Accelerate and deepen our insights into infectious diseases through large-scale 2D and volume electron microscopy combined with artificial intelligence

Abstract text (incl. figure legends and references)

The emergence of new viral variants, which can cause human diseases and pandemics, necessitates a quicker and deeper understanding of the pathogenesis and disease manifestations to develop effective treatments and preventative measures. Viral particles are in the nm range thus their observation requires the nm-resolutions of electron microscopes (EM). EM has been paramount in diagnosing Covid-19 [1] and gaining insights into its disease-causing mechanisms [2]. However, the time-consuming nature and manual operation requirement of conventional EM limit the data throughput, limiting the statistics and understanding over a timespan. Multibeam scanning electron microscope (SEM) technology such as FAST-EM (Delmic B.V., Netherlands) can overcome these bottlenecks. In this work, we apply FAST-EM whole-section imaging to human airway epithelial (HAE) cells 72 hours post-SARS-CoV-2 infection to demonstrate how more profound insights into infectious diseases can be gained faster.

Using FAST-EM, we imaged a whole section of a HAE cell layer spanning an area of 270 x 270 µm2 at a 4 nm pixelsize and at 10 µs/px within only 35 minutes. The same area would take 3.5 hours to image using a single beam SEM with the same pixel size and at 2 µs/px. The EM images revealed the density and the locations of virus-like particles in the HAE cells. At 4 nm resolution hallmarks of SARS-CoV-2 such as electron-dense particles and multilamellar vesicles could be observed (Figure 1).

To perform an accurate examination of the sub-cellular changes caused by different SARS-CoV-2 variants, we analyse the density of the virus-like particles as well as the features that span multiple sections (e.g. multilamellar vesicles, mitochondria, cilia) by performing array tomography and 3D data reconstruction. We employ the available auto-segmentation model empanada MitoNet [3] (Figure 2) and train new segmentation models to extract the rich information present in the FAST-EM data to facilitate downstream quantitative analyses. Our work shows the FAST-EM and artificial intelligence approach as a new and effective way of gaining insights into infectious diseases.

Figure 1 Whole section FAST-EM image of HAE cell layer infected with SARS-CoV-2 enables the study of morphological changes with machine learning. (a) The morphological changes can be studied in the context of the cell layers, 270 x 45 µm², by allowing the identification of different cell types. (b dashed squares) On the cellular level, the impact of the infection is visible once zoomed in. (c, solid squares) The individual mitochondria in the cell layer are identified by employing MitoNet, allowing for quantification of morphological changes in the mitochondria. The scale bar is 20 µm (a) and 5 µm (b-c).