Microbiome of the Rhizosphere: from Structure to Functions

Microbial composition and functioning in the rhizosphere are among the most fascinating but hidden topics in microbial ecology. We generalized bacterial traits regarding community diversity, composition and functions using published 16s rDNA amplicon sequences of 584 pairs of bulk soils vs rhizosphere of crops. The lower bacterial diversity in the rhizosphere (-7% richness) compared to root-free soil reects the excess of available organic substances near the root. The rhizosphere is enriched by Bacteroidetes, Proteobacteria and Cyanobacteria as well as other copiotrophic bacteria (r strategists). Complex but unstable bacterial networks in rhizosphere reect tight microbial interactions and adaptations to uctuating conditions common for r strategists. The dominant dormancy strategy in the rhizosphere is the toxin-antitoxin system, while sporulation is common in bulk soil. Function prediction analysis showed that the rhizosphere is strongly enriched (50–115%) in methanol oxidation, ureolysis, cellulolysis, chitinolysis and nitrogen xation, but strongly depleted in functions related to N-cycling.


Introduction
The rhizosphere harbors a great abundance of diverse microbes, many of which undoubtedly bene t plants by suppressing pathogenic invasion and assisting in nutrient acquisition from the soil 1, 2 .
Understanding the taxonomic and functional components of the rhizosphere microbiome and their differences from bulk soils (here: soil without direct root effects) are crucial for manipulating them for sustainable crop production. The recent development of next-generation sequencing enabled great progress in deciphering the microbiomes in speci c crop rhizospheres. The diversity and composition of the rhizosphere bacterial community depends on both the plant species and soil properties 3,4,5 .
Although plant species, or even genotypes within species, tend to assemble distinct rhizobacterial communities 6,7,8 , these communities can be highly similar even in different environments 9 . A growing number of studies demonstrate that plants have selective effects on rhizobacterial assemblages in the bulk soil reservoir for acquiring speci c functional traits related to plant tness 8,10,11 . The result is that the rhizosphere microbiome considerably extends the functional repertoire of the plant 12 . Despite the different nutrient requirements, disease suppression mechanisms and edaphic habitats of plants, the rhizosphere conditions (with a surplus of available carbon) provide a similar set of conditions for microbial life. All crops are mineral resource-limited organisms and frequently suffer from pathogen invasions. Accordingly, there could be a generalized selective effect -targeted for nutrient acquisition and pathogen suppression -across all crops regardless of their soil environment or climate. Such potential general patterns for rhizosphere versus bulk soils in terms of the taxonomic and functional pro les of bacterial communities remain largely unexplored. Such information, however, is crucial to help understand and manage microbial functions in agroecosystems in support of enhancing global agriculture in the future.
Currently, amplicon sequence analysis of marker genes, typically 16S rRNA in the case of bacteria, helps to characterize the relative abundance of species in soil 13 , and many studies have been conducted to characterize root-associated microbial communities 9 . This has yielded extensive sequencing data targeting the rhizosphere because almost all the published studies are required to deposit their raw data in public gene banks. These high-resolution nucleic acid-based molecular techniques provide excellent insights into the speci c microbiome members living in soil habitats. Research priorities for harnessing rhizosphere microbiomes for sustainable agriculture include determining the functional mechanisms mediating plant-microbiome interactions and de ning the core microbiome of crops 14,15 . The methodological progress made for these research priorities has provided massive data for an integrative analysis and subsequent synthesis. This opens the door for investigating the general principles of rhizosphere microbiome selection from bulk soils. We therefore tackle the challenge to infer the community compositions in rhizosphere and bulk soils, to determine speci c rhizosphere microbiome properties common for the broad range of plant and environmental conditions, and to bridge the general rhizobacterial assemblages with their functions as related to community-aggregated dormancy potential, heterotrophic strategy, and individual nutrient cycling processes.
Herein, to clarify the general differences, we collected the 16s rRNA amplicon-based sequencing data from all available gene banks, and then synthetically analyzed the data using state-of-the-art bioinformatic methods. Bulk soil serves as a microbial reservoir for the rhizo-microbiome 16 and harbors considerable microbial diversities that are distinctly shaped by the environmental factors of the respective microhabitats 17,18 . We thus hypothesized that i) as the rhizobacterial assemblages are predominantly recruited from the corresponding bulk soil, but are preselected by an excess of released root carbon, the bacterial diversity is generally lower and bacterial network displays less complexity in the rhizosphere. Relative to bulk soil, the rhizosphere can fuel its microbiota by providing rich and available energy and carbon resources 19,20 , and root-free soil is frequently carbon limited for microbes (Kuzyakov and Xu, 2013). We therefore further hypothesized that ii) richer copiotrophic bacteria are harbored in the rhizosphere relative to the bulk soil. To this end, we evaluated the community weighted mean 16s rRNA gene copies (rRNA operons) because copiotrophs are assumed to have more rRNA operons than oligotrophs 21,22 . Based on the general pattern of N depletion in the rhizosphere versus bulk soil 23, 24 and on the more intensive N transformations in N-richer conditions, we further hypothesized that iii) the functions responsible for the N cycling-related processes are less represented in the rhizosphere. Plants deposit a signi cant proportion of their photosynthates into the rhizosphere as rhizodeposits and root litter. The rhizodeposits, including amino acids, organic acids, and polymeric carbohydrates such as cellulose and hemicellulose 25 , are not only critical carbon and energy resources for rhizobacteria but are also an important trigger for attracting plant pathogens. We nally hypothesized that v) the functions related to ligocellulose degradation and plant pathogens are overrepresented in the rhizosphere. In this work, we analyzed and synthesized the very broad range of taxonomic and functional features of the microbiome of the rhizosphere as compared to the bulk soil separately from other drivers of compositional variation such as the environment. This yielded general principles of microbial selection around living roots, laying a foundation for harnessing the microbiome for sustainable agricultural production.

Results
Bacterial diversity and community composition in rhizosphere and bulk soils The dataset was collected from the public databases with published papers across continents. Synthesis of the bacterial sequences from cropland paired bulk vs rhizosphere soils under the vegetation of Gramineae, Leguminosae, Solanaceae and Cucurbitaceae (Fig. S1) provided a generalized pattern of alpha-diversity. Bacterial diversity is depleted in the rhizosphere compared to the bulk soil in terms of richness (-7%), Shannon index (-3.9%), evenness (-2.8%) and phylogenetic diversity (-5.2%) (Fig. 1). These alpha-diversity indices lost the signi cant difference under nitrogen (N) addition, indicating a N fertilization-induced compensation of common N depletion in the rhizosphere compared to bulk soil. The absence of signi cances for Gramineae and C4 plants shows similarities in their rhizosphere independent of photosynthesis type. Intriguingly, signi cantly higher phylogenetic diversity was detected in rice rhizosphere, probably re ecting the strongly changing redox environment near rice roots.
Three bacterial phyla -Bacteroidetes, Proteobacteria and Cyanobacteria -were enriched in the rhizosphere, whereas Planctomycetes, Chloro exi, Acidobacteria, Gemmatimonadetes and Nitrospirae were more abundant in the bulk soils ( Fig. 2 and Fig. S3). Even though the relative abundance of Proteobacteria was higher in the rhizosphere, the genera Haliangium, Anaeromyxobacter, Acidovorax and Nitrosospira, belonging to Proteobacteria, were much richer in the bulk soils (Fig. 2). Although there was no difference in the relative abundance of Firmicutes, Bacillus sp. is more common in rhizosphere and Romboutsia sp. prefers the bulk soils. The relative abundance of the phylum Actinobacteria was also similar in the two soils, but some of its genera including Streptomyces, Glycomyces, Arthrobacter, Microbacterium and Lechevalieria are much richer in the rhizosphere, whereas genera including Gaiella, Aeromicrobium, Blastococcus and Solirubrobacter were mainly localized in the bulk soils.
Co-occurrence patterns in rhizosphere and bulk soils Networks of bacterial communities in the rhizosphere (  Table S1). Although the rhizosphere network features many more modules (Table S1), the bacteria community network in the bulk soils is much more robust. This is re ected by the always higher natural connectivity upon removal of any level of nodes (Fig. 3c). In both networks, Proteobacteria and Actinobacteria were dominant in each of the modules ( Fig. 3d and 3e). However, the species highly connecting the nodes among modules differ ( Fig. 3f): in rhizosphere network, the main connectors are a liated with Lacibacter sp., Terrimonas sp., Nocardioides sp., Steroidobacter sp., Diaphorobacter sp., Ralstonia sp., A pia sp. and Arachidicoccus sp., whereas the role of the between-module connectors in the bulk soil network was played by the species Blastococcus sp., Streptomyces sp., Devosia sp., Acidibacter sp. and Bacillus sp.. Dormancy potential and heterotrophic strategies at bacterial community level Dormancy potential, characterized by genes involved in toxin-antitoxin and sporulation, was evaluated to characterize bacterial community-aggregated dormancy strategies in the rhizosphere and bulk soil.
Generally, a 42% higher toxin-antitoxin functional potential is common in the former (Fig. 4a), while a12% higher functional potential for sporulation was determined in the latter (Fig. 4b). Consequently, the bacteria inhabiting the rhizosphere employed distinctly different dominant strategies to enter dormancy than those inhabiting bulk soils. These differences in functional potential regarding toxin-antitoxin were removed by N fertilization, which compensated the N depletion in the rhizosphere. Paddy soil conditions eliminated the bacteria community with a well-developed toxin-antitoxin system in the rhizosphere, revealing a strong effect of land use on the bacterial dormancy potential.
The heterotrophic strategy was characterized by the rRNA operon copy number. Bacteria with more rRNA operon counts are able to sustain a higher maximum growth rate and respond faster to resources. The 6.9% more rRNA operon counts in the rhizosphere (Fig. 4c) indicate that much more fast-growing bacteria (r-strategists) colonize the rhizosphere than the bulk soil. The heterotrophic strategy at the bacterial community-level was also impacted by the land-use regime: the rhizosphere effect on the heterotrophic strategy was weaker in paddies than in upland soils.

Functional signatures in rhizosphere and bulk soils
All bacterial metabolic and ecologically relevant functions are predicted by the FAPROTAX based on the sequences (Fig. 5). Rhizosphere-inhabiting bacteria have increased functional potentials related to plant pathogens (+ 196%), methanol oxidation (+ 116%), ureolysis (+ 97%), cellulolysis (+ 76%), methylotrophy (+ 73%), chitinolysis (+ 62%) and nitrogen xation (+ 52%) compared to bulk soil. Signi cant differences in these functions (except for chitinolysis) were absent in paddy soils. Nitrogen fertilization decreased these rhizosphere effects on function potentials related to cellulolysis, chitinolysis, methylotrophy, and methanol oxidation. In contrast, the bulk soils hosted a bacterial community with 26-44% richer functional potentials related to nitrite respiration, nitrate denitri cation, nitrite denitri cation, nitrous oxide denitri cation, denitri cation, aerobic ammonia oxidation, iron respiration, nitri cation and aerobic nitrite oxidation (Fig. 5). The plant types, land use, and N addition can modify the effect sizes between rhizosphere and bulk soils. For instance, the effects on N cycling-related functions including nitrite respiration, nitrate denitri cation, nitrite denitri cation, nitrous oxide denitri cation, and denitri cation become non-signi cant in Gramineae and under N fertilization. Again, paddy soils showed a speci c response with the absence of the rhizosphere effects on such functions as nitri cation, nitrite respiration, aerobic nitrite oxidation and iron respiration.

Discussion
Bacterial diversity and composition in rhizosphere and bulk soil This meta-analysis is based on pairwise data (rhizosphere vs bulk soils) from amplicon sequencing approaches characterizing the taxonomic and functional features. This yielded fundamental insights into crop rhizosphere microbiomes on an across-continent scale and onto the plant-driven microbial taxa and their functional properties in this unique but uni ed habitat. Our design enabled testing the general in uence of crops on their rhizosphere bacterial community across soil and climate environments. This highlights the bene t of using sequencing data in soils to synthesize general microbiome patterns and to indicate specialized functions and life strategies of microbial taxa based on niche differences between rhizosphere and bulk soils.
The fact that rhizosphere microbiota differ from bulk soil microbiota is well documented 11,15,26,27,28 , and this is attributed to signi cant differences in physico-chemical attributes that result in niche differentiation 4,20,29,30 . In addition to the between-niche difference in terms of the environment, the generalized contrast between bulk soil and rhizosphere was the most important source of the distinction in microbiota composition 11,31 . We observed an overall lower bacterial richness, Shannon index, evenness and phylogenetic diversity for the rhizosphere vs bulk soil microbiota in all environments.
Consequently, bacterial diversity decreases with increasing substrate availability, the conditions common in the rhizosphere. The general view is that the rhizosphere microbiota, a subset of the community in bulk soil, possess certain similar traits across plants, within varied environments. This underlines the selective effect of the rhizosphere, which to a certain extent has generalized consequences for the rhizobacterial assemblages across crops. This even holds true for bacteria belonging to various classes, orders and families. Though the effect size differs between crop groups (Fig. 1), we stress that, even after accounting for genotypic and environmental differences, the selection of microorganisms common for the rhizosphere still displayed certain similar traits 4,8,32 .
Of course, the speci c environmental conditions fundamentally shift the rhizosphere effect size: the rice rhizosphere, for example, harbors phylogenetically broader bacteria than its corresponding bulk soil.
Paddy soils commonly develop anaerobic conditions, and rice plants therefore exhibit a developed aerenchyma. Due to the oxygen release around roots, the Eh and oxygen content in the rice rhizosphere are much higher than in the bulk soil 33, 34 and uctuate strongly. Thus, a broader phylogeny of both anaerobic and aerobic bacteria colonize the rice rhizosphere. These results indicate that environmental heterogeneity, such as root exudates, Eh and oscillating moisture, interact to shape the rhizosphere selective effect.
The particular microbial taxa recruited to the rhizosphere from the soil microbial reservoir can apparently form a particular core microbiome 11,13 . The core microbiome around roots contributes to plant growth and tness 16 . To date, the core microbiome of plants, regardless of whether in the rhizosphere, endosphere or phyllosphere, has been de ned mostly based on taxonomic markers. We, however, emphasize that more attention should be paid to identifying microbes having common functions that are selected for in a general rhizosphere setting. Accordingly, a function-based de nition of the microbiome should facilitate efforts to manipulate communities for useful purposes. Our comprehensive analysis revealed a few predominant taxa that are consistently enriched in the rhizosphere, for example the phyla Bacteroidetes and Proteobacteria (Fig. 2). This result underscores the fact that these phyla are generally adapted to C-rich conditions (common in the rhizosphere) 30,35 , and are consequently very similar across diverse plant species. This nding is not surprising because Bacteroidetes and Proteobacteria utilize labile carbon sources for high metabolic activity, fast growth and propagation-They are generally considered to be r-selected, or weedy fast-growing microbiota whose populations uctuate opportunistically 28 . In contrast to the rhizosphere, the bulk soil is generally enriched by other dominant phyla including Acidobacteria (Fig. 2b), which are oligotrophs 36,37 . Interestingly, a few phylum-level taxa are similar between rhizosphere and bulk soil, but ner taxonomic resolution reveals differences (Fig. 2). A case in point is Firmicutes: its class Bacilli is richer in the rhizosphere, whereas the class Clostridia is more abundant in the bulk soils. Similarly, Actinobacteria is richer in the rhizosphere, but its class Thermoleophilia is more abundant in the bulk soils. Consequently, the resulting generalized pattern that emerges based on the selective effect of the rhizosphere depends on the taxonomic resolution and on the fundamental niches at the level of classes and families. Microbiome formation: from structuring to functions In microbial networks, highly interconnected species are grouped into a module, within which species interactions are more frequent and intensive than with the remaining community. The rhizosphere bacterial network modularity is higher than that in the bulk soil ( Fig. 3 and Table S1). One potential explanation is more pronounced niche differentiations -both spatially and temporally -in the rhizosphere 25 because modules can be interpreted as microbial niches 38,39 . Modularity is one of the main organizing principles of biological networks 40 , and the higher modularity in the rhizosphere indicates a more complex topological structure. Nevertheless, the rhizosphere bacterial co-occurrence network is less robust (Fig. 3c). This generalization is not always valid, for example in the case of a single wheat rhizosphere network that was more stable than that in the bulk soil 26 . This is attributable to the fact that the networks (Fig. 3) are constructed across plant species. Moreover, that particular rhizosphere is characterized by very high temporal dynamics as compared to the more static conditions in the bulk soil 30,41 . Plant species selectively enrich speci c microorganisms by investing in root exudates to feed their rhizosphere microbiota 1,42 . The rhizosphere indigenous microbial community structure often differs remarkably across host species 43 . In a soybean cultivation, the microbial community in the rhizosphere was selected via niche ltering, whereas the bulk soil community arose via neutral (stochastic) processes 44 . The rhizosphere network allocates more network modules for executive functions, but fewer for network robustness, which partly re ects the fast element cycling in the rhizosphere 45,46 .
Rhizosphere and bulk soil are characterized by different dominant strategies of microbial dormancy: sporulation factors and toxin-antitoxin systems (Fig. 4a). The sporulation factors were more abundant in the bulk soil, whereas the toxin-antitoxin systems were enriched in the rhizosphere (Fig. 4b). During plant growth, roots actively and passively release a broad range of organic compounds into the rhizosphere.
These compounds are the driving force for microbial density and activity 29,47,48 . The sporulation factor was abundant in the bulk dryland soils but played nonsigni cant role in paddies (Fig. 4b). Hence, bacterial sporulation is more common in upland soils because the environmental conditions in paddies are more stable and homogeneous, and paddies are always moist. The less role of sporulation played in the rhizosphere (as compared to bulk soil) indirectly con rmed the buffered amplitude of moisture variation, which involves mucilage swelling conditions 49, 50, 51 . The genes related to dormancy/sporulation strongly increase with aridity 52 . Toxin-antitoxin systems are composed by the genes that encode a toxin protein that inhibits cell growth and an antitoxin that counteracts the toxin 53 .
More copiotrophs (e.g. Bacteroidetes and Proteobacteria) inhabited the rhizosphere, as con rmed by their rRNA operon counts, which were considerably higher there (Fig. 4c). Copiotrophs have more operon counts than oligotrophs 54 . The lower rRNA operon copy number is common for K-selected microbiota, and leads to slower growth rates and a more stable population. The bacterial functional trait of rRNA operon copy numbers increases with resource availability 22 . Organisms with multiple operon counts tend to be r-strategists, which are dominant in resource-abundant conditions and respond more rapidly to nutrient inputs 55,56,57 . Therefore, copiotrophs are dominant when resources are abundant, such as in the rhizosphere habitat.
Genes related to nitri cation and denitri cation were higher in the bulk soil (Fig. 5) because the rhizosphere is nutrient depleted due to root uptake 30 . This is indirectly con rmed by the fact that the rhizosphere affects nearly all these N cycle-related functions, whereby this depletion is alleviated by N fertilization (Fig. 5g-q). Nevertheless, cellulolysis, ureolysis and chitinolysis were more intensive in the rhizosphere, re ecting the increased abundance of bacteria degrading these substances. Members of Lysobacter (a liated to Gammaproteobacteria) were abundant in the rhizosphere (Fig. 2) and are known to be chitinolytic bacteria 58 . The methanol oxidation and methylotrophy genes are much more abundant in the rhizosphere than in the soil (except paddies). Methylotrophy is higher in the rhizosphere, but this difference is equalized in paddy soils because of the aerobic microenvironment around rice roots and because the available reduced C is strongly diluted in water excess.
In nature, speci cally in the rhizosphere, plants are constantly challenged by thousands of microbial populations, including commensals, pathogens and symbionts. Plant pathogens and N-xers (e.g. Rhizobium sp., etc.) are enriched in the rhizosphere (Fig. 5) because their reproduction and functioning are dependent on a plant host supply with organics. Although the rhizosphere is a uctuating environment in which the microbiome rapidly evolves in space and time, the accumulating evidence veri es that plants shape their rhizosphere microbiome to their own bene t, making sophisticated use of the microbial functional repertoire 12 . Nonetheless, it remains a challenge to clarify the equilibrium conditions for maintaining plant tness, which involves establishing a balance between the passive attacks of pathogens and the active recruitment of bene cial bacteria.
By integrating sequencing data from several studies, we generalized the main differences in the microbiomes of rhizosphere and bulk soil regarding bacterial diversity, composition, selection of speci c groups, co-occurrence network, and a very broad range of functions (Fig. 6). Bacterial diversity in the rhizosphere is reduced by 2.6-7% and represents a subset of the communities in bulk soil. The rhizosphere community composition is highly enriched with copiotrophic bacteria such as Proteobacteria and Bacteroidetes, while it is strongly depleted in Chloro exi, Acidobacteria and Nitrospirae. As the rhizosphere has an organic C surplus and circulates nutrients quickly, the quicker-growing bacteria are overrepresented with functions related to lignocellulose degradation and plant pathogens, but depleted in the functions responsible for N cycling-related processes (except N xation). The indirect proof of the generalizations presented here is that nitrogen fertilization alleviated nearly all rhizosphere effects on bacterial diversity and functions. This also con rms the common mineral N depletion against the background of excess C around roots. In conclusion, the selective effects of the rhizosphere in shaping microbial communities to some extent overrides the differences between soils, crops and climate. This makes the rhizosphere the strongest factor in forming the composition, structure and functions of the soil microbiome and, thus, a key factor in the cycling of biogenic elements.

Data collection
An extensive literature survey was conducted through the Web of Science database (http://apps.webofknowledge.com/) from 2015 until March 2020. The key words of the literature search were "bulk" and "rhizosphere" and "bacteria". A total of 1081 samples from 29 publications including eld and greenhouse experiments were collected (Fig. S1), comprising 584 pairs of bulk soils vs rhizosphere (Appendix A). These studies examined the bacterial communities by high-throughput sequencing, including Illumina and 454 platforms.
The following criteria were used to select appropriate studies: (1) experiments in agricultural soil were included, studies in grassland, forest, or desert ecosystems excluded; (2) articles with sequencing data in one to-one correspondence between rhizosphere and bulk soil were retained. Nine primer pairs were identi ed from the study metadata (Table S1), with most samples using the primer set of 515F and 806R 28 . Besides the sequencing data, we also collected the following parameters: plant type (C3, C4), plant family (including Gramineae, Leguminosae, Solanaceae and Cucurbitaceae), soil texture and pH, location (i.e., latitude and longitude), mean annual temperature (MAT), mean annual precipitation (MAP) and the nitrogen, phosphorus and potassium fertilization rate.

Bioinformatic analysis and taxonomical annotation
We downloaded the raw sequence data according to the Accession Number from NCBI, and then FASTQ les were obtained for each study. First, raw paired-end sequences of each study were merged with the fastq_mergepairs command and quality ltered with the fastq_ lter command in USEARCH software 59 .
Sequences with a quality score below 20 and shorter than 200 base pairs were ltered. In a next step, the fastx_uniques command was implemented to remove the redundant sequences, and the denoising was conducted by Unosie3 to remove the chimeras and obtain the single nucleotide precision fragments. To integrate studies that used different 16S rRNA gene regions, we adopted a closed-reference work ow, which is a database-dependent approach employing a prede ned set of reference sequences with known taxonomy, to obtain representative sequences and assign taxonomy (Fig. S2). Then, the closed_ref command (sequence identity ≥ 97%) in USEARCH was used to map the above-obtained fragments to non-redundant full-length 16S rRNA sequences with the Silva database (http://www.drive5.com/usearch/manual/sintax_downloads.html). The fragments that could not be mapped were assigned to unknown sequences. After that, the OTU table was constructed according to the mapped full-length 16S sequences. Finally, the mapping OTU result of each data set in Usearch table format was directly imported to R software (version 3.6.0, R Core Team, 2019) and merged into an integrated OTU table for further analyses. All samples were rare ed to the same sequencing depth (10,000 reads per sample) for the downstream analysis.

Statistical analysis
Calculation of the response ratio Declarations Data availability The sequences obtained in this study are previously published and publicly available, which were downloaded from the NCBI Sequence Read Archive. All the accession numbers were listed in Appendix A.