Fermentation Quality Divergence of Alfalfa Silage within the Same Batch: Insights from 16S rRNA Sequencing and Untargeted Metabolomics

Abstract

Alfalfa silage produced from the same batch of raw material frequently exhibits divergent fermentation quality, yet the microbial and metabolic basis of this divergence remains poorly understood. This study aimed to characterize the differences in bacterial community and metabolite profiles between relatively well- and poorly fermented alfalfa silages within the same batch. A total of 100 laboratory-scale silos of wilted alfalfa were prepared and ensiled for 60 days. Based on the pH value and lactic acid (LA) concentration, 49 silos were classified as the relatively well-fermented group (LPGood), 51 silos as the poorly fermented group (LPBad), and six silos were randomly selected from each group for further analysis. The LPGood group showed significantly (p < 0.05) lower pH and higher LA and acetic acid (AA) concentrations than the LPBad group, whereas no significant (p > 0.05) differences were detected in dry matter content or ammonia nitrogen. No significant (p > 0.05) differences in bacterial α diversity were observed between the two groups; however, the genus Lactiplantibacillus was the predominant microorganism in both groups and was significantly enriched in the LPGood group. Metabolomic analysis identified 231 significantly upregulated and 133 significantly downregulated metabolites in the LPGood group relative to the LPBad group (VIP > 1.0, FDR-adjusted q < 0.05), with the LPGood group significantly (p < 0.05) enriched in palmitic acid, oleamide, linoleic acid, succinic acid, and phenylacetic acid. In conclusion, the fermentation quality divergence of alfalfa silage within the same batch was closely associated with the bacterial community composition and metabolic reprogramming, and Lactiplantibacillus, together with its associated metabolic pathways, may serve as potential indicators for monitoring and regulating the fermentation quality of alfalfa silage.

1. Introduction

Alfalfa (Medicago sativa L.), as a high-quality protein forage source, plays a pivotal role in the livestock production. However, it is susceptible to undesirable fermentation which results in the proliferation of spoilage microorganisms and nutritional deterioration owing to the high buffering capacity and low water-soluble carbohydrate (WSC) content of the alfalfa. , These disadvantages not only lead to dry matter (DM) losses and a decline in palatability, but also accelerate the accumulation of toxic metabolites, such as biogenic amines produced by spoilage bacteria. Ensiling depends on anaerobic fermentation by epiphytic lactic acid bacteria (LAB), which convert WSC into organic acids, reduce the pH, and inhibit the growth of spoilage microorganisms and molds. Therefore, higher lactic acid (LA) accumulation and a lower pH have been widely regarded as the key indicators for evaluating the fermentation quality of silage.

In large-scale production practices, the fermentation quality of alfalfa silage often exhibits substantial divergence even when the raw materials are harvested from the same field and ensiled under identical conditions, a phenomenon referred to as fermentation quality divergence. Compared with corn and whole-crop cereal silages, alfalfa silage is particularly prone to such divergence because of the high genetic heterogeneity of its epiphytic microbiota, which results in heterogeneous metabolite profiles and fermentation quality within the same batch of ensiled material. Previous studies on alfalfa have demonstrated that the fermentation quality of alfalfa silage is highly dependent on the dynamics of the epiphytic bacterial community; for instance, alfalfa ensiled under identical conditions can develop markedly different bacterial community structures and fermentation end-product profiles. ,, Multi-omics analyses of alfalfa silage further indicated that fermentation quality is shaped by multiple metabolic pathways rather than by lactic acid fermentation alone. , In this study, the term “micro-ecology” refers specifically to the ensiling micro-ecosystem, that is, the epiphytic microbial community and its metabolic activities within the silo, rather than the broader field-scale ecological factors (e.g., climate, soil, and water resources). An in-depth analysis of the bacterial community succession and metabolic differences underlying the fermentation quality divergence of alfalfa silage is therefore critical for achieving precise regulation of alfalfa silage production. Nevertheless, the key microbial taxa and metabolites associated with this divergence within the same batch of alfalfa silage remain unclear.

Therefore, this study aimed to characterize the differences in fermentation quality, bacterial community composition, and metabolite profiles between relatively well- and poorly fermented alfalfa silages produced from the same batch of raw material, using 16S rRNA gene sequencing integrated with untargeted metabolomics. It should be noted that the present study adopted an end point-based comparative design, in which silage samples were analyzed after 60 days of ensiling and grouped retrospectively according to their fermentation quality; the raw material microbiome and metabolome were not profiled. This study will provide new perspectives for refining the fermentation theory of high-protein forage silage and for developing targeted strategies to improve the fermentation quality of alfalfa silage.

2. Materials and Methods

2.1. Materials and Ensilage Preparation

Alfalfa (M. sativa L.) was harvested at the early-bloom stage from the experimental field of Inner Mongolia Agricultural University (Hohhot, China) and wilted to a dry matter (DM) content of approximately 35% (corresponding to a moisture content of ∼65%). The wilted alfalfawas chopped into 1–2 cm length, And allchopped material was thoroughly mixed to obtain a homogeneous batch. The mixed material was then randomly distributed into 100 laboratory-scale silos (approximately 1 L capacity) at a packing density of about 600 kg/m³. All silos were sealed and stored at ambient temperature (25 ± 2 °C) for 60 days.

2.2. Grouping Criteria and Sampling

After 60 days of ensiling, all 100 silos were opened, and the pH value and organic acid concentrations of each silage sample were determined as described in Section . According to the silage evaluation criteria reported by Ávila and Carvalho and Kung et al., silages with a pH value below 5.20 and a LA concentration above 8.0 g/kg DM were classified as the relatively well-fermented group (LPGood, n = 49), whereas silages with a pH value of 5.20 or above and a LA concentration below 8.0 g/kg DM were classified as the poorly fermented group (LPBad, n = 51). Six silos were then randomly selected from each group (12 silos in total) as biological replicates for subsequent chemical composition, microbial enumeration, bacterial community, and metabolomic analyses. It should be emphasized that, because the pH of alfalfa silage rarely decreases below 4.5 in the absence of additives owing to its high buffering capacity, the LPGood and LPBad groups in this study represent the relative divergence of fermentation quality within the same batch rather than absolutely successful or failed fermentation.

2.3. Chemical Composition and Fermentation Product Analyses

The dry matter (DM) of the raw materials and silages was determined by drying in a forced-air oven at 65 °C to constant weight. The dried samples were then ground through a 1 mm sieve using a grinder (FW100, Taisite Instrument Co. Ltd., Tianjin, China) for the analysis of chemical compositions. The crude protein (CP) content was determined according to the Association of Official Analytical Chemists method. The acid detergent fiber (ADF) and neutral detergent fiber (NDF) contents were measured using the method described by Van Soest et al. The WSC content was quantified using a colorimetric assay with anthrone reagent.

A total of 20 g of alfalfa silage from each replicate was collected and diluted with 180 mL of distilled water. The mixture was stored in a refrigerator at 4 °C for 24 h to extract the fermentation products, followed by filtration through four layers of cheesecloth. The pH of the filtrate was immediately determined using a glass-electrode pH meter. The filtrate was then passed through a 0.22 μm microporous membrane filter,and the concentrations of lactic acid (LA) and acetic acid (AA) were quantified via high-performance liquid chromatography (Agilent 1260 Infinity II, Agilent Technologies, Santa Clara, CA, USA). The ammonia nitrogen (NH₃–N) content of the filtrate was determined using the phenol-sodium hypochlorite colorimetric method.

2.4. Microbial Enumeration

The populations of LAB, yeasts, Escherichia coli, and aerobic bacteria in the raw alfalfa material were enumerated according to the plate-count method described by Bai et al. Briefly, 10 g of fresh sample was homogenized with 90 mL of sterile distilled water and serially diluted 10-fold. The LAB were enumerated on the de Man, Rogosa and Sharpe (MRS) agar after anaerobic incubation at 37 °C for 48 h; the yeasts were enumerated on potato dextrose agar (PDA) after aerobic incubation at 30 °C for 72 h; the E. coli was enumerated on blue light broth agar after incubation at 37 °C for 24 h; and the aerobic bacteria were enumerated on nutrient agar after incubation at 30 °C for 48 h. Colonies were counted and expressed as the logarithm of colony-forming units per gram of fresh matter (lg cfu/g FM).

2.5. Bacterial Community Analysis

Total genomic DNA was extracted from the alfalfa silage samples using the cetyltrimethylammonium bromide method. The DNA integrity was verified via 1.0% agarose gel electrophoresis, and the DNA concentration was determined using a NanoDrop 2000 spectrophotometer (Thermo Fisher Scientific, Wilmington, DE, USA). The V3–V4 hypervariable regions of the bacterial 16S rRNA gene were amplified by PCR using the universal primers 338F (5′-ACTCCTACGGGAGGCAGCAG-3′) and 806R (5′-GGACTACHVGGGTWTCTAAT-3′). The PCR products were purified and subjected to paired-end sequencing on the Illumina platform by LC-Bio Technology Co., Ltd. (Hangzhou, China).

Raw sequencing reads were quality-filtered using the Trimmomatic software to remove low-quality sequences and chimeras. Operational taxonomic units (OTUs) were clustered at a 97% similarity threshold using UPARSE software, and taxonomic classification was performed within the QIIME2 platform. α diversity indices (Chao1 and Shannon) and β diversity (principal coordinate analysis, PCoA) were calculated, and linear discriminant analysis effect size (LEfSe) was further applied to identify distinct microbial biomarkers between the LPGood and LPBad groups (LDA score >3.0, p < 0.05).

2.6. Metabolic Analysis

Metabolite separation and detection were performed using an ultra-high performance liquid chromatography (UHPLC) system (Thermo Vanquish, Thermo Fisher Scientific, Waltham, MA, USA) coupled with a Q-Exactive high-resolution mass spectrometer (Thermo Fisher Scientific, USA). Chromatographic separation was achieved on an ACQUITY UPLC HSS T3 column (100 mm × 2.1 mm, 1.8 μm), with the column temperature maintained at 40 °C. The mobile phase consisted of 0.1% formic acid in water (A) and 0.1% formic acid in acetonitrile (B), administered at a flow rate of 0.3 mL/min for gradient elution. Mass spectrometry data were acquired in both positive and negative electrospray ionization (ESI⁺/ESI^–) modes via fast polarity switching, with a scanning range of m/z 70–1050. The raw LC-MS data were imported into Compound Discoverer 3.1 software (Thermo Fisher Scientific, USA) for peak extraction, deconvolution, peak alignment, and database identification.

Multivariate statistical analysis was conducted using SIMCA 14.1 software (Sartorius Stedim Data Analytics AB, Umeå, Sweden). Principal component analysis (PCA) and orthogonal partial least squares-discriminant analysis (OPLS-DA) models were constructed, and the OPLS-DA model was validated by a permutation test (200 permutations). Differential metabolites were screened based on a variable importance in the projection (VIP) threshold of >1.0 and a p < 0.05. To control the false positive rate arising from multiple comparisons, p values were adjusted using the Benjamini–Hochberg false discovery rate (FDR) procedure, and metabolites with VIP > 1.0 and FDR-adjusted q < 0.05 were considered significantly different. Metabolic pathway enrichment analysis was subsequently performed using the Kyoto Encyclopedia of Genes and Genomes (KEGG) database.

2.7. Statistical Analysis

Statistical analyses were performed using one-way analysis of variance with general linear models in SAS software version 9.2. Significant differences were considered at p < 0.05 level. The correlations among fermentation parameters, dominant bacterial genera, and differential metabolites were evaluated using Mantel tests implemented with the vegan package in R software (version 4.3.0).

3. Results

3.1. Physicochemical Characteristics of Raw Materials

The chemical characteristics and microbial populations of the alfalfa raw materials prior to ensiling were shown in the Table . The DM content (37.80%) of the substrates was within the typical range for alfalfa silage production. The CP (20.05% of DM), ADF (32.14% of DM), and NDF (42.67% of DM) contents showed low variability among the subsamples, indicating that the raw material was homogeneous before ensiling. The microbial populations in the raw materials were 4.73, 6.35, 6.72, and 7.08 lg cfu/g FM for LAB, yeasts, E. coli, and aerobic bacteria, respectively.

1. Nutritional Composition and Microbial Counts of Alfalfa Raw Material .

Item	Alfalfa
DM (%)	37.80 ± 0.25
ADF (%DM)	32.14 ± 1.65
NDF (%DM)	42.67 ± 2.19
CP (%DM)	20.05 ± 0.17
WSC (%DM)	2.43 ± 0.22
LAB (lg cfu/g FM)	4.73 ± 0.22
Yeast (lg cfu/g FM)	6.35 ± 0.17
E. coli (lg cfu/g FM)	6.72 ± 0.34
Aerobes (lg cfu/g FM)	7.08 ± 0.47

^aADF, acid detergent fiber; CP, crude protein; cfu, colony-forming units; DM, dry matter; NDF, neutral detergent fiber; WSC, water-soluble carbohydrates; LAB, lactic acid bacteria.

3.2. Fermentation Quality of Alfalfa Silage

According to the grouping criteria described in Section , 49 of the 100 silos were classified into the LPGood group and 51 into the LPBad group after 60 days of ensiling. The fermentation quality of the two groups is presented in Table . A significantly (p < 0.05) lower pH value was detected in the LPGood group compared with the LPBad group. Similarly, markedly (p < 0.05) higher LA and AA concentrations were found in the LPGood group. No significant (p > 0.05) differences were observed in the DM content or the NH₃–N concentration between the two groups. Propionic acid and butyric acid were not detected in any of the alfalfa silage samples.

2. Fermentation Quality of Alfalfa Silage after 60 Days of Ensiling .

Item	LPBad	LPGood	p-value
DM (%)	36.55 ± 0.58	36.94 ± 0.07	0.611
pH	5.35 ± 0.04	5.10 ± 0.02	< 0.001
Lactic acid	7.51 ± 0.93	9.18 ± 0.46	0.003
Acetic acid	10.01 ± 0.81	11.27 ± 0.16	0.004
NH3–N	0.22 ± 0.11	0.24 ± 0.11	0.822

^aDM, dry matter; NH₃–N, ammonia nitrogen; LPBad, poorly fermented group; LPGood, relatively well-fermented group.

3.3. Bacterial Community Profiles of Alfalfa Silage

A total of 12 alfalfa silage samples from the LPGood (n = 6) and LPBad (n = 6) groups were used to evaluate the bacterial diversity through 16S rDNA gene sequencing. A total of 987,582 raw reads and 986,448 valid reads were obtained from the 12 alfalfa silage samples, respectively. No significant (p > 0.05) differences were observed in the Chao1 and Shannon indexes between the two groups (Table .), The present study analyzed the differences in bacterial diversity between the two groups, and the αdiversity were also calculated. the Good’s coverage index was higher than 99% for all alfalfa silage samples, indicating that the sequencing depth was adequate to represent the bacterial community of each sample, and no significant (p > 0.05) differences were observed among the two groups.

3. Sequencing Data and α-Diversity Indices of the Bacterial Community in Alfalfa Silage.

Item	LPGood	LPBad	SEM	p-value
Sequencing Tags	84254	80343	2093	0.002
Valid Tags	84153	80255	2089	0.002
Chao1 index	748.89	766.14	39.00	0.818
Shannon index	6.11	6.42	0.23	0.589
Goods coverage	0.997	0.998	0.00	1.000

The PCoA plot demonstrated a clear separation of bacterial community profiles between the LPGood and LPBad groups (Figure A). The Venn diagram showed that the two groups shared 1,062 OTUs, with 2,839 and 2,789 unique OTUs identified in the LPBad and LPGood groups, respectively (Figure B). As shown in the Figure C, the two groups exhibited similar bacterial community structures at the phylum level, with the phyla Bacillota and Pseudomonadota (formerly known as Firmicutes and Proteobacteria, respectively) being dominant and collectively accounting for over 99% of the total relative abundance; a significant difference between the two groups was found for the phylum Bacillota. At the genus level, the community structures also remained similar, with the top five dominant genera being Lactiplantibacillus, unclassified Rhizobiaceae, Pantoea, Paracoccus, and Devosia; a significant difference was found for the genus Lactiplantibacillus, which was enriched in the LPGood group (Figure D). The LEfSe analysis further identified the key biomarkers associated with the fermentation quality divergence between the groups. The LPGood group was significantly enriched with typical lactic acid-fermenting bacteria, including Enterococcus, Levilactobacillus, Weissella, and Lactococcus, whereas the LPBad group retained a higher abundance of non-fermentative spoilage-associated bacteria, such as Agrobacterium and Methylobacterium (Figure E).

1.

Microbial community structure. (A) PCoA analysis; (B) Venn diagram; (C) relative abundance of microbial communities at the phylum level; (D) relative abundance of microbial communities at the genus level; (E) LEfSe analysis.

3.4. Metabolites Profiles of Alfalfa Silage

Metabolomic analysis was carried out to further explore the metabolic differences between the LPGood and LPBad groups.. As shown in the Figure A, the PCA analysis revealed that the confidence ellipses of the two groups overlapped, and the samples were clustered relatively closely, indicating minor differences in the overall metabolic profiles between the two groups. The permutation test for the OPLS-DA model showed an intercept of −0.7293 for the Q ² regression line on the Y-axis. Moreover, all Q ² points from the permutation tests were located below the original point, demonstrating excellent predictive ability and robustness of the model, which confirms the reliability of the identified differential metabolites (Figure B). Volcano plot analysis revealed 231 significantly upregulated and 133 significantly downregulated metabolites in the LPGood group compared with the LPBad group (Figure C). Significantly enriched metabolites in the two groups were diversity (Figure D), the LPGood group was significantly (p < 0.05) enriched in biogenic amines (e.g., cadaverine and tyramine) and lipid metabolites (e.g., palmitic acid, oleamide, and linoleic acid), whereas the LPBad group was significantly (p < 0.05) enriched in organic acids (e.g., malonic acid, malic acid, and pyruvic acid) and flavonoids (e.g., genistein, luteolin, and quercetin). The KEGG pathway enrichment analysis under both positive and negative ion modes indicated that the differential metabolites between the LPGood and LPBad groups were significantly enriched in isoflavonoid biosynthesis and the biosynthesis of phenylpropanoids (Figure E).

2.

Metabolic analysis. (A) PCA score plot. (B) OPLS-DA score plot. (C) volcano plot of differential metabolites. (D) VIP scores and heatmap of differential metabolites. (E) KEGG pathway enrichment analysis in positive and negative ion modes.

3.5. Associations among Fermentation Parameters, Bacterial Community, and Metabolites

The Mantel test indicated that the pH was the core factor associated with the fermentation progress, exhibiting significant correlations with both the bacterial community (r < 0.6, p < 0.05) and the differential metabolites (r < 0.3, p < 0.05) (Figure ). The pH was significantly positively correlated with the malic acid (r = 0.61, p < 0.05), malonic acid (r = 0.58, p < 0.05) and l-arginine (r = 0.64, p < 0.05), and negatively associated with cadaverine (r = −0.52, p < 0.05). Furthermore, the LA and AA concentrations showed positive correlations with the metabolites (r ≥ 0.3, p < 0.01), and both were closely associated with the accumulation of energy metabolism intermediates, such as pyruvic acid (r = 0.48, p < 0.05), whereas the DM content had no significant (p > 0.05) correlation with the bacterial community or with the majority of metabolites. Lactiplantibacillus was positively correlated with succinic acid (r = 0.740, p < 0.01), purine (r = 0.700, p < 0.01) and isocitric acid (r = 0.428, p < 0.01). In contrast, Enterobacter and Kosakonia were significantly positively correlated with NH₃–N and cadaverine (r < 0.5, p < 0.01).

3.

Mantel test analysis of correlations among fermentation quality, key metabolites and microbial communities in silage samples. LA, lactic acid; AA, acetic acid; NH₃–N, ammonia nitrogen.

4. Discussion

The fermentation quality of alfalfa silage directly influences its nutritional value and safety as a high-quality protein feed in livestock production, whereas fermentation quality divergence during the ensiling process often leads to severe nutritional losses and the accumulation of harmful metabolites. In this study, 16S rRNA gene sequencing and untargeted metabolomics were comprehensively applied to analyze the bacterial community and metabolomic characteristics of alfalfa silages that exhibited significant divergence in fermentation quality within the same batch.

4.1. Raw Material Characteristics and the Fermentation Quality Divergence of Alfalfa Silage

As presented in Table , the alfalfa raw material in this study exhibited the typical characteristics of high crude protein (20.05% of DM) and low WSC (2.43% of DM) contents, which frequently poses a substantial impediment to the production of high-quality silage. , Furthermore, the epiphytic microbial counts of the raw material showed a low LAB population (4.73 lg cfu/g FM) relative to yeasts (6.35 lg cfu/g FM) and aerobic bacteria (7.08 lg cfu/g FM). These undesirable microorganisms may occupy the niche in the raw material and trigger intense competition for the limited WSC substrate, thereby directing the fermentation toward an undesirable trajectory.

It is noteworthy that the pH values of both groups (5.10–5.35) in the present study were higher than pH 4.2, the threshold generally regarded as indicative of well-fermented silage. This is consistent with previous reports that alfalfa, owing to its high buffering capacity and low WSC content, rarely reaches a terminal pH below 4.5 when ensiled without additives, even after wilting. , Therefore, the LPGood and LPBad groups defined in this study should be interpreted as the relative divergence of fermentation quality within the same batch rather than absolute fermentation success or failure. Within this context, the significantly lower pH and higher LA and AA concentrations in the LPGood group indicate a more extensive lactic acid fermentation, which could be attributed to the heterogeneous distribution of the epiphytic microbial community among individual silos within the same batch. No significant difference was found in the NH₃–N content between the two groups, possibly because the metabolic activity of most proteolytic microorganisms was not markedly affected when the pH remained above 4.50. The absence of significant differences in DM content and α diversity further indicates that the two groups originated from a homogeneous batch of raw material, supporting the notion that the observed divergence arose during the ensiling process rather than from pre-existing differences in the substrate.

4.2. Bacterial Community Composition Underlying the Fermentation Quality Divergence

Previous studies have shown that well-preserved silage depends on long-term preservation through organic acids produced by the metabolic activities of LAB, , as well as on the succession of the bacterial community structure. In the current study, no significant (p > 0.05) differences in the Chao1 and Shannon indices were observed between the two groups, whereas the PCoA plot indicated clear inter-group separation. This discrepancy may be explained by the differential growth of bacterial species under the various pH micro-environments of individual silos. The bacterial community was dominated by the phyla Bacillota and Pseudomonadota, and Lactiplantibacillus occupied the dominant position at the genus level, with a significant difference in its relative abundance between the two groups. These results are in accordance with previous findings that LAB are the foundation of silage fermentation. ,, The LEfSe analysis further demonstrated that the LPGood group was significantly enriched with typical lactic acid-fermenting taxa, such as Enterococcus, Levilactobacillus, Weissella, and Lactococcus, which is consistent with previous research. In contrast, the LPBad group retained higher abundances of Agrobacterium and Methylobacterium. Slow fermentation leads to a lag in the establishment of anaerobic and strongly acidic conditions, so that these undesirable bacteria are not eliminated in a timely manner and continue to compete for ecological niches. The specific enrichment of distinct biomarkers and the differential activation of associated metabolic pathways may therefore underlie the significant divergence in the final fermentation quality between the two groups.

4.3. Metabolic Reprogramming Associated with the Fermentation Quality Divergence

It has generally been considered that successful ensiling requires entering the lactic acid fermentation stage as early as possible so that the pH can decrease rapidly, allowing the silage to be preserved stably over a long period. In the current study, lower pH and higher organic acid contents were detected in the LPGood group, accompanied by relatively minor shifts in bacterial composition; nevertheless, pronounced differences in the metabolite profiles were detected. A total of 2,077 metabolites were identified in the silage samples, which is in accordance with previous reports on rice straw silage, corn silage, paper mulberry leaf silage, and wheat straw silage. ,− Compared with the LPBad group, the concentrations of palmitic acid, oleamide, linoleic acid, succinic acid, and phenylacetic acid were markedly increased in the LPGood group, and metabolic pathways related to carbon metabolism, microbial metabolism in diverse environments, and flavonoid biosynthesis were enriched. This observation is consistent with findings from silages prepared from different raw materials, in which well-preserved silage exhibited increased organic acid- and derivative-related metabolites. , The lipid metabolites enriched in the LPGood group, such as palmitic acid and oleamide, are frequently associated with the active remodeling of the fluidity and rigidity of the cell membrane lipid bilayer by LAB in response to low-pH stress. The higher relative abundance of Lactiplantibacillus in the LPGood group, together with the better fermentation quality, may be attributed to the complex fermentation characteristics and metabolic pathways of LAB in response to substrate types and competitive pressures. Organic acid- and derivative-related metabolites, such as succinic acid, which can influence the pH, were also concentrated in the LPGood group, which in turn may have contributed to the improved fermentation quality. ,

An unexpected finding of this study was that the LPGood group was significantly enriched in biogenic amines, such as cadaverine and tyramine, which are generally regarded as spoilage-associated metabolites. Two explanations may account for this phenomenon. First, amino acid decarboxylation is an important acid-resistance mechanism in bacteria: decarboxylation consumes intracellular protons and thereby helps decarboxylase-positive bacteria maintain cytoplasmic pH homeostasis under acidic stress. The LPGood group was significantly enriched with Enterococcus, a genus well documented for tyramine production through tyrosine decarboxylase activity during fermentation. Therefore, the accumulation of biogenic amines in the LPGood group may reflect the active acid-stress response of acid-tolerant lactic acid-fermenting bacteria rather than extensive spoilage. Second, the terminal pH values of both groups remained above 5.0, which is insufficient to fully inhibit microbial amino acid decarboxylase activity, and biogenic amine formation can proceed even in silages dominated by LAB under such conditions. Nevertheless, from the perspective of feed safety, biogenic amines are undesirable components, and their accumulation in the relatively well-fermented group indicates that the fermentation quality of alfalfa silage in the present study still has considerable room for improvement, for example through inoculation with selected LAB strains to achieve a more rapid and deeper pH decline.

Several limitations of the present study should be acknowledged. First, an end point-based comparative design was adopted, and the bacterial community and metabolome of the raw material at day 0 were not profiled; therefore, the inherent differences in the raw material and the temporal dynamics of the divergence could not be fully resolved, and the relationships reported here are associative rather than causal. Second, although the FDR correction was applied to the differential metabolite screening, the untargeted metabolomic results remain semi-quantitative and require validation by targeted quantification. Third, the grouping of LPGood and LPBad reflects the relative divergence within a single batch of alfalfa rather than absolute fermentation success or failure, and the experiment was conducted in laboratory-scale silos; the generalizability of the findings to farm-scale bunker silos and to different ensiling durations remains to be verified. Future studies should incorporate time-series sampling from day 0 onward and farm-scale validation to elucidate the trajectory and driving forces of fermentation quality divergence in alfalfa silage.

5. Conclusion

This study demonstrates that the fermentation quality divergence of alfalfa silage within the same batch was closely associated with the composition of the bacterial community and the reprogramming of the metabolite profile. The relative abundance of Lactiplantibacillus and its associated metabolic pathways appeared to be the key factors linked to this divergence. Lactiplantibacillus, together with the associated metabolites and pathways, may serve as potential indicators for monitoring and regulating the fermentation quality of alfalfa silage, providing a theoretical basis for improving silage quality through targeted micro-ecological modulation of the ensiling process.

The 16S rRNA gene sequences of the alfalfa silage samples were deposited in the NCBI database with the accession number PRJNA1470148.

J.L.: Investigation, writingoriginal draft, visualization, and data curation. L.H.: Investigation. Y.L.: Investigation. G.G.: Funding acquisition. M.L.: Funding acquisition. S.D.: Conceptualization, supervision, funding acquisition, writingreviewing, and editing.

This research was supported by the Inner Mongolia Autonomous Region High-level Talents Project (NDYB2024-34) and the First-Class Disciplines Project of Inner Mongolia (YLXKZX-NND-039).

The authors declare no competing financial interest.