A comprehensive comparison between Eberhart and Russell joint regression and GGE biplot analyses to identify stable and high yielding maize hybrids
Introduction
Identification of high yielding and stable genotypes across variable environments has been a continued challenge to plant breeders worldwide. The characterization of stable genotypes is often complicated by the frequent occurrence of genotype-by-environment interactions (GEI). Plant breeders, to address the GEI issue, evaluate genotypes in multi-environment trials (MET) inclusive of favorable as well as unfavorable conditions. The MET analysis of variance provides good estimates of genotype (G) and environment (E) main effects and genotype-by-environment interaction (GEI) effects out of which G and GEI effects are mainly relevant to cultivar evaluation. A significant GEI not only results in changes in relative ranking or performance of genotypes from one environment to another, but also hinders breeding progress while selecting and advancing genotypes to subsequent stages of selection (Pham and Kang, 1988). Furthermore, increased GEI is usually associated with decreased correlation between genotypic and phenotypic values which in turn reduces gain from selection.
A genotype is considered to be stable if its performance is relatively constant across varying environments. According to Becker and Leon's (1988) concept of biological or static stability, a stable genotype is the one with minimal variance for yield across different environments. However, this concept has received little attention from breeders and agronomists as they prefer genotypes with high mean yields and the potential to respond to good agronomic inputs and favorable environmental conditions (Becker, 1981). Cultivar or a hybrid with a constant high yield referred to as dynamic stability is the preferred choice in commercial plant breeding.
Several stability analyses have been proposed to handle GEI so as to recommend the genotypes that perform consistently better and yield higher across different locations. Stability indices are either based on regression analysis or principal component analysis (Bernardo, 2002). Some of the most commonly used stability parameters are Finlay and Wilkinson (1963) regression coefficient, Eberhart and Russell (1966) deviation from regression, Shukla (1972) stability variance and Kang (1993) yield stability parameter and the additive main effects and multiplicative interaction model (Gauch, 1992). The additive main effects and multiplicative interaction (AMMI) analysis is an efficient means to determine stable and high yielding genotypes, provides a biplot as well as information on main and interaction effects. Genotypes with a first principal component (PC) axis value near zero represent stable genotypes and their mean performance can be predicted from the main effects model.
According to Flores et al. (1998), there are three groups of stability parameters viz., (a) Group I - statistics which are associated only with yield and show little or no correlation with stability parameters [e.g. Pi method of Lin and Binns (1988) and UPGMA method of Sokal and Michener (1958)], (b) Group II - statistics that consider both yield and stability parameters simultaneously [e.g. AMMI model of Gauch and Zobel (1996) and Kang's (1993) rank-sum method] and (c) Group III - statistics that are associated only with stability and show little or no correlation with yield [e.g. joint regression analysis of Eberhart and Russell (1966) and Shukla (1972) methods].
Parametric methods such as ANOVA and regression provide a measure of GEI. However, an overall response of genotypes to environments and therefore their stability cannot reliably be predicted using these approaches. This is because a genotype's response to environment, in reality a multivariate situation (Lin et al., 1986), is considered as a univariate problem by parametric methods. To circumvent this situation, the idea of clustering genotypes based on their response structure has emerged. The main advantage of this type of classification is that the relative relationships among the genotypes will not change and is independent of any specific set of analyzed data. Recently, Yan et al. (2000) developed a biplot technique named “GGE Biplot” which graphically represents the genotype (G) main effects plus genotype-by-environment interaction (GEI) effects. It was observed that environmental variation was often found to be much higher (>80%) compared to G + GEI variation (<20%); however, G and GE are the two main sources of variation that are relevant to cultivar evaluation. Therefore, the GGE biplot represents the GGE of MET data constructed by simultaneously plotting the two (or more) PC scores of genotypes and environments. The PC scores are generated by subjecting the GGE data to singular value decomposition thereby reducing the noise caused by environment (E) main effects. For genotype evaluation, the basic features of a GGE biplot are as follows: a small circle in the centre of a biplot indicates average environment coordination (AEC) which is the average of environmental PC1 and PC2 scores. The single arrowed line passing through the small circle via biplot origin is called the AEC abscissa with its arrow pointing towards increasing yield. The AEC ordinate (double arrowed line perpendicular to the AEC abscissa passing through the biplot origin) indicates (in) stability. The genotypes are ranked along the AEC abscissa and their stability is projected as the vertical line from the AEC abscissa. A highly unstable genotype would have longer projection on the AEC abscissa irrespective of its direction (Yan, 2001, Yan and Kang, 2003).
Since the Eberhart and Russell (1966) method (from now onwards termed the E-R method) is widely used to make plant breeding advancement decisions, this study was undertaken to compare the E-R method with GGE biplot analysis.
Section snippets
Field evaluation
In 2007, 20 experimental hybrids and five commercial checks were evaluated at 24 locations (Table 1) across the mid-west in the USA. Most of the locations were in Nebraska, Indiana, and Illinois with a very few in Iowa, Kansas, Missouri, Kentucky and Ohio. Of the 20 hybrids, one of the hybrids had consistently poor populations; therefore it was removed from the final analysis. All the experimental hybrids were in the advanced stages of testing. The trial was planted in randomized complete block
ANOVA
The analysis of variance results are presented in Table 2. In this analysis, since there were no replications, the GE interaction effect is confounded with the error term. Therefore, the overall residual effect was used as GEI effect. The ANOVA results indicated that the location (E) and genotype (G) main effects were significant implying a substantial variation among the genotypes as well as locations. The highest percentage of variation was explained by E main effect (>80%) while G and GE
Discussion
The Eberhart and Russell (1966) regression model has been widely used in the past few decades mainly because variability in performance of any genotype could be subdivided into predictable (regression) and unpredictable (var-dev) components. Hypothetically, the E-R method considers both yield (regression) and stability (var-dev), with regression being predicted and to some extent controlled by selecting specific genotypes for specific locations. In this study, the three high yielding genotypes
Conclusion
The results from this study indicated that there exists a significant amount of GE interaction between the 24 genotypes across 24 environments. Although the widely used E-R regression analysis provided useful information on stable genotypes, a broader picture in conjunction with high yield could not be wholly predicted. In this study, where the E-R method's variance deviation from MS could potentially be regarded as a stability parameter, the slope (or the regression coefficient) could not be
Acknowledgements
The authors are immensely grateful to the anonymous reviewers whose comments and suggestions have greatly helped in improving the standard of this manuscript. Authors are also thankful to Andrew Hopkins and John Zwonitzer of Dow Agro Sciences for kindly reviewing this manuscript.
References (20)
- et al.
A comparison of univariate and multivariate methods to analyze G × E interaction
Field Crop Res.
(1998) - et al.
Interrelationship among and repeatability of seven stability indices estimated from commercial cotton (Gossypium hirsutum L.) variety evaluation trials in three Mediterranean countries
Euphytica
(2008) - et al.
Stability analysis in plant breeding
Plant Breed.
(1988) Correlations among some statistical measures of phenotypic stability
Euphytica
(1981)Quantitative Traits in Plants
(2002)- et al.
Stability parameters for comparing varieties
Crop Sci.
(1966) - et al.
The analysis of adaptation in a plant breeding program
Aust. J. Agric. Res.
(1963) Statistical Analysis for Regional Yield Trials: AMMI Analysis of Factorial Designs
(1992)- et al.
AMMI analysis of yield trials
Simultaneous selection for yield and stability in crop performance trials: consequences for growers
Agron. J.
(1993)
Cited by (80)
Genotype × environment interaction and selection of maize (Zea mays L.) hybrids across moisture regimes
2021, Field Crops ResearchCitation Excerpt :In present study, hybrids ZH16929, VH123021, ZH161034, VH131167, ZH17191 and ZH16929-1 were identified as most desirable genotypes across the test environments, indicating the inherent potential of these hybrids for higher yield and wider adaptation under optimal, drought and waterlogged conditions. The relative contributions of mean grain yield and stability by the GGE biplot methodology in our study was similar to previously reported studies on maize (Alwala et al., 2010; Oyekunle et al., 2017). According to Yan (2002), the cosine of the angle between the two environmental vectors explains their association between them.
Genetic parameters and genotype by environment interactions influencing growth and productivity in Masson pine in east and central China
2021, Forest Ecology and ManagementCitation Excerpt :GGE is attracting increasing attention from researchers. It can provide complete and intuitive evaluations for multisite data; furthermore, it can be used to distinguish and represent the test environment and identify the most suitable varieties in a specific site and the varieties with high and stable yields (Alwala et al., 2010). Although the GGE biplot approach has the advantage of graphical GEI, it requires balanced data and homogeneous environmental errors in different sites and a limited fixed effect model, which greatly limits its application in forestry (Yan, 2010).
Regional multi-environment analysis of corn productivity and yield stability as impacted by hybrid maturity
2021, Field Crops ResearchCitation Excerpt :Well informed hybrid maturity decisions can be an essential management adaptation strategy to maximize productivity and yield stability, as well as resource use efficiency in different environments. The analysis of the genotype x environment (G x E) interaction in previous studies has focused on the yield stability of individual corn hybrids across multi-environment trials to aid in breeding selection (Alwala et al., 2010). However, producers are faced with a wide option of hybrid maturities to select from within a given region, and multi-environment studies that evaluate differences across corn hybrid maturities are limited (Assefa et al., 2016).
Pennycress domestication and improvement efforts
2024, Crop ScienceGenotype-by-environment interaction analysis for cotton seed yield using various biometrical methods under irrigation regimes in a semi-arid region
2024, Archives of Agronomy and Soil Science