tive fold induction or suppression of genes in treated
samples versus untreated controls and selecting the
most consistently different signals across the sample
set. A different signature may be established for each
prototypic toxicant class. Once the signatures are determined,
gene-expression profiles induced by unknown
agents in these same model systems can then
be compared with the established signatures. A match
assigns a putative mechanism of action to the test
compound. Figure 2 illustrates this signature method
for different types of oxidant stressors, PAHs, and
peroxisome proliferators. In this example, the unknown
compound in question had a gene-expression
profile similar to that of the oxidant stressors in
the database. We anticipate that this general method
will also reveal cross talk between different pathways
induced by a single agent (e.g., reveal that a compound
has both PAH-like and oxidant-like properties).
In the future, it may be necessary to distinguish
very subtle differences between compounds within
a very large sample set (e.g., thousands of highly similar
structural isomers in a combinatorial chemistry
library or peptide library). To generate these highly
refined signatures, standard statistical clustering techniques
or principal-component analysis can be used.
For the studies outlined in Figure 2, we developed
the custom cDNA microarray chip ToxChip v1.0.