In this study, male C57BL/6 mice were housed 5/cage under controlled conditions with a 12-h light/dark cycle in an ambient temperature (23 ± 2°C) and humidity (55 ± 2%) controlled room. Water and food were provided ad libitum. All animal experiments were performed in accordance with the Xi’an Jiaotong University Guidelines for Animal Experimentation and were approved by the Animal Care and Use Committee of Xi’an Jiaotong University (Xi’an, Shaanxi, China).

The CSR protocol was implemented according to our previous study (Zhang S. et al., 2023). 9- to 10-week-old (n = 78) mice were placed in the automated cylindrical apparatus (KW-BD, NJKEWBIO, Nanjing, China) with a spinning bar at approximately 7 rpm to consistently disrupt their sleep for 20 h (from 8 p.m. to 4 p.m.) for consecutive 1 week. Non-sleep-deprived mice were housed in the same apparatus without the spinning bar and served as controls (CON).

Memory was evaluated using the novel object recognition (NOR; Lueptow, 2017) and the Y-maze test (Kraeuter et al., 2019) in mice at 10–11 weeks (n = 160). In the training session of NOR, the mice were allowed to explore freely in the apparatus with two identical objects for 10 min to form object memory. The testing session was performed 60 min after training. During the testing session, one identical object was swapped out for a new one and the mice freely explored the objects for 10 min. The discrimination index was calculated by dividing the time spent examining the novel object by the total time spent exploring both objects.

In the Y-maze test, mice were allowed to explore freely for 8 min. When the mouse entered the arm with all four paws, it was considered as an arm entry. Consecutive entries into all three arms were deemed to constitute an alternation. Accordingly, the percentage of alternation behavior was measured according to the following formula: % Alternation = (total number of alternations / total number of arms traversed −2) × 100%.

At a young age of 1–2 days, C57BL/6 mice’s cerebral cortices were harvested to get primary cortical astrocytes. Cells were seeded in a T75 flask containing 10 mL of medium (90% DMEM, 10% FBS, and 1% penicillin/streptomycin). To obtain mouse primary astrocytes, two mouse cortices were used per T75 flask preparation and then shaken for 2 h at 250 rpm using an orbital shaker to eliminate most of the microglia. Subsequently, the flask was transferred to an incubator. Cells from independent experiments at 80% confluence were stimulated with lipopolysaccharide (LPS, 100 ng/mL, Sigma-Aldrich, United States; Zhang S. et al., 2023) or TMAO (100 μM, Sigma-Aldrich, United States; Brunt et al., 2021) for 24 h.

Hippocampal tissue was subjected to RIPA buffer supplemented with protease inhibitors. The samples underwent electrophoresis on a 15% SDS-PAGE gel and were subsequently transferred to PVDF membranes. The membranes were incubated with the primary antibody overnight at 4°C after being blocked with PBS containing 5% non-fat dry milk for 2 h at room temperature and were then incubated overnight at 4°C with different primary antibodies including FMO3 (Abcam, ab126711), SREBP2 (ImmunoWay, YN0037), PSD95 (Cell Signaling, 3,450), synaptophysin (Synaptic Systems, 101,011), and ACTB (Abclonal, AC026). Following three washes with TBST, the membranes were subjected to peroxidase-conjugated secondary antibodies (goat anti-rabbit; goat anti-mouse) for 60 min at room temperature. The membrane bands were discovered using an ECL kit (Affinity, Shanghai, China), and the protein band densities were quantified using ImageJ. ACTB was used as an internal control.

The mice were deeply anesthetized and transcardially perfused with ice-cold saline and 4% paraformaldehyde (PFA). Then, the entire brain was fixed with 4% PFA for 6 h, followed by a 30% sucrose solution to dehydrate. Coronal sections of the hippocampus (12-μm thickness) were collected using cryostat instruments. For immunofluorescence staining, the sections underwent three washes with PBS, blocked and perforated for 2 h at room temperature with PBS containing 0.3% Triton X-100 and 5% donkey serum, and then incubated with primary antibodies including GFAP (GeneTex, GTX85454), SREBP2 (SAB, 53501), PSD95 (Abcam, ab238135), and SYP (Synaptic Systems, 101,011) overnight at 4°C. Following three 10 min washes with PBS (pH 7.4), the sections were incubated for 2 h at room temperature with fluorescent dye-conjugated secondary antibodies (donkey anti-chicken Alexa Fluor 488; donkey anti-rabbit Alexa Fluor 594; donkey anti-mouse Alexa Fluor 488) in the dark. Finally, the sections were incubated with DAPI for 10 min, followed by three washes with PBS as described above. All confocal images were acquired using a confocal microscope (Olympus FV3000).

For primary astrocytes, cells were seeded in the confocal dish. Astrocytes were fixed in 4% paraformaldehyde for 10 min at 4°C. Subsequently, astrocytes were washed with PBS, blocked for 30 min at 37°C in blocking buffer, and then incubated with primary antibodies GFAP (GeneTex, GTX85454) and SREBP2 (SAB, 53501) overnight at 4°C. The next day, cells in confocal dishes were washed three times with PBS and incubated with secondary antibodies (donkey anti-chicken Alexa Fluor 488; donkey anti-rabbit Alexa Fluor 594) for 2 h at room temperature in the dark. The rest of the procedure is the same as for tissue immunofluorescence staining.

One side of the hippocampus was used for real-time PCR assays. Total RNA was extracted using a TRIzol reagent (Invitrogen, Carlsbad, United States) and reverse-transcribed into cDNAs using Hifair II 1st Strand cDNA Synthesis SuperMix for qPCR (Yeasen Biotechnology, Shanghai, China). The expression of the target gene was normalized to actin and quantified by the comparative cycle threshold method (2^−ΔΔCT). The primer sequences were as follows: FMO3 forward: 5’-CCCACATGCTTTGAGAGGAG-3′, reverse: 5’-GGAAGAGTTGGTGAAGACCG-3′; SREBP2 forward: 5’-GCCTCTCCTTTAACCCCTTG-3′, reverse: 5’-CCAGTCAAACCAGCCCCCAG-3′.

The hippocampus was trimmed to small pieces (1 mm³) and then was fixed in 2.5% glutaraldehyde in PBS and postfixed in 1% OsO4 at 4°C. The samples were then washed, dehydrated with a graded series of alcohol, incubated in 1% osmium tetroxide for 1 h at room temperature, and embedded in the medium, followed by ultrathin (80 nm) sectioning with an ultramicrotome. Electron micrographs of the hippocampal CA1 region were obtained utilizing a HITACHI electron microscope (HITACHI, H7650, Japan). Ten electron micrographs of each section were randomly taken from each field of view. The density of presynaptic vesicles, PSD length, width, and synaptic cleft width were quantified.

Mice were sacrificed immediately, and the brain tissues were immersed in the fixative (Servicebio, G1101) for more than 48 h. After that, the brain tissues were transferred into a Golgi-Cox staining solution (Servicebio, G1069) and kept in the dark for 14 days. Then, the brain tissues were immersed in distilled water 3 times and immersed in 80% glacial acetic acid overnight to become soft, and then placed in 30% sucrose. The brain sections were cut into 100 μm with a vibratome, pasted on a gelatin slide, and dried in the dark overnight. Images from hippocampal regions were collected using a confocal microscope (Nikon ECLIPSE E100). Dendrites in the bilateral hippocampus were traced from the cell soma to the projection terminus using the Neuron J plugin in ImageJ software. The Sholl analysis plug-in was employed for the analysis, and dendritic complexity was determined by the count of dendritic branch points at regular intervals from the cell bodies. The calculation of spine densities was performed on a single segment of dendrites longer than 10 μm in length from tertiary or higher-order apical and basal dendrites.

The supernatant of the primary astrocyte culture medium and hippocampal tissues of mice were collected. The cholesterol content was investigated using the mouse total cholesterol enzyme-linked immunosorbent assay (ELISA) kits (JM-02912 M1, Jingmeibio, Yancheng, Jiangsu, China), following the manufacturer’s protocols. Protein concentrations of culture medium and hippocampal tissues were both assessed. The results were presented by cholesterol concentrations/protein concentrations.

Serum (40 μL) was added into corresponding organic reagents according to the ratio of serum: methanol: acetonitrile (1:2:2, v/v/v), then vortexed fully for 2 min, and incubated at −20°C for 1 h. The mixtures were centrifuged at 13,000 rpm for 15 min at 4°C to precipitate the protein. The supernatant was lyophilized, and 100 μL 50% ACN was added for reconstitution, centrifuged, and tested on the machine. The hippocampus was immersed in 1.5 mL prechilled MeOH/H₂O (−80°C, 1:1, v/v), and the mixtures were well ground. The mixtures were centrifuged at 14,000 g for 10 min at 4°C and then transferred approximately 1.2 mL of the supernatant to the clean tube. The supernatants were dried in a vacuum concentrator, and all the dried samples were reconstituted with 120 μL MeOH/H₂O (1:1, v/v) solution for LC–MS analysis. Serum and hippocampal concentrations of TMAO were measured using liquid chromatograph coupled with a triple-quadrupole mass spectrometer (TSQ Quantis, Thermo Fisher). D9-TMAO was used as the internal standard.

The SREBP2 (544-1130aa) protein was selected with N-band trx tag for solubilization and C-band his tag for purification. It was constructed into pet32a vector, and Escherichia coli was used as the expression host to product purified SREBP2 proteins. Then, the purified SREBP2 proteins were labeled using Monolith Protein Labeling Kit RED-NHS (NanoTemper, Germany). TMAO was diluted with a 2-fold concentration gradient and incubated with 20 nM purified labeled SREBP2 protein. After loading the samples into NanoTemper glass capillaries, micro-thermophoresis was conducted with 80% light-emitting diode power and 80% MST. The equilibrium dissociation constant (KD) value was detected with the Monolith NT.115 (NanoTemper Technologies, Germany).

TMAO 3D structure of SDF format was downloaded from PubChem data, optimized by ChemBio3D Ultra, and then converted into Protein Data Bank (PDB) format by using AutodockTools. The protein structure of mouse SREBP2 was downloaded from the UniProt database and was imported into AutoDockTools for hydrogenation, calculate charge, distribute charge, specify atomic type, and saved as PBD format. POCASA was applied to predict protein binding sites, and AutoDock Vina was used for docking. Finally, the interaction mode of the docking results was analyzed by PyMOL.

Statistical analyses were performed using GraphPad Prism 7.0. All values, except for Sholl interaction, are presented as mean ± SD. Normal distribution and variance homogeneity were assessed for each dataset. One-way analysis of variance (ANOVA) followed by Bonferroni’s or Turkey’s post-hoc tests was conducted to assess the statistical significance of variations among multiple groups. P values < 0.05 were considered statistically significant (*p < 0.05, **p < 0.01, ***p < 0.001, ****p < 0.0001).

CSR mice were deprived of sleep for 20 h each day (from 8 to 4 pm the next day) for consecutive 1 week (Figure 1A). To explore whether TMAO is involved in CSR-related cognitive decline, we applied liquid chromatography–tandem mass spectrometry (LC/MS) to evaluate TMAO levels. As depicted in Figures 1B,C, the levels of TMAO in both the serum and brain were increased after CSR. To investigate the implication of TMAO on memory, we performed the NOR and Y-maze tests. In the NOR test, the CSR group and the TMAO group both spent much less time exploring a novel object than the control group (Figures 1D,E). At the same time, the CSR + TMAO mice spent a lower time exploring the novel object than the CSR mice (Figures 1D,E). However, there was no change in the total time to explore both objects in the four groups (Figure 1F).

In the Y-maze test, the spontaneous alternation was significantly reduced in the CSR mice and TMAO mice when compared to control mice (CON vs. CSR, p = 0.0006; CON vs. TMAO, p = 0.0016; Figures 1G,H) and was further decreased in the CSR + TMAO mice compared with CSR mice (p = 0.0074). The total distance showed no significant differences among the four groups (Figure 1I). Taken together, these results indicated that TMAO was involved in CSR-induced cognitive dysfunction.

We then assessed the expressions of SYP and PSD95 to explore the effect of TMAO on synaptic dysfunction. At the protein levels (Figure 2A), the expressions of SYP (CON vs. TMAO, p < 0.0001; CSR vs. CSR + TMAO, p = 0.0418; Figure 2B) and PSD95 (CON vs. TMAO, p < 0.0001; CSR vs. CSR + TMAO, p < 0.0001; Figure 2C) were both decreased in the mice supplemented with TMAO diet. In addition, the Western blotting results were corroborated by IF staining for SYP and PSD95 (Figures 2D–G). These data indicated that TMAO induced and aggravated synapse loss in the CSR mice.

Subsequently, the microscopy analysis results (Figure 2H) showed the reduction of PSD length (CON vs. TMAO, p = 0.0229; CSR vs. CSR + TMAO, p = 0.0168; Figure 2I) and width (CON vs. TMAO, p = 0.0006; CSR vs. CSR + TMAO, p = 0.0392; Figure 2J), as well as the number of presynaptic vesicles (CON vs. TMAO, p < 0.0001; CSR vs. CSR + TMAO, p = 0.0001; Figure 2K) were detected in the mice supplemented with TMAO diet. On the contrary, the TMAO-diet mice exhibited an increase in the width of the synaptic cleft (CON vs. TMAO, p < 0.0001; CSR vs. CSR + TMAO, p < 0.0001; Figure 2L).

We next investigated neural morphology by Golgi staining (Figures 2M,N). TMAO led to a significantly lower dendrite length (CON vs. TMAO, p < 0.0001; CSR vs. CSR + TMAO, p = 0.0001; Figure 2O) and lower number of dendrites (CON vs. TMAO, p < 0.0001; CSR vs. CSR + TMAO, p = 0.0129; Figure 2P). Furthermore, Sholl’s analysis revealed a significant reduction in the dendritic complexity of pyramidal neurons in the TMAO mice, which was aggravated in the CSR + TMAO mice (Figure 2Q). A significant reduction in the density of dendritic spines was also observed in the TMAO mice (CON vs. TMAO, p < 0.0001; Figure 2R) and decreased in CSR + TMAO mice (CSR vs. CSR + TMAO, p < 0.0001; Figure 2R). Collectively, these findings demonstrated that TMAO supplement diet related to synaptic proteins loss and dendritic morphology defects.

To evaluate the impact of CSR on cholesterol metabolism, we examined the hippocampal cholesterol content and the expression of its major transcription factor SREBP2. We found that CSR mice displayed a notable reduction in brain cholesterol content (Figure 3A). Furthermore, Western blotting showed that SREBP2 was reduced in CSR mice (Figures 3B,C). Similarly, the qPCR result was consistent with Western blotting (Figure 3D). In addition, qPCR results of HMGCR, a downstream of SREBP2, showed that HMGCR reduced in CSR mice, which indicated that the activity of SREBP2 decreased (Supplementary Figure S1A). To further evaluate whether elevated concentrations of TMAO in serum and brain induced cholesterol deficits, mice were supplemented with 0.12% TMAO. We detected that TMAO caused suppression of cholesterol levels (CON vs. TMAO, p < 0.0001; CSR vs. CSR + TMAO, p = 0.0480; Figure 3A). As expected, TMAO reduced the expression of SREBP2 at the protein level (CON vs. TMAO, p < 0.0001; CSR vs. CSR + TMAO, p = 0.0018; Figures 3B,C) and mRNA level (Figure 3D).

To verify that TMAO-induced reduction in brain cholesterol level is because of the reduction in cholesterol level in astrocytes, we performed in vitro experiments. We stimulate primary astrocytes with 0, 50 μM, 100 μM, 200 μM, or 400 μM TMAO. The Western blotting analysis showed a substantial decrease in astrocytic SREBP2 protein expression following treated with 100–400 μM TMAO (p < 0.05; Supplementary Figure S2A) and the SREBP2 expression showed no significant differences among these three groups (p > 0.05; Supplementary Figure S2B). Accordingly, TMAO (100 μM) and LPS (100 ng/mL) were used to stimulate primary astrocytes. The cholesterol concentration in the culture medium revealed that cholesterol concentration was remarkably lower in the TMAO group than in the CON group (p < 0.0001; Figure 3E), which was aggravated by LPS (LPS vs. LPS + TMAO, p < 0.0001; Figure 3E). As depicted in Figures 3F,G, SREBP2 expression was decreased in the TMAO group compared to the CON group (p = 0.0080) and was further decreased in the LPS + TMAO group (LPS vs. LPS + TMAO, p = 0.0213). Immunofluorescence results were consistent with Western blotting analysis (Figures 3H,I). Above all, we confirmed that TMAO downregulated cholesterol metabolism by inhibiting SREBP2 expression in CSR mice.

Subsequently, we assessed the relationship between TMAO and SREBP2 by molecular docking and MST experiment. The molecular docking predicted results showed that TMAO could form a critical hydrogen bond and hydrophobic force with SREBP2 (Figure 3J), which is the key to the interaction. The result of mean binding energies of docking for the TMAO with SREBP2 was −5.8 kcal/mol, proving that it has a better binding effect. MST is a method that labeled SREBP2 protein with fluorescence and required increasing concentrations of TMAO with an established quantity of SREBP2 protein to determine KD values in solution. TMAO binding to SREBP2 protein yielded a KD value of 5.44 ± 1.89 μM, which belongs to strong binding in the field of protein-small molecule binding (Figure 3K).

Figure 4A shows the experimental process of DIM intervention. To determine the cause of elevated TMAO levels in the CSR mice, Western blotting was applied to evaluate hepatic FMO3 expression level. The results showed that hepatic FMO3 was significantly increased in CSR mice (Figures 4B,C). Consistently, qPCR confirmed a further increased expression of FMO3 mRNA level in CSR mice (Figure 4D). Accordingly, the mice were treated with the FMO3 inhibitor 3,3′-diindolylmethane (DIM; Cashman et al., 1999; Katchamart et al., 2000). Mice were placed on a diet with or without 0.25% DIM supplementation for 4 weeks (Chen et al., 2019). As expected, DIM inhibited hepatic FMO3 protein level in the CSR mice (CSR vs. CSR + DIM, p < 0.0001; Figures 4B,C) as well as mRNA level (Figure 4D). Meanwhile, serum and hippocampal TMAO concentrations were decreased after DIM treatment (CSR vs. CSR + DIM: serum, p = 0.0292; brain, p = 0.0234; Figures 4E,F).

In contrast, Western blotting results showed that the SREBP2 protein expression was significantly increased in the CSR + DIM group than in the CSR group (Figures 4G,H). The mRNA level of SREBP was consistent with Western blotting (Figure 4I). Immunofluorescence results were consistent with Western blotting and PCR analysis (Figures 4J,K). Consistently, brain cholesterol concentrations of CSR mice were also increased following DIM treatment (Figure 4L). Overall, these results confirmed that DIM improved CSR-induced astrocytic cholesterol deficits through TMAO downregulation.

To investigate the impact of DIM on synaptic dysfunction induced by CSR, we assessed the expression of SYP and PSD95 in CSR + DIM mice. At the protein levels (Figure 5A), DIM dramatically improved CSR-induced reduction in SYP (CSR vs. CSR + DIM, p = 0.0007; Figure 5B) and PSD95 (p < 0.0001; Figure 5C). In addition, immunofluorescence staining results showed that the quantification of puncta in PSD95 and SYP and their colocalization showed a considerable loss of synapses in CSR mice (Figures 5D–G), while SYP and PSD95 puncta and colocalized puncta were restored in the CSR + DIM mice compared with CSR mice (Figures 5D–G). These findings suggested that DIM alleviated synapse loss after CSR.

Moreover, electron microscopy analysis was utilized (Figure 5H). In CSR mice, PSD length (Figure 5I) and width (Figure 5J) and the number of presynaptic vesicles (Figure 5K) were all reduced compared with the CON group. On the contrary, there was an observed increase in the width of the synaptic cleft in CSR mice (Figure 5L). Nevertheless, DIM was found to reverse these alterations in synaptic ultrastructure (CSR vs. CSR + DIM; Figures 5I–L).

We next investigated neural morphology by Golgi staining (Figures 5M,N). The CSR group exhibited a statistically significant reduction in dendrite length (Figure 5O) and dendrite number (Figure 5P) of hippocampal pyramidal neurons compared to the control group, which were partly reversed by DIM (Figures 5O,P). Moreover, compared with control mice, the Sholl intersections were fewer in pyramidal neurons of CSR mice at a distance of 30–80 μm from soma to the terminal (Figure 5Q), which were also partly rescued by DIM (CSR vs. CSR + DIM: 50 μm, p = 0.0248), suggesting that DIM mildly enhanced dendrite complexity of CSR mice. In parallel, a significant decrease in the density of dendritic spines was detected in the CSR mice (Figure 5R) but markedly increased in CSR + DIM mice (CSR vs. CSR + DIM: p < 0.0001). Taken together, these findings demonstrated that DIM reversed the depletion of synapse-related proteins and dendritic morphological abnormalities in CSR mice.

To assess the effect of DIM on memory in CSR mice, we performed NOR and Y-maze tests to evaluate the ability of mice. In the NOR test, the CSR + DIM mice explored the novel object for a longer period of time compared to the CSR mice (Figures 6A,B). However, the total time to explore both objects was not different among the four groups (p > 0.05; Figure 6C). In the Y-maze test, the spontaneous alternation was increased in the CSR + DIM mice than in the CSR mice (Figures 6D,E). The total distance showed no significant differences among the four groups (p > 0.05; Figure 6F). These results indicated that DIM effectively attenuated CSR-induced memory deficits.