Plasma metabolomics reveals metabolic signatures associated with Lactobacillus reuteri supplementation in ewes

Introduction

The U.S. sheep industry plays an important role in supporting the economic viability and vitality (American Sheep Industry Association, 2021; National Research Council et al., 2008). However, sheep farming faces compounding production challenges including persistent rising feed costs and escalating antimicrobial resistance pressures (Saleem et al., 2025), thus underscoring the urgent necessity for evidence based, non-antimicrobial alternatives capable of sustaining productivity and safeguarding health in small ruminant production systems (Bava et al., 2024; Ibeagha-Awemu et al., 2025). Probiotics, live, non-pathogenic microorganisms that confer health benefits when administered in adequate quantities (FAO, 2016), have emerged as practically viable alternatives to prophylactic antimicrobial use in ruminant nutrition (Gul and Durante-Mangoni, 2024; Saleem et al., 2024). Prior studies in sheep have demonstrated that probiotic supplementation significantly reduced ruminal ammonia nitrogen concentrations, increased volatile fatty acid production, improved blood intermediates and vitamins, establishing that probiotic mediated ruminal modulation propagates systemically through the portal circulation and hepatic processing machinery beyond the rumen compartment (Haque, 2018; Saleem et al., 2025).

Among the most extensively studied probiotic species in livestock production systems is LR, a gram positive, heterofermentative lactic acid bacterium naturally resident in the gastrointestinal tracts of mammals, including sheep (Walter, 2008; Liu et al., 2024). LR exerts direct antimicrobial action through the enzymatic conversion of glycerol to reuterin (3-hydroxypropionaldehyde), a broad-spectrum antimicrobial compound that disrupts thiol metabolism in competing microorganisms, alongside the biosynthesis of vitamins, exopolysaccharides and amino acids (Schaefer et al., 2010; Yang et al., 2024). These bioactive metabolites act in concert to suppress enteric pathogen proliferation, downregulate pro inflammatory cytokine signaling cascades including NF-κB and MAPK pathways, and reinforce tight junction protein expression to maintain intestinal mucosal barrier integrity (Mu et al., 2018; Wang et al., 2021; Huang et al., 2023). In fact, evidence from Sunit lambs further demonstrated that LR supplementation enhanced rumen fermentation, regulated protein degradation, and optimized muscle aromatic compound profiles, confirming that its systemic influence extends well beyond the intestinal compartment into hepatic, muscular, and circulatory metabolic networks through integrated gut and liver axis signaling (Liu et al., 2023). Similarly, Lactobacillus fermentation products and enzymes such as cellulases, hemi-cellulases, proteases, and ferulic acid esterases have shown promise in improving fiber digestion in dairy and beef cattle (Yang et al., 1999; Beauchemin and Holtshausen, 2010).

Despite these well characterized intestinal and ruminal effects, the systemic plasma metabolomic consequences of LR supplementation in ewes remain entirely uncharacterized, representing a fundamental knowledge gap with both mechanistic and translational significance. The systemic jugular plasma metabolome represents the integrated output of multiple tissues and physiological processes and does not permit direct attribution of individual metabolite changes to portal absorption, hepatic biotransformation, or ruminal fermentation. Nevertheless, it captures the net downstream signal of gut and host metabolic crosstalk, including portal transfer products of microbial fermentation and their hepatic biotransformation into circulating metabolites that modulate peripheral tissue function, immune competence, and oxidative homeostasis (Taiwo et al., 2024; Zhang et al., 2024a). While studies in piglets have established that LR supplementation alters circulating amino acid profiles, lipid metabolite pools, and oxidative stress markers (He et al., 2025; Wang et al., 2025), direct extension to ruminant systems is biologically inappropriate given the fundamental differences in pre-gastric fermentation, nitrogen recycling architecture, and portal metabolite composition. The host specificity of LR, driven by strain level ecological adaptations including host specific adhesin repertoires and metabolic capabilities (Frese et al., 2011; Duar et al., 2017), further renders ewe specific in-vivo investigation an inevitable mechanistic step. Therefore, we hypothesized that dietary LR supplementation would induce measurable and biologically meaningful shifts in the plasma metabolome of ewes. This study characterized the effects of dietary LR supplementation on the performance phenotype and plasma metabolome of ewes to identify discriminating metabolites, associated pathways, and putative biomarkers of supplementation status.

Materials and methods

Animals, feeding, and experimental design

All animal procedures were conducted in accordance with institutional animal care and use guidelines of Fort Valley State University (IACUC protocol number SU R-02- 2025). Eighteen Katahdin ewes (approximately 6 months of age; average BW = 26.32 kg, pooled SE 1.38) were used in a completely randomized design. Animals were individually penned and acclimated for 21 days prior to the start of the experimental period to minimize handling stress and establish baseline feed intake patterns. Following acclimation, ewes were randomly assigned to one of two dietary treatments: (1) a control group (CON; n = 9) receiving a commercially formulated basal diet (Godfrey’s Show and Grow Lamb Feed; minimum 16% crude protein, minimum 2.4% crude fat, maximum 13% crude fiber); or (2) a treatment group (LR; n = 9) receiving the same basal diet supplemented with 10 g of freeze dried LR (2 × 10^9 CFU/g) per ewe per day, administered as a top dress. The 10 g/head/day dose (delivering approximately 2 × 10¹⁰ CFU per ewe per day) was selected based on prior small ruminant work: Moody et al. (2022) demonstrated significantly improved lamb performance at 10 g/head/day of a lactate-producing probiotic paste compared with 5 g or 0 g controls, and Sun et al. (2024) confirmed that LR supplementation in Sunit sheep improves ruminal microbiota composition and downstream tissue outcomes. Feed was offered ad libitum at 4% of individual body weight, adjusted weekly based on recorded body weights, with fresh water available at all times. The experimental period lasted 35 days. Body weights were recorded weekly. Feed offered was weighed and recorded daily using a calibrated platform scale, and orts were collected and weighed before the next feed delivery. No consistent feed refusals were observed across either treatment group, indicating that actual intake approximated feed offered throughout the experimental period.

Blood sample collection and plasma preparation

Blood was collected weekly on days 7, 14, 21, 28, and 35 of the experimental period. On each sampling day, feed was withheld for approximately 12 h overnight to standardize the pre-prandial metabolic state, while water remained available ad libitum. Between 07:00 and 08:30 the following morning and before any feed was offered, approximately 10 mL of whole blood was collected from each ewe by jugular venipuncture using a 21-gauge needle into K₂EDTA vacutainer tubes (BD Biosciences, Franklin Lakes, NJ). All animals were sampled within a 60 to 90 min window, and treatment groups were rotated across the sampling order so that any within-window drift was distributed equally across groups. Tubes were placed on ice immediately after collection, transported to the laboratory within 30 min, and centrifuged at 3,000 × g for 15 min at 4 °C. Plasma supernatant was aliquoted into cryovials and stored at −80 °C until untargeted metabolomic analysis.

Sample preparation for plasma metabolomics profiling

Plasma samples collected on d 7, 14, 21, 28, and 35 were composited for each individual ewe (equal volumes pooled per ewe across the five timepoints) to generate a representative cumulative metabolic profile representative of the entire 35-day intervention period rather than any single-day snapshot, and to reduce technical variance across individual sampling occasions. This pooling design necessarily removes the ability to resolve temporal dynamics of the metabolomic response, a trade off that is discussed in the Limitations and Future Directions subsection. Metabolites were extracted from the composited plasma using a methanol protein precipitation method as previously described (Zhao et al., 2019; Taiwo et al., 2022). Briefly, plasma samples were mixed with methanol at a ratio that precipitated protein, vortexed, centrifuged to remove the protein pellet, and the supernatant was collected and dried. The dried extracts were re-dissolved in 200 µL of water and stored at −80 °C until CIL/LC-MS metabolome analysis was performed.

CIL/LC-MS based metabolomics analysis

In depth untargeted metabolome profiling of the extracted plasma was performed using a chemical isotope labeling (CIL) LC-MS based technique. The technique uses differential ¹²C/¹³C isotope labeling to derivatize metabolites based on their chemical functional groups (amines/phenols, carboxylic acids, and carbonyls), enabling comprehensive sub-metabolome coverage (Zhao et al., 2019). Briefly, extracted metabolites were labeled with light (¹²C) and heavy (¹³C) dansyl chloride reagents targeting amine and phenol groups; sample amounts were normalized by LC-UV quantification of the total labeled metabolite peak area to standardize input across samples prior to injection (Wu and Li, 2012; Zhao et al., 2019). Relative quantification of ¹²C/¹³C-labeled metabolites based on peak-pair ratios was performed on a Bruker Compact quadrupole time-of-flight mass spectrometer (Bruker Daltonics, Billerica, MA) coupled to an UltiMate 3000 ultra-high-performance liquid chromatography system (Thermo Scientific, Waltham, MA), operated in positive-ion electrospray ionization mode with a mass range spanning small-molecule metabolites and typical sub-5-ppm mass accuracy characteristic of the platform. A total of 18 LC-MS data files were generated (9 LR and 9 CON samples), with 3 pooled study quality control (QC) samples analyzed at regular intervals throughout the analytical run. Pooled study QCs were prepared by combining equal aliquots of all 18 study plasma samples and were used to monitor analytical performance and instrument stability across the analytical run. Detailed description of the metabolite labeling, sample normalization, and LC-MS operating conditions have been reported previously (Zhao et al., 2019).

Metabolite data processing and identification

All raw LC-MS data files were processed using IsoMS Pro 1.0 following the procedures described by Mung and Li (2017). Briefly, 12C/13C peak pairs were extracted from each run by the IsoMS software. In this step, redundant pairs (those of adduct ions and dimers) and noise signals (having a singlet peak) were filtered out, retaining only the protonated ion of a peak pair for each true metabolite, and the peak intensity ratio was calculated for each peak pair. The IsoMS-Quant program was then used to determine the chromatographic peak ratio of each peak pair and to generate the final metabolite intensity table (Huan and Li, 2015). Metabolite identification employed a three tier approach: in Tier 1, peak pairs were searched against the CIL metabolite library (containing 1,060 unique endogenous metabolites) based on accurate mass and retention time for positive identification; in Tier 2, a Linked Identity (LI) library containing over 2,000 metabolic pathway related metabolites extracted from the KEGG database was used for high confidence putative identification based on accurate mass and predicted retention time (Li et al., 2013); in Tier 3, the MyCompoundID library (mycompoundid.org; Li et al., 2013) was searched for additional putative identifications. Software tools used in this identification workflow are: IsoMS Pro 1.0 (Nova Medical Testing, Edmonton, Canada; Mung and Li, 2017), IsoMS-Quant (Huan and Li, 2015), the CIL metabolite library (Nova Medical Testing), the KEGG-derived Linked Identity library (Li et al., 2013), and the MyCompoundID library (mycompoundid.org). A total of 8,669 peak pairs were detected across all samples.

Statistical analysis

PS1PPPFinal body weight was analyzed using analysis of covariance (ANCOVA), with treatment as a fixed effect and initial body weight included as a covariate to account for baseline variation among animals. Total weight gain, ADG, and feed: gain ratio were analyzed by one-way GLM with treatment as the fixed effect. For final body weight, initial body weight was included as an additional covariate in the multiple linear regression model to mirror the ANCOVA structure. Significance was declared at ≤ 0.05 and trends at 0.05 < P ≤ 0.10 for performance and correlation analyses. For metabolomics, differentially abundant metabolites were identified by Welch’s two-sample t-test P value < 0.05 on log2-transformed intensities. Processed metabolite intensity values were imported into MetaboAnalyst 6.0 (Pang et al., 2024; Chong et al., 2019) for multivariate and univariate statistical analyses. Prior to statistical testing, features with more than 50% missing values across the 18 samples were removed; the remaining missing values were imputed using the k nearest neighbours (KNN, k = 10) algorithm as implemented in MetaboAnalyst 6.0. In addition, an interquartile-range (IQR) low-variance filter was applied within MetaboAnalyst 6.0 to remove the 40 percent of features with the lowest IQR across samples prior to univariate testing. Retained features were log2 transformed, median normalized to eliminate confounding inter sample technical variation, and auto scaled (mean centered and divided by the feature standard deviation) prior to multivariate and univariate testing (Taiwo et al., 2022). Partial least squares discriminant analysis (PLS-DA) was used to visualize group separation between CON and LR ewes, with model performance evaluated by R2Y, Q2, and permutation testing (n = 2,000); the most discriminating metabolites were identified by variable importance in projection (VIP) scores greater than or equal to 1.0. Differentially abundant metabolites were identified by volcano plot analysis with a fold change threshold of > 1.2 or < 0.833 (= 1/1.2) and a Welch’s two-sample t-test P value < 0.05 on log2-transformed intensities. The complete statistical results for all 1,003 detected metabolites (compound name, group means, fold change, log2 fold change, t statistic, and raw P value) are provided in Supplementary Table This complete 1,003-metabolite table was generated in parallel using R (Welch’s t-test on log2-transformed peak intensities) because the MetaboAnalyst volcano module exports only the metabolites meeting the differential-abundance filter (n = 27) rather than the full detected feature set. As a sensitivity analysis on the multiple-testing burden, Benjamini-Hochberg false-discovery-rate correction was additionally applied in R across the full 1,003-metabolite set using p.adjust with method = ‘BH’; the resulting adjusted q values are reported alongside raw P values in Supplementary Table 1. Pathway enrichment was conducted with the MetaboAnalyst pathway module against the KEGG Ovis aries library using the hypergeometric over representation test and relative betweenness centrality for topology analysis; pathways with ≤ 0.05 and topology impact > 0.10 were considered biologically significant. Only the 27 metabolites identified as significantly differentially abundant (< 0.05), not the full set of 1,003 detected metabolites, were evaluated as putative biomarkers of LR supplementation by receiver operating characteristic (ROC) curve analysis (Xia et al., 2013), with area under the curve (AUC) ≥ 0.85 used to identify individual candidates; combined biomarker panels were assessed using the random forest classifier as implemented in MetaboAnalyst 6.0 with the default settings (500 trees, √p features tried at each split, and Monte Carlo cross validation with 100 iterations resampling two thirds of the samples as the training set and the remaining one third as the test set at each iteration); the reported area under the receiver operating characteristic curve and its 95% confidence interval reflect the cross validated performance across these iterations. The full list of pathways with total metabolite counts, hit counts, raw hypergeometric P values, Holm Bonferroni adjusted P values, and topology impact scores is provided in Supplementary Table 2. Associations between putative biomarkers and performance traits (ADG, feed: gain) were evaluated by Spearman rank correlation on median normalized log2 transformed intensities and by multiple linear regression with LR supplementation status and the putative biomarkers as predictors; variance inflation factors (VIF) were calculated to assess multicollinearity, and isomer 2 of sebacic acid was excluded due to severe multicollinearity with 9-oxononanoic acid. Correlation and regression analyses were performed in R version 4.3.3 (R Core Team, 2024) using dplyr 1.1.4 (Wickham et al., 2023), broom 1.0.8 (Robinson et al., 2024), ggplot2 3.5.2 (Wickham, 2016), car 3.1-2 (Fox and Weisberg, 2019) for variance inflation factors, and pROC 1.18.5 (Robin et al., 2011) for receiver operating characteristic analysis. Homogeneity of regression slopes for the ANCOVA model was tested by including a treatment by initial body weight interaction (P = 0.66) before running the reduced model. Because the Spearman correlations were conducted on a small pre-selected biomarker set (six metabolites nominated from the ROC analysis with AUC ≥ 0.85 and < 0.05) rather than on the full 1,003-metabolite feature set, and because these correlations are exploratory rather than primary hypothesis tests, an additional multiple-testing correction was not applied to the correlation analyses; Spearman P values are reported unadjusted and interpretation focuses on the strongest bivariate signals together with their multiple regression counterparts.

Results

The performance of LR and CON ewes is shown in Table 1 and Figure 1. Initial body weight and daily feed offered were similar between groups (P = 1.00 and P = 0.99, respectively). Total weight gain and ADG were greater in LR ewes than in CON ewes (both P < 0.001), and the feed: gain was lower in LR ewes (P ≤ 0.05), indicating improved feed conversion efficiency. After adjusting for initial body weight by ANCOVA, final body weight was also greater in LR than in CON ewes (adjusted means 36.08 vs 33.73 kg; difference +2.34 kg, P < 0.001), confirming that the treatment effect on cumulative gain was reflected in a heavier finishing weight.

Table 1. Growth performance of Katahdin ewes fed a basal diet (CON) or the basal diet supplemented with Lactobacillus reuteri (LR) over a 35-d feeding period.

¹Pooled standard error of the mean from the fitted model. ²Initial body weight and daily feed offered compared by Welch two sample t-test. Final body weight compared by ANCOVA with initial body weight as a covariate (homogeneity of slopes P = 0.66). Total gain, ADG, and feed: gain compared by one-way GLM with treatment as a fixed effect.

³Adjusted (least squares) means from the ANCOVA at the common initial body weight of 26.32 kg.

Metabolome associated with LR supplementation status

A total of 1,003 metabolites were detected and identified from 8,669 peak pairs after quality filtering, normalization, and removal of features with more than 50% missing values (Supplementary Table 1). The PLS-DA scores plot was performed as a supervised visualization of covariance between metabolite profiles and treatment assignment (components 1 and 2 explained 10.4% and 8.1% of variance, respectively; R2Y = 0.86, Q2 = 0.42, permutation P = 0.683, 1,366 of 2,000 permutations; Figure 2; Supplementary Figure 2), and the PLS-DA is therefore retained solely as a supervised descriptive visualization rather than as evidence of intrinsic treatment-driven group separation. Robust interpretation of the LR metabolic signature accordingly rests on the univariate differentially abundant set, ROC and random-forest biomarker discovery, and phenotype correlations. The unsupervised principal component analysis (PCA) scores plot showed substantial overlap between the CON and LR 95% confidence ellipses along PC1 (16.3% of variance) and PC2 (13.3% of variance), confirming that the intrinsic clustering pattern of the metabolomics data does not cleanly separate the two groups in unsupervised space (Supplementary Figure 1). Twenty-seven plasma metabolites were identified as differentially abundant between the LR and CON groups (P < 0.05). Plasma relative abundance of 9 metabolites, including cyanamide, an isomer of aspartic acid, L-homoserine, 1-pyrroline-4-hydroxy-2-carboxylic acid, and 3-hydroxypicolinic acid, showed higher relative abundance (P ≤ 0.05) in LR supplemented ewes, while 18 metabolites, including 9-oxononanoic acid, isomer 2 of sebacic acid, hydroxylated lecithin, L-fucose, tyramine, and nocardicin A/isonocardicin A, showed higher relative abundance (P ≤ 0.05) in CON ewes (Table 2; Figures 3, 4A, B).

Table 2. Fold change and P value for all 27 differentially abundant plasma metabolites in ewes supplemented with Lactobacillus reuteri.

FC, fold change relative to CON (values > 1 indicate greater abundance in LR ewes; values < 1 indicate greater abundance in CON ewes). P value, raw Welch’s two-sample t-test P value on log2-transformed intensities. CON, ewes fed the basal diet; LR, ewes supplemented with Lactobacillus reuteri.

Nine (9) of the ten top metabolites, 3-Hydroxybenzyl alcohol, L-Ornithine/D-Ornithine, β-alanine, Isopyridoxal, L-Fucose, Hydroxylated lecithin, 9-Oxononanoic acid, Isomer 2 of Sebacic Acid, and Nocardicin A/Isonocardicin A, were enriched in CON and depleted in LR, with CON values clustered near the outer rim (z approximately 0.7 to 0.76) and LR values pulled toward the center (z approximately -0.3). Cyanamide displayed the opposite pattern and was the single metabolite markedly elevated in LR relative to CON.

The results of the ROC analysis identified six metabolites (beta-alanine, L-ornithine/D-ornithine, hydroxylated lecithin, 9-oxononanoic acid, isomer 2 of sebacic acid, and cyanamide) with AUC values ≥ 0.85, providing sufficient specificity and sensitivity to qualify as putative biomarkers of discrimination between LR and CON groups (Figure 5). The box plot distributions of these six putative biomarkers in LR-supplemented and control ewes are provided as Supplementary Figure 2 (see supplementary figure legend for detailed description). The combined six metabolite biomarker panel achieved an AUC of 0.961 (95% CI: 0.778 1.000), consistent with discriminatory information being distributed across multiple candidate biomarker metabolites rather than concentrated in a single feature (Figure 6). Results of the pathway analysis of all metabolites showed that cysteine and methionine metabolism, tryptophan metabolism, primary bile acid biosynthesis, and lysine degradation were enriched (P ≤ 0.05, topology impact > 0.10) in LR supplemented, relative to control ewes (Figure 7). For the four discussed pathways, the specific hit metabolites drawn from the 27 differentially abundant set were as follows: lysine degradation (2 hits from the differentially abundant set: L-ornithine and an isomer of 4-chloro-L-lysine, out of 8 KEGG pathway matches in the detected feature set); cysteine and methionine metabolism (1 hit: Isomer 2 of 2-Amino-2-methyl-1,3-propanediol, out of 11 detected KEGG matches); tryptophan metabolism (0 direct hits from the differentially abundant set, out of 18 detected KEGG matches); and primary bile acid biosynthesis (0 direct hits from the differentially abundant set, out of 6 detected KEGG matches). We note that for the tryptophan metabolism and primary bile acid biosynthesis pathways in particular, structurally related isomers or close analogs of canonical KEGG pathway compounds were present within the broader detected feature set (n = 1,003), and these contributed to the pathway enrichment signal even though no exact canonical pathway member appeared in the 27 differentially abundant set. The enriched pathways should therefore be interpreted as reflecting metabolic contexts within which structurally similar (often isomer-level) metabolites shifted with LR supplementation, rather than as unequivocally demonstrating shifts in the exact KEGG canonical pathway members.

Association of biomarkers with ewes performance traits

To assess the biological importance of the putative biomarkers, spearman rank correlation analysis was performed (n = 18 ewes; 9 CON, 9 LR) between median normalized log2 transformed plasma metabolite intensity and two production traits: ADG (ADG, kg per day) and feed efficiency (feed: gain). ADG was positively correlated with cyanamide (ρ = 0.67, P < 0.01) and negatively correlated with hydroxylated lecithin (ρ = -0.75, P < 0.01), β-alanine (ρ = -0.58, P < 0.05), and isomer 2 of sebacic acid (ρ = -0.47, P < 0.05), with trends toward association for L-ornithine/D-ornithine (ρ = -0.43, P = 0.073) and 9-oxononanoic acid (ρ = -0.45, P = 0.062) (Figure 8). The reverse pattern was observed for feed conversion ratio, where hydroxylated lecithin was the only biomarker with a significant positive correlation (ρ = 0.68, P = 0.002) and cyanamide showed a negative trend (ρ = -0.42, P = 0.081); the remaining four biomarkers showed weaker positive associations that did not reach significance (Figure 9). Since a low feed conversion ratio reflects higher feed efficiency, ewes with elevated plasma cyanamide and depleted lipid mediated and amino acid pool biomarkers, consistent with the LR signature, were associated with the improved growth performance observed in this study. We acknowledge that these correlations were computed across the pooled CON and LR samples (n = 18) and therefore may partly reflect the treatment-driven separation between groups; the reduced within-group sample size (n = 9 per group) provides limited statistical power for formal within-group correlation testing, and the reported correlations should be interpreted primarily as descriptive associations at the whole-cohort level.

To quantify the independent contribution of each biomarker, multiple linear regression models were fitted with LR supplementation status and the putative biomarkers as predictors. Sebacic acid (isomer 2) was excluded due to severe multicollinearity with 9-oxononanoic acid (see Table 3), leaving five biomarkers with variance inflation factors of 1.7 to 3.7 (acceptable multicollinearity). The ADG model was significant (adjusted R² = 0.58, F(6, 11) = 4.86, P < 0.05), with LR supplementation as the only significant independent predictor (β = 0.068; P < 0.05) and 9-oxononanoic acid (β = 0.033; P = 0.11) and hydroxylated lecithin (β = -0.035; P = 0.19) showing the most non-significant trends; cyanamide, β-alanine, and L-ornithine/D-ornithine did not achieve independent significance in the multivariable ADG model; the individual biomarker signals were subsumed under the LR supplementation indicator, consistent with the strong LR treatment effect and the overlap between biomarker and treatment variance. For feed conversion ratio, the multiple regression model did not reach overall significance (adjusted R² = 0.11, F (6, 11) = 1.36, P = 0.312; Supplementary Table 1); only hydroxylated lecithin showed a trend toward association (β = 2.01, P = 0.10), consistent with its bivariate correlation shown in Figure 9. For final body weight, a multiple linear regression that included initial body weight as a covariate (Table 4) explained 96% of the variance (adjusted R² = 0.96, F (7, 10) = 58.01, P < 0.01); LR supplementation (β = 2.83, P < 0.01) and 9-oxononanoic acid (β = 1.50, P < 0.01) were significant independent predictors of final body weight. Consistent with the ANCOVA analysis, no single biomarker was significantly correlated with FBW at the bivariate level (all P > 0.05; Supplementary Figure 3), because FBW is dominated by variation in initial body weight; the biomarker signal for FBW is captured only in multiple regression. Overall, in the bivariate analysis, hydroxylated lecithin showed the strongest single-metabolite correlation with the growth and feed efficiency phenotype (ρ = -0.75 for ADG; ρ = 0.68 for feed:gain), although in the multivariable regression it did not reach independent significance for ADG (β = -0.035, P = 0.19) and only showed a trend for feed conversion (β = 2.01, P = 0.10), while cyanamide elevation and the broader depletion of lipid peroxidation and amino acid pool biomarkers provided additional, partially overlapping signal (Tables 3, 4; Supplementary Table 1).

Table 3. Multiple linear regression of average daily gain (kg per day) on Lactobacillus reuteri supplementation and putative plasma biomarkers in ewes (n = 18).

n = 18. B, unstandardized regression coefficient; SE, standard error; t, t-statistic; VIF, variance inflation factor. LR, Lactobacillus reuteri. Model fit: adjusted R² = 0.58, F (6, 11) = 4.86, P = 0.012. Isomer 2 of sebacic acid was excluded from the regression model due to severe multicollinearity with 9-oxononanoic acid (nearly identical CIL labeled peak intensities across all 18 samples, consistent with co eluting peak pair detection).

Table 4. Multiple linear regression of final body weight (kg) on initial body weight, Lactobacillus reuteri supplementation, and putative plasma biomarkers in ewes (n = 18).

n = 18. Variables defined as in Table 3 footnote. Initial body weight was included as a covariate to mirror the ANCOVA structure used for the primary final body weight test. Model fit: adjusted R² = 0.96, F (7, 10) = 58.01, P < 0.001. Isomer 2 of sebacic acid was excluded from the regression model due to severe multicollinearity with 9-oxononanoic acid.

Discussion

Plasma metabolomic profiling has emerged as a sensitive systemic readout of probiotic activity in livestock, capturing how microbial supplementation reshapes amino acid, lipid, and immune-metabolic pathways at the whole-animal level (Daszkiewicz et al., 2025; Jiye et al., 2005; Saleem et al., 2024; Liu, 2024; Wang et al., 2025). In the present study, LR (LR) supplementation of Katahdin ewes over a 35-day feeding trial produced concurrent improvements in growth performance and a coordinated remodeling of plasma metabolome. LR supplemented ewes had higher ADG and lower feed: gain than controls, consistent with previous reports of Lactobacillus mediated growth promotion in Ossimi lambs (El-Ashker et al., 2018), Nubian goats (Azzaz et al., 2015), and combined B. subtilis and L. casei supplementation of lambs (Khattab et al., 2020), and with comparable benefits in swine.

A striking feature of the differential plasma metabolome was that 18 of the 27 differentially abundant metabolites were lower in LR supplemented ewes than in controls, with only 9 metabolites elevated (cyanamide being the most prominent). This directional pattern does not reflect impaired biosynthesis or systemic depletion; rather, it is most parsimoniously interpreted as increased peripheral uptake and utilization of circulating substrates by metabolically active tissues, particularly skeletal muscle and liver, as substrate is drawn out of the plasma pool to support the elevated anabolic and energetic demand of the improved growth phenotype in LR supplemented ewes. This interpretation is fully consistent with the concurrent gains in ADG (+0.07 kg/d) and reduced feed to gain ratio (5.75 → 4.35) observed in the LR group and is supported by the specific tissue uptake and utilization pathways associated with each of the depleted metabolites developed in the following paragraphs. Several of the depleted metabolites have well characterized tissue uptake pathways that link the plasma metabolomic signal to nutrient absorption efficiency, growth performance, immune competence, and metabolic homeostasis (Idowu et al., 2022; Riaz et al., 2024; Chowdhury et al., 2025).

Five of the six putative biomarker metabolites, β-alanine, hydroxylated-lecithin, L-ornithine, sebacic acid, and 9-oxononanoic acid, showed lower plasma concentrations in LR ewes and converge biochemically on substrate supply for ATP generation in extra-hepatic tissues. β-alanine is the rate limiting precursor for carnosine, a dipeptide enriched in skeletal and cardiac muscle that buffers intracellular H+ generated during anaerobic glycolysis and thereby sustains ATP production during periods of high energy demand (Sale et al., 2010; Boldyrev et al., 2013). Beta-alanine is also incorporated into pantothenate and coenzyme A, an essential cofactor for fatty acid β-oxidation and the tricarboxylic acid (TCA) cycle. Sebacic acid and 9-oxononanoic acid are medium chain fatty acid derivatives that bypass the carnitine shuttle and feed directly into peroxisomal and mitochondrial β-oxidation, yielding acetyl-CoA for the TCA cycle (Mingrone et al., 1989; Reddy and Hashimoto, 2001). Hydroxylated lecithin, an oxidized phosphatidylcholine derivative, releases free fatty acids and choline upon hydrolysis, supplying additional β-oxidation substrate and supporting membrane turnover in metabolically active tissues (van der Veen et al., 2017). L-ornithine enters the TCA cycle through glutamate and α-ketoglutarate and serves as a rate limiting substrate for polyamine biosynthesis, driving cell proliferation and protein anabolism in skeletal muscle and gut epithelium (Pegg, 2009; Kalhan, 2016). Consistent with this convergent biology, dietary Lactobacillus enhances mitochondrial biogenesis and carnosine accumulation in Sunit sheep muscle (Wang et al., 2022). Taken together, the lower plasma concentrations of these five metabolites in LR ewes are consistent with a hypothesis of increased peripheral uptake for mitochondrial energy generation and polyamine-driven tissue anabolism, though direct measurement of tissue-level uptake and flux would be required to formally confirm this interpretation.

When final body weight was analyzed by multiple linear regression with initial body weight as a covariate (Table 4), 9-oxononanoic acid emerged as an independent positive predictor (Beta = 1.50, P = 0.010) alongside LR supplementation (Beta = 2.83, P = 0.003). This within group positive coefficient contrasts with the LR associated between group depletion of plasma 9-oxononanoic acid and likely reflects short term flux through unsaturated fatty acid peroxidation that scales with cumulative energy intake once treatment status is held constant. Thus, the two coefficients suggest that LR shifts the overall plasma 9-oxononanoic acid setpoint downward while heavier animals within each treatment still carry a higher steady state pool, an interpretation that will require validation in larger cohorts and, ideally, absolute quantification of 9-oxononanoic acid and paired tissue level measurements of lipid peroxidation.

The most prominent biochemical theme among the upregulated plasma metabolites was a coordinated remodeling of amino acid and nitrogen metabolism. The elevated plasma concentrations of aspartic acid, L-homoserine, 1-pyrroline-4-hydroxy-2-carboxylic acid, and an ornithine isomer in LR ewes are consistent with a hypothesis of enhanced microbial protein synthesis and portal amino acid supply, which would require paired ruminal, portal, and peripheral plasma sampling for direct confirmation. Aspartic acid occupies a central metabolic node as a urea cycle substrate, a transamination partner in gluconeogenesis, and a precursor for asparagine and pyrimidine biosynthesis. Its upregulation alongside L-homoserine, an obligate precursor for threonine and methionine biosynthesis, indicates enhanced systemic availability of glucogenic and essential amino acid precursors. The pathway level enrichment of cysteine and methionine metabolism and lysine degradation in LR ewes (Figure 7) further confirms this interpretation. Notably, the reduction of the free L-ornithine/D-ornithine pool in LR ewes, along with the enrichment of arginine and polyamine related pathways, is consistent with active redirection of ornithine flux toward polyamine biosynthesis via ornithine decarboxylase and toward proline synthesis via ornithine aminotransferase, each supporting cell proliferation, gut mucosal integrity, and collagen deposition (Wu et al., 2022). These observations parallel the enrichment of arginine biosynthesis and histidine metabolism reported in feed efficient beef steers (Taiwo et al., 2022), suggesting a conserved metabolic signature of enhanced nitrogen utilization efficiency across ruminant species and nutritional interventions.

Furthermore, the elevation of 3-hydroxypicolinic acid in LR ewes, together with pathway level enrichment of tryptophan metabolism, points to activation of the kynurenine immune-metabolic axis. The kynurenine pathway is the pre-dominant route of tryptophan degradation in mammals and a central immunoregulatory hub. Its metabolites, including kynurenine, 3-hydroxyanthranilic acid, and picolinic acid, modulate T cell proliferation, dendritic cell maturation, and macrophage polarization through aryl hydrocarbon receptor (AhR) signaling (Wang et al., 2025). The elevated 3-hydroxypicolinic acid signal is consistent with LR supplementation engaging this immunomodulatory cascade, either through direct microbial production of tryptophan catabolites or through enhanced tryptophan absorption from the gut; targeted validation of the kynurenine pathway intermediates and paired microbiome data will be required to confirm the mechanism. This finding aligns with the broader literature on Lactobacilli as modulators of AhR signaling and provides a plausible mechanistic link between LR supplementation and the improved growth phenotype, though we did not directly measure downstream immune or growth-regulatory outputs of this pathway in the present study; additional targeted assays of kynurenine pathway intermediates and immunometabolic markers would be required to substantiate a direct causal link.

Two additional features of the differential plasma metabolome reflect direct microbial host modulation by LR. First, the downregulation of tyramine, a biogenic amine produced by microbial decarboxylation of tyrosine, may reflect suppression of tyramine producing bacterial populations in the rumen through competitive exclusion or reuterin mediated antimicrobial activity, although direct confirmation would require paired ruminal 16S rRNA or shotgun metagenomic profiling. Tyramine accumulation is associated with gastrointestinal dysfunction and inflammatory signaling, and its reduction would lower the biogenic amine load on host monoamine oxidase detoxification. To our knowledge, this biogenic amine reduction has not previously been reported in the context of probiotic supplementation in sheep. Second, the depletion of Nocardicin A and Isonocardicin A, naturally occurring monobactam antibiotics produced by Nocardia species, is consistent with, but does not by itself demonstrate, reuterin mediated suppression of competing antibiotic producing actinobacteria in the gut microbial community. This interpretation aligns with the documented broad spectrum antimicrobial activity of LR derived reuterin (Schaefer et al., 2010; Yang et al., 2024) and should be regarded as a reasonable working hypothesis pending direct microbiome verification. This observation has potential implications for antimicrobial resistance dynamics, since reduced microbial antibiotic biosynthesis lowers selection pressure for AMR genes within the gut. Concurrent rumen 16S rRNA or shotgun metagenomic data will be required to formally test this interpretation by quantifying Nocardia and other actinobacterial taxa. Until such data are available, all microbial mediated mechanisms proposed above are best considered reasonable hypotheses generated by the plasma metabolomic signal.

The depletion of 9-oxononanoic acid and isomer 2 of sebacic acid in LR ewes is also consistent with reduced systemic lipid peroxidation following probiotic supplementation. 9-Oxononanoic acid is a terminal aldehydic product generated during oxidative cleavage of unsaturated fatty acids and serves as a direct indicator of reactive oxygen species mediated membrane lipid damage (Demirci-Cekic et al., 2022), and sebacic acid is a medium chain dicarboxylic acid produced under conditions of mitochondrial stress. The concurrent reduction of both metabolites indicates a shift in systemic redox balance toward a protective state, which acts alongside the peripheral utilization framing developed for the biomarker set. Cyanamide (FC = 1.66), the single metabolite markedly elevated in LR ewes and a strong individual putative biomarker, may originate from several sources in the ruminant that are not mutually exclusive: (i) reduced microbial degradation of a cyanamide precursor pool under LR mediated remodeling of the rumen microbial community, (ii) altered ruminal or intestinal microbial nitrogen metabolism via cyanogenic and thiourea related pathways that yield cyanamide as an intermediate, (iii) shifts in host and microbial turnover of urea and cyanate derivatives following changes in ruminal ammonia recycling under probiotic supplementation, or (iv) a direct contribution from LR nitrogen metabolism, which cannot be excluded. Each of these routes is compatible with the observed positive correlation between plasma cyanamide and ADG (ρ = 0.67, P < 0.01) but remains a reasonable working hypothesis that will require targeted absolute quantification by external standard calibration, paired ruminal and portal sampling, and 16S rRNA or shotgun metagenomic profiling of the ruminal microbial community for formal confirmation.

Four metabolic pathways, cysteine and methionine metabolism, tryptophan metabolism, primary bile acid biosynthesis, and lysine degradation, were enriched in LR ewes, relative to control. Cysteine and methionine metabolism produces several metabolites including methionine, S-adenosylmethionine (SAM), cysteine, taurine, and glutathione, with the upregulated plasma L-homoserine in LR ewes serving as an obligate precursor for both methionine and threonine biosynthesis (Wu, 2014). Methionine and SAM are essential methyl donors for choline, creatine, carnitine, and phospholipid synthesis, while cysteine and glutathione provide systemic antioxidant defense and redox regulation (Brosnan and Brosnan, 2006; Lu, 2013). LR itself carries the full enzymatic machinery for cysteine biosynthesis, methionine sulfoxide reductase activity, and glutathione handling (Saulnier et al., 2011), and convergent evidence from livestock and human studies supports a probiotic driven modulation of this pathway. In weaned piglets, LR supplementation elevated hepatic taurine and remodeled gut-liver amino acid metabolism (He et al., 2025), and in growing-finishing pigs it raised glutathione peroxidase activity while lowering malondialdehyde (Hou et al., 2015). In ruminants, dietary Lactobacillus enhanced methionine supply in dairy cows, increased plasma trans-sulphuration intermediates including cystathionine, cystine, and taurine (Coleman et al., 2017). In fact, a recent meta-analysis further reported that probiotic supplementation consistently lowers plasma malondialdehyde and raises glutathione-dependent antioxidant capacity across human cohorts (Heshmati et al., 2023). The pathway-level enrichment of cysteine and methionine metabolism in LR ewes is consistent with a cross-species probiotic signature of sulfur amino acid handling; however, because we did not directly assay antioxidant enzyme activities, glutathione redox status, or methyl donor flux in the present study, the biological interpretation of this pathway signal remains hypothesis generating and awaits targeted functional assays.

Tryptophan is an essential amino acid, serving as a precursor for the synthesis of microbial and host metabolites (Alkhalaf and Ryan, 2015). Tryptophan metabolism generates kynurenine, 3-hydroxyanthranilic acid, picolinic acid, and 3-hydroxypicolinic acid (elevated in LR ewes), several of which act as immunomodulatory ligands of the aryl hydrocarbon receptor and influence T cell proliferation, dendritic cell maturation, and macrophage polarization (Cervenka et al., 2017; Roager and Licht, 2018). In addition to the well-characterized indole pathway, through which LR produces indole-3-aldehyde and indole-3-acetic acid as potent aryl hydrocarbon receptor ligands driving host immunomodulation (Zelante et al., 2013; Cervantes-Barragan et al., 2017), our study identifies the kynurenine pathway as a parallel tryptophan route engaged during LR supplementation, evidenced by elevated plasma 3-hydroxypicolinic acid in LR ewes. These two pathways converge on aryl hydrocarbon receptor signaling and immune competence, indicating that LR reshapes systemic tryptophan metabolism through complementary microbial and host routes.

Primary bile acid biosynthesis produces cholic and chenodeoxycholic acids, which solubilize dietary lipids, regulate hepatic and intestinal cholesterol homeostasis, and signal through FXR and TGR5 receptors to modulate host energy balance and immune function (Chiang, 2013; Wahlstrom et al., 2016). Therefore, enrichment of this pathway in LR ewes is consistent with the concurrent reduction of plasma hydroxylated lecithin via enhanced hepatic phospholipid processing and bile mediated excretion. Lysine degradation generates saccharopine, α-aminoadipic acid, and crotonyl-CoA, feeding into acetyl-CoA for mitochondrial energy metabolism and supplying substrate for carnitine biosynthesis (Hallen et al., 2013; Leandro and Houten, 2020); its enrichment in LR ewes is consistent with the broader rerouting of amino acid catabolism toward microbial protein synthesis and host nitrogen efficiency reflected in the upregulated aspartic acid, L-homoserine, and ornithine isomer signals. Together, these four enriched pathways integrate the biomarker level observations into a coherent systemic response, linking nitrogen efficiency, redox and immune balance, and lipid handling in LR supplemented ewes.

Comparable metabolic signatures have been documented across livestock species supplemented with LR. In weaned piglets, LR elevated creatine, ornithine, citrulline, hydroxyproline, and hepatic taurine in the gut-liver axis, with 27 amino acid related metabolites differentially abundant in portal plasma (Yi et al., 2019; He et al., 2025) and enriched primary bile acid biosynthesis in a LR treated alcoholic liver disease model (Zheng et al., 2020). Furthermore, LR preferentially synthesizes the aryl hydrocarbon receptor active tryptophan metabolite kynurenic acid through dedicated microbial enzymatic machinery (Schwarcz et al., 2024). In small ruminants, plasma metabolomics has linked lamb feed efficiency to amino acid pool divergence, citrate, and malate concentrations, with protein turnover and mitochondrial energy production identified as the principal discriminating processes (Touitou et al., 2022). Rumen metabolome analysis of sheep with contrasting ADG has further identified tyrosine metabolism, β-alanine metabolism, and thiamine metabolism among the most differentially enriched pathways (Zheng et al., 2025), corroborating the β-alanine signal observed in the present LR ewe cohort. These findings align with a broader feed efficiency literature in beef cattle, where amino acid metabolism related metabolites including creatine, tyrosine, glycine, glutamine, ornithine, aspartate, lysine, and valine have been associated with feed efficiency phenotypes (Karisa et al., 2014; Clemmons et al., 2017; Jorge-Smeding et al., 2019), and where histidine metabolism and arginine biosynthesis enrichment have been identified in feed efficient beef steers (Taiwo et al., 2022).

Strengths

The present study has several methodological strengths that support the reliability of the reported plasma metabolomic signal. First, animals were assigned to treatments in a completely randomized design with balanced replication (n = 9 per group), preceded by a 21-day acclimation period to stabilize feed intake and reduce residual handling stress. Second, growth performance was analyzed by analysis of covariance with initial body weight as a covariate to control for baseline variation, and the homogeneity of regression slopes assumption was formally tested before fitting the reduced model. Third, plasma metabolomic profiling was conducted using chemical isotope labeling liquid chromatography mass spectrometry, a platform that expands submetabolome coverage of the amine, phenol, carboxylic acid, and carbonyl fractions and improves quantitative comparability of low abundance metabolites through differential isotope labeling. Fourth, we applied an integrated multivariate and univariate pipeline (PLS-DA with permutation testing, volcano analysis with raw P value thresholding, KEGG pathway enrichment with topology weighting, ROC and random forest biomarker analysis, and multiple linear regression with variance inflation factor screening) so that biomarker candidates were nominated only when supported by convergent evidence across analyses. Fifth, correlation and regression analyses linked biomarkers directly to production traits, moving beyond descriptive metabolomics toward phenotype anchored biomarker discovery.

Limitations and future directions

Several limitations should temper interpretation of the present findings. First, plasma samples collected on days 7, 14, 21, 28, and 35 were pooled in equal volumes for each ewe prior to CIL/LC-MS analysis. While this design effectively captures the cumulative systemic metabolic characteristics following 35 days of LR intervention and is well suited to identify stable dietary signatures, it necessarily obscures the longitudinal trajectory of the metabolic response and precludes analysis of the temporal dynamics, kinetic phasing, or possible transient peaks of the plasma metabolome during the intervention; time resolved sampling with individual timepoint analysis is a planned future direction of our research program. Second, the study did not include direct 16S rRNA or shotgun metagenomic sequencing of the ruminal or intestinal microbial community, and all discussion in the present text regarding microbiota mediated mechanisms should therefore be regarded as reasonable working hypotheses generated by the plasma metabolomic signal rather than as demonstrated mechanisms; formal verification will require future integrative studies combining rumen and rectal microbiome sequencing with plasma metabolomics in matched cohorts. Third, all differential abundance and biomarker analyses are based on non-targeted relative quantification of CIL labeled peak pair intensities; for practical application in precision livestock feeding, targeted absolute quantification with authenticated external standards and independent validation in larger multi breed cohorts will be required before deployment as a field diagnostic. Fourth, the reported biomarker-phenotype Spearman correlations were computed across the pooled CON and LR samples (n = 18) and therefore may partly reflect treatment-driven separation between groups; within-group correlation testing was not formally reported because the reduced within-group sample size (n = 9 per group) provides limited statistical power for correlation significance, and formal within-group biomarker-phenotype validation is deferred to larger cohort follow-up work. Fifth, differential abundance was assessed using nominal Welch’s P-values (P < 0.05) without per-metabolite correction for multiple comparisons across the 1,003 tested metabolites. As a sensitivity analysis, Benjamini-Hochberg correction applied across the full 1,003-metabolite set indicated that no individual metabolite reaches q < 0.10 at the present sample size of n = 9 per group (smallest q = 0.707). Consistent with this, and with the permutation-test result for the PLS-DA (P = 0.683), the metabolomics findings are presented throughout as an exploratory, hypothesis-generating characterization of the plasma metabolic response to LR supplementation rather than as confirmatory identification of individual biomarkers. The biological interpretation rests on the convergence of five independent lines of evidence (univariate volcano-filtered differential abundance, VIP-based feature ranking, ROC and random-forest biomarker discovery, phenotype correlations, and biological plausibility) rather than on any single metabolite’s adjusted significance; validation in an independent, larger, multi-breed cohort using targeted, absolutely quantified assays is required before mechanistic or diagnostic claims can be advanced.

In summary, dietary LR supplementation improved growth performance and feed efficiency in Katahdin ewes while inducing distinct changes in the plasma metabolome. These metabolic alterations were characterized by differential abundance of metabolites involved in amino acid, lipid, and bile acid metabolism, with enrichment of cysteine and methionine metabolism, tryptophan metabolism, primary bile acid biosynthesis, and lysine degradation. Six metabolites were identified as candidate biomarkers of LR supplementation, with hydroxylated lecithin showing the strongest single-metabolite bivariate correlation with the growth and feed efficiency phenotype, although it did not reach independent significance in the multivariable regression model. Taken together, these findings demonstrate that plasma metabolomics provides valuable insight into the systemic metabolic responses to probiotic supplementation and identifies potential metabolic indicators of improved animal performance. The differential plasma metabolites and biomarker panel reported here were identified by non-targeted CIL/LC-MS relative quantification; for practical application in precision livestock feeding, targeted absolute quantification with authenticated external standards and independent validation in larger, multi breed sheep populations spanning different production stages and geographic regions is warranted.

Generative AI statement

The author(s) declared that generative AI was not used in the creation of this manuscript.

Any alternative text (alt text) provided alongside figures in this article has been generated by Frontiers with the support of artificial intelligence and reasonable efforts have been made to ensure accuracy, including review by the authors wherever possible. If you identify any issues, please contact us.

Publisher’s note

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.

Supplementary material

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fanim.2026.1926978/full#supplementary-material

Supplementary material Supplementary material at the source

Supplementary material Supplementary material at the source

Supplementary material Supplementary material at the source

Supplementary material Supplementary material at the source

Supplementary material Supplementary material at the source

Supplementary material Supplementary material at the source

Supplementary material Supplementary material at the source

Footnotes

Edited by: Peter Erickson, University of New Hampshire, United States

Reviewed by: Ankita Jena, Fonterra (New Zealand), New Zealand

Feifei Yang, Gansu Agricultural University, China