Abstract
In the mammalian brain, excitatory and inhibitory synapses are generally distinct and have fixed synaptic signs. Therefore, unlike in artificial neural networks, learning in biological networks is thought to be manifested by plasticity mechanisms that modify synaptic weights but not signs. Here, we demonstrate experience-dependent sign switching at synapses between glutamate and GABA co-releasing neurons of the entopedunculus (EP) and their targets in the lateral habenula (LHb). Pairing of reward or punishment with activation of EP co-releasing neurons makes EP-LHb synapses relatively more inhibitory or excitatory, respectively. Synaptic sign switching modulates downstream dopaminergic signaling, correlates with recent dopamine updates, and contributes to reinforcement learning. These data unveil a plasticity mechanism that alters both synaptic signs and weights to rapidly update dopamine release and drive learning.