QFY was prepared by Lunan Pharmaceutical Group Co., Ltd. (Shandong, China, Batch number: 2209001). Major constituents were as follows: Panax ginseng 3.0 kg, cooked Rehmannia Glutinosa 4.50 kg, Angelicae Sinensis 4.50 kg, fried Atractylodis Macrocephalae Rhizoma 2.50 kg, Semen Ziziphi Spinosae 3.0 kg, prepared Polygala 2.50 kg, roasted Glycyrrhiza 1.50 kg.

For the above seven ingredients, Ginseng was refluxed with 60% ethanol twice, each time lasting 1.5 h. Then it was filtered, and the drug residues were reserved for later use. The filtrate was subjected to ethanol recovery and concentrated until the relative density reached 1.03–1.10 (at 60 °C), and then reserved for subsequent steps.

Radix Angelicae Sinensis and fried Atractylodis Macrocephalae Rhizoma were used to extract volatile oil by the water distillation method. The aqueous solution after distillation was collected in another container, and the drug residues were reserved. The volatile oil ethanol solution was included with beta-cyclodextrin, dried, and pulverized for reserve.

The above three kinds of drug residues were combined with the other four ingredients, including cooked Rehmannia glutinosa, etc., and then decocted with water twice, with each time lasting for 2 h. After filtration, the resulting liquid was combined with the concentrated solution of Ginseng and the liquid after oil extraction, and then concentrated to obtain a clear paste. It was then allowed to stand still and centrifuged, followed by further concentration to obtain an extract with a relative density of 1.22–1.28 (at 60 °C). Subsequently, it was dried and pulverized. Finally, the beta-cyclodextrin inclusion complex was added and mixed evenly.

The 5xFAD transgenic mice were acquired from Jackson Laboratories (Stock No.034840, Bar Harbor, ME, USA) and bred to ensure the continuation of the colony. PCR was used to confirm the genotypes of animals: Hemizygous 5xFAD mice and their wild-type littermates were employed for the experiment. The wild-type control group consisted of C57BL/6JXSJL mice without the 5xFAD transgene, genetically matched to the experimental group to minimize background-related variability. The mice were housed in the SPF barrier environment with a 12-h light/dark cycle, a temperature of 22 ± 2 °C, and a relative humidity of 55 ± 10%, with free access to food and water.

When male mice were 4.2 months old, they were randomly split into three groups (n = 21–27): (1) QFY group, comprising 5xFAD mice that were intervened with QFY (2.1 g/kg/d) via intragastric administration once a day; (2) wild type(WT) group, comprising C57/B6XSJL mice that were intervened with equivalent volume of water; (3) 5xFAD group, comprising 5xFAD mice that were intervened with equivalent volume of water. The behavioral experiments were employed from 134 days to 209 days of QFY feeding (Figure 1). After that, the brain and feces were collected for lipofuscin staining, SA-β-gal and Aβ co-staining, metabolomic profiling, and 16S rRNA sequencing.

Six mice were randomly selected from each of the control, model, and QFY groups. Feces were collected from all mice 1 day before sacrifice and stored at −80 °C until testing. The mice were sacrificed by decapitation, and the whole brains were isolated. The whole brains were rinsed thoroughly with water and then split into two halves along the gap between the left and right cerebral hemispheres. Three of the left hemispheres were immersed in alcohol and xylene sequentially. They were then immersed in a 60 °C wax solution for 3 h, transferred to an embedding cassette, and stored for later use once the wax solution solidified. Three of the left hemispheres were fixed in 4% paraformaldehyde, followed by gradient sedimentation in 15 and 30% sucrose solutions sequentially. They were then embedded in optimal cutting temperature compound (OTC) and stored in a −80 °C refrigerator for pathological staining. The cerebral cortex tissue of the right hemisphere was dissected, placed into a centrifuge tube, snap-frozen in liquid nitrogen, and then stored in a −80 °C refrigerator for subsequent untargeted metabolomics analysis.

The left hemisphere samples were cut into 4 μm sections and conventionally dewaxed and hydrated. The ferric chloride solution, potassium ferricyanide solution, and distilled water were mixed at a ratio of 7.5:1:1.5 (Gefanbio, Shanghai, China), and the samples were treated for 8 min. The neutral red staining solution (Gefanbio, Shanghai, China) was used to cover the samples for 8 min after washing with water. Samples were washed with water, dehydrated in a hot air circulating oven, cleared with xylene, and mounted with resin. Results were observed using a light microscope.

Three left hemisphere samples were removed from the −80 °C refrigerator, equilibrated to −20 °C, and cut into 40 μm sections. β-gal Fixative (Solarbio, Beijing, China) was added to cover samples for 15 min after washing with PBS. SA-β-gal dyeing working solution (Solarbio, Beijing, China) was added and maintained at 37 °C overnight after removal of β-gal Fixative and washing with PBS three times. 70% ethanol was used to dissolve crystals, and xylene was added to make the samples transparent. Results were observed using a light microscope.

Samples were hydrated with ethanol and washed with PBST, and immersed in the citrate buffer solution at 99 °C for 25 min. The blocking solution (Absin, Shanghai, China) was introduced for 10 min after washing with PBST three times. Anti-Aβ1–42 was incubated (1:400, Abcam, Cambridge, UK) at 4 °C for the entire night after washing with PBST three times. Enhanced enzyme-labeled IgG polymer (Absin, Shanghai, China) was incubated for 35 min under dark conditions after washing with PBST three times. The DAB kit (Absin, Shanghai, China) was used for color development for 15 min after washing with PBST. Samples were dehydrated and cleared with alcohol and xylene, and sealed with neutral gum.

Results were observed using a light microscope. One section per sample was selected for imaging: hemibrain sections were captured at 2 × magnification, while hippocampal and cortical regions of interest were imaged at 40 × magnification in single-plane mode. Quantification was performed using ImageJ.¹ Briefly, color images were first split into individual channels, and the suitable channel was selected. A consistent threshold was applied across all samples to distinguish positive staining signals from background. Then the positive area fraction was calculated.

The prepared cerebral cortex tissue of the right hemisphere was ground in an Eppendorf tube containing a tungsten bead for 1 min at 65 Hz using a Grinding Mill. Metabolites were extracted via a pre-chilled mixture of methanol, acetonitrile, and water (v/v/v, 2:2:1). The mixture was then ultrasonicated and centrifuged. The supernatants were retrieved and concentrated to dryness under a vacuum environment, then dissolved in 50% acetonitrile. This was transferred into HPLC vials after being filtered with 0.22 μm cellulose acetate for analysis.

A UPLC-ESI-Q-Orbitrap-MS system (UHPLC, Shimadzu Nexera X2 LC-30 AD, Shimadzu, Japan) and Q-Exactive Plus (Thermo Scientific, San Jose, USA) were used to analyze the metabolomic profiling. LC separation was performed through the ACQUITY UPLC® HSS T3 column (2.1 × 100 mm, 1.8 μm) (Waters, Milford, MA, USA). SIMCAP software (Version 14.0, Umetrics, Umeå, Sweden) was utilized to conduct all multivariate data analyses and build models. Pathway analysis was carried out on the differential metabolite data using the MetaboAnalyst database² and the KEGG database.³

The extraction of total genomic DNA samples in feces was accomplished with the use of the OMEGA Soil DNA Kit (M5635-02) (Omega Bio-Tek, Norcross, GA, USA). After detecting the DNA purity, PCR amplification of the 16S rRNA gene at the V3-V4 region was conducted using the forward primer 338F (5′-ACTCCTACGGGAGGCAGCA-3′) and the reverse primer 806R (5′-GGACTACHVGGGTWTCTAAT-3′).

QIIME2 and R packages (v3.2.0) were used following the official tutorials⁴ for sequence data modification and analyses. Sequences were quality filtered, denoised, merged, and chimera removed through the DADA2 plugin. Mafft was employed to align non-singleton amplicon sequence variants (ASVs). Alpha-diversity metrics (Chao1, Shannon) were visualized as box plots, and beta-diversity metrics were visualized via principal coordinate analysis (PCoA) using the diversity plugin. LEfSe (Linear discriminant analysis effect size) was performed to detect differentially abundant taxa across groups with the default parameters.

Spearman’s correlation analysis was performed to analyze the relationship between behavioral data, differential metabolites, and differential gut microbiota in the control, 5xFAD, and QFY groups.

Data were presented as mean ± SEM and analyzed by GraphPad Prism 8. Normality was tested using the Shapiro–Wilk test. Results were analyzed using the unpaired Student’s t-test. Repeated tasks were assessed individually at each time point only. Statistical significance was considered at p < 0.05.

On the 134th, 179th, and 209th days of QFY intervention, grip strength test and rotarod test were conducted to evaluate the forelimb strength and motor coordination ability of male mice (Figure 1). The grip strength/weight was significantly greater in the QFY group relative to the model group at day 179 (p < 0.01) and day 209 (p < 0.01) (Figure 2A). The area under the curve (AUC) of the grip strength/weight was significantly greater in the QFY group relative to the model group (p < 0.01) (Figure 2B). Scatter plot with linear fit showed greater value of slope in the QFY group relative to the model group, indicating that the grip strength/weight decreased more slowly with age in 5xFAD mice with QFY intervention, but there were no significant differences (Figure 2C).

The riding time/weight significantly enhanced in the QFY group relative to the model group at day 134 (p < 0.05) and day 180 (p < 0.05) (Figure 2D). Riding time was normalized to body weight (expressed as s/g) to account for the well-documented negative correlation between body weight and rotarod performance in C57BL/6 J mice (McFadyen et al., 2003). The AUC of the riding time/weight was significantly greater in the QFY group relative to the model group (p < 0.05) (Figure 2E). Scatter plot with linear fit showed a smaller value of slope in the QFY group relative to the model group, indicating that QFY might have a stronger enhancing effect on motor coordination ability in the early stage of 5xFAD mice, but there were no significant differences (Figure 2F).

After QFY treatment for 242 days, a grading score of frailty was employed. The frailty score of the QFY group was lower than that of the 5xFAD group, but there were no significant differences (Figure 2G).

These results indicated that QFY could delay physiological frailty in male 5xFAD mice.

We detected the deposition of lipofuscin, SA-β-gal, and Aβ in the brains of mice (Figure 3A). Compared with WT mice, SA-β-gal-positive cells, amyloid deposition, as well as co-located amyloid deposition and SA-β-gal-positive cells were observed in both the hippocampus (using the CA3 region as an example) and cortexes of 5xFAD mice. These positive signals were less abundant in 5xFAD mice with QFY intervention, and statistically significantly lower levels were observed in three metrics: the Aβ-positive area in the hippocampal CA3 region, the Aβ-positive area in the cortex, and the Aβ/SA-β-gal co-localized area in the cortex (Figures 3B–G).

Studies have shown that the CA3 subregion of the hippocampus exhibits a higher accumulation of lipofuscin compared to other hippocampal subregions, and this accumulation is directly associated with neuronal functional decline (Sharma et al., 1993). Therefore, the CA3 subregion was selected to demonstrate the lipofuscin deposition. Qualitative observations suggest blue lipofuscin deposition appeared in the CA3 regions of the hippocampus of the 5xFAD mice compared with the WT mice. Additionally, a perceptible reduction in lipofuscin staining intensity was observed in the 5xFAD mice with QFY intervention relative to the 5xFAD mice, but this trend of less lipofuscin deposition remains preliminary and requires confirmation through quantitative analyses (Figure 3H). These results implied that QFY also slowed brain aging and alleviated the typical pathological features of AD in male 5xFAD mice.

The alterations in metabolites within the cerebral cortex were investigated using untargeted metabolomics. PLS-DA revealed distinct group separations among the three groups (Figure 4A). The overfitting in the PLS-DA model was evaluated by 200 repetitions of random permutation testing (RPT). The Q2 value was less than the R2 value, indicating the robustness of the PLS-DA model (Figure 4B).

Metabolites exhibiting significant differences were identified using a criterion VIP > 1 in the PLS-DA model and a significance level of p < 0.05. Most of the differential metabolites between the WT mice and the 5xFAD mice, as well as between the 5xFAD mice and the 5xFAD mice with QFY treatment, were lipids and lipid-like molecules. Hierarchical clustering analysis was used to screen significantly different metabolites between the WT mice and the 5xFAD mice, as well as between the 5xFAD mice and the 5xFAD mice with QFY treatment (Figures 4C,D). Several metabolites including proline (p < 0.05), PS (18:1/18:0) (p < 0.05), and PFSA-CI (p < 0.05) were downregulated in the 5xFAD mice relative to WT mice, while PE (18:1/18:1) (p < 0.05) was upregulated in the 5xFAD mice relative to WT mice, and these were all reversed by the intervention of QFY (Supplementary Table 2).

In the pathway enrichment analysis of differential metabolites, arginine and proline metabolism, alanine, aspartate and glutamate metabolism, and butyrate metabolism were screened as key metabolic pathways between WT mice and 5xFAD mice, as well as between 5xFAD mice and 5xFAD mice with QFY treatment (Figures 4E,F).

16S rRNA sequencing was conducted to identify the alterations in the gut microbiota of 5xFAD mice following QFY treatment. A total of 1,795,371 valid sequences and 2,804 ASVs were obtained from the sequencing of 18 samples, and the lengths of the sequences mostly fell within the range of 404–432 bp. The Shannon (diversity index) and Chao indices (richness index) were applied to analyze the α-diversity data, which showed no significant differences among the three groups. With an increase in sampled sequence numbers, the Shannon and Chao curves flattened, demonstrating that the amount of sequencing data was reasonable (Figure 5A). Principal co-ordinates analysis (PCoA) demonstrated a similar trend and closer proximity between the WT mice and the 5xFAD mice with QFY treatment (Figure 5B), while significant variations of gut microbiota were found in three groups at the genus level (Figure 5C). These findings suggested that QFY induced modifications in the overall structure of gut microbiota in 5xFAD mice to a certain extent.

At the genus level, 221 intestinal bacteria were identified among the three groups. The relative abundance analysis of the three groups showed that the main dominant bacterial group was Muribaculaceae (Figure 5D). Moreover, the LEfSe method was applied to single out the remarkably distinct specific bacteria among the three groups, based on the screening criteria of LDA > 2 and p < 0.05 (Figure 5E). 38 specific key genera were found in total, with 16 enriched in control, 11 in model, and 11 in QFY. These subtypes significantly contributed to the discrimination among the three groups. The abundance of 31 intestinal bacterial genera was significantly different between WT mice and 5xFAD mice, as well as between 5xFAD mice and 5xFAD mice with QFY treatment (Figure 5F). Remarkably, the abundance of gut microbiota taxa, such as Ruminococcaceae (p < 0.05), Subdoligranulum (p < 0.05), Bacteroides (p < 0.05), Alistipes (p < 0.05), Rikenellaceae_RC9_gut_group (p < 0.05), Odoribacter (p < 0.05), Desulfovibrio (p < 0.05), Colidextribacter (p < 0.05), Incertae_Sedis (p < 0.01), Lachnospiraceae_UCG-006 (p < 0.05), Clostridia_vadinBB60_group (p < 0.01), Anaeroplasma (p < 0.05), [Eubacterium]_brachy_group (p < 0.05), Paludicola (p < 0.05), [Eubacterium]_nodatum_group (p < 0.05), and Harryflintia (p < 0.05), was lower in 5xFAD mice relative to WT mice, while it was dramatically higher in the QFY group compared to the 5xFAD group. The abundance of Muribaculaceae (p < 0.01), Paraprevotella (p < 0.05), Muribaculum (p < 0.05), UCG-010 (p < 0.05), Erysipelatoclostridium (p < 0.05), Adlercreutzia (p < 0.05), and Gordonibacter (p < 0.05) was higher in 5xFAD mice relative to WT mice, while it was lower in the QFY group compared to the 5xFAD group (Figure 6). These results indicated that QFY modulated the composition of gut microbiota in male 5xFAD mice.

Spearman’s correlation analysis was employed to elucidate the associations among behavioral data, differential metabolites, and genus-level differential gut microbiota. In the correlation analysis between behavioral data and differential metabolites, proline (p < 0.05), PS (18:1/18:0) (p < 0.05), and PFSA-CI (p < 0.05) exhibited significant positive correlations with the grip strength/weight or the riding time/weight, while PE (18:1/18:1) (p < 0.05) showed a significant negative correlation with the grip strength/weight (Figure 7A).

In the correlation analysis between behavioral data and genus-level differential gut microbiota, Ruminococcaceae (p < 0.01), Subdoligranulum (p < 0.05), Bacteroides (p < 0.05), Alistipes (p < 0.05), Rikenellaceae_RC9_gut_group (p < 0.05), and Odoribacter (p < 0.05) were significantly positively correlated with the grip strength/weight (Figure 7B).

In the correlation analysis between differential metabolites and genus-level differential gut microbiota, proline showed significant positive correlations with Ruminococcaceae (p < 0.05), Subdoligranulum (p < 0.01), and Bacteroides (p < 0.05). Moreover, PS (18:1/18:0) showed significant positive correlations with Ruminococcaceae (p < 0.01). In addition, PFSA-CI showed significant positive correlations with Subdoligranulum (p < 0.01). However, PE (18:1/18:1) was significantly negatively correlated with Ruminococcaceae (p < 0.05), Subdoligranulum (p < 0.05), Bacteroides (p < 0.01), and Alistipes (p < 0.05) (Figure 7C).