Soil microbiome indicators can predict crop growth response to large-scale inoculation with arbuscular mycorrhizal fungi

Agricultural intensification has boosted yields but caused biodiversity loss, soil degradation, pollution, greenhouse gas emissions, and eutrophication. Reducing environmental impacts while ensuring food security requires alternatives to agrochemicals. Arbuscular mycorrhizal fungi (AMF) are key beneficial symbionts that enhance plant nutrient uptake and stress tolerance, and can improve soil structure and resistance to drought and disease. However, field AMF inoculation outcomes are highly variable, and the mycorrhizal growth response (MGR) can range from beneficial to detrimental, with limited understanding of when introduced AMF establish and yield benefits occur. This study tested whether large-scale field inoculations with AMF are feasible and whether inoculation success can be predicted from readily measured soil physicochemical parameters and soil microbiome indicators. The goal was to identify predictors of MGR to enable a soil diagnostic tool that informs farmers about likely benefits of AMF inoculation.

Prior work shows AMF form symbioses with most terrestrial plants, exchanging soil-derived nutrients (notably phosphorus) for plant carbohydrates and often enhancing growth in greenhouse and some field settings. Reported benefits include improved nutrient uptake (especially P), better soil structure, nutrient retention, reduced greenhouse gas emissions, and increased drought and disease resistance. Two strategies are used in practice: promoting native AMF via management (low tillage, diversification, organic farming) or introducing AMF inocula into soils. Field inoculation outcomes are context-dependent and variable, with MGR influenced by soil and plant factors; negative effects have been observed in some contexts. Previous studies often suggested P availability mediates AMF benefits and that added P can reduce MGR, but the predictability of inoculation outcomes under field conditions remains limited. No prior large-scale field study had integrated soil characteristics and molecular microbiome profiling to predict conditions under which AMF enhance crop growth.

Design and sites: Field inoculations were conducted over three years in 54 maize fields in northern Switzerland (2018: 22 fields; 2019: 25; 2020: 12). Farmers managed fields per Swiss conventional standards; trials manipulated AMF inoculation and, in a subset, P fertilization. In 2018, a split-plot design compared fertilizer types (NK vs NPK) with control vs AMF inoculation as split plots (8 replicates). In 2019–2020, randomized complete block designs with 8 blocks were used. Buffer rows minimized edge effects. Inoculum and crop: The native AMF Rhizoglomus irregulare isolate SAF22 was produced in greenhouse cultures using Plantago lanceolata in sterilized soil:sand. After sowing maize (variety LG 30.222), seed furrows were reopened and 450 g inoculum (or carrier control) mixed into an 80-cm row segment (~5% v/v). Seeds were replaced; identical seed coatings (standard fungicides, Mesurol) were used across treatments. Fertilization: All fields received N and K; a subset of 2018 plots also received P (triple superphosphate) to test P effects on inoculation success. Fertilizer rates followed Swiss PRIF recommendations. Sampling and measurements: Before fertilization, composite soil samples (20 cores per field) were collected, sieved (2 mm), and analyzed for 52 soil parameters (chemical, physical, biological), with 38 available across all fields. Soil microbial biomass carbon (CMIC) and respiration were measured. At harvest (~4–5 months), biomass of two plants from the center of each plot (16 plants per treatment per field) was measured as fresh and dry weight (strongly correlated). MGR (%) was calculated as the percent change in biomass in AMF plots relative to control plots; fields were categorized by MGR quantiles (low, medium, high). Root samples were collected for AMF colonization (microscopy; magnified intersection method with ink-vinegar staining) and DNA-based profiling. Microbiome profiling: Soil fungal communities were profiled by long-read PacBio SMRT sequencing of the full ITS region using ITS1F/ITS4 primers. Root fungal communities were profiled using AMF-enriching primers SSUmCf/LSUmBr to span SSU–ITS–LSU (~1.5 kb) with touchdown PCR and PacBio SMRT sequencing. Negative controls showed no amplification; positive controls identified rRNA variants of SAF22. Bioinformatics used DADA2 for denoising/chimera removal and DECIPHER for clustering to 97% OTUs. Taxonomy was assigned with UTAX/UNITE; key OTUs were verified by BLAST. Soil and root OTU tables were rarefied (soil: 4,272 reads; roots: 1,660 reads). Community ordination (PCoA, partial dbRDA) assessed year effects and relationships to MGR. Variable selection and modeling: For soil parameters, strongly correlated variables (|r|>0.8) were removed, reducing 38 to 22, then to 15 using three approaches: random forest (1,000 trees), stepwise AIC (stepAIC), and exhaustive glmulti screening (AICc; max 10 predictors). For soil fungi, candidates were identified by indicator species (P<0.1), DESeq2 differential abundance (P<0.1), and random forest importance, yielding 44 sOTUs, further reduced to 13 by glmulti (7 associated with high MGR; 6 with low MGR). Linear regression models combined year, the 15 soil parameters, and the 13 sOTUs. Predictor importance was quantified (relaimpo). Cross-validation used 1,000 iterations of random 90/10 train-test splits, with binary classification of inoculation benefit defined as MGR>12.2% (significant positive effect) vs ≤12.2% (neutral/negative). Statistical analyses were performed in R with standard packages. Additional analyses: AMF establishment in roots was quantified as relative abundance of SAF22 rOTUs in inoculated vs control plots and by total root colonization; correlations with MGR were tested. A subset of fields tested P addition effects and, in 2019, inoculations with other AMF species (Funneliformis mosseae, Claroideoglomus claroideum) individually and combined.

- Mycorrhizal growth response (MGR) across 54 maize fields ranged from −12% to +40%; 14 fields (≈26%) had significant positive responses (+12 to +40%), while 2 fields had significant negative responses (−12%). - Pairwise correlations between individual soil parameters and MGR were few and weak; multivariate modeling was required. - Final predictor sets: 15 soil parameters (including Mg (H2O/EDTA), Mn, Fe, mineralized N (Nmin), ammonium, CMIC, texture, pH, P measures) and 13 soil fungal sOTUs (7 high-MGR associated: e.g., Trichosporon, Myrothecium, Olpidium, Chaetomium, Fusarium, Cladochytrium; 6 low-MGR associated: e.g., Powellomyces, Phaeohelotium, Phaeosphaeria, unknowns). - Full model (year + 15 soil parameters + 13 sOTUs) explained 86% of MGR variation (P<0.001). Relative importance: soil fungi predictors 53%, soil parameters 29%, year 3.7%, unexplained 14.3%. - Reduced model (10 predictors: 6 sOTUs plus Nmin, CMIC, ammonium, Mg (H2O)) explained 68% (P<0.001). Soil-fungi-only model (13 sOTUs) explained 66% (P<0.001). - Binary classification (benefit yes/no at MGR>12.2%): mean accuracy 80% (full model) and 83% (reduced and fungi-only models). - Abundance of pathogenic fungi in soil was the best single predictor group of AMF success (≈33% of explained variance), outperforming nutrient availability; phosphorus explained <2% of MGR variance, and a fertilizer trial showed no consistent P effect. - AMF establishment (relative abundance of SAF22 rOTUs in roots and total root colonization) did not correlate with MGR (rho≈0.08 and 0.01; both ns), although SAF22 abundance and total colonization correlated with each other (rho=0.54, P≈3×10⁻⁵). - Root microbiome shifts supported a pathogen suppression mechanism: in high-MGR fields, inoculation reduced relative abundances of pathogenic/root-associated taxa (e.g., Olpidium, Cladosporium, Mycochaetophora, Pyrenochaeta, Vishniacozyma), alongside reductions of native AMF; AMF diversity did not consistently decrease. - sOTU18 (Trichosporon sp.) was the most important high-MGR predictor; its abundance correlated with poor soil properties (lower organic C/fertility, higher sand). Low-MGR indicator sOTU58 correlated with healthier soil properties (higher respiration, fertility, organic C). - Year effects were minor but present: 2019 had fewer significant positives (16% of fields) vs 2020 (41%). Inoculations with other AMF species or combinations did not significantly differ in MGR from SAF22 in the subset tested.

The study demonstrates that field AMF inoculation effects on maize yield are strongly context-dependent but can be predicted with high accuracy using soil microbiome indicators, particularly the abundance of specific pathogenic fungi, rather than relying on soil nutrient metrics. The combined models explained up to 86% of MGR variation, achieving 80–83% accuracy in a practical yes/no recommendation framework. Contrary to a purely nutritional explanation, inoculation success aligned with pathogen suppression: AMF inoculation reduced root-associated pathogenic taxa in high-MGR fields, consistent with mechanisms including priority effects (rapid AMF establishment), induced systemic resistance, microbiome modulation, and competition for root niches. Phosphorus availability, often implicated in AMF responsiveness, contributed minimally here and P fertilization did not significantly alter outcomes, likely because most soils exceeded deficiency thresholds. Fields with lower soil organic carbon and microbial biomass benefitted more, suggesting AMF inoculation may be particularly valuable where soil health is relatively poor. Notably, AMF establishment metrics did not predict yield response, indicating that functional outcomes (e.g., pathogen suppression) rather than colonization per se determine benefits. The identified fungal indicators provide a feasible basis for rapid soil diagnostics to guide inoculation decisions and enhance the reliability and profitability of microbiome-based agronomy.

This large-scale, multi-year field study shows that maize growth responses to AMF inoculation can be robustly predicted using a small set of soil microbiome indicators and a few soil parameters. Models explained up to 86% of MGR variance and delivered 80–83% decision accuracy for recommending inoculation. Pathogen-associated soil fungi were the dominant predictors, highlighting pathogen suppression as a key mechanism of AMF benefit under field conditions, whereas phosphorus availability was of minor importance. The work provides a practical foundation for developing rapid, affordable soil diagnostic tools (e.g., qPCR or rapid sequencing assays targeting key sOTUs) to guide AMF inoculation. Future research should test different crop varieties and regions, include broader soil types and climates, assess reduced-agrochemical systems, evaluate multiple AMF genotypes/inoculum consortia, and investigate long-term impacts, persistence, and potential invasiveness of inocula. The predictive framework can be extended as a blueprint to other biofertilizers (e.g., Rhizobium spp., Bacillus amyloliquefaciens).

- Geographic and crop scope: Trials were conducted in northern Switzerland on a single maize variety (LG 30.222); generalizability to other varieties, crops, and regions needs testing. - Temporal variability: A year effect was observed; weather and microclimate likely contributed to unexplained variance. - Marker and primer biases: Root profiling used AMF-enriching primers, limiting detection of several pathogens (e.g., Fusarium, Myrothecium) in roots; fungal lifestyle inference from marker genes is limited. - Unmeasured factors: Bacterial microbiome composition, pesticide residues/use, and microclimatic factors were not included and could explain additional variance. - Establishment vs function: While establishment did not predict MGR here, functional assays linking individual pathogens and AMF-provided protection require experimental validation (isolation and pathogenicity testing of key taxa). - Long-term effects: Persistence, invasiveness, and impacts on soil and plant biodiversity of native vs non-native inocula were not assessed over multiple seasons.