Predicting lapses of attention with sleep-like slow waves

The study investigates whether diverse forms of attentional lapses—behavioural sluggishness vs. impulsivity and phenomenological mind wandering (MW) vs. mind blanking (MB)—share a common physiological basis. Prior work links sleep pressure and local sleep-like slow waves during wakefulness to performance errors, but the relationship with subjective experience is unclear. The authors hypothesize: (i) slow waves can predict, at the single-trial level, both sluggish and impulsive behaviours in well-rested individuals; (ii) slow waves are associated with both MW and MB; and (iii) the spatial location of slow waves differentiates sluggishness vs. impulsivity and MW vs. MB. They propose that local slow waves act as a functional switch by transiently perturbing cortical networks, leading to distinct outcomes depending on location.

The paper synthesizes evidence that attentional lapses are prevalent and heterogeneous (MW and MB). Sleepiness and time awake increase both sluggish and impulsive responses and are linked to MW and MB. Local sleep—sleep-like slow waves in specific regions during wakefulness—has been observed in animals and humans and correlates with task errors. Intracranial and scalp EEG studies show wake slow waves reduce neuronal firing and predict errors in tasks recruiting affected regions. The authors relate lapses to transitions toward sleep and hypnagogic phenomena where experiences similar to MW and MB occur. They discuss the Default Mode Network’s debated role in spontaneous thought and propose local sleep may reconcile mixed findings. Conflicting reports on alpha oscillations and pupil size during MW may reflect arousal and context; slow waves may be a clearer marker of sleepiness. Prior work also links local slow-wave modulation to dream occurrence and content, suggesting broader implications for spontaneous experiences across sleep–wake states.

Participants: 32 recruited; 26 included (age 29.8 ± 4.1 years; 10 females). Ethics approved; informed consent obtained. One participant lacked eye-tracking data. Design: Two Sustained Attention to Response Tasks (SARTs): Face SART (Go: neutral faces; NoGo: smiling female face) and Digit SART (Go: digits 1–9; NoGo: digit 3). Continuous stimulus presentation with uniform jittered durations 0.75–1.25 s. Three blocks per task (12–15 min each), with rests; total duration ~103 ± 19.7 min. Random interruptions every 30–70 s probed mental state just before interruption: task-focused (ON), mind wandering (MW), mind blanking (MB), or don’t remember (collapsed with MB due to rarity). Participants rated vigilance on a 4-point scale. Behavioural measures: Go misses (no response before next stimulus), NoGo false alarms (response during NoGo), and reaction times (RTs). Trials with RT <300 ms excluded. Analyses focused on trials within 20 s before probes (also checked 10 s). Physiology: High-density EEG (63 active electrodes; BrainAmp), 500 Hz sampling; online ref AFz; ground Fpz; EOG and ECG recorded. EyeLink 1000 eye-tracking at 1000 Hz (pupil size and eye movements). Preprocessing: high-pass >0.1 Hz (two-pass 5th-order Butterworth), notch 45–55 Hz (4th-order Butterworth), bad channels removed/interpolated; data segmented [-32, 32] s around probes; per-segment mean removed. Pupil size blink-corrected and averaged per trial; for pre-probe analyses, pupil values in the 20 s window were averaged and discretized into 5 bins within participant/task. Slow-wave detection: Re-reference to mastoids (TP9/TP10); downsample 128 Hz; band-pass filter in delta with type-2 Chebyshev (stopband 0.1–15 Hz atten ≥25 dB; passband 1–10 Hz <3 dB). Detect waves via negative peaks; extract start/end (zero-crossings), negative/positive peak amplitudes and times, peak-to-peak amplitude, downward (start to negative peak) and upward (negative to positive peak) slopes. Exclusions: positive peak >75 µV (reduce blink contamination), within 1 s of events with |amplitude| >150 µV, duration <143 ms (>7 Hz). Select top 10% largest absolute peak-to-peak amplitude per electrode as local sleep slow waves (mean threshold ~30.7 ± 1.5 µV). Compute density, amplitude, downward and upward slopes per electrode and window. ERPs compared to task events to differentiate from evoked responses. Statistical analysis: Linear mixed-effects models (LMEs) with subject as random effect and task (Digit vs Face) as fixed effect; predictors included mental state (ON, MW, MB) or slow-wave presence (binary) or slow-wave properties. Model comparisons via likelihood ratio tests with Bonferroni corrections where applicable. Cluster-based permutation tests identified significant electrode clusters (cluster alpha p<0.025, Monte Carlo p_cluster<0.05; Bonferroni-corrected across comparisons). Decision modelling: Hierarchical Bayesian drift-diffusion modelling (HDDM) fit to Go/NoGo SART RT distributions and choices. Parameters: drift rates v_Go, v_NoGo; decision threshold a; non-decision time t (NDT); starting point bias z; drift bias v_bias (|v_Go|−|v_NoGo|). MCMC 8000 samples, 2000 burn-in. Models estimated by mental state (within 20 s pre-probe) and, separately, by slow-wave presence at each electrode (trial flagged if a slow-wave onset occurred between stimulus onset and offset). Posterior predictive checks validated fits. Subject-level point estimates extracted for statistical comparisons.

• Prevalence of mental states (mean ± SEM across N=26): ON ~49% (Faces 49.4 ± 4.9%; Digits 47.2 ± 5.1%), MW ~38–41% (Faces 38.0 ± 4.3%; Digits 40.9 ± 4.8%), MB ~12% (Faces 12.7 ± 3.0%; Digits 11.9 ± 2.9%).
• Baseline performance: Misses on Go trials: Faces 3.1 ± 0.4%; Digits 1.7 ± 0.2%. False alarms (FAs) on NoGo: Faces 35.0 ± 2.5%; Digits 32.5 ± 2.7%.
• Behaviour vs mental state (20 s pre-probe): • Misses: χ²(2)=36.0, p=1.5×10⁻⁸; MW vs ON β=0.011 [0.005,0.016]; MB vs ON β=0.023 [0.015,0.032]; MB vs MW β=0.013 [0.005,0.021]. • FAs: χ²(2)=115.9, p<10⁻¹⁶; MW vs ON β=0.21 [0.17,0.24]; MB vs ON β=0.17 [0.12,0.23]; MB vs MW β=−0.028 [−0.084,0.028]. • RTs: χ²(2)=16.3, p=2.9×10⁻⁴; MB vs ON β=0.019 [0.009,0.030]; MB vs MW β=0.022 [0.011,0.032]; MW vs ON β=−0.0025 [−0.0096,0.0045]. Pattern: MB = sluggish (more misses, slower RTs); MW = more impulsive (more FAs, relatively faster RTs than MB).
• Vigilance: • Subjective ratings: χ²(2)=144.8, p<10⁻¹⁶; MW vs ON β=−0.39 [−0.40,−0.37]; MB vs ON β=−0.53 [−0.55,−0.50]; MB vs MW β=−0.13 [−0.24,−0.02]. • Pupil size (N=25): χ²(2)=18.0, p=1.2×10⁻⁴; MW vs ON β=−0.29 [−0.43,−0.15]; MB vs ON β=−0.22 [−0.45,−0.003]; MB vs MW β=0.065 [−0.16,0.29].
• Global slow-wave properties vs vigilance/pupil: • Vigilance ratings (scalp-averaged slow-wave features over −20–0 s): density χ²(1)=13.1, p=3.9×10⁻⁴, β=−0.074 [−0.114,−0.034]; amplitude χ²(1)=33.1, p=8.5×10⁻⁹, β=−0.023 [−0.031,−0.015]; downward slope χ²(1)=82.1, p<10⁻¹⁶, β=−2.5×10⁻³ [−3.1×10⁻³,−2.0×10⁻³]; upward slope χ²(1)=47.3, p<10⁻¹¹, β=−1.7×10⁻³ [−2.2×10⁻³,−1.2×10⁻³]. • Pupil size: density χ²(1)=12.3, p=4.6×10⁻⁴, β=−0.13 [−0.21,−0.059]; amplitude χ²(1)=7.4, p=6.4×10⁻³, β=−0.0088 [−0.015,−0.0025]; downward slope χ²(1)=15.1, p=1.0×10⁻⁴, β=−6.9×10⁻⁴ [−1.0×10⁻³,−3.4×10⁻⁴]; upward slope χ²(1)=17.6, p=2.7×10⁻⁵, β=−7.6×10⁻⁴ [−1.1×10⁻³,−4.0×10⁻⁴]. Slow-wave amount increased with time-on-task.
• Local slow-wave properties predicting mental states (−20–0 s pre-probe; cluster-permutation): • MW>ON: increased slow-wave density and amplitude over frontal electrodes; increased downward and upward slopes over centro-frontal sites. • MB>ON: increased slow-wave density over frontal areas; increased downward and upward slopes over centro-parietal electrodes (no significant amplitude cluster). • MB>MW: frontal slow-wave amplitude reduced; parietal upward slope increased for MB. Effects strongest within last 5 s before probe.
• Trial-level behaviour vs slow-wave presence (during stimulus window): • Frontal slow waves: faster RTs and more FAs (impulsivity). • Posterior slow waves: slower RTs and more misses (sluggishness). Significant electrode clusters identified for RT, FA, and miss effects.
• Drift-Diffusion Modelling (HDDM): • Mental states: MW vs MB showed lower decision threshold and higher decision bias for MW, consistent with impulsivity (Supplementary Fig. 9). • Global effects of slow waves: reduced decision threshold (α), increased non-decision time (t), and increased starting-point bias (z) toward Go. • Local effects: posterior slow waves reduced drift for Go (v_Go) and drift bias (v_bias); frontal slow waves reduced drift for NoGo (v_NoGo). These account for slower/missed Go responses with posterior slow waves and faster/impulsive responses and FAs with frontal slow waves.

Findings support a unified physiological mechanism for diverse attentional lapses: local sleep-like slow waves in the awake brain precede and accompany both behavioural failures and subjective MW and MB. Spatial specificity explains heterogeneity: frontal slow waves disrupt executive control, biasing toward impulsive, fast Go responses and increased FAs; posterior slow waves impair sensory integration, yielding sluggish responses and misses, and relate to reduced awareness (MB). Slow waves track global sleepiness (vigilance ratings, pupil size, time-on-task), while their location predicts the lapse’s behavioural and phenomenological profile. This framework may reconcile mixed reports linking MW to default mode network activity, alpha oscillations, and pupilometry by introducing local sleep as a key factor mediating arousal and spontaneous thought. Results also inform debates on frontal versus posterior contributions to consciousness: posterior slow waves, often broader in extent, align with transient reductions in awareness (MB), whereas frontal slow waves align with unconstrained thought (MW). Drift–diffusion modelling reveals mechanistic subcomponents influenced by slow waves (threshold, non-decision time, drift rates, biases), linking physiology to decision-making processes.

The study demonstrates that in well-rested individuals performing undemanding tasks, spatially localized, sleep-like slow waves reliably predict attentional lapses in both behaviour and subjective experience. Slow waves serve as a global marker of fatigue and, depending on location, mechanistically explain impulsive versus sluggish behaviours and MW versus MB. This advances a unified account of attentional lapses based on local sleep intrusions during wakefulness. Future work should establish causal and neuronal underpinnings via intracranial recordings, sleep deprivation paradigms, source localisation and simultaneous EEG–fMRI, include task-free resting sessions, and test generalization to naturalistic settings (e.g., driving, classroom). Exploration of neuromodulatory contributions and other mechanisms of sensory decoupling may complement this account. Applications may include real-time brain–machine interfaces to detect and mitigate lapses in educational and professional contexts.

Evidence is correlational; wake slow waves are inferred as sleep-related via converging markers (vigilance, pupil, time-on-task), not directly confirmed. Continuous visual stimulation prevents complete separation of slow waves from task-evoked activity, though ERPs differ. The SART is undemanding and may promote sleepiness, limiting generalizability to more engaging tasks or real-world scenarios. Spatial resolution is limited to scalp EEG; precise sources and neuronal silencing associated with wake slow waves require intracranial confirmation. Other mechanisms (e.g., neuromodulatory changes) may also contribute to attentional lapses, MW, and MB.