ApoE^-/- mice (22 ± 2 g) were obtained from Nanjing Junke Biological Engineering Co., Ltd. Throughout the experiment, adequate food and water were provided. All procedures were approved by the Institutional Animal Care and Use Committee of Harbin Medical University [Protocol (2009)-11]. The use of animals was compliant with the Guide for the Care and Use of Laboratory Animals published by the U.S. National Institutes of Health (NIH Publication No. 85-23, revised 1996). All mice were randomly divided into normal (N), high-fat diet (H), and high-fat diet plus resveratrol (HR) groups. Mice in N, H, and HR groups were administered with standard chow diet (normal, N) or HFD (0.3% cholesterol and 21% (wt/wt) fat) for 24 weeks, respectively. REV (Sigma-Aldrich, Munich, Germany) was administered by oral gavage to the mice (10 mg/kg/day, twice a day) (Chen et al., 2018).

Twenty-four weeks after treatment, the en face aortas and aorta roots of mice were collected. The aforementioned samples were fixed with 4% PFA overnight and dehydrated with 30% sucrose. And then, the samples were embedded in OCT and frozen at −80°C (Chen et al., 2018). For the morphometric analysis, serial sections were cut into 6-μm-thickness slides using a cryostat. The sections were stained with hematoxylin and eosin (HE) for the quantification of the lesion area. Oil Red O staining was performed to indicate the lipid content in the lesions with an Oil Red O staining kit (Nanjing Jiancheng Biology Engineering Institute, Nanjing, Jiangsu, China) according to the manufacturer’s instructions. Aortic lesion size was obtained by averaging the lesion areas in four slides (12 sections) from the same mouse. The lesion area was analyzed as a percentage of the Oil Red O–stained area in the total aorta area. Every four slides from the serial sections were stained with HE, and each consecutive slide was stained with Oil Red O for the quantification of the atherosclerotic lesion area.

The mice were fasted for 12–14 h before blood samples were obtained by retro-orbital venous plexus puncture. Before retro-orbital bleeding was conducted, the topical ophthalmic anesthetic—proparacaine—was applied. The study was conducted in compliance with the NIH Guide. Then plasma was collected by centrifugation and kept at −80°C. Total plasma cholesterol (TC), triglycerides (TG), high-density lipoprotein (HDL), and low-density lipoprotein (LDL) were enzymatically detected according to the manufacturer’s instructions (Nanjing Jiancheng Bioengineering Institute, China).

Liver tissues from the mice were collected for metabolite extraction. The tissues were thawed on ice and grinded with liquid nitrogen. Then the metabolites were extracted (Huang et al., 2021). The repeatability and stability of LC-MS analysis were evaluated using the quality control (QC) sample. The QC samples were prepared by combining equal volume of each extraction. There are 12, 6, 7, and 5 samples in N, H, HR, and QC groups, respectively.

All samples were analyzed using a TripleTOF 5600+ high-resolution tandem mass spectrometer (SCIEX, Warrington, United Kingdom) with both positive and negative ion modes. Chromatographic separation was conducted using an ultra-performance liquid chromatography (UPLC) system (SCIEX, United Kingdom). An ACQUITY UPLC T3 column (100 mm × 2.1 mm, 1.8 µm, Waters, United Kingdom) was applied for the reverse-phase separation. It was introduced for the separation of metabolites as the mobile phase consisted of solvent A (water, 0.1% formic acid) and solvent B (Acetonitrile, 0.1% formic acid). The gradient elution conditions were as follows, with a flow rate of 0.4 ml/min: 5% solvent B for 0–0.5 min; 5%–100% solvent B for 0.5–7 min; 100% solvent B for 7–8 min; 100%–5% solvent B for 8–8.1 min; and 5% solvent B for 8.1–10 min. The column temperature was maintained at 35°C. The TripleTOF Metabolites were measured by a high-resolution tandem mass spectrometer (TripleTOF 5600 Plus; SCIEX, Warrington, United Kingdom). To evaluate the stability of LC-MS, the QC samples were analyzed randomly.

The acquired LC-MS data pretreatment was analyzed using XCMS software. Raw data files were converted into an mzXML format and then processed using the XCMS (Tautenhahn et al., 2012), CAMERA, and MetaX (Wen et al., 2017) in R software. Each ion was identified by the comprehensive information of retention time and m/z. The intensity of each peak was recorded. Then the information was matched to the in-house and public database including HMDB (http://www.hmdb.ca/), METLIN (http://metlin.scripps.edu/), Massbank (http://www.massbank.jp), PubChem (http://ncbi.nlm.nih.gov/) and KEGG (http://www.kegg.com/). Metabolites detected in ≥50% of QC samples or in ≥80% of total samples were collected. The missing data were extrapolated with the k‐nearest neighbor (KNN) algorithm. The data were processed by the probabilistic quotient normalization (PQN) algorithm and corrected by QC-robust spline batch correction (QC-RSC) using QC samples. Metabolic features with standard deviations ≤30% were collected.

Metabolites with a one-way ANOVA p value < 0.05 and PLS-DA VIP >1 were selected to conduct trend analysis using the Short Time-series Expression Miner 1.3.11 (STEM). The metabolites altered in the H group were compared with N group and recovered by REV were collected.

The Student t‐test was used to compare the difference between two groups. P value was adjusted by using the Benjamini–Hochberg method. Principal component analysis (PCA) and partial least-squares discriminant analysis (PLS-DA) were conducted to identify differences between groups using MetaboAnalyst 3.0 (Xia et al., 2015). Features with a VIP >1.0 were collected. Two‐tailed Mann–Whitney U tests and two‐independent sample t-tests were performed using MeV 4.9.0 and PASW Statistics 18 software (SPSS Inc, Chicago, United States), respectively, to evaluate differences in metabolite levels (Saeed et al., 2006). Comparisons among three groups were performed using one-way ANOVA, followed by the Bonferroni post hoc test. A two-tailed p value < 0.05 was considered statistically significant. The data are expressed as mean ± SEM.

Oil Red O staining and HE staining were conducted to evaluate the effect of REV on atherogenesis. High-fat diet (HFD)–fed mice display more prominent features of atherosclerosis (AS), characterized by large atherosclerotic plaques and lipid accumulation in the lesions. However, resveratrol (REV) treatment arrested diet-induced AS that was observed from en face aortas and aortic roots (Figures 1A,C). The quantitative analysis was also conducted and showed its corresponding bottom values (Figures 1B,D). Next, HE staining was also performed to validate the anti-atherosclerotic effect of REV (Figures 1E,F). These results were in accordance with our previous findings that REV possesses anti-atherosclerotic activity (Ye et al., 2019).

Anomalies of serum lipid levels are the main reason for atherogenesis. We observed that long-term administered resveratrol (REV) could obviously reduce the plasma levels of total cholesterol (TC) (Figure 2A), triglycerides (TG) (Figure 2B), and low-density lipoprotein cholesterol (LDL-C) (Figure 2C) and dominantly improve high-density lipoprotein (HDL) (Figure 2D) in the ApoE^-/- mice fed with HFD. Moreover, REV decreased liver lipid accumulation in HFD-fedafter treatment with REV in the liver of ApoE^-/- mice (Figure 2E).

The metabolic profiles of liver tissues from the three groups were analyzed by LC-MS. The quality of detection was analyzed by XCMS software. The total ion chromatogram (TIC) showed the separation of all metabolites in the UPLC. In both the ESI+ and ESI- model, there are differences among these groups, indicating that HFD and REV treatment affected liver metabolite expression (Figures 3A,B). Whether these altered metabolites are involved in the pathogenesis and REV treatment of AS needs further analysis. In mass spectrometry, each substance has specific m/z and rt (Figure 3C).

The PCA showed the different variables among the 3 groups, indicating that the three groups exhibited obviously different metabolites (Figures 4A,B). QC samples exerted high reproducibility with each other. The total ions were acquired from the QC samples. Among the total 16,558 ions obtained from the QC samples, the relative standard deviation (RSD) value of 9,719 ions (58.7% of total ions) was less than 20% (Figure 4C). PLS-DA was performed to further delineate the metabolic differences among the three groups (Figures 4D–G). The data distribution between the high-fat diet (H) and normal (N) groups (Figures 4D,E), as well as high-fat diet (H) and high-fat diet plus resveratrol (HR) groups was different (Figures 4F,G), respectively.

The heat map and volcano plot revealed different metabolite profiles between N and H groups, as well as H and HR groups (Figures 5A–D). Differentially expressed features were identified as |log2 (fold change)| ≥ 1, Q ≤ 0.05, and VIP ≥ 1. A total of 1,146 and 765 differentially expressed features were identified between N and H groups, as well as H and HR groups, respectively (Figures 5A–D; Table 1).

Then the KEGG pathway analysis was also performed. It was shown that the differentially expressed metabolites between N and H group were enriched in pathways including “glycerophospholipid metabolism,” “metabolic pathways,” and “phenylalanine metabolism” (Figure 5E). And the differentially expressed metabolites between H and HR groups was enriched in pathways including “phenylalanine metabolism,” “glycerophospholipid metabolism,” and “primary bile acid biosynthesis” (Figure 5F). There are many overlapped pathways between N/H and H/HR; these pathways are potential pathogenesis mechanisms and therapeutic targets, especially metabolism-related pathways including “primary bile acid biosynthesis,” “phenylalanine metabolism,” and “glycerophospholipid metabolism.”

Five hundred fifty-five metabolites with a one-way ANOVA p < 0.05 and PLS-DA VIP >1 were selected to conduct trend analysis using STEM software. The trend analysis images represent trends in metabolites across the multiple comparison groups, and each small image represents one trend. Metabolites in trends 2 and 5 were selected, and KEGG enrichment was performed using MBRole 2.0 (http://csbg.cnb.csic.es/mbrole2/index.php) (Figure 6A). Metabolites in trend 2 were downregulated by HFD and recovered by REV. On the contrary, metabolites in trend 5 were upregulated by HFD and recovered by REV. The top four enriched pathways of metabolites in trends 2 and 5 were “biosynthesis of unsaturated fatty acids,” “intestinal immune network for IgA production,” “glycerophospholipid metabolism,” and “pathways in cancer” (Figure 6B). The top four enriched pathways of metabolites in trends 2 were “caffeine metabolism,” “intestinal immune network for IgA production,” “small-cell lung cancer,” and “propanoate metabolism” (Figure 6C). And the top four enriched pathways of metabolites in trend 5 were “biosynthesis of unsaturated fatty acids,” “fatty acid biosynthesis,” “linoleic acid metabolism,” and “aldosterone-regulated sodium reabsorption” (Figure 6D).

Based on the KEGG analysis results, the metabolites in pathways with high significance were selected. Xanthine, xanthosine, retinoic acid, succinic acid, and propionic acid were increased in the liver of mice with high-fat diet, whereas REV reversed the expression of these metabolites (Figures 7A–E). On the contrary, some metabolites were decreased in the liver of mice with high-fat diet, the expression of which was reversed by REV, including oleic acid, stearic acid, alpha-linolenic acid, docosapentaenoic acid, and 11,14,17-eicosatrienoic acid (Figures 7F–J). These metabolites are potential biomarkers for liver metabolic reprogramming induced by HFD.