The effect of attention on neuronal responses to high and low contrast stimuli

Abstract

It remains unclear how attention affects the tuning of individual neurons in visual cerebral cortex. Some observations suggest that attention preferentially enhances responses to low contrast stimuli, whereas others suggest that attention proportionally affects responses to all stimuli. Resolving how attention affects responses to different stimuli is essential for understanding the mechanism by which it acts. To explore the effects of attention on stimuli of different contrasts, we recorded from individual neurons in the middle temporal visual area (MT) of rhesus monkeys while shifting their attention between preferred and nonpreferred stimuli within their receptive fields. This configuration results in robust attentional modulation that makes it possible to readily distinguish whether attention acts preferentially on low contrast stimuli. We found no evidence for greater enhancement of low contrast stimuli. Instead, the strong attentional modulations were well explained by a model in which attention proportionally enhances responses to stimuli of all contrasts. These data, together with observations on the effects of attention on responses to other stimulus dimensions, suggest that the primary effect of attention in visual cortex may be to simply increase the strength of responses to all stimuli by the same proportion.

Fig. 1.

Task design. A: task design for the main attention experiment. The location to attend was signaled by either instruction trials or a location cue (yellow annulus) and a series of paired Gabors appeared in the receptive field. On each presentation, the contrast of the Gabors was randomly selected. The animal's job was to detect when a Gabor with a faster drift speed appeared at the cued location, while ignoring any speed changes at the other location. B: task design for the contrast-offset experiment. Gabors appeared at a third location outside the receptive field and the animal's attention was always directed to this location. Preferred and nonpreferred Gabors were presented in the receptive field with randomly selected contrasts that always differed by a factor of 2.

Fig. 2.

The effect of attention on contrast-response functions in middle temporal visual area (MT). A: peristimulus time histograms for a neuron's response to different contrasts. Black histograms are responses with attention to the preferred direction and gray histograms are responses with attention to the nonpreferred direction. The thick line on the x-axis shows the duration of stimulus presentation and gray shaded region is the time window used for calculating firing rates. B: contrast-response functions from the MT neuron. The solid and dotted lines are the best-fitted functions for responses with attention directed to the preferred and nonpreferred stimuli, respectively. Error bars are SE values. C and D: population averages and normalized population averages of responses from all 56 MT neurons. Each neuron's responses are normalized to its maximum mean firing rate. Same format as that in B.

Fig. 3.

Relative performances of the response gain model and the contrast gain model. A: Z-transformed partial correlations between the predictions of each model and the underlying data. Filled circles are neurons for which the prediction of one model is statistically significant and significantly better than the fit of the other model. Dotted lines mark regions where each model meets these criteria (P = 0.05). B: modulation indices for R_max and c₅₀. Black filled circles are neurons with significant changes in R_max (response gain). Gray filled circles are neurons with changes in both R_max and c₅₀. Open circles are neurons with no significant change. Error bars are 95% confidence intervals. An overwhelming majority of neurons was consistent with response gain and none of the significant changes in c₅₀ was in the direction predicted by contrast gain.

Fig. 4.

Dynamics of attentional modulation. A: population average histograms showing the time course of attentional modulation. Each pair of histograms represents the average response of all MT neurons to preferred and nonpreferred stimuli presented in their receptive fields at the given contrast. Same format as that in Fig. 2A. The earliest portions of the response showed little effect of attention. B: the ratio of responses in the two attentional states as a function of time. Responses to stimuli with different contrasts are plotted as a ratio of the responses to the attend-preferred and the attend-nonpreferred conditions. The attentional modulation did not have an instantaneous onset, but instead rose at a rate that increased with contrast. Although the modulation saturated at a common level for high contrast stimuli (a response gain), the modulation did not reach this level during the short stimulus presentation for low contrast stimuli.

Fig. 5.

The contrast-offset experiment. A: the biased competition model of Reynolds et al. (1999) predicts that shifting attention between preferred and nonpreferred stimuli in a neuron's receptive field should cause an effect that is indistinguishable from response gain, as was observed for neurons in MT. B: this model also predicts that attention will produce a contrast gain effect when attention is shifted between a single stimulus inside the receptive field and a distant stimulus. C: the model additionally requires that changing the relative contrast of preferred and nonpreferred stimuli in the receptive field (with attention held constant) will cause response gain without contrast gain. This stimulus manipulation was used to test whether the MT neurons have the property that the model requires. The plots in A–C were made using parameters given in methods. The x-axes are in log units and the vertical scales are arbitrary. D: contrast-response functions for an example neuron with preferred and nonpreferred stimuli in its receptive field. The relative contrast of the two stimuli was varied while attention was held constant. Filled circles and the solid line mark responses recorded when the preferred stimulus had twice the contrast of the nonpreferred stimulus. Open circles and the dashed line mark responses recorded when the nonpreferred stimulus had twice the contrast of the preferred stimulus. Unlike the prediction of the model, a change in the relative contrast of the stimuli shifted the contrast-response function left and right. E: the distribution of modulation indices for c₅₀ between relative contrast conditions in the contrast-offset experiment. A modulation index was computed to quantify the change in c₅₀ that resulted from changing the relative contrast of the stimuli (e.g., between the solid and dashed curves in D). The mean of the modulation indices was 0.32 (a factor of 1.94). F and G: population averages and normalized population averages of responses from 25 MT neurons to preferred and nonpreferred stimuli with different relative contrast. Each neuron's responses are normalized to its maximum firing rate. Same format as that in D.