Lactobacillus plantarum J39 was isolated from the ensiled total mixed ration containing baby corn husk as the main roughage in our previous study (Jaipolsaen et al., 2019). L. brevis BCC42336 and Pediococcus pentosaceus TBRC7603 were obtained from the Thailand Bioresource Research Center, Thailand, and were isolated from fermented Shona cabbage (Cleome gynandra) and fermented bamboo shoot (Dendrocalamus asper), respectively. To prepare starter inocula, all strains were grown in De Man Rogosa and Sharpe (MRS) broth (Lactobacilli MRS Broth, Difco & BBL, United States) at 37°C for 18 h.

An aerial part of the whole-plant corn (Z. mays L.) and Napier grass (P. purpureum) were used as raw plant materials. Both were harvested from local farms in Saraburi, Thailand (14° 38′ 02.5′ N, 101° 11′ 00.3″E). The fresh Napier grass and an above-ground portion of the corn plant were chopped to a length of 2–3 cm and anaerobically packed into 10 × 15-inch polyethylene bags for silage production. Using a Completely Randomized Design (CRD), silage treatments were divided into five groups that included plant material: (i) no inoculation with starter cultures (Control), (ii) inoculated with L. plantarum J39 at 10⁷ CFU/g fresh weight (T1), (iii) inoculated with L. brevis BCC42336 at 10⁷ CFU/g fresh weight (T2), (iv) inoculated with P. pentosaceus TBRC7603 at 10⁷ CFU/g fresh weight (T3), and (v) inoculated with a combination of L. plantarum J39, L. brevis BCC42336 and P. pentosaceus TBRC7603 at 10⁷ CFU/g fresh weight in a 2:1:1 ratio (T4) (Supplementary Figure 1). Each LAB-inoculated group was sprayed with a designated starter culture and mixed thoroughly to ensure homogeneity, and the non-inoculated control group was sprayed with sterile normal saline. Each experimental diet was thoroughly mixed before packing each of 2 kg of freshly chopped Napier grass or freshly chopped whole-plant corn of each individual treatment group into the polyethylene bags. After packing, all bags were vacuum-sealed using a small automatic vacuum sealing machine (Goodluck sealing machine©, model VS100S, China). All bags were stored at room temperature. The samples were collected in a three-point of mixing tray on Day 0 (D0), Day 3 (D3), and Day 7 (D7) after ensiling (Supplementary Figure 1). Then, the collected samples were pooled and placed in small polyethylene bags, vacuum-sealed, and stored at –20°C for fermentation qualities analyses. At the same time, a small portion of the collected samples was immediately ground to a small size (0.5–2 mm) using a laboratory-scale blender. The ground samples were stored in a –80°C freezer for DNA extraction.

The pH values of the fermented feeds were measured using a pH meter (pH Bench F20-Meter with Inlab Semi-Micro pH probe, Mettler-Toledo, Greifensee, Switzerland). For organic acid analysis, 12.5 g of the fresh samples were chopped and homogenized with 100 mL of 0.01 M H₂SO₄ in a lab-blender stomacher (Seward Laboratory, London). Then, one mL of the supernatant of each sample was filtered through a 0.45 μm nylon membrane syringe filter (Anpel laboratory technologies, Shanghai, China). Organic acid profiles were then analyzed using an HPLC–UV system (Alliance^® HPLC e2695 Separations Module and 2998 Photodiode Array Detector, Water, Massachusetts, United States), with an Aminex HPX-87H HPLC column (BIO-RAD^®, California, United States). Analyses were performed isocratically under the conditions of 0.6 mL/min, and 65°C with 0.01 M H₂SO₄ as the mobile phase.

For carbohydrate content analysis, fresh samples were dried at 60°C for 48 h. One gram of the dried sample was homogenized with 10 mL of distilled water for 1 h. After homogenization and centrifugation, the supernatant was filtered through a 0.45 μm nylon membrane syringe filter. Monosaccharides were analyzed using an HPLCUV system (Shimadzu; SIL-10A, Tokyo, Japan). Filtered distilled water in an Aminex HPX-87P HPLC column (BIO-RAD^®, California, United States) was used as the mobile phase. Total starch was analyzed using the K-TSTA 07/11 Megazyme kit according to AOAC International Method 996.11 (AOAC et al., 2007).

Silage quality was evaluated using the following equation Flieg (1938) cited in He et al. (2021).

Flieg’s score = 220 + (2 x%Dry matter–15)–40 × pH

According to the equation, silage was classified as poor, bad, low, medium, and good if the score values were below 20, 21–40, 41–60, 61–80, and above 80, respectively.

Each silage sample was homogenized to fine powder using a blender. An equal amount of 250 mg of each ground sample was used for DNA extraction using the DNeasy^® PowerSoil Kit (Qiagen^®, Germany) according to the manufacturer’s protocol. DNA samples were purified and concentrated using the Genomic DNA Clean and Concentrator (Zymo Research^®, United States) according to the manufacturer’s protocol. DNA purity was quantified using NanoDrop 8000 spectrophotometer (Thermo Fisher Science, United States) and stored at –20°C until use.

The V3 and V4 regions were amplified with the following primer pairs; forward primer (5′ TCGTCGG CAGCGTCAGATGTGTATAAGAGACAGCCTACGGGNGGC WGCAG 3′) and reverse primer (5′ GTCTCGTG GGCTCGGAGATGTGTATAAGAGACAGGACTACHVGGGT ATCTAATCC 3′) (Klindworth et al., 2013). The 16S rRNA amplicon library was constructed by using proofreading Q5^® High-Fidelity DNA polymerase (New England Biolabs^®, United States) with the following PCR cycle parameters: 98°C for 3 min, then 25 cycles of 98°C for 30 s, 54°C for 30 s and 72°C for 30 s, and the final extension was performed at 72°C for 2 min. PCR amplicons were purified using QIAquick^® Gel Extraction Kit (Qiagen^®, Germany) according to the manufacturer’s protocol. All 16S rRNA amplicon libraries were subjected to Illumina MiSeq next-generation sequencing (Macrogen Inc., Korea).

Raw sequences were qualitatively filtered and analyzed using TrimGalore (Krueger, 2015) and FastQC (Simon, 2010). Sequences longer than 200 bp containing homopolymers less than 6 bp, and a mean sequence quality score greater than 20 passed quality control criteria. Sequences were subsequently analyzed using Qiime2 (Hall and Beiko, 2018) and DADA2 (Callahan et al., 2016) to remove chimeras and dereplicate DNA reads, and call amplicon sequence variants (ASVs). ASVs were assigned to the lowest possible taxonomic rank using the BLAST + consensus taxonomy classifier against SILVA version 132. Data analysis was performed using Phyloseq (McMurdie and Holmes, 2013) under the R 3.5.2 environment (Rstudio Team, 2015; R Core Team, 2020). Statistical significance of beta diversity between groups was determined using Permutational Multivariate Analysis of Variance (PERMANOVA) (vegan package, function adonis) (Anderson, 2001). Differences in microbiome composition were estimated using the Kruskal-Wallis H-test, and all p-values were corrected using the false discovery rate method (Benjamini and Hochberg, 1995). A false discovery rate (FDR) of < 0.05 was considered statistically significant.

The statistical analyses were performed using the R 3.6.1 environment. Water-soluble carbohydrate contents and fermentation qualities were statistically analyzed using Analysis of Variance (ANOVA) according to the CRD design. Significant differences (P < 0.05) were calculated by Tukey’s test (Steel and Torrie, 1980). Permutational analysis of variance (PERMANOVA) statistical test was used to compare beta diversity among microbial communities (Oksanen et al., 2019).

Whole-plant corn and Napier grass were fermented with or without starter inoculation. The plant materials were inoculated to yield 10⁷ CFU/g fresh weight of different starter inocula, L. plantarum J39 (T1), L. brevis BCC42336 (T2), P. pentosaceus TBRC7603 (T3), and a combination of L. plantarum J39, L. brevis BCC42336 and P. pentosaceus TBRC7603 (T4). At the same time, normal saline was sprayed to the non-inoculated control group (Supplementary Figure 1). Silages were sampled at D0, D3, and D7 to determine fermentation quality. To determine content of fermentable carbohydrates before ensiling, total sugars and individual monosaccharides, including glucose, xylose, galactose, fructose, and starch content, were determined to compare the initial sugar contents in the different plant raw materials (Table 1). The contents of total sugars, glucose, galactose, and fructose were significantly higher in whole-plant corn than in Napier grass. In contrast, xylose contents were significantly higher in Napier grasses. Whole-plant corn also contained significantly higher levels of starch than Napier grass.

Dry matter content (DM), pH, and organic acid contents were also determined (Tables 2, 3). In whole-plant corn silage, the DM content was significantly decreased in all groups after 7 days of fermentation; however, no treatment effects were observed (Table 2). The pH values of corn silage were significantly decreased in all treatment groups and reached the plateau on D3 after ensiling. In addition, pH values did not differ significantly between treatments within the same days of ensiling. Lactic acid levels were not significantly different between treatments on D3 and D7 after fermentation. In contrast, lactic acid concentrations were significantly higher in all treatments on D7 compared to D3 after fermentation. Butyric acid concentration was undetectable in all treatment groups at both D3 and D7 after ensiling. On the other hand, pyruvic acid content of fresh whole-plant corn (D0) was highest in all treatment groups but decreased significantly after the start of ensiling. Our results clearly showed that starter inocula did not provide significant advantages for the fermentation of corn silage.

In contrast to whole-plant corn ensiling, starter inocula were beneficial for ensiling Napier grass (Table 3). All starter inocula were able to significantly improve the fermentation quality of Napier grass silage (Table 4). The pH values of Napier grass silage in the inoculated treatments (T1–T4) were significantly lower than those of the control group on D3 and D7 after ensiling. In addition, the pH values of Napier grass silage T1 and T4 groups had the lowest pH values on both D3 and D7 after ensiling. Lactic acid concentration in inoculated Napier grass was significantly higher than in the control group. In particular, Napier grass silage T1 and T4 groups had the highest lactic acid levels on D3 and D7 after ensiling. Butyric acid was not detected in all treatments at all-time points. Unlike whole-plant corn, pyruvic acid was not detected in fresh Napier grass (D0); however, a low concentration of pyruvic acid was detected at D3 and D7 after fermentation (Table 3). Interestingly, all starter inocula in this study showed a positive effect on Napier grass ensiling but not on whole-plant corn ensiling (Table 4).

Bacterial composition was monitored at D0, D3, and D7 time points of ensiling using 16S amplicon sequencing. On average, 97,943 ± 16,360 and 103,285 ± 13,114 reads were obtained from whole-plant corn and Napier grass silage, respectively (Supplementary Table 1). The rarefaction curves of all sampling time points reached a plateau, indicating sufficient sequencing depth to describe bacterial profiles during plant ensiling (Supplementary Figure 2). Bacterial taxonomy was classified based on amplicon sequence variants (ASVs) (Supplementary Table 2). Bacterial community structures were determined in five treatment groups: control, T1, T2, T3, and T4 at D0, D3, and D7 of ensiling (Figure 1). Bacterial community profiles in whole-plant corn silages were dominated by phyla Firmicutes, Proteobacteria, and Cyanobacteria (Figure 1A). During the ensiling process of whole-plant corns, the relative abundance of Firmicutes showed increasing trends. In contrast, the relative abundance of Proteobacteria and Cyanobacteria decreased in all groups compared to D0. Silage quality could be indicated by the amount of lactic acid produced by LAB. Here, the relative abundance of LAB increased in both the non-inoculated control and the inoculated groups (T1, T2, T3, and T4), and the dominant genus observed on D3 and D7 of whole-plant corn silage was Lactobacillus (Figure 1B). Weissella, Pediococcus, Enterococcus, and Lactococcus were other predominant lactic acid bacteria that showed decreasing abundance, while dominance shifted to Lactobacillus on D3 and D7. Interestingly, the relative abundance of Pediococcus was higher on D3 than on D0 before decreasing on D7 in all groups. Notably, the whole-plant corn silage inoculated with Pediococcus (T3) was dominated by the genus Lactobacillus, as observed on D3 and D7 of fermentation. Nevertheless, the LAB patterns were mostly similar in the non-inoculated control and inoculated groups. In addition, Kruskal-Wallis H was used to determine significant differences in bacterial abundance before (D0) and during ensiling (D3 and D7) (Figure 1C). The results showed that the abundance of Lactobacillus increased significantly in all groups, while the abundance of Actinobacteria (Brachybacterium and Kocuria), Bacteroidetes (Chryseobacterium), Firmicutes (Exiguabacterium, Weisella, Lactococcus, and Enterococcus), Proteobacteria (Rhodoligotrophas, Methylobacterium, Sphingomonas, Enterobacter, Escherichia, Klebsiella, Acinetobacter, and Stenotrophomonas) were significantly reduced during ensiling.

Bacterial communities in Napier grasses were dominated by three main phyla, Cyanobacteria, Firmicutes, and Proteobacteria, while Actinobacteria dominated only at D0 (Figure 2A). Cyanobacteria were the dominant population in the pre-silage (D0) with an abundance of over 70%. After ensiling, the abundance of Cyanobacteria decreased in all groups. Firmicutes became the dominant group during ensiling in both the non-inoculated control group and the LAB-inoculated groups (T1, T2, T3, and T4). Interestingly, the abundance of Proteobacteria was reduced in the four LAB-inoculated groups but not in the non-inoculated control group. The relative abundance of the major LAB genera, Lactobacillus, Weissella, Pediococcus, Enterococcus, and Lactococcus, were determined and compared (Figure 2B). The dynamic changes of LAB among Napier grass silages were strikingly observed in the LAB-inoculated groups, while little change in the abundance of LAB was observed in the non-inoculated control group. From D0 to D3, the relative abundance of Lactobacillus in T1, T2, and T4 shifted sharply from 1.4 to ∼50%, from 0.7 to ∼30%, and from 0.9 to 31%, respectively. As expected, T3, the Napier grass inoculated with P. pentosaceus, had the highest abundance of Pediococcus at D3 and D7 compared to the other LAB-inoculated groups. Our observations suggest that starter culture strains influenced LAB diversity in ensiling Napier grass. Accordingly, Kruskal-Wallis H-test showed a significant increase of Lactobacillus in T1, T2, and T4, while Pediococcus was significantly present in T3 (Figure 2C). In contrast to the LAB-inoculated groups, the non-inoculated group was not dominated by a single group of LAB but showed a greater diversity of LAB, namely Enterococcus, Lactococcus, and Pediococcus. Both control and LAB-inoculated groups (T1, T2, T3, and T4) showed significant reduction of Proteobacteria (Rhodoligotrophos, Aureimonas, Methylobacterium, Sphingomonas, Pantoea, Acinetobacter, and Pseudomonas) and Actinobacteria (Curtobacterium). However, the abundance of Enterobacter, Escherichia, and Salmonella, which are potentially pathogenic bacterial groups was significantly increased in the control group during Napier grass ensiling.

A principal coordinate analysis (PCoA) based on the unweighted-UniFrac was used to compare the similarity of bacterial profiles during ensiling (Figure 3). PCoA showed a clear separation of the bacterial composition of the whole-plant corn silages at the beginning (D0), D3 and D7 of ensiling (Figure 3A). The bacterial profiles of the control and inoculated groups (T1, T2, T3, and T4) were similar at D0 and showed a shift at D3 and D7. Bacterial profiles of whole-plant corn were similar at D3 and D7 despite the addition of inoculants. In contrast, the PCoA of Napier grass silages showed three distinct clusters (Figure 3B). The bacterial profiles of all five groups were similar at D0. During ensiling, the bacterial communities of the control and inoculated groups (T1, T2, T3, and T4) were clearly separated at D3 and D7, indicating that starter inoculation affected the bacterial diversity of Napier grass silages. Consistently, Permutational Multivariate Analysis of Variance (PERMANOVA) of bacterial communities based on the unweighted-UniFrac distance matrix was used to determine the factors affecting bacterial profiles in the different sample groups (Table 5). Bacterial profiles in whole-plant corn and Napier grass were significantly different (P < 0.05). All four formulations of LAB starter cultures had a significant effect on bacterial diversity in Napier grass silage (P < 0.05), while only a combination of LAB consortia containing L. plantarum, L. brevis, and P. pentosaceus (T4), showed a significant effect on bacterial profiles in whole-plant corn silage. In Napier grass, the different LAB strains had a significant effect on bacterial communities (P < 0.05), except that the group inoculated with a combination of LAB consortia (T4) had similar bacterial profiles to the group inoculated with a single strain, namely L. plantarum (T1) and P. pentosaceus (T3).