Sleep consolidation potentiates sensorimotor adaptation

The study addresses whether sleep actively contributes to the consolidation of sensorimotor adaptation (SMA) memories. While sleep robustly benefits declarative memory and many motor skill learning (MSL) paradigms, prior SMA studies often suggest consolidation occurs with time irrespective of sleep. The authors hypothesize that the apparent lack of sleep benefit in SMA may stem from long, uncontrolled delays between training and bedtime in earlier work. They propose that if sleep occurs soon after learning—while the memory trace is fragile—SMA may benefit from sleep. The purpose is to test how the proximity of sleep to training modulates long-term retention and to determine whether sleep engages active consolidation mechanisms, indexed by NREM sleep spindles and their coupling to slow oscillations (SOs). The work has implications for unifying consolidation mechanisms across memory systems and for optimizing rehabilitation schedules.

Prior research shows NREM sleep enhances declarative memory and can stabilize or enhance MSL, particularly with explicit sequence knowledge. However, SMA studies (force-field and visuomotor adaptation) frequently report comparable retention after equivalent intervals of sleep or wake, suggesting time-dependent but sleep-independent consolidation. Notably, most prior SMA studies did not control the timing between training and bedtime, despite evidence in declarative and MSL domains that shorter training-to-sleep intervals improve consolidation. The authors also reference work showing sleep-dependent benefits in visuomotor learning when training occurs just before sleep, highlighting discrepancies likely attributable to training–sleep proximity. Collectively, the literature motivates testing whether SMA benefits from sleep when sleep closely follows learning, and identifying the time window of memory fragility using interference approaches.

Participants: 290 right-handed adults (150 females; mean age 24.3±4 years) without neurological/psychiatric disorders, recruited from the University of Buenos Aires. Regular sleep schedules were requested and monitored via self-reports. Ethics approval obtained; informed consent provided.

Task and apparatus: Visuomotor adaptation (VMA) center-out task using a right-hand joystick to move a cursor from center to one of 8 targets (1 cycle = 8 trials; 1 block = 11 cycles). Hand vision occluded. Cursor feedback present until target ring; inter-trial interval 1500–2000 ms jittered. Joystick gain set to 1.4 to limit online corrections. Trial types: null (veridical mapping), perturbed (±30° rotation; one control used 45°), and error-clamp (EC; straight-line fake feedback with 10° SD) to assess retention without new learning. Implemented in MATLAB with Psychtoolbox.

Design:

• Experiment 1 (effect of sleep when training–sleep gap is uncontrolled): 111 participants trained (1 null block + 6 blocks of 30° CCW). Five groups tested after 15 min (n=22), 1 h (n=25), 3 h (n=22), 5.5 h (n=22), or 9 h (n=20), all during wake; a sixth group slept and was tested at 24 h (n=23). Training times varied across daytime (no fixed schedule). Retention measured via 2 EC cycles; forgetting curve fit with single exponential to derive wake asymptote.
• Experiment 2 (anterograde interference to map consolidation window): Participants adapted to A (30° CCW; 6 blocks) then B (30° CW; 6 blocks) separated by 5 min (n=15), 1 h (n=20), 6 h (n=19), or 24 h (n=18). Control (n=20) trained only on B. All tested 24 h after B with 2 EC cycles. Napping was prohibited. A control sub-experiment (“overlearning group”; n=20) matched time spent at asymptote to the 5 min/1 h groups (4 blocks on B) to rule out overlearning as a confound.
• Experiment 3 (effect of controlling training–sleep gap): Two main groups trained on B (30° CW; 1 null + 6 perturbed blocks) and had overnight PSG. AM/AM (n=23): trained in morning, slept ~14 h after training (outside window). PM/PM (n=21): trained at night, slept within ~20 min (inside window). Retention assessed 24 h later (2 EC cycles). Circadian control groups: AM/PM (n=20; morning train, evening test, ~9 h wake) and PM/AM (n=10; night train, morning test, ~9 h with sleep; used 45° rotation from prior dataset). Daytime naps discouraged.

PSG acquisition: 11 EEG electrodes (FC1/2/5/6, C3/4, P3/4, Fz, Cz, Pz) referenced to mastoids; EOG and EMG recorded. Sampling at 200 Hz (Alice 5 or BWmini). Manual sleep staging in 30 s epochs (W, N1, N2, N3, REM) following standard criteria. Sleep architecture measures computed.

EEG processing and event detection: Signals bandpass filtered (EEG 0.5–30 Hz; EOG 0.5–15 Hz; EMG 20–99 Hz). SOs (0.5–1.25 Hz) detected via zero-crossing algorithm, 0.8–2 s duration; events > median peak-to-peak amplitude retained. Spindles (10–16 Hz) detected using Hilbert-based instantaneous amplitude with 90th/70th percentile thresholds; duration 0.5–3 s; only fast spindles (≥12 Hz) included. Spindle–SO coupling defined when spindle peak P–P amplitude fell within ±1.2 s of SO trough. Metrics computed during NREM (first sleep cycle).

Sleep metrics: Density of fast spindles (per min NREM) and density of fast spindle–SO couplings (per min NREM). Inter-hemispheric percent change computed as (Left−Right)/Right×100 over homologous pairs (FC1–FC2, FC5–FC6, C3–C4, P3–P4).

Behavioral analysis: Pointing angle per trial (joystick motion angle relative to target line). Outliers removed (>120° or >45° from cycle median). Cycle-wise medians computed. Learning rate estimated by fitting a single exponential y(t)=a·exp(−b·t)+c. Retention computed as EC pointing angle (% of asymptotic pointing angle from last learning block), averaged over 2 EC cycles. Forgetting (Exp 1) modeled with single exponential across groups: y(t)=a·exp(−b·t)+c.

Statistics: Parametric analyses in R. Between-subjects one- or two-way ANOVAs for retention, learning rate (b), and final asymptote. Linear mixed models for sleep metrics with random intercepts for subjects; fixed factor group; Kenward–Roger degrees of freedom; Dunnett’s or Bonferroni-corrected post hoc tests. Alpha=0.05.

Experiment 1 (uncontrolled training–sleep gap):

• All groups adapted similarly (learning rate: F(5,128)=0.85, p=0.52; final asymptote: F(5,128)=1.22, p=0.30).
• Retention declined over wake with time (F(4,106)=13.51, p<0.001); means±SEM: 15 min 79.6±3.1%; 1 h 66.8±3.9%; 3 h 53.6±4.7%; 5.5 h 44.1±4.2%; 9 h 42.0±5.6%.
• Forgetting fit: y(t)=a·exp(−b·t)+c with b=0.44 h⁻¹ (time constant ≈2.25 h), c=40.97%. Asymptote reached ~5.5–6 h post-learning (3 h vs c: t(21)=2.68, p=0.028 Bonferroni; 5.5 h vs c: t(21)=0.74, p=0.94).
• Overnight group (24 h) retention 40.5±4.1% did not differ from wake asymptote (t(22)=−0.122, p=0.90): no net sleep benefit when training–sleep gap is uncontrolled.

Experiment 2 (anterograde interference to map fragility):

• Interference reduced long-term retention (F(4,87)=7.61, p<0.001). Means±SEM: 5 min=18.0±4.3%; 1 h=19.1±4.5%; 6 h=38.6±5.9%; 24 h=41.0±4.8%; Control=48.3±4.5%.
• Strong deficits at 5 min and 1 h (Dunnett vs Control: both p<0.001), dissipating by 6 h (6 h vs Control p=0.41; 24 h vs Control p=0.65). Indicates greatest vulnerability within ~1 h after training.
• Control experiment (overlearning): Overlearning matched to 5 min/1 h groups (mean 24.1±2.4 cycles; F(2,52)=0.326, p=0.72). Despite less overlearning than Control, overlearning group retention did not differ from Control (Dunnett p=0.44), confirming fragility pattern is not explained by time at asymptote.

Experiment 3 (controlled training–sleep gap with PSG):

• Learning comparable across groups (learning rate: F(3,70)=1.46, p=0.23; asymptote: F(3,70)=0.860, p=0.461).
• Training close to sleep improved retention by ~31%: PM/PM=55.5±4.4%; PM/AM=55.9±10.5% vs AM/AM=42.9±4.3%; AM/PM=41.8±5.6%. Two-way ANOVA: main effect of training time (near vs far from sleep) F(1,70)=5.52, p=0.02; no effect of time of test (F(1,70)=0.18, p=0.67). PM groups fell asleep ~21.9±2.8 min after training.
• Sleep architecture did not differ between AM/AM and PM/PM (all p>0.05).
• EEG markers (first NREM cycle): Inter-hemispheric percent change (Left vs Right) increased only when training preceded sleep (PM/PM): • Fast spindle density: PM/PM 12.1±1.6% vs AM/AM 1.6±1.5%; group effect F(1,38.87)=6.48, p=0.015; one-sample vs zero: PM/PM t(20.36)=4.1, p=0.001; AM/AM t(18.8)=0.53, p=1. • Fast spindle–SO coupling density: PM/PM 14.8±1.6% vs AM/AM 1.6±1.1%; group effect F(1,37.74)=6.02, p=0.019; one-sample vs zero: PM/PM t(19.8)=3.5, p=0.004; AM/AM t(19.2)=0.52, p=1.
• AM/AM retention matched the 24 h group in Exp 1 (42.9±4.3% vs 40.5±4.1%; t(43.8)=0.40, p=0.3), confirming the sleep benefit is a net enhancement when sleep follows closely after training.

The results resolve a longstanding debate by demonstrating that SMA consolidation depends on both time and sleep, contingent on training–sleep proximity. When training is distributed and the sleep interval is not controlled, SMA retention follows a time-dependent forgetting curve and sleep confers no net advantage. However, when sleep occurs within the first hour post-training—the period of greatest memory fragility revealed by anterograde interference—long-term retention is enhanced by about 30%. The enhancement is accompanied by increases in contralateral fast spindle density and spindle–SO coupling during NREM, consistent with an active role of sleep in consolidation rather than mere passive protection from interference. These neural findings align with systems consolidation accounts that emphasize coordinated SO–spindle (and potentially ripple) dynamics supporting reactivation and integration of newly acquired information. The convergent time courses of forgetting and consolidation suggest that memory stability and susceptibility to interference are linked. The study reconciles divergent SMA findings in prior literature by identifying training–sleep proximity as a critical determinant and suggests common consolidation mechanisms across declarative, MSL, and SMA domains.

This work shows that sensorimotor adaptation consolidates over time during wake but can be potentiated by sleep when sleep promptly follows training. By identifying a critical vulnerability window (~first hour post-learning) and demonstrating both behavioral enhancement and concomitant electrophysiological markers (increased contralateral fast spindle density and spindle–SO coupling), the study supports an active role of NREM sleep in SMA consolidation. These insights unify consolidation principles across memory systems and have translational implications: aligning rehabilitation or training sessions to precede sleep or planned naps may improve long-term motor outcomes. Future research should causally probe the roles of SO–spindle coupling (e.g., closed-loop stimulation), test generalization across SMA tasks and populations, disentangle active vs passive sleep contributions, and examine hippocampal involvement and neuromodulatory mechanisms underlying the time-dependent sleep benefit.

• The study design does not directly dissociate active consolidation from passive protection by sleep; although EEG markers support active mechanisms, causal manipulations were not performed.
• The synaptic homeostasis hypothesis was not directly tested; the authors note their design precluded directly examining SHY.
• Generalizability is limited to a young, right-handed cohort and to a specific SMA paradigm (visuomotor rotation); effects across other forms of sensorimotor adaptation and diverse populations remain to be established.
• Experiment 1 did not control the training–sleep gap by design (to emulate prior literature), which may introduce heterogeneity in daily contexts and circadian phase at training.
• PSG analyses focused on the first NREM cycle; other sleep periods and dynamics were not analyzed in depth.