We hypothesized that neuronal responses to virtual self-movement would be enhanced during steering tasks. We recorded the activity of medial superior temporal (MSTd) neurons in monkeys trained to steer a straight-ahead course, using optic flow. We found smaller optic flow responses during active steering than during the passive viewing of the same stimuli. Behavioral analysis showed that the monkeys had learned to steer using local motion cues. Retraining the monkeys to use the global pattern of optic flow reversed the effects of the active-steering task: active steering then evoked larger responses than passive viewing.
We then compared the responses of neurons during active steering by local motion and by global patterns: Local motion trials promoted the use of local dot movement near the center of the stimulus by occluding the peripheral visual field midway through the trial. Global pattern trials promoted the use of radial pattern movement by occluding the central visual field midway through the trial. In this study, identical full-field optic-flow stimuli evoked larger responses in global-pattern trials than in local motion trials. We conclude that the selection of specific visual cues reflects strategies for active steering and alters MSTd neuronal responses to optic flow.
Visual motion processing contributes to self-movement perception by analyzing the patterned visual motion in optic flow (Gibson 1950). Optic-flow analysis supports precise heading direction discrimination in humans (Warren et al. 1988) and monkeys (Britten and van Wezel 1998). Such heading information is used during active navigation and contributes to the development of a cognitive map of the environment (Redish 1999). An actively engaged driver learns more about the environment than the similarly exposed passengers in a moving car (Golledge 1999). We hypothesized that active steering might influence the neuronal analysis of optic flow and tested for effects in macaque visual cortical neurons.
Neurons in medial superior temporal (MSTd) cortex respond to the large patterns of visual motion in optic flow (Tanaka et al. 1986; Komatsu and Wurtz 1988a). MSTd neurons show selective responses for the radial center of motion in optic flow (Duffy and Wurtz 1995; Lappe 1996) that contribute to neuronal population responses that represent the current heading direction (Page and Duffy 1999). MSTd neuronal responses are modulated by stimulus sequences to encode the heading direction through the environment (Paolini et al. 2000; Froehler and Duffy 2002).
Behavioral influences on MSTd neuronal responses have been seen in studies of pursuit eye movements (Komatsu and Wurtz 1988b; Ferrera and Lisberger 1997) altering the optic-flow responses of single neurons in monkeys (Komatsu and Wurtz 1988a) (Upadhyay et al. 2000) and middle temporal activation in humans (hMT+) (Goossens et al. 2006). These effects appear to reflect interactions between visual responses evoked by optic-flow and extraretinal signals associated with pursuit control and execution (Newsome et al. 1988; Bradley et al. 1996; Page and Duffy 2003; Churchland and Lisberger 2005). Such interactions might also occur when behavioral tasks link visual signals to appendicular motor control by engaging optic-flow responsive cortical motor control systems (Merchant and Georgopoulos 2006), effects that might be mediated by parietofrontal projections supporting closed-loop manual responses to ongoing visual stimuli such as optic flow. An important unknown in both the pursuit and appendicular control scenarios is the potential role of concomitant extrastriate attentional modulation (Seidemann and Newsome 1999) (Treue et al. 2000) of MSTd's optic-flow responses by the demands of the behavioral tasks.
We hypothesized that engaging the monkey in the manual control of the simulated heading direction in an optic-flow display might reveal motor control and attentional influences on MSTd neuronal responses to those stimuli. We refer to this task as steering because it consists of systematically altering the simulated direction of self-movement based on closed-loop visual feedback. Our initial experiments showed that optic-flow responses reflect the monkeys' strategy in the task; selecting a particular visual cue to guide its manual responses, rather than the manual response task itself. Our subsequent studies suggested that the monkey's selection of particular visual cues might be related to the spatial distribution of visual attention.
Materials and Methods
In this paper, we present the results of 2 experiments conducted on 3 monkeys. The first experiment had 2 versions, distinguished by their using different training protocols, conducted on the same 2 monkeys. Those 2 monkeys were then separated, 1 carried forward to the second experiment and the other transferred to another study. The third monkey was added during the course of the second experiment. All of the experiments are described in the order in which they were conducted.
Single-neuron recording chambers were placed over trephine holes in the parietal calvarium over MSTd of both hemispheres (anterior-posterior −2 mm, medial-lateral ±15 mm, angle 0).
Bilateral perilimbotomies were performed for the subconjunctival insertion of scleral search coils to monitor eye position throughout all experiments (Robinson 1963).
Tungsten microelectrodes (FHC, Inc. Bowdoin, ME) were passed through transdural guide tubes into cortex (Crist et al. 1988).
MSTd was identified by stereotaxic location and the physiological characteristics of single neurons: large receptive fields (>20 × 20°), which included the fovea with direction-selective responses that prefer large moving patterns over moving bars or spots. We recorded all stably isolated single neurons with these characteristics. We also recorded stably isolated single neurons that had complex directionality and ill-defined receptive field boundaries but were located at guide-tube depths and positions that had previously been found to contain neurons with the more typical characteristics of MSTd neurons.
Single-neuron discharges were isolated by template matching (Alpha Omega, Inc. Atlanta, GA), and firing times were stored through the real-time experimental system (REX) experimental control system (Hays et al. 1982). One of the three monkeys is engaged in continuing studies. Recording was completed in both hemispheres of the other 2 monkeys. These monkeys underwent pentobarbital euthanasia and transcardiac formalin perfusion. Histological analysis confirmed that the recordings were made in the anterior bank of the superior temporal sulcus. All procedures were approved by the University of Rochester Committee on Animal Research and were consistent with Society for Neuroscience policies.
The monkey sat fixating the center of a 90 × 90° tangent screen at a distance of 50 cm. It viewed radial optic-flow stimuli while fixation was monitored using magnetic search coil technique (Judge et al. 1980). The optic-flow stimuli presented at the beginning of each trial had 1 of 8 centers of motion at 20° eccentricity and arrayed at 45° intervals around the fixation point. Each stimulus contained 1000 white dots 0.75° wide with a brightness of 1.8 cd/m2 on a background of 0.2 cd/m2. The dots moved in a radial pattern simulating straight-line observer movement toward a frontoparallel plane. Dot speed was a sin2θ function of dot eccentricity, where θ is the angle between the monkey's line of sight and each dot's location on the screen. The dots accelerated as a 2 × cos θ × sin θ function of dot eccentricity (Nakayama and Loomis 1974; Lee and Young 1985; Hatsopoulos and Warren 1991). After implementing these considerations, we maintained an average dot speed of 40°/s in all stimuli.
Monkeys initially moved a joystick to the middle of its range by aligning a visual target within a stationary square and then holding the joystick in this position when the visual cues were removed. The monkeys were then required to fixate (±1.5°) a centered stationary circle presented throughout all trials. We recorded and compiled fixation data from all test conditions finding comparable performance across the monkeys and experiments. These data are provided in the text corresponding to each experiment.
When these behavioral prerequisites were met, the randomly selected optic flow stimulus was presented. After the monkeys brought the radial center of motion to within 3° of the fixation point, and maintained it in that range for 1 s, a liquid reward was delivered to terminate the trial. In training, the monkeys commonly overshot the target and reversed their manual positioning of the joystick to keep the radial center of the optic flow at the center of the display screen to successfully complete the trial. During recording, the monkeys more often slowed the movement of the center of motion as they approached the target position to complete the trial without an overt reversal of the direction of movement of the radial center of motion.
Experiment 1, Active–Passive Comparisons
In the first set of studies, a filled fixation target indicated the onset of an active-steering trial in which the monkey was required to maintain continuous contact with the joystick handle. An open fixation target indicated the onset of a passive-viewing trial in which joystick contact was prohibited. All trials required continuous, centered fixation throughout the stimulus presentation and manual response periods.
Active-steering trials required that the monkey must deflect the joystick horizontally and vertically to move the simulated heading in the radial optic flow to the center of the screen. The optic flow initially appeared with its radial center at 1 of 8 eccentric locations. The optic flow was then modified based on the monkeys' joystick responses so that its radial center of motion progressively moved to the center of the screen within the 4-s time limit for centering the simulated heading (Fig. 1). The appearance of a gradually shifting radial center of motion simulated optic flow seen when an observer steers its heading of self-movement from an eccentric direction to a straight-ahead direction. The monkey maintained centered fixation throughout these trials.
Maximum joystick deflection induced a maximal rate of radial center displacement of 10°/s. The monkeys learned to expedite successful trials by maximizing the appropriate joystick deflection until the radial center of motion approached the center of the screen and then adjusted its response to keep the radial center of motion at that location. In passive-viewing trials, we replayed optic-flow stimuli recorded from the preceding active-steering trial having that radial center of motion while the monkey maintained centered fixation. As in active-steering trials, the monkey maintained centered fixation throughout these trials.
Two versions of Experiment 1 were conducted. In the first version of Experiment 1, the monkeys were allowed to derive their own strategy for steering the radial center of optic flow from its initially eccentric position to the center of the screen. Our interpretation of the results of those studies prompted us to view it as reflecting a local motion–processing strategy in the steering task. In the second version of Experiment 1, the monkeys were drawn in to using a global motion–processing strategy in the steering task. We analyzed data from both versions of Experiment 1 in a mixed-model 2-way analysis of variance (ANOVA) with a between-subject variable of local versus global strategies and a within-subject variable of simulated heading direction.
Experiment 2, Local–Global Comparisons
In the second set of studies, 2 types of active-steering trials were blockwise interleaved, without a passive-viewing condition. Both types of trials required continuous, centered fixation of a filled circle at the center of the screen and continuous contact with the joystick handle. A small arrow located over the fixation point indicated the onset of a trial demanding active steering by central local motion in the optic-flow stimulus (Fig. 6A). A small open circle at 1 of the 8 eccentric target locations indicated the onset of a trial demanding active steering by the surrounding pattern of global motion in the optic-flow stimulus (Fig. 6B).
Both types of trials required that the monkey must rotate the joystick handle to steer the optic flow to conform to the requirements of that trial while maintaining centered fixation. In local-motion trials, dot movement near the center of the screen had to be made to match the direction indicated by the arrow. In global-pattern trials, the eccentric radial center of motion had to be made to match the location of the peripherally placed open circle target location. In both types of trials, the radial center of motion of inward or outward optic flow initially appeared at 1 of the 8 eccentric start positions, the same start positions used in the earlier study, and had to be rotated 90° in the clockwise (CW) or counter-clockwise (CC) direction to align the optic-flow stimulus with the arrow in local motion trials or the circle target in global-motion trials.
It should be recognized that CW and CC rotation of the joystick had different implications in the local motion and global-pattern trials. In local-motion trials, inward versus outward radial patterns required CW versus CC joystick responses. In global-pattern trials, a given inward or outward center of motion required the same direction of joystick deflection to reach a given target location. Thus, the behavioral responses in local motion and global pattern trials were not interchangeable and the monkey had to know which strategy it was to use to execute a response that could be completed in the allotted time. If the monkey initially moved the joystick in the wrong direction, it was impossible for it to reach the goal location within the allocated 5 s. Those trials were classed as response failures, and recording was aborted without the monkey earning a reward. The monkeys were overtrained in these tasks and consistently produced better than the 95% correct trials, a bit poorer at the end of each recording day.
We imposed further constraints on the monkey's strategy by adding an occlusive mask covering selected segments of the stimulus. The occlusive mask was added at the middle of the stimulus presentation period (2500 ms after stimulus onset). In local-motion trials, we masked the peripheral segment of the stimulus, the area outside of the central 20° (Fig. 6A). In the global-pattern trials, we masked the central segment of the stimulus, the area within the central 20° (Fig. 6B).
In both Experiments 1 and 2, blocks of 8 successful trials of each type were interleaved with blocks of 8 successful trials of the other type: In Experiment 1, blocks of 8 active-steering trials were interleaved with 8 passive-viewing trials. In Experiment 2, blocks of 8 local-motion trials were interleaved with blocks of 8 global-pattern trials. All monkeys were highly trained, successfully completing greater than 90% of the trials, before we began neuronal response recordings. We recorded 6–8 repetitions of each trial type in each neuron. All 3 monkeys consistently completed trials in less than 6 s, with similar dynamics of joystick manipulation across all trials.
Neuronal firing rates evoked by outward radial optic-flow stimuli were displayed as raw averages for each stimulus condition. In the first experiment, response data was entered into mixed-model 2-way ANOVA having within-subject effects of radial center of motion (8 locations) and task (active steering vs. passive viewing) and having between-subject effects of behavioral paradigms (local vs. global tasks). In the second experiment, response data was entered into a mixed-model 2-way ANOVA having within-subject effects of radial center of motion (8 locations) and between-subject effects of behavioral paradigms (local vs. global tasks). We report the Greenhouse–Geisser corrected F values (Statistical Package for Social Sciences [SPSS, 2005]).
The preferred stimuli were defined as the stimuli that elicited the largest neuronal response to any of the active (8) or passive (8) conditions. The responses to the preferred stimuli were averaged to create population spike-density response profiles for each condition. The averaged neuronal responses for all directions were displayed as polar plots to illustrate the strength and directionality of the responses for each condition. In the polar plots, 8 thin radial lines were used to represent the responses to each of the corresponding 8 directions of simulated observer movement. The length of those lines is proportionate to the neuronal firing rate during the second 500 ms of stimulus presentation. The thick radial line in each polar plot indicates the vector sum of the 8 individual response vectors. Its direction is in the mean direction of the circular distribution, and its length is proportionate to the mean resultant length of that distribution (Batschelet 1981).
We used Gaussian functions to derive a best-fit curve to neuronal response amplitudes across the 8 outward radial optic-flow–heading direction stimuli. Neuronal response firing rates were measured in the 500- to 1 000-ms interval after stimulus onset, the period corresponding to the monkey's initiating a behavioral response to the stimuli. We quantified the effects of each stimulus and task condition by subtracting the prestimulus background firing rate of that neuron, recorded in that trial, from the firing rate in the response interval. This focused our response measurements on the comparison of stimulus-evoked activity to baseline activity.
Data from 6–8 stimulus repetitions were averaged for each heading direction with the averaged responses submitted to a curve-fitting algorithm (The Mathworks, Inc., Natick, MA, lsqcurvefit). The averaged responses for each neuron were combined by first aligning them on each neuron's preferred heading direction and then averaging responses across all neurons for each stimulus condition. The Gaussian functions yielded amplitude, baseline, tuning width, and preferred radial center parameters based on a least squares algorithm and a full-width half-height measurement of tuning for comparisons across conditions.
We tested whether engaging the monkey in an active-steering task might alter MSTd neuronal responses to simulated optic flow. Three Rhesus monkeys learned to steer using optic-flow stimuli and were the subjects of 2 experiments in which we recorded the activity of 149 MSTd neurons.
Experiment 1: Active Steering versus Passive Viewing
We blockwise interleaved active steering trials with the passive viewing of optic flow. In all trials, the optic flow first appeared with a radial center of motion at 1 of 8 eccentric positions. In active steering trials, the monkey used a joystick to move the radial center of motion in outward radial optic flow toward the centered fixation point. Across all directions, the monkeys showed an average joystick response latency of 273 ± 10 ms with joystick movement velocities 3.47 ± 0.33°/s during the subsequent 500 ms. Continuing to hold that joystick position moved the radial center of motion to the fixation point at the center of the screen within the 4-s time limit (Fig. 1). In passive-viewing trials, the monkey released the joystick and viewed a replay of the optic flow seen during active steering. Eye-position records demonstrated similar performance in the centered fixation task during passive-viewing and active-steering trials, as evidenced by comparable gaze eccentricities (mean ± standard deviation: passive, 0.75 ± 0.18°; active, 0.68 ± 0.24°) across all trials represented in the manual-response tracings of Fig. 1C.
We compared the responses of 50 neurons recorded during active-steering and passive-viewing trials. Most neurons (84%, 42/50) showed significant responses (P < 0.05) to radial optic flow in either the active or passive condition, or both. We found that passive viewing yielded larger responses than active steering (Fig. 2B). After subtracting the baseline activity, recorded during fixation before stimulus onset, responsive neurons showed an average of 77% larger peak responses in passive-viewing trials compared with active-steering trials in the 500- to 1 000-ms interval.
Interleaved Inward and Outward Optic Flow
We considered that smaller optic-flow responses during active steering might reflect the monkey being engaged by manual aspects of the steering task. Alternatively, the monkey might be focusing on aspects of the stimulus other than the patterned optic flow: smaller responses during active steering might reflect the monkey using local-motion cues in the optic flow rather than the global pattern of optic flow for which MSTd neurons are specialized. From the monkey's point of view, it might move the joystick in the direction of the central dot motion until that motion slows and then ceases as the radial center nears the fixation point.
We tested these hypotheses by continuing to engage the monkey in the same manual steering task but altering the utility of local-motion cues in the optic flow by randomly interleaving trials presenting inward or outward radial patterns. Interleaving inward and outward optic flow makes central local motion ambiguous. For example, leftward central motion is seen with outward optic flow having a right-sided center of motion (Fig. 3A, left) and with inward optic flow having a left-sided center of motion (Fig. 3B, left).
The initial presentation of interleaved inward and outward radial motion stimuli resulted in remarkable failure: both monkeys made 100% errors on the inward optic-flow trials. Almost all those errors were the result of the monkeys moving the joystick in the opposite direction required to move the center of motion toward the fixation point. These responses compounded with the usual errors to yield an overall 65–75% error rate, and this frustrating experience led the monkeys to stop working. We interpreted this experience as confirming that the monkey had learned to use the central local-motion cue instead of the global pattern of radial motion in the optic flow.
We then embarked on retraining monkeys to use the global radial motion cues in optic flow. Initial retraining used only inward radial optic flow, but the monkey's high error rates persisted with opposite direction responses, forcing the radial center of motion off the screen. These errors exhausted the maximum response duration and resulted in aborted trials with the programmed data loss. We then greatly constrained the spatial extent of the stimuli to only 20° diameter. Over the following 2 weeks, the monkeys learned to direct the inward radial center of motion in to the center of the screen. Over the following month, we gradually increased the stimulus size back to its uniform full coverage of the stimulus screen, throughout all trials. We then reintroduced outward radial optic-flow stimuli, first in every other day interleaving with inward patterns, then with blockwise interleaving on every training day, and finally with regular trial-by-trial interleaving. The course of retraining back to >90% success rate required 12 weeks in both monkeys.
After retraining with interleaved inward and outward radial optic flow, emphasizing steering by the global pattern of motion (Fig. 3), the monkeys showed longer (t-test, P = 0.004) average joystick response latencies (297 ± 14 ms) than in the preceding study (273 ± 10 ms). Although the new strategy took longer to initiate, their eventual behavioral responses had the same dynamics with initial joystick movement velocities (3.80 ± 0.60°/s) that were similar (t-test, P = 0.08) to those obtained in the previous studies (3.47 ± 0.33°/s). Oculomotor behavior was similar with eye position records, demonstrating comparable performance in maintaining centered fixation during the passive-viewing (0.54° ± 0.43°) and active-steering (0.84°± 0.65°) trials.
We again recorded MSTd neuronal activity during interleaved active-steering and passive-viewing trials. To facilitate comparison to the results of the previous study, we used only outward radial trials in the analysis of neuronal responses to optic flow. In these studies, many MSTd neurons showed substantially stronger optic-flow responses during active-steering trials (Fig. 4A). Most neurons (78%, 31/40) showed significant responses (P < 0.05) to radial optic flow in either the active or passive condition, or both (Fig. 4B). After subtracting baseline activity, these neurons showed an average of 159% larger peak responses (500–1 000 ms after stimulus onset) during active-steering versus passive viewing, the opposite of the previous result.
Comparing Active-Steering Effects
We compared neuronal responses obtained in the active-steering and passive-viewing conditions for neurons recorded before (n = 42) and after (n = 31) we retrained the monkeys to use the global pattern of visual motion, that is, before retraining, when we presented only outward radial flow, versus after retraining, when we interleaved inward and outward radial optic-flow trials but recorded neuronal activity only in the outward radial optic-flow condition. A 2-way ANOVA of averaged neuronal firing rates during the second 500 ms of each stimulus (500–1 000 ms after stimulus onset) had a between-subject variable of local versus global condition (before and after retraining) and repeated measures variables of active steering versus passive viewing and of the 8 initial radial center of motion locations (simulated heading directions). This analysis revealed significant interactions between active steering versus passive viewing across the local (before retraining) versus global (after retraining) conditions (F1,71 = 5.67, P = 0.02).
The impact of the behavioral tasks was even evident in the relatively small responses to centered radial center of motion optic flow, the stimulus that was seen at the end of all successful trials. In the initial studies, the centered radial center stimulus evoked significantly higher mean firing rates in passive-viewing trials compared with active-steering trials (passive = 13.5 ± 0.5, active = 12.3 ± 0.5 2-way ANOVA, as above, showed significant active versus passive effects: F1,238 = 13.76, P < 0.001). In studies conducted after retraining the monkeys, the centered radial center stimulus evoked significantly higher mean firing rates in active-steering trials compared with passive-viewing trials (active = 18.6 ± 1.0, passive = 16.5 ± 0.6, F1,217 = 20.35, P < 0.001).
The comparison of Gaussian fits to the radial center of motion-tuning curves, averaged across neurons in the active and passive trials, revealed different aspects of the experimental effects. In this first series of studies, the Gaussian parameters (active vs. passive: mean ± confidence interval, P value from t-test) yielded higher amplitude (15.6 ± 0.10 vs.19.1 ± 0.08 spikes/s, P = 0.0001) and baseline (8.5 ± 0.07 vs. 9.4 ± 0.06 spikes/s, P = 0.039) measures in the passive condition, with nonsignificantly broader tuning width in passive trials (42.74° ± .405° vs. 46.8° ± 3.15°, P = 0.16). The preferred heading did not change appreciably across conditions (Fig. 5A).
When the monkeys were retrained with interleaved inward and outward radial optic flow, to promote the use of a global pattern–steering strategy, the Gaussian fits showed a nonsignificant larger amplitude during active steering (11.8 ± 0.23 vs. 10.1 ± 0.31 spikes/s, P = 0.067) with a significantly larger baseline during active steering (17.4 ± 0.07 vs. 11.4 ± 0.14 spikes/s, P < 0.0001) and a significantly sharper tuning width in active-steering trials (50.4° ± 9° vs. 91.8° ± 8.1°, P = 0.0002). The preferred heading did not change appreciably across conditions (Fig. 5B).
Together, these findings support the view that training the monkeys to use the global pattern of motion significantly changed the relationship between responses to radial optic flow during active steering and passive viewing. Using local motion, the passive-viewing trials yielded significantly stronger responses with significant amplitude and baseline effects. Using global patterns, the active-steering trials yielded significantly stronger responses with significant tuning width and baseline effects.
The manual control task of moving the joystick to direct the radial center of motion to the fixation point at the center of the screen was the same in both studies. Furthermore, the monkeys showed similar patterns of manual and oculomotor behavior in both studies. This led us to conclude that motoric aspects of the task itself are unlikely to account for neuronal response differences between these studies. However, we had to consider the possibility that the monkey's selecting local motion cues in the first study, and global pattern cues in the second study, might account for the observed differences in the neuronal responses.
Experiment 2: Promoting the Use of Local and Global Cues
We characterized the influence of the monkey's selecting either central local motion cues, versus peripheral global pattern cues, on single-neuron responses to optic flow. To do so, we trained the monkeys to perform pseudorandomly interleaved trials of active steering, using either the central local motion in optic flow or the global pattern of radial motion that dominates the visual periphery when viewing optic flow. In both types of trials, the monkey rotated the joystick's handle to steer the optic flow to match a designated target condition of either central local motion or peripheral radial center of motion.
Trials that required active steering by central local motion began with a small, centered arrow pointing in 1 of 8 directions. We then presented an outward radial optic-flow stimulus having central local motion in a direction that was rotated 90° CW or CC from the direction indicated by the arrow. The monkey turned the joystick to transform the optic-flow field and rotate the direction of dot motion at center of the stimulus. The monkey then adjusted the joystick position until central dot motion was maintained for 500 ms in alignment with the direction indicated by the arrow. After the first 2.5 s, we masked off the peripheral stimulus segment, beyond the central 20°, so that only the central local motion remained as a cue for successfully completing the trial (Fig. 6A).
Trials that required active steering by the radial pattern in the periphery began with a target spot at 1 of the 8 eccentric locations. Outward radial optic flow then appeared with a radial center of motion at a position rotated 90° CW or CC from the target spot. The monkey turned the steering to transform the patterned optic flow and move the radial center of motion along an arc through the periphery so that it came to overlap the target spot. After the first 2.5 s, we masked off the central 20° of the stimulus, so that only the patterned motion in the periphery remained as a cue for successfully completing the trial (Fig. 6B).
After extensive training in alternating between the use of central local motion and radial patterns of optic flow, the monkeys showed some differences in their performance in the 2 paradigms (Fig. 6C).In both types of trials, the average joystick response latencies were longer than in the previous experiments (local motion = 396 ± 66 ms; global pattern = 388 ± 59 ms), with paired comparisons showing a small but consistent advantage for faster responses in the global pattern condition (t = 4.38, P = 0.001). Eye-position records demonstrated performance comparable to that seen in the earlier experiments with the monkeys maintaining centered fixation during active steering by local motion (0.33 ± 0.27°) and by global patterns (0.28 ± 0.11°). These effects suggest the monkeys used different strategies to complete the 2 conditions.
We compared optic-flow responses during active steering by central local motion and active steering by peripheral radial patterns. Most neurons (95%, 56/59) showed significant responses to optic flow (P < 0.05) during either local motion or radial pattern trials, or both (Fig. 7A). Two-thirds of the neurons (66%, 39/59) showed larger average response amplitudes in the radial pattern trials. After subtracting baseline activity, this effect amounted to a net 24% larger peak response during radial-pattern trials than during local-motion trials (Fig. 7B). Thus, across all 59 neurons tested, optic flow evoked larger responses during trials in which the monkey was committed to steering using the peripheral radial pattern of optic flow.
Links to Active-Steering Effects
We observed links between the effects of local-motion steering versus radial-pattern steering and the effects of adding the stimulus masks. In some neurons, adding a peripheral mask caused a dramatic decrement in response activity (Fig. 8B). In other neurons, the peripheral mask evoked increased activity (Fig. 8C).
We measured mask-induced activation as a contrast of the firing rate during the 500 ms before and after the addition of the peripheral mask (Fig. 9A). This revealed a continuum of peripheral mask–induced effects with most neurons showing diminished activity after the addition of the peripheral mask (47/59) and others showing increased activity after the addition of the peripheral mask (12/59).
A relationship between peripheral mask effects and receptive field organization is suggested by receptive field maps. Conventional hand mapping was obtained using projected bar and random pattern stimuli in 14 neurons in which recording time was sufficient. Six neurons yielded maps that defined receptive fields with clear boundaries. Three of those neurons with large, peripheral receptive fields showed response inactivation by the addition of the peripheral occlusive mask. In contrast, 3 neurons with small, central receptive fields showed little effect or a degree of response activation by the addition of the peripheral occlusive mask (Fig. 9B).
In a second analysis of these data, we divided the sample in half to focus on the 30 neurons that showed largest decrease in activity with the addition of the peripheral mask. In this analysis, we compared optic-flow responses during central local-motion steering and radial-pattern steering (Fig. 10A). On subtracting baseline activity, these neurons showed 37% larger responses to the preferred optic-flow stimulus during radial-pattern steering than during local-motion steering.
Gaussian fits to the radial center of motion-tuning curves revealed both amplitude and tuning width changes across the local-motion and radial pattern–steering trials. Gaussian parameters (local motion vs. radial pattern: mean ± CI, P value from t-test) yielded higher amplitude in radial pattern trials (23.8 ± 0.06 vs.30.4 ± 0.07 spikes/s, P = 0.05) and nonsignificant baseline changes (7.6 ± 0.07 vs. 7.7 ± 0.10 spikes/s, P = 0.79) but significantly sharper tuning width in radial pattern trials (50.4° ± 2.25° vs. 44.1° ± 3.15°, P = 0.003). The preferred heading did not change appreciably across conditions. These effects reflect a substantial increase in responsiveness to, and stronger selectivity for, the preferred optic-flow stimulus during radial-pattern steering in neurons that have large, peripheral receptive fields (Fig. 10B).
Task Effects on Optic-Flow Analysis
We engaged trained monkeys in an active steering task that required them to use optic flow to guide manual responses. MSTd neuronal responses to optic flow during active steering were compared with responses recorded while the monkey passively viewed the same stimuli. We hypothesized that active steering would enhance MSTd neuronal response to optic flow. However, our initial studies revealed the opposite effect; MSTd neurons yielded smaller optic-flow responses during active steering than during passive viewing (Fig. 2).
Our finding smaller neuronal responses during active steering was unexpected, especially in the context of the many studies demonstrating increased visual cortical neuronal activity during attentive fixation to visual stimuli (Lynch et al. 1977; Bushnell et al. 1981; Moran and Desimone 1985). A more recent work has confirmed and extended such findings, including studies that showed heightened neuronal excitability when attention was focused on the receptive field of neurons being studied either in area medial superior temporal (MSTd) or adjacent medial temporal (MT) cortex (Treue and Maunsell 1996; Seidemann and Newsome 1999; Cook and Maunsell 2002). The task-dependent differences in the strength of amplitude, baseline, and tuning-width effects seen across our studies support the impression, derived from our recent work, that the proportion of response magnitude and selectivity effects depends on the details of the behavioral task and its links to the stimuli being presented (Dubin and Duffy 2007).
However, some studies have shown that extrastriate visual areas can express diverse responses to attentional manipulations. Feature selective neurons in V4 can show suppression of responses to preferred stimuli when that stimulus does not possess the feature that was cued as behaviorally relevant in that trial (Motter 1994). In the dorsal stream, 7a neurons are less responsive to stimuli presented at the location of attentive fixation with stronger responses to stimuli presented at nonattended sites (Steinmetz and Constantinidis 1995). These mechanisms may maintain vigilance for stimuli appearing in the peripheral visual field during centered attentive fixation, a critical capacity when threats may approach a focused observer.
The mixed results of functional imaging studies of optic-flow activation may be related to the diversity of attentional effects. Optic-flow stimuli yield stronger hMT+ activation during attention to the visual motion rather than simultaneously presented stationary stimuli (O'Craven et al. 1997). In contrast, regional activation by self-movement in a virtual environment is reduced when the observer is engaged in an active-steering task (Walter et al. 2001). The reduced responses during steering have been interpreted as reflecting the subjects attending to motor control aspects of the task. We might speculate that the reduced responses are related to the subjects' cue selection as a perceptual strategy in the task, rather than the manual task itself.
Retraining Reverses Task Effects
We considered the hypothesis that motor control signals generated in the task might be responsible for the reduction of MSTd neuronal responses during active steering. This perspective is supported by analogy to the finding that vestibular neurons respond more vigorously during passive head–body rotation than during manually controlled active head–body rotation (Roy and Cullen 2001) or during volitional head-on-neck movements, possibly reflecting sensory response suppression by motor efference copy (Cullen and Roy 2004). An alternative hypothesis is that MSTd neuronal responses to optic flow were reduced because the monkey was not using the global pattern of optic flow to guide its steering behavior, possibly relying on local motion cues instead.
We compared the motor control and visual motion hypotheses by training the monkey to use the same active-steering versus passive-viewing motor control paradigm with the interleaved presentation of 2 different types of visual motion stimuli, inward and outward optic flow. Our previous psychophysical work had suggested that interleaving inward and outward optic-flow stimuli forces observers to use the global pattern of motion to make decisions about the simulated heading direction (O'Brien et al. 2001). We reasoned that if the monkeys' active motor control for steering in itself was the cause of reduced MSTd neuronal responses, that reduction would persist regardless of the associated visual stimuli.
However, if the monkey had previously learned to ignore the global pattern of optic flow, in favor of a reliance on local-motion cues, then the presentation of inward and outward patterns would present ambiguous stimuli: right center of motion outward flow contains the same central local motion as left center of motion inward flow (Fig. 3A,B, left). In that case, relearning the steering task by relying on global-pattern cues might alter neuronal responses such that active steering would evoke larger responses than passive viewing of the same optic-flow stimuli.
Initially, the monkeys consistently made errors suggesting that they were using local-motion cues. After retraining the monkeys so that they would use global motion, we repeated the neurophysiological studies and found that the effects of active steering had been reversed; active steering now enhanced neuronal responses instead of suppressing them (Fig. 4).
These observations support the view that learned stimulus–response associations can alter the relationship between neuronal activity and behavioral tasks. Specifically, MSTd neuronal responses to identical optic-flow stimuli depend on the monkeys' perceptual strategy for using visual information to control steering. This is not to conclude that engaging the monkey in an active manual control task does not in itself alter the cortical processing of optic flow (Merchant and Georgopoulos 2006), but only that the response changes seen in these studies do not reflect such effects. However, these data were not recorded in the same neurons so we could not resolve whether local and global strategies might change the effects of active steering in individual neurons or might engage different groups of neurons.
Single Neurons and Perceptual Strategies
We trained the monkeys to perform a related steering task using interleaved trials in which the monkey is cued to steer either by central local-motion cues or peripheral radial pattern cues in identical optic-flow field stimuli (Fig. 6). During the first 1 s of the stimulus, an interval when the stimuli in both types of trials are identical, most neurons showed larger responses when the monkey was steering by the radial pattern of optic flow. We did not see irregularities in the monkey's manual responses that might suggest its alternating perceptual strategies during the trials, keeping in mind that local and global trials demanded different response strategies from the start (see Materials and Methods). It is reassuring that if the monkey's strategy were irregular, it would tend to reduce the magnitude of the effects we observed. Hence, we consider our estimate of task effects to be conservative.
We classified these neurons based on whether occlusive masking the peripheral visual field reduced the neuronal response, presumably by preventing the stimulation of an activating segment of the receptive field (Fig. 8). We found that neurons in which a peripheral mask decreased the firing rate tended to show larger responses when the monkey was engaged in steering by the global radial pattern of optic flow. These observations are consistent with the notion that the monkey's perceptual strategy alters optic-flow responses by shifting the balance between neuronal subpopulations having different receptive field structures. Most MSTd neurons have large, peripheral receptive fields consistent with inactivation by peripheral masking (Fig. 9A). MSTd neuronal subpopulations have previously been described as having either synergistic or antagonistic subzones, often with center-surround architecture (Tanaka et al. 1986; Komatsu and Wurtz 1988a; Eifuku and Wurtz 1998). Our studies may be accessing this receptive field organization in a manner that is related to the role they may play in guiding navigation (Logan et al. 2006).
The presence of receptive field subpopulations in MSTd might also explain the results of the first 2 experiments. Neurons with centrally activated receptive fields might be selectively engaged during steering when the monkey is using central local motion (Fig. 9B, right). Their smaller responses could account for the net decrease in response amplitude during the active trials of the first study (Figs 2 and 5A). In contrast, neurons with larger, peripherally activating receptive fields might be selectively engaged during steering by the radial pattern of optic flow (Fig. 9B, left). Their larger responses could account for the net increase in response amplitude during the active-steering trials of the second study (Figs 4 and 5B). Thus, perceptual strategies for steering might influence which neurons are more active in the task.
Attention and Steering by Optic Flow
Differences in neuronal activity during local-motion and radial-pattern steering might reflect the receptive field organization's interactions with attentional mechanisms. We did not obtain quantitative receptive field maps in these studies, but the quantification of the impact of our masking arbitrarily delineated central and peripheral segments of the stimulus suggests broader relations to earlier work on receptive field links to attentional effects. MSTd neurons yield larger responses when the monkey attends to the location of the receptive field (Treue and Maunsell 1996; Seidemann and Newsome 1999; Treue and Martinez Trujillo 1999). A previous example of spatial attention's selection of particular neurons might be evident in the activation of direction-selective MT neurons that are tuned to the binocular disparity of directional cues in a stimulus, supporting direction discrimination by selecting neurons based on their spatial response field in the depth plane (DeAngelis and Newsome 2004).
The optic flow's simultaneous stimulation of the central and peripheral visual field might promote competitive interactions between neurons having different receptive field structures (Moran and Desimone 1985; Spitzer et al. 1988; Reynolds et al. 1999). Competitive interactions in MSTd might be modulated by top–down, potentially inhibitory, signals that tip the balance in favor of a particular segment of the receptive field that might account for the results of both the first and second versions of Experiment 1 as well as Experiment 2. A related phenomenon, or at least an apt analogy, might be seen in the effects of frontal microstimulation on receptive field–specific activation of V4 neurons, an effect that is thought of as being linked to spatial attention (Armstrong et al. 2006). Such mechanisms might contribute to activating MSTd neurons with receptive fields overlapping the attended segment of the visual field.
Alternatively, differences in neuronal activity during local-motion and radial-pattern steering might reflect the relative activation of MT versus adjacent MSTd. The overall spatial scale of a stimulus might also contribute to selective areal activation as seen in the matching of stimulus size to the receptive field dimensions of V4 neurons (Hopf et al. 2006). Such effects might be linked to a searchlight-like mechanism that could match the spatial extent of a selected visual cue (Crick 1984; McAlonan et al. 2006) to the average receptive field dimensions of certain cortical areas, possibly through winner-take-all interareal competition (Lee et al. 1999). In our studies, the monkey's steering by central local-motion cues, rather than by radial-pattern cues, might be equivalent to its engaging in a task that would activate smaller and more centrally located MT-receptive fields versus the larger and more peripherally located MST-receptive fields.
Areal activation of MT versus MST might also be the result of feature-specific attentional mechanisms. The activation of human MT+ depends on the subjects' attending to motion cues, with less activation seen when the same stimuli are presented during tasks that rely on color (Beauchamp et al. 1997) or linguistic (Rees et al. 1997) cues. Likewise, the activation of human MST depends on the subject's attending to the pattern motion stimulus rather than an overlapping stationary pattern (O'Craven et al. 1997). Such feature-specific attentional mechanisms are not exclusive of spatial attentional effects. We have previously presented evidence of coordinated featural and spatial effects on MSTd response properties (Dubin and Duffy 2006). The relative contributions of these mechanisms cannot be resolved in these current studies.
We must be conservative about concluding which, if any, of these mechanisms might pertain to the steering we engage in for everyday vehicular control. Our tasks have much in common with steering for vehicular control, particularly the use of visual feedback for the manual control of the self-movement heading. However, our monkeys are viewing a simulation and are not actually moving. It is relevant that we have previously recorded comparable MSTd neuronal responses during simulated and actual self-movement (Duffy 1998; Page and Duffy 2003). In addition, it may be relevant that our human subjects, engaged in similar tasks (Mapstone et al. 2006), spontaneously refer to it as driving or steering. We have also found links between the ability to interpret these self-movement simulations and human ambulatory (Tetewsky and Duffy 1999; Kavcic et al. 2006) and vehicular (O'Brien et al. 2001) navigation. Nevertheless, the underlying neural mechanisms engaged in everyday steering for vehicular control remain unknown.
This work was supported by the National Eye Institute (EY10287).
We are grateful for the comments of Roberto Fernandez, Voyko Kavcic, Sarita Kishore, Mark Mapstone, Nobuya Sato, and William Vaughn on earlier versions of the manuscript. Sherry Estes and Jennifer Postle assisted with the preparation and training of monkeys. Zarina Ali assisted in the recording of some of these neurons as part of a thesis project at the University of Rochester Undergraduate Program in Neuroscience. Conflict of Interest: None declared.