NeuroImage 43 (2008) Contents lists available at ScienceDirect. NeuroImage. journal homepage:

Transcription

1 NeuroImage 43 (2008) Contents lists available at ScienceDirect NeuroImage journal homepage: Action word meaning representations in cytoarchitectonically defined primary and premotor cortices Natasha Postle a, Katie L. McMahon b, Roderick Ashton a, Matthew Meredith b, Greig I. de Zubicaray b, a School of Psychology, University of Queensland, Brisbane, QLD, Australia b Functional MRI Laboratory, Centre for Magnetic Resonance, University of Queensland, Brisbane, QLD, Australia article info abstract Article history: Received 7 April 2008 Revised 8 July 2008 Accepted 5 August 2008 Available online 16 August 2008 Keywords: Language comprehension Motor function Mirror neurons Somatotopy Verbs Semantic memory Recent models of language comprehension have assumed a tight coupling between the semantic representations of action words and cortical motor areas. We combined functional MRI with cytoarchitectonically defined probabilistic maps of left hemisphere primary and premotor cortices to analyse responses of functionally delineated execution- and observation-related regions during comprehension of action word meanings associated with specific effectors (e.g., punch, bite or stomp) and processing of items with various levels of lexical information (non body part-related meanings, nonwords, and visual character strings). The comprehension of effector specific action word meanings did not elicit preferential activity corresponding to the somatotopic organisation of effectors in either primary or premotor cortex. However, generic action word meanings did show increased BOLD signal responses compared to all other classes of lexical stimuli in the pre-sma. As expected, the majority of the BOLD responses elicited by the lexical stimuli were in association cortex adjacent to the motor areas. We contrast our results with those of previous studies reporting significant effects for only 1 or 2 effectors outside cytoarchitectonically defined motor regions and discuss the importance of controlling for potentially confounding lexical variables such as imageability. We conclude that there is no strong evidence for a somatotopic organisation of action word meaning representations and argue the pre-sma might have a role in maintaining abstract representations of action words as instructional cues Elsevier Inc. All rights reserved. Introduction It is now generally accepted that word meaning is represented in multiple areas of human cortex. This much was proposed by models of language comprehension in the mid-late nineteenth century. The classical Wernicke Lichtheim model considered concepts to be distributed throughout the cerebral cortex, aroused by associations among speech perception and production related memory images in the left hemisphere superior temporal and inferior frontal cortices. As early as 1885, Lichtheim had proposed anatomically distributed and interconnected conceptual centres (see his Fig. 7; Compston, 2006; Smith, 1996), while Freud (1891) would later propose a distributed object concept system linked to word meanings (Henderson, 1992). Modern extensions and revisions of these early models continue to adhere to this precept, often making the additional assumption that representation of word meaning involves distributed networks of motor and somatosensory areas (e.g., Gallese and Lakoff, 2005; Martin and Chao, 2001; Tranel et al., 2001). Some recent models of language comprehension have assumed a direct coupling between action meaning representations and cortical motor areas. One mechanism proposed to give rise to such a coupling Corresponding author. Fax: address: greig.dezubicaray@cmr.uq.edu.au (G.I. de Zubicaray). is Hebbian learning, whereby any two cells or systems of cells that are repeatedly active at the same time will tend to become associated,so that activity in one facilitates activity in the other (Hebb, 1949, p. 70). Hence, the frequent co-presentation of an action word with the execution of the action the word refers to might result in a representation being associated with those motor areas (Pulvermüller, 1996, 2005). Alternatively, a class of visuomotor neurons in premotor cortex mirror neurons that respond congruently when an action is both observed and executed might serve to transform visual information into knowledge coded at an abstract level (Rizzolatti and Craighero, 2004). This mechanism might permit action knowledge to be retrieved by action related words (verbs). Indirect evidence for cortical motor area involvement in action word meaning representation has been provided by transcranial magnetic stimulation (TMS), behavioural and neuroimaging studies. For example, several TMS studies have shown that processing of action related words and sentences is modulated by TMS-induced changes in excitability of left hemisphere cortical motor areas. During application of TMS, Pulvermüller et al. (2005) demonstrated faster lexical decisions to visually presented action words, while Buccino et al. (2005) noted responses slowed when participants listened to effector related action sentences. Oliveri et al. (2004) and Buccino et al. (2005) reported, respectively, that motor evoked potentials (MEPs) recorded from effector muscles were enhanced or reduced when TMS /$ see front matter 2008 Elsevier Inc. All rights reserved. doi: /j.neuroimage

2 N. Postle et al. / NeuroImage 43 (2008) was applied at long or short temporal intervals following action word presentation. Behavioural studies have also shown that action word processing can affect motor performance deleteriously when presented shortly after action onset (b200 ms; e.g., Boulenger et al., 2006). In order to explain these apparently contradictory facilitation and interference effects, Boulenger et al. (2006) have recently suggested the former could reflect post-lexical engagement of motor imagery mechanisms while the latter might be due to competition for common motor resources occurring during lexical access. More recently, Tomasino et al. (2008) reported that TMS applied at both long and short intervals following action word presentation facilitated motor responses when participants were instructed to imagine themselves performing the action, although not for simple reading or frequency judgement conditions. Neuroimaging studies have provided relatively consistent evidence that retrieval of action word meanings activates cortical areas within the vicinity of the left precentral gyrus (e.g., Canessa et al., 2008; Rüschemeyer et al., 2007; Vigliocco et al., 2006; see Pulvermüller, 2005 for an overview of earlier work). A more precise localisation to actual motor areas has been suggested by several recent studies that attempted to link action word meaning representations (e.g., punch, stomp or bite) associated with specific effectors (i.e., hand, foot or mouth) to somatotopically organised motor areas (e.g., Aziz-Zadeh et al., 2006; Hauk et al., 2004; Tettamanti et al., 2005). In order to demonstrate a congruent somatotopic organisation of action meaning representations, two of these functional MRI studies first identified areas involved in action execution (Hauk et al., 2004) and observation (Aziz-Zadeh et al., 2006) within inferior frontal, precentral and middle frontal gyri according to the involvement of their related effectors. These functionally defined motor regions were then interrogated for activity during reading of effector related action words or sentences. These studies reported BOLD signal differences for comparisons between effector action words and a single control condition that were in some cases only indicative of a trend (e.g., effector pairwise comparisons at pb.1; e.g., Aziz-Zadeh et al., 2006) or only significant for two effectors when ROI overlap was the criterion (e.g., Hauk et al., 2004; see Kung et al. (2007) for a systematic critique of ROI overlap analyses). To our knowledge, no neuroimaging study has yet reported significant effects for all three effectors consistent with a complete somatotopic organisation. Unfortunately, macrostructural features such as gyri and sulci tend not to be reliable indicators of cytoarchitectonic borders (Amunts et al., 2007). Premotor (PM) cortex in particular has no macroanatomical landmark to indicate the border between it and the prefrontal cortex anteriorly where more cognitive functions predominate (Geyer, 2003). Neuroimaging and stimulation studies have indicated a ventral to dorsal somatotopic organisation paralleling that of the primary motor cortex (area 4), and like the monkey (e.g., Godschalk et al., 1995), successive overlap of effector representations (Sanes and Schieber, 2001; Schubotz and von Cramon, 2003). However, unlike the lateral surface of monkey PM cortex, the dorsal (PMd) and ventral (PMv) sectors of area 6 in humans have no cytoarchitectonic features that might distinguish them (Picard and Strick, 1996). In addition, within the medial subdivision of PM cortex, the rostrally located presupplementary motor area (pre-sma) shows little evidence of distinct effector related areas, and may instead support more abstract or cognitive functions, while the caudally situated SMA-proper has a clearer rostrocaudal somatotopic organisation (Picard and Strick,1996). Given the issues associated with localising responses accurately to motor areas in neuroimaging studies, we plotted the peak maxima reported by previous investigations in relation to cytoarchitectonic maximum probability maps (MPMs) of human primary and PM cortex derived from microcellular studies of post-mortem brains (i.e., areas 4 and 6; Eickhoff et al., 2006; Geyer, 2003). The MPM technique is robust to areal misclassifications of voxels and provides a sufficient coverage of the cytoarchitectonic volume (Eickhoff et al., 2006). We plotted relatively large 10 mm spheres around the peak maxima from these studies rather than their reported spatial extents as different thresholds or thresholding strategies used across studies result in activations of varying sizes (see Eickhoff et al., 2007). We reasoned that if action meaning representations are indeed organised somatotopically, then one would expect them to be located within the cytoarchitectonic borders of primary motor and PM cortex where motor somatotopy has been demonstrated by stimulation and neuroimaging studies. As can be seen from Fig. 1, this was not the case. In some instances, there was little agreement across studies for peaks reported for a given effector, an issue that we will discuss in more detail later (e.g., Aziz-Zadeh et al., 2006; Hauk et al., 2004; Tettamanti et al., 2005). It is important to note that even if the activity reported by these studies had been located within the cytoarchitectonic MPMs, it would not constitute definitive proof that these areas mediate semantic representations of action words. An alternative interpretation is that this activity may reflect the by-product of imagery of an action (e.g., Jeannerod, 2001), emerging only following identification of the action concept and not being part of the representation of the action word per se (Willems and Hagoort, 2007). For example, Tomasino et al. (2007) reported motor cortical activity when their participants imagined the situation described in an action phrase, but not when they performed a secondary letter detection task designed to prevent imagery. In an attempt to minimise the influence of imagery, studies have typically administered the comprehension task first, followed by action execution (Hauk et al., 2004) or observation tasks (Aziz-Zadeh et al., 2006). A more direct experimental control would be to include a condition involving imageable, concrete words. Comparisons to date have been between action word meanings and abstract word meanings (Aziz-Zadeh et al., 2006; Tettamanti et al., 2005) or visual character strings (Hauk et al., 2004), neither of which elicit comparable imagery. In order to control for motor imagery per se, action words would need to be employed as control stimuli, which would clearly introduce an experimental confound. However, a number of neuroimaging studies have reported that reading concrete words unrelated to motor function evokes activity in left hemisphere motor areas when contrasted with reading of less imageable abstract words (e.g., D'Esposito et al., 1997; Mellet et al., 1998; Pulvermüller and Hauk, 2006), indicating that they might prove adequate for eliciting generic imagery related activity in motor areas. In the present study, we combined functional MRI with cytoarchitectonically defined probabilistic maps of primary and PM cortex to analyse the responses of both movement execution- and observationrelated regions during retrieval of action word meanings associated with specific effectors. The use of both action observation and execution conditions to demonstrate overlapping activity is a necessary requirement for confirming the operation of mirror neurons (see Turella et al., 2008). It is worth noting that the neuroimaging studies cited in support of shared action word meaning comprehension and mirror neuron activity have tended to employ only action observation tasks (e.g., Aziz-Zadeh et al., 2006) or omitted action tasks altogether (e.g., Tettamanti et al., 2005). We also employed a range of conditions controlling for various levels of lexical information, including concrete words of equivalent imageability unrelated to body parts or actions, regular phonology (nonwords) and visual character recognition (hashes). This represents, to our knowledge, the first systematic attempt to test a proposed somatotopic organisation of action word meaning representations with fmri. Materials and methods Participants Eighteen (13 females, 5 males) healthy, right handed, native English speaking volunteers participated in this study. The mean age

3 636 N. Postle et al. / NeuroImage 43 (2008) Fig. 1. Peak maxima from fmri studies reporting overlap of action word meaning representations with left hemisphere effector areas in relation to cytoarchitectonic maximum probability maps (MPMs; Eickhoff et al., 2006; Geyer, 2003) of primary motor (in red) and premotor (in blue) cortices. 10 mm spheres centred on reported maxima are shown overlaid on coronal slices from a single subject's brain in MNI atlas space and colour-coded according to study. was years (SD = 7.21). Informed consent was obtained from all participants and the experimental protocol was approved by the University of Queensland's Medical Research Ethics Committee. Materials Verbal stimuli Action word stimuli were 75 English language effector related verbs comprising 25 each specific to the hand, foot and mouth (see Supplementary material). Following Hauk et al. (2004), both transitive and intransitive verbs were used in order to obtain a sufficient number of stimuli for each effector type. These were presented in infinitive form as single words, consistent with previous studies (e.g., Hauk et al., 2004; Pulvermüller et al. 2005; Rüschemeyer et al., 2007). An additional 25 concrete nouns unrelated to body parts or actions1, 25 nonwords and a series of six hashes (######) were included to control for generic (i.e., non-body part related) meaning, phonological and visual character processing (see Table 1; a full list of stimuli may be requested from the corresponding author). All words were one to two syllables and three to seven letters in length. Stimuli were matched across categories for the number of syllables, number of letters, orthographic and phonological neighbourhood sizes (Davis, 2003), and lexical frequency of the effector related and unrelated words as determined by the CELEX lexical database (Baayen et al., 1995). The effector related and unrelated 1 Although the unrelated words are concrete nouns rather than verbs, studies using both TMS and neuroimaging have demonstrated consistently that the activity elicited in primary and PM cortex during word reading is not modulated by grammatical class (e.g., Bedny and Thompson-Schill, 2006; Gerfo et al., 2008; Oliveri et al., 2004; Vigliocco et al., 2006). words were also matched for imageability (see Table 2). Each effector related word was predominantly associated with its appropriate effector and each unrelated word was not associated with any body part. This was determined by pilot testing of 92 effector related and 33 unrelated words (from which the target words were chosen) by an independent sample of ten native or longstanding English-speakers (7 females, 3 males; M age = 23.90, SD age = 9.30) who indicated what (if any) body part they associated with each word using a similar rating procedure to that of Hauk et al. (2004) with a five point scale (1 = word does not remind me of this body part at all; 5 = word very much reminds me of this body part). These participants also rated each word's imageability on a five point scale (1 = not imageable at all; 5 = very imageable). The nonwords were obtained from the ARC NonWord Database (Rastle et al., 2002) and contained a relatively even mix of letters. Video stimuli Video stimuli consisted of 30 silent movie clips, with 10 each involving the hand, foot or mouth performing simple actions repeatedly for 5 s. The frame was cropped such that only the relevant Table 1 Examples of word and nonword stimuli used in the lexical task Hand words Leg words Mouth words Unrelated words Nonwords Pick Draw Hold Grip Stir Kick Step Limp Wade Hike Chew Suck Kiss Sing Grin Moon Lake Hill Wall Rain Vuxt Oits Bume Tron Biln

4 N. Postle et al. / NeuroImage 43 (2008) Table 2 Matching variables for the lexical stimuli used in the study Matching variables Category Letters Syllables Frequency a Orthographic neighbours Phonological neighbours Imageability Effector association Mouth 4.76 (1.2) 1.2 (.41) (122.82) 6.0 (4.42) 16.2 (8.845) 4.1 (.58) 4.6 (.35) Hand 4.56 (1.08) 1.04 (.2) (134.11) 6.0 (5.95) 15.4 (9.08) 4.11 (.47) 4.48 (.29) Foot 4.6 (1.12) 1.2 (.41) (155.86) 6.0 (4.87) (8.3) 3.94 (.67) 4.6 (.27) Unrelated 4.6 (1.19) 1.28 (.46) (161.84) 7.08 (5.19) 16.2 (10.42) 3.79 (.38) 0.0 (.0) Nonwords 4.6 (1.19) 1.16 (.37) NA NA NA NA NA Data are means with standard deviations in parentheses. NA: Not applicable. a CELEX lexical frequency per million. body part was visible. There were also 10 control videos that depicted a variety of frequently encountered natural and man-made stimuli moving as they would in their natural environments for 5 s. No humans or animals were depicted in these control videos. Behavioural tasks Lexical task Participants completed the lexical task first to minimise the involvement of motor imagery, followed by the video-based movement observation and execution localisation task in separate imaging runs. In the lexical task, stimuli were presented in blocks of five items from each of the six categories (hand, foot, mouth, and unrelated words, nonwords and hashes) in pseudorandom order. A different pseudorandom ordering of conditions was employed across participants. Each item was presented for 3 s such that each block lasted for 15 s. After the presentation of six blocks (one for each category), a fixation cross was presented for 15 s. This sequence of six blocks followed by fixation was repeated until all words had been presented (i.e. 5 times). Participants silently read the words and when the fixation cross or hashes were presented, watched the stimuli without mentally reciting any words. Action observation task The video stimuli were presented for 10 s followed by either a green or red dot (20 mm diameter) for 10 s, and then a blank screen for 5 s. Participants passively viewed the videos. A green dot always followed action videos and a red dot always followed the control videos. This sequence was repeated until all the 40 videos had been presented. The order of presentation of the videos was randomised. Action execution task The green dot presented following an action video cued participants to replicate the movement of the effector in the video for 10 s (using their right hand/foot or mouth; e.g., clench and release fist, curl and release toes, move tongue from side to side). When the red dot was presented they remained still. Post scanning tasks After scanning was completed, participants completed three postexperiment memory tasks where they were required to describe the video stimuli to the experimenter, discriminate the 100 target words from 20 other words (five hand, foot, mouth and unrelated) and discriminate the 25 target nonwords from 25 other nonwords. These memory tasks were included to ensure all participants attended to the different word meanings and processed all words beyond mere visual perception. Image acquisition Images were acquired using a 4 T Bruker MedSpec system. A transverse electromagnetic (TEM) head coil was used for radiofrequency transmission and reception (Vaughan et al., 2002). T2 weighted gradient echoplanar images (EPI) were used to obtain blood oxygen level dependent (BOLD; Ogawa et al., 1990) images during the lexical and video tasks, with an echo time (TE) of 30 ms. During the lexical task, 176 brain volumes were acquired, each volume consisting of 36 planes, with an in-plane resolution of mm and a slice thickness of 3 mm (.6 mm gap), and a repetition time (TR) of 3 s. During the video task, 405 brain volumes of identical resolution to those of the lexical task were acquired, with a TR of 2.50 s. Geometric distortions in the echo planar images caused by magnetic field inhomogeneities at high-field were corrected on the scanner console using a point-spread mapping approach (Zaitsev et al., 2003). The first five volumes were ignored so tissue magnetization could reach a steady state. Foam padding was used to limit head movement. A highresolution MP-RAGE 3D T1 image was acquired within the same session, with an inversion recovery time (TI) of 700 ms, TR of 1500 ms, echo TE of 3.35 ms and a resolution of.73 mm 3. Image analyses The functional images from each participant were realigned using rigid body motion correction and a mean image generated (Freire et al. 2002). This mean image was coregistered with the subject's MP-RAGE 3D T1 image, and the latter normalised to the standard T1 template image in MNI atlas space using statistical parametric mapping software (SPM5; Wellcome Department of Imaging Neuroscience, Queen Square, London, UK). The resulting transformation to atlas space was next applied to the functional timeseries from which the mean had been generated, and the normalised volumes subsequently resampled to 3 mm 3 voxels and smoothed with an isotropic Gaussian kernel (full width half maximum=8 mm). Global signal effects were estimated and removed using a voxel-level linear model (Macey et al., 2004). A conservative decision was made to exclude the imaging data from one participant whose T1-weighted structural image required additional affine transformations to atlas space. The analyses were conducted on the data from the remaining 17 participants. Two sets of fixed effects analyses were conducted for each participant's data within SPM5. The first involved regressors for the four observation and three execution conditions of the video-based task, while the second comprised regressors for the six categories of stimuli in the lexical task. Blocks of stimuli were modelled with a synthetic hemodynamic response function (HRF) according to their respective durations. Serial correlations were modelled in the context of the AR(1) model. Low frequency fluctuations were removed with a 128 s high pass filter. The main contrasts of interest for the videobased task involved comparing the movement observation and execution conditions with the control observation condition and fixation/baseline, respectively. At the second level group random effects analysis, linear contrasts of the parameter estimates were subjected to one-sample t-tests. As we were primarily concerned with detecting somatotopic activation within the boundaries of the left hemisphere primary motor and premotor cortices, search volumes for these areas were constructed from the maximum probability maps of area 4 (4a and 4p inclusive)

5 638 N. Postle et al. / NeuroImage 43 (2008) and area 6 of Eickhoff et al. (2006), respectively (Geyer, 2003). Observation and execution related activation surviving a height threshold of pb.001 and cluster threshold of N25 contiguous voxels within these search volumes was then used to derive cluster-based regions of interest (ROIs) for interrogating the activation elicited in the lexical task. In order to avoid overlapping representations among effectors in either the observation or execution conditions, contrasts for each effector were exclusively masked (at pb.05, uncorrected) with contrasts involving the other two effectors (see Tettamanti et al for a similar approach). For example, the mouth Ncontrol contrast was exclusively masked with both the handncontrol and footncontrol contrasts. The mean percent BOLD signals for each of the six lexical conditions were then extracted from each of the 12 action cluster ROIs (i.e., 2 motor areas 3 effectors 2 observation/execution conditions) using Marsbar software (v0.41; Activations were visualised using MRICron software ( We also performed a similar analysis to that of Tettamanti et al. (2005) for the lexical task data alone using a height threshold of pb.001 and cluster threshold of N25 contiguous voxels within areas 4 and 6. Contrasts for each effector related action word condition were exclusively masked (at pb.05, uncorrected) with contrasts involving the other two effector action word conditions. For example, the mouth wordsnunrelated words contrast was exclusively masked with both the hand wordsnunrelated words and foot wordsnunrelated words contrasts. Finally, we conducted all analyses without masking to identify observation and execution related activation across the language dominant left hemisphere, and contrasted activation in the lexical conditions with visual character strings, following Hauk et al. (2004), using identical height and cluster thresholds. Results Cytoarchitectonically and functionally defined motor areas associated with action execution and observation of specific effectors The analyses of the action execution and observation conditions revealed BOLD signal responses associated selectively with each effector and organised in the expected ventral-to-dorsal manner (mouth ventrally, followed by hand and then foot dorsally) for the cytoarchitectonically defined areas 4 and 6. In addition, for both execution and observation conditions, the activation associated with each effector showed an expected posterior-to-anterior continuity across the lateral surfaces of both areas 4 and 6. Peak maxima for each effector according to execution and observation conditions are provided in Table 3. There is good agreement between peaks detected for action execution and observation according to each effector within areas 4 and 6, consistent with the proposed operation of mirror neurons. Next, we examined mean BOLD signal responses associated with processing of categories of lexical stimuli within the areas identified for each effector in the action execution and observation conditions by Table 3 Peak maxima for execution and observation of effector actions Effector BA4 (primary motor cortex) BA6 (premotor cortex) Execution Observation Execution Observation Mouth 54, 9, 36 (6.18) 54, 9, 36 (4.46) 57, 3, 33 (5.58) 51, 6, 51 (4.35) Hand 39, 30, 60 (7.07) 33, 27, 63 (4.70) 36, 24, 63 (6.74) 36, 27, 66 (4.52) Foot 6, 27, 63 (6.28) 9, 42, 63 (5.44) 3, 12, 60 (6.51) 9, 9, 66 (4.99) Combined 39, 12, 48 (4.51) 33, 33, 51 (4.46) 9, 0, 60 (4.78) 6, 3, 54 (4.97) Z-scores in parentheses. conducting planned comparisons with repeated measures ANOVAs followed by paired t-tests. The use of planned comparisons is justified given the theoretical proposals for congruent effector specific semantic and motor action representations (e.g., Gallese and Lakoff, 2005; Pulvermüller, 2005), and prior comparisons of this type in neuroimaging studies (Aziz-Zadeh et al., 2006; Hauk et al., 2004; Tettamanti et al., 2005). BOLD signal responses for lexical stimuli in cytoarchitectonically and functionally defined mouth, hand and foot action execution areas Only two ROIs showed non-significant trends (pb.1) toward an effect of lexical stimulus category within a cytoarchitectonically defined motor area that showed effector specific activation for action execution. This was the mouth ROI for both area 4 (F[5]=2.02, p=.084) and area 6 (F[5] =1.93, p=.098). Otherwise, no significant effects were found for any ROI interrogated (pn.05 for all). Fig. 2 (top) shows the ROIs interrogated and the corresponding plots of BOLD signal responses according to lexical category. Post hoc contrasts with paired t-tests on responses extracted from mouth area 4 showed relatively increased activation for mouth words versus hashes (t[1,16] =2.47, pb.05, two-tailed) and trends for mouth words having increased activation compared to hand words (t[1,16] =1.98, p=.065) and hand words having reduced activation compared to unrelated words (t [1,16] = 1.87, p=.079). A near identical pattern was revealed within area 6 (mouthnhashes, t[1,16] =2.17, pb.05; mouthnhand, t[1,16] = 2.22, pb.05; handbunrelated, t[1,16] = 1.78, p=.094) (Fig. 2, bottom). All other contrasts were non-significant. BOLD signal responses for lexical stimuli in cytoarchitectonically and functionally defined mouth, hand and foot action observation areas Within the ROIs that showed effector specific activation associated with action observation, we found a significant effect of lexical stimulus category only in the foot observation ROI of area 6 (F[5]= 4.75, pb.001) and a trend toward an effect in the mouth ROI also in area 6 (F[5]=2.27, p=.055). Fig. 3 (area 4 at top and area 6 at bottom) shows the ROIs interrogated and the corresponding plots of BOLD signal responses according to category of lexical stimuli. Within the foot ROI, post hoc contrasts with paired t-tests revealed that foot words had increased BOLD signal relative to hand words (t[1,16] =2.27, pb.05), hashes (t[1,16] =3.59, pb.01), unrelated words (t[1,16] =2.70, pb.05), and nonwords (t[1,16] =3.56, pb.01), although not mouth words (t[1,16] =1.69, p=.11), while mouth words showed a trend toward increasing signal compared to hashes (t[1,16] =2.01, p=.061) and nonwords (t[1,16] =1.90, p=.075). Within the mouth ROI that showed a non-significant trend, post hoc contrasts with paired t-tests revealed that mouth words had increased BOLD signal relative to hashes (t[1,16]=2.39, pb.05). All other contrasts were non-significant. BOLD signal responses for lexical stimuli in cytoarchitectonically and functionally defined action execution and observation areas for combined effectors We next combined the effector execution conditions in order to identify activity common to all effectors. The peak maxima for these regions are summarised in Table 3. The peak response in area 6 was located medially, corresponding to the boundary between the SMAproper and pre-sma according to the meta-analytically derived human motor area template (HMAT; Mayka et al., 2006) after transformation from MNI to Talairach atlas coordinates (using mni2tal conversion; We then used these generic action execution ROIs to interrogate the lexical category stimuli, this time combining the effector related action word conditions. This allowed us to examine whether the general category of action words showed increased activity in motor

6 N. Postle et al. / NeuroImage 43 (2008) Fig. 2. Axial slices depicting activity elicited by execution of mouth (green), hand (red) and foot (blue) actions in primary motor (BA4ap; top) and premotor (BA6; bottom) cortices, and mean percent BOLD signal detected within these regions during presentation of the different categories of lexical stimuli. Error bars indicate SEM. execution areas relative to other categories of lexical stimuli. Repeated measures ANOVAs failed to reveal significant effects of lexical category in area 4 (F[3] =1.83, pn.05). However, a significant effect was found in area 6 (F[3] =6.52, pb.001). Post hoc contrasts with paired t-tests on the responses in area 6 showed action words had increased responses relative to hashes (t[1,16]=3.68, pb.005) and nonwords (t[1,16] =3.27, pb.01), although not to unrelated words (t [1,16]=1.64, pn.05). Fig. 4 (top) shows the ROIs interrogated and plots of BOLD signal responses. A similar analysis was conducted for the effector observation conditions in order to identify activity common to all effectors (Table 2). The peak response in area 6 was located medially in the pre-sma according to the HMAT (Mayka et al., 2006). These generic action observation ROIs were used to interrogate the lexical category stimuli (Fig. 4, bottom). Again, the effector related action word conditions were combined to examine whether the overall category of action words showed increased activity in general motor observation areas relative to other categories of lexical stimuli. Repeated measures ANOVAs failed to reveal significant effects of lexical category in area 4 (F[3]=1.76, pn.05). However, a significant effect was found in area 6 (F [3] =9.17, pb.001). Fig. 4 (top and bottom) shows the ROIs interrogated and plots of BOLD signal responses. Post hoc contrasts with paired

7 640 N. Postle et al. / NeuroImage 43 (2008) Fig. 3. Axial slices depicting activity elicited by observation of mouth (green), hand (red) and foot (blue) actions in primary motor (BA4ap; top) and premotor (BA6; bottom) cortices, and mean percent BOLD signal detected within these regions during presentation of the different categories of lexical stimuli. Error bars indicate SEM. t-tests on the responses in area 6 showed action words had increased responses relative to hashes (t[1,16] =3.88, pb.001), unrelated words (t[1,16] =2.49, pb.05) and nonwords (t[1,16] =3.95, pb.001). BOLD signal responses for effector related action words in cytoarchitectonically defined areas 4 and 6 This analysis failed to reveal any significant activity for the comparisons of the individual effector related action words and unrelated words in either area 4 or 6. We therefore combined the effector related action word conditions to examine whether the overall category of action words showed increased activity relative to unrelated words. This analysis revealed a single cluster of significant activity in the pre-sma (peak x, y, z=0, 12, 57; Z=4.45), confirmed with the HMAT (Mayka et al., 2006). BOLD signal responses for action execution, action observation, and lexical stimuli conditions outside cortical motor areas Bold signal responses elicited across the language dominant left hemisphere by the action execution, action observation, and lexical stimuli conditions are presented in Supplementary material.

8 N. Postle et al. / NeuroImage 43 (2008) Fig. 4. Coronal slices depicting activity elicited by execution (top) and observation (bottom) of combined mouth, hand and foot actions in primary motor (BA4ap; in red) and premotor (BA6; in blue) cortices, and mean percent BOLD signal detected within these regions during presentation of the different categories of lexical stimuli. Error bars indicate SEM. Post scanning test Memory task Of the 25 words within each semantic category, on average, each participant correctly recognised (SD=4.00) of the hand-related, (SD=4.44) of the foot-related, (SD=4.68) of the mouthrelated, (SD= 5.66) of the body part-unrelated and (SD=4.77) of the nonwords. Of the 10 videos within each category, participants correctly recalled approximately 5.28 (SD=1.78) of the hand-related, 5.33 (SD=1.57) of the foot-related, 6.39 (SD=2.38) of the mouth-related and 6.22 (SD=1.93) of the control videos. ANOVAs revealed that no one type of video stimuli was recalled substantially more often than any other (pn.05), and the only significant difference among lexical stimuli was that nonwords were recognised less often than other types of target words (pb.05). Discussion This study is the first to assess whether action meaning representations in the left hemisphere primary and premotor cortices show a congruent somatotopic organisation in relation to both the execution and observation of actions of specific effectors. In addition,

9 642 N. Postle et al. / NeuroImage 43 (2008) it is the first to define these motor areas for interrogation according to a combination of cytoarchitectonic and functional information, and to include a lexical condition controlling for imagery that might be elicited subsequent to the comprehension of word meanings. The use of cytoarchitectonic maps ensured that we interrogated responses solely within objectively defined primary and PM cortical areas, excluding responses in areas of the surrounding frontal cortex responsible for more cognitive functions. While not supportive of a somatotopic mapping of action meaning representations, our results are in agreement with findings from TMS and neuroimaging studies of more general action word processing activating cortical areas within the vicinity of the left precentral gyrus adjacent to the motor areas (e.g., Canessa et al., 2008; Rüschemeyer et al., 2007; Vigliocco et al., 2006; see Willems and Hagoort, 2007 for a review). Primary motor and premotor cortical areas associated with execution and observation of effector specific actions The execution of simple movements with specific effectors resulted in graded activation along the lateral surfaces of the primary and PM cortices with an expected somatotopic organisation. Mouth representations were located ventrally, followed by hand and foot representations dorsally (Penfield and Welch, 1951; Woolsey et al., 1952). The somatotopic organisation extended in a continuous fashion across the lateral surface of area 4 into the caudal portion of area 6 (comprising PMv and PMd). These results are consistent with a now large neuroimaging literature in humans (e.g., Alkadhi et al., 2002; Ehrsson et al., 2003; Stippich et al., 2007; Picard and Strick, 2001; Schubotz and von Cramon, 2003). The observation of movements made with specific effectors likewise resulted in activity organised in a somatotopic fashion comparable to that associated with execution, with locations of peak responses showing good accordance, thus providing evidence similar to that cited in support of a proposed human mirror neuron system (Buccino et al., 2004; see Turella et al., 2008). Somatotopically organised action word meaning representations We did not detect congruent activation for action words associated with specific effectors in the somatotopically organised primary or PM ROIs identified functionally by either execution or observation of actions by those effectors. Nor were responses in effector specific areas selective to action words, as they also showed responses to varying levels of lexical information, especially unrelated imageable concrete meanings and nonwords with regular phonology. The only effector ROI to show a trend in the predicted direction was the PM foot observation ROI. Of interest, this ROI was located in the pre-sma/sma where we found increased activation for generic action words compared to other classes of lexical stimuli (discussed below). It is possible that some foot words were comprehended as whole body actions (e.g., walk, jump), potentially contributing to this activation. Alternatively, as Pulvermüller et al. (2005) suggested, leg words might have a stronger reliance on semantic action-related features compared to other effector words, as leg actions tend to be more similar to each other (e.g., walk, run and jog involve many of the same movements). However, within this ROI foot action words did not show increased signal compared to mouth words, and the latter words also showed trends (psb.08) for increased signal compared to other lexical stimuli. A critical difference between the design of our study and that of previous studies is the use of cytoarchitectural information. We used probability maps derived from microcellular studies of post-mortem brains, ensuring that we interrogated responses evoked solely in primary and PM cortex (Eickhoff et al., 2006; Geyer, 2003). As we noted in the Introduction, the peak maxima reported by previous studies for overlapping action meaning representations did not accord well with the cytoarchitectonic boundaries of the motor areas (see Fig. 1). Moreover, there was little agreement across studies for peaks reported for a given effector, especially foot-related words. In particular, the peaks from the Hauk et al. (2004) study that used single word stimuli did not accord well with the peaks from the studies that used sentence stimuli (Aziz-Zadeh et al., 2006; Tettamanti et al., 2005). This lack of consistency suggests that the activity reported by previous studies might represent stimuli- or task-dependent factors. In addition, we employed unrelated concrete words of equivalent imageability as control stimuli, as they are known to activate left PM cortex when compared with less imageable, abstract words (e.g., D'Esposito et al., 1997; Mellet et al., 1998; Pulvermüller and Hauk, 2006). Previous studies had compared effector action words with abstract words or visual character strings, leaving open the possibility that the activity they observed was due to differences in imagery elicited between conditions (Aziz-Zadeh et al., 2006; Hauk et al., 2004; Tettamanti et al., 2005). The absence of differential activity in the present study indicates that this is likely to have been the case. We also found that nonwords modulated activity in several of the PM cortex action ROIs. Left PM cortex activity is a common finding during nonword reading, and is often attributed to processes of orthography phonology conversion or covert articulation (see Dietz et al., 2005; Mechelli et al., 2005; Snyder et al., 2007). Our data appear consistent with the latter explanation, as the strongest responses for nonwords were observed in the mouth PM cortex ROIs. The pre-sma and the representation of instructions for action Given the absence of compelling evidence of a congruent motor somatotopy for action word meaning representations, we next identified action execution and observation ROIs within areas 4 and 6 that were common to all effectors, and assessed whether these areas showed a selectivity for action word meaning representations in general. The generic action execution and observation ROIs identified in the PM cortex in this manner were both sensitive to the category of lexical stimuli presented (action wordsnunrelated wordsnnonwordsn hashes), with the observation-related ROI showing the clearest dissociation. The peaks of these ROIs were located within the medial wall of the frontal cortex and, when converted to Talairach coordinates, were either on or anterior to a perpendicular line crossing the ventral anterior commissure used to delineate the pre-sma rostrally and SMAproper caudally (the VCA line; Picard and Strick, 1996). The primary motor cortex did not show a similar pattern of responses. The pre-sma is now viewed as having a cognitive motor rather than strictly motor role, due to its representations being relatively effector-independent and its not being connected directly to the primary motor cortex, unlike SMA-proper (Picard and Strick, 1996, 2001). It also has connections with prefrontal cortex, again differing from SMA-proper (Johansen-Berg et al., 2004). As such, our observation of selectively increased activity in this region for the class of action words does not appear consistent with explanations emphasising either Hebbian learning (e.g., Pulvermüller, 1996, 2005) or mirror neuron related mechanisms (e.g., Rizzolatti and Craighero, 2004; Tettamanti et al., 2005) framed in terms of actual motor programs. In their meta-analysis, Grèzes and Decety (2001) noted that the pre-sma was activated primarily during action observation tasks with instructions to imitate. We used this same type of task in the present study. In the monkey, some pre-sma neurons represent relatively specific instructions for action (Hoshi and Tanji, 2004). As the imperative form of a verb serves as an instruction (e.g., sit! ), the activation we observed in the pre-sma might therefore reflect a shared representation for an instruction cue. According to this account, verbs serve as instruction cues by enabling the retrieval of an appropriate motor program, i.e, they represent information required for motor planning. An instruction to imitate similarly requires subjects to retrieve an appropriate motor program from visual information. They then execute the program after a delay. If this account is accurate, one would expect greater activation

10 N. Postle et al. / NeuroImage 43 (2008) associated with action word comprehension in the observation rather than execution ROI, as we found. Given the activation we observed was common to all effectors, such a representation would probably be relatively abstract (e.g., a general instruction to perform an action). As such, it might be considered compatible with a more general notion of embodied cognition, i.e., one that does not assume a direct coupling between action meaning comprehension and cortical motor areas (e.g., Wilson, 2002). In addition, it does not preclude the possibility of a modality-specific representation of action word meaning elsewhere in association cortices (e.g., Hoenig et al., in press). Indeed, our wholebrain analyses indicated that the majority of the BOLD responses elicited by retrieving action word meanings occurred in association cortex adjacent to the motor areas (see Supplementary material). Methodological considerations The maps of somatotopic motor activation reported here were elicited by execution and observation of simple, intransitive movements. Intransitive movements have been used frequently to demonstrate somatotopy in terms of motor execution, observation and imagery (e.g., Alkadhi et al., 2002; Ehrsson et al., 2003; Stippich et al., 2007). While our results accord well with this previous research, it is worth noting that the motor somatotopy reported for observation and imagery of object-related movements does not always match that reported for intransitive movements (e.g., Buccino et al., 2001; Chouinard and Paus, 2006). As Pinker and Jackendoff (2005) point out, activity during observation of object-related movements might be more specific to the semantic goal of the action (i.e., manipulating the object). Much of the research concerning observation of object-related hand movements in humans has revealed activity within frontal cortex along an assumed macrostructural border between PMv and area 44/45 (Broca's area; see Rizzolatti and Craighero, 2004). Additionally, the maps of motor somatotopic activation were derived conservatively using an exclusive masking technique. That is, pixels mutually activated by different effectors were excluded from the ROIs. Previous fmri studies investigating motor somatotopy have employed similar statistical techniques to minimise the influence of overlapping representations (e.g., Alkadhi et al., 2002; Ehrsson et al., 2003; Stippich et al., 2007). Like Tettamanti et al. (2005), we used the exclusive masking technique to identify activation associated with effector specific action word meaning representations. One major limitation of the present study and previous investigations that presented English language verbs in isolation (e.g., Hauk et al., 2004; Pulvermüller et al., 2005) is that there is considerable verb noun homonymy in English (e.g., the word walk can denote both a noun the walk and verb to walk ). Therefore we cannot be certain that the action word stimuli were read as verbs rather than nouns. The implications of this are not entirely clear, as neuroimaging and TMS studies have shown that while primary and PM cortex respond to action word meanings, they are insensitive to manipulations of grammatical class involving these words (e.g., Oliveri et al., 2004; Vigliocco et al., 2006). Ambiguity is also inherent in the earlier studies that used abstract sentence stimuli (Aziz-Zadeh et al., 2006; Tettamanti et al., 2005). As Rüschemeyer et al. (2007) noted, these studies confounded sentential objects with concrete or abstract meanings according to the verb used (e.g., in the abstract sentence He grasped the idea, idea is an abstract noun). Hence, the activation observed could have been due to the action word meaning or object noun meaning, or a combination of both. Our choice of a block rather than an event related design for presenting the lexical stimuli was based upon the superior efficiency of the former in detecting BOLD responses (Bandettini and Cox, 2000; Friston et al., 1999). Direct comparisons of block and event related designs in word reading experiments have likewise shown the expected advantage in terms of response amplitude for the former (e.g., Chee et al., 2003). Recent neuroimaging studies of action word meaning representation have likewise employed block designs (e.g., Vigliocco et al., 2006; single words), some with far fewer stimuli (e.g., only 5 sentences repeated eight times per condition; Aziz-Zadeh et al., 2006; Tettamanti et al., 2005). Nevertheless, the use of a block design could have been suboptimal for detecting early processes associated with accessing word meaning. In order to address this issue, we reanalysed the fmri data as an event related design following Mechelli et al. (2003). The results were essentially the same, although with the expected lower amplitude. However, we acknowledge that techniques with superior temporal resolution to fmri, such as event related potentials (ERPs), are likely to be more sensitive to early processes associated with accessing word meaning, particularly when used in combination with fmri (e.g., Hoenig et al., in press). Finally, as we used fewer words per condition than Hauk et al. (2004), there is a possibility that our results reflect a reduction in power relative to their study. However, we think this explanation unlikely as other fmri studies have used either fewer (23; e.g., Rüschemeyer et al., 2007) or the same number of words per condition to successfully image the neural correlates of single word reading (25; e.g., Cohen et al., 2002). It is also worth noting that the larger lists employed by Hauk et al. (2004) contained words rated by their participants as belonging to more than one category (see their Fig. 1b, p. 302). Summary and conclusions Although a number of recent theories of action word meaning representation have assumed the involvement of a distributed network of cortical somatosensory and motor areas, only a few have proposed a tight coupling with motor somatotopy (e.g., Gallese and Lakoff, 2005; Pulvermüller, 2005). There is considerable evidence from TMS and neuroimaging studies of more general action word processing activating association cortex within the vicinity of the left precentral gyrus, and our results can be considered in agreement with these findings (e.g., Canessa et al., 2008; Rüschemeyer et al., 2007; Vigliocco et al., 2006; see Willems and Hagoort, 2007 for a review). As others have noted, a finding of motor cortex activity during action word comprehension does not prove definitively that semantic representations are being accessed in these areas (Willems and Hagoort, 2007). Even so, only this and a handful of other neuroimaging studies have explicitly examined the issue of a congruent motor somatotopy of action meaning representations and, as we have argued above, the evidence from these studies should be considered in relation to a combination of cytoarchitectonic and functional criteria. These criteria should include controlling for a range of potentially confounding lexical variables, and in the case of a proposed mirror neuron mechanism, include both action execution and observation conditions to demonstrate overlapping activity (Turella et al., 2008). In addition, a demonstration that lesions to discrete motor areas have deleterious effects on the understanding of action words is required in order to support the theory (Willems and Hagoort, 2007), and to date the available lesion evidence is not supportive. Appendix A. Supplementary data Supplementary data associated with this article can be found, in the online version, at doi: /j.neuroimage References Alkadhi, H., Crelier, G.R., Hotz Boendermaker, S., Hepp-Reymond, M.C., Kollias, S.S., Reproducibility of primary motor cortex somatotopy under controlled conditions. Am. J. Neuroradiol. 23, Amunts, K., Schleicher, A., Zilles, K., Cytoarchitecture of the cerebral cortex more than localization. NeuroImage 37, Aziz-Zadeh, L., Wilson, S.M., Rizzolatti, G., Iacoboni, M., Congruent embodied representations for visually presented actions and linguistic phrases describing actions. Curr. Biol. 16, Baayen, R.H., Piepenbrock, R., Gulikers, L., The CELEX lexical database [CD-ROM]. Linguistic Data Consortium, University of Pennsylvania, Philadelphia.