The prevailing view of sleep has long been that it represents a state of profound disengagement from the external world, lacking reactivity to external stimuli. However, accumulating evidence from event-related potentials (ERPs) and intracranial recordings challenges this 'sleep disconnection' dogma. Studies have demonstrated that the sleeping brain retains the capacity for low-level sensory processing across different sleep stages. Furthermore, research indicates that sleepers can process symbolic stimuli at various cognitive levels, encompassing semantic and decision-making processes. The impact of sleep on learning is also significant; learning-related sensory cues presented during sleep can enhance subsequent recall, and new associations can be formed during sleep, influencing behavior even days later. Recent studies have shown the possibility of word-association learning during sleep, potentially generalizing to wakefulness in a cross-modal fashion. While much of this sensory processing during sleep is presumed to be automatic and unconscious, some studies suggest the integration of sensory stimuli into dream content, pointing to the possibility of conscious processing at times. However, a lack of single-trial evidence of stimulus integration during sleep has hindered a comprehensive understanding of the neurophysiological mechanisms underlying this complex phenomenon. The ability to obtain real-time behavioral responses as indicators of subjective experiences would greatly facilitate analysis of brain dynamics associated with sensory and cognitive integration on a trial-by-trial basis. Behavioral responses, previously considered exclusive to wakefulness, are typically excluded from sleep studies or not collected at all. While rare studies have shown manual responses during N1 sleep, the loss of limb muscle tone during deeper sleep stages could mask responses. Facial muscles, less affected by muscle atonia, offer a more suitable measure of behavioral responsiveness. For example, eye movements during REM sleep can signal lucidity in lucid dreamers, and previous work has demonstrated that lucid dreamers can respond to queries during REM sleep using a combination of eye movements and facial muscle contractions. This research builds upon this methodology to explore stimulus integration at the behavioral and neurophysiological levels, initially focusing on lucid dreaming and then extending the investigation to other sleep stages. Participants with narcolepsy were included, due to their characteristic excessive daytime sleepiness, short REM sleep latency, and high frequency of lucid dreams, making them an ideal population for observing multiple lucid dreams in a laboratory setting. Healthy participants (nonlucid dreamers) were also recruited to provide a comparison group.

Prior research has shown evidence of sensory processing during sleep. Studies using ERPs and intracranial recordings have shown that low-level sensory processing is preserved across various sleep stages (Bastuji & García-Larrea, 1999; Ruby et al., 2008; Strauss et al., 2015; Issa & Wang, 2011; Nir et al., 2015). Further work has demonstrated that sleepers can process symbolic stimuli at different cognitive levels, including semantic and decisional stages (Kouider et al., 2014; Andrillon et al., 2016; Andrillon & Kouider, 2020; Xia et al., 2023). The impact of sensory cues presented during sleep on memory consolidation has also been extensively studied, with evidence showing that learning-related sensory cues presented during sleep positively impact subsequent recall (Oudiette & Paller, 2013; Rudoy et al., 2009; Rasch et al., 2007). Moreover, new associations can be learned during sleep, potentially influencing behavior even a week later (Arzi et al., 2012; Arzi et al., 2014). The possibility of word-association learning during sleep and its generalization to wakefulness has also been explored (Züst et al., 2019; Ruch et al., 2022; Koroma et al., 2022). However, the question of conscious awareness during these processes remains a significant area of debate, with conflicting findings reported. Some research points to the incorporation of sensory stimuli into reported dream content (Valli & Hoss, 2019), suggesting conscious processing, but the lack of robust single-trial evidence of stimulus integration and behavioral responses during sleep hinders a deeper understanding of these phenomena.

The study investigated behavioral responses to auditory verbal stimuli across different sleep stages in two groups: 27 participants with narcolepsy (NP) and 21 healthy participants (HP). Polysomnography (EEG, EOG, EMG) continuously monitored sleep stages. During daytime naps, words and pseudowords were presented in a pseudorandomized order, with 1-min stimulation (ON) and 1-min nonstimulation (OFF) periods alternating (Fig. 1). Participants performed a lexical decision task, frowning (corrugator muscle) or smiling (zygomatic muscle) three times depending on the stimulus type (word vs. pseudoword). Facial EMG recorded these responses. After each nap, participants reported their mental content, lucidity, and task recall, and completed an old/new recognition task. Naps were classified as lucid or nonlucid based on self-report; lucid dreamers also signaled lucidity with a frowning-smiling sequence. Responsiveness was visually assessed by inspecting corrugator and zygomatic EMG, blind to sleep stage and stimulus presentation, and validated by an automatic algorithm. Response rates were compared between ON and OFF periods for each sleep stage, and accuracy was calculated for responsive trials (Fig. 3). Reaction times (RTs) were analyzed for correct responses, comparing words and pseudowords and examining the effect of sleep stage (Fig. 4). Lucid and nonlucid REM sleep were compared behaviorally and subjectively (Fig. 4). EEG spectral analysis during baseline and poststimulus periods examined sleep stage characteristics in responsive trials (Fig. 5). Time-frequency analysis and response-locked ERPs investigated brain processing differences between responsive and nonresponsive trials (Fig. 5). EEG markers distinguishing high and low cognitive states (spectral, connectivity, complexity measures) were computed in the prestimulus period to explore their predictive power on responsiveness (Fig. 6). A random forest classifier was trained on these markers in NP to predict responsiveness in both NP and HP (Fig. 7). Finally, lucid REM sleep was further analyzed, comparing electrophysiological markers in responsive and nonresponsive trials, and comparing lucid and nonlucid REM sleep, including the use of temporal generalization to explore conscious processing (Fig. 8).

The study demonstrated that behavioral responses to auditory verbal stimuli were possible across most sleep stages, including N2 and nonlucid REM sleep, in both NP and HP (except N3 in HP). Response rates were significantly higher during ON versus OFF periods across wakefulness and sleep stages (Fig. 3). Participants exhibited significantly greater than chance accuracy in performing the lexical decision task during all sleep stages examined (Fig. 3). Reaction times to pseudowords were significantly slower than to words in both groups, indicating high-level cognitive processing (Supplementary Fig. 13). Participants with narcolepsy displayed higher response rates than HP across all sleep stages during ON periods. In lucid REM sleep, response rates were also higher in ON periods, with accuracy above chance. Reaction times were significantly longer in lucid REM sleep than in other sleep stages. Electrophysiological evidence supported the observation of bona fide sleep during responsive trials, with spectral analyses reflecting typical sleep stage characteristics (Fig. 5). Time-frequency analysis revealed that responsiveness was associated with increased α and β band activity in frontal electrodes, consistent with motor preparation potentials (Fig. 5). EEG markers of high cognitive states (increased complexity and faster oscillations) were higher in responsive trials compared to nonresponsive trials across sleep stages (Fig. 6). A random forest classifier using these markers predicted responsiveness with above-chance accuracy in both groups, particularly when considering only correct responses (Fig. 7), indicating that cognitive processing rather than simply motor capacity drives the EEG differences. In lucid REM sleep, no significant differences were found in these EEG markers between responsive and nonresponsive trials, suggesting a ceiling effect in high cognitive state markers (Fig. 8). Temporal generalization decoding revealed a square-like pattern in stimulus-related brain activity during lucid REM sleep, consistent with a neural signature of conscious access (Fig. 8).

The findings offer compelling evidence for transient windows of sensory connection and high-level cognitive processing during sleep, extending beyond previously documented instances in sleep onset or lucid REM sleep. Behavioral responses occurred within a global background of sleep brain activity, but involved localized activations indicative of cognitive and motor processing. The accuracy and word/pseudoword RT differences suggest that stimuli were processed at a high cognitive level. The association of responsiveness with specific brain dynamics (increased complexity and faster oscillations) that predict responsiveness suggests that a certain level of brain activity is necessary for interaction with external stimuli during sleep. The similarities in EEG marker changes in responsive trials between NP and HP, and the ability of a classifier trained on NP data to predict responsiveness in HP, suggest that these transient windows of reactivity are a general feature of sleep. The higher frequency of such events in individuals with narcolepsy could be attributed to several factors: an enhanced capacity to maintain environmental awareness during sleep, reduced muscle atonia, or an increased tendency for local wake events. The presence of neural signatures of conscious access during lucid REM sleep, combined with subjective reports, strongly supports the conscious processing of external stimuli in this state. The question of consciousness during nonlucid responsive states is open for further investigation, but several lines of evidence suggest the possibility of conscious processing in N2 and nonlucid REM sleep. The findings present a more complex model of the wake-sleep continuum, suggesting that states of consciousness might fluctuate within traditionally defined stages.

This study demonstrates that sleeping humans can transiently integrate external information at a high cognitive level and behaviorally respond, even in non-lucid sleep stages. This challenges the traditional view of sleep as a state of complete disconnection. Specific brain dynamics predict this capacity for responsiveness. The findings hold significant implications for understanding consciousness and cognitive processes during sleep, and pave the way for real-time communication with sleepers, enabling further investigation into sleep-related mental processes. Future research can investigate aspects like metacognition during responsive moments, the extent of new information acquisition during these states, and the causal relationship between neural states and responsiveness.

The study primarily relied on visual inspection of EMG for response detection, although this was validated by an automatic algorithm. It focused on daytime naps, limiting the assessment of N3 sleep and nighttime sleep. The EEG montage may not have been optimal for connectivity measures, and lucidity was based on self-report in NP, requiring confirmation in HP. The study only obtained lucid naps in patients with narcolepsy.