An automated microscopy method has been developed that can provide quantitative data relating to changes in phenotype of a single cell in response to different drugs. The method, reported in Science, profiles the phenotypic effects of different doses of drugs in cell culture, and was able to both categorize blinded drugs and propose targets for drugs with an unknown mechanism of action. The method provides a complementary approach to existing drug-screening technologies, such as transcript analysis and protein-expression studies.

The study of drug effects on specific cells is often limited to measurements taken at a single dose concentration. Recently, however, the idea of using a combination of imaging techniques to look for particular phenotypes in response to a drug has been proposed. In this study, Perlman and colleagues suggest that using large sets of imaging data could provide a cell profile that is analogous to a gene-expression profile, and they present a method based on measurements of different fluorescent images of cell states that can be used to generate extensive dose–response profiles for many drugs.

Fluorescent probes that represented a range of cell biology (for example, structural proteins, kinases and cell-cycle regulators) were used to study the effects of a test set of 100 compounds, including those with known and unknown mechanisms of action, and a toxin with several known targets. Images were collected of ∼8,000 cells per well using automated microscopy, and a set of descriptors recorded for each cell, region and probe, including size, shape and intensity of fluorescent staining.

The authors devised a method of using these descriptors to generate a dosage-dependent profile for each drug. They found that they could readily distinguish drugs with diverse chemical structures that are reported to act at common targets from drugs that work with a different mechanism. Furthermore, as changes in specificity for a target will show as a change in the cell phenotype, but changes in dose will not, the authors developed a titration-invariant similarity score that can profile drugs regardless of the starting dose used in the experiment. The authors speculate that further extension of this method to reflect dependencies between the descriptors used will enable drug profiles to be analysed at a systems level.