Dynamic fecal microenvironment properties enable predictions and understanding of peripartum blood oxidative status and nonesterified fatty acids in dairy cows

The transition period in dairy cows is a critical stage and peripartum oxidative status, negative energy balance (NEB), and inflammation are highly prevalent. Fecal microbial metabolism is closely associated with blood oxidative status and nonesterified fatty acids (NEFA) levels. Here, we investigated dynamic changes in total oxidative status markers and NEFA in blood, fecal microbiome, and metabolome of 30 dairy cows during transition (−21, −7, +7, +21 d relative to calving). Then the Bayesian network and 9 machine-learning algorithms were applied to dismantle their relationship. Our results show that the oxidative status indicator (OSI) of −21, −7, +7 d was higher than +21 d. The plasma concentration of NEFA peaked on +7 d. For fecal microenvironment, a decline in bacterial α diversity was observed at postpartum and in bacterial interactions at +7 d. Conversely, microbial metabolites involved in carbohydrate, lipid, and energy metabolism increased on +7 d. A correlation analysis revealed that 11 and 10 microbial metabolites contributed to OSI and NEFA variations, respectively (arc strength >0.5). The support vector machine (SVM) radial model showed the highest average predictive accuracy (100% and 88.9% in the test and external data sets) for OSI using 1 metabolite and 3 microbiota. The SVM radial model also showed the highest average diagnostic accuracy (100% and 91% in the test and external data sets) for NEFA with 2 metabolites and 3 microbiota. Our results reveal a relationship between variation in the fecal microenvironment and indicators of oxidative status, NEB, and inflammation, which provide a theoretical basis for the prevention and precise regulation of peripartum oxidative status and NEB.


INTRODUCTION
The transition period refers to the 3 wk before and after parturition and is a critical stage for dairy cows (Horst et al., 2021).The peripartum oxidative status, negative energy balance (NEB), and inflammation have been previously identified as major risk factors for high disease incidences in dairy cows during the transition period (Kerwin et al., 2022b).Generally, dairy cows experience a multifactorial and complicated condition consisting of oxidative status, NEB, and additional symptoms (Perrone et al., 2020).Peripartum challenge arises from profound shifts in digestion, endocrine physiology, energy metabolism, reproduction, and lactation (Marei et al., 2022;Vossebeld et al., 2022).Under the combined effects of multiple sources, indicators of oxidative status, NEB, and inflammation show specific longitudinal variants and high individual differences (Sheldon et al., 2019), which prevent full mechanistic exploration and early predictive warnings (Pascottini et al., 2020).
Fecal microbiota have been identified indicative of reliable prediction and regulation of host metabolic challenge (Feng et al., 2022).As individual "fingerprints," fecal microbiota vary between animals but remain stable within a single host, enabling identification of complex and variable host states (Chen et al., 2021).Precise predictions and the role of microbiota have been investigated in multiple large-scale cohorts in humans (Asnicar et al., 2021;Kurilshikov et al., 2021).In peripartum dairy cows, fecal microbiota have been reported to mediate inflammation and oxidative status in the liver, mammary gland, and uterus (Jeon et al., 2017;Hu et al., 2020;Pacífico et al., 2021), and thus they exert important physiological functions for Dynamic fecal microenvironment properties enable predictions and understanding of peripartum blood oxidative status and nonesterified fatty acids in dairy cows Sen-Lin Zhu, 1,2 Feng-Fei Gu, 1,2 Yi-Fan Tang, 1,2 Xiao-Han Liu, 1,2 Ming-Hui Jia, 1,2 Teresa G. Valencak, 1,2 Jian-Xin Liu, 1,2,3 and Hui-Zeng Sun 1,2,3 * reproduction as well as energy metabolism and lactation.Lipid metabolism is highly relevant for NEB and oxidative status in transition dairy cows, and hindgut microbiota could regulate hepatic lipid metabolism by influencing bile acid metabolism (Pacífico et al., 2021).
The mammary gland, as an organ with high metabolic intensity in transition dairy cows, is susceptible to inflammation and oxidative stress under the influence of microorganisms.Hu et al. (2020) found that fecal microbiota alter the blood-milk barrier, resulting in an invasion of neutrophiles and in the exacerbation of inflammation within mammary gland.As a major source of peripartum inflammation, a uterus infection is caused by intestinal pathogens such as Bacteroides, Porphyromonas, Fusobacterium, Prevotella, and Helcococcus (Jeon et al., 2017).Although previous work has shown that hindgut microbiota are involved in metabolic responses of the above mentioned organs, the exact functional role of fecal bacteria for regulating systemic oxidative status, inflammation, and NEB in transitioning dairy cows remains unclear.
Except for microbiota, fecal metabolites are the ultimate microbiome targets in the host, and microbial metabolites might mirror microbial functions (Roehe et al., 2016).With increasing evidence for the relationship between fecal microbiota and host oxidative status, inflammation, researchers focused on exploring potential mediating metabolites secreted by fecal microbiota, such as aromatic AA, short fatty acids, vitamins, and bile acids, among others (Nicholson et al., 2012).Although intestinal metabolic landscapes were observed in transitioning dairy cows exposed to different diets and farm management conditions, precise metabolic changes regulating levels of host oxidative status, inflammation, and NEB still remain to be discovered.In addition, the gut epithelium of dairy cows is covered by a single epithelium layer, which makes it even more susceptible to lumen acidification and lipopolysaccharides than the forestomach with a squamous epithelium (Petri et al., 2021).However, studies containing variation in metabolism and metabolites specifically secreted by fecal microbiota in dairy cows are still rare.Broad availability of metabolite tracing tools makes it possible to distinguish microbial metabolism and nonmicrobial metabolites (Yu et al., 2022).Moreover, advanced machine-learning models have proven relevant for screening variables and predicting host oxidative status and inflammation.
Here, we first explored stages variation of fecal microbial composition, functions, and related metabolites in transitioning dairy cows from 21 d before until 21 d after calving.Next, we investigated the potential correlation between temporal changes in the microenvironment and peripartum oxidative status indicator (OSI) and nonesterified fatty acids (NEFA), as obtained from multiple machine-learning algorithms.Our results provide a theoretical basis for noninvasive, accurate, and early warning of high levels of peripartum OSI and NEFA in dairy cows and give rise to the possibility of preventing and treating peripartum diseases accurately on a large scale.

MATERIALS AND METHODS
All experimental procedures used in this study were approved by the Animal Care Committee of Zhejiang University (Hangzhou, Chitna) and were conducted in accordance with the university's guidelines for animal research.The experimental protocol was approved by the Animal Care Committee of Zhejiang University (Hangzhou, China).

Sample Collection
Briefly, 30 healthy dairy cows in the peripartum period were first selected from a herd of 2,000 dairy cows, with parity (mean = 1.93,SD = 1.00) and BCS on −21 d (mean = 3.21, SD = 0.41) and +21 d (mean = 2.76, SD = 0.4).Body condition was scored following the method described by Edmonson et al. (1989), using the average of a graded scale ranging (1 = thin, 5 = fat) at 3 time points (0600, 1400, and 2000 h), to reduce the error of subjective factors.Body weight was estimated on calving day (BW; 690 kg, SD = 50) and +21 d (BW; 630 kg, SD = 40).Milk yield was recorded after calving on +7 d (39 kg, SD = 11), +14 d (42 kg, SD = 6), +21 d (46 kg, SD = 6).All animals had the same diet, water, and environment.Ingredients of the basal diet before calving and after calving are listed in Supplemental Table S1 (https: / / doi .org/ 10 .6084/m9 .figshare .21606732 .v3;Zhu, 2023a).Alfalfa, steam-flaked corn, and sugar beet pulp were included as replacements of wheat bran, rice, and straw after calving.Moreover, corn silage (28% to 19.8%) and oat hay (19.9% to 7.05%) were declined, with the supplement of soybean meal (10.3% to 17.7%).Blood samples (5 mL) were collected from the coccygeal vein of cows on −21 d (−3 wk), −7 d (−1 wk), +7 d (+1 wk), and +21 d (+3 wk) relative to calving day, each time at 0600 h with ethylenediaminetetraacetic acid (EDTA) evacuated tubes.Blood samples were centrifuged at 3,000 × g for 15 min to collect plasma at 4°C.Rectal fecal samples were collected before the morning feeding at 0600 h using sterilized gloves and stored in 50-mL frozen storage tubes.Fecal samples and plasma samples were flash frozen in liquid nitrogen and then stored at −80°C for subsequent analyses.A total of 120 fecal and 120 blood samples from 30 peripartum dairy cows were collected at 4 time points.For the subsequent validation of the prediction models, additional 85 fecal and 85 blood samples were collected using the same method.Among these 85 samples, 83 samples were used for predicting NEFA levels, 18 samples were used for predicting OSI levels (16 samples were overlapped).
We used FLASH (V1.2.8) to demultiplex the raw sequence data into paired-end fastq files.Clean data were stored in the NCBI SRA database as described in the data available statment.De-noizing and clustering algorithms that generate amplicon sequence variants (ASV) are included in DADA2 embedded in QIIME2 (Callahan et al., 2016).SILVA (138 database, https: / / www .arbsilva.de;Quast et al., 2013) was referenced for taxonomic annotation.At each time point, microbial taxa with a relative abundance greater than 0.01% in more than 15 samples were used for further analysis.The filtered taxonomy matrix was stored in Supplemental Tables S3, S4, and S6 (https: / / doi .org/ 10 .6084/m9 .figshare.21606732.v3;Zhu, 2023a).Alpha and diversities were calculated using QIIME2 (Bolyen et al., 2019).To compare the Bray-Curtis dissimilarity of the microbial community we used the vegdist function in the R package vegan (V2.6-2, https: / / cran .r-project .org/web/ packages/ vegan/ index .html;Dixon, 2003).The adonis test was used to determine significant differences caused by different categories of feed and host related metabolites.Finally, functional components of the microbiome were predicted by PICRUSt2 program.

Co-occurrence Network Analysis
To understand the microbial interaction patterns at all 4 time points, all the genera with an abundance greater than 0.01% in more than half of the samples from at least one of the 4 time points were retained for co-occurrence analysis.Spearman's rank correlation coefficient between the genera were pair-calculated using rcorr function in the Hmics package (v4.6.0,https: / / cran .r-project .org/web/ packages/ Hmisc/ index .html).In the rcorr function, the type arguments was set to Spearman.In the case of ties, the midranks method was used.The P-values were approximated using Fdistribution.Robust correlations (P < 0.05, |rho| > 0.60) were used to construct the networks.

Fecal Metabolite Measurement and Data Analysis
Frozen feces (100 mg) from each sample were thoroughly ground with liquid nitrogen, dissolved with 1 mL of 50% methanol buffer, and incubated for 10 min.After being stored at −20°C overnight to precipitate out protein, the mixture was then centrifuged at 4,000 × g for 20 min at 4°C.The supernatant was injected into the ultra-performance liquid chromatography system (SCIEX, UK) with an ACQUITY UPLC T3 column (100 mm × 2.1 mm, 1.8 μm, Waters, UK) for chromatographic reversed-phase separation.The TOF 5600 Plus high-resolution tandem mass spectrometer (SCIEX, Warrington, UK) with mass set at the 60 to 1,200 Da range was performed to obtain both positive and negative metabolic fragments eluted from the column.Peak detection and annotation were performed by CAMERA (Kuhl et al., 2012), MetaX toolbox in R and the CAMERA package.Each ion was identified by its retention time and mass-to-charge ratio.The raw metabolites matrix was stored in Supplemental Table S9 (https: / / doi .org/ 10 .6084/m9 .figshare.21606732.v3;Zhu, 2023a).

Dynamic Clustering of Microbiota and Metabolites
Kruskal-Wallis rank sum test was performed to select genera with significant changes among the 4 time points.Dynamics patterns in these 91 genera were identified by soft clustering function with Mfuzz in the Mfuzz package (V2.52.0;Kumar and Futschik, 2007).The number of clusters was determined by Dmin function in Mfuzz package with soft clustering for a range of cluster numbers c and reporting the minimum centroid distance.The elbow of the curve was used to determine the optimal cluster number for clustering.We optimized the setting of fuzzifier variables under elbow cluster classification at 3.60, 3.21, 3.86 for genera, host and microbiota co-metabolism, microbial metabolism, respectively.

Metabolites Origin Tracing and Metabolic Function Analysis
Identification of the origin of metabolites was accomplished in MetOrigin (http: / / metorigin .met-bioinformatics .cn/; Yu et al., 2022) with information from 7 well-known metabolite databases, which classified metabolites into 6 groups, including microbiota, host, co-metabolism, food, drug, and environment.We applied Deep MetOrigin Analysis, and chose Bos taurus (cow) as host parameter.Metabolomics pathway analysis (MetPA pathway topology analysis, https: / / www .metaboanalyst.ca/; Xia and Wishart, 2010) included in MetOrigin was applied to each metabolite from each cluster to identify metabolic pathways (P < 0.05).

Construction of Bayesian Networks
Bayesian networks were constructed to infer important interactions between fecal microenvironment and host physiological parameters.From each cluster, the relative importance of the first principal component was calculated to represent the cluster.The network topology was interpreted as successional dependency of these variables throughout the peripartum period (Scutari, 2010).The Bayesian network's structure was constructed using a tabu search algorithm in the bn-learn package (v4.8.3 https: / / cran .r-project .org/web/ packages/ bnlearn/ index .html;Scutari, 2010) after 2,000 bootstrap replicates.We performed model averaging to save significant arcs.Finally, all parameters of the Bayesian network were calculated using bn.fit function.

Microbiability Calculations
Microbiability is used to describe the capacity or ability of the microbiota to influence or contribute to a certain outcome or parameter.The linear mixed-effects model (LMM) was used to correct for fixed effects such as parity and milk yield, and to calculate the contribution of microbial random effects to the variance of blood indicators: where y is the phenotype NEFA, BHB, OSI, HPT, SAA; c is the vectors of the fixed covariates, consisting of parity and milk yield; K is the fixed effects matrix, including coefficients related to c; ms is the random effects vector, representing the random effects from microbiota; u is the random intercept vector, representing the random intercepts for each observation; and e is the error vector, representing the unexplained random errors.Gut microbes in each animal are considered random effects, which follow the distribution ms ~N 0 2 , M m σ ( ) (Xue et al., 2020).

Model Predictions for NEFA and OSI Levels
After model comparison, RF and SVM radial models were selected for predicting NEFA and OSI.We created a SVM radial classification to identify high or low NEFA and OSI levels in host peripheral blood using the tune.svmfunction in the e1071 package.10-fold cross-validation was applied and feature filtering was done for each cross-validation using the rfe function in the caret package.The features within the top 15 highest average coefficients from each cross-validation were selected for final model construction.For OSI levels at +7 d classification, SVM radial classification was used, and common top variables from 5-fold validation were ultimately included in the models.To validate the effectiveness of the model, we collected 85 additional blood and corresponding fecal samples from the peripartum cows (Supplemental Table S2; https: / / doi .org/ 10 .6084/m9 .figshare.21606732.v3;Zhu, 2023a) in the same dairy farm.After measuring their NEFA, OSI and microbial amplicon data using the same methods, we used our trained model for prediction.

Statistical Analysis
Kruskal-Wallis rank sum test was applied to perform comparisons among 4 time points, which was done using the kruskal.testfunction within the package stats in R (V4.1.0).Significant differences from the multiple comparisons were further analyzed by pairwise comparison using Dunn's test, which was conducted with dunntest function in the FSA (V0.9.3, https: / / cran .r-project .org/web/ packages/ FSA/ index .html)package.Following pairwise comparison, the absolute highest values were identified and the number of highest values at −21, −7, +7, +21 d were determined.Correlation analysis was performed using Spearman rank correlation, and the asymptotic P-value was calculated using corPvalueStudent function in Weighted Gene Co-expression Network Analysis (WGCNA; Langfelder and Horvath, 2008).In all models, P-values < 0.05 after Bonferroni correction were considered as significance, and 0.05 ≤ P ≤ 0.10 was considered a significant trend.

Dynamics of Fecal Microbial Composition and Microbiota Abundance During Peripartum Period
A total of 10,095,354 raw reads were obtained from the sequencing data of 120 fecal samples, collected from 30 dairy cows at 4 peripartum time points.After retaining microbial taxa with a relative abundance greater than 0.01%, a total of 8,418,878 reads were retained after quality control, and 32,580 ASV were assigned from these reads (Supplemental Table S3).Our results show that Chao1 indices at −21 and −7 d were significantly higher than at +7 d (P = 0.033 and P < 0.001) and +21 d (P = 0.018 and P < 0.001), −21 d did not differ significantly from −7 d (P > 0.05; Figure 1A).The Shannon indices trend resembled Chao1.The results of microbiota β-diversity among the 4 time points showed significant differences (Anoism P < 0.001, R = 0.49; PERMANOVA P < 0.001, R 2 = 0.18).Bray-Curtis distance from other time points to −21 d increased significantly over time (distance to −21 d: −7 d lower than +7 d, P = 0.0012; +7 d lower than +21 d, P = 0.0012; Figure 1B).In addition, Bray-Curtis distance differed most between −7 to +7 d (PERMANOVA P < 0.001, R 2 = 0.15).
Genera in MicroC3 reached their peak abundance on −7 d, with the most complex distribution of 9 phyla (Figure 1E, Supplemental Table S7).Genera in Mi-croC4 were Firmicutes, Actinobacteria, Bacteroidetes, Fibrobacteres, Spirochaetes, and Tenericutes, which elevated to the highest abundances on +7 d.Microbiota cluster 4 was the only cluster with more Bacteroidetes than Firmicutes (Figure 1E, Supplemental Table S7).

Prediction of Postpartum Phenotypes Based on Fecal Microbiota and Metabolites Using Machine-Learning Models
Nine machine-learning and statistical models were applied to predict blood NEFA and OSI separately.The predictive power of each model with default parameters for the NEFA prediction was determined by an average a-c S uperscript letters in the table represent significant pairwise differences (Dunn's test).Groups sharing the same letter are not significantly different from each other, whereas groups with different letters are significantly different at a predetermined level of significance (P < 0.05).If a group is labeled with 2 letters, it indicates that this group does not significantly differ from 2 distinct groups labeled with each of the individual letters.The superscript letters "ab" indicate that the group labeled with "ab" does not differ significantly from both the groups labeled with "a" and "b." 1 ALT = alanine aminotransferase; AST = aspartate aminotransferase; ALB = albumin; TP = total protein; CREA = creatinine; GLU = glucose concentrations; CHOL = cholesterol ; TG = triglycerides; NEFA = nonesterified fatty acids; SAA = amyloid; CPL = ceruloplasmin; HPT = haptoglobin; CAT = plasma catalase; GSH.PX = glutathione peroxidase; MDA = malondialdehyde; TOS = plasma total oxidative status; OSI = oxidative status indicator; SOD = superoxide dismutase; T-AOC = total antioxidant capacity.
2 Oxidative status indicator = total oxidative status per total antioxidant capacity.
value under the 10-fold cross-validation without feature selection.The highest accuracy of 0.91 was observed in the SVM radial model approach (Figure 7A).After 10 times, feature selection using the training set of 10-fold cross-validation, 4 microbiota and 2 metabolites with the top 10 highest Gini coefficient (Figure 7B).The final SVM radial model revealed that 2 metabolites including 3-hydroxy-2-methylpyridine-4,5-dicarboxylate and pyrocatechol, and 3 microbiota including Mycoplasma, Porphyromonas, and Paludibacter contributed to the discrimination of high and low NEFA status (AUC = 1, accuracy = 1.00 in training set; AUC = 1, accuracy = 1.00 in testing set; Figures 7C-E).Since the accuracy in the test set was equal to that in the training set, to eliminate the problem of over-fitting, 85 additional blood and fecal samples from the peripartum cows were applied for external validation (Supplemental Table S2).A total of 83 additional dairy cows were applied in NEFA validation.Our model had an accuracy of 0.963 for predicting NEFA among the external peripartum data (Supplemental Table S19; https: / / doi .org/ 10 .6084/m9 .figshare.21606732.v3;Zhu, 2023a).The SVM model with radial basis function (kernel) was used for predicting OSI prenatally (Figure 7F).Akkermansia, Desulfovibrio, Oxobacter and m-hydroxycinnamic acid were the common top microbiota from 5-fold crossvalidation and a model built with these features had an accurate postpartum discrimination ability (AUC = 0.969, accuracy = 0.976 in training set; AUC = 1, accuracy = 1.00 in testing set; Figures 7G, 7H, 7I, and 7J).The model test using external samples also showed an accuracy of 0.889 and AUC 0.929.

DISCUSSION
Various factors influencing oxidative status, inflammation and NEB, in the peripartum period of dairy cows are interrelated (Sordillo and Raphael, 2013;Putman et al., 2018).In our study, we observed the significant rise in milk yield and decline in body weight after caving, which have been shown to be an important contributor to the rise of NEFA and BHB after calving.While the body fat mobilized is also a major sources of oxidative status (Sordillo and Raphael, 2013).The upward trend of the concentration of MDA from −7 to +7 d corroborated the previous studies.However, OSI on −21, −7 d shows no significant difference with +7 d.This could due to weaker natural antioxidant defense system, reflected in lower SOD and T-AOC before calving.oxidative status is often accompanied by inflammation, and positive acute phase proteins like SAA and HPT showed similar trend with MDA.Fecal microbiota, important markers of host oxidative status, inflammation, NEB and metabolism (Singh et al., 2017;Vallianou et al., 2021), provide a new tool for understanding and mitigating peripartum oxidative status, NEB and inflammation in dairy cows.
Previous studies have shown that fecal microbiota are better indicators of host health than rumen microbiota in ruminants (Shen et al., 2017;Wallace et al., 2019).
In the present study, we describe the variation between sequential stages in peripartum period in microbiota and metabolites in dairy cows during transition.We found that oligotrophic phyla declined from −21 and −7 to +7 and +21 d, and carbohydrate dependent genera rose at +7 and +21 d, especially on +7 d.Fecal metabolites from each time point mainly consisted of aromatic compounds, lipids and carbohydrates and amino acids.We identified the relationships between relevant pathways within aromatic compounds, lipids and AA metabolism for supplementing and promoting energy metabolism, which may potentially relate to host peripartum oxidative status, inflammation, NEB.The relationship between fecal microenvironment and host OSI and NEFA was confirmed by Bayesian network.Finally, an identification model of host OSI and NEFA with high accuracy based on fecal microbiota and metabolites was constructed.
The temporal variation of fecal microbiota in dairy cows during the peripartum period was carefully described in our study.In contrast to rumen microbiota remodeling between −21 and −7 d as shown in the study of Bach et al. (2019), hindgut microbiota are slower in adapting to the new lumen contents and they show an imbalance on +7 and +21 d (Huang et al., 2020).The decrease in the abundance of oligotrophic phyla such as Verrucomicrobia and increase of copiotrophic Bacteroidetes after parturition is corresponding to feed changes from forage to concentrate (Serena et al., 2018;Bach et al., 2019).At genus level, MicroC4 was the only cluster dominated by Bacteroidetes, with high relative abundances of Bacteroidales unclassified, Lach- Microorganisms generally infiltrate the hindgut metabolic microenvironment.Three different feedrelated categories of metabolites, sterol lipids, organooxygen compounds and phenols, all derived from soybean, grain, hay and their metabolic products, were identified as significant factors shaping the microbiota community during the peripartum period and indicating a major role for feed for microbiota.Sterols lipids were the major factors for microbiota variation in this study.Plant sterols were reported to have antioxidant and anti-inflammatory activities (Vilahur et al., 2019), and can be absorbed in the hindgut (Samtiya et al., 2021).However, plant sterols also caused the reduc-  tion of CHOL (Plat and Mensink, 2002), which could enhance NEB.Therefore, as the major source of sterols lipids, supplementation of soybean meal and grain gradually in refinement stages after calving is more recommended.Furthermore, we addressed the microbial function for dairy cows during the transition periods.More microbiota-related metabolites were upregulated on −21 d, −7 d than on +7 d, +21 d, indicating the higher metabolic intensity during the postpartum period.The metabolic function was partly consistent with the PICRUSt2 results on energy metabolism, and were functionally supported by lipids, amino acids and aromatic compound metabolism, because 16S amplicon functional prediction was less likely to accurately predict microbial function in the hindgut of the animals contrary to human (O'Mara et al., 1997).As for lipid metabolism, glycerophospholipid metabolism was mainly enriched on +7 d (Co-MetaC4).Lysophosphatidylethanolamine and LysoPC were the 2 largest components within the glycerophospholipids, which were upregulated on +7 and +21 d (belonging to Co-MetaC2 and 4), and have been reported to readily cross the enterocyte membrane by passive diffusion (Stahl et al., 1999).Lysophosphatidylcholine can efficiently drive intestinal lipid absorption by promoting lipid uptake and enforcing the lipoprotein effect.Lysophosphatidylethanolamine is known for inducing nonalcoholic fatty liver disease by suppressing liver lipolysis and fatty acid biosynthesis (Nakano et al., 2009;Yamamoto et al., 2022), which can potentially accelerate the mobilization of body fat and the deposition of liver fat between +7 d, +21 d.As for increased AA metabolism on +21 d (Co-MetaC3, MetaC6, 7), we noticed that aromatic AA metabolism was enhanced on +21 d which may be due to the regulation of gut epithelial viability and barrier integrity (Liu et al., 2020).Phenylalanine, tyrosine and tryptophan might relate to the reduction of inflammation and oxidative status on +21 d.For aromatic metabolism, aminobenzoate degradation was the major pathway enriched on −21 d, −7 d.As the newest vitamin B, aminobenzoate is widely used as supplement in dairy cows since it is involved in gluconeogenesis and it increases milk production (Preynat et al., 2009).Interestingly, these major downstream metabolites of metabolic pathways are involved in the TCA cycle and glycolysis.It is interesting that the upregulated energy metabolism between +7 and +21 d were not only dominated by a higher proportion of carbohydrates, but also by intermediate metabolites of energy metabolism from lipids, amino acids and aromatic compounds metabolism.Moreover, between the prenatal and the postnatal periods, the energy metabolism pathways shifted from aromatic compounds metabolism toward lipid and AA metabolism.As recently reported, up-regulated glycolysis of fecal microbiota may interfere with many important biosynthetic pathways that are critical for intestinal cell proliferation as observed in human studies (Wang et al., 2019).Some microbial respiration products, such as peroxisome proliferator activated receptors were reported to be activated by nicotinamide adenine dinucleotide and ATP to induce peroxidase proliferation, thus affecting the intracellular energy balance of the host (Scatena et al., 2008;Tian et al., 2022), as shown in the correlation analysis (Figure 5B, Supplemental Table S17).
Previous studies have detailed the relationships between host blood parameters and fecal microbiota in humans and monogastric animals (Schulfer et al., 2019;Ni et al., 2021).Here, we explored correlation between the dynamics of fecal microenvironment and 3 predominant blood OSI, HPT and NEFA levels during the peripartum period of dairy cows.Interestingly, we noticed that fecal microbiota is one of the potential factors regulating blood indicators.Thus, hindgut microbiota and their metabolites might be one of the potential factors responsible for oxidative status, inflammation and the NEB in peripartum dairy cows.The high accuracy of our machine-learning model relates to reliable fecal parameters for high level of OSI and NEFA prediction and diagnosis.Moreover, the drivers for NEFA prediction including carbohydrate metabolism related genera and phenylalanine metabolism dependent metabolites, provide further evidence for the effects of these pathways on the host.Although our model demonstrates a potential association between prenatal microbiota and postnatal OSI, our data are far from sufficiently explaining the regulatory mechanism.The difficulties with obtaining the relative abundances beyond "-omics" data and the insufficient data volume relating to high costs challenge the sequencing approach to predict OSI and NEFA.We set out to build predictive models using abundance ratios of microbiota and their metabolites for the predominant and stable genera in the fecal.
Taken together, by characterizing the composition and function of microbiota and metabolites in the intestinal microenvironment of dairy cows during the peripartum period, we explored potential correlation and assessed the predictive power of dynamic fecal characteristics with host oxidative status, NEB and inflammation.These insights of ruminant fecal microenvironments between sequential stages in peripartum period not only suggest a relationship between the intestinal microenvironment and the OSI, NEFA level in ruminants, but also complement the importance of further research on hindgut microbiota for ruminant health.Thereby extended knowledge and improved modeling will provide tools to lower and ultimately prevent peripartum oxidative status, NEB in dairy cows.

CONCLUSIONS
Our study reveals variations in hindgut microbiota composition, interactions, and metabolism during the peripartum period in dairy cows.These findings highlight the crucial role of microbiota and their metabolites in peripartum metabolic disturbances during the transition phase.We identified key metabolites associated with co-metabolism pathways such as amino acids, lipids, aromatic compounds, and energy metabolism.These metabolites, including phenol, phenylalanine, LysoPC 15:0, LysoPE 9:0, citric acid, and succinic acid, showed strong correlations with the host's oxidative status and NEB.Specific microbiota clusters (2, 4, and 7) and MetaC2, 4, and 7 were also associated with changes in oxidative status and energy metabolism.Furthermore, 3-hydroxy-2-methylpyridine-4,5-dicarboxylate, pyrocatechol, Mycoplasma, Porphyromonas, and Paludibacter within NEFA and OSI-related clusters demonstrated high accuracy in predicting NEFA (82.4%) and OSI (88.9%) levels.Overall, our study provides a theoretical foundation for implementing preventive protocols to regulate peripartum oxidative status and NEB in dairy cows in the future.
Figure 1.The dynamic changes of the fecal microbiota of 30 cows at −21, −7, 7, and 21 d relative to calving.A: The α diversity (Chao1 and Shannon) indexes of fecal microbiota at 4 time points.B: Clustering and comparison of β diversity index (Bray-Curtis) of 30 dairy cows' fecal microbiota at 4 time points C: Microbial abundance variation at the phylum level.D: Four temporal genera clusters formed by 91 differentially abundant microbiota.E: Phylum sources of the genera within 4 dynamic microbiota clusters.Clusters start with MicroC indicate that clusters formed from differential microbiota.Multiple comparison was done by Kruskal-Wallis rank sum test and pairwise comparison using Dunn's test.Four time points refer to −21, −7, 7, and 21 d relative to calving.

Figure 2 .
Figure 2. The variation of microbial interactions and the fecal metabolome of 30 cows at −21 d, −7 d, +7 d, and +21 d relative to calving.A: The 106 genera's co-occurrence network edge statistics comparison among 4 time points.B: Number of metabolites from different sources analyzed by MetOrigin.C: Distribution and contribution (Adonis R 2 ) of feed-related metabolites significantly affected microbiota community.D: The 6 categories derived from 216 differential microbial sourced metabolites.E: Circle map displayed the summaries of metabolites reaching the highest values at 4 time points.F: Temporal clustering of differential microbial metabolites.The network at each time point was constructed from 106 genera, using (|rho| > 0.6 and P < 0.05) as edge screening criterion.Moreover, the contribution of metabolites to genus level microbiota was analyzed by PERMANOVA (P < 0.05), using the first principal component value of the metabolic abundance matrix as the factor representing the category of metabolites.Multiple comparisons were done by Kruskal-Wallis rank sum test and pairwise comparison using Dunn's test.P-value (*P < 0.05) after Bonferroni correction was set as significant cutoff.Adonis R 2 : in adonis analysis, R-square indicates to what extent the independent variable explaining the variations between samples.

Figure 3 .
Figure 3. Variation of microbial function predicted by PICRUSt2.Microbial function displayed by Metacyc pathway abundance at −21 d, −7 d, 7 d, 21 d relative to calving.A total of 120 fecal samples from 30 cows at 4 time points were used for function prediction.Time 1, 2, 3, 4 refer to −21 d, −7 d, +7 d, +21 d, respectively.The size of red dot represents abundance of Metacyc pathways.

Figure 4 .
Figure 4. Enriched Kyoto Encyclopedia of Genes and Genomes (KEGG) functions of dynamic metabolites clusters.Clusters starting with "Co-Meta" indicate that the metabolites in these clusters are either the host cow or the microbiota.Clusters starting with "Meta" indicate that the metabolites in these clusters are solely from the microbiota.Metabolomics pathway analysis (MetPA pathway topology analysis, https: / / www .metaboanalyst.ca/ ) was applied to identify the pathways (P < 0.05).

Figure 5 .
Figure5.Core metabolic pathways (lipid, AA, aromatic compounds, and energy metabolism) from metabolite clusters with increasing trend at +7 and +21 d relative to calving.A: Kyoto Encyclopedia of Genes and Genomes (KEGG) map of core microbial metabolism pathways of lipid, AA, aromatic compounds and energy metabolism.The metabolites within the differential microbial metabolites are given in blue, and the metabolites shared in energy metabolism are colored in red.Unaffected metabolites from the 4 time points are in gray.B: Correlation between the key metabolites in these pathways and NEFA, HPT, OSI.All of the 120 fecal samples and blood samples were applied for correlation analysis, and the displayed correlation was tested by Spearman's rank correlation with Bonferroni correction (P < 0.05).OSI: Oxidative status indicator.NEFA: Nonesterified fatty acids.HPT: haptoglobin.

Figure 6 .
Figure 6.Correlation assessment of dynamic microbiota, microbial metabolites clusters (Co-Meta clusters, Meta clusters in Figure2) and OSI, NEFA, HPT.A: Bayesian networks generated from HPT, OSI, NEFA, and dynamic microbial and metabolic clusters.B: Functional enrichment of metabolic clusters (MetaC7, MetaC2).C: Variation of NEFA and OSI of the 30 dairy cows between high and low groups based on NEFA and OSI at +7 d (Time 3) relative to calving.The structure of Bayesian networks was analyzed using tabu algorithm with arc strength >0.5 under 2,000 times Botstrapping based on the relative abundance of genera microbiota and blood index.OSI: Oxidative status indicator.NEFA: nonesterified fatty acids.HPT: haptoglobin.

Figure 7 .
Figure 7.The performance of different prediction models of host OSI, NEFA levels in blood based on microbes and microbial metabolites within related clusters (MicroC2, MetaC7 for OSI prediction, MicroC4, MetaC2 for NEFA prediction).A and F: Prediction accuracy using different models and variables for NEFA and OSI respectively.B and G: The selected variables for the support vector machine (SVM) radial model and their importance biomarkers for predicting NEFA and OSI, respectively.C and H: Receiver operating characteristic (ROC) curve of the performance of the SVM radial model for NEFA and OSI in the training data set, respectively.D and I: ROC curve of the performance of the SVM radial model for NEFA and OSI in the testing data set, respectivley.E and J: ROC curve and confusion matrix of the performance of the SVM radial model for NEFA and OSI in external data set, respectively.OSI: Oxidative status indicator.NEFA: nonesterified fatty acids.HPT: haptoglobin.Sensitivity: true positive rate.Specificity: true negative rate.
Zhu et al.: DYNAMIC FECAL MICROENVIRONMENT PROPERTIES

Table 1 .
Zhu et al.: DYNAMIC FECAL MICROENVIRONMENT PROPERTIES Comparison of plasma physiological parameters, inflammation, oxidative status, and phenotypic characteristics

Table 2 .
Zhu et al.:DYNAMIC FECAL MICROENVIRONMENT PROPERTIES Fecal microbiota contribution to certain host parameters of oxidative status, inflammation, and negative energy balance (NEB) in the blood after calving calculated by the linear mixed-effects model after correcting fixed effects such as parity milk yield and weight variation factors 1 (Blanco- Morales et al., 2020)orphyromonadaceae unclassified, Bifidobacterium, Clostridia unclassified on +7 d.These microbiota have been known for the utilization of carbohydrates and production of organic acids(Blanco- Morales et al., 2020).A co-occurrence network comparison revealed vulnerable and uniform interactions between microbiota on +7 d, which further illustrate the remodeling of hindgut microbiota under the effect of a rapid feed dietary shift.Investigating functional dissimilarity of intestinal microbiota by PICRUSt2, we found that the microbial function is kept stable prenatally.Remarkably, energy metabolism including glycolysis and TCA cycle pathway increased between −7 and +7 d, and reached higher values between +7 and +21 d, reflecting the adaptability of microbiota in response to high energy requirements between +7 and +21 d.