Selection of Suitable Genotypes of Urdbean (Vigna mungo L.) for Targeted Environments of Hilly Terrains of India using GGE Biplot and AMMI Analysis

Urdbean [Vigna mungo (L.) Hepper] is an important short duration pulse crop mainly grown in South Asian regions in different seasons under rainfed based agro-ecosystems (Gupta et al., 2001). It is considered as potential nutritious component of vegetarian diet containing 25-28 per cent protein along with other essential vitamins and minerals (Kumar et al., 2014; Kumar et al., 2015). It also plays an important role in sustainable agriculture via enriching the soil by atmospheric nitrogen fixation, which allows it to grow on marginal soils and improve them. On account of its short duration it is ideal candidate crops for cultivation as sole crop in rice fallow situations under residual moisture conditions (Gupta et al., 2016). India is the largest producer and consumer of urdbean cultivated in an area 4.50 mh with 2.83 mt production. The major contributing states are Andhra Pradesh, Maharashtra, Orissa, Madhya Pradesh, Tamil Nadu, Gujarat, Karnataka and Uttar Pradesh (Parihar et al., 2017a; Anonymous 2018). It is being grown over a sizeable area in country, though average productivity is quite low as compared to other pulses (Kumar et al., 2014). The low productivity is resulted due to occurrence of several biotic and abiotic stresses during cropping seasons. To minimize crop yield reduction due to various environmental factors, the quintessential step is to identify genotypes which are more adaptable and stable in performance over different locations and years with high yield potential. Therefore, to reduce the effects of the GE interaction, it is compulsory to analyze the adaptability and stability of each genotype so as to recognize genotypes with expected performance that are responsive to environmental variations (Cruz et al., 2004).
        
Although, several improved varieties of urdbean have been developed; many of them mostly oscillated in terms of yield performance under different locations due to substantial genotype-environment interactions (GEI). GEI complicates the breeding program by ambiguous performance of genotypes across environments which subsequently influenced selection of superior genotypes (Ebdon and Gauch, 2002). In addition, GEI is responsible for poor association between phenotypic and genotypic values of traits of interest, in that way, it condensed expected improvement from selection. Therefore, it is pertinent to understand the genotypic, environmental and their interaction effects to find out suitable genotype for targeted growing environment using multi-environment evaluation. The selection of method for assessing adaptability and stability is dependent on number of accessible environments as well as the type of information and the level of experimental precision required (Cruz et al., 2004). Several approaches have been used over the years to dissect and understand the genotype × environment (G×E) interaction, of them some were based on regression analysis (Eberhart and Russell, 1966; Perkins and Jinks, 1968; Freeman and Perkins, 1971), these models are very restrictive in the type of interaction for which they account. Another popular method is AMMI model which has more flexibility and usefulness for better understanding of GEI. It incorporates both the additive and multiplicative components of the two-way data structure and represents GEI patterns graphically (Mukherjee et al., 2013). The AMMI and GGE analysis explained around 50% of the sum of squares of the genotype × environment interaction, whereas the method of Eberhart and Russell explained very less GEI (Namorato et al., 2009). As compare to AMMI the GGE biplot is superior in term of mega-environment analysis and genotype evaluation because it explains more G+GE. In addition, the discriminating power vs. representativeness view of the GGE biplot is effective in evaluating test environments, which is not possible in AMMI analysis (Yan et al., 2007).  During last decade researcher extensively used  many techniques in different crops including pulses to delineate GEI (Das et al., 2019; Parihar et al., 2017b; 2017c; Parihar et al., 2018; Luo et al., 2015a; 2015b; Yan et al., 2007; Luo et al., 2009). Of them, AMMI and GGE biplot are the most popular and extensively used statistical methods for analysis of crops multi-environment trials data (Hugh 2006; Zhou et al., 2011). Therefore, this investigation was carried out to comprehend the nature and magnitude of genotype × environment interaction and, selection of superior and stable genotypes for targeted environments using AMMI and GGE biplot analysis.

A panel of 20 diverse urdbean genotypes including check variety Uttara (Table 1) was evaluated in a randomized complete block design (RCBD) with three replications at three distinct environments viz., Berthin (E1), Srinagar (E2) and Imphal (E3) under rainfed conditions during rainy season of 2016 with the sowing date of 12.07.2016, 07.07.2016 and 20.07.2016. The experiment was executed at the Experimental Stations of Berthin, CSKHPKV, Himachal Pradesh, Sher-e-Kashmir University of Agricultural Sciences and Technology of Kashmir, Budgam and Central Agricultural University, Imphal, Manipur. Each plot consisted of five rows of 3.5 meter length. Row to row and plant-to-plant distances were 30 cm and 10 cm, respectively. The grain yield was recorded for each plot in each test environment and mean yield was computed in accordance with the experimental design. The PROC GLM procedure available in SAS (Ver. 9.3, SAS Institute, 2011) was used to partition yield variation into genotypes, environments and their interactions. The grain yield data were subjected to AMMI and GGE biplot analyses. The results of the AMMI model analysis were interpreted from the AMMI1 and AMMI 2 graphs that represented the main and first multiplicative (PC1) and second multiplicative (PC2) axes in terms of both genotypic and environmental effect. The GGE biplots were developed from the first two principal components (PC1 and PC2) resultant by using environment-centered yield data to singular valued composition (SVD) as  earlier researcher (Yan et al., 2000; Yan, 2002). To generate graphics which portrayed (i) “which-won-where” pattern, (ii) ranking of cultivars on the basis of yield and stability used PBTools software version 1.4. (STAR, 2014).
 

The AMMI analysis of variance devised that the genotype, environment and genotype × environment effects were significantly influenced the grain yield. Given results demonstrated that the environmental effects contributed maximum part (46.29%) of the total variation, which indicated that environmental factors significantly affect yield performance of tested urdbean genotypes. Such results have also been reported by Ceyhan et al., (2012) and Tolessa et al., (2013) in fieldpea, wherein environmental factors substantial influenced crop performance. In present experiment the variation in grain yield due to environmental components suitably justifying the need of identification of stable genotypes across environments. Further, genotype × environment interactions and genotypes effects explained 25.90% and 22.23% of the total variation, respectively (Table 2). The significant genotype × environment interactions effects witnessed that genotypes yield oscillated in different environments which confirming the necessities of screening urdbean genotypes in multi environment trials in hilly terrains of northern India. While ever genotypes were evaluated for adaptability in various environments, a crossover genotype × environment interaction most often observed (Ceccarelli et al., 2006). The high GE interactions affected performance of genotypes from one environment to another environment which indicates the possibility of identifying different winning genotypes for different environments (Yan 2002, Yan and Tinker 2006). Tarakanovas and Ruzgas (2006) and Abdipur and Vaezi et al., (2014) reported that the more contribution of GEI in total variation in comparison to genotypic effects is owing to genes controlling varied yield related attributes among environments. The GEI often create lots of confusion in selection of particular cultivar in a breeding programme and forces breeder to test genotype stability (Chatterjee et al., 2019).
 

        
Further, the AMMI analysis partitioned the GEI into two IPCAs (IPCA 1, IPCA 2) which explained total variation. The two IPCA represent total variation which justifying the use of AMMI model for the present data set and suitability of two IPCA for proper interpretation of data indicated. Similar finding have been reported by Tadesse et al., (2017) and Getachew et al., (2015) in which two IPCA cumulatively accounted around 88% and 74% of the total GEI, respectively.
 
Assessment of genotypes and locations using GGE biplot
 
GGE biplot partitioned complex GEI into two different principal components (PCs) and presented graphically. The two dimensional ‘which-won-where’ patterns of genotype plus genotype × environment biplot is the most important tool for mega - environments analysis in varietal trials (Yan et al., 2000; Yan et al., 2007). This view of GGE biplot developed adopting environment centered multi-environment data. The GGE biplot captured 99.3% of the total variations through PC1 (91.0%) and PC2 (8.3%) (Fig 1). This indicated that biplot constructed by plotting the first principal component scores (PC1) of genotypes and the locations against their respective scores for second principal component scores (PC2) sufficiently captured the environment centered data. These finding are in agreement with earlier report by Segherloo et al., (2010) in chickpea, wherein the first two principal components explained 95% of the total GGE variation.
 

        
Further, in ‘which-won-where’ pattern of GGE biplot the perpendicular lines divide the polygon into distinct sectors, each having its own winning cultivar which is at the vertex of sector (Yan et al., 2000). Genotypes with lowest and highest grain yield were at different vertices of polygon and contributed maximum in GE interactions. Based on ‘which-won-where’ biplot the genotypes G9 and G20 were the best performing over the locations (Fig 1). While the vertex cultivars identified in this study are G18, G14, G5, G8, G20 and G9. These cultivars are the highest yielders for the environment which is present inside those sectors. The polygon showed that the genotype G18 was contributed most to the interaction, i.e., showing the highest or the lowest contribution in different environments (Fig 1). Also, the three different sector/mega-environments produced in scatter biplot represented the variability in the environments. In mega environment I (Berthin) the best performing genotypes were G20, G8 and G6. Likewise in mega-environment II (Srinagar) the best performing genotype was G5 and in mega-environment III (Imphal) G18 was as a winning genotype. The present findings are in corroboration to the earlier finding in different pulse crops (Parihar et al., 2017b: Parihar et al., 2017c; Parihar et al., 2018; Alam et al., 2014; Karimizadeh et al., 2013: Jeberson et al., 2019a: Jeberson et al., 2019b) in which different mega-environment specific genotypes were identified using which-won-where GGE biplot pattern.
        
The ‘mean versus stability’ GGE biplot is the average-environment coordinates (AEC) view of biplot (Fig 2). The AEC ordinate (‘X’) passes through the plot origin with an arrow indicates the positive end of the axis. The AEC ‘Y’ axis or the stability axis passed the plot origin with both side arrow and was perpendicular to the ATC ‘X’ axis points to greater variability (poor stability) in either direction. The average yield potential of the genotypes is approximated by the projections of their markers to the ATC ‘X’ axis. Genotypes G9 and G20 had the highest mean yield and the G14 had the poorest mean yield. An ideal genotype was the one that had both high mean yield and high stability hence the ideal genotypes were G9 and G20. These results are in accordance to earlier reports (Parihar et al., 2017b; Parihar et al., 2017c; Rezene et al., 2014; Sharma et al., 2012) in which they have identified different genotypes for different traits based on mean versus stability graphical representation of GGE biplot.
 

 
Evaluation of genotypes and locations using AMMI analysis
 
AMMI biplot analysis is a powerful interpretive tool for understanding of GE interactions. AMMI 1 biplots, in which both main effects along with IPCA1 scores (genotype and environment) were plotted against each other (Fig 3).
 

        
It is graph in which displacement along the abscissa indicate difference in additive effects, whereas displacement along the ordinate indicate differences in the interaction effects. Genotypes which come in same groups have similar adaptation while environments which group together influence the genotypes in the same way.  The AMMI 1 biplot accounted 89.4% of the total variations. The environments i.e., Berthin, Imphal and Srinagar were positioned in different quadrants like II, III and IV quadrants, respectively. Genotype or locations presented on the same parallel line, relative to the ordinate, had similar yield, while those located on the right hand side of the midpoint of the axis had higher yields than those on the left hand side. The AMMI 1 biplot showed four groupings of the genotypes, G14, G12 and G10 (generally low yielding and unstable) and the genotype G5 (low yielding and moderately stable). Genotype G2, G11, G15, G16 and G7 were high yielding and stable; G9 and G20 were high yielding but unstable. Similar results were also reported in mungbean where AMMI has divided the genotypes based on the stability as moderate, high and unstable genotypes (Singh et al., 2014; Waniale et al., 2014).
        
Genotypes with IPCA 1 scores near to zero had small (little) interaction across locations, while genotypes with very high IPCA 1 values had sizeable interaction with environments. When an environment and genotype have the similar sign on the PCA axis, their interaction is positive, and vice versa (Negash et al., 2017). The AMMI 1 biplot clearly illustrated that all the twenty entries differed from each other not only for mean grain yield but also for their G × E interaction effects. The genotypes G14, G12, G10, G18 and G1 had low interactions with environments characterized by low IPCA scores (Table 3) and were stable showing broad adoptions across the locations, while the genotypes with higher IPCA scores, G20, G9, G4, G6 and G13, were highly interactive and less stable across locations.
 

        
The AMMI 2 biplot illustrated scores for IPCA 1 and IPCA 2 (Fig 4). The genotypes near the origin are less interactive to environmental interaction and those positioned far from the origins are more interactive and have large G×E interaction. Consequently, the entries G12 and G14 were highly interactive since these were situated away from the origins. The environments E1, E2 and E3 did not form any Group (Fig 4) on the plot and they were also located far from the origin and had different interaction pattern on genotypes. The lines with one side arrow connected to the origin by vector indicate the interactive forces. Hence, all the environments had long spokes which indicate its strong interaction. The Berthin (E1) had high grain yield and high IPCA 2 score with minimal negative IPCA 1 score. Similarly, the low potential environments Srinagar (E2) had negative IPCA score and low grain yield. Thus, the biplot recognized Berthin as the highest yielding environments and Srinagar as the lowest yielding environment for urdbean. The AMMI 2 analysis revealed that genotypes G7, G16 and G13 had wide adaptation. These were less affected by G × E interaction, thus performed well across wide range of locations.
 

To understand effects of genotypes and environments and their interaction on grain yield multi-locations trials are basic requirement. All the tested entries were differed from each other not only for mean grain yield but also for their G×E interaction effects. Both GGE biplot as well as AMMI analysis findings are almost similar in the detection of stable genotypes for the test locations. GGE biplot clearly shown the genotypes G6 and G20 is suitable for environment E1, G5 is suitable for environment E2 and G18 is suitable for environment E3. Suitability of these genotypes predicted through graphical visualization and AMMI has shown the stable genotypes and which one is good environment for seed yield. The study revealed that genotypes VBG 12-034(G-7), IPU 13-3(G-16) and COBG 13-04(G-13) were stable genotypes with wide adaptation based on GGE and AMMI analysis. These genotypes are suitable for hilly terrains of India and can be utilized as donor for future breeding programmes of urdbean. Furthermore, both AMMI and GGE biplot techniques can be used in future for identification of stable and high yielding genotypes across environments.