Predicting Pain Using a Data-Driven Agent-Based Model of the Bilateral Central Amygdala

Sensory processing in the amygdala is a complex, dynamic process. Decades of surgical, electrical, pharmacological, optogenetic, and chemogenetic in vivo manipulations have revealed the nociceptive functions of anatomically- and genetically-restricted neuronal populations. In parallel, molecular and electrophysiological approaches have allowed for high-resolution, temporal examination of nociceptive-induced alterations in amygdala plasticity. Computational integration of this data is critical for future therapeutic development; in practice, these models would allow for in silico prediction of amygdala activity following injury, and in a reciprocal fashion, changes in pain-like behaviors following manipulation of discrete amygdala neuronal populations. To this end, we developed a three-dimensional computer model of the bilateral central nucleus of the amygdala (CeA). We employed agent-based modelling to integrate wet-lab data from two CeA cell populations: Calcitonin Gene-Related Peptide Receptor (CGRPR; Calcrl) expressing cells and Protein Kinase C delta (PKC{delta}; Prkcd) expressing cells. We integrated the spatial location, connectivity, neuronal activity, and electrophysiological properties of these neurons in our realistic bilateral model architecture. Our model captures properties of the amygdala that drive pain modulation, including hemisphere-specific physiological differences, and generates predictions of nociception related to bladder injury. Predictions from the model were compared retrospectively to pain outcomes during manipulation of CGRPR-expressing neurons in whole mice. These comparisons show strong alignment between our model and in vivo outcomes. Significance StatementNociceptive processing in the amygdala emerges from interactions among cell types. No framework has been developed that recapitulates this complexity. Here, we generated a 3-D agent-based computational model of the bilateral central amygdala (CeA) using data that accounts for pain-associated changes in neuronal physiology, cellular identity, anatomical positioning, and temporal specificity. The model revealed that hemispheric differences in intrinsic CeA neuronal excitability are dominant drivers of nociceptive output. Furthermore, when manipulated to mimic injury-associated plasticity, the model accurately predicted behavioral outcomes. This model is a publicly available computational tool that predicts hemisphere-targeted pain therapy efficacy. Moreover, the flexibility of this framework will allow future adaptations to other brain areas and disease contexts.