Dopamine signals integrate innate and learnt valences to regulate memory dynamics

To survive in dynamic environments, animals make behavioral decisions based on innate and learnt information about the valences of sensory cues. In the brains of multiple species, innate and learnt sensory valence signals are initially encoded by distinct neural populations1–7 but then reconverge in downstream brain structures that guide behavioral choices6–9. This reconvergence hinges on the prior acquisition of learnt valence information, which in turn might depend upon innate valence signals. However, it remains unknown whether and how innate sensory valence cues shape the acquisition of learnt valence information. Here we show that in the fruit fly brain, interactions between innate and learnt sensory valence signals within interconnected short- and long-term memory units of the mushroom body jointly regulate memory formation and expression via modulation of dopamine teaching signals. By using time-lapse, in vivo optical voltage imaging to record neural spiking with millisecond-resolution in flies undergoing olfactory associative conditioning, we found that PPL1 dopamine neurons (PPL1-DANs) heterogeneously and bi-directionally encode punishment, reward, and innate and learnt odor-valence cues. Learning modulates these representations in a way that combines innate and learnt valence information and allows the PPL1-DANs to regulate memory storage in their downstream targets, mushroom body output neurons (MBONs). PPL1-γ1pedc and PPL1-γ2α’1 neurons control short-term memory formation. After repeated conditioning, feedback signals carrying short-term memory data from MBON-γ1pedc>α/β to PPL1-α’2α2 and PPL1-α3 allow these dopamine cells to encode previously learnt valences and promote long-lasting memory formation. A computational model constrained by the mushroom body connectome and our spiking data explains how dopamine signals integrate innate and learnt valence data to regulate memory storage, extinction, and the interactions between short- and long-term memory traces. The model yields non-intuitive predictions about the effects of different training protocols, which our experiments confirm. Overall, the mushroom body achieves flexible learning through dopamine-mediated integration of innate and learnt valences in parallel sets of DAN/MBON learning units with feedback interconnections. This hybrid physiologic-anatomic mechanism may be a general means by which ecological information regulates learning and memory in other species and brain structures relying on dopaminergic signaling, including the vertebrate basal ganglia10.