TECHNOLOGY FEATURE

Structural biology: doors open at the European XFEL Vivien Marx

X-ray beams at 27,000 pulses per second promise high-resolution views of macromolecules.

As of September 1, 2017, scientists can and square in the beam are quite rare,” says come to the European X-ray free- Chapman. (EuXFEL) for structural biology pur- Around the world, a number of XFELs suits. They can collect data on are operational or almost there (Box 2, protein nanocrystals and particles such as “Some X-ray free-electron ”), and viruses, protein complexes and single mol- more are being built, for example, in China. ecules. They might create dynamic virus Accelerator-driven FELs and the new ‘movies’ from a series of individual snap- EuXFEL offer scientists exposures on a shots. After 15 years of development, con- femtosecond (10–15 seconds) or even atto- struction and testing, EuXFEL has ‘lased’: second (10–18 seconds) timescale so they Mancuso Jessica it generated a beam brighter than those can measure the structure, variability and Earlier this year, the EuXFEL’s first laser beam produced by all other existing X-ray sourc- dynamics of the experimental objects of their reached the ‘hutch’. es (see Box 1, “EuXFEL at a glance”). In choice1–5. more testing, a beam was successfully sent Strictly speaking, biological objects are developer at the University of Wisconsin to a ‘hutch’, one of several lead-and-steel- never identical to one another, says Janos in Milwaukee, and his colleagues made encased rooms at the facility. Some hutches Hajdu, a biophysicist now at Uppsala movies of a virus (published in this issue6). hold gratings, filters and other beam-tun- University who spent much of his career at Such movies can reveal biologically impor- ing equipment; others have instruments the University of Oxford. Macromolecular tant information, says Ourmazd. In the for measuring samples. From nearby con- structures adopt various conformations. future, the EuXFEL could help researchers trol rooms researchers interact with instru- “Since biology wiggles, it’s all about motion,” map energy landscapes of biological mol- ments in the hutches. he says. This plasticity enables these mole- ecules, identify functional routes along The EuXFEL’s big selling point is its high cules’ biological functions. Some movements those landscapes and make movies of those repetition rate, says Henry Chapman, a happen in milliseconds, some atomic vibra- routes, all of which could, he says, “provide researcher at the Center for Free-Electron tions last only 10 femtoseconds. a multi-lane highway between experiment Laser Science in Hamburg, Germany, which Experiments at the EuXFEL offer the and molecular dynamics.” With EuXFEL is jointly run by the Max Planck Society, opportunity to capture movements down to data it’s possible to get a detailed view of the University of Hamburg and DESY, the the timescale of atomic vibrations. “We are ligand-binding, which underlies all signal- German Electron . With 27,000 living our dream,” says Hajdu. He is part of ing and regulatory processes. X-ray laser pulses per second, the EuXFEL’s a team of early EuXFEL users who will use Chapman sees a way to capture play- pulse rate will make beamtime for nanocrystallography and stud- by-play events, such as those during an it possible to com- ies of single particles such as viruses. One day, enzymatic reaction, on a femtosecond plete an experiment he says, single-cell studies might be possible. timescale by applying XFEL-based time- more quickly and Hajdu has long actively made the scientif- resolved crystallography with hundreds, with less sample ic case for XFELs, including both the Linac even thousands, of microcrystals. Usually than previously Coherent Light Source (LCLS) at Stanford such experiments involve light-triggered possible. “It could University—the world’s first XFEL, which reactions, but there are other fast impulses also help for single- began lasing in 2009—and the EuXFEL. He to try, he says. He and colleagues do fast particle diffractive and many other researchers can now devise sample mixing: a solution with crystals

S. Bajt imaging, which has experiments and strategize about looming meets a solution with a ligand or other been hampered challenges. reactant as the sample courses through The EuXFEL is the place for single- by the fact that the microfluidic channels on its way to an molecule diffraction really good shots XFEL possibilities injection device. “Fast in that context experiments, says where the virus or Using single-particle LCLS–XFEL data, might be as quick as 0.1 milliseconds,” he Henry Chapman. whatever is hit fair Abbas Ourmazd, a physicist and algorithm says. Separately, he notes, research groups

NATURE METHODS | VOL.14 NO.9 | SEPTEMBER 2017 | 843 TECHNOLOGY FEATURE

BOX 1: EUXFEL AT A GLANCE

The facility is 3.4 kilometers long and generates high-luminosity X-rays at a repetition rate of 27,000 pulses per second. Pulses have been produced at 0.15-nanometer —atomic dimensions—and the EuXFEL’s design could get it to 0.05 nanometers. Information for users is here. The EuXFEL is set up as a nonprofit company and has involved collaborating researchers and funding agencies from 11 countries: Denmark, France, Germany, Hungary, Italy, Poland, Russia, Slovakia, Spain, Sweden and Switzerland. The UK was part of the project and then dropped out owing to budget constraints, but it is considering rejoining. The EuXFEL has received UK funding—the UK is, for example, part of the SFX user consortium, as are Sweden, Germany and Slovakia.

The facility’s accelerator whips to nearly the speed of EuXFEL, H. Müller-Elsner light. Then, under the sway of magnets with alternating north and south poles, the electrons undulate and emit X-ray flashes. Electrons reach nearly the speed of light in the superconducting The X-rays and electrons travel together, which pushes the accelerator modules, shown in yellow. electrons into bunches, with each bunch behaving like a single large charged particle. A micro-bunch of 100,000 electrons can experimenters need and then sent to such experiment instrument emit ten billion more X-rays than a single electron. ‘hutches’ as the Femtosecond X-ray Experiments instrument Early users will begin collecting data during their approved (FXE) or the Single Particles, Cluster and Biomolecules and Serial beamtime this fall at the research campus in Schenefeld, Femtosecond Crystallography instrument (SPB/SFX). Germany, near Hamburg. The beam can be tuned as Sources: Henry Chapman, DESY; Janos Hajdu, Uppsala University. use pulse electric fields to find flexible­ to XFELs. Traditionally, they use optical work has been limited to virus particles that parts of a protein. microscopy to find workable crystals. Now are relatively large, she says. The EuXFEL is the place for single- that researchers experiment with “invisibly The high pulse rate enables high-through- molecule diffraction experiments, says small crystals,” it’s faster to deliver a sample put explorations of different states in the Chapman. “And we have a crazy idea on into a beam to see whether it diffracts, says molecules’ energy landscape; researchers how to use the fluorescence light from the Chapman. In less than an hour, thousands can monitor protein complexes in action sample to get structure,” he says. Hundreds of crystallization conditions can be tested. in previously impossible ways. It’s a highly more X-ray are absorbed by Large combinatorial studies can reveal how time-resolved view also of energetically disfa- in the sample than actually scatter to form the structure changes at different tempera- vored transition states that are of great inter- a diffraction pattern. These photo-excited tures or pH levels. est in drug discovery and that help scientists atoms will fluoresce as X-rays with differ- explore function. ent , depending on the element Combining techniques Uetrecht leads a project called in question. He and his colleagues want to To date, research at LCLS has pioneered VISAVIX, which combines mass spec- measure this fluorescence by using pulses to single-molecule imaging and serial trometry and XFEL-based work. The obtain a diffraction pattern. crystallography studies, says Charlotte approach lets her deepen research into Chapman and others want to automate Uetrecht, a scientist at the Heinrich Pette something mass spec has revealed: no sample delivery to the XFEL so that in Institute, which is linked to the Leibniz protein complex, as she says, “comes less than 10 seconds researchers could Institute for Experimental Virology in in a single flavor.” Each one has a vari- obtain a single structure or a snapshot of Hamburg. The EuXFEL will be more intense ety of modifications or flexible regions, a structure in a series. “So we could have in that it lets researchers look at smaller sys- which can lead to broad conformational huge throughput,” he says. This throughput tems and, she says, drive methods for explor- ensembles and different binding equilib- could support fragment screening in drug ing single protein complexes. XFEL-based ria. With VISAVIX, structural analysis is development, improve the research-orient- ed structural biology pipeline and address crystallization and screening bottlenecks. BOX 2: SOME X-RAY FREE-ELECTRON LASERS Fragment screening at the XFEL could EuXFEL “vastly increase” the speed of tasks done FERMI facility at , allowing bigger libraries Linac Coherent Light Source at Stanford (LCLS) of fragments and reducing the amount of Pohang Accelerator Laboratory (PAL) X-ray Free-Electron Laser protein needed. SPring-8 Angstrom Compact Free Electron Laser (SACLA) Researchers can begin to transfer tech- Swiss Free-Electron Laser (SwissFEL) niques from conventional crystallography

844 | VOL.14 NO.9 | SEPTEMBER 2017 | NATURE METHODS TECHNOLOGY FEATURE

that hold more than 50,000 crystals. If they Another issue is background scattering, can scan the chip at 1 kHz, data for a solved which should be lower than it is in most structure could be collected in less than a XFEL-based experiments, says Meents. Just minute, says Meents. That 1-kHz data acqui- reducing scatter will deliver a big boost in sition rate is far from the EuXFEL’s 5-MHz resolution. repetition rate with pulses delivered in ,Univ. of Wisconsin of ,Univ. bunches. Handling pulse trains et al. These ‘bunch trains’ are challenging for A day of beamtime at the EuXFEL is worth users, especially for fixed-target experi- 225 days at LCLS, says Chapman, but it’s ments. For those experiments, Meents a day with some challenges. The Adaptive

A. Hosseinizadeh A. Hosseinizadeh would need to move the sample 10 mm dur- Gain Pixel Detectors must keep up with the ing the 200-nanosecond pause between the data deluge delivered by the X-ray ‘trains’, The X-ray diffraction volume of a virus. pulses in order to shoot the next sample which are 2,700-pulse bunches at 4.7 MHz position. One way to address this is by scan- that repeat ten times a second with 220 ­accelerated because data images can be ning the beam over the sample. To do so, ns between pulses. The detector can store ­captured for just selected samples. one could set up a reflective X-ray mirror a 3,500 frames per second, which is enor- Both mass spec and XFEL work under few meters upstream of the sample. Between mous capacity, he says, but it’s only around vacuum conditions, which simplifies the pulses, the lever arm would move the mir- 10% of a full beam. Even though experi- coupling of these two approaches. Last year ror according to calculated angular values. menters can’t leverage all generated data, Uetrecht and her team ran a test at FLASH, a It’s challenging and still in development, the pulse enables many types of experi- smaller version of the EuXFEL. “We learned he says, but it’s more realistic than trying to ments. A researcher could set up a reaction a lot about the system,” she says, most impor- scan the sample one hundred times faster. and probe it at this rate and capture many tantly that high-intensity radiation did not With optical imaging, most labs take biological processes. For higher resolution, damage the mass spec’s electronics. The multimodal approaches, but with XFELs experimenters can interleave data from group is analyzing their fragment data now, researchers must get as much information as different trains and initiate a reaction at and she hopes to run her first experiments at possible from a single shot exposure. They slightly different times. the EuXFEL in 2018 or 2019. want all the photons: both the elastically and The EuXFEL beam’s high pulse rate is For XFEL-based studies of bovine virus the inelastically scattered and fluorescence beneficial in terms of quick turnaround BEV2, biophysicist Alke Meents developed photons. Detectors need energy resolution and sample consumption, but, says Hajdu, a ‘fixed-target approach’ to ease sample that lets experimenters distinguish between one needs to leave time between shots for loading and reduce the number of crystals photons from different interaction process- the “smoke to clear.” Debris from the first needed7. It uses beamtime more effectively, es. In this vein, Meents and his colleagues shot may still be in the interaction region he says, and can attract more users to serial are currently exploring how to use Compton that researchers are studying. “Ideally one femtosecond crystallography (SFX), a tech- scattering to image larger biological samples would need evenly distributed sets of puls- nique in which many microcrystals are using synchrotron radiation. Combining all es,” he says. The pulse pattern has some his- delivered into the beam to generate a set of of these efforts along with new data analysis tory: it was meant to enable certain types diffraction data. Meents joined Chapman’s methods, he says, should help them catch up of experiments and then became group last year. He needed crystals to get a to the electron microscopy community. “ingrained” in the project, and it’s not ideal detectable signal above noise in the BEV2 With XFELs, scientists can potentially for structural studies. “We have to live with experiment, but XFELs will make it feasible approach the scale reached with electron it for the moment,” he says. Changing it will to ‘image’ non-crystallized particles. microscopy for single-particle imaging, says cost money. Some remedies experimenters Meents is a collaborator on a few EuXFEL Meents, but X-rays damage the sample with can try include beam-tuning to drop pulses proposals—for example, he is involved radiation. Researchers in electron microsco- with the SFX user consortium at the py obtain around 1,000 times more informa- Single Particles, Cluster and Biomolecules tion out of a sample than researchers work- ­instrument—but he has not submitted ing with X-rays before the sample sustains his own proposal. He and his colleagues too much damage. are currently advancing their fixed-target When using XFELs, experimenters can sample-delivery method in which thou- ‘outrun’ radiation damage because they col- sands of crystals are loaded onto a chip and lect data right after X-ray diffraction but raster-scanned. He is exploring how to get before the sample is destroyed. “Despite all the beam to scan ‘over’ the sample, akin to the software methods, which need to be fur- techniques in laser-based micromachining. ther developed for this, I see a great need and The team is taking the idea to nearby potential in having the optimal instrumen- PETRA III, a synchrotron where more tation for such kind of experiments,” says beamtime is available and it’s possible to Meents. The brightest X-ray beams can now A. Meents, CFEL A. Meents, make the major instrument modifications be nanofocused, but experimenters often this experiment requires. The scientists rely on sample holders and instrumentation Structure of a bovine virus determined with a have experimented at LCLS with large chips designed 20 years ago, he says. ‘fixed-target’ approach.

NATURE METHODS | VOL.14 NO.9 | SEPTEMBER 2017 | 845 TECHNOLOGY FEATURE from the pulse train or spacing pulses far- with serial crystal- more telling about the binding of a drug ther from one another. lography. “Together candidate and a protein target. Overall, the EuXFEL holds tremendous with investigation potential for structural biology, says Uetrecht. of time-resolved More XFELs “However, it is not a true single-molecule ­ligand-binding,” The EuXFEL is operational, the Pohang technique yet,” she says. A myriad of images he says, “I am con- Accelerator Laboratory XFEL in South of a single species must still be combined to vinced this will have Korea has lased, and SwissFEL will soon obtain a structure. “It would be awesome to an impact on com- be operational. Tabletop XFELs are under record 3D structures at high resolution in a putational analysis development. LCLS has built a new beam- single shot, thus really for each particle,” she and design of future line called Macromolecular Femtosecond says. Holographic approaches show great Charlotte Uetrecht drug molecules.” Crystallography, and LCLS-II, a facil- potential, especially for large objects such as combines mass ity upgrade, is in the works. As Chapman cells. spectrometry and Data torrent explains, LCLS-II involves the same XFEL-based research. Experimenters at accelerator technology used for the EuXFEL Applied research the EuXFEL will with its high repetition rates. “The competi- Besides the EuXFEL, Europe is home to harvest a torrent of data about their biologi- tion between facilities is absolutely fantastic SwissFEL, part of the Paul Scherrer Institute cal molecules. Data analysis tools and data for users like us,” he says. There are not yet as (PSI) in Villigen, Switzerland. Pilot experi- portals are emerging8 (see “XFEL projects, many XFELs as there are synchrotron sourc- ments start later this year and it will open to tools, data portals” on Methagora). With its es, but XFELs “have proven to be much more the research community in 2018. In contrast repetition rate, the EuXFEL is at the extreme useful and versatile than was ever imagined.” to the EuXFEL’s 27,000 pulses per second, end of the big data challenges now common Structural biology was once a scientific SwissFEL delivers 100 pulses per second, in many fields, says Uetrecht. “Automatic frontier. That changed as the field grew after says Michael Hennig, CEO of leadXpro, a data-sorting to discard empty or back- the first diffraction PSI spin-out. He spent two decades at Roche ground-only images will be essential,” she patterns were gen- directing structure-based research. This says. The detectors at the EuXFEL “feature erated in 1934, says pulse rate difference involves different exper- a veto system to do exactly that.” Hajdu. Unlike elec- iments and measurement strategies. In VISAVIX experiments, there will be tron microscopy, SwissFEL is set up for academic and com- fewer big data challenges given that the which was devel- mercial research, says Hennig. His company researchers are likely to see many “empty oped during World focuses on structural biology and biophysics shots,” says Uetrecht. Because of the low War II, or protein methods for drug discovery, most notably for background provided by the mass spec- crystallography, membrane protein targets such as ion chan- trometer, it will be comparatively easy to dis- which dates back to I. Andersson, Uppsala University I. Andersson, nels, transporters and G-protein-coupled tinguish empty frames and real data. What the early 20th cen- receptors. Working with solubilized mem- could be difficult are the potentially long The XFEL community tury, XFELs are a brane proteins, he and his team use cryo-EM data analysis times for single-particle imag- shows a frontier spirit, young technology: or X-ray crystallography to study binding ing compared to the data-acquisition rate. says Janos Hajdu. the first lasing at the properties of potential drug molecules and “In the long run, I really hope that the col- world’s first XFEL to design future drugs. Hennig wants to find leagues at EuXFEL will provide an integrated was in 2009. “Everything has to be developed lead molecules with “novel chemotypes” with data analysis pipeline,” she says. Given the from scratch,” he says. XFEL-based research pronounced specificity and efficacy. With amount of data generated in experiments, is exciting, and one day soon it will have structures in hand, drug developers can take just the transfer of these data to home insti- industrial applications, too. International on challenging systems as they, for example, tutions will be time-consuming. collaboration in the XFEL community works, search for potential small molecules that One measure of XFEL success, says and he senses “a frontier spirit” that is spark- target protein–protein interactions of trans- Ourmazd, is the degree to which oth- ing new science. porters. He has performed experiments at ers, including pharmaceutical industry XFELs such as LCLS in the United States and researchers, are convinced that the use of Vivien Marx is technology editor for SACLA in Japan, and he and his team have these machines and the existing and emerg- Nature Methods ([email protected]). applied for beamtime at the EuXFEL. ing data analysis algorithms is beneficial. He 1. Pellegrini, C. X-ray free-electron lasers: from Interaction with beamline scientists is criti- wishes data analysis tools were integrated dreams to reality. Phys. Scr. T169, 014004 (2017). cal from sample prep to data analysis, he says. into the planning for the facilities. “We have 2. Neutze, R., Wouts, R., van der Spoel, D., Weckert, E. & Hajdu, J. Nature 406, 752–757 His team might, for example, request adjust- learned to build magnificent machines, but (2000). ments to the beamline settings. “One of the too often neglect to give them brains,” he 3. Chapman, H.N. et al. Nature 470, 73–77 (2011). most thrilling areas of research with XFELs says. 4. Boutet, S. et al. Science 337, 362–364 (2012). 5. Altarelli, M. & Mancuso, A.P. Philos. Trans. R. Soc. are investigations of protein dynamics,” says Especially given the complexity and size of Lond. B Biol. Sci. 369, 20130311 (2014). Hennig. Compared to what is observed with the data sets from XFEL-based experiments, 6. Ourmazd, A. et al. Nat. Methods 14, 875–879 traditional cryo-crystallography, he and his it’s critical to interact with facility staff, both (2017). 7. Roedig, P. et al. Nat. Methods 14, 805–810 team see evidence for different conforma- before and during experiments, says Hennig. (2017). tions and B-factors when structural analysis SFX data analysis is one key area where 8. Maia, F.R.N.C., White, T.A., Loh, N.D. & Hajdu, J. is performed at room temperature such as advances will make electron density maps J. Appl. Crystallogr. 49, 1117–1120 (2016).

846 | VOL.14 NO.9 | SEPTEMBER 2017 | NATURE METHODS