Study population
The 33 enrolled individuals included nine patients diagnosed with AD (57–75 years old), 13 with mild cognitive impairment (MCI) (59–79 years old), four age-matched elderly healthy controls (eHC; 58–71 years old) and five young healthy controls (yHC; 20–30 years old). Two patients with non-AD dementia were also recruited, in an exploratory manner, including one with a clinical diagnosis of corticobasal degeneration and one with progressive supranuclear palsy. All patients were referred to the Memory Clinic at the Department of Geriatric Medicine, Karolinska University Hospital, Stockholm, Sweden, and underwent thorough clinical investigation including medical history, physical examination, laboratory blood tests, Apolipoprotein E (ApoE) genotyping, neuropsychological assessment, cerebrospinal fluid sampling and MRI scans. The diagnosis was based on the consensus of a committee, which included geriatricians, neurologists, clinical neuropsychologists, and specialist nurses.
All AD patients fulfilled the National Institute of Neurological and Communication Disorders, Alzheimer’s Disease and Related Disorders Association criteria [
12] and the DSM-IV criteria for Dementia of the Alzheimer’s type [
13], while all MCI patients met the Petersen criteria [
14]. Eleven MCI patients were classified as amnestic single-domain, and two as amnestic multi-domain [
15]. All amnestic single-domain MCI patients were classified as MCI PIB-positive (
n = 11) based on their amyloid status, as determined by their [
11C]Pittsburgh compound B (PIB) PET scan (see the “[
11C]PIB- and [
18F]FDG-PET image pre-processing” section below), while the amnestic multi-domain MCI patients were classified as MCI PIB-negative (
n = 2). For the purposes of this project, all MCI PIB-positive patients were reclassified into a prodromal AD group in accordance with the new research diagnostic criteria [
16]. Similarly, all PIB-positive patients with clinically diagnosed AD were reclassified into the AD dementia group. This resulted in eleven patients with prodromal AD, nine with AD dementia and two with MCI PIB-negative.
The two patients with non-AD dementia were clinically diagnosed as follows: one patient with predominantly left-sided signs of atypical parkinsonism, progressive apraxia and executive deficits fulfilled the criteria for possible corticobasal degeneration syndrome [
17], and one patient with progressive postural instability and falls, vertical supranuclear opthalmoparesis, dysdiadochokinesia and executive deficits fulfilled the criteria for probable progressive supranuclear palsy [
18].
Nine individuals (five yHC and four eHC) were recruited as healthy controls either via Clinical Trial Consultants AB (Uppsala University Hospital, Uppsala, Sweden) or from patients’ familial circles, after extensive clinical evaluation. They were included in the absence of cognitive complaint, prior head injury, or known neurologic/psychiatric disorder. They were all non-smokers and free from medication.
Neuropsychological assessment
All individuals completed a large battery of neuropsychological tests. Individual performances (except mini mental state examination (MMSE) performance) are expressed as z-scores, from comparison with a reference group of healthy controls [
19].
Global cognition was assessed as the average performance in nine tests, including five subtests from the Weschler Adult Intelligence Scale (i.e. Similarities, Information, Block Design, Digit Span, and Digit Symbol), the Corsi test, the Trail-Making Tests A and B, and the Rey-Osterrieth complex figure copy test. Episodic memory was assessed as the average performance in the following three tests: the Rey-Auditory Verbal Learning learning and 30 min delayed-retention tests, and the Rey-Osterrieth complex figure retention subtest.
Image acquisition
A PET scan using [18F]THK5317 and a 3D-T1-weighted MRI sequence were acquired for all participants. Additionally, all participants except the yHC group underwent a [11C]PIB-PET scan, and all patients (i.e. those with prodromal AD, AD dementia, MCI PIB-negative, and non-AD dementia) underwent an [18F]FDG-PET scan.
The [
18F]THK5317- and [
11C]PIB-PET scans were acquired on an ECAT EXACT HR+ scanner (Siemens/CTI) or a Discovery ST PET/CT scanner (GE) at the Uppsala PET centre, Uppsala, Sweden. Both tracers were synthesised according to standard good manufacturing processes as previously described [
11,
20]. For [
18F]THK5317-PET, 22 frames were acquired over 60 min (6 × 10 s, 3 × 20 s, 2 × 30 s, 2 × 60 s, 2 × 150 s, 4x300 s, and 3x600 s frames) after intravenous injection of 212 ± 42 MBq. For [
11C]PIB-PET, 24 frames were acquired over 60 min (4 × 30s, 9 × 60 s, 3 × 180 s and 8 × 300 s) after intravenous injection of 253 ± 69 MBq. The [
18F]FDG-PET scans were acquired on a Biograph mCT PET/CT scanner (Siemens) at the Department of Nuclear Medicine, Karolinska University Hospital Huddinge, Stockholm, Sweden, with a 15 min static run, 30 min after injection (30–45 min) of 3 MBq/kg. All acquisitions were reconstructed using ordered subset expectation maximisation.
[18F]THK5317-PET image pre-processing
All T1-weighted MRI images were divided into grey and white matter (GM and WM, respectively) tissue classes using SPM8 software’s unified segmentation. The inverse non-linear transformation from this segmentation step was used to wrap a probabilistic atlas [
21] from MNI space into each individual’s native T1 space. The resulting individual atlases were subsequently multiplied using the corresponding binarised probabilistic GM mask, to obtain individual GM atlases. In addition to 30 bilateral regions of interest (ROI) in the probabilistic atlas, two composite bilateral GM ROIs were created, according to the classical neuropathological staging of tau pathology [
22]: a limbic ROI (Braak stages III–IV) comprising the following regions: hippocampi, amygdalas, and parahippocampal, fusiform, middle inferior temporal, orbital and straight frontal gyri, as well as temporal poles and parietal-temporal-occipital junctions; and an isocortical ROI (Braak stages V–VI), comprising all isocortical regions except the precentral and postcentral gyri.
Individual dynamic [
18F]THK5317-PET images were co-registered onto the individual T1-weighted image with PMOD v.3.5 software (PMOD Technologies Ltd., Adliswil, Switzerland). In order to create distribution volume ratio (DVR) images, the reference Logan graphical method was applied to the [
18F]THK5317 images over the 30–60 min scan interval, with cerebellar GM—extracted from the T1 segmentation—as a reference, as previously described [
11].
Additional [18F]THK5317 preprocessing methods
Standard uptake value (SUV) images (40–60 min) were also created for [
18F]THK5317. The SUV was defined as the radioactivity concentration (MBq/mL) divided by [injected dose (MBq)/ patient’s weight (kg)]. Standard uptake value ratio (SUVR) images, using cerebellar GM—extracted from the T1 segmentation—as a reference, were generated for all individuals. Additional SUV images were used to assess the patient with progressive supranuclear palsy, since the latter disease is characterised by the presence of tau pathology in the cerebellum [
23].
The [18F]THK5317 dynamic images were also analysed, using the Muller-Gartner partial volume correction method as implemented in PMOD v.3.5 software, based on the individual T1-weighted images.
[11C]PIB- and [18F]FDG-PET image pre-processing
[
11C]PIB-PET (40–60 min) and [
18F]FDG-PET (30–45 min) images were created and co-registered onto the individual T1-weighted images using SPM8. SUVR images were created using the cerebellar GM as reference for [
11C]PIB and the pons for [
18F]FDG scans. An SUVR threshold of 1.41 for amyloid positivity was applied to [
11C]PIB-PET data [
24]. Additional DVR images for [
11C]PIB retention were generated using the reference Logan graphical method over the 30–60-min scan interval, with cerebellar GM as a reference, as previously described for [
18F]THK5317 DVR.
Statistical analysis — ROI-based comparisons
Differences between all healthy controls (i.e. yHC and eHC), prodromal, and dementia-stage AD for continuous variables were assessed with the non-parametric Kruskal–Wallis one-way analysis of variance (ANOVA), with post-hoc Mann–Whitney Bonferroni-corrected pairwise comparisons; medians and interquartile ranges (IQR) are reported. In detail, to assess comparisons across the three diagnostic groups in the two composite ROIs (limbic and isocortical), Bonferroni-corrected alpha levels of 0.008 were applied (0.05/6 [3 groups × 2 ROIs]). Fisher’s exact tests were used to assess differences between groups for discrete variables.
Within-individual test-retest variability was assessed using the relative difference [(Retest-Test)/Test] in retention between scans in ROIs from the probabilistic atlas. The intraclass correlation coefficient, a measure of reproducibility was applied.
In order to determine which ROIs from the probabilistic atlas were best for discriminating between AD patients (i.e. patients with prodromal and AD dementia) and all healthy controls (i.e. yHC and eHC), receiver operating characteristic (ROC) analysis was performed and the area under the curve (AUC) was determined for each ROI.
All the above-mentioned statistical comparisons were carried out using SPSS v.22.0 software (Armonk, NY: IBM Corp) for Mac OS X. Graphical representations were carried out with the ggplot2 package v.1.0.1 as implemented in R v.3.1.3 software.
Correlation matrices—incorporating the Pearson coefficient—were created to evaluate, two by two, the cortical ROI-based relationships between [18F]THK5317 DVR retention and [18F]FDG SUVR uptake, using the Hmisc v.3.14 and corrplot v.0.73 packages as implemented in R v.3.1.3 software.
Statistical analysis — Voxel-based comparisons
Comparisons of [
18F]THK5317 DVR retention among all healthy controls (i.e. yHC and eHC), prodromal AD, and AD dementia patients at the voxel-level were assessed, two by two, using a non-parametric alternative to two-sample T-tests. Due to the limited sample sizes of each diagnostic group significant differences were assessed with a standard non-parametric procedure based on permutation testing, as implemented in the statistical non-parametric mapping toolbox (SnPM13) [
25]. [
18F]THK5317 DVR images were registered into the MNI space using the corresponding transformation matrix obtained at the T1 segmentation step, and were smoothed (FWHM = 8 mm in all directions) using SPM8. Statistical significance was assessed with permutation tests with 10,000 permutations corrected for multiple comparisons using the false discovery rate test (
p < 0.05).
Individual retention in patients was also compared with the mean retention in the healthy controls with the creation of individual z-score maps. For [
18F]THK5317, mean and standard deviation images were created from the DVR images from the five yHCs. A z-score threshold of 1.96 (95 % confidence interval) was used to determine areas of abnormally high [
18F]THK5317 retention (z-scores > 1.96). [
18F]THK5317 DVR images from each patient were binarised according to the threshold, and were consequently summed in order to illustrate the areas of abnormally high retention in each diagnostic group (
Online Resource 1).
Relationships between tracers with respect to retention at the voxel-level were assessed, two by two, with the BPM (Biological Parametric Mapping) 3.1 toolbox [
26], using the individual PET images registered into the MNI space and smoothed, as discussed above. A cluster threshold of 20 voxels was applied, with no correction for multiple comparisons (
p < 0.001). Two more lenient thresholds (
p < 0.01 and
p < 0.05) were also applied to acknowledge the limited sample size.