A monitoring process modulates the strength of the input (λ) to each group of neurons simulating different trial history conditions: λ increases its value as the number of Stop trials preceding a Go trial increases and decreases its value as the number of Go trials preceding a Go trial increases (Figure 4B). We observe that the model reproduces the same relationship between the probability of failure and SSDs as observed

during the countermanding task, i.e., the probability of failing in the Stop trials increases as the SSD increases (Mirabella et al., 2006) (Figure S3A). To compute decision times in the simulations, we considered that the decision process was terminated when the difference in activity between Go and Stop pools was above a fixed threshold (Roxin and Ledberg, this website 2008). The RT was calculated by adding 150 ms to the decision time, consistent

with the peak in FR Selleckchem Epacadostat observed 150 ms before movement onset in the physiological data (Figure 2A). The mean and SD of RT obtained from the simulations (Figures 4C and S3B) exhibit the same trend as observed in the physiology of PMd (Figures 2C and S2C): the mean and SD of RT in a Go trial are longer/shorter as the number of preceding Stop/Go trials increase. Consistent with our analysis of the physiological data, the different simulated trial history conditions have a similar impact on the variability of the Go pool response (Figure 4D). those This impact of the monitoring signal λ on RT and VarCE can be intuitively understood in terms of the competition between the two neuronal pools Stop and Go through mutual inhibition (Figure 4A). The model is tuned such that the firing rate of the Go pool is not affected by this neuronal competition (Figure S3C), as observed in the response of the neurons we have analyzed (Figure S2E). We observe that, given these assumptions that reflect the physiological properties of PMd, the addition of the monitoring signal leads to the modulation of the effect that

the Stop pool has on the dynamics of the overall network, leading to a change in the mean RT. In addition, when the influence of the Stop pool on the dynamics is increased, the intrinsic noise of the system starts to have a larger impact on the performance and dynamics of the network, resulting in an increase in VarCE and RT variability. Indeed, it has been demonstrated that the neural response variability changes with the strength of the input to this model, due to a shift in the distance from the working point of the system to the bifurcation point (Deco and Hugues, 2012; Roxin and Ledberg, 2008). Here we exploit this effect through the monitoring signal. Hence, perceptual input defines the mean rate, while the history-dependent monitoring signal defines a modulation around this rate expressed in VarCE.