using mu rhythm desynchronization to measure mirror neuron

PAPER

Using mu rhythm desynchronization to measure mirror neuronactivity in infants

Par Nystrom, Therese Ljunghammar, Kerstin Rosander and Claes vonHofsten

Department of Psychology, Uppsala University, Sweden

Abstract

The Mirror Neuron System hypothesis stating that observed actions are projected onto the observer’s own action system assignsan important role to development, because only actions mastered by the observer can be mirrored. The purpose of the presentstudy was to investigate whether there is evidence of a functioning mirror neuron system (MNS) in 8-month-old infants. High-density EEG was used to assess the mu rhythm desynchronization in an action observation task where the infants observed a livemodel. To reduce noise, ICA decompositions were used. The results show a higher desynchronization of the mu rhythm wheninfants observed a goal-directed action than when they observed a spatially similar non-goal-directed movement. Thelocalizations of the sources are in agreement with those proposed by the MNS hypothesis. This indicates that the MNS isfunctioning at this age.

Introduction

It has been suggested that other people’s actions could beunderstood by mapping them onto one’s own motorrepresentation of those actions. The hypothesis is thatthis mapping is accomplished by a neuronal system, theMirror Neuron System (MNS), which is activated bothby the execution of one’s own goal-directed actions andby the perception of someone else performing the sameactions (Rizzolatti, Fadiga, Fogassi & Gallese, 1996;Fadiga & Craighero, 2004). Neurophysiological evidenceindicates that the MNS is a distributed system with atleast three primary nodes, the ventral premotor area(Area 44 in humans), the superior temporal sulcus (STS),and the inferior parietal lobule (IPL) (Rizzolatti et al.,1996; Fogassi, Ferrari, Gesierich, Rozzi, Chersi & Riz-zolatti, 2005). Studies using fMRI have found elevatedactivation in STS, IPL and Area 44 in adult humansduring action observation (Rizzolatti & Craighero,2004). Using MEG, Nishitani and Hari (2002) showedthat observed and imitated lip movements activated firstthe STS, then the IPL, and finally Area 44. All evidence,however, is not in favour of the presence of MNS. Forinstance, while Chong, Cunnington,Williams, Kanwisherand Mattingley (2008) found evidence of mirror neuronsusing fMRI adaptation to examine cross-modal effects ofaction execution and action observation, Lingnau, Ges-

ierich and Caramazza (2009) did not find such evidence.Obviously, it is yet to be determined exactly how andwhen the MNS gets activated.

As the MNS hypothesis states that other people’sactions are understood in terms of one’s own motorprogrammes, it is expected that observed actions are notunderstood until subjects master those actions them-selves. Thus, if the MNS hypothesis is valid, it is expectedthat the development of children’s understanding ofother people’s actions is parallel to their own motordevelopment. As infants become more proficient in themotor domain, they rapidly develop social skills requir-ing understanding of other people’s actions. For instance,towards the end of the first year of life, infants under-stand pointing and they point to objects (Butterworth,2003; Liszkowski, Carpenter & Tomasello, 2006; Toma-sello, Carpenter & Liszkowski, 2007). From 6 months ofage, infants have been shown to imitate other people’sactions (Barr, Rovee-Collier & Campanella, 2005; vonHofsten & Siddiqui, 1993) and from 9 months of agethey perform deferred imitation (Meltzoff, 1988).

One indirect piece of evidence for the MNS hypothesisis found in people’s proactive eye movements duringaction observation. When subjects perform actions, theeyes are moved to the goal proactively (Land & Hayhoe,2001). The purpose of such eye movements is to guidethe hand to the goal. If the MNS hypothesis holds, the

Address for correspondence: Claes von Hofsten or P�r Nystrçm, Department of Psychology, Uppsala University, Box 1225, SE 75142 Uppsala,Sweden; e-mail: [email protected] or [email protected]

� 2010 Blackwell Publishing Ltd, 9600 Garsington Road, Oxford OX4 2DQ, UK and 350 Main Street, Malden, MA 02148, USA.

Developmental Science 14:2 (2011), pp 327–335 DOI: 10.1111/j.1467-7687.2010.00979.x

proactive gaze shifts performed during action perfor-mance should be produced during action observation aswell. Because the eyes are free to move when observingsuch actions, the MNS hypothesis predicts that subjectsshould produce eye movements similar to those pro-duced when they perform the tasks themselves. Inaccordance with this, Flanagan and Johansson (2003)found that adult subjects performed proactive eyemovements to the goal of a displacement action bothwhen they performed the action themselves and whenthey observed someone else perform the action. Suchindirect behavioural evidence for MNS activity in infantswas obtained by Falck-Ytter, Gredeb�ck and von Hof-sten (2006). They found that 12-month-old but not6-month-old infants looked proactively at the goal whena hand transported an object there. For both age groupsof infants, the eyes just tracked the object when the sameobject motion was shown outside the context of anaction. It should be noted that 12-month-olds but not6-month-olds displace an object they have grasped to adifferent location in a goal-directed way. They mightmove the grasped object to the mouth, however. Inaccordance with this, Kochukhova and Gredeb�ck(2009) found that observed hand movements involved ineating actions were accompanied by proactive eyemovements to the mouth of the model in 6-month-oldinfants.

The purpose of the present study was to find neuralevidence of mirror neuron activity in human infants whoobserve a goal-directed action that they master. Theimportance of this question is reflected in several explicitrequests for empirical neurophysiological data (Lepage &Th�oret, 2007; Bertenthal & Longo, 2007; Kilner &Blakemore, 2007). MNS activity has earlier been identi-fied in adults by analysis of the mu rhythm of EEG(Pineda, 2005; Muthukumaraswamy & Johnson, 2004a)and related demonstrations have been performed whenusing MEG (see e.g., Nishitani & Hari, 2000; Hari,Forss, Avikainen, Kirveskari, Salenius & Rizzolatti,1998). There are at least three reasons for associating thedesynchronization of the mu rhythm with MNS. First,these studies show that the mu rhythm oscillates atapproximately 9–13 Hz in adults but is suppressed bothwhen a subject performs an action and observes someoneelse perform the same action. Second, an importantfeature of mirror neurons is that they are tuned to goal-directed actions. The mu rhythm is also modulated by thegoal-directedness of actions. It is more desynchronizedwhen adults either perform or observe goal-directedactions compared to non-goal-directed actions (Muthu-kumaraswamy & Johnson, 2004a; Muthukumaraswamy,Johnson & McNair, 2004). Third, the mu rhythm is asensorimotor rhythm that appears to consist of severalfrequencies with different origins both in motor areasand in parietal sensory areas (Pineda, 2005). This isconsistent with the locations of mirror neurons.

The infant alpha rhythm between 5 and 9 Hz showsstrong resemblances to the adult mu rhythm, and has

therefore been suggested as the infant mu rhythm(Stroganova, Orekhova & Posikera, 1999; Marshall,Bar-Haim & Fox, 2002). Orekhova, Stroganova, Posi-kera and Elam (2006) found that the action-related partof the alpha rhythm emerges at around 5–8 months andincreases in frequency and amplitude over age. Recently,Southgate, Johnson, Karoui and Csibra (2010) testedthe functional reactivity of the infant alpha mu rhythmsin both action execution and action observation tasks.Nine-month-old infants either performed a reachingtask or observed someone else perform it. It was foundthat for both tasks, the mu rhythm was attenuated.Southgate, Johnson, Osborne and Csibra (2009) alsofound that when 9-month-olds observed a reach for ahidden target, the mu rhythm was attenuated, but notwhen the model performed a non-goal-directed handmovement toward the same place or when the goal wasremoved. A third study by van Elk, van Schie, Hunnius,Vesper and Bekkering (2008) showed that the magni-tude of mu rhythm desynchronization in 14- to16-month-old infants who watched action videos wasdirectly related to their own action experience. Strongermu desynchronizations were found for observations ofcrawling compared to walking and the size of the effectwas strongly related to the infant’s own crawlingexperience.

Although the obtained mu rhythm desynchronizationssupport the MNS hypothesis, none of these studies re-ported evidence of the location in the brain where thesedesynchronizations originated. The MNS hypothesisstates that the premotor area should be a source. In thepresent study we tested this hypothesis by independentcomponent analysis (ICA) of high-density EEG record-ings. ICA is a blind source separation technique thataims at finding components that are most statisticallyindependent of each other (Delorme & Makeig, 2004).Since the MNS in infants (measured by the desynchro-nizing mu rhythm) cannot be expected to account formuch variance but still has unique statistical properties,ICA appears to be a suitable tool. Unfortunately, ICAintroduces a component selection problem. Because themu rhythm can be expected to decompose into one or afew components from each subject, only these compo-nents should be considered in the analysis. We used threeconditions to solve the independent component selectionproblem. In one condition, the base-line mu rhythmactivation was estimated when the subject observed anon-moving human model. In the other two conditions,the model moved his arm, either by reaching for andgrasping an object or by placing his hand on the table.Then, the difference in mu rhythm activation between thebaseline and the two movement conditions was used forthe selection of EEG sources. Finally, the differencebetween these two conditions was analysed. If the MNSis activated by the goal-directed reaching movement,then the mu rhythm should be more desynchronized inthis condition compared to the observation of a simplearm displacement.

328 P�r Nystrçm et al.

� 2010 Blackwell Publishing Ltd.

A recent study of 6-month-old infants used a similarmethod to extract mu rhythm activation elicited by videopresentations of a passive and an active actor (Nystrçm,2008). Although no evidence of a consistent difference inthe mu rhythm was obtained for these two situations, anERP analysis of the same data showed differentresponses for goal-directed and non-goal-directedmovements. In the present study we optimized theexperiment in two ways. First, 8-month-old instead of6-month-old infants were studied. If MNS develops ininfancy, its indicators should get more distinct with age.Second, live actors instead of video recorded ones wereused. In earlier studies of adults (J�rvel�inen, Sh�rmann,Avikainen & Hari, 2001; Shimada & Hiraki, 2006) and7-month-old infants (Shimada & Hiraki, 2006), liveactors were found to elicit stronger MNS activationcompared to video recordings. In fact, although Shimadaand Hiraki (2006) found that action observation gave riseto activation in the motor areas in both the live and theTV setting, only the activation in the live setting wassignificant. With these two improvements the presentstudy investigated whether infant MNS activity could beassessed by mu rhythm desynchronization.

Methods

Participants

Thirty-six healthy 8-month-old infants participated.They were tested on a simple reaching task, and allinfants reached successfully for objects presented tothem. The parents were given written information uponarrival at the lab and signed a written consent form inaccordance with the Helsinki Declaration. The experi-ment was approved by the Ethics Committee at UppsalaUniversity. Two infants were excluded before analysisdue to fussing or inattention, and two infants wereexcluded due to technical problems. All parents receiveda gift certificate of 100 Swedish kronor (approximately€9) at a local toy store.

Procedure

A high density 128 channel EEG net (EGI, Corp.,Eugene, OR) of the appropriate size was properly appliedto the infant’s head. The infants were then seated atapproximately 1 metre from a scene where a modeldemonstrated three kinds of events. The subjects sawthese events from the side. On the table, a short railwaytrack with a toy train was placed (see Figure 1). First, themodel demonstrated a baseline condition where he satpassively (static condition). Then, two hand movementconditions were shown. In one of them (goal-directedcondition), the model grasped the toy train and moved itto a new position further away. When the toy was in thenew position the model performed the second condition.In this condition (non-goal-directed condition), the

model moved his hand toward the first toy position (nowempty) and simply placed the hand flat on the table. Asthe railway track was slanted toward the original startingposition, the train returned there after these events whenan electronic trigger was released. Except for the veryfirst condition that for practical reasons was always agoal-directed movement, these conditions were presentedin randomized order by sometimes interleaving the staticcondition and sometimes not, or presenting the goal-directed or non-goal-directed condition twice beforeproceeding to any of the other conditions. The conditionswere repeated until the subject was no longer interestedor started to fuss.

EEG recordings and processing

The 128 EEG channels were sampled at 250 Hz, with ananalogue hardware band pass filter at 0.1 to 100 Hz

Non-goal-directed condition

Goal-directed condition

Figure 1 Photo of the experimental setup from the subjects’viewpoint. Top: the non-goal-directed condition when thehand touches the tabletop. Bottom: the goal-directed conditionjust before touch of the object.

Measuring mirror neuron activity in infants 329


(EGI, Corp., Eugene, OR). During recording, the infantsobserved the model either sitting still or performing thetwo actions described above. The EEG was time-lockedby the model when the hand touched the toy or the table,or after approximately 1 second of sitting passively.Manual triggers hidden from the infants’ view were usedto time-lock the events. The triggers were operated bythe model and triggered by the hand that performed theaction (the triggers were hidden under a cloth onthe table and under the model’s foot behind the tablefor the static condition). The model released the triggerwith the moving hand as it arrived at the table-top orgrasped the object. In cases where the object wasgrasped, the trigger was released by the little finger of thegrasping hand situated on the far side of the object andthus was not visible to the subject. The grasping actionwas performed by the thumb and three of the otherfingers. After training the triggering over a few hundredtrials during pilot testing, the temporal synchronybetween the grasping and the triggering was quite high.The timing precision was determined in a separateexperiment where a trigger was placed on the objectreached for in addition to the trigger used during theaction observation. For a sequence of 182 control mea-sures of the trained experimenter, the mean time differ-ence was 9.7 ms and the standard deviation 27.1 ms.

The experimenter observed the behaviour of thesubject during the whole session, and conditions werenever presented unless the infant was in an attentiveand passive mode. The resulting EEG recordings con-sisted of between 10 and 49 trials in each conditionfrom every subject (mean = 21.06; SD = 7.40). Thedata were transferred to the MATLAB v.7.2 environ-ment and analysed using the EEGLAB v5.03 toolbox.First, 29 of the outermost sensors were removed due tobad contact in most infants, although the net wasproperly positioned. The continuous EEG was thenband pass filtered from 2 to 20 Hz using two-way least-squares FIR filtering to remove noise and to focus onthe frequencies where most brain-related signals appear.The data were segmented into trials from )1s to 1s aftertime-lock (touch of toy or table, or after approximately1 second of resting). A modified artefact rejectionroutine for high density EEG (Junghçfer, Elbert, Tucker& Rockstroh, 2000) removed bad trials and sensorsbased on their maximum values, standard deviationsand range values. Next, the EEG was transformed toaverage reference as recommended by the EEGLABguidelines. Modifications of the artefact rejection rou-tine were removal of interpolation, which could violatethe assumption for the ICA of independent measure-ments from each channel. Bad trials and channels weresimply excluded from the dataset to prevent spreadingof bad data when re-referencing. A natural-gradientlogistic infomax independent component analysis wasperformed on the data (the Runica algorithm; Delorme& Makeig, 2004), which resulted in as many indepen-dent components as remaining channels minus one for

each subject (mean = 88, ranging from 78 to 98 com-ponents).

Selection of independent components

Although each subject’s data were decomposed intomany components only a few of them were assumed toreflect mirror neuron activity. The other componentswere assumed to mask the signal of interest by intro-ducing noise and neural signals not related to the MNS.To reduce noise and to focus on components with mirrorneuron properties, we performed the two-step compo-nent selection procedure described below. First weexcluded components that reflected artefacts by the fol-lowing criteria. Components with any abnormal ICAweight, > 2.7 SD of all weights within a component, wereconsidered artefactual and excluded. The value of 2.7 SDwas decided by visual inspection of all components toretain components with dipole-like scalp projections andto exclude components originating from channel pops ormovement artefacts. Two EEG experts and four novicesvisually inspected the components that showed murhythm desynchronization. The area under the receiveroperating characteristic (ROC) curve (AUC) was 0.95,and justified the use of a maximum of 2.7 SDs in anyICA weight as a rather conservative threshold. Also, themaximum absolute amplitudes of components were cal-culated to identify outlier values that could bias thesubsequent frequency analysis. Trials with abnormalvalues (> 3 SD) were thus excluded and also componentswith less than 10 trials in any condition. Second, weselected components related to mu rhythm desynchro-nization from the remaining components. Frequencyspectra of the three conditions were extracted. To speedup the computations we used Welch’s method (hammingwindows of 256 samples length and 128 samples overlap)instead of the EEGLAB timef function at this point. Theresults were converted to dB values across a 1 Ohm ref-erence load to simplify visual presentation of the powerspectra. Components with a power peak greater than1 dB in the static condition and a decrease in the powerpeak from this value greater than 1 dB for the twomovement conditions were selected for further statisticalanalysis. This procedure selected components thatshowed mu desynchronization in the live movementconditions but without making any distinction betweenthe goal-directed and non-goal-directed conditions. Thepeak power was calculated as the maximum differencebetween the power spectrum and the linear interpolationof the power values at the boundaries of the 8-month-olds’ mu band (5–9 Hz). This definition of the peakpower would account for overall differences in the powerspectra between conditions. The definition of the 5–9 Hzinterval was based on previous studies of infant alpharhythms (Stroganova et al., 1999). An example of thefrequency spectra used to select components is shown inFigure 2 together with the mean spectrum of all selectedcomponents. It should be noted that the highest



amplitude in this frequency interval was 7 Hz in agree-ment with Stroganova et al. (1999). In total, 43 compo-nents with 10–49 trials (mean = 21.06; SD = 7.40) fromeach condition originating from 23 subjects wereselected. Each subject contributed between 1 and 5components (mean = 1.87; SD = 1.12). If the ICA didnot decompose any component with mu characteristicsthe subject was excluded. There could be several reasonsfor this. First, the mu rhythm is subject to individualdifferences. Second, the mu rhythm might not havedeveloped in some subjects. Third, the measurementscould have been contaminated by some minor artefactsthat the ICA could not decompose. In any of these cases,the subject would not have any signal to contribute, andwould be excluded.

The components that were not selected were sub-tracted from the raw EEG to create datasets pruned fromnoise, artefacts and neural activity that were not related

to mu desynchronization. Thereby only the selectedcomponents were represented in the scalp channelactivity of the pruned datasets. These componentsaccounted for 1.4% of the variance in the raw data (eachcomponent ranging from 0.1% to 3.7%).

To control the validity of the selection procedure, theselected components could be interchanged with randomcomponents. The resulting datasets should not show anysignificant effects, as randomly selected signals tend tocancel each other out.

Statistical analysis

First, the statistical procedure was performed channelwise on the raw EEG datasets (n = 32). The same pro-cedure was then repeated on the pruned datasets (n = 23)as a more elaborate analysis of the signal using randomlyselected components. A time ⁄ frequency analysis usingdiscrete wavelet transforms was performed on eachchannel’s (or component’s) conditions using the stan-dard EEGLAB timef function. The window size was 64samples (256 ms) wide, and Hamming windows wasapplied 200 times at an average step of 2.191 samples(8.76 ms). The EEGLAB timef function returned 20frequency bands ranging from 2.0 Hz to 49.8 Hz and 200time points ranging from )800 ms to 800 ms from time-lock. However, as we were mainly interested in the5–9 Hz frequency band, this interval was averaged andthe goal-directed and non-goal-directed conditions werecompared with reference to this measure at every timepoint. As multiple significance tests inflate the risk ofType 1 errors, only clusters of 10 or more adjacent sig-nificant p-values (p < .05) were considered. The mostprominent channel was then analysed in further detailusing pixel wise t-tests in all frequency bands between 2and 20 Hz. To control for multiple testing of these200 · 20 time ⁄ frequency points only clusters of 20 ormore adjacent p-values (p < .05) were considered sig-nificant.

Results

The experimental design was very successful as indicatedby the low exclusion rate. Only two infants out of 34 wereexcluded because of insufficient data. The remaining 32infants watched the model attentively during more than10 trials in each condition. The data from these infantswere analysed using the raw data as well as the data fromselected independent components. In the analysis ofselected components, the t-tests between the goal-direc-ted and non-goal-directed conditions showed a signifi-cant desynchronization of the mu rhythm band for thegoal-directed condition when the hand touched theobject. A global power minimum of )1.6 dB was foundapproximately 10 ms after time-lock (p = .0007) forchannel number 105 in the EGI Geodesic Sensor Net 128channel v2.0. This minimum is found in a significant

Figure 2 Power spectra of component activity. The high-lighted interval (5–9 Hz) was used for determining the murhythm power. The red line marks the ‘static’- condition andthe two blue lines mark the two movement conditions (‘goal-directed action’ and ‘non-goal-directed action’). Top: exampleof mu rhythm power peak estimation in a single component.Bottom: mean of all selected components power spectra.



interval that starts approximately 60 ms before time-lockand remains until 55 ms after time-lock. The channelswith the largest difference between the goal-directed andthe non-goal-directed movement conditions are locatedon the right frontal lobe and in central areas, as shown inFigure 3 (channel numbers 103, 104, 105, 107, 110, 111and 123). Two more channels with significant desyn-chronization were located in the left frontal lobe (overprefrontal cortex, channel numbers 29 and 35) andfinally two over the occipital lobe (channel numbers 71and 76). A magnified view of the most significantchannel is presented in Figure 4, which also shows theamplitudes of 5–9 Hz for all conditions. A detailedtime ⁄ frequency analysis of the most significant sensor ispresented in Figure 5. The difference between the goal-directed and non-goal-directed conditions had a globalminimum of 1.6 dB 10 ms after the hand touches theobject for the frequency 6.8 Hz (p = .0002). The signifi-cant effects start at 75 ms before time-lock and remainuntil 170 ms after time-lock.

To test the validity of the t-tests, a bootstrap analysisusing 5000 permutations was performed. The bootstraptest was chosen as it does not rely on normally distrib-uted data. The significant t-tests were all included in thesignificant bootstrap results. Using the raw EEG data,point wise statistical tests (two-tailed t-tests, n = 32) ofthe mean amplitude differences in the 5–9 Hz frequency

band between the goal-directed and non-goal-directedconditions showed no significant desynchronization inany of the channels analysed. No significant effects werefound in the five runs with EEG control data usingrandomly selected components.

Discussion

The results from the selected components show a greaterdesynchronization of the mu rhythm in 8-month-oldinfants when they observed goal-directed actions com-pared to when they observed non-goal-directed actions.In relation to studies performed on adults, the desyn-chronization onset is very similar (Muthukumaraswamy& Johnson, 2004b). The significant effect remains forabout 120 ms and includes the goal of the action. It is aconservative estimate, as individual variation in pro-cessing time between trials will decrease the effects whenthe mu rhythm desynchronization does not overlap.Thus, it can be concluded that 8-month-old infants showsimilar mu rhythm desynchronization as adults whenobserving goal-directed actions. Nine out of 11 sourceswere located at central areas, where the mu rhythmdesynchronization has been assessed by previous studies(see e.g. Pineda, 2005). Thus, the results support theMNS hypothesis. An interesting feature of the results is

Figure 3 Mean amplitude differences of the 5–9Hz frequency band between the goal-directed and non-goal-directed conditions inthe pruned datasets containing only the selected independent component projections. Each curve represents a channel and thelayout describes the spatial relations of the sensors on the scalp (nose pointing upwards). Significant differences between conditionsare marked with yellow intervals, and these channels are shaded (n = 23, p < .05 and corrected for multiple testing). The significantsensors in right central areas are channel numbers 103, 104, 105, 107, 110, 111 and 123. Two more channels with significantdesynchronization were located in the left frontal lobe (over prefrontal cortex, channel numbers 29 and 35) and finally two over theoccipital lobe (channel numbers 71 and 76). The channel with longest significant interval is marked with a red circle. This sensor ispresented in more detail in Figures 4 and 5.



that seven of the nine sources in the premotor area weresituated on the right hemisphere, whereas Hari et al.(1998) and Southgate et al. (2010) found higher activa-tion on the left side and Nishitani and Hari (2000) foundit bilaterally. Kilner, Marchant and Frith (2009) obtainedmore activation on the contralateral side which indicatesthat the lateralization of the activation may be taskdependent.

While this analysis of EEG source dynamics as inde-pendent components does not require an explicit head

model, one might argue that the signal could reflectposterior alpha waves within the same frequency band(Marshall et al., 2002). One argument against posterioralpha confounding is that visual differences (that wouldaffect the alpha waves) between the tested conditionswere minimized. All objects were present in the scene inall conditions, and the time-locking of the EEG occurredwhen the hand had decelerated and touched the toy trainor the table. Furthermore, the significant sensors in theoccipital area could be explained by the dipole field of

Figure 4 A presentation of the separate conditions in the most significant sensor (marked with a red circle in Figure 3). The leftrow shows data from the raw datasets, and the right row shows data from the pruned datasets that contain only the selectedindependent component projections.

Figure 5 A presentation of the separate conditions in the most significant sensor (marked with a red circle in Figure 3) using thetime ⁄ frequency decomposition returned by the EEGLAB timef function. The frequencies are truncated above 20 Hz since thesefrequencies were filtered out before the ICA. The left row shows data from the raw datasets, and the right row shows data from thepruned datasets that contain only the selected independent component projections. The bottom row shows the point wise statisticalt tests’ p-values (left plot n = 32, right plot n = 23), thresholded at alpha = 0.05, adjusted for multiple comparisons (200 · 20 tests)by removing significant clusters smaller than 20 pixels.



the stronger signal from the right frontal lobe. In theenlarged plots (Figures 4 and 5) all conditions areshown. These plots show that the mu rhythm power hasthe greatest amplitude when the infant observes a passivemodel, that the amplitude is lower when the infantobserves a simple arm movement, and that it is signifi-cantly lower when the infant observes a model graspingan object. All these characteristics have been related tomu rhythm desynchronization in adults and to the adultMNS. This robust desynchronization represents theinfant equivalent of the adult mu rhythm and is tightlylinked to MNS activity. It is in agreement with previousadult mu rhythm studies with reference to the timing ofdesynchronization, the significant frequencies, and thesource locations (Pineda, 2005; Muthukumaraswamyet al., 2004a; Nishitani & Hari, 2000; Hari et al., 1998;Oberman, Hubbard, McCleery, Altschuler, Ramachan-dran & Pineda, 2005). Furthermore, it is also in agree-ment with earlier studies on the frequency of infant murhythm (Orekhova et al., 2006; Gelisse & Crespel, 2006).

Another important issue is whether any other move-ment by the subjects could cause mu rhythm desynchro-nization and thereby contaminate the measurements. Theinfants did move occasionally but these movements wereusually small, evenly distributed over the whole recordingsession, and any resulting unwanted mu rhythm desyn-chronization would just reduce the grand averages (justlike artefacts in conventional ERPs). The strong time-locking of the signal implies that the significant effects arenot related to randomly occurring movements by thesubject. However, if the significant effects are the result ofsystematic movements by the subject, this in itself wouldbe a strong argument that the MNS has been measured.Since the mu rhythm desynchronization is precisely time-locked to the touch of the object, the infant must havepredicted the goal of the action and started to plan thatmotor movement in advance. Also, a synchronized motorresponse could indicate an immature inhibition of mirrorneuron motor resonance.

Although the present study reports how the murhythm is modulated by the goal-directedness of theaction, it also raises questions of how the static baselinerelates to the hand movement conditions. Since the adultMNS has been shown to respond to non-object-directedactions, called intransitive movements (Rizzolatti &Craighero, 2004), it is possible that the infant MNSwould also respond in a similar way. To test this wecompared the non-goal-directed trials with the statictrials using the raw data. We also pooled the goal-directed and non-goal-directed trials and compared themwith the static trials to test the mu desynchronizationbetween the baseline and hand movements. This testshowed strong mu rhythm desynchronization duringhand movements both in frontal and parietal areas.

In summary, the present results together with theresults of van Elk et al. (2008), Southgate et al. (2010)and Southgate et al. (2009) indicate that similar brainprocesses are activated when infants perform and observe

actions. These processes are functioning when infants are8 months of age. Behavioural data support these results.Using the habituation ⁄ dishabituation paradigm, Wood-ward (1998) found that 6-month-old infants are sensitiveto the action goal of others but only when it is performedby a human agent. Sommerville and Woodward (2005)found that 10-month-old infants identify the goal ofaction sequences but only to the extent that they canperform them. The idea that infants represent the acts ofothers and their own acts in comparable terms is notunique to the MNS hypothesis. Infant imitation dem-onstrates that the perception and production of humanaction are closely linked by a ‘supramodal’ representa-tion of action (Meltzoff, 2007a, 2007b). The results onmu rhythm desynchronization indicate that these linksbetween perception and action are not just conceptualbut actually correspond to common brain processes.

Acknowledgements

This research was supported by grants from the SwedishResearch Council to KR (421-2003-1508), from EUintegrated project Robotcub (EU004370) and EU projectContact (EU5010) to CvH. We thank the parents of theinfants who participated in this study. We also thank theEEGLAB development team for expert technical assis-tance and Robert Oostenweld for distribution of thesource localization algorithm.

References

Barr, R., Rovee-Collier, C., & Campanella, J. (2005). Retrievalprotracts deferred imitation by 6-month-olds. Infancy, 7, 263–283.

Bertenthal, B.I., & Longo, M.R. (2007). Is there evidence of amirror system from birth? Developmental Science, 10, 526–529.

Butterworth, G. (2003). Joint visual attention in infancy. In G.

Bremner & A. Slater (Eds.), Theories of infant development(pp. 317–354). Oxford: Blackwell.

Chong, T.T.-J., Cunnington, R., Williams, M.A., Kanwisher,N., & Mattingley, J.B. (2008). fMRI adaptation revealsmirror neurons in human inferior parietal cortex. CurrentBiology, 18, 1576–1580.

Delorme, A., & Makeig, S. (2004). EEGLAB: an open sourcetoolbox for analysis of single-trial EEG dynamics includingindependent component analysis. Journal of NeuroscienceMethods, 134, 9–21.

Fadiga, L., & Craighero, L. (2004). Electrophysiology of actionrepresentation. Journal of Clinical Neurophysiology, 21, 157–169.

Falck-Ytter, T., Gredeb�ck, G., & von Hofsten, C. (2006).

Infants predict other people’s action goals. Nature Neuro-science, 9, 878–879.

Flanagan, J.R., & Johansson, R.S. (2003). Action plans used inaction observation. Nature, 424, 769–771.

Fogassi, L., Ferrari, P.F., Gesierich, B., Rozzi, S., Chersi, F., &Rizzolatti, G. (2005). Parietal lobe: from action organizationto intention understanding. Science, 308, 662.



Gelisse, P., & Crespel, A. (2006). Mu rythm in the infant.Neurophysiologie Clinique, 36, 261–263.

Hari, R., Forss, N., Avikainen, S., Kirveskari, E., Salenius, S.,& Rizzolatti, G. (1998). Activation of human primary motorcortex during action observation: a neuromagnetic study.Proceedings of the National Academy of Sciences, USA, 95,15061–15065.

J�rvel�inen, J., Sch�rmann, M., Avikainen, S., & Hari, R.(2001). Stronger reactivity of the human primary motorcortex during observation of live rather than video motoracts. NeuroReport, 12, 3493–3495.

Junghçfer, M., Elbert, T., Tucker, D.M., & Rockstroh, B.(2000). Statistical control of artifacts in dense array EEG ⁄MEG studies. Psychophysiology, 37, 523–532.

Kilner, J.M., & Blakemore, S.-J. (2007). How does the mirrorneuron system change during development? DevelopmentalScience, 10, 524–526.

Kilner, J.M., Marchant, J.L., & Frith, C.D. (2009). Relation-ship between activity in human primary motor cortex duringaction observation and the mirror neuron system. PLoSONE, 4 (3), e4925.

Kochukhova, O., & Gredeb�ck, G. (2009). Preverbal infantsanticipate that food will be brought to the mouth: an eyetracking study of manual feeding and flying spoons. ChildDevelopment (in press).

Land, M., & Hayhoe, M. (2001). In what ways do eye move-ments contribute to everyday activities? Vision Research, 41,3559–3565.

Lepage, J.-F., & Th�oret, H. (2007). The mirror neuron system:grasping others’ actions from birth? Developmental Science,10, 513–523.

Lingnau, A., Gesierich, B., & Caramazza, A. (2009). Asym-metric fMRI adaptation reveals no evidence for mirrorneurons in humans. Proceedings of the National Academy ofSciences, USA, 106, 9925–9930.

Liszkowski, U., Carpenter, M., & Tomasello, M. (2006). Ref-erence and attitude in infant pointing. Journal of ChildLanguage, 33, 1–20.

Marshall, P.J., Bar-Haim, Y., & Fox, N.A. (2002). Develop-ment of the EEG from 5 months to 4 years of age. ClinicalNeurophysiology, 113, 1199–1208.

Meltzoff, A.N. (2007a). ‘Like me’: a foundation for socialcognition. Developmental Science, 10, 126–134.

Meltzoff, A.N. (2007b). The ‘like me’ framework for recog-nizing and becoming an intentional agent. Acta Psychologica,124, 26–43.

Meltzoff, A.N. (1988). Infant imitation and memory: nine-month-olds in immediate and deferred tests. Child Develop-ment, 59, 217–225.

Muthukumaraswamy, S.D., & Johnson, B.W. (2004a). Chan-ges in rolandic mu rhythm during observation of a precisiongrip. Psychophysiology, 41, 152–156.

Muthukumaraswamy, S.D., & Johnson, B.W. (2004b). Primarymotor cortex activation during action observation revealedby wavelet analysis of the EEG. Clinical Neurophysiology,115, 1760–1766.

Muthukumaraswamy, S.D., Johnson, B.W., & McNair, N.A.(2004). Mu rhythm modulation during observation of anobject-directed grasp. Cognitive Brain Research, 19, 195–201.

Nishitani, N., & Hari, R. (2000). Temporal dynamics of cor-tical representation for action. Proceedings of the NationalAcademy of Sciences, USA, 97, 913–918.

Nishitani, N., & Hari, R. (2002). Viewing lip forms: corticaldynamics. Neuron, 36, 1211–1220.

Nystrçm, P. (2008). The infant mirror neuron system studiedby high density EEG. Journal of Social Neuroscience, 3, 334–347.

Oberman, L.M., Hubbard, E.M., McCleery, J.P., Altschuler,E.L., Ramachandran, V.S., & Pineda, J.A. (2005). EEGevidence for mirror neuron dysfunction in autism spectrumdisorders. Cognitive Brain Research, 24, 190–198.

Orekhova, E.V., Stroganova, T.A., Posikera, I.N., & Elam, M.

(2006). EEG theta rhythm in infants and preschool children.Clinical Neurophysiology, 117, 1047–1062.

Pineda, J.A. (2005). The functional significance of mu rhythms:translating ‘seeing’ and ‘hearing’ into ‘doing’. Brain ResearchReviews, 50, 57–68.

Rizzolatti, G., & Craighero, L. (2004). The mirror neuronsystem. Annual Review of Neuroscience, 27, 169–192.

Rizzolatti, G., Fadiga, L., Fogassi, L., & Gallese, V. (1996).Premotor cortex and the recognition of motor actions. Cog-nitive Brain Research, 3, 131–141.

Shimada, S., & Hiraki, K. (2006). Infant’s brain responses tolive and televised action. NeuroImage, 32, 930–939.

Sommerville, J., & Woodward, A.L. (2005). Pulling out theintentional structure of action: the relation between actionprocessing and action production in infancy. Cognition, 95,1–30.

Southgate, V., Johnson, M.H., Karoui, I., & Csibra, G. (2010).

Motor system activation reveals infants’ on-line prediction ofothers’ goals. Psychological Science, 21, 355–359.

Southgate, V., Johnson, M.H., Osborne, T., & Csibra, G.

(2009). Predictive motor activation during action observationin human infants. Biological Letters, 5 (6), 769–772.

Stroganova, T.A., Orekhova, E.V., & Posikera, I.N. (1999).

EEG alpha rhythm in infants. Clinical Neurophysiology, 110,997–1012.

Tomasello, M., Carpenter, M., & Liszkowski, U. (2007). A newlook at infant pointing. Child Development, 78, 705–722.

van Elk, M., van Schie, H.T., Hunnius, S., Vesper, C., &Bekkering, H. (2008). You’ll never crawl alone: neurophysi-ological evidence for experience-dependent motor resonancein infancy. NeuroImage, 43, 808–814.

von Hofsten, C., & Siddiqui, A. (1993). Using the mother as areference in exploring the world. British Journal of Develop-mental Psychology, 11, 61–74.

Woodward, A.L. (1998). Infants selectively encode the goalobject of an actor’s reach. Cognition, 69, 1–34.

Received: 15 February 2008Accepted: 1 March 2010



Copyright of Developmental Science is the property of Wiley-Blackwell and its content may not be copied or

emailed to multiple sites or posted to a listserv without the copyright holder's express written permission.

However, users may print, download, or email articles for individual use.

using mu rhythm desynchronization to measure mirror neuron

Documents