Heterogeneity of Soil Bacterial and Bacteriophage Communities in Three Rice Agroecosystems and Potential Impacts On Nutrient Cycling

Background: As genetic entities infecting and replicating only in bacteria, bacteriophages can regulate the community structure and functions of their host bacteria, but they are often overlooked because of their relatively low abundance. The ecological roles of bacteriophages in aquatic and forest environments have been widely explored, but those in agroecosystems remains limited. Here, we used metagenomic sequencing to analyze the diversity and interactions of bacteriophages and their host bacteria in soils from three typical rice agroecosystems in China: double cropping in Guangzhou, southern China, rice–wheat rotation cropping in Nanjing, eastern China and early maturing single cropping in Jiamusi, northeastern China. Bacteriophages were isolated and their functions on soil nitrogen cycling and effect on soil bacterial community structure were veried in pot inoculation experiments and Illumina MiSeq sequencing. Results: Soil bacterial and viral diversity and functions varied among the three agroecosystems. Genes detected in communities from the three agroecosystems were associated with typical functions; soil bacteria in Jiamusi were signicantly enriched in genes related to carbohydrate metabolism, in Nanjing with xenobiotics biodegradation and metabolism, and in Guangzhou with virulence factors and scarce in secondary metabolite biosynthesis, which might lead to a signicant occurrence of rice bacterial diseases. In the three ecosystems, 368 species of virus were detected. Notably, over-represented auxiliary carbohydrate-active enzyme (CAZyme) genes were identied in the viruses, which might assist host bacteria in metabolizing carbon, and 67.43% of these genes were present in Jiamusi. In bacteriophage isolation and inoculation experiments, Enterobacter bacteriophage-NJ reduced the nitrogen xation capacity of soil by lysing N-xing host bacteria and changed the soil bacterial diversity and community structure. Conclusions: Our results showed that diversity and function of paddy soil bacteria and viruses varied in the three agroecosystems. Soil bacteriophages can affect nutrient cycling by expressing auxiliary metabolic genes (AMGs) and lysing the host bacteria that are involved in biogeochemical cycles. These ndings form a basis for better understanding bacterial and bacteriophage diversity in different rice agroecosystems, laying a solid foundation for further of soil that production of healthy 0.01% for four bacteriophage groups, and between 0.01% and 1% for 12 bacteriophage groups. Thus, the dominant bacteriophages were mainly distributed in bacteria with moderate abundance. The relationship between the relative abundance of the bacteriophages and that of their hosts is complex. The ratio of viral relative abundance to bacterial relative abundance (VBR) varies widely, between 0 and 27,382. But when VBR was greater than 100 in at least one of the three cropping systems, the abundance of bacteriophage and host bacteria was negatively correlated (R 2 < -0.5). cycling through lysis. Enterobacter phage phiEap-2 was the most abundant in the tested soils from Nanjing, with a relative abundance of 15%. When we extracted all viruses from the Nanjing soils and added them into natural eld soil, the result was similar to adding bacteriophages alone; that is, soil nitrogen-xation capacity was reduced. In addition, the soil microbial community changed, and the bacterial metabolic functions such carbon metabolism and energy metabolism decreased also. These results indicate that bacteriophages can regulate soil bacterial community structure and ecological function and that the differences in soil bacterial function were related to differences in the bacteriophages in the different agroecosystems. which might lead to a signicant incidence of rice bacterial diseases. The 368 virus species that were detected from the three agroecosystems likely affect the soil microbial community structure and function. Bacteriophage isolation and inoculation experiments showed that Enterobacter phage-NJ not only changed the soil bacterial diversity and community structure, but also reduced the nitrogen-xation capacity by lysing nitrogen-xing host bacteria. In addition, auxiliary carbohydrate-active enzyme (CAZyme) genes were notably over-represented in soil viruses, and 67.43% of all CAZyme genes were present in Jiamusi soils, which might assist host bacteria in metabolizing carbon. Our study will contribute to a better understanding of the microbial diversity in the typical rice agroecosystems in China, viral mechanisms that may be involved in this diversity and inform management programs for the safe production of healthy rice. ltered supernatant and puried via

China, rice-wheat rotation cropping in central China, and early-maturing single cropping in northeastern China. We also tested the effects of bacteriophages on soil bacterial community structure and function in amended soils in pot experiments.

Results
Soil microbial communities and differences in three rice agroecosystems In the metagenomic sequencing, 1.10 billion raw reads were obtained. After trimming, 1.07 billion clean reads remained. In Guangzhou, Nanjing and Jiamusi, 511,481, 593,018, 848,794 contigs were obtained, respectively, after clean reads were assembled. Based on similarities to entries in the NCBI-NR database, 97.81% of sequences were classi ed as bacteria, 1.73% as Archaea, 0.27% as Eukaryota, and 0.026% as viruses. In total, 82 phyla, 1904 genera and 12,455 microbial species were identi ed. At the phylum level, Proteobacteria, Actinobacteria, Chloro exi, Acidobacteria and Bacteroidetes were the top ve mostabundant phyla, accounting for more than 78% of all the sequences (Fig. S1A). According to the LEfSe analysis (LDA > 3.5, p < 0.05), soils in Guangzhou and Nanjing were enriched in microbes belonging to Proteobacteria and Actinobacteria, respectively, while soils in Jiamusi were enriched in bacteria belonging to Chloro exi, Acidobacteria and Bacteroidetes (Fig. 1A). At the generic level, there were signi cant differences in the dominant genera (with abundance greater than 1% and without unclassi ed genera). There were 10 dominant genera in Guangzhou, 13 in Nanjing and nine in Jiamusi, only 20% of the total dominant genera were shared by the three rice agroecosystems, and 45% were unique to a single rice agroecosystem (Fig. 1B, Table S3). The top three most-dominant genera were Gemmatimonas, Anaeromyxobacter and Candidatus Solibacter in Guangzhou, Nocardioides, Gemmatimonas and Solirubrobacter in Nanjing, Bradyrhizobium, Candidatus Solibacter and Anaeromyxobacter in Jiamusi (Fig. S1B). It is worth noting that none were shared by the three rice agroecosystems. Our results that the microbial community composition differed among the three rice agroecosystems was con rmed by PCoA analysis. The principal coordinate axis 1 (PCo1) and the principal coordinate axis 2 (PCo2) explained, respectively, 26.66% and 19.22% of the variation in bacterial species Complicated relationship between bacteriophage and bacterial abundance and diversity Viral and bacterial genes were selected out to analysis their relationship. Because of the low abundance of viruses, metagenomic sequencing only detected viruses that were relatively abundant. In total, 14 families, 29 genera and 368 species of viruses were identi ed. As in the most-often characterized viromes in various habitats, the annotated viruses in three rice agroecosystems belonged to families Siphoviridae, Podoviridae, Myoviridae, Phycodnaviridae and Herpesviridae. The relative abundance of Phycodnaviridae was highest in Jiamusi, and Siphoviridae, Podoviridae, Myoviridae and Herpesviridae were highest in Nanjing ( Fig. 2A). Similar with the trend for the number of bacterial species, the number of viral species was highest in Nanjing (218 species), followed by Guangzhou (192 species) and Jiamusi (155 species). The three rice agroecosystems shared 13.39% of the total viral species, and 27.05% of the total viral species were shared by two agroecosystems; however, 59.56% of the viral species were speci c to one agroecosystem (Fig. 2B).
Bacteriophages that infect the same bacterial genus were grouped together. We analyzed the correlation between the relative abundance of dominant bacteriophage groups (relative abundance greater than 1%) and their host bacterial genera (Table 1). There were 19 dominant bacteriophage groups, the relative abundance of the hosts for three bacteriophage groups was greater than 1%, less than 0.01% for four bacteriophage groups, and between 0.01% and 1% for 12 bacteriophage groups. Thus, the dominant bacteriophages were mainly distributed in bacteria with moderate abundance. The relationship between the relative abundance of the bacteriophages and that of their hosts is complex. The ratio of viral relative abundance to bacterial relative abundance (VBR) varies widely, between 0 and 27,382. But when VBR was greater than 100 in at least one of the three cropping systems, the abundance of bacteriophage and host bacteria was negatively correlated (R 2 < -0.5). GZ: rice double cropping system in Guangzhou; JMS: rice single cropping system in Jiamusi; NJ: rice-wheat rotation cropping ecosystem in Nanjing. VBR: th abundance to bacterial relative abundance.
Metagenomic analysis of richness (represented by the Chao index) and diversity (represented by the Shannon index) revealed a signi cant difference in the bacterial and bacteriophage communities of paddy soils in different cropping systems. The Chao index showed that the bacterial richness in Nanjing was signi cantly higher than in Jiamusi and Guangzhou, while the difference between Jiamusi and Guangzhou was not signi cant. The viral richness was positively correlated with that of bacteria (R 2 = 0.662, p = 0.019) (Fig. 3A), and the viral richness of Nanjing was signi cantly higher than that of Jiamusi and Guangzhou. The Shannon index analysis found that the diversity of Nanjing and Guangzhou did not differ signi cantly, but both were signi cantly higher than in Jiamusi. Viral diversity was negatively correlated with that of bacteria (R 2 = -0.876, p = 0.001) (Fig. 3B).

Soil bacterial function and its relationship with environment in three agroecosystems
In the functional annotation of the soil metagenomic reads for the bacterial community using the Kyoto Encyclopedia of Genes and Genomes (KEGG) database, most sequences were associated with metabolism (70.24-72.38%), and far fewer were involved in genetic information processing (8.44-8.93%), environmental information processing (6.79-7.26%), cellular processes (5.49-6.01%), human diseases (3.93-4.57%), and organismal systems (2.71%-2.85%).
According to the LEfSe analysis (Fig. 4A), soil bacteria in Jiamusi were signi cantly enriched in genes associated with carbohydrate metabolism, replication and repair, biosynthesis of other secondary metabolites which were positively correlated with a high soil organic carbon (SOC) content ( Fig. 4B; Table S2).
While soil bacteria in Nanjing were signi cantly enriched in genes associated with xenobiotics biodegradation and metabolism, nucleotide metabolism, folding sorting and degradation, especially for genes involved in biodegradation and metabolism of nitrotoluene, aminobenzoate, naphthalene, steroid and xylene (Fig. S3). The annual frequency and dosage of chemical pesticide application especially herbicides in Nanjing were higher than those in other two areas (Table S2). As a result, pesticide residues in the soil were relatively high, and soils were signi cantly enriched in bacteria with xenobiotics biodegradation genes, presumably to help decompose pesticides in the soil.
Soil bacteria in Guangzhou were signi cantly enriched in genes associated with bacterial virulence, energy metabolism, translation, cell growth and death, which were positively correlated with annual accumulated temperature and annual precipitation. In the eld disease evaluation, we found that the disease indices for rice brown spot, brown streak, bacterial blight, bacterial leaf streak and foot rot were the highest in Guangzhou (Fig. 5A). Similarly, the relative abundance of the pathogenic bacteria Pseudomonas syringae, Acidovorax avenae, Xanthomonas oryzae and Dickeya zeae was signi cantly higher in Guangzhou than in Jiamusi and Nanjing (Fig. 5B) and was positively correlated with temperature and humidity (R 2 > 0.5, p < 0.05). In addition, the analysis of the virulence factor database VFDB showed that the abundance of genes related to factors (adherence-related genes) in the Guangzhou region was signi cantly higher than that in Nanjing and Jiamusi, especially for genes for agella, Trw type IV secretion system, mbriae, Myf/pH6 antigen, and PI-2, accounting for more than 70% of the total adherence-related genes (Fig. 5C). Functional analysis of biosynthesis of other secondary metabolites showed that the genes related to antimicrobial antibiotic secretion (such as biosynthesis of streptomycin, penicillin and cephalosporin, and neomycin, kanamycin and gentamicin) in Guangzhou were signi cantly lower compared with that of Nanjing and Jiamusi (Fig. 5D). Bacterial diseases may be more serious in Guangzhou due to the higher temperature and humidity and lower secretion of antibiotics, which may be conducive to the reproduction of pathogenic bacteria and to the higher relative abundance of adherence-related genes might aid their infection of the plant host.

Abundant auxiliary carbohydrate metabolic genes in paddy viruses
In the functional annotation of viruses in paddy soils using the KEGG database (Fig. 6A), almost all functions in pathway level 2 were found, but the sequences were notably enriched in a limited number of functions; the top three assigned functions accounted for 70% of the annotated sequences. The top two most-abundant functions were nucleotide metabolism, replication and repair, which are critical for the reproduction and survival of viruses. In addition, viruses have genes related to carbohydrate metabolism and the biosynthesis of secondary metabolites; the relative abundances of genes related to carbohydrate metabolism in viruses were negatively correlated (R 2 = -0.498, p = 0.334) with those in soil bacteria (Fig. S4A). Similarly, the abundances of genes related to biosynthesis of secondary metabolites in viruses were negatively correlated (R 2 = -0.500, p = 0.333) with those in soil bacteria (Fig. S4B). These two types of genes carried by bacteriophages might assist host bacteria in carbon metabolism and secondary metabolite synthesis. In the annotation of carbohydrate-active enzymes (CAZymes) using the CAZy database ( Fig. 6B), the species and abundance of viral CAZymes differed signi cantly among the soils of the three cropping systems. Notably, 67.43% of the CAZymes-related genes were present in Jiamusi, which was 9.83 and 2.62 times higher than in Guangzhou and Nanjing. In addition, there were two classes CAZymes (glycoside hydrolases, glycosyl transferases) in Guangzhou, two classes (glycoside hydrolases, carbohydrate-binding modules) in Nanjing, but four classes (glycoside hydrolases, carbohydrate-binding modules, glycosyl transferases, polysaccharide hydrolase) in Jiamusi. Remarkably, polysaccharide lyases (pectate lyase, exopolygalacturonate lyase, thiopeptidoglycan lyase) were unique to Jiamusi, implying potential roles in the decomposition of soil organic carbon.
Enterobacter bacteriophages can reduce the nitrogen xation capacity of soil by lysing host bacteria In Nanjing, the most abundant bacteriophage was Enterobacter phage, with a relative abundance of 15.28%, compared with 0.13% in Guangzhou and 0% in Jiamusi ( Fig. S5A). They were also negatively correlated with the abundance of host bacteria Enterobacter (R 2 = -0.805, p = 0.404) (Fig. S5B). To study its function, the bacteriophage was isolated from the soil in Nanjing and named Enterobacter phage-NJ (Fig. S6A). In the plate lytic experiment, the bacteriophage had lytic activity against nitrogen-xing Enterobacter (Fig. S6B).
To assess the functional effects of Enterobacter phage-NJ on nitrogen xation and uptake, we added Enterobacter phage-NJ to the sterilized soil to which nitrogen-xing Enterobacter had been added. The results showed that the nitrogen-xation capacity decreased by 55%, and plant height, fresh mass and nitrogen content of the rice plants were 14.27%, 32.25%, and 37.17%, respectively, lower than in the controls (Fig. 7B). Similar results were found when Enterobacter phage-NJ was added to untreated eld soil; the nitrogen-xation capacity decreased by 16.49%, and plant height, fresh mass, and nitrogen content were 9.29%, 17.97%, and 18.58%, respectively, lower than in the controls (Fig. 7C).
Rhizosphere microbial diversity in untreated eld soil after the addition of Enterobacter phage-NJ was then studied using 16S amplicon sequencing. After the addition of Enterobacter phage-NJ, the richness and diversity of rice rhizosphere microorganisms was 29.97% and 19.17%, respectively, lower than in the control (Fig. S7A). Moreover, the community structure of rhizosphere microorganisms signi cantly changed; PCoA demonstrated separate clustering (Fig.   S7B). In addition, the metabolic function of rhizosphere microorganisms also signi cantly decreased, especially carbohydrate metabolism, energy metabolism and biosynthesis of secondary metabolites (Fig. S7C).

Discussion
Functional differences of soil bacteri a in the three rice agroecosystems Soil bacteria in three rice agroecosystem performed multiple functions simultaneously and were signi cantly enriched in typical functions that were speci c for an agroecosystem. Soil bacteria in Jiamusi and Nanjing were signi cantly enriched in genes associated with carbohydrate metabolism, xenobiotics biodegradation and metabolism, respectively, while soil bacteria in Guangzhou were enriched in genes associated with virulence factors, but scarce in genes associated with secondary metabolite biosynthesis, which might lead to the signi cant occurrence of rice bacterial diseases. These signi cant differences are likely related to the local environment and cropping system. The effects of environment on soil bacterial function can be divided into direct and indirect effects. Environmental factors especially temperature can directly in uence the functions by accelerating the activity microbes [27,28] or indirectly affect functions by altering the composition of microbial communities [29]. Soils in Jiamusi were signi cantly enriched in genes associated with carbohydrate metabolism, which were positively correlated with the high content of soil organic matter. Soil organic matter can provide nutrients and promote multiplication of microbes, ultimately, indirectly promoting the capacity of carbon metabolism. With lower inputs of plant-derived organic carbon, desert communities have lower relative abundances of genes associated with nutrient cycling and the catabolism of plant-derived organic compounds compared with forests, grasslands, and tundra [30]. Our correlation analysis of environment and function showed that soil bacteria in Nanjing were signi cantly enriched in genes associated with xenobiotics biodegradation and metabolism, which were positively correlated with the large number of pesticide applications. Because of the rice-wheat rotation system in Nanjing, more types and applications of chemicals are used than in the double cropping system in Guangzhou and in the single cropping rice in Jiamusi. Chemical residues are bene cial to the survival of pesticide-degrading microbes and ultimately result in signi cant enrichment of the xenobiotics biodegradation and metabolism functions of soil bacteria in Nanjing. The genes associated with bacterial virulence factors in Guangzhou soils were positively correlated with higher annual accumulated temperature and annual precipitation. One of the most important conditions for rice bacterial diseases is high temperature and humidity [31]. In addition, double cropping rice in Guangzhou provides two growing seasons for hosts of bacterial pathogens, so pathogens have more time and opportunity to survive than Nanjing and Jiamusi. These results indicate that functions of soil bacteria vary in different agroecosystems and that environmental factors play important roles through affect microbial community composition.
Differences in soil viruses among the three agroecosystems and their relationship with host bacteria In our study, there were signi cant differences in the species and composition of paddy soil viruses among the three agroecosystems. In Nanjing, 40.64% more virus species were detected than in Jiamusi and 13.54% more than in Guangzhou. A previous study showed that the bacterial host species and abundance may be a key factor controlling viral abundance and distribution [32]. Similarly, we found that soil viral richness was signi cantly positively correlated with bacterial richness (R 2 = 0.662, p = 0.01). Only 13.39% of the viral species were shared by all three rice agroecosystems, while 59.56% of the viral species were environmentally speci c, which is likely associated with the signi cant differences in temperature, annual precipitation and soil physicochemical properties in the three rice agroecosystems because viral diversity and community structure have been reported to be controlled by factors such as temperature [33], soil moisture [34], and pH [35].
In soils, most viruses are bacteriophages, which can be divided into lysogenic or lytic bacteriophages. The abundance and distribution of bacteriophages and host bacteria in uence each other in a complex relationship based on our results. Bacteriophages are obligate intracellular parasites, so their tness and range limits are in uenced by their hosts, and the bacteriophages in turn can affect host abundance and even the community structure and diversity of the whole bacterial community [36-38]. Lytic bacteriophages can modulate host bacterial populations and diversities by lysing their hosts [16], but lysogenic bacteriophages can increase the host abundance by promoting horizontal gene transfer by lysogenic conversion or transferring packaged bacterial genes to new hosts, which may expand the host metabolic pro le and enhance microbial environmental adaptability [39]. The ratio of viral abundance to bacterial abundance (VBR) has been used as means to infer relationships between viruses and their potential hosts [40]. Lower VBRs means low virion production, and lysogeny is thus considered to be common. Conversely, higher VBR means higher virion production, and lysogeny is considered to be less prevalent; bacteriophages may thus be primarily lytic [41]. In cold deserts, the VBR ranged from 0.15 to 1.66, and the prevalence of bacterial lysogens was between 4.6 and 21.1% [42,43], while in hot deserts, the VBR ranged from 170 to 8,200, and the prevalence of bacterial lysogens was 84%, much higher than in the cold desert soils [44]. In our analysis of the ratio of viral relative abundance to bacterial relative abundance, similar to the results of previous studies [41,42,44]; when the ratio was greater than 100, there was a signi cant negative correlation between the abundance of bacteriophage and bacteria, indicating that bacteriophages may be lytic, which negatively regulates the abundance of host bacteria. However, when the ratio was less than 100, the correlation between bacteriophage and bacteria abundance was not signi cant, perhaps because the bacteriophages may be lysogenic, which have little effect on host bacteria abundance, and/or because bacteriophages overlap in their distribution with multiple susceptible hosts in natural environment, while bacteria can often be infected with several different types of bacteriophages. Furthermore, many factors in uence the abundance of bacteriophages and host bacteria, making the relationship between them more complex.

Bacteriophages inhibit soil nitrogen xation by lysing host bacteria
Lytic bacteriophages have direct and indirect effects on nutrient cycling. They can directly lyse bacteria, which releases organic matter, including nutrients such as carbon, nitrogen and phosphorus. The carbon cycle driven by viruses accounts for 25% of the total carbon cycle in marine agroecosystems [16].
Addition of bacteriophages to soil also increases the NH 4 + concentration via lysis, resulting in a release of inorganic nitrogen, followed by mineralization [45].
Indirectly, bacteriophage lysis can also slow nutrient transformations by reducing the population of key bacteria involved in nutrient cycling [45]. Higher abundance of T4-like phages increases bacterial death, thereby suppressing soil organic carbon mineralization [46]. In wetlands, bacteriophages can infect sulfate-reducing and methanogenic bacteria, which can potentially repress sulfate reduction (and associated carbon mineralization) and methane production, respectively [47]. However, indirect inhibition of biogeochemical cycling by bacteriophages has mostly been inferred by sequencing analysis without relevant experimental veri cation. In our study, Enterobacter phage-NJ, which was isolated from the soil in Nanjing, lysed three species of nitrogen-xing Enterobacter bacteria. When Enterobacter phages-NJ were added to untreated eld soil or sterilized soil with nitrogen-xing bacteria, the soil nitrogen-xation capacity decreased, and rice growth and nitrogen content decreased also. This result provides direct evidence that bacteriophages can inhibit biogeochemical cycling through lysis. Enterobacter phage phiEap-2 was the most abundant in the tested soils from Nanjing, with a relative abundance of 15%. When we extracted all viruses from the Nanjing soils and added them into natural eld soil, the result was similar to adding bacteriophages alone; that is, soil nitrogen-xation capacity was reduced. In addition, the soil microbial community changed, and the bacterial metabolic functions such carbon metabolism and energy metabolism decreased also. These results indicate that bacteriophages can regulate soil bacterial community structure and ecological function and that the differences in soil bacterial function were related to differences in the bacteriophages in the different agroecosystems.
Potential effects of auxiliary metabolic genes of phages on soil microbial function Bacteriophages can affect nutrient cycling not only by lysing the host but also by metabolically reprogramming their hosts via the horizontal transfer of ecologically important genes and the expression of virus-carried auxiliary metabolic genes (AMGs) [48]. AMGs have been most extensively explored in marine ecosystem and are involved in photosynthesis, carbon metabolism, nitrogen metabolism and phosphate uptake [21]. Those genes are hypothesized to increase viral survival and replication by augmenting key steps in host metabolism. Cyanobacteria can photosynthesize in the ocean to obtain energy for growth, but they are susceptible to photoinhibition when sunlight is too intense. Cyanobacteria bacteriophage S-PM2 carries psbA and psbD genes, which encode proteins D1 and D2, key components of photosystem II, which are crucial sites of damage in photoinhibition. Bacterial infection with bacteriophage S-PM2 ensures that photoinhibition is prevented, thereby maintaining photosynthetic activity and energy production for the continued replication of the bacteriophage [18]. Marine viruses contain carbon metabolic genes that encode numerous enzymes involved carbon metabolism and have been shown to increase metabolic ux when glucose is limited [49,50]. Bacteriophages in different ocean environments also harbor niche-speci c AMGs. For example, in sunlit ocean water, photosynthesis-related genes are the main AMGs, but in dark ocean water, genes involved in the tricarboxylic acid cycle and electron transport chain prevail [49].
In mangrove soil [24], permafrost soil [51], and maize-barley rotation soil [20], AMGs involved in carbohydrate metabolism are notably over-represented. In the maize-barley rotation soil, three classes of genes-carbohydrate-binding modules, carbohydrate esterases, glycoside hydrolases-were identi ed. Similarly, in paddy soil, four classes of AMGs of bacteriophages involved in carbon metabolism were identi ed in our study (glycoside hydrolases, glycosyl transferases, polysaccharide lyases, and carbohydrate-binding modules, with the abundance of glycoside hydrolases the highest), indicating that paddy soil viruses primarily participate in the decomposition of organic carbon, breakdown of complex carbohydrates to increase energy production and boost bacteriophage replication. The species and abundance of AMGs also differed among the three agroecosystems we studied. Compared with the bacteriophages in the other two agroecosystems, the bacteriophages in Jiamusi had the greatest diversity of AMGs genes related to carbon metabolism; there were four classes of CAZymes in Jiamusi (polysaccharide lyases, glycoside hydrolases, glycosyl transferases, carbohydrate-binding modules), but only two in Guangzhou (glycoside hydrolases, glycosyl transferases) and in Nanjing (glycoside hydrolases, carbohydrate-binding modules). However, soil bacteria in Jiamusi were signi cantly enriched for carbon metabolism function, likely due to the large diversity of bacteriophage AMGs genes related to carbon metabolism. Remarkably, polysaccharide lyases (pectate lyase, exopolygalacturonate lyase, thiopeptidoglycan lyase) were unique to Jiamusi. Complex carbohydrates or polysaccharides such as cellulose, xylan, pectin, starch, alginate, mannan, and chitin are major components of plant cell walls, which are di cult to degrade.
In Jiamusi, the soil organic matter content is the highest, perhaps because the bacteriophages carry genes for polysaccharide lyases, which might assist host bacteria in degrading organic matter in the soil.

Conclusions
In our analyses to identify and compare the bacterial and viral communities in soils from the three most important rice agroecosystems in China, the communities in Jiamusi were signi cantly enriched in genes related to carbohydrate metabolism, in Nanjing with xenobiotics biodegradation and metabolism, and in Guangzhou with genes associated with virulence factors, but those involved with secondary metabolite biosynthesis were scarce, which might lead to a signi cant incidence of rice bacterial diseases. The 368 virus species that were detected from the three agroecosystems likely affect the soil microbial community structure and function. Bacteriophage isolation and inoculation experiments showed that Enterobacter phage-NJ not only changed the soil bacterial diversity and community structure, but also reduced the nitrogen-xation capacity by lysing nitrogen-xing host bacteria. In addition, auxiliary carbohydrate-active enzyme (CAZyme) genes were notably over-represented in soil viruses, and 67.43% of all CAZyme genes were present in Jiamusi soils, which might assist host bacteria in metabolizing carbon. Our study will contribute to a better understanding of the microbial diversity in the typical rice agroecosystems in China, viral mechanisms that may be involved in this diversity and inform management programs for the safe production of healthy rice. Province in China. In Jiamusi, Nanjing and Guangzhou, the annual precipitation is 510 mm, 1090 mm and 1720 mm, respectively, with 67%, 73%, and 77% mean humidity, respectively, and the annual accumulated temperature of 2700°C, 5400°C and 7500°C, respectively (http://www.weather.com.cn). The three regions different in their rice planting systems (single rice cropping in Jiamusi, rice-wheat rotation in Nanjing, double rice cropping in Guangzhou). Rice was grown at each farm using local practices for cultivation, fertilizer application, and eld management. Chemical pesticides were applied as shown in Table S1. Analyses of soil chemical properties Soil pH was determined using a Mettler-Toledo FE20-Five Easy PlusTM pH meter (Schwerzenbach, Switzerland) in a 1:5 soil-water (w/v) suspension. Soil organic carbon was measured using the chromic acid titration method [53]. Soil available nitrogen was determined by the alkali diffusion method, soil available phosphorus was determined using the sodium hydrogen carbonate solution-Mo-Sb anti spectrophotometric method, soil available potassium was measured by the ammonium acetate method, followed by ame photometric detection [54]. Isolation of Enterobacter bacteriophage from paddy soil Soil sample (10 g) was suspended in 50 mL Luria Bertani (LB) broth (1% tryptone, 0.5% yeast extract, and 1% NaCl) and dispersed by shaking for 12 h at 30°C.

Methods
The soil particles were removed by centrifugation at 4000 ´ g for 15 min, and the supernatant was lter-sterilized through a 0.22 μm lter. One milliliter of ltered supernatant was added to 10 mL of an overnight suspension culture of Enterobacter cloacae (proven to be nitrogen-xing and provided by Agricultural Culture Collection of China) and incubated for 12 h at 30°C [56]. Then bacterial cells were removed by centrifugation at 4000 ´ g for 15 min, and the supernatant was sterilized through a 0.22 μm lter. Then 0.1 mL of the ltered supernatant was mixed with 1 mL of the E. cloacae suspension (OD600 = 1). This mixture was then added to 2.5 mL of soft agar (0.35% agar prepared in 1% tryptone, 0.5% NaCl, 3 mM MgCl 2 , 3 mM CaCl 2 , and 0.04% [wt / vol] glucose) and poured in a uniform layer onto an LB agar plate. Bacteriophage plaques were identi ed after an overnight incubation at 30°C. Bacteriophages were puri ed via ve consecutive transfers of them from individual plaques to new bacterial cell lawns. The puri ed bacteriophage titer was determined by mixing 100 μL of host cells (A600 of 1) and 100 μL of bacteriophage dilution with soft agar and pouring the mixture over LB agar in a plate. Plaques were scored after an overnight incubation at 30°C. Bacteriophages were kept in buffer (10 mM Tris [pH 7.6], 5 mM MgSO 4 ·7H 2 O, 0.01% gelatin) at 4°C.
Each isolated Enterobacter phage was analyzed for its ability to lyse E. ancerogenus and E. ludwigii (proven to be nitrogen-xing and provided by Agricultural Culture Collection of China) using a plate lytic experiment with 0.1 mL bacteriophage mixed with 1 mL of overnight-cultured Enterobacter strain and shaken (200 rpm) at room temperature for 20 min, then 3 mL of soft nutrient agar was added, and the mixture was poured on LB agar plates and incubated overnight at 30°C. Lytic activity was indicated by transparent plaques [57].

Impact of Enterobacter bacteriophages on soil nitrogen-xation capacity
In a greenhouse experiment, the in uence of Enterobacter phages on soil nitrogen xation capacity in untreated eld soil and in sterilized eld soil was investigated. Rice seeds were surface-sterilized and pregerminated, then seedlings with a primary root length of 1-2 cm were planted in pots (6 seeds per pot) with the untreated eld soil. Three treatments with four pots each (each pot lled with 1 kg soil) were set up: (1) N+NJ virus, virus isolated from 1 kg Nanjing soil was diluted 50 times with ddH 2 O and added to each pot after planting rice; (2) N+phage, 1 mL Enterobacter bacteriophages (10 7 PFUs) diluted 50 times with ddH 2 O was added after planting rice; (3) NCK, 50 mL ddH 2 O was added as control. After treatment, rice plants were grown in a greenhouse at 26°C with 14 h of uorescent light and 10 h of darkness for 7 weeks and watered with 20 mL ddH 2 O every 3 days.
Untreated eld soil was autoclaved using the liquid cycle for 30 min at 121 °C, then cooled at room temperature, then we repeated this procedure, sterilized soil was stored at 4 °C until use. Mixed Enterobacter species (E. cloacae-211, E. ancerogenus-212, E. ludwigii-213, ratio 1:1:1) were cultured in LB broth overnight at 30 °C, followed by centrifugation at 1,000 ´ g for 5 min. The pellet was resuspended in ddH 2 O and diluted to an optical density OD600 of 0.02. Rice seeds were surface-sterilized and pregerminated, then seedlings with primary root length of 1-2 cm) were planted in pots (6 seeds per pot) with sterilized soil. Four treatments were set up with four pots per treatment: (1) S+NF, 50 mL Enterobacter strains were added after rice planting, then cultured for a week and added with ddH 2 O; (2) S+NF+virus, 50 mL Enterobacter strains were added after rice planting, then cultured for a week and added with virus isolated from 1 kg Nanjing soil, which was diluted to 50 mL with ddH 2 O; (3) S+NF+phage, 50 mL Enterobacter strains were added after rice planting, then cultured for a week and watered with 1 mL Enterobacter bacteriophages (107 PFUs) was diluted to 50 mL with ddH 2 O; (4) SCK, sterile water was used as the control. After treatment, rice plants were grown in a greenhouse at 26°C with 14 h of uorescent light and 10 h of darkness for 7 weeks and watered with 20 mL ddH 2 O every 3 days.
After 7 weeks, rice height and fresh mass were measured. Then rice plants were dried and grounded, nitrogen concentrations were determined with a modi ed Kjeldahl digestion method. N 2 xation in soil samples was assessed indirectly using the acetylene reduction assay measuring ethylene formation from acetylene.

Illumina MiSeq sequencing
Region V3-V4 of the bacterial 16S rRNA gene was ampli ed from each sample using primers 338F (5'-ACTCCTACGGGAGGCAGCAG-3') and 806R (5'-GGACTACHVGGGTWTCTAAT-3') in a thermocycler PCR system (GeneAmp 9700, ABI, Foster, CA, United States); barcode was added before the primer to build different libraries in the same sequencing pool. The PCR mixture contained 4 μL 5× TransStart FastPfu buffer, 2 μL 2.5 mM dNTPs, 0.8 μL 5 μM forward primer, 0.8 μL 5 μM reverse primer, 0.4 μL TransStart FastPfu DNA Polymerase, 10 ng template DNA, with ddH 2 O added to reach 20 μL. PCR cycling conditions were initial denaturation at 95°C for 3 min; 35 cycles of denaturing at 95°C for 30 s, annealing at 55°C for 30 s and extension at 72 °C for 45 s; and single extension at 72°C for 10 min, and end at 4 °C. The PCR product was extracted from 2% agarose gel and puri ed using the AxyPrep DNA Gel Extraction Kit The raw sequencing reads were demultiplexed, quality-ltered by Trimmomatic and merged by FLASH [58]. Operational taxonomic units (OTUs) with 97% similarity cutoff were clustered using UPARSE (version 7.1, http://drive5.com/uparse/), and chimeric sequences were identi ed and removed. The abundance of OTUs was rare ed to the lowest number of shared OTUs to remove the effect of sequencing depth across samples. The taxonomy of each 16S gene sequence was analyzed using the UNITE (version7.2) database with a con dence threshold of 70% [59]. analysis identi ed remarkable enrichment of taxa from phylum to genus level. Different color nodes represent the microbial groups that are signi cantly higher in the corresponding groups and have a signi cant in uence on the difference between the groups. Light yellow nodes represent the microbial groups that contribute no signi cant difference in different groups or in uence on the difference between groups.      Supplementary Files