Whole genome sequencing identifies elusive variants in genetically unsolved Italian inherited retinal disease patients

Summary

Inherited retinal diseases (IRDs) are a group of rare monogenic diseases with high genetic heterogeneity (pathogenic variants identified in over 280 causative genes). The genetic diagnostic rate for IRDs is around 60%, mainly thanks to the routine application of next-generation sequencing (NGS) approaches such as extensive gene panels or whole exome analyses. Whole-genome sequencing (WGS) has been reported to improve this diagnostic rate by revealing elusive variants, such as structural variants (SVs) and deep intronic variants (DIVs).

We performed WGS on 33 unsolved cases with suspected autosomal recessive IRD, aiming to identify causative genetic variants in non-coding regions or to detect SVs that were unexplored in the initial screening. Most of the selected cases (30 of 33, 90.9%) carried monoallelic pathogenic variants in genes associated with their clinical presentation, hence we first analyzed the non-coding regions of these candidate genes. Whenever additional pathogenic variants were not identified with this approach, we extended the search for SVs and DIVs to all IRD-associated genes.

Overall, we identified the missing causative variants in 11 patients (11 of 33, 33.3%). These included three DIVs in ABCA4, CEP290 and RPGRIP1; one non-canonical splice site (NCSS) variant in PROM1 and three SVs (large deletions) in EYS, PCDH15 and USH2A. For the previously unreported DIV in CEP290 and for the NCCS variant in PROM1, we confirmed the effect on splicing by reverse transcription (RT)-PCR on patient-derived RNA.

This study demonstrates the power and clinical utility of WGS as an all-in-one test to identify disease-causing variants missed by standard NGS diagnostic methodologies.

In this study we confirmed the diagnostic power of whole genome sequencing as an all-in-one test to investigate the missing genetic cause in unexplained inherited retinal disease patients, since it allows the identification of all types of variants, from single nucleotide variants (both in coding and non-coding regions) to structural variants.

Introduction

Inherited retinal diseases (IRDs) are a group of rare, monogenic diseases that cause vision loss. They collectively affect around 2 million people worldwide, with a combined prevalence of 1:2,000 individuals.¹ The different subtypes present diverse yet partially overlapping clinical features, and their genetic etiology is highly heterogeneous with over 280 genes associated with the pathogenesis of different forms (RetNet, http://sph.uth.edu/retnet/; February 2024). In most cases, IRDs follow monogenic inheritance patterns, such as autosomal dominant, autosomal recessive, X-linked, and mitochondrial, nonetheless digenic biallelic and triallelic inheritance patterns, although rare, have also been reported.² IRD-associated genes differ greatly in their inheritance pattern, genotype-phenotype correlations, and pathogenic mechanisms (i.e., gain- or loss-of-function). Incomplete penetrance, hypomorphic alleles, and modifier genes further add to the observed variability.^3,4 This heterogeneity of IRDs complicates the genetic analysis that remains essential to provide differential diagnosis, accurate genetic counseling, and access to specific therapies or clinical trials.

The current genetic diagnostic rate for IRDs is around 50%–70% and almost a third of affected individuals remain genetically unsolved after whole-exome sequencing (WES).^{2,5,6,7,8,9,10} The missing heritability of IRDs can be explained by the limitations of the current next-generation sequencing (NGS) diagnostic approach to detect structural variants (SVs), variants in repetitive GC-rich sequences, or in non-coding regions, or pathogenic variants could be localized in still uncharacterized IRD genes. Reports suggest that the use of whole-genome sequencing (WGS) can improve the diagnostic rate by revealing such elusive variants.^11,12,13,14

Deep intronic variants (DIVs) are localized within introns, conventionally more than 100 base pairs (BP) away from the exon-intron junction,¹⁵ and are generally deleterious by affecting splicing. They can alter transcription regulatory motifs and non-coding RNA genes; however, their most common consequence is pseudoexon (PE) inclusion due to the activation of cryptic splice sites or changes in splicing regulatory elements.¹⁵ The presence of an additional PE in the mRNA can disrupt the reading frame introducing a premature stop codon that leads to mRNA degradation through nonsense-mediated decay.¹⁵ PE inclusion can also lead to the insertion of additional amino acids in the final product, disrupting protein stability.¹⁶ Pathogenic DIVs have been reported for many monogenic diseases and PE inclusion is now considered a frequent cause of disease.^17,18 Recent studies demonstrated that the role of DIVs in the genetic landscape of IRDs is significant and probably underestimated.^11,13,14,19

SVs are conventionally defined as genomic alterations involving DNA regions of at least 1 kb. They comprise balanced alterations, such as inversions and translocations, and genomic imbalances, such as duplications and deletions,²⁰ and have been increasingly recognized as important causes of inherited disease. The number of likely causal SVs has grown over the last years, thanks to the development of NGS-based algorithms that can reliably identify SVs especially in genome sequencing data. The contribution of SVs to the IRD mutational spectrum is estimated between 5% and 15%, but recent evidence suggests that this is likely an underestimate.^21,22,23

In this study, we performed short-read WGS on 33 unsolved IRD subjects to identify causative genetic variants in non-coding regions or to detect SVs that may have eluded the first-tier screening. The cases, selected from a large Italian cohort of 2,790 genetically tested individuals,⁷ had a diagnosis of a suspected autosomal recessive retinal disease (RD) form.

Subjects and methods

Patient selection

Patients were recruited at the Center for Inherited Retinal Dystrophies of the Eye Clinic, University of Campania ‘Luigi Vanvitelli.’⁷ We selected 33 individuals with a suspected autosomal recessive RD form, based on their family history and their clinical presentation, that remained genetically unsolved after the first-tier screening, mostly by NGS (Table S2). In particular, 24 cases were previously screened by ClearSeq Inherited Disease Panel or SureSelect Custom Constitutional Panel (comprising 2,742 and 5,228 disease-causing genes, respectively; Agilent Technologies, Inc.), seven were analyzed using different retinopathy panels comprising between 40 and 137 IRD genes (Retplex³⁶), six were examined by a smMIPs-based analysis (either including 105 genes from the Macular Degeneration panel or only the ABCA4 and PRPH2 genes^37,38,39), and one was screened by APEX-based genotyping microarray for known variants implicated in Leber congenital amaurosis (LCA).⁴⁰ Few cases were screened with more than one platform (see Table S2 for details). The majority of selected cases (30 of 33, 90.9%) harbored a monoallelic pathogenic variant in a gene associated with the clinical presentation. Written informed consent for genetic testing and data sharing was retrieved from the patients (or their parents/legal guardians for minors). All procedures adhered to the Declaration of Helsinki and were approved by the Ethics Board of the University of Campania ‘Luigi Vanvitelli.’

WGS and variant prioritization

Genomic DNA was isolated from peripheral blood using the FlexiGene DNA Kit (Qiagen, Hilden, Germany) according to the manufacturer’s instructions and was analyzed by genome sequencing as previously described.^14,21 Sequencing was performed by BGI Tech Solutions (Warszawa, Poland) using the DNBseq Sequencing Technology, with a minimal median coverage per genome of 30x. Read mapping to the Human Reference Genome build GRCh38/hg38 was performed using Burrows-Wheeler Aligner V.0.7814. Single-nucleotide variant (SNVs) calling was performed using the Genome Analysis Toolkit HaplotypeCaller (Broad Institute). SVs were detected using the Manta structural variant caller.²⁴ Copy-number variants (CNVs) were called with the Canvas Copy-Number Variant Caller.²⁵ SNVs were filtered for a Minor Allele Frequency (MAF) <1% in gnomAD (V2.1.121, http://gnomad.broadinstitute.org/) and in an in-house exome SNV database (over 15,000 alleles). Retained SNVs in IRD-associated genes were prioritized based on their predicted functional effect (i.e., nonsense, frameshift, canonical and non-canonical splice site variants, in-frame deletions/insertions, missense variants, and synonymous variants). Missense variants were prioritized when both in silico tools used to assess their impact predicted a deleterious effect, i.e., CADD-PHRED²⁶ (≥15, range = 0–48) and REVEL²⁷ (≥0.3, range = 0–1). Potential effects on splicing were assessed using SpliceAI²⁸ with default settings, retaining all SNVs with at least one delta score ≥0.2. Potential effects on splicing were also manually inspected using SpliceAI Visual,²⁹ as well as with the Alamut Visual software version 2.1 (Interactive Biosoftware, Rouen, France; http://www.interactive-biosoftware.com) to examine the possible disruption of a branchpoint.³⁰

SVs were prioritized based on their potential to disrupt the regulatory regions in 5′ and 3′ untranslated regions (UTRs) or the reading frame. SVs with an MAF >1% in the 1000 Genomes³¹ and DECIPHER³² were considered as benign. The breakpoint regions of potential pathogenic SVs were confirmed by visualization using the Integrative Genomics Viewer (IGV) software V2.427 (Broad Institute).^33,34

For cases carrying a monoallelic candidate variant in a gene consistent with their clinical presentation, we first investigated for SVs and SNVs in non-coding regions of the gene of interest to identify a presumed missing second pathogenic variant. Whenever additional pathogenic variants were not identified with this approach, we extended the search for pathogenic variations to all IRD-associated genes. For cases in which no primary variant was identified after the initial testing, we investigated SVs and SNVs in both coding and non-coding regions of all IRD-associated genes.

Variant pathogenicity was assessed according to the guidelines of the American College of Medical Genetics and Genomics (ACMG),³⁵ either by manual implementation of the criteria or using the Franklin automated variant classification (http://franklin.genoox.com/).

Variant validation

Sanger sequencing

Potentially pathogenic variants were validated by Sanger sequencing in the proband and in parental samples, whenever available. Sanger sequencing with specific primers was performed using a BigDye v3.1 sequencing kit (Thermo Fisher Scientific, Waltham, MA, USA) on a 3500xl Genetic Analyzer (Thermo Fisher Scientific) according to the manufacturer’s instructions.

RT-PCR analysis

To define the effect of DIVs predicted to interfere with normal splicing (at least one SpliceAI delta-score $\ge$ 0.2), we performed reverse transcription-PCR (RT-PCR) analysis of patient-derived mRNA. Total RNA was extracted from whole blood collected in Tempus Blood RNA Tubes (Thermo Scientific) following the manufacturer’s instructions. Complementary DNA synthesis was performed using the SuperScript VILO cDNA Synthesis Kit (Thermo Fisher Scientific) according to the manufacturer’s instructions. The standard PCR procedure was used to amplify the cDNA. The generated RT-PCR products were analyzed by agarose gel electrophoresis and Sanger sequencing.

Breakpoint analysis and MLPA

SVs were visualized using the Integrative Genomics Viewer (IGV) software V2.427 (Broad Institute). Breakpoints were defined by PCR amplification of the normal fragment (wild-type allele) and the breakpoint-spanning fragment (mutant allele), followed by agarose gel electrophoresis and Sanger sequencing. The oligonucleotide primers used are listed in Table S1. Multiplex ligation-dependent probe amplification (MLPA) was performed as an independent approach to validate the identified SVs, using different kits (SALSA MLPA KIT P328 EYS, SALSA MLPA KIT P361 USH2A mix1, and SALSA MLPA KIT P362 USH2A mix2) according to the manufacturer’s protocol.

Results

Cohort characterization

From a large Italian cohort of genetically tested IRD patients,⁷ we selected 33 unsolved subjects with suspected autosomal recessive inherited RD. The selected cases presented different IRD subtypes (Figure 1): retinitis pigmentosa (RP) was the most frequent phenotype (13 of 33, 39.4%), followed by Stargardt disease (STGD, 8 of 33, 24.2%), Leber congenital amaurosis (LCA, 6 of 33, 18.2%), and Usher syndrome (2 of 33, 6.1%). All patients were previously screened by various diagnostic platforms (see subjects and methods and Table S2).

Figure 1.

Clinical composition of the analyzed cohort

Identification of candidate variants

WGS data analysis allowed us to identify the possible genetic cause of the disease in 11 of 33 patients (33.3%; Table 1). We detected three different DIVs in ABCA4, CEP290, and RPGRIP1 in four patients, one non-canonical splice site (NCSS) variant in PROM1 in two affected siblings, and three SVs (large deletions) in EYS, PCDH15, and USH2A. Finally, we identified one missense variant in ABCA4 and one nonsense variant in RP1.

Table 1.

Overview of the causative genotypes identified in likely solved patients

Patient	Clinical diagnosis	Gene	RefSeq	Allele 1	Allele 1	Allele 2	Allele 2
				(cDNA change)	(Protein change)	(cDNA change)	(Protein change)
PT-1	STGD1	ABCA4	NM_000350.3	c.926del	p.(Pro309GlnfsTer7)	c.5196+1137G>A∗	p.[=,Met1733GlufsTer78]∗
PT-2	STGD1	ABCA4	NM_000350.3	c.2345G>A	p.(Trp782Ter)	c.5603A>T∗	p.(Asn1868Ile)∗
PT-9	LCA	CEP290	NM_025114.4	c.6869del	p.(Asn2290IlefsTer11)	c.6136-436A>G∗	p.Ile2046AspfsTer29∗
PT-10	LCA	CEP290	NM_025114.4	c.6604del	p.(Ile2202LeufsTer24)	c.6136-436A>G∗	p.Ile2046AspfsTer29∗
PT-13	RP	EYS	NM_001142800.2	c.6528C>A	p.(Tyr2176Ter)	c.2137+20590_3444-29847del∗	p.(Val713AspfsTer14)∗
PT-21	USH I	PCDH15	NM_001384140.1	c.2151del	p.(Phe717LeufsTer23)	c.2221-9697_3806+3450del∗	p.(Ala741AspfsTer36)∗
PT-5	STGD1	PROM1∗	NM_006017.3∗	c.221-20T>G∗	p.Asp74AlafsTer1∗	c.221-20T>G∗	p.Asp74AlafsTer1∗
PT-6	STGD1	PROM1∗	NM_006017.3∗	c.221-20T>G∗	p.Asp74AlafsTer1∗	c.221-20T>G∗	p.Asp74AlafsTer1∗
PT-11	RP	RP1∗	NM_006269.2∗	c.1186C>T∗	p.(Arg396Ter)∗	c.1186C>T∗	p.(Arg396Ter)∗
PT-27	LCA	RPGRIP1	NM_020366.4	c.832C>T	p.(Arg278Ter)	c.1468-128T>G∗	p.(Glu489_Pro490insTer3)∗
PT-29	USH II	USH2A	NM_206933.4	c.13275del	p.(Asn4426MetfsTer35)	c.4627+3035_4758+2305del∗	p.(Ile1563ValfsTer8)∗

∗Variants identified by WGS. The pathogenic variants identified in this study have been submitted to the Leiden Open (source) Variation Database (LOVD) (http://www.lovd.nl). Abbreviations: CD, cone dystrophy; LCA, Leber congenital amaurosis; RP, retinitis pigmentosa; STGD1, Stargardt disease; USH II, Usher syndrome type II.

Non-coding variants

The three different DIVs in ABCA4, CEP290, and RPGRIP1 and the NCSS variant in PROM1 were predicted by SpliceAI to influence splicing (at least one delta score ≥0.2). Whenever patient-derived RNA was available, we performed RT-PCR to confirm the effect of the variants directly on the transcript.

ABCA4

PT-1 had a clinical diagnosis of STGD and harbored a monoallelic rare variant in ABCA4 (NM_000350.3:c.926del, p.(Pro309GlnfsTer7)), classified as severe according to the genotype-phenotype correlation model for biallelic pathogenic variants in ABCA4. Using WGS, we detected a frequent DIV in ABCA4 (c.5196+1137G>A) that leads to a 73-bp PE inclusion between exon 36 and exon 37,⁴¹ and the subsequent insertion of a premature stop codon downstream of exon 36. This variant was previously classified as moderately severe,⁴² that in combination with a severe variant can result in cone rod dystrophy (CRD). Additional ophthalmological assessments and segregation analysis (no parental DNA available) should be performed to corroborate the genotype-phenotype correlation.

CEP290

The patients PT-9 and PT-10 had a clinical diagnosis of LCA and carried two distinct monoallelic frameshift variants in CEP290 (NM_025114.4:c.6869del, p.(Asn2290IlefsTer11) and c.6604del, p.(Ile2202LeufsTer24), respectively, Figure 2A), already described as pathogenic in ClinVar (VCV000654872.8, VCV000197597.15, respectively). In these unrelated probands, we identified by WGS the same rare DIV c.6136-436A>G in intron 44 of CEP290. This variant was predicted to activate two cryptic splicing sites in intron 44, presumably causing the inclusion of a 97-bp PE between exon 44 and 45 (Figure 2B). RT-PCR on blood-derived cDNA from PT-10 and a control sample confirmed the presence of an additional fragment in the proband. Sanger sequencing of the RT-PCR products revealed the predicted insertion of 97 bp between exon 44 and 45, which introduces a premature stop codon downstream of exon 44 (p.lle2046AspfsTer29, Figure 2C). The aberrant splicing product was not detected in control samples. Segregation analysis in parental samples confirmed that the variants were present in trans in both probands. Since biallelic variants in CEP290 are associated with Leber congenital amaurosis-10 (LCA10, MIM 611755), we considered these patients genetically explained.

Figure 2.

Pedigrees carrying biallelic CEP290 variants and effect on splicing of the deep-intronic variant c.6136-436A>G

(A) Pedigrees of PT-9 and PT-10 showing the presumed autosomal recessive inheritance pattern of the RD and the segregation of the variants.

(B) Graphical representation of the predicted effect of the DIV (c.6136-436A>G, p.Ile2046AspfsTer29) in CEP290, which interferes with the normal splicing (black dashed line) inducing the insertion of a 97-bp pseudo-exon (PE, red-framed square) between exons 44 and 45 (black squares) of CEP290.

(C) Sanger sequencing of the RT-PCR products obtained from mRNA of patient (PT-10) and control (CTL) samples. The chromatogram shows the sequences of both the wild-type (exons 44–45) and mutant alleles (exons 44-PE-45) in PT-10, while only the wild-type sequence is detected in the CTL sample.

PROM1

PT-5 and PT-6 were two affected siblings with a clinical diagnosis of STGD (Figure 3A). They were previously screened by a smMIPs-based approach for variants across the entire ABCA4 gene and in the exons of PRPH2^38,39, and for a set of 105 genes associated with macular degeneration.³⁷ WGS analysis revealed the presence of a homozygous rare variant (c.221-20T>G) in intron 2 of PROM1 (NM_006017.3), predicted by SpliceAI to affect normal splicing by disrupting the canonical acceptor splice site of exon 3 (SpliceAI delta-score for Acceptor Loss (AL): 0.2, Figure 3B). Considering the position of the variant (20 bp from exon-intron junction), we initially suspected that it could disrupt the branchpoint site; however, inspection with the Alamut Visual software did not reveal any significant alteration at this position (wild-type score: 40.8; variant score: 33.8). We therefore hypothesized that this variant could still affect splicing by disrupting the acceptor site consensus sequence. To confirm the prediction of exon 3 skipping, we performed RT-PCR analysis on patient-derived mRNA extracted from whole blood. PROM1 encodes different isoforms and is expressed from at least five alternative promoters in a tissue-dependent manner.⁴³ The NM_006017.3 isoform (also known as variant 1) represents the longest isoform but is expressed at very low levels in blood. Hence, we were only able to amplify the NM_001145847.1 isoform (also known as variant 2) which, compared to the major isoform, lacks an alternative in-frame exon in intron 4. Since exon 3 is present in all PROM1 isoforms, we hypothesized that the c.221-20T>G variant can also impact the longest isoform. RT-PCR analysis on cDNA derived from whole blood RNA of PT-6 and a control subject, confirmed the presence of a unique aberrant amplicon in the proband (Figure 3B). Sequencing of the RT-PCR products showed complete skipping of exon 3 in the patient, while no splicing aberration was detectable in the control (p.Asp74AlafsTer1, Figure 3C). At the post-genotyping clinical re-evaluation of the patients, their diagnosis was revised to an atypical RP with early-onset maculopathy, consistent with autosomal recessive forms of PROM1-associated disease (MIM 612095).

Figure 3.

Effect on splicing of the non-canonical splice site variant in PROM1

(A) Pedigree of PT-5 and PT-6 showing the suspected autosomal recessive inheritance pattern of the RD.

(B) Graphical representation of the predicted effect of the NCSS variant in PROM1 and agarose gel electrophoresis of the RT-PCR products obtained from mRNA of patient (PT-6) and control (CTL) samples. The c.221-20T>G, p.Asp74AlafsTer1 variant is predicted to induce skipping of exon 3 (red dashed line). RT-PCR on PT-6 yielded only one fragment with lower molecular weight (fragment 1, 579 bp) than the one obtained in the control sample (fragment 2, 635 bp), consistent with the skipping of exon 3 (56 bp).

(C) Sanger sequencing of the RT-PCR products obtained from mRNA of PT-6 and control samples (CTL). PT-6 was homozygous for the c.221-20T>G variant, hence the chromatogram of the fragment 1 only shows the sequence of the mutant allele confirming skipping of exon 3.

RPGRIP1

PT-27 had a clinical diagnosis of LCA and carried a monoallelic nonsense variant in RPGRIP1 (NM_020366.4:c.832C>T, p.(Arg278Ter)), reported in ClinVar as pathogenic (VCV000156387.5). By WGS, we identified an additional DIV in intron 11 of RPGRIP1 (c.1468-128T>G), which had already been reported to cause the inclusion of a 118-bp PE between exon 11 and exon 12, leading to a premature termination codon downstream of exon 11 in a fraction of the RPGRIP1 mRNA.⁴⁴ Since biallelic variants in RPGRIP1 cause Leber congenital amaurosis-6 (LCA6, MIM 613826), we considered this patient genetically solved. Parental samples were not available to verify the variant phase.

Structural variants

From the analysis of WGS data, we identified three large deletions in EYS, PCDH15, and USH2A in patients carrying monoallelic variants in the respective gene.

EYS

PT-13 was diagnosed with RP and carried a heterozygous nonsense variant in EYS (NM_001142800.2: c.6528C>A, p.(Tyr2176Ter)), described as pathogenic in ClinVar (VCV000208578.27). A focused analysis of the EYS locus in the WGS data revealed a 380.9 kb deletion (chr6:64656094-65037024del, GRCh38/hg38) encompassing exons 14 to 21 of EYS (c.2137+20590_3444-29849del, Figure 4A). Breakpoints were defined by PCR amplification of the corresponding region in patient and control samples (using primers A1, A2 and A3; Table S1, and Figure 4B) followed by Sanger sequencing (Figure 4C). A similar deletion (from exon 14 to exon 22), detected by MLPA, had already been described as likely pathogenic.⁴⁵ We therefore concluded that the patient’s phenotype was likely explained by these assumed biallelic variants in EYS (MIM 602772).

Figure 4.

Overview of the deletions detected in EYS, PCDH15, and USH2A

Using the Manta Structural Variant caller, we identified three large deletions in EYS, PCDH15, and USH2A in three patients (PT-13, PT-21, PT-29) who carried monoallelic (likely) pathogenic variants in the respective gene.

(A) Visualization of WGS data in the Integrative Genomics Viewer (IGV) revealed split reads (red colored reads) corresponding to the breakpoints of the deletions identified by Manta in all patients. In PT-13 a 380.9-kb deletion was detected, with breakpoints located in introns 22 and 13 of EYS, causing the deletion of exons 14 to 22. The deletion in PT-21 was spanning 179.2 kb, from intron 28 to intron 18 of PCDH15. In PT-29, a deletion of 77.4 kb encompassing exon 22 of USH2A was detected.

(B) Agarose gel electrophoresis and schematic representation of the PCR products obtained using two different primer sets (Table S1), one amplifying the wild-type alleles (fragments 1, 3, and 5; red numbers) and the other spanning the breakpoint, thus amplifying the mutant alleles (fragments 2, 4, and 6). The fragments spanning the breakpoint were present in all patients (PT-13, PT-21, and PT-29) and in one carrier parent (father of PT-21) and were absent in control (CTL) samples.

(C) The chromatograms obtained by Sanger sequencing of the wild-type (upper) and mutant (lower) alleles showed the breakpoints (vertical line) only in the fragments corresponding to the mutant alleles. The genomic coordinates of either breakpoint are reported below the chromatogram of each mutant allele. All genomic coordinates are referred to the Human Reference Genome build GRCh38/hg38.

PCDH15

PT-21 was a case diagnosed with Usher syndrome type I for the concurrent presence of RP and profound, prelingual hearing loss. Previous testing revealed a rare monoallelic variant in PCDH15 (NM_001384140.1:c.2151del, p.(Phe717LeufsTer23)), classified as likely pathogenic according to the ACMG criteria and inherited from the unaffected mother. Since the clinical presentation of the proband was consistent with a diagnosis of Usher syndrome type IF (USH1F) caused by biallelic variants in PCDH15 (MIM 602083), we analyzed the WGS data focusing on the PCDH15 locus. We identified a 179.2 kb deletion in PCDH15 (chr10:53853726-54032893del, GRCh38/hg38) spanning exons 19 to 28 (c.2221-9696_3806+3449del, Figure 4A). PCR amplification of the encompassing region using primers B1, B2 and B3 (Table S1) revealed an additional amplicon in the proband and his unaffected father, which was absent in the mother (Figure 4B). By sequencing this aberrant product, we defined the breakpoint positions in intron 18 and intron 28 (Figure 4C). The deletion from exon 19 to 28 of PCDH15 is predicted to result in a frameshift, hence the SV was classified as pathogenic according to the ACMG criteria.⁴⁶

USH2A

PT-29 had a clinical diagnosis of Usher syndrome type II and carried a monoallelic frameshift variant in USH2A (NM_206933.4:c.13275del, p.(Asn4426MetfsTer35)), classified as likely pathogenic according to the ACMG criteria. WGS revealed a 77.4-kb deletion (chr1:216094779-216172217del, GRCh38/hg38) spanning exon 22 of USH2A (c.4627+3035_4758+2304del, Figure 4A). PCR amplification of the corresponding region (using primers C1, C2, and C3; Table S1) in the proband and a control sample showed an additional aberrant amplicon only in the affected individual (Figure 4B). Sanger sequencing of the PCR product confirmed the breakpoint positions within intron 21 and intron 22 (Figure 4C). The deletion of exon 22 results in a frameshift, previously reported as pathogenic,⁴⁷ therefore we considered the patient genetically solved since his clinical phenotype was consistent with Usher syndrome type IIA caused by biallelic variants in USH2A (MIM 276901).

Coding variants

The analysis of WGS data also revealed one missense and one nonsense variant that were missed in the first-tier screening.

ABCA4

PT-2 had a clinical diagnosis of STGD and harbored a heterozygous severe variant in ABCA4 (NM_000350.3:c.2345G>A, p.(Trp782Ter)). WGS analysis did not reveal additional pathogenic variants in ABCA4 or in other IRD genes. A less stringent re-evaluation of all ABCA4 variants, regardless of their population frequency, revealed a monoallelic variant c.5603A>T, p.(Asn1868Ile) that was not retained during filtering due to its high MAF (0.04219 in gnomAD v2.1.1). Indeed, this allele accounts for at least 10% of all known ABCA4 disease-causing alleles and has been detected in over 50% of previously considered monoallelic ABCA4 cases.⁴⁸ This variant is an extremely hypomorphic allele, penetrant only when in trans with a severe variant.^3,48 Patients with this variant exhibit distinct clinical characteristics, including a late disease onset and foveal sparing. Considering these findings, post-genotyping re-evaluation of the patient’s clinical presentation confirmed it was consistent with the above-mentioned phenotype. Segregation analysis to verify variant phase was not feasible due to unavailability of family members’ samples.

RP1

PT-11 had a clinical diagnosis of early-onset RP and remained unsolved after an APEX-based genotyping microarray for (at the time) known variants implicated in LCA. In fact, this was the only case in our cohort that had not been previously analyzed by NGS. The microarray analysis had revealed a monoallelic pathogenic variant in CEP290 (NM_025114.4: c.180+1G>A; VCV000461777.26). WGS analysis did not identify additional pathogenic variants in CEP290. Instead, we detected a homozygous nonsense variant in RP1 (NM_006269.2:c.1186C>T, p.(Arg396Ter)), classified as pathogenic in ClinVar (VCV000143134.14). This variant was reported for the first time in 2013⁴⁹ and was therefore absent from the microarray as it was not known at the time of the analysis. The patient’s clinical presentation was consistent with retinitis pigmentosa-1 (RP1) caused by biallelic variants in RP1 (MIM 180100).

Discussion

The high genetic heterogeneity of IRDs complicates the identification of the causative variants. The current genetic diagnostic rate for IRDs is 50%–70%, and a significant number of affected individuals remains genetically unsolved even after comprehensive testing by WES.^2,5,7,8,9,10 Recent studies showed that WGS improves the diagnostic rate by revealing SVs and variants in non-coding regions.^11,13,14 In this study, WGS on unsolved IRD subjects with a suspected autosomal recessive form identified the causative genotype in 33.3% of cases (11 of 33). We identified three different DIVs (ABCA4, CEP290, RPGRIP1), one NCSS variant (PROM1), three SVs (large deletions in EYS, PCDH15, USH2A), and two variants in coding regions (ABCA4, RP1).

Pathogenic DIVs have been reported for many monogenic diseases^17,18 and their role in IRDs has been widely assessed in the last few years,^{11,12,13,14,19} paving the way to the development of innovative therapeutic approaches targeting such variants.^14,50 In line with this, we found likely pathogenic DIVs in four of the 33 patients analyzed (12.12%), accounting for 36.4% of the cases solved by WGS. Two of the four identified DIVs had already been described in literature (RPGRIP1⁴⁴ and ABCA4⁴¹) and one was not reported previously (CEP290). All DIVs were predicted to affect normal splicing leading to the inclusion of PE encoding a premature termination codon that disrupts the mRNA reading frame. For the previously unreported DIVs in CEP290, we confirmed PE inclusion directly on patient-derived mRNA. Interestingly, the two unrelated patients (PT-1 and PT-2) harboring the same DIV in CEP290 originate from the same Italian region. It is therefore tempting to speculate that the c.6136-436A>G DIV may represent a local founder variant.

In two siblings with STGD (PT-5 and PT-6), we identified a homozygous intronic variant (c.221-20T>G) in PROM1 predicted to disrupt a canonical acceptor splice site. This variant had already been detected by targeted gene panel analysis but was filtered out due to the stringent filtering parameters used (both in terms of distance of putative splice site variants from the intron-exon junction as well as in terms of the SpliceAI delta score threshold for retaining potential splice-altering variants, i.e., SpliceAI delta score <0.2). A retrospective re-analysis of WES data from other unsolved cases in the Italian cohort, revealed its presence at homozygous state in two additional unrelated cases, suggesting that this intronic variant could be more frequent in the Italian population than anticipated.

We identified likely pathogenic SVs in 3 of 33 (9.1%) patients, which is in line with the estimated contribution of SVs to the genetic diagnostic rate of IRDs (between 5% and 15%^11,23) and in concordance with previous studies that identified EYS, PCDH15, and USH2A as the genes with highest susceptibility to CNVs.²² Recent reports showed that the power of SV detection from WGS short-read data in IRD patients is still highly underestimated, mainly because pathogenic SVs can escape detection by short-read sequencing or may be misinterpreted and overlooked during the complex process of SV analysis.²¹ Using MLPA, we detected the same deletion of exon 22 of USH2A observed in PT-29, in another unsolved patient. We confirmed that the breakpoints were the same as for PT-29, hence prompting to the same molecular pathogenic mechanism.

Finally, we also identified variants in coding regions (one missense variant in ABCA4, one nonsense variant in RP1). These variants were either not retained during filtering of the initial analysis due to a misclassification of their effect (as in the case of the very frequent hypomorphic allele c.5603A>T, p.(Asn1868Ile) in ABCA4) or were not included in the initial screen (as in the case of the c.1186C>T, p.(Arg396Ter) variant in RP1, that was not present in the APEX-based genotyping microarray used). Therefore, the potential limitations of the technique used for previous genetic testing and variant prioritization should be carefully considered when assessing the possible causes of missing heritability in unsolved IRD patients.

Despite the encouraging success rate of our analysis, still 66.6% of studied cases (22 of 33) remained genetically unresolved after WGS, suggesting that there are pathogenic variants that eluded our analysis. Unsolved patients may carry pathogenic variants that were called but did not pass the stringent prioritization criteria (i.e., genes not yet associated with IRD). It is also possible that some pathogenic variants are in low complexity, GC-rich, or repetitive regions, therefore not called due to their lower coverage. Some patients may carry pathogenic variants within regulatory regions of IRD genes for which interpretation criteria are complex and yet to be established. As mentioned before, IRDs can show a complex etiology with reported cases of hypomorphic alleles and incomplete penetrance that may have contributed to the genetically unexplained quota of patients.

Overall, this study confirms the diagnostic power of WGS as an all-in-one test that allows the detection of all types of variants, from SNVs (both in coding and non-coding regions) to SVs. We believe that the great strides in bioinformatic analysis and variant interpretation criteria will soon lead to an even higher diagnostic yield and clinical utility of WGS to investigate the missing genetic cause in unsolved IRD patients.

Data and code availability

The data generated during this study (pathogenic variants identified) are available at the Leiden Open (source) Variation Database (LOVD) (http://www.lovd.nl).

Web resources

http://www.interactive-biosoftware.com

http://www.lovd.nl

http://franklin.genoox.com

http://sph.uth.edu/retnet