Novel image processing tools and techniques in cryo-electron microscopy

The software resources in cryo-EM have to deal with stacks of hundreds of “8k” images. To process this amount of data, a fast and efficient software tool is required. For this, a performance oriented C++ toolkit named EMKIT was developed. This toolkit can be used to accomplish fast basic operations using in-built classes and methods. Furthermore, complex features can be conveniently built using the toolkit. After the advent of DEDs the amount of data that comes out of the microscopes has drastically increased. Today, there exists a handful of software which can automate the data collection process on microscopes. Target spots on the grid are selected and the microscope automatically records images on these spots. With some tricks the microscopes can now yield a new image every minute. This allows obtaining more data in the limited microscope-time. But it is quite important that this data can be processed in real time to get an idea if the recording is running as expected. Focus was designed to accomplish this task in a user friendly environment.
One of the classical methods in electron microscopy is 2D electron crystallography. Obtaining highly-ordered 2D crystals is difficult and time-consuming. However, 2D crystals diffracting to only 10-12Å can be prepared relatively conveniently in most cases. There is a need to develop image processing algorithms allowing to generate high resolution 3D structure from cryo-EM images of badly ordered crystals. Apart from that, while recording tilted 2D crystals, there is a limitation to the possible extent of tilt. This limitation arises due to considerations such as sample movement, increase in thickness of sample with tilt and varying defocus. The achievable tilt angle is typically 60 degrees. This would mean that in 3D Fourier space, the slices with tilt angle beyond this would be missing. This region in Fourier space turns out to be conical in shape and hence this problem in 2D electron crystallography is known as the “problem of missing cone”. In real space, this makes the densities look elongated in the vertical direction. Apart from the missing cone, data can also be missing in other regions, depending on the tilt sampling.
New methods have been developed which can tackle the problems stated above to some extent including (i) movie-mode unbending, which performs frame-wise unbending in the recorded movie frames; (ii) refinement over sub-tiles of the frames in order to locally refine the crystal tilt geometry within different tile locations on the images; (iii) a projective constraint optimisation refinement for approximating the Fourier data in the region of missing cone. All of these methods were applied to MloK1 membrane protein. MloK1, a cyclic nucleotide-modulated potassium channel from Mesorhizobium loti, is a homologue of human HCN (Hyperpolarization-activated Cyclic Nucleotide-gated) channels important for signal transduction and pacemaking. MloK1 in the presence of lipids forms micrometer-large 2D crystals diffracting only up to 10Å. Using newly developed methods, we determined the three-dimensional (3D) map of full-length MloK1 in the presence of cAMP at the resolution of 4.5Å.