Since the initial identification of recessive mutations in the ABCA4 gene in several large STGD pedigrees [3], a multitude of studies have attempted to assess the mutational spectrum of ABCA4 in various ethnically restricted populations (Italy [11,12]; Spain [13,14]; Japan [15]; Portugal [16]; Hungary [17]; Denmark [18]). While each population appears to carry some founder mutations accounting for a significant proportion of disease alleles, hundreds of extremely rare or private variants have also been identified [19]. This allelic heterogeneity hampers the viability of standard panel-based mutation screening techniques and has lead to the creation of an ABCA4 array designed to genotype over 600 individual DNA variants [19]. While this array is an effective screening tool, the high rate of private mutations in STGD prevents it from being a comprehensive solution.

Another factor making ABCA4 such a difficult diagnostic target is its genomic structure. The large number of coding exons in ABCA4 gene (50 coding exons resulting in a 6.8kb mRNA [Refseq NM_000350]) makes PCR-based sequencing costly and time-consuming. Currently, the cost of PCR and dideoxy-based targeted ABCA4 exon sequencing is approximately half that of WES. The costs of data analysis are difficult to compare, but WES analysis is highly automatable and can be performed in parallel as opposed to the highly labor-intensive analysis required to identify rare variants from individual sequence traces. Approaches using targeted next-generation sequencing hold great potential for ABCA4 mutation screening [20,21], and may reduce the cost of single gene sequencing significantly.

Using a combination of methods to screen for ABCA4 mutations in 121 patients from a heterogeneous patient population in our natural history study of Stargardt Disease, only half (59/121, 49%) of the cases were found to have two known or putative disease-causing ABCA4 mutations on separate alleles. Nearly one third have a single mutation, and the remaining ~20% have no definite or probable disease-causing ABCA4 mutations (manuscript in preparation). Highly sensitive next-generation targeted re-sequencing of ABCA4 has been implemented on an independent sample set with similar results [20]. Whether individuals with putative Stargardt Disease with one or zero detected ABCA4 mutations have non-traditional variation (e.g. promoter variants or epigenetic changes) missed by current mutation screening techniques or mutations in other genes is unclear at this time.

Providing a genetic diagnosis for individuals with putative Stargardt Disease is thus an ongoing challenge. Here we apply whole exome sequencing to nine putative STGD cases to assess the efficacy of this approach for molecular genetic diagnosis of this condition. All patients were previously screened and found to have zero (eight samples) or one (one sample) disease-causing mutation in ABCA4. While some of these probands express ocular imaging phenotypes not typically observed in classic STGD, all were diagnosed clinically as having STGD due to the early age at which signs of macular degeneration were initially identified.

Four out of five ABCA4 mutations were not detected by PCR/dideoxy based exon sequencing and yet were detected in the context of whole exome sequencing in this study. Of the five total disease-causing ABCA4 alleles found by WES, four (all except p.Q636X) have previously been described and are genotyped by the current ABCR array available from Asper Biotech, supporting the use of the genotyping array as a first-pass screening tool due to its low cost.

Two of these mutations were clearly visible as heterozygous variants upon re-analysis of the original sequencing traces. One of the newly identified mutations is visible on the original sequence chromatogram but is located within the first five bases of the read and thus indistinguishable from normal background noise. The chromatogram corresponding to the remaining variant had a high level of noise throughout. Other laboratories have noted similarly missed variants [31]. While currently considered the “gold standard” for molecular genetic diagnosis of SNV-type mutations, PCR-based dideoxy sequencing relies heavily on manual sequence trace inspection and is thus subject to human error. This method should and will continue to be relied upon for many applications, including mutation validation and genotyping of known alleles. However, it is difficult for a molecular diagnostic laboratory to maintain a high level of surveillance quality when performing a large number of these assays, as is required for mutation screening in STGD and other macular dystrophies.

Variant detection by exome sequencing is by comparison far more automated and less susceptible to this type of error. The sources of error in WES are systematic and readily measured. The major sources of erroneous findings by WES are insufficient or uneven read depth (“coverage”), mismapping, and incorrect variant annotation. While these sources of error are complex to resolve, they are highly consistent which has allowed computational advancements to dramatically improve results. In contrast, the confounding issues facing capillary gel electrophoresis-based dideoxy sequencing have remained relatively stable since the inception of this method. Our data supports the contention that WES provides reliable and efficient mutation detection for SNVs and indels in ABCA4, and is an attractive alternative to other methods.

Surprisingly, RDS/PRPH2 mutations were identified by exome-seq in three individuals. This gene has been linked to several other retinopathies including pattern macular dystrophy [30,32-36] and retinitis pigmentosa [37], and it has been previously suggested that mutations in PRPH2 can mimic the STGD phenotype [38] or modify disease severity [5]. However, PRPH2 has not been considered a significant contributor to classic STGD. Indeed, two of the individuals carrying a heterozygous PRPH2 mutation show a “stellate” pattern dystrophy in one eye. While this is consistent with previous reports of PRPH2-related disease [4,28,37-40], it is too subtle a finding to confidently sub-classify patients. The lack of concordance between two eyes of the same individual in both cases (Additional file 6 Figure S4 and Additional file 7 Figure S5) suggests incomplete expressivity of this endophenotype, suggesting that some cases with PRPH2 related disease may show no identifiable signs of pattern dystrophy.

Additional mutations were detected by direct re-sequencing of PRPH2 in a broader cohort of STGD cases. Complicating matters, two out of the eight people with rare missense or nonsense mutations in PRPH2 also have mutations in ABCA4. One of these participants, a compound heterozygote for ABCA4 mutations has two affected relatives who share the PRPH2 mutations but neither of the ABCA4 mutations. For these relatives, the PRPH2 mutation appears to be disease-causing. In the proband with a single ABCA4 mutation and a PRPH2 mutation, the genetic etiology of the disease is less clear. At this time, it is not possible to assess the effect of this potentially digenic scenario, though it has been previously observed [5]. The presence of rare, coding PRPH2 variants in STGD cases with one or more ABCA4 mutations suggests that one must be cautious regarding attributing apparent cases of Stargardt Disease to ABCA4 mutations when only ABCA4 has been screened. Though rare, these cases support the implementation of exome sequencing to detect potentially confounding genetic mutations as well as enable the study of the effect of multi-gene mutation combinations on disease progression and severity.

Mutations in CRB1 have been associated with a wide array of retinal dystrophies, including retinitis pigmentosa and Leber congenital amaurosis [41]. Here we report the first finding of a heterozygous nonsense mutation in CRB1 in an individual with macular degeneration. Given that CRB1 is involved in normal retinal development, this variant is an intriguing functional candidate. A thickening of the fovea (without the cavitations associated with macular edema) observable by OCT in this participant is consistent with other cases of CRB1-related disease. In the absence of a functional validation of the p.K801X variant or analysis of additional cases with nonsense variants in CRB1, it is not possible at present to confidently assign a molecular diagnosis to this case and we thus categorize this variant as being of unknown significance. This inability to causally link a suspect variant with phenotype is a significant limitation of the technique, and will likely be typical of early clinical WES efforts.

While it is attractive to attempt to correlate clinical and phenotypic information with our newly acquired genetic data, sample size is a significant limitation. For example, one might hypothesize that individuals with PRPH2 dysfunction would have abnormal cone/rod responses due to its restricted expression to the outer disc segment of photoreceptors [42]. However, given the range of variability across Stargardt cases and the small number of PRPH2 positive cases, it is not possible to assess any such correlation at present. Collaboration and data sharing amongst multiple STGD research groups is needed to sufficiently increase the sample size to evaluate any genotype-phenotype correlations.

In this study, we suffer from the unavailability of complete family history information and genetic material from related individuals, particularly for participants with variants outside of ABCA4. These limitations are common in clinical practice. While both history and DNA would provide critical insight into mode of inheritance and shed light on the functional consequences of observed genetic variants, we identified a clear molecular basis of disease in six out of nine cases.

Despite near total coverage of most genes, including both ABCA4 and PRPH2 in all nine cases, several of our cases elude a definitive molecular diagnosis. One case has only a single ABCA4 mutation, one has a variant of unknown significance in CBR1 and one sample has no demonstrable mutations in any gene with known association to retinal or macular degeneration. Careful manual inspection of genes carrying potential compound heterozygous, homozygous, and protein-truncating variants in these individuals did not yield any strong functional candidates for novel STGD disease genes. It is now generally accepted that every individual carries a significant burden of potentially pathogenic DNA variants within the known exome [43], and STGD cases are no different in this regard. Which, if any, such variants in genes not currently associated with retinopathy in our sample contribute to disease onset and progression cannot be assessed at this time, nor can we rigorously address the potential of other genetic variants acting in a dominant fashion or in a digenic manner. The exome data provides an opportunity to make a concerted effort to re-analyze these data as new disease genes and genetic modifiers are implicated in retinal and macular dystrophies. In the meantime, several of our cases elude molecular diagnosis. While such cases represent a small proportion of the overall STGD participant population, they may eventually provide invaluable insight into the pathogenesis of STGD and the biology of the retina.

The disease phenotype of STGD can be quite variable, even amongst individuals carrying identical or putatively similar ABCA4 mutations. Age of onset, electroretinogram response, disease progression and disease severity are all clinically relevant parameters of phenotypic variability in these participants. While some genotype-phenotype correlations have been found for population-specific ABCA4 mutations [11], the majority of this variability remains unexplained. One major benefit of exome sequencing over targeted ABCA4 mutation screening is the potential for identifying disease-modifying alleles in other genes. If approached properly, clinical exome sequencing can enable such studies as a byproduct of providing highly sensitive molecular testing.

Knowledge of genome-wide functional variation may be particularly critical to enrollment of participants in future clinical trials. Restoration or replacement of ABCA4 protein function would likely be insufficient for a participant harboring disease-causing mutations in additional genes, such as seen with PRPH2 in this study. Thus the positive identification of two mutations in ABCA4 may not be a proper endpoint for molecular diagnosis in STGD. In contrast, exome sequencing can help place STGD participants into genomic context as it identifies both common and rare functional genetic variation.

Clinicians now have a multitude of options available to them for STGD mutation screening. Targeted approaches continue to grow in sensitivity and shrink in cost [20], but they will always be fundamentally limited to specific genes or alleles. From these small-scale results, we find very strong potential for STGD molecular diagnosis by exome sequencing. Presently, an approach following up array-based ABCA4 mutation screening with exome sequencing is a prudent and cost-effective alternative to direct re-sequencing. In the near future, exome sequencing and indeed whole genome sequencing will likely be sufficiently inexpensive (and potentially covered by health insurance), eliminating the need for STGD-specific mutation screening.