Evaluation of soybean seed shape by elliptic Fourier descriptors.

Round seeds are desirable for food-type soybean [Glycine max (L.) Merr.], and the genetic control of seed geometry is of scientific interest. The aim of this research was to: (1) compare the seed shape variation of 165 soybean cultivars and to quantify the characteristics of the cultivars, and, (2) to evaluate the seed shape of the parental F1 and F2 populations derived from Tamahomare × Peking and Peking × Tamahomare, and to estimate the heritability of the shape factors. For experiment 1, 10 seeds of each cultivar were photographed from the hilum (H) and lateral (L) directions, and the shape from each direction was independently analyzed. For experiment 2, 30 seeds from each of P1 and P2, 100 seeds from F1, and 10 seeds from each of 239 F2 lines were analyzed in the same manner as for experiment 1. The images of the seeds captured were then analysed based on EF-PCA (Elliptic Fourier-Principal Component Analysis) by using the shape analysis software package, SHAPE (Iwata et al., 2003). The EF-PCA analysis describes seed shape mathematically by transforming seed contour coordinates into elliptic Fourier descriptors, and subsequently summarises these descriptors by principal component analysis to evaluate shape with comparatively lower number of principal components. We could separate the shape variation into symmetric variation and asymmetric variation against the center axis of the soybean seed. PC1 (1st principal component) and PC2 of the symmetric variation of the seed shape observed from the H direction indicated the aspect ratio and roundness, respectively, while PC1 of the symmetric variation observed from the L direction indicated the aspect ratio. For the symmetric variation, the contribution ratio of PC1 was over 90%, showing that most of the seed shape variation could be explained by the aspect ratio. The broad sense heritabilities were 0.894, 0.836, and 0.405, for PC1-H, PC2-H, and PC3-H, respectively, and 0.892, 0.839, and 0.877, for PC1-L, PC2-L, and PC3-L, respectively.