Potential and challenges for application of microbiomes in agriculture

Abstract

The plant microbiome can promote plant health and productivity through a multitude of mechanisms. Our understanding of plant–microbiome interaction relies on descriptive natural surveys and experiments performed under simplified laboratory environments. While reductionist approaches are essential to understand mechanisms of plant–microbiome interactions, they risk missing emergent community properties seen in nature. To bridge the gap between basic research and real-world deployment of the microbiome for translational application, one has to consider functional association as well as ecologic principles governing interspecies and interkingdom interactions. In this review, we discuss the beneficial potential of plant microbiomes to enhance plant growth, nutrition, stress tolerance, pathogen protection, and commercial value through the modulation of taste and flavors, using examples from model plants and agriculturally important crops. We then discuss how microbial invasion and persistence in standing communities, trade-offs under multiple stressors, and community instability under host- and environment-imposed modulation should be considered in the rational design of microbial inocula, followed by a scrutiny of the method of microbial delivery. We synthesize ideas on how multiomic data, including genomics, transcriptomes, and metabolomics, can be leveraged to identify strains or target genes of interest for functional studies and how machine learning algorithms can be incorporated to enable prediction of plant–microbiome interactions. Microbiome-based strategies hold promise for improvements in agriculture. Despite the intrinsic complexity of the underlying interactions, interdisciplinary approaches are constantly providing insight into microbiome functioning and assembly principles, which is key toward knowledge-based engineering of the microbiome for increased and sustainable crop performance.

Introduction

The term microbiome refers to a community of microorganisms and their surrounding habitat and encompasses microbial activity and their interaction with their local environment. The collection of microbes within the microbiome is specifically referred to as the microbiota. All eukaryotes harbor abundant and diverse microbiota that influence the physiology of the host. Some microbiota members are pathogens or symbionts that impose profound negative or positive impacts on host fitness; however, these represent a minority. The majority of the microbiome, known as the commensals, modulate host gene expression and phenotypes in more subtle but nevertheless significant ways. For example, expression analysis along different parts of the mouse gut in germ-free and conventionally raised mice has identified up to thousands of genes differentially regulated by the activity of the microbiome (Larsson et al. 2012). Similar transcriptomic changes have been observed in the model plant Arabidopsis thaliana by using simplified synthetic communities (SynComs; Ma et al. 2021; Teixeira et al. 2021), highlighting the general roles of the microbiome in modulating host processes in eukaryotes.

Based on near-saturated transposon-mediated mutagenesis studies on plant-associated rhizobacteria, including Pseudomonas simiae and Rhizobium leguminosarum, it is estimated that up to 2% of microbial genes are involved in competitive root colonization (Cole et al. 2017; Knights et al. 2024). Considering that these studies were each conducted with a mutant population of single bacterial strains representing 1 species and that host–microbiome interactions are by definition more complex due to microbial interspecies interactions, one can safely assume that the number of microbial genes involved is considerably higher than 2%. Indeed, it is not an overstatement to say that the microbiome can modulate almost every aspect of plant physiology, ranging from growth to development, nutrition, and stress tolerance. As such, the microbiome is a resource for functional studies of host–microbiome interactions, for examining potential adaptation mechanisms of both partners, their evolutionary trajectories, as well as translational applications.

Technological breakthroughs have been key to progress in understanding causality in plant–microbiota interactions. This initially included descriptive, culture-independent marker gene–based community profiling and metagenome surveys of plant microbiomes. Applying these techniques within and between plant species in natural but otherwise controlled environments revealed a taxonomic core of the plant microbiota (Bulgarelli et al. 2012; Lundberg et al. 2012; Yeoh et al. 2017). This in turn spurred the establishment of DNA sequence–indexed microbiota culture collections stratified by taxonomy (Bai et al. 2015; Levy et al. 2018; Zhang et al. 2019; Wippel et al. 2021; Durán et al. 2022b). These microbiota culture collections with gnotobiotic plant systems then enabled plant microbiome reconstitution in defined environments for hypothesis testing on microbiota functions (Zengler et al. 2019; Kremer et al. 2021; Northen et al. 2024). Important insights were gained from these simplified microbiota reconstitution systems, mainly obtained with the host A. thaliana: i) bacterial competition for host-derived organic carbon through metabolic niche overlap as an organizing principle for leaf-associated communities (Schäfer et al. 2023), ii) commensal–host homeostasis through interactions with the plant immune system (Lebeis et al. 2015; Xin et al. 2016; Yu et al. 2019 ; Chen et al. 2020; Ma et al. 2021; Pfeilmeier et al. 2021; Teixeira et al. 2021; Paasch et al. 2023; Cheng et al. 2024; Keppler et al. 2025), iii) an essential role of the bacterial microbiota in protecting the host from harmful root-colonizing fungi and oomycetes (Durán et al. 2018), and iv) mobilization of iron for plant nutrition (Harbort et al. 2020).

While reductionist approaches cannot capture the biodiversity seen in field experiments in natural environments, they can be complemented by machine learning (ML) algorithms enabling prediction of plant phenotypes associated with specific microbiomes. These in turn can be leveraged for rational design of bacterial communities for different purposes. Microbiomes hold the potential of enhancing plant health and reducing the use of synthetic fertilizers and pesticides, thus facilitating the move toward more sustainable agriculture. Microbiomes can improve host fitness without direct engineering of plant genomes, which often comes with time-consuming gene editing and plant breeding. With the beneficial potential of an increasing number of microbes verified in the laboratory, major gaps still exist to translate laboratory findings into real-world application. Major challenges center on the difficulty of the specific microbe under investigation (i.e. the focal strains) to invade, persist, and perform biological function upon interactions with a complex environment. This review focuses on the beneficial potential of plant-associated bacterial and fungal microbiomes for improving plant growth, nutrition, disease resistance, and abiotic stress tolerance and their underlying mechanism whenever available (summarized in Fig. 1). We then combine empirical knowledge with ecologic theories to discuss the approaches that can be taken for the practical application of plant-associated microbiomes.

Figure 1.

Examples of beneficial services provided by the plant microbiome. Plant-associated microbes have been reported to improve plant performance, stress tolerance, and quality through a diversity of mechanisms. AMF, arbuscular mycorrhizal fungi; CK, cytokinin; ET, ethylene; GA, gibberelilc acid.

Beneficial services provided by the microbiota

Plant growth promotion and modulation of plant development

Plant growth–promoting rhizobacteria (PGPR) have been studied (Lucy et al. 2004; Lugtenberg and Kamilova 2009; Tabassum et al. 2017; Pieterse et al. 2021) long before their existence and activities could be integrated into the concept of the plant microbiota (Bulgarelli et al. 2013). A given PGPR inoculum must compete with a standing community of root microbiota members. Therefore, studying PGPR activity in monoassociations in laboratory environments does not often translate directly to natural environments.

PGPR can modulate plant growth and development by producing/modulating phytohormone pathways such as auxin, cytokinins, abscisic acid, ethylene, and gibberellin (Orozco-Mosqueda et al. 2023). One of the best-studied examples of a PGPR is Bacillus velezensis GB03, first isolated as B. subtilis A13 in the 1970s from a wheat field in South Australia by Kloepper and colleagues (Jang et al. 2023). Growth promotion by B. velezensis GB03 can be attributed to its ability to induce auxin production through upregulation of nitrilase genes involved in the terminal step of Trp-dependent auxin biosynthesis (Zhang et al. 2007). In addition, GB03 produces volatile compounds such as 2,3-butanediol that promote plant growth possibly through the cytokinin pathway (Ryu et al. 2003). These positive effects, together with the fact that Bacillus spp. can persist under relatively harsh conditions as endospores, make it an effective biostimulant, and GB03 has been the focus of several commercial formulations (Jang et al. 2023).

Plant hormone levels need to be tightly regulated, as excessive amounts or ectopic distribution can have a negative impact on plant growth; in fact, many plant microbiota members are known to inhibit root growth through auxin production. For example, tested members of the root microbiota from the bacterial genus Variovorax can catabolize auxin and reverse negative impacts caused by other community members (Finkel et al. 2020), highlighting that members of a single genus can have a significant impact on plant root architecture. Although auxin signaling is by far the most important pathway underlying plant growth promotion, other pathways are involved. Studies using bacterial communities without any known auxin-related operons but with auxin-deficient mutants collectively suggest that microbes can promote A. thaliana growth and lateral root formation through induction of ethylene signaling, independent from known auxin–ethylene crosstalk (Gonin et al. 2023). Furthermore, an auxin- and ethylene-independent promotion mechanism has been reported (López-Bucio et al. 2007).

PGPR can produce small molecules that modulate host development and growth. Intriguingly, morphogenesis of the marine green algae Monostroma oxyspermum relies on the continuous supply of the small molecule thallusin by microbes, as the leafy thallus undergoes gradual disintegration when thallusin is depleted (Matsuo et al. 2005). Based on a molecular clock study, it is estimated that the Chlorophyta, the group of algae to which M. oxyspermum belongs, shared a most recent common ancestor with the Streptophyta clade, which contains land plants and charophyte green algae (Yoon et al. 2004; Parfrey et al. 2011). These 2 groups diverged approximately 700 to 1,200 million years ago. It is thus interesting to explore whether other microbiome-derived molecules act as morphogens in streptophyte algae or certain basal bryophytes. Some root-associated bacteria including Exiguobacterium can inhibit rice tiller number by producing the cyclo dipeptide Cyclo(Leu-Pro) (Zhang et al. 2025), which binds to the strigolactone receptor OsD14 and activates the signaling pathway. Several bacterial genera have also had a positive impact on rice tiller number, but the underlying mechanisms remain elusive.

Nutrition

In addition to the developmental mechanisms by which PGPR promote plant growth, the microbiome can improve plant performance under stress conditions, such as nutrient limitation. Microbes play an important role in cycling mineral nutrients into bioavailable forms for uptake by roots and allocation to below- and aboveground tissues for growth (Devi et al. 2022). One well-characterized example is biological nitrogen fixation (BNF) by rhizobia bacteria in legume root nodules, in which unavailable atmospheric nitrogen is fixed to ammonium. Following a complex molecular dialogue between legume hosts and soil-borne N₂-fixing rhizobia belonging to Alpha- or Betaproteobacteria, the bacteria are accommodated in large numbers in specialized root nodules where they trade the fixed nitrogen in exchange for carbon from the host. The capacity to induce root nodule formation and perform BNF has also evolved in the genus Frankia, a group of symbiotic Actinobacteria that colonizes nonleguminous plant families including Betulaceae and Myricaceae (Benson and Silvester 1993). While rhizobial and actinorhizal root nodule symbioses are restricted to a relatively narrow group of plants, potential nitrogen-fixing bacteria, called diazotrophs, are found free-living or in the microbiota of many plant species (Zhang et al. 2022; Adachi et al. 2025; Chakraborty et al. 2024). Some nitrogen-fixing cyanobacteria can differentiate into nitrogen-fixing cells known as heterocysts and motile filamentous hormogonia for dispersal (Álvarez et al. 2023). Together, these differentiations allow cyanobacteria to infect different parts of host plants for symbiosis without restriction to the root nodules. In extreme cases, such as the mutualism between extracellular Nostocales cyanobacteria and the aquatic fern Azolla spp., the symbiosis is obligatory, with the symbiotic cyanobacteria being transmitted vertically between plant generations and linked to evidence for reductive bacterial genome evolution (Ran et al. 2010).

Numerous studies have reported improved plant growth through inoculation with diazotrophs, which is often attributed to the bacteria supplying nitrogen to the plant (Giller et al. 2025). However, in all of these studies, the amount of plant nitrogen potentially derived from BNF was much lower than the equivalent values accepted for rhizobia–legume symbioses and often below the thresholds that can be confidently detected through currently available methods (Giller et al. 2025). In one study, a SynCom consisting of 2 Gammaproteobacteria diazotrophs and 2 non–N-fixing strains isolated from maize xylem sap colonized maize and contributed nitrogen to plant tissues (Zhang et al. 2022). Inclusion of 2 nondiazotroph “helper” bacteria increased the nitrogen-fixing activity of the SynCom. Together, these nondiazotrophs can possibly create a hypoxic environment that is favorable for nitrogen fixation (Zhang et al. 2022). ¹⁵N isotopic labeling experiments showed that these SynComs contributed 11.8% of the total N accumulated in maize stems through BNF. Yet, only 0.46% of the total nitrogen of maize was incorporated by SynCom-mediated BNF into leaves and 4.01% into roots (Zhang et al. 2022), which is far below the host nitrogen produced via BNF in legumes such as soybean by N₂-fixing rhizobia under nonfertilized conditions: 88% of total N in roots, 77% in aboveground vegetative biomass, and 49% in beans (Gelfand and Philip Robertson 2015). Similarly, BNF estimates based on ¹³N tracer analysis in roots of the C4 grass Setaria viridis treated with diazotrophic PGPR Azospirillum brasilense or Herbaspirillum seropedicae indicated that only 7% of the plant's daily nitrogen demands were supplied by these rhizobacteria. This value increased with an ammonium-excreting mutant of A. brasilense, but the fraction of BNF-produced nitrogen in the aboveground vegetative biomass of adult plants remains unknown (Pankievicz et al. 2015). Whether and how domestication and breeding have influenced nitrogen fixation efficiency to improve legume–rhizobial symbiosis and microbiome composition is not well understood (Liu et al. 2020b), but nitrogen fixation capacity appears to have increased with soybean domestication (Muñoz et al. 2016). Thus, it remains to be seen whether future rational engineering of diazotroph-mediated BNF in nonlegume hosts can compete with the remarkable efficacy of legume–rhizobial symbiosis.

Unlike legume–rhizobial symbiosis, symbiosis is established between obligatory mycorrhizal fungi (subphylum Glomeromycotina) and a variety of plants. Up to 80% of vascular plants are known to be mycorrhizal, in which arbuscular mycorrhizas constitute the majority (Brundrett and Tedersoo 2018). Arbuscular mycorrhizal fungi (AMF) form distinctive arbuscular structures within the cortical cells of the associated host roots and an extensive extraradical hyphal network. AMF provide nutrients and water to the AM host in exchange for organic carbon including lipids (Jiang et al. 2017; Keymer et al. 2017). Increased nutrient uptake can be simply explained by the enhanced surface area for absorption (Sanders and Tinker 1971) and can sometimes go up to 100% for nutrients such as phosphorus (Smith et al. 2004). However, the rate varies across species (Smith et al. 2004) and is not strictly correlated with extraradical hyphal density (Smith et al. 2000). Hyphal-facilitated phosphorus uptake is also reported in nonmycorrhizal plants (e.g. between Arabidopsis and the fungus Colletotrichum tofieldiae; Hiruma et al. 2016).

Beyond expanding the surface area available for nutrient foraging, extraradical hyphae carry distinct microbial communities (Emmett et al. 2021). AMF hyphae also facilitate migration of rhizobia populations toward the root system of their host plants (He et al. 2024a). These studies highlight the importance of multikingdom interactions and the broader impact of AMF in shaping the surrounding soil microbiome composition. Although nitrogen-fixing Rhizobium and AMF have different host ranges, a common symbiotic signaling pathway is involved (Gherbi et al. 2008; Girardin et al. 2019). The knowledge gained from common symbiotic signaling may help expand the host range of efficient nitrogen-fixing bacteria to nonlegumes. However, similar to the lack of robust evidence that inoculation of nonlegumes with diazotroph-mediated BNF bacteria leads to fixation of agronomically significant quantities of dinitrogen from the atmosphere, a recent meta-analysis revealed that globally sourced commercial mycorrhizal inoculants have consistently fallen short in real-world application (Koziol et al. 2025). In addition to directly providing nutrients to the plant, microbes can improve plant nutrition by modulating signaling pathways involved in nutrient uptake. The presence of the microbiome reduces A. thaliana root endodermis suberization through repression of the ABA signaling pathway, contributing to reduced root diffusion barriers and enhanced nutrient uptake (Salas-González et al. 2021). Through correlative analyses among 20 maize lines and the monitoring of multiple plant performance parameters and associated communities, an inbred maize line was identified to have the highest shoot biomass and nitrogen accumulation with an enrichment of Oxalobacteraceae (Yu et al. 2021). This inbred line exudes significantly more flavones to the rhizosphere, including apigenin and luteolin, which can mediate compositional changes in the microbiome and facilitate nitrogen uptake (Yu et al. 2021). Yet, the causality between increased flavone exudation and enrichment of Oxalobacteraceae for enhanced nitrogen uptake is not directly demonstrated. In a comparative study between rice varieties Oryza sativa indica and japonica with different nitrogen use efficiencies, the nitrogen transporter NRT1.1B shaped microbiome composition (Zhang et al. 2019). Construction and application of an indica-enriched SynCom improved rice growth of an indica variety more in organic nitrogen conditions as compared with a japonica-enriched SynCom (Zhang et al. 2019).

Phosphorus is often limiting for plants, as it is commonly sequestered in insoluble mineral compounds or organic material that plants cannot access. Numerous phosphate-solubilizing or phosphate-scavenging microbes, notably AMF, have been shown to improve plant growth under phosphorus-limiting conditions, and in many cases plant–microbiome interactions are dependent on the phosphate status of the plant host (Breuillin et al. 2010; Hiruma et al. 2016; Aslam et al. 2022; Brisson et al. 2022; Huang et al. 2024; Liu et al. 2024b).

Iron often has limited bioavailability in soils, especially in calcareous soils with a pH usually >7, where it is locked in insoluble Fe³⁺ inorganic compounds. As in the case of phosphorus, root-associated bacteria can play an important role in plant iron nutrition by facilitating iron mobilization (Harbort et al. 2020). The microbe-mediated rescue of A. thaliana iron starvation requires an interaction between the plant and the root microbiota, as it does not occur in plant mutants that are defective in the biosynthesis of catechol coumarins, specialized metabolites secreted by roots. This beneficial trait is strain specific yet functionally redundant across phylogenetic lineages of the microbiota. Although it is not known how these coumarins influence the microbes to facilitate their iron mobilization activity, other secreted coumarins lacking a catechol moiety have antimicrobial activity (e.g. toward soil-borne fungal pathogens Fusarium oxysporum f. sp. raphani and Verticillium dahliae) and thus influence microbiota community composition (Stringlis et al. 2018; Voges et al. 2019). Considering that root-secreted coumarins have at least 2 physiologic functions for the host, as antimicrobials and as cues for microbe-mediated iron nutrition, it is challenging to disentangle observed shifts in the composition of the root microbiome of plant mutants lacking these specialized metabolites when grown in natural soil.

Protection against pathogens

Plant diseases exert a heavy toll on global food production (Savary et al. 2019). In 2019, >2 million tons of pesticides were used (Sharma et al. 2019), with potentially deleterious impacts on ecosystems, including the soil biome (Gandara et al. 2024). It is expected that a reduced use of agrochemicals by biologicals or mimetics of natural compounds will contribute to a more sustainable agriculture. The role of the plant microbiome in improving plant disease resistance has been extensively documented. One of the best-characterized examples is so-called suppressive soil—arguably the most compelling argument that biologicals can work robustly for years under field conditions and for which transferability between soils has been shown in several cases. In contrast to disease-conducive soil, suppressive soil often carries low amounts of pathogens and poses little harm to plants. Even upon inoculation with a high pathogen inoculum, the negative impact of pathogens in the suppressive soil on the plant host tends to diminish over time (Weller et al. 2002; Spooren et al. 2024). Numerous cases of suppressive soils have been reported worldwide, which have revealed a remarkable diversity of the underlying biocontrol agents. The mechanisms of suppression are also diverse and can be summarized as occurring by at least 3 means: antibiosis/antagonism, niche occupation and resource competition, and priming. Although these mechanisms are fundamentally different, microbiomes often use a combination of them to modulate plant disease resistance, as exemplified here.

Common scab is a soil-borne disease caused by the gram-positive bacterial pathogen Streptomyces scabies and related species. These pathogens are characterized by the production of the cell wall cellulose inhibitor thaxtomin A, which has a devastating impact on a variety of root crops, such as potatoes, beets, carrots, and turnips (King and Calhoun 2009). An early study by Menzies and Menzies (1959) demonstrated that contrary to expectations, potatoes grown in a plot of eastern Washington State with a decades-long history of potato monoculture were mostly scab-free. By contrast, potatoes grown in the nearby region with a shorter cultivation history were suffering from scab decline. Importantly, disease suppressiveness in the Washington plot was lost by steaming, providing the first hint that a heat-labile biological factor could be involved. Follow-up studies have gradually revealed the biological agents involved, including nonpathogenic Streptomyces strains (Liu et al. 1995). These Streptomyces bacteria produce antibiotics against the scab-forming S. scabies, pointing to a direct antibiosis mechanism. Antibiotic production is also important for the disease suppression by several Pseudomonas strains against Gaeumannomyces graminis var. tritici, the causal agent of take-all decline. The suppressiveness is related to the capacity of the bacteria to produce the polyketide metabolite 2,4-diacetylphloroglucinol, a broad-spectrum antibiotic that increases bacterial membrane permeability and reduces cell viability in gram-positive and gram-negative bacteria.

In a more recent study by Getzke et al. (2023), the Pseudomonas brassicacearum strain R401 demonstrated broad-spectrum antagonistic activity against multiple bacterial taxa, including the broad host range wilting root pathogen Ralstonia solanacearum. The disease suppression was only partially dependent on DAPG production, highlighting the simultaneous use of multiple mechanisms for disease suppressiveness by R401. Disease suppression by R401 also involves production of the iron chelator pyoverdine, pointing to the importance of resource competition, in this case competition for iron. Iron is essential for biological processes in microbes as well as the host, and a recent study suggested that plants modulate iron bioavailability within their tissues as an immune output. Activation of immunity by treating A. thaliana with the bacterial flagellin-derived elicitor flg22 leads to the elimination of the iron deficiency–responsive signaling peptide Iron Man 1 (IMA1). Elimination of IMA1 suppresses iron acquisition and reduces iron availability in the plant, which is hypothesized to limit microbial growth (Cao et al. 2024). However, direct control of commensal microbiota through IMA1-dependent nutritional immunity remains unexplored. Bacteria have evolved multiple strategies to facilitate iron uptake, including the production of iron-chelating siderophores. In a systematic survey of 2,150 strains isolated from the tomato rhizosphere through the colorimetric chrome azurol S assay, up to 95% of strains exhibit the potential to produce iron-binding molecules including siderophores (Gu et al. 2020). Treatment of supernatant collected from siderophore-producing strains under iron-limiting conditions leads to varying degrees of inhibition of pathogenic R. solanacearum. The addition of iron to these supernatants restores pathogen growth, supporting a role of iron availability in pathogen proliferation. Yet, the addition of bacterial supernatants sometimes facilitates pathogen growth, which is expected as iron-bound siderophores can be taken up by bacteria expressing the corresponding siderophore receptor. Therefore, microbiota-secreted siderophores can be a dual-edged sword inhibiting or facilitating pathogen proliferation (Gu et al. 2020). Interestingly, siderophore-mediated inhibition seems to be linked to phylogeny; that is, closely related strains tend to exhibit stronger inhibitory effects. This result is in line with a resource utilization study that concluded that closely related strains tend to have overlapping niche and substrate preferences (Schäfer et al. 2023). These results highlight the need to consider phylogeny and ecologic theories such as metabolic niche overlap when designing microbial communities for the control of specific pathogens.

The plant microbiome may suppress disease without directly inhibiting pathogen growth but rather by preventing pathogen virulence. Many microbes switch to a pathogenic lifestyle only when their quorum-sensing system is activated by high population densities. Interfering with quorum sensing can therefore prevent disease symptoms (Helman and Chernin 2015). For example, R. solanacearum uses methyl-esterified fatty acids for quorum sensing, and inhibition of this system reduces its virulence when inoculated on tomato (Yoshihara et al. 2020). Soil metagenomes contain genes for lipases that are active against the R. solanacearum quorum-sensing molecules, preventing the pathogen from production of virulence-associated molecules such as exopolysaccharides (Lee et al. 2018). Plant genotypes or environmental conditions that promote expression of these lipases are a possible strategy to reduce the incidence of disease caused by R. solanacearum.

A comparison between rice resistant varieties against Rhizoctonia solani identified a disease-suppressive Aspergillus cvjetkovicii strain (Fan et al. 2024). The underlying mechanism was mapped to a fungal-derived molecule called 2,4-DTBP (2,4-Di-tert-butylphenol). 2,4-DTBP treatment downregulated the R. solani gene AMT1, possibly through interference with the upstream bZIP transcriptional factor. As AMT1 is required for sclerotia formation, downregulation of AMT1 provides a mechanistic link of how a small molecule can suppress R. solani infection.

A similar comparison between resistant and susceptible banana varieties against F. oxysporum, the causal agent of Fusarium wilt, identified specific fungal Trichoderma species responsible for resistance (Liu et al. 2023). Another beneficial fungus, Serendipita vermifera, can provide plant growth promotion and resistance against Bipolaris sorokiniana in Arabidopsis and barley. While the mechanism is unclear, such beneficial effects act synergistically in the presence of the bacterial microbiota and are likely associated with microbial interkingdom interactions rather than host transcriptional modulation (Mahdi et al. 2022).

Although viruses do not fit the definition of living microorganisms, they play pivotal roles in modulating host and microbe physiology. A recent metagenomic study on 4 crop species, including wheat, rice, maize and Medicago, led to the expansion of 9,736 nonredundant viral genomes collectively known as the crop root viral genome collection (Dai et al. 2025). About 60% of the assembled bacterial genomes from this study carry CRISPR spacers, indicative of the history of interaction between the cognate microbial hosts and viruses. Among the entire crop root viral genome collection, up to 82.4% of genomes were detected from a single crop species, suggesting high niche differentiation. Such a knowledge of specificity can be leveraged to develop bacteriophage-based control agents to target pathogens. In a phage therapy study involving R. solanacearum, a 4-phage combination showed the highest biocontrol efficacy (Wang et al. 2019). Even though phage resistance evolved over time, the resistant pathogen tended to grow more slowly than the wild type strain, possibly due to the associated fitness cost of the underlying mutations. In line with the high host specificity inferred by the earlier metagenomic study, the presence of phages did not affect the microbiota composition in the absence of the pathogen, highlighting the feasibility and specificity of the phage-mediated treatment.

For many biological agents responsible for soil disease suppressiveness, robust rhizosphere or soil colonization is required for the suppressive activity. The antagonism of nonpathogenic Streptomyces increased with a higher inoculum titer and the number of antagonists applied (Liu et al. 1995). In the case of take-all decline, biocontrol potential was correlated with the capacity of the Pseudomonas strains to persist in the wheat rhizosphere over 8 plant generations (Raaijmakers and Weller 2001). However, it remains unclear why disease-suppressive strains tend to colonize and persist in one soil but not another. Plant genetics and environmental factors likely play key roles here. In a comparative study between R. solanacearum–resistant and susceptible tomato varieties, differences in rhizosphere microbiomes were observed (Kwak et al. 2018). The authors used a metagenome-assembled genome of a Flavobacterium from the resistant tomato to make predictions regarding its antibiotic resistance and sugar utilization potential. Together, these predictions facilitated the isolation of Flavobacterium TRM1 with disease-suppressive activity (Kwak et al. 2018). Other Flavobacterium species were implicated in controlling soil-borne fungal disease to R. solani in sugar beet, which was shown to be genetically dependent on their potential to synthesize antimicrobials and secondary metabolites encoded by different nonribosomal peptide synthetases (NRPSs) and polyketide synthases (Carrión et al. 2019). Moreover, Chitinophagaceae and Flavobacteriaceae members became enriched in the root endosphere of the Rhizoctonia-suppressive soil, which was associated with enhanced enzymatic activities related to fungal cell wall degradation, and a “minimal consortium” consisting of a Chitinophaga and Flavobacterium strain consistently suppressed fungal root disease in a simplified bioassay. Previous work with the same Rhizoctonia-suppressive soil identified 3 sugar beet rhizosphere-enriched Pseudomonas haplotypes, and a corresponding isolated Pseudomonas strain suppressed the fungal root disease but not its mutant derivative in which 1 NRPS gene is inactivated (Mendes et al. 2011). Hence, it is possible that at least 3 bacterial species are involved in the soil-suppressive trait against Rhizoctonia, 1 of which acts in the rhizosphere and 2 in the root endosphere.

In a field study addressing the effect of crop management on peanut root rot severity caused by F. oxysporum, crop rotation resulted in reduced disease severity as compared with the monocropping group (Zhou et al. 2023). Differences in crop management practices were associated with different microbial communities. Rotation-enriched strains, mostly in the phyla Proteobacteria and Actinobacteria, showed synergistic interactions in suppressing F. oxysporum growth, a root-infecting soil-borne fungal pathogen (Zhou et al. 2023). Root exudates collected from peanuts grown in monocropping and rotation soil have different impacts on microbiomes (Zhou et al. 2023), suggesting that crop management practices modulate microbiome structure possibly through a feedback mechanism involving root exudate chemistry. The abundance of Fusarium-suppressive Trichoderma fungi can also be increased by plant-exuded metabolites including shikimic acid, suggesting possible plant-mediated recruitment processes (Liu et al. 2023).

Foliar infection by Pseudomonas syringae induces root expression of ALMT1, a malic acid transporter, which in turn enriches an immune-priming Bacillus rhizobacterium through positive chemotaxis (Rudrappa et al. 2008). Such a “cry for help” phenomenon along the shoot–root axis is also noted during infection of Arabidopsis by the downy mildew pathogen Hyaloperonospora arabidopsidis (Hpa). Leaf infection leads to a distinctive shift in the root microbiome with enrichment of Microbacterium, Stenotrophomonas, and Xanthomonas spp. This shift in the root microbiome of a primary population of downy mildew–infected plants likely contributes to enhanced protection against the pathogen in a second population of plants growing in the same soil (Berendsen et al. 2018). Remarkably, it turns out that these root-enriched rhizobacteria also associate with Hpa spores, are shared among Hpa cultures, and form 34% of the phyllosphere community of infected Arabidopsis (Goossens et al. 2023). Counterintuitively, although the Hpa-associated bacteria benefit from Hpa infection of the host, they adversely affect Hpa spore formation. With a gnotobiotic Hpa culture, pathogen-infected A. thaliana enrich these bacteria in the root and shoot microbiota, hence forming a soil-borne infection-associated protective bacterial microbiome (Goossens et al. 2023). This protective microbiota appears to be specific against biotrophic Hpa, as foliar infection of A. thaliana with spores of the biotrophic fungal pathogen Golovinomyces orontii induces extensive but distinct local changes in the leaf microbiota and marginal shifts in the root microbiota (Durán et al. 2021). Mutation in a MADS box transcription factor RIN, previously implicated in ethylene signaling and resource allocation, resulted in an overall shift of tomato microbiome structure with a depletion of Actinobacteria (Yang et al. 2023a). The depletion of Actinobacteria is associated with enhanced disease susceptibility possibly due to the ability of Actinobacteria to produce a range of antimicrobials. Metabolomics suggested that exudates of the rin mutant contained lower amounts of riboflavin and 3-hydroxyflavone. Supplementation of these 2 compounds in the rin mutant enhanced the relative abundance of Actinobacteria and promoted disease resistance (Yang et al. 2023a). Together, these examples highlight the potential of metabolic profiling for identifying mechanisms modulating microbiome profiles with beneficial roles in enhancing plant disease resistance.

Drought and salinity stress

The ability of microbes to improve plant abiotic stress tolerance has been extensively reported. Here, we focus on drought and salinity stress tolerance as they are well characterized, partly due to well-controlled perturbation experiments in the field and microbiota reconstitution systems. Across numerous studies, Actinobacteria, especially the genus Streptomyces, are enriched in root and rhizosphere microbial communities under drought (Fitzpatrick et al. 2018; Xu et al. 2018; Qi et al. 2022; Caddell et al. 2023; Faist et al. 2023; Fan et al. 2023; Liu et al. 2024a; Styer et al. 2024; Swift et al. 2024). Although the thick cell walls of these gram-positive bacteria may make them more resistant to drought stress, an increase in the relative abundance of Actinobacteria is not seen in unplanted soil, suggesting that the enrichment is mediated by interactions with the plant host, perhaps indicating another cry for help. The mechanism behind the increase in Actinobacteria is not definitively known; however, it has been proposed that drought-induced changes in host physiology alter the rhizosphere through decreased root exudation, increases in reactive oxygen species (ROS), and reduced iron availability, resulting in an environment where Streptomyces is more competitive (Liu et al. 2024a).

In some cases, the increased abundance of Streptomyces was correlated with improved plant growth and survival under drought, suggesting that the bacteria provide beneficial services for the host (Fitzpatrick et al. 2018; Xu et al. 2021). Indeed, treatment with Streptomyces isolates in gnotobiotic systems does improve tolerance of several plant species to drought and osmotic and salt stresses (Xu et al. 2021; Yang et al. 2023b). The beneficial effect may be due to the production of pteridic acid by the bacteria, which leads to upregulation of genes associated with auxin and abiotic stress responses (Yang et al. 2023b).

However, the presence of Actinobacteria does not always correlate positively with plant performance under drought. A maize study found that although Actinobacteria were generally enriched in the rhizosphere, the presence of Streptomyces was not correlated with plant growth, and other Actinobacterial taxa such as the Nocardiaceae were negatively correlated (Swift et al. 2024). An Actinobacterial order also showed a negative correlation with drought tolerance in potato, being relatively depleted in the roots of a drought-tolerant potato variety, but enriched in a drought-sensitive genotype (Faist et al. 2023). This study took a metagenomic approach, which can reveal functional information about the microbiome, rather than simply phylogenetic information provided by the more common 16S rRNA gene sequencing. Nevertheless, the inconsistencies among studies in which taxa are enriched in the community and correlated with plant growth under drought highlight that caution is required when interpreting correlative community shift data.

Reductionist approaches have been successful in identifying bacteria with beneficial effects under water stress (Yang et al. 2021; Qi et al. 2022; Alwutayd et al. 2023; Acuña et al. 2024). One promising genus is Variovorax, which promotes drought tolerance when inoculated on sorghum and wheat (Qi et al. 2022; Acuña et al. 2024). In Arabidopsis, Variovorax was previously shown to maintain primary root growth in the presence of other microbiota members, which would otherwise inhibit primary root growth and promote lateral root emergence (Finkel et al. 2020). This activity extends to sorghum, where treatment with Variovorax suppressed the root growth inhibition triggered by Arthrobacter (Qi et al. 2022). The Variovorax strains also improved sorghum growth under drought, possibly due to the altered root architecture. A Pseudomonas isolate, SA190, originally isolated from root nodules of the desert legume Indigofera argentea, further increases the drought and osmotic stress tolerance of Arabidopsis and Medicago sativa (Lafi et al. 2016; Alwutayd et al. 2023). The proposed mechanism in this case is modulation of the plant drought response by inducing ABA signaling and increasing the chromosome accessibility of aquaporin genes. This alters plant–water relations and allows the plant to maintain growth and photosynthesis during drought, thus increasing water use efficiency. Since the latter study involved experiments using only monoassociations, it remains to be examined whether this mechanism functions in the context of a microbial community.

Flavors and taste

Fermentation has long been used to increase food nutritional values, as well as to alter flavors and stability for human consumption (Marco et al. 2017). Flavor is one of the most important factors determining the economic values of fermented products. It was recently found that the microbiome modulates this complex trait. Importantly, variation in microbial composition can be used to accurately predict the chemical composition of wines, raising the possibility of using microbiomes as biomarkers to predict terroir (Bokulich et al. 2016). However, bacterial and fungal microbiomes can have different contributions to wine regional flavors. In a study using Chardonnay and Cabernet Sauvignon grapes grown in California, a stronger association was found between bacteria and sensory-active metabolite profiles (Bokulich et al. 2016), but a study on Pinot Noir grapes grown in Australia suggested otherwise, with fungal communities contributing the most to regional distinctiveness (Liu et al. 2020a).

Cheese making involves milk fermentation with starter lactic acid cultures containing a combination of Lactococcus and Streptococcus strains. Dropout experiments of specific microbes from the starter culture of a yearlong cheddar-making experiment demonstrated that Streptococcus thermophilus played a key role in determining the microbial composition and the resultant metabolite profiles (Melkonian et al. 2023). Mechanistically, S. thermophilus enhances Lactococcus community growth by crossfeeding. The Lactococcus community in turn influences cheese flavor by limiting the production of C4 aroma compounds such as diacetyl and acetoin (Melkonian et al. 2023). When in excess, these compounds produce characteristic off-flavors, highlighting the role of microbe–microbe interactions in shaping flavors. Similarly, the composition of microbiomes during each step of coffee bean harvesting, fermentation, and drying contributes to the complex flavor of a cup of coffee (Pereira et al. 2016; Wang et al. 2020a). These studies exemplify the significance of microbial interactions, including competition and cooperation, in flavor formation. It should be noted that the microbiome can also modulate flavor through nutritional processes. For instance, specific microbiomes can facilitate ammonium absorption and influence the nitrogen metabolism of tea plants. Modulation of nitrogen metabolism results in enhanced production of theanine, a nonprotein amino acid, which serves as an important determinant of tea quality (Xin et al. 2024). Thus, the microbiome holds significant potential for fine-tuning the flavors of different products to tailor them to consumer preferences.

Challenges for the deployment of beneficial microbes in an agricultural setting

The development of large-scale culture collections, high-throughput screening with phenotyping platforms, and transcriptome and metabolome profiling has significantly expedited the identification of microbes with beneficial traits under laboratory conditions. However, to harness the potential of the plant microbiome in agroecologic environments, we still have to overcome multiple major challenges: i) the identification of the right microbe/microbial consortium for inoculation; ii) the delivery of focal microbes that can invade and persist in the native microbiome; and iii) regulatory hurdles, including public acceptance and regulatory approval. We focus on the first 2 points in this review (Fig. 2).

Figure 2.

Strategies and challenges for microbiome-based technologies in agriculture. Upper: A pipeline for identifying and characterizing candidate microbes, involving large-scale data collection and reductionist experimental approaches. Lower: Factors that must be considered for successful deployment of microbes.

Microbial invasion and persistence in indigenous communities

For many of the beneficial microbes described so far, growth and stress tolerance activities have been identified in growth systems with highly simplified microbial communities. However, the soil biomes already present in unplanted agricultural soils make it challenging to apply these microbes in practice. Community structure of the plant microbiota is determined by host genotype, soil edaphic factors, and climate (Durán et al. 2022a), which in turn influence a standing local soil biome. A meta-analysis based on 28 field studies summarized the effectiveness of multiple nitrogen-fixing Rhizobium strains on soybean nodulation and yield (Thilakarathna and Raizada 2017). In short, the effectiveness of Rhizobium strains varies a lot. While some of the inoculants gave rise to very positive results (e.g. yield increase up to 109%) as compared with the uninoculated control, some imposed a negative impact. Among all the metadata available, pH seemed to be the strongest driver of inoculant success. The authors also concluded that inoculants tended to be more effective when indigenous rhizobia strains were absent or at a low level.

Although interpretation from such a meta-analysis is compounded by variations in focal strains and experimental conditions, it serves as a good example for subsequent discussion. The focal strain, regardless of whether it is a native or introduced species, will interact with the native microbiota community. Successful invasion and establishment of the focal strain depends on different interactions, including niche overlap competition, antagonism, facilitation, and priority effects (Nemergut et al. 2013). If beneficial microbes are simply added without attention to these factors, they are unlikely to effectively invade and persist in the standing community (Wippel et al. 2021; Edwards et al. 2023; Fonseca-García et al. 2024; Styer et al. 2024). Even a rationally designed SynCom, one composed of microbes that were isolated from sorghum roots and able to grow on sorghum root exudates, showed poor long-term persistence when inoculated on field-grown sorghum (Fonseca-García et al. 2024). The challenge is magnified by the fact that many potentially beneficial microbes are isolated from noncrop plant species and in very different environments, such as deserts (Alwutayd et al. 2023; Acuña et al. 2024). In competitive inoculation experiments of A. thaliana and Lotus japonicus gnotobiotic plant systems and 16 member SynComs composed of microbiota culture collections for these hosts, the SynComs showed a preference for their native hosts (Wippel et al. 2021). Notably, host preference of SynComs consisting of root-derived microbiota strains was a community trait and undetectable in monoassociations. Sequential inoculation experiments revealed priority effects during root microbiota assembly, where established communities were resilient to invasion by latecomers. However, strains with higher host preference tended to have a better capacity to invade an existing community associated with the root but not the rhizosphere of the cognate host, suggesting possible modulation by host factors. Similarly, plants in sterilized soils were preferentially recolonized by microbes from the original soil as compared with microbes from different soil types. Dropping out individual strains from SynComs, followed by the introduction of late colonizers, revealed similar priority effects with strain-specific variations for the leaf microbiota (Carlström et al. 2019). Hence, to maximize the utility of beneficial microbes, plants and microbes will need to be carefully chosen or engineered for effective invasion and persistence within the microbiota. This will require more knowledge of the mechanisms behind host and soil preference.

An additional caveat to consider when choosing microbial strains for deployment is that beneficial strains have the potential to overproliferate and show pathogenic behavior in certain conditions. Commensal bacteria can secrete a repertoire of cell wall–degrading enzymes through the Type II secretion system to facilitate their colonization (Entila et al. 2024; Pfeilmeier et al. 2024). Under normal conditions, the expression of Type II secretion system genes is suppressed by ROS, one of the early defense-associated outputs in plants (Entila and Tsuda 2025). However, when the ability to produce ROS is compromised as in the NADPH oxidase rbohD mutant, these commensal bacteria can turn into opportunistic pathogens and affect the overall community structure of the microbiota (Pfeilmeier et al. 2021; Entila et al. 2024). Similarly, detrimental effects of bona fide microbiota members have been observed in simplified microbiota community contexts (Karasov et al. 2018; Ma et al. 2021). Pseudomonas OTU5 was prevalent and found in high abundance in the leaf endosphere of healthy A. thaliana plants sampled from a natural population. Yet, the same plant genotype inoculated with OTU5 isolates in monoassociation showed severe disease symptoms (Karasov et al. 2018). Pseudomonas R401, which was isolated from roots of healthy A. thaliana growing in natural soil, also caused detrimental effects on plant growth in agar plate assays in monoassociation or simplified SynComs (Ma et al. 2021; Getzke et al. 2023). Therefore, in addition to the ability of a focal strain to invade a standing microbial community, the capacity of the community to suppress potential pathogenicity of the focal strain must be considered when choosing microbial isolates for application. In nature, there are occasional reports of single taxa rising to dominate microbial communities of healthy plants. For example, Pseudomonas strains were reported to become suddenly dominant (up to 44% of all amplicon reads) in the maize rhizosphere after 8 wk (Walters et al. 2018). Similar dominance was observed in sorghum, but the associated “Pseudomonas bloom” consisted of multiple closely related strains instead of a single bacterial lineage (Chiniquy et al. 2021). Further investigation on mechanisms underlying taxon-specific blooms within plant microbiota may provide additional insight of how specific species can outcompete indigenous members without any apparent negative impact on plant health.

Microbial-mediated trade-offs

If a focal microbe is able to invade and persist in the standing microbial community, it may have unexpected consequences for homeostasis of plant signaling or microbial community functioning. When plants prioritize either growth or defense but not both simultaneously, it is referred to as the growthdefense trade-off (Huot et al. 2014). In studies using microbial culture collections isolated from A. thaliana, several groups independently reached the conclusion that the plant growth–defense balance can be modulated by commensal bacteria (Yu et al. 2019; Ma et al. 2021; Teixeira et al. 2021; Eastman et al. 2024; Entila et al. 2024; Keppler et al. 2025). Specifically, up to 40% of the tested commensal bacteria showed activity in suppressing root immune responses and enhanced plant susceptibility to opportunistic pathogens. The outcome of plant growth in response to colonization by SynComs that are only suppressive or nonsuppressive or by consortia containing suppressive and nonsuppressive commensals as found in nature resulted in a distinct transcriptome-based immune status in roots. These findings were used to construct a rheostat model (Ma et al. 2021) in which a dominance of immune-suppressing microbiota strains rendered plants more susceptible to pathogens and a dominance of immune-activating strains led to reduced plant growth. A recent study on maize also identified a set of microbiota-repressed genes associated with leaf growth promotion at specific plant developmental stages. Mutagenesis analyses further validated the positive role in defense and the negative role in growth of some of these maize genes, highlighting the presence of growth–defense regulatory circuits in both divisions of flowering plants: the dicotyledons and monocotyledons (Custódio et al. 2025).

Although activation of immune responses typically results in microbial growth restriction, especially of phytopathogens, such activation can be beneficial to particular commensal bacteria. This idea is supported by a study on B. velezensis strain FZB42 in which FZB42 colonization activated immune responses, including ROS production. ROS in turn stimulated the production of auxin by the bacterium to modulate root architecture and facilitate FZB42 colonization in an immunity-dependent feedback loop (Tzipilevich et al. 2021). In line with the idea that activation of immunity can have a differential impact on microbiota members, elicitor treatment on Arabidopsis led to changes in the relative abundance of only a few strains in SynComs (Ma et al. 2021; Teixeira et al. 2021).

Microbiomes have been reported to modulate the tolerance of plant hosts under abiotic stresses with concomitant changes in defense. For example, under phosphate (Pi) starvation, the transcription factor PHOSPHATE STARVATION RESPONSE 1 (PHR1) upregulated genes involved in the phosphate starvation response (PSR) and alleviated Pi deficiency by promoting Pi uptake (Bustos et al. 2010). Due to still unknown reasons, PSR requires the presence of a synthetic microbiome; that is, the PSR is not activated in axenic plants under phosphate starvation (Castrillo et al. 2017). Activation of PHR1 under phosphate starvation negatively regulates the expression of immune-related genes through the upregulation of the RALF23–FERONIA module (Castrillo et al. 2017; Tang et al. 2022). Such a PHR1-mediated suppression of immunity was originally interpreted as a strategy to promote rhizobacteria colonization to alleviate low Pi stress (Tang et al. 2022). However, as reported by the same group, the negative regulation of RALF23–FERONIA on immunity can be reversed in a RALF23 dosage-dependent manner. In response to high bacterial load or pathogen infection, RALF23 accumulated over a threshold and triggered the cleavage of cytoplasmic-localized FERONIA. The cleavage product then relocalized to the nucleus and activated immunity (Tang et al. 2022). Yet, the microbiome titer and identities that trigger the cleavage and relocalization of FERONIA for immunity activation remain undefined. By contrast, activation of defense by flg22 treatment can inhibit phosphate uptake possibly through phosphorylation of the Pi transporter PHT1;4 (Dindas et al. 2022). Taken together, these studies suggest that the activation of immunity can compromise plant tolerance to abiotic stresses and vice versa, necessitating the consideration of microbial-mediated trade-offs in the presence of multiple stressors.

Differences in microbiota composition due to host genetic variation

Despite differences in the host lineages and their variable habitats, plants are invariably associated with a common set of microbial orders and genera, collectively known as the core microbiota (Yeoh et al. 2017). This shared core microbiota is also found in the soil-derived unicellular photosynthetic algae Chlamydomonas spp., suggesting a common principle of microbiota assembly mediated by photosynthetic organisms (Durán et al. 2022b). Despite the conservation seen in the microbiomes of photoautotrophic eukaryotes, variation in the composition of the microbial communities among different host species can be distinguished at the microbial operational taxonomic unit level (Yeoh et al. 2017). The influence of the genetic makeup of the host on microbiota taxonomic composition is subtle: natural sampling at regional and continental scales with transplant experiments under controlled conditions have shown that the soil biome accounts for arguably the largest source of variation explaining differences in plant-associated microbiome assemblage, followed by environment and plant genotype (Edwards et al. 2015; Wagner et al. 2016; Veach et al. 2019; Thiergart et al. 2020; Tkacz et al. 2020; Wagner 2021; Roux et al. 2023; Bamba et al. 2024; He et al. 2024b; Karasov et al. 2024). Among all the abiotic factors under investigation, water availability (i.e. drought) seems to be the most important one associated with the root microbiome, at least for A. thaliana (Thiergart et al. 2020; Karasov et al. 2024). The evolutionary mechanisms underlying microbiota codiversification with land plant host phylogeny remain poorly understood but must be at least in part attributed to the ability of the host plant to shape its microbiome. Plant roots secrete carbon-rich exudates to nourish their surrounding microbiome, and microbial capacity to metabolize specific carbon sources can explain differences in community composition (Zhalnina et al. 2018; Schäfer et al. 2023).

Root exudate contains numerous specialized metabolites, many of which have yet unknown biological functions. It is nevertheless well established that some of these metabolites serve to recruit or deter microbial colonization. This complex exudate chemistry with the substrate preference of microbes likely serves to fine-tune microbiomes by acting on functional differences among closely related microbes (i.e. at low taxonomic ranks; Wagner 2021). Inter- and intraspecific plant genetic variation plays a significant role in shaping microbiota. For instance, in a field study involving Boechera stricta derived from 5 natural populations grown across central Idaho, the sampling site, followed by plant genotypes and sampling year, accounted for the biggest variation in the β diversity of root- and shoot-associated bacterial communities (Wagner et al. 2016). A fundamental caveat with all microbiota-profiling studies in natural environments is that current methods are insufficient to resolve individual microbial strains or infer functional traits of the microbes; therefore, genetic interactions with the host are only roughly estimated (Wagner 2021). Thus, engineering crop genotypes to favor beneficial microbes is an attractive strategy, but the mechanisms of microbiome assembly and the genetic factors underpinning these mechanisms must first be understood (Cernava 2024).

The identification of host genes involved in modulating microbiota structure, referred to as Microbiome or M genes, has been challenging, as this trait has proved to be extremely polygenic. Studies using intraspecies natural variation have identified numerous plant quantitative trait loci (QTLs) associated with microbiota composition, with little overlap among studies (Horton et al. 2014; Brachi et al. 2022; Oyserman et al. 2022; Andreo-Jimenez et al. 2023; Edwards et al. 2023; Roux et al. 2023). Even within studies, analyses considering different microbiota traits, such as overall diversity or relative abundance of specific taxa, result in different sets of significant QTLs. As an extreme example, a study using Arabidopsis populations in southern France found different microbiota-associated QTLs in plants that had germinated approximately 1 mo apart (Roux et al. 2023).

Many genes linked to significant single-nucleotide polymorphisms or QTLs in these natural variation studies have been proposed as candidate genes affecting the microbiota. However, an extremely small number have been functionally validated, likely due to the small effect size of the individual QTLs. In a genome-wide association study correlating rice phyllosphere microbiota with plant genes (Su et al. 2024), several peroxidase genes implicated in diverse processes were associated with bacterial orders such as Burkholderiales, Pseudomonadales, and Xanthomonadales. In addition, the phenylalanine/tyrosine ammonia-lyase OsPAL02, involved in the synthesis of a lignin precursor 4-hydroxycinnamic acid, was associated with Pseudomonadales. Mutation and overexpression analyses further confirmed the contrasting enriching roles of OsPAL02 on Pseudomonadales but inhibitory roles on Xanthomonadales and Burkholderiales (Su et al. 2024), lending support to the possibility of targeted modulation of microbes, albeit at a higher taxonomic level. It is important to note, though, that knockout and overexpression of genes do not fully capture the genetic variation present within species, and additional approaches such as allele swaps are necessary to confirm that the natural polymorphisms are actually causal for the observed phenotypes.

To improve the chances of identifying useful genetic material for engineering the microbiome in a desired way, different strategies are needed. One approach is to study genetic variation among very closely related plant populations from a narrow geographic origin. Such an approach may increase the representation of rare alleles that nevertheless play important roles in the local adaptation of populations to their microbial environments (Roux et al. 2023). Another strategy is to focus on a more limited set of microbes, as plant genetic factors seem to unevenly modulate the microbiota, with stronger effects on a subset of hub microbial taxa (Brachi et al. 2022). Although many proposed M genes have known immunity-related functions (Andreo-Jimenez et al. 2023; Edwards et al. 2023; Roux et al. 2023), severely immunocompromised mutant plants show only mild alterations in their overall microbiota composition (Lebeis et al. 2015; Chen et al. 2020; Pfeilmeier et al. 2021). By focusing initially on mutants that show increased abundance of rhizosphere Pseudomonas in a simplified plate-based assay, Song et al. (2021) and Song et al. (2023) showed that the immune-related cell surface receptors FERONIA and PSKR target specifically the abundance of Pseudomonas in the rhizosphere through regulation of immune signaling and ROS. The role of FERONIA and PSKR in controlling Pseudomonas relative abundance was then confirmed in a natural soil setting.

M genes provide potential targets to engineer the microbiome through breeding or biotechnology. For example, the contribution of OsNRT1.1B polymorphism to differential root microbiome recruitment and nitrogen use efficiency was reproducible in a field experiment using near-isogenic lines with the indica or japonica OsNRT1.1B allele in the Nipponbare genetic background (Zhang et al. 2019). If target genes are known, CRISPR-Cas9 gene-editing technology provides the potential to improve crop varieties relatively rapidly (Zsögön et al. 2018). However, candidate M genes reported to date, including OsNRT1.1B or OsPAL02, have broad roles in plant growth, and manipulation of these genes would likely come with trade-offs in different environments (Wang et al. 2020b). A possible strategy to mitigate these trade-offs is to include data about microbial communities into existing precision agriculture frameworks, in which many plant and environmental variables are measured (French et al. 2021). The data are then used to inform crop management decisions to optimize plant performance with consideration for the specific plant genotype and environmental conditions.

Shaping of microbiota by metabolites

Differences in metabolite composition are an important mechanism by which host genotype can influence plant–microbe interactions. Several recent studies have analyzed metabolites from plants of different genotypes or under different stresses and attempted to correlate the differences in metabolite profiles with differences in microbiota community composition (Vismans et al. 2022; Caddell et al. 2023; Pantigoso et al. 2023; Thoenen et al. 2023; Yang et al. 2023a; Jin et al. 2024; Liu et al. 2024b; Su et al. 2024). Some of the identified compounds have direct effects on specific microbial taxa that could explain the altered relative abundance of these taxa. Benzoxazinoids (BXs) are a widespread group of specialized and diversified metabolites in grasses, often exuded by roots and exerting diverse roles ranging from allelopathy to insect deterrents and antimicrobial activity in culture, with functionalities determined by BX subclasses (Robert and Mateo 2022). Bacteria that are more sensitive to MBOA, the most abundant BX in the rhizosphere, are enriched in the rhizosphere of BX-deficient maize mutants as compared with wild type plants, suggesting that the presence of BXs shapes the microbiota by reducing the abundance of these sensitive microbes (Thoenen et al. 2023). Plant-secreted compounds can also promote the growth of specific microbes. Under phosphate-limiting conditions, the PSR limits plant growth and alters root exudation (Yang and Finnegan 2010; Ziegler et al. 2016). Rice lines that overexpress PHR2 show a constitutive PSR and release higher amounts of succinic acid and methylmalonic acid even under phosphate-replete conditions. These organic acids stimulate the growth of phosphate-solubilizing Pseudomonas strains, while lactic acid, which is reduced in the PHR2-overexpressing rice, does not (Liu et al. 2024b). As another example, barley varieties that differ in their root exudate profiles show rhizosphere enrichment of Pseudomonas with corresponding metabolic preferences: Pseudomonas strains isolated from an elite variety that secretes high amounts of simple hexose sugars grew better on glucose as compared with acetate, while the opposite was observed in Pseudomonas isolated from a variety with a more complex exudate profile (Pacheco-Moreno et al. 2024). The Pseudomonas communities from the 2 barley varieties also showed differential prevalence of several metabolism-related genes, although the contribution of these genes to the metabolic preferences remains to be fully tested.

However, many metabolites do not seem to modulate community composition through simple feeding or antibiosis but may rather act as infochemicals, altering the interactions of microbes with each other or with the host. Loss of myo-inositol exudation decreases colonization capacity of diverse bacteria on poplar and Arabidopsis roots (O’Banion et al. 2023; Sánchez-Gil et al. 2023). Although myo-inositol can be an energy source for bacteria, Pantoea Δiol mutants that lost the ability to catabolize myo-inositol did not show reduced colonization capacity. Instead, inositol perception increases motility, biofilm formation, and/or iron-scavenging activity of the bacteria, depending on the genus.

To understand metabolic interactions shaping the microbiota, the metabolic activity of the microbial community must be considered through modification of host-derived metabolites and de novo biosynthesis. For example, although MBOA is the most abundant BX in the rhizosphere, it is actually derived from DIMBOA-Glc exuded by maize and can be modified through microbial activity to AMPO, which has a very different toxicity profile (Thoenen et al. 2023; Thoenen et al. 2024). Microbes also produce a huge diversity of exometabolites, some of which have direct antimicrobial activity. As described previously, Pseudomonas R401 produces at least 2 exometabolites with broad-spectrum activity against different bacteria and thereby exerts a strong impact on overall community structure (Getzke et al. 2023). Communities with mutant R401 lacking both specialized metabolites support a higher relative abundance of strains that are antagonized by R401 in pairwise assays in synthetic growth medium as well as reduced relative abundance of the disarmed R401 itself in planta.

Like host metabolites, microbe-derived metabolites can shape microbiota community composition indirectly by acting as signaling molecules. The pathogenic fungus Fusarium oxysporum causes alteration in the root microbiota during infection on tomato roots, which is dependent on the production of fusaric acid by the fungus (Jin et al. 2024). However, the community shift is also dependent on the host, as the change occurred in the systemic compartment of a split-pot system, where the roots and microbes were not exposed to the pathogen. The community shift further varied according to host genotype, with a resistant cultivar showing enrichment for Sphingomonas, a bacterial genus known for its disease-protective activity (Innerebner et al. 2011). Accordingly, the microbial community isolated from roots of an infected resistant cultivar reduced the disease severity on a susceptible cultivar (Jin et al. 2024). The resistant cultivar showed specific changes in root exudate composition, including increased abundance of several compounds that stimulate Sphingomonas growth in culture. While the specific signaling mechanisms are still unknown, it appears that plants can perceive molecules from nearby microbes and respond with metabolic changes of their own, which in turn shape microbial growth.

Given the complexity of metabolic signals and interactions between host plants and the microbiota, using plant metabolites to shape the microbiota in a targeted way appears daunting. However, in several examples discussed here, changes in microbiota composition could be at least partially explained by the results of reductionist pairwise interactions (Getzke et al. 2023; Thoenen et al. 2023). By capitalizing on the empirical knowledge on the microbial utilization of rare metabolites, it is possible to enrich microbial groups in the natural community. Such an idea has been exploited in the transgenic production of opines. Opines are sugar condensates commonly found in Agrobacterium-infected crown galls, in which opines serve as resources to provide an advantage for the opine-inducing and opine-utilizing strains (Guyon et al. 1993). Some non-Agrobacterium cheaters have evolved the capacity to utilize opine. In line with this, transgenic Lotus plants producing opines lead to the enrichment of opine utilizers, which usually exist in low abundance as compared with wild type Lotus (Oger et al. 2004). Yet, it may be difficult to find specific metabolites that can be utilized by only 1 microbial group (Schäfer et al. 2023). In addition, application of nonnatural synthetic compounds such as artificial sweeteners in nature needs to be carefully evaluated for their impact on different ecosystems (Westmoreland et al. 2024).

ML-assisted microbiome study

As described here, the factors governing microbiome assembly, functioning, and interaction with the plant host are highly complex. In response, plant microbiome studies are increasingly generating highly dimensional multiomics data. To make sense of these datasets, we can seek assistance from ML. ML pipelines are powerful tools in microbiome studies for recognizing and classifying large input datasets to train models with predictive power. In a field study covering 54 sites across Switzerland, soil microbiome parameters including fungal abundance could be used to predict plant growth performance after inoculation with AMF (Lutz et al. 2023). The microbiome found on potato tubers is strongly associated with the sprout microbiome after planting (Song et al. 2024; Song et al. 2025), leading to the use of microbiome composition (e.g. at the amplicon sequence variant level) to predict potato vigor grown in the field. The resulting model predicted with approximately 50% accuracy whether the vigor of a seed lot would rank within the highest one-third for samples collected from the same year. However, the accuracy dropped a lot to approximately 18% for prediction across years. Models integrating genomic annotations with empirical carbon source utilization data were recently used to predict the outcomes of interactions between bacterial strains in pairwise interactions and small SynComs. Such an approach leads to the prediction of positive interactions that otherwise cannot be predicted by carbon utilization data alone (Schäfer et al. 2023). ML was also applied to identify microbial patterns correlated with pathogen susceptibility (Emmenegger et al. 2023). To reduce the underlying complexity, one can split large microbial communities into randomly assembled smaller ones, followed by matching phenotyping (i.e. pathogen load quantification). The trait of interest is then correlated with community characteristics such as commensal abundance, community evenness, phylogenetic diversity, and strain identities to select the most important features for model training. This approach successfully predicted strains and their combinations that significantly contributed to pathogen resistance (Emmenegger et al. 2023). In a study using 129 maize accessions, the authors trained an ML model combining genomic markers and microbiota composition data to predict agronomically important traits such as nutrient content and biomass. Although the best model had a relatively low prediction accuracy of 29% (He et al. 2024b), it is not surprising considering the complexity of the underlying interactions among the host, environment, and microbiome.

Although artificial intelligence has opened up a new avenue for understanding plant microbiomes, several notes of caution need to be emphasized. The quality of the dataset used for ML is critical. As the saying goes, “garbage in equals garbage out”: the source data have to be carefully selected. Datasets from microbiota profiling data are often sparse (i.e. containing many zero values), and the number of features per sample is much higher than the number of samples, which may lead to reduced reliability of the resulting model (Hernández Medina et al. 2022). No matter how powerful the model is, it does not provide any underlying mechanistic insight unless it can be correctly interpreted. As in the aforementioned example using microbial patterns to predict pathogen resistance, ML provides no hints regarding the mechanism of resistance conferred by specific microbiota members. In many cases, researchers may choose complex nonlinear models under the assumption that they will perform better than classical linear models (Topçuoğlu et al. 2020). However, a direct comparison of different models trained on microbiota profiling data has shown that linear models may be nearly as accurate as newer complex ML algorithms, with the added advantage that linear models are computationally less expensive to train and can be easily interpreted by feature weights. Thus, when choosing an ML approach to analyze microbiome data, researchers should consider whether a nonlinear model is necessary and how they will interpret their model post hoc. A permutation approach may be used with nonlinear models to interrogate which taxa have the largest effect on the model output, but the results of this analysis have been shown to vary according to model type. Other theoretical approaches, such as SHAP (Shapley additive explanations), can be used to rank the feature importance contributing to a model, providing the mathematical basis for rationalizing the prediction (Lundberg and Lee 2017). ML models do not consider the relative abundance of each feature in isolation but rather in the context of the other microbial populations. Thus, even when information can be extracted from the model—for example, which taxa most influence the model's output—this information may be overly simplistic. Nevertheless, the outputs of ML models provide a basis to generate hypotheses about the microbiome that can then be experimentally validated.

Delivery of microbes

Simply identifying plant-beneficial microbes is not sufficient for their use in agricultural systems: if the desired strains are not found naturally in the field location, they need to be applied as microbial inoculants. Only a limited number of studies have been dedicated to understanding the efficiency and optimal methods for microbe delivery to the plant host at the field scale. Three methods—namely, foliar spray, soil inoculation, and seed coating—are commonly used to deliver microbes to plants (Rocha et al. 2019; Sivaram et al. 2023). Currently, microbial preparations usually focus on spore-forming microbes such as Bacillus, as the spores survive extreme environments and enable long-term survival, making them suitable for long-term storage (Ptaszek et al. 2023). A carrier is usually used that comes in different forms: sterile vs. nonsterile, liquid vs. solid, organic vs. inorganic. These carriers all have their own strengths. However, they generally share properties such as high water-holding capacity to minimize desiccation, protection of microbes under adverse environments, and extension of the microbes’ shelf life. Practical considerations for carrier choice include cost, material availability, biodegradability, ease of application, and rate of release. Peat is by far the most common carrier, but it suffers from source-dependent quality fluctuation. Other carriers include alginate, vermiculite, chitosan, biochar, and different polymers that can be used to encapsulate microbes (Rathore et al. 2013). There is no consensus in the field on which carrier and method of delivery are the best, and empirical field testing is necessary. The use of different forms of carrier can significantly affect the dispersal and distribution of microbes that can be far away from the target crop. Carriers with differing organic matter and mineral nutrient content could modulate the composition of the microbial consortia and the behavior of the microbes and hosts. For example, the use of peat, rock phosphate, and vermiculite as carriers for AMF results in different colonization efficiencies and seed germination rates in a host-dependent manner (Barazetti et al. 2019). Such differences likely reflect the impact of nutrient availability on AMF development; for instance, high phosphate inhibits AMF development (Breuillin et al. 2010). Immobilization of PGPR in chitosan/starch beads increases bacterial survival rates on maize roots as compared with direct liquid application after 21 d (Fernández et al. 2022). The difference is enhanced by using sterile beads as the carrier, possibly indicating the importance of a stabilization effect and reduced competition with native microbes for focal strain survival.

Direct foliar and soil inoculation of microbes is relatively easy but likely not feasible for large-scale field application without extra capital input including machinery. As an alternative, microbes can be inoculated onto young flowers, leading to the vertical transmission of specific microbes to seeds. This method for seed inoculation has been demonstrated on a variety of crops, including wheat and soybean. The PGPR Paraburkholderia phytofirmans PsJN has been successfully inoculated onto young flowers by spray inoculation with zeolite as a carrier. This approach results in a change in seed microbiota composition, enrichment of the focal PGPR strain, and concomitant improvement in plant performance (Mitter et al. 2017).

Concluding remarks

Global agricultural systems currently rely heavily on chemical inputs for achieving high yields (Brunelle et al. 2024). Although these inputs are often inexpensive and easy for growers, their use commonly leads to environmental problems. For example, production of nitrogen fertilizer contributes substantially to greenhouse gas emissions, and chemical pesticides often have toxic effects on nontarget organisms. Microbiome-based strategies offer an alternative to maintain crop productivity while reducing reliance on chemical solutions. Application of beneficial microbes, especially rhizobia and AMF, is already widespread in agricultural systems to improve plant nutrition (Compant et al. 2025). A number of microbiome-derived strains have also been registered for agricultural use, mostly as biopesticides. A commercially available Bacillus amyloliquefaciens QST 713 strain has been shown to improve potato yield and quality characteristics in multiple field studies, although the effect varies substantially with the field site and/or potato variety (Imam et al. 2021; Adamo et al. 2024). In one study, QST 713 reduced symptoms of pathogenic Streptomyces on potato tubers even when applied with only 50% of the usual dose of chemical fungicide, demonstrating the potential for microbial products to reduce chemical inputs (Schirring et al. 2023). Microbes for use as biofertilizers, such as phosphate-solubilizing microbes, are also under development, although their commercial deployment has so far been limited (Soumare et al. 2020). The development of microbiome-based products requires a wealth of knowledge accumulated over decades regarding how the microbes interact with their host plants to colonize the roots and provide nutrients. For example, it has taken >30 years since the report of the nitrogen-fixing endophytic bacteria Gluconacetobacter diazotrophicus (Cavalcante and Dobereiner 1988) for this bacterium to be commercialized in field application as a biofertilizer. However, even when commercial microbial inoculants are brought to market, the claims of their beneficial impact on crop performance are often overstated or even completely unfounded, as in the case of diazotroph bacteria applied to nonleguminous crops (Giller et al. 2025). As discussed previously, the application of beneficial microbiomes is limited by their host preference and interaction with the environment. Thus, development of effective commercial products that leverage the beneficial effects of the microbiome will require more in-depth knowledge of the mechanisms underpinning plant–microbiome interactions.

The plant microbiome has undoubted potential to improve crop performance and stress tolerance, and harnessing this potential will be an important strategy for future sustainable agricultural practices. Commercial application of microbiota products is currently limited due to challenges in identifying the right microbial strains and deploying them effectively at scale, with production in large fermenters often being a cost and scientific bottleneck. Integrating large-scale data collection and ML-assisted prediction has provided a great deal of information about the plant–microbe and microbe–microbe interactions within the microbiota (Compant et al. 2025). Microbiota research should also prioritize reductionist experimental approaches to provide mechanistic insights into the microbiome, which will be necessary to fully leverage the beneficial potential of microbes (Northen et al. 2024). Although available microbiota culture collections do not provide a complete representation of microbiota members identified through culture-independent approaches (Bai et al. 2015), these gaps could be closed through targeted microbe cultivation with customized media. A deeper understanding of the genetic determinants and environmental factors governing microbial invasion and persistence into existing communities is crucial for the design of future knowledge-based inoculants with robust microbiome function. In addition to traditional methods for isolating strains with beneficial traits, recent advances in microbial engineering involving experimental evolution offer possible alternatives to obtain strains with agronomically favorable traits, such as strains with the combination of beneficial traits with high colonization capacity but low pathogenic potential (Li et al. 2021a; Li et al. 2021b). For large-scale cropping systems with diverse standing soil communities, microbiome engineering may be more feasible than direct application of microbial inoculants (French et al. 2021). For this approach, increased understanding of how host genetics affects microbiome assembly will facilitate development of crop varieties with improved microbiome-related traits. Application of metabolites based on empirical resource utilization data is another possible way to enrich targeted microbial groups already existing in the soil, without the need to apply them as an inoculant. Together, these advances should enable the development of microbiome-based strategies in agriculture with robust activities that are less vulnerable to environmental fluctuations.

Funding

C.C. was supported by a research fellowship from the Alexander von Humboldt foundation. K.-W.M. was supported by grants from NSTC (NSTC-113-2311-B-001-009-MY3) and Academia Sinica (AS-IAIA-114-L04). P.S.-L. was supported by the Max Planck Society, Deutsche Forschungsgemeinschaft (DFG, German Research Foundation) under Germany's Excellence Strategy-EXC 2048/1-Project ID: 390686111 and SPP 2125 DECRyPT.

Data availability

No new data were generated or analysed in support of this research.

Dive Curated Terms

The following phenotypic, genotypic, and functional terms are of significance to the work described in this paper:

• rbohD Gramene: AT5G47910

• rbohD Araport: AT5G47910

• myo-inositol CHEBI: CHEBI:17268

• PHR1 Gramene: AT4G28610

• PHR1 Araport: AT4G28610

• ALMT1 Gramene: AT1G08430

• ALMT1 Araport: AT1G08430

• PHR2 Gramene: Os07g25710

• PHR2 Araport: Os07g25710