By Meredith W. Salisbury
For years now, you’ve watched the disciplines that form the foundations of systems biology ramp up to truly impressive throughputs. DNA sequencing went from a radioactive reaction performed tediously to one measured in millions of lanes per year. Gene expression, protein identification, RNA interference — one by one, all these technologies have geared up to a level worthy of being part of the systems biology pantheon.
All, that is, except for imaging. Even today, many labs that otherwise traffic in high-throughput instrumentation still include the inevitable scientist peering hopefully (or quizzically) through a microscope, trying to assess the outcome of some experiment.
Finally, though, the field is showing signs that those days are numbered. Imaging technology has gained in throughput in recent years, and scientists are beginning to ask genome- or proteome-scale questions. With respect to imaging, high throughput can mean viewing cells or whole organisms in three dimensions; in time-lapse images of a certain reaction or a phase like embryogenesis; or even taking hundreds or thousands of images of different cells to compare anything from expression to localization.
In Cambridge, Mass., the evidence is in the Whitehead MIT BioImaging Center, which was set up in 2001 “with the idea that microscopy would be the new tool of discovery in systems biology,” says Director Paul Matsudaira. “There’s a natural transition or evolution in the questions that people will ask for which microscopy or imaging is the natural tool for getting those answers.”
One such question looks at how chromatin is structured and organized within the cell, says Tom Misteli, senior investigator in the cell biology of genomes group within the National Cancer Institute. That’s something that you can’t find through biochemistry or other typical wetlab experiments; imaging will likely be the only way to get at that information. And ramping up to high-throughput imaging will be the only reasonable way to find such information at a systems level. It will also help scientists make the transition from subjective data — a scientist’s judgment call on what he sees in the image field — to an automated system that is “much more accurate than we can do by eye,” Matsudaira says.
Still, that doesn’t mean high-throughput imaging is for everyone. Nikolaus Grigorieff, an associate professor at Brandeis University, spends his time visualizing protein structures within cells — and he says that for such painstaking work, scientists will simply have to plod through the old-fashioned way. “We tend to work on samples that are very hard to get in the first place, and they usually do not behave so well in the microscope,” he says. “Automation where you just take hundreds or thousands of pictures will not get us anywhere.” Grigorieff believes research that relies on very standardized biochemistry and sample prep is more likely to find success in high-throughput approaches. “Everybody’s thinking about high throughput,” he says, “but maybe it’s not always justified.”
Is high throughput right for you? GT spoke with some leaders in the field to get a sense of what kind of science such imaging has enabled, as well as what obstacles still need to be overcome for this to become a mainstay technology. (Predictably, informatics remains the largest bottleneck.) Read on to find out whether your work could benefit from ultra-fast imaging and analysis tools.
Experts who have been watching this field agree that the last 18 months have been a particularly critical time for high-throughput imaging. Before that, they say, this kind of technology was being used by just a handful of people and was fairly early in its development. Ger Brophy, a vice president at imaging instrument provider GE Healthcare Bio-Sciences, says the change he saw at a recent conference compared to what he saw in the field a year and a half ago was significant. “I’ve seen an incredibly accelerated technology acceptance curve in the last 18 months,” he says, pointing out that what was once an esoteric tool is now in the purview of scientists from pharma, biotech, and academia.
“There’s definitely a lot of growth in this area,” says Judy Masucci, director of marketing at Cellomics, which has instrument offerings in the space.
That growth can be traced back to a number of advances in imaging technology, say scientists in the field. “This has all been really spawned by GFP and the ability to look at proteins in living cells,” says Misteli at NCI. GFP and the many subsequent fluorescent tags that have been developed made it possible to do imaging in parallel, says Cristina Montagna, who directs the Genome Imaging Facility at Albert Einstein College of Medicine. Her group has advanced technology that allows them to combine the fluorescent tags to make even more unique tags — that requires special equipment to sense the difference in emission wavelength, but also gives them the ability to up their throughput by labeling every chromosome in just one hybridization experiment.
Other technical improvements have also been key pieces of the puzzle. Gordon Hager at NCI’s Cancer Research Center says that “fluorescent microscopes are getting more standardized. It used to be for every instrument two or three people [in a lab] knew how to use it and that’s it.” Bill Mohler, an assistant professor at the University of Connecticut Health Center, says new approaches to laser excitation as well as tweaks in scanning and detector technology have been important. In particular, the switch “from photomultiplier tubes to CCD camera [has resulted in] higher efficiency for collecting photons,” he says. At Whitehead’s bioimaging center, research scientist James Evans points out that advances in digital cameras and basic graphics power that’s available even on desktop PCs have made a lot of this much more accessible to biologists.
And, like any other suddenly accessible technology, biologists are taking hold and running with it. Imaging, which was once primarily used for information that came in binary format such as traditional high-throughput screening — Did a reaction happen? Was a protein expressed? — has now become a tool that scientists can use to ask far more complex and interesting questions: How does this change over time? What level of gene expression happens depending upon localization in the cell?
You’ve probably heard of at least a few big-ticket projects using large-scale imaging, such as the one between vendor Amnis and Roger Brent’s Molecular Sciences Institute designed to examine signaling pathways in yeast. Another fairly well-known example is the European research project MitoCheck, which aims to study how mitosis is regulated and includes automated microscopy as a core element in its raft of technologies.
Matsudaira’s bioimaging center has managed to stay ahead of the curve in part thanks to deals with technology vendors. The center is a beta site for Cellomics, and has three high-throughput imaging platforms from that company. “We were one of the first academic groups going into high throughput in a very serious way,” Matsudaira says.
For now, the bioimaging center is focusing on fairly basic science while ironing out technical wrinkles, says James Evans. “We’ve been looking at cell structure and morphology and its relation to motility in three dimensions.” Protein localization will be important, too. “We’re hoping to actually map a whole group of proteins related to cell adhesion and the cytoskeleton in 3D over time,” he says. The reason imaging will be crucial to systems biology is in the potential to correlate image data to other data sets, such as those generated by microarrays. “That’s actually what we’re doing right now,” Evans says. “We’re treating cells with a range of concentrations of different drugs and then doing the imaging experiments, but also doing complementary expression experiments to see whether the changes in morphology and adhesion structure in cells is in any way linked to gene expression.”
Sophie Lelièvre at Purdue University tried out high-throughput imaging to further her studies of how cell components are organized within the nucleus. She worked with David Knowles of the Lawrence Berkeley National Laboratory and came up with a new approach to tracking protein location in cells. Their research relied on fluorescently staining cells and feeding those through an algorithm that would detect the average fluorescence and pick out only the brighter than average spots that corresponded to Lelièvre’s protein of interest. To track distribution, they cut the nucleus into thin slices and measured the density of bright spots; then the algorithm generated information that revealed not only the presence of these proteins, but precisely where they were found throughout the cell — all without a scientist squinting through a microscope and hoping for the best. Eventually, Lelièvre says, this work could be used to differentiate phases of cancer within cells.
Many people who are trying to take advantage of high-throughput imaging have gone beyond looking at single cells. “If you get good information from the cell, you can get better information from the organisms,” says Ger Brophy at GE Healthcare. He says that while GE’s imaging technology tends to be oriented to cellular work, customers have started using it for model systems such as zebrafish.
One of the most popular models for imaging has proven to be the trusty nematode. At the University of Washington, Zhirong Bao has worked with Bob Waterston and other team members on a system “where we can map gene expression onto individual cells,” Bao says. Using time-lapse laser confocal microscopy and automated image analysis, he tracks gene expression as the worm develops by taking snapshots of the embryo in slices a micron thick; the slices stack up to provide a 3D view of the organism. Eventually, he says, “we want to profile all 20,000 genes in the worm onto individual cells. … We want to know for each and every cell what genes are used.”
Elsewhere in the C. elegans community, scientists have taken a new twist on an old tool. Denis Dupuy, a member of Marc Vidal’s Dana-Farber Cancer Institute lab, has been studying worms with a flow cytometer. The instrument flows, analyzes, and sorts worms just like smaller flow cytometers do with cells. Dupuy has been using the machine through a collaboration with Union Biometrica, which developed the tool known as Profiler. His work focuses on tracking localization of gene expression in the worm. “This method has been applied to one gene at a time,” Dupuy says of the previous research in this area. “We wanted to crank it up to the genomic level.” After four years of working with the company, Dupuy is writing up his findings in a paper; he says getting funding to support this large-scale project will be the next step.
Rock Pulak, director of biology at Union Biometrica, says the tool has also been used successfully with embryo or larval stages of Drosophila, zebrafish, and mosquitoes, as well as with Arabidopsis seeds and seedlings. He contends that high-throughput imaging is a “first-pass type of approach.” He compares imaging to microarrays: in the same way people go back and follow up important array findings with RT-PCR, he predicts that high-throughput imaging will be critical to show scientists “the broad brushstrokes” which they will go back and verify “using conventional confocal microscopy.” Pulak says that like gene expression, RNAi has proven to be quite complementary to imaging tools.
That merging of data sets is the latest trend in the field, says Mark Collins, senior product manager of bioinformatics at Cellomics. “We’re starting to see people begin to federate together disparate data sources to solve systems biology problems,” he says. “[But] we’re still only on the very first rung of the ladder with respect to how we bring all of this data together to produce systems biology models.”
Rapidly Expanding Data
It’s in fact the imaging data that has become the sticking point in this technology, and will likely scare off a good number of labs otherwise interested in it.
A picture may be worth a thousand words, but how much money is it worth? Researchers looking to bring this kind of technology into their labs will first have to face the tremendous challenge of preparing storage for the data. At the Whitehead MIT BioImaging Center, microscopes running at full capacity could generate a terabyte a day — each. The center currently has 40 terabytes of storage and limits itself to producing about a terabyte per month, but even that could shut out the average scientist. Data expansion due to processing and analysis can be as much as 20 times the volume of your raw data, says Evans.
To give a sense of scale, Director Matsudaira provides some back-of-the-envelope calculations for a genome-wide experiment. Someone looking at every human gene using the 3D image stacking his team performs — 150 slices per stack for 24,000 genes expressing proteins in two colors — would be facing 15 terabytes of data. “All of a sudden now people can’t play this game,” he says. “That’s just the raw data. You haven’t even started analyzing the information.”
Speaking of which, here’s another major hurdle: data analysis. Many labs use software written in-house because scientists feel that what’s on the market is not sufficient. “Every image analysis program has its own nuance. A generic package would only get you so far,” says Washington’s Bao, whose team wrote a program called StarryNight to track cells and detect gene expression.
Meanwhile, UConn’s Bill Mohler is using imaging to study embryogenesis of C. elegans, and his group also worked on software to automate analysis of worm images. To increase throughput, his team loads the image field with 12 embryos at a time — but that means the images come out and need to be separated, synchronized, and aligned for comparison purposes. Mohler’s software does just that, creating a rotating, 3D movie for each individual embryo.
Bao, for one, is on a mission to help improve image analysis — by dragging in experts from other fields. He attended the Genome Informatics meeting at Cold Spring Harbor Laboratory last year in part to encourage computational biologists to take an interest in this technology. “We should get the message out to the computational folks out there that DNA sequence and protein sequence are not the only computable material out there in genomics,” he says. “Images are computable as well.”
Informatics experts aren’t the only thing this nascent field could use. “One of the biggest issues in imaging right now is resolution,” says Misteli. “There are a lot of fundamental questions which could be addressed if we had resolution somewhere between electron and confocal light microscopy but could be used on living cells. That would really be a tremendous step forward.”
As resolution rises, cost will have to come down before scientists can really embrace high-throughput imaging. “It’s still a relatively expensive technique,” says Albert Einstein’s Cristina Montagna, speaking in particular of combining fluorochromes to increase throughput in labeling. “It’s not something that everybody can decide one day they want to buy the kit.”
Community spirit is something else the field could use. Connecticut’s Bill Mohler says a good step forward will come from organizing the community around projects to analyze and annotate images as well as automate correlation of expression of different genes, for instance. Mohler says that this spring his lab will push to get fully funded for imaging work so his scientists can spend more time pushing their research, and this technology, forward.
That sort of community effort to solve these problems could go a long way toward realizing the full potential of imaging in systems biology — especially as better tools are developed to extract meaning from images. “Knowledge is gained from being able to relate the information that we’ve extracted from images to the data that comes from a microarray experiment or a proteomics experiment or the literature,” Matsudaira says.
Looking forward, Brophy at GE Healthcare says he expects that scientists will add chemical genomics to the list of data they correlate with images. The other trend he foresees (and one that will likely cause instant headaches among people already trying to figure out how to handle imaging as it stands right now) is researchers actually pulling together image data from multiple platforms — say, live cell assays and whole model-organism images — to enable “longitudinous studies across different biology platforms.”
But for now, let’s just give scientists a chance to figure out how to pull even one platform of this technology into their labs. That alone is still a sizable challenge, especially considering how little data has emerged from these experiments and been made public so far. Several scientists, such as Denis Dupuy, are hoping that publications will help call attention and respect to the field.
“We’d like to establish some good data sets,” says James Evans at Whitehead, who says such data will be necessary to prove the value of high-throughput imaging to the rest of the community. “At the moment we’re kind of the new kid on the block.”