1Center for Ecological Research, Kyoto University, Otsu, Shiga, Japan. 2PRESTO, Japan Science and Technology Agency, Kawaguchi, Saitama, Japan. 3Department of Biology, Stanford University, Stanford, CA, USA. 4Department of General Systems Studies, University of Tokyo, Meguro, Tokyo, Japan. 5Department of Bioresource Science, College of Agriculture, Ibaraki University, Ami, Ibaraki, Japan. 6Department of Biological Sciences, Nara Institute of Science and Technology, Nara, Japan. 7Genetic Resource Center, National Agriculture and Food Research Organization (NARO), Tsukuba, Ibaraki, Japan. 8Institute for Advanced Biosciences, Keio University, Tsuruoka, Yamagata, Japan. 9Intestinal Microbiota Project, Kanagawa Institute of Industrial Science and Technology, Kawasaki, Kanagawa, Japan. 10Transborder Medical Research Center, University of Tsukuba, Tsukuba, Ibaraki, Japan. 11Institute of Industrial Sciences, The University of Tokyo, Tokyo, Japan. 12Graduate School of Agriculture, Kyoto University, Kyoto, Japan. 13Graduate School of Agricultural Science, Kobe University, Nada-ku, Kobe, Japan. 14Institute of Plant Sciences, University of Bern, Bern, Switzerland. 15Department of Agroecology and Environment, Agroscope, Zurich, Switzerland. 16State Key Laboratory of Plant Genomics, Institute of Genetics and Developmental Biology, Chinese Academy of Science, Beijing, China. 17Centre of Excellence for Plant and Microbial Sciences (CEPAMS), Institute of Genetics and Developmental Biology, Chinese Academy of Science & John Innes Centre, Beijing, China. 18Hokkaido Agricultural Research Center, NARO (National Agriculture and Food Research Organization), Memuro, Hokkaido, Japan. 19RIKEN Center for Sustainable Resource Science, Yokohama, Kanagawa, Japan. 20Graduate School of Life Sciences, Tohoku University, Katahira, Sendai, Japan. 21Department of Ecological Science, Vrije Universiteit Amsterdam, Amsterdam, the Netherlands. 22Present address: RIKEN BioResource Research Center, Tsukuba, Ibaraki, Japan. *e-mail: [email protected]
Identifying the elements that drive ecosystem dynamics is a major challenge; manipulating these elements to produce applicable benefits is even more demanding. Agroecosystems are among the
most complex targets for human manipulation because they consist of entangled webs of interactions formed among macroorganisms (such as plants and insects) and countless microorganisms1–3 (for example, bacteria, archaea, protists and fungi) (Fig. 1). Scientists are now facing the unprecedented task of maximizing the functions of these microbial communities (microbiomes) as novel pest/patho-gen strains, climate change and limited supply (and environmen-tal costs) of chemical fertilizers are increasingly threatening stable agricultural production4–8. Accomplishing this goal, in the face of doubling global crop demand by 2050, requires rapid and immedi-ate solutions9. Deploying microorganisms to increase plant nutrient uptake10–12 and resistance to biotic and abiotic stresses13–17 offers one of the few untapped reservoirs of opportunities to confront sustain-ability issues in agriculture1–3,18.
However, optimization of plant–microbial partnerships is a daunting task given the complexity of plant–microorganism and microorganism–microorganism interactions, and the depen-dence of those interactions on environmental conditions. One of the master regulators of the plant phosphate starvation response is also responsible for some plant defence responses, and can thus drive changes in plant-associated microbiome structure12,19. In some cases, microorganisms are recruited or repelled directly by plant signalling mechanisms, which have evolved to interact with mycorrhizal fungi and nitrogen-fixing bacteria, as well as to oppose
diverse pathogens1,20,21. Thus, a better understanding of complex interactions among plant genotypes, environmental conditions and microbiome structure provides indispensable information in breeding programs3. Alongside conventional breeding approaches, novel sensing technologies are enabling real-time diagnostics and forecasts of plant physiological conditions based on climate and soil physiochemical data22,23. Despite the huge potential, we are just beginning to explore opportunities in which microbiome data can be integrated into ‘smart farming’ practices22,23.
Our aim here is to propose interdisciplinary research strate-gies for optimizing microbiome functioning in agroecosystems. Although a number of plant-growth-promoting microorganisms have been reported, microbial populations introduced to farm-lands often decline rapidly24,25. In addition, they may occasionally outcompete indigenous (resident) microbiomes, potentially having unexpected deleterious effects on ecosystem functions26,27. Here, we propose an interdisciplinary strategy for optimizing plant–micro-biome interactions at the agroecosystem level, beyond classical studies that have focused on direct, pairwise plant–microorganism interactions in controlled experimental conditions. First, we discuss the importance of manipulating microbiome assembly at the ini-tial stages of plant development and then introduce the concept of ‘core microbiomes’, defined herein as sets of microorganisms that form cores of interactions that can be used to optimize microbial functions at the individual plant and ecosystem levels (Fig. 2). These organisms are key in organizing the assembly of plant-associated microbiomes, rather than just promoting host plant growth them-
Core microbiomes for sustainable agroecosystemsHirokazu Toju1,2*, Kabir G. Peay 3, Masato Yamamichi 4, Kazuhiko Narisawa5, Kei Hiruma2,6, Ken Naito7, Shinji Fukuda2,8,9,10, Masayuki Ushio1,2, Shinji Nakaoka2,11, Yusuke Onoda12, Kentaro Yoshida 2,13, Klaus Schlaeppi14,15, Yang Bai16,17, Ryo Sugiura2,18, Yasunori Ichihashi2,19,22, Kiwamu Minamisawa20 and E. Toby Kiers21
In an era of ecosystem degradation and climate change, maximizing microbial functions in agroecosystems has become a prerequisite for the future of global agriculture. However, managing species-rich communities of plant-associated microbi-omes remains a major challenge. Here, we propose interdisciplinary research strategies to optimize microbiome functions in agroecosystems. Informatics now allows us to identify members and characteristics of ‘core microbiomes’, which may be deployed to organize otherwise uncontrollable dynamics of resident microbiomes. Integration of microfluidics, robotics and machine learning provides novel ways to capitalize on core microbiomes for increasing resource-efficiency and stress-resistance of agroecosystems.
selves. Second, we show how informatics pipelines can be used to identify such core microbiomes, which, if deployed correctly, will mediate and organize plant–microorganism and microorganism–microorganism interactions by recruiting indigenous microorgan-isms with diverse functions and suppressing high pathogen loads in the field (Fig. 2). Lastly, we explore ways to distribute and man-age plant-associated core microbiomes in agroecosystems in light of recent advances in medical science, microfluidics, robotics and computer science.
Initial assembly and core microorganismsSignificant progress in linking microbiome structure with host health has been made in medical studies, with increasing knowledge of common features in microbial dynamics28,29 towards effective manipulation of gut microbial communities30 (for example, faecal transplants between human individuals31). In agriculture, however, we have just started to build a general framework for managing the complexity of plant-associated microbiomes1,17,32. The ultimate goal is to identify successful strategies for converting microbiomes from uncontrollable, black box components into rich sources of functions in agroecosystems.
Microbiome typesIf plants were to be randomly colonized by microorganisms in the environment, the potential combinations of coexisting species on plant shoots and roots would be vast. In reality, however, micro-biomes are structured in such a way that individuals of a plant species may be classified into a small number of types (hereafter, ‘microbiome types’) within a local population33. Recent studies have also shown that plant-associated microbiomes are often classified into a few groups depending on crop plant varieties34–36 and even cropping practices36,37. The concept of microbiome types, which
are defined by the discrete or semi-discrete nature of microbiome structures (Fig. 3a,b), first appeared in medical studies on human gut bacteria30,38 and demonstrated how human individuals could be effectively classified into a few clusters based on their gut bacterial compositions, also known as ‘enterotypes’38 (see ref. 39). Invoking community theory, those studies argued that microbiome types are similar to what are known in ecology as ‘alternative stable states’, which represent locally stable, alternative equilibria of community or ecosystem dynamics40.
In general, shifting a biological community from an equilibrium to alternatives requires specific environmental perturbation41. This is why it can be difficult to change a disease-related microbiome type into a healthy microbiome: microbiomes tend to be both resis-tant and resilient, returning to similar community structure after external disturbance or inputs of new species26,30,41. For example, Bifidobacterium or Lactobacillus introduced to human gut as benefi-cial microorganisms are usually replaced by other microorganisms hence their ‘probiotic’ effects do not last without continually supply-ing the host-benefiting bacteria through food or medicine42. Even if shifts from unfavourable (host-damaging) microbiome types occur as consequences of large perturbation (for example, antibiotic or fungicide treatment), such shifts would often result in unexpected fluctuations of microbiome structure43,44. Thus, altering established microbiome types remains a major challenge for human microbi-ome management31.
Fortunately, changing established microbiome structures into alternative ones is not necessarily the main goal of plant microbi-ome management. Because most crop plant species are annuals and their early developmental stages are easily accessible, it is possible to concentrate our efforts on establishment and maintenance of benign microbiome structures, rather than treatment of unhealthy host individuals as needed in medical science. Moreover, while human intestines are fluid environments with frequent perturba-tions (inflow of diet and associated microbial colonizers), environ-ments within plant tissue, especially those within root endospheres, offer some protection from physical perturbations45. Such environ-ments likely allow early microbial colonizers to persist on host plant individuals. An experimental study on rice hosts demonstrated that microorganisms began to colonize the seedling root endospheres within 24 hours of first soil contact and reached a steady microbi-ome state within two weeks36.
Priority effects in initial assemblyIn plant microbiome assembly, small differences in the early colo-nization processes, such as the order of species’ arrival, can result in large differences in community structure46,47 (that is, priority effects48). For example, a phyllosphere bacterium, Pantoea agglom-erans, suppresses a pathogenic bacterium, Pseudomonas syringae (which causes basal kernel blight of barley), only when it is inocu-lated to hosts earlier than the pathogen47. In species assembly, early colonizers often have an advantage because they can use space and resources earlier than other microorganisms and/or because they can produce physical barriers and/or antibiotics that slow coloni-zation of subsequent microorganisms on the plants49 (a process known as niche pre-emption and modification48). Indeed, various taxonomic groups of plant-associated and soil microorganisms are known to produce antibiotics50,51, by which they would exclude competitors from endospheres, phyllospheres or rhizospheres. Similarly, early colonizers may alter jasmonic/salicylic acid lev-els of host plants20,52, potentially changing immune responses of the plants to late colonizers. Moreover, phyllosphere microorgan-isms forming biofilms on leaf surfaces can control assembly of air-borne microorganisms by protecting such late colonizers from UV and desiccation51.
Given the importance of priority effects in plant microbiomes, managing microbial assembly in seeds and seedlings should be
Herbivores
Sap-suckinginsects
Suppression of soil pathogensby rhizosphere microorganisms
N2
Fungi
Recruitment or blockingof air-borne
microorganisms byfoliar endophytes Drought or
high-temperatureresistance enhanced
by symbionts
+
+
×
Nitrogen-fixing Induced defencebacteria
Entomophagousfungi
×
Recruitmentof functional
microorganisms
Nematophagousfungi
Solubilization andtransport of
soil N and P bymycorrhizal fungi
Bacteria
Fig. 1 | Entangled webs of below-ground interactions. In ecosystems, plants not only face biotic/abiotic environmental stresses but also interact with microorganisms with various functions. Nitrogen-fixing bacteria55 and mycorrhizal fungi10, for example, supply solubilized nitrogen (N) and/or phosphorus (P) to host plants, thereby allowing us to reduce chemical fertilizer input into agroecosystems. Some of those symbionts are also known to enhance host plants’ resistance to drought and salinity stress15. Phyllosphere endophytes16,51, disease-suppressive bacteria13 and nematophagous/entomophagous fungi14,69 may play essential, but often overlooked, roles in suppressing populations of air-borne or soil-borne pathogens and pests. Untapped sources of microbial functions will be explored not only in agroecosystems, but also in natural ecosystems87,88.
a research priority. Core microbial species are expected to estab-lish more reproducibly in seeds and seedlings than in adult plants with established resident communities in endospheres and rhi-zospheres45. Because pathogen infection at early growth stages is among the most devastating factors reducing yields of major crop plants (for example, soybean53), control of early microbiome assem-bly could have a significant impact on agricultural production50. The question addressed below is how to take advantage of priority effects in controlling otherwise stochastic processes in early micro-biome succession (Fig. 3c).
Deploying core microorganismsPlant microbiome management based on priority effects can be exe-cuted with three processes: (1) functional species recruitment, (2) pathogen/pest blocking, and (3) core reinforcement (Box 1, Fig. 2). First, beneficial species must be preferentially recruited. This pro-cess is facilitated using microbial species that can recruit many other microbial species with desirable physiological functions1,54. These facilitator microorganisms would be the first to be inoculated in seeds or seedlings as pioneer symbionts, thereby preferentially recruiting communities from native pools of potential microbial symbionts through priority effects. A bacterial strain of Bacillus amyloliquefaciens is known to enhance soybean nodulation by a nitrogen-fixing bacterium, Bradyrhizobium japonicum55, and hence it is a prospective candidate to be used as an early colonizer.
Second, preferential assembly of microbiomes also provides a basis for preventing infection by pathogens and pests. Priority effects facilitate pathogen/pest blocking by allowing resident (early colonizing) microorganisms to slow or prevent the entry of antagonistic late colonizers16,49. In an experiment using tomato, for example, some resident bacterial communities, whose members competed for resources with a soil-borne pathogenic bacterium (Ralstonia solanacearum), significantly reduced the invasion suc-cess of the pathogen into host plants49.
Third, identifying pairs or sets of core microorganisms (that is, core microbiomes, Fig. 2) that form strong facilitative and mutualis-tic interactions with each other can further strengthen such recruit-ment and blocking functions. For example, when co-inoculated to maize or soybean seeds, Azospirillum brasilense Az39 and B. japoni-cum E109 play complementary roles in enhancing early seedling growth of host plants because the former bacterial strain is a better producer of phytohormones regulating root growth (indole 3-acetic
acid and zeatin) and the latter is a better producer of a shoot-growth promoter (gibberellic acid)56. By fuelling early host-growth synergistically, this pair of bacteria is expected to increase plant physiological homeostasis, promoting stable establishment of late colonizing microorganisms.
While screening of plant-growth-promoting microorganisms57 will continue to represent an important research direction, we pro-pose another direction of microbial diversity exploration by focus-ing on roles of core microorganisms organizing community-scale phenomena. Core microorganisms (and core microbiomes), in our definition (see ref 58 for a simpler definition of the phrase), do not necessarily promote plant growth directly. Rather, they are expected to play pivotal roles in organizing assembly of plant-associated microbiomes within and around host plants. This means that a core microorganism may not only be a mutualist but also a commensal-ist or even an antagonist when it is the sole inoculum in a plant performance assay. As microorganisms without direct benefits to hosts can play essential roles in specific community contexts, we put emphasis on optimizing functioning at the microbiome level.
Informatics for nominating core microorganismsWhile a number of studies have demonstrated priority effects in plant microbiome assembly46,48,49, finding potential core microor-ganisms with preferable priority effects is a challenge. To explore candidate core microorganisms, it is important to understand how microbial species in endospheres and rhizospheres interact with each other59. Network theory provides a ‘bird’s-eye’ view of poten-tial microorganism–microorganism interactions49,60,61, by allowing us to understand which species are most likely to mediate interac-tions among many others within microbiomes17,29,62–64. Combined with ‘bottom-up’ approaches for evaluating the persistence and ecological effects of synthetic communities19,3257,65, studies targeting in natura dynamics of microbial interactions will help us optimize relationships between introduced and resident microbiomes in agroecosystems.
Microbial networksDiverse functional groups of microorganisms interact with each other within and around plants, forming complex webs of interac-tions66 (Fig. 3a). Recent studies have demonstrated that the exact physiological effects of microorganisms on host plants will vary considerably depending on the community structure of plant
Functional speciesrecruitment
Pathogen/pest blocking
Core reinforcement
Pathogenic microorganisms (type I)
Functional microorganisms (e.g. N or P provisioning)
Fig. 2 | Managing native biomes using core microorganisms. a, Nominating core microorganisms. Within a microbial network, hub microbial species or strains are defined solely based on network topology data. Among those topological hubs, it is possible to explore candidates of core microorganisms that are expected to mediate interactions between plants and native microbiomes. b–d, In our definition (Box 1), core microorganisms are scored based on their potential for recruiting native microorganisms with diverse physiological and ecosystem functions (b) and potential for blocking pathogen/pest infection (c). Such roles of core microorganisms will be further reinforced by introducing pairs or sets of core microorganisms in facilitative interactions (that is, core microbiomes) (d).
microbiomes67. Such background networks of microorganisms can be complex, not just in diversity but also in structure. For example, plant–microorganism interactions can involve symbionts that are physically nested within another symbiont. These partners can play critical roles: bacteria found inside hyphae of mycorrhizal fungi can contribute to plant–fungal symbioses by fixing atmospheric nitro-gen or promoting fungal mycelial growth11,68. Likewise, microor-ganisms can form unexpected connections with other organisms. For instance, Metarhizium robertsii, a common soil fungus, is both an endophyte and an insect pathogen. Its unique position allows it to provide insect-derived nitrogen to its plant host through fungal mycelia in exchange for host carbon69 (Fig. 1).
To encapsulate these dimensions of microbiome complex-ity, a number of bioinformatic approaches have been developed to analyse microbial diversity and interaction networks based on high-throughput DNA sequencing61,70. Nonetheless, most plant microbiome studies have focused on either bacterial or fungal biomes, rather than both. Moreover, most current studies analyse networks connecting coarsely defined (often at the 97% sequence similarity level) operational taxonomic units that can hide meaning-ful ecological variation, while informatics tools to perform strain- or individual-sequence-level analyses are now becoming available71.
Given the huge number of potential microorganism–micro-organism associations, statistical pipelines for analysing network
Box 1 | Three criteria for nominating core microorganisms
We propose a network theoretical framework for finding optimal core microorganisms based on three-step criteria (outlined in Fig. 2).
Functional species recruitment. In terms of potential roles in func-tional species recruitment, each species (species i) in microorgan-ism–microorganism network data can be scored by a simple index as follows:
∑ σσ
=≠ ≠
Fw w i( )
,ii k l
k l kl
kl
where species k and l are from the set of functional species (MF) in the microbiome (that is, k, l ∈ MF), σkl is the number of shortest paths between species k and l, and σkl(i) is the number of shortest paths between species k and l that pass through the focal species i. Unlike most network theoretical indices63, this index has user-defined weighting parameters; i.e., wk and wl for weighting the functions of species in the microbiome. We incorporated this flexibility into the index because individual farmers may vary in their strategy for pri-oritizing multiple microbial functions. In the management of agro-ecosystems suffering from root-knot nematodes, for example, one may want to put higher weighting values on nematophagous endo-phytes than on mycorrhizal fungi. The weighting parameters may be also used for incorporating results of inoculation experiments comparing plant-growth-promoting abilities among microbial spe-cies/strains12,55. Meanwhile, tools for annotating functional roles are increasingly available for both bacteria70 and fungi81, although their underlying databases need to be continually enriched.
In practice, we need more ecological information for annotat-ing functional microbial species (MF) and setting species-specific weights (wk and wl). In most available microbial networks, a high proportion of species have not yet been ecologically annotated. However, given that species with the same functional roles often coexist in nature (that is, functional redundancy in microbi-omes30), the above network index will help us find core microor-ganisms even with our current knowledge of microbial functions.
Pathogen/pest blocking. Core microorganisms can be used not only for recruiting functional symbiont species, but also for blocking pathogens and pests in agroecosystems. By defining a set of plant pathogens/pests (MP) within a microbial network, a species i may be scored as follows:
∑ σσ
= −≠ ≠
Ew w i( )
,ii m n
m n mn
mn
where species m and n are from the pathogen/pest set MP and σmn, σmn(i), wm, and wn are defined as in the above recruitment index Fi. With user-supplied parameters weighting disease severity among pathogens (wm and wn), each microbial species will be evaluated in terms of its potential roles in blocking the entry of the defined pathogens/pests (m, n ∈ MP) to host plants.
By integrating functional species recruitment and pathogen/pest blocking, the total benefits of a microbial species for its host plant will be scored as follows:
= +B F rE ,i i i
where r is a parameter for balancing two aspects of roles in micro-biomes. The index is then standardized so as to vary from 0 to 1 as follows:
′ =−
−B
B BB B
,ii min
max min
where Bmin and Bmax are the minimal and maximal scores of Bi within a network, respectively.
Core reinforcement. When each microbial species is evaluated in a network, we can further explore sets of microorganisms that promote the formation of robust microbiomes. In exploring best pairs of core species, we can take into account roles of respec-tive species as well as compatibility between focal two species as follows:
= ′ ′R B B C ,ij i j ij
where ′B i and ′B j are total benefit scores of species i and j, respectively, and Cij represents co-occurrence patterns observed in microbial network data (Fig. 2). Several statistical approaches for evaluating co-occurrence patterns for pairs of microbial spe-cies (or strains) (Cij) are now available33,61, allowing us to cal-culate the Rij index and thereby predict reinforcement effects between pairs of microorganisms. However, the index does not represent metabolic interactions involving multiple microbial species. Therefore, the network-based method outlined here should be applied as a complementary, but not an alternative, approach to explore sets of cooperative microorganisms based on metagenomic and metatranscriptomic analyses118 of meta-bolic pathways.
architecture are essential49,60,61. These can be classified into two types. One is based on the ‘snapshot’ information of microbiome structure, in which signs of co-occurrence (or non-coexistence) can be statistically examined in each pair of microbial species33. For this type of analysis, compositions of microorganisms are evaluated simultaneously for tens or hundreds of plant/soil samples and then network links are inferred in pairs of microorganisms that show strong statistical signs of co-occurrence or segregated distribu-tion61. These snapshot analyses are suitable for revealing patterns in microorganism–microorganism co-occurrence, which may indi-cate interactions or shared environmental preferences (that is, niche overlap) among microbial species33,72. Although snapshot networks provide less information about functional inter-microorganism interactions than networks inferred with more sophisticated analy-ses (for example, time-series analytical methods discussed below), they represent overall consequences of direct and indirect interac-
tions, in which host gene expression profiles altered by mycorrhi-zal fungi, for example, may influence colonization success of some pathogens and/or nitrogen-fixing bacteria11,20,21.
Other emerging analytical methods for time-series community analyses (for example, empirical dynamic modelling73–75, sparse S-map76 and transfer entropy77) have the potential to reveal a more dynamic nature of microorganism–microorganism interactions64. For example, a time-series study of mouse gut bacteriome networks suggested that both Allobaculum and Ruminococcaceae bacteria positively influenced a Rikenellaceae bacterium when the mice were young, while the Rikenellaceae bacterium had negative impacts on Clostridiales and Ruminococcaceae bacteria in old mice76. Those analyses provide ways for inferring not only microorganism–micro-organism interaction networks, but also more comprehensive webs of interactions among microbial population dynamics, abiotic envi-ronmental fluctuations and gene expression profiles of respective
3. Assembly of native microorganismsin agroecosystems
Priority effects of inoculated coremicroorganisms in recruiting
functional microorganisms andblocking pathogens
+
Assembly towardsa preferable
microbiome type
Alternative microbiome typesMutualistic microorganisms Pathogenic microorganisms
Other microorganisms
Fig. 3 | Microbial network information for controlling microbiomes in agroecosystems. a, Network depicting potential inter-microorganism associations. Microorganisms frequently co-occurring in the soil are connected. Those that are interlinked can share environmental preferences and/or interact with each other in complementary or facilitative ways. Only the top-100 bacteria and top-100 fungi that were most abundant in the soil samples collected along an environmental gradient in Hawaii86 are shown. b, Microbial network modules and plant microbiome types. Microbial networks often contain modules, which represent groups of frequently coexisting microorganisms. Microbial network modules may represent or be represented by microbiome types of plant individuals. c, Controlling native microorganisms with inoculated core microorganisms. For the first step, modules representing microbiome types of healthy plant individuals are searched within network data (1). Some hub microorganisms of the selected modules are then inoculated to seeds or seedling as candidates of core microorganisms (2). When the inoculated plants are introduced to agroecosystems, those that can coexist with the embedded core microorganisms will be preferentially recruited from the pool of indigenous microorganisms (3).
host and microbial species64. They will also help us evaluate how fluctuations in environmental conditions (for example, tempera-ture, precipitation and edaphic factors) and/or host properties (for example, strains and physiological conditions) affect community dynamics73,74, as well as how priority effects of early colonizers per-sist through host plant development. The precision and applicabil-ity of these analyses, however, depend on the length and stochastic noise of particular time series as well as properties of the target sys-tem’s dynamics (for example, its dimensionality and nonlinearity)78. In addition, relative abundance data obtained from high-throughput sequencing should be used with caution as they provide less infor-mation than absolute abundance data in inferring microorganism–microorganism and microorganism–host interactions79. Methods based on control-DNA calibration of high-throughput sequenc-ing samples80 provide absolute abundance data and are promising for improving our ability to elucidate how microbial interactions change through time.
Network modules and hubsWhen basic information about the topology of a microbial network is obtained, the next step is to evaluate properties of the network structure. These networks can be compartmentalized into a small number of ‘network modules’, in which microbial species/strains are densely linked with each other2,3637,61. Those microbial net-work modules may have corresponding microbiome types, which are defined from the aspect of the microbiome structure of host plant individuals33,37 (Fig. 3b). As network modules can differ in
microbial functions (for example, differential proportions of meth-anogenic microorganisms among microbial network modules of rice36), plant hosts with different microbiome types may differ in physiological states.
With network topology data, we can further identify species that hold key topological positions within the network. These ‘hub’ species coexist with most other species in each module (Fig. 3b). In microbial networks, simple ‘centrality’ metrics63 are often used to score microorganisms based on their topology64. If micro-bial networks are compartmentalized into modules, hub species of modules representing microbiome types of healthy plant indi-viduals are also candidates to become core microorganisms; those hubs may be deployed to organize favourable plant microbiomes. It should be stressed, however, that hub species are designated solely based on positions within network topology, while we define core microorganisms by their high potential to organize microbi-omes in ways that benefit host plants (Fig. 3c, Box 1). Thus, infor-mation on hub species provides only the very first step in core microorganism screening.
Designing core microbiomesWhile challenging, informatics of microbial network structure will help us design core microbiomes to be introduced, with top priority, to plants and agroecosystems. As databases and related informatics tools for inferring functional roles of microorganisms are increas-ing in sophistication and depth70,81, microorganisms in network data can be classified, albeit not comprehensively, into functional
c
a b
OilOil
Mixture ofmicroorganisms
Outputmicrodroplets
Tissue samples ofhealthy plant individuals
Concentrated microbiome Library of candidatecore microbiomes
Dead microorganisms
Living microorganisms
Screening of sets of microorganisms
Centrifuge for separationfrom plant tissue
Microdroplet formation
Seedling containersfor planting machines
Core microbiome 1
Core microbiome 2
Inoculation of coremicroorganisms
on microfluidics devices
Automated planting in agroecosystemsOptimized culture
media for coremicroorganisms
Ramanspectroscopy
screening
Fig. 4 | Preparing and deploying core microbiomes. a, Microbiome cocktails. Symbiont cells obtained from healthy plant individuals can be enriched by centrifuge-based methods92,93. b, Application of microdroplet technologies for isolating and screening core microbiomes. Microdroplet devices, which were originally developed for the isolation of single cells94, will be used for screening pairs/sets of core microorganisms interacting with each other in facilitative ways (that is, core microbiomes). c, Deploying core microbiomes. Single-cell phenotyping/screening based on Raman spectroscopy95 and microfluidics devices90,96,97 can be used for high-throughput inoculation of sets of core microorganisms into seedlings. The use of multiple core microbiomes will increase heterogeneity within agroecosystems, contributing to the recruitment of diverse functional microorganisms and the prevention of pathogen/pest outbreaks.
groups. The basic idea is to use such annotated network data to score microorganisms in terms of their potential positive impact. This can be done in silico by modifying existing network metrics. In Box 1, we illustrate how to identify microbial species that are placed, within a microbial network at central positions, interlinking other microorganisms with positive functions for host plants (Fig. 2b). This approach is also used to score species based on their potential ability to mediate interactions among pathogenic species in a net-work—species with low scores are prospective in avoiding or slow-ing the entry of pathogenic species to host plants (Fig. 2c). Based on the two criteria of functional species recruitment and pathogen/pest blocking, microorganisms in a network can be ranked to nominate candidate core microorganisms.
When each microbial species in a network is scored in terms of its potential for recruiting preferable species and blocking patho-gens/pests, the next step is to explore best pairs or sets of core microorganisms (that is, core microbiomes, Fig. 2d). A best pair of core symbionts, for example, would be explored in a microbial network dataset by taking into account each microorganism’s score for functional species recruitment and pathogen/pest blocking as well as compatibility between two focal microorganisms (Box 1). In other words, exploring pairs or sets of core microorganisms that potentially interact with each other in facilitative or complementary ways62,82, would be a key step for designing core microbiomes. Such strategies of using pairs/sets of microorganisms have been applied in medical studies, wherein inoculated sets of microorganisms had persisted over time and had suppressed pathogens inside the human gut83. In plant science, unpredictability in the persistence levels of microorganisms introduced to plants24,25, and possible deleteri-ous effects of microbial inoculation to ecosystems26,27, are poten-tially reduced by designing optimal pairs/sets of microorganisms. Pioneering attempts to inoculate multiple microbial species have shown varied effects on plant performance15,16, stressing the need for more in planta or in natura studies.
The largest challenge in building informatics pipelines for designing core microbiomes, is collecting and cataloguing microbial functions. This is a daunting task. The majority of species in data-bases, including many endophytic bacteria and fungi, have yet to be classified and many known plant-growth-promoting species have been classified as putative saprotrophs64,81, potentially causing biases in functional annotations. While an increasing number of studies are estimating microbial functions based on genome structure70 and experimental inoculation of microbial strains to crop plants12,15,19, we need to make more effort to collect and curate those data sys-tematically. Nonetheless, novel informatics pipelines for maximiz-ing synergistic microbial interactions based on genome data, and integrating multi-omics data (for example, metagenomic, transcrip-tomic and metabolomic data), are becoming available84,85. Moreover, because microbiomes often involve species with the same, or simi-lar, functions (that is, functional redundancy30), we may not need to annotate all microbial species in a network when we explore some of the potential best sets of core microorganisms.
agroecosystem management with core microbiomesEmerging research focussing on core microbiomes will offer ways to mediate interactions between plants and their external biotic and abiotic environments. However, diversity and functionality of microorganisms, as well as microbial network structure, depend greatly on crop plant species/varieties and field conditions3,2337,86 (for example, resident soil biota, particle size, pH, nutrient avail-ability and tillage cycles). Therefore, we need to accelerate feed-back between informatics and field monitoring to understand how crop production can be determined by predictability of function as linked to identity (for example, similarity in phylum- or order-level compositions of optimal microbiomes among sites or crop plant species/varieties) and regionality in resident microbiome composi-
tions and functions. Moreover, biotic/abiotic conditions can change through time in agroecosystems, potentially having great impact on the success of core microbiome introduction. We, hereafter, discuss how recent technical advances in microfluidics, robotics and com-puter sciences are integrated towards next-generation agriculture, through which we can propose core microbiomes optimized for real-time conditions of local farms and plantations.
logistics of core microbiomesDeveloping standardized methods for collecting and inoculating core microorganims onto hosts will require innovative research lines. While many endophytic bacteria and fungi can be easily iso-lated from the environment65,87,88, there may be core microorgan-isms that remain unculturable with standard techniques. In such cases, the development of microfluidic chips offers promise because these devices can precisely control density, shape and size of micro-bial communities89,90. Physical micro-scale structure prevents the dominance of one species over others, and controlled diffusion of metabolites means that individual bacteria have a greater chance of survival. Such microfluidic devices91 for culturing microorganisms give hope to the possibility of adding novel microbial species/strains to culture collections.
Sophisticated methods for making ‘cocktails’ of microorganisms for host inoculation are likewise essential. In medical science, fae-cal microbiome transplantation has been recognized as a promising therapy for otherwise intractable disease31 (for example, ulcerative colitis). Analogous approaches for crop plants7 would involve devel-oping bulk soil transplants or extracted soil microbiome cocktails using centrifuge-based techniques, whose application to soybean has successfully allowed us to enrich symbiont cells from healthy plant tissue92,93 (Fig. 4a). Microbial cocktails could be further char-acterized and deployed using microdroplet techniques94, in which pairs or sets of microorganisms are isolated (Fig. 4b). To realize more complex environmental conditions optimized for bacterial and/or fungal co-cultures, a series of emerging technologies and devices (such as X-ray computed tomography, three-dimensional printing, photothermal etching and various types of microfluidic chips) will allow us to introduce spatial heterogeneity and chemical gradients at micrometer scales89,90.
In embedding core microbiomes in agroecosystems, automated systems for inoculating plants with selected microorganisms are required (Fig. 4c). Single-cell chemical phenotyping with Raman spectroscopy enables rapid screening of microbial species/strains from mixed microbial cultures95. The sorted microorganisms can be automatically inoculated to plants using the RootChip or lay-ered microfluidic devices90,96,97, which provide high-throughput platforms for placing sterilized seeds/seedlings on the culture/co-culture media of core microorganisms. Chemicals promoting suc-cessful establishment of core microbiomes in host tissue may be simultaneously supplied with such microfluidic devices as inspired by prebiotics, in which galacto-oligosaccharides are used to support Bifidobacterium populations inside host gut98. Core-microbiome-inoculated plants might then be introduced to agroecosystems using unmanned planting machines22.
Portfolios with multiple coresIn maximizing contributions of core microbiomes within agroeco-systems, biodiversity theory99,100 provides key insights. By inoculat-ing differential sets of core microorganisms to individual seeds/seedlings, for example, neighbouring plant individuals in an agro-ecosystem may attract more or less different compositions of func-tional microbial species in the environment (Fig. 4c). The resultant diversity in established microbiomes may work as insurance for crop production against abrupt weather events and/or the spread of specific pathogen strains at the ecosystem level, as predicted by ecological theories of portfolio effects100. Moreover, with high
microbial diversity within and around roots, soil phosphorous or nitrogen will be more efficiently supplied to plants as a con-sequence of functional complementarity among coexisting micro-organisms10,101. Accordingly, taking advantage of such functional complementarity and portfolio effects provided by microbiomes, we propose using multiple core microbiomes, which differ in taxo-nomic and/or functional compositions of constituent microorgan-isms, within a farmland.
When introducing microbial portfolios into real agroecosystems, theoretical studies will help us optimize spatial and temporal con-figurations of crop plant individuals differing in core microbiomes (Fig. 5a). Given that spatial distribution of multiple plant varieties within a farmland (for example, ‘multilines’ of wheat, oat, rice and potatos102) can reduce impacts of pathogen outbreaks103, the best layout of multiple core microbiomes may be explored by simulation studies to prevent the spread of infectious disease. For example, it would be important to examine how spatial autocorrelations in core microbiome structures can determine agroecosystem-level resis-tance to pathogen outbreaks. Likewise, considering that rotation of crop plant species is a key technique for suppressing pathogen and pest populations and avoiding soil nutrient imbalance99,104, studies exploring best microbiome rotation are prospective. Ecosystem-scale models of disease and nutrient dynamics will help us design spatiotemporal configurations of multiple core microbiomes, as well as those of multiple crop plant species and genotypes105, in agroecosystem management.
Smart farming with aI and robotsEven if selected and designed core microbiomes are successfully introduced to agroecosystems, they may occasionally be replaced with unfavourable microbiomes after environmental perturbation (Fig. 5b). Moreover, in continual cropping systems, microbiomes may be easily altered due to outbreaks of native infectious micro-organisms adapted to specific host plants. Shifts to disease-related microbiome types (‘dysbiosis’) have been also reported in medical science106. In patients with diarrhoea and inflammatory bowel dis-ease, for example, gut microbiomes are characterized by decreased taxonomic diversity and by a disproportionate increase of virulent bacteria (for example, Clostridium difficile) after antibiotic treat-ments or other kinds of host physiological perturbation106.
Innovations in data-driven medicine are expected to provide platforms for predicting and preventing dysbiosis. Artificial intel-ligence (AI) based on neural networks, which can outperform expe-rienced radiologists in interpreting medical (for example, X-ray) images107, has come to integrate genome, epigenome, immunome, metabolome and microbiome data—also known as ‘high-definition medicine’, such AI-based methods show ways to prevent and delay disease onset in human individuals108. Developing such multi-omics diagnostic platforms for smart farming (or site-adapted agriculture) is a prerequisite for detecting ‘early warning signals’109 of microbi-ome dysbiosis and related plant physiological disorder in agroeco-systems23. As Newton’s second law was re-discovered by supplying AI with motion-tracking data of various physical systems110, under-
a
b c
Spatial and temporal configuration of core microbiomes in agroecosystems
Detecting early-warning signals of dysbiosis Mobile DNA sequencer(e.g. Oxford Nanopore
SmidgION)
Earlyassembly
Alternativestable state I
Alternativestable state II
Simulating agroecosystemdynamics with AI or
empirical dynamic modelling
Dysbiosis
Time
Tax
onom
ic d
iver
sity
of s
ymbi
onts
Monitoring dataTime Time
Time
Observeddynamics
Predicteddynamics
Predicted dysbiosisat location X
Chemical treatmentsor supplemental
inoculation ofcore microbiomes
Transferingdata
Year 1N Year 2 Year 3
UAVs
Fig. 5 | agroecosystem management. a, Spatiotemporal configurations of core microbiomes. Theoretical studies for optimizing spatial and temporal configurations of core microbiomes may help us prevent the spread and evolution of pathogens in agroecosystems. b, Predicting dysbiosis. The taxonomic diversity of plant-associated microorganisms will increase and reach a plateau during initial assembly, while it can drop in dysbiosis events having deleterious effects on hosts. Time-series analytical methods such as empirical dynamic modelling73,78 may allow us to forecast future dysbiosis. c, Forecasting microbiome dynamics in agroecosystems. Unmanned aerial vehicles (UAVs) and mobile DNA sequencers enable automated monitoring of plant physiology and microbiome structure, providing a basis for simulating agroecosystem dynamics. Regions indicated in red in the UAV image of farmland indicate the potato plants damaged by Phytophthora oomycetes112. Such monitoring and informatics platforms will help in exploring the best timing for introducing crop plants inoculated with core microbiomes. When dysbiosis is forecasted in such smart-farming systems, chemical treatments or other types of operations at specific timing and locations may be feasible with fleet management of (multiple) micro-UAVs.
lying dynamics of microbiomes may be revealed by machine learn-ing algorithms that integrate time-series observational data of plants in field conditions111. Nonetheless, time-series data are much less available in agriculture than in medical science in terms of both quantity and quality; accumulating standardized data with high temporal resolution is a prerequisite for AI-based management of agroecosystem states.
Albeit challenging, emerging technologies for field monitoring of plant physiology and microbiome structure will help us develop platforms for smart farming in the next 5–10 years (Fig. 5c). Imagery from unmanned aerial vehicles (UAVs), for example, now allows physiological phenotyping and disease monitoring at the individual-plant level, revolutionizing the spatial and temporal res-olution of field data112,113. Such UAVs can be equipped with mobile phones and a series of portable analytical devices, providing remote sensing data114. In particular, when combined with mobile DNA/RNA sequencers, UAV-based systems might allow us to monitor microbiomes and transcriptomes of plants under field conditions. Expression profiles of phosphate transporter genes, for instance, may be used as indicators of both nutritional and immune condi-tions of plants12,19. By unifying multiple lines of those monitoring data, AI-based informatics (for example, several machine learning methods111 and transfer learning115) and novel mathematical meth-ods for tracking biological community stability75 will allow us to explore best timing for introducing plants with core microbiomes and to forecast disease outbreaks in agroecosystems (Fig. 5c). Some of those technologies are ready for application in agroecosystems, however, integrating all the different components into a successful interdisciplinary approach is still a fundamental challenge.
ConclusionsHere we have focused on microbiome assembly in early stages of plant development and have outlined informatics pipelines for designing core microbiomes, which are expected to optimize inter-actions between plants and indigenous microorganisms. We have also discussed how the strategy of optimizing microbial functions at the agroecosystem level will be made feasible by the integration of different lines of cutting-edge technologies. Such interdisciplin-ary approaches of ecosystem-level management have the potential to maximize the assets, and limit the damages, of the 450-million-year history of coevolution between land plants and their associated microorganisms116. While native biota are often treated as risk fac-tors potentially causing disease outbreaks in agroecosystems, plants in natural ecosystems have also evolved means for maximizing benefits of microbiomes117. Core microbiome technologies will help crop plants reactivate immune and signalling pathways, and further capitalize on microbial services. Although economies of scale have historically favoured simplified management systems of uniform monoculture stands, it is becoming technologically and economi-cally feasible to introduce various types of spatial and temporal heterogeneity to agroecosystems with the aid of automated plant-ing machinery22. Optimizing spatiotemporal configurations of both plant genetic varieties and core microbiomes will be the key to man-aging resource-efficient and pathogen-resistant agroecosystems.
Published online: 30 April 2018
References 1. Berendsen, R. L., Pieterse, C. M. & Bakker, P. A. The rhizosphere
microbiome and plant health. Trends Plant Sci. 17, 478–486 (2012). 2. de Vries, F. & Wallenstein, M. Below-ground connections underlying
above-ground food production: a framework for optimising ecological connections in the rhizosphere. J. Ecol. 105, 913–920 (2017).
3. Busby, P. E. et al. Research priorities for harnessing plant microbiomes in sustainable agriculture. PLOS Biol. 15, e2001793 (2017).
4. Elser, J. & Bennett, E. Phosphorus cycle: a broken biogeochemical cycle. Nature 478, 29–31 (2011).
5. Callaway, E. Devastating wheat fungus appears in Asia for first time. Nature 532, 421–422 (2016).
6. Howden, S. M. et al. Adapting agriculture to climate change. Proc. Natl Acad. Sci. USA 104, 19691–19696 (2007).
7. Dangl, J. L., Horvath, D. M. & Staskawicz, B. J. Pivoting the plant immune system from dissection to deployment. Science 341, 746–751 (2013).
8. Robertson, G. P. & Vitousek, P. M. Nitrogen in agriculture: balancing the cost of an essential resource. Ann. Rev. Env. Res. 34, 97–125 (2009).
9. Tilman, D., Balzer, C., Hill, J. & Befort, B. L. Global food demand and the sustainable intensification of agriculture. Proc. Natl Acad. Sci. USA 108, 20260–20264 (2011).
10. van der Heijden, M. G. et al. Mycorrhizal fungal diversity determines plant biodiversity, ecosystem variability and productivity. Nature 396, 69–72 (1998).
11. Bonfante, P. & Anca, I.-A. Plants, mycorrhizal fungi, and bacteria: a network of interactions. Ann. Rev. Microbiol. 63, 363–383 (2009).
12. Hiruma, K. et al. Root endophyte Colletotrichum tofieldiae confers plant fitness benefits that are phosphate status dependent. Cell 165, 464–474 (2016).
13. Mendes, R. et al. Deciphering the rhizosphere microbiome for disease-suppressive bacteria. Science 332, 1097–1100 (2011).
14. Zavala-Gonzalez, E. A. et al. Arabidopsis thaliana root colonization by the nematophagous fungus Pochonia chlamydosporia is modulated by jasmonate signaling and leads to accelerated flowering and improved yield. New Phytol. 213, 351–364 (2017).
15. Calvo, P., Nelson, L. & Kloepper, J. W. Agricultural uses of plant biostimulants. Plant Soil 383, 3–41 (2014).
16. Arnold, A. E. et al. Fungal endophytes limit pathogen damage in a tropical tree. Proc. Natl Acad. Sci. USA 100, 15649–15654 (2003).
17. Vorholt, J. A., Vogel, C., Carlström, C. I. & Müller, D. B. Establishing causality: opportunities of synthetic communities for plant microbiome research. Cell Host Microbe 22, 142–155 (2017).
18. Robertson, G. P. et al. Cellulosic biofuel contributions to a sustainable energy future: choices and outcomes. Science 356, eaal2324 (2017).
19. Castrillo, G. et al. Root microbiota drive direct integration of phosphate stress and immunity. Nature 543, 513–518 (2017).
20. Jacott, C. N., Murray, J. D. & Ridout, C. J. Trade-offs in arbuscular mycorrhizal symbiosis: disease resistance, growth responses and perspectives for crop breeding. Agronomy 7, 75 (2017).
21. Oldroyd, G. E. Speak, friend, and enter: signalling systems that promote beneficial symbiotic associations in plants. Nat. Rev. Microbiol. 11, 252–263 (2013).
22. King, A. The future of agriculture. Nature 544, 21–23 (2017). 23. Schlaeppi, K. & Bulgarelli, D. The plant microbiome at work.
Mol. Plant-Microbe Int. 28, 212–217 (2015). 24. Girlanda, M. et al. Impact of biocontrol Pseudomonas fluorescens CHA0 and
a genetically modified derivative on the diversity of culturable fungi in the cucumber rhizosphere. Appl. Env. Micriobiol. 67, 1851–1864 (2001).
25. Streeter, J. G. Failure of inoculant rhizobia to overcome the dominance of indigenous strains for nodule formation. Can. J. Microbiol. 40, 513–522 (1994).
26. Castro-Sowinski, S., Herschkovitz, Y., Okon, Y. & Jurkevitch, E. Effects of inoculation with plant growth-promoting rhizobacteria on resident rhizosphere microorganisms. FEMS Microbiol. Lett. 276, 1–11 (2007).
27. Walsh, U. et al. Residual impact of the biocontrol inoculant Pseudomonas fluorescens F113 on the resident population of rhizobia nodulating a red clover rotation crop. Microb. Ecol. 45, 145–155 (2003).
28. Lloyd-Price, J. et al. Strains, functions and dynamics in the expanded Human Microbiome Project. Nature 550, 61–66 (2017).
29. Bashan, A. et al. Universality of human microbial dynamics. Nature 534, 259–262 (2016).
30. Lozupone, C. A., Stombaugh, J. I., Gordon, J. I., Jansson, J. K. & Knight, R. Diversity, stability and resilience of the human gut microbiota. Nature 489, 220–230 (2012).
31. Paramsothy, S. et al. Multidonor intensive faecal microbiota transplantation for active ulcerative colitis: a randomised placebo-controlled trial. Lancet 389, 1218–1228 (2017).
32. Müller, D. B., Vogel, C., Bai, Y. & Vorholt, J. A. The plant microbiota: systems-level insights and perspectives. Ann., Rev. Genet. 50, 211–234 (2016).
33. Toju, H., Yamamoto, S., Tanabe, A. S., Hayakawa, T. & Ishii, H. S. Network modules and hubs in plant-root fungal biomes. J. R. Soc. Interface 13, 20151097 (2016).
34. Bulgarelli, D. et al. Structure and function of the bacterial root microbiota in wild and domesticated barley. Cell Host Microbe 17, 392–403 (2015).
35. Marasco, R., Rolli, E., Fusi, M., Michoud, G. & Daffonchio, D. Grapevine rootstocks shape underground bacterial microbiome and networking but not potential functionality. Microbiome 6, 3 (2018).
36. Edwards, J. et al. Structure, variation, and assembly of the root-associated microbiomes of rice. Proc. Natl Acad. Sci. USA 112, 911–920 (2015).
37. Hartman, K. et al. Cropping practices manipulate abundance patterns of root and soil microbiome members paving the way to smart farming. Microbiome 6, 14 (2018).
38. Arumugam, M. et al. Enterotypes of the human gut microbiome. Nature 473, 174–180 (2011).
39. Knights, D. et al. Rethinking “enterotypes”. Cell Host Microbe 16, 433–437 (2014).
40. Scheffer, M., Carpenter, S., Foley, J. A., Folke, C. & Walker, B. Catastrophic shifts in ecosystems. Nature 413, 591–596 (2001).
41. Beisner, B. E., Haydon, D. T. & Cuddington, K. Alternative stable states in ecology. Front. Ecol. Env. 1, 376–382 (2003).
42. Kim, S.-W. et al. Robustness of gut microbiota of healthy adults in response to probiotic intervention revealed by high-throughput pyrosequencing. DNA Res. 20, 241–253 (2013).
43. Dethlefsen, L. & Relman, D. A. Incomplete recovery and individualized responses of the human distal gut microbiota to repeated antibiotic perturbation. Proc. Natl Acad. Sci. USA 108, 4554–4561 (2011).
44. Ng, K. M. et al. Microbiota-liberated host sugars facilitate post-antibiotic expansion of enteric pathogens. Nature 502, 96–99 (2013).
45. Hacquard, S. et al. Microbiota and host nutrition across plant and animal kingdoms. Cell Host Microbe 17, 603–616 (2015).
46. Werner, G. D. & Kiers, E. T. Order of arrival structures arbuscular mycorrhizal colonization of plants. New Phytol. 205, 1515–1524 (2015).
47. Braun-Kiewnick, A., Jacobsen, B. J. & Sands, D. C. Biological control of Pseudomonas syringae pv. syringae, the causal agent of basal kernel blight of barley, by antagonistic Pantoea agglomerans. Phytopathology 90, 368–375 (2000).
48. Fukami, T. Historical contingency in community assembly: integrating niches, species pools, and priority effects. Ann. Rev. Ecol. Evol. Syst. 46, 1–23 (2015).
49. Wei, Z. et al. Trophic network architecture of root-associated bacterial communities determines pathogen invasion and plant health. Nat. Commun. 6, 8413 (2015).
50. Weller, D. M. Biological control of soilborne plant pathogens in the rhizosphere with bacteria. Ann. Rev. Phytopathol. 26, 379–407 (1988).
51. Morris, C. E. & Monier, J.-M. The ecological significance of biofilm formation by plant-associated bacteria. Ann. Rev. Phytopathol. 41, 429–453 (2003).
52. Pieterse, C. M. et al. Induced systemic resistance by beneficial microbes. Ann. Rev. Phytopathol. 52, 347–375 (2014).
53. Wrather, J. A. & Koenning, S. R. Effects of diseases on soybean yields in the United States 1996 to 2007. Plant Health Prog. https://doi.org/10.1094/PHP-2009-0401-01-RS (2009).
54. Freilich, S. et al. Competitive and cooperative metabolic interactions in bacterial communities. Nat. Commun. 2, 589 (2011).
55. Masciarelli, O., Llanes, A. & Luna, V. A new PGPR co-inoculated with Bradyrhizobium japonicum enhances soybean nodulation. Microbiol. Res. 169, 609–615 (2014).
56. Cassan, F. et al. Azospirillum brasilense Az39 and Bradyrhizobium japonicum E109, inoculated singly or in combination, promote seed germination and early seedling growth in corn (Zea mays L.) and soybean (Glycine max L.). Eur. J. Soil Biol. 45, 28–35 (2009).
57. Paredes, S. H. et al. Design of synthetic bacterial communities for predictable plant phenotypes. PLOS Biol. 16, e2003962 (2018).
58. Lundberg, D. S. et al. Defining the core Arabidopsis thaliana root microbiome. Nature 488, 86–90 (2012).
59. Barea, J.-M., Pozo, M. J., Azcon, R. & Azcon-Aguilar, C. Microbial co-operation in the rhizosphere. J. Exp. Bot. 56, 1761–1778 (2005).
60. Agler, M. T. et al. Microbial hub taxa link host and abiotic factors to plant microbiome variation. PLOS Biol. 14, e1002352 (2016).
61. Faust, K. & Raes, J. Microbial interactions: from networks to models. Nat. Rev. Microbiol. 10, 538–550 (2012).
62. Coyte, K. Z., Schluter, J. & Foster, K. R. The ecology of the microbiome: networks, competition, and stability. Science 350, 663–666 (2015).
63. Newman, M. E. J. Networks: an Introduction (Oxford University Press, New York, 2010).
64. Toju, H. et al. Species-rich networks and eco-evolutionary synthesis at the metacommunity level. Nat. Ecol. Evol. 1, 0024 (2017).
65. Bai, Y. et al. Functional overlap of the Arabidopsis leaf and root microbiota. Nature 528, 364–369 (2015).
66. Layeghifard, M., Hwang, D. M. & Guttman, D. S. Disentangling interactions in the microbiome: a network perspective. Trends Microbiol. 25, 217–228 (2017).
67. Hartman, K., van der Heijden, M. G., Roussely-Provent, V., Walser, J.-C. & Schlaeppi, K. Deciphering composition and function of the root microbiome of a legume plant. Microbiome 5, 2 (2017).
68. Bianciotto, V. et al. An obligately endosymbiotic mycorrhizal fungus itself harbors obligately intracellular bacteria. Appl. Env. Micriobiol. 62, 3005–3010 (1996).
69. Behie, S. W. et al. Carbon translocation from a plant to an insect-pathogenic endophytic fungus. Nat. Commun. 8, 14245 (2017).
70. Langille, M. G. et al. Predictive functional profiling of microbial communities using 16S rRNA marker gene sequences. Nat. Biotech. 31, 814–821 (2013).
71. Callahan, B. J. et al. DADA2: high-resolution sample inference from Illumina amplicon data. Nat. Methods 13, 581–583 (2016).
72. Delgado-Baquerizo, M. et al. A global atlas of the dominant bacteria found in soil. Science 359, 320–325 (2018).
73. Deyle, E. R., May, R. M., Munch, S. B. & Sugihara, G. Tracking and forecasting ecosystem interactions in real time. P. Roy. Soc. Lond. B Bio. 283, 20152258 (2016).
74. Sugihara, G. et al. Detecting causality in complex ecosystems. Science 338, 496–500 (2012).
75. Ushio, M. et al. Fluctuating interaction network and time-varying stability of a natural fish community. Nature 554, 360–363 (2018).
76. Suzuki, K., Yoshida, K., Nakanishi, Y. & Fukuda, S. An equation-free method reveals the ecological interaction networks within complex microbial ecosystems. Methods Ecol. Evol. 8, 1774–1785 (2017).
77. Schreiber, T. Measuring information transfer. Phys. Rev. Lett. 85, 461–464 (2000).
78. Chang, C.-W., Ushio, M. & Hsieh, C.-H. Empirical dynamic modeling for beginners. Ecol. Res. 32, 785–796 (2017).
79. Vandeputte, D. et al. Quantitative microbiome profiling links gut community variation to microbial load. Nature 551, 507–511 (2017).
80. Smets, W. et al. A method for simultaneous measurement of soil bacterial abundances and community composition via 16S rRNA gene sequencing. Soil Biol. Biochem. 96, 145–151 (2016).
81. Nguyen, N. H. et al. FUNGuild: an open annotation tool for parsing fungal community datasets by ecological guild. Fungal Ecol. 20, 241–248 (2016).
82. van der Heijden, M. G., de Bruin, S., Luckerhoff, L., van Logtestijn, R. S. & Schlaeppi, K. A widespread plant-fungal-bacterial symbiosis promotes plant biodiversity, plant nutrition and seedling recruitment. ISME J. 10, 389–399 (2016).
83. Sheth, R. U., Cabral, V., Chen, S. P. & Wang, H. H. Manipulating bacterial communities by in situ microbiome engineering. Trends Genet. 32, 189–200 (2016).
84. Mee, M. T., Collins, J. J., Church, G. M. & Wang, H. H. Syntrophic exchange in synthetic microbial communities. Proc. Natl Acad. Sci. USA 111, 2149–2156 (2014).
85. Fondi, M. & Liò, P. Multi-omics and metabolic modelling pipelines: challenges and tools for systems microbiology. Microb. Res. 171, 52–64 (2015).
86. Peay, K. G. et al. Convergence and contrast in the community structure of Bacteria, Fungi and Archaea along a tropical elevation–climate gradient. FEMS Microbiol. Ecol. 93, 5 (2017).
87. Narisawa, K., Hambleton, S. & Currah, R. S. Heteroconium chaetospira, a dark septate root endophyte allied to the Herpotrichiellaceae (Chaetothyriales) obtained from some forest soil samples in Canada using bait plants. Mycoscience 48, 274–281 (2007).
88. Usuki, F. & Narisawa, K. A mutualistic symbiosis between a dark septate endophytic fungus, Heteroconium chaetospira, and a nonmycorrhizal plant, Chinese cabbage. Mycologia 99, 175–184 (2007).
89. Aleklett, K. et al. Build your own soil: exploring microfluidics to create microbial habitat structures. ISME J. 12, 312–319 (2018).
90. Li, R., Lv, X., Zhang, X., Saeed, O. & Deng, Y. Microfluidics for cell-cell interactions: A review. Front. Chem. Sci. Eng. 10, 90–98 (2016).
91. Nichols, D. et al. Use of ichip for high-throughput in situ cultivation of “uncultivable” microbial species. Appl. Env. Micriobiol. 76, 2445–2450 (2010).
92. Ikeda, S. et al. Development of a bacterial cell enrichment method and its application to the community analysis in soybean stems. Microb. Ecol. 58, 703–714 (2009).
93. Ikeda, S. et al. Community-and genome-based views of plant-associated bacteria: plant–bacterial interactions in soybean and rice. Plant Cell Physiol. 51, 1398–1410 (2010).
94. Hosokawa, M. et al. Droplet-based microfluidics for high-throughput screening of a metagenomic library for isolation of microbial enzymes. Biosens. Bioelectr. 67, 379–385 (2015).
95. Song, Y., Yin, H. & Huang, W. E. Raman activated cell sorting. Curr. Opin. Chem. Biol. 33, 1–8 (2016).
96. Grossmann, G. et al. The RootChip: an integrated microfluidic chip for plant science. Plant Cell 23, 4234–4240 (2011).
97. Massalha, H., Korenblum, E., Malitsky, S., Shapiro, O. H. SpringerAmpamp; Aharoni, A. Live imaging of root–bacteria interactions in a microfluidics setup. Proc. Natl Acad. Sci. USA 114, 4549–4554 (2017).
98. Monteagudo-Mera, A. et al. High purity galacto-oligosaccharides enhance specific Bifidobacterium species and their metabolic activity in the mouse gut microbiome. Benef. Microb. 7, 247–264 (2016).
99. Tilman, D., Cassman, K. G., Matson, P. A., Naylor, R. & Polasky, S. Agricultural sustainability and intensive production practices. Nature 418, 671–677 (2002).
100. Yachi, S. & Loreau, M. Biodiversity and ecosystem productivity in a fluctuating environment: the insurance hypothesis. Proc. Natl Acad. Sci. USA 96, 1463–1468 (1999).
101. Maherali, H. & Klironomos, J. N. Influence of phylogeny on fungal community assembly and ecosystem functioning. Science 316, 1746–1748 (2007).
102. Mundt, C. Use of multiline cultivars and cultivar mixtures for disease management. Ann. Rev. Phytopathol. 40, 381–410 (2002).
103. Suzuki, S. U. & Sasaki, A. How does the resistance threshold in spatially explicit epidemic dynamics depend on the basic reproductive ratio and spatial correlation of crop genotypes? J. Theor. Biol. 276, 117–125 (2011).
104. Isbell, F. et al. Benefits of increasing plant diversity in sustainable agroecosystems. J. Ecol. 105, 871–879 (2017).
105. Prieto, I. et al. Complementary effects of species and genetic diversity on productivity and stability of sown grasslands. Nat. Plants 1, 15033 (2015).
106. Pham, T. A. N. & Lawley, T. D. Emerging insights on intestinal dysbiosis during bacterial infections. Curr. Opin. Microbiol. 17, 67–74 (2014).
107. Shen, D., Wu, G. & Suk, H.-I. Deep learning in medical image analysis. Ann. Rev. Biomed. Eng. 19, 221–248 (2017).
108. Torkamani, A., Andersen, K. G., Steinhubl, S. R. & Topol, E. J. High-definition medicine. Cell 170, 828–843 (2017).
109. Scheffer, M., Carpenter, S. R., Dakos, V. & van Nes, E. H. Generic indicators of ecological resilience: Inferring the chance of a critical transition. Ann. Rev. Ecol. Evol. Syst. 46, 145–167 (2015).
110. Schmidt, M. & Lipson, H. Distilling free-form natural laws from experimental data. Science 324, 81–85 (2009).
111. Murphy, K. P. Machine Learning: a Probabilistic Perspective (MIT press, London, 2012).
112. Sugiura, R. et al. Field phenotyping system for the assessment of potato late blight resistance using RGB imagery from an unmanned aerial vehicle. Biosyst. Eng. 148, 1–10 (2016).
113. Araus, J. L. & Cairns, J. E. Field high-throughput phenotyping: the new crop breeding frontier. Trends Plant Sci. 19, 52–61 (2014).
114. Quesada-González, D. & Merkoçi, A. Mobile phone-based biosensing: an emerging “diagnostic and communication” technology. Biosens. Bioelectr. 92, 549–562 (2017).
115. Pan, S. J. & Yang, Q. A survey on transfer learning. IEEE Trans. Knowl. Data Eng. 22, 1345–1359 (2010).
116. Brundrett, M. C. Coevolution of roots and mycorrhizas of land plants. New Phytol. 154, 275–304 (2002).
117. Foster, K. R., Schluter, J., Coyte, K. Z. & Rakoff-Nahoum, S. The evolution of the host microbiome as an ecosystem on a leash. Nature 548, 43–51 (2017).
118. Franzosa, E. A. et al. Relating the metatranscriptome and metagenome of the human gut. Proc. Natl Acad. Sci. USA 111, 2329–2338 (2014).
acknowledgementsWe thank Takashi Akagi and three anonymous reviewers for their insightful comments on the manuscript. This work was financially supported by JSPS KAKENHI Grant (26711026), JST PRESTO (JPMJPR16Q6), and the Funding Program for Next Generation World-Leading Researchers of Cabinet Office, the Government of Japan (GS014) to H.T, DOE Award DE-SC0016097 to KGP, and by a European Research Council Grant (335542) to E.T.K.
author contributionsH.T. designed the study and wrote the first draft. H.T. and E.T.K. edited the final version of the manuscript based on discussion with all the authors.
Competing interestsThe authors declare no competing interests.
additional informationReprints and permissions information is available at www.nature.com/reprints.
Correspondence should be addressed to H.T.
Publisher’s note: Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
