Employing high school students, inexpensive video cameras, standard desktop PCs and terabytes of storage, researchers have found new indicators that improve the detection of blood stem cells dividing in tissue culture. Through digital time-lapse photography, the investigators — from the Vancouver-based British Columbia Cancer Agency and the University of British Columbia, and from the University of Waterloo in Ontario — identified features of the individual clones of growing bone marrow cells that correlated with at least one daughter cell that could regenerate the blood system of an irradiated mouse. Thanks to these visual clues, the scientists boosted the efficiency of stem cell detection by two to three times. It is a method that could be extended to other areas because of its generality. “We are using long-term imaging and ‘behavior’ analysis to discover new biomarkers for cells with a desired in vivo activity,” said Eric Jervis, an associate professor of chemical engineering at the University of Waterloo. He noted that the technique ultimately could lead to ways to evaluate the quality of transplants that are first amplified in culture. Such amplification schemes are an essential part of gene therapy methods and of transplant improvement processes. An image of one well of an array of blood stem cell clones at the final point of a four-day incubation shows a visual marker that can be used for classification (inset). The lineage tree generated from this time course is marked with a brown bar along the trace when a cell displays the uropod — or lagging posterior projection — phenotype. From the figure, it is clear that the trait is dynamic. Because more than 50 percent of the cells in this clone displayed the uropod phenotype, it was scored as not containing a repopulation-competent blood stem cell. Image courtesy of Eric Jervis, University of Waterloo. As reported in the May 23 issue of PNAS, the researchers used Sony Corp. video cameras, each with a 1.45-megapixel CCD sensor chip and a digital interface to study blood stem cells. They opted for these cameras after looking at more expensive cooled cameras because they saw only minor differences in performance. The cooled-CCD cameras cost tens of thousands of dollars more per system and, because Jervis was running five microscope setups, the extra cost was too much. To improve the performance of the less expensive cameras, on long exposures the scientists subtracted a background image to remove CCD noise. They mounted the cameras on Zeiss inverted microscopes, achieving stable focus that would last for several days, a task Jervis said was not trivial. Working with highly purified populations of blood stem cells isolated from adult mouse bone marrow, the researchers first manipulated the cells into a specially designed array of 40 microwells on a slide. They deposited a single cell into each well and incubated it for four days. During that time, they took images every three minutes, illuminating the cells only when recording an image. The rest of the time, the cells were not exposed to light. At the end of the observation period, the researchers had accumulated more than 1800 images of each well. They scored each of the images for morphological characteristics, location and parentage. Jervis noted that high school students were hired for the tracking and feature extraction, a method necessitated by the nature of the data. “I worked with several image processing and tracking people, and our movies are just too complex for a computer to make good sense out of the data,” he said. The human visual cortex, Jervis added, is probably the best feature extractor around. What is more, the descriptors can be arbitrary as long as they are consistent. He said that one student likened the cells to a party, with some that hung back, some that mingled, some that were very active and so on. The researchers worked with and analyzed such arbitrary but dependable descriptions. The final enabling technology for the blood stem cell work was data storage — and lots of it. Each image is composed of megapixels of data, and there are thousands for each microwell. Having such a huge amount of data available for analysis was possible only because the cost of storage is now running at less than $1 per gigabyte. Just as importantly, desktop computers are now powerful enough to process this mountain of data efficiently. With their observational runs completed and analyzed, the researchers took the resulting pedigree diagrams and superimposed data on them for visualization. They found several indicators associated with blood stem cell activity, which they had verified by injecting the cells back into mice at the end of the experiment. The first indicator was longer cell cycle times, with those clones that retained stem cells effectively undergoing one fewer cell generation than those that did not. Another marker, a negative one, was the presence of lagging posterior projections, uropodia, in cells in the final 12 h of the four-day culture period. By then, 37 percent of the clones had a majority of their cells exhibiting uropodia. None of the clones in this uropodia group contained blood stem cells. Such visible distinctions could be important. One unproven hypothesis is that these differences emerge from the engaging of tens or hundreds of molecular stages, revealing a great deal of information. “Ultimately, behavior has the potential to tell us much more about a cell than measuring a few critical molecular markers on the cell surface,” Jervis said. The researchers combined these factors to boost the efficiency of detection of clones that contained stem cells from an initial 28 percent to a final figure of 63 percent. They applied them to two separately executed experiments and, in both cases, improved the detection efficiency of stem cells. With the video-tracking technique proven, the group is attempting to apply the methodology to systems such as neural and embryonic stem cells. The capability offered by the technique also could be used in other ways, once more basic research is done. “We should be able to sort cells with best available markers and then put the cells in culture for a few days to improve purity, identify and eliminate those that exhibit a bad behavior, and, eventually, select only those cells for injection that meet an extensive list of requirements,” Jervis said.