Predictive Accuracy of Models Based on Multiple Genomic Attributes in Pinus taeda L

Mapping genotypic variation to phenotypic variation is important for a variety of applications ranging from plant and animal breeding programs to human biomedical genetics. A number of different approaches have been used to model phenotypic variation as a function of allelic variation at a single locus, combinations of alleles at different loci, or the sum total of all variants across the genome. An alternative approach is to use additional sources of information, from different steps in the pathway of information flow from genotype to phenotype. Non-neutral DNA sequence variation can affect gene regulation, so steady-state mRNA transcript levels can be treated as variables affected by genotype and affecting phenotype. Some gene products in turn act to affect levels of metabolic intermediates, so metabolite measurements can similarly be used as variables affected by genotype and affecting phenotype.  We compared the predictive accuracy of statistical models based on genotypes at 3562 SNP loci, alone or in combination with measurements of levels of 199 transcripts or 382 metabolites, using data collected from over two hundred unrelated Pinus taeda individuals. Clonal replication was used to increase the accuracy of estimates of the genetic component of phenotypic variation, to reduce the effect of random environmental variation on the statistical models. Incorporating additional genomic attributes (transcript and metabolite levels) increased the predictive accuracy of models, suggesting that this approach may be useful in applied breeding programs.