Microglial activation in the right amygdala-entorhinal-hippocampal complex is associated with preserved spatial learning in AppNL-G-F mice

BACKGROUND
In Alzheimer`s disease (AD), regional heterogeneity of β-amyloid burden and microglial activation of individual patients is a well-known phenomenon. Recently, we described a high incidence of inter-individual regional heterogeneity in terms of asymmetry of plaque burden and microglial activation in β-amyloid mouse models of AD as assessed by positron-emission-tomography (PET). We now investigate the regional associations between amyloid plaque burden, microglial activation, and impaired spatial learning performance in transgenic mice in vivo.


METHODS
In 30 AppNL-G-F mice (15 female, 15 male) we acquired cross-sectional 18 kDa translocator protein (TSPO-PET, 18F-GE-180) and β-amyloid-PET (18F-florbetaben) scans at ten months of age. Control data were obtained from age- and sex-matched C57BI/6 wild-type mice. We assessed spatial learning (i.e. Morris water maze) within two weeks of PET scanning and correlated the principal component of spatial learning performance scores with voxel-wise β-amyloid and TSPO tracer uptake maps in AppNL-G-F mice, controlled for age and sex. In order to assess the effects of hemispheric asymmetry, we also analyzed correlations of spatial learning performance with tracer uptake in bilateral regions of interest for frontal cortex, entorhinal/piriform cortex, amygdala, and hippocampus, using a regression model. We tested the correlation between regional asymmetry of PET biomarkers with individual spatial learning performance.


RESULTS
Voxel-wise analyses in AppNL-G-F mice revealed that higher TSPO-PET signal in the amygdala, entorhinal and piriform cortices, the hippocampus and the hypothalamus correlated with spatial learning performance. Region-based analysis showed significant correlations between TSPO expression in the right entorhinal/piriform cortex and the right amygdala and spatial learning performance, whereas there were no such correlations in the left hemisphere. Right lateralized TSPO expression in the amygdala predicted better performance in the Morris water maze (β = -0.470, p = 0.013), irrespective of the global microglial activation and amyloid level. Region-based results for amyloid-PET showed no significant associations with spatial learning.


CONCLUSION
Elevated microglial activation in the right amygdala-entorhinal-hippocampal complex of AppNL-G-F mice is associated with better spatial learning. Our findings support a protective role of microglia on cognitive function when they highly express TSPO in specific brain regions involved in spatial memory.


Introduction
Alzheimer disease (AD) is the most frequently diagnosed neurodegenerative disease, with burgeoning incidence rates due to the rising life expectancy in industrialized nations ( Ziegler-Graham et al., 2008 ). The neuropathology of AD is histologically characterized by the triad of accumulation of -amyloid peptide (A ) as extracellular plaques, fibrillar tau aggregates within neurons, and the activation of multiple neuroinflammatory pathways, which is marked by activated microglia expressing high levels of the marker 18-kDa translocator protein (TSPO) ( Heneka et al., 2015 ). In humans, regional heterogeneity and asymmetric spatial distribution of neuropathological AD hallmarks is a frequent finding in PET studies ( Frings et al., 2015 ;Ossenkoppele et al., 2016 ;Tetzloff et al., 2018 ). This seems of particular importance since as the clinical phenotype of AD depends on the regional localization of tau distribution ( Ossenkoppele et al., 2016 ), whereby verbal memory performance is linked to the usually dominant left hemisphere ( Miyashita, 2004 ). Recently, we showed that regional asymmetric distribution of fibrillar amyloidosis is also evident in several mouse models of amyloidosis ( Sacher et al., 2020 ), in which 14%/ 16% of the mice indicated a strong left/ right hemispheric dominance of the pathology. Moreover, lateralized amyloidosis was positively associated in that study with ipsilateral increases in TSPO expression, which is a surrogate of microglial activation ( Sacher et al., 2020 ). However, whether such a hemispheric asymmetry in amyloid deposition and microglia activity is associated with specific cognitive deficits is unknown.
Performance in the Morris Water Maze (MWM) test is a sensitive indicator of spatial learning in mice ( Vorhees and Williams, 2014 ), and is particularly suited to examine AD-model mice because of its high specificity for hippocampal function ( Bryan et al., 2009 ). Spatial memory, as assessed by MWM, involve the integrated functioning of the hippocampus, the entorhinal cortex and associated structures such as the amygdala and the hypothalamus ( Vorhees and Williams, 2014 ). Importantly, these same brain regions have asymmetry in their anatomical structure and in certain molecular markers ( El-Gaby et al., 2015 ;Goto et al., 2010 ;Shinohara et al., 2008 ). This is in line with previous findings that the right hemisphere dominates in spatial learning and memory in rodents ( Shinohara et al., 2012 ;Shipton et al., 2014 ), as in most humans ( Thomason et al., 2009 ). Yet, there has hitherto been no investigation of possible associations between hemispheric-specific levels of AD-related brain changes and spatial learning performance in rodents.
Given this research gap, our aim was first to explore the regional associations of PET-assessed fibrillar amyloidosis and microglial activation with spatial memory performance in the novel App NL-G-F mouse AD model. Based on our previous results ( Ewers et al., 2020 ), we hypothesized that App NL-G-F mice with low fibrillar amyloidosis and high microglial activation would exhibit better spatial memory. Second, we defined regions of interest for brain regions implicated in spatial memory in order to assess hemispheric differences in the associations between A and TSPO-PET with spatial learning performance. We included the amygdala ( Maren, 1999 ;Sarter and Markowitsch, 1985 ), the entorhinal cortex ( Bannerman et al., 2001 ) and the hippocampus ( Bryan et al., 2009 ) as target regions of the spatial learning network. Given the reported dominance of the rodent right hemisphere in spatial learning tasks, we hypothesized a lateralization of correlations favoring the right side. Finally, we correlated left and right lateralization of the PET biomarkers with performance in the spatial learning task and validated our PET results by immunohistochemistry and biochemistry. We hypothesized that App NL-G-F mice with an asymmetry of fibrillar amyloid to the left and an asymmetry of microglial activation to the right would express better spatial learning performance.

Experimental design
All experiments were performed in compliance with the National Guidelines for Animal Protection, Germany, with the approval of the regional animal committee (Regierung Oberbayern) and oversight by a veterinarian. All animal experiments complied with the ARRIVE guidelines and were carried out in accordance with the U.K. Animals (Scientific Procedures) Act, 1986 and associated guidelines, EU Directive 2010/63/EU for animal experiments. Animals were housed in a temperature-and humidity-controlled environment with 12-hour lightdark cycle, with free access to food (Sniff, Soest, Germany) and water. Fifteen female and 15 male App NL-G-F mice were imaged by standardized A -PET and TSPO-PET protocols at a mean (SD) age of 10.3 ± 0.6 months (range: 9.7-12.8 months) ( Overhoff et al., 2016 ). Spatial learning was tested by an established Morris water maze (MWM) protocol ( Sacher et al., 2019 ) within two weeks of PET scanning. To avoid effects of anesthesia, we observed a minimum time gap of five days between the first or last day of spatial learning (including training and probe days) and the closest PET session. The mean (SD) time gap between the probe day and PET examination was 10.6 ± 2.3 days. Ageand sex-matched C57BL/6 wild-type controls were imaged by A -PET and TSPO-PET (each n = 18) and performed the same spatial learning task in MWM (n = 14). Biochemistry and immunohistochemistry were performed in a subset of App NL-G-F and wild-type mice.

Animals
The knock-in mouse model App NL-G-F carries a mutant APP gene encoding the humanized A sequence (G601R, F606Y, and R609H) with three pathogenic mutations, namely Swedish (KM595/596NL), Beyreuther/Iberian (I641F), and Arctic (E618G). Homozygotic App NL-G-F mice progressively exhibit widespread A accumulation first evident at two months of age ( Masuda et al., 2016 ;Saito et al., 2014 ). Deficits in spatial learning are not observed until six months of age ( Whyte et al., 2018 ) but appear in App NL-G-F mice ranging from 8-12 months of age ( Masuda et al., 2016 ;Sacher et al., 2019 ). There have been no reports of relevant neuronal loss and atrophy in this model. Age-and sex-matched C57Bl/6 wild-type mice (n = 18 for each PET modality) were reanalyzed together with App NL-G-F data and served as control material.

PET data acquisition, reconstruction and post-processing
For all PET procedures, radiochemistry, data acquisition, and image pre-processing were conducted according to established, standardized protocols ( Overhoff et al., 2016 ). In brief, 18 F-FBB A -PET recordings (average dose: 11.7 ± 2.1 MBq) with an emission window of 30-60 min after injection were obtained to measure fibrillar cerebral amyloidosis. 18 F-GE-180 TSPO-PET recordings (average dose: 11.4 ± 1.9 MBq) with an emission window of 60-90 min after injection were performed for assessment of cerebral TSPO expression. Isoflurane anesthesia was induced before tracer injection and maintained to the end of the imaging time window.

PET image analysis
All analyses were performed using PMOD (version 3.5; PMOD technologies, Basel, Switzerland). Static 30-60 min and 60-90 min image frames were co-registered to tracer specific templates (mixed App NL-G-F and wild-type templates) by a manual rigid-body transformation (TX rigid ) ( Overhoff et al., 2016 ). In the second step, we performed a reader-independent fine co-registration to the tracer specific templates ( Overhoff et al., 2016 ). Here, the initial manual fused images were normalized by non-linear brain normalization (TX BN ) via the PMOD brain normalization tool (equal modality; smoothing by 0.6 mm; nonlinear warping; 16 iterations; frequency cutoff 3; regularization 1.0; no thresholding). We then applied the concatenation of TX rigid and TX BN to obtain optimal spatial resampling with a minimum of interpolation. Normalization of emission images to standardized uptake value ratio (SUVR) images was performed using the previously validated mesencephalic periaqueductal gray (PAG, 20 mm 3 ) ( Sacher et al., 2019 ). Four ROIs were defined bilaterally in the amygdala (3 mm 3 each), the entorhinal/piriform cortex (5 mm 3 each), the hippocampus (2 mm 3 each) and the frontal cortex (12 mm 3 each, Supplemental Figure 1 ). Target to PAG SUVR values of A -PET and TSPO-PET were obtained for each bilateral region. The TSPO/A ratio was calculated by division of the SUVR values of both tracers in a single region to generate a quantitative index of microglial response relative to a given amount of fibrillary amyloidosis. For each bilateral region, the hemispheric asymmetry index (AI) was calculated for TSPO-PET, A -PET SUVR, and the TSPO/A ratio using the formula:

Behavioral testing
All 30 App NL-G-F mice and 14 age-and sex-matched C57BL/6 wildtype mice underwent a MWM test of spatial learning and memory deficits, which was performed according to a standard protocol with small adjustments ( Bromley-Brits et al., 2011 ) as previously described ( Sacher et al., 2019 ). In brief, the first test day served for acclimatization with the platform visible (5 min per mouse). Then the mice underwent five training days where each mouse had to perform four trials per day with the platform visible on the first training day and the platform hidden under water for all other training days. The test day entailed a single trial with complete removal of the platform. The maximum trial length on all training and test days was set to a maximum of 70 s. The video tracking software EthoVision R ○ XT (Noldus) was used for analyses of escape latency, the platform frequency, and attendance in the platform quadrant at the probe trial. Eight wild-type mice were assessed together with the App NL-G-F model, whereas six wild-type mice were examined two months apart and included to increase to statistical power of the control cohort. Aiming to increase the robustness of MWM results, we conducted a principal component analysis (PCA) of all single MWM readouts as indicated in the statistics section. Thus, we obtained a single quantitative index of MWM performance per mouse for correlation with PET imaging results.

Immunohistochemical and biochemical validation
Biochemical analyses were performed in samples from the entire forebrain in seven App NL-G-F mice (4 female, 3 male) and four wildtype mice (2 female, 2 male). Soluble Trem2 protein was extracted from brain tissue with Tris-buffered saline, and measured by ELISA, using polyclonal sheep antibody for coating (AF1729; R&D Systems) and biotinylated polyclonal sheep antibody (BAF1729; R&D Systems) together with streptavidin-horseradish peroxidase (N-100; ThermoFisher Scientific) for detection.
Brains intended for immunohistochemistry were fixed by immersion in 4% paraformaldehyde at 4 °C for 14 h. Nine App NL-G-F mice (5 female, 4 male) and four wild-type mice (2 female, 2 male) were analyzed. Two representative 50 μm-thick slices per animal were then cut in the axial plane using a vibratome (VT 1000 S, Leica, Wetzlar, Germany). Freefloating sections were permeabilized with 2% Triton X-100 overnight and blocked with I-Block TM Protein-Based Blocking Reagent (Thermo Fischer Scientific). We obtained immunohistochemical labelling of microglia using an Iba-1 primary antibody (dilution; Wako, 1:200), and the A-21244 secondary antibody (Invitrogen, 1:500). 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). Images were acquired on inverted confocal Laser-Scanning Microscope LSM780 (Zeiss). We imaged four different positions in the CA1-Region of the hippocampus of each slice 3-dimensional with a Plan-Apochromat 40x/1.4 Oil DIC M27 (Zeiss) objective The 3dimensional image stacks (2048 × 2048 × 123 pixels (lateral pixel size: 0.17 μm, axial pixel size: 0.40 μm), zoom 0.6, pixel dwell 0.79 μs) were maximum-intensity-projected. The Iba-1 load (area-%) was calculated as the summed area of all Iba-1 positive cells relative to the area of regions of interest using an automated threshold in ImageJ software.

Sample size estimation
An orienting sample size estimation (G * Power, V3.1.9.2, Kiel, Germany) was based on our previous data on PET imaging and spatial learning in App NL-G-F and wild-type mice ( Sacher et al., 2019 ). A t-test estimation with two independent means using = 0.05 and a power of 0.8 for A -PET in the hippocampus computed a minimal necessary sample size of n = 12 per group to detect the observed 3.2 % difference in signal (standard deviation per group: 0.025 SUVr) at 10 months of age between App NL-G-F and wild-type mice. A regression estimation (fixed model) using = 0.05 and a power of 0.8 for a partial R 2 of 0.2 (based on previous PET/MWM correlations) computed a necessary sample size of n = 27.

Statistical parametric mapping (SPM) group comparisons
For both tracers, whole-brain voxel-wise comparisons of PAG-scaled SUVR images between groups of App NL-G-F and wild-type mice were performed by an unpaired Student's t -test as described previously ( Rominger et al., 2013 ).

SPM regression analyses
In preparation for the correlation analyses between spatial learning scores and PET readouts, we extracted the principal component of the MWM test from three spatial learning readouts (escape latency, frequency of attaining platform, time spent in platform quadrant) using SPSS 25 statistics (IBM Deutschland GmbH, Ehningen, Germany). Prior to the PCA, we tested the linear relationship of the data by a correlation matrix, and discarded any items with a correlation coefficient < 0.3. We used the Kaiser-Meyer-Olkin (KMO) measure and Bartlett's test of sphericity to test for sampling adequacy and suitability for data reduction. Components with an Eigenvalue > 1.0 were extracted and a varimax rotation was selected. Subsequently, a voxel-wise regression analysis was performed in App NL-G-F mice using PAG-scaled SUVR images as the dependent variable and the principal component of spatial learning as independent variable. Age and sex served as covariates for the regression analysis. All SPM analyses were corrected for multiple comparisons by a false discovery rate (FDR) correction ( Benjamini and Hochberg, 1995 ). To avoid bias by regional constitution of single voxels ( Chumbley and Friston, 2009 ), we used a 3-dimensional grid of squared regions of interest (0.8 × 0.8 × 0.8 mm each) across the brain to obtain a topological FDR correction ( Chumbley and Friston, 2009 ). An FDR-corrected p -value of 0.05 was set as the threshold for voxel-wise analyses.

Region based regression analyses
SPSS was used for region-based analyses. Regression coefficients ( ) were computed between target region SUVR values of TSPO-PET or A -PET images or the TSPO/A ratio with the principal component of spatial learning performance using a multivariate model in App NL-G-F mice, controlled for age and sex. The association between the regional asymmetry index of the TSPO/A ratio and the principal component of spatial learning in App NL-G-F mice was computed by a multiple regression model together with age, sex and the absolute biomarker value per region (mean of both hemispheres). We applied an FDR-correction to account for multiple testing with predefined regions of interest. Next, we Highest levels of fibrillar amyloidosis were observed by A -PET in entorhinal and piriform cortices, as well as in the thalamus and the striatum, whereas TSPO-expression respresenting activated microglia was predominantly elevated in frontal and parietal cortices and in the hippocampus. ( B ) Few clusters in which A -PET uptake correlated negatively with spatial learning performance when assessed by voxel-wise regression analysis in App NL-G-F mice ( n = 30) and no clusters of positive correlation between A -PET uptake and spatial learning performance. ( C ) Positive correlation clusters between spatial learning performance with higher microglial activation were observed in the App NL-G-F mice ( n = 30) predominantly in the amygdala, the entorhinal and piriform cortices, the hippocampus, and the hypothalamus. All SPM T-scores represent a false discovery rate corrected p-value threshold of < 0.05. All analyses were controlled for age and sex. SPM -statistical parametric mapping, MRI -magnet resonance imaging, A -beta amyloid, TSPO -18 kDa translocator protein, EL -escape latency, FQ -frequency of platform crossings, QU -time spent in the platform quadrant, ( + ) -positive association, (-) -negative association, Reg. -regression analysis, R -right, L = left.
computed the association between sTrem2 or Iba-1 area-% and the principal component of spatial learning in App NL-G-F mice by a regression model together with age and sex. Single MWM readouts, sTrem2 and Iba-1 area-% values were compared between App NL-G-F and wild-type mice by an unpaired Student's t -test. A p -value of less than 0.05 was considered to be significant for rejection of the null hypothesis.

Data/code availability statement
Imaging data are available in nifti format and can be transferred upon request by the corresponding author. Behavioral, biochemical and immunohistochemical data are available in table format on Mendeley Data.

Microglial activation shows stronger associations with impaired spatial learning in App NL-G-F mice than did fibrillar amyloidosis
Voxel-based analyses of A -PET and TSPO-PET SUVR in App NL-G-F mice versus wild-type controls showed the expected widespread fibrillar amyloidosis and microglial activation throughout the brain of the transgenic mice ( Fig. 1 A). We found the highest levels of fibrillar amyloidosis in entorhinal and piriform cortices and the thalamus, whereas microglial activation was predominantly elevated in frontal and parietal cortices as well as in the hippocampus ( Fig. 1 A). Behavioral test results showed that App NL-G-F mice had prolonged escape latency (19.9 ± 12.8 s vs. 13.0 ± 6.7 s, p = 0.023) but only minor differences in platform crossings (3.2 ± 1.8 vs 4.0 ± 2.1, p = 0.219) or time spent in the platform quadrant (17.5 ± 6.2 s vs. 18.3 ± 3.4 s, p = 0.630) when compared to wild-type mice. Given the vulnerability of single MWM readouts to variability, we conducted a PCA to generate a more robust MWM performance index per mouse. The PCA resulted in a KMO sampling adequacy of 0.610 and a significant Bartlett's test of sphericity ( p = 0.001). We found one principal component with an Eigenvalue of 1.730, which explained 58% of the variance. The component matrix was 0.825 for platform crossings, -0.780 for escape latency and 0.664 for time spent in the platform quadrant.
There were only few spatially dispersed voxel clusters where fibrillar amyloidosis correlated negatively with spatial learning performance in App NL-G-F mice and no significant clusters of positive association between fibrillar amyloidosis and spatial learning performance ( Fig. 1 B). Voxel-wise TSPO-PET SUVr in the transgenic mice had strong positive correlations with spatial learning performance predominantly in the amygdala, entorhinal and piriform cortices, the hippocampus, and the hypothalamus ( Fig. 1 C). There were no relevant clusters indicating an association between high TSPO-PET SUVr and worse spatial learning ( Fig. 1 C).

Microglial activation is associated with better spatial learning predominantly in the right amygdala-entorhinal-hippocampal complex
For making detailed comparisons of hemispheric differences in spatial learning associations, we defined homologue regions of interest in brain areas known for their involvement in spatial learning tasks (amygdala, entorhinal/piriform cortex, hippocampus ( Vorhees and Williams, 2014 )), and in the frontal cortex. An elevated TSPO-PET signal in these predefined regions of App NL-G-F mice was associated with better spatial learning when controlled for age and sex ( Table 1 ) in the right amygdala ( = 0.481, p = 0.038) and the right entorhinal/piriform cortex ( = 0.578, p = 0.010). We observed no significant associations between TSPO-PET in the hippocampus and the frontal cortex with spatial learning. We next tested to see if the observed TSPO-PET associations were influenced by the individual regional burden of fibrillar amyloidosis, finding that the corresponding associations for the TSPO/A ratio were similar to those for the stand-alone TSPO-PET readout ( Table 1 ).
Correlation plots for the A -corrected TSPO-PET (TSPO/A ratio) with MWM escape latency are illustrated in Fig. 2 A. The observed associations were also discernible in individual mice with especially high or low TSPO-PET SUVr in the right amygdala-entorhinal-hippocampal complex (compare Figs. 2 B & 2 C). There were no significant associations of region-based fibrillar amyloidosis with spatial learning performance ( Table 1 ).

Asymmetric microglial activation in the amygdala is associated with spatial learning
Next, we asked if the lateralization of microglial activation in individual App NL-G-F mice has an independent impact on the spatial learning performance. To answer this, we computed a regression model with regional asymmetry (assessed by the AI) and spatial learning performance, controlled for the absolute individual amount of TSPO-PET quantification (mean TSPO/A ratio of both hemispheres), age and sex.

Immunohistochemistry and sTrem2 support a protective role of activated microglia on spatial learning
To validate our in vivo PET findings, we performed Iba-1 immunohistochemistry and measures of sTrem2 in brain sections and whole brain lysate. sTrem2 concentration was 4.2-fold higher in App NL-G-F mice when compared to wild-type mice (39.7 vs. 9.5 ng/ml, p < 0.001, Fig.  4 A). sTrem2 and the principal component of spatial learning showed a positive association in App NL-G-F mice after controlling for age and sex ( = 0.707, p = 0.038, Fig. 4 B). The Iba-1 area-% was 4.4-fold higher in App NL-G-F when compared to wild-type mice (10.6 vs. 2.4 %, p < 0.001, Fig. 4 C). The Iba-1 area-% and the principal component of spatial learning showed a positive association in App NL-G-F mice after controlling for age and sex ( = 0.705, p = 0.047, Fig. 4 D, E). In summary, we find congruent data between in vivo TSPO-PET and ex vivo methods.

Discussion
We present the first preclinical in vivo investigation aiming to elucidate regional associations of fibrillar plaque burden and TSPO expression with performance of a spatial learning task in an A mouse model. In App NL-G-F mice, we observed significant correlations of better MWM performance with high TSPO expression in brain regions matching the known networks of spatial memory processing. In contrast, there were distinctly weaker associations of spatial learning scores with fibrillar A burden. Furthermore, the associations between higher TSPO expression and better spatial learning were stronger in right hemispheric brain regions of App NL-G-F mice, and of much lower significance in the left hemisphere. The presence of lateralized TSPO expression in the amygdala predicted spatial learning performance independent of absolute TSPO expression and amyloidosis, and with adjustment for age and sex of the individual App NL-G-F mice.
We found an association between better spatial learning and high TSPO expression in the novel AD mouse model App NL-G-F . We have shown earlier that TSPO expression measured with 18 F-GE-180 PET provides a good surrogate of microglial activation in AD model mice as demonstrated by immunohistochemistry ( Overhoff et al., 2016 ;Parhizkar et al., 2019 ;Sacher et al., 2019 ). These cognitive/behavioral correlates of increased TSPO binding are in line with our preliminary data in a small cohort of PS2APP mice . There, we did not find direct associations of TSPO expression and spatial learning at a single time point, but higher early TSPO expression at 8 months predicted preserved spatial learning at 13.5 months. Our novel data in Microglial activation of App NL-G-F mice ( n = 30) was significantly associated with spatial learning performance in the right amygdala and in the right entorhinal and piriform cortices. Regression analysis was performed with adjustment for age and sex and with false discovery rate correction for eight regions of interest. The error band represents the 95% confidence interval ( B, C ) Water maze heat map comparison (residency on the test day) of representative individual App NL-G-F mice with low ( B: top ) and high ( C: bottom ) Microglial activation in the right amygdala-entorhinal-hippocampal complex. Distribution of fibrillar amyloidosis was similar in both mice. Dashed lines show the platform quadrant and dashed circles show the platform location during training days. TSPO -18 kDa translocator protein, PETpositron-emission-tomography, A -beta amyloid, s -seconds; R = right; L = left. a large group of ( N = 30) App NL-G-F mice aged 11 months implies that present or ongoing microglial activation in response to primary amyloidosis can be beneficial for brain function. Prima facia , one might have expected that elevated expression of microglia, which are the resident brain macrophages, would suggest a more fulminant and damaging inflammatory process. Thus, the present findings might be consistent with the polarized functionality of microglia, which have pro-and antiinflammatory phenotypes ( Chen et al., 2018 ), supporting a protective role of microglia on cognitive function when they express TSPO. Importantly, the present findings of an association between better spatial learning performance in AD-model mice and microglial activation were predominantly observed in the amygdala-entorhinal-hippocampal complex and related regions, which are directly involved in spatial learning. Since the corresponding structures together comprise a proportion of the rodent brain, it is not surprising that we were able to reproduce our PET observations by sTrem2 measures in whole brain lysate. However, the regions involved in spatial learning in the human brain are small in relation to whole brain volume, such that a similar association in AD patients might not manifest in measurable effects on the microglial markers such as soluble Trem2 in cerebrospinal fluid ( Suarez-Calvet et al., 2016 ). This highlights the advantage of region-specific microglia measures afforded by molecular imaging with PET. Iba-1 immunohistochemistry for activated microglia, which also showed a positive association between spatial learning performance with hippocampal microglial activation also validate our in vivo TSPO-PET results.
The increased TSPO expression in the right amygdala and the entorhinal/piriform cortex of App NL-G-F mice was of small magnitude relative to basal expression in wild-type mice, whereas the corresponding increases in frontal cortex and hippocampus were more pronounced. Nonetheless, increased fibrillar A binding was of equal or higher magnitude in the frontal cortex and the hippocampus (see Figure 1 A & Supplemental Figure 2). Longitudinal PET data in App NL-G-F mice indicated slower accumulation of fibrillar A in mice with high TSPO expression at baseline ( Ewers et al., 2020 ), which could indicate that a trait of high activation of microglia in the amygdala-entorhinal-hippocampal complex bodes well for an individual's capacity to cope with A accumulation. This fits with the observation of a cluster-based association Fig. 3. Regression analysis of asymmetry: Partial regression plots illustrate the association between the asymmetry index (AI) of TSPO-PET in different predefined brain regions and spatial learning (principal component, PC) of App NL-G-F mice ( n = 30). Regression analysis was performed with controlling for age, sex and the bilateral value of TSPO-PET. The TSPO/A ratio was used to control for fibrillary regional A levels. False discovery rate correction was performed for the four regions of interest. The error band represents the 95% confidence interval. R = right, L = left. between lower fibrillar A and longer time spent in the platform quadrant ( Fig. 1 ). The present cross-sectional design does not support causal inference, but the results may support the hypothesis that increased TSPO expression is a marker for enhanced microglial removal of A in App NL-G-F , with consequences for preserved brain function. However, we note there were only minor associations of the A -PET signal with other spatial learning readouts. This gives further evidence that the fibrillar A burden detectable by PET may have only limited impact on present cognition, which is also supported by human clinical data ( Jagust et al., 2009 ). We note that soluble or non-fibrillar A may yet impair spatial learning in A mouse models ( Huber et al., 2018 ).
The second goal of our study was to elucidate the functional impact of the asymmetric distribution of PET biomarkers, which we have amply documented in A PET studies of various AD mouse models ( Sacher et al., 2020 ). In this regard, we find clear predominance of right hemispheric associations between TSPO expression and spatial learning readouts. This fits with earlier reports that the right hemisphere is especially involved in spatial learning tasks for rodents ( Shinohara et al., 2012 ;Shipton et al., 2014 ), as is likewise the case in humans ( Thomason et al., 2009 ). The weaker effects observed for left hemispheric regions underpin the hypothesis that the protective effect of microglial activation on spatial learning performance is seemingly specific for the right hemisphere. In this regard, it would be of interest to evaluate if our present findings also hold for brain functions localized in the left hemisphere, such as verbal semantic memory in most humans ( Zahn et al., 2004 ) or auditory brain function in mice ( Ehret, 1987 ). Interestingly, the present pattern of association between TSPO-PET and spatial learning in App NL-G-F matches the preliminary findings in our pilot investigation of PS2APP mice . While that previous study in a small sample of (n = 10) PS2APP mice did not allow us to draw robust conclusions, the current data further strengthen the view that spatial learning performance is preserved when microglia are activated in the amygdala-entorhinal-hippocampal complex of the right hemisphere. Indeed, we saw the strongest behavioral impact of asymmetric TSPO expression in the amygdala. Here, we even found an association of the MWM endpoint with the lateralization of TSPO expression, independent from the absolute amygdaloidal TSPO expression, with correction for effects of amyloidosis, age and sex. Previous structural imaging studies indicate that amygdala volumetric asymmetry increasing in normal human aging ( Pedraza et al., 2004 ), which bears some relation with subjective cognitive decline ( Yue et al., 2018 ). Furthermore, the amygdala has important regulatory functions in memory tasks ( Roozendaal et al., 2004 ), which seemingly fits with the strong associations in App NL-G-F mice between spatial learning performance with asymmetric TSPO expression in amygdala. Notably, we did not see any relationship between asymmetric TSPO expression and asymmetric fibrillar amyloidosis in the amygdala and the entorhinal/piriform cortex of App NL-G-F mice, whereas there was a clearly positive correlation in frontal cortex and hippocampus (Supplemental Figure 2). This could potentially indicate a lesser coupling between amyloidosis and microglial activity in the inferior parts of the amygdala-entorhinal-hippocampal complex when compared to that in the cortex ( Sacher et al., 2020 ). This implies that unknown factors may mediate regional differences in the coupling between A overexpression and the microglial response. This phenomenon might be investigated using cell culture techniques or high-resolution immunohistochemical examination of the spatial association between polarized microglia and amyloid plaques.
One strength of our study lies in multimodal molecular imaging of AD biomarkers in conjunction with a behavioral test of spatial learning in a large cohort of App NL-G-F mice. Future studies of this type could include a longitudinal component to support causal inferences about the impact of baseline microglial expression on responses to amyloid overexpression or determine the biochemical thresholds sufficient for effects on cognition. In addition, one might test the effect of microglial activa- Fig. 4. Biochemical and Immunohistochemical validation: Elevated levels of sTrem2 in App NL-G-F mice (n = 7, A ) were positively associated with spatial learning performance ( B ). Similarly, increased Iba-1 immunoreactivity in the hippocampus of App NL-G-F mice ( n = 9, C ) was positively associated with spatial learning performance ( D ). Regression analysis was performed with controlling for age and sex. The error band represents the 95% confidence interval. ( E ) Individual App NL-G-F examples with low (left) and high (right) Iba-1 immunoreactivity in the hippocampus show deteriorated (left) and preserved (right) spatial learning performance. Scale bar = 20 μm. PC = principal component of Morris water maze (MWM); wt = wild-type. tion with lipopolysaccharide treatment on the progression of amyloidosis and cognitive changes in AD model mice. Ultimately, present results force us to consider that individual mice of the same genetic background are not clones or carbon copies, but that unknown factors determine their individual trajectories with respect to asymmetric expression of biomarkers and cognition. We did not include volumetric analyses in our study design since earlier reports did not reveal relevant brain atrophy in aging App NL-G-F mice ( Masuda et al., 2016 ;Saito et al., 2014 ). However, we note that hybrid imaging systems such as small animal PET/MRI ( Frost et al., 2020 ) could be of interest to study interaction effects between microglial activation, spatial learning deficits, and atrophy in models with neuronal loss, such as tau models. Among the limitations of our study, we note that spatial learning tasks as MWM have a high methodological variance ( Vorhees and Williams, 2006 ). I.e. hitting the place of the platform during first crossing of the basin by chance can lead to false low escape latency, whereas resting at the wall in the platform quadrant can cause spuriously long times spent in the platform quadrant. Therefore, we used a principal component analysis of common MWM readouts to improve the robustness of the data in single animals, as has been used previously in spatial learning evaluations of amyloid mouse models Williams, 2006 , 2014 ;Whyte et al., 2018 ). Present results will support power calculation for the design of subsequent intervention studies entailing a correlation analyses between MWM and PET endpoints, and we note that results in individual mice require cautious interpretation. We also note that A plaques in App NL-G-F mice have only low proportions of fibrillar A , which present the primary binding pocket of this and other fluorinated A PET tracers ( Catafau et al., 2016 ). Thus, A PET may not depict the most toxic component of the increased amyloid expression in App NL-G-F mice. Also, we used the hippocampus as a representative target region for immunohistochemistry validation of microglial activation in the amygdala-entorhinal-hippocampal complex, and note as a limitation of the present study that we did not separately access Iba-1 staining in the amygdala and the entorhinal cortex. The association between spatial learning score and microglial activation in the predefined hippocampus region did not survive correction for multiple comparisons, but the observed hippocampal clusters in the voxel-wise analysis lend credence of an immunohistochemistry validation.
TSPO-PET is not entirely specific for microglial activation, since activated astrocytes and other cells also express TSPO ( Venneti et al., 2013 ;Vivash and O'Brien, 2016 ). Furthermore, despite robust performance in rodents, translation of TSPO-PET towards human imaging is hampered by a polymorphism of the TSPO gene that causes different binding affinity among individuals ( Cumming et al., 2018 ;Mizrahi et al., 2012 ). These current limitations of TSPO-PET leave scope for the development of more specific radiotracers that could distinguish between disease associated and homeostatic microglia ( Götzl et al., 2019 ).

Conclusion
Elevated TSPO expression in the amygdala-entorhinal-hippocampal complex in App NL-G-F mice is associated with better spatial learning performance in the Morris water maze. This association was stronger for mice with right predominance of elevated TSPO expression in the amygdala. These findings indicate that asymmetric microglial activation is important to maintain brain function in the face of genetically-driven amyloid accumulation.

Data_statement
Imaging data are available in nifti format and can be transferred per request by the corresponding author. Behavioral, biochemical and immunohistochemical data are available in