All animal experiments were approved by the local animal care and use committee (LAVES, Niedersächsisches Landesamt für Verbraucherschutz und Lebensmittelsicherheit, Oldenburg, Germany) in accordance with the German animal protection law. Mice were maintained in temperature- and humidity- controlled environment (~ 22 °C, ~ 50%), 12 h light/dark cycle (light on at 7am) with food and water ad libitum. Cages were enriched with wood-chip bedding and nesting material (Sizzle Nest, Datesand). All experiments were performed by investigators unaware of group assignment (‘fully blinded’). Behavioral testing-order was balanced between groups prior to experiments and randomized within groups via random selection by the blinded investigator.
C57BL/6 mice bearing the tamoxifen-inducible diphtheria toxin chain A allele were generated by crossing homozygous Neurod6tm2.1(cre/ERT2)Kan (‘NexCreERT2’, ) with heterozygous Gt(ROSA)26Sortm1(DTA)Jpmb (‘Rosa26-eGFP-DTA’, ) resulting in double heterozygous inducible (‘DTA’) mice and heterozygous NexCreERT2 littermate (‘control’) mice lacking the DTA allele. Detailed PCR-based genotyping protocols are available upon request. Male transgenic mice were weaned at postnatal day 21 and separated by genotype to avoid inclusion effects or aggressive behavior against potentially affected animals. Experiments were performed on adult male mice (starting age 6–8 months) over a period of approximately 2 months.
Tamoxifen (CAS#10540-29-1 T5648, Sigma-Aldrich) was dissolved in corn oil (C8267, Sigma-Aldrich) on injection days at 10 mg/mL. Dependent on the experimental cohort, mice received either 3 or 5 daily intraperitoneal injections of 100 mg tamoxifen/kg body weight.
Blood sampling and high-mobility group box 1 (HMGB1) ELISA
Intermediate blood samples (100µL) were collected 2 weeks after the last tamoxifen injection from the retro-orbital sinus. Terminal blood (500µL) was sampled by cardiac puncture before transcardial perfusion. EDTA plasma aliquots were stored at −80 °C. Plasma HMGB1 concentrations were determined using a commercial HMGB1 ELISA assay (USC-SEA399MU-96, Biozol) according to the manufacturer’s instructions.
Experiments were performed during light phase in the following order: Open field, prepulse inhibition, Morris water maze, SocioBox and complex wheel running (the latter 2 only in 3 × tamoxifen mice).
To evaluate exploratory activity in a novel environment, mice were placed in the center of a gray circular Perspex arena (120 cm diameter, 25 cm height of outer wall). First, time to reach the outer wall (escape latency) was measured, followed by 7 min to freely explore the open field. When exceeding 180 s to reach the outer wall, mice were placed in the periphery zone to start 7 min exploration time. Behavior of the mice was recorded via tracking software (Viewer3, Biobserve). Analyzed parameters were covered distance and the time spent in the central, intermediate and peripheral zones of the open field. Animals were tested at the age of 27 (3 × tamoxifen) and 32 (5 × tamoxifen) weeks with light intensity of 120–140 lx in the center.
Prepulse inhibition (PPI) of the startle response
This paradigm has been described previously in detail . In brief, to evaluate sensorimotor gating, animals were placed in small metal cages to prevent major movements. Cages were placed in sound attenuating cabinets (TSE Systems) on a sensor-attached platform to record movement. After habituation to 65 dB white noise, loudspeakers delivered acoustic stimuli to evoke startle reflexes. Stimuli of different intensity (70, 75, 80 and 120 dB) were used in a pseudo-randomized order. The amplitudes of the startle response were averaged within the various intensities for each mouse. PPI was calculated as percentage of the startle response using following formula: %PPI = 100–[(startle response after prepulse)/(startle amplitude after pulse only) × 100]. For analysis, data from non-performing mice (negative PPI) were excluded.
Morris water maze (MWM)
This test has been described previously in detail [12, 31]. Briefly, to measure spatial learning and memory, mice were placed in a circular tank (120 cm diameter, 60 cm height) filled with opaque water at room temperature (~ 22 °C) with the escape platform (10 cm diameter) approximately 1 cm submerged. Animals’ movement was recorded by video-tracking system (Viewer3, Biobserve). After 2 days of a visual platform task in which extra maze cues were covered by the tank walls, the visual cue (flag on platform) was removed, the platform was relocated and the water height was adjusted so that extra maze cues were visible. For 8 consecutive days, mice had to reach the “hidden” platform in 4 trials per day. Afterwards, the platform was removed and mice were observed during a single “probe trial”. Analyzed parameters were escape latency to platform and covered distance. Additionally, during probe trial, the time spent in, visits and latency to the target quadrant (formerly containing platform) and covered distance were measured. If groups showed no significant differences, the platform was once again relocated for another 4 days of reversal learning, followed by another probe trial. Except for the probe trials, each day consisted of 4 test runs with a maximum of 90 s each and an intertrial interval of 5 min. In between trials, mice were placed in single cages containing paper towels and standing on a heating pad to prevent hypothermia and overexertion of the mice. In absence of a platform to reach, both probe trials consisted of a single test run of 90 s. Animals were tested at the age of 28–30 (3 × tamoxifen) and 33–35 (5 × tamoxifen) weeks with light intensity of 120–140 lx.
A detailed description of this test has been published before . Briefly, to evaluate social recognition and memory, 3 × tamoxifen mice were tested in the SocioBox at the age of 31–32 weeks with light intensity of 10–15 lx. They were placed in a circular apparatus (34 cm inner and 56 cm outer diameter) with 5 small boxes (“inserts”) within the outer wall. For each mouse, the SocioBox experiment consisted of 3 habituation sessions on 3 consecutive days and 2 exposures and 1 recognition trial on day 4. Within each trial, for the first 5 min, mice stayed in a circular partition (19 cm diameter) to prevent immediate exploring of the SocioBox (initiation stage). After lifting of this partition, mice were allowed to freely explore the SocioBox for another 5 min (interaction stage). While the 5 inserts in the outer ring were empty during all 3 habituation trials, they contained a mouse each for interaction purposes (“stimulus mice” or “stimuli”). For both exposure trials, the same 5 stimulus mice in the same position and order were used. For the final recognition test, one stimulus mouse was replaced by a new stimulus mouse (“stranger”), unknown to the test mouse. Perforated fronts of the inserts allowed limited interaction between test mouse and stimulus mice and front walls of each insert were exchanged after each trial. Age and sex matched C3H mice were used as stimuli, based on their reported robust social interaction in test situations . Mouse movement was recorded via tracking software (Viewer3, Biobserve).
This technique and data extraction/processing have been published previously in detail  and were slightly modified for the present application. In short, for SocioBox experiments (performed with 3 × tamoxifen mice), an A655sc infrared thermography camera (FLIR) was mounted 110 cm above the arena, recording images at 640 × 480 pixels and framerate of 2 Hz (habituation 3, exposure 1 + 2) and 5 Hz (memory test), via ResearchIR (FLIR Systems, Oregon, USA) and connected to a computer located in a separate room. Extraction of thermal data was done using OpenCV 4 in Python 3.6. Images were loaded and normalized to values between 0 and 255, with higher values meaning higher temperatures. The SocioBox arena in which the test mouse was allowed to move was defined as the relevant ROI for extracting thermal data. To keep only thermal data from the test mouse, a binary mask for whole body (including tail) was created by applying intensity thresholding and processing steps to decrease image noise. By doing so, one large cluster of connected pixels within this ROI could be detected, constituting the contour of the mouse. Due to the shape and temperature differences, the whole-body area could then be segmented into a central body and a distinct tail area, and the mean temperature of each of those 2 areas could be extracted. Analyzed parameters were temperature changes over time of test mice, latency and duration of interaction with stimulus mice and “strangers” and visits to interaction zones. The Centralization Index (ratio body/tail temperature = T ratio) was used as continuous temperature measure .
Complex wheel running
To stimulate neuronal activation-induced cFos expression in the hippocampus, mice were subjected to a complex running wheel (CRW) set-up for 4 h . Mice were single housed in type III cages (42 × 26 × 18 cm, Tecniplast), equipped with CRW (TSE-Systems) characterized by randomized omitted bars [26, 27]. Mice were habituated to the experimental room and CRW for 2 h prior to dark phase. After dark phase onset, voluntary running was automatically tracked for 4 h via Phenomaster software (TSE-Systems) and the total running distance per mouse calculated. Mice were perfused with Ringer and 4% formaldehyde/PBS, 30 min after removal from the CRW set-up.
Magnetic resonance imaging (MRI)
Mice (5 × tamoxifen) were anesthetized with ketamine and medetomidine (60 mg/kg and 0.4 mg/kg body weight), intubated and kept under 1.5% isoflurane by active ventilation with constant ventilation frequency of 85 breaths/min (Animal-Respirator-AdvancedTM, TSE-Systems). Inside the MR-System, mice were placed in a prone position with head fixed to a teeth and palate holder . All MR measurements were performed at magnetic field strength of 9.4 T (Biospec®, Bruker BioSpin MRI, Ettlingen, Germany) comprising the following methods and acquisition parameters: High-resolution T2-weighted images (2D Rapid Acquisition with Relaxation Enhancement (RARE), TE/TR = 55/6000 ms, 8 echoes, spatial resolution 40 × 40 × 300 µm3), magnetization-transfer (MT) weighted images for volumetric analyses (3D fast low angle shot (FLASH), TE/TR = 3.4/15.2 ms, flip angle 5°, Gaussian-shaped off resonance pulse (off-resonance frequency 7.5 ppm, RF power 6µT), spatial resolution 100 µm isotropic), measurements of blood perfusion by dynamic susceptibility contrast (DSC) MRI (2D RF-spoiled radial multi-echo FLASH : TE1,2,3 = 1/2.15/3.3 ms, TR = 9 ms, flip angle = 11°, spatial resolution = 150 × 150 × 900 µm3, 2 slices, 201 spokes, temporal resolution = 0.55 frames per second, 800 repetitions) and intra-voxel incoherent motion (IVIM) MRI (Stejskal-Tanner pulsed gradient spin-echo sequence, TE/TR = 19/2000 ms, 4 segments, δ = 2.5 ms, Δ = 10 ms, 19 b-values (10, 20, 30, 40, 50, 60, 70, 80, 110, 140, 170, 200, 300, 400, 500, 600, 700, 800, 900 s/mm2) applied in 3 orthogonal directions, spatial resolution 150 × 150 × 400 µm3).
MRI data analyses
MT-weighted images were first converted to NIfTI and preprocessed through denoising and bias field correction  in order to create an unbiased anatomical population template using the python pipeline twolevel_ants_dbm (https://github.com/CoBrALab/). Nonlinear deformation fields of the extracted brains were then used to estimate voxel-wise Jacobian determinants. Student’s t-test was performed on the 3D Gaussian-smoothed maps of Jacobian determinants (FWHM 0.1 mm) for voxel-vise comparison of DTA and control mice. Q-values (false discovery rate (FDR) adjusted p-values) and the respective z-scores were calculated using the 3dFDR function of AFNI . To visualize significant volume reductions in DTA mice z-scores smaller − 2.57 corresponding to a FDR of less than 1% were overlaid on the study template. In order to quantify the volume of selected brain regions, regions of interest (ROIs) including third and lateral ventricles, cerebrum (without hippocampus, ventricles and olfactory bulb), hippocampus, and cerebellum were determined on the study template by manual segmentation using the software package AMIRA (Visage Imaging GmbH, Berlin, Germany). ROIs were then retransformed into the subject space, individually inspected, and, if required, manually corrected. Finally, the respective volume information was extracted.
Diffusion weighted images were co-registered (translation only, imregister function, Matlab 2018b, Natock, MA) and the geometrical mean of the diffusion directions, the apparent diffusion coefficient (ADC) was calculated considering only those images obtained with b-values greater than 200 s/mm2. The ADC was then used to estimate the perfusion fraction (f) and the pseudo-diffusion coefficient (D*) by utilizing all acquired b-value-images .
Maps of R2*-relaxation rate (1/T2*) were estimated assuming a mono-exponential TE-dependent signal decay. A gamma variate function was fitted to the R2*-time curve of the contrast agent bolus-injection in order to estimate the cerebral blood volume (CBV) and the mean transit time (MTT) from which the cerebral blood flow (CBF) was calculated (CBF = CBV/MTT).
Measurements assessing blood–brain-barrier integrity
Blood–brain barrier (BBB) integrity along with brain water content and dry brain mass was evaluated as previously described [5, 34]. Briefly, mice received intravenous injections of Evans blue (50 µg/g body weight, E2129, Sigma-Aldrich) and sodium fluorescein (200 µg/g body weight, F6377, Sigma-Aldrich). After 4 h, mice were anesthetized and transcardially perfused. Brains were collected, frozen on dry ice, weighted and lyophilized. After lyophilization, dehydrated brains were weighted to obtain dry brain mass and to calculate brain water content. Afterwards, tracers were extracted from hemispheres with formamide and quantified in triplicates on a fluorescent microscope (Observer Z2, Zeiss). The concentrations of tracers were calculated using a standard curve and normalized to controls.
Blood flow cytometry
EDTA-blood was collected from the orbital sinus 2 weeks after last tamoxifen injection. Per mouse, 50µL blood were diluted in 50µL PBS and overlaid on 100µL lymphocyte separation medium (1077, PromoCell). After centrifugation, peripheral blood mononuclear cells were isolated, washed and stained for 15 min at 4 °C with the following antibodies: PECy5 anti-CD4 (1:1000, clone H129.19, Biolegend), PECy7 anti-CD8 (1:500, clone 53–6.7, Biolegend), BV510 anti-B220 (1:500, clone RA3-6B2, Biolegend) and PerCpCy5.5 anti-CD11b (1:100, clone M1/70, Biolegend). After staining, cells were washed, suspended in 200µL PBS containing 2% bovine serum albumin (#8076.3, Roth), and filtered through 40 µm cell strainers. Samples were measured on a FACSAria Sorp (BD). Number of lymphocytes were determined using forward and side scatter. Frequency of T-helper cells (CD4+,CD8−), cytotoxic T-cells (CD8+, CD4−), B cells (B220+, CD11b−, CD8−, CD4−), and myeloid cells (CD11b+, B220−, CD8−, CD4−) were determined as percentage of total lymphocytes.
Brain flow cytometry
Mice (3 × tamoxifen) were anesthetized with Avertin (1.36% 2,2,2,-tribromoethanol in ddH2O; 24µL/g body weight) and transcardially perfused with 40 mL Ringer solution (B.Braun). Brains were stored on ice in 10% fetal bovine serum (FBS, #10500–064 Thermo)/DMEM (#41965 Thermo) until all brains were collected. Olfactory bulbs and brain stems were removed and brains mashed through 70 µm cell strainers. To remove myelin, cells were suspended in isotonic Percoll (17-0891-01, GE Healthcare) to a final concentration of 30% and centrifuged. Cells were washed with FACS buffer (2% FBS, 10 mM EDTA in PBS) and filtered through 40 µm cell strainers. Fc-receptors were blocked for 10 min at 4 °C with anti-mouse CD16/32 antibodies (1:100, 14–0161, eBioscience). Cells were stained for 30 min at 4 °C with the following antibody mix: APC anti-CD45 (1:200, clone 104, BioLegend), PE anti-CD11b (1:200, clone M1/70, BioLegend), PECy5 anti-CD4 (1:1000, clone H129.19, BioLegend), PECy7 anti-CD8 (1:500, clone 53–6.7, Biolegend), APCCy7 anti-CD19 (1:200, clone 6D5, BioLegend), PerCP-Cy5.5 anti-CD138 (1:200, clone 281–2, BioLegend). After staining, cells were washed and suspended in 400µL FACS buffer. Per sample, 100µL APC quantification beads (#340487, BD) were added. Samples were measured on a FACSAria Sorp (BD). Cell numbers were corrected for the number of recorded APC beads. Leukocytes (CD45high, CD11b−), microglia (CD45low, CD11bhigh) and macrophages (CD45high, CD11bhigh) were quantified within single cell gate determined by forward and side scatter. CD4+ T-cells and CD8+ T-cells were quantified within leukocyte gate. CD19+ B-cells and CD138+ plasma cells were quantified in CD4− CD8− leukocyte gate.
Mice were anesthetized with Avertin, transcardially perfused with Ringer (B.Braun) and subsequently 4% formaldehyde/PBS. Brains were collected, post-fixed in 4% formaldehyde/PBS for 12 h, dehydrated in 30% sucrose/PBS for 48 h, embedded in optimal cutting medium (Tissue-Tek, #4583, Sakura) and frozen on dry ice. Frozen brains were cut into 30 µm coronal sections on a cryostat (CM1950, Leica) and stored at −20 °C in anti-freeze medium (25% ethylene glycol/25% glycerol/PBS). Quantifications were performed using 4–6 hippocampi from 2–3 sections per mouse. Sections were selected in regularly spaced intervals (every 300 µm) between Bregma coordinates −1.34 to −2.24 mm. Free-floating frozen sections were blocked and permeabilized for 1 h at RT with 5% normal horse serum (NHS, 26,050–088, Thermo) in 0.5% Triton X-100/PBS, incubated overnight at 4 °C with primary antibodies and subsequently stained with corresponding fluorescently-labeled secondary antibodies for 2 h at RT. Nuclei were stained for 10 min at RT with 0.2 µg/mL 4′,6-diamidino-2-phenylindole in PBS (DAPI, D9542, Sigma-Aldrich) and sections were mounted on SuperFrost®-Plus slides (J1800AMNZ, Thermo) with Aqua-Poly/Mount (#18606, Polysciences). The following primary antibodies were used: Mouse anti-GFAP (1:500, NCL-GFAP-GA5, NovoCastra), rabbit anti-Iba1 (1:1000, #019–19741, Wako), rabbit anti-cFos (1:1000, #226003, Synaptic Systems), guinea pig anti-parvalbumin (1:1000, #195004, Synaptic Systems). Corresponding secondary antibodies included: Alexa Fluor 555 anti-rabbit (1:1000, A21428, Thermo), Alexa Fluor 647 anti-mouse (1:1000, A31571, Thermo) Alexa Fluor 633 anti-guinea pig (1:1000, A21105, Thermo). For Fluorojade C staining (AG325, Sigma) of dying neurons, sections were incubated in 0.06% potassium permanganate solution for 10 min. Following a 1 min water rinse, tissue was transferred for 10 min to a 0.0001% solution of Fluorojade C, dissolved in 0.1% acetic acid. Slides were rinsed with ddH2O and dried at 60 °C. Overview images of whole brain sections were obtained on Eclipse-TI 2 epifluorescence microscope (Nikon), equipped with 4 × objective (4x/0.2 NA PLAN APO #MRD00045, Nikon). For quantification of Iba1+ cells and GFAP+ area fraction by densitometry, 1 µm thick optical sections of hippocampi were acquired as tile scans on a confocal laser scanning microscope (LSM 880, Zeiss), furnished with a 40 × oil objective (40x/1.4 NA Plan-APOCHROMAT, #420762–9900, Zeiss). For quantification of parvalbumin+ and cFos+ neurons, 2 µm thick optical sections of hippocampi were acquired as tile scans using the same microscope, equipped with a 20 × air objective (20x/0.8 Plan-APOCHROMAT, #420640–9903 Zeiss). Image acquisition parameters were kept constant within experiments. Quantifications and image processing were performed with FIJI-ImageJ software (Schindelin et al., 2012). Iba1+ cells (mostly microglia), parvalbumin+ cells (inhibitory neurons) and cFos+ neurons were manually counted. GFAP+ area fraction (GFAP+ area/total area of region of interest) was quantified densitometrically upon uniform thresholding and fold change to control animals was calculated. Atrophy in regions of interest (CA1, CA3, dentate gyrus) was determined through manual segmentation. Resulting areas were normalized to the respective average of control animals. CA2/CA3 region is referred to as CA3 in text and figures. Cell counts were normalized to quantified areas. Data obtained from 4 to 6 hippocampi/mouse was averaged for analysis.
Statistical analyses were performed using Prism9 software (GraphPad Software). Results are presented as mean ± standard deviations (SD), unless otherwise stated. Normal distribution of data was assessed using the Shapiro–Wilk test with an alpha error of 0.05. Dependent on data distribution, 2-tailed unpaired Welch’s corrected t-tests or Mann–Whitney U-tests were used to compare groups. Repeated measure data was analyzed using mixed-model ANOVA. P values < 0.05 were considered statistically significant.