Multi‐trait genomic selection for weevil resistance, growth, and wood quality in Norway spruce

Abstract

Plantation‐grown trees have to cope with an increasing pressure of pest and disease in the context of climate change, and breeding approaches using genomics may offer efficient and flexible tools to face this pressure. In the present study, we targeted genetic improvement of resistance of an introduced conifer species in Canada, Norway spruce (Picea abies (L.) Karst.), to the native white pine weevil (Pissodes strobi Peck). We developed single‐ and multi‐trait genomic selection (GS) models and selection indices considering the relationships between weevil resistance, intrinsic wood quality, and growth traits. Weevil resistance, acoustic velocity as a proxy for mechanical wood stiffness, and average wood density showed moderate‐to‐high heritability and low genotype‐by‐environment interactions. Weevil resistance was genetically positively correlated with tree height, height‐to‐diameter at breast height (DBH) ratio, and acoustic velocity. The accuracy of the different GS models tested (GBLUP, threshold GBLUP, Bayesian ridge regression, BayesCπ) was high and did not differ among each other. Multi‐trait models performed similarly as single‐trait models when all trees were phenotyped. However, when weevil attack data were not available for all trees, weevil resistance was more accurately predicted by integrating genetically correlated growth traits into multi‐trait GS models. A GS index that corresponded to the breeders’ priorities achieved near maximum gains for weevil resistance, acoustic velocity, and height growth, but a small decrease for DBH. The results of this study indicate that it is possible to breed for high‐quality, weevil‐resistant Norway spruce reforestation stock with high accuracy achieved from single‐trait or multi‐trait GS.

Conventional versus genomic selection for white spruce improvement: a comparison of costs and benefits of plantations on Quebec public lands

“…We have estimated that with this scenario, there would be a delay of 8 years before having access to improved seedlings for the reforestation program. Thus, this delay, which was included in the simulated rotation for this scenario, is quite similar to the one used by Li and Dungey (2018) in their simulations, that is 7 years. The last scenario of GSSE tested in the present study combines the use of GS in forward mode with multiclonal propagation (Park et al 2016), thus using all available sources of genetic variance for selection, leading to the greatest gains among the five scenarios tested.…”

“…We consider that it would take 34 years to complete the third generation of breeding and have access to the seed for the production of seedlings using conventional means, which was included in the rotation simulation. This generation interval is twice that used by Li and Dungey (2018) in their simulations. However, spruces are known to have longer breeding cycles than those of pine species such as Pinus taeda and P. radiata, with the latter being assumed in their simulations.…”

“…The authors found that the use of genomic-selected third-generation planting stock (18-year cycle with seed produced in seed orchards) was difficult to justify on an economical basis only, when compared with conventional breeding (33-year cycle with seed produced in seed orchards) under the current sawlog market value, but a sensitivity analysis revealed that the GS approach was more financially supported at lower discount rates and through an increase in log price premiums at harvesting. Although not relying on strict financial analysis per se, Li and Dungey (2018) reported results of a study on the expected benefits of genomic selection over conventional forward selection in conifer breeding and deployment, using stochastic simulations. They considered various methods to speed up the deployment of selected material, including top-grafting onto mature seed orchard ortets, somatic embryogenesis, and clonal propagation.…”

A stacking ensemble learning framework for genomic prediction

“…A large number of researches had tried to apply single machine learning methods in genomic prediction [11,14,37,38]. However, the single machine learning methods in the most previous studies only performed well on several traits [13,14,38,39].…”

Genotype-by-environment interaction in coast redwood outside natural distribution - search for environmental cues

“…In this study, diameter at breast height (DBH) was the only attribute representing tree growth. Studies across other forest tree species found strong/moderate correlations between DBH and tree height or volume [36][37][38][39] which provides evidence of using DBH as an ideal proxy phenotype for other essential attributes. However, there is decent genetic variability found in DBH -tree height relationship due to differences in individual's response to environment and competition [40].…”