Principal component analysis (PCA)
The PCA made it possible to differentiate the 20 samples of G.
lutea on the basis of their content of secondary metabolites. The
2D graphical representation of principal-component analysis is
shown in Fig. 4, and represents 95.4% of the total variance in the
data set. The variability of data was generated mostly by the
content of gentiopicroside (values of eigenvectors: 0.66; 0.04)
in the first PC and by loganic acid (values of eigenvectors: 0.09;
0.29) in the second PC. Samples on the lower left hand side of
the PCA score plot (Fig. 4A) came from wild populations (apart
from samples W1 and W2) which are characterised by high levels
of gentiopicroside (Fig. 4B). Samples on the top left hand side
(Fig. 4A) came from experimental cultivations, which were characterised
by high levels of loganic acid and, to a lesser extent, by gentiopicroside
(Fig. 4B). Finally, samples on the middle right hand
side of the PCA score plot (Fig. 4A) were purchased commercially,
and correlated to the other secondary metabolites such as amarogentin,
swertiamarin, sweroside and isogentisin. In conclusion, the
variability found in the data matrix seems to be correlated to the
different origin of the samples. The high content of gentiopicroside
appears to be characteristic of wild-type roots; high levels of
loganic acid and gentiopicroside characterised cultivated samples.
Finally the commercial samples seem to be characterised by lower
levels of loganic acid and gentiopicroside.
4