Steven J. Altschuler and Lani F. Wu, mathematicians skilled in developing models to find meaningful patterns among mountains of data, worked with Timothy J. Mitchison of Harvard Medical School to automate microscopic imaging of drug-treated cells and recast the resulting scans in a computer-friendly format. The result: a method dubbed "cytological profiling" that trains computers to recognize cell status and health from cellular images, virtually automating microscopic scanning for various types of abnormalities.
"The resulting profiles of cellular changes wrought by drugs at various dosages provide information on drug mechanism that is highly relevant to understanding the specificity and toxicity of drugs," says Altschuler, research fellow at the Bauer Center for Genomics Research in Harvard's Faculty of Arts and Sciences. "The information gleaned includes many key indicators of drugs' potential usefulness and limitations as medicines."
"We actually started out on this project thinking that this could be a good research tool," adds Wu, research fellow at the Bauer Center. "We've now discovered, to our surprise, that it may also prove a powerful tool for drug discovery."
High-throughput cytological profiling lets scientists test numerous variables at once, wringing countless discrete cellular measurements from a single experiment. Faced with scores of drugs holding the potential to combat a given disease, researchers could hone in on the most promising drugs in a fraction of the time of current methods.
"This technique employs 'guilt by association' -- if two drugs' cytological profiles look similar, they probably work through similar mechanisms," Altschuler says. "It's particularly useful for understanding drug action because it allows us to look at many concentrations of a drug, which is essential for comparing two drugs that may have different potency but act on the same target."
Since they are hardy and flourish even outside the body, the Harvard team used human cancer cells. They placed the cells in 384 minuscule wells in a plastic dish, injected each well with one of 100 drugs -- both medicines and toxins -- at different concentrations, and finished off the plates with11 chemical probes for different proteins and DNA.
After 20 hours of cell growth, the researchers used automated microscopy to collect some half a million images of the treated cells, followed by approximately 5 billion individual measurements of the size, shape, and quantity of different proteins, DNA, and organelles in each. Software developed by Altschuler, Wu, and colleagues allowed them to convert this copious data into profiles of the effects of each drug, yielding distinctive red-and-green "fingerprints" for each, not unlike the color-coded data from a DNA chip.
However, unlike DNA chips that meld bountiful data into an "average" denoted by dots of color on a grid, cytological profiling preserves individual data points -- so researchers can go back and analyze fine-grain information.
"By allowing quantitative measurement of many proteins and structures in cells over many samples, and systematic comparison between samples, our method brings microscopy into the '-omics' era, like genomics and proteomics," says Mitchison, of Harvard Medical School's Institute of Chemistry and Cell Biology and Department of Systems Biology. "This really allows us a much broader view of how cells are affected by a wide range of perturbations."
Cytological profiling may be especially useful, Mitchison says, for evaluating drug candidatesin areas where making a drug with a highly specific biochemical effect is difficult, such as kinase inhibitors. Future applications may include testing the response of cancer cells with different genetic profiles to a spectrum of anti-cancer drugs, which could help predict clinical responses in individual patients.
Alschuler, Wu, and Mitchison were joined in this research, sponsored by the National Cancer Institute and Howard Hughes Medical Institute, by co-authors Zachary E. Perlman and Yan Feng at Harvard Medical School and Michael D. Slack in the Bauer Center for Genomics Research.