The clinical efficacy of an ES or GS approach for the molecular diagnosis of postnatal developmental disorders is well documented, with diagnostic yields ranging from 10-70% dependent on the primary presentation 27,28. More recent studies have also focused on fetuses with congenital abnormalities identified on ultrasound, and implied that quick turnaround times may aid couples when decisions on elective terminations of pregnancy are to be made 29–31.
However, genomic analyses have not yet been implemented in the clinical setting for cases of perinatal death. A small number of research projects in recent years have sought to elucidate the genetic causes of perinatal death by genomic analyses, returning diagnostic yields ranging from 14-57%14–19,32, suggesting that many of these cases will have an identifiable monogenic basis. While differences in study design and inclusion criteria likely account for some of the variability in diagnostic rate, this discrepancy also reflects the challenges currently faced in the genomic analysis of disorders presenting prenatally. Due to the skewing of existing reference data sets to adult profiles of gene function and expression, interpreting and classifying genetic variants in a prenatal versus postnatal setting is often more challenging due to the current limitations of in utero phenotyping 21,33,34 and availability of appropriate databases and guidelines for fetal and neonatal classifications 21,35 . Traditional standards for variant classification 22 can also be difficult to apply as gestational age (and development) strongly influences whether the representative fetal organ(s) are sufficiently developed to display the characteristic phenotypic features associated with known genetic postnatal disorders 36 with limited databases and guidelines on the classification of fetal and neonatal phenotypes at specific gestational ages 21,35.
In this study, trio/quad ES or GS was prospectively applied for 200 consecutively referred families with perinatal death due to congenital abnormality or in which death was completely unexplained. All cases had undergone standard-of-care full autopsy investigations, and results from microarray and candidate gene testing were negative or non-contributory. Following Mendeliome analysis (n=42) and follow-up research (n=9), a (likely) pathogenic variant in a known gene was identified for 25.5% (51/200) of families and a candidate (VUS/GUS) identified in a further 27 % (54/200), demonstrating the clinical utility of this approach over conventional testing (Figure 1, Figure 2; Supplementary Table 2 and 3). Despite prior molecular investigation of suspected disorders based on phenotype, 60% (63/105) of (candidate) diagnoses were made in known OMIM disease genes, with retrospective phenotype review revealing significant overlap between the proband and the reported phenotype. A further 13.5% (14/104) of (candidate) diagnoses represented novel disease phenotype associations with known disease genes, and the remaining 26.9% (28/104) were made in potentially novel disease genes (Supplementary Table 2). anatomy
Within our cohort, the phenotypic distribution ranged from severe congenital abnormalities to unexplained fetal loss. Compared to the overall distribution of clinical subtypes of perinatal death11, the overall cohort seems skewed towards fetuses with congenital abnormalities and the reported major organ systems (Figure 2, supp Figure 10) are enriched for essential organs detected on early fetal ultrasounds (i.e. brain, urogenital, skeletal and cardiovascular). After excluding cases that were diagnosed by the SNP arrays or gene panels that are part of standard-of-care in Australia, the genomic analyses also indicated a large heterogeneity in the genetic defects underlying perinatal death. Seven recurrently mutated genes (FGFR2, KMT2D, ARSL, TUB1A1, NIPBL, USP9X, and ACTA1) were the likely cause of perinatal death in fourteen families, with variants in the remaining 96 genes being observed only once in our cohort (Supplementary Table 2). While dual diagnoses are expected to contribute to (phenotype expansions of) developmental disorders in 5% of cases 37, reportable candidate variants in two genes were identified in only three families without phenotype expansions. In addition, candidates with possible digenic inheritance were detected in two of the families with a specific clinical presentation.
As observed in other studies on non-consanguineous cohorts of developmental disorders 28, the majority (58.8%; 30/51) of variants leading to definitive diagnoses occurred de novo in the proband. In contrast, the majority (62.9%; 34/54) of candidate variants are autosomal or X-linked recessively inherited variants. The different distribution in causative and candidate variants can (partially) be explained by the additional weight of the criteria for de novo variants (PS2) in the ACMG guidelines 22. Despite studies showing that parental (post-zygotic) mosaicism is an important source of germline ‘de novo' variants in offspring, parents are currently counselled that a recurrence risk for de novo variants is ~1%. As expected based on earlier studies, the phasing experiments for de novo variants showed that ~75% of autosomal de novo variants occurred on the paternal allele. Systematic follow-up of de novo variants by ddPCR revealed that four parents were low-level mosaic in blood (and sperm) for the de novo variant, leaving 39 ‘true’ de novo (candidate) variants, which have likely arisen after gametogenesis. Two paternal mosaic variants indicated a higher recurrence risk compared to the 1% that is commonly counselled in the clinic; with 3% (PED084) and 20% (PED043) respectively 26.
As could be expected, families with informed or measured relatedness seem more likely to yield a (candidate) genetic diagnosis compared to families that are unrelated; 75% compared to 50.5% (not significant) (Supplementary Figure 9). Surprisingly, we identified more (likely) pathogenic variants in female probands (32%; 31/97) compared to male probands (19.4%; 20/103) (Supplementary Table 1 and 2). This may partially be explained by the higher proportion of more mature females recruited (34 females recruited in the last trimester of the pregnancy compared to 22 males). Fetuses of greater maturity are more likely to have a phenotype that is recognisable and more comparable to the phenotypic features in a young infant, for example, and are therefore easier to classify. There is also a higher number of de novo mutations in known disease genes in females (20) compared to male (10) probands in the cohort. Finally, of the 21 quads, in 14 families both siblings were male (66.7%) compared to five families with siblings of both genders (23.8%) and two (9.5%) with both females (Supplementary Figure 10). The number of (candidate) genetic findings from families analysed as quads is remarkably low, indicating that the likely autosomal or X-linked recessive genetic variants were not detected by ES or GS.
Following Mendeliome analysis alone 74.5% (149/200) of families did not receive a definitive diagnosis, of which 47.5% (95/200) remained without a selected candidate for follow-up. The lowest diagnostic yield was observed for cases of perinatal death without congenital abnormalities, of which 91.7% (11/12) remained without a diagnosis, and 66%; (8/12) are unresolved without a candidate (Figure 2, Supplementary Table 2, Supp Fig 1). Notably, there were no LP/P variants identified in previously reported “sudden death” genes in this cohort (e.g. long QT syndrome), despite previous reports suggesting that variants in these genes contribute to unexplained stillbirths 38,39. Overall, the placental genome remains an under-explored area in the understanding of unexplained perinatal death and is an ongoing focus of our research 40. While the MCA pinpointed demographic characteristics that influence the diagnostic yields, larger datasets are required to reach statistical significance for these analyses.
Considering the unique perinatal death cohort described in this study, and the high proportion (24%; 23/96) of diagnoses and candidate variants in (potentially) novel disease genes, follow-up research will remain an important adjunct to clinical genomic analyses. This research is required to characterise all fetal lethal developmental disorders resulting from genetic variants in genes intolerant to variation (i.e. the intolerome) 32. These candidate variants in potentially novel disease genes were prioritised by gene constraint scores and phenotypic overlap (e.g. early lethality) observed in mouse knock-out models and/or similar presentations observed by gene matching approaches (Supplementary Table 6).
The phenotype expansions observed in our cohort include severe prenatal presentations of postnatal disorders, additional clinical manifestations within the same organ system, and other affected organ systems, of which the latter may also be explained by other genetic factors that were missed in our analysis. Interestingly, phenotype expansions were mostly observed in patients with autosomal recessive variants (n=8) compared to autosomal dominant (n=3) and X-linked recessive (n=2) variants (Supplementary Table 2). Generally, the phenotypic presentations in our cohort were more severe compared to what is reported in the literature (Supplementary Table 2; PED012, PED042, PED043)26,41, which may be explained by the resulting variant impact, i.e complete loss-of-function versus reported missense variants which may be partial LoF or hypomorphic variants.
Our results strongly support a clinical role for genomic testing in elucidating the cause of perinatal death, particularly when congenital abnormalities are present. In 21% of cases, genomic autopsy provided a clear diagnosis where standard autopsy could not. Despite this, standard autopsy remains a valued assessment of perinatal death due to its ability to identify non-genetic cause(s) of death (e.g., congenital abnormalities of diabetic embryopathy or evidence of CMV infection), and the advantage of utilising anatomical and histological information obtained at autopsy for the interpretation of candidate variants. Therefore a genomic autopsy is best implemented alongside current standard-of-care measures to help improve diagnostic rates for perinatal death.
In 43 families, the identified (candidate) diagnoses were de novo variants without significant (>1% allele balance) parental mosaicism, and the minimal associated recurrence risk comforted couples for their future pregnancies. In addition to these 43 families, an inherited (candidate) diagnosis was identified in 62 families. Of these, autosomal recessively inherited (candidate) diagnoses were identified in 43 families and X-linked recessive variants in an additional 9; giving a 25% recurrence risk. Indeed, 22.5% (45/200) of families in this study had experienced recurrent perinatal death due to congenital abnormality or unexplained cause (Supplementary Table 1 and 2). Following genomic analysis, a (candidate) diagnosis was identified for 49% (22/45) of these recurrent families, providing options to reduce the risk of further recurrences in future pregnancies.
During the course of our study, at least 24 couples conceived a subsequent pregnancy prior to receiving a result from genomic analysis, providing impetus to reduce turnaround times to improve clinical utility. Driven by this knowledge, and our aim to maximise the impact of our study for clinical care and decision making, we have adjusted our workflows and protocols to provide clinically accredited reports within three months (Figure 3).
Following genomic analysis, 10 families are known to have used the information for reproductive planning, with 5 electing for PGD and 5 electing for PND, facilitating 10 healthy pregnancies and 2 early detections of recurrence (PND) (Table 1). This outcome also highlights the importance of equitable access to high quality assisted reproductive services. Interestingly, one family in this study chose to use PND (PED040, 2 pregnancies) based solely on a candidate variant (VUS)42, resulting in the birth of 2 unaffected children (Table 1). A particularly impressive example of the utility of a molecular diagnosis though, is family PED005 who after 4 consecutive pregnancies affected with Meckel Syndrome were able to use PGD to facilitate the birth of two unaffected children from two separate cycles (Table 1).
We conclude that clinical implementation of a genomic autopsy as part of the current standard-of-care in the investigation of perinatal death would help to reduce the incidence of stillbirth and newborn death by providing options to prevent recurrence. However, an integrated clinical-diagnostic-research setting is beneficial since perinatal death represents an understudied cohort and genetic causes of early lethality remain to be discovered.