Precision in situ cryogenic correlative light and electron microscopy of optogenetically positioned organelles

Advances in structural biology have seen a dramatic switch over the past 15 years from X-ray crystallography-based methods to single-particle cryo-electron microscopy (cryo-EM)-based techniques and now, increasingly, to structural analyses in the cellular context. In situ structural biology not only provides the near-atomic resolution of proteins and/or complexes of interest but also has the advantage of providing the environmental context of how these proteins and macromolecular complexes interact with cellular machinery and organelles of interest, which better informs our understanding of cellular function (Bauerlein and Baumeister, 2021; Wolff et al., 2016; Wu et al., 2020). Although these latest advancements show almost unending possibilities for understanding how proteins and complexes elicit their cellular roles, especially in the context of visual proteomics (Bauerlein and Baumeister, 2021), the application of these methods has been somewhat slowed by several complicating and contributing factors. These include (1) the intricate and expensive interconnected workflows and manual handling steps required between disparate instruments, which often result in damage to the sample and ice contamination; (2) the highly concentrated environment of the cytoplasm, which obscures regions, structures and proteins of interest; and (3) the low signal-to-noise ratio of cryo-EM and cryo-electron tomography (cryo-ET) datasets. As a consequence, the native state cellular architecture remains largely unexplored. Improvements to the ease of data collection, sample scalability and region of interest (ROI) targeting will expedite the adoption of these techniques by the broader scientific community (Bauerlein and Baumeister, 2021; Fu et al., 2019; Wolff et al., 2016; Wu et al., 2020).

Cryogenic correlative light and electron microscopy (cryo-CLEM) combines the resolving power of cryo-EM with the specific illumination of structures of interest by cryo-fluorescence microscopy (cryo-fLM) for the precise localisation of distinct cellular targets. Recent advancements in cryogenic light microscopy stages are now facilitating the combined power of these previously separate techniques. Although there has been an increase in the adoption of cryo-CLEM-based methods in recent years (Carter et al., 2020; Tran et al., 2021; Fu et al., 2019; Klein et al., 2021; Mageswaran et al., 2021; Wu et al., 2020), these techniques remain an emerging field of application. Improved and integrated systems as well as simplified workflows will further facilitate the adoption of these technologies.

Here, we develop a system for simplified correlation of structures of interest by coupling super-resolution Airyscan cryo-confocal microscopy and optogenetic manipulation techniques. We characterise PtK₂ cell morphology and cell substrate interactions to optimise and improve the yield of cryo-ET-compatible regions, and we develop expedited cryo-fLM data acquisition strategies that deliver a simplified correlation workflow for high-precision targeting of ROIs for in situ structural analyses. Finally, we optimise an optogenetic-based system for analysis of perinuclear-localised structures in PtK₂ cells without the requirement for cryo-focused ion beam (cryo-FIB) milling. We redistribute lysosomal-associated membrane protein 1 (LAMP1)-associated organelles from thick perinuclear regions to the thin cell periphery using blue light stimulation and cryptochrome 2 (CRY2)–CIBN conditional association with the kinesin motor KIF5a for nanometre-resolution in situ cryo-ET of late endosomes and lysosomes (here termed endolysosomes). These advancements improve the precision and broaden the applicability of cryo-CLEM for imaging of specific structures of interest in the cell. This method eliminates the requirement for the use of cryo-FIB as an intermediate step to mill cryo-lamellae on grids and thus (1) reduces and simplifies existing manual handling steps and cryo-transfer requirements; (2) improves cryo-ET throughput and scalability; and (3) enhances the cost-effectiveness and accessibility of in situ cryo-EM workflows. This technique, which can proceed from the living cell to cryo-ET of cellular organelles in a day, brings in situ cryo-EM within reach of cell biologists.

We sought to develop an efficient cellular cryo-CLEM workflow that would allow precision targeting of ROIs by cryo-ET for cells grown on a mechanically defined substratum where organelles and/or structures of interest could be analysed at high resolution without cryo-lamella preparation (Fig. 1A,B). High-quality cryo-tomography requires a sample with a thickness of ∼300 nm (or less). Therefore, generation of cryo-tomograms without cryo-FIB milling requires a cell type with a protracted and thin cytoplasm. Towards this aim, we performed a screen of multiple cell lines to determine the most appropriate cell type for in situ cryo-EM. We analysed general cellular morphology and cytoplasmic thickness by imaging ultrathin vertical sections of conventionally processed cells using transmission electron microscopy (TEM). We observed that PtK₂ cells, and to a lesser extent U-2OS and WI-38 cells, possess a thin cytoplasm with abundant organelles extending from perinuclear regions to the cell perimeter, which contrasts with the morphologies of the other screened lines (MDCK, BHK, HeLa, MEF and WI-38; Fig. 2A). As PtK₂ cells are also readily transfected with lipid-based reagents, we chose to pursue their further characterisation.

In situ analyses require that cells can grow on grids for cryo-EM imaging and still maintain a thin cellular periphery. We therefore characterised PtK₂ cells grown on Quantifoil R2/2 gold support carbon film grids. We performed serial blockface scanning electron microscopy (SBF-SEM) and conventional TEM on cells grown on grids. SBF-SEM demonstrated that PtK₂ cells adopt a ‘fried egg’ morphology on grids, with thick nuclear and perinuclear regions that reduce into a thin cytoplasm of ∼100–300 nm thickness extending to the cell perimeter (Movie 1). This region contains abundant cellular organelles (Fig. 2B,C; see dotted circles in B). Segmentation of cells grown on the Quantifoil R2/2 gold support carbon film grids demonstrated numerous regions of a suitable thickness for cryo-ET without the requirement for cryo-FIB lamella preparation (Fig. 2D). The PtK₂ cells also showed excellent binding and uptake of Alexa Fluor 647-conjugated transferrin (transferrin A647), which colocalised with exogenously transfected GFP–Rab5a (Fig. S1A) and GFP–Rab4a (Fig. S1B). This highlights the utility of PtK₂ as an ideal cell line for endolysosomal studies. Additionally, we analysed the transfection efficiency of PtK₂ cells on Quantifoil R2/2 gold support carbon film grids (Fig. S1C) and the plastic surface surrounding the grid (Fig. S1D), finding that transfection efficiency on the plastic surface was slightly higher than that on the grid; however, this difference was not statistically significant (Fig. S1E).

We sought to optimise and maximise the number of compatible ROIs for in situ cryo-ET without cryo-FIB lamella preparation. As the mechanical properties of cell substrates can be tuned to optimise important parameters such as cell spreading and cell–substrate adhesion, we first characterised the compliance or stiffness of the carbon film substrate for different regions of the carbon film across the grid square using atomic force microscopy (AFM) (Fig. 3A, regions 1–3). We observed a higher substrate stiffness for the carbon film overlaying the gold grid bar (region 2) and for the gold grid bar itself (region 3) compared to the Young's modulus of the carbon film in the centre of the grid square (Fig. 3B–D; Fig. S2A; region 1). This result aligns with our previous observations that mature focal adhesions preferentially form on areas supported by the gold grid bars (Cagigas et al., 2021) and observations made by others that cells preferentially align to the grid bar without modification of the substrate (Toro-Nahuelpan et al., 2020). We next explored how changes in carbon substrate thickness could modulate support film stiffness properties. We deposited either 5 nm or 12.5 nm of amorphous carbon onto the front side (Quantifoil carbon side) of a Quantifoil R2/2 gold support carbon film grid. This addition resulted in ∼4-fold increase (5 nm carbon) or ∼8-fold increase (12.5 nm carbon) in the Young's modulus of the carbon film in the centre of the grid square (region 1) compared to that of control grids with no additional carbon deposition (Fig. 3B–D). We observed a 2-fold increase in Young's modulus on the grid bar (region 2) when comparing grids with carbon addition to control grids (Fig. 3C) as well as a reduction in stiffness on the gold grid bar upon addition of carbon (region 3; Fig. 3D). This reduction is a consequence of the addition of the less stiff amorphous carbon layer on top of the gold surface. We also performed Young's modulus measurements at seven locations across the grid square (Fig. S2B), with location 1 near the centre of the grid square and location 7 near the grid bar. We found a linear gradient increase in the Young's modulus measurement from location 1 to 7 (Fig. S2C). Based on these results, we resolved to use front-side deposition of 12.5 nm of amorphous carbon on Quantifoil R2/2 gold support carbon film grids for all our experiments, hereafter referred to as +Carbon grids. These data demonstrate that the thickness of the carbon substrate represents a mechanically tuneable property to control substrate stiffness to potentially manipulate cell adhesion properties.

Given that carbon addition modifies substrate stiffness, and that substrate compliance is an important parameter to control cellular adhesion (Engler et al., 2004), we next explored whether additional carbon deposition could alter cellular distribution across the grid square. We mapped the spatial abundance of the centre of mass of nuclei of cells seeded onto control grids and +Carbon grids (see Materials and Methods; Fig. S3A,B). We found preferential positioning of nuclei at the edge of the grid square for control grids (Fig. S3C,E). This preference was reduced when cells were grown on the stiffened +Carbon substrate (Fig. S3D,E). Cells grown on stiffer substrates achieve greater cell spread and two-dimensional (2D) surface area compared to those seeded on more compliant substrates (Engler et al., 2006). This suggests that with greater substrate stiffness we could potentially increase the total number of areas suitable for cryo-ET by simultaneously reducing the number of cells localised to the edge of the grid square (where the ice is thickest) and by increasing the spread of cells to reduce their thickness at the cell periphery. Substrate compliance is known to organise and remodel the actin cytoskeleton (Doss et al., 2020; Gupta et al., 2015; Prager-Khoutorsky et al., 2011). We therefore assayed stress fibre formation and actin organisation in PtK₂ cells grown on control grids and on +Carbon grids. Cells grown on control grids that localised only to the carbon film with no association with the gold grid bar demonstrated a significant reduction in total phalloidin intensity per cell compared to similarly localised cells grown on +Carbon grids (Fig. 4A–C). We also observed an increase in the proportion of stress fibres relative to whole-cell area (Fig. 4A,B,D) in cells grown on +Carbon grids compared to those grown on control substrates. Strikingly, cells grown on control substrates that localised to the gold grid bars demonstrated increased stress fibre formation compared to cells localised only to the grid square film (Fig. 4A,C,D); this difference was less pronounced in cells grown on the stiffened substrate that localised to the gold grid bar (Fig. 4B,C,D).

These observations led us to investigate focal adhesion formation and stability, comparing cells grown on control and +Carbon grids. We assayed focal adhesion size by using immunofluorescence to detect paxillin, which is a marker of focal adhesions and a regulator of focal adhesion formation and stability. We observed a significant increase in the size of focal adhesions for cells on the carbon film of +Carbon grids compared to those of similarly localised cells on control Quantifoil R2/2 gold support carbon film grids (Fig. 4E–G), suggesting that the additional carbon coat supports focal adhesion maturation. We also observed that focal adhesion size was rescued in cells grown on the control grids where the focal adhesions were localised to the gold grid bar (Fig. 4H); this rescue was less pronounced when comparing between film and grid bar on +Carbon grids (Fig. 4I). Therefore, we analysed the 2D spread area of cells grown on control and +Carbon stiffened substrates and observed a significant increase in 2D area of cells seeded on +Carbon grids (Fig. 4J). Taken together, these data suggest that substrate stiffness is an important consideration for growing cells on grids and may represent a tuneable parameter to modify cell size, shape and substrate adhesion. These optimisation steps could aid data throughput by simultaneously increasing the number of thin peripheral areas suitable for cryo-ET analyses while bypassing the significant bottleneck of cryo-lamella preparation.

Having optimised cell–substrate interactions for improved yield of thin peripheral ROIs for cryo-ET, we next pursued improvements in the speed and accuracy of imaging fluorescently labelled organelles of interest by cryo-CLEM (schematically depicted in Fig. 1A,B). To test our rapid workflow, we pursued a data acquisition strategy that would encompass cryo-preservation to high-resolution data acquisition in less than a day. Cells were seeded onto +Carbon grids at low density to minimise potential cellular overlap. We observed that 50–100 cells seeded per grid allowed for good quality and reproducible back-sided blotting for plunge freezing with sufficient areas for cryo-tomography (Fig. S4A). PtK₂ cells were transfected with EYFP–Mito7 as a marker of mitochondria and then plunge frozen 24 h after transfection. Fig. S4A demonstrates low-magnification cryo-fLM image of a grid of EYFP–Mito7-transfected cells. Areas with thick ice are readily discernible by the high level of autofluorescence (Fig. S4A). The grid was mapped, the resulting images were stitched together, and ROIs were selected based on low expression of EYFP–Mito7 with thin cell peripheries extending toward the centre of the grid square. Confocal z-stacks were acquired at ROIs 1 and 2 using a 100× objective (0.7 NA, 4 mm working distance) with the Airyscan 2 detector. This magnification with a digital zoom of 0.5 was sufficient to acquire whole-grid-square maps at a resolution high enough for targeting mitochondria (Movies 2 and 3). For high-resolution re-registration, both fluorescence and transmitted light channels must be acquired. This strategy allows for the use of the Quantifoil holes as landmarks between imaging modalities. Accurate correlation of the holes in the carbon film between cryo-fLM and cryo-EM imaging provides a sub-2 µm fit corresponding to the diameter of the hole for all fluorescently labelled structures.

Following this, the frozen grid was loaded into the cryo-transmission electron microscope, a low-magnification whole-grid atlas was acquired (Fig. S4B), and whole-grid-square ROIs were identified and re-registered (Fig. S4A,B). Broken grid squares (white asterisks) represent a simple and rapid fiducial marker to correlate cryo-fLM images to cryo-TEM micrographs at the whole-grid scale. Cryo-TEM images at 700× magnification were acquired at the ROIs and fitted to the grid square confocal z-stack using a combination of cellular landmarks and the position of the Quantifoil carbon film holes as correlation markers (Fig. S4C,D). Images were fitted in real time using the Puppet Warp tool in Adobe Photoshop. From this data, electron-transparent regions in the cell periphery could be identified and correlated to EYFP–Mito7 signal for targeted analysis at the individual hole scale (Fig. S4C′,D′; white arrowheads). ROIs 1 and 2 were suitable for cryo-tilt series acquisition (Fig. S4A–D′; green and magenta squares); however, ROIs 3 and 4 (Fig. S4A,B; red and blue squares) represented areas too thick for cryo-ET. This was observed in the z-stack (Movies 4, 5 and 6) and was confirmed through 700× magnification correlation at the grid-square level (Fig. S4I–J′). For expedited analyses, ROI thickness should be screened and excluded during the cryo-fLM imaging. Fig. S4E and F demonstrate an intermediate-magnification (2600×) overlay of ROI 2, showing EYFP–Mito7 signal correlating with electron-dense structures with a mitochondrial morphology. From this overlay, holes of interest (HOIs) were selected for batch cryo-tilt series acquisition. Batch tilt series were acquired overnight at each HOI with 20–30 tilt series acquired per grid. The images acquired for HOIs 1 and 2 demonstrate high-magnification precision targeting of fluorescence signal and correlation with mitochondria (Fig. S4G–H″). This image acquisition strategy results in a highly accurate demarcation of larger organelles of interest quickly. For these assays, we plunge froze live cells in the morning and acquired batch tilt series overnight, starting on the same day.

In the extended peripheral regions of the PtK₂ cell cytoplasm, mitochondria occupy most of the cytoplasmic depth (Fig. 1A,B), which effectively results in a near-2D correlation method. To extend our studies to a cellular structure that does not occupy the entire cytoplasmic volume, and thus an experimental condition that represents a more complex three-dimensional (3D) re-registration process on a smaller structure, we performed cryo-CLEM analyses on GFP-tagged doublecortin X (EGFP–DCX) as a marker of microtubules (Bechstedt and Brouhard, 2012). Using this method, we observed microtubules in our fluorescence images and correlated those signals to individual microtubules in the 3D volume of the reconstructed tomograms (Fig. S5A–H; Movie 7). These data were acquired in under a day, demonstrating that this protocol represents a robust and rapid approach that can be used to target smaller macromolecular complexes in the condensed, but still 3D, volume of the PtK₂ cytoplasm. This direct method of imaging is highly applicable to the study of peripherally localised organelles such as endoplasmic reticulum, mitochondria, autophagosomes and endocytic vesicles as well as cytoskeletal elements.

The major constraint to the method described thus far is that imaging is restricted to the thin cell periphery. This places a physical limitation on our ability to image structures contained within thicker regions of the cell such as endolysosomes and recycling endosomes (Fig. 1A). To enhance the utility of this novel PtK₂ cell system, we expanded our analyses to include imaging of perinuclear-localised organelles without cryo-FIB milling. To do so, we sought to redistribute perinuclear-localised organelles to the peripheral cytoplasm by optogenetic positioning of organelles (OPO) prior to imaging with cryo-CLEM – a combined approach that we term OPO–cryo-CLEM. Systems commonly used to spatially modulate protein localisation include electrogenetic (Murata et al., 2005), chemical genetic (Thang Manh et al., 2013) and photosensitive optogenetic switches (Kramer et al., 2021); with the latter having the best spatial and temporal control (for a review, see Rost et al., 2017). Optogenetic switches consist of a photoreceptor that undergoes a conformational change upon stimulation by a particular wavelength of light, leading to interaction with specific partners via homodimerisation or heterodimerisation with bait and prey tags. Well-characterised optogenetic systems include cryptochrome 2 (CRY2)–CIBN, iLID, and enhanced Magnets (Benedetti et al., 2020; Duan et al., 2015; Guntas et al., 2015).

We have previously utilised the CRY2 (prey)–CIBN (bait) system to modulate endocytic recycling of the T cell receptor after T cell activation (Redpath et al., 2019). When coupled with bait-labelled kinesin motors, it is possible to redistribute prey-labelled organelles to the plasma membrane with inducible anterograde transport (Duan et al., 2015; Nijenhuis et al., 2020). We reasoned that we could utilise the CRY2–CIBN system coupled with anterograde motor transport to redistribute LAMP1-labelled endolysosomes to the cell periphery using the bait-labelled kinesin motor, KIF5a (schematically depicted in Fig. 5A).

We first characterised LAMP1 redistribution with LAMP1–mCherry–CRY2 and KIF5a–GFP–CIBN expression in live and fixed cells. We observed extensive anterograde transport of LAMP1 to the plasma membrane following light stimulation in cells by fluorescence microscopy (Fig. 5B–E; Movie 8). Quantitation revealed a significant proportional redistribution of LAMP1 signal to the peripheral areas of the cell cytoplasm after 3 min of blue light stimulation (Fig. 5F). We then confirmed that LAMP1–mCherry–CRY2 was redistributed to the cell perimeter upon stimulation with blue light prior to plunge freezing by imaging using cryo-fLM and Airyscan 2 detection (Fig. 5G,H). As an additional characterisation, we assessed horseradish peroxidase (HRP)-labelled endolysosomal redistribution using conventional electron microscopy after HRP uptake and light stimulation. PtK₂ cells transfected with both LAMP1–mCherry–CRY2 and KIF5a–GFP–CIBN were subjected to 30 min HRP uptake at 37°C prior to 30 min chase at 37°C to ensure HRP incorporation into LAMP1-positive structures. Cells were stimulated with light and then fixed, and cells of interest were re-registered, as performed previously (Ariotti et al., 2017, 2018). Unstimulated doubly transfected cells showed clustering of HRP-labelled endolysosomes limited to perinuclear areas with minimal association with cytoskeletal elements (Fig. 6A–C). Stimulated cells showed a dramatic association between clustered, HRP-labelled endolysosomes and cytoskeletal elements, as well as HRP-decorated vesicles highly enriched at the plasma membrane (Fig. 6D–F). These data confirm that CRY2–CIBN can be used to reposition LAMP1-positive organelles to the cell periphery. We also confirmed that this method could be used to redistribute Rab11a-positive compartments (Fig. S6; Movie 9).

Finally, we performed OPO–cryo-CLEM and analysed the morphology of re-positioned LAMP1-positive organelles in doubly transfected cells. We observed LAMP1 in multiple locations at the cell periphery (Fig. 7A,B). Three ROIs were selected for batch cryo-ET acquisition (Fig. 7C) and re-registered with our cryo-CLEM methodology; each LAMP1–mCherry–CRY2 structure localised at the cell periphery correlated with an electron-dense vesicular structure at low magnification (Fig. 7D,E). Cryo-tilt series were acquired at ROIs 1–3, reconstructed and overlaid with the cryo-fLM z-stacks. Endolysosomes were readily discernible at each mCherry-labelled vesicle (Fig. 7F–H′; Movies 10, 11 and 12), and these morphologies were consistent with those observed in other recent studies looking at lysosomal structure by cryo-FIB milling and cryo-ET (Cai et al., 2022; Riera-Tur et al., 2022). It is also possible that lipofectamine transfection can result in endolysosomal structures with similar morphologies (Fig. 7G; ROI2) (Pemberton et al., 2024 preprint). To further demonstrate the utility of this method, we redistributed Rab11a-positive recycling endosomes from perinuclear regions to the cell periphery by co-expression of FuGeneRed–Rab11a–CIBN, KIF5a–GFP–CIBN and CRY2-cluster–mCerulean. As with LAMP1, light stimulation resulted in significant redistribution of Rab11a to the cell periphery (Fig. S6), highlighting that OPO–cryo-CLEM is applicable to a diverse set of organelles. These data demonstrate that OPO–cryo-CLEM can be used to redistribute perinuclear-localised organelles to the cell periphery in PtK₂ cells, facilitating cryo-ET without the requirement for cryo-FIB milling and cryo-lamella preparation.

In situ cryo-ET provides the cellular context to structural analyses that is lacking in traditional structural biology-based methods. These assays represent the next step towards the full exploration of the cellular environment at near-atomic scale. To date, the field has been limited in its exploration of the cellular cryo-environment by a combination of factors including the restricted throughput of the techniques used, hardware constraints, and the expensive and challenging operational requirements of interconnected workflows between multiple high-end microscopes operated at cryogenic temperatures. Here, we have developed a robust PtK₂ cell-based system that facilitates structural in situ analyses, from live cells through to batch cryo-tilt series data acquisitions, within 24–36 h. By applying OPO in PtK₂ cells, we show that this system is not just restricted to in situ analysis of peripheral structures but can be also used for perinuclear organelles like endolysosomes. The application of this system will allow for higher-throughput examinations of diverse cellular structures and macromolecular complexes in situ, and can be deployed at a fraction of the cost of the current standard in situ workflow.

This method has the advantage of sidestepping some of the most significant drawbacks associated with current in situ cryo-ET studies. Firstly, our method avoids use of cryo-FIB milling for thinning samples, which represents the largest time bottleneck for the wet lab component of the cryo-ET workflow (Bauerlein and Baumeister, 2021). Here, we developed a technique that avoids the need for cryo-lamella preparation as an intermediate step by utilising the ∼200 nm thickness of the periphery of PtK₂ cells. Developing a protocol that avoids cryo-FIB milling has the flow-on effect of avoiding other limitations associated with the existing methods, including the attrition rate for brittle cryo-lamellae, the problems associated with preferential ice contamination of processed cryo-lamellae and the need for additional manual handling steps between microscopy modalities (Bauerlein and Baumeister, 2021; Wagner et al., 2017). The throughput of this system is further improved by exploiting cryo-correlation between the cryo-confocal microscopy and cryo-ET such that ROIs are specifically targeted, bypassing off-target imaging and image processing. Secondly, this method results in minimal ice contamination of our grids (see Fig. 7; Fig. S4 and S5). By accelerating the workflow such that all grid transfers are performed in under 12 h (plunge freezer, autogrid clipping, cryo-confocal imaging and the cryo-TEM autogrid loading), we could achieve high-quality grids using only liquid nitrogen transfers without the requirement for high- or ultrahigh-vacuum transfer systems. Finally, the current gold standard in situ cryo-ET workflow requires numerous expensive and highly specialised pieces of equipment, including a plunge freezer, a cryo-transfer system, a cryo-fLM-capable system (either as a standalone system or incorporated into a cryo-FIB setup), a cryo-FIB system and a cryo-transmission electron microscope, as well as the highly skilled staff to operate each instrument. This workflow, however, can be implemented within a facility at low cost, especially if the cryo-fLM stage can be used on an existing upright widefield fluorescence or confocal microscope.

The thickness of the PtK₂ cell cytoplasm provides multiple advantages for cryo-EM analysis. PtK₂ cells possess a protracted and thin cytoplasm that provides ample electron-lucent areas for in situ structural analyses and is densely packed with organelles, and they represent a cell type that is highly amenable to genetic modification and expression of fluorescently labelled proteins of interest. Although we have characterised this PtK₂ cell system as an ideal line for in situ cryo-CLEM analyses, we believe that this technique can be applied to other cell types with similar morphological properties. In an initial screen, we showed that MDCK, HeLa and BHK cells are too thick for high-quality cryo-ET imaging. We have previously performed cryo-CLEM on focal adhesions in MEFs (Cagigas et al., 2021), and although sufficiently thin at the cell periphery, MEFs do not possess the ideal extended ‘fried egg’ morphology and as such are only appropriate for analyses directly at the cell perimeter. However, U-2OS and WI-38 cells demonstrated an ultrastructural organisation similar to PtK₂ cells. We suspect that other cells with similar morphology – for example, primary lymphatic endothelial cells (Kriehuber et al., 2001) – would prove an appropriate system for in situ cryo-CLEM and/or OPO–cryo-CLEM analyses if cultured at sub-confluency on grids.

Cells grown on Quantifoil R2/2 gold support carbon film grids preferentially associate with the grid bar (Toro-Nahuelpan et al., 2020). Photo-micropatterning has been developed to circumvent preferential cell adhesion through selective printing of extracellular matrix into the centre of a grid square to direct cell growth to regions better suited for cryo-ET (Toro-Nahuelpan et al., 2020). Here, we sought to provide a characterisation of the cell–grid interface to determine why cells preferentially associate with the grid bar. Our studies demonstrated regional differences in substratum compliance, with the centre of the grid square as the most compliant region of the film and the carbon film over the gold grid bar, as well as the gold grid itself, being orders of magnitude greater in substrate stiffness (Fig. 3). These stiffer areas also demonstrated an increased total number and maturity of focal adhesions, increased phalloidin intensity and greater stress fibre formation (Fig. 4). This indicates that substrate stiffness may represent a tuneable parameter to modify how cells associate with the Quantifoil carbon substratum. As such, we altered carbon film thickness through deposition of additional amorphous carbon and observed that a 12.5 nm-thick carbon layer resulted in an ∼8-fold increase in substrate stiffness that translated into increased focal adhesion formation, increased focal adhesion maturity, increased phalloidin intensity and increased stress fibre formation. We demonstrated that an increase in substrate stiffness altered the position of the nucleus by comparing cells seeded on control grids to cells on +Carbon grids. We also observed that cells on stiffer substrates had an increased 2D cell area. Although not directly tested here, this indicates that additional carbon deposition might represent a low-cost easy option to increase the yield of suitable areas for cryo-ET. Our studies also show that additional carbon deposition can be used experimentally to modify substrate stiffness, allowing investigation of in situ structural changes in mechanosensitive and mechanoresponsive macromolecular assemblies.

In the OPO–cryo-CLEM system we utilised optogenetics to position organelles in the periphery of the cell, making them amenable to cryo-ET. The system could conceivably be modified to redistribute and structurally characterise non-membranous structures or protein complexes. Given previous work using similar optogenetic approaches to increase the clustering of specific structures (Vettkotter et al., 2022), this system could also be used to enhance local concentration and/or clustering of structures of interest per field in in situ cryo-tomograms. This has the potential to increase yield, reduce microscopy time or improve attainable resolution through increased sampling of structures of interest. Although several optogenetic systems have been used to stimulate redistribution of organelles and proteins of interest to specific subcellular compartments (Benedetti et al., 2020; Bugaj et al., 2013; Guntas et al., 2015; Kawano et al., 2015; Kennedy et al., 2010; Lerner et al., 2018; Levskaya et al., 2005, 2009; Swartz et al., 2001), the temporal properties of the CRY2–CIBN system proved favourable for our purpose. The CRY2–CIBN interaction is transient and reversible, with interaction occurring within seconds and reversion taking minutes to hours (Duan et al., 2015), as compared to other systems that can reverse on a seconds to minutes timescale (Benedetti et al., 2020). The slower reversion (off rate) of the CRY2–CIBN system ensured ample time between stimulation and plunge freezing to minimise redistribution of organelles back to the perinuclear space.

As with any new method, certain caveats must be considered before its application. Firstly, the choice of cell system is very important. Thicker cells will not allow sufficient electron penetration without cryo-FIB milling. The size of the organelle of interest also needs to be carefully considered, as thicker organelles will result in lower quality cryo-ET imaging. Here we focused on the redistribution of endolysosomes, which are large organelles that are at the upper limit of what is possible – imaging of larger lysosomes (above 300 nm diameter) using cryo-EM may not be feasible due to the thickness of these structures. Thicker samples can be targeted using scanning transmission tomography techniques such as dual-axis cryo-scanning transmission electron tomography, which can enable imaging of cells and organelles up to 1 μm in thickness (Kirchweger et al., 2023). To avoid issues with electron transparency, we focused our analyses specifically on the smaller endolysosomes recruited to the cell periphery after optogenetic stimulation. This was less of a problem with OPO–cryo-CLEM, as the larger endolysosomes were not actively transported to perinuclear regions, unlike the smaller endolysosomes (see Fig. 5).

It is important to emphasise that for some applications the disruption of the steady-state distribution of an organelle or protein complex during optogenetic repositioning could be disruptive, especially to transient assemblies and highly dynamic structures (for example, sorting tubules) and could even affect cellular processes (Hoepfner et al., 2005). Multiple studies have examined the functional consequences of redirecting organelles to different cell regions. Upon chemogenic redirection to the cell periphery over a 5–30 min timeframe, β2 adrenergic receptors remain in an active signalling state and capable of interaction with recycling mediators such as SNX27 (Willette et al., 2024). Redirection of Rab11a by optogenetic coupling to dynein results in collapse of growth cones, whereas coupling to kinesin results in enhancement of growth cones in neurons following 4–8 min of blue light exposure (van Bergeijk et al., 2015). Lysosome redirection into neuronal synapses via optogenetic coupling with KIF5a using 10 min of blue light exposure results in persistent changes to neuronal long-term potentiation amplitudes (Chen et al., 2023). Using optogenetics to force interactions between lysosomes and mitochondria results in induction of mitochondrial fission within 140 s of blue light exposure (Qiu et al., 2022). These studies highlight that organelles can remain functional following redirection; however, they also reveal that within 2 min following redirection, organelles can change the function of the surrounding environment. It is likely that within this timeframe these organelles may be changed themselves, no longer completely reflecting the structure or function of their original native environment. An advantage of optogenetics is the speed at which redirection can occur. Our redirection of LAMP1- and Rab11a-positive endosomes to the cell periphery occurred within 3 min, following which the samples were snap frozen, suggesting that this technique is well suited for analyses of cargo within a membrane-enclosed structure. By minimising the time taken to redirect an organelle, the functional changes to the organelle and the environment surrounding it can be minimised to ensure that the organelle better reflects its native state. To enhance the possibility of preserving the native state of organelles, an alternate optogenetics system such as iLID, which redirects Rab5-positive endosomes to the cell periphery within 25 s of blue light exposure (Nijenhuis et al., 2020), could be explored. Despite these caveats, the use of OPO–cryo-CLEM has the potential to open a new era of structural characterisation of a wide range of protein and membrane assemblies inside native cells.

Carbon coating was performed on a SafeMatic CCU-10 carbon coater fitted with a CT-010 Carbon Thread evaporation module. Quantifoil R2/2 gold support carbon film grids (Q2100AR2; Electron Microscopy Sciences) were loaded into the vacuum chamber and carbon was deposited under high vacuum with stage rotation. Carbon deposition was automatically detected by the Quartz film monitor sensor and deposition was stopped with the thickness monitor reached 12.5 nm.

The +Carbon and control Quantifoil R2/2 gold support carbon film grids were mounted on a polydimethylsiloxane (PDMS) substrate for analysis in a Bruker Dimension ICON SPM equipped with a Nanoscope V controller. The grid was adhered to the PDMS with the carbon support film facing the probe and allowed to settle on the PDMS for at least 2 days to avoid drifting during AFM measurements. Surface topology and the DMTmodulus were acquired using Peakforce QNM mode, in 1 µm² areas in duplicate at three regions of the grid square using OTESPA-R3 tips (from Bruker AFM tips). Region 1 in the centre of the grid square on the carbon film was selected at the intersection of the two diagonals (Fig. 3A). Region 2 on the carbon film over the gold support bar was selected by drawing a parallel and at the intersection of the line with the grid bar, measurements were taken on either the left or right side. Region 3 in the Quantifoil hole over the gold grid bar (1 μm² areas are represented in Fig. 3A) was chosen in the Quantifoil hole near R2. Average stiffness measurements were generated from each point within the 1 µm² area corresponding to 256 positions in x and y. Significance was determined by unpaired two-tailed Student t-tests. The AFM images were processed with Gwyddion (version 2.59) software.

Low-passage PtK₂ cells (male potoroo kidney epithelial cells; ATCC) were cultured in Eagle's Minimum Essential Medium (MEM; Thermo Fisher Scientific; Table S1) supplemented with 10% foetal bovine serum (Gibco) and MEM Vitamin Solution (Sigma Aldrich). Cells were trypsinised (0.05% Trypsin-EDTA; Sigma Aldrich) and seeded onto control and +Carbon grids as described in Fig. 1B. Briefly, grids were placed in a 35 mm plastic tissue culture dish (Nunclon) and subjected to ultraviolet light for 20 min to ensure sterility. A 40 µl droplet of cells resuspended in MEM was placed over each grid with the carbon side facing up for 2–3 h prior to the gentle addition of 2 ml MEM. Cells were transfected with Lipofectamine 3000 (Invitrogen) as per the manufacturer's instructions 24 h after cell seeding. The plasmids used in this study were as follows: EYFP–Mito7 (Addgene plasmid 56596, deposited by Michael Davidson; http://n2t.net/addgene:56596; RRID:Addgene_56596), EGFP–DCX (Addgene plasmid 32852, deposited by Joseph Gleeson; http://n2t.net/addgene:32852; RRID:Addgene_32852), KIF5A–GFP–CIBN (Addgene plasmid 102252, deposited by Bianxiao Cui; http://n2t.net/addgene:102252; RRID:Addgene_102252), LAMP1–mCherry–CRY2 (Addgene plasmid 102249, deposited by Bianxiao Cui; http://n2t.net/addgene:102249; RRID:Addgene_102249), GFP–Rab5a (Flores-Rodriguez et al., 2011), GFP–Rab4a (Addgene plasmid 56440, deposted by Michael Davidson; http://n2t.net/addgene:56440), FuGeneRed–CIBN–Rab11a (Redpath et al., 2019) and CRY2-cluster–mCerulean (Redpath et al., 2019). HeLa, BHK, MEF, MDCK and WI-38 cells were passaged as described above but grown in DMEM (Thermo Fisher Scientific) supplemented with 10% foetal bovine serum (Gibco) and L-glutamine (Invitrogen). U-2OS cells were grown in McCoy's 5a medium (Thermo Fisher Scientific) supplemented with 10% foetal bovine serum. Further details of cell lines can be found in Table S1.

Sample preparation was performed as described previously (Herms et al., 2013). Briefly, cells were seeded onto Quantifoil R2/2 gold support carbon film grids as described above, fixed in 2.5% glutaraldehyde in in phosphate-buffered saline (PBS; pH 7.4), washed three times in PBS and postfixed in 2% OsO₄ with 1.5% potassium ferricyanide. Samples were then washed in double-distilled H₂O (ddH₂O), incubated in a 1% (w/v) thiocarbohydrazide solution and postfixed again in 2% OsO₄. Cells were washed (ddH₂O) and stained en bloc with 1% uranyl acetate and subsequently with lead aspartate solution (20 mM lead nitrate, 30 mM aspartic acid, pH 5.5). Grids were serially dehydrated in ethanol, serially infiltrated with Durcupan resin and polymerised for 48 h. SBF-SEM imaging was performed as described previously (Rae et al., 2021). Image stacks were assembled and aligned in IMOD (https://bio3d.colorado.edu/imod/). Segmentation was performed with Isosurface rendering in IMOD.

Cells were processed as described above, then Durcopan-embedded blocks were subjected to ultramicrotomy. Vertical ultrathin sections (60 nm) were cut on a UC6 Leica Ultramicrotome and mounted on copper 200 mesh grids (Electron Microscopy Sciences). Grids were imaged on a JEOL JEM-1011 transmission electron microscope fitted with an EMSiS Morada camera.

Cells were seeded onto 35 mm gridded in-plane MatTek dishes. Cells were transfected with LAMP1–mCherry–CRY2 and KIF5a–GFP–CIBN, left for 48 h after transfection, and then pulsed with 5 mg/ml of HRP (Sigma-Aldrich; Table S1) in MEM medium for 30 min at 37°C. The uptake medium was replaced with normal MEM growth medium, and the cells were chased for a further 30 min at 37°C to ensure HRP incorporation into lysosomes. PtK₂ cells were stimulated with blue light for 3 min (control cells were left unstimulated in MEM medium) and fixed in 4% paraformaldehyde (PFA) with 0.1% glutaraldehyde in PBS for 30 min. Cells were washed in PBS and imaged on a Nikon Ti inverted fluorescence microscope with a 63× (NA 1.4) objective. Cells positive by fluorescence for KIF5a and LAMP1 were imaged, and their alphanumeric code recorded by brightfield microscopy. Dishes were then processed using the DAB reaction as described previously (Ariotti et al., 2015), and correlation was performed as described previously (Ariotti et al., 2018).

Cells were seeded on control or +Carbon grids, as described above, for Fig. 4 and Fig. S1, and were seeded onto coverslips for Fig. 5. Cells were fixed with 4% PFA, washed in PBS, permeabilised with 0.1% Triton X-100 and stained with phalloidin–iFluor 488 (ABCAM; Table S1) or DAPI (Thermo Fisher Scientific) as per the manufacturer's instructions. For mounting, grids were inverted onto coverslips with Mowiol (Sigma-Aldrich), and coverslips were then inverted onto slides such that cells were facing toward the coverslip and no fluorescence signal could be obscured by the gold grid bars. For nuclei positioning, mounted grids were imaged on a Nikon Eclipse Ti inverted fluorescence microscope with a 63× (NA 1.4) objective. Grid squares were then divided into a 12×12 matrix of equal size (10 µm×10 µm). The centre of mass of the nucleus was determined using ImageJ and the coordinates of all centres of mass were recorded and overlaid on the matrix. The 144 positions were rotationally averaged into four quadrants of 36 individual coordinates. These positions were then analysed by five groupings relative to the number of matrix positions away from the grid bar. Total phalloidin fluorescence intensity was normalised to background and analysed per cell.

Cells grown on control or +Carbon were fixed in 4% PFA, washed in PBS, permeabilised in 0.5% Triton X-100, washed in PBS, quenched in 50 mM NH₄Cl in PBS for 10 min and blocked in 1% bovine serum albumin in PBS. Grids were incubated with mouse anti-paxillin antibody (Thermo Fisher Scientific, AHO0492) at a dilution of 1:200 in the blocking solution for 30 min, washed in blocking solution and probed with a goat anti-mouse IgG Alexa Fluor 647-conjugated secondary antibody (ABCAM, AB150079). Grids were washed with PBS and stained and mounted as described above. Confocal z-stacks were acquired on a Nikon C2-Si upright confocal with a 63× (NA 1.4) objective.

Stimulation was performed for 3 min with 405 nm laser excitation on either an inverted Nikon Eclipse Ti with a mercury lamp and a 63× (NA 1.4) objective, or an Olympus CKX53 fitted with a CoolLED pE-300^white LED fluorescence module and a 20× (NA 0.4) objective, as described previously (Duan et al., 2015).

PtK₂ cells were plated on 13 mm coverslips and transfected with LAMP1–mCherry–CRY2 and KIF5A–GFP–CIBN or with FuGeneRed–CIBN–Rab11a, KIF5A–GFP–CIBN and Cry2-cluster–mCerulean using Lipofectamine 3000 overnight prior to imaging. Samples were transferred into fresh MEM containing 10% serum and MEM Vitamin Solution and without Phenol Red 1 h prior to imaging. Imaging was performed at 37°C. Simultaneous imaging and optogenetic stimulation was performed with HILO on a Zeiss ELYRA P1 microscope during widefield imaging with 200 mW 405 nm, 200 mW 488 nm and 200 mW 561 nm laser (LAMP1) or a Zeiss 780 LSM with an Argon ion 488 nm and diode 561 nm laser. Samples were imaged for up to 1000 frames (ELYRA; a 63×, NA 1.4 objective) or 100 frames (780 LSM; 100× 1.4 NA oil objective). Quantification of redistribution of endolysosomes was performed as described previously (Duan et al., 2015). Briefly, a central mask of 30 µm was applied to the centre of mass of the nucleus. LAMP1-positive puncta within the mask were proportionally assigned to a central bin and beyond the mask a peripheral bin. Quantification is expressed as a percentage per cell.

Grids were prepared as described above. Plunge freezing was performed on a Leica EM GP with back-sided blotting. The chamber temperature was set to 37°C, humidity was 95% and blot time was varied between 4 s and 6 s with a 0 s post-blot time. Grids were plunged into liquified ethane at −182°C. Grids were transferred to liquid nitrogen for autogrid clipping and were clipped as per the manufacturer's instructions (Thermo Fisher Scientific). Blue light stimulation prior to plunge freezing was performed on an Olympus CKX53 optical microscope fitted with a CoolLED pE-300^white LED fluorescence module. This optical microscope was set up beside the Leica EM GP. Blue light stimulation was performed for 2.5 min and grids were transferred onto the Leica EM GP for immediate freezing. Grids were cryo-preserved within 3–3.5 min of starting light stimulation.

Clipped autogrids were loaded in the Linkam CMS196 cryostage at −196°C and transferred onto the liquid nitrogen cooled bridge. The cryo-stage was mounted onto a Zeiss LSM 900 upright confocal fitted with 10× EC Plan-Neofluar (0.3NA; Zeiss) and 100× LD EC Epiplan-Neofluar (0.75NA; Zeiss) objectives and an Airyscan 2 detector. Whole-grid mapping was performed as follows; the 10× objective at a 0.5 digital zoom was used to image the whole grid with T-PMT transmitted light and fluorescence channels, as this objective and digital zoom allows for fast atlas acquisition with only four images required to image the whole grid. Focal points were set for each grid quadrant using Advanced setup in Zen Blue 3.2 and acquired with preview scan to generate an interactive map of coordinates, and subsequently the whole grid was acquired as a tile set and stitched in Zen Blue. ROIs/grid squares of interest were selected, and the z-stacks were acquired at each ROI using the 100× objective at 0.5 digital zoom with Airyscan detection of both fluorescence and T-PMT channels to image the whole grid square. Higher-magnification z-stacks were acquired at a 1.6× digital zoom using the 100× objective with Airyscan and processed for Airyscan deconvolution. Excitation lasers were 488 nm (GFP/YFP) and 561 nm (mCherry) with pixel dwell time 1.31 µs, and a z-slice optical section thickness between 100 nm and 500 nm. Pixel arrays were 1586×1586 pixels (125.88 µm×125.88 µm) at 50× and 2024×2024 pixels (39.46 µm×39.46 µm) at 160×. Up to 10 ROIs were imaged per grid. Z-stacks required between 30 min to 2 h to acquire all ROIs per grid. We observed that ice contamination increased dramatically when the Linkam CMS196 cryo-stage was used with manually filling. The use of a Linkam Autofill Dewar resulted in minimal ice contamination. Grids were transferred to liquid nitrogen for cryo-EM.

Clipped autogrids were loaded into a 200kV Talos Arctica Cryo-TEM microscope (Thermo Fisher Scientific). Whole-grid batch atlases were acquired to relocate the grid squares of interest. Orientation and rotation were established using broken grid squares, the position of cells and the central grid marker as fiducial markers at low magnification. Atlases were exported to Adobe Photoshop and overlaid with the corresponding low-magnification cryo-fLM map to establish grid squares of interest. Whole-grid-square ROIs were imaged using the Overview magnification at 700× in the Tomography software (Thermo Fisher Scientific), and this data was exported in real time to Adobe Photoshop for intermediate-magnification alignment with maximum-intensity z-projected cryo-fLM z-stacks. Correlation was performed using cellular landmarks and carbon film holes as markers in space (Fig. S4C–D′,I–J′). Aberrations between imaging modalities were corrected using the Adobe Photoshop Puppet Warping tool. The aligned z-projected fluorescence image overlaid on the 700× Overview cryo-EM image was sufficient to determine the positioning of fluorescently labelled structures and electron transparency: this manual alignment method allowed for rapid re-registration at sub-hole resolution to determine holes of interest. Batch cryo-tilt series were acquired with dose-symmetry in the Tomography software. Single-axis tilts were collected from −60° to +60° at 2° increments at 7 µm defocus on a Falcon 3 camera (Thermo Fisher Scientific) operated in linear mode at 22,000× magnification. Total electron dose was kept below 50 e⁻Å⁻² to minimise radiation damage. Tilt increments were aligned through a combination of fiducial tracking of 10 nm gold particles and Patch Tracking in Etomo (University of Colorado; https://bio3d.colorado.edu/imod/). Tilt series were reconstructed using weighted back-projection and filtered for visualisation with SIRT-like filters and nonlinear anisotropic filtering in IMOD. Magnification scaling at the individual hole level was utilised to correlate cryo-fLM with the reconstructed cryo-tomogram.

The authors acknowledge use of the Cryo-Electron Microscopy Facility through the Victor Chang Cardiac Research Institute Innovation Centre, funded by the NSW government, and the Electron Microscope Unit within the Mark Wainwright Analytical Centre (MWAC) at UNSW Sydney; the facilities, and the scientific and technical assistance, of Microscopy Australia and the Centre for Microscopy and Microanalysis, The University of Queensland; and the Australian Cancer Research Foundation (ACRF)/Institute for Molecular Bioscience Cancer Biology Imaging Facility, which was established with the support of the ACRF.

The peer review history is available online at https://journals.biologists.com/jcs/lookup/doi/10.1242/jcs.262163.reviewer-comments.pdf