Learning without a brain—how bacteria store memories and remember the past like artificial neural networks

Conventional wisdom holds that learning requires a brain. However, the capacity to learn is not exclusive to organisms possessing neurons. A recent study in the journal PRX Life demonstrates that a single bacterium can learn from its experiences, store memories of the past, and utilize those memories to anticipate future events.

Bacteria inhabit environments that fluctuate constantly across various timescales. Nutrient levels rise and fall, temperatures change, and threats from antibiotics appear and disappear. To thrive, a bacterium must swiftly react to its current surroundings while simultaneously retaining crucial information about its recent experiences.

Adapting too rapidly exposes the bacterium to changing conditions, while forgetting too quickly prevents it from anticipating recurring threats. This delicate balance intrigued a computational biophysicist who studies how living systems process information and adapt to changing environments. To explore whether single-celled organisms like bacteria can learn from past experiences, the team employed a microfluidic device to monitor the behavior of tens of thousands of individual E. coli cells as they altered the bacteria's nutrient supply at varying rates.

The researchers discovered that bacteria not only respond to the existing nutrient levels in their environment but also retain a record of their nutrient history to navigate changing conditions. If bacteria merely reacted to their present environment, they would respond identically to a sudden influx of food, irrespective of whether their previous environment was stable or fluctuating.

However, when exposed to the same nutrient surge, bacteria that had recently experienced a feast-and-famine environment adapted more rapidly than those originating from a stable environment. Since the immediate conditions were identical for both groups, the researchers concluded that the difference in their behavior stemmed from an internal record of their past, rather than a simple reaction to their immediate surroundings.

In other words, the bacteria's past experiences were shaping their present behavior—indicating that they were learning. To further understand how bacteria stored this information, the team constructed a mathematical model of the internal molecular network regulating bacterial growth. This model not only replicated how bacteria behaved under different nutrient environments but also pinpointed where their memory likely resides.

The model suggested that ribosomes, the molecular factories responsible for building proteins and determining how fast a cell grows, serve as a potential memory storage site. By analyzing the model, the researchers found that some ribosomes responded swiftly to nutrient changes, while others altered more gradually. They reasoned that rapid responders track current conditions, while slow responders retain traces of the past, creating a memory that spans various timescales—from minutes to hours.

This molecular system follows the same computational logic as a gated recurrent neural network, a type of artificial intelligence used to process sequences like speech and sensor data. Central to this process is a gate—essentially a molecular switch that determines how much of an existing memory to retain and how much to overwrite when new information emerges.

While maintaining readiness to adapt consumes resources that could otherwise be allocated to growth, bacteria employ molecular gates to actively regulate the retention and discarding of information. Modern AI systems rely on gates to remember essential details while remaining flexible enough to learn new information. In bacteria, the gate is not software-based but arises from the cell's chemistry.

This strategy also addresses a challenge familiar to AI researchers: how to learn something new without erasing previously acquired knowledge. Bacterial cells manage the balance between memory and flexibility by storing information across multiple timescales. This history-dependent computation does not necessitate neurons. Instead, the study revealed that a network of ribosome populations is sufficient for this purpose.

Long before humans developed artificial intelligence, bacteria had independently evolved a remarkably similar strategy for leveraging past experiences to prepare for future challenges. Using the broadest definition of learning, the findings demonstrate that memory and learning can emerge from chemical networks within a cell, devoid of a brain or neurons.

Moreover, the mathematical model offers a precise, quantitative language for describing how cells process information and make decisions. For artificial intelligence, it provides a biological blueprint for constructing systems capable of continuous learning under stringent energy constraints. These findings may also have implications for medicine.

Many pathogenic bacteria adapt to changing conditions within the body by relying on cellular memory. If this adaptability hinges on memory storage, future drugs designed to combat infection could target the molecular components responsible for enabling cells to store and utilize information.