From long-term progeny trials to genomic selection: empirical prediction and simulation-guided redesign of Scots pine breeding in Germany

Abstract

In Scots pine (Pinus sylvestris L.), long generation intervals limit genetic gain, but mature progeny trials may provide reference populations for genomic selection. We combined repeated phenotypes from two German trials established in 1990 with single-nucleotide polymorphism (SNP) genotypes, parentage auditing, duplicate filtering, and empirically calibrated forward simulations. Of 1,986 phenotype-matched genotyped trees, 1,651 remained after quality control. Across 14 trait-by-age combinations, genomic best linear unbiased prediction (GBLUP) gave modest but consistent improvements in predictive ability for diameter and volume from age 20 onwards. For volume at age 20 (V20), repeated leave-progeny-out predictive ability increased from 0.233 with pedigree-based additive best linear unbiased prediction (ABLUP) to 0.268 with GBLUP, and GBLUP exceeded ABLUP in all 20 matched partitions. Paired jackknife intervals did not establish systematic differences in narrow-sense heritability. ABLUP and GBLUP rankings were highly correlated, but only 75.9% of the top 10% V20 candidates overlapped. Simulations calibrated to empirical V20 variance components and family relationships compared phenotypic selection, ABLUP, progeny testing, direct genomic selection, and genomic selection followed by phenotypic thinning. After five cycles, progeny testing produced the greatest cumulative additive genetic gain (163.3%), whereas direct genomic selection produced the greatest annual gain (1.208% yr−1) because of shorter cycles. Faster genomic pathways accumulated more kinship, and a kinship-aware penalty reduced relatedness with pathway-specific effects on gain. These results support staged genomic implementation with continual reference-population updating, field validation, and explicit coancestry management.