In this initial experiment subjects planned and/or executed tool use gestures with their dominant right hands. Brain areas involved in action planning were isolated by contrasting results of the TOOL-NOGO condition versus CONTROL-NOGO condition, as depicted in . On the basis of findings discussed above, we expected activity in three general regions, all within the left hemisphere: posterior temporal, inferior-middle frontal and posterior parietal cortices.
Figure 1. Experimental design for experiments 1 and 2. All stimuli were presented aurally and subjects' eyes remained closed throughout testing. On half of the trials the instructional cue (IC) identified a familiar tool commonly used unimanually with the dominant (more ...)
Similarly, regions involved in the execution of tool use gestures were isolated by contrasting results of the TOOL-GO versus TOOL-NOGO conditions (). To the extent that the above-mentioned areas are involved in the production of praxis skills, we reasoned that they too should be active, along with established sensorimotor regions of frontal and parietal cortex, basal ganglia and cerebellum.
Thirteen healthy adults (10 females, 3 males) participated in experiment 1. All were right hand dominant as verified by the Edinburgh Handedness Inventory (Oldfield, 1971
), and none had a history of neurological or psychological illness. This protocol was approved by the Committee for the Protection of Human Subjects at Dartmouth College.
Stimuli were presented aurally via a non-commercial, MRI-compatible headphone system. Stimulus timing was controlled by a microcomputer running Presentation software (Neurobehavioral Systems, Davis, CA). Software was triggered by an external signal generated by the MRI-scanner at the onset of data acquisition. Stimuli consisted of the digitally recorded names of 30 familimanipulable tools/utensils, whose uses are performed unimanually with the dominant hand (e.g. knife, hammer, or pencil).
Each subject performed six functional runs with their eyes closed. Runs each consisted of 60 trials and lasted 8 min. An additional 10 s of rest occurred at the beginning and 20 s rest at the end of each run. Counterbalancing of stimuli in each run was optimized so that items from any one condition had an equal probability of preceding or following items from any of the other conditions. The order of runs was counterbalanced across subjects. Each trial began with the subject's left hand resting comfortably at their side and the right hand palm down on their torso. Each trial consisted of three components: (i) an instructional cue (IC); (ii) a variable delay interval; and (iii) a movement cue (MC). Blank time was digitally added to all IC and MC stimuli to create files that were uniformly 1000 ms in length ().
On 50% of trials, ICs identified a familiar tool. When hearing one of these tool ICs, subjects used the delay interval to prepare to pantomime the associated action. If the subsequent MC was a GO signal, they executed the pantomime. If the MC was a NOGO signal, they relaxed until the next IC occurred. An equal number of randomly intermixed trials began with the IC ‘move.’ During the delay interval on these control trials, subjects simply prepared to move their hand in a non-meaningful fashion for ~2 s. Subjects were encouraged not to simply repeat the exact same movement on each trial, but were otherwise unconstrained. If the MC was a GO signal, they would then execute the control movement. If it was a NO-GO signal they remained stationary and waited for the next trial to begin. Subjects were required to perform gestures such that they would be recognizable to observers. In order to minimize motion artifacts, subjects were given practice keeping the upper arm stationary while executing gestures and control movements with the hand, wrist and forearm. Following a GO, the designated action was repeated once before the hand was returned to the starting point in preparation for the next trial. Gestures were visually monitored via closed circuit television to ensure compliance throughout each study.
Magnetic Resonance Imaging
Imaging was performed with a General Electric Horizon whole-body 1.5 T MRI scanner using a standard birdcage head coil. Head movements were minimized by the use of a therma-foam pillow and padding. Prior to each functional run, four images were acquired and discarded to allow for longitudinal magnetization to approach equilibrium. Within each functional run an ultrafast echo planar gradient echo imaging sequence sensitive to blood-oxygenation-level-dependent (BOLD) contrast was used to acquire 25 slices per TR (4.5 mm thickness, 1 mm gap, in-plane resolution = 3.125 3 3.125 mm). The following parameters were used: TR = 2500 ms, TE = 35 ms, flip angle = 90. A high-resolution, T1-weighted, axial fast spin echo sequence was used to acquire 25 contiguous slices(4.5 mm slice thickness with 1.0 mm gap) coplanar to BOLD images: TE = min full, TR = 650 ms, echo train = 2, FOV = 24 cm. High resolution (0.94 x 0.94 x 1.2mm), whole-brain, T1-weighted structural images were also acquired using a standard GE SPGR 3-D sequence.
Structural and functional images were preprocessed and analyzed using SPM99 (http://www.fil.ion.ucl.ac.uk/spm
). Functional data for each individual subject were corrected for differences in time of slice acquisition, and head motion. Functional and structural images were coregistered and transformed into a standardized, stereotaxic space. This resulted in 25 axial slices of isotropic, 3.125 mm3
voxels. Data were smoothed with an 8 mm FWHM, isotropic Gaussian kernel. Data were modeled as variable duration (7 or 9 s) events time-locked to the onset of the IC, using the appropriate duration boxcar convolved with the canonical hemodynamic response function in SPM99.
Results of fixed effects analyses at the level of individual subjects were submitted to second-level random effects analyses with subjects as the random factor. Statistical parametric maps were constructed based on differences between trial types using a t
-statistic. Clusters (K
) consisting of at least eight voxels, separated by a minimum of 8 mm and having t
-values of ≥2.57 (P
< 0.01) were considered statistically significant. Corrections for multiple comparisons were not performed given the a priori
hypotheses concerning a small number of anatomically defined regions of interest as discussed above. Results were converted to the standardized coordinate system of the Talaraich Atlas (Talairach and Tournoux, 1988
) using a nonlinear transformation (http://www.mrccbu.cam.ac.uk/Imaging/mnispace.html
), and displayed on the group mean hi-resolution T1-weighted structural image after normalization. Surface renderings were created using MRICRO software: (http://www.cla.sc.edu/psyc/faculty/rorden/render.html
). Matlab 6.0 (The Mathworks, Natick, MA) was used to perform inclusive masking in order to compare results across experiments.
Region of Interest Analyses
The locations of peak activations in left frontal, parietal and temporal cortices were identified as regions of interest (ROIs) from the statistical parametric map resulting from the random effects comparison of TOOL-NOGO versus CONTROL-NOGO conditions, P
< 0.01, uncorrected for multiple comparisons, K
≥ 10, minimum separation 8 mm. For each subject, BOLD response data from IC onset to 10 s (4 TR
s) post-IC were extracted from all voxels located within 8 mm radius spheres centered on peak locations using the ROI Toolbox v1.7 (http://sourceforge.net/projects/spm-toolbox
) and Matlab 6.0. Values ± 2 SD from the mean of each condition were considered outliers and eliminated. Repeated measures ANOVAs were then performed on individual subjects' time-averaged data, pooled across all voxels within a given ROI, with session as the random variable.
Locations of posterior parietal cortex activations in individual subjects were determined by overlaying co-registered statistical parametric maps (P
< 0.001, K
≥ 10) on each subject's high resolution, T1-weighted anatomical scan and cross checking with an anatomical atlas (Duvernoy,1991
Results and Discussion
Of primary interest is identifying areas activated during the delay interval when subjects are planning a tool use gesture that is then aborted (TOOL-NOGO) versus preparing a random, non-meaningful, limb movement (CONTROL-NOGO). As shown in , group data indicates that conceptualizing tool use gestures primarily activates a subset of regions within the left hemisphere, most notably: (i) posterior parietal cortex within the IPS and extending ventrally into SMG, ANG and ventral SPL;(ii) posterior temporal cortex within the STS and extending ventrally into MTG and inferior temporal gyrus (ITG); (iii) inferior-middle frontal cortex; and, unexpectedly, (iv) dorso-lateral prefrontal cortex (DLFPC) (see ). By contrast, significant activations are absent in posterior right parietal and frontal cortices, although two small clusters in posterior STS and anterior STG did reach significance.
Figure 2. Activations associated with planning and executing right hand tool use gestures. (A) When compared with preparing random hand movements (CONTROLNOGO condition), planning tool use gestures for the right hand (TOOL-NOGO) is associated with major activations (more ...)
Cortical regions showing greater activation during planning of tool use gestures (TOOL-NOGO) versus preparation of control movements (CONTROL-NOGO) for the right hand (experiment 1)
On the basis of earlier findings discussed above, repeated measures ANOVAs were used to analyze data extracted from 8 mm radius spheres centered on the peak regions of activation within left posterior parietal, frontal and temporal cortices. Instructional cue (TOOL versus CONTROL) and MC (GO versus NOGO) were fixed factors. The main purpose of these ROI analyses was to ensure that the differences between experimental and control conditions reflected in the statistical parametric maps result from differences in relative BOLD activations, as opposed to deactivations.
Posterior Temporal Cortex
Activations in left posterior temporal cortex were observed within and adjacent to the STS with a peak in posterior MTG (-49, -49, 11; ). illustrates that IC had a main effect on responses in this region, F(1,12) = 278.0538, P < 0.0001, MSE < 0.00001. Neither the main effect of MC nor the IC x MC interaction had a significant effect, P > 0.05 in both cases. Also, responses in the conditions involving tools were unaffected by the identity of the MC, P = 0.30.
Figure 3. Percent signal change at locations of peak activity in temporal, frontal, and parietal areas during right hand gesture planning. Panels illustrate percent signal change for each condition in data extracted from the locations of peak activity in left posterior (more ...)
Activation of this area is consistent with a previous study of tool use gesture planning (Choi et al., 2001
). As detailed above, activations in this region are also observed in tasks that involve observing or identifying tools and/or the actions with which they areassociated(Martin et al., 1995
; Chao et al.,1999
; Damasio et al., 2001
; Kellenbach et al., 2003
),and damagehereisassociated with category-specific deficits in action naming (Tranel et al., 1997
). Observation of this region in the present task is consistent with these results in that processing the identity of stimuli and accessing the knowledge concerning associated actions is a key component of planning tool use actions.
Peak frontal activation was centered in left inferior frontal gyrus, specifically in pars opercularis (-50, 16, 10), and extended dorsally into the GFm and rostrally into pars triangularis (Amunts et al., 1999
). As shown in , here too there was a greater response when subjects were instructed to prepare a tool use gesture versus a control movement, F
(1,12) = 5.3052, P
= 0.04, MSE = 0.0001. Main effects of MC and interactions between IC and MC were not significant (P
> 0.05), and responses were again unaffected by whether or not a planned tool use gesture was subsequently executed, P
In addition to this peak in inferior frontal cortex, we also observed activations in ventral premotor cortex (inferior pre-central gyrus), and to a lesser extent within inferior GFm. Previous studies of tool use gesture planning observed activations in GFm, but not more inferior regions of frontal cortex (Moll et al., 2000
; Choi et al., 2001
). As detailed earlier, left unilateral activation of all three of these frontal regions has been demonstrated during perceptual and semantic tasks involving tools (Martin et al., 1995
; Grabowski et al., 1998
; Chao and Martin, 2000
; Damasio et al., 2001
; Kellenbach et al., 2003
). Involvement of these regions could reflect activation of motor representations pertaining to the manipulation of tools (Chao and Martin, 2000
), as this region is also active during object grasping and manipulation (Binkofski et al., 1999a
; Ehrsson et al., 2001
In contrast to earlier studies (Moll et al., 2000
; Choi et al.,2001
), we did not detect significant activations in dorsal premotor areas. One small and unexpected activation in left DLFPC was, however, detected. A similar area has been noted in some subjects during gesture planning (Moll et al., 2000
), and previous studies have identified this region as being involved in semantic working memory (Gabrieli et al., 1998
; Poldrack et al., 1999
; Wagner et al., 2001
). As suggested by an anonymous reviewer, it seems reasonable that left DLFPC could therefore be involved in accessing, maintaining and/or manipulating representations of tool use actions that are stored in posterior temporal, inferior frontal and/or posterior parietal cortices. This is particularly crucial in the context of this delayed response task.
Posterior Parietal Cortex
Two large clusters were observed in left posterior parietal cortex along the IPS. The more anterior of these was located in the SMG along the ventral bank of the IPS (-50, -29, 33). As illustrated in , this region showed a greater response on trials where the IC was a tool, F(1,12) = 89.7667, P < 0.00001, MSE = 0.00001. Both the main effect of MC and the interaction between IC and MC were not significant, P > 0.05 in both cases. Likewise, a post-hoc comparison failed to detect any difference between conditions where tool use gestures were merely planned (TOOL-NOGO) versus planned and executed (TOOL-GO), P = 0.40.
The anterior SMG site is generally consistent with results of tasks involving tool observation and/or action semantic processing (Martin et al., 1995
, 1996 Chao and Martin, 2000
; Okada et al., 2000
), processing spatial relations between objects (Damasio et al., 2001
) or gesture planning (Moll et al., 2000
). As will be discussed in detail below, this area is in the vicinity of a region associated with visually guided prehension and/or manipulation of objects (Binkofski et al., 1998
; Chao and Martin, 2000
; Jancke et al., 2001
; Shikata et al., 2001
; Grefkes et al., 2002
; Culham et al., 2003
). Thus, its activation during conceptualization of tool use actions could reflect retrieval of stored attributes associated with grasping and manipulating tools (Chao and Martin, 2000
), and/or the explicit retrieval of knowledge of tool use actions (Kellenbach et al.,2003
shows a similar main effect of IC in the more posterior site, situated in the ventral bank of the IPS at the boundary of SMG and ANG (-38, -52, 56), F
(1,12) = 16.1049, P
= 0.0017, MSE = 0.0001. Again, neither the main effect of MC nor the IC x MC interaction reached significance, P
> 0.05 in both cases. There was again no effect of whether prepared tool use gestures were subsequently executed or aborted, P
= 0.37. Previously, activation in this more caudal region of the IPL has been observed during the planning of tool use gestures (Moll et al., 2000
). As argued below in the comparison of results from experiments 1 and 2, this more caudal region appears to be activated exclusively in tasks involving explicit planning of actions. We therefore hypothesize that it is involved in representing motor programs for acquired tool use skills. This is consistent with results of lesion location studies in IM patients that show greatest overlap in posterior SMG and ANG (Haaland et al., 2000
Individual Differences in Posterior Parietal Cortex
illustrates the relative locations of SMG, ANG, and SPL in left posterior parietal cortex on a 3-D surface rendering of a single subject's high-resolution, T1-weighted, structural MRI. These regions were identified manually in each individual in order to localize peak activation(s) in left and/or right posterior parietal cortices (). Subjects can be grouped into three categories depending on the laterality of these activations. Only one subject did not show involvement of the left posterior parietal cortex. Instead, this individual had a significant activation of the right ANG. The majority of subjects (53.8%) had left unilateral posterior parietal cortex activity. Five out of 13 (38.5%) subjects showed some degree of bilateral posterior parietal cortex activity. In all bilateral cases, however, clusters were larger in the left posterior parietal cortex. Across all subjects, left posterior parietal cortex activations were most common in SMG (84.6%), followed by SPL (46.2%) then ANG(15.3%). For those cases with right posterior parietal cortex activation, ANG (30.7%) was most frequent, followed by both SMG (23.1%) and SPL (23.1%).
Figure 4. Major anatomical divisions of posterior parietal cortex in an individual subject. For purposes of localizing activations in individual subjects, boundaries between SPL (green), SMG (yellow) and ANG (purple) were determined on the basis of anatomical landmarks (more ...)
Locations of posterior parietal activations in individual subjects during planning of tool use gestures for execution with the right hand
The fact that left SMG was the most frequently activated region amongst subjects is consistent with the hypothesis that this region plays a key role in representing memories for skilled praxis (Heilman et al., 1982
). Yet, nearly half of the subjects displayed left SPL activations, which have been reported previously in association with tool use planning (Choi et al., 2001
), and are known to also be involved in the on-line control of reaching (Grafton et al., 1996
). The existence of a single subject with right hemisphere representations mirrors the relatively infrequent reports in the literature of crossed apraxia in righthanders, in which bimanual praxis deficits follow right unilateral lesions (Marchetti and Della Sala, 1997
; Raymer et al., 1999
). Although, as noted in experiment 2, this subject shows bilateral parietal activity for left hand gesture preparation. These individual differences in the location(s) of posterior parietal cortex involvement may contribute to variation in the consequences of parietal damage for skilled praxis (Basso et al., 1980
; Haaland et al., 2000
; Kertesz and Ferro, 1984
). In contrast to frontal lesions, there appears to be considerably more variability in the locations of maximal lesion overlap amongst parietaldamaged apraxics (Haaland et al., 2000
Areas active during gesture execution were isolated by contrasting results of the TOOL-GO versus the TOOL-NOGO conditions (). shows that with the notable exception of left DLFPC, the general regions associated with planning are also active during gesture execution. This includes not only left posterior parietal and inferior-middle frontal regions implicated in sensorimotor transformations during limb movements, but also left posterior temporal cortex. Activation of this later region indicates that the influence of ventral stream representations of tools and/or knowledge of associated actions are also involved during the execution of tool use gestures.
Of course, these areas are in addition to a variety of cortical and subcortical (cerebellum, basal ganglia) structures known to be involved in praxis (Imamizu et al., 2000
; Moll et al., 2000
; Choi et al., 2001
), and/or the representation of acquired motor sequences (Grafton et al., 1998
; Keele et al., 2003
), including: sensorimotor, dorsal and ventral premotor, supplementary motor, posterior parietal, posterior temporal, and ventral prefrontal cortices (). With the exception of contralateral activation in primary sensory motor areas, the bilateral nature of these activations is the most striking difference between activations associated with gesture execution versus planning. A more detailed analysis and discussion of the precise relationship between areas involved in these processes follows experiment 2.
Cortical regions showing greater activation during preparation and execution of tool use (TOOL-GO) gestures versus preparation (TOOL-NOGO) for the right hand (experiment 1)