Clustering and principal component analysis of the samples used for transcript profiling. The analysis was carried out with 911 cytokinin-regulated genes that were selected as described in Methods. (a-e) Analysis using the native dataset. (f-j) Analysis using the dataset after normalization for the organ effect. (a, f) Support tree clustering for the samples. (b-e, g-h) Results of the principal component analyses. (b, g) Three-dimensional representation of the results of the principal component analyses. (c-e, h-j) Two-dimensional plots showing each possible combination of the three dimensions. (c, h) Plots centred on the x-axis; (d, i) plots centred on the y-axis; (e, j) plots centred on the z-axis of (b) and (g), respectively. Each point represents an experimental condition, averaging two microarray replicates of two biological replicates. BA0, control; BA15, 15 min of cytokinin treatment; BA120, 2 h of cytokinin treatment; BA1080, 18 h of cytokinin treatment; CKX1, cytokinin oxidase/dehydrogenase overexpressors. Green, shoot samples; ochre, root samples. The purple markings represent the plane on which the wild-type samples are arranged. The dashed lines in (j) show the factors of global gene expression changes in roots and shoots, respectively. The eigenvalues, which indicate how much of the total variation of the dataset is covered by the respective axis are for b-e: x, 0.490; y, 0.184; z, 0.112; and for g-h: x, 0.269; y, 0.239; z, 0.179. A PCA with a more restricted dataset containing 637 genes (i.e. all genes detected on at least 50% of the arrays) yielded a similar result (data not shown).