SNVs relevant to recurrent pregnancy loss
After removing poor DNA quality samples and samples failing sex-check (five pregnancy losses samples and one family), 3,211,893 SNVs remained for further analysis. Finally, 28,485 impactful SNVs (i.e., missense, frameshift, insertion/deletion, stop gained/retained, and splice region) in all samples from the products of conception (n=16 in three families) were prioritized by Slivar. Using samples that passed quality control (n=16 in three families; Online Supplement and Table S1 ), we identified 87 SNVs involving 75 genes in an embryonic loss sample, 370 SNVs involving 228 genes in three fetal death samples, and 122 SNVs involving 122 genes in two stillbirth samples (Figure 1 and Table 2 ). In Family 1, the SNVs included 11 compound heterozygous, 11 de novo and 92 autosomal dominant in the fetal death cases, and 1 compound heterozygous, 7 de novo and 35 autosomal dominant in stillbirth cases (Figure 1 ). In addition, the SNVs in Family 2 included 6 compound heterozygous, 41 de novoand 40 autosomal dominant in the embryonic loss case, 6 compound heterozygous, 15 de novo and 62 autosomal dominant in the fetal death case, and 6 compound heterozygous, 30 de novo and 43 autosomal dominant in the stillbirth case. Further, the SNVs in Family 4 included 12 compound heterozygous, 5 de novo and 155 autosomal dominant in the fetal death case. Several SNVs identified in our data impact genes that were known to be involved in the development of the embryo and fetus, and congenital abnormalities (e.g.,DICER1,25 FBN2,22FLT4,26 HERC1,27,28 andTAOK129 ).
Among the SNVs we identified, 29 SNVs are predicted as pathogenic (pLI>0.9; LOUEF<0.36), impacting 27 genes, several of which are involved in known diseases (Table S3 ). Specifically, we identified three autosomal dominant and three de novo pathogenic SNVs in fetal death and stillbirth from Family 1, one autosomal dominant and sixteen de novo pathogenic SNVs in embryonic loss, fetal death and stillbirth from Family 3, and one autosomal dominant, one X-linked recessive and three de novopathogenic SNVs in fetal death from Family 4. Given the counts ofde novo SNVs that are higher in losses than live births, we provided details, which included a table of loss-of-function de novo SNVs by pathogenicity and gene impact and exploratory de novo enrichment analysis (Online Supplement and Table S4 ). De novo SNVs were predominantly missense (nonsynonymous) followed by frameshift, splice region, in-framedeletion/insertion and stop gained. The observed mean de novoloss-of-function SNVs in pregnancy losses was higher than that of the expected (2 vs 0.2; p-value=0.01). Moreover, the SNVs were enriched in >1 protein altering genes (p-value<0.001).
Furthermore, among compound heterozygous SNVs we identified, four SNVs in three genes (TM2D1, MUC16, VWA5B2 ) were identified in fetal death from Family 1 but not in any of the live births (Table 4 ). The SNVs were not observed as homozygotes in healthy controls, highlighting their potential relevance to pregnancy loss in our samples. Finally, we conducted exploratory analyses to confirm and validate our findings, which included exploratory SNV rates comparison (Table S2 ), rare-variant association, and Sanger sequencing analyses. The methods and summary of results based on our exploratory analyses are provided in Online Supplement .