Temporal dynamics of ruminal metabolome and fermentation in response to molasses-treated Guinea grass–cowpea silage Authors: Armess Prince Gynth Sossou, Masato Yayota Published: 2026 ABSTRACT Highlights•Molasses reduced the effective degradability of crude protein from 53.7% to 47.6%.•Ruminal metabolites mostly accumulated within the first 4 h of fermentation.•Indole-3-acetate abundance decreased markedly under molasses (log2FC = –9.06).•Molasses supplementation moderated lipid catabolism during ruminal fermentation.•Molasses promoted carbohydrate- and antioxidant-related metabolic responses.Keywords: Tropical forage, Guinea grass – cowpea, Molasses, In vitro fermentation, Ruminal metabolome Introduction Forage, the primary component of ruminant diets worldwide, supplies most of the energy, protein, and structural carbohydrates necessary for ruminants' optimal productivity (Boval & Dixon, 2012; Hatfield & Klascheur, 2020). This is particularly evident in tropical regions (Rudel et al., 2015), where animal feeding systems rely heavily on native or cultivated forage. However, despite their high biomass yield, tropical grasses are characterized by a relatively low water-soluble carbohydrate (WSC) content, which remains a key obstacle to successful ensiling because WSC are vital substrates for lactic acid bacteria (LAB). This, coupled with their low crude protein (CP) and high fiber contents (King et al., 2012; Franco et al., 2017), often results in poor fermentation and, subsequently, reduced nutrient utilization in animals. Combining grasses with forage legumes has proven effective in enhancing the nutritional quality of ruminant diets (Di Miceli et al., 2023; Lai et al., 2023). Supplementation with fermentable additives such as molasses has also been widely used to increase WSC availability, thereby enhancing lactic acid (LA) production, accelerating pH decline, and reducing proteolysis during ensiling (Muck et al., 2018). Moreover, molasses- or inoculant-treated silages have shown reduced fiber fractions and ammonia nitrogen (NH3—N) accumulation (Wang et al., 2018), indicating improved fermentation quality and nutrient preservation. In the animal digestive system, the extent of fermentation achieved during ensiling directly influences ruminal degradation and fermentation kinetics (Deitmers et al., 2024), owing to changes in chemical and microbial composition. Well-fermented silages, characterized by high LA and low NH3—N levels, enhance microbial colonization and nutrient degradability (Kung et al., 2018); whereas poorly fermented silages with high pH or butyric acid suppress rumen microbial activity. Ruminal fermentation dynamics are also influenced by the nature of dietary carbohydrate sources, residual WSC, organic acids, and polyphenolic compounds (Killerby et al., 2022; Cavallini et al., 2025), which affect microbial protein synthesis and gas production. Moreover, it was established that solubility and rate of degradation are the primary determinants of the extent of ruminal degradation (Hanigan et al., 2021). They are crucial for avoiding overestimation of the nutritional value of a feed ingredient by accounting for inevitable losses during digestion and metabolism in animals (McDonald et al., 2010). Guinea grass (Panicum maximum) is a highly adaptable and productive forage grass (Carvalho et al., 2020), whereas cowpea (Vigna unguiculata) is an indigenous tropical legume forage characterized by high CP content and rapid growth. In our previous study, we investigated the effects of different Guinea grass–cowpea ratios and molasses supplementation levels on silage fermentation and ruminal responses (Sossou et al., unpublished). The 65:35 Guinea grass–cowpea ratio resulted in higher CP, organic matter (OM), and metabolizable energy (ME) contents in the silage. Consistent with the findings of Quintero-Anzueta et al. (2021), these changes were associated with higher ruminal NH3—N and branched-chain fatty acids (BCFAs) concentrations. Moreover, supplementation with 8% molasses during ensiling enhanced silage residual WSC, condensed tannin, and rutin concentrations, while improving DM and fiber degradability. Thus, the previous study identified 65:35 Guinea grass–cowpea silage supplemented with 8% molasses as the treatment yielding the most favorable silage quality characteristics and ruminal responses. However, as in many previous studies, it provided only preliminary insights into silage chemical composition and fermentation traits and was limited to evaluating conventional fermentation parameters, including ruminal volatile fatty acid production. The dynamics of nutrient degradability kinetics and microbial metabolism-associated biochemical responses were not investigated. Therefore, a more integrative approach combining temporal degradability assessment with metabolomic analysis was needed to better characterize ruminal responses to molasses-treated Guinea grass–cowpea silage. Metabolomics has emerged as a valuable approach to improving understanding of biochemical transformations occurring during ensiling and ruminal digestion (Zhang et al., 2025). By enabling the identification of biochemical pathways underlying fermentation shifts, metabolomics provides functional insights beyond conventional end-product measurements (Saleem et al., 2012). Recent findings indicate that forage type and silage additives can modify metabolic pathways associated with carbohydrate fermentation, amino acid degradation, and antioxidant metabolism during the ensiling process (Fu et al., 2023; Niu et al., 2025). These diet-associated biochemical transformations are reflected in the ruminal metabolome, which represents a biochemical interface between the diet and the rumen microbiota. Accordingly, forage type (e.g., native forage hay vs. Sudangrass) and forage-to-concentrate ratios markedly altered ruminal metabolites related to amino acids, lipids, organic acids, and carbohydrates (Zhang et al., 2017; Dai et al., 2023; Aldian et al., 2025). Yet, despite growing interest in dietary modulation of the ruminal metabolome, no studies have linked molasses-induced improvements in Guinea grass–cowpea silage to temporal metabolomic shifts in the rumen. Moreover, the relationships between these metabolomic responses, ruminal fermentation characteristics, and nutrient degradability remain poorly understood. Therefore, the present study primarily aimed to characterize temporal shifts in the ruminal metabolome associated with molasses-treated Guinea grass–cowpea silage during in vitro fermentation. A secondary objective was to evaluate the time-course effects on nutrient degradability kinetics and ruminal fermentation characteristics in relation to the observed metabolomic shifts. We hypothesized that molasses, by promoting homolactic fermentation during ensiling, would reduce proteolysis and improve carbohydrate preservation, thereby enhancing ruminal degradability and shifting microbial metabolism toward carbohydrate-dominated pathways. Materials and methods Forage establishment and silage preparation Cowpea and Guinea grass were cultivated from April to August 2024 at the Yanagido experimental Farm of the Gifu Field Science Centre, Gifu University, Japan (35° 27′ 50.59″ N, 136° 44′ 20.15″ E), over a total area of seven ares. The site is characterized by alluvial soil with an annual rainfall of 1856 mm. During the experimental period, temperatures ranged from 19 °C to 31 °C (Weather Spark, 2024). No soil amendment with organic or mineral fertilizer was applied before seeding, and irrigation was performed during the first two months of forage cultivation. The Japanese black-seed cowpea variety “Juroku sasage” was sown manually at a rate of 0.13 kg/a. At 65 days post-seeding, when approximately 50% of the plants were in bloom, the entire cowpea plant (including leaves, stems, and pods) was harvested and wilted for 20 h at ambient temperature to reduce the moisture content. To ensure that both forages could be harvested on the same day, Guinea grass was sown 15 days before cowpea using a manual broadcast spreader (Yamabiko Kyoritsu G-9, Japan) at a rate of 0.12 kg/a; and harvested 80 days post-sowing using a mower (Zenoah TRZ265-NC, Japan) when it reached a height of 1.0–1.5 m. After harvesting, they were wilted for 20 h under the aforementioned conditions. Both forages were then chopped into approximately 5 cm lengths using a chopping machine (CT-X1700, ATEX, Japan). The chopped forages were mixed in a 65:35 Guinea grass:cowpea ratio. The mixture was further treated with molasses (Hayashi Shokai Co., Ltd., Japan), selected for its high content of readily fermentable sugars and its widespread use in improving the ensiling of low-WSC tropical forages and feed utilization in ruminants (Oni et al., 2014; Rashid et al., 2025). It was applied at either 0% or 8% on a fresh matter basis, resulting in two treatments, GC and GCM, each prepared in quadruplicate. The 8% inclusion level, previously reported in tropical forage and ruminant nutrition studies (Tjandraatmadja et al., 1994; Broderick & Radloff, 2004), was selected based on our previous study (Sossou et al., unpublished), in which it yielded the most favorable silage quality among the tested supplementation levels. Molasses was dissolved in 1–2 L of warm water and evenly sprayed onto the forage mixture with continuous mixing in GCM, whereas the same volume of warm water was applied to GC. The silage was packed into 5 kg polyethylene bags by hand compaction, vacuumed-sealed (Kyocera VC-1200, Japan), placed in plastic containers, and stored at 25 °C for 60 days. On day 60, the silage samples were collected for laboratory analyses. In vitro ruminal fermentation, degradability, and gas production kinetics All animal procedures were approved by the Committee for Animal Research and Welfare of Gifu University (Approval ID: AG-P-N-20,240,057 and 20,250,122). Two experimental treatments (GC and GCM) were prepared in quadruplicate, with each independent replicate considered a biological replicate and used as the statistical unit for analysis. At each incubation time point, each biological replicate was incubated in triplicate in 100 mL glass bottles under anaerobic conditions, and these triplicates served as technical replicates. For the 72 h nutrient degradability test, 500 mg of sample was incubated in 40 mL of mixed buffer–ruminal fluid (2:1, v/v) (Makkar et al., 1995). Gas production and estimated ME were evaluated by incubating 200 mg of sample in 30 mL of buffered ruminal fluid. Blank bottles containing only buffered ruminal fluid without substrate were also included for various corrections. Ruminal fluid was collected from four crossbred Saanen × Shiba wethers (41 ± 2 kg body weight) after 15 days of dietary adaptation. The goats were fed 1100 g of Sudan grass hay (as-fed basis), 180 g of alfalfa hay (as-fed basis), and 120 g of steam-flaked corn (as-fed basis) per head per day according to the NRC (2007) requirements for goats. They had free access to licking stones and water. Approximately 250 mL of rumen fluid was obtained from each goat 3 h after the early-morning feeding using an esophageal tube (Tian et al., 2025). To minimize potential saliva contamination, about 20 mL of the initial portion of the collected fluid was discarded, and only the subsequent fraction was retained as inoculum. It was then filtered through four layers of cheesecloth, gently mixed, transferred to a pre-warmed thermos flask, and immediately transferred to the laboratory. McDougall’s solution was used as the buffer and was prepared as described by Makkar et al. (1995). Anaerobic conditions were ensured by continuously flushing the culture-buffered rumen fluid with CO2 until it was dispensed into the bottles. Thereafter, the bottles were sealed with rubber stoppers and aluminum caps and incubated in shaking water baths at 39 °C and 100 oscillations min⁻¹. The bottles were removed after 4, 8, 12, 24, 48, and 72 h of incubation and immediately chilled on ice to stop fermentation. The pH was measured immediately using a glass-electrode pH meter (MP-220, Mettler-Toledo AG, Switzerland). The contents of the glass bottles were moved into 50 mL polyethylene test tubes and centrifuged for 10 min at 4 °C and 3000 × g. Supernatants from technical replicates within each biological replicate were pooled in equal volumes, transferred to 15-mL polyethylene tubes, and stored at −80 °C for metabolomics analysis and at −20 °C for other analyses. Similarly, solid residues from technical replicates were pooled by weight, filtered through sintered silica crucibles, and dried at 60 °C for 48 h in an air-forced dry oven. In vitro dry matter degradability (IVDMD), in vitro organic matter degradability (IVOMD), and in vitro crude protein degradability (IVCPD) were calculated as the loss of DM, OM, and CP weight after the in vitro assays. Gas production during the assay was simultaneously recorded as described previously (Sossou et al., 2025). Cumulative gas production [1] and nutrient degradability [2] data were fitted to the non-linear equations as follows: (McDonald, 1981) where Y is the cumulative gas production at time t (h), B is the asymptotic gas production (mL g⁻¹ DM), C is the rate constant of gas production (h⁻¹), and L is the discrete lag time (h). The half-life of the asymptotic gas production, T (h), was calculated as ln2/C. Effective gas production (EGP, mL g⁻¹ DM) and average gas production rate (AGPR, mL h⁻¹) were calculated using the following equations: (Ørskov & McDonald, 1979) where y is the degradation (%) at time t (h), a is the soluble degradable fraction (%), b is the insoluble but degradable fraction (%), and c is the rate constant of the degradation of b (h⁻¹). Given that 0 h washing losses were not measured, the a-fraction was estimated directly from the fitted model parameters. The effective degradability (ED in %) and potential degradability (PD in %) were calculated using the following equations: where k is the outflow rate from the rumen, which was assumed to be 5% h⁻¹. Carbon dioxide (CO2) and methane (CH4) emissions during the in vitro assay were estimated using the following prediction equations (Wolin, 1960): where A, B, and P are the molar proportions of acetate, butyrate, and propionate, respectively, and V is the volume of buffered ruminal fluid. LC-MS/MS analysis Untargeted metabolomics, individual flavonoids and organic acids in silage, and short-chain fatty acids (SCFAs) concentrations in the buffered ruminal fluid were analyzed using ultra-performance liquid chromatography (UPLC) coupled with quadrupole time-of-flight mass spectrometry (XEVO QToF-MS; Waters Corp., Milford, USA). For untargeted metabolomics, frozen rumen fluid samples corresponding to each biological replicate were thawed at 4 °C and analyzed individually. Five quality control (QC) samples were prepared separately by pooling equal aliquots from all rumen fluid samples to monitor instrument stability and analytical reproducibility throughout the analysis. An amount of 1 mL sample, QC, or blank was spiked with 20 µL of myricetin used as internal standard. 1980 µL of LC–MS-grade methanol was added to the sample, with the mixture vortexed vigorously for 20 min, centrifuged at 16,000 × g and 4 °C for 15 min, and the supernatant carefully collected for analysis. Individual flavonoids quantification was performed using a calibration curve generated from standard solutions of quercetin, kaempferol, luteolin, and rutin (Fujifilm Wako, Japan) in the range of 5–0.001 µM. For quantification, 200 mg of fresh silage was ground in liquid nitrogen (N), extracted with 4 mL of HPLC-grade methanol, heated at 70 °C for 2 h, centrifuged at 3660 rpm for 2.5 min, and filtered through a 0.22 μm polypropylene membrane. The chromatographic separation was achieved using a Waters ACQUITY BEH C18 column (100 mm × 2.1 mm, 1.7 µm) at 40 °C. The mobile phase consisted of liquid chromatography-mass spectrometry (LC-MS)-grade 0.1% formic acid in water (A) and acetonitrile (B) (Fujifilm, Wako, Japan) at a flow rate of 0.4 mL/min. The gradient began at 95% (A), decreased to 80% at 3 min, decreased to 5% at 9 min, was held for 13 min, returned to 95% at 13.10 min, and ended at 16 min. Electrospray ionization (ESI) was performed in both positive and negative modes for untargeted metabolomics, and in the negative mode only for flavonoids, with low and high collision energies of 6 V and 15–45 V, respectively. The capillary, sampling cone, and extraction cone voltages were 3.0 kV, 35, and 4.0, respectively. The source and desolvation temperatures were set at 120 °C and 350 °C, respectively. N gas was used as the cone and desolvation gas at 50 and 1000 L/h, respectively. Argon was used as the collision gas. Organic acids in the silage and individual ruminal SCFAs were analyzed following the previously described LC-MS/MS procedure, with specific modifications detailed below. Samples were extracted into 1.5 mL tubes, centrifuged, derivatized, and analyzed as described by Aldian et al. (2025). Quantification was performed using calibration curves generated from mixed standards of acetate, propionate, butyrate, isobutyrate, valerate, and isovalerate (5–0.005 mM), all purchased from Fujifilm Wako, Japan. Chromatographic separation was performed using a mobile phase consisting of acetonitrile (B) and 0.1% formic acid in water (A), at a flow rate of 0.35 mL/min. The gradient program began at 85% (A) and was maintained for 2 min. It was then gradually reduced to 45% at 11 min, followed by a rapid decrease to 0% at 11.01 min, where it remained constant until 12 min. At 12.01 min, (A) increased to 85% and was maintained at this level for 15 min. ESI was applied in negative mode, with the sampling cone voltage adjusted to 25. Mass spectrometry detection was performed in the MSE centroid full-scan mode with an m/z range of 50–300. Chromatographic data were processed using the QuanLynx module of the MassLynx v4.2 software (Waters Corp., Milford, USA). All UPLC QToF-MS/MS measurements were calibrated using leucine-enkephalin as the reference compound ([M+H]+ = 278.1141 and 556.2771; [M+H]– = 236.1035 and 554.2615), infused at a flow rate of 10 µL/min. Real-time automatic adjustment of the m/z values ensured high mass accuracy across all detected ions. The sensitivity and reliability of flavonoid and SCFAs quantification were assessed by determining the statistical limits of detection (LOD) and quantification (LOQ). These were calculated from calibration curves using the mean slope (S) and the standard deviation of the intercept (σ), applying the formulas: LOD = 3.3 × σ/S and LOQ = 10 × σ/S. Proximate analysis and plant secondary metabolites (PSMs) DM, OM, CP, neutral detergent fiber (aNDFom), and acid detergent fiber (ADFom) contents of the silage samples were analyzed according to AOAC (2005) procedures for DM (930.15), OM (942.05), CP (954.01), aNDFom (2002.04; assay with heat-stable amylase and expressed without residual ash), and ADFom (973.18; expressed without residual ash). Hemicellulose content was calculated as the difference between aNDFom and ADFom as a percentage of DM. WSC was determined using anthrone sulfuric acid colorimetry (Hao et al., 2022). The ME content of the feed was estimated according to Menke’s equation (Menke et al., 1979): where GP = gas production (mL/200 mg DM at 24 h) and CP is the crude protein content in the feed (g/kg DM). To analyze the total tannin (TT) and condensed tannin (CT) contents, 1 g of the dried sample was homogenized in 20 mL of 70% acetone and shaken vigorously for 1 h. The mixture was sequentially centrifuged at 300 rpm for 40 min and 3500 rpm for 15 min. TT was measured by mixing 100 µL of the sample, standard, or blank with 1.6 mL of distilled water, 100 µL Folin–Ciocalteu reagent, and 200 µL of 20% sodium carbonate. The absorbance was read at 750 nm using a UV–vis spectrophotometer (Biospectrometer, Eppendorf, Germany). CT content was measured according to the vanillin–HCl method (Broadhurst & Jones, 1978), in which 0.1 mL of sample, standard, or blank was mixed with 0.6 mL vanillin and 0.3 mL hydrochloric acid. The solution was vortexed sequentially for 15 min, and the absorbance was read at 500 nm. Both TT and CT concentrations were calculated as milligrams of catechin hydrate equivalents per gram of dry plant material (CaE/g DM) based on a standard curve of catechin hydrate (1 mg/mL) using the following equations: (TT: Y = 3.8708x + 0.0312; R² = 0.9982) and (CT: Y = 5.7973x − 0.1623; R² = 0.9951). Aerobic stability test and silage fermentation indices Aerobic stability (AS), assessed at silo opening, was defined as the time (h) required for the silage temperature to rise by 2 °C above ambient air temperature (Wilkinson & Davies, 2013). Using a K/J/T thermometer (PB01–0001, China), thermocouple probes were inserted into the center of each silage container (Kung & Ranjit, 2001). The silos were covered with double layers of cheesecloth to allow airflow while preventing contamination. Temperature readings of both the silage and ambient air (22–23 °C) were recorded every 30 min for 120 h. Ten grams of fresh silage were mixed with 90 mL of distilled water (1:9 material–liquid ratio) in 100 mL conical flasks, sealed, and stored at 4 °C for 24 h. The extract was filtered through four layers of cheesecloth and centrifuged at 10,000 × g for 15 min at 4 °C. The pH of the filtered extract was measured using a pH meter (MP-220, Mettler-Toledo AG, Switzerland), and the concentrations of lactic, acetic, and butyric acids and NH3—N were quantified. Flieg points were calculated to assess silage quality using silage pH values and DM content. Buffering capacity (BC) was determined following Playne and McDonald (1966) and expressed as milliequivalents of alkali required to increase the pH from 4 to 6 per 100 g DM, after correction for a 150-mL water blank. NH3—N concentrations in both silage and buffered ruminal fluid were measured by the phenol–hypochlorite method (Weatherburn, 1967) using a microplate reader (Multiskan FC, Thermo Fisher Scientific) at an absorbance of 596 nm. For LAB enumeration, 10 g of fresh silage was homogenized in 90 mL of sterile water, serially diluted (10⁻¹ to 10⁻⁴) in 0.85% NaCl solution, and 100 μL of each dilution was spread onto MRS agar (Biokar, New Zealand). The plates were incubated at 30 °C for 48 h, and the results are expressed as CFU/g FM. Data processing and analyses All statistical analyses, except for untargeted metabolomics, were performed in R (version 4.4.3; R Core Team, 2024). Before analysis, the data were assessed for normality (Shapiro–Wilk test; p > 0.05) and homogeneity of variance (Bartlett's test; p > 0.05). Data for silage chemical composition and fermentation traits were analyzed using ANOVA within a mixed-model framework, with treatment included as a fixed effect and silage replicate as a random effect. The results were presented as least-squares means with their corresponding standard errors, effect estimates, 95% confidence intervals (CI), and statistical significance was declared at p < 0.05. Cumulative gas production was measured repeatedly from the same incubation bottles over time and analyzed using a mixed-model repeated-measures approach. Incubation bottle was included as a random effect, and a continuous autoregressive covariance structure [CAR(1)] was applied to account for temporal correlation among repeated observations. Nutrient degradability and ruminal fermentation indices were determined using sacrificial bottle sampling, with observations considered statistically independent, and analyzed using the following fixed-effects factorial model: where Yit is the dependent variable, µ is the overall mean, Tᵢ is the fixed effect of treatment, Ht is the fixed effect of incubation time, (T × H)it is the interaction between treatment and incubation time, and εit is the residual error. Pairwise comparisons within incubation time points were performed using Tukey-adjusted p-values. Goodness-of-fit for non-linear gas production and degradability models was assessed using the coefficient of determination (R²) and root mean square error (RMSE). The raw data obtained from LC–MS/MS for metabolomic analysis were converted to .abf format using the Reifycs Analysis Base File Converter (Reifycs, Inc., Tokyo, Japan) and processed with MSDIAL version 4.9 (Tsugawa et al., 2015). Metabolite annotation was performed using accurate mass, retention time, and MS/MS fragmentation matching against the in-house MSDIAL database, with metabolite identities further cross-referenced against the HMDB and KEGG databases. The retention time shift, mass accuracy, and relative standard deviation of the peak areas were set to 0.3 min,5 mDa, and 30%, respectively. Metabolites with a total identification score >700 were considered putatively annotated metabolites (MSI level 2). The resulting aligned data were converted into a .csv file and exported to MetaboAnalyst 5.0 (https://www.metaboanalyst.ca) for downstream analysis. Before analysis, the dataset was normalized to an internal standard, log-transformed, and scaled using Pareto scaling. Multivariate analyses included principal component analysis (PCA) and orthogonal partial least squares-discriminant analysis (OPLS-DA) to identify variables contributing to silage group separation. The OPLS-DA model performance variables R²Y and Q² were all above 0.9. Univariate analyses involved volcano plots and Student’s t-tests, with significance set at p < 0.05. Metabolite trends were compared between GC and GCM using log2 fold change (log2FC), where positive and negative log2FC values indicated higher and lower metabolite abundances under GCM relative to GC, respectively. Variable importance in projection (VIP) scores were extracted from the OPLS-DA model, and metabolites with VIP > 1.0, p 〈 0.05, and log2FC 〉 1.0 or < −1.0 were considered differentially accumulated metabolites (DAMs). Given the exploratory nature of the untargeted metabolomics approach, false discovery rate (FDR) correction was not applied (Tian et al., 2025). A total of 497 and 231 MS peaks were identified in the positive and negative ionizations, respectively, corresponding to 149 annotated metabolites. Results Chemical composition, PSMs and ensiling indices Significant differences were observed in the chemical composition and fermentation characteristics of the GC and GCM silages (Table 1). GCM exhibited higher DM (36.10 vs. 30.30%) and WSC (2.32 vs. 1.29% DM) (p < 0.05) and ME (8.35 vs. 7.52 MJ/kg DM, p < 0.001) contents than GC. In contrast, molasses supplementation lowered the aNDFom and ADFom contents of the silage (p < 0.05). The CP and hemicellulose contents did not differ significantly between the treatments. In addition, both treatments showed no significant differences (p > 0.05) in TT, CT, or individual flavonoid contents, except for quercetin, which was lower in GCM (12.16 vs. 54.11 mg/kg DM; p = 0.001). Fermentation indices revealed a significantly higher lactic-to-acetic acid (LA/AA) ratio, lower pH, and lower acetic acid and NH3—N contents (p < 0.05) in GCM than in GC. Molasses supplementation did not affect LAB counts, BC, or LA content between treatments. AS was markedly improved in GCM (45.00 h vs. 12.33 h in GC, p < 0.001). A slight, but significant, reduction in silage temperature was also noted in GCM. The chemical composition and fermentation-related indices of Guinea grass and cowpea prior to ensiling are presented in Supplementary Material S1. Table 1Chemical composition, plant secondary metabolites, and ensiling traits of molasses-treated Guinea grass–cowpea silage.Table 1 dummy alt textGCGCMSEMa)95% Confidence Intervalp-ValueSilage chemical compositionDM (%)30.3036.101.4685.80 [1.54, 10.1]0.019OM (% DM)88.0588.990.2670.95 [−0.07, 1.96]0.062aNDFom (% DM)51.9848.540.874−3.40 [−6.03, −0.84]0.021ADFom (% DM)35.9033.690.521−2.20 [−3.25, −1.17]0.004CP (% DM)11.2910.840.132−0.45 [−0.98, 0.08]0.076b)ME (MJ/kg DM)7.528.350.1870.83 [0.63, 1.01]<0.001WSC (%DM)1.292.320.2491.03 [0.44, 1.62]0.008Hemicellulose (%DM)16.0814.850.443−1.23 [−3.39, 0.93]0.189Plant secondary metabolites in silageTT (mg CaE/g DM)9.079.640.3270.57 [−1.29, 2.44]0.443CT (mg CaE/g DM)1.221.300.0250.08 [−0.02, 0.19]0.095Rutin (mg/kg DM)3.574.470.5020.90 [−195, 3.76]0.429Quercetin (mg/kg DM)54.1112.169.679−42.0 [−56.8, −27.1]0.001Luteolin (mg/kg DM)23.8728.062.3954.20 [−9.49, 17.9]0.442Kaempferol (mg/kg DM)4.423.240.602−1.18 [−4.55, 2.19]0.386Silage fermentation traitsLAB (log10 CFU/g FM)6.166.150.080−0.008 [−0.52, 0.49]0.965AS (h)12.3345.007.37932.7 [26.2, 39.1]<0.001Flieg Score71.7389.734.22118 [10.1, 25.9]0.003BC (mEq NaOH/100 g DM)34.2830.081.206−4.2 [−8.89, 0.49]0.068Acetic Acid (g/kg DM)24.6918.021.596−6.67 [−10.2, −3.16]0.006Lactic Acid (g/kg DM)75.5786.393.33410.8 [−3.42, 25.10]0.103LA/AA3.064.810.2741.75 [0.98, 2.52]0.003pH4.854.690.039−0.16 [−0.26, −0.06]0.010NH3-N (g/kg DM)4.002.700.300−1.3 [−1.77, −0.83]0.002Temperature ( °C)25.1024.870.054−0.23 [−0.33, −0.14]0.002GC, Guinea grass–cowpea; GCM, Guinea grass–cowpea treated 8% molasses; DM, dry matter; OM, organic matter, aNDFom, ashless neutral detergent fiber treated with amylase; ADFom, ashless acid detergent fiber; CP, crude protein; ME, metabolizable energy; WSC, water soluble carbohydrate; LAB, lactic acid bacteria; AS, aerobic stability; BC, buffering capacity; LA/AA, lactic-to-acetic acid ratio; TT, total tannin; CT, condensed tannin; SEM, standard error of mean. a)Values outside the brackets represent the estimated treatment difference, and values within brackets represent the 95% confidence interval of the estimated treatment difference. b)ME, metabolizable energy was calculated using the formula of Menke et al. (1979). Chemical composition, plant secondary metabolites, and ensiling traits of molasses-treated Guinea grass–cowpea silage. GC, Guinea grass–cowpea; GCM, Guinea grass–cowpea treated 8% molasses; DM, dry matter; OM, organic matter, aNDFom, ashless neutral detergent fiber treated with amylase; ADFom, ashless acid detergent fiber; CP, crude protein; ME, metabolizable energy; WSC, water soluble carbohydrate; LAB, lactic acid bacteria; AS, aerobic stability; BC, buffering capacity; LA/AA, lactic-to-acetic acid ratio; TT, total tannin; CT, condensed tannin; SEM, standard error of mean. a)Values outside the brackets represent the estimated treatment difference, and values within brackets represent the 95% confidence interval of the estimated treatment difference. b)ME, metabolizable energy was calculated using the formula of Menke et al. (1979). Microbial fermentation indices and degradability kinetics Radar plots revealed that most differences in ruminal SCFAs and BCFAs between treatments occurred during the early incubation phase (4–12 h), whereas fermentation profiles became progressively more similar at 24 h (Figs. 1A and 1B). At 4 h, GCM exhibited higher total SCFAs production and greater concentrations of acetate, butyrate, valerate (p < 0.05), and propionate (p < 0.01) than GC, indicating more rapid early fermentation. However, from 8 h onward, SCFAs concentrations were largely comparable between treatments, except for a higher valerate concentration in GC at 12 h (3.72 vs. 3.32 mM; p < 0.05). Similarly, GCM showed higher isobutyrate concentration at 4 h (p < 0.05), whereas total and individual BCFAs concentrations became higher in GC at 8 h and 12 h (p < 0.05). SCFAs and BCFAs are further presented in table format in Supplementary Material S2. Ruminal pH gradually declined over time in both groups, reflecting progressive fermentation acidification, although GCM maintained slightly higher pH values throughout the incubation period, with significant differences observed at 4 h and 24 h. (p < 0.05) (Fig. 1C). Ruminal NH3—N concentrations peaked at 8 h in GC and remained higher than those of GCM at 12 h and 24 h (p < 0.01) (Fig. 1D). Both estimated CO2 and CH4 concentrations increased throughout incubation, with higher values observed in GCM during the initial 4 h (p < 0.05), whereas treatment differences became less pronounced at later incubation times (Fig. 1E and F). Microbial fermentation indices, including short-chain fatty acids expressed in mM (A), branched-chain fatty acids in mM (B), rumen pH (C), ammonia nitrogen in mg/dL (D), carbon dioxide emission (E) and methane emission in mM (F) of untreated Guinea grass–cowpea (GC) vs. the 8% molasse treated (GCM) at different time intervals. The lines and pentagons marked in blue represent the fermentation of GC, and those marked in red represent the fermentation of GCM. Significance was defined using asterisks (* p < 0.05 / ** p < 0.01). The degradability of DM, OM, and CP (Fig. 2A, B, C), along with the nutrient degradability kinetics (Table 2), was significantly influenced by molasses supplementation. DM and OM degradability increased steadily and were significantly higher in GCM at all incubation times, with the most pronounced differences observed at 4 h, 24 h, and 48 h (p < 0.001). In contrast, CP degradability exhibited a distinct pattern, with the highest values recorded in GC, and the differences were most pronounced at 8 h, 12 h, and 48 h (p < 0.001). Regarding DM disappearance kinetics (Table 2), ED was significantly higher in GCM than in GC (51.66 vs. 48.52%; p < 0.001), along with PD (82.77 vs. 81.41%; p = 0.021). The soluble a-fraction was numerically higher in GCM (11.74 vs. 9.00%; p = 0.055), whereas the degradable insoluble b-fraction and degradation rate c were unaffected by molasses. Greater ED (49.48 vs. 45.72%; p = 0.011), coupled with a higher soluble a-fraction (6.79 vs. 3.26%; p = 0.023), was noted for OM disappearance kinetics in GCM, whereas the b-fraction was numerically lower (75.79 vs. 77.76%; p = 0.055). For CP kinetics, ED was significantly lower in GCM (47.57 vs. 53.72%; p < 0.001), despite a higher PD (90.66 vs. 84.54%; p = 0.005). Molasses lowered CP-a (24.40 vs 30.86%; p < 0.001) and sharply decreased the degradation rate constant c (0.027 vs. 0.037 h⁻¹; p = 0.004), but enhanced CP-b (p < 0.001). Cumulative gas production was significantly higher in GCM at all fermentation times (Fig. 2D). GCM exhibited higher asymptotic gas production (189.45 vs. 177.03 mL g⁻¹ DM), faster production rate (0.058 vs. 0.048 h⁻¹), higher EGP (101.96 vs. 86.85 mL g⁻¹ DM) and AGPR (7.89 vs. 5.72 mL h⁻¹), and shorter lag time (0.18 vs. 1.09 h) (Table 2). Dry matter (DM) degradability in % (A), organic matter (OM) degradability in % (B), crude protein (CP) degradability in % (C), and cumulative gas production in mL/g DM (D) of untreated Guinea grass–cowpea (GC) vs. the 8% molasse treated (GCM) at different time intervals. The lines marked in blue represent the degradability/cumulative gas of GC, and lines marked in red represent that of GCM. Significance was defined using asterisks (* p < 0.05 / ** p < 0.01 / *** p < 0.001). Table 2Nutrients degradability and cumulative gas kinetics of molasses-treated Guinea grass–cowpea silage.Table 2 dummy alt textGCGCMSEMa)95% Confidence Intervalp-Value72h nutrients degradability kineticsDMED (%)48.5251.660.6223.14 [2.15, 4.13]<0.001PD (%)81.4182.770.3281.36 [0.29, 2.44]0.021a (%)9.0011.740.7322.70 [−0.08, 5.48]0.055b (%)72.3871.010.437−1.34 [−3.22, 0.54]0.131c (h⁻¹)0.0600.0640.0020.004 [−0.002, 0.011]0.202OMED (%)45.7249.480.6822.99 [0.97, 5.01]0.011PD (%)81.0482.600.3670.74 [−1.05, 2.53]0.350a (%)3.266.790.652.68 [0.53, 4.83]0.027b (%)77.7675.790.527−1.95 [−3.94, 0.05]0.055c (h⁻¹)0.0600.0650.0020.005 [−0.003, 0.012]0.172CPED (%)53.7247.571.177−6.13 [−7.22, −5.05]<0.001PD (%)84.5490.661.3336.13 [2.63, 9.62]0.005a (%)30.8624.401.312−6.47 [−8.98, −3.96]<0.001b (%)53.6066.172.44812.60 [9.58, 15.6]<0.001c (h⁻¹)0.0370.0270.002−0.01 [−0.015, −0.005]0.00472h gas production kineticsB (mL g⁻¹ DM)177.03189.452.80312.4 [10.1, 14.7]<0.001C (h⁻¹)0.0480.0580.0020.01 [0.007, 0.013]<0.001L (h)1.090.180.220−0.97 [−1.2, −0.74]<0.001T (h)14.4011.900.582−2.5 [−3.5, −1.51]0.002EGP (mL g⁻¹ DM)86.85101.963.42615.1 [11.6, 18.6]<0.001AGPR (mL h⁻¹)5.727.890.4892.17 [1.77, 2.57]<0.001GC, Guinea grass–cowpea; GCM, Guinea grass–cowpea treated 8% molasses; DM, dry matter; OM, organic matter; CP, crude protein; B, asymptotic gas production; C, rate constant of gas production; L, discrete lag time; T, half-life of asymptote gas production; EGP, effective gas production; AGPR, average gas production rate; a, soluble degradable fraction; b, insoluble but degradable fraction; c, rate constant of degradation of b; ED, effective degradability; PD, potential degradability; SEM, standard error of the mean. a)Values outside the brackets represent the estimated treatment difference, and values within brackets represent the 95% confidence interval of the estimated treatment difference. Model goodness-of-fit was confirmed using the coefficient of determination R² (R² > 0.98 for IVDMD, IVOMD, IVCPD, and cumulative gas production) and the root mean square error (RMSE = 2.60% for IVDMD, 2.75% for IVOMD, 2.47% IVCPD, and 2.80 mL for cumulative gas production). Nutrients degradability and cumulative gas kinetics of molasses-treated Guinea grass–cowpea silage. GC, Guinea grass–cowpea; GCM, Guinea grass–cowpea treated 8% molasses; DM, dry matter; OM, organic matter; CP, crude protein; B, asymptotic gas production; C, rate constant of gas production; L, discrete lag time; T, half-life of asymptote gas production; EGP, effective gas production; AGPR, average gas production rate; a, soluble degradable fraction; b, insoluble but degradable fraction; c, rate constant of degradation of b; ED, effective degradability; PD, potential degradability; SEM, standard error of the mean. a)Values outside the brackets represent the estimated treatment difference, and values within brackets represent the 95% confidence interval of the estimated treatment difference. Model goodness-of-fit was confirmed using the coefficient of determination R² (R² > 0.98 for IVDMD, IVOMD, IVCPD, and cumulative gas production) and the root mean square error (RMSE = 2.60% for IVDMD, 2.75% for IVOMD, 2.47% IVCPD, and 2.80 mL for cumulative gas production). Ruminal DAMs A total of 26 metabolites were ultimately validated as DAMs between GC and GCM across all time intervals after excluding repeated features. The OPLS-DA and PCA plots presented in this section were generated from positive MSI mode, while the corresponding negative mode plots are provided in Supplementary Material S3. OPLS-DA clearly separated GC from GCM at all fermentation times, and the score plots of all samples were inside the 95% Hotelling’s T-squared ellipse (Figs. 3A–D). At 4, 8, and 12 h, the predictive component T[1] explained 46.0%, 48.8%, and 34.9% of the variance, respectively, whereas at 24 h, it accounted for 21.4%. The model quality was high at all time points (Q² = 0.905–0.961; R²Y = 0.992–0.996), confirming its robustness and predictive reliability. PCA further confirmed the treatment separation in an unsupervised manner (Fig. 3E–H). PC1 explained 37.9–68.3% of the total variance, while PC2 accounted for 11.4–30.6%. In all cases, GC and GCM samples formed distinct, non-overlapping clusters, indicating that molasses supplementation induced consistent metabolic shifts as early as 4 h of fermentation and maintained separation through 24 h. However, based on biological relevance, only rumen-, plant-, and microbial biotransformation-related DAMs are mentioned in the results section (Fig. 4) and are discussed later. Orthogonal partial least squares discriminant analysis (OPLS-DA) (A) and principal component analysis (PCA) (B) of untreated Guinea grass–cowpea silage (GC) and 8% molasses-treated silage (GCM) at 4, 8, 12, and 24 h of in vitro fermentation. Blue dots represent biological replicates of GC, whereas red dots represent biological replicates of GCM. All plots were generated from positive MS ionization data. Rumen differentially accumulated metabolites (DAMs) associated with the microbial fermentation of untreated Guinea grass–cowpea (GC) vs. the 8% molasse treated (GCM) at 4, 8, 12, and 24 h of in vitro fermentation. Red and blue bars indicate increases and decreases in DAMs abundance, respectively. Most DAMs accumulation occurred during the first 4 h of incubation, where several tryptophan-derived indolic metabolites, including indole-3-acetate (log2FC = −9.06) and indole-3-carbinol (log2FC = −3.70), showed lower abundance in GCM relative to GC. This suggests altered amino acid catabolism during the early stage of ruminal fermentation. Similarly, the microbial metabolite 2‑hydroxy-4-methylpentanoic acid (HICA) remained consistently less abundant under GCM across all incubation times (log2FC = −2.61 at 24 h), which may indicate altered branched-chain amino acid metabolism. In addition, lipid-derived metabolites, including ricinoleic acid (log2FC = −1.22), 9-HODE (log2FC = −1.38), and 2-hydroxystearic acid (log2FC = −1.07), were reduced during the early incubation phase, indicating shifts in microbial lipid metabolism. In contrast, the carotenoid-derived metabolite crocetin (log2FC = 2.46) was enriched at both 4 h and 24 h, potentially reflecting enhanced antioxidant-related metabolic activity under GCM. Several PSMs became more prominent during later incubation stages. Biochanin A-β-d-glucoside (log2FC = 2.54), a legume-derived isoflavone, was enriched at 8 h, whereas carveol (log2FC = 1.65), andrographolide (log2FC = 1.46), and mellein (log2FC = 1.24) accumulated preferentially from 12 h onward, suggesting progressive release or microbial transformation of plant-derived bioactive compounds. Furanone derivatives, namely 2-(6‑hydroxy-6-methyloctyl)-2H-furan-5-one and 2-(7‑hydroxy-6-methyloctyl)-2H-furan-5-one, were specifically enriched at 4 h and 12 h, respectively (log2FC = 1.41 and 1.38), suggesting time-dependent shifts in microbial fermentation-derived metabolites. Volcano plots of DAMs at all time intervals are shown in Fig. 5, and a full list of DAMs is provided in Supplementary Material S4. Volcano plots of untreated Guinea grass–cowpea (GC) vs. the 8% molasse treated (GCM) at 4, 8, 12, and 24 h of in vitro fermentation. Red, blue, and gray dots indicate increased, decreased, and insignificant DAMs, respectively. Discussion Ruminal degradability kinetics and fermentation Molasses supplementation significantly enhanced DM and OM degradability across all incubation periods, particularly at 24 h and 48 h. This was likely associated with the additional supply of readily fermentable carbohydrates, which may have promoted microbial colonization and fiber degradation (Wang et al., 2018; Seven et al., 2024), thereby increasing the amount of utilizable energy available to ruminal microbes in GCM. However, because the present in vitro assay did not include 0 h degradability measurements, the rapidly soluble a-fraction was estimated entirely through model fitting without accounting for wash losses. Consequently, the estimated DM-a and OM-a fractions may have been lower than values commonly reported in sacco studies, where 0 h washing contributes to physical particle loss and inflates the soluble fraction. Compared with the Napier grass–Sesbania mixture (750:250 g/kg) evaluated using the in sacco technique (Tessema and Baars, 2004), the present silages exhibited lower estimated rapidly soluble fractions and ED values, whereas slowly degradable fractions (DM-b and OM-b) and PD remained comparable between studies. Our findings therefore suggest that the reduced estimated soluble fraction was primarily methodological in origin and should not be interpreted as evidence of impaired degradability of the silages. The DM-a values obtained in the present study were comparable to those of 8.7% and 10.9% reported for pure Guinea grass and Napier grass, respectively (Guadayo et al., 2019), whereas DM-b was considerably above the range of 34.3–43.1% reported by the same authors. In contrast, our mixed Guinea grass–cowpea silages exhibited DM-b values above the range of 22.2–40.8% reported for ensiled pure legumes, such as annual peanut, perennial peanut, pigeonpea, and cowpea, but had DM-a values substantially below the 30.5–44.8% range (Foster et al., 2011). The moderate DM-a and higher DM-b values obtained in the present study indicate the combined structural complexity of Guinea grass and the more digestible cellular constituents of cowpea. The OM-a range of 4.0–6.7% in our silage was within values reported for pure Guinea grass and para grass (4.6% and 11.0%, respectively) by Guadayo et al. (2019), but well below the typical values of 12.0–24.9% reported for tropical legumes by the same authors. Meanwhile, OM-b (75.8–77.8%) in our study was higher than the reported 36.0–51.6% range for tropical grasses and slightly above the 31.9–48.2% range reported for most legume forages and legume trees (Kamatali et al., 1992; Guadayo et al., 2019). The obtained OM degradation rate (0.060–0.065 h⁻¹) closely matched that of the aforementioned forages (0.047–0.105 h⁻¹). These observations indicate that the slowly degradable OM reservoir in our silage was substantial and that its fermentability was improved by the inclusion of cowpea, despite the limited soluble OM pool. Additionally, mixing Guinea grass with cowpea did not generate a highly soluble OM fraction, as is generally observed for pure legumes; instead, the silage retained a grass-like soluble profile. In contrast, CP kinetics displayed the opposite pattern: a marked reduction in CP-a and CP-c was observed in GCM, ultimately reducing CP-ED despite a higher CP-PD. This kinetic pattern, which is uncommon in single-forage systems, may be associated with two complementary mechanisms. First, molasses supplementation is known to improve fermentation quality during ensiling and reduce proteolysis (Zhang et al., 2024), potentially preserving a greater proportion of true protein in less soluble forms and thereby lowering CP-a and slowing early ruminal hydrolysis. Second, possible carbohydrate–protein interactions during ensiling, together with enhanced microbial assimilation of soluble N during in vitro incubation, may have shifted N away from the instantly soluble pool into microbial pools, increasing CP-PD while reducing early N availability. The CP-a range of 24.4–30.9% observed in the present study was lower than the values reported for pure alfalfa (44.6–56.7%) and ryegrass–clover mixtures (41.6%), but higher than that reported for tannin-rich Acacia mearnsii forage (19.9%) (Valderrama & Anrique, 2011; Mhlongo et al., 2025). In contrast, CP-b values (53.6–66.2%) were substantially higher than those reported for the same forages, indicating that a considerable proportion of protein was available through gradual ruminal degradation. Combined with moderate CP degradation rates (0.027–0.037 h⁻¹), this structural shift toward a smaller CP-a pool and a larger CP-b pool may explain the lower CP-ED values (47.6–53.7%). Despite the higher CP-PD, CP-ED remained lower than values reported for ryegrass–clover mixtures and temperate forages, although it was higher than the 28.7% reported for tannin-rich Acacia mearnsii forage (Valderrama & Anrique, 2011; Mhlongo et al., 2025). Overall, the CP degradation characteristics observed here more closely resemble those of phenolic-rich or structurally complex forages (Hoffman et al., 1993; Frutos et al., 2004), in which reduced proteolysis and a larger slowly degradable protein pool limit rapid ruminal N release despite adequate PD. This suggests a greater proportion of dietary protein escaping rapid ruminal degradation in GCM, potentially favoring post-ruminal protein supply. In addition, the relatively elevated aNDFom and ADFom contents may have further restricted microbial and enzymatic access to substrates (Basha et al., 2015), thereby contributing to the observed degradation patterns. However, because intestinal digestibility and metabolizable protein flow were not evaluated, further in vivo investigations are required to confirm the nutritional significance of these alterations in CP degradability kinetics. The differences observed in nutrient composition and fermentation characteristics between GC and GCM silages provide a mechanistic basis for the degradability patterns observed during ruminal incubation. Molasses supplementation increased DM, ME, and residual WSC contents while improving silage fermentation through a greater LA/AA ratio, lower pH, and reduced NH3—N and AA concentrations. These changes suggest more efficient preservation of fermentable substrates during ensiling and reduced protein degradation prior to ruminal incubation. Consequently, the greater availability of readily fermentable carbohydrates in GCM likely promoted microbial activity during in vitro fermentation, consistent with the higher DM-ED, OM-ED, and gas production observed in this treatment. In contrast, the lower NH3—N concentration in GCM, despite similar CP contents between treatments, indicates reduced proteolysis during ensiling. This interpretation aligns closely with the smaller CP-a fraction and larger CP-b fraction observed in GCM, suggesting that a greater proportion of protein remained in less soluble but potentially degradable forms. The lower pH and AA concentration together with the longer AS of GCM further support this interpretation, as rapid acidification during ensiling is known to limit plant protease activity and microbial protein degradation (McDonald et al., 1991). As a result, proteins preserved in more intact and less soluble forms likely required longer hydrolysis during ruminal fermentation (Borucki Castro et al., 2007), thereby lowering CP-c and CP-ED despite the higher CP-PD. Until 4 h, faster initial fermentation, expressed by higher SCFAs concentrations, was observed in GCM. This is likely driven by the greater availability of readily fermentable carbohydrates supplied by molasses. This stimulates early microbial activity and accelerates carbohydrate fermentation. Such enhanced early microbial growth is also supported by concomitantly higher BCFAs concentrations, as isoacids are essential growth factors for a wide range of rumen bacteria (Allison & Bryant, 1963). The highest estimated CH4 in GCM within the first 4 h following incubation is attributable to its high acetate yield, as acetate production during ruminal fermentation generates hydrogen (Ungerfeld, 2020). The transiently higher BCFAs and valerate concentrations at 8–12 h in GC than in GCM indicate a delayed shift toward amino acid fermentation under carbohydrate-limited conditions, resulting in increased microbial deamination (Russell, 1998). This is further supported by the concomitantly higher NH3—N production observed in GC within the same time interval. Ruminal DAMs Although >700 MS peaks were initially detected across the positive and negative ion modes, only 26 metabolites were ultimately validated as DAMs between GC and GCM across all time intervals after removing redundant features. This relatively small number of DAMs has also been reported in other ruminal metabolome-related studies (Artegoitia et al., 2017; Clemmons et al., 2020). This can be explained by the stringent validation criteria or controlled buffering conditions of the in vitro ruminal system, which may have limited extreme pH-driven metabolic fluctuations. Tryptophan is relatively slowly metabolized by ruminal microorganisms (Candlish et al., 1970), yielding intermediates such as indole, 3-methylindole, and indole-3-acetate (Lewis & Emery, 1962). Thus, the reduction in indole-3-acetate abundance in GCM-derived rumen fluid during the first 4 h of incubation suggests suppression of microbial tryptophan degradation and is consistent with the findings of Gahlot et al. (2026), who reported that sugars suppress bacterial indole production. This likely reflects preferential microbial utilization of readily fermentable sugars supplied by molasses rather than amino acids as an energy source, which is consistent with the higher SCFAs concentrations observed in GCM at 4 h. Similarly, the persistent reduction in 2‑hydroxy-4-methylpentanoic acid (HMPA), a leucine deamination byproduct associated with inefficient protein catabolism (Mamer, 2000), further supports the interpretation that molasses supplementation redirected microbial metabolism away from amino acid degradation toward carbohydrate fermentation. This metabolic pattern aligns closely with the lower ruminal NH3—N concentrations and reduced CP degradability observed in GCM. Several lipid-derived metabolites were also reduced during early incubation under molasses supplementation. The decrease in 9-HODE, a product of linoleic acid oxidation and a marker of enhanced lipid turnover (Putman et al., 2019; Andres Contreras et al., 2020), together with lower ricinoleic acid and 2-hydroxystearic acid concentrations, suggests attenuation of oxidative fatty acid transformation pathways in favor of rapid carbohydrate fermentation. Ricinoleic acid metabolism by ruminal microorganisms through hydrogenation and oxidation into hydroxy and oxo fatty acids has previously been described in dairy goats (Alves et al., 2017), whereas hydroxystearic acids are recognized products of microbial oleic acid transformation (Jenkins et al., 2006). The coordinated reduction of these metabolites in GCM therefore indicates that the greater availability of fermentable sugars may have altered microbial lipid metabolism during early fermentation. This interpretation is consistent with Wu et al. (2021), who reported that carbohydrate supplementation in alfalfa silage modified ruminal microbial activity and accelerated the transformation of unsaturated fatty acids. However, unlike the ruminal instability and fermentation imbalance associated with excessive rapidly fermentable carbohydrate supply (Danese et al., 2026), the present findings indicate that moderate molasses supplementation promoted a more coordinated fermentation response. Collectively, the reduced accumulation of amino acid deamination products, such as indole derivatives, and lipid oxidation-related metabolites, including hydroxystearic acids, may indicate improved N utilization efficiency and moderated oxidative transformation pathways. Moderate enrichment of furanone-type derivatives, including 2-(6‑hydroxy-6-methyloctyl)-2H-furan-5-one and related analogs, was observed at 4 h and 12 h in GCM. These compounds are commonly associated with sugar degradation and secondary microbial metabolism involving carbohydrate-derived intermediates (Schwab, 2013). Their accumulation therefore likely reflects the enhanced availability of fermentable sugars provided by molasses. Similar fermentation-derived formation of furanone compounds has previously been reported during microbial conversion of sugar-derived intermediates (Hauck et al., 2003). Molasses supplementation also influenced the persistence of several plant-derived secondary metabolites identified in the ruminal fluid. Biochanin A-β-d-glucoside (sissotrin) was enriched in GCM at 8 h, suggesting that improved silage fermentation may have favored the preservation of certain isoflavonoids during ensiling. In the rumen, this metabolite serves as a precursor of biochanin A, which has recently been associated with improved N utilization efficiency in dairy goats (Zhang et al., 2026), although the present study does not establish a direct mechanistic relationship. Likewise, increased crocetin and mellein concentrations under GCM suggest greater persistence of carotenoid- and legume-associated secondary metabolites during silage fermentation. Crocetin stabilization has previously been linked to glycosylation or esterification with sugar moieties (Moraga et al., 2004), whereas mellein is a polyketide commonly associated with Fabaceae plants (Valio, 1973) and possesses mild antimicrobial activities (Kokubun et al., 2003). Together, these findings suggest that the improved fermentation environment promoted by molasses may have limited degradation of selected plant bioactive compounds during ensiling and ruminal fermentation. The enrichment of andrographolide and carveol during later incubation stages further supports the possibility that molasses supplementation influenced the persistence of terpene-related metabolites within the ruminal metabolome. Although diterpenoids such as andrographolide are often rapidly degraded in the rumen (Srinivasan et al., 2025), the rapid acidification and improved preservation associated with molasses-treated silage may have contributed to their transient accumulation. Yet, studies supporting the evidence of molasses' effect on terpenoids are extremely rare in the literature. Overall, the metabolomic responses observed here indicate that molasses supplementation shifted ruminal metabolism toward enhanced carbohydrate-driven fermentation while moderating amino acid catabolism, lipid oxidative transformation, and the degradation of selected plant-derived secondary metabolites. Notwithstanding these findings, it is important to acknowledge that ruminal digestion was assessed under in vitro conditions. While this approach allows controlled evaluation of fermentation kinetics and metabolomic responses, it does not fully replicate in vivo ruminal physiology, including digesta passage, host absorption, systemic metabolism, and long-term microbial adaptation, which may affect its direct translation to in vivo conditions. Therefore, the observed metabolic shifts should be interpreted strictly within the context of the present in vitro system. Furthermore, metabolite identification was based on untargeted LC–MS/MS annotation and putative database matching, which may limit structural certainty for some compounds. Although >700 MS features were initially detected, only a relatively small number of metabolites were ultimately validated as DAMs following stringent filtering and validation criteria. Also, the use of a single molasses supplementation level (8%) limits the assessment of dose-dependent fermentation and its associated metabolomic responses. Further in vivo studies integrating metabolomics and animal performances are therefore required to validate and extend these findings under practical feeding conditions. Conclusion Molasses supplementation during ensiling improved the quality and ruminal fermentability of Guinea grass–cowpea silage by enhancing DM and OM degradability kinetics while moderating CP breakdown. Metabolomics further revealed a shift in ruminal metabolism toward carbohydrate-associated pathways, accompanied by reduced lipid and amino acid catabolism and increased abundance of antioxidant metabolites. Collectively, these findings suggest that molasses influences substrate-driven microbial metabolism beyond its role as a readily fermentable energy source under the present in vitro conditions. This confirms its value in ruminant nutrition as a strategic additive for optimizing the nutritional and metabolic characteristics of Guinea grass–cowpea silage. Study approval statement All animal procedures were approved by the Committee for Animal Research and Welfare of Gifu University (Approval ID: AG-P-N-20,240,057 and 20,250,122). Funding This research did not receive any specific grant from funding agencies in any commercial or non-for-profit section. CRediT authorship contribution statement Armess Prince Gynth Sossou: Writing – original draft, Visualization, Methodology, Investigation, Formal analysis, Data curation, Conceptualization. Masato Yayota: Writing – review & editing, Validation, Supervision, Resources. Declaration of competing interest The authors declare that they have no conflicts of interest related to this article. Acknowledgements Sossou Armess Prince Gynth would like to thank the staff of the Yanagido livestock farm, Gifu Field Science Center (Gifu University), for their help and support during land preparation for forage establishment. Footnotes Supplementary material associated with this article can be found, in the online version, at doi:10.1016/j.vas.2026.100745. Supplementary materials Data availability The models and data that support the study findings are available to reviewers and available from the authors upon request. They are not deposited in an official repository. References Associated Data Supplementary Materials Data Availability Statement The models and data that support the study findings are available to reviewers and available from the authors upon request. They are not deposited in an official repository.