Neural correlates of two imagined egocentric transformations

Transcription

1 NeuroImage 35 (2007) Neural correlates of two imagined egocentric transformations Sarah H. Creem-Regehr, Jayson A. Neil, and Hsiang J. Yeh Department of Psychology, University of Utah, 380 S E., Rm 502, Salt Lake City, UT 84112, USA Received 25 September 2006; revised 27 November 2006; accepted 30 November 2006 Available online 16 December 2006 Two egocentric spatial transformation tasks, hand and perspective rotation, were compared using the same visual stimulus within both block and eventrelated functional magnetic resonance imaging (fmri) paradigms. Both involved body-relative judgments but were predicted to vary in the recruitment of the body schema and a motor execution system. The Hand task required the imagined rotation of one s own hand to make a left right handedness decision. In contrast, the Viewer task required a perspective transformation and updating of the parts of a hand as an object. Previous behavioral and neuroimaging work suggested that hand rotations would rely on dynamic and biomechanical processing of body-part relations recruiting a motor processing system, whereas perspective transformations and the updating of object self relations would be supported by primarily visual spatial mechanisms. There was a common neural substrate found for both tasks including the lateral occipital areas, inferior and superior parietal cortex, and the cerebellum. Direct comparisons between the two tasks revealed greater activation in the Hand task in left superior and inferior parietal and premotor cortex and cerebellum, whereas the Viewer task showed greater activation only in the right lingual and fusiform gyri. Degree of rotation also modulated activity in the Hand task in bilateral superior parietal and premotor cortex, but not in the Viewer task. Implications of these regions for the role of dynamic body schema and motor processing in egocentric transformations are discussed Elsevier Inc. All rights reserved. Research on the human ability to imagine spatial transformations has a history that now spans over 30 years (Cooper and Shepard, 1973; Shepard and Metzler, 1971). One reason for continued interest in this topic may be that the process of mentally transforming and updating space is critical to many cognitive tasks such as object recognition, navigation, problem solving, and action planning. Furthermore, the computational question about whether imagined movements are analogs of physical movement has been able to be rigorously studied with the use of response time paradigms and more recently, neuroimaging methods. While mental rotation is often used as a general term for tasks that require imagined transformations in space, evidence suggests that different mechanisms may underlie separate classes of spatial transformations as a function of the frame of reference Corresponding author. Fax: address: sarah.creem@psych.utah.edu (S.H. Creem-Regehr). Available online on ScienceDirect ( that is being transformed. Previous work has made the distinction between rotations of one s perspective, which involves the manipulation of an egocentric frame of reference, and rotations of an object or array of objects, which involves the manipulation of an object-relative frame of reference (Presson, 1982; Wraga et al., 2000, 2005; Zacks et al., 2003b). The present work extends this distinction to examine differences within the egocentric reference frame itself. We directly compared cognitive performance and its neural correlates on tasks that involved body-part (hand) and body (perspective) transformations. Spatial frames of reference are a means of representing locations relative to some spatial framework. Spatial transformations recruit the representation and manipulation of potentially different frames of reference to solve a required task. Much of the early work on mental rotation focused on the human ability to make a decision about the congruency of one rotated object with respect to another. Shepard and Metzler (1971) found that the time required to make a decision about the similarity of the structure of two rotated objects was a function of the angular disparity between the two objects. This classic object mental rotation task involves the rotation of an object-relative frame in which spatial location is represented relative to intrinsic axes of the object. In contrast, tasks relying on a framework specified by axes of the body, recruit the transformation of an egocentric frame (Wraga et al., 2000). Importantly, this broad definition allows for an egocentric axis to be tied to many different body parts, such as the trunk, hand, head, and eyes (Colby, 1998; Howard, 1982). Characterizing imagined transformations as either object-relative or egocentric, although useful as a first-approximation, does not necessarily examine the potential overlap between the two types of processes. Of particular interest is a task that involves mental rotation of body parts (Parsons, 1987a). Although making a handedness or same different decision about a body part is thought to involve the egocentric process of comparing one s own limb to the visual stimulus, behavioral and neuroimaging data suggest that these types of stimuli lead to very similar results as non-body stimuli such as 2D or 3D novel shapes. These types of tasks typically lead to a monotonic response time function in which time to make a decision increases with increasing angular rotation. These response time functions have been interpreted as indicating that the process of mentally rotating an object is analogous to physical rotation of an object, constrained by physics to pass through continuous points in space. Neuroimaging results associated with tasks involving both /$ - see front matter 2006 Elsevier Inc. All rights reserved. doi: /j.neuroimage

2 S.H. Creem-Regehr et al. / NeuroImage 35 (2007) body-part and non-body-part stimuli have indicated activation in several dominant spatial-motor processing regions including the posterior parietal cortex PPC (primarily superior parietal SPL and the intraparietal sulcus IPS), posterior temporal cortex (including MT), supplementary motor area (SMA), lateral premotor cortex PM, primary motor cortex and cerebellum (Alivisatos and Petrides, 1997; Cohen et al., 1996; Kosslyn et al., 1998; Parsons et al., 1995; Podzebenko et al., 2005; Seurinck et al., 2004; Vingerhoots et al., 2002). Whereas both object and body-part stimuli appear to be constrained by the physical properties of rotation, research results emphasize the role of biomechanical representations in body-part transformations. First, Parsons (1987a) found that the response time to make a left right decision about hands or feet (given no explicit instructions on a strategy to use) was highly correlated with the time required to imagine a limb movement (without the left right decision). Both types of judgments were also highly correlated with participant s ratings of the awkwardness of moving into a given limb orientation. Subsequently, Parsons (1994) showed that the time required to physically move to a given hand position was highly correlated with the time to imagine moving the hand and the time required for the left right hand decision. Second, there are mixed results as to the extent of neural motor regions involved in object rotations, and researchers have found that differences in strategies used to mentally rotate objects may recruit egocentric body representations to a greater or lesser extent. For example, Kosslyn et al. (2001) found activation in primary motor cortex for an object rotation task only after giving subjects a strategy to visualize rotation of the object with their own hand. In a related study, Wraga et al. (2003) found greater primary motor/ premotor activation in an object rotation task when it followed a hand rotation task than when it followed another object rotation task. These findings are suggestive of both the flexibility in body strategies used to mentally rotate objects, and the tightly linked association between representations for mental body transformations and physical motor movement. Nevertheless, body-part transformations themselves can be differentiated from whole-body transformations. Although there has been less work on imagined body transformations, investigations have focused on two distinct tasks. The first involves spatial decisions about a visually presented image of a human body. Parsons (1987b) early work suggested that observers imagine their own body rotating to match the visually presented body in order to make a decision. Zacks et al. (1999, 2002a) later compared two types of spatial judgments in the context of visually presented bodies. Using pictures of bodies with an outstretched hand, they required observers to make a same different judgment (e.g., are the two figures presented the same or mirror image?) or a left right judgment (e.g., is the figure s extended hand a left or right hand?). They found very different response time functions for the two types of tasks. The same different task showed a monotonic increase in response time with increasing orientation disparity, as in earlier object-based mental rotation tasks. In contrast, the left right task indicated a flat response time function, likely a result of the ability to perform an egocentric perspective transformation of the body. In related fmri and TMS studies, right-hemisphere dominance in the parietal occipital cortex was found for the object-based task relative to the egocentric perspective transformation task (Zacks et al., 2002b, 2003a). A second, related paradigm has compared object- and perspective-based transformations in the context of spatially updating external objects with both cognitive (Presson, 1982; Wraga et al., 2000) and neuroimaging (Keehner et al., 2006; Wraga et al., 2005; Zacks et al., 2003b) paradigms. The spatial updating paradigm differs from the perspective transformation tasks involving visual bodies because an object or array of objects is presented (without a body), and the observer is explicitly told to imagine either an object or viewer transformation in order to name an external object in a given position. In general, behaviorally, when participants are given a spatial updating task to name an object in a given location after a specified imagined transformation, a large systematic response time advantage has been found for viewer versus object or array transformations (Creem et al., 2001b; Wraga et al., 2000). Furthermore, the rotation functions for a perspective rotation are less dependent on the angle of rotation as long as the rotation coincides with a major axis of the viewer s body. Notably, Creem et al. (2001b) found that the viewer advantage holds for physically impossible viewer rotations that defied the physics of gravity, given an imagined transverse rotation (rotating one s principal axis) with respect to the environment. For example, in one experiment, participants stood facing an array of objects presented on the wall and imagined rotating around the objects (as if walking on the wall) or imagined the rotation of the objects themselves. Several neuroimaging studies have now examined perspective transformations and spatial updating. Creem et al. (2001a) examining only the viewer rotation component of the task using fmri, found activation primarily in the posterior parietal cortex, lateralized to the left hemisphere. Zacks et al. (2003b) comparing object to viewer rotations using a similar spatial updating task, found that right intraparietal sulcus showed larger activation in the object versus the viewer rotation, and a region of the left superior temporal sulcus showed more activation in the viewer versus the object rotation. Most recently, Wraga et al. (2005) found activation in lower level motor regions including the premotor cortex extending to M1 in an object rotation task that was not present in the perspective rotation task. In contrast, the perspective task recruited SMA and occipital temporal regions implicated in motion perception. In all, no whole-body or perspective-based tasks have shown activation in the regions associated with motor preparation/execution that have been dominant in body-part rotation paradigms. Taken together, cognitive and neural investigations of spatial tasks involving body and body-part transformations have shown that mental representations of visually presented objects are in some way tied to physical body representations. However, previous work suggests that differences exist among egocentric transformations. Behaviorally, whereas body-part transformations appear constrained by biomechanical plausibility of movement, perspective transformations appear more physically flexible but constrained by experience. Furthermore, neuroimaging studies of perspective rotation tasks have not demonstrated the activation in premotor and primary motor cortex seen in many of the hand rotation tasks. Our hypothesis is that the body-part rotation tasks require a dynamic representation of intrinsic spatial relations of the body, or in other words, the body schema (Buxbaum et al., 2000; Reed, 2002). In contrast, perspective rotation tasks, while necessarily involving a transformation of the body to some extent, rely on visual spatial processing in an extrinsic coordinate system in order to update the location of an object relative to oneself. The proposed distinction between intrinsic and extrinsic coordinate systems within the egocentric frame of reference has been made in motor control

3 918 S.H. Creem-Regehr et al. / NeuroImage 35 (2007) research (Vindras and Viviani, 1998) and is supported by neuropsychological case studies (Buxbaum et al., 2000; Schwoebel et al., 2001). Intrinsic egocentric coding relies on proprioceptive information of the relative position of body parts, whereas extrinsic egocentric coding represents external target positions relative to the body. In support of this dissociation, Schwoebel et al. (2001) found that a patient with a severe deficit on an intrinsic task of pointing to her own body parts was able to compensate by using the extrinsic coordinate frame of locations in a familiar room. The goal of the present study was to directly compare two egocentric transformation tasks that are hypothesized to recruit different body processing mechanisms, varying in the extent that they recruit the body schema and its underlying motor control system. Given the same visual display (see Fig. 1), subjects performed both perspective and hand transformations. In the perspective rotations, subjects imagined rotating their perspective around the hand and determined whether a certain finger was on the right or the left. In the hand rotations, subjects imagined rotating their own hands to fit the displayed hand then determined whether the displayed hand was a right or a left hand. Previous theoretical and experimental work on spatial transformations led to several predictions. Broadly, it was predicted that imagined hand rotations would recruit regions of the motor system involved in the execution of movements to a greater extent than imagined perspective rotations. More specifically, we predicted increased activation in the lateral premotor cortex, inferior and superior parietal cortex, and the cerebellum in the hand versus perspective task. If any increased activation was found in the perspective versus the hand rotation task, we expected it to be in posterior visual or motion areas. Second, given that both tasks rely on visual spatial transformations, we predicted shared regions of activation across the two tasks in posterior visual and parietal cortex. Third, prior neuropsychological and neuroimaging research on body schema led to the prediction that the two tasks may differ in lateralization of posterior parietal activation. It was predicted Fig. 1. Examples of visual stimuli for the Hand task (a) and Viewer task (b). that the hand rotation task, recruiting a dynamic body representation, would be more left lateralized than the perspective task. Materials and methods Participants Eighteen healthy right-handed volunteers (11 females, 7 males) aged years (average age, 21 years) participated in the study. One additional subject was removed for excessive errors (greater than 20% on both tasks). Two subjects behavioral data were not recorded because of a computer error so sixteen subjects were analyzed for response time and accuracy, but eighteen were included in the imaging analysis. Participants were screened for any contraindications for MRI scanning and handedness was assessed using a modified version of the Edinburgh handedness scale (Oldfield, 1971). The University of Utah Institutional Review Board approved the experimental procedures, and all participants gave their informed consent before beginning the study. Design and procedure While in the MRI scanner, the subject viewed a screen on which an instruction was presented for 1.5 s before a hand was presented in a random orientation surrounded by six spheres for 3 s. Total length of each trial was 4.5 s. Using the bottom of the image as 0, the spheres were placed at 0, 60, 120, 180, 240, and 300 clockwise. Stimuli were created with the modeling program, Blender (www. blender.org). There were 24 different hand stimuli: 6 degrees of rotation 2 positions (palm up/down), 2 hands (left/right). Throughout the experiment the orientation of the spheres never changed, but in every trial one of the spheres was colored blue to indicate which sphere the subjects were to imagine viewing the hand from. Subjects responded using a custom made fiber optic response box for use with the MRI scanner. For the hand rotation task (hereafter named the Hand task), a screen displayed Hand to instruct the subject on the current task. During this task the bottom sphere was always blue (see Fig. 1a). The subjects were instructed to decide whether the hand presented was a right or left hand by imagining rotating their own hand into the position of the hand presented. Previous behavioral studies have found that this is the easiest and most preferred method for subjects to perform this type of task (Parsons, 1987a, 1994). For the perspective rotation task (hereafter named the Viewer task), a screen with the word Thumb or Pinky was displayed. In this task any one of the spheres was blue. The blue sphere was always in the orientation corresponding to the bottom of the hand (where the hand attaches to the wrist, see Fig. 1b). The subjects were instructed to imagine that they were standing at the blue sphere, and from that new imagined perspective to decide whether the previously named hand part, thumb or pinky, was on their right or left. This is also historically the easiest and most preferred method for performing this type of task (Presson, 1982). Before the experiment, subjects were required to familiarize themselves with the tasks by completing a practice version of the experiment outside of the scanner. Then immediately prior to the experiment subjects were reminded of the directions. Each subject participated in four functional runs. Two of the functional runs consisted of the trials presented in a block paradigm (205 frames/ s), while the remaining two functional runs were of the event-related paradigm (280 frames/

4 S.H. Creem-Regehr et al. / NeuroImage 35 (2007) s). The order of block and event-related runs was counterbalanced across subjects. Both paradigms consisted of the same two tasks and used the same stimuli. The event-related paradigm provides for greater flexibility and specificity in processing, while the block paradigm allows for more power and can be used as a standard for comparison. From a cognitive processing emphasis, blocked versus randomized trials could potentially lead to different strategies or performance as a result of more or less switching between tasks. Taken together, the two approaches should provide insight into the methodological differences between the two paradigms. The images were presented on a screen (37 27 in) at the foot of the scanner using an LCD projector (Sharp XG-E12004) and an IBM Thinkpad using the stimulus presentation software E-Prime ( Participants viewed the screen through a mirror placed above their eyes. The images were projected at approximately 5 in. in height. Block paradigm For the block paradigm the possible trial types were separated into the two tasks, viewer and hand. These were each then divided up into two blocks, a 0 rotation block and a block of every other degree (60, 120, 180, 240, and 300 ), following the design of Creem et al. (2001a). The result was four different task blocks, which we labeled: Hand zero (H0), Hand degree (Hdeg), Viewer zero (V0), and Viewer degree (Vdeg). Each task block was 18 s and contained four trials. A fixation cross appeared for 18 s after the completion of the four task blocks. Each run ( s) contained the sequence of the four task blocks plus fixation block, repeated five times. We varied the order in which each of the four blocks was presented, constrained by the rule that a degree group always followed the zero group of a given task. This design led to 80 total trials across the two runs (40 hand and 40 viewer with 20 zero trials and 20 degrees trials in each task). A fixation cross was also presented at the beginning of the run for s. Event-related paradigm The trial order for the event-related runs was determined using the program Optseq2 (Dale, 1999). Optseq2 arranges trials in an order that appears random to the subject while trying to jitter the trial onset times so that the overlaps of the hemodynamic response curves are removed from the image-processing estimate of the induced hemodynamic response. The overall result is a trial order that appears random and where trials from every task and degree are intermixed along with variable fixation times. This method allows a maximum number of trials presented while eliminating possible behavioral effects of expectation of upcoming trials by the subject. The two runs (each 630 s) together included 108 trials (54 hand and 54 viewer with 9 repetitions of each degree of rotation, including 0). MRI acquisition Functional MRI tasks were performed on a Picker Eclipse 1.5- T scanner. EPI images were acquired in a quadrature head coil with slice thickness 5 mm, FOV 55.4 by 25.6 cm, data matrix , repetition time 2.25 s, echo time 35 ms, and flip angle 90. Twenty-five images were acquired during each repetition time. Anatomical images were acquired using a 3D RF-FAST sequence with TE 4.47 ms, TR 15 ms, flip angle 25, bandwidth 25 khz, FOV 25.6 cm, image matrix , and slice thickness 2 mm. Imaging analysis Raw EPI data were ghost corrected, distortion corrected, and reconstructed with in-house Matlab routines to a matrix with square 25.6 cm field of view and in-plane resolution of 4 mm. Statistical analyses were performed using MATLAB (Mathworks, Inc., Natick, MA, USA) and Statistical Parametric Mapping (SPM99, The first 5 images of each task were discarded to ensure that the signal had reached equilibrium. EPI images were aligned to correct for head motion, and anatomical images were co-registered with the EPI images. All images were spatially normalized to the standard Montreal Neurological Institute (MNI) template and smoothed using isotropic Gaussian kernels of 8 mm. Block paradigm Individual and group analyses were performed. We applied a boxcar model convolved with the hemodynamic response function using a general linear model with five stimulus conditions for each participant (Friston et al., 1995). Two linear contrasts were defined on the individual level to test for specific condition effects for hand rotations: Task (Hand-fixation) and Degree (Hdeg H0). Two more linear contrasts were defined to test for specific condition effects for Viewer Rotations: Task (Viewer-fixation) and Degree (Vdeg V0). A third contrast was defined for each individual to compare hand and viewer rotation tasks (Hdeg Vdeg). The individual subject contrasts were used in subsequent group random effects analyses, using onesample t-tests. The statistical threshold for the random effects analyses was set to a minimum of t=4.71 (p< uncorrected threshold) with a minimum cluster size of 50 voxels (p < 0.05 corrected for multiple comparisons on the cluster level). All results are reported in MNI coordinate space. Event-related paradigm In addition to the pre-processing stated above for the block design, slice-timing correction was also applied to the event-related data. Slice timing was based on interleaved acquisition (even and then odd), reference slice 12, sixteen time-bins, and a reference bin of 4. Individual and group analyses were performed. On the individual level, trials were sorted into four trial types, left hand (LH), right hand (RH), left viewer (LV), right viewer (RV) in which left/right refers to the handedness of the hand stimulus presented and hand/viewer refers to the task condition, modeled as a box-car and convolved with a canonical hemodynamic response function. In addition, degree of rotation (0, 60, 120, 180) was modeled as a linear parameter for each trial type. All trials were included in the analysis. Statistical analysis involved a two-step process as described above. Contrasts were defined at the individual level: Hand (H), Viewer (V), RH LH, LH RH, Degree_Hand, and Degree_Viewer. At the group level, random effects one-sample t-tests were performed on the contrasts and a paired t-test was conducted to compare hand and viewer contrasts. The statistical threshold for the random effects analyses was set to a minimum of t=4.71 (p< uncorrected on the threshold level) with a minimum cluster size of 50 voxels (p<0.05 corrected

5 920 S.H. Creem-Regehr et al. / NeuroImage 35 (2007) for multiple comparisons on the cluster level). All results are reported in MNI coordinate space. Behavioral analysis Behavioral analyses for both event-related and block designs of the study were identical. Response accuracy and latency were recorded and analyzed, averaging across the two runs of the same design. A 2 (task) 6 (degree) repeated measures analysis of variance (ANOVA) was conducted on the response time for correct trials and percent error. Both task and degree were within subject variables. Results Behavioral results Block design The behavioral data for the block design are similar to the classical bell-shaped curve response time function found in congruency experiments for hand rotation tasks. The response time function for the Viewer task was also standard in that while it is somewhat similar in shape, response latency was faster than that of the Hand task (see Fig. 2). Overall, the subjects performed the Viewer task (M = ms) faster than the Hand task (M = ms, F(1,15) = 69.2, p < 0.001). The ANOVA also indicated an effect of degree of rotation (F(5,75)=36.17, p < 0.001), as well as task degree interaction (F(5,75) = 3.67, p < 0.01). This indicates that response latency increased with increased degree of rotation (up to 180 ), and the response time patterns differed between the perspective and Hand tasks. Planned contrasts were performed to assess the affect of degree of rotation for each task. For the Hand task, response time increased from 60 to 120 and 120 to 180 (p<0.001). Response time decreased from 180 to 240 (p<0.001). The response time difference between 0 and 60 as well as 240 and 300 was not significant (p>0.18). For the Viewer task, response time increased from 0 to 60 and 120 to 180 (p < 0.01). Response time decreased from 180 to 240 (p<0.001). The response time difference between 60 and 120 as well as 240 and 300 was not significant (p>0.4). Analysis of accuracy data revealed no overall difference between the Hand error (M = 6.25%) and the Viewer error (M=3.83%), but showed a significant effect of degree (F(5,75)= 6.33, p < 0.001), and a task degree interaction (F(5,75) = 2.54, p < 0.05). Planned repeated contrasts were performed to assess the influence of degree on each task. For the Hand task, errors increased from 120 to 180 (p<0.05) and decreased from 180 to 240 (p<0.05); for the Viewer task, errors increased from 120 to 180 (p < 0.05) and there were no other significant differences among degrees. In all, the error patterns looked similar across the two tasks, although the Hand task showed greater percent error at 180. Event-related design As should be expected, the behavioral results for the eventrelated design are similar to those of the block. Again the classic response time function of a bell-shaped curve was present for the congruency paradigm Hand task. The Viewer response time function also was similar to the Hand task only faster (see Fig. 3). The subjects performed the Viewer task (M= ms) more quickly than the Hand task (M = ms, F(1,15) = 60.76, p < 0.001). The ANOVA also revealed an effect of degree of rotation (F(5,75) = 67.36, p < 0.001) and a task degree effect (F(5,75) = 3.78, p < 0.01) indicating that response latency increased linearly with degree of rotation (up to 180 ), but that the response time function pattern differed between the Viewer and Hand tasks. Planned contrasts were performed to assess the affect of degree of rotation for each task. For the Hand task, response time increased from 60 to 120 and 120 to 180 (p<0.001). Response time decreased from 180 to 240 (p<0.001) and 240 to 300 (p<0.05). The response time difference between 0 and 60 was marginally significant (p<0.065). For the Viewer task, response time increased from 0 to 60 and 120 to 180 (p<0.01). Response time decreased from 180 to 240 (p<0.001) and 240 to 300 (p<0.01). The response time difference between 60 and 120 (p > 0.2) was not significant. Analysis of accuracy data showed that overall error was greater on the Hand task (M=9.91%) versus the Viewer task (M=5.26%, F(1, 15)=5.87, p<0.05). The ANOVA also revealed a significant effect of task (F(5, 75)= 12.73, p < 0.001) and task degree interaction (F(5, 75) = 12.73, p < 0.001). Planned repeated contrasts were performed to assess the influence of degree on each task. For the Hand task, errors increased from 60 to 120 (p<0.05) and 120 to 180 (p < 0.001) and decreased from 180 to 240 (p<0.001); for the Viewer task, errors increased from 0 to 60 (p<0.05) and marginally from 120 to 180 (p<0.054) and there were no other significant differences among degrees. As in the block design, the difference in the percent error between Hand and Viewer tasks can be mostly attributed to the 180 rotation. Fig. 2. Mean response time and percent error (±1 SE) for the Hand and Viewer tasks as a function of degree of rotation for the block design runs.

6 S.H. Creem-Regehr et al. / NeuroImage 35 (2007) Fig. 3. Mean response time and percent error (±1 SE) for the Hand and Viewer tasks as a function of degree of rotation for the event-related design runs. Comparison of event-related and block designs As is apparent from a comparison of Figs. 2 and 3, there was an overall response time advantage for the block design (M = ms) compared to the event-related design (M= ). This was confirmed by an additional 2 (design) 2 (task) 6 (degree) ANOVA performed on the data combined from the two designs. The ANOVA revealed a significant effect of design (F(1, 15)=11.42, p<0.01) but no design task interaction (p=0.57), indicating that although response time was faster in the block versus the event-related design, it was a constant shift for both tasks. The same combined design analysis was performed on the accuracy data, revealing no differences between the two designs (block mean=6.42%; event-related mean=6.83%, p=0.76) and no design task interaction (p=0.79). Imaging results Block design Activation for the Hand and Viewer tasks compared to fixation overlapped in lateral occipital areas, left inferior parietal cortex, right superior parietal cortex, right hippocampus, and the cerebellum (see Table 1 for coordinates and Brodmann areas and Fig. 4). However, activation found in the left superior parietal cortex, left precentral gyrus, left inferior frontal gyrus, left insula, and left SMA in the Hand task was not apparent in the Viewer task. The left superior parietal cortex, particularly the region surrounding the anterior intraparietal sulcus, has been shown to be involved in visually directed actions (Culham, 2004). The left precentral gyrus is part of the ventral premotor cortex which shares interconnections with M1 and the posterior parietal cortex, and is directly involved in the preparation and execution of action (Picard and Strick, 2001). This finding is an initial indication to support our hypothesis that brain regions involved in motor control are recruited more for the hand versus the Viewer task. More precise specifications of the activation involved in the hand and perspective transformation process of the tasks can be made by comparing the rotation blocks with the 0 blocks (see Table 2 for coordinates and Brodmann areas). For the Hand task, the analysis revealed significant activation in the left superior parietal cortex and bilateral middle occipital gyrus. In contrast, the Viewer task showed activation in the right superior parietal cortex, left inferior parietal including the supramarginal gyrus, left middle frontal gyrus, and bilateral middle occipital and temporal gyri. It is notable that when comparing the trials involving rotational degrees and those that do not, we find more activation for the Viewer task than the Hand task. This may result from the orientation of the hand stimuli presented to the participant. In the zero-degree trial type for the perspective task, the subject need only transform him- or herself to the front of the display, which is where he or she is physically located. This requires no rotational processing. On the other hand, in the zero-degree trial type for the Hand task, the subject s physical position of his or her hands may not be oriented in an analogous position to the hand presented. For example, in the trials where the hand is presented palm-up the subject must still rotate his or her hand along its longitudinal axis to check for congruency. In this way the zerodegree trials for the Hand task may still require a rotational component, which consequently could have minimized the differences between the zero-degree and rotational degree trials for Table 1 Hand and Viewer main effects (block design) MNI coordinates T value Hand-fixation L middle occipital gyrus R inferior occipital gyrus L precentral gyrus L cerebellum R cerebellum Vermis L Inferior parietal L inferior frontal gyrus L insula L precentral gyrus L medial frontal gyrus (SMA) L superior parietal R hippocampus R superior parietal R middle occipital gyrus Viewer-fixation L middle occipital gyrus R inferior occipital gyrus R cerebellum L thalamus R superior parietal R inferior parietal R hippocampus L hippocampus L inferior parietal

7 922 S.H. Creem-Regehr et al. / NeuroImage 35 (2007) left posterior parietal activation in the Hand task is consistent with left-hemisphere dominance in intrinsic or body-part representations (Tomasino et al., 2003), in contrast to right-hemisphere dominance which has been found for visual spatial descriptions of an external model (Chaminade et al., 2005) and object-based mental rotation (Tomasino et al., 2003; Zacks et al., 2003a). The distinction between body descriptions and intrinsic and extrinsic spatial tasks will be discussed more fully in the Discussion section. The Hand Viewer rotation contrast most directly compares the two tasks. Activation resulted in left superior frontal and precentral gyri, left superior parietal cortex including the precuneus, left insula, and bilateral cerebellum for the hand transformations versus the viewer transformations (see Table 3 for coordinates and Brodmann areas and Fig. 5), again supporting both a left-hemisphere network and increased activation in structures relating to motor execution for the Hand task. There was no greater activation in the Viewer versus the Hand task. Fig. 4. Visual representation of the Hand-fixation and Viewer-fixation contrasts from the block design group random effect analysis (n=18, p<0.05 corrected at the cluster level). Activation clusters are superimposed on a normalized 3D brain model. the Hand task. To further examine this conjecture, an additional contrast (H0 V0) was computed for each individual and used in a subsequent group random effects analysis (p < uncorrected, minimum cluster size = 50). The analysis revealed several regions of activation greater in the Hand zero-degree trials, consistent with a hand-motor transformation, including the supplementary motor area ( 6, 6, 56), and the dorsal premotor cortex (left superior frontal and precentral gyri, 24, 10, 62). There were no regions of activation significantly above threshold in the Viewer zero-degree trials versus the Hand zero-degree trials. One important distinction between these tasks is the lateralization of the superior parietal activation. Superior parietal cortex, particularly the posterior-medial region, has been traditionally associated with mental rotation tasks of objects, hands, and perspective (Bonda et al., 1995; Creem et al., 2001a; Keehner et al., 2006; Podzebenko et al., 2005; Zacks et al., 2003b). The present Table 2 Hand and Viewer tasks compared to zero rotation (block design) MNI coordinates T value Hand rotation Hand zero R middle occipital gyrus L middle occipital gyrus L superior parietal Viewer rotation Viewer zero R middle occipital gyrus R middle temporal gyrus L superior occipital gyrus 19/ L middle occipital gyrus L middle frontal gyrus L inferior parietal R precuneus R superior parietal Event-related design The results of the event-related experiments were consistent with the block design, although there were some differences, likely a function of the inclusion of different parameters in the model (handedness of stimuli and degree as a linear parameter) and potential differences as a result of mixed versus blocked trials. Overall, Hand trials showed significant activation lateralized to the left in the superior and inferior parietal cortex, dorsal and ventral premotor cortex at the precentral gyrus, SMA, thalamus, putamen, insula, and cerebellum; and bilateral occipital temporal cortex and right superior parietal cortex. The Viewer task showed some overlapping regions in left SMA, left inferior and superior parietal cortex, thalamus, and cerebellum, but also showed bilateral dorsal premotor cortex, and bilateral occipital parietal clusters (see Table 4 for coordinates and Brodmann areas). One notable difference here from the block paradigm is the recruitment of the dorsal premotor cortex in the Viewer task. It is our speculation that mixing hand and viewer trials led to a tendency to recruit the body schema in the viewer trials more than was the case with the blocked viewer and hand trials. This conjecture is consistent with previous findings that strategies and prior experience influenced whether motor representations were found in an object rotation task (Wraga et al., 2003). The RH LH and LH RH contrasts were conducted on the Hand task alone, to assess whether there were dissociations when decisions about right and left hands were required. Decisions about right hands showed significant activation in a large cluster centered on the left postcentral gyrus (M1) and the right cerebellum. Decisions about left hands resulted in activation centered on the Table 3 Hand rotation Viewer rotation (block design) a MNI coordinates T value L cerebellum L insula L superior frontal gyrus L precentral gyrus L superior parietal L precuneus R cerebellum a There were no significant activation clusters for Viewer rotation Hand rotation.

8 S.H. Creem-Regehr et al. / NeuroImage 35 (2007) Fig. 5. Activation in the left dorsal premotor and superior parietal cortex resulting from the hand rotation viewer rotation contrast from the block design group random effect analysis (n=18, p<0.05 corrected at the cluster level). Activation clusters are superimposed on a single subject's normalized anatomy. right postcentral gyrus (M1), left cerebellum, right insula, and right supplementary motor cortex (see Table 5 for coordinates and Brodmann areas). The contralateral M1 activation and ipsilateral cerebellum can easily be attributed to the motor response, as a right hand stimulus required a right finger movement to respond, and the left hand stimulus required a left finger movement to respond. No additional activation was found for right hands. However, the left hand stimulus did recruit additional areas of the insula and SMA, which are likely attributed to the more cognitive motor imagery component of the task. Since all subjects were strongly right handed, it is possible that decisions about left hands were less familiar to the subjects body representation and may have required additional cognitive strategies distinct from right hands. The direct comparison between Hand and Viewer tasks led to a clear dissociation between the tasks, consistent with the block design results. The H V paired t-test revealed left inferior and superior parietal and right cerebellum activation whereas the V H comparison revealed only one cluster of activation centered on the right lingual gyrus near the fusiform gyrus (see Table 6 for coordinates and Brodmann areas and Fig. 6). This finding suggests a more visual spatial representation of imagined perspective transformation relative to the motor representation involved in the Hand task. Activation in posterior occipital cortex is consistent with recent findings of Wraga et al. (2005) who found a somewhat more lateral region of the middle occipital cortex in a direct comparison of viewer and object rotation. Finally, the degree parameter for both the Hand and Viewer tasks was analyzed to assess whether there was a linear relationship between magnitude of degree of rotation and response. There was an effect of degree on the Hand task in bilateral dorsal premotor cortex at the middle frontal gyrus, bilateral superior parietal cortex, right inferior parietal, and the middle occipital gyrus (see Table 7 for coordinates and Brodmann areas). For the Viewer task, no areas were significantly above the threshold. This finding of degree of rotation for only the Hand task supports the notion that the hand rotation task recruits a motor control system, whereas the Viewer task does not. Time on task is not a likely explanation of this effect, Table 4 Hand and Viewer main effects (ER design) as RT increased both up to 180 both for the Hand and Viewer tasks, but only the Hand task showed significant linear effect of degree on hemodynamic response. Discussion MNI coordinates T value Hand R inferior occipital gyrus 18/ L precentral gyrus L middle occipital gyrus 18/ L cerebellum L medial frontal gyrus (SMA) L superior parietal L inferior parietal L putamen L insula R superior parietal L precentral gyrus L thalamus L thalamus L hippocampus Viewer R inferior occipital gyrus 18/ L inferior parietal L superior parietal L middle occipital gyrus 18/ L middle frontal gyrus R middle frontal gyrus L medial frontal gyrus (SMA) L cerebellum L thalamus L cerebellum R hippocampus R cerebellum The goal of the present study was to compare two spatial transformation tasks, both involving body-relative judgments but which were predicted to vary in the recruitment of the body schema. The Hand task required the imagined rotation of one s own hand to make a left right handedness decision, a behavioral task which has shown both the influence of biomechanical movement constraints (Parsons, 1994) and the recruitment the neural sub- Table 5 Effects of stimulus handedness on Hand task (ER design) MNI coordinates T value RH LH L postcentral gyrus R cerebellum LH RH R postcentral gyrus L cerebellum R insula R medial frontal gyrus (SMA)

9 924 S.H. Creem-Regehr et al. / NeuroImage 35 (2007) Table 6 Hand Viewer paired t-test (ER design) MNI coordinates T value Hand Viewer L superior parietal L inferior parietal R cerebellum Viewer Hand R lingual/fusiform gyrus strates underlying a motor execution system (Bonda et al., 1995; Kosslyn et al., 1998; Parsons et al., 1995). In contrast, the Viewer task required a perspective transformation and updating of the parts Fig. 6. Activation clusters resulting from the (a) Hand Viewer (L inferior/ superior parietal cortex) and (b) Viewer Hand (R lingual/fusiform gyrus) contrasts from the event-related design group random effect analysis (n=18, p<0.05 corrected at the cluster level). Activation clusters are superimposed on a single subject's normalized anatomy. Table 7 Degree main effect (ER design) a MNI coordinates T value Effect of degree on Hand L middle frontal gyrus L superior parietal L precuneus R superior parietal R middle frontal gyrus R middle occipital gyrus R inferior parietal a There were no significant clusters of activation for the effect of Degree on the Viewer task. of a hand as an object. Behavioral work on viewer transformations has shown that the egocentric frame of reference is rotated quite easily and is not constrained by physically possible movement of the body (Creem et al., 2001b). Neuroimaging investigations have implicated visual spatial processing systems, and regions associated with representing the self (Creem et al., 2001a; Wraga et al., 2005; Zacks et al., 2003b), but not a motor execution system. Our intent was to directly compare these two types of spatial transformations within the same paradigm, predicting some overlap in visual spatial processing mechanisms but dissociations in the extent to which a motor execution system was recruited. An additional goal was to compare two different fmri designs (block and event-related) with the same stimuli and tasks. In general our predictions were upheld, suggesting two partially independent neural systems for egocentric transformations. This claim is supported by several lines of converging evidence. Direct comparisons between Hand and Viewer tasks in both experimental designs showed statistically different regions of activation, categorized both by differences in lateralization and the recruitment of motor processing structures. The Hand task revealed activation in the left caudal dorsal premotor cortex (superior frontal gyrus) extending ventrally to the precentral gyrus (BA 6), left superior and inferior parietal cortex (BA 7 and 40), and the cerebellum. These regions clearly represent a motor execution system, with the absence of primary motor cortex (M1). The presence or absence of M1 in hand decision tasks has been mixed. While some have found M1 activation for mental rotation of objects and hands (Kosslyn et al., 1998, 2001), many other studies have found activation clusters limited to the surrounding premotor cortex (Lamm et al., 2001; Seurinck et al., 2004). The pattern of more left parietal activation for hand rotation and the additional right parietal activation for perspective rotation supports an important distinction between the proposed processes underlying the spatial tasks of updating hand position and one s own perspective. Body-part and perspective transformations not only differ on their adherence to biomechanical or motor constraints, but also in the spatial frame of reference for planning actions (Vindras and Viviani, 1998). One way of conceptualizing the extent to which the body schema is involved in spatial transformations is to ask whether the transformation requires extrinsic coding of an external object location or intrinsic coding of the positions of the body relative to the body itself. Extrinsic coding of objects relative to oneself after an imagined transformation of one s body is seen in object self spatial updating paradigms. The intrinsic system, providing dynamic information about the position of body parts

10 S.H. Creem-Regehr et al. / NeuroImage 35 (2007) seems most relevant to body-part transformations, most likely involving input from proprioceptive and motor systems. Evidence from both neuropsychology and neuroimaging supports a left-lateralized parietal system for motor planning/ preparation (Buxbaum et al., 2005; Tomasino et al., 2003) and intrinsic coordinate control (Chaminade et al., 2005; Schwoebel et al., 2004). Buxbaum and colleagues have argued that patients with ideomotor apraxia (IMA), typically resulting from lesions to left inferior parietal cortex, show deficits not only in skilled arm/hand movements, but in motor planning, particularly when tapping an intrinsic coordinate system. Jax et al. (in press) found that patients with IMA were particularly impaired in imitating postures or grasps directed toward body-relative targets compared to object-relative targets. Goldenberg (2001) has proposed a model of apraxia in which the left inferior parietal lobe is crucial to the imitation of gestures with reference to body-part coding whereas the right hemisphere contributes to gesture production when the task requires discrimination of spatial position. In an fmri study of imitation, Chaminade et al. (2005) found results consistent with this body-part versus spatial position distinction. Reproducing spatial components of an action independent of the limb used (spatial position task) led to right superior parietal activity, whereas reproducing movements using the same limb (body-part coding) activated the left inferior parietal cortex and the insula. In the context of mental rotation, several studies support the distinction between left parietal dominance for body-based transformations and right parietal dominance for object-based transformations. A direct comparison between two types of mental rotation in human lesion patients was conducted by Tomasino et al. (2003). They found a double dissociation between mental rotation of body parts and external objects in that right-hemisphere damaged patients were impaired on object rotation but not hand rotation, whereas left-hemisphere damaged patients were impaired on hand rotation but not object rotation tasks. This double dissociation is mostly consistent with recent findings from Zacks et al. (2003a) who found that cortical stimulation of the right superior parietal cortex disrupted an object-based transformation of a body stimulus, but not an egocentric spatial transformation of one s own body to match the picture. Notably, one of the object rotation tasks in Tomasino et al. (2003) was Ratcliff s (1979) little man task in which drawings of a man holding a black disk in one hand are presented and the subject must decide which hand holds a black disk. A similar stimulus has been used by Zacks et al. (2002a,b) to tap distinctions between object- and perspective-change processes, suggesting that the frame of reference taken in spatial transformations likely influences the cognitive and neural mechanisms involved. In our present study, although decisions about visually presented hands were required in both tasks, there is an important cognitive distinction to consider. Whereas the Hand task required the stimulus to be considered as part of one s own body, processed by an intrinsic and dynamic coordinate system, the Viewer task required updating of the parts of the hand as objects relative to oneself, likely a process subserved by the extrinsic egocentric system. When egocentric transformations require objectrelative decisions, more right parietal processing may be recruited. Updating object location relative to oneself would also be predicted to be less impaired in people with intrinsically defined motor planning deficits such as IMA. This question, potentially best studied using a neuropsychological approach, remains a topic for future investigation. The direct Viewer Hand comparison showed only one region of activation in the ER paradigm (and none in the block design), namely a region of the right posterior occipital cortex (lingual gyrus, BA 18). This finding further supports the interpretation that the Viewer task recruits a visual spatial transformational process rather than a motor process. The task of deciding whether a specific part of the hand (pinky or thumb) is in a certain location relative to oneself likely requires more visual analysis than the body-based process of mental hand rotation. The overall lack of abundant activation in the Viewer versus Hand comparison is consistent with the few other studies which have compared perspective transformations to a control task (Keehner et al., 2006; Zacks et al., 2003b). Decisions about left versus right hands in the Hand task recruited additional activation in the insula and the supplementary motor area beyond that which would be associated with a left hand button response. It may be that the motor systems in these strongly righthanded subjects had an advantage for making decisions about right hands but that left hand decisions required additional higher level imagery processing, recruiting the insula and the SMA. Both have been implicated in cognitive decisions about representations of the body (Chaminade et al., 2005). In particular, the insula has been found to be involved in mental rotation of hands (Bonda et al., 1995) and more generally, an internal representation of the body (Berlucchi and Aglioti, 1997). Final support for the distinction in processing between the two tasks comes from the inclusion of degree of rotation as a linear parameter in the event-related model. The results indicated that degree of rotation had a significant effect on activation clusters in the Hand task, consistent with the hand activation itself. Activation was modulated by increasing degree of rotation in parietal and premotor cortex, with the largest cluster of activation in left superior parietal cortex, extending medially to the precuneus. Keehner et al. (2006) found a similar modulation of activation in the right superior parietal cortex for object rotations in their study which compared object and perspective transformations. In both the present study and Keehner et al. (2006), perspective transformations were not positively related to degree of rotation. Taken together, although the few published studies on the neural mechanisms underlying perspective transformations have varied in the localization of the proposed mechanisms, they portray a common account that appears to support visual spatial, but not motor, processing. The somewhat varied results in localization across studies are likely due to different control tasks, strategies in perspective taking, and/or analysis procedures. Finally, it is useful to compare the results across the block and event-related designs. It is a significant finding that in general, the imaging results from the two paradigms were similar, although more specific analyses such as handedness decisions and parameterization of degree of rotation were allowed with the use of the event-related design. Notably, the behavioral data showed that there were overall response time differences between the two designs, with the block trials performed more quickly than the event-related trials. From a cognitive perspective, this result suggests that there are some costs to switching between the two egocentric transformation tasks and that potentially one task may be performed differently in the context of mixed versus blocked trials. This notion is supported by the finding of premotor cortex activation in the Viewer task, only when intermixed with Hand trials in the event-related design. However, across the two paradigms, the consistent relative response time data as a function of task and degree and the consistency of the imaging results suggest more similarities than differences between the two methodological approaches. In conclusion, the present findings contribute to the understanding of spatial transformations, particularly those involving an