Narrowing the Diagnostic Gap: Genomes, Episignatures, Long-Read Sequencing and Health Economic Analyses in an Exome-Negative Intellectual Disability Cohort

Abstract

PURPOSE:

Genome sequencing (GS)-specific diagnostic rates in prospective tightly ascertained exome sequencing (ES)-negative intellectual disability (ID) cohorts have not been reported extensively.

METHODS:

ES, GS, epigenetic signatures, and long-read sequencing diagnoses were assessed in 74 trios with at least moderate ID.

RESULTS:

The ES diagnostic yield was 42/74 (57%). GS diagnoses were made in 9/32 (28%) ES-unresolved families. Repeated ES with a contemporary pipeline on the GS-diagnosed families identified 8/9 SNVs/CNVs undetected in older ES, confirming a GS-unique diagnostic rate of 1/32 (3%). Episignatures contributed diagnostic information in 9% with GS-corroboration in 1/32 (3%) and diagnostic clues in 2/32 (6%). A genetic aetiology for ID was detected in 51/74 (69%) families. 12 candidate disease genes were identified. Contemporary ES followed by GS cost US$4,976 (95% CI: $3,704; $6,969) per diagnosis and first-line GS at a cost of $7,062 (95% CI: $6,210; $8,475) per diagnosis.

CONCLUSION:

Performing GS only in ID trios would be cost equivalent to ES if GS were available at $2,435, about a 60% reduction from current prices. This study demonstrates that first-line GS achieves higher diagnostic rate than contemporary ES but at a higher cost.

INTRODUCTION

Intellectual disability (ID) affects 3% of the population, of which 15% are moderate to severe and enriched for Mendelian aetiologies.¹ Despite significant molecular diagnostic improvements, approximately half of the individuals with ID remain undiagnosed after exome sequencing (ES).² Genome sequencing (GS) increases coding region coverage consistency, has improved detection sensitivity for structural variants (SVs) including copy number variants (CNVs) and tandem repeat expansions (TREs) and facilitates the detection of noncoding and mitochondrial variation.³ Short-read sequencing (SRS) trio GS as a first-tier stand-alone test has been proposed to maximize diagnoses in Mendelian disorders when cost is not a consideration.⁴

Diagnostic testing costs from published and commercial sources range from US$3,825 - $9,304 (trio ES)⁵ and US$4,963 - $15,000 (trio GS).⁶ Despite the large resources required for the care of people with ID, there is currently limited health economic evidence to support ES versus GS. Studies that evaluate diagnostic rates, costs, sensitivity and cost-effectiveness of these tests are integral to their implementation into clinical practice.⁵ These were therefore assessed in this prospective, tightly ascertained, moderate to severe ID cohort using ES, GS, episignatures and long-read sequencing (LRS). Contemporary ES was repeated on families diagnosed by GS where the original ES was reported as uninformative to investigate the current diagnostic potential for ID unique to GS. A cost for GS trios as a first-line test was also estimated based on the incremental diagnostic rate of GS over contemporary ES in ID from this study.

MATERIALS AND METHODS

Cohort ascertainment

Seventy-four trios (proband and unaffected parents) were recruited to this study (Figure 1) funded through Australian Genomics, the Center for Mendelian Genomics (Broad CMG), and New South Wales Statewide Genomic Service. Families were referred from Australian hospitals to the New South Wales Health Pathology Randwick Genomics (RG) and Victorian Clinical Genetics Services (VCGS) laboratories. Probands with moderate or severe ID, non-contributory microarrays and FMR1 molecular testing were included. Individuals with autism spectrum disorder or prior ES were excluded. All individuals had genetic counselling, and their parents/guardians consented for diagnostic ES and research GS. The study was approved by HREC/16/MH/251.

Figure 1. ID Flagship study.

74 trios with molecularly undiagnosed ID investigated using ES (74 trios), GS (32 trios), EpiSign (32 probands) and LRS (16 probands). The overall ES/GS diagnostic yield is 69% (51/74) with 31% (23/74) unresolved. ID, intellectual disability; ES, exome sequencing, GS, genome sequencing; LRS, long read sequencing.

ES analysis

For original ES, DNA analyzed at RG was extracted from peripheral blood samples and libraries using the Ion AmpliSeq Exome RDY kit (Thermo Fisher Scientific, Waltham, USA), sequence on the Ion PI chip kit (v3) and analyzed on a Life Technologies Proton instrument. Reads were aligned to human genome build hg19/GRCh37, SNVs and indels were identified using TorrentSuite v5.10.1 software and analyzed as reported previously.⁷ The Agilent SureSelect QXT CREv2 kit (Agilent Technologies, Santa Clara, USA) was at VCGS for library preparation, sequenced on an Illumina NextSeq500 (Illumina, San Diego, USA) and variants were analyzed using the Illumina DRAGEN Bio-IT Platform after alignment to hg19/GRCh37. The original ES analyses did not include CNV pipelines. Contemporary ES analysis was performed on selected families at RG to determine GS-unique diagnoses, using library building Exome 2.0 kits (Twist Bioscience, San Francisco, USA) with sequencing on an Illumina NovaSeq 6000 (Illumina, San Diego, USA), mapped to hg38/GRCh38 and analyzed using the Illumina DRAGEN Bio-IT Platform. SNVs, and indels were classified as previously reported,⁸ utilizing the American College of Medical Genetics and Genomics and the Association for Molecular Pathology (ACMG-AMP) guidelines with modifications and incorporating aspects of the scoring system reported by Karbassi et al.⁹ Genomic variants were filtered in the GAIA pipeline⁸ at RG. A commercial genomic analysis pipeline, Alissa Clinical Informatics Platform (Agilent Technologies, Santa Clara, USA), was used for genomic analysis at VCGS. Potential CNVs were identified using DECoN,¹⁰ XHMM,¹¹ the Illumina DRAGEN Bio-IT Platform at RG and an internal CNV detection tool CxGo¹² at VCGS.

GS analysis

GS was performed on ES-negative families at the Broad CMG. DNA libraries were prepared using the PCR-free KAPA Hyper Prep kit without amplification module (KK8505; KAPA Biosystems, Wilmington, USA) with palindromic forked adaptors (Roche, Basel, Switzerland) and run on the NovaSeq 6000 (Illumina, San Diego, CA, USA). Reads were mapped to hg38/GRCh38 using BWA aligner and analyzed using the Genome Analysis Toolkit (GATK) HaplotypeCaller package (v4.0). GS data were analyzed at RG using GAIA, aligned to GRCh38. SV, CNV and the mitochondrial genome were analyzed by a number of different platforms in the study laboratories (See Supplementary Methods). GS data were analyzed by the Broad CMG using the open-source genomic analysis platform seqr.¹³ Clinical phenotypes correlated with gene names were used in PubMed searches (https://pubmed.ncbi.nlm.nih.gov/). Diagnostic outcomes were assessed based on ACMG-AMP criteria. Candidate research genes were deposited into Matchmaker Exchange through seqr¹³ or GeneMatcher.¹⁴

Episignature analysis

Episignature analysis was conducted on ES-negative families with available DNA using EpiSign, a genome-wide DNA methylation test. DNA was extracted from peripheral blood and processed using the Illumina Infinium EPIC bead chip array (Illumina, San Diego, USA) as described previously.¹⁵ Data were analyzed using EpiSign (V3) at the Molecular Diagnostics Laboratory, London Health Sciences Centre, Western University, London, Canada.¹⁶

Long read sequencing analysis

DNA was subject to LRS using Oxford Nanopore Technologies (ONT) at the Garvan Institute Sequencing Platform. Libraries were prepared from high molecular weight DNA using native ligation-based library kit (SQK-LSK110) and sequenced on PromethION R9.4.1 flowcells at one sample per flowcell. ONT sequencing data were converted from FAST5 to BLOW5 format¹⁷ using slow5tools (v0.3.0).¹⁸ Data were base called using Guppy. Targeted genotyping was performed on 48 STR sites associated with neurological disease¹⁹ using a local haplotype aware ONT read assembly spanning each STR site, annotating STR size, motif and other summary statistics using Tandem Repeats Finder (4.09) followed by manual inspection and motif counting.¹⁹

Health economic analysis

Health economic analysis were undertaken from the health care funder perspective at the time of diagnosis to evaluate the cost-effectiveness of performing GS over ES under three different models: i) performing GS on ES-negative families ii) GS-unique diagnoses after contemporary ES iii) performing GS on all families as a first line test. Laboratory costs for ES and GS were sourced from RG and VCGS. These were fixed costs at publicly reimbursement rates in Australia and thus have no uncertainty in the unit costs. All costs are Australian prices unless otherwise stated and are converted to US dollars, based on the exchange rate of AU$ 1 = US$ 0.6577 of 1^st June 2023 (https://www.xe.com/currencytables/). Unit costs are listed in Table S1. The cost-effectiveness of GS relative to ES or contemporary ES was analyzed for the incremental cost per additional diagnosis under three diagnostic pathways with uncertainty assessed using a bootstrapping method. This generated 1000 replicated datasets, randomly drawn with replacement. Incremental costs per additional diagnosis were estimated for each dataset, and a 95% confidence interval was estimated as an uncertainty related to incremental cost from the distribution of incremental cost based on the percentile method. The price reduction of GS was estimated such that it produced the same cost per diagnosis as contemporary ES in ID. As our study was performed to cost the diagnostic rate of different technologies, the base results included the costs that were within the scope of laboratory testing including sample collection, sequencing and data analysis but not including the costs of clinical review. A sensitivity analysis was conducted assuming two clinical geneticist / genetic counsellor visits one at the beginning and one at follow up.

RESULTS

Clinical characteristics of the cohort

This cohort included 74 probands (41 male, 33 female) with a median age of 15 years (6–43 years). ID severity was determined by developmental assessments as moderate (35), severe (37) and profound (2). Non-familial facial features were present in 19/74 (26%), 12/74 (16%) had microcephaly and 10/74 (13%) had seizures (Supplementary Table S2).

ES diagnostic yield

Trio ES analysis was performed in 74 families. Pathogenic or likely pathogenic variants were identified in 42 out of 74 families (57%). These diagnoses included 32 SNVs and 10 indels (Table S3). No diagnosis was obtained in 43% (32/74).

GS diagnostic yield compared to original ES

The ES-negative cases (n=96; 32 families) had trio GS at the Broad CMG to investigate the GS-specific diagnostic rate. This analysis resulted in an additional diagnosis for 9 families (9/74; 12%), which included three SNVs, one indel and five CNVs (Figure S1; Table 1 and S4). Mitochondrial variant, promoter and TRE analyses did not detect any additional diagnoses. Twelve candidate genes were implicated in ID (Table S5). Additional evidence of pathogenicity through collaborative studies facilitated through Matchmaker Exchange via seqr¹³ or GeneMatcher¹⁴ resulted in two families (3%) with possible diagnoses. Twenty-three families (31%) remained undiagnosed after GS testing.

Table 1.

Genome sequencing, episignature and long read sequencing findings in ID Flagship cohort

Proband	GS Gene (HGNC ID)	HGVS Nomenclature		GS ACMG Criteria	GS ACMG Interpretation	Episig nature	LRS
43	NR2F1(7975)	NC_000005.10:g.93585289G>A	NC_000005.10(NM_005654.6):c.266G>A NP_005645.1:p.(Cys89Tyr)	PS2, PM1, PM2, PP3	LP	ND	NT
44	MED13L (22962)	NC_000012.12:g.115991285_116027907dup	NC_000012.12(NM_015335.5):c.480–5305_3670dup NP_056150.1:p.(Ser1224Asnfs*8)	PVS1, PS2	P	BWS	ND
45	XRN1 (30654) (research)	NC_000003.12:g.142425225G>T	NC_000003.12(NM_001282857.2):c.624C>A XP_016862130.1:p.(Asp208Glu)	NA	NA	ND	ND
46	SF3B1 (10768) (research)	NC_000002.12:g.197402101T>G	NC_000002.12(NM_012433.4):c.2107A>C NP_036565.2:p.(Thr703Pro)	NA	NA	ND	ND
47	YARS1 (12840)	NC_000001.11:g.32779421_32779423del	NC_000001.11(NM_003680.4):c.1438_1440del NP_003671.1:p.(Glu480del)	PS2, PM4	LP	ND	ND
		NC_000001.11:g.32787071G>A	NC_000001.11(NM_003680.4):c.689C>T NP_003671.1:p.(Ser230Phe)	PS2, PM2, PP3
48						ND	ND
49	UBR5 (16806) (research)	NC_000008.11:g.102288189C>A	NC_000008.11(NM_015902.6):c.4791G>T NP_056986.2:p.(Glu1597Asp)	NA	NA	ND	ND
50	DYRK1A (3091)	NC_000021.9:g.37512175A>G	NC_000021.9(NM_001347721.2):c.1909A>G NP_001334650.1:p.(Met637Val)	PS2, PP2, BS2	VUS	ND	ND
	POLR3B (30348)	NC_000012.12:g.106433856G>A	NC_000012.12(NM_018082.6):c.1765G>A NP_060552.4:p.(Gly589Arg)	PS2, PM1, PM2, PP2, PP3	LP (no second allele)
51						ND	ND
52						ND	ND
53						ND	ND
54	SCN2A (10588)	NC_000002.12:g.165312617_165332618del	NC_000002.12(NM_001371246.1): c.1034+529_2388+1050del NP_001358175.1:p.(Gln346Serfs*14)	PVS1, PS2, PM2	P	ND	ND
55	POGZ (18801)	NC_000001.11:g.151405203del	NC_000001.11(NM_015100.4):c.3837del NP_055915.2:p.(Lys1279Asnfs*31)	PVS1, PS2, PM2	P	ND	Not analyzed due to low coverage
56	CELF2 (2550) (research)	NC_000010.11:g.11165632A>G	NC_000010.11(NM_001326342.2):c.221A>G NP_001313271.1:p.(Tyr74Cys)	PS2, PM2	Research originally, now publisheda	ND	ND
57						ND	ND
58	UPF3B (20439)	NC_000023.11:g.119808858_119816662del	NC_000023.11(NM_080632.3):c.*18216_*26020del NP_542199.1:p.(?)	PVS1 PM4	LP	ND	Not analyzed due to low coverage
59						ND	NT
60						ND	NT
61	MCM6 (6949) (research)	NC_000002.12:g.135868621T>C	NC_000002.12(NM_005915.6):c.605A>G NP_005906.2:p.(Asp202Gly)	NA	Research originally, now publishedb	NT	NT
62	EHMT1 (24650)	NC_000009.12:g.137699442_137754877dup	NC_000009.12(NM_024757.5): c.22–11525_1369+586dup NP_079033.4:p.(?)	PVS1, PS2	P	Kleefstra	NT
63	KANSL3 (25473) (research)	NC_000002.12:g.96609039C>A	NC_000002.12(NM_001115016.3):c.1409G>T NP_001108488.1:p.(Gly470Val)	NA	NA	ND	ND
	KANSL3 (25473) (research)	NC_000002.12:g.96613488T>C	NC_000002.12(NM_001115016.3):c.795A>G NP_001108488.1:p.(Pro265=)	NA	NA
	FBP2 (3607) (research)	NC_000009.12:g.94593659C>T	NC_000009.12(NM_003837.4):c.68G>A NP_003828.2:p.(Arg23His)	NA	Research originally, now publishedc
64						ND	NT
65						ND	NT
66						ND	NT
67	PPP2R5C (9311) (research)	NC_000014.9:g.101882257T>C	NC_000014.9(NM_001352913.2):c.556T>C NP_001339842.1:p.(Trp186Arg)	NA	NA	ND	NT
68	ZBTB34 (31446) (research)	NC_000009.12:g.126879889_126879890del	NC_000009.12(NM_001099270.4):c.490_491del NP_001092740.2:p.(Pro164Serfs*19)	NA	NA	ND	NT
	NFE2L3 (7783) (research)	NC_000007.14:g.26152577del	NC_000007.14(NM_004289.7):c.79del NP_004280.5:p.(Arg27Alafs*2)	NA	NA
	NFE2L3 (7783) (research)	NC_000007.14:g.26152931G>T	NC_000007.14(NM_004289.7):c.433G>T NP_004280.5:p.(Gly145Cys)	NA	NA
69						UPD7p at /MRXSCJ	NT
70	SHANK2 (14295) / near CREB3L1 (18856)	NC_000011.10:g.46301244_70530165inv	NC_000011.10(NM_012309.5): c.2062–27234_*24171625inv NP_036441.2:p.(?)	PVS1, PS2	P	ND	NT
71						ND	NT
72	ARFGEF3 (21213) (research)	NC_000006.12:g.138298628_138298629del	NC_000006.12(NM_020340.5):c.3671_3672del NP_065073.3:p.(His1224Argfs*11)	NA	NA	ND	NT
73						ND	NT
74	MAP2K4 (6844) (research)	NC_000017.11:g.12129194A>T	NC_000017.11(NM_003010.4):c.947A>T NP_003001.1:p.(Gln316Leu)	NA	NA	ND	NT

GS, genome sequencing; LRS, long read sequencing; ID, intellectual disability; ACMG, American College of Medical Genetics; LP, likely pathogenic; P, pathogenic; NA, not applicable; ND, not detected; NT, not tested; dup, duplication; del, deletion; inv, inversion.

^aItai et al.³⁸

^bSmits et al.³⁹

^cGizak et al.⁴⁰

GS-unique diagnostic yield compared to repeated contemporary ES

Contemporary ES re-analysis including CNV detection was performed at RG to determine the GS-unique diagnostic rate for ID. Three cases with SNVs and five cases with one indel and four CNVs detected by GS were also detected by contemporary ES analysis (Figure S2, Table S6). Compound heterozygous YARS1 variants detected in proband 47 in GS were present in prior ES data but were discarded as the only phenotype in OMIM at the time was Charcot-Marie-Tooth syndrome without correlation to neurocognitive disease. A de novo heterozygous SNV in NR2F1 in proband 43 and a de novo heterozygous indel in POGZ in proband 55 were both present in prior ES data, but they were assessed previously as false positives. A de novo heterozygous variant in CELF2 predicted to result in a missense change was identified by GS and contemporary ES in proband 56 but was not a known Mendelian gene at the time of the original ES. All variants were absent from gnomAD and had in silico scores supportive of pathogenicity. Likewise, four of the five CNVs detected by GS and not detected in the original ES were also detected by contemporary ES using CNV callers. These included a de novo heterozygous partial MED13L 26.6 kb out of frame tandem duplication (NM_015335.5, exons 5 – 16 of 31) in proband 44, a paternal mosaic heterozygous partial deletion of SCN2A in proband 54, a maternally inherited hemizygous UPF3B deletion in proband 58 and a de novo heterozygous EHMT1 partial duplication in proband 62.

GS-unique diagnoses not detected with contemporary ES analysis

One of the five GS-detected CNVs or SVs were not detected by contemporary ES. A large 24Mb de novo heterozygous copy-neutral chromosome 11 pericentric inversion, with break points in intronic regions of SHANK2 and CREB3L1 was detected in proband 70. A subsequent karyotype confirmed the inversion. With known de novo pathogenic loss-of-function variants, the SHANK2 breakpoint is likely to invert exons, resulting in truncation with nonsense mediated decay and loss of protein function. A deletion partially spanning exon 8 and a partial intronic region of ANK3 in an isoform expressed in the brain (NM_020987.5) was detected in proband 73 with the RG GS pipeline, but not with the Broad CMG GS pipeline, contemporary ES pipeline (RG) or ES CxGo (VCGS). This variant was subsequently assessed as a false positive. This brings the total GS-unique diagnoses down from nine (12%) to one (3%) after contemporary ES analysis.

Genome-wide episignature results

Episignatures were investigated in 32 ES-negative probands where DNA was available. The outcomes were a diagnosis in one family and unresolved findings in two families (Table S7). This resulted in a minimum diagnostic yield of 3% and a maximum yield considering all abnormal episignatures of 9%. An EHMT1/Kleefstra syndrome (MIM 610253) episignature was detected in proband 62 and correlated with a GS CNV finding of a partial EHMT1 duplication (Figure 2A). This CNV was also detected through a contemporary ES pipeline with CNV analysis using CxGo. A high confidence episignature interpreted as mosaic Beckwith-Wiedemann syndrome (BWS) (MIM 130650) IC2 on chromosome 11 was detected in proband 44, where a de novo MED13L duplication on chromosome 12 was also identified on GS CNV analysis (Figure 2B). This CNV was assessed as the diagnosis. There is currently no episignature for MED13L, and it is unknown whether the mosaic IC2 episignature may be related to the MED13L event. There were no clinical features consistent with BWS present. An inconclusive episignature for KDM5C was detected in proband 69 (Figure S3A), as well as a likely mosaic uniparental disomy of the paternal chromosome 7 for MEST, SVOPL and HTR5A (hemi-methylation observed at GRB10 and PEG10 regions on chromosome 7 (Figure S3B). A prior SNP array did not detect a mosaic isodisomy in this patient and no rare variants in KDM5C were identified.

Figure 2. Concordance between GS and episignature findings.

A. A Kleefstra syndrome episignature in proband 62 was concordant to the finding of a duplication in EHMT1 using GS. B. A Beckwith-Weidemann syndrome with IC2 episignature in proband 44 and detection of a de novo heterozygous partial out of frame tandem duplication (NM_015335.5, exons 5 – 16 of 31) of MED13L using GS might be related by a yet unknown mechanism.

Long read sequencing results

STR analysis of 48 genes on LRS data from 16 probands, where DNA was available, identified no pathogenic TREs but five probands with variants of potential interest were identified (Table S7). An approximate 275-repeat intronic STARD7 non-reference motif expansion, TTCAG/AACTG, also present in the general population (gnomAD [v3.1.2.]), was identified in proband 49. This expansion was confirmed to be paternally inherited and is likely benign for an ID phenotype. A different intronic STARD7 ATTTC pentamer repeat expansion has been reported in Familial Adult Myoclonic Epilepsy.²⁰

Health economic analyses results

Cost effectiveness analysis provided information on the health and cost impacts of trio GS compared to the current standard of care test, trio ES. The average cost per family under the original ES pathway was estimated at $1,851. Performing GS on 32 original ES negative families increased the average cost per family to $3,956, with a further 9 families being diagnosed (a total of 51 families diagnosed). Had contemporary ES been performed at the beginning of the study, 50 out of 74 families would have been diagnosed by ES. Performing GS on the contemporary ES negative families, therefore, yielded one additional GS-unique diagnosis. Using GS as a first-line test resulted in an average cost of $4,867 per family, with 51 out of 74 families being diagnosed.

The incremental cost per additional genetic diagnosis using GS on the original ES-negative families was estimated at $17,305 (95% CI: $11,125; $40,153). The incremental cost per additional genetic diagnosis using GS on contemporary ES-negative families was estimated at $116,808 (95% CI: $37,314; Dominated (i.e., higher cost but no additional diagnosis)). GS as a first line test had an incremental cost of $223,184 (95% CI: $74,395; Dominated) per additional genetic diagnosis compared to contemporary ES as a first line test. Performing GS on contemporary ES-negative families was less costly compared to performing GS as a first-line test, with the contemporary ES followed by GS pathway costing $4,976 (95% CI: $3,704; $6,969) per diagnosis and first-line GS costing $7,062 (95% CI: $6,210; $8,475) per diagnosis (Table 2). Cost-effectiveness acceptability curves were calculated for a range of willingness-to-pay threshold cost for each additional genetic diagnosis using GS (Figure S4). Since there was only one additional diagnosis for GS in our cohort, the cost-effectiveness acceptability curves show that for a willingness-to-pay threshold of additional $150,000 for each additional diagnosis, there was a 62% probability that GS would be cost-effective if it was performed on contemporary ES-negative families and about 26% if it was performed as a first line test.

Table 2.

Comparisons of average cost, average cost per diagnosis and incremental cost per additional diagnosis under different diagnostic pathways

	Compared to Original ES			Compared to Contemporary ES
	Original ES	Original ES followed by GS	GS (as a first line test)	Contemporary ES	Contemporary ES followed by GS	GS (as a first line test)
Total number of diagnosis	42	51	51	50	51	51
Average cost per family	1,851	3,956	4,867	1,851	3,429	4,867
Average cost per diagnosis (95% CI)	3,261	5,740	7,062	2,739	4,976	7,062
	(2,796; 4,029)	(4,459; 7,680)	(6,210; 8,475)	(2,362; 3,301)	(3,704; 6,969)	(6,210; 8,475)
Incremental cost per additional diagnosis (95% CI)	Reference diagnostic pathway	17,305	24,798	Reference diagnostic pathway	116,808	223,184
		(11,125; 40,153)	(14,879; 55,796)		(37,314; Dominated)	(74,395; Dominated)

ES, exome sequencing; GS, genome sequencing; CI, confidence interval; Dominated, higher cost but no additional diagnosis

Based on Australian costs ($1,851 for ES and $4,867 for GS), the cost of GS trios would need to be reduced by 61.2% (95% uncertainty range: 59.4–61.9%) to be cost equivalent to contemporary ES trios on a cost per diagnosis basis if all families have GS as a first line test. If GS is to be performed only on contemporary ES-negative families (the contemporary ES followed by GS pathway), the cost of GS trios would need to be reduced by 43.7% (95% uncertainty range: 32.2 – 51.4%) for it to be cost equivalent. Based on costs more reflective of those in the US ($2,393 for ES and $6,598 for GS), the approximate cost of GS of $2,435 would be cost equivalent to contemporary ES on a cost per diagnosis basis. This would represent a reduction in the cost of GS of 63.1% (95% uncertainty range: 61.3–63.7%), to become cost-equivalent to contemporary ES.

Sensitivity analysis

Analysis including Australian specific costs of clinical review and genetic counselling, in addition to laboratory costs, shows that the estimated average cost per family would be $2,050 for the contemporary ES pathway, $3,634 for GS following contemporary ES and $5,066 for GS as a first line test. Sensitivity analysis exploring a range of GS diagnostic yields at 5%, 10%, 15% and 20% more than the contemporary ES diagnostic yield showed that the estimated incremental cost per additional genetic diagnosis using GS on contemporary ES-negative families was $38,936 (95% CI: $17,521; Dominated) if the GS diagnostic yield was 5% more than the contemporary ES diagnostic yield and $11,681 (95% CI: $7,964; $23,848) if it was 20% more than the contemporary ES diagnostic yield (Figure S5).

DISCUSSION

This study was assembled to obtain health economic evidence for the most rational genomic testing pathways for people with Mendelian forms of ID. Trio ES was performed on 74 prospective, tightly ascertained families with at least moderate ID, followed by trio GS in 32 ES-negative families, episignature analysis in 32 ES-negative individuals and LRS in 16 ES-negative individuals. A genetic aetiology for ID was detected in 51 out of 74 families (69%) using a combination of older Illumina and Life Technologies ES and current GS, with the original ES contributing 42/74 (57%) and GS-unique diagnoses contributing an additional one diagnoses (1%). Although genomic testing has become available and bioinformatic pipelines more standardised, the time required to complete cohort studies is still significant and results in diagnostic outcome biases due to technological advancement. The GS-unique diagnostic yield was therefore refined by repeated contemporary ES sequencing and analysis. Although GS identified nine events not reported in the original ES, contemporary ES was able to detect 8 of these. The diagnoses were missed due to older sequencing chemistries and unavailability of ES CNV analyses. The one GS-unique diagnosis included a large 24MB de novo copy-neutral inversion on chromosome 11 with breakpoints within SHANK2 and near CREB3L1. The presence of a de novo aetiology for ID enabled accurate low-recurrence risk genetic counselling for this family. Episignature analysis detected an episignature concordant for the EHMT1 duplication event detected on GS/contemporary ES. Episignature testing also provided further diagnostic information in two individuals, consistent with a partial Beckwith-Wiedemann syndrome episignature in one and a possible paternal UPD7 and/or a partial MRXSCJ episignature in the other. The effects of controlling for interim novel gene identifications, refinements to bioinformatic pipelines and improved sequencing quality have increased the relative utility of ES compared to GS. This study has shown that combining contemporary ES with episignature analyses has narrowed the diagnostic gap between ES and GS to maximize ID diagnoses.

LRS more accurately detects more SVs and STR expansions/contractions across the genome than short-read GS.²¹ Much of this variation is highly polymorphic and previously unannotated. Large-scale LRS studies will be required to generate appropriate reference data to assist in interpreting this variation, and functional studies will be required to validate possible pathogenic alleles. Because it is still unknown how much of the additional variation revealed by LRS is clinically relevant, it remains unclear how much of a diagnostic advantage these technologies offer over current first-tier genetic testing.

The GS-specific diagnostic rate for ID appears similar between contemporary ES and GS.²² A recent Dutch study that performed ES and GS on 150 trios with neurodevelopmental disorders, reported a similar diagnostic yield for GS (30.0%) and ES (28.7%).²³ All GS-unique diagnoses reported in that study were CNV events, therefore it may be concluded that the diagnostic benefit of performing GS in ID is to detect CNVs. The higher diagnostic rate (69%) in this study compared to the diagnostic rate of 30% in the Dutch study may be explained by the selection criteria focussing on families with moderate to severe ID in the study reported here. Most GS findings were identifiable on contemporary ES due to better sequencing quality or higher coverage (22%, 2/9), introduction of ES CNV analysis (44%, 4/9) and interim publication of novel Mendelian gene disease relationships (22%, 2/9). Ongoing improvements in bioinformatic pipelines, coding region coverage, SV/CNV detection, consanguinity assessment and integration of HPO-coded phenotypic information can facilitate the identification of missed diagnoses through data re-analysis.^4,24 A significant impact on diagnostic rates for both ES and GS is novel Mendelian gene disease relationship ascertainment over time. This has been estimated to be approximately 244 novel genes per year at the end of 2018 and includes many conditions with ID.²⁵ It would be expected that as the rate of novel gene disease relationship discovery plateaus and eventually falls, the relative utility of reanalysis will be reduced.

ES sensitivity has increased through better sequencing and the inclusion of intronic and non-coding regions where definitive pathogenic variants are described. GS has a simpler library building process, provides more uniform genome coverage and reduces the number of tests for a diagnosis. GS has some drawbacks compared to ES, including decreased sensitivity for detecting mosaic variants due to reduced depth, increased costs for data generation, analysis and storage, and a limited understanding of the functional effects of non-coding variation.²⁶ Costs for confirmatory tests, including RNA-seq for potential splicing variants should also be considered. RNA-seq from relevant tissues, can increase diagnostic yields by 7–36%.²⁷ RNA-seq costs are relatively low but need to be incorporated into a post-GS cost estimates. Although upfront ES costs can be seen as most cost-effective, the additional costs of GS even with small increases in diagnostic rates and only a proportion of eligible families seeking reproductive modulation, the costs are small compared to life-long care for those families with ID.^28,29 Thus, even a single additional diagnosis may have substantial health and cost implications.

The lifetime costs to families living with ID and society have been estimated to be at least US$172,000 per family/year.²⁸ It is reasonable to support the larger upfront GS costs as these are recovered over time, even with a small incremental diagnostic rate over contemporary ES. GS funds and those for patient care are, however, often derived from different sources. Consequently, although the general adoption of GS can be argued, the upfront costs need to be reduced for GS to be adopted as a routine first-tier diagnostic test. The potential for lower cost GS data production is emerging, with costs estimated to be reduced to US$1/Gb³⁰ or US$200 per genome³¹ for high sample volumes exclusive of data analysis and tertiary reporting. This is well below the estimate for cost-equivalence of GS to ES for achieving a genomic report in ID trios of US$2,435 calculated in this study.

The health economic assessments in this study related to ID trios. Other disease types, different sample sizes, gene panels or singleton sequencing will not have the same health economic outcomes and would require independent analysis studies. Despite this, a similar sized cohort (n=87) was reported to have comparable diagnostic yields to this study for ES which was between 20–75% depending on cohort structure, followed by GS performed on the ES-negative cases (n=12) yielded one diagnosis (8%) of a missense and an intragenic deletion of 90Kb.³² The similarity of ES and GS diagnostic rates is becoming apparent and should be taken into consideration when allocating health budget resources for genomic sequencing versus investment in lower cost sequencing technologies.

Useful adjunct methodologies with lower costs than GS, such as episignatures have become available³³ resulting in alternatives to binary decisions between ES and GS. The proportion of genes with episignatures (EpiSign v4) were estimated in this study, the Dutch study,²³ diagnosed ID cases in RG internal databases and the top 200 DECIPHER ID genes ranked by frequency of diagnoses (Figure S6). Of the 51 diagnoses in this study, nine were genes with episignatures, consistent with 12% (9/74) of cases being diagnosable using a comprehensive ID gene panel at lower cost than ES or GS. Similarly, had episignature testing been performed in the Dutch cohort,²³ it could have provided information about the causative gene in 13 out of 150 (7%) cases. The utility of episignature testing is potentially even higher when applied before GS as 3/9 (33%) of the GS-identified genes in this study and 13/45 (29%) in the Dutch study²³ may have been diagnosable by a panel containing ID genes with known episignatures. Forty-nine of the top 200 (24%) ID genes in DECIPHER database have an episignature, whereas, out of 930 cases with diagnoses in 469 ID genes in the RG database, 58 genes (12%) have been assessed as having an episignature. These findings demonstrate that the concurrent use of ES and episignatures could assist in prioritizing the choice of the most relevant post-ES methodology to maximize diagnostic rates and minimize costs. Episignatures may have a higher clinical utility with potential cost reductions if performed with ES to prioritise GS. Episignatures have also been shown to have utility in providing a genome-wide functional test to identify likely genetic etiologies but also in the re-classification of VUS.³⁴ No VUS were identified in this study that had episignature pathogenicity assessment alterations, but this represents additional potential utility for episignatures.³⁵

After extensive testing, 23/74 (31%) families in this cohort remained undiagnosed. RNA-seq has been reported to detect an additional 7–35% of diagnoses and so can be considered to detect abnormally spliced transcripts due to deep intronic or synonymous/missense variants with splicing effects.³⁶ A splice variant was detected in a similar cohort²³ (0.7%) so it is likely that at least one case in our cohort might be diagnosed through RNA-seq. Although the most comprehensive combination of diagnostic techniques could be GS, episignatures and RNA-seq, these are not yet available routinely in most diagnostic settings. Variation relevant to ID and commonly missed by SRS can also be assessed using LRS.²¹ Its utility lies in confirmatory testing and detecting variants inaccessible to SRS, such as large CNVs/SVs and repeat expansions, particularly variation in repeat-rich and high GC-content gene regions, but analysis and interpretation of these data types are still under development.³⁷

The narrowing of the diagnostic gap using contemporary ES analysis and episignature informs the most cost-effective combination of diagnostic techniques for ID. Most variants are predicted to be identified using ES trios analyzed with a contemporary ES pipeline inclusive of CNV analysis. Using complementary genetic technologies after contemporary ES and ES re-analysis such as episignatures and then GS will maximize the diagnostic rates in ID. The GS trio price setting cost-equivalent to ES for ID trio diagnostic rates was assessed as US$2,435, approximately a 60% reduction from current prices. Ongoing technological developments resulting in lower cost GS will increase its relative value compared to ES, to the point where it is likely to become the preferred first-line genomic test.

URLS

Alamut Visual Plus: https://www.sophiagenetics.com/platform/alamut-visual-plus/. CADD Score: http://cadd.gs.washington.edu/. ClinVar: https://www.ncbi.nlm.nih.gov/clinvar/. Eukaryotic Promoter Database: https://epd.epfl.ch/EPDnew_database.php. GeneMatcher https://genematcher.org. gnomAD: https://gnomad.broadinstitute.org/. OMIM: https://www.omim.org/. PubMed: https://pubmed.ncbi.nlm.nih.gov/. RefSeq: https://www.ncbi.nlm.nih.gov/refseq/. seqr: https://seqr.broadinstitute.org/. UCSC Genome Browser: http://genome.ucsc.edu/. VarCards: http://varcards.biols.ac.cn/.

Data Availability

This study generated GS data that is deposited in the NHGRI GREGoR Consortium: Genomics Research to Elucidate the Genetics of Rare Disease - dbGaP Study Accession: phs003047.v1.p1. Proband numbers are mapped to accession numbers (Table S8).