Separate Signals for Target Selection and Movement Specification in the Superior Colliculus

See allHide authors and affiliations

Science  14 May 1999:
Vol. 284, Issue 5417, pp. 1158-1161
DOI: 10.1126/science.284.5417.1158


At any given instant, multiple potential targets for saccades are present in the visual world, implying that a “selection process” within the brain determines the target of the next eye movement. Some superior colliculus (SC) neurons begin discharging seconds before saccade initiation, suggesting involvement in target selection or, alternatively, in postselectional saccade preparation. SC neurons were recorded in monkeys who selected saccade targets on the basis of motion direction in a visual display. Some neurons carried a direction-selective visual signal, consistent with a role in target selection in this task, whereas other SC neurons appeared to be more involved in postselection specification of saccade parameters.

The primate SC plays a major role in the generation of saccades. Many neurons in the intermediate and deep layers of the SC fire a brief burst of action potentials starting approximately 20 ms preceding saccades of a particular range of directions and amplitudes; the region defined by the end points of such saccades comprises the “movement field” of an SC neuron. For each neuron, the location of the movement field in space varies systematically with the location of the neuron in the SC (1–4). Many SC neurons exhibit a “prelude” of activity related to the metrics of an impending saccade up to several seconds before the saccade is actually executed, implicating the SC in higher-level aspects of saccade planning (5–7). It is not known, however, whether these signals play a role in target selection, or simply reflect motor plans formed in response to selection processes elsewhere in the brain.

We investigated the role of the SC in target selection by recording prelude cells in monkeys (Macaca mulatta) trained to select one of two possible saccade targets contingent upon the direction of motion in a visual stimulus presented on a cathode ray tube monitor (Fig. 1) (8). For each SC neuron studied, the geometry of the display was arranged so that one of the targets (T1) lay inside the cell's movement field, while the other (T2) lay outside the movement field.

Figure 1

(A) Geometry of the visual display and (B) the timing of events in the direction discrimination task. The motion stimulus appeared within a circular aperture subtending 7° of visual angle and was usually presented at the center of gaze. Saccade targets were illuminated 300 ms after the monkey acquired the fixation point. Then, 500 to 900 ms later, a 2000-ms motion stimulus was presented, followed by an enforced delay period lasting 1000 to 1500 ms. Disappearance of the fixation point cued the monkey to make a saccade. Fixation breaks aborted trials. Eye position was continuously monitored by the scleral search coil technique (28). Horizontal and vertical eye position was sampled at a rate of 1 kHz and stored at a rate of 250 Hz for off-line analysis.

We recorded the activity of 96 intermediate and deep-layer SC neurons whose prelude activity was greater preceding T1 choices than T2 choices, permitting an experimenter to predict the monkey's decision several seconds before the saccade (9). Figure 2 illustrates the responses of a predictive SC neuron; three aspects of these responses are consistent with a role in target selection. First, predictive activity developed during the stimulus presentation interval while the monkey was formulating its judgment of motion direction. Second, predictive activity developed later and more gradually for low coherence trials than for high coherence trials, consistent with the longer psychophysical integration times required to discriminate weak motion signals (10, 11). Third, this cell lacked a saccade-locked burst, suggesting that it plays only a minor role in saccade execution.

Figure 2

Peristimulus and perisaccade time histograms for the discharge of a single superior colliculus neuron during direction discrimination at three different stimulus coherences. Data from correct trials only are displayed (except at 0% coherence for which correctness is arbitrary). Thick and thin curves illustrate the mean response preceding T1 choices and T2 choices, respectively. The left half-panel of each plot is aligned on motion stimulus presentation. The right half-panels show data from the same trials aligned on saccade initiation. The gap between the two half-panels reflects the timing variability between stimulus offset and saccade initiation. Vertical lines indicate the time of stimulus onset, stimulus offset, and saccade initiation, and the unit sp/s is spikes per second.

Neurons involved in target selection in our task may receive relatively direct sensory inputs concerning the direction of motion in the visual stimulus; the logic of the task dictates that such neurons should be excited selectively by motion flowing toward their movement fields. To test for the presence of such inputs in the SC, we looked for directional visual responses in blocks of trials when the monkey was rewarded for passive fixation (12). Random dot stimuli were presented within a circular aperture surrounding the center of gaze; the direction of coherent motion was either toward or away from the movement field of the SC neuron. Of the 96 choice-predicting SC neurons, 44 yielded directional responses: activity was significantly stronger when motion flowed toward the movement field than away from it (Mann-Whitney U-test: P < 0.05). No cell was significantly more active when motion flowed away from the movement field.

A possible criticism of these experiments is that, by force of habit, our monkeys may have planned saccadic eye movements covertly upon viewing the moving random dots even though they were not required to execute such movements. To control for this possibility, we measured visual direction tuning curves for 22 neurons when the monkey was required to plan saccades to a location outside of the movement field of the cell (Fig. 3A). In this condition, a single saccade target appeared early in the trial, and visual direction tuning curves were measured by presenting stimuli during the overlap period while the monkey awaited a “go” signal before executing the saccade (13). Because the monkey is able to plan the saccade from the beginning of the trial, it is unlikely that random dot motion during the overlap period would elicit covert saccade planning to a different, unrewarded location.

Figure 3

(A) Spatial design of a typical direction tuning measurement. Fixation was at the intersection of the axes. The gray disk shows the spatial extent of the stimulus aperture. The arrow depicts the average end point of 15 saccades elicited by electrical stimulation at the recording site; this provides an estimate of the movement field location of nearby cells. On each trial the monkey was required to make a saccade to a single target located far from the movement field of the cells at the recording site. (B) Averaged and raw responses of an SC neuron to eight directions of motion. Rasters illustrate raw responses to each of the eight directions. The polar plot depicts the average response as a function of motion direction. Background activity is represented by the circle at the origin of the polar plot. Vertical bars in the rasters delimit the 2000-ms stimulus presentation. The arrow indicates the estimated preferred direction of the cell. (C) Scatterplot of preferred direction against angle of electrically elicited saccades.

Direction tuning curves measured during the saccade task (14) did not differ from those measured during passive fixation trials in terms of preferred direction (Wilcoxon signed rank test, P > 0.5), tuning width (0.1 >P > 0.05), or amplitude of response (P> 0.5). The preferred direction of the cell to visual motion measured in the saccade task (arrow, Fig. 3B) corresponded well with the direction of the saccades elicited by electrical stimulation at the same site (arrow, Fig. 3A). This correlation held up well across all 22 neurons tested (Fig. 3C, circular-circular rank correlation coefficient: r = 0.77, P < 0.0001).

These direction-selective responses have not been documented previously in primate SC. An intriguing possibility is that, over the course of training, pathways between visual cortex and SC neurons involved with target selection are modified so as to mediate this learned association (15). Indeed, recordings in a monkey that had not been trained to associate particular directions of motion with particular saccade vectors revealed substantially fewer direction-selective SC neurons (5/35 versus 44/96: z-testP < 0.001) (16).

The two groups of prelude neurons may represent different neural processing levels, one involved in target selection and the other in the specification of saccade parameters. We analyzed the time course of predictive activity in each group of neurons (Fig. 4) (17, 18). In the direction-selective cells, predictive activity developed with a short latency and a rapid time course, suggestive of an early role in saccade planning. The magnitude of the predictive activity during the stimulus presentation was strongly modulated by the motion coherence (Spearman's r = 0.94: P < 0.025). Thus, the activity of these cells reflects not only the animal's decision, but also the strength of the signal upon which the decision is based (19, 20). Finally, these cells exhibited predictive activity up to 500 ms before the presentation of the 0% coherence visual stimulus (Fig. 4A, arrow; permutation tests: P < 0.01 at each time point) (21). This activity may reflect intrinsic bias states which can influence the animal's decision in the absence of strong sensory signals (22).

Figure 4

Ideal observer analysis on pooled data from (A) direction-selective neurons (n = 44) and (B) non–direction selective neurons (n = 52). Trials are aligned on the presentation of the motion stimulus (times 0 through 2 in the left half-panels) and saccade initiation (time 0 in the right half-panels). Curve color corresponds to motion coherence.

Non–direction selective cells differed from direction-selective cells in each of these respects. Predictive activity developed with a longer latency and slower time course (permutation tests: P < 0.025 and P < 0.01, respectively) (23), and was not modulated by stimulus coherence (Spearman's r = 0.26:P > 0.25). In addition, these cells did not exhibit predictive activity in the interval preceding the stimulus presentation (permutation tests: P > 0.1 at each time point).

Analysis of perisaccadic neural activity suggests that the non–direction selective cells are more closely involved with saccade execution. For saccades directed toward the target in the movement field, perisaccadic firing rates (recorded during an interval from 50 ms before until 25 ms after the saccade initiation) were almost three times greater in non–direction selective cells than in direction-selective cells (116 spikes/s versus 40 spikes/s; Mann-Whitney U-test: P < 0.0001).

The direction-selective cells we describe appear appropriate for implementing the association between motion stimuli and saccade vectors that is necessary for correct performance on our task. Mays and Sparks described a high-level class of SC neurons (“quasi-visual” cells) that appear to represent potential targets for saccadic eye movements, but are not linked obligatorily to execution of a saccade (5). Similarly, Basso and Wurtz described SC neurons whose activity reflects the probability that a saccade will be made into their movement fields (24). Either or both of these populations may overlap with our direction-selective neurons.

Our data suggest that at least two levels of processing related to saccade planning are present within the SC. Some cells possess a constellation of properties indicative of a high-level role in decision formation and target selection, while other cells are more directly linked to saccade execution. These two profiles appear to lie at opposite ends of a continuum rather than representing two distinct, nonoverlapping populations of cells. Previous studies have identified neurons in the lateral intraparietal area (LIP) and prefrontal cortex that carry signals appropriate for mediating decision formation, target selection, or saccade planning (15, 19, 20, 2527). It will be important to determine how these formal processes are distributed among the several brain areas and how neuronal populations in these areas interact to accomplish these tasks.

  • * To whom correspondence should be addressed. E-mail: bill{at}


Stay Connected to Science

Navigate This Article