Idea 04/Bio & Medicine/Platform proposal/2 min
Splice-Code Delivery Barcodes
Lipid nanoparticles are aimed at organs. Isoform-resolved spatial atlases let them be aimed at cell states.
If every tissue has a splice signature, then delivery should be indexed to that signature rather than to the organ, turning nanoparticle targeting from a screening lottery into a lookup problem.
01What changed
- Whole-transcriptome, isoform-resolved spatial imaging of single cells in tissue arrived in 2026, exposing transcript variants that bulk or 3-prime spatial methods collapse away.1Cell, 2026
- Structural spatial transcriptomics went further, resolving RNA conformational states during active transcription in intact tissue, which adds a second axis of state information beyond abundance.2JBC abstract, 2026
02The gap
Nanoparticle targeting is still discovered empirically: make a library, inject it, see which organ lights up. The address space is organ-level, even though the biology that determines uptake is cell-state-level. Nobody has built the map from splice signature to particle.
03The idea
Build a two-sided index. On one side, take an isoform-resolved spatial atlas and cluster cells by their splice signature rather than by marker genes. On the other, take an LNP library with varied ionisable lipid, helper lipid and targeting ligand, and measure uptake per cell in the same tissue section.
The first product is a lookup table: for each of, say, thirty isoform-defined populations, the particle formulation with the best uptake ratio against the nearest off-target population. The second product is the inverse map, a predicted formulation for a population that was never profiled, tested prospectively.
The scientific claim worth making is that splice signatures predict uptake better than surface marker panels. That is plausible because uptake depends on endocytic route, and endocytic machinery is differentially spliced across states, for instance in hepatocyte zonation and in tumour versus regenerating tissue.
Start with liver, where zonation gives a clean gradient of isoforms and where LNP biology is best understood, then move to tumour and inflamed tissue. A negative result is nearly as useful: if splice signature does not beat organ-level targeting, that closes a door the field is currently walking through.
04Why now
Isoform-resolved spatial atlases and large LNP libraries became available at the same moment, and neither community reads the other's literature. The join is cheap and has not been attempted.
05The decisive test
Do this
Profile three isoform-defined populations in liver and tumour, screen 60 LNP formulations against them in situ, and test the inverse map prospectively on a held-out population.
Success looks like
At least a five-fold enrichment for three populations over the matched organ-level formulation, and the prospective prediction lands in the top decile of the library.
Kills it
If uptake is explained by surface markers with no residual splice signal, the splice-code hypothesis is wrong and delivery stays marker-based.
06What would break it
- Spatial profiling perturbs tissue and may shift uptake phenotypes.
- Species differences in endocytic splicing could make murine maps non-transferable.
- Isoform calls from short reads are noisy without long-read validation.
07If it works
Delivery becomes designable per cell state, which is the difference between a drug that reaches a tissue and a drug that reaches a cell.