Motor cortex activity predicts response alternation during sensorimotor decisions

Abstract

Our actions are constantly guided by decisions based on sensory information. The motor cortex is traditionally viewed as the final output stage in this process, merely executing motor responses based on these decisions. However, it is not clear if, beyond this role, the motor cortex itself impacts response selection. Here, we report activity fluctuations over motor cortex measured using MEG, which are unrelated to choice content and predict responses to a visuomotor task seconds before decisions are made. These fluctuations are strongly influenced by the previous trial's response and predict a tendency to switch between response alternatives for consecutive decisions. This alternation behaviour depends on the size of neural signals still present from the previous response. Our results uncover a response-alternation bias in sensorimotor decision making. Furthermore, they suggest that motor cortex is more than an output stage and instead shapes response selection during sensorimotor decision making.

Figure 1. Visuomotor decision task.

(a) Participants reported the presence of coherent motion in a display of randomly moving dots with a left- or right-hand button-press. In each trial, the mapping from choice to response hand was newly assigned with a colour cue after the stimulus (choice-response cue). Successive trials were separated by a variable length ITI (median ITI: 1,290 ms). (b) For a red cue (choice-response mapping 1), participants reported the presence and absence of coherent motion with a right and left hand button-press, respectively. The mapping from choice to response was reversed for the green cue (choice-response mapping 2).

Figure 2. Motor cortex activity predicts upcoming responses.

(a) Time-frequency analysis of the difference of source-reconstructed motor cortex activity contralateral minus ipsilateral to each button-press. Beta power (12–30 Hz) shows a characteristic contralateral suppression before the button-press. Z-scores across subjects (n=20 subjects). (b) Beta power (12–30 Hz) immediately before left minus right button-presses (4.5–5.5 s). Beta power suppression is focused on motor cortex. White dashed lines mark the central sulcus. (c) Time-course of beta power (12–30 Hz) in motor cortex contra- and ipsilateral to the button-press. Activity is normalized by the mean across trials. Shaded areas indicate SEM across participants. Black bars mark significant differences, that is, response-predictive activity (−1.0 to 1.1 s, P=0.01; 4.5–6.6 s, P=0.002; two-tailed one-sample cluster permutation tests, n=20). (d) Time-course of response-predictive beta activity, that is, of the difference in beta power between hemispheres contra- and ipsilateral to the button-press. (e) Difference between contra- and ipsilateral beta power averaged across the prestimulus period (−1 to 1.25 s) is significantly different from 0 in both correct (P<0.001) and incorrect trials (P=0.018, two-tailed one-sample permutation tests, n=20).

Figure 3. Motor signals from previous trial affect current trial.

(a) Time-frequency analysis of motor cortex activity contralateral minus ipsilateral to previous trial's button-press in pairs of consecutive trials. Data from previous trial is aligned to button-press. Data from the current trial are aligned to stimulus onset. Data from consecutive trials are concatenated according to median ITI. Z-scores across subjects (n=20). (b) Time-course of beta power (12–30 Hz) in motor cortex contra- and ipsilateral to previous trial's button-press. Activity is normalized by the mean across trials. Shaded areas indicate SEM across participants. Black bar marks a significant difference from 0.7 s after the previous button-press to 4.6 s in the current trial (P=0.002, two-tailed one-sample cluster permutation tests, n=20). (c) Time-course of the difference in beta power contra- and ipsilateral to the previous trial's button-press. (d) Beta power after left minus right button-presses (−1 to 1.25 s). Dashed lines indicate the hand representation of primary motor cortex and the central sulcus, respectively. (e) Beta rebound averaged across the prestimulus period is not significantly different after correct and incorrect choices (P=0.84, two-tailed paired permutation test, n=20).

Figure 4. Response-predictive activity in trials with and without alternation.

(a) Beta lateralization (12–30 Hz; contralateral-ipsilateral to the response) for non-alternation trials (that is, trials where the same button is pressed in the current trial as in the previous trial) is lateralized with a positive sign throughout most of the trial, that is, opposite to the lateralization immediately preceding the button-press at 6.1 s. For alternation trials, the lateralization is negative throughout the entire trial. (b) Beta power (12–30 Hz) across the whole cortex during the prestimulus interval (−1 to 1.25 s) for left minus right button-presses (in the current trial) plotted separately for whether the button-press in the current trial is a non-alternation (that is, repetition of the previous button-press) or an alternation with respect to the previous button-press.

Figure 5. Response alternation and its relation to the beta rebound.

(a) Each participant's tendency to alternate response hands expressed as Pearson's r between opposite hands for consecutive trials. Note that dots are scattered horizontally to avoid overlap. Shaded area denotes SEM. (b) Relationship between response bias and choice accuracy across participants. (c) Relationship between beta rebound (lateralization with respect to previous button-press from −1 to 1.25 s of the current trial) and response alternation across participants. (d) Cortical distribution of the relationship between beta rebound and response alternation across participants. Correlations are masked at P<0.05. Note that the image is symmetric across the midline, because the beta rebound was calculated as the lateralization between corresponding points in both hemispheres.

Figure 6. Correcting response-predictive activity for previous responses.

(a) Time-course of response-predictive beta-lateralization, corrected and un-corrected for previous responses. (b) Response-predictive beta-lateralization, corrected and un-corrected for previous responses averaged across the prestimulus window (−1 s to 1.25 s). Error bars show SEM. Correcting for the previous response reduces prestimulus lateralization significantly (P=0.010, one-tailed paired permutation test, n=20). After correcting for the previous button-press, prestimulus lateralization still predicts the upcoming button-press (P=0.024, one-tailed one-sample permutation test, n=20).

Figure 7. Decision making with known choice-response mapping.

(a) Relationship between beta rebound (lateralization with respect to previous button-press from −1 to 1.25 s of the current trial) and response alternation across participants. (b) Participants' response alternation. Participants are grouped according to above or below median beta rebound. (c) Time-course of response-predictive beta activity, that is, of the difference in beta power between hemispheres contra- and ipsilateral to the button-press. The black bar marks significant response-predictive activity in control trials (2.3–6.1 s, P=0.002, two-tailed one-sample cluster permutation test, n=20). Data from main task re-plotted from Fig. 2c for comparison. (d) Lateralization with respect to the upcoming button-press in the prestimulus window of main and control trials. (e) For subjects with above median beta rebound, prestimulus lateralization predicts the upcoming response in the control task (P<0.001, time window from −1 to 1.25 s, one-tailed one-sample permutation test, n=10), whereas this is not possible in subjects with below median beta rebound (P=0.90, one-tailed one-sample permutation test, n=10).

Figure 8. Differences in beta rebound between main and control task.

(a) Lateralization of beta power with respect to the previous button-press (beta rebound) for main and control tasks across the trial. (b) Beta rebound for the main and control task averaged between the end of the first cue (0.25 s) and stimulus onset (1.25 s) are significantly different (P=0.036, one-tailed paired permutation test, n=20). (c) Beta rebound for the main and control task averaged across the second half of the stimulus and the delay period before the second cue (2.25 s–4.25 s) are significantly different (P=0.010, one-tailed paired permutation test, n=20).