Biology Reference
In-Depth Information
line was cultured in the presence or absence of a drug at fi ve time
points, and this experiment was repeated four independent times
resulting in a 20-sample, 10-gel DIGE experiment. By all conven-
tional methods, the experiment appears to have worked very nicely
as illustrated by high-resolution separations (see Fig. 4a ) and the
presence of some statistically signifi cant univariate changes (see
Fig. 4b ). But PCA on the unfi ltered dataset of 1,012 features
matched across all ten gels clearly indicates that the principal com-
ponents are organizing the samples based on cell passage number
and not by treatment (see Fig. 4c ).
Thus, a simple PCA on the unfi ltered dataset can quickly deter-
mine if something went drastically wrong with the experiment and
in this case saved the investigator considerable time and effort to
follow up potential proteins of interest when clearly the major sources
of variation were unrelated to the biology being manipulated. These
fi ndings raised the likelihood that the few changes “found” by univari-
ate ANOVA arose stochastically and not from the drug treatment.
Fig. 4. PCA showing unanticipated high background noise. Primary mammalian cells were subjected to a drug treatment
during four successive passages in tissue culture. Five time points were taken in each independent experiment, resulting
in 20 samples that were coresolved across ten DIGE gels along with a Cy2-labeled internal standard. ( a ) Representative
DIGE gel from the 10-gel set. ( b ) A protein feature ( white arrow in panel A , identifi ed by mass spectrometry as Hsp27) was
identifi ed as signifi cantly changing (ANOVA p < 0.05). ( c ) PCA performed on the unfi ltered 1,012 features matched across
all ten gels clearly organized the samples into three clusters defi ned by PC1 and PC2, with each cluster containing each of
the 5 time points rather than the anticipated clustering by time point. Closer inspection of the samples revealed that the
samples in each cluster came from the same cell passage number in the experiment. These results indicated that far more
variation was derived from changes occurring in the primary cells during tissue culture, and this experiment was aban-
doned due to the tremendous risk that any observed univariate change (such as was found for Hsp27) simply arose
stochastically rather than due to biologically signifi cant reasons.
Search WWH ::




Custom Search