Synopsis

Keywords: Neurofluids, Alzheimer's Disease

A pathological hallmark of Alzheimer’s disease (AD) is the elevated aggregation of protein amyloid-β (Aβ) in the cerebrum. Recent studies have suggested a role for the parasagittal dural (PSD) space in cerebrospinal fluid (CSF) egress and associated protein clearance. A fully connected neural network was used to generate PSD segmentation masks from 3D T₂-weighted turbo-spin-echo data to assess the relationship between PSD space volume and Aβ burden estimated by ¹¹C-Pittsburgh Compound B in AD participants. PSD space hypertrophy was significantly associated with elevated Aβ levels and was localized to the frontal and parietal subsegments of the PSD.

Introduction

The overall goal of this work is to apply novel MRI methods to explore surrogate measures of cerebrospinal fluid (CSF) egress in the context of amyloid-β (Aβ) retention in Alzheimer’s disease (AD). The amyloid hypothesis proposes that the imbalance of Aβ production and clearance contributes to the development of AD. Studies have demonstrated that production rates of Aβ do not differ in AD¹, and rather, it may be insufficient clearance of Aβ that predominately contributes to aggregation. Increasing evidence suggests that CSF clears central nervous waste products including Aβ via the bulk and possibly recently-proposed glymphatic system². CSF egress from the subarachnoid space has been proposed partly to occur along the parasagittal dural (PSD) space which is located along the sagittal sinus^2–4. As such, changes in PSD space morphology may have relevance to reduced clearance of Aβ in adults with AD. To elucidate this relationship, we measured Aβ accumulation using PET with ¹¹C-Pittsburgh Compound B (PIB) in combination with a novel MR method for estimating PSD space volume to test the primary hypothesis that PSD hypertrophy scales with elevated global Aβ burden. Additionally, we visualize the topography of the hypothesized PSD hypertrophy to understand whether specific portions of the PSD may associate most closely with Aβ burden.

Methods

All participants had a clinical diagnosis of AD or amnestic mild-cognitive-impairment with Aβ-positivity and were recruited from behavioral and cognitive neurology clinics and provided written, informed consent. Acquisition. MRI data were acquired at 3.0 Tesla (Philips) using a body coil radiofrequency transmission and phased array 32-channel reception. The scanning protocol consisted of a 3D T₁-weighted magnetization-prepared-rapid-gradient-echo (TR=8.1 ms, TE=3.7 ms, spatial resolution=1.0x1.0x1.0 mm³) for anatomical reference and a 3D T₂-weighted volume isotropic-turbo-spin-echo-acquisition (TR=2500 ms, TE=331 ms, spatial resolution=0.78x0.78x0.78 mm³) for PSD space volumetric measurement. PET data were acquired using a Digital PET/CT scanner (Philips Vereos) with three-dimensional emission acquisition. Following an intravenous bolus injection of ¹¹C-PIB, a 70-minute dynamic PET scan was acquired and reconstructed with a direct Fourier method with Fourier reprojection (spatial resolution=2.0x2.0x2.0 mm³) algorithm. Analysis. PSD segmentations were generated from 3D T₂-weighted TSE images with a combination of a fully connected neural network based on a stacked U-net architecture⁵. The first layer estimates a binary mask of the peri-sinus space, and the second layer estimates a label map of the PSD and superior sinus lumen based on T₂-weighted signal intensities. The PSD was further segmented into prefrontal, frontal, parietal and occipital subsections. The prefrontal and frontal PSD are distinguished by a plane crossing the pituitary gland and rostrum of the corpus callosum such that the prefrontal PSD lies ventral to the plane and the frontal PSD dorsal. The parietal region is delineated from the frontal PSD by the central sulcus and extends to the parietal-occipital fissure. Finally, the occipital PSD is delineated from the parietal-occipital fissure to the posterior end of the PSD. All PSD volumes are expressed in cubic centimeters (cm³). T₁-weighted data were used to segment region-of-interest masks from the AssemblyNet algorithm⁶. ¹¹C-PIB PET data were used to generate parametric binding potential (BP_ND) maps from a two-step simplified reference tissue model (SRTM2)⁷, using cerebellar gray matter as a reference region, in PMOD (version 4.2, PMOD Technologies LLC, Zürich, Switzerland). Parametric BP_ND maps were transformed to native T₁-weighted space and a region-based voxel-wise partial volume correction was applied⁸. Global Aβ burden was estimated from a mask of commonly observed Aβ aggregation topography⁹. Statistical testing. To test the primary hypothesis that PSD hypertrophies with increasing Aβ burden, a linear regression model was defined with ¹¹C-PIB BP_ND as the dependent variable, PSD volume as the independent variable, and age and biological sex as explanatory variables. Additional analyses of the subsections of the PSD used a similar linear model with PSD subsection volume as the independent variable.

Results

Table 1 summarizes the demographics and imaging outcomes for the 23 participants in the study. The linear model assessing the primary hypothesis revealed a significant, positive relationship between total PSD volume and global BP_ND (R²=0.349, p=0.017; Figure 1). Specifically, increases in total PSD volume were closely associated with increases in global BP_ND of ¹¹C-PIB. A schematic of the PSD subregions and associated scatterplots for each subregion is depicted in Figure 2. Both the frontal and parietal PSD volumes were positively associated with global BP_ND of ¹¹C-PIB and met significant criteria (frontal: R²=0.395, p=0.008; parietal: R²=0.373, p=0.014).

Discussion

We evaluated how parasagittal dural space volume as estimated on 3D, high spatial resolution T₂-weighted MRI relates to global Aβ burden in adults with varying levels of disease burden. Analyses revealed that hypertrophy of the total PSD space is associated with elevations in global Aβ burden specifically in the frontal and parietal subsegments. These findings suggest that CSF egress within these segments of the PSD may play a relevant role in the clearance of Aβ aggregates. This finding is particularly intriguing given that the strongest relationships are seen in subsegments correlated with commonly observed Aβ aggregation topography.

Conclusion

Frontal and parietal parasagittal dural spaces, quantified from recently-validated deep learning algorithms applied to submillimeter 3D T₂-weighted MRI, were significantly larger in Alzheimer’s disease patients with elevated levels of Aβ burden.

Acknowledgements

No acknowledgement found.