Differential associations between neocortical tau pathology and blood flow with cognitive deficits in early-onset vs late-onset Alzheimer’s disease

We included 79 subjects from the Amsterdam Dementia Cohort with either probable AD dementia [27] (n = 68) or mild cognitive impairment (MCI) due-to-AD [28] (n = 11), with positive amyloid-β biomarkers [29‐31]. All subjects underwent a standardized dementia screening, including medical history, extensive neuropsychological assessment, physical and neurological examination, lumbar puncture, blood tests, electroencephalography, and brain MRI. Diagnosis was established by consensus in a multidisciplinary meeting. The diagnosis of MCI/AD met core clinical criteria [27, 28] according to the National Institute on Aging and Alzheimer’s Association (NIA-AA) and were biomarker supported, with an AD-like CSF profile (i.e., tau/Aβ42 fraction > 0.52) [32, 33] and/or an Aβ-PET scan ([¹¹C]PiB or [¹⁸F]florbetaben) showing substantial amyloid accumulation when visually assessed by an experienced nuclear medicine physician (BvB). When both CSF and PET measures of amyloid were available, PET visual read was used to determine amyloid status. Subjects were assigned to either the early- or late-onset AD group, based on a median split (age-at-PET). Since the median age in our sample was 66 years, this led to a similar age cut-off compared to the conventional threshold of 65 years [2, 3, 6, 34]. The sample included eight patients meeting criteria for an atypical variant of AD, including two behavioral variant AD (bvAD) and six posterior cortical atrophy (PCA) patients [31, 35]. All atypical AD patients were part of the early-onset AD group, except one bvAD patient. Exclusion criteria for all participants were (1) dementia not due-to-AD, (2) significant cerebrovascular disease on MRI (e.g., major cerebrovascular accident), (3) major traumatic brain injury, (4) major psychiatric or neurological disorders other than AD, and (5) (recent) substance abuse. The study protocol was approved by the Medical Ethics Review Committee of the Amsterdam UMC, location VU Medical center. All patients provided written informed consent.

Acquisition and processing of [¹⁸F]flortaucipir PET images is described in detail elsewhere [17, 25, 36]. In short, dynamic 130-min [¹⁸F]flortaucipir PET scans were acquired on a Philips Ingenuity TF-64 PET/CT scanner. The scanning protocol consisted of two dynamic PET scans of 60 and 50 min, respectively, with an in-between 20-min break [36]. The first 60-min dynamic scan started simultaneously with a bolus injection of approximately 234 ± 14 MBq [¹⁸F]flortaucipir (injected mass 1.2 ± 0.9 μg). The second PET scan was co-registered to the first dynamic PET scan using Vinci software (Max Planck Institute, Germany). PET data were acquired in list mode and subsequently reconstructed using 3D-RAMLA including standard corrections for dead time, decay, attenuation, random, and scatter. Patients also underwent 3DT1-weighted MRI (Ingenuity TF PET/MR, Philips Medical Systems, The Netherlands) for anatomical and tissue segmentation purposes.

The 3DT1-weighted MRI images were co-registered to their corresponding PET images using Vinci software. Anatomical regions-of-interest according to the Hammers template were subsequently delineated on the MR images and superimposed on the PET scan using PVElab [37]. Using receptor parametric mapping (RPM) with cerebellar gray matter as reference region, we generated binding potential (BP_ND) maps as a measure for tau pathology. Additionally, and similar to previous work [17, 25], R₁ was generated as a proxy for rCBF [18]. Partial volume correction was applied to the PET images using Van Cittert iterative deconvolution methods (IDM), combined with highly constrained back-projections (HYPR) as described previously [17, 38, 39]. Uncorrected data are shown in the main manuscript and partial volume-corrected data are shown in the Supplementary material. In line with a previous study [17], we calculated BP_ND and R₁ in frontal, occipital, parietal, medial, and lateral temporal bilateral cortical lobar regions-of-interest.

In addition to the region-of-interest analyses, we performed voxel-wise analyses to (1) create average images of the different diagnostic groups (early- and late-onset AD without atypical AD cases, PCA, and bvAD), and (2) explore more fine-grained differences in [¹⁸F]flortaucipir BP_ND and R₁ between early- and late-onset AD. Therefore, we warped all native space parametric BP_ND and R₁ images to Montreal Neurological Institute (MNI152) space using the transformation matrixes derived from warping the co-registered T1-weighted MRI scans to MNI space using Statistical Parametric Mapping (SPM) version 12. All warped images were visually inspected for transformation errors and quality control. Average images were calculated using SPM12.

We included eight neuropsychological tests (< 1 year of PET scan), including the Dutch version of the Rey Auditory Verbal Learning Test immediate recall and delayed recall (episodic memory), Digit Span forward and backward (attention/executive functioning), Trail Making Test (TMT) version A and B (attention/executive functioning), Letter Fluency test (D-A-T) (executive functioning), and Category Fluency version animals (language) [40]. TMT-A and -B scores were inverted so that lower scores indicated worse performance. In addition, the Mini-Mental State Examination (MMSE) served as a measure of global cognition [41]. For participants who had TMT-A available but were missing TMT-B, we estimated the TMT-B by multiplying the time needed to complete the TMT-A with the mean TMT-B/A ratio from the respective diagnostic group in the respective cohort. To be included in the cognition analyses, participants were required to have at least 6 out of 9 cognitive tests available, and 72 (91%) participants met this criterion (30 early- and 42 late-onset AD). Among the excluded cases, there was one PCA case (early-onset AD group). sFigure-1 shows the missing values in the cognition-subsample for each neuropsychological test. There were on average 5% missing data in this subsample, for which we performed single imputation (with 5 iterations) for missing data, using all demographic (except for age and APOE4 status), neuropsychological, and neuroimaging variables as predictors [42, 43].

Differences in demographics and clinical characteristics between groups were assessed using independent sample t-tests for continuous variables and χ² for dichotomous data. Differences in neuropsychological test-scores were assessed using ANOVA, adjusting for sex and education.

To assess differences in [¹⁸F]flortaucipir BP_ND and R₁ between early- and late-onset AD groups, we used ANOVA with each of the lobar regions-of-interest (separately) as dependent variable. We repeated these analyses, but now excluding the atypical AD cases (n = 8), to investigate whether potential age-dependent results were not driven solely by these extreme phenotypes. To assess more fine-grained differences, we additionally performed voxel-wise contrasts in SPM12 for BP_ND or R₁ between early- and late-onset AD. All analyses were adjusted for sex. Results from voxel-wise analyses are displayed at both more liberal (i.e., p < 0.001, uncorrected) and more stringent (p < 0.05, family wise error (FWE) corrected) thresholds. Next, we investigated whether associations between BP_ND or R₁ and cognition differed between early- and late-onset AD. We performed linear regression analyses between BP_ND or R₁ and neuropsychological tests (separate models), adjusted for sex and education, including the interaction term “age-at-onset (dichotomous)*BP_ND or R₁”.

Results were considered significant if p-values were ≤ 0.05 for regional analyses. We considered a p-value ≤ 0.10 significant for interaction terms [44]. Additionally, adjustment for multiple comparisons was performed using the Benjamini–Hochberg false discovery rate (FDR) method (indicated by p_FDR) [45]. Statistical analyses were performed using R software, version 4.0.2, and the “mice” package was used for imputation.

Early-onset AD patients showed higher [¹⁸F]flortaucipir BP_ND in lateral temporal (BP_ND 0.56 ± 0.30 vs 0.42 ± 0.30, p = 0.045), parietal (BP_ND 0.84 ± 0.50 vs 0.33 ± 0.29, p < 0.001), occipital (BP_ND 0.65 ± 0.52 vs 0.29 ± 0.23, p < 0.001), and frontal cortex (BP_ND 0.40 ± 0.30 vs 0.16 ± 0.23, p < 0.001) compared to late-onset AD (Fig. 2).There were no differences in medial temporal BP_ND between early-onset AD (0.24 ± 0.14) and late-onset AD (0.25 ± 0.18, p = 0.895). Voxel-wise analyses confirmed higher BP_ND in widespread neocortical regions in early- vs late-onset AD (Fig. 2G). Effects were most pronounced in the precuneus and posterior cingulate, and frontotemporal cortex, as supported by the FWE-corrected results. There were no regions in which late-onset AD showed higher BP_ND compared to early-onset AD (Fig. 2G). Results remained essentially unchanged when PCA and bvAD patients were excluded from the analysis (sFig-3). Furthermore, results from partial volume corrected data yielded highly similar results, although BP_ND values were slightly higher (sFig-4).

By contrast, [¹⁸F]flortaucipir R₁ was lower in late-onset AD compared to early-onset AD in the medial temporal lobe (R₁ 0.66 ± 0.05 vs 0.69 ± 0.05, p < 0.05), but not in any of the neocortical regions (p > 0.05) (Fig. 3). Voxel-wise analyses showed lower R₁ in the (medial) temporal lobe (surviving FWE-correction), and subtly lower R₁ in the medial frontal cortex in late-onset compared to early-onset AD (Fig. 3G). In contrast, parieto-occipital regions showed lower R₁ in early-onset AD compared to late-onset AD, but this did not survive FWE-correction. Results remained essentially unchanged when atypical cases were excluded from the analysis (sFig-5). Results from partial volume corrected data yielded highly similar results, although R₁ values were slightly higher (sFig-6).

Linear regression analyses showed that, in general, higher [¹⁸F]flortaucipir BP_ND was strongly associated with worse scores on a variety of cognitive tests (Fig. 4A). Age-at-onset moderated these associations such that in early-onset AD, associations between lateral temporal, parietal, and occipital BP_ND with TMT-A, and parietal BP_ND with TMT-B were stronger than in late-onset AD (all p_interaction < 0.10: Fig. 4). In late-onset AD, associations between medial temporal BP_ND and Digit Span backward and Letter fluency test (D-A-T) were stronger than in early-onset AD (Fig. 4). Results from partial volume corrected data were essentially comparable (sFig-7).

In general, lower [¹⁸F]flortaucipir R₁ was associated with worse scores on multiple cognitive tests (Fig. 5). Age-at-onset moderated these associations such that in early-onset AD, associations were stronger than in late-onset AD (all p_interaction < 0.10: Fig. 5). More specifically, this was the case for the association between lateral temporal R₁ and Digit Span backward, TMT-A, TMT-B, and Letter fluency test (D-A-T), and the associations between occipito-parietal R₁ and TMT-A, TMT-B, and Category fluency Animals, as well as the association between occipital R₁ and MMSE (Fig. 5). Results from partial volume corrected data yielded some (test-specific) differences compared to the results from the uncorrected PET data (sFig-8).