Hidden Splicing Variants in Inherited Retinal Degeneration: Discovery and Functional Insight

Abstract

Purpose: To enhance the molecular diagnosis of inherited retinal degeneration (IRD) by systematically identifying pathogenic splicing variants and characterizing their transcript-level consequences.

Methods: We analyzed 738 IRD families who underwent targeted gene panel sequencing. A splicing variant detection pipeline, integrating two computational algorithms (SpliceAI and dbscSNV_ADA) with functional validation via minigene assays, was implemented to detect splice-disrupting variants beyond canonical sites.

Results: Splicing variants accounted for 14% of genetically diagnosed families. Of these, 4% were newly identified through our combined computational and experimental platform. Notably, 28% of all splice-disrupting variants, located in noncanonical, exonic, or deep-intronic regions, would likely have been missed by conventional analysis pipelines, which typically prioritize protein-coding changes and canonical splice sites, and often lack systematic evaluation of splicing effects beyond these regions. Five recurrent splice-disrupting variants were observed across multiple families, including EYS:c.5644+5G>A, which caused exon truncation and was found in 11 unrelated families. Functional assays confirmed aberrant splicing, and the associated phenotypes were consistent with known disease presentations.

Conclusions: This study demonstrates the utility of a combined splicing variant detection platform in uncovering hidden pathogenic variants and improving IRD diagnostic yield. These findings have implications for refining genetic testing and guiding the development of splicing-targeted therapies.

Figure 1.

Genetic diagnostic outcomes and classification of splicing variants in IRD cases. (A) Pie chart illustrating the genetic diagnostic outcomes of 738 IRD cases. A total of 14% (105/738) were diagnosed with splicing variants, including 10% (71/738) with well-documented splicing variants and 4% (34/738) with experimentally validated splicing variants. (B) Stacked bar chart showing the distribution of 128 splicing-variant alleles identified in 105 families, classified by variant type and genomic location. Of these 128 alleles, 72% were well documented variants and 28% were experimentally validated. Among the well-documented variants, 45% (58/128) were located in canonical splice dinucleotides and 27% (23/128) in NCSS sequences. Among the experimentally validated variants, 17% (22/128) were located in NCSS sequences, 3% (4/128) in exonic regions, and 8% (10/128) in deep-intronic regions.

Figure 2.

Aberrant splicing event detected exclusively by NGS analysis. (A) Gel electrophoresis of the COL2A1: c.1942-6C>G variant reveals no detectable size difference in the separated bands. (B) Transcript sequencing visualization using the Integrative Genomics Viewer (IGV) demonstrates a 5-bp extension in the mutant transcript compared to the wild-type (WT). (C) Sashimi plot analysis highlights aberrant splicing induced by the COL2A1: c.1942-6C>G variant, leading to intron retention.

Figure 3.

Retinal images and splicing analysis of patients with homozygous variants EYS: c.5644+5G>A and CNGB1: c.1122-3344G>A. The top panel (A–C) represents a patient with the homozygous EYS: c.5644+5G>A splicing variant, while the bottom panel (D–F) corresponds to a patient with the homozygous CNGB1: c.1122-3344G>A variant. (A, D) Fundus autofluorescence (FAF) imaging and color fundus photography, revealing retinal atrophy and vascular attenuation. (B, E) The OCT scans of both eyes (OD: right eye; OS: left eye), demonstrating structural abnormalities such as outer retinal thinning and macular degeneration. (C, F) Transcript analysis, comparing wild-type (WT) and mutant splicing patterns. Sashimi plots highlight aberrant splicing events caused by the variants, including exon truncation due to the EYS: c.5644+5G>A variant and pseudoexon (PE) activation induced by the CNGB1: c.1122-3344G>A variant.

Figure 4.

Comparison of SpliceAI and dbscSNV_ADA predictions for variants located in noncanonical splice sequences. A heatmap showing the prediction scores of SpliceAI (blue), dbscSNV_ADA (yellow), and outcomes for minigene validation (gray) for consensus noncanonical splice site variants. Darker colors indicate higher predicted scores. ROC curve comparing the sensitivity and specificity of SpliceAI (blue) and dbscSNV_ADA (orange) in predicting splice-altering effects within consensus noncanonical splice sites, including the splice acceptor site (SAS; –20 to –3 of the intron and first 2 bp of the exon) and splice donor site (SDS; last 2 bp of the exon and +3 to +6 of the intron). The AUC values indicate the performance of each model, with SpliceAI achieving an AUC of 0.67 and dbscSNV_ADA an AUC of 0.83.

Figure 5.

Distribution of predicted splicing variants identified in this study. A total of 31 predicted splicing variants were identified and categorized based on their genomic positions relative to consensus splice site regions. Variants located within deep-intronic regions (outside of the splice consensus motifs) are shown in yellow. Variants positioned in splice region consensus motifs are grouped into the splice donor site (SDS; last 2 bp of the exon and +3 to +6 of the intron) and splice acceptor site (SAS; −20 to −3 of the intron and first 2 bp of the exon) and are highlighted in green. Variants falling within internal exonic regions (excluding the first and last 2 bp) are shown in white. Variants that were experimentally validated to induce aberrant splicing in this study are indicated with an asterisk.