At the Center for Molecular Microscopy, our primary technique is focused ion beam scanning electron microscopy (FIB-SEM, otherwise known as ion abrasion scanning electron microscopy, or IA-SEM), often in conjunction with correlative light microscopy, to image large samples in 3D and at nanoscale resolutions. More recently we have deployed array tomography (AT) to large volume samples.
At the CMM we have acquired a Zeiss Xbeam 550 FIB-SEM and a Zeiss Gemini 450 SEM for volume EM projects. For correlative microscopy we have a Zeiss LSM upright microscope equipped with an AiryScan detector. We also have a full suite of EM preparation equipment including a Leica high-pressure freezer and freeze-substitution device and an RMC Boeckeler ATUMtome. We have access to TEM microscopes, which we share with the EML core laboratory.
We have developed protocols to render challenging samples amenable to volume EM imaging. Most recently, we have adapted high-pressure freezing (HPF) and quick freeze substitution (QFS) protocols to trap architectural intermediates in the rapidly developing C. elegans embryo (Rahman et al 2020).
We have standardized protocols that allow for easy correlative imaging (LM + FIB-SEM) of adherent cells (Murphy, et al. 2014) and continue to work on methods that will allow more efficient FIB-SEM imaging of adherent cells.
Efficient volume EM
While several groups have utilized “brute force” volume EM imaging of massive chunks of tissue (typically for neuroanatomy or connectomics projects) with great success, we have focused on rapidly and efficiently imaging targeted features of interest at high resolution, while also occasionally sampling the larger cellular context. The strategy of high-resolution “ROI imaging” and intermediate resolution “keyframe imaging”, developed in collaboration with Zeiss Inc and Fibics Inc, is now a central feature of FIB-SEM tomography acquisition software ATLAS3D (Narayan, et al., 2014).
Processing and Visualization
We utilize in-house scripts to register and process acquired image stacks in order to generate isotropic image volumes; these are executed at the NIH Biowulf cluster.
Segmentation – the extraction and visualization of specific features from an information-rich volume EM dataset – continues to be a challenge outside of neuronal tracing successes in the field. While we still rely on semi-automated and manual segmentation using commercial or open-source platforms, we are actively pursuing machine learning and deep learning based segmentation and visualization approaches, with the aim to quantify various architectural parameters. We have previously used a simple DL based pipeline to segment features from a FIB-SEM image volume of sporulating bacteria (Cheng et al 2018).