Research areas in the Snell Laboratory

In a series of publications associated with the International Year of Crystallography, the Snell Laboratory discussed advancements in crystallization techniques and the construction of chemical approaches for obtaining crystals using pre-prepared crystallization screens. They also detailed the valuable information gained from visual outcomes of crystallization experiments and various methods for measuring these outcomes. The laboratory has studied crystallization across volume extremes, developing methods to optimize volumes for neutron diffraction studies and to minimize volumes for modern X-ray sources such as X-ray Free Electron Lasers. Proximity to the National Crystallization Center has provided numerous opportunities for extensive research in this field.

Uses of crystals have exploded and different techniques require different volume of crystals, from large for neutron studies, on the order of 10s of micros for synchrotron X-ray work, a few hundred nanometers for micro-electron diffraction, and down to the single molecule for small angle scattering or single particle cryo-electron microscopy. The laboratory works on understanding how to work with this entire range.

The Snell Laboratory has also been a pioneer in integrating machine learning and artificial intelligence technologies, creating one of the first comprehensive training sets for the automated classification of crystallization outcomes. This collaborative effort, involving research institutes, academia, government laboratories, the pharmaceutical industry, and GoogleBrain, resulted in a successful algorithm still used for classifying outcomes today. The potential predictive capabilities were highlighted in a review of computational crystallization approaches. Additionally, their investment in metadata capture during the Protein Structure Initiative led to early studies identifying physicochemical features of different crystallization mechanisms.

Crystallization screening involves diverse chemistries, often poorly captured in the literature and challenging to relate across multiple condition screens. In collaboration with the Collaborative Crystallization Centre in Australia and others, the Snell Laboratory helped establish the need for a consistent definition of crystallization components for effective machine learning. They demonstrated that identifying chemically similar and diverse conditions could be combined with outcome classification to pinpoint different optimization regions and mechanistic functions of targets.

The Snell Laboratory has also been a global leader in studying the impact of gravity on crystallization. Growing crystals in orbiting spacecraft reduces acceleration and minimizes convection over the crystal surface during growth. They developed sophisticated techniques to analyze the physical perfection of these space-grown crystals, discovering that while long-scale perfection (mosaicity) improves, short-scale perfection (atom-to-atom repeats) is strongly influenced by forces such as Brownian motion. This finding has implications for growing larger, more perfect crystals for neutron diffraction studies, although the benefit of a microgravity environment may be limited for improving short-scale quality with advancing X-ray sources.

Education is another area of significant interest for the laboratory. In an initiative designed to teach the scientific method to undergraduate students, various chemicals sourced from supermarkets were used as crystallization reagents with selected model proteins. Students developed hypotheses, made observations, and compared outcomes to existing literature. One notable case involved fish paste facilitating the crystallization of a target protein, which paralleled literature reports of crystallization using a specific amino acid. This experiment highlighted the protein’s selective crystallization contacts within the cocktail, providing a valuable teaching moment and a deeper understanding of the ‘silver-bullet’ approach to crystallization. Dr. Snell is actively involved in education, regularly teaching workshops around the world.

The Snell Laboratory has been analyzing the metal identity and content in metalloprotein samples used to obtain existing models in the Protein Data Bank (PDB). Techniques such as Particle Induced X-ray Emission (PIXE) and X-ray Fluorescence Spectroscopy (XRFS) at ion beams and synchrotron sources have been utilized. PIXE analysis of 87 metalloproteins revealed that 48% contained a metal in the model not present in the original protein sample. Of these, only 15% could be explained by promiscuous metals in the crystallization conditions, while 33% remained unexplained.

The laboratories X-ray fluorescence experiments at the Stanford Synchrotron Radiation Lightsource (SSRL), utilizing low X-ray absorption capillaries containing 400 nL of solution, with only 40 nL exposed to the beam. Remarkably, we successfully identified and measured metal content down to μM concentration. Leveraging 10 multiple exposures, we achieved a signal 8 times higher than noise for 1 μM Zn concentration. We further optimized automation for longer exposures and streamlined the process. Employing NIST trace standards for six individual metals and a mixture of four metals, we accurately determined metal identity and concentration down to 1 ppm for each sample.  The NIST traceable single metal solutions, initially at a concentration of 1000 μg/ml, were appropriately diluted for the experiments (1x, 10x, 100x, and 1000x). We summed a total of 10 scans for 1x and 10x dilutions, 50 scans for 100x, and 100 scans for 1000x dilutions. This sensitivity of 1 ppm proves adequate for both identifying and quantifying the metal, comparable to PIXE capabilities. However, the requirement for 100x scans per individual sample proved time-consuming, while sample changeover was inconvenient. Additionally, data processing was time-intensive, and utilizing a regular crystallography beamline not tailored for such studies resulted in data of lower quality than could be achieved on a beamline dedicated to X-ray fluorescence. We have adopted an alternative approach that is just as sensitive and can evaluate hundreds of samples at a time. Work on this is in progress.

Spectrum showing PIXE data collected on a log scale and the calculated fit to the experimental data. Kα and Kβ emission peaks are seen for metals present. From Grime et al., 2020.

Structural models with electron density maps of the metal sites of one chain in 3DCP. The original metals are shown on the left, and  the setals identified by PIXE on the right. An SO₄ ion is clearly resolved in the electron density. Chains A and B were similar. Accurately modeling the correct metal can reveal mechanistic information that was otherwise missed. From Grime et al., 2020