Thalamic dynamics orchestrate the recovery of tonic alertness during nocturnal sleep inertia

Abstract

At dawn, we experience a shift from slumber to sentience following neurophysiological transitions, termed as sleep inertia (SI). Although resting-state fMRI studies have discovered brain reorganizations during SI, the neural basis underlying impaired alertness in SI remains unclear. We conduct simultaneous EEG-fMRI recordings on 26 adults with repeated measures of pre-sleep, nocturnal sleep and consecutive post-sleep awakenings. Using the psychomotor vigilance task (PVT) to probe tonic alertness, we discover that cingulo-opercular network (CON) activation, inclusive of thalamus, troughs upon awakening, and increments along with the post-sleep awake duration. The dynamic recovery of thalamus activity during SI depends on prior sleep architecture and the awake duration, mediating the PVT performances on awakening. Although CON connectivity remains stable, the connectivity changes between thalamus and frontoparietal network (FPN) are associated with changes of thalamic activation and PVT performances during SI. Collectively, thalamic activity and its coupling with the FPN support the restoration of tonic alertness during SI, providing a concise framework for the neural mechanisms underlying cognitive recovery upon awakening.

Sleep inertia (SI), waking from sleep, impairs tonic alertness temporarily. Thalamic activity recovers dynamically during SI, mediating PVT performance. Thalamo-FPN coupling specifically supports alertness restoration, framing SI cognitive recovery

Introduction

Forty-two percent of adolescents¹ experienced difficulty when waking up, during the transition from sleep to full wakefulness, a process known as sleep inertia (SI). This transition is characterized with subjective sleepiness, disorientation, and impacted cognitive performances^2,3, including declined response speed, impaired error-monitoring, and decision-making functions. SI could last from a few minutes to 30 min or longer^4,5, which significantly affect individuals execute safety-critical tasks after waking, such as surgeons performing emergent operations at night or workers required to make immediate decisions. The recovery of the SI depends on the method of waking (forced- or self-awakening), sleep duration, sleep composition, prior sleep deprivation, and mood disorders; and it is exacerbated in patients with depression or idiopathic hypersomnia³. A core function affected by SI is tonic alertness⁶, the ability of maintaining sustained attention of the surroundings and preparing responses. Tonic alertness is the bridge in cognitive control hierarchy connecting the arousal basis and phasic attentional functions⁷. Unlike the prolonged durations of hypo-vigilance after sleep deprivation⁸ or delirium after anesthesia⁹, the alertness level during SI recovers in a relatively short period. Therefore, investigating the neurophysiological mechanisms underlying the alertness recovery in SI can potentially provide clinical insights into treatments for patients with disorders of consciousness¹⁰.

Studies investigating tonic alertness during SI typically involve neuropsychological tests and/or neurophysiological measures². Among various neuropsychological measures^2,3,6, psychomotor vigilance task (PVT) is a common task requiring the visual– or auditory–motor action with tonic alertness for external-oriented monitoring^11,12, to assess hypo-vigilance states with robustness¹³, sensitivity¹⁴ and without a learning effect¹⁵. SI impairs multiple facets of PVT performance: response speed decreases upon awakening^5,16–18, and metrics such as the standard deviation of reaction time¹⁸, fastest and slowest responses⁵, and lapse counts¹⁸ were all sensitive to the acute SI effect. These behavioral metrics would reflect the maintenance of tonic alertness during SI.

In neurophysiology, the underlying brain mechanisms of SI have been investigated using regional cerebral blood flow (CBF), electroencephalography (EEG), and functional magnetic resonance imaging (fMRI) studies. After waking, declined CBF in the heteromodal neocortical areas indicates a delayed recovery for attention in SI¹⁹. Increased EEG sleep-like delta power in the posterior brain region is regarded as the neural signature of SI^20,21. Waking up from NREM sleep, the resting-state (rs)-fMRI connectivity within the sensorimotor network is disrupted^22,23. After an afternoon nap, the increased functional connectivity between the default mode network (DMN) and dorsal attention network (DAN) indicates a lack of segregation between brain networks, but no association has been observed with behavioral indices²⁴. However, contrasting results were reported in a recent nocturnal nap study²⁵, showing that DMN–DAN connectivity significantly decreased during SI. Despite that, Chen and colleagues reported that during SI, the negative correlation between the frontoparietal network (FPN) and EEG-based vigilance index disappeared²⁵, suggesting that the FPN may play a functional role in expediting the tonic alertness during SI.

To date, neuroimaging studies have provided only indirect neurophysiological evidence based on rs-fMRI brain-network studies rather than direct evidence using the coinstantaneous neuroimaging modality measuring the task-involved alertness function during SI. Literature supported that when maintaining high alertness on the task engagement (tonic alertness), the cingulo-opercular network (CON, composed of anterior insula, dorsal anterior cingulate cortex, and thalamus^7,26,27,) activity increased, and the DAN activity increased with selective attention (phasic alertness) instead²⁸. Therefore, it is speculated that CON activity may reflect the dynamic changes of tonic alertness during SI. However, one resting-state study reported that CON-based functional connectivity remains unchanged during the nap inertia²⁴, which conflicts against previous evidence of time-varying tonic alertness during SI. To date, the linkage between the time-varying dynamics of tonic alertness during SI and the corresponding brain functionality remains missing. Thus, in this study we investigated SI-related PVT performances and brain functionality, including both task-evoked brain activation and resting-state functional connectivity in fMRI, immediately upon waking. Specifically, we focused on tonic-alertness–related CON functions to test whether brain–behavior associations change dynamically across the wakefulness period during sleep inertia.

As a transition period, SI effect is weathered by two factors, the passive resistance from the sleep (e.g., the duration from sleep offset and the length of NREM sleep stages²⁴) and the active effort to reach full wakefulness (e.g., the trait/state factors²⁹ that determine the recovery speed). Therefore, we hypothesized that the functional indices (brain activity and connectivity) of CON present time-varying dynamics during sleep inertia, affecting the tonic-alertness performances along with the two factors. We precisely aligned fMRI measures to each participant’s exact awakening time and collected the depth of prior sleep through sleep scoring to characterize SI dynamics. Here we show that PVT-induced brain activities of the CON, including thalamus, troughed immediately after waking and slowly incremented with the waking time during SI. The thalamus activities mediated the relationship between awake duration, prior sleep history and PVT response speeds. Beyond brain activity, the connectivity changes between CON (specifically in thalamus) and the FPN shows a tight association with the recovery of PVT response speed instantly upon awakening.

Results

Twenty-six participants completed all sleep experimental procedures with simultaneous EEG-fMRI recordings, containing one pre-sleep session, one sleeping session, and three consecutive measures after awakening (Fig. 1a). Among the 26 participants, 15 participants were segregated into the Inertia group (Table S1), and 11 participants (Table S2) were taken as the Prolonged Wakefulness (PW) group, who had overlong scoring of “wake” status before the cessation of sleep session ( > 20 min). The first measurement of post-sleep session was significantly closer to the end of sleep among participants in the Inertia group (awake duration: 6.1 ± 1.4 min) as compared to those participants in the PW group (awake duration: 61.9 ± 10.8 min). No difference was observed in their time in bed (TIB) during sleep sessions (T_24.2 = .46, p = 0.65) between the Inertia (131.9 ± 12.9 min) and PW (123 ± 13.9 min) groups (Table S1 and Table S2), despite PW group has significantly shorter total sleep duration (T_24.2 = 3.37, p = 0.003).

Fig. 1. Experiment procedure and PVT performance in Inertia group.

a The experiment with simultaneous EEG-fMRI recording across all scans. The interview and questionnaire sections were conducted at least three days prior to the experiment. b Median speed (reciprocal of response time) across 4 sessions (Pre, A1, A2, and A3) and the trends of it for each participant. c Mean standard deviation of speed across 4 sessions and the trends of it for each participant. Significant increase on it was observed at A3. d Mean speed trends across 4 sessions of the fastest 20% (fSPD, in red), intermedia 60% (iSPD, in black), and slowest 20% (sSPD, in blue) trials. A significant interaction was observed between the measurement points (A2 and A3) and the speed conditions (fSPD and iSPD). However, this interaction was not significant when the analysis was performed using sSPD and iSPD instead. T1=high-resolution structural scan; R=resting-state scan (5-min); P = PVT scan (6-min); SLEEP=scan of sleep session; T(start) refers to participants’ regular bedtime; T(end) refers to the ending time of SLEEP; Pre=pre-sleep session ( ~ 30 minutes prior to T0); A1=1st, A2=2nd, A3=3rd session, in approximately 5, 20, 35 minutes after T(end); STD=standard deviation.

PVT Performances during SI

Across the four sessions, the median PVT speed and the standard deviation of the speed did not exhibit significant effect of time (Fig. 1b, c) in the Inertia group. The paired t-test of the standard deviation of the speed between the A1 and A3, and A2 and A3 revealed significant increase of the alertness variability in the late wakefulness (Fig. 1c). By separating PVT speeds into three ranges, we found a significant interaction indicating an accelerating trend in fSPD from A1 to A3 (F_1,14 = 12.7, p = 0.003) and from A2 to A3 (F_1,14 = 13.41, p = 0.003), as compared with the performance change in iSPD trials (Fig. 1d). However, no differences were observed between the trend in sSPD and the trend in iSPD. In the PW group, we did not observe any significant time effects and the interactions on any index of PVT performance, except the median speed which presented a trend of decrease during the measurements after wakeup (Fig. S1a).

Neural correlates of SI influenced by awake duration and sleep composition

The ROI-based mixed-effects model revealed a significant effect of time (Fig. 2a) over PVT activation in CON (F_3,72 = 5.45, p = 0.002), as well in all sub-nodes, i.e., thalamus (F_3,72 = 8.51, p < 0.001), insula (F_3,72 = 4.45, p = 0.006), and dACC (F_3,72 = 3.34, p = 0.024). However, there was no significant difference found between groups, neither the interactions between factors of time and groups. Voxel-wise result (Fig. 2b) exhibited the effect of time, confirming that the significant changes existed over the three CON sub-nodes, besides other noticeable regions reported in Table S3. PVT activation in CON decreased during early measurements (i.e., A1 and A2) after sleep, and this effect of time was more evident on thalamus and insula than on dACC, as observable from both Fig. 2a, b. By examining both the awake duration from sleep offset to the end of the sleep session and the amount of NREM 2&3 sleep prior to awakening, we captured their respective influences on the decreased activation of the CON at the network level, as well as focally in the thalamus (Fig. 2c, d) and insula (Fig. S2), specifically at A1, the earliest measurement after awakening. Specific to the Inertia group, shorter awake duration and longer NREM 2&3 sleep are correlated with lower beta weights of these regions during A1 PVT phase. After FDR correction over repeated correlations on four measurements, the significant positive and negative correlations are survived only at thalamus (Fig. 2c, d). In the PW group, no such correlation patterns are disclosed (Figs. 2c, d and S2). Moreover, the coefficient values were continuously weakened when the cutoff was loosened, and more participants from the PW group were included into the correlation analysis (Fig. S3a and S3b).

Fig. 2. SI-associated alterations of PVT activity in CON (n = 26) and its correlation with sleep and wakeup history.

a ROI-based mean beta weights across 4 sessions in the CON, and its three key sub-nodes, namely the bilateral thalamus, bilateral insula and dACC. b Time effect of PVT activation across 4 measurements based on the F-test of rm-ANOVA model (F > 6.011, 3dClustSim-corrected p < 0.05 with uncorrected p < 0.001, k > 54). Pearson correlation between PVT beta weight and c awake duration and d length of NREM stage 2 and 3. The left two scatter plots presented the planned correlation of the Inertia group (black dots, n = 15 independent samples) between beta weight of CON and thalamus at A1 with awake duration, and the inset statistics are uncorrected. Light gray dots indicate the PW group (n = 11 independent samples). Two line-plot panels on the right show the correlation coefficients across 4 sessions in both groups. The star signs indicated significance after FDR correction. * means p < 0.05, ** means p < 0.01. dACC=dorsal anterior cingulate; Pre=pre-sleep session; A1=1st, A2=2nd, A3=3rd session after sleep.

Critical role of thalamic activation at the early wakefulness

In the Inertia group, we found that averaged CON activation in PVT persisted across sessions, including insula and dACC, but except for that of thalamus, which was absent in A1 and A2 (Fig. 3a, top 4 rows). PVT also activated other essential regions related to vision and motor functions, such as the left postcentral gyrus, supplementary motor area, visual cortex, and cerebella. From the rm-ANOVA model (F > 6.53, 3dClustSim-corrected p < 0.05 with uncorrected p < 0.001, k > 66), a significant time effect was observed in the bilateral thalamus (Fig. 3a, last row; see Table S4 for other clusters). The ROI analysis (Fig. 3b, top panel) revealed a prominent SI effect, namely gradual recovery of declined brain activity, in both the CON and thalamus levels. Notably, the right inferior frontal gyrus (r-IFG) exhibited a continuing decline after awakening (Fig. 3a, b), which was different from the pattern of bilateral insula as shown above (Fig. 2b). Relatively, the PW group did not exhibit any significant difference in the PVT-associated activation of the averaged CON level or its subregions (Fig. S4). Figure 3c presents the correlation between brain activation and the PVT speed across the four sessions. The CON activation exhibited a positive association with PVT speed at A2 (r = 0.64, p-fdr < 0.05) and marginally at A3 (r = 0.52, p-fdr = 0.098, p-unc = 0.049), whereas the thalamus activity tended to correlate with the PVT speeds during the early wakefulness conditions of A1 and A2 (r = 0.55, p-fdr = 0.068, p-unc = 0.032 and 0.034 respectively). However, this positive brain–behavior relation during SI was not observed in the PW group (Fig. S4c).

Fig. 3. SI-associated alterations of PVT activity in the Inertia group (n = 15 independent samples) and its correlation with PVT performance.

a PVT activation across 4 sessions (FWE-corrected p < 0.05, k > 20) and effect of time based on the F-test of rm-ANOVA model (F > 6.53, 3dClustSim-corrected p < 0.05 with uncorrected p < .001, k > 43). b ROI-based mean beta weights across 4 sessions in the CON, the bilateral thalamus, and right inferior frontal gyrus (r-IFG). c Pearson correlation between beta weight and the PVT median speed in each session. Red line indicates significant correlation with p < 0.05, while the red dash line represents marginally significance 0.05 < p < 0.10.

In line with the above analyses, mediation analysis separately disclosed two pathways that thalamic activation during A1 PVT phase mediates the association between sleep history and alertness performance. Thalamic activation fully mediates (1) the association (Fig. 4a and Table S5) between awake duration and PVT median speed (indirect effect estimate=06, p = 0.036); (2) the association (Fig. 4b and Table S5) between length of NREM 2&3 sleep and PVT fSPD (indirect effect estimate = −0.006, p = 0.042). Other planned mediation analyses did not yield any significant mediation effect. Of note, the mediation model of thalamus activation mediating the relationship between length of NREM 2&3 sleep and PVT median speed was marginally significant (indirect effect estimate=059, p = 0.067); and the significant path c in this model (Table S5) suggested a negative direct effect without introducing the mediator (estimate = −0.008, p = 0.03).

Fig. 4. Model with thalamus activation as a mediator at PVT A1 phase in the effect of sleep history on alertness performance.

a Mediation from thalamus activation between awake duration and PVT median speed. b Mediation from thalamus activation between length of NREM 2&3 sleep and PVT fast mean speed (top 20 percentiles). In both panels, the solid black lines with arrow indicate significant effects. The line and gray lines suggest the non-significance of that path. Beta estimates and the standard errors in the parentheses are presented along the names for each path. Paths a, b, c, c’ and ab are standing respectively for the (1) effect from predictor variable on mediating variable, (2) effect from mediating variable on criterion variable, (3) the total effect from predictor variable on criterion variable, (4) the direct and (5) the indirect effects from predictor variable on criterion variable after introducing the mediating variable. * means p < 0.05, *** means p < 0.001. For the details of 95% confidence intervals, see Table S4.

Association between task activity and resting connectivity in CON during SI

Regarding the role of the CON in tonic alertness, Fig. 5a displays the FC conjunction results of the CON (FC_CON) across the four conditions (top panel), the time effect reflected by rm-ANOVA (middle panel, also see Table S6), and the index/degree of restoration change (RC, A1-A3) (lower panel, also see Table S7) in the Inertia group (|T|> 2.70, 3dClustSim-corrected p < .05 with uncorrected p < 0.005 and k > 152); we did not observe significant changes of within-CON connectivity after waking from sleep (Fig. 5a, b). Instead, FC_CON exhibited time-varying connectivity changes to the bilateral postcentral gyri and right medial temporal pole (Table S6). However, during the A2 and A3 sessions (Fig. 5c), the FC strength of r-IFG connected with the CON was positively correlated with PVT r-IFG activation (r = 0.62, p-fdr < 0.05). Again, no significant findings were observed in the PW group (Fig. S5). Besides FC, we estimated the correlation between the RC degree of PVT thalamus activity with that of ReHo and ALFF calculated locally at thalamus, but no significant relationship was observed between resting-state indices and task activation in both Inertia (Fig. 5d) and PW (Fig. S5d) groups.

Fig. 5. SI-associated alterations of resting-state functional connectivity (FC) in the Inertia group (n = 15 independent samples) and its correlation with corresponding PVT activation.

a Top: seed-based FC_CON (Z-transformed) from the conjunction across the 4 sessions (FWE-corrected, p < 0.05, k > 20); Middle: the significant FC_CON revealed by rm-ANOVA across the 4 sessions (F > 6.53, 3dClustSim-corrected p < 0.05); Bottom: the paired t-test for the RC degree (A1–A3, |T|> 2.70, 3dClustSim-corrected p < 0.05). b ROI-based mean FC strength across the 4 sessions in the CON, bilateral thalamus, and right inferior frontal gyrus (r-IFG). c Pearson correlation between FC and the corresponding PVT beta weight of each ROI. Red line indicates the correlation test is significant as the p value is less than 05 level after FDR correction. d Association between the thalamic connectivity change (ΔFC: A1–A3 or A2–A3) and thalamic activation change (Δβ: A1–A3 or A2–A3). In post-sleep A1 and A2 sessions, no significant correlation was found between the RC degree of PVT activation and the RC degree of each resting-state indices (i.e., FC connected with CON, ReHo, and ALFF). RC=Restoration Change.

Inter-network cross-talk between thalamus and FPN predicts the recovery from SI

Because the FPN presents a SI-related dynamic change after sleep offset²⁵, we also examined whether the coordination between the thalamus and FPN was involved in the recovery of tonic alertness function. The results of rm-ANOVA indicated that the FC_thalamus-FPN per se did not exert any SI effect or time effect in both groups (Fig. S6), and the FC_thalamus-FPN did not exhibit a significant relationship with PVT thalamus activity (Fig. 6a). However, it is noticeable that participants (Fig. 6a, black dots) who did not show SI effect with PVT on the individual level had a relatively higher FC_thalamus-FPN level at A1 than those who did have (light gray dots); while the opposite pattern is revealed at the sub-panel of A3. Through correlation analysis, an inverse relationship was observed between the RC degree of FC_thalamus-FPN and the RC degree of PVT thalamus activity immediately upon awakening (A1), but not in the A2 session (Fig. 6b). In terms of the behavior level, the RC degree of FC_thalamus-FPN also exhibited a significant positive correlation with the RC degree of PVT fSPD in the A1 session (Fig. 6d), although the FC_thalamus-FPN per se did not exhibit a significant association with fSPD in either of the conditions (Fig. 6c). Again, such relationship was not significant in the PW group (Fig. S7b and S7d).

Fig. 6. Functional connectivity between thalamus and frontoparietal network (FC_thalamus-FPN) and its relationship with PVT thalamic activation and PVT performance during SI in the Inertia group (n = 15 independent samples).

a Scatter plots between FC_thalamus-FPN and thalamus PVT activation across the 4 sessions; b Association between the RC degree of FC_thalamus-FPN (ΔFC: A1–A3 or A2–A3) and the RC degree of thalamus PVT activation (Δβ: A1–A3 or A2–A3). Significant negative correlation was detected at the A1 condition (blue line); c Scatter plots between FC_thalamus-FPN and PVT fSPD across the 4 sessions; d Association between the RC degree of FC_thalamus-FPN (ΔFC: A1–A3 or A2–A3) and RC degree of PVT fSPD (ΔfSPD: A1–A3 or A2–A3). Significant positive correlation was detected at the A1 condition (red line). Participants with evident SI effect are displayed in gray dots, and those without SI effect are shown in black dots. RC=Restoration Change.

Discussion

Beyond the circadian rhythm and homeostatic drive, the sleep inertia (SI) serves as the third process to better describe the fluctuation in sleepiness and alertness³⁰. The SI effect is typically evaluated through behavioral tests, such as PVT, which measures SI-related hypo-vigilance. While most neuroimaging studies of SI have focused on resting states before and after sleep, task-based fMRI identifies brain regions specifically engaged in a task, whereas rs-fMRI explores the spontaneous properties of brain inter-network integrities rather than specific cognition. However, technical challenges have led to a trend of observing SI effects in resting states without task engagements, compounded by the efforts from actively restoring the cognition deficiencies and passively adapting to wakefulness. To address these challenges, we used a three-way synchronized system to link task performance and brain mechanisms during SI. We successfully mapped the dynamic activity of the CON and thalamus during SI (Fig. 3b), which corresponded to task performances in a time-varying manner (Fig. 3c). Especially at the earliest A1 test point, within CON thalamus is the pivotal mediator that conveys the impact from prior sleep and awake history and expresses on the measured alertness performance. The FC_CON did not exhibit any time effect (Fig. 5b) or association with PVT-induced brain activity (Fig. 5c). However, the recovery rate of cross-network FC_thalamus-FPN was negatively correlated with PVT thalamus activity (Fig. 6b) and PVT fSPD (Fig. 6d) upon wakening, suggesting that the FPN plays an indirect role in counteracting alertness impairments during SI by interacting with the thalamus.

Executing PVT requires the ability to continuously monitor the external environment, supported by the tonic alertness function, which has been shown to be impaired during SI^5,14. However, the phenomenon based on reaction speed was not observed in the current nocturnal nap design, possibly due to confounding factors such as the short duration of sleep, lack of slow-wave sleep in the MRI environment, mixture of self- or forced-awakening¹⁷, or the limited effect size of SI when assessed using vigilance tasks³¹. Besides, the PVT performances exhibited subtle differences between two groups, where the Inertia group exhibited long sleep time (with slow-wave sleep) and short awake time before A1, but the PW group exhibited short sleep time (without slow-wave sleep) and long awake time before A1. The Inertia group sustained their PVT speed across repeated measures and even increased in fast responses (fSPD) (Fig. 1d), indicating the dynamic dissipation of SI. In contrast, the PW group showed a decreasing trend in post-sleep PVT performance (Fig. S1a) under the conditions of fragmented sleep and prolonged wakefulness, implying accumulated sleepiness when they underwent the repeated measures. From these observations, it is noted that both factors, the prior sleep architecture and the post-sleep wakefulness, contribute to the recovery of tonic alertness and brain functionality during SI.

Previous studies^26,28, have proposed that the CON plays a critical role in supporting tonic alertness by continuously suppressing internal and external distractions⁷. In well-rested individuals, higher tonic alertness demands are associated with increased CON activity^26,28. Consistently, the Inertia group in our study showed a positive correlation between CON activation and PVT speed, emerging from A1 to A2 and diminishing by A3 (Fig. 3c). This temporal pattern may reflect the gradual restoration of CON–PVT correlation as a progressive dissipation of acute sleep inertia. Notably, this correlation was not evident immediately upon awakening (A1), suggesting that the dynamic uncoupling between CON activity and PVT performance may represent one neural signature of sleep inertia. This interpretation aligns well with EEG findings showing reduced beta-band (17–24 Hz) activity during early wakefulness^7,10,—an electrophysiological marker previously linked to CON function³². In contrast, the PW group did not show evidence of CON–PVT positive correlation at their A1 time point (Fig. S4c), different from the typical positive association between CON activity and tonic alertness. Notably, their A1 occurred, on average, nearly one hour after awakening, where the acute SI effects should have largely dissipated. Thus, the absence of CON–PVT positive correlation in the post-sleep sessions of the PW group was speculated to be affected by the prolonged wakefulness (or cumulative sleepiness) during their circadian trough^33–35. In short, the proposed SI dissipation in the dynamic relations between PVT and CON activity may offer a coherent ecological framework and account for both pre-awakening sleep architecture and the duration of post-awakening wakefulness.

Our study did not find any time-varying changes in the FC within CON, except for an increase in within-network thalamic connectivity during late wakefulness (Fig. 5a; all FC differences compared between A1 and A3 are shown in Table S7). Meanwhile, we did not observe any correlation between the RC degrees of task activities and resting-state indices in the thalamus during post-sleep sessions (Fig. 5d), or them per se (Fig. 5c). This lack of correlation is unexpected since previous studies have shown a positive association between task activation and resting-state indices during wakefulness^36,37. This finding suggests that the SI transition period is unique and exhibits an uncoupling condition between task-specific brain activity and spontaneous resting-state properties during the recovery from SI.

During early wakefulness, the bilateral thalamus in the CON region showed decreased PVT activation and that tended to correlate with PVT response speed (Fig. 3c). This suggests that thalamus activation is critical for maintaining tonic alertness during severe periods of SI. The thalamus is known as a relay hub with high and reciprocal thalamocortical connectivity, gating the external input and mediating communication throughout the cerebral cortex^38,39. Previous studies have shown that thalamus activity is positively associated with changes in sleep-wake transition⁴⁰ microsleep⁴¹, sleep deprivation⁴², narcolepsy⁴³, and arousal/vigilance fluctuation^44,45. Our findings support the idea that thalamus activity, although attenuated, plays a critical role in for maintaining tonic alertness during severe SI periods.

Moreover, we confirm the mediator role of the disclosed neural correlate of SI in linking the prior sleep history and awake duration in a time-sensitive manner. Immediately on awakening (A1), the observed thalamic activation was positively associated with the awake duration (Fig. 2c), reflecting a linear brain recovery during SI and being consistent across participants in the SI group⁴⁶. Relatively, participants in the PW group, who had substantially longer awake durations, did not show this association (Fig. S3). This finding aligns well with the literature indicating that SI dissipates rapidly, often within ~30 min⁶, which highlights the importance of the awakening time on SI to exclude individuals who may have already moved beyond the acute SI window (by synchronizing the awakening timing through a forced-awakening protocol or by applying a minimum awake-duration cut-off as in the current study). In contrast, the negative correlation (Fig. 2d) between thalamic activation and NREM 2&3 sleep was more stable by excluding participants with extended wakefulness, yet only detectable at the A1 session (Fig. S3). Prior work has shown that the overall duration of sleep or a nap has been shown to modulate the severity of SI-related behavioral impairment^47,48, particularly when greater amounts of slow-wave sleep occur before awakening^14,49. These findings support the conceptualization of SI as a transient state shadowed by residual sleep-like features, such as persistent slow cortical activity^32,50, and segregation of large-scale brain networks^24,51, as suggested by resting-state neuroimaging evidence. Within our task-based framework, we further propose that the first attempt to perform an alertness-demanding task may actively counteract these lingering sleep features, potentially accelerating the dissipation of SI. This hypothesis warrants further confirmation, particularly in experimental designs that can disentangle the respective contributions of prior sleep architecture and immediate wakefulness history when interpreting early post-awakening neural and behavioral dynamics.

Compared to the awake duration factor, the composition of NREM 2 and 3 sleep was accumulated in a longer temporal span far from the PVT measurement points, which may explain why its mediation model with PVT median speed was marginal (Table S5) despite a total negative effect was still detected. However, when the optimal PVT performance from fSPD was used, the mediation path for NREM 2 and 3 sleep as the predictor variable turned significant (Fig. 4b). This implies that fSPD index, compared to median or mean of a whole PVT trials body, is more sensitive to subtle influence from the factor of prior sleep pattern, such as the NREM 2 and 3 sleep. In the scenario with complex nocturnal nap experiment of current design, this speculation was further supported by the facts that behaviorally only fSPD expressed the recovery sign from SI, and the recovery contribution from FPN could precisely predicted the changes of fSPD along SI dissipation. Altogether, the evidence advocates fSPD as a more sensitive index to monitor SI effect for future studies, especially when the effect is potentially small or contaminated by confounders, such as sleepiness from sustained attention.

Previous research on SI²⁵ led us to investigate the relationship between FC_thalamus-FPN and subsequent PVT performance and brain activity during early wakefulness. FPN is known to be involved in attentional processing⁵² and is considered a flexible hub for higher cognitive control⁵³. Our findings suggest that individuals with stronger FC_thalamus-FPN connectivity in the A1 session could be less affected by SI (black dots in Fig. 6a, c), more likely to carry a quick recovery of thalamic activation (neural substrate of SI) at A3 (negative correlation in Fig. 6b). For those with evident SI effect (gray dots in Fig. 6a, c), the rebound of fSPD (behavioral sign of SI recovery) might even occurred before the recovery of their thalamic activation. The accompanied change in fSPD (positive correlation in Fig. 6d) suggests that participants perhaps have certain degrees of maneuverability to speed up their recovery by intentionally leveraging top-down control through FPN, which resulted in the prompt assistance to a subset of optimal alertness trials. Specifically, this active “redialing” attempt (between FPN and thalamus) may boost the readiness to attend external sensory stimuli, thus overcoming transient cognitive impairments caused by SI. Nevertheless, it is currently unclear whether mainly the differences in how quickly people can wake up (a trait factor) or their willingness to wake up rapidly (a state factor) or both of these factors contribute to the variation in FC_thalamus-FPN among individuals.

During early SI (A1), the r-IFG, a core node in the FPN and a juncture bridging FPN and CON, exhibited relatively high task activation (Fig. 3b). The FC between this region and the CON predicted its activation level during the post-sleep PVT sessions (Fig. 5c), and this pattern was consistent with studies of sleep deprivation where increased r-IFG activity was compensatory^54,55, and not positively correlated with task performances^55,56. Previous research has suggested that r-IFG activity provides phasic attention reorientation when attention tasks become challenging⁵⁷, hinting at a similar compensatory role of r-IFG activity during waking processing in SI. However, further studies are needed to verify this hypothesis.

This study contains several limitations. First, participants were instructed to self-terminate the sleep session freely based on their subjective perception, resulting varied wakefulness length in the end across participants, which in turn may have affected the precision of aligning SI dynamics based on our measurements. Future studies could consider a continuous alertness-monitoring paradigm, such as a visuomotor tracking task⁴¹, to improve alignment precision across participants. Second, the sample size was limited based on the technical difficulties, and confounding factors^14,58, such as sex and age, were not explored. The sex distribution in our data was nearly identical in both groups (Inertia: 11/15 male; PW: 8/11 male, both ~73%), so any sex-related influence is likely balanced across conditions. Future studies are warranted to explore the sex differences in SI. Third, the resting and task data, thus the indices generated, were acquired with a slight time delay of several minutes, which remains a time-varying problem given the rapid-changing brain functionality during SI and the effect of task execution on subsequent resting state⁵⁹. Lastly, the study only focused on the tonic alertness using PVT, and the task-specific involvement of other brain networks in SI recovery (e.g., DAN with visuo-spatial attention task⁶⁰) remains unexplored and warrants further investigation.

Methods

Participants

A total of 58 young adults of East Asian descent were recruited from an online pool and screened through in-person interview. Participants were selected who had regular sleep-wake schedules, minimal sleep disturbances, no habit of midday napping, no history of neurological and psychiatric disorders. This research protocol was approved by the Research Ethics Committee of National Taiwan University (No. 201512ES054). All participants provided their written consent forms prior to the experiment, and all ethical regulations relevant to human research participants were followed.

Experimental procedure

Participants underwent a battery of assessments at least three days before the experiment, including the actigraphy for assessing regular bedtime and four standardized questionnaires evaluating sleep habits, circadian preference, daytime sleepiness, cognitive status. These comprised the Morningness–Eveningness Questionnaire (MEQ), Pittsburgh Sleep Quality Index (PSQI), Epworth Sleepiness Scale (ESS), and the Mini-Mental State Examination (MMSE), with individuals scoring below 26 on the MMSE excluded due to potential cognitive impairment. Bedtime regularity was determined using an actigraphy worn on the non-dominant hand (SOMNOwatch^TM plus, SOMNOmedics GmbH, Randersacker, Germany) for at least three consecutive days. On the day of the experiment, participants were instructed to refrain from napping and consuming caffeinated or alcoholic beverages. They arrived at the MRI center of National Yang Ming Chiao Tung University 2 h before their regular bedtime to align the sleep session as close as possible to their normal sleeping time, minimizing the influence of circadian and homeostatic processes on sleep and SI effect^3,6. After EEG cap preparation, participants were advised to use the restroom before entering the scanning room. In the MRI room, they lay on memory foam with customized cushions, used earplugs, and wore a pneumatic belt and finger-tip oximeter to record respiration and cardiac pulsations. The main sleep session could last up to approximately 3 h due to technical limitations.

Before the sleep session, the experiment began with a 6-min anatomical scan, followed by a 5-min open-eye resting-state fMRI scan and a 6-min PVT fMRI scan conducted before sleep session (Pre). During the resting-state scan, a white cross on a black background served as the fixation point. The PVT was designed with 72 trials, a red square target appeared on the center of the screen for a maximum dwelling duration of 1 s at pseudorandom inter-trial intervals (1, 3, 5, or 7 s) using Psychtoolbox and MATLAB. Participants were instructed to press a button using their right thumb as soon as they saw the red square, and there was no performance feedback. The combination of resting-state and PVT scans was repeated in the same order at approximately 5, 20, and 35 min after the sleep session ended, corresponding to A1, A2 and A3, respectively (Fig. 1a). By using this repeated-measure design, we were able to prolong the monitoring of SI effects for up to 50 min (i.e., the end of A3), even though SI effects typically dissipate 30 min after waking⁴. To isolate the recovery effect from SI, the final measurement (A3) was treated as the main SI-relieved wakefulness condition, rather than pre-sleep wakefulness, because of its relatively similar circadian phase and sleep pressure as the SI-severe A1 condition.

Because the sleep time was an uncontrollable factor in this protocol, we could only confirm that participants were actually experiencing SI when we received their sleep-staging results (details described below) after finishing all EEG preprocessing procedures. Eventually, 32 of 58 participants were excluded because of their inability to fall asleep in the scanner (n = 6); excessive EEG noise level for sleep scoring (n = 5), excessive motion (n = 5), total sleep time of less than 5 min (n = 1), interruption by the need to urinate (n = 2), and incomplete data acquisition (n = 12). The remaining 26 participants slept for average 43.90 ± 43.74 min inside the scanner, with a sleep-onset latency of 18.19 ± 15.85 min and 24.44 ± 31.76 min spent in NREM stages 2 or 3. Furthermore, inevitable scenario occurred when some participants woke up in the middle of the sleep session and failed to sleep again. Such cases with short sleep duration and prolonged wakefulness shall not be considered as the sleep inertia for that they have passed the SI window by the time they performed the first post-sleep (i.e., A1) session. Therefore, we segmented the 26 participants into two groups: 15 participants (age: 24.5 ± 0.66; 11 males) were classified into the Inertia group (Table S1) for their immediate awakening from sleep session, and 11 participants (age: 24.5 ± 0.73; 8 males) into the Prolonged Wakefulness (PW) group (Table S2) for their extended time in wakefulness. Because SI effect typically dissipates within ~30 min⁴, the participants in the PW group were selected for those who had more than 20 min of continuous wakefulness prior to the cessation of the sleep session (based on sleep scoring) without the acute SI effect. Importantly, both groups underwent the identical experimental procedure during the scan, and the grouping was carried out based on the sleep-scoring performances retrospectively.

Psychomotor vigilance task

Considering the sensitivity of all indices possibly derived from PVT⁶¹, we calculated the median speed of responses (1/RT) as the estimate of tonic alertness across task sessions. With additional interest on within-session fluctuation of performance, standard deviation of speed, and mean response speeds from fastest 20% (fSPD), intermedia 60% (iSPD), and slowest 20% trials (sSPD) were then obtained. A repeated measures ANOVA (rm-ANOVA) test was applied on all indices of interest separately in the Inertia and PW groups. The paired two-tailed T-test was planned as post hoc test, and the results were then corrected through multiple comparisons using the false discovery rate (FDR) method. The two-way rm-ANOVA, 3 (fSPD, iSPD, sSPD) × 2 (A1/A2, A3), was conducted on the aforementioned results to determine the respective contributions to the significant increase in the standard deviation of speed in A3 session (Fig. 1c).

Simultaneous EEG-fMRI recordings

Simultaneous EEG–fMRI signals were recorded using an MR-compatible EEG system (Brain Products, Gilching, Germany) and a 3 T Siemens Tim Trio system (Erlangen, Germany) with a 12-channel head coil. The 32-channel EEG cap contained 27 EEG channels, two electrooculography (EOG) channels, two electromyography (EMG) channels, and one electrocardiogram (ECG) channel. High-resolution T1-weighted anatomical images (3D MPRAGE; matrix size = 192 × 192 × 176; 1 mm³ isotropic voxels; flip angle [FA] = 9°; repetition time [TR] = 1900 ms, echo time [TE] = 2.28 ms, and inversion time [TI] = 900 ms) were obtained prior to the functional scans for localization. To carry out the synchronization between the 3 computers (EEG, MRI, and task stimulation), the EEG signals were synchronized with MR triggers using a Brain Products specialized cable, and in-house MATLAB scripts were used to send the markers of stimuli and responses to the EEG reception computer.

Functional scans were subsequently obtained utilizing a single-shot, gradient-recalled echo planar imaging (GE-EPI) sequence aligned along the AC–PC line for whole-brain coverage in both resting-state, PVT, and the sleep sessions. The participants in the first patch were scanned with the following GE-EPI parameters: field of view = 220 mm, in the first batch with 15 participants, the fMRI parameters were TR/TE/FA = 2750 ms/ 35 ms/ 83°, matrix size = 64 × 64, 35 slices with 3.4 mm thickness in the resting-state session, and TR/TE/FA = 2000 ms/ 35 ms/84°, matrix size = 64 × 64, 31 slices with 4 mm thickness in the PVT scans. In the second batch with 11 participants, another fMRI setting was applied across task and resting scans: TR/TE/FA = 2610 ms/ 30 ms/ 77°, matrix size = 64 × 64, 32 slices with 4 mm thickness. Due to a scanner-imposed upper limit on the number of volumes per run (4096), the maximum continuous recording of sleep session was constrained approximately 3 h.

Data preprocessing

Offline artifact removal was implemented for EEG data on Vision Analyzer 2.0 (Brain Products), following by step orders of marker timing, upsampling (to 5000 Hz), correction of gradient (adaptive average subtraction), and removal of cardioballistic artifacts. A sliding template algorithm referred from ECG was adopted for calculating respective pulse artifacts (time-locked to R peak), then subtracted from 27 EEG channels. After the two artifact-removal steps, all-channel data were down-sampled to 250 Hz in Analyzer and then low-passed at 40 Hz in the EEGLAB⁶². Only EEG channels were further re-referenced through the average reference method. Another version of scoring-ready data was also provided, on which potential component of physiological noise sources (ECG, EMG, EOG) were removed additionally after the decomposition from independent component analysis using EEGLAB. Thereafter, a licensed sleep technician compared then selected a version of the data, and manually scored sleep stages according to American Academy of Sleep Medicine (AASM) criteria on every 30-s epochs⁶³.

Preprocessing for all MRI data was accomplished by AFNI⁶⁴ and FSL⁶⁵. As an anatomical reference, T₁-weighted images were skull-stripped, then registered with functional data, and normalized to standard Montreal Neurological Institute (MNI) template in an isotropic resolution of 2 × 2 × 2 mm³. For the resting-state dataset, cardiac and respiratory fluctuations were modeled and removed by RETROICOR method⁶⁶. The preprocessing procedure for resting-state dataset further including motion correction, despiking, 3rd-order detrending, confounding nuisance regressors (6 motion parameters, pulse, respiration, white matter, and cerebrospinal fluid signal). Afterward, rs-fMRI data were low-pass filtered with 0.1 Hz and smoothed with a Gaussian kernel (FWHM = 6 mm). For PVT sessions, images underwent motion correction, slice-timing, and smoothing with the same parameter as the resting-state dataset. Functional images from all sessions were coregistered the T₁-weighted images and then normalized to the MNI space accordingly.

Task-based brain activity and resting-state functional connectivity

In the PVT data, the first-level activation was estimated using an event-related generalized linear model for all the hit trials. The trials' onset time was modeled with 0 duration and temporally convolved with the canonical hemodynamic response function. In addition to the stimuli regressor, one constant, one linear trend, and six head motion regressors were included in the design matrix. After model fitting, the beta maps related to PVT events were generated for each participant and then collected for later analysis of random effects. In the resting-state data, we applied seed-correlation analysis to obtain voxel-wise functional connectivity (FC) of the CON during 4 sessions for each participant²⁸. For later statistical comparison at the second level, the Fisher-z transform was applied to the correlation maps, and the default significance level was set as two-tailed p < 0.05 with multiple comparison methods. Two other indices, the amplitude of low-frequency (ALFF, 0.01–0.1 Hz) and Regional Homogeneity (ReHo, 27 voxels in neighborhood), were also calculated for measuring local properties of rs-fMRI data.

Voxel-wise and ROI analyses

In group analysis, the significant functional activation maps at each session were obtained through the one-sample t-test, with family-wise error rate (FWE) correction and cluster size > 20 voxels (FWE-corrected p < 0.05). One-way rm-ANOVA models were applied to PVT activation and resting-state functional index maps across four sessions, with correction using the 3dClustSim method considering the autocorrelation (at uncorrected p < 0.001 with cluster level p < 0.05). A paired t-test was used for planned contrast between early SI (A1) and latest wakefulness (A3) states to identify maximal inertia effect. This calculation of subtracting A1 from A3 was conducted for both neuroimaging and behavioral indices, referred to as an index of restoration changes (RC) during SI recovery. This pipeline was used firstly including all the participants, then separately applied to the Inertia and PW groups.

Regions of interest (ROI) analysis was conducted on nodes defined by the Power-264 atlas⁶⁷, including the bilateral thalamus (#224, #225), bilateral insula (#55, #60), dorsal anterior cingulate cortex (dACC, #47, #54), and right inferior frontal gyrus (r-IFG, #209). In addition, we constructed a network-level CON mask comprising all nodes labeled as belonging to the CON or located within the thalamus, based on our hypothesis regarding its close involvement in tonic alertness. To test the difference between groups and the effect of time over PVT activation (beta weight), a mixed-effects ANOVA model (N = 26, within-participant factor: four measurements; between-participant factor: Inertia or PW groups) was used. For the two groups, respectively, the rm-ANOVA test was used to obtain the effect of time over beta weight in PVT sessions and Fisher Z-transformed FC in resting-state sessions. In post hoc analysis, paired T-test with FDR correction was conducted for significant findings.

Statistics and reproducibility

Pearson correlations were used to explore interrelationships between sleep indices (awake duration from sleep offset and sleep length of NREM 2 and 3), PVT median speed, and fMRI indices in all ROIs (PVT beta weights and resting-state FC). FDR correction method was implemented for repeated correlation approaches across different measurement points (i.e., Pre, A1, A2, and A3) or different sample sizes when pooling participants from two groups. Adjusted p values were presented in figures. Planned mediation analyses were conducted with eight alterations, combining from two predictor variables (X: awake duration or NREM 2 and 3 sleep length), two mediating variables (M: thalamic or CON’s activation at PVT A1), two criterion variables (Y: median speed or fSPD at PVT A1). Statistical inference about the path significance was relied on bootstrapping approach with 10000 re-samples in 95% confidence intervals.

Pearson correlation was also used to assess the associations of the changes (RC degrees from A1 to A3) between tasks PVT beta weight and the three resting-state indices. The mean FC connecting bilateral thalamus ROI with all the other nodes in FPN was examined for the potential effect of SI with rm-ANOVA test. For the post hoc comparison the paired T-tests were used with FDR correction. The relationship of the RC degrees during SI between the FC_thalamus-FPN (derived from resting state) and two signs (cerebral and behavioral) for SI recovery, namely the PVT beta weight of thalamus ROI and the PVT fSPD at A1 and A2 phases, was also examined separately.

Ethical statement

The study protocol was approved by the Research Ethics Committee of National Taiwan University (No. 201512ES054). All participants provided their written informed consent to participate in this study, and all ethical regulations relevant to human research participants were followed.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.

Author contributions

Conception and experiment design: Wu CW, Chen CC, Lee HC; Experiment operation and data collection: Chen S, Kung YC; Data analysis: Chen S, Hsiao FC; Manuscript drafting: Chen S, Kung YC, Wu CW; All authors reviewed and approved the manuscript.

Peer review

Peer review information

Communications Biology thanks Alireza Chamanzar and Helene Vitali for their contribution to the peer review of this work. Primary Handling Editors: Joao Valente. A peer review file is available.

Data availability

The data that support the findings of this study are publicly available in Figshare⁶⁸. The raw MRI datasets and data generated in the analysis are available from the corresponding author upon reasonable request as part of scientific collaboration with adherence to standards of good scientific practice, although restrictions may apply due to privacy reasons and ongoing research projects. Data sharing will require an MTA and is limited to non-commercial use.

Code availability

The customized codes for this study are publicly available in Figshare⁶⁸.

Competing interests

All authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Supplementary information

The online version contains supplementary material available at 10.1038/s42003-026-09839-w.