The new england journal of medicine Review Article Frontiers in Medicine Next-Generation Sequencing to Diagnose Suspected Genetic Disorders David R. Adams, M.D., Ph.D., and Christine M. Eng, M.D.​​ linical next-generation sequencing is being used frequently in From the Office of the Clinical Director, medical practices in which genetic testing has traditionally taken place — National Human Genome Research In- stitute, and the Undiagnosed Diseases for example, medical genetics and medical subspecialties such as neuroge- Program, National Institutes of Health, C Bethesda, MD (D.R.A.); and the Depart- netics. Emerging diagnostic applications include rapid-reporting approaches in intensive care settings (especially neonatal and pediatric)1 and use early in the course ment of Molecular and Human Genetics, 2 Baylor College of Medicine, and Baylor of complex disease. Large-scale projects in the United States, China, and else- Genetics — both in Houston (C.M.E.). where are exploring and developing the role of clinical next-generation sequencing Address reprint requests to Dr. Adams at in precision medicine.3,4 This suggests a future in which genomic data will influence the Undiagnosed Diseases Program, Na- tional Institutes of Health, Bldg. 10, Rm. medical decision making for a diverse and growing group of patients (see video). 10C103E, 10 Center Dr., Bethesda, MD 20892, or at david . adams@ nih . gov. Clinical Next-Generation Sequencing N Engl J Med 2018;379:1353-62. as a Diagnostic Tool DOI: 10.1056/NEJMra1711801 Copyright © 2018 Massachusetts Medical Society. The laboratory techniques that are used in clinical next-generation sequencing have been described in numerous reviews5; proposed guidelines for their application to diagnostic testing have been published.6 The technology generates accurate and reliable sequence information for most parts of the genome. In a comparison of data from exome sequencing and Sanger sequencing (considered the standard of sequenc- ing) for 684 participants in five genes, the validation rate for the exome sequencing results was 99.97%. Furthermore, discrepant results in high-quality exome sequenc- An illustrated glossary and a ing regions were more likely to be correct in the exome sequencing data than in the video overview of first round of Sanger sequencing data.7 next-generation A clinical next-generation sequencing test can be designed to target a panel of se- sequencing are lected genes, the exome (all known genes, or approximately 1 to 2% of the genome), available at NEJM.org or the entire genome. Gene panels target curated sets of genes associated with specific clinical phenotypes. Phenotypes may be narrow, with 4 genes in the panel for familial hypercholesterolemia, or broad, with more than 1000 genes in the panel for intellectual disability. Clinical exome sequencing targets approximately 22,000 protein-coding genes. Clinical genome sequencing is untargeted, generating se- quence data from a region that is 50 to 100 times as large as that covered by exome sequencing and that includes regulatory, intronic, and intergenic regions (Fig. 1). Clinical decision making about which test to order is an area of active research. Genome sequencing generates more uniform sequencing in some regions than does exome sequencing. Emerging analytic approaches can use genome sequencing to detect structural variants and expansion of short nucleotide repeats associated with disease. However, bioinformatic tools for genome sequencing are overall less devel- oped than those available for exome sequencing. In addition, the cost of genome sequencing remains higher than that of exome sequencing, partly because of the cost of data management and analysis. n engl j med 379;14 nejm.org October 4, 2018 1353 The New England Journal of Medicine Downloaded from nejm.org at HOUSTON ACADEMY OF MEDICINE on October 4, 2018. For personal use only. No other uses without permission. Copyright © 2018 Massachusetts Medical Society. All rights reserved. The new england journal of medicine Gene 1 Gene 2 Gene 3 DNA Exon Exon Exon Exon Exon Exon Exon Intergenic Intron Difficult-to- region sequence region Sanger Sequencing NGS Gene Panel (only a selection of genes are targeted) Depth-of-coverage graph Aligned reads at base resolution: Gene 2 not targeted for sequencing ... AATCTGACA ... ... AATCTGACA ... ... AATCTGACA AATCAGACA Aligned NGS ... fragments ... AATCAGACA... (“reads”) Exome Sequencing Depth-of-coverage graph Aligned NGS fragments (“reads”) Genome Sequencing Intergenic Depth-of-coverage graph Exon region Intron Aligned NGS fragments (“reads”) The primary goal for any diagnostic genetic test tify potential risk variants for genetic disease that is the identification of DNA sequence variants that is absent or has not been diagnosed at the time of may be confidently associated with the presenting testing; these results are referred to as secondary, signs and symptoms. Other test results may iden- incidental, or medically actionable findings. Pa- 1354 n engl j med 379;14 nejm.org October 4, 2018 The New England Journal of Medicine Downloaded from nejm.org at HOUSTON ACADEMY OF MEDICINE on October 4, 2018. For personal use only. No other uses without permission. Copyright © 2018 Massachusetts Medical Society. All rights reserved. Next-Generation Sequencing and Genetic Disorders Figure 1 (facing page). Clinical Next-Generation ment categories: pathogenic, likely pathogenic, Sequencing (NGS) Test Types. likely benign, benign, and variant of unknown Exome, genome, and panel NGS tests have different significance. Databases of previously assessed vari- genomic coverage characteristics. NGS gene panel ants, such as ClinVar, have been established to tests cover a set of genes defined by the clinical diag- collect and distribute information about previous- nostic laboratory. The panel will typically cover genes ly interpreted variants.14 ClinVar uses a categorical associated with a set of related medical conditions (e.g., heritable epilepsy disorders). Exome sequencing rating system to indicate the level of evidence for covers the majority of known genes, including genes submitted interpretations. Variants are also priori- that have not yet been associated with human disease. tized on the basis of association with the pheno- Genome sequencing covers a majority of both genes type of the patient, although the possibility of and intergenic regions. Each test type has an associat- phenotypic heterogeneity and blended phenotypes ed pattern of false negative results. For instance, a gene panel may not include a mutated gene and an (more than one mendelian disorder manifesting 15 exome may miss deep intronic splice mutation. In ad- in an individual patient) must be considered. dition, some regions of the genome are difficult to se- Clinical laboratories primarily report variants quence with any existing method. in genes for which the gene–disease association is well established. In other cases, the proposed association will be novel, creating an “N = 1” situa- tients with these risk variants may benefit from tion (in which the diagnosis cannot be claimed early screening and management efforts. Guide- to be definitive) and the opportunity to establish lines for the clinical reporting of this category of a new gene–disease association.10 The risk of findings have been published.8 falsely associating diseases with genes and vari- ants is regularly illustrated by the reclassification Variant Classification of previously established pathogenic variants as the result of improvements in frequency databas- Next-generation sequencing generates thousands es.16 One innovative way to locate additional cases of sequence variants that must be filtered and pri- is through the use of matching databases. Gene- oritized for clinical interpretation, which results in Matcher (https://genematcher.org/), DECIPHER the reporting of a limited number of variants per (https://decipher.sanger.ac.uk/), and Phenome- report. This process may differ slightly among Central (https://www.phenomecentral.org/) iden- individual laboratories, but it generally includes tify matching cases with the use of deidentified annotation of variants, application of frequency data, such as gene names or disease features.17-19 filters and database searches to enrich for rare The Matchmaker Exchange protocol allows match- variants and eliminate common variants, and pre- es between such databases.20 These tools are pub- diction of functional effect. Clinical evaluation of licly available and do not require computational a DNA sequence variant includes an assessment of expertise. potential effects on the function of one or more genes and an assessment of the evidence support- Diagnostic R ate ing attribution of the illness at presentation to the and Testing Strategy affected gene or genes.9 Both assessments benefit from strong association information (e.g., variant Gene Panels to disease and absence of variant to absence of Gene panels (selected genes sequenced by a next- disease).10 However, such evidence may be difficult generation sequencing method) often have higher to obtain for rare variants or diseases. diagnostic rates than exome sequencing or ge- Variants are evaluated according to evolution- nome sequencing, being designed to maximize ary conservation, population frequency, and mod- coverage, sensitivity, and specificity for the includ- eled (or measured) effect on protein function. ed genes. An exception may occur in the context of Large-scale genomic sequencing databases, includ- greater