The impaired clearance of CSF in AD
The impaired clearance of the CSF and interstitial fluid (ISF) are thought to play a role in the deposition of Aβ in the AD brain [
1,
30‐
32]. Abnormalities in several physiological clearance mechanisms that potentially underlie Aβ removal from the brain have been shown in animal models. Recent transgenic AD mouse studies have demonstrated age-related reductions in ISF and CSF clearance as well as CSF clearance deficits prior to Aβ accumulation [
7,
33]. As reviewed by Tarashoff et al. [
34], Aβ levels are affected by the deficits in endothelium function [
34], enzymatic degradation [
35‐
37], intramural periarterial drainage of Aβ [
38], glymphatic paravascular clearance [
7], lymphatic and immune clearance [
10], and CSF absorption [
39].
PET studies show that small molecular weight PET tau and amyloid tracers demonstrate rapid brain penetrance and clearance, with over 70% of the injected dose cleared from the brain by the study end [
21,
40,
41]. We previously exploited this feature in our studies as a potential CSF clearance biomarker [
16]. We reported in a small sample using THK5117 that ventricular CSF clearance is reduced in AD and inversely associated with brain Aβ levels. Now, we replicate our findings using another PET radiotracer, THK5351, and with a larger sample. Moreover, we observed for the first time correlations between CSF clearance and brain atrophy and cognitive functioning as related to AD.
MRI phase contrast studies have shown reduced CSF flow at the acqueduct in AD as compared to mild cognitive impairment (MCI) [
42] and associated with cognitive deficits [
43]. However, the phase contrast method measures pulsatile velocity rather than CSF flow and the results have been inconsistent [
44,
45]. Intrathecal MR contrast studies directly show CSF flow and the glymphatic transport of Gd-DTPA through the brain ISF [
46,
47]. While intrathecal contrast injection combined with dynamic contrast MRI appears to optimze the measurement of the CSF clearance, the intrathecal administration of contrast is invasive and it has limited application in clinical practice. Compared to these MRI measures, our intravenous dynamic PET measure is minimally invasive. Overall, the ventricle supplies the bulk of CSF to the brain [
48] and therefore the rate of tracer removal from the ventricle is an attractive biomarker of the global CSF clearance.
With similar reasoning, Silverberg et al. used an invasive method with a ventricular catheter to test the hypothesis that impaired CSF dynamics were associated with AD [
32]. They estimated the CSF production rate using intrathecal pressure changes before and after a volume of CSF was removed. However, this method is indirect and invasive, and a method that does not perturb the very system that is being measured would be preferable. More recently, in human, using a
13C
6-leucine labelled Aβ and continuous lumbar spine CSF sampling, Bateman et al. observed that Aβ clearance was reduced 33% in AD but the Aβ production rate was unaffected [
3,
6]. Consistent with Bateman et al., we observed a 18% reduction relative to NL group when CSF clearance was measured with
18F-THK5351 and 27% when measured with
11C-PiB PET.
Impaired CSF clearance and brain amyloid
Our dynamic PET data suggest ventricular CSF clearance could be a useful biomarker to monitor CSF flow dysfunctions. Overall, our results are consistent with prior evidence showing that the increased residence time of Aβ contributes to its aggregation and fibrillization in the extracellular space [
49]. We find reduced CSF clearance in AD for both THK5351 and PiB PET radiotracers, moderately strong associations with the extent of brain Aβ.
We observed that the two PET clearance measures were significantly correlated (r = 0.66, n = 24, p < 0.01). It is important to observe high correlation across tracers, even though the magnitude of brain binding is four-fold greater for PiB than for THK5351. This observation suggests that the VCSF-slope measure is independent of the global (specific and non-specific) tracer binding. Further highlighting the value of our observation, the clearance correlation with the amyloid burden was seen both in the total group and separately in the AD group.
However, the vCSF-slope measure does not inform us on specific pathways for clearing metabolic waste products from the brain parenchyma. Animal studies suggest that such pathways may involve: (a) perivascular CSF routes (for larger molecules) [
47]; (b) intramural periarterial drainage [
50]; (c) periaxonal/perineural routes [
51] and (d) the vascular pathway via the BBB (for small size molecules) [
1,
52,
53]. Aβ from the CSF is eliminated along perivascular pathways and enters the parenchyma along the pial-glial-basement membranes and this process is driven by vascular pulsations, which decrease in AD, potentially explaining the reduced clearance of the THK5351 and PiB PET radiotracers in the present study [
50,
54,
55].
The 11C-PiB tracer, which also demonstrated utility as a CSF clearance agent, appears to be partially confounded by disease-related binding detected in the time frame used to estimate CSF clearance. We believe this is reflected in the greater estimated PiB clearance 27% vs 18% for THK5351, since some of the PiB tracer enters Aβ plaques in the 10–30 min time window. Overall, as compared with PiB, THK5351 has an advantage as a clearance agent. This supports the validity of the method and point towards a preference to the THK5351 for clearance estimations.
Impaired CSF clearance is associated with decreased cognitive function
The CSF clearance measure and the brain amyloid binding were both associated with cognitive function. Intriguingly, in the subgroup analysis, the association of CSF clearance and cognitive function was significant in the NL group (r = − 0.83, n = 9, p < 0.01), while the correlation between brain amyloid binding and cognitive function was significant in the AD group (r = 0.58, n = 15, p < 0.05). Previous studies [
56,
57] show that as many as 50% of ADAS-Cog component subscales demonstrate undesirable ceiling effects in subjects with mild or moderate AD. This may explain why the vCSF-slope fails to show a significant correlation with ADAS-Cog within our AD group. In spite of small sample size (n = 9) the correlation between vCSF and ADAS-cog was very strong in the control group. This finding is consistent with the observation that CSF clearance deficits may occur in presymptomatic stages of disease progression. Moreover, reduced clearance appear to be a risk factor for abnormal brain amyloid deposits. Here our small normal sample studied cross-sectionally, was at a disadvantage, as our healthy elderly volunteers did not demonstrate sufficient variation in amyloid binding for the correlations to reach significance. Most of the normal participants had very low binding levels. Nevertheless, these data suggest the hypothesis that CSF clearance measures have potential as a predictive biomarker at the pre Aβ lesion disease stage. However, in the absence of longitudinal data, this remains speculative. Of interest, a previous animal study showed clearance deficits prior to Aβ lesions [
7].
Confounds and study limitations
Ventricular CSF clearance could be confounded by both specific and non-specific binding of the tracer in the brain. However, our results suggest that over the time interval studied, clearance rates were independent of binding effects for THK5351. This is supported by the observation that
18F-THK5351, unlike
11C-PiB, did not show a global binding effect at the 10–30 min time interval. Additional evidence justifying that tracer brain binding has a small effect on CSF clearance rate is based on the high within-subject correlations for
18F-THK5351 and PiB (r = 0.66, p < 0.01), even though the tracers have different binding distribution volumes [
58,
59]. Precise quantification would require using an 'inert' and permeable radiotracer that has no brain binding but undergoes rapid transit across the blood, CSF and the blood–brain barriers [
60]. Future highly permeable radiotracers that do not exhibit brain binding may advance this work. Overall, the results suggest that brain tracer binding had minimal effect on vCSF-SLOPE for
18F-THK5351 and
11C-PiB.
Conventional compartmental PET models are designed for tracer binding and typically don’t include a CSF compartment or estimate its rate of clearance. More advanced compartmental models have been proposed [
18], but have not been validated. New models for CSF dynamics that include tracer absorption recycling from blood and the exchange between the CSF and parenchymal interstitial space compartments are needed.
We evaluated several other possible confounds, including choroid plexus binding and partial-volume errors. Neither tracer showed choroid plexus binding that could potentially bias the ventricular clearance estimates. The ventricular partial volume error, due to contamination by proximity to brain, was minimized by individually sampling the ventricle 4 mm from the brain and with subsequent partial volume corrections [
61]. Partial volume correction did not change the results. Another possible confound, the enlarged ventricular volume in AD may cause tracer dilution, thereby altering the clearance function. This was also tested and found not to affect the findings. Because of the MAO-B binding contamination, we determined the THK5351 tracer to be poorly suited for determining the cerebral tau burden. As a result, we don’t have reliable tau positivity information on study participants.
Overall, our cross-sectional findings are consistent with the hypothesis that CSF clearance is reduced in AD. Moreover, these data support a mechanism whereby the abnormal deposition of Aβ in the brain may be due to the failure of CSF clearance [
3,
32]. However, a longitudinal sample is needed for estimating the directionality of the relationship between impaired clearance, brain Aβ deposits and the neurodegeneration we observed.