A semiautomated whole-exome sequencing workflow leads to increased diagnostic yield and identification of novel candidate variants.

Advancing the clinical utility of whole-exome sequencing (WES) for patients with suspected genetic disorders is largely driven by bioinformatics approaches that streamline data processing and analysis. Herein, we describe our experience with implementing a semiautomated and phenotype-driven WES diagnostic workflow, incorporating both the DRAGEN pipeline and the Exomiser variant prioritization tool, at an academic children's hospital with an ethnically diverse pediatric patient population. We achieved a 41% molecular diagnostic rate for 66 duo-, quad-, or trio-WES cases, and 28% for 40 singleton-WES cases. Preliminary results were returned to ordering physicians within 1 wk for 12 of 38 (32%) probands with positive findings, which were instrumental in guiding the appropriate clinical management for a variety of patients, especially in critical care settings. The semiautomated and streamlined WES workflow also enabled us to identify novel variants in candidate disease genes in patients with developmental delay and autism and immune disorders and cancer, including ANK2, BPTF, BCL11A, FOXN1, PLAA, ATRX, DNAJC21, and RAD50 Together, we demonstrated the implementation of a streamlined WES workflow that was successfully applied for both clinical and research purposes.

INTRODUCTION

Whole-exome sequencing (WES) has been enormously successful for both novel disease gene discovery (Bamshad et al. 2011; Ku et al. 2012) and in the clinical diagnostic setting to determine the genetic causes of rare Mendelian (monogenic) disorders (Lee et al. 2014; Yang et al. 2014; Farwell et al. 2015). Over the past few years, with the wide application of WES, a rich framework of experimental and analytical approaches has been established for elucidating the genes responsible for Mendelian disorders (Bamshad et al. 2011). Technical and informatics advancements have led to reduced cost and turnaround time, as well as increased diagnostic yield and impact on patients’ outcomes. The success rate has varied significantly across laboratories primarily because of (1) technical differences in capturing and sequencing approaches, (2) the patient population being studied and possible ascertainment bias, (3) the availability of large reference databases, (4) the bioinformatics pipelines used, and, most importantly, (5) a highly variable practice of variant interpretation, which has been largely manual and has not consistently followed unified guidelines. The overall WES diagnostic rate in unselected, consecutive patients has been reported to be ∼25% (Lee et al. 2014; Yang et al. 2014; Farwell et al. 2015; Dragojlovic et al. 2018). The efficiency of data analysis is limited by the need for multiple manual data interpretation steps, limited communication with ordering physicians, and, most importantly, limited adaptation of robust informatics tools to facilitate prioritization of variants.

Herein, we describe our clinical and research findings with implementing a semiautomated and phenotype-driven WES diagnostic workflow at a large academic children's hospital with an ethnically diverse pediatric patient population. The workflow was implemented, validated, and optimized using validation samples with known molecular diagnoses and then applied prospectively to clinical cases. We implemented the Edico DRAGEN Genome pipeline for rapid (∼4-min) variant calling, used standardized phenotypic description Human Phenotype Ontology (HPO) terms for primary gene selection, and incorporated Exomiser for automated variant prioritization. Using these approaches, results can be obtained in 72 h and can thus positively impact patient management in critical care settings. This workflow has a proven ability to identify novel variants and novel candidate disease genes automatically, especially for patients with developmental and immunologic disorders.

RESULTS

Clinical Presentation and Family History

The 106 clinical cases were referred from nine clinical divisions at the Children's Hospital Los Angeles including Medical Genetics (39), Neurology (26), Immunology (17), Neonatology and Pulmonology (12), Hematology & Oncology (5), Endocrinology (3), Dermatology (2), General Pediatrics (1), and Ophthalmology (1) (Fig. 1A). These included 40 singletons (proband only), 60 trios (typically proband and both parents), 5 duos (proband and one parent), and 1 quad (proband, parents, and affected sibling). Of the 106 probands, there were 63 (59%) males and 43 (41%) females. Patients’ age at the time of referral varied from newborn to 22 yr (median age 3 yr) (Supplemental Table 1). Patients presented with a range of phenotypes, as indicated by the 521 HPO terms mapped to patients’ clinical phenotypes (Supplemental Table 1). The most common clinical indications included “global developmental delay” (29%), “seizures” (21%), “generalized hypotonia” (19%), “microcephaly” (10%), “cognitive impairment” (9%), “short stature” (9%), “cleft palate” (8%), “immunodeficiency” (8%), “autism” (8%), “failure to thrive” (8%), “abnormality of the cerebral white matter” (8%), “delayed speech and language development” (7%), and “cryptorchidism” (7%) (Fig. 1B). The majority of patients had negative chromosomal microarray (CMA) results and many had single gene or gene panel testing that was uninformative. Three patients with positive CMA results were referred for WES because of concern for a second genetic disorder.

Figure 1.

Descriptive statistics of the 106 clinical WES cases. The 106 clinical cases were referred from nine divisions (A). Patients presented with a range of phenotypes. The most common 12 clinical phenotypes are listed (B).

WES Molecular Diagnoses and Findings

A total of 40 molecular diagnoses were obtained in 38 of 106 patients, achieving an overall diagnostic rate of 36%. The diagnostic rate was 28% (11/40) for singleton-WES and 41% (27/66) for duo-, quad-, and trio-WES, respectively. The diagnostic yield for trio-WES is equal to or higher than 12 of the 13 exome studies of pediatric patient populations (Dragojlovic et al. 2018). Of note, 42% (18/43) of patients between 1 and 5 years old had pathogenic or likely pathogenic diagnostic findings, higher than the diagnostic rate of 27% (6/22) for patients younger than 12 mo, and 34% (14/41) for patients between 6 and 22 yr (Table 1; Supplemental Table 1). Further prospective studies with a significantly larger patient population are required to draw any conclusions from this interesting observation.

Table 1.

Whole-exome sequencing results in 38 patients with molecular diagnoses and in four patients with likely pathogenic findings

Study ID	Study type	Gender	Age	Gene	Zygosity	Transcript ID	DNA change	Protein change	Classification	Inherited from	Inheritance	MIM ID	Exomiser rank
1	Trio	Male	4 yo	GRIN1	Het	NM_007327.3	c.2531G > T	p.Arg844Leu	Likely pathogenic	De novo	AD	614254	1
11	Trio	Female	4 yo	STAT1	Het	NM_007315.3	c.1154C > T	p.Thr385Met	Pathogenic	De novo	AD	614162	1
15	Trio	Male	8 yo	CREBBP	Het	NM_004380.2	c.5603G > A	p.Arg1868Gln	Likely pathogenic	De novo	AD	180849	1
				COL5A2	Het	NM_000393.3	c.1842G > A	p.Met614Ile	VUS	Mother			Not ranked
21	Singleton	Male	9 yo	CLN6	Homo	NM_017882.2	c.395_396delCT	p.Ser132Cysfs*18	Pathogenic	Mother and father	AR	601780	2
34	Trio	Female	7 yo	TMCO1	Homo	NM_019026.4	c.646_647delGC	p.Ala216Profs*53	Likely pathogenic	Mother and father	AR	213980	1
				DNAH11	Homo	NM_001277115.1	c.10472G > A	p.Arg3491His	VUS	Mother and father		611884	14
43	Trio	Female	5 yo	STAT1	Het	NM_007315.3	c.1154C > T	p.Thr385Met	Pathogenic	De novo	AD	614162	1
				SCNN1A	Het	NM_001159576.1	c.1604G > A	p.Arg535Gln	VUS	Father			Not ranked
47	Quad	Male	7 yo	TMCO1	Homo	NM_019026.4	c.616C > T	p.Arg206*	Pathogenic	Mother and father	AR	213980	1
52	Trio	Male	2 yo	FOXP3	Hemi	NM_014009.3	c.1190G > A	p.Arg397Gln	Pathogenic	Mother	XLR	304790	1
55	Duo	Male	6 yo	STAT1	Het	NM_007315.3	c.821G > A	p.Arg274Gln	Pathogenic	Nonmaternal	AD	614162	1
61	Trio	Male	2 yo	SCN1A	Het	NM_001202435.1	c.3327dupC	p.Ser1110Glnfs*14	Pathogenic	De novo	AD	607208	1
65	Duo+ sibling	Female	17 yo	FKBP10	Homo	NM_021939.3	c.831dupC	p.Gly278Argfs*95	Pathogenic	Mother and father	AR	259450; 610968	1
72	Trio	Male	22 yo	ATRX	Hemi	NM_000489.4	c.7156C > T	p.Arg2386*	Pathogenic	Mother (het)	XLR	309580	1
82	Trio	Female	20 mo	ALG13	Het	NM_001099922.2	c.50T > A	p.Ile17Asn	Likely pathogenic	De novo	XLD	300884	1
83	Trio	Female	19 mo	CHN1	Het	NM_001822.5	c.667G > A	p.Ala223Thr	Likely pathogenic	Father	AD	604356	Not ranked
				WFS1	Het	NM_006005.3	c.1999C > T	p.Gln667*	Pathogenic	Mother	AD	600965	2
				KIF1A	Het	NM_001244008.1	c.3052C > T	p.His1018Tyr	VUS	De novo			2
85	Duo	Female	2 yo	QARS	Het	NM_005051.2	c.1430A > G	p.Tyr477Cys	Likely pathogenic	Nonmaternal	AR	615760	1
				QARS	Het	NM_005051.2	c.1132C > T	p.Arg378Cys	Likely pathogenic	Mother		615760	1
88	Singleton	Male	9 yo	PTEN	Het	NM_000314.5	c.203A > G	p.Tyr68Cys	Pathogenic	Unknown	AD	158350; 605309	1
94b	Trio	Female	3 yo	PLAA	Homo	NM_001031689.2	c.1487-1G > A	p.?	Likely pathogenic	Mother and father	AR	617527	10
102	Trio	Male	22 mo	PACS1	Het	NM_018026.3	c.607C > T	p.Arg203Trp	Pathogenic	De novo	AD	615009	5
104	Singleton	Male	5 wo	MTM1	Hemi	NM_000252.2	c.1262G > T	p.Arg421Leu	Pathogenic	De novo	XLR	310400	1
109	Singleton	Male	2 mo	SFTPC	Het	NM_001172357.1	c.163C > T	p.Leu55Phe	Likely pathogenic	Mother	AD	610913	2
				DES	Het	NM_001927.3	c.407T > A	p.Leu136His	VUS	Father			1
115	Singleton	Male	2 wo	ABCA3	Homo	NM_001089.2	c.2279T > G	p.Met760Arg	Pathogenic	Unknown	AR	610921	1
126	Singleton	Male	8 yo	NSD1	Het	NM_022455.4	c.3839G > A	p.Trp1280*	Pathogenic	De novo	AD	117550	1
				FGF14	Het	NM_004115.3	c.486dup	p.Arg163Glnfs*7	Likely pathogenic	Father	AD	609307	27
127b	Trio	Female	5 yo	ANK2	Het	NM_001148.4	c.1417C > T	p.Arg473*	Likely pathogenic	De novo			1
130	Trio	Male	1 yo	SZT2	Het	NM_015284.3	c.4854-2A > G	p.?	Likely pathogenic	Mother	AR	615476	4
				SZT2	Het	NM_015284.3	c.5949_5951delTGT	p.Val1984del	Likely pathogenic	Father		615476	4
132	Trio	Male	15 yo	DNAJC21	Homo	NM_194283.3	c.544C > T	p.Arg182*	Pathogenic	Mother and father	AR	617052	3
				TP53a	Het	NM_000546.5	c.742C > T	p.Arg248Trp	Pathogenic	De novo			1
133	Singleton	Male	3 yo	COL17A1	Homo	NM_000494.3	c.4145_4148delAGAG	p.Glu1382Alafs*40	Pathogenic	Likely mother and father	AR	226650	1
135b	Singleton	Male	14 yo	RAD50	Het	NM_005732.3	c.2165dupA	p.Glu723Glyfs*5	Pathogenic	Unknown			36
144	Trio	Male	4 yo	SETD2	Het	NM_014159.6	c.5152G > A	p.Glu1718Lys	Likely pathogenic	De novo	AD	616831	1
				SPAST	Het	NM_014946.3	c.489A > G	p.Ile163Met	VUS	De novo		182601	2
156	Trio	Male	5 yo	BPTF	Het	NM_004459.6	c.5715_5716delAG	p.Val1906Glufs*15	Pathogenic	De novo	AD	617755	2
163	Trio	Male	6 yo	RPS26	Het	NM_001029.3	c.1A > C	p.Met1?	Pathogenic	De novo	AD	613309	1
165	Singleton	Male	10 yo	BCL11A	Het	NM_022893.3	c.152G > C	p.Cys51Ser	Likely pathogenic	Unknown	AD	617101	52c
167	Trio	Female	3 yo	SPTA1	Homo	NM_003126.2	c.7132C > T	p.Gln2378*	Likely pathogenic	Mother and father	AR	266140; 270970	1
				G6PD	Het	NM_000402.4	c.653C > T	p.Ser218Phe	Pathogenic	Mother (het)		134700; 300908; 611162	Not ranked
186	Singleton	Male	D.N.	PKHD1	Homo	NM_138694.3	c.1480C > T	p.Arg494*	Pathogenic	Likely mother and father	AR	263200	1
193	Trio	Female	2 yo	EIF2B5	Homo	NM_003907.2	c.943C > T	p.Arg315Cys	Pathogenic	Mother and father	AR	603896	1
194b	Singleton	Male	22 do	FOXN1	Het	NM_003593.2	c.1247delC	p.Pro416Glnfs*134	Likely pathogenic	Unknown		601705	20
198	Trio	Female	34 mo	ARID1B	Het	NM_020732.3	c.5968C > T	p.Arg1990*	Pathogenic	De novo	AD	135900	1
200	Singleton	Female	6 yo	PGAP3	Homo	NM_033419.4	c.851A > G	p.His284Arg	Pathogenic	Likely mother and father	AR	615716	1
				MYO15A	Homo	NM_016239.3	c.1454T > C	p.Val485Ala	VUS	Likely mother and father			35
				DEPDC5	Het	NM_001242896.1	c.2785G > A	p.Gly929Ser	VUS	Unknown			56
205	Trio	Female	9 mo	RAC2	Het	NM_002872.3	c.184G > A	p.Glu62Lys	Likely pathogenic	De novo	AD	608203	1
215	Trio	Male	3 yo	PPP2R1A	Het	NM_014225.5	c.773G > A	p.Arg258His	Pathogenic	De novo	AD	616362	1
216	Trio	Male	3 yo	SLC9A6	Hemi	NM_001042537.1	c.1728-19_1728-3del	p.?	Likely pathogenic	De novo	XLD	300243	1
219	Trio	Male	14 yo	FAM111B	Het	NM_198947.3	c.1462delT	p.Cys488Valfs*21	Likely pathogenic	De novo	AD	615704	6
224	Singleton	Male	11 mo	XIAP	Hemi	NM_001167.3	c.562G > A	p.Gly188Arg	Likely pathogenic	Mother (het)	XLR	300635	1
				GATA2	Het	NM_001145661.1	c.445G > A	p.Gly149Arg	VUS	Mother			3

(yo) Year old, (mo) month old, (wo) week old, (do) day old, (D.N.) deceased newborn, (Het) heterozygous, (Homo) homozygous, (Hemi) hemizygous.

aPotentially somatic alteration.

bPatients with potential diagnoses.

c52nd place using Exomiser version 7.2.1 at the time of initial analysis in 2016; 1st place using Exomiser version 10.1.1 in the end of this study.

Thirty-eight patients received a total of 40 molecular diagnoses, of which 52.5% (21/40) were autosomal dominant (AD) diseases, 32.5% (13/40) were autosomal recessive (AR) diseases, and 15% (6/40) were X-linked diseases (Tables 1 and 2). De novo variants accounted for 67% (14/21) of AD conditions. Of the six patients with X-linked diseases, three had de novo variants (Tables 1, 2; Supplemental Table 1). Two patients (Patient 83 and 126) had dual molecular findings (Table 1). Preliminary results were returned to the ordering physician in as few as 3 d and within 1 wk for 32% (12/38) of patients with positive findings, most of whom were in the neonatal intensive care unit. The timely delivery of the molecular diagnoses had an impact on the clinical management of these patients (see below). Follow-up family segregation analyses, particularly for singleton-WES cases, were helpful in further clarifying the variants’ pathogenicity. For example, the pathogenic variant classification was further supported for Patients 104 and 126 in whom the variant was confirmed to be de novo in the proband using targeted Sanger sequencing on both parents. In contrast, we reclassified a hemizygous BCORL1 variant from variant of uncertain significance (VUS) to likely benign (Patient 69) because of its cosegregation in two other unaffected male siblings (WES study IDs: 148 and 177) (Supplemental Table 1).

Table 2.

Modes of inheritance observed across 40 molecular diagnoses in 38 patients

Mode of inheritance	Number of diagnoses	Percent of diagnosis
Autosomal dominant		21 (52.5%)
De novo	14
Inherited from one of the parents	4
Inheritance unknown	2
Nonmaternal	1
Autosomal recessive		13 (32.5%)
Homozygous	11
Compound heterozygous	2
X-linked recessive and dominant		6 (15%)
De novo	3
Hemizygous inherited from the heterozygous mother	3

Interestingly, eight patients had a finding in genes associated with cancer predisposition (Table 3). These included pathogenic variants in DNAJC21, RAD50, and ATRX identified in patients with cancer, a PMS2 nonsense variant detected in a patient with a family history of colon cancer, and MEN1, RET, BRCA2, and CHEK2 variants in patients without a clinical or family history of cancer (Table 3). Notably, we were the first to report ATRX as a potential cancer predisposition gene for osteosarcoma using our WES platform (Patient 72) (Ji et al. 2017). The association between a germline pathogenic ATRX variant and osteosarcoma was subsequently reported (Smolle et al. 2017; Masliah-Planchon et al. 2018).

Table 3.

Pathogenic and likely pathogenic variants identified in genes associated with cancer predisposition

Exome study ID	Gender	Age	Clinical phenotypes	Gene	Zygosity	Variant identified	Inherited from	Referring source
132	Male	15 yr	Short stature, webbed neck, white patch on the tongue, intellectual disability, acute myeloid leukemia, pancytopenia, and pectus excavatum	DNAJC21	Homozygous	NM_194283.3: c.544C > T (p.Arg182*)	Mother and father	Heme/Oncology
135	Male	14 yr	History of unilateral retinoblastoma and pilocytic astrocytoma, mild ataxia, brachydactyly, dental malocclusion, high, narrow palate, pigmented freckles and right shoulder granular tumor	RAD50	Heterozygous	NM_005732.3: c.2165dupA (p.Glu723Glyfs*5)	Unknown	Genetics
72	Male	22 yr	Dysmorphic features, microcephaly, severe intellectual disability (nonverbal), mild anemia, cryptorchidism and osteosarcoma	ATRX	Hemizygous	NM_000489.4: c.7156C > T (p.Arg2386*)	Mother (heterozygous)	Heme /Oncology
174	Male	2 wk	Respiratory insufficiency and pulmonary hypertension. Maternal family history of colon cancer	PMS2	Heterozygous	NM_000535.5: c.1927C > T (p.Gln643*)	Mother	Neonatology/Pulmonology
193	Female	2 yr	Developmental delay, leukodystrophy, vanishing whiter matter and dystonia	MEN1	Heterozygous	NM_130803.2:c.1633C > T (p.Pro545Ser)	Mother	Neurology
127	Female	5 yr	Microcephaly, global developmental delay, delayed speech and language development, aggressive behavior, seizures, epileptic spasms, infantile spasms, hyperammonemia, and intellectual disability	RET	Heterozygous	NM_020975.4: c.1907C > T (p.Thr636Met)	Father	Genetics
87	Male	16 yr	Ectopic ossification, muscle weakness, nephrolithiasis, asthma, headache, migraine, obesity, short attention span, and striae distensae	BRCA2	Heterozygous	NM_000059.3:c.4936_4939delGAAA (p.Glu1646Glnfs*23)	Father	Endocrinology
110	Male	20 mo	Colitis, hematochezia, inflammation of the large intestine, and thrombocytopenia	CHEK2	Heterozygous	NM_007194.3: c.1100delC (p.Thr367Metfs*15)	Unknown	Immunology

Exomiser Results for WES Clinical Cases and Identification of Novel Candidate Variants

For the 43 reported pathogenic and likely pathogenic variants in 38 clinical cases, Exomiser ranked 72.1% (31/43) variants as the top candidate in 30 cases, 20.9% (9/43) in second to sixth place in seven cases, and 6.9% (3/43) at 27th place or not ranked in three cases (Table 1). Together, 93% (40/43) of pathogenic or likely pathogenic variants were ranked in the top 6 by Exomiser. The Exomiser rank for sequencing variants identified in the clinical cases is shown in Table 1.

Updated Molecular Diagnoses Using Recent Gene–Disease Associations

The highest rank by Exomiser initially for candidate variants in BPTF, DNAJC21, PLAA, and BCL11A were the second place, third place, 10th place, and the 52nd place for Patient 156, 132, 94, and 165, respectively (Table 4). There were no published disease associations with these genes when WES was initially performed, and therefore according to ACMG guidelines, they should not have been clinically reported. However, the nature of the variants indicated that these variants were potentially disease-causing. These four genes were later shown to have clear disease associations (Dias et al. 2016; Tummala et al. 2016; Hall et al. 2017; Stankiewicz et al. 2017). Stankiewicz et al. first described the gene disease association for BPTF, an AD neurodevelopmental disorder with dysmorphic facies and distal limb anomalies (MIM: 617755) (Stankiewicz et al. 2017), which largely overlaps with our patients’ clinical presentation (Patient 156). This patient had a de novo truncating variant in BPTF. A homozygous canonical splice-site variant in PLAA was identified in a family (case 94) with two affected children suspected to have an AR disease. The PLAA gene was very recently shown to be associated with an AR neurodevelopmental disorder with progressive microcephaly, spasticity, and brain anomalies by Falik Zaccai et al. and Hall et al., subsequent to the original clinical study for this family (Falik Zaccai et al. 2017; Hall et al. 2017). Similarly, since the time of the original analysis for Patient 132 new studies have been published that described an AR bone marrow failure syndrome (MIM: 617052) caused by germline biallelic pathogenic variants in DNAJC21 (Tummala et al. 2016; Dhanraj et al. 2017; Bluteau et al. 2018). The disorder is characterized by onset of pancytopenia in early childhood. Patients may have additional variable nonspecific abnormalities, including poor growth, microcephaly, and skin anomalies (Tummala et al. 2016; Dhanraj et al. 2017), which also fits the clinical presentation of our patient. The proband had a homozygous nonsense variant in DNAJC21 inherited in an AR manner from both parents who were heterozygous for this variant. Patient 165 had a novel BCL11A missense variant identified in a highly conserved region of the gene where missense variants are depleted in the general population (ExAC missense constraint z-score = 5.52). Since the time of initial exome analysis, new publications have described the AD Dias–Logan syndrome (MIM: 617101), caused by de novo heterozygous pathogenic variants in the BCL11A gene (Dias et al. 2016). The disorder is characterized by delayed psychomotor development, intellectual disability, variable dysmorphic features, including microcephaly, downslanting palpebral fissures, strabismus, and external ear abnormalities, as well as asymptomatic persistence of fetal hemoglobin, which overlaps the patient's clinical presentation. Revised clinical reports were therefore issued that included the novel pathogenic or likely pathogenic variants for these patients. Our findings enrich the phenotypic characteristics of these newly established syndromes.

Table 4.

Potential molecular diagnosis at the time of WES analysis

Study ID	Study type	Gender	Gene	Chromosomal position (GRCh37)	HGVS DNA reference	HGVS protein reference	Variant type	Predicted effect	dbSNP/dbVar ID	Genotype	Parent of origin	Exomiser rank	Supporting references
156a	Trio	Male	BPTF	Chr 17:65,914, 863_65,914,864	NM_004459.6: c.5715_5716delAG	p.Val1906Glufs*15	Deletion	Frameshift	NA	Heterozygous	De novo	2	Stankiewicz et al. 2017
132a	Trio	Male	DNAJC21	Chr 5:34,937,536	NM_194283.3: c.544C>T	p.Arg182*	Substitution	Nonsense	rs771063992	Homozygous	Mother and Father	3	Tummala et al. 2016Dhanraj et al. 2017Bluteau et al. 2018
94b	Trio	Female	PLAA	Chr 9:26,913,946	NM_001031689.3: c.1487-1G>A	p.?	Substitution	Splice-site	rs1426488816	Homozygous	Mother and Father	10	Falik Zaccai et al. 2017Hall et al. 2017
165a	Singleton	Male	BCL11A	Chr 2:60,773,339	NM_022893.3: c.152G>C	p.Cys51Ser	Substitution	Missense	NA	Heterozygous	Unknown	52d	Dias et al. 2016Cai et al. 2017Yoshida et al. 2018
72a	Trio	Male	ATRXc	Chr X:76,776,310	NM_000489.4: c.7156C>T	p.Arg2386*	Substitution	Nonsense	rs122445099	Hemizygous	Mother (heterozygous)	1	Masliah-Planchon et al. 2018Ji et al. 2017Smolle et al. 2017
135b	Singleton	Male	RAD50	Chr 5:131,931,451	NM_005732.3: c.2165dupA	p.Glu723Glyfs*5	Duplication	Frameshift	rs397507178	Heterozygous	Unknown	36	Heikkinen et al. 2006Walsh et al. 2011Lhota et al. 2016Lin et al. 2016
127b	Trio	Female	ANK2	Chr 4:114,186,083	NM_001148.4: c.1417C>T	p.Arg473*	Substitution	Nonsense	NA	Heterozygous	De novo	1	Koch et al. 2008Willsey et al. 2013Iossifov et al. 2012Farwell et al. 2015
194b	Singleton	Male	FOXN1	Chr 17:26,861,836	NM_003593.2: c.1247delC	p.Pro416Glnfs*134	Deletion	Frameshift	NA	Heterozygous	Unknown	20	Dominant inheritance not yet been published

aDiagnoses supported by later publications.

bCases with candidate variant/potential diagnoses.

cAssociation between osteosarcoma and the ATRX germline pathogenic variant was not known at the time of WES analysis.

d52nd place using Exomiser version 7.2.1 at the time of initial analysis in 2016, 1st place using Exomiser version 10.1.1 in the end of this study.

Identification of Candidate Variants in ANK2, RAD50, and FOXN1

Unlike the above described cases, exome analysis revealed a de novo heterozygous nonsense NM_001148.4:c.1417C > T (p.Arg473*) variant in the ANK2 gene for Patient 127 (Table 4) , who was referred for testing at age 5 yr with severe global developmental delay, seizures, aggressive behavior, and microcephaly. Pathogenic variants in ANK2 are known to be associated with AD ankyrin-B-related cardiac arrhythmia and long QT syndrome 4 (MIM: 600919), which did not explain the patient's primary clinical concerns. However, Exomiser ranked this variant in the first place likely because of its de novo and truncating nature. RNA-seq expression data from the Genotype-Tissue Expression Project suggests that ANK2 is expressed in brain tissue. Animal model experiments demonstrated that ank2 is required for synaptic stability in Drosophila (Koch et al. 2008). Although the ANK2 gene currently has not yet been associated with any human neurodevelopmental phenotypes in the OMIM database, there is more evidence supporting a correlation between a de novo ANK2 variant and a human autism spectrum disorder and neurological phenotypes (Iossifov et al. 2012; Willsey et al. 2013; Farwell et al. 2015; Stessman et al. 2017). We therefore consider this de novo truncating variant in ANK2 as a strong candidate to explain the patient's clinical phenotypes. More published data are needed to further support its association with a human neurological disorder.

Patient 135 had multiple primary tumors including unilateral retinoblastoma, astrocytoma, and a right shoulder granular cell tumor. RB1 gene sequencing and deletion/duplication testing was negative. WES revealed a heterozygous NM_005732.3:c.2165dupA (p.Glu723Glyfs*5) variant in RAD50 (Table 4). Germline heterozygous RAD50 loss-of-function mutations have been described in association with hereditary predisposition to cancers of the breast, ovary, fallopian tube, and peritoneum (Heikkinen et al. 2006; Walsh et al. 2011). Additionally, this variant is absent from the population data set gnomAD. However, based on currently available data, precise cancer risks associated with this gene are not well-described. It is also unclear whether germline pathogenic variants in RAD50 are associated with an increased risk for retinoblastoma and pilocytic astrocytoma seen in this patient. Additional studies on the tumor and parental studies may further inform the association between the pathogenic variant in RAD50 and the clinical presentation seen in this patient. We therefore considered this RAD50 frameshift variant to be another candidate variant for this patient's predisposition to cancer.

Our Exomiser analysis also highlights potential discovery of a new inheritance pattern for known disease genes. An example is the identification of a heterozygous NM_003593.2:c.1247delC (p.Pro416Glnfs*134) frameshift variant in the FOXN1 gene in a 22-d-old male with severe T-cell immunodeficiency, aplasia of the thymus, with normal B and NK cells (Patient 194; Table 4). Pathogenic variants in FOXN1 are associated with AR T-cell immunodeficiency, congenital alopecia, and nail dystrophy (MIM: 601705). Although exome coverage for the FOXN1 exonic regions and splice-site junctions (5 bp into introns) was 100% at a minimum of 10× depth of coverage, a second possible variant, as expected for an AR disorder, was not identified by this analysis. The possibility this individual harbors a second variant, including a deletion or duplication of one or more exons or a variant in the intronic or regulatory regions that is undetectable by this test, cannot be excluded. However, a single heterozygous FOXN1 truncating variant without a second hit using whole-genome sequencing was observed in a different patient with similar phenotypes (data not published), which suggests a potential dominant inheritance for disease. Although no functional studies have been performed to test the effects of this particular variant on FOXN1 protein function or stability, this variant is expected to replace the carboxy-terminal portion of the FOXN1 with 134 aberrant amino acids. Additionally, the variant is not present in the population database gnomAD. Given that this is a frameshift variant in a gene in which frameshift variants are likely to cause disease, and the absence of this type of variant in the general population, we consider this variant likely pathogenic. Further functional and parental studies are warranted to further clarify its pathogenicity and its potential dominant inheritance.

WES Secondary Findings

Secondary findings deemed to be medically actionable were reported in nine of 101 patients who opted in to receive secondary findings (Supplemental Table 1). These included variants within the 59 genes recommended by the ACMG (Kalia et al. 2017), as well as pathogenic or likely pathogenic variants in MEFV, MYOC, and CHEK2. As expected, for incidental findings, the Exomiser rank was much lower than for those relevant to patients’ clinical phenotypes, with only 4/11 variants ranked between top 2 to 5.

Clinical Impact of Our Rapid WES Test

An accurate genetic diagnosis is essential for effective clinical management and may lead to modified treatment strategies, especially for pediatric patients compared to adults. We have implemented a WES workflow that allows for rapid diagnosis leading to improvements in patient management as exemplified by the following examples.

Patient 224 was an 11-mo-old male who presented with emesis, failure to thrive, hematochezia, and low serum concentration of IgG since age 5 mo. Endoscopy showed moderate to severe rectosigmoid colitis with ulcerations. The clinical picture was believed to reflect very early onset inflammatory bowel disease (VEO-IBD) prompting consideration of rare variants in NOD2 or IL10RA and IL10RB. Expedited WES was pursued and a maternally inherited XIAP NM_001167.3:c.562G > A (p.Gly188Arg) likely pathogenic hemizygous variant was identified. The results were returned to the ordering physician within 5 d. Follow-up targeted Sanger sequencing for the XIAP variant indicated that the mother was a heterozygous carrier for this variant who was not clinically affected. Pathogenic variants in XIAP are associated with X-linked recessive lymphoproliferative syndrome, type 2, which is characterized by primary immunodeficiency due to hemophagocytic lymphohistiocytosis and IBD. Allogeneic hematopoietic cell transplantation is the only curative treatment for this condition, and the patient's father is currently undergoing an evaluation to be a potential donor. This case highlights the ability to use WES to search beyond a narrow panel-based approach to obtain a diagnosis. In this case, and in similar ones reported previously (Kelsen et al. 2015; DeWeerdt 2016), a correct, early diagnosis markedly changed management (allogeneic hematopoietic cell transplantation) and prevented the patient from having to endure ineffective treatments.

Patient 104 was a 5-wk-old male with neonatal onset diffuse hypotonia in conjunction with muscle weakness, microcephaly, and respiratory distress. The infant was unable to be weaned from oxygen, had inspiratory stridor, and required nasogastric tube feeding. Decisions regarding placement of a gastrostomy tube and utility of a muscle biopsy for diagnosis prompted a request for an expedited WES, which was conducted within 1 wk. Exome sequencing revealed a hemizygous NM_000252.2:c.1262G > T (p.Arg421Leu) pathogenic variant in MTM1, encoding myotubularin. Pathogenic variants in MTM1 are associated with X-linked recessive myotubular myopathy (XLMTM). XLMTM presents prenatally with decreased fetal movements and polyhydramnios and affected males have significant motor delays and most fail to achieve independent ambulation. After informed discussion with the family, the treatment team proceeded with G tube placement and plans for long-term ventilatory support were discussed. The health-care management plans for this patient will include long-term follow-up care with a pulmonologist and neurologist. Annual evaluations with an ophthalmologist are indicated to screen for ophthalmoplegia, ptosis, and myopia. For this patient, the family had consented to receive incidental findings. A heterozygous NM_000218.2:c.1085A > G (p.Lys362Arg) likely pathogenic variant in the KCNQ1 gene was detected as an incidental finding. Pathogenic variants in KCNQ1 are classically associated with long QT syndrome (MIM: 192500). This finding directly impacts follow-up and management of the patient, and further analysis revealed that the variant was inherited from the patient's father, who was then referred to a cardiologist for evaluation.

DISCUSSION

Incorporation of Exomiser into the Clinical Diagnostic Workflow Can Potentially Improve the Overall Diagnoses

Previous data have shown that phenotype-driven exome analysis and comparison of human to model organism phenotypes markedly improves the ability to identify candidate disease-causing variants (Zemojtel et al. 2014; Smedley et al. 2015). However, this approach has yet to be proven in a clinical diagnostic setting. Our group customized and validated multiple bioinformatics tools, including Exomiser, for clinical diagnosis. The algorithms use model species (e.g., mouse, fish, rat), functional data, and protein–protein association data for inferring gene–variant–disease association through phenotype-driven associations (Smedley et al. 2015). By effectively utilizing available phenotypic information, functional data, and genomic knowledge, phenotype-driven strategies can improve efficiency and diagnostic yield. We used an earlier version of Exomiser (version 7.2.1) at the time of WES analysis and the latest version at the end of this study (version 10.1.1). Novel variants in candidate genes, including BPTF, DNAJC21, PLAA, BCL11A, ANK2, FOXN1, and RAD50, were identified or supported using this approach, which combined with updated literature sources allowed us to resolve an additional three cases. Our diagnostic yield of 36% includes the four patients with pathogenic or likely pathogenic variants in ATRX, BPTF, BCL11A, and DNAJC21 as the phenotypes of these patients are consistent with the current literature (Dias et al. 2016; Tummala et al. 2016; Dhanraj et al. 2017; Smolle et al. 2017; Stankiewicz et al. 2017; Masliah-Planchon et al. 2018). In a conservative fashion, at this time we have not included the patients with variants in ANK2, RAD50, FOXN1, and PLAA as the gene–disease association has not been fully established (ANK2, RAD50); the clinical phenotypes of our patient do not fit completely with what have been previously described (PLAA); or the observed potential inheritance pattern has not yet been proven (FOXN1). We anticipate that others may identify and report patients with similar clinical phenotypes that have pathogenic variants in at least some of these genes, which would further increase our diagnostic rate.

Cancer Predisposition

In this case series, a total of five patients had a clinical indication for cancer predisposition (Patient 58, 72, 132, 135, and 174). Pathogenic or likely pathogenic variants were identified for four of the five patients (Table 3); three of which (ATRX, DNAJC21, and RAD50) were novel at the time of analysis. Additionally, four pathogenic or likely pathogenic variants were identified in known cancer predisposition genes as incidental findings, including variants in RET, MEN1, BRCA2, and CHEK2. The total number of patients with findings in a cancer gene is higher than expected, given the unselected nature of this patient population. Identifying these patients may lead to modified treatment strategies in cases of syndrome-related increased toxicity or resistant disease. Cancer surveillance is warranted for at-risk relatives. Ultimately, systematic monitoring of patients and family members who have germline cancer predisposition variants may allow for detection of cancers at their earliest and most curable stage, thereby improving patient outcomes.

Conclusion

We have implemented a phenotype-driven semiautomated workflow for WES at an academic children's hospital, which enables us to efficiently utilize WES in a frontline diagnostic setting. Aside from achieving a high diagnostic yield, we were able to reduce the turnaround time for clinical diagnosis, which impacts clinical management for a number of patients. Furthermore, the incorporation of Exomiser into the workflow has allowed us to identify a number of strong candidate variants and disease genes—some of which were supported by later publications. Combined, we established a successful WES workflow for increased diagnostic yield in the clinical setting, and strong evidence for novel variants in a research setting.

METHODS

WES Workflow and Exomiser Validation

We established the workflow using 27 validation cases (Supplemental Table 2). These include the HapMap trio samples (NA12878, NA12891, and NA12892), 25 singleton-WES, as well as 2 trio-WES (parents and affected child) cases with known pathogenic, likely pathogenic, and VUS selected by the clinical laboratory directors. The samples were coded and the data were processed and analyzed in a blinded manner to determine the analytical and clinical accuracy, respectively.

The FreeBayes variant caller failed to call a 13-bp deletion in the 5′ region of RB1 exon 17 (NM_000321.2) for one case (Supplemental Table 2, validation case 19). For the other 26 cases with 34 previously reported variants, including 27 pathogenic and likely pathogenic variants and seven VUSs, manual review of the exome data in a blinded manner achieved (26/26) 100% concordance rate with the prior diagnoses.

Additionally, we used Exomiser for 26 validation cases (except validation case 19) with known positive results. For pathogenic and likely pathogenic variants, Exomiser ranked 76% (22/29) variants in first place and 24% (7/29) in the second to sixth place. One likely pathogenic MLH1 variant (Validation case 16) was reported as an incidental finding and was ranked in the sixth place under the AD model because of partial phenotype overlap. For the seven VUSs, Exomiser ranked three in the first place, one in the second place, one in the seventh place, one in the 16th place and one unranked (Supplemental Table 2). For validation case 12, Exomiser initially ranked the known pathogenic variant in seventh place using a general HPO term “HP:0000819 Diabetes mellitus”. When a more specific HPO term “HP:0004904 Maturity-onset diabetes of the young” was used, however, the ranking was elevated to first place.

Clinical WES for Patients

A total of 106 consecutive, unselected patients were referred to the Center for Personalized Medicine (CPM) for WES testing from inpatient units and outpatient clinics at the Children's Hospital Los Angeles (CHLA) (Supplemental Table 1). Genetic counseling and consent was required prior to WES and parents were provided with an option to receive incidental or secondary findings (Kalia et al. 2017; Green et al. 2013). This study was reviewed and approved by the CHLA Institutional Review Board.

Exome Sequencing Library Preparation, Sequencing, and Bioinformatics Pipeline

DNA was extracted from peripheral blood using a commercially available kit (Promega Maxwell RSC DNA Extraction Kit). The WES library was generated using the Agilent SureSelect Human All Exon V6 plus a custom mitochondrial genome capture kit, designed in collaboration with Agilent for consistent coverage across the mitochondrial genome (Falk et al. 2012). Paired-end 2 × 100 bp sequencing was performed using the Illumina NextSeq 500 sequencing system. The overall exome sequencing workflow is illustrated in Figure 2, and the bioinformatics methodologies are detailed in the Supplemental Information. In brief, we used a combination of the bcbio pipeline (https://github.com/chapmanb/bcbio-nextgen), custom scripts and open-source bioinformatics tools to fully automate quality control (QC), sequence alignment, coverage analysis, variant calling, and initial filtering based on ExAC (Exome Aggregation Consortium, http://exac.broadinstitute.org/) allele frequencies. The coverage statistics are summarized in Supplemental Figure 1 and detailed in Supplemental Table 3. The variant statistics are detailed in Supplemental Information.

Figure 2.

Clinical exome workflow at the Center for Personalized Medicine, Children's Hospital Los Angeles. Patients with a suspected genetic disease are recommended to undergo clinical exome sequencing after genetic counseling based on clinical evaluation and other available genetic test results. Variants generated from the bioinformatics pipeline for manual review are also processed via Exomiser. Candidate variants were then reviewed at the Exome Review Board before Sanger confirmation and final reporting.

Variant Calling with Edico DRAGEN Genome Pipeline

We also implemented the DRAGEN (Dynamic Read Analysis for Genomics) Genome pipeline from Edico Genome for fast FASTQ to VCF variant analysis, at roughly 4 min per exome. The DRAGEN provides hardware-accelerated analysis of next-generation sequencing (NGS) data using a field-programmable gate array (FPGA). We validated the DRAGEN Genome pipeline and achieved 99.4% and 99.9% sensitivity and specificity in overall variant calling when compared with Genome In A Bottle (GIAB; https://www.nist.gov/programs-projects/genome-bottle) calls, and 100% concordance rate for clinically significant variants for all 37 cases with positive findings.

Standardized Clinical Phenotype Encoding With HPO

For each patient, a list of phenotypic key words was generated by a clinical geneticist and a genetic counselor after review of the clinical and family history, requisition form and relevant medical notes which were then mapped to standardized HPO terms (Supplemental Tables 1, 2; Köhler et al. 2014, 2017; Robinson et al. 2014) and captured in CPM's PhenoTips instance (Girdea et al. 2013). A custom HPO annotator tool was used to facilitate the conversion of free-text clinical descriptions to matching HPO terms through natural language processing and weighted mapping of the descriptions against the tiered HPO terms, synonyms, and term definitions. With a free-text input of clinical phenotype descriptions, the HPO annotator prompts a list of potentially matching HPO terms through weighted mapping of the descriptions against the tiered universe of HPO term names, synonyms, and term definitions. Likely matching HPO terms are ranked by composite scores of semantic similarity. Those deemed to most accurately reflect the patient's clinical phenotype terms were selected, commented, and captured into CPM's PhenoTips instance. For downstream analysis, we also developed a custom and web-based system that generates the primary gene list based on HPO's frequent term-to-gene associations. In addition to the genes directly generated by the input term, we found genes for all the descendent terms, which are more specific, following the HPO tree structure.

A primary gene list was generated based on HPO's term-to-gene associations (Supplemental Fig. 2; Girdea et al. 2013).

Variant Filtration, Annotation, Analysis, and Interpretation

Variant filtration included three major steps. First, variants were restricted to the exonic regions and splice-site junctions (5 bp into the intron). Second, synonymous variants that were not previously reported as pathogenic in the Human Gene Mutation database (HGMD) and not predicted to alter splicing were filtered out. Last, common variants with minor allele frequency (MAF) of >1% were filtered out. MAF is estimated using combined reference data sets: Genome Aggregation Database (gnomAD, http://gnomad.broadinstitute.org) (after March 2017) and Exome Aggregation Consortium (ExAC, http://exac.broadinstitute.org) (before March 2017). The remaining variants were annotated and analyzed using a commercial platform (Agilent Cartagenia Bench Lab NGS v5.0). For WES-trio, all de novo, homozygous, hemizygous (if male), and compound heterozygous variants were manually reviewed. For inherited heterozygous variants, only variants within the primary gene list were manually reviewed. For proband-only WES, all homozygous, hemizygous, and potentially compound heterozygous variants were manually reviewed. For heterozygous variants, only those within the primary gene list were manually reviewed. Additionally, all variants previously reported as disease-causing in the HGMD or ClinVar databases were reviewed. If a disease-causing variant was not identified, truncating variants in genes known to cause human diseases were reviewed. Each variant was interpreted with extensive database and literature review to consider clinical relevance, inheritance pattern, family segregation, disease mechanism, and variant level evidence (e.g., population allele frequency, evolutionary conservation, computational prediction, variant location) following ACMG standards and guidelines (Richards et al. 2015). A variant interpretation checklist was used to ensure consistency.

Integration of Automated Exomiser Prioritization

In parallel with manual review, we incorporated an automated variant prioritization tool, Exomiser (Smedley et al. 2015) (version 7.2.1 and subsequently 10.1.1), into our WES diagnostic workflow. The Exomiser application ranks genes and variants by combining clinical phenotype relevance and predictions as to whether the variant is deleterious (Smedley et al. 2015). Variant prioritization is done with two methods, PhenIX and hiPHIVE, for each of three inheritance models: AD, AR and X-linked. PhenIX is based on known gene–disease associations in humans and is therefore more suitable for clinical diagnosis; hiPHIVE uses supporting evidence from functional studies in model species and protein-to-protein associations, and therefore has the potential for novel disease gene discovery. Four filters (target, frequency, pathogenicity, and inheritance) were applied to the candidate variants. All variants that were filtered out were not ranked. The remaining candidate variants were numerically ranked from the first to the final number of variants.

Exome Board Review

Relevant candidate variants selected from both manual review and the top 20 candidate variants ranked by Exomiser were reviewed in the context of patients’ clinical phenotypes at the Exome Review Board. This board includes ordering physicians, clinical geneticists, laboratory directors, bioinformaticians, and genetic counselors.

Sanger Sequencing Confirmation and Follow-Up Family Segregation Studies

Sanger sequencing was performed to confirm all the clinically relevant variants prior to final reporting and for follow-up family segregation studies.

ADDITIONAL INFORMATION

Data Deposition and Access

Variants reported in this study have been detailed in the Supplemental Tables. Variants have also been deposited at NCBI ClinVar (https://www.ncbi.nlm.nih.gov/clinvar/) under accession numbers SCV0006800 92.1, SCV000692541.1–SCV000692559.1, and SCV000854432.1–SCV000854564.1. Raw sequencing data is not available under the IRB used to perform this study.

Ethics Statement

All testing performed was part of clinical patient management and not on a research basis. The study was approved by the Children's Hospital Los Angeles Institutional Review Board (CHLA-17-00374: CPM Exome Study).

Acknowledgments

The authors gratefully acknowledge extremely valuable scientific discussions with our clinical colleagues Drs. Alvaro Serrano, Amy Yeh, Andrew Dietz, Ashley Margol, Ashwini Lakshmanan, Caitlin Montgomery, Catherine Karimov, Cecilia Fu, Cheryl Lew, Christopher Kuo, Cindy Wong, Daniela Schweitzer, Deborah Holder, Deborah Krakow, Debra Stern, Denizhan H. Akan, Derek Wong, Diana Ferman, Divya Vats, Elizabeth Vandyke, Erin Lee, Etai Adam, Eugenia Ho, Eunice Hagen, Gail Hubbard, Jamie Frediani, Jay Desai, Jennifer Cotter, Jonathan Tam, Joseph A. Church, Julie Jaffray, Jyes Querubin, Kathryn Fletcher, Kathy Langevin, Kiarash Sadrieh, Leigh Ramos-Platt, Leo Mascarenhas, Linda Randolph, Manvi Bansal, Mark Borchert, Mark Selleck, Matt Lallas, Mattew Oberley, Megan Langille, Melinda Braskett, Michele Van Hirtum-Das, Naim Salama, Neena Kapoor, Nusrat Ahsan, Pedro Sanchez, Phillippe Friedlich, Pisit Pitukcheewanont, Quyen Luc, Rachel Chapman, Rebecca DeMaria, Rebecca Ellen Meyer, Robert Vasko, Ronald Ferdman, Saranya Veluswamy, Srikumar Nair, Susanna Sorrentino, Tena L. Rosser, Theodora Stavroudis, Thomas Hofstra, Tiffany Restelli, Van Hirtum-Das, and Wendy Mitchell. In addition, we thank David Ruble, Yuxia Zhan, Cindy Fong, Vandana Mehta, Dejerianne Ostrow, and the clinical laboratory team for their technical assistance.

Author Contributions

J.A.B. and X.G. conceived and supervised the study. J.J. led the study and performed the bulk of variant interpretations according to the ACMG guideline. J.B., D.T.M., and M.B. carried out all analytical tasks while validating the exome data analysis pipeline. M.B. implemented exome data analysis pipeline and performed sequence analysis and variant calling. L.S. implemented bioinformatics tools for mapping clinical descriptions to HPO terms and performed Exomiser analysis. T.T. computationally determined the genetic ancestry of all cases. C.Q., G.R., and S.C.S. contributed to phenotype ascertainment and variant interpretation. All authors participated in the Exome Review Board and contributed to acquisition, analysis, and interpretation of data. J.J., J.A.B., and X.G. drafted the manuscript. All authors contributed to the development of the manuscript.

Funding

This study was supported by the Department of Pathology and Laboratory Medicine at the Children's Hospital Los Angeles.

Competing Interest Statement

The authors have declared no competing interest.