Le, Tuong Huu, José V. Pardo, and Xiaoping Hu. 4 T-fMRI study of nonspatial shifting of selective attention: cerebellar and parietal contributions. J. Neurophysiol. 79: 1535–1548, 1998. Regional blood oxygenation in the cerebellum and posterior cerebral cortices was monitored with functional magnetic resonance imaging (fMRI) at four Tesla while 16 normal subjects performed three tasks with identical visual stimulation: fixation; attention focused upon either stimulus shape or color and sustained during blocks of trials (sustained attention); and rapid, serial shifts in attention between stimulus shape or color within blocks of trials (shifting attention). The stimuli were displayed centrally for 100 ms followed by a central fixation mark for 900 ms. Each stimulus was either a circle or a square displayed in either red or green. Attention shifting required switching between color and shape information after each target detection and occurred on average once every three seconds. Subjects pressed a response key upon detecting the target; reaction time and response accuracy were recorded. Two protocols for T2*-weighted echo-planar imaging were optimized, one with a surface coil for the cerebellum alone and the other with a volume coil for imaging both cerebellum and posterior brain structures (parietal, occipital, and part of temporal cortices). Because fMRI of the cerebellum is particularly susceptible to cardiac and respiratory fluctuations, novel techniques were applied to isolate brain activation signals from physiological noise. Functional activation maps were generated for contrasts of 1) sustained attention to color minus fixation; 2) sustained attention to shape minus fixation; and 3) shifting attention minus sustained attention (to color and shape; i.e., summed across blocks of trials). Consistent with the ease of these tasks, subjects performed with >80% accuracy during both sustained attention and shifting attention. Analysis of variance did not show significant differences in false alarms or true hits across either attentional condition. A subgroup of subjects whose performance data were recorded during ten minutes of continuous practice did not show significant changes over time. Both contrasts between the conditions of sustained attention to color or to shape as compared with the fixation condition showed significant bilateral activation in occipital and inferior temporal regions (Brodmann areas 18, 19, and 37). The anterior medial cerebellum was also significantly activated ipsilateral to the finger used for responding. The principal comparison of interest, the contrast between the condition of shifting attention and the condition of sustained attention produced significant and reproducible activation: lateral cerebellar hemisphere (ansiform lobule: Crus I Anterior and Crus I Posterior; left Crus I Posterior); cerebellar folium; posterior superior parietal lobule (R and L); and cuneus and precuneus (R and L).
Advances in the neurobiology of the human attention system implicate a complex and anatomically distributed network of cortical and subcortical regions with isolable cognitive operations (Posner and Dehaene 1994). A region of the anterior cingulate becomes active during tasks that, regardless of the specific nature of the stimulus, require attention for the selection process during verb generation (Posner et al. 1988); during competing processing alternatives (Pardo et al. 1990); and during the simultaneous processing of multiple attributes of stimuli (Corbetta et al. 1991). Stereotactic lesions of this region can cause deficits in selective attention in humans (Janer and Pardo 1991). On the basis of studies of patients with lesions (Freidman-Hill et al. 1995; Petersen et al. 1989; Posner et al. 1984; Robertson et al. 1997) and of neuroimaging of normal volunteers (Corbetta et al. 1993, 1995; Iacobini et al. 1996; Vandenberghe et al. 1996), the superior parietal cortices participate in the covert orienting of visuospatial attention, in feature conjunction during visual search, in spatial coding for object perception, in stimulus discrimination at varying spatial locations, and in spatial compatibility between stimuli and responses: these studies converge upon a fundamental role of the parietal cortex in attentional processes involving spatial codes (Andersen et al. 1985). A heteromodal network, encompassing regions of prefrontal, parietal, and temporal cortices of the right hemisphere, activates during tasks requiring sustained attention to visual or somatosensory information (Pardo et al. 1991). Part of the activity within this network supports spatial working memory (Jonides et al. 1993; McCarthy et al. 1996) and has striking parallels to working memory systems in nonhuman primates (Goldman-Rakic 1987). Maintenance of the alert state has a right-hemisphere dominance (Whitehead 1991) and may map to the right ventrolateral prefrontal cortex (Paus et al. 1997). Several subcortical regions also contribute to attentional processing. The pulvinar nucleus of the thalamus may participate in the modulation of activity in visual cortices to enable filtering of irrelevant distractors (LaBerge and Buchsbaum 1990). The midbrain reticular formation and intralaminar nuclei of the thalamus may engage states of high vigilance and arousal (Kinomura et al. 1996; Paus et al. 1997).
Recent data suggest that the cerebellum also plays a role in attention. Various studies (for recent review, see Schmahmann 1996) now suggest that the cerebellum has functions beyond those of motor processing. However, many of the tasks used to study such functions do not allow ready dissection of more elementary cognitive operations. Akshoomoff and Courchesne (1992, 1994) first hypothesized that the cerebellum may mediate rapid shifts in attention both between and within sensory modalities . They reported performance and event-related potential (ERP) data from patients with neocerebellar damage showing a selective impairment in detecting a target immediately after a shift in the attended attribute.
Studies of patients with lesions have inherent limitations. First, the disease process affects often multiple brain regions. Second, lack of control over the extent and precise placement of the lesion precludes definitive anatomic localization. Third, the plastic reorganization that occurs after brain injury can confound interpretations concerning the key structures relevant to the successful execution of a task. Fourth, the specific neural circuits important for task execution, which interact with the lesion location, elude definition. Therefore, the neuroimaging of normal volunteers while performing tasks analogous to those of Akshoomoff and Courchesne could provide an important test of the hypothesis of cerebellar involvement in shifts of attention.
Magnetic resonance imaging at 4 Tesla of signals arising from tissue oxygenation (blood-oxygen-level-dependent, BOLD), which reflects neuronal activity, is a noninvasive technique that provides both high signal-to-noise ratio and high spatial resolution (Ogawa et al. 1990; Thulborn et al. 1982). Unfortunately, functional imaging of the cerebellum presents special challenges: high sensitivity to artifacts arising from quasiperiodic cardiac and pulmonary fluctuations and from the magnetic susceptibility of biological tissue. These difficulties were overcome recently through the development of several new techniques (Hu et al. 1995; Hu and Le 1996). The research presented here exploits this progress to visualize the components of the cerebellum and related structures that mediate switching of visual attention within a single, central location in space. A preliminary report of this work was presented in abstract form (Le and Hu 1996). A recent paper using a similar strategy but with different methods and analyses provides converging data with some of the cerebellar results reported here (Allen et al. 1997).
Sixteen healthy volunteers, aged 20–40 yr, with normal or corrected-to-normal visual acuity were studied after providing written informed consent according to the policies of the Institutional Review Board of the University of Minnesota. Six right-handed (2 male, 4 female), one left-handed (male), and one ambidextrous (female) subjects participated in study 1, which used a surface coil. Six right-handed (2 male, 4 female) and two left-handed (1 male) subjects participated in study 2, which employed a volume coil.
Visual stimuli were displayed on a screen with a video projector connected to a personal computer located outside the magnetic shielding. A mirror mounted above the eyes of the supine subject permitted viewing of the screen within the confines of the MRI scanner. The stimuli consisted of a green circle, a red circle, a green square, and a red square. Each stimulus was displayed foveally in the center of the screen for a duration of 100 ms, followed by a central fixation cross hair for a duration of 900 ms (Fig. 1, left). The stimuli were presented randomly with the following probabilities: red circle, 1/9; green circle, 2/9; red square, 2/9; and green square, 4/9. The visual angle subtended by the stimuli and cross hair was 1.4° of the central visual field. The brief stimulus duration and overlapping size and location of the stimuli and cross hair were selected to assist central fixation and to preclude eye movements. The long interstimulus interval minimized masking effects.
Each subject performed four types of tasks: visual fixation, sustained attention to color, sustained attention to shape, and shifting attention (adapted from Akshoomoff and Courchesne 1994). The order of each task was randomized for each imaging session. Subjects were instructed to maintain central fixation during all tasks. Instructions specific to a task were displayed for one second at the start of the block of trials. The task conditions were randomized. A target appeared on average once every three seconds because the interstimulus interval was one second and the probability of a target was 1/3 (e.g., for sustained attention to the color red: 1/9 red circle + 2/9 red square = 1/3). Therefore subjects with perfect accuracy pressed the response key with an average frequency of 1/3 Hz.
An imaging session (lasting ∼1 h) typically consisted of three runs (one such run is shown in Fig. 1, right, a). Each run lasted about 11 min and contained 10 trial blocks (e.g., Si or shifting attention; C or sustained attention to color), each lasting 70 s.
Subjects were instructed to maintain fixation upon the cross hair and to ignore the shapes and colors. The response key was not used. This task controlled for visual processing.
SUSTAINED ATTENTION TASKS.
Sustained attention to color: subjects were instructed to maintain central fixation and to respond by pressing the response key with the index finger of their dominant hand whenever the target color, red, appeared. Subjects were to ignore any shape information.
Sustained attention to shape: subjects were instructed to maintain central fixation and to respond by pressing the response key with the index finger of the dominant hand whenever the target shape, a circle, appeared. Subjects were to ignore any color information.
Both of these tasks controlled for sensory processing, motor responses, target detection, and instructional set without any requirement for shifting attention during the block of trials.
SHIFTING ATTENTION TASK.
Subjects were instructed to maintain central fixation and to respond to targets by pressing the response key with the index finger of their dominant hand. Choice of target depended upon alternately, after each response, shifting attention between shape and color information. Only the red color and circular shape served as targets for a given block of trials. For example, subjects responded to the first red target independent of shape, then switched to detecting a circle regardless of color. Upon detecting a circle, the subject switched to detecting any red stimulus and so forth. Therefore the subject would alternate between detecting red stimuli and circular stimuli. If the subject performed with perfect accuracy, the shift in attention to another attribute occurred on average once every three seconds (i.e., the average target frequency, see above).
This task contains not only the components of the sustained attention tasks described above, but an additional component requiring the shifting of attention between feature dimensions. This attention shifting task does not involve visuospatial shifts in attention as all stimuli were foveal and were restricted to the central 1.4° of the visual field and individual stimuli were displayed for less time than that necessary for eye movements (Hallett and Lightstone 1976). No visual discrimination occurs in any spatial location except at the locus of fixation. Attention shifting across features remains postured upon a single stimulus. Only one feature is attended at any time, so attention to conjunction of features does not occur. The restriction of targets to just one color and one shape during any block makes the shifting task comparable in difficulty to the sustained attention task (see Behavioral results in results).
BEHAVIORAL DEPENDENT MEASURES.
Each subject spent 10–20 min practicing the tasks before entering the MRI scanner. The performance of each subject was recorded with a personal computer. The reaction time (RT) was defined as the time between the onset of the stimulus and the pressing of the response key. Three other performance parameters were calculated for each subject Equation 1 Equation 2 Equation 3THF provided a measure of sensitivity and the FAF provided a measure of specificity. (Metz 1978). The IDA yielded an overall measure of performance by combining both sensitivity and specificity factors.
One-way analyses of variance (ANOVA) were performed on THF, FAF, IDA, and RT as dependent variables and the attentional conditions (sustained and shifting) as independent variables. Only RTs for correct responses were used.
Magnetic resonance imaging
All fMRI experiments were conducted on a four Tesla whole body system with a 1.25-m-diam bore (SISCo/Siemens; Sunnyvale, CA). The magnet was equipped with a head gradient insert that produces a maximum gradient of 3 G/cm with a rise time of 180 μs in all three axes. To minimize gross motion, a customized foam insert was used to constrain the head (Alpha Cradle Foaming Agent, Smithers Medical Products, Akron, OH). Inversion-recovery–prepared ultrafast gradient echo T1-weighted images (Haase et al. 1986) were acquired for the localization of the region of interest (ROI) and for subsequent stereotactic transformation (see Stereostatic transformation). Static magnetic field (B o) homogeneity was optimized after selection of the region of interest by manual shimming to a linewidth of <15 Hz. The process of localizing the ROI and shimming took 15 min. Subsequently, fMRI experiments utilized consecutive acquisition of the apparent transverse relaxation time T2*-weighted echo-planar images (EPI) (Mansfield 1977) during trial blocks of task performance.
STUDY 1: FOUR-SEGMENT EPI WITH SURFACE COIL.
A circularly polarized surface coil (Ackerman et al. 1980) placed behind the occiput provided optimal signal-to-noise ratio (SNR) for the cerebellum. Eight to ten contiguous coronal slices encompassing the cerebellum were obtained with the following imaging parameters: slice thickness = 5 mm, repetition time (TR)/echo time(TE) = 3.5 s / 30 ms, and field of view (FOV) = 20 × 20 cm. A four-segment EPI sequence (Kim et al. 1996) with a matrix size of 128 × 128 provided an effective voxel resolution of 1.6 ×1.6 × 5 mm. This sequence employed a variable flip angle approach (Frahm et al. 1985) with an intersegment TR of 62 ms. Between 180 and 200 T2*-weighted images per slice were acquired during a typical block of trials.
STUDY 2: SINGLE-SHOT EPI WITH VOLUME COIL.
A quadrature birdcage volume coil (Hayes et al. 1985) provided uniform SNR in both the posterior cerebrum and the cerebellum. Single-shot T2*-weighted EPI (Mansfield 1977) yielded an effective voxel resolution of 3.1 × 3.1 × 5 mm (matrix size: 64 × 64). Fourteen to sixteen contiguous coronal slices starting at the level of the posterior commissure and containing the cerebellum, occiput, parietal lobe, and part of the temporal lobe were acquired as follows: slice thickness = 5 mm, TR/TE = 3.5 s/30 ms, α = 20, FOV = 20 × 20 cm. Each experimental run consisted again of 180–200 T2*-weighted images per slice.
The only preprocessing of the fMRI images involved the removal of artifactual signals because of respiration and cardiovascular pulsations, which did not affect image resolution. The cardiac signal was monitored with a pulse oximeter (NONIN, Plymouth, MN) attached to the finger. Respiration was gauged with a tension transducer attached to a flexible belt surrounding the upper abdomen. Both measures were sampled at 100 Hz. The details of the implementation and their validation have been described previously (Hu et al. 1995). Nyquist ghosts in EPI images were minimized by using phase encoded-reference calibration (Hu and Le 1996).
STATISTICAL SIGNIFICANCE THRESHOLD.
Student's t-tests were applied voxelwise by using the variance for each voxel. The significance thresholds for the difference images were adjusted toP < 0.05 after Bonferroni correction for the total number of voxels studied. These correspond to uncorrected P values of 1.6(10)−5 and 3.4(10)−7, respectively, for study 1 and 2. Additionally and without regard to distributional properties or variance homogeneity, bootstrap analyses using the scans from same behavioral condition for the difference image (i.e., as both minuend and subtrahend) indicate an average of three false clusters per thresholded difference image. For comparison, typical difference images between different behavioral conditions show, using these thresholds, from 20 to 50 activated clusters. As a further safeguard against Type I errors and to ensure that the results are generalizable, only responses occurring in the majority of subjects (i.e., ≥4 of 6 subjects) are listed in Tables 3-5 (showing from 7 to 16 foci).
The contrast between the condition of sustained attention to color (or shape) and the fixation condition was calculated as those pixels passing the significance threshold in the t-score difference image in “bold” signal between adjacent trial blocks C (sustained attention to color) and V (visual fixation), i.e., C − V, Fig. 1, right, b [or S (sustained attention to shape) and V, i.e., S − V, Fig. 1, right, b]. The contrast between the conditions of shifting attention (Si) and of sustained attention was determined by those pixels showing significant activation common to both the contrast between the adjacent trial blocks involving the conditions of shifting attention (Si) and of sustained attention to color (C) and the contrast between the adjacent trial blocks involving the conditions of Si and of sustained attention to shape (S). In other words, the comparison between the conditions of Si and sustained attention involved the logical “and”' operation, i.e.,(Si − C) “and” (Si − S)—see Fig. 1, right, a.
The reproducibility of significant activation across subjects in study 1 and study 2 was calculated after normalization to the “equivalent cerebellum” (see next subsection) and to the Talairach atlas (see next subsection), respectively. The pixelwise frequency of subjects with t-scores exceeding the significance threshold were displayed (e.g., see Fig. 3 c).
Stereotactic transformation of the fMRI data from study 2 was performed by using the atlas of Talairach and Tournoux (1988) with our implementation of the linear methods described previously (Fox et al. 1988). The routines are available publicly (Strupp 1996).
A stereotactic atlas for the cerebellum remains under development (Schmahmann et al. 1996). In study 1, which focused upon the cerebellum, the full extent of the brain and the anterior commissures was not visualized so that routine linear normalization to the Talairach atlas (as performed for study 2, above) was not possible. Therefore, a linear proportional-grid, as in the Talairach and Tournoux atlas (1988) was used to anatomically normalize the cerebellum alone across subjects. Briefly, the “equivalent cerebellum” was generated by a bounding box including the full extent of the cerebellum based upon projections to three orthogonal planes: 1) parasagittal plane through the midline of the cerebellum; 2) axial plane passing through the superior border of the cerebellum; and 3) coronal plane passing through the anterior border of the cerebellum. The reference volume (see Fig. 2) extended laterally 60 mm to either side of the cerebellum's midline; axially 60 mm (superior to inferior); and coronally 45 mm (anterior to posterior). The cerebellum from each subject was bounded by a similar box and subsequently warped linearly to match the “equivalent cerebellum.” The coronal sections displayed in the results are the average after linear interpolation of all the normalized cerebella. The anatomic designations within the cerebellum were generated from the Talairach and Tournoux atlas (1988), the atlas of Angevine et al. (1961), and an analysis of the individual structural images with the atlas of Schmahmann et al. (1996).
Performance data were collected for only 11 of the 16 subjects because of technical difficulties. Table 1 displays the performance parameters of these 11 subjects. Accuracy was greater than 80% (i.e., IDA > 0.8) for all subjects. There were no significant differences in accuracy (F[1,10] = 0.089, P = 0.77), true hits (F[1,10] = 0.74, P = 0.40 ), false alarms (F[1,10] = 0.40, P = 0.54), or reaction time (F[1,10] = 0.047, P = 0.83) between the conditions of sustained vs. shifting attention. There was a trend for greater intrasubject variance in reaction time during the shifting as compared to the sustained attention condition (F[1,10] = 2.91, P = 0.10). The performance during practice of four subjects was recorded and analyzed; it did not change over 10 min of continuous practice (data not shown).
STUDY 1: CEREBELLAR IMAGING WITH FOUR-SEGMENT EPI AND SURFACE COIL.
This study focused specifically upon the cerebellum. Signal-to-noise was maximized with a surface coil over the occiput. Regions outside the cerebellum, at a distance from the surface coil, could not be detected sensitively. Eight subjects were studied, but the data from one subject, a left-handed male, were discarded because of gross motion artifacts (leaving 6 right-handed and 1 ambidextrous subject, all using their right hand for responding).
Figure 3 a displays the cerebellar activation during attention switching as compared to sustained attention from a representative, right-handed female. The robustness of the suppression of cardiac and respiratory artifacts is apparent in the image quality. On the right side, Crus I Anterior (CrIA, ↖) and Posterior (CrIP) of the ansiform lobule are activated, whereas CrIP is mostly activated on the left side. Several foci mapped also to the posterior and medial cerebellum at the folium. The time course of the signal change for the right lateral cerebellar region (CrIA) (↖) is displayed in Fig. 3 b. Signal increase coincided temporally with blocks of trials requiring attention shifts. Stereotactic coordinates for the activated regions were not obtained in this study because only posterior structures were visualized with the occipital surface coil. Comparison across subjects relied upon linear interpolation to the “equivalent cerebellum.”
Table 2 displays the regions, volumes, and frequencies of significant cerebellar activation from the contrast between the condition of shifting attention vs. sustained attention when comparing across subjects after normalization to the equivalent cerebellum. Right lateral cerebellar activation (ansiform lobule Crus I) was predominant and occurred in all subjects (6 right-handed; 1 ambidextrous by using the right hand). Several other cerebellar regions were activated although less consistently and with smaller volumes.
Reproducibility maps (frequency of significant activation across subjects per pixel) were generated after stereotactic normalization of the cerebellum across subjects for the contrast between the conditions of shifting attention and of sustained attention (Fig. 3 c). The map demonstrates that a region of the right lateral cerebellum (Crus I) was most consistently activated during attention shifting. Activation in the folium (posterior cerebellar region close to midline) was also reproducible (5 of 7 subjects). Posterior parietal regions also showed reproducible activation across subjects, but this region rests outside the optimal field of view for the occipital surface coil and can not be properly normalized stereotactically on the basis of cerebellar proportions.
Notably, regions of cerebellar activation occurring during attention shifting did not overlap with those active during sensorimotor processing. The contrast between the conditions of sustained attention (requiring the pressing of a key with the index finger of the dominant hand) and of visual fixation (without motor output), showed significant activation, as expected (Nitschke et al. 1996), in the ipsilateral anterior lobe of the cerebellum (6 of 7 subjects; data not shown, but see section with volume coil). One right-handed male failed to produce significant cerebellar activation in this region. In the contrast between the conditions of sustained attention and of visual fixation, none of the subjects exhibited significant activation in the posterior and lateral cerebellar regions activated during attention switching.
STUDY 2: IMAGING OF POSTERIOR SYSTEMS WITH SINGLE-SHOT EPI AND VOLUME COIL.
Study 1 identified a nonmotor role for the right lateral cerebellum during shifts of attention between visual features. Additionally, parietal regions were recruited bilaterally during attention shifts. The surface coil used in Study 1 was small, was positioned over the cerebellum, and did not permit visualization of the commissures for normalizing stereotactically. Thus it provided optimal signals only from the cerebellum. Study 2 applied a volume coil during the same task in another group of volunteers with the following rationale: 1) to replicate the finding of right lateral cerebellar and parietal involvement in switching attention; 2) to image the brain for stereotactic transformation to Talairach coordinates; 3) to optimize the visualization of parietal activation during attention shifts by using a larger volume coil. The weaker signals arising from the cerebellum with the whole-head coil were compensated by decreasing the matrix size by one-fourth, resulting in lower spatial resolution. A total of eight subjects participated in this study, but data from two right-handed females were discarded because of motion artifacts (leaving 4 right-handed and 2 left-handed subjects, each subject using their dominant hand for responding).
SHIFTING ATTENTION VERSUS SUSTAINED ATTENTION.
Figure 4 (top left) exhibits a representative activation map for the contrast between shifting attention and sustained attention for one subject, a left-handed male. The activation of the right lateral cerebellum (ansiform lobule CrIA) is highlighted by ➭. A complex network of regions occurs within both posterior superior parietal lobules. The visual cortices are not activated, consistent with an equivalent level of sensory processing between the shifting and sustained attention tasks. The anterior lobe of the cerebellum, likewise, is not activated, indicating the motor aspects were balanced across tasks.
Table 3 displays in order of decreasing volume the regions of significant activation and the reproducibility across subjects for the contrast between the conditions of shifting attention vs. sustained attention. As in study 1, the right lateral cerebellum was consistently activated across all subjects. Two centroids in this region mapped as a doublet with Talairach coordinates (x, y, z: 41 and −61, −26 and 46, −56 and −28). The left lateral cerebellum was activated less consistently, as was the cerebellar folium. Posterior superior parietal lobule, cuneus, and precuneus (BA 7) activated robustly and consistently in both hemispheres.
Figure 4 (bottom right) displays the reproducibility map for the contrast between the conditions of shifting attention and of sustained attention. The location of the loci of activated clusters seen in at least four of the six subjects for this contrast is shown in Table 4. These include the cerebellum (CrIA and CrIP of the ansiform lobule, and folium), parietal cortices (posterior superior parietal lobule, precuneus and cuneus, inferior parietal lobule), and the parietal occipital sulcus (BA 19/7). The activated regions of the cerebral cortex are shown schematically in Figure 5.
SUSTAINED ATTENTION VERSUS VISUAL FIXATION.
Figure 4 (bottom left) displays a representative activation map for the contrast between the conditions of sustained attention to color and of visual fixation for one subject, a left-handed male (same subject as in Figure 4, top left). The activation arising from sensorimotor processing localizes ipsilaterally to the finger used for responding and remains restricted to the anterior medial cerebellum. No activation occurs in posterior and lateral cerebellum. This contrast exhibits little parietal activation as compared with the contrast between the conditions of shifting attention and of sustained attention.
Figure 4 (top right) shows the reproducibility map for the contrast between the conditions of sustained attention to color and of visual fixation; the equivalent map for the condition of sustained attention to shape and of visual fixation are shown in Fig. 4 (middle). The regions reproducible in at least four of six subjects and their locations are summarized in Table 5. The activations occur in extrastriate visual cortices, including lingual, fusiform, and inferior temporal gyri (see Fig. 5). Here, the low reproducibility of activation in the anterior medial cerebellum, as expected from pressing the response key with the dominant hand, results from the mixed handedness of the volunteers (4 right-handed, 2 left-handed). Moreover, reanalysis of the data for only the four right-handed subjects reveals an ipsilateral anteromedial cerebellar response in the contrast between the conditions of sustained attention to color and visual fixation (data not shown). The activation arising from sensorimotor processing localizes ipsilaterally to the finger used for responding and remains restricted to the anterior medial cerebellum (culmen, Larsell lobules IV–V) at stereotatic coordinates (x, y, z: 21, −35, −21).
Behavioral dependent measures
The subjects did not show significant differences in reaction time, true hit rates, false positive hit rates, or accuracy index between the conditions of shifting attention and of sustained attention. The comparable performance across the two conditions reflects several aspects of the tasks used here: 1) the shifting task was made relatively easy by the use of only one color (red) and one shape (circle)—switching could occur only between these two targets and 2) the target rate was set to 0.3 Hz, which is just below the rate of 0.4 Hz when the true hit rate begins to fall (Akshoomoff and Courchesne 1992). Frequencies of attention shifting >0.4 Hz are negatively correlated with the true hit rate; their use could confound the interpretation of the imaging data (Akshoomoff and Courschesne 1992). The time course of nonspatial attention shifting was not studied here, but the dwell time of human visual attention is ∼400 ms (Duncan et al. 1994). Given the relatively large intertarget interval (3 s on average), subjects should have shifted their nonspatial attention well before a target appeared. So, the equivalent performance for shifting versus sustained attention under the conditions used in this study appears reasonable.
Cerebellum in attention shifting
These data identify the posterior lateral cerebellar hemispheres (particularly CrIA and CrIP of the ansiform lobule) and its medial extension into the folium as important during nonmotor shifts of attention. These cerebellar regions were shown to be distinct from those related to sensorimotor processing, which mapped to more medial and anterior areas at the culmen (Larsell lobules IV–V) ipsilateral to the hand used for responding (Nitschke et al. 1996). The results support the hypothesis (Akshoomoff and Courchesne 1994) that the cerebellum is involved in shifting attention. The results show striking convergence with a recent report by Allen et al. (1997).
Allen et al. (1997) used task paradigm similar to that used here, except that they did not use a subtractive approach. They isolated the attention shifting operation by avoiding overt motor responses through a silent counting strategy, which they argued minimized motor contributions. Their analysis strategies were very different than those employed here. Yet, the location of the cerebellar responses converge well across the two studies (here, left cerebellar responses x, y, and z were −41, −67, and −29, respectively; their responses were −37, −63, and −22). Perhaps the major difference in results between the two studies relates to laterality issues. Here, the right cerebellar responses were greater and more reproducible than those on the left side, whereas they found greater activation on the left side. Whether or not this difference reflects sampling bias, task modifications, or technical changes between the studies remains to be elucidated.
We hypothesize that the folium activation in the present study may arise from an interaction with the visual system during attention switching across visual features. Earlier studies, largely in cats, reported evoked responses to both auditory and visual stimuli in this region even in preparations devoid of extraocular eye muscles or cerebral cortex (Snider 1950). However, visual evoked responses have not been reported in the lateral cerebellar hemispheres; the fundamental cognitive operation performed here remains uncertain.
The cerebellar activity does not appear related to motor responding or timing (Ivry 1993) because the condition of sustained attention controlled for these components and the behavioral data were not significantly different between attentional conditions. Neither is the role of the cerebellum system in error detection and correction (Fiez et al. 1992; Flament et al. 1996) used to explain coordinated movements and motor learning, tenable for these tasks, which produced no differences in errors or other performance measures. Likewise, the sensory aspects of the attentional tasks were matched (Gao et al. 1996).
The neocerebellum plays a more general role in cognitive abilities. Schmahmann (1996) has argued for parallel dual processing streams through the cerebellum; the present data are consistent with this formulation. First, a sensorimotor stream feeds forward from motor, premotor, supplementary motor, postcentral, and BA 5 cortices through the red nucleus to the olivary (climbing fiber) system to the cerebellar cortex of the medial anterior lobe and lobulus simplex as well as the paramedial lobule. Feedback occurs from the efferent Purkinje axons through the deep cerebellar nuclei to the thalamus and thence back to the cortices relevant to sensorimotor processing. The activation of the medial part of the anterior lobe of the cerebellum during sensorimotor processing is consistent with this first stream. Second, a cognitive stream arises from associative and paralimbic cortices and feeds forward through the corticopontine tracts onto the pontocerebellar circuits to the neocerebellar cortex. The activation of the lateral posterior cerebellar cortex during the switching of attention across stimulus features would follow this second stream.
Whether or not the cerebellum performs a unitary cognitive operation across both motor and cognitive streams remains uncertain. The highly repetitive structural motifs and the extensive cortical interconnections provoke such hypotheses. Cerebellar contributions to several cognitive paradigms have been reviewed recently (Schmahmann 1996; Thach 1996) and increasingly ascribe modulatory and tuning functions to the lateral cerebellar hemisphere. By analogy to the cerebellar contribution to sensorimotor processing, Thach (1996) suggests that the cerebellum's modulatory role in cognition reflects context linkage and the shaping of responses through learning: cortical information could flow to the lateral cerebellar hemispheres over mossy fibers transmitting information via parallel fibers to thousands of Purkinje cells, while error feedback in climbing fibers modulates the strengths of parallel fiber to Purkinje cell synapses. Whether or not climbing fibers participate in the cognitive, nonmotor, functions of the cerebellum in humans remains controversial (Schmahmann 1996).
In the specific context of attentional processing, Akshoomoff and Courchesne (1994) propose that the cerebellar cortex participates in the “rapid sequential changes and adjustment of neural activity in order to proceed from one condition to another.” Allen et al. (1977) hypothesize that the neocerebellum's participation in attention arises from the need to predict, prepare for, and adjust to imminent information acquisition, analysis, or action. The neocerebellum, through its connections via the thalamus to the prefrontal and posterior parietal cortices, might enable the coordination of mental skills regardless of any specific function in selective attention. The parallel decrease in activity with rehearsal between the right lateral neocerebellum and left prefrontal and anterior cingulate cortices (e.g., Raichle et al. 1994) would be consistent with these hypotheses: attention and other cognitive procedures in the prefrontal cortex diminish along with decreased need for tuning the response in the right lateral cerebellar cortex as a task becomes automated with practice. The present methods did not permit visualizing prefrontal activity. However, the frontal lobe must play an important role in tasks requiring attention shifts. The left dorsolateral prefrontal cortex connects via the thalamus to the right deep cerebellar nuclei, which in turn comprise the efferent pathway of the neocerebellum (Middleton and Strick 1994). Patients with frontal lesions show gross impairments in switching attention between features as contrasted with performance during sustained attention (Owen et al. 1991, 1993). The left dorsolateral prefrontal cortex plays also an important role in the Wisconsin Card Sort Test (WCST), which requires switching between number, color, and shape information (Berman et al. 1995; Nagahama et al. 1996). Both of these imaging studies of WCST also had significant activation of the right lateral cerebellum. In addition, a recent metaanalysis (Shulman et al. 1997) of visual processing experiments, that include language, memory, and attention, found consistent right thalamus and right lateral neocerebellar foci in common across nine different experiments. The authors stated that the right lateral cerebellum response was of a nonmotor origin but did not suggest a specific role for that locus. Whether or not the right lateral cerebellum is involved in the attentional process or related to a more general molulatory or tunning mechanism (Thach 1996) remains to be elucidated.
Cerebellar function in cognition probably extends beyond the above modulatory/tuning hypothesis—particularly in its mnemonic functions. For example, a recent report of motor learning provides evidence that the cerebellum supports memory for a complex sensorimotor skill (Shadmehr and Holcomb 1997). In contrast to previous literature (see review in Thach 1996), which finds that the neocerebellum decreases in activity as learning occurs, the study finds that another region of the cerebellum (more medial to the CrIA and CrIP responses reported here) increases in activity during the retrieval of a motor memory after a period of consolidation.
Parietal contributions to shifting attention
The activation of posterior parietal structures in concert with the cerebellum could be predicted from the known connectivity between the two brain regions (reviewed in Schmahmann and Pandya 1989). The parietal cortex projects to the pons, which feeds forward to the cerebellum (Schmahmann and Pandya 1989). The lateral neocerebellum, in particular, appears interconnected with the parietal cortex (Brodal 1979). Also, parietal and cerebellar cortices interact functionally (Allen and Tsukuhara 1974; Oka et al. 1975; Sasaki et al. 1975).
Converging studies indicate that the parietal cortices, particularly superior and posterior (BA 7) regions, contribute to attention processes with cognitive operations involving spatial codes. Studies of nonhuman primates show activity consistent with spatial coding (Anderson et al. 1985). The presence of neglect after lesions to parietal cortex has been documented extensively (Pascual-Leone et al. 1994). During visuospatial orienting of attention, patients with parietal lesions show marked impairment at disengaging attention after an invalid spatial cue (Posner et al. 1984). Positron emission tomography imaging studies of normal volunteers have reported that the superior parietal cortex activates upon orienting visuospatial attention to the contralateral field regardless of the direction of orienting (Corbetta et al. 1993). During tasks requiring visual search for conjunction of features, the superior parietal cortex increases in activity bilaterally (Corbetta et al. 1995). The superior parietal cortices may mediate the transformations involved in spatial-response incompatibility (Iacobini et al. 1996). One patient with bilateral superior parietal lesions shows illusory conjunction of features, particularly when multiple stimuli occur simultaneously (Friedman-Hill et al. 1995). Additional studies of patients with parietal lesions and Balint's syndrome highlight the importance of spatial processing in sustaining visual perception (Robertson et al. 1997). These findings, taken together, support the hypothesis that superior, posterior parietal cortices perform computations upon spatial codes involved in the binding of features (and stereotypic responses) for an attended stimulus.
The posterior superior parietal activity found in the present studies does not readily fit into the above conceptualization. No visuospatial orienting of attention occurred in these foveal tasks. Only one stimulus was present at any time so competition between stimuli on the basis of location could not happen. The tasks here never required conjunction of features, the response depended at any instant upon a single attribute, i.e., either color or shape. Although an element of stimulus-response incompatibility may be involved (i.e., after shifting attention, responding to the previous stimulus feature must be suppressed), spatial components do not appear important. Also, the subjects were studied after considerable practice and with stable performance. In addition, eye movements of several study participants were measured outside the MRI scanner in a simulated environment. Eye movements, if any, were below the sensitivity of the electro-occulogram (EOG) equipment. Nevertheless, we can not exclude the possibility that rapid switching of attention across stimulus features might require a greater degree of spatial focusing than when the attended feature remains constant in time.
Alternatively, the present results argue that the function of the parietal cortices may be broader and not necessarily restricted to computations involving spatial codes. The presence of a rapidly changing stimulus with frequent shifts of attention between stimulus features presents an unusual situation not found readily outside the laboratory. Under usual circumstances, the visual and attention systems rely heavily upon spatial information for perception. However, location is only one of many features that must be bound in the creation of a percept. We hypothesize that the superior parietal lobule, with its extensive connections to visual association and prefrontal cortices (Caminiti et al. 1996), may participate in the selection and binding of stimulus features even when spatial codes may not be required. For example, the current task of switching attention sequentially across stimulus features may require additional binding of visual with temporal information.
Recently, the superior parietal cortex was shown to increase in activity bilaterally during discrimination tasks requiring attention to a conjunction of grating orientation and grating location (Vandenberghe et al. 1996). Of particular interest, the comparison between the conditions of orientation discrimination and of location discrimination did not show significant additional superior-parietal activation. Changes in the direction of covert attention, localized and sustained throughout blocks of trials, also did not produce increases in superior parietal blood flow. In contrast, the superior parietal cortex did not activate during a visual search when attention was divided across multiple features of stimuli distributed throughout the visual field (Corbetta et al. 1991). An explanation consistent with all data is that unfocused, divided attention for detection may not have the same neural substrates as focused attention for discrimination with frequent switching across features.
The absence of activation in lower visual sensory areas in the contrast between the condition of shifting attention and of sustained attention is noteworthy. Multiple regions in visual (BA 18, 19) and inferior temporal (BA 37) cortices did activate during the condition of sustained attention to color or shape as compared with the condition of visual fixation. Recruitment of these regions has been reported previously during passive stimulation tasks (Zeki et al. 1991), during color and shape discrimination tasks (Gulyás and Roland 1991), and during tasks requiring attentional modulation of shape and color information (Corbetta et al. 1990). Activation in these regions occur also during the WCST (Berman et al. 1995; Nagahama et al. 1996). In contrast, “top-down” processing during the shifting of attention across visual features recruited here medial parietal cortices in addition to those required for perception of targets during sustained attention to color or shape. This activation of the medial parietal lobe, including cuneus and precuneus, can not yet be interpreted definitively because of the debate concerning its potential role in high-level visual representation (e.g., Buckner et al. 1995; Fletcher et al. 1995).
In summary, these data provide evidence that the neocerebellum, posterior superior parietal cortex, and cuneus/precuneus participate during the switching of attention across visual features presented foveally. In the current context, previous work would suggest tentatively that the cuneus/precuneus may be involved in visual associative processes; the neocerebellum switching and fine tuning the mental skill; the posterior parietal cortex participating in the selection and binding of visual and temporal features.
We thank Dr. Jeremy Schmahmann (Harvard University), who aided our anatomic interpretations and provided preliminary information from the cerebellum atlas; Dr. Peter Andersen, Dr. Peter Erhard, J. Strupp, and G. Adriany for technical assistance; and volunteer subjects for patience and perseverance. Helpful discussions with Drs. Kamil Ugurbil, Charles Nelson, and Toshinori Kato are greatly appreciated.
This work was supported by the National Center for Research Resources at the National Institutes of Health (Grants RR-08079 and MH-55346) and the Department of Veterans Affairs.
Address for reprint requests: X. Hu, Dept. of Radiology and Center for Magnetic Resonance Research, University of Minnesota Medical School, 385 East River Rd., Minneapolis, MN 55455.