![](http://media.springernature.com/relative-r300-703_m1050/springer-static/image/art%3A10.1038%2Fnrd714/MediaObjects/41573_2002_Article_BFnrd714_Figa_HTML.jpg)
So many toxins to try, so little time to test them. If only one could sample just a little bit of each before deciding which to select for further investigation. For toxicologists, the advent of genomic technologies, such as microarrays for gene-expression analysis, has given rise to the hope that the toxicity of any substance could be rapidly predicted just by examining its influence on the expression of the subset of genes most likely to be affected. In the December issue of Molecular Pharmacology, Chris Bradfield and colleagues show that the number of predictor genes needed for a diagnostic toxicity test might in fact be remarkably small.
Many toxins of regulatory concern fall into one of a limited number of classes, each of which affects a defined set of biological processes. Assuming that most, if not all, toxic substances alter gene-expression levels, one should therefore be able to spot the likelihood of toxicity in an unknown compound by screening for its ability to alter the expression of a diagnostic subset of genes. To assess how large this subset needs to be, the authors used complementary DNA microarray analysis to determine changes in the levels of gene transcripts expressed in the livers of mice exposed to members of five well-characterized toxicological categories. Starting by observing the fluctuations in 1,200 genes, they used an iterative selection algorithm to whittle this number down to just 12 diagnostic gene transcripts, changes in which were able to predict toxicological class to 100% accuracy.
An interesting extension of this finding was that increasing the number of transcripts in the diagnostic set above 12 quickly reduces the predictive accuracy of the test. The suggestion that less is actually more when it comes to transcript profiling will be good news for pharmaceutical companies trying to decide how much data to submit with a new drug application. Focusing on a limited number of target genes will give rise to a much more manageable amount of data, and allow greater reproducibility between different screening facilities.
The present study was designed to reduce, as far as possible, the potential for variability in response. Inbred mice were used to minimize the influence of genetic polymorphisms, and in most cases only single, acute doses of toxin were studied. Although any diagnostic transcript set will need to be verified against more traditional biomarkers, such as blood chemistry, the chances of being able to screen tiny amounts of unknown compounds for toxicity against a small set of diagnostic genes in the near future look good (see the perspective article by Ulrich and Friend on p84 of this issue). Until then, relatively large amounts of compound will continue to be screened by non-genomic methods. For now, better make mine a double.
References
ORIGINAL RESEARCH PAPER
Thomas, R. S. et al. Identification of toxicologically predictive gene sets using cDNA microarrays. Mol. Pharmacol. 60, 1–6 (2001)
FURTHER READING
Ulrich, R. & Friend, S. Toxicogenomics and drug discovery: will new technologies help us produce better drugs? Nature Rev. Drug Discov. 1, 84–88 (2002)
Related links
Rights and permissions
About this article
Cite this article
Smith, A. What's your poison?. Nat Rev Drug Discov 1, 5 (2002). https://doi.org/10.1038/nrd714
Issue Date:
DOI: https://doi.org/10.1038/nrd714