Introduction
The ultimate aim of any breeding programme is to develop a stable performing line/hybrid for a targeted trait. The importance of phenotypic stability of genotypes in plant breeding has been widely advocated by several researchers so far. As the environment is highly dynamic in nature, it exerts the major effect on phenotypic expression of traits which result in the variation in the performance of a genotype across and within the location over the seasons (Koundinya et al., Reference Koundinya, Pandit, Ramesh and Mishra2019; Sharma et al., Reference Sharma, Leskovar, Crosby and Ibrahim2020). Hence breeders always look for the genotypes with least interaction with the environment and exhibit a stable performance thereby optimizing the resource-use efficiency for better yields (Ahmad et al., Reference Ahmad, Hassan and Bakhsh2005). For this, breeders resort to multi-environment trials (METs) which help in estimating the effect of genotype × environment interaction (GEI) on the trait(s) and based on the degree of GEI, stable performing lines can be selected. The success of these METs depends on how accurately the effect of GEI has been analysed.
Different methods, viz. univariate and multivariate, have been proposed for the same, such as two-way analysis of variance (ANOVA), linear regression analysis, principal component analysis (PCA) and additive mean effect and multiplicative interaction (AMMI), GGE (Genotype + Genotype × Environment) model. The AMMI model revolutionized the GEI analyses by dividing the G × E component of variation into principal components (PCs) thereby making a graphical representation of the stability of genotypes (Zobel et al., Reference Zobel, Wright and Gauch1988). GGE biplots display the genotype and GEI at the same time besides finding the mega environments and identifying the genotypes which performed better in which environment (Yan and Tinker, Reference Yan and Tinker2006). To represent the entire variation in all the PCs, instead of PC1 alone as in AMMI, a new statistic weighted average of absolute score (WAAS) is developed which takes into account the total GEI and assign a score to genotypes. This WAAS is used in the Y-axis and mean values in the X-axis in the biplot to identify stable genotypes with better performance (Olivoto et al., Reference Olivoto, Lúcio, da Silva, Marchioro, de Souza and Jost2019a). In combination breeding programmes, breeders concentrate on multitrait improvement. But, these stability statistical models indicate genotype stability for only one trait (Koundinya et al., Reference Koundinya, Ajeesh, Hegde, Sheela, Mohan and Asha2021). Hence, a multitrait stability index (MTSI) is developed to identify stable and better performing genotypes based on all the traits under consideration (Olivoto et al., Reference Olivoto, Lúcio, da Silva, Sari and Diel2019b). Previous literature review suggested that a few studies have been carried out on determining the phenotypic stability of seed yield and its attributing traits by Jain and Patel (Reference Jain and Patel2012); Pawan et al. (Reference Pawan, Garg and Manoj2016) and Teja et al. (Reference Teja, Saidaiah, Kumar, Geetha and Bhasker2022) using linear regression models but studies on improving and identifying stable vegetable podded cluster beans is scarce. Further, as said earlier linear regression models leave considerable variation without taking into account. In our experiment, the data were analysed by using advanced statistical tools, viz. WAAS and MTSI.
Cluster bean being an underutilized legume vegetable, native to north-west India and Africa has immense potential as a nutritious vegetable for diet diversification in the climate change scenario. It is also a drought-tolerant and green manuring crop adaptable to tropical climatic conditions. The vegetable cluster bean variety Pusa Navbahar, bred in the 1980s by Indian Agricultural Research Institute, New Delhi is still the ruling variety in India. Hence, to develop a high-yielding and nutritionally superior vegetable-type cluster bean we initiated our research work. Cluster bean germplasm collected from all over India were evaluated for their yield and horticultural traits and the genetic variability and character association for the quantitative traits were studied (Mishra et al., Reference Mishra, Aghora and Senthil Kumar2020). Further, the divergence present in the germplasm was studied and genotypes were grouped into different clusters based on their genetic distance (Mishra et al., Reference Mishra, Aghora and Venugopalan2019). It was observed that all the vegetable-type cluster beans were grouped into cluster I, characterized by fleshy, smooth and medium-long pods with slow-growing seeds. These vegetable-type lines were further evaluated for yield traits and five lines were shortlisted based on their performance. Pure line selection was continued in these shortlisted lines to achieve uniformity as they were local collections from different Indian states (online Supplementary Table S1). To identify a stable and best-performing line among them, the following investigations were carried out.
Materials and methods
Experimental material, experimental design and test location
Experimental materials include five advanced breeding lines developed in cluster bean i.e. IIHRCB 22-1-1, IIHRCB 23-1-2, IIHRCB 26-2-1, IIHRCB 27-1-3 and IIHRCB 32-1-3 (pedigree information is provided in online Supplementary Table S1) along with check Pusa Navbahar. The materials were grown at the Vegetable Research farm of ICAR-Indian Institute of Horticultural Research, Bengaluru (13°7′ N, 77°29′ E). The experiment was laid out in the field in randomized block design with four replications following 30 cm spacing between the rows and 10 cm spacing between plants. Total 40 plants per replication per genotype were maintained. The weather of the experimental site is sub-tropical climate. The experiment was laid out in two different seasons in two consecutive years, viz. Kharif, the rainy season and Summer, the dry and hot season. The weather parameters recorded during the experimental period are given in online Supplementary Table S2. The day temperatures during Summer in 2019 and 2020 were above 30°C, while the day temperatures were below 30°C in Kharif in 2019 and 2020. Rainfall was almost double during Kharif 2019 and 2020 when compared with the Summer 2019 and 2020. Humidity was also high during Kharif 2019 and 2020 than Summer 2019 and 2020.
Observations and statistical analysis
Observations were recorded on yield and yield attributing traits like number of clusters per plant (NCP), number of pods per cluster (NPC), number of pods per plant (NPP), single pod weight (SPW) and yield per hectare (YPH). NCP, NPC and NPP are measured as count data on 10 randomly selected competitive plants while SPW in g and YPH were calculated in t/ha. There were five harvestings in each season. The data over these five harvestings were added to obtain final NCP, NPC, NPP, SPW and YPH.
Statistical analysis
The data obtained from various parameters were subjected to the AMMI model analysis as per the statistical model given by Zobel et al. (Reference Zobel, Wright and Gauch1988). WAASs were calculated as per Olivoto et al. (Reference Olivoto, Lúcio, da Silva, Marchioro, de Souza and Jost2019a) by using the following formula. They were calculated from the interactive principal component axes (IPCAs) of G × E Interaction from AMMI ANOVA.
where WAASi = weighted average of absolute scores of the i th genotype or environment; IPCAik = absolute score of the i th genotype or environment in the k th IPCA and EPik = magnitude of the variance explained by the k th IPCA.
The nominal mean values were plotted against environmental IPCA1 to obtain which-won-where graphs. The nominal mean values were calculated as per the below formula:
where Ŷij = nominal mean for i th genotype in the j th environment, μi = grand mean of the i th genotype, IPCA1i = IPCA1 score of the i th genotype and IPCA1j = IPCA1 score of the j th environment.
MTSI was proposed by Olivoto et al. (Reference Olivoto, Lúcio, da Silva, Sari and Diel2019b). The scores obtained in factor analysis were used to calculate MTSI based on the genotype–ideotype distances. MTSI was calculated based on all the traits under the study, viz. NCP, NPC, NPP, SPW and YPH by using the following formula:
where Fij = j th score of the i th genotype and Fj = j th score of the ideotype.
Software
All the statistical analyses were performed using R open-source software 4.0.2 using the package Metan developed by Olivoto (Reference Olivoto2019).
Results
AMMI ANOVA
In the present experiment, it was observed from AMMI ANOVA that GEI is highly significant for all the traits (online Supplementary Table S3). Significant variation was observed for genotypes (G), environment (E) and their interaction (GEI) for all the traits under study. For the traits, yield per hectare and pods per plant, maximum variation was contributed by the genotypes which were 50.41 and 66.47% and for pods per cluster maximum variation was contributed by the environment (33.73%) while for the number of clusters per plant and single pod weight, error component had more variation (37.41, 33.43%) than genotypes, environments and their interaction. The effect of the environment was found to be low for the number of clusters per plant (5.99%) followed by pods per plant (17.50%). The entire GEI variation was captured by the first three IPCAs depicting that there was no noise in the model. All the PCs were found to be significant for all the traits. The first two IPCAs contributed a maximum of 87.9, 97.3, 94.6, 98.6 and 85.6% variation for yield per hectare, number of clusters per plant, pods per plant, pods per cluster and single pod weight, respectively, leaving a small but significant amount of variation in the third IPCA.
Genetic parameters
The genetic parameters estimated are presented in online Supplementary Table S4. A moderate genotypic coefficient of variation (GCV) was observed for yield per hectare (21.14%) and pods per plant (20.40%) while very low GCV values were observed for number of clusters per plant (7.75%), pods per cluster (9.71%) and single pod weight (6.83%) implying that the extent of variation existed was low. This was due to the inclusion of only five accessions in the experiment. Similarly, moderate-to-low PCV values were observed for the traits as 21.89, 10.10, 20.92, 11.86 and 8.29% for YPH, CPP, PPP, PPC and SPW, respectively. A slightly higher PCV values indicated the less influence of environment on these traits. A very high heritability was observed for YPH (93.30%) and PPP (95.11%) while a moderate heritability was observed for CPP (58.96%), PPC (66.94%) and SPW (67.95%). Moderate-to-low genetic advance over mean (GAM) values were recorded for all the traits pointing that there would not be much improvement in mean value through selection. Lack of high heritability coupled with high GAM suggested that selection within these accessions for these traits would not be fruitful. High coefficient of variations and high heritability coupled with high genetic gain for number of pods per plant, number of clusters per plant and pods per cluster was observed by Mishra et al. (Reference Mishra, Aghora and Senthil Kumar2020) in cluster bean when analysed with 38 accessions. The low genetic variation and genetic advance in present experiment was due to the use of the top five high yielding accessions selected from a diverse population. Hence, there would not be any significant further improvement through selection.
Mean versus WAAS biplot
The mean versus WAAS biplots for all the traits are presented in Fig. 1. The stable and fairly stable lines are given in Table 1. In our studies, the accession IIHRCB 27-1-3 was found to be stable for yield per hectare but its mean value was below average. Two accessions IIHRCB 22-1-1 and Pusa Navbahar have been considered fairly stable as their WAAS was near zero and had mean values more than average (>17.7 t/ha). Similarly, for single pod weight also IIHRCB 27-1-3 was observed as stable with a below-average mean value. The fairly stable accessions for this trait were IIHRCB 22-1-1 and IIHRCB 23-1-2 because of their near-to zero WAAS value and above-average mean values (>2.66 g). None of the accessions was qualified as stable for pods per plant. Although genotype IIHRCB 23-1-2 had the least WAAS, it was not recommended as it had below-average mean value (<80.01). For the number of clusters per plant, genotype IIHRCB 26-2-1 could be recommended as a stable genotype because of its near-to zero WAAS value combined with the above-average mean value (>17.22). In the case of pods per cluster and the number of clusters per plant, the genotype IIHRCB 32-1-3 had near-to zero WAAS value yet was not qualified to be recommended as a stable genotype because of the below-average mean value (<4.68). The accessions IIHRCB 26-2-1 and IIHRCB 22-2-1 for pods per cluster and IIHRCB 22-2-1 and Pusa Navbahar for the number of clusters per plant were considered fairly stable. The genotypes present in the second quadrant of the biplot are suitable for favourable conditions; such genotypes for various traits were Pusa Navbahar for pods per plant, Pusa Navbahar and IIHRCB 32-1-3 for single pod weight and IIHRCB 26-2-1 for yield per hectare.
Figure 1. (a) Mean versus WAAS biplot for individual traits: (A) clusters per plant and (B) pods per cluster. (b) Mean versus WAAS biplot for individual traits: (C) pods per plant; (D) single pod weight and (E) yield per hectare.
Table 1. Genotypes identified for various conditions based on mean versus WAAS biplot
The WAAS values of environments indicated that Summer 2021 had least interaction with the genotypes while Summer 2020 had maximum interaction with the genotypes for cluster per plant, pods per cluster and single pod weight. Kharif 2021 had maximum interaction with the genotypes for yield per hectare while Kharif 2020 had maximum interaction with the genotypes for pods per plant. Summer 2020 and 2021 interacted similarly for yield per hectare whereas Kharif 2020 and 2021 interacted similarly for the number of clusters per plant. For the remaining traits, all four environments interacted differently. The Summer season in 2020 and 2021 had a negative effect on the environment except for the number of clusters per plant which was found more during Summer. Kharif 2021 was found to be a highly favourable environment for all the traits except for the number of clusters per plant. The low temperature and high rainfall during Kharif season promoted vegetative as well as reproductive growth. The temperatures >30°C during summer season negatively affect the pollination and fertilization which might be the reason for the highest pods per cluster, yield per hectare and pods per plant during Kharif season. The available information was not sufficient to explain the more number of clusters per plant during Summer season.
Which-won-where
In general, these plots are used for identifying the genotypes which performed better in which environment (Fig. 2). The accession IIHRCB 22-2-1 had maximum nominal mean value for yield per hectare, single pod weight and pods per cluster during Summer 2020 and 2021. For yield per hectare and single pod weight, during Kharif 2020 and 2021, IIHRCB 22-2-1 had maximum mean values suggesting that it was an all-time winner for these characters. In the case of pods per cluster, it had the maximum nominal mean value in Kharif 2021 and the second highest in Kharif 2020 following IIHRCB 23-1-2. IIHRCB 26-2-1 recorded the maximum nominal mean value for pods per plant in all seasons except for Summer 2021. It had the second-highest nominal mean value during the Summer 2021 following IIHRCB 22-2-1. Akin to it, IIHRCB 26-2-1 had the maximum number of clusters per plant during Kharif 2020, Kharif 2021 and Summer 2021 and was having the second-highest mean value during Summer 2020 following IIHRCB 23-1-2. The accessions IIHRCB 32-1-3, IIHRCB 27-1-3 and IIHRCB 23-1-2 had the lowest mean values for yield per hectare and pods per plant across all seasons.
Figure 2. Which-won-where plots for individual traits in different environment: (A) clusters per plant; (B) pods per cluster; (C) pods per plant; (D) single pod weight and (E) yield per hectare.
Multitrait stability index
MTSI was calculated based on yield per hectare, single pod weight, pods per plant, pods per cluster and the number of clusters per plant. The factor analysis revealed that the first two PCs contributed 48.6 and 39.5%, respectively, which together explained 88% variation. The traits' number of clusters per plant, pods per plant and pods per cluster contributed 48.6% variation as they were associated with the first PC while yield per hectare and single pod weight contributed 39.5% variation (Table 2). After varimax rotation, the communality ranged from 63.8% for pods per plant to 98.9% for single pod weight. Based on Fig. 3, the genotype IIHRCB 22-1-1 could be selected or recommended at 15% selection intensity. The selection of this stable and superior genotype would result in a selection differential of 2.68, 22.0, 18.2, 35.0 and 10.5% for the characters number of clusters per plant, pods per plant, pods per cluster, yield per hectare and single pod weight, respectively.
Table 2. Communalities and selection differential of various characters based on MTSI
Xo, mean of genotypes; Xs, mean of selected genotypes; SD, selection differential.
Figure 3. Selection based on MTSI.
Discussion
AMMI ANOVA
The principal objective of any experimental design is to assign experimental units to treatments for precise quantification of errors and achievement of accuracy which improves recommendations, repeatability, selection and genetic gains. For analysis of experiments, the ANOVA model is commonly used to compare the mean response value of treatments at different levels of factors. It further partitions the total variance among treatments into three sources, i.e. genotypic main effect (G), environmental main effect (E) and GEI effect but does not analyse the GEI. This GEI is very complex. Hence, when the GEI is significant, analysis and interpretation of GEI are crucial to identify the stable genotype. Out of the various methods used for the analysis of GEI, we used the AMMI model as it combines the ANOVA for genotype and environment main effect with PCA of the GEI effect.
The significant GEI for all the traits under study explained that all the traits were influenced by the environment. The effect of environment on the phenotypic expression of the traits was found to be high for pods per cluster, more than the genotype effect, suggesting that this trait was influenced more by the changes in the environment. The traits, number of clusters per plant and single pod weight had more variance from the error component indicating that there was significant noise in the model which was explained by the factors not under consideration. Although the amount of variation due to the environment was low, it had a significant effect on the number of clusters per plant and pods per plant, it had a significant effect. The first two IPCAs captured 80–99% variation for all the traits implying that maximum GEI variation was captured by them. Previously significant GEI was observed by Rashidi et al. (Reference Rashidi, Farshadfar and Jowkar2013) and Shimray et al. (Reference Shimray, Bharadwaj, Patil, Sankar, Kumar, Reddy, Singhal, Hegde, Parida, Roorkiwal and Varshney2022) in chickpea; Ashwini et al. (Reference Ashwini, Ramesh and Sunitha2021) in horse gram; Kvitschal et al. (Reference Kvitschal, Filho, Scapim, Vidigal, Pequeno, Sagrilo and Rimoldi2006) and Adjebeng-Danquah et al. (Reference Adjebeng-Danquah, Manu-Aduening, Gracen, Asante and Offei2017) in cassava.
Mean versus WAAS biplot
The AMMI1 biplot considers only the first PC leaving a significant amount of interaction variance in the remaining PCs. Hence, WAAS which is based on all the PCs in the AMMI ANOVA was used in the AMMI1 biplot. WAAS biplot between WASS and mean value (Y) facilitates the simultaneous selection for performance and stability for an individual trait. Stable genotype is those with a WAAS value of zero or near-to zero as their interaction with the environment is nil or very little. But mere stability is of no use unless it is coupled with high mean values. Therefore the genotypes with a WAAS value of zero or near-to zero with above average mean value are recommended for general cultivation (Koundinya et al., Reference Koundinya, Ajeesh, Hegde, Sheela, Mohan and Asha2021). Based on the above-said criteria, only one accession was qualified for recommendation, i.e. IIHRCB 26-2-1 for trait, number of clusters per plant. As the number of clusters is an important yield component, it may impart homoeostasis in yield over the seasons (Koundinya et al., Reference Koundinya, Pandit, Ramesh and Mishra2019). The component compensation was previously explained by Eberhart and Russell (Reference Eberhart and Russell1966) who suggested selecting the genotypes which have stability for yield component traits. The biplots also helped in identifying the fairly stable genotypes with WAASs slightly higher than zero with above-average mean values. These genotypes could also be recommended for general cultivation as their interaction with the environment is less. Few genotypes interacted with the environment at a higher degree but had very high mean values. These genotypes are suitable for highly favourable conditions.
When two seasons are flocked together in the same quadrant, they are considered to have similar interactions with genotypes in both seasons for the trait under consideration. In the present experiment, there was no clear demarcation between the two major seasons, viz. Kharif and Summer. All four environments interacted differently with the genotypes for the majority of the traits. For clusters per plant, pods per cluster and single pod weight the Summer season interacted differently with the genotypes in 2020 and 2021. The effect of Summer season was found to be positive only for the number of clusters per plant when it had a negative effect on all the other traits which were favoured by Kharif 2021. Inferences on environment interaction with the genotypes based on WAAS versus mean biplot were previously studied by Koundinya et al. (Reference Koundinya, Ajeesh, Hegde, Sheela, Mohan and Asha2021).
Which-won-where
Olivoto et al. (Reference Olivoto, Lúcio, da Silva, Marchioro, de Souza and Jost2019a) provided a graphic interpretation by plotting the trait nominal means (Ŷij) against environmental IPCA1 values. The accession which records the maximum mean value across all environments is called an ‘all-time winner’. In our experiment, the genotype IIHRCB 22-2-1 had maximum mean values for yield per hectare and single pod weight in all seasons suggesting that it was an all-time winner for these characters. There was no all-time winner for other traits. The accessions IIHRCB 32-1-3, IIHRCB 27-1-3 and IIHRCB 23-1-2 had the lowest mean values for yield per hectare and pods per plant across all seasons suggesting their low productivity might be due to their genetic makeup. Based on these plots it is very difficult to differentiate Summer and Kharif as two mega environments as their interaction with genotypes was not in a specific pattern. Previously, Koundinya et al. (Reference Koundinya, Ajeesh, Hegde, Sheela, Mohan and Asha2021) identified genotypes for different environments and mega environments through these graphs.
Multitrait stability index
Olivoto et al. (Reference Olivoto, Lúcio, da Silva, Sari and Diel2019b) proposed a tool MTSI for identifying stable genotype based on multiple traits. It is based on the genotype–ideotype distance using the scores obtained in factor analysis. The considered ideotype has the highest WAAS Y values for all the observed traits (Olivoto et al., Reference Olivoto, Lúcio, da Silva, Sari and Diel2019b). Communalities indicated that single pod weight had maximum heritability (98.9%) followed by pods per cluster (93.6%) while pods per plant had the lowest heritability (63.8%). Lower MTSI values refer to the stable genotypes based on multiple traits. In this study, IIHRCB 22-1-1 obtained the lowest MTSI value indicating it as the most stable genotype among all the traits. It was supported by Table 1 as well as these genotypes were classified as a fairly stable and better performers for all the traits. The selection of IIHRCB 22-1-1 would improve the traits number of clusters per plant (2.68%), pods per plant (22.0%), pods per cluster (18.2%), yield per hectare (35.0%) and single pod weight (10.5%) over the ideal ideotype as they have higher selection differentials. Previously, Zuffo et al. (Reference Zuffo, Steiner, Aguilera, Teodoro, Teodoro and Busch2020) also used MTSI to select stable soya bean genotypes under drought and salinity situations and Singamsetti et al. (Reference Singamsetti, Shahi, Zaidi, Seetharam, Vinayan, Kumar, Singla, Shikha and Madankar2021) for selection of maize hybrids in different soil moisture regimes. All the previous studies stated that the MTSI is a useful tool for simultaneously selecting stable genotypes based on multiple traits.
Conclusion
Stable performance of a genotype across environment/location/season is always the major criteria for promoting it as a variety/hybrid. However, estimation of this stability requires crucial analysis of GEI along with genotypic performance per se. Further, superiority in multiple traits across environment provides an added advantage to the grower. Hence, in this study using the AMMI model-based WAAS and MTSI tools, we identified cluster bean genotype IIHRCB 22-1-1 (Fig. 4) as the most promising and stable vegetable-type genotype for both Summer and Kharif cultivation in sub-tropical regions of India. Other important plant descriptor for this selected line includes, medium-tall, single stemmed plant with dark green, smooth, trifoliate leaves; pink flowers in racemose inflorescence develops into string-less, smooth, meaty, long pods (12–14 cm) borne in cluster at each node. Crop duration is 90–100 days.
Figure 4. Cluster bean line IIHRCB 22-1-1.
Supplementary material
The supplementary material for this article can be found at https://doi.org/10.1017/S1479262124000169
Acknowledgements
The authors are indebted to Director, ICAR-IIHR for extending all necessary facilities and funding by granting permission to carry out the research work.
