Animals were treated, and samples were collected in strict accordance with the Guidelines for the Care and Use of Laboratory Animals of China. All procedures were approved by the Institutional Animal Care and Use Committee of Sichuan Agricultural University (no. DYY-2018203039). This study was conducted on a modern dairy farm in Sichuan, China. The sampling period was from January to April 2022, with an average temperature of 12 ~ 22°C. The heifers were housed in the same pen during the trial, fed and watered ad libitum, and fed a total mixed ration (TMR) twice daily. TMR diet formulation and dry matter intake were reported by previous studies (Luo et al., 2023). The primiparous cows were examined daily by a licensed veterinarian and milked at 06:00, 14:00, and 21:30. Cows were transferred to the pre-parturient barn 3 weeks before calving and fed the pre-parturient TMR diet, after which they were transferred to the neonatal barn and fed the post-parturient TMR diet. The trial enrolled 90 healthy Holstein cows and recorded disease information for the entire transition period, from 3 weeks prepartum to 3 weeks postpartum. The removal of 14 cows due to premature delivery, dystocia, and accidental death resulted in 76 cows with normal deliveries.

Blood and feces were collected before the morning feeding. Serum and plasma were separated using negative pressure and sodium heparin blood collection tubes, respectively. Feces and blood were collected at three time points: 7 days before calving, the day of calving, and 7 days after calving. A portable ketosis detector (WD1621, Nova Bio Vet, Waltham, MA, United States) was used to monitor blood β-hydroxybutyric acid (BHB) levels in cows 7 days after calving. Cows with plasma BHB > 1.2 mmol/L at 7 days postpartum and with no other perinatal diseases were assigned to the ketosis group (Duffield, 2000). Cows without any recorded diseases throughout the transition period were defined as the healthy group. Ten cows from the ketotic and healthy groups were randomly selected based on BHB concentrations at 7 days postpartum. Samples from the ketotic and healthy groups 7 days postpartum were named KET+7 (n = 10) and HEA + 7 (n = 10), respectively. Serum, plasma, and fecal samples were collected on the day of calving and 7 days before calving based on cow numbers. The ketotic and healthy groups, on the day of calving, were named KET-0 (n = 10) and HEA-0 (n = 10), respectively. The ketotic and healthy groups, 7 days before calving, were named KET-7 (n = 10) and HEA-7 (n = 10), respectively. Serum samples from three time points were tested for biochemical indices. Plasma and feces samples from the KET-7 and HEA-7 groups were selected for metabolomic and 16S rRNA amplicon sequencing.

Serum NEFA, BHB, glucose, insulin, and triglycerides (TG) concentrations were determined using commercially available kits from the Nanjing Jiancheng Bioengineering Institute (no. A042-2-1, no. E030-1-1, no. F006-1-1, no. H203-1-2, and no. A110-2-1, respectively). The revised quantitative insulin sensitivity check index (RQUICKI) was calculated from the plasma NEFA, glucose, and insulin concentrations and used to assess insulin resistance in dairy cows (Xu et al., 2014). All tests were performed according to the manufacturer’s instructions.

Fecal samples were pretreated for metabolite separation using an ultra-high-performance liquid chromatography (UHPLC) system (1290 Infinity LC, Agilent Technologies, Santa Clara, CA, United States). A 2.1 mm × 100 mm ACQUIY UPLC BEH Amide 1.7 μm column (Waters, Ireland) was used for this system. Mass spectrometry was performed using a quadrupole time-of-flight (Q-TOF) 6600 system (AB SCIEX, Toronto, Canada). UHPLC-Q-TOF MS was performed at Shanghai Applied Protein Technology Co., Ltd. The sample pretreatment, equipment parameters, and metabolite identification methods were the same as those used in our previous report (Luo et al., 2022).

Total DNA from the feces was extracted using CTAB, and primers were used to amplify the V3-V4 region of the 16S rRNA gene (Logue et al., 2016). The amplification products were detected through 2% agarose gel electrophoresis and recovered using a recovery kit (Beckman Coulter Genomics, Danvers, MA, United States). Purified amplification products were evaluated using a bioanalyzer (Agilent 2100) and Illumina (Kapa Biosciences, Woburn, MA, United States) library quantification kits. The library was sequenced using a NovaSeq 6000 sequencing platform (Illumina). Sequencing data were spliced, filtered, and noise-reduced to obtain amplicon sequence variant (ASV) and an abundance table. Alpha and beta diversity analyses were performed based on the ASV sequences and abundance tables. Species annotation was based on ASV (feature) sequence files using the SILVA (Release 138¹) and NT-16S databases. The sequences were spliced and filtered to obtain ASV feature sequences and ASV abundance tables.

Cold methanol/acetonitrile (1:1, v/v) extract was added to the plasma samples and the isotope standard and was centrifuged at 14,000 × g for 20 min at 4°C. The supernatant was collected in a new centrifuge tube and vacuum-dried. For liquid chromatography-mass spectrometry (LC–MS) analysis, the samples were re-dissolved in 100 μL acetonitrile/water (1:1, v/v) solvent and transferred to LC vials. Plasma metabolites were isolated using a UHPLC system (1290 Infinity LC; Agilent Technologies). C18 columns (Waters, C18-2.1 mm × 100 mm, 1.7 μm) were used for this system. MS was performed using a 6,500 system (AB SCIEX). UHPLC-MRM-MS was performed at Shanghai Applied Protein Technology Co., Ltd.

Statistical analysis was performed using SPSS software (version 27.0; IBM SPSS Inc., Chicago, IL, United States). Data are presented as the mean and standard error of the mean. Random forest predictive modeling and ROC analysis were performed using the R language package (version 4.3.2) and SPSS, respectively, to screen for potential warning biomarkers of ketosis. Spearman’s correlation analysis was performed using the OmicStudio tool at https://www.omicstudio.cn.

Metabolites were analyzed for variability using the Student’s t-test and multivariate statistical analysis. Multivariate statistical analyses included principal component and orthogonal partial least squares discriminant analyses. p-values were obtained using the Student’s t-test. p-values were adjusted using the Benjamini-Hochberg method. Orthogonal partial least squares discriminant analysis was used to determine variable importance in projection (VIP). Screening for differential metabolites in the feces was based on VIP > 1 and p < 0.05.

Regarding the fecal 16S rRNA amplicon sequencing, the Shannon index represents the α diversity of the two comparison groups. The variability in α diversity was reflected by the p-value of the Student’s t-test. The β-diversity of the two comparison groups was obtained based on Bray–Curtis distances. The variability in β-diversity was illustrated by p-values based on the anosim approach. Differential genera were screened based on linear discriminant analysis (LDA) and the Wilcoxon–Mann–Whitney test. The fecal microbiota was screened using LDA score > 2 and p < 0.05 as the screening criteria to explore the differential microbiota of ketosis. Finally, the variability of bile acids in the plasma was based on the p-value obtained from the Student’s t-test.

Energy-related indices in cows in the ketosis and healthy groups were measured at three time points: 7 days before calving, on calving day, and 7 days after calving, to observe changes in energy metabolism in the cohort. No significant changes were observed 7 days before calving in NEFA, BHB, glucose, insulin, RQUICKI, or TG in the serum of cows in either the KET-7 or HEA-7 groups (Figure 1). Serum NEFA, glucose, and insulin levels were significantly higher (p < 0.05), and RQUICKI and TG levels were significantly lower (p < 0.05) in the KET-0 that compared to those of HEA-0 group. Serum NEFA and BHB were higher (p < 0.01), and RQUICKI was lower (p < 0.01) in the KET+7 than that those in the HEA + 7 group. The results of ROC analysis showed that NEFA, BHB, glucose, insulin, RQUICKI and TG had low warning potential (AUC < 0.7) at 7 days before calving (Supplementary Table S1). It indicates that 7 days prepartum is not suitable for early warning with energy related indicators. Serum energy-related indicators suggest that the cows were already experiencing metabolic differences on the day of calving and that ketotic cows had greater energy deficits than healthy cows and had begun to mobilize lipolysis.

To screen the early warning of ketosis at 7 days before calving, the fecal metabolomics was performed using LC–MS/MS. In the positive and negative ion modes, 1,234 and 790 metabolites, respectively, were identified in the raw data after peak area extraction, missing-value removal, and secondary mass spectrometry. Principal component analysis showed that the quality control samples clustered together, indicating good stability of the instrument during the detection process. The model parameters for orthogonal partial least squares discriminant analysis in the positive and negative ion modes were R²Y = 0.9922 and Q² = −0.06286, and R²Y = 0.9386 and Q² = −0.1267, respectively. When R²Y > 0.5 and Q² < 0, the model was considered reliable for the two comparison groups, and the analysis was continued (Supplementary Figure S1).

In the positive and negative ion modes, 95 and 49 differential metabolites, respectively, were screened with VIP > 1 and p < 0.05. The types of differential metabolites were categorized and visualized (Figure 2A; Supplementary Table S2). Differential metabolites had the highest percentage of lipid and lipid-like molecules (32.64%), followed by organic heterocyclic compounds (19.44%) and benzenoids (14.58%). Fatty acids contain a higher percentage of lipids and lipid-like molecules. Since bile acids are involved in lipid digestion and absorption, we were concerned about the changes in bile acids in the feces. A total of 20 bile acids were detected in the feces, with significant differences in beta-muricholic acid (β-MCA), gamma-muricholic acid (γ-MCA), 12-ketodeoxycholic acid (12-KDCA), isodeoxycholic acid (Iso-DCA), taurolithocholic acid (TLCA), and histidine-conjugated chenodeoxycholic acid (His-CDCA) between the KET-7 and HEA-7 groups (Figure 2B; Supplementary Table S3). The 20 bile acids tested were categorized according to their source and structure into unconjugated primary bile acids (UnconPBA), conjugated primary bile acids (conPBA), unconjugated secondary bile acids (UnconSBA), and conjugated secondary bile acids (conSBA). The major bile acid in the feces of the KET-7 group was UnconSBA (75.17%), followed by UnconPBA (16.6%). The differential bile acids were primarily derived from UnconSBA. Finally, the KET-7 group showed a significant difference in UnconPBA compared to that of the HEA-7 group, which may be closely related to the increased proportion of CA in the KET-7 group (Supplementary Figure S2).

Twenty bile acids were screened for potential warning biomarkers using the random forest model and ROC analyses (Figure 3; Supplementary Table S4). The random forest model analysis used MeanDecreaseGini (Gini index) as an indicator to evaluate the accuracy of the metabolite prediction model. The two most abundant bile acids were His-CDCA and iso-DCA (Figure 3A). The ROC curves were evaluated for accuracy using the area under the curve (AUC). His-CDCA and iso-DCA had AUC values of 0.85 and 0.84, respectively, and had the potential to warn of ketosis (Figure 3B). Additionally, a significant correlation existed between bile acids in feces 7 days before calving and plasma biochemical indices 7 days after calving (Figure 3C). For example, fecal bile acid iso-DCA was significantly and positively correlated with NEFA (r = 0.49, p = 0.03) and BHB (r = 0.49, p = 0.03). His-CDCA in feces was significantly correlated with plasma NEFA (r = 0.65, p = 0.002) and BHB (r = 0.46, p = 0.04) levels. TLCA in feces was significantly correlated with plasma BHB (r = 0.49, p = 0.002) and NEFA (r = 0.66, p = 0.03). The ratio of unconjugated to conjugated bile acids was significantly correlated with plasma BHB (r = 0.65, p = 0.002) and NEFA (r = 0.46, p = 0.04). Thus, fecal untargeted metabolomics confirmed that metabolic differences in fecal bile acids had already occurred in ketotic cows 7 days before calving. These metabolic differences may be closely related to lipid metabolism. Alterations in fecal bile acids may contribute to the development of ketosis. These results indicate that iso-DCA is potential warning metabolites for ketotic cows 7 days before calving.

Significant metabolic differences were identified in secondary bile acids in ketotic cows 7 days prepartum using fecal untargeted metabolomics. Furthermore, the formation of secondary bile acids was closely related to microbial metabolism. Therefore, the microbiota was identified using 16S rRNA amplicon sequencing to verify the relationship between microbiota and bile acids 7 days prepartum. The α- and β-diversity analyses were performed based on ASV sequence and abundance tables. The ASV numbers and Shannon indices of the KET-7 and HEA-7 groups did not differ significantly (Figures 4A,B). Principal coordinates analysis showed that the 2 comparative groups clustered together, suggesting no difference in microbial structure between the KET-7 and HEA-7 groups (Figure 4C).

The dominant phyla were Firmicutes (71.56 ± 1.03% vs. 72.9 ± 0.72%) and Bacteroidota (22.35 ± 1% vs. 21.09 ± 0.88%), followed by Verrucomicrobiota (1.52 ± 0.23% vs. 1.93 ± 0.34%) and Proteobacteria (1.28 ± 0.21% vs. 1.11 ± 0.07%). In the KET-7 and HEA-7 groups, no differences were observed in the dominant phyla (Supplementary Figure S3; Supplementary Table S5). The dominant genera were Oscillospiraceae UCG-005 (22.16 ± 0.75% vs. 20.81 ± 0.38%) and Rikenellaceae-RC9-gut-group (10.3 ± 0.7% vs. 8.47 ± 0.46%), followed by Lachnospira (5.86 ± 0.27% vs. 5.66 ± 0.52%), Oscillospiraceae UCG-010 (4.55 ± 0.46% vs. 6.07 ± 0.59%), Eubacterium-coprostanoligenes-group (4.99 ± 0.36% vs. 5.27 ± 0.2%), Christensenellaceae-R-7-group (3.38 ± 0.14% vs. 3.52 ± 0.2%), Ruminococcus (4.01 ± 0.25% vs. 4.14 ± 0.15%), Monoglobus (3.04 ± 0.16% vs. 3.15 ± 0.16%), Alistipes (2.83 ± 0.19% vs. 3.09 ± 0.15%), and Firmicutes (2.24 ± 0.13% vs. 2.15 ± 0.13%) (Figure 4D). Compared with the HEA-7 group, the relative abundance of Rikenellaceae-RC9-gut-group increased in the KET-7 group (0.05 < p < 0.1), whereas the relative abundance of Oscillospiraceae UCG-010 significantly decreased (p < 0.05).

The top 10 differential species in terms of LDA analysis at the genus level are shown in Figure 5A. The relative abundances of Agathobacter, Prevotella-9, Desulfotomaculum, Eubacterium, Acetitomaculum, Lactobacillus, and SP3-e08 were significantly higher in the KET-7 group than in the HEA-7 group, and the relative abundances of Enterococcus, Lachnospiraceae-AC2044-group, and Lachnospiraceae-NK4A136-group were significantly higher in the HEA-7 group than in the KET-7 group (Supplementary Figure S4).

The top 10 differential genera were subjected to ROC analysis to screen for potential warning biomarkers of ketosis. The AUC of Lactobacillus and Prevotella-9 were 0.88 and 0.85, respectively (Figure 5B). The association between bile acids and differential genera in feces was investigated using Spearman’s correlation analysis (Figure 5C). A strong correlation was observed between the different fecal genera and bile acids. For example, the fecal differential genera Prevotella-9 and Lactobacillus were positively correlated with 12-KDCA (r = 0.55, p = 0.012 and r = 0.69, p = 0.001, respectively), β-MCA (r = 0.55, p = 0.012 and r = 0.68, p = 0.001, respectively), γ-MCA (r = 0.53, p = 0.016 and r = 0.67, p = 0.002, respectively), iso-DCA (r = 0.35, p = 0.12 and r = 0.58, p = 0.008, respectively), and His-CDCA (r = 0.37, p = 0.1 and r = 0.44, p = 0.055, respectively). In addition, Prevotella-9 and Lactobacillus showed significant positive correlations with UnconPBA (r = 0.51, p = 0.02; and r = 0.58, p = 0.007, respectively) and Un/conPBA (r = 0.56, p = 0.01; and r = 0.52, p = 0.02, respectively). Therefore, alterations in bile acids may be closely related to specific microbial alterations. Therefore, Lactobacillus and Prevotella-9 may be potential warning biomarkers in ketotic cows 7 days before calving.

Abnormalities in the metabolism of bile acids in the feces were identified. Plasma bile acid levels were quantified using targeted metabolomics to validate the relationship between bile acid metabolism and ketosis. A total of 23 bile acids were identified in the plasma, of which the CDCA, LCA, NorDCA, Iso-DCA, and 12-KLCA levels were significantly different in KET-7 and HEA-7 groups (Figure 6A). The plasma bile acids in the KET-7 group were predominantly conPBA (59.75%), followed by UnconPBA (19.57%). In addition, the variation in differential bile acids in the plasma, consistent with the feces, was primarily derived from UnconSBA. Finally, the KET-7 group showed a significant difference in UnconSBA compared to the HEA-7 group (Supplementary Figure S5). Analysis using random forest plots and ROC showed that lithocholic acid had the highest Gini index and AUC values (Figures 6B,C; Supplementary Table S6). Spearman’s correlation analyses were performed on differential bile acids in the plasma and feces using energy-related indices to investigate the association between bile acids and plasma energy-related indices (Figure 6D). Significant correlations were observed between bile acid levels and energy-related indices. For example, plasma LCA showed a significant negative correlation with NEFA (r = −0.45 and p = 0.047) and BHB (r = −0.48, p = 0.03). Plasma NEFA was significantly negatively correlated with NorDCA (r = −0.61, p = 0.005) and 12-KLCA (r = −0.63, p = 0.004). Therefore, changes in plasma bile acid levels may be closely related to the development of ketosis. In the future, a larger sample size should be used to verify the potential of LCA to warn of ketosis.

Bile acid levels were significantly lower in the plasma and higher in the feces of the KET-7 group than in those of the HEA-7 group (Figures 2B, 6A). Eleven shared bile acids were detected in plasma and feces: CA, CDCA, DCA, LCA, GDCA, TCDCA, Iso-DCA, TLCA, GCDCA, GCA, and TDCA. These shared bile acids were standardized and the following was observed: LCA and CDCA were significantly lower in the plasma of the KET-7 group than in that of the HEA-7 group, whereas no differences were observed in the feces; TLCA was significantly lower in the feces of the KET-7 group than in that of the HEA-7 group, whereas no difference was observed in the plasma; and iso-DCA was significantly lower in the plasma and significantly higher in feces of the KET-7 group than in those of the HEA-7 group (Supplementary Figure S6). These results suggest the presence of shared bile acids in the feces and plasma. Secondary bile acids may play a crucial role and highlight a specific link between gut microbiota and the development of ketosis.