Genomic prediction for hastening and improving efficiency of forward selection in conifer polycross mating designs: an example from white spruce

Abstract

Genomic selection (GS) has a large potential for improving the prediction accuracy of breeding values and significantly reducing the length of breeding cycles. In this context, the choice of mating designs becomes critical to improve the efficiency of breeding operations and to obtain the largest genetic gains per time unit. Polycross mating designs have been traditionally used in tree and plant breeding to perform backward selection of the female parents. The possibility to use genetic markers for paternity identification and for building genomic prediction models should allow for a broader use of polycross tests in forward selection schemes. We compared the accuracies of genomic predictions of offspring's breeding values from a polycross and a full-sib (partial diallel) mating design with similar genetic background in white spruce (Picea glauca). Trees were phenotyped for growth and wood quality traits, and genotyped for 4092 SNPs representing as many gene loci distributed across the 12 spruce chromosomes. For the polycross progeny test, heritability estimates were smaller, but more precise using the genomic BLUP (GBLUP) model as compared with pedigree-based models accounting for the maternal pedigree or for the reconstructed full pedigree. Cross-validations showed that GBLUP predictions were 22-52% more accurate than predictions based on the maternal pedigree, and 5-7% more accurate than predictions using the reconstructed full pedigree. The accuracies of GBLUP predictions were high and in the same range for most traits between the polycross (0.61-0.70) and full-sib progeny tests (0.61-0.74). However, higher genetic gains per time unit were expected from the polycross mating design given the shorter time needed to conduct crosses. Considering the operational advantages of the polycross design in terms of easier handling of crosses and lower associated costs for test establishment, we believe that this mating scheme offers great opportunities for the development and operational application of forward GS.

Evaluation of the efficiency of genomic versus pedigree predictions for growth and wood quality traits in Scots pine

“…The execution of GS in animal and crop breeding programs, such as dairy cattle, oat, maize and wheat, increased genetic gains [44,73]. Implementation of GS in tree breeding is underway with recent publications in eucalypts [61,[74][75][76][77], white spruce [78][79][80], black spruce (Picea mariana [Mill.] BSP) [60], interior spruce [39,70], Norway spruce [68,81,82], loblolly pine [58,83,84], lodgepole pine (Pinus contorta Douglas) [85] and maritime pine [66,67].…”

“…Predictive ability ( r 1 ) has been widely used but normally shows lower values than r 2 , r 3 and r 4 for different traits ( Table S1 ) as it uses the individual phenotypes as approximation to TBV and could be comparable to heritability [ 62 , 68 , 78 , 87 , 88 ]. Predictive accuracy ( r 2 ) and theoretical accuracy ( r 3 ) have been used less frequently [ 40 , 74 , 78 ]; the former one is considered as an unbiased estimate of accuracy of selection from n-fold cross-validation, since the correlation between an individual phenotype and its TBV cannot be higher than the square root of heritability [ 52 , 89 ]. The r 3 [ 63 ] was usually used to evaluate the models with full datasets, however in the current study it was used in the cross-validation analyses as well, since different PEV were obtained for each fold, thus estimations can be evaluated in the same way as the remaining methods.…”

“…Higher prediction efficiencies were obtained for traits with moderate heritabilities (above 0.30) which is in line with the positive correlation between trait heritabilities and r 1 reported in loblolly pine [ 58 ] and maritime pine [ 66 ]. Chen et al [ 68 ] concluded that values of narrow-sense heritability were more similar to values of r 1 than to r 4 , as r 1 involves both phenotypic and genetic values, however using r 2 instead of r 1 could remove influence of heritability since it is considered an unbiased estimation of the accuracy of selection from n-fold cross-validation [ 52 , 78 , 89 ].…”

The Genetic Basis of Tomato Aroma

“…During tomato fruit ripening, free fatty acids, mainly linolenic and linoleic acid, are derived from the catabolism of acylglycerides from disintegrating cellular membranes, by the action of lipases [87,88]. Linolenic and linoleic acid can be further catabolized by means of β-oxidation, α-oxidation, or the lipoxygenase pathway [68,[105][106][107].…”

“…Genomic prediction is particularly useful for traits for which phenotyping is expensive, difficult or time consuming, since no phenotyping is needed for selection, once a good prediction model based on data of a representative training population is available. This technique has been widely applied in animal selection [83][84][85], while its practical application in plant breeding is still limited to major crops, such as maize and wheat [82,86,87], in which QTLs for important traits, such as yield, have already been fixed in the elite germplasm [88], or to tree crops, where early selection is very useful and cost-effective [89].…”