Correlative single‐molecule localization microscopy and electron tomography reveals endosome nanoscale domains

Abstract Many cellular organelles, including endosomes, show compartmentalization into distinct functional domains, which, however, cannot be resolved by diffraction‐limited light microscopy. Single molecule localization microscopy (SMLM) offers nanoscale resolution but data interpretation is often inconclusive when the ultrastructural context is missing. Correlative light electron microscopy (CLEM) combining SMLM with electron microscopy (EM) enables correlation of functional subdomains of organelles in relation to their underlying ultrastructure at nanometer resolution. However, the specific demands for EM sample preparation and the requirements for fluorescent single‐molecule photo‐switching are opposed. Here, we developed a novel superCLEM workflow that combines triple‐color SMLM (dSTORM & PALM) and electron tomography using semi‐thin Tokuyasu thawed cryosections. We applied the superCLEM approach to directly visualize nanoscale compartmentalization of endosomes in HeLa cells. Internalized, fluorescently labeled Transferrin and EGF were resolved into morphologically distinct domains within the same endosome. We found that the small GTPase Rab5 is organized in nanodomains on the globular part of early endosomes. The simultaneous visualization of several proteins in functionally distinct endosomal sub‐compartments demonstrates the potential of superCLEM to link the ultrastructure of organelles with their molecular organization at nanoscale resolution.

only a small region at a time was in focus. Additionally, because this setting often produced a slight gap between the section and the glass, we could not always apply a full TIRF illumination, leading to reduced overall SMLM quality. Furthermore, the copper EM-grid and the active contents of the photo-switching buffer, probably mainly the thiols, led to a progressing change in pH-value and thiol concentration during acquisitions, leading to suboptimal conditions for SMLM after the first measurement. We tried to circumvent this issue by using gold EM-grids but observed similar effects, probably due to progressing adhesion of thiols to the extended gold surface of the grids.
Consequently, the SMLM quality, regarding photo-switching and single-molecule brightness, in superCLEM is slightly inferior compared to sections directly deposited on glass. For future endeavours the use of intrinsically blinking high-performance dyes could further improve the method.
Drying the back-side of a grid with the filter paper after SMLM imaging is probably the most delicate step of the whole protocol with great impact on the ultrastructural preservation. If the front side of a section is accidentally exposed to drying (in the absence of methylcellulose), its ultrastructure will be irreversibly destroyed. Additionally, the formvar film can get torn during this procedure. Especially during the early stages of the protocol development we lost a significant portion of samples around this step. We assure potential users that with some patience and practice this step can be performed with sufficient success rates, although a certain loss ratio will remain.
Due to intrinsic restrictions of the individual methods may cause that signals cannot fully or not certainly be assigned to an underlying ultrastructure. These restrictions include the finite angular range ('missing wedge') in electron tomography, shrinkage of sections after drying and the correlation of 2D fluorescence data with 3D electron tomography.
We experience the missing wedge as a problem especially for tubular structures, as it hard to observe their connections with the central vesicles due to the limited information that can be acquired by tomography.
On the other side, we could also observe tubular structures associated with the central vesicle of an endosome that did not have an obvious counterpart in the SMLM image (e.g. Supplementary Figure 7). Given the density of SMLM Tfn signal in tubules, these most likely represent other recycling tubules that contain different cargos or tubules that are destined for retrograde transport.
We observed a significant axial shrinkage of sections after drying, estimating a remaining 1/3 of their initial thickness. This allowed us to analyse sections that were originally thicker than the commonly used 300 nm, though it is likely that some distortions to the ultrastructure took place. Tokuyasu sections are so far not frequently used for electron tomography and the structural collapse along a z-axis is probably the main reasons. On the positive side, the membranes are well preserved and appear clearly white due to the negative staining. As pointed out in the main manuscript, the delicate structures of tubules originating from the globular part of endosomes are consistent with resin embedding, where no drying and the connected axial shrinkage can take place. So we are confident that shrinkage will lead to ultrastructural compression rather than major distortion, but users have to keep in mind that this may not be the case for alternative target structures.
Finally, the correlation of two-dimensional SMLM data to the three-dimensional EM tomograms is only possible under certain conditions. The usage of densely seeded fiducial markers in the imaging plane allows the mapping of SMLM data to the ultrastructure with nanometer precision laterally, but not axially, where signals can be assigned to certain likely structures. As illustrated in Suppl. Figure S5, the lateral overlay precision based on the fiducials seems limited to approximately 40 nm. This is true for the specific setting of this quantification, i.e. aligning a large field of view as illustrated in Suppl. Figure 8. We called this step the rough alignment in the main manuscript which is mainly used to pinpoint the endosome of interest with certainty. Therefore, the before stated 40 nm can be seen as an absolute lower bound for the overlay accuracy regarding the final alignment, which consists of a composite of the initial, fiducial based overlay and a subsequent manual refinement. Distinct structures like ILVs and singular tubules can be interpreted as intrinsic fiducial structures and allowed us to fine-tune the lateral and angular positioning of the SMLM signal, without changing the scale of the SMLM signal.
Concordantly, based on a priori knowledge about the proclivity of Tfn to reside in recycling tubules and EGF in ILVs we experience that after endosome segmentation particular cargo could be assigned to particular compartment with certain probability.
Adapting the workflow to three-dimensional SMLM will allow a complete correlation between imaging modi, but potential users should be aware of chromatic shifts and other artefacts that complicate multi-colour three-dimensional SMLM. Alternatively, one could use ultra-thin Tokuyasu sections to virtually eliminate the possibility of axial uncertainties. In the case of tubulo-vesicular morphology of endosomes, we found that 100 nm thick sections do not contain sufficient ultrastructural information to reliably reconstruct high-quality SMLM images of endosomes (especially tubules). In our estimate, the usage of thin sections has potential for compact organelles or more uniform distribution of the molecules of interest. Figure S1: Validation of the GFP-Rab5c BAC HeLa cells.

Supplementary Information
(A) Immunoblot of GFP-Rab5c cells with high GFP-Rab5c expression with an anti-Rab5c antibody. The GFP-tagged protein has an expected size of ~50 kDa. The percentage of each population was 30% relative to the total Rab5c amount. The GAPDH band was used as loading control.