Synthetic fusion promoters broaden rod-and-cone targeting in human retinal organoids

What the work claims

Tran et al. claim that promoter fusion is a faster and more controllable route to pan-photoreceptor gene-therapy vectors than de novo synthetic-promoter screens or black-box machine-learning designs. They constructed two fusion promoters, Pikali (PR1.7 upstream of GRK1) and Nocchu (GRK1 upstream of PR1.7), each roughly 2 kb. In human iPSC-derived retinal organoids, Pikali transduced about 45% of all photoreceptors and 55% of cones, while Nocchu transduced about 30% of photoreceptors and 35% of cones. Both outperformed the parental GRK1 and PR1.7 promoters in balancing coverage and expression level, and Pikali achieved significantly higher reporter mRNA than GRK1.¹

How it works

The starting promoters are PR1.7, a 1.7 kb truncated human L-cone opsin promoter that drives strong, cone-biased expression, and GRK1, a roughly 300 bp rhodopsin-kinase promoter that is rod-biased and small enough to preserve AAV cargo space. The authors reasoned that concatenating the two complete promoters would let each contribute its own transcription-factor binding landscape, producing a single regulatory element that recruits both rod and cone transcriptional machinery. Pikali places PR1.7 first; Nocchu reverses the order. Both constructs are packaged into AAV2-7m8 vectors carrying enhanced GFP as a reporter.

The validation cascade moves from mouse retina to human retinal organoids to non-human primate retinal explants. In wild-type mouse retinas after subretinal injection, GRK1 expression is restricted to rods, while PR1.7 is bright in cones with occasional rod signal. The two fusion promoters produce broader, more mixed photoreceptor expression. In human iPSC-derived retinal organoids infected at day 100 and analysed at day 150, the relative performance becomes clearer. Flow cytometry using a CRX-driven H2B-mCherry reporter to mark all photoreceptors shows Pikali at roughly 45% GFP-positive photoreceptors, Nocchu at about 30%, GRK1 at just over 20%, and PR1.7 at only a very small fraction. When cones are separately gated with an ARR3-mCherry reporter, Pikali reaches about 55% of cones, whereas Nocchu and GRK1 are both near 35% and PR1.7 is only about 2% despite cones making up more than 60% of photoreceptors in these organoids.

RT-ddPCR measurements of GFP mRNA normalised to GAPDH confirm the visual ranking: Pikali highest, Nocchu and PR1.7 comparable and slightly lower, GRK1 weakest, with a statistically significant difference between Pikali and GRK1. In macaque retinal explants maintained ex vivo for two weeks, overall transduction is low, but the promoter-specific patterns mirror those seen in organoids, with Pikali and Nocchu showing broader photoreceptor expression than the parental elements. The authors also used a fine-tuned nucleotide-transformer classifier to show that the fused sequences retain predicted promoter activity across both parental domains, offering a computational rationale for the design.¹

Where a skeptic should push

The headline result is convincing as a promoter-engineering proof of concept, but the work has not yet shown therapeutic efficacy. All expression data use GFP as a reporter, not a therapeutic transgene. A promoter that efficiently drives GFP may not drive a larger, more complex coding sequence with the same fidelity, especially given AAV's roughly 4.7 kb packaging limit. The fusion promoters themselves are about 2 kb, leaving limited room for a coding sequence and regulatory elements.

The human retinal organoid model is also a double-edged benchmark. These organoids are cone-rich; more than 60% of photoreceptors are cones at the infection stage. That composition flatters cone-biased or pan-photoreceptor promoters and may not reflect the adult human retina, where rods dominate. The non-human primate explant data are valuable because they come from a rod-dominated, mature retina, but the explants suffer from low overall transduction and culture-induced cone vulnerability, so the numbers are descriptive rather than quantitative.

Statistical power is modest. Flow cytometry quantification is based on three independent experiments. The study does not report dose-response curves, toxicity, or expression stability over longer periods. Finally, the mechanisms that make Pikali outperform Nocchu remain largely inferential: the orientation matters, but the precise cis-regulatory logic is not dissected experimentally.

Implication for retinal organoid gene-therapy design

For retinal gene therapy, the central lesson is that human iPSC-derived retinal organoids can reveal promoter behaviours that differ from mouse retina and that matter for clinical translation. GRK1 looks rod-specific in normal mice but is weak and cone-permissive in human organoids; PR1.7 is strong per cone but reaches very few cells. A simple fusion strategy corrects both problems, producing compact promoters that are compatible with AAV and perform reasonably well across photoreceptor subtypes. This positions organoids as an essential filter between small-animal promoter validation and expensive non-human primate or clinical studies.

The opportunity is to treat promoter design as a modular engineering problem. Instead of screening millions of synthetic sequences or trusting AI-generated motifs, teams can combine validated regulatory elements in defined orientations and use organoids to rank them. That approach is especially attractive for mid- to late-stage retinal degeneration, where rods are already lost and the remaining cones may express lower levels of canonical rod promoters. A promoter that retains activity in both cell types could rescue a broader clinical window.

The threat is overfitting to the organoid. Retinal organoids are immature, cone-heavy, and avascular. A promoter that scores well in organoids may lose activity or specificity in the adult, rod-rich, fully vascularised human retina. The study itself shows that macaque explants behave differently. If sponsors skip non-human primate validation, they risk advancing a vector that fails in patients. Another risk is cargo size: every base pair spent on the promoter is base pair not spent on the therapeutic gene, regulatory untranslated regions, or enhancers. A 2 kb promoter is viable for AAV but tight for larger transgenes.

The bottom line for drug-discovery teams is that retinal organoids are now good enough to rank compact, human-relevant promoters, but the final selection still requires a disease-relevant functional readout and confirmation in a mature primate retina.

The bottom line

Pikali and Nocchu are a practical addition to the retinal gene-therapy toolkit. They demonstrate that fusing two validated promoters can outperform either parent in balancing rod-and-cone coverage in human retinal organoids. What would cement their value is packaging a therapeutic transgene, measuring functional protein levels, and confirming durable expression in a rod-dominant non-human primate model. What would limit them is evidence that the fusion architecture imposes a cargo-size or expression-stability penalty that offsets the breadth advantage.