An evaluation of eight cycles of replicated recurrent selection in maize

Responses to S1 per se (S1) and reciprocal full-sib (RFS) selection in maize were compared after eight cycles by evaluating a replicated recurrent selection study involving the Nebraska B (NB) and Nebraska Stiff Stalk (NS) synthetics. Selection was based on a multiplicative index which included adjusted yield, upright plants, and retained ears. Replication of breeding methods provided estimates of selection response in light of “replicate effects” which included genetic drift caused by small effective population size (Ne = 20). The random-mated populations, the selfed (F = 0.5) populations, and the variety crosses of cycles 0, 5, and 8 were evaluated in 1994 and 1995 in environments similar to those in which the populations had been selected. Linear responses in the index of 5.75% and 6.46% cycle-1 were observed for the S1 and RFS variety crosses, respectively. Although both methods have been equally effective for improving the index of the variety cross, S1 selection was preferred due to greater relative improvements in the selfed populations than those from RFS selection. RFS selection provided greater increases in heterosis for grain yield, whereas S1 selection resulted in greater reductions in inbreeding depression. Detection of selection method effects was hampered by the presence of considerable variation for response to selection among replicates. Inconclusive results, reminiscent of previous comparative studies, were observed as responses in the index of the variety cross were similar for S1 and RFS selection in two of the replicates, but favored RFS selection in the third replicate. Also, responses in the index of the best replicates were 30% and 60% greater than responses of the poorest replicates for S1 and RFS selection, respectively. Our results suggest choice of optimum breeding methods made based upon non-replicated selection are unreliable due to “replicate effects” which include random drift effects and selection environment effects. Our results and those of previous evaluations also suggest that sampling error caused divergence of the replicates in the initial cycle of selection that significantly affected the subsequent response curve. Based on this evidence, we recommend replicated selection as a method which could provide more expeditious gains if selection among replicates was practiced as soon as superior replicates could be identified