NeuroImage 104 (2015) 221–230

Contents lists available at ScienceDirect

NeuroImage

j ourna l homepage: www.e lsev ie r .com/ locate /yn img

Full Length Articles

Visual gravity cues in the interpretation of biological movements:neural correlates in humans

Vincenzo Maffei a,c,⁎, Iole Indovina a,c, Emiliano Macaluso d, Yuri P. Ivanenko c,Guy A. Orban e,f, Francesco Lacquaniti a,b,c

a Centre of Space BioMedicine, University of Rome Tor Vergata, via Montpellier 1, 00133 Rome, Italyb Department of Systems Medicine, University of Rome Tor Vergata, via Montpellier 1, 00133 Rome, Italyc Laboratory of Neuromotor Physiology, IRCCS Santa Lucia Foundation, via Ardeatina 306, 00179 Rome, Italyd Laboratory of NeuroImaging, IRCCS Santa Lucia Foundation, via Ardeatina 306, 00179 Rome, Italye Laboratorium voor Neuro- en Psychofysiologie, K.U. Leuven, Medical School, Campus Gasthuisberg, Herestraat 49/1021, BE-3000 Leuven, Belgiumf Department of Neuroscience University of Parma, Via Volturno 39, 43100 Parma, Italy

⁎ Corresponding author at: IRCCS Santa Lucia FoundaRome (Italy). Fax: +39 06 51501482.

E-mail address: v.maffei@hsantalucia.it (V. Maffei).

http://dx.doi.org/10.1016/j.neuroimage.2014.10.0061053-8119/© 2014 Elsevier Inc. All rights reserved.

a b s t r a c t

a r t i c l e i n f o

Article history:Accepted 4 October 2014Available online 12 October 2014

Keywords:Biological motionAction observationPredictive codeMismatch detectionDCMGravity

Our visual system takes into account the effects of Earth gravity to interpret biologicalmotion (BM), but the neuralsubstrates of this process remain unclear. Here wemeasured functional magnetic resonance (fMRI) signals whileparticipants viewed intact or scrambled stick-figure animations of walking, running, hopping, and skippingrecorded at normal or reduced gravity. We found that regions sensitive to BM configuration in the occipito-temporal cortex (OTC) were more active for reduced than normal gravity but with intact stimuli only. Effectiveconnectivity analysis suggests that predictive coding of gravity effects underlies BM interpretation. This processmight be implemented by a family of snapshot neurons involved in action monitoring.

© 2014 Elsevier Inc. All rights reserved.

Introduction

Subtle features of biological motion (BM) are perceived fromimpoverished visual stimuli, such as point-light or stick-figure animations(Johansson, 1973). The interpretation of these stimuli presumably de-pends on local motion cues from the limbs, as well as on configural cuesabout changing body shape (Bertenthal and Pinto, 1994; see Blake andShiffrar, 2007). These configural and motion cues, processed respectivelyin ventral and dorsal cortical pathways, probably are integrated in OTC(Giese and Poggio, 2003; Jastorff and Orban, 2009; Peelen et al., 2006).

BM interpretation may also depend on cues about the underlyingforces. Most movements are governed by Earth gravity, in addition tomuscle and inertial forces, and there is behavioral evidence that theeffects of gravity relative to the visual scene are taken into account inBM interpretation (Runeson and Frykholm, 1981). For instance, the rec-ognition of BM is strongly impaired when these stimuli are presentedupside down (the inversion effect) (Sumi, 1984). This effect is mainlydue to a reversal of the normal gravitational cues (Shipley, 2003;Westhoff and Troje, 2007). Moreover, gravitational cues allow to re-trieve the size of moving animals (Jokisch and Troje, 2003).

tion, via Ardeatina 306, 00179

Both themotion and configural parameters of BM rely on gravity. In-deed the time-course of elevation angles, hence the changes of BMshape over time, aswell as the foot swing and stance phases are affectedby the gravity level. This was shown by changing the effective gravityacting on the lower limbs during locomotion (Ivanenko et al., 2011;Sylos-Labini et al., 2013). Several findings show that the gravitationalsignals embedded in BM are critical for the analysis of local kinematics(Chang and Troje, 2009; Troje andWesthoff, 2006), and that configuralinformation may modulate this process (Bardi et al., 2014; Hirai et al.,2011). However, to date, it has not been investigated whether gravita-tional information contributes also to the analysis of BMconfigural cues.

Here we considered the possibility that gravity may guide the inter-pretation of BM configural cues. To address this issue, we measured thefMRI brain signals associated with different human gaits (walking,running, hopping, skipping) inwhich the body oscillates like a pendulumor moves ballistically under gravity (Cavagna et al., 1976). These gaitswere first recorded at two different effective gravity levels, that is, atnormal Earth's gravity (1g) or at simulated reducedgravity (16%of Earth'sgravity, or 0.16g, similar to theMoon gravity). These gaits were then ren-dered as intact or scrambled stick-figures and presented to naïve ob-servers. The rationale for using several different types of gait was that ofincreasing the chances of recording brain activity related to featuresshared across gaits, and of enhancing the BM activation signals (Jastorffand Orban, 2009). A crucial feature in our experimental approach was

222 V. Maffei et al. / NeuroImage 104 (2015) 221–230

that the overall changes in gait characteristics resulting from simulatedhypogravity are unique to this condition, and cannot be produced bymodifying muscle or inertial forces in normogravity (Donelan andKram, 1997; Ivanenko et al., 2002b, 2011; Sylos-Labini et al., 2013). Atthe same time, however, locomotion in hypogravity complies with thegeneral kinematic constraints of natural BM, i.e., the two-thirds powerlaw relating limb velocity to path curvature and the planar covarianceamong the changes in segment angles (Ivanenko et al., 2002a,2002b).

We tested whether gravity modulated the activity in the brain sitessensitive to the BM configuration, by examining conditions differing ingravity. These sites were identified by comparing intact vs. scrambledstimuli (Jastorff and Orban, 2009; Peelen et al., 2006). Further, byusing dynamic causal modeling (Friston et al., 2003), we explored themechanisms underlying the influence of gravity on BM configuralprocessing. It has been previously shown that the effect of gravity onobject- and ego-motion reveals a distributed network of cortical andsubcortical areas, whose primary sites are in TPJ (temporo-parietaljunction) and insula (Bosco et al., 2008; Indovina et al., 2005, 2013;Maffei et al., 2010; Miller et al., 2008). Thus, we investigated whetheran inter-regional connectivity between cortical areas sensitive to BMconfiguration (Jastorff and Orban, 2009; Peelen et al., 2006) and thosesensitive to gravity could mediate the interpretation of BM. An alterna-tive possibility is that an automatic and early process detecting gravitycues of BM (Bardi et al., 2014; Troje and Chang, 2013) occurs locally inregions selective to BM.

Materials and methods

Participants

Fifteen healthy subjects (6 females and 9 males, age 20−29 years),naïve to the experimental goals, gave written informed consent toparticipate in the fMRI experiments approved by the Ethics Committeeof Santa Lucia Foundation.

Stimuli and procedures

The stimuli were generated by recording 116 trials in 8 subjects (4females and 4 males, 25–45 years; not involved in the fMRI experi-ments) walking (22 trials), running (28), hopping (22) or skipping(44) on a treadmill at 3–6 km/h (general procedures as in Ivanenkoet al., 2011). In half of the trials, subjects were tilted on the coronalplane by 81° respect to the vertical, with each leg suspended in an inde-pendent exoskeleton (see Supplemental Information for more details).The resulting effective gravitational acceleration was 0.16g (similar toMoon gravity). Note that we do not claim of being able to reproduceexactly the locomotion on the Moon (that nobody has recorded so far,except for low quality videos of astronauts moving around in heavyspacesuits), but only of producing a locomotionwith an effective gravityon the lower limbs equal to 16% of normal gravity. In the remainingtrials, subjects stepped in the vertical position (normal gravity, 1g) atmatched speeds. The recorded movements of the thigh, shank and footwere then rendered in lateral view as white stick-figures (7° visual-angle (v.a.) in height) against a black background (Fig. 1a). We usedstick-figures because in a pilot psychophysical experiment with point-lights we found that observers often misinterpreted hip and knee jointsas head and hands, respectively. Moreover, stick-figures are consideredas appropriate as or even more appropriate than point-light displays toisolate the role of intrinsic motion by removing detailed informationabout body morphology (Troje, 2013). In addition to intact stick-figures, we created spatially-scrambled stimuli by changing the stickstarting position (Fig. 1a, see also Supplemental Information). This wasdone by adding a random shift to themiddle point of each stick segment.The amplitude of this shift ranged from 35 pixels to 180 pixels. The finalsize of scrambled configuration matched that of original (intact) stimuliin terms of height (300 pixels, 7° v.a.) andwidth (200 pixels, 4.6° v.a.) of

the final area covered by the moving stick. Thus the kinematics of eachscrambled segment was identical to that of the corresponding segmentfrom intact stimuli, with the overall movement excursion being alsomatched.

The stimuli were projected (1024×768 pixels, 60-Hz) on a translu-cent screen inside the scanner behind the participant's head, viewed viaa 45°-tilted mirror fixed on the head coil. In each trial, subjects fixated ared square (0.25° v.a.) at the screen center. The task was a simple two-alternative forced choice discrimination (Beauchamp et al., 2002, 2003),with subjects deciding if the stimulus contained normal or reduced grav-ity by pressing one of two buttons (fORP, Current Designs). To familiarizeparticipants with the stimuli and task, theywere trained two days beforefMRI, sitting in a chair and viewing a vertical display.

The experimental design was 2 × 2 factorial, crossing “configura-tion” (intact or scrambled) and “gravity” (normal or reduced gravity).Each stimulus condition (1g_intact, 1g_scrambled, 0.16g_intact,0.16g_scrambled) was presented 116 times, for a total of 464 trialspresented in six runs acquired for each participant. Each of the 6 runsincluded either 19 or 20 repetitions of each of the four conditions. Thestimuli (duration 2.0 ± 0.5 s, mean ± SD) were presented in pseudo-random order in a run, alternating rightward and leftward locomotionacross runs (balanced across subjects). Inter-trial intervals wererandomized with a long-tailed (geometric) distribution, resulting in amean onset asynchrony of 5.1 s (range 2.8–11.1 s). During fMRI exper-iments, eye movements were recorded (60 Hz) with an ASL 504 eye-tracking system (Bedford, MA).

Data acquisitions

MR images were acquired with a Siemens Magnetom Allegra 3 Thead-only scanning system (Siemens Medical Systems, Erlangen,Germany), equipped with a quadrature volume RF head coil. Participantswere provided with noise suppression equipment (ear plugs andheadphones) and lay supine with the head immobilized with foamcushioning. Whole brain BOLD echoplanar imaging (EPI) functionaldata were acquired with a 3 T-optimized gradient echo pulse-sequence(TR=2.47 s, TE=30ms; flip angle= 70°; FOV=192mm, fat suppres-sion). 38 slices of BOLD images (volume) were acquired in ascendingorder (64 × 64 voxels, 3 × 3 × 2.5 mm3, distance factor: 50%; inter-slicegap = 1.25 mm; slice thickness = 2.5 mm), covering the whole brain.For each participant, a total of 1032 volumes of functional data were ac-quired in six consecutive runs. At the end of each run (lasting 6' 30"),the acquisition was paused briefly.

fMRI data preprocessing

Data and statistical analyseswere performed using the SPM8 software(Wellcome Trust Centre for Neuroimaging, London, UK) implemented inMATLAB7.0 (TheMathWorks Inc., Natick,MA)using standardprocedures(Lindquist, 2008;Mikl et al., 2008). After discarding the first four volumesof each run, images were corrected for head movements, normalized tothe standard SPM8 EPI template in the Montreal Neurological Institute(MNI) space, resampled to 2 × 2 × 2 mm3 isotropic voxel size, and spa-tially smoothed with a full-width half-maximum (FWHM) isotropicGaussian kernel of 8 × 8 × 8 mm3. Voxel time series were processed toremove autocorrelation using a first-order autoregressive model andhigh-pass filtered (128-s cut-off, first-order autoregressive model).

fMRI analysis

Statistical analysis was performed in two stages. For every partici-pant, BOLD responses were estimated with a general linear model. Thefour experimental conditions were modeled with a boxcar function(1.5s duration, corresponding to the average response time) time-locked to the stimuli onset and convolved with SPM8 canonical hemo-dynamic response function. To control for potential differences in

Fig. 1.Visual stimuli. (a) Still-frames from intact and scrambled animations; the two images correspond to the same stick-figure. The elevation angle of the foot (α) is shown for illustrativepurposes only. (b) Time course of the changes ofα over the normalized gait cycle for different gaits. Normal gravity in orange, reduced gravity in cyan. Horizontal continuous and dottedlines denote duration of stance and swing, respectively. (c) Top: mean (±s.e.m.) phase-lag of α between normal- and reduced-gravity conditions. Bottom: mean difference (±s.e.m.) ofstance duration (as percentage of gait cycle) between normal- and reduced-gravity conditions. * Significant differences at p b 0.01 (t-value N 2.8, two-sample t-test) between normal- andreduced-gravity conditions.

223V. Maffei et al. / NeuroImage 104 (2015) 221–230

motion energy between conditions, we computed the net visual motionenergy of every trial, based on frame-by-frame displacements of thelimb segments (Kim et al., 2011). Motion energy and head movementparameters were then included in the fMRI analysis as covariates of nointerest. A second-level random effects analysis (ANOVA) modeled thefour conditions of interest plus the main effect of subject, usingnon-sphericity correction. Linear compounds (contrasts) were used tocompare the condition effects based on between-subjects variance.

We investigated the effect of gravity-cues on BM by testing for the in-teraction between “gravity” and “configuration” in the regions sensitiveto either BM configuration or gravity. First, we identified the regions ofinterest (ROIs) as areas activated by the intact BM stimuli (main ef-fect of configuration: [1g_intact + 0.16g_intact] N [1g_scrambled+ 0.16g_scrambled]) or the gravity-cues (main effect of gravity:[1g_intact + 1g_scrambled] N [0.16g_intact + 0.16g_scrambled]),

using SPM. To deal with multiple comparison problem (i.e., to controlfamily-wise error rate), we used a standard approach (Lindquist, 2008;Woo et al., 2014) based on cluster-extent thresholding (Friston et al.,1994). We initially established a voxel-level threshold to define clusters(i.e., groups of suprathreshold contiguous voxels). As recommended, weset this a-priori threshold at p-value b 0.001 (Woo et al., 2014), thedefault value in SPM. Depending on the size of each cluster, SPM8gives the probability that the specified cluster occurs by chance; namelythe p-value corrected for family-wise error rate (p-corr). In ourdataset, the number of voxels in the smallest cluster corresponding top-corr b 0.05 was k = 239, in accordance with theoretical predictions(Woo et al., 2014). Each of the clusters (11 clusters, see Table 1) thatwas significantly activated according to this procedure was used as asingle ROI. Within each of these ROIs, we then tested the interactionbetween “gravity” and “configuration” using MarsBar. To verify that

Table 1Stereotaxic position (MNI) of local maxima.

(a) Configuration (c) Interaction

x y z Z-value k p-corr (p-uncorr)

pITG 50 −70 −14 6.13 895 b0.002 (b0.001)MOG/IOG 44 −84 −14 5.23Fusiform g. 42 −46 −26 4.6

(b) GravityMOG −22 −102 10 5.86 938 n.s. (0.06)

Lingual g. −14 −94 −10 4.55IOG/Lingual g. −30 −96 −14 3.35

Lingual g. 22 −96 −4 4.32 760 n.s (0.29)SOG 16 −96 18 4.26

Putamen −28 −22 4 5.13 1006 n.s. (0.11)Insula −40 2 −8 3.49

Hippocampal −28 −6 −16 4.16Putamen 32 −10 −8 5.32 2340 n.s. (0.12)Insula 48 −12 2 3.99

Hippocampal 28 −10 −16 4.58MTG −62 −42 −12 4.55 608 n.s (0.23)TPJ −56 −52 32 5.39 1103 n.s. (0.11)TPJ 52 −56 28 4.56 1101 n.s. (0.71)

64 −48 40 4.35IFG (BA11) −26 34 −18 4.3 245 n.s. (0.12)MFG (BA46) 36 18 42 4.54 921 n.s. (0.52)SFG/MFG 18 48 30 4.74 239 n.s. (0.35)

(a,b): Brain regions significantly more activated by (a) intact than scrambled and by (b) 1gthan0.16g stimuli (p-corrb 0.05, corrected formultiple comparisons at cluster level). Stereo-taxic position of local maxima (MNI space) are reported together with their respective Z-values (in bold). Italics indicate cluster size (k) in voxels. (c): The interaction “configuration”× “gravity”was significant only in the ROI selective for intact stimuli. Results reported for theROIs are corrected for multiple comparisons (uncorrected p-values are in brackets, n.s.: notsignificant). pITG: post. inferior temporal gyrus; IOG: inferior occipital gyrus; MOG: middleoccipital gyrus; SOG: superior occipital gyrus; MTG: middle temporal gyrus; TPJ: temporo-parietal junction; IFG: inferior frontal gyrus;MFG:middle frontal gyrus; SFG: superior frontalgyrus.

224 V. Maffei et al. / NeuroImage 104 (2015) 221–230

the statistical test was unbiased from the selection criteria used todefine the ROIs, we estimated the orthogonality of the correspondingcontrast-vectors taking into account the design matrix (X), as de-scribed in Kriegeskorte et al. (2009). Since temporal dependence isnot a concern in second-level between subjects analysis, we usedthe criterion: CselectionT · (XTX)−1 · Ctest = 0. Here, Cselection

were the contrast vectors corresponding to the main effects of “grav-ity” and “configuration”, whileCtest was the contrast vector of the inter-action. This product was equal to zero (i.e. orthogonal contrast-vectors)for both CgravityT (XTX)−1Ctest andCconfigurationT (XTX)−1Ctest demonstratingthat the interaction (statistical test) and the main effects (selectioncriteria) were independent, thus excluding any circularity in theanalysis.

In addition, we also assessed the interaction between “gravity” and“configuration” by considering five regions that previous studies associ-ated with specific aspects of BM processing (Jastorff and Orban, 2009).The five spherical ROIs (8 mm radius, i.e. the same size as theFWHM kernel of the smoothing procedure) included three regionssensitive to BM configural cues (LOS/MOG [x,y,z = 35 −89 −7],pITG [44 −68−22], midFus [41 −49 −21]), and two regions sensitiveto BM kinematic cues (pITS [51 −75 −2], pSTS/STG [63 −43 10]).Note that, in a previous study, midFus and pITS responded also to the in-teraction between kinematics and configuration (Jastorff and Orban,2009). The center of each ROI (black circle in Fig. 2b) was the center-of-mass (average coordinates) of all sites classified in the specific region(Table 1 exp 1–3 in Jastorff and Orban, 2009). This analysis circumventedissues of circularity, since the ROIs were defined by an independent dataset (Kriegeskorte et al., 2009).

MarsBar statistical analysis makes use of the ROI average signal ex-tracted from each subject. Thus a single statistical test is calculated foreach ROI and contrast of interest. As a consequence, to prevent themul-tiple comparisons problem, the p-values reported were FDR-corrected

for the number of ROIs: 11 clusters activated in the present study (seeTable 1) and 5 ROIs selected from Jastorff and Orban (2009), Ntotal = 16.

Control analysis (1)

In the main analysis we included the motion energy as covariate ofno interest. To further check that the principal results were not affectedby motion energy, we separately analysed the subset of data (25%)matched for motion energy between 1g and 0.16g conditions. Thestatistical models were identical to those of the main dataset, exceptthatmotion energywas not included as a co-variate. Main effects andinteractions were calculated in each functional ROI using MarsBar ROItoolbox and the results were FDR-corrected for the number of ROIs.

Control analysis (2)

We assessedwhether unspecific changes in kinematic parameters oflocomotion, independent of changes in gravity, might also reproduceour fMRI results. Thus, a simple increase in the speed of locomotion atEarth's gravity is normally associated with corresponding changes inlimb kinematics, such as a shortening of the stance phase (Ivanenkoet al., 2002a; Osaki et al., 2008) although these changes are distinctlydifferent from those associated with a change of gravity (Sylos-Labiniet al., 2013) as remarked above. To this aim, we ordered the 1 g trials(intact and scrambled) on the basis of decreasing locomotion speed.The first 58 trials (50% intact and 50% scrambled) were classified asfast, the remaining 58 trials as slow: the average speed of slow trialswas 4.15 ± 1.57 km/h, while that of fast trials was 5.07 ± 1.32 km/h.Such a difference in average locomotion speed was associated with asignificant difference in motion energy (t(56) = 8.5, p b 0.001), andalso in motion trajectories (Ivanenko et al., 2002a; Osaki et al., 2008),as indicated by a significant increase in stride length (slow trials:57 ± 19 cm; fast trials: 70 ± 13 cm; t(56) = 2.32, p-value = 0.029),and foot vertical excursion (slow trials: 26 ± 3 cm; fast trials: 34 ±8 cm; t(56) = 4.67, p-value b0.001). At the first level of fMRI analysis,we re-analysed the fMRI data using the additional factor “speed” (slowor fast). All the remaining analyses were identical to those of the mainanalyses described above, except that motion energy was not includedas a covariate. At the second level of fMRI analysis, we imported the im-ages corresponding to the factors: intact slow, intact fast, scrambled slowand scrambled fast, calculated exclusively on 1g data. We tested themain effect of “speed”, main effect of “configuration” and the interaction“speed” × “configuration” in each functional ROI using MarsBar ROItoolbox and the results were FDR-corrected for the number of ROIs.

Dynamic causal modelling (DCM)

Themain result of the intra-regional analyseswas a significant inter-action in the region selective to configuration cues (i.e. OTC), whereactivity was higher for 0.16g_intact than 1g_intact. This pattern maysuggest an error detectionmechanism, as formulated by predictive cod-ing theory (Friston, 2005; Kilner et al., 2007; Rao and Ballard, 1999;Schultz et al., 1997). Within this framework, we hypothesised that theeffect of gravity on biological motion would be mediated by inter-regional influences between the brain regions involved in our task.We explored this using Dynamic Causal Modelling (Friston et al.,2003). In order to constrain the number of possible models we consid-ered a subset of 3 regions: 1) OTC (pITG, center at [50 −70 −14]);2) posterior insula ([48 −12 2]); 3) temporo-parietal junction (TPJ,[52−56 28]). The last two regions, belonging to the peri-sylvian cortex,have consistently been shown to be involved in the visual analysis ofgravity (Bosco et al., 2008; Indovina et al., 2005, 2013; Maffei et al.,2010; Miller et al., 2008).

Since DCMs are fitted to subject-specific BOLD time series, we usedfunctional constraints to ensure the comparability of our models acrosssubjects. Accordingly, the subject-specific maps of the effects of interest

Fig. 2. Statistical parametric mapping of GLM analysis. (a) Main effects (p-corr b 0.05) of “gravity” (red) and “configuration” (green). Maps are projected onto inflated surfaces of thehuman PALS atlas (Caret). Meta-analysis of regions selective for intact BM is outlined in blue (Jastorff and Orban, 2009). The area enclosed in the rectangle is zoomed in (b) and plottedonto the flat map of human PALS atlas. (b) Black circles represent spherical ROIs associated to BM configural cues (LOS, pITG, midFus) and to BM kinematic cues (pITS, pSTS/STG). Forillustrative purpose, panel (c) shows the estimated activity (±s.e.m. between-subjects) of the indicated brain sites (orange circle in panel a–b, see Table 1) for the four types of stimuli.Note that the peak localization and estimated activity were calculated from the same dataset showed in panels a and b, thus they were not mutually independent. OTC: occipito-temporalcortex; OTS: occipito-temporal sulcus, TPJ: temporo-parietal junction; Fus.: fusiform gyrus; pITS: posterior-inferior temporal sulcus; pITG: posterior-inferior temporal gyrus; pSTS/STGposterior-superior temporal sulcus/gyrus; LOS: lateral occipital sulcus.

225V. Maffei et al. / NeuroImage 104 (2015) 221–230

(SPM-F) were thresholded at p-uncorr b 0.05. For each of the threeregions of interest, all voxels within 8 mm of the activation maximadefined at group level (Table 1) were chosen.

A set of 80 models was specified by changing the structure ofconnections (and their modulation) between OTC (i.e. pITG) and peri-sylvian regions (i.e. posterior insula and TPJ) in all possible ways. Allmodels had the effect of “intact configuration” as driving input for OTCand of “Earth (1g)” as driving input for the peri-sylvian regions. Connec-tions weremodulated by the effect of “gravity on intact stimuli” (1g_in-tactminus 0.16g_intact).Modelswere classified in 4 families dependingon the direction of the modulatory connectivity between pITG and theperi-sylvian regions (Fig. 3a–d): the F1-family had the forwardmodula-tory connectivity only (pITG to peri-sylvian regions, 20models); the F2-family had the backwardmodulatory connectivity only (peri-sylvian re-gions to pITG, 20 models); F3-family had both forward and backwardmodulatory connectivity (25 models); and finally the F0-family (15models) had only a within-region (local) modulation in pITG but nei-ther forward nor backward modulated connectivity (Ewbank et al.,2011). The family F0 was included to assess the alternative hypothesisthat (local) mechanisms within pITG mediated the detection of thegravitational cues. The bayesian model selection (BMS) implementedinDCM10 (SPM8)was used to compute the group evidence and to iden-tify the preferred models (Stephan et al., 2010).

Results

Behavior

Consistent with previous results (Ivanenko et al., 2011), we foundthat the characteristics of all gaits performed at reduced gravity(0.16g) differed slightly but systematically from kinematics at normalgravity (1g) at matched speeds (Fig. 1b-c, see Supplemental Table 1).Thewaveforms of the angularmotion of the limb segmentswere similarfor both conditions, but 0.16g involved significantly shorter stanceduration and phase of foot elevation relative to 1g (two sample t-test,all t-values N 2.8, all p-values b 0.01, see also Supplemental Table 1).

Observers discriminated accurately whether locomotion was at 1gor 0.16g, for both intact (training: d′ = 3.4 ± 0.95 SD; fMRI: d′ =3.6 ± 0.92) and scrambled stimuli (training d′ = 3.0 ± 0.79; fMRId′ = 3.2 ± 1.09), but did so significantly better with intact stimuli (allpaired t-test, p-value b 0.02, t(14) N 2.73). The difference betweenfMRI day and training day was not significant for either intact(p-value = 0.21, t(14) = 1.13) or scrambled stimuli (p-value = 0.27,t(14) = 1.3).

The occurrence of learning processes during the experimentwas un-likely, since feedback of performance was not provided to subjects.However, to directly assess this possibility, we compared the number

Fig. 3. Four families of dynamic causal models. (a) F0 with only a local modulatory connectivity in OTC; (b) F1 with forward modulatory connectivity; (c) F2 with backward modulatoryconnectivity; (d) F3 with forward and backward modulatory connectivity. The red outline denotes the winning family and the (e) best-fitting model with positive forward and negativebackward modulatory connectivity (see results). Continuous arrows: fixed connections. Dotted arrows: connections not modulated/absent. Black arrows: modulatory effect on fixedconnections. Grey arrows: driving inputs. I×G: Intact × (1 g − 0.16 g).

226 V. Maffei et al. / NeuroImage 104 (2015) 221–230

of errors between the first and the last session. We did not find anysignificant difference between session 1 (mean error ± SD: 6.8 ± 5.02)and session 6 (mean error ± SD: 6 ± 4.88) of the training day(p-value = 0.50, t(14) = 0.68), between session 6 of training dayand session 1 (6.8 ± 5.22) of fMRI experiment (p-value = 0.54,t(14) = 0.63), or between session 1 and session 6 (7.3 ± 6.81) offMRI experiment (p-value = 0.74, t(14) = 0.33). These results indicatethat there was no evident learning across sessions/days.

We separately analysed a sub-sample of trials (25%) matched formotion energy in the normal and reduced gravity conditions. Intactstimuli were discriminated better than scrambled also in the motion-energy matched sample (paired t-test, p-value b 0.05, t(14) = 2.15).

Analyses of eye-movements confirmed that participants maintainedwell fixation in the scanner, as shown by the low (3.4± 2.6 SD) averagenumber of saccades (amplitude N 2°, duration N 100 ms) per minute,without any significant effect of gravity, configuration or their interac-tion (ANOVA, all p-value N 0.1, F-value b 2.8).

fMRI – ROI definition

Regions sensitive to BM configural cuesThe main effect of configuration revealed significant activation of

OTC (encompassing pITG, Fusiform gyrus and IOG/MOG) in the righthemisphere (p-corr b 0.05, green in Fig. 2a–b, Table 1a). Instead theleft hemisphere showed only a statistical trend toward significance(p-uncorr b 0.001); peak at [−44 −84 −12]; Z = 5.09), consistentwith previous studies reporting hemispheric asymmetries (Grossmanand Blake, 2002; Jastorff and Orban, 2009; Peelen et al., 2006).

Regions sensitive to gravity cues

The main effect of gravity revealed significant activation of bilateralTPJ and insular cortex (i.e. pery-sylvian cortex). Note that the peaks intemporo-parietal junction were more rostral than the TPJ sites relatedto false belief (see Schurz et al., 2013). Additionally we found significantactivations in frontal, lingual and occipital cortex, hippocampal regionsand putamen (p-corr b 0.05, red in Fig. 2a, signal plots in Fig. 2c,Table 1b). These activations largely overlap with those previously asso-ciated with the perception of object motion (Indovina et al., 2005;Maffei et al., 2010; Miller et al., 2008) or ego-motion (Indovina et al.,2013) at normal gravity.

fMRI – ROIs analysis

The primary scope of the present study was to verify how gravityaffects the analysis of BM configural cues. We investigated this issueby looking for an interplay between configuration and gravityfirstwithinthe ROI identified from the main effect of either configuration or gravity(see Methods). This analysis showed a significant interaction in theROI related to configuration, namely the right OTC (p-corr b 0.05,corrected for the total number of ROIs; Fig. 2a–b). Critically, the pat-tern of activity in this area showed a greater BOLD signal for locomo-tion in hypogravity than in normal gravity (see for instance theactivity plot of pITG in Fig. 2c). By contrast, none of the ROIs relatedto normal gravity (main effect of gravity, Table 1b) showed signifi-cant interactions, even at a lenient threshold (p-uncorr = 0.05,Table 1c, see also Supplemental Fig. 2).

227V. Maffei et al. / NeuroImage 104 (2015) 221–230

Together with regions activated in the current experiment, we alsotested the interaction between configuration and gravity in five ROIsthat previouswork associatedwith the processing of BM shape andmo-tion (see Methods). Coherently with the analysis using ROIs derivedfrom the present experiment, we found that the interaction was signif-icant in the two ventral ROIs belonging to OTC sensitive mainly toconfigural cues (i.e. pITG and midFG) and in pITS (all p-corr b 0.01,see Fig. 2b), but not in pSTS/STG and LOS (all p-uncorr N 0.12).

To test whether the four different gaits may have contributed differ-ently to the interaction effect in these 3 ROIs (see above),we built a newfirst level design matrix nowmodeling also the factor of “gait”. This fac-tor included 4 levels corresponding to walk, run, hop and skip. At thesingle-subject level, the interaction contrast was calculated separatelyfor each gait and the corresponding four contrast-images for eachsubject were assessed in a second-level (RFX) model. OmnibusF-contrasts tested for any difference between the interaction effectsacross the 4 gaits. These tests did not reveal any significant effect inour ROIs (p-values = 0.31, 0.32 and 0.08 respectively for pITG, midFusand pITS, suggesting that all four gaits contributed to the overall interac-tion effect.

Control analysis (1)

Differences in the net visual motion energy between conditionscould hardly account for our results, becausemotion energywas includedin themodel of brain responses reported so far (seeMethods). To double-check, we performed a control analysis based only on the sub-sample oftrials exactly matched for motion energy between 1g and 0.16g condi-tions. This analysis replicated the main effect of gravity previously re-ported in all regions of the gravity network (all p-corr b 0.05), exceptfor MTGwhose activation was not significant after correction for multi-ple comparisons (p-uncorr = 0.02). Moreover, in none of these regionswas the interaction significant, even at a low uncorrected threshold ofp-uncorr ≤ 0.1 (but p-uncorr = 0.04 for MOG and p-uncorr = 0.03for the insulae). The right OTC exhibited both a significant main effectof configuration and a significant interaction between configurationand gravity (all p-corr b 0.05). These results further confirm that ourprincipal findings do not depend on differences in motion energy be-tween conditions.

Control analysis (2)

We checked whether unspecific changes in kinematic parameters oflocomotion, unrelated to the gravity level,might reproduce ourfindings,in particular the interaction in OTC. To this aim, we divided the 1g trialsin two groups of trials, slow or fast trials on the basis of locomotion speed(see Methods). We re-analysed the fMRI data using “configuration”(intact or scrambled) and “speed” (slow or fast) as factors. As a conse-quence of speed variation, the motion energy, stride length and motiontrajectorieswere also different (Ivanenko et al, 2002a; Osaki et al, 2008).The main effect of “configuration” was significant only in OTC (p-corrb0.05), while themain effect of “speed” (fast vs slow speed)was not sig-nificant (p-uncorr N 0.27), except for the left MOG and right LingualGyrus ROIs belonging to lower-level visual areas (p-corr b 0.05). Impor-tantly, the interaction “speed” × “configuration”was not significant (allp-uncorr N 0.17) in any ROI. Thus, our key findings were not related to asimple kinematic manipulation such as that associated with speedchanges, but theywere specifically associatedwith a change in the grav-ity level.

DCM

The activation pattern of pITG in OTC showed increased responses tointact stimuli when movements were not compatible with normalgravity. This result is compatible with an error detection process. Inthe framework of predictive coding, this may entail inter-regional

signalling between high-order regions where the expected stimuliare represented and lower-order regions that detect the mismatchbetween incoming and expected stimuli. The activity in lower-levelareas is mediated by backward connections from higher-level areas,while the internal representation of the environment (i.e. expectation)can be revised by error signals forwarded from lower-level areas(Egner et al., 2010; Friston, 2005; Kilner et al., 2007; Rahnev et al.,2011; Rao and Ballard, 1999).

We used Dynamic Causal Modelling (Friston et al., 2003) to investi-gate whether the pattern of activity in OTC (i.e. pITG) can be explainedby a modulation of the inter-regional connectivity (forward, backwardor both) between OTC and peri-sylvian regions. Eighty models weresubdivided in four different families depending on the connectionsmodulated (Fig. 3a–d). Our DCMs also included an alternative hypothe-sis involving only intra-regional (i.e. local) effectswithinOTC, cf. modelsin family F0 (Fig. 3d and Supplemental Information).

We found that family F3 accounted for 0.99 of the posterior proba-bility. All models in family F3 had a modulation of the bi-directionalconnection between OTC and pery-silvian regions (Fig. 3c), and noneof these models included a modulation of the intra-regional (i.e. local)connectivity. In particular threemodels belonging to family F3 (posteri-or probability of 0.53, 0.34 and 0.12 respectively for model 1, model 2and model 3) fitted the data significantly better compared to all otherremaining models (Bayesian factors: model 1 N4.1; model 2 N3.6; andmodel 3 N2.6; see also Supplemental Information). Model 1 had fixedbi-directional connection between OTC and insula and between insulaand TPJ. Model 2 and Model 3 had in addition a bi-directional connec-tion between OTC and TPJ. Importantly, all these models belongingto family F3 had the bi-directional connection between OTC andpery-silvian regionsmodulated in a similar way: (1) a backward con-nection from peri-sylvian regions to OTCmodulated negatively and (2) aforward connection fromOTC to peri-sylvian regionsmodulatedpositive-ly (e.g. Fig. 3d–e, see also Supplemental Table 2 and Supplemental Fig. 3).This finding was in accordance with theoretical model of predictive codedescribing a functional asymmetry between the forward connections(driving error-related signals to higher regions) and backward connec-tions (delivering signals about predictions to lower regions) (Friston,2005; Park and Friston, 2013).

Discussion

In this studywe found that gravitymodulated the activity evoked byBM configural cues in the occipito-temporal cortex. Our findings sup-port the idea of a crucial role of gravity in the interpretation of biologicalmovements. Furthermore, both the pattern of BOLD signal and theconnectivity analysis suggest that a predictive code is involved in thisprocess, although it is probably not the only mechanism. Indeed gravityinfluences differently the processing of localmotion and configural cues.The former process seems to take place via an automatic bottom-upmechanism (Troje and Chang, 2013), while our study suggests that thelatter process occurs via a top-down mechanism. However, notice thatour protocol did not address the possible interaction between gravityand local motion cues in BM.

Behavior

Behavioral results showed that the observers discriminated accu-rately locomotion performed at normal versus reduced gravity,confirming previous evidence that gravity effects are taken into accountin the analysis of BM (Troje and Chang, 2013). An alternative possibilityis that they did not use the gravity cues, but during training they simplylearned to discriminate two different categories of visual stimuli. How-ever we deem this unlikely. First, it should be remarked that we neverprovided any feedback of performance (during training or fMRI),which is typically required for learning. Moreover, the finding that

228 V. Maffei et al. / NeuroImage 104 (2015) 221–230

discrimination performance did not differ significantly between ses-sions/days argues against the occurrence of a learning process.

Finally, the discrimination was accurate both when the observersand the display were vertical during training and when they werehorizontal in the scanner, indicating that the task could be accom-plished in the reference frame of the visual scene irrespective of the sub-ject orientation relative to physical gravity (Chang et al., 2010).

Gravity may bias the activity of snapshot neurons

fMRI results showed that gravity cues affect the processing of BMin two sets of regions: those involved mainly in processing of BMconfiguration cues belonging to OTC (pITG and midFus), and thosein which kinematic cues interact with configuration cues in the dor-sal stream (i.e. pITS).

The right OTC responded to intact stimuli. This area has been shownto be activated by BM in several previous studies (Han et al., 2013;Peelen et al., 2006; Peuskens et al., 2005; Saygin et al., 2004; Vainaet al., 2001). Note that the extent of activation elicited by the effect ofconfiguration in our study is smaller than that showed previously (forinstance compare Fig. 2 in the present study with Fig. 3 in Jastorff andOrban, 2009). This could be due to residual configural cues present inthe scrambled stimuli (so that the configural effect is underestimated)or it might depend on the fact that our stimuli showed only the lowerbody segments.

Jastorff and Orban (2009) have shown that this region is dominatedby the main effect of configuration in BM. Subsequent studies (Jastorffet al., 2012) have provided evidence that this ventral branch of the BMactivation, processing the shape information contained in BM stimuli,corresponds to the lower bank of the STS of the monkey. According toan influential model (Giese and Poggio, 2003), BM pattern detectorstemporally smooth and summate the activity of snapshot neurons,which are thought to process shape (Beintema and Lappe, 2002;Lange and Lappe, 2006) together with the activity of optic flow patternneurons which are thought to process local motion. Recent single cellstudies in monkeys have localized snapshot action-selective neuronsin the lower bank of STS. The response of some snapshot neurons de-pends on the sequential temporal order, so that individual neuronsfire strongly only if the corresponding body shapes occur in the correcttemporal context (Singer and Sheinberg, 2010; Vangeneugden et al.,2011). The evidence provided by Jastorff et al. (2012) suggests thatthe activation obtained in the configuration main effect reflects the ac-tivity of these snapshot neurons. This is further supported by the factthat most locomotion-selective STS neurons respond to the lower halfof the body (Vangeneugden et al., 2011) which we used in our stimuli.Our results suggest that snapshot neurons respond more to intactthan to scrambled stimuli, but more so for reduced than normal gravity.

The recognition of BM is strongly impaired when stimuli aredisplayed upside down (Sumi, 1984). This phenomenon has been relatedto the sensitivity of the visual system to gravity (Pavlova and Sokolov,2000). Sokolov et al. (2010) noted that stimuli displayed in upright andupside down (inverted display) retain the same relational structure andamount of motion. The main difference between upright and invertedBM stimuli is related to kinematics. The kinematics of upright stimuli iscongruent with walking under natural conditions, while, in the case ofinverted BM, kinematics violates physical gravity (Pavlova et al., 2004;Troje and Westhoff, 2006). In previous studies, a sensitivity to inversioneffect has been found in STS (Grossman and Blake, 2001), and also inoccipito-temporal cortex (Jokisch et al., 2005; Pavlova et al., 2004;Peuskens et al., 2005). In particular, an inverted point-light walker(i.e. a stimulus with a non-gravitational kinematics) elicits greater activ-ity in OTC than normal BM stimulus (Grezes et al., 2001). This findingappears consistent with the present results, namely that OTC respondsto BM stimuli under a condition of abnormal gravity.

In addition to OTC, the interaction was significant also in pITS, a sitelinking kinematic information with body representation in the dorsal

stream (Jastorff and Orban, 2009). This result supports previous behav-ioral finding showing a gravity effect on the analysis of BM local kine-matics (Bardi et al., 2014; Hirai et al., 2011; Troje and Westhoff, 2006).

Predictive mechanisms may underlie BM interpretation

Remarkably, we found that the activity elicited by intact stimuli washigher for hypogravity than for normal gravity conditions in OTC foci. In asimilar vein, recent studies found stronger responses in OTC for observa-tion of extraordinary/irrational actions compared to ordinary actions (deLange et al., 2008; Jastorff et al., 2011; Maffei et al., in press). This patterndoes not fit well with the possibility that the neural activity in pITG isdriven mainly by feature selectivity (here, normal gravity). Instead, itseems more compatible with mismatch detection mechanisms, such asthose of the predictive coding model (Friston, 2005; Kilner et al., 2007;Rao and Ballard, 1999). The results of DCM analysis support this hypoth-esis. Predictive coding theory posits that the brain is organized in hierar-chical levels interacting with each other. At each cortical level, there arerepresentational units, receiving a prediction about the incoming stimulifrom higher areas, and error units encoding differences between predict-ed andmeasured stimuli (Rao andBallard, 1999; Summerfield and Egner,2009). The error units are supposed to contribute more strongly to theBOLD response than prediction units (Egner et al., 2010). Our results sug-gest that the error units aremore activewhen viewing locomotion underhypogravity than normal gravity. Given the strong sensitivity of a sub-population of snapshot neurons to the temporal context (Singer andSheinberg, 2010; Vangeneugden et al., 2011), phase shifts resultingfrom reduced gravity (Fig. 1c) should be easily detected by these putativeerror units. Notably, similarmechanisms have been previously suggestedto operate in action observation network when a mismatch emerges be-tween appearance (human) and motion (not biological) (Cross et al.,2012; Saygin et al., 2012).

DCM shown an asymmetric modulatory effect of gravity on back-ward and forward connectivity. Similarly, in a previous study (Rahnevet al., 2011) it has been shown that expectation modulated positivelythe forward connection and negatively the backward connection be-tween a sensory region (MT+) and an higher level region. This scheme,coherent with predictive code (Friston, 2005; Park and Friston, 2013),seems underlie the change of prediction in high-level regions (Rahnevet al., 2011). In particular, the finding of a negative modulatory effectof gravity on backward connection (inducing an enhancement in OTCactivity during locomotion at reduced gravity) may represent the top-down signal sent toward the error units and required to detect the mis-match between expected and measured gravity (Summerfield andEgner, 2009). Conversely, themodulation of gravity on the forward con-nection had the opposite sign. This modulation may reflect the signalforwarded to the prediction units in higher-level area. So the occurrenceof an unexpected stimuli (e.g. intact at reduced gravity) elicits a de-crease of activity (i.e. inhibition) in peri-sylvian regions according totheoretical model. Indeed predictive coding supposes that error unitsmodulate the response of predictor units by inhibiting the neuronsmaking erroneous predictions (Koster-Hale and Saxe, 2013), so thatthe internal representation of statistical regularities of the environmentcan be updated.

Regions sensitive to gravitational kinematics

The brain regions sensitive to gravity for both intact and scrambledstimuli included peri-sylvian areas (temporo-parietal and insularcortices), frontal and occipital cortex, hippocampus and putamen.These regions do not overlap with those activated by the contrast ofbiological movements with rigid translation of bodily configurations(Jastorff and Orban, 2009) or the contrast of biological with artificial(robotic) motion (Saygin et al., 2012), consistent with the fact thatall our stimuli involved biological kinematics recorded from actuallocomotion.

229V. Maffei et al. / NeuroImage 104 (2015) 221–230

It might be argued that the activations associated with gravity mayreflect somedifference in task load. In such case, since d-primewas betterfor intact than scrambled stimuli, we should expect also an effect of con-figuration in these same areas. Instead the effect of configuration and theinteraction were not significant in any of the regions showing a maineffect of gravity. Thus our data argue against this possibility. Instead, theregions sensitive to normal gravity largely overlap with those previouslyassociated with the perception of object motion (Indovina et al., 2005;Maffei et al., 2010; Miller et al., 2008) or self-motion (Indovina et al.,2013) under normal gravity.We hypothesize that these regions internal-ize the effects of Earth gravity on visual motion in general.

Conclusion

We suggest that gravity affects BM processing in two ways. In addi-tion to an early automatic and evolutionary ancient process that facili-tates local kinematics analysis (Johnson, 2006; Troje and Chang, 2013;Vallortigara and Regolin, 2006), gravitational cues embedded in locomo-tion may also be involved in the interpretation of configural cues. Thelatter process, according to previous hypothesis about the mechanismsunderlying action understanding (Kilner et al., 2007; Koster-Hale andSaxe, 2013), seems rely on a predictive code. In particular the finding ofa modulatory connections between peri-sylvian regions and OTC givesnew insights on how biological movements interpretation may work atearly stages of brain processing.

Supplementary data to this article and representative movies of thestimuli can be found online at http://dx.doi.org/10.1016/j.neuroimage.2014.10.006.

Acknowledgments

The authors declare no competing financial interests. The workwas supported by grants from the Italian Ministry of Health (RC andRF-10.057), Italian Ministry of University and Research (PRIN grant2010MEFNF7_002), and Italian Space Agency (CRUSOE andCOREA). We thank Dr. Francesca Sylos-Labini for help in collectingkinematic data for the movies.

References

Bardi, L., Regolin, L., Simion, F., 2014. The first time ever I saw your feet: inversion effect innewborns' sensitivity to biological motion. Dev. Psychol. 50, 986–993.

Beauchamp, M.S., Lee, K.E., Haxby, J.V., Martin, A., 2002. Parallel visual motion processingstreams for manipulable objects and human movements. Neuron 34, 149–159.

Beauchamp, M.S., Lee, K.E., Haxby, J.V., Martin, A., 2003. FMRI responses to video andpoint-light displays of moving humans and manipulable objects. J. Cogn. Neurosci.15, 991–1001.

Beintema, J.A., Lappe, M., 2002. Perception of biological motion without local imagemotion. Proc. Natl. Acad. Sci. U. S. A. 99, 5661–5663.

Bertenthal, B., Pinto, J., 1994. Global processing of biological motion. Psychol. Sci. 5,221–225.

Blake, R., Shiffrar, M., 2007. Perception of human motion. Annu. Rev. Psychol. 58, 47–73.Bosco, G., Carrozzo, M., Lacquaniti, F., 2008. Contributions of the human temporoparietal

junction andMT/V5+ to the timing of interception revealed by transcranial magneticstimulation. J. Neurosci. 28, 12071–12084.

Cavagna, G.A., Thys, H., Zamboni, A., 1976. The sources of external work in level walkingand running. J. Physiol. 262, 639–657.

Chang, D.H., Troje, N.F., 2009. Acceleration carries the local inversion effect in biologicalmotion perception. J. Vis. 9 (19), 11–17.

Chang, D.H., Harris, L.R., Troje, N.F., 2010. Frames of reference for biological motion andface perception. J. Vis. 10, 22.

Cross, E.S., Liepelt, R., Hamilton, A.F., Parkinson, J., Ramsey, R., Stadler, W., Prinz, W., 2012.Robotic movement preferentially engages the action observation network. Hum.Brain Mapp. 33, 2238–2254.

de Lange, F.P., Spronk, M., Willems, R.M., Toni, I., Bekkering, H., 2008. Complementary sys-tems for understanding action intentions. Curr. Biol. 18, 454–457.

Donelan, J.M., Kram, R., 1997. The effect of reduced gravity on the kinematics of humanwalking: a test of the dynamic similarity hypothesis for locomotion. J. Exp. Biol.200, 3193–3201.

Egner, T., Monti, J.M., Summerfield, C., 2010. Expectation and surprise determine neuralpopulation responses in the ventral visual stream. J. Neurosci. 30, 16601–16608.

Ewbank, M.P., Lawson, R.P., Henson, R.N., Rowe, J.B., Passamonti, L., Calder, A.J., 2011.Changes in “top-down” connectivity underlie repetition suppression in the ventralvisual pathway. J. Neurosci. 31, 5635–5642.

Friston, K., 2005. A theory of cortical responses. Philos. Trans. R. Soc. Lond. B Biol. Sci. 360,815–836.

Friston, K.J., Harrison, L., Penny, W., 2003. Dynamic causal modelling. Neuroimage 19,1273–1302.

Friston, K.J.,Worsley, K.J., Frackowiak, R.S., Mazziotta, J.C., Evans, A.C., 1994. Assessing the sig-nificance of focal activations using their spatial extent. Hum. Brain Mapp 1, 210–220.

Giese, M.A., Poggio, T., 2003. Neural mechanisms for the recognition of biological move-ments. Nat. Rev. Neurosci. 4, 179–192.

Grezes, J., Fonlupt, P., Bertenthal, B., Delon-Martin, C., Segebarth, C., Decety, J., 2001. Doesperception of biological motion rely on specific brain regions? Neuroimage 13,775–785.

Grossman, E.D., Blake, R., 2001. Brain activity evoked by inverted and imagined biologicalmotion. Vis. Res. 41, 1475–1482.

Grossman, E.D., Blake, R., 2002. Brain areas active during visual perception of biologicalmotion. Neuron 35, 1167–1175.

Han, Z., Bi, Y., Chen, J., Chen, Q., He, Y., Caramazza, A., 2013. Distinct regions of right temporalcortex are associated with biological and human-agent motion: functional magneticresonance imaging and neuropsychological evidence. J. Neurosci. 33, 15442–15453.

Hirai, M., Chang, D.H., Saunders, D.R., Troje, N.F., 2011. Body configuration modulates theusage of local cues to direction in biological-motion perception. Psychol. Sci. 22,1543–1549.

Indovina, I., Maffei, V., Bosco, G., Zago, M., Macaluso, E., Lacquaniti, F., 2005. Representationof visual gravitational motion in the human vestibular cortex. Science 308, 416–419.

Indovina, I., Maffei, V., Pauwels, K., Macaluso, E., Orban, G.A., Lacquaniti, F., 2013. Simulatedself-motion in a visual gravity field: Sensitivity to vertical and horizontal heading inthe human brain. Neuroimage 71, 114–124.

Ivanenko, Y.P., Grasso, R., Macellari, V., Lacquaniti, F., 2002a. Control of foot trajectory inhuman locomotion: role of ground contact forces in simulated reduced gravity. J.Neurophysiol. 87, 3070–3089.

Ivanenko, Y.P., Grasso, R., Macellari, V., Lacquaniti, F., 2002b. Two-thirds power law inhuman locomotion: role of ground contact forces. Neuroreport 13, 1171–1174.

Ivanenko, Y.P., Labini, F.S., Cappellini, G., Macellari, V., McIntyre, J., Lacquaniti, F., 2011.Gait transitions in simulated reduced gravity. J. Appl. Physiol. 110, 781–788.

Jastorff, J., Orban, G.A., 2009. Human functional magnetic resonance imaging revealsseparation and integration of shape and motion cues in biological motion processing.J. Neurosci. 29, 7315–7329.

Jastorff, J., Clavagnier, S., Gergely, G., Orban, G.A., 2011. Neural mechanisms of under-standing rational actions: middle temporal gyrus activation by contextual violation.Cereb. Cortex 21, 318–329.

Jastorff, J., Popivanov, I.D., Vogels, R., Vanduffel, W., Orban, G.A., 2012. Integration of shapeand motion cues in biological motion processing in the monkey STS. Neuroimage 60,911–921.

Johansson, G., 1973. Visual perception of biological motion and a model for its analysis.Percept. Psychophys. 14, 195–204.

Johnson, M.H., 2006. Biological motion: a perceptual life detector? Curr. Biol. 16,R376–R377.

Jokisch, D., Troje, N.F., 2003. Biological motion as a cue for the perception of size. J. Vis. 3,252–264.

Jokisch, D., Daum, I., Suchan, B., Troje, N.F., 2005. Structural encoding and recognition ofbiological motion: evidence from event-related potentials and source analysis.Behav. Brain Res. 157, 195–204.

Kilner, J.M., Friston, K.J., Frith, C.D., 2007. Predictive coding: an account of the mirror neuronsystem. Cogn. Process. 8, 159–166.

Kim, J., Park, S., Blake, R., 2011. Perception of biological motion in schizophrenia andhealthy individuals: a behavioral and FMRI study. PLoS One 6, e19971.

Koster-Hale, J., Saxe, R., 2013. Theory of mind: a neural prediction problem. Neuron 79,836–848.

Kriegeskorte, N., Simmons, W.K., Bellgowan, P.S.F., Baker, C.I., 2009. Circular analysis insystems neuroscience: the dangers of double dipping. Nat. Neurosci. 12, 535–540.

Lange, J., Lappe, M., 2006. A model of biological motion perception from configural formcues. J. Neurosci. 26, 2894–2906.

Lindquist, M.A., 2008. The statistical analysis of fMRI data. Stat. Sci. 23, 439–464.Maffei, V., Macaluso, E., Indovina, I., Orban, G., Lacquaniti, F., 2010. Processing of targets in

smooth or apparent motion along the vertical in the human brain: an fMRI study. J.Neurophysiol. 103, 360–370.

Maffei, V., Giusti, M.A., Macaluso, E., Lacquaniti, F., Viviani, P., 2014. Unfamiliar walkingmovements are detected early in the visual stream: an fMRI study. Cereb. Cortexhttp://dx.doi.org/10.1093/cercor/bhu008 in press, Available online February 13, 2014.

Mikl, M., Marecek, R., Hlustik, P., Pavlicova, M., Drastich, A., Chlebus, P., Brazdil, M., Krupa, P.,2008. Effects of spatial smoothing on fMRI group inferences. Magn. Reson. Imaging 26,490–503.

Miller, W.L., Maffei, V., Bosco, G., Iosa, M., Zago, M., Macaluso, E., Lacquaniti, F., 2008. Vestib-ular nuclei and cerebellum put visual gravitational motion in context. J. Neurophysiol.99, 1969–1982.

Osaki, Y., Kunin, M., Cohen, B., Raphan, T., 2008. Relative contribution of walking velocity andstepping frequency to the neural control of locomotion. Exp. Brain Res. 185, 121–135.

Park, H.J., Friston, K., 2013. Structural and functional brain networks: from connections tocognition. Science 342, 1238411.

Pavlova, M., Sokolov, A., 2000. Orientation specificity in biological motion perception.Percept. Psychophys. 62, 889–899.

Pavlova, M., Lutzenberger, W., Sokolov, A., Birbaumer, N., 2004. Dissociable cortical pro-cessing of recognizable and non-recognizable biological movement: analysinggamma MEG activity. Cereb. Cortex 14, 181–188.

230 V. Maffei et al. / NeuroImage 104 (2015) 221–230

Peelen, M.V., Wiggett, A.J., Downing, P.E., 2006. Patterns of fMRI activity dissociate over-lapping functional brain areas that respond to biological motion. Neuron 49,815–822.

Peuskens, H., Vanrie, J., Verfaillie, K., Orban, G.A., 2005. Specificity of regions processing bi-ological motion. Eur. J. Neurosci. 21, 2864–2875.

Rahnev, D., Lau, H., de Lange, F.P., 2011. Prior expectationmodulates the interaction betweensensory and prefrontal regions in the human brain. J. Neurosci. 31, 10741–10748.

Rao, R.P., Ballard, D.H., 1999. Predictive coding in the visual cortex: a functional interpre-tation of some extra-classical receptive-field effects. Nat. Neurosci. 2, 79–87.

Runeson, S., Frykholm, G., 1981. Visual perception of lifted weight. J. Exp. Psychol. Hum.Percept. Perform. 7, 733–740.

Saygin, A.P., Wilson, S.M., Hagler, D.J., Bates, E., Sereno, M.I., 2004. Point-light biologicalmotion perception activates human premotor cortex. J. Neurosci. 24, 6181–6188.

Saygin, A.P., Chaminade, T., Ishiguro, H., Driver, J., Frith, C., 2012. The thing that should notbe: predictive coding and the uncanny valley in perceiving human and humanoidrobot actions. Soc. Cogn. Affect. Neurosci. 7, 413–422.

Schultz, W., Dayan, P., Montague, P.R., 1997. A neural substrate of prediction and reward.Science 275, 1593–1599.

Schurz, M., Aichhorn, M., Martin, A., Perner, J., 2013. Common brain areas engaged in falsebelief reasoning and visual perspective taking: a meta-analysis of functional brain im-aging studies. Front. Hum. Neurosci. 7, 712.

Shipley, T.F., 2003. The effect of object and event orientation on perception of biologicalmotion. Psychol. Sci. 14, 377–380.

Singer, J.M., Sheinberg, D.L., 2010. Temporal cortex neurons encode articulated actions asslow sequences of integrated poses. J. Neurosci. 30, 3133–3145.

Sokolov, A.A., Gharabaghi, A., Tatagiba, M.S., Pavlova, M., 2010. Cerebellar engagement inan action observation network. Cereb. Cortex 20, 486–491.

Stephan, K.E., Penny, W.D., Moran, R.J., den Ouden, H.E., Daunizeau, J., Friston, K.J., 2010.Ten simple rules for dynamic causal modeling. Neuroimage 49, 3099–3109.

Sumi, S., 1984. Upside-down presentation of the Johansson moving light-spot pattern.Perception 13, 283–286.

Summerfield, C., Egner, T., 2009. Expectation (and attention) in visual cognition. TrendsCogn. Sci. 13, 403–409.

Sylos-Labini, F., Ivanenko, Y.P., Cappellini, G., Portone, A., Maclellan, M.J., Lacquaniti, F.,2013. Changes of gait kinematics in different simulators of reduced gravity. J. Mot.Behav. 45, 495–505.

Troje, N.F., 2013. What is biological motion?: Definition, stimuli and paradigms. In:Rutherford, M.D., Kuhlmeier, V.A. (Eds.), Social perception: detection and interpreta-tion of animacy, agency, and intention. MIT Press, Cambridge.

Troje, N.F., Chang, D.H.F., 2013. Shape-independent processes in biological motionperception. In: Johnson, K.L., Shiffrar, M. (Eds.), People watching: social, percep-tual, and neurophysiological studies of body perception. Oxford University Press,pp. 82–100.

Troje, N.F., Westhoff, C., 2006. The inversion effect in biological motion perception: evi-dence for a “life detector”? Curr. Biol. 16, 821–824.

Vaina, L.M., Solomon, J., Chowdhury, S., Sinha, P., Belliveau, J.W., 2001. Functional neuro-anatomy of biological motion perception in humans. Proc. Natl. Acad. Sci. U. S. A. 98,11656–11661.

Vallortigara, G., Regolin, L., 2006. Gravity bias in the interpretation of biological motion byinexperienced chicks. Curr. Biol. 16, R279–R280.

Vangeneugden, J., De Maziere, P.A., Van Hulle, M.M., Jaeggli, T., Van Gool, L., Vogels, R., 2011.Distinct mechanisms for coding of visual actions in macaque temporal cortex. J.Neurosci. 31, 385–401.

Westhoff, C., Troje, N.F., 2007. Kinematic cues for person identification from biologicalmotion. Percept. Psychophys. 69, 241–253.

Woo, C.W., Krishnan, A., Wager, T.D., 2014. Cluster-extent based thresholding in fMRIanalyses: pitfalls and recommendations. Neuroimage 91, 412–419.