Open Collections

Patterns of genotype-environment interactions and sensitivity to genomic selection in the lodgepole pine breeding program in British Columbia

University of British Columbia

Lodgepole pine (Pinus contorta Dougl. ex. Loud.) is an important tree species in British Columbia, Canada, both commercially and ecologically. Much of the seed used for reforestation after harvesting and wildfires is produced by seed orchards that rely on an extensive breeding program. Tree breeders require tools to adjust quickly to unforeseen challenges exacerbated by a changing climate and two approaches were examined: mulit-environment trial (MET) analyses and genomic selection. A MET dataset was used to analyse height growth data across five breeding zones using a factor analytic model for estimating additive (co)variance among test sites and the use of molecular markers to predict breeding values for height growth and wood quality traits (average wood density, earlywood density, latewood density, latewood proportion and microfibril angle) was assessed for a subsample of 1,569 trees from four progeny test sites using 19,584 single-nucleotide polymorphism markers. Test sites were clustered into four main breeding zones based on genetic correlations between sites. Climate change projections were applied to the four new zones which suggested that southern zones will expand. Predicted breeding values from Bayesian models (Bayes B and C) and best linear unbiased prediction (BLUP) using a hybrid matrix (H-matrix) were compared to models that used an average relationship matrix (ABLUP) and a realized relationship matrix (GBLUP). Bayesian models had similar prediction accuracies compared to ABLUP and GBLUP when models were confined to a closed population (> 0.74), but accuracy decreased substantially when relatedness was controlled between the training and validation populations (> 0.25). The models also worked well across environments and test cycles and were similar to GBLUP for ranking trees within families. HBLUP was equivalent to ABLUP and GBLUP models for genetic parameters and prediction accuracy, but had very low within-family rank correlations (0.08 average Spearman rank correlation across traits). Bayesian models can be used to predict breeding values and rank trees that have no phenotypic data. In contrast, HBLUP should not be used for ranking trees within families in the absence of phenotypic data, but can effectively improve estimates of variance components and breeding value estimates.

Text

2020-04-20

Attribution-NonCommercial-NoDerivatives 4.0 International

http://hdl.handle.net/2429/74101

Forestry

University of British Columbia

UBCV

http://creativecommons.org/licenses/by-nc-nd/4.0/