Root renovation: how an improved understanding of basic root biology could inform the development of elite crops that foster sustainable soil health

Additional keywords: effectors, meta-analysis, plant breeding, plant-microbial interaction, rhizosphere, root exudates, sequencing, small RNA.

Plant roots are major architects of the soil, and are a key focus of plant breeding aimed at developing more sustainable crops (Kell 2011). Through sheer physical strength, roots sift through the soil making it more porous to air and water while also reinforcing it against erosion (Gyssels et al. 2005). The deeper the roots, the more water that is transported through the soil column whereas more fine and shallow roots impact nutrient use by the plant (Callesen et al. 2016). Roots also secrete different compounds that alter the chemical make-up of the soil. This can alter soil aggregate size and can solubilise fertilisers, both attributes linked to increased health and vigour of plants (Keiluweit et al. 2015; Rillig et al. 2015; Jin et al. 2017). Due to these clear benefits, a great deal of effort has been expended with the aim of breeding crops that have specific root architectures. This key investment by the global research community has resulted in the production of new crop varieties that are less reliant on irrigation, fertilisation and pesticide application while also increasing the yields and profitability to growers (Bardgett et al. 2014; Ju et al. 2015; Meng and Yao 2015).

Plant roots, however, do not grow in a sterile substrate in natural environments, nor are they alone in supporting plant health and fitness. An overarching theme that emerges from ecology is the central role of the interaction between plant roots and soil microbes (used as an umbrella term here to include fungi, oomycetes and bacteria) in determining how plants grow and react to different environmental stressors. For this reason, rather than discussing plants and microbes separately, some have begun using the term ‘holobiont’. This term includes the biomolecular network of a host organism and its attendant microbiome, when referring to plants and the organisms with which they interact (Bordenstein and Theis 2015). The diversity of the microbial community encapsulated within the holobiont is known to correlate significantly with plant performance (van der Heijden et al. 2008; Prober et al. 2015), but identifying and interpreting the impact of individual microbes within this community is complicated by the fact that each organism can directly or indirectly affect plant health. One broad class of soil microbes that associate with plants, dubbed ‘beneficial/mutualistic’ microbes, recycle and alter soil nutrients into a format usable by plants, act as nutrient and signalling conduits between plants, break down or immobilise toxic compounds and boost plant immunity to devastating diseases. Conversely, another class of microbes are defined as ‘pathogenic/parasitic’ due to their ability to funnel nutrients away from plants and/or to release toxic compounds that cause disease and reduce yield. One approach to significantly and sustainably increase the health and productivity of our crops is to breed crops with root traits that encourage and harness the population of beneficial soil microbes while simultaneously resisting pathogenic microbes.

With respect to manipulating plant–microbe interactions, research into disease resistant plants and rootstocks has been a priority within the research community. Plant pathogens are a persistent and integral part of agroecosystems that are constantly co-evolving with crops and are one of the major causal agents of crop/forestry productivity losses worldwide. Studies have estimated that the annual damage and control costs for plant pathogens are over $20B in crop production and $7B in forestry (Pimentel et al. 2005; Murray 2009). Using conventional breeding and genetic technologies, we now have a wide range of disease resistant crops that have led to reductions in the use of chemical control agents (Zhu et al. 2000; Phipps and Park 2002). Such newly bred lines, however, come with an ecological cost. Studies show that domestication and breeding for disease resistance has negatively impacted the diversity of soil microbes and curtailed the ability of certain crops to associate with beneficial microbes (Germida and Siciliano 2001; Mutch and Young 2004; Sangabriel-Conde et al. 2015; Szoboszlay et al. 2015; Plett et al. 2016). This effect may be due in part to increased general plant immunity (Plett et al. 2016) as well as changes to root architecture and communication with microbes through altered root exudation (Pérez-Jaramillo et al. 2016). The resultant loss of biodiversity is a critical problem facing agroecosystems: should microbial diversity in agricultural soils decrease to such an extent that functional redundancy between different microbial taxa is lost, then further loss of any one functional group may severely impact fertility in these systems. This view is supported by meta-analyses such as Nielsen and colleagues (2011) and modelling by other groups such as Hunt and Wall (2002). Furthermore, as the presence of certain microbiota can act as an ecological barrier to pathogenic organisms or to adverse abiotic environmental conditions, loss of these organisms will have both a direct and indirect impact on plant health (Petersen et al. 2002; Mendes et al. 2011; van Elsas et al. 2012). One approach that has been undertaken to improve microbial diversity and health within soils is to alter soil management practices. Specifically, growers have been encouraged to intentionally inoculate soils with beneficial microbes that improve plant health through improved nutrition or through repression of disease (Abd-Alla et al. 2014; Shen et al. 2015; Xiong et al. 2017). These ‘bio-fertilisers’ or ‘bio-pesticides’ are the result of isolating one or two beneficial microbial strains from one soil and then mass-producing them for use in all agricultural situations. For some applications, such as the inoculation of legumes with rhizobial bacteria, this has worked well (Javaid 2009; Wiseman et al. 2009; Cozzolino et al. 2013). In other instances, as the inoculants are not properly adapted to all field conditions, some inoculation strategies have failed (Adholeya et al. 2005; Verbruggen et al. 2013; Owen et al. 2015). A further issue with such a strategy is that the inoculum typically needs to be re-applied on a regular basis, calling into question its sustainability. Another option is to use plants with discrete root traits that specifically encourage the growth of beneficial microbes already present in the soil, thereby increasing soil and plant health without the addition of microbial inoculants. To that end, could we use our current knowledge of root biology, and the physical and chemical traits of plant roots, to manipulate the balance of different microbial lifestyles within the soil, or do we need to produce new knowledge in this area to make such an endeavour feasible?

In order to obtain new crop varieties that maintain past value-added phenotypes (e.g. increased yield, disease resistance) as well as integrate new traits aimed at improving soil health through tailored plant–microbe interactions, we could continue to use traditional breeding practices. Plant lineages that have demonstrated high levels of colonisation by beneficial microbes could be crossed with crop varieties that are currently used within the industry and the progeny screened for the maintenance of agriculturally relevant phenotypes as well as the new, microbially associated phenotypes. However, as opposed to easily observable phenotypes such as plant height, flowering time or yield, assaying the status of plant–microbe interactions at the root–soil interface requires laborious sampling, processing and specialised personnel. If, however, we could identify basic genetic loci in plants that are linked to fostering beneficial plant–microbe interactions in the soil while also repressing pathogenic and parasitic microbes, we could generate a simple genetic screen to identify plants with value-added belowground attributes. The results obtained from such basic scientific research could then be integrated into traditional and molecular breeding practices to more holistically manage plant and soil health.

We posit that one of the key plant traits that we need to understand in order to populate such a panel of favourable genetic markers, is how plants identify and communicate with soil microbes. Several recent reviews have thoroughly covered the current state of our knowledge concerning how plants and microbes interact (Hacquard et al. 2017; Glazebrook and Roby 2018; Plett and Martin 2018). Therefore, in the present work we will only briefly highlight ongoing research into how plants perceive and communicate with soil microbes, and how microbes manipulate hosts during the colonisation process. We will then use these studies as a platform to identify areas of research and techniques that are necessary for us to significantly advance our understanding of how plant genetics and root traits could be used to impact soil microbial communities to better support sustainable plant vigour and productivity.

One of the first steps in advancing the ideal of developing crops that alter soil microbial communities to support plant growth and vigour is to understand how plants (a) perceive microbes, and (b) differentiate between ‘good’ and ‘bad’ microbes. Given the fact that plants survive when in contact with the immense diversity and complexity of the soil microbial community, plants must be able to both perceive and respond to the presence of microbial chemical signals or by-products in the rhizosphere (i.e. the soil within close proximity to roots). Based on the plant perception pathway(s) induced by these microbial signals, the plant should theoretically have dedicated perception mechanisms that stimulate distinct responses towards different groups/lifestyles of microbes (Fig. 1a). Examples of distinct pathways include the perception of nodulation signals from rhizobial bacteria and of pathogenic microbes separately by MtLYK9 and MtLYR4 (Bozsoki et al. 2017). In addition to these distinct perception pathways, other studies have demonstrated that some plant cell-wall based receptors can perceive signals from multiple microbes exhibiting different lifestyles (Fig. 1a). A study on Medicago truncatula Gaertn. demonstrated that NFP, a receptor for rhizobial bacterial Nod-factor signals, is also responsive to pathogenic signals (Rey et al. 2013). Similarly, the rice receptor OsCERK1 fosters symbiotic mycorrhizal infection while also controlling the induction of defence responses against pathogens (Miyata et al. 2014; Zhang et al. 2015). Despite these few comparative examples, the main foundation of our understanding of the microbial recognition process in plants is built upon one-on-one plant–microbe interactions. Therefore, the key question of how a plant distinguishes mutualists from pathogens in a mixed community remains unresolved.

Fig. 1.  A conceptual figure of the plant responses towards microbes of different lifestyles. (a) A simplified theoretical mechanism that plants may adopt to distinguish pathogens from mutualists. Within this simplified framework, detrimental organisms produce pathogen-specific signals that are perceived by unique plant receptors that activate defence responses, while beneficial microbes produce mutualist-specific signals that are perceived differently from pathogenic signals to activate symbiotic pathway(s) and/or suppress defence responses in plants. (b) The top two panels demonstrate the continuum of different lifestyle descriptions of microbes that interact with plants and their corresponding effect on plant health. The bottom panel of the figure depicts a possible overlapping perception mechanism that may be utilised in plants to identify commensal microbes whereby commensals or endophytes may produce a cocktail of signals that is neither purely pathogenic nor mutualistic. Such a mixture of signals would induce different plant pathways that, while partial colonisation of plant tissues is enabled, neither partner organism benefits nor does either partner show a detriment.

When trying to unravel this conundrum, however, we run into a philosophical problem; in nature most host–microbe interactions cannot be discretely categorised into beneficial (i.e. good) or pathogenic (i.e. bad). This is true for both inter- and intra-species analyses with certain microbes exhibiting a continuum of effects on plant hosts spanning from mutualistic to endophytic/commensal and pathogenic (Newton et al. 2010; Stergiopoulos and Gordon 2014). This is true for genera such as Fusarium sp. (Gordon and Martyn 1997; Ploetz 2006; Stergiopoulos and Gordon 2014), for mycorrhizal fungi (Hoeksema et al. 2010) and for a range of plant endophytic microbes (Porras-Alfaro and Bayman 2011). Therefore, the existence of an indistinct delineation between beneficial and pathogenic microbes may also mean that there is overlap in how both lifestyles of microbe interact with, and are perceived by, the plant (Fig. 1b). But how might this work? Does the plant sense a variety of signals from these enigmatic microbes and, through the integration of the intensity of each signal mount an appropriate response? How might we test such enigmatic perception mechanisms in crop plants? It is likely that the answer to this lies in combining several ‘-omics’ platforms in a systems biology/ecology approach.

One of the major impediments to identifying plant perception networks that might identify one specific class or lifestyle of microbe from another is that the majority of plant–microbe interaction studies are performed in isolation with one host and one microbe. Although such studies have been, and will continue to be, a necessary staple in studying the molecular aspects underpinning inter-kingdom interactions they do limit the broader applications of these findings (Crowther et al. 2018). Further, these studies may only use a narrow range of techniques and not combine multiple -omic approaches. Could we, however, leverage previously published work from a diversity of plant–microbe interactions and a range of experimental platforms to identify specific pathways and proteins whose role is to differentiate between the presence of beneficial and detrimental microbes? We feel that this is a feasible and under-explored research avenue. A scan of the PubMed Central database for proteomic, metabolomic, transcriptomic, and genomic papers on key plant–microbe interactions published in the last 10 years displays a wealth of possible datasets to mine for such analyses (Fig. 2). With recent advances in data analytical techniques, one way we could analyse these publically available data would be to conduct meta-analysis to distinguish molecular patterns involved in plant perception for distinct sets of microbes, be it different microbial classes (e.g. bacteria, fungi) or different functional groups (e.g. biotroph, saprotroph, hemitroph). A conceptual schematic workflow for such an endeavour using RNA sequencing data is presented in Fig. 3. In short, relevant datasets as those mined from PubMed Central, would be retrieved and curated based on their experimental parameters. From each dataset, a matrix of differentially expressed genes (DEGs) would be calculated and then integrated followed by several different analyses including comparative clustering analysis and network analyses (Fig. 3). A current roadblock to performing such meta-analysis is how to overcome the variations among different datasets arising from technical aspects (different protocols and technological platforms), statistical aspects (different data structure and distribution) and biological aspects (genetic variation). Should these challenges be overcome, one could make use of robust statistical and computational methods to extract desirable patterns and trends hidden in the complex matrix of combined datasets. In recent years, unsupervised and supervised machine learning strategies have frequently been used to cluster and predict key patterns that classify different groups: for example, identifying key volatile signatures amongst human pathogenic microbes (Palma et al. 2018), key genomic factors related to the lifestyles of plant microbiome bacteria (Martínez‐García et al. 2016) as well as genetic pathways related to biotic stress from those related to abiotic stress (Shaik and Ramakrishna 2014). The possibilities for using such a methodology are far-ranging and could be used to test new hypotheses by reusing public datasets, overcoming the issues of small sample sizes and increasing the statistical power of the results (Camacho et al. 2018). The results of these meta-analyses could then be experimentally verified in planta using transgenically altered plants or systemically integrated biosensors as in Pini et al. (2017). The outcome of these studies would be a greatly refined list of genetic traits in plants that enable them to perceive microbes within the environment and, potentially, also shed light on if and how plants are able to differentiate between beneficial, commensal, and detrimental microbes.

Fig. 2.  A search of the PubMed Central online database for potentially relevant papers useful for initial consideration within a meta-analysis from the years 2008–18. An example of the general query format utilised to pull the relevant data was: ‘metabol* [All Fields] AND rhizob* [All Fields] AND host [All Fields] AND plant [All Fields] NOT review [All Fields]’. For metabolomics we utilised metabol*, for proteomics we used the search term ‘proteo*’, for transcriptomics we used the search term ‘transcriptome*’ and for genomics we used ‘genom*’; all bars are differentially coloured to distinguish between the four techniques. The Venn diagram above each set of bars indicates the number of publications that integrate all four methodologies within the search terms. With regards to the microbial models used in the search, for rhizobial bacteria we used ‘rhizob*’, for arbuscular mycorrhizal fungi we used ‘arbusc*’, for Pseudomonas we used ‘Pseudomon*’ while excluding common human pathogens, for Bacillus we used ‘Bacillu*’, for Trichoderma we used ‘Trichoderm*’, for Phytophthora we used ‘Phytophth*’, for Magnaporthe we used ‘Magnaporth*’ for Botrytis, Melampsora and Ustilago we used the full name in the search. Data accessed 17 May 2018.

Fig. 3.  An example meta-analysis workflow for transcriptomic data. Within this workflow, previously published transcriptomic datasets correlated with a specific research question are collected and curated from public databases. These raw datasets are then normalised using the same data analysis package, in this instance expressed as differentially expressed genes (DEGs). Once a common ratio-metric rational such as DEGs are applied to all datasets, the data from different studies can be integrated and transformed into a common matrix. Finally this matrix can undergo meta-analysis with various approach such as simple clustering through to more complex network analysis.

The composition of root exudates changes upon perception of a given microbe. Root exudates then, in turn, impact different microbial genera based on their individual biology (Kamilova et al. 2006; Escudero et al. 2014; Guo et al. 2015). In effect, this shapes the composition of the microbial community in the rhizosphere (Broeckling et al. 2008; Coskun et al. 2017; Guo et al. 2017; Sasse et al. 2018). The identification of specific root signals that impact microbial communities in a manner that supports plant health, and an understanding of what controls the synthesis of these signals, is another avenue to generate plants that naturally foster a healthy microbial community.

Root exudates contain a mixture of substances including primary and secondary metabolites, nucleic acids, enzymes, mucilage, ions and small peptides (Bais et al. 2006). Several research studies have sought to identify the function of compounds in root exudates with regards to their influence on microbial communities. Such studies have had to overcome certain drawbacks including how metabolic content of root exudates vary depending on the plant species (Kidd et al. 2018), between cultivars of the same plant species (Guo et al. 2015; Mönchgesang et al. 2016b), the developmental stage of the plant (Gransee and Wittenmayer 2000; Chaparro et al. 2013), the part of root system involved (Pini et al. 2017), soil and other environmental conditions (Bowsher et al. 2015; Holz et al. 2017; Liu et al. 2017) and the microbial community in the soil (Guo et al. 2015). Several root exudate compounds are associated with functions related to defence against pathogens (Baetz and Martinoia 2014) and/or to symbiotic facilitation (Li et al. 2016). For example, sesquiterpenes and strigolactones enhance mutualistic symbiosis with arbuscular mycorrhizal fungi (AM fungi) (Akiyama et al. 2005; Akiyama and Hayashi 2006) and rhizobial bacteria (Steinkellner et al. 2007), but these chemical signals may also attract parasitic weeds (Akiyama and Hayashi 2006). Rutin, a flavonoid, promotes hyphal growth in the mycorrhizal genus Pisolithus sp. (Lagrange et al. 2001) and induces gene expression in other mutualistic fungi (Plett and Martin 2011), but it can also stimulate growth of the pathogenic fungi Alternaria alternata and Fusarium solani (Kalinova and Radova 2009). Therefore, the development of new innovations in both the identification of metabolites and the ability to assess their simultaneous impact on whole microbial communities and community function is needed.

Proteins also form a critical part of the substances secreted by the root. Protein secretome studies of plant roots reveal that several short peptides released from plant roots also participate in plant–microbial interaction, including to defend against pathogens and to facilitate beneficial symbionts interaction (Marmiroli and Maestri 2014). For example, a class of cysteine-rich, short-chain, predominantly cationic peptides known as ‘defensins’, act as chemical defences against pathogenic microbes. Interestingly, legumes produce nodule-specific cysteine-rich (NCR) peptides that resemble defensins and also have an antibacterial property. This antibacterial property appears to target bacteria other than beneficial rhizobia (Tiricz et al. 2013). Further, these NCR peptides modulate the differentiation of rhizobia into bacteroids, keep the bacteroids alive inside nodules, and govern nitrogen fixation efficiency (Van de Velde et al. 2010; Horváth et al. 2015; Montiel et al. 2017; Yang et al. 2017). It is likely that the differential effect of these peptides is based on structural differences; in Medicago, different NCR classes are induced by a pathogen rather than by a mutualistic mycorrhizal fungi (Uhe et al. 2018), and the number as well as diversity of NCR genes have been found to correlate with the diversity in rhizobial bacteroids that a given legume is able to host (Montiel et al. 2017). Aside from legume-rhizobia symbiosis, poplar roots produce secreted proteins that enter and accumulate in the nucleus of the ectomycorrhizal fungus Laccaria bicolor, impacting its hyphal growth and branching (Plett et al. 2017). Besides peptides, plants also actively exude amino acids directly to their rhizosphere, which can be induced by certain microbial signals (Phillips et al. 2004; Lesuffleur and Cliquet 2010). The direct exudation of l-amino acids are also thought to act as easily accessible carbon resources to promote microbial growth (Phillips et al. 2004) while their enantiomers, d-amino acids, are generally considered toxic and difficult to be metabolise, although their function remains uncertain (Hener et al. 2018).

In addition to secondary metabolites and peptide-based signalling molecules, emerging research is bringing new insights into a potentially novel class of signalling molecules in the form of RNA. Small RNAs (sRNAs) are short, noncoding RNAs that regulate the expression of target genes transcriptionally or post-transcriptionally (Ku et al. 2015). The machinery of sRNA regulation exists in eukaryotic organisms including animals, plants, fungi and oomycetes. In plants, different classes of sRNAs, specifically microRNAs (miRNAs) and small interfering RNAs (siRNAs) play important regulatory roles in plant immune systems (Padmanabhan et al. 2009). Although sRNAs are well established regulators of gene expression internally in animals, plants and fungi, it has been shown in recent years that sRNAs can also be exported into extracellular spaces (Cai et al. 2018; Tsatsaronis et al. 2018). sRNAs, also known as ‘cross-kingdom RNA-interference (RNAi)’, can be secreted by the plant and transported into fungal tissues and suppress gene expression and virulence in pathogens and have been demonstrated in barley, wheat and cotton plants (Nowara et al. 2010; Baulcombe 2015; Zhang et al. 2016b). To date, there is no evidence supporting the natural occurrence of sRNA trafficking between plant and soil-borne microbes, although host-induced gene silencing (HIGS) has been proven to be plausible in the rhizospheric environment; cotton, tomato and Arabidopsis plants with transgenetically-acquired ability to produce pathogen-targeting sRNA could resist infection of soil-borne fungal pathogen Verticillium dahliae (Zhang et al. 2016a; Song and Thomma 2018). Although the sRNA trafficking vectors were uncertain in the above studies, a new research study demonstrated in Arabidopsis that plant sRNAs travelled into fungal cells through secretion of exosome-like extracellular vesicles (Cai et al. 2018). Further, current studies on plant–microbial cross-kingdom RNAi focus on the plant–pathosystem aboveground. To our knowledge, it is unknown whether plants exude sRNAs to regulate their interaction with beneficial soil-borne microbes in a similar manner.

There is still a wide knowledge gap concerning the existence and role of cross-kingdom RNAi in the rhizosphere, but it seems that the specificity of sRNAs as signals within the rhizosphere are theoretically superior to the aforementioned metabolite signals as sRNAs bind to specific transcripts with complementary sequences. They therefore have the capacity to modulate plant–microbe interactions in a very specific manner. Furthermore, given that the half-lives of double-stranded sRNAs in soil are estimated to range between 14 and 30 h, sRNAs released into the soil are considered to have minimal risk to persist in the environment (Dubelman et al. 2014). For these reasons, the discovery of sRNA crosstalk has inspired innovation of new environmental-friendly sRNA-based fungicides that target specific fungal disease (Wang et al. 2016) and encouraged the use of HIGS in crop protection against soil-borne pathogens (Koch and Kogel 2014; Zhang et al. 2016a; Song and Thomma 2018).

Although published research suggests that root exudates (be they metabolites or proteins) help to recruit certain microbes while suppressing others, the specificity of these signalling molecules is often largely unknown. This is due to the fact that most studies seeking to characterise the role of a single chemical exudate are typically performed using a biosystem with a single plant species associated with a single microbial symbiont. Thus, the specificity of these signalling molecules in targeting a particular microbe/microbial group(s) within the rhizosphere is untested. To address this gap, research groups are beginning to attempt the characterisation of metabolic exudates directly in a natural community (Swenson et al. 2015, 2018; Trivedi et al. 2016). Typical approaches involve application of the chemical to the soil or use of a mutant plant that over-produces the desired signal followed by meta-transcriptomic and meta-genomic analysis of the microbial community (Yang et al. 2002; Liang et al. 2014; Nam et al. 2014). If this is performed on a synthetic community made up of microbes of known identity with fully sequenced genomes, this works well in understanding how the plant-based signal can affect multiple microbes simultaneously at the level of both microbial population turn-over and at the level of organismal gene expression. If, however, the soil to which the chemical signalling agent is applied contains a natural microbial community with many unknown or understudied members, it is very difficult with conventional meta-omic approaches to link differences in trancriptomic regulation to a specific microbe. This is because current fragment-based meta-omic approaches have difficulties in accounting for the presence of uncharacterised microbes that do not have sequenced genomes. Assembling microbial genomes with meta-omics data from environmental samples is difficult because of uneven genome coverage across the microbial community. In addition, amplicon sequencing approaches with phylogenetic markers such as 16S rRNA/internal transcribed spacer (ITS) are often associated with bias and ambiguity issues (Elbrecht and Leese 2015; Tedersoo et al. 2015; Tremblay et al. 2015). Therefore, a new approach is needed.

Single-cell sequencing has dramatically advanced the depth and resolution of genomic/transcriptomic information from a pool of a whole microbial community down to the individual cell. With single-cell sequencing techniques, cell-to-cell variation can be accounted for, despite the minuscule amount of DNA and RNA each cell contains (Kashtan et al. 2014; Buettner et al. 2015; Kolodziejczyk et al. 2015). Given that plant–microbial interactions in the rhizosphere can be highly variable spatiotemporally, single-cell sequencing is a desirable tool for understanding the molecular signalling networks at a cellular level and makes it possible to directly associate a transcriptional response to a discrete microbe when in the presence of a certain root exudate. Such systems have been trialled successfully with microbial cells and have demonstrated the great potential in differentiating variation within a microbial community no matter if the microbes present are widely studied (e.g. Pseudomonas sp.) or novel. Particularly, single-cell sequencing can effectively solve the genome assembly issues associated with unknown microbes commonly seen in traditional metagenomic studies. Given that it is often challenging to culture the majority of uncharacterised microbial species, single cell sequencing enables culture-free sequencing of genomes for these organisms (Stepanauskas 2015; Woyke et al. 2017). Several studies have demonstrated the use of single-cell sequencing directly on environmental samples to study uncharacterised microbial species in atypical microbial communities, including Oceanospirillales in oil spill-contaminated sea water (Mason et al. 2012), Hydrotalea in highly acidic mine drainage (Medeiros et al. 2017) and Poribacteria, endosymbionts in marine sponges (Siegl et al. 2011). However, the application of single-cell sequencing is restricted to a microbial community where cells can be isolated relatively easily (Woyke et al. 2017). One major limitation of single-cell sequencing applications on the rhizospheric community is the isolation and sorting for individual microbial cells-of-interest from the topologically complex soil-plant interface. Further investigation is needed to overcome this technological hurdle for applying single-cell sequencing in rhizospheric environments.

In addition to single cell genomics, other innovative protocols are being developed that can further dissect the response of individual members of the rhizosphere to root exudates at a cell-by-cell basis. For instance, the invention of single-cell mass spectrometry (MS) made real-time profiling of metabolites in individual plant cells possible (Fujii et al. 2015). This technology could be employed in understanding the cellular metabolic fluxes within microbes upon treatment with a root-based signalling molecule. Alternatively, a study by Pini et al. (2017) has made use of ‘bacterial biosensors’ – genetically modified bacteria with bioluminescent reporters specific to a range of metabolites to understand root exudation in real time. These bacteria were allowed to colonise pea roots in the rhizosphere and any metabolites that were exuded from the plant could be detected and visualised (Pini et al. 2017). Therefore, new technologies are becoming available that we can adapt to understand the identity and the impact of root exudates at both the level of the rhizosphere community level right down to the single cell. These studies will revolutionise our understanding of signalling within the rhizosphere and will identify specific root exudates that could be value-added traits in crops due to their targeted effect in manipulating the microbiome of the soil.

Once the identity of plant-based signals that manipulate the soil microbiome in a desired manner are known, we must also identify the mechanisms by which the signals are synthesised in order to breed for plants with the appropriate levels of these signals. Although an array of candidate metabolites, sRNAs and other molecules that manipulate microbial communities are already known, the molecular events that lead up to their production is often unknown. This could be because -omics data from a single molecular level is typically insufficient to understand the full picture of complicated pathways by which chemical compounds that are not produced via the traditional DNA > RNA > protein pathway are manufactured in the root. Although genomic resources are becoming available for a wide range of plants, transcriptomes can be informative only in understanding protein or metabolite signals that have known synthetic pathways. To step beyond this traditional form of analysis, recent research studies have begun to gain new understanding into the synthesis of metabolites and sRNAs by integrating multi-omics datasets through the use of advanced computational modelling on both new experimental work and publically available datasets. For instance, by assessing the metabolomic profiles of the root exudates and genomic variation across >300 accessions of Arabidopsis, a study correlated the synthesis of a range of metabolites with specific gene clusters (Mönchgesang et al. 2016a). Another study combined comparative genomics, RNA-seq, small RNA-seq and degradome data in an effort to understand the interspecies RNA silencing between grapevines and an oomycete pathogen (Brilli et al. 2018). In a study concerning the mutualistic interaction between AM fungi and Lotus japonicus L., combined analysis of lipidomic and transcriptomic data was used to elucidate the lipid-derived chemical signalling pathways involved in this symbiotic relationship (Vijayakumar et al. 2016). In all, these examples demonstrate that data from multiple molecular levels can be complementary to each other to identify the pathway by which root cells produce signalling compounds during interaction with soil microbes.

Should we use the knowledge advances derived from the methods proposed above to breed plants with enhanced discrimination between beneficial and pathogenic microbes as well as choosing varieties that produce signalling compounds that manipulate the soil microbiome in a way that sustainably supports plant health, we could still be stymied by a major aspect of microbial biology: the fact that microbes can deploy counter-measures that manipulate plant perception and/or signalling pathways (Asai and Shirasu 2015; Plett and Martin 2015; Wang et al. 2015). In response to plants, microbes are known to produce metabolic, nucleic acid or proteomic ‘effectors’ that target these pathways in a plant-specific manner. The majority of effectors studied to date have shown that they are necessary ‘keys’ used by microbes to colonise plant tissues. Therefore, without knowledge of the plant pathways targeted by these microbial effectors, we could produce plant varieties that, despite fostering microbial diversity in the soil and properly differentiating between microbial lifestyles, could be completely unable to enter into symbiosis with mutualistic microbes or could become hyper-susceptible to a pathogen. Therefore, we must understand the arsenal of microbial effectors in the soil and determine how best we may leverage this knowledge to manipulate microbial access to plant tissues in crop plants.

Effector proteins encoded by microbes are typically species- or clade-specific and have a very narrow range of plants upon which they have an impact. These effectors are small (typically <300 amino acids), secreted proteins that are produced by a microbe in response to plant signals. For example, effectors produced by pathogenic fungi such as Fusarium oxysporum, Cladosporium fulvum and Ustilago maydis can suppress host immunity by inducing alkinisation of root tissues (Masachis et al. 2016) scavenging chitin (de Jonge et al. 2010) and by inhibiting generation of reactive oxygen species (ROS) respectively (Hemetsberger et al. 2012). Mutualistic symbionts also deploy effectors that suppress the host plant defence system (Torto-Alalibo et al. 2009; Plett et al. 2011, 2014; Okazaki et al. 2013, 2016). For example, in mutualistic AM fungi the effector SECRETED PROTEIN7 (SP7) is induced in the presence of a host, taken up by host cells where it interacts with an ethylene response factor (ERF19) and represses hormonally controlled plant defences (Kloppholz et al. 2011). In rhizobial symbioses, effector proteins such as NopL reduce plant immunity and promote nodulation (Bartsev et al. 2004) whereas other effector proteins have been found to affect the host range of specific rhizobial genera (Kambara et al. 2009).

Microbial small RNAs (sRNAs), much like proteins, may also play effector-like roles during the interaction with host plants. sRNA classes of size 18–23 nt or 28–35 nt have been implicated in the virulence of several fungal pathogens (Vetukuri et al. 2012; Fahlgren et al. 2013; Weiberg et al. 2013). In terms of plant–microbial interaction, a set of Botrytis cinerea sRNAs (Bc-sRNAs) act like proteinaceous effectors to suppress immunity in Arabidopsis and tomato (Weiberg et al. 2013). The knowledge of microbial effector biology, be these protein- or RNA-based effector signals, and how effectors aid microbial colonisation of plant tissue has been one of the biggest advances in our understanding of the factors controlling plant–microbe interactions in the last decade. These studies will also prove to be a critical break-through that will be essential in coming years if we wish to genetically select plant genotypes that foster some microbial symbioses while repelling others.

Breeding plants that are responsive (or resistant) to all microbial effectors within the soil is not a reasonable pursuit. However, recent work has begun to suggest that there may be ‘hub’ proteins in the plant targeted by the effectors of multiple microbes (Fig. 4). For instance, the JAZ proteins in plants that are responsible for the perception of the hormone jasmonic acid are targeted by effectors of both pathogenic and mutualistic microbes (Jiang et al. 2013; Gimenez-Ibanez et al. 2014; Plett et al. 2014; Weßling et al. 2014). Therefore, if we were to understand which plant proteins were ‘hubs’ – key pathways responsible for mediating multiple plant–microbe interactions of interest and that were targeted by microbial effectors – we could breed plants with these hub proteins in mind. Before we can select for these hub proteins, however, we will have to (i) know what microbes are present in the soil; (ii) understand their effector complement; and (iii) have identified and studied the plant protein targets of these microbial effectors within plants.

Fig. 4.  A conceptual figure of how both pathogenic and mutualistic microbes may use effector-like proteins and sRNAs to target one plant protein ‘hub’ responsible for the control of signalling networks related to plant–microbe interactions. Within this figure, a central plant protein that regulates plant–microbe signalling can simultaneously be targeted by a range of microbial effectors (either protein in nature or through sRNA interference) to alter downstream plant response(s). How these effectors target or modulate such hub proteins would theoretically correspond to the lifestyle of the interacting microbes in question. Within this example, both protein- and RNA-based effectors of pathogens activate the protein, whereas effectors of mutualistic microbes act to suppress the activity of the hub protein.

Single-cell genomics applied within the rhizosphere could not only advance our knowledge of microbial responses to plant signals but could also take us a large step towards understanding what effector-like sequences are present within the discrete genomes of these microbes. Further work could then be performed to characterise the mode of action of these effector proteins in key crops. Such characterisation would include identifying and determining the interaction kinetics between microbial effectors and plant proteins/signalling pathways. Ideally, these advances would identify ‘hub’ plant proteins targeted by multiple microbial effectors (Fig. 4) and sequence polymorphisms in these hub plant proteins that could ameliorate or disrupt these interactions. Should this be the case, plants encoding these variations, or transgenic plants with site-specific gene mutation, could be used to study the impact of host genetics on the receptivity/responsiveness of plants to specific effectors. These data could be used in conjunction with knowledge of plant modes of microbial perception and plant signalling pathways to generate crops that would enable us to not only modify soil microbial communities for the betterment of farming sustainability but also alter plant responsiveness to microbial effectors thereby tailoring access to plant tissues by soil microbes.

Consideration of plant–microbe interactions within the soil, and how we can modify these interactions to benefit the sustainability of our agricultural practices, calls for a continued and focussed research effort on the part of the academic community and continued dialogue between academia and the agricultural industry. Altogether, the studies outlined in the previous sections could give us the key genetic loci of plants that could become value-added traits in crop breeding that will be one aspect by which we could re-create vibrant, diverse microbial communities in our agricultural soils. But is this only an academic design or a realistic aim? The pressures on current breeding programs are massive, and the number of desirable phenotypes required by industry are numerous. For example, in the Australian chickpea breeding program alone, there are 16 traits that have varying levels of desirability in new plant lines and five traits that must be maintained in any new variety produced (Fig. 5a). How, then, can we think to add new genetic loci into this breeding framework that aim to manipulate soil microbial communities and plant–microbe interactions? We would suggest that this depends upon how we view plant breeding: do we want to breed plants from a plant-centric view or from a holobiont-centric view? If we take a step back and consider the scientific literature to date, acknowledging that plant genetics are key to plant performance, it is also obvious that the majority of plant phenotypes desired in agricultural crops are also heavily influenced by microbes (Fig. 5b). Therefore, if we consider the holobiont-centric breeding plan, could we both obtain desirable plant phenotypes while simultaneously improving soil health? In this respect, we could consider a new plant breeding paradigm where we would seek to improve plant performance by choosing plants with genetics aligned to the promotion of beneficial rhizobacteria (symbiotic or free-living). For example, Pseudomonas, and Bacillus communities can repress plant diseases, sequester heavy metals, promote plant nutrition and enhance growth due to the release of growth promoting factors (Lucy et al. 2004; Idris et al. 2007; Rajkumar et al. 2012). We might also aim to improve the frequency of Drechslerella, Drechmeria and Stropharia fungi due to their nematopathic characteristics (Xu et al. 2011; Zouhar et al. 2013; Liu et al. 2014). There are many further examples that could be listed as found in Fig. 5, together demonstrating that identification of, and selection for, plant genetic loci that modulate plant–microbe interactions during a breeding program is not just an academic exercise, but one that could have significant positive impacts on agricultural practices while still meeting industry-set breeding priorities. Should this aim be then combined with complementary strategies, such as alterations to agricultural soil management practices, we could see a significant step forward in our ability to sustainably manipulate microbial communities in soils to our benefit.

Fig. 5.  A conceptual model of how both plant physiological and plant–microbe interactions could be integrated in a complementary manner within a breeding program to achieve holobiont-based plant breeding practices. (a) We list the recent plant characteristics sought within the Australian chickpea breeding industry and their level of desirability (source: K. Hobson, Pulse Breeding Australia). (b) We identify potential microbes within the soil that, if encouraged, could support the plant physiological characteristics desired by the chickpea industry.

We began this review by discussing how plant domestication, and consideration of the plant in isolation of its natural environment, have led to modern crop plants that, in some cases, come with an ecological cost because they may not be able to fully benefit from the soil microbiome. We posited that this is due, in part, to a breakdown in communication between plants and microbes. Such a confluence of events, together with agricultural management practices that also hinder microbial growth, has led to the observed decrease in microbial diversity in many agricultural systems that, in the long term, could have far-reaching consequences on the sustainability of our soils. But, these changes are not irreversible. As we have also covered above, recent research is making steady headway in considering the plant ‘holobiont’ – the view whereby the plant and its attendant microbes are considered as a contiguous whole, and where perturbation of one organism impacts the whole ecological system (Zilber‐Rosenberg and Rosenberg 2008; Vandenkoornhuyse et al. 2015; Sánchez-Cañizares et al. 2017). We have outlined above some adaptations to current research streams that we feel could have significant impacts upon our understanding of how plant–microbe interactions occur and how they ultimately affect soil health. We encourage investigation of how plant genetics affect the perception of multiple different microbial lifestyles simultaneously rather than individual microbes at the exclusion of all others. We also promote the thoughtful and targeted use of meta-analyses to leverage past work in these fields to guide future research effort. Finally, we hope that the interaction between plants and microbes will be considered not only in one direction (i.e. only plants affecting microbial physiology), but that these interactions be considered to the full extent of their dynamic back-and-forth relationships where microbes can have just as large an impact on plant function. Should such a view take hold, whereby the impact of plant–soil–microbe interactions are all considered, and extend to regulators, industry, and right down to individual landholders we could well see a re-invigoration of how plant breeding, varietal selection and land management are performed. Benefits from these proposed approaches are far-reaching and, hopefully, will bring a positive long-lasting impact on the sustainability, productivity and profitability of our agroecosystems.

The authors declare no conflicts of interest.