Microbial Network and Soil Properties Are Changed in Bacterial Wilt-Susceptible Soil

Bacterial wilt disease is caused by the pathogen Ralstonia solanacearum and is a widespread devastating soilborne disease leading to huge economic losses worldwide. The soil microbial community is crucial to the capacity of soils to suppress soilborne diseases through complex interactions. Network analysis can effectively explore these complex interactions. In this study, we used a random matrix theory (RMT)-based network approach to investigate the changes in microbial network and associated microbial interactions in a bacterial wilt-susceptible soil (BWS) in comparison to a healthy soil (HS). We found that the structure and composition of the microbial network in BWSs were quite different from those of the HS. The BWS network had fewer modules, edges, and key microorganisms and lower connectivity than the HS network. In the BWSs, apparently the topological role of microbes was changed and key microorganisms were shifted to specialists.

ABSTRACT

Bacterial wilt disease is a devastating disease of crops, which leads to huge economic loss worldwide. It is hypothesized that the occurrence of bacterial wilt may be related to changes in soil chemical properties and microbial interactions. In this study, we compared the soil chemical properties and microbial network structures of a healthy soil (HS) and a bacterial wilt-susceptible soil (BWS). The contents of available nitrogen, potassium, and phosphorus and the soil pH in the BWS were significantly lower than those in the HS. BWS showed nutrient deficiency and acidification in comparison with the HS. The structure and composition of the BWS network were quite different from those of the HS network. The BWS network had fewer modules and edges and lower connectivity than the HS network. The HS network contained more interacting species, more key microorganisms, and better high-order organization and thus was more complex and stable than the BWS network. Most nodes and module memberships were unshared by the two networks, while the ones that were shared showed different topological roles. Some generalists in the HS network became specialists in the BWS network, indicating that the topological roles of microbes were changed and key microorganisms were shifted in the BWS. In summary, the composition and structure of the microbial network of the BWS were different from that of the HS. Many microbial network connections were missing in the BWS, which most likely provided conditions leading to higher rates of bacterial wilt disease.

IMPORTANCE Bacterial wilt disease is caused by the pathogen Ralstonia solanacearum and is a widespread devastating soilborne disease leading to huge economic losses worldwide. The soil microbial community is crucial to the capacity of soils to suppress soilborne diseases through complex interactions. Network analysis can effectively explore these complex interactions. In this study, we used a random matrix theory (RMT)-based network approach to investigate the changes in microbial network and associated microbial interactions in a bacterial wilt-susceptible soil (BWS) in comparison to a healthy soil (HS). We found that the structure and composition of the microbial network in BWSs were quite different from those of the HS. The BWS network had fewer modules, edges, and key microorganisms and lower connectivity than the HS network. In the BWSs, apparently the topological role of microbes was changed and key microorganisms were shifted to specialists.

INTRODUCTION

In China, long-term continuous monoculture has led to the occurrence of severe soilborne disease in many croplands due to the accumulation of pathogens and plant autotoxins in soil (1–3). Continuous cropping has led to an outbreak of bacterial wilt disease in a large area of China (4). Bacterial wilt disease, caused by the pathogen Ralstonia solanacearum, is a widespread devastating soilborne disease characterized by a rapid and fatal wilting symptom in the host plant (5). R. solanacearum can infect more than 400 plant species, including many Solanaceae plants (eggplant, pepper, tomato, tobacco, etc.) (6), leading to economic losses of up to 100% of current yields in banana, 90% in tomato and potato, 30% in tobacco, and 20% in groundnut. Bacterial wilt disease is difficult to control, and infected fields can rarely be reused (7). Biological control of bacterial wilt is a good alternative to chemical control. At present, most of the patented biological control agents are made of beneficial bacteria. Specific antagonistic microorganisms (e.g., Pseudomonas spp., Bacillus spp., Streptomyces spp., Acinetobacter spp., Burkholderia spp. and Paenibacillus spp.) can be used for biological control of bacterial wilt disease. These act through the production of antibacterial metabolites, competition, and the induction of host systemic resistance against the pathogens. These specific beneficial microorganisms have been isolated mostly from soils or the rhizosphere of plants (8, 9). As previously reported, soil microorganisms play important roles in the suppression of soilborne disease (10).

It is known that the naturally existing population of microorganisms in soil can significantly reduce the severity of bacterial wilt in plants (11). Soil and plant health depend largely on the composition and diversity of the soil microbial community (12). The diversity of soil microbial communities is crucial to the capacity of soils to suppress soilborne diseases (13, 14). For example, the abundances of the plant-beneficial bacteria Arthrobacter and Lysobacter have significant negative correlations with bacterial wilt disease (11, 15). By modifying the microbial community, organic amendments to soil can increase the activities of soil microorganisms that are antagonistic to pathogens (16). Diverse bacterial communities containing high abundances of beneficial microorganisms can more effectively suppress plant pathogens. Several beneficial bacteria are known to be antagonistic to R. solanacearum (17). Previously, we also found that the composition and structure of the soil microbial community are altered in soils infected with bacterial wilt disease, in comparison with healthy soils (18, 19).

While the composition of soils changes in response to bacterial wilt disease, we do not know how the microbial network of bacterial wilt-susceptible soil (BWS) changes in comparison to that of healthy soil (HS). Different microbial species in soil microbial communities have complex interactions (such as symbioses, parasitism, competition, or predation), forming a complex interrelated ecological network (20). Network analysis can offer new insights into these microbial interactions as well as the identities of keystone microorganisms, along with their structure of a complex microbial community. The topological properties of the microbial network (modularity and complexity) are good indicators of the stability and diversity of a microbial community (21). Network analysis has been used to explore complex interactions between microorganisms in many ecosystems, including soils. Recently, the phylogenetic molecular ecological networks (pMENs) based on the random matrix theory (RMT) approach have been developed to study microbial networks. The pMENs provide us with a powerful tool to study microbial interactions and their responses to environmental changes (22–25). For example, it has been found that the phylum Actinobacteria contains key microorganisms that lend stability to microbial networks and significantly impacts a variety of soil properties (the percentages of C or N, the proportion of soil moistures, and soil pH) (23).

Long-term continuous monoculture of tobacco in the last 20 years has caused severe tobacco bacterial wilt disease in many areas of Enshi State, Hubei Province, China. In this study, we used high-throughput sequencing technology in combination with an RMT-based network approach to investigate the changes in microbial interactions in response to tobacco bacterial wilt disease and to further elucidate the relationships between microbial connectivity and specific soil properties. Two groups of soil samples were collected from tobacco fields located in Enshi State. One was a HS with a very low incidence of bacterial wilt disease (<15%). Another was a BWS with a high occurrence of bacterial wilt disease leading to high rate of mortality (86.66%). The cooccurrence or coexclusion patterns of microbial species in HS and BWS microbial communities were investigated and compared by network analysis. We hypothesized that the structure and composition of BWS networks is fundamentally different from those of the HS network. The HS network was more integrated, stable, and complex and had more interactions, therefore making it better able to exclude the pathogen R. solanacearum from the plant environment.

RESULTS

Healthy soil suppresses tobacco bacterial wilt disease.

Disease incidence (percentage of plants with wilting symptoms) and disease index of tobacco bacterial wilt disease in the BWS were significantly (P < 0.01) higher than those in HSs (Table 1). The disease index (extent of disease per plant) of the BWS was 74.88, suggesting that severe bacterial wilt disease occurred in the BWS. The disease index of the HS was much lower (0.12), suggesting that HS well suppressed bacterial wilt disease.

TABLE 1.

Tobacco properties and soil chemical properties

Properties	Value fora:
	HSs (n = 17)	BWSs (n = 20)
Disease incidence (%)	0.12 ± 0.32 Bb	86.66 ± 22.75 Aa
Disease index	0.12 ± 0.33 Bb	74.88 ± 18.95 Aa
SOM (g/kg)	27.96 ± 6.48 a	26.69 ± 5.74 a
AP (mg/kg)	34.64 ± 13.39 a	23.43 ± 12.40 b
AN (mg/kg)	176.56 ± 23.56 a	147.76 ± 23.90 b
AK (mg/kg)	315.87 ± 90.95 a	246.02 ± 53.20 b
pH	6.47 ± 0.79 a	5.98 ± 0.37 b
Soil moisture content (%)	20.15 ± 3.42 a	21.53 ± 1.74 a
Bulk density (g/cm3)	1.16 ± 0.12 a	1.23 ± 0.14 a
Total porosity (%)	55.79 ± 4.12 a	53.51 ± 4.47 a
Soil water-stable aggregates (mm)	1.87 ± 0.63 a	1.83 ± 0.51 a

^aValues are means ± standard error. Different letters within the same row indicate a very significant difference (P < 0.01; uppercase) or a significant difference (P < 0.05; lowercase) between HSs and BWSs.

Chemical properties of HS are different from BWS.

No significant difference was observed for soil organic matter (SOM) content, soil moisture content, bulk density, total porosity, and soil water-stable aggregates of HSs and BWSs. The contents of alkali-hydrolyzable nitrogen (AN), available phosphate (AP), and available potassium (AK) of the BWS were all significantly (P < 0.05) lower than those of the HS (Table 1). The soil pH of BWS decreased 0.49 ± 0.37 U in comparison to that of HS and was significantly (P < 0.05) lower than that of HS.

Composition of soil microbiome.

We rarefied the data set by randomly selecting reads from a sample to count the number of operational taxonomic units (OTUs) represented using mothur software to an even sequence depth of 20,000 reads per sample. The total number of OTUs binning at 97% similarity found in all the samples was 4,461. Rarefaction curves showed that the sequence depth was sufficient to access the most abundant taxa in each soil (see Fig. S1 in the supplemental material). Thus, we focused on interaction networks of the more abundant taxa. The analysis of the estimated richness indices (e.g., observed species, Chao1, and abundance-based coverage [ACE]) revealed that the HSs harbored higher bacterial richness. The Shannon diversity index indicated that the bacterial diversity of the HS was significantly higher (P < 0.05) than that of the BWS (Table S1).

The plant-beneficial bacterial species (e.g., Actinoallomurus, Actinocorallia, Actinospica, Bacillus, Blastococcus, Burkholderia, Conexibacter, Lysinibacillus, Lysobacter, Rhodanobacter, Spirillospora) known to inhibit the growth of pathogens are more abundant in the HSs than in the BWSs (Fig. S2). But the abundance of the pathogen Ralstonia in BWSs was higher than in HSs. Network analysis found that these beneficial bacterial species have more negative interactions in the HSs (26) than in the BWSs (5).

The abundances of 23 bacterial genera (Acetivibrio, Acidisoma, Anaerolinea, Azohydromonas, Bacteroides, Chondromyces, Clostridium, Cryptanaerobacter, Denitratisoma, Hyalangium, Methylocystis, Nakamurella, Neochlamydia, Noviherbaspirillum, Oryzihumus, Parachlamydia, Pelomonas, Rhizobacter, Sideroxydans, Sinosporangium, Streptococcus, Terrabacter, and Variovorax) were significantly different among BWSs and HSs (Fig. S2).

Network structures and compositions are different in HS and BWS networks.

We built the networks through the Molecular Ecological Network Analysis (MENA) pipeline and used the OTU abundance expressed as standardized relative abundances (Data Set S1) in the pipeline to construct networks. The HS network contained more nodes and edges (266 nodes, 514 edges) than the BWS network (204 nodes, 180 edges). There are more negative interactions in HSs (225) than in BWSs (27). Nodes with close interactions formed a module. A total 13 modules (with ≥5 nodes) were obtained in the HS network, while the BWS network had 9 modules (with ≥5 nodes) (Fig. 1). The two largest modules were both found in the HS network (H3, 50 nodes; H1, 44 nodes), and the HS network typically had larger and more modules, nodes, and edges than the BWS network. Network complexities, as measured by the node number and the average connectivity of the nodes (22), were considerably different between HSs and BWSs. The HS network was more complex than the BWS network. It was speculated that nutrient (phosphorus, nitrogen, potassium) deficiency increased bacterial wilt occurrence and reduced the soil microbial interactions.

FIG 1.

Overview of microbial networks. Graphs of the HS network (A) and the BWS network (C) are shown. Nodes of different colors belong to different bacterial phyla. Blue edges represent negative interactions between nodes. Red edges represent positive interactions. Modules with five or more nodes are included. Module sizes in the HS network (B) and the BWS network (D) are specified individually.

Connectivity, the number of links between a node and other nodes, provides information on how strongly an OTU is connected to other OTUs (23). The connectivity distribution, correlated with soil variables, gradually followed a power law, with R² values being 0.91 and 0.94 for HS and BWS networks, respectively, indicating the scale-free property of two networks (Table 2). Various network indexes, including average connectivity, average clustering coefficient, average geodesic distance, and modularity, were used to describe the topology properties and structure of networks. Connectivity is the number of links (edges) of a node with other nodes. Geodesic distance is the shortest path between two nodes. The clustering coefficient shows how well a node is connected with the neighbor nodes. The average connectivities were 3.87 and 1.77 for HS and BWS networks, respectively. A high average connectivity means a more complex network, suggesting that the HS network is more complex than the BWS network. The average geodesic distances were 4.15 and 4.07 for HS and BWS networks, respectively, indicating the small-world behavior of the two networks (24). The average clustering coefficient of the HS network (0.19) was higher than that of the BWS network (0.09), suggesting that the nodes of the HS network were clustered closer than those of the BWS network. The average clustering coefficients of empirical networks (∼0.09 to 0.19) were higher than those of the corresponding random networks (∼0.01 to 0.04), which also suggested the small-world property of the two networks. In the random network, OTUs associate randomly, resulting in a network with little clustering. The random networks were generated using the Maslov-Sneppen procedure by rewiring all of the links of a pMEN with the same numbers of nodes and links to the corresponding empirical pMEN (23).

TABLE 2.

Topological properties of the empirical pMENs of microbial communities in HS and BWS and their associated random pMENs^a

Condition	Empirical networks						Random networks
	Stb	Network size (no. of nodes)	Avg connectivity	Avg geodesic distance	Avg clustering coefficient	Modularity (no. of modules)	Avg geodesic distance ± SD	Avg clustering coefficient ± SD	Avg modularity ± SD
HS	0.83	266	3.87	4.15	0.19	0.57 (40)	3.74 ± 0.07	0.04 ± 0.01	0.49 ± 0.01
BWS	0.81	204	1.77	4.07	0.09	0.87 (57)	5.77 ± 0.48	0.01 ± 0.00	0.84 ± 0.01

^aThe random networks were generated by rewiring all of the links of a pMEN with the identical numbers of nodes and links to the corresponding empirical pMEN.

^bS_t, similarity threshold.

Modularity measures the degree to which a network is organized into clearly delimited modules. The modularity (M) values of the HS (0.57) and BWS (0.87) networks were higher than the threshold value (0.4) for a modular structure and the corresponding randomized networks (∼0.49 to 0.84), suggesting that the two networks are modular (24, 28). These results suggested the scale-free, small-world, and modularity properties of the HS and BWS networks. The clustering coefficients, connectivities, and modularities were all different between the HS and BWS networks, suggesting that the topological property of the BWS network was different from that of the HS network. The BWS network had a lower connectivity and clustering coefficient than the HS network, suggesting that the BWS network was less stable than the HS network.

The majority of the nodes in the HS and BWS networks belonged to 10 different bacterial phyla. Among them, Acidobacteria, Actinobacteria, and Proteobacteria were more dominant (Fig. 2A). The proportions of Actinobacteria, Gemmatimonadetes, Proteobacteria, and Verrucomicrobia in the HS network were higher than those in the BWS network. The proportions of Acidobacteria, Bacteroidetes, Firmicutes, and Planctomycetes in the BWS network were higher than those in the HS network. Only 90 nodes were shared by the HS and BWS networks (Fig. 2B). Most of the nodes (290 nodes) were unshared by the two networks, suggesting that most microbial species had their special niches under the HS and BWS soils. The HS network possessed more specific nodes (176 nodes) than the BWS network (114 nodes). These results suggested that the compositions of nodes of the HS network were different from those of the BWS network.

FIG 2.

Relative abundances of different nodes in HS and BWS networks. (A) Proportion of nodes belonging to different phylogenetic groups; (B) Venn diagrams indicating the number of nodes shared and not shared by the HS and BWS networks.

The topological roles of nodes in the BWS network are different from those in the HS network.

The topological roles of nodes were defined by the within-module connectivity (Z_i) and among-module connectivity (P_i) (23, 25, 29–32) (Fig. 3). The majority of nodes (97.3%) were peripherals (specialists). Among them, 80% of the peripherals had no edges at all outside their own modules (i.e., P_i = 0). The connectors had low Z_i but high P_i values and were highly connected with other modules. The module hubs had high Z_i but low P_i values and were highly connected with many nodes in their own modules. A total of 2.7% of nodes were module hubs (4 nodes) and connectors (8 nodes) (both named as generalists). No network hub (supergeneralist) was found in the two networks. More generalists (3 module hubs, 8 connectors) were found in the HS network than in the BWS network (1 module hub). The 11 generalists in the HS network belonged to Actinobacteria, Chloroflexi, Firmicutes, Gemmatimonadetes, and Proteobacteria, respectively. Two module hubs (OTU 18 and OTU 122) belonging to Actinobacteria were closely related to Gaiella spp. and Conexibacter spp., respectively (Fig. 3). OTU 19 belonging to Gemmatimonadetes was closely related to Gemmatimonas spp. Four connectors (OTU 73, OTU 568, OTU 1023, and OTU 1613) belonging to Proteobacteria were closely related to Kofleria spp., Syntrophobacteraceae, Stella spp., and Bauldia spp., respectively. Three connectors (OTU 8, OTU 14, and OTU 123) belonged to Firmicutes. OTU 123 was closely related to Bacillus spp. Another connector (OTU 252) belonged to Chloroflexi. The connectors and module hubs are key microorganisms and maintain network stability and play important roles. These generalists were key microorganisms in the HS network. Only one module hub (OTU 375) was found in the BWS network. OTU 375 belonging to Acidobacteria was closely related to Acidobacteria_Gp6.

FIG 3.

Topological roles of nodes. Threshold values of Z_i and P_i for categorizing nodes are 2.5 and 0.62, respectively. Generalists are labeled with OTU number, module number, and phylum. Six OTUs identified as generalists in the HS network but as specialists in the BWS network are indicated in green.

It was worth noticing that some nodes (OTU 8, OTU 14, OTU 19, OTU 73, OTU 122, and OTU 123) were identified as generalists in the HS network but as specialists in the BWS network. This suggested that the same bacterial species play different roles in these two networks. In the BWS, the topological roles of bacteria were changed and the key microorganisms were shifted, in comparison to the HS. This might make the BWS network become unstable and key microorganisms possibly lose their activities such as suppressing disease in BWS.

Analysis of module H4 of the HS network.

As an example of network analysis, module H4 of the HS network is illustrated in Fig. 4. Module H4 contained 18 OTUs. Among them, OTU 31 and OTU 39 were the most dominant (Fig. 4A). The relative abundances of 18 OTUs were the highest in the fourth HS sample, followed by the tenth and eleventh HS samples (Fig. 4B). The key OTUs in module H4 were identified as OTU 90, OTU 218, OTU 222, OTU 252, and OTU 548. Their abundances changed following the same trends among all HS samples (Fig. 4C). Close interactions were found among different OTUs. For example, OTU 90 had positive interactions with OTU 222, OTU 252, and OTU 324. OTU 218 had positive interactions with OTU 292 and OTU 709 but a negative interaction with OTU 59 (Fig. 4D). Positive interactions indicate the symbiosis or synergies of different microbial species. Negative interactions suggest that competition, antagonism, or parasitism existed between microbial species. In module H4, Proteobacteria had the greatest number of memberships (7 OTUs), followed by Actinobacteria (4 OTUs) (Fig. 4E). As the key OTUs, OTU 222 belonging to Actinobacteria was closely related to Ilumatobacter spp. OTU 90 and OTU 218 belonging to Proteobacteria were closely related to Rhodoplanes cryptolactis and Sphingomonas chloroacetimidivorans, respectively. OTU 252 belonging to Chloroflexi was closely related to Chloroflexus spp. OTU 548 belonged to Verrucomicrobia.

FIG 4.

Analysis of module H4. (A) Heat map of standardized relative abundances (SRA) of OTUs in different soil samples. The gradient from −5 to +5 represents the log₁₀ values of the SRA of OTUs. Red and green correspond to the OTUs with high and low relative abundances, respectively. (B) SRA of OTUs (y axis) across different soil samples (x axis). Phi represents variance explained by OTUs. (C) Module memberships that consistently coexist in these soil samples. (D) Interactions among different nodes. Blue lines represent negative interactions. Red lines represent positive interactions among different nodes. Different node colors represent different bacterial phyla, as in Fig. 1. (E) Phylogenetic relationship of different nodes in module H4, which is constructed by the neighbor-joining approach with 1,000 bootstrap values.

Correlation between modules, nodes, and soil variables.

Many modules of the HS network were significantly correlated with soil variables (Fig. 5). Module H2 was significantly positively correlated to the AN content (r = 0.6, P = 0.01). Modules H1, H2, and H4 were significantly positively correlated to soil pH, suggesting that these three modules were stimulated by soil nitrogen and pH. Module H9 was significant negatively correlated to the SOM and AK contents. H11 was significant negatively correlated to soil pH.

FIG 5.

Correlations between module and soil variables. The numbers in each plot are the correlation coefficients (r), with the significance (P) given in parentheses. Red represents a positive correlation; green represents a negative correlation. (A) HS network; (B) BWS network.

In the BWS network, modules B6 and B4 were significantly positively correlated to the AP content. Modules B5 and B1 were significantly positively correlated to soil pH value. Modules B6 and B4 were significantly negatively correlated to soil pH value. Module B3 was significantly negatively correlated to the AN and AP contents. Module B2 was significantly negatively correlated to the AK content. Soil pH was well correlated with nearly half of the modules (8 of 17) in the two networks, suggesting that soil pH is the key environmental variable associated with the HS and BWS networks.

A total of 31 nodes shared among HS and BWS modules are positively correlated with the SOM, AP, AN, and soil pH value. Especially, 25 nodes are positively correlated with soil pH value, suggesting that soil pH is the main soil property affecting bacterial abundance and stimulating bacteria (Table 3).

TABLE 3.

Correlation analysis between OTU abundance and soil chemical properties

OTU	Correlation
	SOM	AP	An	AK	pH
OTU13	–0.359a	–0.181	–0.105	–0.15	0.371a
OTU19	0.118	–0.201	0.112	–0.103	0.444b
OTU31	–0.04	0.398a	0.053	0.171	–0.296
OTU40	–0.263	–0.176	–0.078	–0.329a	0.442b
OTU44	0.337a	0.356a	–0.065	0.095	–0.489b
OTU50	0.085	0.147	0.082	0.068	0.329a
OTU61	0.061	–0.208	0.151	0.035	0.502b
OTU80	0.078	0.055	0.166	–0.017	0.328a
OTU91	0.205	–0.315	0.021	0.080	0.373a
OTU94	–0.223	–0.086	–0.019	–0.246	0.475b
OTU97	0.200	0.097	0.423b	0.118	0.335a
OTU103	0.034	–0.052	0.091	–0.105	0.504b
OTU108	–0.178	–0.097	–0.089	–0.194	0.569b
OTU122	0.07	0.356a	–0.028	0.082	–0.461b
OTU152	0.117	0.316	0.450b	0.006	0.184
OTU177	0.026	–0.363a	–0.208	–0.051	0.346a
OTU179	–0.191	–0.147	–0.036	–0.225	0.500b
OTU180	0.257	0.455b	0.092	0.205	–0.386a
OTU188	–0.169	–0.247	–0.128	–0.030	0.511b
OTU191	–0.116	–0.076	0.067	–0.301	0.398a
OTU206	0.133	–0.284	0.209	0.083	0.374a
OTU218	–0.028	0.124	0.152	0.217	0.495b
OTU222	0.043	0.084	0.197	0.005	0.410a
OTU229	–0.215	–0.277	–0.248	–0.209	0.385a
OTU234	0.005	0.002	0.087	–0.327a	0.433b
OTU236	–0.202	0.132	0.092	0.090	0.380a
OTU283	–0.099	0.026	0.136	–0.226	0.575b
OTU308	0.135	0.054	0.383a	0.155	0.378a
OTU367	0.100	–0.197	0.364a	–0.020	0.098
OTU379	0.151	–0.061	0.122	–0.101	0.436b
OTU631	–0.160	–0.138	0.008	–0.239	0.493b

^aValue represents significant (P < 0.05) correlation between OTU abundance and soil property.

^bValue represents very significant (P < 0.01) correlation between OTU abundance and soil property.

DISCUSSION

In this study, we constructed microbial networks of the HSs and BWSs based on high-throughput sequencing data of 16S rRNA genes. Previous studies just analyzed the microbial species, abundance, and diversity (18, 26). Microbial network studies mainly analyze the interactions among different microbial species in the complex soil ecosystems. The interactions among microbial species are vital to ecosystem stability, especially to the soil microbial community (23, 33). Thus, the microbial network represents a new approach to analyze interactions of microbial populations. This study can help us to better understand the change of the microbial community in BWS in comparison to that in HS.

Based on microbial network analysis, it was found that the HS network was more complex and stable and contained more interacting microbial species (266 nodes, 514 edges) than the BWS network (204 nodes, 180 edges). Nodes in the HS network were connected with more edges (interactions) than the BWS network. In the HS network, different OTUs were tightly connected and formed a more stable network, which likely is better able to adapt to environmental changes (25, 34, 35). More interactions among microbial species help soil microorganisms fulfill functions such as participating in nutrient cycling, promoting plant growth, and suppressing pathogens (36, 37). More interacting bacteria in the HS network also suggests that there is more exchange of metabolites and information among microbial species, which makes the HS network work more efficiently than the BWS network (38). Hence, the microbial network of HSs could well explain the lower incidence of pathogens such as R. solanacearum in them. In contrast to the HS network, microbial populations of the BWS network were connected by fewer edges (interactions), and thus the exchange of materials and information between microbial species was possibly hampered and decreased in BWSs. As a result, the disease incidence and disease index of tobacco bacterial wilt disease in BWS were significantly higher than those in HS (Table 1, Fig. S3). We speculated that the better organized microbial network of HS can well suppress bacterial wilt disease while the unstable microbial network of BWS could lead to bacterial wilt disease.

Besides, the HS network had more key microorganisms (11 generalists) than the BWS network (1 generalist) (Fig. 3). Generalists are the key microorganisms in the microbial network and play important roles in the network. More generalists made the HS network more stable and ordered than the BWS network. More generalists, as keystone taxa, possibly made the exchange of materials and information among microbial species in the HS network occur more frequently than in the BWS network (25). Except for one generalist, most nodes of the BWS network were identified as peripherals (specialists), which were less connected with other nodes. Fewer key microorganisms meant that the structure of the BWS network was probably looser and more unstable. The exchange of materials and information among microbial species might be deficient and blocked in the BWS.

Compared with HS, the roles of microbial species were shifted in BWS. For example, generalists (such as OTU 8, OTU 14, OTU 19, OTU 73, OTU 12,2 and OTU 123) in the HS network became specialists in the BWS network (Fig. 3), suggesting that the ecological roles of key microorganisms in BWS were obviously different from those in HS. In the HS network, some generalists were plant-beneficial microorganisms (e.g., Bacillus and Actinobacteria), which can efficiently inhibit R. solanacearum (4, 39, 40). For example, OTU 122 belongs to Actinobacteria, which includes known antagonistic microorganisms against many plant pathogens (e.g., Fusarium oxysporum, Pyricularia oryzae Cavara, Pseudomonas, Rhizoctonia solani, R. solanacearum) (4, 27). OTU 123 was closely related to Bacillus spp. Bacillus species produce many biologically active compounds, including various antimicrobial substances (such as polyketides, lipopeptides, siderophores, and peptides) (41, 42). Cyclic lipopeptides produced by Bacillus (e.g., fengycin, iturin, mycosubtilin, surfactin) can inhibit various plant pathogens (41–46). Role shifts of these plant-beneficial generalists might lead to a BWS network that is unable to inhibit the pathogen R. solanacearum. Besides, the majority of OTUs play different roles in HS and BWS. Soil microorganisms possibly changed their physiological and metabolic characters in order to adapt to the different ecosystems of HS and BWS, which is consistent with other reports (25). Therefore, role shifts of OTUs might be related to the loss of suppressiveness to bacterial wilt and the prevalence of bacterial wilt when HS turned to BWS.

The proportion of nodes belonging to Actinobacteria in the HS network was higher than that in the BWS network (Fig. 2), suggesting that Actinobacteria were enriched in HS. Some Actinobacteria play important roles in degrading recalcitrant compounds (such as lignin and cellulose) and produce thousands of secondary metabolites, such as antibiotics that can effectively suppress various plant pathogens. For example, Streptomyces rochei and Streptomyces virginiae can efficiently control tobacco and tomato bacterial wilt diseases (4, 39, 47–50). Actinobacterial OTUs have more negative edges toward other OTUs in HS (138 negative edges) than in BWS (20 negative edges). Thereby, the enrichment of nodes belonging to some Actinobacteria in HS might be related to their strong suppressiveness to bacterial wilt disease. The obvious reduction in the number of nodes belonging to some Actinobacteria in BWS might be related to the higher disease incidence of bacterial wilt disease.

We also found that soil pH was the key soil variable associated with these two microbial networks. Soil pH was positively correlated with modules H2, H1, and H4 of the HS network but negatively correlated with modules B4 and B6 of the BWS network (Fig. 5), and 25 nodes shared among HS and BWS modules are positively correlated with soil pH value (Table 3), suggesting that soil pH is the main driver of bacterial network and bacterial abundance. This is consistent with previous studies which found significant correlations between soil pH and microbial community (37, 51–53). Bacterial diversity, richness, and community structure are adversely affected by soil acidity (54). Here, pH of BWS was significantly lower than that of HS (Table 1). We speculate that soil acidification negatively affects the stability of the BWS network.

In conclusion, this study found that the HS network was better organized and contained more interacting species and generalists than the BWS network. The balance of the microbial network was broken in BWSs. The BWS network was less stable and complex than the HS network. In comparison to HS, the roles of nodes were shifted in BWS.

MATERIALS AND METHODS

Soil sampling.

The study sites were located in the Enshi State (29°92′ to 29°97′N, 109°34′ to 109°40′E), Hubei Province, China. This area is exposed to a subtropical humid climate with an annual average temperature of 16°C and an annual rainfall of 1,400 to 1,500 mm. The soil type is yellow-brown soil (classified as Alfisols). The soil texture is qualified as loam. Tobacco has been continuously planted in this area for the last 15 years. Soil samples were collected from 37 different plots before the cultivation of tobacco in May. One composite soil sample was collected for each plot. In each plot (44 by 15 m), soils were randomly sampled at 15 different sites from tillage layer soils (0- to 20-cm depth and 10-cm diameter) and then were well mixed to form a composite soil sample. Samples were homogenized through a 2-mm sieve and then divided into two subsamples: one for DNA extraction and another for chemical property analysis.

Soil samples included two groups: the soils infected with R. solanacearum, the causative agent of tobacco bacterial wilt disease, termed bacteria wilt soils (BWSs), and soils with minimal levels of R. solanacearum, termed healthy soils (HSs) (Table 1; see Fig. S3 in the supplemental material) (18). Overall, 17 HS and 20 BWS samples were obtained.

Disease incidence and disease index of tobacco bacterial wilt disease.

From 1,000 tobacco seedlings per plot, a total of 60 tobacco seedlings were randomly selected from each plot at 90 days posttransplantation, and the disease incidence and disease index of bacterial wilt disease were recorded (18). Disease incidence was calculated by the percentage of diseased tobacco plants in each plot. Disease index was evaluated using the following disease score method: 0 = no symptom, 1 = less than one-half of tobacco leaves wilted, 3 = one-half to two-thirds of tobacco leaves wilted, 5 = more than two-thirds of tobacco leaves wilted, 7 = all leaves wilted, and 9 = stems collapsed or tobacco plants died (18). The incident index was calculated using the following formula:

$$\text{disease index}=[{\displaystyle \sum _{}(r\times N)}/(n\times R)]\times 100$$

where r is the disease severity, N is the number of infected tobacco plants with a rating of r, n is the total number of tobacco plants tested, and R is the value of the highest disease severity in each plot.

Soil characteristics.

Soil pH, contents of soil organic matter (SOM), alkali-hydrolyzable nitrogen (AN), available phosphate (AP), and available potassium (AK) were determined as previously described (18, 19). AK content was detected by flame photometry. The Mo-Sb colorimetry method was used to assay AP content. SOM content was determined by titrimetry with potassium dichromate. Soil pH was determined in a 1:2.5 soil-to-water suspension with a pH meter. AN content was determined by the alkaline hydrolysis diffusion method. The soil moisture content, bulk density, total porosity, and soil water-stable aggregates were also investigated.

High-throughput sequencing of bacterial 16S rRNA genes.

DNA was extracted from 0.4 g of soil using a FastDNA spin kit in accordance with the manufacturer’s procedures (18). The primer pair 338F (5′-ACTCCTACGGGAGGCAGCA-3′) and 806R (5′-GGACTACHVGGGTWTCTAAT-3′) was used for amplifying the V3-V4 hypervariable regions of the bacterial 16S rRNA gene. Amplicons were sequenced on an Illumina MiSeq platform (Illumina, Inc., USA) at Bohao Biotechnology Company, Shanghai, China. The sequence quality was statistically analyzed by CASAVA1.8. Raw sequence data were filtrated using the FASTX Toolkit 0.0.13 software package to remove low-quality sequences with a Q value (a measure of sequencing quality) of <20 and with less than 35 bp (18). Average read length was approximately 250 bp. Operational taxonomic units (OTUs) were assigned at 97% similarity level and were used to perform rarefaction analysis and calculate richness and diversity indices (55). Chao and Shannon indices were calculated to measure microbial richness and diversity.

The sequence numbers for individual samples were different. We standardized the OTU distribution matrix into the relative abundance based on the following equation (23):

$${{\displaystyle \mathit{RA}}}_{\mathit{ij}}=\frac{{{\displaystyle S}}_{\mathit{ij}}}{{\displaystyle {\sum }_{j=1}^n{{\displaystyle S}}_{\mathit{ij}}}}\times 100$$

where i is the i^th sample, j is the j^th out, and S_ij is the number of sequences in the i^th sample and j^th OTU. ${\displaystyle {\sum }_{j=1}^n{{\displaystyle S}}_{\mathit{ij}}}$ is the sum of sequences in the i^th sample. Thus, OTU abundances were normalized to the standardized relative abundances (SRA).

Construction and analysis of the microbial network.

We used an online tool, the Molecular Ecological Network Analysis Pipeline (MENAP), to construct the phylogenetic molecular ecological networks (pMENs) (http://ieg2.ou.edu/MENA). First, a matrix of OTU SRA (see Data Set S1 in the supplemental material showing a standardized OTU matrix for HS and BWS), a matrix of soil variables, and an OTU annotation file were prepared. The SRA matrix was submitted to MENAP to construct networks (24, 56). Second, modules were detected by the greedy modularity optimization. Three files were generated for network graph visualization by Cytoscape 3.4.0 software (57), which offers a good visualization interface of the complex networks (58, 59, 62). The network graph was represented using different OTUs (nodes) with positive or negative interactions (edges). Positive interactions indicate that the abundances of these OTUs changed following the same trend across different soil samples. Negative interactions indicate that the abundances of those OTUs changed following the opposite trend in different soil samples (25, 60).

A phylogenetic molecular ecological network (pMEN) is a representation of various microbial interactions (e.g., predation, competition, and mutualisms) in an ecosystem in which different microbial species (nodes) are connected by pairwise interactions (links). pMENs describe the cooccurrence of OTUs across different soil samples. pMENs were constructed for the microbial network of HS and BWS, respectively. Each network was composed of different modules. A module is a group of microbial species (nodes) that strongly interact with each other but rarely interact with nodes in other modules (60). Modularity (M) measures the degree to which a network is organized into clearly delimited modules.

The topological role of each node (OTU) was defined by two parameters: within-module connectivity (Z_i) and among-module connectivity (P_i). Z_i described how well a node was connected to other nodes within its own module. P_i described how well a node was connected to different modules (29). Threshold values of Z_i and P_i for categorizing nodes are 2.5 and 0.62, respectively. According to the threshold values of Z_i and P_i proposed by previous research (23, 25, 29–32), the nodes were divided into four categories. (i) Peripheral nodes (named specialists) had low Z_i (<2.5) and low P_i (<0.62) values. These nodes had only a few links and almost always were connected to the nodes within their own modules. (ii) Connectors (named generalists) had low Z_i values (<2.5) but high P_i values (>0.62). They were highly connected with other modules. (iii) Module hubs (also named generalists) had high Z_i values (>2.5) but low P_i values (<0.62). They were highly connected with many nodes in their own modules. (iv) Network hubs (named supergeneralists) had both high Z_i (>2.5) and P_i (>0.62) values. Generalists (connectors, module hubs) and network hubs are the key microorganisms, which maintain the network stability and play important roles (38).

Statistics analysis.

Differences in soil properties and network indexes between HS and BWS were compared by a least-significant-difference (LSD) test. Statistics of networks, including network properties, were calculated. For assessing statistical significance of network indexes, a total of 100 randomly rewired networks were generated. The differences in indexes between pMEN and random networks were detected by Z test for the HS and BWS networks, respectively. Correlation coefficients for networks were calculated on the basis of global network properties, the centrality of individual nodes, module separation, and modularity (56). The Mantel test was conducted to discern the relationships between module and soil properties using the R vegan package (61).

Accession number(s).

The Illumina MiSeq sequence data have been deposited at DDBJ/EMBL/GenBank under accession number KCKA00000000.