|Home | About | Journals | Submit | Contact Us | Français|
Previous studies have shown that the visual responses of neurons in extrastriate area V4 are enhanced prior to saccadic eye movements that target receptive field (RF) stimuli. We used receiver-operator characteristic (ROC) analysis to quantify how well V4 neurons could discriminate stable RF stimuli targeted by visually-guided saccades or ignored during saccades elsewhere. We found that discrimination was transiently enhanced prior to saccades to RF stimuli whereas it was reduced prior to saccades elsewhere. Similar to what is observed during covert attention and after frontal eye field microstimulation, the changes in stimulus discrimination were due at least in part to changes in response magnitude. In addition, we found evidence of an increased reliability of responses when saccades were made to the RF stimulus. These results highlight the similarity of mechanisms driving covert spatial attention and the preparation of visually-guided saccades.
Many studies have examined the effects of covert visual attention on the responses of neurons within macaque area V4, where visual activity is biased in favor of attended stimuli (Moran and Desimone, 1985; McAdams and Maunsell, 1999; Reynolds and Chelazzi, 2004). The attention-driven changes in visual activity in this area are paralleled by the modulation observed during the selection of receptive field (RF) stimuli as targets for saccadic eye movements (Fischer and Boch, 1981; Moore, 1999; Mazer and Gallant, 2003), suggesting a similarity of mechanisms underlying covert and overt spatial attention (Rizzolatti et al., 1987; Moore et al., 2003). More recent studies have addressed this apparent similarity by employing electrical microstimulation to perturb neural activity within saccade-related structures to alter the performance of monkeys on attention tasks (Moore and Fallah, 2001; Moore and Fallah, 2004; Cavanaugh and Wurtz, 2004; Muller et al., 2005; Cavanaugh et al., 2006) or the visual activity within V4 (Moore and Armstrong, 2003). The latter studies have shown that the changes in V4 responses following brief pulses of microstimulation delivered to spatially corresponding frontal eye field (FEF) representations are indistinguishable from the changes observed during covert attention (Armstrong et al., 2006; Armstrong and Moore, 2007).
Among the similarities between the effects of covert attention and FEF microstimulation is that both alter the ability of V4 neurons to discriminate among RF stimuli. Although past studies of presaccadic modulation of V4 activity have argued that targeting of RF stimuli by saccades (i.e. during overt attention) also alters visual discriminability (Moore et al., 1998), this has not been demonstrated. We therefore reexamined the visual activity of V4 neurons during a task in which monkeys made saccades to either the RF stimulus or to another location. We used receiver-operator characteristic (ROC) analysis to quantify how well V4 neurons could discriminate the orientation of stable RF stimuli during saccade preparation. We found that discrimination was transiently enhanced prior to saccades to RF stimuli whereas it was reduced prior to saccades elsewhere. These changes in stimulus discrimination are due in part to changes in response magnitude, similar to what is observed during covert attention and after FEF microstimulation, and may also be due to an increase in response reliability. These results suggest that the modulation of visual responses observed in V4 during the preparation of saccades is driven by mechanisms similar to those giving rise to covert attention effects, and similar to those perturbed during FEF microstimulation.
Two male monkeys (macaca mulatta, 8–12 kg) were used in these experiments. All experimental procedures were in accordance with National Institutes of Health Guide for the Care and Use of Laboratory Animals, the Society for Neuroscience Guidelines and Policies. General surgical procedures have been described previously (Zipser et al., 1996).
Visual stimuli were displayed on a 34 × 27 cm Sony video monitor that was driven by a Number Nine graphics board (640 × 480) at a 60 Hz, non-interlaced, refresh rate. The video display was positioned 57 cm in front of the monkey. Visual stimuli consisted of light and dark bars with one of four orientations (0°, 45°, 90°, or 135° θ) presented at the center of a V4 neuron’s RF. The fixation spot was a small (0.25° diameter) circle displayed at the center of the video display but was displaced (>5.0°) and used as a saccade target on some behavioral conditions.
Monkeys performed a visually-guided, delayed saccade task which was initiated by fixation to within < 1.0° of the central fixation spot. Immediately following fixation, an oriented bar stimulus appeared in the RF of the neuron under study and remained there until the end of the trial. Following the onset of the RF stimulus, the monkey was required to maintain fixation for a delay of 0.5 to 1 s, while it waited for the appearance of a saccade target (0.25° diameter) at one of two locations distant from the RF. In two-thirds of the trials (away conditions), the target appeared, the fixation spot was extinguished and the monkey was rewarded for making a saccade to the target. In the remaining one-third of trials (toward condition), the saccade target did not appear. Instead, when the fixation spot was extinguished and the monkey was rewarded for saccades to the RF stimulus. Both conditions were identical until the cue to saccade (disappearance of the fixation spot) and were randomly interleaved. During all behavioral trials, eye position was measured via the scleral search coil method, and digitized at 200 Hz for offline analysis.
The activity of single V4 neurons was recorded via glass-coated platinum-iridium electrodes lowered into the dorsal surface of the prelunate gyrus. Neural activity was sampled at 32 kHz, digitized and stored. The waveforms of single neurons were isolated by offline clustering (DataWave Technologies).
For each neuron, the preferred orientation was defined as that which evoked the maximum response, and the non-preferred orientation was defined as that which evoked the minimum response, out of 4 possible orientations (0°, 45°, 90° and 135° θ). Neurons were considered orientation-selective if the preferred and non-preferred responses summed across the initial 600 ms of stimulus presentation (i.e. prior to saccade preparation) were significantly different using a t-test (significance level P < 0.05). Data from both the toward and away conditions were collapsed together to maximize the statistical power of the selectivity measurement. ROC analysis was carried out on the distributions of neuronal firing rates measured during the execution of the delayed saccade task. The areas under ROC curves were used as an index of stimulus discrimination and were calculated as in previous studies (Britten et al., 1992; Armstrong and Moore, 2007). Specifically, we computed the average firing rate in a moving 50 ms window, from RF stimulus onset to saccade onset. We then computed the probability that the firing rate in each stimulus condition exceeded a criterion. The criterion was incremented from 0 to the maximum firing rate, and the probability of exceeding each criterion was computed. Thus, a single point on the ROC curve is produced for each increment in the criterion, and the entire ROC curve is generated from all of the criteria. The area under the ROC curve is a normalized measure of the separation between the two firing rate distributions obtained with the preferred and non-preferred RF stimuli, and provides a measure of how well the neuronal response discriminates the two stimuli. Differences in ROC areas, at the population level, were assessed by way of nonparametric tests on paired samples.
The analysis of presaccadic activity during abortive saccades consisted of extracting all trials in which the monkey broke fixation and made a saccade (> 2°) either to a location within or near the RF stimulus (< 5.0°), or to another location. Abortive saccades were only considered if they occurred after the onset of the RF stimulus and before the offset of the fixation spot.
We studied the activity of 90 single neurons in area V4 of two monkeys performing a visually-guided, delayed saccade task in which the receptive field stimulus for a given neuron could be the target of a saccadic eye movement. On a given trial, the monkey made saccades either to a stable stimulus in the RF of a V4 neuron (toward) or to a target outside the RF (away). Of the 90 neurons recorded, 63 were orientation-selective and were used for further analyses of the effect of saccade preparation on stimulus discrimination. Figure 1 shows the average normalized response of orientation-selective neurons during both phases of the saccade task. Following stimulus onset, there was a clear difference in response magnitude elicited by the preferred and non-preferred orientation. However, this response difference appeared to diminish later in the trial, even though the stimulus in the receptive field remained present. Furthermore, as previously reported (Moore et al., 1998; Moore, 1999), the response difference appeared to increase within 100 ms of saccades directed to the RF stimulus, as compared to saccades directed elsewhere. Prior to saccades to targets outside the RF, the preferred and non-preferred response difference was relatively unchanged or appeared to be slightly diminished. The previous accounts of presaccadic modulation of V4 responses have noted both the apparent reemergence of selectivity prior to saccades into the RF and the apparent decline prior to saccades elsewhere (Moore et al., 1998; Moore, 1999). However the discriminability of visual responses was not explicitly quantified.
To better quantify the changes in stimulus discrimination during saccade preparation, an ROC analysis was performed on the responses of the 63 orientation-selective neurons during both the stimulus onset and the presaccadic phase of the trial (Figure 2). The ROC curve, computed from each neuron’s distribution of responses to the preferred and non-preferred orientations, quantifies the extent to which the responses discriminate between the two RF stimuli. The area under the ROC curve is equal to the performance expected of an ideal observer discriminating among stimuli based solely on the neuron’s response (Green and Swets, 1966), and it is used here as an index of visual response discriminability. The mean ROC area of the orientation-selective neurons is shown for both the stimulus-aligned and the saccade-aligned data sets. Following stimulus onset, the mean ROC areas for both toward and away conditions rose to ~0.70 (~70% performance), where they remained stable for 100 ms. Subsequently, the mean ROC area of both conditions declined over the ensuing 300 ms to ~0.57. Thus despite the continued presence of the oriented bar stimulus in each neuron’s RF during the delay period, the ROC areas were reduced by more than half of that measured during the initial response. During the latter part of the delay period (−350 to −200 ms relative to saccade onset), the mean ROC areas for both conditions remained stable and indistinguishable from one another (P > 0.3, Wilcoxon signed rank test). However, within the last 100 ms prior to saccade initiation, the ROC areas of the two conditions diverged sharply. Saccades to the RF stimulus were preceded by an increase in ROC area (P < 0.03, Wilcoxon signed rank test, −50 ms versus delay period) while saccades to targets in the opposite hemifield were preceded by a decline in ROC area (P < 0.03, Wilcoxon signed rank test, −50 ms versus delay period). In both cases, the peak increment/decrement in ROC area occurred within ~50 ms of saccade onset, at which time the mean ROC area of the two conditions differed significantly (P < 10−3, Wilcoxon signed rank test). Closer to, and following, saccade onset, the mean ROC areas of the two conditions converged again toward levels measured during the delay period. Thus, the dynamics in presaccadic stimulus discrimination depended on whether the RF stimulus was used as the target of the saccade. Importantly, the performance of the population at discriminating preferred from non-preferred stimuli was enhanced when those stimuli were targeted as compared to when they were non-targets.
The ability of the population of V4 neurons to discriminate between the preferred and non-preferred RF stimuli depends both on the mean response elicited by the two stimuli and on the response reliability. Presaccadic changes in stimulus discrimination could either be due to changes in response magnitude or changes in response reliability or both. To determine which is the case, we first examined the difference between the mean presaccadic responses to preferred and non-preferred stimuli and compared the difference in the toward condition to that of the away condition. As previously reported (Moore et al., 1998), saccades toward the RF stimulus were preceded by a significantly greater difference in preferred and non-preferred responses than saccades away (P < 0.04, Wilcoxon signed rank test). Thus, the improvement in presaccadic stimulus discrimination was at least partly due to an increased difference in the response magnitude of visual responses.
We measured the reliability of visuals responses in two ways. First, we examined the relationship between the mean presaccadic responses and the variance in response across trials for the population of selective neurons (Figure 3a). Mean response and variance data were taken from a 50 ms window centered at −50 ms (−75 to −25 ms) relative to saccade onset for both toward and away conditions. The data from the two conditions were then fit with power functions and compared as in previous studies (McAdams and Maunsell, 1999; Bichot et al., 2001; Armstrong and Moore, 2007). The relationship between mean response and variance did not differ between the two saccade conditions. Both the power terms (powertoward= 1.3, poweraway= 1.3; P > 0.7, Wilcoxon signed rank test) and the coefficients (coeff toward= 0.87, coeffaway= 0.94; P > 0.4) were statistically indistinguishable for saccades made toward or away from the RF. Thus, by this metric, it appeared that the improvement in stimulus discrimination prior to saccades to the RF stimulus was not due to an increase in response reliability, suggesting that it was due only to changes in the response magnitude.
However, a recent study reported evidence of an increased reliability of V4 responses during covert attention as measured by the Fano factor (Mitchell et al., 2007). The Fano factor is simply the ratio of spike count variance to the mean spike count, and because it can be computed on an individual neuron basis, it may be a more sensitive measure of response variability. Thus, we also computed the Fano factor for responses in the toward and away case for the same 50 ms time window used to measure ROC areas (Figure 3b). Consistent with the findings of Mitchell et al. (2007), we found a difference between the two behavioral conditions. The Fano factor was significantly reduced prior to saccades toward the RF, as compared to the away condition (toward = 1.08; away = 1.23, p < 0.02, Wilcoxon signed rank test). Moreover, the magnitude of the decrease (median = 15.7%) was comparable in size to that reported during covert attention in the above study (~10%). Thus, the presaccadic changes in stimulus discrimination appeared to be due to changes in both the magnitude and the reliability of the visual response.
Although the cue to saccade in both toward and away trials included the offset of the fixation spot, only the latter condition included the onset of a visual target on the display. Thus, there remains the possibility that the changes in presaccadic activity, and stimulus discrimination, were primarily visually-driven, rather than due to differing saccade plans. While one may not expect a small (0.5 deg.) target flashed in the hemifield opposite of the RF to elicit the modulation observed, and indeed it is has been shown that surround effects in area V4 require sufficient size and contrast beyond that of the targets used here (Desimone et al., 1993), the dependence of the presaccadic modulation on stimulus events remains an open question. To address this question, we examined neuronal activity during trials in which the fixation offset (cue to move) had not yet occurred but the monkey initiated a saccade and aborted the trial. Data from the entire population of visually responsive neurons (n=90) were used to extract trials on which the monkey aborted the trial and made a saccade to the RF stimulus (n = 239 trials) or to a location outside of the RF (n = 471 trials). These abortive saccade trials were used to test the effect of saccade direction (toward vs. away) on presaccadic activity in the absence of any differences in the visual display (Figure 4). Similar to what is observed during correct toward and away trials in which the direction of the saccade is cued, the un-cued, abortive saccades to the RF stimulus are preceded by a significant enhancement in visual activity, compared to when they are made to locations outside of the RF (P < 0.01, one-sided Kolmogorov-Smirnov test). Thus, the preparation of a saccade appeared to be sufficient to drive the enhancement of V4 responses when RF stimuli were targeted.
We found that the discrimination of RF stimuli by V4 neurons, as measured by ROC analysis, is significantly altered prior to saccadic eye movements. The discriminability of orientation selective neurons was transiently increased within 50 ms of saccades directed to the RF stimulus. In contrast, stimulus discrimination was transiently decreased prior to saccades made to other locations. The opposing changes in presaccadic stimulus discrimination resulted in a substantial difference in the ability of V4 neurons to discriminate the RF stimulus depending on the direction of the imminent saccade. The changes in stimulus discrimination resulted in part from significant changes in the difference in response magnitude between preferred and non-preferred RF stimuli. This observation is consistent with those of previous studies of the effects of covert attention on V4 responses (Spitzer et al., 1988; McAdams and Maunsell, 1999a; McAdams and Maunsell, 1999b), as well as in studies that have evoked changes in visual responses of V4 neurons with FEF microstimulation (Armstrong and Moore, 2007). Both experimental manipulations improve V4 response discriminability, and do so at least in part via changes in response magnitude.
A recent study found that the effects of covert attention differ between regular-spiking and fast-spiking neurons in V4 (Mitchell et al., 2007). Fast-spiking neurons, which correspond to presumed locally-projecting interneurons, exhibit more dramatic effects of attention than regular-spiking neurons which should largely consist of distally projecting cells. Moreover, the effects of attention on the responses of both regular-spiking neurons and fast-spiking responses included an increase in response reliability, particularly in the latter class of neurons. This result differs from the report of McAdams and Maunsell (2007b) in which no changes in response reliability were observed using the population-level relationship between mean spike count and spike count variance as a measure. We also found no evidence of a difference in response reliability between the toward and away conditions using the population-level measure (figure 3a), similar to our observed effects of FEF microstimulation (Armstrong et al., 2007). However, using the Fano factor as the measure of reliability, we found evidence of increased reliability of visual responses in the toward condition. Thus, evidence of changes in response reliability appears to depend on the way in which it is measured. This dependence may be due to a greater sensitivity of the Fano factor analysis, which allows for a within comparison for each neuron, in contrast to the population-level analysis used in the other method. Thus, it appears that for both covert attention and saccade preparation, the increased discriminability of V4 visual responses occurs as a result of increased reliability as well as increased response magnitude.
The similarities between the effects of covert attention, saccade preparation (overt attention), and FEF microstimulation are consistent with the hypothesis that the filtering of visual signals within cortex is due at least in part to an interaction of visual and saccade-related signals during the preparation of saccades (Moore et al., 2003). This interaction could involve the direct feedback projections from the FEF to areas within extrastriate cortex (including V4) (Stanton et al., 1995), an influence of projections from the superior colliculus via the pulvinar nuclei (Wurtz et al., 2005), an influence of the lateral intraparietal area (LIP) (Goldberg et al., 2006), or all of the above. Nevertheless, the interaction appears to be sufficient to bring about perceptual benefits of the type observed during covert attention when saccades are withheld (Moore and Fallah, 2001; Moore and Fallah, 2004; Cavanaugh and Wurtz, 2004; Muller et al., 2005; Cavanaugh et al., 2006). This interaction also appears to facilitate the influence of visual target features, such as orientation (Moore, 1999) or direction of motion (Schafer and Moore, 2007), on the metrics of saccades when they are actually made. Furthermore, the timing of presaccadic increases in discriminability is consistent with the presumed synaptic delays of signals sent from extrastriate cortex to the FEF and the SC, which mediate the triggering of the saccade during overt attention. The enhancement of visual responses during overt and covert attention, and the increase in the discriminability of RF targets suggest that the interaction of visual and saccade-related signals is reciprocal and that it results in the simultaneous selection of target features and the appropriate/optimal saccade metrics with which to direct gaze. A challenge of future studies will be to identify the specific neurons that contribute the cross-areal interactions that give rise to the combined visual and oculomotor benefits of attentional selection.
We thank D.S. Aldrich for technical assistance. This work was supported by NIH Grant EY14924, the Pew Charitable Trust, the Sloan Foundation, and the McKnight Foundation. M.H. Chang was supported by a Stanford Bio-X graduate fellowship.
Publisher's Disclaimer: This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final citable form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.