Sensory integration in interoception: Interplay between top-down and bottom-up processing

Although the neural systems supporting interoception have been outlined in general, the exact processes underlying the integration of visceral signals still await research. Based on the predictive coding concept, we aimed to reveal the neural networks responsible for the bottom-up (stimulus-dependent) and top-down (model-dependent) processing of interoceptive information. In a study of 30 female participants, we utilized two classical body perception experimentsdthe rubber hand illusion and a heartbeat detection task (cardioception), with the latter being implemented in fMRI settings. We interpreted a stronger rubber hand illusion, as measured by higher proprioceptive drift, as a tendency to rely on actual sensory experience, i.e., bottom-up processing, while lower proprioceptive drift served as an indicator of the prevalence of top-down, model-based influences. To reveal the bottom-up and topedown processes in cardioception, we performed a seed-based connectivity analysis of the heartbeat detection task, using as seeds the areas with known roles in sensory integration and entering proprioceptive drift as a covariate. The results revealed a left thalamus-dependent network positively associated with proprioceptive drift (bottom-up processing) and a left amygdala-dependent network negatively associated with drift (top-down processing). Bottom-up processing was related to thalamic connectivity with the left frontal operculum and anterior insula, anterior cingulate cortex, hypothalamus, right planum polare and right inferior frontal gyrus. Top-down processing was related to amygdalar connectivity with the rostral prefrontal cortex and an area involving the left frontal opercular and anterior insular cortex, with the latter area being an intersection of the two networks. Thus, we revealed the neural mechanisms underlying the integration of interoceptive information through the interaction between the current sensory experience and internal models. (c) 2021 Elsevier Ltd. All rights reserved.