AI microscope flips the popular understanding of cell movement

A group of scientists led by Virginia Tech Professor Amrinder Nain has discovered that the behavior of cells in petri dishes can differ dramatically from those in living tissues. The findings, which have been published in Proceedings of the National Academy of Sciences, could explain several long-standing medical mysteries.

When cells in your body encounter scar tissue or the edges of a tumor, the stiffer tissue causes them to reduce their pulling forces by up to 90 percent. This behavior is the opposite of what has been traditionally taught and published in textbooks and may call into question previous research based on petri dish studies.

“We’ve been studying cells in completely artificial conditions for decades,” said Nain, a professor of mechanical engineering. “It would be like trying to understand bird behavior in cages, then being shocked to discover they behave differently when freed.”

The research potentially clarifies why cancer cells can invade soft tissue along specific pathways or why some wounds heal differently from others. It also offers new avenues for making engineered tissues perform more like organic ones.

Flat readings

For decades, scientists have studied cell forces using methods that fundamentally alter the measurements they’re trying to obtain. Traditional techniques require either killing cells, paralyzing them with chemicals, or bombarding them with fluorescent light that eventually causes them to die.

“Cells move at a snail’s pace, even compared to a snail. They move only about the width of a human hair per hour, so you can imagine our surprise when we discovered they can drastically tune their forces during movement,” said Nain. “The forces they generate are tiny, and measuring them without disturbing the cells was considered impossible.”

The traditional method for studying cells has been to place them on flat surfaces such as a petri dish. But in that environment, cells can flatten into shapes that look like fried eggs rather than wrapping around fibers and orienting themselves as they do in real tissues. Although estimates of cell forces have been made for cells in fiber networks, no technique existed for tracking forces in live cells in 3D.

Enter artificial intelligence (AI) and fiber mechanics. To pull the disciplines together, Nain partnered with two Virginia Tech colleagues: computer science Associate Professor Anuj Karpatne and mechanical engineering Assistant Professor Sohan Kale. The team created something unprecedented: a standard microscope-enabled strategy that uses deep learning AI to estimate the forces of cells nearly invisible to the human eye.

“The suspended fibers act as an ultra-sensitive sensor. Their deflections gave a direct readout of the tiny forces that cells exert,” said Kale.

“Imagine trying to track a transparent jellyfish swimming through a net while both are moving and changing shape,” said Karpatne. “It’s hard because they both distort and obscure each other, especially in the critical places around their boundaries. That’s what we asked the AI to do: track shape-shifting cells on deforming fiber networks.”

Breakthrough

The breakthrough came when graduate students Abinash Padhi, now a postdoctoral researcher at the University of Chicago, and Arka Daw, now a scientist at Amazon, developed a clever workaround. They trained the AI using fluorescent images, then taught it to extract the same information from regular microscope images that don’t harm cells.

What they found challenged the textbooks. On flat surfaces like the Petri dishes used in virtually all cell research, cells encountering stiffer materials pull harder. But in fiber networks analogous to those in the body, cells do the opposite: they relax when moving into stiffer regions, dropping their forces by 90 percent. The key? Direction matters more than stiffness.

“Cells care more about where they can pull than how hard the surface is,” said co-author Farid Alisafaei from New Jersey Institute of Technology. “It’s like climbing on a loose net: you have to bunch together to get a firm grip, and there is no sense pulling hard until you know which way to pull.”

This phenomenon, called tension anisotropy, was recently proposed by a team including Alisafaei and Guy Genin from the National Science Foundation–funded Center for Engineering MechanoBiology at Washington University in St. Louis. Tension anisotropy has been challenging to verify until now. The results of this study verified that cells definitively sense the direction of mechanical forces, not just their strength.

The team’s AI-powered microscope, called Deep-Learning-Enabled Fiber-Force Microscopy, revealed three discoveries.

• First, it showed that cells anchor to fibers everywhere, not just at the edges. Unlike on flat surfaces, where cells only grip at their borders, in 3D fiber networks, they form attachments throughout their bodies.
• Second, the microscope showed that force patterns predict stem cell fate. The microscope detected distinct force signatures long before stem cells showed any visible signs of what they would become. Cells destined to become fat cells suppressed their forces early; those becoming bone cranked up forces later.
• Third, the team discovered that cell contractility, or its ability to change force, can increase tenfold. Rather than pulling steadily, cells deploy their forces dramatically during movement, a phenomenon entirely invisible to traditional methods that only capture averages.

These findings shed light on medical mysteries that have puzzled physicians for decades, including why some wounds heal differently depending on tissue orientation, and why lab-grown organs behave differently than expected. “We’ve been developing drugs and treatments based on how cells behave on flat surfaces,” said Genin. “It’s time to test everything again in realistic conditions.”

Team mmbers are beginning to collaborate with oncologists to test whether disrupting directional cues could stop cancer metastasis without toxic chemotherapy. They’re also working with tissue engineers to design implants that give cells the right directional signals for healing.

“This isn’t just about having a better microscope,” Nain concluded. “It’s about realizing we’ve been playing by the wrong rulebook. Now we have an opportunity to write the right one.”

The work was funded in part by the National Science Foundation and the National Institutes of Health, the Virginia Tech Institute for Critical Technology and Applied Science and Virginia Tech Macromolecules Innovative Institute, and the Human Frontier Science Program.

Original study: DOI: 10.1073/pnas.2424047122