Voluntary modulation of mental effort investment: an fMRI study

Abstract

Mental effort is a common phenomenological construct deeply linked to volition and self-control. While it is often assumed that the amount of exertion invested in a task can be voluntarily regulated, the neural bases of such faculty and its behavioural effects are yet insufficiently understood. In this study, we investigated how the instructions to execute a demanding cognitive task either "with maximum exertion" or "as relaxed as possible" affected performance and brain activity. The maximum exertion condition, compared to relaxed execution, was associated with speeded motor responses without an accuracy trade-off, and an amplification of both task-related activations in dorsal frontoparietal and cerebellar regions, and task-related deactivations in default mode network (DMN) areas. Furthermore, the visual cue to engage maximum effort triggered an anticipatory widespread increase of activity in attentional, sensory and executive regions, with its peak in the brain stem reticular activating system. Across individuals, this surge of activity in the brain stem, but also in medial wall cortical regions projecting to the adrenal medulla, positively correlated with increases in heart rate, suggesting that the intention to willfully modulate invested effort involves mechanisms related to catecholaminergic transmission and a suppression of DMN activity in favor of externally-directed attentional processes.

Figure 1

The Stroop task. Example screenshots for congruent and incongruent trials are shown on the left and the right side, respectively. The correct button presses are indicated by arrows. Note that the order of the labels at the bottom of the screen was randomly assigned at each trial, and that in the incongruent trials (but not in the congruent trials) the font color of such labels was also discordant with their text.

Figure 2

Effect of volitional effort (EXR, RLX) and stimulus type (congr, incongr) on response times (RT). In this and subsequent repeated-measures interaction plots, error bars display Fisher’s Least Significant Difference.

Figure 3

Correlation of individual EXR-RLX changes in response times (RT) and number of errors per run, for incongruent trials only. The value of Pearson’s correlation coefficient (r) is indicated in the plot.

Figure 4

Subjective task-load ratings (NASA-TLX instrument) for the EXR and RLX conditions. The Performance subscale has been relabeled as ‘negPerform’, to remind the reader that its scores are high for ratings of poor performance. Error bars represent 95% confidence intervals for the mean scores. Paired comparisons marked with an asterisk are significant at a two-tail α < 0.05, Bonferroni-corrected.

Figure 5

Average heart rate during task and rest blocks in the two endogenous effort modulation conditions.

Figure 6

Objective performance v.s. subjective experience in the Stroop task. The left graph shows the across-subject correlation between EXR-RLX differences in response times (RT) to congruent trials and EXR-RLX differences in the TLX-NASA Effort subscale. The right graph shows the correlation between EXR-RLX differences in response times (RT) to incongruent trials and and EXR-RLX differences in the TLX-NASA negPerform subscale. Pearson’s correlation coefficients are reported within the plot area.

Figure 7

Activation map for the contrast cueEXR-cueRLX. Top, middle, and bottom rows illustrate representative slices in axial, sagittal, and coronal view, respectively. In this and other brain maps, the underlying anatomy is the average of the Talairach-warped anatomical volumes of all participants.

Figure 8

Relationship between individual EXR-RLX changes in heart rate (HR) and in the BOLD response to the effort cue. On the left side, the HR changes are regressed onto the BOLD average contrast of parameter estimates in a 3mm-radius sphere centered on the global activation peak from the cueEXR-cueRLX contrast (Talairach coordinates: 8, −22, −8). On the right side, similar graphs are displayed for the anterior and posterior midcingulate clusters (aMCC, pMCC) from the whole-brain analysis. Pearson’s correlation coefficients (r) and 95% confidence intervals (CI) are indicated in the plots.

Figure 9

Statistical parametric maps for different contrasts. The top two rows illustrate task-related activation, compared to passive fixation, in the EXR and RLX conditions; the third row from top illustrates the main effect of endogenous effort modulation; the fourth and fifth row split the latter effect by stimulus type; finally, the bottom row shows the interaction of voluntary effort engagement and stimulus type, as identified by the [(EXR-RLX)/incongr −(EXR-RLX)/congr] contrast. Note that the top colourbar refers to the first two rows from top, and the bottom colourbar to the remaining four rows.

Figure 10

Average BOLD response to correct trials, split by level of endogenous effort and trial type, in each of the clusters from the EXR-RLX Z-map. Abbreviations: R = right, L = left, RL = bilateral.

Figure 11

Interaction of voluntary effort investment and trial type on the BOLD response to correct trials in each of the clusters from the cond × stim Z-map.

Figure 12

Areas jointly sensitive to exogenous and endogenous effort modulation. The map was obtained by a binary intersection of the thresholded statistical maps corresponding to the two effects. Red indicates joint positive activation, blue joint negative activation.

Figure 13

Areas jointly sensitive to EXR-RLX effects during the preparatory cue and during the actual task execution. The map was obtained by a binary intersection of the thresholded statistical maps corresponding to the two effects. All clusters correspond to positive activations in the original EXR-RLX contrasts.

Figure 14

Across-subjects EXR-RLX changes in RT vs. EXR-RLX changes in cerebellar BOLD activation. The displayed linear relationship was significant at a Bonferroni-corrected p = 0.025.

Figure 15

Correlation of individual EXR-RLX BOLD changes in the cerebellar cluster and in the number of errors per run, for incongruent trials only. The value of Pearson’s correlation coefficient (r) is indicated in the plot.