Collection, pre-processing and on-the-fly analysis of data for high-resolution, single-particle cryo-electron microscopy

Abstract

The dramatic growth in the use of cryo-electron microscopy (cryo-EM) to generate high-resolution structures of macromolecular complexes has changed the landscape of structural biology. The majority of structures deposited in the Electron Microscopy Data Bank (EMDB) at higher than 4-Å resolution were collected on Titan Krios microscopes. Although the pipeline for single-particle data collection is becoming routine, there is much variation in how sessions are set up. Furthermore, when collection is under way, there are a range of approaches for efficiently moving and pre-processing these data. Here, we present a standard operating procedure for single-particle data collection with Thermo Fisher Scientific EPU software, using the two most common direct electron detectors (the Thermo Fisher Scientific Falcon 3 (F3EC) and the Gatan K2), as well as a strategy for structuring these data to enable efficient pre-processing and on-the-fly monitoring of data collection. This protocol takes 3–6 h to set up a typical automated data collection session.

Introduction

The use of cryo-EM to determine near-atomic-resolution structures of macromolecular complexes has grown dramatically in the past 5 years, led by improvements in microscope hardware, detector technology and image processing methods^1–4. Two companies manufacture high-end cryo transmission electron microscopes (cryo-TEMs) aimed at the life science market (Thermo Fisher Scientific (formerly FEI) and JEOL), and several manufacture direct electron detectors (Thermo Fisher Scientific, Gatan and Direct Electron)^5–7 suitable for visualization of frozen hydrated specimens in this resolution range. A variety of hardware combinations can thus be used to obtain high-resolution data. However, the vast majority of structures deposited in the EMDB at higher than 4-Å resolution have been achieved using either the Gatan K2 or the Thermo Fisher Scientific F3EC direct electron detectors on Titan Krios microscopes⁸.

Increasing numbers of institutions are investing in cryo-EM technology, but the operation of these instruments and the management of the enormous data flows they create present a series of challenges. Many Titan Krios sites are operational 24 h/d, normally ∼80% or more of the time. With current detector technology, this can mean producing ∼5 Tb of data per 24 h per microscope, and this figure may increase dramatically as new detector technologies come online. These datasets present enormous practical challenges, including how and when to move them around, and how they should be organized, but such considerations are essential to improving the efficient use of machine time, as they facilitate on-the-fly pre-processing of data (including motion correction and contrast transfer function (CTF) estimation) in parallel with data collection. Through such approaches, the quality of data can be assessed in near real time, allowing data collections unlikely to yield meaningful results to be halted early or to have their parameters altered. Pre-processing also allows essential steps in cryo-EM structure determination to be executed in parallel with data collection, reducing the time and therefore cost from data collection to structure.

Based on the experiences of the Astbury Biostructure Laboratory (ABSL), here we present our complete protocol to go from an optimized, frozen, cryo-EM grid to a pre-processed single-particle dataset. In our view, this is generally the least time-consuming and best-characterized section of the cryo-EM workflow, as specimen preparation is still a major bottleneck for many cryo-EM projects^9,10, and full processing of the data to get the most out of the dataset can take months of dedicated, expert work. The protocol describes the standardized operating procedures developed and used at the ABSL by all users: internal and external, and academic and industrial. The procedures are robust and tested, and have been used to successfully collect data on a very wide range of macromolecular complexes, including small protein complexes (<150 kDa), membrane proteins (in both detergent and nano-discs), viruses and amyloid fibrils. Using these procedures, we (and others) have determined multiple structures in the 2.5-to 5-Å resolution range (Table 1) and published their structures in peer-reviewed publications^11–14. The protocol is aimed at a broad audience, from non-expert users looking to set up automated data collection using EPU software, to experienced users needing an aide-mémoire, to new facility managers looking for information on implementations of on-the-fly image processing.

Table 1. Detector choices and electron doses for example samples.

Sample (kDa)	Buffer	Detector (pixel size)	Total dose (e–/Å2) (dose per fraction)	Defocus range	Resolution (Å)	Reference
Cowpea mosaic virus (4,080 kDa)	10 mM sodium phosphate, (pH 7.0)	F3EC (integrating) (1.065)	67.5 (1.5)	–0.5 to –3.5	2.7	12
Cytochrome bc1 membrane protein (480 kDa)	25 mM Tris (pH 7.5), 100 mM NaCl, 0.5 mM, 0.015% n-dodecyl-β-D-maltoside (DDM)	K2 (counting) (1.065)	44 (2.2)	–1 to –4	4.4	11
Cytochrome bc1 membrane protein (480 kDa) with inhibitor bound	25 mM Tris (pH 7.5), 100 mM NaCl, 0.5 mM EDTA, 0.015% DDM	F3EC (integrating) (1.065)	66 (1.13)	–1 to –4	3.7	S. P. Muench, University of Leeds, unpublished
Polyketide synthase module (230 kDa)	200 mM HEPES, 200 mM NaCl	F3EC (integrating) (1.065)	111 (1.4)	–0.75 to 3	3.7	N.A.R., data not shown
Escherichia coli ribosome (2.5 mDa)	10 mM HEPES-KOH (pH 7.5), 50 mM KCl, 10 mM NH4Cl, 10 mM Mg(OAc)2, 1 mM dithiothreitol (DTT)	F3EC (integrating) (1.065)	63 (1.4)	–0.8 to 2.9	3.2	N.A.R., data not shown
Ageratum yellow vein virus (3110 kDa)	100 mM sodium phosphate buffer (pH 7.0)	F3EC (integrating) (1.065)	110 (1.4)	–0.5 to –2.5	3.3	27
Amyloid fibrils	25 mM sodium phosphate 25 mM sodium acetate (pH 2.5)	K2 (counting) (1.065)	49.9 (1.2)	–1.25 to –3	3.9	28
Coxsackievirus A24v: ICAM1 complex	Tris-buffered saline	F3EC (integrating) (1.065)	60 (1.5)	–0.5 to –3	3.9	14
BK Polyomavirus + GT1b oligosaccharide	10 mM Tris, 50 mM NaCl, 0.01 mM CaCl2 (pH 7.8)	F3EC (integrating) (1.065)	50 (1.3)	–0.6 to –5	3.4	29
Bacterial nutrient transporter (320 kDa)	10 mM HEPES, 100 mM NaCl, 0.03% (vol/vol) DDM (pH 7.5)	K2 (counting) (1.065)	77.9 (1.6)	–1.2 to 2.6	3.7	N.A.R., data not shown
Prespliceosome	See reference	K2 (counting) (1.065)	63.52 (3.176)	– 0.8 to 4	4	30
Saccharomyces cerevisiae imidazole glycerol phosphate dehydratase	50 mM Tris, 30 mM NaCl (pH 8.0)	F3EC (integrating) (1.065)	50 (1.3)	– 0.6to –5.0	3.2	13
Feline calicivirus	PBS	F3EC (integrating (1.065)	63 (1.26)	–1.2 to 3.5	3	31
50S large ribosome subunit from Staphylococcus aureus (~1.5 MDa)	10 mM Tris-HCl (pH 7.51), 60 mM NH4Cl, 12 mM Mg (OAc)2, 200 mM NaCl	F3EC (integrating) (1.065)	46 (1.3))	–1.1 to –2.9	2.9	N.A.R., data not shown
Protein complex (400 kDa)	20 mM MES (pH 6.5), 500 mM NaCl, 2 mM Tris(2-carboxyethyl) phosphine	F3EC (integrating) (1.065)	83 (1.2)	–1.5 to 3.5	4.1	E. Zeqiraj, University of Leeds, unpublished
Chaperone complex of RuvB-like AAA+ ATPase (390 kDa)	25 mM HEPES, 140 mM NaCl, 10 mM 2-mercaptoethanol	K2 (counting) (1.065)	48 (1.2)	–1.5 to 2.5	3.7	Ó. Llorca, Spanish National Cancer Research Centre, unpublished
Beta galactosidase (440 kDa)	Not disclosed	F3EC (counting) (0.66)	61 (0.8)	–0.7 to 2.3	2.82	J. Reeks and P. A. Williams, Astex Pharmaceuticals, unpublished
Dynein/dynactin/HOOK3 complex (2.6 MDa)	25 mM HEPES (pH 7.2), 150 mM KCl, 5 mM DTT, 1 mM ATP, 0.005% Tween 20	K2 (counting) (1.065)	48 (1.2)	–1.8 to 3	5.7	A. P. Carter, Medical Research Council Laboratory of Molecular Biology, unpublished

Overview of the procedure

The procedure consists of 12 main stages (Fig. 1). First, grids are transferred to the microscope (Steps 1–11). Initial checks are then performed to ensure the correct software is loaded and that the microscope is in the correct mode (Steps 12–15). The grids are then checked to identify the most suitable ones for data collection (Steps 16–17). The beam-setting parameters in the EPU software that are to be used during automated collection are then set (Steps 18–21, Box 1), and the image shift calibrations are performed (Steps 22–25). A low-magnification atlas is then taken of the grid to help identify regions of appropriate ice thickness for data collection (Step 26). The areas for automated collection are then selected (27–30). Direct alignments are checked (Step 31), a gain reference is taken (if required), and the final imaging parameters are set (Steps 32–37). Final checks are then performed to ensure no key steps have been missed (Steps 38 and 39); then the automated collection is started (Steps 40 and 41). The data generated go through organization and pre-processing to enable monitoring of data quality during the session and decrease the time from specimen to structure (Steps 42–51).

Fig. 1. Flowchart of the procedures.

a,b, Flowchart describing the main steps of the procedure for F3EC (a) and that for the energy-filtered K2 Summit (b), with approximate timings. We expect this procedure to take 3–6 h, although timing will be specimen dependent.

The protocol uses Thermo Fisher’s Titan Krios electron microscope and EPU automation software with either F3EC or Gatan K2 detectors to collect data, saves the data to storage systems and pre-processes the data using RELION v.2.1^15,16, MotionCor2¹⁷ and Gctf¹⁸. It also performs statistical analyses to assess and maximize data quality. It is essential that users choose the most appropriate microscope hardware, imaging conditions and data collection schemes to answer the scientific question, as these choices are sample specific. We offer generic advice on this, with information on direct electron detector choice (Box 2) and examples of imaging conditions for a variety of samples (Table 1).

Box 2. Choosing your detector.

The most appropriate detector choice, mode and parameters will be sample specific, and in reality, users’ choices may be limited by the hardware available at their facility. All the direct electron detectors currently on the market, in all modes, are capable of producing sub-3-Å structures of a range of biological specimens. Most of the time, it is the sample, not the detector choice, that limits the final resolution. There are more comprehensive accounts of direct electron detector characterization and performance^5,7; here we offer a quick guide to choosing a detector. This protocol is written for F3EC and K2; therefore, we focus on choosing between these models while mentioning other detectors.

Integrating/counting

During integrating mode, a signal on the detector is summed, whereas in counting mode, each incident electron is individually detected with pixel (counting) or sub-pixel (super resolution) accuracy. For data collection, the F3EC and Direct Electron DE-64 operate in integrating and counting modes, whereas the K2 operates in counting and super-resolution modes. Counting mode detectors have better detective quantum yield (DQE) curves as compared with those of integrating mode detectors, especially at low spatial frequencies. This essentially means that for the same number of electrons applied to the specimen, a counting mode image will have more contrast, meaning it is easier to see and align the particles. However, counting mode collection is typically 1.5–5× slower as compared with integrating mode collection, primarily due to longer exposure times.

Optimal doses for each detector in each mode

In counting mode, the performance of the detector is highly dose dependent, with lower dose rates resulting in better DQE curves. A trade-off is required between dose rate and exposure time because very long exposure times may bring their own problems, such as mechanical movement of the stage. The F3EC is optimally used with 0.5–0.7 e⁻/pixel/s, whereas the K2 is optimally used with 3–10 e⁻/pixel/s in counting mode and 1–3 e⁻/pixel/s in super-resolution mode. To obtain a reasonable total signal in the final image (>35 e⁻/Ų), typically 60-to 90- or 8-to 13-s exposures, respectively, are needed. This results in the collection of ∼25–30 (F3EC) and 45–60 (K2) micrographs per hour for each detector in counting mode.

In integrating mode, dose rates can be much higher, and performance of the detector is linear across a greater range. We generally use the F3EC in integrating mode between 40 and 100 e⁻/pixel/s, with a 1-to 2-s exposure. Depending on other data collection parameters, this results in 70–150 images an hour.

General rules

• Choosing your detector and mode involves a data quantity/quality trade-off.

• For >400-kDa specimens with compact globular structures for which you are expecting better than 3.5-Å resolution, such as ribosomes or icosahedral viruses, we recommend F3EC in integrating mode.

• For 100-to 400-kDa specimens, we recommend (if possible) taking images in different modes on different detectors to compare contrast and make an informed decision. Typically, these smaller specimens benefit from the counting mode’s increase in contrast, but integrating mode combined with high dose can be used (Table 2), especially if your protein preparation is heterogeneous and data processing will benefit from the faster collection, which will in turn result in more micrographs and subsequently more particles.

• If you are tilting the stage during data collection, you will benefit from using an energy-filtered K2.

• Anything <100 kDa probably requires a phase plate (Box 3).

• In your final movie, aim for a dose of 1–2 e⁻/Ų/fraction, for larger specimens. If you are pushing for sub-4-Å resolution, use a number closer to 1 e⁻/Ų/frame.

This protocol is for single-particle data collections and is not restricted to any particular sample type. The protocol for can be readily adapted to any Thermo Fisher Scientific microscope with an autoloader, including the Talos Arctica and Glacios models with EPU installed. The data pre-processing workflow described in this protocol can be adapted for any electron micrograph movies created by any direct electron detector from any microscope. It is also modular, so that different software packages can be readily interchanged. This workflow specifically describes data collection for processing by single-particle analysis, but changes to the protocol would allow it to be adapted for other types of data collection, notably tilt series collection.

Limitations

In this protocol, we describe how we utilize our specific hardware setup at ABSL using the Titan Krios microscope equipped with an F3EC or an integrated K2 with our specific data storage systems and processing hardware. This will need to be adapted for each individual hardware setup at new facilities. This protocol describes a ‘standard’ single-particle data collection. In some cases in which the specimen creates specific challenges, such as a preferred orientation, the method could be easily altered to include collection of tilted data¹⁹.

There are a range of other software packages that can be used for automated data collection for single-particle analysis, notably SerialEM²⁰ and Leginon²¹. These are both attractive options that can offer benefits, including increased speed of data collection and greater flexibility. For data transfer and pre-processing, there are a huge variety of options, many of which could achieve comparable outcomes. Many Titan Krios sites use programs such as SCIPION²² or Focus²³, or their own scripts as wrappers to call external programs for CTF estimation or motion correction. A range of programs for both motion correction and CTF estimation are available, including Unblur and Summovie²⁴, as well as CTFFIND4²⁵.

Experimental design

One of the first decision points is the choice of hardware, particularly the choice of electron detector (Box 2), and whether to use a phase plate (Box 3). This choice may be predetermined based on the user’s access to equipment. For clarity, the protocol described below is for the F3EC. For the K2, the full protocol can be found in Supplementary Methods 1. A flowchart of the individual protocols for each detector, and their estimated timing, is shown in Fig. 1.

Box 3. Use of a Volta potential phase plate.

A Volta potential phase plate (VPP) is a thin, amorphous carbon film positioned at the back focal plane of the objective lens³². The beam in a parallel state is brought to crossover at the VPP, where beam interactions with the carbon surface cause an ‘on-the-fly’ local Volta potential, creating a phase shift between the scattered and unscattered electrons, which increases continuously with accumulated dose on the VPP. This dramatically increases the contrast of the specimen, which can enable the visualization and alignment of smaller particles. When a specimen has an ordered density of ∼<100 kDa, use of a phase plate is usually the only way to create enough contrast to align and classify particles with sufficient accuracy to obtain a high-resolution structure. Useful phase shifts for single-particle work are between 20° and 120°. The protocol presented here can be easily modified to include use of the VPP (Supplementary Methods 3).

In this protocol, we assume the microscope is properly aligned; therefore, microscope column alignment is not discussed. We do assume that any user looking to follow this protocol and set up data collection on high-end TEMs will have had training in the basic operation of TEMs and be familiar with the principles of microscopy. Accessing training in the operation of (high-end) TEMs is a major challenge for cryo-EM as a rapidly expanding field, and the publication and sharing of training resources and protocols is one contribution that can be made to tackling this challenge.

This workflow assumes that grids have been thoroughly pre-screened and identified as suitable for automated data collection, hence we include only limited information on grid screening. Without a sample of sufficient quality (clean, vitreous ice, good particle distribution showing range of orientations, minimal heterogeneity or aggregation of sample), even with the most optimal microscopy, a high-resolution structure cannot be achieved. Many variables can be altered when optimizing sample preparation; these have been reviewed elsewhere^9,10.

Materials

Reagents

• Liquid nitrogen (LN₂) ! CAUTION LN₂ can cause cryogenic burns and asphyxiation in confined spaces.

Equipment

Clipping grids

• Negative-pressure tweezers (Dumont, no. 5)

• Positive-pressure tweezers (Dumont, cat no. SS140)

• Cryo-EM grids of specimen of interest

• C-clip ring (Thermo Fisher Scientific, cat. no. 1036173)

• C clip (Thermo Fisher Scientific, cat. no. 1036171)

• C-clip insertion tool (Thermo Fisher Scientific, cat. no. 1115575)

• Autogrid container (Thermo Fisher Scientific, cat. no.1084591)

• Autogrid assembly station (Thermo Fisher Scientific, cat. no. 1130697)

• NanoCab (Thermo Fisher Scientific, cat. no. 1121822)

• Krios loading station (Thermo Fisher Scientific, cat. no. 1130698)

• Autogrid tweezers (Thermo Fisher Scientific, cat. no. 1121750)

• Cassette tweezers (Thermo Fisher Scientific, cat. no. 1121751)

• Krios cassette (Thermo Fisher Scientific, cat. no. 1121816)

Microscopy

• Thermo Fisher Scientific Titan Krios microscope (Thermo Fisher Scientific)

• TEM software (Thermo Fisher, v.2.9.1)

• TEM Imaging & Analysis (TIA; Thermo Fisher, cat. no. 4.17)

• FluCam Viewer (Thermo Fisher, cat. no. 6.9.1)

• EPU software (Thermo Fisher, v.1.11.0)

• Electron detector: F3EC and buddy camera with live view such as a Ceta 16 M (10480xx, Thermo Fisher) or BioQuantum K2 Summit direct electron detector (Gatan) GMS 3/DigitalMicrograph (DM; v.3.22; Gatan, http://www.gatan.com/products/tem-analysis/gatan-microscopy-suite-software)

Optional

• AutoCTF v.0.6.9 (Thermo Fisher)

• Volta Potential phase plate (Thermo Fisher)

Computing

▲ CRITICAL All computers described in this protocol are connected with a 10-Gb fiber ethernet and operate on a UPS (uninterruptible power supply).

• Falcon 3EC direct electron detector with manufacturer-provided 60-Tb offload server or Gatan K2 with manufacturer-provided PC with RAID and a 60-Tb offload server (Dell, model no. PowerEdge R730)

• GPU server for on-the-fly analysis (Dell, model no. PowerEdge 7910) with 4× QUADRO M4000 8-Gb GPU (NVIDIA)

• GPFS storage node (Dell, model no. PowerVault MD3860i) on storage servers (Dell, model no. PowerEdge R430)

• Scripts for data processing. The scripts described in the manuscript, along with a modified copy of pipeliner.cpp for RELION, are available for download at https://github.com/Leeds-ABSL/ABSL_pipeline. A description of the scripts used is available in Supplementary Note 1.

Procedure

▲ CRITICAL This procedure is designed to act as an aide-mémoire for more experienced users and a more complete guide for the non-expert. All microscope users should expect local rules or procedures to be in place, as well as local variations in the layout of software. Please check with local facility management before using this protocol.

▲ CRITICAL Step 12 and beyond describe a protocol specific to the F3EC workflow. When using the K2 Summit with energy filter, follow the procedures in Supplementary Methods 1.

Transfer of cryo-EM grids to the microscope • Timing 30 min

▲ CRITICAL Steps 1–11 are visualized in Supplementary Video 1.

▲ CRITICAL To reduce contamination on grids, Steps 1–9 should ideally be carried out in a dehumidified environment using clean liquid nitrogen (LN₂) freshly decanted from a pressure vessel or clean onion dewar. All tools should be dry and at room temperature (20–21 °C) before cooling in LN₂. Tools must be warmed back to room temperature and dried in a warming cabinet or heat block between uses. Throughout the process, levels of LN₂ should be kept topped up to appropriate levels to reduce on-grid contamination and reduce the risk of grid devitrification.

1. Retrieve cryo-EM grids from LN₂ storage.

2. Use tweezers to load the C clips into the C-clip insertion tool. With the base of the tool on a flat, clean surface, press down on the C-clip insertion tool so that the C-clip becomes positioned at the rim of the tool.

   ▲ CRITICAL STEP Ensure that anything which will enter into the vacuum of the microscope is not touched with bare hands, as oils will deteriorate the conditions in the vacuum. This includes C clips, C-clip rings and the cassette.

   ▲ CRITICAL STEP Inspect the C clip inside the tool to ensure that it has not become deformed during this process. The C clip should sit flush around the rim of the tool. If any perturbations are seen, for example, an end of the C clip is bent or the clip does not sit fully flush around the rim of the tool, reject this C clip.

3. Cool the autogrid assembly station to LN₂ temperature. Leave for several minutes for the temperature to equilibrate. Following equilibration, the level of the LN₂ should either allow the transfer of the cryo-EM grids (Step 4) to be completed under nitrogen vapor or in a thin layer of LN₂. The station may need to be topped up throughout the procedure to ensure that samples remain vitrified.

4. Transfer the cryo-EM grids to be clipped to the autogrid assembly station.

5. Insert C-clip rings into each of the four positions of the autogrid assembly station, ensuring that they are oriented so that the flat side is positioned against the base of the station. Cool the autogrid tweezers for manipulating grids and C-clip insertion tools (pre-loaded with C clips, as described in Step 2) to liquid nitrogen temperature. Gently transfer the grid to the C-clip ring. Move the station around to the ‘closed’ position and insert the C-clip ring tool over the top, ensuring that it is straight. Press the button on the top of the tool to release the C clip. Remove the C-clip ring tool and turn the station to the open position.

6. Use the autogrid tweezers to flip the autogrid assembly 180° to ensure that the grid is properly and securely clipped. It is recommended that this be repeated twice for each grid. Once clipped, grids are referred to as ‘autogrids’.

   ▲ CRITICAL STEP It is vital that grids be securely clipped. If the autogrid assembly falls apart within the microscope (in either the autoloader or the octagon), it can cause serious problems. If the grids are bent, they may not be securely clipped; therefore, these should be rejected.

   ? TROUBLESHOOTING

7. Repeat Steps 5 and 6 for each grid that is to be clipped.

8. Transfer the autogrids to the autogrid container. In the autogrid container, autogrids should be oriented so that the flat base of the C-clip ring faces away from the notch.

9. Take a clean, room-temperature autoloader cassette and transfer it to the loading station. Cool the loading station and a NanoCab to LN₂ temperature. Once cooled, transfer the autogrid container containing the autogrids to the station and use the autogrid tweezers to position the autogrids in the slots of the cassette so that the flat side of the autogrid is facing the gold band of the cassette. Once loaded, visually inspect to check if the grids are all fully inserted down into the slot. Use the autogrid tweezers to press very gently against the top side of the autogrid and check that it springs back into position, indicating it is seated properly against the springs.

   ▲ CRITICAL STEP If you apply too much force, the grid will be dislodged from the cassette slot; therefore, apply only slight pressure.

   ? TROUBLESHOOTING

10. Dock the pre-cooled NanoCab, full of LN₂, against the loading station. Slide the cartridge arm over the cassette and hold down the button on the arm to grasp the cassette. Slide the cassette into the NanoCab, release the button on the arm and withdraw. Undock the NanoCab. Use a pair of blunt tweezers to press down on the cassette to ensure it is properly seated against the bottom of the NanoCab (the cassette should not move upon pressing firmly down).

11. Take the NanoCab and insert it into the autoloader of the microscope. Click ‘Dock/Undock’ in the user interface (UI) software. When the loading process is complete, remove the NanoCab and close the microscope doors. It is useful to monitor the ‘Autoloader’ menu and autoloader vacuum display to ensure this completes successfully.

    ? TROUBLESHOOTING

Perform initial microscope checks • Timing 10 min

• 12Check that all the required software is open (UI, FluCam Viewer, TIA and EPU).
• 13Ensure you are in TEM mode (‘UI’ > ‘Setup’ > ‘Beam settings’ > ‘TEM mode’; ‘TEM mode’ should be yellow), Autozoom is off (‘UI’ > ‘Setup’ > ‘Beam settings’ > ‘Tab out’ > ‘Autozoom’; ‘Autozoom’ should be gray), the autoloader turbo is auto off (‘UI’ > ‘Autoloader’ > ‘Options’, check ‘Turbo Auto Off’). Check the UI for any error messages, as indicated by red icons with a black cross. If errors are present, consult with a member of the facility staff.
• 14Check that the gun lens, high tension and extraction voltage in the UI are set to the values recommended by your facility management (based on recent alignment of the microscope).
• 15Check in the UI software that the vacuum values are green and all autoloader temperatures are colder than −170 °C (this may take >10 min after loading specimens).

Identify the grid for data collection • Timing 0.5–4 h

• 16Start grid inventory (‘UI’ > ‘Autoloader’ >‘Tab out’ > ‘Inventory’) and check whether the number and positions of the grids in the microscope correspond to those loaded. Once the inventory has finished, select the desired grid in the autoloader and select ‘load’. When complete, ‘cartridge successfully loaded’ will be displayed in the UI, the autoloader menu will not display anything in the log area, and the grid will be shown as located in the column if you hover the mouse over that slot in the ‘Autoloader’ menu. The slot position in the autoloader will be colored yellow.

  ? TROUBLESHOOTING

• 17Confirm that the grid is suitable for automated data collection, i.e., its appearance is consistent with your screening images. If the grid has been previously loaded into a microscope, assess the quality of the ice to ensure that there is no increase in surface ice contamination or any evidence of devitrification.

  ▲ CRITICAL STEP Here, we assume that the grids loaded have been prescreened and identified as suitable for data collection. The ‘ideal’ ice conditions vary dramatically for different samples; for more guidance, see refs ^9,10.

  ? TROUBLESHOOTING

Determine the beam-setting presets • Timing 30 min

• 18Navigate to an intact square that you are unlikely to use for data collection (a dry square works best, and areas of thick ice should be avoided). Set the eucentric height manually at low magnification to bring the specimen roughly to eucentric height. In EPU, navigate to ‘Preparation’ > ‘Acquisition Settings and Optics’ > ‘Presets’ and push the ‘Grid Square’ settings to the microscope by clicking ‘Set’ and then view the specimen on the FluCam Viewer. Use the joystick to center the image (5-mm circle on the FluCam Viewer) on a visible feature such as ice contamination. Tilt the stage to 20° by navigating to ‘UI’ > ‘Stage’ > ‘Set Alpha’ (this becomes yellow when activated) and use the z-axis buttons on the right-hand panel to move the feature back to the center of the screen. Click ‘Set Alpha’ again to return the stage to 0° tilt. If this has been done correctly, the visible feature should still be centered. If not, repeat the process.
• 19In the EPU software, each beam-setting preset should be checked to ensure that the variables are optimal for collection, as these will vary from experiment to experiment (Box 1, Supplementary Tables 1 and 2 for F3EC; Supplementary Table 3 for K2). These are set in the ‘Presets’ dropdown menu (‘EPU’ > ‘Preparation’ > ‘Acquisition and Optics Settings’ > ‘Presets’). For each setting, on the FluCam Viewer, ensure that the beam fully illuminates the entire detector, that the beam is parallel (this can be checked in ‘UI’ > ‘Beam Settings’ > ‘Parallel’, ‘Spreading’ or ‘Condensing’) and that you are using the dimmest beam needed to complete the Atlas, Grid Square and Hole/Eucentric tasks. This exposes your specimen to as small an electron dose as possible before the data-acquisition exposure. Small adjustments to variables such as the illuminated area can be made later in the session, but changes to spot size can mean repeating steps and can cause delays in data acquisition.
• 20
  Check that the electron dose is appropriate for your detector mode choice and data collection goals (see Box 2 for information on detector dose and Table 1 for example parameters). To set the detector mode, navigate to ‘EPU’ > ‘Preparation’ > ‘Acquisition and Optics Settings’ > ‘Presets’ > ‘Data acquisition’, and select either ‘Counted’ or ‘Linear’. To calculate electron dose, move to an area where there is no specimen, i.e., a broken grid square. Push the data-acquisition beam settings to the microscope by clicking ‘Set’ in EPU. Press ‘Eucentric focus’ on the hand panel. With the FluCam Viewer, check that the beam is centered over the detector; if not, use the beam shift direct alignment to correct (Supplementary Methods 2). The condenser apertures should be set in the way you intend to use them during data acquisition. In the UI, go to the ‘Camera’ menu and check that ‘F3EC’ is selected and inserted (yellow). Then navigate to the ‘Bias/Gain’ tab and select ‘Reference Image Manager’. Make sure the beam is unblanked.

  • For an F3EC integrating mode data collection session, select ‘normal’ from the ‘Available Reference Images’ in the Falcon ‘Reference Image Manager’ and press ‘Measure Dose’. We typically use 40–100 electrons (e⁻)/pixel/s.
  • For an F3EC counting mode data collection session, select one of the two electron-counting reference images from the ‘Available Reference Images’ in the Falcon ‘Reference Image Manager’ and press ‘Measure Dose’. A setting of 0.5–0.7 e⁻/pixel/s is recommended.
    This will give a reading in electrons/physical pixel/second.
    ? TROUBLESHOOTING
• 21To work out the dose per Ų, first calculate your Ų value; i.e., at 75-k magnification, each physical pixel represents 1.065 Å (calibrated for ABSL Krios 1, F3EC combination), and 1.065 ×1.065 = 1.13 Ų. Now divide your dose per physical pixel per second by the Ų to get e⁻/Ų/s; e.g., if you have 50 e⁻/pixel/s, 50/1.13 = 44.3 e⁻/Ų/s. To get the total dose, multiply this figure by the length of exposure in seconds, for example for a 2-s exposure, the total accumulated dose per exposure would be 44.3 × 2 = 88.6 e⁻/Ų. If you need to, alter the data-acquisition parameters to achieve the desired dose. Be mindful to keep the illuminated area as small as possible to (i) allow multiple exposures per hole, if appropriate, and (ii) reduce the likelihood of the beam pre-exposing neighboring acquisition areas. However, the beam should be large enough to prevent any beam fringing from appearing at the edges of the image. Beam spot size, illuminated area and, if needed, condenser 2 aperture can be changed to fine-tune the desired dose.

Box 1. Beam-setting presets.

Beam-setting presets in EPU are a convenient way of setting beam-setting parameters to perform specific tasks and switching between the magnifications used during automated data collection. They are a set of parameters relating to the beam, including spot size and illuminated area. You can push these settings to the microscope by selecting it in ‘EPU’ > ‘Preparation’ > ‘Acquisition and Optics Settings’ > ‘Presets’ (select option from dropdown menu) > ‘Set’. These will need to be altered depending on the properties of the grid you are using. Here is a description for each:

• Atlas. This is usually optimized to be as fast as possible and does not typically need to be varied according to different grid types.

• Grid square. This is adjusted according to mesh size. It should show one entire grid square within the field of view with no neighboring squares.

• Hole/Eucentric. When grids with a regular array of holes are being used, this magnification should contain a single, complete hole and, at most, small fractions of neighboring holes. Choose a magnification with which it is not possible to take an image just of carbon (i.e., too-high magnification), as this may lead to the software skipping holes. For lacey carbon grids, this parameter is not required for automated data collection. When possible, the spot size should be kept the same as that used for data acquisition, to minimize changes to lens settings during acquisition. As the specimen will be pre-exposed to the electron beam during hole imaging, the total dose should be kept to a minimum by reducing exposure time and limiting the illuminated area, to prevent neighboring holes being inadvertently exposed to the beam.

• Data acquisition. These settings depend on your desired settings for data acquisition, but usually a magnification resulting in 1–1.35 Å/pixel is chosen for most single-particle projects (although some choose a smaller physical pixel size), with illuminated area/spot size chosen to deliver the desired amount of dose to the detector.

• Autofocus and drift measurement. When possible, focus and drift measurement presets should be kept the same as those for data acquisition to reduce the number of changes to lens power. In F3EC counting mode, focus/drift measurement calculations can be carried out in integrating mode for speed.

Example beam-setting presets using different detectors and modes can be found in Supplementary Tables 1–3.

Image shift calibrations • Timing 10 min

• 22Navigate to an intact grid square. Center on a feature of interest that will be visible at the ‘Atlas’ and ‘Data Acquisition’ magnifications, for example, a piece of ice contamination. Roughly set eucentric height as described in Step 18. Run the auto function ‘set eucentric height by stage tilt’.

  ? TROUBLESHOOTING.

• 23In EPU, set the data-acquisition beam settings for the microscope. On the FluCam Viewer, check that the feature is visible in the center of the screen. Ideally, center a distinctive feature, such as the point of an ice crystal. Select ‘Eucentric focus’ on the right-hand panel. In EPU, navigate to ‘Preparation’ > ‘Calibrate Image Shifts’ and click ‘Start Calibration’.
• 24In EPU, an image at data-acquisition magnification will appear with a marker in the center. If the marker is placed on your recognizable feature, click ‘proceed’. If you want to move it, double-click in the image and click ‘re-acquire’ until the feature is in the center of the image.
• 25The microscope will now change to the next-highest magnification in the beam presets and take an image. In the second image, double-click so that the same feature is centered identically to the first image and click ‘re-acquire’. Repeat until the feature is identically centered, and then click ‘proceed’. Repeat this step until it says ‘image shift calibration finished successfully’ in the EPU log on the right.

Atlas the grid • Timing 15 min

• 26Ensure the objective aperture is not inserted (‘Apertures’ > ‘Objective’ > ‘None’). In the EPU, navigate to ‘Atlas’ > ‘Session Setup’, click on ‘Create new sample’ and name your session. We recommend a format such as ‘Username_myprotein_date_ATLAS’. Save images in MRC format and store the atlas directly on the F3EC offload server (usually designated ‘Z:’ in a standard Krios installation). Click on ‘Acquire’ to acquire the atlas.

Select the square for data acquisition and define the template • Timing 1–2 h

• 27In EPU, navigate to ‘EPU’ > ‘Session Setup’ and click on ‘New Session’. We suggest a session name such as ‘Username_myprotein_date_EPU’. Select ‘manual selection’, save images in MRC format and save the data directly to the F3EC offload server (usually ‘Z:’; a directory with the session name will be created and all data will write there. If you choose a subdirectory, the data will be split between different directories). Choose the type of grid, i.e., ‘Quantifoil’ and the size of the holes or Lacey grids from the drop-down menu, then click ‘Apply’.
• 28Go to ‘EPU’ > ‘EPU’ tab > ‘Square Selection’; if all squares are green, click ‘Unselect all’ on the top left. Open tiles (hover over atlas image, right-click and choose ‘open tile’) and look to see which areas of the grid have appropriate ice thickness for data acquisition. For most specimens, you can tell at this low magnification if an area is dry (no ice) or the ice is too thick (Fig. 2a). When you have identified a suitable square (Fig. 2b), select it by hovering over the square, followed by a right-click; then choose ‘add’; then right-click and choose ‘move stage to grid square’.
• 29Bring the grid square to eucentric height by following Step 18 for manual adjustment, and then running the auto function ‘set eucentric height by stage tilt’ as described in Step 22. Select ‘Eucentric focus’. Go to ‘EPU’ > ‘EPU’ > ‘Hole Selection’ and click ‘acquire’. This will save your x, y and z coordinates.
• 30
  Set the template for automated acquisition, including areas and variables for autofocus, drift measurement and image acquisition. This process varies depending on the type of grid hole, which can be either regular, such as with Quantifoil and C-flat (option A), or irregular, such as with lacey carbon (option B).

  1. For grids with a regular array of holes, such as Quantifoil and C-flat

     1. Go to ‘EPU’ > ‘EPU’ > ‘Hole Selection’ and click ‘acquire’. The first time you do this, click ‘measure hole size’. Move and adjust the yellow circles so that they match the hole size, then click ‘find holes’. Repeat this until the software finds the hole sizes correctly (use the zoom function to see more accurately). This needs to be done only once per session.
     2. The ‘filter ice quality’ histogram on the right can be adjusted by moving the red histogram lines in order to adjust and refine hole selection. Use the ‘Select’ tools menu at the top to fine-tune hole selection. For example, remove holes that are empty or in which the ice is too thick/thin/contaminated and holes that are close to the grid bar (Fig. 2c).
        ? TROUBLESHOOTING
     3. Go to ‘EPU’ > ‘EPU’ > ‘Template Definition’. Click ‘Acquire’ and ‘Find and Center Hole’.
        ? TROUBLESHOOTING
     4. Change the ‘Delay after Stage Shift’ and the ‘Delay after Image Shift’ times to between 1 and 5 s, depending on the stability of the stage. 1 s is generally sufficient, but this time can be increased if unacceptable drift is observed (as measured during motion correction in Steps 42–51).
     5. Click ‘Add acquisition area’, then click anywhere on the template image. The outer circle represents the illuminated area; the inner square represents the exposure area (Fig. 2e,f). On the top right, add your defocus range for the acquisition (add the defocus you would like, making sure it is a negative number if you wish to work in standard defocused mode imaging); see Table 1 for examples.
     6. Move the acquisition area to the desired location. Depending on the sample, hole size and user preference, you can do one exposure in the middle of the hole, ideally with the illuminated area covering the whole hole (this may help reduce the effect of charging), or add multiple exposures around the hole, taking care not to overlap illuminated areas into neighboring exposure areas (Fig. 2e,f).
     7. If you want to add additional exposure areas, click ‘Add Acquisition Area’, click the template image and move the exposure to the desired location (double-checking that they have retained the defocus list). When choosing the number of acquisitions to take around a hole, bear in mind that the beam diameter shown in EPU can vary by ±∼10%, depending on the accuracy of alignment, so it is safer to leave some space between the acquisition areas or check that the physical beam and the virtual beam in EPU coincide, by burning a hole in the carbon to confirm the true size of the illuminated area.
     8. Click ‘Add Autofocus Area’ and click anywhere on the image. Move the autofocus area to the carbon surrounding your hole. Standard practice is to autofocus every 5–15 µm, depending how large the variation in height is across the grid square (with more uneven grids, we recommend autofocusing more often). Focus using the objective lens and ensure ‘autostigmate’ is set to ‘no’.
     9. Click ‘Add Drift Measurement Area’. Perform a drift measurement once per grid square, and set the threshold to 0.05 nm/s. If your microscope has known stage stability issues, you may have to relax this threshold and/or perform drift measurement more often. The drift measurement area should overlap directly with the autofocus area. Make sure neither drift nor autofocus area overlap with an acquisition area either in this hole or neighboring holes (Fig. 2e,f).
     10. Check the template layout by running the ‘template execution’ function. This is a good idea in order to see if you need to move your acquisition areas (e.g., too much/not enough carbon in images) or would like to assess particle distribution.
  2. For irregular arrays of holes, such as lacey carbon

     1. In ‘EPU’ > ‘EPU’ > ‘Area Selection’, choose the spacing between acquisitions. It is recommended that you add ∼300–400 nm to the data-acquisition illuminated area (e.g., for an illuminated area of 0.9 µm, choose spacing of 1.3 µm), to ensure the illuminated areas do not overlap. Click ‘View Pattern’. Use area selection tools to add or remove acquisition areas (Fig. 2d).
     2. Change the stage shift delay to 5 s. A shorter or longer time can be used, depending on the stability of the stage. As collection on irregular carbon typically means a greater number of stage shifts, a longer stage shift delay is recommended as compared with that used for regular arrays.
     3. Set the autofocus recurrence to ‘after distance’. Depending on the height variation of your grid, every 8–15 µm is recommended. Focus using the objective lens, and ensure that ‘autostigmate’ is set to ‘no’.
     4. In ‘Data Acquisition Area Settings’, add your defocus range for the acquisition (add the defocus you would like, making sure it is a negative number if you wish to work in standard defocused mode imaging). You have to do this for only one grid square; it remembers for subsequent squares.
     5. Use the bottom histogram (filter ice quality) on the right and the exposure area selection tools to optimize hole selection in order to exclude areas of suboptimal/no ice.

Fig. 2. EPU setup.

a, Typical Atlas view with thick (orange), appropriate thickness (blue) and dry (white) areas indicated (which will vary by sample). b, Square selection on an Atlas. Each square should be inspected to ensure that it is not broken (data collected on broken squares may have more motion, affecting data quality) and that the ice thickness is appropriate for the specimen. Grid squares that have been collected are in blue; orange indicates collection in progress, and green indicates areas to be collected. c, Grids with regular hole (green circles) selection. Holes close to the grid square bars (which typically are poorly vitrified) are deselected. d, Lacy carbon with thin continuous film acquisition area selection. Outer green circles represent beam diameter; inner squares represent exposure area. Note that large contaminants, areas at the edge of the square and areas where the carbon support/hole ratio is poor are deselected. e, Template with single shot per hole and whole hole illumination. Scale bar, 1 μm. f, Template with multiple shots per hole. In e and f, autofocus and drift measurement areas (purple) are overlaid; green circles represent the illuminated area and inner squares the exposure area. Scale bars, 50 μm (a,b); 35 μm (c); 10 μm (d); 1 μm (e,f).

Check direct alignments • Timing 30 min

• 31Perform direct alignments on the grid (Supplementary Methods 2).

  ! CAUTION In many electron microscopy facilities, direct alignments are carried out only by facility staff. We provide a guide to performing the basic direct alignments in Supplementary Methods 2, but you should check local procedures in your EM facility before attempting these.

  ▲ CRITICAL STEP Some direct alignments cannot be adequately performed on UltrAuFoil grids²⁶, and so if using these, perform direct alignments on a carbon grid or cross-grating before starting data collection.

Perform gain reference and set final imaging settings • Timing 10–120 min

• 32The gain reference should be of sufficient quality that when an image is taken using the parameters chosen for data acquisition, with no specimen obstructing the beam, a completely featureless image is obtained. The procedure for collecting a new gain reference in integrating mode is very quick to perform, so this can be done for each data collection run. In counting mode, the procedure takes 1.5 h, so collection of a new gain reference is not recommended unless required. To collect a new gain reference, make sure the FluScreen is lifted and that you are over a hole with absolutely no obstruction in the field of view. Ensure that the beam is not blanked.
• 33Go to ‘EPU’ > ‘Preparation’ > ‘Acquisition and Optics Settings’ > ‘Presets’ > ‘Data acquisition’ and click ‘Set’.
• 34Check that the beam is centered and that there is no beam fringing visible in the image. If necessary, adjust using the beam shift direct alignment (Supplementary Methods 2).

  ▲ CRITICAL STEP If beam fringes are present in the gain reference, they will be seen in every image of the data collection, even if the illuminated area is subsequently expanded. Take great care to ensure that the beam diameter is sufficiently large so no beam fringing is seen.

• 35
  In the UI, choose ‘Camera’ (check that F3EC is selected and inserted) > ‘Bias/Gain’ tab > ‘Reference Image Manager’. When using F3EC in integrating mode, follow option A; when using F3EC in counting mode, follow option B.

  1. F3EC in integrating mode • Timing 10 min

     1. In ‘Falcon Reference Image Manager’, select ‘normal’ from the ‘Available Reference Images’, check that the exposure time is 10 s and ‘images to average’ is 1.
     2. Click ‘Measure Dose’. You might have to click ‘Measure Dose’ twice to get a reliable dose reading. Check that this is consistent with the earlier value checked at Step 20.
     3. Select ‘normal’ gain reference and click ‘Acquire selected gain reference’.
     4. Once complete, acquire a test image in the UI: choose ‘Camera’ > ‘Acquire’ (use 2-s exposure). Inspect the image for signs of beam fringing at the edges. Check that the fast Fourier transform has no features. If the inspection reveals something wrong (e.g., beam clipping), rectify the problem and retake the gain reference until the flat field image is completely featureless.
  2. F3EC in counting mode • Timing 1.5 h

     1. In ‘Falcon Reference Image Manager’, select ‘pre-EC’ from the ‘Available Reference Images’ and click ‘measure dose’. The reading should be 0.5–0.7 e⁻/pixel/s. Change the spot size and/or beam intensity in order to correct the dose.
     2. Take a counting mode image in TIA to ensure that there is no beam clipping in the image (select ‘UI’ > ‘Camera’, then select the ‘counting’ checkbox and 60-s exposure, then select ‘Acquire’); if there is beam clipping, then expand the illuminated area. If the image appears as a flat image with no features, there is no need to take a gain reference.
     3. If a new gain reference is required, in ‘Falcon Reference Image Manager’, click ‘pre-EC’ from the ‘Available Reference Images’, check that the exposure time is 60 s and ‘images to average’ is 45; then click ‘Acquire selected gain reference’.
     4. Once complete, select ‘post-EC’ and ‘Acquire selected gain reference’.
     5. Once complete, acquire a test image (‘UI’ > ‘Camera’ (ensure that the ‘counting’ box is checked) > ‘Acquire with 60 s exposure’). Inspect to ensure that you have a flat field image. If the inspection reveals something wrong (e.g., beam clipping), rectify the problem and retake the gain reference until the flat field image is completely featureless.
• 36With no obstruction in the field of view, take a dose measurement and use this to calculate your final electron dose parameters as in Step 21.
• 37In ‘EPU’ > ‘Preparation’ > ‘Acquisition and Optics Settings’ > ‘Presets’ > ‘Data Acquisition’, check that you are collecting fractions. In ‘EPU’ > ‘Preparation’ > ‘Direct Detector Dose Fractions’, set the number of fractions you would like to split your exposure into and click ‘Equal Dosage’ and ‘Validate’ to ensure that the values are compatible with the software. We recommend a number of fractions that results in between 1 and 2 e⁻/A²/fraction, as this seems to represent a good trade-off between the signal in each frame necessary to perform motion correction and sufficient frequency to correct for beam-induced movement.

Final checks • Timing 10 min

• 38Before automated acquisition begins, perform a final check of variables that can affect data quality or prevent common mistakes. In ‘UI’ > ‘Autoloader’ > ‘Turbo’ > ‘Options’, click ‘Turbo auto off (Default)’. Ensure that the Titan Krios enclosure is shut.
• 39Check the microscopy parameters. Ensure that the aperture series is as desired, the beam is centered in the data-acquisition beam settings and no beam fringes appear in the image. Check that you are collecting fractions, if desired. Ensure that the disk you are writing to has sufficient space available for the entire planned data collection.

Start automated collection • Timing 5 min; data typically collect for 24–72 h

• 40In ‘EPU’ > ‘EPU’ > ‘Automated Acquisition’, click ‘Start’ to begin the EPU run. We suggest you now move on to start data organization and on-the-fly image processing steps, then come back to select more grid squares as described in Steps 28–30, but note that the template needs to be set only once for each data collection.

  ? TROUBLESHOOTING

• 41Fill out a data-acquisition report (Supplementary Table 4).

Data transfer, organization and on-the fly processing • Timing 20 min; it carries on for the duration of data collection

▲ CRITICAL Data must be moved from the microscope’s limited offload server (at ABSL, ∼60 Tb) to a storage system (at ABSL, ∼5 Pb). To do this, we use a custom copying script (ABSL_OTF.sh; Supplementary Note 1). The following steps (42 and 43) describe the usage of this script, which was designed for the computational setup at ABSL but can be readily adapted to other hardware setups.

• 42Open a terminal window and navigate to a directory in which to process the data. Create a directory for processing, i.e., ‘myprotein_date’ (i.e., run the command: mkdir myprotein_date).
• 43Run the script ‘ABSL_OTF.sh’ with the appropriate arguments. Run the following command:

  sh ABSL_OTF.sh <runtime in mins> <which microscope (krios1 or gatan)>

  <what you want directory to be called (date is automatically added)>

  <name of EPU directory>

  For example, sh ABSL_OTF.sh 2880 krios1 myprotein Username_myprotein_-date_EPU

  Allow the script to run; data will be transferred over in blocks every 30 s.

  ▲ CRITICAL The network implementation at ABSL is based o an 10-Gb ethernet, allowing sustained transfer speeds of ∼1 Tb/h. Slower network speeds will create a backlog of untransferred/ unprocessed data that might take several hours/day to be cleared after the data collection finishes.

• 44Begin on-the-fly processing of the data. In our workflow (Fig. 3), a slightly modified version of RELION v.2.1¹⁶ (Supplementary Note 2) is used for on-the-fly processing. ABSL_OTF.sh copies data to a storage location and then a Raw_data directory in a separate, designated processing directory, and then it creates symbolic links to the raw images (unaligned fraction images in .mrc stack format, written by EPU). This serves to protect the original data so that users cannot accidently delete or modify the original files.
• 45Open RELION. Set up an import job. Select ‘Raw_data/*.mrc’ as the input files. Schedule, but do not run, this job.
• 46Set up a motion correction/dose weighting job in RELION. Select ‘Import/job001/movies.star’ as the input files. Set the other motion-correction parameters, including dose weighting, as desired. Schedule, but do not run, the job.
• 47Set up a CTF determination job in RELION. Select ‘MotionCor/job002/corrected_micrographs.star’ as the input file. Schedule, but do not run, the job.
• 48Use the ‘Autorun’ function to run the scheduled jobs for the duration of the data collection run by navigating to ‘Autorun’ > ‘Run scheduled jobs’. Set the ‘Run the jobs how many times?’ value and ‘Wait at least in between (in minutes)?’ parameters. Begin running scheduled jobs with the ‘execute’ button.
• 49Assess the quality of the data. ABSL_OTF.sh will prepare a diagnostic image named ‘micrograph_analysis_0.png’ every time a cycle of CTF estimation is completed (Fig. 4).
• 50Finish the on-the-fly processing. After the data collection has finished, allow ABSL_OTF.sh to run until all data have been transferred. Once all files have been motion-corrected and CTF-estimated, stop the scheduled jobs in RELION by navigating to ‘Autorun’ > ‘Stop running scheduled jobs’. Allow any active RELION jobs to finish running.
• 51The data are now ready for downstream processing steps. As part of this workflow, users can also use RELION’s automated particle picking and 2D classifications as described in ref. ¹⁶.

Fig. 3. On-the-fly data processing pipeline (Steps 42–51).

Data are copied from their write-on F3EC and K2 locations to a storage location. Symbolic links are then made to the processing directory, where RELION batch jobs are used to motion-correct and perform CTF estimation. The outputs from this are plotted by micrograph analysis for the user to inspect.

Fig. 4. Example output of the micrograph analysis script.

a, A scatter plot of the two orthogonal defocus measurements provides a quick visual assessment of the range of defocus values in the dataset. b–d, Histograms describe the overall dataset estimated resolution (b), astigmatism (c) and phase shift (d). e,f, Plots show estimated resolution (e) and astigmatism (f) values for each micrograph in order as they were acquired, expressed as a percentage of the mean values for the entire dataset. Large changes in these values over time suggest that a problem may have occurred during the data-acquisition run. g, Plot of phase shift for each microscope in order of acquisition allows the tracking of the change in phase shift as the plate becomes charged and the microscope moves to new phase plate positions. The non-phase-shift version of the script produces identical output, minus d and g.

Troubleshooting

Troubleshooting advice can be found in Table 2.

Table 2. Troubleshooting table.

Step	Problem	Possible reason	Solution
6	Autogrid assembly comes apart upon checking	Grid is bent, putting pressure on the assembly Manufacturing faults in C clips	Make sure only visibly flat grids are clipped; do not attempt to clip grids that have been bent or folded over When there are repeated problems with a single batch of C clips, there may be an issue with the manufacturing of the C clips. We recommend keeping track of what batches of C clips and C-clip rings are being used, and contacting the manufacturer if this is suspected
9	Autogrid assembly comes out of position in cassette	Autogrid was not loaded into correct position in the cassette	Make sure that when the autogrid is picked up with the autogrid tweezers, they fully cover the grid; this helps to ensure that when loading it into the cassette it is sufficiently inserted. When in the cassette, the autogrid tweezers can be used to gently grip and push the autogrid directly downward to ensure it is fully in position before pushing to ensure it is correctly seated against the springs of the cassette
11	NanoCab does not dock properly to microscope	Cassette is not properly seated in NanoCab	Use the back of a pair of tweezers to press firmly against the cassette to ensure that it is fully seated against the base of the NanoCab. The pin on the top of the NanoCab should be protruding
16	Grid inventory does not match up with samples that were loaded, or when grid is loaded, no grid can be seen	Autogrid assembly has come apart	The safest option is to unload all grids from the microscope octagon (if loaded) and then autoload and assess if the autogrid assembly has come apart. Try to account for all components (grid, C clip, C-clip ring), as these can cause problems in the octagon or autoloader of the microscope, for example by blocking valves
17	Grids appear to have surface ice contamination	Crystalline ice in LN2	Use LN2 freshly decanted from a pressure vessel into a clean, dry dewar in Steps 1-11
	Grids appear to have surface ice contamination	Crystalline ice formed in LN2 during clipping procedure	Where possible, perform Steps 1-11 in a dehumidified environment, ideally <20% relative humidity. Cool the clipping station immediately before use and perform the clipping as quickly as possible to minimize the time for water in the air to condense on the cold LN2
22, Supplementary Methods 1, step 17	Autofunction ’eucentric height by stage tilt’ fails	The specimen Z position is too far from the eucentric height	Set ’grid square’ magnification and manually set the eucentric height, using the ’stage tilt’ and ’z axis’ buttons on the right-hand panel
	The autofunction ’eucentric height by stage tilt’ fails	Not enough signal in the image	Ensure there is carbon in the image. Set the image binning to 2 in the ‘hole/eucentric’ preset. Increase the brightness of the beam by decreasing the illuminated area and/or using a brighter (lower number) spot size
20	Measured dose in the Reference Image Manager gives no value or a very low value	Software bug or dose was measured while screen was retracting or while beam was blanked/F3EC retracted	Ensure the FluScreen is retracted, the beam is not blanked and the F3EC is inserted. Press the ‘measure dose’ button again
30A(ii), Supplementary Methods 1, step 25A(ii)	Hole selection tools are grayed out	Incorrect session settings	In EPU, under ‘session set up’, ensure ‘manual’ is selected
30A(iii), Supplementary Methods 1, step 25A(iii)	Software unable to find hole	Signal is too low to reliably find the hole	Set image binning in ‘hole/eucentric’ preset to 2. Increase the signal in the image by increasing the ‘hole/eucentric’ preset exposure time, or brightness of the beam, by decreasing the illuminated area, using a brighter (lower number) spot size (although it is recommended that the spot size be kept the same as for data acquisition where possible)
	Hole appears to be a different size than the yellow circle/software unable to find hole	Hole size is incorrect	Go back to Step 30(i) and ensure that the hole size is correctly measured. Sometimes ice halo effects can produce a misleading image on the lower-magnification square image, so use of the zoom tool is recommended. Ideally, this should be done on a thin ice/dry area for the most accurate results
40, Supplementary Methods 1, step 29	Motion-correction analysis shows that images have unacceptable motion/drift	Source of vibration is present	If the enclosure of the Titan Krios is open, more drift than usual may be seen. Ensure all doors are properly closed. Another common source is the autoloader turbo; ensure this is off (Step 38)
	Images being collected are seen to have motion, and motion-correction analysis shows unacceptable drift	Mechanical drift because of grid or stage	If the support film of the grid is cracked, more drift may be seen in the images. Try to pick grid squares that have no cracks or broken areas. Some stages are less stable than others. If the stage is the problem, try increasing the stage settling time (Step 30A(iv), 30B (ii)) or perform drift measurement more often (Step 30A(ix)).
	Images appear to have charging (often seen as localized areas of ‘drift’ or where image appears blurred)	Reasons for charging are often unclear	One strategy that can be tried is to use whole-hole illumination. Set up beam settings and template (Fig. 2, Step 30A) so that at data-acquisition magnification, the beam illuminates the whole hole. This appears to reduce charging in some cases
	Images being collected are seen to have (by CTF	Changes in z height across the grid	Autofocus more frequently (Step 30A(viii), 30B(iii))
	analysis) defocus outside the desired range	Autofocus procedure resulting in slightly different defocus as compared with calculated results	Change the autofocus range (Step 30A(viii), 30B(iii))
Supplementary Methods 1, step 7	Unable to find zero-loss peak	Not enough dose on K2	Check if K2 is inserted and the beam is centered over the GIF. With the slit out, in Digital Micrograph, take an image in linear mode and ensure you have thousands of counts on the detector. Use a larger C2 aperture and brighter spot, if needed

Timing

The time taken to set up automated data collection is variable, depending on the hardware used, the length of the automated data collection session and the experience of the user. Although this protocol may take as little as 1 h for a short collection set up by an experienced user, typically 3–6 h would be standard (Fig. 1). For some samples, especially when there is substantial variation in particle distribution across a single grid, much more time may be needed for picking acquisition areas.

Steps 1–11, transfer of cryo-EM grids to the microscope: 30 min

Steps 12–15, perform initial microscope checks: 10 min

Steps 16 and 17, identify the grid for data collection: 0.5–4 h

Steps 18–21, determine the beam-setting presets: 30 min

Steps 22–25, image shift calibrations: 10 min

Step 26, atlas the grid: 15 min

Steps 27–30, select the square for data acquisition and define the template: 1–2 h

Step 31, check direct alignments: 30 min

Steps 32–37, perform gain reference and set final imaging settings: 10–120 min

Steps 38 and 39, perform final checks: 10 min

Steps 40 and 41, start automated collection: 5 min; data typically collect for 24–72 h

Steps 42–51, data transfer, organization and on-the-fly processing: 20 min; it carries on for the duration of data collection

Anticipated results

By following this protocol, the user should be able to produce high-quality electron micrographs for single-particle data analysis and pre-process the results in close to real time. Monitoring the micrograph analysis output permits visualization of the estimated defocus, resolution in the micrographs, astigmatism and, where relevant, the phase shift. An example micrograph analysis is shown in Fig. 4. The user can then make informed decisions about alterations to defocus range and objective stigmation. For phase plate data, useful phase shifts are between 20° and 120°, therefore, on-the-fly analysis of the phase shift allows a user to ensure that data are collected in the optimal range by altering the frequency with which the phase plate position is changed.

The micrograph analysis script uses the resolution estimate from Gctf. Gctf v.1.06 estimates the resolution of a micrograph as the resolution at which the cross-correlation coefficient (CCC) between the eqi-phase average and actual micrograph power spectrum falls to 0. We feel this overestimates the resolution and prefer to use a CCC cutoff of 0.5, as implemented in the ABSL_EPA_CC_threshold.py script. The estimated resolution allows users to determine if their datasets have high-resolution features, which can be a good general indicator of data quality. However, this has two caveats. First, the resolution estimate is reliant on signal in the images, so images of grids with continuous carbon film will appear to have higher resolution than images of vitreous ice containing a small protein. The second caveat is that this estimate is based on signal transferred through the imaging system and recorded on the detector. This is not necessarily signal from the biological specimen. It should also be noted, therefore, that although this resolution estimate is a good indicator of the quality of the micrograph (and the dataset), and thus a metric of microscope performance, the structure of the macromolecular complex being imaged might not be solvable to high resolution.

These on-the-fly analyses can also act as an early warning if there is a deterioration in microscope performance as a result of instability in air temperature or chilled water on the lenses. This is typically seen as a grid-square-independent deterioration in the resolution over time and/or changes in objective stigmation.

Concluding remarks

The aim of this protocol is to permit collection of high-quality single-particle data and their facile organization, storage and pre-processing as a prelude to 3D structure determination. Assuming a high-quality, stable and homogeneous macromolecular complex is imaged, this protocol, combined with single-particle image processing techniques, will lead to a high-resolution cryo-EM 3D reconstruction.

Reporting Summary

Further information on research design is available in the Nature Research Reporting Summary.

Data and code availability

An example dataset for cowpea mosaic virus has been uploaded to EMPAIR (EMPAIR-10205) and the corresponding reconstruction has been uploaded to EMD-3952; the output from the micrograph analysis script can be seen in Supplementary Fig. 1. The scripts described in the protocol, along with a modified copy of pipeliner.cpp for RELION, are available for download at https://github.com/Leeds-ABSL/ABSL_pipeline. A description of the scripts used is available in Supplementary Note 1. RELION v,2.1 is available on a GPLv2 license.