Choosing probe genotypes for the analysis of genotype-environment interaction in winter wheat trials

Genotype-environment interaction was analyzed in French multi-environment wheat (Triticum aestivum L.) trials using probe genotypes and bi-additive factorial regression. Probe genotypes are specific genotypes in which the comparisons of yield components to reference values describe the most-important environmental factors that limited grain yield. The time-period until flowering was described by the deviation of kernel number from a threshold number while the grain-filling period was described by the reduction of thousand-kernel weight from a potential value. The aim of this paper was to determine the convenient number and the characteristics of probe genotypes to include in wheat breeding trials. Two sets of genotypes were used to model genotype-environment interaction: set 1 with 12 varieties tested in 18 environments and set 2 with ten lines tested in 14 environments. Set 2 was used for validation. Seven probe genotypes described the environments by providing environmental covariates, namely differences in yield components, for further analysis of interaction in set 1 and set 2. Interaction was modelled with bi-additive factorial regressions including differences in yield components. Several rounds of models were fitted to determine the optimal number of probe genotypes (i.e. environmental covariates) to introduce. From the seven probe genotypes, all the possible combinations including one to seven genotypes were studied. Significance of the combinations was tested with critical values obtained from simulations through 1,000 random permutations. Taking into account the information available on the probe genotypes, one would think that two, three or four probe genotypes would be sufficient, otherwise the number should reach four or five genotypes. In all cases, these numbers will provide models more-parsimonious than the classical AMMI model. The important information to be known on the probe genotypes prior their first multilocation experiment is: interaction pattern, earliness, and differences in yield component. Tested for the first time, a quadruplet is better than a triplet because the probability of choosing complementary genotypes increases with their number.