5D Library › Two-Process Sleep Architecture, Slow-Wave Glymphatic Clearance and Thermoregulatory Phase Control · 中文
Chapter 4. Does Sleep Wash the Brain? Diffusion, Flow and an Open Dispute
The claim and its first evidence
The brain has no lymphatic vessels inside its tissue, yet it must remove the waste its cells produce, including proteins such as amyloid-β. In 2012 Iliff, Nedergaard and colleagues described a route by which cerebrospinal fluid enters the brain along the spaces around arteries, exchanges with the interstitial fluid, and leaves along veins, and called it the “glymphatic” system (Iliff et al. 2012). In 2013 Xie and colleagues reported that in mice the extracellular space expanded during natural sleep and anaesthesia — the volume fraction from about 0.14 to 0.23 — and that the clearance of injected amyloid-β was faster (Xie et al. 2013). “Sleep washes the brain” has been a popular sentence ever since.
Evidence
Supporting evidence. In mice, CSF flow in the perivascular spaces around surface arteries has been tracked directly with particles, at a mean speed of the order of 20 µm/s, driven largely by arterial pulsation (Mestre et al. 2018). In humans, functional MRI shows large, slow oscillations of CSF during non-REM sleep, coupled to the slow waves of the EEG (Fultz et al. 2019). Neither measurement is of flow through the tissue itself, which is the part in dispute.
Evidence
Contrary evidence. Using a fluorescent tracer and measurements designed to separate diffusion from flow, Miao, Franks, Wisden and colleagues found in mice that the clearance of the tracer from the brain was reduced, not increased, during sleep and anaesthesia (Miao et al. 2024). Earlier, Smith, Verkman and colleagues found that solute transport through the brain’s tissue was consistent with diffusion and did not depend on the water channel aquaporin-4 that the glymphatic model gives a central role (Smith et al. 2017), and a modelling study on reconstructed tissue found that interstitial transport occurs by diffusion rather than bulk flow at realistic pressure gradients (Holter et al. 2017). The field has not settled; this chapter does not settle it either. It asks what the dispute turns on.
Diffusion or flow?
The Péclet number compares transport by bulk flow at speed over a distance with diffusion at effective diffusivity , where is the tortuosity of the extracellular space, about 1.6 in brain tissue (Syková and Nicholson 2008). For diffusion dominates, for flow does. The seventh program computes from the Stokes-Einstein relation for three illustrative molecular sizes.
Calculation
# Chapter 4. Diffusion or flow? The Peclet number Pe = v L / D_eff compares
# transport by bulk flow at speed v over distance L with diffusion at
# effective diffusivity D_eff = D / lambda^2 (lambda = tortuosity of the
# extracellular space, taken as 1.6). Free diffusivity from Stokes-Einstein, D
# = kT / (6 pi eta r), at 37 C with the viscosity of water (0.69 mPa s), for
# three illustrative hydrodynamic radii. Bulk-flow speeds in the brain's
# interstitium are not established; we scan them.
# Then: what does a rise in extracellular volume fraction alpha from 0.14 to
# 0.23 change? Diffusivity D_eff does not contain alpha. A Kozeny-Carman
# permeability, k ~ alpha^3 / (1 - alpha)^2, does - and so does flow driven by
# a fixed pressure gradient.
import math
KT = 1.380649e-23 * 310.15
ETA = 0.69e-3
LAM = 1.6
if __name__ == "__main__":
print("A. Effective diffusivity and time to diffuse"
" a distance L (t = L^2 / 2")
print("D_eff)")
print(" radius D_eff (m^2/s) L = 50 um L = 100 um L = 200 um")
Ds = {}
for r_nm, label in (
(1.5, "small peptide"),
(3.0, "mid-sized protein"),
(5.5, "antibody-sized"),
):
D = KT / (6 * math.pi * ETA * r_nm * 1e-9) / LAM ** 2
Ds[r_nm] = D
ts = [
"{:8.0f} s".format((L * 1e-6) ** 2 / (2 * D))
for L in (50, 100, 200)
]
print(" {:>4.1f} nm {:.2e} {}".format(r_nm, D, " ".join(ts)))
print(" ({})".format(label))
print(
"\nB. Peclet number over L = 100 um for a range of bulk-flow speeds"
)
print(" v (um/s) small peptide antibody-sized")
for v in (0.01, 0.1, 0.5, 1.0, 5.0, 20.0):
print(" {:7.2f} {:9.3f} {:9.3f}".format(
v, v * 1e-6 * 100e-6 / Ds[1.5], v * 1e-6 * 100e-6 / Ds[5.5]))
for r in (1.5, 5.5):
print(
" Pe = 1 at v = {:.2f} um/s for radius {:.1f} nm".format(
Ds[r] / 100e-6 * 1e6, r
)
)
a0, a1 = 0.14, 0.23
kc = lambda a: a ** 3 / (1 - a) ** 2
print("\nC. Extracellular volume fraction 0.14 (awake) -> 0.23 (asleep)")
print(" change in D_eff (pure diffusion, tortuosity unchanged): x 1.00")
print(
" change in Kozeny-Carman permeability: x {:.2f}"
.format(kc(a1) / kc(a0))
)
print(
" change in interstitial flow speed at fixed pressure: x {:.2f}"
.format(kc(a1) / kc(a0) / (a1 / a0))
)
print("\nReading: whether a larger extracellular space"
" speeds clearance depends on")
print("which regime the tissue is in. If transport"
" is diffusive (Pe << 1) the")
print("volume change by itself does almost nothing;"
" only if bulk flow matters")
print("does it act, and then strongly. The regime is"
" the measurement in dispute.")
# A. Effective diffusivity and time to diffuse a distance L (t = L^2 / 2
# D_eff)
# radius D_eff (m^2/s) L = 50 um L = 100 um L = 200 um
# 1.5 nm 8.57e-11 15 s 58 s 233 s
# (small peptide)
# 3.0 nm 4.29e-11 29 s 117 s 467 s
# (mid-sized protein)
# 5.5 nm 2.34e-11 53 s 214 s 855 s
# (antibody-sized)
#
# B. Peclet number over L = 100 um for a range of bulk-flow speeds
# v (um/s) small peptide antibody-sized
# 0.01 0.012 0.043
# 0.10 0.117 0.428
# 0.50 0.583 2.138
# 1.00 1.166 4.277
# 5.00 5.832 21.383
# 20.00 23.327 85.532
# Pe = 1 at v = 0.86 um/s for radius 1.5 nm
# Pe = 1 at v = 0.23 um/s for radius 5.5 nm
#
# C. Extracellular volume fraction 0.14 (awake) -> 0.23 (asleep)
# change in D_eff (pure diffusion, tortuosity unchanged): x 1.00
# change in Kozeny-Carman permeability: x 5.53
# change in interstitial flow speed at fixed pressure: x 3.37
#
# Reading: whether a larger extracellular space speeds clearance depends on
# which regime the tissue is in. If transport is diffusive (Pe << 1) the
# volume change by itself does almost nothing; only if bulk flow matters
# does it act, and then strongly. The regime is the measurement in dispute.Section A shows that diffusion over the distances between vessels is not slow: a small peptide crosses 100 µm in about a minute, an antibody-sized molecule in a few minutes. Section B shows how little flow it takes to matter: for the small peptide over 100 µm, at only 0.86 µm/s; for the antibody-sized molecule, at 0.23 µm/s. Whether interstitial flow of that order exists is exactly what the contrary studies doubt.
Section C is the key to reading the dispute. A rise in the extracellular volume fraction from 0.14 to 0.23 does not change the effective diffusivity at all, unless tortuosity also changes. It does change the permeability of the tissue to flow — by a factor of 5.53 under a Kozeny-Carman relation — and the interstitial flow speed at a fixed pressure gradient by 3.37. So the 2013 finding implies faster clearance only if flow, not diffusion, is what clears the tissue.
Four hypotheses, four predictions
The eighth program solves the transport equation for a 100-µm slab of tissue under five scenarios, each one hypothesis about what sleep changes.
Calculation
# Chapter 4. Half-clearance time of a solute from a 100-um slab of tissue:
# tracer starts uniform, leaves at one face (the perivascular drain,
# concentration held at 0), and clean fluid enters at the other. Transport by
# diffusion (D_eff from p07, small peptide) plus optional bulk flow v toward
# the drain. Explicit finite volumes, 50 cells, upwind advection. Each
# "asleep" scenario is one hypothesis about what sleep changes; the program
# does not choose between them - it shows how far apart their predictions are.
import math
L, NX = 100e-6, 50
DX = L / NX
D_AWAKE = 8.57e-11 # small peptide, tortuosity 1.6 (p07)
def half_time(D, v):
"""Conservative finite volumes: clean fluid enters at x = 0, the drain
face at x = L
holds concentration 0 (half a cell away from the last cell centre)."""
dt = 0.2 * DX * DX / D
if v > 0:
dt = min(dt, 0.2 * DX / v)
c = [1.0] * NX
m0, t = sum(c), 0.0
while sum(c) > 0.5 * m0:
F = [0.0] * (NX + 1)
for j in range(1, NX):
F[j] = v * c[j - 1] - D * (c[j] - c[j - 1]) / DX
F[NX] = v * c[NX - 1] + D * c[NX - 1] / (DX / 2)
c = [c[i] + dt * (F[i] - F[i + 1]) / DX for i in range(NX)]
t += dt
return t
if __name__ == "__main__":
scen = [
(
"diffusion only; sleep changes nothing measurable",
D_AWAKE,
0.0,
D_AWAKE,
0.0,
),
(
"diffusion only; tortuosity 1.6 -> 1.5 in sleep",
D_AWAKE,
0.0,
D_AWAKE * (1.6 / 1.5) ** 2,
0.0,
),
(
"slow flow 0.3 um/s, x 3.37 in sleep (Kozeny-Carman)",
D_AWAKE,
0.3e-6,
D_AWAKE,
0.3e-6 * 3.37,
),
(
"fast flow 2 um/s, x 3.37 in sleep",
D_AWAKE,
2e-6,
D_AWAKE,
2e-6 * 3.37,
),
(
"flow 2 um/s, HALVED in sleep (the 2024 direction)",
D_AWAKE,
2e-6,
D_AWAKE,
1e-6,
),
]
print("scenario")
print(" awake t1/2 asleep t1/2 sleep speeds clearance x")
for name, Da, va, Ds, vs in scen:
ta, ts = half_time(Da, va), half_time(Ds, vs)
print(name)
print(
" {:8.0f} s {:8.0f} s {:5.2f}".format(ta, ts, ta / ts)
)
print("\nReading: the same anatomy predicts anything"
" from no effect of sleep, to")
print("about a doubling of clearance, to a slow-down,"
" depending on two quantities")
print("that are not yet agreed: the interstitial"
" flow speed and what sleep does")
print("to it.")
# scenario
# awake t1/2 asleep t1/2 sleep speeds clearance x
# diffusion only; sleep changes nothing measurable
# 23 s 23 s 1.00
# diffusion only; tortuosity 1.6 -> 1.5 in sleep
# 23 s 20 s 1.14
# slow flow 0.3 um/s, x 3.37 in sleep (Kozeny-Carman)
# 20 s 16 s 1.27
# fast flow 2 um/s, x 3.37 in sleep
# 12 s 6 s 2.16
# flow 2 um/s, HALVED in sleep (the 2024 direction)
# 12 s 16 s 0.76
#
# Reading: the same anatomy predicts anything from no effect of sleep, to
# about a doubling of clearance, to a slow-down, depending on two quantities
# that are not yet agreed: the interstitial flow speed and what sleep does
# to it.If transport is diffusive and sleep changes nothing measurable, sleep has no effect (×1.00). If sleep lowers tortuosity slightly, from 1.6 to 1.5, clearance is a little faster (×1.14). If there is slow flow that sleep speeds by the Kozeny-Carman factor, ×1.27; if there is fast flow, ×2.16. If there is fast flow that sleep halves — the direction the 2024 study points — sleep slows clearance (×0.76).
The same anatomy therefore predicts anything from a doubling of clearance in sleep to a slow-down. The quantities that decide between them — the interstitial flow speed and what sleep does to it — are the ones not yet agreed. That is a more useful conclusion than either slogan: it says which measurement would settle the question.
Hypothesis · not established
The author’s hypothesis (not established): in the same animals, with the same tracer and the same anaesthesia-free protocol, the measured change in clearance between sleep and wake is predicted by the measured change in interstitial flow speed and not by the change in extracellular volume fraction alone. What would show it to be wrong: a change in clearance that tracks the volume fraction when flow speed is measured and unchanged.
Note
What this chapter does not claim. It does not claim that sleep has no role in brain waste clearance, nor that it has one. It claims that the popular version (“sleep washes the brain”) is stated more confidently than the evidence allows, and that the evidence divides according to a measurable quantity. Nothing here supports any product or practice sold as improving brain clearance.
Summary
The glymphatic hypothesis and the finding that the extracellular space enlarges in sleep made “sleep washes the brain” popular; later studies found transport through tissue consistent with diffusion and, in one, reduced clearance during sleep. The physics shows why the evidence can point both ways. Diffusion over inter-vessel distances takes about a minute for a small peptide, and flow matters only above about 1 µm/s. An enlarged extracellular space leaves diffusivity unchanged but raises permeability to flow about fivefold, so it speeds clearance only if flow dominates. The predicted effect of sleep ranges from ×0.76 to ×2.16 depending on two unmeasured quantities.
References
Fan, Y. (2026). Two-Process Sleep Architecture, Slow-Wave Glymphatic Clearance and Thermoregulatory Phase Control: Adenosine Homeostasis, Flip-Flop Switch Bistability, Core-Distal Heat Flux and Sleep-Stage Markov Chains [双过程睡眠架构、慢波胶质淋巴清除与体温节律相位控制] (Version 1.0.0). 5D Library. https://doi.org/10.5281/zenodo.24090162 · Licensed under Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International (CC BY-NC-ND 4.0).