New insights in metabolism modelling to decipher plant–microbe interactions

Summary

Plant disease outbreaks, exacerbated by climate change, threaten food security and environmental sustainability world‐wide. Plants interact with a wide range of microorganisms. The quest for resilient agriculture requires a deep insight into the molecular and ecological interplays between plants and their associated microbial communities. Omics methods, by profiling entire molecular sets, have shed light on these complex interactions. Nonetheless, deciphering the relationships among thousands of molecular components remains a formidable challenge, and studies that integrate these components into cohesive biological networks involving plants and associated microbes are still limited. Systems biology has the potential to predict the effects of biotic and abiotic perturbations on these networks. It is therefore a promising framework for addressing the full complexity of plant–microbiome interactions.

Plant–microbiome interaction is complex

Plants constantly interact with a plethora of microorganisms, which can be beneficial, neutral or detrimental to plant growth and health (Bais et al., ²⁰⁰⁴; Haichar et al., ²⁰¹⁴). The mechanisms supporting these associations are, to a large extent, mediated by plant‐derived metabolites. Indeed, as primary producers in the food chain, plants produce various organic compounds that create a nutrient‐rich environment conducive to intense colonisation by a wide variety of microbes (Fatima & Senthil‐Kumar, 2015). For example, plants exert selective forces on soil microbial communities through root exudation, which involves the release of organic carbon into the soil, directly available for microbial growth such as a wide range of primary and secondary metabolites including carbohydrates, organic acids and amino acids (Pantigoso et al., ²⁰²²). Root exudates are also composed of specific compounds, which play an important role as signalling molecules during root–microbe interaction (Haichar et al., ²⁰¹⁴; Koprivova & Kopriva, ²⁰²²). Within the rhizosphere, this is a cyclic process, with substrates flowing between different compartments of the complete soil–plant–microbe system. Hence, metabolism is the underpinning force that sustains plant–microbe interaction (Holden, 2019).

Despite advances, our understanding of the molecular mechanisms underlying plant–microbe interactions is still incomplete. While several molecular mechanisms of the pattern‐triggered immunity and effector‐triggered immunity defence systems have been identified, highlighting the importance of nucleotide‐binding leucine‐rich repeat receptor genes (Jones et al., ²⁰²⁴), there is still limited knowledge about the role of metabolism in these interactions. Organisms are complex systems with thousands of components and mechanisms that allow pathogens and beneficial microbes to colonise the host. Hence, interactions are not the product of isolated genes, but instead emerge from the concerted operation of molecular networks. Therefore, identifying components and the systemic behaviour of networks is necessary for a better understanding of gene function and regulation. To this end, system biology is an appropriate way to study plant–microbe interaction in a more comprehensive fashion.

Systems biology has gained prominence with the development of ‘omics’ techniques combined with mathematical modelling. In this field, genome‐scale metabolic models of microorganisms are widely used (Feist & Palsson, 2008; Kim et al., ²⁰¹⁷). These models allow for the characterization and prediction of microbial function based on genome information, fundamental knowledge, and empirical observations of metabolic capacity. They aim to deliver a holistic understanding of metabolism at the genome scale.

Metabolic modelling can be applied to a known pure culture, synthetic mixed culture, or complex environmental mixed culture (Batstone et al., ²⁰¹⁹). These models have been employed to analyse metabolic interactions between organisms, including host–microbiome relationships. Furthermore, they already have practical applications, helping industry to improve biochemical production by optimising genotype design or refining growth media conditions, as well as identifying novel drug targets against microbes (Zimmermann et al., ²⁰²¹).

This review aimed to provide an overview of systems biology applications to decipher our understanding of plant–microbe interactions at the network scale. The goal was to model and predict the outcome of these interactions. Current approaches, applications, and challenges of systems biology to describe beneficial and pathogenic interactions with plants at different scales, from model strains to complex community (microbiome), will be discussed.

Challenges in modelling bacterial metabolism during interaction with plant

Systems biology seeks to unveil the properties of living organisms emerging at the network level (Peyraud et al., ²⁰¹⁷). To understand the dynamics and the biological functions of a network, system biology proceeds to ‘mathematical model reconstruction’. It integrates detailed information (omics, physiological and kinetic data, etc.) about each component and their interactions within a molecular network. This comprehensive data integration allows for the construction of mathematical models that represent the dynamic interactions within biological systems. Mathematical models are used routinely to help understanding complex systems by simulating various conditions and predicting the system's response to different perturbations. Different model types exist, the three main ones are constraint‐based model used for genome‐scale metabolic model, logical model used for regulatory network model, and kinetics model that can be used for both. Their use depends on the modelling purpose and the data availability (Fig. 1). Constraint‐based models, which simulate steady‐state flux distributions, are historically employed to study metabolic networks. Genome‐scale metabolic models (GEMs or M‐models) are widely used and exist for numerous organisms nowadays. Flux balance analysis (FBA) (Orth et al., ²⁰¹⁰) is the most used approach to analyse these models. It determines the optimal solution for a system to reach optimal rates of an ‘objective function’, that is, a biological function to be optimised like growth or the secretion of a specific molecule (Orth et al., ²⁰¹⁰). Genomic‐scale metabolic models with FBA have been useful to predict microbial growth rates, substrate uptake and gene deletion strategies for microbes. In the following, we are presenting three main challenges in modelling microbes interacting with plants that require development of new modelling methods with a focus on improving prediction based on genome‐scale metabolic models, see Fig. 1. These challenges are deeply intertwined and the modelling frameworks presented below gain interest by their integration.

• First, the speed of substrate consumption, that is, substrate uptake rates, has to be parametrised to accurately predict the population dynamics of the microbes in their ecological niche. However, such data are rarely available or could be fastidious to collect experimentally for all substrates. Hence, several modelling frame works were developed to parametrise the speed of nutrient consumption. Models of metabolism and macromolecular expression (ME models) provide, in addition to metabolism, a quantitative approach to gene expression through the integration of protein synthesis and posttranscriptional modifications inside the cell (Lerman et al., ²⁰¹²; Thiele et al., ²⁰¹²). They improve predictions of constraint‐based models. Nevertheless, this new model type tends to make the model computationally heavier and significantly increases the resolution time.

Fig. 1.

Model types tailored to address specific modelling challenges, highlighting their roles and advances to understanding complex systems. Schematic view of modelling approaches in order to overcome modelling drawbacks and to decipher biological mechanisms.

More recently, a new ME model formalism called Expression and Thermodynamics Flux (ETFL) has been proposed to reduce computational time by c. 10 times (Lloyd et al., ²⁰¹⁸; Salvy & Hatzimanikatis, ²⁰²⁰). In addition to the study of metabolism and the cost of protein synthesis, proposed in earlier ME models, ETFL incorporates thermodynamic constraints into a mixed‐integer linear problem formulation, enabling a rapid solution (Salvy & Hatzimanikatis, 2020). Expression and Thermodynamics Flux has started to be developed for eukaryotic organisms as well (Oftadeh et al., ²⁰²¹).

Alternatively, other methods were developed to integrate enzymatic constraints and proteomics data in classical FBA. One of them is gecko (Domenzain et al., ²⁰²²). Finally, new methods based on ME models or ETFL have been recently developed to simulate cellular dynamics under changing environments (Yang et al., ²⁰¹⁹; Salvy & Hatzimanikatis, ²⁰²¹). Recently, Schroeder et al. (2024) suggested gathering models that integrate the synthesis and cost of intracellular proteins under the umbrella term Resource Allocation Model (RAM). Even though these methods are promising, they are more fastidious to build than GEMs. Hence, they have not yet been applied to microbes interacting with plants.

• The second challenge in modelling plant–microbes' interactions is that not only small molecules drive the interactions but also many macromolecules are involved. Such macromolecules, like effectors, plant cell wall degrading enzymes, or exopolysaccharides, are often secreted by the microbes in the plant environment. Few frameworks were developed to model macromolecule biosynthesis, secretion, and activity to study specific aspects of the plant–microbe interaction. For instance, adding a macromolecule module to the metabolic network was used to predict the cost of secretion of pathogenicity determinants by bacterial pathogens or plant cell wall degradation by necrotrophic fungi (Peyraud et al., ^{2016, 2019}). However, the way of modelling such a macromolecule module is not yet standardised. Also, the diversity in the structural and chemical properties of plant polymers, like hemicellulose and pectins, is challenging to model in order to predict the degradation by plant cell wall degrading enzymes (Pieczywek et al., ²⁰²³). Enhancing the prediction capacity of nutritional interactions will not only facilitate a better understanding of the metabolic exchanges between plants and microbes but also aid in integrating biomechanics modelling (Léger et al., ²⁰²²).

• Interactions between plants and microbes are highly dynamic and present spatial heterogeneity and environmental variability. Hence, gene reprogramming is key to consider for predicting the outcome of the interactions. Hybrids modelling of M‐model and regulatory network models allows a better description of the behaviour of microbes in various environments they colonise, from the rhizosphere, phyllosphere, to different plant tissues (Peyraud et al., ²⁰¹⁸). Several methods have been developed to integrate regulatory networks with metabolic models, like time‐dependent FBA (rFBA) (Covert & Palsson, 2002) or regulatory steady‐state analysis (Marmiesse et al., ²⁰¹⁵). The main methods used to reconstruct regulatory networks are logical modelling with Boolean or multi‐state modelling. Probabilistic methods have also been used to account for the inherent stochasticity in gene expression and regulatory interactions. For example, such a modelling framework was implemented to model the virulence regulatory network of the pathogen Ralstonia solanacearum (Peyraud et al., ²⁰¹⁸). This study demonstrated how the virulence regulatory network impacts the robustness of the pathogen's metabolism, highlighting the intricate connections between gene regulation and metabolic function. Despite the advancements, the integration of regulatory and metabolic models for plant‐pathogen interactions remains limited by the available data found in the literature. More research is needed to develop and apply these integrated models to a broader range of plant–microbe systems.

Challenges in modelling plants

Modelling plants takes up several challenges. Firstly, a plant cell contains multiple organelles where some metabolites can translocate between them and where different metabolic reactions occur. The second main challenge is to build a multi‐cellular model. Indeed, in a specific tissue, the cells can interact and communicate through metabolite exchanges. Finally, the plant is a multi‐organ organism, meaning that cells can have different genetic expressions. Also, the interactions between several organs of the plant must be included to gain a more comprehensive understanding of general plant metabolism, like sink/source exchange of nutrients (Sampaio et al., ²⁰²²).

Organ centred models

The first and most curated metabolic model of a plant cell is the one from Arabidopsis thaliana (Poolman et al., ²⁰⁰⁹; Arnold & Nikoloski, ²⁰¹⁴). Shortly thereafter, other models from different species harbouring different metabolisms compared with A. thaliana, like monocot or C₄‐photosynthesis, were developed such as maize (De Oliveira Dal'Molin et al., ²⁰¹⁰; Saha et al., ²⁰¹¹; Bogart & Myers, ²⁰¹⁶; Chowdhury et al., ²⁰²²), rice (Lakshmanan et al., ²⁰¹³; Poolman et al., ²⁰¹³), barley (Grafahrend‐Belau et al., ²⁰⁰⁹), tomato (Yuan et al., ²⁰¹⁶), and rapeseed (Hay & Schwender, 2011; Pilalis et al., ²⁰¹¹). The metabolic reconstruction of these species may be challenging due to their metabolism being more or less distant from A. thaliana. Different organs, such as leaves, seeds, or roots, were studied with these models by fixing regulatory Boolean network rules and/or flux constraints. Although metabolic models of one plant tissue are useful for studying specific features of species, they do not allow for fully computing interactions between two or more organs at the same level.

Multi‐organ models

To overcome the limitations cited previously, Gomes De Oliveira Dal'Molin et al. (2015) developed the first multi‐tissue metabolic model of A. thaliana to study the interactions between leaf, stem, and root. Several other multi‐organ models followed for A. thaliana (Gerlin et al., ²⁰²¹), maize (Seaver et al., ²⁰¹⁵), tomato (Gerlin et al., ²⁰²²), and Medicago truncatula (Pfau et al., ²⁰¹⁸; diCenzo et al., ²⁰²⁰), to cite only a few. Mostly, these authors duplicate the metabolic model, change organ‐specific pathways and fluxes, and simulate the interactions with a pool of common metabolites exchanged between the models. Multi‐tissue models can be used at a dynamic level using the dFBA approach (Shaw & Cheung, 2018). The study highlighted the nitrogen (N) journey in the root and leaf, among other observations. Indeed, the simulation results indicate that N is stored in the root during the night and then transported and used in the leaf during the day. Such studies permit a better understanding of whole plant metabolism and are thus useful for engineering experiments (Shaw & Cheung, 2020).

Difficulties remaining

Despite recent progress, some difficulties remain, especially regarding available information on plant secondary metabolism or roles of some proteins. This leads to incomplete and incorrect metabolic networks, even for the most curated plant model, for example A. thaliana (Zamani Amirzakaria et al., ²⁰²²). Moreover, most of the models map the overall metabolism of a plant cell with little or no consideration for metabolic variations between cell types or environmental conditions (Sampaio et al., ²⁰²²). Recently, the use of transcriptomic data has allowed metabolic differences to be inferred between environmental stress conditions, such as heat vs cold stress, or unlimited vs limited nitrogen conditions (Chowdhury et al., ^{2022, 2023}). For multi‐organ models, some compartments like apoplast or rhizosphere, where many plant–bacteria interactions occur, are rarely modelled. Also, the biomass reaction as well as the uptake and secreted exchange fluxes for each organ have to be parameterized with literature or experimental measurements (Gerlin et al., ²⁰²²; Goelzer et al., ²⁰²⁴). To obtain accurate predictions, a significant amount of data is required. To overcome these issues, Goelzer et al. (2009, 2011) proposed a new modelling method named Resource Balance Analysis (RBA). Initially used for bacterial models (Goelzer et al., ²⁰¹¹), this method does not require constraints for fluxes and metabolite concentrations but instead gives a result with the optimal allocation of resources to cellular processes. Used very recently for the plant model of A. thaliana leaf (Goelzer et al., ²⁰²⁴), RBA succeeded in quantitatively predicting an accurate metabolism in different complex environments. Also, the integration of heterogeneous multi‐omics data could improve the predictions of multi‐organ models (Sampaio et al., ²⁰²²). However, these types of data are sometimes challenging to analyse and integrate into metabolic network constraints. In this context, combining machine learning (ML) with metabolic modelling may allow the improvement of plant model prediction capabilities, like the one used by Bai et al. (2024) to identify genes involved in plant specialised metabolite synthesis.

Advancement in modelling for plant–microbe interactions

Several metabolic models of microbes interacting with plants and for different plant species are now available to the scientific community (Gerlin et al., ²⁰²¹). It is therefore appealing to reuse these constructions to build metabolic models of plant–microorganism interaction, in order to better understand the molecular dialog between both partners, trophic exchanges during the interaction, and how promoting (in the case of plant growth‐promoting rhizobacteria) or restricting (in the case of phytopathogens) interactions affect the plant.

In this review, we decided to highlight metabolic models to predict interaction outcomes between plant rhizobacteria, plant phytopathogenic bacteria, and plant leaf microbiome (Fig. 2).

Fig. 2.

New insights in modelling plant–microbe interactions. Major advances have been made in predicting interaction outcomes between plant phytopathogenic bacteria, plant rhizobacteria, and the plant leaf microbiome. BCAA, branched‐chain amino acids; GEM, genomic‐scale metabolic models; PGPR, plant growth‐promoting rhizobacteria.

Metabolic reconstruction as tool to investigate key mechanisms within N‐fixing symbiosis

Bioengineering nitrogen (N) fixation in nonleguminous crops is being considered increasingly to enhance food production yields and reduce pollution due to soil fertilisation. The quantitative study of nitrogen‐fixing symbiosis in leguminous crops is thus an important area of research.

Three plant–symbiont interaction models about nitrogen fixation were constructed and are available (Pfau et al., ²⁰¹⁸; diCenzo et al., ²⁰²⁰; Holland et al., ²⁰²³). The models proposed by Pfau et al. (2018) and diCenzo et al. (2020) are based on the legume species M. truncatula, which possesses indeterminate nodules. The findings of both studies illustrated the impact of symbiosis on metabolic flux and plant growth. Pfau et al. (2018) emphasised the impact of oxygen availability on nitrogen fixation, whereas diCenzo et al. (2020) demonstrated that the metabolic costs associated with symbiosis are linked to the maintenance of nitrogenase enzyme activity. Holland et al. (2023) presents the first metabolic model for soya bean (Glycine max) nodule with its associated microsymbiont Bradyrhizobium diazoefficiens. The authors predicted the nitrogen fixation cost of c. 4.13 g C g⁻¹ N, which when implemented into the crop scale model ‘Soybean‐BioCro’ developed by Matthews et al. (2022) translated to a grain yield reduction of c. 27% compared with a nonnodulating plant receiving its nitrogen from the soil. Interestingly, they estimated the profitability (i.e. return in investment) of fertilisers vs inoculant application and evidenced that nodulated soya bean shows much higher profitability than fertilised soya bean. In addition, it can be reasonably assumed that, even if the gain is reduced for a plant that obtains nitrogen via nodulation in comparison with a plant that absorbs nitrogen from the soil through fertilisers, the issue of the environmental impact of nitrogen fertilisers, and in particular greenhouse gas emissions, will be resolved by plant–bacteria interactions.

In nutrient‐scarce regions planted with cereals such as in Africa, engineering N‐fixing symbiosis into cereals is of great interest. Indeed, developing a metabolic model of nodulated cereals such as maize or millet would be a powerful tool to design synthetic nodules. In practical terms, such a model could help identify metabolic barriers in maize that hinder nodulation and optimise symbiotic metabolic pathways to maximise yield through successive simulations with various gene sets.

Metabolic reconstruction to understand pathogen nutrition during infection cycle

To date, only one plant–pathogen interaction model was published by Rodenburg et al. (2019). They reconstructed an integrated metabolic model of the oomycete pathogen Phytophthora infestans and tomato by combining two previously published models for both species (Yuan et al., ²⁰¹⁶; Rodenburg, ²⁰¹⁸). To obtain more insight into the metabolic process during the infection cycle, dual‐transcriptome data covering a full infection cycle of P. infestans on tomato were obtained and integrated into the model. The transcriptomic data were used to generate submodels representative of the sampling time postinoculation, which are subsets of the full model. Using the Integrative Network Inference for Tissues algorithm to relate gene expression to metabolic activity, the authors evidenced that as the infection progresses, P. infestans becomes increasingly dependent on metabolites derived from the necrotic tomato tissue rather than relying on its own metabolism. Indeed, differences in reaction enrichment between the initial and late submodels allowed the authors to demonstrate that during early infection, P. infestans synthesises amino acids de novo, but later scavenges them from the tomato. This model enabled identifying pathogen dependencies on the host, paving the way for the study of new control strategies against P. infestans. This reconstruction could be improved in the future with the addition of the regulation and gene expression networks of P. infestans (see Section II) and further physiological and metabolomic data. However, it represents an important step in studying pathogen nutrition in planta, and the model's methodology should be applied to other pathogen–host systems.

Metabolic reconstruction to understand the constraints imposed by plant immunity on the pathogen during colonisation

Plants possess a sophisticated immune system that is capable of preventing the establishment of infection by invading microbes. One consequence of this plant immune response is the alteration of metabolite composition, which in turn affects the ability of microbes to cause disease. Recently, a computational model of Pseudomonas syringae metabolism, implemented with transcriptomic profiles obtained 5 h postinoculation into leaves of Arabidopsis plants that have been mock‐treated or pre‐immunized (Nobori et al., ²⁰¹⁸), was used to reveal changes in bacterial metabolism imposed by plant immunity (Tubergen et al., ²⁰²³). This model predicted that branched‐chain amino acids (BCAAs) play a central role in Pst DC3000 pathogenesis. The precise molecular mechanisms by which BCAAs affect gene expression in Pst DC3000 remain unclear. As it has been demonstrated for other bacterial species, it may be reasonably assumed that BCAAs bind to Lrp and directly modulate its activity on gene expression (Tani et al., ²⁰⁰²). This prediction was confirmed by both in vitro and in planta experiments. Indeed, in BCAA‐supplemented minimal medium, the expression of virulence marker genes was repressed. Similarly, a reduction in the expression of virulence marker genes was observed at 3 h postinoculation when Pst DC3000 was co‐infiltrated with BCAAs in naive plants. This finding reinforces a fine‐tuned connection between metabolism and virulence. In addition, the observed suppression of virulence factors by BCAAs may indicate that plants enhance BCAAs concentration in leaf apoplast to induce an increase in effective intracellular levels in pathogens, thereby interfering with the Lrp‐mediated transcription of virulence genes.

Metabolic models to predict plant microbiome assembly: applications and implications

Plants interact also with complex communities called microbiomes. These communities can promote plant growth and health and therefore have the potential to make agriculture more sustainable. However, we currently only have a rudimentary understanding of the interspecies interactions that shape these microbial communities. In a groundbreaking study, Schäfer et al. (2023) advanced our knowledge by identifying an organising principle for the bacterial communities residing on the leaves of A. thaliana. They employed metabolic interaction models to explore the dynamics of these leaf‐associated bacteria. By analysing the carbon source preferences of 224 bacterial strains from plant leaves and using metabolic genome‐scale models to simulate pairwise interactions, the study provided new understanding of how these microbes compete and cooperate. It is crucial to draw attention to the fact that this model is built from in vitro data and not on substrate consumption in planta. This emphasises the need to perform and integrate more data from microbiome activities in planta within metabolic models. Gaining insight into the interactions between the microbiota and the plant will allow for the design of targeted microbiomes for a wide range of applications, especially in agriculture and the environment.

Perspectives

As discussed above, the first reconstructions of plant microbe models have provided new insights into plant–microbe interactions. Systems biology and metabolic modelling are likely to play a key role in advancing our understanding of plant–bacteria interactions in the future. The limited number of available plant microbe models today can be attributed to current technical and material constraints. Firstly, as discussed in Section III, plants are more complex to model than bacteria due to their cell compartmentalization, which makes the modelling process time‐consuming. Moreover, the new fine‐grained RAMs such as RBA, ME model and ETFL, developed to improve prediction capabilities are more complex to build but also require much more data (Schroeder et al., ²⁰²⁴). Most importantly, there is still a significant lack of data for less studied organisms, either because they are time‐consuming to collect or because there are no high‐throughput experiments to measure them. Under these conditions, multi‐omics integration between different studies is essential. However, in the absence of standard procedures for data acquisition and processing, it requires extensive manual curation from the literature. Therefore, a major community effort is needed to standardise protocols, data formats, and computational tools to enable efficient multi‐omics integration (Kim & Tagkopoulos, 2018; Pazhamala et al., ²⁰²¹). The resulting high‐quality data obtained should then be stored in public data libraries, updated and curated by the community over time (Macklin, 2019).

Nevertheless, where knowledge is still lacking, ML tools and their broad range of applications have the power to fill the gaps in GEMs. Machine learning has been used, for example, to predict transcription start sites or annotate functional genes from genomics (Mishra et al., ²⁰¹⁹). Accurate reconstructions are highly dependent on the availability and quality of catalytic rate constant (K _cat) data. Collections of K _cat values are present in enzyme databases, like BRENDA, but they are far from covering all the diversity of existing organisms and enzymes, as no high‐throughput experimental assays exist to measure them. Several deep‐learning approaches have been developed lately to overcome this data deficiency (Li et al., ²⁰²²; Kroll et al., ²⁰²³; Yu et al., ²⁰²³). As an example, DLK_cat (Li et al., ²⁰²²) estimates K _cat values from the enzyme sequence and their substrate structures. They have successfully improved enzyme‐constrained GEMs (ecGEMs) reconstruction, and this will certainly help us enhance network models quality. Regarding plant–pathogen interactions, ML was used in the last decade to identify new components of the plant immune system, pathogen effectors and their targets (Mishra et al., ²⁰¹⁹). Importantly, the prediction of missing data values is not the only way of using ML in system biology. Machine learning can be used to analyse metabolic model outputs and generate or improve constraint‐based models (Zampieri et al., ²⁰¹⁹). Therefore, the combination of systems biology and ML, especially deep learning (Camacho et al., ²⁰¹⁸), holds the potential to enhance our understanding of plant–microbe interactions (Mishra et al., ²⁰¹⁹) and, ultimately, addresses the fundamental questions in the field.

Some of the key topics currently being explored are how plants discriminate between pathogenic and beneficial bacteria, how plants select their microbiota to promote their health, and how potentially highly harmful pathogens appear to exist as part of the healthy plant microbiome. Integrating the plant, microbiota, and pathogens into a single model will provide answers to these questions by unravelling the molecular mechanisms underlying the interactions among all the involved partners. In the present situation, it could also assist us in determining whether microbiota can aid plants in adapting to climate change. Ultimately, this knowledge will pave the way for the development of biocontrol strategies such as tailored microbiomes, which could also be leveraged to help plants adapt to climate change. Modelling plant–microbe interactions is thus of great interest for food security. Alongside biocontrol approaches and climate change adaptation, plant–microbe models also have an important role to play in precision agriculture (Gebbers & Adamchuk, 2010). Crop simulation models have been constructed since 1980 to predict agricultural system performance (Jones et al., ²⁰¹⁷). They simulate crop growth and yield for a wide range of crops considering various abiotic factors such as weather data, local soil conditions, and crop management (Hoogenboom et al., ²⁰¹⁹). However, these models cannot predict the effects of biotic stresses or account for the potential beneficial impacts of the crop microbiome or symbiosis, as they do not yet integrate the molecular aspects of plant physiology, let alone plant–microbe interactions. Hybrid models, integrating molecular networks and agronomical models, will make it possible to simulate the effects of abiotic parameters on biotic factors (and vice versa) and to measure their overall impacts on crop development. Such models will help to identify plant candidate genes for new plant breeding programmes (Lavarenne et al., ²⁰¹⁸). This will broaden the range of crop management options available, as metabolic modelling will enable us to influence microbiome composition, to bioengineer nitrogen‐fixing symbioses, etc. Finally, hybrid models will be able to guide us on the best management practices to develop sustainable agriculture.

Competing interests

None declared.

Author contributions

FezH conceived the idea for the review. FZH, CB, AC, WN, SR and RP contributed to its design and writing.

Disclaimer

The New Phytologist Foundation remains neutral with regard to jurisdictional claims in maps and in any institutional affiliations.