Efficiency of multi-trait genomic selection in two contexts of Eucalyptus breeding

In this study we explore the performance of multi-trait genomic selection (MT-GS) using the single-trait version (ST-GS) as the baseline. We implemented the approaches in two Eucalyptus breeding schemes. In Congo, where commercial plantations of E. urophylla*grandis are dedicated to pulp production, we considered height increment between 8 and 18 months (H8 18), critical trait to avoid weed competition, and volume at 55 months (V55), the target trait, to analyse the interest of MT-GS. We used 3303 SNP to define the relationship matrices and to estimate breeding values of 1130 cloned hybrid progenies. The prediction accuracy was estimated through a cross validation process (75% of tree in the training set). For H8 18, the accuracy was 0.354 and 0.370 for ST-GS and MT-GS respectively. For V55, the accuracy attained 0.414 and 0.424 for ST-GS and MT-GS respectively. Although MT-GS presented higher accuracies, estimates were not significantly di↵erent. This result was explained by the low heritability of V55 and H8 18 (h2=0.26 and h2=0.13) and the relatively high genetic correlation (⇢a =0.77). In Madagascar, where E. robusta small-scale plantations are grown for fuel wood, the volume at 49 months (V49), the total lignin (TL) and the holo-cellulose (Holo) are the targeted traits for breeding. 2919 SNP were used to define the relationship matrix and estimate breeding values of 415 individuals of a provenance trial. Accuracies were estimated through the same procedure. For V49, the prediction accuracy attained 0.30 for both ST-GS and MT-GS. For TL, the prediction accuracy was near zero whatever the approach (0.05 for ST-GS and 0.04 for MT-GS).