Discussion
Our pilot WGS study identified 87 SNVs involving 75 genes in embryonic loss (n=1), 370 SNVs involving 228 genes in fetal death (n=3), and 122 SNVs involving 122 genes in stillbirth (n=2) samples, as potentially related to pregnancy loss, across three families. The SNVs included twenty-two de novo , six autosomal dominant and one X-linked recessive mutation(s) that had high pathogenicity scores (pLI>0.9; LOUEF<0.36). In addition, our findings for higher counts of de novo SNVs in losses compared with live births, excess of genes with >1 loss-of-function de novo SNVs (p-value=0.01), and occurrence of multiple de novoevents in a single gene in samples from losses, implicate de novoSNVs in the pathogenesis of pregnancy loss. Furthermore, several of the identified SNVs impact genes (e.g., DICER1,25FBN2,22 FLT4,26HERC1,27,28 and TAOK129 ) that were known to be involved in the development of the embryo and fetus, and that are associated with congenital abnormalities, highlighting the potential role of SNVs in phenotypes that may share a common pathway with recurrent pregnancy loss. Lastly, we identified missense compound heterozygous SNVs impacting genes (e.g.,VWA5B2 ) in two fetal death samples that were absent from live births and population controls, providing evidence for haplosufficient genes relevant to recurrent pregnancy loss.
Previous genetic studies of pregnancy losses are limited for several reasons, including (1) lack of access to paternal DNA samples, which would make interpretations difficult without distinguishing inherited form from de novo variants,30 (2) unavailability of pedigrees with products of conception from chromosomally normal losses and live-births, or (3) unavailability of high-quality data and protocols for DNA restoration and variant detection.31-33 Loss-of-function risk variants and inherited variants in intolerant genes (i.e., genes that are critical for human development, conditions incompatible with life resulting in fetal demise)16,23,34 were not identified, possibly due to limited sample size and focus on families with recurrent, rather than sporadic losses.
Recently, whole-exome sequencing of stillbirth in maternal-offspring duos was conducted to identify variants in intolerant genes that were impossible to ascertain with karyotype or microarray.23 Though the study was limited in ascertaining de novo from inherited variants, due to unavailability of paternal DNA, genes were reported by the authors that are either lethal, known to cause disease, or increase stillbirth risk (e.g., CCR5 , FAT1 , FLNB , INPP5K ,MYO1C , PLOD2 ). Importantly, these genes overlap with findings in our data. For example, we identified a de novomissense chr3:58141895:C:T in FLNB , a gene known for its role in atelosteogenesis type 1, in a stillbirth of Family 3. In addition, we identified autosomal dominant missense chr17:1471262:C:T inMYO1C , a gene linked to deafness and cytoskeletal development, in a stillbirth of Family 1. Given that the SNVs were not identified in live births, the findings warrant validation to confirm potentially lethal variants in FLNB and MYO1C genes causing chromosomally normal stillbirths.
Recently, Kline et . al . similarly hypothesized that chromosomally normal losses are caused by rare variants in several different genes, some of which are incompatible with development to the fetal stage.22 The authors reported damaging variants in several genes that are relevant to recurrent pregnancy loss, including FBN2 (Fibrillin 2). Notably, we identified a de novo in-frame deletion involving FBN2 in fetal death in Family 4 that was not identified in any of the live births across families. Although Kline et . al . identified compound heterozygous variants of FBN2 in embryonic loss, our SV analysis in stillbirth in Family 4 (see Online Supplement ) confirmed a de novoSVs (chr5:128335405) impacting FBN2 , suggesting variants disrupting the FBN2 gene may be incompatible with development to the fetal stage. Furthermore, FBN2 is well described for its role in congenital contractual arachnodactyly and embryonic and fetal development, and may be a potential candidate worth investigating in larger studies.35,36
Given the small participant sample with WGS data in our pilot study, it is noteworthy that we identified variants in several genes (e.g.,DICER1,25 FBN2,22FLT4,26 HERC1,27,28 andTAOK129 ) that were previously identified by genetic studies of pregnancy loss. To confirm our findings, we conducted several validation and confirmatory analyses. First, we compared our data to a population of healthy controls (gnomAD) and identified rare (gnomAD AF<0.006) compound heterozygous SNVs in four genes (TM2D1, MUC16, VWA5B2 ) across two families that were not observed as homozygotes in healthy gnomeAD controls. This finding suggested that variants in haplosufficient genes may contribute to fetal demise in offspring of two healthy parent carriers. Given that our filtering approach is cantered on allele frequencies and predicted impact, and is agnostic to the phenotype of interest, the identification of gene candidates associated with congenital and developmental phenotypes is notable. Although we demonstrated some sharing of SNVs across families (e.g., compound heterozygous SNVs in four TM2D1 , MUC16 ,VWA5B2 were shared across two families), losses may not have common etiologies.22,37 As such, this finding suggests that different genes may play a role at different developmental epochs and across families.16 Second, we explored validation of VWA5B2 gene by Sanger sequencing and confirmed that WGS in our sample confidently called its compound heterozygous SNV (chr3:184236380:T:C). However, further interpretations from our Sanger sequencing results were hindered by the DNA extraction quality and require sequencing of additional samples with higher DNA quality.
Compound heterozygous variants have been previously implicated in pregnancy loss38 and present a scenario in which each parent is purportedly healthy but carries variants in the same gene(s) that may be incompatible with life. As such, functional validation of compound heterozygous variants may provide a clearer picture of the genetic landscape of recurrent pregnancy loss, especially recurrent cases. De novo variants in highly conserved or constrained genes also may lead to pregnancy loss. However, a de novo mutation has a much lower recurrence rate than recessive or dominant inherited disorders.39,40 Impactful X-linked recessive variants, for example, a missense X-linked SNV (chrX:108591181:C:A) impactingCOL4A5 (Alport syndrome 1 gene) and possibly relevant to fetal death (Online Supplement Table S3 ), may also serve as candidates for validation. Importantly, genetic diagnoses based on impactful variants following various modes of inheritance may be used to provide a prognosis based on data from other families with similar mutations.41,42 Confirmation of genes relevant to pregnancy loss will also identify critical pathways and novel therapeutic targets for improving pregnancy outcomes.
Our study has several limitations. The higher counts of de novoSNVs we observed in pregnancy losses compared with live births could result from sequencing error, reflected from degradation of placenta samples due to FFPE. FFPE samples have small fragment sizes and very uneven coverage, contributing to false positive SNVs/SVs. For example, low quality libraries (high DNA degradation) from two samples may have contributed to the large number of de novo SNVs observed in losses in our data. To validate SNVs in our data, we conducted exploratory Sanger sequencing analysis. Results showed poor validation for de novo (data not shown) but confirmed several compound heterozygous calls (Table 3 ) that were not confidently called in our samples. Furthermore, we used Slivar, a method that is strictly a filtering strategy, and the utility of the output relies on high-quality input variants. Future studies utilizing freshly obtained placenta samples for WGS may address elevated sequencing error potentially contributed by FFPE.
Strengths of our study include some prospective collection of samples from losses and live births. This may improve strategies for determining the ‘intolerome’, conditions incompatible with life resulting in fetal demise, and potential to improve database of lethal genes and phenotypes, which are poorly represented. Although our study is underpowered to compare rates of SNVs/SVs between losses and live births, our study serves as a requisite feasibility step in exploring genes relevant to pregnancy loss. Thus, the findings from our pilot study will provide justification for conducting WGS using larger parent-offspring families with potential to identify SNVs causing pregnancy loss.
In summary, the findings reported herein provide evidence for genetic variants (including several in previously recognized genes) relevant to unexplained pregnancy loss in families. WGS of DNA from larger numbers of families (including parent-offspring DNA from affected and unaffected pregnancies) may help identify lethal genes contributing to sporadic and recurrent pregnancy loss. Elucidating pregnancy loss causing genes may lead to biomarkers useful for risk stratification, the identification of genes relevant to normal and abnormal pregnancy, and novel therapeutic targets.