The accuracy of trait measurements greatly affects the quality of genetic analyses. In automated phenotyping pipelines, phenotyping errors are often treated as random effects that can be controlled by increasing population sizes and/or numbers of replications. In contrast, some work has indicated that error in phenotype measurements from high throughput technologies may be partially under genetic control (Liang et al. 2018). Consistent with this hypothesis, we observed substantial non-random, genetic contributions to phenotyping errors for five tassel traits collected using an image-based phenotyping platform. Phenotyping accuracy relative to manually collected ground truth data varied according to whether a tassel exhibits “open” or “closed” branching architecture. A GWAS conducted for this open vs. closed tassel trait detected three distinct trait-associated SNPs (TASs); two of these are adjacent to known inflorescence genes. Surprisingly, TASs identified via GWASs conducted on five tassel traits that had been phenotyped both manually (i.e., ground truth) and via an image-based method exhibit little overlap. Further, it was possible to identify TASs obtained via GWAS on the differences between ground truth and automated measurements (i.e., phenotyping errors), indicating that this limited overlap is under genetic control. Similar results were obtained in an analysis of data collected from another image-based phenotyping pipeline. Therefore, this study suggests that phenotyping errors cannot always be controlled simply by increasing population size and/or replication number. Further, our ability to identify candidate genes associated with tassel architecture via GWAS on phenotyping errors that were not detected via GWAS of either automated or manually collected tassel trait data, highlights a complementary method of gene discovery.
Coauthors: Aaron Kusmec – Iowa State University;Seyed Vahid Mirnezami – Iowa State University;Lakshmi Attigala – Iowa State University;Srikant Srinivasan – Indian Institute of Technology Mandi;James Schnable – University of Nebraska-Lincoln;Maria G. Salas Fernandez – Iowa State University;Dan Nettleton – Iowa State University;Baskar Ganapathysubramanian – Iowa State University;Patrick Schnable – Iowa State University
