Nonlinear computation by a habenula-driven recurrent inhibitory network in the raphe

Serotonin (5-HT) neurons in the dorsal raphe nucleus (DRN) receive a constellation of long-range inputs, yet guiding principles of local circuit organization and underlying computations in this nucleus are largely unknown. Using inputs from the lateral habenula (LHb) to interrogate the processing features of the DRN, we uncovered 5-HT1A receptor-mediated recurrent connections between 5-HT neurons, refuting classical theories of autoinhibition. Cellular electrophysiology and imaging of a genetically encoded 5-HT sensor revealed that these recurrent inhibitory connections spanned the raphe, were slow, stochastic, strongly facilitating, and gated spike output. These features collectively conveyed highly non-linear dynamics to this network, generating excitation-driven inhibition and winner-take-all computations. In vivo optogenetic activation of LHb inputs to DRN, at frequencies where these computations are predicted to ignite, transiently disrupted expression of a reward-conditioned response in an auditory conditioning task. Together, these data identify a core computation supported by an unsuspected slow serotonergic recurrent inhibitory network.