A serial millisecond crystallography (SMX) facility has recently been implemented at the macromolecular crystallography beamline, MX2 at the Australian Synchrotron. The setup utilizes a combination of an EIGER X 16M detector system and an in-house developed high-viscosity injector, “Lipidico.” Lipidico uses a syringe needle to extrude the microcrystal-containing viscous media and it is compatible with commercially available syringes. The combination of sample delivery via protein crystals suspended in a viscous mixture and a millisecond frame rate detector enables high-throughput serial crystallography at the Australian Synchrotron. A hit-finding algorithm, based on the principles of “robust-statistics,” is employed to rapidly process the data. Here we present the first SMX experimental results with a detector frame rate of 100 Hz (10 ms exposures) and the Lipidico injector using a mixture of lysozyme microcrystals embedded in high vacuum silicon grease. Details of the experimental setup, sample injector, and data analysis pipeline are designed and developed as part of the Australian Synchrotron SMX instrument and are reviewed here.

Macromolecular crystallography is currently one of the most important methods for determining the structure of molecules. In 2017, over 90% of structures deposited in the Protein Data Bank (PDB) were determined using X-ray diffraction.1 The brightest X-ray source one can use for macromolecular structure determination is the X-ray free-electron laser (XFEL) which utilizes X-ray pulses that are sufficiently intense to produce high-quality diffraction while being of short enough duration to terminate before the onset of substantial radiation damage.2–4 The first serial femtosecond crystallography (SFX) experiments5,6 introduced a number of innovations. These include data collection from a continuously flowing suspension of crystals in a liquid jet,7 software for efficiently identifying crystal hits from SFX datasets consisting of 10’s of thousands of diffraction patterns,4 and indexing and merging data.8 

Since the first protein structure determination at an XFEL,5,6 technical development has been extremely rapid. In particular, the challenge of delivering fresh crystals fast enough to record diffraction patterns before they are destroyed by the high energy XFEL pulse has resulted in a wide variety of new sample-delivery methods, broadly classified as either fixed target or moving target methods.9 Here we focus on moving target methods, of which the fast flowing gas dynamic virtual nozzle (GDVN)7 is the most well-established. In spite of the success of the GDVN, liquid jet speeds on the order of 10 m/s have traditionally resulted in a very low crystal hit rate by the beam. This is due to the fact that around 80 mm of the sample stream never interacts with X-rays due to the gap between consecutive XFEL pulses running at 120 Hz. Therefore, the GDVN is only suitable for proteins that can be mass produced and delivered in narrow, fast flowing liquid jet. On the other hand, high-viscosity injectors (HVIs),10,11 where the protein crystals are embedded in a highly viscous carrier medium, result in a much higher fraction of the sample interacting with the X-ray beam and thus contributing to the dataset. In this type of injector, the sample mixture is extruded at a rate on the order of a few hundred micrometers per second. With a detector collecting images at 50 Hz, only 4–8 μm of the sample stream is then extruded between each image exposure, leading to a significant reduction in sample wastage compared to the GDVN. Thus, for a dataset collected using a HVI, usually <1 mg protein is required for structure solution compared to 50–200 mg protein for the GDVN.

Several synchrotron beamlines are now able to generate protein crystallography datasets with an exposure time on the order of milliseconds. This makes them ideally suited to HVI approaches, which allows the samples to move through the beam on similar timeframes. The recent integration of the HVI at synchrotrons using detectors which have millisecond image acquisition capabilities has spawned a new field of serial millisecond crystallography (SMX). SMX is typically performed at the 3rd generation synchrotron microfocus beamlines, under room temperature conditions. The increased availability of synchrotron sources compared to XFELs makes SMX attractive as a complementary technique to SFX that enables efficient, rapid data collection applied to the three most common crystallographic techniques: high-resolution structure determination, ligand soaking, and de novo phasing.12 

SMX was first established by Stellato et al.13 at the P11 beamline of the third-generation synchrotron source PETRA III (DESY, Hamburg). Stellato and co-workers collected almost 1.5 × 106 individual diffraction frames at a 25 Hz frame rate while a suspension of lysozyme microcrystal was pushed by a syringe pump to a 100 μm-inner-diameter glass capillary that was mounted on a motorized stage and placed horizontally in the X-ray interaction region.13 Later, SMX using a HVI was demonstrated by Botha et al. with lysozyme crystals embedded in a lipidic cubic phase (LCP) matrix at the PXII beamline at the Swiss Light Source (SLS).11 In the same year, Nogly et al. demonstrated SMX at the ESRF Microfocus Beamline ID13 (Grenoble, France) with crystals from the light-driven proton-pump bacteriorhodopsin (bR) also embedded in LCP.14 These HVI SMX experiments use significantly less sample than the equivalent liquid jet experiments as discussed earlier. However, a concern associated with having the sample moving more slowly through the beam compared to liquid jet is the increased risk of inducing radiation damage within the crystal. Even on millisecond time scales, radiation damage due to radical diffusion, quenching, and recombination processes can occur. Eliminating these effects requires an optimization of the crystal transit time through the beam allowing for high-resolution data to be collected while spreading the absorbed X-ray dose over larger crystal volumes. The balance between the low sample consumption (i.e., low flow rates) and radiation damage has been achieved in previous successful SMX-HVI experiments which have not reported any influence due to structural degradation in the X-ray beam. This is even true for samples which contain radiation sensitive residues that are highly prone to damage from X-rays.11,12,14,15

Since SMX is typically performed at room temperature, in contrast to protein crystallography under cryogenic conditions, SMX can be used to probe dynamics.16 This paves the way to time-resolved SMX studies being performed.17 Although not yet proven experimentally, the combination of HVIs with high-frame rate detectors (such as the EIGER) should enable the realization of time-resolved SMX. By optimizing the crystal transit time against the detector frame rate, it should be possible to resolve millisecond structural dynamics.

Continuous improvement in both detector technology and experimental setup has seen the data rates for SMX significantly improve in recent years. Botha et al. and Nogly et al. were able to demonstrate SMX with data collection rates of 10 Hz and 17 Hz, respectively. While at the Swiss Light Source (SLS) in 2017, Weinert et al. were able to collect SMX data with a detector running at 50 Hz. Here, at the Australian Synchrotron, we demonstrate SMX with an EIGER X 16M detector18 running at a frame rate of 100 Hz (10 ms exposure time) using a new HVI, Lipidico. Key features of the Lipidico setup are as follows: it utilizes commercially available syringes to act as the sample reservoir; the sample is extruded from the same syringe it was prepared in order to eliminate the need for multiple sample transfers; it is portable, and we can carry the whole device using the handle and mount it on the beamline within minutes; it has a wide range of sample extrusion velocities from 60 mm s−1 down to 60 μm s−1 to meet the sample delivery requirements for fast and slow imaging rates. We have also incorporated a new type of hit-finding algorithm19 into the SMX data analysis pipeline. This allows for fast, efficient processing of the very large multi-TB datasets typically produced by the EIGER X 16M. Here we describe and demonstrate a new facility for SMX at the Australian Synchrotron. This instrument is based on the combination of the custom-built Lipidico injector and the EIGER X 16M detector while the data pipeline incorporates a novel hit-finding algorithm based on the principles of “robust-statistics.”

Hen egg-white lysozyme was purchased from Sigma-Aldrich. Lysozyme crystals were grown using the batch method crystallization. A 10:1 ratio of crystallization buffer (1.5M lithium chloride, 0.2M sodium acetate, 40% polyethylene glycol 6000, pH 4.8) to 250 mg/ml lysozyme (in 50 mM sodium acetate solution, pH 4.0) was mixed and vortexed immediately. The sample was incubated overnight at 22 °C, allowing the crystals to grow. This process resulted in crystals with diameters between 15 and 20 μm in size. For delivery of the crystals using Lipidico, the lysozyme crystals were mixed with Dow corning high vacuum silicon grease to increase the viscosity of the sample. Standard LCP syringe mixing protocols20 were used to mix the lysozyme crystals with the silicon grease. Briefly, two gas tight Hamilton glass syringes (Hamilton Company, 7656-01) were used and syringe 1 contained 42 μl of silicon grease which was coupled to syringe 2 containing 28 μl of lysozyme crystals. The sample was mixed by pushing the plunger of the lysozyme crystals toward the silicon grease side, gently forcing the crystals to move into the grease. The plunger compressing the grease was then depressed forcing the grease back to the lysosome crystal side. This process was continued until the sample was homogeneous. All the sample was pushed into one syringe, the coupler was disconnected, and a 110 μm i.d. needle (Hamilton Company, 7803-05) was attached to the syringe. The syringe containing the sample was then mounted directly onto the Lipidico injector. The final crystal count in the silicon grease was estimated to be 4.3 × 105 ± 5.5 × 104 crystals/ml using the ImageJ particle analysis tool.21 

We developed the syringe injector “Lipidico” (Fig. 1) to disperse highly viscous liquids in a micrometer stream vertically into the X-ray beam. The drive device of Lipidico is a 24 V DC motor with only 16 mm in diameter that is connected to a gearbox with a 1:231 geared reduction, and the gearbox is connected to a high precision fine thread screw with 320 μm pitch that pushes against the syringe plunger of a standard Hamilton syringe. The drive line of the syringe injector is made of Maxon Motor components and consist of a GPX 19 Planetary Gearhead, ENX 10 Encoder, EPOS2 Position Controller, DC-max 16 S Precious Metal Brushes DC motor, all assembled on a custom made aluminum frame that fits an in-line motorized stage.22 The motorized stage works as the alignment tool of the vertical sample stream with the X-ray beam (Fig. 1).

FIG. 1.

Lipidico, the new high viscosity injector (HVI) built for the Australian Synchrotron. (a) A CAD illustration showing Lipidico HVI. Dimensions: 463 × 431 × 193 mm (W × H × D) (b) A photo of Lipidico mounted on the MX2 beamline at the Australian Synchrotron. The Lipidico injector design makes it possible to shift from cryocrystallography to serial crystallography within the same beamtime shift.

FIG. 1.

Lipidico, the new high viscosity injector (HVI) built for the Australian Synchrotron. (a) A CAD illustration showing Lipidico HVI. Dimensions: 463 × 431 × 193 mm (W × H × D) (b) A photo of Lipidico mounted on the MX2 beamline at the Australian Synchrotron. The Lipidico injector design makes it possible to shift from cryocrystallography to serial crystallography within the same beamtime shift.

Close modal

For SMX data collection, the beam size at the sample position was 12 × 22 μm (W × H) in full width half maximum (FWHM). The X-ray energy was 13 keV, and the flux at the focus was 2.4 × 1012 photons s−1. Standard 100 μl Hamilton syringes were loaded with approximately 20 μl of the grease-lysozyme crystal sample. The sample was extruded through a 110 μm i.d. needle (Hamilton Company, 7803-05). The sample stream was continuously monitored with an in-line camera (Fig. 2). The experiments were performed at 25 °C. The detector distance was set at 300 mm from the sample interaction point, and the detector was running at a frame rate of 100 Hz. The data collection was made in five data blocks containing 12, 150, 150, 300, 509 data files, respectively, and each file had 200 frames which resulted in 2400, 30 000, 30 000, 60 000, and 101 800 frames in each run and 224 200 frames in total. We estimate the crystal transit time through the beam to be 350–420 ms based on the formula: (crystal size + beam height)/stream velocity. Given the velocity range 100–120 μm s−1, crystal size between 15 and 20 μm, and crystal transit time 350–420 ms, we can expect to collect 35–42 frames per crystal with a 100 Hz frame rate.

FIG. 2.

Sample syringe mounted on Lipidico. Gas tight 100 μl Hamilton syringe with a 12 mm long and 110 μm i.d. needle mounted onto Lipidico. Inset: Sample injection region with a stream of viscous media extruded from the needle. The X-ray beam position is indicated by the dashed red arrow.

FIG. 2.

Sample syringe mounted on Lipidico. Gas tight 100 μl Hamilton syringe with a 12 mm long and 110 μm i.d. needle mounted onto Lipidico. Inset: Sample injection region with a stream of viscous media extruded from the needle. The X-ray beam position is indicated by the dashed red arrow.

Close modal

The collected frames were preprocessed using a peak finding algorithm based on robust statistics19 to filter out for analysis only those frames which contain diffraction patterns (hit finding step, Fig. 3). The following parameters were set in the algorithm for data analysis: minimum acceptable SNR 6, minimum number of peak pixels 1, maximum number of peak pixels 25, which were optimal for our dataset. The peak finding algorithm reduced the dataset to 6614 frames; corresponding to a 2.95% hit rate. Frames and hit rate per data block were 258 (10.75%), 2074 (6.91%), 993 (3.31%), 1163 (1.95%), and 2120 (2.08%), respectively.

FIG. 3.

Detected peaks by the peak finding algorithm.19 Peaks identified by the algorithm are denoted by black circles on an arbitrary diffraction image. Resolution rings are also displayed on the image showing diffraction peaks that extend up to 2 Å.

FIG. 3.

Detected peaks by the peak finding algorithm.19 Peaks identified by the algorithm are denoted by black circles on an arbitrary diffraction image. Resolution rings are also displayed on the image showing diffraction peaks that extend up to 2 Å.

Close modal

The output from the peak finding algorithm was saved in a single HDF5 file with an associated peak list and used as the input file for the CrystFEL 0.6.3. program suite8,23 for indexing and merging the diffraction patterns. The CrystFEL indexing programs used to determine the lattice parameters were “mosflm-axes, mosflm-latt-nocell, mosflm-nolatt-cell, and dirax-axes;” the “partialator” programs were used to merge reflections.23 We calculated figures of merit using “check-hkl” and “compare-hkl” in CrystFEL. Before refinement, we estimated the resolution using the correlation coefficient between the random half datasets CC1/2 and the signal-to-noise ratio I/σ(I) as described by Evans and Murshudov24 (Fig. 5).

The CrystFEL output was converted to the MTZ format for further analysis, and refinement was carried out using the CCP4 suite. Phaser25 was used for the molecular replacement to determine the structure where turkey egg-white lysozyme was used as the initial search model (PDB:1LJN). Refmac526 was used from the CCP4 suite for the structure refinement. Statistics for data collection and refinement are detailed in Table I together with published data from Refs. 12 and 15 for comparison.

TABLE I.

SMX data-collection statistics. This table provides a comparison of the SMX data collected at the Australian Synchrotron and two different facilities: Swiss Light Source (SLS) and the Advanced Photon Source (APS).

Data collection and processing
CurrentWeinert et al.12 Martin-Garcia et al.15 
Light source, beamline Australian synchrotron, MX2 SLS, X06SA APS GM/CA 23-ID-D 
Sample Lysozyme Lysozyme Lysozyme 
Crystal size (μm) 15–20 15 × 10 5–10 
Crystal density (No. per ml)    
Viscous media High vacuum grease LCP LCP 
Temperature (K) 300 293 298 
Detector Dectris EIGER X 16M Dectris EIGER X 16M PILATUS3 6M 
Frame rate (Hz) 100 50 10 
Sample–detector distance (mm) 300 n.a. 300 
X-ray energy (keV) 13 12.4 12 
Flux (photons s−12.4 × 1012 3.9 × 1011 3.0–4.1 × 1011 
Beam size (W × H) (μm) 12 × 22 5 × 5 10 
Average extrusion velocity (μm s−1110 250 120 
Total data collection time (h) ∼6 0.3 12 
Average crystal exposure (transit) time (ms) 385 n/a 85 
Space group P4321P4321P4321
Unit-cell a = b, c (Å) 78.68, 38.48 78.6, 38.9 79.1, 38 
α = β = γ (deg) 90 90 90 
No. of frames 224 200 58 000 364 724 
No. of frames in reduced data set 6 614 35 471 124 800 
Hit rate (%) 2.95 61 34 
No. of indexed frames 4 852 27 000 18 648 
Indexed of reduced data set (%) 73.4 76.1 14.9 
Resolution range (Å) 78.68–1.83 (2.05–1.83) 24.84–1.5 (n/a) 35–2.05 (2.10–2.05) 
Completeness (%) 99.44 (99.15) 94.7 (49.23) 99.8 (n/a) 
I/σ(I) 5.08 (1.30) 8.35 (0.72) 11.2 (0.4) 
Redundancy 73.97 (51.34) 452.9 (4,6) 873.3 (43.9) 
CC 0.99 (0.81) 0.99 (0.53) 0.978 (0.44) 
CC1/2 0.96 (0.47) 0.996 (0.17) n/a 
Rsplit 14.09 (93.57) n/a 10.7 (242.8) 
Refinement 
Measurements (reflections) in total 643 960 9 140 532 29 412 
Reflections used in refinement 9 435 20 181 7 164 
Resolution range (Å) 17.74–1.83 (2.05–1.83) 24.84–1.5 (n/a) 35–2.05 (2.10–2.05) 
Rwork/Rfree (%) 18.7/26.3 15.8/19.9 22.8/26.8 
No. of atoms 1 059 n/a 1 023 
Protein 1 001 n/a 1 002 
Water and others (ligands or ions) 58 n/a 21 
Average B factor (A249.42 37.0 34.9 
PDB ID 6MQV 5NJM 5UVJ 
Data collection and processing
CurrentWeinert et al.12 Martin-Garcia et al.15 
Light source, beamline Australian synchrotron, MX2 SLS, X06SA APS GM/CA 23-ID-D 
Sample Lysozyme Lysozyme Lysozyme 
Crystal size (μm) 15–20 15 × 10 5–10 
Crystal density (No. per ml)    
Viscous media High vacuum grease LCP LCP 
Temperature (K) 300 293 298 
Detector Dectris EIGER X 16M Dectris EIGER X 16M PILATUS3 6M 
Frame rate (Hz) 100 50 10 
Sample–detector distance (mm) 300 n.a. 300 
X-ray energy (keV) 13 12.4 12 
Flux (photons s−12.4 × 1012 3.9 × 1011 3.0–4.1 × 1011 
Beam size (W × H) (μm) 12 × 22 5 × 5 10 
Average extrusion velocity (μm s−1110 250 120 
Total data collection time (h) ∼6 0.3 12 
Average crystal exposure (transit) time (ms) 385 n/a 85 
Space group P4321P4321P4321
Unit-cell a = b, c (Å) 78.68, 38.48 78.6, 38.9 79.1, 38 
α = β = γ (deg) 90 90 90 
No. of frames 224 200 58 000 364 724 
No. of frames in reduced data set 6 614 35 471 124 800 
Hit rate (%) 2.95 61 34 
No. of indexed frames 4 852 27 000 18 648 
Indexed of reduced data set (%) 73.4 76.1 14.9 
Resolution range (Å) 78.68–1.83 (2.05–1.83) 24.84–1.5 (n/a) 35–2.05 (2.10–2.05) 
Completeness (%) 99.44 (99.15) 94.7 (49.23) 99.8 (n/a) 
I/σ(I) 5.08 (1.30) 8.35 (0.72) 11.2 (0.4) 
Redundancy 73.97 (51.34) 452.9 (4,6) 873.3 (43.9) 
CC 0.99 (0.81) 0.99 (0.53) 0.978 (0.44) 
CC1/2 0.96 (0.47) 0.996 (0.17) n/a 
Rsplit 14.09 (93.57) n/a 10.7 (242.8) 
Refinement 
Measurements (reflections) in total 643 960 9 140 532 29 412 
Reflections used in refinement 9 435 20 181 7 164 
Resolution range (Å) 17.74–1.83 (2.05–1.83) 24.84–1.5 (n/a) 35–2.05 (2.10–2.05) 
Rwork/Rfree (%) 18.7/26.3 15.8/19.9 22.8/26.8 
No. of atoms 1 059 n/a 1 023 
Protein 1 001 n/a 1 002 
Water and others (ligands or ions) 58 n/a 21 
Average B factor (A249.42 37.0 34.9 
PDB ID 6MQV 5NJM 5UVJ 

We used the program RADDOSE-3D27,28 to estimate an average diffraction weighted dose, which was calculated to be ∼0.32 MGy. The calculations assumed a grease container thickness of 50.00 μm to simulate the sample environment which results in a beam attenuation fraction of 0.10.

Our syringe injector uses a gas-focused free syringe-needle combination as previously implemented in the “DAPHNIS” application for SFX at the SACLA XFEL (Japan).29,30 One advantage of using this syringe-needle combination is that it eliminates any dead volume sample loss during transfer to an external sample reservoir. In general, when performing experiments with a HVI, the stream must be closely monitored and its speed must be adjusted during the measurement to ensure that a stable stream is produced. To improve stream stability, the traditional injectors presented by Botha et al.11 and Weierstall et al.10 utilize a helium sheath gas flowing over the sample capillary to support the sample stream as it comes out. Without the sheath gas, the viscous sample stream can curl up against the injector tip. Lipidico does not require any gas-focusing, and two solutions were developed to avoid this issue. The first approach was to induce a counteracting electrostatic charge on a small piece of polystyrene ∼5 mm under the needle to produce a downward acting Coulomb force on the sample. The second approach used a Venturi funnel under the needle that produce a weak suction effect to direct the viscous stream. We have tested the Lipidico injector with a number of alternative carrier media. Figure 4 demonstrates the stability of the stream for three different inert carriers: petroleum jelly, monoolein, and silicon grease.

FIG. 4.

Demonstration of the stream stability for three different inert carriers: (a) petroleum jelly, (b) monoolein (LCP), and (c) silicone grease as comparison.

FIG. 4.

Demonstration of the stream stability for three different inert carriers: (a) petroleum jelly, (b) monoolein (LCP), and (c) silicone grease as comparison.

Close modal

Lipidico combines a drive line, made up of a DC motor, gearbox, and a very fine threaded screw that pushes against the plunger on a standard Hamilton syringe. This combination makes it possible to use a broad range of stream velocities. In the current configuration, the maximal sample velocity is approximately 60 mm s−1 and this flow rate may be of requirement for performing time-resolved studies or with ultrafast data acquisition times in the microsecond range. Current HPLC pump based injectors have a lower maximal extrusion velocity, which makes them impossible to match such fast data-acquisition rates. The slowest extrusion velocity achievable with Lipidico is just 60 μm s−1, which may be used in situations where many frames per crystal are desired. When changing the extrusion velocity with Lipidico, there is no time-lag between setting the new set point velocity from the remote controlling computer to an actual change of injection velocity. The direct and instantaneous control of the injection velocity makes the injector very easy to run and ensures a steady and stable sample stream applicable for many crystallographic approaches and relevant detector frame rates.

The need for longer crystal exposure times with synchrotron sources, using monochromatic X-rays, combined with a desire to capture one diffraction frame per crystal has previously restricted the data-acquisition rates to typically <20 Hz. Weinert et al.12 successfully recorded multiple frames per crystal with an EIGER X 16 M detector at a frame rate of 50 Hz.12 Here, we further extend this limit further, collecting crystal diffraction data at a 100 Hz data-acquisition rate, which is 2–10 times the rate of previous HVI based SMX experiments.11,12,14,15 With a 10 ms exposure time, it was possible to collect the 224 200 frames in an effective measurement time of only 37 min and 20 s (2242 s). During the first Lipidico experiment presented here, the data were collected in five separate data blocks, totaling around 6 h of beamtime. During normal user operation, however, the amount of beamtime required would be substantially reduced. After the hit finding step, the data were reduced to 6614 frames, of which 4852 were indexed by CrystFEL(0.6.3).8,23 The observed reduction in the hit rate here (2.95%) compared to other reported SMX experiments (Table I) is attributed to a low crystal density; here, we used a crystal density of ∼105 ml−1 compared to ∼107 ml−1 reported in the literature.11 It is expected that the hit-rate increases substantially upon optimization of the crystal density and homogeneity. However, the high hit rate that was demonstrated by Weinert et al.12 by their “crystal scanning” approach with a wide and thin beam appears currently to be the best practice to achieve high hit rates with HVI based SMX.

A key question to answer in SMX experiments is what is the minimum amount of data required for structure determination? Weinert et al. demonstrated that the cross correlation between the final refined positions and simulated annealing ligand omit maps reaches 90% of the final cross correlation after only 4000, 900, and 900 indexed patterns, respectively, for the three different ligands they studied. Thus, whether a ligand is bound or not could be determined with comparatively few images.12 Furthermore, Stellato et al. studied the required number of indexed patterns based on Rwork and Rfree and found that ≥5000 patterns are required for structure solution.13 The present experiment indicates that, depending on the crystallography of the system, these predictions of the number of patterns required of structure solution are reasonable.

To further analyze our data quality and to estimate the resolution limit, we followed the approach by Evans and Murshudov24 and plotted the signal-to-noise, I/σ(I), and the correlation coefficient, CC1/2, against resolution. The advantage of a correlation coefficient is that it has a well defined range: 1 for a good correlation and 0 for no correlation. CC1/2 is generally close to 1 at low resolution and falls sharply to near zero at higher resolution as the intensities become weaker, as shown in Fig. 5, below 2 Å where signal-to-noise (I/σ(I)) falls to below 1.0. There is at present no general consensus on the optimum criteria for interpretation of these plots and how to estimate the point at which adding additional high-resolution data does not contribute anything useful: hence the true “resolution” of a dataset is still debated in the literature. By the time CC1/2 has fallen to around 0.2–0.4, or I/σ(I) to around 0.5–1.5, there is little information remaining;24 however, it is difficult to establish definitive criteria prior to refinement, wherein in the present case the resolution could be extended to below 2 Å.

FIG. 5.

Plot of data-processing statistics against resolution, for lysozyme data collected at the Australian Synchrotron. CC1/2 is close to 1 at low resolution and falls sharply at a resolution <2 Å, as the intensities become weaker and I/σ(I) falls below 1.0.

FIG. 5.

Plot of data-processing statistics against resolution, for lysozyme data collected at the Australian Synchrotron. CC1/2 is close to 1 at low resolution and falls sharply at a resolution <2 Å, as the intensities become weaker and I/σ(I) falls below 1.0.

Close modal

During our analysis of the peak list from the peak finding step, we found that adjacent frames with crystal hits arrives in clusters in the data stream. We present an example of this by plotting the hit sequence from run number 3 that contains 30 000 frames and is 300 s in duration in Fig. 6, where each line is a frame detected by the peak finding algorithm. The distribution of hits resembles a barcode pattern and does not follow a uniform probability distribution over the course of a dataset. The median length of a cluster with hits in the five data blocks was 40 ± 18 frames (at 95% confidence interval), which is consistent with number of frames per crystal estimated from the crystal transit time above. The exact positions of the lines in a cluster of hits follows from the parameters we set in the peak finding algorithm which were optimized for the dataset. Crystal clustering leading to “data blocks” is most likely due to inhomogeneous sample mixing. This effect can potentially be reduced through using a smaller (e.g., 50 μm I.D.) nozzle. However, it can also be likely eliminated through optimization of the sample preparation conditions for future experiments. The three-way coupler presented by James and Standfuss31 to further improve sample homogenization provides an effective approach that could be adopted.

FIG. 6.

Crystal clustering detected in the HVI stream. (a) Plot of the peak list from the peak finding algorithm of run 3, depicted as a barcode pattern. Adjacent frames with crystal hits show up in clusters over the course of the dataset, as exemplified here in a 300 s long run with 30 000 frames. (b) A zoomed in section between frames 9100 and 9300 of the data stream in (a).

FIG. 6.

Crystal clustering detected in the HVI stream. (a) Plot of the peak list from the peak finding algorithm of run 3, depicted as a barcode pattern. Adjacent frames with crystal hits show up in clusters over the course of the dataset, as exemplified here in a 300 s long run with 30 000 frames. (b) A zoomed in section between frames 9100 and 9300 of the data stream in (a).

Close modal

Hydrophobic media, such as LCP,32 grease,30 and vaseline11 are very versatile and compatible with most crystallization conditions. They form structures (LCP) or emulsions (grease and vaseline) where the crystals are contained in their hydrophilic environment surrounded by the hydrophobic molecules that constitute the carrier media. The background scattering created from the hydrophobic molecules show up in the diffraction pattern as a ring around the beamstop, at low resolution, and at higher resolution around 3–5 Å. To reduce these background effects and increase compatibility with variety of protein samples, efforts have been made to use carrier media such as agarose,33 hyaluronic acid,34 NaCMC and F-127,35 hydroxyethyl cellulose matrix,36 and polyacrylamide.37 For a recent review about sample delivery medium selection for serial crystallography, see Ref. 38.

In general, the background intensity profile is attenuated and less prominent with a hydrophilic carrier; however, a diffuse, sometimes broad, background scattering ring can show up due to the prevalent water bonded structures in hydrophilic assemblies. Peak finding algorithms find similar number of indexable diffraction patterns collected from hydrophilic as well as hydrophobic carrier media and yields similar data quality.35 Hence, when selecting the appropriate carrier media, the priority is on identifying a chemically compatible system which provides optimum injection conditions for the protein crystals.

We used a data-acquisition strategy that captured around 40 images per crystal, with a slow flow rate that resulted in an average X-ray exposure time of 380 ms. At room temperature, this gives rise to a higher accumulated radiation dose per crystal. To estimate the dose, we used the program RADDOSE-3D.27,28 This version included contributions from the surrounding carrier matrix. Based on the average amount of time the crystal spends in the X-ray beam, RADDOSE-3D predicted an average diffraction weighted dose of 0.32 MGy equivalent to ∼8 kGy/image and a dose rate of 831 kGy s−1. This dose rate is comparable to that used by Owen et al.39 to study the X-ray-induced decay of protein and virus crystals in room temperature macromolecular crystallography studies. To study potential radiation damage in our case, we calculated the electron density map around the disulfide bonds. For the results presented here, no signs of radiation damage were observed (Fig. 7). No measurable radiation damage, even for structures containing radiation sensitive residues, has been reported in previous HVI based SMX experiment.11,12,14,15 Notably, Cherezov et al. demonstrated that high viscous sample carrier like lipidic cubic mesophases can cause an inverse dose-rate effect, which would result in less radiation damage to the crystals.40 

FIG. 7.

A representative image of the electron density map (2Fo-Fc, 1σ) surrounding the disulfide bonds of lysozyme, showing no signs of radiation damage. The omit density map (Fo-Fc) was generated at 3σ and is shown in red.

FIG. 7.

A representative image of the electron density map (2Fo-Fc, 1σ) surrounding the disulfide bonds of lysozyme, showing no signs of radiation damage. The omit density map (Fo-Fc) was generated at 3σ and is shown in red.

Close modal

In conclusion, we have described and demonstrated new capability for SMX at the Australian Synchrotron. This instrument is based on the implementation of a novel injection system “Lipidico” capable of a very wide range of sample delivery flow rates and avoiding the need for sample transfer. The data analysis pipeline included the use of a novel hit-finding algorithm based on the principle of robust-statistics. Together these elements constitute an efficient and flexible SMX setup at the Australian Synchrotron for the user that can accommodate a variety of SMX and potentially even time-resolved experiments.

P.B., M.H.J., C.D., and B.A. acknowledge support from the Australian Research Council Centre of Excellence in Advanced Molecular Imaging (Grant No. CE140100011) (http://www.imagingcoe.org/). This research was undertaken together with beamline scientists Dr. Tom Caradoc-Davies and Dr. Jun Aishima at the Australian Synchrotron in Melbourne, Australia.

1.
BioSync
, “
A structural biologist’s guide to high energy data collection facilities
,” http://biosync.sbkb.org/index.jsp; accessed on
5 February 2018
.
2.
R.
Neutze
,
R.
Wouts
,
D.
van der Spoel
,
E.
Weckert
, and
J.
Hajdu
,
Nature
406
,
752
(
2000
).
3.
L.
Lomb
,
T. R.
Barends
,
S.
Kassemeyer
,
A.
Aquila
,
S. W.
Epp
,
B.
Erk
,
L.
Foucar
,
R.
Hartmann
,
B.
Rudek
,
D.
Rolles
 et al,
Phys. Rev. B
84
,
214111
(
2011
).
4.
A.
Barty
,
C.
Caleman
,
A.
Aquila
,
N.
Timneanu
,
L.
Lomb
,
T. A.
White
,
J.
Andreasson
,
D.
Arnlund
,
S.
Bajt
,
T. R.
Barends
 et al,
Nat. Photonics
6
,
35
(
2012
).
5.
H. N.
Chapman
,
P.
Fromme
,
A.
Barty
,
T. A.
White
,
R. A.
Kirian
,
A.
Aquila
,
M. S.
Hunter
,
J.
Schulz
,
D. P.
DePonte
,
U.
Weierstall
 et al,
Nature
470
,
73
(
2011
).
6.
S.
Boutet
,
L.
Lomb
,
G. J.
Williams
,
T. R.
Barends
,
A.
Aquila
,
R. B.
Doak
,
U.
Weierstall
,
D. P.
DePonte
,
J.
Steinbrener
,
R. L.
Shoeman
 et al,
Science
337
,
362
(
2012
).
7.
D.
DePonte
,
U.
Weierstall
,
K.
Schmidt
,
J.
Warner
,
D.
Starodub
,
J.
Spence
, and
R.
Doak
,
J. Phys. D: Appl. Phys.
41
,
195505
(
2008
).
8.
T. A.
White
,
R. A.
Kirian
,
A. V.
Martin
,
A.
Aquila
,
K.
Nass
,
A.
Barty
, and
H. N.
Chapman
,
J. Appl. Crystallogr.
45
,
335
(
2012
).
9.
L.
Chavas
,
L.
Gumprecht
, and
H.
Chapman
,
Struct. Dyn.
2
,
041709
(
2015
).
10.
U.
Weierstall
,
D.
James
,
C.
Wang
,
T. A.
White
,
D.
Wang
,
W.
Liu
,
J. C. H.
Spence
,
R. B.
Doak
,
G.
Nelson
,
P.
Fromme
,
R.
Fromme
,
I.
Grotjohann
,
C.
Kupitz
,
N. A.
Zatsepin
,
H.
Liu
,
S.
Basu
,
D.
Wacker
,
G. W.
Han
,
V.
Katritch
,
S.
Boutet
,
M.
Messerschmidt
,
G. J.
Williams
,
J. E.
Koglin
,
M. M.
Seibert
,
M.
Klinker
,
C.
Gati
,
R. L.
Shoeman
,
A.
Barty
,
H. N.
Chapman
,
R. A.
Kirian
,
K. R.
Beyerlein
,
R. C.
Stevens
,
D.
Li
,
S. T. A.
Shah
,
N.
Howe
,
M.
Caffrey
, and
V.
Cherezov
,
Nat. Commun.
5
,
3309
(
2014
).
11.
S.
Botha
,
K.
Nass
,
T. R.
Barends
,
W.
Kabsch
,
B.
Latz
,
F.
Dworkowski
,
L.
Foucar
,
E.
Panepucci
,
M.
Wang
,
R. L.
Shoeman
 et al,
Acta Crystallogr., Sect. D: Biol. Crystallogr.
71
,
387
(
2015
).
12.
T.
Weinert
,
N.
Olieric
,
R.
Cheng
,
S.
Brünle
,
D.
James
,
D.
Ozerov
,
D.
Gashi
,
L.
Vera
,
M.
Marsh
,
K.
Jaeger
 et al,
Nat. Commun.
8
,
542
(
2017
).
13.
F.
Stellato
,
D.
Oberthür
,
M.
Liang
,
R.
Bean
,
C.
Gati
,
O.
Yefanov
,
A.
Barty
,
A.
Burkhardt
,
P.
Fischer
,
L.
Galli
,
R. A.
Kirian
,
J.
Meyer
,
S.
Panneerselvam
,
C. H.
Yoon
,
F.
Chervinskii
,
E.
Speller
,
T. A.
White
,
C.
Betzel
,
A.
Meents
, and
H. N.
Chapman
,
IUCrJ
1
,
204
(
2014
).
14.
P.
Nogly
,
D.
James
,
D.
Wang
,
T. A.
White
,
N.
Zatsepin
,
A.
Shilova
,
G.
Nelson
,
H.
Liu
,
L.
Johansson
,
M.
Heymann
,
K.
Jaeger
,
M.
Metz
,
C.
Wickstrand
,
W.
Wu
,
P.
Båth
,
P.
Berntsen
,
D.
Oberthuer
,
V.
Panneels
,
V.
Cherezov
,
H.
Chapman
,
G.
Schertler
,
R.
Neutze
,
J.
Spence
,
I.
Moraes
,
M.
Burghammer
,
J.
Standfuss
, and
U.
Weierstall
,
IUCrJ
2
,
168
(
2015
).
15.
J. M.
Martin-Garcia
,
C. E.
Conrad
,
G.
Nelson
,
N.
Stander
,
N. A.
Zatsepin
,
J.
Zook
,
L.
Zhu
,
J.
Geiger
,
E.
Chun
,
D.
Kissick
 et al,
IUCrJ
4
,
439
(
2017
).
16.
J. S.
Fraser
,
H.
van den Bedem
,
A. J.
Samelson
,
P. T.
Lang
,
J. M.
Holton
,
N.
Echols
, and
T.
Alber
,
Proc. Natl. Acad. Sci. U. S. A.
108
,
16247
(
2011
).
17.
P.
Nogly
,
V.
Panneels
,
G.
Nelson
,
C.
Gati
,
T.
Kimura
,
C.
Milne
,
D.
Milathianaki
,
M.
Kubo
,
W.
Wu
,
C.
Conrad
 et al,
Nat. Commun.
7
,
12314
(
2016
).
18.
A.
Casanas
,
R.
Warshamanage
,
A. D.
Finke
,
E.
Panepucci
,
V.
Olieric
,
A.
Nöll
,
R.
Tampé
,
S.
Brandstetter
,
A.
Förster
,
M.
Mueller
 et al,
Acta Crystallogr., Sect. D: Struct. Biol.
72
,
1036
(
2016
).
19.
M.
Hadian-Jazi
,
M.
Messerschmidt
,
C.
Darmanin
,
K.
Giewekemeyer
,
A. P.
Mancuso
, and
B.
Abbey
,
J. Appl. Crystallogr.
50
,
1705
(
2017
).
20.
M.
Caffrey
and
V.
Cherezov
,
Nat. Protoc.
4
,
706
(
2009
).
21.
C. A.
Schneider
,
W. S.
Rasband
, and
K. W.
Eliceiri
,
Nat. Methods
9
,
671
(
2012
).
22.
See https://www.kuseldesign.com.au/Contact.html for specific OEM information contact coauthor Michael Kusel.
23.
T. A.
White
,
V.
Mariani
,
W.
Brehm
,
O.
Yefanov
,
A.
Barty
,
K. R.
Beyerlein
,
F.
Chervinskii
,
L.
Galli
,
C.
Gati
,
T.
Nakane
 et al,
J. Appl. Crystallogr.
49
,
680
(
2016
).
24.
P. R.
Evans
and
G. N.
Murshudov
,
Acta Crystallogr., Sect. D: Biol. Crystallogr.
69
,
1204
(
2013
).
25.
A. J.
McCoy
,
R. W.
Grosse-Kunstleve
,
P. D.
Adams
,
M. D.
Winn
,
L. C.
Storoni
, and
R. J.
Read
,
J. Appl. Crystallogr.
40
,
658
(
2007
).
26.
G. N.
Murshudov
,
P.
Skubák
,
A. A.
Lebedev
,
N. S.
Pannu
,
R. A.
Steiner
,
R. A.
Nicholls
,
M. D.
Winn
,
F.
Long
, and
A. A.
Vagin
,
Acta Crystallogr., Sect. D: Biol. Crystallogr.
67
,
355
(
2011
).
27.
O. B.
Zeldin
,
M.
Gerstel
, and
E. F.
Garman
,
J. Appl. Crystallogr.
46
,
1225
(
2013
).
28.
C. S.
Bury
,
J. C.
Brooks-Bartlett
,
S. P.
Walsh
, and
E. F.
Garman
,
Protein Sci.
27
,
217
(
2017
).
29.
K.
Tono
,
E.
Nango
,
M.
Sugahara
,
C.
Song
,
J.
Park
,
T.
Tanaka
,
R.
Tanaka
,
Y.
Joti
,
T.
Kameshima
,
S.
Ono
,
T.
Hatsui
,
E.
Mizohata
,
M.
Suzuki
,
T.
Shimamura
,
Y.
Tanaka
,
S.
Iwata
, and
M.
Yabashi
,
J. Synchrotron Radiat.
22
,
532
(
2015
).
30.
M.
Sugahara
,
E.
Mizohata
,
E.
Nango
,
M.
Suzuki
,
T.
Tanaka
,
T.
Masuda
,
R.
Tanaka
,
T.
Shimamura
,
Y.
Tanaka
,
C.
Suno
,
K.
Ihara
,
D.
Pan
,
K.
Kakinouchi
,
S.
Sugiyama
,
M.
Murata
,
T.
Inoue
,
K.
Tono
,
C.
Song
,
J.
Park
,
T.
Kameshima
,
T.
Hatsui
,
Y.
Joti
,
M.
Yabashi
, and
S.
Iwata
,
Nat. Methods
12
,
61
(
2015
).
31.
D.
James
,
T.
Weinert
,
P.
Skopintsev
,
A.
Furrer
,
D.
Gashi
,
T.
Tanaka
,
E.
Nango
,
P.
Nogly
, and
J.
Standfuss
,
J. Visualized Exp.
144
,
e59087
(
2019
).
32.
E. M.
Landau
and
J. P.
Rosenbusch
,
Proc. Natl. Acad. Sci. U. S. A.
93
,
14532
(
1996
).
33.
C. E.
Conrad
,
S.
Basu
,
D.
James
,
D.
Wang
,
A.
Schaffer
,
S.
Roy-Chowdhury
,
N. A.
Zatsepin
,
A.
Aquila
,
J.
Coe
,
C.
Gati
,
M. S.
Hunter
,
J. E.
Koglin
,
C.
Kupitz
,
G.
Nelson
,
G.
Subramanian
,
T. A.
White
,
Y.
Zhao
,
J.
Zook
,
S.
Boutet
,
V.
Cherezov
,
J. C. H.
Spence
,
R.
Fromme
,
U.
Weierstall
, and
P.
Fromme
,
IUCrJ
2
,
421
(
2015
).
34.
M.
Sugahara
,
C.
Song
,
M.
Suzuki
,
T.
Masuda
,
S.
Inoue
,
T.
Nakane
,
F.
Yumoto
,
E.
Nango
,
R.
Tanaka
,
K.
Tono
 et al,
Sci. Rep.
6
,
24484
(
2016
).
35.
G.
Kovácsová
,
M. L.
Grünbein
,
M.
Kloos
,
T. R.
Barends
,
R.
Schlesinger
,
J.
Heberle
,
W.
Kabsch
,
R. L.
Shoeman
,
R. B.
Doak
, and
I.
Schlichting
,
IUCrJ
4
,
400
(
2017
).
36.
M.
Sugahara
,
T.
Nakane
,
T.
Masuda
,
M.
Suzuki
,
S.
Inoue
,
C.
Song
,
R.
Tanaka
,
T.
Nakatsu
,
E.
Mizohata
,
F.
Yumoto
 et al,
Sci. Rep.
7
,
703
(
2017
).
37.
J.
Park
,
S.
Park
,
J.
Kim
,
G.
Park
,
Y.
Cho
, and
K. H.
Nam
,
Sci. Rep.
9
,
2525
(
2019
).
38.
K. H.
Nam
,
Int. J. Mol. Sci.
20
,
1094
(
2019
).
39.
R. L.
Owen
,
D.
Axford
,
J. E.
Nettleship
,
R. J.
Owens
,
J. I.
Robinson
,
A. W.
Morgan
,
A. S.
Doré
,
G.
Lebon
,
C. G.
Tate
,
E. E.
Fry
 et al,
Acta Crystallogr., Sect. D: Biol. Crystallogr.
68
,
810
(
2012
).
40.
V.
Cherezov
,
K. M.
Riedl
, and
M.
Caffrey
,
J. Synchrotron Radiat.
9
,
333
(
2002
).