Covariance Nonstationarity is Evident in Spatial Transcriptomics and Provides a New Categorization of Spatially Varying Genes
This study demonstrates that covariance non-stationarity is a prevalent and biologically significant feature in spatial transcriptomics that violates common Gaussian process assumptions, thereby necessitating a re-evaluation of how spatially varying genes are characterized and categorized.
Original paper licensed under CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). This is an AI-generated explanation of a preprint that has not been peer-reviewed. It is not medical advice. Do not make health decisions based on this content. Read full disclaimer
Imagine looking at a city from high above. From that distance, the streets might look like a uniform grid, and the buildings might seem to follow a predictable pattern. But if you zoom in, you see that the neighborhood changes completely. One block is a bustling market, the next is a quiet park, and another is a dense cluster of skyscrapers. The rules that govern how people move and interact in the market are different from those in the park. For a long time, scientists studying the microscopic world inside our bodies have often treated tissues like that distant city view, assuming that the rules governing how cells talk to each other are the same everywhere. They have assumed that if two cells are close together, they are likely to behave similarly, and that this relationship holds true no matter where you look in the tissue. This assumption, known as stationarity, has made the math easier, but it may have been hiding a more complex reality.
In the field of spatial transcriptomics, researchers map where genes are active within a tissue sample, creating a detailed picture of the body's internal geography. This technology has exploded in popularity, allowing scientists to see not just which genes are turned on, but exactly where they are doing their work. To make sense of these maps, scientists use statistical models to find genes that change their activity depending on their location. These are called spatially varying genes. The standard approach has been to use a single, uniform rule to describe how gene activity spreads across the tissue. However, a new study suggests that this uniform rule is often wrong. The researchers found that in many parts of the body, the way genes influence their neighbors changes from place to place, much like how the rules of a busy city street differ from those of a quiet suburb.
The team, led by researchers at the University of Victoria, set out to test whether this uniform assumption was holding up in real biological data. They examined twelve different tissue samples from humans and mice, including samples from breast cancer, ovarian cancer, and various parts of the brain. Instead of forcing all the data to fit a single, unchanging pattern, they allowed their models to change their behavior depending on the location. They compared the standard, unchanging models against more flexible ones that could adapt to local differences. The results were striking. In some tissues, nearly half of the genes showed signs that their behavior was not following a single, uniform rule. In the human breast cancer samples, for instance, fifty percent of the genes favored the flexible, changing model. In other tissues, the number was lower, but still significant, ranging from three percent to fifty percent. This means that for a large number of genes, the old assumption of a uniform pattern was missing a crucial piece of the puzzle.
When the researchers looked closer at the genes that did not follow the uniform rule, they found a clear biological story. These genes were not random; they were often the ones involved in the body's defense systems, such as immune responses and inflammation. In cancer tissues, these genes were frequently linked to the complex interactions between tumor cells and their surrounding environment. The study suggests that these genes are active in specific, localized pockets rather than spreading evenly across the whole tissue. For example, some genes showed high activity only along the edges of a tumor or in small clusters of cells, while others were active in a broad, shifting pattern that changed as you moved across the sample. This local variation is likely a reflection of the tissue's own complexity, where different groups of cells form distinct neighborhoods with their own unique chemical environments.
The researchers also checked to make sure their findings were reliable. They created thousands of simulated datasets based on the real data to see if their new method could correctly identify the changing patterns without making mistakes. The tests showed that their approach was highly accurate, correctly spotting the changing patterns most of the time while rarely flagging a gene as changing when it was actually uniform. This gives them confidence that what they are seeing in the real tissue samples is a genuine feature of biology, not just a glitch in the math.
Perhaps the most important takeaway is that ignoring these local changes can lead scientists to miss important biological signals. If a researcher assumes a gene behaves the same way everywhere, they might fail to notice that it is actually doing something critical in just one small part of the tissue. By recognizing that the rules of gene interaction can change from place to place, scientists can now categorize genes in a new way. Instead of just asking if a gene is active in a specific spot, they can ask how the gene's relationship with its neighbors changes as you move through the tissue. This new perspective offers a more nuanced view of how tissues are organized, revealing that the body is not a uniform landscape, but a collection of distinct, dynamic regions where the rules of life are constantly shifting.
Drowning in papers in your field?
Get daily digests of the most novel papers matching your research keywords — with technical summaries, in your language.