High-speed raster-scanning synchrotron serial microcrystallography with a high-precision piezo-scanner

The design and first measurements with a high-speed piezo-positioner-based goniometer for raster-scanning serial macromolecular micro-focus crystallography at synchrotron storage rings are presented.


Introduction
In macromolecular crystallography (MX), serial crystallography has been the method of choice to overcome limitations in sample dose life under X-ray irradiation. Before the advent of cryocrystallography (Hope, 1988;Henderson, 1990), the merging of partial datasets from multiple crystals was used to deal with the limited dose life (Kendrew et al., 1960) and continues to be used in room-temperature crystallography for hard-to-freeze crystals, such as virus crystals (Fry et al., 1999;Axford et al., 2012;Rossmann, 2015).
At free-electron lasers, the 'diffraction before destruction' approach  has crystals surviving only a single exposure and this therefore dictates serial crystal delivery; the crystal sample is destroyed during the measurement and must be replaced by a new sample for further diffraction measurements. With the latest generation of high-brilliance storage-ring sources, beamlines are being constructed that enable XFEL-like serial crystallography to be performed at storage-ring facilities . The appeal of these instruments is that they enable crystallographic data collection from microcrystals that would previously have been too small to yield useable diffraction data.
In a pioneering experiment at beamline P14 at the PETRA3 electron storage ring, the methods from FEL serial-femtosecond crystallography and synchrotron-based MX were successfully combined . In these measurements, a 3 Å -resolution structure was obtained by rastering a frozen slurry of microcrystals of cathepsin B across a microfocus beam and merging the diffraction images from hundreds of microcrystals to make a complete crystallographic dataset. Experiments at the ESRF beamline ID13 have extended this scanning serial method to room-temperature data collection by containing the microcrystals in an Si 3 N 4 membrane sandwich to avoid the drying of the samples while limiting background scattering (Coquelle et al., 2015).
These scanning serial crystallography methods dramatically reduce sample consumption when compared with techniques employing liquid jets (Weierstall, 2014) or flow capillaries . For crystals not grown in lipidic cubicphase media, they avoid the need to introduce the crystals into a viscous medium for the high-viscosity extrusion injectors (Nogly et al., 2015) and the associated background scattering. The acquisition of several sequential rotation exposures in the cryo-rastering experiment simplifies data treatment and obviates the need for Monte Carlo averaging of still images and a further reduction in the amount of crystals required to record a complete dataset. The increased number of reflections recorded in a rotation image and the known partialities of the recorded reflections aid indexing and data integration Gati et al., 2014). Even in cases where larger crystals of a compound are available, the advantages of microcrystallography may make this method appealing by avoiding averaging over crystal inhomogeneities, reduced mosaicity (Bowler et al., 2010), reduced radiation damage attributed to photoelectron escape in the case of micrometresized crystals (Sanishvili et al., 2011) and increased uniformity during ligand diffusion (Stagno et al., 2017).
Recently constructed synchrotron storage rings achieve ever lower horizontal electron-beam emittances from 1.2 nm rad for PETRA III (Wanzenberg et al., 2015), over 0.9 nm rad for NSLS-II (Willeke, 2015) to 0.3 nm rad for MAX IV (Eriksson et al., 2016), permitting the design of MX beamlines that deliver microfocus beams with high flux densities. Specifically, the FMX beamline in sector 17-ID-2 of NSLS-II currently delivers 3 Â 10 12 photons s À1 into a 1 mm (V) Â 1.5 mm (H) beam focus (Fuchs et al., 2016). This flux density surpasses the brightest current MX beamlines by up to two orders of magnitude. For the unattenuated beam at FMX at 12.66 keV, the estimated maximum dose reaches the 30 MGy Garman limit (Owen et al., 2006) in as short as 40 ms, while considering a protein crystal matched to the beam size.
A common factor to raster-scanning serial crystallography approaches at storage rings reported to date Coquelle et al., 2015) are data collection times in the hours range as a result of available dose rates and goniometer speeds. Fixed-target supports with crystals loaded onto regularly spaced grids enable raster scans with significantly increased hit rates close to 50% (Owen et al., 2017) and up to >90% (Roedig et al., 2017), and thereby achieve data collection times under 10 min.
In addition to fixed rasters, the new goniometer at FMX supports fly-scanning mode without stopping the goniometer or closing the shutter. This comes at much higher collection speeds that overcome the lower hit rates. In addition, crystals can be loaded onto unstructured membranes to facilitate sample-holder loading without boundary conditions on the crystal shapes. With the new goniometer's scanning speed, we demonstrated data collection times of a few minutes, with the actual open-shutter time under 20 s. By bringing raster-scanning serial crystallography collection times from the hours range into the time range of a classical single-crystal data collection, these experiments become an efficient new way of collecting diffraction datasets from previously intractable crystals. In addition, MX at such extreme dose rates promises two further key advantages: an increase in time resolution and the potential to outrun radiation-damage effects (Warkentin et al., 2017(Warkentin et al., , 2013Owen et al., 2012).
The focus of our present work is to enable rapid-scanning serial macromolecular crystallography experiments under cryo-conditions and room temperature at the full flux of the fully focused FMX beamline. The system we developed has the potential for application well beyond the protein crystallography community; X-ray imaging and microscopy will benefit from this development. Full-field transmission X-ray microscopes require fast and accurate rotation to acquire three-dimensional tomograms (Wang et al., 2012;Hwu et al., 2013). Nanometre-scale resolution scanning fluorescence imaging systems also require accurate rotation accompanied by a fast fly-scanning capability (Nazaretski et al., 2015(Nazaretski et al., , 2017Holler et al., 2014). The capability of the system we developed to perform infinite rotation makes it even more appealing for in situ, in operando measurements performed during transmission, fluorescence, diffraction or ptychography experiments (Liu et al., 2016;Nazaretski et al., 2013Nazaretski et al., , 2014Shapiro et al., 2014).

Instrument design
To realize the promise of experiments at the FMX beamline's highest achievable dose rates, sample positioning and movement requires uncompromising design decisions for the goniometer spindle and experimental station. Here we present the newly developed FastForward Goniometer, a design that is not limited by the requirements of multi-axis macromolecular crystallography goniometers.
The experimental boundary conditions inform the requirements for the goniometry to be developed. The 750 Hz maximal detector frame rate of the FMX's Eiger 16M detector in a 4M region of interest (ROI) coincides well with a 3% fraction of the estimated 40 ms half-life. For the goniometer translations, a raster pitch with a 1 mm beam size yields the scanning-speed requirement of 1 mm s À1 . This is a lower bound when considering overlapping beam tails, as well as potential flux upgrades to the beamline, e.g. by increasing the monochromator bandwidth. The chosen design can achieve linear scanning speeds of >10 mm s À1 . The main challenge for achieving fast rastering performance are the resulting high line-scanning frequencies. At 1 mm s À1 , for a 100 mm raster size, the associated 10 Hz scanning frequency will lead to unacceptable mechanical resonances in typical MX goniometer systems. Here we present the design and demonstrate the performance of a piezo positioner-based goniometer with a 200 mm Â 200 mm Â 200 mm travel range and 25 nm bidirectional repeatability, which can raster the full scan area in 20 s with a 1 mm step size. Combined with a 4 mm travel range coarse piezo-walk centering stage, it is applicable to both classical cryo-crystallography and high-speed raster-scanning serial crystallography. The FastForward Goniometer system was developed in collaboration with Physik Instrumente and utilized modified NEXLINE and NanoCube stages as subparts. To achieve the desired scanning speed and precision, as well as to fit the system into the limited space of the FMX endstation, the scanner utilizes three piezo actuators arranged diagonally with respect to the laboratory coordinate system (see Fig. 1b). When compared with a stack of individual linear-motion stages, this diagonal arrangement yields a much smaller form factor of the entire assembly. When arranged diagonally, the stiffness of the stage we developed, which determines the highest scanning frequency it can achieve, is much higher than that of a simple motion stack. Moreover, the combined motion of all three piezo actuators allows a performance of not only translational but also rotational motion around an arbitrarily selected pivot point. Owing to the fact that shear piezo actuators are utilized in the scanner we developed, its scanning range is limited to 200 mm in all three directions. To overcome the travel-range limitation, the fine-scanning stage is mounted on top of coarse linear actuators with a travel range of 4 mm in the Y and Z directions, respectively [see Fig. 1(a) for a schematic of the setup]. Fig. 1(c) demonstrates the system that we developed installed at the FMX endstation, all key components enumerated. The piezo scanner system is mounted on a horizontal high-precision air-bearing rotational axis (Nelson Air SP150), which in turn resides on top of an XYZ stage assembly. This scanning system allows for infinite rotation around the x axis, i.e. in Fig. 1(a); therefore all actuators and encoders of the scanner are connected through a sliding cable connector (the so-called slip ring) which is a part of the Nelson Air rotary stage. Upon completion, the manipulation system we developed was first characterized in the metrology laboratory of NSLS-II and then tested at the FMX endstation.

Sample preparation
Rod-shaped bovine trypsin crystals and proteinase K microcrystals were chosen as the test crystals for our experiments. Trypsin crystals were grown on siliconized glass cover slips (hanging drop) by equilibrating 15 mg ml À1 trypsin (Sigma-Aldrich; T1426) plus 5 mg ml À1 benzamidine and 10 mM calcium chloride in 20 mM HEPES (pH 7.0) with 3.75% PEG 3350 and 5% glycerol over a reservoir containing 15% PEG 3350 and 20% glycerol. Large trypsin crystals formed after $7 days, surrounded by thin trypsin needles. Proteinase K crystals were grown on siliconized glass cover slips (hanging drop) by equilibrating 25 mg ml À1 proteinase K (Worthington Biochemical; LS004224) in 40 mM calcium chloride, 400 mM sodium   nitrate and 50 mM BisTris (pH 6.5) over a reservoir containing 160 mM calcium chloride, 1.6 M sodium nitrate and 200 mM BisTris (pH 6.5). To generate 5 mm-sized microcrystals, each crystallization drop was allowed to air dry for $5 min, until an opaque layer of sub-micrometre crystals formed, and then the cover slip was mounted over the reservoir. The surface layer of crystals continued to grow overnight to a final size of 5 mm and were cryo-protected by 30% ethylene glycol prior to harvesting.
The sample crystals were harvested from the mother liquor using loops or meshes on SPINE caps. The samples were then flash-cooled and stored in liquid nitrogen until mounted on the goniometer using a permanent magnet. In a final system update, a smart magnet (Arinax), which can detect the presence of mounted specimens, will replace the permanent magnet to enable automatic sample changing. A nitrogen coolant gas stream (Cryostream 800 system, Oxford Cryosystems) provided a sample temperature of 100 K.

Instrument characterization
Prior to deployment at the beamline, the system that we developed was thoroughly characterized at the metrology laboratory of NSLS-II and basic performance characteristics were inferred.
3.1.1. Stiffness. The stiffness of a system, usually characterized by its fundamental resonance frequencies, determines the highest scanning frequency the system can achieve. We utilized a three-axis laser interferometer (Attocube FPS3010) to measure displacements of the scanning stage in the X, Y and Z directions. The recorded time domain signal was Fourier transformed and the frequency domain spectrum was obtained, see Fig. 1(d).
The fundamental resonance frequencies in the X, Y and Z directions correspond to 380 Hz, 275 Hz and 290 Hz, respectively, which allows us to achieve scanning frequencies exceeding 100 Hz.

Resolution and repeatability.
The resolution and repeatability of the piezo scanner were directly measured using the same three-axis interferometer. In order to improve the background noise and stability of the measurements, the piezo scanner was mounted on a home-built aluminium support structure installed on a pneumatically suspended optical table. To minimize the influence of acoustic noise, the system we developed was kept inside a double-wall sound-absorbing enclosure. As shown in Fig. 2(a), steps of 10 nm in all three directions are clearly resolved. The noise floor of approximately 10 nm in all three directions was mainly determined by environmental contributions. When fully assembled and powered through a slip-ring, the background noise increased to 25 nm because of the friction-based mechanical connection inside the slip-ring.
Similar measurements over a travel distance of 1 mm were also performed to determine the repeatability of the motion. Fig. 2(b) demonstrates a 1 mm repetitive motion in the Xdirection, with a repeatability better than 5 nm. A 2 nm slow drift observed during the 10 s wait time at each step (see inset in Fig. 2b) was caused by thermal fluctuation. For larger steps (in excess of 100 mm) the measured bidirectional repeatability was determined to be better than 25 nm. All three axes demonstrated similar performances. For coarse motions, repeatability was determined to be 135 nm and 77 nm for the y axis and z axis, respectively, for a travel range >2 mm.
In the endstation, the background-noise measurement using external interferometry was not feasible with the space constraints. Here, readings from the piezo stages' internal encoders were utilized to estimate the noise level, which was 150 nm peak-to-peak.  rotational errors by measuring the trajectory of a diamondturned reference cylinder installed on top of a rotary stage during rotational motions. The cylinder is 30 mm tall and 25 mm in diameter with the surface roughness better than 0.2 nm and circularity better than 30 nm. The setup, reference cylinder and measurement procedure are similar to those described by Xu et al. (2014) and Kim et al. (2013). Fig. 2(c) demonstrates rotational errors obtained during the measurements. The total runout error is 1 mm peak-to-peak with a 150 nm asynchronous error (see inset in Fig. 2c). The synchronous (repeatable) error will be incorporated into the control algorithms and compensated for by a piezo-scanner during scanning measurements. The axial runout was determined to be 40 nm peak-to-peak.
3.1.4. Accuracy of high-speed scanning. The trajectory of a high-speed scan was measured by external interferometry. A typical two-dimensional scan involves fast motion in the X-direction and slow motion in the Y-direction. Fig. 2(d) shows a comparison between the targeted position and the actual position measured by the interferometer at a 40 Hz scanning frequency. The scanning range was maximized and reached AE100 mm in the X-direction. The bidirectional repeatability was determined to be frequency dependent, varying from <100 nm at 20 Hz scanning frequency and reaching the maximum of 300 nm at 40 Hz scanning frequency and above.

Beamline experiments
Following laboratory characterization, the piezo scanner was installed at the FMX endstation for crystallographic data collection. During the measurements, diffraction data were recorded by an Eiger 16M detector, which can achieve up to a 133 Hz full frame rate and 750 Hz frame rate in a 4M ROI.
For evaluation of the performance of the goniometer and scanner system in crystallographic experiments, the most common combined-motion data collection strategies were characterized.
3.2.1. Vector scan. In order to reveal potential systematic errors in the hardware and/or software, the piezo scanner was first tested by performing vector scans along rod-shaped crystals of bovine trypsin with average dimensions of 25 mm Â 25 mm Â 200 mm. In vector scans, the goniometer performs a combined motion of the rotation axis and the Xf, Yf and Zf motions to translate the crystal across the beam in a continuous exposure.
The experiment was conducted at a photon energy of 12.66 keV, a flux of 8.8 Â 10 10 photons s À1 , a beam size of 2 mm (V) Â 4 mm (H) FWHM and the detector operating in full-frame mode. After centering the selected crystal with respect to the X-ray beam using the on-axis microscope, diffraction images were collected in a continuous fly scan over a rotation range of 180 with 0.1 per frame at an exposure time of 50 ms. During the rotation, the sample was translated 0.5 mm deg À1 to spatially distribute the radiation dose over a 90 mm range. The average dose in the exposed region was calculated to be 3.6 MGy via RADDOSE-3D (Zeldin et al., 2013). Demonstrating the smoothness of the goniometer movement, the scale factor of the processed dataset varies between 0.47 and 0.52 (Fig. S1 of the supporting information).
The diffraction patterns from the vector scan were indexed and integrated using the fast_dp processing pipeline (Winter & McAuley, 2011;Kabsch, 2010;Grosse-Kunstleve et al., 2002;Winn et al., 2011). The quality and internal consistency of the data were judged on the basis of standard hI/(I)i statistics, as well as on the basis of CC(1/2), the correlation percentage between intensities from random half-datasets (Karplus & Diederichs, 2012). The diffraction data were processed to a resolution of 1.52 Å , with an overall R merge of 0.041 and a CC(1/2) of 99.9% (Table 1). The hI/(I)i in the highest-resolution shell was 1.6. Structure refinement was conducted via PHENIX (Adams et al., 2010) with an initial model from PDB entry 3t26. The resultant R free = 0.189.
3.2.2. Raster scan. For high-speed raster-scanning serial crystallography data collection, stiff MiTeGen polyimide sample supports were utilized for stability during scanning and to maximize the scanning speed. We had greatest success with efficient crystal loading and low background scatter using MicroLoop and customized MicroWell pins (Guo et al., 2018) from MiTeGen. Diffraction experiments were conducted at a photon energy of 12.66 keV, beam size of 1 mm (V) Â 2 mm (H) FWHM, maximum unattenuated flux of 1.8 Â 10 12 photons s À1 at a ring current of 300 mA at the time of the experiment and with the detector in 4M ROI mode to achieve the required frame rates.
Raster scans of the trypsin crystals with detector frame rates of up to 750 Hz were conducted in a test experiment to calibrate the synchronization between the piezo scanner and detector. After the raster scan, Spotfinder (Sauter, 2010(Sauter, , 2011 was utilized to evaluate the quality of each diffraction pattern. A heat map, based on the spot count determined by Spotfinder, demonstrated the location and the morphology of the protein crystal. The synchronization of spindle and piezo positioners to the detector was adjusted until the heat map of the raster scan matched the shape of the crystal (Fig. 3a). After the initial calibration, the synchronization worked for all frame rates and scanning speeds.
Unlike serial femtosecond X-ray crystallography at FELs, which takes thousands of 'still' diffraction patterns from crystals, the sample in our experiment is rotated during raster scan data collection. As a result of the moment of inertia of the rotary stage, the sample is rotated continuously in one direction to synchronize the rotation with the high-speed raster scanning. For all raster experiments, we used proteinase K microcrystals with average sizes of 5-10 mm.
We developed protocols for the two main-use cases of raster scanning, namely 'rastering for crystal location' and 'rastering for data collection'.
In 'rastering for crystal location', the aim is to scan the ROI and locate all micrometre-sized crystals. A raster scan with a 2 mm step and 2.5 ms exposure time per frame scans an area of 80 mm Â 40 mm in 2 s (Fig. 3b). Instead of still exposures, the scan was conducted in a 'nearly still' way with a rotation of 0.002 per frame to minimize the position mismatch induced research papers by the rotation while retaining the rotation axis as the experiment synchronization master; 10% of the maximum flux resulted in an average dose of 30 kGy. The resultant heat map (Fig. 3b inset) shows the position of each crystal.
'Rastering for data collection' is optimized to achieve the highest data collection speed; therefore, the heat-map function is disabled in this type of raster scan. Various combinations of parameters including raster area, step size and oscillation width were characterized in a set of initial experiments. Given the size of the protein crystals, a 1 mm step size and 0.2 oscillation width were determined empirically to be the optimum settings to enable indexing of the partial datasets (i.e. diffraction data from a 1 wedge would be collected by scanning across a 5 mm-wide crystal). Multiple raster scans were performed sequentially over the ROI. The total rotation during a data collection raster scan, as dictated by the number of steps and the oscillation width per step, was limited in the À45 to +45 angular range around the face-on view to minimize multiple crystal exposures. In the current experiments, each raster section was entered separately in the data collection GUI; in the future, a large millimetre-scale raster scan will be automatically partitioned to sub-rasters and performed by combining the fine and coarse scanning stages. Different detector frame rates from 200 Hz to 750 Hz, corresponding to exposure times from 5 ms to 1.334 ms, were tested. Detailed settings of these raster scans are listed in Table S2. The beam attenuation was adjusted based on the exposure time in order to maintain a constant dose (0.30 MGy calculated by RADDOSE-3D). At the turning points of the raster lines, the dose deposited on the crystals is higher because of the slower goniometer translation. The border pixels were not evaluated in the data processing.

Data processing
The raster data from multiple microcrystals were processed using a Python scripted workflow, which employs DOZOR (Svensson et al., 2015), XDS (Kabsch, 2010), POINTLESS and AIMLESS (Winn et al., 2011). XDS was chosen for its robust handling of poor quality images, such as poor spot shape and weak diffraction, and for its capability to run highly parallelized on our compute cluster. The command line input and output streams of DOZOR and XDS greatly facilitate their integration with Python.  Table 1 X-ray data-collection and refinement statistics for crystallography experiments performed using a piezo scanner at the FMX beamline. Diffraction patterns acquired from multiple raster scans were first processed by DOZOR, which estimated the diffraction signal of each frame and assigned a scalar figure of merit (Svensson et al., 2015). Based on DOZOR's figures of merit, the data were divided into multiple partial datasets, with a typical partial dataset containing diffraction patterns from a 1-2 wedge. Each partial dataset was then separately indexed and integrated in XDS to determine the space group and unitcell dimensions from each indexable dataset. Inspired by the Nearest-cell PDB database search tool (Ramraj et al., 2012), the unit-cell size of each dataset was transformed from crystallographic coordinates (a, b, c, , , ) into the default orthogonal system (Bernstein et al., 1977;Callaway et al., 1992) to better visualize the indexed partial datasets and remove outliers. The resultant diagonal vectors of the unit cell were displayed in a three-dimensional plot (Fig. 3c) and were clustered using the K-means clustering algorithm (Lloyd, 1982). Each cluster was processed separately in the subsequent processing. The outliers, as determined by their distance from the center of the cluster, were excluded from subsequent processing.
The remaining datasets were combined via XSCALE without merging and assuming the space group is P1. The result was evaluated by POINTLESS to determine the best-fitted space group and the data merged again by XSCALE utilizing the given space group. The SciPy functions Linkage and Dendrogram (Jones et al., 2001) were utilized to perform the hierarchical clustering of the correlation coefficients from XSCALE, using the distance definition described by Giordano et al. (2012) and Santoni et al. (2017). The resultant dendrogram was plotted via Python matplotlib (Hunter, 2007) for further outlier removal (Fig. 3d). This two-step outlier rejection, i.e. unit-cell clustering followed by correlation coefficient clustering, was found to be necessary as the correlation coefficient clustering was not able to completely distinguish the difference in unit cells. Compared with other unit-cell clustering methods that use the edge lengths (Santoni et al., 2017) or face diagonal lengths (Giordano et al., 2012), this coordinate transformation method considers both the length and orientation of the volume diagonal vector, and worked very well for the datasets presented.
After outlier removal, the data merged by AIMLESS were used for structure refinement, which was manually performed in PHENIX. PDB entry 2id8 was selected as the starting model for E. album proteinase K. COOT (Emsley et al., 2010) was used for viewing the electron density maps, adding ligands and adjusting the solvent. The refined structures were validated using the program PROCHECK (Laskowski et al., 1993).
As shown in Table 1, data of proteinase K with a 200 Hz, 500 Hz and 750 Hz detector frame rate were processed. A total of 30 raster scans were collected at each frame rate, resulting in 391, 385 and 499 partial datasets, respectively. Among those partial datasets, 279, 236 and 316 were found to be indexable, giving an average indexing rate of 65%. After removing those with incorrect unit-cell sizes, a hierarchical cluster-analysis dendrogram with a cutoff limit at a distance of 0.60 was utilized for the final outlier removal. The final datasets, merged from 225, 214 and 299 partial datasets, respectively, were processed to a resolution of 2 Å , limited by the detector distance and area in ROI mode. The corresponding hI/(I)i in the highest-resolution shell was 3.2 AE 0.2. The overall CC(1/2) was 95.9%, 93.8% and 95.5%, respectively, with 83.6%, 78.5% and 81.8% in the highest-resolution shell. The structure refinement also showed similar R free and B factors. Crystallography data collection and processing at FMX. Raster scans of (a) a bovine trypsin rodshaped crystal and (b) 5-10 mm-sized proteinase K crystals loaded on MiTeGen loops. The rasterscan areas are shown as red-dashed rectangles; the corresponding heat maps acquired after the raster scans are shown in the insets. The grid coloring is the diffraction pattern spot count. (c) Clustering of the serial crystallography partial datasets based on unit-cell size. The data shown were acquired from raster scans of proteinase K microcrystals with a 200 Hz detector frame rate. Of these 279 partial datasets, 234 (blue dots) were indexed to c = 106.58 AE 0.41 Å , with 99.6% completeness ( Table 1). The remaining 45 partial datasets (orange dots) with c = 110.06 AE 0.84 Å did not yield a complete dataset and were excluded from refinement. The unit-cell size was converted to Cartesian coordinates for better visualization. The data were clustered into two groups using a K-mean clustering algorithm. (d) Hierarchical cluster-analysis dendrogram for the proteinase K partial datasets acquired at a 750 Hz detector frame rate. The cutoff limit was set to 0.6 (dashed line).
At three data collection speeds, both the X-ray diffraction data and the structure refinement demonstrate similar statistics, indicating equal data quality. The data taken at a 750 Hz detector frame rate were collected with 100% transmission using the full flux of the FMX beamline. While scanning the 30 mm Â 15 mm area with a 1 mm step, the piezo scanner achieved a 12.5 Hz scanning frequency, surpassing our regular goniometer by more than one order of magnitude. This is currently still limited by the beamline flux, which will be further increased with the NSLS-II ring current from 350 to 500 mA, and by higher dose rates achievable at longer wavelengths.

Conclusions
We have developed the FastForward Goniometer, a highspeed, high-precision, piezo-based goniometer system to perform serial microcrystallography using the full flux of the fully focused FMX beamline. The laboratory characterization demonstrated that the system we developed is capable of achieving greater than 40 Hz scanning frequency over a 200 mm scanning range in all three directions with 300 nm repeatability. For scanning frequencies below 5 Hz, the bidirectional repeatability was below 25 nm. The performance of the goniometer system was illustrated by conducting highspeed raster-scanning serial microcrystallography with the detector frame rates varying between 200 Hz and 750 Hz. The resultant protein structures, which were solved via a Python scripted automated data-processing workflow, indicated equally high data quality, proving reliable performance of the piezo-scanner at various scanning frequencies. This unprecedented experimental speed significantly reduces serial crystallography data collection time at synchrotrons, allowing utilization of the full brightness of the emerging synchrotron radiation facilities.