Northeast Structural Genomics Consortium
At A Glance
Name: Gaetano Montelione
Position: Director, Northeast Structural Genomics Consortium, since 2000. Professor of molecular biology and biochemistry, Rutgers University, since 1998. Faculty, Rutgers University, since 1989.
Background: Postdoc, Gerhard Wagner's lab, University of Michigan, 1987-1989.
PhD in physical chemistry, Cornell University, 1987. Worked in Howard Scheraga's lab at Cornell, and Kurt Wuthrich's lab at the Swiss Federal Institute of Technology in Zurich.
A team of researchers led by State University of New York at Buffalo professor Thomas Szyperski (see ProteoMonitor 3/3/2003) recently published a paper in the Proceedings of the National Academy of Sciences describing their progress in determining protein structures via high-throughput GFT-NMR. The work was sponsored by the Northeast Structural Genomics Consortium, which is the only large-scale center funded by the Protein Structure Initiative (see PM 2/18/2005) with a strong NMR component.
ProteoMonitor caught up with Gaetano Montelione, the head of the Northeast Structural Genomics Consortium, to find out more about his background and the development of high-throughput NMR.
Can you describe the role of the Northeast Structural Genomics Consortium?
I'm the head of the Northeast Structural Genomics Consortium, and this is a large, multi-institutional project aimed at determining large numbers of protein structures. It's actually an $80 million grant, because we have $30 million the first five years, and now we have another $50 million for the second five years. Certainly the automated NMR is a part of it, but it's only a small part actually.
What are the other parts of it?
The other part of it is, we determined 200 structures in the past five years, and these were done about half by NMR and half by X-ray crystallography. So there are a lot of robotics methods involved, as well as sample production. There's bioinformatics involved and target selection. There's function discovery that we've done, and a comparison of the value of X-ray crystallography versus NMR. That paper has just now been accepted in the Journal of the American Chemical Society. That paper shows that we actually tried to crystallize 250 proteins that we also took NMR spectra for. And we found that there was no correlation between the quality of the spectra and the success in crystallization. What that means is things that give beautiful spectra don't [necessarily] crystallize. So one can determine those structures by NMR. And things that give miserable NMR spectra do [sometimes] crystallize, and you get the crystal structures. So this has a very big impact for proteomics, structural biology and pharmaceutical design. The pharmaceutical companies have cut back a lot on their NMR efforts because NMR is slow, they feel it's not automated, and also [they thought that] anything you get a good spectrum of, you're going to get a crystal structure of. We've shown all of those things are not true. The NMR process now can be very fast, it can be automated, and many things that give beautiful spectra don't give crystals.
What was your role in the study published recently in PNAS?
As far as the PNAS article led by Thomas, I'm a collaborator, and the main thing that we did was develop some of the software for automated NMR data analysis. You see, his contribution is very fast data collection, which is very critical. But then once the data is collected, you have to analyze it, which can be very slow. We've developed tools to automatically analyze NMR data. Those programs are called AutoAssign and AutoStructure. Those have been published in the last year or so, and they allow us to go in a day from the data to the structure. That's very important. So whereas traditionally it took maybe a year to do a structure between the data collection and the analysis, now all the data can be collected in a couple of days and the data can be analyzed in a day or two. So now you see papers like the one we just had in PNAS, where the average time for a structure was about five days.
And that's in large part due to the software?
Well, it's actually three things. It's a hardware development of what's called cryogenic probes — these probes that have very cold detection circuitry that gives them much better sensitivity. The second thing is the GST-NMR that Thomas developed, which is a way to collect the data that is much more efficient and allows you to collect multi-dimensional spectra rather than two-dimensional spectra. So this is a big improvement in the speed of collecting the data, and it complements the increased sensitivity of the cold probes. The third thing is once you get the data, you've got to analyze it. So what we've been doing over the last several years is developing software tools to make that analysis automated and fast and reliable. It's really those three pieces coming together that gives the high-throughput structure determination approach.
Is your background originally in structural biology?
I'm trained in physical chemistry. I got my PhD at Cornell University. I did protein NMR spectroscopy for my PhD. I trained there with Howard Scheraga. And then I went to work with Kurt Wuthrich in Zurich, who later got the Nobel Prize for developing protein NMR as a tool for structure determination. I was in Wuthrich's lab for about a year, and when I left there, that's when Thomas started there.
And then from there to a laboratory at the University of Michigan with a guy named Gerhard Wagner. And Gerhard was actually another one of Wuthrich's proteges. He trained as a graduate student with Wuthrich and stayed for several years as a research associate. So we all are connected with this lab in Zurich.
Then in 1990, I came to Rutgers University, and now I'm a distinguished professor of molecular biology and biochemistry.
When did you start working on the GFT-NMR?
Well, as I said, my major contribution was the automated NMR data analysis part. That we started working on seven years ago. And we started collaborating with Thomas about three years ago.
When I was with Gerhard Wagner, we developed a technique called triple resonance NMR. It's a way to simultaneously observe a proton that has carbon, hydrogen and nitrogen signals in the protein at the same time. And that is the basis for GFT-NMR. So I was actually the first person to do what's called triple resonance NMR. And that was a big improvement and formed the idea that you could automate the process. Another group at NIH actually did a lot more development on triple resonance NMR while I was getting my lab started, but I was the first person to do those experiments. I was the first to publish them.
Thomas followed up on those experiments, initially just doing the same things we were doing, but then he put a new twist on it, and the twist is GFT-NMR. And that really is a big step forward. So triple resonance NMR in and of itself was a big step, and now the GFT flavor, the GFT version of triple resonance NMR is an even bigger step.
Were you involved in the development of the hardware at all?
Well, some of the hardware that's used has pulsed field gradients, and pulsed field gradients were developed in the early 90's. I was involved in developing that hardware, together with engineers at Varian NMR. So that's one part of the hardware. And the original triple resonance NMR required hardware, and I was involved in designing that hardware.
How did you get involved in the NMR software development?
Well when we saw with triple resonance NMR the quality of the data we could get, we realized that it could be automated. So then we started to collaborate with computer scientists here at Rutgers to develop software to — instead of manually analyzing data — to do it in a fully-automated way. And that's taken some time in development, but overall it's been very successful.
Is your NMR software commercially available?
We license it to commercial groups. We give it away free to academic groups. Commercial groups pay money to use the software.
For the second part of the Protein Structure Initiative project, what do you plan on doing, now that you've solved 200 structures?
We're going to continue to enhance GFT-NMR and automated data analysis — that's certainly a part of it. A lot of the new development involves pushing to larger proteins and dealing with integral membrane proteins, where the spectra are much more heavily overlapped and difficult to analyze. So we need to bring in new algorithms and improved GFT pulse sequences.
More generally, in the structural genomics project, we produce hundreds of protein samples from different organisms, and our success rate with bacteria is about 10-fold higher than our success rate with eukaryotic proteins. So a lot of the effort also involves developing improved technology for making human and other eukaryotic proteins that are suitable for either crystallization or NMR data collection.
So a lot of the new technology — there's an NMR technology component, but there's also a strong protein expression technology component.
Also, in the crystallography, we're trying to develop improved ways to do robotic, high-throughput crystallization.
How do you decide which method — NMR, or crystallography — to use to determine a structure?
Well, in our consortium, we try both in parallel.
Is there value to having structures solved by both?
Yes, because sometimes there are distortions due to crystal packing. Other times, having two structures, we can use that to refine and improve our methods. Another thing is NMR is very sensitive to dynamics and flexibility, so even if you have the crystal structure, it's useful to do NMR because you can learn about protein flexibility.
Given that both NMR and X-ray crystallography are automated now, which is faster?
In the most favorable cases, I think the crystallography method is still faster. I've seen cases where you get crystal structures in a day. And the NMR is a day or two. So in the best cases, they're comparable.