In-vivo Measures of Tau Burden are Associated with Atrophy in Early Braak Stage Medial Temporal Lobe Regions in Amyloid Negative Individuals

Abstract

Introduction:

It is unclear the degree to which tau pathology in the medial temporal lobe (MTL) measured by ¹⁸F-Flortaucipir positron-emission tomography (PET) relates to MTL subregional atrophy and whether this relationship differs between Amyloid-β positive (A+) and negative (A−) individuals.

Methods:

We analyzed correlation of MTL ¹⁸F-Flortaucipir uptake with MTL subregional atrophy measured with high-resolution magnetic resonance imaging (MRI) in a region of interest (ROI) and regional thickness analysis and determined the relationship between memory performance and PET and MRI imaging measures.

Results:

Both groups showed strong correlations between ¹⁸F-Flortaucipir uptake and atrophy, with similar spatial patterns. Effects in rhinal cortex recapitulated Braak staging. Correlations of memory recall with atrophy and tracer uptake were observed.

Discussion:

Correlation patterns between tau burden and atrophy in the A− group mimicking early Braak stages suggests that ¹⁸F-Flortaucipir is sensitive to tau pathology in primary age-related tauopathy. Correlations of imaging measures with memory performance indicate that this pathology is associated with poorer cognition.

Introduction:

Alzheimer’s Disease (AD) is defined by the presence of two core pathologic species, amyloid plaques and neurofibrillary tangles (NFTs) composed of the proteins amyloid-β and tau, respectively. NFTs are thought to be more directly related to neuronal loss and synaptic dysfunction and, thus, to the cognitive symptoms of the disease relative to amyloid plaques (Bobinski et al., 1997; Fukutani et al., 1995). PET ligands now exist that allow for in vivo visualization of both amyloid plaques and tangles, and these, along with cerebrospinal fluid measures of amyloid-β and phosphorylated tau, are now being used to define Alzheimer’s Disease in research settings (Jack et al., 2018).

Braak and Braak (Braak & Braak, 1991) described six stages of NFT spread that begins within the medial temporal lobe (MTL), specifically in a region which they referred to as transentorhinal cortex, which largely coincides with Brodmann Area 35 (BA35) and represent the medial portion of perirhinal cortex (PRC). From the transentorhinal cortex (Braak Stage I), NFTs spread medially in the entorhinal cortex (ERC; Braak Stage II) followed by the CA1 subfield of the hippocampus (Braak Stage II/III). Eventually, NFTs extend outside of the MTL to heteromodal association areas and then throughout much of the cortex (Braak IV-VI). Notably, NFTs appear to develop and generally follow this topographic pattern both in the presence or absence of amyloid-β pathology, but their spread is more aggressive when there is concomitant amyloid-β pathology (L. Wang et al., 2016). Conversely, NFT pathology in the absence of cerebral amyloid was recently termed Primary Age Related Tauopathy (PART) (Crary et al., 2014). By definition, PART includes individuals who are Braak Stage ≤ IV and with amyloid plaque burden consistent with CERAD score of 0 (“Definite” PART) or 1 (“Probable”).

It remains uncertain whether current tau tracers are sensitive to PART (Lowe et al., 2016; Maass et al., 2018; Marquié et al., 2015). Further, it is not clear the degree to which MTL measures of tau relate to atrophy within specific MTL subregions in both amyloid-β positive (A+) and negative (A−) individuals and whether these are consistent with Braak staging. If this is the case in those without cerebral amyloid, this would provide convergent validity to the sensitivity of a given tau tracer to PART.

The current study leverages the multi-site acquisition of amyloid (¹⁸F-florbetapir or ¹⁸F-florbetaben) and tau (¹⁸F-flortaucipir) PET and high-resolution T2-weighted MRI (0.4×0.4×2.0 mm³) specifically prescribed for MTL subregional segmentation, collected as part of the Alzheimer’s Disease Neuroimaging Initiative (ADNI). ¹⁸F-Flortaucipir has demonstrated consistent correlation with local brain atrophy in patients across the AD continuum and binding to the paired helical filamentous tau that make up the NFTs associated with AD in post-mortem studies (Bejanin et al., 2017; S.R. Das et al., 2018; Lowe et al., 2016; Marquié et al., 2015). While PET lacks the resolution for measuring tau burden within specific MTL subregions, we will use a summary measure of MTL cortex uptake and relate it to structural changes in these more granular regions as measured from MRI. In the largely cognitively normal and prodromal AD cohort studied here, we predict strongest correlations between structural changes in the earliest Braak stage regions (BA35, ERC, CA1) with the summary measure of MTL cortex ¹⁸F-Flortaucipir uptake in both A+ and A− individuals.

Methods:

Participants:

Data used in this study were obtained from the ADNI database. The study cohort consisted of 185 participants; 113 labeled as A− and 72 labeled as A+ based on their Amyloid-PET scans. Amyloid-β status (A+ vs A−) was determined from participants’ latest available ¹⁸F-Florbetapir (N=183) or ¹⁸F-Florbetaben (N=2) scans using an SUVR threshold of ≥ 1.11 for ¹⁸F-Florbetapir or 1.08 for ¹⁸F-Florbetaben computed from a well-established composite ROI (Landau et al., 2013). Quantiles of the interval between MRI and PET scans in days were (“+” means PET earlier, “-“ means MRI earlier) : median = −5, 10% = −707, 25%= −41, 75% = 1, 90% = 8, range −1159 to 692. We utilized SUVR data made available by ADNI based on the post-processing steps described at http://adni.loni.usc.edu/methods/pet-analysis-method/pet-analysis/. We chose the most recent amyloid PET scan, rather than, in a few cases, the closest one to the MRI, reducing the possibility of some participants accumulating amyloid and crossing the threshold to A+ after the MRI.

A summary of participants’ demographic information and psychometric measures are reported in Table 1.

Table 1.

Participants characteristics. Mean and standard deviation are shown. Logical memory score was obtained from the NEUROBAT_LDELTOTAL variable in ADNI.

Group (N)	A− CN (68)	A− MCI (30)	A+ CN (41)	A+ MCI (30)	A+ Dementia (16)
Age (Years)	69.52 (5.86)	70.82 (5.87)	72.58 (5.87)	71.86 (7.52)	70.48 (8.74)
Education (Years)	16.57 (2.42)	16.17 (3.34)	17.05 (2.36)	16.13 (2.75)	15.56 (2.78)
Sex (M/F)	36/32	13/17	22/19	13/17	10/6
Logical Memory	14.49 (3.46)	10.33 (4.11)	14.8 (3.22)	7.03 (5.28)	2.19 (3.64)
MMSE	29.21 (1.19)	28.63 ( 1.45 )	28.78 (1.51 )	27.2 (2.17)	21.33 (4.98)
ApoE status (E4 Carrier %)	17.6%	13.3%	43.9%	60%	50%

Image Acquisition:

A high-resolution T2-weighted structural MRI specifically optimized for imaging the MTL (in-plane resolution of 0.4×0.4 mm², slice thickness 2.0 mm) was used for making subregional volume and thickness measurements. The high in-plane resolution makes it possible to visualize the internal structure of the hippocampus and to separate MTL cortical gray matter from dura mater, making these scans ideal for subregional MTL morphometry. ¹⁸F-Flortaucipir PET scans acquired closest in time to the structural MRI were used for estimating tau burden in MTL (quantiles of the interval between MRI and PET scans in days were: median = −14, 10% = −565, 25% = −64, 75% = 0, 90% = 1.6, range −1377 to 521). Tau PET imaging consisted of a continuous 30 minute brain scan (6 frames of 5 minute duration) 75 minutes following approximately 10 mCi of ¹⁸F-Flortaucipir injection. ¹⁸F-Flortaucipir PET images were downloaded from the ADNI data archive in the most fully post-processed format with the image description “Coreg, Avg, Std Img and Vox Siz, Uniform Resolution”.

Image Processing:

T2-weighted MRI was automatically segmented using the multi-atlas labeling technique implemented in the ASHS software (Yushkevich et al., 2015), providing labels for the hippocampal subfields CA1, CA2, CA3, Dentate Gyrus (DG) and subiculum (SUB), and regions of extrahippocampal MTL cortex including entorhinal cortex, perirhinal cortex, subdivided into Brodmann area 35 36, and parahippocampal cortex (PHC). These segmentations were used to generate volumetric measurements of the hippocampal subfields. In addition, a multi-template shape analysis technique that explicitly accounts for the existence of multiple discrete variants of the collateral sulcus (Ding & Van Hoesen, 2010) was applied to the segmentations, resulting in point-wise correspondence maps of the MTL subregions across all subjects in an MTL-specific template space. These maps are then used to generate summary thickness measurements for the extrahippocampal cortical ROIs, and to conduct regional thickness analysis within the ribbon that includes the aforementioned MTL cortical subregions in addition to hippocampal subfields CA and subiculum. All MRI images, automatic MTL segmentations and quality of coregistration between structural MRI and PET images were visually checked for acceptable quality. 37 datasets were excluded from 222 participants with MRI and both PET modalities).

The post-processed ¹⁸F-Flortaucipir PET images were generated by averaging co-registered individual frames, re-oriented in a standardized image space such that the anterior-posterior axis of the subject is parallel to the AC-PC line, followed by scanner-specific filtering to generate an image with a uniform isotropic resolution of 8 mm FWHM. More details about the pre-processing employed is described at http://adni.loni.usc.edu/methods/pet-analysis-method/pet-analysis/. Post-processed PET images were then registered to subject’s T1-weighted structural MRI using ANTs (Avants, Epstein, & Gee). The following ANTs parameters were used: Metric: Mattes mutual information (weight=1, number of bins=32), Transformation model: Rigid (gradient step = 0.2), Smoothing levels = 4×2×0, Shrink factor = 4×2×1. The MRI scan was parcellated into cerebellar, cortical, and subcortical ROIs using a multi-atlas segmentation method (H. Wang et al., 2011). The atlas set and the resulting segments are described in (Landman & Warfield, 2012). Mean PET tracer uptake in cerebellar gray matter was used as a reference region and a standardized uptake value ratio (SUVR) map was generated for each participant. A summary measure of MTL cortex tau burden was computed as the average ¹⁸F-Flortaucipir SUVR in a composite MTL cortical ROI consisting of BA35 region of perirhinal cortex and ERC. We did not include hippocampus in this summary measure given its proximity to choroid plexus and known off-target binding of ¹⁸F-Flortaucipir (Lee et al., 2018).

Statistical Analysis:

Partial linear correlation was used to assess the relationship between MTL cortex tau burden and subregional atrophy in A+ and A− participants separately. Volumes of hippocampal subfields (CA1, CA2, CA3, DG, and SUB) and thickness of extrahippocampal cortices (BA35, BA36, ERC, and PHC), averaged over both hemispheres, were correlated with MTL cortex tau burden. Age, sex, and time between MRI and ¹⁸F-Flortaucipir PET scan were used as nuisance covariates in all analyses. In addition, intracranial volume (ICV) was a nuisance covariate in analyses involving hippocampal subfield volumes. Bonferroni correction was used for multiple comparisons correction. An additional interaction analysis was carried out in the entire cohort where an interaction term between composite amyloid-PET SUVR and ¹⁸F-Flortaucipir MTL cortex was used as a predictor of atrophy in the linear model. In addition to these ROI analyses, regional thickness analysis was performed in the MTL-specific template. The summary measure of MTL cortex tau burden was correlated with local thickness using the same linear correlation framework as used in ROI analysis. To account for multiple comparisons, this analysis employed cluster-level family-wise error rate (FWER) correction (Hayasaka & Nichols, 2003) with clusters defined using an empirical threshold of uncorrected p < 0.05. Permutation testing with 1000 permutations (with clusters pooled between the left and right hemispheres) was used to assign a corrected p-value to each cluster.

In addition to the multimodality imaging correlations, atrophy and tau measures were also correlated with delayed recall on the Wechsler Memory Scale-Revised Logical Memory Test (Wechsler, 1987) in a partial correlation framework. Cut-offs for this measure determine MCI status in ADNI. For this analysis, age, education, and time between cognitive testing and imaging were used as nuisance covariates. In addition, ICV was used as covariate when volumetric atrophy measure was used. As a supplementary analysis, these correlations were repeated with the Rey Auditory Verbal Learning Test (AVLT; AVLT 30-minute Delayed Recall and AVLT d-prime recognition score) (Rey, 1964).

Sensitivity Analysis:

In order to mitigate against the possibility that individuals with sub-threshold levels of amyloid accumulation might be erroneously classified in the A− group, the primary analysis of ¹⁸F-Flortaucipir uptake vs. atrophy correlations were repeated with a lower amyloid threshold of 1.0 to determine group status. For this analysis, two subjects with ¹⁸F-Florbetaben scans were excluded.

Results:

ROI analysis:

Results of ROI analysis examining relationship between tau burden measured by ¹⁸F-Flortaucipir and volume or thickness are presented in the left two columns in Table 2. Broadly, several ROIs showed significant correlations in both the A− and A+ groups. In the A− group, BA35 and ERC emerged as statistically the most significant, in absolute terms, with CA1 and subiculum also showing significant, but less strong correlations and subiculum not surviving Bonferonni correction. Only PHC and DG did not reach the threshold for statistical significance consistent with these regions having less early NFT pathology based on Braak staging. Scatterplots of BA35 and CA1 are displayed in Figure 1. Note that one outlier was removed from the A− group who had much higher ¹⁸F-Flortaucipir MTL uptake (SUVR=1.98) and significant atrophy. These effects were not present when the analysis was restricted to only cognitively normal A− individuals, likely due to a limited range of variance for the imaging measurements.

Table 2.

Partial correlations of MTL subregional atrophy with MTL cortex tau burden (left) and logical memory scores (right) in the A− and A+ groups. The last row of each table also shows correlation between MTL cortex tau burden and logical memory. Covariates include time intervals between all pairs of measures included in a given analysis, age, sex, ICV (for volumes but not for thickness), and years of education (for memory correlations).

A− Group
	Partial correlation with summary MTL cortex tau burden		Partial correlation with logical memory scores
ROI	p-value	Partial r	p-value	Partial r
BA35	0.0001**	−0.38	0.003*	0.34
BA36	0.022	−0.24	0.03	0.26
ERC	9.00E-05**	−0.39	0.001**	0.36
PHC	0.05	−0.20	0.55	0.10
CA1	0.002*	−0.31	0.35	0.15
DG	0.06	−0.20	0.68	0.12
SUB	0.01‡	−0.26	0.40	0.13
MTL Tau			0.005*	−0.28
A+ Group
	Partial correlation with summary MTL cortex tau burden		Partial correlation with logical memory scores
ROI	p-value	Partial r	p-value	Partial r
BA35	1.50E-05**	−0.45	7.5E-08**	0.55
BA36	0.004*	−0.31	0.002*	0.33
ERC	0.0006*	−0.36	0.0004*	0.38
PHC	0.04	−0.22	0.02	0.26
CA1	1.94E-05**	−0.45	4.7E-08**	0.56
DG	6.11E-05**	−0.43	1.3E-05 *	0.46
SUB	0.001*	−0.35	9.8E-05*	0.41
MTL Tau			3.0E-09**	−0.59

Bonferroni-corrected p-values are indicated as :

^**p < 0.001

^*p < 0.05

^‡p< 0.1.

Figure 1.

Scatterplots showing MTL cortex tau burden vs. MTL subregional atrophy measures in A+ and A− groups.

A very similar pattern was observed in the A+ group in which all of the regions except for PHC displayed a significant correlation with MTL cortex ¹⁸F-Flortaucipir uptake. BA35 thickness displayed the strongest correlation, in absolute terms, of the extrahippocampal regions followed by ERC and BA36. In the hippocampus proper, CA1 volume displayed the strongest correlation with MTL cortex ¹⁸F-Flortaucipir uptake followed by DG and subiculum. In general, partial correlations were of somewhat higher magnitude in the A+ compared to A− group, as well as having a greater range of SUVR values for MTL cortex uptake (Figure 1).

Supplementary Analysis:

The partial correlation analyses were repeated to rule out effects of certain confounding factors. These were: 1) Sensitivity Analysis: When participants were assigned to A+ or A− groups based on a lower and more conservative ¹⁸F-Florbetapir composite threshold of 1.0, the results remained largely similar. These data are included as supplementary material (Table S1), 2) Interaction Analysis: When an interaction term between amyloid and tau tracer uptake was added to the partial correlation analysis, no significant interaction was found (Table S2), 3) Provenance covariate: When site of acquisition was added as covariate, it did not have a significant effect, and 4) Apolipoprotein E (ApoE) status: Adding ApoE e4 carrier status as a covariate did not have a significant effect. Another secondary analysis examined the correlation of global amyloid tracer uptake with MTL cortex ¹⁸F-Flortaucipir uptake in both groups, with age as a covariate. Significant correlation was only found in A+ (p=0.001), not in A− (p=0.6) group.

Relationship with memory performance:

Delayed recall performance on the Logical Memory Test was significantly correlated with both subregional atrophy and ¹⁸F-Flortaucipir uptake in both A− and A+ groups. The atrophy correlations were strongest, in absolute terms, in the rhinal cortices. In A−, only BA35 and ERC effects survived multiple comparisons correction (Table 2).

Measures of the AVLT test scores also showed strong correlations with atrophy and ¹⁸F-Flortaucipir uptake in A+, with BA35 showing the strongest effects in absolute terms among the cortical regions, and CA1 among the hippocampal subfields. Correlations with AVLT scores were weaker in A−. The only significant correlations were found in BA35 and ERC, but these effects did not survive Bonferroni correction (Table S3).

Regional thickness analysis:

Regional thickness analysis helps visualize tau-atrophy correlations within the MTL cortical ribbon consisting of the extrahippocampal cortical ROIs and extending into SUB and CA subfields of hippocampus. In contrast to the ROI analysis, this technique allows for exploration of regional clusters of significant correlations that can extend across subregional boundaries. Figure 2 recapitulates the results of ROI analysis in the MTL cortex, displaying strong clusters of significant correlation of ¹⁸F-Flortaucipir uptake with thinning in BA35 and ERC clusters in both groups, regardless of amyloid status. In the hippocampus, significant clusters can be seen straddling CA1 and SUB in the A+ group, with only smaller clusters present in the hippocampus in the A− group.

Figure 2.

T-statistic map of partial correlation between tau burden and local thickness in the MTL cortical ribbon, with time between MRI and PET scans and age as covariates. Significant clusters after FWER correction are circumscribed with black contours. There is considerable overlap between A+ and A− clusters on both sides. However, hippocampal clusters in A− tend to extend farther distally from the subiculum-CA boundary in the body region near the CA1/CA2 boundary, e.g. as indicated by the red circle on the left hemisphere.

Figure 2 also displays the overlap of these significant correlations between the two groups. Remarkably, there is considerable overlap in BA35 and ERC, particularly around the region Braak and Braak referred to as transentorhinal cortex. In the CA1 and subiculum region, although most of the areas of high correlations in A− overlapped with the A+ clusters, they did not reach statistical significance with FWER correction. There is some tendency for the region of significance in the A− group compared to A+ group to be shifted more towards the CA1/CA2 boundary compared to the A+ group (see red circle in Figure 2). In both the A+ and A− groups, the maps of the relationship between ¹⁸F-Flortaucipir uptake and MTL thinning exhibit remarkable symmetry between the left and right hemispheres.

Discussion:

In this study we present strong evidence for the in vivo sensitivity of ¹⁸F-Flortaucipir to tau burden in PART and its linkage to structural changes in the MTL and memory. A defining feature of disease stage in Alzheimer’s Disease is the spread of NFTs in the MTL and beyond, which have come to be described as Braak stages based on the landmark work of Braak and Braak (Braak & Braak, 1991). A relatively stereotyped pattern of involvement within specific regions of the MTL constitute the first three Braak stages and support the notion of differential vulnerability of the integrity of these structures to the earliest pathologic changes of AD. While this pattern of NFT spread is linked to AD, biologically defined by the presence of both amyloid-β plaque and NFT pathology (Hyman et al., 2012; Montine et al., 2012), Braak and Braak originally described a large number of individuals with minimal or no cerebral amyloid-β who also exhibited NFTs consistent with early Braak stages (Braak & Braak, 1997). Indeed, by the mean age of the present cohort, more than 80% of the individuals in the Braak and Braak cohort were classified as at least Braak Stage I or II, regardless of the presence of amyloid-β.

These individuals who exhibited NFTs with none to minimal cerebral amyloid-β were more recently given the moniker of Primary Age Related Tauopathy (PART), a pathological category defined by Braak Stage ≤ IV and either absent or sparse amyloid plaque burden (Crary et al., 2014). While the clinical implications of PART remain uncertain, and may reflect a primary driver of “normal” age-associated decline, some post-mortem work has suggested that the degree of NFT pathology in PART is associated with ante-mortem cognitive decline and hippocampal atrophy (Jefferson-George, Wolk, Lee, & McMillan, 2017; Josephs et al., 2017). Indeed, Crary and colleagues described progressively lower MMSE scores with higher Braak stage in PART (Crary et al., 2014).

The current study leveraged advances in in vivo imaging to estimate MTL cortex tau burden with the PET tracer, ¹⁸F-Flortaucipir (Xia et al., 2013) and measures of MTL subregional structural integrity with MRI (Yushkevich et al., 2015). We examined individuals from ADNI who had both an ¹⁸F -Flortaucipir tau PET and an amyloid-PET (¹⁸F -Florbetapir or ¹⁸F –Florbetaben) scan, as well as a high in-plane resolution T2-weighted scan specifically prescribed for MTL subregional measurement. A number of studies have reported that uptake of ¹⁸F-Flortaucipir in the MTL cortex correlates with hippocampal, entorhinal, and transentorhinal volume or thickness in A+ individuals (Das et al., 2018; Maass et al., 2018; Wang et al., 2016). Further, we recently demonstrated that BA35 and ERC displayed increased longitudinal thinning in relationship to higher ¹⁸F-Flortaucipir SUVR in MTL cortex (Xie et al., 2018) in A+ individuals, although that study did not find any effects in A−. The current findings provide further support for tau burden in the MTL being disproportionately associated with the regions affected earliest by NFTs based on the pathological literature.

It is less clear the degree to which ¹⁸F-Flortaucipir is sensitive to the NFTs of PART. Autoradiographic studies have reported binding of ¹⁸F-Flortaucipir in PART cases that varies based on the degree of maturity of the tangles (Lowe et al., 2016). In vivo studies have provided mixed results with regard to the degree to which tracer uptake correlates with hippocampal or MTL structural measures in A− individuals with some studies showing a relationship (Maass et al., 2018) and others not (Das et al., 2018; Wang et al., 2016). These discrepencies may reflect issues with sample sizes and, thus, power across studies. Finally, some work has supported a relatively specific age-related increase in ¹⁸F-Flortaucipir uptake in A− individuals consistent with the increased prevalence and degree of NFT deposition in PART (Maass et al., 2018; Schöll et al., 2016).

The current study utilized an MRI acquisition and automated segmentation approach that allows for more granular measurement of MTL subregions and has been previously shown to be particularly sensitive to early changes in this region associated with AD pathology (Wolk, Das, Mueller, Weiner, & Yushkevich, 2017; Yushkevich et al., 2015). Furthermore, thickness measurements determined in a pointwise manner across the MTL surface were specifically developed to account for anatomic variants common to the collateral sulcus that impact the lateral boundaries of ERC, BA35 and BA36 (Xie et al., 2017). In this context, we observed significant correlation of MTL cortex tracer uptake with cortical thinning most prominently in BA35 and ERC. CA1 and subiculum volume, while weaker in absolute terms than BA35 and ERC, were the regions with strongest correlation within the hippocampus proper.

These findings appear to support the sensitivity of ¹⁸F-flortaucipir to PART and suggest that PART has consequences for the integrity of the MTL. While the relationship described could reflect some kind of non-specific tracer uptake related to neurodegeneration rather that PART, the fact that the regions most strongly related to this uptake are regions associated with NFTs in Braak stage I and II suggest a relatively specific pattern best explained by NFT pathology. Moreover, the pattern of findings was quite similar to that of the A+ group in which NFTs almost certainly modulate these regional effects. In both groups, BA35 and ERC are the MTL cortical regions that show the strongest effects in ROI analysis. More impressively, the pink label in Figure 2 shows remarkable overlapping regions of significant correlation of local thickness and MTL cortex ¹⁸F-Flortaucipir uptake in the two groups, particularly in the transentorhinal region. This finding is remarkable, as PART and AD-related NFTs are thought to largely display the same spatiotemporal trajectory as demonstrated here. The presence of MTL atrophy in the setting of PART is also consistent with antemortem imaging in pathologically determined cases in which higher Braak stage was associated with greater anterior hippocampal atrophy (Josephs et al., 2017). One potential difference between the A+ and A− group in the pattern of atrophy was in the hippocampus proper in which the A+ group had stronger correlations. This may reflect the greater range of tracer uptake and tau pathology in A+. Since NFT pathology begins in the MTL cortex and then spreads to the hippocampus, the more restricted range in A−, may explain the relatively isolated effects in the MTL cortex in this group. However, it should be noted that some individuals with dementia in the A+ group actually had relatively low tau burden within the range of those in the cognitively normal group. These individuals are more likely to have other concomitant pathologies such as TDP-43, or diminished reserve resulting in their dementia level impairment with less NFTs. Additionally, there was a hint that there was relatively greater atrophy associated with MTL SUVR more superio-medially towards the CA1/CA2 boundary in the A− group relative to the greater subicular/CA1 correlation in the A+ group. While clearly not definitive, this finding is consistent with some recent work suggesting PART is associated with relatively greater CA2 burden than AD-related NFTs (Jellinger, 2018).

It is worth noting that the fact that the A− group included individuals with MCI is consistent with autopsy studies of PART in which higher Braak stages are associated with poorer cognition (Crary et al., 2014). Moreover, the regions which displayed the strongest relationship of atrophy to MTL cortex tracer uptake, BA35 and ERC, in the A− group were the regions correlated with delayed recall performance on the Logical Memory Test, similar to that of the A+ group. This finding is consistent with autopsy data in which Logical Memory performance declines more rapidly in individuals with PART who have higher Braak stages of NFT pathology (Jefferson- George et al., 2017). Notably, another study did not find a relationship between ERC tau burden and cognitive decline in A−, but included only cognitively normal controls (Sperling et al., 2018). Similarly, the reported correlations were not present when the A− group was limited to cognitively normal individuals in the current dataset. It is notable that the correlations of MTL subregional atrophy with logical memory were somewhat stronger in absolute terms than correlations with ¹⁸F-Flortaucipir uptake in the A− group, which could reflect that contribution of non-tau related factors also affecting structure and function in BA35 and ERC. Additionally, differences in cerebral blood flow may introduce noise in the tau measures. Cerebral blood flow is also associated with cognitive aging (Rabin et al., 2018; Vemuri et al., 2017). Contrary to prior work in A+ individuals (Maass et al., 2018), a structural measure, BA35, also correlated as strongly as tracer uptake with memory function, which may be due to the enhanced sensitivity of the more granular measurements used here, or potentially greater accuracy in the segmentations. The relationships of AVLT memory scores were similar to those of logical memory. The only significant effects, which did not survive Bonferroni correction, in the A− groups were in BA35 and ERC, the two regions of early NFT pathology. In contrast, the A+ group also showed strong correlations in hippocampal subfield CA1, a region where pathology spreads later in the disease course. Taken together, this shows how the brain-behavior relationships in the MTL follow a similar spatiotemporal trajectory in both A− and A+ groups.

There are a couple of additional caveats and limitations in the present study. One is that the SUVR cutoff for A+ and A− PET scans tends to reflect the sensitivity for Thal Aβ phase of 2 or greater or CERAD neuritic plaque frequency of “greater than sparse” (Clark et al., 2012; Murray et al., 2015). Thus, the A− group may include those with “possible” PART (Thal stage 1 or 2, sparse CERAD plaque frequency) rather than exclusively those with “definite PART (Thal stage 0, no neuritic plaques). To mitigate against this potential confound and recent work suggesting subthreshold uptake is associated with ¹⁸F-Flortaucipir uptake and interacts with cerebral amyloid levels in relationship to entorhinal thinning (Knopman et al., 2019; Leal, Lockhart, Maass, Bell, & Jagust, 2018), we performed a sensitivity analysis where a lower SUVR cutoff of 1.0 was used. This resulted in 15 additional individuals being classified as A+. The results were essentially unchanged with this more conservative cutoff, suggesting that individuals with subthreshold amyloid burden did not drive the effects in the A− group. Notably, a direct correlation between global amyloid tracer uptake and MTL cortex tau tracer uptake was only present in the A+ group, further suggesting that there was not a clear link between subthreshold amyloid and NFTs in the A− group. Nonetheless, we cannot completely rule out the possibility that some in the A−, even with a stricter SUVR cutoff, would fall in the “possible” PART category given limitations in the sensitivity of the modality. We also found that ApoE e4 carrier status, which also may influence relationships between tau and amyloid with MTL atrophy, did not impact findings when included as covariate. However, as expected, the proportion of e4 carriers was lower in the A− group.

It is also possible that the generally stronger correlations between our MTL cortex summary measure of ¹⁸F-Flortaucipir uptake and extrahippocampal regions rather than hippocampus reflects the fact that the summary ROI did not include the latter region. This choice was made to avoid the off-target uptake in the choroid plexus which can confound hippocampal measurements. We also note that the limited spatial resolution of PET introduces some amount of signal mixing from neighboring regions even in the summary MTL cortex measure.

In conclusion, the current findings support the notion that ¹⁸F-Flortaucipir is sensitive to the tau pathology associated with aging, i.e. PART, and that the degree of this pathology results in measurable atrophy in early Braak regions with consequences to cognitive function. As such, one can track this potential contributor to age-related cognitive decline and differentiate from that driven by preclinical AD with multimodal imaging. Finally, this work also supports the notion that structural measurements of regions affected by early NFT pathology may serve as sensitive surrogates of tau-mediated neurodegeneration regardless of amyloid status.

Systematic review:

The authors reviewed the literature using traditional resources such as PubMed and google scholar, as well as from recent conference presentations. There has been considerable work on relating in-vivo tau PET measures with atrophy in structural MRI. These studies have been appropriately cited.

Interpretation:

Evidence for strong correlations between MTL subregional atrophy and logical memory performance with ¹⁸F-Flortaucipir uptake that follow a strikingly similar topographic pattern in both amyloid negative and positive individuals recapitulating early Braak stages suggests that this tracer is sensitive to tau pathology in primary age- related tauopathy (PART).

Future Directions:

The manuscript opens up possibilities of testing hypotheses examining effects of tau pathology at a more granular level within the medial temporal lobe in PART, including relating tau burden with longitudinal atrophy rates in MTL subregions, as well as other functional imaging markers.