Structural biologists are undertaking increasingly challenging projects including the study of membrane proteins and complex multi-component machines. Structural investigations are also transitioning beyond solving a single static structure, to the application of a series of sequential structural snapshots to provide details of the atomic positions and motions that define the relationships involved in molecular recognition, transition state stabilization, and other aspects of the biocatalytic process. The success of these experiments requires careful optimization of samples and experimental setups, often involving multiple experiments at the laboratory bench and the beamline, where automation serves as an enabling technology to efficiently deliver multiple crystals and meet stringent timing requirements.

Developments at SSRL and LCLS-MFX will be presented that tackle challenges involved in the use of very small and radiation-sensitive crystals. To facilitate the handling and optimization of delicate crystals, new in situ crystallization and remote data collection schemes have been released that avoid direct manipulation of crystals, support robotic sample exchange, and allow full rotational access of the sample in a controlled humidity environment. By simplifying crystal handling and transport at near-physiological temperatures, these technologies remove barriers to enable more widespread use of serial crystallography methods for studies of metalloenzyme structure and protein dynamics. Data analysis tools that provide rapid feedback for experimental optimization during fast-paced experiments will also be described.

Right from its initial conception, the micro-focus beamline I24 at Diamond Light Source has looked beyond the assumption that an experimenter’s structural question would be answerable with a single, well diffracting, cryo-cooled sample. A multi-crystal approach to data collection has become a modus operandi. Initially attention was focused on small volume and weakly diffracting samples that would typically receive a destructive X-ray dose before complete and redundant data could be recorded. To help tackle this requirement, pipelines for rapid collection and intelligent merging of thin wedges of data from multiple crystals have been developed. Additionally, Serial Synchrotron crystallography (SSX) has become a core activity, with the intention of probing structural dynamics obtainable within protein crystals at room temperature. This brings the requirement for many thousands of crystals, each contributing only a tiny proportion of the final dataset and providing a challenge in terms of collection and processing. I present latest results from SSX and multi-crystal experiments, describe the tools available for users of the beamline and consider optimum methods for successful many-crystal experiments.

Infrared (IR) absorption spectromicroscopy is a powerful, non-invasive probe that provides access to spatio-chemical information at the micron scale. The physical basis of IR spectroscopy lies in the oscillations of dynamic dipole moments in chemical bonds, with resonant frequencies in the IR spectral region of 4,000-400 cm-1 wave numbers. The bending or stretching of chemical bonds between atoms with different electronegativities, such as O-H or C=O, will lead to intense absorption and thus provide a unique fingerprint of specific chemical groups within the sample. The presence - or absence - of specific spectral fingerprints provide the opportunity to locate and identify chemical processes throughout the sample, and track these processes in time or as a function of external perturbation.

We can perform these types of measurements using a Synchrotron Fourier Transform Infrared (SFTIR) spectromicroscopy setup, as it provides orders of magnitude more photons than traditional bench-top machines1. Even with an ultra-bright IR source as provided at the Advanced Light Source, acquisition times typically take multiple hours. These long acquisition times are in part caused by the size of the field of view, as compared to the probe size: users typically analyze a 70 um by 100 um sample using a regular grid with a spacing of 1 um. With an acquisition time of 4 seconds per pixel, we would need about 8 hours to measure the full sample. Given that access to instruments is scarce, compounded by the desire to characterize different samples, being able to speed up data acquisition is of paramount importance.

Here we present a strategy that drastically increases the efficiency of SFTIR spectromicroscopy by coupling the data collection with Gaussian Process based surrogate model 2,3. This approach models the full hyperspectral datasets across the entire field of view, including regions we haven’t measured yet, by multivariate Normal distribution. By analyzing this distribution, we gain insight into which future measurement locations provide the greatest reduction in total uncertainty, and can also predict various quality metrics of the surrogate model we can use to end an experiment. Preliminary experiments show we can increase the throughput of SFTIR experiments by a factor of ~20.

References

1. Holman, H. ‐Y N. & Martin, M. C. Synchrotron Radiation Infrared Spectromicroscopy: A Noninvasive Chemical Probe for Monitoring Biogeochemical Processes. Advances in Agronomy 79–127 (2006) doi:10.1016/s0065-2113(06)90003-0.

2. Chang, H. et al. Building Mathematics, Algorithms, and Software for Experimental Facilities. in Handbook on Big Data and Machine Learning in the Physical Sciences 189–240 (World Scientific, 2020).

3. Noack, M. & Zwart, P. Computational Strategies to Increase Efficiency of Gaussian-Process-Driven Autonomous Experiments. in 2019 IEEE/ACM 1st Annual Workshop on Large-scale Experiment-in-the-Loop Computing (XLOOP) 1–7 (2019).

The rotation method is the most common approach of collecting macromolecular diffraction data. In the days of image plates and charge-coupled device detectors (CCDs), substantial readout time and noise made sophisticated data collection strategies necessary. The correct starting angle of data collection would help minimize the number of images. A rotation increment of up to 1°/image served to raise weak reflections above the detector noise. Datasets took hours to collect.

This is not a sensible way of collecting data anymore. Hybrid Photon Counting detectors, which are installed on essentially all MX beamlines around the world and on many laboratory diffractometers, are free of dark current and readout noise and limited only by Poisson counting statistics. Using rotation increments of around 0.1°/image (fine slicing) decreases the measured background and increases the signal to noise of the experiment. With fast detectors, full 360° datasets can be collected in seconds to a few minutes.

Does the new standard of 360° of data collected at 0.1°/image excuse crystallographers from thinking and optimizing their experiments? Not at all. We show how the full-rotation approach to data collection can accommodate such scenarios as extremely radiation-sensitive samples and experimental phasing. Solving structures by single-wavelength anomalous dispersion from atoms native to the sample becomes possible even with data collected at room temperature. A successful experimental strategy comprises adjustments to beam energy, photon flux, detector distance, starting angle, number of full rotations, orientation of the crystal, and many more.

The recording of data at the highest possible quality makes all subsequent steps of data processing, phasing and model building easier. It will result in a more precise atomic model to answer the biological questions that prompted the structural work. Despite the apparent simplicity of the full-rotation method, data collection, the last experimental step of MX, is as critical as ever. There is no excuse for walking away with less than best data.

A recent surge in reports of crystals exhibiting elastic flexibility has changed the way we view these materials. With potential applications in flexible electronics, in depth research is required to understand why some crystals can be tied into knots, while others shatter under an applied force. Different rationales for elastic flexibility have been proposed: many crystals have been engineered to impart flexibility through isotropic interactions, although other elastic crystals have anisotropic interactions [1]. Clearly, the different interactions present result in diverse bending mechanisms. The mechanism of flexibility in elastic crystals can be resolved on an atomic-scale by use of micro-focused synchrotron radiation [2]. By examining the localised crystal structure at multiple positions across a bent crystal, the deformations of the cell parameters can be quantified (Fig. 1). Isotropic and anisotropic crystals have been analysed using this technique to determine their respective mechanisms.

Unfortunately, structural mapping quickly produces large volumes of data, and manual processing would be inefficient when there are only small changes to the data. Instead, software was developed to automatically process these datasets. It is capable of taking raw frames and providing finalised CIF files with results graphically analysed. This allows for greater insight into these elastic crystals, as more data can be analysed in a reasonable time frame. This software, CX-ASAP, consists of a series of independent modules which can be placed together into an auto-processing pipeline. The advantage of this modular approach, is the fact that it is applicable to a wider range of large crystallographic dataset analysis, such as variable temperature experiments. The main consideration of this software is the limit of computer knowledge, as there are key steps during the automation where user input is mandatory for reliable results.

[1] Ahmed, E., Karothu, D. P. & Naumov, P. (2018). Angew. Chem. Int. Ed. Engl. 57, 8837-8846.

[2] Worthy, A., Grosjean, A., Pfrunder, M. C., Xu, Y., Yan, G., Edwards, G. & Clegg, J. C. (2018). Nat Chem. 10, 65-69.

Keywords: flexible crystal; elastic crystal; automation; mechanisms; synchrotron

The author wishes to acknowledge the work of Dr Arnaud Grosjean for preliminary automation work.

The decision on the high-resolution cutoff has an apparent impact on the quality of a structure model. To determine the optimal cutoff automatically, we developed a software tool PAIREF [1]. The program performs the paired refinement protocol that allows linking the data and structure model quality. This analysis goes beyond the conventional criteria based on the indicators of data quality only (e.g. I/σ(I), R_meas).

PAIREF is freely available for multiple platforms and can be run from the command-line or graphical user interface. Two refinement engines are currently supported: REFMAC5 from the CCP4 software suite [2] and PHENIX.REFINE [3]. The program creates a compact comprehensive report. The final decision on the cutoff is based on several statistics that are calculated and monitored: R-values, correlation coefficients, optical resolution, merging statistics, etc. The consequent comparison between CC_work and CC* allows the assessment of overfitting. Moreover, a unique feature of the program is the complete cross-validation scenario: the protocol is run in parallel for each free-reflection set selection individually which leads to averaged, more general and meaningful results.

During the work on PAIREF, we confirmed previous findings and proved that useful signal can be often still present in the high-resolution data not fulfilling the obsolete conventional criteria. To give an example: In the particular case of interferon gamma from Paralichthys olivaceus (PDB entry 6f1e), the cutoff was originally applied at 2.3 Å, according to the criterion for I/σ(I) higher than 2 in the highest resolution shell. Nevertheless, we ran paired refinement up to 1.9 Å and observed a systematic decrease in R_free while including data up to 2.0 Å [1]. Hence, the structure was improved, despite very poor statistics relating to the last resolution shell 2.1-2.0 Å (I/σ(I) = 0.1, CC_1/2 = 0.03).

Furthermore, we similarly examined the high-resolution data from endothiapepsin (PDB entry 4y4g). This structure was originally solved at 1.44 Å resolution. However, we could observe a significant improvement in the quality of electron density of the partially occupied fragment after refinement up to 1.20 Å (Fig. 1). This observation was in harmony with corresponding drops in R_free [1].

Generally, the quality of a structure model can benefit from the involvement of even weak high-resolution data. Thus, the application of paired refinement could be recommended for any structural project in X-ray macromolecular crystallography. PAIREF provides automation of the routine and gives all the relevant statistics for users to make a precise decision on the cutoff.

Figure 1. Improvement in omit maps of the partially occupied fragment B53. Electron density after refinement up to 1.44 (purple) and 1.20 Å (orange) is shown at a level of 0.56 eÅ⁻³. Atomic coordinates were adapted from PDB entry 4y4g.

[1] Malý, M., Diederichs, K., Dohnálek, J. & Kolenko, P. (2020). IUCrJ 7, pp. 681–692.

[2] Winn, M. D., Ballard, C. C., Cowtan, K. D., Dodson, E. J., Emsley, P., Evans, P. R., Keegan, R. M., Krissinel, E. B., Leslie, A. G., McCoy, A., McNicholas, S. J., Murshudov, G. N., Pannu, N. S., Potterton, E. A., Powell, H. R., Read, R. J., Vagin, A., & Wilson, K. S. (2011). Acta Cryst. D 67, pp. 235–242.

[3] Adams, P. D., Afonine, P. V., Bunkóczi, G., Chen, V. B., Davis, I. W., Echols, N., Headd, J. J., Hung, L. W., Kapral, G. J., Grosse-Kunstleve, R. W., McCoy, A. J., Moriarty, N. W., Oeffner, R., Read, R. J., Richardson, D. C., Richardson, J. S., Terwilliger, T. C., & Zwart, P. H. (2010). Acta Cryst. D 66, pp. 213–221.

This work was supported by the MEYS CR (projects CAAS – CZ.02.1.01/0.0/0.0/16_019/0000778 and BIOCEV – CZ.1.05/1.1.00/02.0109) from the ERDF fund, by the Czech science foundation (project 18-10687S), and by the GA CTU in Prague (SGS19/189/OHK4/3T/14).