Abstract
Mexican native maize (Zea mays ssp. mays) is adapted to a wide range of climatic and edaphic conditions. Here, we focus specifically on the potential role of root anatomical variation in this adaptation. Given the investment required to characterize root anatomy, we present a machine-learning approach using environmental descriptors to project trait variation from a relatively small training panel onto a larger panel of genotyped and georeferenced Mexican maize accessions. The resulting models defined potential biologically relevant clines across a complex environment that we used subsequently for genotype-environment association. We found evidence of systematic variation in maize root anatomy across Mexico, notably a prevalence of trait combinations favoring a reduction in axial hydraulic conductance in varieties sourced from cooler, drier highland areas. We discuss our results in the context of previously described water-banking strategies and present candidate genes that are associated with both root anatomical and environmental variation. Our strategy is a refinement of standard environmental genome-wide association analysis that is applicable whenever a training set of georeferenced phenotypic data is available.