University of California, San Diego
Name: Trey Ideker
Title: Assistant Professor of Bioengineering, University of California, San Diego
Professional Background: 2003 — present, assistant professor of bioengineering, University of California, San Diego; 2001 — 2003, Pfizer Fellow in Computational Biology, Whitehead Institute for Biomedical Research/Massachusetts Institute of Technology; 1996 — 2001, Leroy Hood Laboratory, Institute for Systems Biology/University of Washington.
Education: 2001 — PhD, Molecular Biotechnology, University of Washington; 1995 — MSc, Electrical Engineering and Computer Science, Massachusetts Institute of Technology; 1994 — BSc, Electrical Engineering and Computer Science.
The University of California, San Diego, has distinguished itself over the past few years as a hub of activity for researchers using ChIP (chromatin-immunoprecipitation)-on-chip applications to investigate biological pathways.
One UCSD investigator is Trey Ideker. Recently, Ideker, along with collaborators at the Whitehead Institute for Biomedical Research/Massachusetts Institute of Biology, and the State University of New York, Albany, used ChIP-chip to map the transcriptional network controlling DNA damage response to an environmental toxic agent, in this case methyl methanesulfonate.
The group published its results, which the authors claim have significant ramifications for both systems biology and disease research, in the May 19 issue of Science [A systems approach to mapping DNA damage response pathways. Science. 2006 May 19;312(5776):1054-9]. To learn more about those results and their potential, BioArray News discussed the experiment as well as ChIP-chip with Ideker this week.
What is your primary focus at the Ideker Lab?
We specialize in the systems modeling and analysis angle. That's where my expertise is. We are right at the interface between computer science and biology.
We are working with a yeast model, and our expertise is really in the systems analysis. One question that exists is, 'How do you actually take ChIP-chip data and build models or circuit diagrams of different pathways using them? And here we have gone after DNA damage response because it's just about the most important, basic cellular process I can think of.
DNA damage is important for three main diseases: One is, of course cancer — DNA lesions underlie most cancers. Another is problems with DNA repair pathways, which are important for a variety of genetic disorders like Warner's syndrome and [other] aging related disorders. But most important from my funding perspective is how environmental toxins work in damaging DNA. So we are actually funded for this study by the National Institute for Environmental Health Sciences.
Walk me through the experiment.
The goal of the paper was to map DNA damage response in a large-scale way. We took 30 transcription factors as entry points into that pathway that we have shown or others have shown are important for the overall response.
We then used ChIP-chip to measure the promoter binding regions after damage is induced and then compared those binding profiles to the promoters that are bound before damage is induced. So you can look at how the network is rewired after damage.
To validate, we used a gene-expression experiment to knock out the transcription factors and make sure that it is in fact activating or repressing the genes that the model says it is regulating. And by and large the overlap between the gene expression profiles and the ChIP-chip data is not huge, so again, that's a caveat for people that just want to look at ChIP-chip data alone.
In the end we looked only at interactions that are verifiably functional.
Why did you use methyl methanesulfonate to trigger DNA damage?
It's a very popular damaging agent for the study of basic DNA repair. There are different ways of damaging DNA, and MMS is used because it is an alkylating agent and it turns out that several chemotherapies are alkylating agents as well.
What kinds of arrays did you use for the ChIP-chip experiments?
This work was all done using arrays printed by the MIT/Whitehead Facility, which is basically run by Rick Young's lab. So we purchased them from Rick Young. More recently we have been using Agilent ChIP-chip arrays, but not as part of this study. All the data for this study were wrapped up about a year ago.
What is your take on current commercial arrays?
Agilent arrays are the only ones I have experience with. But I know NimbleGen is in the market and basically anyone that makes whole-genome tiling arrays these days could also be in the market.
The main issue with this application is that there's a really long protocol and it remains to be seen whether or not companies will be successful in getting labs to adopt and apply this protocol. I can tell you for us it's been quite a bear, and for anyone who has experience with ChIP-chip, it's a fairly finicky process and it takes a lot of high-level training and trial and error to get it to work.
How do you think the technology can be refined so it can be more successful?
I am not passing judgment on current technologies and I don't know whether or not the technology is refined to the point whether a company can make money off of it, but that's exactly what Agilent, NimbleGen, and others will find out.
We are happy and we love the Agilent arrays, but because of the actual process — the immunoprecipitation, amplification, and so on — it takes a technician in my lab the better part of a year to really get good ChIP-chip results.
Compared to gene expression arrays it's much more difficult.
What has this paper advanced?
Well, there are basically two advancements to this paper. One is the more straightforward one in that we are basically mapping the pathways or set of pathways that make up DNA damage response. That will then have applications in medicine.
But there is also the systems biology advancement. A major problem with ChIP-chip data is that it is very noisy. It's not the only technology to have this problem. Certainly, gene expression arrays have this problem too.
As a rule, this kind of large-scale technology is good for when you want to paint a big picture, but when you actually ask the question, 'Is this gene differentially expressed?' there is a false positive and negative rate associated with that.
So if a ChIP-chip experiment indicated 500 genes as differentially expressed, the question remains which of those are important for the pathway. And if they even are, who knows how many of these binding events in the cell have any functional consequence whatsoever? So you have that two-tiered noise problem.
In our paper, we integrated the ChIP-chip data with systematic gene expression data in response to knockouts of every transcription factor in the system. So there are 30 transcription factors we used to enter into this DNA-damaged pathway and we got ChIP-chip data for each one of them. And then we go through the same 30 factors, knock them out, and validate which targets of each factor determined with ChIP-chip are actually changing in the gene expression study.
So, to paint a big picture, it's this integration of data which we believe can now be used to define these different cellular models. We are interested in technologies that can be used to construct models where the interactions aren't just taken raw data off the ChIP-chip experiment, but are actually systematically validated.
What can be done with this model?
We would like to use this model to find new proteins to target with drugs — to see what node in the model you can remove or abrogate its function with a drug and get a desired response. The model can predict cellular responses now and its connections predict how a cell will respond to different stimuli like DNA damage.
Now we should be able ask whether or not we can impair a pathway. For example, since chemotherapies work by and large by damaging DNA, I would like to be able to go into my cancer cell and basically take out its ability to repair DNA. That way the chemotherapy would work better. So we could find the proteins or series of proteins to target with drugs and render those nodes non-functional.
We also believe the lowest hanging fruit on the disease angle will be can you use these models diagnostically or prognostically.
How can you better promote pathway models like the one you have created?
Well, I think what's really needed is a database of pathway models. Right now, our models aren't exactly textbook. Even our validated models should be viewed as hypotheses. We are not completely raw data, and we are not just an invalidated hypothesis, but we have validated with just one type of validation. And what really needs to be done is to take these things and run them through the ringer.
What you need to do that is a database of pathway hypotheses. Not canonical pathways as you would find in some kind of textbook, but things that are on their way to becoming fact. But that doesn't exist, and given it doesn't exist, all of our ChIP-chip data is in the ArrayExpress database and we have our own website where we do provide these models. But, I think the real payoff for these kinds of pathway maps is yet to come.