Unsupervised machine-learning identifies latent pyrenoid states linked to mitotic remodeling defects and CO2-dependent growth

Inside the microscopic world of single-celled algae, life depends on a delicate, invisible machinery that captures carbon dioxide from the air and turns it into food. This process, known as photosynthesis, relies on a specific enzyme that acts like a molecular factory worker, grabbing carbon atoms to build sugars. However, this worker is often inefficient, and to make it work better, some algae have evolved a special compartment, a dense cluster of proteins that concentrates carbon dioxide right where it is needed. This structure, called a pyrenoid, behaves like a liquid droplet floating inside the cell, constantly shifting and reforming. The challenge for these tiny organisms is that this liquid structure must survive the violent event of cell division. When the cell splits in two to create offspring, the pyrenoid must be dismantled, sorted, and rebuilt in each new daughter cell. If this reorganization goes wrong, even in subtle ways that are hard to see, the new cells might struggle to grow or survive, especially when carbon dioxide is scarce.

Scientists have long wondered whether small, hidden flaws in how these structures are rebuilt during cell division could lead to bigger problems later on. To answer this, researchers turned to a common green alga, Chlamydomonas reinhardtii, and developed a new way to look at its internal structures. Instead of relying on human eyes to spot differences, which can miss the faintest irregularities, they taught a computer to recognize what a healthy pyrenoid looks like. They started by taking thousands of images of normal, healthy cells and using a type of artificial intelligence to learn the full range of what "normal" looks like. This computer system, trained on nearly five thousand images of wild-type cells, learned to identify the standard shape and texture of the pyrenoid. To make this training even more robust, the researchers expanded their dataset, creating over one hundred thousand image variations to ensure the computer understood every possible normal state.

With this digital standard in place, the team began a massive search through a library of twenty-one thousand mutant algal strains, each carrying a random genetic change. They fed images of these mutants into the computer system, which compared them against the learned standard of normality. The computer flagged seventeen specific strains that showed unusual patterns, identifying them as having defects in their pyrenoid structure. What made this approach powerful was its ability to detect subtle deviations that a human observer would likely overlook. By analyzing exactly where the computer saw a difference, the researchers could map out the specific local flaws in the pyrenoid, revealing irregularities that were too complex to classify by eye alone.

When the researchers watched these mutant cells in real-time, they saw the consequences of these hidden defects. In several of the identified strains, the pyrenoid failed to break apart correctly during cell division, or the pieces containing the essential enzyme did not distribute evenly to the new cells. In other cases, the structure failed to reassemble properly after the cell split. These mechanical failures in the cell's division process translated directly into growth problems. The mutant algae grew normally when carbon dioxide was plentiful, but as the supply of carbon dioxide dropped, their growth slowed significantly. This confirmed that the subtle structural flaws in the pyrenoid made the cells much more vulnerable when resources were tight.

The study also pointed to specific genes responsible for these issues. By mapping where the random genetic changes occurred, the researchers identified a gene called STT7, which produces a protein known to regulate how the alga captures light. To confirm this link, they created new lines of algae where this gene was deliberately turned off. These modified algae showed the same strange patterns in their pyrenoid structure as the original mutants, and they lacked the STT7 protein entirely. This evidence suggests that the STT7 protein plays a direct role in maintaining the shape and function of the pyrenoid, connecting the regulation of light harvesting to the physical integrity of the carbon-concentrating structure.

This work demonstrates that unsupervised machine learning can uncover biological defects that are too subtle for traditional observation. By letting a computer define the boundaries of normal and then searching for anything that falls outside them, scientists can find new genetic players involved in complex cellular processes. The findings show that the health of a cell's growth and its ability to thrive in low-carbon environments are deeply tied to the precise, often invisible, mechanics of how it rebuilds its internal structures during division. The ability to detect these latent states opens a new window into understanding how life maintains its essential machinery through the chaos of reproduction.