Tau pathology and relative cerebral blood flow are independently associated with cognition in Alzheimer’s disease

Patients were recruited from the Amsterdam Dementia Cohort of the Alzheimer Center Amsterdam [14]. All subjects underwent a standardized dementia screening, including medical and neurological examination, informant-based history, assessment of vital functions, screening laboratory tests, neuropsychological evaluation, MRI, lumbar puncture and/or amyloid-β positron emission tomography (PET), after which diagnoses were determined in a multidisciplinary consensus meeting [14]. For this study, patients with a diagnosis of Alzheimer’s disease (AD) dementia [15] or mild cognitive impairment (MCI) due to AD [16] were included. For all subjects, AD biomarkers in cerebrospinal fluid (CSF) and/or Aβ PET were abnormal (CSF Aβ42 < 813 pg/mL [17] and/or abnormal Aβ PET (on visual read)). According to the NIA-AA Research Framework [18], all subjects are considered in the AD pathophysiological continuum. Subjects were excluded if they had severe traumatic brain injury, abnormalities on MRI likely to interfere with segmentation of tau PET and participation in drug trial with a tau or Aβ-targeting agent.

All procedures were in accordance with the ethical standards of the Medical Ethics Review Committee of the Amsterdam UMC VU Medical Center and with the 1964 Helsinki Declaration and its later amendments or comparable ethical standards. Informed consent was obtained from all individual participants included in the study.

All participants underwent a single dynamic [¹⁸F]flortaucipir PET scan at the Amsterdam UMC VU Medical Center on an Ingenuity TF PET-CT scanner (Philips Medical Systems, Best, The Netherlands) within 1 year from their neuropsychological examination. [¹⁸F]Flortaucipir was synthesized at the Amsterdam UMC VU Medical Center, using a protocol described in detail previously [19]. The scan protocol started with a low-dose CT for attenuation correction, followed by a 234 ± 14 MBq [¹⁸F]flortaucipir bolus injection (injected mass 1 ± 1 μg). Simultaneously with tracer injection, a 60-min dynamic emission scan was initiated. After a 20-min break and following a second low-dose CT for attenuation correction, an additional dynamic emission scan was performed during the interval 80–130 min post-injection. During scanning, head movements were restricted by a head holder with band and head position was regularly checked. PET scans were reconstructed using a matrix size of 128 × 128 × 90 and a final voxel size of 2 × 2 × 2 mm³. All standard corrections for dead time, decay, attenuation, randoms and scatter were performed. Both scan sessions were co-registered into a single dataset of 29 frames (1 × 15, 3 × 5, 3 × 10, 4 × 60, 2 × 150, 2 × 300, 4 × 600 and 10 × 300 s), in which the last 10 frames belonged to the second PET session.

In addition, all subjects underwent structural MRI on a 3.0 Tesla (3 T) Philips medical systems’ Ingenuity TF PET-MRI. The protocol included an isotropic structural 3D T1-weighted image using a sagittal turbo gradient-echo sequence (1.00 mm³ isotropic voxels, repetition time = 7.9 ms, echo time = 4.5 ms, and flip angle = 8°), and a 3D fluid-attenuated inversion recovery (FLAIR) image (1.04 × 1.04 × 1.12 mm voxels, repetition time = 4800 ms, echo time = 278.8 ms, flip angle 90°).

Using Vinci software (Max Plank Institute, Cologne, Germany), T1-weighted MR images were co-registered to their individual PET scans in native space. To delineate cortical gray matter regions-of-interest (ROIs) on the co-registered MR images, the Hammers template [20] incorporated in PVElab software was used (which uses the default settings of SPM to define gray matter). To generate voxel-wise parametric images of non-displaceable binding potential (BP_ND) and R₁, receptor parametric mapping (RPM) [21] with cerebellar gray matter as reference region was applied to the dynamic 130 min PET data [22]. Our group previously demonstrated that, when compared to full kinetic modelling, RPM is the most optimal simplified parametric method for [¹⁸F]flortaucipir [23] with excellent test-retest repeatability [24]. PET images were partial volume-corrected using Van Cittert iterative deconvolution methods (IDM), combined with highly constrained back-projection (HYPR) [25]. A moving frame composite image was used for HYPR to better sustain the temporal information while denoising [26]. Uncorrected data are presented throughout the paper and partial volume-corrected data are presented in the Supplementary material.

For voxel-wise analyses, using Statistical Parametric Mapping (SPM) version 12 software (Wellcome Trust Center for Neuroimaging, University College London, UK), we warped all native space parametric BP_ND and R₁ images to Montreal Neurological Institute (MNI152) space, by using the transformation matrixes derived from warping the co-registered T1-weighted MRI scans to MNI space. Warped images underwent quality control for transformation errors.

For regional analyses, the following bilateral ROIs were created a priori based on the Hammers atlas [20] (in subject space): medial temporal (hippocampus, parahippocampal and ambient gyri, anterior temporal lobe medial part), lateral temporal (superior temporal gyrus, middle and inferior temporal gyri), parietal (inferolateral remainder of parietal lobe, superior parietal gyrus, gyrus cinguli posterior part), occipital (cuneus, lingual gyrus, lateral remainder of occipital lobe) and frontal (middle frontal gyrus, orbitofrontal gyri, superior frontal gyrus) regions.

As a measure of vascular pathology, white matter hyperintensities (WMHs) were visually rated by an experienced rater on subjects’ FLAIR image using the Fazekas scale, with scores ranging from 0 to 3 [27].

Cognitive domain scores were created by averaging Z-transformed test-scores (based on the current sample) of corresponding tests for memory (Immediate Recall of the Dutch version of the RAVLT, Delayed Recall of the Dutch version of the RAVLT and Visual Association Test-A), executive functioning (Stroop Colour Word test III, Phonemic Verbal Fluency (D-A-T), Digit Span Backwards and Trail Making Test (TMT)-B), language (Category Fluency Animals and Visual Association Test-Naming) and attention (TMT-A, Stroop Colour Word test I and II and the Digit Span Forward) [28]. Tests on which lower scores indicated better performance (TMT-A and -B, Stroop Colour Word test I, II and III) were inverted. Domain scores were only calculated if two or more tests within a domain were available.

To assess the correlations between [¹⁸F]flortaucipir BP_ND and R_1, with age, sex, education and Fazekas score, a correlation matrix was created using Spearman correlations. A p value below 0.05 was considered statistically significant.

To examine the regional associations between [¹⁸F]flortaucipir BP_ND and R_1, linear regression analyses, adjusted for age and sex, were performed. To assess the contribution of white matter damage in these associations, analyses were additionally adjusted for Fazekas score.

To assess voxel-wise associations between [¹⁸F]flortaucipir BP_ND or R₁ and cognition, voxel-wise regression analyses using SPM12 were performed. Analyses were adjusted for age, sex and education. A p value below 0.001 (uncorrected) was considered statistically significant for voxel-wise analyses. Additionally, a more conservative family-wise error (FWE) correction at p < 0.05 was applied.

To investigate regional associations between [¹⁸F]flortaucipir BP_ND or R₁ and cognition (dependent variables), linear regression analyses, adjusted for age, sex and education (model 1), were used. Subsequently, we entered [¹⁸F]flortaucipir BP_ND and R₁ simultaneously in the model to assess their independent associations with cognition (model 2).

A total of 71 subjects (MCI due to AD: n = 10, and AD dementia: n = 61) with a mean age of 66 ± 8 years and MMSE score of 23 ± 4 were included (Table 1). By study design, all subjects had abnormal amyloid biomarkers. [¹⁸F]Flortaucipir BP_ND values were highest in parietal (0.55 ± 0.43) regions and R₁ values were lowest in medial temporal regions (0.68 ± 0.06) (Table 1). [¹⁸F]Flortaucipir BP_ND and/or R₁ showed statistically significant correlations with age and education (Table 2), but not with sex.

Higher [¹⁸F]flortaucipir BP_ND was associated with lower R₁ within the lateral temporal (stβ − 0.32 [95%CI − 0.56 to − 0.08]), parietal (− 0.43 [− 0.72 to − 0.14]) and occipital (− 0.53 [− 0.78 to − 0.29]) ROI (Table 3). Higher BP_ND in the occipital ROI was also associated with lower R₁ in the parietal ROI (− 0.38 [− 0.64 to − 0.12]). All associations remained significant after FDR correction (Table 3). Figure 1 shows a selection of scatterplots for these associations. Addition of Fazekas scores to the model did not notably change the results (Supplementary Table 1).

Voxel-wise analyses (model 1) revealed that, in general, higher [¹⁸F]flortaucipir BP_ND was associated with worse cognition (Fig. 2). More specifically, higher medial temporal BP_ND associated with worse memory performance, and higher (orbito-)frontoparietal BP_ND with worse scores on executive functioning (Fig. 2). Higher inferior temporal BP_ND associated with worse language performance and higher (middle-)frontoparietal and occipital BP_ND with worse attention scores (Fig. 2). After FWE correction, sparse associations with higher BP_ND in the medial temporal regions and worse memory performance remained, as well as higher BP_ND in the temporal (fusiform cortex) regions and worse language scores (data not shown). Associations between higher BP_ND in the parietal and frontal regions and worse attention also survived FWE correction (data not shown).

Overall, lower R₁ associated with worse cognition (Fig. 2). In more detail, lower fronto-temporoparietal R₁ associated with worse scores on executive functioning and to a sparser extent with worse language performance (Fig. 2). Lower temporoparietal R₁ associated with worse attention scores (Fig. 2). None of the associations survived FWE correction (data not shown).

Regional linear regression analyses (model 1) revealed that higher medial temporal BP_ND was associated with worse memory performance (− 0.43 [− 0.66 to − 0.20]), higher lateral temporal BP_ND with worse scores on executive functioning (− 0.26 [− 0.52 to − 0.02]) and language (− 0.37 [− 0.66 to − 0.11]), and higher parietal BP_ND with worse executive functioning (− 0.46 [− 0.81 to − 0.23]), language (− 0.34 [− 0.76 to − 0.03]) and attention (− 0.50 [− 0.89 to − 0.25]) (Table 4; Fig. 3). Higher BP_ND in the occipital ROI was associated with worse memory (− 0.27 [− 0.54 to − 0.00]), executive functioning (− 0.26 [− 0.53 to − 0.01]), language (− 0.40 [− 0.73 to − 0.14]) and attention (− 0.37 [− 0.67 to − 0.10]) performance, and higher BP_ND in the frontal ROI with worse executive functioning (− 0.34 [− 0.65 to − 0.13]) and attention (− 0.33 [− 0.67 to − 0.08]). After FDR correction, the majority of significant associations remained (Table 4; Fig. 3).

Lower lateral temporal and parietal R₁ was associated with lower scores on executive functioning (0.27 [0.04 to 0.50]; 0.36 [0.15 to 0.60]), language (0.30 [0.04 to  0.57]; 0.28 [0.02 to 0.57]) and attention (0.33 [0.08 to 0.57]; 0.48 [0.26 to 0.71]) (Table 4; Fig. 3). Lower R₁ in the occipital ROI was associated with worse language (0.28 [0.03 to 0.57]) and attention (0.31 [0.07 to 0.59]) performance. By applying the FDR correction, the significant associations between lower R₁ in the parietal ROI and worse executive functioning remained, as well as the significant associations between lower R₁ in the lateral temporal, parietal and occipital ROI and worse attention scores (Table 4; Fig. 3). Scatterplots for a selection of these associations are presented in Fig. 1.

Finally, to examine the independent effects of tau pathology and rCBF on cognitive functioning, linear regression analyses including both [¹⁸F]flortaucipir BP_ND and R₁ were performed (model 2) (Table 4). Results revealed that higher medial temporal BP_ND was independently associated with worse memory (− 0.44 [− 0.67 to − 0.20]), higher lateral temporal BP_ND with worse language (− 0.31 [− 0.60 to − 0.04]) performance, and higher parietal BP_ND with worse scores on executive functioning (− 0.36 [− 0.70 to − 0.11]) and attention (− 0.36 [− 0.72 to − 0.10]). Higher occipital BP_ND was independently associated with worse language (− 0.33 [− 0.68 to − 0.03]) and higher frontal BP_ND with lower scores on executive functioning (− 0.33 [− 0.64 to − 0.11]) and attention (− 0.31 [− 0.66 to − 0.06]). Most significant associations survived FDR correction (Table 4). For [¹⁸F]flortaucipir R₁, lower values in the lateral temporal ROI were independently associated with worse attention (0.28 [0.03 to 0.54]), while low R₁ values in the parietal ROI were with lower scores on executive functioning (0.27 [0.06 to 0.50]) and attention (0.39 [0.17 to 0.62]). After FDR correction, the association between lower R₁ in the parietal ROI and worse attention remained (Table 4).

Overall, partial volume-corrected data yielded slightly higher values for both [¹⁸F]flortaucipir BP_ND and R₁ (Supplementary Table 2), but results from regression analyses remained essentially comparable (Supplementary Tables 3 & 4; Supplementary FIGs. 1 & 2).

An important finding in the present study is that tau pathology and rCBF, at least in part, independently contribute to cognitive deficits in AD. A previous study demonstrated that tau pathology was also independently associated with specific cognitive impairment in AD in the context of neurodegeneration [29]. This leads to the notion that tau pathology may impact cognitive performance directly, but also indirectly through a variety of mechanisms [29]. One such mechanisms might be rCBF, since some of the associations found between [¹⁸F]flortaucipir BP_ND and cognition in the present study disappeared when R₁ was included in the model simultaneously. Other factors possibly explaining the tau pathology-independent associations between rCBF and cognition in AD might be the presence of other down- or upstream pathological factors like tau-independent atrophy, vascular pathology or other proteinopathies. Vascular damage for example might lead to impaired rCBF, possibly causing an increase in amyloid-β accumulation, which in turn can lead to inflammation and neuronal dysfunction, leading to cognitive deficits [30]. Further research is required, however, to gain knowledge about the mechanisms explaining the tau-independent relationships between rCBF and cognition in AD.

Strong (regional) associations between tau pathology and cognitive deficits in AD have been established by multiple (imaging) studies [10‐12, 31], and results of the present study are generally in line with previous findings. As expected, tau pathology in the medial temporal regions showed strong associations with memory, while tau pathology in temporoparietal regions was associated with language. High levels of tau pathology in frontal regions were associated with more anteriorly based cognitive functions, like executive function and attention.

Although the association between CBF and cognition in AD has not been studied using this [¹⁸F]flortaucipir PET approach before, other studies investigated these associations by using [¹⁸F]FDG PET or MRI techniques (such as arterial spin labeling (ASL)) to measure (proxies of) CBF [13, 32]. These studies demonstrated that in AD, reduced CBF is generally associated with worse global cognition [13, 32] and not with domain-specific cognitive impairments (memory, executive function, language, attention and visuospatial functioning) [13]. Nonetheless, it was also demonstrated that most associations with cognition were found for low CBF in parietal and occipital regions, while least associations were found for temporal and frontal CBF [13]. This is in line with our results, although we also found multiple associations between cognition and low CBF in lateral temporal regions. This might be explained by the fact that the former study used a ROI covering the entire temporal cortex and did not differentiate between medial and lateral temporal regions. Both studies used subjects from the Amsterdam Dementia Cohort and used a similar approach to assess cognition, but another striking similarity between the former and present study is the range of regression coefficients for significant associations between CBF and cognition. Standardized regression coefficients ranged from 0.22 till 0.42 across the cognitive domains in the former study [13], and ranged from 0.27 till 0.48 in the present study, indicating comparable effect sizes.

Relative CBF is tightly correlated with measures of metabolic activity such as [¹⁸F]FDG PET [4, 5, 9]. Earlier studies investigating [¹⁸F]flortaucipir and [¹⁸F]FDG PET in AD found considerable overlap between higher levels of tau tracer uptake and lower levels of metabolic activity [11, 33], with moderate correlation coefficients across 30 predefined brain regions. The present study used R₁ as proxy for rCBF, and in line with the previously described study [33], we also found spatial overlap between high levels of tau pathology and low rCBF, with comparable standardized regression coefficients. The overlap of high levels of tau pathology and low levels of rCBF was in both studies not completely uniform across all brain regions, suggesting that both measures represent complementary aspects of AD pathology [33]. A potential explanation might be that tau pathology may develop prior to or even (partially) drive impaired metabolic activity or CBF, creating a time-lag between both pathological mechanism leading to topographical differences [34]. Alternatively, other pathological processes besides tau pathology may contribute to impaired metabolic activity or CBF, such as other proteinopathies. Vascular pathology has been linked to AD [35] and might have an impact on for example rCBF. However, in our study, the influence of vascular pathology showed to be negligible, since no correlation between Fazekas score and tau pathology or rCBF was found, and addition of Fazekas score to the regression model assessing the association between tau pathology and rCBF did not notably change results.

This study has several strengths, including the use of [¹⁸F]flortaucipir R₁ as a measure of rCBF, since this tracer has not been used in this context before, while [¹⁸F]flortaucipir currently is the most widely used tracer for tau pathology in AD. Another strength is that both measures were derived from a single dynamic [¹⁸F]flortaucipir PET scan, thereby circumventing the need for a dual-tracer study and avoiding bias caused by time-lags between measures of tau pathology and rCBF. Furthermore, analyses were repeated with partial volume-corrected data, and results remained essentially comparable; hence, we feel that our findings are not biased by atrophy to a large extent.

This study also has some limitations. The AD patients in this study were relatively young, which might hamper generalizability of results to older patient populations. Also, because our sample included only ten ‘MCI due to AD’ patients, further research is needed to elucidate potential differences in the BP_ND-R₁ relationship between diagnostic groups. Furthermore, it might be difficult to draw firm conclusions about the performance of [¹⁸F]flortaucipir R₁ compared to other measures of CBF due to the lack of a golden standard for measuring CBF. At last, this study has a cross-sectional design, which excluded the possibility to investigate whether the associations found between tau pathology, rCBF and cognition in AD represent causality. Therefore, longitudinal designs are required.