Atomic resolution structure determination by the cryo‐EM method MicroED

The electron cryo‐microscopy (cryoEM) method MicroED has been rapidly developing. In this review we highlight some of the key steps in MicroED from crystal analysis to structure determination. We compare and contrast MicroED and the latest X‐ray based diffraction method the X‐ray free‐electron laser (XFEL). Strengths and shortcomings of both MicroED and XFEL are discussed. Finally, all current MicroED structures are tabulated with a view to the future.


Introduction
Structural biologists use various methods to reveal inter-and intra-molecular interactions to understand the complicated chemical and physical processes of life. The pursuit of visualizing such detailed interactions has predominately relied on X-ray crystallography. However, cryo-electron microscopy (cryo-EM) has in recent years gained incredible momentum with ground-breaking advances in detector technology 1 and the development of new methodologies. 2 Modern cryo-EM diversified into four methods: tomography, single-particle reconstructions, twodimensional (2D) electron crystallography and MicroED (Fig. 1). Cryo-electron tomography was developed to study whole cells and large organelles albeit at relatively modest resolutions of 1 nm. 7,8 In single-particle reconstructions imaging thousands of isolated and purified particles in vitrified ice followed by motion correction, averaging and reconstruction can yield structures close to 2Å in resolution for wellbehaved samples. 4,9 The last two methods in cryo-EM rely on having crystalline material. Electron crystallography can use imaging and/or diffraction from highly ordered 2D crystals to obtain structures of membrane proteins embedded in a lipid bilayer. This method has yielded the structure of the water channel aquaporin-0 and its surrounding membrane at 1.9 Å resolution from 2D crystals only a single protein layer thick. 5,10 Finally, in MicroED vanishingly small threedimensional (3D) crystals of biological material are studied by electron diffraction under cryogenic conditions to reveal unprecedented atomic detail such as the positions of protons in a protein. 6,11 In this review, we highlight the unique strengths of MicroED and summarize current advances in this method.

Brief Overview of MicroED
As with any crystallography experiments, MicroED relies on having well-ordered crystals. The same crystallization robots and setups are used to screen for crystal growth as in X-ray crystallography. However, crystals for MicroED are typically billions of times smaller in volume than those used for X-ray crystallography. A discussion of how to find such small crystals in the crystallization drops was recently published along with a detailed protocol for data collection and processing. 12,13 Once crystals are identified in the crystallization drops (either by negative stain EM, optical means or powder diffraction) they are placed on an EM grid, plunged into liquid ethane for freezing and viewed in a cryo-electron microscope.
The examination in cryo applies three different dose rates sequentially. An initial grid screening at an ultra low dose rate (< 10 26 e 2 Å 22 s 21 ) with low magnification in bright field is performed to survey for thin crystals. Next, the microscope is switched to over-focused diffraction mode at < 10 23 e 2 Å 22 s 21 , where each crystal is inspected individually. If a crystal shows appreciable diffraction, a final dose rate of 0.01-0.05 e 2 Å 22 s 21 is applied to collect the MicroED data set.
During data collection, crystals can be either tilted discretely or rotated continuously in the electron beam (Fig. 2). The initial MicroED data were collected as a series of still exposures, each of which captured one diffraction pattern at a certain discrete angle. 2 As the stage was tilted each time with an increment of 0.1-18, a complete dataset of up to 908 worth of patterns per crystal were collected for determining the structure of lysozyme. With such experimental setup all reflections are only partially recorded, which hampers scaling and merging. This issue was solved by an improved data collection scheme for MicroED called continuous rotation. 12,14 This improved method yields data where reflections are fully recorded over a contiguous sequence of images, as diffraction occurs continuously while the crystal is rotating in the electron beam and the data is recorded as a movie on a fast complementary metal-oxide semiconductor (CMOS)-based detector. 12 Data processing has also improved since the initial proof-of-principle study. Initial still diffraction data was processed using an in-house developed program specifically designed for lysozyme. 15 However, with continuous rotation each frame of the movie contains a wedge of data, typically between 0.1-0.58, which is analogous to images collected using the rotation method in X-ray crystallography. 16 As such, continuous rotation MicroED data can be processed using standard X-ray crystallography data reduction software 17 such as MOSFLM 18,19 and XDS. 20 Detailed guidelines on data processing can be found in reference. 13

Comparisons Between MicroED and XFEL
The ratio of inelastic to elastic scattering in an X-ray experiment is high in comparison with electrons; and X-rays deposit more destructive energy onto the sample. 21 Therefore, in traditional X-ray crystallography very large crystals are needed to obtain interpretable diffraction patterns and to withstand the large radiation dose that accumulates during the collection of a complete rotation series from a single crystal. 22 Newly developed technologies use a more intense and short X-ray pulse to allow data collection from smaller crystals. Data sets are then obtained by merging the intensities integrated on thousands of diffraction patterns originating from millions of crystals. 23 Serial femtosecond crystallography (SFX) at an X-ray free-electron laser (XFEL) can provide highresolution diffraction data from small crystals. The high intensity beam obliterates the crystal after one exposure, but before the crystal is destroyed diffraction is collected. 24 This so-called "diffract before destroy" has become increasingly popular and a number of structures have already been determined by this diffraction method. 23,[25][26][27][28][29][30][31][32] Of course those were facilitated by large instrument development 33-35 and software development for data reduction. [36][37][38][39] The main strengths of an XFEL experiment are that crystals can be smaller than the usual crystals used for traditional X-ray crystallography at home sources or synchrotrons 23 and that time resolved studies of dynamical processes could be conducted. [40][41][42] The shortcomings are that the cost of the experiment is prohibitively high, instrument beam time availability is very low, and difficulties processing the large amount of data that is rapidly produced. Furthermore, recent studies indicate that even SFX data is affected by radiation damage. 43 Since the pulse is too short for the crystal to rotate during the exposure each crystal only provides one still diffraction pattern before it is destroyed, and the reflections that are collected are always partially recorded leading to problems with scaling. Further complications arise from the variance in the exposed crystal volume in each shot, and fluctuations in the pulse intensity. Scaling is achieved by merging the reflections originating from thousands of diffraction patterns. 44,45 Finally, sample delivery is commonly achieved via a nozzle that sprays grams of crystalline material at the diffracting beam, therefore the sample requirement is high and the delivery nozzle can quite frequently clog up stopping the experiments. [46][47][48] MicroED promises to overcome many of the obstacles encountered by an XFEL while maintaining many of the strengths and benefits. The quantity of crystalline material in a MicroED experiment can be much smaller than in an XFEL experiment. The smallest crystals used successfully were of a fragment of a-synuclein. The structure was determined at 1.4 Å resolution from crystals that were only 50 nm thick. 6 With careful data collection and analysis, structures with very limited beam damage can be determined from diffraction patterns of radiation-sensitive biological material recorded using very low electron dose (down to 10 26 e 2 /Å 2 /s) under cryogenic conditions. 14 The equipment needed for a MicroED experiment is relatively cheap and readily available, and no modifications are necessary to the electron microscope. 12 Time resolved and dynamic studies of biological systems can likewise be achieved as activators and inhibitors, pH or light can be used to start or stop a chemical reaction in the crystals right before plunge freezing into ethane and data collection. Such studies have already been conducted by electron crystallography of 2D crystals decades ago. 49,50 A single nanocrystal is sufficient for an entire data set to be collected and determined by MicroED. 14,17 Here the data is collected by continuous rotation, so full reflections are recorded, and data analysis can be done using standard crystallographic software like MOSFLM 19 and XDS 20 without modifications to the data reduction software. 14 Data collection and data analysis take 10 minutes, respectively, followed by standard structure refinement. Finally, the highest resolution achieved by MicroED so far surpasses what was achieved in XFEL as well as any other cryo-EM method.
Some of the possible shortcomings in MicroED are shared with any crystal-based methods. Mic-roED depends on having well ordered crystals. If such crystals cannot grow then no crystallographic based method can be applied for structure solution. It is currently unclear what is the largest asymmetric unit that can be investigated by MicroED. To date the largest reported was that of catalase where two catalase proteins occupied the asymmetric unit amassing at 500 kDa. 17 It is possible that for even larger asymmetric unit cells, large detector chips should be used to allow sufficient separation between diffraction spots for effective indexing. Better sample preparation methods are necessary for MicroED. The current cryo grid preparation methods that are blotting based can damage delicate crystals and the water-air interface that is created  can likewise cause proteins to unwind and aggregate. Injection or laminar flow sample delivery methods should be developed to eliminate this issue and to increase the throughput of crystal screening.

Examples
Several protein structures have been determined by MicroED so far (Fig. 3). Those are summarized in Table I and Figure 4. The first structure solved was that of lysozyme at 2.9 Å resolution by still diffraction 2 and then at 2.5 Å resolution by continuous rotation. 14 This was followed by the 3.2 Å resolution structure of catalase, which was determined from a single nanocrystal only 8 protein layers thick, 17,51 and Ca 21 /ATPase. 51 In late 2015 the first novel structures determined by MicroED were published. 6 Fragments of the toxic core of a-synuclein were determined at 1.4 Å resolution. Previous studies have shown that asynuclein is the main component of neuronassociated aggregates or Lewy bodies that cause neurodegenerative diseases such as Parkinson disease. The formation of these aggregates relies on an 11-residue segment of a-synuclein, termed NACore. These NACore crystals, which were smaller than the wavelength of light, were in fact invisible in light microscopy, and the structure determined from 50 nm thick crystals. Such small crystals were not suitable for any X-ray based diffraction studies including XFEL, but they did yield a high-resolution structure by MicroED where H-atoms were observed for the first time by cryo-EM (Fig. 3). A close inspection of the NACore structure revealed that the twist of the b-sheets creates a tension for the protofilament of amyloid aggregates, which likely restrains the growth of the crystals. This new structural information shed light upon amyloid nucleation and could have implications for treatment of neurodegenerative diseases through inhibitor design strategies.
Recently the structures of four prion peptides were determined at 1 Å resolution by MicroED and direct phasing methods. 11 Prior to these examples, all other MicroED structures were phased by molecular replacement, where phases from a previously determined structure are applied to the measured structure factor amplitudes and refined to provide a new solution. 53 However, when the resolution obtained is high enough direct methods can be used for solving the phase problem. 54 This ab initio approach to phasing relies on the relationship between structure factors in reciprocal space, as well as constraints in real space. 54 This method requires very accurate measurements of the diffracted intensities at high resolution, typically around 1 Å . The fact that four structures could be determined by such methods from MicroED data indicates that the intensities collected in a continuous rotation Mic-roED experiment are very accurate and do not suffer from dynamical artifacts as previously wrongly suggested by others. 55

Future Direction
As MicroED matures additional structures will be solved and provide new biological insights from crystals that are vanishingly small. Future developments in the method should focus on phasing and sample delivery. While molecular replacement is powerful, other phasing methods must be employed if no search model is available. Direct phasing can be done if the resolution obtained is close to 1 Å . Otherwise phasing by isomorphous methods using heavy metals may be a viable option. Phasing by imaging crystals can also be achieved or by using low-resolution density maps obtained by single particle methods. Sample delivery needs to be improved and may include electro spray or capillary based methods in the future. Electron scattering factors need to be optimized to account for the scattering observed in a MicroED experiment and the entire method should be automated from data collection to structure determination. With such advances we expect a bright future for MicroED in the structural biology community.