Predicting lapses of attention with sleep-like slow waves

Abstract

Attentional lapses occur commonly and are associated with mind wandering, where focus is turned to thoughts unrelated to ongoing tasks and environmental demands, or mind blanking, where the stream of consciousness itself comes to a halt. To understand the neural mechanisms underlying attentional lapses, we studied the behaviour, subjective experience and neural activity of healthy participants performing a task. Random interruptions prompted participants to indicate their mental states as task-focused, mind-wandering or mind-blanking. Using high-density electroencephalography, we report here that spatially and temporally localized slow waves, a pattern of neural activity characteristic of the transition toward sleep, accompany behavioural markers of lapses and preceded reports of mind wandering and mind blanking. The location of slow waves could distinguish between sluggish and impulsive behaviours, and between mind wandering and mind blanking. Our results suggest attentional lapses share a common physiological origin: the emergence of local sleep-like activity within the awake brain.

Fig. 1. Experimental design and hypotheses.

a Participants performed both a SART (Sustained Attention to Response Task) on faces stimuli (NoGo trials: smiling faces) and a SART on digits (NoGo trials: 3). Face/Digit presentation was continuous (new face/digit every 0.75–1.25 s). Images of faces were obtained from the Radboud Face Database (see “Methods”). b Every 30 to 70 s, participants were interrupted and instructed to report their mental state (see “Methods” and Supplementary Methods). Most importantly, they were asked to indicate whether they were focusing on the task (task-focused: ON), thinking about nothing (mind blanking: MB) or thinking about something other than the task (mind wandering: MW). High-density EEG and pupil size were continuously recorded throughout the task. c Proportion of mental states reported during probes categorized as task-focused (ON, green), mind wandering (MW, orange) and mind blanking (MB, blue) during the tasks with Digits (circles for each individual participant; filled surfaces for smoothed density plot) and Faces (diamonds and surfaces with horizontal stripes). Grey diamonds and circles show the average across participants.

Fig. 2. Low arousal is associated with attentional lapses characterized by different behavioural outcomes.

Proportion of misses (a) and false alarms (b) in the 20 s preceding task-focused (ON, green), mind wandering (MW, orange) and mind blanking (MB, blue) during the tasks with Digits (circles for each individual participant; filled surfaces for smoothed density plot) and Faces (diamonds and surfaces with horizontal stripes). The markers’ size is proportional to the number of reports for each participant (same for c–e). Grey diamonds and circles show the average across participants, weighted by the number of reports (same for c–e). c Distribution of reaction times (RT) for Go Trials (left: Face; right: Digit) in the 20 s preceding ON, MW and MB reports. Box plots show the mean (central bar), the lowest and highest individual data points (end of the whiskers) and the lower and higher quartiles (edges of the box). On top of each plot are shown the smoothed density plot for the different mental states. d Vigilance scores (subjective ratings provided during probes) associated with ON, MW and MB reports. e Discretized pupil size (see “Methods”) in the 20 s preceding ON, MW and MB reports. In a–e, stars show the level of significance of the effect of mental states (likelihood ratio test, see “Methods”; ***p < 0.005; a: p = 1.5 × 10⁻⁸; b: p < 2 ×  10⁻¹⁶; c: p = 2.2 × 10⁻⁴; d: p < 2 × 10⁻¹⁶; e: p = 1.2 × 10⁻⁴). N = 26 except for panel (e), where N = 25.

Fig. 3. Properties of slow waves.

a Average waveform of the slow waves detected over electrode Cz during the behavioural tasks (red, left; N = 26 participants). The average waveform of slow waves detected during sleep (blue, right), extracted from another dataset (see Supplementary Methods), is shown for comparison. Slow waves were aligned by their start, defined as the first zero-crossing before the negative peak (see “Methods”). b Scalp topographies of the density of slow waves (arbitrary units) detected in wakefulness (top) and sleep (bottom). c Scalp topographies of wake slow waves properties (first column: temporal density; second: peak-to-peak amplitude; third: downward slope (D-slope); fourth: upward slope (U-slope); see “Methods”) averaged across participants (N = 26). d Scalp topographies for slow-wave density (first column), amplitude (second), downward slope (D-slope, third) and upward slope (U-slope, fourth) for the different mental state (ON (task-focused), MW (mind wandering) and MB (mind blanking)).

Fig. 4. Local properties of slow waves are predictive of mental states.

Locally, based on each individual electrode, we performed mixed-effect analyses, following with permutation analysis to quantify the impact of slow-wave properties on mental states. First column: density; second: amplitude; third: downward slope (D-slope); fourth: upward Slope (U-slope). These slow-wave properties were extracted for each electrode and used to compute the t values (shown in each topographical plot) from mixed-effect models on the following comparisons: a MW > ON, b MB > ON and c MB > MW (N = 26 participants). Black dots denote significant clusters of electrodes (p_cluster <0.05 corrected for 12 comparisons using a Bonferroni approach, see “Methods”). ON task-focused, MW mind wandering, MB mind blanking.

Fig. 5. Local occurrences of slow waves are associated with modulations of performance.

Mixed-effects models were used to quantify the correlation between slow-wave occurrence and reaction times (a), false alarms (b) and misses (c) at the single-trial level. Topographies show the scalp distribution of the associated t values (N = 26 participants). Black dots denote significant clusters of electrodes (p_cluster <0.05 corrected for three comparisons using a Bonferroni approach, see “Methods”).

Fig. 6. Global and local effects of the occurrence of slow waves on sub-components of decision-making.

Reaction times in the Go/NoGo tasks were modelled according to a Hierarchical Drift-Diffusion Model (see “Methods”). a–f Topographical maps of the effect of slow waves (i.e. whether or not a slow wave was detected for each trial and for a specific electrode) on the parameters of decision-making: drift Go [v_Go] (a), drift NoGo [v_NoGo] (b), drift bias [v_Bias] (c), threshold [a] (d), non-decision time or NDT [t] (e), decision bias [z] (f). The effect of slow-wave occurrence was estimated with LMEs (see “Methods”) and topographies show the scalp distribution of the associated t values (N = 26 participants). Black dots denote significant clusters of electrodes (p_cluster<0.05, Bonferroni-corrected, see “Methods”).