Associate professor, chemistry and biochemistry
University of Texas
At A Glance
Name: Edward Marcotte
Position: Associate professor, chemistry and biochemistry, University of Texas at Austin; co-director, UT Center for Systems and Synthetic Biology
Background: Assistant professor, chemistry and biochemistry, UT-Austin, 2001-2005; co-founder and principal scientist, Protein Pathways, 1999-2000; Postdoc, University of California, Los Angeles, 2000; PhD, UT-Austin, 1995
Edward Marcotte of the University of Texas at Austin is the latest in a bevy of researchers developing "cell chips," or ordered cellular arrays, to facilitate high-content analysis of single cells for functional genomics and proteomics applications. Marcotte and colleagues have developed spotted cell microarrays, and are coupling the technology with automated imaging to measure cellular phenotypes and protein localization on a genome-wide scale.
Marcotte's lab has already demonstrated the utility of the method by using it to measure changes in yeast cell phenotypes in response to mating pheromone, work that was published in the January 31 online edition of Genome Biology. Now, Marcotte and colleagues are turning their attention toward more complex measurements, such as sub-cellular localization of proteins; and other cell types, such as bacterial and mammalian; and have just been awarded a four-year, $326,000-per-year grant from the National Institutes of Health to further develop the cell chip technology.
This week, Marcotte discussed with CBA News his most recent research, the increasingly popular cell array format, related imaging and informatics challenges, and commercial potential for the technology.
Where did the interest for this work begin?
The interest came from trying to figure out genome-scale methods for understanding functions of genes. Obviously, sub-cellular localization and cellular phenotype are very important indicators of gene function. We were thinking about various methods that would allow us to measure those aspects of a gene's function across all the genes in a cell. We've investigated things like mass spectrometry-based proteomics; bioinformatics and looking at protein interaction networks; and of course, DNA microarrays; and we started thinking about high-throughput microscopy as a complementary technique that would give us interesting and valuable information about gene function.
What made you decide to develop the cellular array format?
Our version of the cell microarrays came from two things. One was the availability of spotting robots that we and neighboring labs use to manufacture DNA microarrays. The other was the availability of controlled strain collections, like the yeast deletion collection and the GFP-tagged library, as well as the approaching availability of siRNA-treated mammalian cell collections. We began thinking about how best to take advantage of those. There had been earlier reports of cell chips for mammalian cells, and even though at the time we were thinking about yeast strains, we wanted to identify generic technologies. It occurred to us that we could use the spotting robots to try and print the cell collections, and by doing that, we could very quickly manufacture slides that encompassed the entire yeast deletion strain collection. The principle was that we could print off replicate slides, just like DNA microarrays, and then each slide could be used in a separate imaging assay, where we could use differential stains or immunohistochemistry to interrogate each slide for a different marker or protein. At that point, we would then see the genetic determinants of localization or expression of that marker across the deletion collection, and then each slide could give us a different genetic screen. We though that we could use it for investigating genetic interactions and modulators of sub-cellular localization.
You alluded to other cell chips for mammalian cells. CBA News has covered similar research from the Whitehead Institute (6/8/2004) and Max Planck Institute (3/8/2005). Are you familiar with this work, and how is your technology different?
The Genome Biology paper describes yeast collections, and none of the other work was being done with yeast collections. If you extend it into the future, and spotting human cells, the real difference is that each cell we spot can be genetically distinct. The current generation of the siRNA human cell microarrays consists of identical cells cultured over spotted DNA or siRNA, or some variation thereof. But the cells cultured across those are all genetically identical, and then take on different characteristics at each spot.
We can spot any variants at each spot. They could be identical, different strains, different cell types, different organismal lineages any particular variation and then we can look at the variation across them. It's a very general technique that's not limited to particular cell types, and it's complementary to the other style in that it allows us to investigate a different axis of variation, if you will.
What types of techniques did you use to compare and contrast the thousands of images you obtained from these assays?
For the data in the Genome Biology paper, we would usually collect a brightfield or DIC image, and one or two fluorescent channels mostly single channels for nuclear stains. And we would collect about 10,000 images, or about two images per 5,000 strains. More recently we've been doing sub-cellular localization assays, and collecting two or three fluorescent channels, which takes us up to 15,000 or 20,000 images per slide. For the very first ones, we actually ended up doing manual screening of phenotypes by different individuals. They would give quantitative scores and then compare the scores given by different curators. So it was not automated. Since then we've had to go back and develop image-segmentation software to pick out the cells, and begin extracting quantitative information from the slides. For yeast, this has been extremely difficult. There are a number of groups developing software like this, and it turned out to be a real chore to segment the images of the yeast cells and extract fluorescence information. For higher cells, it's a much more straightforward problem.
Did you consider high-content imaging platforms for your work? You used fluorescence microscopy driven by Molecular Devices' MetaMorph image-analysis software in the Genome Research paper.
We did all of our imaging essentially on a Nikon [fluorescence] microscope with an automated stage, piezo-electric driven objective, and under the control of MetaMorph, and we would write appropriate scripts. The trick was figuring out what grid to follow initially. So we would actually scan the slides using a DNA microarray scanner to find the grid of spots, and then give those coordinates to the microscope and have it autofocus on each one.
We looked, but didn't choose to go with a high-content automated imaging platform, for whatever reason. But this method has worked out to be good enough and generic enough for us it seems to be working well, and is actually quite simple to do. At this point we've collected something like 400,000 or 500,000 microscope images of yeast cells. Something we anticipated early on, and did have to end up doing, was to develop the database infrastructure to store all of these. There, we looked toward some of the groups developing open-source versions of this. They weren't developed early enough, however, to be useful to us, so we ended up developing our own platform for cell microarrays that stored, handled, and allowed us to curate and parse all of the data that were generated.
What made you decide to look at yeast's genetic response to pheromones as a test case?
We've been working from easier assays to harder assays. The first, most obvious assays to us were morphology changes, so we weren't even looking at sub-cellular information, but simply cell shape changes. The first thing we did is look at the deletion collection and make sure that the shapes we saw were consistent with what had been seen previously large cells, small cells, elongated cells, and those sorts of things. The second thing we did was attempt to change the shapes. Pheromone response in yeast is a classic assay in which the yeasts grow into a polarized mating projection, and have a very characteristic shape change. By adding pheromone to the deletion collection, we then had a visual screen for which cells failed to change shape, and those are defects somewhere along the pathway for pheromone response. It was essentially a very nice proof of concept for the method, and we ended up discovering a number of new genes in the pathway.
Now you're moving onto sub-cellular localization studies. What types of modifications will you have to make for this?
One thing we've done is imaged the GFP-tagged yeast strain collection. Now, each strain is genetically identical, but has a different protein tagged for GFP. Images from one of these microarrays give us readouts of where all of the proteins are located each strain informing us about a different protein sub-cellular localization. By manipulating the library in some way, we can then see how all the proteins respond by changing their sub-cellular localization of expression patterns. So it gives us a readout of the spatial dynamics of the proteome in the cells. We've also moved toward developing immunohistochemistry against the cells by printing spheroplasted cells. In yeast, to do immunohistochemistry, you have to first remove the cell wall. In this case we had to fix the cells, remove the cell wall, and print the spheroplasts themselves, and then we can screen the slides with a fluorescently tagged antibody against something that we're interested in. Both of these have moved us into the difficulty of extracting quantitative information from fluorescent images.
And that's something that you're addressing now?
Yes, we are developing software to extract the data and recognize localization. We're building upon the work of a number of people in the field, such as Bob Murphy at Carnegie Mellon (see CBA News, 2/15/2005) and others who are describing quantitative features of sub-cellular localization, and then using algorithms to try to find proteins with similar localization and changes in localization.
Are the genome-wide localization assays what you recently won an NIH grant for?
Yes, that's a four-year grant that just started. Given the funding difficulties in the NIH right now, it was trimmed a bit from what we asked for. But it covers five years to get the spotted cell microarray platform going, and one of the goals of the grant was to be able to measure protein and RNA levels from single cells. That, of course, is a dilemma for the detection of a lot of the fluorescent signal that we're looking at, and requires amplification techniques to amplify the fluorescent signal from low-copy RNAs. But we were trying to get single-cell detection of protein and RNA from the same cells, so we could start looking at things like stochastic variation across cells.
Is this something you plan to extend to other cell types?
We've already extended it to bacterial and mammalian cells. For bacteria, we're in the middle of genetic screens for fluorescent sub-cellular phenotypes in E. coli; and we've also shown that we can print mammalian cells. It's very early stage for each, but we already have proof of concept for them.
Are you applying for patents on any aspects of this, and is there commercial potential?
We have applied for patents on the spotted microarray technique, and we think that it's very generic and can be applied to lots of different circumstances. Actually, what spurred the patent was the notion that one group could print replicate slides. Imagine printing the deletion strain collection prepped for immunohistochemistry, printing off lots of replicates, and then distributing those slides to other groups who could screen for the particular thing they're looking for. That seemed to be a very simple commercial product that could come out of this and could be of use in the same way that proteome chips are being marketed.