Cracking the Serotonin Code: How Your Brain Predicts Future Rewards
Post by Rachel Sharp
The takeaway
Serotonin neurons signal a prospective code for value – a prediction of near-future rewards that explains why these neurons respond to both rewards and punishments. This unifying theory reconciles seemingly contradictory previous explanations and provides a framework for understanding serotonin's role in learning and behavior.
What's the science?
Serotonin has puzzled scientists for decades. This chemical messenger in the brain has been linked to everything from mood and sleep to learning and decision-making. Despite this, as well as its crucial role in mental health, scientists have struggled to figure out just how serotonin-producing neurons function and respond to stimuli.
The brain receives most of its serotonin from a cluster of cells deep in the brainstem called the dorsal raphe nucleus. Previous theories suggested that serotonin neurons might signal reward, surprise, salience (how significant something is), or even uncertainty, but none of these theories alone could explain all the observed patterns.
A new study published in Nature by Harkin and colleagues might have finally cracked the code. This study introduces a model for serotonin signalling that unifies these perspectives - a "prospective code for value" - that predicts future rewards while efficiently compressing information. The researchers explain unanswered questions like why serotonin neurons activate in response to both rewards and punishments, and why they respond more strongly to surprising rewards, but show no such preference for surprising punishments.
How did they do it?
The researchers built a mathematical model of serotonin neuron activity based on reinforcement learning principles, where "value" represents an estimate of total future reward. Their key insight was that serotonin neurons don't just encode raw value, but rather they filter this signal through a process called spike-frequency adaptation – where neurons gradually decrease their firing rate in response to sustained stimulation.
This filtering creates what's called a "prospective code" that emphasizes surprising value changes while also compressing slow, expected fluctuations in value. Overall, the proposed model predicts that serotonin neurons will:
Show temporary activation when a reward-predicting cue appears
Activate at the end of a punishment (when value increases as punishment ends)
Respond more strongly to surprising rewards than expected ones
Show similar responses to both expected and unexpected punishments
To test their model, the researchers analyzed multiple datasets of serotonin neuron recordings from mice during experiments in which animals learned to associate cues with delayed rewards. Finally, they compared their model to competing theories that base the coding of serotonergic activity solely on reward, surprise, and salience.
What did they find?
The prospective value coding model successfully accounted for previously unexplained aspects of serotonin neuron activity:
Responses to rewards and punishments: The model explains why serotonin neurons increase their firing when either a reward begins or a punishment ends: both represent increases in value.
Context modulation: The model accounts for why baseline serotonin activity is higher in reward-rich environments and lower in punishment-rich environments: the activity reflects the average value of each context.
Surprise preference for rewards but not punishments: The model clarifies why serotonin neurons show stronger activation to unexpected rewards but similar responses to both expected and unexpected punishments: previous theories have been unable to account for this.
Finally, when tested against previous theories using quantitative analysis of real neural data, the prospective code for value outperformed competing models in predicting actual serotonin neuron activity.
What's the impact?
By providing a unified computational framework for serotonin function, this research reconciles competing theories and establishes a clear connection to critical aspects of reinforcement learning.
Understanding serotonin's role as a prospective code for value has implications for both basic neuroscience and clinical applications. Many psychiatric disorders involve disruptions in the serotonin system, including depression, anxiety, and obsessive-compulsive disorder. A clearer understanding of serotonin's computational role could lead to more targeted treatments and better explanations of how existing treatments work.
Beyond clinical relevance, this research illuminates a fundamental principle of neural processing: adaptation mechanisms. Rather than simply representing information directly, neurons transform signals in ways that make them more useful to downstream brain regions. This principle of efficient coding likely extends beyond the serotonin system to other modulatory systems throughout the brain, suggesting a common computational strategy across neural circuits.