Prior to initiating studies, all animal care and experimental protocols were approved by the Committee on Animal Experimentation, Faculty of Veterinary Medicine, Hokkaido University (No. 19–0079), which was awarded Accreditation Status by the Association for Assessment and Accreditation of Laboratory Animal Care International (AAALAC International). C57BL/6 N mice were obtained from CLEA Japan (Tokyo, Japan). All mice were maintained at 22 ± 4°C under a 12/12 h light/dark cycle (7 am–7 pm) with ad libitum food and water access. Male mice (11–14 weeks) were used for experiments.

LPS (#L2630, Escherichia coli O111:B4) was purchased from Sigma-Aldrich (St. Louis, MO, United States). A stock solution (10 mg/mL) was prepared in distilled water and stored at −20°C until injection. Mice received a single-dose intraperitoneal (i.p.) injection of LPS (1 or 3 mg/kg body weight) in saline (10 mL/kg body weight) at 7 pm (at the end of the light cycle). Mice were used for each experiment after the recovery periods outlined for specific experiments described as on days 0–7.

All behavioral tests and measurements of body weight and water intake were conducted at 7 pm. We selected seven behavioral tests commonly used in LPS-injected mouse models to evaluate brain functions and abnormalities such as spontaneous activity, anhedonia, anxiety-, depression-like behavior, and memory (Yin et al., 2023). In all behavioral tests, all apparatuses were wiped with 70% EtOH after each mouse use to prevent smell from affecting animal behavior.

A ToxinSensor™ Chromogenic LAL Endotoxin Assay Kit (GenScript, NJ, United States) was used to quantify endotoxin levels of the plasma and brain. Mice were anesthetized by inhalation of isoflurane (Pfizer, New York, NY, United States). Blood samples were taken from anesthetized animals by direct cardiac stick and mixed with EDTA. After perfusion with 10 mL saline, the whole brain was isolated, weighed, and homogenized in 1 mL distilled water. Blood and brain samples were centrifuged at 1,500 × g, and supernatants were used as samples. Endotoxin levels were quantified according to manufacturer’s instructions. All equipment used for the experiment was endotoxin-free. Endotoxin levels were quantified as endotoxin units (EU) in plasma volume (EU/mL) and as EU per brain wet weight (EU/g).

BBB permeability was evaluated using Evans blue dye, which binds to albumin in blood, as described (Manaenko et al., 2011). Evans blue (2% w/v) was injected i.p. (4 mL/kg body weight), and 3 h later, mice were anesthetized by inhalation of isoflurane and perfused with 10 mL saline. Then, the whole brain was isolated, weighed, and homogenized in 100% trichloroacetic acid. The brain sample was incubated at 4°C for at least 12 h and then centrifuged at 500 × g, and the supernatants were used for measurement of absorbance at 610 nm using a microplate reader (SH-1000; Corona Electric, Hitachinaka, Japan). Brain Evans blue amounts were calculated using a standard curve and normalized to brain wet weight (g Evans blue/g wet brain weight). Brain dry weight was measured after the wet brain was incubated at 60°C for at least 12 h. The water content (%) was determined from the difference in the wet and dry weights divided by the wet weight.

Mice were anesthetized by inhalation of isoflurane and perfused with 4% paraformaldehyde (PFA) in phosphate-buffered saline (PBS). Whole brains were isolated and fixed with 4% PFA overnight at 4°C. Brains were then glued to a slicer stage (LinearSlicer Pro7, Dosaka EM, Kyoto, Japan), flooded in PBS, and cut into 50 μm coronal slices. Slices were blocked with a blocking buffer composed of 10% goat serum, 0.5% Triton X-100, and 0.05% sodium azide in PBS at room temperature (RT) for 2 h. Slices were incubated with primary antibodies (Supplementary Table S1) at 4°C for at least 12 h, and then with Alexa Fluor 488- (#A11008, 1:500, Thermo Fisher Scientific, MA, United States) and/or 555- (#A21422, 1:500) conjugated goat anti-rabbit or mouse antibody at RT for 90 min. Slices were mounted onto glass slides with DAPI-Fluoromount G (SouthernBiotech, Birmingham, AL, United States). Images were captured using a laser scanning confocal microscope (LSM 700, Carl Zeiss, Oberkochen, Germany) with a 40 × lens objective. The images were comprised of 30 μm Z-stacks consisting of 31 optical slices of 1 μm thickness by maximum intensity projection. The mean intensity or areas of GFAP, Iba-1, and CD68 were analyzed by ImageJ software (National Institutes of Health, MD, United States) and normalized to the control, which was arbitrarily set to a value of “1.0.”

PathoGreen™ Histofluorescent Stain (Biotium, CA, United States) was used to detect neuronal degeneration. Experiments were performed according to manufacturer’s instructions. Briefly, the brain was perfused and fixed with 4% PFA overnight at 4°C. Then, the brain was cut into 50 μm coronal slices, which were dried at 60°C for 30 min. Slices were incubated with 0.06% KMnO₄ and 1,000 × PathoGreen™ stock solution at RT for 10 min each. After air drying, slices were incubated with xylene and mounted onto glass slides. Images were captured using a laser scanning confocal microscope with a 40 × lens objective.

Golgi-Cox staining was performed as described previously, but with partial modification (Bayram-Weston et al., 2016). The brain was perfused and fixed with 4% PFA overnight at 4°C, and incubated with a solution consisting of 1.04% Hg₂Cl₂, 1.04% K₂Cr₂O₇, and 0.83% K₂CrO₄ in distilled water at RT for 2 weeks. The brain was incubated with 25% sucrose at RT for 6 h and subsequently cut into 100 μm coronal slices. After air drying, slices were incubated with 28% ammonia for 10 min. Slices were then incubated with xylene and mounted onto glass slides. Images were captured using a fluorescence microscope (BZ-9000, KEYENCE, Osaka, Japan) with a 100 × lens objective. Morphological evaluation of dendritic spines was conducted automatically using the application “Reconstruction¹” as described (Risher et al., 2014). Five neurons were randomly selected from slices obtained from one mouse, and one dendrite >30 μm was measured per neuron. The average of the dendritic spine values per mouse was used as an independent data point.

Morphological features of astrocytes and microglia were assessed using confocal Z-stack images of GFAP and Iba-1 staining, respectively. Astrocytes were analyzed using SMorph as described previously (Sethi et al., 2021), and microglia were analyzed using ImageJ software as described (Young and Morrison, 2018). All glial cells in one image (approximately 60 cells) per mouse were analyzed. Principal component (PC) analysis was performed using the statistical analysis software JMP® 17 (SAS Institute, Inc., Cary, NC, United States) for astrocytic and microglial morphology using the morphological parameters shown in Supplementary Figures S2, S3.

Total RNA was extracted from whole hippocampi using RNAiso Plus (Takara Bio, Tokyo, Japan). To remove genomic DNA and synthesize cDNA, RNA samples were then incubated with qPCR RT Master Mix with gDNA Remover (TOYOBO, Osaka, Japan). Real-time PCR was performed using Thunderbird SYBR qPCR Mix (TOYOBO), oligonucleotides, and the cDNA reaction solution. Primer sequences are provided in Supplementary Table S2. Thermal cycles were performed using an Eco Real-Time PCR System (Illumina, San Diego, CA, United States). Cycling conditions were 95°C for 1 min (initial denaturation), followed by 40 cycles of denaturation (95°C, 15 s), annealing, and extension (63°C, 30 s). RNAs without reverse transcription were used as negative controls to determine if DNA contamination was present, as indicated by lack of amplification by real-time PCR. Melt curve analysis confirmed that only one amplicon was produced by each reaction. The expression levels of factors relative to Gapdh were calculated using the ΔΔCq method and experimental results were normalized to the control, which was arbitrarily set to a value of “1.0.”

All studies were designed to generate groups of equal size using randomization and blinded analysis. Data are expressed as means ± S.E.M. (n = number of independent measurements) of at least five independent experiments (biological replicates). Statistical comparisons between two groups were conducted using the unpaired Student’s t-tests. For multiple comparisons, one-way ANOVAs followed by the Dunnett’s test or Tukey’s test were used. Post-hoc tests were performed only if F achieved p < 0.05 and there was no significant inhomogeneity of variance. p < 0.05 was considered statistically significant. All statistical analysis was performed using the statistical analysis software JMP® 17.

We first conducted behavioral tests in mice 24 h after LPS injection, as reported in many previous studies. LPS injection (1 mg/kg, i.p.) decreased time spent in center, increased total freezing time in the open field test, and increased total immobility time in the forced swimming test (Figures 1A,B). In addition, sucrose preference decreased during the first 24 h following LPS injection (Figure 1C), but notably, body weight and water intake decreased significantly (Figures 1D,E). Total distance traveled in the open field test, an indicator of spontaneous activity, also decreased (Figure 1A). These results were indicative that the anxiety- and depression-like behavior observed in these behavioral tests could reflect the effects of systemic symptoms.

Next, we examined whether mice had recovered from systemic symptoms 7 days after LPS injection (3 mg/kg, i.p.). Body weight and water intake decreased after LPS injection, which was lowest on day 1 or 2 but recovered by day 7 (Figures 2A,B), and LPS injection 7 days prior did not affect the total distance traveled in any behavioral tests (Figures 2D,E, 3A; Supplementary Figures S1B,C). These results suggested that LPS-injected mice recovered from systemic symptoms on day 7. We then conducted multiple behavioral tests on day 7. LPS injection 7 days prior did not affect the sucrose preference test, forced swimming test, open field test, or elevated plus maze test, which were conducted to evaluate anhedonia-, depression-, and anxiety-behaviors (Figures 2C–E). Prior LPS injection also did not affect the results of the Y-maze test, which was conducted to evaluate working memory (Figure 3B). Next, NORT and OLRT were conducted to evaluate short-term memory. In NORT, the control group exhibited a significant increase in the percentage of exploration time of the novel object in the test section, compared with that of the original object in the training section (Figure 3C). Meanwhile, the LPS group exhibited no increase in the percentage between original and novel objects. There was also a significant difference in the percentage between the control and LPS groups in the test section. Furthermore, we measured the time difference between the first exploration of the novel object and the original object by mice in the test section. The control group explored the novel object earlier than the original object, but the LPS group hardly exhibited any difference between the time to explore each object (Figure 3D). In OLRT, the percentage of exploration time increased for the moved object in the control group, but not in the LPS group (Figure 3E). There was no difference in object recognition time between the two groups (Figure 3F). These results suggest that short-term memory impairment was present on day 7 following LPS injection, after recovery from systemic symptoms.

As short-term memory impairment was detected on day 7 post-LPS, multiple pathologies were investigated. Plasma LPS level increased to 1.5 ± 0.1 EU/mL 24 h after LPS i.p. injection (Figure 4A). Brain LPS levels were also significantly increased 24 h after LPS injection. Because this finding was suggestive of BBB breakdown, we evaluated BBB permeability by measuring migration of Evans blue into the brain and measuring brain water content as an indicator of cerebral edema. Neither index changed significantly on days 0, 1, 3, or 7 post-LPS injection (Figures 4B,C), suggesting that permeability of the BBB did not change. To evaluate neuronal survival in the CA1 area of the hippocampus, which is associated with short-term memory, the density of NeuN-positive cells was quantified, and neuronal cell death was detected by PathoGreen™ staining. The densities of live and dead neurons were not changed on days 0, 1, 3, and 7 post-LPS injection (Figures 4D,E). Based on these results, the experimental conditions of LPS injection used in this study increased the amount of toxin in the blood and brain, but did not affect the BBB or neuronal survival.

Subsequently, we examined whether LPS affected CA1 neuronal activity. The density of cells positive for c-Fos, a measure of neuronal activity, was significantly decreased on day 7 (Figure 5A). The density and morphology of dendritic spines, which are composed of synapses, affect neuronal circuits and plasticity, contributing to memory and learning (Runge et al., 2020). Golgi-Cox staining revealed that spine density did not change on day 7 (Figures 5B,C), but spine morphology was altered, which was characterized by elongated spines (Figure 5D). Because immature spines have an elongated morphology, spines were classified according to their morphology. The percentage of immature spines (long thin and thin) increased and the percentage of mature spines (mushroom and branched) decreased on day 7 (Figure 5E). Taken together, these findings demonstrate that LPS caused suppression of neuronal activity accompanied by increased immature CA1 dendritic spines on day 7.

Spine density tended to decrease transiently on day 3 (Figure 5C). Because activated glial cells potentially phagocytose spines, leading to decreased spine density (Stein and Zito, 2019), we subsequently investigated glial cell activation. GFAP and Iba-1, indicators of astrocyte and microglial activation, respectively, were detected by immunostaining (Figure 6A). In astrocytes, the density of GFAP-positive cells and mean GFAP fluorescence intensity increased on day 3 and recovered to baseline levels by day 7 (Figures 6B,C). Because astrocytic activation is strongly associated with morphological changes, we evaluated process morphology with the SMorph application, and performed PC analysis of 16 parameters (Supplementary Figure S2). The first two PCs described approximately 56% of the observed variability in the data (Figure 6D). PC1 was increased on day 7, which significantly correlated with projected area and total branch length, and PC2 was increased on day 3, which was significantly correlated with average length of terminal branches and average branch width (Supplementary Figure S4A). According to these parameters, process widths and lengths increased on day 3, and process number and total length increased on day 7. In microglia, like in astrocytes, the density of Iba-1-positive cells and the mean fluorescence intensity of Iba-1 increased on day 3 and recovered to baseline levels on day 7 (Figures 6E,F), and the soma size of microglia increased significantly on days 1 and 3, which is characteristic of activated microglia (Figure 6G). We also performed PC analysis using six parameters for microglia (Supplementary Figure S3). The first two PCs accounted for approximately 85% of the observed variability in the data (Figure 6H). PC1 was increased on days 1 and 3, which was significantly correlated with the number of branches and junctions, and PC2 was increased on days 1 and 7, which was significantly correlated with average and maximum length of branches (Supplementary Figure S4B). According to these parameters, the number of processes increased on day 3, and process length significantly increased on days 1–7 in microglia (Supplementary Figure S3).

Subsequently, we measured mRNA levels in the hippocampus for several factors considered to be indicators of glial cell activation (Ikeshima-Kataoka, 2016; Jurga et al., 2020; Escartin et al., 2021; Guo et al., 2022). For the seven factors (Gfap, Aqp4, Thbs1, Maob, Sox9, Slc1a2, and Slc1a3) that are indicators of reactive astrocytes, RT-qPCR analysis revealed that only Gfap significantly increased on day 1, and Aqp4 non-significantly increased on day 3 (Figure 7A). No significant differences between groups were detected for any of the astrocytic activation factors on day 7. For the six microglial M1 markers (Aif1, Il1b, Ccl2, Tnf, Fcgr3, and Cd86), there were significant increases in all the factors on day 1 and no factors significantly differed between groups on day 7 (Figure 7B). The M1 markers Aifi, Fcgr3, and Cd86 were also significantly increased on day 3. For the five M2 markers, four factors (Chil3/4, Il10, Ccl22, and Mrc1) were significantly increased on days 1 or 3 (Figure 7C). Mrc1 remained significantly increased on day 7. Contrastingly, Retnla significantly decreased on days 1, 3, and 7. These data suggest that the robust LPS-induced inflammation peaked on days 1–3 and approximately recovered to baseline levels by day 7. The decrease in dendritic spine density and glial cell inflammation were both maximal on days 1–3. We thus hypothesized that activated microglia phagocytosed the spines, and performed immunostaining for CD68, the microglia-specific phagocytic marker (Figure 7D). CD68-positive cells did not express the astrocyte marker GFAP or the neuronal marker NeuN (Supplementary Figure S5). Mean CD68 fluorescence intensity and area of CD68 increased significantly on day 3 (Figures 7E,F). It is likely that microglia phagocytosed spines on days 1–3, resulting in an increase in immature spines on day 7.

Finally, we also examined the inflammatory responses to LPS in other brain regions, including the dentate gyrus, perirhinal cortex, entorhinal cortex, basolateral amygdala, and paraventricular hypothalamic nucleus, which include the regions involved in short-term memory. Glial cells were activated in all regions examined on day 3 (Figure 8A). Meanwhile, CD68 fluorescence intensity was increased in the hippocampal dentate gyrus without significant changes in the other regions on day 3 (Figures 8B,D). c-Fos expression was decreased in the hippocampal dentate gyrus on day 7, while no difference was observed in the perirhinal cortex or entorhinal cortex (Figure 8C). Contrastingly, LPS increased c-fos expression in the basolateral amygdala and paraventricular nucleus (Figure 8C). These results suggest that glial cell activation occurred throughout the brain, but that there were regional differences in microglial phagocytosis and neuronal activity in response to LPS challenge.