Cerebral Cortex Advance Access originally published online on October 24, 2007
Cerebral Cortex 2008 18(6):1272-1280; doi:10.1093/cercor/bhm158
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
The Representation of Spatial Attention in Human Parietal Cortex Dynamically Modulates with Performance
1 Department of Cell Biology, Neurobiology and Anatomy, 2 Department of Radiology, Medical College of Wisconsin, Milwaukee, WI 53226, USA, 3 Department of Human Movement Sciences, University of Wisconsin Milwaukee, PT Program, Milwaukee, WI 53201-0413, USA
Address correspondence to Wendy E. Huddleston at Department of Human Movement Sciences, University of Wisconsin Milwaukee, PT–Pavilion 350, PO Box 413, Milwaukee, WI 53211, USA. Email huddlest{at}uwm.edu.
 |
Abstract |
|---|
 Top Abstract IntroductionMaterials and MethodsResultsDiscussionSupplementary MaterialFundingReferences |
|---|
The control and allocation of attention is an essential, ubiquitous neural process that gates our awareness of objects and events in the environment. Neural representations of the locus of spatial attention have been previously demonstrated in parietal cortex. However, the behavioral relevance of these neural representations is not known. While undergoing functional magnetic resonance imaging, subjects performed a covert spatial attention task that yielded a wide range of performance values. Voxels in parietal cortex selective for attended target location also dynamically modulated, becoming more or less responsive as performance levels changed. Surprisingly, this relationship was not linear. Responses peaked at intermediate performance levels and dropped both when performance was very high and when it was very low. Such dynamic modulation may represent a mechanism for organizing neural control signals according to behavioral task demands.
Key Words: behavior • functional magnetic resonance imaging • vision
|
Introduction |
|---|
|
TopAbstractIntroductionMaterials and MethodsResultsDiscussionSupplementary MaterialFundingReferences |
|---|
In a crowded environment, spatial attention allows the observer to focus on behaviorally important information at a particular location in space while ignoring irrelevant information. Top-down, or voluntary, attentional control signals within the brain can alter the neural activity in primary sensory cortex so that task-relevant information is preferentially processed. Within the human visual system, voluntary attentional modulation has been robustly demonstrated (Tootell et al. 1998
; Brefczynski and DeYoe 1999; Somers et al. 1999; Huk and Heeger 2000; Hopfinger et al. 2001; Boynton 2004), and this modulation occurs in a spatially specific manner concordant with cortical retinotopic organization (Tootell et al. 1998; Brefczynski and DeYoe 1999; Kastner et al. 1999). To mediate such spatial selection, we hypothesize that the neural signals controlling attention are themselves organized as a spatial representation but that this organization is established dynamically to meet task demands at any given moment. If not needed, this organization may wane or disappear until again required by the task at hand.
Parietal cortex is a candidate region for the source of the attentional signals controlling earlier visual areas because it has direct neural connections to occipital cortex (Lewis and Van Essen 2000), and it has been implicated in a variety of attention-demanding tasks including task switching (Kimberg et al. 2000; Sohn et al. 2000; Rushworth et al. 2001, 2002), attention to features (Corbetta et al. 1995; Vandenberghe et al. 2001; Shulman et al. 2002), and shifts of visuospatial attention (Corbetta et al. 1995; Yantis et al. 2002). Also, maps of attentional space with a consistent topography have been identified in parietal cortex of monkeys (Ben Hamed et al. 2001; Crowe et al. 2004; Wardak et al. 2004; Heider et al. 2005; Raffi and Siegel 2005; Bisley and Goldberg 2006) and in humans (Sereno et al. 2001; Silver et al. 2005; Swisher et al. 2007). However, it is unclear how these maps relate to performance of a visuospatial attention task. Serendipitously, we obtained a possible clue to this relationship. In pilot experiments preceding the work reported here, we observed marked differences in the robustness of attention-related functional magnetic resonance imaging (fMRI) signals in parietal cortex of different subjects, with some showing strong spatial tuning for attended target location, whereas others had much weaker tuning. Remarkably, we found that we could convert subjects with poor tuning into subjects with strong tuning or vice versa if the task parameters were modified individually for each subject so as to affect their performance on the task. The study reported here seeks to document this phenomenon and presents an initial exploration of its neural basis and significance for attention-related behavior. We demonstrate that blood oxygen level–dependent (BOLD) fMRI signals within parietal cortex relate to performance in a nonlinear manner during a covert spatial attention task and that this relationship is not adequately explained by individual stimulus or task-related parameters.
Preliminary accounts of this work have been reported in abstract form (Organization for Human Brain Mapping Annual Conference 2004 [#TH49], Society for Neuroscience Annual Meeting 2004 [#709.8], and 2005 [#129.1]).
|
Materials and Methods |
|---|
|
TopAbstractIntroductionMaterials and MethodsResultsDiscussionSupplementary MaterialFundingReferences |
|---|
Subjects
Nine subjects (23–53 years, 4 female, 2 left handed) participated in the experiment. All subjects provided informed consent as approved by the Institutional Review Board of the Medical College of Wisconsin. All participants reported normal or corrected-to-normal vision and no neurological deficits.
Stimulus, Task, and Experiment Variants
All stimuli were created with a Visual Stimulus Generator (VSG 2/3) computer video board (Cambridge Research Systems, Rochester, UK). As illustrated in Figure 1, the visual stimulus consisted of a ring of white disks (subtending 0.4 degrees visual angle) and an additional disk at the center fixation point, similar to that of Sereno et al. (2001). The circular array had an average eccentricity of 17°. The position of each disk in the circular array randomly jittered every 500 ms, and each disk could disappear at random intervals. Subjects viewed the stimulus through custom MRI compatible optics with a large field of view (56°) or with prisms and a back projection screen.
|
The task trial sequence is outlined in Figure 1. Subjects were cued to attend to a particular target location by the appearance of a small marker extending from the fixation point for 500 ms. Subjects were instructed to immediately attend to the target disk upon presentation of the cue. The cue then disappeared and was followed by a 4500-ms delay. The flickering and jittering of the disk array was sufficient to require the subjects to monitor the disk for the entire delay (and this is typically what the subjects reported anecdotally). At the end of the delay, a tone sounded and the subject was required to make a two-alternative forced choice button press, indicating whether the covertly attended disk was present (button A) or absent (button B) at the time of the tone. The disk was present 50% of the time. The distractor array was present during the entire scan run, but each individual distracter was present 85% of the time. The experiment was designed such that subjects were cued to attend successively to 1 of 4 target locations in a blocked fashion. During each block, subjects performed the present/absent task at a single location for 3 trials while maintaining fixation, and then rested while fixating and making sham button presses at the appropriate time for 3 trials. Subjects completed 2 blocks for each attended location in the 5-min MRI scan. Subjects repeated the task with different distractor densities (16, 24, or 32 total disks in the array) to alter performance levels. Subjects completed 7 runs for each experimental condition on each day. Due to technical difficulties, 1 subject completed 6 runs in each task condition and another subject 5 runs in each task condition. However, the number of runs within subjects was always consistent for comparisons between task conditions. That is, the same numbers of runs (5, 6, or 7) were compared between task conditions for each subject.
Preliminary data were collected on 2 of the subjects outside the scanner to monitor eye-tracking movements (EyeLink II data collection at 250 Hz, SR Research Ltd, Mississauga, Ontario, Canada) during this task. Performance levels obtained outside the scanner during eye tracking were comparable to those obtained inside the scanner. Eye-tracking data did not show evidence of task-correlated eye movements. Moreover, if subjects overtly shifted their gaze to foveate the target rather than covertly attend to it, the observed spatial tuning of individual voxels for target location (see Results) would have been destroyed, but was not.
Scanning Parameters
All BOLD fMRI experiments were completed on a 1.5 T GE Signa scanner (GE Medical, Milwaukee, WI), equipped with a custom radio frequency gradient head coil (Medical Advances, Inc., Wauwatosa, WI) suited for whole-brain imaging. A GE-EPI sequence was used to collect BOLD fMRI signals. Imaging parameters included: field of view 64 x 64, time echo 40 ms, time repitition 2 s, 6 mm axial slices, 3.75 x 3.75 mm in-plane resolution, and 90° flip angle. Spoiled GRASS T1-weighted (0.97 x 0.97 x 1.1 mm resolution) anatomical images were collected during each scan session. Functional scans were coregistered to the anatomical images for comparisons within each subject and for subsequent display and analysis.
Data Analysis
The 3dDeconvolve analysis in the AFNI software package (Cox 1996) was used to estimate hemodynamic response functions (HRFs) for each attended location and then to identify voxels having a significant response to 1 or more cued target locations. The full statistical model consisted of 4 different reference functions, 1 for each of the 4 attended target locations in the present experiment. An overall F-statistic was calculated based on the full model, indicating significant response modulation to any of the cued target locations. General linear tests for each of the individual reference functions were performed to calculate different partial F-statistics for each voxel. For all subjects, the experiment was repeated using different disk densities, and data sets from 6 of the 9 subjects were collected on multiple days. Data collected across days were indistinguishable from data collected in a single day; so all data were pooled for subsequent analysis. A statistical threshold was independently selected for each subject and used for each experimental condition to maintain the level of Type I error within subject (0.00001 < P < 0.001), uncorrected for multiple comparisons.
Spatial Specificity Index
An amplitude measure was calculated as the area under the curve of the estimated HRF for each attended location. An area under the curve measure was used rather than a peak amplitude measure as it is less sensitive to spurious spikes of activity and better represents the sustained amplitude of the BOLD signal, particularly in a blocked design as used here. This value was then compared across the 4 attended locations to establish the preferred target location for each voxel. Voxels were binned for subsequent analysis based on their preferred target location. The amplitude measure of the HRF for a voxel's preferred location was normalized to minimize the effect of varying baselines between runs and brain regions. To perform the normalization, the amplitude measure of the HRF for the preferred location was divided by the sum of the areas under the curve of the HRFs for all 4 locations. This normalized value provided an index of the difference between the largest amplitude of the preferred location relative to the signal amplitude for all locations and was coined the spatial specificity index (SSI). If the HRF amplitude for the preferred location greatly exceeded the others, the SSI would approach 1. If the amplitude of the 4 HRFs were approximately equal, the SSI would approach 0.25. In this way, differences in signal amplitudes between the 4 attended locations across voxels and tasks could be compared from region to region or session to session. The SSI, then, represents any difference in parietal activation between preferred and nonpreferred target locations for a given voxel. The SSI is sensitive to relative changes in signal amplitude between target locations but is rather insensitive to overall changes in the BOLD signal across all target locations in a given task condition.
Signal Quality
We wanted to analyze another measure of the BOLD signal, in addition to the amplitude-dependent SSI described above, to more robustly quantify the effects of spatial attention on human parietal activation. We opted for a statistical measure (a partial F-statistic) to represent the quality of the BOLD signal (signal quality [SQ]) in terms of the ability of our reference functions to explain variance within the signal. We compared the partial F-statistic (SQ) among voxels preferring different target locations in an attempt to capture alternative mechanisms by which parietal voxels might encode a "preferred location" other than differential activity (signal amplitude) between preferred and nonpreferred target locations. To be clear, this analysis did not involve specific determinations of signal and noise to determine SQ. Instead, we calculated the partial F-statistic on a voxel-wise basis specific to the target location preferred by each respective voxel. We calculated general linear tests and partial F-statistics for each attended location and for a general nonspatial attention effect (3dDeconvolve plugin, AFNI [Cox 1996]) on a voxel-wise basis. Statistically, the partial F-statistics (SQ) represent the level of contribution of the reference function for the preferred target location for each voxel to the fit of the full model across all conditions, in our case the reference functions for each of the 4 target locations.
Signal Quality Changes across Attended Locations
Absolute SQ values between subjects varied greatly and thus needed to be normalized to allow pooling of data across subjects. To accomplish this, the mean SQ data (voxel-wise partial F-statistics) plotted by attended location were fit with a second-order polynomial function for each hemisphere of each subject separately. All data sets were then normalized to a standard value assigned to the center of each curve. A one-tailed Student's t-test was performed to identify significant differences in normalized SQ between ipsilateral and contralateral attended locations for each region of interest (ROI) (
= 0.05).
One concern was that the analysis used to assign a preferred target location to each voxel might potentially assign voxels a preferred location when in reality there were only slight differences in SQ, especially for the 2 contralateral locations relative to each other. Consequently, a post hoc analysis was performed to determine if there were significant differences in SQ between the 2 contralateral locations on a voxel-wise level. A dependent model permutation test (Good 2000) was performed for each pair of targets in the contralateral hemifield (e.g., top left vs. bottom left targets in right parietal cortex). We tested the probability that the SQ for the 2 locations was distinctly different from one another. Two thousand permutations were run using NPStat (v3.8, University of Victoria, British Columbia, Canada) to estimate the actual probability of a Type I error. This analysis confirmed significant differences between the 2 contralateral target locations, and thus, all further data analyses were performed between the 4 attended target locations.
|
Results |
|---|
|
TopAbstractIntroductionMaterials and MethodsResultsDiscussionSupplementary MaterialFundingReferences |
|---|
Nine subjects performed the visuospatial attention task while undergoing fMRI. For all subjects, the location of activity included cortex medial to the intraparietal sulcus (IPS) bilaterally, as well as along both banks of the IPS posteriorly (Fig. 2A). Activation also included the inferior parietal lobule (IPL) bilaterally in all but 1 subject. Voxels preferring contralateral target locations were represented in the highest density, though ipsilateral locations were represented as well.
|
Data were subjected to a cluster analysis in which the centers-of-mass for the areas of activity in parietal cortex were calculated. Figure 2A illustrates the activation pattern for 2 representative subjects. The areas of activity for all subjects typically fell within the SPL (Talairach coordinates left: x = –26 mm, y = –51 mm, z = 47 mm; right: x = 25 mm, y = –59 mm, z = 50 mm; 5 subjects), precuneus (Talairach coordinates left: x = –17 mm, y = –65 mm, z = 42 mm; right: x = 16 mm, y = –54 mm, z = 49 mm; 3 subjects), or the IPLs (Talairach coordinates left: x = –35 mm, y = –52 mm, z = 49 mm; right: x = 34 mm, y = –43 mm, z = 44 mm; 4 subjects). Four of the 9 subjects had 2 separate regions of activation within parietal cortex of each hemisphere. Three subjects had multiple regions of activity in 1 hemisphere and single clusters in the opposite hemisphere, and 2 subjects had single clusters bilaterally that encompassed both the superior parietal lobules (SPLs) and IPL (Fig. 2A,B). Four of the 9 subjects had a greater area of activation in the right hemisphere when compared with the left, though the remaining subjects had no such asymmetry. No subject had a greater area of activation in the left hemisphere than the right hemisphere.
These activated regions served as general ROIs for all further analyses. The target-location ROIs were composed of subsets of voxels within these general ROIs. Quantitative analyses were performed on ROIs defined in a 2-step process. First, voxels demonstrating a significant modulation with attention to any of the 4 target locations (as compared to the fixation-only blocks) were included. Thresholds for identifying significant fMRI modulation were based on the number of false positives, both within and outside the brain. Functional significance thresholds were adjusted for each subject and each scan session to maintain the number of false positives constant (<1%). Voxels exceeding the functional threshold were included in the ROI. In the second step of this process, only voxels exceeding the functional threshold posterior to the postcentral sulcus, anterior to the parietooccipital sulcus, and dorsal to the temporoparietal junction were included in the bilateral parietal ROIs.
Much of parietal cortex activated bilaterally with the present task, and we needed to determine if this region could be further divided into functionally heterogeneous subregions. In particular, we wanted to determine if specific areas within parietal cortex responded differentially to changes in performance. To assess this, we compared data from 4 subjects who completed the attention task with 2 levels of disk density on the same day. We color coded the sign and magnitude of SQ change between the 2 disk density conditions and then looked for obvious clustering according to the level and direction of change, if any. We found intervoxel differences between the 2 task conditions, but no pattern of clusters could be identified consistently across the 4 subjects tested. Therefore, the parietal cortex was evaluated in its entirety as a single ROI. It is important to note that this analysis was not intended to differentiate between potentially different spatial representations of attended location (i.e., multiple attention maps) but rather if particular regions of parietal cortex were potentially more selective for changes in performance. We found no significant differences with respect to the results reported here, so for the present purposes we treated these ROIs as uniform, acknowledging that they undoubtedly contain multiple subdivisions that can be defined in other ways (Lewis and Van Essen 2000; Sereno et al. 2001; Sereno and Huang 2006; Swisher et al. 2007).
Data from this experiment agree with previous reports of multiple maps of attention in parietal cortex. As indicated by the filled arrowheads in Figure 2A, the zone of activation we observed coincides with the center of mass reported for activation of right parietal cortex by Sereno et al. (2001). In some subjects, the areas of activation within SPL overlapped with the area identified as IPS2 by Silver et al. (2005) (open stars in Fig. 2A). However, the center-of-mass for our activation was always anterior to the published location of IPS1 (filled stars in Fig. 2A), though IPS1 could be encompassed by the posterior extent of our ROI. The medial aspects of our ROI's also correspond nicely with the more recently identified topographic regions (Swisher et al. 2007).
Signal Characteristics
Figure 3 shows typical characteristics of the BOLD signal averaged over several voxels within the right parietal cortex of 1 subject. As the subject shifted attention from target to target (gray scale code below x axis and in inset in Fig. 3), the signal amplitude increased and then returned to baseline during interleaved blocks of fixation trials ("F" in Fig. 3). Note the presence of a general task-related signal for all target locations and an additional, spatially specific, increase for the contralateral locations. Also, there was a significant difference between top and bottom locations in contralateral space, indicating specificity for locations within a hemifield as well (P < 0.0005, dependent design permutation test [Good 2000]). Thus, these voxels had 2 distinct components to the BOLD response: 1) a spatially nonspecific, task-related component and 2) a spatially specific component.
|
Further details of the attended target-location signals are presented in Figure 4. Based on the amplitude of the BOLD signal, approximately 70–80% of the attentionally responsive voxels in parietal cortex showed increased activity when attending to contralateral locations (Fig. 4A,B) with the remaining 20–30% showing increased activity when attending to ipsilateral locations (left parietal cortex F3,28 = 9.60, P < 0.0005; right parietal cortex F3,28 = 17.40, P < 0.000005). We did not observe any obvious and consistent hemispheric asymmetry in the representation of target locations during the attention task. Previous investigators have proposed that the well-known tendency for right- but not left-sided lesions to cause a left-sided neglect may reflect an asymmetry in the hemispheric representation of attentional space (Heilman and Van Den Abell 1980
; Heilman et al. 1986; Ringman et al. 2004). We found no evidence of this, at least within the medial superior parietal cortex and anterior IPL activated in this study.
|
Average SQ was calculated for each set of voxels identified as "preferring" a specific target location (based on amplitude of the BOLD signal) and are plotted in Figure 4C,D. The partial F-statistics (SQ) from those voxels preferring contralateral locations were, on average, more than 50% greater than from voxels preferring ipsilateral locations. For example, the average SQ for voxels preferring the top and bottom right target locations in left parietal cortex had significantly higher mean SQ than the SQ from voxels preferring the top and bottom left target locations (left parietal cortex one-tailed t-test, t-stat = –7.60, P < 0.00000001). The same was true for right parietal cortex (right parietal t-stat = 7.9, P < 0.000001). Such differences between voxels preferring contralateral locations and ipsilateral locations were much weaker in the SSI (Fig. 4E,F), although it was significant (ipsilateral 9% lower than contralateral, left parietal cortex F3,28 = 3.40, P < 0.05; ipsilateral 10% lower than contralateral, right parietal cortex F3,28 = 4.40, P < 0.01). Consequently, the changes in spatial specificity among attended target locations could not solely account for changes in SQ. These results suggest that, within each hemisphere, contralateral target locations are represented by a preponderance of high-quality signals (high SQ), although ipsilateral target locations are represented as well.
Effects of Performance and Task Manipulation
Next, we wanted to determine if the parietal attention-related signals were modulated by changes in task or performance, thereby providing evidence for their behavioral relevance. For these analyses, we restricted our computations to voxels identified as preferring contralateral locations. Five subjects repeated the experiment at additional distractor densities (16, 24, or 32 disks; 1 subject—5 data sets, 2 subjects—4 data sets, 2 subjects—3 data sets). Overall, we found the spatial extent of attention-related activation varied considerably as performance ranged from poor to excellent. This is illustrated in Figure 2B where average accuracy (% correct) for all target locations is indicated beneath each brain image. The greatest number of voxels was activated at intermediate performance levels with fewer activated as performance either improved or declined from the peak. Thus, the robustness of task-related modulation changed as performance varied, and this was consistent in both hemispheres for all subjects.
To quantify these behavioral effects, the peak SQs for groups of voxels representing each contralateral target location (i.e., location-specific ROIs) were plotted against psychophysical performance separately for voxels in left and right parietal cortex (Fig. 5A,B). (Shading of individual data point symbols identifies the curve to which it belongs. Shape of each symbol represents corresponding stimulus disk density as described below.) Each SQ value represents the median of the top 10% of SQ values observed in those voxels preferring a given target location. That is, they represent the best SQ for each attended location. The absolute values of the SQ differed between subjects, so they were normalized for these graphs (as described in Materials and Methods). As illustrated in Figure 5, SQ was highest for intermediate levels of task performance and lower both when performance was very poor or very good.
|
To confirm the second-order nature of the inverted "U" function in Figure 5, we applied a least squares fit to the data using both linear and second-order polynomial models. Correlation coefficients for each of the models are reported in Table 1 (Values in parentheses are correlation coefficients when 1 outlier was removed.). Clearly, the second-order polynomial model fit the data much better than the linear model in all subjects and in both hemispheres.
|
We did not see the inverted U relationship in the spatial specificity indices (Fig. 5C–F) in parietal cortex. Panels C and D in Figure 5 illustrate the average specificity index for exactly the same voxels as used in the SQ analysis (panels A and B). As illustrated in Table 1, however, the correlation for the second-order model was poorer than for the SQ data. To rule out a potential bias caused by selecting only voxels with the top 10% SQ values, we repeated the analysis but selected voxels having the top 10% SSI (Fig. 5E,F). Regardless of the selection process, the relationship between performance and specificity was poor and did not reflect the inverted U–shaped function observed in the SQ data. These findings suggest that although the quality of the BOLD modulation (SQ) varied significantly with performance, spatial specificity did not.
The inverted "U"–shaped functions plotted in Figure 5 came from 3 subjects who completed the attention task 4–5 times with 3 different disk densities. Data from 2 other subjects, in which there were 3 different task conditions, followed this same trend for 7 of the 8 target location ROIs tested (data not shown). Four additional subjects performed the task with only 2 task conditions. In these cases, we fit a generalized inverted U curve to the data to determine if it would accurately predict the observed SQ values given each subject's actual performance values. Figure 6 illustrates this process. For 3 of the subjects, a single generalized curve (black curve in Fig. 6) was created by fitting a third-order polynomial to the pooled data with 1 outlier point removed (r = 0.828). (A third-order polynomial was used to allow for asymmetry in the slopes of either side of the primary curve.) Figure 6 illustrates an example from 1 subject in which the measured performance values were 73% and 92% correct (dashed arrows). Based on the generalized curve, the predicted slope was negative as was the observed slope between the 2 observed SQ values. Such predictions were made for all 4 target locations in all 4 subjects. The resulting SQs varied across the 2 conditions in a direction that could be correctly predicted from the associated performance values for 75% of the ROI sites tested in 3 of these subjects. For 1 subject, an alternate generalized curve (gray curve in Fig. 6) was fit to his data as performance generally exceeded 80% accuracy. Using this curve, the predictions were again correct for 3 of the 4 target locations. In sum, for all 9 subjects, the peak SQ consistently occurred at intermediate performance levels and rarely occurred at the lowest or highest performance levels.
|
It is important to note that the various performance values observed for each subject arose not only from different disk densities but also from random variations across experimental sessions and across groups of voxels preferring different target locations. Because of this, different performance values could be observed for the same disk density, or conversely, different densities could sometimes yield the same performance values. To emphasize this, individual data points in Figure 5 are represented by 1 of 3 shapes, each corresponding to a different disk density. Close inspection of the distribution of shapes shows that performance tends to improve linearly as disk density decreases, and this is confirmed directly in Figure 7A. (Note the x axis values for Figure 7A are arranged in descending order to roughly correspond to the performance axis of Fig. 5). In contrast to performance, average SQ was largely unaffected by disk density per se or, if anything, fell slightly as density decreased (Fig. 7B) though this latter relationship was not statistically significant. In sum, although performance correlates inversely with stimulus disk density, such stimulus-related effects alone do not account for the inverted U relationship between SQ and performance illustrated in Figure 5. This also suggests that if task difficulty varies directly with disk density, as one might expect, then changes in difficulty also do not account for the results of Figure 5.
|
An alternate interpretation of Figure 5 might be that as task difficulty rises and performance starts to suffer (right half of inverted U function), the subject's arousal increases and SQ improves. However, when the task becomes too difficult (left half of inverted U function), the subject "gives up," so arousal suddenly drops and SQ declines. To assess this possibility, we examined response latency data across performance levels (Supplementary Fig. S1). As task difficulty increases, one might expect the reaction time to systematically increase as subjects take longer to be certain of the target status. If subjects were giving up, one would expect a sudden change in latency at the poorer performance levels, perhaps increasing if subjects become careless or possibly decreasing if subjects simply press any button immediately when the response tone sounds. Although latency did tend to increase over the whole performance range, there was no evidence of a sudden or disproportionate increase or decrease at the poorest levels, suggesting a sudden change in the subjects’ strategies. For one subject in particular (Supplementary Fig. S1, BCW), little or no change in latency with performance occurred, yet SQ changed 3- to 4-fold over the same performance range (Fig. 5, BCW).
|
Discussion |
|---|
|
TopAbstractIntroductionMaterials and MethodsResultsDiscussionSupplementary MaterialFundingReferences |
|---|
In this study, most voxels in parietal cortex exhibited a general task-related activation reflecting differences between passively fixating a central marker and actively attending to a peripheral cued target and tracking its status (present/absent). In addition, many though not all, parietal voxels respond selectively to different attended target locations with the majority preferring contralateral targets but some preferring ipsilateral targets. We observed no asymmetry between right and left hemispheres in this respect. Voxels activated in the foregoing manner occupied cortex within, and medial to, the IPS extending anteriorly as far as the postcentral sulcus. In most subjects (7/9), this region consisted of at least 2 distinguishable clusters in one or both hemispheres. These zones likely include the topographically organized zone identified by Sereno et al. (2001
), the zone defined by Silver et al. (2005) as IPS2 and possibly IPS1, and at least portions of IPS3 and IPS4 as defined by Swisher et al. (2007). Most importantly, though, this study provides strong human-derived evidence for a quantitative relationship between neural signals in parietal cortex and performance accuracy in an endogenously driven visuospatial attention task. The number of activated voxels and the quality of the fMRI signal were consistently modulated according to task performance but in a nonmonotonic fashion that peaked at intermediate levels of performance accuracy. These SQ variations reflect differences in performance caused in part by task manipulation (disk density) and by individual differences in performance associated with different target locations and different experimental sessions.
A link between parietal cortex and visuospatial attention has been documented in a number of earlier studies in both humans (Posner et al. 1982; Corbetta et al. 1990; 1998; Muller et al. 1998) and animals (Goldberg and Segraves 1987; Gottlieb et al. 1998; Colby and Goldberg 1999; Cook and Maunsell 2002; Bisley and Goldberg 2006). Additionally, previous research has provided preliminary evidence for a relationship between spatial attention and identification of correct versus incorrect trials in humans (Sapir et al. 2005). However, evidence for a quantitative relationship between "spatially specific" parietal signal characteristics and performance has been lacking in humans. The success of the present study in uncovering such a quantitative relationship may derive partly from its focus on the analysis of spatially specific subsets of voxels most likely to represent each attended location. The study also benefits from its focus on SQ rather than spatial specificity alone, which was only weakly related to performance. In this respect, it is important to stress that only those signals representing the upper 10% of the observed SQ range were compared with performance. The advantage of this whole approach is that it tends to highlight those neural signals most likely to specifically represent each attended location and to represent the best quality signals associated with each of those locations. Presumably, it is these spatially specific, high-quality signals that set the limits of behavioral performance, and so are most likely to exhibit a consistent relationship with behavior. The surprise was that the relationship between SQ and performance was nonmonotonic in that the SQ values peaked at intermediate performance levels and fell when performance was too good or too bad (yielding an inverted U function). To our knowledge, this is a unique result, and it provides important clues as to the properties of parietal signals that are linked to behavior.
The interpretation of the SQ/performance curve is complex. To perform well on the task used in this study, the observer must employ covert attention to isolate and track the cued target thereby reducing interference from distracters (noise) and permitting an accurate assessment of the presence or absence of the target. If accurate assessment of the target status is directly related to the quality (partial F-statistic) of a target-related signal, then the F-statistic should fall monotonically as accuracy declines, but this was not observed. On the other hand, increasing disk density increases the need for precise attentional isolation of the target. A signal representing the attentional control or its modulatory effect would then need to improve correspondingly. However, our results show that, on average, SQ is largely unaffected by disk density per se though performance tends to fall as disk density increases. Why then would SQ drop as performance declines? Plausibly, this may reflect the effect of a second factor, observer ability. As task demand increases, the observer tries to increase the precision of attentional isolation, so SQ of the spatially encoded control signal improves. As disk density surpasses the ability of the observer to isolate and track the target, the attentional control/effect becomes erratic (noisy) and SQ falls. In this scenario, then, SQ peaks when the task is neither too easy nor too hard but when it is just right, thereby matching the observer's ability to the attentional precision demanded by the task. Because the observer's ability can vary from day to day and for different locations within the target array (an observation reported by many subjects), a range of performance values naturally ensues so that performance is related not just to disk density but to the interplay of both task- and observer-related factors.
The inverted U function relating SQ and performance is reminiscent of the well-known Yerkes-Dodson curve relating performance and arousal (Yerkes and Dodson 1908). Increasing arousal initially assists performance but then impedes performance as the observer becomes overly aroused. In such case, performance is poor for both high and low arousal levels. So, if SQ in the present study simply reflects arousal, we would expect to find both the highest and lowest values of SQ associated with the poorest performance, but this was not the case. Arousal also affects reaction time (Freeman 1933) with reactions times being the slowest at the highest and lowest levels of arousal. However, we found reaction times increased monotonically as performance declined, consistent with a progressive increase in difficulty without achieving a level of too little or too much arousal. Thus, neither reaction time nor difficulty appears capable of accounting for the nonlinear relationship between SQ and performance that we observed. Altogether, it seems clear that none of the most obvious single-factor interpretations is likely to account for the relationship we have observed between parietal physiology and performance.
We found the fact that the spatial specificity measure of the fMRI signal varied only modestly with changes in performance, while more than 3-fold changes occurred in the partial F-statistic over the same range of performance quite interesting. Intuitively, we often think of SQ in the context of a signal representing target information in a suboptimal environment, so the SQ value could represent degraded target information with an associated impairment in judging the status of the target. As mentioned above, this would lead to the expectation that SQ should be monotonically related to performance, and this was not the case. However, if the fMRI signal represents an attentional control signal or represents a signal used to modulate visual processing (e.g., gain control signal), then the quality of that signal would be related to the precision of control or the precision of the modulatory effects. To the extent the parietal voxels exhibiting these signals encode or represent the location (and/or extent) of the "window" of attention, SQ could then represent variation in the location (and/or extent) of the attentional window. Interestingly, spatial specificity remains good as long as there is at least a minimum SQ present. Consistent spatial specificity across performance may be a necessary characteristic for a dynamically modulated representation of attended target location. By demonstrating a link between parietal physiology and behavior, this study provides motivation and a new perspective for a more comprehensive analysis of the role of parietal cortex in attention related behavior.
|
Supplementary Material |
|---|
|
TopAbstractIntroductionMaterials and MethodsResultsDiscussionSupplementary MaterialFundingReferences |
|---|
Supplementary material can be found at: http://www.cercor.oxfordjournals.org/.
|
Funding |
|---|
|
TopAbstractIntroductionMaterials and MethodsResultsDiscussionSupplementary MaterialFundingReferences |
|---|
National Institutes of Health grants EB00843, EY13801, and MH019992 to E.A.D. and RR00058 to the Medical College of Wisconsin.
|
Acknowledgments |
|---|
We thank Doug Ward for his assistance with statistics. Conflict of Interest: None declared.
|
References |
|---|
|
TopAbstractIntroductionMaterials and MethodsResultsDiscussionSupplementary MaterialFundingReferences |
|---|
Ben Hamed S, Duhamel JR, Bremmer F, Graf W. Representation of the visual field in the lateral intraparietal area of macaque monkeys: a quantitative receptive field analysis. Exp Brain Res (2001) 140:127–144.[CrossRef][ISI][Medline]
Bisley JW, Goldberg ME. Neural correlates of attention and distractibility in the lateral intraparietal area. J Neurophysiol (2006) 95:1696–1717.
Boynton GM. Adaptation and attentional selection. Nat Neurosci (2004) 7:8–10.[CrossRef][ISI][Medline]
Brefczynski JA, DeYoe EA. A physiological correlate of the ‘spotlight’ of visual attention. Nat Neurosci (1999) 2:370–374.[CrossRef][ISI][Medline]
Colby CL, Goldberg ME. Space and attention in parietal cortex. Annu Rev Neurosci (1999) 22:319–349.[CrossRef][ISI][Medline]
Cook EP, Maunsell JH. Attentional modulation of behavioral performance and neuronal responses in middle temporal and ventral intraparietal areas of macaque monkey. J Neurosci (2002) 22:1994–2004.
Corbetta M, Akbudak E, Conturo TE, Snyder AZ, Ollinger JM, Drury HA, Linenweber MR, Petersen SE, Raichle ME, Van Essen DC, et al. A common network of functional areas for attention and eye movements. Neuron (1998) 21:761–773.[CrossRef][ISI][Medline]
Corbetta M, Miezin FM, Dobmeyer S, Shulman GL, Petersen SE. Attentional modulation of neural processing of shape, color, and velocity in humans. Science (1990) 248:1556–1559.
Corbetta M, Shulman GL, Miezin FM, Petersen SE. Superior parietal cortex activation during spatial attention shifts and visual feature conjunction. Science (1995) 270:802–805.
Cox RW. AFNI: software for analysis and visualization of functional magnetic resonance neuroimages. Comput Biomed Res (1996) 29:162–173.[CrossRef][ISI][Medline]
Crowe DA, Chafee MV, Averbeck BB, Georgopoulos AP. Neural activity in primate parietal area 7a related to spatial analysis of visual mazes. Cereb Cortex (2004) 14:23–34.
Freeman G. The facilitative and inhibitory effects of muscular tension upon performance. Am J Psychol (1933) 26:602–608.
Goldberg ME, Segraves MA. Visuospatial and motor attention in the monkey. Neuropsychologia (1987) 25:107–118.[CrossRef][ISI][Medline]
Good P. Permutation tests: a practical guide to resampling methods for testing hypotheses (2000) New York: Springer.
Gottlieb JP, Kusunoki M, Goldberg ME. The representation of visual salience in monkey parietal cortex. Nature (1998) 391:481–484.[CrossRef][Medline]
Heider B, Jando G, Siegel RM. Functional architecture of retinotopy in visual association cortex of behaving monkey. Cereb Cortex (2005) 15:460–478.
Heilman KM, Bowers D, Valenstein E, Watson RT. The right hemisphere: neuropsychological functions. J Neurosurg (1986) 64:693–704.[ISI][Medline]
Heilman KM, Van Den Abell T. Right hemisphere dominance for attention: the mechanism underlying hemispheric asymmetries of inattention (neglect). Neurology (1980) 30:327–330.
Hopfinger JB, Woldorff MG, Fletcher EM, Mangun GR. Dissociating top-down attentional control from selective perception and action. Neuropsychologia (2001) 39:1277–1291.[CrossRef][ISI][Medline]
Huk AC, Heeger DJ. Task-related modulation of visual cortex. J Neurophysiol (2000) 83:3525–3536.
Kastner S, Pinsk MA, De Weerd P, Desimone R, Ungerleider LG. Increased activity in human visual cortex during directed attention in the absence of visual stimulation. Neuron (1999) 22:751–761.[CrossRef][ISI][Medline]
Kimberg DY, Aguirre GK, D'Esposito M. Modulation of task-related neural activity in task-switching: an fMRI study. Brain Res Cogn Brain Res (2000) 10:189–196.[CrossRef][Medline]
Lewis JW, Van Essen DC. Mapping of architectonic subdivisions in the macaque monkey, with emphasis on parieto-occipital cortex. J Comp Neurol (2000) 428:79–111.[CrossRef][ISI][Medline]
Muller MM, Picton TW, Valdes-Sosa P, Riera J, Teder-Salejarvi WA, Hillyard SA. Effects of spatial selective attention on the steady-state visual evoked potential in the 20-28 Hz range. Brain Res Cogn Brain Res (1998) 6:249–261.[CrossRef][Medline]
Posner MI, Cohen Y, Rafal RD. Neural systems control of spatial orienting. Philos Trans R Soc Lond B Biol Sci (1982) 298:187–198.[ISI][Medline]
Raffi M, Siegel RM. Functional architecture of spatial attention in the parietal cortex of the behaving monkey. J Neurosci (2005) 25:5171–5186.
Ringman JM, Saver JL, Woolson RF, Clarke WR, Adams HP. Frequency, risk factors, anatomy, and course of unilateral neglect in an acute stroke cohort. Neurology (2004) 63:468–474.
Rushworth MF, Passingham RE, Nobre AC. Components of switching intentional set. J Cogn Neurosci (2002) 14:1139–1150.
Rushworth MF, Paus T, Sipila PK. Attention systems and the organization of the human parietal cortex. J Neurosci (2001) 21:5262–5271.
Sapir A, d'Avossa G, McAvoy M, Shulman GL, Corbetta M. Brain signals for spatial attention predict performance in a motion discrimination task. Proc Natl Acad Sci USA (2005) 102:17810–17815.
Sereno MI, Huang RS. A human parietal face area contains aligned head-centered visual and tactile maps. Nat Neurosci (2006) 9:1337–1343.[CrossRef][ISI][Medline]
Sereno MI, Pitzalis S, Martinez A. Mapping of contralateral space in retinotopic coordinates by a parietal cortical area in humans. Science (2001) 294:1350–1354.
Shulman GL, d'Avossa G, Tansy AP, Corbetta M. Two attentional processes in the parietal lobe. Cereb Cortex (2002) 12:1124–1131.
Silver MA, Ress D, Heeger DJ. Topographic maps of visual spatial attention in human parietal cortex. J Neurophysiol (2005) 94:1358–1371.
Sohn MH, Ursu S, Anderson JR, Stenger VA, Carter CS. Inaugural article: the role of prefrontal cortex and posterior parietal cortex in task switching. Proc Natl Acad Sci USA (2000) 97:13448–13453.
Somers DC, Dale AM, Seiffert AE, Tootell RB. Functional MRI reveals spatially specific attentional modulation in human primary visual cortex. Proc Natl Acad Sci USA (1999) 96:1663–1668.
Swisher JD, Halko MA, Merabet LB, McMains SA, Somers DC. Visual topography of human intraparietal sulcus. J Neurosci (2007) 27:5326–5337.
Tootell RB, Hadjikhani N, Hall EK, Marrett S, Vanduffel W, Vaughan JT, Dale AM. The retinotopy of visual spatial attention. Neuron (1998) 21:1409–1422.[CrossRef][ISI][Medline]
Vandenberghe R, Gitelman DR, Parrish TB, Mesulam MM. Location- or feature-based targeting of peripheral attention. Neuroimage (2001) 14:37–47.[CrossRef][ISI][Medline]
Wardak C, Olivier E, Duhamel JR. A deficit in covert attention after parietal cortex inactivation in the monkey. Neuron (2004) 42:501–508.[CrossRef][ISI][Medline]
Yantis S, Schwarzbach J, Serences JT, Carlson RL, Steinmetz MA, Pekar JJ, Courtney SM. Transient neural activity in human parietal cortex during spatial attention shifts. Nat Neurosci (2002) 5:995–1002.[CrossRef][ISI][Medline]
Yerkes Y, Dodson J. The relation of strength of stimulus to rapidity of habit formation. J Comp Neurol Psychol (1908) 18:459–482.[CrossRef]
CiteULike 
Connotea 
Del.icio.us What's this?
| ||||||||||||||||||||||||||||||||