Precursor processes of human self-initiated action

Abstract

A gradual buildup of electrical potential over motor areas precedes self-initiated movements. Recently, such "readiness potentials" (RPs) were attributed to stochastic fluctuations in neural activity. We developed a new experimental paradigm that operationalized self-initiated actions as endogenous 'skip' responses while waiting for target stimuli in a perceptual decision task. We compared these to a block of trials where participants could not choose when to skip, but were instead instructed to skip. Frequency and timing of motor action were therefore balanced across blocks, so that conditions differed only in how the timing of skip decisions was generated. We reasoned that across-trial variability of EEG could carry as much information about the source of skip decisions as the mean RP. EEG variability decreased more markedly prior to self-initiated compared to externally-triggered skip actions. This convergence suggests a consistent preparatory process prior to self-initiated action. A leaky stochastic accumulator model could reproduce this convergence given the additional assumption of a systematic decrease in input noise prior to self-initiated actions. Our results may provide a novel neurophysiological perspective on the topical debate regarding whether self-initiated actions arise from a deterministic neurocognitive process, or from neural stochasticity. We suggest that the key precursor of self-initiated action may manifest as a reduction in neural noise.

Keywords: Externally-triggered action; Human; Readiness potential; Self-initiated action; Stochastic fluctuations.

Fig. 1

Timeline of an experimental trial. Participants responded to the direction of dot-motion with left and right keypresses. Dot-motion could begin unpredictably, after a delay drawn from an exponential distribution. A. In the ‘self-initiated’ blocks participants waited for an unpredictably occurring dot-motion stimulus, and were rewarded for correct left-right responses to motion direction. They could decide to skip long waits for the motion stimulus, by making a bilateral keypress. They thus decided between waiting, which lost time but brought a large reward, and ‘skipping’, which saved time but brought smaller rewards. The colour of the fixation cross changed continuously during the trial, but was irrelevant to the decision task. B. In the ‘externally-triggered’ blocks, participants were instructed to make bilateral skip keypresses when the fixation cross became red, and not otherwise.

Fig. 2

EEG activity prior to skip actions. The red and blue lines represent self-initiated and externally-triggered skip conditions, respectively. Data is time-locked to the skip action (black vertical line), baseline-corrected in a 10 ms window around the skip, and recorded from FCz electrode. The average time of the skip instruction (fixation cross changing to red) in the externally-triggered condition is shown as a grey vertical line. A. Grand average RP amplitude ± standard error of the mean across participants (SEM). B. Standard deviation across trials averaged across participants ± SEM. Shaded grey area shows a significant difference between standard deviation traces across central electrodes, detected by cluster-based permutation test.

Fig. 3

EEG activity prior to response to coherent dot motion direction. The red and blue lines represent activity in self-initiated and externally-triggered blocks, respectively. Data is time-locked to the response to coherent dot motion direction (black vertical line), baseline-corrected in a 10 ms window around the response, and recorded from FCz electrode. The average time of the coherent dot motion onset is shown as a grey vertical line. A. Grand average ERPs ± SEM across participants. B. Standard deviation across trials, averaged across participants ± SEM across participants.

Fig. 4

A–C. Results of sensitivity analysis. Effects of changing parameters of a stochastic accumulator model on SD across 1000 model runs. (A) Drift gradually changed from 0.05 (cyan) to 0.15 (blue) in 0.02 steps, while other parameters were kept fixed. (B) Leak gradually changed from 0.3 (magenta) to 0.7 (blue) in 0.1 steps, while other parameters were kept fixed. (C) Change in noise gradually changed from −0.05 (yellow) to 0.05 (green) in 0.02 steps, while other parameters were kept fixed. D-F. The best fitting parameters to real mean RP amplitude in self-initiated (red) and externally-triggered (blue) conditions. Asterisks show significant difference (p < 0.001). Error bars show SD across participants. G. Effect sizes (d_z) for the between-condition difference in fitted drift, the leak and the change in noise parameters. Error bars show 95% confidence interval.

Fig. 5

(A) Observed SD across trials averaged across participants ± SEM. Data are baselined to a 10 ms window around the skip and are recorded from FCz electrode. (B) Simulated SD across trials averaged across participants ± SEM. The red and blue lines represent activity in self-initiated and externally-triggered blocks, respectively. The black vertical line is the moment of skip action. (C) Correlation between observed and simulated EEG convergence. EEG convergence was measured by subtracting the area under the SD curve in self-initiated from the externally-triggered condition.

Fig. 6

Within-trial EEG variability. (A) SD was measured within 100 ms windows for each trial and each condition. Red and blue bars show within-trial SD in each time bin in self-initiated and externally-triggered conditions, respectively. The solid red and blue lines show the linear fit to the time bins in self-initiated and externally-triggered conditions, respectively. (B) The slope of the change in within-trial variability was then compared between the self-initiated (red) and externally-triggered (blue) skip conditions. Error bars show SEM across participants.

Fig. 7

Percentage change in total EEG power compared to baseline (2.5–2 s prior to action) in self-initiated (A), and externally-triggered skip conditions (B). In each condition, the percentage change in power was computed 1–0.5 s prior to skip action, and from 15 to 30 Hz based on previous literature (region of interest shown by black box). (C) The percentage change from baseline was compared between the self-initiated (red bar) and externally-triggered (blue bar) conditions. Error bars show SEM across participants.