Figure 10.

Simulation of neuronal population activity in motor cortex. We simulated a 2-D, center-out task with different movement speeds and directions. The population activity was modeled as a sum of N motor-cortical neurons, the discharges of which consisted of (1) a linear speed tuning (scaled by factor S), (2) a velocity cosine tuning (scaled by factor V) with a random PD and (3) a Gaussian noise term. We investigated the SNR (y-axis) of directional tuning (blue) and speed tuning (red) as a function of the neuronal population size N (x-axis). (A) Noise model indicated a baseline SNR in absence of any speed or velocity tuning. (B) Velocity tuning model. The SNR of directional tuning was independent of the population size. The speed-tuning SNR remained at the baseline level. (C) Model with velocity and speed tuning. For small populations, the directional SNR (blue) dominated the speed SNR (red), as observed in SUA studies. The situation reversed in large neuronal populations (N > 10.000) and the speed SNR dominated that of movement direction. The directional SNR even decreased for large populations (N = 10⁶), because the strong speed effect masked the cosine velocity tuning. (D) Analysis of the reversal points (color-coded), defined as minimum population size where speed SNR is significantly greater than the directional SNR, as a function of model parameters. The velocity cosine tuning was fixed (V = 1) and the speed tuning strength S was varied together with the task-unrelated Gaussian noise level σ. In summary, the model showed that the tuning properties of neuronal populations might be quite different, depending on their size, from the level of SUA.