Perturbed population states: neuron loss and the recovery of behavioral performance

Brain networks have the remarkable ability to adapt to neuron loss; however, little is known about how population spiking activity is affected quantitatively and even less is known about how populations adapt to restore behavior. Here, we analyzed the impact of neuron loss on population correlation structure using numerical simulations, a spiking neural network model, and in vivo spiking data from a lesion technique in which focal neuron loss is created through chronically implanted electrode arrays in awake-behaving animals. This easily-repeated perturbation allows neuroscientists to track the affected neuron population, before and after injury, establishing causal links between models of recorded activity and skilled task behaviors. The extent of perturbation to the largest 10 eigenvalues of the population spike count covariance matrix was significantly correlated with the extent of behavioral impairment, both in the model and in vivo. Despite the loss of individual neurons, correlation structure in the population returned alongside recovery of reaching behavior. These same correlations also provided a way to estimate neuron loss, using only small recorded sub-samples of the true underlying population. The results are anticipated to foster discussion among both systems and computational neuroscientists that are interested in population state models of cortex, and those interested in reverse engineering neural networks by building upon null network models to produce recovery patterns observed in vivo.