Plant Developmental Stage Drives the Differentiation in Ecological Role of Plant Bacterial and Fungal Microbiomes

Plants live with diverse microbial communities which profoundly affect multiple facets of host performance such as nutrition acquisition, disease suppression and productivity, but if and how host development impacts the assembly, functions and microbial interactions of crop microbiomes are poorly understood. Here we examined both bacterial and fungal communities across soils (rhizosphere and bulk soil), plant epiphytic and endophytic niches (phylloplane, rhizoplane, leaf and root endosphere), and plastic leaf of fake plant (representing environment-originating microbes) at three developmental stages of maize at two contrasting sites, and further explored the potential function of phylloplane microbiomes based on metagenomics. Our results suggested that plant developmental stage had a much stronger inuence on the microbial diversity, composition and interkingdom networks in plant compartment niches than in soils, with the strongest effect in the phylloplane. Air (represented by fake plants) was an important source of phylloplane microbiomes which were co-shaped by both plant development and seasonal environmental factors. Further, we demonstrated that bacterial and fungal communities in plant compartment niches exhibited contrasting response to host developmental stages, with higher alpha diversity and stronger deterministic assembly within bacterial microbiomes at the early stage but a similar pattern within mycobiomes at the late stage. Moreover, we found that bacterial taxa played a more important role in microbial interkingdom network and crop yield prediction at the early stage, while fungal taxa did so at the late stage. Metagenomic analyses further indicated that phylloplane microbiomes possessed higher functional diversity and functional genes involved in nutrient provision and disease resistance at the early stage than the late stage.


Abstract Background
Plants live with diverse microbial communities which profoundly affect multiple facets of host performance such as nutrition acquisition, disease suppression and productivity, but if and how host development impacts the assembly, functions and microbial interactions of crop microbiomes are poorly understood. Here we examined both bacterial and fungal communities across soils (rhizosphere and bulk soil), plant epiphytic and endophytic niches (phylloplane, rhizoplane, leaf and root endosphere), and plastic leaf of fake plant (representing environment-originating microbes) at three developmental stages of maize at two contrasting sites, and further explored the potential function of phylloplane microbiomes based on metagenomics.

Results
Our results suggested that plant developmental stage had a much stronger in uence on the microbial diversity, composition and interkingdom networks in plant compartment niches than in soils, with the strongest effect in the phylloplane. Air (represented by fake plants) was an important source of phylloplane microbiomes which were co-shaped by both plant development and seasonal environmental factors. Further, we demonstrated that bacterial and fungal communities in plant compartment niches exhibited contrasting response to host developmental stages, with higher alpha diversity and stronger deterministic assembly within bacterial microbiomes at the early stage but a similar pattern within mycobiomes at the late stage. Moreover, we found that bacterial taxa played a more important role in microbial interkingdom network and crop yield prediction at the early stage, while fungal taxa did so at the late stage. Metagenomic analyses further indicated that phylloplane microbiomes possessed higher functional diversity and functional genes involved in nutrient provision and disease resistance at the early stage than the late stage.

Conclusions
Our results suggest that host developmental stage profoundly in uences plant microbiome assembly and functions, and the bacterial and fungal microbiomes take a differentiated ecological role at different plant development stages. This study provides empirical evidence for host exerting strong effect on plant microbiomes by deterministic selection to meet the physiological requirement of plant developmental stages. These ndings have implications for the development of future tools to manipulate microbiome for sustainable increase in primary productivity.

Background
Plants live with large and diverse prokaryotes and eukaryotes (i.e. plant microbiomes) which have coevolved with their hosts and profoundly impact a range of aspects of plant performance [1][2][3][4]. For example, some bene cial bacteria and fungi like nitrogen xing, antagonistic bacteria and mycorrhizal fungi in the rhizosphere and plant compartments, can deeply in uence plant growth and health via promoting nutrient acquisition, protecting against pathogen attacks, and increasing tolerance to environmental stress [5][6][7][8]. Recent studies suggested that plant microbiome assembly and host health are largely in uenced by complex and dynamic interactions between the host, microbes, and the environment, but the ecological processes that govern plant-microbiome-environment interactions remain poorly understood [3,9]. A better understanding of the mechanisms and temporal dynamics of plant microbiome assembly, functions and co-occurrence networks is of signi cant importance for the development of microbiome-based solutions for sustainable crop production systems [10][11][12].
Assembly of plant microbiomes starts soon after sowing and develops with plant growth under the in uence of deterministic (e.g. selection mediated by biotic and abiotic factors) and stochastic (e.g. random dispersal and drift events) processes [2,8,13]. It has been suggested that diverse microbial taxa can colonize different plant compartment niches through dispersal from soil, seed and air initially, and then form a dynamic community under the integrative effects of multiple host and environmental factors [4,8,[14][15][16]. On the one hand, the plant host has strong selection effects on its microbiomes via host immune system, genetic networks and plant exudates that vary among different compartment niches and host species [13,[17][18][19][20]. Further, multiple environmental factors such as climate, edaphic properties (e.g. soil pH and nutrients) and human perturbations (e.g. agricultural management regimes), also play important roles in driving plant microbiome assembly [21][22][23][24][25]. It has been reported that plant microbiomes were mainly determined by compartment niche and host species at the plant level, with the phylloplane and rhizoplane acting as an important interface between the host and the environment [26][27][28][29][30]. Some recent studies also highlighted the signi cant contribution of plant developmental stages on plant microbiome assembly [31][32][33]. For example, it was suggested that rice root microbiota varied dramatically during the vegetative stages but stabilized after reproduction [33]. This is expected given that plant physiological requirement and composition of plant exudates vary with its developmental stages [9,13,34,35]. However, we lack empirical evidence for underpinning the mechanisms of microbiome assembly across soil, plant epiphytic and endophytic niches along with plant developmental stage in eld under the interactive in uence of climate, edaphic factors and fertilization regimes.
In addition to host and environmental factors, the assembly and stability of the plant microbiome were also strongly affected by microbe-microbe interactions [2,36,37]. Potential microbial interactions across different habitats can be characterized using microbial co-occurrence network analysis [37][38][39]. Previous studies have reported that the network hubs (hub taxa) which frequently interact with other taxa in microbial co-occurrence networks, may act as mediators and gatekeepers for microbial communities and play a crucial role in plant microbiome development, host nutrient acquirement and tness [13,37,40,41]. Accordingly, a recent study on Arabidopsis root-microbiota showed that microbial interkingdom interactions among bacteria, fungi and oomycetes, could greatly promote Arabidopsis survival in which bacteria played a vital role in protecting plants against pathogens and maintaining microbial interkingdom balance for plant health [41]. The latter suggests the microbial interactions are closely associated with plant tness. Knowledge is lacking however on bacterial-fungal interactions along the soil-plant continuum and how these respond to changes in plant developmental stages, and to what extent these complex interkingdom interactions affect the microbiome dynamics and host performance.
In this study, maize grown in two main agricultural production areas with contrasting soil type and climate condition in China, and received different fertilization practices was used as a study model. The dynamics of both bacterial and fungal communities were investigated over three developmental stages (seedling, tasseling and mature stages) across 432 samples from soils (bulk soil and rhizosphere), multiple plant compartment niches (rhizoplane, root endosphere, phylloplane, leaf endosphere, and grain), and plastic leaf (represented local climate and air environment). Speci cally, our aims were to (1) uncover the interactive effect of plant development and environmental factors such as soil type, climate and fertilization practice on microbiome assembly along the soil-plant continuum; (2) identify the interkingdom network patterns and ecological role of bacterial and fungal communities across plant developmental stages. We hypothesized that (1) plant development would dominate over environmental factors in shaping plant microbiomes and vice reverse for soil microbiome, and the phylloplane microbiome would be more dynamic as a result of dual effect from host selection and seasonal environmental factors; (2) The microbial interkingdom network patterns and plant microbiome functions would alter as per the physiological requirement of plant developmental stage, and bacteria and fungi may alternatively function at different stages.
Sample collections in two sites were performed at June, August and September of 2017, corresponding to maize seedling, tasseling and mature stage, respectively. For each time point, leaf, root, rhizosphere soil, and bulk soil samples were collected from each plot following our previous method [29,42]. The plastic leaf samples were also collected from each time point and maize grain samples were collected at the mature stage. As the study mainly aimed at the temporal dynamics of soil and crop microbiomes, we focused on six compartments (bulk soil, rhizosphere, rhizoplane, root endosphere, phylloplane and leaf endosphere) in three treatments (Control, 80%N, and 80%NS). We also included phylloplane samples from other four treatments (N, 80%NI, 80%NKle, 80%NSB), as our previous studies suggested that the phylloplane niche was a hotspot of plant-microbes-environment interactions and affected by both host and environments [29]. In total, we collected 432 samples for microbial community analysis. More information on fertilization treatment, eld management and sampling is provided in supplementary materials "Method S1" and our previous publications [29,42].
All samples for molecular work were transported to the lab on dry ice and stored at -80 °C until further processing. Soil physicochemical characteristics (e.g. pH, NH 4 + -N and NO 3 --N) and enzyme activities related with C, N, and P cycling (e.g. nitrogenase activity and phosphatase) were measured according to previous protocols [42][43][44].

DNA extraction
The rhizosphere and bulk soil DNA were extracted from 0.4 g soil using the PowerSoil DNA Isolation Kit (MO BIO Laboratories, Carlsbad, CA, USA) according to manufacturer's instructions. The epiphytic and endophytic microbial cells from the leaves (10-15 g) and roots (3-5 g) were collected following our previous method [29], and subjected to DNA extraction using the PowerSoil DNA Isolation Kit. The epiphytic DNA from the plastic leaves (10-15g) was obtained using the same method for the maize leaf epiphytic DNA collection. For maize grain, ~ 5 g sample was ground using sterile mortars and pestles with liquid nitrogen and then DNA was extracted from the 0.4 g resulting powder using the PowerSoil DNA Isolation Kit.

Amplicon sequencing and bioinformatic analysis
Bacterial 16S rRNA gene V5-V6 region was ampli ed using primers 799F and 1115R [45], and fungal ITS2 region was ampli ed using primers fITS7 [46] and ITS4 [47]. Sequencing was performed on the Illumina MiSeq platform with a Paired-End protocol. The raw sequences were quality-ltered using USEARCH (v10.0) [48] as previously described [29,42], and all correct biological reads (i.e. zero-radius operational taxonomic units, ZOTUs) were picked at 100% similarity using unoise3 command [49] with default parameters. Bacterial and fungal sequences were classi ed using SILVA (v13.2) and UNITE (v8.0) databases, respectively. Bacterial ZOTUs assigned to chloroplast, mitochondrial and viridiplantae, and fungal ZOTUs assigned to plant or protist were removed. Both bacterial and fungal ZOTUs represented by less than 2 sequences were removed to avoid possible biases. In total, 11,665,748 bacterial and 23,156,931 fungal high quality reads from 432 samples were retrieved and sorted into 18,602 bacterial and 9,299 fungal ZOTUs (i.e. phylotypes; analogous to amplicon sequence variants). Both bacterial and fungal alpha-and beta-diversity were calculated in QIIME [50]. Bacterial and fungal ZOTU tables were then rare ed to 3130 and 33000 reads for the alpha diversity estimates, respectively. For beta diversity analyses, ZOTU tables were normalized using the cumulative-sum scaling (CSS) method [51].

Metagenomic sequencing and data mining
To further characterize phylloplane microbiome functions, we selected 9 maize phylloplane (based on the N treatment, 3 replicates × 3 stages) and 9 plastic leaf (3 replicates × 3 stages) DNA samples from the site QJ for metagenomic sequencing using the Illumina NovaSeq platform with a Paired-End protocol.
Raw sequences were quality-ltered using Trimmomatic (v0.39) [52], and sequences belonging to the maize genome were removed by mapping the data to the maize reference genome (NCBI: NC_050096.1) with Bowtie2 (v2.1.0) [53]. Finally, an average of 8.6 Gb of clean data was retrieved for each sample.
Microbial interkingdom network analysis at bacterial and fungal genus level was performed using the CoNet [62] in Cytoscape (v3.5) [63] based on Spearman correlation scores (Spearman's r > 0.7 or r < −0.7; P < 0.01). Both bacterial and fungal genera present in at least 10 samples were retained for the network analysis [42]. The networks were visualized in Gephi [64]. Bacterial functional pro les were predicted using functional annotation of prokaryotic taxa (FAPROTAX) [65], and fungal functional groups were predicted using the program FUNGuild [66]. The Source Model of Plant Microbiome (SMPM) was estimated using SourceTracker (v1.0) [67]. Differential abundance analysis was performed using EdgeR's generalized linear model (GLM) approach [68]. Random forest analysis was conducted using the "randomForest" R package [69]. Mantel test was performed to explore the correlations between microbial communities, soil physicochemical characteristics, and soil enzyme activities using the "vegan" package [42]. All statistical analyses were carried out in R (http://www.r-project.org). Nonparametric statistical test (Kruskal-Wallis test or Wilcoxon test) was performed to evaluate the alpha-diversity difference and the taxonomical difference among different niches and stages. More information on the network analysis, random forest modeling analysis and source tracking analysis are detailed in our previous publication [29,42].

Results
Diversity and community assembly of bacterial and fungal microbiomes across three plant developmental stages The linear mixed model analysis suggested that plant developmental stages had a greater in uence on both bacterial and fungal Chao1 richness in plant compartment niches (phylloplane, leaf endosphere, rhizoplane and root endosphere) than those in the rhizosphere and bulk soils (Table S1). For plant compartment niches, bacterial Chao1 richness decreased from the seedling stage to the mature stage, while fungi showed opposite pattern ( Fig. 1a and Fig. S1a, b).
Null model analysis showed that the relative contribution of deterministic (|βNTI| ≥ 2) and stochastic (| βNTI| < 2) processes in crop microbiome assembly were greatly affected by plant developmental stage, particularly for the phylloplane and leaf endosphere (Fig. 2a, b). At the seedling stage, a higher relative contribution of deterministic processes mainly belonging to heterogeneous selection in plant compartments was observed in bacterial communities (~71%) than in fungal communities (~47%). Conversely, the effects of deterministic processes decreased for bacterial communities (to ~53%) but increased for fungal communities (to ~64%) at the mature stage (Fig. 2a, b). Collectively, deterministic processes exerted a greater in uence on crop bacterial microbiome at the early stage and on mycobiome at the late stage, respectively. By contrast, microbial communities in the rhizosphere (71-81%), bulk soil (61-76%) and plastic leaf (87-100%) were mainly driven by deterministic processes over the time (Fig. 2a,  b).

Temporal dynamics of microbial interkingdom co-occurrence networks
We further performed co-occurrence network analysis to assess the impact of host developmental stage on bacterial-fungal interkingdom interactions along the plant-soil continuum. Our results showed that microbial interkingdom network patterns shifted clearly across three developmental stages, with differentiated bacterial and fungal roles in the networks during plant development (Fig. 3a-e). Speci cally, bacterial taxa had higher network connectivity (i.e. network degree) than fungal taxa at the seedling stage, while the pattern was reverse at the mature stage ( Fig. 3a-d). In contrast, fungal network connectivity increased from 2.2 at the seedling stage to 17.8 at the mature stage ( Fig. 3a-d). Moreover, the proportion of negative network edges (mainly representing bacteria-fungi interkingdom correlations) markedly increased from 6.1% at the seedling stage to 50.5% at the mature stage ( Fig. 3a-c, e). We further de ned the "network hubs" as node with high values of degree (> 50) and closeness centrality ( > 0.3) in the network, and found 3 network hubs (bacteria: 3; fungi: 0) at the seeding stage, and 2 (bacteria: 1; fungi: 1) at the tasseling stage, and 10 (bacteria: 4; fungi: 6) at the mature stage ( Fig. 3d; Table S3). Similar patterns were also recorded in microbial interkingdom functional networks based on function prediction of amplicon sequencing data, with more bacterial functional network hubs (e.g. group "nitrate respiration") at the rst two stages (Fig. S3a-e). In contrast, the highest number of fungal functional network hubs was identi ed at the mature stage and mainly represented by group "Saprotroph" (Fig. S3ad).
As for each niche, microbial interkingdom network patterns in plant compartments changed distinctly across three developmental stages, particularly for the phylloplane and leaf endosphere (Fig. S4a-c). The fungal network connectivity and the proportion of fungal nodes signi cantly increased from the seedling stage to the mature stage in all four plant compartmens (Fig. S4a-c). Conversely, the network patterns in the rhizosphere and bulk soil were relatively stable over the three stages (Fig. S4a-c).

Temporal dynamics of microbiome composition across soil, plant epiphytic and endophytic niches
Taxonomic classi cation showed that both bacterial and fungal communities in plant compartment niches varied distinctly across three developmental stages, but not in the rhizosphere and bulk soils (  6e-8). Notably, we found that Actinobacteria in plant compartments was more abundant at the seedling stage (26.1%) than at the tasseling stage (15.9%) and the mature stage (12.1%, P < 0.001) (Fig. 4a, Fig. S5a). In addition, the relative abundance of Dothideomycetes in the rhizoplane increased from the seedling stage to the mature stage (P < 0.001), while the Dothideomycetes in the root endosphere showed an opposite pattern (Fig. 4a, Fig. S5a). Notable, Dothideomycetes dominated mycobiomes in the plastic leaf and maize phylloplane in both sites and showed no visible variation among three developmental stages (Fig. 4a). Differential abundance analysis at ZOTU level further showed that some members within bacterial families Burkholderiaceae, Microbacteriaceae, Streptomycetaceae and Rhizobiaceae were signi cantly enriched in at least two plant compartments (e.g. phylloplane and rhizoplane) at the seedling stage (Fig. S6a, b; Table S4). Moreover, some members within fungal families Coniothyriaceae, Mycosphaerellaceae, Sporidiobolaceae and Symmetrosporaceae were signi cantly enriched in at least two plant compartments at the mature stage (Fig. S6a, b; Table S4), and some genera within these families were identi ed as hubs in microbial interkingdom network at the mature stage ( Fig. 3a-d; Table S3).
Furthermore, the random forest modeling analysis indicated that bacterial community composition at the seedling stage is a strong predictor for crop yield while fungal community composition at the mature stage did so (Fig. 4b). Mantel test showed that bacterial communities of the rhizosphere and bulk soils in both sites had signi cant correlations with soil N cycling related enzyme activities like nitrogenase and potential nitri cation rate across three developmental stages. In contrast, fungal communities had signi cant correlations with soil enzyme activities related with C and P cycling like β-glucosidase and phosphatase ( Fig. S7a-b; Table S5).
The Source Model of Plant Microbiome (SMPM) showed that maize-associated bacterial and fungal communities were mainly derived from bulk soils and gradually ltered at different plant compartment niches, and the trends were similar across three developmental stages ( Fig. S8a; Table S6). Maize grain potentially selected majority of taxa from leaves (bacteria: 46.5%, fungi: 28.4%) and roots (bacteria: 39.6%, fungi: 58.1%) (Fig. S8a). Remarkably, environment-originated (represented by the plastic leaf) microbiomes were important sources of maize phylloplane microbiomes, with an increasing contribution from 52.6% to 87.2% for bacteria and from 86.6% to 92.4% for fungi across three stages (Fig. S8b).
The functional pro les of phylloplane microbiomes Metagenomic analysis indicated that the functional composition (i.e. NMDS ordinations of KEGG Orthology) of maize phylloplane microbiome signi cantly differed from that of plastic leaf of fake plant (R 2 = 44.3%, P < 0.01), and the developmental stage also had signi cant effect on phylloplane microbiome functions for both maize (R 2 = 50.0%, P < 0.01) and fake plant (R 2 = 84.1%, P < 0.01) (Fig.  5a). Importantly, the maize phylloplane possessed higher microbiome functional diversity (i.e. Chao1 richness based on KO, CAZy and COG) at the seedling stage than at other stages, while the plastic leaf showed opposite pattern (Fig. 5a). Differential abundance analysis showed that the numbers of speci cally enriched functional traits (including KO, CAZy and COG) in maize phylloplane was signi cantly higher at the seedling stage than at any other stages, with functional gene involved in disease resistance (K13457) signi cantly enriched at the seedling stage (Fig. 5b). Moreover, functional genes involved in nitrogen xation (e.g. nifH, nifD, nifK) and nitrate reduction (e.g. nrfA and napA) in maize phylloplane were more abundant at the seedling stage, while nitrous oxide reductase gene (nosZ), N assimilation gene (nasA, nasB) and C degradation (e.g. xylA and amyA) and P transport (e.g. pstA and pstB) related genes were more abundant at the other two stage (Fig. 5c). Some functional attributes related with methyl-accepting chemotaxis (K03436), inorganic ion transport (COG_P) and defense mechanisms (COG_V) were identi ed as the biomarker functions for the maize phylloplane while chitin synthase (K00698) was identi ed for fake plant by LEfSe (Fig. S9a). Although both maize and fake plant shared 39-55% of phylloplane ZOTUs across three stages (Fig. S9b), a stronger depleted effect was observed in the maize phylloplane when compared with fake plant. Some ZOTUs mainly belonging to Alphaproteobacteria and Actinobacteria, like ZOTU7 (Microbacterium) and ZOTU9985 (Sphingomonas), were signi cantly enriched in the maize phylloplane over three stages (Fig. S9b).

Plant development strongly in uences the assembly of plant microbiomes
Uncovering the ecological principles and processes that underpin plant microbiome assembly and developmental dynamics is essential to advance fundamental understanding of co-evolution and future application of the crop microbiome to sustainable increase in farm productivity [9,10,70]. Our results demonstrate that maize microbiome assembly is mainly in uenced by compartment niche and developmental stage regardless of farming regions and fertilization regimes. Further, plant microbiomes are more sensitive to plant developmental stages than soil microbiomes in terms of multiple microbial attributes (i.e. alpha diversity, community structure, assembly processes and interkingdom networks).
These ndings are consistent with previous studies showing that plant compartment is a determining factor shaping the assembly of plant-associated microbiomes [26][27][28][29]42], and that plant seasonal status has signi cant effects on microbiomes in grasses phyllosphere and Arabidopsis rhizosphere [32,71]. Further, metagenomic analysis in our study revealed that the functional diversity and enriched functional traits in the maize phylloplane varied across three stages. There was a stronger depletion effect in the maize phylloplane in comparison to fake plant phylloplane, and functional attributes involved in methylaccepting chemotaxis and defense mechanisms were signi cantly enriched in the maize phylloplane. Together these results indicate that the plant host exerts a strong selection effect to recruit and lter speci c microbial taxa and functions from nearby species pool during plant development [18,72,73].
Complementary to the previous nding that host selection via plant compartment niche and host genetics plays a dominant role in shaping plant microbiomes assembly [20,26,29,31,42], this work provides novel evidence that plant developmental stages profoundly in uence not only plant microbiome assembly but also their functions.
The effects of plant developmental stage represented the dynamic effects of plant metabolism, exudation and immune-associated traits [9,13,74], and plant-associated microbes have strong chemotaxis activities towards plant signal molecules such as organic acids and sugars [34,71,[75][76][77]. For instance, a recent work revealed that wheat root-released organic carbon varied dramatically across wheat growth stages and correlated with different microbial taxa [34]. It was also suggested that plant exudates and volatiles like coumarins, benzoxazinoids and triterpenes play key roles in shaping plant microbiomes during host development [17,72,76]. The effect of plant developmental stages on microbiome in this study included the effects from season-dependent environmental factors like air, dust and climate. Our results showed that both maize and fake plant phylloplane microbiomes had similar temporal patterns and shared more than one third of ZOTUs at each stage. Further, fake plant phylloplane microbiome contributed an increasing proportion as the source of the maize phylloplane microbiome over the time, indicating a part contribution of environmental temporal factors to crop microbiome assembly. These results presented strong eld evidence showing that local air, dust and rainwater are the main sources of crop microbiome in phyllosphere. These ndings signi cantly advance our knowledge on the source, driving force and potential function of phyllosphere microbiome, and further corroborated that the phylloplane acts as an important interface between the host, microbes, and the environment [30,[78][79][80].
Our results also showed that plant developmental stage had signi cant effects on the rhizosphere and bulk soil microbiomes, though it was much weaker than the site effects, implying that plants also have profound in uence on soil microbiomes via strong rhizosphere effect [6,35,74]. Collectively, by examining the temporal dynamics of bacterial and fungal microbiomes in the soil-plant continuum of maize and fake plants phylloplane in geographically distant sites, this study considerably expanded our knowledge on the succession of plant microbiomes and their potential function under different temporal and spatial scales in eld.
The differentiation in ecological roles of bacterial and fungal microbiomes across plant developmental stages Bacteria and fungi have coevolved with their host for more than 400 million years and greatly contribute to numerous aspects of plant health and productivity [1,8,41]. In this study, bacterial-fungal interkingdom interaction patterns distinctly shifted across three developmental stages. Bacterial community possessed higher alpha diversity and network connectivity at the seedling stage while fungal diversity was higher at the mature stage. Moreover, bacterial and fungal taxa dominated network hubs at the seedling stage and the mature stage, respectively. These suggested that the host can selectively modulate microbial interactions to meet its requirement during plant development, as microbial network hubs were supposed to play crucial roles in maintaining plant health and nutrient [37,40]. In addition, bacterial taxa at the seedling stage were better predictors of crop yield while fungal taxa at the mature stage did so. Metagenomic analysis further corroborated that maize phylloplane microbiome possessed higher functional diversity at the seedling stage than the other two stages. Importantly, more abundant genes associated with disease resistance were enriched at the seedling stage while C degradation and P transportation related genes were enriched at two late stages, which further support the ability of plants to manipulate microbial assembly as per physiological requirements [1,73]. Based on the limited knowledge on the plant microbiome, it has been proposed that the dynamics of plant microbiome composition are a re ection of the current needs of the host plant [3,40,73], and represent the consequence of subtle changes in microbial selection strategy exerted by the host during plant development [1,13,73,81]. Our results therefore support that bacteria may take a more important ecological role in the plant microbiome and host performance at the early stage, while fungi do so at the late stage. This nding is supported by the Null model analysis, which demonstrated the dominant effect of determinism on bacterial community and of stochasticity on fungal community at the seedling stage, but a reverse pattern at the mature stage.
We further found that the negative edges representing bacterial-fungal interkingdom correlations in network increased over the time, implying an increasing competition relationship between bacteria and fungi along plant development stages. It was suggested that microbial competitive interaction could positively in uence microbiome stability [40,82,83]. Our study provides more empirical evidence on this and further support the argument that the host may facilitate host tness and plant-microbiome balance by deterministic host selection during plant development. These ndings provide new insights into complex interactions among the plant, microbes and the environment and provide essential information for the future development of tools to manipulate crop microbiome.

Keystone bacterial and fungal taxa and their ecological functions at different developmental stages
Our results suggest that the composition and potential functions of plant microbiomes change across plant growth and bacterial phyla Actinobacteria and Bacteroidetes were more sensitive in response to plant developmental stage. More abundant Actinobacteria were observed at the seedling stage than at two late stages in plant compartments. Actinobacteria are well known as antagonistic bacteria excreting antibiotic compounds that provide protection against plant pathogens [84][85][86]. For example, a recent work showed that the enrichment of protective microbes like Actinobacteria in the rhizosphere could facilitate disease suppression [85]. Furthermore, some ZOTUs within families Burkholderiaceae, Streptomycetaceae and Rhizobiaceae were signi cantly enriched in plant compartment niches at the seedling stage. The members within Burkholderiaceae and Rhizobiaceae are important diazotrophs and plant growth promoting bacteria (PGPR) [1,6,12], and the members within Streptomycetaceae are wellknown antibiotic-producing bacteria [41,87,88]. Coincidently, metagenomic analysis for the maize phylloplane suggested that functional genes involved in disease resistance and N 2 xation were signi cantly enriched at the seedling stage. In addition, bacterial communities in the rhizosphere and bulk soils showed signi cant correlations with nitrogenase activity across three developmental stages, and the bacterial functional group "nitrite respiration" was identi ed as the network hubs at the seedling stage. These ndings provide evidence for the ecological importance of crop bacterial microbiomes in maintaining plant health and nutrient requirement at the early stage.
For fungal community, classes Sordariomycetes and Dothideomycetes were more sensitive to plant development, with Dothideomycetes increasing over the time in the rhizoplane but decreasing in the root endosphere. Previous works have shown that Sordariomycetes and Dothideomycetes are the most dominant fungal taxa in soils and plant compartments, respectively, and that class Dothideomycetes comprises a highly diverse range of fungi including endophytes, epiphytes and plant pathogens [42,89]. In addition, many members within Dothideomycetes are also identi ed as saprotrophic fungi functioning in wood and leaf-litter decomposition and nutrient cycling [89,90]. Notably, Dothideomycetes predominated phylloplane mycobiomes of both fake plant and maize in two distant study sites across three developmental stages. It was suggested that Dothideomycetes are the dominant fugal taxa of air microbiomes [90]. This indicated that Dothideomycetes in fake plant and maize phylloplanes might be mainly dispersed from air. The reverse dynamics of Dothideomycetes in the rhizoplane and root endosphere indicated multiple ecological roles of Dothideomycetes during host growth, however, its biological functions need further exploration. Furthermore, some fungal ZOTUs a liating within families Coniothyriaceae, Mycosphaerellaceae and Symmetrosporaceae were identi ed as network hubs and signi cantly enriched in plant compartments at the mature stage. Some members of families Coniothyriaceae and Mycosphaerellaceae within Dothideomycetes are important saprobes with celluloseand carbohydrate-degrading ability [90,91]. Coincidently, we found that most network hubs in both taxonomic and functional networks of the mature stage belonged to fungal functional group "Saprotroph". Moreover, fungal communities in the rhizosphere and bulk soils had signi cant correlations with C cycling related enzymes like β-glucosidase across three developmental stages. These results suggest that fungal taxa play key roles in regulating plant C cycles like decomposition of plant residues at the late stage. This indicates that crop mycobiomes may play an increasing ecological role as the decomposers with the aging of the plant, and the host plant may be passively occupied by saprophytic fungi as the consequence of reduced host immunity.
Collectively, our study demonstrates that plant is able to recruit speci c microbial taxa with desire functions to meet their health and nutrient requirements at different developmental stages. However, the molecular mechanisms governing plant-microbiome interactions during host development and the ecological and biological functions of crop microbiomes in facing climate challenge and achieving sustainable agriculture are not fully understood and need further exploration [10,92,93].

Conclusions
By examining the temporal dynamics of bacterial and fungal communities across soils, multiple plant compartments and fake plant phylloplane at two geographically distant sites, this study provides a systematic understanding on the succession of microbiome composition and their potential functions during plant development. Our results demonstrate that plant developmental stage has a much stronger in uence on multiple microbial attributes (i.e. alpha diversity, community structure, determinism/stochasticity patterns and interkingdom networks) in plant compartment niches than in soils, with the strongest effect in the phylloplane. We further found that air is an important source of phylloplane microbiomes, which were strongly co-shaped by plant development and seasonal environmental factors. Furthermore, we demonstrate that the ecological role of bacterial and fungal microbiomes signi cantly shifts with plant development, along which bacteria take a more important role in microbial interkingdom networks and crop yield prediction at the early stage while fungi take an increasing role at the late stage. Metagenomic analyses further revealed that maize phylloplane microbiome possessed higher functional diversity and functional genes involved in disease resistance and nutrient provision at the early stage than at the late stage. In contrast, more abundant N assimilation and C degradation genes were observed in maize phylloplane microbiomes at the late stage, corresponding to an increasing ecological importance of saprophytic fungi in plant microbiomes. Additionally, we found a dominant effect of determinism on bacterial microbiomes at the early stage and on mycobiomes at the late stage in plant compartments. Together these results suggest that the host has a strong selective modulation effect on their microbiomes which is strongly modi ed by the plant developmental stages. These ndings signi cantly advance our fundamental understanding of plantmicrobiome interactions and provide critical new knowledge for future synthetic community research, and the development of microbiome tools to enhance plant protection and agriculture production in a sustainable way.