11,535 Publications · 65,411 Citations

Question: is the capillary network of cortex placed to serve the neurons and synapses that use the most energy? Approach: we put a public blood-vessel segmentation (TriSAM) of the MICrONS cubic millimetre of mouse visual cortex in the same frame as 93,778 typed nuclei, 337 million detected synapses and the in vivo two-photon scores of about 13,000 co-registered neurons. Pericyte nuclei sat on the vessels (99% within 2 um), which confirms the frame. Results: microglia, astrocytes, OPCs and oligodendrocytes sat nearer to vessels than local random placement predicts (by 2.3, 1.0, 0.6 and 0.4 um). All 11 neuron types sat at chance (-0.28 to +0.20 um); basket cells sat 0.36 um farther than pyramidal cells in the same 50 um block. With somata and volume edges removed, neuropil synapse density stayed within -2.4% to +0.8% of the depth mean from 4 to 30 um from the nearest vessel, as in a shifted-map control. Capillary volume did not follow synapse density across depth (rho = -0.17) or between blocks at one depth (rho = 0.042; shift null -0.003 +- 0.028). Excitatory neurons with more input synapses sat slightly nearer to vessels (-0.32 um per 1.74-fold more inputs; 95% CI of the standardized slope -0.078 to -0.043); inhibitory neurons did not. Vessel distance did not predict response reliability, model accuracy or orientation selectivity (15 tests, |slope| <= 0.052 SD per SD, none after Holm). Exploratory: vessel volume in the light cone above a neuron lowered its measured digital-twin accuracy (top 10%: -0.031 against the depth mean, about 11% lower), but vessel volume below did not (-0.006), which fits an optical shadow and is a confound for cell selection in functional connectomics. Limitations: one animal and area, fixed tissue, a zero-shot vessel mask at 2 um voxels, nucleus position as a soma proxy, automated synapse counts, and calcium scores rather than energy use. The package has code, per-cell distance tables, all result tables and figures.