Attentional modulation of intrinsic timescales in visual cortex and its underlying network mechanisms

Neurons process information on different timescales across the neocortex that are related to the dynamics of their intrinsic activity fluctuations [1]. While areal differences in intrinsic timescales reflect the functional specialization of cortical areas, it is unknown whether the timescales can adjust rapidly and selectively to the demands of a cognitive task. We studied the timescales of spiking activity recorded from local neural populations within cortical columns of primate area V4 during two spatial attention tasks and a fixation task. In all tasks, the autocorrelation of intrinsic activity deviated from a single exponential decay, commonly assumed for estimating the timescale, and instead, indicated a multiplicity of timescales. We characterized these timescales using a precise Bayesian estimation method based on Approximate Bayesian Computations (ABC) [2] (Fig 1A). This method confirmed that at least two distinct (fast and slow) timescales were present in the local population dynamics (Fig 1B). The slow timescale increased on trials when monkeys attended to the receptive fields of recorded neurons (att-in), yet the fast timescale did not change (Fig 1C). To identify the possible mechanisms underlying the multiplicity of timescales and their flexible modulation, we developed a recurrent network model with binary units representing cortical minicolumns [3] and local spatial connectivity among them (Fig 1D). The activity of model minicolumns exhibits two distinct timescales similar to V4 data. The fast timescale arises from vertical interactions within a minicolumn and the slow timescale is induced by horizontal interactions among minicolumns (Fig 1E, top). These timescales depend on the network topology, and the slow timescale vanishes from the local dynamics of networks with random connectivity (Fig 1E, bottom). We derived an analytical relationship between the timescales and connectivity parameters, to identify model parameters capturing the timescales of V4 data. The model indicates that the modulation of slow timescales during attention arises from a slight increase in the efficacy of network recurrent interactions (Fig 1F), that can be explained by top-down inputs and neuromodulatory mechanisms during attention [4,5]. Our results revealed that targeted neural populations integrate information over variable timescales following the demands of a cognitive task and link the network structure, functional brain dynamics, and flexible behavior