Applying next generation sequencing with microdroplet PCR to determine the disease-causing mutations in retinal dystrophies

A set of 82 genomic DNA samples was included in this study. This cohort includes 74 patients with clinical diagnosis of Choroidermia (CHM), CRD, CSNB, LCA, MD, Occult Macular Dystrophy (OCMD), and RP from the NEI Ophthalmic Genetics Clinic or tested in the NEI DNA Diagnostic Laboratory. Clinical evaluations included measurement of best-corrected visual acuity, visual field assessment (either kinetic or static perimetry depending on patient presentation), fundus biomicroscopy and indirect ophthalmoscopy. Optical coherence tomography (Stratus OCT 3 or Cirrus, Carl Zeiss Meditec, Inc., Dublin, CA) was performed. Color and autofluorescence fundus imaging was obtained. International Society for Clinical Electrophysiology of Vision (ISCEV) standard electroretinography responses were obtained using a Burian-Allen contact lens electrode and an LKC-2000 system (LKC Technologies, Gaithersburg, MD). Systemic examinations were performed whenever necessary. Genomic DNA was isolated from peripheral leukocytes using the Gentra Puregene Blood Kit (Qiagen Inc., Valencia, CA) in accordance with the manufacturer’s protocol. Whenever available, a blood sample from affected and unaffected family members was collected for co-segregation analysis. Pedigrees were constructed based on patient interviews. This study was reviewed and approved by the Combined Neuroscience Institutional Review Board of the National Institutes of Health and informed consent was obtained from each participant as adhering to tenets of the Declarations of Helsinki.

A total of 82 samples including 74 patients and 4 unaffected family members from 67 unrelated RD families were included in this study. In addition, 3 mutation positive samples that were previously studied in this lab and one DNA control sample acquired from Coriell Repository were included in this study [12]. Clinical examinations were performed in the NEI Ophthalmic Genetics Clinic and patients were clinically diagnosed with CHM, CRD, CSNB, LCA, MD, OCMD, and RP. Patient samples referred from outside clinics were previously tested in the NEI DNA Diagnostic Laboratory.

NGS sequencing data were processed and analyzed through a bioinformatic pipeline by the Genomic Workbench and the Alamut Visual. An average of 1.14 million reads was generated per sample, an average of 59.5% of reads mapped to the targeted regions. Bioinformatic analysis indicated that this procedure was able to reach 99% of coverage on target sequences with averages of 97% at 1X, 95% at 10X, and 90% at 20X. The average read depth was 119X. The read depth per sample was evaluated at 1X, 10X, and 20X and summarized in the Fig. 1.

On average, our bioinformatic pipeline generated 500–600 variants per sample. After filtering out common polymorphisms with a frequency > 2% in any of the variant databases queried, including 1000 Genome build 201,105 and 201,011 (Genomes Project 2010) and dbSNP130 (NCBI dbSNP database build 130), an average around 30–60 variants remained per sample. The remaining variants were further filtered based on our secondary functional selection as stopgain or stoploss variants, missense variants, synonymous variants with predicted splicing effects, and intronic variants within the 25 bp of exon flanking regions with predicted splicing effects. We had an average of 10 candidate variants per sample for further Alamut pathogenicity analysis and reporting.

Sanger sequencing was used to analyze whether a variation segregated with the disease phenotype in the corresponding family (whenever additional family members were available). Using this approach, we confirmed 63 previously determined pathogenic variants and 28 novel variants predicted as pathogenic or likely pathogenic in our analyses (Additional file 3: Table S3).

The clinical correlation was performed after mutations and variations were determined and verified by Sanger sequencing. We found that phenotype-genotype relationship could be established in 38 of the 67 probands (56.7%) and the additional affected family members were confirmed with the familial mutations (Table 1). We included 6 samples from a large Occult Macular Dystrophy (OCMD) family. In the three affected members, we confirmed the known mutation, p.R45W in the RP1L1 gene and excluded the potential of additional pathogenic variants in other retinal disease related genes. We also confirmed the mutation carrying status in the other three unaffected family members.

We found inconsistences between the genotypes and the clinical diagnoses in 4 samples (Table 2). The confirmed mutations in these 4 patients had been reported with different retinal disorders for the genotype-phenotype relationship in the literature [13‐15], but not reported for conditions that our patients were diagnosed with. Patient RD1–12 was diagnosed with sporadic CRD. Two C2orf71 gene pathogenic variants were confirmed: the p.W505* was reported in unrelated patient [15] and the p.S1090Ifs*17 was novel and not found in populations (ExAC: Exome Aggregation Consortium, Cambridge, MA) (http://​exac.​broadinstitute.​org/). Additionally, three heterozygous stop-gain mutations in unrelated genes (BBS4, TYRP1, and SLC45A2) were found (Table 2). The C2orf71 gene mutations were reported in patients with autosomal recessive RP [16], not CRD. Our review of the patient’s clinical conditions still concluded a diagnosis of CRD (Fig. 2a and b). Patient RD11–05 was diagnosed with sporadic RP. Exam was positive for a preserved central visual field island with 20/32 acuity and some macular cystic changes. Electroretinography was consistent with retinitis pigmentosa. We identified a known mutation in the GUCA1A gene, p.P50L, which has been reported in patients with autosomal dominant CRD (Downes et al. 2001), not RP. Her clinical manifestation was consistent with RP (Fig. 2c). Patient RD14–05 was diagnosed with sporadic RP. Best corrected visual acuity measured at 20/100 and central visual field was limited to 10 degrees with a few peripheral islands. Findings were consistent with advanced RP (Fig. 2d and e). We found two variants in the TRPM1 gene, p.W398R and p.R1305H. The p.W398R was novel and was predicted as pathogenic, but the p.R1305 was found in African population with an allele frequency of 2% in our most recent ExAC search (rs13380059). Patient RD11–06 was also diagnosed with RP at age of 74 yr. and no family history was reported. Best corrected visual acuity was at 20/800. The patient reported a 10 year history of decline in peripheral and central visual function and denies any problems with vision as a child. Visual field was limited to a paracentral small island and electroretinography responses were extinguished (Fig. 2f). The patient was found carrying one candidate allele per gene in three genes including 2 reported mutations and 1 reported variant in the dbSNP database in heterozygous state. One of the reported mutations, p.P575L in the GUCY2D gene, has a very high allele frequency in the African population in ExAC (unlikely pathogenic). The other known mutation, p.P406L in the TYR gene, has been reported in patients with autosomal recessive Oculocutaneous Albinism Type I. The reported variant, IVS9 + 1G > A in the TYRP1 gene, is predicted as pathogenic for a likely splicing error, but mutations in the TYRP1 have been reported with autosomal recessive Oculocutaneous Albinism Type III.

We also found single heterozygous mutation status as insufficient to confirm the genotype-phenotype relationship in another 5 patients. The patients RD4–05 and RD11–02 were found heterozygous for novel EYS gene mutations, p.W2046* and p.Y1893Rfs*12 respectively. The mutations in the EYS gene have been reported with autosomal recessive RP. Since possible gene or partial gene copy number variation or deep intronic mutations were not excluded in this study for these patients, it was insufficient to conclude the correlation in these 2 patients. A similar situation presented itself for patients RD14–03 and RD15–01 who carried single heterozygous mutations in the ABCA4 gene, IVS + 5G > A splicing error and p.I1562T respectively (Table 2). It was reported that a portion of patients with the Stargardt disease were bearing single heterozygous mutations in the ABCA4 gene [17, 18]. It could be considered as consistent with the clinical diagnosis of RP, even if the second mutation in their ABCA4 gene was not identified. Adding further complexity to the interpretation, patient RD14–03 was also found carrying two additional USH2A gene variants, p.S2867 L and p.V3518I. Of these USH2A gene variants, the p.S2867 L has a very low allele frequency in populations (ExAC) (http://​exac.​broadinstitute.​org/) and is predicted as pathogenic, while the p.V3518I has an allele frequency of 2% in the African population and 6 homozygous individual have been identified in the African population(ExAC) (http://​exac.​broadinstitute.​org/). The disease-responsible gene for patient RD14–03 remains unidentified. The patient RD15–01 was found with additional pathogenic variant in the RPGRIP1 gene, p.P585S - previously reported in patient with autosomal recessive CRD [19].

We have also identified additional previously reported pathogenic variants in some of the probands for whom we established genotype-phenotype relationship in this study (Table 3). These additional pathogenic alleles could be sufficient for a clinical correlation by itself (RD14–08 and RD12–02), or have been reported multiple times in literature (RD20–07, RD12–02, RD12–01, RD13–02, RD12–07, and RD13–08, or are predicted as obligate disease causing variants (RD11–03). For example, the p.C984R in the GUCY2D gene was not reported in the literature and predicted as pathogenic. Mutations in the GUCY2D have been reported in autosomal dominant CRD. The p.C984R is a valid pathogenic variant by itself.

To technically verify the sequencing results, we included 4 previously tested DNA samples. Two of them were tested in the resequencing chip by Affimatrix, in which we had previously identified and validated one missense mutation in the GUCY2D gene (RD1–01) and no mutation in a Coriell DNA sample RD1–03, ND0068*B1 [12]. The other 2 samples were from previous Sanger sequencing: one 1 bp deletion in the CNGB3 gene in one sample (RD13–08) and two OCA gene mutations in the other sample (RD13–07). All of the mutations were correctly identified through the bioinformatics pipeline. In addition to the previously identified mutation, we also detected additional pathogenic variants in the RD13–08 (Table 3).