Glymphatic System: Brain Waste Clearance

• Glymphatic system is defined as a brain-wide fluid transport network that moves CSF through perivascular spaces to clear metabolic waste.
• It utilizes multiple mechanisms—including sleep modulation, arterial pulsatility, and peristaltic effects—to drive CSF–ISF exchange and waste removal.
• Advanced imaging and computational models enable in vivo quantification of PVS markers and transport dynamics, linking GS function to neurodegenerative diseases.

The glymphatic system (GS) is a glia-modulated, brain-wide, lymphatic-like clearance pathway that moves cerebrospinal fluid (CSF) along perivascular spaces (PVS) into the parenchyma, facilitates CSF–interstitial fluid (ISF) exchange, and supports removal of metabolic waste including amyloid‑β and tau (Zeng et al., 9 Dec 2025). In the literature summarized here, the GS is treated simultaneously as an anatomical network centered on periarterial and perivenous routes, a physiological transport system shaped by sleep, pulsatility, osmotic and hydrostatic forces, and an imaging and modeling target for quantifying brain-wide waste clearance in vivo (Elkin et al., 2018). Across recent work, the GS is also framed as a clinically relevant mechanism in neurodevelopment, neurodegeneration, small vessel disease, hydrocephalus-spectrum disorders, and optic nerve physiology, while remaining the subject of active debate regarding the relative roles of diffusion, convection, peristalsis, and oscillatory dispersion (Valnes et al., 2018).

1. Definition, anatomical organization, and physiological role

The GS is consistently defined as a brain-wide fluid transport network that moves CSF through and around brain tissue to remove metabolic waste (Elkin et al., 2018). In this framework, CSF flows from reservoirs at the brain’s surface into peri-vascular spaces surrounding arteries, penetrates brain tissue, and ultimately drains into lymphatic vessels in the meninges and neck vasculature (Elkin et al., 2018). A key function attributed to this system is clearance of soluble amyloid‑β and tau proteins, and impairment of this clearance is linked in the cited literature to toxic protein accumulation, neurodegeneration, cerebral small vessel disease, cognitive decline, depression, autism, alcohol-related brain injury, Alzheimer’s disease, Parkinson’s disease, idiopathic normal pressure hydrocephalus, and other CSF disorders (Zeng et al., 9 Dec 2025).

Perivascular spaces are central to this conception. They are described as fluid-filled channels surrounding penetrating arteries and veins in the brain parenchyma and as a core anatomical component of the GS (Sinclair et al., 2024). Within the GS hypothesis, PVS provide low-resistance pathways for CSF to enter the brain, intermix with ISF, and promote convective clearance of solutes such as amyloid‑β, tau, and other metabolites (Sinclair et al., 2024). MRI-visible enlarged PVS are therefore increasingly interpreted as imaging biomarkers of glymphatic dysfunction or overload, especially in aging, neurovascular pathology, and small vessel disease (Zeng et al., 9 Dec 2025).

The same organizing principles recur beyond the brain parenchyma. In the optic nerve, a multidomain electro-osmotic model represents CSF entry via arterial perivascular space, passage through glia and extracellular space, and exit through venous perivascular space, with astrocytic endfeet furnishing two coupled exchange routes through AQP4-mediated membranes and gaps between endfeet (Xiao et al., 25 Oct 2025). This suggests that, within the literature at hand, the GS is not merely a descriptive label for periarterial influx and perivenous efflux, but a broader transport architecture linking perivascular conduits, astrocytic interfaces, extracellular space, and lymphatic drainage.

2. Transport mechanisms and driving forces

Several distinct physical mechanisms are used to explain GS transport, and the literature does not reduce them to a single driver. One recurrent claim is that GS activity is strongly sleep-dependent: animal and human work cited in the adolescent study indicates that glymphatic activity peaks during non‑REM/slow-wave sleep and is suppressed or impaired by sleep deprivation or fragmentation (Zeng et al., 9 Dec 2025). That sleep dependence is mechanistically linked to CSF–ISF flow, arterial pulsatility, and AQP4 polarization in the same study’s sequential mediation analysis, which identifies a pathway from sleep to PVS burden to brain volume to behavior as the more plausible route among the tested models (Zeng et al., 9 Dec 2025).

A second line of work emphasizes pulsatility. The modeling study on arterial PVS flow proposes that the outer PVS boundary oscillates due to brain pulsations in addition to arterial wall motion driven by a blood pressure wave (Holba et al., 28 Apr 2025). In that lubrication-theory framework, brain pulsations substantially magnify net axial CSF flows created by arterial wall motion, and the predicted net mean flows are almost entirely positive, directed toward deeper brain, across a broad range of clinically normal parameters (Holba et al., 28 Apr 2025). The same work reports that dilated PVS, thicker arterial walls, and higher arterial stiffness strongly reduce net flow, thereby offering a hydrodynamic rationale for the clinical association between enlarged PVS and impaired clearance (Holba et al., 28 Apr 2025).

A third mechanism is oscillatory dispersion in the interstitium. The 2026 oscillatory random-walk model treats interstitial spaces as a region of dynamic oscillatory flow driven by cardiac and respiratory cycles (Raghavan et al., 19 Feb 2026). In that formulation, molecules advected through a randomly branched interstitial network undergo mechanical dispersion, yielding an effective diffusivity larger than molecular diffusion alone when observed over times long compared with an oscillation cycle (Raghavan et al., 19 Feb 2026). The model estimates cardiac-only contributions on the order of $20\,\mu\mathrm{m}^2/\mathrm{s}$ and combined cardiac-plus-respiratory contributions in the range $80\text{–}300\,\mu\mathrm{m}^2/\mathrm{s}$, which is proposed as one explanation for MRI observations of faster-than-expected brain-wide tracer spread (Raghavan et al., 19 Feb 2026). This suggests a mechanistic bridge between diffusion-dominant interpretations at the micro-scale and effective large-scale transport seen in human tracer MRI.

A fourth line concerns peristalsis. The 2026 experimental study on short annular geometries demonstrates that long-wavelength peristaltic deformation of an inner wall in a short annular channel can produce net axial transport in the direction of wave propagation despite predominantly oscillatory local motion (Salach et al., 27 May 2026). Because cerebral PVS are annular and much shorter than the arterial pulse wavelength, this result is directly relevant to the question of whether arterial wall motion can produce net periarterial transport under physiologically relevant geometric constraints (Salach et al., 27 May 2026). Taken together, these studies portray the GS as a transport regime shaped by sleep state, arterial wall motion, brain-boundary motion, electro-osmotic transendfoot exchange, and oscillatory dispersion, rather than by a single canonical pump.

3. Imaging markers and in vivo quantification

The most prominent MRI-derived structural marker in the cited literature is PVS burden. In the adolescent ABCD analysis, higher white-matter PVS burden is used as a proxy index of GS function, with normalized PVS volume defined as total PVS volume divided by total white matter volume (Zeng et al., 9 Dec 2025). PVS segmentation there uses T1w and T2w structural MRI, Enhanced PVS Contrast maps derived from filtered T1w/T2w, Frangi vesselness filtering, a white matter mask, and thresholding of the vesselness map at $1\times10^{-7}$, followed by robustness analyses across probabilistic PVS templates (Zeng et al., 9 Dec 2025). The study is explicit that this is an indirect, structural, and static marker rather than a direct measurement of CSF–ISF exchange or solute clearance (Zeng et al., 9 Dec 2025).

Automated PVS segmentation is itself a substantial research area because manual visual rating is time-consuming, subjective, and poorly scalable (Sinclair et al., 2024). The nnUNet-based PINGU model was trained on a heterogeneous sample from six different datasets and reports internal 5-fold cross-validation performance of voxel and cluster level Dice scores of $0.50 \pm 0.15$ and $0.63 \pm 0.17$ in white matter, and $0.54 \pm 0.11$ and $0.66 \pm 0.17$ in basal ganglia (Sinclair et al., 2024). On unseen sites, performance declines, but PINGU still outperforms publicly available alternatives, especially in the basal ganglia, which the paper identifies as an area of PVS related to vascular disease and pathology (Sinclair et al., 2024). Earlier Frangi-based work likewise treated PVS as small tubular structures and optimized segmentation parameters against neuroradiological ratings using ordered logit models, with segmentation-based PVS burden estimates correlating with visual assessments up to Spearman’s $\rho = 0.74$ (Ballerini et al., 2017).

Dynamic tracer MRI constitutes a second major in vivo strategy. In rodent GlymphVIS, temporal contrast-enhanced MRI after cisterna magna infusion is treated as a time series of tracer densities, and generalized regularized optimal transport is used to infer a denoised density field $\rho(t,x)$ and time-varying velocity field $v(t,x)$ satisfying an advection–diffusion equation (Elkin et al., 2018). That framework reduces registration mean-squared and infinity-norm errors by up to a factor of $80\text{–}300\,\mu\mathrm{m}^2/\mathrm{s}$0 relative to the prior state-of-the-art method and recovers known glymphatic pathways such as peri-arterial transport along the middle cerebral artery, as well as pathway reservoirs in the basal cistern and interpeduncular cistern (Elkin et al., 2018). A related regularized optimal mass transport framework extends this approach with accelerated numerical methods and Lagrangian post-processing, generating pathlines, speed maps, Péclet maps, and velocity flux vectors that distinguish advection-dominated periarterial and CSF routes from diffusion-dominated parenchymal transport (Chen et al., 2022).

Human intrathecal gadolinium MRI has also been used as a clinically accepted proxy for glymphatic and CSF transport dynamics (Rieff et al., 2024). In the U-net prediction study, gadobutrol administered into the lumbar spinal canal is tracked with serial 3D T1-weighted MRI, and the 24-hour peak tracer distribution is predicted from earlier scans (Rieff et al., 2024). Models trained using only the first two hours post-injection achieve test performance comparable to models trained with later-stage scans, and neuroradiologist ventricular reflux grading of predicted 24-hour images reaches a mean accuracy of $80\text{–}300\,\mu\mathrm{m}^2/\mathrm{s}$1 with Krippendorff’s alpha values on predictions alone of $80\text{–}300\,\mu\mathrm{m}^2/\mathrm{s}$2 under $80\text{–}300\,\mu\mathrm{m}^2/\mathrm{s}$3 training (Rieff et al., 2024). This suggests that early tracer dynamics carry substantial information about later CSF/glymphatic distribution patterns, although the study itself limits claims to prediction of spatial distribution rather than direct quantification of clearance rates (Rieff et al., 2024).

4. Computational and mathematical frameworks

Recent GS research is heavily computational, with inverse problems, reduced-order models, optimal transport, and multidomain electrodiffusion all appearing as core methodologies. In regularized optimal transport formulations, tracer density $80\text{–}300\,\mu\mathrm{m}^2/\mathrm{s}$4 evolves according to an advection–diffusion equation,

$80\text{–}300\,\mu\mathrm{m}^2/\mathrm{s}$5

while the inferred flow minimizes a kinetic-energy functional plus a data-fidelity term (Elkin et al., 2018). This same line of work supports time-continuous interpolation between observed MRI frames and explicit trajectory analysis, thereby converting intensity-based image sequences into physically constrained transport fields (Elkin et al., 2018).

A related theoretical development is the Fisher–Rao regularized transport formulation, which introduces an augmented velocity

$80\text{–}300\,\mu\mathrm{m}^2/\mathrm{s}$6

so that the advection–diffusion constraint becomes a conservation law in terms of $80\text{–}300\,\mu\mathrm{m}^2/\mathrm{s}$7 (Elkin et al., 2019). In this framework, Lagrangian pathlines are computed by integrating

$80\text{–}300\,\mu\mathrm{m}^2/\mathrm{s}$8

and the resulting trajectories distinguish transport under dexmedetomidine plus low-dose isoflurane from transport under high-dose isoflurane, with stronger peri-arterial and parenchymal transport under the sleep-like anesthetic state and more vigorous nasal efflux under high-dose isoflurane (Elkin et al., 2019). The paper also shows equivalence between the advection–diffusion regularized OMT problem and a Fisher–Rao regularized or Schrödinger-bridge formulation, clarifying the theoretical status of these GS flow reconstructions (Elkin et al., 2019).

The 2026 subject-specific reconstruction framework advances this line by estimating spatially varying velocity, diffusivity, and clearance parameters from contrast-enhanced MRI while explicitly enforcing mass conservation (Bakiler et al., 1 May 2026). The concentration field satisfies

$80\text{–}300\,\mu\mathrm{m}^2/\mathrm{s}$9

with Robin boundary clearance

$1\times10^{-7}$0

and the velocity is decomposed into a data-informed part and a corrective gradient field to recover a weakly divergence-free flow (Bakiler et al., 1 May 2026). The system is discretized using immersed isogeometric analysis with quadratic B-spline basis functions, and forward simulations with the recovered fields show close agreement with experimental observations while preserving tracer conservation (Bakiler et al., 1 May 2026). A plausible implication is that physically constrained inverse problems can move the field beyond arbitrary-unit flow visualizations toward subject-specific quantitative transport maps.

Multidomain continuum modeling appears in the optic nerve studies. The 2024 potassium-clearance model represents axons, glia, extracellular space, and arterial, venous, and capillary PVS as coupled domains with volume fractions, fluid velocities, ion concentrations, and membrane fluxes (Xiao et al., 2024). The 2025 electro-osmotic extension adds hydrostatic, osmotic, and electro-diffusive coupling and shows that AQP4-mediated transport and gaps between glial endfeet implement two coupled glymphatic exchange routes (Xiao et al., 25 Oct 2025). Parameter sweeps in that model show that reducing AQP4-mediated permeability or PVS permeability elevates pressure, suppresses radial exchange, and slows clearance, with PVS-mediated export emerging as the dominant route for small and moderate neutral solutes (Xiao et al., 25 Oct 2025).

Reduced-order modeling addresses the scaling problem of patient-specific GS simulation. Geometry Reduced Order Modeling uses MRI-based inter-brain mappings to transfer previously computed solutions onto new geometries and is applied to two glymphatic-function example problems across 101 human MRI scans, including a patient with no known neurological disease and a patient with idiopathic normal pressure hydrocephalus displaying significantly enlarged ventricles (Solheim et al., 11 Jun 2025). Because the available text for that paper is only a stub, any stronger statement about its governing equations or numerical performance would exceed the provided data.

5. Development, sleep, cognition, and disease associations

The adolescent study provides the clearest developmental account. In 6,800 participants from the ABCD baseline cohort, adolescents with insufficient sleep, defined as $1\times10^{-7}$1 h/night, had higher normalized white matter PVS burden than those with sufficient sleep, defined as $1\times10^{-7}$2 h/night (Zeng et al., 9 Dec 2025). Reported mean normalized PVS values were $1\times10^{-7}$3 in the sleep-insufficient group and $1\times10^{-7}$4 in the sleep-sufficient group, with a group-difference estimate $1\times10^{-7}$5, $1\times10^{-7}$6, $1\times10^{-7}$7, and Cohen’s $1\times10^{-7}$8 (Zeng et al., 9 Dec 2025). Shorter sleep duration and worse subjective sleep disturbance were both associated with higher PVS burden, and mediation analyses showed that PVS burden partially mediated sleep effects on crystallized intelligence, episodic memory, executive function, visual processing, and psychosis symptom severity, with indirect proportions up to $1\times10^{-7}$9 (Zeng et al., 9 Dec 2025).

That study also reported broader structural and behavioral associations: adolescents with insufficient sleep had reduced cortical, subcortical, and white matter volumes, poorer cognitive performance across multiple domains, and elevated psychopathology, with the largest cognitive effect in crystallized intelligence and the largest psychopathology effect in general problems (Zeng et al., 9 Dec 2025). Sequential mediation favored the order sleep $0.50 \pm 0.15$0 PVS $0.50 \pm 0.15$1 brain volume $0.50 \pm 0.15$2 behavior over the reverse ordering, although the authors explicitly cautioned that the data are cross-sectional and therefore not causal proof (Zeng et al., 9 Dec 2025). This suggests that, at least in adolescence, the GS may function as a developmental vulnerability point linking sleep adequacy to neurodevelopment and mental health.

In adult and aging-related disease, enlarged PVS are repeatedly associated with pathology. The PINGU and Frangi-filter papers both frame MRI-visible PVS as markers of glymphatic dysfunction, neurovascular-unit pathology, and small vessel disease (Sinclair et al., 2024). The PINGU synthesis specifically distinguishes basal ganglia PVS as more linked to vascular and small-vessel disease, hypertension, stroke, and vascular dementia, whereas white-matter PVS are more often associated with amyloid and neurodegenerative pathology (Sinclair et al., 2024). The hydrodynamic study on brain pulsations offers a mechanistic rationale for that association by showing that net axial CSF flow decreases sharply as PVS thickness increases and also declines with arterial stiffening and thickening, both of which are characteristic of vascular aging and hypertension (Holba et al., 28 Apr 2025).

Human CSF tracer MRI extends the disease relevance beyond PVS burden alone. Ventricular reflux grading, defined on a $0.50 \pm 0.15$3–$0.50 \pm 0.15$4 scale from no supra-aqueductal reflux to lasting 24-hour ventricular enrichment isointense with subarachnoid CSF, is treated as a clinically meaningful surrogate of altered CSF and glymphatic flow (Rieff et al., 2024). High ventricular reflux grades are linked in that study to raised intracranial pressure pulsatility, shunt-responsive idiopathic normal pressure hydrocephalus, reduced tracer clearance, and disturbed CSF dynamics (Rieff et al., 2024). The deep-learning results therefore position GS-related tracer patterns not merely as descriptive images, but as disease-sensitive phenotypes that can potentially be forecast from early post-injection scans (Rieff et al., 2024).

The optic nerve models broaden the disease context further by connecting glymphatic-like transport to ionic homeostasis, pressure regulation, and size-dependent solute clearance (Xiao et al., 25 Oct 2025). They do not provide clinical outcome statistics, but they do argue mechanistically that impaired AQP4 function or reduced PVS permeability would slow clearance and thereby furnish a substrate for CNS pathology (Xiao et al., 25 Oct 2025). A plausible implication is that some disease associations attributed to “glymphatic failure” may arise from specific bottlenecks—AQP4 polarization, PVS patency, or transendfoot exchange—rather than from a unitary loss of whole-system transport.

6. Controversies, surrogate measures, and open problems

A central controversy concerns whether diffusion alone can explain brain-wide tracer distribution. The human PDE-constrained optimization study explicitly tested a diffusion-only model for intrathecal Gadobutrol spread and found that the apparent diffusion coefficient required to fit the MRI data was $0.50 \pm 0.15$5 larger than the DTI-derived estimate in grey matter and $0.50 \pm 0.15$6 larger in white matter (Valnes et al., 2018). The authors concluded that diffusion may not be the only mechanism governing transport and interpreted the elevated effective diffusivity, especially in white matter, as supportive of a slow glymphatic contribution at longer time scales (Valnes et al., 2018). At the same time, the paper did not claim direct proof of a specific bulk-flow pathway, only that classical diffusion as captured by DTI-derived estimates was insufficient (Valnes et al., 2018).

The PVS marker literature introduces a second controversy: MRI-visible PVS burden is often used as a glymphatic surrogate, but the adolescent study is explicit that PVS burden is indirect, static, and not a direct measure of CSF–ISF exchange rate, solute clearance, or GS dynamics (Zeng et al., 9 Dec 2025). Enlarged PVS may also reflect compensatory enlargement, vascular pathology, or non-GS processes (Zeng et al., 9 Dec 2025). The same paper reports that PVS volume was positively correlated with brain volumes in this age group, implying that increased PVS is not straightforwardly equivalent to tissue damage (Zeng et al., 9 Dec 2025). This cautions against treating PVS burden as a direct scalar readout of “glymphatic performance.”

A third unresolved question concerns the dominant physical driver of periarterial transport. The literature summarized here contains supportive evidence for net inward periarterial flow from both theory and experiment: brain pulsations magnify net axial flow in PVS (Holba et al., 28 Apr 2025), and long-wavelength peristalsis in short annular channels yields net transport in the direction of wave propagation (Salach et al., 27 May 2026). Yet the diffusion-versus-convection debate persists because oscillatory mechanical dispersion can also generate substantial effective diffusivity without requiring sustained directional interstitial bulk flow (Raghavan et al., 19 Feb 2026). This suggests that apparently incompatible positions in the GS debate may partly concern different spatial scales and observables: directional periarterial inflow, oscillatory parenchymal dispersion, and effective tissue-scale transport need not be mutually exclusive descriptions.

Methodological limitations are equally prominent. Subjective sleep measures, absence of polysomnography or actigraphy, and cross-sectional design constrain causal inference in the adolescent work (Zeng et al., 9 Dec 2025). U-net prediction of 24-hour tracer distribution is limited to 2D slice-based outputs and does not directly estimate clearance kinetics (Rieff et al., 2024). Optimal-transport reconstructions depend on assumptions such as proportionality between MRI intensity and tracer density, a scalar diffusion coefficient, and advection–diffusion closure (Elkin et al., 2018). PVS segmentation methods remain vulnerable to domain shift across scanners and sites (Sinclair et al., 2024). The subject-specific inverse framework addresses some of these issues through mass-conserving velocity reconstruction and smooth B-spline regularization, but it is still calibrated on contrast-enhanced imaging rather than direct endogenous waste measurements (Bakiler et al., 1 May 2026).

Future directions recur across the cited works. The adolescent study explicitly calls for objective sleep measures and direct GS measures such as DTI‑ALPS and CSF-flow imaging (Zeng et al., 9 Dec 2025). The CSF-tracer prediction study points to 3D architectures, modeling beyond 24 hours, and integration with multimodal biomarkers (Rieff et al., 2024). The hydrodynamic and peristaltic studies imply that anatomically realistic eccentricity, porosity, permeability, and 3D flow measurement remain open experimental targets (Holba et al., 28 Apr 2025). The inverse and reduced-order modeling studies suggest that large-scale subject-specific GS inference may become feasible when physically constrained imaging models, smooth discretizations, and geometry transfer are combined (Bakiler et al., 1 May 2026). Taken together, the present literature portrays the glymphatic system as an increasingly quantifiable but still conceptually plural field, in which anatomy, sleep, vascular mechanics, interstitial transport, and imaging surrogates are converging without yet yielding a single settled model of brain waste clearance.