The confound this design attacks is old and simple. If an isoform's abundance differs between two brain regions, that could mean individual cells adjust their splicing according to where they sit, or it could mean the two regions are stocked with different cell types that each have fixed preferences. A measurement that pools several cells cannot tell those apart. Visium's 55-micrometre spot is wider than the average cell diameter in mouse brain, so each spot behaves as a small pseudo-bulk sample that may contain several cell types [5]. Resolution here is the control that makes the comparison possible.
The arithmetic of the improvement is worth stating in the currency that matters. Against the same group's earlier 10-micrometre Spl-ISO-Seq [3], the new barcode pitch is 20-fold finer on a side and 400-fold finer in area [18]. Against a Visium spot it is 110-fold finer on a side [19]. The barcode count moves by a similar order: from 80,000 to more than 450 million is roughly 5,600-fold [20].
The region-agnostic analysis is the more interesting half of the method, because it does not inherit the anatomist's map. It found spatial patterns that region-by-region comparison missed, Ighm among them [9]. Anyone who has watched a signal vanish because it straddled two atlas boundaries will recognise why that matters.
What the evidence supports is narrower than the framing usually applied to results like this. The abstract reports that many spatial isoform signals are not driven by cell-type composition alone, which leaves room for composition to still play a partial role; no fraction is given for how many survive conditioning on cell type [10]. A finer cell-type annotation could also absorb some of the residual, since subtypes below the labels used may themselves be spatially segregated. And nothing here was perturbed. Position and isoform usage covary within a cell type; no cell was moved to see whether its splicing followed. The thing this does not tell you is which direction the arrow points, or whether a third variable such as local signalling drives both. This is also adult mouse brain [7], and Snap25 in mouse excitatory neurons is not a statement about human disease tissue.
The part most likely to be used by other labs is the software rather than the wet-lab protocol. The authors report that Spl-IsoQuant-2 and Spl-IsoFind apply to many spatial and single-cell protocols, with reproducibility between platforms including Visium HD and Stereo-seq [12], and that the chemistry runs on both PacBio and Oxford Nanopore reads [11]. That is the difference between a result and a method: a group with existing Stereo-seq or Visium HD sections can re-ask the composition question against data already on disk.
Read strictly, the resolution claim is a hardware fact and it is settled. The biology claim is a well-designed inference whose denominator is not yet visible, and that is the piece still worth having answered.
