The Technology That Came in from the Cold, Part 1 February 4, 2000

Due to circumstances beyond my control, last week's My View could not be presented. So, this week's will contain a little more content than usual but I hope you enjoy it. It describes a fascinating technology that has the potential to open up new vistas in our understanding of proteins and how they interact.

What a weird year so far. It has been sunny in Seattle while the eastern US is blanketed by frost and snow. My sister-in-law in Raleigh reported that the wind chill factor was 3 ° and that there had been 20 inches of snow on the ground. Maybe this is what global warming will bring. We will have weather like LA, Anchorage will become warm and LA will fry. So, while we enjoy the nice winter weather we have been having, I thought I would talk about a cool technology that has suddenly gotten hot. It is called cryo-electron microscopy. And it is being used to determine protein structure in ways that have not been accessible until now.

A protein's structure determines its function. Often, by investigating what the structure is, we can understand how it interacts with other proteins, how its biological properties are developed and how to modify its activities. Two major technologies have been used for directly examining protein structure. These are X-ray crystallography and Nuclear Magnetic Resonance (NMR).

X-ray crystallography is the gold standard. It can determine the structure of a protein to atomic resolution, allowing us to view every atom in the protein. If any of you have had exposure to the technology, you remember lots of equations, but the basics are simple. Because of the nature of a crystal, all the molecules are lined up in exactly the same orientation in 3 dimensions. The electrons present in each molecule can bend, or diffract, the incoming X-rays. Since the molecules are all in the same orientation, the diffractions are additive, resulting in small amounts of detectable X-rays off the main axis of the beam. The intensities of the diffracted beams can be used to back-calculate the positions of the electrons, and, thus, the structure of the protein. If, that is, you can determine one key important piece of data, the phase of the X-ray beam.

A wave of any type can be described by its wavelength (the distance between wave peaks), amplitude (the height of the wave, or its intensity) and phase (at what point between a peak and a trough the wave is at when it is detected). The X-ray acts as a wave, and, while we know its wavelength or we can record the intensity of the diffracted beam, we can not directly determine the phase. Without the phase, no back-calculation. There are a lot of tricks, using big-mouth names like isomorphous replacement, to get around this but they are not trivial. Of course, you can only use the X-ray beam if you have big enough crystals. In many cases, particularly for large proteins, this is not possible, thus making structure determination using X-ray crystallography moot.

These difficulties have limited the usefulness of X-ray crystallography to relatively small proteins or proteins that are easily crystallized. The presence of heterogeneities, such as glycosylation differences, make it much more difficult to generate crystals. In addition, except in some limited cases, the image is static. You can not easily look at intermediates in a cellular process. Conformational changes can prevent a protein from being crystallized with its ligand bound. A second approach, NMR, does not require crystals — it can be done in solution.

NMR uses the nucleus of the atoms (thus the nuclear in it name), not the electrons. It shares some superficial principles in common with X-ray crystallography. The signal is enhanced by aligning all the nuclei in 3 dimensions using a magnetic field instead of a crystalline lattice. Instead of looking at the perturbation of an X-ray beam, the nuclei are perturbed with a radio pulse. By altering the timing of the radio pulses, or how many of them you use, or any variety of different protocols, you can learn many things about the local environment of the nuclei. One of these is the distance between two atoms. In a simple system, with a few atoms in the molecule, you can get enough distances to accurately determine where each atom is (i.e. A is 2 Å from B which is 2 Å from C which is 2 Å from A tells you they form a triangle). In the case of a protein, hundreds of such 'distance constraints' must be applied to find models that fit all of them.

And that is what is generated by NMR, models. NMR does not directly produce a single structure. It results in a group of models that all fit the constraints provided by the data. For small molecules, the models can accurately reflect the actual structures of the protein in solution. However, more atoms generally require a larger magnetic field to separate out all the signals. In addition, not all nuclei produce useful signals, so lots of tricks using stable isotopes of carbon or nitrogen are needed. To get an accurate structure of a molecule as large as a protein you need a big magnet and a lot of distinct signals. Because of practical limitations, NMR can not routinely determine the structure of molecules much larger than 50-60 kilodaltons. And getting high resolution pictures containing amino acid sidechains can be very difficult.

Many of the drawbacks for both X-ray crystallography and NMR are being addressed. Better crystallization protocols or more ingenious use of computers continue to push back the boundaries. Against this background, I would now like to introduce the technique of cryo-electron microscopy. This technology is complementary to both X-ray crystallography and NMR, providing novel views of macromolecules. This technology uses proteins in solution without the need for stable isotope incorporation. It can produced detailed images of structures from 50 Å to several thousand Å across.

An electron microscope uses the scattering of an electron beam to detect small structures. We are used to seeing it used to detect small organelles or cellular membranes. However, its resolving capabilities are quite good and it has been used to visualize complex proteins, such as antibodies and ribosomes. In fact, the first models of a ribosome were based on electron micrographs. This was done by painstakingly examining each image of a ribosome on the micrograph by eye. The first structure of the ribosomal particle was proposed by Lake 20 years ago. Antibodies raised to ribosomal proteins could be visualized bound to the ribosomes, in order to construct models of their locations on the ribosome. Remember, all this was done BEFORE the application of computational approaches. So even the published models were done by hand.

Now, normal electron microscopy literally burns the sample away. So usually the sample is stained first with a heavy metal derivative, so-called 'negative staining', to maintain an image of the structure even after the electron beam destroys it. Cryo-EM does not examine a stained image. It uses a rapidly cooled aqueous solution containing the sample, thus the "cryo". This produces a vitrified 'ice' in which the protein molecules are suspended in random orientations. In order not to destroy the sample before data collection, the contrast is turned way down by lowering the intensity of the electron beam. This makes identification of each image quite difficult, but the electron micrographs can be digitized and a computer can examine each molecule. Applying all sorts of fancy algorithms (involving arcane Fourier transforms, etc.) the computer can overlay all the molecules that are in similar orientations. Since all the orientations could be seen at once in a micrograph, the computer can take them all and "sew" them together into a complete structure. And adding more data in the form of more pictures produces better signal to separate from the noise. So, by examining say thousands of images, the computer can generate structures that show much more detail than any one molecule on the micrograph. These structures are not only biologically relevant but actually can tell you something about the biology that is taking place. Here is a site that gives a great tutorial.

Of course, a computational analysis is critical for using this approach. And new algorithms are being produced all the time. Take a look at some raw data and compare it to the final model. If it was not for the fact that it is relatively easy to check the model using other approaches (e.g. epitope mapping), it would be hard to believe that such high resolution models could be generated from such apparently low-resolution images. It kind of reminds me of some of the image reconstruction tools displayed in a James Bond film. You know, wait for the computer to completely resolve a single pixel into a high resolution picture.

Cryo-electron microscopy does not suffer from some of the problems of crystallography or NMR. You do not need to generate crystals or use weird isotope replacement to get your data. You can use normal protein in a normal buffer suspension. Instead of extensive protocols to align the molecules in the same orientation BEFORE data collection, it simply examines single protein particles. A computer then aligns the molecules AFTER data collection and assembles a structure. Its main drawbacks are the size of the proteins that can be examined and the resolution of the resulting model. It is much better at determining structures of large molecular complexes megadaltons in size than it is for proteins that are under 100 kilodaltons. However, it can easily distinguish small molecules in a large complex (e.g. tRNAs) and the size limit is being lowered. The resolution can be increased by using higher contrast images or using enhanced processing. Theoretical limits indicate that it may be possible to elucidate the structures of proteins 20-30 kiloDaltons in size. And, as I will discuss next week, the resolution is being enhanced to below 10 Å, allowing secondary structures to be examined.

Next week I will discuss some actual examples recently published using cryo-EM. These range from icosohedral viral particles almost 2000 Å across with MW over 1 gigadaltons made up of over 370 proteins, through 30S, 50S and 70S ribosomal particles, to multimers of a 130 kilodalton protein. This technology has the potential to provide somewhat blurry, but nonetheless biologically important, views of proteins. In combination with other computational approaches (such as threading), this technique may be able to give us a clear view of how proteins interact to form the large multisubunit structures that control most of the cellular processes. As a preview, check out this Quicktime movie of translation on a ribosome that used actual structures with tRNAs bound. It simply could not have been done using any other approach.