Networks of interactions among microbes further shape the establishment and maintenance of the microbial community (Horton et al. 2014; Layeghifard et al. 2017; Hassani et al. 2018). In those networks, highly connected genera or “hubs” play an important ecological role in the establishment of the community and the regulation of microbial assembly (Agler et al. 2016; Layeghifard et al. 2017; Hassani et al. 2018). To determine the structure of the rice leaf microbiome, we inferred microbial co-occurrence networks and identified critical hubs (Additional file 2). The criteria to select the hubs was based on a combination of the network properties such as weighted degree, betweenness centrality, modularity class, clustering, and eigen-centrality (Additional file 3: Table S5). We found that the rice microbe network can be defined by 12 hubs: Clostridium, Mycoplasma, Bacillus, Buchnera, Prochlorococcus, Helicobacter, Methylobacterium, Chamaesiphon, Azotobacter, Kineococcus, Acidovorax, and Pseudomonas (Fig. 2a, Additional file 3: Table S5). Interestingly, connectivity of a genus within the network did not correlate with their abundance. For example, the highly abundant genera Xanthomonas and Burkholderia were not identified as hubs. The genera Kineococcus or Helicobacter, with less than 1% abundance, played a role in shaping the network of interactions (Agler et al. 2016).
The hubs Methylobacterium, Pseudomonas, Bacillus, Kineococcus, Azotobacter, Acidovorax, and Clostridium have also been reported as commensal, beneficials, or pathogens in other plants (Cottyn et al. 2009; Bian et al. 2012; Horino et al. 2015; Chen et al. 2018; Lai and Huang 2018). The hubs Chamaesiphon and Prochlorococcus could be part of the rice microbiome, as the paddy field conditions like shallow water and sunlight could create an ideal environment for algae growth. The hub Buchnera, could be part of aphids and rice interaction, as Buchnera is an aphid endosymbiont. The hubs Mycoplasma and Helicobacter, associated with human diseases, have not been associated with plant microbiomes. We hypothesized that these bacteria, generally present in animals and humans, are hubs in the rice microbiome due to a strong influence of agricultural practices, like irrigation water or crop-human interaction.
The networks can be further sub-structured into modules, which represent a group of organisms that has more interactions within the group than with other members of the network (Layeghifard et al. 2017). We found seven modules. Based on the bacterial composition of some modules, microbial genetic ancestry or ecological niche seems to shape them (Fig. 2a, Additional file 3: Table S6). For example, one module was enriched with Cyanobacteria, another module had bacteria associated to animals and other modules mainly had Alphaproteobacteria or Betaproteobacteria associated with plants or soils (Additional file 3: Table S6). Overall, taxonomy status or ecological niche rather than the abundance of individual taxa appears to define the interactions within the microbial community. Moreover the number of modules in the network suggested a highly stable network since a microbial community appears to reach an equilibrium when its network of interactions had a small number of modules (Layeghifard et al. 2017).
The functional and nutritional capacities of the microbes partly define the networks of interactions among microbes (Agler et al. 2016; Layeghifard et al. 2017; Hassani et al. 2018). To understand if biological functions in the bacterial communities associate to the ecological network, we predicted functional categories for the microbial taxa (Additional file 2). We found 85 predicted KEGG functional categories, where 28 had more than 1% abundance (Fig. 2b, Additional file 3: Table S7). Transcription, translation, primary metabolism, flagellar assembly, environmental adaptation (ECM receptor interaction) and secondary metabolism (terpenoids, antibiotics, and xenobiotics) were the most abundant pathways. The enrichment of these KEGG pathways linked with the abundant genera as well as the bacterial hubs. For example, the enrichment of pathways related to human diseases reflected the presence of animal pathogens in the rice microbiome. Moreover, Pseudomonas, facultative anaerobes, methanogenic bacteria, and some gram-positive bacteria, like Bacillus and Streptomyces, have the capacity to metabolize xenobiotic, terpenoids and antibiotics compounds (Additional file 3: Table S2). The xenobiotics metabolism KEGG category suggested an adaptation of the bacterial to an environment where chemicals, like pesticides, might be applied to crop fields (Tipayno et al. 2017). The presence of functional categories common to other leaf microbiomes, restated the idea that functions and communities are not random (Delmotte et al. 2009; Xiao et al. 2017). Moreover, our results aligned with the idea that metabolic functions and interaction within the microbiome regulate the establishment of the community (Agler et al. 2016; Louca et al. 2017).
We then determined the microbial co-occurrence networks and functional categories for each agCh and agPh dataset. Interestingly, agCh and agPh assembled networks with similar average degree and modularity (Additional file 1: Figure S3A). We identified seven highly connected hubs in each dataset, five of which were common to agCh and agPh networks (Additional file 1: Figure S3A, Fig. 2). Furthermore, we found no significant differences between agCh and AgPh functional profiles, where both datasets shared 22 of 24 KEGG level pathways (Additional file 1: Figure S3B). The data suggest that rice leaf microbiome assemble communities with similar structure, independently from the available microbial diversity. Overall, the similarity of the network structures and the functional redundancy of the leaf microbiome dataset supported the idea that key microbial groups might regulate the establishment by providing essential functions in the community (Hassani et al. 2018). This result also aligned with other studies where core biological functions of the microbiome are associated to different plant tissues or plant substrates (Vorholt 2012; Agler et al. 2016; Louca et al. 2017; Xiao et al. 2017).