Repetitive transcranial magnetic stimulation reveals a role for the left inferior parietal lobule in matching observed kinematics during imitation

Apraxia (a disorder of complex movement) suggests that the left inferior parietal lobule (IPL) plays a role in kinematic or spatial aspects of imitation, which may be particularly important for meaningless (i.e. unfamiliar intransitive) actions. Mirror neuron theories indicate that the IPL is part of a frontoparietal system that can support imitation by linking observed and stored actions through visuomotor matching, and have less to say about different subregions of the left IPL, or how different types of action (i.e. meaningful or meaningless) are processed for imitation. We used repetitive transcranial magnetic stimulation (rTMS) to bridge this gap and better understand the roles of the left supramarginal gyrus (SMG) and left angular gyrus (AG) in imitation. We also examined whether these areas are differentially involved in meaningful and meaningless action imitation. We applied rTMS over the left SMG, over the left AG or during a no‐rTMS baseline condition, and then asked participants to imitate a confederate's actions whilst the arm and hand movements of both individuals were motion‐tracked. rTMS over both the left SMG and the left AG reduced the velocity of participants’ finger movements relative to the actor during imitation of finger gestures, regardless of action meaning. Our results support recent claims in apraxia and confirm a role for the left IPL in kinematic processing during gesture imitation, regardless of action meaning.


Introduction
Neuroscientific research implicates the left parietal lobe in imitation (Molenberghs et al., 2009;Caspers et al., 2010), but neither the precise area nor its exact role is fully established. Some neuropsychologists studying the apraxiasdisorders of complex movement in which different types of imitation can be impairedsuggest that the left parietal lobe controls kinematic (Buxbaum et al., 2014) or spatial (Goldenberg, 2009) aspects of imitation and that the inferior parietal lobule (IPL) in particular is critical for imitating meaningless actions. Theories of social interaction based on the putative human mirror neuron system, by contrast, indicate that the IPL is part of a frontoparietal system that links observed and stored actions through visuomotor matching (Rizzolatti et al., 2014), in terms of movements (Iacoboni & Dapretto, 2006;Iacoboni, 2009) or goals (Hamilton, 2008(Hamilton, , 2014. In this way, the IPL could provide a scaffolding for imitation. These theories, however, often say less about the specific role of IPL subregions [supramarginal gyrus (SMG); angular gyrus (AG)] or how the left IPL might or might not be differentially involved in meaningful and meaningless action imitation.
Furthermore, neuroscience-driven work on imitation in healthy individuals is often defined by the experimental scanning environment. Experimental paradigms typically use a single participant responding to pre-recorded stimuli in the confines of an fMRI scanner, which tends to limit the imitative capacity afforded by viewing and acting with the entire arm and hand. We wanted to better understand imitation as a dynamic social experience, an approach that may be essential to understand realistic imitation behaviour (Reader & Holmes, 2016). In addition, in research with both healthy individuals and apraxia patients, there is little use of motion-tracking for characterising imitation in an objective fashion (with some exceptions, e.g. Braadbaart et al., 2012;Campione & Gentilucci, 2011;Gold et al., 2008;Hayes et al., 2016;Hermsd€ orfer et al., 1996;Kr€ uger et al., 2014;Pan & Hamilton, 2015;Reader & Holmes, 2015;Sacheli et al., 2012Sacheli et al., , 2013Sacheli et al., , 2015aWild et al., 2010;Williams et al., 2013), despite the fact that kinematics are an important element of social interactions (Krishnan-Barman et al., 2017). With this in mind, we used a two-person motion-tracking approach in this experiment to better understand the links between actor and imitator behaviour.
To bring work on imitation in neuropsychology and healthy participants together, we used non-invasive brain stimulation to arbitrate between the roles of the two broad left IPL subregions SMG and AG in imitation and, secondary to this, to test whether these regions are differentially involved in meaningful and meaningless action imitation. In particular, we were interested in examining both imitation accuracy and kinematics in an exploratory fashion. We applied repetitive transcranial magnetic stimulation (rTMS) to the left SMG, left AG, or during a no-rTMS baseline, and then asked participants to imitate a confederate's actions in a two-person, ecologically valid and naturalistic motion-tracking paradigm.

Participants
We recruited 12 participants from the University of Reading and the surrounding area (mean AE SE age = 23.2 AE 1.1 years, five males, two left-handed). Left-handed participants were not excluded as, in the SMG at least, praxis representation is not related to handedness (Kr oliczak et al., 2016). The experiment was approved by the University of Reading ethics committee (ref: UREC 15/49); participants gave written, informed consent; the experiments were conducted in accordance with the Declaration of Helsinki (as of 2008).

Stimuli and apparatus
Positions of the participant's right arm and hand and a confederate's left arm and hand were recorded using a Polhemus Liberty motiontracking system (Polhemus Inc., Colchester, VT, USA) recording 16 channels (eight per person) with six degrees of freedom (x, y, z, azimuth, elevation and roll) at 240 Hz. Trackers were attached to the shoulder (acromial end of the clavicle), elbow (olecranon), wrist (pisiform) and the thumb and finger tips, using adhesive medical tape or Velcro TM . rTMS was applied using a PowerMAG 100 (Mag & More GmbH, Munich, Germany) with a 70-mm figure-of-eight coil.
The experiment was controlled and data acquired using custom software written in MATLAB 2014b (MathWorks, Inc.) and using the ProkLiberty interface (https://code.google.com/p/prok-liberty/). We used LabMan and the HandLabToolbox to document and control experiments and analyse data. The associated repositories are, or will be, freely available at https://github.com/TheHandLaboratory, whilst raw data are available on request.
Thirty gestures were used as stimuli. This included five meaningful hand, five meaningful finger and 20 matched meaningless gestures. For each meaningful gesture, two matched meaningless gestures were created. In the case of finger gestures, this was performed by changing the fingers used and/or the orientation of the hand. In the case of hand gestures, this was performed by changing the orientation or position of the hand. We used more meaningless than meaningful gestures to reduce the number of times participants were exposed to these actions, reducing the likelihood that they would associate them with a particular meaning. The finger gestures signified 'okay', 'thumbs up', 'shoot', 'peace' and 'silence'. The hand gestures signified 'salute', 'stop', 'shock', 'looking into the distance' and 'I'm listening' (Fig. 1A). Emblematic gestures were used as, unlike pantomimed actions which imply an object and require continuous motion, emblematic gestures are point to point which is more appropriate for motiontracking (i.e. it is straightforward to extract a single velocity curve for analysis).
During the imitation task, participants sat opposite a confederate at a round plastic table (diameter = 85 cm), approximately 100 cm apart (Fig. 1B). A Blu Tackâ start point was located 20 cm away from each person's abdomen. To inform the confederate of the actions they needed to perform, a computer screen (unobservable by the participant imitator) was placed parallel to the table, approximately 50 cm to the left of the actor.

TMS site localisation
Visualisation of the participant's brain used T1 weighted MR images alongside Brainsight 2.2.13 (Rogue Research Inc., Montreal, Canada). Due to SMG and AG size, stimulation locations were based on guidance from previous experimental activation and cytoarchitectonic maps (Caspers et al., 2006(Caspers et al., , 2008. The stimulation site for left SMG was area PF, located by finding the dorsal extension of the posterior end of the Sylvian fissure and the anterior end of the intraparietal sulcus, drawing an imaginary line between them and stimulating the centre of this line. Evidence suggests that area PF usually falls within these limits (Caspers et al., 2006). As AG activation in neuroimaging studies of imitation appears to be less frequent than SMG activation, the stimulation site was the centre of the left AG, aiming to cover both PGa (anterior) and PGp (posterior). The AG site was located half way between the dorsal extension of the posterior superior temporal sulcus and the intraparietal sulcus. Mean AE 95% CI locations are shown in Fig. 1C. For both AG and SMG, the coil was oriented orthogonal to the main orientation of the gyrus limits. The location of the coil in the no-rTMS baseline condition was placed directly between the AG and SMG positions, but held parallel against the head, such that no or minimal stimulation of the brain should occur.

rTMS parameters
Monophasic rTMS was applied to the left SMG and left AG, and in the no-rTMS baseline condition (coil held away from the head) at 1 Hz and 110% of distance adjusted resting motor threshold (RMT, Stokes et al., 2007). RMT (Rossini et al., 1994) was obtained at the start of the first session. Mean AE SE RMT was 69 AE 4.1% of maximum stimulator output (MSO). The distance from M1, SMG, and AG to the scalp was measured using Brainsight. The no-rTMS site distance was measured from the cortical tissue directly underlying the no-rTMS site to the scalp. Stimulation intensity was limited to a maximum of 85% MSO in order to prevent the coil overheating. Mean AE SE stimulation intensity in each condition was as follows: SMG = 70 AE 4.1%, AG = 72 AE 4.4%, and no-rTMS = 71 AE 3.9% MSO.

Design and procedure
Participants took part in three sessions split at least a week apart. On a single day, rTMS was applied twice (once for meaningless and once for meaningful) for 15 min (900 pulses at 1 Hz) either to the left SMG, left AG, or in the no-rTMS baseline condition, in counterbalanced order across participants. After each rTMS application, participants took part in either a meaningful or meaningless action imitation task. Meaningless and meaningful actions were segregated into their own separate blocks (each following a single rTMS application), as there is evidence to suggest that performing novel and known actions in a sequence could recruit a single processing route, whilst presenting them separately recruits separate routes (Tessari & Rumiati, 2004; but see Press & Heyes, 2008;A.T. Reader, V.M. Rao, A. Christakou & N.P. Holmes, unpublished data). Task order was counterbalanced across stimulation sites. Imitators were not given detailed instructions regarding the task constraints and were simply told to imitate the confederate, with the aim that this would ensure naturalistic performance in the task.
Both confederate and imitator began with their thumb and forefinger gripping their start points. In both meaningful and meaningless imitation tasks, action images appearing on a computer screen (in a random order, one per trial) signalled the confederate to perform the displayed action, which they performed and maintained briefly before returning their hand to the start point. Five seconds after presentation of the image to the confederate, a tone indicated the participant to imitate the action they had observed, which they performed in a mirror fashion (i.e. using their right hand to copy the confederate's left hand). After completing the action, they returned their hand to the start point before the next trial. Following a single application of rTMS, each meaningful action was presented six times, or each meaningless action was presented three times, giving a total of 60 trials per condition and TMS site. Imitation was performed in same-sex dyads, with either a male confederate or one of two female confederates. The same confederate was used as actor for every condition of a given participant.
Following the third rTMS session, participants were presented with a questionnaire featuring the meaningful and meaningless images in pseudorandom order. They were asked to state whether they thought each gesture had a meaning and, if it did, to provide an explanation. This was done with the aim of excluding participants if they failed to meet an arbitrary 60% agreement with our meaningful and meaningless action categorisation, but no participants failed this criterion. Mean AE SE agreement on the meaningful gestures was 75.8 AE 7.83%, and the mean percentage of meaningless gestures described as meaningful was 5.83 AE 1.83%.

Data analysis
An automated script was used for pre-processing and extraction of variables. The analysis routines processed the position data from each trial of each participant and rejected artefacts (e.g. trials with missing samples or spikes resulting from electromagnetic interference) before filtering with a bidirectional low-pass fourth-order Butterworth filter (cut-off frequency, 15 Hz). 5.3% of trials were removed due to incorrect start times or artefacts. The imitator's and actor's 3D velocity over their primary movement (movement onset to gesture completion, mean AE SE duration = 1021 AE 34.3 ms) were resampled to 240 samples and then correlated to provide a measure of imitation accuracy (Reader & Holmes, 2015). The 3D velocity profiles for each of the imitator's trackers were correlated with each of the actor's corresponding trackers: shoulder, elbow, wrist, thumb, index finger, middle finger, ring finger and little finger. 3D velocity profiles were used as they provide a measure of the change in the 3D position of the trackers over time and, thus, the formation of the final hand posture over time. This was considered preferable to using the x, y and z position values as it allowed us to reduce the number of statistical comparisons and the likelihood of false positives.
To use parametric statistics on the resulting r-values, they were converted into Z-values using Fisher's transformation (Z = 0.5*ln ((1 + r)/(1Àr))), where ln is the natural logarithm. JASP (version 0.8.0.0, JASP Team) was used to perform two-way repeated measures ANOVAs on the means of all relevant trials for each of these variables across each crossed condition (Table 1). Preliminary analysis indicated that hand gestures were biasing the results (i.e. the mean Z-values for all digits were similar as the digits generally moved together). Because of this, we split the hand and finger gestures before examining accuracy and then corrected for multiple comparisons using Bonferroni correction, reducing our alpha used to determine a statistically significant result to 0.025. We then performed the following analysis on the hand and finger data separately.
To assess whether there were time-dependent significant differences in the stimulation-site related main effects or interaction, t-statistic plots were created to examine changes in the imitator (i.e. regardless of the actor) 3D velocity for each instance (Fig. 2). We took this time-series-driven approach to inform us of possible differences in peak kinematic values in separate trackers, without the inflated type 1 error that would occur were we to examine multiple kinematic parameters in multiple trackers. In cases where the t-value was at a significant level for any sequence of samples in the time series, we performed permutation testing on the relevant data.
Permutation testing was performed over 10 000 iterations to create a custom empirical null distribution of the length of samples with significant t-statistics, which was then used to decide whether an observed sequence was significantly long. This is similar to the use of cluster-based statistics in fMRI, where a fixed, arbitrary threshold is used for creating clusters, and then, a second threshold is calculated for determining how large a cluster needs to be before it is statistically significant. On each iteration, the condition labels for each participant's data were pseudorandomised, and the original analyses were then repeated exactly, to obtain t-statistics, and sequences of significant t-statistics for the difference between 'SMG' and 'AG' conditions, under the null hypothesis. From this, we were able to assign a pvalue to our actual results by seeing what proportion of the tail of the distribution was greater (or lesser) than or equal to the actual result. We examined the minimum length of sequences of continuous values in which |t| > 2.201 (i.e. statistically significant at a samplewise P < 0.05) and also the P-values associated with the sequences of timepoints in our recorded data where |t| > 2.201.
Where significantly long sequences were found, we examined the standard kinematic parameters that occurred during those periods to confirm whether the differences were derived from the SMG-AG comparison or whether there was further information to be gained from the no-rTMS baseline. Imitator peak kinematic parameters were examined using one-tailed post hoc paired t-tests. To check that any differences were derived from imitator rather than actor performance, we ran the same analysis on the actor peak values using two-tailed paired t-tests. For all post hoc paired t-tests, Hedges' g rm was chosen to report effect sizes in repeated measures comparisons (Lakens, 2013).

Results
No significant effects of stimulation site nor interaction between stimulation site and meaning were observed in imitation accuracy (Table 1). In hand gestures, the shoulder and elbow positions were significantly more correlated for meaningless than for meaningful actions. This effect was also significant in the same direction for the shoulder in finger gestures and, only when uncorrected for multiple comparisons, in the elbow. Figure 2A shows the t-statistic plots for the main effect of SMG vs. AG in hand gestures. Figure 2B shows the t-statistic plots for the interaction between stimulation site and meaning in hand gestures. Figure 2D shows the t-statistic plots for the interaction between stimulation site and meaning in finger gestures. No significantly long sequences were observed in these data. Figure 2C shows the t-statistic plots for SMG vs. AG in meaningful and meaningless actions for finger gestures. Permutation testing for the thumb revealed a significant sequence between 59 and 108 samples (P = 0.035). The index finger showed a significant sequence between 65 and 119 samples (P = 0.022). The ring finger showed a significant sequence between 72 and 116 samples (P = 0.041). The little finger showed a significant sequence between 67 and 114 samples (P = 0.036). The middle finger sequence between 70 and 106 samples was not significantly long (P = 0.054).
The mean peak velocity (PV) for the digits in finger gestures was examined post hoc as the above significant sequences overlapped with this kinematic parameter. Figure 3 emphasises this difference in the original data for the thumb. One-tailed t-tests for imitator mean digit PV (Fig. 4) indicated a Bonferroni-corrected significant (P < 0.025) difference, where stimulation over AG resulted in a greater mean digit PV than over SMG (t(11) = 2.23, P = 0.024, g rm = 0.207), and a similar difference where stimulation over AG resulted in a greater mean digit PV than the no-rTMS baseline (t(11) = 2.10, P = 0.030, g rm = 0.182). There was no significant difference in mean digit PV between stimulation over SMG and the no-rTMS baseline (t(11) = À0.503, P = 0.303, g rm = 0.0465).
We then used two-tailed t-tests to perform the same analysis on the PV of the actor in their finger gestures (Fig. 4), which revealed a Bonferroni-corrected significant (P < 0.025) difference between mean digit PV in the AG condition and no-rTMS baseline (t(11) = 2.91, P = 0.014, g rm = 0.529). There was no significant difference in mean digit PV following stimulation over SMG and the no-rTMS baseline (t(11) = 1.89, P = 0.086, g rm = 0.331) or between SMG and AG conditions (t(11) = À1.71, P = 0.115, g rm = 0.222). This suggested that actor behaviour was biased, so we also decided post hoc to examine the imitator PV relative to the actor PV to try and control for the effects of this bias.
We examined the imitator mean digit PV relative to the actor mean digit PV in finger gestures using two-tailed t-tests and a Bonferronicorrected alpha criterion of 0.025 (Fig. 5). There were no significant differences using this corrected criterion. However, mean digit relative PV was reduced following SMG stimulation compared to the no-rTMS baseline (t(11) = À2.37, P = 0.037, g rm = 0.335) and also following AG stimulation compared to the no-rTMS baseline (t(11) = À2.31, P = 0.041, g rm = 0.281). There was also no significant difference in mean digit relative PV between SMG and AG stimulation (t(11) = À0.316, P = 0.758, g rm = 0.0424).

Discussion
We tested participants' ability to imitate meaningful and meaningless hand and finger gestures following rTMS over the left SMG, left AG, or after a no-rTMS baseline. Whilst there were no differences in imitation accuracy between stimulation sites, we observed that participants' digit PV was lower relative to the actor in finger gestures following left SMG or left AG stimulation, though with a larger effect size in the SMG condition. These results provide the first causal evidence, using brain stimulation in healthy individuals, for a role of the left IPL in processing the kinematics of finger movements during gesture imitation.
Whilst stimulation did not influence imitation accuracy, there was some evidence for differences between accuracy in meaningful and meaningless action performance. Interestingly, participants matched the confederate's shoulder and elbow movements to a significantly greater degree in meaningless actions. Meaningless actions may rely more on matching action kinematics (e.g. Rumiati & Tessari, 2002;Tessari & Rumiati, 2004;Wild et al., 2010) than meaningful actions. The fact that effects were only observed in the shoulder and elbow may reflect the fact that the differences in accuracy were easier to detect in proximal effectors with lower degrees of freedom.
Examining t-statistics over time, we found significantly long periods during which imitator finger velocity was significantly lower following rTMS over left SMG compared to left AG during imitation of finger gestures, but regardless of action meaning. This effect was also reflected in mean digit PV. To account for possible differences in actor behaviour, we examined imitator mean digit PV relative to the actor mean digit PV and found that participants  showed significantly reduced mean digit PV relative to the actor following SMG and AG stimulation compared to baseline. This result seems to indicate that during the imitation of finger gestures, rTMS to the left SMG or left AG reduces digit velocity relative to the observed actor. Further study is necessary to understand the underlying processes altered in this experiment, but these results have the potential to bring together both findings from apraxia and discussions of the putative mirror neuron system in healthy individuals.
As noted in the introduction, some suggest that the putative human mirror neuron system supports our imitative capacity. The IPL is one area of this proposed frontoparietal system  that has been suggested to support imitation through visuomotor matching of seen actions and those that are already in the motor repertoire. This theory provides one explanation for our result in the absence of any stimulation-related effects specific to action meaning. That is, disturbed visuomotor matching following stimulation over left IPL could reduce the velocity with which the fingers shape complex postures relative to the actor. However, this effect does not necessarily stem from an rTMS-induced inhibition of direct matching. Rather, by considering claims made in the mirror neuron literature and neuroimaging studies of healthy individuals, recent discussions in apraxia, and recognising the possibility that different areas of the IPL may subserve different aspects of imitation, we may be able to provide a more holistic explanation for the observed data. Buxbaum et al. (2014) found that damage to the left IPL was associated with deficits in kinematic (rather than postural) aspects of movement for novel and tool-related actions. Similar results were found in a more recent voxel-based lesion-symptom mapping study by Dressing et al. (2016). Buxbaum et al. (2014) suggested that the left IPL computes 'movement plans [as] dynamic changes in the relative spatial positions of body parts needed to reach a goal configuration' (p. 13). If this is the case, changes in effector movement relative to the goal (the actor movement) are a possible consequence of left IPL stimulation. Our results indicate that this is true for both intransitive meaningful and meaningless actions. Whilst meaningless action imitation may be more reliant on kinematic processing (Buxbaum et al., 2014), kinematic information might still be relevant to the way in which one must replicate a meaningful action. To take  an extreme example, the imitator is not likely to ignore explicit but irrelevant kinematics, such as a particularly slow or rapid action which does not assist in the development of the final posture (see , for evidence that possibly supports this).
Our results seem, then, to confirm the importance of the left IPL in meeting the kinematic requirements of the observed action, over and above the meaning of that action. The sensitivity of the IPL to observed movement kinematics (Becchio et al., 2012) might substantiate this claim, and as movement necessarily contains kinematic features, our results may help to explain why left IPL activity is frequently reported during imitation in healthy individuals, regardless of the type of action imitated (e.g. Tanaka & Inui, 2002;M€ uhlau et al., 2005;Molenberghs et al., 2010;Jack et al., 2011). There is also evidence, at least for tool-related actions, that damage to the left IPL is more reliably associated with deficits in action performance, rather than action recognition (Tarhan et al., 2015), potentially confirming the priority that the left IPL places on processing movement during imitation, rather than semantic information (which may be more reliably served by the temporal lobe, e.g. Dressing et al., 2016;Kal enine et al., 2010).
It is worth stressing here that our assertion is made in the light of our use of intransitive emblematic gestures, whilst the term 'meaningful gesture' in apraxia literature commonly refers to action pantomimes (i.e. pretending to perform a tool-use hand action without the tool in hand). A more detailed discussion of the distinctions between object-directed and emblematic gestures is provided by Buxbaum et al. (2005). However, there is evidence that damage to the IPL can be associated with deficits in both the imitation of communicative and tool-use gestures (e.g. Dressing et al., 2016), implying that our findings may be applicable beyond the specific action types used in this experiment. We do, of course, suggest further testing this hypothesis.
It is also worth noting that the size of the IPL indicates that it is not solely involved in the kinematic matching process we have proposed. Indeed, this was part of our motivation for using cytoarchitectonically defined brain regions for stimulation. It is perhaps more feasible that the left parietal lobe is involved in multiple processes for imitation and that these processes may be dependent on the type of action to be imitated. For example, the parietal operculum (POp), anterior to area PF, or the anterior intraparietal sulcus (aIPS), superior to area PF, have also been found to be involved in imitation. Specifically, some have suggested that, during imitation, the left POp is involved in comparing information about the imitator's body with the observed actor's body (Mengotti et al., 2013;Kr€ uger et al., 2014) and that activity in this area is correlated with the subsequent accuracy of the imitative action (Kr€ uger et al., 2014). The left aIPS, however, has been suggested to guide object-directed hand movements (Tunik et al., 2007) and, importantly, also appears to represent the goals of observed object-directed actions (Hamilton & Grafton, 2006Hamilton, 2008Hamilton, , 2014Sacheli et al., 2015a). As such, the aIPS could support the imitation of object-directed actions because it ostensibly provides a 'common representational system for the actions of self and other ' (p. T84, Tunik et al., 2007). The sensitivity of the aIPS for both observed and performed action goals (Oosterhof et al., 2010) is in keeping with claims that this area may be part of the putative human mirror neuron system (Tunik et al., 2007). Lastly, Goldenberg & Randerath (2015) suggested that the left IPL could have a role in the apprehension of spatial relationships. They report that damage to this area can result in shared deficits in the imitation of meaningless hand gestures, which require placing the hand in relation to other parts of the body whilst the finger positions remain invariant, and in tasks such as the Token Test which require patients to classify objects based on their spatial and physical properties.
With the above in mind, we can distinguish four potential elements of imitation served by the left parietal lobe. Left area PF and/or area PGthe parts of the SMG and AG stimulated in this experimentcould be involved in creating movement plans based on the spatiotemporal requirements of the to-be-performed action (i.e. the kinematics), regardless of whether the action is imitative. The POp could then ensure that these spatiotemporal requirements are met during imitative scenarios, by comparing the imitator's body to that of the actor (Mengotti et al., 2013;Kr€ uger et al., 2014). This checking process may be essential for meaningless intransitive actions, in the absence of objects to provide context, and could provide one explanation for the defective imitation of meaningless gestures frequently associated with left IPL damage (Goldenberg, 2009). The aIPS, in contrast, could be useful for informing imitation in object-directed scenarios (but see Martin et al., 2016), where visuomotor matching between the observed and to-be-performed action can be done at the goal rather than movement level (Hamilton, 2014). There may also be a role for the left IPL in processing spatial relationships during imitation. This may be particularly important for the imitation of meaningless hand gestures (Goldenberg & Karnath, 2006;Goldenberg & Randerath, 2015), as this task could be reliant on the decomposition of visual information about the observed gesture into 'simple spatial relationships between a limited number of defined body parts' (p. 47, Goldenberg & Randerath, 2015). The absence of specific effects for meaningless hand gesture imitation in our experiment is discussed later.
In the light of the discussion above, it is also necessary for us to comment on the similar effects observed for both of our active stimulation sites. Unfortunately, our study was not able to reveal more about the possible division of labour between the left SMG and the left AG. The existence of similar effects in both the left SMG and the left AG may not necessarily reflect similar roles for each of these regions, but perhaps the connectivity between them. For example, the left AG stimulation may have had an effect on information processing between the AG and SMG, with the greater effect size observed following SMG stimulation (Hedges' g rm = 0.335 vs. 0.281) partially supportive of this claim. As the AG is anatomically connected to the posterior SMG (Seghier, 2013), rTMS over both areas might induce a reduction in efficient SMG functioning if information regarding the kinematics of the observed action is passed in a posterior-anterior (AG to SMG) fashion. It is also worth noting that a previous study indicated that the AG could be involved in both meaningful and meaningless action imitation (Vanbellingen et al., 2014). However, the location of AG in that experiment was more ventral and anterior to ours. An alternative explanation for the similar effects observed for both stimulation sites is that our results reflect an influence of unspecified peripheral stimulation of another area of the IPL, for example cytoarchitectonic area PFm, which lies between PF (in the SMG) and PGa (AG). Interestingly, a previous study by Weiss et al. (2013) using transcranial direct current stimulation (tDCS) found that anodal stimulation over left area PFm improved participants' gesture matching ability. However, the authors suggested that their result could have been driven by a combined effect of SMG and AG stimulation, considering the central position of PFm and the size of tDCS electrodes. Weiss et al. (2013) stated that both parts of the IPL may have to be stimulated simultaneously in order to facilitate gesture matching. Considering our results, the same could be true for influencing gesture imitation and further highlights the importance of considering in detail the role of different regions of the IPL, along with their interactions.
One response to this problem is to suggest further neurostimulation research aimed at disentangling the relative contributions of the seven parcellated regions of the IPL (PFt, PFop [the POp], PFcm, PF, PFm, PGa and PGp), along with the aIPS. Such an endeavour would be essential in order to clarify how different areas of the left parietal lobe may or may not interact and their relative contributions to different types of action and action imitation. In addition, neuroimaging approaches that compare brain activity with actual imitative performance (i.e. Frey & Gerry, 2006;Kr€ uger et al., 2014), across different types of action, are also likely to be invaluable. In general, we suggest that, where possible, researchers try to report in more detail at least the distinction between the SMG and AG, if not their parcellated subregions, rather than stating a role for the IPL in general.
A potential limitation of this study, and another possible explanation for the consistent effects observed in both the SMG and AG, is that these results simply reflect a TMS general influence, in the absence of an active control site. TMS can cause changes in behaviour distinctly unrelated to changes in cortical activity, whereby the clicking sound of the TMS coil alone can facilitate or inhibit reaction times dependent on the time of stimulation within a trial (Duecker et al., 2013;Meteyard & Holmes, 2018). However, as far as we are aware, this has only been found for online TMS protocols, and we are not aware of data suggesting that the offline approach we used is also likely to result in 'mere presence' effects of rTMS. In addition, the no-rTMS session involved much of the same conditions as in the active stimulation sessions (e.g. 15 minutes of waiting during TMS application, clicking sound).
Finally, it is interesting that our primary result was found specifically for finger gestures, rather than hand gestures. This is partly at odds with some previous findings in apraxia, although our stimuli were not modelled on previous distinctions between hand and finger gesturesour main aim was to ensure there was a sufficient number of different emblematic stimuli for participants to copy. There is some evidence to suggest that defective finger gesture imitation is more associated with left inferior frontal gyrus damage, compared to defective hand gesture imitation, which is more associated with left IPL damage (Goldenberg & Karnath, 2006). As stated above, some discuss the role of the left IPL in meaningless hand gesture imitation in terms of 'body part coding', where hand gesture imitation is reliant on the spatial mapping of the hand in relation to other body parts (e.g. Goldenberg & Karnath, 2006;Goldenberg & Randerath, 2015). However, some experimenters have failed to find a dissociation for brain regions related to defective hand and finger imitation (Achilles et al., 2017). It is also possible that the movement to attain the posture, rather than the final hand position alone, may be an important factor. For example, whilst the Goldenberg (1996) assessment for meaningless gestures (as used by Karnath, 2006 andRanderath, 2015) considers only performance on the final gesture posture, Buxbaum et al. (2014) examined dynamic movements; hence, their suggestion that dynamic change in body part position is important. It is possible that the difference between kinematic and postural elements of movement may take precedence over the difference between hand and finger gestures, although further research would be needed to clarify this. If this is the case, the fact we only observed kinematic effects in the finger velocities for finger-specific gestures may reflect the fact that the dynamic change in each of the fingers independently is more complex than the movement of the hand as a whole.
In conclusion, we found that the left IPL is involved in matching observed digit velocity in action imitation, regardless of action meaning. More work is needed to expand on how imitation kinematics are processed in the left IPL and the relative contributions of the SMG and AG and their respective cytoarchitectonic subregions. These nuances can be examined using neuronavigated TMS. Our results highlight that brain stimulation may help close the gap in understanding imitation in apraxia and in healthy people, particularly if it is combined with large-scale motion-tracking.