Local microbial yield-associating signatures largely extend to global differences in plant growth

Area of Science:

• Microbiome research
• Agricultural science
• Machine learning applications

Background:

• High-throughput DNA sequencing enables microbiome analysis for agricultural applications.
• Previous studies explored soil microbiomes for predicting crop yield and soil health locally.
• Generalizability of local microbiome-yield relationships to global scales remains unclear.

Purpose of the Study:

• To investigate if soil bacterial patterns predicting plant growth locally can be generalized globally.
• To develop and validate machine learning models for predicting agricultural productivity using soil microbiomes.

Main Methods:

• Measured soil bacterial microbiome composition in a maize field in Germany with high spatial resolution.
• Correlated microbiome data with high-resolution maize yield measurements.
• Applied machine learning (LASSO regression) to build predictive models and validated them using global datasets and vegetation index data.

Main Results:

• A locally trained LASSO model predicted approximately 65% of maize yield variation.
• The model, using 26 bacterial genera, showed correlations with global yield and plant growth metrics.
• It predicted up to 37% of global vegetation variation (NDVI), with specific genera like Hyphomicrobium being key predictors.

Conclusions:

• A globally conserved set of soil bacterial taxa correlates with vegetation and plant growth.
• These findings suggest the potential for using soil microbiomes to predict plant growth and agricultural productivity on a global scale.
• Specific bacterial genera consistently contribute to predicting plant growth across different regions.