Modelling Plant Breeding Programs as Search Strategies on a Complex Response Surface

Abstract

The concept of an adaptation (fitness) landscape has been used to explain evolutionary processes. The landscape is a response surface for the genetic space defined by a genotype-environment system and evolution of populations through natural selection a search for higher peaks in this space. This is an appealing framework for other disciplines interested in issues of search and optimisation. One such application is the genetic improvement of traits in plant breeding. Here, breeding programs can be viewed and analysed as search strategies that are used to explore the surface of an adaptation landscape to find higher adaptive positions. The current theoretical framework considers genetic improvement as a hill climbing process on a smooth single peaked adaptation landscape. However, there is strong evidence to suggest that due to the effects of genotype-by-environment (G × E) interactions and epistasis, the landscapes encountered by plant breeders are in fact rugged and multi-peaked. Simulation methodology was used to compare two selection strategies currently used in plant breeding and investigate their capacity to confront the difficulties associated with the influences of G × E interaction and epistasis: (i) selection of genotypes based on performance in a single environment (mass selection), and (ii) selection of genotypes based on performance in several environments (multi-environment testing). A third selection strategy was proposed for genetic improvement on more complex adaptation landscapes. This selection strategy (shifting search strategy) was based on Wright’s ‘Shifting Balance Theory’.