Cell-type-specific inhibitory circuitry from a connectomic census of mouse visual cortex

Key results

1. 01

   ShownInterneurons specialised for inhibiting other interneurons (InhTCs) form two groups. Most target dendrite-targeting (somatostatin-like) cells, as expected. A second group, which the authors report as new, selectively targets perisomatic (basket-like) cells, with larger synapses and more synapses per connection.

   Of 29 InhTCs, 21 sent most inhibitory-directed output to distal-dendrite-targeting cells (DistTCs; mean 74%) and 8 to perisomatic-targeting cells (PeriTCs; mean 82%). The 8 contacted 56/58 PeriTCs (mean 10.5 synapses per target). Their PeriTC synapses were 69% larger (median) than those onto other subclasses, with 3.1 vs 1.5 synapses per connection (p = 1.1 × 10^-5, t-test). · Inhibition of inhibitory neurons; B. Basket cell disinhibition

2. 02

   ShownClustering excitatory neurons on synaptic input features as well as dendritic shape gave morphological types (M-types). These separate upper from lower layer 2/3 and distinguish visually similar cells, such as two layer 2 types that differ markedly in the number and density of synaptic inputs.

   Consensus clustering of 29 features gave 17 M-types; 142/143 (99%) expert-labelled layer 6 cortico-thalamic (CT) cells fell into L6tall types. L2a vs L2b: 4758 vs 2649 synaptic inputs, 1.04 vs 0.72 syn/µm median density, 4532 vs 3510 µm dendritic length. Layer 5 near-projecting (L5NP) cells had the fewest dendritic inputs and the lowest input density. · Dendritic synaptic input properties define…; Excitatory feature extraction and clustering

3. 03

   ShownInhibition is spread unevenly across M-types, and much of the difference tracks overall synaptic input. Somatic and dendritic inputs rise together, cells that receive more perisomatic-targeting input tend to receive more dendrite-targeting input, and intermingled L2a and L2b cells get very different amounts.

   Of 70,884 inhibitory-to-excitatory column synapses, PeriTCs and DistTCs averaged 581 and 596 per cell. Somatic vs dendritic input across M-types: r = 0.96, p = 5 × 10^-10. PeriTC and DistTC input correlated positively within 12/18 M-types. L2b cells received 37.7 ± 0.27 intracolumnar inhibitory synapses vs 94.8 ± 0.58 for L2a. · Coordination of inhibition across excitatory…; Figure legends

4. 04

   SuggestedClustering interneurons by how their output divides among excitatory M-types gave 18 motif groups. Most groups mix subclasses, so perisomatic and dendrite-targeting cells jointly inhibit the same target types, often with sublaminar precision.

   15/18 groups (156/163 cells) held at least two subclasses. Within-column output budgets correlated with dataset-wide budgets based on predicted M-types (Pearson r = 0.90), although interneurons made more than 4 times as many synapses outside the column. In layers 2–4 each M-type received strong inhibition from 2-3 groups; most layer 5 groups targeted a single M-type. · Cellular contributions of inhibition

5. 05

   SuggestedSpecific targeting comes from both precise axon placement and synaptic selectivity, weighted differently by subclass. A layer-2-specific group is largely explained by where its axons overlap targets, whereas dendrite-targeting cells need selectivity beyond a depth- and compartment-matched null model.

   In 17/18 motif groups the Selectivity Index (observed vs shuffled synapses) was consistently above or below 1 for some M-types, yet layer-2-specific Group 1 showed no consistent positive selectivity. PeriTCs had the lowest Kullback-Leibler divergence from the null model; perisomatic inputs were more spatially segregated across M-types than distal/apical ones (p = 3.0 × 10^-19, Mann-Whitney U). · Synaptic selectivity; E. Peters’ rule and White’s exceptions; Figure legends