At A Glance
Tim Zachaerewski, Associate Professor, Biochemistry and Molecular Biology, Michigan State University
Education: 1990-1992 — Post-doctoral studies – Receptor research in Pierre Chambon’s Lab, Institute de Chemie Biologique, Strasbourg, France
1990 – PhD, Toxicology, Texas A&M
1986 – BS, University of Guelph, Ontario.
Tim Zacharewski, an associate professor of biochemistry and molecular biology at Michigan State University, is a long-time user of microarrays in his work, and has served as an advisor to the US Food and Drug Administration. His lab investigates mechanistic toxicology, which involves elucidating how synthetic and natural chemicals elicit adverse effects.
BioArray News spoke with Zacharewski at the recent pharmacogenomics regulatory workshop in Washington, DC (see p. 1) to learn how he uses microarrays and discuss his views of the future of the technology.
What brought you to Washington for this workshop?
I think there is tremendous potential for the technology in a regulatory setting, in terms of drug approval, as well as in terms of environmental health.
How many people do you have working in your lab, and how do you find help and fund your work?
We have two post-docs, five graduate students, four technical staff — including undergraduates who work one semester and then take classes the next. It averages around that range. We have some $400,000 to $500,000 a year in grants.
One of the challenges [in recruiting] is identifying graduate students or postdocs with the right tool sets, and skills. We are in a toxicology, preclinical environment that is computationally intensive, and statistically intensive. You want [workers] to have experience with biochemistry. All people in our lab do in vitro, as well as in vivo work. Like the technology is holistic, the training has to be holistic too. You see what is going on in setting up the experiment. You learn how to handle the animals, how to do treatment, do tissue culture, do the dissection. You prepare the samples, do the hybridization, the development and maintenance of the array. You are involved in the running of the array assay, analysis of the data, and interpretation of the data. You are going to take it from soup to nuts so that you have a full understanding. Not everybody has good bench skills, but they might be great computationally. They will eventually get self-directed, but they will have to suffer through the other sets. Those without computational skills and statistical skills will still be heavily involved in the terms of the analysis. Interpretation will always fall back to that individual. After you have identified your set of active genes, or you have ranked and prioritized them, then it’s a manner of putting that in biological context.
When did you first start using microarrays?
I guess I was seduced by the technology. My department is a pretty eclectic department with about 40 faculty involved. I got into it about five years ago because the guy across the hall from me, Christophe Benning, a plant biochemist, was doing gene-expression [research] in brassica, to try and determine expression patterns that contributed to oil production in brassica seed. So the technology, and the research laboratory, and the infrastructure were available. There was a presentation in the department —– I mentioned before [that] it was seducing —– I could see the potential, so I initiated preliminary studies through some seed funding, and I was able to get my first grant through the EPA, of all agencies. Then the NIH started to get more involved.
What kind of platform are you using?
We have used Affymetrix GeneChips but now we are actually producing our own cDNA microarrays. We are going more towards cDNA microarrays just because of costs and the flexibility that [cDNA arrays] provide as well. Our human and rat arrays [printed at 10,000-spot density per array] have been designed so that there are orthologous genes that are represented on each one of the arrays, with the goal of examining cross-species comparison, extrapolation of mechanism[s] between species, and the conservation of mechanisms and responses across species as well.
One of the reasons why we went with cDNA arrays was really to be able to customize our arrays a little bit more. The other reason is because the Michigan Life Sciences corridor provided access to clone sets at no cost to me. So, I was able to go in and cherry-pick those specifically to customize them. So, we have arrays that are rich with genes that may or may not be associated with the specific mechanism of action based upon some in silico searches. We have selected genes specifically to try and maximize the degree of overlap of orthologous genes across those arrays as well as to investigate cross-species comparisons through gene-expression response.
What is your throughput of arrays per year and what is your average cost per array?
This year [throughput] should be 1,000. Next year it should be 1,500. We can probably do up to 2,000 a year.
Including labeling, it costs about us about $100 to $125 per array, not including labor. GeneChips cost me about $800 a sample — for the chip itself, the labeling, and a service fee to get access to the reader and the hybridization.
You mentioned the Michigan Life Sciences Corridor. How did you benefit from that initiative?
The availability of infrastructure. We have had access to clone sets; we have expansion of technologies, services, and equipment available to us — robots, spotters, and real-time PCR. With that, we can now go in and do some verification of changes in gene expression. On the protein level, we can use proteomic approaches, whether that is mass spec-based or protein arrays. We have some bioinformatics support that has been really valuable as well.
What questions are you trying to answer?
I’m small-ligand focused, asking: How does ligand structure affect gene expression? We are primarily focused on receptor-mediated mechanisms of action, and to even get more specific, estrogen receptor-mediated mechanisms. There is a lot of flexibility in terms of the small molecules that will interact with those receptors, so we can then ask: How do those changes in gene-expression contribute to the physiological response, or the toxicological response?
We are interested in host-receptor binding, and interaction with a regulatory or responsive gene, and looking at ‘How do the changes in gene expression contribute to the physiological response?’ From there we can understand the mechanisms, or the molecular phenotype that is contributing to the physiological response and the identified biomarkers that are going to be more predictive. I say ‘more predictive’ because if you have a mechanistic understanding of the biomarker, then the predictive value will increase. That is where we are trying to go, in addition to getting a molecular phenotype, and put the changes in gene expression into biological context. Some people refer to it as phenotypically anchoring your gene expression data. And, ultimately, to go toward representing a biological network: What is the cascade of molecular events that contribute to the physiological response? What is the interaction between the genes, what are the pathways that are activated [and] inactiviated?
Our data is different than what a lot of people are focusing on. A lot of people are focusing on profiling cancer, for example, different stages, and different types. We are looking at the effects of the small molecule, and we are doing things like time course, dose response. I think you would be hard-pressed to find a microarray study looking at dose responses published in the peer- reviewed literature. With no model, what is study design? What are the analysis approaches you should use? How do you analyze those results? There is not much guidance to give us clues to what we are doing. We are sort of groping in the dark here.
We try to make our best effort and hope that the reviewers recognize that, given the resources we have available to us. We aren’t a big pharma; we are not at one of the premier institutes within the NIH systems. I don’t have a school of statisticians at my fingertips that are familiar with the doctrine associated with gene expression. I do collaborate with a statistician, Chris Gennings, at Virginia Commonwealth, and we have been working together for about four years. It’s been a painful four years —– her trying to get up to speed on the technology, and the data, and the formats —– and me in my lab trying to figure out the statistics, which are pretty sophisticated and not something you would see in a grad program. We are starting to get into general linear mixed models, we are using an empirical- Bayesian model as well – to identify genes and to do some normalization as well. There are opportunities to get into reverse engineering, support vector machines are used more to identify potential relationships or associations between genes, so it sort of reconstructs the pathway.
What are your models?
Right now, we are using continuous cell lines, hematoma cells, rat and mouse. In addition, we are doing rat and mice in vivo as well, and human, rat, and mouse in vitro. We have enough to keep us busy for a while. The continuous cell lines are of value as a learning exercise. If we can learn how to handle the experiments, analyze the data and triple the results with a simpler model, then we will progress into more complex models. It’s surprising, actually, that the in vivo models are the easiest ones to work with, and are a little more robust in their reproducibility.