Bioinformatic Extraction of Functional Genetic Diversity from Heterogeneous Germplasm Collections for Crop Improvement

Efficient utilization of genetic variation in plant germplasm collections is impeded by large collection size, uneven characterization of traits, and unpredictable apportionment of allelic diversity among heterogeneous accessions. Distributing compact subsets of the complete collection that contain maximum allelic diversity at functional loci of interest could streamline conventional and precision breeding. Using heterogeneous population samples from Arabidopsis, Populus and sorghum, we show that genomewide single nucleotide polymorphism (SNP) data permits the capture of 3-78 fold more haplotypic diversity in subsets than geographic or environmental data, which are commonly used surrogate predictors of genetic diversity. Using a large genomewide SNP data set from landrace sorghum, we demonstrate three bioinformatic approaches to extract functional genetic diversity. First, in a candidate gene approach, we assembled subsets that maximized haplotypic diversity at 135 putative lignin biosynthetic loci, relevant to biomass breeding programs. Secondly, we applied a keyword search against the Gene Ontology to identify 1040 regulatory loci and assembled subsets capturing genomewide regulatory gene diversity, a general source of phenotypic variation. Third, we developed a machine-learning approach to rank semantic similarity between Gene Ontology term definitions and the textual content of scientific publications on crop adaptation to climate, a complex breeding objective. We identified 505 sorghum loci whose defined function is semantically-related to climate adaptation concepts. The assembled subsets could be used to address climatic pressures on sorghum production. To face impending agricultural challenges and foster rapid extraction and use of novel genetic diversity resident in heterogeneous germplasm collections, whole genome resequencing efforts should be prioritized.