Sleep Function

Speculative If the function of sleep is clearance, and if the sign of that effect is the one Science reported in 2013 rather than the one Nature Neuroscience reported in 2024, then the rate at which a brain disposes of amyloid-β and tau is set by a behaviour that is free, non-invasive, and already under partial voluntary control. That is the whole prize, and the conditional is not decoration: two measurements of the same quantity point in opposite directions, and section 7 says what would separate them.

Established The magnitude is best stated as the effect one night of sleep is measured to have on a human biomarker, because that is the only quantity this field has so far produced in people. About 31 hours awake raised ¹⁸F-florbetaben binding roughly 5% in the thalamus and hippocampus of 20 adults aged 22 to 72. Intrathecal gadobutrol was still elevated in the cortex and white matter of seven sleep-deprived humans at 48 hours, after a full recovery night, against 17 rested controls. A 38-participant randomised crossover published on 27 January 2026 found that adding sleep-active physiology to the model explained an extra 49.1% (Aβ40) to 56.0% (Aβ42) of the variance in overnight plasma biomarker change among amyloid-positive participants. Those are the numbers on one side. On the other is a mouse measurement saying the flow runs backwards.

Frontier Who collects this is unusual, because the intervention is not a product. Sleep cannot be patented, sold by the dose, or reimbursed; the beneficiaries are whole populations and the losers, if the clearance story holds, are the makers of every sedative that suppresses the oscillation the clearance is supposed to ride on. Speculative The one commercial route anyone has tested — phase-locked acoustic stimulation of slow oscillations — produced a 24% increase in slow-wave-coupled spindles in 32 older adults and no group-level difference in plasma Aβ42/Aβ40. The physiology moved and the biomarker did not, which is the honest state of the commercial case.

Handwave The condition, stated plainly: this upside is real only if brain-wide solute clearance increases in sleep, and the strongest direct test of that proposition in eleven years found the opposite by about 30%. Until one experiment runs both assays in the same animals across the same transitions, the upside here is bounded by a quantity whose sign no laboratory agrees on — and that is section 5's subject, not a rhetorical flourish.

Sleep is defined behaviourally, not physiologically, and the definition has four parts that have not changed in decades: a period of immobility or a species-typical posture, rapid reversibility, an elevated arousal threshold, and homeostatic regulation — deprive the animal and it takes the state back. Nothing in that definition mentions a function. The question this brief is about is what the state is for, and it is genuinely open.

Established The phenomenon is close to universal among animals with nervous systems, which is the single strongest argument that it does something. Sleep or sleep-like states have been reported in mammals, birds, reptiles, fish, insects, cephalopods, spiders, nematodes and cnidarians — jellyfish and Hydra included, animals with nerve nets and no brain. Frontier The durations, however, vary by more than an order of magnitude and track ecology rather than physiology: frigatebirds sleep about 40 minutes a day on the wing, wild elephants about 2 hours against 4 to 5 in captivity, wild sloths about 9.5 hours against 15 or more in captivity. A function that is universal but whose dose varies twentyfold is a function nobody has correctly specified.

Established Four candidate functions carry the literature, and they are not variants of one idea. Clearance: sleep opens the interstitial space and flushes metabolic waste, amyloid-β among it. Synaptic renormalisation: waking potentiates synapses and sleep scales them back down, restoring capacity and signal-to-noise. Memory consolidation: sleep replays and redistributes the day's traces from hippocampus to cortex. Metabolic repair: sleep resolves an oxidative debt incurred by being awake. Speculative They predict different experiments, they are supported by different measurements, and no result yet published rules any of them out.

Speculative This brief takes the position that sleep-function research is pre-paradigmatic in a specific and diagnosable way: the field has not agreed what quantity it is measuring. “Clearance” means one thing if you inject a tracer into cerebrospinal fluid and watch it enter the brain, and a different thing if you inject it into tissue and watch it leave. Handwave Those two experiments can both be competently run, both be correct about their own observable, and yield headlines with opposite signs — which is exactly what happened, and it is the anchor of everything below.

Established The anchor: two measurements in journals of the same rank, in the same species, disagree about the direction of sleep's effect on brain clearance. Both are below, at full precision, because the precision is where the disagreement actually lives.

Established Xie and colleagues, Science, October 2013, in male C57BL/6 mice aged 10 to 12 weeks. Real-time tetramethylammonium iontophoresis put the interstitial volume fraction at 23.4 ± 1.9% in naturally sleeping mice (n = 6) against 14.1 ± 1.8% in awake mice (n = 4), P < 0.01. Tortuosity was 1.3 to 1.8 in every state with no significant difference, so the change is volume rather than geometry. Cerebrospinal-fluid tracer influx over 30-minute sessions was about 95% lower in awake animals. Radiolabelled amyloid-β cleared with a rate constant of 0.02 min⁻¹ asleep against 0.01 min⁻¹ awake — twofold faster, across 23 to 29 mice. The paper's conclusion is that the restorative function of sleep may be a consequence of enhanced removal of neurotoxic waste.

Frontier The commission that produced this brief named the 2013 result as “~60% interstitial expansion” in sleep, and that attribution needs one correction which turns out to matter. The “more than 60%” figure is not the natural-sleep comparison. It is the anaesthesia arm: 13.6 ± 1.6% awake rising to 22.7 ± 1.3% in the same ten mice after ketamine/xylazine (100 mg/kg and 10 mg/kg), P < 0.01, paired t-test. The natural-sleep comparison is between different animals — four awake, six asleep, unpaired — and the abstract folds the two together as “natural sleep or anesthesia are associated with a 60% increase in the interstitial space”. Speculative The consequence is not pedantic. The state in which the 2013 expansion measurement is strongest and best controlled is ketamine/xylazine anaesthesia — which is precisely the state in which the 2024 paper reports the largest reduction in clearance. The two results collide hardest where each is most confident.

Established Miao and colleagues, Nature Neuroscience, May 2024, in male C57BL/6J mice. Instead of watching a tracer enter the brain from cerebrospinal fluid, they put 0.5 µl of Alexa Fluor 488 at 5 mg/ml (about 570 Da) into the caudate putamen over 10 minutes and measured its arrival and decay in the frontal cortex by fibre photometry. Clearance fell in natural sleep (P = 0.016) and under three structurally unrelated anaesthetics — dexmedetomidine 200 µg/kg (P = 0.0029), ketamine/xylazine 100 + 20 mg/kg (P = 0.0015) and pentobarbital 50 mg/kg (P = 0.037). Peak clearance in awake controls ran at 70 to 80%. The authors' institution puts the effect at about 30% reduction in sleep and about 50% under anaesthesia (institutional press release, UK Dementia Research Institute). The paper's stated conclusion is that brain clearance is markedly reduced, not increased, during sleep and anesthesia. An Author Correction on 14 June 2024 fixed one column of source data for Figure 3b and changed nothing in the claim.

Frontier The dispute is now formal and on the record, which is unusual and useful. On 11 March 2025 Nature Neuroscience published a Matters Arising from fourteen authors including Maiken Nedergaard, Douglas Kelley, John Thomas and Steven Goldman, stating that the conclusion of that study — brain clearance is reduced during sleep and anesthesia — is not supported by the data presented, and raising objections to experimental methodology, analytical rigour and mathematical assumptions. Frontier The same issue carried the reply, whose opening position is that all the technical objections raised are misunderstandings, misleading or incorrect, and that the critics' aim was to discredit rather than to question. Speculative Two laboratories, one journal, one day, and no agreed experiment: that is the pre-paradigmatic signature this category exists to record.

Frontier The specific technical objections are worth stating, because they are testable rather than rhetorical. Reported in trade coverage of May 2024, updated March 2025: that intraparenchymal injection needs more controls for glial scarring; that the cannula should have been clamped for 30 minutes after injection to prevent backflow; that the sleep condition was recovery sleep rather than spontaneous sleep; that no established perivascular route connects caudate putamen to frontal cortex, so the measurement may not be on a clearance pathway at all; and, from Jonathan Kipnis, that needle insertion damages tissue and raises intracranial pressure. Frontier Against that, Jason Rihel at University College London is quoted finding the mathematical modelling and the diffusion controls persuasive. Speculative Every one of those objections is an experiment somebody could run, and none of them has been run as an arbitration.

Frontier Beneath the sleep question sits an older one that has never been settled: whether parenchymal transport is convective at all. Smith and colleagues in eLife (2017) injected 10, 70 and 2,000 kDa dextrans into cisterna magna and into parenchyma directly and found penetration strongly size-dependent in tissue and size-independent only in paravascular spaces — the signature of diffusion inside the parenchyma and convection only along vessels. Aquaporin-4 deletion produced no significant difference in ovalbumin penetration in mice or rats, and similar Aβ_1–40 distribution in Aqp4^+/+ and Aqp4^−/− animals. Their conclusion is that the results do not support the proposed glymphatic mechanism of convective solute transport in brain parenchyma. Frontier Other groups have reported aquaporin-4-dependent transport, and the disagreement has survived a decade of replication attempts.

Frontier Two 2024–2025 results identify a driver, and they agree with each other and with the 2013 direction. Jiang-Xie and colleagues in Nature (2024) showed that synchronised neuronal firing in non-REM sleep generates large-amplitude, self-perpetuating ionic waves in interstitial fluid; chemogenetically flattening those waves largely impeded cerebrospinal-fluid infiltration and clearance, and transcranial optogenetic stimulation generating them artificially substantially potentiated perfusion — a causal, bidirectional manipulation, which is rare here. Frontier Hauglund and colleagues in Cell (2025) report that the frequency of norepinephrine oscillations in non-REM sleep predicts glymphatic clearance, and that zolpidem suppresses both, with microdialysis showing about a 50% drop in prefrontal norepinephrine for several hours. Speculative If that holds, a hypnotic prescribed to produce sleep suppresses the mechanism sleep is supposed to be for, which is the single most consequential claim in this brief and rests on one laboratory.

Established The human record is thinner, more indirect, and mostly on the 2013 side. Fultz and colleagues in Science (2019) measured a large cerebrospinal-fluid oscillation at 0.05 Hz — roughly every 20 seconds — appearing in non-REM sleep in 13 subjects, a 5.52 dB increase in CSF signal power (95% CI 2.33 to 7.67, P = 0.003) against 11 awake, with slow-delta EEG peaking 6.4 s before the CSF peak and cortical BOLD anticorrelated with ventricular CSF at r = −0.48 at a 2 s lag. Established Ringstad and colleagues in JCI Insight (2018) gave 0.5 ml of 1.0 mmol/ml gadobutrol intrathecally to 8 reference subjects and 9 with idiopathic normal-pressure hydrocephalus, finding centripetal enhancement peaking at 24 hours and clearing over four weeks — and stating honestly that they lacked the scans between 6 to 9 and 48 hours needed to isolate a sleep effect.

Established The nearest thing to a human test of the anchor is Eide and colleagues in Brain (2021), and it is small. Seven participants kept awake for one night against 17 allowed to sleep, same intrathecal gadobutrol; tracer was still elevated in frontal, temporal, parietal and cingulate cortex, in white matter and in limbic structures at 24 hours, and still elevated at 48 hours after a recovery night. Their conclusion: one night of total sleep deprivation impairs molecular clearance from the human brain, and humans do not catch up on lost sleep. Frontier Their own stated limitations are severe: no EEG, so micro-sleep in the deprived arm cannot be excluded; limited temporal resolution; and an injected tracer that moves against its own concentration gradient, unlike an endogenous metabolite.

Frontier Two 2025–2026 human results are the newest evidence and they both favour clearance in sleep. A randomised crossover in Nature Communications (27 January 2026), 38 analysed participants aged 49 to 66 with 25 sleep and 22 wake visits plus a 13-person replication, reported that a neuro-glymphatic model explained an additional 49.1% to 56.0% of variance in overnight plasma Aβ40 and Aβ42 change in amyloid-positive individuals, concluding that sleep-active physiological processes enhance overnight glymphatic clearance of Alzheimer's biomarkers into plasma. Frontier A 12-person randomised crossover preprint (June 2025) found cerebrospinal Aβ and tau consistently lower after sleep than after afternoon sampling or sleep deprivation, supporting a model in which slow-wave sleep enhances CSF turnover. Speculative Both are small, both are indirect, and neither measures the quantity the mouse studies disagree about.

Frontier The human proxy the field leaned on for five years does not appear to measure what it claimed. The diffusion-tensor ALPS index was adopted rapidly because it needs no injection. A 2024 critique in Neuroradiology notes that perivascular spaces account for about 1% of white matter tissue, that most cortical vessels do not connect directly to the white-matter vessels ALPS measures, and concludes that the validity of DTI-ALPS as a marker of glymphatic brain clearance is highly questionable. Established A 2026 Radiology study compared ALPS against intrathecal contrast-enhanced MRI in 56 participants and found a negative association with deep white-matter enhancement at 24 hours (β = −48, P < 0.001), no significant association at 48 hours anywhere, and concluded that ALPS likely reflects local white-matter integrity rather than glymphatic clearance. A large share of the human glymphatic literature is built on this index.

Established The other three candidate functions have measurements too, and they are not weaker. De Vivo and colleagues in Science (2017) reconstructed 6,920 synapses from 8,427 spines in mouse motor and somatosensory cortex and found the axon-spine interface 18.9% smaller after sleep than after spontaneous wake and 17.5% smaller than after enforced wake, with the effect confined to the smaller 80% of synapses — the largest fifth changed by +0.7% and +2.0% and did not scale. Established Wang and colleagues in Nature (2018) identified 80 mostly synaptic sleep-need-index phosphoproteins whose phosphorylation tracks sleep need, and showed that inhibiting SIK3 reduces both their phosphorylation and non-REM slow-wave activity. Sleep pressure has a molecular readout, and it is synaptic rather than metabolic.

Frontier The metabolic account gained its strongest result in 2025 and it is in flies. Sarnataro, Miesenböck and colleagues in Nature located sleep pressure in a mismatch between electron supply and ATP demand in the mitochondria of dorsal fan-shaped body neurons: installing an alternative oxidase from Ciona intestinalis to bypass complex IV relieved basal sleep pressure; uncoupling proteins Ucp4A and Ucp4C reduced sleep (P ≤ 0.0381); a light-driven proton pump that makes ATP from photons rather than electrons precipitated sleep; mitochondrial fusion increased baseline sleep and rebound (P ≤ 0.0450) while fission decreased sleep (P < 0.0001). Their conclusion is that sleep, like ageing, may be an inescapable consequence of aerobic metabolism. Speculative That is a different theory of sleep from all three above, and it makes no prediction about brain clearance whatsoever.

Established And the result that most embarrasses every brain-centred account. Vaccaro and colleagues in Cell (2020) found that flies deprived of sleep died around day 20 against about 40 for controls, that reactive oxygen species accumulated in the gut, that antioxidant overexpression in the gut restored near-normal lifespan while overexpression in the nervous system did almost nothing, and that most tissues including brain were indistinguishable between deprived and control animals. Sleep-deprived mice showed the same intestinal ROS elevation. Speculative The lethal consequence of sleep loss, in the one organism where it has been traced, was not in the brain.

Established The comparative record sets a floor under how universal any proposed function can be. Lyamin and Siegel recorded four juvenile northern fur seals and found REM sleep falling from about 80 minutes a day on land to 3 minutes a day in seawater — a 96.4 ± 1.0% reduction sustained for 10 to 14 days — with a rebound of only 28 ± 17 minutes against 895 ± 147 minutes lost, about 3 ± 2%. Frontier A state can be suppressed by 96% for two weeks with no debt repaid, which is very hard to reconcile with REM discharging any vital obligation. Siegel's standing alternative, that sleep is adaptive inactivity optimising the timing of behaviour rather than performing a restorative function, has never been refuted and is rarely engaged.

Frontier Question one, and everything else is downstream of it: what quantity is “brain clearance”? The 2013 experiment measures tracer entering brain from cerebrospinal fluid and the expansion of the space it enters. The 2024 experiment measures tracer leaving a tissue depot. Speculative These are not the same observable, and a state that increases inflow while decreasing net efflux is physically possible — indeed one of the 2024 authors has publicly suggested precisely that reading of the older data. Handwave If both papers are right about their own measurement, the field's central claim was never about a single quantity, and a decade of citations to “sleep clears the brain” has been citing an ambiguity.

Frontier Question two: is parenchymal transport convective or diffusive? The 2017 size-dependence result and the aquaporin-4 nulls say diffusive in tissue and convective only along vessels; the glymphatic programme says convective throughout. Speculative The distinction is not academic: diffusion cannot be modulated by a brain state, and bulk flow can. If parenchymal transport is diffusive, then sleep can only act by changing the volume, the concentration gradient or the perivascular pump, and the mechanism has to be re-specified. Speculative A wholly diffusive parenchyma with a convective perivascular network is compatible with most of the data on both sides and is the version fewest people argue for.

Frontier Question three: which anaesthetics are models of sleep and which are not? Established The 2013 paired interstitial-expansion measurement was made under ketamine/xylazine, which produces high-amplitude slow waves that superficially resemble non-REM. Speculative The 2024 paper found clearance suppressed under three anaesthetics with three different molecular targets, which argues the effect is of the state rather than of the drug. Handwave But nobody has established that any anaesthetic is a valid model of physiological sleep for fluid dynamics, and the entire literature on both sides leans on that unexamined assumption. This is the seam with Consciousness Research, which has the same problem for a different reason.

Frontier Question four: is sleep one function or several that happen to share a behaviour? Clearance, synaptic renormalisation, memory consolidation and mitochondrial repair each have a measured result behind them, and the comparative review that surveys them explicitly declines to pick one, stating that sleep supports many functions for the brain and body. Speculative The multi-function answer is the most likely and the least useful, because a state with four functions and no dominant one cannot be optimised, dosed or targeted. Handwave It is also unfalsifiable in its usual form, which is why this brief treats it as a placeholder rather than a theory.

Frontier Question five: why do the durations vary twentyfold? Elephants at 2 hours, frigatebirds at 40 minutes on the wing, fur seals at 3 minutes of REM for a fortnight, and humans carrying single-gene variants who sleep four to six hours with no measured deficit. Speculative Every restorative theory predicts that sleep need scales with something — brain mass, metabolic rate, synaptic load — and none of the proposed scalings survives contact with the comparative data. Speculative Siegel's adaptive-inactivity account explains the variation easily and explains the homeostatic rebound badly; the restorative accounts do the reverse.

Frontier Question six: does the sleep-dementia association need clearance at all? The epidemiological association between poor sleep and later dementia is robust, but the causal arrow is contested in both directions — amyloid pathology disrupts sleep, and the earliest tau deposition is in brainstem arousal nuclei. Speculative If the association is mostly reverse causation, every clearance mechanism could be wrong and the epidemiology would be unchanged. Frontier The 2024 authors have said as much publicly: good sleep may reduce dementia risk for reasons other than clearing toxins. Neurodegeneration Reversal carries the pathology side of this.

Speculative Question seven, and the one with the most immediate consequence: does sedation that is not sleep carry the same benefit? If norepinephrine oscillation is the pump, then a hypnotic that suppresses norepinephrine produces sleep architecture without the function. Speculative This is a prescribing question affecting tens of millions of people and it rests, today, on one paper from one laboratory in one species. Handwave Nothing in the human literature tests it.

Frontier The bottleneck is not biological. It is that every available assay perturbs the thing it measures, and no two assays measure the same thing. Everything below follows from that single fact.

Frontier Bottleneck one: the tracer is an intervention. Cisterna-magna injection raises intracranial pressure and can itself drive flow; intraparenchymal injection makes a needle track, causes glial reaction, and creates a concentration gradient no metabolite has. Speculative Both objections are correct and they apply to opposite sides of the anchor disagreement, which is why neither side's methodological criticism of the other is answerable within the existing methods. Handwave There is no tracer-free in vivo measurement of brain solute clearance, and its absence is the single most important gap on this page.

Frontier Bottleneck two: anaesthesia contaminates the comparison in both directions. A large share of the mouse imaging literature is collected under anaesthesia because head-fixed optical imaging of a freely sleeping mouse is hard. Speculative Anaesthesia is used both as a proxy for sleep and as a separate condition, sometimes in the same paper, and ketamine/xylazine specifically produces the slow-wave signature that makes it look like a good proxy. Handwave Until the field agrees that anaesthesia is an independent state rather than deep sleep, half the measurements are uninterpretable.

Frontier Bottleneck three: natural sleep versus recovery sleep. A mouse imaged during spontaneous mid-light-phase sleep and a mouse imaged in rebound after deprivation are in different states with different slow-wave amplitude, different norepinephrine dynamics and probably different fluid dynamics. Speculative This is one of the explicit objections raised against the 2024 paper, and it is a genuine confound that the 2013 paper also carries — its awake and asleep animals were measured at midday, in different individuals.

Frontier Bottleneck four: the human assays are the wrong instrument for the question. Intrathecal gadolinium requires a lumbar puncture, an off-label contrast agent and a 48-hour scan schedule, so cohorts are single-digit to low-double-digit and drawn from patients being investigated for something else. Speculative Fast fMRI measures a cerebrospinal-fluid flow surrogate at the fourth ventricle and not clearance. Handwave DTI-ALPS requires nothing and appears to measure white-matter integrity, which is why it is popular and why the 2026 comparison against intrathecal MRI found essentially no association at 48 hours.

Established Bottleneck five: the field measures small molecules and reasons about aggregating proteins. Alexa Fluor 488 is about 570 Da; gadobutrol about 605 Da; amyloid-β_1–42 about 4.5 kDa and it oligomerises. Speculative The 2017 size-dependence result says transport in tissue falls off sharply with molecular weight, so a tracer chosen for brightness may have transport physics that a pathological peptide does not share. Handwave Every headline about sleep clearing Alzheimer's proteins rests on an extrapolation across roughly an order of magnitude in mass that nobody has validated.

Speculative Bottleneck six is sociological and it is the binding one. The two laboratories exchanged a Matters Arising and a Reply on the same day in March 2025 and neither changed position. Handwave There is no adversarial collaboration, no agreed protocol, no pre-registration and no shared animal cohort, and none of these requires an instrument that does not exist. What is missing is an agreement, and agreements are cheaper than microscopes.

Frontier This subject waits on results in fields that are not studying sleep, which is the usual shape of a pre-paradigmatic problem.

Frontier Dependency one: a label-free measurement of interstitial solute transport. The candidate routes are magnetic-resonance methods sensitive to endogenous metabolites, optical methods reading an intrinsic contrast, or an implantable sensor reading a native analyte rather than an injected dye. Speculative None of these is a sleep project; each would be built by an imaging-physics or sensor group for its own reasons, and sleep research would inherit it.

Frontier Dependency two: chronic awake head-fixed imaging with polysomnography. Scoring genuine non-REM and REM by EEG and EMG while imaging fluid dynamics at second resolution in an unanaesthetised mouse removes the largest confound on this page. Speculative The hardware exists in pieces across several laboratories and has not been assembled into a standard rig.

Speculative Dependency three: the circuit-level account of what a brain state is. If norepinephrine oscillation is the driver, sleep-function research depends on locus coeruleus physiology, which belongs to arousal research. Speculative Consciousness Research shares the dependency and the same methodological hazard: both fields are trying to characterise a state using instruments calibrated in a different state.

Speculative Dependency four: a settled account of amyloid and tau clearance kinetics in humans. Established Stable-isotope labelling kinetics gives production and clearance rates for Aβ in cerebrospinal fluid, and it is a different measurement from tracer transport. Handwave Nobody has run isotope kinetics against a sleep manipulation with the statistical power to detect a 30% effect, and that study requires no new physics at all.

Frontier The experiments below rank by how much a result would change this brief's assessment, not by how hard they are. The first one is cheap and nobody has run it.

Speculative Tier one, and the decisive experiment for this brief: both clearance assays in the same animals across the same state transitions, pre-registered, with each laboratory's predicted direction recorded in advance. The same mice, EEG- and EMG-scored so that wake, spontaneous non-REM and ketamine/xylazine are separated rather than pooled, receive a cisterna-magna influx measurement and an intraparenchymal efflux measurement in counterbalanced order, with a third arm receiving both. Frontier The experiment has three possible outcomes and all three are informative: influx up and efflux down means the two papers measure different quantities and the field's central claim needs restating; both up or both down means one laboratory has a methodological artefact and it can be localised; no effect in either means the mouse is a poor model for a question that was always about humans. Handwave No instrument in that protocol needs inventing. What is missing is the agreement to run it, which is why this brief treats the bottleneck as institutional rather than technical.

Frontier Tier two: the arbitration controls that each side has demanded of the other, run as a single study. Cannula clamped against unclamped after intraparenchymal injection; glial-reaction histology at the injection site against a sham track; spontaneous sleep against rebound sleep after deprivation; caudate-putamen-to-frontal-cortex transport against a route both parties accept as perivascular. Speculative Each of the four is a one-figure experiment and together they would close most of the March 2025 exchange. Speculative That none has been published in the eighteen months since is the most informative fact about this field's incentives.

Speculative Tier three: a tracer-free measurement, because both sides' objections are objections to tracers. A method reading an endogenous solute — lactate, glucose or an intrinsic magnetic-resonance contrast — without an injection would settle the artefact argument outright rather than adjudicating it. Handwave No such method exists at the required spatial and temporal resolution, and no group has published a specification for one, so this tier is a requirement rather than a plan.

Speculative Tier four: molecular-weight series rather than a single tracer. Run the same protocol with tracers at roughly 0.5, 5, 40 and 500 kDa in the same animals across the same states. Frontier The prediction is sharp and published: if parenchymal transport is diffusive, penetration falls with the square root of the diffusion coefficient and the state effect should shrink with mass; if it is convective, the state effect should be roughly size-independent. Speculative This is the one design that tests the convection question and the sleep question with the same animals, and it would also tell us whether the extrapolation from a 570 Da dye to a 4.5 kDa peptide is legitimate.

Speculative Tier five: the human experiment that does not need an injection. Stable-isotope labelling kinetics of Aβ and tau, in a randomised crossover of one night of sleep against one night of deprivation, powered to detect a 30% change in fractional clearance rate. Frontier It measures an endogenous molecule at its endogenous concentration, it has an established assay, and it is the only design on this page that could make a claim about human amyloid without an intrathecal needle. Speculative The randomised crossovers published in 2025 and 2026 used plasma and cerebrospinal concentrations rather than kinetics, which is a weaker measurement of the same idea.

Speculative Tier six: the norepinephrine prescribing test, which has a natural experiment already running. Zolpidem is prescribed to millions; if it suppresses the norepinephrine oscillation that drives clearance, long-term users should show a biomarker signature. Frontier A matched-cohort analysis of plasma p-tau217 and Aβ42/40 in chronic hypnotic users against matched insomnia patients on non-sedative treatment is an observational study using assays that already exist. Handwave Confounding by indication is severe, the effect would be small, and no such analysis has been published — but the population is already exposed, which makes this the cheapest evidence available and the most neglected.

Speculative Tier seven: falsify the clearance account by dissociation. If a manipulation produces slow-wave sleep with the fluid dynamics abolished, or the fluid dynamics with the sleep abolished, the functional claim can be tested directly rather than correlationally. Frontier The 2024 optogenetic result already achieves half of this in one direction by generating ionic waves in a non-sleeping animal. Speculative The other half — sleep with the pump off, and a cognitive or biomarker endpoint measured — is the experiment that would tell us whether clearance is a function of sleep or merely a correlate of it.

Established And the experiments that are not tests of sleep function and are regularly reported as though they were. Cross-sectional DTI-ALPS comparisons between patient groups, epidemiological associations between self-reported sleep duration and later dementia, and single-night biomarker changes without a clearance measurement are studies of association in populations where the causal arrow is contested. Speculative They are consistent with the clearance account, consistent with reverse causation, and consistent with a third factor, which is another way of saying they do not discriminate.

Frontier There is no product here and this page will not invent one. Every requirement below is a requirement for an instrument that would answer a question.

Speculative The specification for the decisive test, stated from the two protocols it has to reconcile. Chronic EEG and EMG electrodes for state scoring in a head-fixed mouse; a cisterna-magna cannula for influx and a stereotaxic parenchymal cannula with a clamp for efflux, in the same animal; two-photon or macroscopic fluorescence at the influx site and fibre photometry at the efflux site, sampled faster than the 20-second oscillation the human work identifies; a tracer series spanning roughly three orders of magnitude in molecular weight; and pre-registered analysis code, because the March 2025 exchange was substantially about analysis rather than about data. Handwave Every component of that list exists in a catalogue. The assembly does not exist because no single laboratory has an incentive to build the rig that could prove it wrong.

Established The human side has a specification too, and it is already built and already constraining. Fast fMRI at sub-400 ms repetition time with simultaneous EEG resolves a 0.05 Hz cerebrospinal-fluid oscillation and a 6.4-second lead of slow-delta over flow. Intrathecal gadobutrol at 0.5 ml of 1.0 mmol/ml with scans out to four weeks resolves brain-wide tracer transport in humans. Frontier The first measures a flow surrogate and not clearance; the second measures clearance and requires a lumbar puncture, which is why its cohorts are seven and seventeen rather than seven hundred. Neither instrument is the limiting factor; the tolerance of healthy volunteers for spinal needles is.

Speculative The engineering requirement that would change the field is a wearable or bedside assay of clearance. Nothing in the current literature specifies one. Brain-Computer Interfaces and Precision Neuropsychiatry are developing chronic recording and closed-loop stimulation for other reasons, and this subject would inherit both. Handwave An inherited instrument is not a programme, and this page does not claim one.

Frontier The nearest page on this map is Human Hibernation, and the two are deliberately inverse. That brief is about inducing a hypometabolic state on purpose and asks whether large mammals can be driven into torpor at all; it treats brain state as something to be engineered. This page asks why an ordinary brain state exists in the first place and treats it as something to be explained. Speculative The seam is that both depend on the same unresolved question of whether an induced state and a natural one share a physiology, which the torpor brief meets as a large-mammal scaling problem and this one meets as the anaesthesia confound.

Frontier The methodologically closest neighbour is Consciousness Research, for a reason worth stating. Established That brief's central finding is that every objective marker the field owns was calibrated against verbal report and so cannot outrun it. Speculative This brief has the same defect in a different medium: every measure of brain clearance was calibrated against an injected tracer and cannot outrun the injection. Memory Engineering owns the consolidation function that is one of the four candidates here and is not duplicated on this page.

Speculative Three neighbours inherit something concrete and none of them produces evidence for this page. Neurodegeneration Reversal would inherit a modifiable upstream variable if the clearance account survives, and inherits nothing if it does not. Longevity Therapies shares the mitochondrial-ROS mechanism that the 2025 fly work proposes, which is the one place where sleep and ageing become the same question rather than neighbouring ones. Aging Societies carries the population arithmetic that makes a small per-night effect worth measuring at all.

Frontier One adjacency runs the other way and is usually missed: anaesthesiology. The 2024 result came from an anaesthesia laboratory, not a sleep laboratory, and the three-anaesthetic design is an anaesthetist's instinct rather than a sleep researcher's. Speculative The clinical question of whether general anaesthesia impairs clearance in elderly surgical patients is the field's most consequential application and it is owned by neither community.

Frontier The institutional requirement here is not money. It is an arbitration mechanism, and none exists. Established The National Institutes of Health Sleep Research Plan, last updated in December 2021, organises the field around five strategic goals and nine critical opportunities, coordinated through the National Center on Sleep Disorders Research. Speculative It does not name brain clearance as a priority, and it predates the disagreement this brief is about by two and a half years.

Frontier What the field actually has for resolving a contradiction is the Matters Arising, and March 2025 is a clean demonstration of its limits. A comment from fourteen authors and a reply from the original group appeared on the same day; the comment said the conclusion was not supported by the data and the reply said every technical objection was a misunderstanding, misleading or incorrect. Speculative The format produces a durable public record of the disagreement and cannot produce a measurement, which is what was needed.

Speculative The requirement that would change the field is an adversarial collaboration with a pre-registered protocol and a funder willing to pay for a result that embarrasses somebody. The model exists elsewhere: the adversarial-collaboration format that produced the largest test in consciousness research is described in Consciousness Research, and it damaged both theories it tested, which is what success looks like. Handwave Nothing in the sleep field's funding structure rewards that, and the two laboratories involved have every ordinary incentive to keep publishing separately.

Frontier A second institutional requirement is a reporting convention, and it is nearly free. A convention requiring every clearance paper to state which observable it measured — influx from cerebrospinal fluid, efflux from parenchyma, or net concentration — in the abstract, and to score state by electroencephalography rather than by assumption, would have prevented most of this argument. Speculative Journals impose such conventions routinely for clinical trials and have not for this.

Speculative Third, a regulatory requirement that nobody has asked for. If hypnotics suppress a clearance mechanism, that is a post-marketing safety question for a drug class dispensed at population scale. Handwave No regulator has opened it, no manufacturer has been asked to, and the evidence base is one mouse paper — which is precisely how a real signal would look at this stage and also how a false one would.

Frontier The live ethical issue is that a contested mouse result is already in public health messaging. “Sleep washes toxins out of your brain” is repeated as settled science in sleep-hygiene advice, in consumer sleep-tracker marketing and in dementia-prevention campaigns. Speculative The underlying measurement has been directly contradicted in the same species by a study whose conclusion is that clearance falls by about 30% in sleep, and the contradiction has not reached any of that messaging.

Speculative The specific harm is not that people sleep more than they need to. It is that a mechanistic claim is being used to assign responsibility: if sleep clears the brain, then a shift worker, a new parent or a person with untreated apnoea is accumulating damage through a personal failure to sleep. Handwave That framing converts a structural condition of labour into an individual health behaviour, and it does so on the strength of a number whose sign is disputed. Aging Societies carries the population version of this.

Frontier The prescribing question is sharper and affects identifiable people now. If the 2025 norepinephrine result generalises, hypnotics that suppress noradrenergic oscillation deliver the architecture of sleep without its putative function, and patients are being told they are getting the benefit. Speculative The honest position is that the evidence is one paper in mice and that the clinical question is open, which is a different statement from either reassurance or alarm. Speculative Precision Neuropsychiatry owns the prescribing infrastructure.

Speculative There is a research-ethics dimension in the animal work that the dispute makes concrete. Both sides' preferred controls — more sham-injection cohorts, more anaesthetic arms, molecular-weight series — mean more animals. Handwave A single properly designed arbitration study would almost certainly use fewer animals in total than the sequence of mutually unconvinced replications that is the alternative, which makes the institutional failure an animal-welfare question as well as a scientific one.

Frontier Finally, the equity point about who is studied. Human clearance measurements come from patients undergoing lumbar puncture for suspected cerebrospinal-fluid leak or normal-pressure hydrocephalus, and from small volunteer cohorts in high-income countries. Speculative The populations with the most disrupted sleep — shift workers, people in noisy or insecure housing, people with untreated apnoea — are absent from every dataset on this page.

Established The civilizational stake is not that we might learn to sleep less. It is that roughly a third of a human life is spent in a state whose purpose is unknown, and the economy has already priced it as waste. Speculative Shift work, always-on service industries and the expectation of asynchronous availability all assume that sleep is compressible. Speculative If it is a restorative process with a measurable output, that assumption is a slow, uncosted transfer from individual health to industrial convenience.

Handwave If the strongest version of the metabolic account is right — that sleep, like ageing, is an inescapable consequence of aerobic metabolism — the consequence runs the other way and is harder. Sleep would then not be an adaptation that could in principle be engineered around but a thermodynamic tax on running a brain on oxygen. Handwave That would make every proposal for long-duration crewed spaceflight, for pharmacological sleep reduction, and for indefinite biological lifespan subject to the same constraint, and it would be the single most important structural fact about being an animal. The evidence for it is currently a set of manipulations in fly neurons.

Speculative The near-term civilizational consequence is narrower and concerns trust rather than physiology. The glymphatic story is one of the most successfully communicated neuroscience findings of the last fifteen years, and it is now in textbooks, consumer products and clinical advice. Speculative If the direction turns out to be wrong, or the claim turns out to have conflated two different measurements, the correction will be expensive in public credibility in exactly the domain — dementia prevention — where credibility is scarcest. Frontier That is an argument for settling it sooner rather than for suppressing the dispute.

Speculative And the distributional point. If sleep quality turns out to be a modifiable input to dementia risk, the interventions available — quiet, dark, secure, temperature-controlled sleeping space, predictable schedules, treated apnoea — are distributed exactly as housing and employment security are distributed. Speculative A confirmed clearance mechanism would convert a known inequality in sleeping conditions into a measurable inequality in neurological outcome, which is a stronger political claim than the field currently makes and would be entailed by its own result.

These horizons track measurement and arbitration rather than technology, because there is no engineering programme here to track:

• 10 yr: Frontier Expect the anchor disagreement to be resolved or abandoned within this window, most likely by a third laboratory running both assays rather than by either principal conceding. Speculative Expect the human literature to move off DTI-ALPS as the 2026 comparison against intrathecal MRI propagates, with a reanalysis wave that weakens several published patient-group differences. Speculative Expect the norepinephrine-hypnotic question to produce its first human observational study, and expect it to be underpowered and confounded by indication.
• 25 yr: Speculative This is the window in which a tracer-free clearance measurement plausibly exists, and it is the only development that would make the current argument retrospectively unnecessary. Speculative The most likely outcome on this horizon is not that one side wins but that “clearance” is decomposed into two or three separately measured quantities and the original question dissolves. Speculative Expect sleep function to remain plural: several mechanisms, each with evidence, none dominant.
• 50 yr: Handwave The defensible statement is about shape rather than date. A field that cannot agree on its observable either acquires a better instrument or fragments into sub-fields that stop citing each other, and both have happened to neighbouring problems. Speculative If the mitochondrial account consolidates, sleep research merges with metabolism and ageing research and stops being a distinct field.
• 100 / 250+ yr: Handwave Beyond useful forecasting. Handwave Any specific claim about engineered sleep reduction, pharmacological substitution for sleep, or sleep-free cognition on this horizon is an assertion, and this page does not make one — the single-gene human short sleepers are the only existence proof anyone has, and they sleep four to six hours rather than none.

• Depends on No brief on this map produces a result that unblocks this one, and the absence is informative. Consciousness Research is the nearest methodological relative and supplies a cautionary pattern rather than a prerequisite: a field whose every marker is calibrated against the thing it was meant to replace. Human Hibernation shares the induced-versus-natural-state problem and is not blocking, because torpor work proceeds whether or not sleep function is settled. Memory Engineering owns the consolidation function and would be reinterpreted, not blocked, by a result here.
• Requires (not on this map) Six conditions, in the order of the tokens above. A tracer-free measurement. Every objection either laboratory raises against the other is an objection to an injected tracer; a method reading an endogenous solute would make the argument moot instead of adjudicating it, and no group has published a specification for one. The both-assays experiment. Cisterna-magna influx and intraparenchymal efflux in the same electroencephalographically scored animals across wake, spontaneous non-REM and ketamine/xylazine, with predicted directions registered in advance; no component needs inventing. The collaboration. A written agreement between the two groups on protocol, analysis code and what each result would mean — the cheapest item on this list and the one whose absence does the most damage. The rigs. Chronic head-fixed optical imaging with simultaneous electroencephalography and electromyography in unanaesthetised mice, sampling faster than the 20-second oscillation the human work identifies; the pieces exist and the standard rig does not. The human kinetics. Stable-isotope labelling of amyloid-β and tau across a randomised sleep-versus-deprivation crossover, powered for a 30% change in fractional clearance rate — endogenous molecules, established assay, nobody has run it. The payer. Without a reimbursement route, no sponsor funds the multi-year sleep-intervention trial with a neurological endpoint that would convert a mechanism into a medicine.
• Enables A confirmed clearance mechanism would give Neurodegeneration Reversal a modifiable upstream variable and would put a safety question to every sedative-hypnotic in use; a refutation would remove a decade of mechanistic scaffolding from dementia-prevention advice and enable nothing, which is the more likely and the less discussed outcome. A result either way would reweight Memory Engineering and Neuroplasticity Engineering against the synaptic account. No typed enabling edge is claimed, because a mechanism whose sign is disputed enables nothing on a schedule.
• Adjacent Anaesthesiology, magnetic-resonance physics, fluid mechanics and comparative zoology supply most of the numbers on this page, and none of those communities is trying to explain sleep. Within this map: Human Hibernation for the induced-state problem, Consciousness Research for the calibration problem, Longevity Therapies for the mitochondrial account, Brain-Computer Interfaces and Precision Neuropsychiatry for the chronic-recording hardware this subject would inherit, and Aging Societies for the population arithmetic.

Frontier Claim: “sleep washes toxins out of the brain” is established science. Established It is a hypothesis with a landmark 2013 measurement behind it and a 2024 measurement of the same species pointing the other way, plus a formal published dispute in March 2025 that neither side conceded. Speculative The honest statement is that brain-wide solute clearance changes with brain state, that the direction is disputed for sleep and anaesthesia, and that the human evidence is indirect and mostly favours the original direction. That is a weaker claim than the one in circulation and it is the one the evidence supports.

Frontier Claim: the 2024 contradiction has been refuted. Established It received a Matters Arising from fourteen authors and an Author Correction that fixed one column of source data in Figure 3b. Frontier Neither is a refutation. The correction did not touch the claim, and the Matters Arising was answered on the same day by a reply calling every technical objection a misunderstanding, misleading or incorrect. Speculative A dispute in which both parties published simultaneously and neither moved is a dispute, not a resolution.

Frontier Claim: the glymphatic system is an anatomically established pathway. Established Perivascular transport along vessels is well supported; convective bulk flow through the parenchyma is not. Frontier The 2017 eLife test found tissue penetration strongly size-dependent — the diffusion signature — and found no significant aquaporin-4 dependence for ovalbumin or Aβ_1–40 in mice or rats. Speculative Other groups report aquaporin-4 dependence, so the correct summary is that a decade of attempted replication has not settled whether the parenchymal half of the mechanism exists.

Frontier Claim: DTI-ALPS lets us measure glymphatic function in humans without an injection. Established A 2026 Radiology comparison against intrathecal contrast MRI in 56 participants found no significant association at 48 hours in any region and concluded the index likely reflects local white-matter integrity. Frontier A 2024 critique adds that perivascular spaces are about 1% of white-matter volume and that most cortical vessels do not connect directly to the vessels ALPS measures, concluding that its validity as a marker of glymphatic clearance is highly questionable. Speculative The index is cheap, non-invasive and widely used, and those three properties explain its adoption better than its validation does.

Frontier Claim: sleep is for the brain. Established In the one organism where the lethal mechanism of sleep deprivation has been traced, it was not: flies died around day 20 against 40 for controls, reactive oxygen species accumulated in the gut, gut-targeted antioxidants restored near-normal lifespan, nervous-system-targeted ones did almost nothing, and brain tissue was indistinguishable between deprived and control animals. Speculative Sleep-deprived mice showed the same intestinal signature. Any theory of sleep that is exclusively neural has to explain that result and mostly does not cite it.

Frontier Claim: REM sleep is essential and its loss causes rebound and pathology. Established Four northern fur seals suppressed REM by 96.4 ± 1.0% for 10 to 14 days in seawater and recovered 3 ± 2% of what they lost. Speculative A vital process that can be suspended for a fortnight and not repaid is not obviously vital, and this result is fifteen years old and still largely unanswered by the restorative literature.

Frontier Claim: everyone needs about eight hours. Established Sleep duration varies more than twentyfold across animals with the same basic neurophysiology, and human carriers of single-gene variants sleep four to six hours habitually without detectable deficit. Speculative The eight-hour figure is a population average of a trait with large genetic variance, and treating it as a physiological requirement is a category error. Handwave It does not follow that anyone can train into short sleep; the carriers did not choose it.

Frontier Claim: you can catch up on lost sleep. Established The 2021 human tracer study found clearance still impaired at 48 hours after one deprived night and a recovery night, and its authors state that humans do not catch up on lost sleep. Frontier It is seven participants against seventeen controls, with no electroencephalography to exclude micro-sleep, so it is the best available evidence and it is thin. Speculative Subjective recovery and biomarker recovery are different measurements and the literature routinely conflates them.

Speculative Claim: sleeping pills deliver the benefits of sleep. Frontier The 2025 norepinephrine result reports that zolpidem suppresses the noradrenergic oscillation associated with clearance and produces about a 50% fall in prefrontal norepinephrine for hours. Speculative If that generalises to humans, sedation delivers sleep architecture without the putative function. Handwave It is one paper in mice, the human data do not exist, and both reassurance and alarm are currently unsupported.

Speculative Claim: poor sleep causes dementia. Frontier The association is robust; the causal direction is not. Amyloid and tau pathology disrupt sleep, and the earliest tau deposition is in brainstem arousal nuclei, so reverse causation is a live explanation for much of the epidemiology. Speculative The authors of the 2024 contradiction have said publicly that good sleep may reduce dementia risk for reasons other than clearing toxins, which is a claim their own result does not touch either way.

Handwave Claim: sleep will eventually be pharmacologically compressible or replaceable. Speculative Nothing on this page supports it. The existence proof is a handful of single-gene short sleepers at four to six hours; the mitochondrial account, if right, says the requirement is a consequence of aerobic metabolism rather than an adaptation. Handwave A technology that removes a thermodynamic tax is a different kind of proposition from one that removes an adaptation, and the field cannot currently say which it is dealing with.