About the Author

Three-Dimensional Particle Tracking in Cells

By Jillian Kunze

Jay Newby of the University of Alberta expressed his excitement at the prospect of correlating massive quantities of data with cell structures during his virtual minisymposium presentation at the 2020 SIAM Annual Meeting, which took place last week. While working as a postdoctoral researcher at the University of North Carolina at Chapel Hill, he collaborated with graduate student Grace McLaughlin on a project to track particles within cells. Their goal was to understand the organization of cytoplasm—the semifluid substance outside of a cell’s nucleus—across space and time. 

Newby and McLaughlin’s initial study focused on the random motion of the cytosol, the cytoplasm’s liquid component that lies outside of all cell organelles. Newby was particularly interested in the cytosol’s diffusivity, or the rapidity at which fluids could spread through it, as diffusivity is related to the poorly understood mechanism of asynchronous division of multiple nuclei within the same cell. Because the nuclei lie close together in syncytia, or multinucleated cells, the proteins that drive the cell cycle can travel through the cytosol between nuclei and synchronize them. Therefore, understanding the cytoplasm’s diffusivity is very important for understanding the causes of nuclei coupling. Newby and his colleagues concentrated on Brownian motion in the cytosol; they investigated the variance of increments because this parameter depends on fluid properties. 

To probe the cytoplasm, the group used genetically encoded multimeric nanoparticles (GEMs). GEMs, which express in cells as fluorescent particles of a fixed size, are a reliable method for particle tracking in biological systems. Newby’s team observed the GEMs and estimated their positions, then superimposed them into a three-dimensional (3D) video to track their movement. A complicating factor was the fact that hyphae—the thin, single-cell filaments of the syncytia in question—are often close to each other and can be difficult to separate when analyzing video data. In addition, measuring the GEMs’ position does not correlate the particles’ motion to their location within the cellular structure. To handle this issue, the researchers used surface segmentation to couple the position measurements with the cell’s physical structure.

A still from a three-dimensional (3D) video tracking the movement of genetically encoded multimeric nanoparticles (GEMs) within a hypha.

Identifying the individual tracks of GEMs in the cells proved difficult, as GEMs are small and highly mobile and 3D microscopy is slow when compared to two-dimensional microscopy. So the researchers instead focused on the variance of increments of Brownian motion from frame to frame, which is all they needed for a maximum-likelihood estimation of diffusivity. 

Newby and his collaborators processed the images by performing particle localization with a convolutional neural network, a class of deep neural networks that scientists often use to analyze visual images. Unfortunately, doing so ate up a lot of computing power — just one particle tracking video could be over 100 gigabytes. Since all of these data sets could not fit onto a single machine, the team utilized a tracking pipeline on Google Cloud with Apache Beam, an open-source resource for the creation of pipelines with data-parallel processing. They aimed to make this pipeline as automated as possible to remove human bias, though some parts did still need to be done by hand. To validate the pipeline, the group first qualitatively compared its output to the known locations of hyphae in the cell, as it was easy to visually identify any problems. 

The researchers overlaid the hypha locations with the output of the pipeline to visually validate the pipeline.
For a more quantitative approach to test the automated pipeline, Newby and his group used synthetic test videos to investigate whether the pipeline could accurately resolve a low-diffusivity region directly adjacent to a high-diffusivity region. The results were not perfect, as the pipeline underestimated the higher diffusivity, but it still performed well when approximating the relative diffusivity between the two regions. 

By using the pipeline to analyze GEMs data, Newby and his collaborators found that diffusivity was heterogeneous within cells and variable between cells. They saw more heterogeneity between hyphae than initially expected and discovered that the cytosol at the hyphae tips was more viscous than throughout the rest of the hyphae. In addition, a change of diffusivity occurred in the cytosol surrounding the nuclei throughout the cell cycle; this cytosol exhibited a lower diffusivity when the nuclei were in certain phases. The results of this study and future investigations with 3D cell particle tracking could lead to a deeper understanding of cytosol’s mysterious activity.  

   Jillian Kunze is the associate editor of SIAM News.