Fig. 6
Modulation of cortical inhibition could underlie population dynamics. a We simulated the activity of ~1 mm2 of the somatosensory cortex as 77,169 leaky integrate-and-fire neurons with biologically realistic cell-type-specific connectivity (~0.3 billion synapses)32,33. The model consists of four cortical layers (2/3, 4, 5, and 6), each with a population of excitatory (“E”, orange pyramids) and inhibitory (“I”, teal circles) neurons. A population of simulated thalamic neurons provides excitatory synaptic input to layers 4 and 6 to simulate touch. This graphical representation of the model displays the major excitatory and inhibitory projections (connection probabilities > 0.04 are drawn, full connectivity matrix in Methods). b Raster plot showing activity of a random subset (2%) of the modeled neurons, during two simulated touch trials (each row is a single neuron, orange/teal dots indicate the spike times of excitatory/inhibitory neurons, and neurons are sorted by layer, simulated touch, indicated by black bars. ‘Non-modulated’ and ‘modulated’ trials were interleaved). c Population dynamics, when we simulate neuromodulation by oxytocin release during social touch. When we do not modulate the network during touch (Vr + 0.0 mV, 1.0 × Rm, top left corner), the responses during modulated touch and non-modulated touch are the same (reset voltage of the interneuron populations, Vr, and membrane resistance of the excitatory populations, Rm, indicated outside plots; each dot indicates the touch response of 2000 single random neurons, the brown line shows the best model, and a small inserted plot indicates the Bayesian information criterion of three models fitted to the data: a ‘Bias model’ with only a bias and the slope fixed at unity; a ‘Potentiation model’ with no bias, but a free-slope parameter; a ‘Full model’ allowing for both a bias and a potentiation of responses, cf. Fig. 5a). The simulations invite three conclusions: increasing the inhibitory drive by depolarizing interneurons leads to a potentiation of touch responses, just as we observed in our data (first column: slope different from unity, no shift, cf. Fig. 5b). Increasing the input resistance of excitatory neurons simply leads to a bias in responses (first row: upward shift from unity line). When both effects are applied simultaneously, the effects “cancel out”, and the touch response is approximately normalized (diagonal: the best model again essentially falls on the unity line)
