Using an artificial intelligence (AI) cell classification technique, Northwestern Medicine investigators found that viruses can control structural and genetic polarity inside the cell nucleus. The findings, published in Nature, highlight the importance of genome organization during infection and the extent to which AI can help scientists identify complex intracellular processes.
Viruses can control cells in many ways, from viral proteins present in the nucleus directly controlling gene expression to proteins working on the cell’s surface or in the cytoplasm to control cell signaling networks. But how and why the nucleus is moved and reorganized under various conditions, including during viral infection, has remained a matter of investigation, according to the authors.
A core issue in studying any intracellular process is that there’s often considerable heterogeneity in what is happening in each individual cell within a cell culture, according to Derek Walsh, PhD, professor of Microbiology-Immunology and senior author of the study.
“During infection, for example, you can have some uninfected cells, where some infections fail, and in those that are infected, each cell can be at a different stage of infection. This is very hard to experimentally control or synchronize, and standard approaches such as blotting simply give an ‘average’ of what is happening at a given time when you harvest cells,” Walsh said.
Investigators were also able to image individual cells, but that process requires analyzing copious amounts of cells in order to get an accurate picture of what is really happening, which can be very prone to error, according to Walsh.
“Being human, it’s also hard not to focus on the more striking and obvious phenotypes that you see, which can create more subjective analysis or unintentionally exaggerated phenotypes that are simply easier for us humans to see and work with,” Walsh added.
To simplify this process, a team led by Dean Procter, PhD, a postdoctoral fellow in the Walsh laboratory, developed automated cell imaging systems that use AI based networks called convolutional neural networks to identify and analyze infected cells.
“Frustrated by the limitations of previously available image analytics tools, we sought to develop analysis pipelines that leveraged the recent advances in computer vision technology that are rapidly changing our world,” Procter said.
Specifically, the investigators provided the system large training datasets to learn how to identify infected cells and different stages of infection in the cells. Once the network was trained, the team programmed a microscope to scan and image entire coverslips containing cell cultures. The system then sorted and classified which cells were infected and at what stage of infection they were at.
The investigators were then able to program the system to identify certain parameters, such as the brightness and location of specific proteins in the nucleus, and generated either “line scans” of intensities that it measured or an “average projection” of the entire specified area in the cell.
“When it does this over thousands and thousands of cells, what you end up with is a user-independent and completely unbiased ‘spatial western blot’ for cells that are infected, leaving aside uninfected cells or cells that are not relevant to your analysis,” Walsh said.
Using the system, the investigators identified an extensive regulatory pathway from samples of infected cells. This pathway generates strong acetylated microtubules — tubular structures present in the cytoplasm — which attach to the cell’s nuclear membrane and intranuclear proteins to control actin filaments. This, in turn, reorganized the cell nucleus internally, controlling its structural and genetic polarity.
“What was surprising was that a virus can form microtubules in the cytoplasm that effectively grab hold of the nuclear surface, and then uses this to reorganize the inside of the nucleus in a form of ‘outside-in’ control. Another surprising aspect was the discovery that nuclear actin filaments are involved,” Walsh said.
The findings, according to Walsh, may improve the understanding of the fundamental mechanisms of genome organization in infected cells and how this contributes to overall infection.
“We also hope that by providing a relatively unbiased ‘neural network definition’ of the stages of human cytomegalovirus replication, we can begin to build a community-sourced standard that is routinely used,” Procter said.
This work was supported by the National Institutes of Health grants R01AI141470 and P01GM105536.