Cross-Sectional Comparison of Small Animal [18F]-Florbetaben Amyloid-PET between Transgenic AD Mouse Models

We aimed to compare [18F]-florbetaben PET imaging in four transgenic mouse strains modelling Alzheimer’s disease (AD), with the main focus on APPswe/PS2 mice and C57Bl/6 mice serving as controls (WT). A consistent PET protocol (N = 82 PET scans) was used, with cortical standardized uptake value ratio (SUVR) relative to cerebellum as the endpoint. We correlated methoxy-X04 staining of β-amyloid with PET results, and undertook ex vivo autoradiography for further validation of a partial volume effect correction (PVEC) of PET data. The SUVR in APPswe/PS2 increased from 0.95±0.04 at five months (N = 5) and 1.04±0.03 (p<0.05) at eight months (N = 7) to 1.07±0.04 (p<0.005) at ten months (N = 6), 1.28±0.06 (p<0.001) at 16 months (N = 6) and 1.39±0.09 (p<0.001) at 19 months (N = 6). SUVR was 0.95±0.03 in WT mice of all ages (N = 22). In APPswe/PS1G384A mice, the SUVR was 0.93/0.98 at five months (N = 2) and 1.11 at 16 months (N = 1). In APPswe/PS1dE9 mice, the SUVR declined from 0.96/0.96 at 12 months (N = 2) to 0.91/0.92 at 24 months (N = 2), due to β-amyloid plaques in cerebellum. PVEC reduced the discrepancy between SUVR-PET and autoradiography from −22% to +2% and increased the differences between young and aged transgenic animals. SUVR and plaque load correlated highly between strains for uncorrected (R = 0.94, p<0.001) and PVE-corrected (R = 0.95, p<0.001) data. We find that APPswe/PS2 mice may be optimal for longitudinal amyloid-PET monitoring in planned interventions studies.


Introduction
The steadily growing number of patients suffering from Alzheimer's disease (AD) will place a great burden on healthcare systems in the coming decades, baring development of an effective intervention therapy [1]. Molecular imaging of β-amyloid with positron emission tomography (PET) has given new insight into the progression of AD pathology and has entered clinical diagnostic use [2]. Furthermore, PET imaging is increasingly used for detecting cerebral amyloidosis in transgenic mouse models of AD [3,4]. Small animal PET studies of longitudinal design afford monitoring of the rate of β-amyloid accumulation, and present the possibility of testing interventions for attenuating plaque formation. However, there are some controversies in the literature with regard to the fitness of amyloid-imaging in different AD-mouse models. For example, one PET study failed to reveal any specific signal [5], and another did not detect the known increase of β-amyloid in an aging AD mouse model [6]. Other investigations successfully monitored the progression of amyloidosis in APP23 mice [7] or the early onset of amyloidosis in 5xFAD mice [8]. Comparisons between recent small animal β-amyloid PET studies are difficult, due to the diversity of mouse models, radiotracers, imaging protocols, and quantification strategies [9]. Hence, we wanted to undertake a PET study with an established β-amyloid tracer in several AD mouse strains, applying standardized methods, and correlating PET findings with histological analysis. We therefore aimed to make PET recordings with the [ 18 F]-labelled β-amyloid tracer florbetaben [10][11][12] in four different AD mouse models at a range of ages, with the main focus on APPswe/PS2 mice. Our additional objective was to generalize our validation of partial volume effect correction (PVEC) for quantitation of β-amyloid burden in APPswe mice [13] for APPswe/PS2 mice. Furthermore, we conducted a comprehensive search of the existing literature in order to review systematically the several AD mouse models investigated by florbetaben and other PET tracers for β-amyloid.

Animal models
All experiments were performed in compliance with the National Guidelines for Animal Protection, Germany, with approval of the local animal care committee of the Government of Oberbayern (Regierung Oberbayern), and overseen by a veterinarian. Anaesthesia was performed with isoflurane 1.5%. Euthanasia was performed in deep narcosis by cervical dislocation. Three newly explored and one previously described [3] AD mouse models were assessed by PET at different ages, and ultimately by autoradiography ex vivo and histological methoxy-X04 staining ( Table 1). Data from the three newly examined models was assessed using groups of mice at different ages, while historical data from APPswe mice had been acquired longitudinally. A total group of 22 age-matched C57Bl/6 mice (WT) served as control material. Details of transgenic strains follow: 2.1.1 APPswe/PS2 (PS2APP). The transgenic B6.PS2APP (line B6.152H) is homozygous for both the human presenilin (PS) 2, N141I mutant and the human APP K670N, M671L mutant. APP and PS 2 transgenes are driven by mouse Thy-1 and mouse prion promoters, respectively. This line had been created by co-injection of both transgenes into C57Bl/6 zygotes, as previously described [14]. Homozygous B6.PS2APP mice were crossed with (C57Bl/6 x DBA/ 2) x (C57Bl/6 x DBA/2) mice to generate the 152F2 mice, which are hemizygous for both transgenes, and show first plaques in the cortex and hippocampus at 6 months of age [15]. We made PET recordings in 152F2 mice (N = 32), i.e. five animals aged five months, seven animals aged eight months, six animals aged ten months, two aged 12 months, six aged 16 months and six aged 19 months; a subset was used for additional histological analysis. The six ten-month old animals were used for test-retest experiments within an interval of one week.
2.1.2 APPswe/PS1G384A (G384A). APPswexPS1G384A mice were generated by crossing of two pre-existing transgenic mouse lines. APPswe are characterized as stated above, while PS1G384A carries a transgene for mutant PS 1 (G384A mutation; both lines are driven by a Thy-1 promoter). G384A mice (carrying the APP23 transgene) accumulate a significant amount of the 42 amino acid form of ß-amyloid (Aß42) by the time they are two months old; amyloid plaques are at first sparsely distributed, and later on spread from deeper to superficial cortical layers, with highest density in the frontal and somatosensory cortex [16]. We obtained PET recordings in two animals aged five months, and a single animal aged 16 months.
2.1.3 APPswe/PS1dE9 (APP/PS1dE9). APP/PS1dE9 mice [17] overexpress the Swedish mutation of APP, together with PS 1 deleted in exon 9 each driven by the mouse PrP promoter. Overexpression of the transgene construct leads to overproduction of Swedish mutant APP and the PS1 Δexon9 splice variant with concomitant increase in parenchymal β-amyloid load. These mice develop the first plaques at four months of age mainly in cortical areas and the hippocampus [18]. We obtained PET recordings from two mice aged 12 months and two animals aged 24 months.
2.1.4 APPswe. Transgenic APPswe mice overexpressing human amyloid precursor protein (APP) with the Swedish double mutation (K670N, M671L) driven by the mouse Thy1.2 promoter were generated as described earlier [14]. These mice were backcrossed to C57BL/6 mice to generate a line with less than 5% DBA2 background (as determined by microsatellites genotyping). Mice heterozygous for the transgene start to accumulate ß-amyloid at the age of approximately nine months and develop ß-amyloid plaques around the age of 12 months, mainly in the cortical mantle. Historical PET data [3] from 15 APPswe mice imaged at four different ages (10 months: N = 5; 13 months: N = 12; 16 months: N = 8; 20 months: N = 5) was reprocessed using the same standardized parameters as for the newly examined models.

Radiochemistry
The [ 18 F]-florbetaben precursor (Piramal Imaging, Berlin) was radiolabeled by the method of Zhang et al. [11], with slight modifications. This procedure yields a radiochemical purity exceeding 98% and specific activity of 50-90 GBq/μmol at the end of synthesis. In brief, [ 18 F]fluoride was produced by a cyclotron using the 18  , with mobile phase consisting of CH 3 CN/ammonium formate buffer (0.1 M, with 5 mg/mL sodium ascorbate) 60/40 delivered at a flow rate 5 ml/min. The fraction with the desired product, which eluted with a retention time of 12 min, was collected and diluted with 12 ml water containing 10 mg/ml sodium ascorbate. This solution was passed through a Sep-Pak Light C18 cartridge and rinsed with 10 ml of 20% ethanol in water (also containing 10 mg/ml sodium ascorbate) to separate the hydrolysis by-product from the [ 18 F] florbetaben final product, which was then eluted from the cartridge with 1-2 ml ethanol. For mouse experiments, the eluate was divided into 100 μl portions, which were formulated with 900 μl of stabilization solution (1.78 ml PEG400, 6.72 ml H 2 O, 44 mg ascorbic acid, 288 mg sodium ascorbate), followed by adjustment of pH to 7.4 with 75 μl of concentrated phosphate buffer (Braun, Melsungen, Germany). The total synthesis time was 75 min and radiochemical yield was 18% (decay corrected). The product was examined for purity by analytical HPLC using an XDB-C18 analytical column (150×4.6 mm, 5 μm; Agilent Technologies), and mobile phase consisting of CH 3 CN/ammonium formate buffer (0.1 M, with 5 mg/mL sodium ascorbate) 60/40, delivered at a flow rate of 1 ml/min, giving a retention time of 6.2 min, as monitored with serial UV and radiodetection.

Amyloid PET
Image acquisition and reconstruction followed a standardized protocol, with minor modifications from the previously-published methods [3]. Mice were anesthetized with isoflurane (1.5%, delivered via a mask at 3.5 L/min in oxygen) and received bolus injection of 10. While still deeply anesthetized, mice were killed by cervical dislocation immediately after the PET scan (62 min p.i.), followed by rapid removal of the brain (68 ± 4 min p.i.). The image reconstruction procedure consisted of a 3-dimensional ordered subset expectation maximization (OSEM) with four iterations and 12 subsets followed by a maximum a posteriori (MAP) algorithm with 32 iterations. Scatter and attenuation correction were performed and a decay correction for [ 18  For VOI-based analyses, a cerebellar VOI comprising 64 mm³, two bilateral frontal cortex VOIs comprising 15 mm³ each and two bilateral hippocampal VOIs comprising 7mm³ each, were employed for calculation of [ 18 F]-florbetaben cortex-to-cerebellum (SUVR CTX/CBL ) and hippocampus-to-cerebellum standardized uptake value ratios (SUVR HIP/CBL ). Cortical VOIs were configured larger in comparison to the initial [ 18 F]-florbetaben validation study [3] as they proved higher robustness in a recent investigation [13]. The initial oval-shaped VOI was superceded with an anatomically delineated VOI, more accurately accounting for the borders of the neocortex.
Test-retest estimates were acquired in a subset of six PS2APP animals at ten months of age within one week.

Partial volume effect correction
PVEC was first performed in a subset of five PS2APP animals for which ex vivo autoradiography was available. After its successful validation, PVEC was applied to the entire dataset. In particular, we used a VOI-based approach implemented in PMOD [19,20] with a full-widthat-half-maximum (FWHM) of 1.72 mm, and a previously-evaluated VOI-mask for ten volumes, consisting of two neocortical VOIs and one cerebellar VOI, along with further VOI comprising the rest of the cerebrum, and an additional six extra-cerebral VOIs (Harderian glands, frontal, superior, basal, spinal, and background). PVEC was additionally performed using a VOI-mask including two hippocampal VOIs instead of the two neocortical VOIs [13]. The neocortical, hippocampal and cerebellar VOIs used for the VOI-mask were identical to those used for calculation of the uncorrected PET data. As described by Rousset et al., regional point spread functions were calculated through integration of single tissue domains' point spread function, and used for computation of weighting factors representing the contributions of the set of ten tissue domains. Coefficients of a geometric transfer matrix (GTM) were calculated, and PVE-corrected radioactivity concentrations were calculated in the defined VOIs by multiplication of the original PET data by the inverted GTM. The PVE-corrected images thus contained only those regions with VOI-mask definitions, for comparison with VOI-based results for uncorrected PET data. Finally, VOIs were employed for calculation of PVE-corrected [ 18 F]-florbetaben cortex-to-cerebellum (SUVR CTX/CBL ) and hippocampus-to-cerebellum standardized uptake value ratios (SUVR HIP/CBL ).

Ex vivo autoradiography
One hemisphere of five PS2APP animals (8 mo: N = 1; 12 mo: N = 2; 19 mo: N = 2) and of one G384A mouse (16 mo: N = 1) killed at 62 min p.i. ( Table 1) was frozen by immersion in isopentane (-40°C); 20 μm thick sagittal slices were cut with a Leica CM 1510-1 Cryostat (Leica Microsystems, Nussloch Germany), and mounted on glass slides within 60 minutes post mortem. An imaging plate (Fujifilm; BAS cassette2 2025) was exposed to the slides for ten hours, scanned at 25 μm resolution with Raytest equipment (CR 35 BIO, Dürr Medical, Germany), and analyzed with AIDA image analyzing software (V4.50). Cerebellar and frontal cortical regions-of-interest (ROIs) were drawn analogous to PET-VOIs, and mean radioactivity concentrations per mg tissue equivalent were used for calculation of autoradiographic SUVR CTX/CBL and SUVR HIP/CBL (mean of > 15 slices per animal). As a smaller cortical VOI was used in the initial PET study of APPswe mice [3], we reanalyzed slices from the APPswe mice so as best to match the approach used for the new material. Error-(%) in uncorrected and PVE-corrected in vivo PET results was calculated relative to this high resolution "gold standard" of autoradiographic results ex vivo.

Histochemical analyses
Cerebral hemispheres intended for histochemistry were fixed by immersion in 4% parafomaldehyde at 4°C for one week. Several representative 100 μm thick slices in the sagittal plane about 1.5 mm from the midline were cut from each fixed hemisphere using a vibratome (VT 1000 S, Leica, Wetzlar, Germany). The slices were permeabilized overnight at room temperature in 2% Triton X-100 in phosphate buffered saline (PBS) at pH 7.4. Fibrillary ß-amyloid plaques were stained with the fluorescent dye methoxy-X04 (0.01 mg/ml in PBS at pH 7.4 for 15 min) [21]. The unbound dye was removed by three washing steps with PBS, and the slices were then mounted on microscope slides with fluorescent mounting medium (Dako, Germany). 3D image stacks for each hemisphere were acquired on a confocal laser scanning microscope (LSM 780, Jena, Zeiss, Germany). Imaging of the whole slice was performed in tile scan mode which allows automatic stitching of an array of fields of view. For this imaging, the methoxy-X04 was excited at 405 nm, and the emitted light was collected from 410 to 585 nm. The area and number of plaques were automatically counted using Imaris software (Imaris 7.4.2; Bitplane) in a region (0.49 ± 0.06 mm³) defined to match dimensions of the frontal cortex VOI from PET image analysis. Plaque load was calculated as the summed area of all plaques relative to the frontal cortex area. As a smaller cortical VOI was used in the initial PET study of APPswe mice [3], we reanalyzed slices from the APPswe mice so as best to match the approach used for the new material. The histological analyses were performed by an operator blind to the PET results.

Statistics
Group comparisons of uncorrected and PVE-corrected VOI-based PET results measured as SUVR CTX/CBL and SUVR HIP/CBL were performed using one-way ANOVA and the Tukey posthoc test for multiple comparisons, calculated by IBM SPSS 22 Statistics. Groups of APP-Swe and PS2APP mice with N 5 were compared against their younger littermates and against the WT group. Comparisons for groups with N < 5 were reported descriptively with their littermates and respective WT animals. For correlation analyses and test-retest analyses, Pearson's coefficients of correlation (R) were calculated. A threshold of p < 0.05 was considered to be significant for rejection of the null hypothesis.

Literature review
We performed a Pubmed search with the keywords "AD mouse model" and "amyloid PET" aiming to identify all relevant studies. Mouse strain characteristics, the particular radiotracer, and PET image acquisition and reconstruction parameters, as well as summary autoradiographic findings were recorded. Furthermore, Aß42/40 protein levels as well plaque load as functions of mouse age were recorded, when available.
3.1.2 Hippocampal findings. SUVR HIP/CBL was 1.00 ± 0.03 in WT mice; corresponding hippocampal measurements in PS2APP mice aged five, eight or ten months, G384A mice aged five months, APP/PS1dE9 mice aged 12 and 24 months, and APPswe animals at 16 and 20 months were slightly higher than in WT, with SUVR HIP/CBL in the range of 1.04-1.08. Among the PS2APP mice, significantly higher SUVR HIP/CBL values were observed in groups of six mice aged 16 months (1.28 ± 0.07; p < 0.001) and aged 19 months (1.31 ± 0.12; p < 0.001) when compared to their littermates at five months of age. Increased SUVR HIP/CBL was also seen in the single 16 month old G384A mouse (1.13).

Partial volume effect correction and ex vivo autoradiography
Applying the PVEC increased the SUVR CTX/CBL measurements by 30 ± 11% (range: 12-53%) in the five histologically analyzed PS2APP animals, with the more pronounced effect seen in those animals with higher uncorrected SUVR CTX/CBL . Values of SUV in the cerebellum VOI increased by 8 ± 4% with PVEC, propagating to a nonlinear relationship between PVE-corrected and uncorrected SUVR CTX/CBL, which was described by y = 1.16x 1.65 . Corrected SUVR CTX/CBL was 1.18 in the single PS2APP mouse aged eight months and age-dependently increased to 1.47/1.55 (+25/31%) in the two mice aged 12 months and to 1.68/2.04 (+42/73%) in two mice aged 19 months ( Fig. 2A). Applying PVEC with the hippocampal VOI-mask increased SUVR HIP/CBL by 20 ± 5%, in a nonlinear relationship described by y = 1.09x 1.34 . SUVR CTX/CBL by autoradiography ex vivo was 13-30% higher in PS2APP animals (1.18 at eight months, 1.36/1.53 at 12 months, and 1.60/2.05 at 19 months) relative to PET results without PVEC; this difference indicates an overall mean 22 ± 7% underestimation of amyloid burden by PET. PVEC reduced this mean bias to a non-significant overestimation by 2 ± 5% (Fig. 2B).
Autoradiographic SUVR HIP/CBL was 1.11 in an eight month old PS2APP animal, and increased to 1.28/1.38 in two littermates aged 12 months and 1.27/1.55 in the pair aged 19 months. Uncorrected PET data showed a bias of −5 ± 4% versus +14 ± 12% with PVEC, indicating increased bias and variance. SUVR CTX/CBL by autoradiography ex vivo in the single G384A mouse was 1.41, i.e. only 3% higher than the PVE-corrected PET SUVR CTX/CBL of 1.37, and 29% higher than the uncorrected PET SUVR CTX/CBL of 1.09 in the right hemisphere.

PVEC application to the entire dataset
Uncorrected and PVE-corrected SUVR CTX/CBL of all mice are provided in Table 1. PVEcorrected values distinctly increased the estimated slope, especially in the PS2APP strain and the difference between young and old PS2APP animals (Fig. 3).

Histochemical analyses
3.5.1 Plaque analysis. PS2APP mice had close to 1300 mainly compact plaques in the frontal cortex at all ages studied, but the average size and maximum plaque diameter increased with, leading to a plaque load of up to 12.2% in 19 month old animals; plaques were also abundant in olfactory bulb and thalamus. G384A mice showed a fourfold increase in number of plaques from 5 to 16 months of age, with a concomitant increase in mean diameter. APP/PS1dE9 mice revealed a sparse distribution of small plaques in cerebral cortex, and also in cerebellum. Reanalysis of material from APPswe mice aged 20 months showed distinctly larger plaques (max. diameter 243 μm) compared to the three other strains (all < 100 μm). Plaque load in 20 month old APPswe mice (6.3 ± 3.1%) was comparable to that of the 16 month old G384A mouse (7.6%). For details see Table 2. 3.5.2 Cross-sectional plaque load correlation. We found a very high correlation between cortical plaque load in all 22 hemispheres stained with methoxy-X04 and the uncorrected SUVR CTX/CBL (R = 0.94, p < 0.001; Fig. 4A) as well as with the PVE-corrected SUVR CTX/CBL (R = 0.95, p < 0.001; Fig. 4B) as assessed by [ 18 F]-florbetaben PET. The only outlier was a single 24 month old APP/PS1dE9 mouse (arrow), in which too low SUVR CTX/CBL was apparently due to bias in PET results arising from plaques in the cerebellum reference region.

Discussion and Review of the Literature
We present the first cross-sectional small animal PET study with a single [ 18 F]-labelled radiotracer for β-amyloid under uniform standardized experimental settings in several established AD mouse models. We found distinct differences in time course and anatomic pattern of βamyloid deposition among the four AD mouse strains investigated. PET findings of cortical [ 18 F]-florbetaben binding in AD mouse models matched precisely the distribution of plaques verified by histology with methoxy-X04 staining. Of the four AD mouse strains examined, PS2APP mice were judged most suitable for future intervention studies, based on the presence of [ 18 F]-florbetaben binding in mice aged only 8 months, continued substantial β-amyloid accumulation to at least 19 months, and lack of specific binding in the cerebellum reference region.

PS2APP animals
A previous study with [ 18 F]-florbetapir in double transgenic APP/PS1 mice showed an increasing PET signal from three to five months, but no further progression to 12 months [6]; the transgenic animals attained a constant SUVR CTX/CBL for [ 18 F]-florbetapir of only 1.1. This PET finding was in contrast to histopathological plaque load, which increased 3-fold from four to eight months of age, and further doubled by 19 months of age [22]. In contrast, our PS2APP mice showed 8% elevated [ 18 F]-florbetaben SUVR CTX/CBL at eight months relative to WT mice, and then a continuous progression of cortical binding to 19 months. The PET findings are in line with histology, and with earlier reports of β-amyloid protein levels in the PS2APP strain [15]. Compared to the human pathology compact plaques of different sizes represent the majority in old PS2APP mice, whereas diffuse and cored plaques are predominating in AD patients [23]. Similarly there was general agreement between PET signal from [ 11 C]-Pittsburgh Compound B ([ 11 C]-PiB), plaque load and Aß42 protein in homozygous ARTE 10 mice, which  Plaque load percentage (PL-%) and plaque numbers in the frontal cortex VOI are reported (columns 4 and 5), as well as the density of plaques (column 6). Exemplified sagittal slices from methoxy-X04 staining of ß-amyloid plaques are shown together with illustrations of the plaque size distribution in S1 were evaluated at ages up to 21 months [4]. A recent report of 5xFAD mice examined with [ 11 C]-PiB and [ 18 F]-florbetapir at 11 months of age showed distinctly elevated cortical uptake of both tracers (+20%) compared to WT mice [8]. Thus, the report of a plateau of [ 18 F]florbetapir uptake in aging APP/PS1 mice may be anomalous.

G384A animals
Double transgenic mice carrying mutations of APP and the G384A mutation are reported to develop solid amyloid plaques as early as two months of age, when the APP23 strain was used for the cross-breeding [16]. In our study methoxy-X04 staining revealed solid plaques at a low plaque load of 1.5% in G384A mice aged five months including the APPswe mutation. However we did not see consistently elevated [ 18 F]-florbetaben PET signal in these animals, but only in a single mouse aged 16 months which presented a 5-fold higher plaque load compared to five months. We concede that we examined relatively few G384A mice, i.e. two animals at 5 months and a single animal aged 16 months. As the extent of amyloidosis in transgenic AD models can be highly variable, the seeming increase in the single G384A mouse has to be interpreted with caution. Nonetheless, the evidence for a positive ß-amyloid PET signal was unambiguous in the aged animal of this strain, as its signal clearly exceeded the upper bound of the 95%-CI in WT mice. There is abundant precedent in the literature for discrepancies between age of plaque onset and detectability with small animal amyloid PET. Thus, several explorations of APP23 mice imaged with [ 11 C]-PiB [7,24,25] showed that transgenic mice with only an APP mutation can indeed serve for β-amyloid imaging. Sparse plaque formation was evident histologically in APP23 mice as early as six months, and aged animals exhibited numerous large dense-core plaques, high Aß40 level and a remarkable plaque load (Table 3). Nonetheless, elevated [ 11 C]-PiB  Table 3. Comprehensive overview of histological studies in ten AD mouse strains which have been investigated in PET studies to date, including the present work.

APP/PS1dE9 animals
From our data in four mice, the APP/PS1dE9 model appeared unsuitable for PET quantitation of relative cerebral amyloid burden due to the presence of ß-amyloid pathology in the reference region (cerebellum). Furthermore, cortical plaque levels in even the most aged mice (24 mo) only slightly exceeded our detection threshold for [ 18 F]-florbetaben PET (noted above), with the caveat that one must take into the small sample size. In concordance with this finding in vivo, the plaque load ( Table 2) and Aß42 protein levels ( Table 3) in the APP/PS1dE9 strain appear far lower than those of APP23, G384A and PS2APP mice, which likely accounts for the lower [ 18 F]florbetaben PET signal in 24 month old APP/PS1dE9 mice when compared to other mouse strains at this age. Earlier reports on APP/PS1dE9 mice did not report the presence of cerebellar plaques, albeit these studies were confined to younger animals [17,26], or only investigated forebrain for histological analyses [27]. However, a very recent longitudinal [ 11 C]-PiB PET study of APP/PS1dE9 mice reported cerebellar plaques in 19 month old mice [7]. In our hands, aged APP/PS1dE9 mice had comparable ß-amyloid levels in cerebellum and in cortical areas, such that the SUVR CTX/CBL actually declined from 12 to 24 months of age, ultimately falling below the lower bound of the 95%-CI in WT mice. The cerebellum has been used as a reference region in nearly all recent small animal β-amyloid PET studies, although its histological suitability was previously established only for ARTE 10 and APPswe mice [3,4]. The plaque-free pons might be proposed as an alternate reference region, but due to its small size in mice, we expect that its use would bring considerable penalty in precision of binding ratios. Furthermore, plaques in APP/ PS1dE9 mice had 3-fold higher plaque area in a 3D6-immunohistochemistry study which also detects diffuse β-amyloid deposits, in contrast to thioflavin S staining, which is only sensitive to dense aggregates [26]. Our approach of using a radiotracer and a histological marker with similar binding characteristics gave more concordant results, in that low plaque load matched with low in vivo [ 18 F]-florbetaben binding. Therefore, we agree with Snellman et al. (2013) that low SUVR CTX/CBL magnitude likewise reflected inherently lower radiotracer binding in 24 month old APP/PS1dE9 mice when compared to other AD mouse strains at this age.

Partial volume effect correction
A previous report on [ 18 F]-florbetapir PET in APP/PS1 mice reported 20-30% higher SUVR CTX/CBL in autoradiographic results ex vivo compared to PET findings [6]. We likewise found nearly identical discrepancies by method in young and aged PS2APP animals, as well as the re-evaluated APPswe mice, which we attribute to errors in regional quantitation arising from uncorrected PVEs. Application of the Rousset method for cross-talk correction essentially removed the bias in our PET estimates [13]. The present generalization of the PVEC algorithm to another mouse model helps establish the validity of the approach to improved quantitation; there have been but few investigations of PVEC in the rodent brain. We find that cortical spillover is otherwise a significant factor in underestimating β-amyloid signal, especially in mice with high plaque load. The systematic use of PVEC will help to reduce quantification bias in future studies, and should improve sensitivity of PET for detecting effects of interventions. Unlike in neocortex, uncorrected PET results in hippocampus were more congruent with the autoradiographic results, no doubt due to the VOI being surrounded by tissues with similar radioactivity concentrations, resulting in lesser net spill-out of signal. Indeed, PVEC for hippocampus resulted in overcorrection and excessive variance, which is unsurprising for such small structures (6.6 mm³). Despite the limitations of PVEC for the mouse hippocampus, in vivo β-amyloid imaging of that structure may yet be possible without correction, as suggested by our finding of increasing SUVR HIP/CBL in aged PS2APP animals.

Cross-sectional plaque load correlation
The reason for the general finding of low binding of β-amyloid PET radiotracers in the brains of young transgenic mouse models of AD, or even absence of specific signal in some strains, despite abundant β-amyloid levels, is a matter of considerable controversy. Maeda et al. [25] even reported lower in vitro binding of [ 11 C]-PiB to plaques in double transgenic PS-1/APP mice aged eight months compared to Tg2576 animals aged 23 months with similar Aß42 levels; they concluded that plaque formation in accelerated models impart fewer radiotracer binding sites due to post-translation factors resulting in lower AßN3-pyroglutamate composition. We found an excellent cross-sectional correlation of PET results and methoxy-X04 staining for our four AD models (Fig. 4). This probably reflects the similar binding/staining properties of [ 18 F]-florbetaben and methoxy-X04, as both detect fibrillar but not diffuse β-amyloid depositions [10,21]. Although different mouse strains in this investigation had a high variability of plaque sizes, numbers and morphology, the relationship between stained fibrillar β-amyloid intensity and radiotracer binding was nearly linear. Of note is that in cortex of AD patients the compact plaque type occurs at only 10%, with cored and diffuse plaques dominating. Furthermore all three plaque types differ with respect to Aβ species, with diffuse and compact plaques containing mostly Aβ42, while cored plaques contain Aβ40 [23].
We conclude that histological analyses with this Congo red derivate are indeed predictive of [ 18 F]-florbetaben binding sites to be imaged by PET, with the caveat that the in vivo method is insensitive to low plaque loads (<1.5%). This is probably related to factors such as low specific binding signal, the limited spatial resolution of PET, and image noise. Additionally, the impact of different specific activities of [ 18 F]-florbetaben have not yet been tested systematically in preclinical settings, and might conceivably have influenced the detectability threshold in our study [25], however in a preliminary analysis in a series of PS2APP mice no conspicuous effect on scan results as a function of time from end of synthesis could be observed (data not shown).
Comparisons between β-amyloid PET studies in AD mouse models must be made with some caution, due to diverse methodological differences, including the particular radiotracers, emission recording time-frames and reconstruction parameters. Cortex-to-reference region estimates for WT animals fall within a narrow range in a number of recent small animal PET studies using different radiotracers irrespective of age (Table 4), which predicts general stability of the method. On the other hand, PET signals in transgenic animals are more difficult to

Protein quantification and plaque load
Based on our literature review, high Aß42 protein levels (> 100 ng/mg wet brain) have been found in APP23, G384A, PS2APP, ARTE10 and 5xFAD mice, even at ages less than 12 months in the latter four strains ( Table 3). In contrast, Tg2576, APPswe, and APP/PS1dE9 mice had far  less protein accumulation, even in aged animals. In animals for which plaque load measurements were also available, there was a close correspondence between histological load and Aß42 protein levels. Aß42 concentration in APP/PS1-21 mice increased from 11 ng/mg wet brain at three months of age to 67 ng/mg wet brain at eight months of age, and was not assessed at greater ages, whereas the histological plaque load in these animals further doubled at 19 months of age. Protein levels and plaque load results derive from different publications using a variety of methodologies for the assessment. Comparisons between different mouse strains should therefore be made with some caution.

Limitations
The findings reported for G384A and APP/PS1dE9 mice are qualified by the small sample size per time point in these strains. Further PET scans could not be obtained in these strains due to unavailability of additional mice.

Conclusion
Present findings support the use of double transgenic PS2APP mice for early detection of cortical β-amyloid deposition by means of [ 18 F]-florbetaben PET, in conjunction with PVEC. Uniform standardized procedures offer the capability for valid comparisons of β-amyloid PET results in different AD mouse strains. Interpretation of PET results in specific models is facilitated in a cross-sectional context, which removes most sources of differences between studies. The question if single or multi transgenic models are better suited for future treatment approaches likely depends on the type of intervention to be tested.