In one MICrONS mouse, the other targets of an axon predict its connections beyond pairwise response similarity; the extra signal is in the top shared response modes

Question: mouse visual cortex neurons with similar responses connect more often (like-to-like rule). A pairwise rule uses only the axon and the candidate neuron. We tested if the other targets of the same axon add information. Data: public MICrONS functional connectomics tables (Ding et al. 2025, v1): 142 proofread axons, 11,035 candidate neurons, 292,855 pairs with axon-dendrite proximity, 8,118 connections. Method: for each axon and candidate j we computed ST, the mean response similarity of j to the other targets of the axon (leave-one-out), and SN, the same for the non-target candidates. Poisson regression with the similarity of j to the axon, proximity, distance and scan controls, and fixed effects for the postsynaptic neuron, axon x layer and axon x area; standard errors clustered by both neurons; plan written before the first fit. Results: the rate ratio per SD of ST was 1.18 (95% CI 1.10-1.26) for in vivo signal correlation, 1.38 (1.25-1.52) and 1.30 (1.20-1.40) for two digital-twin measures (Holm p < 1e-6). SN had a rate ratio below 1, so the signal is the target versus non-target contrast. The effect stayed with all three axon similarity measures together (1.16), cross-scan neurons only, receptive-field controls and in all projection classes and layers. It was not larger for axons with low response reliability. Most of it is in the top 5 shared response modes: the mode part of ST had 1.21 (1.12-1.31), the residual part 1.04 (0.99-1.09). Simulations of a pure pairwise rule gave no false positives (0/24); noise in the axon response gave mean 1.04 at reliability 0.4 and 1.10 at reliability 0.2. At the pair level, the pairwise rule with cell effects explains about 90% of the extra similarity between two targets of one axon. Limitations: one animal; a replication attempt in the V1 Deep Dive mouse (640 connections) did not detect the pairwise or the co-target term (movie signal 0.92, 0.76-1.12); very unreliable or structured noise in the axon response can give part of the effect; the mode split is post hoc.

MICrONS · functional connectomics · like-to-like connectivity · shared input · higher-order connectivity · co-target similarity · electron microscopy · two-photon imaging · mouse visual cortex · V1 Deep Dive · Allen Institute · structure-function · fixed effects · response modes