Theta-activity in anterior cingulate cortex predicts task rules and their adjustments following errors

Abstract

Accomplishing even simple tasks depend on neuronal circuits to configure how incoming sensory stimuli map onto responses. Controlling these stimulus-response (SR) mapping rules relies on a cognitive control network comprising the anterior cingulate cortex (ACC). Single neurons within the ACC convey information about currently relevant SR mapping rules and signal unexpected action outcomes, which can be used to optimize behavioral choices. However, its functional significance and the mechanistic means of interaction with other nodes of the cognitive control network remain elusive and poorly understood. Here, we report that core aspects of cognitive control are encoded by rhythmic theta-band activity within neuronal circuits in the ACC. Throughout task performance, theta-activity predicted which of two SR mapping rules will be established before processing visual target information. Task-selective theta-activity emerged particularly early during those trials, which required the adjustment of SR rules following an erroneous rule representation in the preceding trial. These findings demonstrate a functional correlation of cognitive control processes and oscillatory theta-band activity in macaque ACC. Moreover, we report that spike output of a subset of cells in ACC is synchronized to predictive theta-activity, suggesting that the theta-cycle could serve as a temporal reference for coordinating local task selective computations across a larger network of frontal areas and the hippocampus to optimize and adjust the processing routes of sensory and motor circuits to achieve efficient sensory-motor control.

Fig. 1.

Task selectivity of LFP power. (A), Proportion of sites with individually statistically significant task selectivity for the pro- or antisaccade task based on a cluster-based randomization test with multiple comparison correction. The statistical map reveals that task selectivity emerges in a narrow theta-frequency band (5–10 Hz) before target stimulus onset. The white box indicates the time-frequency region of interest used for later analysis. (B) Average task selectivity in the preparatory period for sites showing significant task selective theta-activity. (C) Average task selectivity plotted separately for sites preferring the prosaccade task (Upper) and the antisaccade task (Lower).

Fig. 2.

Trial-by-trial (i.e., ROC) task prediction for power at different frequency bands, and its latency for trials relative to the change in task rule. (A) ROC values for successive time bins during the preparatory period. Shaded areas denote SEM. (B) P value evolution for the ROC values in A. Gray area shows the ≤2.998 = −log(0.05) region where trial-by-trial prediction of the task is not significant. Colors denote different frequency bands as indicated in A. (C) Latency of significant ROC prediction based on theta-power for trial sets relative to the task rule change. (D) Average TSI of theta-power for trial sets relative to the task rule change. Shaded areas show SEM.

Fig. 3.

Behavioral correlates of theta-frequency modulation for task-selective sites. (A) Average time-frequency distribution of the TSI for error trials (plotted in a color range similar to the TSI on correct trial shown in Fig. 1B). (B) Average (±SEM) TSI for correct and error trials for theta-power (5–10 Hz) in the −0.4 to 0 s before stimulus onset. Two stars denote statistical significant difference at P < 0.01. (C) Average TSI for correct trials following other correct trials (Upper) and for correct trials following errors (Lower). (D) (Upper) Average TSI of theta-power on correct trials following previous correct (blue) or error (red) trials calculated for successive time windows before stimulus onset. (Bottom) Statistical inference of TSIs, with significant values above gray shaded area.

Fig. 4.

Spike–LFP phase consistency in the theta-band can be task-selective and is stronger when theta-activity is larger. (A) Six example trials of a sparsely firing ACC neuron and the LFP (passband 5–10 Hz) during the preparatory period during pro- (red) and antisaccade (blue) trials. (B) Scatterplot of spike–LFP phase consistency for pro- (x axis) and antisaccade (y axis) trials. Filled circles show spike–LFP pairs with Rayleigh significant (P < 0.05) phase consistency for only the antisaccade task (blue), the prosaccade task (red), or both tasks (black). (C) Spike–LFP phase consistency for median split subsets of trials according to theta-power for either task separately, showing that, for the antisaccade task, spikes are on average significantly (P < 0.05) more synchronized to LFP theta when theta-power is stronger.