Introduction

Forage production is seasonal in many parts of the world. To make the greens available throughout the year, ensiling is a common way to preserve the green fodder. Ensiling is an anaerobic bacterial-based fermentation process. During fermentation, sugars are converted by epiphytic lactic acid bacteria (LAB) mainly to lactic acid and acetic acid. This lowers the pH and creates an environment where the resulting silage is preserved. However, biochemical, and microbiological incidents can arise at the different stages of ensiling which result in high variability in fermentation quality. Poorly fermented silage would induce economic losses, affect animal performance, and even threaten animal and human health [1].

Silage fermentation process involves a variety of bacterial communities and biochemical reactions. Existing studies on deciphering silage fermentation mainly depends on the dynamics of bacterial community [2, 3]. Nevertheless, what ecological mechanisms govern the formation and development of bacterial community structures in silage is poorly known. From the ecological perspectives, microorganisms establish communities according to deterministic or stochastic processes. Deterministic theories suggest that local, niche-based processes, such as environmental filtering, biotic interactions and interspecific trade-offs largely determine the patterns of species diversity and composition [28]. Epiphytic bacterial richness displayed no statistical difference during the harvest period, suggesting that epiphytic microbiota is highly conserved during plant growth. Bacterial diversity decreased with the delay of harvest, presumably associating with the decreasing leaf ratio, since leaf contributes most to the diversity of epiphytic microbiota [29]. Epiphytic microbiota structures underwent dynamic change during the harvest period. This could be due to the altered plant status including hormonal and physiological changes as matured [30]. Three epiphytic genera (Enterobacter, Erwinia, Curtobacterium) exhibited increasing relative abundances with the delay of harvest. Most members of these bacteria are pathogenic to plants [31]. The increases in their proportions probably reflected a decrease in plant resistance.

Generally, diverse bacterial communities are formed in field and LAB development during silage fermentation will simplify bacterial community resulting in a decline in alpha-diversity [13]. Greater silage bacterial diversities with the delay of harvest suggested a decreased effect of LAB on dominating silage microbiota. The bacterial community shifted from Proteobacteria to Firmicutes after fermentation. This was closely associated with the combined stress of low pH and anaerobiosis during fermentation [32]. Lactobacillus, Pediococcus, unclassified_f_Enterobacteriaceae, and Hafnia–Obesumbacterium constituted the predominated genera in the silages. Lactobacillus and Pediococcus are desirable bacteria in silage fermentation. Their flourishment has been shown to promote the establishment of acidic environment and the suppression of undesirable bacteria [33]. In contrast, unclassified_f_Enterobacteriaceae and Hafnia–Obesumbacterium are the undesirable bacteria competing with LAB for limited WSC contents [11]. Silages dominated by these bacteria often exhibit high pH and extensive protein degradation [11, 34]. This explained their positive relationships with pH and NH3-N contents (Fig. S2B). Delaying harvest increased the proportions of Hafnia–Obesumbacterium, explaining why the silage fermentation quality decreased with the delay of harvest.

Bacterial assembly processes and driving factors

In microbial ecology, it is widely accepted that microbiota assembly patterns in different habitats can be explained by deterministic and stochastic processes, based on niche and neutral theories, respectively [35]. Null model analysis provides a way to explore whether communities are randomly assembled or non-randomly aggregated or segregated, and to identify the underlying mechanisms for microbial assembly [10, 17, 36]. In this study, null model analysis was for the first time applied to silage ecosystem to reveal the ecological processes controlling bacterial assembly. The results showed that the assembly of rare subcommunity was primarily controlled by stochastic processes. This is likely due to the small population sizes of the rare species, which make them easily to be impacted by demographic stochasticity [37]. In contrast, abundant species often occupy core niche positions and therefore they are strongly impacted by deterministic filtering [38]. Our results showed that stochastic processes also played important roles in the succession of abundant subcommunity in the silages harvested in southern China. The stochastic processes consider that random changes shape microbial communities and that their fluctuations are random, including unpredictable interference, random birth and death, and dispersal probability [39]. This was corroborated by the low explanatory power of measured variables in RDA analysis (Fig. S2A). It is worth noting that, deterministic processes, more specifically, heterogeneous selection, were more prominent in abundant bacteria with the delay of harvest. Generally, heterogeneous selection is determined by dynamic selection under biotic or abiotic conditions and can lead to large changes in microbial community [40]. This implies that the characteristic changes during forage maturity influenced the rules governing the assembly of bacteria during silage fermentation.

To evaluate the potential drivers of producing trends in phylogenetic assembly of abundant bacteria in the oat silages, the βNTI values were correlated with the changes in measured variables during silage fermentation. The results revealed the importance of DM and WSC contents as critical factors that impact the balance between stochastic and deterministic processes in the assembly of abundant bacteria. The DM content associates with the moisture in silage, which can structure microbial communities through many indirect pathways. For example, changes in solute diffusion and water potential due to varying moisture levels contribute to distinct variations in microbial community [41]. The bacterial competition for resources is regulated by the dissolved nutrient level [42]. Therefore, it is not surprising that WSC content can shape the turnover of bacterial community during fermentation. In addition to DM and WSC contents, the change in pH value was also an important driver of the bacterial assembly. This could be attributed to the neutral nature of endocellular pH of most microorganisms. However, unlike our expectation, the βNTI values were negatively correlated with pH values, indicating that greater pH changes promoted stochastic assembly during silage fermentation. The disagreement between our expectation and the observation was probably explained by the fact that, besides direct effects, pH may indirectly affect the bacterial community by altering the solubility of elements (e.g., phosphorus, aluminum, and iron) [42]. Bacterial species may respond differently to the direct and indirect effects of silage pH decline especially under extremely wet conditions. Similarly, studies on freshwater lakes and agricultural soils also reported the increased importance of stochasticity in acidic environments [6, 43]. Our results further suggested the negative relationship between Flieg’s scores and βNTI values. It revealed that deterministic filter could increase the heterogeneity of community through the selection of undesirable species with stronger competitive abilities in silages.

Conclusions

This study quantified the bacterial assembly processes in oat silages harvested in southern China using null models. Significant differences in raw material characteristics were observed among harvest days. The fermentation qualities of silages tended to decrease with the delay of harvest. During silage fermentation, stochastic processes were the primary contributor to the assembly of rare subcommunity, while abundant subcommunity was controlled both by stochastic and deterministic processes. Delaying harvest increased the dominance of deterministic assembly of abundant subcommunity. The changes of three variables (DM, WSC and pH) have significant relationships with the assembly of abundant bacteria in oat silages harvested in southern China. Furthermore, significant negative correlation was found between Flieg’s scores and the βNTI values. This study revealed the ecological processes controlling the bacterial assembly during silage fermentation, which provides new insights into the mechanisms underlying the construction of silage bacterial community.