Animals and experimental design
The present study was conducted following the Regulations for the Administration of Affairs Concerning Experimental Animals of the State Council of the People’s Republic of China. The Committee on Experimental Animal Management of the Chinese Academy of Agricultural Sciences, (Beijing), approved our research protocol (Approval No. 46/28.05.2016).
We conducted this research at the Institute of Grassland Research of the Chinese Academy of Agricultural Sciences Research Station in 2016 during the summer. The site is located at the Xilin River basin in the Inner Mongolia Autonomous Region of China (116°32’ E, 44°15’ N). Its predominant natural vegetation included three species of grass: Leymus chinensis, Stipa krylovii, and S. grandis.
In June 2016, a total of 48 Uzhumchin wethers (average age: 24 months; baseline live weight (LW):33.2 ± 3.9 kg) were obtained from the Xinshengjia Sheep Breeding Company (Xilinhot, Inner Mongolia, China). We randomly assigned the wethers to either the light grazing (LG) (4 sheep/plot) or the overgrazing (OG) (12 sheep/plot) groups. The experimental site comprised six plots (1.33 ha/plot, three plots per grazing group). Two different stocking rates were used, namely, 3.0 (LG) and 9.0 sheep/ha (OG) (Fig. 1). The wethers were allowed to graze without interruption in the plots during the duration of the experiment (June 5 to September 3, 90 days). At the start of the research study, sheep were treated with an anthelmintic to eliminate the infection of internal parasites  and given water and mineral lick stones ad libitum.
To identify the chemical constituents of the sheep diet, we collected plants every month of the grazing experiment. The protocol for plant collection was as described elsewhere  and the details are provided as supplementary material. The plants were assessed for different chemical composition indices such as crude protein (CP), gross energy, nitrogen free extract (NFE), neutral detergent fiber (NDF), acid detergent fiber (ADF), as well as acid detergent lignin (ADL), as described elsewhere (a single pooled sample per each plot, n=3) . The LW of the animals was identified on the study’s first and the last day (after 12 h of fasting) and used it to calculate the mean daily gain. Blood was then collected from four sheep selected at random from each plot in the groups (n=12 for each group) via venipuncture of the jugular vein and into vacuum tubes. The blood was then centrifuged twice at 2,000 g at 4°C for 30 min and then at 400 g at 4°C for 10 min to isolate the sera, which were then kept at -80°C to await further analysis.
Biochemical, immune response and inflammation analyses
For biochemical, immune response and inflammatory variables of the serum, the concentration of alanine aminotransferase (ALT), aspartate aminotransferase (AST), alkaline phosphatase (ALP), albumin (ALB), total bilirubin (TBIL), blood urea nitrogen (BUN), glucose (GLU), and nonesterified fatty acid (NEFA) were measured using a fully automatic biochemistry analyzer (Hitachi 7020, Tokyo, Japan); interleukin 8 (IL-8) and immunoglobulin G (IgG) were determined using a corresponding diagnostic kit (Nanjing Jiancheng Bioengineering Institute, Nanjing, China) according to the instructions of the manufacturer.
Each sample was slowly thawed at 4°C. One volume of plasma (50 μL) was mixed with three volumes of cold mixtures of methanol/acetonitrile/H2O (2:2:1, v/v/v) to precipitate proteins following adequate vortexing, incubation (-20°C for 1 h) and centrifugation (15 min at 13,000 rpm and 4°C). The supernatant was collected, dried in a vacuum, redissolved in 50 μL of acetonitrile/water (1:1, v/v), and centrifuged at 14,000 rpm for 15 min, 4°C. A pooled quality control (QC) sample was created by taking the same amount of each sample and mixing the amounts together. We employed QC samples to evaluate the system’s stability and performance before sample loading and during the whole experimental process (every five samples).
We conducted sample analyses using an ultra-high-performance liquid chromatography (UHPLC) system (Agilent 1290 Infinity LC system, USA) coupled with a premium quadrupole time-of-flight mass spectrometer (QTOF MS) system (AB Sciex TripleTOF 6600, MA, USA). Hydrophilic interaction liquid chromatography (HILIC) separation of metabolite samples was performed on a reversed-phase C18 ACQUITY UPLC BEH column (2.1 mm × 100 mm, 1.7 μm; Waters, Ireland). In both the electrospray ionization (ESI) positive and negative modes, the flow rate was 400 μL/min, with solvent A composed of 25 mM ammonium acetate + 25 mM ammonium hydroxide and solvent B composed of acetonitrile. The gradient consisted of 85% B for 1 min, a linear reduction to 65% B in 11 min, a reduction to 40% B in 0.1 min, holding this concentration for 4 min, and then an increase to 85% B in 0.1 min, with a 5 min re-equilibration period.
MS was operated with an AB Sciex TripleTOF mass spectrometer equipped with heated ESI positive and negative modes (AB Sciex TripleTOF 6600). The conditions of the ESI source were the following: ion source gas 1 (gas 1), 60; ion source gas 2 (gas 2), 60; curtain gas (CUR) , 30; source temperature, 600°C, and ion spray voltage floating (ISVF), ± 5,500 V. For MS-only acquisition, the instrument used an m/z range of 60-1,000 Da, with an accumulation time for the TOF MS scan was set at 0.20 s/spectra. For auto MS/MS acquisition, the instrument was set to within an m/z range of 25-1,000 Da, with an accumulation time of 0.05 s/spectra. We conducted the product ion scan with information-dependent acquisition (IDA) at a mode of high sensitivity. Collision energy (CE) was set at 35 V and ± 15 eV. The declustering potential (DP) was at ± 60 V.
Data processing and analysis
The raw MS data using an instrument-specific format (.wiff) were converted to the common data format (MzXML) with the ProteoWizard tool msconvert (version 3.0.10051) and then processed by the program XCMS for feature assessment, retention time correction, as well as alignment. Metabolite identification was performed based on the accurate mass (< 25 ppm) and MS/MS data that were matched to our standard database. In ion features that were extracted, the subsequent analysis included only the variables showing > 50% of the nonzero measurement values in one or more groups.
After normalization that used total peak intensity, we uploaded the processed data to SIMCA-P (version 14.1, Umetrics, Umea, Sweden) for multivariate data analysis, including Pareto-scaled principal component analysis (PCA) as well as orthogonal partial least-squares discriminant analysis (OPLS-DA). Seven-fold cross-validation as well as response permutation testing were employed to assess model robustness. The variable importance in the projection (VIP) value of every variable in the OPLS-DA model was assessed to determine its the contribution to classification. Furthermore, metabolites with a VIP value > 1 were assessed using the Student’s t-test at the univariate level to determine their significance [the normality test of data was conducted using SPSS Statistics 17.0 (SPSS, Inc., USA)], and a P value < 0.05 was deemed as statistically significant.