Early Tau Burden Correlates with Higher Rate of Atrophy in Transentorhinal Cortex

INTRODUCTION

The recently proposed ATN system for staging Alzheimer’s disease (AD) characterizes individuals by the presence of amyloid-β (Aβ), tau-based neurofibrillary tangles (NFT), and structural neurodegeneration [1]. Investigating the relationships between these biomarkers is important to understand the underlying pathophysiology of AD. Compared to Aβ-related pathology, NFTs are more directly linked to neuronal and synaptic loss [2]. The ¹⁸F-AV-1451 [3] PET tracer promises in vivo quantification of the tau-based paired helical filaments of NFTs. Associations between ¹⁸F-AV-1451 uptake and macroscopic structural change, expected given the link between NFT pathology and neurodegeneration, were reported in recent cross-sectional studies that examined NFT-neurodegeneration relationships at the scale of the whole brain [4–7]. The current study instead examines these relationships at a granular level in the medial temporal lobe (MTL), the area of the cortex in which the spread of NFTs in the cortex originates in AD. This more focal analysis is relevant for developing and understanding biomarkers for preclinical AD, in which NFT pathology and neurodegeneration are thought to be mostly confined to the MTL [8].

A novel aspect of the current study is that it uses a longitudinal measure of neurodegeneration, as opposed to a cross-sectional measure that is likely suboptimal because it may be affected by subject variation in the volume or thickness of brain structures due to non-AD related effects, such as developmental differences. Instead, MRI-based measures of longitudinal atrophy are more directly related to active neuronal injury and thus are likely to be more closely linked to NFT burden. However, longitudinal atrophy has not been well studied in relation to tau biomarkers. Only one prior study [9] demonstrated such a relationship in a whole-brain analysis. A finding of significant association of ¹⁸F-AV-1451 uptake with localized measurements of longitudinal structural change in the earliest regions of NFT pathology (Braak stages I-II [8]) would support the potential role of these measures in monitoring of very early AD progression.

Moreover, the degree to which the linkage between NFT pathology and neuronal injury is related to the presence of cerebral amyloid is unclear, particularly within the MTL where NFT pathology is also common in the absence of amyloid, so-called primary age-related tauopathy (PART).

This study explores the association between longitudinal measures of structural neurodegeneration within the regions of earliest NFT pathology—Brodmann areas 35 (BA35), entorhinal cortex (ERC; the transentorhinal cortex is located at the ERC/BA35 boundary), hippocampus and Brodmann areas 36 (BA36; BA35 and BA36 are subregions of perirhinal cortex)—and tau burden measured by ¹⁸F-AV-1451 PET across the MTL as a whole. This relationship is examined separately within Aβ negative (Aβ–) and positive (Aβ+) subjects to determine the degree to which it is modulated by cerebral amyloid.

METHODS

Neuroimaging data processing

Quantify region of interest (ROI) and pointwise longitudinal rate of structural change

The pipeline used for quantification of rate of structural change from the MRI scans is specially designed for MTL substructures, which overcomes crucial limitations of conventional approaches that are often optimized for whole-brain analysis. Figure 1 is a schematic flowchart that summarizes the pipeline and highlights its unique aspects. In brief, it contains the following three steps.

OPEN IN VIEWER

Fig.1

Schematic flowchart of the tailored pipeline that quantifies region of interest (ROI) and pointwise longitudinal rate of structural change. The pipeline is specially designed for the medial temporal lobe (MTL) substructures with the three unique aspects highlighted in the figure. CS, collateral sulcus; ERC, entorhinal cortex; BA35 and BA36, Brodmann areas 35 and 36.

1) Segmentation of ROI: A multi-atlas segmentation algorithm was used to segment eight MTL ROIs: bilateral hippocampus, ERC, BA35, and BA36 in the proximal MRI of each subject. The algorithm warps a set of expert-labeled MRI scans called “atlases” to the target MRI scan and combines them into a consensus segmentation [10]. Segmentation of MTL cortex in T1-weighted MRI is challenging because it is hard to separate the dura mater from gray matter in T1-weighted MRI. As shown in Fig. 1, the dura mater is juxtaposed to inferior MTL cortex and has similar signal intensity to gray matter in T1-weighted MRI. It is commonly mislabeled as gray matter by conventional methods, which may confound findings of research studies. By contrast, in T2-weighted MRI, dura mater can be easily separated from gray matter due to its dark appearance (Fig. 1). Uniquely, our atlases are assigned a separate label for the dura mater using information from both T1-weighted and T2-weighted MRI images in the same subjects. This was shown to improve the segmentation accuracy over conventional segmentation methods that do not account for this confound explicitly [11]. A composite “MTL-cortex” ROI was generated by combining ERC, BA35 and BA36 of each hemisphere.

2) Establishing pointwise correspondence of MTL cortex: Quantifying regional effects within MTL cortex requires anatomically meaningful pointwise correspondence between all subjects. To achieve this, the subjects’ MTL cortical ROIs were coregistered in a groupwise fashion to a common template using a specialized algorithm that accounts for large variability in the folding patterns of the MTL cortex by generating and linking multiple intermediate templates [12, 13]. Our prior study showed that the algorithm is able to establish anatomically meaningful correspondence while the conventional approaches cannot [12, 13]. Pointwise and summary measures of cortical thickness were derived from the template fitting. Each hemisphere was processed separately.

3) Computation of annualized atrophy rate: Annualized atrophy rate in hippocampal volume and in the thickness of the cortical ROIs was estimated using symmetric diffeomorphic registration [14] between the two MRI scans, as implemented in Automatic Longitudinal Hippocampal Atrophy software. Conventional approaches commonly choose one MRI scan as the fixed image and perform registration on the other one, which may introduce bias in longitudinal change estimation. Instead, the symmetric diffeomorphic registration brings the two MRI scans into a “middle” image space and was shown to be unbiased with respect to the ordering of the MRI scans and highly sensitive to small longitudinal changes [15].

MTL ¹⁸F-AV-1451 uptake

The pre-processed ¹⁸F-AV-1451 scans were first rigidly aligned to the proximal MRI of the same subjects, and then converted to a standardized-uptake-value-ratio (SUVR) maps using mean ¹⁸F-AV-1451 uptake in cerebellar gray matter as reference. ¹⁸F-AV-1451 uptake in the left/right “MTL-cortex” ROI was then derived.

Amyloid-β status

The Florbetapir PET scans were converted to SUVR maps using whole cerebellum as reference. A global Florbetapir SUVR was computed by averaging mean SUVR in anterior/posterior cingulate, precuneus, middle/superior temporal cortex, middle frontal, inferior parietal, and supra marginal gyrus. Subjects with global SUVR≥1.11 were regarded to be Aβ+.

RESULTS

Among Aβ+ subjects, MTL ¹⁸F-AV-1451 uptake was significantly positively correlated with annualized atrophy rate in MTL cortex (Table 1), with the largest effect in the right BA35 (scatter plot in Supplementary Figure 1), but not in bilateral hippocampus. Pointwise analysis results (Fig. 2A) demonstrate that the correlation of annualized atrophy rate with tau burden was most prominent in the bilateral transentorhinal cortices (lateral ERC/BA35), the area that Braak and Braak [8] identified as the earliest cortical site of NFT deposition (Fig. 2B). No significant correlation was found in Aβ– subjects in both ROI and pointwise analyses.

OPEN IN VIEWER

Fig.2

A) Regional statistical analysis results of annual longitudinal change versus medial temporal lobe ¹⁸F-AV-1451 uptake within Aβ+ subjects (n = 33) from the Alzheimer’s Disease Neuroimaging Initiative (ADNI). Bottom row shows the corresponding label maps. The correlation was most prominent in the bilateral transentorhinal cortices (lateral ERC/BA35), the earliest cortical site of NFT deposition [Braak I, shown in (B), adapted from Braak and Braak (1995) [8]]. No significant correlation was found in Aβ- subjects (n = 36). Black contours highlight significant clusters after corrected for multiple comparisons in cluster level (p < 0.05, clusters were defined using an empirical threshold of p = 0.01). NFT, neurofibrillary tangle; ERC, entorhinal cortex; BA35 and BA36, Brodmann areas 35 and 36.

DISCUSSION

We found significant positive associations between MTL ¹⁸F-AV-1451 uptake, a putative measure of NFT burden, with antecedent annualized atrophy rate in several MTL regions in individuals with evidence of cerebral amyloid. This finding is consistent with the notion that increased accumulation of tau pathology is linked to more robust neurodegeneration. Moreover, pointwise analysis further localized the strongest effect to the transentorhinal cortex. Finally, the lack of this relationship in the absence of cerebral amyloid suggests that the active neurodegeneration associated with NFT pathology may be mediated by the presence of brain amyloid.

These findings are consistent with Braak staging of NFT pathology [8]. Over the course of AD progression, sites of earliest NFT pathology are likely to accumulate more NFTs and to undergo more neurodegeneration than other MTL regions. This difference should be particularly pronounced in preclinical and prodromal AD, where NFT pathology is largely circumscribed to early Braak regions. The finding of strongest correlations between annualized atrophy rate and ¹⁸F-AV-1451 uptake in the bilateral transentorhinal cortices fits this model, and provides further validation for the clinical significance of ¹⁸F-AV-1451 in tracking AD-related neurodegeneration in the MTL.

No significant correlation was found in Aβ- subjects. PART is a common feature of Aβ- individuals in the age range of this cohort with NFT burden usually spanning Braak stages I-IV and, thus, including the regions examined here. Some have argued that a proportion of the Aβ- cognitively normal and MCI patients with evidence of AD-like neurodegeneration, often referred to as having suspected non-Alzheimer’s pathophysiology (or SNAP) [17], may have PART as the driver of their neurodegenerative change. However, the lack of correlation here between atrophy rate with MTL ¹⁸F-AV-1451 uptake suggests that NFTs may have a more limited effect on the rate of neurodegeneration in the absence of cerebral amyloid. If this is the case, perhaps a longer timeframe than the current study would be necessary to reveal this relationship. Alternatively, it could be that there is little or no relationship between NFT pathology and neurodegeneration in the absence of cerebral amyloid (but note Josephs et al. [18] for evidence of increased antemortem MTL atrophy with increasing Braak stage in postmortem PART cases). Finally, it is also possible that ¹⁸F-AV-1451 binds less avidly to more “immature” tangles that are potentially more predominant in PART [19]. Understanding the linkage between ¹⁸F-AV-1451 and other tau tracers and PART is critical given the potential importance of this condition to age-associated cognitive changes [20].

A novel aspect of this study is the use of a tailored pipeline optimized for measuring cortical MTL substructures. Quantifying longitudinal atrophy in these regions is challenging due to large anatomical variability of sulcal patterns, particularly collateral sulcus, which influence the location of boundaries between BA35 and ERC (i.e., transentorhinal cortex) [21]. Further, when using T1-weighted MRI, dura mater, which has similar intensity as gray matter, produces an additional confound in this region. Conventional methods that do not account for these confounds, produce measurement noise that could obscure the early structural change revealed in the current analysis [11].