Protein Crystallization & X-ray Crystallography Outsourcing Services Company

X-ray Crystallographic Structure Determination Workflow

To streamline and maximize the speed and efficiency of our services, our staff follows specific protocols throughout all crystallization and protein X-ray structure determination experiments. To ensure the quality of client samples, we have also developed general guidelines that you can download for your convenience. A product sheet for one of our crystallization-grade Protein Shop proteins demonstrates the quality required for crystallization-grade proteins.

Below is an overview of the stages of protein X-ray structure determination. On our educational site, you can find more information on the basics of X-ray crystallography.

Overview of X-ray Crystallography

More than 100 years have passed since William H. Bragg and William L. Bragg (father and son) explained the diffraction patterns obtained by irradiating crystals with X-rays and derived Bragg’s law. A few years later, in 1915, they were awarded the Nobel Prize in Physics. In 1926, the first crystals of an enzyme, urease, were reported by James B. Sumner, which can be considered the starting point for structural biology.

For many years, low-intensity laboratory X-ray sources available during the early stages of protein X-ray crystallography made collecting high-quality diffraction data very slow and, for small, weakly diffracting crystals, entirely impossible. Much time had to be spent optimizing crystallization conditions to obtain larger crystals suitable for the X-ray sources of that time. However, since the mid-1990s, with the introduction of synchrotrons, enormous advances have fundamentally changed how we do X-ray crystallography. With modern 4th-generation synchrotrons capable of generating high-brilliance microfocus beams, advanced robotics and automation, fast detectors, and high-speed diffraction data processing, high-throughput X-ray crystallography has become routine. Today, we can determine protein structures using much smaller crystals more efficiently, much faster, and at a considerably lower cost.

The process of protein structure determination by X-ray crystallography includes five main stages:

• Crystallization (initial screening and crystal hit optimization)
• Crystal harvesting, testing, and X-ray data collection
• Obtaining phases, calculating the electron density map, and building the structural model
• Refinement and overall quality assessment against Protein Data Bank (PDB) criteria
• In the case of a ligand complex, we need to apply specific criteria to ligand-protein complex structure validation and quality assessment

1- Crystal Testing & X-ray Diffraction Data Collection. The “Synchrotron-Only” Approach to Data Collection

In an earlier blog post, we discussed how AstraZeneca’s team decided to switch from combining a laboratory X-ray source (for preliminary crystal tests) and synchrotron radiation for data collection to a “synchrotron-only” approach. At SARomics Biostructures, both crystal testing and diffraction data collection are primarily conducted at the protein crystallography beamlines BioMax and MicroMax at the 4th-generation synchrotron MAX IV Laboratory in Lund. For data collection and processing, we use the crystallographic software packages XDS and autoPROC (Global Phasing). With a synchrotron nearby, we enjoy quick access to beamlines, eliminating the need to ship crystals over long distances to remote synchrotrons and the risk of crystal damage during transport. In a note below, we discuss the benefits of synchrotron X-ray crystallography.

Generally, to minimize radiation damage from modern high-intensity synchrotron beams, data collection must be conducted at cryogenic temperatures. To this end, after harvesting from drops, crystals are transferred to a cryo-solution and flash-frozen at liquid nitrogen temperature. The cryo-solution prevents ice formation around a crystal when it is mounted in a special loop before insertion into an X-ray beam. Ice interferes with data collection (ice crystals scatter X-rays) and may also damage the protein crystals, resulting in poor data quality. However, as discussed in our blog article, there are instances in which room-temperature protein crystallography can be advantageous.

2- Phase Determination and Electron Density Calculation

Calculating an electron density map, into which the structural model will be built, requires phase information. X-ray detectors register only the intensities of the diffracted waves; the phase information is lost during this process and must be retrieved using other methods. The method of heavy-atom replacement for solving the phase problem was first developed by J. C. Kendrew et al., who reported the structure of myoglobin in 1958, and M. F. Perutz et al., who published the structure of hemoglobin in 1960. Both scientists were awarded the Nobel Prize in Chemistry in 1962.

Currently, heavy-atom replacement is rarely used. Instead, molecular replacement is the most popular method. Molecular replacement requires an experimental 3D structure of the protein or a close homolog (typically with sequence identity above 40-50%) to calculate initial phases and an electron density map. At times, computer-modeled structures (e.g., from AlphaFold) may also be used to solve the phase problem. However, this requires that the conformation of the computed structure (especially for multidomain proteins) match that of the crystal structure being studied. If molecular replacement fails, we must use de novo structure determination. For this, typically, a selenomethionine (SeMet)-labeled protein sample, which we can produce upon the client’s request, enables structure determination using synchrotron radiation and anomalous X-ray scattering. Crystallographic software packages Phaser and Molrep are typically used to phase macromolecular crystal structures using molecular replacement and single-wavelength anomalous dispersion.

Once the phases are available, researchers can calculate an electron density map that reveals the positions of atoms in the crystal. At this stage, an atomic model of the structure can be built, e.g., using the protein’s amino acid sequence.

3. Model Building, Refinement, and Model Overall Quality Assessment

The initial model built into the electron density map must be examined and further adjusted to better fit the experimental data. This process is called refinement. An improved model provides better phases and allows calculation of a better map that may reveal new details (e.g., missing loop density, missing density for parts of a ligand, or new solvent molecules). The crystallographic R-factor and R-free monitor refinement progress and assess the overall agreement between the model and the experimental data. However, we should keep in mind that the resolution of the X-ray data ultimately determines both the quality and level of detail in the interpretation of the structure. Crystallographic software such as REFMAC (CCP4) and BUSTER (Global Phasing), along with the graphics program Coot (CCP4), are used for model building and refinement.

Generally, a higher resolution allows more reliable model interpretation, whereas for lower-resolution models (e.g., under 2.5 Å), the interpretation of the electron density map may be more subjective. However, higher resolution does not automatically guarantee good model quality. Many high-resolution structures may have poor overall quality. For this reason, in addition to the R-factor and resolution, the PDB provides other factors for quality assessment. They include the Ramachandran plot and deviations of bond lengths and bond angles from their ideal values, collectively referred to as overall geometry (stereochemistry) criteria. Our educational site provides a detailed discussion on X-ray crystallographic structure validation and quality criteria assessment.

4. Ligand-Protein Complex Structure Validation

The correct interpretation of the electron density observed in a protein’s binding site can sometimes be challenging. Many cases have shown that a density at the binding site, initially interpreted as a bound ligand, was actually due to water molecules, buffer components, or precipitant molecules. Therefore, it is essential to validate the quality of the atomic model of the ligand-protein complex. This validation is crucial for all subsequent steps, especially in computational methods used in structure-based drug discovery projects, which depend heavily on the overall quality of the 3D structure. Global statistics, such as resolution, the R-factor, and model geometry discussed above, ensure that the structure is well-refined and meets the general quality criteria set by the PDB. However, even well-refined protein structures may contain problematic regions with poor electron density, resulting in poor local model quality. In the case of a bound ligand (small molecule), poor electron density may result from partial occupancy of the binding site as well as conformational flexibility of the ligand in the crystal. Therefore, it is important to inspect the bound ligand using so-called real-space (in contrast to reciprocal space) quality indicators. Below we discuss some of the details.

1- While building the ligand into the electron density, we need to ensure that the electron density covers all ligand atoms and that no unaccounted-for density remains within the binding site. Unaccounted-for density, e.g., when running crystallographic fragment screening, may indicate that the bound molecule is not what we expected to find there.

2- The fit of the model to the electron density can be assessed by examining the Real Space Correlation Coefficient (RSCC) and the Real Space R-Value (RSR). These values are calculated per residue and can be obtained from the PDB validation report. RSCC is generally higher for higher-resolution and lower-B-factor models – a perfect fit of the model to the electron density would yield an RSCC of 1.0. The lower the RSCC, the greater the uncertainty in the ligand model’s position relative to the electron density. Generally, values between 0.8 and 1.0 are considered satisfactory.

3- On some occasions, model bias may also result in spurious electron density, which is mistakenly interpreted as evidence of ligand presence. If in doubt, we can verify the presence of the ligand by first removing it from the coordinate file, running a few cycles of refinement, and then calculating a so-called omit electron density map (a map with the ligand molecule omitted from the coordinate file). If the ligand is present in the binding site at a high occupancy, an omit map will clearly reveal its density.

4- In addition, we need to examine the interactions between the ligand and the amino acid side chains within the protein-binding site. For example, chemically unrealistic interactions or atom-atom distances that are too short reveal potential problems with the ligand’s positioning.

5- We also need to ensure that the ligand model complies with the expected stereochemical parameters of the chemical structure.

6- The atomic displacement factors (B-factors or temperature factors) of the ligand should be compared to the average B-factors of the protein atoms within the binding site. Too-high ligand temperature factors should also alert us to potential problems.

A detailed discussion of ligand validation in macromolecular structures determined by X-ray crystallography in the PDB can be found in the excellent papers by Smart et al. (2018) and Pozharski et al. (2017).