Genomic prediction of growth and wood quality traits in Eucalyptus benthamii using different genomic models and variable SNP genotyping density

Genomic selection (GS) is poised to revolutionize eucalypt tree improvement by shortening breeding cycles and increasing selection intensities. This could be particularly valuable for alternative, non-mainstream Eucalyptus species that are still in the initial stages of breeding. Eucalyptus benthamii is important for its adaptation to frost-prone subtropical regions. In this work, we compared seven genomic prediction models, six Bayesian and one frequentist GBLUP (Genomic Best Linear Unbiased Prediction) with the conventional pedigree-based ABLUP approach. Models were evaluated for their ability to estimate heritabilities and predict wood quality traits (wood density, extractives, lignin, and carbohydrates content) and volume growth in 77 open-pollinated families of Eucalyptus benthamii. We also evaluated predictive abilities and heritabilities using variable numbers of SNP in the models. Heritabilities ranged from 0.09 (extractives content) using Bayesian Lasso (BL) to 0.55 (wood density) using ABLUP. Predictive abilities (PA) ranged from 0.12 (for volume using ABLUP) to 0.44 (for wood density using three Bayesian models). All seven genomic models performed similarly well and better than the pedigree model for all traits, except extractives content. Subsets of 5000-7000 SNPs yielded heritabilities and PAs nearly as large as using all 15,293 SNPs. However, a low-density SNP panel might not be economically and technically advantageous compared to the current high-density multi-species Eucalyptus EUCHIP60k. Our results support a positive outlook to implement GS to accelerate Eucalyptus benthamii breeding for adaptation to frost-prone regions.

Automated summary

Genomic selection shows great potential for improving eucalypt tree breeding, especially for non-mainstream species. This study compared different genomic prediction models and found that they performed well in predicting wood quality traits and volume growth. The models outperformed the conventional pedigree-based approach, except for extractives content. Additionally, using a lower density SNP panel yielded similar predictive abilities and heritabilities compared to a higher density panel, but may not be economically advantageous.