The rumen fluid used in this experiment was collected from 6 Hu sheep of approximately 2 years old with permanent rumen fistulas, and the sheep were housed at the Animal Experimental Center of Hebei Agricultural University. The diet was formulated according to NRC2007 (50 kg body weight, 200 g daily gain, the metabolizable energy, and crude protein content in the diet was 11.09 MJ/kg and 12.28% as dry matter basis, respectively). The feed was collected and dried at 65°C for 48 h to achieve air-dried samples, sieved through a 10-mesh sieve, and stored in a dry and sealed environment for nutrient analysis (Table 1).

Four treatment groups were set up, namely, RF group (fresh rumen fluid), FRF group (frozen rumen fluid at −80°C for 1 month, ice water bath for 8 h, and preheated at 39°C for 30 min before used), CRF group (continuously batch-cultured rumen fluid for 23 days), and MRF group (3 of 4 fresh rumen fluid mixed with 1 of 4 continuously batch-cultured rumen fluid).

Before rumen fermentation in vitro, the bacterial content in each group was counted. In brief, after being stained with ammonium oxalate-crystal violet and saline multiplicity dilution, the bacteria were counted using a hemocytometer under 1,000x magnification. The bacterial content was calculated as follows: Bacterial Count/ml = N/S*D*16*10*100, where N was the number of counting grid squares, S was the total number of squares counted, and D was the dilution factor. As the bacteria content in different groups was also different, the bacterial content in each group was regulated to 7.45 × 10⁹/ml, with rumen buffer solution for rumen fermentation in vitro.

Six replicates were set up for each group, and the rumen fermentation with anaerobic condition was performed following the previous protocol (Li et al., 2022) with some modifications: 3 g of diet containing 5% urea (dry matter basis) was used as fermentation substrate, and the fermentation lasted for 48 h. Gas production was measured using a graduated syringe; 2 mL of fluid was collected at 2, 4, 6, 8, 10, 12, 24, and 48 h, respectively. At the end of the fermentation, the bags were placed in an ice water bath immediately to terminate fermentation; The nylon bag containing feed residues was rinsed three times with distilled water and then dried at 65°C for 48 h; the liquid fraction was also collected and stored at −80°C for determination of pH, MCP, and bacterial composition.

The pH of the fluid fraction was determined using an EL20-type acidimeter. The MCP protein was determined following the method described by Makkar et al. (1982), and the concentration of NH₃-N was measured with the phenol-hypochlorite colorimetric method (Lv et al., 2019). Gas production parameters were calculated using the LELAG mathematical model: GP = b(1–2.718(−c(t-LAG))), where b represented theoretical gas production (ml), c represented gas production rate, and t represented time point (Wang et al., 2011). The diet substrate and undegraded residuals were dried at 65°C for 48 h, and the dry matter (DM) was determined following the AOAC method (AOAC, 2016). The digestibility of dry matter was calculated according to the formula as follows: [(Initial Dry Matter Content–Final Dry Matter Content)/Initial Dry Matter Content] × 100%.

Total microbial DNA was extracted from the rumen fluid samples after 48 h of rumen fermentation in vitro using the E.Z.N.A. Soil DNA Kit (Omega Bio-tek, Inc., United States), and the quality and concentration of the DNA were assessed using the Nanodrop 2000 (Thermo Fisher Scientific, Inc., United States). The V3-V4 regions of the 16S rRNA gene were amplified with the primers 338F (5′-ACTCCTACGGGAGGCAGCAG-3′) and 806R (5′-GGACTACHVGGGTWTCTAAT-3′) using an ABI 9700 PCR instrument (Applied Biosystems, Inc., USA). The PCR amplification system consisted of 1 μL of forward primer (5 μM) and reverse primer (5 μM), 3 μL of DNA template (30 mg total template), 3 μL of BSA (2 ng/μl) and ddH₂O, and 12.52 μL of Taq Plus Master Mix. The PCR amplification program was as follows: pre-denaturation for 5 min (95°C), denaturation for 45 s (95°C), annealing for 50 s (55°C), extension for 45 s (72°C) for 28 cycles, and extension for 10 min (72°C). The PCR products were purified using the Nucleic Acid Purification Kit (Agencourt AMPure XP, Beckman Coulter, Inc., United States), and finally, high-throughput sequencing was performed on the Illumina Miseq platform (Illumina, Inc., United States), following our previous study (Li et al., 2022).

After filtering out the low-quality reads with a quality score lower than 20 over a 50 bp sliding window and sequences with reads shorter than 50 bp with Trimmomatic (v0.36; Gong et al., 2019), the sequencing raw data were assembled using Pear (v0.9.6) software. Vsearch (version 2.0.3) was then used to remove short sequences and chimeras in assembled sequences. Representative sequences were aligned at the Silva138 database with a 97% sequence similarity threshold, to generate operational taxonomic units (OTUs). The α-diversity and β-diversity indexes were analyzed using QIIME (v1.8.0) software.

The rumen fermentation parameters after 48 h of rumen fermentation in vitro, including DM digestibility, 48 h cumulative GP, theoretical GP, pH, and MCP, were analyzed using one-way ANOVA, and comparison was performed with Duncan’s method using SPSS 26.0 software; the area under the curve (AUC) of dynamic changes in NH₃-N content in rumen fluid was calculated, changes in rumen microbial community structure were analyzed based on principal component analysis and ANOSIM statistics, and RandomForest analysis was performed with the RandomForest package using R software (v4.1.2) with ntree = 10,000, mtry = p/3, where p is the number of bacterial taxa, and signature bacteria were screened by increasing mean squared error (%IncMSE) and cross-validation. Data were expressed as means and standard error of means (SEM), and p < 0.05 was considered significantly different.

During the first 13 days of gradually increasing urea addition from 0 to 5%, the gas production showed obvious fluctuations from 58.87 mL/g to 77.18 mL/g. While the urea addition was stabilized at 5% after 13 days of fermentation, the gas production became relatively stable at approximately 70.26 mL/g. The dry matter digestibility remained stable with an average of approximately 34.74% over 23 days of continuous rumen batch culture in vitro (p > 0.05, Figure 2).

After 23 days of continuous batch culture, the bacterial counts in the CRF group were approximately 3.4 × 10¹⁰, significantly lower than that in fresh rumen fluid (7.54 × 10⁹) but still higher than that in frozen rumen fluid (2.48 × 10¹⁰; p < 0.05; Table 2).

To confirm the function of the microbial community after continuous rumen batch culture in vitro, the fermentation parameters of four groups were compared. The results showed that both the 48 h cumulative and theoretical gas production of the CFR group were significantly lower than those of the RF and MRF groups (p < 0.05), but no significant difference was detected in the DM digestibility, pH, and MCP concentration between the groups (p > 0.05; Table 3).

As the microbiota of rumen fluid was domesticated after continuous batch culture in vitro, we wondered whether the NH₃-N utilization efficiency would increase. Here, we observed the fold change in NH₃-N concentrations against its initial concentration at 0 h of fermentation, which significantly differed between the groups after 48 h of in vitro fermentation. The highest NH₃-N concentration in the FRF, RF, and MRF groups was after 6–8 h of batch culture, with NH₃-N concentrations rising 4.2, 3.1, and 2.4 times, respectively, while that in the CRF group was at 4 h, with NH₃-N rising approximately 1.5 times (Figure 3A). The area under the curves of dynamic NH₃-N content in the CRF group was also significantly lower than that in the RF, FRF, and MRF groups (Figure 3B; p < 0.05).

With high-throughput sequencing of 16S rRNA genes, 29 phyla, 62 classes, 130 orders, 205 families, and 337 genera were identified. Both Chao1 and Shannon indexes of the CRF group and FRF group showed significantly lower compared with that of the RF group (p < 0.05; Figure 4). A significant difference in the microbial community structure among the RF, FRF, CRF, and MRF groups with PCoA analysis was also observed (R = 0.9798; ANOSIM, p = 0.001; Figure 5). The PC1 axis explained 48.48% of the variance, while the PC2 axis explained 26.72% of the variance of microbial community structure (Figure 5). This finding suggested that continuous batch culture or frozen treatment could lead to a reduction in microbial diversity.

At the phylum level, Bacteroidota, Firmicutes, and Proteobacteria were the top three dominant phyla (relative abundances higher than 5%); however, the relative abundances of the dominant phyla differed between groups (p < 0.05). Bacteroidota were the most dominant phylum in the RF group and MRF group, accounting for 60.21 and 60.78%, respectively, which were significantly higher than that of the FRF and CRF groups (33.54 and 24.97%, P<0.05). Firmicutes were the second most abundant phylum, on average, and it was the most dominant phylum in the FRF group and CRF group, accounting for 51.03 and 37.53%, respectively, which were significantly higher than that in the RF group and MRF group (25.37 and 14.98%, P<0.05). Proteobacteria had the highest abundance in the CRF group, accounting for 33.81%, which was significantly higher than that in the other three groups (p < 0.05), with 9.7, 11.48, and 20.6%, respectively. In addition, compared with the RF group, Spirochaetota and Verrucomicrobiota showed a significant decrease in the abundance in the CRF group (P<0.05; Table 4).

At the genus level, Prevotella, Succinivibrio, F082, and Rikenellaceae_RC9_gut_group were the dominant genera with relative abundances higher than 5% in both of the RF and MRF groups; For the FRF group, Streptococcus, Bacteroides, Megasphaera, Prevotella, Succinivibrio, and Rikenellaceae_RC9_gut_group were the dominant genera. For the CRF group, Succinivibrio, Prevotella, and Megasphaera were the dominant genera. The relative abundances of Prevotella, F082, and Rikenellaceae_RC9_gut_group in the CRF group were significantly lower than that in the RF and MRF groups (p < 0.05). In addition, the abundance of Succinivibrio in the CRF group was significantly higher than that in the RF and MRF groups by 5.4 and 1.9 times, respectively (p < 0.05); the relative abundance of Megasphaera in the CRF group was significantly higher than that in the RF and MFR groups by 25.5 and 9.1 times, respectively (p < 0.05; Table 5).

The core bacteria refer to the bacteria that exist in over 80% of all the samples. We observed that 155 OTUs out of the total 2,261 OTUs were core bacteria in all samples, and these core OTUs accounted for 77.5, 86.9, 90.1, and 85.0% in the abundance in the RF, FRF, CRF, and MRF groups, respectively. Moreover, the core bacteria that were identified mainly belonged to the five dominant genera, namely, Succinivibrio, Prevotella, Streptococcus, F082, and Megasphaera, whose average abundances among all samples were 20.1, 12.7, 9.1, 7.0, and 5.1%, respectively. In the CRF group, the core bacteria of Succinivibrio, Prevotella, and Megasphaera were effectively retained after 23 days of continuous batch culture, and the abundance of Succinivibrio was significantly higher than that in the other three groups (p < 0.05; Figure 6).

The area under the curve (AUC) was used to identify bacterial taxa that were associated with the NH₃-N utilization efficiency. We observed that eight genera were highly associated with the AUC of dynamic changes in NH₃-N content by cross-validating the error curves, including Streptococcus, Succinivibrio, Clostridium_sensu_stricto_1, p.251.o5, Oxalobacter, Bacteroidales_UCG.001, p.1088.a5_gut_group, and Elusimicrobium, which could explain 75.72% of the variation of the AUC of dynamic change in NH₃-N content in different groups (Figure 7).