Novel Variant Findings and Challenges Associated With the Clinical Integration of Genomic Testing: An Interim Report of the Genomic Medicine for Ill Neonates and Infants (GEMINI) Study

Key Points

Question  Can a targeted genomic sequencing platform diagnose neonates and infants suspected of having a genetic disorder as accurately as rapid whole-genomic sequencing?

Findings  In this comparative effectiveness study of 113 infants, diagnostic and/or phenotypically related variants of unknown significance were returned for 51 patients (45%), while 62 (55%) had negative results; results were concordant between platforms in 73% of patients. Of 51 positive cases, 67% differed in the reported result because of technical limitations of the targeted platform, interpretation of the variant, and/or filtering discrepancies.

Meaning  The diagnostic capabilities of genomic sequencing technologies have the ability to affect clinical care but have significant limitations that must be better understood.

Importance  A targeted genomic sequencing platform focused on diseases presenting in the first year of life may minimize financial and ethical challenges associated with rapid whole-genomic sequencing.

Objective  To report interim variants and associated interpretations of an ongoing study comparing rapid whole-genomic sequencing with a novel targeted genomic platform composed of 1722 actionable genes targeting disorders presenting in infancy.

Design, Setting, and Participants  The Genomic Medicine in Ill Neonates and Infants (GEMINI) study is a prospective, multicenter clinical trial with projected enrollment of 400 patients. The study is being conducted at 6 US hospitals. Hospitalized infants younger than 1 year of age suspected of having a genetic disorder are eligible. Results of the first 113 patients enrolled are reported here. Patient recruitment began in July 2019, and the interim analysis of enrolled patients occurred from March to June 2020.

Interventions  Patient (proband) and parents (trios, when available) were tested simultaneously on both genomic platforms. Each laboratory performed its own phenotypically driven interpretation and was blinded to other results.

Main Outcomes and Measures  Variants were classified according to the American College of Medical Genetics and Genomics standards of pathogenic (P), likely pathogenic (LP), or variants of unknown significance (VUS). Chromosomal and structural variations were reported by rapid whole-genomic sequencing.

Results  Gestational age of 113 patients ranged from 23 to 40 weeks and postmenstrual age from 27 to 83 weeks. Sixty-seven patients (59%) were male. Diagnostic and/or VUS were returned for 51 patients (45%), while 62 (55%) had negative results. Results were concordant between platforms in 83 patients (73%). Thirty-seven patients (33%) were found to have a P/LP variant by 2 or both platforms and 14 (12%) had a VUS possibly related to phenotype. The median day of life at diagnosis was 22 days (range, 3-313 days). Significant alterations in clinical care occurred in 29 infants (78%) with a P/LP variant. Incidental findings were reported in 7 trios. Of 51 positive cases, 34 (67%) differed in the reported result because of technical limitations of the targeted platform, interpretation of the variant, filtering discrepancies, or multiple causes.

Conclusions and Relevance  As comprehensive genetic testing becomes more routine, these data highlight the critically important variant detection capabilities of existing genomic sequencing technologies and the significant limitations that must be better understood.

An estimated 10% to 25% of neonates in neonatal intensive care units have an undiagnosed genetic disorder.^1-3 Because of the nonspecific presentation of many genetic disorders, affected neonates (1) have a 40% longer hospitalization,⁴ (2) may not receive a diagnosis, (3) may be misdiagnosed, and (4) often have a prolonged diagnostic odyssey.⁵ The Newborn Sequencing in Genomic Medicine and Public Health (NSIGHT) trials demonstrated the important role that rapid sequencing can have in providing a timely genetic diagnosis to improve neonatal outcome.^3,6-9 However, these platforms remain expensive and involve complex ethical dilemmas.¹⁰ A targeted sequencing approach aimed at disorders presenting in the first year of life could limit incidental findings and reduce costs. However, the diagnostic capability of such a platform is unproven and must be compared with more standardized whole-genomic sequencing (WGS) technology prior to routine use.

The Genomic Medicine for Ill Neonates and Infants (GEMINI) Study (NCT03890679) is an ongoing, multiyear, multisite trial funded through the US National Center for Advancing Translational Sciences with a targeted enrollment of 400 neonates or infants younger than 1 year suspected of having an undiagnosed genetic disorder. The GEMINI Study is comparing the diagnostic yield of a novel targeted genomic sequencing platform (NewbornDx; Athena/Quest Diagnostics) with rapid WGS (rWGS). Specifically, the targeted genomic sequencing platform interrogates 1722 actionable genes for disorders that present in the first year of life. Patients and their parents undergo simultaneous testing on the targeted genomic sequencing platform and rWGS. Interpretation on both platforms is rapid (≤14 days) and targeted based on proband phenotype to establish a timely diagnosis and to avoid detecting unrelated incidental findings.

Recruitment for the GEMINI Study began in July 2019. Enrollment was exceeded by more than 30%, and 51 novel variants were identified that had never been previously reported in the literature. There was a 67% discordance between laboratories for infants found to have a diagnosis or variant possibly related to phenotype. Discordance was often caused by each laboratory’s interpretation regarding the relative significance of that variant to disease presentation (ie, pathogenic [P], likely pathogenic [LP] or variant of unknown significance [VUS]). As these data emerged, it was important to share these interim findings for the benefit of undiagnosed infants with similar phenotypes and to highlight existing limitations regarding these quickly emerging technologies.

Parental written consent for participation in the GEMINI study was obtained with central institutional review board approval from Johns Hopkins University with approval at each participating hospital: Tufts Medical Center (Boston, Massachusetts), Rady Children’s Hospital (San Diego, California), University of Pittsburgh Medical Center Children’s Hospital (Pittsburgh, Pennsylvania), Mount Sinai Kravis Children’s Hospital (New York, New York), North Carolina Children’s Hospital (Chapel Hill), and Cincinnati Children’s Hospital Medical Center (Cincinnati, Ohio). Hospitalized infants younger than 1 year with a suspected, undiagnosed genetic disorder were eligible for enrollment. Neonates were excluded if they were born at fewer than 23 weeks’ gestation, had a major congenital infection, or had a genetic diagnosis that fully explained all phenotypic findings. Infants were classified as urgent if they (1) required mechanical ventilation, (2) exhibited severe neurological complications, (3) were hemodynamically unstable, or (4) were categorized as such at the request of the site’s principal investigator. Urgent cases underwent ultrarapid sequencing and analysis with a preliminary report generated within 72 hours of specimen arrival.

Although trio testing was preferred, enrollment was dependent solely on the proband. Parents must opt in to receive secondary findings approved by the American College of Medical Genetics and Genomics (ACMG) for their infant and themselves.¹¹ Because of the phenotypic-driven interpretation, secondary findings were not sought, but rather were incidental findings of the analysis and not always detected. Secondary findings were only reported for the proband if they were (1) on the ACMG list and present in childhood or (2) they presented in childhood and there is a specific treatment available.¹² Nonpaternity is never revealed. Incest is reported to appropriate authorities for all enrolled minor mothers. In most cases, a family met with a geneticist or genetic counselor at the time of enrollment when a 3-generation pedigree was obtained to inform sequencing interpretation.

The patient provided 1 mL of whole blood in EDTA tubes for rWGS and 5 dried blood spots on filter paper (0.5 mL; Perkin Elmer; Health Sciences Spot Saver Cards; GR2261007) for the targeted genomic sequencing platform. Parents provided 3 mL of whole blood in EDTA tubes. Blood for rWGS was shipped on ice to Rady Children’s Institute of Genomic Medicine; blood for the targeted genomic sequencing platform was shipped at ambient temperature to Athena/Quest Diagnostics. To facilitate rapid interpretation, human phenotype ontology (HPO) terms were provided by the site to the laboratories for each patient.¹³ HPO terms are a standardized vocabulary of phenotypic human abnormalities that accurately describe the individual being evaluated and are used to perform a targeted interrogation of the genome.¹³ Pertinent demographic and clinical data were recorded. Race and ethnicity were recorded from the medical record based on parental self-reporting. Clinical utility of findings was assessed after return of results via a survey of the physician of record. Changes in clinical management, medications, surgeries, other therapies, and diagnostic testing were recorded.

Clinical rWGS and ultrarapid WGS laboratories were accredited by the College of American Pathologists and certified through the Clinical Laboratory Improvement Amendments. The methods have been published in detail.³ HPO terms were mapped to simple genetic diseases with VAAST (Fabric Genomics).¹⁴

Genome sequences were aligned to human genome assembly GRCh37 (hg19), and variants were identified with the DRAGEN Platform (Illumina).¹⁴ Structural variants were identified with Manta and CNVnator and filtered to retain those coding regions of known disease genes with allele frequencies less than 2% in the Rady Children’s Institute of Genomic Medicine database.¹⁴ Nucleotide and structural variants were automatically annotated and ranked using Opal Clinical (Fabric Genomics) and manually interpreted iteratively by clinical molecular geneticists according to standard clinical guidelines.¹⁴ Genomic sequence interpretation was performed as singleton probands. Infants undiagnosed as singletons were reanalyzed as trios.¹⁴ If a provisional diagnosis was made with a treatment identified to prevent morbidity or mortality, it was immediately conveyed to the caregivers. Causative variants were confirmed by Sanger sequencing or chromosomal microarray.

The targeted genomic sequencing platform¹⁵ testing was performed in a College of American Pathologists–accredited, Clinical Laboratory Improvement Amendments–certified, and New York State–licensed laboratory by Athena/Quest Diagnostics. Genomic DNA was extracted using QIAmp DNA methods (Qiagen). Custom oligonucleotide probe libraries (Agilent SureSelect) captured genomic DNA regions of interest. Sequencing was performed on a NextSeq 500 (Illumina) using paired-end 75-base pair reads. Libraries were sequenced to a global mean targeted coverage of approximately 300 times and a local coverage of approximately 99% of bases 20 times or more. Sequencing reads were mapped and aligned to the reference genome GRCh37 (hg19), followed by position sorting and variant calling using Edico Dragen version 2.6.5 (Illumina). Opal Clinical software (Fabric Genomics) was used for variant interpretation and HPO-driven prioritization of causal variation. Candidate variants were assessed for pathogenicity using a standardized framework.¹⁶ Data were gathered from multiple sources. Evidence was reviewed by a variant scientist, clinical molecular geneticist, geneticist, and genetic counselor. Plausible causal variants in genes related to phenotype were identified based on a systems approach of disease severity and body system combined with the application of phenotypically driven variant ontological reranking in the Fabric Genomics platform.¹⁷ Assessment of variants includes inheritance pattern, frequency of variant, variant consequence, and reports in public databases. All variants were confirmed by Sanger sequencing.

Variants were classified as P, LP, or VUS based on HPO terms provided and each laboratory’s interpretation in accordance with ACMG guidelines.¹⁸ A VUS was only reported if located in a gene that was casually related to a genetic disease whose expected clinical features in infancy clearly overlap with the observed phenotypes in the proband. All variants were reported to ClinVar at yearly intervals per protocol. Discordant results between laboratories were defined as variant discrepancies that differed between clinically significant (P and LP) and VUS and variant discrepancies between VUS and not reported. In cases of discordance, the infant was classified into the highest level of variant classification. Analysis took place from March to June 2020.

To date, 113 of the targeted 400 patients (28%) have been enrolled (eFigure in the Supplement). Pertinent clinical and demographic data of enrolled patients are listed in Table 1. Overall, 116 parents (79%) who were approached consented to enrollment. A total of 102 infants (90%) were analyzed as part of a duo or trio on the targeted genomic sequencing platform while rWGS reflexed to a duo or trio for 71 infants (63%). Gestational age of patients ranged from 23 to 40 weeks and postmenstrual age from 27 to 83 weeks. Sixty-seven patients (59%) were male. Enrollment per site were as follows: Tufts Medical Center, 14; Rady Children’s Hospital, 27; University of Pittsburgh Medical Center Children’s Hospital, 7; Mount Sinai Kravis Children’s Hospital, 7; North Carolina Children’s Hospital, 5; and Cincinnati Children’s Hospital Medical Center, 53. Twenty-five cases (22%) were classified as urgent and underwent ultrarapid sequencing.

Diagnostic and/or VUS variants were returned for 51 patients (45%), while 62 (55%) were reported as negative. Results were concordant between platforms in 83 cases (73%). Thirty-seven patients (33%) had a P or LP variant consistent with a specific genetic diagnosis, and 14 patients (12%) had at least 1 VUS detected by 1 or both sequencing platforms (Figure). Patients undergoing urgent testing had 9 P/LP variants (36%) and 6 VUS (24%) (eTable 1 in the Supplement). Four infants (3%) had more than 1 diagnosis, 5 (4%) had a diagnosis and a VUS, and 2 (2%) had more than 1 VUS. A trisomy, tetrasomy, or a chromosomal deletion or duplication was identified in 9 patients (8%) by rWGS. Although 51 variants (9 P, 17 LP, and 25 VUS) have not previously been reported in the literature to our knowledge, 17 of these have been reported in ClinVar and/or gnomAD variant databases. Twelve infants (11%) had de novo variants, highlighting the significance of this inheritance pattern in this patient population. Of 51 patients with variants classified as P, LP, or VUS, results of the 2 testing platforms were discordant in 34 (67%) because of technical limitations of the platforms, variant interpretation, or both (Figure). Technical discrepancies included (1) a diagnosis of a trisomy or tetrasomy (n = 3), chromosomal duplication (n = 2), or chromosomal deletion (n = 4), which could not be detected with the targeted genomic sequencing platform; (2) the gene was not present on the targeted genomic sequencing platform (n = 8); and (3) limited coverage of the gene on the targeted genomic sequencing platform (n = 1). There was 1 computational filtering discrepancy caused by a mapping quality threshold between laboratories. Genomic results from each laboratory for infants with a diagnosis (n = 37) are provided in Table 2 including differences in variant interpretation, technical limitations, and those that were fully concordant. eTable 2 in the Supplement includes include all remaining infants with a VUS only (n = 14). Median age at diagnosis was 22 days (range: 3-313 days); 21 infants (58%) received a diagnosis in the neonatal period (≤28 days of life). Significant changes based on a diagnosis occurred in 29 infants (29 of 37 [78%] with P/LP variant[s] detected; 29 of 113 patients [26%] tested; Table 3) including redirection of care from comfort to specific therapy (n = 3), redirection of care to comfort (n = 3), and/or change in medical, surgical, subspecialist, diagnostic testing, or other therapeutic management (n = 23). Of 113 parents, 105 (93%) opted to receive secondary findings for their infants, with 110 of 113 mothers (90%) and 75 of 83 fathers (90%) opting to receive their secondary findings. Secondary findings were reported among 7 trios (Table 4). Three parents and other family members were newly diagnosed with a genetic condition based on the infant’s diagnosis.

The GEMINI Study has provided rapid genomic sequencing results to 113 patients, identifying positive findings in 51 (45%) and a molecular diagnosis in 37 (33%). The NSIGHT trial found a 43% diagnostic rate in critically ill neonates with WGS.⁷ A 2018 meta-analysis exploring the clinical utility of WGS and whole-exome sequencing in older children with suspected genetic disorders revealed a diagnostic yield of 41% and 36%, respectively.¹⁹ Beyond independent validation, the GEMINI Study affirms the significant effect that rapid phenotypically driven genomic sequencing can have on clinical care. Although enrollment is ongoing, the GEMINI Study has already (1) directly informed clinical care in 29 of 37 newly diagnosed infants (78%); (2) diagnosed 3 parents and related family members; (3) identified 51 novel variants; and (4) identified clinically actionable secondary findings in patients and their parents. The majority of results were provided within the first 28 days of life, demonstrating a substantial reduction in time to diagnosis. With a 79% enrollment rate, the GEMINI Study reveals a strong parental desire for testing in neonates suspected of having a genetic disorder. The reasons parents declined enrollment included a fear of blaming 1 partner, belief that a neonate did not have a genetic disorder, and disinterest in pursuing genetic testing or participating in research.

Importantly, results have directly informed clinical care and improved outcomes, including the identification of secondary findings. During a phenotypically driven interpretation of a trio, a BRCA2 pathogenic variant was found in a mother unaware of her carrier status and a RYR1 pathogenic variant was identified in a neonate at risk of malignant hyperthermia.²⁰ Interestingly, the GEMINI Study has identified genetic conditions in parents who have had lifelong signs/symptoms without a clear cause. One father reported a history of chorea and respiratory morbidities consistent with brain-lung-thyroid syndrome (pathogenic variant in NKX2-1) after his infant’s testing established the genetic diagnosis.²¹ It is likely that the rapid diagnostic capabilities of these testing platforms will translate into improved outcomes for both parents and their children.

Unlike many prior studies exploring the diagnostic capabilities and clinical utility of next-generation sequencing,^4-8,22,23 neonates and their parents in the GEMINI Study undergo simultaneous genetic analysis and variant interpretation on 2 distinct platforms. The challenges associated with discrepant clinical interpretation have previously been reported²⁴ and are caused in part by the required compilation of subjective, manual, and complex assertions that are collected from diverse sources.¹⁸ In published comparisons, discordance in variant classification between clinical laboratory directors ranged from 12% to 71%.^24-28 The GEMINI Study further highlights the challenges of integrating this technology into care. Although there was a 73% diagnostic concordance between platforms, infants with a genomic variant had discordant reports from the 2 laboratories 67% of the time. While 56% of these discrepancies were caused by the technical limitations of the targeted genomic panel, many were due to the unique variant interpretation used by each laboratory. These data are important for 2 reasons. First, despite the use of a targeted, neonatal-specific genomic platform, some neonates will require more comprehensive genomic coverage (ie, chromosomal microarray, whole-exome sequencing, rWGS). While the targeted genomic sequencing platform is capable of detecting small copy number variants (<1000 kilobases) associated with microduplications and microdeletions, the platform currently does not leverage any copy number or structural variant detection. Second, the discrepant interpretation of variant results provided by each laboratory prompted us to report preliminary findings before study completion. Each laboratory uses the same reported phenotypes and HPO terms to direct genomic interpretation. Computational settings used to filter and rank the variants identifies some as possibly causative and can fail to identify actual contributory variants. However, despite ACMG guidelines for the interpretation and reporting of variants detected on next-generation sequencing platforms,¹⁸ interpretation of the same variants at each laboratory also contributed to discordance. The ACMG guidelines are based on the association between reported findings in variant databases and/or the literature with their accompanying phenotypes. Therefore, by reporting our preliminary findings, along with the HPO terms that informed variant classification, we hope to improve variant detection and reporting for infants with similar phenotypes and highlight the strengths and potential limitations of these genomic platforms. This served as the primary impetus for this interim report.

The capabilities of each platform may also inform clinical interpretation. Patient 8 had 2 different P diagnoses, 1 detected by each platform. Based on the HPO terms, the targeted genomic sequencing platform diagnosed the infant with a maternally inherited pathogenic variant in FOXC2. On further discussion, it was determined that 6 family members, including the mother, had symptoms consistent with the FOXC2 variant, which causes lymphedema-distichiasis syndrome.²⁹ Conversely, despite detecting the FOXC2 variant, Rady Children’s Institute of Genomic Medicine determined that a paternally inherited 22q11.2 duplication (58 genes) in the same neonate was solely responsible for the infant’s phenotype of cleft palate, microretrognathia, and dysmorphic facies.³⁰ The infant’s diagnosis is a result of both genetic findings, made possible by simultaneously running both platforms. This demonstrates the need for more structured reporting guidelines as infants may present with more than 1 genetic disorder.

Although all patients have physical, clinical, and/or metabolic signs/symptoms highly suggestive of a genetic disorder, a specific cause was not identified for the majority of patients enrolled. The inability to diagnose patients is likely multifactorial. A rapid, phenotypically driven genome interpretation limits examination of the entire genome. Only genes known to be associated with clinical findings are interrogated on both platforms. Additionally, most genetic information obtained through WGS is not analyzed. Thus, it is possible that infants may actually have a genetic disorder that will be identified as analytical techniques and variant databases become more robust and/or the neonate develops additional phenotypic findings. It is also likely that some infants do not have a genetic cause for their clinical presentation. Teratogens and environmental exposures during key periods in gestation and/or epigenetic modifications may be contributory. Finally, the results remind us of our narrow understanding of the genome, including the role that introns likely play in genetic disease. Although technology is no longer a barrier to rapid genomic sequencing, we remain limited by our understanding and interpretation of these complex biologic processes.

This study is limited, in part, by its targeted, phenotypic-driven analysis based on our current understanding of the human genome. By only interrogating areas of the genome known to result in the patient’s presenting symptoms, a rapid return of results may be provided at the expense of a diagnosis.

Preliminary results of the GEMINI Study revealed that 51 of 113 infants (45%) had an important genetic variant detected. Fifty-one variants were novel and previously unpublished. While there was an overall 73% concordance between platforms for patients tested, of those with a positive finding, 67% received discordant results from the different methods. By testing 2 platforms simultaneously, GEMINI has highlighted the need for rapid dissemination of findings to better inform the field about novel variants and highlighted the existing variability in genomic sequencing technologies.

Corresponding Author: Jill L. Maron, MD, MPH, Mother Infant Research Institute, Tufts Medical Center, 800 Washington St, PO Box 394, Boston, MA 02111 (jmaron@tuftsmedicalcenter.org).

Accepted for Publication: September 30, 2020.

Published Online: February 15, 2021. doi:10.1001/jamapediatrics.2020.5906

Correction: This article was corrected on March 29, 2021, to fix typos in the funding program name and on June 21, 2021, to fix errors in Table 4.

Author Contributions: Dr Maron had full access to all of the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.

Concept and design: Maron, Kingsmore, Dimmock, Vockley, Diacovo, Stroustrup, Kurfiss, Davis.

Acquisition, analysis, or interpretation of data: Maron, Kingsmore, Wigby, Chowdhury, Dimmock, Poindexter, Suhrie, Vockley, Gelb, Stroustrup, Powell, Trembath, Gallen, Mullen, Tanpaiboon, Reed, Kurfiss, Davis.

Drafting of the manuscript: Maron, Kingsmore, Vockley, Trembath, Kurfiss, Davis.

Critical revision of the manuscript for important intellectual content: All authors.

Statistical analysis: Trembath, Tanpaiboon, Kurfiss.

Obtained funding: Maron, Kingsmore, Vockley, Davis.

Administrative, technical, or material support: Maron, Kingsmore, Wigby, Chowdhury, Dimmock, Poindexter, Diacovo, Stroustrup, Trembath, Gallen, Mullen, Tanpaiboon, Reed, Kurfiss.

Supervision: Maron, Kingsmore, Poindexter, Suhrie, Vockley, Davis.

Conflict of Interest Disclosures: Dr Maron reported grants from National Center for Advancing Translational Sciences (NCATS) during the conduct of the study. Dr Kingsmore reported grants from National Institutes of Health (NIH) during the conduct of the study. Dr Dimmock reported grants from NIH during the conduct of the study; personal fees from BioMarin Pharmaceutical, Audentes Therapeutics, and Ichorion Therapeutics outside the submitted work; and had a patent to US8718950B2 licensed. Dr Diacovo reported grants from University of Pittsburgh during the conduct of the study. Dr Stroustrup reported grants from NIH during the conduct of the study and outside the submitted work. Dr Gallen reported other from Tufts University during the conduct of the study and other support from Quest Diagnostics outside the submitted work. Dr Mullen reported other from Tufts University during the conduct of the study and other support from Quest Diagnostics outside the submitted work. Dr Tanpaiboon reported other from Quest Diagnostics during the conduct of the study. Dr Kurfiss reported grants from NCATS during the conduct of the study. Dr Davis reported grants from NIH during the conduct of the study. Drs Tanpaiboon, Mullen, and Mr Gallen receive compensation in the form of salary and stock from Athena Diagnostics/Quest Diagnostics. No other disclosures were reported.

Funding/Support: This study was supported by the National Center for Advancing Translational Sciences (NCATS) (grants U01TR002271 and UL1TR002544).

Role of the Funder/Sponsor: NCATS did not participate in the design and conduct of the study, the collection, management, analysis, and interpretation of the data or manuscript preparation and review. NCATS did approve the submission of interim findings of the ongoing Genomic Medicine in Ill Neonates and Infants (GEMINI) study.

Additional Contributions: We thank the families who graciously participated and the GEMINI team of investigators across all participating sites for their ongoing efforts to ensure the successful completion of this study.