Principal Component and Cluster Analysis in Mungbean [Vigna radiata (L.) Wilczek]

Mungbean [Vigna radiata (L.) Wilczek] is also known as green gram, gold gram, moong. It is a diploid (2n=2x=22), self-pollinating grain legume belonging to Fabaceae family. Mungbean is a warm-season crop that provides a good opportunity for crop diversification due to its ability to grow quickly, allowing adaptation to multiple cropping systems and rotations of crops. In the economic year 2023, India produced over 3 million metric tonnes. (www.statista.com). Despite numerous fluctuations, the trend has been upward. Currently, India leads the production and it is followed by China and Myanmar (Nair et al., 2014). Mungbean is an excellent source of high-quality protein, fiber, minerals andvitamins.Mungbean’s drought tolerance, low input requirements andshort growth cycle make it a worthwhile crop for cultivation, especially in areas with environmental stress. Furthermore, it is more adaptable and digestible than other pulses (Parida and Das, 2005).
       
There are significant constraints hampering the development of the mungbeanprogramme. For example, the limited genetic diversity that hinders mungbean productivity. Needless to say, there is a need to broaden the genetic base of the released varieties (Win et al., 2020). Another, yield, a complex quantitative trait, is the result of multiple morphological and physiological traits are always influenced by environmental factors (Gupta et al., 2023). Therefore, to develop new varieties of mung bean, an assessment of phenotypic diversity or characterization of morphological and agronomical characters plays an important role in the selection of desirable parents in the genetic improvement or plant breeding programme.
       
Multivariate techniques are an efficient tool for interpreting the genetic variation present in the germplasm, selecting it anddevelop the strategies to incorporate useful diversity in their breeding programmes. Several multivariate techniques are used to estimate genetic diversity, including metroglyph analysis, D² statistics, cluster analysis, principal component analysis, principal coordinate analysis, canonical analysis, factor analysis andcorrespondence analysis. Amon them,multivariate, cluster andprincipal component analyses are commonly used to estimate quantitative trait variability and identify superior genotypes. (Jeberson et al., 2017).Therefore, the present investigation to assess the genetic diversity using principal component analysis, on the basis of agronomical characters, classify the germplasm in the similar group and identify the superior genotypes which could be used as parents in breeding program.

The research investigation was conducted at the Agricultural Experimental Farm,(AEM), Division of Genetics and Plant Breeding, RKMVERI, Narendrapur, West Bengal, over a span of two years, from 2021 and  2022 in summer season. The AEM, Narendrapur lies at 22°43'N latitude and 88°40'E longitude belongs to New Alluvial Zone at an elevation of 8m above mean sea level.
       
A total 30 diverse germplasm obtained from the ICAR-Indian Institute of Pulse Research, Kanpur which are presented in Table1. The experiment was laid out in the randomized complete block design (RCBD) with three replications. The size of unit plot was 3 m×2.2 m. The distances of 45 cm between rows, 15 cm within the plants, 75 cm between plots and 1 m between replications. The experimental field was prepared; genotypes were sown and followed various cultural operations for better growth of seedlings. The thinning was followed by 12 days after sowing and left one plant on a hill. Five randomly selected plants were harvested per plot when colour of almost all the pods become black or brown. Pods of each plant were kept separately in paper bags and sun dried. Threshing was done by hand and strict care was taken to avoid mechanical mixture of seeds.
 

       
11 yield attributing characters were recorded and presented in Table 2. The recorded data were analyzed for descriptive statistics (mean, maximum, minimum, Standard error of difference, coefficient of variation). Pearson’s simple correlation coefficient was studied among yield attributing characters. The traits were analyzed as multivariate by using procedures of principal component analysis (PCA) and cluster analysis (CA) with the help of R software version 4.3.2. PCA was carried out using a correlation matrix to reveal the relationship between quantitative traits that are correlated by converting them into uncorrelated traits known as PCs (Johnson and Wichern, 1988). PCA provided information about the relative importance of each trait for screening germplasm in this research. The other multivariate analysis, most widely used clustering technique, D² statistics were employed suggested by Mahalanobis (1936) and Rao (1952).
 

The simple descriptive statistics including mean value, maximum and minimum value and coefficient of variation (CV) of the 11 morphological characters studiedpresented in Table 3. The statistical analysis showed that wide range of variation indicated in the germplasm for the characters. High variation was recorded in DTF(36.50-53.16), DTM(63.50-71.83), PH (43.12-76.06 cm), NPB1.480-4.34), BPP(5.47-9.91),NCP(5.37-10.65), NPP(16.20-42.18), LOP(6.76-8.72 cm), NSP(9.33-14.11), HSW(2.55-4.50 g) andSYP.(4.32-13.76 g). Among genotypes, VIRAT, SIKHA, SAMRAT and IPDI-539 showed highest SYP and NPP. Early flowering was recorded inSIKHA, IPM2K-14-9 and MH2-15. TMB 96-2 had consistently good performer for NPB, BPP, NCP and TSW. While, EC5200-34 had maximum NSP and LOP, it was followed by Suketi-1. CV is the parameter to measure the genetic variability for the characters. High CV (>10%) was observed for SYP and it was followed by NCP represent the substantial amount of genetic variability and with the accordance of the findings of Gupta et al., (2023). Moderate CV (5-10%) was observed for NPB and it was followed by NPP, NSP and HSW. Low variability was found in DTF and it was followed by BPP and PH. Similar findings were corroborated by Sarkar and Kundagrami (2016) and Kaur et al., (2017).
 

       
The correlation coefficient among all yield attributing characters isanalyzed by Pearson’s Correlation are represented in Fig 1. The dark blue and red cell indicates highest and lowest significant positive and negative correlationcoefficient respectively. The highest significant positive correlation had revealed between SYP and NPP, it was followed in between DTF and NPB, betweenSYPand NPB also with HSW, between NPB and NCP, between DTF and DTM and between PH and NPP respectively. These characters could be considered for the future breeding programme in mungbean. On the other side, DTF had significant negative correlation coefficient between NPP and also exhibited with SYP. Any successful plant breeding program involves careful selection of suitable parents. Parents with greater genetic distance are likely to produce more variation. The genotypes with wide variation provide more scope for selection (Canci and Toker, 2014). According to Bos and Caligari (1995), more genetic variation in traits leads to greater genetic gain. Yimram et al., (2009) reported that significant variation in mungbean correlated with growth, phenology, yield components and grain yield.
 

       
Cluster analysis was conducted to explain the genetic relationships among various genotypes and to identify suitable parents for the breeding programs. The importance of genetic diversity in parental genotypes is emphasized for the enhancement of breeding initiatives. Utilizing a hierarchical classification approach with Ward linkage clustering method based on 11 yield attributing traits, 30 germplasm were categorized into six distinct and clearly defined clusters in Table 4. Among these, the highest genotypes namely, nine had in cluster 1 followed by cluster 5 contributed six genotypes, cluster 2, 4 and 6 represented four genotypes each and the cluster 3 had only three genotypes.Genotypes within the same cluster are indicative of a closer genetic relationship compared to those belonging to different clusters. Fig 2 showed a dendrogram that the distribution of genotypes across clusters revealed significant genetic variability. The average intra and inter-cluster D² values were estimated among 30 Germplasm (Table 5). The least intracluster distance was observed in cluster I (2.625) it indicates minimum difference among the germplasm occupied in this cluster. The maximum intra cluster D² value was observed in cluster VI (4.985) followed by cluster III (3.883), cluster II (3.381), cluster IV (3.192) and cluster V (2.912) indicating that maximum differences occur among the genotypes that fall in these clusters. Furthermore, the maximum intercluster distance was between cluster III and cluster VI (7.263) followed by cluster V and cluster VI (6.750), cluster I and cluster VI (6.239) and cluster IV and cluster VI (6.087). It indicates that genotypes lied in these clusters are genetically diverse. According to Falconer (1964), the greater divergence between parental genotypes parallels to increased heterosis in crosses. It is advantageous to undertake crosses between genotypes originating from distant clusters. Therefore, it would be desirable to select a diverse array of segregants to produce highly heterotic crosses, which could be suitable for subsequent selective breeding. Therefore, it would be desirable to select genotypes present in these diverse clusters for the breeding programme. Improvement in crop yield and its associated traits constitutes the fundamental goal of any breeding program. Therefore, the evaluation of cluster diversity related to seed yield and its contributing attributes is essential for the purpose of genotype selection. (Gupta et al., 2023). In the present experiment, significant variations were observed among the clusters for the majority of the traits.
 

 

 

       
The mean values of seed yield and its components in various clusters are shown in Table 6. Cluster I revealed the low to moderate range of mean values for most of the characters. DTF (44.05), NPB (2.21), NCP (6.28), NPP (21.02), LOP (7.53 cm), NSP (10.63), HSW (3.19 g) and SYP (6.31g) were recorded minimum. Cluster II had the lowest mean values for BPP (6.48) and highest mean values for LOP. (8.01 cm). Cluster III had lowest mean values for most of the characters. This cluster possessed the accessions showed less DTF, least mean values for PH (49.49 cm), NPB (1.91) and HSW (3.01 g). It had highest mean values for NSP (12.60). Cluster IV showedlowest mean values for DTM (64.95), NCP (5.82) and LOP (7.17 cm). Cluster V showed highest mean values for DTF (47.93), DTM (70.88) and lowest mean values for NPP (20.11) and NSP (10.53). Among the clusters, the Cluster VI had highest mean values for most of the characters like PH (66.61 cm), NPB (3.70), BPP (8.68), NCP (8.22), NPP (33.88), HSW (4.12 g) and SYP (11.93 g). From the above comparison, cluster II, IV and VI had better cluster mean for SYP and its attributing characters. This will be very useful for future plant breeding programme and improvement of a new variety. Fig 3 revealed the percentage of contribution of each character towards the total diversity. The minimum contribution (<5%) was recorded in NSP (1.4%), NCP (2.5%), DTM (3.2%) and LOP (3.9%). The maximum contribution for the total divergence was recorded from BPP (25.7%), SYP (14.5%), PH (13.8%) and NPP (12%). Similar results were the findings of Ajay et al., (2012), Gokulakrishnan et al., (2012) and Jadhav et al., (2023).
 

 

       
In multivariate statistics, principal components analysis (PCA) possesses the capability to convert a set of potentially correlated variables into a reduced set of uncorrelated variables referred to as principal components (PC). Eigenvalues are commonly employed to ascertain the number of factors to retain. If the eigenvalue is less than one, it indicates that the explanatory efficacy of the principal components is inferior to the average explanatory efficacy of the original variables (Jadhav et al., 2023). PCA, only the pertinent principal components were extracted from the 11 characters. The identification of key contributors to variance was facilitated by examining the characters with high loading values on PC₁. Eigen values greater than one can serve as an inclusion criterion. Principal components with eigen values less than unity wereconsidered non-significant. In the present investigation, out of 11 principal components first three PC i.e. PC₁ to PC₃ which eigen value more than one extracted from the data contributed 67.44% from the total variation (Table 7 and Fig 4). PC₁ showed 37.05% followed by PC₂ (17.18%), PC₃ (13.21%).The eigenvalue and variance associated with each principal component gradually decreased, while cumulative variability increased gradually. Table 7 shows the principal components results along with each character loading score. Scree plot graph depicted in Fig 4 explained the percentage of variance associated with each PCs obtained by drawing a graph between percentage of variances and principal component numbers. In PC₁, the high positive component loading from SYP(0.450), NPP (0.426), NPB (0.422), HSW (0.396) and NCP (0.333). Whereas, high negative loading components was observed in PC₁ with DTF and DTM. The high positive loading components signified the effectiveness of the selection process for traits associated with yield on PC₁. This observation illuminates that PC₁ accounted for the predominant variability in traits linked to yield. Similarly, in second principal component, DTM (0.526), DTF(0.380), BPP (0.288), NCP(0.217) and PH(0.183) were contributed major variation. High negative loading components were LOP (-0.473) and NSP (-0.408). Furthermore, the important characters, DTF (0.542), NSP (0.431), LOP (0.383), DTM (0.362) and NCP (0.247) were contributed more variation in PC₃.The characters like BPP (-0.375) and it was followed by NPP(-0.117) and SYP (-0.039) showed negative loading. Principal component analysis proves advantageous for breeders in preparing the targeted breeding programs based on informed insights into the specific groups where particular traits hold greater significance. From the above loading scores, high positive loading components and the highest cluster mean for the various traits were found common and get confirmation their diversity possesses by the germplasm.
 

 

       
The biplot (Fig 5) and the loading plot (Fig 6) showed the SYP, NPB, NPP, HSW and DTM are far from the origin and had higher loading and great influence on the variation. Therefore, the loading plot reflecting the contributions of the characters to PC₁ and PC₂. Genotypes that were proximate and overlapped on the biplot exhibited similar properties, while those that were distant and remote from the origin demonstrated genetic variation.Genotypes from divergent clusters like TMB 96-2, SAMRAT, SIKHA, VIRAT, SUKETI-1, PDM 04-123,EC5200-34 and COGC-912 were scattered far apart on the plot.
 



       
The phenotypic expression of individual genotypes was explained through the principal component (PC) scores, as delineated in the Table 8. On the basis of PC scores, it is possible to suggest exact selection indices, the strength of which can be determined by the variability each principal component can explain. According to Singh and Chaudhary (1977), high values for variables within a specific genotype are represented by a correspondingly high PC score attributed to that genotype within the associated component. The highest PC scores of positive values >1.5 in each PC could be used them as selection indices.
 

       
The genotypes such as AKM 96-2, IPDI-539, PRATIKSHA NEPAL, SUKETI-1, TMB 96-2, SAMRAT, SIKHA and VIRAT had high yielders with high PC scores. These genotypes were also good performer for the other associated yield traits. The characters with high variability areemphasized by PC analysis. Therefore, rigorous selection can be designed to quickly increase yield.

From the experiment, a significant amount of genetic diversity was revealed and it was highlighted in clusters using hierarchical cluster analysis. The descriptive analysis of yield and its attributing traits had pointed out some important characters like BPP, PH, NPP and NPB as the important characters to identify and classify the diversity from the germplasm. ThroughPCA, it was helpful to identify traits and genotypes from similarities and differences among the genotypes. According to the finding of the research the identified diverse genotypes from the clusters like AKM 96-2, IPDI-539, PRATIKSHA NEPAL, SUKETI-1, TMB 96-2, SAMRAT, SIKHA and VIRAT were superior in yield and associated traits. Therefore, the information of this result will be very useful with these characters for future mungbean breeding program.

All the authors acknowledge the Indian Institute of Pulses research (ICAR-IIPR), Kanpur for providing the germplasm for research.

All authors declare that they have no conflict of interest.