A new theory of population coding in the cerebellum

小脑群体编码理论 为了准确地运动，大脑依赖于内部模型来预测运动的感觉结果。 命令.这一观点的证据来自人类行为实验[1-7]和动物损伤研究[8- 10]。 11]，这表明形成内部模型的关键结构是小脑。然而，在小脑中， 通常很难将单个浦肯野细胞（P细胞）的尖峰活动与行为联系起来：而对于某些任务， 像平滑追踪眼球运动一样，P细胞的活动是眼球速度的简单函数[12]，对于大多数其他 例如扫视[13，14]、手腕运动[15]或手臂运动[16-19]等运动，很难将其与 个体P细胞的活动对行为的影响。小脑的解剖学表明，P细胞以小群体的形式组织， [20]一个是一个是一个的。这种解剖学意味着， 小脑的计算单位不是单个P细胞，而是一群P细胞，它们聚集在一起， 单个输出神经元。因此，小脑中的群体编码具有特定的解剖学意义： 收敛到单个输出神经元上，一起编码行为的一个方面[21]。关键问题是 识别活脑中每个种群的成员。最近，我们展示了一种方法来解决这个问题， 问题[22]：共享相同复杂尖峰调谐的P细胞可能属于相同的群体。然而，在这方面， 复杂尖峰调谐识别是异常困难的：复杂尖峰是具有可变 波形持续时间。事实上，目前的方法依赖于复杂尖峰的手动标记， 不能扩展到多触点探头。在这里，三个专门研究绒猴、小鼠和猕猴的实验室 共同开发自动检测和归因复杂尖峰的算法。这些算法 专注于尖峰的频域分类，并将在高密度多触点探头上进行测试。 总之，算法和实验程序应该显着提高神经科学家的能力， 解决小脑中的群体编码问题，最终导致更好地理解 小脑学会精确地控制我们身体的运动。

A theory of population coding in the cerebellum In order to move accurately, the brain relies on internal models that predict the sensory consequences of motor commands. Evidence for this idea comes from human behavioral experiments [1-7] and animal lesion studies [8- 11], suggesting that the critical structure for forming internal models is the cerebellum. However, in the cerebellum it is often difficult to relate spiking activity of individual Purkinje cells (P-cells) with behavior: while for some tasks like smooth pursuit eye movements the activity of P-cells is a simple function of eye velocity [12], for most other movements such as saccades [13,14], wrist movements [15], or arm movements [16-19], it is difficult to associate activity of individual P-cells to behavior. Anatomy of the cerebellum suggests that P-cells organize in small groups, together projecting onto a single output nucleus neuron [20]. This anatomy implies that the fundamental computational unit of the cerebellum is not a single P-cell, but a population of P-cells that together converges onto a single output neuron. Thus, population coding in the cerebellum has a specific anatomical meaning: P-cells that converge onto a single output neuron together encode an aspect of behavior [21]. The critical problem is to identify the membership of each population in the living brain. Recently, we demonstrated a way to approach this problem [22]: P-cells that share the same complex spike tuning likely belong to the same population. However, identification of complex spike tuning is exceptionally difficult: complex spikes are rare events that have variable waveform durations. Indeed, the current approach relies on manual labeling of complex spikes, something that cannot be scaled to multi-contact probes. Here, three labs with expertise in marmosets, mice, and macaques have come together to develop algorithms that automate detection and attribution of complex spikes. These algorithms focus on the frequency-domain classification of spikes, and will be tested on high density multi-contact probes. Together, the algorithms and experimental procedures should significantly improve the ability of neuroscientists to tackle the question of population coding in the cerebellum, ultimately resulting in better understanding of how the cerebellum learns to precisely control movements of our body.