Alternative selection methods and explicit or implied economic-worth functions for different traits in tree breeding

Abstract

Tree breeders must almost always address multiple traits. That entails choosing selection methods and weighing up known or assumed economic-worth functions for traits. Classically, selection methods include independent culling levels, sequential culling for traits within generations, culling for different traits in successive generations (tandem selection), and selection indices, but elements of individual methods can be used in combination. Adverse genetic correlations among economic traits create a need to know the comparative economic worth of gains in different traits and can severely constrain feasible breeding goals. A traditionally cited optimal method is the Smith-Hazel selection index (and more generalised variants) based on good genetic-parameter information and known, linear economic-worth functions. The theoretical optimality, however, depends on various assumptions which are often tacit and unrecognised, and violated. Optimally cost-efficient selection will also depend on evaluation costs and ages of expression for different traits and may need to go well beyond selection indices. More complex criteria of optimality may also arise, making true optimisation very challenging and often unachievable. Where the genetic gain is delivered as mixes of segregants produced by seed propagation, non-linearities of economic-worth functions may not be readily exploited. However, non-linearities that include intermediate optima for trait values, or optima that are conditional on values for other traits (namely restricted domains in ‘multi-dimensional space’), might be exploited much better in clonal forestry systems. With adverse genetic correlations and differences among environments in the expression of genetic variation in individual traits, optimal solutions may entail selection for specific environments or production systems, on a finer scale for deployment than in breeding populations. Genomic selection offers accelerated genetic gains, and may mitigate effects of adverse genetic correlations, but depends strongly on high-quality phenotypic information on the expression of different traits at different ages in specific environments. This technology and its potential combination with gene editing promise enhancements of multi-trait selection. Remote-sensing technologies can now yield huge volumes of field data and much expanded candidate populations, but in some trade-off with effective heritabilities of remote-sensed phenotypes.