Television shows about medical examiners and psychological profilers are all the rage these days. Wouldn't you like to see one about “cytological” profilers? No, not Alois Alzheimer in a period drama, peering at plaques and tangles through a compound microscope, but twenty-first-century pharmacologists reading dose-response curves and colorful "heat" plots on their computer monitors.
Well…maybe it wouldn't be a hit show on television. But pharmacology detectives may end up glued to their computer monitors, thanks to a new methodology that allows high-throughput, microscopy-based profiling of drug effects in cells. The techniques, described in the 12 November issue of Science by Steven Altschuler, Lani Wu, and colleagues at Harvard University, may even have an unexpected bonus—the ability to identify novel therapeutic value in drugs both familiar and unfamiliar.
Microarray analysis of genes and proteins has garnered a great deal of attention in recent years (see ARF related news story). However, as first author Zachary Perlman and colleagues allude to, for all its speed and volumes of data, cellular phenotyping with microarrays cannot describe cell biology in the same way as microscopy. But microscopy as traditionally practiced, nose to microscope, is slow work. In their article, Perlman and colleagues combine an assembly line approach to treating and photographing cells in culture, with algorithms for crunching multivariate data on different drugs, dosages, and cellular responses.
To test the technique, the authors cultured human cancer cells in 384-well plates and treated them with one of 91 different drugs—13 different concentrations made by threefold dilutions. After 20 hours they fixed the cells and probed them with fluorescent antibodies for DNA and a sampling of 10 proteins from across the spectrum of cell biology (DNA + 2 protein probes per well). As many as 8,000 cells were photographed per well, and these images were analyzed by computers assigned the task of measuring different descriptors for each labeled cell. These descriptors, based on the DNA and antibody staining, included such characteristics as size/shape of nucleus or cytoplasmic annulus and ratio of nucleus to cytoplasm, as well as information about the levels and distribution of target proteins, based on the intensity or characteristics (e.g., speckling) of antibody staining. The researchers reduced this mountain range of data to useful images such as dose-response curves for each drug, probe, and descriptor, as well as a series of compound profiles for each drug—"heat maps" that indicate the magnitude of change in different descriptors by the intensity of color.
Perlman and colleagues also developed a titration-invariant similarity score (TISS) for each drug, which allowed them to compare compounds independent of the concentration ranges being used. The idea here is that the technique can recognize dose-dependent trends irrespective of the dose. In other words, if two compounds that bind to the same biological site, a hormone receptor, for example, are tested, one in the micromolar range and one in the nanomolar range, they should have nearly identical profiles. The TISS was successful in grouping compounds of similar biological mechanism. The hope for this particular aspect of the methodology is that it might reveal previously unrecognized biological mechanisms of old drugs, or help identify uses (or dangers) for new compounds.
There are numerous opportunities for improving and expanding on this methodology, according to Perlman and colleagues, and they conclude that, "This analysis, extended to work in tissues or clinical samples, offers the potential to speed the identification of toxic compounds during therapeutic drug development and the targeting of drug effects to specific subtypes of cells."—Hakon Heimer
No Available Further Reading
- Perlman ZE, Slack MD, Feng Y, Mitchison TJ, Wu LF, Altschuler SJ. Multidimensional drug profiling by automated microscopy. Science. 2004 Nov 12;306(5699):1194-8. PubMed.