Abstract
Body ownership and the sense of agency are two central aspects of bodily self-consciousness. While multiple neuroimaging studies have investigated the neural correlates of body ownership and agency separately, few studies have investigated the relationship between these two aspects during voluntary movement when such experiences naturally combine. By eliciting the moving rubber hand illusion with active or passive finger movements during functional magnetic resonance imaging, we isolated activations reflecting the sense of body ownership and agency, respectively, as well as their interaction, and assessed their overlap and anatomic segregation. We found that perceived hand ownership was associated with activity in premotor, posterior parietal, and cerebellar regions, whereas the sense of agency over the movements of the hand was related to activity in the dorsal premotor cortex and superior temporal cortex. Moreover, one section of the dorsal premotor cortex showed overlapping activity for ownership and agency, and somatosensory cortical activity reflected the interaction of ownership and agency with higher activity when both agency and ownership were experienced. We further found that activations previously attributed to agency in the left insular cortex and right temporoparietal junction reflected the synchrony or asynchrony of visuoproprioceptive stimuli rather than agency. Collectively, these results reveal the neural bases of agency and ownership during voluntary movement. Although the neural representations of these two experiences are largely distinct, there are interactions and functional neuroanatomical overlap during their combination, which has bearing on theories on bodily self-consciousness.
SIGNIFICANCE STATEMENT How does the brain generate the sense of being in control of bodily movement (agency) and the sense that body parts belong to one’s body (body ownership)? Using fMRI and a bodily illusion triggered by movement, we found that agency is associated with activity in premotor cortex and temporal cortex, and body ownership with activity in premotor, posterior parietal, and cerebellar regions. The activations reflecting the two sensations were largely distinct, but there was overlap in premotor cortex and an interaction in somatosensory cortex. These findings advance our understanding of the neural bases of and interplay between agency and body ownership during voluntary movement, which has implications for the development of advanced controllable prosthetic limbs that feel like real limbs.
Introduction
When you raise your arm, you automatically experience that it was you who caused the arm to lift and that the moving arm is your own. These two experiences blend so naturally during everyday voluntary behavior that we rarely think of them as distinct. However, in philosophy, cognitive science, and cognitive neuroscience, there is a long tradition of studying the sense of being in control of and causing bodily action through volition (i.e., the sense of agency; Jeannerod, 2003; Haggard, 2017), and the immediate perceptual experience of limbs and body parts as one’s own (i.e., the sense of body ownership; Petkova and Ehrsson, 2010; Ehrsson, 2020), as distinct processes. Body ownership and agency are both considered to be fundamental aspects of self-consciousness and are critical for defining what it means to be a conscious embodied agent distinct from the environment. However, most previous studies have focused on these two experiences in isolation using different experimental paradigms, so little is known about how they combine during voluntary movement.
Body ownership is considered to depend on the integration of visual, somatosensory, and other sensory bodily signals into coherent multisensory percepts of one’s own body through mechanisms of multisensory integration (Ehrsson et al., 2004; Blanke et al., 2015; Samad et al., 2015; Ehrsson, 2020), whereas agency relates to the association between voluntary action and outcome and has been linked to the match between the expected sensory consequences of movement and their sensory feedback (Frith et al., 2000) and the experience of volition during voluntary movement (Haggard, 2017). Previous functional magnetic resonance imaging (fMRI) studies have identified brain areas associated with the sense of body ownership and the sense of agency, where body ownership is associated with activity in a set of premotor-parieto-cerebellar regions (Ehrsson et al., 2004, 2005; Guterstam et al., 2013; Gentile et al., 2013; Limanowski and Blankenburg, 2016) and with agency related to activations in the right inferior parietal cortex, temporoparietal junction (TPJ), pre-supplementary motor area (SMA), insula (Farrer and Frith, 2002; Farrer et al., 2003; Schnell et al., 2007; David et al., 2008; Yomogida et al., 2010; Chambon et al., 2013), superior temporal gyrus (STG; Nahab et al., 2011; Uhlmann et al., 2020), and left primary sensorimotor cortex (Sperduti et al., 2011). However, previous agency imaging studies have focused on agency over external sensory events that occur as a consequence of bodily movement rather than agency experienced directly over one’s moving limbs, and body ownership studies have not investigated movement (but see Tsakiris et al., 2010). Therefore, the precise functional neuroanatomical relationship between ownership and agency during simple voluntary movement remains unclear.
Here, we used the rubber hand illusion (RHI; Botvinick and Cohen, 1998) elicited by finger movements—the moving RHI (Kalckert and Ehrsson, 2012)—to investigate the neural bases of body ownership and agency within a single fMRI paradigm. To elicit this bodily illusion, the participants perform a repetitive finger movement with their hidden index finger while they observe a rubber hand placed in full view making the corresponding finger movements. After a few synchronous movements, the participants start to experience the moving rubber hand as their own and that they are directly controlling its movements voluntarily (Kalckert and Ehrsson, 2012, 2014). By manipulating the relative timing of the real and rubber hand finger movements (synchrony or asynchrony), the type of movement (active or passive), and the spatial–anatomic orientation of the rubber hand with respect to the real hand (congruent or incongruent), the sense of body ownership and agency can be individually manipulated (Kalckert and Ehrsson, 2012). Thus, we implemented a 2 × 2 × 2 factorial within-subjects experimental design with these three factors to identify active neuronal populations that reflect body ownership, agency, and their potential interaction. We hypothesized that ownership and agency should be associated with activity in different neural circuits, which is in line with previous studies, but also that their combination should be associated with overlapping and stronger activation in certain frontoparietal regions because of the integration of the two sensations.
Results
Behavioral experiment
The results from the behavioral pretest experiment replicated the main findings from the original article on the moving rubber hand illusion and confirmed that our behavioral paradigm worked as expected (Kalckert and Ehrsson, 2012), but in a full 2 × 2 × 2 factorial within-subject design. The results confirmed that the sense of body ownership and sense of agency can be dissociated behaviorally, as we had expected (Fig. 4A). In the AMSTCO condition, the participants experienced both a sense of body ownership and agency of the rubber hand (i.e., the mean rating scores of these two sensations were both positive, meaning that, on average, the participants affirmed both these experiences in the moving RHI condition with active finger movements). Furthermore, in the PMSTCO condition, the moving RHI condition with passive finger movements, the participants experienced a sense of body ownership (positive rating score) of the rubber hand but denied experiencing a sense of agency (negative mean agency score). Finally, in the AMSTIO condition, the participants experience a sense of agency over the rubber hand but no sense of body ownership (positive agency and negative ownership scores). In the control conditions, the participants did not report sensing body ownership or agency, and the mean ownership and agency scores were negative (Fig. 4A).
A, The results from the behavioral experiment. These results show a double dissociation between the sense of body ownership and sense of agency in our full factorial design. The AMSTCO condition displayed high ratings for both sense of body ownership and sense of agency. The PMSTCO condition showed high ownership ratings and low agency ratings, whereas the AMSTIO condition showed high agency ratings and low ownership ratings. Bars represent mean ratings, and error bars indicate the SEM. B, Ownership and agency indices calculated by subtracting the pooled ownership and agency control ratings from the pooled ownership and agency ratings, respectively. Bars indicate the means, and error bars indicate the SEM.
We then compared the ownership to the ownership control ratings and found significantly higher ratings of the ownership statements compared with the control statements in the AMSTCO condition (W = 349; p < 0.001; rank–biserial correlation, 0.989) and the PMSTCO condition (W = 314; p < 0.001; rank–biserial correlation, 0.932). The same analysis for the sense of agency showed significantly higher ratings of the agency statements compared with the agency control statements in the AMSTCO condition (W = 351; p < 0.001; rank–biserial correlation, 1.00) and the AMSTIO condition (W = 351; p < 0.001; rank–biserial correlation, 1.00). The individual ratings for each statement and condition are given in Extended Data Table 4-1.
Table 4-1
Mean ratings and standard deviation of each statement from the behavioral experiment. Download Table 4-1, DOCX file.
We then directly tested the hypothesis that the sense of body ownership depended on synchronous visuosomatosensory feedback when moving the finger as well as spatial congruency between the orientations of the rubber hand and the participants’ real hand (Botvinick and Cohen, 1998; Tsakiris et al., 2010; Ehrsson, 2012; Kalckert and Ehrsson, 2012). To this end, we analyzed the ownership indices [the difference between the ownership score and ownership control score] in a 2 × 2 × 2 ANOVA (Fig. 4B, ownership and agency indices across the eight conditions). The factors movement type (active/passive), timing of movements (synchronous/asynchronous), and orientation of the rubber hand (congruent/incongruent) were entered into the analysis. The results showed a significant main effect of movement (F = 6.63; df = 29, 1; p = 0.016; η2 = 0.012), a significant main effect of timing (F = 41.276; df = 29, 1; p < 0.001; η2 = 0.216), and a significant main effect of orientation (F = 17.645; df = 29, 1; p < 0.001; η2 = 0.091). Importantly, the interaction between timing and orientation was significant (F = 31.933; df = 29, 1; p < 0.001; η2 = 0.109), in line with the spatial and temporal multisensory rules of illusory rubber hand ownership (Kalckert and Ehrsson, 2012) and our operationalization of ownership in the fMRI factorial experimental design. There was no significant interaction between timing and movement type (F = 0.894; df = 29, 1; p = 0.353; η2 = 0.002). However, the interaction between movement type and orientation was also significant (F = 5.982; df = 29, 1; p = 0.022; η2 = 0.008), which suggests higher ownership ratings during the active finger movements when the rubber hand was in a spatially congruent orientation. Moreover, there was a significant three-way interaction among timing, movement type, and orientation (F = 6.421; df = 29, 1; p = 0.018; η2 = 0.013). This three-way interaction suggests enhanced ownership of the rubber hand in the active synchronous congruent condition when participants experience both ownership and agency over the moving rubber hand compared with the passive synchronous congruent condition, when people only experience illusory ownership, and thus provides behavioral support for examining the interaction of ownership and agency in our factorial fMRI design. In line with this, post hoc pairwise comparisons between the AMSTCO and PMSTCO conditions in terms of ownership index (t = 3.155; df = 29; p = 0.004; Cohen’s d = 0.607) and ownership scores (t = 2.413; df = 29; p = 0.023; Cohen’s d = 0.464) further revealed significant differences in both cases. This is an interesting behavioral finding that suggests that active finger movements provide a stronger cue for body ownership than passive finger movements, which has a bearing on an ongoing debate in the behavioral literature on whether body ownership and agency interact in the moving rubber hand illusion or if they are completely independent (Dummer et al., 2009; Tsakiris et al., 2010; Walsh et al., 2011; Kalckert and Ehrsson, 2012, 2014, 2017; Riemer et al., 2013; Hara et al., 2022).
We hypothesized that the sense of agency is dependent on synchronous visuomotor feedback (i.e., the match between predicted sensory consequences of the active movement and sensory feedback) as well as on participants actively moving the index finger (i.e., voluntarily generating the movements; Kalckert and Ehrsson, 2012, 2014). To this end, we analyzed the agency indices (i.e., the difference between the agency scores and the agency control scores) in a 2 × 2 × 2 ANOVA. The three factors of movement type (active/passive), timing (synchronous/asynchronous), and orientation (congruent/incongruent) were entered in the analysis (Fig. 4B). As expected, the results showed a significant main effect of movement type (F = 42.244; df = 29, 1; p < 0.001; η2 = 0.207) and a significant main effect of timing (F = 107.572; df = 29, 1; p < 0.001; η2 = 0.255), which suggests that both active movements and synchronous seen and felt movement enhanced agency ratings. There was no main effect of orientation (F = 0.021; df = 29, 1; p = 0.886; η2 = 0.00002), indicating that the orientation of the rubber hand did not influence agency. Importantly, the interaction between synchrony and movement type was significant (F = 36.751; df = 29, 1; p < 0.001, η2 = 0.132), in line with the hypothesis and our operationalization of agency as this two-way interaction in our fMRI design. The interaction between movement type and orientation was not significant (F = 0.406; df = 29, 1; p = 0.530, η2 = 0.0005), nor was the interaction between synchrony and orientation (F = 0.379; df = 29, 1; p = 0.251, η2 = 0.001). The three-way interaction among synchrony, movement type, and orientation was also nonsignificant (F = 1.560; df = 29, 1; p = 0.223, η2 = 0.002). These latter results are consistent with the hypothesis that agency does not depend on the orientation of the rubber hand and that agency can be operationalized as an interaction between movement type and temporal congruence, only arising for active movements with synchronous visual feedback. Overall, the questionnaire results from our behavioral experiment confirmed that our selective manipulation of ownership and agency in the moving rubber hand illusion worked as expected, and was in line with established multisensory and cognitive constraints and provided behavioral support for examining the interaction of ownership and agency in the fMRI data (see below).
The sense of body ownership is associated with activity in multisensory frontal and parietal regions as well as in cerebellar regions
To identify activations associated with the sense of ownership of the rubber hand in both the active and passive conditions, we used the contrast [(PMSTCO – PMATCO) – (PMSTIO – PMATIO)] + [(AMSTCO – AMATCO) – (AMSTIO – AMATIO)]. In line with our hypothesis, this contrast revealed significant activation peaks in the left premotor cortex, posterior parietal cortex, and cerebellum (p < 0.05, FWE corrected for multiple comparisons; Fig. 5, Table 2). The premotor activations were located in the precentral gyrus at a location that corresponds to the dorsal premotor cortex (PMd; −34, −10, 64; p < 0.05, FWE corrected; Fig. 5), and parietal lobe activations were observed in the supramarginal gyrus (SMG; −60, −48, 38; p < 0.05, FWE corrected; Fig. 5). Activation peaks were also observed in the primary motor cortex (precentral gyrus) and the primary somatosensory cortex (postcentral gyrus) at sites that corresponded very well to peaks identified in the localizer experiment (see above). However, since no a priori hypotheses existed for these regions and they did not survive correction for multiple comparisons at the whole-brain level, they are reported with their uncorrected p values. We also observed activity in the intraparietal cortex (p < 0.001, uncorrected) but more posteriorly than we had predicted based on previous work. In the subcortical structures, we observed significant activity in the Crus I (lobule VIIa; 40, −74, −34) and vermis (lobule VIIa; 4, −68, −46) of the cerebellum (p < 0.05, FWE corrected; Fig. 5). Finally, we observed a large active cluster in the left dorsolateral prefrontal cortex (Fig. 5A; p < 0.001 uncorrected). No clusters survived correction for multiple comparisons at the whole-brain level (FDR corrected). Further statistical details on the anatomic locations in MNI space of the above-mentioned peaks are shown in Figure 5 and Table 2.
Activation peaks for the main contrasts
A, Overview of the brain regions that display activation reflecting the sense of body ownership over the rubber hand defined by the contrast [(PMSTCO – PMATCO) – (PMSTIO – PMATIO)] + [(AMSTCO – AMATCO) – (AMSTIO – AMATIO)]. For display purposes only, the activations are projected onto a three-dimensional rendering of a standard brain with a threshold of p < 0.005 (uncorrected for multiple comparisons, k ≥ 5). RH, Right hemisphere; LH, left hemisphere; Occ, occipital view; CS, central sulcus. B, Bar charts displaying the parameter estimates (a.u.) and SEs for the major peaks of activation. The coordinates are given in MNI space. The peaks are displayed in representative sections indicated by a dotted white circle on an activation map (p < 0.005, uncorrected for display purposes). L, Left; R, right; PrCG, precentral gyrus; PoCG, postcentral gyrus; IPS, intraparietal sulcus; SMG, supramarginal gyrus. Asterisks indicate activation peaks that survive small-volume correction (*p < 0.05, corrected; **p < 0.01); the peaks without an asterisk did not survive correction and are reported in Table 2 with their uncorrected p value. All peaks from the contrast are reported in Extended Data Table 5-1. Condition key: first letter A or P (active or passive) with subscript M (movement); second letter S or A (synchronous or asynchronous) with subscript T (timing); and third letter C or I (congruent or incongruent) with subscript O (orientation).
Table 5-1
All peaks from the ownership contrast [(PMSTCO-PMATCO)-(PMSTIO-PMATIO)] + [(AMSTCO-AMATCO)-(AMSTIO-AMATIO)]. All peaks that survived the threshold p<0.005 (uncorrected, k≥10) are reported. Three peaks located in the white matter were excluded from this table. Download Table 5-1, DOCX file.
Correlation between subjective ownership ratings and ownership contrast
In a complementary descriptive approach, we followed up on the above ownership interaction contrast by examining whether those BOLD effects also correlated with the subjective ratings in the ownership statements. To this end, we performed a multiple regression analysis using the ownership ratings from each participant to search for voxels whose parameter estimates could be predicted from the behavioral contrast (see Materials and Methods). We identified four such regions whose parameter estimates were significantly correlated with the behavioral contrast (Fig. 6). The activity in the left PMd (−24, −12, 70; p < 0.05, Fig. 6) and cerebellum was significant after FWE correction (cerebellum; −26, −46, −26; p < 0.05, Fig. 6), whereas the activity in the postcentral gyrus and postcentral sulcus was not (p < 0.001, uncorrected; Fig. 6).
Correlation between behavioral ownership ratings (x-axis) and parameter estimates from the ownership contrast (y-axis; in a.u.) in the left precentral sulcus (PrCS; −24, −12, 70), left postcentral gyrus (PoCG; −24, −40, 68), left postcentral sulcus (PoCS; −22, −38, 70), and left cerebellum (−26, −46, −26). Pearson’s r and p values are given in each respective correlation plot. The peaks are displayed as activation maps (p < 0.005, uncorrected) on representative sections of an average anatomic section and are indicated with a dotted white line.
The sense of agency is associated with activity in the left precentral and postcentral gyri as well as right superior temporal gyrus
We then examined activations that reflect the sense of agency, that is, increases in activity dependent on actively generated movements as well as synchronous sensory feedback from the moving finger regardless of whether the hand was experienced as part of one’s body or not. To this end, we used the contrast [(AMSTCO – PMSTCO) – (AMATCO – PMATCO)] + [(AMSTIO – PMSTIO) – (AMATIO – PMATIO)], which represents agency across the congruent and incongruent conditions. In line with our hypotheses, we observed a significant activation peak in the left premotor cortex (−38, −8, 62; p < 0.05, FWE corrected; Fig. 7, Table 2) and an activation in the right superior temporal gyrus that almost reached significance (58, −24, 12; p = 0.051, FWE corrected; Fig. 7, Table 2). This cluster is the second largest (k = 347) in this contrast (the largest one being the left superior temporal gyrus), and its location is very close to the peak from the localizer experiment around which the small volume correction was made, which is why we chose to report it despite the p value of 0.051. We also observed increases in activity in the intraparietal cortex bilaterally as well as the left superior temporal gyrus and left postcentral gyrus (p < 0.001, uncorrected), but these activations did not survive correction for multiple comparisons and are thus only mentioned for descriptive purposes.
A, Overview of the brain regions that display activation reflecting the sense of agency defined by the contrast [(AMSTCO – PMSTCO) – (AMATCO – PMATCO)] + [(AMSTIO – PMSTIO) – (AMATIO – PMATIO)]. For display purposes only, the activations are projected onto a three-dimensional render of a standard brain with a threshold of p < 0.005 (uncorrected for multiple comparisons, k ≥ 5). RH, Right hemisphere; LH, left hemisphere; STS, superior temporal sulcus; CS, central sulcus. B, Bar charts displaying the parameter estimates (in a.u.) and SEs for the major peaks of activation. The coordinates are given in MNI space. The peaks are displayed in representative sections indicated by a dotted white circle (p < 0.005, uncorrected for display purposes). L, Left; R, right; PrCG, precentral gyrus; PoCG, postcentral gyrus; IPS, intraparietal sulcus; STG, superior temporal gyrus. *Activation peaks that survive small-volume correction (p < 0.05, corrected); the peaks without an asterisk did not survive correction and are reported in Table 2 with their uncorrected p value. All peaks from the contrast are reported in Extended Data Table 7-1.
Table 7-1
All peaks from the ownership contrast [(AMSTCO-PMSTCO)-(AMATCO-PMATCO)] + [(AMSTIO-PMSTIO)-(AMATIO-PMATIO)]. All peaks that survived the threshold p<0.005 (uncorrected, k≥10) are reported. Download Table 7-1, DOCX file.
Agency and ownership overlap in the precentral gyrus
To test for areas that showed increases in activity reflecting both ownership and agency, we used a conjunction analysis with the two two-way interaction contrasts described above for ownership and agency (Friston et al., 1999; Fig. 8A). The analysis revealed a significant activation peak in the precentral gyrus (PMd, −38, −8 62; p < 0.05, FWE corrected; Fig. 8A).
A, Conjunction analysis between the agency contrast and ownership contrast revealed overlapping activation in the left PMd. The significant activation peak (p < 0.05, corrected) is displayed on a representative section (p < 0.005, uncorrected) and is indicated with a dotted white line. B, PPI analysis of regions displaying increased connectivity with the seed region in the left postcentral gyrus (−38, −28, 52). The left SMA displays a task-specific increase in connectivity with the left postcentral gyrus (SMA; t = 3.56; p = 0.001, uncorrected). The peak is displayed as part of an activation map (p < 0.005, uncorrected) and is indicated with a dotted white line. The activation maps are presented on representative sagittal and coronal sections of a mean anatomic MRI image made up of all participants’ structural brain scans.
Interaction between ownership and agency revealed activation in the somatosensory cortex
To test for interaction between ownership and agency, we used the contrast [(AMSTCO – PMSTCO) – (AMATCO – PMATCO)] – [(AMSTIO – PMSTIO) – (AMATIO – PMATIO)]. This corresponds to the three-way interaction among movement type (active/passive), timing (synchronous/asynchronous), and rubber hand orientation (congruent/incongruent), and thus reveals neural responses unique to the combination of ownership and agency in the moving rubber hand illusion condition (AMSTCO). The results show significant activation in the left primary sensorimotor cortex with a significant peak of activation located in the postcentral gyrus at the level of the hand representations (−38, −28, 52; p < 0.05, FWE corrected; Fig. 9) and three further peaks in the postcentral gyrus that did not survive corrections for multiple comparisons (p < 0.005; Fig. 9, Table 2).
A, Overview of the brain regions that display activation reflecting the unique combination of agency and body ownership as defined by the contrast [(AMSTCO – PMSTCO) – (AMATCO – PMATCO)] – [(AMSTIO – PMSTIO) – (AMATIO – PMATIO)]. For display purposes only, the activations are projected onto a three-dimensional rendering of a standard brain with a threshold of p < 0.005 (uncorrected for multiple comparisons, k ≥ 5). RH, Right hemisphere; LH, left hemisphere; IPS, intraparietal sulcus; PoCS, postcentral sulcus; CS, central sulcus. B, Bar charts displaying the parameter estimates (in a.u.) and SEs for the major peaks of activation. The coordinates are given in MNI space. The peaks are displayed in representative sections indicated by a dotted white circle on an activation map (p < 0.005, uncorrected for display purposes). L, left; R, right; PoCG, postcentral gyrus. *Activation peaks that survive small-volume correction (p < 0.05 corrected); the peaks without an asterisk did not survive small-volume correction and are reported in Table 2 with their uncorrected p value. All peaks from the contrast are reported in Extended Data Table 9-1. Condition key: first letter A or P (active or passive) with subscript M (movement); second letter S or A (synchronous or asynchronous) with subscript T (timing); third letter C or I (congruent or incongruent) with subscript O (orientation).
Table 9-1
All peaks from the ownership x agency interaction contrast [(AMSTCO-PMSTCO)-(AMATCO-PMATCO)]-[(AMSTIO-PMSTIO)-(AMATIO-PMATIO)]. All peaks that survived the threshold p<0.005 (uncorrected, k≥10) are reported. Download Table 9-1, DOCX file.
Table 10-1
All peaks from the inverse of the ownership x agency interaction contrast [(AMSTCO-PMSTCO)-(AMATCO-PMATCO)]-[(AMSTIO-PMSTIO)-(AMATIO-PMATIO)]. All peaks that survived the threshold p<0.005 (uncorrected, k≥10) are reported. Download Table 10-1, DOCX file.
We should clarify here that the somatosensory activation under discussion can probably not be explained by somatosensory attenuation (Zeller et al., 2015; Kilteni and Ehrsson, 2017, 2020) or gating (Angel and Malenka, 1982; Post et al., 1994; Voudouris et al., 2019; Kilteni and Ehrsson, 2022) because we observed an increase in activity, not a reduction. Moreover, we controlled the amplitude of the movements, and there were no significant differences in movement frequency between conditions (see below; Fig. 13). Therefore, it is unlikely that low-level differences in motor output or somatosensory feedback confounded our S1 findings. We also think it is implausible that differences in tap force between the active and passive movements could explain our results because participants were trained to apply gentle taps and the experimenter reproduced such gentle taps in the passive condition; furthermore, the effect of active versus passive movements are matched in the three-way interaction contrast (as well as in the agency and ownership interaction contrasts).
Next, we examined the opposite direction of the three-way interaction contrast of movement type, synchrony, and orientation [(AMSTCO – PMSTCO) – (AMATCO – PMATCO)] – [(AMSTIO – PMSTIO) – (AMATIO – PMATIO)]. This contrast revealed only one activation in the left middle occipital gyrus and one smaller activation in the right middle occipital gyrus (Fig. 10, Table 2), but neither of these activations survived correction for multiple comparisons.
To investigate which brain regions are associated with the sense of agency of external objects as opposed to bodily objects, we defined a contrast that was the inverse of the three-way interaction [(AMSTCO – PMSTCO) – (AMATCO – PMATCO)] – [(AMSTIO – PMSTIO) – (AMATIO – PMATIO)]. The results show activation in the left middle occipital gyrus (p < 0.001, uncorrected; did not survive correction for multiple comparisons) and right middle occipital gyrus (p = 0.002, uncorrected). The coordinates are given in MNI space. L, left; R, right; MOG, middle occipital gyrus. The peak is displayed in a representative section and indicated by a dotted white circle on an activation map (p < 0.005, uncorrected for display purposes; k ≥ 5). The bar chart represents the parameter estimates (in a.u.) for the peak.
Psychophysiological interaction analysis of functional connectivity
Our results reported above revealed activation in the postcentral gyrus (S1) associated with the combined experience of illusory ownership and agency (three-way interaction). This made us curious if there could be changes in functional connectivity between S1 and other brain areas that could help us understand this finding further. Thus, in a post hoc exploratory PPI analysis of the functional connectivity in the three-way interaction of the factors timing, movement type, and orientation, we investigated the task-specific connectivity changes between the section of the postcentral gyrus under discussion (−38, –28, 52) and the rest of the brain. We found that the sense of ownership in the presence of a sense of agency increased the functional coupling between the left primary sensory cortex and the ipsilateral SMA (−2, −6, 64; t = 3.56; p = 0.001, uncorrected; Fig. 8B). In the rest of the brain, no active clusters were observed apart from one in cerebellum (R VIIb; 28, −68, −46; t = 3.51; p = 0.001, uncorrected).
Activations in the insular cortex and right temporoparietal cortex reflect visuoproprioceptive synchrony and asynchrony, respectively
In the previous literature, it has been suggested that the right angular gyrus located in the temporoparietal region is involved in the loss of agency when there is a mismatch between the expected sensory consequences of self-generated movement and the sensory feedback (Farrer and Frith, 2002; Farrer et al., 2003; Tsakiris et al., 2010). Furthermore, it has been reported that the insular cortex shows increases in activation when people experience agency (Farrer and Frith, 2002; Farrer et al., 2003). However, in our main planned contrasts reported above, we did not find any changes in activation in these two regions, even at the level of uncorrected p values (p < 0.005). To examine this apparent inconsistency further, we looked at the main effect of synchrony [(AMSTCO + AMSTIO + PMSTCO + PMSTIO) – (AMATCO + AMATIO + PMATCO + PMATIO)] and the main effect of asynchrony contrasts [(AMATCO + PMATCO + AMATIO + PMATIO) – (AMSTCO + PMSTCO + PMSTIO + AMSTIO); i.e., areas that show greater activation when visual feedback and finger movements are synchronous or asynchronous regardless of the senses of ownership or agency (i.e., across active and passive movements and across anatomically congruent or incongruent hand orientations)]. Interestingly, we found a large and significant activation (t = 3.66; p = 0.022, FWE corrected) located in the right angular gyrus of the TPJ region (50, −50, 32) that reflected the asynchronous relation between movement and visual feedback (main effect of asynchrony; Fig. 11A). In contrast, the synchrony of finger movements and visual feedback of the finger movement of the model hand (main effect of synchrony) was associated with significant activation (t = 3.71; p = 0.020, FWE corrected) of the left insular cortex (−38, −2, 10; Fig. 11B). Thus, rather than reflecting the sense of agency or the loss of agency by mismatching sensory feedback, our results suggest that the insular cortex and right temporoparietal cortex are involved in the basic detection of synchronous or asynchronous multimodal stimuli.
A, Activation in the right angular gyrus represented by the main effect of asynchrony: (AMATCO + PMATCO + AMATIO + PMATIO) – (AMSTCO + PMSTCO + PMSTIO + AMSTIO). B, Activation in the left insular cortex represented by the main effect of synchrony: (AMSTCO + PMSTCO + PMSTIO + AMSTIO) – (AMATCO + PMATCO + AMATIO + PMATIO). The coordinates are given in MNI space. The peak is displayed in a representative section and is indicated by a dotted white circle on an activation map (p < 0.005, uncorrected for display purposes).
Activation in the supplementary motor cortex reflects the main effect of active versus passive movements
Another area suggested to be involved in agency in previous fMRI studies, including agency in the moving RHI (Tsakiris et al., 2010), is the SMA. However, this area did not show any agency-related activity in our agency contrast described above, not even at p < 0.005 uncorrected. However, when we examined the main effect of movement type, contrasting all active movement versus all passive movement conditions in the current design, we observed significant activation of the SMA (AMSTCO + AMATCO + AMSTIO + AMATIO) – (PMSTCO + PMATCO + PMSTIO + PMATIO; Fig. 12). This region seems to be important for generating movements voluntarily, thereby indicating its role movement planning, programming, and volition more generally (Roland et al., 1980; Fried et al., 1991; Makoshi et al., 2011). However, we found no evidence for specific involvement in the sense of agency of the moving rubber hand.
Main effect of movement type (active or passive). Using the contrast (AMSTCO + AMATCO + AMSTIO + AMATIO) – (PMSTCO + PMATCO + PMSTIO + PMATIO), we compared all active movement conditions to all passive conditions (regardless of ownership or agency; yellow–red color scale for activation; top row). Active movement was associated with significant activations in the left supplementary motor area (−4, −4, 58; t = 4.98; p < 0.001 uncorrected), left precentral gyrus (PMd; −42, −10, 60; t = 7.82; p < 0.001, FDR corrected; data not shown), left precentral gyrus (M1; −40, −18, 56; t = 9.20; p < 0.011, FDR corrected; data not shown), right cerebellum (lobule VI; 20, −50, −24; t = 9.23; p < 0.001, FDR corrected; data not shown), left thalamus (−14, −22, 4; t = 5.90; p = 0.026, FDR corrected; data not shown), and right angular gyrus (34, −50, 24; t = 5.79; p = 0.033, FDR corrected; data not shown). We also compared all passive movement conditions to all active movement conditions, (PMSTCO + PMATCO + PMSTIO + PMATIO) – (AMSTCO + AMATCO + AMSTIO + AMATIO). Passive movements were associated with a relative increase in neural activity compared with active movements in the bilateral medial frontal cortex (only right shown in section: 10, 44, −2; t = 5.8; p < 0.001, uncorrected; left medial frontal cortex: −6, 46, −2; t = 5.18; p < 0.001; blue–green color scale for activation). The peaks are displayed in a representative section and are indicated by a dotted white circle on an activation map (p < 0.005, uncorrected for display purposes). All peaks from the contrast are reported in Extended Data Table 12-1. RH, Right hemisphere; LH, left hemisphere; SFG, superior frontal gyrus; MFG, medial frontal gyrus.
Table 12-1
The peaks from the localizer experiment used to define the ROIs in the current study. Download Table 12-1, DOCX file.
When we looked for areas showing greater activity in the passive movement conditions than in the active ones, we found a large activation in the medial prefrontal cortex in a region associated with default mode activity (Raichle et al., 2001; Buckner et al., 2008; Tacikowski et al., 2017), autobiographical episodic memory (Maguire, 2001; Svoboda et al., 2006; Bergouignan et al., 2014), and self-related information processing (Qin and Northoff, 2011; Tacikowski et al., 2017). The most straightforward interpretation is that since participants did not have an active task in this condition (they just relaxed their hand, and the experimenter generated the finger movements), the activity was higher in the default mode, thus explaining the relatively higher activity in this medial prefrontal region compared with the active movement conditions when the participant had a task to move their finger repeatedly. This activation also corresponds well to similar activity observed in the passive finger movement condition in the study of Tsakiris et al. (2010), which these authors attributed to ownership (Fig. 12).
Controlling for the number and frequency of taps in the different conditions
Using the optical sensor placed under the index finger of the participants, the number of taps as well as the frequency of taps for each condition could be analyzed. The analysis was performed on the time periods included in the fMRI analysis (i.e., excluding the time before illusion onset and the corresponding time periods for conditions without illusion). A one-way ANOVA revealed no significant differences across conditions for the frequency of taps (mean, 1.53 Hz; F = 0.636; df = 7; p = 0.725; Fig. 13). Moreover, when the frequencies of taps were analyzed using the same 2 × 2 × 2 design as the fMRI experiment, we found no significant main effect of movement type (F = 2.519; df = 19, 1; p = 0.129; η2 = 0.014), no significant main effect of timing (F = 2.353; df = 19, 1; p = 0.142; η2 = 0.007), no significant main effect of orientation (F = 2.390; df = 19, 1; p = 0.139; η2 = 0.041), and no significant interactions (movement type × timing: F = 0.928; df = 19,1; p = 0.348, η2 = 0.008; movement type × orientation: F = 0.152; df = 19,1; p = 0.701; η2<0.001; orientation × timing: F = 2.215; df = 19,1; p = 0.152; η2 = 0.006; movement type × timing × orientation: F = 0.430; df = 19,1; p = 0.520; η2 = 0.003).
The number and frequency of taps across conditions. The bars represent the mean number and frequency of taps for all conditions for the period excluding the illusion onset times (see Materials and Methods). Error bars indicate the SEMs. The analysis of the frequencies of taps revealed no significant main effects and no significant interactions, and there were no differences in frequencies across conditions. The exact values for each condition are given in Extended Data Table 13-1.
Discussion
This study has three main novel findings. First, the neural substrates of ownership and agency were largely distinct, with body ownership associated with increases in activity in the premotor cortex, posterior parietal, and cerebellar regions, and the sense of agency related to increased activity in the superior temporal cortex and dorsal premotor cortex. Second, one active section of the dorsal premotor cortex was associated with both agency and body ownership, indicating a cortical site where ownership and agency information may be combined. Third, there was an interaction between body ownership and agency in the somatosensory cortex so that its activity was higher when participants experienced both sensations. This was accompanied by higher ownership ratings, suggesting an agency-induced ownership enhancement of somatosensory cortical activity specific for voluntary movement. Collectively, these findings extend our knowledge of the neural basis of body ownership and agency, and reveal their functional interaction and the relative neuroanatomical overlap and segregation during simple movement, which advances our understanding of how bodily self-consciousness is implemented in the human brain.
The sense of body ownership during movement: integration of spatiotemporally congruent visuoproprioceptive signals in premotor-parietal-cerebellar regions
The present study extends the previous neuroimaging literature on the neural basis of body ownership (Ehrsson et al., 2004; Petkova et al., 2011; Brozzoli et al., 2012; Gentile et al., 2013; Guterstam et al., 2013, 2019; Limanowski and Blankenburg, 2016; Preston and Ehrsson, 2016; Chancel et al., 2022b) into such experience arising from the sensory feedback of movement. The sense of ownership of the moving rubber hand was associated with significant activations in the left premotor cortex (precentral gyrus), posterior parietal cortex (left supramarginal gyrus), and right lateral cerebellum. These activations probably reflect the integration of spatially and temporally congruent visual information from the moving rubber hand and kinesthetic–proprioceptive information from the hidden real hand because the neural response was specifically related to the conditions when the rubber hand was placed in an anatomically congruent condition and the seen and felt movements were synchronous (i.e., when the visual and kinesthetic–proprioceptive information obeyed the temporal and spatial rules of body ownership; Ehrsson, 2012; Kalckert and Ehrsson, 2012; Blanke et al., 2015; Samad et al., 2015; Chancel et al., 2022a), controlling for agency effects and effects related to active versus passive movement.
The difference between visuokinesthetic integration, which was studied herein, and visuotactile integration, which was investigated in previous RHI studies, can probably explain the differences in precise localization of the activation peaks in the premotor cortex compared with a previous study (Ehrsson et al., 2004). Although activations have been seen in both ventral and dorsal aspects of the premotor cortex in previous RHI studies (Gentile et al., 2013; Guterstam et al., 2019), the most consistent activations tend to have been located in the ventral premotor cortex (Ehrsson et al., 2004; Gentile et al., 2013; Guterstam et al., 2013, 2019; Limanowski and Blankenburg, 2016; Grivaz et al., 2017). The dorsal premotor cortex is active during passive hand and arm movements (Zhavoronkova et al., 2017), finger tapping (Ullén et al., 2003; Bengtsson et al., 2009), and illusory hand and arm movements triggered by muscle tendon vibration (Naito et al., 1999, 2005), which is consistent with a role in multisensory representation of the upper limb in space. The current activation in the SMG (p < 0.05, corrected) is consistent with the findings of earlier body ownership illusion studies based on visuotactile stimulation (Gentile et al., 2013; Petkova et al., 2011), and the current intraparietal cortex activation is located in a section of this sulcus associated with multisensory integration in perihand space (Lloyd et al., 2003; Makin et al., 2007; Brozzoli et al., 2011) and illusory hand ownership (Chancel et al., 2022b). We also observed activity in the ipsilateral lateral cerebellum that is in line with previous fMRI studies on various versions of the rubber hand illusion based on visuotactile stimulation (Ehrsson et al., 2004, 2005; Guterstam et al., 2013) and limb movement illusions (Ehrsson et al., 2005; Hagura et al., 2009). Importantly, the current findings extend the previous literature on body ownership and body representation by demonstrating a role for these premotor-parietal-cerebellar regions in the sense of limb ownership during movement.
The sense of agency in one’s own bodily movement: premotor and superior temporal cortex
We could isolate activity in the dorsal premotor cortex and superior temporal cortex reflecting agency over limb movement while controlling for unspecific effects related to multisensory synchrony–asynchrony detection, active versus passive movement, and body ownership. The dorsal premotor area has been reported in previous studies on the sense of agency over sensory events caused by voluntary movement (David et al., 2008; Yomogida et al., 2010; Nahab et al., 2011; Sperduti et al., 2011; Haggard, 2017), so our finding extends this to agency over perceived own bodily movement. The dorsal premotor cortex is anatomically connected to and receives input from the dorsolateral prefrontal cortex regarding intentions and the initiation of voluntary action in the context of an overall action plan (Passingham, 1993; Koechlin et al., 2003; Abe and Hanakawa, 2009; Yamagata et al., 2012) and receives multisensory input from the posterior parietal cortex regarding one’s own body as well as external sensory events; the dorsal premotor area can also influence movement execution in M1 and receive feedback from this area through direct corticocortical connections (Porter and Lemon, 1995; Dum et al., 2002). The dorsal premotor cortex is thus in an excellent position, anatomically and physiologically, to play a central role in the sense of agency by integrating and comparing signals related to voluntary motor commands and sensory feedback, consistent with our findings.
Interestingly, the section of the dorsal premotor cortex associated with agency also showed body ownership-related activity, as revealed in our conjunction analysis. This finding suggests that the neural bases of body ownership and agency are not completely distinct (Tsakiris et al., 2010), and that least one cortical area is involved in both processes. Different neuronal populations within the dorsal premotor cortex could implement the formation of a coherent multisensory representation of the hand in space (ownership), the generation of voluntary motor commands, and the matching of the outcomes of those commands with the sensory feedback and predictions (agency) or the same neuronal population within this area may implement both of these mechanisms (which could be tested in future studies with BOLD adaptation or multivoxel pattern analysis). Our findings suggest a more intimate relationship of the representations of body ownership and agency in the premotor cortex than commonly assumed and indicate that more attention should be devoted to this region in future studies on the neural mechanisms of agency of bodily action.
Previous neuroimaging studies have suggested that the superior temporal cortex plays a role in the sense of agency, but they reported that activation in the superior temporal gyrus reflected the loss of agency when controlling a virtual limb (Nahab et al., 2011; Uhlmann et al., 2020). However, these studies did not control for multisensory synchrony–asynchrony, the visual appearance (and identity) of the hand, or body ownership. In contrast, we found a relative activity increase that reflected gaining agency of the moving rubber hand, although all experimental conditions were deactivated compared with the resting baseline. The current activation peak is located more ventral and anterior to the deactivations in previous studies (Nahab et al., 2011; Uhlmann et al., 2020), making direct comparisons difficult. Although the precise functional role of the superior temporal cortex in agency is unclear, this region has been associated with action observation (Kilintari et al., 2014), visual processing of biological motion (Saygin, 2007), and perception of causality between sensory events (Blakemore et al., 2001), which collectively point toward a function of supporting the (visual) perception of causality relationships between the seen finger movement and the executed finger action, which presumably is an important component of the agency experience.
Interaction of body ownership and agency in the somatosensory cortex
Our analysis revealed somatosensory activity that was uniquely related to the situation when both ownership and agency were experienced over the moving rubber hand (interaction between ownership and agency). In principle, this activity could reflect a change in body ownership caused by agency or a change in agency caused by ownership. We think the former is more likely because the behavioral data showed a significant corresponding interaction effect in the questionnaire hand ownership ratings but not in the agency ratings. Thus, the somatosensory activity may be related to a change in the somatic feeling of the rubber hand illusion when this illusion is produced by visuomotor–kinesthetic correlations during active movements as opposed to visuokinesthetic correlations during passive movements. Motor commands and efferent signals can influence limb movement sensations (Gandevia et al., 2006; Walsh et al., 2010), and thus, we theorize that information related to the active motor command signals made the ownership experience more vivid by boosting kinesthetic sensations from the finger movements of the rubber hand. Such motor command signals could originate from premotor areas and influence the somatosensory cortex via corticocortical connections, which is supported by the finding of increased functional connectivity between the SMA and S1 in the active synchronous congruent condition when both ownership and agency were experienced (Fig. 8). Alternatively, agency might influence the multisensory integration process that determines body ownership by facilitating combination over segregation by influencing the prior probability of a common cause (Samad et al., 2015; Chancel et al., 2022a), although it remains unclear how this would lead to enhanced S1 activation rather than increased premotor or posterior parietal activity. The somatosensory activity might also reflect a special component of agency over one’s bodily movements—“bodily agency”—perhaps reflecting differences between own movement-related somatosensory predictions and predictions about external (e.g., visual) events that are indirectly caused by voluntary action (Frith et al., 2000). According to this view, somatosensory activity would reflect somatosensory predictions during bodily agency, whereas visual cortical activity would reflect visual predictions associated with “external agency” over the nonowned (rotated) rubber hand (Fig. 10). Regardless of the underlying mechanism and conceptualization, our finding links somatosensory activity to the combination of ownership and agency during voluntary limb movement.
