Increased Medial Temporal Tau Positron Emission Tomography Uptake in the Absence of Amyloid-β Positivity

Key Points Question What is the longitudinal trajectory of older individuals who show positron emission tomography–assessed medial temporal lobe (MTL) tau deposition in the absence of amyloid-β (Aβ) pathology (A− TMTL+)? Findings In this cohort study of 969 older participants, A− TMTL+ individuals displayed moderate tau accumulation mainly restricted to the MTL, which was paralleled by cerebrospinal fluid phosphorylated tau increases and colocalized atrophy progression; no significant Aβ accumulation was observed. By contrast, Aβ-positive individuals showed pronounced and cortically widespread tau accumulation, which was accompanied by extratemporal cortical atrophy and significantly faster cognitive decline. Meaning The findings of this study suggest that individuals with A− TMTL+ do not appear to be on a pathologic trajectory toward Alzheimer disease.

A myloid-β (Aβ) plaques and tau neurofibrillary tangles are the hallmarks of Alzheimer disease (AD). 1,2The presence of neurofibrillary tangles has been observed to be tightly linked to increased Aβ load. 3However, the presence of neurofibrillary tangles in the medial temporal lobe (MTL) has also been observed in older individuals without substantial Aβ pathology, 4 a condition that has been termed primary agerelated tauopathy (PART). 5][14] Positron emission tomography (PET) studies have also consistently shown increased MTL tau PET signal in a subset of individuals with negative Aβ PET scans, [15][16][17] which may reflect PART, among other possible conditions. 18The in vivo PET-based identification of these individuals also allows studying their future clinical and pathologic progression.
In this study, we used data from a large multicohort sample to study the longitudinal pathologic characteristics and future clinical course of Aβ PET-negative (A − ) individuals who show increased MTL tau PET signal (TMTL + ).Specifically, we studied baseline characteristics and longitudinal changes in cognition, neuroimaging, and cerebrospinal fluid (CSF) biomarkers in these individuals and contrasted them to biomarkernegative controls as well as to individuals with an AD-typical Aβ-and tau-positive PET profile.

Study Design
Data used in the preparation of this article were obtained from the Alzheimer's Disease Neuroimaging Initiative (ADNI), Harvard Aging Brain Study (HABS), 19 and AVID-A05 study cohorts (eMethods in Supplement 1).Informed written consent was obtained from all participants or their corresponding caregivers.All protocols were approved by each cohort's respective institutional ethical review board.This study followed the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) reporting guideline.Data were collected between July 2, 2015, and August 23, 2021.We included all participants who had undergone concurrent structural magnetic resonance imaging, Aβ PET, tau PET, and clinical evaluation within a 6-month window (N = 1093).Participants were further classified into 4 groups according to PET-based Aβ (A) and tau (T) status, as described in the Neuroimaging section: A − TMTL − (n = 250), A − TMTL + (n = 264), A + TMTL + (n = 451), and A + TMTL − (n = 128).Additionally, a subcohort of 16 healthy younger controls (maximum age, <39 years) with concurrent magnetic resonance imaging and tau PET scans from the AVID-A05 study was included for the definition of the tau PET positivity threshold.
A subset from the ADNI study had baseline and follow-up CSF biomarkers available (described in the CSF Biomarkers section), and all participants had baseline cognitive data.
Subsets of the study participants underwent follow-up neuroimaging (mean [SD], 2.36 [0.76] years for Aβ PET and 1.83 [0.84] years for tau PET) and cognitive assessments (eMethods in Supplement 1).Participants' characteristics are provided in the Table.

Neuroimaging
Magnetic resonance imaging acquisition details for ADNI, HABS, and AVID-A05 are reported in the eMethods in Supplement 1. Magnetic resonance images were segmented with FreeSurfer, version 7.1.1 and Statistical Parametric Mapping 12 (SPM12, Wellcome Department of Imaging Neuroscience, Institute of Neurology).FreeSurfer-derived regions of interest (ROI) were merged to generate masks resembling regions affected by neurofibrillary tangle pathology in Braak stages I/II, III/IV, and V/VI (eMethods in Supplement 1). 20,21FreeSurferbased cortical thickness maps were coregistered to the fsaverage template and smoothed with a 2-dimensional isotropic gaussian filter of 12 mm full width at half maximum.
PET acquisitions followed study-specific protocols that are detailed in the eMethods in Supplement 1. Tau-PET scans were acquired using [ 18 F]flortaucipir (FTP), and Aβ-PET scans were acquired using either [ 18 F]florbetapir (ADNI and AVID-A05), [ 18 F]florbetaben (ADNI), or [ 11 C]Pittsburgh compound B (HABS) radiotracers.The multicentric PET scans were preprocessed using an in-house-developed pipeline that replicated the ADNI pipeline for PET scanner harmonization. 22,23Scannerspecific gaussian filters were applied to each PET image (regardless of PET imaging modality) to reach a uniform isotropic resolution of 8 mm.
For FTP-PET scans, region-based voxelwise 24 partial volume correction was applied using the PETPVC toolbox 25 and Baker atlas. 26Global standardized uptake value ratio (SUVR) in Aβ-PET scans was quantified using the centiloid scale 27 (eMethods in Supplement 1).In addition, cortical surface SUVR maps were generated for all PET scans using FreeSurfer, 28,29 coregistered to the fsaverage template, and smoothed with a 2-dimensional isotropic gaussian filter of 10 mm full width at half maximum.

Key Points
Question What is the longitudinal trajectory of older individuals who show positron emission tomography-assessed medial temporal lobe (MTL) tau deposition in the absence of amyloid-β (Aβ) pathology (A − TMTL + )?

Findings
In this cohort study of 969 older participants, A − TMTL + individuals displayed moderate tau accumulation mainly restricted to the MTL, which was paralleled by cerebrospinal fluid phosphorylated tau increases and colocalized atrophy progression; no significant Aβ accumulation was observed.By contrast, Aβ-positive individuals showed pronounced and cortically widespread tau accumulation, which was accompanied by extratemporal cortical atrophy and significantly faster cognitive decline.

Meaning
The findings of this study suggest that individuals with A − TMTL + do not appear to be on a pathologic trajectory toward Alzheimer disease.
To minimize the effect of subthreshold Aβ burden in the A − TMTL + study group, [30][31][32] Aβ positivity was defined using a conservative cutoff of 12 centiloids. 27This cut point proved to optimally discriminate between Thal phases 0 to 1 and 2 to 5 33 and it is therefore lower compared with traditional cut points based on discrimination of AD neuropathologic change levels (24.4 centiloids 33 ) or reliable worsening (19 centiloids 34 ).The tau-positivity threshold was defined as the 95th percentile of regional entorhinal cortex (ERC) SUVR values in the younger control cohort 34 (SUVR = 1.21) (eFigure 1 in Supplement 1).

Statistical Analysis
Statistical analysis of differences between the A − TMTL − vs A − TMTL + and A + TMTL + study groups was performed using generalized linear models (GLMs) controlled for age, sex, cohort (ADNI, HABS, and AVID-A05), and baseline centiloid values in the case of A − TMTL + vs A − TMTL − comparisons.Effect sizes were measured using Cohen d, and group differences between cortical maps were corrected for multiple comparisons using the FreeSurfer clusterwise correction for multiple comparisons.Longitudinal rates of change were computed using linear mixed-effect models with participant-specific intercepts and slopes (eg, V k ~time + (time|participant), where V k is the value on the kth vertex of a cortical map).First, we investigated vertex-wise and ROI-based group differences in baseline FTP SUVRs.Vertex and ROI-based group differences were also computed for the FTP SUVR longitudinal rates of change.Additionally, group differences in longitudinal centiloid accumulation were similarly investigated.Analysis of baseline and longitudinal differences in CSF Aβ42/40 and p-tau181 biomarker levels used analogous statistical models, but values were log-transformed before analysis to account for the exponential progression of CSF biomarker levels.Baseline and longitudinal differences across groups in cognitive metrics were studied separately for cognitively unimpaired and cognitively impaired individuals because of the different neuropsychological instruments that are best suited to detect the subtle cognitive changes in participants without impairment and more overt cognitive changes in those with impairment.As post hoc sensitivity analyses, we repeated the previous analyses with higher cut points for Aβ (24 centiloids) and tau PET positivity (mean +2.5 SD of the ERC FTP, SUVR = 1.27).Moreover, we assessed the outcome of using a larger MTL ROI comprising the ERC and amygdala.
In addition to the comparisons of dichotomized A and TMTL groups, complementary analyses were performed to assess continuous associations of baseline ERC FTP SUVR with vertex-wise cortical thickness patterns across all A − individuals, using GLMs adjusted by sex, age, cohort, and baseline centiloid.Analogously, associations between baseline ERC FTP SUVR and cognitive performance were studied across the A − subcohort with equally adjusted GLMs.Statistical tests were 2-sided, and P < .05 was considered statistically significant.The strength of the associations was assessed using the Pearson partial correlation coefficient (r).

Tau and Aβ Accumulation
Analysis of baseline FTP SUVR contrast maps (Figure 1A) noted increased tau burden in A − TMTL + individuals to be most pronounced in the MTL and extending into the inferior temporal lobe and the ventromedial prefrontal cortex, while A + TMTL + individuals showed the AD-characteristic pattern of widespread cortical tau accumulation across temporal, parietal, and frontal areas.In vertex-wise longitudinal FTP SUVR analyses, A − TMTL − individuals showed little increase of tau accumulation over time, whereas the A − TMTL + cohort displayed a moderate increase of tau uptake restricted to the MTL and inferior temporal regions (Figure 1B).By contrast, A + TMTL + participants showed a pronounced and widespread increase of tau accumulation.These differences were confirmed in direct statistical contrasts between the TMTL + groups and the A − TMTL − group (Figure 1C).An ROI-based FTP SUVR analysis showcased similar results (eFigure 3 in Supplement 1), with A − TMTL + participants showing statistically significant albeit moderate longitudinal (mean [SD], 1.83 [0.84] years) tau PET increases that were largely limited to the temporal lobe, whereas those with A + TMTL + showed faster and more cortically widespread tau PET increases.

Neurodegeneration
Compared with the A − TMTL − group, A − TMTL + participants (Figure 3A) showed cortical thinning at baseline mainly restricted to the MTL, whereas A + TMTL + participants showed more widespread cortical thinning extending to the lateral temporal lobe, the posterior cingulate, and the parietal and frontal lobes.This pattern was also reflected in ROI-based analyses, with A − TMTL + individuals showing significant cortical thinning in Braak stages I/II and Braak stages III/IV only (Figure 3B).The complementary analysis using continuous tau PET measures confirmed an association between ERC FTP SUVR and medial temporal neurodegeneration across A − individuals (eFigure 5 in Supplement 1).In longitudinal analyses, A − TMTL + individuals showed faster cortical thinning compared with the A − TMTL − group that was largely restricted to the MTL, while accelerated cortical thinning in the A + TMTL + group further extended to the lateral temporal, parietal, and frontal lobes (Figure 3C).Similarly, ROI-wise analyses (Figure 3D) showed significantly faster cortical thinning in A − TMTL + participants, mostly in Braak stages I/II.

Sensitivity Analyses
Overall, the results derived from the sensitivity analyses were consistent with the main results presented in this study.Similar patterns of tau PET SUVR, CSF biomarkers, atrophy, and clinical change were found across the A TMTL groups when changing the Aβ PET cut point to 2 4 centiloids (eFigures 7-10 in Supplement 1) and when changing the Braak stages I/II SUVR cut point to 1.27 (eFigures 11-14 in the Supplement 1).Analyses using a larger MTL ROI (ERC plus amygdala) yielded a slightly different distribution of A TMTL groups (eFigure 15 in Supplement 1) and showed that the Aβ-and tau-accumulation patterns were similar to those obtained with the ERC ROI.

Discussion
In this study, we explored in detail the pathologic and clinical course of older individuals who display PET-measured tau accumulation in the MTL in the absence of Aβ pathology (A − TMTL + ), a condition reminiscent of pathologically defined PART. 5 In a large multicentric cohort of almost 1000 older individuals, we found that increased MTL tau PET signal without notable Aβ pathology is relatively common in older individuals and is associated with further longitudinal tau PET uptake increase, which remains largely restricted to the MTL.These tau PET increases colocalize with progressive MTL neurodegeneration, are associated with only subtle changes in global cognitive performance, and are not accompanied by notable accumulation of Aβ pathology over time.Using a tau PET cutoff defined in healthy younger individuals, we observed that tau PET-measured MTL accumulation in the absence of Aβ is a common condition in older individuals, representing 51% of A − individuals in this study.The frequency of tau PET positivity in this A − sample is consistent with a previous study using a similar method (67%), 17 and it is substantially higher compared with previous studies using larger temporal ROIs without partial volume correction (approximately 17%-20%). 15,37The discrepancy may be explained by use of extra-MTL ROIs without partial volume correction, which results in a lack of sensitivity to MTLspecific signal.
The degree to which FTP-PET can detect PART remains a subject of debate.9][40][41] This may lead to the conclusion that FTP-PET cannot detect PART-related tau deposition, which is, by definition, Braak stage IV or less.Yet, FTP showed binding to neurofibrillary tangles from PART brains in autoradiography studies 42,43 and, therefore, FTP-PET may detect a subset of PART cases with suprathreshold neurofibrillary tangle density.To date, the number of PART cases in the available PET-to-autopsy studies is low (n = 3) 38 and we cannot exclude that PART could be detected with FTP-PET in a subset of individuals.This hypothesis is consistent with the fact that the prevalence of tau PET positivity among older A − individuals in our study (51%) is considerably lower than the prevalence of PART in this age range in neuropathologic studies. 5The topography of our findings is also consistent with PART: in line with recent studies, 15,44,45 our results showed that increased tau PET signal in A − TMTL + individuals was largely limited to the MTL.Both baseline and longitudinal increases in tau PET signal in A − TMTL + individuals were found to be paralleled by increases in CSF p-tau181 levels, suggesting that these signal increases reflect actual increases in tau burden.Together, these results suggest that PART may be an important neuropathologic substrate for many A − TMTL + individuals in our study, although probably not the only one. 15e also acknowledge that pathologic entities other than PART may lead to abnormal FTP-PET signal in the MTL among Aβ-negative individuals.Although FTP shows high specificity for AD-type tau aggregates in autoradiography studies, 42,46,47 extensive increases in cortical FTP-PET signal in the absence of Aβ can occur in patients with AD dementia or mild cognitive impairment, which likely represent tanglepredominant dementia. 15,480][51][52] The binding mechanisms remain unclear, although binding to non-AD tau as well as to neurodegenerative processes that colocalize with TDP-43 deposition might result in nonspecific FTP binding. 49,53Therefore, we cannot exclude the possibility that limbic-predominant age-related TDP-43 encephalopathy, which is associated with neurodegeneration in the MTL, might also result in abnormal FTP-PET signal in the same regions, although the influence of limbic-predominant age-related TDP-43 encephalopathy on FTP-PET appears to be limited. 54These considerations suggest that the A − TMTL + group likely represents both a clinically and pathologically heterogeneous group.Thus, increased FTP-PET signal in the MTL in the absence of Aβ should not be considered a specific marker of PART.Despite these limitations, our findings are valuable and contribute to understanding the clinical and pathologic course of the A − TMTL + group as a whole.Yet, given its high frequency among cognitively unimpaired individuals and those with mild average cognitive decline, the clinical significance of the A − TMTL + profile remains uncertain.Additional work is needed to identify the subset of A − TMTL + individuals who will experience more relevant clinical outcomes.
3][14] Herein, we noted that A − TMTL + individuals did not show significant Aβ accumulation over the available follow-up period, thus arguing against the possibility that this condition reflects an early tau-first subtype of AD. 55 Longitudinal cortical thickness analysis demonstrated that A − TMTL + individuals have moderate and restricted MTL atrophy progression, whereas atrophy is more accelerated and spreads to widespread neocortical regions in A + TMTL + participants.These results suggest that tau accumulation in Aβnegative individuals is not a benign process but is associated with increased neurodegeneration, 56 although the rates of progression are significantly slower compared with rates in A + TMTL + participants.
In cognition analyses, we did not find significant differences in baseline performance or longitudinal decline between A − TMTL + individuals and the A − TMTL − controls, whereas a significantly faster decline was observed in cognitively impaired A + TMTL + individuals.However, a more sensitive analysis using continuous tau measures showed an association of ERC tau uptake with worse cognition and faster cognitive decline also in cognitively impaired A − participants.These results suggest that, in the absence of Aβ, tau accumulation in the MTL has only subtle effects on cognition and does not herald the pronounced cognitive decline typical for AD.Further research with longer follow-up might be necessary to delineate the long-term consequences of ongoing tau accumulation in the absence of Aβ.

Limitations
This study has limitations.The first of these is the lack of autopsy data of A − TMTL + individuals, which leaves the exact association between the PET-defined A − TMTL + group and PART to be determined.Second, we relied on cutoffs for group definition.While centiloid cutoffs for denoting Aβ status are well established, 27 a number of different methods and cutoffs for defining tau PET positivity have been used in the literature, resulting in highly variable proportions of the different A and TMTL groups. 57Herein, we applied a commonly used method for objectively defining biomarker cutoffs based on data from healthy younger controls, 34 and several of our principal findings were replicated in complementary continuous analyses that are independent of cutoff definition.Third, to achieve robust sample sizes of the less-prevalent A − TMTL + individuals, we pooled data across different cohorts.While the possible influence of multicentric data acquisitions was minimized by harmonizing imaging preprocessing, it limited our ability to analyze domain-specific cognitive decline, as neuropsychological instruments differed across cohorts.Fourth, follow-up time for the evaluation of both longitudinal clinical and biomarker measures was relatively short.Fifth, the cohorts included in our study represent selective research cohorts that may not reflect the general population, and our findings should be replicated in more diverse cohorts.

Conclusions
The results presented in this longitudinal cohort study suggest that individuals with MTL tau accumulation in the absence of Aβ follow a separate, less malign, pathologic course compared with that of typical AD.While these individuals showed progressive tau accumulation and neurodegeneration, this process was comparably slow, remained largely restricted to the MTL, and was associated with only subtle changes in global cognitive performance.Moreover, these individuals did not show notable Aβ accumulation over followup, arguing against the possibility that this A − TMTL + condition reflects a tau-first subtype of AD.Further studies are warranted that specify the exact association of this common PET-defined condition with pathologic PART.
Cognitive performance in cognitively unimpaired individuals was assessed using a modified version of the Preclinical Alzheimer Cognitive Composite 36 (PACC) derived as the sum of the z scores of the Mini-Mental State Examination total score, Log-Transformed Trail Test B, and Logical Memory Delayed Recall (PACC-3).The PACC-3 is designed to detect the first signs of cognitive decline in otherwise asymptomatic individuals.Cognitive performance in cognitively impaired individuals (combined mild cognitive impairment and AD dementia) was assessed using the Alzheimer's Disease Assessment Scale-Cognitive Subscale (ADAS-Cog 11).

Figure 1 . 8 A
Figure 1.Cross-Sectional and Longitudinal Characterization of Amyloid-β (A) and Tau (T) Positron Emission Tomography Accumulation in the Medial Temporal Lobe (MTL) Group differences in cross-sectional FTP SUVR patterns A

Figure 4 . 15 P = . 19 A
Figure 4. Baseline and Longitudinal Characterization of Cognitive Performance