Activity in LIP, But not V4, Matches Performance When Attention is Spread

Abstract

The enhancement of neuronal responses in many visual areas while animals perform spatial attention tasks has widely been thought to be the neural correlate of visual attention, but it is unclear whether the presence or absence of this modulation contributes to our striking inability to notice changes in change blindness examples. We asked whether neuronal responses in visual area V4 and the lateral intraparietal area (LIP) in posterior parietal cortex could explain the limited ability of subjects to attend multiple items in a display. We trained animals to perform a change detection task in which they had to compare 2 arrays of stimuli separated briefly in time and found that each animal's performance decreased as function of set-size. Neuronal discriminability in V4 was consistent across set-sizes, but decreased for higher set-sizes in LIP. The introduction of a reward bias produced attentional enhancement in V4, but this could not explain the vast improvement in performance, whereas the enhancement in LIP responses could. We suggest that behavioral set-size effects and the marked improvement in performance with focused attention may not be related to response enhancement in V4 but, instead, may occur in or on the way to LIP.

Figure 1.

Change detection task. After fixating a central spot, 1, 2, 4, or 8 oriented bars were flashed in a circular array around the fixation point, with one bar placed in the center of the receptive field (RF) of the neuron symbolized by a dashed circle. The animals had to keep fixation during the presentation of this array (array 1). After a gap of 100–300 ms, the oriented bars reappeared (array 2) but one of the bars could have changed orientation by rotating 90° (change trial), in which case the animals had 1 s to make an eye movement to the rotated bar to be rewarded (saccade indicated by a dashed arrow). If no rotation occurred, the animals had to keep fixation to be rewarded (No change trial). The right panels show the arrangement of the different set-sizes: 1 bar (S1), 2 bars (S2), 4 bars (S4), and 8 bars (S8).

Figure 2.

Behavioral performance in the change detection task. (A, B) Mean percentage of correct change trials (hits) is plotted as a function of set-size (number of stimuli in the array) for the individual animals during the V4 sessions (A) and the LIP sessions (B). (C, D) Mean percentage of correct no-change trials (correct rejections: CR) is plotted as a function of set-size for the V4 sessions (C) and the LIP sessions (D). (E, F) Mean reaction times for the individual animals during the V4 sessions (E) and the LIP sessions (F). All error bars represent SEM. Different symbols indicate each animal’s performance.

Figure 3.

V4 and LIP neuronal responses to set-size during the change detection task. (A, B) Examples responses from single cells in V4 (A) and LIP (B) from all completed trials are plotted as a function of time from both array 1 and 2 onsets for different set-sizes indicated by the different colors. (C, D) mean population responses for the subset of neurons tested with all 4 set-sizes for V4 (C) and LIP (D). Horizontal dark bars on the x-axis indicate significant bins in which there was a significant main effect of set-size (P < 0.01, ANOVA using 100 ms bins every 1 ms; F[3,35] for V4 and F[3, 65] for LIP).

Figure 4.

Mean responses of the V4 (black triangles) and LIP (gray circles) populations as a function of set-size during 2 distinct temporal periods after array 1 onset. Each point represents the mean (± SEM) response of the averaged responses for all neurons tested with that particular set-size (n = 93, 137, 173, and 82 in V4, and n = 126, 145, 157, and 106 in LIP, for S1, S2, S4, and S8, respectively). Solid lines represent responses 50–150 ms after array 1 onset, while dashed lines represent mean responses 200–400 ms after array 1 onset.

Figure 5.

Response to array 2 and metrics of how well the populations could differentiate change from no-change trials. (A, B) The mean population responses from all trials in which the change occurred in the receptive field (black) and from all no-change trials (gray) as a function of time aligned on the array 2 onset for V4 (A) and for LIP (B). The shaded area shows the window in which the 2 responses were compared statistically (see text). (C, D) ROC analysis. Mean area under ROC curves for different set-sizes from responses in the V4 population (C) and in the LIP population (D). Each ROC curve was computed using the responses from change trials in which the preferred orientation was presented in the receptive field and no-change trials in which the nonpreferred orientation was presented in the receptive field. A value of 0.5 indicates chance, while a value greater than 0.5 indicates higher responses to change trials compared with nonchange trials. Vertical colored dashed lines show when the value became significantly different from 0.5 (permutation test) and the solid black line on the x-axis shows times when the means among the 4 traces were significantly different (P < 0.01, ANOVA, degrees of freedom 464 in V4 and 513 in LIP). (E, F) Mean ± SEM difference indices computed from single cells in V4 (E) and LIP (F). A difference index of 0 indicates that the neuron’s mean responses in the 2 conditions were the same, while a positive index indicates that the neuron’s mean response in the change condition was greater than in the no-change condition. In all panels, arrows on the x-axis represent the grand mean reaction times for each specific condition.

Figure 6.

Response to array 2 as a function of behavioral outcome for all neurons in which all 4 set-sizes were tested. (A) The mean response of 36 neurons in V4 from hit trials (thick lines) and correct rejection trials (thin lines) plotted as a function of set-size. (B) The mean response of 66 neurons in LIP from hit trials (thick lines) and correct rejection trials (thin lines) plotted as a function of set-size. (C) The responses from hit trials in LIP aligned by saccade onset.

Figure 7.

Effect of biasing attention to one location when 4 items were in the array. (A, E) Open circles show the mean (±SEM) percentage of hit trials plotted against the mean (±SEM) false alarm rate for set-sizes of 2, (S2), 4 (S4), and 8 (S8) in the location that would later have the high reward associated with it from V4 sessions (A) and LIP sessions (E). The solid points show the mean (±SEM) percentage of hit trials plotted against the mean (±SEM) false alarm rate at the high reward location (S4h). (B, F) Similar plots to (A) and (E), but for performance at the locations away from the high reward location in standard trials (S2, S4, and S8) and in the high reward condition (S4h) from V4 (B) and LIP (F) sessions. (C, G) Measures of criterion from the V4 (C) and LIP (G) sessions. Open black circles show criterion measures from all locations in all sessions. Red points show criterion measures from the high reward location in the S4 condition (open circles) and in the high reward block (closed diamond). Blue points show the criterion measures from the locations away from the high reward location in the S4 condition (open circles) and in the high reward block (closed diamond). (D, H) Mean responses in array 1 from blocks with 4 items in the array (S4) and from blocks with the high reward condition (S4h). The shaded area shows the window in which the 2 responses were compared statistically (P-values from the Wilcoxon sign-rank test are presented).

Figure 8.

Discriminating change from no-change trials under conditions of spread and biased attention. (A, B) ROC analysis. Mean area under ROC curves for the standard block of set-size 4 (S4) and the block in which the animal received a higher reward for correct saccades to the receptive field (S4h) from responses in the V4 population (A) and in the LIP population (B). Vertical dashed lines show when the value became significantly different from 0.5 (permutation test) and the solid black line on the x-axis shows times when the means between the 2 traces were significantly different (P < 0.01, ANOVA; n = 173 and 86 for S4 and S4h in V4 and n = 157 and 108 for S4 and S4h in LIP). (C, D) Mean ± SEM difference indices computed from single cells in V4 (C) and LIP (D) from the same 2 conditions. Arrows on the x-axis represent the grand mean reaction times for each specific condition. (E, F) The mean response of 36 neurons in V4 (E) and 66 neurons in LIP (F) from hit trials (thick lines) and correct rejection trials (thin lines) in the S4 and S4h conditions.