“Cryo-EM will Make a very big Difference”
(February 6th, 2015) It is rumoured that structural biology is undergoing a revolution. Once dominated by X-ray crystallography methods, cryo-electron microscopy (cryo-EM) is now transforming the field. Sjors Scheres is one of the driving forces behind this revolution.
Selected by Nature as “one of the people that mattered in 2014”, the structural biologist discovered his preference for working behind a computer screen rather than at the bench early in his career. After a postdoc with Jose-Maria Carazo in Madrid, Scheres was invited to establish his lab at the MRC Laboratory of Molecular Biology (LMB) in Cambridge, UK, the birthplace of modern structural and molecular biology. At the LMB, Scheres developed RELION, image processing software for cryo-EM that would change the field.
LT: You started your research career by doing bench science. Why did you decide to switch to software development?
I did a five-year PhD, half of which was a failed attempt to solve a protein structure, and the other half was programming. When I first did the protein crystallisation project to determine the structure of a transcription factor, I liked the data collection and analysis of crystallography, but I didn’t really like the wet lab part. So when the whole project didn’t work out, I decided I would rather have a theoretical project. I worked with Piet Gros on the protein structure refinement methods for X-ray crystallography and I liked that much better.
LT: What got you interested in cryo-EM?
I thought method development in X-ray crystallography was rather mature; much of it was already on automation, so that wouldn’t be a good field to start my career as a method developer. So I looked for a younger field and then came across electron microscopy (EM). During my postdoc in Madrid, I was working on image processing rather than X-ray diffraction data. Jose Maria Carazo suggested I could work on structural heterogeneity. In cryo-EM, we take images of purified proteins on a thin layer of ice from different angles, and then reconstruct the projected images into a 3D structure. The problem is that many of the proteins that we study adopt more than one conformation in solution, so if we put them all into one reconstruction, we get a blurry image of the 3D structure. We call that structural heterogeneity in our data. What you should do is to separate out the projection images of these 3D structures, and get many reconstructions at the same time from the same data set. At that time, people didn’t know how to do this.
LT: Is this when you started to work on RELION?
No, not really. I was interested in maximum likelihood statistical methods, which had proven much better for structure refinement in crystallography than conventional least square methods. In EM, they were still using these methods. […] When I was in Madrid, I implemented the maximum likelihood method in XMIPP, which was a software package developed in the lab. From the people there I learned how to do programming, and also how to use supercomputers, do parallel computing and so on. And then when I moved to Cambridge in 2010, I started my own programme, RELION, and that was a new framework that works even better.
LT: You postdoc supervisor has strong links with industry, but RELION is open source…
Jose Maria has strong links with industry but, for example, XMIPP is open source and RELION contains pieces of code from XMIPP. That’s the idea of open source, that you can take pieces of code and develop it further. To develop an image-processing package, besides the new algorithm you have to write (which is the new exciting stuff), there is a whole lot of work to be done, more trivial things. That would be a lot of work to write from scratch.
LT: How did RELION change the field of structural biology?
The field of cryo-EM is undergoing a revolution. There are two reasons for that: one is the development of a new detector for electrons, and that I think is the most important one. Before people were taking pictures on photographic film, or on CCD cameras, which would introduce quite a bit of noise in the imaging process. The other thing that has improved is better image processing and RELION fits into that category. What’s special about RELION is that it can find an optimal way of filtering the data automatically, whereas with older programmes, you had to be quite experienced in image processing to get good results. With RELION you don’t need to be an expert, that’s the main contribution. Also you can separate distinct 3D structures from a single dataset with RELION, that’s very important. At the moment, we’re working on making the technique work for smaller protein complexes, and also for those that are floppy, things that have not just two or three conformations, but a whole continuous spectrum of conformations
LT: Any structural biologist should be able to use RELION, but how easy (or hard) is it to use?
We have a wikipage that has a lot of documentation and a tutorial. Usually, people that go through the tutorial and the documentation can use it. We also have an email list where people can ask each other and me questions about problems that might arise. The one limitation in using it is that it’s computationally expensive and you need to learn to use clusters, and sometimes that’s a steep learning curve for more biochemical-oriented structural biologists.
LT: Will cryo-EM replace X-ray crystallography?
Over the past decades crystallography has contributed much more than cryo-EM, but the trend will reverse. One good thing about cryo-EM is that you can work with much smaller amounts of protein. So for those proteins in the cell occurring in very low copy number, and when over-expression is not possible, cryo-EM will make a very big difference. But I don’t think crystallography will become obsolete. For domain structures it will still be much easier and faster to use crystallography. But the relative amounts of structures that are done with crystallography or EM will shift dramatically. For larger proteins cryo-EM will be the driving force in structural biology in the coming decades.
Images (2): MRC LMB