Fig 1.
A: Schematics and quick facts about the detailed, large-scale cortical microcircuit that was used several times before to study the relationship of cortical structure and function [25, 35, 36]. B: Schematics of the assembly detection pipeline from the spiking activity of 186,665 excitatory neurons in the circuit. C: Analysis of the connectivity of cell assembly neurons. D: Derivation of assembly membership probability based on different features of structural connectivity.
Fig 2.
In vivo-like activity in silico.
A1: Raster plot of the microcircuit’s activity with 628,620 spikes from 98,059 individual neurons and the population firing rates below. The y-axis shows cortical depth. (As cortical layers do not have the same cell density, the visually densest layer is not necessarily the most active—see a2 bottom.) A2: Single cell firing rates (in excitatory and 3 classes of inhibitory cells) and layer-wise inhibitory and excitatory population firing rates in evoked (showed in a1) and spontaneous (not shown) activity. B: Top: pyramid-like overlap setup of VPM patterns, then the centers of the VPM fibers in flat map space. Bottom: raster plots of VPM fibers forming each of the patterns for the stimulus stream in a1 (i.e., from pattern A at 2000 ms to pattern J at 6500 ms). On the right: same for non-specific (POm) input.
Fig 3.
A: Population firing rate of excitatory neurons with the determined significance threshold. B1: Hierarchical clustering of the cosine similarity matrix of activation vectors of significant time bins (above threshold in A, see Methods). B2: Clustered significant time bins ordered by patterns presented. Colors indicate assembly identity and will be used systematically throughout the remainder of the manuscript. C: Number and location of neurons in each cell assembly: flat map view on top, depth-profile below. D: Jaccard similarity of cell assemblies and number of neurons participating in different number of assemblies. E: Input-output map: Input distance is calculated as the Earth mover’s distance of the VPM fiber locations, this metric is small when the activated fibers have large intersections and large otherwise (see Fig 2B and Methods). The output distance is the normalized Euclidean distance of pattern evoked time bin cluster counts (counts of different colors in the matrices above in B2, Methods) this measure is minimized if the same assemblies are active at the same time.
Fig 4.
Connectivity determines cell assembly membership.
A-B: Thalamic connectivity, C-D: recurrent connectivity (indicated with schematics from Fig 1D). A: First: Effect of thalamic innervation (from VPM and POm nuclei) on participation in cell assemblies. Solid lines indicate the mean and the shaded areas indicate 95% confidence intervals. Second: probability of membership in an exemplary middle assembly against mean common POm indegrees with respect to all assemblies. Third and fourth: nI (normalized mutual information, see Methods) of mean common thalamic indegree and assembly membership. B: Probability of membership in exemplary early (first), middle (second), and late (third) assemblies against pattern indegree with respect to all patterns. Fourth: nI of pattern indegree and assembly membership. C: Simplex counts within assemblies and random controls (same number of neurons with the same cell type distribution), where a simplex of dimension k is a motif of k + 1 neurons fully connected in feed-forward fashion. D: Probability of membership in exemplary early (first), middle (second) and late (third) assemblies against 0-indegree with respect to all assemblies. Fourth: nI of 0-indegree and assembly membership. E: First: Probability of membership in an exemplary early assembly against k-indegree with respect to the same assembly. Inset: k-indegree is given by the number of k-simplices within an assembly (orange nodes) completely innervating a given neuron (black). nI of k-indegree and assembly membership for k = 1 (second), k = 2 (third) and k = 3 (fourth).
Fig 5.
Synapse clustering coefficient determines cell assembly membership.
A: Probability of membership in exemplary early (first), middle (second), and late (third) assemblies against synapse clustering coefficient (SCC, see Methods) with respect to all assemblies. Fourth: nI (normalized mutual information, see Methods) of SCC and assembly membership. B: Combined effect of SCC and indegree (as in Fig 4d). First: Probability of membership in an exemplary early assembly, against 0-indegree with respect to the same assembly, grouped by SCC significance (see Methods). Second: Joint distribution of SCC and 0-indegree. Third: nI of SCC and assembly membership conditioned on 0-indegree. Fourth: Relation of nI and conditional nI grouped by postsynaptic early, middle, and late assemblies (i.e., rows of the matrices in A fourth, and B third). C: Simulation results for 10 selected neurons per assembly (with the highest 0-indegree and significant clustering; red arrow in B left) with modified physiological conditions. Left: correlation of spike times with the assembly. Right: Single cell firing rates.
Fig 6.
A: Jaccard similarity based hierarchical clustering of assemblies from different simulation instances. B: Spike time reliability (see Methods) of neurons at different levels of coreness (see Methods). Missing boxes indicate missing categorical data at given levels of coreness. C: Locations of neurons in the union (all instances) and core (above expected, see Methods) of exemplary (pattern A responsive) early consensus assembly. D:nI (normalized mutual information, see Methods) of connectivity features and consensus assemblies membership.
Fig 7.
Summary of nIs of all connectivity features and assembly membership.
Assemblies from single simulation on top and consensus assemblies from 10 simulations on the bottom. Only within-assembly interactions (diagonals of nI matrices) shown, except for the patterns, where for each column (postsynaptic assembly) we used the maximum value (strongest innervating pattern). Not only the colors, but the radii of the pies code for features: Red (VPM) longer one depicts common-innervation (as yellow for POm) while the shorter one direct innervation from patterns. Green: Simplex dimension increases as length decreases (longest: 0-indegree, shortest: 3-indegree). Blue: longer one is the SCC conditioned on 0-indegree, while the shorter one is the same, but unconditioned (and that is why it always overlaps with 0-indegree).