Glymphatic system

The glymphatic system is a proposed brain-wide perivascular fluid-transport and solute-clearance process in which subarachnoid cerebrospinal fluid (CSF) enters periarterial spaces, exchanges with brain interstitial fluid (ISF), and contributes to removal of interstitial solutes toward CSF, perivascular, perineural, venous, and lymphatic outflow routes.¹² The strongest direct mechanistic evidence comes from tracer experiments in rodents. Human studies establish sleep-coupled CSF and vascular dynamics and provide some tracer-clearance evidence, but no non-invasive clinical test currently measures all three stages—CSF influx, parenchymal exchange, and net efflux—with validated specificity.

The broad existence of perivascular CSF–ISF transport is not the main controversy. The unresolved questions are how much solute moves by advection, dispersion, or diffusion within parenchyma; how AQP4 facilitates transport; which efflux routes dominate; and whether sleep increases net clearance rather than only tracer influx or fluid oscillation. The name “system” should not be read as a discrete vessel network equivalent to the peripheral lymphatic system.

Boundary: glymphatic transport versus meningeal lymphatics

These are connected but distinct anatomical/process layers:

Layer	Location and structure	Primary role	What it is not
Glymphatic / paravascular transport	Perivascular spaces and extracellular compartments within the brain, closely associated with [[astrocytes	astrocytic]] endfeet	CSF influx, CSF–ISF exchange, redistribution and clearance of interstitial solutes
Meningeal lymphatics	True lymphatic endothelial vessels in the dura, especially near venous sinuses and meningeal arteries	Drain CSF-associated fluid, macromolecules, and immune cells toward cervical lymph nodes	The route by which CSF first enters or crosses brain parenchyma

Independent mouse studies in 2015 identified functional dural lymphatic vessels, and later histology and contrast-enhanced MRI demonstrated homologous dural lymphatic structures in humans and common marmosets.²³⁴ Glymphatic transport may deliver material from parenchyma back to CSF and border compartments; meningeal lymphatics are one downstream exit route. Other routes—including cribriform/perineural drainage, arachnoid granulations, vascular basement membranes, and transport across the blood–brain barrier—operate in parallel. “Glymphatic clearance” therefore should not be used as a synonym for all brain waste removal.

Mechanistic model

1. CSF influx

Iliff et al. used intracisternal fluorescent tracers, two-photon microscopy, and radiotracer assays in mice to describe rapid movement of subarachnoid CSF along periarterial spaces into brain tissue.¹ Perivascular spaces provide lower-resistance routes than neuropil. Tracer arrival depends strongly on injection method, anesthesia, arousal state, and vascular dynamics; tracer influx is not itself proof of net waste clearance.⁵⁶⁷

2. CSF–ISF exchange

At penetrating vessels, water and solutes exchange between perivascular and interstitial compartments. Small solutes diffuse readily through extracellular space; oscillatory fluid movement can add dispersion, and pressure gradients may add local advection. Whether there is substantial, coherent convective bulk flow across the neuropil remains contested. Computational reconstruction of neuropil predicts diffusion-dominated interstitial solute transport, while live perivascular imaging supports directed flow within periarterial spaces.⁸⁷

The defensible synthesis is scale- and compartment-specific: directed flow in perivascular conduits can coexist with diffusion and dispersion as major modes within the narrow extracellular space. A single label such as “bulk flow” obscures this distinction.

3. Efflux and disposal

Interstitial solutes can return to CSF/perivascular compartments and then leave through meningeal lymphatic and perineural routes, vascular basement membranes, venous-associated spaces, or blood–brain barrier transport. Iliff et al. reported slower clearance of intraparenchymally injected mannitol and amyloid-β in Aqp4-null mice, connecting the pathway to interstitial efflux.¹ However, route proportions vary by solute size, charge, binding, brain region, and experimental preparation. A result showing faster CSF tracer entry does not establish faster exit of an endogenous protein.

Aquaporin-4: facilitation, not a solute pump

Aquaporin-4 (AQP4) is a water channel concentrated at astrocytic perivascular endfeet. It does not transport amyloid-β, tau, or other macromolecules directly. The original mouse work found that Aqp4 deletion reduced CSF–ISF exchange and interstitial-solute clearance.¹ A later study reported diffusion-dominated, AQP4-independent transport under its experimental conditions,⁹ prompting a five-center re-examination in mice and rats. That multicenter study again found impaired tracer influx or efflux in Aqp4-deficient animals and attributed part of the discordance to anesthesia, age, tracer delivery, and tissue injury from intraparenchymal injection.¹⁰

Thus, AQP4 involvement is supported across multiple laboratories, but the mechanism and effect size are not settled. Loss of the perivascular polarization of AQP4 may matter more than total abundance, and a global knockout does not isolate perivascular water permeability from other consequences of lifelong AQP4 deficiency. AQP4 dependence does not by itself prove trans-astrocytic bulk flow. gap/contradictory-evidence

Physiological regulation

Sleep, arousal, anesthesia, and norepinephrine

Xie et al. reported that natural sleep or ketamine/xylazine anesthesia in mice expanded extracellular space and increased CSF tracer exchange, with faster removal of interstitial amyloid-β than during wakefulness.⁵ Anesthetic regimen matters: across six regimens, tracer influx tracked greater electroencephalographic delta power and lower heart rate, so “anesthetized” cannot be treated as a uniform physiological condition.⁶

In freely sleeping mice, Hauglund et al. found coupled infraslow oscillations in norepinephrine, cerebral blood volume, and CSF during non-rapid-eye-movement sleep. Optogenetic manipulation supported a causal sequence in which norepinephrine-linked slow vasomotion pumps CSF influx; 5 mg/kg zolpidem suppressed the oscillations and reduced intracisternal fluorescent-tracer influx, while a separate intraparenchymal radiotracer experiment related clearance to sleep microarchitecture.¹¹ This favors preserved sleep microarchitecture over a simple equation of unconsciousness with clearance.

Net clearance during sleep is nevertheless actively disputed. Miao et al. injected a small fluorescent dye into mouse caudate-putamen and reported unchanged spreading but reduced disappearance during sleep and three anesthetic conditions—the opposite direction from the canonical model.¹² The paper received a source-data correction, and a 2025 Matters Arising exchange disputed its flow model, sampling, analysis, and the broader assumption that CSF tracer entry measures clearance; the authors defended their interpretation.¹³ The experiments differ in tracer, injection compartment, endpoint, and spatial scale. Current evidence supports sleep-linked changes in fluid dynamics, but does not justify treating increased influx, oscillation amplitude, and net parenchymal efflux as interchangeable. gap/contradictory-evidence

Circadian timing

In anesthetized mice, CSF-tracer influx and brain-to-blood clearance of intraparenchymal Evans blue were greater during the habitual rest phase, whereas mandibular lymph-node filling after cisterna-magna injection showed the opposite rhythm and was greater during the active phase. Influx, Evans-blue clearance, lymph-node filling, and perivascular AQP4-polarization rhythms persisted under constant light; Aqp4 deletion abolished the day–night difference in influx and lymph-node filling.¹⁴ The study measured rhythmic astrocyte-enriched gene expression but did not itself disrupt the astrocyte molecular clock. Because mice are nocturnal and most experiments occur during their light/rest phase, clock time is a major reproducibility variable. Human circadian control of net solute clearance has not been established directly. gap/needs-human-replication

Vascular pulsatility, vasomotion, and respiration

Particle-tracking velocimetry in awake mice showed net periarterial CSF flow synchronous with the cardiac cycle; experimentally altered arterial pulsatility changed flow, and hypertension increased backflow and reduced net movement.⁷ Slower vasomotor oscillations appear especially important during natural sleep.¹¹

In 10 healthy adults, real-time MRI showed that forced inspiration strongly displaced ventricular CSF, while cardiac pulsation contributed a faster but smaller component.¹⁵ These large-compartment CSF movements may alter pressure gradients relevant to perivascular exchange, but they do not demonstrate parenchymal solute clearance or even net ventricular flow. Respiration, arterial pulse, vasomotion, venous volume, and neural activity operate at different frequencies and should be measured rather than collapsed into a single “pump.”

Aging

Kress et al. compared young (2–3 months), middle-aged (10–12 months), and old mice (18 months in the main Results and figure captions; 18–20 months in the abstract). Old mice showed reduced CSF tracer penetration, approximately 40% less clearance of intraparenchymal amyloid-β, approximately 27% lower penetrating-arteriole wall pulsatility, and loss of perivascular AQP4 polarization.¹⁶ These observations connect vascular and astrocytic aging with impaired transport, but do not establish which change is upstream.

Human aging evidence is less direct. In 35 patients undergoing serial MRI after intrathecal contrast for clinical indications, older age was associated with slower 39-hour tracer clearance from predefined glymphatic-pathway regions and putative meningeal lymphatic vessels; clearance measures in the two compartments were correlated.¹⁷ This was an observational, selected clinical cohort rather than a healthy lifespan sample. Other age-associated imaging measures mix clearance biology with atrophy, small-vessel disease, white-matter geometry, and ventricular enlargement. Longitudinal human tracer studies across healthy adulthood are lacking. gap/needs-human-replication

The main aging relevance is to loss-of-proteostasis: impaired removal could contribute to accumulation of amyloid-β, tau, and other extracellular solutes. It is one route among several, alongside blood–brain barrier transport, enzymatic degradation, phagocytosis, and intracellular autophagy. The causal share attributable to glymphatic impairment in sporadic neurodegeneration is unknown.

Human evidence

Human evidence has advanced beyond anatomy, but each modality answers a narrower question:

Simultaneous EEG and fast functional MRI demonstrated that non-rapid-eye-movement sleep couples slow neural activity, hemodynamic changes, and large CSF oscillations.¹⁸ This establishes sleep-linked hydrodynamics, not solute removal.
A 2026 multimodal EEG–functional-MRI–near-infrared-spectroscopy study found that neurovascular and water-signal coupling becomes more bidirectional during human sleep, consistent with a larger role for vasomotor-driven hydrodynamic waves; it did not measure tracer or waste clearance.¹⁹
In 24 patients already undergoing clinical evaluation for possible CSF disorders, intrathecal gadobutrol MRI found greater 24-hour brain tracer retention after total sleep deprivation (n=7) than after unrestricted sleep (n=17). The groups were not randomized, sleep was not objectively staged, and intrathecal contrast use limits healthy-population generalization.²⁰
A 2026 randomized crossover analysis of 39 healthy adults aged 49–66 linked sleep physiology, an investigational in-ear impedance-derived estimate of parenchymal resistance, and morning plasma amyloid/tau changes in a direction consistent with sleep-active brain-to-blood clearance.²¹ The inference depends on a compartment model and indirect device/biomarker measures; the reported significance tests were not multiplicity-adjusted, this was not direct tracking of endogenous proteins from brain to plasma, and multiple authors disclosed financial or stock-option interests.

Together, these data support state-dependent human brain-fluid physiology and make sleep-dependent clearance plausible. They do not yet provide a calibrated individual-level measure of glymphatic capacity or show that increasing a proxy improves cognition or slows neurodegeneration.

Measurement hierarchy

Evidence tier	Example	What it can establish	Main limitation
1. Direct tracer transport with temporal sampling	Animal two-photon microscopy, radiotracer recovery, serial tissue/lymph-node sampling	Compartment-specific influx, distribution, or efflux under controlled perturbation	Invasive injection perturbs pressure/tissue; anesthesia and species effects
2. Serial intrathecal contrast MRI	Human gadobutrol enrichment and washout over 24–48 h	Macroscopic movement and persistence of a CSF tracer in vivo	Off-label/invasive; semiquantitative; usually CSF-disorder cohorts; not an endogenous protein
3. Dynamic fluid/vascular coupling	Fast MRI + EEG, phase-contrast MRI, near-infrared water signals	CSF displacement and coupling to neural, vascular, respiratory, or sleep states	Movement/oscillation is not net clearance
4. Static or local imaging proxies	[[methods/dti-alps	DTI-ALPS]], visible perivascular spaces, free-water measures	Reproducible imaging associations within a matched protocol
5. Downstream associations	Plasma/CSF biomarkers, sleep scores, cognition, disease status	Hypothesis generation and clinical correlation	Multiple causal paths; cannot validate the transport mechanism alone

DTI-ALPS deserves special restraint. It is a local periventricular directional-diffusivity ratio, not a flow or clearance assay. In a 2026 direct comparison, it had limited correspondence with intrathecal contrast-enhanced MRI and no association with 48-hour tracer dynamics.²² Disease studies should report “higher/lower ALPS index,” reserving “glymphatic dysfunction” for a qualified interpretation supported by orthogonal evidence.

Disease associations

Disease-specific proxy associations require method-specific interpretation: animal tracer or genetic perturbation can test a transport mechanism, whereas cross-sectional DTI-ALPS or visible-perivascular-space associations cannot establish clearance failure or its causal direction.

In ME/CFS, one 2026 single-center study reported a lower bilateral DTI-ALPS index in 31 cases than 27 controls and correlations with single-item sleep-disturbance and concentration ratings.²³ The separate WHODAS cognition score and fatigue association were null. This is a preliminary local diffusion phenotype, not evidence that ME/CFS causes glymphatic clearance failure, waste accumulation, neuroinflammation, or accelerated aging. gap/needs-replication

Intervention hypotheses

No intervention is clinically validated to improve aging outcomes by selectively enhancing the glymphatic system; accordingly, druggability-tier is null.

Hypothesis	Evidence boundary
Preserve physiological [[interventions/lifestyle/sleep	sleep]] and treat diagnosed sleep disorders
Maintain vascular health and pulsatility	Hypertension and altered pulsatility impair perivascular flow in animals; no trial shows that changing a glymphatic endpoint mediates cognitive benefit
Modulate norepinephrine-linked vasomotion	Causal mouse evidence; sedatives/hypnotics can distort natural oscillations, so drug-induced unconsciousness cannot be assumed beneficial
Restore AQP4 polarization or enhance meningeal lymphatics	Preclinical target; no selective, validated aging therapy or human target-engagement assay
Controlled breathing	Forced inspiration changes ventricular CSF displacement in humans,¹⁵ but net parenchymal solute clearance and durable clinical outcomes remain untested

These are research hypotheses, not reasons to use a sedative, alter breathing, or pursue an unvalidated drug. A convincing intervention study would show target engagement with a direct transport measure, an orthogonal endogenous-solute endpoint, and a clinically meaningful outcome.

Model-organism extrapolation

Dimension	Assessment	Rationale
Dural meningeal lymphatic anatomy conserved in humans?	yes	Human and common-marmoset histology and MRI support homologous dural lymphatic structures⁴
Rodent tracer kinetics quantitatively transferable?	no	Brain scale, cortical geometry, CSF production, posture, sleep architecture, vascular dynamics, and anesthesia differ
AQP4 mechanism conserved?	plausible but not causally established in living humans	Human post-mortem localization and genetics are not equivalent to conditional intervention
Sleep-linked fluid oscillations replicated in humans?	yes	EEG–MRI and multimodal studies show coupled CSF/hemodynamic dynamics
Sleep-enhanced net endogenous-solute clearance replicated in healthy humans?	partial/contested	Human tracer and biomarker studies support the model but remain indirect, selected, or model-dependent

Preclinical studies should report strain, sex, age, clock time, arousal scoring, anesthetic regimen, injection volume/rate/pressure, head position, tracer properties, and whether the endpoint measures entry, redistribution, or exit. Without these details, apparently contradictory studies may be measuring different processes.

Limitations and gaps

Net transport mechanism: the relative contributions of advection, dispersion, and diffusion across perivascular versus interstitial compartments are unresolved. gap/contradictory-evidence
Sleep directionality: increased CSF influx and oscillations during sleep are better supported than increased net parenchymal efflux; the Miao 2024 result and 2025 exchange require direct replication with matched tracers and separate entry/exit measurements. gap/contradictory-evidence
Human reference standard: no safe, standardized, non-invasive assay measures influx, exchange, and efflux across the whole human brain. gap/no-mechanism
Healthy aging trajectory: longitudinal direct-tracer data across adult human aging do not exist. gap/needs-human-replication
AQP4 causality in humans: association of depolarized AQP4 with aging/disease does not establish that restoring polarization improves clearance or outcomes. gap/needs-human-replication
Route accounting: the fractions cleared through meningeal lymphatics, cribriform pathways, vascular basement membranes, blood–brain barrier transport, and CSF-to-blood routes remain solute- and state-dependent.
Clinical mediation: no randomized trial has shown that changing a validated glymphatic measure mediates reduced dementia incidence, slower cognitive decline, or longer healthspan. gap/long-term-unknown

Footnotes

doi:10.1126/scitranslmed.3003748 · Iliff JJ et al. · in-vivo + ex-vivo tracer study · model: adult mice, including Aqp4-null · established periarterial CSF influx, interstitial-solute clearance, and AQP4-sensitive transport ↩ ↩² ↩³ ↩⁴
doi:10.1038/nature14432 · Louveau A et al. · anatomical + functional tracer study · model: mouse · identified functional lymphatic vessels in dura; published corrigendum added historical references (doi:10.1038/nature16999) ↩ ↩²
doi:10.1084/jem.20142290 · Aspelund A et al. · anatomical + functional study · model: mouse · independently described dural lymphatic drainage of brain ISF/macromolecules ↩
doi:10.7554/eLife.29738 · Absinta M et al. · contrast-enhanced MRI + immunohistochemistry · model: humans and common marmosets · demonstrated dural structures with lymphatic endothelial markers alongside venous sinuses and their non-invasive MRI visualization ↩ ↩²
xie-2013-sleep-glymphatic-clearance · Xie L et al. · experiment-specific cohorts; terminal clearance n=77 · in-vivo two-photon + iontophoresis + radiotracer clearance · model: male C57BL/6 mice, 10–12 weeks, during wake, natural sleep, and anesthesia · reported larger extracellular space, increased CSF tracer exchange, and faster injected-solute clearance during sleep/anesthesia · doi:10.1126/science.1241224 ↩ ↩²
doi:10.1126/sciadv.aav5447 · Hablitz LM et al. · in-vivo tracer + EEG/physiology across six anesthetic regimens · model: mouse · influx correlated with high delta power and low heart rate ↩ ↩²
doi:10.1038/s41467-018-07318-3 · Mestre H et al. · two-photon particle-tracking velocimetry · model: awake mice · demonstrated pulsatile net periarterial CSF flow and impaired net flow with hypertension ↩ ↩² ↩³
doi:10.1073/pnas.1706942114 · Holter KE et al. · three-dimensional neuropil reconstruction + computational fluid modeling · predicted diffusion rather than pressure-driven bulk flow in parenchymal extracellular space ↩
doi:10.7554/eLife.27679 · Smith AJ et al. · in-vivo/ex-vivo tracer experiments · model: mouse · reported diffusion-dominated, AQP4-independent parenchymal solute transport under the tested conditions ↩
doi:10.7554/eLife.40070 · Mestre H et al. · five-center in-vivo/ex-vivo replication + meta-analysis · model: Aqp4-deficient mice and rats · supported AQP4-sensitive influx/efflux while identifying anesthesia, age, injection, and tissue injury as heterogeneity sources ↩
doi:10.1016/j.cell.2024.11.027 · multimodal recording, optogenetics, vascular manipulation, fluorescent-tracer influx, and intraparenchymal radiotracer clearance · model: freely sleeping mice · linked norepinephrine oscillations and slow vasomotion to CSF influx; 5 mg/kg zolpidem reduced oscillations and fluorescent-tracer influx ↩ ↩²
doi:10.1038/s41593-024-01638-y · intraparenchymal small-dye photometry + histology · model: male mice · reported reduced net dye clearance during sleep and dexmedetomidine, ketamine/xylazine, or pentobarbital anesthesia; source-data correction doi:10.1038/s41593-024-01698-0 ↩
doi:10.1038/s41593-025-01897-3 · 2025 Matters Arising challenging Miao et al.’s methods/analysis · reply defending the study: doi:10.1038/s41593-025-01898-2 ↩
doi:10.1038/s41467-020-18115-2 · Hablitz LM et al. · time-of-day tracer influx + intraparenchymal Evans-blue clearance + lymph-node filling + Aqp4 deletion · model: mice under anesthesia · glymphatic influx/clearance peaked in the rest phase while lymph-node filling peaked in the active phase; rhythms persisted under constant light ↩
doi:10.1523/JNEUROSCI.3246-14.2015 · n=10 · real-time MRI with forced-breathing and breath-hold protocols · model: healthy human ventricular CSF dynamics · identified a strong inspiration-linked displacement component, not net flow or parenchymal solute clearance ↩ ↩²
kress-2014-glymphatic-aging · Kress BT et al. · experiment-specific n=4–20/group · in-vivo + ex-vivo age comparison · model: mice aged 2–3, 10–12, and 18 months in the main experiments · reported impaired CSF–ISF exchange/injected amyloid-β clearance, reduced arterial pulsatility, and lost perivascular AQP4 polarization with age · doi:10.1002/ana.24271 ↩
doi:10.1002/ana.25670 · n=35 · observational serial intrathecal-contrast MRI · model: humans undergoing clinical evaluation · older age associated with slower 39-hour clearance measures in predefined glymphatic-pathway regions and putative meningeal lymphatic vessels ↩
doi:10.1126/science.aax5440 · Fultz NE et al. · simultaneous EEG + fast fMRI · model: sleeping humans · demonstrated coupled neural, hemodynamic, and CSF oscillations during non-rapid-eye-movement sleep ↩
doi:10.1073/pnas.2510731123 · n=24 · Väyrynen T et al. · simultaneous EEG, functional MRI, and functional near-infrared spectroscopy · model: healthy sleeping humans · found sleep-state changes in directed coupling among electrophysiology, blood oxygenation, and water signals; no tracer endpoint · PubMed links a 2026 commentary (doi:10.1073/pnas.2608470123) ↩
doi:10.1093/brain/awaa443 · n=24 (7 total-sleep-deprivation, 17 unrestricted sleep) · non-randomized intervention with serial intrathecal-gadobutrol MRI · model: humans under evaluation for possible CSF disorders · greater brain tracer retention after sleep deprivation ↩
doi:10.1038/s41467-026-68374-8 · n=39 analyzed · randomized crossover sleep vs total-sleep-deprivation analysis · model: healthy adults aged 49–66 · investigational impedance/EEG device + plasma amyloid/tau + compartment modeling; indirect/model-dependent clearance inference; multiple authors disclosed financial or stock-option interests ↩
dti-alps · Mossige I et al. · n=56 · prospective-observational comparison of DTI-ALPS with intrathecal gadobutrol-enhanced MRI · limited correspondence and no 48-hour tracer-dynamics association · doi:10.1148/radiol.252070 ↩
thapaliya-2026-me-cfs-dti-alps · n=58 analyzed (31 ME/CFS, 27 controls) · cross-sectional · lower bilateral DTI-ALPS index in ME/CFS with limited symptom-item correlations and important null findings · doi:10.3389/fnins.2026.1875420 ↩

🧬 Aging Wiki

Explorer

glymphatic-system