Most irreversible blindness results from retinal disease. To advance our understanding of the etiology of blinding diseases, we used single-cell RNA-sequencing (scRNA-seq) to analyze the transcriptomes of ~85,000 cells from the fovea and peripheral retina of seven adult human donors. Utilizing computational methods, we identified 58 cell types within 6 classes: photoreceptor, horizontal, bipolar, amacrine, retinal ganglion and non-neuronal cells. Nearly all types are shared between the two retinal regions, but there are notable differences in gene expression and proportions between foveal and peripheral cohorts of shared types. We then used the human retinal atlas to map expression of 636 genes implicated as causes of or risk factors for blinding diseases. Many are expressed in striking cell class-, type-, or region-specific patterns. Finally, we compared gene expression signatures of cell types between human and the cynomolgus macaque monkey, Macaca fascicularis. We show that over 90% of human types correspond transcriptomically to those previously identified in macaque, and that expression of disease-related genes is largely conserved between the two species. These results validate the use of the macaque for modeling blinding disease, and provide a foundation for investigating molecular mechanisms underlying visual processing.
PURPOSE: Current sequencing strategies can genetically solve 55-60% of inherited retinal degeneration (IRD) cases, despite recent progress in sequencing. This can partially be attributed to elusive pathogenic variants (PVs) in known IRD genes, including copy-number variations (CNVs), which have been shown as major contributors to unsolved IRD cases. METHODS: Five hundred IRD patients were analyzed with targeted next-generation sequencing (NGS). The NGS data were used to detect CNVs with ExomeDepth and gCNV and the results were compared with CNV detection with a single-nucleotide polymorphism (SNP) array. Likely causal CNV predictions were validated by quantitative polymerase chain reaction (qPCR). RESULTS: Likely disease-causing single-nucleotide variants (SNVs) and small indels were found in 55.6% of subjects. PVs in USH2A (11.6%), RPGR (4%), and EYS (4%) were the most common. Likely causal CNVs were found in an additional 8.8% of patients. Of the three CNV detection methods, gCNV showed the highest accuracy. Approximately 30% of unsolved subjects had a single likely PV in a recessive IRD gene. CONCLUSION: CNV detection using NGS-based algorithms is a reliable method that greatly increases the genetic diagnostic rate of IRDs. Experimentally validating CNVs helps estimate the rate at which IRDs might be solved by a CNV plus a more elusive variant.