High-Throughput Low-Dose Biomolecule Imaging in Liquid Phase Electron Microscopy

Liquid Phase Electron Microscopy (LPEM) has demonstrated high-resolution structural information comparable to cryo-EM [1] and also yields dynamic information. Imaging biomolecules in the liquid phase depends on the encapsulation method and electron dose. This work aims to tackle these key areas to advance the use of LPEM to study biomolecular structure and dynamics. Liquid cells (LCs), typically carbon (Fig. 2B) or silicon nitride (Fig. 2C) based, are used to enclose liquids and protect samples from the vacuum environment inside electron microscopes. Carbon cells include graphene and amorphous carbon cells, which both encapsulate samples into inhomogeneous liquid pockets and can wrinkle during preparation, reducing the effective viewing area [2]. SiNx LCs are mass-produced via nanofabrication processes, are more robust, and have liquid throughout the contents of the window, but yield lower resolution than carbon-based LCs. Resolution in SiNx LCs was primarily limited by window bulging, so we developed a bulge-free SiNx LC system that has addressed this issue and eliminated bulging. The resolution in our LCs is now only limited by the thickness of the SiNx window membranes (⁠≈25 nm each). This work demonstrates the successful nanofabrication of 5-nm-thin SiNx windows measured with Electron Energy Loss Spectroscopy (EELS) and presents the next steps required to image biological systems at high magnification. EELS results demonstrate a total thickness of 9.5 nm for two windows, using the calculated inelastic mean free path of Si3N4 (⁠λIMFP⁠) of 123.4 nm [3], and measured t/λIMFP of 0.077 (Fig. 1). Electron transmission estimations show that 10 nm SiNx approaches the transmission of carbon-based LCs at high beam energies (Fig. 2A) [2, 4]. Thus, thin SiNx LCs have the potential to confer resolution similar to carbon LCs, while supplying uniform liquid layer thickness (t) across the viewing area, and high throughput. This opens vast opportunities for high-throughput imaging, notably, the study of critical emerging diseases. The next step in high-throughput biomolecule imaging is addressing sample degradation caused by beam-induced radiolysis [5]. One solution is to reduce the dose by sparse sampling and reconstructing images with inpainting [6]. Sparse scans can be achieved through scan control in Scanning Transmission Electron Microscopy, and inpainting can be performed using an algorithmic approach [7], dictionary learning [6], or deep learning [8]. In this study, images of dioleoyl-phosphatidylcholine (DOPC) liposomes taken with our LPEM system were used to demonstrate inpainting. A random mask (Fig. 3B) and spiral mask [9] (Fig. 3E) were applied to simulate sparse sampling to remove 80% of the respective original data. Inpainting was performed using the Telea algorithm in the OpenCV library [7, 10]. High agreement between original (Fig. 3A&D) and inpainted (Fig. 3C&F) images validates this approach for acquiring low-dose, high-resolution data. Although still in its early stages, this research will prioritize the investigation of biological specimens using thin SiNx LC windows, alongside exploring the potential of combining this new technology with inpainting techniques [11]. EELS spectra of the thin SiNx liquid cell assembly. A) The full EELS spectrum; B) The same spectrum on an enlarged scale showing the core loss peak. Estimation of electron transmission through different liquid cells (LCs). A) Electron transmission estimation data for different LCs, calculated using previously described methods [2] and data [4]. B-C) Schematic for carbon LCs. D-E) Schematic for SiNx liquid cells. B and C show a general view, where the grey lines indicate the cross-sections shown in C and E. In E two black circles are the o-ring (omitted in D). Blue represents liquid, and the SiNx is yellow in D-E. Inpainting images of liposomes in liquid phase. A&D) Images of DOPC liposomes assembled in SiNx liquid cells with a 200 nm liquid spacer and 60 nm total SiNx windows. B&E) 20% of data remaining from A&D after applying random (B) or spiral (E) masks. C&F) Inpainted images from B&E respectively using the Telea algorithm [7, 10].

Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.