A massive new dream database reveals clues about consciousness during sleep

An international collaboration of scientists has created the largest-ever public database of brain activity recordings and accompanying dream reports. In a first analysis of this resource, the researchers confirmed that dreaming is not exclusive to the rapid eye movement stage of sleep, finding that when conscious experiences occur in deeper sleep, the brain exhibits patterns of activity that more closely resemble wakefulness. The study was published in the journal Nature Communications.

For millennia, dreams have been a source of fascination, but their systematic study is a modern endeavor with significant scientific applications. Investigating subjective experiences during sleep can inform research into consciousness itself, aid in understanding memory consolidation, and provide insights into sleep disorders like sleepwalking. Despite this importance, progress has been hampered by a persistent challenge: dream studies are resource-intensive, often resulting in small sample sizes that are difficult to compare across different laboratories.

To address this limitation, a consortium of 53 researchers from 37 institutions across 13 countries came together to build the Dream EEG and Mentation database, known as DREAM. Coordinated by Monash University in Australia and supported by organizations including the Bial Foundation, the project aimed to centralize and standardize decades of dream research.

The goal was to create a large-scale, publicly accessible resource that would allow scientists to conduct more robust and comprehensive analyses of the dreaming brain. Giulio Bernardi of the IMT School for Advanced Studies Lucca in Italy, a contributor to the project, noted that the effort represents a decisive step in the scientific exploration of human consciousness by gathering vast amounts of research into a single place.

The team assembled data from 20 different studies, resulting in a collection of more than 2,600 records of awakenings from 505 participants. For each record, the database contains electroencephalography, or EEG, recordings, which measure the brain’s electrical activity using electrodes placed on the scalp. Some records also include magnetoencephalography, a related technique that measures magnetic fields produced by the brain’s electrical currents. These neurophysiological recordings are paired with a report from the participant upon waking.

Free daily newsletter

To make the diverse datasets comparable, the researchers established a unified classification system for the reports. Upon awakening, if a participant recalled any subjective experience, it was labeled as “experience”. If they felt they had been dreaming but could not recall any specific content, a phenomenon sometimes called a “white dream,” it was classified as “experience without recall”. If they reported no conscious experience at all, it was marked as “no experience”. This standardized approach allows for powerful, large-scale comparisons that were previously not possible.

With the database established, the researchers performed an initial set of analyses to demonstrate its utility. First, they examined the relationship between dream reports and the stages of sleep. Sleep is not a uniform state; it cycles through distinct stages defined by specific patterns of brain activity. The best-known stage is rapid eye movement, or REM, sleep, characterized by an active brain, muscle paralysis, and quick eye movements. The other stages are collectively known as non-REM, or NREM, sleep, which progresses from light sleep (N1 and N2) to deep, slow-wave sleep (N3).

The analysis of the DREAM database confirmed a well-established pattern on a large scale. Reports of “experience” were most frequent following awakenings from REM sleep. During NREM sleep, the likelihood of recalling an experience decreased as sleep deepened. Awakenings from light N1 sleep yielded more dream reports than those from deeper N2 sleep, which in turn yielded more than awakenings from the deepest N3 stage. This finding reinforces the idea that while REM sleep is a fertile ground for dreaming, it is not the only stage where conscious experiences occur.

The team then explored what makes NREM sleep with dreams different from NREM sleep without dreams. They applied an automated sleep-scoring algorithm to the EEG data. Instead of assigning a single sleep stage to a period of brain activity, this algorithm provides a probability distribution, estimating the likelihood that the brain is in a wake, REM, or NREM state at any given moment.

During NREM sleep periods that were followed by a report of a dream, the algorithm detected an increased probability of wake-like brain activity compared to NREM periods with no reported experience. It suggests that dreaming during NREM sleep may occur when the brain enters a hybrid state that is neither fully asleep nor fully awake.

Finally, the researchers investigated whether it was possible to predict the presence of a dream from brain activity alone. They used artificial intelligence algorithms, training them on features extracted from the EEG signals in the 30 seconds before an awakening. These features included standard measures, like the power of brainwaves at different frequencies, as well as more complex, nonlinear characteristics of the signal. The algorithms were tasked with distinguishing between EEG patterns that led to a report of “experience” and those that led to “no experience”.

The models were able to make this distinction with an accuracy greater than chance for both NREM and REM sleep. For REM sleep, the classifiers performed particularly well, achieving a high degree of accuracy in predicting whether a participant was having a conscious experience. This result indicates that objective, measurable signatures of dreaming exist within the brain’s electrical activity. It opens the door to future technologies that could potentially identify moments of dreaming in real time without having to wake the sleeper.

The creators of the DREAM database acknowledge that this initial work is a starting point. The primary purpose of the study was to present the database itself and demonstrate its potential. Future research can use this resource to ask more detailed questions. For example, while the current analysis could predict the presence or absence of a dream, substantially more data may be needed to identify the neural correlates of specific dream content, such as seeing a face or hearing a voice.

The database is designed to be a living project, open to new contributions from researchers around the world. By providing a large, standardized, and open platform, the consortium hopes to accelerate the pace of dream research. This collaborative approach may help scientists answer fundamental questions about the function of dreams and the nature of consciousness itself. By studying the brain as it generates entire worlds from within, researchers gain a unique window into the processes that give rise to subjective experience.

The study, “A dream EEG and mentation database,” was authored by William Wong, Rubén Herzog, Kátia Cristine Andrade, Thomas Andrillon, Draulio Barros de Araujo, Isabelle Arnulf, Somayeh Ataei, Giulia Avvenuti, Benjamin Baird, Michele Bellesi, Damiana Bergamo, Giulio Bernardi, Mark Blagrove, Nicolas Decat, Çağatay Demirel, Martin Dresler, Jean-Baptiste Eichenlaub, Valentina Elce, Steffen Gais, Luigi De Gennaro, Jarrod Gott, Chihiro Hiramatsu, Bjørn Erik Juel, Karen R. Konkoly, Deniz Kumral, Célia Lacaux, Joshua J. LaRocque, Bigna Lenggenhager, Remington Mallett, Sérgio Arthuro Mota-Rolim, Yuki Motomura, Andre Sevenius Nilsen, Valdas Noreika, Delphine Oudiette, Fernanda Palhano-Fontes, Jessica Palmieri, Ken A. Paller, Lampros Perogamvros, Antti Revonsuo, Elaine van Rijn, Serena Scarpelli, Monika Schönauer, Sarah F. Schoch, Francesca Siclari, Pilleriin Sikka, Johan Frederik Storm, Hiroshige Takeichi, Katja Valli, Erin J. Wamsley, Jennifer M. Windt, Jing Zhang, Jialin Zhao & Naotsugu Tsuchiya.