Independent repetition suppression in macaque area V2 and inferotemporal cortex

Abstract

When a complexly structured natural image is presented twice in succession, first as adapter and then as test, neurons in area TE of macaque inferotemporal cortex exhibit repetition suppression, responding less strongly to the second presentation than to the first. This phenomenon, which has been studied primarily in TE, might plausibly be argued to arise in TE because TE neurons respond selectively to complex images and thus carry information adequate for determining whether an image is or is not a repeat. However, the idea has never been put to a direct test. To resolve this issue, we monitored neuronal responses to sequences of complex natural images under identical conditions in areas V2 and TE. We found that repetition suppression occurs in both areas. Moreover, in each area, suppression takes the form of a dynamic alteration whereby the initial peak of excitation is followed by a trough and then a rebound of firing rate. To assess whether repetition suppression in either area is transmitted from the other area, we analyzed the timing of the phenomenon and its degree of spatial generalization. Suppression occurs at shorter latency in V2 than in TE. Therefore it is not simply fed back from TE. Suppression occurs in TE but not in V2 under conditions in which the test and adapter are presented in different visual field quadrants. Therefore it is not simply fed forward from V2. We conclude that repetition suppression occurs independently in V2 and TE.NEW & NOTEWORTHY When a complexly structured natural image is presented twice in rapid succession, neurons in inferotemporal area TE exhibit repetition suppression, responding less strongly to the second than to the first presentation. We have explored whether this phenomenon is confined to high-order areas where neurons respond selectively to such images and thus carry information relevant to recognizing a repeat. We have found surprisingly that repetition suppression occurs even in low-order visual area V2.

Keywords: V2; inferotemporal; macaque; repetition.

Graphical abstract

Figure 1.

Task design. A: on each trial, while the monkey maintained gaze on a central fixation point (FP), 2 images were presented in succession in the visual field contralateral to the recording hemisphere. The gray squares indicate the 2 locations at which images could appear during recording in TE. In the trial illustrated here, the first (adapter) and second (test) images differed in both identity and location. B: during recording in V2, 1 of the possible locations was centered on the neuronal receptive field (RF) and the other possible location was symmetrically placed with respect to the horizontal meridian. An invisible occluder (OC) prevented impingement of images on the fixation point under conditions in which the RF was close to fixation. The location of the RF in this panel is the average across all neurons tested in this way. C: in a subset of late V2 sessions, images cropped to disks 3° in diameter were centered on the RF and 3° above it. D: on any given trial, the adapter and test could match or differ with respect to identity and could match or differ with respect to location. The test image always appeared in the lower quadrant. The borders of the panels (dashed or solid, blue or orange) indicate how curves representing population firing rate under the 4 conditions are coded in Figs. 2A, 4A and 8A.

Figure 2.

Repetition suppression in TE. A: population mean firing rate as a function of time during the trial under 4 conditions determined by whether the adapter was in the upper quadrant (different location) or lower quadrant (same location) and whether the test image was identical to the adapter (same image) or not (different image). B: counts of neurons in which firing rate under the same-image condition differed significantly from firing rate under the different-image condition. Counts are subdivided according to whether firing rate was lower under the same-image condition (suppression) or the different-image condition (enhancement). C: repetition suppression as a function of time when adapter and test were at different locations (orange) and at the same location (blue). The measure of suppression is the population mean of firing rate under the different-image condition minus firing rate under the same-image condition. Ribbon is ±SE. The underlying bars indicate the periods during which the measures were significantly different from 0 as determined by a cluster-based permutation test. The P value for each test is juxtaposed to the corresponding bar. Each triangle indicates, for the plot of corresponding color, the estimated time of onset of suppression, computed by the method diagrammed in the inset. This was the time of the zero intercept of a line connecting the beginning and end points of the rising phase of the smoothed response. The rising phase (red) was defined as ending at the peak of the response and beginning at the first local minimum before the rise toward the peak. Note that for a local minimum above or below 0, the time of the zero intercept will not be the same as the time of the local minimum itself. Images in this experiment were placed at locations depicted by gray squares in Fig. 1A. Curves were smoothed by convolution with a 10-ms Gaussian kernel. Analysis of data from individual animals established the consistency of trends evident in the combined data as shown in the Supplemental Material (https://doi.org/10.6084/m9.figshare.19131758.v1).

Figure 3.

The trough-rebound dynamic in TE was enhanced when the test image matched the adapter. A: the trough-rebound index for the response to the test image matching the adapter is plotted against the trough-rebound index for the response to the adapter itself. B: the trough-rebound index for the response to the test image matching the adapter is plotted against the trough-rebound index for the response to the test image not matching the adapter. In each plot, each point represents data from 1 neuron. The numbers are the counts of points above and below the identity diagonal.

Figure 4.

Repetition suppression in V2. A: population mean firing rate under 4 conditions. B: counts of neurons in which firing rate depended significantly on whether the test matched the adapter or did not. C: mean suppression as a function of time after test image onset. Images in this experiment were placed at locations depicted by gray squares in Fig. 1B. All conventions as in Fig. 2. Analysis of data from individual animals established the consistency of trends evident in the combined data as shown in the Supplemental Material (https://doi.org/10.6084/m9.figshare.19131758.v1).

Figure 5.

The trough-rebound dynamic in V2 was enhanced when the test image matched the adapter. A: the trough-rebound index for the response to the test image matching the adapter is plotted against the trough-rebound index for the response to the adapter itself. B: the trough-rebound index for the response to the test image matching the adapter is plotted against the trough-rebound index for the response to the test image not matching the adapter. All conventions as in Fig. 3.

Figure 6.

Repetition-related effects occur earlier in V2 than in TE. A: population suppression as a function of time after test onset with adapter and test at the same location. The TE curve is replotted from the blue curve in Fig. 2C. The V2 curve is replotted from the blue curve in Fig. 4C. B: cumulative frequency with respect to time after test onset at which neurons exhibited half-maximal suppression with adapter and test at the same location. The plots represent data from 93 TE neurons and 99 V2 neurons. Ten TE neurons and 14 V2 neurons were excluded because of failure to meet analysis criteria (see materials and methods). C: trough-rebound dynamic when the test matched the adapter in identity and location. The TE curve is replotted from the solid blue curve in Fig. 2A. The V2 curve is replotted from the solid blue curve in Fig. 4A. Arrows indicate times of trough minima. The curves in B and C were normalized to the value at time 0 and the maximum within a window spanning 0–350 ms. We show only the early phase of the response in these panels so as to make clear the timing of early events.

Figure 7.

V2 neurons respond with reduced strength to a test image matching the adapter in location even when it does not share the adapter’s identity. Poststimulus suppression is plotted against prestimulus suppression for 113 neurons. The reduction in response strength is apparent in the tendency for observations to lie above 0. The reduction is not simply a result of the response riding on a suppressed prestimulus baseline, as indicated by the observation that the intercept of the best-fit line lies well above 0 and by detailed analysis described in the text.

Figure 8.

Repetition suppression in V2 with small closely spaced images. A: population mean firing rate under four conditions. B: counts of neurons in which firing rate depended significantly on whether the test matched the adapter or did not. C: mean suppression as a function of time after test image onset. Images in this experiment were placed at locations depicted by gray disks in Fig. 1B. All conventions as in Fig. 2. Analysis of data from individual animals established the consistency of trends evident in the combined data as shown in the Supplemental Material (https://doi.org/10.6084/m9.figshare.19131758.v1).