Retinal Diseases Caused by Mutations in Genes Not Specifically Associated with the Clinical Diagnosis

Purpose When seeking a confirmed molecular diagnosis in the research setting, patients with one descriptive diagnosis of retinal disease could carry pathogenic variants in genes not specifically associated with that description. However, this event has not been evaluated systematically in clinical diagnostic laboratories that validate fully all target genes to minimize false negatives/positives. Methods We performed targeted next-generation sequencing analysis on 207 ocular disease-related genes for 42 patients whose DNA had been tested negative for disease-specific panels of genes known to be associated with retinitis pigmentosa, Leber congenital amaurosis, or exudative vitreoretinopathy. Results Pathogenic variants, including single nucleotide variations and copy number variations, were identified in 9 patients, including 6 with variants in syndromic retinal disease genes and 3 whose molecular diagnosis could not be distinguished easily from their submitted clinical diagnosis, accounting for 21% (9/42) of the unsolved cases. Conclusion Our study underscores the clinical and genetic heterogeneity of retinal disorders and provides valuable reference to estimate the fraction of clinical samples whose retinal disorders could be explained by genes not specifically associated with the corresponding clinical diagnosis. Our data suggest that sequencing a larger set of retinal disorder related genes can increase the molecular diagnostic yield, especially for clinically hard-to-distinguish cases.


Introduction
Inherited retinal diseases are a heterogeneous group of disorders that lead to retinal dysfunction and visual impairment. Retinitis pigmentosa (RP) is a group of progressive retinal dystrophies affecting about 1 in 3000 individuals [1,2]. RP causes night blindness and progressive loss of peripheral vision in early stages and loss of central vision later in life. Leber congenital amaurosis (LCA) represents a group of severe retinal disorders causing profound visual disability recognizable shortly after birth or within the first year of life. LCA affects about 1 in 50,000 people and is characterized by early onset visual impairment, nystagmus, and non or poorly recordable responses in the electroretinogram (ERG) [3]. Familial exudative vitreoretinopathy (FEVR) is a retinal disorder associated with defective retinal angiogenesis. FEVR is characterized by avascularity in the peripheral retina with variable clinical presentations, from no symptoms to early onset blindness [4]. To date, pathogenic variants in about 55, 19, and 5 genes are known to cause non-syndromic RP, LCA, and FEVR, respectively [5][6][7].
Targeted next-generation sequencing (NGS) has been used extensively for the molecular diagnosis of retinal diseases [8,9]. The diagnostic yields of targeted NGS panels range from 36% to 82% for RP 18% to 40% for LCA, and 49% for FEVR [6, [10][11][12][13][14]. It has been reported that patients with a descriptive clinical diagnosis of retinal disease may carry pathogenic variants in genes not specifically associated with that diagnosis due to the substantive phenotypic overlap and genetic heterogeneity [6, [15][16][17]. For example, apparently non-syndromic patients with retinitis pigmentosa may carry pathogenic variants in the Bardet-Biedl syndrome gene, BBS1 [18]. Patients with severe visual impairments can have pathogenic variants in pattern dystrophy gene PRPH2 [6]. Thus, tests focused on a specific group of genes for a particular clinical diagnosis may not detect variants in genes not typically associated with that condition. Despite a few reports in research settings, this phenomenon has not been evaluated systematically in clinical diagnostic laboratories that fully validate all target genes to minimize both false negatives and false positives [12].
Previously, our laboratory analyzed 98 RP, 13 LCA, and 12 FEVR samples by targeted capture NGS. A total of 207 ocular disease genes were captured and sequenced for each of these samples (S1 File). However, we focused the sequence analysis on 66 RP, 19 LCA, and 4 FEVR genes that have been clinically validated and are well known to be associated with the corresponding disorders. As a result, definitive molecular diagnoses were previously established in 73% (72/98) of RP, 46% (6/13) of LCA, and 25% (3/12) of FEVR cases, which are similar to previously published results mentioned above (S1 Table).
We hypothesized that a portion of the unsolved cases might be caused by pathogenic variants in other retinal disease genes not analyzed initially. Since the sequence data of 207 ocular disease-related genes are readily available, we analyzed the remaining genes of the 42 unsolved cases in this study. Our data underscore the clinical and genetic heterogeneity of retinal disorders and suggest that sequencing a larger set of related retinal disease genes can increase the molecular diagnostic yield.

Patient samples
A total of 42 DNA samples tested negative for pathogenic variants in the clinically validated 66 RP, 19 LCA, or 4 FEVR genes at CLIA-certified and CAP-accredited Baylor Miraca Genetics Laboratories (BMGL) were further analyzed as described below. The subsequent analyses were performed by protocols approved by Institutional Review Board for Human Subject Research of Baylor College of Medicine, and complied with the tenets of the Declaration of Helsinki. Patient information was de-identified prior to the analysis.

Sequencing analyses and variant interpretation
Our targeted capture NGS approach has been described recently [12]. Briefly, a customdesigned DNA probe library was used to capture target exons and 20bp of the flanking intron regions of 207 ocular disease genes (S1 File). Indexed captured samples were pooled to be loaded onto each lane of the flow cells for sequencing on a HiSeq2000 (Illumina, Inc., San Diego, CA, USA) with 100 cycle single-end reads. Clinical validations were performed for 66 RP, 19 LCA, and 4 FEVR genes that are well-known to be associated with the corresponding disorders (https://www.bcm.edu/research/medical-genetics-labs/, test code 2190, 5090, 5250). Those regions with coverage <20X, usually GC rich or highly repetitive, were covered by PCR/ Sanger sequencing. An average of 1000X per base sequence depth was achieved and 3-12 candidate variants were obtained per sample [12]. American College of Medical Genetics guidance was used for the interpretation of sequence variants [19]. Pathogenic variants were confirmed by Sanger sequencing.

Copy number variation analysis
Analysis and detection of exonic CNVs were performed according to our recently published method [20]. Briefly, normalized coverage of each exon of a test sample was compared to the mean coverage of the same exon in the reference samples. The exons with possible CNVs were depicted automatically. The script for the detection of CNVs is deposited at https://sourceforge. net/projects/cnvanalysis. Candidate CNVs were confirmed by a custom-designed oligonucleotide CGH array [21].

Summary of identified pathogenic variants
Pathogenic variants in other retinal disease genes not previously analyzed were identified in five RP, two LCA, and two FEVR cases, accounting for 19% (5/26), 29% (2/7), and 22% (2/9) of unsolved RP, LCA, and FEVR cases, respectively (Tables 1 and 2). Additionally, single heterozygous pathogenic variant in autosomal recessive disorders were identified in two RP patients (data not shown). All the reported variants were confirmed by Sanger sequencing. Taken together, variants in other retinal disease genes were identified in 21% of (9/42) unsolved patients (Table 1).

Patients with variants in syndromic retinal disease genes
Six of these patients have variants in syndromic retinal disease genes. Syndromic features other than an isolated retinal dystrophy may be detected or may develop later in life than the time at which the patient is evaluated for visual impairment. Thus, the additional systemic features may not have evolved or may be overlooked at the time of the initial ophthalmologic evaluation. These constitutional features may also be less evident than expected. Patient 1 harbors a heterozygous reported nonsense change, c.1677C>A (p.Y559 Ã ), and a heterozygous novel . This patient had widespread rod and cone dystrophy but did not have obesity, speech pathology, intellectual disability, polydactyly, or hypogonadism ( Fig 1A). However, the family subsequently disclosed that he was born with a horseshoe kidney and had had surgical repair of an ureterocele, which has been observed in BBS [27]. Patient 2, a 25-years-old woman, was referred for molecular diagnosis of non-syndromic RP and was found to have a heterozygous well established pathogenic variant, c.1169T>G (p.M390R), and a heterozygous nonsense variant, c.1645G>T(p.E549 Ã ), in the BBS1 gene. Defects in BBS1 cause Bardet-Biedl syndrome 1 (BBS1) [MIM: 209900], an autosomal recessive and genetically heterogeneous ciliopathy characterized by retinitis pigmentosa, obesity, kidney dysfunction, polydactyly, behavioral dysfunction, and hypogonadism. It has also been previously described that BBS1 mutations can result in a wide spectrum of phenotypes, including apparently nonsyndromic retinitis pigmentosa, if other clinical features are not carefully sought for [18]. Patient 3 carries a homozygous, well known pathogenic variant, c.1169T>G (p.M390R), in BBS1 gene. This patient had widespread rod and cone dystrophy but did not have obesity, developmental delay, speech pathology, intellectual disability, or renal defects ( Fig 1B). However, after revealing the results for BBS1 mutation, the parents disclosed the previous excision of a small extra digit, consistent with polydactyly in BBS [12]. Patient 4 carries compound heterozygous novel  c.625-2A>G, and a one copy whole gene deletion, in the NPHP1 gene (Fig 2).

Patients with variants in non-syndromic retinal disease genes
The working clinical diagnoses of some of these patients were confounded by their ages when first evaluated by the ophthalmologist. For example, patient 7 was referred for RP testing but was found to have a heterozygous variant, c.341C>T (p.T114I), in the GUCA1A gene after our expanded analysis. This variant was reported previously in a single patient with cone dystrophy but has never been reported in public databases [24]. It is known that mutations in GUCA1A cause autosomal dominant cone dystrophy or cone-rod dystrophy (CRD)  [32]. Since this patient was evaluated at 1 year of age, he may have been too young for a unique clinical diagnosis. Definitive molecular diagnosis also reveals the wide clinical spectrum of non-syndromic retinal diseases. Patient 9 was referred for FEVR testing because he had typical FEVR features including tractional retinal detachment, vitreous hemorrhage, retinal dragging, and peripheral avascular retina. However, our analysis identified a well-known hemizygous pathogenic variant, c.214G>A (p.E72K), in the X-linked retinoschisis (XLRS) gene RS1 [25]. XLRS is characterized by splitting of the neural retina (schisis). Schisis occurs in the inner retinal layer and is usually different from the retinal detachment in FEVR, which is the split between the neural retina and the retinal pigment epithelium [33]. It has been reported that some XLRS patients with RS1 mutations had atypical fundus findings, including tractional retinal dragging, exudative detachment, and vitreous hemorrhage, all of which are consistent with the clinical presentation of patient 9 [34,35]. Therefore, our data demonstrate that molecular diagnosis can refine or modify the descriptive clinical diagnosis and subsequently change the counseling for associated features, other complications, and recurrence risks for both the patient and the family.

Discussion
Our study suggests that a substantial portion of undiagnosed and apparently non-syndromic retinal dystrophy cases can be explained by pathogenic variants in genes not specifically associated with the corresponding clinical diagnosis. We evaluated this hypothesis systematically by analyzing sequence variants in 207 ocular disease-related genes and identified pathogenic variants in genes not specifically associated with the corresponding clinical diagnosis in 9 out of 42 cases that could not be explained by smaller set of disease-specific genes. These cases account for 19% (5/26) of RP, 29% (2/7) of LCA, and 22% (2/9) of FEVR cases in our unsolved patient cohort (S1 Table). Our analysis increases the overall diagnostic rate from 73% to 78% for RP. The increases in diagnostic rate for LCA and FEVR are much higher, but the overall solved rates remain much lower than that for RP. The increases may be underestimates, because in this study we only focused on defined pathogenic variants and excluded variants of unknown significance. Nevertheless, our results suggest that sequencing a larger set of retinal disorderrelated genes can increase substantially the diagnostic yield and help to identify unexpected genotype-phenotype correlations.
Our study provides valuable reference to estimate the fraction of clinical samples whose retinal disorders may be explained by genes not specifically associated with the corresponding clinical diagnosis. Similar findings have been reported elsewhere [6, [15][16][17][36][37][38][39][40][41]. Our approach has unique advantages. We ensured 100% coverage by fill in the low or no coverage regions such as ORF15 of RPGR gene, or regions with high GC content and/or homologous sequences, by Sanger sequencing of specifically amplified PCR products. In addition, the average coverage depth is consistently at~1000X per base, that allows the detection of heterozygous exonic CNVs [20]. The CNV detection algorithm has been integrated into the routine analytical pipeline for clinical application that has been validated in parallel with exon targeted oligonucleotide array CGH [20]. These unique features of our panel-based NGS approach greatly improve clinical sensitivity. With 100% coverage and the ability to detect SNVs and CNVs simultaneously, a negative result from deep NGS panel analysis suggests that the disease-causing variants are unlikely in the target regions, and other options such as WES may be considered. WES has been used for the molecular diagnosis of retinal disorders, however, it does not ensure 100% coverage and is not validated clinically for CNVs [42][43][44][45]. Thus, in our experience, if a sample with isolated retinal disorder is negative for all candidate genes in the targeted panel, the yield of additional WES analysis is minimal. Indeed, four of the samples in this study also had clinical WES in our laboratory, and no additional reportable variants were identified.
Early identification of mutations in syndromic retinal disease genes of apparently non-syndromic and isolated retinal disease patients could lead to timely and pre-planned management before acute presentation of more serious features of the complete syndrome [16,46]. Among the nine patients with variants in genes not specifically associated with the corresponding clinical diagnosis, six carry variants in syndromic retinal disease genes. It has been reported that pathogenic variants in syndromic retinal disease genes can lead to wide spectrum of phenotypes, from non-syndromic retinal dystrophy to full syndromes [6, 16,18]. In addition, it is not unusual that ophthalmologists tend to attend more to the ocular findings of the patient, while extra-ocular syndromic features were either unsought by the enquiring physician, unreported by the patient or family, or not yet developed at the time of eye evaluation. Therefore, it is important for physicians to gather clinical data comprehensively and be aware of the phenotypic overlapping among many retinal disorders. The remaining three patients carry variants in other non-syndromic retinal disease genes. Many factors, such as age at evaluation, wide variance in the phenotypic spectrum of diseases, genetic modifiers, and environmental exposures, may confound the incisive distinction of subtle differences between two clinically similar retinal phenotypes. In all these situations, sequencing a larger set of related retinal disease genes can help to capture variants in unexpected genes, increase the molecular diagnostic rate, reach a definitive clinical diagnosis, and lead to accurate prognosis and improved management of the patient.
Other genetic and technical factors may account for the remaining molecular etiology of retinal diseases. First, highly GC-rich, highly repetitive and/or homologous regions could not be captured, sequenced, and aligned unambiguously by targeted capture NGS [47,48]. For example, the open reading frame 15 (ORF15) of RPGR gene, which is a RP mutational hotspot, contains a~300 bp highly repetitive region that cannot be unambiguously analyzed by conventional capture NGS [12,49,50]. Long-range PCR followed by NGS may be used to identify variants in these regions. Second, nucleotide changes not at the canonical splice site, or changes in regulatory regions such as promoter, or 3' and 5' untranslated regions, may be disease-causing. For example, we have added specific probes to capture the frequent intronic pathogenic variant c.2991+1655A>G in CEP290 in our panel [51]. Third, exonic deletion/ duplications have been shown to cause retinal diseases [52][53][54]. We have developed recently a method to detect exonic CNVs efficiently with capture based NGS data [20]. Here, we successfully identified a one copy whole gene deletion of NPHP1 in patient 5 (Fig 2). Our data suggest that exonic CNV analysis should be included in the NGS panel-based clinical testing of retinal diseases to increase the diagnostic yield. Lastly, novel disease genes, yet to be identified, may account for other unsolved retinal disease cases. To identify novel disease genes, whole exome sequencing or whole genome sequencing may be considered.
Supporting Information S1 File. The genes and transcripts included in the capture design. (TXT) S1