Figure 1: A human brain with the hippocampus highlighted. RIKEN researchers have developed a model that may explain how context-dependent behavior arises in the hippocampus. © KATERYNA KON/SCIENCE PHOTO LIBRARY
How animals may modify their behavior depending on their context has been modeled mathematically by two RIKEN neuroscientists1. Their simple but biologically plausible model could shed light on mental disorders such as autism and schizophrenia.
Animals, including humans, commonly behave differently depending on the situation they are in. For example, hearing the word "Fire!" could cause someone to evacuate a building or to pull the trigger of a gun depending on the context.
That implies that the brain must combine sensory input and context when deciding how to react.
An important brain area for this context-dependent behavior is the region known as the hippocampus. In particular, neurons in the hippocampus-especially place cells that register location-undergo reorganization in different contexts. This process is known as hippocampal remapping.
But it's not known how context-dependent behavior sequences arise in the hippocampus. Researchers have developed various theoretical models to explain context-dependent behavior, but they suffer from shortcomings such as not being realistic from a biological standpoint.
"The main thing we want to try to understand is how the brain can achieve flexible behavior without resorting to highly sophisticated learning rules," says Taro Toyoizumi of the RIKEN Center for Brain Science (CBS).
Now, Toyoizumi and Yoshiki Ito, also of CBS, have modeled flexible behavior in animals using a simple reinforcement learning model that makes sense biologically.
Despite its simplicity, the model can reproduce various observations in both mice and people.
"It's surprising that this concept hadn't been proposed before," says Toyoizumi. "It uses the well-known associative memory model to retrieve an appropriate context, defined as the combined activity pattern of the sensory-encoding and context-encoding neurons."
The model consists of two network modules: a context selector that stores possible contexts as attractors, and a sequence composer that learns to create context-dependent sequences. These sequences inform courses of actions and predictions about future outcomes via reward-based learning. When the prediction differs from what actually occurs, the context selector triggers remapping in the sequence composer.
Despite being a very basic model, it could offer valuable insights into mental disorders, the researchers think. Patients with schizophrenia or autism often have problems with processing sensory input and flexibility in behavior, but the underlying causes of these symptoms are unknown. A possible explanation is that the balance between the numbers of sensory-encoding and context-encoding neurons could be skewed in these conditions.
"A good balance between the two is crucial for performance," says Toyoizumi. "We conjecture that if one of them is too big, it may give rise to these mental disorders."
"Restoring the balance of neural representations in the brain may help improve psychiatric symptoms," Ito adds.
