A curious concept of CNS clearance

Over the past decade, multiple lines of research have shown that sleep decreases amyloid-β and tau burden compared with wakefulness, and glymphatic clearance is increased during sleep in both human and mouse brains^1–6. A recent study by Miao et al.⁷ has questioned these findings. We here raise concerns regarding experimental methodology, analytical rigor, and theoretical and mathematical assumptions in the Miao et al.⁷ study. The conclusion of that study – brain clearance is reduced during sleep and anesthesia – is not supported by the data presented.

1. Conceptual misunderstanding of brain clearance.

Metabolic waste clearance from the brain is a process whereby proteins are degraded in situ (by ubiquitination or autophagy) or exported from the brain to the periphery⁸. Brain clearance is experimentally defined as a decline in reference molecules in brain homogenates or as their evacuation to peripheral tissues⁸. Miao et al.⁷ instead measure dye displacement between caudate-putamen and an optical probe placed in the prefrontal cortex, thus defining clearance as redistribution of the tracer within brain. This is fundamentally flawed. Moving garbage around your home will not eliminate your waste problem. Waste must be flushed out of the brain to be cleared.

2. Misrepresenting modeling of brain clearance mechanisms.

Clearance of the tracer occurs via a combination of advection (clearance by bulk fluid motion) and diffusion in brain. Miao et al.⁷ estimate the tracer diffusivity by fitting experimental measurements to Eq. 8, which is not a solution of a diffusion or advection–diffusion equation and is at odds with centuries of established physics⁹. Miao et al.⁷ give no justification for the mathematical form of Eq. 8, which attempts to model advective transport as a reaction term. The authors argue that their measurements are inconsistent with a naive approximation of transport by pure diffusion, thus an unspecified mechanism that is neither diffusion nor advection must be responsible. They define this mechanism as “clearance”, contradicting accepted definitions, ignoring the possibility that diffusivity and advection may change between wake and anesthesia/sleep, dismissing established research, and introducing confusion into the field.

3. Misalignment between modeled and actual flow.

Advection effects are quantified by fitting Eq. 6 to measurements from experiments akin to the one outlined in Extended Data Figure 5 in Miao et al⁷. However, Eq. 6 presumes a flow emanating radially from the origin, which is physically impossible because it fails to conserve fluid mass. Radial flow also contradicts the actual flow direction (Fig. 1), entering from the top and exiting at the bottom. Since the modeled flow is impossible and differs significantly from the experimental one, fitting Eq. 6 will not quantify transport parameters.

Fig. 1. Discrepancy between flow velocity model and actual velocity.

Equation 6 models velocity as purely radial, emanating from the origin, whose location is not stated but presumably is the centre of the observation region (right). However, the experiment being modeled involved no such outward flow; rather, the velocity pointed downward (left). Depending on the location, downward flow is either aligned with, orthogonal to, or in direct opposition to outward flow. Additionally, the modeled flow fails to conserve fluid mass, making it impossible. Equation 6 cannot be used to model the experiment shown in Extended Data Figure 5 of Miao et al.⁷ or any others.

4. Misrepresented brain states.

The brain states presented in this study are incorrectly described, and figure design and labels are misleading. Figures 2d-f in Miao et al.⁷ indicate cohorts under different anesthesia conditions for 12 hours, however, the mice were anesthetized only once with a single bolus injection at the start of the experiment, which is not mentioned in the legend, or the result and methods sections. PENTO induces an anesthetic plane for 10–300 minutes, KET-XYL for 30–45 minutes, and DEX (with KET) for 20–30 minutes, meaning that these cohorts were in a state of post-anesthesia recovery and NOT anesthetizied for 92% (11 of 12 hours) of the experiments¹⁰. This is important, because our published data restricted the studies of glymphatic influx and clearance to deeply anesthetized mice and can therefore not be compared to the data of Miao et al.⁷. The data in Fig. 2 must be represented as post-anesthesia rather than anesthesia to avoid confusion in the field. Additionally, the recordings in Figure 2g of Miao et al.⁷ were done during sleep deprivation and rebound sleep. Post-anesthesia arousal is associated with hyperactivity in mice¹¹, and rebound sleep differs from normal sleep and requires several days for complete recovery¹².

Fig. 2: Data visualization, unequal sampling, and improper statistical analysis may cause interpretation errors in diurnal variation of diffusion.

(A) The original manuscript Figure 1e (green shaded boxes, data ranges). (B) Data re-graphed to show data variability and four-hour intervals (red dots, individual animals; error bars mean ± SEM; gray box, dark phase). (C) Tukey boxplot showing statistical outliers in the ZT0–4 and ZT20–24 bins (red dots). (D) Data re-plotted without outliers in C, with the significant ZT8–12 and ZT2–24 comparison (inset) (red dots, individual animals; error bars mean ± SEM; gray box, dark phase; *p = 0.0413).

5. No confirmation of consistent injection volumes.

Real-time 3D MRI and SPECT analyses from our group reveal significantly lower tracer injection into the brains of awake mice compared to sleeping or anesthetized mice¹³. Miao et al.⁷ fails to validate the injected tracer volumes, thus does not capture this difference. In Figure 2 of Miao et al.⁷, a fiberoptic detects tracer at a distant pinhole, while Figure 3 measures tracer content in fixated brains at 3- or 5-hours post-injection only, missing earlier timepoints. This leads to the wrong conclusion of greater clearance when, in reality, less tracer was injected in the awake group. The authors also replace the infusion cannula with a dummy post-injection, which relieves pressure and likely allows tracer to escape. Furthermore, the injection site’s peak fluorescence along the AP axis varies between mice (Figure 3b in Miao et al.⁷). Without consistent injectate volume and position, comparisons of brain clearance between groups cannot be trusted.

6. Invalid tracer compromises diffusion measurements.

Unconjugated Alexa Fluor crosses the blood–brain barrier (BBB), invalidating its use for studying parenchymal diffusion¹⁴. The authors ignore efflux to the vascular compartment. The signal that Miao et al.⁷ detect with an optical fiber in prefrontal cortex is a mixture of Alexa Fluor in both the vascular compartment and the tissue. As such, it cannot be used to calculate dispersion in brain, which is the fundamental basis for all the equations utilized.

7. Injected volume far exceeds physiologically available space.

Volumetric analysis shows that caudate–putamen averages 20.5 ± 0.5μL in mice¹⁵, with the extracellular space composing 3.1 μL. Miao et al.⁷ inject 10 μL in Figure 1 (FRAP experiments), nearly half the regional volume and 3-fold larger than the extracellular space. Such extreme fluid volumes will impose mechanical and osmotic stress, which will have brain-wide impacts and alter neural activity. The injury induced by large volume injection in Figure 1 in Miao et al.⁷ likely varies between awake and anesthetized mice, as anesthetics are potent neuroprotective agents.

8. No assessment of brain damage and inflammation.

Miao et al.⁷ do not include histological analysis of injury and reactive gliosis from multiple cannulations. Miao et al.⁷ also fails to show the cannula placement, a standard practice in the field. This omission is critical, as their design involves repeated large-volume tracer injections and 30 seconds of UV exposure to the cortex, likely inducing immune responses and artifacts. The authors need to document these effects to clarify their potential impact on the findings.

9. Unclear data visualization and improper analysis.

The over-scaled y-axis in Figure 1de downplays data variability. (Fig. 2AB), and interquartile range analysis reveals statistical outliers (Fig. 2C). Reanalysis shows a non-significant one-way ANOVA (F (5, 62) = 1.361, p = 0.2513). However, the statistics in Miao et al.⁷ are inappropriate for testing diurnal variations in brain clearance: For time series analysis of individual mice the appropriate parametric test is a Cosinor Analysis (a modified general linear model). This requires exact ZT and equal sample sizes. Miao et al.⁷ does not report exact ZT and has significantly less datapoints at night. Appropriate non-parametric tests would also need ZT reported. To utilize the current dataset, we ran two-sided t-tests on diffusion coefficients at timepoints 12 hours apart. Comparing early day and night or midday and midnight was not significant (ZT0–4 vs. ZT12–16: t(24) = 0.2382, p = 0.8137; ZT4–8 vs. ZT16–20: t(20) = 1.281, p= 0.2148). There are significant differences in diffusion coefficients between late day and night (ZT8–12 vs. ZT20–24: t(18) = 2.197, p = 0.0413; Fig 2D) indicating an underlying diurnal variation.

In sum, several significant flaws have been identified in Miao et al.⁷. Its design is fundamentally uninformative, and the results are confounded by experimental artifacts. Consequently, the data fails to support meaningful conclusions, whether regarding glymphatic clearance or other aspects of brain fluid transport.