In four mice, the published transcriptomic axis of V1 interneurons predicted state modulation mainly between subclasses and types, and added no held-out-mouse prediction beyond subclass labels

Bugeon et al. (2022, Nature) reported that one transcriptomic axis (tPC1) of mouse V1 inhibitory neurons largely predicts how brain state modulates each cell, also within subclasses, and that tPC1 aligns with Patch-seq physiology, layer 1 axon fraction and cholinergic receptor expression. We re-tested these claims with the released data (figshare v2: 1,065 coppaFISH-typed cells, 4 mice, 17 sessions) and the published source data. Our reproduction matched the published subtype state modulation (r = 0.994), the tPC1 gene loadings (r = 0.9999) and the within-subclass ANCOVA (F = 15.7; published 14.5). Controls for soma depth, transcript detection and activity level did not remove the within-subclass link (partial r 0.13 to 0.14, 963 cells). But most of the link was between groups: partial r was 0.41 with only session terms, 0.13 within subclass, 0.055 within type and 0.052 within subtype (95% intervals for type and subtype included zero). The within-subclass slope was different between subclasses (p = 0.0004): positive in Lamp5 (r = 0.24) and Sst (0.43), near zero in Pvalb (0.07), and negative in Vip (-0.17). It was near zero in the mouse with the most cells (n = 406). In leave-one-mouse-out tests, tPC1 alone gave a held-out R2 of 0.149, five subclass labels 0.159, subclass labels plus tPC1 0.153, and 11 types 0.199; the test-retest reliability of cell means was about 0.65. The published within-type continua repeated in each mouse with enough cells and after a depth control (Lamp5 neurogliaform subtype index r = -0.40, published -0.39; Ndnf, Cck; Sst-Calb2 index r = 0.42). At the subtype level, 15 of 24 Patch-seq and receptor correlates had q < 0.05 across 35 subtypes, but within subclasses only spike adaptation (r = 0.54) and Chrm3 expression (r = 0.62) stayed. Layer 1 axon fraction did not predict subtype state modulation (r = -0.19 across, 0.25 within subclass; 26 subtypes). An axis from library-normalized log counts, which the paper did not use, kept a small within-type link (r = 0.11); this result is exploratory. Limits: four mice, the authors' fixed deconvolution and neuropil settings, and Patch-seq and receptor values that are subtype means from other animals.

cortical interneurons · brain state · locomotion · transcriptomic cell types · coppaFISH · Patch-seq · mouse V1 · replication · leave-one-mouse-out · Lamp5 · Vip · Sst · Pvalb · cholinergic receptors · axon morphology · structure-function