Goal representations dominate superior colliculus activity during extrafoveal tracking

Abstract

The primate superior colliculus (SC) has long been known to be involved in saccade generation. However, SC neurons also exhibit fixation-related and smooth-pursuit-related activity. A parsimonious explanation for these seemingly disparate findings is that the SC contains a map of behaviorally relevant goal locations, rather than just a motor map for saccades and fixation. This explanation predicts that SC activity should reflect the behavioral goal, even when the behavioral response is not fixation or saccades, and even if the goal does not correspond to a visual stimulus. We tested this prediction by using a tracking task that dissociates the stimulus and goal locations. In this task, monkeys tracked the invisible midpoint between two peripheral bars, such that the visual stimuli were peripheral but the goal was foveal/parafoveal. We recorded from SC neurons representing peripheral locations associated with the stimulus or central locations associated with the goal. Most neurons with peripheral response fields did not respond differently during tracking than during passive viewing of the stimulus under fixation; most neurons with central response fields responded more during tracking than during fixation, despite the lack of a visual stimulus. Moreover, the spatial distribution of activity during tracking was larger than that during fixation or tracking of a foveal stimulus, suggesting that the greater spatial uncertainty about the invisible goal corresponded to more widespread SC activity. These results demonstrate the flexibility with which activity across the SC represents the location, as well as the spatial precision, of behaviorally relevant goals for multiple eye movements.

Figure 1.

Behavioral paradigms and recorded SC sites. A, In the extrafoveal tracking task, monkeys were presented with two peripheral bars moving together sinusoidally along the axis connecting their centers to the display center. The monkeys had to infer the invisible midpoint between the two bars and track it. Extrafoveal stimulation trials were identical except that there was also a central, stationary fixation spot. B, The left panel shows a mathematical model of the anatomical map of the SC (Ottes et al., 1986; Quaia et al., 1998) with the approximate locations of our recorded neurons indicated. The black squares indicate neurons that were assessed during the experiment to be peripheral enough to represent the location of the contralateral bar. The gray circles indicate neurons that were assessed to be central enough to represent locations associated with the invisible midpoint. The right panel shows the initial location of the contralateral bar. For the gray neurons in the left panel, peripheral bar locations (gray circles in the right panel) were at more than twice the preferred eccentricities of these neurons. The figure indicates sites on a single SC map, but we recorded from neurons in either the right or left SC.

Figure 2.

Sample neurons having the inferred goal of extrafoveal tracking inside their response fields (RF). A–F, Each column plots the responses of one neuron during memory-guided saccades (A, B), extrafoveal stimulation (C, D), and extrafoveal tracking (E, F). A, B, The left traces of each column show eye position (black for horizontal and gray for vertical) and neuronal responses (rasters and spike density) aligned on target onset. The right traces in each column show these same variables aligned on saccade onset. These two neurons exhibited saccade-related activity for small saccades and also exhibited fixation-related activity (data not shown). The thin dashed lines indicate SEM. C, D, Each column plots eye position (top two rows) and average spike density functions in the extrafoveal stimulation condition for each neuron individually. The black and gray traces show responses for the rightward and leftward starting phases of stimulus motion, respectively. The neurons were active, but there was no strong modulation in their activity level as a function of time or stimulus phase. E, F, Same as C and D, but for extrafoveal tracking. The neurons were more active (on average) during extrafoveal tracking than during extrafoveal stimulation, and their activity level was clearly modulated as a function of time. Red indicates veridical goal trajectory in each phase. For all horizontal and vertical eye positions, upward deflections indicate rightward and upward, respectively. Also, the eye position scale bars in E and F apply for the curves of C and D as well.

Figure 3.

Sample neurons having one of the peripheral bars inside their response fields (RF). This figure is organized like Figure 2. For the neuron in A, the contralateral bar was placed at 10.5, −2° (x, y), and for the one in B, it was at 8, 0° (x, y). Both of these locations were inside the respective RFs of the neurons, as indicated by the strong activity for saccades to similar locations. C, D, The neuron in the left column exhibited a visual burst at trial onset (arrow) (supplemental Fig. 1, available at www.jneurosci.org as supplemental material) followed by low-level, relatively unmodulated responses during maintained fixation. The second neuron was similar except that it had no visual response at trial onset. E, F, The responses of each neuron during extrafoveal tracking appeared similar to its responses during extrafoveal stimulation (for detailed comparison, see Fig. 4).

Figure 4.

Spatial tuning of the sample neurons of Figures 2 and 3 for stimulus or goal location. A, For the neurons in Figure 3, we recalculated responses during sustained extrafoveal tracking (black curves) and extrafoveal stimulation (gray curves) as a function of the retinal eccentricity of the contralateral bar along the tracking axis. The results of each neuron (left and right) demonstrate that the bar inside the response field (RF) did not recruit much more activity when it was being used to guide extrafoveal tracking than when it was passively viewed. B, The same analysis, but for the neurons of Figure 2, and now showing responses as a function of the location of the invisible midpoint between the two bars. 0° on the x-axis indicates the vertical meridian. During extrafoveal tracking (solid black curves), each neuron increased its responses when the goal occupied retinal locations approaching the center of the contralateral RF (estimated with the dotted line from visually guided saccade data), and decreased it otherwise. During extrafoveal stimulation (gray curves), there was no apparent change in the responses of the same neurons with changes in the (ignored) midpoint location. Error bars denote SEM.

Figure 5.

Spatial tuning for goal location that was consistent across different experimental conditions. Each row shows data from a single neuron. The left panel of each row plots responses as a function of the instantaneous retinotopic location of the inferred movement goal during extrafoveal tracking. As with other analyses, this analysis excluded activity in the vicinity of catch-up saccades. The right panel of each row plots saccade-prelude activity as a function of delayed visually guided saccade end points. A, Activity of the sample neuron of Figures 2, A, C, and E, and 4B, left. B, Activity of the sample neuron of Figures 2, B, D, and F, and 4B, right. C, D, Activity of two additional neurons from our group with central response fields. As can be seen, each neuron exhibited a preference for a particular retinotopic goal location during extrafoveal tracking, and this preference was correlated with the preference of the neuron during saccades.

Figure 6.

Summary of population activity during sustained extrafoveal tracking and extrafoveal stimulation. Extrafoveal tracking responses are plotted against extrafoveal stimulation responses for peripheral (left) and central (right) neurons. Responses were measured when one of the bars (left) or the invisible midpoint between the two bars (right) was at the best location that drove each neuron, and these locations were matched retinally across the two eye movement conditions (i.e., comparisons of neuronal responses across conditions were made for similar locations of the peripheral visual stimuli). Most neurons with a stimulus inside their response fields (RF) did not differentiate between extrafoveal stimulation and extrafoveal tracking. However, most central neurons were more active during extrafoveal tracking, despite the absence of visual stimuli at their preferred eccentricities, than during extrafoveal stimulation. The filled symbols denote significant differences between tracking and fixation (p < 0.05, t test).

Figure 7.

Similarities and differences between extrafoveal tracking and foveal tracking. A, Sample eye position traces and neuronal responses during extrafoveal tracking from the same neuron of Figure 2, B, D, and F. Neuronal responses are now shown as average spike densities (with SEM envelopes) and as raw spike rasters. B, Responses of the same neuron when the midpoint between the two bars was shown explicitly with a small spot. The neuron was still modulated, but to a lesser extent. C, These observations were also made when we measured responses as a function of retinotopic goal location along the tracking axis. In both conditions, the neuron exhibited similar preference for goal location, regardless of whether the goal was visible or not. The curve for foveal tracking has fewer sampled positions on the x-axis because of the higher spatial precision of tracking in this condition. The red curves in A and B indicate veridical goal location, and the dotted black curve in C indicates our estimate of the response field (RF) profile of the neuron at the displayed locations.

Figure 8.

Spatial locus of SC activity immediately after trial onset in the extrafoveal tracking condition. We plotted average spike density functions for the first 200 ms of extrafoveal tracking trials (i.e., before tracking had started). Neurons with peripheral response fields and having one of the bars inside these fields often exhibited a transient burst of activity after bar onset (black curves). Neurons with more central response fields had relatively high levels of activity at trial onset (during maintained fixation of a small spot), and these neurons exhibited a transient reduction in their activity shortly after bar onset (gray curves). The histogram on the bottom shows first saccade latencies (to initiate tracking), and the data point labeled “>500” shows the average activity during sustained extrafoveal tracking (after any initiation saccades had completed). The inset shows the approximate recorded locations on the SC map of the two groups of neurons described in this figure. The thin curves and error bars denote SEM.

Figure 9.

Spatial distribution of SC activity during sustained extrafoveal tracking, foveal tracking, extrafoveal stimulation, and fixation. A, In the extrafoveal tracking condition, we binned all neurons according to their preferred eccentricity and then plotted the average firing rate in each eccentricity bin when the inferred movement goal was aligned with the center of gaze. The activity was centered on the fovea and could be described with a Gaussian function fit. The map on the right of the panel shows the putative distribution of SC activity based on this Gaussian fit, if circular symmetry is assumed (red is maximal activity; blue is minimal activity). B, Similar observations were made for the foveal tracking condition, except that the spatial extent of activity was smaller, suggesting a smaller population of active neurons in this condition (compare putative SC map to that in A). C, Extrafoveal stimulation had an even narrow distribution of SC activity. D, The smallest population of recruited SC neurons that we could estimate was that during fixation of a small spot with no other peripheral stimuli. The size of the activity distribution estimated in this condition was similar to that obtained previously [Munoz and Wurtz (1995b), their Figs. 7H, 12] with a similar behavior. E, Summary of the SD of the Gaussian function fits for all four conditions shown in A–D. This SD is shown in degrees of visual angle (left y-axis) and in millimeters of SC coordinates according to the mathematical model of Equation 1 (right y-axis).

Figure 10.

Behavioral correlate of the changes in the spatial extent of SC activity representing goal location. We computed the SDs of horizontal and vertical eye position relative to the tracked goal (i.e., the two-dimensional spatial precision of eye movements) and plotted them against the SD of the Gaussian function fits of SC activity shown in Figure 9. When a larger population of SC neurons was recruited to represent the behavioral goal, there was also a greater variability of two-dimensional eye position around this goal. Error bars denote 95% confidence intervals (across sessions), and the dashed gray line is the line of unity slope. Each data point on the x-axis (in increasing order) corresponds to the extrafoveal stimulation, foveal tracking, and extrafoveal tracking conditions, respectively.

Figure 11.

Assessing shifts in the population of active neurons during extrafoveal tracking, extrafoveal stimulation, and foveal tracking with shifts in the retinotopic location of the midpoint between the two bars. A, We performed an analysis similar to that of Figure 9 but for different retinotopic goal locations during extrafoveal tracking. We plotted the activity of our neurons when the goal was at a 1.5° eccentricity relative to gaze on either the contralateral (gray) or ipsilateral side (black), along the axis of motion. Our neurons changed their activity with goal location (for samples illustrating this, see Figs. 2, 5) in a manner that was consistent with the entire center of mass of the active population shifting to reflect this goal location. B, The same analysis on the same set of neurons revealed much smaller changes in neuronal activity during the extrafoveal stimulation condition when the monkeys could ignore the (invisible) midpoint. C, During foveal tracking, we again observed shifts in the center of mass of the active population as the goal (now visible) shifted retinotopically. The smaller range of tracking errors in this condition meant that, to have enough data to perform this analysis, we could only test for shifts associated with 1.25° goal locations as opposed to 1.5°.

Figure 12.

Activity of SC neurons during the extrafoveal tracking blink condition. A, For nine neurons with peripheral preferred eccentricities, we recorded activity while one of the bars was transiently blanked for 200 ms (top schematic). The bar that was blanked was either the one inside the response field of the neuron (contra blink, dashed oval in schematic) or the one outside it (ipsi blink). The top row of data show the average normalized eye velocity traces for the nine neurons in the baseline extrafoveal tracking condition (gray) and in the blink condition (black). Each trace is the average for one neuron, with normalization done based on the gray curves. Eye velocity was reduced by bar blanking, but it did so similarly regardless of which bar was removed (compare contra and ipsi blink plots). The second row of data plots the average normalized spike density curves across neurons (with SEM envelopes) when the contra and ipsi bars were blanked (black curves). The gray curves show the average responses of the same neurons when no bars were blanked. When the contra bar was blanked, a visual burst occurred shortly after the bar reappeared (arrow). B, The same analyses were made for 10 neurons representing the invisible goal location. These neurons maintained their activity during the blank interval, regardless of which bar was removed (compare black and gray neuronal responses; arrow during blink). When the contra bar was removed, the neurons also exhibited a short-lived reduction after the peripheral bar reappeared (arrow after blink). RF, Response field.