10.12.15: Alfonso Renart

Recurrent connections in cortical circuits appear to reduce the dimensionality of temporal fluctuations in population activity (Okun et al., Nature, 2015). However, the state-dependence of this phenomenon is not clear. We explored this issue in population recordings of spontaneous activity from the auditory and somatosensory cortices of Urethane anaesthetised rats, focusing on periods of cortical activation, which are also characteristic of attentive wakefulness. PCA reveals that the correlation structure of the population is low-dimensional, with a gap in the amount of explained correlation between the first and subsequent principal components. Although low-dimensional variability has been found previously in cortical circuits, the correlation structure in our recordings is, unlike previous findings, not related to fluctuations in the global firing rate of the network. Indeed, the distribution of loadings onto the high-variance mode (PC₁) across the population is wide and roughly centered at zero. Thus, the dynamics of the population is ‘competitive’: neurons with high positive or high negative loadings are negatively correlated with each other. Because PC₁ is approximately orthogonal to the mean population activity, the population-averaged correlation is close to zero, as we showed previously (Renart et al., Science, 2010). What mechanisms could underly this dynamical behaviour? Competitive amplification has typically been modeled in a 3-population network with two excitatory populations interacting directly and through mutual inhibition. Competition arises when the network operates close to a pitchfork bifurcation. We developed a different mechanism using non-normal amplification in the same 3-population network. We show that a model using non-normal amplification naturally generates idiosyncratic but robust features of the real data, such as positive correlations within the two populations of clearly different magnitudes and a time delay in the negative correlations between them. Our work shows that the low-dimensional dynamics of cortical circuits is strongly reconfigured across different brain-states, and suggests that non-normal amplification drives competitive spontaneous temporal fluctuations during states of cortical activation.