Skip to main content

Network dynamics of auditory cortex and the impact of correlations on the encoding of sensory information

Final Report Summary - NETDYNCORTEX (Network dynamics of auditory cortex and the impact of correlations on the encoding of sensory information.)


During this project we focused on the following research lines:

1. We have first characterized spontaneous spiking activity in the auditory cortex of rats under urethane anesthesia. We performed a quantitative characterization of the statistics of the durations of up and down periods observed in epochs when the brain underwent slow waves similar to those observed during slow wave sleep. The analysis revealed that these periods are more irregular than previously thought challenging the idea that up and down states reflect and slow oscillation in the activity of the cortex. Moreover, we found no traces of a fatigue mechanism (e.g. spike frequency adaptation) during up periods putting into question the role of this type of processes, commonly thought underlie the generation of these spontaneous pattern. We proposed a novel computational network model for how up and down dynamics are generated based on the existence of two stable attractors, an active and a silent attractor, and fluctuations in the activity causing stochastic transitions between them. These results shed light into the basic mechanisms giving rise to spontaneous brain activity and could help to elucidate its function.

2. We also investigated how slow changes in brain state, observed spontaneously under urethane anesthesia, impact the statistics spontaneous activity. In particular we quantified changes in the statistics of up and down periods and proposed a network model showing up and down alternations that could account for those changes. We found that brain state changes had a systematic effect on the average duration, irregularity and correlation and that this effect could be reproduced by slow variations in the model parameters. These findings set the mechanistic basis of the changes occurring in cortical circuits when the brain undergoes brain state transitions.

3. In a third project we investigated the origin of the so called pair-wise noise correlations between cortical neurons. These correlation refer to the fact that the variability in the spiking pattern displayed by cortical neurons is not independent but nearby cells tend to share a fraction of this variability (i.e. their activity co-caries positively). We used population recordings from rat auditory cortex during spontaneous and stimulus evoked conditions. We found that correlation were mostly due to the existence of brief periods during which all neurons stopped firing. This was a fundamentally different explanation to previous studies which argued that correlations were mostly caused by the existence of shared inputs in the anatomy of the cortical micro-circuit. We used a simple network model showing stochastic transitions into the silence that could account for the stimulus-evoked dynamics of correlations and their dependence on brain state.

4. We conducted a study on the role of neuronal fluctuations on perceptual decision making in which we developed a computational model of a hierarchical network representing a sensory and a decision circuit. The aim of the study was to reproduced the classical motion discrimination task performed in monkeys and was able to explain the relation between single neuronal variability in sensory area MT and behavioral decisions. This relation, commonly called choice probability, exhibits a temporal time course which was inconsistent with the previous feed-forward model describing the underlying mechanisms (Shadlen et al, J. Neuroscience 1996). We developed a a hierarchical network model which shows that the non-linear dynamics of sensory integration can resolve the contradiction. The novelty of our work is that, by coupling a standard decision-making circuit (Wang, Neuron 2002) with the standard model of sensory circuitry, i.e. a balanced network (Renart et al., Science 2010), we could reproduce the basic phenomenology of the dynamics of choice probability and sensory integration. Moreover, the model allowed us to dissect several sources of neuronal noise correlations, a key element to obtain choice probability (Haefner et al, Nat. Neurosci. 2013). We found that anatomically shared inputs, a mechanism previously thought to play a crucial role, does not cause significant choice probability, whereas a different mechanism thought to play a marginal role (stimulus fluctuations) is in fact a major source of correlations and choice probability. The model allowed us to derive specific predictions that we validated in single-unit and paired recordings from monkey area MT (from classical experiments performed by Newsome, Movshon and collaborators; Britten et al, Vis. Neurosci. 1996; Zohary et al, Nature 1994), and in newly obtained psychophysical data.

5. Finally we developed an auditory discrimination task in rats to investigate the impact of expectation on perceptual decision-making. We trained rats to learn the sensory statistics of different environments and use that information to predict future stimuli and thus enhance their ability to discriminate and ultimately maximize reward rate. We found that rats can learn and used these statistics but they are constantly varying the weight given to the prediction in their decisions based on the recent history of choices and rewards. In particular, a recent sequence of rewarded choices following the stereotypical pattern found in a given environment, increases the belief of animals to predict future stimuli following that same pattern. In contrast, error trials are followed by an apparent lack of expectation bias, as if the animal had momentarily lost confidence on the its internal statistical model of the environment.

Overall our results demonstrate that cortical circuits exhibit stochastic non-linear dynamics whose underlying mechanisms, we are only starting to understand. This neuronal stochasticity seems to correlate with behavior but the interpretation of this relation is difficult as the sources of the neuronal variability are diverse. One possible hypothesis is that part of the variability observed in sensory areas reflects the priors about the environment that the systems is constantly trying to predict. The task we have developed using rats could be a powerful tool to investigate this possibility.