Multivariate analysis of diversity of landrace rice germplasm

Transcription

1 Multivariate analysis of diversity of landrace rice germplasm K.A. Sanni, 1* I. Fawole, 2 A. Ogunbayo, 1 D. Tia, 1 E.A Somado, 3 K. Futakuchi, 1 M. Sié, 1 F.E. Nwilene 4 and R.G. Guei 5 1 Africa Rice Center (AfricaRice), 01 B.P. 2031, Cotonou, Benin; 2 Bell University of Technology, Ota, Nigeria; 3 African Development Bank, Tunis, Tunisia; 4 AfricaRice Nigeria Station, c/o IITA, P.M.B. 5320, Oyo Road, Ibadan, Nigeria; 5 FAO, Viale delle Terme de Caracalla, Rome 00100, Italy. Abstract Multivariate analysis involves observation and analysis of more than one statistical variable at a time. The variability of 434 accessions of rice (Oryzasativa L.) germplasm from Côte d Ivoire was evaluated for 14 agromorphological traits in upland conditions at M bé, Côte d Ivoire (7 5 N; 5 1 W), using augmented experimental design, and analyzed with multivariate methods. The unweighted variable pair group method of the average linkage cluster analysis (UPGMA), canonical discriminant analysis (CAN) and principal component analysis (PCA) were used to analyze the data obtained. This enabled assessment of the extent and pattern of variation of the germplasm, and identification of the major traits contributing to the diversity. Seven cluster groups were obtained from the 14 agro-botanical traits using UPGMA. CAN showed the contribution of each trait to the classification of the rice accessions into different cluster groups. The first three principal components explained about 58.41% of the total variation among the 14 characters. The results of CAN and PCA suggested that traits such as plant height, leaf length, number of days to heading and maturity, tillering ability, panicle length and grain size (weight, length and width) were the principal discriminatory characteristics. It is concluded that variation exists in the germplasm, which provides opportunities for this collection to be useful for genetic improvement. Introduction Rice has become a commodity of strategic significance and the fastest-growing food source in Africa, such that its availability and price are now a major determinant of the welfare of the poorest people, who are the least food-secure consumers in Africa (Nwanze et al., 2006). It is now grown and consumed in more than 40 African countries, where about 20 million farmers are engaged in its production and about 100 million people depend on rice directly for their livelihood. However, self-sufficiency in African rice production is declining as demand increases, so there is an urgent need to increase and improve the continent s production of rice in order to satisfy the high demand. Striving for African self-sufficiency in rice production not only depends on cultural practices and management, but also on obtaining suitable rice varieties, which could be drawn from existing germplasm collected and conserved by genetic resources centers (Ng et al., 1988). Rice has been grown in Africa for thousands of years; according to Portères (1976), African rice (O. glaberrima Steud.) was first domesticated in the inland delta area of the Niger river in about 1500 BCby the indigenous inhabitants of the area. However, Asian rice was subsequently introduced to West Africa by the Portuguese in about AD1500, since when Asian rice has been replacing O. glaberrima in many parts of the region (Portères, 1950).Some landraces and varieties that have endured through many generations have become adapted to the local environment (through natural and human-assisted processes). However, grain yield potential of landraces is low and unable to supply the population-driven consumption needs of modern times. African rice research organizations recognize the value of traditional landraces, and the array of genetic diversity they hold, as invaluable inputs for the development of better-adapted and higher-yielding varieties. In many cases, the variability is not adequately used in crop improvement programs, because it has not been specifically defined and the genetic potentials are hidden. In order to promote rice germplasm use as well as guidelines for future collections, plant breeders must understand the patterns of variability and grouping of the available germplasm. In the process of conservation, from acquisition to characterization and evaluation, different kinds of data related directly or indirectly to genetic diversity between genotypes or population are produced for exploitation in many ways. Three multivariate methods (Dillon and Goldstein, 1984; Hair et al., 1987) were used to describe and group sets of individual rice accessions, simultaneously taking into account several characteristics and the relationships between them. The study aimed to determine the level of germplasm variation and to identify and classify groups of accessions with different genetic diversity. Materials and methods The germplasm used in this study consisted of 434 landrace accessions of O. sativa collected from three locations Danane, Touba, and Boundiali in Côte d Ivoire. Danane is located in a high-rainfall zone near the Liberian border. Touba is in a forest savannah transition zone near the border with the Republic of Guinea, * Corresponding author (k.sanni@cgiar.org). Second Africa Rice Congress, Bamako, Mali, March 2010: Innovation and Partnerships to Realize Africa s Rice Potential 1.1.1

2 while Boundiali is further north in the savannah. There were 211 landrace accessions collected from Danane, 61 from Touba, and 162 from Boundiali. For the purposes of easy identification and retrieval, accession numbers composed of three letters (to represent the site), plus a serial number were assigned to the collections. Thus, landraces collected from Danane bore the prefix DNN, while those from Touba and Boundiali bore the prefixes TBA and BDL, respectively. The trial was conducted in the upland field of Africa Rice Center (AfricaRice) at M bé, Côte d Ivoire (7 5 N; 5 1 W). Because of the limited quantity of seeds, an augmented design (Federer, 1956) with eight checks was used. The eight checks were: WAB56-104, Morobérékan, IDSA6, IDSA10, NERICA 2, NERICA 1, WAB638-1, and WAB Each plot consisted of four rows, each 1 m 5 m long, with a distance of 40 cm between plots. The accessions were direct-seeded at a plant spacing of cm, with two seedlings per hill. A basal dose of NPK ( ) was applied at the rate of 200 kg/ha during land preparation. Urea was applied at the rate of 65 kg/ha as top-dressing, once at tillering and again at booting. Hand weeding was done regularly to minimize weed infestation. Bird and rodent damage were controlled using bird scarers and fencing, respectively. Fourteen quantitative characters were measured according to methods in the descriptors for rice O. sativa (IRRI, 1980). Variables considered in the descriptive and multivariate analyses were morphological (plant height, number of tillers, number of fertile tillers, leaf length, leaf width, and culm diameter at maturity); phenological (days to 50% flowering and days to harvest maturity, both measured from the day of sowing); and grain traits (panicle length, grain length, grain width, grain shape, and weight of 100 grains). Statistical analysis Data were statistically analyzed to identify groups among the landraces that were genetically different or similar based on the 14 traits measured. Canonical discriminant analyses (CAN) were used to determine the relative contribution and linear associations among the traits. Two methods, namely the unweighted pair group method of average linkage (UPGMA) and principal component (PCA), were used to classify the 434 accessions into groups (clusters). UPGMA was performed using the Numerical Taxonomic and Multivariate Analysis System statistical software (NTSYS, version 2; Rohlf, 1990), while CAN and PCA were done using SAS/PC version 8 (SAS Institute, 1999). Results and discussion Pearson s coefficient Pearson s correlation coefficient (r) is a measure of the strength of linear association between two variables. The coefficients of correlation for pairs of characters in the landrace collection are presented in Table 1. Among the variables studied, the highest correlation corresponded to the number of days to heading (DH) and days to maturity (Mat) (r = 0.98). Traits presenting highly significant correlations with these two characteristics were tiller fertility ratio (Ftr, r = 0.67 and 0.68, respectively), leaf width (LW, r = 0.52 and 0.54, respectively) and 100-grain weight (Gwgt, r = and 0.308, respectively). The negative correlations show that Ftr, LW and Gwgt tended to decrease as the DH and Mat increased. The positive correlations between the grain weight and grain width (r = 0.59) and grain length (r = 0.56) indicate that the wider and/or longer the grain, the heavier it is.. The highly positive correlation between total number of tillers and fertile tillers (r = 0.87) showed that the fertile tillers tend to increase along with the total number of tillers. Principal component analysis The result of the PCA explained the genetic diversity of the rice collection. Proper values measure the importance and contribution of each component to total variance, whereas each coefficient of proper vectors indicates the degree of contribution of every original variable with which each principal component is associated. The higher the coefficients, regardless of the direction (positive or negative), the more effective they will be in discriminating between accessions. There are no standard tests to prove significance of proper values and the coefficients. In this study, we chose to follow the criterion used by Clifford and Stephenson (1975) and corroborated by Guei et al. (2005), which suggested that the first three principal components are often the most important in reflecting the variation patterns among accessions, and the characters associated with these are more useful in differentiating accessions. According to this criterion, the first three components account for more than 58% of total variation (Table 2), giving a clear idea of the structure underlying the variables analyzed. However, the criterion of Raji (2002) was chosen to determine the cutoff limit for the coefficients of the proper vectors; this criterion treated coefficients greater than 0.3 as having a large enough effect to be considered important, while traits having a coefficient less than 0.3 were considered not to have important effects on the overall variation observed in the present study. The first principal component accounted for more than 24% of total variance, whereby number of days to heading and days to maturity were the variables that contributed most negatively. In contrast, the variables Second Africa Rice Congress, Bamako, Mali, March 2010: Innovation and Partnerships to Realize Africa s Rice Potential

3 Table 1. Correlation matrix of 14 characters used in characterizing the rice collection Ht LL LW DH Mat Tt Ft Ftr Cdia PL GL GW GS LL 0.383** LW ns DH 0.337** ** Mat 0.316** ** 0.979** Tt ** Ft ** Ftr ** 0.668** 0.679** 0.308** Cdia 0.319** PL 0.321** GL GW GS ** 0.746** Gwgt ** 0.308** ** 0.593** Cdia, culm diameter; DH, number of days from planting to 50% heading; Ft, number of fertile tillers; Ftr, ratio of Ft/Tt; GL, grain length; GS, grain shape; GW, grain width; Gwgt, 100-grain weight; Ht, plant height; LL, leaf length; LW, leaf width; Mat, number of days from planting to maturity; PL, panicle length; Tt, total number of tillers. Second Africa Rice Congress, Bamako, Mali, March 2010: Innovation and Partnerships to Realize Africa s Rice Potential 1.1.3

4 Table 2. Coefficients and vectors associated with the first three principal components First Second Third Eigenvalue % variance Cumulative % total variance Coefficient vector Plant height (cm) Leaf length (cm) Leaf width (cm) % heading (days) % maturity (days) Total tillers (no.) Fertile tillers (no.) Fertile tiller ratio Culm diameter (mm) Panicle length (cm) Grain length (mm) Grain width (mm) Grain shape (ratio) Weight of 100 grains (g) contributing most positively were yield components with respect to grain width and weight of 100 grains (Table 2). As a result, the first component differentiated those accessions that head and mature earlier in the season and also registered those with high yield-component values. The first component identified mainly phenological variables presenting negative contributions. Guei et al. (2006) observed similar performance for phenological variables of rice. The second principal component accounted for more than 20% of total variance. Variables highly and positively correlated were grain length and weight of 100 grains. The second component thus identified yield component variables presenting positive contributions and the main characters responsible for classification. The third principal component accounted for 13% and was associated with panicle and stem diameter, plant height, grain diameter, and weight of 100 grains, thus differentiating those accessions with good architecture, plant height, thick stems, large panicles, and medium to large grains (Table 2). These findings agree with Caldo et al. (1996), who report that maturity, heading, plant height, culm length, leaf length, and tillering ability were the major factors contributing to the variation of parental lines of modern Philippine rice cultivars. Takeda (1990) also reports that grain size as indicated by weight, volume, or length is one of the most important agronomic traits in rice. Cluster analysis Analyses performed by UPGMA divided the 434 landraces into seven clusters. The landraces were not evenly distributed among the clusters,and their places of origin were distributed across the clusters. For example, there were 220 landraces in Cluster 1 and 100 in Cluster 2. These two clusters contained 73% of the landraces in the study. Members of the Cluster 1 comprised 133 from Danane, 52 from Boundiali, and 35 from Touba. Cluster 2 consisted of 71 landraces from Boundiali, 22 from Touba, and 7 from Danane (Table 3). Means of the agrobotanical traits that characterize individual clusters are presented (Fig. 1A D). Clustering based on phenology and morphological characteristics The landraces in Clusters 4, 5 and 6 were late maturing. The mean duration to 50% flowering for each group was 154 days for cluster 4, 174 for cluster 5 and 160 days for cluster (Fig. 1A). The accessions in Cluster 2 had a mean duration to flowering of about 84 days, and were earliest among all the landraces. Mean plant height per group varied little, and was between 125 and 145 cm across the groups, except for members of Cluster 3, which had mean height of <100 cm (Fig. 1A). The number of tillers produced per plant was highest (7 8 tillers per plant) for the landraces in Clusters 2, 4 and 5 (Fig. 1B). But, while Clusters 4 and 5 had only about four fertile tillers, there were seven fertile tillers in Cluster 2. Total number of tillers was lowest in Cluster 7. The accessions in this group had a mean of four tillers per plant. Culm diameter was similar across all clusters at cm (Fig. 1B). Leaves were longest and widest among landraces in Cluster 1 (Fig. 1C). Members of this group also had the longest mean panicles (25 cm). Variability in mean panicle length was 5 cm, and Cluster 7 had the shortest panicles (mean = 20 cm) Second Africa Rice Congress, Bamako, Mali, March 2010: Innovation and Partnerships to Realize Africa s Rice Potential

5 Value Cluster A Plant height(cm) 50% flow ering (days) Maturity (days) Value Cluster B Tiller number (/plant) Fertile tillers (/plant) Culm diameter (s core) Value Cluster C Leaf length (cm) Leaf w idth (x 20cm) Panicle length (cm) Value Cluster D Grain weight (g/grain) Grain length (/4mm) Grain width (mm) Grain shape (score) Figure 1. Characteristic means of seven similarity clusters generated by UPGMA based on the classification of 434 landraces of O. sativa, using different agro-botanical characteristics. Number of members were 220, 100, 2, 74, 35, 1 and 2, sequentially, in Clusters 1 through 7. For the purpose of scaling, data for leaf width are multiplied by 20, while data for grain length were divided by 4. All other data are means for cluster groups. Clustering based on grain characteristics Classifications based on grain characteristics are shown in Figure 1D, and Cluster 3 stands out with respect to these characters. The mean grain weight for the landraces in Cluster 3 was 4 g/grain, whereas in Clusters 5, 6, and 7, grain weight was less than 2 g/grain. Although Cluster 6 had the lowest mean grain weight, the landrace had a mean grain length of 9.6 cm, which was the longest. Application in breeding objectives The overall objectives are to aid utilization of the landraces in rice improvement work. Therefore, Table 3 shows the number of varieties from each location that were grouped in each cluster. Table 4 summarizes the value of each of the clusters in identifying breeding stocks based on the traits studied. No single cluster may be suitable for all traits. For example, tall plant types were found in Cluster 1, whereas high tillering landraces could be found in Cluster 2.Early maturing plants occur in Cluster 2. The landrace in Cluster 6 is characterized by thick culm, long grains, but narrow leaf, infertile tillers, and lightweight grains. Table 3. The spread of 434 Oryza sativa landraces across seven clusters. The landraces were collections from Boundiali, Danane, and Touba in Côte d Ivoire. Cluster analysis was based on 14 agro-botanical characteristics Collection site Code Cluster 1 Cluster 2 Cluster 3 Cluster 4 Cluster 5 Cluster 6 Cluster 7 Total Boundiali BDL Danane DNN Touba TBA Total Canonical discriminant analysis Canonical discriminant analysis simultaneously examines the differences in the morphological variables and indicates the relative contribution of each variable to accession discrimination (Vaylay and van Santen, 2002). Second Africa Rice Congress, Bamako, Mali, March 2010: Innovation and Partnerships to Realize Africa s Rice Potential 1.1.5

6 Table 4. Descriptive guide to assist in the use (in breeding programs) of 434 Oryza landraces with respect to listed agro-botanical characteristics Cluster N Plant height Leaf length Leaf width (cm) Culm diameter No. Fertile Panicle length Grain length Grain width (cm) (cm) (mm) tillers(/plant tillers (cm) (mm) (mm) Grain weight (g/gain) ) (/plant) Tall Short Long Short Wide Narrow Wide Narrow High Low High Low Long Short Long Short Wide Narrow Heavy Light Early late Cluster Cluster Cluster 3 2 Cluster 4 74 Cluster 5 35 Cluster 6 1 Cluster 7 2 Maturity (days) Second Africa Rice Congress, Bamako, Mali, March 2010: Innovation and Partnerships to Realize Africa s Rice Potential

7 Quantitative variables were considered as independent, and the clusters identified by cluster analysis as dependent variables. The first six discriminatory functions were statistically significant according to chi-square test at a probability of Proper values and the distribution of their variances indicated that the first two functions accounted for more than 89% of total variance. Wilks lambda coefficients for these two functions were the lowest, indicating an almost perfect discrimination regarding the remaining functions. In contrast, the R 2 indicates the percentage of variance explained by the model (Table 5). The significant (P<0.001) canonical correlation (Rc) between the accessions and the first canonical variate (Rc = 0.91) and second canonical variate (Rc = 0.62) indicates that the canonical variates can explain the differentiation of the accessions (Table 6). Table 5. Discriminatory functions that distinguish between clusters of landrace rice accessions Function Proper Variance (%) R 2 coefficient Wilks ג χ 2 df P value Proportion Cumulative < < < < < Table 6. Canonical discriminant coefficients (CAN) showing the contribution of each character used in the classification of the rice collection into different cluster groups Character CAN1 CAN2 Ht LL LW DH Mat Tt Ft Ftr Cdia PL GL GW GS Gwgt Canonical correlation (Rc) P level of significance Variance accounted for (%) Cumulative variance accounted for (%) For explanation of abbreviations, see Table 1. Canonical loadings measure the simple linear correlation between an original independent variable (trait) and the canonical variate. This implies that the canonical loading reflects the variance that the observed variable shares with the canonical variate, and can be interpreted in assessing the relative contribution of each variable to each canonical function (Hair et al., 1987). The first canonical discriminant function is dominated by a large loading from number of days to 50% heading and maturity, followed by grain width (Table 6). The second function is dominated by a large loading from panicle length, followed by plant height, number of tillers, leaf length, grain length, and grain width. It is therefore evident in the canonical discrimination that the composition of the accessions differs chiefly in number of days to 50% heading, maturity, grain characters, number of tillers, plant height, leaf length, and panicle length. The result of the canonical discriminant analysis confirmed the results of both the PCA and cluster analysis. These findings are similar to the observations by Sanni et al. (2008). Conclusion The rice landraces studied showed considerable variability for most of the characteristics that could be exploited for crop improvement. Cluster analysis based on plant morphology suggested that the accessions could be Second Africa Rice Congress, Bamako, Mali, March 2010: Innovation and Partnerships to Realize Africa s Rice Potential 1.1.7

8 grouped. Such groupings are useful to breeders in identifying possible landraces that may be used as parents in breeding for any of the morphological traits that were studied. Above all, the information generated will reduce the overall time required by plant breeders to screen large populations for potential breeding stock. References Caldo RA, Sebastian LS and Hernandez JE Morphology-based genetic diversity analysis of ancestral lines of rice in Philippine rice cultivars. Philippines Journal of Crop Science 21(3): Clifford HT and Stephenson W An Introduction to Numerical Classification. Academic Press, London. 229 p. Dillon WR and Goldstein M Multivariate Analysis: Methods and applications. John Wiley and Sons, New York. Federer WT Augmented designs. Hawaiian Planter s Rec. 55: Guei RG, Sanni KA, Abamu FJ and Fawole I Genetic diversity of rice (Oryza sativa L.). Agronomie Africaine 5: Hair JF Jr, Anderson RE and Tatham RL Multivariate Data Analysis (2nd edn). Macmillan, New York. IRRI Minimum List of Descriptors and Descriptor-states for Rice Oryza sativa L. Manual. International Rice Research Institute, Los Baños, Philippines. 20 p. Ng NQ, Chang TT, Vaughan DA and Zuno-Alto Veros C Africa rice diversity: Conservation and prospect for crop improvement. p In: Crop Genetic Resources of Africa II. Nwanze KF, Mohapatra S, Kormawa P, Keya S and Bruce-Oliver S Perspective: Rice development in sub-saharan Africa. Journal of the Science of Food and Agriculture 86: Portères R Vieilles agricultures de l Afrique intertropicale. L Agronomie Tropicale 5: Portères R p In: Harlan JR, de Wet JMJ and Stemler ABL eds. The Origin of African Plant Domestication. Mouton, The Hague. Raji AA Assessment of genetic diversity and heterotic relationships in African improved and local cassava (Manihot esculenta Crantz) germplasm. PhD thesis. University of Ibadan, Nigeria. Rohlf FJ NTSYS-pc. Numerical taxonomy and multivariate analysis system. Applied Biostatistics, New York. Sanni KA, Fawole I, Guei RG, Ojo DK, Somado EA, Sanchez I, Ogunbayo SA and Tia DD Geographical patterns of phenotypic diversity in Oryza sativa landraces of Côte d Ivoire. Euphytica 160: SAS Institute SAS/STAT Guide for Personal Computers, Version 8 Edition. SAS Institute, Cary, NC p. Takeda K Inheritance of grain size and its implication for rice breeding. p In: Rice Genetics II. Proceedings of the second International Rice Genetics Symposium, May IRRI, Manila. Vaylay R and van Santen E Application of canonical discriminant analysis for the assessment of genetic variation in tall fescue. Crop Science 42: Second Africa Rice Congress, Bamako, Mali, March 2010: Innovation and Partnerships to Realize Africa s Rice Potential