All experimental protocols were approved by the Animal Care Committee of Xinjiang Agricultural University (No. 2020024).

Twelve Kazakh rams with rumen fistula [location: Changji, China; age: 8 months old; average initial body weight (BW): 36.09 ± 1.81 kg] were selected for this study. The sheep were randomly allocated to either the CON group (n = 6) or NA group (n = 6). Sheep in the CON group were fed a basal diet (concentrate supplement to forage ratio 70:30), and sheep in the NA group were fed a basal diet that was supplemented with 130 mg/day niacin (Shanghai Yuanye Biotechnology Co., Ltd., Shanghai, China). The experiment lasted for 35 days: days 1–14 were the adaptation period, days 15–35 were the test period. Basal diets were formulated according to the Meat Sheep Feeding Standards (NY/T816-2021) (Table 1).

During the experiment, dry matter intake (DMI) and BW (determined at 07:30 am, before the morning feeding) were recorded. All samples were oven dried at 65°C for 48 h to a constant weight. The dried samples were ground and passed through a 1-mm sieve for subsequent analysis.

On days 15 and 35 of the experiment period, all trial sheep were weighed before the morning feed (07:30 am). Ruminal fluid samples were collected from all trial sheep on days 34 to 35 of the experiment. All ruminal fluid samples were collected 0 h before morning feeding and 1, 3, 5, and 7 h after feeding from six sheep (three sheep in the CON group and three sheep in the NA group) every day.

To avoid salivary contamination, 80 mL of ruminal fluid was collected through the ruminal fistula and filtered through four layers of gauze. The first 30 mL of the ruminal fluid was discarded, leaving 50 mL. Ruminal pH was immediately measured using a portable pH meter (Anscitech Co., Ltd., Wuhan, Hubei, China), and ruminal fluid pH was measured 0 h before morning feeding and 1, 3, 5, and 7 h after feeding. Fifteen milliliters of ruminal fluid were squeezed through four layers of cheesecloth, transferred into sterile tubes, and stored at −20°C for subsequent analysis. A total of 12 sheep ruminal fluid samples were collected in this study.

Each sample was analyzed for dry matter (DM; method No. 930.15) by drying at 105°C to a constant weight, ether extract (EE; method No. 989.05) using Soxhlet extraction with diethyl ether and crude protein (CP; method No. 976.05) using the Kjeldahl method with a Buchi Distillation Unit B-324 (Büchi Labortechnik AG, Flawil, Switzerland), as described by the Association of Official Analytical Chemists (13). Cellulose, hemicellulose, and lignin contents were determined according to the Van Soest method (14) using fiber bags and an ANKOM 220 Fiber Analyzer (ANKOM Technology Company, Macedonia, NY, United States).

Frozen ruminal fluid samples were thawed and centrifuged at 12,000 rpm for 10 min at 4°C to obtain a clear supernatant. Then, 1 mL of the supernatant was mixed with 0.25 mL of a metaphosphoric acid standard solution (25 g/100 mL) for subsequent analysis. The concentration of ammonia-N (NH₃-N) was determined using the phenol-sodium hypochlorite colorimetric method described by Broderick and Kang using a UV spectrophotometer (UV-1801, Beijing Beifen Ruili Analytical Instrument Co., Ltd., Beijing, China) (15). The VFA concentrations were analyzed using gas chromatography according to the method described by Mirzaei-Alamouti et al. (16). The VFA was quantified using a high-performance gas chromatograph (GC-2014; Shimadzu Corporation, Kyoto, Japan) that was equipped with a hydrogen flame detector and capillary column (Agilent Technologies, Santa Clara, CA, United States; 30 m long, 0.32-mm diameter, 0.50-μm film thickness). The remaining liquid samples were then immediately frozen in liquid nitrogen, transported to the laboratory, and stored at −80°C for further analysis of microbiota and metabolites. The intermediate products of lactic acid metabolism and lactate metabolic enzyme activity in the rumen fluid were determined based on samples taken at 0 h before feeding and at 1, 3, and 5 h after feeding. Ruminal lactate metabolism intermediates were determined using liquid chromatography (LC-20ADXR; Shimadzu Corporation, Kyoto, Japan), and related enzymes were determined using kits (Nanjing Jiancheng Bioengineering Institute, Nanjing, China).

DNA was extracted from ruminal fluid samples using a DNA extraction kit, and the V3–V4 region of the 16S rRNA gene in the conserved region was amplified. Samples with a bright main strip were chosen. The amplified length was approximately 419 bp (Biomarker Technologies, Rohnert Park, CA, United States). Samples were then sequenced on an Illumina NovaSeq 6000 (Illumina, San Diego, CA, United States) and subjected to quality inspection; high-quality raw data were trimmed. The decontamination-optimization process yielded effective data. The 97% similarity level was clustered using USEARCH software to obtain operational taxonomic units (OTUs) (17). Sequences were diluted 50,000-fold and normalized. Alpha diversity analysis was performed using the Mothur software¹ to calculate species richness and diversity (18) using the Chao1, Ace, Shannon, and Simpson indices and distance matrices of samples obtained using weighted and unweighted algorithms. Principal component analysis (PCA), principal coordinate analysis, and significant species difference analysis were performed using QIIME software with an unweighted UniFrac distance matrix for the principal coordinate analysis.

The ruminal fluid samples were thawed, further ground, and extracted, and metabolites were extracted into the extraction solution (mobile phase methanol-acetonitrile, volume ratio of 1:1), sonicated for 10 min, and then kept at −20°C under static conditions. After 1 h, the samples were centrifuged at 12,000 rpm for 15 min at 4°C. Subsequently, 500 μL of the samples was placed into an Eppendorf tube and dried in a vacuum concentrator for 10 min, and 160 μL of extraction solution (acetonitrile-water volume ratio 1:1) was added to the dried metabolite to reconstitute it. It was vortexed for 30 s, placed in an ice-water bath and sonicated for 10 min, and then centrifuged at 4°C at 12,000 rpm for 15 min. Next, 120 μL of supernatant was carefully removed and placed into a 2-mL injection vial, and 10 μL of each sample was mixed into quality control samples for onboard detection (19). The liquid chromatography/mass spectrometry (LC-MS) system for metabolomic analysis consisted of a Waters Acquity-Class Plus ultra-high-performance liquid chromatograph (Waters, Milford, MA, United States) in series with a Waters Xevo G2-XS QTOF high-resolution mass spectrometer (Waters) and Waters Acquity-UPLC HSS T3 column (1.8 μm 2.1*100 mm) (20). The positive ion mode comprises a mobile phase A containing 0.1% formic acid aqueous solution and a mobile phase B containing 0.1% formic acid acetonitrile. The negative ion mode comprises a mobile phase A containing 0.1% formic acid aqueous solution and a mobile phase B containing 0.1% formic acid acetonitrile. The injection volume was 1 μL.

The Waters Xevo G2-XS QOF high-resolution mass spectrometer (Waters) can perform primary and secondary mass spectrometry data acquisition in the MSe mode under the control of an acquisition software (Mass Lynx V4.2; Waters). Dual-channel data acquisition for low collision energies can be performed simultaneously in each data acquisition cycle. The low collision energy was 2 V, the high collision energy range was 10–40 V, and the scanning frequency was 0.2 s for a mass spectrum. The ESI ion parameters were as follows (21): capillary voltage, 2,500 V (positive ion mode) or −2,000 V (negative ion mode); cone voltage, 30; Vion source temperature, 100°C; desolvation gas temperature, 500°C; flow rate of backflushing gas, 50 L/h; flow rate of desolventizing gas, 800 L/h; and collection range of mass-to-nucleus ratio, 50–120.

Analysis of variance was performed using SPSS software (version 27.0; SPSS Institute Inc., Cary, NC, United States). For initial body weight, final body weight, average daily gain, DMI, pH, NH₃-N, and VFA, t-tests were used to analyze the differences between the microbiome groups, and Spearman’s correlation was performed for the top 20 flora. The correlation between rumen microorganisms and rumen metabolites [variable importance in projection (VIP) >1, fold change (FC) >1 or (FC) <1, and p < 0.05] was evaluated; the data shown represent the mean and standard error of the mean. Statistical significance was set at p < 0.05 or p < 0.01. The R programming language was used to draw principal coordinate analysis (PCoA), Venn, non-metric multidimensional scaling (NMDS), and correlation analysis diagrams. The α-diversity index was plotted using GraphPad Prism 10.0.

As shown in Table 2, there were no significant differences in the initial BWs, final BW, and average daily gain between the CON and NA groups (p > 0.05), whereas the DMI (p < 0.05) in the NA group increased significantly when compared to that of CON group.

Compared to the CON group, ruminal fluid pH in the NA group increased significantly at 3, 5, and 7 h after feeding (p < 0.01) (Figure 1). As shown in Figure 2, the NH₃-N (p < 0.01), propionate (p < 0.01), and butyrate (p < 0.05) concentrations were significantly higher in the NA than those in the CON group. No significant differences were observed in acetate and total VFA concentrations in the rumen between the two groups (p > 0.05).

As shown in Figure 3, all indicators such as pyruvate content (Figure 3A), pyruvate kinase activity (Figure 3B), and lactate content (Figure 3C) in the CON group and the NA group showed a trend of first increasing and then decreasing as the time after feeding prolonged. Compared with the CON group, the pyruvate content (Figure 3A) of the NA group was significantly increased 0 h before feeding (p < 0.05); the lactic acid content (Figure 3C) of the NA group was significantly reduced 1 h and 3 h after feeding (p < 0.05); the activity of lactate dehydrogenase (Figure 3D) in the NA group was significantly reduced at 3 h and 5 h after feeding (p < 0.01); the NADH content (Figure 3F) in the NA group was significantly reduced (p < 0.01) at 3 h after feeding. There was no significant difference in other indicators between the two groups (p > 0.05). Other indicators between the two groups [pyruvate kinase (Figure 3B), NAD⁺ content (Figure 3E), succinic acid content (Figure 3G), succinate dehydrogenase activity (Figure 3H), malic acid content (Figure 3I) and malic acid dehydrogenase activity (Figure 3J)] was not significantly different at each time point (0 h before feeding, 1, 3, and 5 h after feeding) (p > 0.05).

In total, 959,870 raw reads were obtained from 12 samples. There were 930,233 effective reads, and the average length of the sequences was 419 bp. There were 26 and 37 OTUs unique to the CON and NA groups, respectively, and 1,137 OTUs common to both groups (Figure 4A). PCoA showed good separation of samples in each group (Figure 4B), with R² and p-values of 0.712 and 0.007, respectively, and NMDS analysis further demonstrated significant differences between groups (Figure 4C). The Chao1, Ace, Shannon, and Simpson indices of the two groups were not significantly different (p > 0.05) (Figure 5).

The taxonomic distribution of rumen microbiota at the phylum and genus levels is shown in Figure 6. Within the bacterial population, 18 phyla and 26 genera were identified in ruminal samples. Bacteroidetes was the predominant phylum, with a relative abundance of 43.70%, followed by Firmicutes (36.25%) and Proteobacteria (12.77%) (Figure 6A). At the genus level (Figure 6B), Prevotella (21.47%) was the predominant genera, followed by Succinivibrionaceae-UCG-002 (6.88%), and Quinella (5.95%).

A PCA plot was used to visualize trends and outliers. The two groups were not separated when viewed using the first two principal components. To further examine metabolic changes, orthogonal projection to latent structure-discriminant analysis (OPLS-DA) was performed. The results showed that the R²Y and Q²Y of the ruminal fluids (positive and negative) were 0.705, 0.857, 0.695, and 0.28, respectively. The two groups were separated based on OPLS-DA score plots (Figures 7A,C). Additionally, the models showed good predictability, with no overfitting (Figures 7B,D).

According to the statistical analysis and VIP values obtained by OPLS-DA, 831 metabolites were identified, and 36 specific differential metabolites were screened according to the principle: p < 0.05, (VIP) >1, (FC) value >1, or (FC) <1. Twenty-one of these metabolites were found to be at higher levels in the NA group than those in the CON group (p < 0.05), while 15 differential metabolites in the CON group were at higher levels than those in the NA group (p < 0.05). The compounds, p-value, VIP values, and FC values of each metabolite are shown in Table 3. The main metabolites differentially expressed between the CON and NA groups were amino acids, organic acids, sugars, and lipids. The levels of amino acids (such as L-histidine, norleucine, methylphenylalanine, aspartic acid, and phenylalanine), organic acids (such as glutaric acid, succinic acid, propionic acid, and L-malic acid), glucose and galactose (such as α-D-glucosamine-1-phosphate and 2-aminogalactopyranose), and lipids (o-cymene) in the NA group were higher than those in the CON group (p < 0.05). The levels of other amino acids (tetrahydropteroyl tri-L-glutamate), organic acids (maleic acid), glucose, galactose (chitobiose), and lipids (coumaric acid) in the NA group were lower than those in the CON group (p < 0.05).

According to the KEGG metabolic pathway display (Figure 8), there were 20 metabolic pathways for the positive ionization mode and 32 metabolic pathways for the negative ionization mode. This included pathways for the citric acid cycle, pyruvate metabolism, cysteine and methionine metabolism, glutathione metabolism, and arginine and proline metabolism.

Spearman’s correlation coefficient was used to calculate the correlation coefficient between each index (Figure 9). Correlation analysis between rumen fermentation parameters and rumen metabolites showed that NH₃-N was positively correlated with Trp-Pro (p = 0.046, r = 0.594) and negatively correlated with glutathione (p = 0.415, r = −0.720), maleic acid (p = 0.409, r = −0.629). Propionate was positively correlated with Trp-Pro (p = 0.032, r = 0.350), pyruvic acid (p = 0.013, r = 0.706). pH was positively correlated with Trp-Pro (p = 0.005, r = 0.751), hexanoic acid (p = 0.005, r = 0.754), Gln-Glu-His (p = 0.010, r = 0.709), allysine (p = 0.005, r = 0.730), Pro-Phe-Val (p = 0.009, r = 0.716), and sebacic acid (p = 0.007, r = 0.726) and negatively correlated with Tyr-Leu-Ile (p = 0.010, r = −0.709), Glu-Val-Ile-Arg (p = 0.018, r = −0.667), glutathione (p = 0.023, r = −0.646), maleic acid (p = 0.042, r = −0.593). Butyrate was negatively correlated with Tyr-Leu-Ile (p = 0.004, r = −0.783), Glu-Val-Ile Arg (p < 0.001, r = −0.853), maleic acid (p = 0.035, r = −0.622). The total volatile fatty acids was positively correlated with Pro-Arg-Asp (p = 0.028, r = 0.629) and negatively correlated with allysine (p = 0.046, r = −0.594) (Figure 9B).

Correlation analysis between rumen microorganisms and rumen metabolites showed that Christensenellaceae_R_7_group was negatively correlated with sebacic acid (p = 0.017, r = −0.685), allysine (p = 0.026, r = −0.650), Pro-Phe-Val (p = 0.035, r = −0.622), Trp-Pro (p = 0.015, r = −0.699), Gln-Glu-His (p = 0.017, r = −0.825) and hexanoic acid (p = 0.016, r = −0.692). NAD was positively correlated with Quinella (p = 0.007, r = 0.755) and Fretibacterium (p = 0.046, r = 0.594). Ala-Ile-Val-Phe was negatively correlated with Prevotella (p = 0.017, r = −0.685), and Succinivibrio (p = 0.044, r = −0.589) (Figure 9C).