Analysis of Grain Yield of Early-Maturing Yellow-Endosperm Maize Hybrids under Nitrogen Stress and Optimal Conditions Using AMMI Biplot

MATERIALS AND METHODS

Germplasm and experimental site: Twelve yellow-endosperm maize inbred lines were crossed using the diallel mating design without the reciprocals²³ to generate 66 single-cross hybrids. The hybrids were evaluated along with four check hybrids in varied soil N conditions in 2014 and 2015 to determine the response of the hybrids to changing N condition with respect to variability in grain yield of the hybrids. The trial was conducted in Ibadan (3.56°E, 7.33°N and 168 m asl), Nigeria. The experimental soil was purposely depleted of its native N from continuously planting maize at a very high population density without application of N fertilizer and uprooting and removing the biomass completely after each cropping. This depletion procedure was repeated until the soil N had been completely removed. Soil analysis was carried out to confirm the N status after each depletion process. The soil was depleted to about zero level of N. Mean annual rainfall and mean temperature of the experimental site during the experiment were 141.3 mm and 25.8°C, respectively, for 2014 and 103.3 mm and 26.6°C, respectively, for 2015 (Appendix 1).

Experimental layout and crop management: The 66 hybrids along with four hybrid checks were planted in a 10×7 lattice design with three replicates in each year. Plots consisted of two 5 m long rows, with 0.75 m inter-row spacing and 0.5 m spacing between plants within a row. Three seeds were sown and thinning was done two Weeks After Planting (WAP) to two plants per hill to obtain a plant population density of 53,333 plants ha^–1.

Appendix 1:	Monthly mean amount of rainfall and mean temperature at the experimental sites during the trials from 2013 to 2015
Temp: Mean temperature

There were three N conditions: 0, 30 and 90 kg N ha^–1 denoting zero N, low N and optimal N, respectively. The fertilizer was applied in the form of NPK 15:15:15 at 30 kg ha^–1 to each of low N and optimal N plots at 2 WAP. The optimal N plots received 60 kg N ha^–1 in the form of urea to bring the total available N to 90 kg ha^–1 two weeks later (4 WAP). Urea was not applied to zero N plots but all the plots received 60 kg P ha^–1 as single super phosphate (P₂O₅) and 60 kg K ha^–1 as muriate of potash (K₂O). Three N conditions applied in each of the two years constituted the six environments for the trial. Standard cultural practices were adopted for field maintenance, harvesting and seed processing.

Data collection and analysis: Ears were harvested when dry. The grains were shelled and weighed, after which moisture content was determined using a digital moisture tester. Grain yield in kg ha^–1 was computed using grain moisture content = 15%, harvested plot area = 7.5 m² and 1 ha = 10,000 m² as follows²⁴:

where, GWT is the grain weight and MC is the grain moisture content at harvest.

The grain yield data were subjected to analysis of variance (ANOVA) using SAS separately for each N condition and across N conditions, over the two years. Hybrids and N conditions were considered fixed effects, whereas replicates and years were considered random. The AMMI model was used to analyze the grain yield of the hybrids across N conditions andto investigate G×E effects. Statistical model, according to Yang et al.²⁵, used for the AMMI analysis was as follows:

where, Υ_ij is the mean of the ith genotype in the jth environment, μ is the overall mean, τ_i is the effect of the ith genotype, δ_j is the effect of the jth environment.

λ_k (λ₁>λ₂>…λ_t): scaling constants (singular values) that allow the imposition of ortho-normality constraints on the singular vectors for genotypes [α_k = (α_1k,……., α_gk)] and for environments [γ_k = (γ_1k, ……, γ_ek)], such that Σ_iα^2ik =Σ_jγ² jk = 1 and Σ_iα_ikα_ik' = Σ_j γ_jk γ_jk' =0 for k≠k .

The α_ik and γ_jk for k = 1, 2, 3.... were designated as primary, secondary, tertiary ... effects of genotypes and environments, respectively, ε_ij is the residual error.

For any genotype-environment combination, the main effect equals the genotype mean plus the environment mean minus the grand mean. The interaction is the product of genotype PCA and environment scores. When a genotype and environment have the same sign on the PCA axis, their interaction is positive, if different, their interaction is negative. On the biplot, genotypes and environments with large PCA1 scores either positive or negative have high interactions while those with PCA1 score of zero or nearly zero have small¹⁰.

RESULTS

Grain yield of the hybrid maize in varied N conditions: Analysis of variance for each year showed that significant variation in genotypes existed under the three N conditions (Table 1). The significant difference in genotypes under no N condition in 2014 was p<0.05 and p<0.001 in 2015. The results further showed that significant variation was very high (p<0.001) for the trait under low N and optimal N conditions.

Table 1:	Mean squares from the analysis of variance of grain yield for the yellow endosperm maize
ns: Non-significant, *p<0.05, ***p<0.001, df: Degree of freedom, CV: Coefficient of variation

Table 2:	Mean squares from the combined analysis of variance of grain yield for the yellow endosperm maize
ns: Non-significant, *p<0.05, **p<0.01, ***p<0.001, df: Degree of freedom, CV: Coefficient of variation

Coefficient of variation (CV) for individual N condition ranged from 14.34% for optimal N in 2015 to 35.40% for no N condition in 2014. In the ANOVA for individual N condition for each year, highest mean square error was found for no N in 2014 while the least was observed for 90 kg N ha^–1 in 2015.

Combined ANOVA, however, showed highly significant differences (p<0.001) for years (Y), genotypes (G), N condition (N), genotypes×year (G×Y), genotypes×nitrogen condition (G×N) and year×nitrogen condition (Y×N) (Table 2). The CV was 21.07% for combined ANOVA. Mean grain yield across locations was about 2636 t ha^–1.

Mean grain yield of the hybrids was lower in 2014 than in 2015 under the three N conditions (Table 3). Mean grain yield was about 1495 kg ha^–1 under No N, about 2341 kg ha^–1 under low N and about 4073 kg ha^–1 under optimal N condition. Under no N conditions, grain yield ranged from 431 kg ha^–1 for TZEI10×TZEI124 to 1537 kg ha^–1 for TZEI146×TZEI124 in 2014 and from 1017 kg ha^–1 for TZEI12×TZEI16 to 3183 kg ha^–1 for TZEI146×TZEI128 in 2015 (Table 3). Under low N condition, grain yield ranged from 803 kg ha^–1 for TZEI10×TZEI124 to 2451 kg ha^–1 for TZEI12×BD74222 in 2014 and from 859 kg ha^–1 for TZEI12×TZEI16 to 4382 kg ha^–1 for BD74165×BD74161 in 2015. However, under optimal condition, grain yield ranged from 2505 kg ha^–1 for TZEI124×BD74161 to 4530 kg ha^–1 for TZEI13×TZEI128 in 2014 and from 3288 kg ha^–1 for TZEI13×TZEI10 to 6453 kg ha^–1 for TZEI16×BD74165 in 2015. The TZEI146×TZEI12, TZEI12×BD74222, TZEI146×TZEI128, TZEI146×TZEI16, TZEI146×TZEI124, TZEI12×BD74161, TZEI10×BD74165, TZEI16×BD74165 and BD74165×BD74161 were among the highest yielding genotypes across N environments, each of them yielded about or greater than 3000 t ha^–1.

AMMI analysis of grain yield of the maize hybrids: First PCA scores for environment ranged from -36.34 for optimum N condition in 2015 to 23.31 for zero N in 2014 while from -37.34 for TZEI16×BD64165 to 31.33 TZEI13×TZEI10 (Table 3). The PCA1 of the N conditions in 2015 were negative while those of 2014 were positive. The ANOVA for grain yield of the maize hybrids across environments is shown in Table 4. About 67.2% of the total sum of squares was attributable to environments, whereas 12.1% was attributed to genotypes and 11.4% to G×E interaction. Variances attributable to environments, genotypes and G×E were highly significant (p<0.001). Partitioning of the G×E into the two interaction principal component axes (IPCA 1 and 2) revealed that IPCA 1 had highly significant (p<0.001) effect, whereas the effect of IPCA 2 was significant at p<0.01. The IPCA 1 and IPCA 2 jointly accounted for 9.6% of the total sum of squares, while the residual accounted for only 1.8% of the total sum of squares.

Table 5 shows main effects, interaction and final yield of top three and least two grain yielding maize hybrids for each N condition in the two years. The result showed strong evidence that G×E interactions were significant in both years. Hybrid TZEI12×BD74222 exhibited least absolute interaction (26.0) while TZEI146×TZEI128 exhibited highest absolute interaction (674.1). Least interactions were obtained for no N in 2015 than others while highest were in optimal condition in 2015. Under no N condition, the effects of genotype on the grain yield of the hybrids were higher in 2015 than 2014 but the absolute interaction effects were lower in 2015.

Table 3:	Mean grain yield of best 19 and poorest 10 yellow endosperm maize hybrids with four checks
N: Nitrogen, PCA: Principal component axis, LSD: Least significant difference

Table 4:	Additive main effect and multiplicative interactions analysis of variance for grain yield of the yellow endosperm maize hybrids
**p<0.01, ***p<0.001, df: degree of freedom, †: Values in parentheses are percentage of G×E

Hence, the final yields were higher in 2015. Though the trend was similar under the two remaining N conditions but the interaction effects were variable.

Figure 1 represents the AMMI biplot showing the association among the best 19 and 10 poorest yielding hybrids under the varied N environments. In the AMMI biplot, about 83.2 and 8.4% of the total variation were explained by environments and genotypes, respectively. The AMMI biplot is divided into four quadrants, lower yielding environments in the two left quadrants and the higher yielding environments in the two right quadrants.

Table 5:	Grain yield of the top three and least two yellow endosperm maize estimated by additive main effect and multiplicative interactions model
Only TZEI124×BD74165 (61) was located in the low-yielding quadrant, whereas TZEI13×TZEI146 (1), TZEI13×TZEI128 (4), TZEI146×TZEI124 (18), TZEI14×TZEI25 (67) and TZEI128×BD74161 (44) were placed in the high-yielding quadrant. Hybrids TZEI128×BD74161 (44), TZEI14×TZEI25 (67), TZEI124×BD74222 (63) were placed close to the origin of the plot (low PC1) with high main effects. Although TZEI12×TZEI11 (32), TZEI11×TZEI8 (46), TZEI124×BD74161 (62) and TZEI124×BD74165 (61) had low PC1, they had low yield. TZEI13×TZEI146 (1), TZEI10×BD74165 (28) and TZEI16×BD74165 (58) were the farthest from the origin (absolute high PC1) with high mean effects. TZEI13×TZEI10 (2) and TZEI11×TZEI16 (47) among others had relatively high PC1.

DISCUSSION

Varying fertilizer levels create micro-environments and hybrids respond differently to the environments because of the differences in the genotypic constitution of the hybrids. The significant effects for year, nitrogen condition and their interactions demonstrated the divergent conditions that resulted in differences for environmental means causing variation in the hybrids. The various interactions of genotypes with environments introduce confusion in selection of promising genotypes. Statistical analyses are routinely carried out to separate the G×E interaction effects that result from the performance of different hybrids in the varying environments and to study the hybrid response and adaptability to be able to make precise decisions.

Fig. 1:	AMMI biplot of the grain yield of best 19 and poorest 10 grain yielding early maturing yellow-endosperm maize hybrids with four checks

Numerous scientists^13,14,18 have also reported confusing influence of environments on the expression of maize traits during breeding programmes by significant effects of G×E.

The CVs for individual N condition ranged from low to high indicating substantial amount of variability in the response of the hybrids to various N conditions. They were lower for optimal N, followed by low N and more variable for no N condition. The high CVs for low N and no N conditions may be attributed to the poor fertility status of the soil, which might have caused fluctuations in maize yield across replications. The relatively low values under other N conditions and for combined ANOVA, however, indicate uniformity of the management practices of the experiment. Significant variations among genotypes resulting from wide range of CVs pointed the differences in the ability of the genotypes to respond to changes in environment. The variability created can be exploited for maize improvement for grain yield under varying environmental conditions. Observations have also been made in this line^1,2,23.

Mean grain yield of the hybrids was lower in 2014 than in 2015 under the three N conditions. Variation in the mean yield is expected because crop genotypes perform differently in different growing condition because of variation in their response to changes in the effects of the growing condition. Amount of rainfall was higher in 2014 than in 2015 but mean rainfall was higher during the early and vegetative growth stage of the crop in 2015. Besides, rain was comparatively more consistent in 2015 than 2014. This could contribute to the differences in the yield between the two years. The significant variation among hybrids under all the N conditions suggests sufficient genetic variability that could be exploited by identifying promising hybrids in the various N conditions. Moreover, the significant effects of the genotypes, N condition, year, G×N, G×Y and Y×N interaction in the combined analysis shows that the parameters are important in the breeding process. The significant interaction effects suggested that each factor cannot independently explain all the variation observed. This resulted in different performances of the genotypes in the evaluated environments. Hence, further analysis of the data was essential to obtain information on the response of the hybrids to changes in N conditions. Several reports have suggested more detailed studies of G×E interaction to recommend superior crop genotypes^{7,11,15,19,26,27}.

In this study, the grain yield of the hybrids varied considerably among the environments. Hybrids TZEI12×BD74161 (37), BD74165×BD74161 (64) and TZEI16×BD74165 (58) had grain yield greater than 3000 kg ha^–1 under low N, while TZEI13×TZEI11 (5) that yielded least under optimal N had higher grain yield than the best hybrid (TZEI128×BD74161) under no N conditions. This result confirmed the importance of soil N in determining grain yield of maize^1,2,23. The AMMI yield estimates are adjusted hence, the final yields were different from the treatment means. More precise yield estimates have been reported to increase the probability of success selection^19,28. The AMMI has been used to identify maize hybrids suitable for varying environments. The significant mean squares of the first two Interaction Principal Component Axis (IPCA) further suggest that the maize hybrids did not yield consistently under the three N conditions for the two years and that some hybrids were responded variously the different N conditions. This was also demonstrated by the 84% of the 11.4% sum of squares of G×E captured by AMMI as jointly explained by IPCA1 and IPCA2. Sum of square for the IPCA1 were about 4.5 times larger than that of the residual. This also revealed differences in grain yield among the maize hybrids across N conditions in the two years because of the presence of significant G×N as well as Y×N. This finding supported imperative that specific hybrids be identified for specific N condition.

The AMMI biplot for the maize in varied N conditions explained a total of 91.6% of the total sum of squares of the total variation. Considering the fact that the PC1 score of a particular hybrid in the AMMI analysis is an indication of the responsiveness of the hybrid to environment. Hybrids TZEI128×BD74161 (44) and TZEI124×BD74222 (63) were high yielding and their yield fluctuated less than the remaining hybrids across N conditions. So Thus, their yields were relatively high in all the N conditions. Hybrids TZEI12×TZEI11 (32), TZEI11×TZEI8 (46), TZEI124×BD74-161 (62) and TZEI124×BD74165 (61) were low yielding and did not respond actively to changes in N conditions because they were near the origin of the AMMI biplot (low PC1). The hybrids can be fitted well by an additive model.

However, hybrids TZEI13×TZEI10 (2) and TZEI11×TZEI16 (47) were low yielding because they had low main effect values on the AMMI biplot but they largely responded to zero N and low N conditions in 2014 (0N_14 and 30N_14) by having high IPC1 values. When a genotype and an environment have the same sign on the PCA axis, their interaction is positive, if different, their interaction is negative^12,16. Therefore, hybrids TZEI13×TZEI10 (2), TZEI11×TZEI16 (47) and TZEI124×BD74165 (61) had positive interaction with N deficient (no N and low N) condition, while TZEI13×TZEI146 (1), TZEI13×TZEI128 (4), TZEI146×TZEI124 (18), TZEI10×BD74165 (28) and TZEI16×BD74165 (58) had positive interaction to optimal N condition. Hence, the yield of the hybrids were more stable under suitable cultivation conditions of environment created by climate and soil N conditions.

The effects of genotype on the grain yield of the hybrids were higher in 2015 than 2014 but the interaction effects were variable. Variation in the amount of rainfall cannot be overemphasized in this study because the interaction of rainfall might have affected utilization of the available N in the two years, thus, the hybrids were grouped by years. This can be explained by the amount of rainfall that was markedly different in the two years. Though the year 2014 had higher amount of rainfall than year 2015, the rainfall distribution was more even in 2015 especially during the growing period of the maize. In situations like this, several N conditions in a more diverse weather conditions may be required to separate the hybrids based on their tolerance to the stress. This kind of environment can be created with same N levels in a location for more than two years or same N levels in more than one location in a year. It has been reported that weather conditions affect performance and selection of cereals especially maize^27-31. Similarly, Oluwaranti et al.³⁰ also reported significant influence of weather on production of maize in the tropics.

The AMMI biplot for the maize in varied N conditions explained a total of 91.6% of the total variation. High yielding hybrids TZEI128×BD74161 (44) and TZEI124×BD74222 (38) as well as low yielding TZEI12×TZEI11 (32), TZEI11×TZEI8 (46), TZEI124×BD74161 (62) and TZEI124×BD74165 (61) responded less to change in N conditions. Hybrids TZEI13×TZEI10 (2), TZEI11×TZEI16 (47) and TZEI124×BD74165 (61) had positive interaction with N deficient (no N and low N) condition, while TZEI13×TZEI146 (1), TZEI13×TZEI128 (4), TZEI146×TZEI124 (18), TZEI10×BD74165 (28) and TZEI16×BD74165 (58) had positive response to optimal N condition.

CONCLUSION

High yielding attributes of hybrids TZEI128×BD74-161 and TZEI124×BD74-222 were stable across the N conditions. Hybrids TZEI13×TZEI10, TZEI11×TZEI16 and TZEI124×BD74-165 had positive interaction with N stress (zero N and low N) condition, while TZEI13×TZEI146, TZEI13×TZEI128, TZEI146×TZEI124, TZEI10×BD74165 and TZEI16×BD74165 had positive response to optimal condition. Thus, the hybrids are suggested for the various N conditions.

SIGNIFICANCE STATEMENTS

This study identified high yielding attributes of hybrids TZEI128×BD74161 and TZEI124×BD74222 as stable across the three N conditions. They can therefore be recommended for cultivation in both deficient and optimal N conditions. Hybrids TZEI13×TZEI10, TZEI11×TZEI16 and TZEI124×BD74165 interacted with N stress condition, while hybrids TZEI13×TZEI146, TZEI13×TZEI128, TZEI146×TZEI124, TZEI10×BD74165 and TZEI16×BD74165 responded to optimal condition. Hence, the sets of hybrid are suitable for cultivation under N stress and optimal conditions, respectively. The hybrids identified for the various N conditions will help seed companies to make informed choices on the varieties to produce for farmers’ use. Thus, the constraints to maize production due to soil fertility can be reduced.

ACKNOWLEDGMENT

The authors thank personnel involved in the Maize Improvement Programme of Institute of Agricultural Research and Training, Ibadan, Nigeria for their help with field work. Many thanks to IITA and CIMMYT for providing seeds for the study. This research was supported by the Institute of Agricultural Research and Training, Obafemi Awolowo University, Ibadan, Nigeria.