Interoceptive predictions in the brain

For decades, neuroscientists viewed the brain as a passive stimulus-response organ that waits for sensory input before acting. This Opinion article challenges that view by presenting the Embodied Predictive Interoception Coding (EPIC) model, which proposes that the brain actively generates predictions about both the external world and the body's internal state. Rather than perception following sensation, the authors argue that interoceptive experiences—our feelings about heart rate, breathing, hunger, and other bodily states—largely reflect the brain's predictions about what the body should be experiencing, constrained by actual ascending sensory signals.

The EPIC model integrates Bayesian active inference principles with Barbas and colleagues' structural model of corticocortical connections, examining how predictions and prediction errors flow through different layers of cortical tissue. The authors propose that agranular visceromotor cortices (including the anterior cingulate and anterior insula) issue predictions about the body's internal state to maintain homeostasis, while granular sensory cortices in the mid- and posterior insula compute prediction errors when actual interoceptive signals diverge from predictions. This hierarchical organization explains how the brain minimizes discrepancies between expected and actual bodily sensations through three mechanisms: modifying predictions, generating motor commands to change sensations, or adjusting attention to incoming signals.

The authors propose that disruptions in interoceptive predictions may represent a common vulnerability factor underlying various mental and physical illnesses. They illustrate this with depression, where structural and functional abnormalities in agranular visceromotor regions may lead to chronic imbalances in how the brain predicts the body's needs, ultimately resulting in the physiological dysregulation and behavioral changes characteristic of depressive illness. The model suggests that conditions like depression, anxiety, and metabolic disorders may share aberrant interoceptive prediction as a common neural mechanism.

This framework has significant clinical implications for understanding and treating mental and physical illnesses. The EPIC model suggests that therapies like cognitive behavioral therapy and deep brain stimulation may work by correcting aberrant interoceptive predictions and their downstream effects on autonomic, endocrine, and immune function. The authors call for empirical verification of the model through resting-state connectivity analyses, high-resolution fMRI, and electrocorticography to map the interoceptive system in the human brain and test specific hypotheses about prediction and error computations within cortical layers.