Abstract
STUDY QUESTION Are rare genetic variants in the conceptus associated with chromosomally normal pregnancy loss?
SUMMARY ANSWER The proportion of probands with at least one rare variant is increased in chromosomally normal loss conceptuses compared with controls.
WHAT IS ALREADY KNOWN Among non-consanguineous families, one study of seven chromosomally normal losses to four couples with recurrent pregnancy loss (RPL) and a case report of a family with RPL of which one was known to be chromosomally normal identify compound heterozygote variants in three different genes as possibly causal. Among consanguineous families, RPL of chromosomally normal pregnancies with non-immune hydrops fetalis (NIHF) has been attributed to recessive variants in genes previously implicated for NIHF and new candidate genes.
STUDY DESIGN, SIZE, DURATION The starting sample was 52 chromosomally normal losses to 50 women, identified in 2003-2005 as part of a cohort study on trisomy and ovarian aging. The analytic sample comprises 19 conceptus-parent trios with DNA from 17 biologic parents (cases). The control group derives from the National Institutes of Mental Health’s National Database for Autism Research (NDAR). It comprises 547 trios of unaffected siblings of autism cases and their parents.
PARTICIPANTS/MATERIALS, SETTING, METHODS We use exome sequencing to identify rare variants in the coding region of the genome. We defined variant rarity in two ways: ultra-rare (absent in gnomAD) and rare (heterozygote <10−3 in gnomAD). For autosomal recessives, we further required that the variant was absent as a homozygote in gnomAD. We compare the rates of rare predicted damaging variants (loss of function and missense – damaging) and the proportions of probands with at least one such variant in cases versus controls. Secondarily, 1) we repeat the analysis limiting it to variants in genes considered causal in fetal anomalies and 2) we compare the proportions of cases and controls with damaging variants in genes which we classified as possibly embryonic lethal based on a review which was blinded to case-control status.
MAIN RESULTS AND THE ROLE OF CHANCE The rates of ultra-rare damaging variants (all de novo) are 0.21 and 0.17 in case and control probands, respectively. The corresponding rates for rare potentially pathogenic de novo variants are 0.37 and 0.24, respectively; for autosomal recessive variants they 0.11 and 0.03. The proportions of probands with at least one rare potentially damaging variant were 36.8% among cases and 22.9% among controls (odds ratio (OR) = 2.0, 95% CI 0.9, 3.0). Secondary analyses show no damaging variants in fetal anomaly genes among case probands; the proportion with variants in possibly embryonic lethal genes was increased in case probands (OR=14.5, 95% CI 3.4, 61.1). Cases had variants in possibly embryonic lethal genes BAZ1A, FBN2 and TIMP2. Post hoc review of these cases suggests that CDH5 may also be an embryonic lethal gene.
LIMITATIONS, REASONS FOR CAUTION The number of case trios (n=19) limits the precision of our point estimates. We observe a moderate association between rare damaging variants and chromosomally normal loss with a confidence interval that includes unity. A larger sample is needed to estimate the magnitude of the association with precision and to identify the relevant biological pathways.
WIDER IMPLICATIONS OF THE FINDINGS Our data add to a very small literature on this topic. They suggest rare genetic variants in the conceptus may be a cause of chromosomally normal loss.
TRIAL REGISTRATION NUMBER N/A.
STUDY FUNDING/COMPETING INTERESTS(S) Exome sequencing of case trios was performed by Baylor-Hopkins Center for Mendelian Genomics through National Human Genome Research Institute grant 5U54HG006542.
Data used in the preparation of this manuscript were obtained from the National Institute of Mental Health (NIMH) Data Archive (NDA). NDA is a collaborative informatics system created by the National Institutes of Health to provide a national resource to support and accelerate research in mental health. Dataset identifier(s): src_subject_id. This manuscript reflects the views of the authors and may not reflect the opinions or views of the NIH or of the Submitters submitting original data to NDA.
No author has a competing interest.
Introduction
Pregnancy loss <20 weeks gestation (miscarriage), is the most common adverse outcome of pregnancy, occurring in about 30% of pregnancies that survive to implantation and 10-20% of pregnancies that survive to six weeks (Zinaman et al., 1996; Wang et al., 2003; Wilcox, 2010). Chromosomal abnormality in the conceptus is the cause of at least 50% of losses, with the proportion determined mainly by maternal and developmental (e.g., prefetal, fetal) ages of the sample (Kline et al., 1989; Warburton, 2000; Hardy et al., 2016). The abnormality usually results in embryonic or fetal lethality (hereafter, embryonic lethality).
The remaining pregnancy losses are chromosomally normal. Aside from an association with maternal age and a few rare conditions, there are few established risk factors for these losses (Kline et al., 1989; Cohain et al., 2017). It has long been hypothesized that variations in single genes are a cause of chromosomally normal loss. Embryonic lethal variants may arise de novo or they may be inherited. We hypothesize that de novo variants predominate for three reasons. First, the increased risk of loss after chromosomally normal loss is modest, on the order of 1.4-2 fold (Boue et al., 1975; Lauritsen, 1976; Kline et al., 1989). Second, among women with repeat losses, the increased risk of a second chromosomally normal loss is modest (Warburton et al., 1987). Recent studies suggest a higher recurrence risk for chromosomally normal loss, but do not adjust for maternal age (Sugiura-Ogasawara et al., 2012; Nikitina et al., 2020). The increased recurrence risk for chromosomally normal loss is detectable among women <35, but not among older women because of the increasing risk of trisomy with age (Warburton, 1987). Third, the morphologic phenotypes are heterogeneous. Most chromosomally normal losses are of pregnancies that did not reach the fetal stage of development (<8 weeks post-conception, hereafter, prefetal). In one embryoscopic study, among 56 chromosomally normal, predominantly prefetal, losses, 29% are classified as normal embryos, 36% as growth disorganized embryos and the remainder with combined or isolated defects (Philipp et al., 2003) (our computations). Of the latter group, eight embryos/fetuses (14% of the 56) had structural anomalies, with the remainder involving a phenotype that likely reflects autolysis in utero or amniotic adhesion or band anomalies.
Most searches for variants in single genes that disrupt in utero development, ordinarily using exome sequencing (ES), focus on structural anomalies, whether diagnosed prenatally or at birth or later. These studies show that genes associated with anomalies among pediatric patients are associated with the same anomalies detected earlier in fetal development (Lord et al., 2019; Petrovski et al., 2019); see also Best et al (Best et al., 2018). Some of these variants may have variable phenotypes, extending to the phenotypes seen in chromosomally normal losses. By analogy with aneuploid losses, however, we hypothesize that chromosomally normal loss is caused by rare variants in many different genes, with some incompatible with development to the fetal stage.
In pregnancy loss, many studies focus on the woman, seeking to identify maternal genotypes associated with recurrent pregnancy loss (RPL). These studies focus on genes related to selected processes (e.g., immune response, metabolism, coagulation, angiogenesis). Findings are inconsistent, with positive studies reporting only modest associations (Rull et al., 2012; Pereza et al., 2017; Shi et al., 2017).
Few studies have sought to identify genetic variants in losses occurring before a diagnosed structural anomaly. A major limitation of most studies is the absence of information on the chromosomal constitution of the conceptus. Pregnancy loss studies have focused on women with RPL, assuming implicitly that repeat losses have the same cause. This assumption is reasonable in consanguineous couples, but less so otherwise. A widely cited study of consanguineous couples with RPL ascertained the sample by the presence of lethal non-immune hydrops fetalis (NIHF) in at least two pregnancies (Shamseldin et al., 2015; Shehab et al., 2017). Of 19 probands studied, five were chromosomally normal losses of which two were homozygous for genetic variants known to cause NIHF and three for novel candidate embryonic lethal variants. Case reports of consanguineous families with RPL and an index chromosomally normal loss also identify recessive or X-linked variants as causal (Shehab et al., 2017). Several case reports or series in non-consanguineous couples that are sometimes classified as RPL studies base ascertainment on recurrence of a structural anomaly in two pregnancies (see Table 2 in review by Colley et al 2019 (Colley et al., 2019)). The underlying assumption is that RPL and structural anomalies in the same family have a common cause. It is unclear whether this assumption is valid. Even if it is, findings from these studies will not generalize to most women or couples with RPL, nor with sporadic chromosomally normal loss.
In non-consanguineous couples, repeat losses may have different causes (e.g., a mix of chromosomally normal and abnormal or abnormal of different types). Several studies hypothesize that the maternal (and in one instance paternal) genotype is related to RPL through repeat triploidy (Filges et al., 2015) or trisomy (Bolor et al., 2009; Mizutani et al., 2011; Hanna et al., 2012; Sazegari et al., 2014). In these studies, although the parental genotype is examined, the hypothesis is that aneuploidy in the conceptus caused the loss. Of the few studies attempting to discover embryonic lethal variants in single genes in RPL, only three (Qiao et al., 2016; Fu et al., 2018; Mitra et al., 2020) draw on conceptuses known to be chromosomally normal.
A study of seven chromosomally normal loss conceptuses (Qiao et al., 2016) identifies as potentially causal two compound heterozygote variants in repeat chromosomally normal losses to two couples. A similar finding stems from a case report of a couple with three first trimester losses (Mitra et al., 2020), at least one of which was chromosomally normal. These two reports identify two genes that are associated with structural anomalies (DYNC2H1 and DHCR7) suggesting that, in some instances, variants in genes associated with structural anomalies may reflect the lethality of already identified features of the associated syndromes or expansion of the phenotypes to include arrested embryonic development without structural anomalies.
The largest published study includes 19 chromosomally normal loss conceptuses (Fu et al., 2018). The investigators assessed the occurrence of damaging SNVs in 286 candidate genes considered related to embryonic lethality. Interpretation is limited by the absence of parental sequencing data (to determine whether variants in the conceptus are de novo or inherited), a comparison group, and description of how the genes were selected.
In this paper we add to a small body of literature to estimate the association of rare potentially damaging SNVs with chromosomally normal loss. We draw on an unselected sample in which we obtained DNA from the chromosomally normal conceptus (proband) and parents. We compare the frequency and types of rare potentially damaging SNVs in the 19 trios (cases) with those in 547 unaffected siblings of autism case trios (controls).
MATERIAL AND METHODS
Study population
Cases: The sample is drawn from a consecutive series of pregnancy losses <18 weeks developmental age identified at a single hospital (2003-2005) (Warburton et al., 2009). The current study includes all chromosomally normal losses for which we had stored conceptus and maternal DNA and the woman consented to use of stored samples in future studies on reproduction.
We submitted 61 conceptus and maternal DNA samples to Baylor Hopkins Center for Mendelian Genetics for ES (Supplement Table S1a sets out the derivation of the sample). Of the 61 samples, nine were excluded from analysis prior to ES (Table S1a, footnote 2). The remaining 52 chromosomally normal losses occurred to 50 couples; two couples each had two chromosomally normal index losses.
Forty-seven (49 losses) women agreed to future contact. In 2016-2018 we attempted to contact these women to update their obstetric histories and to obtain paternal saliva samples (for analysis of DNA). We updated the obstetric histories of 25 women (27 losses) and obtained DNA from saliva from 17 biologic fathers (19 losses).
The primary analytic sample comprises the 19 trios. In addition, we had permission to analyze DNA from all conceptus − mother duos. Of the 33 duos, 28 have good quality ES data (Table S1b and see below). We use these data to define the sample of variants included in the analysis.
We use pathology laboratory data and reported last menstrual period, gestational age and sonography findings to classify losses as prefetal or fetal. The pathology data are of limited utility in part because products of conception were obtained by suction curettage. We define as prefetal, losses that occurred <71 days gestation (from last menstrual period) or in which the conceptus was judged by ultrasound to be non-viable or to have stopped development <71 days. We define as fetal, losses that occurred at ≥71 days and in which ultrasound findings indicated viability at ≥71 days or a fetus or fetal parts were identified in pathology. Among the 19 trios, 12 were prefetal, 5 fetal and 2 of unknown morphology. Mean gestational age was 74 days (range 30-119). At the follow-up interview, mean maternal age was 46 years (range 39-55) and of the 17 women, 13 (76%) reported two or more pregnancy losses (including the index loss).
Controls
We selected controls on January 13, 2017 from the National Institutes of Mental Health’s National Database for Autism Research (NDAR) (Hall et al., 2012; NIH, 2020). The control sample comprises unaffected siblings of autism cases with exome sequencing performed on DNA from the sibling, mother, and father. We further required the sibling be aged 60-251 months at interview (to limit the chance of including a child with undiagnosed autism). For families with more than one sibling in the NDAR database, we include only one sibling (the oldest at time of interview). In total, 1475 sibling-mother-father trios met these criteria. From these, we selected the trios where the sequencing platform and exome coverage in the sibling and parents was comparable to that used for cases (see below). Of the 1475, 587 families met these criteria. Of these, the genetic sequencing data files turned out to be corrupted (n= 26) or ES data of poor quality (n=14, see below), resulting in a sample of 547 control trios.
Exome Sequencing
For cases, the proband and maternal exome was captured using the Agilent SureSelect HumanAllExon V4.51 MbKit_S03723314 kit (Agilent v4 capture kit). The paternal exome was captured using the Agilent SureSelectXT Clinical Research Exome. For controls, the exome was captured using NimbleGen custom array or EZExomeV2.0(Sanders et al., 2012).
Variant-calling
We used New York Genome Center’s automated analysis pipeline to jointly call variants in cases and controls (Arora et al., 2019). We aligned sequencing reads to the human reference GRCh37 using BWA-MEM v0.7.15 (DePristo et al., 2011; Van der Auwera et al., 2013). We called and recalibrated variants using the GATK best-practices.
Exclusions
We used the PLINK’s genome-wide IBD sharing algorithm (Purcell et al., 2007; Chang et al., 2015) to check for relationships within and between pedigrees and homozygosity of the X-chromosome to compare phenotypic with genotypic sex. We exclude families that fail either the relationship or sex checks. Among cases, we exclude five conceptus − mother duos (Table S1b). Among controls, we exclude 14 trios.
Case-control comparability
The exome capture library kits used in cases capture 51 MB of the genome, while the kit for controls captures 37 MB of the genome. To avoid bias in detecting putatively causal variants, we limit our analysis to the intersection of the two capture kits such that each qualifying variant was called (as homozygous reference or non-reference) in at least 80% of 123 case samples (includes 19 trios and 28 conceptus − mother duos) and 80% of 1641 control samples (547 trios). This eliminates variants that occur in off-target regions or regions covered unevenly between cases and controls.
Primary analysis
We define as damaging variants with predicted loss of function (i.e., nonsense, splice site, and frameshift, hereafter, LoF) and missense variants with REVEL>0.50 (Ioannidis et al., 2016). For de novo and X-linked variants we carry out the analysis twice, first using a strict rarity criterion (i.e., the variant in the proband did not occur in gnomAD (Karczewski et al., 2020), hereafter “ultra-rare”) and second, relaxing the rarity criterion (i.e., overall minor allele frequency (MAF) in gnomAD <10−3, “rare”). For homozygous and compound heterozygote (CH) variants, we also required that the variant be absent as a homozygote in gnomAD. We hypothesize that embryonic lethal variants are incompatible with survival to birth and normal development. Thus, in our analyses, we exclude de novo variants in probands that are also observed in another unrelated member of the sample (parents or probands), excepting variants that occur in more than one proband in the same case or control group.
We compare cases and controls for: 1) the rates of variants of specific types (LoF, missense – damaging, missense – tolerated and synonymous) and mode of inheritance and 2) the proportion of probands with at least one damaging (LoF or missense – damaging) variant. We use synonymous variants to determine whether case and control data are comparable. We test the significance of differences in case-control differences in rate ratios with a conditional two-tailed binomial test and the case-control differences in the proportions with a two-tailed Fisher’s exact test. We use P<0.05 as the threshold in testing the statistical significance of each comparison and provide 95% confidence intervals for the odds ratios (ORs) in the second comparison. We do not adjust for multiple comparisons. We provide data for cases and controls in the supplement (Supplement Tables S2-S5) to allow others to test or generate hypotheses.
Secondary analyses: First, we repeat the analysis using the rare criteria, limiting the analysis to 1340 genes classified as causing or likely to cause fetal anomalies and taking account of allele type associated with disease (Lord et al., 2019). Second, we compare the frequency of damaging variants that occurred in genes likely to be lethal to the embryo. To classify genes and variants as suspect, two investigators (AT and VJ) reviewed the literature on all genes that occur in damaging variants in our sample without knowledge of case-control status. In PubMed, they searched for “GeneX NOT cancer NOT tumor NOT carcinoma”. Depending on the initial search results, the added terms including pregnancy or miscarriage or abortion or stillbirth or development. For genes without a well-established association with human disease or for which the initial search yielded very little information, they expanded the search to animal or in vitro models. They evaluated the variants based on the guidelines of the American College of Medical Genetics and Genomics (Richards et al., 2015) under the assumption that the variant occurred in a case proband. Since we were for the most part evaluating genes, rather than specific variants, they did not use the five variant classification categories of the College. Rather, they classified variants in four categories based on what is known about the gene. The categories are: known lethal (there are none), “likely” to be lethal (none) to the embryo, “possibly” lethal or “not likely”. Of the 154 genes (159 variants) in which damaging variants occurred, they classified 10 as possibly lethal (Supplement S6).
Results
Cases and controls have similar rates of synonymous missense variants. Cases and controls do not differ significantly in either the rates of LoF and damaging de novo or recessive, including compound heterozygous (CH), variants nor in the proportion of probands with at least one damaging variant, whether we define variants as ultra-rare or rare (Table I). Among probands, 36.8% of cases had at least one rare damaging variant compared with 22.9% of controls (OR = 2.0, 95% CI 0.9, 3.0).
Genetic variants among 19 chromosomally normal losses (cases) and 547 unaffected siblings of children with autism (controls): a-c) Rate ratios for variants classified by rarity, mode of inheritance and type and d) Odds ratio relating presence of any potentially damaging variant to chromosomally normal loss
Secondary analyses: No cases and 15 controls have at least one damaging variant in a gene associated with fetal anomalies. Among controls, the damaging variants occur in 13 different genes (Supplement S2, S4).
Three cases and seven controls have at least one damaging variant in a gene classified as “possibly” embryonic lethal (OR = 14.5, 95% CI 3.4, 61.1, P=0.003). Among cases, two variants, in BAZ1A and TIMP2A, are de novo, and one, in FBN2, is a compound heterozygote (Supplement S7).
One case had a de novo nonsense variant in BAZ1A, c.[1156C>T]; p.[Arg386Ter]. BAZ1A (Bromodomain Adjacent to Zinc Finger Domain, 1A; MIM#605680) is involved in chromatin remodeling. The variant has a pLi of 1 (o/e 0.1; 8 observed, 82 expected in gnomAD). While BAZ1A does not currently have an OMIM gene-disease association, two reports describe BAZ1A missense variants in individuals with multisystemic syndromic phenotypes (Zaghlool et al., 2016),(Weitensteiner et al., 2018). Thus, we speculate that a nonsense variant in BAZ1A might disrupt embryonic development. This same case had a de novo missense variant in RRAD (Supplement S8), however, given that there are >50 heterozygous individuals in gnomAD, we think it unlikely that this variant is related to pregnancy loss.
One case had a de novo damaging missense TIMP2 variant, c.[553G>A]; p.[Gly185Arg]. TIMP2 (Tissue Inhibitor of Metalloproteinase 2; MIM#188825) is an inhibitor of Matrix Metalloproteinase 2 (MMP2; MIM#120360), which is a type IV collagenase. The MMP2/TIMP2 complex is involved in several gestational processes including implantation and placentation (Staun-RamShalev, 2005). We hypothesize that perturbation of these processes may be the cause of the pregnancy loss.
One case had two damaging missense variants in FBN2, c.[4457G>T];[2927G>T] and p.[Ser1486Ile];[Gly976Val]. Autosomal dominant variants in FBN2 (Fibrillin 2; MIM#612570) are associated with Congenital Contractural Arachnodactyly (Beals syndrome, CCA; MIM#121050), which includes skeletal, joint, and muscular manifestations (Callewaert, 1993 updated 2019; Meerschaut et al., 2020). Both severe and mild phenotypes have been observed, and inter- and intrafamilial variability is recognized (Callewaert, 1993 updated 2019). In our case, we did not clinically examine the parents and thus do not know whether they manifest a mild unrecognized form of CCA. Since pathogenic variants in FBN2 can result in fetal akinesia, it is possible that biallelic pathogenic variants in this gene may disrupt embryonic development.
This same case proband had a de novo damaging missense variant in CDH5 (Supplement S8). We did not classify this gene as possibly embryonic lethal a priori in part because we accidently missed relevant literature from mouse models. CDH5 (Cadherin 5; MIM#601120) does not currently have a gene-disease association in OMIM. The c.535G>A (p.Asp179Asn) variant identified in the conceptus is absent in gnomAD, although several variants within the same region occur at low frequencies. CDH5 is involved in cell differentiation and vascular endothelial development. On discovering that the CDH5 variant occurred in a case, we re-examined the literature. In homozygous mouse knockout models, CDH5 is an embryonic lethal; absence is associated with defects in endothelial cell organization and angiogenesis (Gory-Faure et al., 1999; Crosby et al., 2005). Thus, the one case has de novo damaging missense variants in two genes, FBN2 and CDH5, each of which might disrupt embryonic development.
Of the seven controls with at least one variant in a possibly embryonic lethal gene, two had another damaging variant (one in SLC4A1 and the other in WDR19). Both genes are classified as unlikely to be embryonic lethal (Supplement S9).
We studied two chromosomally normal losses to two couples in the sample. In one family, neither conceptus had a potentially damaging variant. In the other family, one conceptus had a potentially damaging variant and one did not.
Discussion
Our data provide support for the hypothesis that rare genetic variants in the conceptus may cause pregnancy loss. Our sample, however, was too small to estimate the strength of association with precision. Thus, we observed a moderate (OR=2.0), not statistically significant, increase in the proportion of cases with at least one rare potentially pathogenic variant compared with controls. In support of the inference that rare variants are a cause of some chromosomally normal losses, the point estimate for the rate ratios for de novo LoF variants is stronger than that for missense – damaging variants (2.3 versus 1.6).
When we limited the analysis to genes associated with fetal anomalies (Lord et al., 2019), no case proband has a damaging variant compared with 15 (2.7%) of control probands. Among controls, damaging variants occurred in 13 genes.
One limitation of our study is that we did not assess conceptuses for potentially pathogenic copy number variants. We do not consider this a significant limitation because these variants are rare in chromosomally normal loss conceptus (about 5%) (Zhou et al., 2016).
A priori, among potentially damaging variants, we classified 10 genes as possibly lethal in utero. Variants in possibly lethal genes occurred in three cases and seven controls (OR = 14.5). Among cases, the possibly lethal genes are BAZ1A, FBN2 and TIMP2. If the associations with variants in BAZ1A and FBN2 are causal, we hypothesize a mechanism related to disruption of embryonic development. For TIMP2 the literature favors mechanisms that interfere with implantation or placentation. Of the remaining 16 cases, four had damaging variants in genes that we did not classify as a possibly embryonic lethal. These include cases with de novo damaging variants in ARSH, FAM131B and MORN5 and a compound heterozygous variant in SERGEF. Post hoc, in a review of the three case probands with possibly embryonic lethal genes, we discovered that one also had a variant in CDH5. Although we now think that CDH5 is a possibly embryonic lethal gene, we do not include this gene in our case-control comparison. Because the CDH5 variant occurred in a case already classified as having a variant in a possibly embryonic lethal gene, inclusion would not alter our estimate of the strength of association.
FAM131B and MORN5 are not currently listed in OMIM; their protein function is unknown; we did not find any reports of associations with human or mouse phenotypes. ARSH (Arylsulfatase H; MIM#300586) and SERGEF (Secretion-Regulating Guanine Nucleotide Exchange Factor; MIM#606051) are listed in OMIM, but without a disease association; we did not find any evidence of associations with human or mouse phenotypes. Current knowledge is insufficient to speculate on a role for any of these genes in pregnancy loss.
Our analyses have several strengths. First, our data from controls—unaffected siblings of autism probands—are consistent with those reported in another similar control sample, supporting the validity of our analysis. In a study of congenital heart disease (Jin et al., 2017), at least one potentially damaging de novo or autosomal recessive variant was detected in 26% of unaffected siblings of autism proband controls. In our sample, the corresponding rate is 23%. Second, we classified the genes in which damaging variants occurred as possibly embryonic lethal or not without knowledge of case-control status. Thus, unlike prior observations in this area, our comparison is not vulnerable to unintended biases that accompany knowledge of the pregnancy outcome. (Our reconsideration of CDH5 is an exception to this statement since it is done with knowledge of case-control status.) Third, our sample is larger than any other reported study of chromosomally normal loss conceptus trios thus adding to a small body of literature on such losses in a non-consanguineous sample. On the other hand, our data comprise only 19 trios, limiting not only our ability to make a precise estimate of the strength of associations, but also to identify likely causal variants. Since we hypothesize that many genetic variants contribute to chromosomally normal loss, we did not expect to find damaging variants in the same gene in more than one case in our sample.
In sum, our data suggest that the odds of a lethal variant might be increased by as much as two-fold in chromosomally normal loss conceptus compared with controls. This result requires replication in a larger sample both to confirm the point estimate and to identify biological processes in chromosomally normal loss.
Data Availability
All raw and processed de-identified data will be shared upon request, as per the consent specified in our IRB approval.
ACKNOWLEDGEMENTS
We acknowledge the late Dorothy Warburton who conceived this study with us. We acknowledge Judy Chih-Yu and Vanessa Felice who prepared the DNA samples, Amelia Zuver and Troy Layouni who set up the fieldwork databases, and Sarah Robbins who collaborated in the planning stage of the analysis. We thank Kathy Hardy for her insightful feedback on this paper. We thank the women and men who participated in this research.
Footnotes
Author added; Supplemental files updated (new table S6 added) Discussion section revised to reflect updated results