Image Analysis Tools for Continuous Label-free Quantification of Proliferation and Adipogenic Differentiation of Mesenchymal Stem Cells

In the world of modern biology, scientists are constantly trying to understand how cells grow and change. Two of the most important processes they study are proliferation, which is simply the way cells multiply to increase their numbers, and differentiation, where a generic cell transforms into a specialized one, such as a fat cell. For decades, researchers have relied on methods that require stopping the experiment, killing the cells, and staining them with colorful dyes to see what is happening. While these methods provide a snapshot, they are destructive; once you look at the cells, you cannot watch them again. This makes it difficult to understand the smooth, continuous story of how a cell population evolves over time. Furthermore, the chemicals used to make cells visible can sometimes damage them or create false signals, leading to inaccurate counts.

A growing field of research is now turning to a different approach: watching cells without touching them at all. By using advanced cameras and computer programs, scientists can observe living cells as they grow and change, capturing thousands of images over days or weeks without ever adding a single drop of dye. This "label-free" method promises a clearer, more honest view of cellular life. It is particularly vital for emerging industries like cultivated meat, where scientists need to grow large quantities of animal fat and muscle in a lab, and for medical research into obesity and diabetes, where understanding the precise timing of fat cell formation is crucial. The challenge, however, has been teaching computers to recognize these cells accurately in the images, as cells often clump together and look very similar to one another.

In a recent study, researchers set out to test two different ways of using this label-free technology to track mesenchymal stem cells, which are versatile cells capable of turning into fat. These cells were taken from both humans and pigs, reflecting the dual interest in human health and the production of sustainable meat. The team compared two distinct strategies. The first was a built-in software system provided by the manufacturer of the live-cell imaging device, designed to be easy to use and integrated directly into the microscope. The second was a custom-made computer model, trained by the researchers themselves using a powerful, open-source tool called Cellpose, which is designed to help computers "see" and separate individual cells in complex images.

The researchers grew these cells in the lab, watching them multiply and then guiding them to become fat cells. They took pictures of the cultures every hour for several days, creating a massive library of images showing the cells at every stage of their life. They then ran these same images through both the built-in software and their custom model to see which one could count the cells and measure their growth more accurately. When it came to counting cells as they multiplied, both methods worked well, but they had different strengths. The built-in software was excellent at handling crowded scenes where cells were packed tightly together, providing a reliable count even when the dish was full. The custom model, however, was superior at lower densities, where it could clearly distinguish individual cells and count them with high precision, even when they were just starting to grow.

The story became more interesting when the cells began to turn into fat. As the cells differentiated, they started to fill with lipid droplets, which are tiny bubbles of fat inside the cell. The built-in software was very good at spotting these mature fat cells because they look different and have a higher contrast than the other cells. It could tell the researchers how much of the dish was covered by these mature fat cells. However, it struggled to see the early stages of the process, missing the cells that had just started to change but had not yet filled with enough fat to stand out. The custom model, on the other hand, was trained to recognize the shape of the cells themselves, not just the fat inside them. This allowed it to spot the very first signs of differentiation, identifying cells that were beginning to commit to becoming fat long before they were fully mature. This gave the researchers an earlier and more detailed view of the transformation.

The study also tested whether these computer models could work across different species and different types of microscopes. The custom model, which was initially trained on images of pig cells, could not automatically recognize human cells without being retrained, highlighting that these computer programs need specific examples to learn. However, once the researchers trained a new version of the model on human cells, it worked just as well. Furthermore, they proved that these models were not limited to the specific high-tech imaging system used in the lab. They successfully applied their custom models to images taken with standard, manual microscopes, showing that this technology can be used in many different laboratories, not just those with expensive, automated equipment.

One of the most significant findings was the ability to see the hidden variations within a single culture dish. Because the imaging system took pictures of many different spots in the dish over time, the researchers could see that fat cells did not all form at the same time or in the same places. Some areas of the dish were full of fat cells, while others remained empty. This kind of spatial detail would have been impossible to see with traditional methods, which usually take a single picture from the center of the dish at the very end of the experiment. The continuous monitoring also revealed that different batches of cells, even from the same type of animal, behaved differently. Some batches turned into fat quickly and efficiently, while others struggled, a variation that could be critical for producing consistent cultivated meat or for understanding why some people are more prone to obesity than others.

Ultimately, the study demonstrated that label-free imaging is a powerful tool that removes the need for messy, destructive staining procedures. It allows scientists to watch the entire life cycle of a cell population in real time, from the moment they are seeded to the moment they become specialized fat cells. While the built-in software offers a convenient, plug-and-play solution that works well for mature cells, the custom-trained models provide a level of flexibility and detail that is unmatched, capable of spotting the earliest changes and adapting to different cell types and microscope setups. By combining these approaches, researchers can now track the dynamics of cell growth and fat formation with a clarity that was previously out of reach, offering new insights for both medical science and the future of food production. All the computer models developed in this study have been made available to the public, allowing other scientists to use and improve upon these tools for their own research.