Evolutionary computing to assemble standing genetic diversity and achieve long-term genetic gain

Loss of genetic diversity in elite crop breeding pools can severely limit long-term genetic gains, and limit ability to make gains in new traits, like heat tolerance, that are becoming important as the climate changes. Here we investigate and propose potential breeding program applications of optimal haplotype selection (OHS), a selection method which retains useful diversity in the population. OHS selects sets of candidates containing, between them, haplotype segments with very high segment breeding values for the target trait. We compared the performance of OHS, the similar method optimal population value (OPV), truncation selection on genomic estimated breeding values (GEBVs), and optimal cross selection (OCS) in stochastic simulations of recurrent selection on founder wheat genotypes. After 100 generations of inter-crossing and selection, OCS and truncation selection had exhausted the genetic diversity, while considerable diversity remained in the OHS population. Gain under OHS in these simulations ultimately exceeded that from truncation selection or OCS. OHS achieved faster gains when the population size was small, with many progeny per cross. A promising hybrid strategy, involving a single cycle of OHS selection in the first generation followed by recurrent truncation selection, substantially improved long term gain compared with truncation selection, and performed similarly to OCS. The results of this study provide initial insights into where OHS could be incorporated into breeding programs. Core Ideas Plain language summary Breeders use selection strategies based on genetic and phenotypic information to choose parents that will improve agriculturally-relevant traits (eg. grain yield) in their progeny. Generally, this involves estimating breeding values (scores) for each candidate parent. This study investigated an alternative ‘haplotype stacking’ approach called Optimal Haplotype Selection (OHS), which instead estimates breeding values for each unique genome segment in the population, then selects a group of parents who, between them, carry the haplotypes with the highest estimated breeding value at each chromosomal segment. In simulations, OHS gives improvements close to existing methods when populations are small, and outperforms them in the long term (100+ generations). Using just one generation of OHS boosts the performance of other methods in the short and long term. Breeders might consider adopting haplotype stacking in their programs, once techniques to do so are established. ### Competing Interest Statement The authors have declared no competing interest. * GEBV : genomic estimated breeding value GS : genomic selection OCS : optimal cross selection OHS : optimal haplotype selection OPV : optimal population value SNP : single nucleotide polymorphism