| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|
|
|
|||||||
|
|||||||||
1Program in Neuroscience and 2Department of Neurobiology and Anatomy, Wake Forest University School of Medicine, Winston Salem, North Carolina 27157
Submitted 16 December 2003; accepted in final form 31 January 2004
 |
ABSTRACT |
|---|
 TOP ABSTRACT INTRODUCTIONMETHODSRESULTSDISCUSSIONACKNOWLEDGMENTSREFERENCES |
|---|
|
INTRODUCTION |
|---|
|
TOPABSTRACTINTRODUCTIONMETHODSRESULTSDISCUSSIONACKNOWLEDGMENTSREFERENCES |
|---|
Previously, we reported on the visual- and saccade-related properties of neurons in several central thalamic nuclei as revealed through the use of a visually guided, delayed saccade task (Wyder et al. 2003a
). This task, which coupled a specific sensory stimulus to a specific saccadic response by an instructed delay, revealed that many thalamic neurons represented the times of occurrence and locations of visual stimuli and the saccades made to acquire them. In addition, we found that many thalamic neurons maintained veridical spatial information throughout the instructed delay period during which the monkey was required to withhold responding to the stimulus. Spatially selective delay-period activity may be the hallmark of a neuron that participates in "higher-order" aspects of sensorimotor function. In many regions, including the lateral intraparietal area (LIP), supplementary eye field (SEF), frontal eye field (FEF), and dorsolateral prefrontal cortex (PFCdl), the temporal evolution of delay period activity has been shown to correlate with cognitively driven events such as movement selection (Glimcher and Sparks 1992), motor planning (Barash et al. 1991a,b; Bracewell et al. 1996; Mazzoni et al. 1996; see Andersen 1995 for review), spatial attention (see Colby and Goldberg 1999 for review), and perceptual judgment (see Glimcher 2001; Schall and Thompson 1999; Shadlen and Newsome 1996 for reviews). Common to all of these studies is the demonstration that the magnitude of delay-period activity is a function of both the physical stimulus and the context in which the stimulus occurs.
The observation that central thalamic neurons maintain spatial tuning throughout an imposed delay period, though suggestive, is not strong evidence for involvement in a context-dependent process of linking sensory stimuli to saccadic commands. Such activity might simply indicate the presence of a visual stimulus within the neuron's receptive field, independent of task context. The objective of the present study was to determine whether, in fact, these neurons carry information about both the stimulus and its relevance within the context of the behavioral task. To do so we evaluated the activity of single central thalamic neurons in association with both a singletarget delayed saccade task and a 2-stimulus saccadic choice task. Whereas in the single-target task, the lone visual stimulus [whether within or beyond the neuron's response field (RF)] was known from the outset to be the target of an eventual saccade, the choice task contained a period of ambiguity during which the 2 simultaneously present stimuli (one within and one outside of the neuron's RF) had equal potential to become either the saccadic goal or an irrelevant distracter. This period was followed by a cue identifying the target and distracter.
The single-target and choice tasks permitted comparison of activity evoked by the same physical stimulus in the neuron's RF in 4 different behavioral contexts: 1) a single stimulus known to be the saccade target; 2) 2 stimuli, one in and one out of the RF, each with the potential to become a saccade target (or distracter); 3) 2 stimuli, a known target in the RF and a known distracter out of the RF; and 4) 2 stimuli, a known distracter in the RF and a known target out of the RF.
The logic of the single-target/choice task comparison is straightforward, with "stimulus-bound" activity predicted to be relatively unaffected by the changing context and more "goalrelated" activity differentiating between potential targets (and distracters), known targets, and known distracters. We included "memory" variants of the single-target and choice tasks to further distinguish stimulus-related and goal-related activity. Whereas the prediction for purely stimulus-related activity is relatively trivial—a cessation of activity when the RF stimulus is extinguished—that for apparently "goal-related" activity is more interesting. In principle, the representation of a saccadic goal could be stimulus-independent, persisting after the stimulus has disappeared, or stimulus-dependent, specifying the locations of currently visible goals only.
There are to date very few published accounts of the visuo-oculomotor properties of neurons in central thalamus (Schlag and Schlag-Rey 1984; Schlag-Rey and Schlag 1984; Wyder et al. 2003a; see Sommer 2003 for review) and, to our knowledge, only preliminary findings have shown the capacity for delay-period activity to carry information related to behavioral context (Schall and Thompson 1994). In brief, we observed some central thalamic neurons that were consistent with each of the stimulus/goal-related outcomes described above. One group signaled the presence of a visual stimulus independent of task goals, a second group signaled the location of a current saccadic goal independent of the continued presence of the stimulus, and a third group signaled the presence of currently visible saccadic goals. The latter 2 groups indicate that delay-period activity in central thalamus can convey combined information about sensory stimuli (both past and present) and the behavioral context in which they occur. Considered along with the anatomical position of central thalamus, these data suggest that activity within cortico-subcortical loops plays a role in the context-dependent linkage of sensory signals and saccadic commands.
Some of these data previously appeared in preliminary form (Wyder et al. 2003b).
|
METHODS |
|---|
|
TOPABSTRACTINTRODUCTIONMETHODSRESULTSDISCUSSIONACKNOWLEDGMENTSREFERENCES |
|---|
All surgical and experimental protocols complied with the National Institutes of Health Guide for the Care and Use of Laboratory Animals, USDA regulations, and the policies set forth by the Wake Forest University School of Medicine Animal Care and Use Committee (ACUC). Details of the surgical and recording procedures were previously described (Wyder et al. 2003a). Briefly, 3 rhesus monkeys (Macaca mulatta) were prepared for chronic single-unit recording. Before behavioral training, under general anesthesia, an MRI-compatible titanium post was attached to the skull and an eye coil was implanted in one eye (Judge et al. 1980). During subsequent training/recording sessions, the post served to restrain the monkey's head while the eye coil provided an analog signal of eye position (Fuchs and Robinson 1966; Robinson 1963). Recovery from the initial surgery required 2–4 wk, during which time analgesics and antibiotics were administered as required.
Fully recovered animals were trained on the behavioral tasks (see following text). Once trained to criterion levels of performance, a second surgery was performed to place a recording cylinder (Crist Instrument) over the central thalamus. Daily recording sessions began on full recovery (2–3 wk).
Recording procedures
Eye position was sampled and stored at 500 Hz. Neural activity was recorded using parylene-coated, tungsten microelectrodes (Micro Probe) having impedances of between 1.0 and 1.5 M
at 1 kHz. Electrodes were inserted through a dura-piercing cannula and advanced to the thalamus by a hydraulic microdrive. Activity was monitored using an oscilloscope and an audiomonitor, and the action potentials of single neurons were isolated using a time/amplitude window discriminator. Spike times were stored at a resolution of 10 µs.
Behavioral tasks
During training and subsequent recording sessions, monkeys were seated in a primate chair in a very dimly lit room. The stimulus display consisted of an array of light-emitting diodes (LEDs). At a viewing distance of 57 in., adjacent LEDs were separated by either 1 or 2° of visual angle (Cartesian coordinates) and maximum horizontal and vertical stimulus eccentricities were 24 and 21°, respectively. Standard operant methods were used to train monkeys to look toward visual targets for liquid reward (drop of water or juice).
This report presents neural data associated with performance of 2 variations of a saccadic choice task (Fig. 1). The primary objective of this experiment was to evaluate the degree to which the "delay period" activity of central thalamic neurons reflects behavioral context. The logic of the tasks (which are detailed below) is straightforward. In each version of the choice task the RF stimulus remains physically invariant, but changes in behavioral relevance are dictated by a cue that designates it as either the target of a rewarded saccade or as an irrelevant distracter.
|
MEMORY-GUIDED CHOICE TASK. The memory-guided choice task (Fig. 1B) was identical to the visually guided version up to and through the after-cue period (Panel 3). At this point, instead of extinguishing the fixation point to mandate a saccade, both the target and distracter were extinguished (Panel 4), requiring the monkey to remember the location of the target or movement vector throughout an additional interval (memory period; 600 ms, fixed; 500 or 700 ms, variable). The memory period terminated with the GO signal (offset of fixation light), signaling the monkey to make a saccade to the remembered location of the target.
Visually guided and memory-guided choice trials were randomly interleaved and equally probable. Variable intervals and/or random interleaving of trial types were a critical element of the experimental design and prevented monkeys from predicting the time of the GO signal. At a minimum (20/49 neurons), temporal uncertainty was created by interleaving fixed-interval visually guided and memory-guided choice trials resulting in randomized Cue-GO intervals of 300 (visually guided) and 900 ms (memory-guided: 300-ms cue interval plus 600-ms memory period). At a maximum (29/49 neurons), 6 Cue-GO intervals, ranging from 300 to 1,350 ms, were randomized by interleaving variable-interval visually guided and memory-guided choice trials.
SINGLE-TARGET TASKS. The spatial selectivity of each neuron was also examined in association with single stimulus (target) visually guided and memory-guided saccade tasks. In these tasks (described previously, Wyder et al. 2003a), the fixation light was illuminated red, and a single color-matched eccentric target was subsequently presented. As for the choice tasks, monkeys maintained fixation on the central stimulus for a fixed (1,000 ms) or variable (750 or 1,000 ms) delay period, after which the fixation spot was extinguished (GO signal) and a saccade to the target required. On visually guided trials, saccades were directed to a persistent visual stimulus, whereas, on memory trials, the target was extinguished before the GO signal and the saccade directed to the location of the now-absent target.
The single-target tasks were used to estimate the direction selectivity of task-related modulations and provided the primary bench-mark for evaluating the activity recorded on choice trials (see following text). For approximately one half of the neurons, the 2 single-stimulus and 2 choice tasks were randomly interleaved and equally probable (P = 0.25), whereas for the remaining neurons, single-stimulus and choice trials were presented in separate blocks. When blocked, single-target trials preceded choice trials.
STIMULUS LOCATION. In all tasks, stimulus locations were selected from a circular array of 8 possible locations (see Wyder et al. 2003a for details). Target direction was randomized across trials and stimuli were presented at a fixed eccentricity of either 6, 10, or 20° (i.e., the radius of the circle). Eccentricity was chosen based on an on-line estimate of that which produced the strongest task-related modulation.
Data analysis procedures
DIRECTION SELECTIVITY. A neuron's RF was delineated by estimating a preferred direction for each eccentricity tested. Tuning functions were generated from activity recorded during performance of the single-target tasks and the eccentricity that yielded the greatest direction selectivity was considered for subsequent analyses. Our methods of evaluating direction tuning were previously described in detail (Wyder et al. 2003a). Briefly, average firing rates for each of the 8 target directions were calculated separately for both a visual and a saccade-related epoch. Each epoch was 100 ms in duration and corresponded to intervals of maximal task-related modulation (see Wyder et al. 2003a for details). Plots of average firing rate versus target direction were then fit with Gaussian functions, the means of which provided estimates of preferred direction. The visual epoch was used to estimate preferred direction whenever possible. However, in cases of weak or absent stimulus-related modulation, tuning was based on the motor epoch.
Figures 2 and 3 illustrate, for an example neuron, the procedures for determining preferred direction (Fig. 2) and for evaluating the capacity of single thalamic neurons to discriminate between target and distracter stimuli (Fig. 3). Figure 2 depicts data from single-target visually guided saccade trials for a single neuron. The polar plot of Fig. 2A compares average firing frequency as a function of target direction for a poststimulus epoch (filled symbols) to that for a prestimulus baseline period (open circles), with stimulus-related activity clearly illustrating a preference for rightward (contralateral) stimuli. In Fig. 2B, stimulus-related firing as a function of direction is fit with a Gaussian function that yielded an estimate of 334° as the preferred direction for this neuron.
|
|
), it was common for 2 or 3 points (target/stimulus locations) of similar firing rate to straddle the peak (within-RF) of the Gaussian function. Analogously, the asymptote of the Gaussian function was typically fit to 2 or 3 points of a similarly low firing rate. In all but 2 cases (47/49 neurons), firing rates were each based on averaging the activity across 2 stimulus directions for which rates were comparable. In the remaining 2 cases, sufficiently sharp direction tuning permitted use of a single stimulus direction. Figure 2C plots both rasters (top) and average firing frequency histograms (bottom) for a target within (black; 315 and 0°) and opposite (gray; 135 and 180°) the RF. Task events are depicted above each set of rasters. Activity profiles are shown aligned on stimulus/target onset (left panel) and saccade onset (right panel). Note that, beginning about 100 ms after stimulus presentation (t = 0), activity increases for the within-RF target (black) and decreases (transiently) for the oppositely directed non-RF target (gray). This difference in firing is maintained throughout the delay period and subsequent reaction time (beginning with GO signal), leading up to saccade onset (t = 0; right panel), at which time, preferred-direction activity is abruptly suppressed (100-ms baseline and poststimulus epochs used to construct Fig. 2, A and B are indicated by horizontal lines above average firing histogram).
EVALUATING THE NEURAL DISCRIMINATION OF TARGETS AND DISTRACTERS. The neural discrimination analysis hinged on comparing activity for the same neuron both before and after the stimulus in the RF was revealed to be either a target or a distracter. Figure 3 illustrates this for the same neuron depicted in Fig. 2. Rasters and average firing frequency histograms (Fig. 3A) are shown aligned on stimulus onset (left) and saccade onset (right) for activity associated with the visually guided choice task. Trials are separated according to whether the stimulus in the RF was ultimately revealed to be the target (red) or distracter (blue). For comparison, the average firing frequency histograms for the single-target trials (black: single target within RF; gray: single target opposite RF) of Fig. 2 are also shown here (trials in which neither a target nor a distracter was within the RF of the neuron were not considered for subsequent analyses).
Corresponding task events for choice trials are shown in the panels above each set of rasters. Stimulus probabilities were set such that, on any given trial, the stimulus in the RF was equally likely to become a target (red rasters) or a distracter (blue rasters) and, critically, targets and distracters were equally likely to be red or green (see bifurcation in the task panels above each set of rasters). In this example, there were 147 trials in which the RF stimulus became the target (76 red, 71 green) and 148 trials in which the RF stimulus became the distracter (68 red, 80 green). The random assignment of color (red/green) and stimulus identity (target/distracter) ensured that, if present, differences in firing rate attributed to stimulus identity (target or distracter) could be distinguished from those attributed to preference for the color of the RF or fixation stimulus (see following text).
As expected, activity during the before-cue period (t = 0–700 ms) did not differentiate between stimuli that, only later in the trial, would become distinguishable on the basis of their relationship to the color of the fixation light (i.e., target or distracter). However, about 200 ms after presentation of the cue, activity evolved to differentiate between a target or distracter within the RF with the difference in level for these conditions approximating or even exceeding that for single-target trials (black: target within RF; gray: target opposite RF). A neuron was considered to have "discriminated" between a target and distracter if, by 300–350 ms after the cue, a statistically significant (t-test; P < 0.05) difference in firing rate had developed. For the example neuron, the mean firing rate for a target in the RF was 95.8 ± 23.2 spikes/s and for a distracter in the RF was 42.6 ± 34.4 spikes/s, a significant difference.
The degree and time course of target-distracter differentiation was further quantified by adapting an analysis previously used for neurons in the FEF and the superior colliculus (SC) (McPeek and Keller 2002; Thompson et al. 1996). The resulting discrimination functions provided a means to visualize and quantify the time course of discrimination, providing bases for comparing individual neurons within the sample and for comparing this sample to comparable data from other reports (e.g., McPeek and Keller 2002; Thompson et al. 1996; see DISCUSSION). We evaluated the degree to which the "target" and "distracter" response profiles differed in 5-ms increments, for each interval calculating the probability that the firing rate on "target" trials would exceed that on "distracter" trials (or vice versa in cases of target-related suppression). The resulting probability value, referred to as the discrimination index (DI), is analogous to the area under a ROC curve and provides an index of how reliably the presence of a target or a distracter in the response field could be predicted based on the response of the neuron.
For each trial, instantaneous firing frequency was calculated by taking the reciprocal of each interspike interval. The resulting plots of instantaneous firing frequency versus time were then binned at 1 ms and subjected to a moving average to yield an estimate of mean firing frequency in 5-ms increments. At each 5-ms time point, the value of mean firing frequency was the product of averaging across 10 (1-ms) bins (5 ms before to 5 ms after). For each 5-ms interval the DI was calculated to give the probability that the firing rate on target trials exceeded that on distracter trials according to the formula
 |
Values of DI are plotted in Fig. 3B for the time period beginning 100 ms before the cue interval. Note that values hover near 0.5 during the before-cue interval, begin to rise about 200 ms after the cue, and rise monotonically over the subsequent 150 ms, to reach an asymptotic value of near 0.9. We quantified this time course by fitting functions of DI versus time with the modified cumulative Weibull function given by
 |
). Here max and min are, respectively, the maximum and minimum DI values calculated for a given neuron (Thompson et al. 1996). They were obtained by averaging the DI values during the first and last 50 ms, respectively, of the interval over which the Weibull function was fit. Generally, good fits were obtained using the interval from cue onset until 100 ms after the end of the cue interval (400 ms total), but small adjustments to this interval were made in some cases to accurately capture the discrimination process. The onset and offset of discrimination were estimated as the time at which the best-fit Weibull function reached 25 and 75% of its maximum value, respectively. For the neuron shown in Fig. 3, the discrimination function rose from 0.49 to 0.91, reaching 25% of maximum at a postcue time of 245 ms and 75% of maximum at a postcue time of 315 ms. For comparison, Fig. 3, C and D illustrate the same analysis, performed on the same data set, but with trials parsed according to the color of the RF stimulus (Fig. 3C) or the color of the fixation stimulus (Fig. 3D). In both analyses, a preference for red is indicated by DI > 0.5 and a preference for green by DI < 0.5. The DI function for RF stimulus color (Fig. 3C) hovers above 0.5, suggesting a weak preference for red over green stimuli. In contrast, the DI function for fixation stimulus color (Fig. 3D) declines (to <0.5) after the cue changes, indicating a preference for a green fixation stimulus. Compared with the effect of target/distracter identity, the influences of color are subtle. At 300–350 ms after the cue, the target/distracter DI value differed from 0.5 by 0.34, whereas deviations for RF-color and fixation color were much less pronounced at +0.1 and -0.13, respectively. For all neurons that discriminated between targets and distracters, a 2-way ANOVA was performed to test for main effects of RF stimulus color, fixation stimulus color, and their interaction.
SINGLE-TARGET DISCRIMINATION FUNCTIONS. To provide a bench-mark for comparing the degree of discrimination achieved on choice trials, DI analyses were also applied to single-target trials by comparing trials in which the lone stimulus was in the RF to trials in which the stimulus was opposite the RF. Differences between responses to single targets within and opposite the RF were quantified.
Histology
During a single, final recording session, electrolytic lesions were made by passing 10 µA for 20 s at several locations. Lesion sites were chosen to mark the locations and boundaries of the regions where neurons were recorded. One week postlesion, monkeys were sedated with ketamine, administered an overdose of sodium pentobarbitol, exsanguinated, and perfused with heparanized saline and 4% paraformaldehyde. The brain was blocked, equilibrated in 30% sucrose, and frozen sections were cut at 50 µm thickness. Every other section was mounted and stained for Nissl substance (cresyl violet).
|
RESULTS |
|---|
|
TOPABSTRACTINTRODUCTIONMETHODSRESULTSDISCUSSIONACKNOWLEDGMENTSREFERENCES |
|---|
). The present study is based on a subset of a larger sample that contributed to an earlier report (Wyder et al. 2003a). The majority of neurons were recorded bilaterally from one monkey (n = 42) with minor contributions from 2 additional monkeys (n = 4 and n = 2). Functional distinctions among central thalamic neurons
As described in the INTRODUCTION, used in conjunction, the behavioral tasks used here had the potential to reveal important functional distinctions among neurons in central thalamus. In principle, when the activity of the same neuron is considered across single-target and choice trials and visually guided and memory-guided trials, it should be possible to determine whether the task-related activity of that neuron represents the locations of visual stimuli (independent of task requirements), the locations of only those visual stimuli identified as saccadic goals, or the locations of saccadic goals independent of the presence of a visual stimulus.
The neurons depicted in Figs. 4 and 5 illustrate the range of "context" dependency exhibited by this sample of central thalamic neurons. At one extreme is a neuron (Fig. 4) that responded unconditionally to the presence of a visual stimulus in its response field. At the other extreme is a neuron (Fig. 5) having a response that was conditional on the knowledge that its response field contained a saccadic goal.
|
|
Interpreting sustained delay period activity solely on the basis of single-target visually guided trials is problematic. One plausible interpretation is that delay-period activity signaled the continued presence of a saccadic goal in the response field. Alternatively, this activity, which seemed to increase as the trial progressed, could have been a signal of motor preparation for the impending saccade. Note also that these possibilities are not mutually exclusive, in that a sensory-contingent goal-related signal could have yielded to a motor preparation signal as the impending saccade drew nearer (e.g., see Sato and Schall 2003; Thompson et al. 1996).
The activity profiles associated with the choice task and memory-guided tasks appear to rule out the "goal-related" interpretations detailed above, instead suggesting that the activity of this neuron signaled the locations of sensory stimuli largely independent of task objectives. For example, when 2 stimuli were present, one within and one opposite the RF (Fig. 4A; red and blue traces), activity was virtually identical to that for a single target in the RF (black trace). Thus the neuron did not distinguish between a stimulus that was known to be a target (black trace) from one that merely had the potential to become a target (red and blue traces). More telling is the after-cue epoch shown in Fig. 4B. Here, it is apparent that activity failed to discriminate between a stimulus that had been identified as a target (red trace) and one that was known to be a distracter (blue trace). This ambivalence to the task relevance of the stimulus even carried into the saccadic period (Fig. 4C), which showed roughly equivalent perisaccadic activity for saccades directed toward or away from the RF.
The activity of the same neuron on memory trials provided further evidence that the presence of a visual stimulus, independent of context, was the primary determinant of this neuron's response. Figure 4D shows that activity ceased abruptly about 100 ms after the stimuli were extinguished and remained low throughout the saccadic period [although there is some hint of residual saccade-related activation for saccades in the preferred direction (black and red traces)].
The neuron shown in Fig. 5 provides a strong counterexample, in that its activity was conditional on knowledge that the stimulus specified the location of a saccadic goal. Furthermore, once established, this representation did not require the continued presence of the stimulus. On choice trials (red and blue traces), the presence of 2 stimuli (each a potential target) during the before-cue portion of the delay period yielded a response rate intermediate to that for single targets (black and gray traces). However, about 250 ms after the cue identified the target and distracter (Fig. 5B; cue), the activity profiles diverged, with activity increasing if the stimulus in the RF was revealed to be the target (red), and decreasing if revealed to be the distracter (blue).
Figure 5D shows that, whether for single-target or choice trials, direction-selective activity persisted in the absence of the stimulus, maintaining a similarly high level of differentiation during the memory period for both trial types. Likewise, the saccade-related modulation of this neuron was virtually identical, whether for saccades to visual goals or to their remembered locations (Fig. 5, C and E). In either case, sustained activity preceding movement to a preferred-direction goal was sharply suppressed just before saccade onset, a motor-related response pattern not uncommon among neurons in the central thalamus (Schlag and Schlag-Rey 1984; Schlag-Rey and Schlag 1984; Wyder et al. 2003a).
Key differences between the example neurons are apparent in Fig. 6, which quantifies and compares the degree of discrimination for single-target trials (target in RF vs. target opposite to RF) to that achieved for target/distracter discrimination on choice trials. This comparison is made for the neurons of Figs. 4 and 5 (Fig. 6, A and B) and for a third neuron (Fig. 6C) that showed a conditional response that was intermediate to those discussed above. The DI (see METHODS; Fig. 3B) is plotted as a function of time for activity synchronized on relevant task events including stimulus onset (left column), cue change (middle column), and stimulus offset (right column; memory trials only). Generally, we found that DI values > 0.65 corresponded to statistically significant differences in firing rate (see Fig. 8 below) and a dotted line demarcating this value is provided as a reference on each graph. The discrimination functions shown in Fig. 6A confirm that the "stimulus-contingent" neuron of Fig. 4A is very selective for direction, discriminating at a high level (DI 
0.9) for a target within versus a target opposite the RF for as long as a stimulus is present (thin trace, left and middle columns). In contrast, on choice trials (thick trace) the persistence of DI values near 0.5 throughout the delay period, even after the cue change (top row, middle column), reflects the fact that this neuron fired equivalently for a target or distracter in its RF. On single-stimulus memory trials, the strong preference for a stimulus in the RF was lost because this activity required the physical presence of the stimulus (thin trace; top row, right column).
|
|
0.8; middle row; middle column) that was maintained throughout the memory interval (middle row; right column). On choice trials (thick trace) the discrimination function began to rise about 250 ms after the cue was presented (middle row; middle column), with target/distracter discrimination on choice trials ultimately reaching the same level as the within-RF/opposite-RF discrimination on single-target trials. In both cases, a high level of discrimination was maintained throughout the memory period (middle row; right column). The neuron shown in Fig. 6C was intermediate; this neuron was direction selective on single-target trials (thin trace) and selective for targets versus distracters after the cue on choice trials (thick trace; bottom row; middle column), but showed discrimination that was strongly dependent on the continued physical presence of the visual goal (bottom row; right column). As such, the activity of this neuron showed a dual dependency, conditional on both the presence of a visual stimulus and the knowledge that the stimulus was the goal of an impending saccade. The activity profiles for this neuron, along with those from a similar neuron, are shown in Fig. 7, A–E and F–J, respectively. The DI functions shown in Fig. 6C (light traces) indicate that neurons of this type were selective for direction on single-target trials for as long as the stimulus was visible. Accordingly, single-target firing rates converged about 150 ms after stimulus offset (Fig. 7, D and I; black and gray traces). The stimulus dependency of the target/distracter discrimination is more obvious on choice trials in which an even larger difference in firing rate, developed during the after-cue period, dissolved within the same time frame (Fig. 7, D and I; red and blue traces). The after-cue discrimination for these neurons was the product of both increases (for targets, red traces) and decreases (for distracters, blue traces) in firing from a before-cue level that was intermediate to that for single targets within (black traces) and opposite (gray traces) the RF.
|
To characterize discrimination for individual neurons and for the sample as a whole, DI functions were narrowed to 3 values each for single-target and choice trials. These values were computed by averaging DIs over selected 50-ms epochs. For choice trials, these consisted of one interval before cue presentation, one after cue presentation, and one during the memory period. For single-target trials, the temporally corresponding epochs consisted of 2 from the delay-period (one early, one later; see following text) and one from the memory period. The sample intervals are indicated by horizontal line segments in Fig. 6 (middle and right columns). The before-cue period corresponded to the 50 ms leading up to the cue (change in fixation color). The analogous single-target period is referred to as "early delay" and is the interval from 350 to 300 ms before the end of the visual delay period.
To capture target/distracter discrimination at its fullest development, the after-cue period corresponded to the first 50 ms after the end of the delay period (i.e., 300–350 ms after presentation of the cue). Although the beginning of this sample period coincided with offset of the fixation stimulus (nonmemory trials) or peripheral stimuli (memory trials), visual afferent delays on the order of 100 ms (see Wyder et al. 2003a) preclude the influence of these sensory events on activity during this initial 50 ms. In subsequent text (and figures), this epoch is referred to as "after-cue" (choice) and "late delay" (single-target). Finally, to obtain a steady-state measure of "memory-period" DI, uncontaminated by stimulus (offset)-related transients, the final sample was drawn from well within the memory period (from 450 to 500 ms).
Figure 8 shows the distributions of DI values associated with single-target (left column) and choice (middle column) trials for each of the 3 sample epochs (Rows 1–3) for the entire sample of neurons. On the single-target tasks, it is apparent (using 0.65 line as reference) that many neurons were tuned for direction and thus discriminated between a single stimulus within and opposite the RF during both the early and late visual delays (Fig. 8, A and B). Overall, 69% (34/49) and 57% (28/49) of the early and late visual delay epochs yielded significant differences (open bars) in mean firing rate (t-test; P < 0.05), with the means of these distributions nearly identical (early: 0.69; later: 0.70) and indicative of reasonably good spatial selectivity during the period that the stimulus was present. Significant differences in firing rate were less prevalent during the memory interval with 35% (17/49) of neurons discriminating between a stimulus previously present within or opposite the RF. Accordingly, the mean DI was nearer to 0.5 (0.56).
On choice trials, DI values were distributed symmetrically around 0.5 (0.49 ± 0.08) for the before-cue interval, with only 3 (46/49) instances of a significant difference in firing rate (t-test; P < 0.05). This, of course, is the expected outcome given that the stimulus configurations for the trials that compose these 2 groups were identical with the grouping based on a future change (after cue) in stimulus status (to target and distracter). In contrast, after the cue was presented to signify one stimulus as the target and the other as the distracter, more than one third (17/49; 35%) of the sample developed a significant difference (open bars) in mean firing rate, as evidenced by the rightward shift in mean DI (Fig. 8E, 0.60 ± 0.14). We once again emphasize that the development of discrimination across Fig. 8, D and E reflects a change in RF-stimulus significance, not a change in the physical properties of the stimulus within or opposite to the RF (see Fig. 1 for task description).
On choice trials, the incidence of target/distracter discrimination during the memory interval was identical to that during the after-cue period, with 35% of the cases (14/40) having significantly greater activity for targets than distracters in the preferred direction (Fig. 8F). Overall, the number of observations (n = 40) was slightly reduced because of exclusion of neurons for which there were fewer than 7 successful memory trials for each location. For 3 neurons, significant discrimination resulted from greater activity for distracters in the preferred direction (DI < 0.35; open bars, Fig. 8F). This some-what counterintuitive finding was the result of an apparently active response suppression triggered by offset of the RF target on memory trials.
The scatter plots of Fig. 8, G–I relate DI values for the comparable epochs on single-target and choice trials. As expected, DI values are predominantly below the line of equality in Fig. 8G, indicative of discrimination during the early visual delay period on single-target trials, but not during the beforecue period on choice trials. In Fig. 8, H and I, DI values are more evenly distributed about the line, indicating that, after the cue on choice trials, discrimination between a target and distracter in the RF reached a level similar to that for a single target within and single-target opposite the RF for many neurons.
Differential discrimination across sample epochs as a basis for categorization
The majority of our sample (28/49; 59%), like the 3 example neurons, were selective for direction during the late visual delay period on single-target trials (see Fig. 8B). Our main findings were that many of these neurons did discriminate between targets and distracters and tended to conform to one of the types depicted in Fig. 6, B and C. Most "nondiscriminating" neurons were similar to the neuron depicted in Fig. 6A.
For just over half of the neurons with delay-period activity (15/28; 54%), activity after the cue (during the after-cue epoch) on choice trials evolved to discriminate a target from a distracter. For 13 of these target/distracter "discriminating" neurons, there were a sufficient number of memory trials to further define them on the basis of whether discrimination persisted after the stimuli were extinguished. Most (8/13; 62%) were similar to the example shown in Figs. 5 and 6B, with target/distracter differentiation persisting late into the memory period. The other 5, however, followed the pattern of the example neurons shown in Figs. 6C and 7 such that differentiation was lost soon after the targets were extinguished on choice trials.
The remaining 13 neurons with direction-selective delay period activity failed to discriminate between a target and distracter during the after-cue period on choice trials. Of these, 12 could be further characterized on the basis of memory trials. Most of these neurons (7/12) could be described as "stimulus-bound," failing to convey spatial information in the absence of the target on either single-target or choice memory trials; these neurons resembled the example depicted in Figs. 4 and 6B. Four of the 5 remaining neurons maintained spatial selectivity during the memory period on single-target trials, whereas 3 of the 5 developed the selectivity during the late memory period on choice trials. For these latter 3 neurons, differentiation occurred very late relative to when the cue was provided and may have reflected preparation for the impending saccade.
Summarizing the breakdown described above, of the 25 neurons that showed direction selectivity during the late delay period, and which could be further defined on the basis of activity on choice and memory trials, 20 conformed to one of the 3 example discrimination profiles shown in Fig. 6A (n = 7), 6B (n = 8), and 6C (n = 5). The number of units in the latter 2 categories could be augmented to n = 9 and n = 7, respectively, by considering 3 additional neurons that approached statistical significance for direction selectivity during the single-target delay period and achieved significance for both the after-cue and memory periods (n = 1; e.g., Fig. 6B) or for the after-cue period alone (n = 2; e.g., Fig. 6C). Unlike these 3, the majority (16/20) of those that did not have direction-selective delay-period activity also failed to discriminate between targets and distracters during the analogous after-cue period on choice trials, as expected.
As noted above, most neurons with spatially selective delay period activity on single-target trials conformed closely to one of the 3 patterns depicted in Fig. 6. Rather than an artificial parsing of a response continuum based on statistical criteria, these groups appeared to be the basis for true qualitative distinctions. Consistency among members within each group is evident by comparing the averaged DI functions shown in Fig. 9, A–C to those for the example neurons shown in Fig. 6, A–C. For the plots in Fig. 9, DI values were averaged across neurons with similar response profiles, as described above. Although averaging substantially reduced the noise, the magnitude and timing of the modulations that characterize the groups remained largely intact.
|
Tasks were designed so that neural selectivity for the color of either the eccentric stimulus or fixation stimulus would not confound evidence for target/distracter discrimination (see METHODS; Fig. 3). Targets and distracters were not associated with a particular eccentric or fixation stimulus color, but were defined by the relationship between the colors presented in the RF and at fixation (match vs. nonmatch). We had no reason to anticipate strong preferences (either neural or behavioral) for a particular stimulus or fixation color as such would not have aided (and could have hindered) task performance. On the whole, we found little evidence of a general preference for color among neurons that were selective for targets versus distracters. When trials were instead parsed according to RF stimulus color or fixation stimulus color, DI values deviated little from 0.5. Based on the same after-cue time interval (300–350 ms) used to compute target/distracter DI values, DI values for RF-stimulus color and fixation stimulus color deviated from 0.5 by ±0.08 and ±0.07, respectively. These deviations are small by comparison to an average deviation of ±0.25 for target/distracter differentiation.
A 2-way ANOVA (RF stimulus color x fixation stimulus color) was performed to further examine the relative potency of the effects of RF stimulus color, fixation stimulus color, and stimulus identity (target vs. distracter selectivity reflected in magnitude of the interaction) on target-/distracter-selective neurons. The ANOVA evaluated activity for the same 50 ms after-cue period used to test for significant differences in target versus distracter firing rates (Fig. 8E). The activity of relatively few target-/distracter-selective neurons showed significant main effects for RF-stimulus color (6/16), fixation stimulus color (3/16), or both (3/16). In contrast, in all but one case (which approached significance at P = 0.054), the presence of a significant interaction confirmed selectivity for targets (color matches) versus distracters (color nonmatches). Moreover, in each of the 6 cases in which a main effect of color was observed, the interaction was much larger than the main effect(s). This difference is evident in the corresponding DI values. For neurons that showed a main effect of RF-stimulus color (n = 6), DI values deviated by ±0.12 from 0.5, compared with ±0.28 for target/distracter differentiation for these same neurons. Similarly, for neurons showing a main effect of fixation-stimulus color (n = 3), DI values deviated by ±0.18 from 0.5, compared with ±0.34 for target/distracter differentiation.
Timing of target–distracter differentiation
The activity of each of the neurons contributing to the average DI functions in Fig. 9, B and C was differentially modulated in response to the cue identifying the locations of the target and distracter relative to its RF. In principle, this evolving activity reflects the processes of detecting the cue change, registering the cue color, and identifying the matching stimulus (or by elimination, the nonmatching stimulus). The time course of neural target–distracter discrimination was quantified by estimating the time at which activity began to reflect the changed status of the stimuli and the time at which this gradually increasing difference in activity approached its maximum. As shown in Fig. 3B and described in METHODS, for each neuron, the onset and offset of differentiation were given as the times at which a best-fit cumulative Weibull function reached 25 and 75% of its maximum value. Weibull fits were generally quite good for neurons that discriminated between targets and distracters; correlation coefficients (plots of predicted vs. observed) ranged from 0.81 to 0.98, with a mean of 0.94. The estimates derived from these fits are plotted in Fig. 10. Mean onset time (Fig. 10A) and mean offset time (Fig. 10B) were 171 (±77) and 253 (±60) ms after presentation of the cue, respectively. On average, the DI function rose from minimum (25%) to maximum (75%) over a period of 82 (±42) ms (Fig. 10C). A best-fit line to the scatter plot relating 75 and 25% points on the cumulative Weibull function had less than unity slope at 0.67 and an intercept of 139 ms, suggesting that later starting increases in the DI function required slightly less time to reach their maximum (Fig. 10D). There was no evidence that timing for neurons that differentiated during the memory interval (filled symbols) and those that did not (open symbols) were different by these measures.
|
Neurons that differentiate between targets and distracters in the RF could contribute to the process of identifying and/or selecting the appropriate stimulus for an upcoming saccade. Presumably then, trials in which the RF stimulus was apparently misclassified would be associated with either less-robust target/distracter differentiation or perhaps even neural discrimination of opposing sign (consistent with incorrect choice). Unfortunately, monkeys made very few choice errors (on average 6 error trials per recorded neuron), providing no opportunity to make statistically meaningful comparisons for an individual neuron. However, we could evaluate error trials by pooling across all target-/distracter-selective neurons (n = 16). To do so, we compared normalized firing rates for the 4 possible outcomes: 1) correct: target in RF, saccade to target; 2) correct: distracter in RF, saccade to target; 3) error: target in RF, saccade to distracter; and 4) error: distracter in RF, saccade to distracter. For each neuron, the mean firing rate during the after-cue period (300–350 ms after cue) was calculated for each of the 4 conditions and normalized to the rate obtained for correct trials to targets in the RF (set to 1.0). These data indicate that, on average, differentiation on error trials was both less robust (firing rate difference not significant; t-test; P > 0.05) and opposite in polarity to that observed on correct trials. On correct trials, mean firing rate for a distracter in the field (n = 785) was about half (0.52) that for a target in the field (1.0; n = 787). Firing rates were intermediate for error trials with errant saccades to a distracter in the field (n = 56) associated with a higher rate (0.73) than failures to saccade to a target in the field (0.62; n = 42). These data suggest that choice errors resulted when neural activity provided ambiguous or erroneous information about the identity of the stimulus in the RF. We emphasize, however, that a task that deliberately elicits a greater proportion of errors would be needed to provide a rigorous test of this hypothesis.
Recording sites
Figure 11 plots the locations of 45 of the 49 neurons included in this report. Two of the remaining 4 units were recorded at a location 0.5 mm rostral to the section shown in Fig. 11A, and for the final 2 units, histology was not available; however, MRI images of electrode placement indicate that these units were located in the central thalamus. Recordings in one monkey ranged from A.P. 6.5 to A.P. 8 (Fig. 11, A–C, section 6.5 not shown), and in the second monkey they were restricted to A.P. 11 (Fig. 11D). Anterior–posterior levels are estimated based on Olszewski (1952). Units were recorded mainly in the central lateral (CL) and paracentral (Pc) nuclei, and in paralaminar regions of the ventral lateral (VL) and ventral anterior (VA) nuclei. In addition, a few units were recorded in the lateral dorsal (LD) nucleus and in paralaminar regions of the medial dorsal (MD) nucleus.
|
|
DISCUSSION |
|---|
|
TOPABSTRACTINTRODUCTIONMETHODSRESULTSDISCUSSIONACKNOWLEDGMENTSREFERENCES |
|---|
), we demonstrated that many neurons in central thalamus maintain spatial information throughout the instructed delay period of a visually guided delayed-saccade task. These earlier findings established that the activity of central thalamic neurons can bridge the gap between the sensory encoding and motor execution phases of a delayed-saccade task; however, they did not define a role for these neurons in the context-specific linking of sensory signals to motor commands. As the primary mediators of information transfer from subcortical structures to frontal cortex, we have postulated that central thalamic nuclei could play key roles in the processes that link specific stimuli with specific actions, as required by present circumstances and current behavioral objectives. Consistent with this idea, we report here that the delay-period activity of some central thalamic neurons is a function of both the sensory stimulus and its relevance for guiding a rewarded saccade. Functional distinctions among neurons with delay-period activity
Comparison of delay-period activity across single-target, choice, and memory-trial types revealed that individual thalamic neurons differed in ways suggestive of fundamentally distinct contributions to visuomotor control. In its simplest form, delay-period activation appeared to signal the presence of a stimulus in a particular location (i.e., within the RF), largely independent of task goals (Figs. 4, 6A, 9A). Although selective for stimulus location, such a neuron responded similarly for a potential target, known target [with (choice) or without (single-target) an opposing distracter], or a known distracter in its RF. Consistent with activation determined primarily by the physical stimulus (not context), a neuron of this type ceased firing when execution of a saccade (visually guided trials) or start of the memory period removed the stimulus from the RF.
Strictly stimulus-related delay period activity is simply a tonic, location-specific response to a persistent visual stimulus. Such a neuron, if involved in representing stimuli for the purpose of informing future action, must do so only at the very earliest stages of the sensorimotor translation, contributing neither to perceptual discrimination (target selection) nor to movement selection. Despite only a remote link to motor command formation, tonic visual neurons are not uncommonly observed in what are ostensibly visuomotor areas. For example, a similar stimulus dependency was recently reported for a subset of SC-projecting FEF neurons (Sommer and Wurtz 2001). As noted by Sommer and Wurtz (2001), the fact that such neurons project to the SC, where saccadic motor commands are formed, suggests that "unqualified sensory" information reaches fairly low levels of the saccadic command chain.
In contrast to the apparently low-order stimulus coding exhibited by some central thalamic neurons, we found the delay-period activity of others to be the product of both the location of the stimulus and its status as either a saccade target or an irrelevant distracter (Figs. 5; 6, B and C; 7; 9, B and C). Common to all neurons of this type was activity that evolved to discriminate a target from a distracter in its RF, with this process beginning within 150 to 200 ms of receiving the informative cue. Although both the timing and degree of discrimination attained (i.e., maximum DI value) varied, there were no clear-cut distinctions to be made on the basis of these quantitative criteria. On the other hand, "discriminating" neurons could be subcategorized on the basis of whether they continued to represent the location of the identified goal after the stimuli (target and distracter) were extinguished on choice memory trials. Thus the delay period activity of some thalamic neurons was conditional on both the physical presence of a stimulus in the RF and the knowledge that the stimulus was the target of an impending saccade.
Unlike the neurons with strictly stimulus-dependent activity described above, a thalamic neuron that differentiated between saccadic goals and irrelevant stimuli must contribute to later stages in the process of linking specific visual stimuli (targets) to rewarded saccades. Such a neuron (e.g., Figs. 5; 6, B and C) could have been either an active participant in the perceptual decision process that identifies the RF stimulus as a target or distracter on the basis of its color, or it may have been involved in the postperceptual process of planning a saccade to the identified target.
The neural correlates of perceptual decision-making (e.g., target selection) and postperceptual motor planning processes are logically and practically dissociable (see Schall 2003 for recent review). In FEF, for example, neurons have been shown to participate in one, the other, or both (Murthy et al. 2001
