利用TRAP2小鼠和化學遺傳學抑制探討選定腦區在兩機率選項作業學習過程中的角色

Abstract

機率學習是透過嘗試和錯誤學會機率概念的認知學習過程，在這個過程中個體需要透過在一系列選項中選擇其一，並在獲得獎勵或懲罰後，不斷重複前述過程，以學習蘊含在選項背後的機率概念。這樣的行為在現實生活中是相當必要的。目前已有研究指出一些精神疾病的患者在這樣的學習歷程中，有和正常個體不同的行為模式，因此去研究大腦各部位如何參與形成這樣的機率學習認知歷程就是一個相當重要的課題。因為在人類身上能進行的實驗操弄有限，難以了解這樣複雜的行為背後的因果關係，因此本實驗利用小鼠作為模型來進行一連串的研究。 在這篇研究中，小鼠將在一個我們自製的行為裝置中完成機率學習，這個金屬行為裝置由matlab code 控制arduino 使小鼠能夠在裝置內完成半自動化的學習。機率學習的過程主要分為三大階段：行為塑造階段(shaping phase)、第一次機率學習(acquisition)、第二次機率學習(testing phase)。 在機率學習的過程當中，小鼠在選擇左右兩側的選項後會有不同的機率得到獎勵，分別為百分之二十或八十。在第一部分的實驗中我們將在此機率學習的過程中，利用TRAP2小鼠來觀察其在前額葉皮層(medial prefrontal cortex) 、背內側紋狀體(dorsal medial striatum)、背外側紋狀體(dorsal lateral striatum) 學習第一天以及最後一天的神經活動動態變化。第一部分的實驗結果顯示，在機率學習的初始階段，內側眼匡皮質(medial orbitofrontal cortex) 以及背內側紋狀體的活躍程度較高，在學習歷程的最後一天這兩個腦區的活動程度皆較低；相對而言，前邊緣皮質(prelimbic cortex)和背外側紋狀體在整個機率學習過程中保持相同的活躍程度。這樣的發現暗示著內側眼匡皮質(medial orbitofrontal cortex) 以及背內側紋狀體早期活躍與否可能是此機率學習的關鍵。 為了近一步釐清找到的特定腦區在此作業學習歷程中不同階段的重要性，我們透過化學遺傳學抑制進行後續操弄。由於已經有相關研究指出前額葉皮質有解剖及功能上的投射至背內側紋狀體，因此在第二部分的實驗中我們選擇對下游的背內側紋狀體進行化學遺傳學的抑制操弄，看看在學習的早期及晚期該腦區的活躍程度變化是否有功能上的意義。在第二部分的實驗中，共有四個組別：全期抑制組(all-time inhibition group)、早期抑制組(early-inhibition group)、晚期抑制組(late-inhibition group)和控制組。分別抑制整個機率學習歷程，或者只抑制早期或晚期的學習歷程，控制組則在整個學習階段都沒有被抑制。根據第二部分的實驗結果，全期抑制組和早期抑制組的表現沒有差異且相比於晚期抑制以及控制組差，這說明了背內側紋狀體在機率學習的早期更為重要且可能負責在該時間段將當關鍵的功能。我們利用reward-noreward 模型來分析小鼠的行為後發現，早期抑制和晚期抑制的組別存在一個參數consistency的統計差異。 本研究的發現顯示，背內側紋狀體在機率學習的早期更為重要，且在早期抑制背內側紋狀體會導致較差的學習結果，而這樣的結果很可能是影響了reward-noreward model中的consistency所導致。根據此篇研究，使用不同的定義來切割機率學習的不同階段，或許是釐清機率學習這樣複雜行為的一個研究方法。

Probabilistic learning is the process of acquiring knowledge through trial and error to choose the optimal alternative from a set of alternatives with varying probabilities of receiving a reward or punishment. This concept has been applied to Neuroscience, Cognitive Psychology, and Behavioral Economics to investigate the neural and computational mechanisms underlying learning and decision-making. Understanding how people learn from probabilistic feedback has implications for improving decision-making processes in real-world situations. Probabilistic learning tasks are essential for adjusting to daily life, but individuals with cognitive abnormalities demonstrate distinct patterns in these tasks in comparison to healthy controls. Compared to animal studies, human research has many limitations. The mouse is a good model organism in science, which offers advantages for studying choice behavior, such as genetic tractability, small size, naturalistic behaviors, and insights into neural mechanisms. In this study, a handmade chamber controlled by MATLAB and Arduino systems was utilized. Three phases comprised the 2-choice probabilistic learning task: the shaping phase, the probabilistic acquisition phase, and the testing phase. In the first experiment, TRAP2 mice were used to examine the dynamics of the following 4 selected brain regions (including the medial orbitofrontal coretexcortex (MO), the prelimbic cortex (PrL), dorsal medial striatum (DMS), and dorsal lateral striatum (DLS) in probabilistic learning based on previous studies. The mice were presented with varying probabilities on the left and right sides to receive a reward. During the acquisition phase, the reward probability was 80% for one choice and 20% for the other side. The mice underwent three blocks of 25 trials each day, and the acquisition criteria is that the overall average accuracy is above 75%. Taking advantage of TRAP2 mice, neurons that were active on the first and last day of acquisition phase were labeled. The results of the first experiment indicated that there were significantly increased neural activities in the MO and DMS on the first day of acquisition phase whereas no significant difference was found on the last day. In contrast, there was no significant difference on either the first or the last day in the PrL and DLS. For further manipulation, we analyzed the choice strategy ratio to determine a time point that divided the acquisition procedure into the early and late phases. Because previous researches demonstrated that mPFC has direct projection to DMS, we took advantage of chemogenetic to selectively inhibit DMS activity during different time point in the acquisition phase in the second experiment to evaluate the time and functional significance of the DMS. There were 4 groups in Experiment 2. The all-time inhibition group is inhibited chemogenetically throughout the entire acquisition phase, whereas the early-inhibition group and late-inhibition group were only inhibited chemogenetically during the early and late phases, respectively. The DMS activity level of the control group was not inhibited throughout the acquisition. The all-time inhibition and early-inhibition groups required more learning trials to achieve the acquisition criteria than the late-inhibition and control groups. Our result indicated that DMS is important in the early phase of probabilistic acquisition. Besides, we fitted the behavioral data to a reward-no reward model and discovered a statistically significant difference between early- and late-inhibition on a parameter for consistency. Together, these findings revealed that DMS plays an important role at the beginning of probabilistic acquisition, and that inhibition of DMS in the early phase resulted in a worse acquisition outcome and lower consistency. Accordingly, using different definitions to separate the different stages of probabilistic learning may be a research method to clarify complex behaviors such as probabilistic learning.