At A Glance:
- NAME: Stanley Fields
- AGE: 47
- POSITION: Investigator, Howard Hughes Medical Institute, since 1997
- Professor, Department of Genome Sciences and Medicine, University of Washington, Seattle, since 1995
- PRIOR EXPERIENCE: Faculty member, State University of New York at Stony Brook, 1985-95
- Postdoc, UCSF, 1981-85; studied with Ira Herskowitz
- PhD, University of Cambridge, MRC Laboratory of Molecular Biology, UK, 1981; studied with George Brownlee
So you invented the yeast two-hybrid system?
The method goes back to 1987. I was at the State University of New York at Stony Brook, where I was a faculty member. There was a seed grant available from the technology group at Stony Brook to develop some idea that would have commercial applications. Based on the availability of that grant, and our work in transcriptional regulation at the time, we came up with the assay, to use transcriptional factors as a means to detect protein interactions. It was really the seed grant that crystallized our thinking. So we did put in a grant, which was not funded. [However,] the technology is now available for licensing from The Research Foundation of the State of New York.
We were working on a signaling pathway in yeast that ultimately led to transcriptional changes in response to the mating pheromone that the opposite cell type of yeast secretes. We were particularly interested in the terminal point of that signaling pathway, which was this transcription factor, and how it worked, and how it might get modified by these signaling events. This was in the mid ‘80s, and there was a lot of work, especially from Mark Ptashne’s lab, that showed that these proteins often had two domains, one responsible for DNA binding, one for transcriptional activity. Both the experiments from his lab and from Steven McKnight’s lab showed that proteins can function as transcriptional activators when bound to a protein that was itself bound to DNA, rather than directly by binding to DNA. Those were the general features of transcription factors that we were aware of that allowed the two-hybrid idea to work.
When did people start using it on a genomic scale?
Many people have used it for single proteins. I don’t think that there are that many genomic applications yet. Early on after the original publication of the first two-hybrid search, which was in 1991 [Chien et al., PNAS 88(21):9578-82], Paul Bartel and I started to think about how we could scale up the assay to handle more than one protein at a time. The original example we did this with was phage T7, an E. coli phage that has about 55 predicted proteins. We started around 1993 to set up an assay where we could look at all of the phage proteins and their interacting partners at once. We worked that out over the next few years, and it was published in 1996 [Bartel et al., Nature Genetics 12(1):72-7]. Around the time when I moved to Seattle, which was in 1995, we started looking at approaches to scale up from that to look at interactions of yeast proteins on a genomic basis. We collaborated with Curagen on that approach, and as we started to scale up yeast, there were other labs doing genomic two-hybrid screens, including Ito’s lab in Japan looking at yeast proteins, Marc Vidal looking at C. elegans proteins, and Pierre Legrain in France working on yeast proteins. Everybody [has] tried different kinds of approaches, and at this point, there have been several different large scale two-hybrid studies. However, I would say that the number of people doing that compared to the number doing just single proteins of interest is still a tiny fraction.
How can contradicting results from large protein interaction datasets be reconciled?
I wouldn’t say they contradict each other, I think they are complementary, and they often yield different pieces of data. Not only do the two-hybrid and mass spec sets not always agree with each other, but if you look at multiple two-hybrid or multiple mass spec studies, there is substantial lack of overlap among them. I think the reason for that is partly the fact that nothing is saturating; we still don’t know a lot of the protein interactions. They all have some background of false positives as well as the false negatives that don’t come up, and then they vary in their sensitivity and protein expression levels and so on.
The other feature to keep in mind is, two-hybrid is always direct, one protein to one protein, so in almost all cases, you are just looking at pairs, and they are divorced from their native context: they are expressed at high levels in the yeast nucleus as parts of hybrid proteins. Mass spectrometry and the purification of protein complexes doesn’t rely on one-to-one interactions but on a whole complex coming down. So if you pull down ten proteins, you don’t know whether protein A binds to B or to C or to D, and so forth. Depending on how it is analyzed, sometimes all the pairwise combinations are included, but most of those don’t exist in Nature.
The answer is to [view] what comes out of the two-hybrid, out of the mass spec as candidates, and then look at the same time at protein co-expression data, array data, protein localization data, and various computational approaches. By doing all of that, you can assign probabilities that given proteins are likely to interact or not, and several groups have now published papers on these kinds of computational approaches. If you combine different sets using different approaches, then you get data of higher quality [due to] the fact that complementary approaches have identified the same piece of information.
Can you tell me about the malaria project you are involved in?
That’s the only one where we are using traditional two-hybrid on a large scale. It’s a collaboration between us and Myriad Proteomics in Salt Lake City [where Paul Bartel is now employed]. The idea is to build libraries of both binding domain and activation domain fusions, and then Myriad has a high-throughput technology to carry out many two-hybrid searches. We are hoping to do, ideally, tens of thousands of Plasmodium two-hybrid searches with them. None of it is for commercial interest. The data will all be public, and none of the interactions will be patented. We are trying to get funding from the NIH to pay for that, and this is something where Myriad’s capacity and worldwide need match up very well.
[The aim] initially would be to just infer something about function of the various proteins. Ideally those functions would then get converted somewhere along the way into potential drug targets. In addition, we are working on that as part of a structural genomics consortium based at the University of Washington that aims at solving the structures of proteins from Plasmodium, trypanosomes, and Leishmania.
Tell me about the Yeast Protein Linkage Map project.
We had built an array of about 6,000 two-hybrid transformants that fuse each of the predicted yeast open reading frames to the GAL4 activation domain, which we used in the original work with CuraGen. In the original publication, we published about 200 searches using that array, so 200 proteins were tested against the 6,000. We have since gone on to do close to about 1,000 searches overall. Most of these have been collaborations between us and various yeast groups out in the community who have an interest in a particular set of proteins or even a single protein. They have contacted us and then we have worked with them and we have run the searches in the lab here. Some have been bigger efforts, on the order of 10 to 50 proteins, but in most cases it’s been on the order of one to five proteins.
You say in a recent review article that it’s important for the proteomics community to work “hand in hand with those focused on biological problems”...
[The Yeast Protein Linkage Map] is, I think, one example of that; there are many others now. Our effort in this is part of a group here called the Yeast Resource Center that’s funded by the NCRR, the National Center for Research Resources. The Yeast Resource Center is interested in collaborating with the community at large, and does this both for two-hybrid, but also for approaches that use mass spectrometry, microscopy, and protein prediction. I think that proteomic approaches clearly have their [greatest] power when they are combined with dedicated biologists who are interested in a specific process.
What are your favorite variations on the yeast two-hybrid theme?
I don’t know if I have a favorite. They answer different questions, and I think they all have potential strengths. We have been doing a lot of work with a membrane protein interaction system [invented by Alexander Varshavsky’s lab at Caltech and adapted by Igor Stagljar’s lab in Switzerland] that looks like it has considerable potential to identify new interactions of membrane proteins.
How does it work?
The way it works is that instead of fusing two bits of transcription factors to the proteins that you are studying, you fuse the N-terminal and C-terminal domains of ubiquitin to the ends of two membrane proteins. In addition, the C-terminal domain of ubiquitin is also linked to a transcription factor. When two proteins come together in a membrane, then the ubiquitin halves get together and form a quasi-native ubiquitin species that gets recognized by proteases that are present in the cell. Those proteases cleave the C-terminus of ubiquitin, and that cleavage then releases the transcription factor, which goes into the nucleus to turn on a reporter gene.
Have you used this successfully yet?
We have an array of about 700 proteins, which are all fused to this N-terminal domain of ubiquitin. At this point we have tested a couple of hundred proteins against this array by fusing those proteins to the C-terminal end of ubiquitin with the transcription factor. We are detecting interactions, but as with all of these assays, there are false positives and false negatives. We are hoping to submit this later this year for publication.