Multivariate Diversity Analysis of Quantitative Traits of Mung Bean [Vigna radiata (L.) Wilczek] Genotypes

Mung bean is an important pulse crop grown by poor farmers in marginal and drought-prone areas of Ethiopia. Information on the extent of genetic divergence in mung bean is vital to identify diverse genotypes for crop improvement and the ecient utilization of the existing genetic resources. Therefore, the objectives of the study were to assess the extent and pattern of morphological diversity among the mung bean genotypes and to identify the traits contributing to the genetic diversity using multivariate analyses. The experiment was conducted using 60 mung bean genotypes at Jinka Agricultural Research Center during the 2018 cropping season. The rst seven principal components explained 80.1% of the total variation. Almost all the studied traits were important contributors to the divergence. The cluster analysis based on quantitative traits revealed four distinct groups. The highest inter-cluster distance was recorded between cluster I and cluster IV (D 2 = 43.16 units). The minimum inter-cluster distance was noted between cluster III and cluster IV (D 2 = 12.16 units). The maximum and minimum intra-cluster distances D 2 were recorded within cluster I (D 2 = 6.49 units) and cluster III (D 2 = 3.53 units), respectively). The range of intra and inter-cluster distance was 3.53 to 6.49 units and 12.16 to 43.16 units, respectively. Hence, the high genetic distance exhibited within and among clusters has to be exploited through the crossing and selection of the most divergent parents for future mung bean breeding programs.


Introduction
Mung bean [Vigna radiata (L.) Wilczek] is an important short-season grain legume, well suited to smallholder production under adverse climatic conditions (Vijayalaksmi et al., 2003). It is a cheap source of dietary protein with high levels of iron compared to many other legumes (Nair et al., 2012).
Morphological traits are routinely used for estimating genetic diversity and to analyze genetic relationships among the genotypes (Pandey, 2007). Basnet et al. (2014) reported that evaluation of germplasm is useful not only in the selection of core collection but also important for breeding programs. Nair et al. (2013) suggested that future progress in mung bean breeding requires imperative action to identify mung bean genotypes with favorable agronomic traits for further improvement programs. Therefore, broad genetic diversity is needed to achieve mung bean breeding goals. Sarkar and Kundagrami (2016) indicated that mung bean genotypes exhibited a wide range of variation for most of the studied traits. Lestari et al. (2014) reported that the collection of diverse local cultivars and their subsequent genotyping would enhance germplasm diversity and provide information, both of which are bene cial for developing collection strategies and breeding purposes with desirable agro-morphological characteristics. Lavanya et al. (2008) indicated that the multivariate techniques may be an e cient tool in the quantitative estimation of genetic variation to select germplasm more systemically and effectively and to develop strategies to incorporate useful diversity in their breeding programs.
Quanti cation of the magnitude of the genetic diversity and classifying it into homogeneous groups to facilitate identi cation of genetic variability enable breeders to select traits of interest for the mung bean improvement program. The analysis of genetic diversity in germplasm collections can also facilitate the classi cation and identi cation of groups of accessions with superior characteristics to be used for breeding purposes (Mohammadi and Prasanna, 2003;Tosti and Negri, 2005;Bhandari et al., 2017).
Hence; characterization of genetic diversity needs comprehensive statistical procedures, such as D 2 statistics and multivariate analysis, namely principal component analysis and cluster analysis that have been considered as the best tools for choosing promising genotypes for a future breeding program in mung bean. Therefore; these techniques are used for clustering a large number of genotypes into homogenous groups, the genetic distance among and within clusters, and principal component analysis for identifying the most important contributing traits for the genetic diversity and identifying superior cowpea genotypes (Mohammadi and Prasanna, 2003;Oliveira et al., 2003;Aremu et al., 2007;Doumbia et al., 2013;Abe et al., 2015;Molosiwa et al., 2016;Udens et al., 2016). Furthermore, limited studies have been reported on mung bean globally (Ghafoor et al., 2001;Rahim et al., 2008;Pandiyan et al., 2012;Nainu et al., 2020) and Yirga et al. (2020) in Ethiopia. However, most of these studies were conducted using a very limited number of genotypes on fewer quantitative traits. Moreover, there is no su cient information on the genetic diversity of mung bean genotypes by using multivariate analyses. Therefore, the objectives of the present study were to assess the extent of genetic diversity among 60 mung bean genotypes using phenotypic traits, to cluster the genotypes into homogenous groups, estimate the genetic distance and relationships between genotypes, and identify the major traits contributing to the diversity using multivariate techniques.

Descriptions of the Study Area
The eld experiment was conducted from March to June 2018 at Jinka Agricultural Research Center (JARC) during the main cropping season. Jinka Agricultural Research Center is located 729 km southwest of Addis Ababa at 36 0 33' 02.7" E, 05 0 46' 52.0" N, and at an altitude of 1420 m above sea level. The maximum, minimum, and average temperatures of the center for ten years (2009-2019) are 27.68 0 C, 16.61 0 C, and 22.14 o C, respectively; while the mean annual rainfall is 1381 mm. The soil type of the center is Cambisols (Mes n et al., 2017).

Experimental Materials
A total of 60 mung bean genotypes were used for this study. Of these, 44 were obtained from Melkassa Agricultural Research Center (MARC), and 16 genotypes were collected from Southern Nations, Nationalities, and People's (SNNP) region.

Experimental Design and Procedures
The experiment was laid out using a 6 × 10 alpha lattice design. The plot size was 3 m long, 0.3 m between rows, and 0.05 m between plants. It consists of ve rows accommodating 60 plants per row. The distance between plots, intra blocks, and replications was 1, 1.5, and 2 m, respectively.

Data Collection
The descriptor of mung bean developed by the International Board for Plant Genetic Resources (IBPGR, 1980) was followed for data collection. The data collected on a plot basis include days to owering (days), days to maturity (days). The data collected on the plant basis from ten plants were: plant height (cm), number of primary branches per plant, number of pods per plant, number of seeds per pod, pod length (cm), peduncle length (cm), number of pods per cluster, terminal lea et length (cm) and terminal lea et width (cm). The data were collected from the central three rows for the determination of seed yield, hundred seed weight (g), seed yield per plot (g), biomass (g), and harvest index (%).

Data Analyses
The means of seventeen traits were used and standardized to a mean of zero and a unit variance to avoid biases due to differences in the scales of measurements (Sneath and Sokal 1973). A series of multivariate analyses such as Minitab Statistical Software (Minitab, 2010) and SAS software (SAS, 2008) were used. The principal component analysis was performed using the correlation matrix to determine principal components, proportions of eigenvalues, and the scores of the principal components.
Hierarchical (Ward, 1963) cluster analysis was performed to group genotypes and construct a dendrogram by using Minitab software. The Euclidean distance was measured for the dissimilarity of the genotypes. The average intra and inter-cluster distances were calculated using the generalized Mahalanobis's D 2 statistics (Mahalanobis, 1936). The R 2 (RSQ), Cubic Clustering Criteria (CCC), pseudo-F statistics (PSF), and pseudo-T 2 statistics were considered for de ning optimum cluster numbers (Milligan and Cooper, 1985). The contribution of each trait to divergence as described by Sharma (1998)

Principal Component Analysis
After identifying the variation of morphological traits then that should be further evaluated at a genetic level (Rabhani et al., 1998). The patterns of variation and the relative importance of each of the 17 quantitative traits in explaining the observed variability were assessed through principal component analysis ( Table 1). The PCs having the eigenvalues of greater than unity are signi cantly contributed to the variation explained for the speci c traits under study, indicates that these PCs contributed an additional increment of the speci c trait at least by one unit. Chat eld and Collins (1980) suggested that principal components with eigenvalues less than one should be eliminated from PCA as they are not signi cant. Jager et al. (1983) suggested that the principal component was computed from the correlation matrix and genotype scores obtained from the rst components (which has the property of accounting for maximum variance) and succeeding components with latent roots greater than the unity. According to the principal component analysis, seven principal components with the eigenvalues more than unity were considered and accounted for about 80.1% of the total variance among genotypes (Table 1), thus the seven principal component scores might be used to review the original seventeen variables in any further analysis. Similarly, Belul et al. (2014) reported that regarding the acceptable level of proportion in variation if a proportion of variation higher than 75% of the total variation is considered for characterization and evaluation of genetic collections for all legumes. Correspondingly, Nainu et al.
(2020) reported that 78% variation was explained by the rst ve PCs for eleven quantitative traits on eighty-one mung bean genotypes. Also, Pandiyan et al. (2012) indicated that 63.79% variation was justi ed by the rst three principal components for 18 quantitative characters of six hundred and forty-six mung bean accessions. Likewise, Ghafoor et al. (2001) took the rst four principal components with eigenvalues > 1, contributed 78.7% of the total variance amongst forty mung bean genotypes. Similarly, Nainu et al. (2020) used the rst three principal components with eigenvalues greater than 1 contributed 78% of the total variance amongst eighty-one mung bean genotypes.
The relative contributions of the PCs are explained by individual PCs, but the highest share was explained by the rst principal component PC1 (22.4%), which contributed a high proportion of total variation and the rest six PCs had the proportion of 57.7% of the total variance, respectively. The rst principal component (PC1) with eigenvalue 4.246 accounted high proportion of the total variance (22.4%) and the remaining three principal components viz., PC2, PC3, and PC4 with eigenvalue 2.861, 2.468, and 1.926 recorded 15.1, 13.0, and 10.10% of the total variance, respectively, while the other PCs (PC 5 to PC 7) had weak or no discriminatory power (Table 1). These results are in agreement with Mahalingam et al. (2020) who reported that the rst PC with the eigenvalue of 2.6 accounted for 32.60% of the total variation and the remaining three PCs with the eigenvalue of 1.7, 1.1, and 1.0 recorded 20.4, 11.3 and 10.2 %, respectively and accounted for 53.3% of the total variance and had the most discriminatory power while the rest four PCs had weak discriminatory power on seventy-four mung bean genotypes, similarly reported by Jeberson (2017) on 74 mung bean genotypes.
The present study depicted that all the studied traits have contributed signi cantly to the variations towards the PCs through their loading effects. Thus, the most important descriptors were those associated with the rst seven principal components. The criterion of Raji (2002) was chosen to determine the cut-off limit for the coe cients of the proper vectors. According to this criterion, traits with coe cient values greater than 0.3 were considered as those traits having a large effect and to be considered important while traits having a coe cient value lesser than 0.3 were considered not to have important effects on the overall variation observed in the present study. As suggested by Hair et al. (2009) who observed that the loading effect of any traits greater than 0.3 was regarded as meaningful.
In this investigation, seed yield per hectare (0.451) and harvest index (0.464) in PC1 had high loading effects (Table 1)  The extent of variance and relationship among different traits as explained by the loading plot ( Fig. 1) displayed the magnitude of the relationship between the quantitative traits. The genotypes allotted in quadrant-I were similar by their terminal leaf length, terminal leaf width, peduncle length, pods per plant, and harvest index. These traits had a relatively strong association and have a positive contribution to the discrimination of genotypes. In the second quadrant, genotypes were strongly associated with a single trait namely, the number of seeds per pod. Genotypes assigned in the third quadrant were strongly associated with the number of primary branches per plant, biomass yield, and days to maturity.
Genotypes found in the fourth quadrant were similar for seed yield per hectare, plant height, seed yield per plant, pod length, pods per cluster, and hundred seed weight (Fig. 1). Moreover, the scattered plot showed that the genotypes were distributed more or less equally in all quadrants. The scatter plot consisting of four quadrants illustrated the distribution of mung bean genotypes according to the diversity of their morphological characteristics. These could be used by breeders to identify the presence of genetic variability within the tested genotypes and select the donor parents for speci c traits. Genotypes overlapped in the two principal axes have a similar phenotypic expression of the traits.
The loading plot indicates the similarity and differences among the studied traits. In the biplot, the traits found near to the origin such as the number of primary branches per plant, pods per cluster, and terminal leaf length have smaller loading effects and also had little in uence. Traits like pods per plant, peduncle length, days to maturity, pods per plant, hundred seed weight, and harvest index are far from the origin and had higher loading and great in uence in this classi cation. Among the studied traits, days to maturity, seed yield per plot, and harvest index have higher loading effects, indicating that these traits had great in uence. Therefore, the loading plot re ecting the contributions of the characters to PC1 and PC2. Furthermore, genotypes classi ed using quantitative traits were explained by the rst two dimensions PC1 and PC2 (Fig. 2). Genotypes closer to each other and overlapping on the loading plot had similar characteristics whereas; those genotypes far apart from each other and those found distantly far from the origin are genetically diverse.
Most of the genotypes were concentrated in the second, third, and fourth quadrant and few genotypes in the rst quadrant. The genotypes G7, G9, G60, and G36 were realized to have diverged greatly from the other genotypes.

Genetic Divergence Analysis
The genetic distances among the genotypes based on the studied 17 quantitative traits were estimated using D 2 . The result revealed that the D 2 between the clusters was highly signi cant (P ≤ 0.01), suggesting high diversity among genotypes. Therefore; based on the D 2 results, the studied 60 genotypes were grouped into four clusters with a variable number of genotypes, revealing the presence of a considerable amount of genetic diversity.

Clustering of genotypes
The analysis of variance was highly signi cant among the divergent genotypes for all the 17 traits under study, which revealed the presence of considerable variability among the genotypes. This nding is in agreement with the report of Sen and De (2017) who observed highly signi cant differences among the mung bean genotypes for all the studied 13 traits and thus indicated the presence of substantial amounts of diversity among the genotypes. This suggested that adequate opportunity is available for the selection of superior genotypes aimed at enhancing the genetic yield potential of mung bean. Based on this, 60 mung bean genotypes were grouped into four groups based on the similarity of the studied traits ( Fig. 4.3). Similarly, Popoola et al. (2017) reported that the 26 accessions of the Vigna vexillata were segregated into three clusters, cluster 1 consists of 15 accessions subdivided into two groups, cluster 2 is comprised of 10 accessions while cluster 3 has one accession. Musalamah et al. (2006) classi ed 75 mung bean genotypes into 11 major groups. Similarly, Yimram et al. (2009) classi ed 344 mung bean accessions into 5 major and 1 minor cluster. Similarly, Sen and De (2017) reported that the 30 mung bean genotypes were grouped into six clusters. While Sandeep et al. (2014) grouped fty cowpeas genotypes into twelve clusters.
Understanding the level of genetic diversity in germplasm is helpful to plant breeders as it supports their decision on the selection of parental genotypes and is important in widening the genetic base of cowpea breeding (Prasanthi et al., 2012). The pattern of distribution of genotypes among various clusters re ected the presence of considerable genetic variability among the genotypes under study. The genotypes belonging to the same cluster are designated to be more closely related than those belonging to different clusters. Monogenotypic clusters indicated that such genotypes might have completely different genetic makeup from the remaining genotypes and each other, thus leading to the formation of the separate cluster. A maximum number of genotypes were comprised of cluster I (20), followed by cluster II (17) and cluster IV (14). The minimum genotypes (9) comprised Cluster III (Fig. 3). A dendrogram summarizing the homogeneities between mung bean genotypes based on 18 traits is depicted in Fig. 3. The tested genotypes were grouped into different clusters with various numbers of genotypes. The number of genotypes per cluster varied from 9 to 20 genotypes into each cluster. Cluster I was the largest cluster comprising 20 genotypes, followed by Cluster II and Cluster IV contained 17 and 14 genotypes, respectively, while the rest cluster III contained 9 genotypes.

Cluster mean performance
As improvement in seed yield and other related characters is a basic objective in any breeding program, thus cluster means need to be considered for the selection of genotypes. The mean values of 17 quantitative traits per cluster are presented in Table 2. In this study, the mean values varied among clusters for all traits. Genotypes that took shorter days to ower and to attain maturity were found in Cluster-I, while those genotypes took extended time for owering and maturity found in Cluster-II, Cluster-  I and Cluster IV while the  minimum was recorded on Cluster II and Cluster III. The highest biomass yields were exhibited on Cluster   III and Cluster IV while the minimum was noted on Cluster I and Cluster II. Generally, Cluster IV showed the best performance for most of the traits in general and showed the highest mean performance for pod length, plant height, pods per cluster, pods per plant, hundred seed weight, seed yield per hectare, biomass yield, and harvest index. Therefore, cluster IV would be preferable for the selection of parents with high mean values and more effective for the improvements of genotypes. On the contrary, Cluster II and Cluster III had minimum values for yield and yield-related traits; this showed the poorest performance of traits.
In general, there was a highly signi cant variation in mean performance among clusters in most traits. Therefore, this is a good opportunity to select potential parents across the cluster for speci c traits for future mung bean improvement. In general, the variation observed among 60 mung bean genotypes for the quantitative traits suggests that the presence of inherent genetic diversity among mung bean genotypes. Similar results have also been reported by various authors (Bhist et al., 1998;Rahim et al., 2008;Tarika et al., 2009;Abna et al., 2012;Pandiyan et al., 2012;Kingsly et al., 2015).
In this study, the levels of trait contribution for inter-cluster divergence studies for mung bean genotypes showed that CTIC (%) of ≥ 25% as a high contributor, ≥ 10% < 25% as a medium contributor, and < 10% as a little contributor for inter-cluster divergence. On the contrary, Tesfaye et al. (2019) reported that the levels of trait contribution for inter-cluster divergence for cowpea ≥ 15% as a high contributor, ≥ 8% < 15% as a medium contributor, and < 8% as a little contributor for inter-cluster divergence in Ethiopia. In this study, as presented in Table 2, the CTIC (%) for seed yield per hectare (38.50%), seeds per pod (31.25%), pods per plant (30.15%), and harvest index (29.07%), indicating that these traits were the major contributors for a genetic divergence to the entire genotypes. The CTIC (%) for days to owering (1.87%), days to maturity (1.19%), and terminal leaf width (7.9%), signifying that these traits were slight contributions to the genetic divergence. Generally, the variation observed among the 60 mung bean genotypes indicated that the studied agronomic traits might reveal diversity among mung bean genotypes. This nding is supported by the previous reports of ( Molosiwa et al., 2016;Moolendra et al., 2018) on cowpea genotypes.

Intra and inter-cluster distance
The values for squared distances (D 2 ) indicated that there were signi cant differences observed between clusters I with Clusters III and IV (P ≤ 0.01), while no signi cant variations were observed among Cluster I and Cluster II, Cluster II with Cluster III, Cluster II with Cluster IV, Cluster III with Cluster IV. The range of intra and inter-cluster distance was 3.53 to 6.49 units and 12.16 to 43.16 units, respectively (Table 3). The highest average inter-cluster D 2 was recorded between cluster I and Cluster IV (D 2 = 43.16 units) followed by Cluster I and Cluster III (D 2 = 31.63 units), suggesting the highest genetic divergence existing between the genotypes of these clusters and expected to give a higher frequency of better transgressive segregants or desirable combinations for development of useful genetic stocks or varieties. Therefore, these two clusters were more genetically divergent from each other. As per the inter-cluster distance (from Cluster I, III, and IV), selection of parents for hybridization program among genotypes from diverse Clusters would achieve novel recombinants that increase e ciency for improvement of seed yield in mung bean. The nearest inter-cluster distance was found between clusters III and IV (12.16 units) followed by Clusters II and III (13.24 units). Those genotypes that showed non-signi cant variations among the clusters did not have a diverse gene pool and they have a narrow genetic base. Thus, crossing parents from these clusters might not give higher heterotic value in the future hybrid breeding program in the subsequent generations and will not get a wide range of variability in the segregating population.
The maximum intra-cluster distance (D 2 ) was recorded in Cluster II (6.49 units) followed by cluster III (5.70 units) and cluster IV (4.63 units), respectively. The present study indicates that the genotypes in clusters II and III were more divergent than any other clusters. Thus, the genotypes belonging to the distant clusters could be used for the mung bean breeding program to get a wider range of variability. Also, genotypes from these two distinct Clusters II and III could be utilized as a parent for a hybrid breeding program or recombinant breeding program owing to their wider within-group distance. While the lowest D 2 was recorded in Cluster IV (3.53 units), which showed the presence of less genetic variability or diversity within these clusters. In general, the intra-cluster distance was much less than the inter-cluster  2011) marked that such intra-and inter-cluster distances might arise due to the differential genetic makeup of the genotypes. The intra-cluster distance values indicate the closeness of the genotypes falling in the same cluster. The clusters exhibiting an intra-cluster distance of 0.00 reveal to be monogenotypic and consequently less heterogeneous; on the other hand, high intra-cluster D 2 values indicate a more genetic divergence between genotypes belonging to the same cluster and therefore more heterogeneous. According to Murthy and Arunachalam (1966) success of the hybridization followed by selection depends largely on the choice of parents showing high genetic diversity for traits of interest. Therefore, such intra-cluster heterogeneity among the genotypes constituents' obtained in the present experiment might serve as a guideline to choose parents for the recombination breeding program. It indicated that these cluster pairs were most divergent or in other words, the genotypic constituent of these cluster pairs comprised the genes from most distantly related parents in respect of the characters studied. The genotypes belonging to different clusters separated by high estimated statistical distance may be used in the hybridization program for crop improvement as well as for studying the inheritance pattern of different characters in mung bean. The above results further revealed that considering individual characters the genotypes were more divergent than those considering a constellation of characters.

Conclusions
Morphological characterization of mung bean genotypes based on quantitative traits showed that the existence of polymorphism on mung bean genotypes. Plant height, the number of primary branches, the number of clusters per plant, the number of pods per cluster, and the number of pods per plant are the major yield contributing characters and hence during selection, attention should be given to these characters for the development of high yielding mung bean genotypes. Pod length was the most important trait that positively contributed to principal component analysis to detect phenotypic diversity. The rst seven principal component analyses detected the major contributing traits accounted for 80.1% of total variations and are informative enough to discriminate against 60 mung bean genotypes. The total genotypes were clustered into four distinct groups and the highest inter-cluster D 2 was recorded between cluster I and cluster IV followed by cluster I and cluster III, indicating that the inter-cluster distances were more genetically divergent from each other. The range of intra and inter-cluster distances was 3.53 to 6.49 units and 12.16 to 43.16 units, respectively. Hence, the high genetic distance exhibited within and among clusters has to be exploited through the crossing and selection of the most divergent parents for future mung bean breeding programs.
Intra cluster distance was much lesser than inter-cluster distance showing the presence of high genetic divergence among the clusters. Seed yield per hectare, pods per plant, seeds per pod, and harvest index were the most discriminate traits for grouping the entire genotype into different clusters. In general, the present investigation indicated that high genetic distances exhibited within and among clusters are an opportunity to exploit through breeding via crossing and selection of the most divergent parents from the clusters.
Furthermore, the adoption of biochemical and molecular approaches might show enough genetic diversity among mung bean genotypes for further variety development.