USH2A Molecular Research InitiativeProject 01 · USH2A G1526R Molecular Mechanism

Patient-initiated · computationally characterized · experimentally unresolved

Molecular evidence

Molecular Evidence

Last scientific review: 2026-10-12.

RNA Splicing Evidence

SpliceAI, Pangolin and AlphaGenome agree on enhancement at the cryptic donor-associated site and reduction at the annotated donor-associated site. This agreement concerns site-level predictions. Only AlphaGenome supplies the explicit donor–acceptor pair in these runs; Pangolin’s executed classification head does not separately identify donor and acceptor types. 6 7 8 Supporting claims: C010prediction

For the 1-Mb input, AlphaGenome returned the minus-strand intron [216097213,216175301), using a 0-based half-open convention. This corresponds to last-retained donor base 216175302 and first-retained acceptor base 216097213, both 1-based genomic coordinates. The annotated donor is 216175252. Supporting claims: C011predictionC016conditional

The exon-22 acceptor ranks first by ALT output among the cryptic donor’s 98 returned partners in both selected tracks. However, the annotated exon21–exon22 junction still has a higher ALT prediction. The preferred cryptic pairing must not be described as the dominant RNA product. Supporting claims: C013prediction

Proposed cryptic junctionREF predictionALT prediction
Embryonic camera-type eye0.00064938724972307680.005134492181241512
Adult brain0.000255330698564648630.0007750487420707941

These are original model outputs. They are not measured read counts, RNA percentages or probabilities of complete transcripts. Supporting claims: C014prediction

The 1,048,576-bp input covers this junction; the 131,072-bp input excludes its acceptor and cannot test the same pair. Embryonic camera-type eye is not a retina or photoreceptor assay, and adult brain is an additional context rather than retinal validation. Supporting claims: C012prediction

Competing pairings remain visible in the evidence tables: intronic acceptors 216111456 and 216111459, exon-23 acceptor 216089139, an internal exon-22 acceptor at 216097159, a weak short-terminal-exon candidate at 216175102, and a weak whole-exon21-skipping comparator. Some pairings do not define a complete onward RNA path. Supporting claims: C015prediction

Exon 21 has 231 nucleotides. A predicted donor 50 nucleotides before its annotated end leaves 181 nucleotides and joins to exon 22. The diagram labels both donors, the acceptor and the two coordinate conventions.
FIG-03 · Predicted cryptic donor and proposed 50-nt event. On the minus strand, genomic coordinates decrease in transcript direction. The predicted donor’s last retained base is 216175302; the annotated donor’s is 216175252. Joining the predicted donor to exon-22 acceptor 216097213 conditionally removes c.4578_4627 (50 nt). Base labels are 1-based; the raw intron is [216097213,216175301), 0-based half-open. Hatched bases are omitted only under the proposed reconstruction.

Limitations: Not to scale. The pair is predicted; the complete RNA and its abundance have not been observed. Other junctions are assumed unchanged.

Claims shown in this figure: C011C016C019 Evidence: USH2A-E002, USH2A-E006, USH2A-E007 References: REF-001, REF-008
Text description of the figure

A minus-strand exon diagram distinguishes the predicted donor at 216175302 from the annotated donor at 216175252. Joining the predicted donor to exon-22 acceptor 216097213 removes 50 nucleotides from the end of exon 21 under the stated reconstruction.

Exon 21 has 231 nucleotides. A predicted donor 50 nucleotides before its annotated end leaves 181 nucleotides and joins to exon 22. The diagram labels both donors, the acceptor and the two coordinate conventions.

The diagram is wider than the screen: scroll it horizontally (drag, swipe, or arrow keys once focused) to read every panel.

FIG-03 · Predicted cryptic donor and proposed 50-nt event. On the minus strand, genomic coordinates decrease in transcript direction. The predicted donor’s last retained base is 216175302; the annotated donor’s is 216175252. Joining the predicted donor to exon-22 acceptor 216097213 conditionally removes c.4578_4627 (50 nt). Base labels are 1-based; the raw intron is [216097213,216175301), 0-based half-open. Hatched bases are omitted only under the proposed reconstruction.

Limitations: Not to scale. The pair is predicted; the complete RNA and its abundance have not been observed. Other junctions are assumed unchanged.

Conditional RNA Consequences

Using the predicted cryptic-to-exon22 junction while retaining all other long-reference processing deletes c.4578_4627, exactly 50 nt. The deleted interval is transcript bases 5017–5066 or genomic bases 216175252–216175301, inclusive. This mature RNA is reconstructed from the prediction and reference transcript; AlphaGenome did not directly output it. Supporting claims: C016conditional

The resulting reading frame encodes Arg1526–His1527–stop1528, corresponding to a 1,527-amino-acid product if translated. The premature stop ends 124 nt before the next junction, a positional configuration compatible with potential NMD. The presence of such a stop does not demonstrate RNA decay or protein production. Supporting claims: C017conditionalC018conditional

The predicted deletion-containing splice product has not been experimentally observed in this project. Actual abnormal RNA abundance is unknown. No RNA assay was performed, so this limitation must not be presented as a negative experimental finding. Supporting claims: C019unresolved

The deleted 50-base sequence and a codon table compare reference, correctly spliced G1526R and the hypothetical RNA. Only the reconstructed RNA changes codon 1527 to histidine and codon 1528 to a stop. NMD and protein production remain unmeasured.
FIG-04 · Conditional RNA consequence and translation. The exact 50-nt deletion shifts the reading frame. Translation of the reconstructed sequence yields Arg1526–His1527–stop1528 and 1,527 amino acids if protein is made. The stop ends 124 nt before the next junction in the reconstructed long transcript. This is compatible with potential NMD, not a demonstration of decay.

Limitations: The table is calculated sequence evidence, not mass spectrometry or an RNA measurement. NMD efficiency and actual product amounts are unknown.

Claims shown in this figure: C016C017C018C019 Evidence: USH2A-E007 References: REF-001
Text description of the figure

The exact 50-nt deletion changes the frame. The codon table yields Arg1526, His1527 and stop1528, giving 1,527 amino acids only if translated. The stop is 124 nucleotides before the next junction, compatible with potential NMD but not proof of decay.

The deleted 50-base sequence and a codon table compare reference, correctly spliced G1526R and the hypothetical RNA. Only the reconstructed RNA changes codon 1527 to histidine and codon 1528 to a stop. NMD and protein production remain unmeasured.

The diagram is wider than the screen: scroll it horizontally (drag, swipe, or arrow keys once focused) to read every panel.

FIG-04 · Conditional RNA consequence and translation. The exact 50-nt deletion shifts the reading frame. Translation of the reconstructed sequence yields Arg1526–His1527–stop1528 and 1,527 amino acids if protein is made. The stop ends 124 nt before the next junction in the reconstructed long transcript. This is compatible with potential NMD, not a demonstration of decay.

Limitations: The table is calculated sequence evidence, not mass spectrometry or an RNA measurement. NMD efficiency and actual product amounts are unknown.

Protein Structural Analysis

Six WT/G1526R pairs of local ColabFold/AlphaFold2 models were compared using matched model parameters and seeds. The construct spans residues 1364–1709. Local Cα RMSDs were 0.152–0.341 Å, partly overlapping the 0.132–0.230 Å variation among WT models; this does not establish a large local backbone disruption. 9 10 Supporting claims: C021predictionC022prediction

Target pLDDT was 88.69–90.56 in WT and 78.00–83.56 in G1526R models. These confidence values are not native-function scores. Small predicted structural differences also do not exclude altered folding, processing or cellular behavior. Supporting claims: C023predictionC029unresolved

The primary Rosetta calculation failed geometry qualification in 26 of 30 mutant structures. Its large positive values were excluded from native-stability conclusions. Two more-flexible calculations removed gross geometry flags and gave +13.706/+9.135 REU, but protocol dependence and positive reverse controls still prevented a calibrated native folding-energy estimate. 11 Supporting claims: C024predictionC025prediction

ThermoMPNN forward means were +1.288 and +1.306 on its predicted kcal/mol scale for two input contexts; reverse means were approximately +0.060 and +0.058. This is a limited, model-dependent destabilization lead, not a measured native free-energy change. Its scale is not interchangeable with Rosetta REU. 12 Supporting claims: C026predictionC025prediction

Three paired 20-ns WT/G1526R MD comparisons, plus a preserved 5-ns preparation sensitivity, total 125 ns. Local RMSD differences across pairs reversed direction in the common 10–20 ns window (−0.664 to +1.185 Å). No uniform dynamic defect or equilibrium convergence was established in the modeled isolated LG1 domain. 13 Supporting claims: C027predictionC028prediction

AlphaFold, Rosetta and MD do not establish native protein function in this study. Supporting claims: C029unresolved

View the raw-junction review archive · Browse evidence and methods

Three plots show local backbone RMSD, complete-LG1 RMSD and residue-1526 confidence. Matched WT/G1526R displacements overlap within-WT model variability, and mutant confidence is lower. Neither result is a functional assay.
FIG-05 · Matched local protein structure comparison. Unchanged source figure comparing six matched WT/G1526R local predictions with within-allele model variability. Local residues are 1518–1534; complete-LG1 RMSD is also shown. Bars are medians, and connecting lines pair model/seed conditions. The 15 within-allele pairwise comparisons share models and are not independent biological replicates. pLDDT is model confidence, not energy or function.

Limitations: Local predicted structures do not establish native folding energy, cellular behavior or normal function. The model ensemble is descriptive.

Claims shown in this figure: C021C022C023C029 Evidence: USH2A-E008 References: REF-009, REF-010
Three plots show local backbone RMSD, complete-LG1 RMSD and residue-1526 confidence. Matched WT/G1526R displacements overlap within-WT model variability, and mutant confidence is lower. Neither result is a functional assay.

The diagram is wider than the screen: scroll it horizontally (drag, swipe, or arrow keys once focused) to read every panel.

FIG-05 · Matched local protein structure comparison. Unchanged source figure comparing six matched WT/G1526R local predictions with within-allele model variability. Local residues are 1518–1534; complete-LG1 RMSD is also shown. Bars are medians, and connecting lines pair model/seed conditions. The 15 within-allele pairwise comparisons share models and are not independent biological replicates. pLDDT is model confidence, not energy or function.

Limitations: Local predicted structures do not establish native folding energy, cellular behavior or normal function. The model ensemble is descriptive.