Incorporation of Molecular Marker and Spatial Data into Analysis of Clonally-Replicated Progeny Tests

Different analytical procedures can be incorporated to the current forest tree breeding with the goal to maximize genetic gain and allow for improvement of the selection. Three approaches were developed to explore alternative analytical tools in clonal data of loblolly pine (Pinus taeda L.), replicated across different test environments. First, spatial analysis was applied to the field tests which were affected by spatial autocorrelation among individuals from the same clone or between genetic entries replicated within sites. A linear mixed model was examined to compare various variance-covariance structures for prediction of genetic merit and variance components of a total of 453 cloned replicated progeny planted across 16 sites. The use of spatial autoregressive plus independent residual combined with factor analysis structure for heterogeneous variance-covariances was superior in model fit statistics and genetic parameters estimation, compared with the default independent and identically distributed random effects. The factor analytic structure, which separate genetic effects into common and specific components, represented an useful framework than standard assumption, with efficient modeling of GxE interaction, variance component estimation and allowing the estimation of high broad-sense heritability.;Second, the incorporation of molecular markers information into forest breeding was done by analyses of SNP genotypes at over 3,400 polymorphic loci in a set of progeny from a structured mating design with 13 families, using lignin content, cellulose content, height and volume traits with differing heritabilities. Cross-validation strategies using a subset of the individuals to train a genomic selection model using all markers followed by prediction of genetic value for the remaining subset of individuals were used to explore the utility of genomic selection in a small breeding population of loblolly pine. Mean predictive values, measured as the correlation across all families between genome-estimated genetic value and measured genetic value from clonally-replicated field tests, ranged from 0.30 to 0.83 across the four traits and four different cross-validation scenarios evaluated. Estimation of genetic value based on pedigree information alone yielded similar accuracies across all families. Prediction accuracies of models that use only a subset of markers associated with phenotypes were generally comparable with the accuracies of the model using all markers, but they do not need to be strictly associated with the phenotype. For lignin and cellulose, the genomic selection scenario was efficient under different relative lengths of the breeding cycle, which would allow cost-effective applications in tree breeding programs.;Third, using molecular marker information, we compared genomic estimated breeding values based markers (G matrix) with estimates of the average additive genetic covariances (A matrix) from pedigree. We analyzed volume of 165 clones obtained from nine families. The estimates from realized genomic relationships used 3,461 biallelic SNP markers and two methods to generate a G matrix containing those genomic co(variances). The accuracy of the predictions for both G and A based models was compared for cross-validation methods of sampling clones either within family or at random. Using the genomic covariances resulted in smaller standard errors for the genomic estimated breeding values. The observed accuracies of the predictions with the genomic matrix based on allele frequencies or regression were similar and higher than accuracies from the only-pedigree base model. Realized genomic relationships could predict Mendelian segregation among full-sibs individuals, which was not the case for average relationships. Selection based on GBLUP would increase genetic gain in applied forestry breeding, due to important saving in breeding cycle length and cost per unit time.