Portal de Periódicos da CAPES

In case you missed it: The Prenatal Diagnosis editors bring you the most significant advances of 2019

2019; Wiley; Volume: 40; Issue: 3 Linguagem: Inglês

10.1002/pd.5632

1097-0223

Lyn S. Chitty, Lisa Hui, Alessandro Ghidini, Brynn Levy, Jan Deprest, Tim Van Mieghem, D. W. Bianchi,

Fetal and Pediatric Neurological Disorders

This year, the editors of Prenatal Diagnosis met on a damp autumn day in rural England to make plans for the Journal's next year (Figure 1). Following our now well-established tradition, we identified the "hot" topics relevant to prenatal diagnosis that had emerged over the past year to collate into a brief review to help our readers in case you had missed it. This year we cover developments that may change our clinical practice, including fragmentomics (a biological feature of cell-free DNA); non-invasive prenatal diagnosis (NIPD) for monogenic disorders; artificial intelligence and machine learning; the prenatal phenotypes emerging with increased use of fetal exome sequencing; recent developments in fetal surgery; and the artificial placenta. These are all areas that are set to change the way we deliver prenatal diagnosis and obstetric care. We hope you enjoy the read. The biological phenomenon of circulating cell-free nucleic acids has created enormous opportunities for groundbreaking discoveries in clinical medicine. The field of prenatal diagnosis has been the first to successfully translate cell-free DNA (cfDNA) into wide-scale clinical practice with non-invasive prenatal testing (NIPT). We know that the fetal portion of cfDNA in maternal plasma comes from the cytotrophoblast1 and that, in most circumstances, the maternal haematopoietic cells contribute the majority of the circulating cfDNA.2 Despite the enormous progress made during the first two decades of cfDNA, a major remaining challenge is how to determine the fetal origin of cfDNA fragments without relying on unique fetal sequences, such as Y-chromosome specific sequences or polymorphisms. Being able to measure the relative proportion of placental and maternal DNA—the so-called fetal fraction (FF)—is now recognised as a crucial quality assurance metric.3, 4 However, laboratories currently use a variety of methods that are not directly comparable, with the Y-chromosome-based method remaining the "gold standard" despite only being applicable to 50% of pregnancies.5 Another challenge related to the problem of reproducibly measuring FF is the determining the tissue of origin of aberrant cfDNA when a source other than the placenta is suspected, for example, where a chaotic plasma genomic profile obtained during NIPT raises suspicion of a maternal neoplasm6, 7 If a suspected tumoural source of cfDNA could be inferred from the maternal plasma cfDNA, it would improve the management of these pregnant women. The emerging knowledge base on the fundamental biology of cfDNA has been called "fragmentomics." Compared to maternal cfDNA, cfDNA of fetal origin is shorter,8, 9 differentially methylated,10, 11 contains a "nucleosome footprint" that reflects a nonrandom fragmentation pattern,12 and has characteristic end sites that differ from maternal fragments.13 In 2019, pioneering researchers at the Li Ka Shing Institute of Health Sciences consolidated these understandings of fetal cfDNA in a publication that opens further opportunities for prenatal diagnosis, oncology, and transplant medicine. In this proof-of-principle study, the authors proposed a new method for nucleosome positioning profiling and quantitative determination of the relative contributions of various tissues in plasma DNA by fragmentation pattern analyses.14 In this elegant paper, the investigators capitalise on the knowledge that cfDNA fragmentation is not random but occurs in tissue-specific patterns that reflect open chromatin regions and nucleosome position and is deducible by sequencing coverage patterns. Fetal and maternal cfDNA fragments also have their own pattern of "preferred" end sites. This results in differences in the read densities of sequences corresponding to the orientation of the upstream and downstream ends of cfDNA molecules in relation to the reference genome ("orientation aware"). When compared with the "gold standard" Y-chromosome-based method of FF measurement from first trimester plasma samples, the orientation-aware cfDNA fragmentation value (OCF) approach showed strong correlation with FF. In the next decade, we should expect an acceleration of clinical applications in cfDNA based on the rapid gains in the new field of fragmentomics. This will not be limited to refining the performance of NIPT through better FF measurement or the occasional individualised analysis of cfDNA samples from women with suspected occult malignancies but has huge diagnostic potential in organ transplant recipients and cancer patients. The cfDNA revolution that occurred in prenatal screening and diagnosis is thus set to expand to other fields in medicine. NIPD for monogenic disorders has been an established clinically accredited service in the UK since 2012 for selected conditions in families at known increased risk, either because of a family history or ultrasound findings suggestive of a monogenic disorder.15 It is less widely used elsewhere in the world, where it is largely delivered on a research basis rather than as an accredited service. This year saw further publications expanding the potential for NIPD for recessive disorders. The first one reported the use of next generation sequencing (NGS) in pregnancies at risk of sickle cell disease.16 One advantage of the approach described is that it does not require the paternal genotype. However, while the sensitivity of this test was 100% in the presence of sufficient FF, performance was less good at lower thresholds. The authors proposed that this test could be used to triage for invasive testing that would only be required to confirm positive results.16 The second paper described the clinical implementation of relative haplotype dosage for the definitive NIPD of cystic fibrosis, with an average turnaround time of 5.75 days and a low inconclusive rate. These authors reported excellent performance, but it has the disadvantage of requiring samples from both parents and an affected child or sibling, as it is based on a linkage approach.17 Both of these papers described NIPD in pregnancies at known increased risk; however, 2019 saw the publication of two papers describing the screening of low-risk pregnancies for monogenic disorders.18, 19 These papers have been used to support at least three commercial companies launching platforms publicly available for screening pregnancies at general risk for monogenic disorders. The first report uses an NGS approach to screen 422 pregnancies for pathogenic or likely pathogenic variants in 30 genes associated with dominant monogenic disorders that can lead to significant adverse outcomes, including skeletal dysplasias, Noonan spectrum disorders, Rett syndrome and many others.18 Positive results were reported in 35 cases, of which 31 had an abnormal ultrasound scan and four a positive paternal history. Follow-up was only available in 147 of the 422 pregnancies tested (34.8%), which confirmed 20 true positives and 127 true negatives. The second publication also used an NGS-based method for molecular counting to screen for sickle cell disease, cystic fibrosis, spinal muscular atrophy, alpha-thalassemia, and beta-thalassemia.19 The approach used in this report was to first screen the pregnant mothers' germline DNA, and the cfDNA in maternal plasma is reflex tested in mothers subsequently identified as carriers to determine the fetal genotype. Much of the data reported in this paper comes from modelling experiments using spiked-in genomic DNA, which may be a useful starting point for assay development but, as discussed above, we know that fetal cfDNA is different from maternal cfDNA or postnatal DNA.14 There appears to be a high inconclusive rate in this study, although the FF was reported at greater than 10% in most cases. Of note, none of the genotypes reported included pathogenic mutations. Furthermore, most of the genes included in this study have multiple pathogenic mutations, and it is unclear how many of these were included in the assay described. These papers published in 2019 clearly indicate that NIPD for monogenic disorders is being extended to both cover more conditions but also to reach out to families at general risk through commercial platforms. The evolution of NIPD in high-risk pregnancies to include more conditions is clearly needed, as many families are at significant risk, and we know that both parents and health professionals value safer prenatal diagnosis. However, any extension must be accompanied by robust validation. New dominant mutations occur in around 1 in 600 individuals and account for significant mortality and morbidity, potentially justifying screening in low-risk pregnancies. The recessive conditions described in the paper by Tsao et al are all serious, with some amenable to neonatal therapy, and perhaps soon in-utero therapy will be available, again potentially justifying attempts for better screening and diagnosis. However, it is of concern to see commercial tests launched with relatively little validation, including in one instance no pathogenic cases. Furthermore, as we move into screening low-risk pregnancies, we need to understand what parents want and indeed where they perceive the limit for testing to be. How can we counsel them appropriately to ensure informed parental decision making? How do we explain the uncertainty that accompanies some diagnoses, and who is going to do the counselling? What is certain is that we are likely to see more publications in 2020 and beyond describing screening for monogenic disorders using analyses of cfDNA. We must hope that commercial companies will adhere to the strict standards required for accredited laboratories around the world before launching tests that may raise unrealistic expectations and require significant pretest and posttest counselling. Artificial intelligence (AI) and machine learning are emerging technologies that have already had major influences on our daily lives (eg, personalised news feeds, online shopping algorithms, and robots judging gymnastic competitions). Deep learning is a subcategory of machine learning; it is based on analyses of actual data and has the capacity to continuously improve its performance with additional data and feedback loops. During 2019, papers began to appear that applied this technology to areas of interest to readers of Prenatal Diagnosis, including improving blastocyst selection for transfer,20, 21 analysis of placental volume,22 analysis of multi-omics to predict risk of preterm delivery,23 and reducing the time for analysis and reporting of newborn genome sequencing data.24 A critical need exists to improve embryo selection in order to increase the chance of a live birth following fertility treatment. Time-lapse video imaging is a new technology that allows embryos to remain in culture under stable temperatures and metabolic conditions while being continuously observed and recorded.20 In two recent studies, investigators hypothesised that artificial intelligence could be used to objectively analyse tens of thousands of video images, thereby improving the approach to selection of the best quality embryos. In the first, expert embryologists curated a high quality data set of developing embryos that was used to train the system.20 The trained algorithm could distinguish between good and poor quality embryos with 96.4% accuracy. However, this algorithm could not successfully distinguish between those embryos likely to result in a live birth and those that would not. By contrast, another group performed a retrospective analysis of videos obtained from 10 638 fresh and vitrified embryos to develop a model of predicting pregnancy to the fetal heart beat stage.21 The diagnostic accuracy of the model, as measured by average area under the curve (AUC), was 0.93. This promising deep learning approach does not take into account traditional ways of assessing embryos, such as morphology grade, biologic, or metabolic activity. However, it has the potential to improve stability of embryo culture, laboratory efficiency, and ultimately, rates of healthy live births. As an organ that is critical for fetal growth and development, it seems that the placenta would be an excellent target for the development of machine learning tools for the early prediction of fetal growth restriction. Looney et al at Oxford used deep learning to develop a fully automated three-dimensional ultrasound (3D-US) segmentation technique to estimate placental volume in the first trimester. The system was used to generate first trimester placental volumes, which correlated with infant birth weights at term. While the estimated detection rates of small for gestational age babies (23%) increased over prior nonautomated studies, the system is still not good enough to use as a routine screening tool.22 Prediction of the risk of preterm birth is another major goal of obstetrics. Short cervical length is currently the best available marker, but it has a low positive predictive value. In a recent study, investigators tested the hypothesis that adding metabolomics, proteomics, cytokine analysis, demographic, and clinical data to transvaginal sonographic measurements demonstrating a cervical length of less than 15 mm would do a better job of predicting those pregnancies at risk for preterm birth.23 This was a small, proof-of-principle study that suggested that adding multiple layers of clinical and laboratory data was better than using sonographic data alone. Larger studies are needed to validate these preliminary results. As prenatal genome-wide sequencing gradually becomes incorporated into clinical care,25-27 practical lessons can be learned from the emerging approaches being used in intensive care units.24 In prior paediatric clinical sequencing studies, the mean time from exome sequencing to diagnosis was 16 days. In April, however, a team from Rady Children's Hospital in San Diego incorporated rapid automated DNA sequencing, deep phenotyping from the electronic health record (EHR), and AI to automatically annotate and interpret sequencing results. The programme was able to prospectively diagnose clinical conditions in three of seven neonatal intensive care unit patients in a median time of 20 hour; treatment changed in all three as a result of the diagnosis. Automated genome analysis and interpretation has the potential to increase widespread adoption of prenatal and paediatric genomic medicine, areas in which there are shortages of experts available and analysis is labour intensive. A major hurdle to be overcome in prenatal medicine will be the gaps in knowledge regarding the fetal clinical presentations of monogenic diseases. Fortunately, with the increased prenatal utilisation of chromosome microarrays and exome sequencing, more cases are being described in the literature that demonstrate new knowledge regarding fetal phenotypes.26, 28, 29 This should help to facilitate automated prenatal diagnoses in the future, reducing turnaround time and expanding management options. This past year, we have seen a significant increase in the number of sequencing reports focused on fetuses with anomalies. In particular, two large prospective cohort studies, one from the United States and one from the UK (the PAGE study), reported on trio whole exome sequencing (WES) in unselected pregnancies with fetal structural abnormalities.30, 31 Both studies showed an overall added diagnostic yield of about 10% with higher yields in fetuses with multisystem anomalies and in selected anatomical systems, for example, skeletal.30, 31 Of great interest is the observation that "expanded" phenotypes are being encountered in fetuses who are being diagnosed using next generation sequencing methodologies. Since the clinical descriptions of most genetic diseases derive from assessment of paediatric and adult patients presenting with medical issues, we are only now beginning to appreciate how those same genetic diseases may present in the prenatal period. For example, Meier et al identified a fetus with a known pathogenic de novo variant in FGFR2, which is typically associated with Apert syndrome.32 The fetus presented with agenesis of the corpus callosum and bilateral syndactyly of the hands and feet but no sign of craniosynostosis, which is a hallmark feature of Apert syndrome. Ferretti et al. described a similar situation where a fetus with poor intrauterine growth and a pathogenic missense variant in NSD1 only demonstrated classic signs of Sotos syndrome at 1 year of age.33 Similarly, autosomal dominant mutations in RERE are associated with a neurodevelopmental disorder with or without anomalies of the brain, eye, or heart. This was only clinically evident in a 6-month old child followed-up after a prenatal WES finding of a de novo frameshift variant in RERE.31 In this case, the only prenatal finding was an isolated nuchal translucency (NT) of 3.5 mm. In some instances, the fetus may "grow into the phenotype" during the pregnancy as reported by Arora et al who described a fetus with an NT that increased from 2.9 mm at 12 weeks to 7 mm at 17 weeks.34 Additional abnormalities were noted at 28 weeks and included moderate unilateral ventriculomegaly (14 mm), Dandy-Walker malformation, and multiple joint contractures (later shown to be arthrogryposis multiplex congenita).34 The fetus expired 2 days post birth, and WES revealed an inherited homozygous novel splice site variant in COG8, a gene associated with a congenital disorder of glycosylation. Another example of an expanded fetal phenotype comes from De Graer et al who reported novel features (diaphragmatic eventration and duplication of the distal part of the small bowel) of the PIK3CA-related overgrowth spectrum (PROS).35 As new fetal phenotypes are reported in association with a variety of novel mutations, we need to ensure that enough evidence exists before causally linking a specific variant with specific fetal anomalies. Such evidence usually stems from functional and animal model studies as well as the enrichment of the variant in unrelated fetuses with concordant phenotypes and the absence of the variant in normal individuals.32 Current clinical practice in postnatal patients dictates that, among other things, detailed phenotyping is necessary for accurate interpretation of sequence variants revealed by WES. This presents a major challenge during pregnancy as the anomalies that we expect in paediatric or adult patients may not be apparent (eg, seizures and intellectual disability) or may not yet have manifested at the gestational period when the fetus is assessed. Furthermore, subtle or different features may be discernible in the prenatal period, which may be different to those observed in the postnatal stage.32 Large-scale efforts to generate public databases with well-curated fetal phenotypes and their associated WES findings will greatly enhance our ability to provide meaningful WES results to patients identified with fetal anomalies. Fetal therapy contributes significantly to the literature, and this year we highlight maternal issues as well as technological developments that could enhance our ability to deliver in-utero therapy. In 2019, two papers described the potential maternal complications of fetal surgery.36, 37 The first was a systematic review of 166 studies reporting either open or endoscopic in-utero surgery. This showed that the risk for any maternal complication was higher (20.9%) for open fetal surgery compared to endoscopic surgery (6.2%). When considering only severe complications, the difference was smaller: 4.5% for open and 1.7% for endoscopic surgery.36 However, in this review, around a quarter of studies reviewed were excluded as they failed to comment on maternal outcomes. Of note, this review also showed that open fetal surgery increased the risk of preterm labour in subsequent pregnancies but not uterine dehiscence or rupture. This is in contrast to the second study that reported maternal outcomes for 77 women undergoing hysterotomy for fetal spina bifida repair. Here they showed that the risk for rupture (9.6%) and perinatal death was similar to that seen after classical caesarean section.37 This is an important driver in the further development of endoscopic techniques. The other maternal issue being addressed is whether and how human immunodeficiency virus (HIV) or viral hepatitis seropositive mothers can be offered fetal surgery, with regard to the risk of intraoperative viral transmission to the fetus. Shamshirsaz et al propose an ethical framework to extend inclusion criteria for fetal surgery to those pregnant patients with HIV and hepatitis B or C infection with low or undetectable viral loads.38 Moehrlen et al report the first successful fetal surgery for fetal spina bifida with passive in-utero HBV vaccination.39 For complex surgeries, like spina bifida repair, fetoscopy requires amnio-insufflation, but the use of dry carbon dioxide has historically been questioned because it causes fetal acidosis. Recent work demonstrated that this can be mitigated by maternal hyperventilation, along with heating and humidifying the gas.40, 41 Dry CO2 also has an effect on the fetal membranes but, based on biomechanical studies published this year, heating and humidifying the gas may also reduce the risk for membrane rupture.42, 43 While the debate continues over whether or not minimally invasive surgery is equally neuroprotective as open surgery and which surgical techniques have the least prematurity risk,44-46 it is worth considering the experience required to complete the learning curve for prenatal spina bifida repair. Centres performing high volumes of in-utero surgery report that it takes up to 30 (open) or 60 (fetoscopy) procedures to become proficient.47 This raises questions about the viability of smaller programmes, and emphasises the need for appropriate high-fidelity training models for this (and other) fetal procedures.48, 49 Imaging, and increasingly MRI, is key in selecting cases suitable for fetal therapy as it is used for planning, simulating, and optimising fetal surgery. An early application is likely to be in the treatment of twin-to-twin transfusion syndrome where a combination of Doppler and MRI are used to determine placenta and vascular structure,50 and, using computer vision and deep learning techniques, the ideal entry point and the laser trajectory can be determined.51 New methods such as photoacoustic imaging, which can be incorporated into current fetoscopes, can also be used to enhance the identification of placental vessels to improve success of ablation.52 Thus, while we have seen advances in data collection to demonstrate likely maternal complication rates, more needs to be done in this area to help determine the optimal approaches, including long-term outcome studies. In addition, as we see new technologies being applied, we must ensure good, (randomised) prospective trials are performed to obtain objective evidence of the best approaches to use. We have chosen to feature the artificial placenta in this year's "In case you missed it" as a few important manuscripts describing the methodology and physiology of this technology were published in 2019, bringing it closer to clinical translation. Creating an artificial placental and uterine environment as a life-support device for extremely preterm neonates has been an endeavour since the 1960s. It is now 2 years since the first successful attempts at keeping severely preterm lambs alive for up to 28 days in an artificial uterine environment were reported.53 In the last year, significant further progress has been made in both demonstrating the physiological effects of the artificial environment and the feasibility in younger animals. In 2019, several groups have shown that important components of fetal physiology are maintained on artificial placental support and develop more naturally than in a ventilated newborn. Stable cardiovascular physiology has been shown both in short- and in long-term experiments using a pumpless system.54, 55 Even after longer term support, animals can be transitioned to normal care53 and maintain normal behaviour and mitochondrial respiration.56 Previous experiments with a pump-driven system had already shown that fetal splenic development57 and cerebral oxygenation can be maintained58 and that white matter integrity is preserved with this type of life support.59 Even though still limited, these experiments give preliminary support for the safety of the artificial placenta. Further supporting a translation to the human setting, the group at the Children's Hospital Of Philadelphia (CHOP) has now demonstrated that fetal umbilical catheterisation can also successfully be achieved in significantly younger lambs (85 to 96-day gestation) than previously reported. In addition, the research group in Perth has confirmed the feasibility of a pumpless system that can be applied to lambs at 95-day gestation and weights of 600 g. This is comparable to the 23-week-old human fetus, which would be the clinical target population.60 Finally, early evidence with artificial placental systems has increased our understanding of normal fetal physiology, for example, the importance of erythropoietin to support red blood cell development and prevent fetal anaemia (and subsequent need for transfusions).61 Experiments with chronic hypoxemia in this system have advanced our knowledge of the pathophysiology of this condition.62 Over the next 5 to 10 years, we expect further dissemination of the artificial placenta system throughout research labs worldwide, providing us with more information on the physiology and pathophysiology of different conditions. Major hurdles still need to be overcome before this technology can translate to clinical care, but, if it does, this would be a gigantic leap forward in the care of extremely preterm newborns. Additionally, and of relevance to the readers of Prenatal Diagnosis, the ex-utero fetal environment could become an ideal platform for fetal intervention, as limitations due to preterm birth or transmaternal access would be easily overcome. The views expressed in this article are those of the author(s) and not necessarily those of the NHS, the NIHR, or the Department of Health. LSC is partially funded by the NIHR Biomedical Research Centre at 84 Great Ormond Street Hospital. The views expressed in this article are those of the author(s) and not 80 necessarily those of the NHS, the NIHR, or the Department of Health. N/A