Sleep onset is a creative sweet spot

The ability to think creatively is paramount to facing new challenges, but how creativity arises remains mysterious. Here, we show that brain activity common to the twilight zone between sleep and wakefulness (nonrapid eye movement sleep stage 1 or N1) ignites creative sparks. However, hitting the creative sweet spot requires individuals balancing falling asleep easily against falling asleep too deeply.

Behavioral task

Participants were presented with eight-digit strings, composed of three possible numbers (1, 4, and 9), and given a task to find the final solution of each string.

• The task consisted of transforming the string into a response string, through a sequential application of two rules: The “same” rule (if two successive digits are the same, the response is this digit) and The “different” rule(if they are different)
• To come up with the solution, participants had to apply these rules in a stepwise manner, starting with the first two digits and then using their first response together with the next digit to determine the second response and so on until the end of the string.

Key Findings

The discovery of a hidden rule is 2.7 times more likely after spending only 1 min of N1 during an incubation period, compared to a similar period of quiet rest including only wakefulness.

• Spectral analyses substantiate these findings and unravel a “creative sweet spot,” consisting in a medium level of alpha and a low level of delta. This creative sweet spot largely overlaps with the standard N1 stage, but not always, as it was also identified in wake solvers who never entered N1.

Participants

103 healthy participants (73 females, age 23.23 ± 3.58 years) recruited, screened out for exclusion criteria such as excessive daytime sleepiness, history of sleep, and neurological or psychiatric disorders

• Participants who fall asleep easily were selected, as measured by the Epworth scale
• To facilitate sleep onset, we asked participants to sleep about 30% less than usual during the night preceding the experiment (either by going to bed later or waking up earlier) and to avoid stimulants on the day of the experiment

Hypnagogic Experiences

To account for the time factor, we also compared the amount of reported hypnagogia at the bottle drop with the one obtained in an additional control experiment, in which subjects took a 30-min break in a dark bedroom; they were regularly awakened by a sound and asked to describe their mental content.

• We then calculated the percentage of reported hypnotia when the awakening sound occurred after MSEs to be in comparable conditions to the current study.

The Eureka moment

To detect sudden changes in solving time (taken here as a marker of insight), we used an algorithm provided by MATLAB named “findchangepts”, which detects abrupt changes in signals.

• By definition, the algorithm always finds a point in which solving time decreases the most even if the “drop” is small or corresponds to a random, not sustainable deviation in reaction time in one trial.

Results

We scored each participant’s break using standard sleep scoring criteria (34) and divided participants into three groups based on their vigilance state during the break (see demographic and sleep parameters in table S1): the “Wake” group, the “N1” group (subjects with at least one 30-s epoch of N1 but without any signs of deeper sleep stages, N = 24), and the “N2” group, including three who directly fell into N2 without passing by N1.

• All groups were exactly in the same conditions during the incubation period. Subjects in the N1 group were awake most of the time.

A single minute of N1 inspires insight

There was a significant effect of the group (wake versus N1 versus N2) on the percentage of participants who found the hidden rule after the break (Fisher’s test, P < 0.001)

• The percentage of individuals gaining insight was similar in the Wake and N2 groups, suggesting that there is no relationship between specific sleep/wake trajectory and general insight abilities

Neurophysiology of the sweet spot

Sleep onset is a complex, dynamic process (35, 36), potentially encompassing multiple transitions between different substages (31), each with subtle variations in physiological activity (e.g., alpha/theta, muscle relaxation).

• To better understand the critical factors for boosting insight, we calculated the power spectra for the entirety of the resting period and assessed whether it could predict subsequent Eureka moments in all subjects.
• The power in the delta (3.2 to 4.4 Hz) and alpha (9 to 9.8 Hz) bands were both predictive of insight
• We found a negative, quadratic effect of alpha power on insight (Fig. 3, A and B), meaning that subjects with the highest levels of delta power had the lowest probability of insight

The first stage of non-REM sleep (N1) has received little attention, and its cognitive role is largely unknown.

However, a recent study showed that 10 min of “awake quiescence” (i.e., a quiet rest spent in a dimly lit room with reduced sensory stimulation) more than doubled the number of subjects who discovered a hidden rule compared to ten min of active wake (12).

• N1 is accompanied by involuntary, spontaneous, dream-like perceptual experiences (18, 19) that incorporate recent wake experiences (20, 21) in a creative way by binding them with loosely associated memories
• These hypnagogic experiences could be considered as an exacerbated version of awake spontaneous thoughts (e.g., mind-wandering)
• In line with this hypothesis, we recently reported an increased creative potential in patients with narcolepsy (26), a population with frequent transitions toward sleep during the day
• Thalamic deactivation in N1 often precedes that of the cortex by several minutes suggesting that executive abilities are not completely abolished during this stage

EEG spectral analyses

extracted power spectrum over contiguous epochs of 30 s

• used Welch’s method to compute power spectra density between 1 and 30 Hz using windows of 6 s with 50% overlap and a frequency resolution of 0.2 Hz
• time-resolved power spectrum around bottle drops and random time drops (−50 to 10 s around these times)
• Windows that exceeded an absolute amplitude of 750 μV before a drop were excluded from this analysis

Acknowledgments: We thank S. Leu, P. Dodet, J. Maranci, B. Dudoignon, and M. A. Paller for help with sleep scoring.

Funding: This work was supported by Doctoral School ED3C (to C.L.), a research grant from Société Française de Recherche et Médecine du Sommeil (C.O.), and INSERM research endowment (to D.O.).

• Author contributions: Conceptualization, Methodology, Investigation, Writing, Data and Materials Availability, Supervision, and Competing interests: The authors declare that they have no competing interests.

Experimental procedure

The protocol was subdivided into four main phases: training, pre- and post-training phases

• Training consisted of 10 trials with the hidden rule
• Pre-training consisted of 2 blocks of 30 trials each
• Break consisted of a 20-min break in a dark room without sensory stimulation
• Post-testing consisted of nine blocks of thirty trials each and lasted 64 min on average
• To control for any groups’ differences in the vigilance state, participants also performed a 3-min PVT (40) at the beginning of each phase

Object

The ideal object needed to meet the following criteria: makes a noise when falling, is light weight to avoid cramped arms, is slippery to facilitate the fall, is large enough so that the fist cannot close on it, and prevents the object from falling

• We tested many objects (spoon, small steel spheres, stress balls) before finding the right one.

A reliable marker of sleep onset

Holding an object while napping is propitious to capturing creative sparks

• Of the 63 drops, 26 (41.27%) occurred after N1.
• When we examine the wake-to-sleep transition taking into account microsleep episodes (MSEs), which are not considered in the standard sleep scoring method over 30-s-long windows, this proportion rises to 77.78%
• The bottle drop appears more as a marker of the accumulation of MSEs rather than of consolidated N1
• Further tested the ability of Edison’s method to capture sleep onset by performing spectral analyses restricted to the period around the bottle drop
• We observed a clear power increase in the delta band 24 s before the drops (significant cluster: [−23.8, −0.2] s, Pcluster < 0.0001; Fig. 4A), an increase that was superior to the one that would be observed by chance (randomly generated time drops, see Materials and Methods).
• This delta increase was congruent with subjective reports as most subjects (81.48%) reported that they had been drifting to sleep when the bottle dropped.

A delayed Eureka moment

The “Eureka” moment did not occur immediately following the resting period, but rather after 94 trials on average, regardless of the group [mean Wake= 82.67, N1 = 101.55, N2 = 106.73, Wilcoxon signed-rank test, z = −6.33, P < 0.001]

• This finding, along with the fact that the average length of the Post phase was approximately 1 hour (leaving plenty of time to discover the rule after the dissipation of sleep inertia), strongly suggests that the lower rate of insight in the N2 group cannot be explained by sleep inertia.

EEG recordings

Subjects were continuously monitored with video polysomnography during the experiment. The montage included three EEG channels (FP1, C3, and O1), EOG with electrodes placed on the outer canthi of the eyes, chin EMG, a microphone, and infrared video recordings (Brainet, Medatec Ltd., France).

Sleep scoring

Two scoring methods were applied: The standard sleep scoring guidelines from the American Academy of Sleep Medicine (AASM) (34) and a microsleep scoring method, similar to the one recently developed by Hertig-Godeschalk et al. (42).

• EEG was scored continuously to identify MSEs, defined here as any short window of at least 3 s of theta activity with alpha loss.

Statistical Analysis

Interjudges’ agreement was evaluated with the Cohen’s Κ test.

• Fisher’s and chi-square tests were used in the analysis of contingency tables and Kruskal-Wallis tests for ordinal variables
• Nonparametric statistics were applied when variables could not be approximated to the normal distribution (Shapiro-Wilk test).
• All tests were two-tailed and a probability level of less than 0.05 was considered significant.

Source