The facility is an HHMI-wide shared facility with state-of-the-art instrumentation and strong staff expertise dedicated to structural biology, offering services to cover the entire CryoEM workflow. There are two primary types of work the facility does for users.
1. High-quality high-throughput data collection on ready-to-image cryo grids prepared and provided by users with CryoEM experience
In this case, the facility does not help calibrate samples, prepare grids, or process data. Users are required to supply CryoEM images from their cryo grids beforehand to qualify for data collection time on one of the two Janelia Titan Krios microscopes. The purpose of this requirement is to ensure that quality of cryo grids and samples are of acceptable quality, to use microscope capacity at the facility most responsibly. The user can either come to Janelia or work remotely. In either case, facility staff will operate the microscope, optimize imaging conditions/parameters and set up automated data collection while users can focus on evaluating and selecting targets for data collection.
2. A lab can provide a sample in solution
In this case, the facility calibrates the sample and carries out most or all of the CryoEM-related work on suitable samples, from cryo grid preparation/screening, to large volume data collection, to image processing and 3D reconstruction. Facility staff will interact closely with lab members and the cryo grid preparation step may take multiple rounds of iteration. This type of work is research-oriented by nature and should be considered as a collaboration between the lab and the facility.
Data Collection
For cryoEM work, high quality high throughput data collection under optimized conditions is arguably the most critical step with obtaining good sample being a prerequisite for any possible success. Therefore, the primary function of the cryoEM facility was initially set to be data collection. To ensure that data is collected under optimal conditions with efficient scheme and high throughput, the cryoEM facility staff operates the microscope, optimizes imaging conditions and sets up automated data collection for outside users. This relieves users from all the setup work which can be difficult for a general user as modern electron microscopes becomes more and more complicated and sophisticated. As a result, they can concentrate on checking the quality of the sample and identifying good areas for data collection.
Depending on the sample and experimental goal, data collection needs to be done with different methods and imaging conditions/parameters. The two primary data collection methods widely used by users and fully supported by the facility are single particle analysis and electron cryo-tomography. Single particle analysis is an averaging technique and applies to structurally homogenous samples. Electron tomography is typically used on structurally inhomogeneous/unique biological samples whose three-dimensional structure can not be elucidated by other structural biology methods.
Sample preparation
As a core facility to serve the entire HHMI community, we strike to help as many HHMI labs as possible. We have expanded our services to cover the entire cryoEM work flow from sample preparation, to data collection to image processing and 3D reconstruction.
With the hardware/software setup, staff experiences/knowledge and optimized operation at the facility, the most important deciding factor for a cryoEM project to succeed is the sample itself, including both intrinsic properties of the sample and qualities of cryo grids. For single particle work the important intrinsic properties of the sample are structural homogeneity and rigidity, sample size and symmetry. Only if the sample is structurally homogenous can averaging method used in single particle analysis produces good result. A larger sample typically produces more signal/contrast in an image. The presence of symmetry provides more units for averaging.
For cryoEM the sample needs to be plunge frozen on a cryoEM grid typically covered with a holey carbon film. Quality of the cryo grid such as ice thickness, particle distribution, absence of contaminants and stability of the support film will have significant impacts on both the quality and throughput of imaging. Therefore, it is critical to optimize sample preparation even though it may be a tedious process. It would be a mistake to shortcut this critical step and collect data on a suboptimal grid.
Preparing suitable cryoEM grids takes experiences and dedication. Staff at the cryoEM facility have rich experiences working with a wide variety of biological samples and are determined to spend time and efforts to find the optimal freezing conditions. The freezing conditions can be different for different samples. It usually takes multiple rounds of trials and errors to find the best freezing condition for a new sample.
For cryo grids preparation users only need to send in purified sample in solution either frozen on dry ice or on ice pack depending on the stability of the sample. We will calibrate the sample first with negative staining EM method. If the sample look promising, then we will test freeze cryo grids, screen the grids on the FEI cryo T12 microscope and go back to freezing grids under different conditions if necessary till optimal conditions are found. Then multiple cryo grids will be prepared under the optimal conditions and stored in liquid nitrogen for data collection on one of the two Titan Krios microscopes.
Image Processing and 3D reconstruction
We do image processing for both tomography and single particle analysis. Image processing and 3D reconstruction for tomography is typically simpler and quicker than single particle work. So we mainly devote this section to a very simplified overview of image processing for single particle analysis which is a complicated process involving many intermediate steps. Some steps are time consuming and computation intensive. There are different software packages available dedicated to various steps of the process. We have set up dedicated hardware and software and implemented efficient protocols to carry out the entire workflow of image processing and 3D reconstruction.
The first major step of image processing for single particle is the so-called motion correction. With direct electron detectors each data point is typically a movie stack of many short-exposure frames from the same area of sample. There are motions of the sample between these frames which will introduce a blurring effect and limit resolution if the frames are simply summed up. Therefore, frames (or even individual particles) in a movie stack need to be aligned to get rid of the motion-induced blurring effect first. After motion is corrected certain frames in a movie stack can be summed up for better signal/noise. These need to be done for all the movie stacks collected. In the end only the sum images will be used for later steps.
After motion correction the next step is contrast transfer function (CTF) correction. The images obtained on an electron microscope are inherently modulated by the optical settings of the microscope. CTF describes how an electron microscope modify the sample signal in the Fourier space and its influence heavily depends on the optical settings. CTF correction fits the power spectrum of each image with a theoretical model to find various optical parameters (such as defocus and astigmatism) for the image. These parameters will be taken into account during later processing steps.
The next step is boxing out particles from all the images collected. Typically, a few thousands of particles need to be hand picked first to generate reference/template for automated picking. Depending on sample features, distribution and cleanness of areas imaged, the results of automated picking might or might not be satisfactory and inspections of the picked particles might be needed to discard bad or wrongly picked ones.
Then the selected particles need to go through the so-called 2D classification. Each particle is compared to the rest to see if they look similar and belong to the same class. Particles within the same classes are averaged to generate class averages with improved signal to noise ratio, overcoming the inherent low signal to noise ratio in individual images. After 2D classification and the building of a low resolution 3D initial model (either from a known relevant structure or from the 2D class averages), 3D classification and refinement are carried out to generate a 3D reconstruction. Although the actual algorithm is different for different software packages on 2D classification, 3D reconstruction and refinement, most programs carry out these step in an iterative fashion. These steps are computation intensive, time consuming and might take multiple runs. Supervision with caution from human operator is usually needed.