Table 1.
PCA results. (A) Decomposition of the correlation matrix expressed as eigenvalues, difference between adjacent eigenvalues, proportion of variance explained by each component and cumulative variance. (B) Component pattern (loadings). Each loading corresponds to the correlation coefficient between original variables (samples) and corresponding component.
| (A) | ||||
| Component | Eigenvalue | Difference | Proportion | Cumulative |
| 1 | 5.970 | 4.816 | 0.498 | 0.498 |
| 2 | 1.154 | 0.248 | 0.096 | 0.594 |
| 3 | 0.906 | 0.307 | 0.076 | 0.670 |
| 4 | 0.599 | 0.049 | 0.050 | 0.720 |
| (B) | ||||
| Sample | PC1 | PC2 | PC3 | PC4 |
| p136.1 | 0.273 | 0.364 | 0.300 | 0.053 |
| p136.2 | 0.230 | 0.495 | 0.224 | −0.037 |
| p136.3 | 0.275 | 0.364 | 0.275 | 0.068 |
| m136.1 | 0.305 | −0.040 | −0.225 | 0.358 |
| m136.2 | 0.213 | 0.298 | −0.600 | −0.689 |
| m136.3 | 0.246 | 0.268 | −0.462 | 0.531 |
| p229.1 | 0.320 | −0.200 | 0.178 | −0.160 |
| p229.2 | 0.318 | −0.178 | 0.226 | −0.166 |
| p229.3 | 0.316 | −0.159 | 0.201 | −0.140 |
| m229.1 | 0.318 | −0.285 | −0.176 | 0.133 |
| m229.2 | 0.325 | −0.307 | −0.052 | 0.068 |
| m229.3 | 0.296 | −0.235 | −0.064 | −0.113 |
PC1 = ‘tissue attractor’ [30]. This component is by far the most relevant in terms of percent of variance explained (around 50%) and has all positive loadings; thus, it can be considered as a ‘size’ component [31] stemming from the largely invariant tissue-specific average genome expression profile. PC2 = ‘cell line effect’ this component accounts for approximately 10% of total variance; it is a ‘shape’ component [31] having both positive and negative loading correspondent to a directionality of effect having as poles (opposite loadings, see Table 1B) the lsc136 and lsc229 cell lines. This implies the presence of a neat ‘effect line’ (no exception to this loading sign rule, Table 1B) pointing to the presence of genes with a correlated expression with opposite up (down) expression in the two lines independently of the PKH26+/PKH26− phenotype. PC3 = ‘quiescent/replicative effect’ this component accounts for 7.5% of total variance and discriminates PKH26+ and PKH26− (p vs. m) phenotypes in terms of loading sign. Again, this is a shape component pointing to the presence of a clear opposite pattern of gene activation/repression discriminating the two conditions. It is worth noting that principal components are each other orthogonal by construction, thus implying the above effects (tissue attractor, line effect, phenotype effect) do not superimpose. This means that the quiescent/dormant discrimination happens through the same underlying mechanism in both lines.