The brain’s drive to enter REM sleep follows the same fundamental pattern in species with vastly different sleep architectures, a new cross-species computational analysis shows.
Lead: A team of mathematicians and neuroscientists has developed a data-driven measure of REM sleep propensity and used it to show that the dynamics of REM sleep entry are remarkably similar across humans, rats, and mice, despite the radical differences in how these species organize sleep.
What they found: The study, published in Frontiers in Neuroscience, introduces a REM sleep propensity measure defined as the probability of entering REM sleep before accumulating additional minutes of non-REM (NREM) sleep. Rather than treating REM sleep as a simple on-off state, the measure captures how the likelihood of REM onset evolves in real time during a sleep episode.
Here is how it works. Standard sleep stage annotations divide the night into NREM and REM periods. The researchers took each moment during NREM sleep and asked: given how much NREM has already accumulated in this cycle, what is the probability that the next sleep stage transition will be to REM, rather than staying in or returning to NREM? The resulting curve describes how REM propensity rises and falls over the course of a sleep episode, offering a dynamic picture that standard summary statistics like “minutes of REM” or “REM latency” cannot provide.
Applying this metric to sleep recordings from humans, rats, and mice, the researchers found several conserved features. Across all three species, the propensity for REM sleep at the moment of REM onset positively correlated with how long the subsequent REM bout lasted. In other words, when the brain is more “primed” to enter REM, the resulting REM period tends to be longer. This finding replicates and extends a previous result in mice, confirming that the relationship holds up in rats and humans as well.
The analysis also sheds light on the well-known human phenomenon in which the percentage of REM sleep increases across the night. Conventional wisdom has often attributed this to progressively longer REM bouts as the night goes on. The team found that this is only part of the story. The increase involves a more subtle shift in the balance between two types of sleep cycles: single NREMS-REMS cycles that stand alone and sequential cycles that occur back-to-back. Short sequential cycles and longer single cycles coexist in all three species, but on different timescales, and their interplay changes across the night in humans. This nuance would be invisible to analyses that simply average REM duration across the night.
Why it matters: REM sleep has been studied intensively for decades, but most quantitative work has focused on the duration and timing of REM episodes rather than on the underlying propensity that drives them. By formalizing REM propensity as a directly computable probability, this study provides a math-based tool that can be applied uniformly across species, sleep stages, and experimental conditions.
The cross-species conservation the team uncovered suggests that the basic control mechanisms governing REM sleep entry may be shared across mammals, even in species like rats that sleep in short polyphasic bouts compared to the consolidated monophasic sleep of humans. This opens the door for rodent models to inform human sleep research at a deeper mechanistic level than previously assumed. If the same propensity dynamics drive REM transitions in mice, rats, and humans, then insights from rodent experiments on the neural circuitry of REM sleep may translate more directly to human sleep than many researchers have expected.
Because the approach is purely computational and relies on standard sleep stage annotations, it can be retroactively applied to existing datasets, making it an inexpensive and scalable addition to the sleep researcher’s toolkit. Any lab with polysomnography data and stage scoring can compute REM propensity curves without new equipment or experimental protocols.
The study also introduces a framework for analyzing NREMS-REMS cycles as a continuum rather than categorizing them into rigid types. By treating each cycle’s position in a sequence, its duration, and its REM propensity as continuous variables, the analysis captures variability that is lost in traditional cycle classification schemes.
Limits: The study is a descriptive computational analysis, not an interventional one. It does not identify the specific neural circuits or molecular signals that generate the observed propensity dynamics. The measure also depends on the quality and resolution of sleep stage scoring, which can vary across labs. The authors note that while the cross-species similarities are striking, the underlying mechanisms could still differ in ways the propensity measure does not capture. Additionally, the human data analyzed came from healthy young adults, and the rodent data from laboratory strains under controlled conditions, so whether these patterns hold across different ages, health conditions, or environmental contexts remains an open question.
Bottom line: REM sleep propensity, quantified as a probability that evolves over the course of a sleep episode, reveals deep structural similarities in how humans, rats, and mice transition into REM sleep. The finding suggests that a shared computational logic may govern REM entry across mammalian species, providing a new framework for studying both normal sleep and disorders that disrupt REM timing, such as narcolepsy, REM sleep behavior disorder, and the sleep fragmentation seen in aging.
Source: Akhavan N, Ginsberg AG, Cruz MEC, et al. A data-driven measure of REM sleep propensity for human and rodent sleep. Frontiers in Neuroscience. 2026;20:1844209. doi: 10.3389/fnins.2026.1844209. Open access via PMC13286888.