PhD, Systems Biology
Name: Ariel Cohen
Position: PhD in systems biology, Weizmann Institute of Science, Rehovot, Israel, 2004 to present
Background: MSc in computational biology and bioinformatics, Weizmann Institute of Science, 2003 to 2004
In a study appearing in the Dec. 5 online edition of Science, a team of Israeli and American researchers describe a method they have developed that uses proteomic dynamics to study how cells respond to the chemotherapeutic camptothecin.
In particular, they sought to study the effect of the drug on the dynamics of the proteome and how the dynamics differ from cell to cell. To look at the drug response, they measured the level and localization of about 1,000 proteins in individual cells over time, and found a diverse protein response to the drug.
They saw “rapid localization changes of proteins depending on the drug mechanism of action, followed by slower wide-ranging temporal changes of protein levels,” they said in the article.
For most proteins, there was only moderate cell-cell variability in drug response, but for a set of proteins responses were widely different between individual cells. Some of these proteins are involved in cell-fate decisions, and at least two proteins show “cell-to-cell differences that are correlated with the fate of the cell,” the researchers wrote.
“Thus examining spatiotemporal proteome dynamics in individual cells offers clues about what is special about the subpopulation of cells that escapes the drug action,” they added.
This week ProteoMonitor spoke with Ariel Cohen, the corresponding author of the study, about the work and its implications. Below is an edited version of the conversation.
Is this the first study to use this protein-dynamics approach to see how cells respond to drugs?
That divides into two parts: First of all, how you look at proteins responding, before we go into the drug. And usually when people look at proteins responding, they make a soup of the cells and then they use mass spec or 2D gels and so forth, and the problem with those approaches is they are very hard to do over time – the protein dynamics – and, of course, you lose the single-cell data.
Now, when people came to do single-cell analysis, they’re usually not looking at that amount of proteins. Usually, they choose one, two, or three proteins in the system and they delve into that. So what’s new about this approach is two things: First of all, that you can look in single-cell resolution, but at a vast amount of proteins.
Now, regarding the drug response, I don’t know of any work that has been done looking at the single-cell level at such an amount of proteins.
What made you and your colleagues decide that this might be a good way to study drug response?
We had a belief that a lot of very important biology is missed when you don’t look at single cells. We know that most of biology is looking at soups of cells. We thought that’s OK for most of the time, but there is very important, interesting biology that is being missed out.
Now, some work has been done looking at the single-cell data, but usually it’s going toward noise analysis, which is kind of a new prospect that’s been arising these last few years. But it’s very hard to see how this type of noise, the differences between cells, is related to phenotype.
So there is some work that has been done that showed these relationships, but this is the first [study] that I know of that has shown the relationship with the response to the drug.
You mentioned how this approach improves over traditional proteomics methods like mass spec and 2D gels. Did you consider using any high-throughput cellular analysis methods, like any of the high-content screening platforms?
We believe that high-content screening usually has to do with the perturbation you do in the cells, and usually just several parameters – it could be an order of 10, maybe hundreds of different morphological types. … We believe that once you try to focus down on a subset of these small molecules, you can get a very nice readout using our technology. So I would think you could do some high-throughout screening, meaning you can screen the response of many proteins, but not the response of so many molecules. But once you do focus on a subset of molecules, it should be fairly easy to run the analysis that we’ve done with just this one drug.
Can you briefly walk through how the method works?
We infect cells with a reporter. The reporter goes into the host genome, and in some cases it goes into sites coding for a [protein]. This is all random; we don’t know where it’s going [in the genome]. In about 1 percent of the cases, it goes in the right frame shift in the right strand and so forth, and into the coding region of a gene. And we added to the region coding for the fluorescent protein splice-acceptor sites and splice-donor sites. So the cells don’t know there’s something new inside and they just treat this coding region as another exon, so when they transcribe the gene and later on translate it, then they just do it for this reporter in this part of the gene.
Now, we don’t know where it integrated, so we sort the population using [fluorescence-activated cell sorting], and we see all the glowing cells, and these are the cells where the infection really worked out and it turned into a protein tagged with a fluorescent [label], and then we sort these cells into single wells, 384 single-well plates, so each well has just one single cell.
And then we expand this single cell and use 3’ [Rapid amplification of cDNA end by PCR] between the fluorescent tag and the poly(A) signal in the RNA.
This is the way we create the reporter library. And what we do with this library is not just freeze it away … we decided to look at it, and to look at it, we needed high-throughout microscopy, which means you need a microscope that has automatic focusing. It needs to know how to get back to the same X, Y, and Z coordinates in a certain time resolution, so you can follow the cells over time.
And once you have this data, this movie data, the third thing you need to do is actually analyze it. We’re talking about terabytes of information. This is thousands and thousands of movies, spanning over three, four, five days of growth of the cells, so you have to have an automatic pipeline that can deal with all this data. So we just throw it into storage and we have custom software we wrote that takes these movies and does all sorts of image correction, it segments the cells and their intracellular compartments, and it knows how to track different segments over time.
And in the end, it knows how to build a database of all this information, so the end product is just graphs with data regarding the behavior of individual cells you’ve tracked over time.
Why did you choose camptothecin as the drug for this study?
First of all … we wanted a drug with known mechanism of action. We were hoping we could figure out a few things regarding its mechanism of action, and we wanted to reassure ourselves regarding what was already known.
The second reason was that we had the target-tagging library. The library creation came hand-in-hand with all the filming and the movies done in the microscope, so the library was actually expanding during the experiment. So because we had this drug target at the beginning, we decided to run with it. And it turned out to be a very good choice.
Even though it’s not used in the clinic against this lung cancer, the dynamics we found were very interesting and have implications for other drugs used in the clinic against this cell line, against this type of cancer, actually, and in other cancer types.
Were the results you saw in line with what you expected?
We did not expect the fast degradation of proteins that we saw. We did not expect the localization to be telling the details of the story more than the changes in the amount of proteins. We’ve been working with other drugs, and the protein response is not trivial.
I think the last and most important is that we did not expect to find a single protein that would be correlated with cellular fate. Today, people are speaking about networks of proteins being correlated with cell fate, and it’s really like finding a needle in a haystack.
So we believe our approach can start with a thousand proteins, scan them, and then focus on one or two very important proteins at the end of the pipeline.
How do you envision this information being used for drug discovery? Is it basic information about mechanism of action, or can this knowledge be used to modify drugs to improve efficacy or improve safety?
I’ll tell you what we believe, but I’m not sure if everybody thinks this. We believe that this technology should become a basic tool in R&D for drug research in preclinical stages, in very basic R&D. We believe this technology can do several things.
First of all, we believe it can very nicely find biomarkers for different cellular fates. That’s something being discussed. We think it can offer novel drug targets, because we’re not actually looking at specific systems. Everybody is looking at some signal transduction pathways, several receptors. All of pharma is very focused, and I think such a technology, such an approach allows you to look at places and pathways that you didn’t expect would be relevant to the drug you’re using or to the cancer you’re trying to treat, and so forth.
Now, we didn’t show this yet, but we believe that you can [use this] to provide cytotoxicity profiles of different drugs because we think we can correlate the behaviors of cytotoxic compounds and non-cytotoxic compounds, and to actually become a diagnostic tool for different compounds used in preclinical R&D. It could help sort these things out.
I think the last interesting direction, and something we’re focusing on now, is that you can get very deep insights into combinations of drugs. Today drugs are not being used independently but actually as a cocktail, and we think such a system could help you unravel the basic mechanisms of how these combinations are working within the cell. They can actually give you an insight into the cell. It’s like really opening the cell and showing you what’s happening — not only looking at the cell fate, but actually the whole process all the way up to the cell fate.
So you could study all the players involved at once?
It’s not all the players, but we believe it’s some percentage. We believe that we’ve tagged about 20 percent of the viable proteins, or the proteins that are taking part in the cellular action. We believe that there are several thousands of genes working within a given cell line. So although there are around 30,000 genes, we believe it’s around 5,000 to 10,000 proteins that are actually being expressed.
Are you looking at ways to study more proteins with this method?
We’re now thinking in several directions. First of all is expanding the cell line, expanding to other cell lines. A further direction is to turn this method into an even higher-throughout method, so can actually work with 96 384-well plates. We think you can aim at scanning the library and seeing the dynamics of proteins and cells for over several days, and you can have such scans within a week or several weeks. So we believe it’s a doable scanning method within a reasonable time frame.
We can envision how the system can be used to understand better different molecules, different drugs that are being used now, maybe even why some drugs are affecting part of the population and not the other part.
Which of these options do you think your group will be looking at first?
I think turning it into a high-throughput mechanism is one of our goals. Another goal is to elucidate what we’ve already seen. We think it has implications in the clinic, using the data we already have and the approach we’ve already taken with this specific cell line. We think we can have some deep insights into other drugs that are being used against this cell line.
Obviously, there are some ideas arising from the paper, the data we’ve shown in the paper.
Are you working with any partners for this work?
Actually, this project is in collaboration with another lab at the institute, and we’re looking at collaborations with clinicians, doctors, to check our ideas within patient samples to see whether our ideas hold water in the clinic and not just in the cell line.
Is there anything else that you think is worth noting about this method and its potential?
I think there are surprises waiting along the way in single-cell analysis. There’s a lot of buzz around personalized medicine, but I think personalized medicine will be only one stage, and there will be a lot of buzz to see how individual cells of the same person are reacting to drugs. I think that’s relevant to the realm of cancer stem cells – are they there or are they not and so forth.
So I think there are exciting things waiting.