French Andrew S, Pfeiffer Keram
Department of Physiology and Biophysics, Dalhousie University, PO Box 15000, Halifax, Nova Scotia, B3H 4R2, Canada.
Department of Behavioral Physiology and Sociobiology, University of Würzburg, Biocenter Am Hubland, 97074, Würzburg, Germany.
Biol Cybern. 2018 Oct;112(5):403-413. doi: 10.1007/s00422-018-0763-0. Epub 2018 Jun 18.
In a previous study, we used linear frequency response analysis to show that naturalistic stimulation of spider primary mechanosensory neurons produced different response dynamics than the commonly used Gaussian random noise. We isolated this difference to the production of action potentials from receptor potential and suggested that the different distribution of frequency components in the naturalistic signal increased the nonlinearity of action potential encoding. Here, we tested the relative contributions of first- and second-order processes to the action potential signal by measuring linear and quadratic coherence functions. Naturalistic stimulation shifted the linear coherence toward lower frequencies, while quadratic coherence was always higher than linear coherence and increased with naturalistic stimulation. In an initial attempt to separate the order of time-dependent and nonlinear processes, we fitted quadratic frequency response functions by two block-structured models consisting of a power-law filter and a static second-order nonlinearity in alternate cascade orders. The same cascade models were then fitted to the original time domain data by conventional numerical analysis algorithms, using a polynomial function as the static nonlinearity. Quadratic models with a linear filter followed by a static nonlinearity were favored over the reverse order, but with weak significance. Polynomial nonlinear functions indicated that rectification is a major nonlinearity. A complete quantitative description of sensory encoding in these primary mechanoreceptors remains elusive but clearly requires quadratic and higher nonlinear operations on the input signal to explain the sensitivity of dynamic behavior to different input signal patterns.
在之前的一项研究中，我们使用线性频率响应分析表明，对蜘蛛初级机械感觉神经元进行自然主义刺激所产生的反应动力学与常用的高斯随机噪声不同。我们将这种差异归因于动作电位从感受器电位的产生，并提出自然主义信号中频率成分的不同分布增加了动作电位编码的非线性。在这里，我们通过测量线性和二次相干函数来测试一阶和二阶过程对动作电位信号的相对贡献。自然主义刺激使线性相干向低频转移，而二次相干始终高于线性相干，并随着自然主义刺激而增加。在初步尝试区分时间相关和非线性过程的顺序时，我们通过两个块结构模型拟合二次频率响应函数，这两个模型由幂律滤波器和交替级联顺序的静态二阶非线性组成。然后，使用多项式函数作为静态非线性，通过传统数值分析算法将相同的级联模型拟合到原始时域数据。具有线性滤波器后跟静态非线性的二次模型比相反顺序更受青睐，但显著性较弱。多项式非线性函数表明整流是主要的非线性。对这些初级机械感受器中感觉编码的完整定量描述仍然难以捉摸，但显然需要对输入信号进行二次及更高阶的非线性运算，以解释动态行为对不同输入信号模式的敏感性。
