A group of scientists affiliated with Harvard University and Harvard Medical School has published a method to perform high-throughput cell-based drug profiling in cells that may provide a blueprint for future academic drug-discovery efforts.
The research also supports the power of high-content analysis — as opposed to high-throughput screening — in drug discovery, particularly in determining toxicity and mechanism of action.
The study, which appears in the Nov. 12 issue of Science, “suggests that large sets of unbiased measurements might serve as high-dimensional cytological profiles analogous to transcriptional profiles,” the paper notes.
As stated in the article, “the method [is] based on hypothesis-free molecular cytology that provides multidimensional single-cell phenotypic information yet is simple and inexpensive enough to allow extensive dose-response profiles for many drugs.”
To be sure, academic laboratories are becoming increasingly involved in drug discovery. These labs generally do not intend to, nor do they have the resources to compete with drug-discovery firms or pharmaceutical companies. While such private endeavors can invest huge amounts of money in drug discovery, they also ultimately must keep an eye on the potential payoff, something with which academic researchers need not be concerned.
And although the lack of resources at academic drug-discovery labs can sometimes be a hindrance, it can often spur creative solutions to problems that may not be explored at companies.
The method underscores the idea that these labs, and drug-discovery labs in general, may not always need to invest in “bigger, better, and faster” high-throughput and high-content screening platforms to perform effective drug profiling.
Another advantage that academic labs have is that they can typically draw on the expertise of specialized departments within or between institutions, or draw in collaborators with diverse backgrounds that can contribute to specific parts of the problem.
The Harvard research is a prime example of this philosophy. Steven Altschuler, the corresponding author on the paper, is currently employed at the Bauer Center for Genomics Research, which he joined recently from Rosetta Inpharmatics. He and Bauer colleagues Michael Slack and Lani Wu bring data analysis and mathematics backgrounds to the table.
Lead author Zachary Perlman and co-authors Yan Feng and Timothy Mitchison are affiliated with Harvard Medical School’s Institute of Chemistry and Cell Biology and Department of Systems Biology, and brought to the study expertise in pharmacology, cell biology, and biochemistry.
In the Science paper, the researchers assembled a test set of 100 compounds: 90 were drugs of known mechanism of action; six were blinded alternate concentrations of the known drugs; one was a toxin with multiple biological targets; and three were drugs of unknown action.
The researchers analyzed 13 three-fold dilutions of each drug on cultured HeLa cells that had been fixed and stained for a variety of cell components and processes. They used a standard Nikon fluorescence microscope with automation capabilities to take images of up to approximately 8,000 cells from each well of a 384-well plate.
The key to the study was an analysis method based on a statistical test called the Kolmogorov-Smirnov statistic, which provides a way to determine differences between two data sets without making assumptions about the distribution of data. This allowed the researchers to compare experimental and control distributions from the same plates.
The Harvard approach yielded dose-dependent effects of 100 drugs on cultured cells, which the researchers calculated from approximately 109 data points extracted from more than 600,000 cellular images. According to Slack, the image processing portion took several days, while the actual statistical analysis was completed in about one day.
Comparatively, other high-throughput screening method such as fixed cells, biochemical assays, or gene expression profiling survey hundreds of thousands of compounds over the course of a year by gathering upwards of half a million data points per day. Even live-cell, high-content imaging approaches — which are slower by nature — can generate hundreds of thousands of images per day, from which data points are then extracted.
Approaches such as the one advocated by the Harvard group actually have more in common with high-content screening than high-throughput single-parameter assays. An important point of the research is that it can more readily reveal off-target effects of drugs, as well as subtle dose-dependent responses — hot-button issues in light of recent failures of previously approved drugs.
The researchers provided as an example drugs that were known to have multiple targets, such as histone deacetylase inhibitors and the general kinase inhibitor staurosporine, and were therefore expected to show complex dose-response behavior. “Such phenotypic complexity may help explain why toxicity at high doses is common even for therapeutic drugs that are apparently highly selective at the level of target binding,” the paper states.
The Harvard scientists also claim that their method can group compounds together that have similar mechanisms. This is important because several of the compounds that belonged to one group had poorly characterized mechanisms of action. Therefore, the researchers reasoned, “our methods can … thereby suggest mechanism for new drugs.
“Extensions of cytological profiling to reflect dependencies among descriptors will allow more sophisticated analysis of drug responses at systems levels,” the researchers wrote.
The researchers hope that their work will serve as a launching pad for future academic drug-discovery experiments, and suggested several ways the research can be built upon.
Specifically, they hope to improve and extend the method with “better lab automation, broader drug reference sets, different types of perturbation (such as RNAi), improved strategies for cell segmentation, more sophisticated feature extraction, different sets of antibody probes and cells, the inclusion of more time points and live cell imaging, and the integration of complementary profiling strategies,” the paper states.