Experimental design

This translational study used preclinical (n = 224) and human (n = 59) TSPO-PET imaging data, which were consistently analyzed with ICCs for assessment of regional synchronicity of the tracer signal (Fig. 1A and B). Mice with microglia depletion via colony-stimulating factor 1 receptor (CSF1R) inhibition using PLX5622 were analyzed in contrast to age-matched placebo controls as a proof-of-concept experiment. Regional synchronicity of TSPO-PET data was investigated in mice with microglia dysfunction and in AD mouse models of Aβ or tau pathology to recapitulate key AD-related brain changes. Human TSPO-PET scans were obtained from the ActiGliA cohort study [4], and we analyzed ICC in dependence of AD stages and cognitive performance. We performed single-cell radiotracing (scRadiotracing) after radiotracer injection with forebrains and hindbrains of wild-type (WT) and AD mice to determine the cellular sources of TSPO-PET tracer binding and the biological basis of regional signal desynchronization.

Fig. 1

Study design. In both mouse (A) and human (B) studies, TSPO-PET images were registered to a tracer specific template. Based on extracted mean values, inter-correlation-coefficients (ICCs) were calculated. In the human study (B), we additionally calculated a microglia synchronicity index (desynchronization index, DI) for each subject on a single volume-of-interest (VOI) basis (see detailed explanation further in text and in Fig. 9A). For each VOI, we compared the DIs between studied cohorts and correlated it with two cognition scores. C Overview of mouse cohorts. The numbers in green indicate the number of mice in corresponding cohorts. Light green color stands for mice with pharmacologically depleted microglia (PLX5622). WT = wild type; AD = Alzheimer’s disease; CTRL = healthy control

Animal experiments

The experiments have been approved by the local animal care committee of the Government of Upper Bavaria (Regierung Oberbayern, approval numbers: ROB-55.2–2532.Vet_02-15–210, ROB-55.2–2532.Vet_02-19–26), overseen by a veterinarian and in compliance 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. The ARRIVE essential ten checklist is provided with the manuscript. Animals were housed in a temperature- and humidity-controlled environment with 12 h light–dark cycle, with free access to food and water. Each mouse was treated as one individual. WT mice were used as controls. All the procedures were performed at the Department of Nuclear Medicine, LMU University Hospital, LMU Munich. An overview of all included small animal study groups is provided in Fig. 1C and Supplementary Table S1.

In particular, female WT (n = 28) (Fig. 2A), APPswe/PS2 (PS2APP, n = 20) (Fig. 2B), and APPswe/PS1deltaE9 (deltaE9, n = 16) (Fig. 2C) mice were used for pharmacological depletion of microglia by CSF1R inhibition [19] to test whether ICCs of TSPO-PET change when microglia are cleared from the brain. The mice received TSPO-PET at 6–9 (WT), 11.5 (PS2APP), or 5.5–6.5 (deltaE9) months of age and cages were randomly stratified into treatment (WT: n = 14; PS2APP: n = 10; deltaE9: n = 8) or vehicle (WT: n = 14; PS2APP: n = 10; deltaE9: n = 8) groups. CSF1R inhibition was performed by administration of PLX5622 (1200 ppm) orally in chow (Sniff Spezialdiaeten GmbH, Soest, Germany) for seven weeks (four weeks for deltaE9), and follow-up TSPO-PET scans were performed in the last week of treatment. For the WT and the PS2APP cohort, brain extraction for immunohistochemistry analyses was performed on the last day of treatment, and Iba1 staining was performed to validate microglia depletion as reported elsewhere [20].

Fig. 2

Mean TSPO-PET uptake (global mean-scaled) in each study cohort. Microglia depletion study in mice: A WT mice with microglia depleted by PLX5622 injection (WT PLX5622) and age-matched WT mice with placebo injection (WT Placebo) at 6–9 months of age, B Aβ mouse model (PS2APP) with microglia depleted by PLX5622 injection (PS2APP PLX5622) and age-matched mice with placebo injection (PS2APP Placebo) at 11.5 months of age, C Aβ mouse model (deltaE9) with microglia depleted by PLX5622 injection (deltaE9 PLX5622) and age-matched mice with placebo injection (deltaE9 Placebo) at 6 months of age. D Dysfunctional microglia study in mice: mice with deficient Trem2 gene (Trem2−/−) and age-matched WT mice at 12 months of age. E Dysfunctional microglia study in an Aβ mouse model (APPPS1): with intact Trem2 (APPPS1 Trem2+/+) and deficient Trem2 (APPPS1 Trem2−/−) at 12 months of age. F Study of mouse models at the onset of neuropathology: an Aβ mouse model (AppNL−G−F), a tau mouse model (P301S), and age-matched WT mice at 2–2.5 months of age. G Study of mouse models with moderate neuropathology: two Aβ mouse models (AppNL−G−F and APPPS1), a tau mouse model (P301S), and age-matched WT mice at 5–6 months of age. H Study of a mouse model of acute ischemic stroke: mice 7 days after photothrombotic surgery (Stroke) and sham surgery (Sham) at 2 months of age. I Human AD continuum study: subjects with prodromal AD, AD dementia, age-matched control subjects (CTRL test), and young control subjects used for calculation of the normal synchronicity (CTRL train). n represents the number of subjects; the mean age is shown on the bottom right of each image

To investigate the impact of dysfunctional microglia on microglia synchronicity, we analyzed TSPO-PET scans of female APPPS1 (APPPS1-21) mice [21] and age- and sex-matched WT with intact or deficient Trem2 (APPPS1 Trem2+/+, n = 11; APPPS1 Trem2−/−, n = 11; Trem2+/+, n = 17; Trem2−/−, n = 17) at 12 months of age (Fig. 2D and E) [22]. The presence of dysfunctional microglia and Aβ pathology was validated by CD68 and X-34 staining, respectively, in [22].

For investigation of microglial synchronicity at very early pathology stages, we compared AppNL−G−F mice (APP knock-in Aβ mouse model) and P301S mice (tau mouse model) with age matched WT mice at 2–2.5 months of age using a sample size of n = 12 TSPO-PET scans per group (all female) (Fig. 2F).

For investigation of microglial synchronicity at stages of moderate neuropathology, we compared AppNL−G−F mice, APPPS1 mice (transgenic Aβ mouse model) and P301S mice with age-matched WT mice at 5–6 months of age using a sample size of n = 14 TSPO-PET scans per group (all female) (Fig. 2G). Validation of early and moderate neuropathology was performed by Aβ staining for AppNL−G−F and APPPS1 mice in [15] and by AT8 staining for P301S mice in [23].

To study the impact of acute neuroinflammation on microglia synchronicity, we compared male WT mice (n = 6) at 7 days after photothrombotic stroke surgery to age- and sex-matched WT mice (n = 6) that underwent a sham surgery (Fig. 2H) [24]. The mice were 2–2.5 months of age at the time of the surgery. Photothrombotic stroke surgery was performed as described previously [24,25,26]. Briefly, mice were anesthetized with isoflurane (30% O2, 70% N2O) and placed in a stereotactic frame, maintaining body temperature at 37 °C. Dexpanthenol eye ointment was applied. Mice received 10 μl/g body weight of 1% Rose Bengal (Sigma Aldrich) in saline intraperitoneally 5 min before anesthesia. After a skin incision exposed the skull, the lesion site was marked in the left hemisphere (1.5 mm lateral, 1.0 mm rostral to bregma). A 2.0 mm diameter circular area was exposed to a 25 mV laser (561 nm, Cobolt Jive 50) for 17 min, 10 min post-Rose Bengal injection. The sham procedure was performed in the same fashion, but without laser illumination. Buprenorphin and Carprofen were administered for analgesia. Post-surgery, mice were placed in a 37 °C warming chamber for 20 min.

Each PET scan contained a mixture of different animal groups, i.e., transgenic mice and WT mice, using different randomized positions per scan. In each comparison group, we used an equal number of animals per cohort to ensure comparability of the results.

Animal PET imaging and analysis

All small animal positron emission tomography (μPET) procedures followed an established standardized protocol for radiochemistry, acquisition, and post-processing [27]. In brief, [18F]GE-180 TSPO μPET (TSPO-PET) with an emission window of 60–90 min post-injection was used to measure cerebral microglial activity. All analyses were performed by PMOD (V3.5, PMOD Technologies, Zurich, Switzerland) using a tracer-specific template for spatial normalization [28]. PET data were analyzed using a user-independent automatized brain normalization step to exclude operator bias in the PET data analysis. Normalization of injected activity was performed by global mean scaling, resulting in a standardized uptake value ratio (SUVR). Additionally, for each cohort, we defined a best reference region (BR) as the region that showed the lowest Cohen’s d between the transgenic/treated cohort and the corresponding WT/placebo cohort when comparing their global mean-scaled tracer uptake. Furthermore, the previously validated myocardium correction method for TSPO-PET [29] was used for validation by a brain-independent quantification method in the depletion experiment. μPET mean values were extracted from all volumes of interest (VOIs) of the Mirrione atlas [30] (excluding midbrain, inferior colliculi, hypothalamus, central grey matter, superior colliculi, and olfactory bulb), with further subdivision of the neocortical target region (visual, auditory, entorhinal, sensorimotor, and somatosensory cortex, all split into left and right) [31]. The final mouse VOI set consisted of 21 regions (9 bilateral). Cerebellum, brain stem, and right and left sensorimotor cortex VOIs were manually cropped to correct for signal spill-in from the brain ventricles and Harderian glands. All small animal PET experiments were performed with isoflurane anesthesia (1.5% at time of tracer injection and during imaging; delivery 3.5 L/min).

Human PET imaging

34 Aβ-positive (determined by a visual Aβ-PET read) patients across the AD continuum (17 prodromal AD consisting of 5 with subjective cognitive decline (SCD) and 12 with mild cognitive impairment (MCI), 17 AD dementia) and 12 age- and sex-matched cognitively normal controls (controls, CTRL) were enrolled from the ongoing interdisciplinary and prospective AD study "Activity of Cerebral Networks, Amyloid and Microglia in Aging and AD (ActiGliA)" [4]. 13 younger controls were additionally included as a training cohort (Fig. 2I). Detailed characteristics of all included subjects are shown in Supplementary Table S2. Controls were defined as subjects without objective cognitive impairment (clinical dementia rating [CDR] global score = 0, Consortium to Establish a Registry for Alzheimer’s Disease Neuropsychological Battery [CERAD-NB] total score ≥ 69), and no indication of Aβ pathology on PET (negative visual read) and/or CSF examination (Aβ42/40 ratio ≥ 5.5%). All subjects were scanned at the Department of Nuclear Medicine, LMU, using a Biograph 64 PET/CT scanner (Siemens, Erlangen, Germany). The Aβ-PET status was acquired in all ActiGliA subjects using [18F]flutemetamol Aβ-PET. PET acquisition and PET data analyses (ethics-applications: 17–569 & 17–755) were approved by the local institutional ethics committee (LMU Munich) and the German radiation protection (BfS-application: Z 5—22,464/2017–047-K-G) authorities. All participants provided written informed consent before the PET scans. Before each PET acquisition, a low-dose CT scan was performed for attenuation correction. Emission data of TSPO-PET were acquired from 60 to 80 min [32] after the injection of 191 ± 10 MBq [18F]GE-180 as an intravenous bolus. The specific activity was > 1500 GBq/μmol at the end of radiosynthesis, and the injected mass was 0.13 ± 0.05 nmol. Images were reconstructed using a 3-dimensional ordered subsets expectation maximization algorithm (16 iterations, 4 subsets, 4 mm gaussian filter) with a matrix size of 336 × 336 × 109, and a voxel size of 1.018 × 1.018 × 2.027 mm3. Standard corrections for attenuation, scatter, decay, and random counts were applied. TSPO-PET data were preprocessed using PMOD (PMOD Technologies, Zurich, Switzerland). Spatial normalization was performed to a tracer specific template in the Montreal Neurological Institute (MNI) space analogous to mouse data processing. All images were normalized by global mean scaling and smoothed with a Gaussian filter of 6 × 6 × 6 mm3 to account for intersubject differences in anatomy. Additionally, similarly to the mouse study, we defined a BR for the AD cohorts as the region that showed the lowest Cohen’s d between the prodromal AD / AD dementia cohort and the CTRL cohort when comparing their global mean-scaled tracer uptake. The Schaefer 200 atlas was used for region definition. VOIs belonging to the temporal and parietal lobe were defined in this work as AD-signature network regions (94 VOIs), while VOIs belonging to the motor and sensory cortex (27 VOIs) were defined as motor-sensory network and used as control regions, since the corresponding brain areas only minor affected by AD neuropathology (see detailed description in Supplementary Table S3).

TSPO binding status

Genotyping was performed at the Department of Psychiatry of the University Hospital LMU Munich and Regensburg. Genomic DNA was extracted from whole blood using a SQ Blood DNA kit von Omega Bio-Tek (Norcross, GA, USA) according to the manufacturer’s protocol. DNA quality was assessed by optical absorbance and gel electrophoresis. TaqMan quantitative polymerase chain reaction assays were used for amplification, Sager method for sequencing. Binding affinity of the [18F]GE-180 TSPO ligand is affected by the co-dominant rs6971 (Ala/Thr) single nucleotide polymorphisms (SNP) and needs to be considered in the imaging analysis [33]. Ala/Ala carriers are high-affinity binders (HAB), Thr/Thr carriers are low-affinity binders (LAB), and Ala/Thr carriers are mixed-affinity binders (MAB). Only HAB and MAB subjects were allocated for the PET-analysis, whereas n = 6 subjects with LAB status were excluded a priori.

Human assessments of cognitive performance

Neurocognitive testing was performed using the mini-mental status examination (MMSE) and the clinical dementia rating (sum of boxes) (CDR SOB). All neurocognitive testing was performed by trained psychologists.

Assessment of TSPO-PET ICC

ICC assessment was performed for all the studied groups using a modified version of the Python script reported in Grosch et al. [34]. In brief, the code calculates Pearson’s correlation [35] for each VOI pair and displays the Pearson’s R values as a 2D array, which additionally can be filtered based on the p- and R-value of the correlations.

In this work, to obtain robust ICC estimates, we additionally performed bootstrapping by resampling with replacement, i.e.:

Each subject had an equal probability to be included in a resampled dataset,

Each subject could be included in a resampled dataset more than once,

The size of each resampled dataset equaled the size of the original dataset.

10,000 resampled datasets were generated for each mouse cohort and 1000 for each human cohort. For each resampled dataset, we calculated Pearson’s R for all the VOI pairs. The R values were then transformed using Fisher’s R to Z transformation [36]:

$$Z=\text\left(R\right)$$

to normalize the distribution of the R values obtained from the resampled datasets. The mean Z of the 10,000 resampled datasets (1000 for human data) is further referred to as ICC. ICC values > 0.5 and < -0.5 with p < 0.005 (mouse cohorts) and p < 0.001 (human cohorts) were defined as significant connections. A connection between two cortical VOIs was defined as a cortical connection, between two subcortical VOIs as a subcortical connection, between a cortical and a subcortical VOI as a cortical-subcortical connection.

Individual assessment of microglia desynchronization

To assess microglia desynchronization in AD continuum patients relative to a cognitively normal human cohort on a single subject level, we introduced a quantitative score that we called desynchronization index (DI). DI of a single VOI was estimated as follows (Fig. 9A and Fig. 1B):

1.

For 13 CTRL subjects of the young training cohort, we extracted mean [18F]GE-180 uptake values of all the AD-signature network VOIs.

2.

Using these values, we calculated a linear fit between each VOI pair (mean of 10,000 bootstraps), representing the ICC.

3.

For an individual subject to be assessed, mean [18F]GE-180 uptake values of the same VOIs were extracted.

4.

For each VOI pair, we calculated the perpendicular distance d from the subject’s pair of values to the corresponding linear fit calculated in step 2.

5.

The DI of VOI i was defined as a sum of the perpendicular distances of all the pairs with this VOI:

where N is the number of VOIs.

DIs were calculated for 12 CTRL, 17 prodromal AD, and 17 AD dementia subjects. All the above-mentioned calculations were performed using a custom-made Python script (Numpy and Pandas libraries). As a control analysis, the same calculations were additionally performed using the motor-sensory network VOIs instead of the AD-signature network VOIs.

DI was then additionally calculated for all the investigated mouse cohorts. The calculation steps were identical to human, with a single difference that, to calculate the DI in the reference cohorts (listed in Supplementary Table S4), we used the leave-one-out approach: in step 1, all the subjects except for one test subject were used and in steps 3–5 the DIs were calculated for this one left-out subject; the procedure was then repeated for all the cohort subjects.

To compare the proposed DI analysis to the conventional SUVR analysis, we calculated the average SUVR and DI for each region across all subjects in both the mouse and human cohorts, followed by division of these average values by the corresponding average values in a reference cohort (Supplementary Table S4). These ratios were then compared between SUVR and DI. A higher ratio indicates a greater ability of either SUVR or DI to distinguish between normal and abnormal cohorts.

Single-cell radiotracing

App NL−G−F (n = 8) and wild-type (n = 8) mice at an age of 6–7 months underwent scRadiotracing after TSPO tracer injection [37, 38]. Dedicated procedures for mouse models of neurodegenerative diseases were described previously [39] and modified as listed below. Forebrains and hindbrains were processed separately. Adult Brain Dissociation Kit (Miltenyi Biotec, 130–107-677) was used for brain dissociation according to the manufacturer's instructions. Adult mouse brains were dissected, briefly washed with PBS (Gibco), cut into small pieces, and dissociated with enzyme mix 1 and 2 using gentleMACS™ Octo Dissociator (Miltenyi Biotec, 130–096-427). The dissociated cell suspension was applied to pre-wet 70 µm cell strainer (Miltenyi Biotec, 130–110-916). The cell pellet was resuspended using cold PBS and cold debris removal solution. Cold PBS was gently overlaid on the cell suspension. After centrifugation at 3,000 × g and 4 °C for 10 min with acceleration at 9 and deceleration at 5, the two top phases were removed entirely. The cell pellets were collected and resuspended in 1 mL cold red blood cell removal solution followed by 10 min incubation. Cell pellets were collected for positive isolation of microglia or astrocytes, negative isolation of neurons, and proportion analyses of endothelial cells, oligodendrocytes and remaining cells (Fig. 9A). 5% of the single cell suspension were used to characterize proportions of successfully collected microglia, astrocytes, and neurons.

Microglia were isolated from the single cell suspension using CD11b MicroBeads, human and mouse (Miltenyi Biotec, 130–049-601) and a MACS separation system (Miltenyi Biotec). The prepared cell pellets were resuspended in 90 µL of D-PBS/0.5% bovine serum albumin (BSA) buffer per 107 total cells. In total, 10 µL of CD11b MicroBeads per 107 total cells were added and incubated for 15 min in the dark at 4 °C. Cells were washed by adding 1—2 mL of buffer per 107cells and centrifuged at 300 × g for 5 min. The cell pellets were resuspended in 500 µL of D-PBS/0.5% BSA. The pre-wet LS columns (Miltenyi Biotec, 130–042-401) were placed onto a QuadroMACS Separator (Miltenyi Biotec, 130–090-976). The cell suspensions were applied onto the column. The columns were washed 3 times with 3 mL of D-PBS/0.5% BSA buffer. The flow-through containing the unlabelled cells was collected as the microglia-depleted fraction. The columns were removed from the magnetic field, and microglia were flushed out using 5 mL of D-PBS/0.5% BSA buffer.

Astrocytes were isolated from the CD11b(-) fraction using ACSA2 MicroBeads (Miltenyi Biotec, 130–097-678) and a MACS separation system (Miltenyi Biotec). The prepared cell pellets were resuspended in 80 µL of AstroMACS separation buffer (Miltenyi Biotec, 130–117-336) per 107 total cells. In total, 10 µL of FcR blocking reagent were added and incubated for 10 min in the dark at 4 °C. Afterwards, 10 µL of anti-ACSA2 MicroBeads were added and incubated for 15 min in the dark at 4 °C. Cells were washed by adding 1 mL of AstroMACS separation buffer and centrifuged at 300 × g for 5 min. Cell pellets were resuspended in 500 µL of AstroMACS separation buffer. The pre-wet MS columns (Miltenyi Biotec, 130–042-201) were placed at OctoMACS Separator (Miltenyi Biotec, 130–042-109). The cell suspensions were applied onto the column, followed by washing 3 times with 500 µL of AstroMACS separation buffer. The flow-through was collected containing non-astrocytic cells as an astrocyte-depleted fraction. The columns were removed from the magnetic field, and the astrocytes were flashed out using 3 mL of AstroMACS separation buffer.

Neuronal isolation was performed in the CD11b(-)/ACSA2(-) fraction. Neuron Isolation Kit, mouse (Miltenyi Biotec, 130–115-390), was used according to the manufacturer's instructions. The prepared cell pellets were resuspended in 80 µl of PBS-0.5% BSA buffer per 107 total cells. 20 μL of Non-Neuronal Cells Biotin-Antibody Cocktail was added and incubated for 5 min in the dark at 4 °C. Cells were washed and centrifuged at 300 × g for 5 min. Cell pellets were again resuspended in 80 μL of PBS-0.5% BSA buffer per 107 total cells. 20 μL of Anti-Biotin MicroBeads were added and incubated for 10 min in the dark at 4 °C. The volume was adjusted to 500 µl per 107 total cells with PBS-0.5% BSA buffer and then proceed to magnetic separation. The pre-wet LS columns (Miltenyi Biotec, 130–042-401) were placed at QuadroMACS™ Separator (Miltenyi Biotec, 130–090-976). The cell suspensions were applied onto the columns. The columns were washed with 2 × 1 ml PBS-0.5% BSA buffer. The flow-through containing the unlabelled cells were collected as the neuron-enriched fractions. The columns were removed from the magnetic field, and the non-neuronal cells were flushed out using 3 ml of PBS-0.5% BSA buffer.

Gamma emission measurements

CD11b(+) microglia-enriched, ACSA2(+) astrocyte-enriched, neuron-enriched and depleted fractions were analysed to determine radioactivity of harvested cells. Radioactivity concentrations of cell pellets were measured in a high sensitive gamma counter (Hidex AMG Automatic Gamma Counter) relative to the injected activity, with decay-correction to time of tracer injection for final activity calculations.

Flow cytometry

Flow cytometry staining was performed at 4 °C. Each microglia-enriched and astrocyte-enriched cell pellet was resuspended in 100 µL of cold D-PBS containing fluorochrome-conjugated antibodies recognizing mouse CD11b or ACSA2 (Miltenyi Biotec, 130–113-810 and 130–116-247) in a 1:50 and 1:9 dilution respectively and incubated for 10 min at 4 °C in the dark for quality control. Neuron-enriched and depleted were resuspended in 45 µl of D-PBS/0.5% BSA buffer and fluorochrome-conjugated antibodies recognizing CD90.2, CD31 and O4 (Miltenyi Biotec, 130–123-066, 130–123-813 and 130–117-711) were added in a 1:9 dilution and incubated for 10 min at 4 °C in the dark. Samples were washed with 1 mL of D-PBS and centrifuged for 5 min at 400 × g. Finally, cell pellets were resuspended in 500 µL of D-PBS and samples were used for flow cytometry using a MACSQuant® Analyzer. Acquired data included absolute cell numbers and purity of enriched CD11b( +) and ACSA2( +) cells in microglia/astrocyte-enriched samples as well as proportions of CD90.2 (neurons), CD31 (endothelial cells) and O4 (oligodendrocytes) for neuron-enriched and depleted samples.

Single-cell RNA-sequencing in APP23 and WT mice

Single-cell RNA(scRNA) sequencing was performed for microglia isolated from cortex, hippocampus, and cerebellum of male C57BL/6 J (WT) and APP23 mice of 17 and 27 months of age, as previously described [40]. In brief, freshly isolated brain regions were homogenised mechanically, at 4 °C, and myeloid cells were separated using a Percoll gradient. Cells were stained with antibodies against CD45 and CD11b as well as the amyloid-binding dye Methoxy-X04 (MX04, abcam), which labels plaque-associated microglia that have taken up aggregated amyloid-β [41]. Microglia were identified as CD11bhigh/CD45intermediate, and plaque-associated cells as MX04positive (MX04 +) by flow cytometry, using a Sony SH800Z FACS sorter. Single microglia were deposited into 384-well plates and sequenced using SmartSeq2 chemistry. After quality control, data analysis of high quality cells was performed using the R package Seurat v3 and v4 [42] and Wilcoxon rank sum test with Benjamini–Hochberg correction for pairwise statistical comparison of TSPO expression in microglia in the vicinity (MX04 +) and distant (MX04-) from plaques.

Statistical analysis

A simulation analysis was conducted to analyze appropriate cohort size and data robustness for ICC analysis. A random number generator was used to reduce the cohort sizes by one mouse at a time, and a new ICC matrix was calculated for each cohort. For this analysis we used an FDG-PET cohort of P301S mice and matching WT controls. The root mean square error (RMSE) was calculated (RMSE = √(∑(ICCP301S – ICCWT)2/ nICC). Subsequently, the sample size-dependent effect size was calculated for individual regions using Cohen’s d and we observed robust results of ICCs at a sample size > 11 mice. Unless otherwise specified, all statistical analyses were performed using Scipy and Pingouin libraries (Python 3.7). To compare median absolute ICCs between cohorts, Wilcoxon signed-rank test (two-tailed) was used, as the underlying ICC distributions were shown to be non-Gaussian, the corresponding results are reported as Q2 (Q1, Q3), where Q2 is the second quartile, or the median, Q1 is the first quartile, Q3 is the third quartile. p < 0.05 was used as significance threshold. To test for differences in DI between the human cohorts, one-way analysis of variance (ANOVA) (numerator degrees of freedom = 2, denominator degrees of freedom = 43, significance threshold p < 0.05) and unpaired t-tests (two-tailed, significance threshold p < 0.05) were used. The normality of the DI distributions was assessed using Shapiro–Wilk test [43]. Based on the DIs from the VOIs where a significant difference in DI between the cohorts was reported, we performed a principal component analysis (PCA) (Scikit-learn library, Python 3.7) and tested for differences in the first principal component (PC1) between the cohorts by means of one-way ANOVA and unpaired t-tests. To assess the relationship between the DI and the MMSE and CDR SOB score, we calculated Pearson’s R and the p-value of the correlation. Benjamini–Hochberg false-discovery rate (FDR) adjustment [44] was applied to these p-values. pFDR < 0.05 was used as a threshold for significant correlations. Additionally, we correlated PC1 to the above-mentioned scores. To analyze differences between SUVR and DI in discrimination of mouse transgenic/treated cohorts versus WT/placebo cohorts as well as human subjects with AD versus CTRL subjects, we performed paired t-tests (two-tailed, significance threshold p < 0.05). To evaluate differences in cellular TSPO tracer uptake and proportion of isolated cell fractions between WT and AppNL−G−F mice, we performed an unpaired t-test, p < 0.05 was used as significance threshold. Multiple regression was used to determine contributions of neurons (CD90.2), endothelial cells (CD31), oligodendrocytes (O4) and remaining cells (negative for all markers) to the radioactivity in the cell pellet.