The Aaron Klug Centre for Imaging and Analysis

Lecture 3: CryoEM

Recent developments in cryo-electron microscopy have enabled the determination of three-dimensional structure at both the macromolecular and cellular level; this, in turn, has led to a number of significant biological insights signalling the re-emergence of electron microscopy as major tool in structural biology.

Electron microscopy will be remembered by many as a laborious technique in which skilled practitioners somehow managed to get meaningful information in spite of mistreatment of biological tissue which involved crosslinking, dehydration, embedding in epoxy, sectioning and staining. These techniques were prone to the production of false structures and certainly did not lead to the reliable imaging of objects of molecular dimension. Furthermore, information about the third dimension was usually inferred and depicted in drawings. Much has changed!

The favoured method of specimen preparation now involves cryo-fixation – in which the material is subjected to rapid cooling by immersion in liquid ethane at -170°C. For small specimens this results in vitrification of the water i.e. solidification without the formation of crystalline ice thereby enabling the molecules to be trapped in their natural hydrated state. Thin films of vitrified water in which macromolecules and macromolecular complexes are suspended can then be inserted directly onto the vacuum of the electron microscope using specially constructed holders which maintain the specimen at liquid nitrogen temperature. Sufficient contrast to image the molecules can be obtained by underfocussing the microscope.

These techniques have led to spectacular images of viruses, large macromolecular complexes and macromolecules having masses greater than 300kD.

The vitrification of specimens greater than a micrometre or so in size presents problems as the cooling does nor occur sufficiently rapidly and crystalline ice forms which destroys the surrounding structure. The key to overcoming this problem was the realization that attainable cooling rates would result in vitrification if the pressure were increased to 2×108 Pa – the pressure at which freezing point depression is maximized. Machines now exist which enable tissue samples with dimensions of up to 0.5mm to be vitrified at high pressure – this size is useful for almost all cell biological investigations. Reliably obtaining usable cryo-sections of cryo-fixed material remains an unsolved problem. However, this has been circumvented by the process of freeze substitution, which involves dissolving out the water in the solid state and replacing it with resin which can then be polymerized so that the sample can be handled conventionally. The gain with this procedure is substantially better specimen preservation without the problems associated with low temperature specimen examination. The penalty is that one does not have the a priori confidence that the specimen has undergone no potentially destructive treatment.

Having overcome the problems of artifacts introduced by sample preparation, the next issue is the creation of three dimensional images. The image obtained in the transmission electron microscope is the sum of the density contained in all the planes in the object parallel to plane of the image, i.e. it is a two dimensional projection of a three dimensional object. The problem in creating three dimensional images is devising a method of separating these planes. This can be done by obtaining projections of the object from a number of different angles. In cases where the object under examination is unique, the only way this can be done is by tilting the stage in the electron microscope and recording an image at every angle of tilt. This process and the subsequent reconstruction of the images is called tomography. It has only recently become practical to do this on a routine basis due to advances in electronic image recording devices and computer control for electron microscopes. With these new instruments, the microscope can be programmed to collect a tilt series rapidly and without human intervention. Indeed, the bottleneck in the process has moved to the interpretation and it has become important to develop adequate software tools to help segment and analyse these incredibly rich images. At the present time, application of tomography to mitochondria and the Golgi apparatus, for example, have produced startling insights that have forced us to reevaluate our understanding of these organelles.

In the case of macromolecules or other objects in which it can be assumed that every object is identical, an entirely different strategy is possible. Macromolecules dissolved in aqueous media usually exist in completely random orientations, thus views of fields of macromolecules embedded in a thin film of vitreous ice provide views of molecules from every possible direction. The problem is to determine the relative orientations of the molecules. This is done by comparing the image of each molecule against a set of projections generated from a three dimensional model of the structure. After the whole set of images has been classified in this way, a new model is generated by back projecting the experimental images. Convergence is usually achieved after iteratively repeating the classification and model generation – this is judged by the stability of the membership of the classes. This technique is inappropriately named the single particle method.

The resolution of structures obtained either by tomography or by the single particle method depends on the closeness with which angular space is sampled and, of course, the signal to noise ratio in the images. The issues that need to be confronted are slightly different in each case. In electron microscope tomography it is impossible to view the specimen from all angles because at high tilt angles the beam is obscured by the specimen holder. This leads to inhomogeneous resolution in the final image. However, use of prior knowledge of the contrast in the specimen in the reconstruction algorithm can minimise the effect of this in the final image. In the best examples, macromolecular complexes on the scale of GroEL can be identified inside the bacterial cytoplasmic matrix.

In the case of reconstructions made by the single particle method, reliable structures of asymmetric objects at 2.5nm resolution can be achieved from a few thousand images. The number of images required can be reduced if the object has intrinsic symmetry, thus GroEL with its 7-fold symmetry can be reconstructed to this resolution from approximately 1000 images and as few as 20 images may be required for viruses with icosohedral symmetry. Practically achieving higher resolutions than this involves a combination of instrumental design features and sophisticated image processing. Commonly available (and relatively inexpensive) transmission electron microscopes are unsuitable for high resolution work. Key features which are needed are field emission guns, which have higher temporal coherence than conventional tungsten filaments, accelerating voltages higher than 200kV which lead to a reduction of contrast reducing elastic-inelastic double scattering and high stability stages to reduce specimen drift. Additional improvements can be gained by using an energy filter to remove the inelastically scattered electrons. Microscopes with these features have been used to obtain maps of sufficient resolution to enable a polypeptide chain to be fitted in several cases in which the proteins form two-dimensional crystals. These are special cases which allow averaging over several million molecules using essentially crystallographic techniques combined with tomographic principles, but they illustrates the capability of the system. The most well known examples include bacteriorhodopsin, alpha beta tubulin and photosystem II.

Among the problems that need to be overcome computationally in the application of the single particle method at high resolution, is the correction of the contrast inversion of certain spatial frequencies that occurs as a result of recording images at the significant distance underfocus required to actually find the particles. Current technology allows a resolution of 0.8 to 1.0nm in the hands of most laboratories with the right equipment. It is believed that the single particle method should be capable of yielding 0.45nm resolution. However a serious problem is that extracting the higher resolution data from the noise requires very large numbers of images (each of which contain a large number of pixels) and the classification of these images presents a challenge for today’s supercomputers.

Improved high resolution information must depend on increasing the signal to noise ratio. It is a law of statistics that noise can be reduced by increasing the electron dose, however this in turn increases damage to the specimen. It has been discovered empirically that substantially less damage occurs at liquid helium temperatures and a number of microscopes with which allow imaging between 10 and 20K have recently been installed in major Japanese, European and American laboratories.

The success of EM structure determination has clearly earned it a place alongside X-ray crystallography and NMR in the arsenal of the structural biologist. Detailed comparisons of the X-ray and EM structures of ribosomes at low resolution and of bacteriorhodopsin at intermediate resolution have shown that the maps are similar. This is not unexpected since both techniques yield images as a result of the interaction with atoms in the macromolecule albeit that the underlying physics is slightly different. Observed differences in the maps were either due to errors or were legitimate differences caused by changes in the flexible regions of the structure being constrained by crystal packing forces. Other real differences were due to the different weightings given to low and high resolution data by the crystallographers and microscopists, and the different criteria used for resolution determination.

An understanding of the maps obtained from both techniques allows results from EM and X-ray crystallography to be pooled in different ways to accelerate structure determination and to extend the range of structures that can be determined. Thus in the case of the ribosome, the intermediate resolution EM maps were used to provide initial phases for the X-ray structures and in the case of large multicomponent structures such as pyruvate dehydrogenase, each of the individual components (the atomic co-ordinates of which are known) could be fitted into the low resolution EM structure, enabling approximate atomic models of the whole enzyme to be assembled. The most important niche for structure determination by the single particle method is in the study of conformational changes that occur in the operation of macromolecular complexes. Having the particles freely suspended in solution prior to vitrification has unique advantages – substrate can sprayed onto enzyme solutions less than a millisecond before vitrification enabling the transitional state to be trapped and imaged. Additionally the particles are free from the constraints imposed by crystal packing and thus are more likely to adopt their biologically active conformation. Movements of domains can clearly be visualised and in cases where component X-ray structures are available, atomic interpretations are possible. The multicomponent dynamic systems studied successfully to date include chaperonins, the ribosome, the nicotinic acetylcholine receptor, a number of different motor systems and pneumolysin (a membrane pore forming toxin) to name but a few.

Functional studies increasingly point to the importance of large multicomponent assemblies. Cryo-electron microscopy is uniquely positioned to visualise these either inside the cell or in isolated preparations thereby ensuring it a productive future in cellular and molecular structural biology.