Figure 9: Structure and activity patterns of the neural network model.

(a) Schematic representation of the network wiring process. Left: Initial random connectivity between each Martinotti cell (green circle) and the pyramidal cell population (blue triangles). Middle panels: Establishment of final connectivity by selecting and strengthening the main connections (for simplicity only 1 main connection is shown; inhibitory strength λ=0). Connections with neighbouring units are removed (pink area), reciprocal excitatory connections are added (blue). Right: resulting inhibitory effect of each pyramidal cell on the rest of the population. Phase selective suppression of activity is mediated by collective Martinotti-cell activity. (b–e) The dynamic properties at the excitatory PC-to-MC synapses are governed by two parameters, b1, controlling the persistence of synaptic modulation, and b2, controlling the strength and direction of the modulation (cf. Methods and Supplementary Fig. 5). To examine the importance of facilitating synapse dynamics, b1 and b2 are chosen to model either facilitating ((b,c) b1=0.00025; b2=0.5) or depressing excitatory synapse dynamics ((d,e) b1=0.04; b2=−0.5). (b) Spontaneous activity profile formation in PC. In blue, example of raw activity over time of the pyramidal cell population (arranged according to directional selectivity) in the absence of selective external input. Fed with white noise input, the system evolves over time from an initial random configuration (left) towards a stable configuration of concentrated activity (‘bump’ formation, right). In green, same as above, but for simultaneous activity of MC (positions on the circle are randomly assigned). (c) Persistence of directional tuning after external input removal. Starting from random activity (left), PC around a given direction are transiently provided with an additional external input (red arrows). The final state of the network is then observed after removing this additional input (grey background). No directional selectivity appears in MC. (d) Replacing facilitating by depressing dynamics at excitatory synapses impairs the spontaneous development of a coherent activity bump in the network. (e) Depressing excitatory synapse dynamics do not support the persistence of directional activity correlated to an initial external input.