Rice Root-Associated Diazotrophic Community Succession is Driven by Growth Period Combined With Fertilization

Biological nitrogen xation (BNF) and nitrogenous fertilizers are two crucial ways for nitrogen input in the rice paddy system. Little is known about the effect of nitrogenous fertilizers on root-associated diazotrophs. Here, we investigated the succession of total and active diazotrophs in rhizosphere soil and rice roots under four fertilization treatments (control, NPK, NPK + pig manure, and NPK + rice straw) during three key growth periods (tillering, heading and mature). The 15 N isotope dilution experiment veried that root-associated diazotrophs could supply N for rice under low nitrogen nutrition by BNF (%Ndfa=11.51). Niche differentiation of diazotrophs existed at the rhizosphere soil-root interface. At both DNA and RNA levels, growth period had stronger effects on the community composition of endophytic diazotrophs than rhizosphere diazotrophs. The Chao1 and Shannon indices of total endophytic diazotrophs dramatically increased at the vegetative stage, and underwent relatively minor changes at the reproductive stage. Furthermore, the community structures of total endophytic diazotrophs were more stabilized in the reproductive stage than the vegetative stage. The number of OTUs shared by rhizosphere soil and roots increased during rice growth. Compared with CK, NPK reduced the relative abundances of Pelobacter and increased the relative abundances of Azoarcus. Pig manure not only improved soil nutrients (OM, TN, TP, AN, AP and NO3-N), but reduced the effect of chemical fertilizers on the community composition of natural rhizosphere diazotrophs. the combined effects of growth period and fertilization on root-associated diazotrophs the sustainable


Abstract Background
Biological nitrogen xation (BNF) and nitrogenous fertilizers are two crucial ways for nitrogen input in the rice paddy system. Little is known about the effect of nitrogenous fertilizers on root-associated diazotrophs. Here, we investigated the succession of total and active diazotrophs in rhizosphere soil and rice roots under four fertilization treatments (control, NPK, NPK + pig manure, and NPK + rice straw) during three key growth periods (tillering, heading and mature).

Results
The 15 N isotope dilution experiment veri ed that root-associated diazotrophs could supply N for rice under low nitrogen nutrition by BNF (%Ndfa=11.51). Niche differentiation of diazotrophs existed at the rhizosphere soil-root interface. At both DNA and RNA levels, growth period had stronger effects on the community composition of endophytic diazotrophs than rhizosphere diazotrophs. The Chao1 and Shannon indices of total endophytic diazotrophs dramatically increased at the vegetative stage, and underwent relatively minor changes at the reproductive stage. Furthermore, the community structures of total endophytic diazotrophs were more stabilized in the reproductive stage than the vegetative stage. The number of OTUs shared by rhizosphere soil and roots increased during rice growth. Compared with CK, NPK reduced the relative abundances of Pelobacter and increased the relative abundances of Azoarcus. Pig manure not only improved soil nutrients (OM, TN, TP, AN, AP and NO3-N), but reduced the effect of chemical fertilizers on the community composition of natural rhizosphere diazotrophs.

Conclusions
Growth period demonstrated a stronger in uence on the root-associated diazotrophs than fertilization practices. The community structures of total endophytic diazotrophs were more stabilized in the reproductive stage than the vegetative stage, and the alpha-diversity indices and complex of network structures of endophytic diazotrophs increased at the vegetative stage. Replacing chemical fertilization with pig manure not only increased soil nutrients, but regulated rhizosphere diazotrophic community structures. Understanding the combined effects of growth period and fertilization on root-associated diazotrophs presents the basis towards the sustainable crop production.

Background
As a critical macronutrient, nitrogen (N) often limits plant yields in the agricultural ecosystem [1]. Over the past several decades, chemical fertilizers increase crop production, however, nitrogen-use e ciency (NUE) in rice is only about 39% [2] and excessive N causes soil acidi cation and environmental pollution [3]. Alternatively, BNF is an available and ecofriendly way to provide N for plants, which can reduce atmospheric dinitrogen to biologically available ammonium by nitrogen-xing bacteria and archaea. Numerous Diazotrophs could colonize and reproduce in the bulk, rhizosphere soil and rice tissues [4][5][6][7], and increased crop productivity through BNF [8], producing antimicrobials [9], and improving abiotic stress tolerance [10].

BNF plays an important role in N supply during rice growth
The rice yields of three fertilization treatments (NPK, NPKM, NPKS) were signi cantly higher than CK, while the yields of CK kept consistent since 2015 (Fig. 1a). Non-symbiotic BNF supplied 37.32% crop N for rice [26]. Thus, we inferred that nitrogen-xing bacteria of CK played an important role to maintain rice production. The 15 N isotope dilution method was used to evaluate the contribution of endophytic diazotrophs in the nitrogen nutrition supply for rice growth ( Table 1). The root length, plant height and fresh weight of rice of C 20 and T 20 had no signi cant difference but were signi cantly lower than T 10 . The N concentration among the three treatments had no signi cant difference, but δ 15 N of C 20 (33.26%±1.10%) and T 20 (33.85%±0.25%) were signi cantly higher than T 10 (29.43%±0.49%). It certi ed that endophytic nitrogen-xing bacteria could supply N for rice by BNF under low nitrogen nutrition (%Ndfa=11.51).
The diversity of total rhizosphere and endophytic diazotrophs were shaped by growth period, year and fertilization Illumina sequencing was performed to investigate the community compositions of rhizosphere and endophytic diazotrophs and their response to different fertilization managements in different growth periods. 318 samples collected from the eld (120 rhizosphere soil DNA samples, 119 root DNA samples, 48 rhizosphere soil RNA samples and 31 root RNA samples) were sequenced, and after normalization, each sample had 22552 sequences. These sequences were clustered into 706 operational taxonomic units (OTUs) at 90% identity.
For alpha diversity analyses, richness index (Chao1), evenness index (Shannon) and phylogenetic diversity index (PD) of rhizosphere diazotrophs were signi cantly higher than endophytic diazotrophs at the DNA level (Additional le 4: Table   S4), which was consistent with the results of rarefaction curves (Additional le 5: Figure S1). Interestingly, total rhizosphere diazotrophs had the highest α-diversity at the tillering stage, while total endophytic diazotrophs had the lowest α-diversity at this stage (Additional le 4: Table S4). The effects of growth period on all tested α-diversity indices were statistically signi cant, but fertilization only had signi cant effects on the evenness index (Table 2). For soil and root DNA samples, almost all α-diversity indices of diazotrophs among four fertilization treatments (CK, NPK, NPKM and NPKS) had no signi cant difference (Additional le 6: Table S5, Additional le 7: Table S6).
The phylogenetic trees of 50 dominant OTUs based on the sequences of nifH genes were generated (Fig. 2, Additional le 8: Figure S2). The relative abundances of dominant OTUs of total rhizosphere diazotrophs were notably different from total endophytic diazotrophs. Moreover, the whole community structure of total rhizosphere diazotrophs was notably different from and more stable than total endophytic diazotrophs (Additional le 9: Table S7, Fig. 1b). We visualized and quanti ed the differences between diazotrophic communities (β-diversity) using non-metric multidimensional scaling (NMDS) and permutational multivariate analysis of variance (PERMANOVA). The growth period was a primary driver of the total rhizosphere (R 2 =0.22, P<0.001) and total endophytic diazotrophic β-diversity (R 2 =0.33, P<0.001) ( Fig. 1c-d, Additional le 10: Figure S3a-b, Table 2), revealing a stronger temporal effect on total endophytic diazotrophs than total rhizosphere diazotrophs. However, total rhizosphere diazotrophs were more sensitive to the year and fertilization management (R 2 =0.19 and 0.11, P<0.001) than total endophytic diazotrophs (R 2 =0.06 and 0.05, P<0.001). Year, growth period and fertilization management showed signi cant interactions in pairs.
In these four cluster trees (Fig. 1e, Additional le 12: Figure S5), soil DNA samples and root DNA samples were divided into two clusters, and samples from the same growth period were frequently clustered together except for soil samples in 2016. However, PERMANOVA and ANOSIM suggested that the growth period signi cantly affected the total rhizosphere diazotrophic communities in 2016 (data not shown). Generally, the communities of either total rhizosphere or endophytic diazotrophs of NPK and NPKS treatments were clustered together, and they were separated from CK and NPKM treatments (Fig. 1e, Additional le 12: Figure S5a). Furthermore, the soil samples of CK were quite different from the other three fertilization treatments, and differences between CK and NPKM were smaller than CK and NPK or CK and NPKS (Additional le 13: Table S8). We noted that the relative abundances of Pelobacter of NPKM were lower than that of CK and higher than that of NPK, but Azoarcus was the opposite (Additional le 14: Figure S6). Furthermore, for Bradyrhizobium, Geobacter, Heliobacterium, Rubrivivax and Desulfomonile, their abundance of of NPKM were between that of CK and NPK at most sampling points.
The rhizosphere soil physicochemical indices OM, TN, TK, TK, AN, AP, AK, pH, NO3-N and NH4-N were tested (Additional le 15: Table S9). Soil pH value of CK was highest among four fertilization treatments. NPKM increased the content of organic matter (OM), total N (TN), total P (TP), available N (AN) and available P (AP) and NO3-N. Fertilization management, year and period had signi cant effects on all tested physicochemical parameters (P<0.01, Additional le 16: Table S10), except that the period had no signi cant effects on TN and NH4-N (P>0.05). Most rhizosphere and endophytic diazotrophic α-diversity indices had negative correlation with TP and AN at six sampling points (Additional le 17: Table  S11). The Mantel test showed that AK, TK and NH4-N were signi cantly negatively correlated with the community structures of total rhizosphere and endophytic diazotrophs (Additional le 18: Table S12). Correlation analysis showed that pH had a signi cant positive correlation with Pelobacter, but had a signi cant negative correlation with Thermodesulfovibrio in rhizosphere soil (Additional le 19: Figure S7).
The community structures of total endophytic diazotrophs were more stabilized in the reproductive stage than the vegetative stage The similarity distances (1-Bray dissimilarity) of diazotrophic community structures among three rice growth periods were calculated at each fertilization treatment ( Fig. 3a-b). For root samples, the similarity distances between the tillering and heading stages (red nodes) were higher than the tillering and mature stages (green nodes), but lower than the heading and mature stages (blue nodes). However, this trend was not observed in soil samples.
Tracking diazotrophic community changes at the class level revealed that many classes showed distinct temporal dynamics in root samples ( Fig. 3c-d). For example, Alphaproteobacteria and Deltaproteobacteria enriched as time went on, but Betaproteobacteria decreased. The endophytic diazotrophic abundance of "others" (mainly unclassi ed taxa) remarkably increased, and was close to rhizosphere diazotrophs at the mature stage. The abundance of shared OTUs between soil and root samples in soil samples (Pink ows) apparently increased during rice cultivation, but slightly decreased in root samples.
The linear discriminate analysis (LDA) effect size (LEfSe) was used to identify the distinguishing diazotrophs during different growth periods in paddy soil and rice roots [27]. The numbers of the discriminating taxa of rhizosphere diazotrophs, whose LDA scores between different growth periods greater than 2, were 172 and 96 in 2015 and 2016, and of the endophytic diazotrophs were 149 and 97 (data not shown). The discriminating genus signi cantly increasing at the same sampling period over two years were ltered and heatmaps were drawn to demonstrate more particular information of the responses of these genera to the growth period ( Fig. 3e-f). For the rhizosphere soil samples, the relative abundances of Treponema, Chloroherpeton, Paludibacter, Coraliomargarita and Desulfobulbus were signi cantly higher at the tillering stage than at the heading and mature stage (Fig. 3e), and they were correlated positively with AK (Additional le 19: Figure S7a-b); the relative abundances of Oscillatoria, Cyanothece and Coleofasciculus were signi cantly higher at the heading stage, and Oscillatoria was correlated negatively with NO3-N; the relative abundances of Heliobacterium, Sinorhizobium and Frankia were signi cantly higher at the mature stage, and they were correlated positively with NO3-N. For root samples, the relative abundance of Tolumonas was signi cantly higher at the tillering stage than at the other stages (Fig. 3f); the relative abundances of Thiorhodospira, Sulfurospirillum, Pelosinus and Selenomonas were signi cantly higher at the heading stage, and Thiorhodospira was correlated negatively with pH and NO3-N; the relative abundance of Rhizobium, Desulfovibrio, Geobacter, Frankia and Acetobacterium were signi cantly higher at the mature stage, and they were correlated negatively with AK (Additional le 19: Figure S7c-d).
Diversity and community structures of active nitrogen-xing bacteria Consistent with the total root-associated diazotrophs, the community structures of active rhizosphere diazotrophs were notably different from active endophytic diazotrophs (Additional le 9: Table S7); growth period was a primary driver of the active rhizosphere (R 2 =0.15, P<0.001) and active endophytic diazotrophic β-diversity (R 2 =0.17, P<0.001); fertilization management had a higher effect on the β-diversity of active rhizosphere diazotrophic (R 2 =0.14, P<0.001) than endophytic diazotrophs (R 2 =0.10, P>0.05) (Additional le 20: Table S13, Additional le 10: Figure S3c-d). The rhizosphere diazotrophic community structures of CK and NPKM had no signi cant difference, and both of them had signi cant differences with the rhizosphere diazotrophic community structures of NPK and NPKS. (Additional le 13: Table S8). The relative abundances of Geobacter of NPKM were higher than that of CK and lower than that of NPK in soil RNA samples (Additional le 14: Figure S6). Furthermore, TP, and AN, and had negative correlation with α-diversity indices of active rhizosphere and endophytic diazotrophs, and NH4-N and NO3-N had signi cant negative correlation with active rhizosphere diazotrophs (Additional le 17: Table S11). The Mantel test showed that TK, AN, AP, pH and NO3-N were signi cantly correlated with the community structures of active rhizosphere diazotrophs, and NO3-N had a signi cant correlation with the community structures of active endophytic diazotrophs (Additional le 18: Table S12).
Richness, Evenness and PD indices of all DNA samples were higher than corresponding RNA samples (Additional le 4: Table S4). The β-diversity of the total and active diazotrophic communities was investigated as well (Additional le 10: Figure S3e). The soil DNA samples clustered together, whereas the soil RNA samples, root DNA samples and root RNA samples were more dispersed. To compare their aggregation difference, the similarities (Z scores) in between-group diazotrophic community composition were visualized in bar plots. Z scores of the soil RNA samples, root DNA samples and root RNA samples had not signi cant differences but they were signi cantly lower than the soil DNA samples (Fig.   4a).
The rhizosphere DNA and RNA samples separated on the cluster tree (Fig. 4c), but root DNA samples and their corresponding RNA samples frequently clustered together (Fig. 4d). The correlation coe cient of the total and active diazotrophic communities in soil (Spearman correlation: r= 0.244, P<0.05) was slightly smaller than that in roots at the OTU level (Spearman correlation: r= 0.271, P<0.05). In Fig. 4b, Z scores of the soil DNA and RNA samples at the heading stage were highest, but there was no signi cant differences between Z scores of the soil DNA and RNA samples at tillering and mature stages, and Z scores of the root DNA and RNA samples at heading and mature stages.
The abundance changes of diazotrophs at the RNA and DNA levels were calculated in the rhizosphere soil and rice roots (Additional le 21: Figure S8a-b). Burkholderia, Anaeromyxobacter, Bradyrhizobium and Pelobacter predominated in the total and active rhizosphere diazotrophic communities. NifH gene expressions in Burkholderia and Anaeromyxobacter were somewhat higher than that of the total rhizosphere diazotrophic community, whereas nifH gene expressions in Bradyrhizobium and Pelobacter were opposite. Burkholderia, Azoarcus, Thiorhodospira and Bradyrhizobium predominated in the total and active endophytic diazotrophic communities. Growth period and fertilization appreciably affected the nifH gene expressions of endophytic diazotrophs. For instance, the abundance changes of Azoarcus at the RNA and DNA levels at the heading and mature stage were the opposite. The abundance changes of Burkholderia, Thiorhodospira and Bradyrhizobium at the RNA and DNA levels of CK and the other three fertilization managements were the opposite. Due to high between-sample variations, which might be caused by environmental differences, there was a few or no statistically signi cant difference of diazotrophic abundance at the RNA and DNA level in the soil or roots.

Root-enriched nitrogen-xing bacteria
To investigate the distribution of nitrogen-xing bacteria in rice roots and rhizosphere soil, the relative abundances of 39 dominant genera in these two niches were compared by the STAMP software. The size of the bubbles in the chart (Fig. 5) represents the relative abundance of 39 dominant genera in rice root samples, and the color of the bubbles represents the difference between the square root of the relative abundances of these genera in the roots and soil.
At the DNA and RNA levels, the relative abundances of Thiorhodospira, Rhizobium and Burkholderia in all root samples at three growth periods were almost signi cantly higher than that in corresponding soil samples (P<0.05); the relative abundances of Paludibacter, Magnetospirillum, Clostridium and Rhodospirillum in root samples at the heading and mature stages were signi cantly higher than that in corresponding soil samples; the relative abundances of Sulfurospirillum at the heading stage, Azoarcus and Frankia at the mature stage in root samples were signi cantly higher than that in corresponding soil samples. However, Bradyrhizobium, Pelobacter, Heliobacterium, Geobacter and Anaeromyxobacter had signi cantly higher relative abundance in the rhizosphere soil than that in rice roots.
Co-occurrence patterns of total and active root-associated diazotrophs Co-occurrence networks were built to explore potential interactions and niche-sharing of diazotrophs in paddy soil and rice roots during three key growth periods (Fig. 6, Additional le 22: Figure S9). In each of these networks, nodes represent OTUs (relative abundance>0.01%) and edges represent signi cant co-occurrence relationships (Spearman |r|>0.65 and P<0.01). Topological properties of co-occurring network were calculated (Additional le 23: Table. S14). Overall, the ecological networks of rhizosphere and endophytic diazotrophs had different network sizes and degree of connectivity, and showed distinct time related successions at three key rice growth periods. Consistent with the α-diversity analyses (Additional le 4: Compared with networks at the tillering stage, the rhizosphere and endophytic diazotrophic assemblages formed larger and more complex networks at the heading and mature stages. Networks of total rhizosphere and endophytic diazotrophs at the heading and mature periods exhibited similar network topological structures (nodes, links, average path distance, average clustering coe cient, average degree and modularity) in 2016. However, for active diazotrophs, soil RNA network at the mature stage was most complex, and root RNA network at the heading stage was more intricate. The percent of positive correlations in total rhizosphere diazotrophic networks (57.31%±3.63%) was much less than that in total endophytic diazotrophic networks (83.05%±7.52%), active rhizosphere diazotrophic networks (88.81%±9.22%) total endophytic diazotrophic networks (88.43%±15.64%).

Discussion
Rice is the dominant crop in the middle and lower reaches of the Yangtze River Basin in China and produces more than 65% rice yield of the total country. Rice productivity is usually limited by nitrogen [28]. Nitrogenous fertilizer and BNF to a large extent raise rice productivity and make up for the depletion of soil nitrogen nutrient stocks caused by crop uptake [26]. Fertilization management [29][30][31] and growth period [32][33] varied the diazotrophic community compositions in the agricultural ecosystem. However, the combined effects of fertilization and growth period on diazotrophic communities in the paddy eld still remain largely unknown. In this study, we investigated the combined effects of fertilization and growth period on the community composition of rhizosphere and endophytic diazotrophs in the paddy ecosystem at the DNA and RNA levels using amplicon sequencing. The strength of this experiment is that revealing niche differentiation of diazotrophs and including the effects of fertilization and temporal changes into the experimental design.

Different responses of rhizosphere and endophytic diazotrophs
Rice root exudates induce a chemotactic response for rhizosphere [18] and endophytic bacteria [34], and defense responses of plants limit bacterial invasion [35]. Thus, speci c bacteria can penetrate root tissues and colonize in plants [36]. The OTU richness indices of endophytic diazotrophs were more variable than rhizosphere diazotrophs, as depicted by the rarefaction curves (Additional le 5: Figure S1). It is probably caused by sporadic and non-uniform colonization of endophytic diazotrophs in the roots [37]. The community composition of rhizosphere and endophytic microbiome were different in various plants [36,[38][39][40]. In this present study, the community compositions of rice rhizosphere diazotrophs were signi cantly different from rice endophytic diazotrophs at the DNA and RNA levels (Additional le 9: Table S7), suggesting that rice roots represent a unique niche for diazotrophic communities. Phylogenetic analysis showed that Alphaproteobacteria (25.29%) was the dominant class and Bradyrhizobium (24.8%) was the dominant genus in rice rhizosphere soil. Similar observations were reported in rice bulk soil [16] [41]. For root samples, Betaproteobacteria (34.95%) was the dominant class in rice roots [42][43]. The previous study has reported endophytic diazotrophs were dominated by Gammaproteobacteria (66-98%) before ooding, whereas after ooding Betaproteobacteria was the dominant class (26-34%) [44].
The rhizosphere diazotrophic communities were more diverse than the endophytic diazotrophic communities at the DNA and RNA levels (Additional le 4: Table S4) [36,39], and the community structures of rhizosphere diazotrophs were notably different from endophytic diazotrophs (Additional le 9: Table S7) [40,45]. Moreover, compare with year and fertilization management, growth period (temporal variations) was the dominant factor affecting the community composition of rhizosphere and endophytic diazotrophs ( Table 2, Additional le 20: Table S13). Previous studies had reported that the diversity of free-living diazotrophic and total bacterial populations were more sensitive to temporal and seasonal effects than fertilization management in agricultural soil [46]. Notably, growth period had a higher signi cant effect of the endophytic diazotrophic communities than that of the rhizosphere diazotrophic communities, but fertilization management had a stronger effect on the community composition of rhizosphere diazotrophs than endophytic diazotrophs. Taken together, growth period, year, and fertilization affect soil and root diazotrophic communities differently.
We also found some root-enriched nitrogen-xing bacteria whose abundances in roots were signi cantly higher than that in soil (Fig. 5). Thiorhodospira, Rhizobium and Burkholderia were root-enriched diazotrophs at three growth periods, and Paludibacter, Magnetospirillum, Clostridium and Rhodospirillum were root-enriched diazotrophs at the heading and mature stages. They occupied high relative abundance in rice roots at the DNA (46.29%±14.93%) and RNA (47.77% ±13.01%) levels. Burkholderia and Rhizobium were common plant endophytic diazotrophs [47][48][49], and some strains of these genera have been used as plant growth-promoting bacteria (PGPB) and biocontrol bacteria [50][51][52]. Besides, Burkholderia sp. BV6, Burkholderia sp. CJ42 and Rhizobium sp. CJ54 were isolated from surface-sterilized rice roots in this work (Additional le 2: Table S2). Sulfurospirillum is a reductively dechlorinating bacterium [53]. Thiorhodospira is an purple sulfur bacterium, but its nitrogenase activity has not been tested [54]. It is widely admitted that endophytic diazotrophs are capable to x N more e ciently than the rhizosphere or rhizoplane diazotrophs [55], so we speculate that these root-enriched diazotrophs play an important role for rice growth and development in the whole life cycle of rice.
Shifts of the total community composition of rhizosphere and endophytic diazotrophs during rice growth Previous work showed that temporal variations in the abundance and diazotrophic community composition, but these studies were carried out either with low resolution detection methods or ignoring endophytic diazotrophs [32][33][56][57]. We studied the structural variability in the root-associated diazotrophs during a rice growing season. Our research demonstrated that the community structures of total endophytic diazotrophs varied dramatically during the vegetative stages and stabilized at the reproductive stage (Fig. 3b), which was consistent with the results of rice root microbiota by 16S rRNA gene amplicons sequencing [21][22]. Additionally, the richness, evenness and PD indices of total endophytic diazotrophs signi cantly increased at the vegetative stage, but the α-diversity indices of total rhizosphere diazotrophs had no signi cant changes (Additional le 4: Table S4). Thus, we deduced that abundant diazotrophs recruited from the paddy soil colonize and reproduce in rice roots at the vegetative stage, which caused large changes in diazotrophic community structure. We see support for this idea as the proportion of nitrogen-xing bacteria shared by soil and root samples increased clearly over time, and the percentage of unclassi ed diazotrophs in roots was close to that in soil (Fig.   3e). Considering that root exudation increased with rice growth until panicle initiation and decreased from the owering period to the maturing stage [58], the community succession of rice endophytic diazotrophs is probably driven by root exudation. However, further studies are needed to investigate the regulation mechanism of root exudates on the endophytic diazotrophic community. Moreover, when applying nitrogen-xing bacteria as bio-fertilizers, it should be taken into account whether they could colonize e ciently in the roots at a particular time.
Pig manure application maintained the community composition of natural rhizosphere diazotrophs Nitrogen de ciency promotes root growth, but decreases shoot biomass, shoot and root nitrogen content [59][60].
Furthermore, the e ciency of BNF is promoted with low N concentration [61], but rapidly reduced or even inhibited under a high concentration of ammonium [62]. The 15 N isotope dilution experiment veri ed that low nitrogen nutrition promoted elongation of rice roots, and rice inoculated with a mixture of endophytic diazotrophs obtain a higher level of BNF contribution with a low N content (Table 1).
Previous researches showed that applying chemical or organic fertilizers decreased the diversity of dominant diazotrophs [31], but we found that fertilization management was inconsequential to the α-diversity indices of total diazotrophs in rhizosphere soil and rice roots (Additional le 6: Table S5, Additional le 7: Table S6), which was consistent with Wakelin and Ogilvie's observations [63][64]. According to the correlated analysis, AN and TP had signi cantly negative correlation with α-diversity indices of total rhizosphere diazotrophs (Additional le 17: Table S11).
Given that the β-diversity of rhizosphere diazotrophs were more affected by fertilization than that of endophytic diazotrophs ( Table 2, Additional le 20: Table S13), we focused on the effect of fertilization on rhizosphere diazotrophs.
Compared with NPK and NPK + rice straw (NPKS), the rhizosphere diazotrophic community of NPK + pig manure (NPKM) and control (CK) had higher similarities at the DNA and RNA levels (Additional le 13: Table S8). Thus, the application of pig manure not only increased soil nutrients (OM, TN, TP, AN, AP and NO3-N) (Additional le 15: Table S9) [65], but reduced the effect of chemical fertilization on the natural rhizosphere diazotrophic community structure.
Active rhizosphere diazotrophic communities are less diversity but more divergent than total rhizosphere diazotrophic communities We used RNA-derived nifH gene Illumina sequencing to elucidate the accurate diazotrophic community structure and characterize the nifH expression pro le of rice root-associated diazotrophs. The α-diversity indices of rhizosphere and endophytic diazotrophs in the DNA samples were less than their corresponding RNA samples (Additional le 4: Table S4).
The community structures of total and active diazotrophs in soil or roots were had evident deviations [66], and the community structures of active rhizosphere diazotrophs were more volatile than that of total rhizosphere diazotrophs (Fig.  4a, Additional le 10: Figure S3e). This is consistent with previous observations that seasonal variation and management practices had more effect on the active diazotrophs than total diazotrophs in bulk and rhizosphere soil [29,[67][68]. Our results suggest that AN, TP, NH4-N and NO3-N had signi cant negative correlation with α-diversity indices of active rhizosphere diazotrophs (Additional le 17: Table S11). Our work extend previous studies by speci cally characterizing the distribution and transcriptional activity of active root-associated diazotrophs in the paddy eld.

The temporal dynamics of root-associated diazotrophic networks
Habitat heterogeneity in uenced the bacterial [69] and fungal [70] network interactions. In our work, rhizosphere diazotrophs formed larger and more complex networks than endophytic diazotrophs at the DNA and RNA levels. The strong host-speci c screening effect is imposed by plants on soil microbes, resulting in that only speci c microorganism can colonize roots [71]. Previous alpha diversity analyses showed that the richness and evenness indices of rhizosphere diazotrophs were higher than endophytic diazotrophs (Additional le 4: Table S4), meaning number of OTUs in soil was greater than that in roots. The percent of positive correlations in total rhizosphere diazotrophic networks were lowest in all networks, meaning that total rhizosphere diazotrophs had more competitive relationships than total endophytic diazotrophs, active rhizosphere diazotrophs and endophytic diazotrophs (Fig. 6).

Conclusions
This study provides a comprehensive understanding of the combined effects of growth periods and fertilization management on the diversity, composition and co-occurrence network of root-associated diazotrophs in the paddy eld during the whole growth period. We proved that the structural variability of rhizosphere diazotrophs is much lower than that of the endophytic diazotrophs, and growth period and fertilization management are undeniably important factors affecting nitrogen-xing bacteria community. Speci c diazotrophs appear to be recruited from the paddy soil to colonize and reproduce in rice roots during rice growth. Furthermore, the community structures of total endophytic diazotrophs were more stabilized in the reproductive stage than the vegetative stage. Follow-up studies need to investigate the effects of plant age, developmental progression, climatic variations and cropping management on diazotrophs in the paddy system, reveal the colonization process of diazotrophs, and further deepen our knowledge of root-rhizosphere soildiazotrophs interactions.

Hydroponic experiment
Fresh rice roots were sampled from the experimental eld, and surface-sterilized with 75% ethanol for 1 min and 1% NaClO for 4 min. Then the roots were ground with a mortar, and made into the rice root grinding uid. Oryza sativa seeds were sterilized with 5% sodium hypochlorite for 1 h and then germinated. After 16 days, roots of rice seedlings were sampled and surface-sterilized, then rice root DNA was extracted. The endophytic diazotrophs was undetected using polF/polR primers [72]. The remaining 36 rice seedling (three-leaf stage) were divided into three groups. Thereinto, the T 10 and T 20 treatments were inoculated by soaking roots in the rice root grinding uid for 12 h at room temperature, and the excess inoculum was washed off with tap water. Seedlings of the control group (C 20 ) were soaked using autoclaved root grinding uid. Next, all rice seedlings were transferred into the 5L-beakers and each beaker had two seedlings. They were cultured in the Yoshida's culture solution [73] with some modi cation. The concentration of 15 N-labeled (NH 4 ) 2 SO 4 (50 atom% excess) of C 20 and T 20 treatments was 20 ppm, while of T 10 was 10 ppm. The nutrient solution was changed every ve days and pH remained 5.5-6.5 throughout the experiment.
After planting for 62 days, fresh weight, root length and plant height were measured. Destructive sampling of the whole plant was undertaken for determination of total N content and 15 N concentration. The 12 seedlings were divided into three groups, and every group was dried in forced ventilation oven at 60℃ for 6-8 h, grinded and sieved (0.25 mm). Total N of the dry matter was determined using Kjeldahl digestion. 15 N concentration was measured by Finnigan-MAT-251 mass spectrometer (Bremen, Germany) in the institute of Soil Science, Chinese Academy of Sciences, and the precision is ± 0.20‰. The percentage of N derived from the atmosphere by BNF (% Ndfa) was calculated according to the equation: % Ndfa = [1− R t / R c ] ×100. R t is the 15 N/ total N ratio of the treatment, and R s is the 15 N/ total N ratio of control.
Endophytic diazotrophs isolated from rice roots Ten-fold serial dilutions of the rice root grinding uid mentioned above were plated on YMA or Döbereiner's N-free media.
After 3-5 d of incubation, isolates with different morphology were transferred and streaked on corresponding agar plates. the protocol. Their 16S rRNA and nifH gene were ampli ed by PCR using 27F/1492R and polF/polR primers. PCR products were ligated into pMD19-T-Simple Vector (TaKaRa, Dalian, China), sequenced, and identi ed their closest phylogenetic relatives [74]. Nitrogen xation activity was measured using the modi ed ARA [75]. The isolates were cultivated in semisolid Döbereiner's media in sterilized test tubes, and a nal concentration of 1% acetylene was added after growth for 2 days at 30°C. The concentration of ethylene was detected 24 h later. Non-inoculated test tubes were used as negative controls and every isolate had three replicates. Ethylene was measured with a gas chromatograph (Shimadzu GC-2010, Kyoto, Japan) equipped with a Porapak N packed column and a ame ionization detector (FID). The injector, detector and oven temperatures were 180°C, 220°C and 100°C, respectively. The carrier gas was N 2 (highest available purity) at a ow rate of 10 ml/min, and the supply of H 2 and air for the FID were 30 and 300 ml/min, respectively. The retention time of ethylene was 5.23 min, and the retention time of acetylene was 7.15 min. Isolates were cultured in YMA medium with or without 0.5 g/l Trp for 24 h. After centrifugation, the IAA production of the supernatant was determined by Salkowski reagent (12 g FeCl 3 per liter in 7.9 M H 2 SO 4 ) [76]. CAS solution was used for the quantitation of siderophore production of supernatants [77]. Isolations were inoculated on the solid National Botanical roots, the rhizosphere soil was thoroughly homogenized and sieved (2 mm). Then the soil was divided into two parts: part one was stored at -80℃ for DNA and RNA extraction; part two was air-dried and sent to the Qiyang Red Soil Experimental Station (26°45′N, 111°52′E, Hunan Province, China) for soil physical and chemical analysis [78]. Soil nitrate-nitrogen (NO 3 -N) and ammonium-nitrogen (NH 4 -N) were analyzed using a continuous ow analytical system [79]. Before DNA and RNA extraction, roots were sterilized with 75% ethanol for 1 min and 1% NaClO for 4 min. The last washing solution was coated in TSA solid medium to detect whether the roots were completely disinfected. The primers nifH1 (5'-barcode-TGYGAYCCNAARGCNGA-3') and nifH2 (5'-ADNGCCATCATYTCNCC-3') were used to amplify a 359-bp sequence of nifH gene [80]. Barcode is an eight-base sequence unique to each sample. PCR reactions were performed in triplicate to minimize PCR bias which contained 4 μL of 5×FastPfu Buffer, 2 μL of 2.5 mM dNTPs, 0.8 μL of each primer (5 μM), 0.4 μL of FastPfu Polymerase, and 10 ng of template DNA or complementary DNA [81]. Puri ed PCR products were pooled in equimolar and paired-end sequenced (2×250 bp) on the Illumina MiSeq platform (Biozeron, Shanghai, China) according to the standard protocols.

Sequence processing and statistical analyses
Raw fastq les were quality-ltered with the following criteria: (1) Bases of each read whose quality scores<20 were discarded. Only the reads with perfectly matched barcodes, primers<2 nucleotides mismatch and containing no ambiguous character were kept. (2) Then the barcodes and primers were deleted. (3) The remaining forward and reverse reads with at least 10-bp overlap were combined into a single sequence using FLASH. Combined sequences of less than 250-bp were discarded. The quanti ed reads were translated into protein sequences and corrected potential frameshifts using FrameBot program ( http://fungene.cme.msu.edu/FunGenePipeline/) and the corresponding FunGene database (http://fungene.cme.msu.edu/) (min length=100 amino acids, hmm =50%) as a reference [43,[82][83]. Reads having inframe stop codon (s) were manually removed. Then the protein reads were removed singleton, dereplicated and clustered into OTUs using the UPARSE pipeline at 90% identity [84][85][86]. Chimeric sequences were then identi ed and removed using UPARSE in de novo mode. Finally, OTUs that contained one sequence were removed. To assign putative taxonomy, representative sequences were aligned to a closest-match sequence of the FunGene database by BLASTp [87][88].
Chao1, Shannon and PD indices were calculated by Mothur software after sequences were normalized. Rarefaction curves, bar plots, cluster trees, heatmaps and bubble plots were drawn using R (Version 3.4.0). The Mental tests, PERMANOVA and ANOSIM analysis were conducted using the mental(), adonis() and anosim() functions from the Vegan package in R, respectively. NMDS was conducted on the Bray-Curtis distance matrix using the vegan R package.
Spearman correlation analysis between diazotrophic taxa and soil physicochemical indices was performed using the corrplot library in R. To identify the phylogenetic a liation of nifH sequences, neighbor-joining (NJ) trees were constructed using the molecular evolutionary genetics analysis (MEGA) software for the 50 most abundant OTUs together with selected reference sequences. To track the dynamics of individual OTUs among different samples, Sankey plots were constructed based on D3.js (v.5.14.2) (d3js.org) [89]. Bacterial taxonomic ow at the class level was drawn based on the OTU table of all samples, and for OTUs shared by soil and root samples, the OTU table was further ltered to OTUs whose abundance is less than 0.05%. The extended error bar plots were carried by STAMP software (Statistical Analysis of Metagenomic Pro les; https://beikolab.cs.dal.ca/software/STAMP) to lter fertilization-sensitive and root-enriched biomarkers. The online LEfSe program (http://huttenhower.sph.harvard.edu/galaxy/root?tool_id=PICRUSt_normalize) was performed to nd biomarker whose relative abundance signi cantly changed during rice growth. The co-occurrence network was constructed with the 'psych' package by using the Spearman correlation, and the correlations with a