The Role of Genome Sequencing in the NICU

Abstract

Genetic diseases test the functionality of an infant’s genome during fetal-neonatal adaptation and represent a leading cause of neonatal and infant mortality in the US. Due to disease acuity, gene locus and allelic heterogeneity, and overlapping/diverse clinical phenotypes, diagnostic genome sequencing in neonatal intensive care units has required development of methods to shorten turnaround time and to improve genomic interpretation. Between 2012 and 2021, 33 clinical studies have documented diagnostic and clinical utility of first-tier, rapid or ultra-rapid whole genome sequencing through cost-effective identification of pathogenic genomic variants which change medical management, suggest new therapeutic strategies, and refine prognosis. Genomic diagnosis also permits prediction of reproductive recurrence risk for parents and surviving probands. Using implementation science and quality improvement, deployment of a Genomic Learning Healthcare System will contribute to reduction of neonatal and infant mortality through integration of genome sequencing into best practice neonatal intensive care.

Introduction: the unique genomics of neonatology

The perinatal period, defined as six months before to 12 months after birth, is unique in terms of healthcare consequences across the lifetime. During this period, the functionality of a child’s genome is tested by the biologic requirements of fetal-neonatal transition and extra-uterine physiologic adaptation. During pregnancy and the immediate post-partum period, many fetal physiologic processes are provided, or compensated for, by the maternal or placental genome. In addition, genome-encoded developmental regulation continues to activate and silence genes and gene pathways throughout the first months and years of life. For example, drug metabolism is quite different in newborns and older children due to lack of expression of the key metabolic enzyme cytochrome P450 2D6 (CYP2D6) (92). Finally, delivery of viable fetuses may occur at diverse, post-conception time points. Thus, in caring for newborns, neonatologists must consider age both as days since conception and days since delivery. Much remains to be learned about the factors that regulate the ontogeny of gene expression and silencing associated with neonatal and gestational ages (92). Thus, genetic, genomic, and functional genomic regulation of fetal development and neonatal adaptation to the extra-uterine environment is uniquely challenging in three ways. Firstly, the timing of onset of monogenic disease is regulated by the continuum of causal gene expression in the fetus or newborn during the perinatal period as suggested by examples of genetic disruption of liver and cerebral function (68, 88). Secondly, the timing of onset of monogenic disease is influenced by the continuum of compensatory effects of maternal or placental genes. For example, in the setting of primary neonatal immunodeficiency, maternal immunoglobulins may protect a newborn from infections for months (19, 95). Thirdly, the adaptive physiologic changes that occur at delivery immediately expose latent genetic defects. Third trimester echocardiography in fetuses with congenital heart disease, for example, can define cardiac anatomy but may not be able to predict neonatal cardiac adaptation to extra-uterine life.

While we can now routinely decode and analyze a human genome in a day, the contributions of many genetic defects to disease penetrance, severity, onset, complications, response to treatment, and outcomes need to be defined to permit clinically actionable genomic results and remain under active investigation (40)(75). Similar to comparing a perfect space rocket at t minus 5 minutes for launch and t plus five minutes, the phenotypically normal fetus may develop life-threatening genetic disease in the neonatal period whose treatment and prognosis may be difficult to predict based on fetal phenotype. For example, knowledge of the underlying genetics in developmental epileptic encephalopathies does not necessarily translate today into clinically actionable guidance for treatment or prognosis. While we have identified genetic variants associated with >7,000 genetic diseases, for a majority, we do not yet understand the encoded mechanisms of action. Another challenge to use of genomic information for sick infants is that different variants in a single gene may be associated with different disease phenotypes (76), and similar disease phenotypes can be associated with different genes (55). The former is particularly perplexing. Variants in a voltage-gated potassium channel gene, KCNQ2, for example, map to both Developmental and epileptic encephalopathy type 7 (DEE7, Mendelian Inheritance in Man (MIM) #613720) and Benign neonatal seizures type 1 (BNS1, MIM#12100, also known as Myokimia). The treatment and prognosis of these two disorders are quite different. BNS1 presents with seizure onset at 2-8 days of life, seizures respond to first-line antiepileptic medications, such as phenobarbital, and most remit by 6 weeks. DEE7 has similar age of seizure onset but requires much more aggressive antiepileptic therapy and is associated with developmental delay and intellectual disability. In summary, while the greatest potential contribution of genome sequencing to human health is currently for newborns, the unique physiologic and genomic characteristics of the newborn plus our incomplete understanding of encoded mechanisms of action, treatment guidance, and prognosis present bottlenecks to achieving the full potential of timely genetic diagnosis for reducing fetal and neonatal morbidity and mortality. However, the importance of the contributions of genetic disease to neonatal and infant mortality has accelerated study and integration of genome sequencing into the neonatal intensive care unit (NICU) to improve infant outcomes.

The value of genome sequencing in the NICU –

Although many of the physiologically supportive clinical practices of neonatal intensive care help to improve survival of critically ill infants with genetic diseases, many affected infants have unique disease mechanisms associated with pathogenic genomic variants that are rare or novel due to reduced reproductive fitness (49) and that disrupt function and/or anatomy of multiple organs (91). An evidence-based clinical practice guideline from the American College of Medical Genetics and Genomics (ACMG) supports the clinical utility and desirable effects of whole exome (WES) and whole genome sequencing (WGS) on active and long-term clinical management for pediatric patients <1 year of age with 1 or more congenital anomalies (56). NICUs have recognized the critical need for diagnostic discovery of genomic causes of newborn diseases to individualize treatment, establish prognosis, predict reproductive recurrence risk, and improve infant outcomes (44, 106). Leveraging collaborations with geneticists, genetic counselors, obstetricians, genomicists, and bioinformaticians, advances in genome sequencing technology (especially reduction in turnaround time (75)), computational strategies for identifying pathogenic variants (61), and reduced sequencing costs, NICUs have begun transitioning from a phenotype first to a genotype first approach for genetic diagnosis in critically ill infants by integrating rapid trio WES and WGS into early diagnostic testing that will permit faster diagnosis, more precise prognosis, and rapid identification of individualized treatment options (43, 75). In addition, genotype first diagnosis provides families estimates of reproductive recurrence risk for future pregnancies and for the surviving proband and siblings. To illustrate the diagnostic and therapeutic value of WES and WGS for NICU patients with congenital anomalies, we will briefly describe an example, epileptic encephalopathy due to thiamine metabolism dysfunction syndrome.

Epileptic encephalopathy due to thiamine metabolism dysfunction syndrome –

Although infections, trauma, and environmental factors are associated with epileptic encephalopathies, genetic etiologies contribute significantly to their frequency and pathogenesis (6, 93). Due to purifying selection, genetic causes are heterogeneous. Discovery of a monogenic etiology in an infant with urWGS can inform precision therapy as illustrated by a recently reported case (75). Briefly, a 5-week-old, previously healthy male infant of consanguineous parents whose reproductive history included a previous child who expired with epileptic encephalopathy presented with epileptic encephalopathy. Ultra-rapid (within 16.5 hours) WGS revealed homozygous, known pathogenic (ClinVar VCV000533549 .2) frameshift variants in SLC19A3, a thiamine transporter known to be a monogenic cause of thiamine metabolism dysfunction syndrome 2 (THMD2) (MIM#607483). Based on these findings, treatment with thiamine and biotin was initiated with almost immediate resolution of seizures. At 7 months of age, the infant is reportedly thriving. Untreated THMD2 leads to rapid neurologic deterioration and death (52). This patient illustrates the importance of integrating rapid- or ultra-rapid-WGS for critically ill infants with congenital anomalies that include epileptic encephalopathies, metabolic disorders, and structural birth defects.

Integrating Genome Sequencing in Neonatal Intensive Care Units

The opportunity for genetics and genomics to reduce neonatal morbidity and mortality has occurred over the last 2 decades after neonatal intensive care units (NICUs) developed life-sustaining obstetric, surgical, and medical strategies for infants born at the edge of viability, with life-limiting congenital anomalies, with neonatal infections, and with severe intrapartum insults. In the early 20^th century, newborns were delivered at home and survived if they were able to establish respiration, maintain oral intake, and avoid infection. Between 1930 and 1950, deliveries shifted from homes to hospitals to gain access to maternal anesthesia and analgesia. Hospital staff and families recognized that survival of premature infants could be significantly improved by providing thermoregulation, oxygen, hand hygiene, and more intensive physiologic monitoring. This recognition prompted the establishment of the first NICUs in the 1960s. With increased neonatal specialization and favorable reimbursement, rapid expansion of the number of NICUs occurred over the past 3 decades to about 800 units in the United States (30). Concurrent with broad deployment and refinement of neonatal intensive care, the neonatal mortality rate (death within the first 28 days of life) has fallen from 18.7 (1960) to 3.8 (2018) per 1,000 live births. This survival improvement is attributable to collaboration of multidisciplinary teams of nurses, physicians (neonatologists, obstetricians, anesthesiologists, and pediatric subspecialists), respiratory, occupational, and physical therapists, social workers, pharmacists, and clinical investigators focused on the unique physiologic challenges of extra-uterine adaptation. These teams developed, implemented, and deployed quality improvement frameworks with specific care practices, interventions, and quantitative clinical guidelines specific for the newborn’s unique transitional physiology and disease susceptibilities. For example, interventions to improve pulmonary outcomes for premature infants (e.g., antenatal glucocorticoid administration, surfactant replacement therapy, and less invasive ventilation strategies) combined to reduce a previously common cause of neonatal mortality in preterm infants, respiratory distress syndrome, from 20% (1980) to 2% (2019) of infant deaths. As a consequence, the current, most frequent, population-based causes of infant deaths in the United States based on birth certificate data are genetic, and include congenital malformations, deformations, and chromosomal abnormalities, which accounted for 20.6% of 20,921 total infant deaths in the United States in 2019 (2, 36). Many, but not all, of these are caused by single locus genetic diseases. Approximately 10% of the US annual birth cohort or ~400,000 newborns are now admitted to NICUs, costing at least $26 billion (2007 dollars) and accounting for up to 50% of the total United States pediatric health care expenditure (34). In addition, 10% to 25% of critically ill infants in NICUs are estimated to have an undiagnosed single gene locus disorder which may be missed due to the nonspecific presentations of many genetic diseases in the newborn period and to limitations in reimbursement for comprehensive inpatient genetic testing (35, 58, 105, 106).

Prior to availability of genome sequencing in the NICU, genetic diagnosis in critically ill newborn infants relied on both a prenatal and postnatal phenotype first approach. Strategies to assess fetal phenotype, including ultrasound, maternal biochemical testing, and fetal magnetic resonance imaging, provide fetal characteristics that may not be specific to individual genetic diagnoses (35). Historically, fetal and neonatal genetic testing was highly selective and employed chromosomal analysis (karyotyping, fluorescence in situ hybridization, and chromosomal microarray (CMA)) and Sanger sequencing of exons of specific genes. Samples for fetal DNA extraction included chorionic villi, amniocentesis fluid, fetal umbilical blood, or maternal cell free DNA. Diagnoses were limited to aneuploidy, large copy number variants, and recurrent single locus genetic diseases with pathognomonic features (10, 35, 41, 47, 50, 104). Until very recently, there was considerable uncertainty in estimates of the incidence or types of genetic diseases among infants and children hospitalized for critical illness (54, 59). Furthermore, the etiology of common presentations was frequently regarded as polygenic, influenced heavily by the common disease/common variant hypothesis and results of genome-wide association studies. We now know that infant-onset conditions such as epilepsy, developmental delay, intellectual disability, hearing loss, and inflammatory bowel disease are actually each reflective of at least one thousand separate single locus genetic diseases. In addition to a lack of comprehensive testing modalities, patenting of disease genes created barriers to diagnostic testing (39, 46).

Comprehensive genetic testing first became available with the advent of next generation sequencing in 2005 (57). The first individual human genomes were sequenced at a cost of at least a million dollars during the period from 2007 – 2009 (42, 53). This uniquely exciting time in human genetics generated multiple novel realizations. For example, comprehensive sequencing revealed that genomes contained many more variants than suspected (about 6 million per individual). Disease gene discovery was less computationally challenging when trio genome sequencing (parents plus proband) replaced the traditional approach of linkage analysis and positional cloning. An exciting development in 2009 and 2010 was the use of WES and gene panel exome sequencing (71). The exome refers to the collection of all exons of ~20,000 protein-coding genes and comprises about 2% of the genome. While exomes must be sequenced much more deeply than genomes (90-fold versus 20-fold) for comprehensive coverage (because of coverage skewing), they decreased the cost of genome sequencing about tenfold (46). The era of the $1000 research genome led to many discoveries. Almost overnight, human genetics went from being descriptive to quantitative. Advances in bioinformatics facilitated the development of high performance computing strategies necessary for genomic data management and interpretation. We realized that human genomes had many more de novo variants than previously recognized (about 80 per genome), and that resultant dominant disorders contributed the majority of single locus genetic disease diagnoses in outbred populations (48). The success of positional cloning for identification of recessive conditions led to the erroneous conclusion that they were much more common than dominant disorders. The first use of exome sequencing to diagnose genetic diseases was in 2009 (15, 70), and the first gene panel sequencing test for diagnosis and carrier testing was in 2011 (7). Two landmark papers at that time showed the ability of WGS- or WES-based diagnosis dramatically to change childhood outcomes (4, 107). Many physician-scientists who had trained in adult internal medicine believing it to be the most rigorous clinical discipline (for example, one of us (SFK)) suddenly realized that the next decade would be dominated by pediatric molecular discoveries! We realized to our great chagrin that 27% of variants in our databases had been misclassified as pathogenic (7). As a community, we had been “looking under the lamp-post” for our lost diagnostic keys and thereby over-interpreting pathogenicity.

Rapid, diagnostic WGS of infants with suspected genetic diseases in NICUs became possible in 2012 (83). Until that time, time to result by genome or exome sequencing was several months, limiting diagnostic use to outpatients (Figure 1). Faster sequencing and bioinformatics, together with semi-automated diagnostic interpretation, decreased time to genome sequencing result to 50 hours.

Figure 1.

The current inflection point in precision neonatology for single locus genetic diseases. The cost of research-grade genome sequencing (black line) and time to result of diagnostic rWGS (blue line) have been decreasing for ten years. The number of known genetic diseases (yellow line) has increased dramatically since the advent of next generation sequencing. The number of approved genetic therapies for childhood-onset genetic diseases (teal line) is rapidly increasing.

In the past ten years, the turnaround time, scalability, cost, and diagnostic yield of rapid, diagnostic WGS have continued to improve iteratively (Figure 1). Minimum turnaround time is now 13.5 hours (75). WGS is scalable to entire populations (29). Minimum cost of a research genome is now ~$500 (37). Diagnostic capacity has steadily expanded to include disorders of the mitochondrial genome, structural and copy number variants, simple sequence repeat expansions, imprinting disorders, and loci featuring pseudogenes. Sensitivity and specificity for single nucleotide variants now exceed 99.5% (74). Rapid diagnostic exome sequencing and rapid diagnostic gene panel exome sequencing became available shortly after rapid genome sequencing. They remain less expensive than genome sequencing ($2000-$2500 for rapid, diagnostic gene panels, $4000-$5000 for rapid exome sequencing, and $8,000-$10,000 for rapid genome sequencing). They cannot, however, be performed as rapidly as genome sequencing, because they require exon enrichment and amplification steps. While their diagnostic yield has been similar to that of genome sequencing, as our ability to detect pathogenic, non-exonic variants improves, genome sequencing will inevitably become superior.

Rapid Genome Sequencing Technology

The technology, infrastructure, and computational strategies that enable rapid NICU genome sequencing have become highly standardized (Figure 2). Firstly, parental consent is sought. This consent is necessary because of the possibility of harm for the infant. In the United States, most potential harm was mitigated by the Genetic Information Nondiscrimination Act (GINA) of 2008. The scope of testing must also be determined with parental consent. Parent – infant trio testing is faster and has a higher potential diagnostic yield than singleton testing (17). Some parents, however, such as armed forces service members, are not protected by GINA. Parents must also decide about whether they wish incidental findings to be returned (66, 67). These genomic variants are not related to the infant’s current illness but which have significant consequences for future health (67). Finally, there is the option of rapid and ultra-rapid testing. The latter is reserved for the subset of NICU infants in whom time to result is critical, either because of the severity of their illness or because of time constraints of medical decision making or significant interventions (such as starting extra-corporeal membrane oxygenation (ECMO)).

Figure 2.

Steps and minimum times of rapid genetic disease diagnosis by whole genome sequencing and implementation of precision neonatology. EHR, electronic health record.

Diagnostic genome sequencing requires two inputs. Firstly, the clinical features of the child’s illness are extracted from the medical record. This extraction is often performed computationally by natural language processing. The observed clinical features, typically as Human Phenotype Ontology (HPO) terms, are then compared computationally with the expected clinical features of all known genetic diseases to create a quantitatively prioritized, differential diagnosis list. Alternatively, the clinical features are used to select from a set of virtual gene panels. The second input is genome sequence from DNA extracted from 50 μL of peripheral blood. The DNA is then prepared for sequencing, a process called library preparation, which involves random fragmentation into ~500 nucleotide pieces, and attachment of short DNA probes on either end. Genome sequencing is then performed. For exome or gene panel sequencing, two additional steps, namely enrichment of exons and amplification of the remaining material, are performed. Sample preparation from blood sampling to start of sequencing takes from 2 hours to 2 days. The longer preparation time is required for exome and panel sequencing. Sequencing is almost always by synthesis, resulting in a pair of 100 – 150 nucleotide sequences (reads) from either end of each fragment. Genomes are sequenced to at least 30-fold coverage (~100 gigabases (GB) of DNA sequence), while exomes are sequenced to at least 90-fold coverage (~10 GB). Sequencing takes 11 hours – 2 days, depending on instrument settings, and is performed from single samples all the way to 50 samples per run. The remaining steps are computational and either performed in a cloud environment or on local high-performance computing instruments. First, the nucleotide of each position on each read is determined, together with a quality score. In general, quality must be better than 1 error in a 1,000 (Quality (Q) score of >30). Secondly, each pair of sequences is mapped to the corresponding, unique region of the genome (alignment), and a mapping quality score is determined. Thirdly, the aggregate sequence at each position is determined based on the consensus of ~30 reads in genome sequencing. This involves determining the DNA sequence, zygosity, and copy number. All nucleotides and regions that differ from the reference genome are identified (variant calling). In a genome, this is typically over 3.5 - 5 million single nucleotide substitutions (single nucleotide variants, SNVs), 750,000 – 1 million insertions and deletions of size 1 – 50 nucleotides (indels), and 20,000 structural and copy number variants that range in size from 50 nucleotides to entire chromosomes (Table 1). Variants are mapped to genes, genes to diseases, and diseases to the differential diagnosis calculated based on the infant’s clinical features. To diagnose a genetic disease, the search space is almost always limited to those genes. Pathogenicity prediction algorithms are used to predict the effect of each variant on gene or protein function. About 99% of variants are predicted not to affect function (Table 1). Mendelian Inheritance in Man (MIM) lists 7,036 genetic diseases that map to 4,545 genes (72). Next, due to the low frequencies of neonatal genetic diseases associated with purifying selection pressure, all variants that are common in populations are discarded. With a few exceptions, the cutoff is typically 1% minor allele frequency (MAF), removing 95% of variants (Table 1). These steps generate a short list of variants that are evaluated one by one, based on a large number of additional considerations, such as diplotype zygosity, allele frequency, computational pathogenicity prediction, and de novo occurrence. Each variant is compared with several reference databases of disease-causing and non-disease-causing variants. Dependent upon the patient’s age, location, such as the NICU, and clinical features, a genetic disease diagnosis is made in 10% to 50% of cases. In 5% of cases, two genetic diseases co-exist. In addition to findings that are considered diagnostic, in about 20% to 35%, variants of uncertain significance (VUSs) in genes associated with the phenotypic characteristics of the child’s presentation are identified. These are often called VUS-suspicious, and are typically also reported. In 5-10% of cases, there will be incidental findings (diagnostic findings that are considered unrelated to the child’s current illness). This process, genome interpretation, can be fully automated, taking 5-15 minutes, with retention of about 80% sensitivity and about 50% specificity, values which are currently insufficient to permit autonomous performance (20). Expert manual interpretation by specialist staff including physicians and molecular laboratory directors typically takes one hour to one day. Results likely to change management in a highly beneficial manner are typically returned immediately, verbally as a provisional report. About one third of results undergo orthogonal confirmatory testing to verify, for example, that two variants in a recessive condition are in trans, de novo status when parental genomic sequences are not available, or to exclude the possibility that variants are false positives.

Table 1.

Comparison of the median analytic performance of rWES (n=95) and rWGS (n=118). Wilcoxon signed-rank adjusted p-values

rWES or rWGS	% Coding Nt with ≥10X Coverage	Total Variants	SNVs	Indels	Coding variants	Rare Variants (MAF <1%)	Variants in MIM disease genes	Missense Variants	Non-sense Variants	Altered canonical splice sites	Frame-shift indels	Disrupted start codons
rWES	94.5	38,901	35,465	3,401	23,421	2,703	670	558	14	15	46	3
rWGS	98	4,669,310	3,792,213	881,699	26,080	240,648	48,231	687	16	82	85	4
Fold difference	1.04	121	107	258	1.12	85	69	1.3	1.3	5.4	1.8	1.3
p-value		<.0001	<.0001	<.0001	<.0001	<.0001	<.0001	<.0001	<.0001	<.0001	<.0001	0

Diagnostic and Clinical Utility of Rapid Genome Sequencing

The diagnostic yield of rapid genome sequencing in infants and children in ICUs was evaluated in 33 clinical studies between 2012 and November 2021 (Table 2) (3, 13, 16, 21, 24, 25, 27, 28, 32, 43, 63, 64, 73, 79, 82–84, 86, 87, 96–98, 103, 109, 110). Most were cohort studies. Fifteen evaluated rWGS, 18 rWES, and one rapid panel testing. While the inclusion criteria, clinical settings, and hospital systems varied, all evaluated the diagnostic yield in children in ICUs with suspected genetic diseases. Average turnaround time varied from 0.8 days to 60 days. The weighted average rate of genetic disease diagnosis was 36% (range 19% - 83%, n=2874). Twenty-nine studies (n=1037) also evaluated acute clinical utility, as measured by changes in management upon return of results. The weighted average rate of change in management was 27% (range 7% - 60%). Ten studies (n=487) examined changes in outcome following those changes in management. The weighted average rate of change in outcome was 19% (range 8% - 30%).

Table 2.

Studies of the diagnostic performance, clinical utility and change in outcome of rWES, rWGS and rapid panel tests in seriously ill children in intensive care units

Ref.	Year	Study Type	Test Type	Enrollment Criteria	Size	Dx Rate	Change in Management	Change in Outcome	TAT (d)
83	2012	Cases	urWGS	NICU infants with suspected genetic disease	4	75%	n.d.	n.d.	2
103	2015	Cohort	rWGS	<4 mo of age; Suspected actionable genetic disease	35	57%	31%	29%	23
63	2017	Cohort	rWES	<100 days of life; Suspected genetic disease	63	51%	37%	19%	13
96	2017	Cohort	rWGS	Infants; NICU, PICU; susp. genetic diseases	23	30%	22%	22%	12
64	2018	RCT	rWGS, SOC	<4 mo of age; Suspected genetic disease	32	41%	31%	n.d.	13
25	2018	Cohort	rWGS	infants; Suspected genetic disease	42	43%	31%	26%	23
92	2018	Cohort	rWES	Acutely ill children with suspected genetic diseases	40	53%	30%	8%	16
90	2018	Cohort	rWGS	Children; PICU and Cardiovascular ICU	24	42%	13%	n.d.	9
82	2019	Cohort	rWGS	4 months-18 years; PICU; Suspected genetic diseases	38	48%	39%	8%	14
28	2019	Cohort	rWGS	Suspected genetic disease	195	21%	13%	n.d.	21
17	2019	Cases	urWGS	Infants; ICU; Suspected genetic disease	7	43%	43%	n.d.	0.8
32	2020	Cohort	rWES	<6 mo; ICU; hypotonia, seizures, metabolic, multiple congenital anomalies	50	58%	48%	n.d.	5
24	2019	Cohort	rWES	NICU; infants; susp. genetic disease	25	72%	60%	n.d.	7.2
110	2019	Cohort	rWES	PICU and other; children; susp. genetic disease	40	53%	43%	n.d.	6
97	2020	Cohort	rWES	NICU & PICU; complex	130	48%	23%	n.d.	3.8
27	2020	Cohort	rWES	Critical illness; medical genetics selected	46	43%	52%	n.d.	9
13	2020	Cohort	rWES	PICU; < 6 years; new metabolic/neurologic disease	10	50%	30%	n.d.	9.8
87	2020	Cohort	rWES	ICU	368	27%	n.d.	n.d.	n.d.
16	2020	Cohort	rWES	>1 year; ICU and inpatient	102	31%	27%	n.d.	11
79	2020	Cohort	rWES	Various	41	32%	n.d.	n.d.	7
3	2020	Implem	rWES	<18 yr; NICU and PICU	108	51%	44%	n.d.	3
86	2020	Cohort	rWES	Infants; NICU, PICU; susp. genetic diseases	18	83%	61%	n.d.	14
98	2020	Cohort	rWES	Infants; NICU, PICU; susp. genetic diseases	33	70%	30%	30%	1
			rWGS		94	19%	24%	10%	11
11, 23, 43	2019-2021	RCT	rWES	Infants; disease of unknown etiology; with in 96 hours of admission	95	20%	20%	18%	11
			urWGS		24	46%	63%	25%	4.6
58	2021	Crossover	rWGS, panel	Infants; disease of unknown etiology	113	33%	26%	n.d.	n.d.
22	2021	Implem	rWGS	Medicaid infants; unknown etiology; within 1 week of admission	178	43%	31%	n.d.	3
73	2021	Cohort	rWES	Critically ill; 6 days-15 years; susp. genetic diseases	40	43%	31%	n.d.	5
84	2021	Cohort	rWES	NICU, PICU, infants; sup. Genetic diseases	61	43%	11%	n.d.	60
77	2021	RTDCT	rWGS, WGS	0-120 days old; ICU; susp. genetic disease	354	31%	25%	n.d.	15
109	2021	Crossover	rWES	Critically ill infants with conditions suggestive of geneticaly heterogeneous disorders	202	20%	n.d.	n.d.	20
			rWGS		202	37%	7%	n.d.	7
21	2021	Cohort	rWGS	NICU, PICU units with probable genetic disease and in urgent need for etiological diagnosis to guide medical care	37	57%	n.d.	n.d.	43
Weighted Average					2874	36%	27%	18%

n.d.: not done; d: days

A meta-analysis compared the diagnostic and clinical utility of WGS, WES, and CMA in 20,068 children with suspected genetic diseases through August 2017 (17). It found that the diagnostic utility of WGS (41%, 95% CI 34-48%) and WES (36%, 95% CI 33-40%) was not significantly different, but was greater than CMA (10%, 95% CI 8-12%). Diagnosis was significantly more likely for trios than singletons (odds ratio 2.04, 95% CI 1.62-2.56). Interestingly, hospital-based interpretation led to a higher rate of diagnosis by WGS/WES (42%, 95% CI 38-45%) than that of reference laboratories (29%, 95% CI 27-31%). WGS results had greater clinical utility (27%, 95% CI 17-40%) than CMA (6%, 95% CI 5-7%).

There have been three randomized, controlled trials (RCTs) of rapid genome sequencing in infants in ICUs. The first, Newborn Sequencing in Genomic Medicine and Public Health (NSIGHT1), compared the rate of genetic disease diagnosis with rWGS plus standard genetic tests to that of standard genetic tests alone (including exome sequencing) in 65 infants aged <4 months in a regional NICU and pediatric intensive care unit (PICU) with illnesses of unknown etiology (77). The study was terminated early due to loss of equipoise: 15% of controls underwent compassionate cross-over to receive rWGS, demonstrating the difficulty of diagnostic RCTs. The rate of genetic diagnosis within 28 days of enrollment (the primary end point) was higher with rWGS (31%) than controls (3%). Median time to diagnosis was significantly less with rWGS (13 days) than controls (107 days). This study established that rWGS increased the proportion of NICU/PICU infants who received timely diagnoses of genetic diseases relative to standard genetic tests.

The second RCT, NSIGHT2, evaluated effectiveness of rWES, rWGS, and urWGS as first-tier tests in 213 seriously ill infants with diseases of unknown etiology, representing 46% of NICU admissions (a broader proportion than previously studied) (43). The analytic performance of rWGS/urWGS was superior to rWES (Table 1). Their diagnostic rates and times to result of rWGS and rWES were not different (~20% and median of 11 days, respectively). Both, however, were inferior to urWGS (46% diagnosis, median time to result 4.6 days). The incremental diagnostic yield of reflexing to trio after negative singleton analysis was only 0.7%, a result that was at odds with the previously published meta-analysis. A second report from the NSIGHT2 RCT demonstrated that clinicians perceived rapid genome sequencing to be useful (the primary NSIGHT2 study end point) in 77% of infants (Figure 3) (23). Interestingly, both positive (93%) and negative tests (72%) had clinical utility. The explanation of the latter was that negative rapid genome sequencing results were useful in decreasing the posterior probability of genetic disease, enabling clinicians to focus on non-genetic etiologies. Results of rapid genome sequencing changed clinical management in 28% infants and outcomes in 15%.

Figure 3.

Comparison of the traditional approach to etiologic diagnosis of genetic diseases in NICU infants with rWGS-informed precision medicine. Values are from the NSIGHT2 RCT (11, 23, 43)

A third report from the NSIGHT2 RCT evaluated parental perceptions of clinical utility, adequacy of consent, and potential harms and benefits (Figure 3) (11). When rWGS was first introduced, there were concerns that parents of newborns in ICUs would be unable to provide informed consent and that testing would cause anxiety, be associated with decisional regret, depression, and lead to interference with bonding. However, more than 90% of NICU infant parents felt adequately informed to consent to diagnostic genome sequencing. Despite only 23% of infants’ receiving diagnoses, 97% of parents reported that genome sequencing was useful, and median decisional regret was zero (range 0-100). Evaluation of potential harms of testing revealed that only 2% perceived harm (1% related to a negative result and 2% related to stress or confusion). This study established that when rapid genome sequencing was performed as a first-tier diagnostic test in almost one half of regional NICU infants, most results were considered useful, and harms were rare and mild.

The third RCT was a multicenter, time-delayed trial of 354 infants in ICUs with suspected genetic disease aged <4 months who were randomized to receive WGS results within either 15 or 60 days after enrollment (31). Among infants who received WGS results within 15 days, 31% had diagnoses, and 21% had consequent changes in management at 60 days after enrollment, compared with 15% and 10%, respectively, among those who received results within 60 days. This study established the superiority of use of rWGS as a first-tier diagnostic test.

In summary, research studies have established an evidence base for the diagnostic and clinical utility of rWGS as a first tier diagnostic test for NICU infants with diseases of unknown etiology. This evidence base contributed to publication by the ACMG of a clinical guideline that supports the clinical utility and desirable effects of WES and WGS on active and long-term clinical management for pediatric patients <1 year of age with 1 or more congenital anomalies (56).

Implementation Science and Quality Improvement

While research studies are ideal for testing hypotheses, they feature biases that can lead to their results’ failing to be repeated in general experience. Inclusion and exclusion criteria and requirement for informed parental consent, for example, can result in enrollment that is not representative of NICU populations. Furthermore, research studies employ staff such as genome-literate research nurses and genetic counselors who may upskill NICU teams, enabling practices that would not be possible in their absence. Translation of novel, evidence-based practices into routine clinical use typically takes decades. The field of implementation science has developed strategies to facilitate adoption of evidence-based practices (60, 78, 101, 102). It is defined as “the scientific study of methods to promote the systematic uptake of research findings and other evidence-based practices into routine practice, and, hence, to improve the quality and effectiveness of health services” (5). Implementation science involves comprehensive identification of the key barriers to adoption, their organization within a framework, and then real-world implementation studies that explore how to refine implementation to optimize effectiveness and facilitate adoption. Sites in the US, UK, and Australia started implementation science studies of rapid genome sequencing in 2017 motivated by the desire to scale use by NICU teams and achieve sustainable improvement in outcomes (14). The resultant framework developed by Rady Children’s Institute for Genomic Medicine (RCIGM) is shown in Figure 4 (45). Assessment of barriers to adoption across ~70 NICUs served by RCIGM revealed the need to implement rWGS within a learning, rapid precision medicine delivery system (RPM). Based on the successful framework for implementation of traditional newborn screening (8), RPM had four components: 1. Onboarding of health systems, upskilling of NICU teams, and early identification of infants in need during admission (items 1-7, Figure 4); 2. Diagnosis, comprising rWGS ordering, sequencing, interpretation and results review with ordering pediatricians (items 8-13, Figure 5); 3. Acute management guidance and precision medicine delivery (items 15 and 16, Figure 5), and 4. Learning based on monitoring outcomes, data science and quality improvement (items 14, 17 and 18, Figure 4) (45). Implementation science studies of rapid genome sequencing have started to be published. For example, Project Baby Bear, a payer-funded, implementation project, evaluated the clinical and economic impact of the RCIGM system of rWGS-based RPM (22, 26). Rapid WGS was utilized as a first line diagnostic test in Medicaid infants with diseases of unknown etiology in five regional California NICUs. The majority of infants were from underserved populations. Of 184 infants enrolled, 40% received a diagnosis by rWGS that explained their admission with a median turnaround time of 3 days. In 32% of infants, rWGS led to changes in medical care. rWGS testing and resultant precision medicine cost $1.7 million (~$21,000 per diagnosis) but led to ~$2.5 million cost savings ($13,526 per infant tested, Figure 5). Sensitivity analysis showed that most of the cost savings were lost if turnaround time was lengthened to 14 days. Thus, Project Baby Bear confirmed the diagnostic and clinical utility previously shown in research studies and demonstrated net reduction in healthcare utilization costs associated with relatively broad indications for first tier rWGS. Analysis of the implementation process revealed the need for an rWGS champion in each NICU, educational needs and strategies, negotiating decision-making roles and processes, workflows and workarounds, and perceptions about rWGS. A similar implementation pilot in 12 hospitals demonstrated the feasibility of rWES with 3-day turnaround time in critically ill pediatric patients with suspected monogenic conditions in the Australian public health care system (3). Genetic diseases were diagnosed in 51% of infants and management was changed in 76% of those receiving diagnoses and 11% of infants with negative reports. An earlier implementation project by the same group in two Australian centers showed similar rates of diagnostic and clinical utility, while the cost per diagnosis was $10,453, and net cost savings were $10,600 per infant tested (90).

Figure 4.

Genome-informed neonatology is a learning healthcare delivery system with four components: 1. Education, engagement, and equipping (yellow line); 2. Rapid, diagnostic whole genome sequencing (green line); 3. Translation into precision neonatology (blue line); 4. Therapeutic innovation (blue line). Learning feedback loops are shown.

Figure 5:

Cost effectiveness of precision neonatology is dependent on time to result. a. Cost savings during the initial hospitalization from first-tier use of rWGS for NICU infants with suspected genetic diseases in Project Baby Bear. b. Total cost savings.

Policies and Guidelines

Three national health systems (the National Health Service in England and in Wales and Australian Genomics) are adopting rWGS in critically ill, inpatient infants with suspected genetic diseases. In the US, California Blue Shield was the first major payer to issue a coverage policy for rapid, diagnostic WES/WGS in NICU infants and trios in July 2019 (Box 1) (12). Rapid testing was defined as an average turnaround time of less than 14 days, but usually less than 7 days. The policy called for immediate verbal reporting of results to the clinician if changes in management were likely. The criteria for reimbursable testing are shown in Box 1 and reflect those used in published research studies such as Project Baby Bear and NSIGHT2. A new current procedural terminology (CPT) code was authorized for rWGS (0094U). In March 2020, the policy was accepted by Anthem/Blue Cross/Blue Shield of America and has since been ratified by ten regional plans. In September 2021, Michigan became the first US state to reimburse rWGS in critically ill Medicaid infants in NICUs and PICUs as a carve-out from the intensive care diagnosis related codes (DRGs). The Michigan coverage policy is very similar to that of Blue Cross/Blue Shield. The CPT code 0094U was implemented and priced at $6,275 for probands and $10,750 for trios. In July 2021, Governor

Box 1. Policy for coverage of rWGS and rWES in infants in ICUs by Blue Shield California.

Rapid whole exome sequencing or rapid whole genome sequencing, with trio testing when possible, meets the definition of medical necessity for the evaluation of critically ill infants in neonatal or pediatric intensive care with a suspected genetic disorder of unknown etiology when both (1 & 2) of the following criteria are met:

1. At least one of the following criteria is met:

   1. Multiple congenital anomalies (e.g. persistent seizures, abnormal ECG, hypotonia);
   2. An abnormal laboratory test or clinical features suggests a genetic disease or complex metabolic phenotype (e.g, abnormal newborn screen, hyperammonemia, lactic acidosis not due to poor perfusion); or
   3. An abnormal response to standard therapy for a major underlying condition.
2. None of the following criteria apply regarding the reason for admission to intensive care:

   1. An infection with normal response to therapy;
   2. Isolated prematurity;
   3. Isolated unconjugated hyperbilirubinemia;
   4. Hypoxic Ischemic Encephalopathy:
   5. Confirmed genetic diagnosis explains illness;
   6. Isolated Transient Neonatal Tachypnea;
   7. Nonviable neonates.

Newsom signed California Assembly Bill 114 into law, providing $6 million to reimburse rWGS in Medicaid infants in California in 2022. A coverage policy has not yet been issued by the California Department of Health. Feasibility projects of rapid WGS and WES are underway in NICU infants in many high and middle income countries (89).

Future Directions

Many research and implementation studies are currently underway that will continue to refine our understanding of the use of genome sequencing in NICUs and in other pediatric intensive care settings (Pediatric Intensive Care Units and Pediatric Cardiac Intensive Care Units). It is already apparent, however, that rWGS-based precision medicine has considerable diagnostic and clinical utility and cost effectiveness in infants in ICUs. This evidence will drive an era of broad adoption of rWGS-informed precision medicine for infants in NICUs. We anticipate issuance of guidelines from professional bodies that support use of rWGS as a first-tier diagnostic test in critically ill infants with diseases of unknown etiology. We anticipate that guidelines will recommend testing shortly after admission, turnaround time of 3 days, and reports to include all classes of variants. We anticipate rWGS-informed precision medicine will become part of the core curriculum for neonatology fellowship. In parallel, there will continue to be broader reimbursement of rWGS by Medicaid and private payers. It will be important for coverage policies to include reimbursement as a carve-out payment rather than inclusion in current DRGs.

A key unanswered question at present is how broadly genome sequencing should be used in NICUs. The NSIGHT2 study showed diagnostic and clinical utility when genomic sequencing was used in 46% of admissions to a regional NICU. It was the first to quantify the value of negative genomic sequencing. Further studies are needed to examine optimal breadth of use in level II, III, and IV NICUs.

The research published to date still underestimates the prevalence of single locus genetic diseases in infants in ICUs. An unpublished study of postmortem infants from one of us found a high rate of undiagnosed genetic disorders, many of which had effective treatments in a county with relatively broad use of NICU and PICU rWGS (45, 81). Genetic diseases continue to be discovered at a rapid pace, and the quality of WGS continues to improve rapidly. For example, it is likely that diagnostic yield will increase by 5 – 15% through use of combined long-and short-read WGS (65). Long-read WGS is more expensive than short-read WGS. It is not as good at detection of SNVs but far superior for characterization of SVs and CNVs. The emerging picture of these variants is that they are generally more complex than we were aware and frequently are combination events that include deletions, insertions, and rearrangements at a single locus. Also in the relatively near future, we will change from read alignment to read assembly, generating assembled individual genomes (85). It is likely that this will increase yield by another 5 – 15%, particularly in racial and ethnic groups that did not contribute to the current reference human genome. An intermediary to genome assemblies may be digital, automated ethnic ancestry delineation in short-read WGS and then alignment to the best from a large set of diverse reference genomes (38). Lastly, as we enter a new era of integrative, clinical omics, we are starting to see the value of integrating WGS with functional omics, such as RNAseq (51), proteomics, and metabolomics. Currently, there are a very large number of variants for which we cannot predict pathogenicity. Most of these are intergenic or intronic. These omics technologies allow us to evaluate the functional consequences of such variants. The two uses of such technology are to diagnose unsolved cases and to build new databases of functional omics-annotated variants. Ultimately, there will be very few remaining VUSs.

Another trend that is gaining momentum is the use of artificial intelligence (AI) to improve scalability and increase adoption of rapid genome sequencing and precision medicine in infants in ICUs. In addition to the emerging uses of AI discussed previously, we anticipate algorithm-based, automated EHR alerts triggered in infants with high likelihood of underlying genetic disease and at high risk of mortality (akin to EHR sepsis alerts) (80). An unpublished study from one of us (SFK) describes the development of an AI-informed, automated system for provision of acute management guidance for newly diagnosed genetic diseases. It was designed for use by front-line intensivists and neonatologists, particularly in hospitals that lack a full array of subspecialists and super-specialists where there may be delays in implementation of optimal treatments for ultra-rare genetic diseases. AI tools such as these will be critical in democratization of rWGS-based precision medicine to most birthing-hospital associated NICUs.

The most exciting future development will be accelerated development of genetic therapies (Figure 1). Effective gene therapy for Spinal Muscular Atrophy type 1 (SMA1) is a staggering accomplishment. We are realizing, however, that the system shown in Figure 4 is critically needed in order to diagnose and treat hypotonic babies in the first week of life. A byproduct of increasing use of rWGS is unparalleled natural history studies of specific infant-onset genetic diseases that accelerate drug development both by identification of end-points and indications and by serving as real-world evidence for submissions investigational new drug applications to the Food and Drug Administration (FDA) in the US (1). It must be understood that the majority of infant-onset genetic diseases was either only recently discovered or was not generally diagnosable in the absence of rapid genome sequencing. Thus, many genetic diseases for which drug development was not previously feasible now have potential therapeutic strategies, including gene therapy, anti-sense oligonucleotides, small molecule, genome editing, and repurposed FDA approved drugs. In addition to better therapies, natural history studies will improve our ability to predict prognosis for critically ill infants, communicate with parents, anticipate complications, and stratify patients within genetic disorders as is now happening with Duchenne Muscular Dystrophy and Cystic Fibrosis.

Finally, and beyond the scope of this review, is newborn screening for genetic diseases by rWGS (NBS-by-WGS) (9) and increasing use of rWGS for fetal diagnosis. We anticipate that healthy newborns will start to be widely screened by WGS for ~500 infant onset, severe genetic diseases that have effective treatments within the next five years. Professional society opinions support the use of prenatal exome sequencing in cases with specific fetal anomalies, and we anticipate imminent testing of fetal rWGS to improve diagnostic success (18, 35, 69, 94, 99)

Next steps: broad implementation of genome sequencing and associated research –

Genetic heterogeneity of fetal and neonatal phenotypes, overrepresentation of novel and rare genomic variants in the NICU population, the high fraction of affected infants in NICUs with undiagnosed congenital anomalies or metabolic disorders, the significant contribution of genetic disorders to morbidity and mortality among infants, and the proven record of rWGS/rWES to change clinical management and identify individualized therapeutic strategies make the integration of genome sequencing into NICU clinical care a high priority (101, 102). As illustrated by the framework developed by Rady Children’s Institute for Genomic Medicine (RCIGM) described above (45), acceleration of this integration will require an intentional commitment to implementation of a sustainable Genomic Learning Healthcare System (GLHS) through implementation science which uses knowledge and experience derived from NICU and prenatal clinical care for cycles of continuous improvement and research to improve patient outcomes through enhanced diagnostic success, clinical efficiency, and discovery of genotype-guided therapeutics (60–62, 101, 102, 108). Multiple stakeholders including parents, payers, bioinformaticians, geneticists, genetic counselors, neonatologists, obstetricians, pediatric subspecialists, developmental biologists, model organism investigators, experts in strategies to rescue variant-encoded disruption (e.g., with gene therapy, anti-sense oligonucleotides, repurposed drugs approved by the FDA, small molecules, or genome editing), clinical investigators, genomicists, and institutional leaders will need to participate in development of the GLHS for NICUs. The challenges to GLHS implementation include facilitating a cultural shift among providers from phenotype- to genotype-first diagnosis, development of governance structures that can adapt to new ethical and operational questions, standardization of genomic and phenotypic information in EHRs, more reliable computational and functional evaluation of novel and rare VUSs that are clinically actionable, and development of consent strategies that permit inclusion of an individual’s clinical and genomic data while protecting patient confidentiality (61, 100, 108). However, NICUs have a long record of commitment to quality improvement through implementation science and, more recently, of value-based quality initiatives (33). With the already available evidence of clinical value of r- and ur-WES and WGS in the NICU, GLHS implementation will lead to deployment of these best practices to improve outcomes for critically ill newborn infants and children and reduce costs of their care.

Terms and Definitions:

ACMG

American College of Medical Genetics and Genomics

AI

Artificial intelligence

BNS1

Benign neonatal seizures type 1

CDH

Congenital diaphragmatic hernia

CMA

Chromosomal microarray analysis

CPT

Current procedural terminology

DEE7

Developmental and epileptic encephalopathy type 7

DRG

Diagnosis related code

ECMO

Extracorporeal membrane oxygenation

FDA

Food and Drug Administration

GB

Gigabase

GINA

Genetic Information Nondiscrimination Act

GLHS

Genomic Learning Healthcare System

HPO

Human Phenotype Ontology

Indel

Insertion and deletion

MAF

Minor allele frequency

MIM

Mendelian Inheritance in Man

NICU

Neonatal Intensive Care Unit

NSIGHT

Newborn Sequencing in Genomic Medicine and Public Health

PICU

Pediatric intensive care unit

RCT

Randomized controlled trial

RPM

Rapid precision medicine

rWES

Rapid whole exome sequencing

rWGS

Rapid whole genome sequencing

SNV

Single nucleotide variant

TAT

Turnaround time

THMD2

Thiamine metabolism dysfunction syndrome 2

urWGS

Ultra-rapid whole genome sequencing

VUS

Variant of uncertain significance

WES

Whole exome sequencing

WGS

Whole genome sequencing