Long-term running exercise improves cognitive function and promotes microglial glucose metabolism and morphological plasticity in the hippocampus of APP/PS1 mice

APPswe/PSEN1dE9 mice were obtained from the Animal Model Institute of Nanjing University, China. To exclude the influence of sex differences on cognitive impairment in AD [40], only male mice were used in this study. Forty 10-month-old male APP/PS1 mice and forty wild-type littermates were randomly divided into sedentary groups (WT_Sed and AD_Sed, n = 20/group) and running groups (WT_Run and AD_Run, n = 20/group). Single housing had an impact on social behavior [41]. To minimize the influence of a single housing on the results, each mouse from each group was single caged in standard polypropylene cages (290 × 180 × 160 mm) throughout the experiment. Each running mouse had free access to a running wheel (12 cm in diameter) for 3 months, which was connected to a counter to record the running distance (Wuhan Yihong Technology, China). Body weight was measured weekly throughout the study. Mice were intraperitoneally injected with 5-bromo-2′-deoxyuridine (BrdU) (10 mg/kg; Sigma) for 7 days and sacrificed 28 days after the last injection [42]. Mice were housed at 22 ± 1 °C with 55 ± 5% humidity on a 12-h light–dark cycle with ad libitum access to water and food. All animal procedures followed double-blind principles and the National Institutes of Health Guide for the Care and Use of Laboratory Animals (NIH Publication No. 85–23).

New object recognition (NOR) [43] was performed in a 30 × 30 × 45 cm³ open box. All mice at the age of 13 months were placed in the center of the box to acclimate for 10 min per day for 3 days. On the test day, mice were first placed in the box for 10 min with two identical objects and then placed in the box again for 10 min with one of the objects, which was replaced with a new object 1 h later. The time exploring the object (nose tip toward the object at a distance ≤ 2 cm) was recorded; exploration times of less than 8 s were excluded [44]. The discrimination index (DI; DI = total time spent with a new object/total time spent with two objects) was calculated.

The Morris water maze (MWM) [45] was used to assess spatial learning and memory and utilized a circular pool (120 cm in diameter) and a platform (10 cm in diameter) divided into four quadrants filled with white water (22–25 ℃). In the positioning navigation experiment (days 1–6), the platform was fixed in a quadrant and hidden 1 cm below the water. Each mouse was placed on the platform for 15 s of brief training and then placed into a quadrant of water facing the pool wall to find the platform. If the platform was not found within 60 s, the mouse was guided to the platform for another 15 s of training. Each mouse was tested in all four quadrants in a random order once each day in each quadrant. In the space exploration experiment on day 7, the platform was removed, and the mice were placed in two different positions away from the platform and allowed to swim for 60 s. A video tracking system recorded each mouse's swimming speed, escape latency and frequency across the platform.

Five mice in each group were given PET scans at the age of 10 months and 13 months. After fasting for 12 h, mice were anesthetized with isoflurane (4% for induction, 2% for maintenance) supplemented with oxygen, and [18]F-FDG was intraperitoneally injected. The injection dose and body weight of each mouse were recorded. Approximately 40 min later, the mice were anesthetized again and scanned using a nanoScan PET/MRI system (Mediso, Hungary). During scanning, mice were placed on a 37 ℃ constant temperature bed, and their respiratory rates were continuously monitored. MRI-based attenuation correction PET images were reconstructed with Nucline software (Bioscan, USA). PET images were matched to a predefined mouse brain atlas template (Additional file 1: Fig. S1a), and the standardized uptake value (SUV) of the volume of interest (VOI) was calculated for semiquantitative analysis using POMD v.3.4 (PMOD Technologies, Switzerland) [SUV = VOI activity concentration (Bq/cm)/(injected dose (Bq)/body weight (g))].

Mice were intraperitoneally anesthetized with 1% pentobarbital (Sigma). The brains of mice (three mice in each of the following groups: WT_Sed, AD_Sed, and AD_Run) were directly dissected, and the hippocampus was isolated on ice and soaked in RNA protect tissue reagent (QIAGEN). Total RNA was extracted with TRIzol (Invitrogen). Ribosomal RNA (rRNA) was removed, and the RNA was fragmented into small pieces. The cleaved RNA fragments were copied into first-strand cDNA followed by second-strand cDNA synthesis. This process removes the RNA template and synthesizes a replacement strand, incorporating dUTP in place of dTTP to generate ds cDNA. The incorporation of dUTP quenches the second strand during amplification. After quality and quantity evaluation, the cDNA library was created, and paired-end sequencing was performed on a BGISEQ-500/MGISEQ-2000 System (BGI-Shenzhen; Beijing Genomic Institute in Shenzhen, China). Raw data were filtered with SOAPnuke v.1.5.2. The clean reads were mapped to the reference genome using HISAT2 v.2.0.4. Bowtie2 v.2.2.5 was applied to align the clean reads to the gene set, a database built by BGI-Shenzhen, and the expression levels of the genes were calculated with RSEM v.1.2.12. Differential expression analysis was performed using DESeq2 v.1.4.5 [46]. Gene set enrichment analysis (GSEA) was performed using GSEA v.4.1.0 [47].

Five mice in each group were transcardially perfused with saline containing heparin followed by 4% paraformaldehyde solution. The brains were removed and divided into two hemispheres, fixed with paraformaldehyde and preserved at 4 °C. The brains were dehydrated with sucrose solutions (10%, 20%, and 30%) in 0.1 M phosphate-buffered saline (PBS), embedded in optimal cutting temperature (OCT) compound, and quickly frozen for 10 min. One hemisphere of the brain was cut into 50 μm thick sections, and another hemisphere was cut into 30 μm thick sections using a frozen slicer (Leica, Germany). All 50 μm sections containing the hippocampus were randomly divided into 5 equal series starting from the first section (n = 12–14 sections/each series), and the 30 μm sequences were divided into 20 equal series (n = 4–5 sections/each series) and stored in alcohol at − 20 ℃.

One series of 50 μm sections was washed twice with 0.01 M PBS and permeabilized in PBS containing 0.3% Triton X-100/0.1% Tween-20 (PBST) for 1 h, incubated with 0.3% H₂O₂ for 20 min to eliminate endogenous peroxidase activity, and blocked in PBST with 10% goat serum/1% fetal serum at 37 °C for 2 h after heat-induced antigen retrieval in citrate buffer for 30 min. Then, the sections were incubated with rabbit anti-IBA1 antibody (Abcam, ab153696, 1:1000) at 4 °C for 48 h, followed by incubation with biotin-labeled goat anti-rabbit IgG and horseradish peroxidase-labeled working solution (ZSDB-BIO, China) at 37 °C for 2 h. Next, sections were visualized with 3,3-diaminobenzidine (DAB), mounted on slides, counterstained with hematoxylin, dehydrated with gradient alcohol series (75%, 80%, 95%, 100%), cleared with xylene, sealed with neutral gum and coverslipped.

The total number of microglia in the hippocampal DG, CA1, and CA3 regions was determined via the optical fractionator method using a morphometry system consisting of a microscope with a camera (Olympus, Japan), a high-precision microcator (ProScan, UK), and stereological analysis software (New CAST, Denmark). Stereological counting and analysis were performed by a double-blind procedure. Images were captured under a × 4 lens. Outlines of the DG, CA1, and CA3 regions were delineated according to The Mouse Brain in Stereotaxic Coordinates [48]. The stereological probe (counting point) was superimposed onto the image, and the volume of the region of interest was calculated using Cavalieri's principle [49]. Under the × 100 oil lens, the unbiased counting frame was randomly and equidistantly placed in the delineated area of interest: the area sampling fraction (asf) was 8%, the counting height was 15 μm and the guard height was 3 μm. The number of positive cells in each counting frame was counted successively: the z-axis position of the first focused microglia was set as 0 μm, and counting began when the z-axis moved below the guard height to 18 μm. The actual thickness (t) was recorded, and the overall average sampling fraction of each region was calculated (asf = 15 μm/t). The counted number was summed (ΣQ⁻), and the total number was calculated: N = ΣQ⁻ × (1/5) × (1/asf) × (1/tsf) [50].

One series of the 30 μm sections mentioned above was permeabilized, subjected to antigen retrieval, blocked and then incubated with the following primary antibodies at 4 °C for 36 h: mouse anti-GLUT5 (Santa Cruz, sc-271055, 1:100), mouse anti-TREM2 (Santa Cruz, sc-373828, 1:100), mouse anti-CD68 (Abcam, ab955, 1:500), rabbit anti-IBA1 (Abcam, ab153696, 1:500) and rabbit anti-PSD95 (CST, 3450, 1:500) followed by incubation with the corresponding fluorescent secondary antibodies (1:200) at 37 °C for 2 h: DyLight-549 goat anti-mouse IgG and DyLight-488 goat anti-rabbit IgG. Sections were stained with DAPI and sealed with an anti-fluorescence quencher.

For BrdU staining, after blocking, the sections were incubated with primary antibodies at 4 °C for 36 h: rabbit anti-IBA1 (Abcam, ab153696, 1:500) and chicken anti-NeuN (Sigma, ABN91, 1:400). Sections were treated with 2 M HCl for 50 min, washed with 0.1 M sodium tetraborate buffer and incubated with rat anti-BrdU antibody (Abcam, ab6326, 1:300) at 4 °C for 24 h. Then, the sections were incubated with DyLight-549 goat anti-mouse IgG, Alexa Fluor 488 goat anti-chicken IgG, and DyLight-649 goat anti-rat IgG at 37 °C for 2 h and sealed. For thioflavin S staining, sections were immersed in 50% alcohol containing 0.03% thioflavin S (Sigma) at room temperature in the dark for 10 min, cleaned with 50% alcohol 3 times, and sealed with an anti-fluorescence quencher.

Images were captured using a confocal microscope (Nikon) under a × 60 oil lens and a superresolution microscope system (Olympus) under a × 20 lens. The image threshold was uniformly set using ImageJ v.6.4. “Skeleton analysis” was used to measure the morphology of the microglia [51]. The number of positive cells was determined by “particle analysis” [52], and the number of colabeled cells was counted manually with a double-blind procedure. 3D images were observed and masked using Imaris v.9.0 [53].

The hearts of ten mice in each group were exposed, and blood was drawn from the right atrium into an anticoagulant tube containing heparin and centrifuged for 15 min at 3000×g. The plasma was extracted and stored at − 80 °C. The level of sTREM2 in plasma was determined with a mouse sTREM2 ELISA kit (J&L, JL20435, China).

Protein solutions from three mice in each group were denatured with 5 × loading buffer at 95 °C for 10 min and stored at − 20 °C. Twenty to forty micrograms of total protein sample were separated by SDS-polyacrylamide gel electrophoresis using a Bio–Rad protein assay and transferred onto polyvinylidene fluoride (PVDF) membranes. Membranes were blocked in Tris-buffered saline/0.1% Tween buffer (TBST; 25 mM Tris–Cl, 125 mM NaCl, 0.1% Tween-20) with 5% skim milk powder at room temperature for 2 h and incubated with primary antibodies at 4 °C overnight: mouse anti-GLUT5 (Santa Cruz, sc-271055, 1:300), mouse anti-GLUT3 (Abcam, ab150299, 1:1000), mouse anti-ADAM10 (Santa Cruz, sc-28358, 1:500), mouse anti-SYK (Santa Cruz, sc-1240, 1:300), rabbit anti-phospho-SYK (Tyr525/526) (CST, 2710, 1:300), mouse anti-β-tubulin and mouse anti-β-actin (Abmart, 1:1000). The corresponding secondary antibody was incubated at room temperature for 2 h (1:4000). The results were analyzed using SuperSignal West Pico Chemiluminescent Substrate (Thermo Fisher Scientific, USA).

Total RNA was extracted with TRIzol (Invitrogen). cDNA was synthesized using the PrimeScript RT reagent Kit (Takara) in a reaction of 1 μg of total RNA per reaction. Relative mRNA levels were quantified by qRT–PCR using the SYBR® Green Premix Pro Taq HS qPCR Kit (Accurate Biology, China) with the following specific primers: Trem2-F: CCAAGGAATCAAGAGACC, Trem2-R: CAGTGAGGATCTGAAGTTG; Tyrobp-F: TGCGACTGTTCTTCCGTGAG, Tyrobp-R: TTCCGCTGTCCCTTGACCT; Cd68-F: AGGCTTTTCATTTCCTCTTCCA, Cd68-R: TCCTCTGTTCCTTGGGCTAT; Cd74-F: CGACCTCATCTCTAACCA, Cd74-R: GTACAGGAAGTAAGCAGTG; Itgax-F: ACTGTTCACCACCCAAAGTG, Itgax-R: TGTCCCCTTGTTTTCTCCCA; Spp1-F: TTTCTGATGAACAGTATC, Spp1-R: GAAGATGAACTCTCTAAT.

There was no significant difference in the average running distance per day (F = 0.031, p = 0.862) and bodyweight of mice (F = 0.099, p = 0.960) between the two mouse genotypes (Table 1). The MWM and NOR tests were conducted to evaluate cognitive function. In the MWM test, there was no difference in the average swimming speed in the place navigation test from days 1 to 6 among each group (F = 0.388, p = 0.762, Fig. 1a), but the escape latency was significantly changed among the four groups (F = 7.273, p = 0.000, Fig. 1b). Compared with AD_Sed mice, the escape latency of WT_Sed mice was significantly shortened (post hoc, p = 0.002, Fig. 1b), and the escape latency of AD_Run mice was significantly shortened (post hoc, p = 0.000, Fig. 1b). In the spatial probe test on day 7, the number of times crossed the platform also changed significantly among the four groups (F = 5.824, p = 0.001, Fig. 1c). The number of times the AD_Sed mice crossed the platform was less than that of the AD_Run mice (post hoc, p = 0.024, Fig. 1c), and the number of times the AD_Sed mice crossed the platform was less than that of the WT_Sed mice (post hoc, p = 0.001, Fig. 1c). There was a significant difference in the DI of the NOR test among the four groups (F = 6.719, p = 0.000, Fig. 1d). The DI significantly increased in AD_Run mice compared with AD_Sed mice (post hoc, p = 0.001, Fig. 1d), and the DI significantly increased in WT_Sed mice compared with AD_Sed mice (post hoc, p = 0.004, Fig. 1d). These findings demonstrated that long-term voluntary running markedly protected learning and memory loss in 13-month-old APP/PS1 mice.

There were significant differences in the volumes of DG (F = 6.490, p = 0.004) and CA1 (F = 7.558, p = 0.002) regions among the four groups, while there was no significant difference in the volume of CA3 region (F = 2.909, p = 0.067) (Fig. 1e). The volumes of the DG and CA1 regions were significantly increased in AD_Run mice compared with AD_Sed mice (post hoc, p(DG) = 0.021, p(CA1) = 0.008, Fig. 1e). The volumes of the DG and CA1 regions were significantly increased in WT_Run mice compared with WT_Sed mice (post hoc, p(DG) = 0.010, p(CA1) = 0.008, Fig. 1e). However, there was no significant difference in the DG, CA1, and CA3 volumes between WT_Sed mice and AD_Sed mice (Fig. 1e). These results indicated that running exercise increased the hippocampal volume in the DG and CA1 regions of APP/PS1 mice relative to their sedentary littermates, as it did in wild-type mice. The Aβ plaque-positive area in the hippocampus of the AD_Run mice was notably lower than that in the hippocampus of AD_Sed mice (p = 0.013 (≤ 20 μm in diameter), p = 0.000 (> 20 μm in diameter), Fig. 1f), suggesting that running exercise significantly reduced Aβ plaque deposition in the hippocampus of APP/PS1 mice.

As shown in Fig. 2a, 115 genes were upregulated and 10 genes were downregulated in AD_Sed mice compared with WT_Sed mice. These upregulated genes included some disease-associated microglial genes, such as Trem2, Tyrobp, Spp1, Clec7a, Ccl3, Cd74, and Cd68, suggesting extensive microglial activation during AD progression. However, there were few DEGs in AD_Run mice compared with AD_Sed mice (Fig. 2b). The top differentially expressed genes are shown in the heatmap (Fig. 2c). These results indicated that the effect of running exercise on the transcriptional level of hippocampal genes was too small to determine molecular mechanisms in APP/PS1 mice, inferring that regulation caused by running exercise may occur at the posttranscriptional level.

To further observe the slight changes in gene expression, gene set enrichment analysis (GSEA) was performed. The top differential KEGG pathways in GSEA are shown in Fig. 3a, b. We found that several glucose metabolism pathways were reduced in AD_Sed mice compared with WT_Sed mice, including the “pentose phosphate pathway”, “fructose and mannose metabolism” and “glycolysis/gluconeogenesis” (Fig. 3a, c). In contrast, “starch and sucrose metabolism”, “pyruvate metabolism”, “glycolysis/gluconeogenesis”, “pentose phosphate pathway”, and “fructose and mannose metabolism” were upregulated in AD_Run mice compared with AD_Sed mice (Fig. 3b, c). These results suggested that hypometabolism may exist in the hippocampus of APP/PS1 mice, whereas running exercise promoted hippocampal glucose metabolism in the hippocampus of APP/PS1 mice. In addition, several synaptic pathways were reduced in AD_Sed mice compared with WT_Sed mice but upregulated in AD_Run mice compared with AD_Sed mice, such as the “GABAergic synapse” (Fig. 3a, b). PSD95, a postsynaptic density marker, was stained and the mean intensity of PSD95 showed significant difference in the DG (F = 9.167, p = 0.012), CA1 (F = 7.405, p = 0.019), and CA3 (F = 8.179, p = 0.015) regions among the four groups (Additional file 2: Figs. S2a, 2b). The mean intensity of PSD95 in the DG, CA1, and CA3 regions of AD_Sed mice was significantly lower than that in WT_Sed mice (post hoc, p(DG) = 0.005, p(CA1) = 0.020, p(CA3) = 0.005, Additional file 2: Fig. S2b), and the mean intensity of PSD95 in the DG and CA3 regions of AD_Sed mice was also lower than that in AD_Run mice (post hoc, p(DG) = 0.008, p(CA3) = 0.009, Additional file 2: Fig. S2b). These results suggested that running exercise may improve cognitive function by reducing synapse loss in the hippocampus of APP/PS1 mice.

To evaluate glucose metabolism in the hippocampus, [18]F-FDG-PET was performed. Our results showed a significant difference in the FDG uptake of the hippocampus among the four groups (F = 19.647, p = 0.000, Fig. 4a, b). The FDG uptake (SUV) was significantly enhanced in AD_Run mice compared with AD_Sed mice (post hoc, p = 0.001, Fig. 4b), confirming the significant enhancement of glucose metabolism in the hippocampus of APP/PS1 mice induced by running exercise. The FDG uptake (SUV) in WT_Run mice was also much higher than that in WT_Sed mice (post hoc, p = 0.000, Fig. 4b), indicating that running exercise also increased hippocampal glucose metabolism in WT mice. In addition, hippocampal glucose uptake was significantly correlated with the DI from the NOR test (p = 0.006, Fig. 4c).

As shown in Fig. 5a, GLUT5 was mainly expressed on microglia. There was a significant difference in the ratio of GLUT5⁺/IBA1⁺ microglia in the DG region (F = 25.520, p = 0.000), but not in the CA1 (F = 1.830, p = 0.220) or CA3 (F = 3.026, p = 0.094) regions among the four groups (Fig. 5a, b). The ratio of GLUT5⁺/IBA1⁺ microglia in the DG was significantly increased in AD_Sed mice compared with WT_Sed mice (post hoc, p = 0.002, Fig. 5b). Furthermore, the ratio of GLUT5⁺/IBA1⁺ microglia in the DG was significantly increased in AD_Run mice compared with AD_Sed mice (post hoc, p = 0.021, Fig. 5b). The protein level of GLUT5 was also changed significantly in the hippocampus among the four groups (F = 7.087, p = 0.012, Fig. 5c, d). Although the expression of GLUT5 was not significantly different between WT_Sed mice and AD_Sed mice (post hoc, p = 0.189, Fig. 5c, d), the expression of GLUT5 was significantly increased in AD_Run mice compared with AD_Sed mice (post hoc, p = 0.017, Fig. 5c, d). However, there was no significant difference in the GLUT3 expression among the four groups (F = 1.240, p = 0.357, Fig. 5c, e). These results indicated that running exercise may improve cognition through enhancing microglial glucose metabolism in APP/PS1 mice.

As shown in Fig. 6a, running exercise did not affect several microglial genes at the mRNA level that were upregulated in APP/PS1 mice, such as Trem2, Tyrobp, and Spp1. The mRNA levels of Trem2, Tyrobp, Cd68, Cd74, Itgax, and Spp1 in the hippocampus were also verified using qRT–PCR. There were significant differences in the mRNA levels of Trem2 (F = 9.257, p = 0.001), Tyrobp (F = 9.030, p = 0.001), Cd68 (F = 8.145, p = 0.002), Cd74 (F = 19.357, p = 0.000), and Itgax (F = 39.541, p = 0.000) among the four groups, while there was no significant difference in the mRNA levels of Spp1 among the four groups (F = 1.463, p = 0.252) (Additional file 3: Fig. S3a–f). The mRNA levels of Trem2, Tyrobp, Cd68, Cd74, and Itgax was significantly increased in AD_Sed mice compared with WT_Sed mice (post hoc, p = 0.000, p = 0.017, p = 0.006, p = 0.000, p = 0.000). However, there were no significant differences in the level of these genes between AD_Run mice and AD_Sed mice except for a decrease in Cd74 in the hippocampus of AD_Run mice compared with AD_Sed mice (post hoc, p = 0.001, Additional file 3: Fig. S3d). Therefore, we evaluated the protein levels of TREM2 and sTREM2. There were significant differences in the protein levels of TREM2 in the hippocampus (F = 4.440, p = 0.012, Fig. 6c) and the level of sTREM2 in the plasma (F = 3.307, p = 0.031, Fig. 6e), but there was no significant difference in the level of sTREM2 in hippocampus among groups (F = 1.856, p = 0.162, Fig. 6d). We found more TREM2-positive fluorescent labels in the microglia of AD_Run mice (Fig. 6b). The levels of TREM2 in the hippocampus were significantly increased in AD_Run mice compared with AD_Sed mice (post hoc, p = 0.003, Fig. 6c). These results indicated that running exercise upregulated the expression of the TREM2 protein in the hippocampus of APP/PS1 mice. However, the levels of sTREM2 were significantly reduced in AD_Run mice compared with AD_Sed mice in the plasma (post hoc, p = 0.027, Fig. 6e), suggesting that running exercise may reduce TREM2 shedding and prevent sTREM2 release into the blood in APP/PS1 mice. Moreover, there was a significant negative correlation between plasma sTREM2 levels and hippocampal FDG uptake (p = 0.020, Fig. 6f), which indicated that TREM2 hydrolysis was negatively associated with hippocampal glucose metabolism. Inhibiting the loss of TREM2 may be one of the mechanisms by which running exercise improved glucose metabolism in the hippocampus and rescued cognitive decline in APP/PS1 mice.

The protein level of SPP1, a TREM2-dependent protein [33] (F = 18.293, p = 0.001, Fig. 6h), ADAM10, a protease of TREM2 (F = 6.679, p = 0.014, Fig. 6j), and p-SYK (F = 12.638, p = 0.002, Fig. 6k) were significantly different among groups, but TYROBP (DAP12) expression was unchanged significantly (F = 2.108, p = 0.178, Fig. 6i). SPP1 was significantly increased in AD_Run mice compared with AD_Sed mice (post hoc, p = 0.000, Fig. 6g, h). ADAM10 showed lower expression levels in AD_Sed mice than in WT_Sed mice (post hoc, p = 0.002, Fig. 6g, j) and AD_Run mice (post hoc, p = 0.042, Fig. 6g, j), which suggested that there might be other reasons for the increased hydrolysis of TREM2 in APP/PS1 mice. The ratio of p-SYK (Tyr525/526)/SYK increased significantly in AD_Run mice compared with AD_Sed mice (post hoc, p = 0.002, Fig. 6g, k,), suggesting that running exercise enhanced the phagocytosis of microglia by activating the TREM2/SYK signaling pathway.

The total number of microglia in the DG (F = 41.644, p = 0.000) and CA1 (F = 7.907, p = 0.002) were significantly different among the four groups (Fig. 7a, b). The total number of microglia in the DG and CA1 increased significantly in AD_Sed mice compared with WT_Sed mice (post hoc, p(DG) = 0.001, p(CA1) = 0.020, Fig. 7a, b), suggesting significant microglial proliferation in the hippocampus of APP/PS1 mice. Furthermore, the total number of microglia in the DG was significantly increased in AD_Run mice compared with AD_Sed mice (post hoc, p = 0.000, Fig. 7a, b). There was no significant difference in the total number of microglia in the CA1 region between AD_Sed mice and AD_Run mice (post hoc, p = 0.188, Fig. 7a, b). There was no significant difference in the total number of microglia in the CA3 region among the four groups of mice (F = 1.738, p = 0.202, Fig. 7a, b). These results indicated that running exercise led to an increase in microglia in the DG of APP/PS1 mice. There was also a significant difference in the density of IBA1 microglia in the DG region among the four groups (F = 39.990, p = 0.000, Fig. 7c). The density of IBA1 microglia in the DG of AD_Run mice was higher than that in the DG of AD_Sed mice (post hoc, p = 0.005, Fig. 7c). The endpoints of microglial branches (F = 28.368, p = 0.000) and the process length of microglia (F = 46.012, p = 0.000) were changed significantly in the DG region among the four groups (Fig. 7d–f). The endpoints of microglial branches were reduced in the DG of AD_Sed mice compared with WT_Sed mice (post hoc, p = 0.000, Fig. 7d, e). In addition, the endpoints of microglial branches were increased in the DG of AD_Run mice compared with AD_Sed mice (post hoc, p = 0.003, Fig. 7d, e). Similarly, the process length of microglia decreased in the DG of AD_Sed mice compared with WT_Sed mice (post hoc, p = 0.000, Fig. 7d, f), and the process length of the microglia increased in the DG of AD_Run mice compared with AD_Sed mice (post hoc, p = 0.044, Fig. 7d, f). Collectively, our results showed that running exercise increased the number of microglia and improved the number and length of microglial branches in the DG of APP/PS1 mice.

To further determine whether the effect of running exercise on the number of microglia is due to prompted proliferation or survival, microglia were labeled with BrdU, a marker of proliferation [42], and NeuN was used to delineate hippocampal subregions. There was a significant difference in the number of BrdU⁺/IBA1⁺ microglia in the DG region among the four groups (F = 22.959, p = 0.000, Fig. 8a, c). The number of BrdU⁺/IBA1⁺ microglia in the DG was significantly increased in AD_Sed mice compared with WT_Sed mice (post hoc, p = 0.000, Fig. 8a, c), suggesting significant microglial proliferation in the hippocampus of APP/PS1 mice. Although the number of BrdU⁺ cells in the DG of the hippocampus was increased in AD_Run mice compared with AD_Sed mice (post hoc, p = 0.044, Fig. 8a, b), there was no difference in the number of BrdU⁺/IBA1⁺ microglia in the hippocampus between AD_Run mice and AD_Sed mice (post hoc, p = 0.709, Fig. 8c). These results suggested that there was significant microglial proliferation in the DG of APP/PS1 mice and that running exercise did not further promote microglial proliferation. The increase in microglial number in the DG may be due to the promotion of microglial survival.

CD68, a marker of microglial activation, and IBA1 were stained. There were significant differences in the ratio of CD68⁺/IBA1⁺ microglia in the DG (F = 116.905, p = 0.000), CA1 (F = 32.310, p = 0.000), and CA3 (F = 28.301, p = 0.000) regions among the four groups (Fig. 9a, b). The ratio of CD68⁺/IBA1⁺ microglia in the DG, CA1, and CA3 regions was significantly increased in AD_Sed mice compared with WT_Sed mice (post hoc, p(DG) = 0.000, p(CA1) = 0.000, p(CA3) = 0.000, Fig. 9b), while the ratio of CD68⁺/IBA1⁺ microglia in the DG region of AD_Run mice was lower than that of AD_Sed mice (post hoc, p = 0.011, Fig. 9b). Moreover, the ratio of the CD68/IBA1-positive area around Aβ plaques (≤ 15 μm) in the hippocampus of AD_Run mice was significantly lower than that in AD_Sed mice (p = 0.025, Fig. 9a, c). These results suggested that running exercise might reduce the proportion of microglial activation in APP/PS1 mice.