A Long-range Recurrent Neural Network Mediates Threat Induced Innate Sensorimotor Integrations

摘要 动物对厌恶刺激的先天行为反应涉及感官的多感觉整合 指导决策过程的信息。我们的长期目标是加深我们的基本知识 构成威胁诱导防御行为基础的感觉运动整合的神经机制。 我们将首先关注先天的三突触、长程递归神经网络(RNN)连接的功能作用 有躯体运动皮质的情绪区域。前脑的几个区域，包括丘脑、杏仁核和 额叶皮质区域参与表达与恐惧相关的行为和记忆。了解如何 这三个区域的相互作用对于破译恐惧引发行为的基本机制至关重要。最近， 我们发现，由mPFC(Hidden)中的自反馈连接形成的远程RNN 单位，HU)，整合来自上游情绪区域的输入(输入单位，Iu，包括基础侧支 杏仁核、岛叶皮质)，并进一步支配运动皮质投射神经元(输出单位，OU)。RNN是一种 具有自反馈(闭环系统)连接的网络。循环电路能够放大、图样 完成和记忆。需要检验的中心假设是先天的长程RNN，这是 由：位于杏仁基底外侧核(BLA)和岛叶皮质(IC， 即威胁检测单元)、mPFC中的决策隐藏单元(即恐惧整合单元)以及下游 位于躯体运动皮质(SMO，即恐惧输出单位)的动作启动中心在恐惧中起着关键作用 防御行为基础下的诱导感觉运动整合。理解三者中的神经元动力学 RNN在先天防御行为中的关键节点，对于破译RNN在威胁中的作用至关重要- 回避行为选择。提出了一个具体的目标和两个综合的子目标来测试这一点 假设：目的1.确定RNN环路是否参与先天防御行为。这一目标将考验 中心假设：次级目标1a。研究脑神经元电活动动力学和时间关系 防守时的输入单元(Iu，即BLA)、隐藏单元(HU，即mPFC)和输出单元(OU，即SMO 回避行为。次级目标1b。测试RNN关键节点激活的充分性和必要性， HU和RNN的最终输出节点：SMOS，在调节防御回避行为。成功 执行此试探性建议将帮助我们获得概念验证数据，从而实现 旨在了解mPFC在传感器电机集成中的关键作用的全面项目，特别是 RNN在情绪相关决策中作为基本计算单位的作用。

Summary An animal’s innate behavior responses to aversive stimuli involves multisensory integration of sensory information to guide the decision-making process. The long-term goal is to deepen our fundamental knowledge of the neural mechanisms underlying sensorimotor integration that underlies threat-induced defensive behavior. We will first focus on functional role of an innate tri-synaptic, long-range recurrent neural network (RNN) linking emotional regions with somatic motor cortex. Several forebrain regions, including the thalamus, amygdala and frontal cortical regions, is involved in expression of fear-related behaviors and memories. Understanding how these three regions interact is critical to deciphering the basic mechanisms of fear-triggered behavior. Recently, we discovered that a long-range RNN, which is formed by a self-feedback connectivity in the mPFC (Hidden Unit, HU), integrates inputs from upstream emotional regions (Input Unit, IU, which includes basal lateral amygdala, insular cortex) and further innervates motor cortex projecting neurons (Output Unit, OU). RNN is a network with self-feedback (closed-loop) connections. Recurrent circuits are capable of amplification, pattern completion and memory. The central hypothesis to be examined is that the innate long-range RNN, which is composed of: fear related emotional centers located in the basal lateral amygdala (BLA) and insular cortex (IC, i.e., a threat detection unit), decision making hidden unit in mPFC (i.e., a fear integration unit), and a downstream action initiation center located in the somatic motor cortex (sMO, i.e., a fear output unit), plays a key role in fear induced sensorimotor integration underlying defensive behavior. Understanding neuronal dynamics in the three key nodes of the RNN during innate defensive behavior, is crucial for deciphering the role of the RNN in threat- avoidance behavior selections. A specific aim with two comprehensive sub-aims are proposed to test this hypothesis: Aim 1. To determine if the RNN circuits are involved in innate defensive behavior. This aim will test the central hypothesis by: Sub-aim 1a. Examining the neuronal activity dynamics and temporal relationships in the Input Unit (IU, i.e. BLA), the hidden unit (HU, i.e. mPFC) and Output Unit (OU, i.e. sMO) during defensive avoidance behavior. Sub-aim 1b. Testing the sufficiency and necessity of activation of the key node of the RNN, HU and the final output node of the RNN: sMOs, in mediating defensive avoidance behavior. Successful execution of this exploratory proposal will help us obtain proof-of-concept data that will enable the pursuit of a full-scale project aimed at understanding the key role of mPFC in sensorimotor integration, and in particular the role of RNNs as fundamental computation units in emotion-related decision making.