Stimulus-specific neural encoding of a persistent, internal defensive state in the hypothalamus

Summary

Persistent neural activity has been described in cortical, hippocampal, and motor networks as mediating short-term working memory of transiently encountered stimuli^1–4. Internal emotion states such as fear also exhibit persistence following exposure to an inciting stimulus^5,6, but such persistence is typically attributed to circulating stress hormones^7–9; whether persistent neural activity also plays a role has not been established. SF1⁺/Nr5a1⁺ neurons in the dorsomedial and central subdivision of the ventromedial hypothalamus (VMHdm/c) are necessary for innate and learned defensive responses to predators^10–13. Optogenetic activation of VMHdm^SF1 neurons elicits defensive behaviors that can outlast stimulation^11,14, suggesting it induces a persistent internal state of fear or anxiety. Here we show that VMHdm^SF1 neurons exhibit persistent activity lasting tens of seconds, in response to naturalistic threatening stimuli. This persistent activity was correlated with, and required for, persistent thigmotaxic (anxiety-like) behavior in an open-field assay. Microendoscopic imaging of VMHdm^SF1 neurons revealed that persistence reflects dynamic temporal changes in population activity, rather than simply synchronous, slow decay of simultaneously activated neurons. Unexpectedly, distinct but overlapping VMHdm^SF1 subpopulations were persistently activated by different classes of threatening stimuli. Computational modeling suggested that recurrent neural networks (RNNs) incorporating slow excitation and a modest degree of neurochemical or spatial bias can account for persistent activity that maintains stimulus identity, without invoking genetically determined “labeled lines”¹⁵. Our results provide causal evidence that persistent neural activity, in addition to well-established neuroendocrine mechanisms, can contribute to the ability of emotion states to outlast their inciting stimuli, and suggest a mechanism that could prevent over-generalization of defensive responses without the need to evolve hardwired circuits specific for each type of threat.