Deep residual networks for crystallography trained on synthetic data.

The use of artificial intelligence to process diffraction images is challenged by the need to assemble large and precisely designed training data sets. To address this, a codebase called Resonet was developed for synthesizing diffraction data and training residual neural networks on these data. Here, two per-pattern capabilities of Resonet are demonstrated: (i) interpretation of crystal resolution and (ii) identification of overlapping lattices. Resonet was tested across a compilation of diffraction images from synchrotron experiments and X-ray free-electron laser experiments. Crucially, these models readily execute on graphics processing units and can thus significantly outperform conventional algorithms. While Resonet is currently utilized to provide real-time feedback for macromolecular crystallography users at the Stanford Synchrotron Radiation Lightsource, its simple Python-based interface makes it easy to embed in other processing frameworks. This work highlights the utility of physics-based simulation for training deep neural networks and lays the groundwork for the development of additional models to enhance diffraction collection and analysis.

A capillary-based microfluidic device enables primary high-throughput room-temperature crystallographic screening.

A novel capillary-based microfluidic strategy to accelerate the process of small-molecule-compound screening by room-temperature X-ray crystallography using protein crystals is reported. The ultra-thin microfluidic devices are composed of a UV-curable polymer, patterned by cleanroom photolithography, and have nine capillary channels per chip. The chip was designed for ease of sample manipulation, sample stability and minimal X-ray background. 3D-printed frames and cassettes conforming to SBS standards are used to house the capillary chips, providing additional mechanical stability and compatibility with automated liquid- and sample-handling robotics. These devices enable an innovative in situ crystal-soaking screening workflow, akin to high-throughput compound screening, such that quantitative electron density maps sufficient to determine weak binding events are efficiently obtained. This work paves the way for adopting a room-temperature microfluidics-based sample delivery method at synchrotron sources to facilitate high-throughput protein-crystallography-based screening of compounds at high concentration with the aim of discovering novel binding events in an automated manner.

Tuning Catalytic Bias of Hydrogen Gas Producing Hydrogenases.

Hydrogenases display a wide range of catalytic rates and biases in reversible hydrogen gas oxidation catalysis. The interactions of the iron-sulfur-containing catalytic site with the local protein environment are thought to contribute to differences in catalytic reactivity, but this has not been demonstrated. The microbe Clostridium pasteurianum produces three [FeFe]-hydrogenases that differ in "catalytic bias" by exerting a disproportionate rate acceleration in one direction or the other that spans a remarkable 6 orders of magnitude. The combination of high-resolution structural work, biochemical analyses, and computational modeling indicates that protein secondary interactions directly influence the relative stabilization/destabilization of different oxidation states of the active site metal cluster. This selective stabilization or destabilization of oxidation states can preferentially promote hydrogen oxidation or proton reduction and represents a simple yet elegant model by which a protein catalytic site can confer catalytic bias.

The crystal structure of dGTPase reveals the molecular basis of dGTP selectivity.

Deoxynucleotide triphosphohydrolases (dNTPases) play a critical role in cellular survival and DNA replication through the proper maintenance of cellular dNTP pools. While the vast majority of these enzymes display broad activity toward canonical dNTPs, such as the dNTPase SAMHD1 that blocks reverse transcription of retroviruses in macrophages by maintaining dNTP pools at low levels, Escherichia coli (Ec)-dGTPase is the only known enzyme that specifically hydrolyzes dGTP. However, the mechanism behind dGTP selectivity is unclear. Here we present the free-, ligand (dGTP)- and inhibitor (GTP)-bound structures of hexameric Ec-dGTPase, including an X-ray free-electron laser structure of the free Ec-dGTPase enzyme to 3.2 Å. To obtain this structure, we developed a method that applied UV-fluorescence microscopy, video analysis, and highly automated goniometer-based instrumentation to map and rapidly position individual crystals randomly located on fixed target holders, resulting in the highest indexing rates observed for a serial femtosecond crystallography experiment. Our structures show a highly dynamic active site where conformational changes are coupled to substrate (dGTP), but not inhibitor binding, since GTP locks dGTPase in its apo- form. Moreover, despite no sequence homology, Ec-dGTPase and SAMHD1 share similar active-site and HD motif architectures; however, Ec-dGTPase residues at the end of the substrate-binding pocket mimic Watson-Crick interactions providing guanine base specificity, while a 7-Å cleft separates SAMHD1 residues from dNTP bases, abolishing nucleotide-type discrimination. Furthermore, the structures shed light on the mechanism by which long distance binding (25 Å) of single-stranded DNA in an allosteric site primes the active site by conformationally "opening" a tyrosine gate allowing enhanced substrate binding.

The Stanford Automated Mounter: pushing the limits of sample exchange at the SSRL macromolecular crystallography beamlines.

The Stanford Automated Mounter System, a system for mounting and dismounting cryo-cooled crystals, has been upgraded to increase the throughput of samples on the macromolecular crystallography beamlines at the Stanford Synchrotron Radiation Lightsource. This upgrade speeds up robot maneuvers, reduces the heating/drying cycles, pre-fetches samples and adds an air-knife to remove frost from the gripper arms. Sample pin exchange during automated crystal quality screening now takes about 25 s, five times faster than before this upgrade.

High-density grids for efficient data collection from multiple crystals.

Higher throughput methods to mount and collect data from multiple small and radiation-sensitive crystals are important to support challenging structural investigations using microfocus synchrotron beamlines. Furthermore, efficient sample-delivery methods are essential to carry out productive femtosecond crystallography experiments at X-ray free-electron laser (XFEL) sources such as the Linac Coherent Light Source (LCLS). To address these needs, a high-density sample grid useful as a scaffold for both crystal growth and diffraction data collection has been developed and utilized for efficient goniometer-based sample delivery at synchrotron and XFEL sources. A single grid contains 75 mounting ports and fits inside an SSRL cassette or uni-puck storage container. The use of grids with an SSRL cassette expands the cassette capacity up to 7200 samples. Grids may also be covered with a polymer film or sleeve for efficient room-temperature data collection from multiple samples. New automated routines have been incorporated into the Blu-Ice/DCSS experimental control system to support grids, including semi-automated grid alignment, fully automated positioning of grid ports, rastering and automated data collection. Specialized tools have been developed to support crystallization experiments on grids, including a universal adaptor, which allows grids to be filled by commercial liquid-handling robots, as well as incubation chambers, which support vapor-diffusion and lipidic cubic phase crystallization experiments. Experiments in which crystals were loaded into grids or grown on grids using liquid-handling robots and incubation chambers are described. Crystals were screened at LCLS-XPP and SSRL BL12-2 at room temperature and cryogenic temperatures.

Crystal structure of the pristine peroxidase ferryl center and its relevance to proton-coupled electron transfer.

The reaction of peroxides with peroxidases oxidizes the heme iron from Fe(III) to Fe(IV)=O and a porphyrin or aromatic side chain to a cationic radical. X-ray-generated hydrated electrons rapidly reduce Fe(IV), thereby requiring very short exposures using many crystals, and, even then, some reduction cannot be avoided. The new generation of X-ray free electron lasers capable of generating intense X-rays on the tenths of femtosecond time scale enables structure determination with no reduction or X-ray damage. Here, we report the 1.5-Å crystal structure of cytochrome c peroxidase (CCP) compound I (CmpI) using data obtained with the Stanford Linear Coherent Light Source (LCLS). This structure is consistent with previous structures. Of particular importance is the active site water structure that can mediate the proton transfer reactions required for both CmpI formation and reduction of Fe(IV)=O to Fe(III)-OH. The structures indicate that a water molecule is ideally positioned to shuttle protons between an iron-linked oxygen and the active site catalytic His. We therefore have carried out both computational and kinetic studies to probe the reduction of Fe(IV)=O. Kinetic solvent isotope experiments show that the transfer of a single proton is critical in the peroxidase rate-limiting step, which is very likely the proton-coupled reduction of Fe(IV)=O to Fe(III)-OH. We also find that the pKa of the catalytic His substantially increases in CmpI, indicating that this active site His is the source of the proton required in the reduction of Fe(IV)=O to Fe(IV)-OH.

Mapping the conformational landscape of a dynamic enzyme by multitemperature and XFEL crystallography.

Determining the interconverting conformations of dynamic proteins in atomic detail is a major challenge for structural biology. Conformational heterogeneity in the active site of the dynamic enzyme cyclophilin A (CypA) has been previously linked to its catalytic function, but the extent to which the different conformations of these residues are correlated is unclear. Here we compare the conformational ensembles of CypA by multitemperature synchrotron crystallography and fixed-target X-ray free-electron laser (XFEL) crystallography. The diffraction-before-destruction nature of XFEL experiments provides a radiation-damage-free view of the functionally important alternative conformations of CypA, confirming earlier synchrotron-based results. We monitored the temperature dependences of these alternative conformations with eight synchrotron datasets spanning 100-310 K. Multiconformer models show that many alternative conformations in CypA are populated only at 240 K and above, yet others remain populated or become populated at 180 K and below. These results point to a complex evolution of conformational heterogeneity between 180--240 K that involves both thermal deactivation and solvent-driven arrest of protein motions in the crystal. The lack of a single shared conformational response to temperature within the dynamic active-site network provides evidence for a conformation shuffling model, in which exchange between rotamer states of a large aromatic ring in the middle of the network shifts the conformational ensemble for the other residues in the network. Together, our multitemperature analyses and XFEL data motivate a new generation of temperature- and time-resolved experiments to structurally characterize the dynamic underpinnings of protein function.

Goniometer-based femtosecond crystallography with X-ray free electron lasers.

The emerging method of femtosecond crystallography (FX) may extend the diffraction resolution accessible from small radiation-sensitive crystals and provides a means to determine catalytically accurate structures of acutely radiation-sensitive metalloenzymes. Automated goniometer-based instrumentation developed for use at the Linac Coherent Light Source enabled efficient and flexible FX experiments to be performed on a variety of sample types. In the case of rod-shaped Cpl hydrogenase crystals, only five crystals and about 30 min of beam time were used to obtain the 125 still diffraction patterns used to produce a 1.6-Å resolution electron density map. For smaller crystals, high-density grids were used to increase sample throughput; 930 myoglobin crystals mounted at random orientation inside 32 grids were exposed, demonstrating the utility of this approach. Screening results from cryocooled crystals of ß2-adrenoreceptor and an RNA polymerase II complex indicate the potential to extend the diffraction resolution obtainable from very radiation-sensitive samples beyond that possible with undulator-based synchrotron sources.

Remote access to crystallography beamlines at SSRL: novel tools for training, education and collaboration.

For the past five years, the Structural Molecular Biology group at the Stanford Synchrotron Radiation Lightsource (SSRL) has provided general users of the facility with fully remote access to the macromolecular crystallography beamlines. This was made possible by implementing fully automated beamlines with a flexible control system and an intuitive user interface, and by the development of the robust and efficient Stanford automated mounting robotic sample-changing system. The ability to control a synchrotron beamline remotely from the comfort of the home laboratory has set a new paradigm for the collection of high-quality X-ray diffraction data and has fostered new collaborative research, whereby a number of remote users from different institutions can be connected at the same time to the SSRL beamlines. The use of remote access has revolutionized the way in which scientists interact with synchrotron beamlines and collect diffraction data, and has also triggered a shift in the way crystallography students are introduced to synchrotron data collection and trained in the best methods for collecting high-quality data. SSRL provides expert crystallographic and engineering staff, state-of-the-art crystallography beamlines, and a number of accessible tools to facilitate data collection and in-house remote training, and encourages the use of these facilities for education, training, outreach and collaborative research.

New paradigm for macromolecular crystallography experiments at SSRL: automated crystal screening and remote data collection.

Complete automation of the macromolecular crystallography experiment has been achieved at SSRL through the combination of robust mechanized experimental hardware and a flexible control system with an intuitive user interface. These highly reliable systems have enabled crystallography experiments to be carried out from the researchers' home institutions and other remote locations while retaining complete control over even the most challenging systems. A breakthrough component of the system, the Stanford Auto-Mounter (SAM), has enabled the efficient mounting of cryocooled samples without human intervention. Taking advantage of this automation, researchers have successfully screened more than 200 000 samples to select the crystals with the best diffraction quality for data collection as well as to determine optimal crystallization and cryocooling conditions. These systems, which have been deployed on all SSRL macromolecular crystallography beamlines and several beamlines worldwide, are used by more than 80 research groups in remote locations, establishing a new paradigm for macromolecular crystallography experimentation.

Diffraction-based automated crystal centering.

A fully automated procedure for detecting and centering protein crystals in the X-ray beam of a macromolecular crystallography beamline has been developed. A cryo-loop centering routine that analyzes video images with an edge detection algorithm is first used to determine the dimensions of the loop holding the sample; then low-dose X-rays are used to record diffraction images in a grid over the edge and face plane of the loop. A three-dimensional profile of the crystal based on the number of diffraction spots in each image is constructed. The derived center of mass is then used to align the crystal to the X-ray beam. Typical samples can be accurately aligned in approximately 2-3 min. Because the procedure is based on the number of ;good' spots as determined by the program Spotfinder, the best diffracting part of the crystal is aligned to the X-ray beam.