Long-term memory storage by the interaction of structural and synaptic plasticity in recurrent neuronal networks

为了在复杂多变的环境中生存，人类和动物通过形成记忆来学习过去的经验。因此，一种触发记忆形成的经历会导致大脑网络中的神经元激活，导致神经元之间的连接发生变化，而神经元之间的连接代表着这种经历的记忆。这一记忆项目的保留被认为取决于这些变化的持久性。然而，最近的实验表明，突触每天都会被创建和移除，而大部分突触是在几周的时间尺度上交换的。这就提出了一个问题，可以持续数年甚至数十年的长期记忆能在这样一个可变的衬底上存储和保持多久。具体地说，尽管有突触的更替，但长期记忆是如何从神经元活动和改变神经元连接的不同可塑性过程之间的相互作用中产生的，即结构可塑性和突触可塑性，结构可塑性描述突触的产生和移除，突触可塑性描述突触传递效率的变化。为了在网络层面上研究这种复杂的相互作用，我们提出了一种使用相关过程的数学模型的理论研究。我们假设，活动和所描述的可塑性机制之间的相互作用导致了细胞群体中的集体动力学，这可以由传入的刺激控制：在高刺激水平下，神经元活动和连接性之间的正反馈导致高度相互连接的细胞簇-所谓的细胞组合-被认为是记忆项目。在低刺激水平下，这些细胞之间的连接将被修剪，从而忘记记忆。在中等刺激水平，我们预计已经形成的组件可以维持很长一段时间。更准确地说，我们假设这些组件的活动和连接性之间的正反馈使它们稳定在突触翻转的情况下，这样它们就可以保留在长期记忆的时间尺度上。为了验证这些假设，我们将使用网络模拟和分析数学方法，这将使我们能够确定组装稳定性的决定因素。这项研究的结果将是一个动态记忆模型，它可以以一种依赖输入的方式形成或移除记忆，并在突触翻转的情况下将它们保持在长期记忆的时间尺度上。

In order to survive in complex and changing environments, humans and animals learn from past experiences by forming memories. Hereby, an experience, which triggers memory formation, causes neuronal activation in the brains networks leading to changes of the connections between neurons, which represent the memory of this experience. The retention of this memory item is believed to depend on the persistence of these changes. However, recent experiments show that synapses are created and removed on a daily basis and large fractions of synapses are exchanged on a timescale of weeks. This poses the question, how long term-memories, which can persist for years and decades, can be stored and maintained on such a variable substrate. Specifically, it has to be clarified how, in spite of synaptic turnover, long-term memory can emerge from the interaction between neuronal activities and different plasticity processes involved in changing the neuronal connectivity namely structural plasticity, describing the creation and removal of synapses, and synaptic plasticity, describing changes of the synaptic transmission efficacies. To investigate this complex interaction on the network level, we propose a theoretical study using mathematical models of the involved processes. We hypothesize that the interaction between activity and the described plasticity mechanisms leads to a collective dynamics in populations of cells, which can be controlled by the incoming stimulation: At high stimulation levels, a positive feedback between neuronal activities and connectivity leads to the formation of highly interconnected clusters of cells - so-called cell assemblies - which are considered as memory items. At low stimulation levels, the connections between these cells will be pruned, such that the memory is forgotten. At intermediate stimulation levels, we expect that already formed assemblies can be maintained for long period of time. More precisely, we hypothesize that the positive feedback between activity and connectivity in these assemblies stabilizes them against synaptic turnover, such that they can be retained on the timescale of long-term memory. To test these hypotheses, we will use both network simulations and analytical mathematical methods, which will allow us to identify the determinants of assembly stability. The outcome of this study will be a dynamic memory model which can form or remove memories in an input-dependent fashion and maintain them on the timescale of long-term memory despite synaptic turnover.