New hardware and workflows for semi-automated correlative cryo-fluorescence and cryo-electron microscopy/tomography.

Correlative light and electron microscopy allows features of interest defined by fluorescence signals to be located in an electron micrograph of the same sample. Rare dynamic events or specific objects can be identified, targeted and imaged by electron microscopy or tomography. To combine it with structural studies using cryo-electron microscopy or tomography, fluorescence microscopy must be performed while maintaining the specimen vitrified at liquid-nitrogen temperatures and in a dry environment during imaging and transfer. Here we present instrumentation, software and an experimental workflow that improves the ease of use, throughput and performance of correlated cryo-fluorescence and cryo-electron microscopy. The new cryo-stage incorporates a specially modified high-numerical aperture objective lens and provides a stable and clean imaging environment. It is combined with a transfer shuttle for contamination-free loading of the specimen. Optimized microscope control software allows automated acquisition of the entire specimen area by cryo-fluorescence microscopy. The software also facilitates direct transfer of the fluorescence image and associated coordinates to the cryo-electron microscope for subsequent fluorescence-guided automated imaging. Here we describe these technological developments and present a detailed workflow, which we applied for automated cryo-electron microscopy and tomography of various specimens.

Introduction

Correlative light and electron microscopy (CLEM) combines the advantages of fluorescence microscopy (FM) and electron microscopy (EM). FM provides positional as well as dynamic information on specific biomolecules. EM provides detailed, high-resolution information on cellular ultrastructure and protein structure, while also revealing the environment surrounding the molecule of interest. This makes EM and its extensions to 3D volume imaging a powerful tool for structural biology as well as the method of choice for detailed analysis of cellular morphology (Carroni and Saibil, 2016, Lučič et al., 2013).

A biological sample needs to undergo preparatory steps before it can be imaged by EM. The goal is to maintain the sample in a close-to-native state and ensure optimal structure preservation while enabling it to enter the high vacuum microscope column. In “traditional” EM the sample is dehydrated and embedded in resin prior to staining with heavy metals and imaging at room temperature. In contrast, in cryo-EM the sample is imaged in its fully hydrated, vitrified state. The samples can be prepared by quickly plunging them into ultra-cold liquefied gases (Dubochet et al., 1988, Taylor and Glaeser, 1974), or by freezing them under high-pressure (McDonald, 2009). For samples that are too thick for cryo-EM imaging (above ∼500 nm), thin slabs can be cut from the sample by cryo-sectioning (Al-Amoudi et al., 2004) or the sample can be milled with a focussed ion beam (Rigort et al., 2012). Contrast is generated directly by the electron density differences within the molecules. Hence, cryo-EM permits the morphological and structural characterisation of biological samples near to their native state and at high-resolution (Carroni and Saibil, 2016). In electron tomography (ET), 2D projection images acquired at different specimen angles are combined to generate a 3D reconstruction of the imaged volume (Irobalieva et al., 2016, Lučič et al., 2013).

In recent years various CLEM techniques using traditional sample preparation methods have provided remarkable insights into different areas of cell biology (Briggs and Lakadamyali, 2012, Bykov et al., 2016, Gibson et al., 2014, Redemann and Müller-Reichert, 2013, Sjollema et al., 2012). To maximize the accuracy with which fluorescence signals are identified in an EM image, both FM and EM need to be performed on the identical specimen (Kukulski et al., 2012, 2011). This raises a particular challenge in correlating FM with cryo-EM: the FM imaging needs to happen at temperatures below −140 °C where the cryo-EM sample does not devitrify. In addition, the sample must be shielded from atmospheric humidity during imaging and loading, while mechanical motion and optical distortions due to cooling or temperature gradients must be minimized. A number of solutions have been developed that add a special cryo-stage to a standard or adapted fluorescence microscope for cryo-CLEM (Zhang, 2013). Currently available solutions that fit an inverted fluorescence microscope feature a glass slide that separates the cold sample from the microscope objective lens (Sartori et al., 2007, Schellenberger et al., 2014). The minimum working distance (WD) reported for an inverted cryo-stage system is 410 μm with a 0.9 numerical aperture (NA) objective lens (Arnold et al., 2016). Solutions that attach to an upright microscope geometry do not require a separating glass slide (Liu et al., 2015, van Driel et al., 2009). The optical performance of all of these systems is limited by use of an immersion-free objective lens with relatively high WD and correspondingly low NA (Briegel et al., 2010, Koning et al., 2014). In a previous study using an upright geometry, we have shown that a short-WD dry objective lens with a high NA can be used for high-accuracy cryo-CLEM with a specially designed stage in which the objective approaches the sample closely (Schorb and Briggs, 2014).

We have developed a new dedicated cryo-FM system for CLEM. Our main goals during development of this system were: 1) to improve the transfer of the sample to and from the cryo-stage to reduce contamination by atmospheric moisture and to avoid sample loss due to warming or otherwise failed transfers; 2) to provide a mechanically and thermally stable stage appropriate for imaging multiple samples over long periods of time; 3) to maximize the optical performance; and 4) to incorporate the new hardware and its control software into a workflow that allows automated FM scans of entire cryo-EM grids and easy transfer of FM images and coordinates to the EM for use during data collection. Here, we describe and assess the new cryo-FM system. We provide a detailed workflow and protocol for using the cryo-FM system to automatically image an entire EM grid; for transfer of coordinates from FM to the software controlling the electron microscope; for cryo-EM or cryo-ET data acquisition; and for the image registration steps required to maximize the accuracy of correlation.

Results

Design of the stage and transfer shuttle

The cryo-FM stage is designed such that it can be mounted on a commercial upright microscope body (DM6FS, Leica Microsystems, Wetzlar, Germany). In this configuration the stage is fixed in Z but can move in X and Y directions, while the objective is inserted into the stage from above and moves axially for focusing. An overview of the stage and transfer shuttle design is given in Fig.1. Inside the stage, liquid nitrogen (LN), provided by an external pump, cools a metal block that supports the specimen. The interior of the stage is thermally isolated from the stage casing by the use of non-conductive materials. A high NA (0.9), short WD objective enters the stage through a port in the stage lid that tightly surrounds the objective (Fig.1G). The objective and the enclosing lid move relative to the flat stage surface during lateral (X and Y) movements while maintaining a seal to prevent influx of air. The front part of the objective is made of ceramic material with little thermal conductance to minimize heat transfer to the sample during imaging (Fig.1C). To avoid condensation inside the objective from environmental humidity, the spring mechanism for the front lens assembly was removed and the whole objective lens was sealed. In addition, a glass slide was added in the parfocal adapter that connects the objective lens and the microscope body to avoid build-up of condensation on the objective back lens. There is no intermediate cover glass between the objective and the sample, allowing the front lens to approach to 280 μm from the sample.

Fig. 1

The cryo-FM stage and transfer shuttle. A Photograph of the microscope, cryo-stage (CS) and the mounted transfer shuttle (TS), the LN-pump (P) in a 5L dewar (D), the pump controller (PC) and the fluorescence light source (FL). B Cutaway drawing of the cryo-stage (CS) and mounted transfer shuttle (TS). The rod (R) with the gripper (G) is extending through airlocks into the stage. C Photograph of the modified cryo-microscopy objective lens. The black and white ceramic cover towards the front of the lens drastically reduces thermal conductance. D Detail of the transfer shuttle. The cartridge that holds the EM grid is marked with a red box. The cartridge loading station (CLS) is shown immediately left of it; the gripper (G) that inserts the cartridge into the cryo-stage can be seen in the center. The circular slots at the bottom hold grid storage boxes. E Side view of the transfer shuttle attached to the cryo-stage during the loading/unloading of a specimen. F Upper panel shows a close-up view of the specimen inside the cryo-stage. The gripper (G) reaches inside the stage to place or pick up the cartridge from the cooled support shown on the left. Lower panel shows stage with lid on and objective inserted. G Detailed view of the cryo-stage with inserted objective. The black collar around the objective prevents condensation from entering the stage through the objective tube. The transparent lid is closed with covers to prevent stray light from entering the chamber.

The sample is loaded into the stage using a transfer shuttle that can be mounted onto the side of the stage (Fig.1A, B, E). The transfer shuttle is filled with LN that cools the loading assemblies and provides a cold, dry environment to transfer samples from storage containers to the cryo-FM stage and vice versa. A transparent lid with spacers covers the transfer station when not in active use.

Inside the transfer shuttle, the vitrified samples on cryo-EM grids are loaded into a specialized, commercial, flat, square, copper cartridge for cryo-FM imaging (Leica 16707511109) (Movie S1). The transfer shuttle contains three spaces for standard cryo-EM grid storage boxes (Fig.1D). A grid can be removed from a storage box and mounted into the cryo-FM cartridge using a dedicated loading station. In the loading station thin copper clips that secure the grid inside the cartridge (Fig.1D) are raised, allowing the grid to slide into its position within the cartridge, and lowered to secure it in place. These copper clips have minimized material thickness in order to allow the objective to approach the sample as closely as possible. (Other available cartridge systems for cryo-EM secure the grid using a clip ring assembly whose thickness prevents FM imaging at low WD.) This configuration provides mechanical stability and protection during loading and handling of the sample. The grid stays mounted in the cartridge during transfer and imaging. The cartridge loading station is detachable and can be baked-out separately from the transfer shuttle.

The transfer shuttle incorporates an extendable rod with a gripper at its tip into which the cartridge can be inserted. To transfer the sample into the cryo-FM, the shuttle is docked to the stage, the rod is extended through a port in the wall of the shuttle directly into the cryo-FM stage and the gripper is released, leaving the cartridge in the stage (Fig.1F). The transfer shuttle is removed before imaging. During imaging the cartridge sits on top of an LN-cooled copper block. The design is such that vibrations from the pump or the evaporation of LN are minimized. A temperature sensor regulates the pumping rate and LN flow. A heater element is embedded in the copper block for adjusting the temperature and for baking out the system after use. A hole in the block underneath the specimen allows brightfield imaging. The vaporising dry nitrogen gas creates an overpressure atmosphere in the stage and prevents contamination by frost from condensing atmospheric humidity.

The microscope is controlled using a dedicated CLEM module that we integrated into Leica’s existing MatrixScreener HCSA (High Content Screening Automatization) software. This module provides the required functionality for generating grid-wide scans of cryo-EM specimens. The generated data format allows a direct exchange of images and coordinate lists with the software controlling the electron microscope.

A workflow for correlative cryo-FM and cryo-EM

Here we give an overview of the workflow for correlative cryo-FM and cryo-EM using the system. In the Appendix we present a detailed step-by-step protocol. A schematic illustration with approximate times is shown in Fig. 2.

Fig. 2

The cryo-CLEM workflow and estimated times for plunge-frozen specimens. The individual steps of the workflow are show in the left column, with estimated durations listed on the right. Procedures that should happen without a longer interruption or that need to be repeated are grouped in gray boxes.

Sample and grid preparation

For cryo-EM imaging the sample is applied to metal (copper, gold) grids covered by a thin holey carbon support film. When choosing suitable grids for the cryo-CLEM experiment, the field of view (FOV) of the camera/microscope combination should be considered. It is convenient if an entire grid square fits within the FOV. In our case, with a FOV of 179 × 134 μm, we choose either 200 or 300-square mesh grids. We also observed that some types of support film, such as Quantifoil (Quantifoil Micro Tools GmbH, Groβlöbichau, Germany), exhibit auto-fluorescence particularly in the green channels when cooled, while in others (C-Flat, Protochip Inc., Morrisville, NC, USA) the effect is not as strong. This also needs to be considered when choosing the appropriate grid type.

Aligning and registering images between FM and EM is done in three steps. First, a rough registration is preformed using grid-wide landmarks that are visible in both imaging modalities. Irregularities in individual grid squares such as defects in support film or arrangements of sections/cells are typically sufficient landmarks. Alternatively, patterns of regular landmarks are available on commercial “finder-grids”. Second, we refine the coordinate registration between FM and EM locally to acquire high-resolution EM images or tomograms precisely at the desired positions. To maximize the correlation accuracy we perform a third, post-imaging registration step that makes use of plastic microspheres that are fluorescent in multiple channels and whose electron density is sufficient to identify them in EM images (Kukulski et al., 2011, Schorb and Briggs, 2014). We apply these fluorescent beads (TetraSpeck, Thermo Fisher Scientific, Waltham, MA, USA) to the support film of the grids before plunging as described in (Schorb and Briggs, 2014). The choice of beads is influenced by the fluorescence properties of the sample. The size and thus the brightness of the TetraSpecks should approximately match the expected intensity of the target signal in the desired channel. We typically use beads with a diameter of 100 nm while for very dim specimens 50 nm TetraSpecks that are 8 times less bright can be used. We also add 10 nm colloidal gold beads as alignment markers for cryo-ET or for the precise registration of images acquired at different EM magnifications (Kukulski et al., 2011). We vitrify the specimen by plunge freezing. The exact protocol for preparation of the EM grid will necessarily depend upon the sample the user wishes to study.

Transfer of sample into cryo-FM stage

The cryo-FM transfer shuttle is cooled by liquid nitrogen and all transfer steps are performed in a cold, dry nitrogen gas atmosphere. The grids are transferred into the shuttle in standard, round, cryo-EM grid boxes. A single grid is loaded into the imaging cartridge. The imaging cartridge has two thin clips that secure the grids. These clips are released for loading and unloading of the grid using the cartridge loading station (CLS) (Fig.1D and Movie S1). To move the cartridge from the CLS into the microscope stage, it is picked up with the loading rod. The transfer shuttle as a whole is moved to the cryo-light microscope and docked to the entry port on the side of the cryo-stage. After opening the two shutters that seal the port, the loading rod is inserted into the cryo-stage. Once the cartridge is released in place, the loading rod is retracted, the shutters are closed and the transfer shuttle is removed.

Cryo-light microscopy

To screen the overall condition of the specimen regarding ice quality and thickness, we perform a brief inspection by brightfield imaging. In order to register the grid with the coordinate system of the microscope we identify the grid’s center mark and store it in the software. As an EM grid is never perfectly flat, the next step is to generate a “focus map” of the grid to determine its topography. To speed up the procedure in the absence of bright fluorescence signals, a first coarse scan using brightfield illumination can be performed across the grid. Subsequently, the focus map is generated by autofocus scans using the fluorescence signal from the fiducial beads or the specimen.

The entire process from loading the grid into the cartridge until the grid-wide focus map is generated takes about 15 min per grid (Fig. 2). During data collection, we need to collect multiple images at different focal levels at each position (a “Z-stack”) to compensate for differences in sample height across the FOV. We therefore define the number of focal levels to be collected depending on the thickness and flatness of the sample (a typical Z-stack size is 10 μm with 1 μm increment). We also define the number of colour channels and adjust the illumination intensity and camera exposure time for each colour channel. The intensity settings usually stay the same for different grids from one batch while the settings for the Z-stack may differ from grid to grid.

Based on these settings, the CLEM acquisition software can run an automated scan of the entire grid, acquiring a multi-colour, multi-Z image stack at each XY-position of the grid. It uses image overlap to stitch the single tiles into a map of the full specimen. The time required to scan the grid depends on the number of desired channels as well as their exposure time and the Z-stack size. Typically a full acquisition takes about 10–30 min per fluorescence channel. Once the acquisition is complete, the cartridge is removed from the cryo-FM and the grid retrieved via the transfer shuttle and stored. The image data can be transferred to another computer for analysis while other grids are imaged.

The Cryo-FM system is compatible with different cameras. Images in this manuscript were collected on a Leica DFC-365FX (Figs. 3A–C, 5, 6) or a Hamamatsu ORCA-Flash4.0 (Figs. 3D–F, 4).

Fig. 3

Cryo-FM imaging of frozen-hydrated sections and whole vitrified cells. A Stitched cryo-fluorescence mosaic image, as generated during the automated grid scan, of vitreous sections of high-pressure frozen yeast cells with alpha-subunit of COPI fused to mCherry. Two section ribbons of 60 nm (left) and 100 nm (right) nominal thickness are shown. Scale bar: 200 μm. B Zoom in on the grid square marked in A. Scale bar: 20 μm. C High magnification view of the box marked in B. The cytoplasm is slightly autofluorescent in the green channel, mCherry signal appears in red, TetraSpeck beads appear white/magenta and are fluorescent in green, red and far red (channel displayed in blue). Scale bar: 2 μm. D Cryo-fluorescence mosaic of a grid with plunge-frozen HeLa cells expressing COPII coat component Sec23 labeled with EYFP (Verissimo et al., 2015) grown on a gold EM grid. Scale bar: 200 μm. E Zoom in on the grid square marked in D. Scale bar: 20 μm. F High Magnification view of the box marked in E with EYFP signal displayed in green and TetraSpeck beads in blue (far red channel). Scale bar: 2 μm. All images shown are maximum projections of Z-stacks.

Fig. 5

The procedure to set up automated EM acquisition at fluorescent POIs. A Cryo-fluorescence mosaic of the whole grid, merge of red and green channels. The yellow circles indicate grid-wide reference landmarks such as torn support film. Scale bar: 200 μm. The inset A2 is a magnified view within circle 2 showing a typical grid-wide landmark: a corner of a hole in the support film. Scale bar: 10 μm. B Full cryo-EM grid map acquired by SerialEM. The same landmarks as in A are marked. The inset B2 shows the same feature as marked in A2. Scale bar: 10 μm. C Magnified FM view of a single grid square. Virus particles appear as green spots while the multi-colour fluorescent beads (TetraSpecks) are visible in both channels. Yellow circles indicate landmarks used to locally align this image with the corresponding EM acquisition. As local landmarks we typically use pairs or clusters of fiducial beads, which appear as bright fluorescent signals. White boxes indicate fluorescent signals marked as target POIs to be automatically acquired in EM. The blue rectangle indicates the part of the image used for the post-acquisition high-accuracy correlation illustrated in Fig. 6A. Scale bar: 200 μm. D Cryo-EM mosaic acquisition of the same grid square. Yellow circles indicate local landmarks used to locally register the FM image of the grid square (C). White boxes indicate the positions where high-magnification images were recorded based on the transformed positions of the fluorescent POIs. The blue rectangle indicates the region in the grid square where the post-acquisition high-accuracy correlation illustrated in Fig. 6B was performed. Scale bar: 10 μm. E Magnified view of the landmark feature c, which is a cluster of three ∼100 nm TetraSpeck beads. White arrows indicate the 10 nm gold fiducials. Scale bar: 200 nm. F Magnified view of the cryo-FM image showing a fluorescence signal of interest. The white box corresponds to the FOV of 1.24 μm of the high-magnification EM image automatically acquired at this spot barely covering 10 × 10 FM pixels.

Fig. 6

Post-acquisition high-accuracy correlation. A The area of interest in the FM image (marked with blue box in Fig.5C) surrounding the signal of interest circled in red. The TetraSpeck beads that are used for the high-accuracy correlation are marked with yellow circles. The blue area indicates the location of the image shown in panel B. Scale bar: 5 μm. B Single medium-magnification cryo-EM image of the region corresponding to the blue boxes in Figs. 5C, 5D and 6A, acquired after the high-magnification imaging at the POI is finished. The positions of the TetraSpeck beads corresponding to those in A are marked in yellow. The blue square indicates the position and FOV of the high-magnification image in C. The red circle is centered on the predicted coordinates of the feature of interest calculated by the post-acquisition correlation procedure using the fiducial beads. Scale bar: 2 μm. C High-magnification image at the POI. The red circle marks the coordinates predicted by post-acquisition high-accuracy correlation. The circle has a radius of 50 nm. Scale bar: 100 nm. D to G images and analogous coordinate predictions for four other particles. Image E was acquired on the carbon support film. Radius of the white prediction circles: 50 nm. FOV: 500 nm.

Fig. 4

Correlated cryo-FM and cryo-ET. HeLa cells were transfected with the HIV-1-derived plasmids pCHIV and pCHIV.mCherry (Lampe et al., 2007), as described in (Baumgärtel et al., 2011) and vitrified by plunge-freezing. A Grid square overviews. TetraSpeck beads appear in green (imaged in far red) and HIV Gag-mCherry signal in red. Scale bar: 20 μm. B Zoom into the marked areas of interest. Scale bar: 2 μm. C Superposition of the cryo-EM montage of the selected area and the transformed fluorescence image after correlation. Scale bar: 2 μm. D Central section through a tomogram collected at the marked point of interest. Upper panel shows an individual mature HIV-1 particle. The dashed circle marks the predicted position of the fluorescent signal as determined by post-acquisition high-accuracy correlation. Lower panel displays virus assembly sites at the plasma membrane of the cell. Tomograms were collected on a Titan Krios electron microscope (FEI) equipped with Quantum 967 LS energy filter and K2 direct detector (Gatan) at a pixel size of 1.32 Å and reconstructed using IMOD (Kremer et al., 1996). Scale bar: 50 nm.

Annotation of FM images to define positions of interest

The cryo-FM images are next analysed to determine the positions of interest (POI) where EM acquisition should take place. This can be done by manual selection of fluorescent signals (as was done in this study), or combined with image analysis routines (e.g. particle picking) within the FM control software (LAS X) or using another software. We also mark the coordinates of grid-wide landmarks, such as broken grid squares or marks on grid bars. These are called “Grid reference positions” within the software. The two coordinate lists are saved in a file format that can be directly read by the software package used for EM.

Cryo-electron microscopy

The grids are then loaded into the EM. We control the EM using the software package SerialEM (Mastronarde, 2005). It is capable of mapping imported images and associated coordinate systems onto the EM’s internal stage coordinates (Briegel et al., 2010, Mastronarde, 2016). Its “navigator files” can store the coordinates of selected points as well as links to image files and associations to other coordinate systems. The coordinate lists and cryo-FM images can be opened and visualised inside SerialEM.

We collect an initial overview montage of the entire grid using SerialEM. Within the EM montage we then mark the grid-wide landmarks previously selected in the FM images. Using the matching pairs of landmarks as registration points, SerialEM then transforms the grid-wide FM coordinates onto the EM coordinate frame. The accuracy of the grid-wide coordinate registration is sufficient to target grid-squares containing POIs. We next record montaged EM maps at an intermediate magnification covering each grid square of interest and perform a second, local, coordinate registration from FM to EM on the grid square level using the fluorescent beads as fiducial markers. The POIs previously identified by FM have now been located in the EM coordinate frame. High-magnification cryo-EM images or tomograms can then be collected at all POIs in an automated manner.

Post-acquisition high-accuracy correlation

After the acquisition, the FM and EM images can be more accurately aligned using the TetraSpeck beads as fiducial markers (Fig. 6). Based on coordinates of these fiducial markers, the mathematical transformation that relates the FM and EM images can be calculated. Identification of the beads in the FM and EM images and alignment of the images can be directly performed using the scripts described in (Kukulski et al., 2011) which are implemented in MATLAB (MathWorks, Natick, MA, USA). These scripts, with modifications that adapt for file formats and further improve ease of use, were used for the experiments described here and are available for download from http://www.embl.de/download/briggs/cryoCLEM/index.htm. With this post-acquisition bead-based registration we can achieve high correlation accuracy of ∼100 nm or less. Other software packages such as Icy (de Chaumont et al., 2012, Heiligenstein et al., 2016) can also be used.

Experimental performance

The design of the stage and particularly the transfer shuttle has reduced contamination of the sample by atmospheric humidity during transfer as compared to our previous instrument (Schorb and Briggs, 2014). Inside the cryo-stage, contamination is negligible and multiple grids can be imaged over 6 h without the need of a bake out. The cryo-FM stage provides mechanically stable imaging conditions. During the typical imaging period for one grid (about one hour for acquisition of a full grid map in three channels) mechanical drift is minimal, and is not sufficient to affect the focussing of tiles or the alignment of the full mosaic. For the duration of a single exposure we do not observe vibrations of the stage that would blur the image.

Together with the associated software that allows automated cryo-FM imaging, the new system provides an efficient cryo-CLEM workflow. The key advantages of this workflow are the ability to run automated mosaic cryo-FM scans covering the complete grid, and the ability to subsequently run automated cryo-EM data collection on the selected POIs. The entire process, from sample preparation and cryo-FM imaging to targeted acquisition of 300 high-magnification cryo-EM images or 25 tomograms can be performed within 2 days. Freezing and FM of 6 grids can be achieved in one standard working day. Signals of interest are then identified and a second day is spent on loading the EM, mapping grids, aligning the coordinate systems, setting up the imaging conditions and running the automated acquisition which can continue overnight.

Proof of principle applications

Representative cryo-FM grid-scans of plunge-frozen cells and vitreous sections are shown in Fig. 3. An example of the use of the system to perform targeted cryo-ET is shown in Fig. 4. In this example, cells transfected with a Human immunodeficiency virus (HIV-1) variant where the Gag protein is tagged with mCherry (Lampe et al., 2007), were vitrified by plunge freezing and imaged by cryo-FM. Target positions were identified based on the presence of red signals in the cryo-FM image that were located within regions of appropriate ice thickness in montaged intermediate magnification cryo-EM maps. These positions were imaged by cryo-ET revealing virus assembly sites and recently released virus particles.

To demonstrate the use of the system for automated data collection, we applied the full workflow to plunge-frozen, fluorescent p22 bacteriophage particles as previously described in (O’Neil et al., 2012, O’Neil et al., 2011; Schorb and Briggs, 2014). These provide fluorescent point sources that can be unambiguously identified in cryo-EM, and therefore allow quantitative assessment of the correlation performance. We generated full mosaic scans of two grids by cryo-FM (Fig.5A, acquisition parameters and detailed protocol, see Appendix). Based on the FM images we identified 9 grid squares of interest, distributed over the two grids, and marked 423 fluorescent signals as POIs (Fig.5C). We then collected high-magnification images of the POIs. For this experiment we used an FEI Polara microscope (FEI Company, Hillsboro, OR, USA) equipped with GATAN US4000 CCD (Gatan Inc., Pleasanton, CA, USA) camera operated at 100 kV at 31000x magnification with 3.5 μm underfocus and an electron dose of 20 e/Å2 per image. The whole experiment from freezing to data collection was performed in 2.5 days.

To verify the success of the correlation procedure, we analysed a subset of 100 acquired images. In these we found 87 virus particles within the FOV (1.24 μm squared). In 5 images there was aggregated material that is likely to have given rise to the fluorescence signal while 1 image contained a fluorescent bead instead of a virus. In 5 images we found nothing, and for 2 images the EM acquisition failed. For 98 successful image acquisitions, 93 contained the fluorescence signal source, a hit-rate of 95%. To assess the possibility of imaging a fluorescent specimen by pure chance, we set up 20 acquisitions randomly positioned in a neighbouring grid square. In these we found 1 virus and 1 fiducial bead within the FOV while 18 images contain no visible feature (10% hit-rate). We next used the deviation of the positions of the fluorescent objects from the center of the images to estimate the accuracy of image acquisition. This deviation is approximately normally distributed with a standard deviation of 227/256 nm (in XY direction), which is consistent with the observed 95% hit-rate. This distribution indicates that data could be acquired with a FOV of 900 nm while still expecting an 80% hit rate. The largest source of inaccuracy during this step is imperfect registration of images during image montaging within the SerialEM software. We have been able to obtain improved targeting precision when using other software such as IMOD to generate image montages (Kremer et al., 1996).

We then performed post-acquisition high-accuracy coordinate registration on a subset of the positions exactly as described in (Kukulski et al., 2011, Schorb and Briggs, 2014). Most users take about 5 min per registration for this step. For 52 high-magnification images, where the predicted location of the fluorescence signal is clearly associated with a virus-like particle visible in the image, we measured the deviation of the predicted coordinate from the center of the observed particle (Fig. 6). This deviation is normally distributed for each coordinate axis with standard deviations of 31 and 37 nm in X and Y. We note that the obtained accuracy for a specific sample will be dependent on multiple factors, including the intensity of the fluorescent signal (and therefore the ability to define its center) and the distribution, density and intensity of the Tetraspecks.

Discussion

Here we have introduced new hardware and software for correlative cryo-FM and cryo-EM and described a workflow by which they can be used for semi-automated imaging. The stage and transfer shuttle permit use of a 0.9 NA short working-distance objective. The system offers high stability and reduced contamination to facilitate imaging. The instrument together with the dedicated software provides the capability of doing automated scans of full cryo-EM grids. Together these developments have provided us with a dramatic increase in throughput and reliability when compared to our previous system (Schorb and Briggs, 2014). The workflow enables an offline selection of coordinate lists for both landmarks and POI before going to the EM. These lists can then be directly imported to the EM control software and used as basis for automated acquisition.

The cryo-CLEM system can be used to study specimens where there is only a single target signal per grid or numerous events separated only by the diffraction limit of the fluorescence microscope. Where the fluorescent signal can be identified above background it should be possible to locate it in the electron microscope. We envisage that the ease-of-use of the workflow makes it feasible to add cryo-FM as a routine step between sample preparation and cryo-EM/ET data collection. In particular for cellular and other heterogeneous samples this will facilitate efficient and targeted cryo-EM data collection, while providing complimentary information.

Conflict of interest statement

M.S., L.G., R.L. and J.A.G.B have filed patents related to this work. Intellectual property therein has been licensed by Leica Microsystems who have commercialized the cryo-FM system and released it as a product.