In the August issue of Nature Biotechnology, a team led by Ben Cravatt published a paper describing their approach to small molecule profiling of proteins (see p. 3). In an interview, Cravatt discussed both his background and research in more detail:
AT A GLANCE
NAME: Ben Cravatt
POSITION: Associate professor, chemical biology, Scripps Research Institute, and co-founder and member of scientific advisory board, ActivX Biosciences
PRIOR EXPERIENCE: Worked with Richard Lerner and Dale Boger as a graduate student at Scripps Research Institute
How did you begin working in the field of proteomics?
I was a biology major but I worked in a chemistry lab [at Stanford University] as an undergraduate. I went to Scripps as a graduate student with the goal of doing significant amounts of synthentic chemistry but also integrating that approach into biology. I worked with Richard Lerner and Dale Boger primarily. One of my main areas of research that came out of my graduate work was to understand how enzymes regulate signaling molecules in the brain. A significant portion of our lab doesn’t do proteomics; it focuses on individual enzymes that we think are interesting, and goes through and studies their structure, function, and does knock-out mice and makes inhibitors.
But in the course of those studies we encountered several instances where we wished we had the ability to more broadly profile enzyme activities in complex samples and crude proteomes, to look for new enzymes and help us understand whether the inhibitors we were making were selective for the enzymes. That’s where the proteomics came in. I felt we could do that with chemistry–that we could make chemical probes that would label the active sites of large classes of enzymes directly in a proteome, and then you could just visualize changes in activity based on selective labeling events. That really started a year or two into my academic career–as an independent faculty member–which was probably ‘98 or ‘99.
How is your approach different from the way others are using chemical probes to look at proteins?
One of the other people who is heavily involved in this field is Ruedi Aebersold, and Ruedi and I are quite close collaborators. Ruedi has been focusing on trying to find a global way to quantitate protein levels, so he uses isotope tagging, which is really a very powerful approach, but it won’t distinguish an active from an inactive protein. Our goal is to capture a significant but not unreasonably large fraction of the proteome, and then at the same time be able to screen for changes in activity–regardless of whether or not an enzyme was changing in abundance or not. If its active site chemistry and structure was altered in a way that would prevent it from reacting with the probes, or made it inactive, we would see that. That was our main goal, especially for enzyme classes like proteases, where post-translational modification regulates their activity. We felt it was pretty difficult to work with mRNA levels or protein levels and really understand whether those sorts of enzymes were actually active. We don’t look at nearly as large a portion of the proteome as something like what Ruedi does, but I think we get an enriched level of information from the portion of the proteome we look at.
What’s the strategy for coming up with the different types of probe chemistries?
There’s two strategies. The Nature Biotechnology [20, (8), pp 805-809] paper covers one, and then we have a paper coming out in PNAS which is the first big biological application of this kind of approach. That uses another type of probe. Taking the latter one first, the simplest thing is to exploit or take advantage of the last 50 or 75 years of pretty intensive enzymological research, where people have come up with affinity reagents that bind the active sites of enzymes. Biochemists have been using these things for years to help them sort out of what type of class an enzyme might belong to.
Most of those studies in the past were done on single enzymes in a test tube, and we said, ‘Well, maybe some of these reagents could actually be coupled with a biotin tag or a fluorophore and allow you to use them in the whole proteome to visualize these enzymes.’ In that case you’re taking advantage of known chemistries and just reconfiguring them to make probes for proteomics-compatible research. If you look at the PNAS paper, we used fluorophosphonate-based probes to target the serine hydrolase superfamily. The family is large, composing probably one to two percent of the proteome, so you can get a significant amount of proteomic information, enough to depict the state of cancer cells just from that probe alone. But for a significant fraction of the proteome there aren’t the sorts of well-defined affinity agents and that’s where the approach we describe in Nature Biotechnology comes in, which is where we make libraries of candidate probes. We give them a certain amount of reactivity, we give them some variable binding groups, and then we couple them with fluors. Then we let the proteome teach us what it likes to react with, as long as we screen in a relatively intelligent manner, which just means we look for probe-specific and heat-sensitive reactivities. Pretty much without exception we came across interesting protein-small molecule interactions that usually take place on the active sites of enzymes. It’s a pretty rapid way to discover new proteomics-compatible probes from different enzyme classes, because we do all the screening directly in the whole proteome–we don’t do anything with purified enzymes initially.
The paper talks about a two-tiered way of screening proteins. How does that work exactly?
We’re actually writing up a paper right now where we’ve merged those two to some degree and made trifunctional probes that have a fluor and a biotin on them. There are other approaches. ActivX Biosciences has sublicensed a lot of this technology and has complementary ways to get proteins out for identification that don’t use biotinylated probes at all. That’s a proprietary approach and it’s pretty cool, but for us we still have to use either a two-tiered approach or trifunctional probes. We like the trifunctional probes a lot because it allows you to see by fluorescence when you avidin-precipitate a protein. That paper will probably go out in the next couple weeks or so. But both the work described in Nature Biotechnology and PNAS used the two-tiered approach.
How do you know that you’re actually binding the protein on an active site?
It’s a little easier with FP [fluorophosphonate] probes to target certain hydrolases because there’s so much work that’s been done [already]. We can show that no inhibitors of those enzymes compete off the labeling. It’s pretty straightforward. With Greg Adam’s approach [published in the Nature Biotechnology paper], he’s had to do a lot more work. He had to find ways to compete off reactivities with things that we know bind the active site. He looked for if the enzyme was co-factor dependent, would co-factors affect labeling, or if you had a substrate, would the substrate affect labeling. In every case where he’s looked, which is a significant number, he’s been able to recreate those events. But in the end the only way that we can validate beyond that is to go in and actually identify the residue that’s being labeled. That’s one of the things we’re doing in the future.
Do you have to necessarily do x-ray crystallography?
It depends on whether members of the enzyme classes have been solved by x-ray crystallography. If they haven’t, then yes you probably do, but in some cases there’s already a core member that’s already been solved, so you if you label a certain residue you can map it on to the core structure, and see whether it’s the active site or not.
What’s the potential for multiplexing these types of profiling experiments?
I’ve actually talked with Ruedi about this as well, and ActivX is working hard in this area. The 1D gel approach is powerful due to its throughput, but its not that high-resolution. We’re trying to avoid 2D gels at all costs because they are just so slow. If you look at the PNAS paper we say that to do that biological study it would have taken us 500 2D gels. It’s an outrageous amount of work–we couldn’t have afforded to do it! The cool thing is that there are other ways that one can separate proteins once you’ve labeled them at a single site. You could do what Ruedi does–break them up and run peptides out and separate those out. That’s something we’re now working on–looking at new ways to separate other than just by 1D gel. Obviously if you do use 1D gels, one way you can gain a little bit is to use different fluors, which we describe in the Nature Biotechnology paper.
That’s powerful, but it’s somewhat limited because there’s only so many fluors that give orthogonal imaging capacity on a given gel. There’s probably only about three or four fluors you can use, so you’re not going to multiplex a thousand times over with that sort of approach. But it still works for us; it helps. If you have a proteome that’s of limited quantity, where we don’t have an unlimited supply of it, we’ll sometimes throw multiple probes in with different fluor tags on just so we can get the most information out of whatever limited sample we have. But what’s kind of nice about the method is that it doesn’t require that much sample at all. In the initial screening, we can get proteomic information out of 20 micrograms of protein or 15 or 10 micrograms of protein. It’s a pretty small quantity that we need for detecting differentially expressed proteins.