Quantifying microbiota impact on plant traits for the guidance of breeding programs

Understanding the drivers of plant phenotypic variation is central to crop improvement. Root-associated microbiota are increasingly recognized as important drivers of plant growth and disease resistance, yet their quantitative contribution to phenotypic variation remains unclear. Here, we extend the classical genotype–environment framework by explicitly incorporating soil microbiota as a distinct factor. By combining contrasting genotypes, soil matrices and microbiota in a full-factorial experiment, we show that microbiota contribute substantially to plant trait variation, either directly or through interactions with genotype and environment, depending on the trait. We conclude by discussing the implications of these direct and interaction effects for the design of plant breeding programs. Understanding the sources of phenotypic variation in plants is central to both fundamental biology and agricultural innovation. Classically, plant phenotype (P) is modeled as the additive and interactive effects of plant genotype (G) and the environment (E), formalized as P = G + E + G × E (Falconer, 1989). This framework underlies most quantitative genetic approaches, allowing identification of loci associated with important traits. However, a substantial fraction of organism phenotypic variance often remains unexplained by genetic information alone – a gap referred to as ‘missing heritability’ (Maher, 2008). A convincing explanation could be that, besides plant's genes, additional information sources can be inherited across generations; the so-called ‘inclusive heritability’ (Danchin et al., 2011). In that context, one increasingly acknowledged, yet rarely mobilized, contributor to phenotypic variance is plant-associated microbiota (Lemanceau et al., 2017). Specifically, soil microbial communities are known to influence plant physiology, growth, and health, hence being now proposed by microbial ecologists as a promising way to breed plants (Wei Trivedi et al., 2022). However, plant geneticists are reluctant to incorporate microbiota into breeding programs, given the already extensive experimental designs involved and the limited knowledge of how strongly microbiota influence plant phenotype compared to plant genotype. Soil microbiota is often considered to be entirely part of the environment (E), due to its strong correlation with soil physicochemical parameters (pH, organic matter, etc.) (Bahram et al., 2018; Delgado-Baquerizo et al., 2018). However, rhizosphere and endosphere microbiota were found to be strongly influenced by host genotype, consistently across different soils, giving rise to the concept of a ‘core microbiota’ (Turnbaugh et al., 2009; Lundberg et al., 2012; Lemanceau et al., 2017). Other studies have highlighted that both plant species and soil parameters contribute to shaping the rhizosphere microbiota (Berg Buitenhuis et al., 2019). Hereafter, we explicitly distinguish additive microbiability (M) from microbiota-driven interaction effects (G × M, E × M and G × E × M), which capture complementary, non-additive pathways by which microbiota modulate host phenotype in a genotype- and environment-dependent manner. Here, the classical model was enhanced to include the microbiota (M), yielding the extended formula: P = G + E + M + G × E + G × M + E × M + G × E × M (Oyserman et al., 2021). This captures the idea that microbiota can act as an independent factor (M), but also as a factor dependent on host genotype (G × M), environment (E × M), or both (G × E × M). This distinction has major implications for both plant (Nerva et al., 2022; Shen et al., 2024) and microbial breeding (Mueller Col for Columbia; Cvi for Cape Verdi Islands), E (three autoclaved soils from Burgundy: Auxonne, Breteniere, Champdôtre) and M (three entire microbial communities extracted from the three soils before autoclaving, used as inoculants) — to evaluate if microbiota's effect on plant was dependent (or not) on the plant genotype and/or the abiotic environment (Supporting Information Fig. S1). At the end of the experiment, we checked that inoculated microbiota changed the structure of the bacterial and fungal rhizosphere communities. We also verified that our design had enough replicates to detect significant effects thanks to a power analysis and quantified the effects of G, E, M and their interactions on shoot dry biomass and leaf necrosis size. The bacterial and fungal rhizosphere communities at the end of the experiment were primarily shaped by M and E, with G playing a minor but significant role (Fig. 1a,c). For bacteria, E had the strongest influence on explained variance (40.8%, P < 0.001), followed by M (17.7%, P < 0.001) and the E × M interaction (14.7%, P < 0.001), for a total of 73.2% with only these two factors and their interaction (Fig. 1a; Table S1). For fungi, M had the strongest effect (25.6%, P < 0.001), followed by E (17.2%, P < 0.001) and the E × M interaction (11.8%, P < 0.001), for a total of 54.6% with only these two factors and their interaction (Fig. 1c; Table S1). Specifically, for the effect of E, EChampdôtre strongly influenced bacterial communities (Fig. 1b, circles apart from triangles and squares for EBretenière and EAuxonne), and fungal communities to a lesser extent (Fig. 1d). This finding aligns with recent work underlying the major role of soil abiotic environment in reshaping transplanted microbial communities. For instance, Bamba et al. (2024) observed that E (e.g. salt-treated vs control media) significantly shaped root (and not rhizosphere) microbiota variance by 22%. Regarding the effect of M, bacterial communities after inoculation by MChampdôtre were strongly segregating from those inoculated with MBretenière and MAuxonne (Fig. 1b). Fungal communities that received MChampdôtre or MBretenière were more stable across environments than those inoculated with MAuxonne (Fig. 1d). This also echoes previous findings (Bamba et al., 2024) that M (microbial inoculants from adjacent salt-treated or untreated fields) was responsible for 22% of root microbiota variance. G effects on rhizosphere microbiota were significant but minor (bacteria: 0.76%, fungi: 1.10%, Fig. 1a,c; Table S1), close to previous findings showing 4% of root microbiota variation due to G (Bamba et al., 2024). Taken together, these results indicate that despite the well-documented influence of soil physicochemical parameters (Bahram et al., 2018; Delgado-Baquerizo et al., 2018), host genotype (Yadav et al., 2023), and their interaction (Berg Fig. S4). The dominant influence of E over M and G likely reflects differential soil fertility: the lowest shoot biomass was observed in the soil of Champdôtre, with the lowest N, P, K, and C content, and the highest shoot biomass was observed in the Auxonne soil which had the highest K content (however, the Bretenière soil had the highest N and P content). Other properties like texture, pH, CaCO3, and CEC could also interfere. M effect may be due to the involvement of microbiota in nutrient cycling and G effect due to differences in nutrient uptake efficiency. The G × E × M interaction explained only 4.0% of the variance (Fig. 2b), with notable genotype-specific responses depending on the soil: in EAuxonne, only GCan was affected by M (MAuxonne: +249.2%, MBretenière: +207.3% vs MChampdôtre); in EBretenière and EChampdôtre, only GCol responded to M (MAuxonne: +197.3% and MBretenière: +306.1% vs MChampdôtre). Hidden by the G × E × M interaction, there was a small significant G × M interaction (2.0%), confirming that genotypes were not responding in the same way to M across all soils (Fig. S5). Taken together, these findings suggest that microbiota effects on shoot biomass were of the same importance as G effects, primarily genotype-independent and mediated through resource availability and soil context. Pathogen resistance was strongly modulated by host genotype, with microbiota acting through genotype-specific interactions. In total, 63.5% of the variance in leaf necrosis size was explained by the model. The most influential factor was G (28.9%), followed by G × E × M (14.8%), E (6.3%), and G × M (5.6%, Fig. Table S1). M had significant effect However, when summing all its significant interactions, M for of necrosis size variance. This is higher than G × M and G × E × M contributions observed for Lotus japonicus traits and for shoot and and for root respectively) (Bamba et al., 2024). GCan showed the necrosis size and compared to GCol and Fig. consistently with its higher than genotypes et al., The strong influence likely reflects of host in interactions. depending on the microbiota GCol can as as more (Fig. indicating that M can for a of genetic was by M of necrosis size of GCan was mainly influenced by E (e.g. effect of MChampdôtre EChampdôtre and and GCol was the most to M, with from to depending on E (Fig. The G × M interaction that, with genotypes in the in terms of size of GCan GCol et al., whereas MChampdôtre the of GCol GCol and MBretenière its to the level of the most genotype GCol = (Fig. M with pathogen through genotype- and environment-dependent the idea that microbiota effects on disease resistance were mediated by the plant strongly contribute to plant phenotypic variance and may part of the in plant traits. We found that microbiability was to for M to 14.1% microbiota-driven interaction effects for shoot biomass, comparable with G effects. M interaction effects explained to of variance for leaf necrosis a effect of This is close to results in animal like in which the microbiota explained of variance, and host genotype both effects being largely independent (Difford et al., 2018). results the potential of soil microbial communities to contribute to the unexplained phenotypic variance – often referred to as ‘missing heritability’ (Maher, et al., 2011). studies experimental and microbiota et al., are needed to determine this variance is host genotype effects on microbiota are well (Yadav et al., 2023), our findings for into the of plant traits mediated by microbial communities (Mueller et al., 2019). The of autoclaved not soils could also modulate microbial effects, their due to In inoculants effects but could community In soils, may the of a in estimating M effects in autoclaved studies at M effects in the would require an and such as the of species in to community structure (Liu & 2024) and microbiota et al., findings that microbiota influenced plant depending on the targeted plant trait biomass or pathogen which has important implications for Specifically, the trait in should guide how microbiota are into programs. In evaluate genotypes, with three across to over to – to to et al., This results in × × to × × M as a factor can the experimental this for we considered genotypes, if is not a We considered to distinct microbiota comparable in to – with three on our shoot biomass was mainly influenced by direct microbial effect (M), largely independent of genotype (G) and environment G × M effect and E × M despite strong differences environments (E In to the most microbial screening to microbiota on a single genotype at a single with three replicates would require only M × × × to M × × × screening and Conversely, microbial effect on pathogen resistance dominant G × E × M and G × M interactions. for these would require the previous design, by it by the number of genotypes and to from M × × × to M × × × designs However, if only E × M or G × M interactions are significant not G × E × M), designs but For E × M to microbiota across to environments would require M × × × to M × × × For G × M genotypes with to microbiota would require M × × × to M × × × These reach for breeding that not all environmental conditions, particularly are to pathogen this information could be used to the number of environments the number of E, E × M and G × E × M traits with limited G and E is key to For traits G × E × M interactions, targeted experimental strategies or screening likely be needed before Arabidopsis thaliana a model organism with a at the level and for interactions, was used in this were on their to the pathogen et al., = and Cvi with Can for Canary Col for Cvi for Cape Verdi were from the Arabidopsis at distinct soils were in the a soil organic from Auxonne a soil organic from a at Bretenière a organic from before all soils for at and at were autoclaved at for at for to to at for in to that could after at for to to autoclaved a parameters and for at Soil was dry soil and soil microbial community inoculants (M), of soil at after to were in of to the of the corresponding were from soil a soil were to for and the only in were used as our soil microbial community inoculants, or to a was to the soil inoculants soil: Bretenière soil: soil: of the inoculants were by and the by to a in an of of dry soil were used to of dry soil for this at to an The of or in the was thus not likely to a significant effect on plant However, to that observed effects were due to in the inoculants, of inoculants, inoculants, and were allowing us to effects from inoculation abiotic in inoculants inoculants, two of at for and of inoculated We found that inoculants had the same effect as whereas inoculants had a significant effect (Fig. the of the soil was the lowest of for Bretenière and for we also verified through a experiment that the not the of the inoculated soils (Fig. had a small effect on the soil pH, as compared with differences soils and for Auxonne and Bretenière when compared to Champdôtre, respectively). et al., a was for its relevance to thaliana et al., The was for in the at on at with were in through and to × in were by of the on the the after and or due to received fungal growth, plants were and in the of were to A G × E × M design was (Fig. three soil abiotic environment three three extracted from This in with biological replicates plant Fig. S1). The entire design was to include plants plant Fig. S1), in and The plants that not the pathogen as an on the leaf of a control only the the were to potential factors due to our inoculation in the and were in a glasshouse for in the Auxonne and soils, in the soil due to a growth due to pH, which was to in was with of dry were to of were only one was in in to have the most were three a with the same of one to to of were three a detection of on of was following the At the end of the experiment or in the and were and at in an to shoot dry was measured on the plant inoculated with and At rhizosphere soil was in by root to only soil and at was extracted from the rhizosphere soil from all replicates of with the were quantified and to in and fungal communities were the following in were in a and the to total were by and × on the by used a the experimental were for replicates and rhizosphere soil as the plants were The and the and analysis of and were the by the in with the with et al., 2018) were & with a at were through with the et al., of control screening for and detection of host from Arabidopsis For the et al., was to by with to thaliana were two for and for the of of the total and of the total we to all and the and and for A analysis was on the followed by a on of a on of and a of with a and were were the and were at and for and to the (Fig. below these were not considered for were for the and for This and biological replicates to × E × for the and The bacterial and fungal The of not the of detected effects (Fig. were 2023). The was into two control plants were to plant dry shoot biomass, and plants were for size as a for disease For the biomass all plants were considered In the pathogen replicates were on to by the size et al., For biomass, was verified by the of the of model and by of the shoot biomass not a was after identification of the optimal the 2021). A for on data to the of three on in to was thanks to a of < of biomass size data were had variance and were thus were to the effects of genotype soil environment (E), and microbiota (M) on both biomass and disease after a The proportion of variance explained by effect was as the proportion of of squares to the total of variance (Table S1). were from the same soil for the dry biomass, and the same genotype for the necrosis < For only significant an effect of the microbiota (M) were (Fig. 2). soil shoot or genotype necrosis are in Fig. the of observed effects, a power analysis was a simulation = & Fig. S2). in microbial community structure were with a analysis on the for bacterial and fungal communities. A analysis was to the effect of the pathogen inoculation on the structure of microbial effects The variance analysis was hence the following model for both the bacterial and fungal × E × M, with (Table S1). were through a analysis the to model followed by the We would like to and of the for for their and help with plant We from the resource for Arabidopsis thaliana at This work was by a from and and by – – & – & data – & The plant trait experimental design and the have been on the and are The have been in the data of the biomass data of the pathogen data of the biomass data of the pathogen and are Fig. design to the G × E × M model. Fig. analysis for the dry biomass and necrosis size. Fig. of the soil abiotic environment on plant dry shoot Fig. of the soil microbiota (M) on plant dry shoot Fig. of genotype × microbiota interaction on plant dry Fig. of genotype on leaf necrosis size. Fig. of genotype × microbiota interaction on the leaf necrosis size. Fig. shoot biomass of plants inoculated with and soil microbial Fig. A experiment to of from the Fig. of the bacterial and fungal Fig. structure of and on Fig. Shoot dry biomass of Arabidopsis thaliana and necrosis size due to Table for the microbial communities and the plant traits. Table properties of the three is not responsible for the content or of Information by the than should be to the The is not responsible for the content or of information by the than should be to the corresponding for the The remains with to in and in