Developments in cryo-FIB Sample Preparation: Targeting in Cryo-Lift-Out Preparation of Tissues and Machine Learning Models for Fully Automated On-Grid Lamella Preparation

Cryo-electron tomography (cryo-ET) of cryo-focused ion beam (cryo-FIB) prepared lamellae has provided unprecedented insights into cellular ultrastructure and opened the door to in situ structural biology of large macromolecular complexes. This technique, however, still has a number of constraints: relatively low throughput, difficult targeting, and the need for user expertise in many workflow steps, especially when preparing samples by cryo-lift-out. Technological advances and a growing extent of automation are therefore essential to enable the routine application of cryo-FIB lamella preparation in the context of cell and developmental biology research [1]. We will be discussing our recent advances in gas injection system free cryo-FIB lift-out transfer [2–3] and how the targeting of specific biological events can be achieved in cryo-FIB lift-out using multi-modal imaging. In the pursuit of targeting specific events in tissue, the volume of the transferred material in lift-out can be increased compared to traditional lift-out methods. This allows for cryo-FIB scanning electron microscopy (SEM) volume imaging of the lift-out volume in order to identify and retain the feature of interest (e.g. specific organelles) in the final ∼200 nm thin lamella. Additionally, this volume imaging throughout the milling process provides contextual information that would otherwise get lost as the material is ablated (Figure 1 A-D). If, however, organelles poor in SEM imaging contrast are to be studied, in-chamber fluorescence light microscopes can aid in targeting structures during lift-out and final lamella preparation (Figure 1 E-G). Furthermore, we will discuss how these developments can also be leveraged to prepare multiple sections from a single cryo-lift-out transfer. In addition, we will demonstrate how we develop and deploy the open-source platform SerialFIB as a solution to streamline and automate workflows developed for various microscope manufacturers [4]. Finally, we will discuss how novel cryo-FIB instrumentation, namely plasma ion sources and automated sample exchange [5] as well as machine learning algorithms can be leveraged to enable fully automated cryo-FIB lamella preparation without the need for user input (Figure 2), paving the way for high-throughput visual proteomics of unicellular organisms. Targeting in cryo-lift-out using FIB-SEM volume imaging and in-chamber fluorescence microscopy. A: Cryo-lift-out of the target material from a bulk high-pressure frozen sample using a Kleindiek cryo-gripper. B: FIB image of the extracted material after volume imaging. The yellow square shows the region that was milled during volume imaging. C: SEM image of the lamella surface. Once the feature of interest is visible in SEM, the lamella can be prepared by removing a damage layer from SEM imaging and fine milling from the bottom. *: Nucleus, PM: plasma membrane of a follicle cell in the D. melanogaster egg chamber, MT: mitochondria, ER: endoplasmic reticulum, LD: lipid droplet. D: Segmentation of the images obtained during FIB-SEM volume imaging for end-pointed of the lamella preparation. E: Cryo-lift-out using the needle system. Asterisk denotes an adaptor copper block attached to the lift-out needle for stronger attachment of the target material during lift-out. F: After attachment to the receiver half-moon grid, the lift-out lamella is released from the needle by FIB milling a line pattern, indicated in yellow. G: Fluorescence overlay (Nup358 tagged with GFP [11]) onto the SEM imaging of the lamella prepared in F. The protein that is tagged localizes to the nuclear envelope, thus the lipid droplets visible in the SEM are excluded from the inside of the nucleus. Automated site selection in cryo-FIB lamella preparation of a C. reinhardtii grid. A: Original image as obtained during grid screening. B: U-Net segmentation of the image in A. C: Overlay of A and B. D: Highest scoring result from grid library template matching (yellow) overlayed onto the original image. Squares define potential lamella sites which will be checked for overlap with the grid bar segmentation and minimum distance to the edges of the image. Note that due to the limited training data set size, false positives are detected but discarded as they are not central enough in the image to be considered. E: FIB and F: SEM image of the fully automatically prepared lamella from C. reinhardtii cell.