Sloan-Swartz Center for Theoretical Neurobiology and Center for Integrative Neuroscience, Department of Physiology, University of California , San Francisco, San Francisco, California 94143.
Howard Hughes Medical Institute and Department of Neurobiology, Duke University School of Medicine, Durham, North Carolina 27710.
eNeuro. 2014 Nov 12;1(1). doi: 10.1523/ENEURO.0004-14.2014. eCollection 2014 Nov-Dec.
A single extra spike makes a difference. Here, the size of the eye velocity in the initiation of smooth eye movements in the right panel depends on whether a cerebellar Purkinje cell discharges 3 (red), 4 (green), 5 (blue), or 6 (black) spikes in the 40-ms window indicated by the gray shading in the rasters on the left. Spike trains are rich in information that can be extracted to guide behaviors at millisecond time resolution or across longer time intervals. In sensory systems, the information usually is defined with respect to the stimulus. Especially in motor systems, however, it is equally critical to understand how spike trains predict behavior. Thus, our goal was to compare systematically spike trains in the oculomotor system with eye movement behavior on single movements. We analyzed the discharge of Purkinje cells in the floccular complex of the cerebellum, floccular target neurons in the brainstem, other vestibular neurons, and abducens neurons. We find that an extra spike in a brief analysis window predicts a substantial fraction of the trial-by-trial variation in the initiation of smooth pursuit eye movements. For Purkinje cells, a single extra spike in a 40 ms analysis window predicts, on average, 0.5 SDs of the variation in behavior. An optimal linear estimator predicts behavioral variation slightly better than do spike counts in brief windows. Simulations reveal that the ability of single spikes to predict a fraction of behavior also emerges from model spike trains that have the same statistics as the real spike trains, as long as they are driven by shared sensory inputs. We think that the shared sensory estimates in their inputs create correlations in neural spiking across time and across each population. As a result, one or a small number of spikes in a brief time interval can predict a substantial fraction of behavioral variation.
一个额外的尖峰就会产生影响。如图所示，在右侧面板中，小脑浦肯野细胞在 40ms 窗口（由左侧栅格中的灰色阴影指示）中发射 3（红色）、4（绿色）、5（蓝色）或 6（黑色）个尖峰时，平滑眼动的起始眼动速度的大小取决于此。尖峰序列富含信息，可以提取这些信息来指导毫秒级时间分辨率或更长时间间隔的行为。在感觉系统中，信息通常是根据刺激来定义的。然而，特别是在运动系统中，理解尖峰序列如何预测行为同样至关重要。因此，我们的目标是系统地比较眼球运动系统中的尖峰序列与单个运动中的眼动行为。我们分析了小脑绒球复合体中的浦肯野细胞、脑干中的绒球靶神经元、其他前庭神经元和外展神经元的放电。我们发现，在一个短暂的分析窗口中增加一个尖峰可以预测平滑追踪眼动起始的试验间变化的很大一部分。对于浦肯野细胞，在 40ms 的分析窗口中增加一个额外的尖峰，平均可以预测行为变化的 0.5 个标准差。最优线性估计器对行为变化的预测略优于短窗口中的尖峰计数。模拟结果表明，只要它们由共享的感觉输入驱动，单个尖峰能够预测行为的一部分的能力也会从具有与真实尖峰序列相同统计特性的模型尖峰序列中出现。我们认为，输入中的共享感觉估计在时间和每个群体中产生了神经尖峰的相关性。因此，在短时间间隔内的一个或少量的尖峰可以预测行为变化的很大一部分。
