Relationships Between Cognition and Neuropathological Tau in Alzheimer’s Disease Assessed by 18 F Flortaucipir PET

INTRODUCTION

The pathologic hallmarks of Alzheimer’s disease (AD) include extracellular amyloid-β plaques (Aβ) and hyperphosphorylated tau neurofibrillary tangles (NFTs). Both Aβ and NFTs are required to establish a neuropathologic diagnosis of AD [1]. Current evidence strongly supports the role of Aβ as a disease an initiating event [2], even occurring years before symptom onset, with progressive and widespread accumulation throughout neocortex [3]. NFT accumulation has also been well-described in autopsy studies [4], and is thought to follow Aβ by 5–10 years [5, 6]. More recently, Jack et al. [7] suggested that tau and Aβ pathology may develop independently, with Aβ accelerating pre-existing tau pathology. Both clinical and pathological observations suggest that tau aggregates may be found in aging individuals in the absence of Aβ [8] and may produce mild amnestic cognitive changes (known as primary age-related tauopathy) [9–11].

Although Aβ has been associated with cognitive decline, the relationship between either the quantity or the spatial distribution of Aβ and clinical disease stage has not been consistently supported [12, 13]. In contrast, tau accumulation is more closely associated with the degree of neuronal loss as well as the clinical presentation of the disease, increasing cognitive deficits, and declining functions of daily living [12]. Autopsy findings have indicated stronger correlations between NFT number and pre-mortem cognitive performance than correlations between counts of Aβ plaques and cognition in those same individuals [14–16]. Across a substantial body of literature, tau pathology has also been shown to be more closely related to cognitive impairment, predominantly memory decline, than amyloid in postmortem studies [12, 17–19].

The advent of molecular imaging agents providing quantitative measures of Aβ and tau have allowed researchers to explore these proteins in AD in vivo. Therapeutic trials now routinely employ molecular imaging to screen prospective subjects as a component of enrollment criteria and to monitor treatment response. Several groups have reported progress in developing PET tracers for imaging tau deposition. These include a series of ¹⁸F-labelled arylquinoline derivatives [20] and ¹¹C-labelled phenyl/pyridinyl-butadienylbenzothiazoles/ benzothiazoliums [21]. A third group described the 5H-pyrio[4, 3-b] indole, 18FT807 (subsequently known as 18F-AV-1451 or flortaucipir) as a PET tracer for detection of tau pathology [22, 23]. Flortaucipir was reported to have high affinity and selectivity for PHF-tau, but not Aβ [22], and initial clinical trials [23] suggested greater uptake in AD and mild cognitively impaired subjects compared to controls. Favorable test-retest reproducibility of the tracer was also reported [24]. Recent studies have focused on comparing in vivo tracer distribution [25–29] to the pattern of NFTs seen in postmortem studies [4, 30], especially for individuals with positive Aβ scans [31]. The relationship between tau tracer positron emission tomography (PET) signal and cognitive performance has also been explored, though results have varied across studies, with most studies reporting substantial and significant correlations between neocortical standardized uptake value ratio (SUVr) and Mini-Mental State Examination (MMSE) [27, 33]. McSweeney et al. (2019) also found that cerebrospinal fluid (CSF)-pTau181 correlated with flortaucipir uptake [33]. Others have found no significant relationship between neocortex SUVr and cognition, but report associations between regional tracer retention and specific tasks [25] or with specific AD clinical phenotypes [34, 35].

The neuropathological literature and early PET studies suggest that the spatial location and density of tau may be indicative of the degree of neurodegeneration, synaptic dysfunction, and the character of cognitive deficits. We hypothesized the density and distribution of pathological tau measured by flortaucipir would correlate with cognitive impairment across a range of domains in a regionally distinct and domain specific manner. This study explores relationships between the spatial distribution of flortaucipir and a range of cognitive functions assessed by neuropsychological measures. We focused on amyloid positive subjects to gain insight into the role of tau within subjects whose biomarker profile is consistent with AD as a cause for cognitive impairment. It is also worth noting that flortaucipir was recently approved by the FDA for PET imaging of the brain to estimate the density and distribution of NFTs in adult patients with cognitive impairment who are being evaluated for AD.

MATERIAL AND METHODS

As part of the phase II open label multicenter study (NCT# 02016560 18F-AV1451-A05) evaluating the safety and imaging characteristics of flortaucipir, PET scans were administered to 222 cognitively healthy, mild cognitive impairment (MCI), and AD subjects at baseline and 18 months [31]. The phase II study was designed to examine relationships between flortaucipir tau imaging, florbetapir PET amyloid status, and baseline and longitudinal cognitive deterioration. Subjects were diagnosed based on NIA-AA diagnostic criteria for MCI and AD [36, 37], with criteria for age ≥50 and MMSE > 10. Clinical diagnoses were made during screening, prior to study procedures, based on NIA-AA diagnostic criteria for MCI or AD dementia Participants provided written informed consent and both informed consent and the protocol were approved by the relevant Institutional Review Boards. For this report, we analyzed baseline flortaucipir scans in the subset of 84 Aβ+ subjects, including healthy controls (n = 5), and those with clinical diagnoses of MCI (n = 47) or AD (n = 32). Amyloid negative subjects were not included in these analyses. Demographic data are provided in Table 1.

PET scans were acquired 80–100 min after the injection of approximately 370 MBq flortaucipir in four 5 min frames. Imaging was performed across 25 imaging centers within the United States. In addition to tau PET, florbetapir (two 5 min frames beginning 50 min after a 10 mCi dose,±60 days from the flortaucipir scan) and 3D T1 magnetic resonance imaging (MRI) (±30 days) were acquired for each participant. PET data were reconstructed with an iterative or row-action maximum likelihood algorithm with an image matrix size of 128×128 or 200×200, pixel size of 2–2.67 mm×2–2.67 mm, slice thickness of 2–4.25 mm, and post-reconstruction Gaussian filter of 3–5 mm or a relaxation parameter of normal or sharp filter.

Cognitive assessments included baseline MMSE, Alzheimer’s Disease Assessment Scale (ADAS-cog₁₁) (as well as its subscores), and a neuropsychological test battery. That battery included the Wechsler Memory Scale-Revised Logical Memory (WMS-R), Trail Making Test, Boston Naming Test, Digit Symbol Substitution Test (DSST), Animal List Generation, WMS-R Digit Span (Forwards and Backwards), American National Adult Reading Test, Clock Drawing Test, Judgment of Line Orientation, and WMS-R Logical Memory II (Delayed Recall). Function was assessed with the Functional Activities Questionnaire (FAQ).

RESULTS

Subjects were an average age of 73.8±9.4 years with mean MMSE and ADAS-cog scores appropriate for their clinical diagnoses; controls (MMSE = 30; ADAS = 6), MCI (MMSE = 27; ADAS = 11), and AD (MMSE = 21; ADAS = 21).

Correlations were seen between flortaucipir SUVr (in either MUBADA or precuneus) and MMSE, ADAS-Cog, and FAQ (Table 2). Correlations of tau PET with cognition/function ranged from 0.56 for MUBADA/MMSE to 0.28 for the precuneus/FAQ (all significant). For florbetapir, correlations ranged from 0.34 with ADAS to 0.17 with FAQ (not significant). Adjusting for each other when assessing correlations with cognition, both amyloid SUVr and MUBADA tau SUVr remained independently associated with cognition (models included baseline amyloid SUVr (Table 3, top row) or MUBADA tau SUVr (Table 3, bottom row)).

The top row of panels A and B of Fig. 1 show correlation maps of the relationship between flortaucipir SUVr distribution and ADAS-cog or MMSE scores, respectively, displayed such that all voxels with r≥0.3 are assigned a color. The bottom rows of these panels show the binarized VOI that results from identifying all voxels with r > 0.5. The correlation map for ADAS is dominated by the inferior lateral temporal lobes and is left lateralized. Precuneus and left lateral parietal regions are also involved. In contrast, the MMSE relationship is primarily focused in the right inferior temporal lobe, with a pattern of precuneus and parietal lobe similar to ADAS, though right rather than left dominant and generally weaker. Regression plots resulting from correlating the cognition VOIs for ADAS and MMSE to the test scores are shown in Fig. 2. No correlation between voxel-wise flortaucipir SUVr and FAQ was seen, even though there was a significant relationship seen for the MUBADA and precuneus VOIs. That is, despite a significant correlation between FAQ and SUVr in both a large VOI (MUBADA) and a smaller anatomic VOI (precuneus), at the voxel-wise level no meaningful pattern of correlation was observed, perhaps due to high individual voxel variability. Panel C of Fig. 1 shows the weighted voxels contributing to the MUBADA VOI. Note that these voxels best discriminate AD from healthy control based on a discriminant analysis [24, 44] and are similar to those seen for the correlation with ADAS.

OPEN IN VIEWER

Fig. 1

Spearman correlation maps of the relationship between flortaucipir distribution and ADAS-cog (A) and MMSE (B) scores overlaid on an atlas MRI (top rows for each measure) and the binarized VOI that results from identifying all voxels with ρ> 0.5 (bottom rows for each measure). The last row (C) shows the MUBADA VOI for comparison (color intensity reflects relative weighting of voxels within the region). Note, all images follow radiologic convention; correlations coefficient values are displayed on the MNI 152 brain atlas in 1 cm transaxial increments.

OPEN IN VIEWER

Fig. 2

Regression plots derived from binarized VOIs representing the relationship between test scores and flortaucipir SUVR. Correlation coefficients for the same analysis using the MUBADA VOI are provided parenthetically. These data are modeled to remove the effects of age.

Correlation patterns differed across the cognitive domains (more tau = worse performance). These pat-terns largely appeared to follow known spatial distributions of relevant functional neuroanatomy, though there were some exceptions. For tests of episodic memory (WMS-R: Logical Memory I for Immediate Recall; WMS-R: Logical Memory II for Del-ayed Recall and ADAS-Cog11 Word Recall Task for Delayed Recall) correlation images predominantly include temporal lobe, left side dominant, with the strongest signal in the area of the left amygdala and hippocampus (Fig. 3A). Correlations with left parietal and frontal regions are also evident, except for Delayed Recall which showed an exclusive bilate-rally-symmetric anterior mesial (amygdala and hippocampus) correlation.

For tests of attention/executive function (DSST, Trails A, B, B-A) frontal involvement is present, though additional areas are engaged reflecting the complicated nature of each test (Fig. 3B). The DSST displayed correlation within orbitofrontal, superior frontal, right posterior temporal and parietal cortices. Trails A demonstrated correlations spanning bilateral orbital frontal cortex as well as temporal-parietal regions (right > left). Trails B correlations with tau were limited to prefrontal cortex (orbital frontal primarily). The derived Trail B –Trail A score exhibited an exclusive relationship with the prefrontal cortex.

OPEN IN VIEWER

Fig. 3

Pearson correlation maps of relationships between flortaucipir distribution and cognitive measures related to episodic memory (A: WMS-R Logical Memory I Immediate Recall, ADAS Cog11 Word Recall and WMS-R Logical Memory II Delayed Recall), attention/executive function (B: DSST, Trails A, B, B-A) and spatial functions (C: Clock Draw –Copy, Ideational Praxis, Constructional Praxis). Color scale is as in Fig. 1.

For tests related to spatial function (Clock draw –Copy, Ideational Praxis, Constructional Praxis) correlations are dominated as expected by parietal foci. Clock Draw –Copy correlations are dominantly bilateral parietal (L > R; Fig. 3C). Ideational Praxis correlations demonstrated a locus in the right posterior parietal (intraparietal sulcus/angular gyrus) and posterior temporal (fusiform and middle temporal gyrus) regions. Constructional Praxis shared a similar focal relationship in the right posterior temporal parietal cortices with Ideational Praxis, though correlations were broader and extended inferiorly through the right temporal lobe.

All remaining cognitive correlation images are provided in the Supplementary Material (Supplementary Figures 1–3). Many of these tests assessed multiple cognitive domains, and similarly have complex correlation relationships with flortaucipir SUVr. Both Comprehension of Spoken Language and Remembering Test Instructions displayed correlations with tau in frontal lobe extending from superior to orbital regions. Comprehension showed extensive bilateral frontal involvement and small anterolateral temporal foci, while Remembering was right frontal dominant (Supplementary Figure 1). The Clock Draw - Command test elicited the highest correlation coeffici-ents among all tests, with a bilateral representation throughout the temporal-parietal areas and superior and orbital frontal involvement, sharing many of the same relationships exhibited by total ADAS and MMSE. Word Recognition and Digit Span showed similar relationships with flortaucipir uptake to those seen for test of episodic memory (Fig. 3A), with small left temporal lobe foci (antero-mesial for Word Recognition and postero-mesial for Digit Span (Supplementary Figure 2).

Boston Naming and Animal Generation correlated with flortaucipir SUVr in left inferior/anterior temporal lobe somewhat like Word Recognition, but more lateralized (Supplementary Figure 2). Objects and Fingers correlated with orbital frontal cortex (left dominant) and bilateral temporal pole. All three of these correlations are relatively weak. Judgement of Line Orientation showed extensive right temporal, parietal and frontal correlations. The rim of higher r values surrounding almost the entire brain, which are present exclusively in this correlation map, are unexplained. A modest left temporo-parietal pattern was seen for Adult National Reading Test, while little to no correlations between tau and scores were seen for Commands, Orientation and Word Finding Difficulty (Supplementary Figure 3).

DISCUSSION

These data suggest a relationship between the density and pattern of tau NFTs (as represented by flortaucipir SUVr) and cognitive impairment. Both a large, predominantly posterior neocortical VOI (MUBADA) and a small anatomically defined VOI (precuneus) demonstrated significant correlations with ADAS, MMSE, and FAQ. Note in particular the similarity between the ADAS correlation map and the weighted voxel distribution of MUBADA (Fig. 1). Given that MUBADA was designed to separate Aβ+ AD from other groups in a manner not dissimilar from the role of ADAS (designed to provide an objective assessment of cognitive status across domains relevant to AD), similarity in their distributions is reasonable. Similarly, Cho et al. reported that AD and MCI patients showed significant correlation between neocortical flortaucipir and Clinical Dementia Rating Scale Sum of Boxes (CDR-SB) and MMSE [32]. Maruyama et al. [21] found that neocortical [(¹¹)C]PBB3 (another putative PET tau tracer) retention was significantly correlated with MMSE, CDR, and CDR-SB. Results are consistent with neuropathology literature describing the role of tau distribution in disruption of specific cognitive domains [18, 48], and more generally indicates that NFTs predict MMSE score independently of neuron numbers and amyloid load [15, 49]. Further, Braak stage [30] correlates with cognitive status [50].

However, we also saw relationships between global amyloid burden (florbetapir) and cognition and function. Some relationship between the two biomarkers, and a shared relationship to cognitive impairment is reasonable. Relationships between levels of both Aβ and NFTs are well established [29, 51], and there is a general hypothesis that amyloid burden precedes and accelerates pathological tau [7]. We previously reported [31] in a larger cohort of both Aβ+ and Aβ- subjects (from which this sample is drawn) a significant correlation between flortaucipir and florbetapir SUVr. Similarly, Schöll et al. found increasing Aβ (PiB) with increasing flortaucipir-derived Braak stage [28] as well as an inverse rel-ationship between global cognition and tau that was related to cortical Aβ. Johnson et al. [27] also found that performance on MMSE and CDR-SB was related to flortaucipir SUVr in inferior temporal cortex and more weakly to Aβ (PiB) across in MCI and AD subjects. Generally, autopsy findings have indicated stronger within-subject correlations between pre-mortem cognitive performance and NFTs than with Aβ [14–16]. While it has been noted that Aβ is associated with cognitive decline, strong correlations are not generally reported between the quantity of cortical Aβ and cognition [12, 48]. It is also relevant that Pontecorvo et al. demonstrated that the tau signal overrides the amyloid signal with respect to cognitive variables when both are in the regression model [31]. Further, Ossenkoppele et al. [34] only found a weak relationship between flortaucipir SUVr and PiB DVR in a small sample of AD patients with varying clinical phenotypes. Cross-sectional studies [52–54] generally report robust inverse associations between amyloid and cognition (primarily memory) in healthy controls and MCI, but not in AD, perhaps because amyloid burden is thought to plateau beyond the MCI stage [7]. Limited relationships between amyloid burden and cognition are likely also in part due to the presence of non-amyloid associated pathologies in the clinically diagnosed cohorts that contribute cognitive impairment without amyloid.

This study focused on understanding the multi-do-main aspects of the relationship between the pattern of NFT deposition and cognition in AD. Voxel-wise correlation patterns between cognitive test scores and flortaucipir SUVr followed domain-specific neuroanatomical distribution (Fig. 3, Supplementary Figures 1–3). This is not unique to our study, though prior publications have not explored these relationships over this range of domains. For example, Oss-enkoppele et al. [34] found that worse performance on domain-specific neuropsychological tests was associated with greater flortaucipir uptake in accordance with predicted functional anatomy: memory with medial temporal lobes, visuospatial function with occipital, right temporo-parietal cortex and language with left-dominant temporal cortex. Brier et al. [25] reported significant associations between global and temporal lobe uptake and specific tasks in older controls (N = 36) and a small sample of AD subjects (N = 10). Temporal lobe SUVr was the strongest predictor of episodic memory, semantic memory and visuospatial processing, but not of working memory. They suggested temporal lobe tau was sufficient to predict cognitive impairment in specific domains across the early disease stages represented by their subjects. We identified spatial relationships between tau and cognitive performance for a large battery of cognitive and functional tests covering several cognitive domains. Note that while cognitive tests of a specific domain have a common core element (e.g., episodic memory recall), they also have different task requirements (e.g., simple word recall versus story recall that induces more frontal engagement). Our data seem to reflect both core components and other regions that might be recruited for specific aspects of each task (see below).

For episodic memory, we observed left lateral temporal lobe dominant correlations between tau and Immediate Recall, left mesial temporal lobe for Word Recall and focal, bilateral amygdala and hippocampus correlations for Delayed Recall (Fig. 3A). While the left temporal lobe was the focus of each of these relationships, the specific character of that relationship varied by cognitive assessment. Similarly, Cho et al. [32] found in their subset of Aβ+ subjects, SUVr in mesial temporal regions correlated with both verbal and visual memory function, while SUVr values in superior and middle temporal and posterior cingulate regions correlated with verbal memory. Scholl et al. [28] found that cross-sectional and longitudinal decline in episodic memory was strongly related to mesial temporal SUVr, furthering our understanding of the role of tau pathology in memory decline in aging.

For tests of attention/executive function, DSST and Trails A (Fig. 3B) correlations focused in bilateral orbitofrontal, posterior parietal and right posterior temporal regions. The right posterior temporal correlation, possibly driven by the working memory component of Trails A, may relate to results reported by Nickel et al. [55]. They found, in patients with mesial temporal lobe epilepsy, that right temporal hypometabolism (FDG PET) was associated with significantly impaired performance on the Halstead-Reitan Trail Making Test (similar to Trails A).

Trails B, on the other hand, demonstrated dominantly frontal lobe correlations, as expected given more complex cognitive processes than Trails A (set shifting, cognitive flexibility, and working memory) that are associated with prefrontal cortex function [56–58]. The correlation image for Trails B-A primarily localized to the left dorsolateral prefrontal cortex (Supplementary Figure 2) in a pattern somewhat overlapping the findings of Zakzanis et al. [59] and Moll et al. [60] on activation to Trails B-A tests adapted for functional MRI (fMRI) in healthy adults.

For tasks focused on spatial processing, parietal lobe correlations were observed. Clock Draw –Copy correlations were dominated by bilateral parietal regions (L > R; Fig. 3C). While the Clock Draw - Command condition requires numerous cognitive functions (as suggested by the widespread neocortical correlations, Supplementary Figure 1), the copy condition draws heavily on visuospatial functions [61, 62] and is used for assessing parietal lobe lesions. Thus, the parietal localization of the tau/test score correlation image is not unexpected. In normal subjects executing a closely related fMRI task, Trojano et al. [63] found bilateral posterior parietal lobe activation in regions similar to our findings, as did Ino et al. [64] in an fMRI study of clock drawing.

Ideational Praxis (assessment of visuospatial skills/perceptual orientation) correlated with right posterior parietal (intraparietal sulcus/angular gyrus) and posterior temporal (fusiform and middle temporal gyrus) regions (Fig. 3C). As with Clock Draw, Ideational Praxis/tau correlations demonstrated decreased performance with increasing tau in brain regions activated on functional imaging studies in healthy controls performing related tasks. For example, Creem-Regehr and Lee [65] used fMRI to image activation in healthy controls to imagining the grasping of tools (one component of the ideational praxis test –“imagine grasping an envelope”) and found foci of activation in essentially identical regions (posterior parietal and fusiform). Ruby et al. [66] used regional cerebral blood flow imaging to examine activation in healthy volunteers of script-based motor planning, again similar to the ideational praxis test. They observed bilateral parietal lobe activation near angular and right fusiform gyri, similar to the correlation map for Ideational Praxis, though we did not observe a left parietal focus.

Overall, our data are consistent with the relationship between tau pathology and cognition reported in the neuropathology literature, the lesion literature, and the available tau imaging literature. This consistency is encouraging, and further supports the role of tau in the progression of cognitive decline. Further, domain-specific patterns of regional correlations between elevated tau and poor performance is supported by studies using fMRI and regional cerebral blood flow imaging revealing where normal subjects activate cortex to successfully perform tasks. The comparisons are not exact, perhaps in part because the functional activation tasks are not direct duplicates of the tests in our cognitive and functional battery. Nonetheless, the similarities are striking and are summarized in Fig. 4.

OPEN IN VIEWER

Fig. 4

Relationship of representative correlation images with published literature regarding neuronal activation in healthy volunteers to tasks related to the tests used in the current study. Reprinted from Cognitive Brain Research [65] with permission from Elsevier; Cortex [66] with permission from Elsevier; Neuropsychologia [59] with permission from Elsevier; Arquivos de Neuro-psiquiatria [60] with permission from Arquivos de Neuro-psiquiatria; and Neuroscience Research [64] with permission from Elsevier.

Agreements with the literature on in vivo spatial relationships between tau and cognition are not universal. For example, we found that the MMSE rel-ationship was primarily focused in the right inferior temporal lobe, with additional signal in left fusiform/parahippocampus and bilateral parietal lobe and precuneus. In contrast, Ossenkoppele et al. [34] found that worse MMSE scores were associated with higher flortaucipir uptake in bilateral orbitofrontal cortex (left > right) and left anterior temporal cortex. Patient populations differ between the studies, which may ex-plain some of the differences in findings. Other examples of mismatches or mixed-matches between our findings and those in the literature were discussed above. While mismatches may relate to differences in subject populations, actual cognitive measures, or analytic approaches, our data are also from a relatively small sample and should be verified by further studies in larger samples.

Additional limitations of this study include re-latively small sample size and the lack of amyloid-negative subjects. Of course, amyloid-negative subjects were explicitly excluded from this sample in order to focus on AD-spectrum subjects. It would be valuable to explore the multi-domain nature of the impact of tau regional distribution in a larger sample, as well as in more subjects very early in the AD diagnostic spectrum. There is also a lack of ethnic diversity in this sample, so application to a more diverse population would be speculative.

These findings do suggest that the pattern of NFT pathology is related to the character of cognitive and functional impairment and might support the concept of assessing both baseline impairments and their longitudinal evolution at a single subject level. Specifically, treatment trials may want to consider whether broad measures, such as ADAS or MMSE, would be sensitive to potential treatment responses as effectively as subject-specific assessments that recognize the role of the tau pattern in dictating the cognitive domains most likely to evolve without intervention.

CONCLUSIONS

Voxel-wise SUVr values correlated with cognitive scores such that greater intensity of flortaucipir uptake was associated with worse performance in a domain-dependent spatial distribution. This supports the hypothesis that PET imaging of neuropathologic tau deposits may reflect underlying neurodegeneration in AD. However, it will be important to further characterize associations among tau, amyloid, and structural imaging biomarkers. Further work incorporating longitudinal studies will also be necessary to understand whether correlations reported herein relate to cognitive progression.