The complementarity of serial femtosecond crystallography and MicroED for structure determination from microcrystals

Abstract

In recent years, nano and microcrystals have emerged as a valuable source of high-resolution structural information owing to the invention of serial femtosecond crystallography (SFX) with X-ray free electron lasers and microcrystal electron diffraction (MicroED) using electron cryomicroscopes. Once considered useless for structure determination, nano/microcrystals now confer significant advantages for static and time-resolved structure determination from a wide variety of difficult-to-study targets. MicroED has been used to obtain sub-Ångstrom resolution maps in which hydrogen atoms can be clearly resolved from only a few nano/microcrystals, while SFX has been used to probe protein dynamics following reaction initiation on time scales from femtoseconds to minutes. We review these two complementary techniques and their abilities for high-resolution structure determination.

Introduction

Serial femtosecond crystallography (SFX) with X-ray free-electron lasers (XFELs) and Micro-electron diffraction (MicroED) have emerged as two powerful microcrystallography techniques for structural biology (Fig. 1). Recent breakthroughs in MicroED and SFX exploit the limited crystal size while taking distinct approaches to mitigating radiation damage: ultra-low electron doses and ultrafast XFEL pulses, respectively. For MicroED, the use of ultra-low doses on thin nanocrystals (sub-micron thickness) has minimized multiple scattering events and absorption due to the strong interaction between electrons and matter, enabling structure determination at atomic resolution from unprecedentedly small protein crystals [1,2]. Such small crystals are currently almost unusable for X-ray crystallography, due to the much lower elastic scattering cross-section (probability of elastic scattering) in the X-ray case, necessitating extremely high X-ray fluxes. The high ratio of inelastic to elastic X-ray scattering severely exacerbates radiation damage, but XFEL pulses can be sufficiently brief to outrun secondary radiation damage [3]. The main advantage is that protein microcrystal sizes routinely used in SFX range from 5 – 30 microns, allowing rapid reaction initiation by photoactivation, chemical mixing or other methods, to probe protein dynamics on femtosecond to multi-second timescales, at room temperature, with no structural damage. Here, we outline the distinct yet complementary nature of these techniques, and review how they have opened the door to many new exciting structural studies.

Fig. 1. Overview of structure determination by XFEL and MicroED.

Both techniques use similar microcrystalline starting material and produce similar high-resolution structures; however, the overall workflow is significantly different between the two. Both XFEL and MicroED have their own unique strengths and weaknesses, which are described in this review.

Serial Femtosecond Crystallography

One of the critical bottlenecks in the field of protein crystallography is the growth of large, well-ordered crystals suitable for conventional X-ray crystallography. For every elastic X-ray scattering event that produces useful structural data, an order of magnitude more X-rays are absorbed [4], depositing energy in the crystal through a chain of inelastic scattering events, leading to local and global damage. In large crystals the radiation dose is spread over many identical protein molecules. However, for difficult targets, such as membrane proteins or protein complexes, the process of optimizing conditions for large, high quality crystals can take a prohibitively long amount of time and resources. In addition, numerous types of protein crystal disorder, such as misaligned mosaic blocks or variations in unit cell within the crystal, are less likely to manifest in microcrystals, further compelling the development of microcrystallography.

Significant improvements to microfocus synchrotron beamlines, including brighter, more tightly focused beams, and fast-readout detectors to handle the much shorter crystal lifetimes, have facilitated room-temperature data collection from gradually smaller microcrystals [5]; however, unless radiation can be outrun, there is a limit to the smallest crystal that can produce useful structural information [6,7]. Microcrystallography at XFELs, however, adheres to the ‘diffraction-before-destruction’ approach [8], using XFEL pulses of 20 – 50 fs duration to yield crystallographic data before the hydrated, room-temperature sample inevitably explodes from the very high levels of ionization. Only “still” diffraction snapshots, with partial reflections, can be recorded during such ultra-brief exposures. To obtain accurate structure factors, thousands of indexed diffraction patterns are merged, averaging over the random orientation, size distribution and potential anisomorphism of the microcrystals, as well as the stochastic shot-to-shot variation of the intensity, spectrum and focus of the XFEL pulses [9]. Despite the seemingly insurmountable uncertainties inherent in this approach [10,11], in less than a decade, there were several developments in data analysis, such as orientation refinement, scaling, post-refinement and detector geometry calibration to sub-pixel accuracy, which decreased the requisite number of SFX diffraction patterns for accurate measurement of structure factors by about an order of magnitude [12–15]. Merged SFX data allows highresolution room-temperature structure determination (Fig. 2) and is now sufficiently accurate to measure very small differences in structure factors, as needed for de novo phasing [16–20], ultrafast photoactivated time-resolved (TR) SFX [21–25] or TR SFX of ligand binding [26–28]. Enabled by the two-color lasing capability at XFEL facilities [29], the two-color diffraction scheme was also proposed to further reduce the error from random crystal sizes and orientations [30].

Fig 2. Recent high-resolution structures determined by XFEL and MicroED.

(A) High- resolution Photoactive Yellow Protein (PYP) structure determined by XFEL. This structure (PDB ID: 5HD3) was one of multiple PYP structures determined as part of a time-resolved (TR)-SFX experiment. The fine time resolution provided by TR-SFX combined with the high-quality and high-resolution of the diffraction data facilitated the identification of early structural changes in the PYP chromophore upon exposure to light [21]. (B) Representative MicroED structures. Due to improved methodology, structures determined by MicroED have been determined at increasing resolutions. The example structures shown here from Sup35 (PDB ID: 5K2E [52]) and Trypsin (PDB ID: 5K7R [48]) were determined using standard MicroED data collection and processing procedures and data originating from 6 and 10 crystals, respectively. Density maps (2Fo-Fc) in blue are contoured at 1.5σ for all structures shown in both A and B.

Numerous room-temperature sample delivery options have been developed over the last decade to facilitate rapid replenishment of protein microcrystals in the XFEL path. These include microfluidic injectors to deliver microcrystals in jets of liquid or high-viscosity media, raster scanning of fixed targets [31], or electrospinning [32]. Most commonly, solvated microcrystals are injected in a micron-scale liquid jet from a gas-dynamic virtual nozzle (GDVN) [33]. For photoactivated proteins (or for release of photoactivated caged molecules), a femtosecond or nanosecond laser flash is used to initiate the reaction upstream of the XFEL interaction region [34]. Simply changing the distance between pump and probe (or the jet velocity) provides a range of time points from femtoseconds to seconds, limited by the pump laser duration, flow rate and jet stability downstream of the nozzle. Compared to time-resolved crystallography at 3^rd generation synchrotrons, pump-probe SFX on microcrystals enables a higher extent of reaction initiation by avoiding the uneven reaction initiation in larger crystals due to pump laser absorption. The time resolution is also increased by at least 3 orders of magnitude (100 fs versus 100 ps). To study ligand-induced structural changes, mix- and-inject systems have been developed to control the reaction time prior to diffraction [35]. Mix-and-inject SFX involves microfluidic mixing upstream of the X-ray probe [26–28], and the time resolution accessible in chemical mixing is determined by the mixing time and diffusion rates [36]. Finally, the development of the high-viscosity injector [37] was a critical breakthrough for SFX by (a) significantly decreasing the sample volume that was wasted between XFEL shots relative to a liquid injector, with a low repetition rate XFEL up to 120 Hz and (b) facilitating delivery of fragile microcrystals of membrane proteins, such as G protein-coupled receptors (GPCRs) in lipidic-cubic-phase (LCP), their common growth medium [38].

While SFX using XFELs has very exciting and unique advantages, there are still some hurdles to overcome. The key impediment to the widespread use of XFELs in structural biology is the scarcity of available beam time. Currently, there are only five hard X-ray laser facilities operating in the world, the Linac Coherent Light Source (LCLS) at SLAC in the USA [39], SPring-8 Angstrom Compact free electron Laser (SACLA) in Japan [40], the Pohang Accelerator Laboratory X-ray Free-Electron Laser (PAL-XFEL) in Korea [41], the European XFEL (EuXFEL) in Germany [42] and the SwissFEL in Switzerland [43]. Only a small number of XFEL experiments can be performed concurrently at any XFEL facility. In addition, the slowly increasing number of XFEL facilities does not meet the needs of the global user community. For these reasons, XFEL beamtime will continue to be highly competitive and oversubscribed.

Structure Determination by MicroED

MicroED is a recently developed electron cryomicroscopy (cryoEM) technique [1,2,44], which uses a conventional transmission electron cryomicroscope (cryo-TEM) to determine structures of biomolecules using electron diffraction (Fig. 2B) rather than imaging, as is used in single-particle cryoEM. Relative to X-rays, electrons deposit significantly less radiation damage per useful scattering event. This more efficient scattering allows MicroED to be used to determine high-resolution structures from crystals that are only hundreds of nanometers thick. Since the publication of the first MicroED structure, there has been continuous optimization of the method to allow the collection of higher quality data and resulting structures. This is highlighted by the fact that the first proof-of-concept MicroED lysozyme structure in 2013 was determined at 2.9 Å resolution [45], and this was improved shortly thereafter in 2014 with the implementation of continuous rotation data collection [46,47], which yielded an increase in resolution to 2.5 Å. Fast forward a few years later, and with further methods development and improved data processing, similar lysozyme nanocrystals are now capable of yielding a final refined structure at 1.8 Å resolution [48].

One of the greatest benefits of using MicroED is that it can be performed using commonly available and relatively economical (~$1.5 M) cryo-TEMs that are equipped with a sensitive high-speed camera. Detailed protocols for performing MicroED data collection and processing are also available [49,50]. The frozen-hydrated nanocrystals are prepared in a fashion that is identical to single-particle cryo-EM sample preparation [51]. Briefly, an aliquot of nanocrystals are pipetted onto a holey carbon EM grid, excess buffer is blotted away and the sample is rapidly plunged into liquid ethane so that the buffer is vitrified preventing the formation of crystalline ice. This process currently limits MicroED to samples at cryogenic temperatures, or dry samples at room temperature. Crystal screening is performed in imaging mode, and the ability to visualize the crystals prior to the collection of data is a major advantage of MicroED over other micro-crystallographic methods. Crystals are assessed based on their, size, shape, and orientation, which greatly increases the data collection rate during a MicroED session. When high-quality crystals are identified, ultra-low dose diffraction data sets are collected by continuously tilting the crystals from 0 to ~60° with respect to the incident electron beam, while the detector constantly records the diffraction data. This creates a continuous series of diffraction frames collected from single crystals, which are similar to data collection in conventional X-ray crystallographic experiments. Therefore, MicroED data can be integrated and scaled using standard programs for X-ray crystallography and structure determination can be also performed within commonly used crystallographic program suites, with the understanding that the atomic scattering amplitudes are quite different for electrons and X-rays (see article by Marques et al., in this issue).

With MicroED, care must be taken to ensure the crystals are not too thick, otherwise the effects of dynamic scattering and the absorption of the electron beam by the sample will be too severe for structure determination. Past studies have confirmed that crystals on the order of 400 nm thick can be used for high-quality structure determination; however, more detailed studies are underway to define the upper limit of crystal thickness. Also, provided the crystals diffract to a sufficiently high resolution and the number of atoms per molecule is not excessive, the recorded intensities are accurate enough to phase the data by direct methods [52,53]. For MicroED of macromolecules, a current limitation is the lack of a robust method for experimentally phasing novel structures. Thus far, the majority of structures determined by MicroED have relied on the molecular replacement method using the X-ray structure of a known homologous protein. Currently, there is a drive within the MicroED community to develop phasing procedures, with the most straightforward choices being isomorphous replacement methods or phasing by calculating the Fourier transform of a 3D density map of the macromolecule derived by EM imaging and tomographic 3D reconstruction of the microcrystals. Much of the initial MicroED validation and methods development have been performed on model protein microcrystals [46,48,49,54–56]; however, several novel structures have now been determined using the method. Relevant to applications in drug discovery, a recent MicroED structure of the CTD-SP1 construct of HIV-1 Gag with and without a bound maturation inhibitor showed how the inhibitor stabilizes the 6-helix bundle of the domains, thereby preventing access of HIV-1 protease to perform cleavage between CTD and SP1 [57]. One of the most impressive applications of MicroED has been the study of peptide fragments that are related to various neurological diseases that had resisted structural determination by all other methods. These very high-resolution structures provided new insights into the structural basis of amyloid aggregation and toxicity [58–60] and have recently led to the design of potential therapeutic agents [61]. One particular structure of a prion protofibril was determined to 0.72 Å resolution [62]. Because of the high resolution and sensitivity of electron diffraction for hydrogen atoms [63], the MicroED map allowed the accurate placement of hydrogen atoms in the atomic model, which revealed a network of stabilizing hydrogen bonds in the structure [62]. More recently, MicroED was used to efficiently solve atomic-resolution structures of small organic molecules, pharmaceuticals and natural products [53,62,64,65]. Beyond the study of organic and biological molecules, MicroED has also been used to determine the organization of novel metallic nanoparticle structures [66].

The synergy between XFEL and MicroED

While both XFELs and MicroED are capable of producing high-resolution structures from crystals that are much too small for traditional X-ray crystallography, each approach has its own relative strengths and challenges (Table 1). Researchers can choose the technique that will best answer the structural questions of their project (Fig 3). The methods often complement each other, and there have already been examples of the two methods being successfully used to support each other [59,67,68].

Table 1.

Comparisons of SFX and MicroED

	SFX	MicroED
Sample size	Typically 5–15 μm	Less than ∼500 nm thicka
	Smallest crystals used for 2 Å structure determination were ∼0.2 × 0.2 × 0.4 μm3 spheroids [76]	Limited due to absorption and dynamic scattering
Wavelengths/energies	∼5–12 keV	200–300 keV
Phasing options	Molecular replacement, isomorphous replacement, single/multiwavelength anomalous dispersion, high-intensity radiation damage induced phasingb	Molecular replacement, direct methods, imaging,b isomorphous replacementb
Resolution		Proteins: 1.6–3.2 Å
	Proteins: up to 1.45 Å	Peptides/organic molecules: 0.7–1.4 Å
Temperature	Room temperatured	Cryogenic, room temperaturec
Sample delivery/environment	In air or in vacuum; liquid or high-viscosity microfluidic injection; raster scanned fixed target; electrospinning; acoustic droplet injection with conveyor belt drive Several	In the vacuum of the cryo-TEM. Samples deposited on carbon coated EM grid and moved via the stage of the cryo-TEM As little as 100 nL
Volume of sample required	100’s of μl to mLs
Radiation damage	Outrun secondary damage	Limited to ∼2–5 e−/A2 for high-resolution
Time-resolution	fs–s	ms–s
Data collection time	Tens of minutes to hours. Could be on the order of seconds at new MHz-rate XFELs.	A few minutes per data set
Data volume	GBs (processed), TBs (raw)	MBs (processed), GBs (raw)
Software	cctbx.xfel, crystfel, Standard X-ray crystallography data processing and refinement programs	Modified X-ray Crystallography data processing and refinement programs
Access to facilities	Five hard XFELs online globally.	Standard cryo-TEM equipped with high-speed CMOS detector
	Competitive beamtime

^aVaries depending on unit cell size and orientation. Ultimately limited by absorption and dynamic scattering.

^bThese are potential phasing methods, to date they have yet to be implemented.

^cRoom temperature data collection is only valid for crystals that can survive being dried.

^dHydrated microcrystals sprayed into vacuum in micron-scale liquid jet cool extremely rapidly.

Fig 3.

Roadmap for performing XFEL and MicroED Studies.

There is also considerable overlap between the XFEL and MicroED communities in the development of novel methods and improvements for the two techniques. This connection has been particularly important for crystallization and handling of nano and microcrystal samples. Many of the crystal growth conditions and sample handling procedures used for serial crystallography experiments have been adopted to create samples for MicroED [46,48,69,70]. Recently, several approaches for fragmenting large but poorly-ordered crystals into smaller higher quality microcrystals were reported [48]. The resolution and quality of data could be dramatically improved using fragmentation, yielding several high-resolution MicroED structures. While this approach focused on MicroED sample preparation, the same methods can be easily adopted for XFEL analysis, provided that larger quantities of crystals can be grown in batch.

In the future, further combinations of the strengths of each technique promise to bring great new insights to structural biology. For example, a unique attribute of electron diffraction is that electrons are very sensitive to charge and chemical bonding [71–75] (and see article by Marques et al., in this issue). The proper modeling of charge and bonding effects for MicroED data is an area of active research. When improved model and scattering factors for MicroED are implemented, an experiment could be conducted which uses the incredible time resolution made possible by XFELs along with charge and bonding information that can be revealed by MicroED analysis to study enzymatic mechanisms in exceptional detail. In this way, XFELs could help elucidate the mechanism at very fine time scales, while MicroED would allow the direct visualization of charges and bonding in transitions states that occur at longer time scales (ms to s).

Conclusions

The use of microcrystals in XFEL crystallography and MicroED has allowed the study of new samples, which were previously not accessible by traditional methods. Because of this, microcrystallography is quickly becoming an indispensable tool for the structural biologist. Both methodologies will benefit from continued collaboration between XFEL and MicroED researchers, where the strengths of each technique can be used in tandem to answer new questions and conduct novel experiments. The continued development and implementation of these methods – both individually and in combination – promises to bring exciting new discoveries to the field of structural biology.