Analysis of Genotype by Environment Interaction of Improved Pearl Millet for Grain Yield and Rust Resistance

Pearl millet is grown by inhabitants of the semi-arid zones. Due to the unpredictable climatic conditions the genotype-by-environment interaction (GEI) makes it hard to select genotypes adapted to such conditions. The study objectives therefore were to analyse the patterns of GEI and to identify superior genotypes for grain yield and rust resistance. Seventy six genotypes were planted in four environments in 4×19 alpha design with two replications. The ANOVA results showed that main effects of environments were significant (p ≤ 0.05) for grain yield and highly significant (p ≤ 0.001) for rust resistance while the main effects of the genotypes and their interactions with environments were also important for grain yield and rust severity at 50% physiological maturity. The GGE biplot analysis revealed that environments associated with more rains received during vegetative phase performed better than those receiving more rains during post-anthesis phase. The winner in the best environment for grain yield was ICMV3771×SDMV96053 while Shibe×CIVT9206 and Shibe×GGB8735 were the best for rust resistance.


Introduction
Pearl millet is adapted to environmentally marginalised conditions worldwide (Bashir et al., 2014) and a multipurpose (IFAD, 1999) cereal for people living in semi-arid areas in Uganda (Lubadde et al., 2014).However, on-farm productivity is low partly due to the effect of rust disease.The economical approach to control rust is through resistance breeding (Singh, 1990) and selecting genotypes adapted to low input and drought-prone environments (Vadez et al., 2012).Unfortunately, the potential performance of improved genotypes under marginal conditions is always obscured by the effect of genotype by environment interaction (GEI) (Yan & Racjan, 2002); leading to selection of genotypes not suitable for particular environments (Cooper & Delacy, 1994) and subsequently leading to low yield.It is therefore important to assess GEI effect before releasing varieties (Gupta & Ndoye, 1991;Haussmann et al., 2012).Several methods have been adopted to assess GEI in pearl millet breeding but the GGE-biplot analysis was used in this study because of the ability to graphically better explain the genotype and genotype by environment components of variation and being more efficient in discriminating genotypes and environments (Yan et al., 2007).

Experimental Materials and Study Environments
The experimental materials are shown in Table 1.The seventy six genotypes were evaluated in four pearl millet growing environments in Uganda.They included 60 single cross hybrids developed by crossing six male parents with ten female parents in a North Carolina2 design.The environments were defined as seasons by sites combinations.Environments; E1 was Kitgum site and 2012 second rains; E2 was Kitgum site and 2013 first rains; E3 was Serere site and 2012 second rains while E4 was Serere site and 2013 first rains.

Female parents
Male parents Crosses

Experimental Sites and Field Layout
The Kitgum location was 03°13′N, 032°47′E, 969 m.a.s.l.while Serere location was 01°32′N, 033°27′E, 1140 m.a.s.l.The genotypes were replicated twice in 4×19 alpha mating design.The materials were planted in four rows each of 5 m long and 60 cm × 30 cm spacing.Fertiliser application was according to Khairwal et al. (2007) and inoculation procedure done according to Thakur et al. (2011).

Data Collection and Analysis
Data was collected on 36 randomly selected plants per plot using the IBPGR and ICRISAT (1993) manual and traits were rust resistance at 50% physiological maturity (PSM 50 ) determined according to Tooley and Grau (1984) and grain yield (kg ha -1 ).The rust resistance was determined from rust severity data collected from the third leaf from top of the plant.This PSM 50 trait was used instead of the area under disease progress curve (AUDPC) since it had a significant effect on grain yield.The trait also seems to be more realistic since it is determined when there is no more change in grain yield.The analysis was done using the Breeding Management System 3.0 (IBP-BMS, 2014) and Genstat (Payne et al., 2012).

Combined Analysis of Variance for Assessing GEI and Performance of Environments
The combined analysis of variance (ANOVA) results (Table 2) indicate the main effects of environments being significant (p ≤ 0.05) for grain yield and highly significant (p ≤ 0.001) for rust resistance at 50% physiological maturity.The main effects of the genotypes and the interaction effects between genotypes and environments were also significantly (p ≤ 0.05) important for the two traits.Results further show that generally the coefficients of determination (R 2 ) estimated from the AMMI model were low for the traits; an indicator that a greater variation was due to the environments.This is corroborated by the rainfall pattern variation observed in each environment (Figure 1).The performance of environments was influenced by the rainfall amount received where the best performing environments in terms of grain yield and rust resistance received lower rainfall amounts.Coincidentally in the best performing environments (E1 and E3), most rainfall was received during the vegetative phase while in the poor performing environments most rainfall was received during the flowering phase.When rainfall is received during flowering, there is disruption of the pollination process since the pearl millet is predominantly outcrossing, with support by wind, a probable reason why the environments performed poorly in terms of grain yield.Heavy rainfall during flowering also causes reduced seed set and poor grain quality (DPP, 2011) in addition to promoting rust and consequently low grain yield.The variation in performance highlights the importance of environments in genotype performance and consequently GEI in trait expression.Rainfall pattern is one of the factors also reported by Gebre (2014) as being a source of variable performance of improved genotypes.The environments being important in genotype performance has also been reported in several pearl millet studies (Ezeaku et al., 2014;Misra et al., 2010;Gupta et al., 2013).The ANOVA adequately identified GEI as a significant source of variation but it is not able to explore the nature (Matus-Cadiz et al., 2003) of the GEI which may mask the true performance of genotypes in certain environments (Crossa, 1990) and thus the need to explore more methods; for which case GGE biplot was adopted.Note.LSD testing done at α = 0.05; ** = highly significant with p ≤ 0.001, * = significant with p ≤ 0.05.
Generally the genotypes associated with high rust resistance were also highly unstable in terms of grain yield and associated with the unstable environments E2 and E4 (Figure 6).These observations emphasize the importance of GEI and adopting selection for specific environments.In many pearl millet studies the GGE biplot has also been used to identify pearl millet mega environments to reduce number of test environments (Gupta et al., 2013;Ishaq et al., 2014).Mashiri et al. (2014) adopted the GGE biplot technique to estimate environmental effects for days to flowering, plant height and physiological maturity (Bashir et al., 2014).In addition, Gebre (2014) and Mustapha and Bakari (2014) used the GGE biplot analysis to identify stable genotypes with high grain yield while Bashir et al. (2014) identified best performers for grain yield.Thus, the practicality in using the GGE biplot merits its use in selecting for stability and adaptability of genotypes for grain yield and other yield-related traits.

Conclusion
The study focused on establishing the genotype by environment interaction effect, characterising environments and genotypes.The ANOVA results showed that the effects of environments, genotypes and genotype x environment interaction (GEI) were important in trait expression and performance of genotypes.In addition, it was observed that amount of rainfall received at both vegetative and post-anthesis phases had an effect on grain yield and disease severity.Finally, the GGE biplot was useful in concisely characterising the environments and the genotypes.It characterised the environments in terms of stability and productivity.This resulted in grouping of mega environments with E2 and E4 being ideal for rust discrimination while E1E3 was the best for grain yield; implying that environment-specific selection should be adopted.

Figure 1 .
Figure 1.Total amount of rainfall received during the evaluation period and performance of environments Source for rainfall data: Department of Meteorology, Ministry of Water and Environment, Uganda.

Figure 2 .
Figure 2. Performance of environments in terms of rust severity grain yield = 2355 kg ha -1

Figure 3 .Figure 6 .
Figure 3. Ranking genotypes based on both means and stability for grain yield across environments