Predicting Select Soil Health Genes using Hyperspectral Reflectance in Nematode-Infected and Drought Stressed Greenhouse Cotton

John P Brooks ^1,2* Martin Wubben ^1,2 Renotta Smith ^1,2 Josh Waldbieser ^1,2 Sathishkumar Samiappan ³ Purushothaman Ramamoorthy ³ Raju Bheemanahali ³

• ¹ Agricultural Research Service, United States Department of Agriculture, Washington D.C., United States
• ² Crop Science Research Laboratory, Agricultural Research Service (USDA), Mississippi, Mississippi, United States
• ³ Geosystems Research Institute, Mississippi State University, Mississippi State, United States

Predicting, or correlating, soil microbiome metrics with above ground phenotypic plant measurements would enable rapid diagnosis of soil microbiome imbalances. Rapid plant measurements through remote sensing are a leading innovation in agriculture and have reduced the need for labor-intensive plant and soil measurements. In the current study we utilized cotton (Gossypium hirsutum) as a plant model whereby stress was induced by drought and root-knot nematode (RKN; Meloidogyne incognita) infection to induce a change in the soil microbiome which would be reflected as a plant phenotypic response. The experiment was a randomized complete block design with two cotton genotypes (RKN-susceptible or RKN-resistant) and four stress combinations. Rootzone samples were collected upon plant termination and quantified for five soil health genes: 16S rRNA, 18S rRNA, ureC, phoA, and cbbLR. Plant physiology, biomass, and remote sensing hyperspectral readings were previously reported. Overall, RKN infection and plant genotype treatments had little effect on genes. Interestingly, drought stress increased most gene abundances, while plant physiological and biomass measurements decreased, indicating microbiome response to plant stress. Hyperspectral reflectance, through machine learning, accurately predicted the presence of drought stress with an area under the receiver operating characteristic curve value of 0.864. Furthermore, the readings were able to predict the abundance values for all genes except 18S rRNA within one standard deviation of ground truth levels. This study demonstrated that there are key plant characteristics that are registered via hyperspectral wavelengths which can be used to accurately predict soil health gene abundance. While the use of hyperspectral readings and soil microbiome status to inform plant health and vice versa are still in their infancy, the current study provides us with future directions towards this end.