Differential rumen and hindgut microbiome and metabolome in Holstein female calves with divergent feed efficiency

Abstract

Background

Significant environmental problems have challenged animal agriculture, improving feed efficiency in animals has become a vital research direction for sustainable agriculture. Bacteria play a critical role in the feed efficiency of animals. However, our current understanding of bacteria communities in the gastrointestinal tract of high-feed efficiency animals and their metabolic mechanisms remains unclear.

Results

Twenty Holstein female calves were used in this multi-omics study that integrated metagenomic and metabolomic analyses of 20 Holstein female calves to investigate feed efficiency, as measured by residual feed intake (RFI). From an initial cohort of 84 calves, the 10 with the highest RFI (HRFI, low efficiency) and the 10 with the lowest RFI (LRFI, high efficiency) were selected at 84 days of age. Rumen fluid, feces, and serum samples from these calves were collected for subsequent analyses. We found that LRFI calves harbored rumen and fecal microbiomes with significantly different community structures and co-occurrence networks compared to HRFI calves. Multi-omics integration identified robust microbial and metabolite biomarkers discriminating RFI groups. These microbiomes were functionally linked to differential nutrient utilization, LRFI calves were characterized by enhanced starch and protein digestibility coupled with propionate-oriented fermentation, associated with key species like Erysipelotrichaceae_bacterium and Hungatella_sp. Conversely, HRFI calves showed higher fat digestibility and acetate production. Notably, serum glutamate was enriched in LRFI calves despite lower intake, correlating with potential microbial metabolites (ribitol, taurine). Subsequent validation confirmed that glutamate supplementation in mice improved nitrogen metabolism and gut barrier function.

Conclusions

In summary, this multi-omics study reveals that high feed efficiency in calves is associated with distinct microbial ecosystems characterized by functions such as starch degradation and propionate production, where glutamate metabolism serves as a central node.

Download video file ^{(40.7MB, mp4)}

Video Abstract

Supplementary Information

The online version contains supplementary material available at https://doi.org/10.1186/s40168-026-02446-1.

Introduction

Ruminant farming is indispensable to global livestock production, with ~ 60% of feed derived from lignocellulosic biomass (e.g., grasses, crop residues, and leafy forages) that remains non-edible for human consumption [1]. Through microbial fermentation, ruminants convert these underutilized plant materials into animal products, thereby enhancing food security via agricultural waste valorization and alleviating land-use pressures associated with conventional feed production. However, ruminants display inherently lower feed conversion efficiency compared to monogastric animals, coupled with elevated emissions of reactive nitrogen and methane [1, 2]. Consequently, advancing feed efficiency in ruminants has become a critical priority to reconcile productivity, environmental sustainability, and economic viability in modern livestock systems.

Bacteria are crucial for enhancing feed efficiency because they are involved in both feed digestion and the regulation of gastrointestinal homeostasis. Recent studies have highlighted the correlation between variations in rumen bacteria in dairy cows and their production characteristics, such as feed efficiency [3, 4], methane production [5, 6], and milk composition [7]. Elolimy et al. (2020) reported that due to differences in the composition of gut microbes, calves with high-feed efficiency had greater utilization of energy, vitamins and amino acids compared to calves with low feed efficiency [8]. Bonilha et al. (2017) found that modifications in rumen bacterial composition in cows with high-feed efficiency contributed to improvements in the digestibility of dry matter, organic matter, and neutral detergent fiber, ultimately enhancing feed efficiency [9].

RFI serves as a metric for assessing feed efficiency, which is determined by calculating the difference between actual and predicted feed intake, which represents a phenotypic characteristic with moderate heritability [10]. The predicted feed intake is computed based on the dietary intake needed to meet the maintenance and production requirements of animals; therefore, positive RFI values indicating low feed efficiency and negative RFI values indicating high-feed efficiency [11–13]. RFI has been adopted as an indicator within genetic selection by several countries [14, 15].

The rationale for investigating RFI in Holstein female calves stems from two critical dimensions: (1) while calf rearing represents a low-profit phase in dairy operations, growth performance during this period exerts a pronounced influence on milk yield during the subsequent 305-day lactation period [16]. Enhancing feed efficiency during this critical window is therefore essential for optimizing growth trajectories and long-term productive outcomes. (2) RFI demonstrates both moderate heritability and sustainable phenotypic expression [17, 18], with LRFI calves maintaining superior milk efficiency throughout lactation [19]. Therefore, assessing RFI in early-life female calves provides dual benefits: reducing nursery phase costs while augmenting future milk yield, production efficiency, and genetic advancements through accelerated selection pressure. Notably, the post-weaning period represents a critical developmental window during which the gastrointestinal microbiota of calves establishes relative stability and becomes increasingly resistant to perturbation [20–23]. Therefore, understanding the microbial features associated with high feed efficiency during this phase may provide actionable insights for early-life intervention. Existing feed efficiency studies have predominantly focused on ruminal microbiota [13, 24–26], leaving gaps in understanding the role of hindgut microbial communities in nutrient utilization dynamics.

Consequently, for the reasons detailed above, in this study, to align the research more closely with actual farm production processes and ensure that the findings can be applied to farms in the future, we selected 80 Holstein female calves for investigation and examined how rumen and hindgut bacteria and their metabolites influence feed efficiency via bacteria, metabolites and host interactions.

Results

Compositional profiles of the rumen and fecal bacteria

Following the completion of the experiment, we calculated the RFI values for all the calves involved. Subsequently, we selected the calf samples exhibiting the highest and lowest RFI values for detailed analysis, in order to better understand the factors associated with efficiency of feed utilization. A total of 20 rumen fluid and 20 fecal samples were collected from 20 dairy calves (Fig. 1A). To characterize the microbial communities in the gastrointestinal tract, we analyzed the domain-level composition of rumen and fecal samples. In rumen samples, the microbial community was dominated by Bacteria, accounting for 95.28% of the total sequences. Viruses represented 2.77%, followed by Archaea (1.50%), Eukaryota (0.41%), and unclassified sequences (0.05%). In fecal samples, Bacteria comprised 88.77% of the sequences. Viruses accounted for 9.61%, while Archaea represented 1.52%. Eukaryota and unclassified sequences constituted minor fractions of 0.09% and 0.01%, respectively (Fig. S1 and Table S1). The taxonomic analysis results of bacteria revealed the presence of top 20 genera in the rumen fluid and feces of calves (Fig. 1B). Our analysis revealed that within fecal samples, the mean relative abundance of Rikenellaceae_RC9_gut_group bacteria in the HRFI group was 41.02%. Conversely, in rumen fluid, the average proportion of Erysipelotrichaceae_UCG-002 in the LRFI group was notably higher, reaching 68.56% (Fig. 1B). Significant differences were observed between HRFI and LRFI groups on day 84 in the beta diversity of the calves' rumen fluid (corrected P < 0.01) and fecal (corrected P < 0.01) bacterial communities (Fig. 1C) using principal coordinates analysis (ANOSIM was used to test for differences between the groups). This was further confirmed by the significant differences in the Shannon index (rumen fluid: corrected P < 0.01; fecal: corrected P < 0.01) (Fig. 1C).

Fig. 1.

Experiment flow, fecal and rumen microbial compositional profiles, and phenotypic differences of 20 screened calves. A The study work flow including animals, types of samples collected, sample collection time points, testing methods, number of calves used in different test. B Microbial composition genus level in rumen and feces of HRFI and LRFI. C Bacterial compositional profiles of HRFI and LRFI calves based on genus visualized using principal-coordinate analysis (PCoA) and comparison of alpha diversity indices between HRFI (n = 10) and LRFI (n = 10) calves. D Difference of residual feed intake (RFI) between HRFI (n = 10) and LRFI (n = 10) calves. E BW and ADG at different days or weeks in the HRFI (n = 10) or LRFI (n = 10) calves. F Solid dry matter intake (SDMI) and liquid dry matter intake (LDMI) at different weeks in the HRFI (n = 10) or LRFI (n = 10) calves. G Difference of diarrhea frequency between HRFI (n = 10) and LRFI (n = 10) at different weeks of age. Diarrhea Frequency = ∑Incidence of diarrhea in experimental calves/(Calf numbers × days of trial) × 100%. (P < 0.05, denoted by *; P < 0.01, denoted by **, P < 0.001, denoted by ***)

The difference in the residual feed intake coefficient between the HRFI (0.177 ± 0.05 kg DMI/d) and LRFI (−0.117 ± 0.05 kg DMI/d) groups was 0.294 kg dry matter intake (DMI) per day (nominal P < 0.01) (Fig. 1D and Fig. S2). The growth performance results, which are presented in Fig. 1E, indicate that there were no significant differences in calves body weight (BW) (corrected P = 0.94) and average daily gain (ADG) (corrected P = 0.63) between the HRFI and LRFI groups throughout the trial. The HRFI group exhibited a higher dry matter intake for solid feed (SDMI) than the LRFI group throughout the trial period (corrected P < 0.01). Under experimental induction feeding protocols, no significant difference in liquid dry matter intake (LDMI) was observed (corrected P = 0.28) (Fig.1F). Diarrhea frequency was quantified across the experimental period, revealing no measurable variation between groups (corrected P = 0.54) (Fig. 1G). These results collectively indicate that the observed dry matter intake discrepancies were unrelated to diarrheal events.

Bacterial interaction network

We constructed a bacterial interaction network for the top 100 most abundant bacteria at the genus level in both groups to elucidate the associations among these dominant bacteria. Compared with HRFI, the bacteria in the LRFI group displayed lower average degree (HRFI vs. LRFI = 6.59 vs. 4.68) in rumen (Fig. 2A), however, the opposite was true in the feces (Fig. 2B), where LRFI exhibited a higher average degree (HRFI vs. LRFI = 7.60 vs. 10.73). In addition, the degrees of each genus are listed in Table S2 and Table S3. Combining the results of differentially abundant bacteria at the genus level, we found that in the rumen, Erysipelotrichaceae_UCG-002 was enriched in LRFI (corrected P < 0.01) (Fig. 2C) and was a genus with a higher degree (Table S2); Olsenella, Eubacterium_nodatum_group, and Erysipelotrichaceae_UCG-009 were enriched in HRFI (corrected P < 0.05) and were genera with higher degrees; whereas in the feces, UCG-005 and Marvinbryantia (Fig. 2D) were genera with higher degrees (Table S3), enriched in HRFI (corrected P = 0.01) and LRFI (corrected P = 0.05), respectively.

Fig. 2.

Interaction network based on the top 100 bacterial genus. A The interaction network of HRFI and LRFI genus in rumen fluid. B The interaction network of HRFI and LRFI genus in feces. The size and color of the nodes represent the magnitude of the degree: the larger the size and the redder the color, the higher the degree. The color of the edges indicates correlation, blue for negative correlation and red for positive correlation, while thicker edges signify greater weight. Only P < 0.05 and the correlation coefficient >|0.5| for the top 100 most abundant genus are retained. Wilcoxon rank-sum test revealed the most differentially microbial compositional profiles in the rumen fluid (C) and feces (D) microbiome at the genus level. The bacteria marked in red on the Y-axis are those enriched and with higher degree in the LRFI group, while those marked in blue are enriched and with higher degree in the HRFI group

Differential metabolites analysis

We identified 2,280 and 2,172 metabolites in the rumen and feces of 20 calves, respectively (Table S4), and further analyzed the differentially expressed metabolites between low and high RFI calves in these two digestive sites.

The PLS-DA model for the rumen metabolome achieved a cumulative Q² of 0.556 and R²X of 0.394 (Fig. S4A), In contrast, the fecal metabolome model showed a lower predictive ability (R²X = 0.279; Q² = 0.418). PCoA showed a significant difference in rumen metabolite composition between HRFI and LRFI calves (R = 0.41, corrected P < 0.01), while the hindgut metabolite composition exhibited a trend toward separation (R = 0.11, corrected P = 0.06) (Fig. S3). A total of 68 differential metabolites were found (Variable Importance in Projection, VIP > 1; corrected P < 0.05) in the rumen fluid; among them, 57 differential metabolites were downregulated and 11 differentially expressed metabolites were upregulated (Fig. S4B and C) in the LRFI group. A total of 48 differential metabolites were found (VIP > 1; corrected P < 0.05) in the feces, of which 27 were downregulated and 21 were upregulated (Fig. S4B and C) in the LRFI group.

Integrated analysis of metabolites and microbes

To investigate the potential of rumen and gut bacteria and metabolomic parameters as diagnostic markers, we used DIABLO to integrate 16S rRNA gene sequencing data and metagenomic data with metabolomics data at the genus and species levels, respectively, to identify multi-omics features (Fig. 3). By discovering correlated and predictive features across datasets, we distinguished high and low RFI phenotypes. In the rumen, we identified 5 signature genera (AUC = 0.94; Accuracy 98.1%) and 5 signature metabolites (AUC = 0.96; Accuracy 97.6%) (Fig. 3A and Table S5) and analyzed the relationships between different omics features. Among them, only Erysipelotrichaceae_UCG-002 was significantly positively correlated with metabolite features such as ribitol (cutoff r = 0.7) (Fig. 3E and Table S5). At the species level, we identified 16 signature species (AUC = 0.97; Accuracy 98.4%) and 6 signature metabolites (AUC = 0.92; Accuracy 97.7%) (Fig. 3C and Table S6), where two metabolites, N-Carboxyethyl-gamma-aminobutyric acid and ribitol, were found to be positively correlated with Erysipelotrichaceae_bacterium and the other 8 signature species and negatively correlated with 4 signature species (cutoff r = 0.7) (Fig. 3E and Table S6).

Fig. 3.

Integrated analysis of microbiome and metabolome. Loading plots from multiblock sPLS-DA applied to A rumen metabolome and genus-level microbiome, B fecal metabolome and genus-level microbiome, C rumen metabolome and species-level microbiome, and D fecal metabolome and species-level microbiome data to distinguish RFI groups. Colors indicate the group in which the median is maximum for each feature. AUC based on multiblock sPLS-DA performed on the study for the data set after 2 components. The function calculates the AUC for HRFI vs. LRFI. E circosPlot of the most significant variables from each dataset and their pairwise correlations. Variables are represented on the side of the circosPlot, where colors indicate the type of data (metabolome and microbiome), and external lines display the measured levels with respect to each outcome category (HRFI and LRFI). In the center, orange lines signify a positive correlation between the linked variables, and gray lines a negative one

In feces, we also identified 5 signature genera (AUC = 0.90; Accuracy 95.2%) and 5 signature metabolites (AUC = 0.93; Accuracy 95.8%) (Fig. 3B and Table S7), where only Rikenellaceae_RC9_gut_group was significantly positively correlated with 5 signature metabolites such as PE.16.1.0.0.3 (cutoff r = 0.7). At the species level, we identified 23 signature species (AUC = 0.96; Accuracy 94.6%) and 5 signature metabolites (AUC = 0.96; Accuracy 94.9%) (Fig. 3D and Table S8), among which 14 positive correlations and 35 negative correlations were found (cutoff r = 0.7). Pathway mapping of the aforementioned signature metabolites (Table S4) revealed that in the rumen (Fig. S5A), metabolites enriched in the LRFI group primarily belonged to Amino acids, peptides, and analogues, Carbohydrates and carbohydrate conjugates, and Carboxylic acid derivatives, with the broader categories being Organic acids and derivatives and Organic oxygen compounds. No related pathways were detected for metabolites enriched in the HRFI group. In feces (Fig. S5B), metabolites enriched in the LRFI group primarily belonged to Carbonyl compounds, with the broader category being Organic oxygen compounds. In contrast, metabolites enriched in the HRFI group primarily belonged to Glycerophosphoethanolamines and Monoterpenoids, with the broader category being Lipids and lipid-like molecules.

Procrustes analysis revealed a significant association between the microbiome and metabolome in the rumen (M² = 0.800, R = 0.447, P = 0.048), indicating that the two datasets share a coherent structure. In contrast, no significant concordance was observed in feces (M² = 0.905, R = 0.308, P = 0.485), suggesting a weaker relationship in the hindgut (Fig. S6).

Differences in nutrient utilization and volatile fatty acids (VFAs) production

Considering that the metabolites generated by bacteria are directly associated with their substrate interactions, we proceeded to examine whether there were significant differences in nutrient digestion and utilization between the two groups. The results indicate that, relative to the HRFI group, the LRFI group of calves exhibits higher digestion rates for dry matter (nominal P < 0.01), protein (nominal P < 0.01), and starch (nominal P = 0.02), whereas the HRFI calves display a higher digestion rate for fat (nominal P < 0.01) (Fig. 4A).

Fig. 4.

Differences in nutrient utilization and the concentration of VFAs in the gastrointestinal tract among Holstein female calves with different RFI and their relationship with marker bacteria. A Difference of nutrient digestibility between HRFI (n = 10) and LRFI (n = 10) calves on 84 days of age. B Concentration of rumen and fecal VFAs in HRFI (n = 10) or LRFI (n = 10) Holstein female calves on day 7, day 35 and day 84. (P < 0.05, denoted by *; P < 0.01, denoted by **). C Spearman correlation analysis of the relationship between rumen and hindgut species in Holstein female calves and different nutrient utilization rates and VFAs (n = 20). The bacteria names on the X-axis in the figure are in red for propionic acid-related bacteria and in blue for acetic acid-related bacteria

On the other hand, the results of rumen VFAs analysis reveal that at 84 days of age, LRFI calves exhibit lower acetate concentration (corrected P < 0.01) but higher propionate concentration than HRFI calves (corrected P < 0.01). Although no difference in butyrate concentration was observed at 84 days of age, the rumen of HRFI calves exhibited a higher butyrate concentration at 7 days of age (corrected P < 0.01). In the results of VFAs analysis in the hindgut, we found that at 7 and 84 (corrected P < 0.05) days of age, the proportion of propionate in the LRFI group was significantly higher than that in the HRFI group. For butyric acid, significantly higher concentration was observed in the LRFI group compared to the HRFI group only at 7 and 35 days of age (corrected P < 0.01) (Fig. 4B).

Further correlating the characteristic species with nutrient digestion and gastrointestinal fermentation, we found that Hungatella_sp. and Erysipelotrichaceae_bacterium showed a significant positive correlation with propionate concentration (nominal P < 0.05). Parvibacter_caecicola, Acetatifactor_sp. DSM 110981, and Candidatus_Choladocola_avistercoris showed significant positive correlations with acetate concentration (nominal P < 0.05). On the other hand, in the hindgut, we observed that Halobacillus_fulvus and Synergistales_bacterium had a significant positive correlation with acetate production (nominal P < 0.05), while Acidipropionibacterium_sp. and Bacillus_sp. F19 not only showed a significant positive correlation with propionate production but also with starch (nominal P < 0.05) (Fig. 4C). The above results suggest that the aforementioned bacterial species and genera may be important contributors to, or indicators of, the observed differences in nutrient digestion and fermentation.

Host serum metabolome analysis

In order to further investigate how bacteria and metabolites in the rumen and hind-gut affect host metabolism, we conducted a study on host serum metabolism of 20 screened calves. According to the results depicted in the volcano map, 47 differential metabolites were identified in the serum (VIP > 1; corrected P < 0.05). Among these, 25 exhibited downregulated expression, whereas 22 showed upregulated expression (Fig. 5A). Metabolic pathway analysis of these 47 significantly distinct serum metabolites revealed the enrichment of one pathway related to D-glutamine and D-glutamate metabolism. (-ln P > 3, pathway impact > 0.1, Fig. 5B). In the analysis of pathways involving differential metabolite enrichment, 31 pathways associated with L-glutamate enrichment were identified. Nodes related to L-glutamate constituted 81.6% of the total nodes (Fig. 5D). The AUC value for L-glutamate was 0.84 (Fig. 5C).

Fig. 5.

Host serum metabolite analysis. A Volcano plot of differential metabolites (P < 0.05; VIP > 1; FC > 1) of host serum between HRFI (n = 10) and LRFI (n = 10). B Pathway enrichment analysis performed using the significantly different serum metabolites between LRFI (n = 10) and HRFI (n = 10) calves. C Ranking of serum metabolites with different AUC values and AUC values of L-Glutamate. D Sankey diagram shows the relationship between different hierarchical pathways and the enrichment of differential metabolites

Functional gene analysis and pathway construction

Based on the above results, in order to explore the production pathway of related metabolites and their relationship with glutamate, we mainly focused on amino acid, energy, lipid, and carbohydrate metabolisms related to feed efficiency. The functions of rumen and fecal bacteria were elucidated using Kyoto Encyclopedia of Genes and Genomes (KEGG) profiles.

In the KEGG profiles, 20 differential third-level pathways were detected between LRFI and HRFI in the rumen (Fig. S7B; LDA > 3). Calves with LRFI showed elevated expression levels of functional genes linked to starch and sucrose metabolism pathways, particularly in the phosphorylation and degradation of sucrose, cellobiose, maltose, and trehalose. Additionally, the results revealed that HRFI calves exhibited higher expression levels of methane-producing enzymes and genes than LRFI calves. In contrast, enzymes and genes associated with acetyl-coenzyme production and fatty acid synthesis exhibited higher expression levels in the HRFI group than in the LRFI group (corrected P < 0.05; Fig. 6A; Table S9). At the same time, the functional pathway observation showed that the rumen of LRFI calves could produce propionate through propanoate metabolism, which was consistent with the results of VFAs on day 84. Specifically, ribitol, generated through pentose and glucuronate interconversion, was significantly enriched in the rumen of the LRFI group (Fig. S4C; VIP > 1; corrected P < 0.05).

Fig. 6.

The rumen and hind-gut function pathway. A Rumen-enriched metabolic pathways involved in methane metabolism, pentose and glucuronate interconversion biosynthesis, Valine, leucine and isoleucine biosynthesis, oxidative phosphorylation, TCA cycle, taurine and hypotaurine metabolism, primary bile acid biosynthesis and secondary bile acid. B Hindgut-enriched metabolic pathways involved in pyruvate metabolism, methane metabolism, fatty acid biosynthesis, pentose and glucuronate interconversions, and starch and sucrose metabolism. C Linear regression analysis was performed between ribitol (n = 10/group), taurine (n = 10/group), and glutamate (n = 10/group). Red module indicates genes, enzymes, or metabolites that are enriched in LRFI. Blue module indicates genes, enzymes, or metabolites that are enriched in HRFI. Those enriched in the host serum of LRFI are depicted in green module. Gray module indicates no difference between the two groups

In the KEGG profiles, 28 differential third-level pathways were identified in the KEGG profiles of LRFI and HRFI groups in the feces (Fig. S7A; LDA > 3). The abundance of genes encoding enzymes involved in methane metabolism; propanoate metabolism; pentose and glucuronate interconversions; valine, leucine, and isoleucine biosynthesis; and taurine and hypotaurine metabolism was significantly higher in the feces of LRFI calves than in HRFI calves (corrected P < 0.05; Fig. 6B; Table S9). The functional pathway observation showed that the hindgut of LRFI calves could produce propionate through propanoate metabolism, which was consistent with the results of VFAs on day 84. Specifically, taurine, generated through taurine and hypotaurine metabolism, was significantly enriched in the hindgut of the LRFI group (Fig. S4C; VIP > 1; corrected P < 0.05). The abundance of genes encoding enzymes involved in the TCA cycle was significantly higher in the feces of HRFI calves than in LRFI calves (P < 0.05; Fig. 6B; Table S10).

Glutamate effectiveness verification

To further investigate whether the key metabolic metabolite, glutamate, detected in the serum was due to feed intake or generated by LRFI calves, we first compared the glutamate intake of two groups of calves (Fig. 7A), finding that the daily glutamate intake in the LRFI group was significantly lower than that in the HRFI group (corrected P < 0.01). Despite the lower intake, the apparent digestible glutamate in the LRFI group was consistent with that in the HRFI group, without significant differences (nominal P = 0.48) (Fig. 7B). Therefore, the difference in calf feed efficiency was not driven by higher glutamate intake. Instead, LRFI calves could achieve a glutamate metabolic state consistent with or even superior to that in HRFI calves with lower glutamate intake. This strongly suggests that glutamate levels may reflect an internal metabolic efficiency result or biomarker, rather than a simple nutritional driver.

Fig. 7.

Mice model verification of glutamate effects. A Comparison of glutamate intake in calves with different RFI. B Comparison of glutamate digestibility in different RFI groups. C Flowchart of the mouse model verification experiment for glutamate effects. D Feeding intake in different treatment groups of mice. E Differences in body length and actual comparison images among different treatment groups of mice. F Comparison of body weight throughout the experiment in different treatment groups of mice. G Serum glutamate and glutamine concentrations in different treated mice. H Analysis of nitrogen metabolism-related indicators in different treatment groups of mice. I Analysis of lipid and energy metabolism indicators in different treatment groups of mice. J Difference in albumin concentration among different treatment groups of mice. K Analysis of intestinal damage indicators in different treatment groups of mice. L Comparison of IGF-1 concentrations in different treatment groups of mice. (P < 0.05, denoted by *; P < 0.01, denoted by **; a, b indicates superscript letters, and their different representations have significant differences)

To directly verify whether glutamate has a functional role in improving feed efficiency-related phenotypes, we designed and completed a mouse control feeding experiment (Fig. 7C). The results showed that, compared to the Na group, glutamate feeding did not significantly promote feed intake (Fig. 7D), but improved trends in body length (Fig. 7E), weight (Fig. 7F) (P = 0.06). Blood biomarker detection revealed that, compared to CON and Na, glutamate feeding significantly increased serum glutamate and glutamine levels (corrected P < 0.05), while blood urea nitrogen (BUN) was significantly reduced (corrected P < 0.05), and total protein (TP) was significantly elevated (corrected P < 0.05) (Fig. 7G and H). However, it had no significant impact on basic energy indicators such as glucose, triglycerides, cholesterol, and albumin (ALB) (Fig. 7I, J). Compared to CON and Na, glutamate feeding significantly reduced diamine oxidase (DAO) levels (corrected P < 0.05) (Fig. 7K). Compared to CON, insulin-like growth factor-1 (IGF-1) was unaffected by glutamate feeding, but chloride sodium reduced serum IGF-1 concentration (corrected P < 0.01). In summary, in calves, high-efficiency individuals were not efficient because they consumed more glutamate; instead, it was their intrinsically superior metabolic system, such as the potential microbial and ribitol or taurine network discovered in this study, which more efficiently utilized dietary nitrogen sources and maintained higher glutamate metabolic levels. Glutamate is the output marker and key functional node of this efficient system. The mouse experiment proved that elevating glutamate levels could directly produce physiological effects associated with the efficient phenotype. This validated that glutamate is not a passive marker in the entire pathway, but has a positive biological function.

Analysis of the pathways and functions

As observed in the overarching experiment (as depicted in Fig. 8), ribitol, which was enriched in the LRFI group, exhibited a positive correlation with the propionic-acid-associated bacteria within the rumen, such as Hungatella_sp. and Erysipelotrichaceae_bacterium. At the same time, ribitol was also shown negative correlations with acetic-acid-associated bacteria. Ribitol undergoes metabolic processes via the pentose phosphate pathway (ko00030) and pentose-glucuronate interconversion (ko00040), ultimately yielding 5-phosphoribose-1-pyrophosphate (PRPP). Consequently, PRPP contributes to L-glutamate production via histidine metabolism (ko00340). Analysis of the metabolome data from all calves revealed a statistically significant positive (R² = 0.1994; nominal P = 0.03) correlation between ribitol and glutamate (Fig. 6C).

Fig. 8.

The characteristic total digestive microbiome and associated metabolic pathways of high feed efficiency youngfemale calves. The figure shows the characteristic microbial species and metabolite interaction modules of DIABLO analysis, as well as the characteristic metabolites, including ribitol and taurine, which can produce glutamate through specific metabolic pathways. It also illustrates the functions of glutamate after entering the bloodstream, related to protein and nitrogen metabolism

Within the hindgut, the propionic acid-associated bacteria, such as Acidipropionibacterium_sp. and Bacillus_sp._F19, exhibited negative correlations with lipids and lipid-like molecules. Formamide catalyzes the generation of L-glutamate through nitrogen metabolism (ko00910), whereas taurine, which was enriched in the LRFI group, promotes L-glutamate production through taurine and hypotaurine metabolism (ko00430). Analysis of the metabolome data from all calves revealed a statistically significant positive (R² = 0.1268; nominal P = 0.04) correlation between taurine and glutamate (Fig. 6C).

Upon entering the bloodstream, L-glutamate participates in various metabolic pathways, such as the amino acid biosynthesis, nitrogen metabolism, arginine biosynthesis, D-amino acid metabolism, and protein digestion and absorption in the host (Fig. 5D). This result is consistent with the results of the mouse experiment. Through gastric gavage of glutamate, we also observed a significant increase in total protein concentration and a decrease in BUN in the mouse serum (Fig. 8).

Discussion

In dairy production systems, Holstein female calf rearing constitutes a non-profitable yet critical growth phase. The growth status during this period determines 305-day milk yield in subsequent lactation period [16], underscoring the importance of optimizing feed efficiency and nutrient utilization for calf development. Gut microbiota plays pivotal roles in host nutrition by delivering essential nutrients and enhancing feed conversion efficiency [27, 28]. The natural colonization of calf gastrointestinal microbiota progresses through three distinct ontogenetic stages: an initial exponential phase (0–1 week), a transitional phase (1–12 weeks), and a stable phase (post-12 weeks) [20–23]. This phased developmental pattern presents strategic opportunities for microbiota manipulation to improve nutrient absorption in calves.

Through integrated network analysis, we identified key bacterial taxa with differential abundance in the LRFI individual's rumen and hindgut. Erysipelotrichaceae UCG-002, enriched in the LRFI group, not only has a higher abundance but also shows greater connectivity in the network, suggesting it may occupy a central position in this efficiency-related microbial network. Existing studies indicate that this bacterium is involved in pyruvate, lactic acid, and glucose metabolism [29], so it may be associated with starch digestion and propionate production through interspecies interactions [30].

Similarly, in the fecal network, the Rikenellaceae_RC9_gut_group enriched in the HRFI group and the Marvinbryantia enriched in the LRFI group both exhibit high connectivity. Previous studies have reported that Rikenellaceae_RC9_gut_group is positively correlated with fat deposition [31], while Marvinbryantia is related to the utilization of soluble fibers such as pectin [32]. Therefore, they may respectively serve as the core nodes in the fat-fermentation network and carbohydrate-fermentation network in the feces of HRFI and LRFI calves. The energy cost difference between lean and fat tissue deposition means that fat accumulation usually requires more metabolic investment [33], which may be related to the lower growth efficiency under isocaloric conditions.

Interestingly, the co-occurrence networks revealed distinct ecological patterns between the two gastrointestinal sites. In feces, positive interactions predominated in both HRFI and LRFI groups, suggesting that cooperative relationships are a general feature of the hindgut ecosystem [34]. In the rumen, however, group-specific differences emerged, while HRFI calves maintained a positive-to-negative ratio favoring cooperation, LRFI calves exhibited a near-equal balance, potentially reflecting a more competitive microbial environment optimized for efficient nutrient utilization [35].

The PLS-DA model for the rumen metabolome demonstrated good predictive ability (cumulative Q² = 0.556, R²X = 0.394), indicating that the identified metabolic features reliably distinguish between HRFI and LRFI calves. In metabolomics studies, a Q² value above 0.5 is generally considered indicative of robust model performance [36]. The significant separation observed in PCoA (R = 0.41, corrected P < 0.01) further supports that rumen metabolic profiles are distinctly associated with feed efficiency. In contrast, the fecal metabolome model showed lower predictive ability (Q² = 0.418, R²X = 0.279), and PCoA revealed only a trend toward separation (R = 0.11, corrected P = 0.06). This is expected, as fecal metabolome is inherently more complex and variable than rumen fluid, integrating contributions from host digestion and microbial metabolism along the entire intestinal tract [37]. Additionally, the hindgut microbiota may have a more indirect influence on feed efficiency compared to the rumen, the primary site of nutrient fermentation in ruminants. Collectively, these results suggest that rumen metabolic profiles are more directly linked to feed efficiency, while fecal metabolome changes, though subtler, still provide valuable insights into hindgut metabolic adaptations associated with this phenotype.

To further validate, this study used the DIABLO framework to integrate microbiome and metabolome data, effectively distinguishing between HRFI and LRFI phenotypes. The effectiveness of this strategy is consistent with previous studies, such as a DIABLO-based metabolomics classifier that demonstrated extremely high discrimination ability in colorectal cancer research (AUC = 0.98) [38], supporting the idea that microbial metabolites may serve as direct links between microbiome composition and host phenotype.

This study also found that Erysipelotrichaceae_UCG-002 is positively correlated with ribitol, suggesting this key bacterium may participate in connecting carbohydrate metabolism with the pentose pathway involved in phosphoribosyl pyrophosphate (PRPP) production [39], which is a critical precursor for histidine synthesis and purine metabolism, ultimately connected to glutamate and nitrogen metabolism networks. In the hindgut, Rikenellaceae_RC9_gut_group is positively correlated with multiple lipid metabolites, directly confirming the core association between this microbial community and fat digestion metabolism.

At the species level, correlation analysis showed that the enrichments of Hungatella sp. and Erysipelotrichaceae_bacterium in the rumen of LRFI calves are positively associated with propionic acid concentration, suggesting their potential involvement in propionate production. Although there is a lack of direct literature evidence, at the genus level (Erysipelotrichaceae_UCG-002), this bacterium has been confirmed in several ruminant studies to be positively correlated with diet starch content [29], rumen microbial stability restoration [40, 41], and starch utilization, as well as propionic acid and lactic acid concentrations [42, 43]. Therefore, the higher propionic acid concentration in the rumen of LRFI calves may be partly related to the activity of these bacteria. Meanwhile, in the rumen of HRFI calves, bacteria related to acetic acid, such as Acetatifactor sp., were enriched, consistent with the presence of an acetate-oriented microbial metabolic profile in their digestive tract. Acetic acid is not only a primary energy substrate but also a precursor for lipid synthesis [44]. In contrast, propionate is the predominant gluconeogenic precursor in ruminants; it is efficiently converted to glucose in the liver, thereby providing a more direct and energetically favorable substrate for growth compared to acetate [45, 46]. The high acetate environment in the hindgut of HRFI calves may promote intestinal fat digestion and absorption [47], which could potentially result in more fat being used for body fat deposition rather than efficient growth [33], which aligns with the HRFI phenotype.

In the hindgut, Acidipropionibacterium sp. and Bacillus sp. F19 are positively correlated with starch digestibility and propionic acid production. Many Bacillus species are efficient producers of extracellular amylases [48], while Acidipropionibacterium acidipropionici is a recognized propionic acid producer [49, 50], and its supplement can increase host intestinal propionic acid levels and improve energy metabolism [51]. This suggests that the microbial community in the hindgut of LRFI calves may be organized in a manner that facilitates efficient cooperation in starch digestion and propionate production. It is noteworthy that some strains, such as Parvibacter caecicola, have not yet been reported to be directly related to acetate metabolism. Currently, our understanding of the specific metabolic functions of many intestinal bacteria remains limited [52], and the results of this study provide new clues for a detailed exploration of their pure culture characteristics, metabolic pathways, and their roles in nutrient allocation in ruminants.

The stronger association between microbiome and metabolome in the rumen compared to feces is consistent with the rumen’s primary role in microbial fermentation, where microbial activity directly drives metabolite production [53]. The fecal metabolome, by contrast, is a composite of host and microbial processes along the entire gut [54, 55], which may dilute the direct microbiome and metabolome linkage.

Regarding the digestibility phenotype, previous studies have reported that low RFI dairy cows have higher dry matter and organic matter digestibility [9, 56, 57], as well as higher protein digestibility [56, 58]. This study observed consistent results in calves, indicating that the relationship between RFI and digestibility may be consistent among ruminants. LRFI calves had lower feed intake, which may reduce rumen outflow rate [59], providing bacteria with more time for digestion. This mechanism is consistent with, and could be one factor explaining, the observed higher dry matter digestibility in LRFI calves [9, 60].

Analysis of the host serum metabolome revealed that glutamate, a metabolite enriched in LRFI, plays a crucial role in most of the differential metabolic pathways. This study uncovered a key paradox: LRFI calves have lower glutamate intake but higher serum glutamate levels. This indicates that differences in feed efficiency stem from the host's internal metabolic efficiency rather than external intake differences. Serum metabolomics confirmed that glutamate is enriched in the LRFI group and active in pathways such as amino acid biosynthesis, nitrogen metabolism, and protein digestion and absorption, providing mechanistic explanations for its central role in maintaining body weight and muscle growth [61]. Furthermore, glutamate contributes to energy production via participation in the tricarboxylic acid cycle [62], and its supplementation has been shown to improve intestinal barrier function in calves [63]. To validate the active functional role of glutamate, we conducted a mouse intervention experiment. Exogenous glutamate supplementation reduced blood urea nitrogen, increased serum total protein, improved nitrogen metabolism, and enhanced gut health, directly confirming its physiological functions as a precursor for protein synthesis and energy substrate for intestinal epithelium. In summary, we propose an integrative mechanistic hypothesis: the advantage of LRFI calves may be underpinned by their overall optimized metabolic network. This network more efficiently directs nitrogen sources toward glutamate-centered biosynthetic pathways and reduces losses by maintaining intestinal health, thus converting less intake into high-efficiency growth. Therefore, glutamate is not only a biomarker of efficient metabolic state but also a functional core driving high nitrogen utilization and organismal maintenance.

Functional analysis shows that in the hindgut of LRFI calves, genes involved in NADH degradation, methanol decomposition, branched-chain amino acid synthesis, and propionic acid metabolism have higher activity. NADH oxidation is the main energy production process, and its more active degradation may be related to more efficient energy metabolism and improved feed utilization [64]. Additionally, in the hindgut of LRFI calves, methane is more readily converted to formic acid, methyl-CoM, and carbon dioxide [65], directing carbon flow toward energy production rather than gas loss. The pathways for branched-chain amino acid synthesis are also more prominent [66], possibly supporting protein synthesis and growth needs. Notably, taurine enriched in the hindgut of LRFI calves can produce glutamate through taurine and hypotaurine metabolism pathways. Taurine plays a role in inhibiting pathogenic microorganisms and supporting intestinal immunity in the early stages [67], but in this study, there was no significant difference in diarrhea incidence between the two groups of calves, suggesting that its role may be more focused on glutamate synthesis rather than direct antimicrobial effects.

Within the rumen, LRFI calves exhibited high expression levels of functional genes associated with starch and sucrose metabolism pathways, specifically in the phosphorylation and degradation of sucrose, cellobiose, maltose, and trehalose, which are key steps in glycolysis and energy acquisition [68]. In contrast, methanogenesis genes in the rumen and hindgut of HRFI calves are more active [65], possibly related to their higher acetate levels, as acetate metabolism produces hydrogen that is an important substrate for methanogenesis [69–72].

Genes associated with acetyl-CoA and fatty acid synthesis are more highly expressed in HRFI calves, consistent with metabolomic and digestibility results, indicating that HRFI calves have a stronger capacity for fat digestion and utilization. Fatty acid synthesis is an energy-consuming process [73], and some of this energy may not be used for growth, further reducing feed efficiency.

In addition, we found an enrichment of ribitol in the LRFI calf rumen, and this metabolite can produce glutamate through the equally enriched pentose/glucuronic acid interconversion pathway. The taurine from the hindgut combined with this indicates that microbial metabolites from the rumen and hindgut may be absorbed and influence the host’s serum metabolic profile, thereby regulating growth and metabolism. We also further investigated the potential propionic acid production mechanism in the rumen and hindgut of LRFI calves. In the rumen, the production of propionate was derived from two primary pathways: (1) the entry of metabolic products of starch and sucrose degradation into propionate metabolism, which is consistent with digestion capacity of marker bacteria in the rumen; and (2) the integration of acetyl-CoA generated via pyruvate metabolism into propionate metabolism, a pathway also observed in the hindgut.

Further supporting a specialized metabolic configuration, we observed a higher number of downregulated than upregulated microbial pathways in the rumen of high-efficiency calves. This pattern suggests that the rumen microbiome of high-efficiency calves undergoes metabolic streamlining, wherein microbial communities focus resources on a few core functions, such as starch degradation and propionate production, while suppressing others [74]. The concurrent downregulation of multiple pathways, such as methane metabolism, fatty acid biosynthesis, aligns with the concept that efficient animals harbor microbiomes optimized for energy harvest rather than functional diversity [75]. This is consistent with previous reports that high-efficiency ruminants exhibit less diverse but more specialized microbial communities with simplified metabolic networks [76].

Overall, this study systematically investigated the rumen and hindgut microbiota of LRFI and HRFI calves, identifying key bacterial taxa implicated in feed efficiency regulation and elucidating their functional roles. High-feed efficiency calves exhibited four distinct metabolic signatures: (1) enhanced starch digestion capacity in the gastrointestinal tract; (2) metabolic reprogramming favoring propionate fermentation pathways; (3) improved amino acid metabolic efficiency; and (4) reduced methanogenic potential. These findings establish a framework for developing targeted interventions to manipulate gut microbiota colonization patterns and optimize feed conversion in calves.

However, this study has two main limitations. First, the cross-sectional design with sampling at a single time point (84 days) precludes analysis of microbiota succession dynamics. Although previous studies have well established that the gastrointestinal microbiota of calves reaches relative stability during the relevant developmental stage [20–23], longitudinal studies from birth are still needed to determine the temporal sequence between microbial assembly and the emergence of divergent feed efficiency. Second, while fecal microbiota transplantation (FMT) is the gold standard for establishing causality, this study did not perform such intervention. Instead, we employed an orthogonal strategy, glutamate supplementation in mice, to demonstrate that a key microbiota-linked metabolite can causally improve host metabolism. Nevertheless, FMT experiments using LRFI-derived microbiota are required to directly test whether the microbial community as a whole is sufficient to transfer the efficient phenotype. Addressing these limitations represents a critical next step toward translating our correlative findings into causal mechanisms.

Conclusion

This study revealed significant differences in the digestive tract microbiota and their metabolic outputs between LRFI and HRFI calves through multi-omics integrated analysis, and identified key correlated features. The research showed that the microbiota in the rumen and hindgut of LRFI calves had compositional and functional profiles suggesting an inclination centered on starch degradation and propionate production. The enrichment of key bacterial groups was associated with higher starch digestion efficiency and propionate concentrations. In contrast, the microbiota of HRFI calves was more related to the fat metabolism and acetate production pathway. Further metabolomics analysis revealed that serum glutamate levels in LRFI calves were significantly elevated. This increase could not be attributed to differences in feed intake but was positively correlated with endogenous precursors such as ribitol from microbial metabolism in the rumen and taurine from hindgut microbial metabolism, suggesting that microbial activity may influence the host's nitrogen metabolism network. Functional validation experiments showed that glutamate supplementation could directly improve host nitrogen utilization efficiency and intestinal barrier function, supporting its central role in this metabolic network. In summary, the study found that differences in feed efficiency among calves are closely related to distinctly different functional microbial niches within the digestive tract. LRFI calves' microbiota was characterized by efficient carbohydrate fermentation and optimized nitrogen metabolism. These findings provide new perspectives and potential regulatory targets for future microbial management strategies to improve feed efficiency in ruminants.

Methods

Animal management and sampling

A total of 160 periparturient multiparous cows (parity 2.71 ± 0.53; previous 305-day milk yield: 11,050 ± 202 kg; body condition score: 3.45 ± 0.19; mean ± SD) were enrolled in this study. Neonatal calves were evaluated immediately after dam calving, and healthy calves (30–50 kg birth weight) were selected after exclusion criteria were applied: twins, neonates exhibiting weakness (inability to stand, lethargy, or depression), and stillborn calves. A total of 84 Holstein female calves were included in the study. Calves were completely weaned off the milk on day 57 and remained in individual hutches for the whole trial period. Calves had free access to a calf starter without antibiotic and fresh water from day 3. To eliminate the effect of dietary factors, all calves were fed the same starter diet until the end of the study period (day 84). If the calves died before the end of the trial, a new calf was immediately raised, and all data from the dead calf were discarded. Three experimental calves were replaced. Following the experiment, the RFI of the calves was determined using their growth performance data collected throughout the entire experimental duration. The detailed calculation procedure is outlined in section “Residual feed intake calculation”, 43 calves exhibited positive RFI, and 41 showed negative RFI, 20 calves were selected (initial body weight = 41.7 ± 4.1 kg; mean ± SD; serum total protein ≥ 5.5 g/dL) (Table S11) and divided into HRFI (top 10 RFI value, low feed efficiency; n = 10) group and LRFI (bottom 10 RFI value, high feed efficiency; n = 10) group based on the results of RFI value. Milk feed procedures are shown in Table S12. BW and Average daily gain (ADG) were measured or calculated biweekly before the morning feeding. Fecal score, starter and milk intake were recorded daily. Fecal scoring criteria are shown in Table S13. Representative starter was collected once a week for further determination of dry matter (DM), crude protein (CP), ether extract (EE), crude Ash (Ash), neutral detergent fiber (NDF), and acid detergent fiber (ADF) following the methods of AOAC [77]. Milk samples was also collected weekly for further determination of DM, CP, Fat and Solid non-fat (SNF) by Milk composition analyzer (YOUYUNPU spectrum photoelectric Technology Co., Ltd., Shandong province, China), the ingredients and nutrients are shown in Table S14. The glutamate in the feed and feces is hydrolyzed using acid hydrolysis, which breaks down the feed protein into free amino acids under the action of hydrochloric acid, and is then separated and measured by ion exchange chromatography (High-Performance Liquid Chromatograph Chromaster® Amino Acid Analysis System, Japan). Rumen fluid was collected using an esophageal tube (2 mm wall thickness, 6 mm internal diameter; Anscitech Co., Ltd., Wuhan, China) 3 h after morning feeding on day 84 for subsequent 16S RNA gene sequencing, metabolome and metagenome analyses. Blood samples were collected on day 84, before the morning feeding session, for subsequent metabolomic analysis. Jugular venipuncture was performed using vacutainer collection tubes without anticoagulants. Fecal samples were collected twice a day from day 80 to day 84 for digestibility determination. Acid-insoluble ash (AIA) was used as an internal indicator in the feed and feces to determine the total tract apparent digestibility of all calves by the following formula [78].

$$\text{Nutrient digestibility}=100-\frac{100\times (\text{nutrient in feces},\%\times \text{AIA in DM consumed},\%)}{\text{AIA in feces},\%\times \text{nutrient in consumed DM}\%}$$

Four additional fecal samples collected on day 84 were divided into 5 ml centrifuge tubes and stored at −80 °C for subsequent 16S RNA gene sequencing, metabolome, and metagenome analyses. It should be noted that the aforementioned metagenomic and metabolomic assays were not conducted on all the calves. After all the samples from the calves had been collected, based on the results of RFI values, we selected 10 calves from each group, forming subgroups with both differences in RFI values and differences in microbiota structure and composition, for further metagenomic and metabolomic comparisons.

Residual feed intake calculation

An RFI coefficient of 84 Holstein female calves from birth to 84 days of age was determined by subtracting the expected feed intake from the actual feed intake. To estimate the expected feed intake, models were constructed to predict the DMI based on ADG and mid-test metabolic body weight (MMW) by using the PROC REGRESSION procedure in SAS 9.2 (SAS Institute Inc., Cary, NC, USA). The DMI of starter and pasteurized milk served as the dependent variable, whereas the ADG and MMW served as the independent variables.

In this model, RFI represented the residuals derived from a multiple regression model that regressed the DMI on ADG and MMW:

• MBW (kg) = the weight at the middle of the trial (body weight at day 42) [19, 79]

• ADG (kg/day) = (body weight at 84 days of age—birth weight)/84.

• Actual DMI (kg/day) = the average daily dry matter intake of the calf during the entire trial period.

• Predict $\text{DMI}=0.05140\times MB{W}^{0.75}+0.97005\times \text{ADG}-0.37440({\text{R}}^2=0.8161)$

In which β0 is the y-intercept, β1 is the partial regression coefficient of ADG, β2 is the partial regression coefficient of MMW, and ɛ is the error term. Among them, the RFI coefficient for each calf (kg DMI/day) was calculated as the difference between the actual and predicted DMI.

Determination of VFAs in rumen fluid and feces

Rumen fluid and feces of 20 calves at 7, 35, and 84 days of age were used for volatile acid detection. Rumen fluid was collected with esophageal tube (2 mm wall thickness, 6 mm internal diameter; Anscitech Co., Ltd., Wuhan, China) 3 h after morning feeding. The first 20 mL of rumen fluid was discarded to reduce the chances of saliva contamination. The rumen fluid was filtered through four layers of cheese cloth, then divided and placed into two 15 mL centrifuge tubes. Rumen fluid was then stored at − 20 °C for determination of the concentration of VFA with gas chromatography [80]. Centrifugation of rumen fluid samples in 15-mL centrifuge tubes was carried out to remove feed particles and impurities (5,400 rpm for 10 min at 4 °C). Subsequently, 1 mL of the supernatant obtained from centrifugation and 0.2 mL of a 25% metaphosphoric acid solution containing the internal standard 2-Ethylbutyric Acid were precisely added to a 1.5-mL centrifuge tube. The mixture was then homogenized and incubated in an ice-water bath for over 30 min. Afterward, another round of centrifugation was performed (10,000 rpm for 10 min) to eliminate protein precipitates within the samples. Finally, the supernatant was collected and reserved for gas chromatograph detection. To assess fecal VFAs content, 2 g of freshly collected feces are diluted fivefold with 8 mL of ultrapure water. The mixture is homogenized and subsequently centrifuged at 5,400 rpm for 10 min at 4 °C. Subsequently, 1 mL of the supernatant is then collected. The subsequent steps are the same as the detection of volatile acids in rumen fluid.

Glutamate effectiveness verification

A total of 30 4-week-old male BALB/c mice were used in this study. After a 7-day acclimatization period, the mice were randomly assigned to three experimental groups (n = 10 per group): (1) the Control group (CON), receiving oral gavage of purified water; (2) the Monosodium Glutamate (GluNa) group, receiving GluNa at a dose of 30 mg/kg body weight via oral gavage; and (3) the Sodium Chloride (Na) group, receiving NaCl at an equimolar sodium dose of 9.4 mg/kg body weight (calculated based on the sodium content in GluNa) via oral gavage. The experimental period lasted for 14 days. During the trial, body weight was measured every 2 days, and daily feed intake was recorded. At the end of the experiment, blood samples were collected from all mice via retro-orbital bleeding under anesthesia. Serum was separated by centrifugation and stored at −80 ℃ until analysis.

Serum biochemical parameters, including total protein (TP), albumin (ALB), blood urea nitrogen (BUN), glucose (GLU), total cholesterol (TC), and triglycerides (TG), were measured using commercial assay kits (Wuhan Sizhengyuan Biotechnology Co., Ltd., Wuhan, China) following the manufacturer’s instructions on a fully automated biochemical analyzer (BS-370E, Shenzhen Mindray Bio-Medical Electronics Co., Ltd., Shenzhen, China). The concentrations of non-esterified fatty acids (NEFA), insulin-like growth factor-1 (IGF-1), endotoxin (LPS), glutamine (Gln), and glutamate (Glu) were quantified using specific enzyme-linked immunosorbent assay (ELISA) kits (Nanjing Jiancheng Bioengineering Institute, Nanjing, China). Diamine oxidase (DAO) activity was determined using a spectrophotometric assay kit from the same manufacturer.

DNA extraction, 16S rRNA gene sequencing, and data processing for microbiota

Genomic DNA was extracted from rumen and fecal samples on day 84 of 20 selected calves using the QIAamp DNA FAST STOOL Mini Kit (CAT. 51604) according to the manufacturer’s instructions. The DNA extract was checked on 1% agarose gel, and DNA concentration and purity were determined using NanoDrop 2000 UV–vis spectrophotometer (Thermo Scientific, Wilmington, USA). The hypervariable region V3-V4 of the bacterial 16S rRNA gene was amplified with primer pairs 338 F (5'-ACTCCTACGGGAGGCAGCAG-3') and 806R(5'-GGACTACHVGGGTWTCTAAT-3') by an ABI GeneAmp® 9700 PCR thermocycler (ABI, CA, USA), then prepared for paired-end sequencing. The amplicon libraries were sequenced on an Illumina platform using the NEXTFLEX Rapid DNA-Seq Kit with a 2 × 301 bp paired-end configuration. The PCR product was extracted from 2% agarose gel and then purified using the AxyPrep DNA Gel Extraction Kit (Axygen Biosciences, Union City, CA, USA) according to manufacturer’s instructions and quantified using Quantus™ Fluorometer (Promega, USA).

The raw 16S rRNA gene sequencing reads were demultiplexed, quality-filtered by fastp version “0.20.0” [81] and then merged by FLASH version “1.2.7” [82] with the following criteria: (i) the 300 bp reads were truncated at any site receiving an average quality score of < 20 over a 50 bp sliding window, and the truncated reads shorter than 50 bp were discarded, reads containing ambiguous characters were also discarded; (ii) only overlapping sequences longer than 10 bp were merged according to their overlapped sequence. The maximum mismatch ratio in the overlapping region was 0.2. Reads that could not be assembled were discarded; (iii) samples were distinguished according to the barcode and primers, and the sequence direction was adjusted for exact barcode matching and two nucleotide mismatches in the primer matching. A 97% similarity cut-off [83, 84] was clustered using UPARSE version “7.1”(Edgar, 2013), and the chimeric sequences were identified and removed. The taxonomy of each OTU representative sequence was analyzed using the RDP Classifier version “2.2” [85] against the 16S rRNA database using a confidence threshold of 0.7, the annotation information for all OTU samples is detailed in Table S15. For 20 samples, an average of 60,005.20 ± 14,590.97 raw paired-end sequences were obtained per sample. After quality filtering, paired-end sequence merging, and chimera removal, an average of 53,878.68 ± 13,031.50 high-quality sequences were retained per sample (Table S16). The sequencing depth of all samples met the requirements for subsequent analysis.

Metabolite extraction, quality control, and analysis

We conducted a comprehensive metabolome analysis of serum, rumen fluid and fecal samples on day 84 from selected 20 calves. Each 50 mg fecal and 50 μL serum and rumen fluid samples was accurately weighed, and the metabolites were extracted using a 400 µL methanol: water (4:1, v/v) solution with 0.02 mg/mL L-2-chlorophenylalanin as internal standard. The mixture was allowed to settle at −10 ℃ and fecal samples were treated by High-throughput tissue crusher Wonbio-96c (Shanghai Wanbo Biotechnology Co., Ltd.) at 50 Hz for 6 min, then followed by ultrasound at 40 kHz for 30 min at 5 ℃. The samples were placed at −20 ℃ for 30 min to precipitate proteins. After centrifugation at 13,000 × g at 4 ℃ for 15 min, the supernatant was carefully transferred to sample vials for LC-MS/MS analysis. All samples were combined with an equal amount of supernatant to create QC samples for machine testing.

A UHPLC-Q Exactive system (Thermo Fisher Scientific) was used for the LC–MS analysis. Chromatographic conditions: 2μL of the samples were separated by HSS T3 column (100 mm × 2.1 mm i.d., 1.8 μm) and then entered into mass spectrometry detection. After mass spectrometry detection was completed, the raw data were converted to the mzXML format using ProteoWizard and processed using an in-house program developed using R and is based on XCMS for peak detection, extraction, alignment, and integration. The R package and BiotreeDB (V3.0) were used for metabolite identification. The metabolites were searched and identified using the HMDB (human metabolome database) (http://www.hmdb.ca/), KEGG (Kyoto Encyclopedia of Genes and Genomes) (https://www.kegg.jp/), and MetOrigin online server (https://metorigin.met-bioinformatics.cn).

Metagenomic sequencing, gene prediction, taxonomy, and functional annotation

The DNA extract was fragmented to an average size of approximately 400 bp using a Covaris M220 (Gene Company Limited, China) for paired-end library construction. A paired-end library was constructed using the NEXTFLEX® Rapid DNA Seq (Bioo Scientific, Austin, TX, USA). Adapters containing the full complement of the sequencing primer hybridization sites were ligated into the blunt ends of the fragments. Paired-end sequencing was performed on an Illumina NovaSeq (Illumina Inc., San Diego, CA, USA) using NovaSeq Reagent Kits, according to the manufacturer’s instructions (www.illumina.com). The constructed library had an average insert size of 505.53 ± 4.20 bp. The data were analyzed on the online Majorbio Cloud Platform (www.majorbio.com). The paired-end Illumina reads were trimmed of adapters, and low-quality reads (length < 50 bp, quality value < 20, or having N bases) were removed using fastp [81] (https://github.com/OpenGene/fastp, version 0.20.0), resulting in high-quality clean data for subsequent analysis. To minimize the impact of host DNA contamination on microbial community analysis, quality-controlled reads were aligned to the Bos taurus reference genome (ARS-UCD1.2) using Bowtie2 (version 2.4.5) [86] in end-to-end sensitive mode. Metagenomic data were assembled using MEGAHIT [87] (https://github.com/voutcn/megahit, version 1.1.2), which uses succinct de Bruijn graphs. Contigs with a length of more than 300 bp were selected as the final assembly results and used for further gene prediction and annotation. Open reading frames (ORFs) from each assembled contig were predicted using MetaGene [88] (http://metagene.cb.k.u-tokyo.ac.jp/). Predicted ORFs with a length ≥ 100 bp were retrieved and translated into amino acid sequences using the NCBI for Biotechnology Information translation table (http://www.ncbi.nlm.nih.gov/Taxonomy/taxonomyhome.html/index.cgi?chapter=tgencodes#SG1.) A non-redundant gene (NR) catalog was constructed using CD-HIT [89] (http://www.bioinformatics.org/cd-hit/, version 4.6.1) with 90% sequence identity and 90% coverage. Reads after quality control were mapped to the non-redundant gene catalog with 95% identity using SOAPaligner [90] (http://soap.genomics.org.cn/, version 2.21) and gene abundance in each sample was evaluated (Supplementary Dataset 1). Representative sequences of the non-redundant gene catalogs were aligned to the NCBI NR database with an e-value cutoff of 1e-5 using Diamond [91] (http://www.diamondsearch.org/index.php, version 0.8.35) for taxonomic annotations (Supplementary Dataset 1). The KEGG annotation was conducted using Diamond [91] (http://www.diamondsearch.org/index.php, version 0.8.35) against the Kyoto Encyclopedia of Genes and Genomes database (http://www.genome.jp/keeg/) with an e-value cutoff of 1e⁻⁵.

Statistical analysis

Power analysis, with G*Power software (version 3.1.9.7), was used to determine the adequate number of calves per treatment, and we obtained a power (1 − β) of 0.80 and a type-I error probability (α) of 0.05. The effect size f was determined from a previous study [8] to be 0.76 for DMI, 1.10 for Relative abundance of the differentially abundant bacteria and 1.41 for peak intensity of the differential metabolites. Based on the power analysis results, a minimum of 8 calves for each group for the phenotype study, 5 calves for each group for the bacterial composition study and 6 calves for each group for the metabolome analysis were required to meet the power requirements.

Parametric and non-parametric tests

All continuous phenotypic variables (body weight, ADG, DMI, digestibility coefficients, VFA concentrations, blood physiology and biochemistry) were tested for normality using the Shapiro–Wilk test prior to group comparisons (PROC UNIVARIATE and PROC DISCRIM version 9.2; SAS Institute Inc., Cary, NC, USA). Variables that did not violate the normality assumption (P > 0.05) were analyzed using parametric tests (Student’s t‑test for two‑group comparisons, mixed models for repeated measures). For variables that deviated significantly from normality (P < 0.05), we used the corresponding non‑parametric alternatives (Wilcoxon rank‑sum test for two‑group comparisons: wilcox.test function in R). Data of BW were analyzed based on days (day 1, 14, 28, 42, 56, 70, and 84). ADG and starter DMI data were pooled biweekly and weekly, respectively, and analyzed based on week (weeks 1–12). Rumen and fecal VFAs were analyzed based on days (days 7, 35, and 84). Data were analyzed using a mixed model (PROC MIXED version 9.2; SAS Institute Inc., Cary, NC, USA). The model included the fixed effects of treatment, time (day or week), and their interaction (treatment × time), with calf as a random effect. Fecal scores for each calf were recorded daily and used to calculate diarrhea frequency. Data on diarrhea frequency were analyzed using a GLIMMIX procedure in SAS (version 9.2, SAS Institute Inc., Cary, NC, USA) [92]. Unpaired t-test (PROC TTEST version 9.2; SAS Institute Inc., Cary, NC, USA) was used to analyze the difference in digestibility between the HRFI and LRFI calves. Significant differences were declared at P ≤ 0.05, with trends indicated at 0.05 < P ≤ 0.10.

Student’s t-test (t.test function in R) was used to analyze indicators of alpha diversity, with the FDR-adjusted P value < 0.05 considered significant and 0.05 ≤ P < 0.10 considered a tendency. For beta diversity, a PCoA of the bacterial and metabolite profiles was conducted using UniFrac distance matrices in R. Beta diversity based on Bray–Curtis distances was calculated and tested using an analysis of similarity (ANOSIM). ANOSIM was performed using the vegan package in R, with the Bray–Curtis distance algorithm and 999 permutations. Expression of enzyme in functional pathways, rumen fluid and fecal bacteria were compared using the Wilcoxon rank-sum test with an FDR-adjusted P < 0.05, which was considered significant.

Microbial network analysis

To compare co-occurrence patterns between high-feed efficiency and low-feed efficiency calves, we constructed separate SparCC networks for the HRFI and LRFI groups within each sample type (rumen and feces) using the Sparse Correlations for Compositional data (SparCC) algorithm [93, 94] implemented via the sparccboot function from the SpiecEasi package in R. The input was a genus-level abundance table (rows = samples, columns = genera). To assess the statistical significance of the inferred correlations, a bootstrap procedure with 100 iterations was performed using the sparccboot function. The resulting p-values and correlation coefficients were extracted using the pval.sparccboot function. Significant correlations were defined as those with a nominal P < 0.05 and an absolute correlation strength |r|> 0.5. A microbial co-occurrence network was built using the significant correlations. In this network, nodes represent microbial genera, and edges represent significant positive or negative correlations. The weight of each edge was defined as the absolute value of the correlation coefficient. Genera with no significant connections to any other node were removed from the final network using the delete.vertices function. The networks within LRFI and HRFI groups were visualized using Gephi v0.10.1 (https://gephi.org).

Correlation analysis

A data matrix comprising the actual concentrations of individual VFAs and the normalized gene count of key microbes was used for Spearman correlation analysis. The analysis result of Correlation heatmap plot was generated using the R software packages "corrplot" through the CNSknowall (https://cnsknowall.com). In the heatmap, hierarchical clustering of both rows and columns was performed using the complete linkage method with Euclidean distance as the similarity measure. The relationship between ribitol, taurine and glutamate was determined using linear regression and correlation models (PROC REG and CORR, version 9.2; SAS Institute, Inc., Cary, NC, USA). Associations with nominal P < 0.05 were considered statistically significant.

Multi-omics integration analysis of microbial and metabolic features

To investigate the microbial and metabolic signatures associated with RFI and their interactions, we performed an integrated multi-omics analysis. The gut microbiome dataset and the metabolome dataset were integrated using the DIABLO framework within the mixOmics package in R [95, 96]. First, to enhance model robustness, features with near-zero variance across all samples were filtered out from the raw microbial and metabolomic data using the nearZeroVar() function from the caret package. A design matrix was constructed to specify the correlation structure between the two data blocks (microbiome and metabolome). We set all non-diagonal entries to 0.1 using the matrix function, and the diagonal to 0. This value was chosen to allow moderate association between blocks while avoiding overfitting. The filtered data were then structured into two data blocks: one for microbial features and another for metabolic features, with the RFI groups (high/low) serving as the supervised outcome (Y). A supervised multiblock partial least squares discriminant analysis (sPLS-DA) model (block.splsda function) was applied. The model was initially fitted with up to 5 components and the design matrix defined above. The optimal number of latent components in the model was determined via tenfold cross-validation repeated 50 times using the perf function, selected based on the lowest balanced error rate (BER). Subsequently, the number of features to retain in each data block for each component was tuned using a grid search with the tune.block.splsda function. The search range for each block was set to c(5:9, seq(10, 18, 2), seq(20,30,5)), as suggested in the package documentation to explore a broad range of feature combinations while balancing computational cost. The optimal keepX parameters (list.keepX) were selected based on the lowest BER. The final DIABLO model was constructed with the selected ncomp and list.keepX using the block.splsda function. Using the final model, key microbial and metabolic features contributing most to the discrimination between RFI groups were identified and visualized using the selectVar and plotLoadings functions. A cross-correlation network between the two omics datasets was visualized via the circosPlot function, retaining only correlations with an absolute value greater than 0.7. Finally, the predictive performance of the model, represented by the Area Under the Receiver Operating Characteristic curve (AUC), was calculated and evaluated for each data block using the auroc function, accuracy was calculated using the perf function to perform cross‑validation on the final DIABLO model, this function outputs both the BER and the overall error rate (Overall.ER) for each component under different distance metrics. From these error rates, the classification accuracy is calculated as: accuracy = 1 − Error Rate. A fixed random seed (set.seed(123)) was set for all analytical procedures involving random processes to ensure reproducibility.

To compare the overall concordance between the microbiome and metabolome datasets in the rumen and feces, we performed Procrustes analysis using the protest function in the vegan R package. Briefly, principal coordinate analysis (PCoA) based on Bray–Curtis distances was applied to each dataset separately, and the resulting ordinations were rotated to maximize alignment. The significance of the concordance was assessed using 999 permutations.

Metabolite pathway enrichment analysis

Significantly different metabolites were selected based on the variable importance in the project (VIP) obtained using the PLS-DA model and the P-value of Student’s t-test: metabolites with VIP > 1 and corrected P < 0.05. Pathway enrichment analysis of differentially abundant metabolites was performed on the Majorbio Cloud Platform (www.majorbio.com) with Scipy.stats (Python package) (https://docs.scipy.org/doc/scipy/). Metabolites were mapped to the KEGG database (v2021.09.18). The significance of enrichment for each pathway was tested using a Fisher’s exact test, followed by strict Bonferroni correction for multiple testing. Pathways with a corrected p-value (adj. p) < 0.05 were considered significantly enriched. To identify key metabolites within the significant pathways, a topological analysis was conducted based on the relative-betweenness centrality metric, which helps pinpoint hub compounds that occupy central positions in the pathway network.

In the above P-values, whenever multiple comparisons were involved, the P-values were corrected. In this paper, corrected P-values are referred to as corrected P, while uncorrected P-values are referred to as nominal P.

Authors’ contributions

T.Y.C., X.J.X. and L.S.R. conceived the study, conducted animal experiments, prepared the original draft, and analyzed the output data. R.P., X.J.Z., M.S., Y.M.X., J.J.W., J.Y.M., and Y.M.Z conducted fecal nutrient determination and data analysis. L.S., S.L.L., and Z.J.C. contributed to manuscript revisions. L.S. and Z.J.C. also provided financial support. W.W., J.G., and M.M. provided supervision. All the authors read and approved the final manuscript.

Funding

This study was supported by the National Key Research and Development Program (No.2021YFF1000703-03), the Key Research and Development Program of Ningxia (2024BBF01006) and Pinduoduo-China Agricultural University Research Fund (PC2023B02002).

Data availability

The 16S rRNA gene sequences from rumen and fecal samples are available in the NCBI Sequence Read Archive with the accession ID: PRJNA1222629 (https://www.ncbi.nlm.nih.gov). The Metagenome sequences can be accessed in the NCBI Sequence Read Archive under the accession ID: PRJNA1123313 (https://www.ncbi.nlm.nih.gov). The Metabolomic data have been deposited in the OMIX, China National Center for Bioinformation/Beijing Institute of Genomics, Chinese Academy of Sciences under the accession ID: PRJCA036201 (https://ngdc.cncb.ac.cn/omix/releaseList).

Declarations

Ethics approval and consent to participate

This work has received approval for research ethics from the Ethical Committee of China Agricultural University (Case number: AW82211202-1–1) and Ethics Committee of CONTEC Medical Laboratory Services Hebei Co., LTD (Case number: MDL2026-01–16-03), a proof/certificate of approval is available upon request.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.