Well-isolated V1 neurons were recorded from two rhesus monkeys while performing a task that required fast detection of a visual stimulus. The recordings were obtained by using an implantable, multielectrode/microdrive array with ultra-thin electrodes 30
that allowed us to study each neuron for several hours. The recording stability combined with the excellent spike isolation allowed us to measure the response properties from each neuron with great level of detail (Supplementary Fig. 1
). In addition to quantifying response modulations to spatial attention and task difficulty, we also measured orientation tuning, direction selectivity, spatial frequency tuning, receptive-field structure (by reverse correlation analysis), response latency, color selectivity, contrast sensitivity, spike width, interspike-interval distribution, response linearity (F1/F0) and spontaneous firing rate for each neuron. A main goal of this study was to characterize the functional properties of neurons that are modulated by spatial attention and task difficulty, so as to identify the specific types of neurons that mediate attentional effects.
Effects of spatial attention and task difficulty
During the recordings, monkeys held a bar and fixated a small cross while attending to one of five peripheral drifting gratings, which was cued at the beginning of each trial (). The five gratings had different spatial locations but they were all identical in orientation and direction of movement, spatial frequency, temporal frequency and size (matching the response properties of the neuron studied). After a randomized period of time that lasted 1.5–3 seconds, the cued grating changed color and the monkey indicated the color change by releasing the bar as fast as possible (shorter reaction times resulted in larger rewards; reaction times longer than 500 ms were not rewarded). Response modulations to spatial attention and task difficulty were measured within the 500 ms preceding the color change (1–2.5 seconds after the cue was turned off, , top). The color change could be either easy or hard to detect (due to the change in its color/luminance contrast) and it could take place either inside or outside of the neuron’s receptive field (, see Methods). The level of task difficulty was adjusted in each recording session to obtain reaction times that were significantly longer during the hard task than the easy task (P < 0.05 in Wilcoxon test). The cue was blocked for both spatial location and difficulty for 20 or more consecutive trials so that the monkeys could maximize their effort within a trial block (i.e. if the first trial was difficult to detect, the monkeys knew that the following trials would be also difficult and would need to increase their effort to maximize reward). We studied visual responses from 92 neurons at two or more different levels of task difficulty and two spatial locations of attention. We also measured in detail the response properties of each neuron (average recording duration: 1.6 hours; see Supplementary Fig. 1
Behavioral task and attentional response ratios measured in V1 single cells during hard and easy tasks
During the easy task, the monkeys detected the color change rapidly (average reaction time: 337 ± 47 ms) and made few mistakes (percentage of rewarded bar releases: 98%), but only 8 out of 92 V1 neurons (8%) were modulated by spatial attention (, top; see methods for data on each monkey). During the hard task, the reaction times were significantly higher (394 ± 61 ms, P < 0.0001 Wilcoxon test), errors were more frequent (percentage of rewarded bar releases: 86%) and 39 out of 92 neurons in V1 (42%) were modulated by spatial attention (, bottom; all cells modulated by attention during the easy task were also modulated during the hard task). We selected the cells that were significantly modulated by attention for further analysis (n = 39, red bars in , P < 0.05 in Wilcoxon test, see Methods for measurements of significance). Throughout the text, visual responses measured in the four different conditions of task difficulty and spatial attention will be referred to as E (responses during the easy task), H (responses during the hard task), I (responses while the focus of attention was inside the receptive field) and O (responses while the focus of attention was outside the receptive field).
The modulation of V1 visual responses by spatial attention and task difficulty are illustrated with two representative cell examples in . The top of the figure shows the peristimulus time histograms (PSTHs) of the visual responses driven by the last grating cycle before the color change (see for details about the temporal course of stimulus and response). Each PSTH shows responses to different levels of difficulty (easy in blue, hard in red and intermediate in black) and different spatial locations of attention (left: attention inside the receptive field; right: attention outside the receptive field). The panels below the PSTHs illustrate different functional properties measured in each neuron (i.e. spike width, interspike interval distribution, receptive field, direction selectivity and spatial frequency tuning). As illustrated in this figure, the most pronounced effect of increasing task difficulty in cell_a (from an easy to a hard task) was response enhancement when spatial attention was located inside the receptive field. In contrast, the most pronounced effect of increasing task difficulty in cell_b was response suppression when attention was located outside the receptive field. Cell_a and cell_b differed in several other response properties. Cell_b had wider spike waveforms and a tighter interspike-interval distribution than cell_a. Moreover, unlike cell_a, cell_b was direction selective, color selective (not shown) and it could not be visually driven with sparse noise stimuli (light and dark spots presented for 40 ms within the receptive field). Other properties such as spatial frequency tuning and Fourier harmonic F1/F0 ratio were similar in both cells.
Examples of two V1 cells whose responses were modulated by task difficulty and spatial attention
On average, the attentional ratio [(I − O)/(I + O)] of attention-modulated V1-cells was 0.09 during the hard task, which corresponds to a mean response enhancement of 22% [(I − O)/I * 100] when spatial attention was located inside versus outside the receptive field. In contrast, the average attentional ratio during the easy task was significantly lower, 0.03 (P < 0.0001 Wilcoxon test), which corresponds to a modest enhancement of just 7% when attention was inside versus outside the receptive field (). Clearly, increasing task difficulty made the attentional modulations much stronger within area V1, from a 7% modulation, which matches some V1 measurements in previous studies 21
, to a 22% modulation, which approaches the modulations measured in higher cortical areas 8, 10, 21
Spatial attention and task difficulty modulations of V1 visual responses
The effect of task difficulty on visual responses was spatially specific. Increasing task difficulty (from an easy to a hard task) enhanced the visual responses of cells with receptive fields at the focus of attention and suppressed the responses of cells with receptive fields outside the focus. We measured the difficulty ratio [(H − E)/(H + E)] of the responses from the attention-modulated cells at two spatial locations of attention (). When spatial attention was inside the receptive field, most difficulty ratios were positive (30 out of 39, P = 0.0008, Chi-square test) indicating that V1 neurons responded more strongly during the hard task than the easy task (response enhancement). In contrast, when spatial attention was located outside the receptive field, most difficulty ratios were negative (28 out of 39, P = 0.006, Chi-square test) indicating that V1 neurons responded less strongly during the hard task than the easy task (response suppression). The magnitudes of the response enhancement and response suppression were significantly correlated to each other (r = 0.56, P < 0.001, ). That is, neurons that showed the strongest response suppression when spatial attention was outside the receptive field showed the weakest response enhancement when spatial attention was inside the receptive field. Conversely, neurons that showed the strongest response enhancement when attention was inside the receptive field showed no response suppression when attention was outside the receptive field. This suggests that the response suppression and response enhancement driven by an increase in task difficulty could be mediated by different populations of neurons. We investigated this idea further by performing a detailed study of the functional properties of neurons modulated by attention.
Functional properties of V1 neurons
Intrigued by the differences in the response properties of cell_a and cell_b, we divided our population of attentionally modulated cells into two groups: difficulty-enhanced and difficulty-suppressed neurons. We called difficulty-enhanced neurons those cells that showed a net enhancement in visual response with task difficulty, such as cell_a (n = 20) and difficulty-suppressed neurons those that showed a net suppression of visual responses, such as cell_b (n = 19). By definition, difficulty-enhanced neurons generate the strongest visual responses when the task is hard (, left) and difficulty-suppressed neurons generate the strongest visual responses when the task is easy (, left). We noticed that difficulty-enhanced neurons, such as cell_a, were usually non-directional selective, had narrow spike-widths and broad interspike interval distributions, as illustrated by four cell examples in (middle and right). In contrast, difficulty-suppressed neurons, such as cell_b, were directional selective, had broader spike-widths and tighter interspike interval distributions (, middle and right).
Modulation by task difficulty in V1 cells
To quantify the relation between task-difficulty modulations and neuronal functional properties, we calculated a summed difficulty ratio for each cell as (Hs − Es)/(Hs +Es), where Hs = IH + OH and Es = IE + OE. A positive ratio indicates that the cell responded more strongly during the hard task than the easy task (independently of the location of attention) and a negative ratio indicates that the cell responded more strongly during the easy task. Consistent with the examples shown above, the summed difficulty ratio was correlated with the cell direction selectivity both in the sample of cells modulated by spatial attention (r = −0.58, P < 0.001, n = 39, ) and in the entire sample (r = −0.25, P < 0.02, n = 92, not shown). Moreover, the summed difficulty ratio was correlated with the cell’s spike width (r = −0.58, P < 0.001, ), and the peak of the interspike-interval distribution (r = 0.45, P < 0.004, ). We also found a correlation between the summed difficulty ratio and contrast sensitivity (r = 0.51, P <0.03) although the sample of cells tested for contrast was smaller (n = 19; ). Interestingly, the regression lines from some of these correlations seemed to link two separate clusters of cells rather than a continuum ().
Response modulations to spatial attention and task difficulty are correlated with the direction selectivity, spike width, interspike interval and contrast sensitivity of the cell
To estimate the significance of this observation, all the scatter plots from were tested for bimodality 31
(Hartigan tests were run in histograms with X-axes parallel to the regression lines). These tests revealed a significant bimodal distribution in the projection defined by direction selectivity and the summed-difficulty-ratio (, P = 0.018), supporting the notion that difficulty-enhanced and difficulty-suppressed neurons form two separate clusters. Interestingly, neurons whose responses were not modulated by spatial attention had intermediate properties between these two clusters ( and ). Other response properties were also compared between clusters but were not significantly different. These included mean firing rate, response linearity (F1/F0 ratio), color sensitivity, spatial frequency tuning (peak and bandwidth), orientation tuning (circular variance), receptive field size, response latency and visual responses to brief dark/light spots (Supplementary Fig. 2
Figure 6 Response properties that distinguish V1 cells classified as difficulty-suppressed (summed difficulty ratio < 0), difficulty-enhanced (summed difficulty ratio > 0) and non-modulated (no significant modulation by spatial attention as defined (more ...)
Functional differences between two populations of V1 neurons, difficulty-suppressed and difficulty-enhanced
Ideally, the attentional modulations should be measured at multiple levels of task difficulty to quantify the correlation between task difficulty and response magnitude. However, it was not technically possible to measure multiple levels of task difficulty and acquire detailed quantification of the response properties from each cell, due primarily to the amount of time that the animals were willing to work. To address this limitation, in a group of cells (n = 59) we gave priority to the measurement of attentional modulations over response properties and studied three or more levels of difficulty at two spatial locations. Within 22 attention-modulated cells in the group, seven of them were classified as difficulty-enhanced (summed difficulty ratio > 0) and six as difficulty-suppressed (summed difficulty ratio < 0). We calculated separate correlation coefficients for difficulty-enhanced and difficulty-suppressed neurons by measuring the mean firing rate at three different levels of difficulty (hard, medium and easy) and then normalizing the mean firing rate after dividing by the maximum mean rate obtained under the three conditions. Consistent with the results reported above, the mean firing rate of the difficulty-enhanced neurons was positively correlated with task difficulty only when the spatial attention was inside of the receptive field (attention inside the receptive field: r = 0.66, P < 0.0001; attention outside the receptive field: r = 0.07, P = 0.69). In contrast, the mean firing rate of the difficulty-suppressed neurons was correlated with attention difficulty only when attention was outside of the receptive field (attention inside the receptive field: r = 0.12, P = 0.55; attention outside the receptive field (r = −0.77, P < 0.0001; ).
The magnitude of the visual responses was correlated with the level of task difficulty