The field of sleep research was launched nearly 75 years ago, after the discovery of rapid eye movement sleep (REM) upended the prevailing belief that sleep was a passive state. Studying REM sleep revealed that the brain was doing something, even if scientists weren’t sure what. Today, most sleep researchers will agree that there is no one consensus on why we sleep, but many postulate that REM sleep—the stage when we dream—allows our brains to encode and reorganize information. That’s different from non-REM sleep, which is largely focused on repairing the brain from daily stress.

Now, a multidisciplinary team of researchers has pinpointed a period in early childhood when kids experience a dramatic decline in REM sleep. Their analysis, published today in Science Advances , reveals that the purpose of sleep shifts in children between 2 and 3 years old—from a process of reorganization and learning to a process of repairing the brain from daily stresses and wear.

“One reason this study was so successful,” says Van Savage , co-author of the study and biomathematician at the University of California, Los Angeles, “is because it brings together people from different backgrounds, like mathematical physical theory, statistical data analysis, and traditional sleep researchers who look at the data differently than before.”

Savage’s team was eager to parse competing theories about what our brains do when we sleep, how sleep is related to body size, and why we need more sleep early in life . But they would try a different technique than most: using a quantitative framework to analyze existing datasets.

Savage and his team sorted through previous research that could help clue them into the role of sleep. They gathered data from over 60 studies that included information on the sleeping brains of children and teens, and a few other animals, to see if pattern held true across mammal species. The team created a mathematical model to analyze data that could provide insight into what was happening during sleep, including the brain’s metabolic rate, the overall size of the brain and the time spent in REM sleep versus non-REM sleep.

After processing their data, they found a consistent pattern: human infants showed a dramatic decline in the amount of REM sleep—the stage associated with reorganization—around 2.4 years old. Beyond this age, toddlers’ rely more heavily on non-REM sleep, suggesting sleep becomes primarily for repair. Researchers also looked at time spent in REM sleep in three other mammals as they age: rats, rabbits and Guinea pigs, but didn’t find this sudden change. This discovery emphasizes how critical sleep is for developing human brains, says Savage, and how the function of sleep can shift with our brain’s changing needs.

Savage describes the brain’s reorganization process as a forest with branches and connections that REM sleep grows and prunes. An infant’s brain has to do an incredible amount of learning in the first few years of life, which requires forming, strengthening and trimming connections. “Early on, the brain is moving connections all around and really is fluid,” he says. “Later on, it's not that you can't change them at all, but they're changing much more slowly.”

Savage says the later repair phase is like taking out the trash or fixing the internal parts of a computer so that it can run smoothly again. Non-REM sleep allows our brain to clear waste build-up and keep it running normally. The brain is our most energy-consuming organ, explains Savage, and sleep is needed to repair damage to our hard-working neurons . He emphasizes that reorganization doesn’t stop––we’re able to learn new things throughout adulthood––but damage-control becomes the primary function.

“This paper is really cool because it finds this very unexpected, specific event happening in development,” says Sara Aton , a neuroscientist and sleep research at the University of Michigan who was not involved in the study. She’s optimistic about the clues to sleep function that quantitative analyses can reveal. “I hope that people will look at the modeling approach as a useful tool,” says Aton.

Others are skeptical about using this type of framework. “In my view, it's not a mathematical problem, it's a problem of understanding circumstances,” says Jerome Siegel , who studies REM sleep in mammals and was not involved in the work. He’s skeptical because models like the one developed by Savage can fail to consider factors like day length, diet and climate, which impact the sleep patterns of humans and other animals. Siegel says that there simply may not be enough data to apply to this kind of framework to get accurate results.

Aton says she’s intrigued by the discovery of this transition “because that indicates that something really important might be happening right at that transition point...the question is, what is it?”

Neither Aton nor Savage knows what sparks this transition. Savage thinks the shift might have something to do with the uniquely human challenge of having big-brained babies. By six months old, a baby’s brain has reached over half its adult size , and by age two, it’s closer to 90 percent. “If we're trying to look at how things change with brain size, that's where all the action is taking place,” says Savage.

Savage and his team decided to look at the sleep patterns of children because the brain is shifting dramatically early in life, including a rapid change in a type of brain tissue known as white matter by age 2. He also points to developmental milestones around age 2 and 3, like language learning, which may provide clues to the brain’s transition. Why there is such a sharp decrease in REM sleep, and why at around age 2-and-a-half, still remains a mystery.

Next, Savage wants to do a thorough investigation of other animals’ sleep patterns, which could tell him more about the biological origin and purpose of sleep. He thinks the sleep pattern transition might happen much earlier in other species—maybe even before birth. But his team will need a lot more data on the sleeping brains of other mammals before they can begin to find answers. “If you had good data for another mammal as it grew,” says Savage, “my guess is you would see something similar.”