Imaging Aβ and tau in early stage Alzheimer’s disease with [¹⁸F]AV45 and [¹⁸F]AV1451

Fifteen subjects were included in the analysis, with a diagnosis of mild AD (with no AD pathophysiological evidence) according to national institute of aging-Alzheimer’s association (NIA-AA) criteria [18‐21], aged between 54 and 83 years, with a mini mental state examination (MMSE) score of 21–29 and a modified Hachinski ischemic score (HIS) of less than 4 [22]. All subjects underwent a series of assessments including clinical, cognitive, gait and ophthalmological assessments, as well as molecular markers in cerebrospinal fluid (CSF), blood, urine, PET imaging, magnetic resonance imaging (MRI), magnetoencephalography (MEG) and electroencephalography (EEG). The study was approved by the institutional review board and all subjects signed an informed consent form (ICF).

All subjects underwent a 3D T1-weighted structural MRI and two dynamic PET scans to image Aβ and tau on separate days. Acquisitions were conducted in accordance with the international conference on harmonisation (ICH) guideline for good clinical practice (GCP) and the ethical principles that have their origins in the declaration of Helsinki. The institutional review board/ethics committee reviewed and approved the protocol and ICFs as well as any advertising and subject materials before any subjects were enrolled. A written, signed and dated ICF was provided before any protocol was performed.

The MRI scan was performed on a Siemens 3T Tim Trio with a 32-channel phased array head coil. A 1 mm isotropic whole-brain structural 3D T1-weighted MPRAGE [23] was acquired using TI = 880 ms, TR = 2000 ms and FA = 8° with a parallel imaging factor of 2 in 4 m: 54 s. For the PET scans, subjects were positioned in the PET scanner after the insertion of a venous cannula in an antecubital or forearm vein, and a head-fixation device was used to minimize head movement during data acquisition. The PET scans were acquired on two Siemens PET/CT (computed tomography) scanners (Hi-Rez Biograph 6 and Biograph 6 TruePoint with TrueV, Siemens Healthcare, Erlangen, Germany), though for consistency each subject had both of their PET scans on the same scanner. A low-dose CT scan was performed immediately before each PET scan in order to estimate attenuation. Each subject received a single intravenous bolus of [¹⁸F]AV45 (150 ± 24 MBq) for a 60 min Aβ PET scan and [¹⁸F]AV1451 (163 ± 10 MBq) for a 120 min tau scan. The dynamic images were reconstructed using a 2D filtered back projection (FBP) algorithm resulting in a 128 × 128 matrix with 2 mm isotropic voxels. Corrections were applied for attenuation, randoms, scatter and radioactive decay.

Two different classes of image analysis approaches were investigated in this study: static and dynamic. In both approaches, the T1-weighted MRI scan was used to obtain anatomical information for each subject. Each subject’s whole brain was extracted using the Oxford centre for functional MRI of the brain (FMRIB) software library (FSL) [24] brain extraction tool (BET) [25] and the corresponding grey matter probability maps were created using statistical parametric mapping (SPM5) software [26]. A template MRI (ICBM152 [27]) was then nonlinearly warped to subject’s MRI using SPM5 and the resulting deformation was applied to a brain anatomical atlas [28] consisting of 119 brain regions to obtain the brain region boundaries for each subject. This individualized atlas was used at a later stage to derive regional tissue time activity data from the dynamic PET images.

The dynamic PET images (2 × 2 × 2 mm) were initially motion corrected by rigidly registering each frame to a reference frame (13–15 min p.i.) using mutual information and subsequently rigidly transformed into alignment with the MRI and individualized atlas. The chosen reference frame contains both information on delivery and uptake allowing for both early and late frames to be aligned effectively. Finally, to be able to calculate different parameters, the atlas in the individualized Montreal neurological institute (MNI) space is down-sampled to match PET voxel size.

Regional time activity curves (TACs) were generated using the atlas and dynamic PET images. The SRTM and reference Logan graphical analysis with cerebellum grey matter as reference region were applied to the regional TACs to estimate the BP_ND, used to quantify the amount of tracer binding to the target proteins. Parametric images for both measures were also created by applying the models to each voxel [16].

The SRTM model equation is given by,where C_T(t) is the activity concentration in the target tissue, C_R(t) is the activity in reference tissue, k₂ is the efflux rate constant from target tissue, R₁ is the ratio of the delivery in target region to reference region. R₁, k₂ and BP_ND are estimated for each region.

The reference Logan graphical analysis method is given by,where BP_ND is the binding potential, ‘int’ is the regression intercept and t^* is the equilibrium time.

The distribution of estimated regional BP_ND values using the two reference tissue analysis methods from tau and Aβ PET scans over all subjects was compared to assess the performance of each method.

For the static analysis, SUVR values (Eq. 3) were calculated using cerebellum grey matter as the reference region. Regional average SUVR values were calculated for 20 min time windows for a range of different start times post injection (p.i.) (between 0 and 60 min for [¹⁸F]AV45 and between 0 and 120 for [¹⁸F]AV1451) from static images created by averaging the frames in each time window.where SUV_Target is the SUV of the target region and SUV_Reference is the SUV of the reference region.

Additionally, the time stability of both dynamic (BP_ND) and static (SUVR) outcome measures and their relationship were investigated. The BP_ND values were calculated for several scan durations by reducing the scan duration with steps of 10 min and the SUVR values were calculated for 20 min time windows created by splitting the scan period into 20 min time windows. The relation between these two measures was assessed by performing a regression analysis. Additionally, the correlation between regional Aβ and tau signals was explored.

All the above mentioned analysis was performed using molecular imaging and kinetic analysis toolbox (MIAKAT™, version 4.0.2, http://​www.​miakat.​org/​MIAKAT2/​index.​html).

TACs and SRTM (Eq. 1) kinetic fits were successfully generated for all 119 atlas regions of interest (ROIs) for both Aβ and tau dynamic PET scans for all subjects. The SRTM with cerebellum grey as reference region was fitted to the TACs for each region (Fig. 1a, b) and the three parameters (R₁, k₂ and BP_ND) were estimated per region. The TACs show that the tau signal from [¹⁸F]AV1451 has slower kinetics and less regional distinction in comparison to the Aβ signal from [¹⁸F]AV45. Similarly, reference Logan graphical analysis (Eq. 2) with cerebellum grey as reference tissue was fitted to the TACs (Fig. 1c, d) and the two parameters (BP_ND and int) were estimated per atlas region for each subject. Both models showed a good fit to the data for both Aβ and tau dynamic PET scans.

To investigate the approximate time at which regional BP_ND values stabilize, the mean regional BP_ND values derived using SRTM averaged over all subjects was plotted against scan duration for both Aβ and tau dynamic PET scans (Fig. 2a, b). Mean regional BP_ND estimates become stable after 30 min p.i. for [¹⁸F]AV45 and after 80 min p.i. for [¹⁸F]AV1451 for all regions. Similar analysis was performed to the same PET scans using reference Logan graphical analysis model with cerebellum grey as reference region and mean regional BP_ND values were estimated for all subjects using a range of starting time points (t*) for the regression analysis (Fig. 2c, d). For t* values of over 30 min, the BP_ND values are stable for Aβ scans, whereas stability occurs at t* greater than 70 min for tau. The estimated regional BP_ND values using the two dynamic models were compared over all subjects for Aβ and tau dynamic PET scans (Fig. 3). Similar regional BP_ND values were estimated by both dynamic models but overall, the regional BP_ND values estimated by SRTM were higher on average with larger variation across subjects.

The static analysis involved the calculation of regional SUVR values for 20 min scan windows spanning the full scan duration using Eq. 3 with the grey matter cerebellum as the reference region. The time stability of mean regional SUVR was assessed for several regions for all subjects (Fig. 4) for Aβ and tau PET scans which showed that the values became stable for scan windows of ~ 30–50 min p.i. for [¹⁸F]AV45 and ~ 80–100 min p.i. for [¹⁸F]AV1451.

To investigate the relationship between the dynamic and static measures, the coefficient of determination (R²) was calculated for all atlas regions across all subjects between BP_ND values, obtained from SRTM and reference Logan graphical analysis methods, and SUVR values of several 20 min time windows. The relationship between these two outcome measures started to stabilize from scan windows of ~ 40–60 min p.i. for [¹⁸F]AV45 and ~ 80–100 min p.i. for [¹⁸F]AV1451. The linear regression analysis of the regional BP_ND and SUVR values for the above mentioned time windows showed a high correlation between the two measures using both dynamic methods for both Aβ (SUVR_{40 − 60} = 0.97 × BP_{ND_SRTM} − 1.2, R²_SRTM = 0.75; SUVR_{40 − 60} = 1.11 × BP_{ND_Logan} − 1.3, R²_Logan = 0.7; p < 0.01) and tau (SUVR_{80 − 100} = 1.028 × BP_{ND_SRTM} − 1.13; R²_SRTM = 0.88; SUVR_{80 − 100} = 1.27 × BP_{ND_Logan} − 1.25; R²_Logan = 0.71; p < 0.01) PET scans.

The correlation between estimated regional Aβ and tau tracer signals using each of SRTM, reference Logan graphical analysis and SUVR was explored. Twelve different atlas regions with potential of having high Aβ and tau uptake were selected based on the literature and R² was calculated for each outcome measure (BP_ND or SUVR) for all selected region pairs (Fig. 5). In particular, for the quantitative dynamic analysis methods (SRTM and reference Logan), Aβ signal in the thalamus was highly correlated with tau signal in thalamus (R²_SRTM = 0.76, R²_RefLogan = 0.84, p < 0.05), hippocampus (R²_SRTM = 0.74, R²_RefLogan = 0.73, p < 0.05) and amygdala (R²_SRTM = 0.56, R²_RefLogan = 0.55, p < 0.05). For SUVR, whilst the overall pattern of correlations was similar, R² was reduced (hippocampus: R²_SUVR = 0.58, thalamus: R²_SUVR = 0.46 and amygdala: R²_SUVR = 0.57). The tracer uptake pattern in whole brain, for both [¹⁸F]AV45 and [¹⁸F]AV1451, are shown in Fig. 6 for all 12 subjects who completed full 60 min of the Aβ PET scan and at least 110 min of the tau PET scan. Overall, the tau signal was very low in all subjects but there was a strong signal observed in the striatal areas. Amyloid signal on the other hand was strong in all brain regions and was not directly related to MMSE scores (R² = 0.006).