Development of High-Throughput Clinical Testing of RPGR ORF15 Using a Large Inherited Retinal Dystrophy Cohort

Abstract

Purpose: Mutations in the ORF15 region of RPGR account for approximately half of all X-linked retinitis pigmentosa cases. However, a robust high-throughput method for the detection of ORF15 mutations has yet to be validated. We set out to develop the first clinically validated next-generation sequencing (NGS) method for the detection of mutations in this difficult-to-sequence region, including test accuracy and coverage data.

Methods: As part of a blind-test, 145 research samples, previously tested by Sanger sequencing, and 81 clinical samples were evaluated using NGS of long-range PCR products fragmented with Illumina's Nextera library preparation kit (method 1), or with Centrillion's OneTube technology, supplemented with duplication analysis using an ORF15-specific in-silico array (method 2). DNA fragments were analyzed using Agilent's DNA 1000 assay, and sequencing was done on Illumina's MiSeq 2×150 or HiSeq2500 2×100. NextGENe by SoftGenetics was used for data analysis and variant calling.

Results: The Nextera library preparation method produced 24 cases of discordance due to (in order of decreasing occurrence) false-negatives, incorrectly called variants, and a false-positive. Subsequent use of a new, OneTube NGS library preparation method, supplemented with duplication analyses, resolved discordance between Sanger and NGS data in all cases. This improvement in variant detection accuracy was largely attributed to improvement in random fragmentation offered by the enzymatic OneTube method, resulting in more complete coverage of the highly repetitive ORF15 region. Minimum coverage was roughly 320 reads for Nextera and 6800 reads for OneTube (normalized for total read counts).

Conclusions: This paper documents the first clinically validated NGS method for reliable, high-throughput sequencing of RPGR ORF15. Sensitivity and specificity of the new method were 100%, with the caveat of unclear zygosity calling for one large duplication case. These findings demonstrate a reliable and practical implementation for NGS-based diagnosis of RPGR ORF15 mutations. They also provide the foundation for targeted, high-coverage sequencing of any other repetitive regions within the genome.

Retinitis pigmentosa (RP; OMIM #268,000) is the most commonly diagnosed inherited retinal dystrophy (IRD). It is clinically and genetically heterogeneous, with at least 64 causative genes currently identified (http://www.sph.uth.tmc.edu/retnet/; provided in the public domain by The University of Texas Health Science Center, Houston, TX, USA).¹ It is estimated that the more severe X-linked form of RP (xlRP) constitutes 10% to 20% of all RP cases^2–6, however, the actual figure is likely to be higher. It has been reported that roughly 9% of families thought to have an autosomal dominant form of RP (adRP)⁷ and 15% of male sporadic cases⁸ can be attributed to mutations in X-linked genes retinitis pigmentosa 2 (RP2; MIM 300757, http://omim.org/entry/300757; provided in the public domain by Johns Hopkins University, Baltimore, MD, USA) and retinitis pigmentosa GTPase regulator (RPGR; MIM 312610, http://omim.org/entry/312610). Overall, RPGR mutations account for >70% of xlRP cases and, as such, is the most common RP gene.^9–12 

RPGR has been characterized extensively.^13–16 It encodes several isoforms, but only the largest of these, Isoform C (NM_001034853), is highly expressed in the retina and involved in the pathogenesis of RP (https://ghr.nlm.nih.gov/gene/RPGR#normalfunction; provided in the public domain by National Institutes of Health, Bethesda, MD, USA).¹⁰ This isoform, also known as RPGR ORF15, spans 4767 nucleotides encoding a 1152-amino acid protein (NP_001030025). Over 60% of all RPGR mutations are clustered to its unique terminal exon, ORF15 (c.1754-3459), which encodes a 567-amino acid C-terminus, rich in glutamic acid and glycine.¹⁰ The occurrence of these mutations is thought to be due to slippage of DNA polymerase on the highly repetitive, 1kb, purine-rich region (c.2184-3162). 

Therefore, accurate detection of ORF15 mutations is central to the diagnosis of xlRP and, consequently, to subsequent genetic counseling and family planning considerations. A robust, accurate and scalable test for ORF15 also will be necessary for implementation of personalized medicine strategies, such as for acceptance into gene-therapy clinical trials and for the prescription of approved treatments that may arise from these trials.^5,17 

Despite next-generation sequencing (NGS) becoming the clinical standard for the genetic testing of most IRDs, clinical testing of ORF15 still relies on traditional Sanger sequencing.^18–21 This is largely due to difficulties arising from the highly repetitive nature of ORF15. 

Recently, a NGS-based method was developed for sequencing long-range PCR (LR-PCR) products of ORF15, exhibiting a mutation detection rate (MDR) of 31% in 51 patients.²² However, the absence of well-documented clinical validation via blind, independent testing means that test sensitivity, specificity, and reproducibility remain largely unknown. 

We report the first clinically validated NGS method for the sequencing of ORF15, one of the most difficult-to-sequence regions in the genome. 

Materials and Methods

DNA Samples

A total of 226 DNA samples were tested. These samples were obtained from pedigrees that contained individuals clinically diagnosed with xlRP or that showed a pattern consistent with X-linked retinal disease. 

De-identified samples for 145 individuals from 52 pedigrees were sourced from the Australian Inherited Retinal Disease Registry and DNA Bank.²³ Samples were collected and analyzed in accordance with the Declaration of Helsinki. Ethics approval was granted by the Sir Charles Gairdner Hospital Human Research Ethics Committee (Human Ethics Approval Number 2001-053). Samples were sourced from affected and unaffected males and females, including carrier females, from RP families with a clear or suspected X-linked pattern of inheritance. 

These DNA samples had previously been Sanger sequenced by the Australian Inherited Retinal Disease Registry (AIRDR); 40 had tested negative for ORF15, while ORF15 mutations had been detected in the remaining 105 samples (54 from affected males and 51 from females with or without symptoms of RP). They were provided to the Molecular Vision Laboratory (MVL) for NGS testing, without any accompanying information. 

An additional 81 samples from male patients clinically diagnosed with X-linked RP were used for further validation of this method. ORF15 mutations identified in these samples by NGS were later confirmed by targeted Sanger sequencing. 

NGS testing of all 226 samples was done by the MVL. Concordance of Sanger sequencing and NGS results for the blind-tested research samples was evaluated by the AIRDR in Australia. The MVL evaluated the clinical samples. 

Target Enrichment, NGS Library Preparation, and Sequencing

Long range PCR (LR-PCR) was used to amplify a 2064 base pair (bp) region of the RPGR gene containing ORF15. DNA (400–500 ng) was amplified in a total reaction volume of 50 μL using Takara LA Taq DNA polymerase (# RR002M) and forward and reverse primers, AGCAGCCTGAGGCAATAGAA and CAAAATTTACCAGTGCCTCCT (5′-3′) respectively. The PCR program used was 96°C for 3 minutes, 30 cycles of 94°C for 30 seconds, and 68°C for 15 minutes, followed by 72°C for 5 minutes, with a final hold at 4°C. LR-PCR products were purified by QIAquick PCR Purification Kit (https://www.qiagen.com/us/; provided in the public domain by Qiagen, Hilden, Germany). 

NGS libraries were prepared using the Nextera DNA Library Preparation Kit (method 1; https://www.illumina.com/; provided in the public domain by Illumina, San Diego, CA, USA) or the OneTube NGS library preparation kit (method 2; http://www.centrilliontech.com/; provided in the public domain by Centrillion Technologies, Palo Alto, CA, USA). The profiles of DNA fragments were analyzed using the DNA 1000 Assay on the Bioanalyzer 2100 (https://www.agilent.com/home; provided in the public domain by Agilent Technologies, Santa Clara, CA, USA). Samples were sequenced on Illumina MiSeq using the 2 × 150 bp MiSeq Reagent Kit v2 or Illumina HiSeq2500 using TruSeq SBS Kit v3-HS (2 × 100 bp) plus TruSeq PE Cluster Kit v3-cBot-HS. Samples were allocated with a minimum of 400,000 reads, yielding a target average coverage of at least 20,000 reads for the ORF15 region. 

Bioinformatics and Data Analysis

FASTQ files were generated from Illumina's BaseSpace Sequence Hub and aligned using NextGENe by SoftGenetics, LLC (http://www.softgenetics.com/NextGENe.php; provided in the public domain by State College, PA, USA). VCF and BAM files were exported to GeneticistAssistant by SoftGenetics (http://www.softgenetics.com/GeneticistAssistant.php) for variant interpretation and mutation identification. Alignment criteria were set to 85% overall base matching percentage and variant detection at 5% minor allele frequency. 

Duplication analysis was done using an artificial reference sequence consisting of 160 bp contigs separated by a 50 bp homopolymer “A.” Contigs were centered on the duplication breakpoint, defined as the junction of the duplicated regions, and provided with a flanking sequence to reach a contig length of 160 bp. The sequence was generated using a script, stepping through each position from c.2000 to c.3300 and iterating over all duplication sizes from 1 to 200 bp, for a total of 260,000 possible duplications tested. The sequence also can be generated omitting in-frame duplications for frameshift-only analysis. Alignment criteria were set to 100% overall base matching percentage with no allowance for indels. Duplication hits were defined as contigs with >100 aligned reads. Zygosity testing was done on the specific duplication contig only, with alignment criteria relaxed to 95% and allowing for indels. 

Results

NGS Library Preparation From LR-PCR Products

During development of this method for the sequencing of ORF15, the Nextera method was used initially for fragmentation of the LR-PCR product. However, several inconsistencies between Nextera NGS and Sanger sequencing results were detected. These included 12 false-negatives and 1 false-positive. In a further eight cases, mutations were incorrectly identified. Two benign duplication variants also were either incorrectly called or not detected (Table). We suspected this discordance was due to the repetitive sequence in ORF15 preventing the transposon-based Nextera fragmentation method from generating a well-represented sequencing library. 

Table

 

Concordance in Variant Data Between Sanger Sequencing and NGS of RPGR ORF15 is Significantly Improved With OneTube-NGS and Duplication Analysis

 

We then tested the OneTube enzymatic method for library preparation. Average read length of adapter-ligated fragments was much smaller using OneTube compared to Nextera, with peaks observed at 340 and 600 bp, respectively. Using OneTube NGS, we then retested all but two cases (see Table) randomized with a group of Nextera-Sanger concordant controls. As a result, variants were correctly identified in 21 of the 24 discordant cases. Previously concordant results also were confirmed. 

Coverage of ORF15 and Mutation Detection Accuracy

Of the ORF15 mutations identified, 65% were concentrated within the difficult-to-sequence, highly repetitive region (c.2184-3162), for which Nextera and OneTube NGS data highlight a relative lack of coverage (Fig. 1). Minimum coverage using OneTube NGS (∼6800 reads) was significantly higher compared to Nextera (∼320 reads), while average coverage of the entire exon was comparable at approximately 36,000 and 32,000 reads for OneTube and Nextera, respectively (Fig. 1). In setting a coverage threshold of 500 reads as a quality control metric, OneTube NGS achieved 100% coverage of ORF15, while Nextera NGS achieved 96.8%. These results highlight a critical gap in coverage in a region in which ORF15 mutations were concentrated. All Sanger-identified mutations that went undetected using the Nextera method were localized to this region (Fig. 1). 

Figure 1

 

Mutation coverage curves and data for ORF15 of RPGR from NGS of LR-PCR products fragmented with Nextera (top) and OneTube (bottom). Vertical red lines represent the position of missed mutations using Nextera (top). Blue bars represent the position and number of unique variants using OneTube (bottom; secondary y-axis). Note the considerable increase in minimum coverage using OneTube (red boxes). Coverage data from a representative sample.

Manual inspection (using NextGENe Viewer) of the mutations initially missed by Nextera-NGS revealed that the mutation sites coincided with highly repetitive areas containing sequence quality issues and alignment difficulties, resulting in many single nucleotide variants being flagged by the software with varying allele frequencies. We found that poor sequence quality masked some of the mutations, highlighting the difficulty in separating true mutations from false-positives under these circumstances. Gaps in coverage also were associated with a higher proportion of sequence data being derived from the ends of reads, where run-specific artifacts commonly are found. When these occur in a significant proportion of available reads at a given location, true-positives can be difficult to distinguish from false-positives. With OneTube-NGS data, we demonstrated that these issues could be overcome with a more uniform distribution of reads staggered across the region of interest, coupled with sufficient depth of coverage to minimize the effect of individual artifacts. 

Duplication Analysis

Given the increased prevalence of large duplications within repetitive regions, and the remaining three cases of discordance, duplication analysis was performed using an ORF15-specific in silico array. This method detected the remaining frameshift duplication (c.2144_2216dup) and two benign, in-frame duplications (c.2820_2840dup, c.2721_2744dup), concordant with Sanger sequencing data (Table). Specifically, under strict alignment criteria, approximately 3000 reads aligned perfectly to the 73 bp (c.2144_2216dup) contig (Fig. 2a), while less than 10 reads mapped to other contigs. Further analysis was successful in determining zygosity for the 21 bp (c.2820_2840dup; Fig. 2b) and 24 bp (c.2721_2744dup) duplications, but not for the larger 73 bp duplication (c.2144_2216dup). For a 73 bp duplication, the wild-type allele in the case of a heterozygous duplication would be expected to appear as a 73 bp deletion. However, alignment difficulties, owing to deletion size approaching the size of our read length (100 bp), limited the zygosity calling confidence for larger duplications with our current pipeline. 

Figure 2

 

Duplication analyses. (a) Duplication detection using alignment to an artificial reference sequence. Perfect alignment over this unique duplication junction indicates the presence of c.2144_2216dup within ORF15 of RPGR in this sample. (b) Duplication zygosity testing of a mixed sample containing a negative control and a sample homozygous for ORF15 benign duplication, c.2820_2840dup. The wild-type sequence appears as a 21 bp deletion against the reference sequence for this duplication, while sequence containing c.2820_2840dup, shows complete alignment.

Therefore, the combined method of OneTube fragmentation, supplemented with duplication analysis, successfully detected all Sanger-identified ORF15 variants among the blind-tested Australian cohort of suspected xlRP pedigrees, in which ORF15 mutations were causative for disease in approximately 50% of cases (Lamey et al., manuscript in preparation, 2018). 

Discussion

Mutation of the highly repetitive and unstable ORF15 region of RPGR is widely reported to cause 25% to 70% of xlRP cases.^2–7 As for other similarly repetitive regions, ORF15 has proven refractory to variant detection using traditional NGS methods, and Sanger sequencing of ORF15 can be labor-intensive, time-consuming, and subject to allele dropout. Coupled with increasing clinical volumes and the demand for a more timely turnaround of test samples, there is an urgent need for an accurate, high-throughput mutation detection method to assist in the diagnosis and management of xlRP. 

We addressed this shortfall and present the first validation of a complete NGS-based ORF15 test within a standardized clinical pipeline. This follows recently published work using high-fidelity PCR for ORF15 NGS, which represented a significant step forward in ORF15 screening in the research setting, with improvements in coverage and mutation detection.²² 

The marked improvement in accuracy using the OneTube fragmentation method can be attributed to its superior coverage of this region. Based on this study, depth of coverage is a main factor affecting the accuracy of NGS of repetitive regions, such as ORF15. Our minimum coverage (∼7000 reads) is significantly higher than that for recently reported NGS-based ORF15 screening methods (1–2000 reads).²² Using our approach of blind-testing against a large number of Sanger-sequenced samples from an xlRP cohort, and comparing the variant detection rate and accuracy of OneTube versus Nextera, we demonstrated the amount of coverage required for successful clinical NGS of this region, and the inadequacy of the Nextera fragmentation method in this instance. This approach distinguished the validation of our protocol from similar recent work,²² and exemplified the importance of such clinical validation in NGS method development. 

The OneTube method also was successfully validated against over 50 female samples from suspected xlRP pedigrees. This is a significant step forward, as female samples can be notoriously difficult to analyze by Sanger sequencing due to the prevalence of in-frame polymorphic indels.²⁴ Benefits include informed genetic counseling and the provision of family planning options. 

The short-read length of NGS fragments also presents a challenge in the analysis of highly repetitive regions, in which large deletions and duplications relative to read length are more common. Large deletions typically are detected by normal variant calling. However, large duplications are masked by alignment across the region, with the only distinguishing feature being a single, duplication-specific breakpoint between duplicated regions. Consequently, highly repetitive regions demand stricter sequencing requirements, and the resulting bottleneck in the bioinformatics pipeline is becoming increasingly problematic. 

With the aim of overcoming this issue for NGS-based screening of repetitive regions, custom duplication analysis was performed. This successfully identified all three Sanger-identified ORF15 duplications that previously were undetected. This distinguishes our pipeline from previous work,²² resulting in detection of large duplications by high throughput ORF15 screening, which has not been reported previously. This absence in the literature may be due to the inability to detect large duplications or to sample size differences. 

Collectively, these results emphasized the robustness of our method, with the caveat of uncertain zygosity in one case. While further development is underway to optimize zygosity calling, rare cases of uncertain zygosity likely will be resolved via independent Sanger sequencing of NGS-identified variants, as per our standard pipeline for our other clinical panels. 

Continuing Development of NGS-Based Diagnostics: ORF15 and Beyond

These results highlighted the weaknesses of previous screening methods and suggested the improvements that can be made. We suspect that many critical mutations within ORF15 have been underreported previously, and we hope that this advancement will lead the way for a comprehensive, accurate, and practical implementation of NGS-based diagnosis for difficult regions within the genome. Beyond its application to RPGR ORF15, we believe the LR-PCR–based NGS method outlined here shows exciting potential in its ability to target any specific region within the genome for accurate, specific, low-cost, and high-coverage sequencing. 

We hope that our data will provide a useful benchmark for other groups trying to sequence difficult genomic regions. This method already has been applied to finding breakpoints in several patients with large deletions identified by array CGH analysis (data not shown) and forms the basis for whole gene sequencing assays currently in development for several critical genes in clinical trial pipelines. 

In conclusion, we demonstrated the efficacy of the new OneTube sample preparation method, supplemented with duplication analysis, in achieving robust coverage of the entire ORF15 region, with 100% mutation detection sensitivity and specificity for our sample population within a standardized clinical pipeline. A key outcome of this study is the evidence that clinical validation should include previously tested samples in the development of new variant detection methods. 

Acknowledgments

Supported by Retina Australia, the Western Australian DNA Bank, Department of Medical Technology and Physics, Sir Charles Gairdner Hospital, and the National Health & Medical Research Council (NHMRC) of Australia (#1116360). 

Disclosure: J.P.W. Chiang, Molecular Vision Laboratory (I, E); T.M. Lamey, None; N.K. Wang, None; J. Duan, None; W. Zhou, P; T.L. McLaren, None; J.A. Thompson, None; J. Ruddle, None; J.N. De Roach, None