Structure determination of macromolecular complexes using cryoEM has become routine for certain classes of proteins and complexes. Here, this pipeline is summarized (sample preparation, screening, data acquisition and processing) and readers are directed towards further detailed resources and variables that may be altered in the case of more challenging specimens.
Cryo-electron microscopy (cryoEM) is a powerful technique for structure determination of macromolecular complexes, via single particle analysis (SPA). The overall process involves i) vitrifying the specimen in a thin film supported on a cryoEM grid; ii) screening the specimen to assess particle distribution and ice quality; iii) if the grid is suitable, collecting a single particle dataset for analysis; and iv) image processing to yield an EM density map. In this protocol, an overview for each of these steps is provided, with a focus on the variables which a user can modify during the workflow and the troubleshooting of common issues. With remote microscope operation becoming standard in many facilities, variations on imaging protocols to assist users in efficient operation and imaging when physical access to the microscope is limited will be described.
Single particle CryoEM
To investigate life at a molecular level we must understand structure. Many techniques to probe protein structure are available, such as NMR, X-ray crystallography, mass spectrometry and electron microscopy (EM). To date, the majority of structures deposited to the Protein Databank (PDB) have been solved using X-ray crystallography. However, from ~2012 onwards, cryo-electron microscopy (cryoEM) became a mainstream technique for protein structure determination and its use increased dramatically. The total number of EM maps deposited to the Electron Microscopy Databank (EMDB) (as of Dec 2020) was 13,421 compared with 1,566 in 2012 (Figure 1, www.ebi.co.uk). In 2012 the number of atomic coordinates modelled in cryoEM density maps, deposited to the PDB was just 67 but as of Dec 2020, 2,309 structures have been deposited so far, a 35-fold increase. This underlying growth in the quality and quantity of cryoEM density maps produced, sometimes referred to as the ‘resolution revolution'1, was caused by a coalescence of advances in multiple areas: the development of new cameras for imaging known as direct electron detectors; new software; and more stable microscopes2,3,4.
Figure 1: Cumulative submissions to the EMDB from 2012 to December 2020. Please click here to view a larger version of this figure.
Single particle analysis (SPA) is a powerful tool to generate biological insight in a wide variety of sample types by elucidating high resolution structures of isolated complexes5, 6 including viruses7,8, membrane proteins9, 10, helical assemblies11 and other dynamic and heterogeneous macromolecular complexes12, 13, the sizes of which vary by orders of magnitude (from 39 kDa14, 15 to tens of megadaltons). Here, a protocol for a standard pipeline for cryoEM SPA from sample to structure is described.
Prior to embarking upon this pipeline, a purified sample should be subjected to biochemical analysis to assess its chances of downstream success. Preparation of a suitable sample is arguably the biggest barrier to SPA, particularly for transient and heterogeneous (both compositional and conformational) complexes. The macromolecular complex preparation should contain as few contaminants as possible, at sufficient concentration to yield many particles in each cryoEM micrograph, and in a buffer composition well suited to cryoEM analysis. Certain buffer constituents, including sucrose, glycerol and high (~> 350 mM concentrations of salts, depending on the sample size, properties and other buffer constituents) can interfere with the process of vitrification or reduce the signal-to-noise ratio in images, hindering structure determination16.
Typically, as a minimum, size exclusion chromatography (SEC) and SDS- PAGE gel analysis should be used to assess sample purity17, 18, but circular dichroism, functional assays, SEC coupled with multi-angle light scattering, and thermal stability assays are all useful tools for qualitative analysis of macromolecular complex preparations prior to cryoEM analysis. However, the results from these biochemical analyses may yield little insight into the structural heterogeneity of the sample and its behaviour on a cryoEM grid. For this reason, negative stain EM is routinely used as a quick, cheap and powerful tool for assessing compositional and conformational heterogeneity, and therefore a good way of ascertaining which elution fraction from a purification is most promising, or screening different buffer compositions19, 20. Once a promising sample has been identified, we can proceed to the SPA cryoEM pipeline. Negative stain does not always align with the subsequent results seen in cryoEM; sometimes a sample looks poor by negative stain but improves when seen in vitreous ice in cryoEM. In contrast, sometimes samples look excellent during negative stain steps but require significant further optimisation when progressing to cryoEM. However, in the majority of cases negative stain provides a useful quality control step.
Vitrification
The harsh environment within the vacuum system of the electron microscope causes both dehydration and radiation damage to unfixed biological specimens21. Therefore, to image the sample in a native-like state, the biological specimen must be preserved prior to imaging. For purified preparations of macromolecular complexes, vitrification is the method of choice to enable its visualization by cryoEM whilst preserving the atomic details of the complex. The discovery of vitrification as a method of sample preparation was a fundamental advancement in electron microscopy of biological specimens, for which Dubochet was recognized in the 2017 Nobel Prize in Chemistry. Sample vitrification involves creating a thin layer of solution containing the specimen of interest, typically tens of nm thick, suspended on a cryoEM grid support. The thin film is then frozen extremely rapidly in a cryogen such as liquid ethane at ~-175 °C. The freezing rate is ~106 °C/s, fast enough that amorphous, or vitreous ice forms, suspending the specimen in a thin, solid film22.
The initial variable to consider is the cryoEM grid support chosen23. An EM grid typically consists of an amorphous carbon film with perforations (either regular or irregular), over a support structure. The support structure is typically a circular metal grid 3.05 mm in diameter, usually made from copper, but other metals such as gold, or molybdenum (which has preferred thermal expansion properties24) can be used. Sometimes, an additional thin, continuous support is applied across the grid, such as graphene, graphene oxide or a thin (~1-2 nm) amorphous carbon layer. While standard cryoEM grids (most commonly 400-200 mesh copper with a perforated (1.2 µm round holes separated by 1.3 µm (r1.2/1.3), or 2 µm separated by 2 µm of carbon (r2/2)) carbon support- although many different patterns are available) have been used in the vast majority of structures reported to date, novel grid technologies with improved conductivity and reduced specimen movement have been reported25. Selected grids are subjected to a glow-discharge/plasma cleaning treatment to render them hydrophilic and amenable to sample application26.
Following glow-discharge, the next stage is thin film formation. This thin film is most commonly formed using filter paper to remove excess liquid from the grid. While this can be carried out manually, a number of plunge freezing devices are commercially available, including the Vitrobot Mk IV (Thermo Fisher Scientific), EM GP II (Leica) and CP3 (Gatan). With these devices, ~3-5 µL of sample in solution is applied to the EM grid, followed by blotting away excess solution using filter paper. The grid, with a thin film suspended across it, is then plunged into liquid ethane cooled by liquid nitrogen (LN2) to ~-175 °C. Once frozen, the grid is maintained at a temperature below the devitrification point (-137 °C) prior to, and during imaging.
Specimen screening and data collection
Following vitrification of a cryoEM grid, the next stage is to screen the grid to assess its quality and to determine whether the grid is suitable to proceed to high resolution data collection. An ideal cryoEM grid has vitreous ice (opposed to crystalline ice) with the ice thickness just sufficient to accommodate the longest dimension of the specimen, ensuring the surrounding ice contributes as little noise to the resulting image as possible. The particles within the ice should have a size and (if known) shape consistent with biochemistry, and ideally be monodisperse with a random distribution of particle orientations. Finally, the grid should have enough areas of sufficient quality to satisfy the desired data collection length. Depending on the specimen, this may take many iterations of vitrification and screening until optimal grids are produced. Both fortunately and unfortunately, there are a huge range of variables that can be empirically tested to alter particle distribution on cryoEM grids (reviewed in16,27). In this manuscript, representative results for a membrane protein project10 are shown.
Once a suitable grid has been identified, data collection can proceed. Several models of cryo-transmission electron microscopes for biological specimens are optimized to collect high-resolution data in an automated fashion. Typically, data is collected on 300 kV or 200 kV systems. Automated data collection can be achieved using software including EPU (Thermo Fisher Scientific)28, Leginon29, JADAS30 and SerialEM31, 32. An automated data collection with modern detectors typically results in terabytes (TB) of raw data in a 24 h period (average datasets are ~ 4 TB in size).
Due to the COVID-19 restrictions in place on much of the world (time of writing December 2020), many microscopy facilities have moved to offering remote access. Once the grids have been loaded into the autoloader of a microscope, data acquisition can be conducted remotely.
Image processing and model building
Where a data collection session may be typically 0.5-4 days, subsequent image processing may take many weeks and months, depending on the availability of computing resources. It is standard for initial image processing steps, namely motion correction and contrast transfer function (CTF) estimation to take place 'on the fly' 33, 34. For downstream processing, there are a plethora of software suites available. Particles are 'picked' and extracted from micrographs35, 36. Once particles are extracted, a standard protocol would be to process the particles through several rounds of classification (in both two dimensions (2D) and three dimensions (3D) and/or focused on specific regions of interest) to reach a homogeneous subset of particles. This homogeneous subset of particles is then averaged together to produce a 3D reconstruction. At this point data is often corrected further to produce the highest quality map possible, for example through CTF refinement, distortion corrections37 and Bayesian polishing38 . The outcome of this image processing is a 3D cryoEM map of the biological specimen of interest. The resolution range reached in a 'standard' automated single particle experiment from a grid of sufficient quality, with data collected on a 300 kV microscope system is typically between 10 Å and 2 Å depending on the size and flexibility of the protein complex. With an ideal specimen, resolutions of ~1.2 Å have now been reached using SPA workflows5. While this protocol details steps towards obtaining an EM density map, once this is in hand it can be further interpreted through fitting and refining a protein model (if resolution is < 3.5 Å) or building de novo39. Data associated with structure determination experiments can be deposited into online public repositories, including EM density maps (Electron Microscopy Data Bank)40, resulting atomic coordinates (Protein Data Bank)41 and raw datasets (Electron Microscopy Public Image Archive)42.
In this protocol, the outer-membrane protein complex RagAB (~340 kDa) from Porphyromonas gingivalis is used as an example macromolecular complex10 (EMPIAR-10543). For those new to cryoEM, support for samples through this pipeline from sample to structure is available, subject to peer review, through funded access schemes such as iNEXT Discovery and Instruct.
1. Grid Vitrification
NOTE: For all steps in steps 1 and 2, ensure that all tools are clean, dry and at room temperature before cooling them to LN2 temperature, using freshly decanted LN2 to reduce ice contamination. Where possible, work within a humidity-controlled environment with < 20% relative humidity. Ensure appropriate personal protective equipment and H&S documentation is in place before commencing work.
2. Clipping grids for loading into an autoloader microscope
3. Secure remote log in to microscopes
NOTE: With COVID-19 controls at the time of writing, but also with environmental concerns associated with international travel, more microscopy facilities have been offering services where the user operates remotely. The method of implementation for this will vary according to the local IT configuration of each facility, and the needs of its internal and external user community. Here the process for remotely accessing cryoEMs at eBIC and controlling the microscope through EPU software is described.
4. Loading samples into an autoloader microscope and screening for ice and sample quality
NOTE: In this section a microscope with an autoloader and EPU software is used for sample screening, but this can be achieved using other software and a side entry system and cryoEMs from other manufacturers.
5. Single particle cryoEM data collection (with a focus on remote operation)
NOTE: A detailed protocol for data acquisition with EPU is described in the manufacturers manual and elsewhere28. Here modifications of this protocol for remote operation (namely reducing use of the hand panels to conduct tasks and using software-based alternatives) are highlighted.
6. Image processing to yield EM density map
NOTE: The majority of cryoEM facilities offer pre-processing of micrograph movies 'on the fly'. There are a wide variety of software packages and approaches available for this including RELION pipelines28,33, cryoSPARC43, Scipion34 and WarpEM44. A RELION based pipeline is described here and it is assumed that the user has moved the micrograph movies to an appropriate storage location with access to computing resources. An overview of the process and representative results for a membrane protein project are provided, a detailed description and step by step tutorial can be found on the RELION homepage: https://www3.mrc-lmb.cam.ac.uk/relion.
When screening, grids can be discarded at the atlas stage, where features resolved at low magnification mark the grid as not suitable for data acquisition. For example, if a grid has been subject to significant mechanical damage with the majority of grid squares broken (Figure 2A), or where the grid appears to be 'dry', with no vitreous ice (Figure 2B). Such grids are typically identifiable as the edges of the grid squares appear sharp and distinct. Across the majority of grids made using the plunge freezing device, a gradient of ice is observed (Figure 2C,D). Particle distribution, depending on the specimen of interest, can vary dramatically with ice thickness and so screening a range of grid squares to assess particle distribution is recommended. Tools have been implemented within EPU software during the atlas screening step to help the user identify grid squares of similar or different ice thickness, which can be particularly useful to users who are new to examining cryoEM grids (Figure 2E, F).
Figure 2: Example low magnification 'atlas' montages from screening sessions. A) A grid which has suffered significant damage with the majority of grid squares broken – unsuitable for collection. B) A dry grid with no vitreous ice – unsuitable for collection. C) A grid demonstrating an ice gradient with ~ 50% of the grid useable. D) An ice gradient with ~ 33% of the grid useable. Both C and D, are suitable for data collection if the usable grid squares have an ice thickness appropriate for collection, and there are enough acquisition areas to satisfy the minimum duration of a collection (e.g., 24 h) E) An example atlas with range of ice thicknesses. F) The same atlas presented in E but with, grid squares categorized and coloured by EPU software according to ice thickness. Please click here to view a larger version of this figure.
When screening particle distribution, ensure that imaging parameters, such as magnification and total electron dose, are similar to those expected to be used during data acquisition in order to provide an accurate picture of expected results. During screening, an ideal particle distribution is monodisperse with a range of particle orientations visible (depending on the specimen and existing knowledge of the particle's morphology, this may be challenging to ascertain) (Figure 3A). The ice should be as thin as possible while accommodating the particles largest dimension, if ice is too thin it can melt when illuminated with the electron beam. This causes excessive motion in the micrograph, and areas that display this characteristic should be avoided (Figure 3B). From collective experience, this effect is most commonly observed when there is detergent in the buffer. This can result in very thin ice at the centre of the hole and so particles can be physically excluded and forced towards the edge. This effect is observed in Figure 3C, but in this case it is not an extreme example and these images would still usefully contribute to a dataset. Finally, the ice needs to be vitreous; exclude any areas of the grid (or grids) where the majority or all of the images taken show crystalline ice (Figure 3D) from data acquisition. Often, non-vitreous ice is observed at the edge of grid squares. Readers are referred to detailed reviews of the variables that can be altered during grid vitrification16 and descriptions of particle behaviour in the thin film environment46, 47 for further information.
Figure 3: Representative micrographs showing differing particle distributions. A) An 'ideal' distribution of monodisperse particles adopting a range of orientations. B) Overly thin ice in the middle of the hole that it deforms upon exposure to the electron beam causing excessive motion in the micrograph. This effect is most often observed when detergent is present in the buffer C) Where ice is thinner in the centre of the hole, this physically excludes particles from the centre, causing crowding of particles towards the hole edge. In this case it is not extreme enough to prevent these images being useful, but it suggests it is worth screening slightly thicker areas. D) Ice is not vitreous, data should not be collected on areas which look like this example micrograph. Please click here to view a larger version of this figure.
On-the-fly image processing can help to pick up errors and problems with data acquisition and so is always recommended where possible. For example, excessive motion within micrographs may indicate that the autoloader turbo pump is active, or data is being collected on a cracked grid square where ice is moving significantly in the electron beam, indicating the grid square should be skipped. On the fly CTF estimation can reveal circumstances where a positive focus point (rather than defocus) is applied (where CTF estimation programs and parameters to find these points are used), and determine the phase shift where a Volta phase plate48 is used. On the fly image processing pipelines often include a graphical summary of the data (Figure 4A) to make it easier for users to assess micrograph quality quickly and decide if data collection amendments are required.
Selection of particles from micrographs, whilst avoiding 'false positives' such as contamination or the grid support film can require optimisation. However, particle pickers such as crYOLO often work sufficiently well using default parameters for a 'first pass' of the data (Figure 4B), enabling progression to 2D class averaging where it can be easier to assess the quality of the data and the likelihood of downstream success. For most projects, 2D classification of ~> 10k particles should start to reveal classes which have secondary structure detail. To proceed to 3D, the 2D classification stage should typically reveal classes representing a range of particle orientations. If a preferred orientation is revealed, more iterations of sample preparation16 or further data acquisition with the sample tilted may be required49. All classes which show secondary structure detail should be chosen to take forward to 3D analysis, while 'junk' particles are discarded (Figure 4C).
Figure 4: Initial image processing steps. A) Output from an 'on the fly' image processing script. B) Example micrograph (left) with appropriately auto-picked particles identified using the crYOLO general model (right, with particles bounded by red squares) Scale bars (white) are 50 nm. C) Results from 2D classification showing classes which were discarded in the red square, and classes from which particles were selected for further processing in green. Please click here to view a larger version of this figure.
A small subset of particles can be used to generate an initial model (Figure 5A). This initial model can then be used as a starting model in 3D classification and refinement. In the case of RagAB, the dataset contained three distinct conformers which can be separated during 3D classification (Figure 5B). Particles contributing to each of these classes can then be treated independently and used to refine an EM density map which can then be subject to further interpretation and model building.
Figure 5: Generating 3D EM density map. A) Typical initial model generated using RELION. B) 3D classification over 5 classes showing separation of particles into three distinct conformational states: open-open (green), open-closed (blue), closed-closed (purple). C) Process of mask creation. The map from 3D refinement (left) should be visualized in chimera. The volume viewer can then be used to identify the lowest threshold at which the map is free from disjointed, noisy density (middle). This threshold value is input as initial binarization threshold in the RELION Mask creation job. An example mask output is shown in grey (right). D) High resolution EM density map of the open-closed state of RagAB (EMD-10245), filtered and colored by local resolution (Å). Please click here to view a larger version of this figure.
In this protocol we have described a basic pipeline applicable to specimens amenable to routine SPA. While this filter paper blotting method of thin film formation and vitrification is undoubtedly successful given its use in the vast majority of SPA projects to date, it comes with a number of disadvantages. These include sample wastage, the slow timescales (seconds) required to form the thin film and freeze the specimen, reported irreproducibility27 and reported negative effects of using filter paper to blot away excess liquid50. Recently, new technologies have been developed to improve reproducibility of thin film production51, 52. Other technologies have been developed which reduce the time between sample application and vitrification53,54,55. While filter paper-based methods for thin film formation remain most ubiquitous method of SPA cryoEM sample preparation at the time of writing, these new technologies may bring a range of benefits in terms of efficiency and reproducibility of grid vitrification, as well as creating new opportunities to bring in additional experimental dimensions, such as time resolution and rapid mixing prior to vitrification.
The process of grid screening for most users is presently a qualitative process which involves the acquisition of low magnification atlases followed by taking high-magnification images across the grid to assess particle distribution. While this is a sufficiently robust approach for some types of specimen, it can be difficult to assess by eye if the specimen is indeed what the researcher is hoping to image or has a preferred orientation, for example with small (<200 kDa) samples or where the low-resolution morphology makes it hard to identify by eye if a range of particle distributions are present. For some projects, it is impossible to determine if the specimen is as desired, for example where a ligand is bound or where the sample is being screened to assess if a small (e.g., 10 kDa) subunit is still present in association with a complex. For these projects, fully automated pipelines for data analysis combined with a 'short' 0.5 – 1-h collections, that can proceed through image processing steps to 2D classification or even 3D classification and refinement would help efficiently determine if a longer collection is warranted. These pipelines are still under development and are not widely implemented at present, but they have the potential to improve the efficiency of cryoEM grid screening, especially for challenging specimens.
Improvements in direct electron detectors, as well as modifications in microscopy combined with advances in image processing such as image shift data collection, have increased the throughput and quality of images produced during data acquisition. This increase in the rate of data being collected highlights the need for thorough screening of cryoEM grids ahead of many TB of data being acquired.
CryoEM SPA has become a truly mainstream structural biology technique, and in many cases the 'go to' approach for some classes of specimens, such as heterogeneous and labile macromolecular complexes. While the protocol here describes a basic overview of the SPA pipeline, each section covered here (grid vitrification and screening, cryoEM and image processing) is a topic in its own right and worthy of exploration during the development of an SPA project. As sample preparation and microscopy technologies progress, and new image processing algorithms and approaches come online, SPA will continue to develop as a pipeline, assisting researchers in gaining insight into complex biological systems.
The authors have nothing to disclose.
This work was supported by the iNEXT-Discovery (Grant 871037) funded by Horizon 2020 program of the European Commission. J B. R. White is funded by the Wellcome Trust (215064/Z/18/Z). The FEI Titan Krios microscopes were funded by the University of Leeds (UoL ABSL award) and Wellcome Trust (108466/Z/15/Z). We thank M Iadanza for use of his micrograph analysis script. We acknowledge Diamond Light Source for access and support of the cryo-EM facilities at the UK's national Electron Bio-imaging Centre (eBIC) funded by the Wellcome Trust, MRC and BBRSC.
Blunt tweezers | Agar Scientific | AGT5022 | |
Cryo EM round storage box | Agar Scientific | AGG3736 | |
CryoEM autogrid boxes | ThermoFisher Scientific | 1084591 | |
CryoEM grids | Quantifoil | N1-C14nCu30-01 | |
Ethane gas | Boc | 270595-F | |
LN2 foam dewar | Agar Scientific | AG81760-500 | |
LN2 storage dewar | Worthington industries | HC 34 | |
Pipette | Gilson | 10082012 | |
Pipette tips | Star labs | s1111-1706 | |
Syringe | BD | BD 300869 | |
Type II lab water | Suez | select fusion | |
Vitrobot | ThermoFisher Scientific | 1086439 | |
Vitrobot filter paper | Whatman | 1001-055 | |
Vitrobot styrophome container assembly | ThermoFisher Scientific | 1086439 | |
Vitrobot tweesers | ThermoFisher Scientific | 72882-D | |
Software | |||
EPU | ThermoFisher Scientific | 2.8.1.10REL | |
TEM server | ThermoFisher Scientific | 6.15.3.22415REL | |
Tia | ThermoFisher Scientific | 5.0.0.2896REL | |
Titan krios microscope | ThermoFisher Scientific | Titan Krios G2 |