K (Reyan No. 4), C (South China No. 7), and B were cultivated in the experimental base of the Chinese Academy of Tropical Agricultural Sciences (19°58′E, 19°52′N). The first cut K of the vegetative stage (~1.5–1.8 m in height) was harvested and chopped into about 2 cm pieces by a grass-shredding machine, and then C and B in the growing stage were cut in the same manner (Donghong No. 1, Donghong Mechanical Equipment Co., Ltd., China). The number of L (Snow Brand Seed Co., Ltd., Japan) was 1.0 × 10⁵ CFU/g (Zi et al., 2021).

In total, six different treatments were carried out in our current research as follows: B: Broussonetia papyrifera, BL: Broussonetia papyrifera + L. plantarum, C: cassava foliage, CL: cassava foliage + L. plantarum, K: king grass, and KL: king grass + L. plantarum. According to the manufacturer's guidelines, the inoculants were dissolved in sterile distilled water and then mixed thoroughly with the grass. An equal amount of sterile distilled water was added to the control group. Every treatment was carried out in triplicate. Briefly, 200 g of K, C, and B was blended with inoculants, and the mixture was placed in plastic bags (35 × 12 × 5 cm; Guozhong Packing Co., Ltd., Haikou, China). A total of 18 bags (six treatments × three replicates) were sealed by a vacuum packaging machine (DZQ-380B; Ansheng Machinery Co., Ltd., Quanzhou, China) and stored at normal temperature (25~30°C). The chemical composition, organic acid, and microbial community were determined after 30 days of fermentation.

Samples were oven-dried at 65°C for 72 h to determine the DM content, and dried materials were ground for chemical analysis. CP and ether extract (EE) were determined according to the Guidelines of the Association of Official Analytical Chemists (AOAC, 2005). The concentration of neutral detergent fiber (NDF) and acid detergent fiber (ADF) and the content of WSC were determined as described previously (He et al., 2020).

The fermentation quality of silage was determined using distilled water extracts. Briefly, 50 g wet silage was blended with 200 ml distilled water, followed by incubation at 4°C for 24 h and then filtration. The pH (Thunder magnetic phs-3c pH meter) and concentrations of LA, AA, propionic acid (PA), butyric acid (BA), and NH₃-N were measured as established previously (Mao et al., 2014).

Microbial counts were analyzed using the plate count method as described previously (Li et al., 2018). Briefly, 20 g silage samples were mixed with 180 ml sterilized saline, and then LAB, coliform, yeasts, and molds were enumerated on Man, Rogosa, and Sharpe (MRS) agar, violet red bile agar, and rose Bengal agar, respectively.

The impacts of L. plantarum were investigated using a two-way ANOVA in SAS 9.3 software (SAS Institute Inc., Cary, NC, United States). Duncan's multiple range test was adopted to identify significant differences, and P < 0.05 was considered statistically significant.

Table 1 shows the chemical and microbial compositions of C, K, and B. In this study, for C and K, the concentration of NDF and ADF were higher compared with previous reports, while the contents of DM, OM, CP, and ether extract (EE) were lower or comparable to the previous reports (Li et al., 2019). For B, the level of CP was comparable to the previous reports, and the level of NDF was higher compared with previous reports, while the levels of ADF and DM were lower (Si et al., 2018). The content of WSC plays a significant role in assessing fermentation quality. The content of WSC was 5.21, 7.21, and 6.43% in C, K, and B in this study. This can ensure the fermentation quality in theory (Ni et al., 2018). However, in addition to WSC, LAB quantity also needs to reach the threshold. Well-preserved silage needs no less than 10⁵ CFU/g FM of LAB (Cai et al., 1998). The number of LAB, coliform, yeasts, and mold in the fresh C was 3.89, 1.53, 2.06, and 0 Log₁₀ CFU/g FM, respectively. It was 4.22, 1.77, 2.78, and 3.04 Log₁₀ CFU/g FM, respectively, in fresh K and 3.52, 2.08, 2.63, and 0.78 Log₁₀ CFU/g FM, respectively, in fresh B. Compared with harmful bacteria, LAB were not dominant in quantity among the three forages, at the beginning of ensiling, yeasts would grow and compete for energy materials with LAB and could not ensure good silage quality, illustrating that silage inoculants, such as LAB inoculants, were imperative for silage.

Table 2 provides the fermentation characteristics of three species of tropical forages. For fermentation quality, the pH of ensiling is the most important key index. The pH was remarkably decreased after fermentation in CL, KL, and BL compared with the control groups. However, the pH values of the inoculant-treated groups were ≥4.2, and related research have indicated a similar pH of L. plantarum treatment (Yang et al., 2019; Ren et al., 2020). The reason may be that there was not enough WSC for LAB growth and reproduction, no more LA was produced, and the pH value cannot be reduced to a lower level (Yan et al., 2019). In addition, inoculants reduced the contents of NH₃-N compared with the control group. NH₃-N is a universal index of protein hydrolysis, and it usually reflects the levels of peptide bond hydrolysis and amino acid or peptide deamination. The smaller the value, the less the protein is decomposed, which means a better storage quality (Wang et al., 2019). The remarkably decreased content of NH₃-N in the inoculant-treated silage might be a result of lower pH values, which could inhibit the activities of some harmful bacteria and proteases and effectively conserve nutrients (Wang et al., 2019).

In CL, the content of LA was remarkably elevated, and there was no significant change in acetic acid (AA), besides the content of PA in the inoculant-treated groups reduced by about a fifth compared with the control group. Furthermore, the content of BA was reduced to 0. In KL, the content of LA, AA, and PA was similar to the control group, and BA was never detected. In BL, the content of LA was remarkably increased while AA was significantly decreased than the control group, and there was no significant change in PA. Furthermore, the content of BA was never detected. Furthermore, the content of BA was reduced to 0 in the treatment of all three treatments, manifesting that K, C, and B treated by L. plantarum were well-preserved, which was consistent with a previous study (Lv et al., 2020).

These results revealed that the addition of L. plantarum in the ensiling process could efficiently promote the fermentation quality.

A total of 1,439,466 raw reads were obtained. After splicing and filtering, 1,311,196 clean tags were generated. Each sample produced at least 69,299 clean tags, with an average of 72,844 clean tags.

Table 3 presents the α-diversity of the bacterial diversity of K, C, and B silages. The average Good's coverage for all samples was over 0.99, suggesting that the sequencing process was sufficient to describe the changes in the bacterial community. Shannon and Simpson stand for diversity, and Chao1 and ACE represent richness. Figure 1 is principal coordinates analysis (PCoA), i.e., principal coordinate analysis, which is a non-constrained data dimensionality reduction analysis method, which can be used to study the similarity or dissimilarity of sample community composition. It was used to search the relevance among the community structures of bacteria during ensiling. Next, the changes in the bacterial community during fermentation were evaluated by a high-throughput sequencing technique based on the 16S ribosomal DNA gene. Figure 2 presents the relative abundance of bacterial communities in C, K, and B. The linear discriminant analysis (LDA) effect size (LEfSe) method was used in Figure 3 to examine the discrepancies of microbial communities between groups and probe the specific bacteria in each group (LDA score > 4.0).

In BL, there was no significant change in diversity, while the richness was decreased compared with the control group. PCoA analysis showed the control group and silages treated with L. plantarum were not separated, indicating that the microbial community was not changed remarkably during the ensiling process. The combination of α-diversity and β-diversity showed that the effect of bacterial diversity and community structure of L. plantarum on B was limited. At the phylum level, the total contents of Firmicutes and Proteobacteria were more than 90% in both B and BL. Similar results have been found in Eucalyptus nutans of four different altitudes over the Tibetan Plateau (Ding et al., 2020). The same phenomenon has been observed in cassava foliage (Li et al., 2020). With inoculant L. plantarum, the content of Firmicutes was decreased, while the relative abundance of Proteobacteria was increased. At the genus level, the dominating bacteria in the B and BL groups were Weissella and Enterobacter. In addition, Lactobacillus, Pantoea, Escherichia-Shigella, and Raoultella were also included. Previous research has shown that Weissella initiates the silage fermentation, and then the predominant bacteria gradually turn into Lactobacilli, which are more tolerant to low pH (Cai et al., 1998). However, Lactobacillus was less (16.1–8.3%), and Enterobacter was more (28.4–31.4%) compared with the control group. As the core microorganism in the silage process, Lactobacillus, the beneficial bacteria, can produce lactic acid and determine the quality of silage (Cai et al., 1998). Enterobacteriaceae can grow under anaerobic conditions and trigger protective mechanisms under adverse conditions, including low pH. In this study, Enterobacter was the subdominant bacteria after ensiling, existing in silages that are non-pathogenic (Santos et al., 2016). However, their increase is undesirable because they compete with LAB for WSC at the initial stage of ensiling. Other studies also confirmed that Enterobacter is harmful to silage because it can compete with LAB for nutrients through fermenting lactic acid into acetic acid and producing ammonia, leading to nutrient loss (Silva et al., 2016). Our finding showed that L. plantarum had a weak competitive effect in this treatment, and the high content of oxygen or the high content of Escherichia coli attached to the raw materials might lead to the stable period dominated by Enterobacter (Chen et al., 2021), which might also explain why there was no significant change in bacterial community composition with or without inoculants. In our study, LAB was not the dominant bacteria in B. It is possible that the competitiveness of LAB already existing or the inoculant L. plantarum was weaker than that of the dominant bacteria but the reduction of relative abundance did not affect the fermentation quality of BL. It can be seen from the content of LA and AA. The LDA showed that Proteobacteria was the most abundant phylum in the BL group. Gammaproteobacteria was the most abundant class, Enterobacteriaceae and Leuconostocaceae were the most abundant family, Enterobacteriales was the most abundant order, and Weisella, Enterobacter, Pantoea, Escherichia_Shigella, and Raoultella were the most abundant genera. Moreover, uncultured_bacterium_g_Weisella,endophytic_bacterium_HA01, Pantoea_ sp_LI0_RH04, uncultured_ bacterium_ g_ Escherichia_ Shigella, and uncultured_ bacterium_ g_ Rautella were the most abundant species.

In CL, the diversity and richness were increased compared with the C group. For β-diversity, Figure 1 showed that C and CL were separated. The aforementioned diversity analysis showed that L could affect the bacterial diversity and community structure of C silage. At the phylum level, Firmicutes and Proteobacteria were the dominating bacteria. With inoculant L. plantarum, the content of Proteobacteria was decreased, while Firmicutes was increased. At the genus level, Lactobacillus accounts for a very high proportion in C and CL groups. In particular, C and CL groups contained Pseudomonas. When L. plantarum was added, its relative abundance became less, while the relative abundance of Lactobacillus and Enterobacter was increased, and Acinetobacter appeared. Pseudomonas is harmful to silage quality and nutrient preservation. In this group, the L. plantarum treatment notably reduced the relative abundance of Pseudomonas, and such a finding was similar to a previous study (Zi et al., 2021). The LDA showed that, in the C group, Pseudomonadaceae was the top abundant family, Pseudomonas was the most abundant genus, and uncultured _bacterium_ g_ Pseudomonas was the most abundant species. In the CL group, Lactobacillus_ paraplantarum was the most abundant species.

In KL, the diversity and the richness were decreased. For β-diversity, Figure 1 indicated that a clear separation of bacterial communities was discovered in K and KL. Therefore, L. plantarum could affect the bacterial diversity and community structure of K silage. It could be explained that L. plantarum decreased the pH, inhibited the growth of harmful microorganisms, and promoted the growth of LAB species. At the phylum level, Firmicutes and Proteobacteria had absolute advantages in relative abundance in K and KL, and compared with K, the content of Firmicutes was decreased while Proteobacteria was increased. Besides, Cyanobacteria appeared in the KL group. At the genus level, Lactobacillus accounts for a very high proportion in K and KL groups. There were Gluconacetobacter and Acinetobacter in the KL group, and Gluconacetobacter is widely used in wine and vinegar production. It is a bacterial strain that consumes ethanol and sugar to produce acetic acid (Raspor and Goranovic, 2008). Therefore, the higher the relative abundance of Gluconacetobacter, the higher the acetic acid content. Acinetobacter species are aerobic bacteria. However, previous studies have found that some Acinetobacter can survive in an anaerobic environment with acetic acid as substrate (Fuhs and Chen, 1975), which consumes energy from carbohydrates and results in DM loss (Pitt, 1986). Another research has validated this finding, evidenced by increased the content of AA, the abundance of Acinetobacter, and the loss of DM in silage treated with Lactobacillus brucelli (Oliveira et al., 2017). In this study, when L. plantarum was added to K, and the relative abundance of Acinetobacter or Lactobacillus was increased or decreased, respectively. This finding might be attributed to the high content of acetic acid in the KL group, which provided a suitable growth environment for Acinetobacter. The microbial genus level of K had been a relatively single community mainly composed of Lactobacillus. When L. plantarum was added as an inoculant, the amount of available matrix remains unchanged, which may aggravate intraspecific and interspecific competition so that bacteria with stronger acid production, growth ability, and stress tolerance became dominant bacteria. However, the relative abundance of Lactobacillus decreased as a whole. To explore this problem, it is better to identify the community composition to the species level. The LDA showed that, in the K group, Firmicutes was the most abundant phylum, Bacilli was the most abundant class, Lactobacillales was the most abundant order, Lactobaillaceae was the most abundant family, Lactobacillus was the most abundant genus, and uncultured_ bacterium_ g_ Lactobailus and Lactobacillus_ panis were the most abundant species. Alphaproteobacteria was the most abundant class in the KL group, Acetobacterales was the most abundant order, Acetobacteraceae was the most abundant family, Gluconacetobacter was the most abundant genus, and Lactobacillus_ Casei, Giluconacetobacter_ Liquefaciens, and Lactobacillus_ ginsenosidimutans were the most abundant species.