The Case of the Missing Mouse Pax6 Neuron: A Cross-Species Mix-Up Rewrites Brain Cell Taxonomy

For decades, PAX6 has been one of the most celebrated names in developmental biology. The transcription factor first made its mark as the gene behind a famous eyeless mouse mutant, and its structure and function have since been traced from insects to mammals, orchestrating early cortical patterning and neurogenesis wherever it appears. Given that deep evolutionary pedigree, it seemed almost self-evident that when adult primate single-cell brain atlases described a distinct interneuron subclass bearing the Pax6 name, the same cell type should exist in the mouse, the most thoroughly sampled species in all of neuroscience. Yet it does not. Mouse atlases, despite their extraordinary resolution, have never carried an annotated Pax6 interneuron subclass. That conspicuous absence has now been resolved, and the answer is turning a long-standing annotation puzzle into a lesson about how cell types should be named, compared, and understood across species.

A team led by Jonathan M. Werner, Hamsini Suresh, Leon French, and Jesse Gillis, publishing in iScience, set out to determine whether the primate Pax6 interneuron subclass is a genuine primate-specific innovation or simply a mouse cell type hiding in plain sight under a different label. Their strategy was a sweeping meta-analysis of caudal ganglionic eminence (CGE) and medial ganglionic eminence (MGE) interneuron subclasses, drawing on data from the Allen Institute, the Broad Institute, Yale, and UT Southwestern. The collection spanned human, chimpanzee, gorilla, macaque, and marmoset alongside mouse, covering both 10x and Smart-seq platforms, single-cell and single-nucleus preparations, and 18 mouse and 3 primate brain regions. Using replicability-based cross-dataset alignment, the researchers re-evaluated subclass identity and then tested the resulting homologous groups against electrophysiological and spatial data.

The first clue came from an unexpected direction: the gene SNCG, the namesake of the Sncg interneuron subclass. In mouse, SNCG is a strong and reliable marker for the Sncg subclass. In primates, however, the team found that SNCG shows positive differential expression in MGE-derived rather than CGE-derived subclasses, a striking inversion confirmed independently in both the Yale and UT Southwestern primate datasets. Meanwhile, the mouse Sncg subclass showed mild positive differential expression of PAX6 itself. Taken together, these patterns hinted not at a lost mouse lineage but at a misannotation: the labels themselves may have drifted across species, even as the underlying cell types remained conserved.

To test that idea rigorously, the researchers examined how many marker genes are actually needed to anchor a subclass identity across species. Individual markers, they found, can be treacherous. Top single markers for subclasses such as Pvalb, Lamp5, Lamp5 Lhx6, and Sncg showed weak or inconsistent differential expression in their expected mouse counterparts. But when markers were aggregated, the picture changed dramatically: sets of just ten primate subclass markers in aggregate reliably recovered the expected trends in mouse. Most tellingly, primate Pax6 markers were more enriched in the mouse Sncg subclass than primate Sncg markers were, a result that directly supports a cross-species misannotation rather than a true loss of Pax6-like interneurons in the rodent brain.

The decisive step was a one-vs-best MetaNeighbor analysis, a stringent test that maps each mouse cluster to its best and next-best primate subclass matches and quantifies how well the cluster distinguishes its top match. The result was unambiguous: robust one-to-one matches emerged between mouse clusters and all nine primate interneuron subclasses, including Pax6. These homologous matches were supported by all eight mouse datasets and spanned all 18 brain regions examined. When the researchers compared the original mouse labels to their primate-aligned identities, the story crystallized: nearly all previously annotated mouse Sncg clusters actually align with the primate Pax6 subclass, along with subsets of Lamp5 and Vip clusters. Conversely, the mouse clusters most similar to the primate Sncg subclass were largely annotated as mouse Vip clusters. The homologous mouse Pax6 subclass, in other words, is largely composed of what had been called the mouse Sncg subclass, plus a PAX6-expressing subset of Lamp5 cells.

Crucially, the reassignment is not merely a relabeling exercise; it recovers real transcriptional structure. Under the homologous annotations, Pax6-labeled mouse clusters form a coherent, well-separated group, and the Vip and Lamp5 clusters that remap to Pax6 are more similar to each other than to the remaining Vip and Lamp5 clusters. The same held true for the homologous Sncg subclass, which forms a distinct outgroup among Vip clusters. The team also identified new marker genes that better capture subclass identity across species: FREM1 emerged as a high-performing marker for the homologous mouse Pax6 subclass, while SORCS3, MAML3, and CDH24 proved to be conserved Pax6 markers across primates and the combined primate-mouse analysis. For the Sncg subclass, CXCL14 was identified as the optimal cross-species marker, though its lack of specificity, showing strong signal for the Pax6 subclass as well, underscores how poorly conserved Sncg-defining markers really are.

The researchers then asked whether the reassigned Pax6 cells show distinctive properties beyond their transcriptomes. Using a large mouse Patch-seq dataset, they found that homologous labels explain slightly more electrophysiological variance than the original labels, though the improvement is modest compared to the near-global increase in explained gene-expression variance. Electrophysiological similarity alone collapsed onto a Vip-Lamp5 continuum, failing to cleanly organize Pax6, Sncg, or Lamp5-Lhx6 cells, a reminder that transcriptomic measurements currently outresolve electrophysiology for subclass-level distinctions. Still, one consistent signature emerged: Pax6 cells were distinguished from all other CGE subclasses by repolarization-related features, specifically lower rheobase spike width and a lower rheobase upstroke-downstroke ratio, indicating a slightly faster firing pattern. Spatial analyses of a whole-brain mouse atlas added further context, showing that MGE subclasses preferentially localize with deep-layer cell types and CGE subclasses with superficial-layer types, with both lineages including cortical and hippocampal clusters, though no subclass showed a sharply specific spatial niche.

Perhaps the most provocative twist of the study concerns the Sncg subclass itself. While Pax6 turned out to be conserved and mislabeled, Sncg showed the opposite profile: highly replicable across primate datasets from Yale and UT Southwestern, yet weakly matched in mouse, where its best homolog is a Vip outgroup. The pattern extends beyond rodents. In tree shrew, an evolutionary intermediate between rodents and primates, interneurons showed clear enrichment for primate Vip and Pax6 markers but only mouse, not primate, Sncg markers. Rat cortex told the same story, with the Sncg homolog corresponding to a Vip outgroup. Ortholog and paralog analyses added an evolutionary mechanism: human Vip outgroup clusters switch from expressing more Vip orthologs in non-human primates to more Sncg orthologs in non-primates, while human Vip and Sncg clusters converge on Vip paralogs relative to mouse. The authors propose that the primate Sncg subclass evolved from an ancestral Vip outgroup precursor, making it a primate-specific elaboration rather than a conserved mammalian subtype, a hypothesis that assessments of more distant species will need to confirm.

The broader message reaches well beyond one misplaced label. Cellular identity, the authors argue, is distributed across the transcriptome, and small marker gene sets that work well within a species can degrade rapidly with evolutionary distance. Very few genes maintain both strong signal-to-noise and broad coverage across multiple mammalian species, meaning individual genes can shift specificity without disrupting the broader transcriptional program that preserves a cell type. Aggregate transcriptional similarity, benchmarked by replicability across technologies, datasets, brain regions, and above all species, offers a far more stable foundation for defining and comparing cell types. The study also highlights a humbling gap: neither electrophysiology nor spatial mapping currently recapitulates the fine structure of transcriptomically defined CGE subclasses, and the authors argue that internal transcriptomic consistency should be the first benchmark for recognizing a cell type, with other modalities layered on afterward. As single-cell atlases balloon in scale, the misplaced mouse Pax6 interneuron stands as a vivid warning that names can mislead, and that the deepest truths about brain cell diversity may only emerge when species are allowed to speak to one another through their full transcriptomes.

Subject of Research: Cross-species transcriptomic reassignment of the Pax6 and Sncg cortical interneuron subclasses in mouse and primates

Article Title: The misplaced mouse Pax6 interneuron subclass: A cross-species transcriptomic reassignment

Article References: Werner, J. M., Suresh, H., French, L., & Gillis, J. (2026). The misplaced mouse Pax6 interneuron subclass: A cross-species transcriptomic reassignment. iScience, 29(10), Article 117632. https://doi.org/10.1016/j.isci.2026.117632

Image Credits: AI Generated

DOI: 10.1016/j.isci.2026.117632

Keywords: Pax6, interneurons, single-cell transcriptomics, cross-species comparison, cell-type taxonomy, Sncg, caudal ganglionic eminence, marker genes, brain evolution, Patch-seq, spatial transcriptomics, mouse brain atlas