Tau PET patterns mirror clinical and neuroanatomical variability in Alzheimer’s disease

See Sarazin et al. (doi:10.1093/brain/aww041) for a scientific commentary on this article.

The PET tracer [¹⁸F]-AV-1451 allows visualization of tau pathology in living subjects. Ossenkoppele et al. employ the tracer in patients with distinct Alzheimer's disease variants to investigate correlates of tau deposition. Pathological aggregation of tau, but not amyloid-β, is linked to patterns of neurodegeneration and clinical manifestations of Alzheimer’s disease.

See Sarazin et al. (doi:10.1093/brain/aww041) for a scientific commentary on this article.

The PET tracer [¹⁸F]-AV-1451 allows visualization of tau pathology in living subjects. Ossenkoppele et al. employ the tracer in patients with distinct Alzheimer's disease variants to investigate correlates of tau deposition. Pathological aggregation of tau, but not amyloid-β, is linked to patterns of neurodegeneration and clinical manifestations of Alzheimer’s disease.

Abstract

See Sarazin et al. (doi:10.1093/brain/aww041) for a scientific commentary on this article.

The advent of the positron emission tomography tracer ¹⁸F-AV1451 provides the unique opportunity to visualize the regional distribution of tau pathology in the living human brain. In this study, we tested the hypothesis that tau pathology is closely linked to symptomatology and patterns of glucose hypometabolism in Alzheimer’s disease, in contrast to the more diffuse distribution of amyloid-β pathology. We included 20 patients meeting criteria for probable Alzheimer’s disease dementia or mild cognitive impairment due to Alzheimer’s disease, presenting with a variety of clinical phenotypes, and 15 amyloid-β-negative cognitively normal individuals, who underwent ¹⁸F-AV1451 (tau), ¹¹C-PiB (amyloid-β) and ¹⁸F-FDG (glucose metabolism) positron emission tomography, apolipoprotein E (APOE) genotyping and neuropsychological testing. Voxel-wise contrasts against controls (at P < 0.05 family-wise error corrected) showed that ¹⁸F-AV1451 and ¹⁸F-FDG patterns in patients with posterior cortical atrophy (‘visual variant of Alzheimer’s disease’, n = 7) specifically targeted the clinically affected posterior brain regions, while ¹¹C-PiB bound diffusely throughout the neocortex. Patients with an amnestic-predominant presentation (n = 5) showed highest ¹⁸F-AV1451 retention in medial temporal and lateral temporoparietal regions. Patients with logopenic variant primary progressive aphasia (‘language variant of Alzheimer’s disease’, n = 5) demonstrated asymmetric left greater than right hemisphere ¹⁸F-AV1451 uptake in three of five patients. Across 30 FreeSurfer-defined regions of interest in 16 Alzheimer’s disease patients with all three positron emission tomography scans available, there was a strong negative association between ¹⁸F-AV1451 and ¹⁸F-FDG uptake (Pearson’s r = −0.49 ± 0.07, P < 0.001) and less pronounced positive associations between ¹¹C-PiB and ¹⁸F-FDG (Pearson’s r = 0.16 ± 0.09, P < 0.001) and ¹⁸F-AV1451 and ¹¹C-PiB (Pearson’s r = 0.18 ± 0.09, P < 0.001). Voxel-wise linear regressions thresholded at P < 0.05 (uncorrected) showed that, across all patients, younger age was associated with greater ¹⁸F-AV1451 uptake in wide regions of the neocortex, while older age was associated with increased ¹⁸F-AV1451 in the medial temporal lobe. APOE ϵ4 carriers showed greater temporal and parietal ¹⁸F-AV1451 uptake than non-carriers. Finally, worse performance on domain-specific neuropsychological tests was associated with greater ¹⁸F-AV1451 uptake in key regions implicated in memory (medial temporal lobes), visuospatial function (occipital, right temporoparietal cortex) and language (left > right temporoparietal cortex). In conclusion, tau imaging—contrary to amyloid-β imaging—shows a strong regional association with clinical and anatomical heterogeneity in Alzheimer’s disease. Although preliminary, these results are consistent with and expand upon findings from post-mortem, animal and cerebrospinal fluid studies, and suggest that the pathological aggregation of tau is closely linked to patterns of neurodegeneration and clinical manifestations of Alzheimer’s disease.

Introduction

The ability to detect fibrillar amyloid-β plaque depositions using ¹¹C-Pittsburgh compound B (PiB; Klunk et al., 2004) or fluorinated amyloid PET tracers (Vandenberghe et al., 2010; Barthel et al., 2011; Clark et al., 2011) has had a major impact on the field of Alzheimer’s disease. Among many other applications, amyloid-β imaging has been incorporated into diagnostic criteria for various stages of the disease (Albert et al., 2011; McKhann et al., 2011; Sperling et al., 2011; Dubois et al., 2014), has substantial impact on clinical decision-making (Ossenkoppele et al., 2013a; Sanchez-Juan et al., 2014) and patient management plans (Schipke et al., 2012; Grundman et al., 2013), and has shown potential as a surrogate outcome measure in clinical trials tailored to reduce cerebral amyloid-β plaque burden (Salloway et al., 2014; Liu et al., 2015). A yet unresolved issue after a decade of amyloid-β imaging research, however, is the disconnection between the diffuse distribution of amyloid-β pathology throughout the neocortex (Rabinovici et al., 2010; Wolk et al., 2012; Lehmann et al., 2013; Jung et al., 2015) and the selective patterns of brain atrophy and glucose hypometabolism that strongly correlate with clinical symptoms (Rabinovici et al., 2010; Ridgway et al., 2012; Lehmann et al., 2013; Madhavan et al., 2013). Thus, although amyloid-β deposition may be a prerequisite to developing Alzheimer’s disease dementia, at some point during the disease course additional factors are likely involved in determining regional neurodegeneration and symptomatology.

Tau, a microtubule-associated protein that aggregates into intracellular neurofibrillary tangles in Alzheimer’s disease (Weingarten et al., 1975; Braak and Braak, 1991), has often been suggested as a facilitator of the downstream effects of amyloid-β (Desikan et al., 2012; Jack and Holtzman, 2013). Tau pathology has a devastating effect on synaptic function (Gomez-Isla et al., 1997; Beharry et al., 2014; Spires-Jones and Hyman, 2014) and relates more strongly to cognitive functions during life than does amyloid-β (Arriagada et al., 1992; Nelson et al., 2012; van Rossum et al., 2012; Rolstad et al., 2013). However, current knowledge about relationships between tau pathology and other disease measures is mainly derived from animal, neuropathological and CSF studies. The recent advent of a PET tracer (i.e. ¹⁸F-AV1451, formerly called ¹⁸F-T807), which shows high affinity and selectivity for paired helical filament tau pathology in vitro (Chien et al., 2013; Xia et al., 2013; Marquie et al., 2015), now allows in vivo assessment of regional tau load.

Mapping the distribution of tau pathology may further our understanding of disease mechanisms in distinct clinical variants of Alzheimer’s disease. These clinical variants include posterior cortical atrophy (PCA, ‘visual variant of Alzheimer’s disease’; Crutch et al., 2012), logopenic variant primary progressive aphasia (logopenic variant PPA, ‘language variant of Alzheimer’s disease’; Gorno-Tempini et al., 2011), corticobasal syndrome (Alzheimer’s disease pathology is the causative pathology in up to 25% of cases; Dickson et al., 2002; Lee et al., 2011), and memory-predominant and behavioural/dysexecutive variants of Alzheimer’s disease (Ossenkoppele et al., 2015c). Age at onset has also been associated with clinical and anatomical variability; while patients with late-onset Alzheimer’s disease (≥65 years of age) typically present with amnestic symptoms and medial temporal lobe neurodegeneration, patients with early-onset (<65 years) tend to exhibit greater non-amnestic deficits and neocortical-predominant atrophy (Rabinovici et al., 2010; Mendez et al., 2012; Ossenkoppele et al., 2012; Smits et al., 2012; Moller et al., 2013; Cavedo et al., 2014). Furthermore, patients with Alzheimer’s disease who carry the apolipoprotein E (APOE) ϵ4 allele have shown greater medial temporal lobe vulnerability (Pievani et al., 2009; van der Flier et al., 2011; Ossenkoppele et al., 2013b; Lehmann et al., 2014) and post-mortem tau pathology (Beffert and Poirier, 1996; Tiraboschi et al., 2004) compared to non-carriers.

In a recent case study of a patient with PCA, ¹⁸F-AV1451 was found to be more closely linked to hypometabolism and cognitive deficits than was amyloid-β (Ossenkoppele et al., 2015d). Here we extend these findings in a larger sample and across multiple phenotypes, studying associations between tau pathology and factors that underlie heterogeneity in the manifestation of Alzheimer’s disease. We hypothesized that: (i) ¹⁸F-AV1451 would show distinct patterns in clinical variants of Alzheimer’s disease, in early-onset versus late-onset Alzheimer’s disease and in APOE ϵ4 carriers versus non-carriers; and (ii) the amount and distribution of ¹⁸F-AV1451 retention would correlate with neuropsychological performance.

Materials and methods

Participants

Twenty patients were recruited from the University of California San Francisco (UCSF) Alzheimer’s Disease Research Center. This represents a consecutive series of recruited research participants who met clinical criteria for probable Alzheimer’s disease dementia (McKhann et al., 2011) or mild cognitive impairment (Albert et al., 2011) due to Alzheimer’s disease, who underwent ¹⁸F-AV1451 PET scans between June 2014 and June 2015. All patients underwent a medical history and physical examination, a structured caregiver interview, brain MRI and neuropsychological testing. Seven patients met specific diagnostic criteria for PCA (Mendez et al., 2002), five for logopenic variant PPA (Gorno-Tempini et al., 2011), and one for corticobasal syndrome (Armstrong, 2014). Based on chart reviews and evaluation of the neuropsychological reports, the remaining patients were classified as amnestic (n = 5) and behavioural/dysexecutive variant Alzheimer’s disease (n = 1; Ossenkoppele et al., 2015c). One patient (59-year-old male) presented with a history of focal cortical deficits (i.e. agnosia, aphasia, apraxia) with relative preservation of memory function. This patient did not meet formal criteria for any of the Alzheimer’s disease variants, and will be referred to as ‘non-amnestic Alzheimer’s disease’. Nineteen (95%) patients had a visually positive PiB-PET scan as assessed according to previously published procedures (Rabinovici et al., 2010). Amyloid-β status was unknown for one patient with a clinical diagnosis of logopenic variant PPA. Exclusion criteria were: (i) meeting core clinical criteria for another dementia; (ii) clinically significant cerebrovascular disease; (iii) major systemic disease; (iv) history of a neurological disorder causing dementia; or (v) recent history of substance abuse.

In addition, 15 cognitively normal controls were recruited through the Berkeley Aging Cohort. The eligibility criteria for controls include a minimum age of 60, preventing a more accurate age matching of controls to the relatively young patients in the present study. Further eligibility criteria included normal performance on cognitive tests, absence of neurological or psychiatric illness that potentially affects brain structure and function (including white matter abnormalities) and lack of major medical illnesses and medications that affect cognition. To enhance contrasts between Alzheimer’s disease and normal ageing, we excluded 12 additional control studies with ¹⁸F-AV1451 and ¹¹C-PiB who met criteria for preclinical Alzheimer’s disease (Sperling et al., 2011) based on positive PiB-PET [distribution volume ratio (DVR) ≥ 1.08; Mormino et al., 2012]. Informed consent was obtained from all subjects or their assigned surrogate decision-makers, and UCSF, University of California Berkeley, and the Lawrence Berkeley National Laboratory (LBNL) institutional review boards for human research approved the study.

MRI

All patients and controls underwent MRI on a 3 T Siemens Tim Trio scanner or 1.5 T Magnetom Avanto scanner, respectively. Sequences included T₁-weighted magnetization prepared rapid gradient echo (MP-RAGE) for structure and fluid attenuation inversion recovery (FLAIR) to assess cerebrovascular disease [(Villeneuve et al., 2014; Ossenkoppele et al., 2015a), see specifications in Supplementary material]. MP-RAGE sequences were processed using FreeSurfer 5.1 (Fischl et al., 2002) to define native-space reference regions and cortical regions of interest.

Positron emission tomography

PET scans were performed at LBNL on a Siemens Biograph 6 Truepoint PET/CT scanner in 3D acquisition mode, and on a ECAT EXACT HR scanner for a small subset of ¹¹C-PiB (n = 3) and ¹⁸F-FDG (n = 3) PET scans in controls. A low-dose CT/transmission scan was performed for attenuation correction prior to all scans. ¹¹C-PiB and ¹⁸F-FDG PET scans were performed as previously described (Lehmann et al., 2013). ¹⁸F-AV1451 was synthesized and radiolabelled at LBNL’s Biomedical Isotope Facility. Approximately 10 mCi of ¹⁸F-AV1451 was injected intravenously, and subjects participated in one of two acquisition schemes: 0–100-min post-injection full dynamic scans (4 × 15, 8 × 30, 9 × 60 s, and 2 × 3, 16 × 5 min frames) followed by 120–150 min (6 × 5 min frames: 13 controls, 15 patients), or 75–115 min (8 × 5 min frames: two control subjects, five patients). PET data were reconstructed using an ordered subset expectation maximization algorithm with weighted attenuation. Images were smoothed with a 4 mm Gaussian kernel with scatter correction and evaluated prior to analysis for patient motion and adequacy of statistical counts. All PET scans were acquired within an average of 68 ± 109 days (¹⁸F-AV1451 to ¹⁸F-FDG: 105 ± 122 days, ¹⁸F-AV1451 to ¹¹C-PiB: 95 ± 121 days, ¹⁸F-FDG to ¹¹C-PiB: 0 ± 2 days).

Image analyses

PET images were co-registered to the subjects’ MP-RAGE using Statistical Parametric Mapping (SPM) version 8 (Wellcome Trust Centre for Neuroimaging, Institute of Neurology at University College London). Dynamic 90 min ¹¹C-PiB data were analysed using Logan graphical analysis with FreeSurfer-derived grey matter cerebellum as the reference region, yielding voxelwise DVR (Logan et al., 1996). One patient did not undergo ¹¹C-PiB PET, and two patients had no ¹¹C-PiB dynamic data available (but were positive on visual read of semiquantitative parametric images), yielding a total of 17/20 patients with ¹¹C-PiB DVR data. ¹⁸F-FDG PET images were summed, and standardized uptake value ratios (SUVR) were calculated for the 30–60-min post-injection interval using mean activity in the pons (manually edited from FreeSurfer-derived brainstem) as the reference region (Minoshima et al., 1995). Sixteen patients had ¹⁸F-FDG PET data available. Consistent with the initial report on human ¹⁸F-AV1451/¹⁸F-T807 PET (Chien et al., 2013), SUVR images at t = 80–100 minutes post-injection were created by normalizing summed activity from the realigned frames to mean activity in cerebellar grey matter, an area relatively spared of neurofibrillary tangle pathology even in advanced Alzheimer’s disease (Braak and Braak, 1991; Marquie et al., 2015). SUVR_80–100 showed the highest correlation across all time intervals with non-displaceable binding potentials derived from analyses with simplified reference tissue model 2 (SRTM2) in 23 controls and 15 patients with Alzheimer’s disease with dynamic data available (data not shown), and was therefore the interval of choice. We performed partial volume correction on all tracers using the Geometric Transfer Matrix approach (Rousset et al., 1998) and report both corrected and uncorrected data for region of interest analyses.

APOE status

Genotyping of the APOE allele (rs429358 and rs7412) was performed using a TaqMan® Allelic Discrimination Assay on an ABI 7900HT Fast Real-Time PCR system (Applied Biosystems) according to a previously described methodology (Coppola et al., 2012). APOE ϵ4 status was determined in 19/20 patients.

Neuropsychology

Within 1 year of the ¹⁸F-AV1451 PET scan (mean interval: 112 ± 83 days), 18/20 patients underwent a neuropsychological test battery that covered four major cognitive domains (Kramer et al., 2003; Lehmann et al., 2013; Ossenkoppele et al., 2015a): memory [immediate recall, delayed free recall, delayed cued recall and delayed recognition of the California Verbal Learning Test (9 items), and delayed recall of the Benson Figure], visuospatial or parietal function (copy of the Benson Figure, design fluency, calculations, face matching, and number location of the Visual Object and Space Perception battery), executive function [Modified Trail-making time and accuracy, Stroop interference (n correct and n errors), Digit Span backward, verbal fluency (D-words), and abstraction], and language [sentence repetition and comprehension, verbal agility, letter/word reading, picture vocabulary, category fluency (animals), and the Boston Naming Test]. Raw test scores were z-transformed using means and standard deviations (SD) of a group of 574 cognitively normal individuals [42.5% male, mean age: 66.1 ± 11.4, education: 17.2 ± 2.4, Mini-Mental State Examination (MMSE): 29.2 ± 1.1, all with a Clinical Dementia Rating (CDR) of 0 and a Geriatric Depression Scale (GDS) score ≤ 5] recruited at UCSF. Subsequently, these z-scores were averaged across all neuropsychological tests within a domain, resulting in composite scores for each cognitive domain.

Statistics

Baseline demographic and clinical features

Differences in baseline characteristics between groups were assessed using ANOVA with post hoc Bonferroni tests for continuous variables and χ² and Kruskal-Wallis with post hoc Mann-Whitney U-tests for dichotomous or categorical data.

Voxelwise contrasts between patients and controls

For voxelwise statistical analyses in SPM8, parametric PET images were spatially normalized to Montreal Neurological Institute space, followed by smoothing using an 8 mm isotropic Gaussian kernel. Then, voxelwise contrasts were performed to examine differences in ¹⁸F-AV1451, ¹¹C-PiB and ¹⁸F-FDG PET uptake in patients with PCA against healthy controls. Similar contrasts were performed between patients with logopenic variant PPA and amnestic Alzheimer’s disease against controls for ¹⁸F-AV1451 only, as ¹¹C-PiB and ¹⁸F-FDG were not available in all patients. The contrasts were thresholded at P < 0.05 family-wise error corrected, without adjustment for nuisance variables.

Region of interest analyses for ¹⁸F-AV1451, ¹¹C-PiB and ¹⁸F-FDG PET

Mean SUVR (¹⁸F-AV1451 and ¹⁸F-FDG) and DVR (¹¹C-PiB) values were extracted for FreeSurfer-defined frontal (superior and middle gyri and orbitofrontal cortex), lateral temporal (superior, middle and inferior gyri), lateral parietal (inferior, superior and suprmarginal gyri), medial parietal (posterior cingulate and precuneus), occipital (calcarine, cuneus and lateral occipital cortex), medial temporal (hippocampus, entorhinal cortex and parahippocampus), and basal ganglia (putamen, caudate and pallidum). These regions were selected as we aimed to assess the full spectrum of Braak stages (Braak and Braak, 1991) as well as ‘off-target’ tracer uptake in basal ganglia structures (Marquie et al., 2015; Johnson et al., 2016). We also calculated a hemispheric asymmetry index [AI, similar to Rabinovici et al. (2008) and Frings et al. (2015)] for each region of interest using the formula AI [%] = 200 × (R − L)/(R + L) for ¹⁸F-AV1451 and ¹¹C-PiB, or AI [%] = −200 × (R − L)/(R + L) for ¹⁸F-FDG. Thus, negative outcomes indicate more tau and amyloid-β, or more severe glucose hypometabolism in the left hemisphere, while positive values reflect right lateralized asymmetry. Differences in uptake values were assessed using ANOVA with post hoc Bonferroni tests.

Next, we performed Pearson correlations between the three PET tracers across 30 a priori selected left and right cortical [i.e. frontal, parietal, lateral temporal and occipital: based on Ossenkoppele et al. (2015d)] FreeSurfer regions of interest in native MRI space across 16 patients with Alzheimer’s disease with all tracers available. To control for interdependency resulting from multiple measurements (i.e. regions of interest) per subject, we additionally performed mixed effects models (i.e. a single model for each tracer comparison) assuming random intercepts and fixed slopes, using the maximum likelihood method for parameter estimation and likelihood ratio tests to assess significance.

Relationships between ¹⁸F-AV1451 with age, APOE and cognition

We conducted separate whole-brain voxelwise linear regressions between ¹⁸F-AV1451 uptake and age, APOE ϵ4 allele status (carrier or non-carrier), MMSE, and composite scores on four different cognitive domains (memory, visuospatial, executive and language functions) across all patients with Alzheimer’s disease. These analyses were thresholded at P < 0.05 uncorrected for multiple comparisons. In the model exploring effects of APOE ϵ4 allele status on voxelwise ¹⁸F-AV1451 uptake, analyses were performed with and without adjusting for global amyloid-β burden (measured as mean ¹¹C-PiB retention in prefrontal, lateral temporal, parietal, and cingulate cortices; Ossenkoppele et al., 2014). Other analyses were performed without adjustment for nuisance variables. Finally, we performed region of interest analyses to test associations between regional ¹⁸F-AV1451 uptake patterns and age, memory, visuospatial performance and language ability. Rank correlations (Spearman’s rho or Kendall’s τ in case of ties) were calculated between age and the ratio of hippocampal to cortical ¹⁸F-AV1451, memory score and hippocampal ¹⁸F-AV1451, visuospatial score and occipital ¹⁸F-AV1451, and language and left temporoparietal ¹⁸F-AV1451. We examined the regional specificity of these correlations by relating composite cognitive scores to ¹⁸F-AV1451 uptake in control regions of interest: memory score and cortical (i.e. weighted average of frontal, temporal, parietal and occipital cortex) ¹⁸F-AV1451, visuospatial score and cortical ¹⁸F-AV1451, and language score and right temporoparietal/whole cortex ¹⁸F-AV1451. A schematic overview and visual representations of the FreeSurfer-defined regions of interest are provided in Supplementary Table 1 and Supplementary Fig. 1, respectively.

Results

Patients

Demographic and clinical characteristics are presented in Table 1. The Alzheimer’s disease patients were on average relatively young (64 ± 8 years on average across phenotypes) and in a mild stage of the disease (mean MMSE: 21 ± 6). As expected, patients with PCA showed more severe impairment on visuospatial tasks than patients with logopenic variant PPA or an amnestic-predominant presentation (P < 0.05), while patients with logopenic variant PPA performed worse on language tests than the other two groups (P < 0.05).

Table 1.

Demographic and clinical characteristics according to diagnostic group

	Posterior cortical atrophy	Logopenic variant PPA	Amnestic Alzheimer’s disease	Non-amnestic Alzheimer’s disease	Behavioural/dysexecutive variant Alzheimer’s disease	Corticobasal syndrome	Controls
n	7	5	5	1	1	1	15
Age	63 ± 10	65 ± 10	67 ± 8	59	59	60	79 ± 7a
Sex (male/female)	4/3	1/4	3/2	1/0	1/0	0/1	6/9
Education (years)	15 ± 1	18 ± 2	17 ± 3	24	16	16	17 ± 1
Handedness (R/L/A)	6/1/0	3/2/0	4/0/0 (1 NA)	1/0/0	0/0/1	1/0/0	14/0/1
PiB status (+/−)	7/0	4/0 (1 NA)	5/0	1/0	1/0	1/0	0/15
PiB global DVR	1.86 ± 0.13	1.84 ± 0.38	1.79 ± 0.43	1.86	1.94	1.64	1.05 ± 0.03
APOE ϵ4 (+/−)	4/3	2/5	3/5	0/1	0/1	−	2/12
CDR	0.9 ± 0.2	0.6 ± 0.2	0.7 ± 0.3	0.5	1	−	−
MMSE	23 ± 4	18 ± 8	20 ± 5	27	21	16	29 ± 1a
Memory z-score	−2.5 ± 1.3	−2.6 ± 1.5	−3.4 ± 1.0	−1.5	−2.5	−	−
Visuospatial z-score	−5.9 ± 2.9b	−1.9 ± 1.9	−2.0 ± 1.4	−0.8	−1.4	−	−
Executive z-score	−2.9 ± 2.0	−3.7 ± 2.5	−1.9 ± 1.9	−0.8	−2.4	−	−
Language z-score	−2.1 ± 0.9	−4.0 ± 1.6c	−1.7 ± 1.4	−0.3	−0.3	−	−

Differences in baseline characteristics between groups (n > 1) were assessed using ANOVA with post hoc Bonferroni tests for continuous variables and χ² and Kruskal-Wallis with post hoc Mann-Whitney U-tests for dichotomous or categorical data.

^aControls > patients, P < 0.05.

^bPosterior cortical atrophy < Logopenic variant primary progressive aphasia and amnestic Alzheimer’s disease, P < 0.05.

^cLogopenic variant primary progressive aphasia < posterior cortical atrophy and amnestic Alzheimer’s disease, P < 0.05.

A = ambidextrous; CDR = Clinical Dementia Rating; R = right; L = left; NA = not available.

Control subjects

The healthy controls were substantially older (79 ± 7 years) than the Alzheimer’s disease patients and 14% carried an APOE ϵ4 allele. In line with a recent study (Johnson et al., 2016), visual inspection and regional quantification of ¹⁸F-AV1451 images (Table 2 and Supplementary Fig. 2) showed some degree of uptake in medial temporal lobes (i.e. entorhinal cortex, parahippocampal gyrus and hippocampus) in all normal subjects. According to the staging proposed by Braak and Braak (1991), this pattern extended into fusiform gyrus and inferior temporal cortex in several subjects. In addition, most controls showed tracer uptake in striatum, substantia nigra and choroid plexus thought to represent ‘off-target’ tracer binding unrelated to tau (Marquie et al., 2015).

Table 2.

¹⁸F-AV1451, ¹⁸F-FDG and ¹¹C-PiB retention values and asymmetry indices in regions of interest

	PCA		Logopenic variant PPA		Amnestic Alzheimer’s disease		Controls
	SUVR	AI (%)	SUVR	AI (%)	SUVR	AI (%)	SUVR	AI (%)
18F-AV1451
Frontal	1.57 ± 0.33	8.8 ± 14.9	1.76 ± 0.59	−11.8 ± 18.5	1.55 ± 0.39	−2.5 ± 14.0	1.07 ± 0.06d	0.4 ± 1.4
Lateral temporal	2.09 ± 0.47	8.8 ± 10.6	2.43 ± 0.47	−4.2 ± 10.0	2.05 ± 0.28	−8.4 ± 17.1	1.11 ± 0.05d	−0.1 ± 2.1
Lateral parietal	2.42 ± 0.50	3.6 ± 12.3	2.28 ± 0.74	−13.5 ± 14.5	2.03 ± 0.47	−5.8 ± 20.0	1.08 ± 0.08d	−0.4 ± 2.8
Medial parietal	2.47 ± 0.55a	9.7 ± 13.3	2.11 ± 0.48	−12.4 ± 15.5	1.93 ± 0.45	−4.8 ± 18.6	1.12 ± 0.09d	1.2 ± 2.1
Occipital	2.16 ± 0.43b	11.3 ± 8.7	1.65 ± 0.17	−7.5 ± 10.8	1.42 ± 0.31	1.3 ± 17.9	1.06 ± 0.06d	1.3 ± 2.0
Medial temporal	1.43 ± 0.20	1.9 ± 5.5	1.41 ± 0.25	−1.9 ± 7.1	1.70 ± 0.33c	−4.9 ± 7.3	1.18 ± 0.06d	−0.5 ± 4.6
Basal ganglia	1.33 ± 0.14	3.0 ± 4.2	1.21 ± 0.23	−3.3 ± 7.3	1.44 ± 0.11	−3.4 ± 7.9	1.47 ± 0.15e	−1.3 ± 2.6
	n = 7		n = 5		n = 5		n = 15
18F-FDG
Frontal	1.43 ± 0.06	3.0 ± 6.6	1.51 ± 0.03	−10.8 ± 6.2	1.40 ± 0.18	−6.6 ± 9.1	1.48 ± 0.08	−1.1 ± 2.2
Lateral temporal	1.13 ± 0.09f	11.3 ± 14.6	1.14 ± 0.20	−12.3 ± 0.8	1.18 ± 0.04	−9.0 ± 14.1	1.34 ± 0.12	0.3 ± 2.6h
Lateral parietal	1.07 ± 0.08f	16.3 ± 17.3	1.25 ± 0.15	−14.9 ± 1.5	1.32 ± 0.08	−6.4 ± 12.9	1.48 ± 0.14	0.2 ± 2.3h
Medial parietal	1.34 ± 0.06f	15.2 ± 13.4	1.35 ± 0.36	−18.6 ± 6.1	1.51 ± 0.14	−5.9 ± 10.0	1.65 ± 0.11	−3.0 ± 2.2i
Occipital	1.28 ± 0.23g	11.7 ± 17.4	1.72 ± 0.24	−6.5 ± 3.2	1.64 ± 0.10	−1.6 ± 6.2	1.54 ± 0.11	−2.0 ± 2.1j
Medial temporal	1.02 ± 0.09	5.8 ± 8.6	1.10 ± 0.03	−6.1 ± 4.3	1.03 ± 0.03	−7.5 ± 7.9	1.10 ± 0.06	−0.6 ± 2.9j
Basal ganglia	1.43 ± 0.04	1.7 ± 8.2	1.60 ± 0.17	−5.3 ± 2.5	1.46 ± 0.10	−5.0 ± 4.6	1.51 ± 0.13	0.1 ± 2.9
	n = 7		n = 2		n = 4		n = 13
	DVR	AI (%)	DVR	AI (%)	DVR	AI (%)	DVR	AI (%)
11C-PiB
Frontal	1.84 ± 0.16	1.6 ± 4.7	1.91 ± 0.50	−3.1 ± 6.8	1.61 ± 0.29	4.9 ± 5.6	1.00 ± 0.04k	0.4 ± 1.3
Lateral temporal	1.65 ± 0.16	−1.9 ± 6.1	1.73 ± 0.39	−0.1 ± 2.5	1.51 ± 0.26	3.8 ± 3.4	0.99 ± 0.02k	−1.0 ± 3.0
Lateral parietal	1.73 ± 0.12	−3.1 ± 5.2	1.85 ± 0.26	1.9 ± 3.8	1.61 ± 0.31	2.12 ± 4.5	1.04 ± 0.03k	−0.9 ± 2.4
Medial parietal	1.97 ± 0.15	−1.4 ± 2.6	2.04 ± 0.25	0.6 ± 9.1	1.82 ± 0.33	−0.1 ± 0.6	1.10 ± 0.04k	1.7 ± 3.7
Occipital	1.50 ± 0.14	2.8 ± 5.6l	1.58 ± 0.23	−5.2 ± 2.2	1.31 ± 0.17	3.4 ± 3.5	1.04 ± 0.03k	1.0 ± 3.4
Medial temporal	1.14 ± 0.08	−1.9 ± 6.2	1.21 ± 0.12	2.8 ± 5.0	1.10 ± 0.08	7.5 ± 3.1	0.98 ± 0.05k	0.8 ± 3.0
Basal ganglia	1.68 ± 0.16	1.3 ± 4.3	1.70 ± 0.31	−0.7 ± 2.7	1.60 ± 0.18	1.6 ± 4.5	1.23 ± 0.06k	−0.8 ± 1.8
	n = 7		n = 3		n = 4		n = 15

Data are presented as group means ± SD; see Fig. 3 for individual data. AI for ¹⁸F-AV1451 and ¹¹C-PiB were calculated using AI [%] = 200 × (R − L) / (R + L), and AI [%] = -200 × (R − L) / (R + L) for ¹⁸F-FDG. Negative values indicate greater AV1451 and PiB, and reduced FDG uptake in the left compared to the right hemisphere. Differences between groups for SUVR were assessed using ANOVA with post hoc Bonferroni tests at P < 0.05: ^aPCA > amnestic Alzheimer’s disease; ^bPCA > logopenic variant PPA and amnestic Alzheimer’s disease; ^cAmnestic Alzheimer’s disease > PCA and logopenic variant PPA; ^dControls < PCA, logopenic variant PPA and amnestic Alzheimer’s disease; ^eControls > logopenic variant PPA; ^fPCA < amnestic Alzheimer’s disease; ^gPCA < logopenic variant PPA and amnestic Alzheimer’s disease; ^hControls ≠ PCA, logopenic variant PPA and amnestic Alzheimer’s disease; ⁱControls ≠ PCA, logopenic variant PPA; ^jControls ≠ PCA; ^kControls < PCA, logopenic variant PPA and amnestic Alzheimer’s disease; ^lPCA ≠ logopenic variant PPA.

¹⁸F-AV1451 images of distinct Alzheimer’s disease variants

Figure 1 shows a selection of individuals with distinct phenotypes of Alzheimer’s disease illustrating the neuroanatomical correspondence between ¹⁸F-AV1451 uptake and the clinical presentation. ¹⁸F-AV1451 uptake was most prominent in clinically affected posterior regions in PCA (Fig. 1A), the language-dominant left hemisphere in logopenic variant PPA (Fig. 1B) and the basal temporal cortex and medial temporal lobe structures in amnestic-predominant Alzheimer’s disease (Fig. 1C). A patient presenting with dysexecutive, visuospatial and language complaints with preserved memory function showed ¹⁸F-AV1451 uptake in diffuse parts of the neocortical association cortex with relative sparing of the medial temporal lobes (Fig. 1D). A patient with the behavioural/dysexecutive variant of Alzheimer’s disease presenting with impaired organizational skills and judgement, inappropriate behaviour, diminished personal hygiene, hyperorality and newly acquired preference for sweet food, showed ¹⁸F-AV1451 uptake in classical temporoparietal regions and—in line with Ossenkoppele et al. (2015c)—relative sparing of the frontal cortex (Fig. 1E). Finally, a ¹¹C-PiB-positive patient with corticobasal syndrome (mainly affecting the left hemibody) who initially presented with cognitive impairment (i.e. memory and executive dysfunction) showed an asymmetric ¹⁸F-AV1451 pattern with greater uptake in the right hemisphere and unique involvement of primary sensorimotor cortex (Fig. 1F), which was spared in other patients.

Figure 1.

¹⁸F-AV1451 uptake in individuals with distinct phenotypes of Alzheimer’s disease. SUVR ¹⁸F-AV1451 images (neurological orientation) in (A) a 59-year-old female (MMSE: 28) with posterior cortical atrophy [note that this a different patient than presented in Ossenkoppele et al. (2015d)]; (B) a 77-year-old female (MMSE: 17) with logopenic variant PPA; (C) a 71-year-old female (MMSE: 23) with amnestic Alzheimer’s disease; (D) a 59-year-old female (MMSE: 27) with non-amnestic Alzheimer’s disease; (E) a 59-year-old male (MMSE: 21) with a behavioural presentation of Alzheimer’s disease, and (F) a 60-year-old female (MMSE: 16) with a corticobasal syndrome affecting the left hemibody.

Voxelwise contrasts between patients and control subjects

Next, we performed voxelwise contrasts for ¹⁸F-AV1451, ¹¹C-PiB and ¹⁸F-FDG between patients with posterior cortical atrophy and controls. In line with our hypothesis and a previous case study (Ossenkoppele et al., 2015d), patients with PCA showed substantial inverse spatial overlap between greater ¹⁸F-AV1451 and reduced ¹⁸F-FDG uptake, specifically affecting brain regions implicated in higher-order visual processing, particularly in the right hemisphere (Fig. 2A). This is further illustrated by overlaying the binarized ¹⁸F-AV1451 and ¹⁸F-FDG significance maps (Fig. 2D). Additionally, ¹⁸F-AV1451 uptake was elevated in several brain regions where ¹⁸F-FDG uptake did not differ from controls, including primary visual cortex and posterior cortex in the left hemisphere. Contrary to the regional specificity of ¹⁸F-AV1451 and ¹⁸F-FDG, ¹¹C-PiB retention was distributed throughout the neocortex and implicated both clinically affected and relatively spared regions (e.g. extensive binding in frontal cortex; Fig. 2A).

Figure 2.

Tracer uptake patterns in patients with different clinical Alzheimer’s disease variants. (A) Voxelwise contrasts, thresholded at P < 0.05 after family-wise error correction and without covariates, indicating regions in which patients with PCA had greater ¹⁸F-AV1451 (green), reduced ¹⁸F-FDG (red), and elevated ¹¹C-PiB (blue) compared to healthy controls. (B and C) Contrasts for ¹⁸F-AV1451 uptake between amnestic Alzheimer’s disease (B) and logopenic variant PPA (C) patients against healthy controls at P < 0.05 after family-wise error correction and without covariates. In panel D, we binarized the ¹⁸F-AV1451 and ¹⁸F-FDG (P < 0.05 family-wise error corrected) contrasts between patients with PCA and controls, and show in cyan regions where both ¹⁸F-AV1451 and ¹⁸F-FDG differed from controls, and in blue brain regions where only the ¹⁸F-AV1451 contrast was significant. In E, individual ¹⁸F-AV1451 images of five logopenic variant PPA patients are displayed in order of their MMSE scores.

In line with the clinical presentation, the ¹⁸F-AV1451 pattern in patients with an amnestic-predominant presentation most prominently affected the medial temporal lobes and lateral temporoparietal cortex (Fig. 2B). When assessing the parametric images and AIs of the five logopenic variant PPA cases individually, three patients showed the predicted left > right asymmetry, one was fairly symmetric, and one showed right > left ¹⁸F-AV1451 uptake (Table 2 and Fig. 2E). Notably, the two patients with most advanced disease severity showed the greatest ¹⁸F-AV1451 uptake in the right hemisphere and in anterior brain regions. On group-level voxelwise analyses, the ¹⁸F-AV1451 pattern in patients with logopenic variant PPA was rather symmetric, with uptake mainly observed in bilateral temporoparietal regions that are typically involved in language function (Fig. 2C). Voxelwise ¹¹C-PiB and ¹⁸F-FDG contrasts between amnestic Alzheimer’s disease and logopenic variant PPA patients against controls are shown in Supplementary Fig. 3. In general, the results are similar to those seen in patients with PCA (i.e. widespread neocortical ¹¹C-PiB binding, while ¹⁸F-FDG uptake is more region-specific), but should be interpreted with caution given the smaller sample size due to missing scans.

Region of interest analyses between ¹⁸F-AV1451, ¹¹C-PiB and ¹⁸F-FDG PET

Mean uptake values and AIs of ¹⁸F-AV1451, ¹¹C-PiB and ¹⁸F-FDG in eight regions of interest for each Alzheimer’s disease phenotype and the controls are shown in Table 2, Fig. 3, Supplementary Table 2 and Supplementary Fig. 4 (partial volume corrected data).

Figure 3.

¹⁸F-AV1451 uptake and hemispheric asymmetry in regions of interest. (A) ¹⁸F-AV1451 SUVR values for each Alzheimer’s disease patient and controls in seven bilateral regions of interest. (B) The degree of asymmetric tracer uptake within each region of interest as the percentage difference in SUVR value for left compared to right hemisphere: asymmetry index [%] = 200 × (R − L)/(R + L). Table 2 and Supplementary Table 2 show the group means and differences of ¹⁸F-AV1451 standardized uptake value ratios and asymmetry indices.

Bonferroni-corrected ANOVAs showed that patients with PCA had greater ¹⁸F-AV1451 retention in occipital and medial parietal (i.e. posterior cingulate and precuneus) regions of interest and reduced ¹⁸F-FDG uptake in occipital and lateral parietal regions compared to patients with amnestic and logopenic variant PPA (Table 2 and Fig. 3). Furthermore, patients with amnestic Alzheimer’s disease had increased ¹⁸F-AV1451 uptake in medial temporal lobe structures compared to the other Alzheimer’s disease groups. No differences between the Alzheimer’s disease phenotypes were found for ¹¹C-PiB. ¹⁸F-AV1451 and ¹⁸F-FDG AIs showed R > L asymmetry for PCA, L > R asymmetry for logopenic variant PPA and subtle L > R asymmetry for amnestic Alzheimer’s disease patients across regions of interest (Table 2). No consistent patterns were observed for ¹¹C-PiB. Partial volume correction of the PET data resulted in substantially greater tracer retention values but did not alter any of the differences described above.

Pearson correlation across the different Alzheimer’s disease phenotypes (n = 16 with all tracers available) in 30 FreeSurfer-defined regions of interest showed a strong association between increased ¹⁸F-AV1451 and reduced ¹⁸F-FDG uptake (Fig. 4A, Pearson’s r = −0.49 ± 0.07, P < 0.001). In addition, there were modest positive associations between ¹¹C-PiB DVR and ¹⁸F-FDG SUVR (Fig. 4B, Pearson’s r = 0.16 ± 0.09, P < 0.001) and between ¹⁸F-AV1451 SUVR and ¹¹C-PiB DVR (Fig. 4C, Pearson’s r = 0.18 ± 0.09, P < 0.001). These results were supported by mixed effects models testing relationships between each of the three PET tracers, holding region of interest constant as a fixed effect and subject as a random effect. Mixed effects models showed a significant negative relationship between ¹⁸F-AV1451 and ¹⁸F-FDG [β = −0.23 ± 0.01, χ²(1) = 201.4, P < 0.001], and supported positive associations with smaller effect sizes between ¹¹C-PiB and ¹⁸F-FDG [β = 0.29 ± 0.06, χ²(1) = 23.5, P < 0.001] and between ¹¹C-PiB and ¹⁸F-AV1451 [β = 0.78 ± 0.15, χ²(1) = 26.5, P < 0.001]. The relationships between PET tracers in distinct cortical regions (i.e. frontal, parietal, lateral temporal and occipital) are reported in Supplementary Table 3.

Figure 4.

Direct region of interest comparisons between PET tracers. Across 16 Alzheimer’s disease patients, regions with greater ¹⁸F-AV1451 uptake were strongly associated with lower ¹⁸F-FDG metabolism (A). Modest, positive associations were also observed between ¹¹C-PiB and ¹⁸F-FDG (B) and between ¹⁸F-AV1451 and ¹¹C-PiB (C).

Repeating the Pearson correlation analyses using partial volume corrected data, we found a similarly strong association between increased ¹⁸F-AV1451 and reduced ¹⁸F-FDG uptake (Supplementary Fig. 5A, Pearson’s r = −0.41 ± 0.07, P < 0.001). In contrast to uncorrected data, the modest positive association between ¹¹C-PiB DVR and ¹⁸F-FDG SUVR was no longer significant (Supplementary Fig. 5B, Pearson’s r = −0.01 ± 0.09, P = 0.88) and the association between ¹⁸F-AV1451 SUVR and ¹¹C-PiB DVR was somewhat stronger after applying partial volume correction (Supplementary Fig. 5C, Pearson’s r = 0.37 ± 0.08, P < 0.001). Mixed effects models of partial volume corrected data showed a significant negative relationship between ¹⁸F-AV1451 and ¹⁸F-FDG [β = −0.11 ± 0.01, χ²(1) = 70.0, P < 0.001], no relationship between ¹¹C-PiB and ¹⁸F-FDG [β = 0.02 ± 0.04, χ²(1) = 0.2, P = 0.642], and a significant positive relationship between ¹¹C-PiB and ¹⁸F-AV1451 [β = 1.52 ± 0.11, χ²(1) = 163.8, P < 0.001]. The relationships between partial volume corrected PET tracers in distinct cortical regions (i.e. frontal, parietal, lateral temporal and occipital) are reported in Supplementary Table 3.

Associations of ¹⁸F-AV1451 with age and APOE ϵ4 status

Whole-brain voxelwise linear regressions with ¹⁸F-AV1451 SUVR and age as continuous variables, thresholded at P < 0.05 uncorrected without covariates, showed that younger age was associated with greater ¹⁸F-AV1451 uptake in wide regions of the neocortex (Fig. 5A). Several frontal and medial parietal regions survived at P < 0.01 uncorrected (Supplementary Fig. 6A). The reverse contrast revealed that older age was associated with increased ¹⁸F-AV1451 uptake specifically in the medial temporal lobes (left > right). Consistent with our voxelwise results, region of interest analysis measuring the ratio of hippocampal-to-cortical ¹⁸F-AV1451 uptake showed a positive correlation with age [Fig. 5A, τ = 0.54, P = 0.001; Supplementary Fig. 7A (partial volume corrected), τ = 0.41, P = 0.012].

Figure 5.

Patterns of ¹⁸F-AV1451 retention in Alzheimer’s disease associated with age and APOE ϵ4 status. Result from voxelwise linear regression are displayed at P < 0.05 uncorrected for multiple comparisons, without covariates for age (A), and adjusted for global amyloid-β burden for APOE ϵ4 status (B).

Voxelwise analysis between APOE ϵ4 status (dichotomous: carriers, non-carriers) and ¹⁸F-AV1451 SUVR (continuous), controlling for global amyloid-β burden and thresholded at P < 0.05 uncorrected, showed that APOE ϵ4 carriers had greater ¹⁸F-AV1451 uptake in bilateral medial temporal and right temporoparietal cortex (Fig. 5B). Supplementary Fig. 6B shows that the right medial temporal lobe survived a slightly more stringent threshold of P < 0.01 uncorrected. There were no significant clusters with elevated ¹⁸F-AV1451 in APOE ϵ4 non-carriers. The results remained essentially the same when we did not adjust for global amyloid-β burden, and APOE ϵ4 carriers and non-carriers did not differ in age (63.6 ± 9.6 versus 65.2 ± 8.0 years, P = 0.35).

Associations of ¹⁸F-AV1451 with neuropsychological performance

We performed separate voxelwise linear regressions of MMSE and cognitive domain composite scores on ¹⁸F-AV1451 SUVR at a statistical threshold of P < 0.05 uncorrected for multiple comparisons and without covariates. Worse MMSE scores were associated with higher ¹⁸F-AV1451 uptake in bilateral orbitofrontal cortex (left > right) and left anterior temporal cortex (data not shown). Region of interest analysis showed no relationship between MMSE and whole cortex (i.e. weighted average of frontal, temporal, parietal and occipital cortex) ¹⁸F-AV1451 (τ = 0.10, P = 0.568, partial volume corrected: τ = −0.03, P = 0.849). Patients with lower memory composite scores had higher tracer retention in hippocampus and neighbouring medial temporal regions, with some involvement of inferior temporal cortex (Fig. 6A). Region of interest analyses showed a trend toward worse memory performance with increased ¹⁸F-AV1451 uptake in hippocampus (Spearman’s rho = −0.40, P = 0.106; partial volume corrected Spearman’s rho = −0.45, P = 0.063), while no relationship was observed between memory and whole cortical ¹⁸F-AV1451 (Fig. 6A, Spearman’s rho = 0.05, P = 0.844; partial volume corrected Spearman’s rho = −0.19, P = 0.456). Worse visuospatial performance correlated with increased ¹⁸F-AV1451 in bilateral occipital lobe extending into right temporoparietal regions (Fig. 6B). This was confirmed in region of interest analyses showing an association between worse visuospatial performance and elevated ¹⁸F-AV1451 in occipital cortex (Spearman’s rho = −0.52, P = 0.028; partial volume corrected Spearman’s rho = −0.51, P = 0.032), but not in whole cortex (Fig. 6B, Spearman’s rho = 0.11, P = 0.674; partial volume corrected Spearman’s rho = −0.06, P = 0.818). We did not observe a relationship between worse executive function and greater ¹⁸F-AV1451 uptake in any region. Finally, lower language composite scores were associated with elevated ¹⁸F-AV1451 uptake in left temporoparietal regions typically implicated in language function, along with smaller clusters in right anterior temporal and inferior frontal cortices (Fig. 6C). In region of interest analyses, worse language performance was correlated with increased ¹⁸F-AV1451 uptake in left temporoparietal cortex (Spearman’s rho = −0.53, P = 0.026; partial volume corrected Spearman’s rho = −0.62, P = 0.007) but not in right temporoparietal cortex (Spearman’s rho = −0.23, P = 0.366; partial volume corrected Spearman’s rho = −0.37, P = 0.126) or whole cortex (Spearman’s rho = −0.16, P = 0.53; partial volume corrected Spearman’s rho = −0.33, P = 0.185). Supplementary Fig. 6 C–E shows the regions that survived voxelwise linear regression at a more conservative threshold of P < 0.01 uncorrected, and Supplementary Fig. 7B–D shows the partial volume corrected region of interest analyses.

Figure 6.

Associations between regional ¹⁸F-AV1451 and cognitive performance. Voxelwise contrasts (top; P < 0.05 uncorrected) show associations between increased ¹⁸F-AV1451 retention and worse performance on memory (A), visuospatial (B), and language testing (C). Region of interest analyses (bottom) demonstrate that cognitive impairment was associated with focal rather than global increases in ¹⁸F-AV1451 in regions underlying specific cognitive domains.

Discussion

In this study, we used the novel PET tracer ¹⁸F-AV1451 in distinct phenotypes of Alzheimer’s disease to examine associations between tau pathology and neuroanatomical and clinical heterogeneity in vivo. We found that tau pathology, as measured by ¹⁸F-AV1451 PET, preferentially occupied brain areas that are critical for cognitive functions uniquely affected in distinct variants of Alzheimer’s disease, and strongly co-localized with hypometabolic regions. In contrast, amyloid-β pathology, as measured by ¹¹C-PiB PET, affected both clinically affected and unaffected regions and showed weak positive (without partial volume correction) or absent (with partial volume correction) associations with regional glucose metabolism. Older age was associated with greater tau pathology in medial temporal lobe structures, while younger patients had greater ¹⁸F-AV1451 uptake in the neocortical association cortex. APOE ϵ4 carriers had greater ¹⁸F-AV1451 uptake than non-carriers. Finally, increased ¹⁸F-AV1451 uptake was associated with worse performance on various cognitive domains in regionally specific patterns, in line with established brain-behaviour relationships. Taken together, these results are consistent with findings from post-mortem, animal and CSF studies, and provide preliminary evidence that the aggregation of tau is closely linked to patterns of neurodegeneration and the clinical manifestations of Alzheimer’s disease.

Distinct variants of Alzheimer’s disease

In previous studies of typical and atypical clinical presentations of Alzheimer’s disease, there has been a dissociation between the rather diffuse distribution of amyloid-β plaques and the highly specific patterns of neurodegeneration that strongly correlate with symptomatology (Rabinovici et al., 2010; Wolk et al., 2012; Lehmann et al., 2013; Jung et al., 2015). It has often been suggested that tau pathology, rather than amyloid-β, may be driving disease manifestation (Desikan et al., 2012; Jack and Holtzman, 2013). Tau imaging with ¹⁸F-AV1451 in conjunction with amyloid-β imaging now enables detailed characterization of the two hallmark Alzheimer’s disease pathologies in vivo. Visual inspection of individual images (Fig. 1) and formal group-level voxelwise analyses (Fig. 2) revealed that tau selectively targeted brain regions that are clinically affected, while amyloid-β was distributed throughout the association cortex (Table 2). For example, ¹⁸F-AV1451 was predominantly retained in occipital, parietal and occipitotemporal cortices in patients with PCA, clinically characterized by deficits in primary and higher-order visual processing. Furthermore, patients with an amnestic-predominant presentation showed prominent medial temporal lobe involvement, the patient with a behavioural/dysexecutive presentation had a classical temporoparietal pattern of ¹⁸F-AV1451 uptake with relative sparing of the frontal cortex, in line with the atrophy pattern shown in patients with this phenotype (Ossenkoppele et al., 2015c), and the patient with corticobasal syndrome uniquely showed involvement of the peri-rolandic cortex and had greater ¹⁸F-AV1451 contralateral to the affected body side. Given the positive ¹¹C-PiB scan, the patient with corticobasal syndrome most likely has Alzheimer’s disease, rather than corticobasal degeneration (a 4-repeat tauopathy), as the causal aetiology. Age-related comorbid amyloid-β pathology is unusual at the patient’s age (16–18% across cognitively normal persons aged 60; Jack et al., 2014; Jansen et al., 2015) and Alzheimer’s disease pathology may be a more frequent cause of corticobasal syndrome in young patients (Ossenkoppele et al., 2015b). Furthermore, glucose hypometabolism was more prominent in temporoparietal than in frontal cortex, which has been associated with underlying Alzheimer’s disease pathology in patients with clinical corticobasal syndrome (Sha et al., 2015). Finally, ¹⁸F-AV1451 uptake in this patient fell within the Alzheimer’s disease range, which is well above the levels we have observed in other patients with suspected corticobasal degeneration (data not shown) and is in line with a recent post-mortem binding study showing that ¹⁸F-AV1451 binds to paired helical filaments of tau found in Alzheimer’s disease, but not to straight tau filaments found in corticobasal degeneration and other 4-repeat tauopathies (Marquie et al., 2015).

Based on previous neuroimaging studies (Rabinovici et al., 2008; Madhavan et al., 2013; Teichmann et al., 2013; Rogalski et al., 2014; Ossenkoppele et al., 2015a) and the specific language presentation, we hypothesized that patients with logopenic variant PPA would show ¹⁸F-AV1451 uptake in temporoparietal language regions in a clearly asymmetric fashion with greater tau in the left hemisphere compared to right hemisphere. Visual inspection of individual images and a quantitative laterality index indicated that three logopenic variant PPA showed the predicted left > right pattern, one was fairly symmetric and one patient had greater ¹⁸F-AV1451 uptake in right than in left hemisphere. This is in line with a neuropathological study of primary progressive aphasia due to underlying Alzheimer’s disease (Gefen et al., 2012) showing that, while most patients showed higher neurofibrillary pathology in the left hemisphere, a significant minority (14–29% depending on the brain region) had greater tangle pathology in the right hemisphere. At group-level voxelwise analyses, however, the ¹⁸F-AV1451 uptake pattern appeared to be fairly symmetrical, largely covering bilateral temporoparietal regions. This finding might be due to the small and heterogeneous sample comprising a wide range of demographic and clinical features. Notably, the two patients with lowest MMSE scores had greater involvement of the right hemisphere and frontal ¹⁸F-AV1451 uptake, potentially indicative of tau pathology selectively targeting the left hemisphere in early disease stages and subsequently spreading anteriorly in the left hemisphere and into inter-connected contralateral brain regions with advancing disease (Ridgway et al., 2012; Rohrer et al., 2013; Rogalski et al., 2014; Ossenkoppele et al., 2015a).

Age and APOE ϵ4

Consistent with previous studies showing selective vulnerability of the medial temporal lobes in late-onset Alzheimer’s disease and greater cortical neurodegeneration in early-onset Alzheimer’s disease (Rabinovici et al., 2010; Mendez et al., 2012; Smits et al., 2012; Moller et al., 2013; Cavedo et al., 2014), we found that older age was associated with greater ¹⁸F-AV1451 uptake in bilateral medial temporal structures, whereas younger age was associated with elevated ¹⁸F-AV1451 in occipital, parietal and frontal regions. These different uptake patterns as a function of age highly correspond to the greater overall and neocortical-predominant neurofibrillary tangle burden observed in patients with early-onset Alzheimer’s disease at autopsy (Nochlin et al., 1993; Bigio et al., 2002; Marshall et al., 2007). Moreover, we found that APOE ϵ4 carriers had greater ¹⁸F-AV1451 uptake in medial and lateral temporal and parietal lobes, even after accounting for global amyloid-β burden. This is consistent with some (Beffert and Poirier, 1996; Tiraboschi et al., 2004) but not all (Schmechel et al., 1993; Greenberg et al., 1995) post-mortem studies, and in line with a study showing that APOE ϵ4 directly facilitates phosphorylation of tau resulting in increased filamentous tau load (Harris et al., 2003). Taken together, the present study indicates that factors often linked with heterogeneity in the clinical manifestation of Alzheimer’s disease (i.e. age-at-onset and APOE ϵ4 status) are associated with distinct patterns of tau pathology. Our results are also consistent with the neuropathological observation of ‘hippocampal-sparing’ Alzheimer’s disease, where greater tangle burden in cortical versus medial temporal regions is observed in a subset of patients, and is correlated with early age-of-onset, non-amnestic clinical presentations, and lower prevalence of the APOE ϵ4 allele (Murray et al., 2011).

Neuropsychological performance

Expanding upon autopsy and CSF studies describing global associations between tau pathology and cognition (Arriagada et al., 1992; Nelson et al., 2012; van Rossum et al., 2012), we found regionally specific relationships between greater ¹⁸F-AV1451 and reduced memory, visuospatial and language functions. In addition, lower MMSE scores were associated with greater ¹⁸F-AV1451 uptake in orbitofrontal and anterior temporal cortices. While these may not be considered core Alzheimer’s disease regions, they could underscore an anterior spread of tau with advancing disease that coincides with cognitive decline. We did not observe any regional associations between ¹⁸F-AV1451 and poor executive function scores, which could be due to difficulty isolating executive function deficits from other cognitive deficits (i.e. visuospatial, language and memory) that may have affected performance on executive function tasks. Similarly, the bilateral ¹⁸F-AV1451 pattern associated with worse language performance could be associated with transneuronal spread of tau (Clavaguera et al., 2009; de Calignon et al., 2012; Raj et al., 2012; Zhou et al., 2012) from left hemisphere to right, or with a selective vulnerability of language regions (Rogalski et al., 2008; Miller et al., 2013) to the toxic effects of pathological tau.

Partial volume correction

To assess for confounding effects of atrophy on PET quantification we performed partial volume correction. This yielded substantially higher ¹⁸F-AV1451 uptake values in patients, while uptake values remained low by comparison in controls. Additionally, regional asymmetry of ¹⁸F-AV1451 uptake was increased after partial volume correction in PCA (right > left) and logopenic variant PPA (left > right) patients but remained equally symmetric in controls. In comparisons between PET tracers in patients, we observed a strong, negative relationship between ¹⁸F-AV1451 and ¹⁸F-FDG that survived partial volume correction. Furthermore, the association between ¹⁸F-AV1451 and ¹¹C-PiB was strengthened after partial volume correction. In contrast, the mildly positive relationship that we initially observed between ¹¹C-PiB and ¹⁸F-FDG disappeared after partial volume correction, suggesting this association was largely driven by partial volume effects. Correlations between ¹⁸F-AV1451 uptake patterns and neuropsychological performance in patients were similar with and without partial volume correction.

Strengths and limitations

A main strength of the present study is the inclusion of a variety of Alzheimer’s disease phenotypes, which allowed us to study relationships between tau pathology and neuroanatomical and clinical heterogeneity. There are also some limitations. First, this is a preliminary study with a relatively small sample size. Future studies including more patients and a better balance across Alzheimer’s disease variants are needed. Second, due to the small sample size, some voxelwise analyses (mainly the associations between tau pathology with age, APOE ϵ4 status and cognition) were conducted at fairly liberal thresholds and should be interpreted with caution, although the relative lack of noise (i.e. focal regional correlations opposed to scattered small clusters) and the convergence of region of interest analyses are encouraging. Third, some patients were missing ¹¹C-PiB and ¹⁸F-FDG data (Table 2 and Supplementary Fig. 3), so we could only perform a fair comparison across tracers in patients with PCA (Fig. 2). Fourth, we did not assess relationships between molecular markers and brain atrophy because T₁-weighted MRI was acquired at different scanners for patients and controls. Finally, as the healthy controls were significantly older than our relatively young Alzheimer’s disease patients, we did not adjust our voxelwise contrasts (Fig. 2) to avoid collinearity between age and ¹⁸F-AV1451 uptake.

Future directions

Tau imaging will help address fundamental questions about the mechanisms that drive clinical, anatomical and functional diversity in Alzheimer’s disease. Key topics for future research include the further study of cross-sectional and longitudinal relationships between tau, amyloid-β and neurodegeneration, and how each of these factors contributes to clinical progression. Understanding which protein is the major force driving neurodegeneration and cognitive decline is highly relevant for developing and testing effective therapies, and the advent of novel tau PET tracers (Villemagne et al., 2015) will for the first time allow in-depth testing of the entire amyloid cascade hypothesis (Hardy, 2009) in vivo.

Glossary

Abbreviations

AI = 

asymmetry index

DVR =

distribution volume ratio

MMSE =

Mini-Mental State Examination

PCA =

posterior cortical atrophy

PiB =

Pittsburgh compound B

PPA =

primary progressive aphasia

SUVR =

standardized uptake value ratio

Funding

This research was funded by Marie Curie FP7 International Outgoing Fellowship (628812; to R.O.); The donors of (Alzheimer’s Disease Research), a program of BrightFocus Foundation (to R.O.); Tau Consortium (to G.D.R. and W.J.J); National Institute on Aging grants (R01-AG045611; to G.D.R.), (R01-AG034570; to W.J.J.), (P50-AG023501; to B.L.M.), and (K23-AG038357; to K.A.V.); John Douglas French Alzheimer’s Foundation (to G.D.R. and B.L.M.); State of California Department of Health Services Alzheimer’s Disease Research Centre of California grant (04-33516; to B.L.M); Swedish Medical Association and the Swedish Foundation of Nuclear Medicine (to M.S.). Avid Radiopharmaceuticals enabled use of the ¹⁸F-AV1451 tracer, but did not provide direct funding and was not involved in data analysis or interpretation.

Supplementary material

Supplementary material is available at Brain online.