Modulation of responses to moving small-field objects by variable background activity.

A. Raster plots show selected responses to a translating small-field square (dashed line: stimulus window) which are modulated by relatively low states (left subplot) and relatively high states (right subplot) of precedent background activity (BA). The relevant period of background activity may roughly cover the last 500 ms (grey shading) preceding the stimulus. B. Correlation analysis in CPU neurons. Spike counts within the first (second) 700 ms-window during stimulus presentation were plotted against those in the 700 ms-window of background activity directly preceding the stimulus. Data points close to or above the bisecting line correspond to trials that lacked the inhibitory response typical for CPU neurons; strongest responses are reflected by data points closest to the x-axis. Data were obtained from 3 CPU1- and 4 CPU2 neurons and cover 62 responses to the first stimulus in sequences of single small-field squares translating against a blank background; different plot marks correspond to different neurons. Plot marks are not scaled to the frequency of observations, thus identical values from the same experiment appear as a single data point. Dotted lines show linear regressions. C. Response amplitudes compared to different states of background activity. Box plots show distributions of relative response amplitudes (changes in spike rate relative to different levels of background activity). These were calculated for responses to the first stimulus in a respective sequence of small-field squares translating against a blank background (N responses from n neurons). For estimation of relative response amplitudes, we subtracted the spike rates of different levels of background activity (1s bins), obtained from the same respective cell throughout the entire experiment. These levels include the median and two additional quantiles of the background activity's spike count distribution that lie beyond the median in the same side of the distribution as expected for responses. In case of the inhibitory responses in CL- and CPU neurons, the 25% and 2.5% quantile of the spike rate distribution were used. For the excitatory responses of TB neurons, the 75% and 97.5% quantile provide relevant normalization. Indicated above each box-plot is the result of a two-sided sign test for zero median (* pcorr<0.05, ** pcorr<0.01, *** pcorr<0.001, ns not significant). Note that a non-significant test result here does not imply that responses did not involve a significant change in spike rate as compared to the local background activity that actually preceded the responses (see Fig 4A). While responses are relatively robust compared to the median levels of background activity (left columns of subplots), the additional response-type specific normalizations (middle and right columns) reveal that more extreme levels of background activity may result in masking of responses, as the difference to zero change decreases substantially and may become non-significant. Notches in box plots indicate 95% confidence interval of the median. Some outliers were truncated for the sake of better visualization.