occurs in a well-defined pattern (isolated) or (b) more than one major unrelated birth defect in separate organ systems or a complex sequence occurring in a well-defined pattern with one or more unassociated major birth defects (multiple). A clinician trained in pediatric cardiology also classified each congenital heart defect (CHD) as “simple” (a well-defined specific defect), “associations” (a common, uncomplicated combination of two or three cardiac defects that often occur together), or “complex” (a combination or pattern of independent defects in multiple cardiac structures; Botto, Lin, Riehle-Colarusso, Malik, & Correa, 2007; Rasmussen et al., 2003). Each child with a CHD was classified based on both cardiac and noncardiac defects. The detailed classification protocol used by NBDPS clinicians allows for stratified analyses well-suited to identify potential causal variants (i.e., precise phenotypes reduces etiologic heterogeneity and allows assessment of children by defect severity). Results from children with more than one of the included birth defects will be analyzed with each case group separately.

As there is currently no “standard” ES protocol, and NBDPS specimens were collected up to two decades ago using less than optimal methods, we first sought to optimize the method for sequencing exomes from self-collected cytobrush-derived DNA specimens. This method optimization would allow the development of a workflow that could be used across phenotypes, thereby ensuring consistency. Pilot studies were conducted using 9 wet brush-derived and 51 dry brush-derived DNA specimens from 20 children with rare intestinal anomalies, colonic atresia or stenosis, and their parents (60 specimens total). The 20 case children were ascertained from surveillance systems in 8 of 10 Centers (New Jersey and New York were not represented).

DNA extraction from wet brushes was completed at each Center’s laboratory using several protocols (Rasmussen et al., 2002). DNA extracted from one wet brush per participant was mailed to the NBDPS biorepository; DNA extracted from the second wet brush per participant was retained at the Center’s laboratory. DNA extraction from dry brushes was modified such that the two brushes collected per participant were processed separately (Gallagher et al., 2011). DNA extraction from one dry brush per participant continued to be completed at each Center’s laboratory using several protocols and was retained locally. DNA extraction from the second dry brush per participant was completed at a CDC laboratory using Puregene (Gentra, Germantown, MD) or a phenol chloroform method (Garcia-Closas et al., 2001), modified by separating the aqueous and organic phases using Phase Lock Gel barrier tubes according to the manufacturer’s instructions (Fisher Scientific, Pittsburgh, PA); the resulting DNA was mailed to the NBDPS bio-repository. DNA specimens included in the ES projects were stored at the NBDPS biorepository (1997–2011: CDC and ATSDR Specimen Packaging, Inventory and Repository in Lawrenceville, GA; 2011-present: Fisher BioServices, Rockville, MD).

DNA specimens were shipped from the NBDPS bio-repository to a National Human Genome Research Institute laboratory (Dr. Lawrence Brody, Bethesda, MD) for processing and initial quality control (QC). Modest DNA yields prompted an evaluation of whole genome amplification (WGA) of these archived specimens and use of WGA product as input for ES. Genomic DNA (gDNA, 50 ng) from all 60 specimens was shipped to QIAGEN (Germantown, MD) for WGA using REPLI-g^® WGA reagents. The QIAGEN lab reported that 56% (5 of 9) of the wet brush specimens and 100% (51 of 51) of the dry brush specimens were successfully amplified.

Fifty-six of 60 specimens passed QC at QIAGEN and were sequenced at the National Institutes of Health (NIH) Intramural Sequencing Center (NISC, Rockville, MD; https://www.nisc.nih.gov/). Briefly, libraries were prepared from 1 μg of each WGA product using the KAPA DNA Library Preparation Kit (KAPA Biosystems, Boston, MA). A 25-base pair (bp) sequencing run was completed to assess unique alignment to the human reference sequence. Four of the five wet brush specimens that passed initial QC yielded 40% or fewer reads that aligned uniquely to the human reference sequence (range 0.5–39.3%). These results suggested that these specimens were contaminated with high levels of nonhuman DNA. Among the 51 dry brush specimens, all WGA products sequenced were “usable” but had limitations, such as low library diversity (a measure of unique reads), high mismatch rate in comparison to the human reference, or poor concordance based on mapping the paired reads from the ends of the sequenced WGA fragment to the reference genome (i.e., the mapped distance between the two reads was not within the expectations for the size range of the fragment).

Exomes were captured using SeqCap EZ Human Exome + UTR kit v3.0 (Roche NimbleGen, Madison, WI), and sequencing was completed on Illumina’s HiSeq 2500 system (Illumina, San Diego, CA), which covered 96 Mb and employed 126-bp paired-end read sequencing. Image analysis and base calling were performed using Illumina Genome Analyzer Pipeline software (version 1.18.64.0) with default parameters.

Sequence reads were mapped to hg19 using Illumina’s efficient large-scale alignment of nucleotide databases for machine QC and were realigned using NovoAlign. Binary alignment map (BAM) files were generated. Genotypes were called using a probabilistic Bayesian algorithm, most probable genotype (MPG; Teer et al., 2010), and annotated using ANNOVAR (Wang, Li, & Hakonarson, 2010). An MPG score ≥10 for at least 85% of the target exome sequence was considered minimum coverage (an MPG score of 10 corresponds roughly with 10× to 20× depth of coverage), and bases with PHRED quality scores <20 were excluded. These metrics were applied only to evaluate unamplified gDNA versus WGA product.

To determine the quality of ES data from WGA products, 1 μg of unamplified gDNA from 10 of the 56 specimens was sequenced, and each pair of results derived from the same specimen was compared (Table 2). Overall, dry-brush-derived DNA specimens yielded higher quality DNA and performed better than wet-brush-derived DNA specimens. Additionally, ES quality metrics were better when input was native gDNA versus WGA product. Although some WGA products performed reasonably well when compared to unamplified DNA (specimens 7–10), there was no correlation between gDNA yield and WGA product performance; however, the percent of precapture reads that aligned to human reference sequence correlated with mean depth of coverage and MPG scores. Based on these results, the decision was made to move forward using unamplified gDNA from dry brush specimens as input for ES.

Given the low quantity of gDNA present in these specimens, a low input library preparation protocol (Accel-NGS^® 2S Plus DNA Library Kit, Swift BioSciences, Ann Arbor, MI) was evaluated. This method was applied to 100 ng gDNA from six dry-brush-derived DNA specimens of children with colonic atresia or their parents and six replicates of HapMap control specimen NA12878 (Coriell Institute, Camden, NJ). All other steps were performed as described above. The percent of on-target bases with MPG scores ≥10 ranged from 95.9 to 96.4% for high-quality control replicates and 93.7 to 96.7% for NBDPS specimens. Total variants observed compared to the reference genome ranged from 86,243 to 87,583 for controls and 83,269 to 89,264 for NBDPS specimens. Average depth of coverage in the target region ranged from 58.2× to 67.0× for controls and 38.1× to 65.7× for NBDPS specimens. Based on the similarity of results from high-quality control replicates and NBDPS specimens using the low input protocol, as well as their similarity to results from 10 unamplified gDNA specimens sequenced using the standard library preparation protocol (requiring 1 μg DNA), all specimens derived from NBDPS dry brush trios and included in one of the ES projects were sequenced using unamplified gDNA and a low input library preparation protocol.

Following method optimization, we used specimens from children with colonic atresia and their parents to complete a proof-of-principle study to demonstrate the feasibility of using the optimized methods for variant discovery and analysis.

The prevalence of colonic atresia is estimated at 1 in 20,000 (Mirza, Iqbal, & Ijaz, 2012) to 1 in 66,000 live births (Davenport, Bianchi, Doig, & Gough, 1990). Approximately one-half of children with colonic atresia have additional defects (El-Asmar, Abdel-Latif, El-Kassaby, Soliman, & El-Behery, 2016; Etensel et al., 2005). If isolated colonic atresia is identified and treated early, the prognosis is good with an overall mortality of <26%; however, mortality increases considerably if there is a delay in treatment beyond 3 days (Etensel et al., 2005). Colonic atresia cases are classified according to the continuity of the bowel and mesentery (Bland-Sutton, 1889) into three types (Types I, II, and III, in order of severity from least to most severe) and a catch-all category of “not otherwise specified.” It is estimated that 75% of colonic atresias are proximal to the splenic flexure in the ascending colon (Winters, Weinberger, & Hatch, 1992). Furthermore, most cases have Type III (Powell & Raffensperger, 1982), characterized by a “V” shape in the mesentery with proximal and distal blind ends, which can have more postoperative repair complications compared to the other types of colonic atresia.

The etiology of colonic atresia is unknown. The prevailing theory involves a vascular insult (accident) to the mesenteric vessels during development (Louw & Barnard, 1955). Another theory is genetic; there is one report of a familial occurrence of colonic atresia with three isolated cases among first-degree relatives (Benawra, Puppala, Mangurten, Booth, & Bassuk, 1981). To our knowledge, there are no published genetic studies that include isolated human colonic atresia cases; however, mouse models suggest three putative candidate genes for human colonic atresia: fibro-blast growth factor 10 (Fgf10) and its receptor (Fgfr; Fairbanks et al., 2005) and Cdx2 (Gao, White, & Kaestner, 2009). Most reports of colonic atresia are case/surgical reports so environmental exposure and pedigree information are sparse, although several studies have reported cases of colonic atresia associated with intrauterine varicella infection (Alexander, 1979; Hitchcock, Birthistle, Carrington, Calvert, & Holmes, 1995; Sauve & Leung, 2003). Using NBDPS data, several other exposures have been associated with colonic atresia, including periconceptional cold/flu with fever (Waller et al., 2018) and periconceptional genitourinary infections (Howley et al., 2018).

Following the methods outlined in the Appendix, BAM files from nine trios (nine probands and their unaffected parents) with sufficient dry-brush-derived DNA that were successfully sequenced at NISC were transferred to the University of Washington Center for Mendelian Genomics (UW-CMG) for reprocessing and analysis. Freemix was used to estimate contamination levels of specimens (Jun et al., 2012). One trio was removed for evidence of contamination (i.e., a proband with freemix estimate >6%), resulting in eight useable trios. Using GEMINI v0.20.2 (Paila, Chapman, Kirchner, & Quinlan, 2013), variant filters included a depth ≥ 6, genotype quality ≥20, allele frequency ≤ 0.005 across populations represented in reference databases (1,000 Genomes phase 3 [1000 Genomes Project Consortium et al., 2015], Exome Aggregation Consortium [ExAC v0.3; Lek et al., 2016], Exome Sequencing Project 6,200 [v2; NHLBI Exome Sequencing Project, n.d.], or the UK10K February 15, 2016 release ([UK10K Consortium et al., 2015]), and a predicted medium to high impact severity on the gene/protein (GEMINI, 2017). These filters yielded 156,716 total single nucleotide polymorphisms (21,570 singletons) with a 2.4 transition/transversion (Ti/Tv) ratio and 20,374 indels (3,921 singletons) for the eight full trios. Summary statistics for each individual are provided in Table 3.

Given that little is known about genetic risk for colonic atresia, two independent analyses were conducted. The first analysis considered all Mendelian inheritance models, except for autosomal dominant. After variant filtration, no gene had prioritized variants from more than one family, and the following plausible modes of inheritance were represented: de novo, compound heterozygous, X-linked de novo, and X-linked recessive. After excluding variants with evidence of mismapping or sequencing errors, 31 genes were identified with possible pathogenic variants for the eight trios and are shown in the Supporting Information Table S1 by modes of inheritance. Each variant was visualized in Integrative Genomics Viewer (Robinson et al., 2011) and interrogated for predicted functional impact, conservation of alignment position, low frequency in control populations, and biological plausibility. There were 13 predicted protein-truncating or loss/gain of function variants (e.g., splice site, stop-gain, frameshift, and in-frame indel) and an additional 3 potentially pathogenic variants (>15 PHRED-scaled Combined Annotation Dependent Depletion [CADD] score [Kircher et al., 2014]; these 16 variants are bolded in the Supporting Information Table S1).

For the next analysis, we assessed all variants in the three candidate genes implicated in colonic atresia from animal studies: FGF10, FGFR2, and CDX2. Prior to filtering for predicted pathogenic coding variants, we identified 2 variants in FGF10, 30 in FGFR2, and 6 in CDX2. All variants had PHRED-scaled CADD scores ≤15, indicating little evidence of conservation or predicted pathogenicity (data not shown). Moreover, the variants were not very rare (population frequency > 0.005) and some occurred in many parents and probands, which suggest that they are unlikely to be causal for the rare defect of colonic atresia. Further follow-up was not conducted.

Given the paucity of genetic studies on colonic atresia, the small number of trios analyzed, and that no genes had rare candidate variants across multiple families, we were unable to make definitive conclusions on the likely mode of inheritance or potential risk variants in the NBDPS probands with colonic atresia who could be studied. One variant in SCARA3 was submitted to Matchmaker Exchange (Philippakis et al., 2015) because of its high PHRED-scaled CADD score (23.8) and biological plausibility. Given the sporadic occurrence, colonic atresia could result from environmental effects, genetic effects, or a combination of genetic and environmental effects; mosaicism; or possible somatic mutations.

Following the proof-of-principle study, exomes of case children with one of nine other birth defects and their parents were selected for ES at NISC using the optimized methods (i.e., dry-brush-derived gDNA and a low input library preparation protocol). The additional case groups included the following: anterior segment dysgenesis eye defects, primary congenital glaucoma, transverse limb reduction defects, split hand/foot malformation, cloacal exstrophy, bladder exstrophy, anophthalmos or microphthalmos, sacral agenesis, and biliary atresia. UW-CMG sequenced exomes of selected CHD trios (tricuspid atresia, Ebstein anomaly, hypoplastic left heart syndrome, and heterotaxy with and without CHDs) using dry-brush-derived gDNA and a low input library preparation protocol optimized in their laboratory (ThruPLEX DNA-seq Kit, Rubicon Genomics, Ann Arbor, MI).

As in the proof-of-principle study, BAM files from each case group sequenced at NISC were transferred to UWCMG so that each case group was processed separately using the same pipeline (details in Appendix). UW-CMG used peddy [Pedersen & Quinlan, 2017] to check sex, ancestry (using Principal Components Analysis), and pedigrees/relationships and annotated the variant call format files with the ENSEMBL Variant Effect Predictor (v89; McLaren et al., 2016). A summary variant report was prepared for each case group. This report included a list of genes identified using variant filtration in GEMINI (Paila et al., 2013) under each mode of inheritance (homozygous recessive, compound heterozygous, de novo, X-linked recessive, and X-linked de novo) except autosomal dominant, in multiple families and for each family. In addition, UWCMG provided a report for each case group describing copy number variants (CNVs) identified using CoNIFER (Krumm et al., 2012). Upon project completion, these data will be shared broadly in public repositories; as examples, aggregate variant and broad phenotype data will be shared through dbGaP and Geno2MP (Chong et al., 2015); likely pathogenic and pathogenic variants through ClinVar (Landrum et al., 2014); and candidate genes with the MatchMaker Exchange (Philippakis et al., 2015) via MyGene2 (Chong et al., 2016); and the CMG website.

When enough specimens are available per defect, rare variant association testing, such as burden and kernel-based testing, will be conducted within ancestry groups. Rare variants will be validated by Sanger sequencing and potentially included in functional studies. Additionally, the rich environmental exposure data collected from NBDPS participants can be mined to assess exposures that might modify genetic effects, although small numbers limit the robustness of such an assessment for some defects. We plan to publish all results, including negative findings, as for some of these phenotypes, these might be the only current exome-sequenced cohorts with numbers large enough to conduct these analyses, making the results important to include in the peer-reviewed literature.

NBDPS sequencing projects face several challenges. First, the modest DNA amounts can be of suboptimal quality from self-collected buccal cells (Gallagher et al., 2011), potentially leading to false negative results. Second, collection and storage of specimens for up to two decades can negatively affect DNA stability and integrity (Madisen, Hoar, Holroyd, Crisp, & Hodes, 1987; Visvikis, Schlenck, & Maurice, 1998). Third, sample sizes are modest due to the rarity of the outcome under study, limiting power for analysis of gene–environment interaction effects. Other analytic challenges remain, such as distinguishing noise from signal and assessing background levels of de novo variants.

However, the NBDPS does provide a unique resource for genomic studies of birth defects with several strengths. NBDPS specimens are ethnically diverse: over 35% of participants self-identified as a race or ethnicity other than non-Hispanic White. This is in contrast to the majority of publicly available genomic reference data that are from populations of European origin (i.e., non-Hispanic White). Because of our family trio study design, we will be able to use exome data from parents of all race–ethnicities who have no family history of the specific birth defect under study as controls. Analysis of exomes of all race2013ethnicities will allow improved understanding of risk factors for these defects and possible preventive measures for all populations.

Our strategy is to focus initially on identifying highly penetrant de novo variants and other candidate variants under all possible Mendelian models of inheritance; however, many of these birth defects will likely be the result of multiple moderate effect variants working together (i.e., oligogenic model). Due to the collaborative nature of these projects, we will be able to assess potential pleiotropy across case groups. As we will have limited opportunity to identify more modest effect variants with our sample sizes, we seek opportunities to collaborate with investigators worldwide who have specimens from families with these defects. We are also poised to expand beyond our initial set of phenotypes.

The NBDPS sequencing projects exemplify collaborative efforts between Centers in multiple regions of the United States. This ensures that diverse populations are ascertained. The study is characterized by collaboration between government agencies and academia, as well as between interdisciplinary scientists (i.e., epidemiologists, geneticists, clinicians, and biostatisticians). The combination of data available from these NBDPS trios (e.g., pregnancy exposure information, genetic data, birth defects clinical information, and family histories) is a unique resource to help advance our understanding of the biological pathways involved in fetal development and to elucidate possible gene–environment interaction effects associated with birth defect risks. Such advances could inform effective birth defect prevention strategies.