Dietary Supplementation of Inulin Ameliorates Subclinical Mastitis via Regulation of Rumen Microbial Community and Metabolites in Dairy Cows

ABSTRACT Subclinical mastitis (SCM) is one of the highly infectious diseases in dairy cows with the characteristics of high incidence and nonvisible clinical symptoms. The gastrointestinal microbiota is closely related to mastitis. Inulin is a prebiotic fiber with functions in improving intestinal microbial communities and enhancing the host’s immunity. However, the impact of dietary inulin on the rumen inner environment remains unknown. The current study investigated whether inulin could relieve SCM by affecting the profiles of ruminal bacterial and metabolites in dairy cows. Inulin inclusion rates were 0, 100, 200, 300, and 400 g/day per cow, respectively. Inulin increased milk yield, milk protein, and lactose and reduced the somatic cell counts (SCC) in milk. In serum, the concentration of proinflammatory cytokines, such as interleukin-6 (IL-6), IL-8, tumor necrosis factor α (TNF-α), and malondialdehyde (MDA) were decreased, and IL-4 and superoxide dismutase (SOD) were increased. Meanwhile, inulin increased the concentration of propionate, butyrate, and lactic acid (LA), while it decreased NH3-N in rumen. The propionate- and butyrate-producing bacteria (e.g., Prevotella and Butyrivibrio) and several beneficial commensal bacteria (e.g., Muribaculaceae and Bifidobacterium) as well as metabolites related to energy and amino acid metabolism (e.g., melibiose and l-glutamate) were increased. However, several proinflammatory bacteria (e.g., Clostridia UCG-014, Streptococcus, and Escherichia-Shigella) were decreased, accompanied by the downregulation of lipid proinflammatory metabolites, for example, ceramide(d18:0/15:0) [Cer(d18:0/15:0)] and 17-phenyl-18,19,20-trinor-prostaglandin E2. In the current study, the above indicators showed the best response in the 300 g/day inulin group. Overall, dietary supplementation of inulin could alleviate inflammatory responses in cows with SCM through improving the rumen inner environment. IMPORTANCE The correlation between mastitis and the gastrointestinal microbiome in dairy cows has been demonstrated. Regulating the profile of rumen microorganisms may contribute to remission of subclinical mastitis (SCM). Supplementation of inulin in the diets of cows with SCM could increase the abundance of short-chain fatty acid (SCFA)-producing bacteria and beneficial commensal bacteria in rumen and meanwhile the levels of amino acids and energy metabolism. Conversely, the abundance of ruminal bacteria and metabolites with proinflammatory effects were decreased. Our study suggests that the improvement of the rumen internal environment by inulin supplementation could ameliorate inflammatory responses during SCM in dairy cows and thus improve lactation performance and milk quality. Our results provide a theoretical basis for regulation measures of SCM in dairy cows.

IMPORTANCE The correlation between mastitis and the gastrointestinal microbiome in dairy cows has been demonstrated. Regulating the profile of rumen microorganisms may contribute to remission of subclinical mastitis (SCM). Supplementation of inulin in the diets of cows with SCM could increase the abundance of short-chain fatty acid (SCFA)-producing bacteria and beneficial commensal bacteria in rumen and meanwhile the levels of amino acids and energy metabolism. Conversely, the abundance of ruminal bacteria and metabolites with proinflammatory effects were decreased. Our study suggests that the improvement of the rumen internal environment by inulin supplementation could ameliorate inflammatory responses during SCM in dairy cows and thus improve lactation performance and milk quality. Our results provide a theoretical basis for regulation measures of SCM in dairy cows.
ECM and FCM had no significant difference between the I-2 and I-3 groups. However, milk electrical conductivity (EC) (P = 0.083) presented a downward trend with the addition of inulin, which was lowest in the I-3 group. Meanwhile, milk urea nitrogen (MUN) (P = 0.046), blood urea nitrogen (BUN) (P = 0.035), and milk somatic cell count (SCC) (P , 0.01) values were significantly decreased and lowest in the I-3 group.
Inflammatory response and lipopolysaccharide concentration. As shown in Table 2, inulin supplementation elevated the concentrations of interleukin-4 (IL-4) (P = 0.042) and superoxide dismutase (SOD) (P = 0.041) and decreased the levels of IgG (P = 0.013), IL-6 (P = 0.035), IL-8 (P = 0.034), tumor necrosis factor-a (TNF-a) (P = 0.042), prostaglandin E 2 (PGE 2 ) (P = 0.036), and malondialdehyde (MDA) (P = 0.022) in serum. Among them, the highest value of SOD, as well as the lowest values of IgG, TNF-a, and MDA, appeared in the I-2 and I-3 groups. The lowest value of PGE 2 was found in the I-3 and I-4 groups. Meanwhile, the greatest concentration of IL-4 and least concentrations of IL-6 and IL-8 were found in the I-3 group. Compared with the control group, the lipopolysaccharide (LPS) concentration in rumen (P = 0.046) was reduced and the concentration in serum (P = 0.054) demonstrated a downtrend (P = 0.054) in the I-3 group (Fig. 1).
Effects of inulin addition on rumen fermentation parameters. As shown in Table 3, the concentrations of propionate (P = 0.021), butyrate (P = 0.034), total volatile fatty acids (TVFA) (P = 0.041), and lactic acid (LA) (P = 0.037) in the rumen of SCM cows were increased after inulin addition and were highest in the I-3 group (the concentration of LA demonstrated no significant difference between the I-2 and I-3 groups). The concentration of isobutyrate (P = 0.084) had a tendency to increase, while the concentration of NH 3 -N (P , 0.01) and the ratio of acetate to propionate (A/P) (P = 0.041) were reduced and lowest in the I-3 group (A/P demonstrated no significant difference between the I-2 and I-3 groups). Additionally, the pH value in rumen (P = 0.083) and the concentration of rumen urea nitrogen (RUN) (P = 0.064) both demonstrated a decreasing trend with inulin supplementation.
Effect of inulin on diversity, richness, taxonomic annotation, and composition of the ruminal microflora community. A total of 1,303,222 high-quality 16S rRNA gene sequences were obtained from 30 rumen fluid samples with Good's coverage of 99% in all samples. Meanwhile, the rarefaction curves revealed that the increment of a DMI, dry matter intake; ECM, energy-corrected milk; FCM, fat-corrected milk; F/P, fat-to-protein ratio; MUN, milk urea nitrogen; BUN, blood urea nitrogen; SCC, somatic cell counts; EC, electrical conductivity of milk; FCR, feed conversion ratio; SEM, standard error of mean; Con, control group; I-1, inulin-1 group with an inulin addition level of 100 g/day per cow; I-2, inulin-2 group with an inulin addition level of 200 g/day per cow; I-3, inulin-3 group with an inulin addition level of 300 g/day per cow; I-4, inulin-4 group with an inulin addition level of 400 g/day per cow; a, b, c , within a row, different letters differed significantly (P , 0.05 the number of species and diversity index gradually tended to be flat with an increase in the number of reads sampled (Fig. S1 in the supplemental material), which implied that the amount of sequencing data was sufficient to reflect most of the microbial diversity information in the samples. As shown in Table S2 and Fig. 2, compared with the control group, the Sobs (P = 0.041) and Chao (P = 0.035) indexes, which reflect the microbial richness, as well as Shannon (P = 0.037) and Simpson (P = 0.021) indexes, which indicate bacterial diversity, were significantly increased in SCM cows fed inulin and peaked in the I-3 group. After sequence clustering and quality filtering, a total of 2,151 operational taxonomic units (OTUs) with .97% similarity were identified. Using the RDP classifier Bayesian algorithm to perform taxonomic analysis on 97% similar OTU representative sequences, 18 phyla and 266 genera of bacteria were obtained (Table S3 and S4). At the phylum level, Bacteroidota (51.5 6 1.65%), Firmicutes (36.8 6 1.47%), Proteobacteria (6.85 6 0.94%), Patescibacteria (2.16 6 0.10%), and Actinobacteriota (1.49 6 0.07%) were predominant across the 5 groups (Fig. 3A). At the genus level, Prevotella (18.6 6 1.30%), Ruminococcaceae  3B). Weighted normalized UniFrac and unweighted UniFrac distance algorithms were used to perform the principal coordinates analysis (PCoA) of rumen microbial communities ( Fig. 4A and B). Difference tests among groups were analyzed via the analysis of similarities (ANOSIM) method, with R acting as a statistic. In Fig. 4, an R of .0 indicated that the intragroup distance was less than the intergroup distance, which meant the grouping was valid. Furthermore, the difference of microbial composition among control and inulin treatment groups was significant (P = 0.001). Hierarchical clustering analysis of the top 50 bacteria in relative abundance was conducted to further investigate the effects of inulin on ruminal microbial profiles (Fig. S2). In the present study, the distribution of ruminal bacteria was divided into two major categories. Similar to the results of the PCoA, the profiles of ruminal bacteria in the control and I-1 groups could not be clearly distinguished, with Streptococcus, Ruminococcus, Neisseriaceae, Coprococcus, Ruminococcaceae, Escherichia-Shigella, Staphylococcus, etc., being the most abundant. However, with the increase of inulin addition levels, the ruminal bacterial profile was significantly changed. Compared with the control group, Prevotella, Muribaculaceae, Christensenellaceae R-7 group, Butyrivibrio, SEM, standard error of the mean; Con, control group; I-1, inulin-1 group with an inulin addition level of 100 g/day per cow; I-2, inulin-2 group with an inulin addition level of 200 g/day per cow; I-3, inulin-3 group with an inulin addition level of 300 g/day per cow; I-4, inulin-4 group with an inulin addition level of 400 g/day per cow; a, b, c , within a row, different letters differed significantly (P , 0.05). Bifidobacterium, Lactobacillus, etc., were the predominant bacteria genera in the I-3 and I-2 groups. Meanwhile, the relative abundances of Treponema, Candidatus Saccharimonas, Succiniclasticum, Prevotellaceae_UCG-001, Lachnospiraceae NK4A136 group, and Eubacterium hallii group in the I-4 group were higher than in the control group. Differentially abundant ruminal bacteria among control and different inulin addition groups. The differentially abundant bacterial communities among control and inulin treatment groups are presented in Table 4. At the phylum level, Bacteroidota (false-discovery rate [FDR]-adjusted P = 0.033), Proteobacteria (FDR-adjusted P = 0.048), and Actinobacteriota (FDR-adjusted P = 0.037) were increased in the inulin supplementation groups compared with the control group and were mostly abundant in the I-3 group (except for Proteobacteria, which demonstrated no significant difference between the I-2 and I-3 groups). The relative abundance of Firmicutes (FDR-adjusted P = 0.037) was decreased with the addition of inulin and was lowest in the I-3 group.  Genus-level classification of rumen microbial communities showed higher abundances of Prevotella (FDR-adjusted P = 0.013), Muribaculaceae (FDR-adjusted P = 0.033), Bacteroidales BS11 gut group (FDR-adjusted P = 0.042), Bacteroidales RF16 group (FDR-adjusted P = 0.045), Butyrivibrio (FDR-adjusted P = 0.031), Lactobacillus (FDR-adjusted P = 0.042), and Bifidobacterium (FDR-adjusted P = 0.043) in inulin groups than in the control group, and most of these bacteria were most abundant in the I-3 group. However, the relative abundances of Ruminococcaceae (FDR-adjusted P = 0.035), Oscillospira (FDR-adjusted P = 0.042), Clostridia_UCG-014 (FDR-adjusted P = 0.035), Coprococcus (FDR-adjusted P = 0.036), Streptococcus (FDR-adjusted P = 0.048), Staphylococcus (FDR-adjusted P = 0.045), Neisseriaceae (FDR-adjusted P = 0.046), and Escherichia-Shigella (FDR-adjusted P = 0.045) were reduced with the supplementation of inulin, and the lowest abundances of the above bacteria were also detected in the I-3 group.
Metabolic profiling of ruminal fluid samples. An untargeted metabolomics analysis based on liquid chromatography-mass spectrometry (LC-MS) technology was used to investigate the metabolite profiles in rumen fluid samples. Total ion chromatograms (TICs) of quality control (QC) samples at positive and negative ion modes are shown in Fig. S3A and B, respectively. Through overlapping comparison, the response intensity and retention time of each chromatographic peak was found to be basically overlapped, which reflected the reliability of the experiment data quality. To reflect the overall difference among different groups and the degree of variability among samples within the group, a principal-component analysis (PCA) was conducted based on an unsupervised multivariate statistical analysis method. As shown in Fig. 6A and B, in both positive and negative ion modes, the different inulin treatment groups and control group could be separated except for the I-1 group, which was closer to the control group than other treatment groups.
To further visualize the differences in metabolites among different groups, partial least squares discriminant analysis (PLS-DA) and orthogonal partial least squares discriminant analysis (OPLS-DA) were performed based on a supervised discriminant analysis method to conduct multigroup variance analysis and pairwise comparison at positive and negative ion modes, respectively (Fig. S4, S5, and S6). In PLS-DA and OPLS-DA score plots, the data at positive and negative ionization analysis were separated distinctly among control and inulin treatment groups. Additionally, R 2 X(cum) and R 2 Y (cum) represent the cumulative interpretation rates to the X and Y matrices of the model, respectively; Q 2 (cum) represents the predictive ability of the model. The closer these three indicators are to 1, the higher the predicting ability of the model would be. The values of R 2 X(cum), R 2 Y(cum), and Q 2 (cum) were all above 0.7. Moreover, permutation testing indicated good stability and reliability of the model with an R 2 Y of .0.7 and a Q 2 Y of ,0.
After filtering and optimizing, a total of 1,637 metabolites (748 and 889 in positive and negative ion modes, respectively) were identified from the 30 rumen fluid samples. Classification information of metabolites was obtained via comparing with the Human Metabolome Database (HMDB) 4.0. As shown in Fig. S7, at the superclass level, metabolites belonging to lipids and lipid-like molecules (28.45%), organic acids and derivatives (19.86%), organoheterocyclic compounds (14.95%), phenylpropanoids and polyketides (14.19%), and organic oxygen compounds (10.43%) were predominant (Fig. S7A). At the subclass level, amino acids, peptides and analogues (16.95%), carbohydrates and carbohydrate conjugates (7.59%), and fatty acids and conjugates (3.99%) were the main components, and 11.81% of metabolites were not identified according to the taxonomic information (Fig. S7B).
Metabolic pathway analysis of differentially abundant metabolites. The KEGG database was used to annotate the metabolic pathways of differentially abundant metabolites involved ( Table 5). The major metabolic pathways that the increased metabolites in inulin groups involved were galactose metabolism, amino acid metabolism and biosynthesis, protein digestion and absorption, the glycolytic pathway, etc. However, sphingolipid metabolism, arachidonic acid metabolism, choline metabolism, glycerophospholipid metabolism, amino acid/peptide hydrolysis, etc., were the main metabolic pathways that the high-abundance metabolites in the control group were involved in.
Correlation analysis between significantly differentially abundant metabolites and ruminal bacteria. Based on Spearman's correlation coefficients, the correlations between significantly differentially abundant metabolites and bacteria in rumen are shown in Fig. 8, and correlation coefficients (r) and FDR-adjusted P values are listed in Table S10. Melibiose was positively associated with Bifidobacterium (r = 0.610, FDRadjusted P = 0.014) and Lactobacillus (r = 0.581, FDR-adjusted P = 0.035) but negatively associated with Streptococcus (r = 20.524, FDR-adjusted P = 0.036) and Escherichia-Shigella (r = 20.594, FDR-adjusted P = 0.025). L-Glutamate was positively correlated Each column represents a sample, each row represents a metabolite, each color represents the relative expression of metabolites in the sample, and the corresponding relationship between the color gradient and the value is shown in the gradient color block. VIP bar graphs of metabolites are on the right. The length of the bar represents the contribution value of the metabolite to the difference between the two groups. The larger the value, the greater the difference between the two groups. The color of the bar indicates the significance of the metabolite difference between the two groups, that is, the FDRadjusted P value (*, 0.01 , FDR-adjusted P value # 0.05; **, 0.001 , FDR-adjusted P value # 0.01; ***, FDR-adjusted P value # 0.001). Con, control group; I-1, inulin-1 group with an inulin addition level of 100 g/day per cow; I-2, inulin-2 group with an inulin addition level of 200 g/day per cow; I-3, inulin-3 group with an inulin addition level of 300 g/day per cow; I-4, inulin-4 group with an inulin addition level of 400 g/day per cow.

DISCUSSION
BM is not merely a local inflammatory response that occurs in the udder (7-10, 15). It can a strong immune response throughout the body and changes physiology and behavior, including impaired lactation function, slowed rumen motility, less feeding time, and reduced rumination frequency (16). Currently, little information is available on the effect of inulin on lactation performance in dairy cows. The large amount of volatile fatty acids (VFAs) produced by rumen microbial fermentation of inulin might provide energy substrates for lactation (16,17). Decreased milk lactose secretion was considered to be one of the indicators of mastitis (18). Increased permeability of the blood-milk barrier caused by inflammation promotes certain ions entering the milk from the blood, which might cause an osmotic gradient that is not conducive to the secretion of milk lactose (18). However, as the precursor of lactose synthesis, the increased concentration of propionate produced by inulin fermentation in the rumen might promote the synthesis of milk lactose (19). Additionally, inulin increased the diversity and richness of rumen microbes in this study; thus, the elevated synthesis of microbial protein in rumen may contribute to the synthesis of milk protein in the mammary gland (20). Further evidence for the increase of rumen microbial protein content after inulin supplementation was the decrease of rumen NH 3 -N concentration. Consistent with previous studies, inulin can be used as a substrate by hemicellulose- FIG 8 Correlation analysis between significantly differentially abundant ruminal microorganisms and metabolites among control and inulin-treated groups. Each row in the graph represents a genus, each column represents a metabolite, and each lattice represents a Spearman correlation coefficient between a bacterium and a metabolite. Red represents a positive correlation, while blue represents a negative correlation.
degrading bacteria in rumen to support their growth and nitrogen utilization, so reduced NH 3 -N concentrations have been observed in rumen with inulin supplementation (21)(22)(23). Moreover, Gressley et al. (23) found that the concentrations of MUN and BUN in lactating dairy cows were decreased after inulin supplementation, which also reflected the reduction of rumen NH 3 -N. The concentrations of RUN, MUN, and BUN could reflect the utilization of protein in the rumen (23). However, RUN only presented a downward trend after inulin treatment in our experiment. During mastitis, the transport of LPS from the rumen to blood could increase the release of proinflammatory cytokines and elevate the content of immunoglobulin in the blood (24,25). In the present study, inulin treatment reduced the concentration of LPS in rumen and blood. It's reported that inulin supplementation could decrease gut permeability and reduce LPS translocation to ameliorate systematic inflammation (26), but direct evidence of gut permeability and integrity still needs further research. Additionally, decreased levels of proinflammatory factors (PGE 2 , IL-6, IL-8, and TNF-a) and increased levels of the anti-inflammatory factor IL-4 might indicate that inulin supplementation could alleviate inflammation via reducing proinflammatory cytokines (26). Meanwhile, it might also explain the reduced levels of milk SCC in the inulin groups.
For ruminants, the rumen is the dominant organ housing the largest number of microorganisms, which directly affects the digestion and metabolism of nutrients and immune function (27). Previous studies reported that rumen microbial diversity and concentration of SCFAs in cows with SCM were significantly reduced, which might be due to the changed structure in the rumen microbial community (8,15). Consistent with previous studies, inulin supplementation significantly increased the concentration of TVFAs in the rumen of dairy cows with SCM, especially butyrate, propionate, and LA. The concentration of acetate did not change significantly, and A/P decreased significantly (27,28). Acetate is regarded as the precursor of milk fat synthesis (29); thus, milk fat was also not observed to be significantly changed after inulin treatment in the current study.
Changes in concentration of SCFAs were closely related to the inulin effect on the ruminal bacteria profile. In the present study, supplementation of inulin significantly increased diversity and richness of the rumen microbiome. Inulin could be converted to microbial glycogen by ruminal bacteria and protozoa, which would affect VFAs and microbial production (30). At the phylum level, inulin addition increased the relative abundance of Bacteroidota and Proteobacteria yet decreased Firmicutes and Actinobacteriota. Bacteroidota, such as Prevotella, Bacteroidales BS11 gut group, Bacteroidales RF16 group, and Muribaculaceae, produce mainly propionate by fermenting dietary fiber (31)(32)(33). Muribaculaceae was the member of the beneficial commensal bacteria in the rumen and intestine, and participated in the formation of the mucus layer, maintaining of barrier function and being related to the innate immune system (34). Zhong et al. (15) found that compared with high-SCC dairy cows, Proteobacteria were enriched in the rumen of cows with low SCC, among which Succinivibrionaceae were the main cause of the difference. This was consistent with our findings. Succinivibrionaceae are also succinate-producing bacteria (35). Thus, the above increased propionate-producing bacteria in inulin groups might explain the positive correlation with milk lactose (19). Importantly, several studies suggested that propionate plays an important role in hosts' susceptibility to bacterial and fungal infections (9,36) and could prevent LPS-induced mastitis by regulating the blood-milk barrier (11).
Firmicutes, such as Butyrivibrio, Lachnospiraceae NK3A20 group, Lactobacillus, and Eubacterium ruminantium group, could effectively ferment inulin to produce butyrate and LA (31,37). In addition, inulin could promote the proliferation of Bifidobacterium in the gastrointestinal tract, which may inhibit the secretion of IL-8 and stimulate the production of IL-10 (38). Although Bifidobacterium could not directly produce butyrate through fermentation of inulin, it could stimulate other butyrate-producing bacteria, such as the Eubacterium ruminantium group (39). Butyrate was the crucial nutrient inducing the expression of tight junction proteins and could affect the inflammatory response by regulating the release of inflammatory cytokines (40). Meanwhile, LA produced by Lactobacillus through fermentation of inulin could prevent the invasion and colonization of pathogens in the intestinal tract (41).
As the common pathogens in milk of cows with mastitis, Streptococcus, Staphylococcus, Escherichia-Shigella, and the conditional pathogen Clostridia have also been identified in the rumen of cows with SCM in this study. Recently, the entero-mammary endogenous pathway of microorganisms has been increasingly reported (10). Gastrointestinal microorganisms have been demonstrated to affect inflammation in distal intestinal tissues, including the udder (4)(5)(6). Studies reported that changes in fecal microbial communities of cows with mastitis were similar to those in the milk, with increases in Enterococcus, Streptococcus, and Staphylococcus and decreases in Lactobacillus (7,10). Consistent with Wang et al. (8), the conditional pathogen Neisseriaceae in the intestine was significantly enriched in the rumen of cows with SCM. An important virulence mechanism of Neisseriaceae is the ability to shed outer membrane blebs containing LPS (42). In this study, we observed higher abundance of Ruminococcaceae and Oscillospira in the rumen of cows with SCM. Ma et al. (7) also found that these two bacteria at high levels in the feces of cows with mastitis. The upregulated Ruminococcaceae and Oscillospira in the rumen of cows with mastitis might be related to infection and inflammation, which still needs to be further explored. In this study, these inflammation-related ruminal bacteria were positively correlated with milk SCC and negatively correlated with milk components, highlighting their adverse effect on lactation performance.
Correlations between the rumen microbial profile and mastitis have been confirmed by several studies (7)(8)(9)15). In the current study, dietary supplementation of inulin mainly increased the abundance of propionate-and butyrate-producing bacteria and beneficial commensal bacteria in the rumen of cows with SCM and reduced the abundance of inflammation-inducing bacteria within the rumen. Consistent with Li et al. (26), our results suggest that the anti-inflammation effect of inulin might closely relate to the modulation of ruminal microbiota.
Variations in the rumen microbial community might further lead to changes in metabolic activities. The increase in D-fructose, melibiose, N-acetyllactosamine, and DHAP (10:0) after inulin addition upregulated a series of metabolic pathways related to energy metabolism, such as carbohydrate digestion and absorption, galactose metabolism, fructose and mannose metabolism, and the glycolytic pathway. Xi et al. (1) detected a significant decrease in carbohydrates and energy-related metabolites in milk at the peak of IMI. Ma et al. (7) found a consumed glycolytic metabolism pathway in the fecal metabolites of cows with mastitis. In addition, the inflammatory response caused by pathogens in the rumen might lead to local energy consumption (1,7). Inulin supplementation elevated the energy levels of cows with SCM, which was essential for maintaining the synthesis of milk components and lactation. Of note, melibiose is also reported to participate in immune responses, which could strongly affect the response of Th cells to ingested antigens (43). Moreover, the increase of L-glutamate, Ltyrosine, and benzoic acid in the inulin groups facilitated glutathione metabolism, tyrosine metabolism, phenylalanine metabolism, protein digestion and absorption, thiamine metabolism. Elevated amino acid metabolism in the rumen of cows with SCM fed inulin might be attributed to the degradation of increased rumen microbial protein.
The thiamine metabolic pathway was also previously observed to be inhibited in the feces and milk of dairy cows with mastitis (7,44). As a cofactor for many biochemical reactions, vitamin B might participate in the inhibition of inflammation (7). Furthermore, in the gastrointestinal tract amino acids could be used by microbiota activated by inulin to synthesize SCFAs (45). For instance, butyrate could be derived from glutamate, which could maintain intestinal barrier function (46). During mastitis, tyrosine and phenylalanine are also decreased in milk (1). L-Tyrosine could be produced by Bifidobacterium and Lactobacillus (1). Benzoic acid is the precursor of hippuric acid, which inhibits the growth of E. coli (1, 47).
On the other side, the downregulated pathways were mainly linked to lipid metabolism, including sphingolipids, arachidonic acid, choline metabolism, and glycerophospholipid metabolism. Decreased lipid metabolites contained ceramide, 17-phenyl-18,19,20-trinor-prostaglandin E 2 , 12-oxo-20-trihydroxy-leukotriene B4, LysoPC, and 9,12,13-TriHOME, which may be involved in the host proinflammatory response. Specifically, ceramide is a sphingolipid inflammatory mediator that participates in LPSmediated mitochondrial destruction to induce apoptosis (48). 17-Phenyl-18,19,20-trinor-prostaglandin E 2 was classified as a member of the prostaglandins and related compounds (49). PGE 2 could accelerate E. coli-induced mastitis in dairy cows during early lactation (49). Moreover, in the present study, a high concentration of PGE 2 was also detected in the serum of cows with SCM after inulin treatment. 12-Oxo-20-trihydroxy-leukotriene B4, LysoPC, and 9,12,13-TriHOME were all capable of aggregating neutrophils. 12-Oxo-20-trihydroxy-leukotriene B4 was the lipid omega-oxidation metabolite of leukotriene B4 (LTB4), which was the main metabolite of neutrophil polymorphonuclear leukocytes. During BM, LTB4 could induce the infiltration of neutrophils in the bovine mammary gland and lead to a surge in SCC (50). Despite the fact that LysoPC and 9,12,13-TriHOME were reported to induce the recruitment of monocytes and the production of proinflammatory cytokines, limited information regarding their association with BM has been reported (51,52).
At present, limited studies have investigated the effect of inulin on these lipid proinflammatory metabolites in the rumen of cows with SCM. We speculated that it might be related to the regulation of lipid metabolism by inulin. Consistent with those studies on humans and other animals, consumption of inulin could cut down the concentration of triglycerides and cholesterol, which is mediated through the reduced activities of hepatic lipogenic enzymes, such as ATP citrate lyase, acetyl coenzyme A (acetyl-CoA) carboxylase, and fatty acid synthase (12,13).
Based on the correlation between BM and the rumen inner environment, modulating the microbial community and the abundance of metabolites in the rumen may contribute to SCM remission to a certain extent. Throughout the study, modulation of inulin in the microbial community and metabolites in rumen affected the level of inflammatory cytokines and LPS as well as milk composition in cows with SCM. Moreover, during the period of SCM, increased milk SCC indicates elevated neutrophils in milk which result in oxidation reactions and the increase of liquid peroxidation products (53). In the present study, the antioxidant activity of inulin might relate to the scavenging of reactive oxygen species and hydroxyl free radicals (54), which need further study. On the other hand, the results of lactation performance, immunity indices, rumen fermentation, and microbial diversity in cows with SCM did not increase linearly with an increase of inulin supplementation in the present study. When the addition level was over 300 g/day per cow, most of the indicators showed a downward trend, which might be due to the metabolic adaptability of dairy cows to inulin. Currently, limited information is available on the appropriate dosages of inulin in the diets of dairy cows. Overall, in the current study, dietary supplementation of 300 g/day of inulin per cow has a better effect in alleviating inflammation reactions in cows with SCM.
In summary, our findings revealed that dietary supplementation of inulin could increase the abundance of propionate-and butyrate-producing bacteria and beneficial commensal bacteria, promote the level of metabolites associated with energy and amino acid metabolism, and lower the abundance of inflammation-related bacteria and levels of lipid proinflammatory metabolites in the rumen of cows with SCM. Thereby, inulin attenuates inflammatory and oxidative stress responses in the serum, increases the concentration of propionate and butyrate in the rumen, and improves the lactation performance of cows with SCM. Overall, the mechanism of inulin in alleviating SCM in dairy cows might be through the improvement of the rumen microbial community and their metabolites.

MATERIALS AND METHODS
Ethics statement. All experimental designs and protocols were approved by the Animal Ethics Committee of the Chinese Academy of Agricultural Sciences (Beijing, China; approval number IAS-2020-92) and were in accordance with the recommendations of the academy's guidelines for animal research.
Animals, feedstuffs, and experiment design. The current study was conducted in an intensively managed dairy farm located in Beijing, China (39°919N, 116°669E). According to milk somatic cell counts (SCC), California mastitis test (CMT), milk electric conductivity (EC), and clinical symptoms in udders (55) and by combining with the basic body condition (milk SCC = 684,000 6 27.7 cells/ml; the result of CMT was weakly positive and positive; milk EC = 7.83 6 1.07; no clinical symptoms in udders; days in milk = 137 6 8.7 days; parity = 3.43 6 0.38; milk yield = 29.7 6 1.96 kg/day), a total of 30 Holstein dairy cows with SCM in mid-lactation were selected and randomly divided into 5 groups (n = 6). Inulin addition levels of 0 (control group), 100 (I-1 group), 200 (I-2 group), 300 (I-3 group), and 400 (I-4 group) g/ day per cow, respectively, were used. The criteria for CMT results are referenced in Wang et al. (8). Inulin used in the current study was supplied by Langfang Academy of Agriculture and Forestry Sciences (Hebei, China). All groups were fed the same total mixed rations (TMR) as basic diet, with a concentrateto-forage ratio of 40:60. The TMR ingredients and nutritional compositions are shown in Table S1 in the supplemental material. Cows were housed in individual tie stalls and offered TMR ad libitum 3 times a day at 07:30, 13:30, and 19:30 with free access to food and water. Due to the hygroscopicity of inulin, it was weighed and rolled into pills with the same shape and size. A feeder (Boehringer-Ingelheim, Biblach, Germany) was used to dose the pills via the mouth 3 times a day at the time of offering TMR. Feed residuals were collected to calculate dry matter intake (DMI). The experiment lasted for 8 weeks, with 2 weeks for adaptation and 6 weeks for the trial.
Milk sampling and analysis. Cows were milked 3 times a day at 06:30, 12:30, and 18:30 through an automatic milking system (Afimilk Ltd., Kibbutz Afikim, Israel). Milk yield data were recorded every day by a dairy herd management software program (Afimilk Ltd., Kibbutz Afikim, Israel). The milk yield reported was the average of each week. Milk samples from each cow were sampled in the morning, at 12:00, and in the evening and mixed with a ratio of 4:3:3 on the last day of the experiment (day 56) (56). Mixed milk samples from each cow were collected into two 50-ml sterile centrifuge tubes and stored at 4°C. Milk composition, including milk fat, milk protein, lactose, and SCC, was determined by near infrared spectroscopy using a milk composition analyzer (PRO40SEC, Daimaker Technology Co., Ltd., Beijing, China). The milk EC was measured by a conductivity meter (MIK-EC8.0, Meikong Automation Technology Co., Ltd., Hangzhou, China). The concentration of milk urea nitrogen (MUN) was analyzed using a MUN test kit (SK081, Bio Lebo Technology Co., Ltd., Beijing, China).
Serum collection and analysis. Blood sampling was conducted 1h before morning feeding on the last day of the experiment (day 56). Two tubes of blood samples (5 ml per tube) from each cow were sampled through the tail vein. The blood samples were placed on ice for 30 min. After coagulating, the blood samples were centrifuged for 15 min at 3,000 Â g at 4°C to separate the serum. The serum sample from each cow was collected and stored at 220°C for the further detection of blood urea nitrogen (BUN), total protein (TP), albumin (ALB), globulin (GLB), inflammatory cytokines, and lipopolysaccharide (LPS). BUN was analyzed using a BUN test kit (Meilian Biotechnology Co., Ltd., Shanghai, China). The concentrations of TP and ALB were analyzed by a TP colorimetric test kit (biuret method) (Elarite Biological Technology Co., Ltd., Wuhan, China) and an ALB assay kit (bromocresol green colorimetry) (Ameijie Technology Co., Ltd., Wuhan, China), respectively. The concentration of GLB was calculated as the difference between the TP and ALB. The concentration of IL-6, IL-8, IL-10, IL-4, IL-2, TNF-a, prostaglandin E 2 (PGE 2 ), glutathione peroxidase (GSH-Px), superoxide dismutase (SOD), and malondialdehyde (MDA) were measured by corresponding bovine enzyme-linked immunosorbent assay (ELISA) kits (Enzyme Link Biotechnology Co. Ltd., Shanghai, China). The concentrations of IgA, IgG, and IgM were determined by immunoturbidimetry with bovine immunoglobulin ELISA kits (Cloud-Clone Technology Co., Ltd., Wuhan, China). The concentration of LPS in serum was detected by a general LPS ELISA kit (ZY-LPS-Ge, Ze Ye Biotechnology Co., Ltd., Shanghai, China). The above measurements were acquired according to the manufacturers' instructions.
Rumen fluid sampling and measurement. On day 56, ruminal fluid samples from each cow were sampled 3 h after morning feeding using a rumen fluid sampler with a 1.5-m hose and a metal filter (GCYQ-1-A, Shanghai SDI Scientific Instrument Co., Ltd., Shanghai, China) and a 200-ml syringe. Ruminal fluid was drawn with the syringe by inserting the rumen sampler through the mouth into the rumen. The first two tubes of ruminal fluid were discarded to prevent saliva contamination. Approximately 150 ml of rumen fluid from each cow was sampled. Four layers of sterile gauze were used to filter the rumen fluid samples. The pH value of the filtrate was measured immediately using a portable pH meter (WLB-TC-PH, Wailaibo Biotechnology Co., Ltd., Beijing, China). The filtrate from each cow was divided into five 10-ml sterile centrifuge tubes to determine the concentrations of volatile fatty acids (VFAs), lactic acid (LA), NH 3 -N, rumen urea nitrogen (RUN), LPS, and ruminal microorganisms and metabolites. The ruminal fluid samples used to analyze the microbiome and metabolites were stored at 280°C, and the others were stored at 220°C. The concentrations of VFAs, including acetate, propionate, butyrate, isobutyrate, valerate, and isovalerate, were analyzed by gas chromatography (Agilent 7890A, Agilent Technologies, CA, USA). The concentration of NH 3 -N was detected via alkaline sodium hypochlorite-phenol spectrophotometry. The concentration of LA was measured by a lactate assay kit (MAK064, Sigma-Aldrich, St. Louis, MO, USA). The concentration of RUN was measured by the diacetyl-anthracene colorimetric method using a urea test kit (CJT77101, Chejeter Technology CO., Ltd., Beijing, China). The concentration of LPS in rumen was detected by a general LPS ELISA kit (ZY-LPS-Ge, Ze Ye Biotechnology Co., Ltd., Shanghai, China) (17). Analysis of the ruminal microbiota and metabolites was performed by 16S rRNA sequencing and untargeted metabolomics based on liquid chromatography-mass spectrometry (LC-MS), respectively (8).
DNA extraction, PCR amplification, and sequencing. Rumen microbiome genomic DNA from the 30 samples was extracted according to the instructions of the FastDNA spin kit for soil (MPbio, CA, USA) in accordance with the manufacturer's instructions, and the details were referred to in our previous study (9). To extract total DNA, 1 ml of rumen fluid, 978 ml of sodium phosphate buffer, and 122 ml of MT buffer were added to the lysing matrix E tube, which was centrifuged at 14,000 rpm for 10 min. The supernatant was transferred to a 1.5-ml centrifuge tube, and 250 ml of polyphenylene sulfide (PPS) was added, which was mixed well and centrifuged at 14,000 rpm at ambient temperature for 5 min. The supernatant was transferred into a 2-ml tube with 900 ml of binding matrix and mixed thoroughly, then 500 ml of 5.5 M guanidine isothiocyanate solution was added. The solution was transferred to a SPIN filter and 500 ml of SEWS-M was added, which was then centrifuged at 14,000 rpm for 1 min. The residual solution was removed and dried for 3 min. After adding 100 ml of 55°C preheated DES (DNase/pyrogen free water) eluate for 5 min, the solution was centrifuged at 14,000 rpm at ambient temperature for 2 min. Total DNA was obtained after discarding the SPIN filter. The purity and concentration of DNA were detected by a spectrophotometer (NanoDrop2000, Thermo Fisher Scientific, Waltham, USA). DNA integrity was determined by 1% agarose gel electrophoresis at 5 V/cm for 20 min.
The primer pairs 338F (59-ACTCCTACGGGAGGCAGCAG-39) and 806R (59-GGACTACHVGGGT WTCTAAT-39) targeting the hypervariable region V3-V4 of the ruminal bacterial 16S rRNA gene were used to conduct PCR amplification. The PCR system and program were the same as in our previous study (8). PCRs were performed in triplicate. The PCR products were detected by 2% agarose gel electrophoresis. Purification and quantification of PCR products were performed using an AxyPrep DNA gel extraction kit (Axygen Biosciences, Union City, CA, USA) and a Quantus fluorometer (Promega, Madison, USA), respectively. Subsequently, the libraries were processed using the purified amplicons using a NEXTFLEX rapid DNA-seq kit (Bioo Scientific, USA) and were further sequenced on a MiSeq PE300 platform (Illumina, San Diego, CA, USA) (57).
Processing of sequencing data. Quality control (QC) for the raw sequencing data was performed using Trimmomatic software (version 0.39). Paired-end reads were merged using FLASH (V1.2.11, http:// ccb.jhu.edu/software/FLASH/). The Uparse software (version 7.0.1090 http://www.drive5.com/uparse/) was used for operational taxonomic unit (OTU) clustering: (i) extracting nonrepetitive sequences from optimized sequences (http://drive5.com/usearch/manual/dereplication.html); (ii) removing single sequences without repetitions (http://drive5.com/usearch/manual/singletons.html); (iii) performing OTU clustering for nonrepetitive sequences (excluding single sequences) according to 97% similarity, removing chimeras, and obtaining representative sequences of OTUs; and (iv) selecting sequences with .97% similarity to the representative sequence and generating OTU form. The RDP classifier (version 2.11) Bayesian algorithm (https://sourceforge.net/projects/rdp-classifier/) was used to perform taxonomic analysis for OTU representative sequences with greater than 97% similarity. The microbial community composition was counted at different taxonomic levels, which were then compared with the 16s rRNA database Silva (release 132, http://www.arb-silva.de). Alpha diversity indexes were calculated by Mothur software (version 1.30.2, http://www.mothur.org/wiki/Calculators). Based on the alpha diversity indexes (Shannon and Sobs), the rarefaction curves were drawn using the vegan package in R software (version 2.12.1). The distance algorithms of the principal coordinates analysis (PCoA) were weighted normalized UniFrac and unweighted UniFrac, with analysis of similarities (ANOSIM) as difference tests among groups. According to the relative abundance of ruminal bacteria, a Kruskal-Wallis H test was used to conduct significance tests of differences among groups at phylum and genus levels, respectively. The P value was corrected for false-discovery rate (FDR). Hierarchical clustering analysis (HCA) of bacteria with a top 50 relative abundance was conducted using Scipy packages of Python (Version1.0.0), and the clustering manner was average (58).
LC-MS metabolomics analysis. Prior to LC-MS, the rumen fluid samples were thawed and mixed thoroughly. The sample preparation steps were consistent with our previous research (10). The supernatant was transferred to an autosampler vial for LC-MS analysis. In addition, 20 ml of supernatant was transferred from each sample in each group to mix as a QC sample in this group. The platform for LC-MS analysis was the UHPLC-Q Exactive HF-X system (Thermo Fisher, MA, USA). Each rumen fluid sample (10 ml) was separated by an Acquity UPLC HSS T3 chromatographic column (100 mm Â 2.1 mm inside diameter [i.d.], 1.8 mm; Waters, Milford, USA) to perform mass spectrometry detection. The following chromatographic conditions were used: mobile phase A, 95% water 1 5% acetonitrile, containing 0.1% formic acid; mobile phase B, 47.5% acetonitrile 1 47.5% isopropanol 1 5% water, containing 0.1% formic acid. The following separation gradient parameters were used: 0 to 3.5 min, mobile phase A was decreased from linear 100% to 75.5%, mobile phase B was increased from linear 0% to 24.5%; 3.5 to 5 min, mobile phase A was decreased from linear 75.5% to 35%, mobile phase B was linearly risen from 24.5% to 65%; 5 to 5.5 min, mobile phase A was decreased from linear 55% to 0%, mobile phase B was linearly risen from 65% to 100%; 5.5 to 7.4 min, mobile phase A was maintained at 0%, mobile phase B was maintained at 100%; 7.4 to 7.6 min, mobile phase A was linearly risen from 0% to 48.5%, mobile phase B was linearly decreased from 100% to 51.5%; 7.6 to 10 min, mobile phase A was linearly risen from 48.5% to 100% and maintained for 2.2 min, mobile phase B was linearly decreased from 51.5% to 0% and maintained for 2.2 min. The flow rate was 0.40 ml/min, injection volume was 2 ml, and the column temperature was 40°C. The samples were ionized by electrospray, and mass spectrometry signals were collected in positive and negative ion scanning modes, respectively. The following mass spectrometry conditions were used: mass scan range (m/z), 70 to 1,050; heater temperature, 425°C; capillary temperature, 352°C; aux gas flow rate, 13 arbitrary units (arb). The spray voltages for positive and negative ion mode were 3,500 and 23,500 V, respectively, and the cyclic collision energy was 20 to 60 eV.
Data processing in metabolomics. The LC-MS raw data were imported into the metabolomics processing software Progenesis QI (Waters Corporation, Milford, USA) for baseline filtering, peak identification, integration, retention time correction, peak alignment, etc., and a data matrix, including the information of retention time, mass-to-charge ratio, and peak intensity were obtained. The sum normalization method was used to normalize the response intensity of the samples' mass spectrum peaks to obtain a normalized data matrix. Meanwhile, the MS and MS-MS data were annotated with the public metabolism database HMDB (http://www.hmdb.ca/) to obtain information about the metabolites.
The R package ropls (version1.6.2) was used to perform principal-component analysis (PCA), partial least squares discriminant analysis (PLS-DA), and orthogonal least partial squares discriminant analysis (OPLS-DA). The 7-fold cross validation and response permutation testing were conducted to evaluate the predictive ability and stability of the model. In addition, Student's t test and multiple of difference analysis were used for difference testing. The selection of significantly different metabolites between two groups was determined based on the variable weight value (VIP) obtained from the OPLS-DA model and the Student's t test P value. The P value was corrected by FDR adjustment. The metabolites with a VIP value of .1 and an FDR-adjusted P value of ,0.05 were defined as significantly different metabolites. The clustering and distance calculations for metabolites were analyzed by hierarchical clustering and Euclidean distance, respectively. The selection of significantly different metabolites among the five groups was accomplished through a Kruskal-Wallis H test. Metabolic pathway annotation was performed by using the KEGG database (https://www.kegg.jp/kegg/pathway.html). Spearman's correlation analysis was performed using Scipy packages of Python (version1.0.0). The P value was corrected by FDR adjustment. FDR-adjusted P values of ,0.05 signified a significant correlation. The range of coefficient of association (r) was from 21.0 to 1.0, and an r of .0 or an r of ,0 represented a positive and negative correlation, respectively (59).
Statistical analysis. DMI, lactation performance, rumen fermentation parameters, inflammatory cytokines, and oxidative stress indexes data in serum were analyzed by one-way analysis of variance (ANOVA) and Student's t test in SPSS statistics 22 (IBM, New York, USA). Significance was declared at a P value of #0.05; a tendency was declared at 0.05 , P , 0.10.
Data availability. 16S rRNA raw data from the study were deposited in the NCBI Sequence Read Archive (SRA) under accession no. PRJNA746433.

SUPPLEMENTAL MATERIAL
Supplemental material is available online only. SUPPLEMENTAL FILE 1, PDF file, 4.8 MB.