A SNP-hunting ASO silences mutant rhodopsin, in organoids and in mice

What the work claims

This is a translational proof-of-concept study from Spaag, Wu, Yun and colleagues at the University of Tübingen, Columbia, Radboud University Medical Center, the Telethon Institute of Genetics and Medicine and partner clinics. The strategy, SNARE (SNP-guided Silencing of Aberrant RHO Expression), exploits rs7984, a benign c.-26A/G polymorphism in the RHO 5' untranslated region. Among 445 RHO-adRP patients across four European cohorts, 28.5% (n=127) are heterozygous for this SNP, and phasing showed the pathogenic variant sits in cis with the c.-26A allele in roughly half of them, so an ASO targeting one SNP allele can in principle serve many mutations at once. The authors estimate RHOligo-A alone could address about 15.4% of RHO patients, and adding the complementary G-allele ASO would reach roughly 28.5%.¹

The lead candidate emerged from a 32-gapmer screen (16-mers with a 10-base DNA gap flanked by modified RNA wings) run in engineered HEK293T reporter lines, then validation moved to human material: patient-derived iPSC retinal organoids carrying the class III variant R135W, with a CRISPR-corrected isogenic wild-type control, and finally a new mouse in which the endogenous Rho locus was fully humanized to carry the c.-26A SNP together with P347L, a prevalent European variant in a different mechanistic class.¹

How it works

Gapmer ASOs recruit RNase H1 to degrade the RNA strand of a DNA-RNA hybrid, and a single mismatched base pair in the gap is enough to tilt cleavage strongly toward the matched allele. That is the whole trick: the SNP, not the mutation, is the drug target. In organoids treated by gymnotic uptake at 5 or 15 µM, the authors quantified allele selectivity not by absolute knockdown but by the enrichment of the non-targeted transcript using targeted amplicon sequencing; A-ASO_5, later methylated to reduce immunogenicity and renamed RHOligo-A, shifted the allelic ratio in both configurations, with non-target transcript increases around 16 to 27% depending on dose.¹

The phenotypic readout in organoids was spatial: rhodopsin normally concentrates apical to the outer limiting membrane, and the R135W mutant mislocalizes. Shell-by-shell quantification across two independent differentiation batches (77 wild-type, 74 mutant and 22 treated organoid images) showed significantly restored RHO levels in three inner-nuclear-layer shells after 20 days, and a trend toward restoration in the outer-segment region that did not reach significance.¹ In the humanized mouse, a single intravitreal dose at postnatal day 21 to 27 produced bulk RHO transcript knockdown of 30 ± 5% (low dose, 10 µM) and 41 ± 11% (high dose, 50 µM) at 10 days, but allele-resolved sequencing showed the selectivity was real: 64 ± 3% knockdown of the c.-26A mutant allele versus 15 ± 4% of the wild-type partner allele at high dose. Scotopic rod b-waves recovered from 65 ± 14 µV in control eyes to 147 ± 15 µV in high-dose treated eyes at 1 month, and outer nuclear layer preservation reached about 79% in treated eyes versus roughly 53 to 55% in controls over months 1 to 3.¹

Where a skeptic should push

The single most load-bearing assumption is that single-nucleotide discrimination observed in vitro survives the messy pharmacology of an eye. In the reporter screen, good candidates still knocked down the counter-allele by roughly 10 to 30%, and in vivo at high dose the wild-type allele lost 15 ± 4% of its transcripts. The authors argue haploinsufficiency is tolerated, which is true, but they also concede that the lower boundary of safe RHO abundance is undefined; notably, wild-type humanized mice showed reduced ERG responses at the high dose, which the authors attribute to off-target RHO knockdown rather than toxicity, a distinction that matters commercially and clinically.¹

Generalization is the second problem, and it is structural. The organoid work uses one patient, one variant (R135W), one iPSC line and its isogenic control. The in vivo efficacy uses a different variant (P347L) in a mouse engineered to carry the human SNP the drug needs, meaning the model was built around the therapeutic's discriminator. Each system validates the drug in a context pre-selected to be compatible; no model tests both a non-SNP-matched patient background nor a second donor's organoids. And the organoid phenotype that improved was largely in the inner nuclear layer, while the disease-relevant outer-segment rescue remained a non-significant trend.

The safety battery is genuinely thoughtful, with a caveat that is itself a lesson. RNA-seq of treated wild-type organoids flagged 219 differentially expressed genes, more than the 133 seen with a scrambled control; intersecting with in silico off-target predictions nominated 26 candidates, but every one vanished after correcting for retinal pigment epithelium, a common organoid differentiation by-product, contaminating samples differently. A dedicated RPE assay then showed zero differentially expressed genes for RHOligo-A against a 394-gene positive-control ASO. In other words, the clean safety answer depended on recognizing that the organoid readout was silently confounded by its own cellular contamination.¹

Where retinal organoids sit in ASO development

The non-obvious opportunity is a division of labor. Retinal organoids will not predict what a mouse can show about rescue of vision, and this paper uses them for what they are uniquely good at: endogenous, human, allele-resolved target engagement. The organoid supplied the patient-genotype pharmacology (gymnotic uptake, amplicon-sequenced allele ratios), a human-cell transcriptomic safety panel sensitive enough to require RPE correction, and a localization phenotype tied to disease mechanism. That is a template for allele-specific silencing programs across gain-of-function inherited disease, where the same SNP-first logic applies wherever a common benign polymorphism sits in cis with heterogeneous pathogenic variants.

The threat is over-reading organoid safety and efficacy. A transcriptome with 219 shifted genes that only becomes clean after a post-hoc cell-composition correction should temper anyone planning to run ASO triage on raw organoid RNA-seq; without the isogenic controls and the orthogonal RPE assay, this program could have nominated phantom off-targets or missed real ones. And the coverage math deserves cold water: a headline of mutation-agnostic therapy resolves, after phasing, to about 15.4% of RHO patients for RHOligo-A alone, with the combined two-ASO strategy reaching 28.5%, contingent on routine SNP phasing in diagnostic pipelines that many clinics do not currently perform.

The bottom line

Established: gapmer ASOs can discriminate the c.-26A/G SNP with useful allele selectivity across reporters, patient-derived retinal organoids and a fully humanized mouse; a single intravitreal dose of RHOligo-A in that mouse drove mutant-allele suppression, near-normal rod ERG responses at 1 month, and about 79% structural preservation at 3 months; and the molecule passed a multi-layer in vitro safety assessment including a cytokine release panel in PBMCs from nine donors. Unestablished: efficacy in any human, performance on a second donor or variant class in organoids, the dose window between mutant silencing and wild-type suppression, and durable safety of repeat intravitreal gapmer dosing, which the authors say still requires formal pharmacokinetics and toxicology. What would confirm the platform is allele-selective knockdown with functional benefit in a second, independently derived patient organoid line and an SNP-unmatched humanized control. What would break it is a narrow therapeutic index once wild-type RHO suppression is measured directly in primates.