The surprising result from the University of Basel’s new Nature study is not merely that silencing two populations of sleep-drive neurons cut NREM sleep by nearly 70%. It is that most mice survived, remained highly engaged and retained long-term memories—although 17% died—raising the possibility that the brain’s pressure to sleep is not identical to the body’s underlying need for it.

Researchers cut non-rapid eye movement sleep in mice by nearly 70 percent by chronically inhibiting two populations of brainstem neurons. Most treated animals survived for weeks, stayed unusually active and retained one learned association, while 16.7 percent died.

The result, reported by William Joo and colleagues in a Nature paper published on August 19, 2026, does not make sleep optional. It suggests that a circuit generating the pressure to sleep can be altered separately from at least some functions that prolonged sleep loss usually disrupts.

This is one study in genetically modified mice, not settled consensus or evidence about what people can safely tolerate. This is reporting on animal research, not medical advice.

The study began with 162 mapped mouse brains

The researchers compared brain activity during ordinary sleep and wake cycles, six hours of forced wakefulness and three hours of recovery. They used two methods to keep mice awake, periodically misting them so they groomed or introducing a new object every 15 minutes. Both approaches can add stimulation, so finding similar patterns across them helped separate time awake from one specific procedure.

After each condition, the team stained for FOS, a protein that appears in recently active neurons, then cleared and imaged whole brains with light-sheet microscopy. The resulting sleep-wake atlas covers 162 brains across 26 conditions.

Two candidate regions stood out because their activity rose during prolonged wakefulness and declined during recovery sleep. One was the median raphe, or MR, in the brainstem. The other was the anterior medial preoptic area, or aMPO, farther forward in the brain.

Activating wake-responsive cells in either region increased the duration of NREM sleep and the strength of slow delta waves in the electroencephalogram, or EEG. Inhibiting the cells reduced attempts to sleep during deprivation and weakened the usual compensatory sleep afterward. The University of Basel’s account of the study describes these populations as monitoring prolonged wakefulness. The experiments support a more precise claim: the cells are part of circuitry that turns extended waking into measurable sleep drive.

The same region contains opposing kinds of neuron

A brain region is not a single switch. Within the median raphe, the researchers separated neurons by the chemical signals they use. GABAergic neurons and serotonergic neurons both became more active during prolonged wakefulness and promoted NREM sleep when activated. Turning both on together produced a stronger, longer effect than activating either alone.

Glutamatergic neurons in the same region behaved differently. Activating them promoted wakefulness, while inhibiting them increased NREM sleep. That cell-type split matters: saying the median raphe “controls sleep” would flatten a local circuit containing opposing functions.

The finding also joins rather than replaces other proposed sleep-pressure circuits. A separate 2026 Current Biology study of another brainstem population found that wake-active GABA neurons in the oral pontine reticular nucleus changed their activity after deprivation and helped produce recovery sleep. Sleep homeostasis is unlikely to reside in one anatomical dial.

Chronic inhibition changed the gauge itself

For the long-term experiment behind the headline, the Basel team focused on GABAergic and serotonergic cells in the median raphe. They delivered a virus that made both populations express Kir2.1, a potassium channel that decreases neuronal excitability. This was chronic co-inhibition, not surgical removal and not ordinary sleep deprivation.

The main EEG comparison included 20 control mice and 17 surviving Kir2.1 mice. In the treated survivors, NREM sleep fell by nearly 70 percent relative to controls, leaving them awake for more than 6.5 additional hours each day. Their average wake bout lasted more than twice as long, and the reduction persisted over several weeks.

The EEG result was as important as the clock. Ordinarily, delta power during NREM sleep rises after a long wake bout and recovery sleep increases after deprivation. Kir2.1 mice accumulated delta power more slowly, made virtually no sleep attempts during a six-hour deprivation session and showed no meaningful rebound above their already reduced baseline.

A useful review of sleep and circadian homeostasis cautions that delta power, NREM duration, REM duration, molecular responses and gene expression recover on different timescales. The gauge is not the whole condition. What this experiment altered was a group of established behavioral and EEG indicators of sleep pressure. It did not measure every process for which sleep may be needed.

One in six treated mice died

The paper reports 16.7 percent lethality within 28 days of viral injection. An EEG trace from one mouse captured an almost complete absence of sleep before death. The authors write that even more severe sleep loss was probably responsible, but that is an interpretation rather than a demonstrated mechanism of death.

Older animal work makes the danger plausible. In a controlled 1983 Science experiment on prolonged sleep deprivation in rats, severely sleep-deprived animals developed systemic pathology and died while stimulus-matched controls did not. Methods and species differ, so the studies cannot be equated. Together they make one conclusion difficult to sustain: suppressed sleepiness is not proof of physiological safety.

The survivors did not simply lie still while EEG software mislabeled them as awake. They traveled farther in open-field and maze tests, kept interacting with novel objects during deprivation and abraded wooden blocks more heavily over 18 days. Those are concrete signs of behavioral engagement. They are not a comprehensive health assessment.

“Retained memory” means one conditioned association

The memory experiment used contextual fear conditioning. Three weeks after viral injection, mice explored a chamber and received five brief foot shocks. Researchers returned them to the same chamber without shocks after one day and again after 14 days, measuring freezing as evidence that the context-shock association remained.

All treated mice showed freezing above their own pre-shock baseline at both recalls. At one day, however, they froze modestly less than controls and moved more before the first freeze. By day 14, their overall freezing was comparable with the control group, although they still traveled farther.

That supports long-term storage of one aversive associative memory. It does not demonstrate intact working memory, navigation, attention, skill learning or autobiographical memory, which mice cannot report. As ScienceBlog’s earlier look at infantile amnesia and memory retrieval emphasized in a different context, encoding, storage and access are separable questions. One successful task cannot stand in for the whole of memory.

There is another selection boundary. Behavioral testing necessarily describes the mice that survived the manipulation. The 16.7 percent that died cannot contribute reassuring memory or engagement scores.

Sleep pressure may not equal every need for sleep

The authors raise an intriguing possibility: perhaps the surviving mice were not carrying ordinary sleep debt while somehow resisting it. Perhaps they accumulated less of the neural state measured as sleep pressure. Their lower delta buildup, missing rebound and absent sleep attempts fit that interpretation.

But the experiment does not identify a single underlying quantity called “need” and compare it directly with “pressure.” Pressure was inferred from EEG and behavior. Need is a broader physiological idea involving whatever sleep restores or protects across the brain and body. The deaths suggest that at least some costs may continue even when familiar pressure signals are blunted.

I think that distinction is the study’s real value. It creates a model in which the command to sleep can be weakened without simply using stimulants to overpower sleepiness. Researchers can now ask which molecular, immune, metabolic and cognitive consequences track the missing sleep itself, and which track the circuit that normally makes the animal stop and sleep.

Nothing here supports trying to reduce sleep or ignore persistent sleep problems. Those are matters for a qualified clinician, and these mice underwent invasive genetic manipulation that killed a substantial minority.

The experiment therefore ends with a sharper question, not a shortcut. The brain’s pressure to sleep and the body’s full need for sleep may not be identical, but 17 percent mortality is powerful evidence that they are not safely divorced.