Direct comparison of recombinant inbred line (RIL) and parental line grain yield in a relatively low yielding Environmental Group (A) with that of a moderate yielding environmental group (F). Certain RILs exhibit a greater magnitude of genotype × environment interaction (G × E) effect (quadrants I and III) while others show similarities in relative performance between environmental groups (quadrants II and IV). 

Contexts in source publication

Context 1

... yield exhibited nearly a twofold range in magnitude among the 23 individual trials ( Table 1). The highest mean grain yield was 12.40 ± 1.01 Mg ha −1 in the 2006 Alma 112,000 plants ha −1 trial while the lowest grain yield was 6.22 ± 0.77 Mg ha −1 from the 2007 Elora 37,000 plants ha −1 trial. When averaged across all 23 trials, mean grain yield of the RIL hybrids was 9.23 ± 2.00 Mg ha −1 while the mean grain yields of the parental hybrids were 9.23 ± 1.96 Mg ha −1 , 10.27 ± 2.02 Mg ha −1 , and 9.38 ± 2.53 Mg ha −1 for the hybrids involving CG60, CG108, and CG60 × CG108, respectively. Combined ANOVA indicated that environments and G × E were signifi cant sources of variation; however, genotypes were not a signifi cant source of variation due to the rather large G × E eff ects. To address the G × E issue, an AMMI model was utilized as an aid to identify trials in which the pattern of G × E was similar (Fig. 1). Overall, the fi rst three principal components in the AMMI analysis explained 41.5% of the total G × E variation. The fi rst principal component (PC) captured 18.5% of total G × E variation, while the second and third PCs captured 12.8 and 10.2%, respectively. Using the AMMI analysis as a guide, ANOVA was used to determine if signifi cant diff erences for G × E were present within various combinations of trials (Table 2). When data from individual trials, which were clustered in the AMMI analysis, were combined and no signifi cant G × E was present, those particular trials were considered a distinct environmental group (EG). By using this approach the 23 trials were resolved into eight unique EGs (referred to as groups A through H). The numerous EGs detected in this study are not entirely surprising given the highly vari- able nature of southwestern Ontario (Loffl er et al., 2005). Grain yield BLUP estimates across EGs ranged between a grand minimum of 4.07 Mg ha −1 and a grand maximum of 11.52 Mg ha −1 (Table 3). (Table 3). Group B was the highest yielding EG at 10.33 Mg ha −1 although it only consisted of one trial. The highest yielding multitrial EG was H at 9.99 Mg ha −1 , and the lowest yielding EG was A at 6.50 Mg ha −1 . Trans- gressive segregants were observed in all EGs though no RIL exceeded the high parent (always CG108) in groups A, C, and H. Interestingly, the modifi ed single-cross parental line (CG60 × G108) showed the poorest grain yield performance in group A yet was close to the mean value in all other EGs. This pattern was also observed in the initial study from which CG60 and CG108 were chosen (Lee et al., 2006). Often, individual RILs performed well in certain groups and poorly in others (Fig. 2 and 3), thus depicting the nature of G × E interaction within this elite breeding cross. The magnitude of the Vg estimates varied across EGs (Table 3). The largest estimates of Vg were observed for groups A, D, and G, while the smallest estimates were observed in groups C, F, and H. In general, the smaller estimates were associated with the EGs that contained more individual yield trials (groups F and H) and may represent more frequently observed patterns of G × E. The magnitude of the Vg estimates suggests that no large eff ect QTL would be detected for most of the EGs, with the possible exception of groups A and G. In total, nine single-eff ect QTL and four epistatic interactions were detected across seven EGs (Table 4). Many of these QTL colocalize to regions identifi ed in other studies (e.g., Veldboom and Lee, 1994, 1996; Ajmone-Marsan et al., 1995; Ribaut et al., 1997; Melchinger et al., 1998; Austin and Lee, 1998; Austin et al., 2000). The proportion of total phenotypic variation individually accounted for by these QTL ranged from 4.6 to 18.3%. In general the single-eff ect QTL and epistatic interactions were specifi c to an EG. One of the single-eff ect QTL was detected in two of the EGs: umc1413 in groups D and G, with CG108 being the favorable allele in both groups. Only one mem- ber of an epistatic interaction was detected across EGs: umc1172 in groups C and F with CG60 being the favorable allele in both groups. The MLM for group A, one of the EGs with a larger Vg estimate, explained 52% of the phenotypic variation and contained three single-eff ect QTL and one epistatic interaction. In the two EGs with the smallest Vg estimates, F and H, only 19 and 18% of the phenotypic variation was explained, respectively, and this was almost exclusively due to epistatic interactions. These results are consistent with the rather extreme G × E eff ects that were observed and are consistent with the expectations drawn from the Vg estimates for each EG. If the interpretations presented in several recent studies are correct (Duvick, 1984; Frey, 1996; Calderini and Slafer, 1998) and continuation of the unprecedented increase in grain yields is not likely to be sustained at the same rate, what occurred during the past 60 yr that led to this diffi culty? This experiment was designed to examine one of the approaches utilized in inbred line development in the commercial maize breeding sector—breeding within a family within a heterotic pattern. While these observations were made on a single elite breeding cross, the observations are likely applicable across most inbred line families that have been under within heterotic pattern, within family pedigree selection for several generations (i.e., within a closed breeding program). They may extend even more broadly to within heterotic patterns in general, again within a closed breeding program. What we observed in this elite breeding cross identifi ed two major obstacles that plant breeders must contend with for continued genetic progress in the elite breeding pool. Obstacle number one is, without a doubt, the sometimes rather extreme G × E. Will all elite breeding crosses exhibit G × E to the same degree? Perhaps, but the eff ect on selection could be minimized by testing in multiple years. Only early generation selections, which are routinely based on 1-yr data, may prove problematic. Within our data set there appear to be two main EGs, F and H, suggesting that some G × E patterns are perhaps more common than others. The lower Vg estimates observed in these two groups further support this hypothesis. Group F, with seven individual trials, encompasses two planting densities, 2 yr, and all three locations. The fi ve individual trials in Group H include all three densities and locations, but only 1 yr. Were there individuals within the RILs that would have been consistently higher performing regardless of the G × E? When the CVs for each of the 128 RILs were examined, there did not appear to be any RIL with a low CV and a high average grain yield (data not shown). Given these observations, testing across multiple years even in early generations of selection may be necessary. The second obstacle is what appears to be a rather severe Bulmer Eff ect resulting in all of the large single- eff ect QTL being fi xed. This is not surprising consider- ing that this is the goal of breeding—fi xing those alleles responsible for the largest genetic contributions to the trait of interest. The fi rst QTL fi xed in any breeding cross are likely those that have a sizable eff ect on the trait regardless of G × E. Compounding the fi xation of large single eff ect favorable alleles is that extensive linkage disequilibrium (LD) within a family is also occurring, the other aspect to a Bulmer Eff ect. The elite inbred lines used in this study were part of a larger study documenting the extent and nature of identity by descent (IBD) in three families of inbred lines derived from ...

Context 2

... the modifi ed single-cross parental line (CG60 × G108) showed the poorest grain yield performance in group A yet was close to the mean value in all other EGs. This pattern was also observed in the initial study from which CG60 and CG108 were chosen ( Lee et al., 2006). Often, individual RILs performed well in certain groups and poorly in others (Fig. 2 and 3), thus depicting the nature of G × E interaction within this elite breeding ...