Systems Biology:
Method Automates Capture of Cell Image Data
Medical Education
HMS Hospitals Hone Plans for Deeper Cuts in Resident Hours
Clinical Research:
Earliest Cancer Trials May Offer Lower Risk But Lower Benefit
Ambulatory Care:
Postmarketing Study Answers Questions on Statin Safety
Sodium Channel Modulates Calcium-Based T Cell Activation
Drug Ads Need Plainer Language to Explain Risks
Proceedings of the HMS Faculty Council
HMS Appoints Connors as Board of Fellows Chair
Lynn Eckhert Takes Over as AAMC Chair
Dean's Community Service Awards
Broad and Novartis Announce Joint Program to Decode Genetics of Type 2 Diabetes
Center for Large-scale SNP Analysis Backed at Broad Institute
Judge Baker Appoints New President, Opens New Facility
Richmond Award Honors Antismoking Activists
NIH Roadmap Supports Training in Genetics and Complex Diseases
Lefkopoulou Lecturer Describes Approach to Incomplete Data in Longitudinal Studies
Children's Wins $2.5m in Health Surveillance Grants
Cultural Misunderstandings Can Be Opportunities for Learning
Medical Simulator Goes to the Head of the Class
Method Automates Capture of Cell Image Data
May Support Drug Discovery Through Documenting Drug Action on Whole CellsBy computerizing the capture and analysis of nearly half a million snapshots of single cells, researchers from HMS and Harvard University have generated the most comprehensive pictures to date of how drugs and toxins affect the health of human cells in culture.
To transform pictures into numbers, the researchers trained computers to look at photographs of cells labeled with fluorescent probes, left, and pick out nuclei, outlined in the middle panel. Green nuclei are mitotic cells; red nuclei are nonmitotic cells. Thousands of automated measurements of cell shape, size, and the abundance and localization of specific proteins combine to generate a complex fingerprint, right, of drug action. (Image courtesy of Zach Perlman)
The work represents a collaboration of researchers at the Bauer Center for Genomics Research in the Faculty of Arts and Sciences, the HMS-based Institute of Chemistry and Cell Biology (ICCB), and the HMS Department of Systems Biology. This new type of drug profiling, based on automated microscopy and image analysis, can reveal information about the biological mechanism and toxicities of new drugs and eventually may be used to speed discovery efforts.
"This work illustrates the potential of microscopy to go beyond a description of what something looks like to being a detailed quantitation of what's going on in the cell," said Timothy Mitchison, the Hasib Sabbagh professor of systems biology, codirector of the ICCB, and head of the effort to automate the collection of thousands of cell images.
"The quantitative measurement of many proteins in many cells, and systematic comparison between samples, brings microscopy into the 'omics' era," added Steven Altschuler, who along with Lani Wu and Michael Slack at the Bauer Center developed the software that made sense of the pile of cell pictures. The findings appear in the Nov. 12 Science.
From Images to Information
To test the idea that whole cell responses could be profiled in a high-throughput assay, Mitchison and ICCB fellow Yan Feng first selected 100 chemical compounds known to affect different biochemical steps in cell growth and metabolism. Using the ICCB's robots and automated microscopes, they treated human cancer cells in mini-dishes with 13 different concentrations of each compound. Then they probed cell physiology using fluorescent stains for DNA and 10 proteins involved in replication, cell structure, or signaling pathways. Automatic fluorescent microscopes generated pictures covering 70 million cells, representing all combinations of compounds, concentrations, and probes.
Taking snapshots in the dark, Timothy Mitchison (rear), Zachary Perlman (front), and their dutiful robot (left) work to collect thousands of images for a high-throughput analysis of drug effects on human cancer cells. (Photo by Jeff Cleary)
The next step was to wring every possible bit of information from the cell pictures using new image analysis techniques. While any high-school biologist can recognize a cell under the microscope, training a computer to do the same is not a trivial problem, explained Zachary Perlman, a biophysics graduate student and first author on the paper. To pick out cells, Perlman had the computer zoom in on the bright blue-stained DNA in the nucleus and use that landmark to make 93 different measurements on cells treated with each drug dose. Some parameters, like the amount of DNA or the location of one of the stained proteins, had obvious biological significance, but other measurements, like the size or shape of the nucleus did not. "Our goal was to measure everything we could quickly," said Perlman. "We wanted to get a large set of unbiased measurements and see if that told us anything new about how the drugs were affecting the cells."
Perlman's efforts effectively converted cell pictures into numbers--and lots of them. Through a cross-river collaboration, Bauer fellows Altschuler and Wu took the roughly 1 billion data points and, with Slack, developed a number-crunching algorithm to organize the mass into a biologist-friendly format. The final output looks a lot like data from DNA chips that track the activity of many genes simultaneously, but in this case, the changing colors represent the levels or location of proteins or DNA in the intact cell. Most importantly, Altschuler and Wu's programs enable the investigators to compare the complex, dose-dependent effects of many different drugs in a single experiment.
Catching Drugs in Action
Altogether, the measured parameters add up to a fingerprint of drug action on the whole cell. Comparisons between known drugs showed similar fingerprints among drugs that hit similar biological targets, suggesting that this approach will be useful to identify targets of new drugs or predict early on which drugs will be toxic to cells.
| "This work illustrates the potential of microscopy to go beyond a description of what something looks like to being a detailed quantitation of what's going on in the cell." |
"This impressive piece of work sets a new standard for analyzing the effects of drugs on cells," said Robert Murphy, a cell and computational biologist at Carnegie Mellon University. "This is what people in systems biology are pushing for--to automate collection of biological data on a large scale and combine that with computational analyses to accurately make comparisons between the things being analyzed."
Pharmaceutical companies have expressed interest in the technique, said Mitchison, which should be easy to implement and improve as better robots, imaging systems, and software are developed. In the Harvard community, Mitchison looks forward to making automated imaging and analysis widely available to researchers through the continued development of the ICCB.
For his own research, Mitchison wants to use the screen to understand how cancer drugs act on cells with different genetic flaws. The current study used garden-variety cancer cells because they are easy to grow and manipulate on a large scale, but Mitchison wants to measure the response of a range of cancer cells to a precisely chosen set of antitumor drugs, in hopes of getting information that would point to a more targeted use of drugs in the clinic.
--Pat McCaffrey