Simple models including energy and spike constraints reproduce complex activity patterns and metabolic disruptions
Fig 5
Decrease in health can be partly compensated by homeostatic mechanisms and be invisible from the statistical properties of background activity, as shown by the behavior on an excitatory and inhibitory population with N = 1000 neurons in the asynchronous irregular (AI) state.
A. For such a network, changes in the neuronal health, modeled by a decrease in the α parameter, do not appear in the background activity of the raster (non-grayed areas), where the activity of both excitatory (red circles) and inhibitory (gray triangles) neurons remain very similar. To see the actual consequences of the decrease in health, one must look at the response of the network to an additional input, which is shown in the grayed areas on panel A. In response to a threefold increase in the rate of Poisson input between 1400 and 1650 ms, the activity of 100 excitatory neurons (marked by the orange area) progressively switches from continuous tonic firing (top) to well-separated bursts (bottom). B. More quantitative analyses also confirm that the background activity remains close to Poissonian, with coefficients of variation (CVs) around 0.7–0.8, and asynchronous, with an average cross-correlation (CC) smaller than . C. The distributions of firing rates over 5 seconds remain almost identical and centered around 2 Hz; dotted and dashed lines respectively denote the quartiles and medians.