Dynamic ¹¹C-PiB PET Shows Cerebrospinal Fluid Flow Alterations in Alzheimer Disease and Multiple Sclerosis

Abstract

Cerebrospinal fluid (CSF) plays an important role in solute clearance and maintenance of brain homeostasis. ¹¹C-Pittsburgh compound B (PiB) PET was recently proposed as a tool for detection of CSF clearance alterations in Alzheimer disease. The current study investigates the magnitude of ¹¹C-PiB PET signal in the lateral ventricles of an independent group of Alzheimer and mild cognitive impairment subjects. We have also evaluated multiple sclerosis as a model of disease with CSF clearance alterations without amyloid-β tissue accumulation. Methods: A set of 11 Alzheimer and 12 mild cognitive impairment subjects and a set of 20 multiple sclerosis subjects with matched controls underwent MRI and dynamic ¹¹C-PiB PET. Lateral ventricle regions of interest were generated manually from MRI data. PET data were analyzed using cerebellum or a supervised reference region for the Alzheimer and multiple sclerosis data sets, respectively. The magnitude of ¹¹C-PiB signal in the lateral ventricles was calculated as area under the curve from 35 to 80 min and SUV ratio (SUVR) from 50 to 70 min. Compartmental modeling analysis was performed on a separate data set containing 11 Alzheimer and matched control subjects; this analysis included an arterial input function, to further understand the kinetics of the lateral ventricular ¹¹C-PiB signal. Results: ANOVA revealed significant group differences in lateral ventricular SUVR across the Alzheimer, mild cognitive impairment, and healthy control groups (P = 0.004). Pairwise comparisons revealed significantly lower lateral ventricular SUVR in Alzheimer subjects than in healthy controls (P < 0.001) or mild cognitive impairment subjects (P = 0.029). Lateral ventricular SUVR was significantly lower in multiple sclerosis subjects than in healthy controls (P = 0.008). Compartmental modeling analysis revealed significantly lower uptake rates of ¹¹C-PiB signal from blood (P = 0.005) and brain tissue (P = 0.004) to the lateral ventricles and significantly lower ¹¹C-PiB signal clearance out of the lateral ventricles (P = 0.002) in Alzheimer subjects than in healthy controls. Conclusion: These results indicate that dynamic ¹¹C-PiB PET can be used to observe pathologic changes in CSF dynamics. We have replicated previous work demonstrating CSF clearance deficits in Alzheimer disease associated with amyloid-β deposits and have extended the observations to include ventricular CSF clearance deficits in mild cognitive impairment and multiple sclerosis.

There has been great interest surrounding cerebrospinal fluid (CSF) dynamics since the existence of the glymphatic system was proposed in 2012 (1–3). The glymphatic system has been suggested as being largely responsible for the clearance of waste from the brain (1,4). The original description of the glymphatic system proposed that CSF penetrates the brain via paraarterial spaces; enters the brain parenchyma, where it combines with interstitial fluid and collects waste and other solutes; and returns to the subarachnoid space or clears through the vascular and lymphatic systems via paravenous spaces (Fig. 1A). This system is thought to be analogous to the body’s peripheral lymphatic system with the additional involvement of glial cells (1). Although there is still debate over the mechanism underlying the glymphatic system (2,3,5), there is an overall agreement that a para- or perivascular clearance system of the brain exists and that it is closely linked to the production and flow of CSF.

FIGURE 1.

Interrelationship between clearance and fluid systems of brain. (A) Glymphatic system. (B) CSF system. (Illustrations by Julia J. Schubert.)

CSF is produced predominantly by the choroid plexuses of the ventricular system. Interstitial fluid of surrounding brain tissues, excluding those of the circumventricular organs, exchanges quite freely with ventricular CSF because of the presence of gap junctions in the ependymal cell lining of the adult ventricular system that allow free diffusion of small molecules (6). CSF generally has a net positive flow through the ventricular system and out to the subarachnoid space that surrounds the brain and spinal cord (Fig. 1B). CSF is eventually cleared to the venous system through arachnoid villi, to the cervical lymphatics through the cribriform plate (7,8), or to the recently discovered meningeal lymphatics that line the dural sinuses (9). The transastrocytic water movements associated with CSF flow into and CSF/interstitial fluid flow out of the brain parenchyma are facilitated by aquaporin-4 water channels, which are localized along perivascular astrocytic end-feet that contribute to the blood–brain barrier (1).

The glymphatic system has been shown to contribute to clearance of amyloid-β (Aβ) (1,10,11), the main component of the amyloid plaques found in Alzheimer disease brain. Significant accumulation of Aβ has been observed after inhibition of glymphatic transport (10), and deletion of the aquaporin-4 gene suppresses clearance of soluble Aβ (1). Phase-contrast MRI has also shown significantly decreased CSF flow in multiple sclerosis patients compared with healthy controls (12,13), and associations have been observed between decreased CSF flow and conversion rate from clinically isolated syndrome to clinically defined multiple sclerosis and relapse rate in relapsing-remitting multiple sclerosis (12). Further, loss of perivascular aquaporin-4 localization has been observed in an experimental autoimmune encephalomyelitis mouse model of multiple sclerosis (14). These findings support a link between CSF flow alterations and neurologic diseases such as Alzheimer disease and multiple sclerosis.

Despite the current interest in CSF dynamics, noninvasive in vivo methods for measuring glymphatic function are limited, and most existing data have been collected using animal models (4,10,14). ¹¹C-Pittsburgh compound B (PiB) is a neutral lipophilic benzothiazole PET tracer and is used to image Aβ plaques in Alzheimer disease (15) and, more recently, to quantify in vivo myelin loss and regeneration in multiple sclerosis (16). ¹¹C-PiB PET has also recently been used to quantify CSF clearance in humans (8). Small-molecule radiotracers enter the CSF of the lateral ventricles directly from the blood through a thick epithelial cell layer of the choroid plexuses and from the brain parenchyma via diffusion through the ependymal cells that line the adult ventricular system (6). We aim to replicate previous results that showed decreased lateral ventricular ¹¹C-PiB signal magnitude in Alzheimer disease compared with controls (8) and to extend the method to mild cognitive impairment and multiple sclerosis patient groups. From previous research, each of these patient groups is expected to have altered CSF dynamics. However, because of differences in the pathogenesis of Alzheimer disease and multiple sclerosis, Aβ deposition is expected in only the Alzheimer disease and mild cognitive impairment groups. Inclusion of the multiple sclerosis group allows for further testing of our dynamic PET method for measuring CSF dynamics without the confounding factor of Aβ accumulation in brain tissue. We also aim to further understand the kinetics of the ¹¹C-PiB PET signal in the lateral ventricles by performing compartmental modeling analysis. Combined, these results will allow for appropriate interpretations about the lateral ventricular ¹¹C-PiB signal magnitude and how these measurements may relate to CSF-mediated clearance in healthy and diseased populations.

MATERIALS AND METHODS

Participants

To investigate differences in CSF clearance in Alzheimer disease and multiple sclerosis, analysis was performed on 2 data sets. One data set included 11 Alzheimer disease patients (6 women and 5 men; mean age, 66.5 ± 4.1 y), 12 mild cognitive impairment patients (4 women and 8 men; mean age, 68.6 ± 7.9 y), and 12 age- and sex-matched healthy controls (5 women and 7 men; mean age, 64.4 ± 6.6 y) who were recruited from Hammersmith Hospital NHS Trust, the National Hospital for Neurology and Neurosurgery, St. Margaret’s Hospital, and Victoria Hospital. Recruitment of the healthy controls was aided by enrolment of spouses of the Alzheimer disease subjects. The Alzheimer disease and matched healthy control data have previously been reported alone (17) and with comparison to the mild cognitive impairment data (18). Clinically probable diagnoses of Alzheimer disease were assigned on the basis of the criteria of the National Institute of Neurologic and Communicative Diseases and Stroke/Alzheimer’s Disease and Related Disorders Association and Diagnostic and Statistical Manual of Mental Disorders, fourth edition. All Alzheimer subjects also met the National Institute on Aging/Alzheimer’s Association criteria. Alzheimer subjects aged 55 to 79 with a clinical diagnosis of Alzheimer disease before enrolment were included in the study. Patients and controls with a history of mental health issues, significant white matter microvascular disease on MRI, a contraindicative MRI result, a history of drug or alcohol abuse, or any other neurologic causes were excluded. Full details on inclusion and exclusion criteria for the Alzheimer subjects were reported in the original publication on this data set (17). Mild cognitive impairment subjects were chosen on the basis of exclusion criteria similar to those for the Alzheimer subjects and fulfilled Petersen’s criteria for amnestic mild cognitive impairment (19). Six of the 12 mild cognitive impairment subjects were previously classified as amyloid-positive.

A second data set included 20 relapsing-remitting multiple sclerosis patients (13 women and 7 men; mean age, 32.3 ± 5.6 y) and 8 age- and sex- matched healthy controls (5 women and 3 men; mean age 31.6 ± 6.4 y). All patients were diagnosed with multiple sclerosis according to the revised McDonald criteria (20) and had at least one gadolinium-enhancing lesion when they entered the study. Additional information on inclusion and exclusion criteria can be found in the original report on this data set (16).

These studies were approved by the local ethics committees, and participants gave written informed consent before data collection.

Clinical Assessments

Alzheimer Disease/Mild Cognitive Impairment (AD/MCI) Data Set

Detailed neurologic assessments were performed for 9 Alzheimer subjects. These assessments included taking a patient history from a close relative, routine blood analysis, and electroencephalogram. The following neuropsychometric assessments were also performed: Mini-Mental State Examination (21), Warrington short recognition memory tests for words and faces, Alzheimer Disease Assessment Scale Word List Learning test and 30-min delayed recall (22), immediate and delayed recall of modified complex figure (23), Digit Span forwards (24), Trail Making Part A (25), clock drawing (26), copy of modified complex figure (23), 30-item Boston Naming Test (27), letter fluency (F-A-S test) (28), and category fluency (animals, birds, and dogs). All mild cognitive impairment subjects also underwent a comprehensive assessment that included neurologic examination, neuropsychologic testing, and MRI.

Multiple Sclerosis Data Set

The Expanded Disability Status Scale (29) and Multiple Sclerosis Severity Scale (30) were used to rate 19 multiple sclerosis subjects at the time of enrolment.

Demographic data from the Alzheimer, mild cognitive impairment, multiple sclerosis, and matched control groups are included in Supplemental Table 1 (supplemental materials are available at http://jnm.snmjournals.org).

PET and MRI Data Acquisition

AD/MCI Data Set

All subjects had 90-min ¹¹C-PiB PET on a Siemens ECAT EXACT HR+ scanner with 3-dimensional acquisition and an axial field of view of 15.5 cm. All subjects were given an intravenous bolus injection of ¹¹C-PiB (mean, 365 ± 24 MBq) at the start of each scan. Image reconstruction and data processing, including scatter correction, were performed using standard Siemens software. All subjects also had MRI, which was performed with a 1.5-T GE Healthcare scanner. The T1-weighted structural MRI data were used for region-of-interest (ROI) segmentation. Additional information on the PET and MRI protocols can be found in the original reports on these data (17,18).

Multiple Sclerosis Data Set

A High-Resolution Research Tomograph (CPS Innovations) was used to perform a 90-min ¹¹C-PiB PET scan on all subjects. All subjects were given an intravenous bolus injection of ¹¹C-PiB (mean, 358 ± 34 MBq) at the start of each scan. An intraslice spatial resolution of about 2.5 mm full width at half maximum was achieved, with 25 cm axial and 31.2 cm transaxial fields of view. A Poisson ordered-subset expectation maximization algorithm with 10 iterations was used for image reconstruction. To assist in reducing the effects of partial volume in the multiple sclerosis data set (31), the reconstructed images were smoothed with a filter implementing point-spread function. All subjects also underwent MR scanning on a 3-T Siemens Trio 32-channel Total Imaging Matrix system. T1-weighted structural MRI data collected before gadolinium injection were used for ROI segmentation in this work. The PET image acquisition, reconstruction, and quantification techniques have been previously described (32). Additional information on the PET and MRI protocols can be found in the original report on this data set (16).

Data Analysis

AD/MCI Data Set

Preprocessing of the PET and MRI data from the AD/MCI data set was performed and time–activity curves were generated using MIAKAT software (version 4.2.6) (33). MIAKAT is implemented in MATLAB (version R2015b; The MathWorks, Inc.), and the preprocessing pipeline uses tools from SPM12 and FSL (version 5.0.9) analysis toolboxes (34) to perform brain extraction, tissue segmentation, rigid and nonlinear registration to the Montreal Neurological Institute template (35), ROI definition, and motion correction. An additional gray matter ROI was defined that excluded the cerebellum reference region. The data sets of one Alzheimer and one mild cognitive impairment subject did not pass quality control because of misregistration artifacts and were excluded from the AD/MCI analysis.

Multiple Sclerosis Data Set

Motion correction of the multiple sclerosis data set was performed by realigning each PET frame to a common reference space, as has been previously described (36). T1-weighted MR images were registered to the ¹¹C-PiB PET images. A priori designated ROIs were used for a supervised reference region and a gray matter ROI used in the multiple sclerosis analysis. The gray matter ROI excluded reference and cerebellar gray matter for comparison with the Alzheimer data set. Time–activity curve extraction for the multiple sclerosis data set was performed using in-house MATLAB (version R2017a) scripts.

For both data sets, manual lateral ventricle ROIs were generated for all subjects using the subject T1-weighted structural MRI data and the ITK-SNAP (www.itksnap.org) snake tool (37), following previously described guidelines for lateral ventricle extraction (38). The lateral ventricle ROIs were then eroded by 2 voxels (5.2 mm) using the erode function given by the FSL “fslmaths” utility package (version 5.0.9) (34) to reduce partial-volume effects of the surrounding tissues. An example lateral ventricle ROI is shown in Supplemental Figure 1.

SUV ratios (SUVRs) from 50 to 70 min were calculated for the gray matter and lateral ventricle ROIs for each subject from both data sets using cerebellum or a supervised reference region for the AD/MCI and multiple sclerosis data sets, respectively. For consistency with a previous study (8), the area under the curve (AUC) from 35 to 80 min (AUC_35–80) was also calculated for the lateral ventricle ROI for a subset of the AD/MCI data set. AUC_35–80 was also calculated for the lateral ventricle ROI for all subjects in the multiple sclerosis data set.

Compartmental Modeling Analysis

Participants

A data set that included 11 Alzheimer disease subjects and 11 age- and sex-matched healthy controls was included for sole use in compartmental modeling analysis. These data have been previously reported (39). All Alzheimer subjects in this data set met the National Institute of Neurologic and Communicative Diseases and Stroke/Alzheimer’s Disease and Related Disorders Association criteria for probable Alzheimer disease and had Mini-Mental State Examination scores greater than 16 at study inclusion. Healthy controls were recruited by advertisement or during participation in a separate long-term follow-up study in the clinic. This study was approved by the local ethics committee, and all participants gave written informed consent before data collection.

PET Data Acquisition

All subjects underwent 90-min ¹¹C-PiB PET with online arterial sampling. PET data were collected on an ECAT EXACT H+ scanner (Siemens/CTI). Metabolite-corrected arterial input functions were determined using the measured percentage of radioactive parent compound in plasma. Further details about the PET protocol can be found in the original report on this data set (39).

Compartmental Model

The final model used to describe ¹¹C-PiB PET kinetics in the lateral ventricles is shown in Figure 2. This model includes 2 compartments to account for the signal from both bound and unbound ventricle pools and 2 input functions. One input function corresponds to the whole-brain gray matter, which describes the tracer transport from the brain tissue into the lateral ventricles, and one input function corresponds to the parent plasma arterial input function, which describes the tracer transport from the blood into the lateral ventricles. This model was determined on the basis of the known biologic restraints of the CSF system (6).

FIGURE 2.

Kinetic model used in compartmental modeling analysis. K_clearance includes total clearance from lateral ventricles to blood, tissue, and rest of ventricular system.

Data Analysis

Compartmental modeling analysis was performed with SAAM II software (40). The lateral ventricle ROI used for time–activity curve extraction was defined using a registered Montreal Neurological Institute template and the ITK-SNAP snake tool (37) with additional manual drawing when necessary. This data set was used because it included an arterial input function necessary for accurate signal measurements from the blood, which is required for compartmental modeling analysis. Visual inspection of the Montreal Neurological Institute template–generated lateral ventricle ROIs overlaid on individual subjects’ PET images showed good overlap with the areas known to be lateral ventricle CSF in all cases. The rate constants K_clearance, K_1tissue, K_1blood, k₃, and k₄ represent the rate of ¹¹C-PiB signal between tissues in the system. K_clearance takes into account the clearance of signal from the lateral ventricles to blood, surrounding tissues, and the rest of the ventricular system because it is not possible to differentiate between these clearance pathways. The SAAM II software was used to iteratively fit the time–activity curve data to the model, and final values of the rate constants were recorded for each subject.

Statistical Analyses

All statistical analyses were performed in SPSS (version 24.0). The Shapiro–Wilk W test was used to test for normality of the data. ANOVA was used to investigate group differences in lateral ventricle and gray matter ¹¹C-PiB signal in the AD/MCI data set. Paired differences in lateral ventricle ¹¹C-PiB SUVRs between groups in the AD/MCI and multiple sclerosis data sets were investigated using independent-samples t tests. Paired differences in lateral ventricle ¹¹C-PiB AUC_35–80 between groups were investigated using independent-samples t tests for the AD/MCI data set and the Mann–Whitney U test for the multiple sclerosis data set. Paired differences in gray matter ¹¹C-PiB SUVRs between groups were investigated using the Mann–Whitney U test for the AD/MCI data set and an independent-samples t test for the multiple sclerosis data set. Spearman correlation was used to investigate the relationship between gray matter and lateral ventricular ¹¹C-PiB signal as well as between clinical scores and lateral ventricular ¹¹C-PiB signal in Alzheimer disease and multiple sclerosis patient groups. Group differences in K_clearance, k₃, and k₄ were investigated using independent-samples t tests. Group differences in K_1tissue and K_1blood were investigated using the Mann–Whitney U test. Brain size and ventricle size were tested as possible covariates.

RESULTS

Lateral Ventricular ¹¹C-PiB Signal Magnitude in Relation to Diagnosis

ANOVA revealed significant group differences in lateral ventricle SUVRs across the Alzheimer disease, mild cognitive impairment, and healthy control groups (F_2,30 = 6.86, P = 0.004), which remained significant when corrected for ventricle size (F_2,29 = 3.34, P = 0.050). Brain size was not found to be a significant covariate in the AD/MCI data set. Additional pairwise comparisons revealed a significantly lower magnitude of lateral ventricular ¹¹C-PiB, as measured by SUVR, in Alzheimer disease subjects (mean, 0.37; SD, 0.09) than in healthy controls (mean, 0.57; SD, 0.15) (t₁₈ = 4.08, P < 0.001) or mild cognitive impairment subjects (mean, 0.49, SD, 0.14) (t₁₉ = 2.36, P = 0.029). No significant difference in lateral ventricle SUVR was observed between mild cognitive impairment subjects and healthy controls (t₂₁ = 1.37, P = 0.185) (Fig. 3A). Lateral ventricle AUC_35–80 was significantly lower in Alzheimer disease subjects (mean, 0.47; SD, 0.13) than in healthy controls (mean, 0.72; SD, 0.20) (t₁₈ = 3.32, P = 0.004) or mild cognitive impairment subjects (mean, 0.65; SD, 0.18) (t₁₇ = 2.47, P = 0.024). ANOVA also revealed significant group differences in lateral ventricle AUC_35–80 across the Alzheimer disease, mild cognitive impairment, and healthy control groups (F_2,26 = 5.53, P = 0.010) (Fig. 3B).

FIGURE 3.

Group differences in lateral ventricle ¹¹C-PiB measurements. (A) Lateral ventricle ¹¹C-PiB SUVR in healthy controls (HC), mild cognitive impairment (MCI), and Alzheimer disease (AD). (B) Lateral ventricle ¹¹C-PiB AUC_35–80 in HC, MCI, and AD. (C) Lateral ventricle ¹¹C-PiB SUVR in HC and multiple sclerosis (MS). (D) Lateral ventricle ¹¹C-PiB AUC_35–80 in HC and MS. Significant differences (P < 0.05) are indicated by an asterisk.

Lateral ventricle ¹¹C-PiB SUVRs were significantly lower in multiple sclerosis (mean, 0.77; SD, 0.19) than in healthy controls (mean, 1.01; SD, 0.21) (t₂₆ = 2.87, P = 0.008), which remained significant when corrected for ventricle size (F_1,25 = 4.35, P = 0.047) (Fig. 3C). Brain size was not found to be a significant covariate in the multiple sclerosis data set. There were no other significant differences in AUC_35–80 measurements between groups.

Correlations with Tissue ¹¹C-PiB

We observed significant negative correlations in ¹¹C-PiB SUVRs between gray matter and lateral ventricle ROIs in mild cognitive impairment subjects (r = −0.664, P = 0.026) and AD/MCI-matched control subjects (r = −0.909, P < 0.001), which was not observed in Alzheimer disease subjects (r = −0.479, P = 0.162) (Fig. 4A). No significant correlations were observed between gray matter ¹¹C-PiB SUVRs and lateral ventricle AUC_35–80 in Alzheimer disease subjects (r = −0.182, P = 0.614), mild cognitive impairment subjects (r = −0.150, P = 0.700), or AD/MCI-matched control subjects (r = −0.115, P = 0.751) (Fig. 4B).

FIGURE 4.

Correlations between gray matter ¹¹C-PiB SUVR and lateral ventricle ¹¹C-PiB SUVR (A) and AUC_35–80 (B) in Alzheimer disease (AD), mild cognitive impairment (MCI), and healthy controls (HC). In A, r = −0.479 and P = 0.162 for AD, r = −0.664 and P = 0.026 for MCI, and r = −0.909 and P < 0.001 for HC. In B, r = −0.182 and P = 0.614 for AD, r = −0.150 and P = 0.700 for MCI, and r = −0.115 and P = 0.751 for HC.

We did not observe any significant ¹¹C-PiB signal correlations between gray matter and lateral ventricle regions in the multiple sclerosis subjects (SUVR: r = 0.045, P = 0.850; AUC_35–80: r = −0.170, P = 0.945) or multiple sclerosis–matched control subjects (SUVR: r = 0.333, P = 0.420; AUC_35–80: r = −0.238, P = 0.570).

Gray Matter ¹¹C-PiB Signal Magnitude in Relation to Diagnosis

ANOVA revealed significant group differences in gray matter SUVRs across Alzheimer disease, mild cognitive impairment, and healthy control groups (F_2,30 = 14.53, P < 0.001). Gray matter ¹¹C-PiB signal, as measured by SUVR, was significantly higher in Alzheimer disease subjects (mean, 3.68; SD, 1.14) than in healthy controls (mean, 1.53; SD, 0.28) (U = 119.0, P < 0.001) or mild cognitive impairment subjects (mean, 2.52; SD, 1.15) (U = 87.0, P = 0.024) and was significantly higher in mild cognitive impairment subjects than in healthy controls (U = 116.0, P = 0.002) (Fig. 5A). Gray matter ¹¹C-PiB SUVRs were not significantly different between multiple sclerosis subjects and healthy controls (t₂₆ = 1.32, P = 0.198) (Fig. 5B).

FIGURE 5.

Group differences in gray matter ¹¹C-PiB SUVRs. (A) Gray matter ¹¹C-PiB SUVR in healthy controls (HC), mild cognitive impairment (MCI), and Alzheimer disease (AD). (B) Gray matter ¹¹C-PiB SUVR in HC and multiple sclerosis (MS). Significant differences (P < 0.05) are indicated by an asterisk.

Correlations with Disease Severity Measures

AD/MCI Data Set

We did not observe any significant correlations between Mini-Mental State Examination and lateral ventricle SUVRs in the Alzheimer disease group (r = −0.042, P = 0.914). No other significant correlations were observed between disease severity measures and lateral ventricle ¹¹C-PiB measures in the Alzheimer disease group.

Multiple Sclerosis Data Set

We did not observe any significant correlations between Multiple Sclerosis Severity Scale (r = 0.075, P = 0.762), Expanded Disability Status Scale (r = −0.003, P = 0.991), or disease duration (r = −0.144, P = 0.555) and lateral ventricle SUVRs. We observed significant correlations between lesion load, as defined by the volume of white matter lesions, and lateral ventricle SUVRs (r = −0.493, P = 0.032) and AUC_35–80 (r = −0.595, P = 0.007) (Fig. 6). We did not observe any other statistically significant correlations between disease severity measures and lateral ventricle ¹¹C-PiB signal measures in the multiple sclerosis group.

FIGURE 6.

Lateral ventricle ¹¹C-PiB measures and lesion load correlations. (A) Correlation between lesion load and lateral ventricle SUVRs in multiple sclerosis (MS) (r = −0.493, P = 0.032). (B) Correlation between lesion load and lateral ventricle AUC_35–80 in MS (r = −0.595, P = 0.007). Lesion load is defined as volume of white matter lesions.

Compartmental Modeling Analysis

The final compartmental model used in our analysis reliably fit the lateral ventricle time–activity curve data. Example fits of the lateral ventricle time–activity curve data from one healthy control and one Alzheimer disease subject using the final compartmental model are shown in Supplemental Figure 2.

Results from compartmental modeling analysis revealed significant group differences in rates of ¹¹C-PiB signal exchange across tissues between Alzheimer disease and healthy control subjects. The rate of signal from blood to the lateral ventricles (K_1blood) (U = 13.0, P = 0.005), from tissue to the lateral ventricles (K_1tissue) (U = 12.0, P = 0.004), and from the lateral ventricles to blood, tissue, and the rest of the ventricular system (K_clearance) (t₁₈ = 3.70, P = 0.002) was significantly lower in Alzheimer disease subjects than in healthy controls (Fig. 7). We did not observe any other significant differences between groups.

FIGURE 7.

Results from compartmental modeling analysis: healthy controls and Alzheimer group differences in K_1blood, K_1tissue, and K_clearance rates. Error bars represent 95% confidence intervals. Significant differences (P < 0.05) are indicated by an asterisk.

Simplified compartmental models, including one without specific binding in the lateral ventricles and one without a brain tissue input function, were also compared with the final compartmental model used in our analysis. The final model used was superior to simplified models, as determined by comparison of Akaike information criterion scores.

DISCUSSION

Our observation of decreased ¹¹C-PiB signal in the lateral ventricle ROIs in Alzheimer disease and multiple sclerosis patient groups compared with healthy controls indicates that dynamic PET measures can be used to observe pathologic changes in CSF dynamics. We have successfully replicated a previous study that showed decreased lateral ventricle ¹¹C-PiB signal in Alzheimer disease compared with healthy controls (8). Here, we have also shown that lateral ventricle ¹¹C-PiB is significantly lower in multiple sclerosis subjects than in healthy controls and that mild cognitive impairment subjects have ¹¹C-PiB measures intermediate between those of healthy controls and Alzheimer disease subjects. We replicated previous work showing an inverse correlation between gray matter and lateral ventricle ¹¹C-PiB binding in Alzheimer disease, and we show that this correlation is present in mild cognitive impairment subjects and AD/MCI-matched healthy controls. Our compartmental modeling analysis reveals that the reduced lateral ventricle signal is likely due to decreased tracer entering the lateral ventricles through the blood and through the brain tissue in Alzheimer disease compared with healthy controls. This analysis also shows that tracer is cleared from the lateral ventricles to the surrounding tissues and the ventricular system at a lower rate in Alzheimer disease subjects than in healthy controls. Our results also indicate that the reduction of tracer entering and leaving the lateral ventricles in patient groups is independent of Aβ deposition, as indicated by the lateral ventricle PET results from the multiple sclerosis data set, which is not expected to have significant Aβ accumulation in tissue. This is further supported by our observation of no significant difference in ¹¹C-PiB signal in gray matter between multiple sclerosis patients and controls, in contrast to the higher ¹¹C-PiB signal observed in Alzheimer disease gray matter. Altogether, these results suggest that CSF-mediated tissue clearance is reduced in Alzheimer disease and multiple sclerosis subjects compared with healthy controls.

Increasing age is considered a major risk factor in the development of Alzheimer disease (41), and Aβ deposition has been associated with increasing age in cognitively normal individuals without Alzheimer disease (42). The observed inverse relationship between gray matter and lateral ventricle ¹¹C-PiB signal in mild cognitive impairment and control groups could indicate that tissue CSF clearance alterations are associated with Aβ deposition before Alzheimer disease onset. Under the pathogenic conditions of Alzheimer disease, there are likely additional factors related to the disease that contribute to further Aβ deposition (43), which weakens the linear relationship between CSF clearance and tissue Aβ deposition. This could explain why we did not observe the inverse relationship between CSF clearance and Aβ deposition measures in the Alzheimer disease group. However, because of the greater heterogeneity in tissue ¹¹C-PiB measures in the Alzheimer disease and mild cognitive impairment groups, we may be lacking the statistical power necessary to measure the correlations in these groups. Additional mechanistic analysis with a larger cohort would be helpful in drawing further conclusions about these results. We did not observe an inverse relationship between gray matter and lateral ventricle ¹¹C-PiB in multiple sclerosis or multiple sclerosis–matched controls, likely because of the younger age of the cohort or a different mechanism of the disease that does not typically involve Aβ accumulation. The positive correlation between lesion load and ¹¹C-PiB signal in the multiple sclerosis group may indicate that CSF clearance alterations play a role in ongoing disease activity in multiple sclerosis.

Previous literature has reported decreased Aβ clearance from the brain in Alzheimer disease (44), but it has been unclear whether this decrease is due to reduced membrane transport from tissue to the CSF or from reduced clearance of the CSF through the CSF system. Compartmental modeling analysis allows us to make further inferences regarding where changes in Aβ clearance are occurring and which tissue types and systems may be responsible. We see that the change in Aβ clearance in Alzheimer disease is likely due to both reduced clearance of Aβ from the tissue to the CSF and reduced clearance of CSF through the CSF system. It remains unclear whether alterations to one of these processes precede alterations to the other. Additional compartmental modeling analyses using data from other patient groups (including multiple sclerosis, clinically isolated syndrome, mild cognitive impairment, and subjects with first signs of Aβ accumulation) would be useful for drawing further conclusions about our results.

The explanation as to why CSF-mediated clearance is reduced in neurologic diseases such as Alzheimer disease and multiple sclerosis is not entirely clear. Previous research has shown that sleep is important for glymphatic system function and consequently the clearance of waste from the brain (45). Sleep disorders are common in both Alzheimer disease (46) and multiple sclerosis (47) and may play a part in the onset of disease and could contribute to the ongoing disease processes. Additionally, synaptic activity has been linked to increased levels of Aβ in interstitial fluid (48), and voluntary exercise has been shown to increase clearance of Aβ by the glymphatic system in aged mice (49). These findings suggest that physical activity also plays a role in brain clearance. Further, physical activity has been shown to be effective in preventing the onset of and improving outcomes in Alzheimer disease (50) and may also contribute to improved sleep quality in health (51) and neurologic disease (52,53). Previous work has shown reduced sleep quality before the onset of cognitive decline in Alzheimer disease (54). However, additional research is still required to investigate whether inactivity or sleep disturbances precede the onset of disease, which may help further explain the initial pathophysiology that leads to disease onset.

Additional explanations for alterations in CSF-mediated clearance in aged and Alzheimer disease brain could be attributed to changes in the barriers that exist between brain tissue, blood, and CSF. Age is associated with cellular atrophy and decreased cell height of the choroid plexuses, which is exacerbated in Alzheimer disease (55). The production of CSF and removal of solutes from the CSF by the choroid plexuses are active processes (56), and the cells become less energy-efficient with age (57). Age-related changes to the cells of the choroid plexuses contribute to decreased clearance of solutes to the blood as well as decreased CSF production, resulting in slower CSF turnover and reduced CSF-mediated brain clearance. The ependymal cells that line the ventricular system also flatten with age and exhibit greater dispersion in cilia expression (58), which normally contributes to flow of CSF and communication across the ependymal layer (59). These age-related changes to the ependymal layer likely contribute to decreased CSF-mediated brain clearance.

A previous MRI study revealed irregularities in the ependymal layer in early multiple sclerosis that have been attributed to ependymal perivenular inflammation (60). Inflammation of the choroid plexuses in multiple sclerosis has also been observed (61). Under healthy conditions, immune surveillance of the brain is partly accomplished by lymphocyte entry to the CSF through the choroid plexuses (62). Initial entry of reactive lymphocytes through the choroid plexuses was observed before lymphocyte infiltration into brain tissue across the blood–brain barrier in a mouse model of multiple sclerosis (63), which provides support for the involvement of the choroid plexuses in multiple sclerosis disease onset. The alterations to the ventricular barriers in multiple sclerosis could be both a cause and a consequence of impaired CSF-mediated clearance mechanisms, and further work is required to better understand the relationship and chronology of these changes.

Additional work will be done to investigate whether our ¹¹C-PiB PET results are specific to Alzheimer disease and multiple sclerosis or whether CSF clearance is also altered in other neurologic disorders. Further, we will look into use of other dynamic PET tracers and integration of quantitative MRI and CSF biomarker measures for assessing and understanding CSF-mediated clearance. It will also be helpful to explore how manipulation of known glymphatic system modulators affects the dynamic PET measures, which will allow for more informed interpretations of the results given from these PET measures in the future.

Our study had some limitations. Although we still observed statistically significant differences between groups when including ventricle size as a covariate in the AD/MCI and multiple sclerosis data sets, this may not be an appropriate way to correct for ventricle size. Larger ventricle size is an inherent feature of Alzheimer disease, and we observe significant group differences in ventricle size across the Alzheimer disease, mild cognitive impairment, and healthy control groups in the current analysis (P = 0.041), most notably between Alzheimer disease and healthy controls (P = 0.001). Simply adding ventricle size as a covariate may, therefore, be statistically inappropriate for use in our model because we are likely removing variance between groups that exists due to disease. Additional work should be performed to investigate more appropriate ways to accommodate for differences in ventricle size in the future.

We are restricted in our compartmental modeling analysis by the absence of structural image data in the form of MRI or CT for the selected Alzheimer disease data set. We produced all lateral ventricle ROIs for the compartmental modeling using a registered Montreal Neurological Institute template for each subject. Registered Montreal Neurological Institute templates are not as reliable for defining brain structures as MRI or CT images, and our lateral ventricle ROIs for compartmental modeling analysis are therefore prone to error. Our selection of ¹¹C-PiB PET data sets was restricted to those that included an arterial input function that is required for compartmental modeling analysis. However, the invasive nature of continuous arterial blood sampling discourages use of arterial input functions in human studies. Ideally, future PET studies will be designed to include arterial blood sampling during PET data collection for use in compartmental modeling analysis.

Our compartmental model has several limitations. The model uses time–activity curve data from all gray matter for the tissue pool that exchanges with the lateral ventricles. However, we expect direct exchange only between the deep gray matter structures and the lateral ventricular CSF because of their proximity to each other. However, these regions of direct exchange are difficult to define, and signal from whole gray matter is expected to behave similarly to the signal in the deep gray matter structures alone. Therefore, signal from whole gray matter was used as a representative measure for the gray matter that exchanges with the lateral ventricle CSF. We also tested a model using signal from cerebellar gray matter as a region that is not expected to have specific binding in Alzheimer disease (Supplemental Fig. 3) to confirm that the differences in rate constants between groups are not driven by differences in gray matter signal due to specific binding to Aβ. A model that entirely excluded input from the gray matter tissue to the lateral ventricles (Supplemental Fig. 4) resulted in poorer fits of the lateral ventricle time–activity curve data, indicating that this simplified model is inferior in its ability to represent the system. Finally, it is unclear what the specific binding in the lateral ventricles represents in our model. This bound pool may be accounted for by tracer binding to the ventricle walls or choroid plexuses or, less likely, to compounds within the CSF. We found that a model that excluded the ventricular bound pool (Supplemental Fig. 5) did not reliably fit the lateral ventricle time–activity curve data, indicating that this simplified model does not accurately represent the system.

CONCLUSION

The results from this work confirm that dynamic ¹¹C-PiB PET can be used to assess CSF dynamics in health and neurologic disease. In this work, we present a successful replication of previous results that showed CSF-mediated clearance deficits in Alzheimer disease associated with amyloid-β deposits (8). We also extend the observation to include CSF-mediated clearance deficits in mild cognitive impairment and multiple sclerosis. Our results provide further support for a promising method for assessing brain clearance in neurologic disease. We hope that this and future work in this area will improve the understanding of the pathogenesis of neurologic diseases such as Alzheimer disease and multiple sclerosis.

DISCLOSURE

The multiple sclerosis data that were analyzed in this work were originally collected in a study that received funding from the European Leukodystrophy Association (grant 2007-0481), INSERM-DHOS (grant 2008: recherche Clinique et translationnelle), Assistance Publique des Hôpitaux de Paris, and the “Investissements d’avenir” (grant ANR-10-IAIHU-06). The AD/MCI PET and MRI scans were funded by the Medical Research Council (grant WMCN_P33428), and the original studies were in part funded by Alzheimer’s Research U.K. (grant WMCN_P23750). Mattia Veronese and Federico E. Turkheimer received funding from the MRC-U.K. PET Methodology Programme (grant nG1100809/1), an ARSEP travel grant, and the National Institute for Health Research Biomedical Research Centre at South London and Maudsley National Health Service Foundation Trust and King’s College London. Benedetta Bodini received financial support from ANR MNP2008-007125, the ECTRIMS post-doctoral research fellowship, and Fondation d’Aide pour la Recherche sur la Sclerose en Plaques. The Alzheimer’s disease data used in compartmental modeling analysis were collected in a study that received funding from the National Institute of Aging (grant R01AG17761). Paul Edison has received funding from the Medical Research Council and is currently funded by the Higher Education Funding Council for England (HEFCE). He has also received grants from Alzheimer’s Research, U.K.; Alzheimer’s Drug Discovery Foundation; Alzheimer’s Society, U.K.; Novo Nordisk; GE Healthcare; Astra Zeneca; Pfizer; Eli Lilly; and Piramal Life Sciences. The amyloid tracer used for the AD/MCI data set was made available by GE Healthcare. The original study that collected the Alzheimer’s disease data used in our compartmental modeling analysis received research support from GlaxoSmithKline. No other potential conflict of interest relevant to this article was reported.

KEY POINTS

• QUESTION: Do cerebrospinal fluid flow alterations exist in multiple sclerosis and Alzheimer disease?

• PERTINENT FINDINGS: This case-control study further supports the use of dynamic ¹¹C-PiB PET to assess cerebrospinal fluid dynamics in health and disease. We observed significantly lower lateral ventricle ¹¹C-PiB signal in multiple sclerosis and Alzheimer disease patient groups than in healthy controls, indicating that alterations to cerebrospinal fluid–mediated brain clearance may be present in these diseases.

• IMPLICATIONS FOR PATIENT CARE: These findings improve our understanding of pathologic mechanisms of multiple sclerosis and Alzheimer disease and support the use of our imaging method as a biomarker for investigating cerebrospinal fluid–mediated brain clearance.