Directional Tuning

We obtained for 55 cells directional tuning curves based on reverse correlation of the spike trains to the RDPs which changed directions every 80msec. Based on this method the average direction-selectivity index for these cells was 0.83 (± 0.04) (cf.

figure 3.1, A). The distribution of direction indices is similar to those previously reported for the population of neurons within area MT (Mikami, Newsome, and Wurtz, 1986; Snowden, Treue, and Andersen, 1992) . For 38 cells (69.1%) directional tuning could be reliably fit with a Gaussian model (chi square, p<0.05). Examples of Gaussian fits are shown in figure 3.2 (dashed red lines in upper panels, see below for details). The average standard deviation (sigma) of the Gaussian fitted tuning was 48.4^o (cf. figure 3.1,B). The isolated neurons were tuned to various preferred directions (cf. inlet in figure 3.1, C).

mean: 48.4º

sigma of gaussian tuning difference of pref. dir (stavg vs classical tuning)

Figure 3.1: A: Distribution of directional indices for cells which could be held long enough for the mapping experiment. DI’s are based on the peak and trough of the reverse correlogram obtained with the MRC stimulus. For the 55 cells the mean number of MRC tuning trials was 21.4 (range: 6-55) with each of twelve directions shown on average 11.4 times.B: Distribution of the standard deviation (sigma) of the circular Gaussian model fit to the values of the reverse correlogram at the latency of the peak response. Sigma reflects the degree of selectivity with values ranging near 45^o indicating a fine tuning typical for area MT. C: Deviation of the preferred direction of neurons measured with the MRC stimulus (spike-triggered average, or stavg) and with the classical method with longer duration of uniform motion. The inlet represents the distribution of preferred directions obtained for tuning curves based on all cells tested with the reverse correlation method. It illustrates that there is no sampling bias of neurons within a limited range of direction tuning.

Tuning correspondence of classical and MRC method

In order to validate the reverse correlation method on our own sample we also obtained tuning curves based on up to twelve directions with a standard method for 39 cells. For these cells we presented RDPs moving in only one direction during each trial and recorded the average firing rates during an 750msec interval (cf.

methods section, p. 54). These data were used to estimate the direction tuning of the cells in order to select the preferred and null direction used in the mapping experiment which commenced after this initial tuning test. With this standard method fewer tuning curves were obtained (compared to the MRC method) because we usually recorded these data as long as needed to determine the preferred and null direction of the cell. Therefore, we rarely recorded responses to more than eight directions in more than three trials per direction which were the inclusion criteria for the present analysis (i.e. at least 3 trials for each of at least eight different directions). Despite this small dataset neuronal responses mostly showed direction tuning similar to the direction tuning obtained with the reverse correlation method:

The direction index based on this standard method was on average 0.88 (± 0.04) correlating significantly with the DIs obtained with the MRC stimulus (r = 0.67, p<0.001). Similarly, circular Gaussian models (cf. p. 54) fit to both data sets provided almost identical estimates of the preferred direction of the neurons tested

B C

Figure 3.2: Examples of spike-triggered directional tuning. A-C: Tuning curves (up-per panels) and flow charts of reverse correlograms (lower panel) for three example cells. The color code of the flow charts reflects the probability that a particular direction (in different rows) is correlated with spike counts at time ∆t. High/ low values at negative time bins therefore reflect that the respective directions preceded spikes at the given time with a higher/lower probability. Slicing the flow chart cor-relograms at the peak response across directions (red line) reveals the directional tuning of the neuron. The upper panel illustrates these spike-triggered tuning curves (dotted red line: fitted Gaussian models; straight red line: observed values). Blue lines illustrate the tuning obtained for the same cell with the classical method (aver-age firing rates to directions of motion shown for extended time periods (>800ms)).

The high overlap of the tuning curves obtained with different measures validates the motion reverse correlation method. Grey straight and dashed lines reflects the mean and two times the noise (standard deviation) of the correlogram (computed as average of values in the positive time range), which provide a statistical tool to accept response estimates as statistically significantly different from random counts.

The examples illustrate an almost perfect overlap of directional tuning obtained with different methods. The third example (C) illustrates that the method also works well on single neurons with a notable baseline firing rate and a tuning of a lower signal-to-noise ratio (than in the first two cases). Note that in the examples shown, the spike-triggered average tuning was obtained in trials interleaved with the RF mapping experiment, while the classical method was always run prior to the mapping experiment.

with both methods: Figure 3.1, C, plots the distribution of the difference in degrees of the preferred direction derived from the Gaussian fit to the responses of the standard and the reverse-correlation method. Note that only cells were included for the comparison of the Gaussian parameters for which at least three trials could be obtained for at least eight direction. The histogram illustrates strong agreement on the direction tuning of individual cells (no significant difference between conditions, t-test, p>0.05). This point is further illustrated by the single neuron examples shown in figure 3.2. Analysis of the responses to the MRC stimulus also provided an estimate of the latency of individual neurons: The peak value of the reverse correlograms over the sample of 55 cells occured on average at -107ms (± 5.6ms)

indicating that most spikes are elicited 107ms after the preferred direction of the neuron was shown. This response latency estimate fits well to the average sensory latency reported for area MT, which is in the range of about 75 to 100msec for moving stimuli which simultaneously entails a motion-onset and a luminance-onset (87msec for large field moving RDPs, Raiguel et al. (1999); 94msec for moving bars, Raiguel et al. (1989); and about 75msec for flashed light bars, Schmolensky et al.

(1998)). In summary, the set of recorded neurons was fine tuned for the direction of motion, measured with different methods and at different time points and a high degree of consistency of results from both methods. The narrow tuning curves with an average Gaussian sigma of 48.4^o, together with a sensory peak-response latency of on average 107ms reflect properties expected for neurons in area MT.