Keystone Taxa Lactiplantibacillus and Lacticaseibacillus Directly Improve the Ensiling Performance and Microflora Profile in Co-Ensiling Cabbage Byproduct and Rice Straw

Ensiling has been widely applied to cope with agricultural solid waste to achieve organic waste valorization and relieve environmental pressure and feedstuff shortage. In this study, co-ensiling of cabbage leaf byproduct and rice straw was performed with inoculation of Lactiplantibacillus plantarum (LP) to investigate the effects of inoculation on ensiling performance and microflora profiles. Compared to the control, LP inoculation preserved more dry matter (DM) content (283.4 versus 270.9 g·kg−1 fresh matter (FM) on day 30), increased lactic acid (LA) content (52.1 versus 35.8 g·kg−1 dry matter on day 15), decreased pH (3.55 versus 3.79 on day 15), and caused accumulation of acetic acid (AA), butyric acid (BA), and ammonia. The investigation showed that LP inoculation modified microflora composition, especially resisting potential pathogens and enriching more lactic acid bacteria (LAB) (p < 0.05). Moreover, Lactiplantibacillus and Lacticaseibacillus were identified as the keystone taxa that influenced physicochemical properties and interactions in microflora. They were also the main functional species that directly restrained undesirable microorganisms (p < 0.05), rather than indirectly working via metabolite inhibition and substrate competition (p > 0.05). The results of this present study improve the understanding of the underlying effect of LP inoculation on improving silage quality and facilitate the bio-transformation of cabbage byproduct and rice straw as animal feed.


Introduction
With population and living standard rapidly rising, a series of social problems have been more prominent, including a shortage of food, food-feed competition for arable land, and a vast amount of waste from agriculture and the food processing industry [1,2]. Vegetable wastes (e.g., leaves of cabbage, cauliflower, and amaranth) and rice straw are two common types of waste generated from consumption of leaf vegetables and rice as important food sources. According to the statistics, 800 million tons of vegetable waste and 900 million tons of crop straw are annually generated from harvesting to sale in China [2][3][4]. Generally, vegetable wastes are highly moist and perishable, while rice straw is dry and degrades slowly. This would pose huge challenges to the downstream processing of The values are shown as the mean ± standard deviation of three replicates.

Silage Set-Up and Sampling
To ensure suitable moisture and enough energy for FM, the formula of mixed raw material was 180 g cabbage leaf, 40 g rice straw, and 40 g corn flour per bag. To inoculate more equably, 5 mL LP inoculation, as in Section 2.1., was diluted with 100 mL 0.8% sterile NaCl (Sinopharm, Shanghai, China) solution and shaken well. Diluted LP inoculation of 5 mL was added per bag for the treatment group (LPGP, LP content: 5.9 × 10 5 CFU·g −1 FM), while equal-volume sterile NaCl solution was added in raw material to replace the inoculation and served as the control group (CKGP, without inoculation). The mixed raw material was packed into PET plastic bags (23 cm × 30 cm) and sealed with a vacuum sealer (Blueberry 320X, Shanghai Inuo Packaging Materials Co., Ltd., Shanghai, China). Silage bags were incubated at 30 • C for 3, 7, 15, and 30 days. A total of 30 silage bags was divided into 2 treatments and 3 replicates for 5 sampling times (2 × 3 × 5 = 30). The silage sample of 20 g was taken from each treatment for analysis of microbial community and physicochemical parameters.

Analytical Sample Preparation
The first lixivium was prepared by each mixing ensiling sample of around 20 g with 180 mL sterile NaCl solution (0.8% w/v) in a 500 mL conical flask. The mixture was agitated at 30 • C for 2 h using a rotary shaker at 200× rpm and filtered through four layers of sterile medical gauze under negative pressure. The filtrate was collected, centrifuged at 4 • C for 20 min at 10,000× g to obtain the sediment for DNA extraction and sequencing. The acquisition of the second lixivium was similar to the first one except that NaCl solution was substituted with sterile water. The supernatant was gathered after centrifugation and used for analysis of physicochemical parameters [17]. The silage sample was oven-dried at 65 • C until the weight was stable, and then ground and sieved through a 1 mm screen for subsequent nutritional composition analyses.

Analyses of the Physicochemical Properties
The physicochemical parameters included the principal nutritional composition (the content of crude protein (CP), crude fat (CF), neutral detergent fiber (NDF), acid detergent fiber (ADF)) and fermentation characteristics (DM, WSC, pH, organic acids (FA, LA, AA, PA, and BA), ethanol, and ammonia nitrogen/total nitrogen content).
The silage samples were oven-dried at 65 • C for 72 h to determine the DM content [2]. As in Section 2.3., 0.2 g dry sample was digested under 430 • C with H 2 SO 4 (18.4 mol/L; Sinopharm, Shanghai, China) for 1 h and then the CP content was determined using a Kjeldahl nitrogen analyzer [10]. CF content was determined as the DM loss via Soxhlet extraction (B-811, BUCHI, Flawil, Switzerland) using ethanol as solvent and oven-dried at 65 • C for 72 h. The defatted sample was subsequently used to measure ADF and NDF contents using an automatic Fibretherm (C. Gerhardt, Königswinter, Germany) per the manufacturer's instructions [2,18]. For fermentation characteristics, the contents of organic acids and ethanol were detected using a high-performance liquid chromatography system (20AVP, Shimadzu Corp., Kyoto, Japan) with a RID-10A refractive index detector (AminexHPX-87H column (300 × 7.8 mm), Bio-Rad, Hercules, CA, USA). The oven temperature was set at 65 • C and 0.005 mol/L H 2 SO 4 solution was used as mobile phase at a velocity of 0.8 mL/min. The retention time of LA, FA, AA, PA, BA, and ethanol was 9.9, 10.3, 11.3, 13.0, 14.7, and 16.2 min, respectively. All the standard reagents of organic acids and ethanol were purchased from Sigma-Aldrich Co., Ltd. (Burlington, Vermont, USA). Total organic acid (TOA) was calculated using the summation of LA, FA, AA, PA, BA. The silage quality was evaluated using Flieg's score based on the organic acid ingredient (score ≥ 81, very good; 61 ≤ score < 81, good; 41 ≤ score < 61, medium; 21 ≤ score < 41, bad; score < 21, very bad) [4]. Ammonia-N was quantified using Nessler's reagent (Hach, Loveland, CO, USA) and a spectrophotometer (DR2800; Hach, USA) at 420 nm. The second lixivium, as in Section 2.3, was used to determine pH with a digital pH meter (PB-10, Sartorius, Arvada, CO, USA) per the manufacturer's instructions. Around 20 mg dry sample, as in Section 2.3 was parcelled using tin paper and used to detect total nitrogen using an elemental analyzer (PerkinElmer SERIES ll 2400, Waltham, MA, USA). The WSC was extracted from a fresh sample by boiling water and quantified via a microplate reader (BioTek, Winooski, VT, USA) at 630 nm using the anthrone method [2,19].

DNA Extraction and MiSeq Sequencing
For the microbial community, total DNA was extracted using an E.Z.N.A . ® soil DNA Kit (Omega Bio-Tek, Norcross, GA, USA) per the manufacturer's instructions. DNA purity and concentration were measured using a NanoDrop 2000 UV-vis spectrophotometer (Thermo Scientific, Wilmington, NC, USA). All DNA samples were stored at −80 • C until required. PCR amplification of prokaryotic 16S rDNA and eukaryotic internal transcribed spacer (ITS) regions were performed as described previously [10]. The V3-V4 hypervariable regions from 16S rRNA were amplified barcoded fusion primers 338F (ACTCCTACGGGAGGCAGCAG) and 806R (GGACTACHVGGGTWTCTAAT). The amplification of the ITS region from ITS rRNA used ITS1F (CTTGGTCATTTAGAGGAAG-TAA) and ITS2R (CTTGGTCATTTAGAGGAAGTAA). The DNA quality was confirmed by 1.5% agarose gel electrophoresis. PCR products were sent to Majorbio Bio-pharm Technology Co., Ltd. (Shanghai, China) for further purification, extraction, and sequencing, as described previously [20].

Statistical Analysis
Fermentation characteristics and nutrition content were analyzed using two-way ANOVA for a 2 × 5 (2 treatments × 5 sampling times) full factorial experimental design with three replicates (IBM SPSS 26.0, New York City, NY, USA). The significant differences between the two groups were determined by the Tukey test (α = 0.05 and P critical = 0.05). Correlations among dominant genera (RA > 1%) were evaluated using Spearman correlation analysis. Only important interrelationships were considered (|Spearman coefficient| > 0.7, p < 0.01) in the co-occurrence network and visualized using Gephi 0.9.2. The relationships among the physicochemical indexes and dominant taxa (the top 5 bacterial and fungal genera) in these two groups were evaluated using RDA (Canoco 5.0, Ithaca, NY, USA). The direct and indirect effects among the functional microflora, inhibited microflora, microbial metabolite, and fermentation substrate were evaluated using partial least squares path modeling (PLS-PM) (SmartPLS 3.0, Boenningstedt, Schleswig-Holstein, Germany) [24]. Functional and inhibited microflora were identified using factor analysis (IBM SPSS 26.0, New York City, NY, USA).

The Physicochemical Properties of Silage
Upon the completion of ensiling process, physicochemical properties including fermentation characteristics and nutritional composition are shown in Tables 2 and 3, respectively. The pH value and content of DM and WSC significantly declined as the fermentation time elongated while the content of LA, AA, and ethanol significantly increased in all groups. The pH was nearly neutral at the initial phase and slowly decreased to below 4.0 on day 15. Interestingly, the LP inoculation resulted in a lower pH value (3.55~5.87 in LPGP vs. 3.79~5.93 in CKGP, p < 0.001). In particular, pH rapidly decreased to below 4.0 on day 3 in LPGP. The DM content rapidly declined from 309.5 to 270.9 g·kg −1 FM in CKGP. Furthermore, the LP inoculation preserved more DM content (283.4~310.3 g·kg −1 FM in LPGP vs. 270.9~309.5 g·kg −1 FM in CKGP, p < 0.001). The WSC content decreased rapidly on day 3 and kept stable after day 7 in CKGP. The LPGP presented higher WSC content at day 3 than CKGP (p < 0.05), while there was no significant difference after day 7 (p > 0.05). The organic acid and ethanol content increased first to the maximum (35.8 g·kg −1 DM) and then declined to 22.3 g·kg −1 DM at the end in CKGP. Regarding LPGP, the LP inoculation resulted in higher LA content (0~52.1 g·kg −1 DM in LPGP vs. 0~35.8 g·kg −1 DM in CKGP, p < 0.001) of silage samples, while there were lower contents of AA, and BA (p < 0.001), which led to higher LA/TOA and lower AA/TOA, BA/TOA (Table 4). Regarding the silage quality based on the organic acid ingredient, only the samples on day 15 were evaluated as "good" (Flieg's score > 60) in CKGP and other scores were evaluated below 60. However, the LP inoculation improved the silage quality (very good; Flieg's score ≥ 81) during the overall fermentation process. Furthermore, LP inoculation played no significant role in ethanol production. Ammonia-N significantly accumulated in CKGP (p < 0.05; 6.4~11.9% during ensiling), and the LP inoculation significantly decreased ammonia-N content during the whole process (p < 0.05, below 7% during ensiling).   For nutritional components, the CF content decreased while the contents of ADF and NDF increased in both groups (Table 3). Among the various nutritional components, lignocellulose was less preferred and left, leading to an increasing mass ratio in DM in both groups. Moreover, the LP inoculation played no significant role in the accumulation of CF and CP (p > 0.05) but significantly decreased the content of ADF and NDF (p < 0.05) on day 3 and day 15. The ADF content also significantly decreased at the mature phase (day 30, p < 0.05).

Silage Bacterial and Fungal Composition
Based on 16S and ITS rDNA sequencing, the coverage index of all samples was above 0.99 (Table 5), which indicated the DNA sequencing results were representative of the microbial community. The richness was evaluated using Chao1 and Ace indexes while the diversity was done by Shannon and Simpson indexes. The Shannon index of the bacterial community significantly declined, and the Simpson index of the bacterial community significantly increased in LPGP compared with that in CKGP (p < 0.05); moreover, there was no significant difference between Shannon and Simpson indexes of fungi, Chao1 and Ace indexes of bacteria and fungi (p > 0.05).
Regarding the fungal community, the raw material was mainly rich in diverse saprotrophs fungi (such as unclassified_g__Wallemia, Aspergillus penicillioides, RA 99.6%). However, in CKGP, a few epiphytic animal-pathotroph fungi rapidly accumulated during the ensiling process with RA dynamically changing from 16.7% to 53.4% and kept an RA of 53.4% on day 30 (such as unclassified_g__Fusarium, unclassified_g__Trichosporon). With the LP inoculation, the RA of animal pathogens was significantly reduced compared to CKGP (2.6-18.1%, p < 0.05) and presented an RA of 8.8% at the mature phase. In total, we found LP inoculation significantly declined the animal-pathogenic fungal community during ensiling.

Co-Occurrence Network Analysis for Correlations in Microbial Community
To better understand how the exogenous LP inoculation interacted with the natural microbial community, co-occurrence networks in two groups were constructed based on the correlations among dominant genera in Figure 3. Notably, the topological characteristics of the co-occurrence network became more complex with the LP inoculation, especially for the average weighted degree (2.996 in CKGP and 6.372 in LPGP, Table S1). Furthermore, the node number, edge number, and diameter of the network increased by 19.4%, 45.0%, and 66.7% in LPGP compared with CKGP. Interestingly, the most abundant genera in CKGP (Pediococcus, unclassified_f__Enterobacteriaceae, unclassified_f__Dipodascaceae, Fusarium, Aspergillus) had few correlations with others, especially negative correlations (Figure 3a). However, concerning Weissella and Penicillium, whose average RA (10.63% and 2.6%, respectively) was relatively lower in CKGP, there existed obvious negative correlations (degree 11 and 5, respectively) with other taxa, such as Pseudomonas. Interestingly, with LP inoculation, Lactiplantibacillus and Lacticaseibacillus turned out to be the most abundant genus (total RA 83.74% from days 3 to 30), which were blinding as keystone taxa with the most negative correlations (degree 12). Furthermore, bacterial genera presented more positive correlations with each other in LPGP compared with CKGP and declined during anaerobic fermentation (Figure 3). Considering LPGP, there were more LABs (LP, RA 3.5%) than in CKGP at the initial phase (day 0). Upon the completion of the ensiling process, silage with LP inoculation enriched more LABs (such as LP, Lcb. rhamnosus, RA 93.7%) and fewer potential pathogens (RA 1.2%).
Regarding the fungal community, the raw material was mainly rich in diverse saprotrophs fungi (such as unclassified_g__Wallemia, Aspergillus penicillioides, RA 99.6%). However, in CKGP, a few epiphytic animal-pathotroph fungi rapidly accumulated during the ensiling process with RA dynamically changing from 16.7% to 53.4% and kept an RA of 53.4% on day 30 (such as unclassified_g__Fusarium, unclassified_g__Trichosporon). With the LP inoculation, the RA of animal pathogens was significantly reduced compared to CKGP (2.6-18.1%, p < 0.05) and presented an RA of 8.8% at the mature phase. In total, we found LP inoculation significantly declined the animal-pathogenic fungal community during ensiling.

Co-Occurrence Network Analysis for Correlations in Microbial Community
To better understand how the exogenous LP inoculation interacted with the natural microbial community, co-occurrence networks in two groups were constructed based on the correlations among dominant genera in Figure 3. Notably, the topological characteristics of the co-occurrence network became more complex with the LP inoculation, especially for the average weighted degree (2.996 in CKGP and 6.372 in LPGP, Table S1). Furthermore, the node number, edge number, and diameter of the network increased by 19.4%, 45.0%, and 66.7% in LPGP compared with CKGP. Interestingly, the most abundant genera in CKGP (Pediococcus, unclassified_f__Enterobacteriaceae, unclassi-fied_f__Dipodascaceae, Fusarium, Aspergillus) had few correlations with others, especially negative correlations (Figure 3a). However, concerning Weissella and Penicillium, whose average RA (10.63% and 2.6%, respectively) was relatively lower in CKGP, there existed obvious negative correlations (degree 11 and 5, respectively) with other taxa, such as Pseudomonas. Interestingly, with LP inoculation, Lactiplantibacillus and Lacticaseibacillus turned out to be the most abundant genus (total RA 83.74% from days 3 to 30), which were blinding as keystone taxa with the most negative correlations (degree 12). Furthermore, bacterial genera presented more positive correlations with each other in LPGP compared with CKGP and declined during anaerobic fermentation (Figure 3).

Redundancy Analysis for Correlations of Dominant Microbes and Physicochemical Properties of Silage
The correlation between dominant taxa and physicochemical properties of silage in CKGP and LPGP was evaluated via redundancy analysis (RDA, Figure 4). The first two axes of RDA accounted for 82.20% and 76.46% of the variance between dominant genera and physicochemical properties in CKGP and LPGP, respectively. WSC (71.2%) and AA (5.0%) were identified to be the key physicochemical properties explaining the succession of dominant microorganisms in CKGP (p < 0.05, Figure 4a). In CKGP, WSC was negatively correlated with RA of dominant genera (Enterobacter, unclassified_f__Enterobacteriaceae, Weissella, Pediococcus, Lactiplantibacillus, Trichosporon, Fusarium, and unclassified_f__Dipodascaceae) and contents of major microbial metabolites (LA, AA, BA, ethanol, and ammonia), and positively correlated with DM content, pH, and RA of Aspergillus. In LPGP, WSC (52.8%) and LA (16.2%) were the key physicochemical properties explaining the microflora succession of silage (p < 0.05). WSC showed strong negative correlations with RA of keystone taxa (Lactiplantibacillus and Lacticaseibacillus), and contents of microbial metabolites, and was positively correlated with pH, and RA of Aspergillus, Enterobacter, Bacillus.

Redundancy Analysis for Correlations of Dominant Microbes and Physicochemical Properties of Silage
The correlation between dominant taxa and physicochemical properties of silage in CKGP and LPGP was evaluated via redundancy analysis (RDA, Figure 4). The first two axes of RDA accounted for 82.20% and 76.46% of the variance between dominant genera and physicochemical properties in CKGP and LPGP, respectively. WSC (71.2%) and AA (5.0%) were identified to be the key physicochemical properties explaining the succession of dominant microorganisms in CKGP (p < 0.05, Figure 4a). In CKGP, WSC was negatively correlated with RA of dominant genera (Enterobacter, unclassi-fied_f__Enterobacteriaceae, Weissella, Pediococcus, Lactiplantibacillus, Trichosporon, Fusarium, and unclassified_f__Dipodascaceae) and contents of major microbial metabolites (LA, AA, BA, ethanol, and ammonia), and positively correlated with DM content, pH, and RA of Aspergillus. In LPGP, WSC (52.8%) and LA (16.2%) were the key physicochemical properties explaining the microflora succession of silage (p < 0.05). WSC showed strong negative correlations with RA of keystone taxa (Lactiplantibacillus and Lacticaseibacillus), and contents of microbial metabolites, and was positively correlated with pH, and RA of Aspergillus, Enterobacter, Bacillus.

PLS-PM Analysis to the Effect of Investigated LP Inoculation
To better understand the augmentation effects from LP inoculation, we further explored the intricate relationships among functional microflora, inhibited/undesirable microflora, fermentation substrates, and microbial metabolites using PLS-PM. Both the direct and indirect effects among different latent variables were evaluated ( Figure 5). The goodness of fit over 0.60 indicated good predictive power of these two models, and similar direct effects were found in both models (p < 0.05). The functional microflora was negatively related with fermentation substrates (coefficient = −0.817 in CKGP and −0.894 in LPGP, respectively) and was positively correlated with microbial metabolites (coefficient = 0.821 in CKGP and 0.814 in LPGP, respectively), and microbial metabolites were negatively related with inhibited microflora (coefficient = −0.332 in CKGP and −0.279 in LPGP, respectively).  In CKGP, functional microflora was observed to be indirectly correlated with undesirable microflora by fermentation substrates (coefficient = −0.455, p = 0.002) and microbial metabolites (coefficient = −0.273, p = 0.033, Figure 5a). However, the direct correlation between functional microflora and inhibited microflora was not significant (coefficient = −0.884, p = 0.387). On the contrary, in LPGP, keystone taxa (Lactiplantibacillus and Lacticaseibacillus) became the main functional microorganism (Figure 1a) and negatively correlated with inhibited microflora in a strongly direct way (coefficient = 0.762, p < 0.001; Figure 5b). However, the indirect correlations with inhibited microflora by fermentation substrate (coefficient = 0.012, p = 0.923) and microbial metabolites (coefficient = −0.227, p = 0.063, Figure 5b) were not significant.

Discussion
Vegetable waste, such as cauliflower and cabbage leaf byproducts, are ubiquitous in landfills but rarely reported to be bio-transformed as animal feed, such as silage [2]. LP could be an efficient microbial inoculation to enhance the fermentation quality of ensiling different materials [2,9,11,25,26] but the underlying effect and intricate relationships have been little investigated. This present study disposed of cabbage byproducts as added-value silage and was the first to focus on the underlying effect of LP improving silage quality and microbial community functional diversity in food microbial areas.

The Effects of LP Inoculation on Physicochemical Properties of Silage
The minimum DM (>200 g·kg −1 FM) and WSC content (>50 g·kg −1 DM) were reported necessary as LABs transformed the WSC to organic acid, e.g., LA, to decrease the pH and preserve the forage successfully [9,27]. Thus, the formula of initial materials was suitable in this study. During the ensiling process, the major functional microorganisms such as LABs and yeasts would degrade organic matters, especially those readily degradable ones (e.g., WSC), to organic acids and ethanol [2,6]. Hence, the recalcitrant lignocellulosic components accumulated while the WSC, DM, CF content decreased after fermentation (Tables 2 and 3). The LP inoculation significantly increased the DM content of silage samples, which indicated a better performance on conservation and valorization of organic matters ( Table 2). The higher DM content preserved in LPGP could be explained by the LP inoculation increasing the LA content, which could decrease pH and resist undesirable and fast-metabolic microbiome, such as pathogens and deterioration [6]. The LP inoculation also resulted in a higher content of LA and lower content of AA and BA in LPGP, due to the enhanced homolactic fermentation. Consequently, more organic matters would be preserved, which was consistent with increasing DM content in this study [2,9]. Flieg's score based on the organic acid ingredient was reported as the important evaluation proxy for silage quality, which could indicate the odor characteristic in a way [4,28]. In this study, the LP inoculation remarkably increased Flieg's score (Table 4), and the silage quality was evaluated as "very good". Hence, inoculation of exogenous LP could be a feasible method to transform cabbage waste and rice straw into animal feed. The inoculation significantly reduced ammonia-N content, which could be explained by the fact that more LABs and LA could effectively restrain undesirable bacteria such as Enterobacteria from degrading protein and oligopeptide to ammonia [6,9]. Moreover, ammonia-N below 7% and beyond 10% indicated a successful silage fermentation and severe nutritional loss, respectively [29]. The LP inoculation effectively inhibited the nutritional loss during the silage.
Moreover, the LP inoculation did not facilitate the accumulation of CF and CP, while it basically maintained the value of these two important nutrition indicators ( Table 3). The ADF and NDF content declined during the silage; in particular, the ADF content significantly declined at the end of co-ensiling (p < 0.05), which was different from Mu et al.'s report that LP played no significant role in ADF and NDF content in amaranth and rice straw silage [9]. This different result could account for more organic acid to accelerate the lignocellulose hydrolysis and the different raw materials in the silage study [27].

Dynamic Variations of Microbial Composition and Functional Diversity
The specific rDNA sequencing method was widely applied to investigate microbial community succession in silage [25]. The coverage index of all samples was almost 1.0 indicating the DNA sequencing results were representative of the microbial communities (Table 5). According to the declining Shannon and increasing Simpson indexes of bacterial community in LPGP, LP inoculation significantly decreased the diversity of bacterial communities during the fermentation (p < 0.05), while it played an insignificant role in bacterial community richness and fungal community diversity and richness. These results could be explained by the fact that exogenous LP inoculation resisted the undesirable bacteria and interfered with bacterial community diversity, which was consistent with what Keshri et al. reported in wheat silage [25]. As Figure 1a shows, undesirable microorganisms were enriched in CKGP, such as Enterobacteriaceae (unclassified_f__Enterobacteriaceae, Enterobacter) and Pantoea. Both of them were reported for their proteolytic activity, which could lead to inefficient LA productivity, higher pH, and higher ammonia content [6,30]. Generally, inadequate epiphytic LABs (<10 5 CFU·g −1 FM) and abundant aerobic bacteria (>10 6 CFU·g −1 FM) could result in low-quality silage due to the inefficient lactic fermentation and aerobic deterioration during ensiling [2,6], which was consistent with our results in CKGP. With the LP inoculation, Lactiplantibacillus and Lacticaseibacillus became the most abundant genus and LABs (total RA 83.74% from days 3 to 30), which was also frequently found and provided a stable fermentative environment in other silage [6]. Interestingly, the main LABs shifted from LP during the first 15 days to Lcb. rhamnosus during the last 15 days ( Figure S2a). Therefore, LP and Lcb. rhamnosus could play an important role in the first and late half phase, respectively, during co-ensiling of cabbage byproduct and rice straw. Furthermore, LP and Lcb. rhamnosus were facultative anaerobic and fastidious anaerobic LABs, respectively, so the available oxygen content in the first 15 days could be more suitable for facultative anaerobic LP. Lcb. rhamnosus was enriched more in the last 15 days, which might result from less O 2 and more CO 2 [31] . To consider the blinding effect of these two taxa, we assumed that Lactiplantibacillus and Lacticaseibacillus were the keystone taxa in LPGP and then focused on the effect of keystone taxa on improving silage quality. Pseudomonas was more abundant in LPGP and was also regarded as undesirable bacteria due to its ability to produce biogenic amines [10]. However, its RA was relatively low in both groups. Pediococcus was the most dominant LAB in CKGP. Comparatively, Lactiplantibacillus and Lacticaseibacillus became the most dominant taxa with the LP inoculation. Moreover, Lactiplantibacillus and Lacticaseibacillus were reported as more tolerant of low pH and more effective homofermentative LABs than Pediococcus during co-ensiling of amaranth and rice straw [9], which was consistent with our results. Comparatively, the exogenous LP inoculation could significantly improve the RA of homofermentative and efficient LABs in silage while decreasing the undesirable bacteria (p < 0.05).
Based on previous reports, Trichosporon and Fusarium are pathogenic fungi of humans and crops and should be prevented in final silage products [32,33]. However, they were enriched dominantly in CKGP. LP inoculation remarkably enriched more Wallemia, which were reported as saprotroph fungi and resulted in lower total RA of undesirable fungi in this study, such as Aspergillus, Trichosporon, and Fusarium [10,34,35]. Therefore, LP inoculation could inhibit these mold and pathogenic fungi, which could reduce the accumulation of mycotoxins and improve the safety quality of silage [36]. However, limited researchers noticed the role of LP inoculation in the fungal pathogens in silage production.
In the present study, the epiphytic microbiome including obligately aerobic, sporeforming, and potentially pathogenic bacteria, and a few animal-pathotroph fungi were rapidly enriched in CKGP during the ensiling process. As previously reported, aerobic microbes could result in aerobic deterioration and poor nutrition preservation. They hardly survived in an anaerobic environment, and usually showed a declining RA along the ensiling process [6,17]. The spore-forming Bacillus in this study was facultative anaerobic and could tolerate low pH and the anaerobic environment to some extent. Their spores could exist in silage and ruminant intestinal tract, which were closely associated with aerobic silage deterioration and spore contamination in livestock products (e.g., milk) [6]. Furthermore, epiphytic animal-pathotroph fungi, such as Trichosporon and Fusarium, and Gram-negative Enterobacter, unclassified_f__Enterobacteriaceae in CKGP, usually produced multiple mycotoxins and endotoxin when present in animal feeds; these toxins could result in poor-quality dairy cow performance and endanger both animal and human health [6,7]. Therefore, the bacterial species characterized as aerobic/spore-forming/potentially pathogenic and animal-pathotroph fungi are detrimental in silage production. With the LP inoculation, these aforementioned undesirable microorganisms were significantly resisted. However, there are few studies that systematically focus on microbial functional diversity related to silage production.

Correlations of Dominant Microbes and Physicochemical Properties of Silage
As Banerjee et al. reported, negative correlations of co-occurrence networks indicated possible competition for resources and growth inhibition, while positive indicated common predators within microbial taxa [14]. Interestingly, the most abundant genera in CKGP (Pediococcus, unclassified_f__Enterobacteriaceae, Fusarium, Aspergillus) had few correlations with others, especially negative correlations (Figure 3a). However, Weissella and Penicillium whose average RA (10.63% and 2.6%, respectively) was relatively lower were identified as the keystone taxa in CKGP due to their obvious negative correlations (degree 11 and 5, respectively) with inhibited taxa, such as Pseudomonas. Interestingly, with LP inoculation, keystone taxa (Lactiplantibacillus and Lacticaseibacillus) turned out to be the most abundant genus (RA 67.8%). Based on that, it was plausible that the keystone taxa in CKGP with low RA were not sufficient to restrain the undesirable microbe. On the contrary, LP inoculation significantly increased the RA of keystone taxa (Lactiplantibacillus and Lacticaseibacillus) and improved the overall functionality of the microbial community in LPGP.
Correlations among dominant taxa and physicochemical indexes in CKGP and LPGP were similar (Figure 4). DM, pH, and WSC content were negatively related to organic acids, ethanol, and ammonia content as well as some dominant taxa. These correlations indicated the aforementioned dominant genera could metabolize WSC and other organic components to organic acids, ammonia, and ethanol, therefore consuming DM, reducing pH, and inhibiting undesirable taxa, such as Aspergillus in CKGP and Aspergillus, Bacillus, and Enterobacter in LPGP. However, dominant taxa and the correlations among dominant taxa changed obviously with the inoculation of LP (Figure 4b). These results could be explained by the fact that exogenous LP inoculation increased the RA of keystone taxa (Lactiplantibacillus and Lacticaseibacillus), which could provide efficient homofermentation and resist undesired taxa (Aspergillus, Enterobacter, Bacillus). These results were consistent with previous studies that LP inoculation increased the RA of Lactiplantibacillus in silage, improved silage microbial community and fermentation metabolites [2,9].

Intricate Relationships of LP Inoculation Augmentation Effects
With LP inoculation, more keystone taxa (Lactiplantibacillus and Lacticaseibacillus) were enriched, which effectively inhibited undesirable microflora and improved the fermentation quality of silage. As PLS-PM analysis revealed ( Figure 5), the similar relationships between CKGP and LPGP indicated that functional microflora could utilize fermentation substrate and facilitate the production of organic acids, ammonia, and ethanol. Meanwhile, these metabolites could resist some undesirable microflora, such as Aspergillus and Bacillus. These results were consistent with the discussion about correlations between dominant taxa and physicochemical indexes in Section 4.3.
Notably, functional microflora in CKGP indirectly resisted undesirable microflora by fermentation substrate competition and microbial metabolite inhibition rather than direct interactions between two types of microorganisms. Comparatively, functional microflora (keystone taxa: Lactiplantibacillus and Lacticaseibacillus) in silage inoculated with LP primarily worked through directly inhibiting undesirable microflora, rather than indirect effects including competition for fermentation substrate and inhibition by producing mi-crobial metabolites. These hypotheses were different from the conventional perspective that LABs transform WSC into organic acids to reduce pH and inhibit the undesirable microorganisms [2,6,9], and are reasonably based on the current analysis results.

Conclusions
In co-ensiling of cabbage byproduct and rice straw, the inoculation of exogenous LP significantly improved the ensiling quality indicated by the elevated contents of DM and LA, and reduced the contents of ammonia, AA, BA, ADF, and the pH value, which was closely associated with the effective homolactic fermentation system. Specifically, LP inoculation significantly influenced the structure of the microbial community, improved the proportion of functional types needed by successful ensiling, and resisted the undesired microorganisms, especially aerobic, spore-forming, and pathogenic taxa. With LP inoculation, Lactiplantibacillus and Lacticaseibacillus were observed as the keystone taxa and major functional species, which directly inhibited undesirable microbes and improved the fermentation characteristics.
Supplementary Materials: The following are available online at https://www.mdpi.com/article/10 .3390/microorganisms9051099/s1, Figure S1: Effect of Lactiplantibacillus plantarum additive on the microbial community dynamics at the phylum level of cabbage and rice straw silage. Figure S2: Effect of Lactiplantibacillus plantarum additive on the microbial community dynamics at the species level of cabbage and rice straw silage. Table S1: Topological characteristics of co-occurence network., Spreadsheet S1: The bacterial phenotypes annotation using DacDive database and Bergey's Manual of Systematic Bacteriology., Spreadsheet S2: The fungal trophic modes and ecological guilds annotation using FUNGuild.

Data Availability Statement:
The datasets generated for this study can be found in Sequence Read Archive under BioProject, PRJNA686717 (https://www.ncbi.nlm.nih.gov/sra/, accessed on 20 December 2020).