Cucurbit Genetics Cooperative Report 18:10-12 (article 5) 1995
Principal Component Analysis for Traits Selection in Cucumber Breeding
Hongwen Cui, Meng Zhang, and Huanwen Meng; Junjun Deng
Department of Horticulture, Northwestern Agricultural University, Yangling, Shaanxi, 712100, P.R. China; Xi'an Vegetable Research Institute, Shaanzi, 71200, P.R. China
It is not easy to select for numerous traits during crop improvement (3). Principal component analysis (PCA) is a statistical method by which many traits can be scaled into a few comprehensive indices (1). These indices are not subject to trait corre_{8} integrated with quantitative genetics, can increase selection efficiency. Therefore, a study was designed to examine 12 traits using PCA as a tool for enhancing cucumber improvement strategies.
Methods. Ten inbreds with different genetic backgrounds were selected for intercrossing using an incomplete diallel mating design. The F_{1} hybrids were planted in a two-way randomized block design with 3 replications in order to decrease experimental error (2). Ten plants of each cultigen were randomly chosen to evaluate 19 traits during the growth period. Twelve of these traits were selected for analysis based on previous studies. The traits included: 1) total yield per plant (X_{1}); 2) early yield per plant (X_{2}), 3) number of harvested fruits in the early stage (X_{3}); 4) average fruit weight in the early stage (X_{4}); 5) leaf number at first harvest (X_{5}); 6) the node position of the first pistillate flower (X_{6}); 7) leaf number in the last stage (X_{7}); 8) number of effective branches (X_{8}); 9) the days from sowing to pistillate flowering of 50% of the plants (X_{9}); 10) fruit length (X_{10}); 11) downy mildew disease index (X_{11}); and 12) fruit developing average rate (X_{12}).
Results. PCA was conducted using 12 traits by generating a genetic correlation matrix (Tables 1 and 2). The results indicate that the first five components explained 98.6% of the total phenotypic variation of the 12 traits, while the first three components explained 42.5%, 31.0%, and 18.7% of the observed variation, respectively. The vectors indicate the weight of each eigenvector, and do not describe the effect of individual component traits. A factor loading matrix was constructed to more accurately describe the components of each trait using three eigen vectors (Table 3). The first three components were analyzed as follows according to trait properties and the relative importance of components.
The traits X_{2}, X_{3}
, X_{6} , X_{8}, and X_{9} produced large loading values for the first component and allw ere significant. This component array of traits accounted for 84.4% of the total variance of thephenotypic variationl This trait array represents the early maturity and early yield, and therefore this component was designated the "early-maturity component."
The traits X_{1}, X_{5}, X_{7}, X_{10} , and X_{11} produced large loading values for the second component and were also significant in their contribution to the observed phenotypic variance This trait array explained 76.6% of the total phenotype variance. This array described total yield and its composition, and was designated the "yield component."
Traits X_{4}, X_{6} , and X_{12} produced large loading values for the third component which were significant in their contribution to the size and rate of fruit development. These three traits made up 68.9% of the total phenotypic variance for this component which was designated the "Fruit weight component."
Conclusion.The traits which had large loading values ion the first three principal vectoring components should be made selection criteria in cucumber breeding programs which emphasize improvement for early maturing, high yielding (number and weigh of fruit) lines and hybrids. The relative importance of each trait can be characterized by the rank order of their contribution (%) to explaining the observed phenotypic variation.
Table 1. The eigenvalues and percentage of genetic correlation matrix describing the effect of 12 traits in cucumber (Cucumis sativus L.)
Component |
1 |
2 |
3 |
4 |
5 |
Eigenvalue |
5.1 |
3.7 |
2.2 |
0.6 |
0.3 |
Percent (PCI) |
42.5 |
31.0 |
18.7 |
4.7 |
2.7 |
CPCT^{2} |
42.5 |
73.5 |
91.2 |
95.9 |
98.6 |
^{z}Cumulative percentage.
Table 2. The eigenvectors resulting from the principal component analysis incucumber (Cucumis sativus L.)
Traits |
Vector 1 |
Vector 2 |
Vector 3 |
X_{1} |
0.2197 |
0.4218 |
0.1270 |
X_{2} |
-0.4105 |
-0.0007 |
0.2369 |
X_{3} |
-0.4364 |
0.0646 |
0.0287 |
X_{4} |
0.2178 |
-0.1983 |
0.4648 |
X_{5} |
0.1148 |
-0.3620 |
-0.3181 |
X_{6} |
0.4021 |
-0.1792 |
-0.1311 |
X_{7} |
0.2151 |
0.4503 |
0.0786 |
X_{8} |
0.4124 |
0.0758 |
0.0786 |
X_{9} |
0.3909 |
-0.1275 |
0.2331 |
X_{10} |
-0.0008 |
0.3454 |
0.4359 |
X_{11} |
-0.0355 |
-0.4357 |
0.2291 |
X_{12} |
-0.0050 |
-0.2857 |
0.5324 |
Table 3. The factor loading matrix constructed from eigenvalues of 12 traits in cucumber (Cucumis sativus L.)
Traits |
Vector 1 |
Vector 2 |
Vector 3 |
X_{1} |
0.4396 |
0.8133 |
0.1852 |
X_{2} |
-0.9272 |
-0.1406 |
0.3455 |
X_{3} |
-0.9859 |
0.1245 |
0.0418 |
X_{4} |
0.4920 |
-0.3824 |
0.6777 |
X_{5} |
0.2953 |
-0.6978 |
-0.4638 |
X_{6} |
0.9082 |
-0.3455 |
-0.1911 |
X_{7} |
0.4859 |
0.8682 |
0.1146 |
X_{8} |
0.9317 |
0.1461 |
-0.1145 |
X_{9} |
0.8831 |
-0.2458 |
0.3399 |
X_{10} |
0.0018 |
0.6659 |
0.6356 |
X_{11} |
0.0800 |
-0.8400 |
0.3340 |
X_{12} |
-0.0114 |
-0.5509 |
0.7763 |
Note: 1r0.051 = 0.514; 1r_{0.01} 1 = 0.641.
Literature Cited:
- Cui, Jongwen and Uongtao Qi. 1989. The application of principal component analysis to cucumher hreeding. Acta Northwestern Agricultural University. No. 17 (3).
- Ling, Shuhong. 1988. The studies of two-way randomized block design. Acta An'hui Agricultural College. No. 3:65-70.
- Ma,Yuhu. 1984. The base of quantitative genetics in plant breeding. Jingsu Science and Technology Publishing House.