Technical Development of Electron Cryomicroscopy and Contributions to Life Sciences

Introduction

The 2017 Nobel Prize in Chemistry was awarded to Jacques Dubochet (University of Lausanne, Switzerland), Joachim Frank (Columbia University, USA), and Richard Henderson (MRC Laboratory of Molecular Biology, UK), for their pioneering works in the development of electron cryomicroscopy and image analysis for the structural analysis of biological macromolecules, such as proteins and nucleic acids. The three-dimensional (3D) structure of biomolecules is the basic and important information not only for life sciences but also for medical sciences and drug design, and electron cryomicroscopy (cryoEM) has become a powerful tool for high-resolution structural analysis over the past several years and now established its position as one of the essential techniques for structural analysis in addition to conventional X-ray crystallography and NMR. Electron cryomicroscopy receives much attention because it can achieve near-atomic resolution in structural analysis only with a very small amount of solution samples, as small as a few tens of μgs, without the need of crystallization. The Nobel Prize was awarded to recognize the contributions of the above three researchers as the founders of this technology. Here, we describe the concept of electron cryomicroscopy and image analysis as a technique for structural biology, the history and present state of our own technological development, and the future potential of this technique for life and medical sciences through further development.

Structural analysis of biological macromolecules by electron cryomicroscopy

The basic mechanisms that drive and support biological activities are highly shared by diverse organisms, from microorganisms, such as bacteria and yeast, to multicellular organisms, such as animals and plants as well as humans with higher-order brain functions. All of these functions are determined based on the structures of proteins and nucleic acids with 3D arrangements of so many atoms, from a few thousands to tens of thousands. Moreover, their structures are not solid like bulk materials of metals and ceramics but are very dynamic and flexible and are designed to function by actively utilizing thermal fluctuations. One of the major challenges in life science is the elucidation of mechanisms that determine and express these functions, and it is necessary to look at the 3D structures of so many biological macromolecules in various states that are involved in various biological functions. The number of 3D structures we need to solve would range from a few hundreds of thousands to a few million.
The powerful feature of cryoEM, especially single particle image analysis, is that there is no need for sample crystallization that is essential for X-ray crystallography and that there is virtually no upper limit in the size of molecular complexes unlike NMR. However, since the majority of the interatomic bonds that maintain their 3D structures are non-covalent bonds, such as hydrogen bonds, salt bridges and van der Waals contacts, the structures are very sensitive to electron beam irradiation damage, a few orders of magnitude worse than that of metals, ceramics and semiconductors. Therefore, to record high-resolution images of biological macromolecules in their native state without much damage, it is necessary to embed them in an amorphous ice thin film by rapidly freezing their aqueous solutions and record their images by a transmission electron cryomicroscope (cryoTEM) with a specimen stage cooled by liquid helium or liquid nitrogen to a low enough temperature to prevent the vitreous ice from converting to crystalline states. Even at such low temperatures, the electron dose that can be irradiated without much damage to the atomic-level structures is limited to 20 to 30 e^-/Å2, and so cryoEM images tend to be extremely poor due to the intrinsic statistical noise, which is due to a relatively small number of electrons detected in each pixel of an image detector, and the Landau noise, which is due to a large distribution of signal levels of individual electron detection. Since individual molecular images recorded by a cryoTEM are the 2D projection images of molecules embedded in the amorphous ice film in various orientations, it is also necessary to collect a large number of images that are sufficient to cover different orientations with even distributions in order to reconstruct the 3D image at high resolution. So, it is essential to efficiently collect as many molecular images as possible, classify them into each orientation of projection, and obtain an averaged image for each orientation by aligning the position and orientation of the molecular images to increase the signal level while reducing the noise. This procedure is called 2D class average. Then, the relative relationships between the 2D class averaged images in the orientation of projection are determined, and finally the 3D image of the molecule can be reconstructed (Fig. 1). To achieve high-resolution structural analysis, it is important to use a cryoTEM and an image detector both capable of efficiently collecting high-quality, high-resolution cryoEM images. High-precision image-analysis programs and high-speed computers are also required. We will describe the history of our own development of cryoTEM systems to solve such problems and some of the achievements by using them.

Fig. 1　Schematic diagram explaining the process of single particle image analysis.

Dolphins represent biomolecules embedded in a thin film of vitreous ice in various orientations. CryoEM images correspond to their 2D projections with high noise levels. After the S/N is greatly increased by going through 2D classification and average of many 2D projections, a 3D image can be reconstructed.

Advances in cryoTEM, image detector and methods of data collection and analysis

Field emission electron gun

The pioneer of the cryoEM field in Japan is Yoshinori Fujiyoshi (Visiting Professor, Nagoya University). In collaboration with JEOL, he designed and developed a cryoTEM with a very stable, liquid helium-cooled specimen stage for the first time in the world to minimize the electron beam irradiation damage to the limit to enable recording of high-quality cryoEM images of biological macromolecules [1]. I started collaborating with him from around 1990 to start using electron cryomicroscopy for the structural analysis of the bacterial flagellar filament to understand its assembly and supercoiling mechanism. We had been using X-ray fiber diffraction and X-ray crystallography to those days because the achievable resolution by electron cryomicroscopy was rather limited then, but I recognized the potential of the cryoTEM with the liquid helium-cooled specimen stage for high-resolution structural analysis. He taught us the basics of cryoEM techniques from cryoEM grid preparation to minimum-dose imaging of frozen-hydrated ice-embedded protein complexes. Then, a year or so later, I was offered a job from Tsuneharu Nitta, Director of the Central Research Laboratories of Panasonic (Matsushita Electric Industrial Co., Ltd. at the time), to start my own laboratory in a subdivision of the Advanced Technology Research Laboratory, which Panasonic was planning to establish in KeiHanNa Science City as its new basic-research oriented R&D center. The subdivision was called the International Institute for Advanced Research (IIAR), and I was promised to have a generous funding for the development of new equipments for X-ray structural analysis and electron cryomicroscopy. So I asked Yoshinori Fujiyoshi to join this new Panasonic institute to start the IIAR together as an advanced structural biophysics research center. We asked JEOL to introduce a newly developed device, a Schottky-type field emission electron gun (Thermal FEG), to the above-mentioned cryoTEM with the liquid helium-cooled specimen stage. This was the very first cryoTEM with an FEG and was named JEM-3000SFF (Generation 3: G3) (Fig. 2) [1]. Panasonic established the new KeiHanNa institute in 1994, and we moved into the new building and set up our laboratories with this new cryoTEM. We expected a significant improvement in the quality and resolution of EM images by the high coherence of its electron beam by field emission. The improvement was actually remarkable, showing up much higher resolution signals in every cryoEM image we collected. The Fujiyoshi group aimed to solve the structures of membrane proteins, such as bacteriorhodopsin and aquaporin, in two-dimensional crystals, and our group aimed at analyzing the structure of large helical assemblies of macromolecules, such as the bacterial flagellar filaments. Because only photographic films were available as the image detector in those days, the efficiency and throughput of high-quality image data collection were very poor. So it took more then several years for the structural analyses to reach near-atomic resolution, but we were able to obtain many impactful results by the mid-2000s, with atomic-resolution structures of membrane protein 2D crystals and macromolecular complexes [2-6]. Our structural analysis of the bacterial flagellar filament attained a resolution close to 4 Å by careful image analysis of highly-selected high-quality cryoEM images of the filaments corresponding to only 40,000 flagellin molecules, and the polypeptide backbone folding and large side chains were clearly resolved (Fig. 3) [6].

Fig. 2　Development history of cryoTEMs in our group.

Newly incorporated elemental technologies and equipments are indicated in red characters.

Fig. 3　Molecular structure of the flagellar filament revealed by cryoEM image analysis.

The bacterial flagellum is a motility organelle with a rotary motor and a helical filamentous propeller. The flagellar filament is a large helical assembly of tens of thousands of flagellin molecules. By using the JEM-3000SFF cryoTEM and photographic films as the image detector, we collected cryoEM images and solved the structure of the flagellar filament at around 4 Å resolution. The main chain folding and many large side chains were clearly visualized for the first time by cryoEM image analysis of biomolecules, and this allowed us to build a complete atomic model of this huge protein assembly.

CCD camera and Ω-type energy filter

The Graduate School of Frontier Biosciences was established in Osaka University in 2002, and we moved our laboratory to the Nanobiology building in 2004. Just before that, we obtained a government funding from the MEXT as part of its supplementary budget and asked JEOL to introduce a couple of newly developed devices to the cryoTEM with the liquid helium-cooled specimen stage. This was JEM-3200FSC (Fig. 2) and was an improved version of Fujiyoshi’s fourth generation cryoTEM (G4) [1] that was introduced to the Riken Harima Research Institute at the SPring-8 site. One of the new devices was an improved version of the in-column Ω-type energy filter, and this was introduced to improve the signal to noise ratio (S/N) of cryoEM images by removing the majority of inelastically scattered electrons, which lost the coherence due to energy loss and therefore contribute only to the high background noise and not to the high-resolution EM image. We found nearly two-fold improvement in the image S/N just by this energy filter [7]. Another device was a 4K × 4K CCD image detector (TVIPS F415MP), which was already used for electron diffraction recording but not for imaging due to its lower resolving power than photographic films. Its resolving power is lower because each electron forming an EM image has to be converted to photons by a thin layer of scintillator on top of the glass-fiber coupling block and the image of each electron becomes blurred by the electron scattering within the scintillator. Even with this disadvantage, the CCD markedly improved the efficiency and throughput of high-quality image data collection because we can see the EM image and its Fourier transform immediately after exposure, and this allows us to make a quick and reliable judgement on the quality and resolution of individual cryoEM images during data collection [7]. We were also able to dramatically improve the efficiency of high-quality image data collection further by increasing the specimen temperature from 4 K to around 50 K by stopping the supply of liquid helium from the in-column tank to the top-entry specimen pod. Although the radiation damage can be minimized at 4 K, most of the cryoEM images suffered from local, directionally biased image blurring due to the charge up upon electron beam irradiation because of the extremely poor electrical conductivity of the thin specimen ice film at the extremely low temperature. It was less than a few % of collected images that could be used for image analysis. We solved this difficult problem by increasing the specimen temperature to about 50 K, and almost all of the collected images became sufficiently high quality to be used for image analysis [7].
These technological improvements and advances made previously multi-year projects completed within a couple of weeks from data collection to 3D image reconstruction and allowed us to solve the structures of many different, interesting biomolecular assemblies, such as the bacterial flagellar hook, muscle actin filament, the ParM filament that segregates plasmids for bacterial cell division, the thin needle tube of the virulence type-III secretion system of pathogenic bacteria and the actomyosin rigor complex, all at 5 - 7 Å resolutions. We were able to build reliable atomic models to gain insights into the mechanisms of their functions by docking and refining the available crystal structures to the 3D maps [7-12]. The structure of the stacked disk formed by the tobacco mosaic virus coat protein was solved at 3.8 Å resolution also within a couple of weeks to allow many of the side chains to be visualized (Takashi Fujii, unpublished).

Side entry liquid helium-cooled stage

he liquid helium-cooled specimen stage developed by Fujiyoshi in the 1980’s was the top entry type. It was mechanically very stable, but electron cryotomography could not be done because the stage had no tilting mechanism. Special contrivance was made for imaging 2D crystals in many different tilt angles to cover much of the 3D Fourier space for high-resolution 3D image reconstruction, but it was still time consuming. So Fujiyoshi decided to redesign his top entry stage to implement a tilting mechanism and applied for a MEXT budget called Special Coordination Funds for the Promotion of Science and Technology for its development by JEOL, and I joined this project as a co-proposer. The one developed in this project is the cryoTEM of Generation 6 (G6) (Fig. 2). JEOL built two of them and delivered to Kyoto University and Osaka University in 2006 [1]. One of our main research subjects is the bacterial flagellar motor, which rotates the supercoiled flagellar filament as a helical propeller to produce thrust for bacterial swimming motility. We need detailed structural information of the flagellar basal body spanning the cell membranes to understand the mechanism of motor rotation, but the important components of the motor, such as the stator units, are dissociated from the basal body during its isolation and purification by detergent solubilization because of their weak binding, and therefore the structure of the flagellar motor cannot be observed in its functional states when isolated from the cell. So what we aimed to do with this new cryoTEM was to establish the method of electron cryotomography (ECT) to observe the in situ structures of biomolecular complexes inside the cells. We tried to establish the ECT method with this new cryoTEM with a tilt mechanism to visualize the functional motor structure in the cell membranes by quickly freezing the entire bacterial cells on a specimen grid and record many of their tilt images to reconstruct 3D images of the cells with the flagellar basal bodies. However, since the diameter of the Salmonella cells are nearly 1 μm thick and the specimen thickness for the electron beam to pass through becomes twice thicker at 60º tilt, which is too thick even for 300 keV electrons to pass through and scatter elastically to form EM images of good quality, we had to wait for several years to visualize the in situ structure of the basal body until we have introduced a genetic engineering technique to produce “mini-cell” to make Salmonella cells much smaller than the wild-type [13].
In order to further improve the efficiency and throughput of image data collection for single-particle image analysis, we decided in 2011 to change the specimen stage to the side entry type of JEOL’s original design with a potential to make automated data collection possible by computer control. This modified version of cryoTEM is G6N (Generation 6 with New Modification) (Fig. 2).
The introduction of this specimen stage together with a different type of objective lens pole piece has resulted in improved usability and image resolution beyond our expectations. One of the standard methods to examine the highest possible image resolution of a cryoTEM is to take an EM image of a test specimen, such as an amorphous platinum-iridium (Pt/Ir) alloy thin film, under a relatively large defocus condition of about 1 μm and look at its Fourier transform to see how far the Thon ring extends (see Fig. 4). This is what we routinely do to examine the resolution of cryoEM images of frozen-hydrated ice-embedded specimens, and such a large defocus is necessary to enhance the image contrast of low resolution to make ice-embedded biomolecules visible for image analysis. By using a thin film of alloy, we are not limited by a low electron dose that we must use to avoid radiation damage to see the weak Thon ring signal at the highest possible resolution. By such a test of image resolution performed on the installation completion date of the cryoTEM G6N, we were able to see the Thon ring extending beyond 2.0 Å resolution. In order to make the Thon ring extend to the highest possible resolution under such defocus conditions, the electron optical system of the cryoTEM must be set to produce a parallel illumination beam on the specimen. It was fortunate that the objective lens pole piece of this cryoTEM was designed to fulfill such requirement under almost any imaging conditions that users set up to collect images even being unconscious of parallel illumination condition.
However, there were some other problems. Although the specimen stage was liquid helium-cooled, the specimen temperature was slightly higher than that of the G4 top-entry type, and the temperature could not be elevated to 50 K as we did with the G4 top-entry type. In addition, the supply of helium was limited around the world and prices were rising. Our microscope facility was connected with a helium recovery pipe to the Low Temperature Center of Osaka University in the same campus, but our helium recovery line frequently caused trouble on the Low Temperature Center operation by contaminating air. So we decided to stop using liquid helium around 2012. We filled liquid nitrogen in both the liquid nitrogen tank and the liquid helium tank of the cryoTEM G6N and found that the specimen temperature can be kept stably at 77 K, exactly at the liquid nitrogen temperature for 7 to 8 hours. By fully utilizing the CCD camera-controlling software provided by TVIPS, nearly automated data acquisition became possible, and this made cryoEM image data collection very efficient.

Fig. 4 Thon ring of the Pt/Ir thin film image recorded with the prototype CRYO ARM™.

A cryoEM image of the Pt/Ir thin film was taken under 1 μm defocus with the prototype CRYO ARM™ operated at 200 kV. The Thon ring in its Fourier transform extended to 1.8 Å resolution.

CMOS-based direct electron detector camera

And finally, the bright time of the cryoEM field had come along in 2013 by the arrival of a CMOS-based direct electron detector camera. David Agard and his colleagues at the University of California, San Francisco, had been developing an EM image detector in collaboration with Gatan, one of the major EM camera manufacturers by repurposing a CMOS device for X-ray image detectors developed by a group at the University of California, Berkeley. Gatan completed the development of this camera system and made it commercially available in 2013 as K2 Summit. The CMOS imaging chip had wonderful specifications, having 4K by 4K pixels, being robust against direct irradiation of high-energy electrons accelerated to 300 keV or even higher, and showing a minimal image blur due to electron scattering within the very thin electron-detecting semiconductor layer, and best of all, the data acquisition rate of 16 million pixel image was 400 frames per second, which made it possible to carry out single electron counting. Because the total electron dose typically used for cryoEM image recording is 20 to 30 e^-/Å over one second exposure, the number of electrons coming onto the detector plane per each frame becomes limited and countable if the dose rate is lowered by about 10 fold by making the exposure time 10 fold longer. Single electron counting gives us a great advantage in reducing the image noise level. One of the major noise sources in cryoEM image recording is the statistical noise, which is large relative to the signal because the number of electrons forming individual cryoEM images is small due to the low-dose imaging to avoid radiation damages. There is no way to escape from it. However, single electron counting can minimize the Landau noise, which is an intrinsic noise of detector caused by a large distribution in the signal amplitude that any type of energy-accumulating image detector, such as CCD, produces for individual electron detection.
Yifan Cheng at the University of California, San Francisco, made the most of the performance of this CMOS camera system and devised a way to collect sharp high-quality cryoEM images of proteins by movie-mode imaging and motion correction. He and his colleagues successfully analyzed the 3D structure of a membrane protein, the TRPV1 receptor ion channel, which senses heat and spiciness, from a small amount of solution sample that eluded crystallization over many years in spite of much effort by a group of his colleagues, and published two papers in Nature at the end of 2013 [14, 15]. The structure was solved at 3.4 Å resolution by analyzing about 100,000 single-particle images of the protein picked up from about 1,000 cryoEM images obtained from a small amount of sample solution containing the detergent-solubilized protein. After individual frame processing for single electron counting, they added up every 80 frames to make a cryoEM movie of 5 frames/sec and then added up these movie frames with motion correction to minimize the image blur caused by a mechanical drift of the specimen stage and the distortion of ice film caused by electron irradiation to make the final cryoEM image very sharp [16].
We were able to introduce Gatan K2 Summit to our cryoTEM G6N (Fig. 2) in 2015 and tried to utilize its high performance as an image detector. The movie mode image recording allowed us to capture the structures of various biomolecules including membrane proteins at near-atomic resolution, and the highest resolution we attained was 2.7 Å for the structure of the thin needle tube of the Shigella type III virulence-factor secretion system (Takashi Fujii & Yurika Yamada, unpublished). Since the attainable resolution depends largely on the structural stability of the sample molecule and the thickness of the ice thin film on the cryoEM grid, it should be possible to achieve resolution exceeding 2.0 Å with a better specimen and grid.

Development of CRYO ARM™

The FEI company (Thermo Fisher Scientific since 2017), one of the major manufacturers of electron microscopes, started developing high-performance user-friendly cryoTEMs for researchers in life sciences around the end of the 1990’s and produced cryoTEMs called Polara in 2004 and Titan Krios in 2008. Titan Krios is in particular a user-friendly system, capable of storing 12 cryoEM grids in a magazine, allowing users to load each grid onto the specimen stage cooled by liquid nitrogen to around 90 K by automatic loading device (autoloader), and allows automated image data collection by specifying areas for image recording after users judged the quality of grid. Titan Krios has been highly evaluated by many users in the cryoEM field over the world because of its high performance in attaining high resolution and user-friendliness both in the 3D observation of cellular structures by ECT and single particle image analysis because both techniques require many cryoEM image data to be collected.
JEOL developed JEM-ARM200F (ARM: Atomic Resolution Microscope) in 2009 as a high resolution TEM for material research and received high evaluation in the world. So we thought JEOL should develop an ARM-based cryoTEM utilizing its ultra-high resolution electron optics by developing necessary devices, such as a highly-stable computer-controlled liquid nitrogen-cooled specimen stage that can maximize the performance of high-resolution electron optics, a cryoEM grid autoloader that is more user-friendly and convenient to use than that of Titan Krios, and a liquid nitrogen automatic filling system, and also by adding the Ω-type in-column energy filter to minimize the inelastically-scattered electrons to increase the S/N of the recorded images. The Ω-filter is also useful for quickly measuring the thickness of ice film for the evaluation and judgement of the quality of the EM grid. Just at the right timing, a large funding named State-of-Art Research Infrastructure Establishment Program was announced by JSPS, and the proposal and application for the funding by Toshio Yanagida as Director of Riken Quantitative Biology Center (QBiC) (also, Specially Appointed Professor and Professor Emeritus of Osaka University; Director of the Center for Information and Neural Networks) was approved in 2010 for the preparation for the establishment of Riken QBiC. This budget was allocated to Riken and Osaka University for their collaborative research activities. We proposed to use part of the budget to develop a user-friendly high-resolution cryoTEM and obtained an approval to ask JEOL to develop it. We conveyed our idea, vision and required specifications to the technical team of JEOL, and they agreed to start the development. JEOL planned to complete a prototype cryoTEM by the spring of 2014, and we had many discussions repeatedly over many meetings. The nickname we came up for this new cryoTEM was CryoARM, and JEOL named it CRYO ARM™ (Fig. 2). One of the goals we set was to exceed 2.0 Å as the attainable resolution in the structural analysis of biomolecular complexes.
Although the development was delayed due to various circumstances and reasons, a prototype cryoTEM was finally installed in May 2016 in the Nanobiology building. According to the initial design plan made at the time when a Cold FEG was not yet available, this prototype CRYO ARM was installed with a Thermal FEG with an acceleration voltage of 200 kV. We examined the Thon ring in the Fourier transform of the Pt/Ir thin film images recorded under 1 μm defocus and confirmed that the Thon ring extended beyond 1.8 Å resolution (Fig. 4). Our important mission started from this point. We prepared cryoEM grids of many different biomolecular specimens and collected cryoEM images to analyze their structures by single particle image analyses to evaluate the performance of the cryoTEM for the resolution. We also carefully examined and evaluated various aspects of the cryoTEM, such as the user-friendliness of cryoTEM operation and controlling software, quickness and smoothness of the manual operation with the cryo-workstation for cryoEM grid transfer as well as autoloader operation, the efficiency of data collection by automated image recording, and the points to be improved in the automated data acquisition software JADAS towards completely-automated data collection, to feed them back to JEOL engineers. As sort of we predicted, so many mechanical troubles frequently occurred with the autoloader and cryo-workstation, and JEOL engineers had to redesign the systems and parts, sometimes with new materials, and bring them from Akishima, Tokyo, to replace with the old ones and make adjustments of sensors and actuators to fix those troubles. It was almost every other week for about half a year from the installation. The specimen stage was largely redesigned for higher stability, and the TEM system controlling software was also improved including the introduction of one-click buttons on display to achieve parallel illumination and coma-free optics alignment. The cryoEM method holds up with so much knowledge accumulated over the years of its development, and a cryoTEM system can work well only by implementing and realizing all those essential know-how and specifications. JEOL engineers worked really hard to make this cryoTEM system work as we desired. It was February 2017 when every function of this prototype CRYO ARM™ finally started working stably, and we never had any serious troubles since then.
We usedβ-galactosidase as a test sample to see the reachable resolution of structural analysis with image data collected with CRYO ARM™. We also continued our evaluation of the operation stability of the hardware and software for further improvement and especially focused on JADAS, an automated data acquisition software program that JEOL has been developing over the years, to make it something that every cryoEM user can use comfortably for efficient data collection. JADAS became practically usable after several updates, and in the summer of 2017 we were able to collect 2,500 images from a cryoEM grid ofβ-galactosidase over 3 days. We picked up about 350,000 single particle images from them, selected about 88,000 good ones by going through 2D and 3D classifications by