Previously published manuscripts on pregnancy loss were identified through a literature search using PubMed. The search criteria included keywords and Medical Subject Headings terms; “pregnancy loss,” “stillbirth,” “fetal death” or “fetal death,” and “placenta.”

Manuscripts were eligible if they were full-text articles written in English, published between January 01, 2000 and January 01, 2020, and conducted in humans or human cell lines. We chose a search period of the last 20 years in order to identify publications that potentially utilized NGS approaches during a time when accessibility to the technologies increased. In addition, we incorporated publications investigating the placenta as there is likely a genetic contribution to placental insufficiency in some pregnancy losses (16, 17). Since the placenta is genetically similar to the fetus, it allows examination of both maternal and paternal contributions, as well as de novo mutations to pregnancy loss. We excluded case studies, studies with very small sample sizes (n < 4), descriptive articles or commentaries, infertility/non-spontaneous abortion studies and studies with non-genetic etiologies. Systematic review articles that met search criteria were further explored for relevant studies referenced therein.

Studies report their primary outcomes over a broad range of gestational ages [e.g., conventional definitions of stillbirth and Early Pregnancy Loss (EPL)], therefore, in the present review, we aimed to summarize studies by the pregnancy outcomes: EPL (including peri-implantational, pre-embryonic, embryonic, fetal death and early fetal death), late fetal death and losses >20 weeks' gestation, unclassified fetal death (losses assessed across gestation) and RPL. RPL is most commonly defined as ≥3 losses with ≤1 intervening live birth (5), however, we included studies that defined RPL as ≥2 losses. The recommended classifications allow evaluation across various developmental epochs, classifying losses by stage and etiology of loss (14) to help identify common pathways (15). Furthermore, the classifications may identify studies that report genetic factors with different mechanisms, e.g., genes essential for embryonic lethality and functional genes essential for human development (e.g., cardiomyopathy).

In this systematic review, we summarized the studies according to PubMed ID, first author last name and initial, year of publication, pregnancy loss outcome, predictor(s), method of assessment or study design, sample size, and tissue. We provide studies that identified candidate genes with functional pathways. For studies that did not report specific pathways, we conducted an Online Mendelian Inheritance in Man (OMIM) search to identify the roles of the reported genes in disease or functional pathways. Finally, we focused our discussion toward studies that report findings based on genetic factors that are likely causal (e.g., single-gene, autosomal and/or recessive de novo or inherited mutations, “intolerome,” copy number variations [CNVs], single nucleotide polymorphisms [SNPs]) (13). We summarized multi-omic studies, e.g., studies based on proteins and methylated genes that have different mechanisms than single-gene mutations or CNVs. The literature search was cross-examined by Authors. All conflicts were discussed and resolved before proceeding to systematic review.

The PRISMA 2020 checklist was utilized to ensure the manuscript conformed to the systematic review definition. Of note, this study has not been registered with a specific review protocol. There are no randomized clinical trials on genetics of pregnancy loss. Risk of bias was not assessed, principle summary measures were not utilized, and synthesis of data for a meta-analysis was not performed.

Our search yielded 580 potential records. The PRISMA flow diagram is provided in Figure 1. After title and abstract review, 38 records were excluded after additional filters for articles that are not full text, based on non-human studies, and not identified as English articles. After title/abstract review, additional 446 records were excluded because they were either descriptive/commentaries, studies with small sample size (n < 4), qualitative studies, systematic or comprehensive reviews, studies based on infertility and non-spontaneous abortion, or ambiguous with critical information missing. After full-text review, 53 full-text articles that were based on non-genetic factors associated with pregnancy were excluded. In the present study, we included 54 studies that reported findings based on genetic/multi-omic etiologies involved in pregnancy loss.

Nine studies (17%) examined genetic factors in relation to EPL (Table 1). Most of the studies identified dysregulated miRNAs, epigenetic regulators which may have important role in placental development and function. The largest of these, with sample size reaching 105 participants, showed that miR-378a-3p is downregulated in early pregnancy loss (n = 50) compared with normal (n = 55) decidua (24). Hosseini et al. detected other dysregulated microRNAs (e.g., miR-135a) in maternal plasma and villous cells of women (n = 16) who had EPL, but the comparison group were women (n = 8) who underwent abortions (23). Using endocervical specimens collected prior to EPL (n = 20), altered protein expression patterns of extra villous trophoblast (EVT), which plays a role in proper implantation and placentation, were detected in cases compared to controls (21). The authors' ability to obtain EVT cells early from ongoing pregnancies and determine the eventual pregnancy loss occurrences may have allowed opportunities to discover novel biomarkers through global analytic approaches (21).

Since a standardized definition of stillbirth has not been agreed upon, studies examining loss >20 weeks' gestation were lumped together and classified based on their specific cutoffs. One study examined self-reported miscarriage or stillbirth as the primary outcome over a broad range of gestational ages and 10 studies (18%) examined losses >20 weeks' gestation as the primary outcome (Table 2). Of these 10 studies, cutoffs of 20, 22, 23, 24, and 32 weeks were utilized (Table 2). Seven studies examined the associations of genes involved in maternal thrombophilia with losses >20 weeks' gestation. The largest of these, with sample size reaching 1,830 participants, performed a candidate gene analysis (30). The only positive association was with maternal homozygous SNP in FVL (Factor V Leiden) gene (2/488 [0.4%] vs. 1/1380 [0.0046%]; OR = 87.4; 95% confidence interval [95%CI]: 7.9–970.9). The investigators concluded that these heritable thrombophilia genetic markers were not associated with losses >20 weeks' gestation. In another candidate gene study, pregnancy loss >22 weeks' gestation was associated with carriers (n = 96) of allele A of rs1800783 eNOS (endothelial nitric oxide synthase 3) gene in placental tissue. The eNOS gene may be critical for pathways involved in placental growth (28). Furthermore, a genome-wide analysis using high-resolution Illumina SNP arrays identified 24 putative novel CNVs in placental and fetal samples (n = 54) (27). Using a larger study with similar methodology, Reddy et al. detected normal, abnormal (pathogenic), and variants of unknown significance CNVs in 396 (74.4%) samples from pregnancy loss >20 weeks' gestation (including samples with anomalies) (3). The remainder of the studies examining losses >20 weeks' gestation utilized other techniques such as quantitative reverse transcriptase polymerase chain reaction (qRT-PCR), immunohistochemistry, and Western blot.

Thirty-two studies (59%) examined RPL, including pre-embryonic, embryonic, and fetal losses, as the primary outcomes. There was variation in the definition of RPL across studies, with some using a minimum of two losses (34, 35) and others using a minimum of three (17, 36–42) (Table 3). The majority of RPL studies were hypothesis-based, i.e., conducted a candidate gene approach to examine SNPs in selected genes, a priori, and with plausible pathophysiologic pathways. Haplotype analysis conducted by Rogenhofer et al. showed that maternal blood M2 haplotype carriers with RPL (n = 100) in ANXA5, annexin 5 gene involved in coagulation, had a 3.4-fold increased RPL risk compared to controls (n = 500) and a 2.1-fold increased RPL risk compared to randomly selected population controls (n = 533) (47). SNP-prevalence analysis conducted by Jin et al. showed RPL cases (n = 112) carried the rs2249825 G allele in HMGB1 (high mobility group box 1) gene in maternal whole blood more frequently than controls (n = 118) (48). Seyedhassani et al. compared the frequency of mutations in BAX gene, a pro-apoptotic gene, among RPL women (n = 67) and controls (n = 70) and showed associations between A(-179)G mutation in the BAX promoter and RPL (41). Quintero-Ronderos et al. sequenced the complete coding region of THBD, the endothelial cell receptor for thrombin gene, in women affected by RPL (n = 262) and showed THBD-p.Trp153Gly mutation might be related to RPL (54). Lastly, Masini et al. analyzed the genotype and allele frequencies of thrombin-activatable fibrinolysis inhibitor (TAFI) SNPs among women with (n = 86) and without (n = 72) RPL. Genotype and allele frequencies of TAFI +505 and +1583 SNPs were significantly different in women with RPL compared to controls (38).

Genome-wide association studies of RPL were also conducted to highlight genetic variants with relevant functional pathways. For example, Kasak et al. examined placental and parental genome-wide CNV profiles of idiopathic RPL trios (n = 25 parental blood, n = 13 placental) and duos (n = 8 maternal blood, n = 9 placental), and detected CNVs in NUP98 and MTRR genes (7). NUP98 (Nucleoporin 98 And 96 Precursor) and MTRR (5-Methyltetrahydrofolate-Homocysteine Methyltransferase Reductase) genes are implicated in embryonic stem cell development and folate metabolism, respectively (7). Another genome-wide association study was reported by Yu et al. (31) but the study identified DNA methylation and gene expression, mechanisms that are also modulated by environmental factors (31). The study suggested hypo-methylation in CREB5 gene in the decidual tissue was associated with RPL (58).

Next generation sequencing approaches further identified deleterious mutations that are likely causal. For example, by conducting whole exome sequencing (WES) using parental blood and placental chorionic villi samples, Qiao et al. (53) detected compound heterozygous deleterious mutations affecting DYNC2H1 and ALOX15 genes, both critical for early development, in two out of four families with RPL. Among unrelated women (n = 49) affected by RPL, Quintero-Ronderos et al. conducted WES in maternal leukocytes and detected 27 coding variants in 22 genes among 41% of the women. The affected genes, which were enriched by potentially deleterious sequence variants, belonged to distinct molecular cascades playing key roles in implantation (55). Furthermore, Shehab et al. conducted WGS analyses using maternal blood, unaffected offspring blood and fetal tissue in families (n = 7) with recurrent fetal death and detected a frameshift mutation in FOXP3 gene. The authors confirmed the mutation in the affected fetal tissue using Sanger sequencing.

Three studies were based on unclassified pregnancy loss, assessed over a broad range of gestational ages (Table 4). Cochery-Nouvellon et al. conducted a candidate gene study using 3,218 case (experienced embryonic loss at <10 weeks and fetal loss ≥10 weeks gestation) and 6,436 control mother-father pairs, the largest 1:2 matched case-control family-based study included in our review (66). The authors reported that the A6936G allele of PROCR, an endothelial protein C receptor gene involved in coagulation (Table 5), in maternal and paternal blood is associated with fetal death. The authors confirmed the association between candidate gene Factor V Leiden (F5), also involved in coagulation, and fetal loss, but pointed out that relationship between thrombophilias and pregnancy loss varies according to ethnicity and loss type. Alonso et al. (64) also examined mutations in the F5 gene in first-trimester abortions (at ≤ 12 weeks of gestation), second-trimester abortions (at 13–22 weeks of gestation), and fetal death (at ≥23 weeks) of mothers (n = 75). The presence of thrombophilia in 75% of the women combined with a mutation in F5 gene was marginally associated with intrauterine fetal death (P = 0.04; OR = 12; 95%CI: 1.44–102).

Among the 54 studies included in this review, 26 (48%) examined placental tissue (e.g., chorionic villous tissue and trophoblast cells) and reported placental genetic factors associated with pregnancy loss across the developmental epochs. Two studies (4%) incorporated samples from parent-offspring trios (maternal, paternal and fetal/placental) and identified genetic factors related to recurrent losses. Twenty-three studies (53%) examined genetic factors assessed in the maternal tissue samples only (Figure 2). Multiple genetic pathways associated with embryonic and fetal survival may play a role in pregnancy loss. The reported pathways are essential for placental function, epigenetic reprogramming, embryonic development and several critical cellular functions (Table 5).

Eight studies reported genetic/multi-omic etiologies of EPL, however, the studies examined miRNAs, including other epigenetic regulators and proteins that require utilization of expensive targeted assays (e.g., qRT-PCR and immunohistochemistry). Epigenetic mechanisms may play an important role in placental development and function, but are also modulated by environmental factors (7). Indeed, the etiology of many pregnancy losses could be multifactorial, including genetic and environmental factors; however, in some couples, pregnancy loss can be inherited as a Mendelian trait (i.e., monogenic form) (67). Despite the strong genetic underpinnings underlying EPL (10, 68), evidence for causal genetic variants is lacking.

Among genome-wide association studies of pregnancy loss at 20 weeks' gestation or more, two studies utilized chromosomal microarray, a higher resolution and enhanced sensitivity method that allowed unbiased detection of pathogenic abnormalities (3, 27). These studies by Reddy et al and Harris et al detected 24 putative novel CNVs in 54 placental and fetal samples from losses >20 and 22 weeks' gestation, respectively, and genetic abnormalities explained 41.9% of idiopathic cases (3, 27). A recent study, that was not included in our review due to its publication date, improved these findings by utilizing NGS approach that allowed detection of the de novo lethal mutations and the “intolerome” (i.e., genes that are critical for human development, the loss of which is incompatible with life) (11). Using the maternal and fetal samples, enrichment of loss-of-function variants in genes that are intolerant to variation in the human population were observed. This suggested dramatic and progressive increases in the proportion of losses >20 weeks' gestation with likely causative genetic abnormalities, however, the genetic etiologies of 40% of idiopathic cases remain to be elucidated. Due to unavailability of paternal samples in the previous studies, they could not detect compound heterozygous variants, distinguish pathogenic de novo from inherited variants and consequently could not explain significant proportion of idiopathic cases. Additional efforts were made by Cochery-Nouvellon et al. (66) that utilized mother-father duos with larger sample size. However, the study was a candidate gene study and showed limited evidence of association between coagulation pathway genes and unclassified pregnancy loss.

Among thirty-two studies that reported genetic etiologies of RPL, making up the majority of studies included in this review, two utilized an NGS approach in families to identify deleterious mutations that are likely causal (53, 56). Using WES analysis in parental blood and placental chorionic villi samples, Qiao et al. (53) detected compound heterozygous deleterious mutations affecting DYNC2H1 and ALOX15, genes critical for early development, in two out of four families with RPL. By conducting WGS followed by Sanger sequencing validation analyses, Shehab et al. (56) detected frameshift mutation in FOXP3 gene that is critical for the function of regulatory T cells in families affected by recurrent intrauterine fetal death. Other genes such as loss-of-function risk variants and inherited pathogenic mutations in intolerant genes were not identified, potentially due to the lack of larger parent-offspring trio studies.

While chromosomal microarray, the current clinical guideline for genetic evaluation of losses >20 weeks' gestation, enhanced the ability to detect microdeletions and duplications beyond the resolution of standard karyotype, additional detailed diagnostic yields will require utilization of NGS approach. Efforts are underway to apply this technology to losses >20 weeks' gestation (69).

With the advent of NGS, monogenic disorders (including de novo, inherited autosomal-dominant/autosomal-recessive mutations, and SNPs) that are either lethal, known to cause disease, or dramatically increase risk of pregnancy loss in families can be identified. De novo mutations occur as likely penetrant variation in a Mendelian gene and could explain sporadic cases of pregnancy loss. Point mutations, other genetic variations such as CNVs (genomic deletions or duplications), as identified by studies in this review, may also occur de novo. The added contribution of novel de novo missense variants to losses >20 weeks' gestation was estimated by pulling all rare and damaging novel missense variants in the study (11). Therefore, the authors estimated a bound on the diagnostic yield in known genes associated with losses >20 weeks' gestation between the previously reported yield (4.5%) vs. the present yield (13.4%; 36/268 cases). However, without parental genotype information, the study remained at the lower bound of the diagnostic yield. Consistent with other diagnostic studies, the diagnostic yield using parent-offspring trios is estimated to be up to three-fold higher compared with studies that use singletons (70).

Combined with identification of de novo mutations, other single gene abnormalities may be used to provide prognosis based on data from other patients with similar mutations (71). Such monogenic forms may be associated with extreme phenotypes and early losses, but this is not always the case. Studies that show familial aggregation of pregnancy loss may help clarify whether losses that occur early in gestation and a positive family history exists, suggesting autosomal-dominant transmission of risk alleles. To prove whether the mutations appeared in the germline of the probands as de novo mutations, parental DNA assessment is required (67).

Challenges still remain in clinical applications of genome sequencing and validating the results from sequencing using maternal cell-free DNA, chorionic villus sampling and amniocentesis. Suggested strategies to overcome these challenges include serial assessment of genotypes, phenotypes and ‘omics data over the course of the pregnancy (e.g., genomics, transcriptomics, metabolomics) (10, 68). Molecular diagnostic evaluations rely on databases (e.g., OMIM) and guidelines of the American College of Medical Genetics and Genomics with characteristics designed to enrich for pathogenicity in Mendelian disease genes (11). In these databases, lethal phenotypes are especially poorly represented. Other strategies for gene discovery, including determination of the “intolerome” are likely to reveal new genotype-phenotype correlations and shed light on the human “intolerome,” conditions incompatible with life resulting in fetal demise (11, 13). Studies that incorporate DNA sequencing in affected and unaffected families, designed as case-control trio studies, will help in determination of the “intolerome” by identifying novel embryonic-lethal or fetal-lethal variants that are not seen in unaffected families. Using WGS in parent-offspring trios, 60–80 high confidence de novo mutations per individual can be identified (67). Compared with WES, WGS may further expand the spectrum of causal de novo mutations by allowing for a better coverage of the exome and identification of non-coding variants.

Although PubMed search is a comprehensive retrieval tool appropriate for systematic review of journal research in health care, other search methods (e.g., Embase, Web of Science) were not utilized. Restricted MeSH terms applied in PubMed may have excluded other studies pertinent to the present systematic review. To provide a more comprehensive review of the literature, we reviewed and included studies within review articles that matched eligibility in our search criteria. In addition, we independently explored OMIM to report and confirm genetic pathways and functional effects of the reported genes.

Experts have recommended categorization of pregnancy loss as: <10 weeks gestational age (termed early pregnancy loss), 10–19 weeks and 6 days of gestation (termed fetal death), and 20 or more weeks gestation (termed stillbirth). EPL was further subdivided into peri-implantational loss before 5 weeks, pre-embryonic loss from 5 to 5 weeks and 6 days of gestation, and embryonic loss from 6 to 9 weeks and 6 days of gestation (14). Similarly, fetal death can be subdivided into early fetal death, defined as losses between 10 and 15 weeks and 6 days of gestation, and late fetal death, losses from 16 to 19 weeks and 6 days of gestation (14). These classifications may identify studies that report genetic factors with different mechanisms, e.g., genes essential for embryonic lethality and functional genes essential for human development (e.g., cardiomyopathy). Additionally, assessment of losses at different stages of pregnancy may help identify pathways essential for in utero survival at critical stages of development.