Computational Models of Incentive-Sensitization in Addiction: Dynamic Limbic Transformation of Learning into Motivation

Abstract

Incentive salience is a motivational magnet property attributed to reward-predicting conditioned stimuli (cues). This property makes the cue and its associated unconditioned reward ‘wanted’ at that moment, and pulls an individual’s behavior towards those stimuli. The incentive-sensitization theory of addiction posits that permanent changes in brain mesolimbic systems in drug addicts can amplify the incentive salience of Pavlovian drug cues to produce excessive ‘wanting’ to take drugs. Similarly, drug intoxication and natural appetite states can temporarily and dynamically amplify cue-triggered ‘wanting’, promoting binge consumption. Finally, sensitization and drug intoxication can add synergistically to produce especially strong moments of urge for reward. Here we describe a computational model of incentive salience that captures all these properties, and contrast it to traditional cache-based models of reinforcement and reward learning. Our motivation-based model incorporates dynamically modulated physiological brain states that change the ability of cues to elicit ‘wanting’ on the fly. These brain states include the presence of a drug of abuse and longer-term mesolimbic sensitization, both of which boost mesocorticolimbic cue-triggered signals. We have tested our model by recording neuronal activity from mesolimbic output signals for reward and Pavlovian cues in the ventral pallidum (VP), and a novel technique for analyzing neuronal firing “profile”, presents evidence in support of our dynamic motivational account of incentive salience.