Building the bodily self‐awareness: Evidence for the convergence between interoceptive and exteroceptive information in a multilevel kernel density analysis study

Abstract Exteroceptive and interoceptive signals shape and sustain the bodily self‐awareness. The existence of a set of brain areas, supporting the integration of information coming from the inside and the outside of the body in building the sense of bodily self‐awareness has been postulated, yet the evidence remains limited, a matter of discussion never assessed quantitatively. With the aim of unrevealing where in the brain interoceptive and exteroceptive signals may converge, we performed a meta‐analysis on imaging studies of the sense of body ownership, modulated by external visuotactile stimulation, and studies on interoception, which involves the self‐awareness for internal bodily sensations. Using a multilevel kernel density analysis, we found that processing of stimuli of the two domains converges primarily in the supramarginal gyrus bilaterally. Furthermore, we found a right‐lateralized set of areas, including the precentral and postcentral, and superior temporal gyri. We discuss these results and propose this set of areas as ideal candidates to match multiple body‐related signals contributing to the creation of a multidimensional representation of the bodily self.

conflict generated when a synchronous brushing is administered to someone's hand, occluded from vision, and on a visible nearby rubber hand: under such circumstances, subjects "feel touches on the rubber hand" as if this was belonging their body (Botvinick & Cohen, 1998).
On the other hand, interoceptive signals also play a role in the construction of a coherent sense of the self (Craig, 2002;Critchley & Harrison, 2013). Interoception has been mainly studied via paradigms like the heartbeat perception, but there are several other sources of information for different bodily sensations, which have been investigating using self-evaluated assessments (interoceptive sensibility) and behavioral tests (interoceptive accuracy; Garfinkel, Seth, Barrett, Suzuki, & Critchley, 2015).
Signals coming from the inside and outside of the body may interact, and the interplay between the two domains is currently unclear.
From a behavioral point of view, Tsakiris, Jimenez, and Costantini (2011) have shown that interoceptive awareness may modulate the malleability of the sense of the self. They administered the RHI paradigm to a group of subjects showing low or high interoceptive accuracy, as measured by the Heartbeat Perception task, to find the counterintuitive finding that the less interoceptively accurate participants had a stronger incorporation effect of the rubber hand. Later, Suzuki, Garfinkel, Critchley, and Seth (2013) have shown that both subjective and objective measures of virtual-hand ownership are enhanced by cardio-visual feedback in time with the actual heartbeat, as compared to asynchronous feedback. They also have shown that these measures, correlated with individual differences in interoceptive sensitivity, were modulated by the integration of proprioceptive signals instantiated using real-time visual remapping of finger movements of the virtual hand (Suzuki et al., 2013).
Although it is not clear whether a supramodal form of bodily selfawareness exists, this evidence definitively suggests the existence of forms of integration between interoceptive and exteroceptive signals.
While it is possible that integration of these domains might occur by forms of structural and functional connectivity of separate specialized regions, the logic of the human brain organization also allows one to hypothesize the existence of higher-order neural systems where signals of the different sources are integrated (Zeki & Shipp, 1988).
To date, a quantitative investigation on this hypothesis and differences and commonalities between the two domains is lacking. Moseley, Gallace, and Spence (2012) have proposed a theoretical framework for the interaction between interoceptive and exteroceptive information in the bodily self-awareness. They proposed the existence of a body matrix, a set of neural structures that "serve to maintain the integrity of the body at both the homeostatic and psychological levels and to adapt to changes in our body structure and orientation" (Moseley et al., 2012). They have also hypothesized that the posterior parietal and insular cortices may play an important role in the body matrix. In a more recent theoretical review (Park & Blanke, 2019), it has been suggested the existence of at least two subnetworks supporting the integration between interoceptive and exteroceptive signals in bodily self-awareness: a premotor cortexintraparietal sulcus-insula network preferentially processing signals for self-identification, and a temporal parietal junction-posterior cingulate cortex-intraparietal sulcus network preferentially processing signals for self-location. Interestingly, the authors have proposed an overlap of these two subnetworks in the intraparietal sulcus (Park & Blanke, 2019). However, explicit evidence on the anatomical localization of cortical regions linking exteroceptive and interoceptive signals is limited and discrepant to what previously hypothesized. Recently, Blefari et al. (2017) in an fMRI experiment administered the full body illusion paradigm coupled with cardiovascular response and heartbeat awareness task to a group of healthy volunteers: they found and proposed that the Rolandic operculi, in coordinates falling inside ventral premotor cortex, may be the critical regions subserving the integration of external and internal bodily signals. This interesting finding, in line with early fMRI reports on the RHI paradigm (Ehrsson, Spence, & Passingham, 2004), is clearly in need of replication as it would move the emphasis on a premotor view of bodily self-awareness.

| Aim of the study
In the current study, we aimed at unveiling the neural networks underpinning exteroceptive and interoceptive signals, seeking for modality-specific and shared systems. As said, the overarching aim of this work was to test the existence of a shared neural system subtending the making of bodily self-awareness. To this scope, we submitted to a quantitative meta-analysis, the available neuroimaging studies concerning body ownership (external visuotactile information) and interoception (internal physiological information), which included psychological illusions involving diverse body parts and different approaches to assess interoception. We first performed two separate meta-analyses investigating which brain regions are associated with the sense of body ownership and the interoceptive awareness according to these paradigms. We then examined whether body ownership and interoception share common neural substrates. Based on previous findings, we hypothesized to find evidence for a role of highorder associative areas, for example, the parietal regions, where the merging of external and internal information is made possible by the convergence of different inputs. Because of the specific choice made on the to-be meta-analyzed materials, we will argue that any such regions may contribute to the building of a bodily self-awareness.

| Study selection
To select relevant neuroimaging studies concerned with brain activity related to interoception or body awareness, an extensive database search of peer-reviewed functional neuroimaging studies with no initial restrictions (regarding the type of publication or publication date) was conducted. We relied on the following sources: MEDLINE library, life science journals, and online books indexed in PubMed. Also, we reviewed the reference lists of these articles to find additional reports The data sets and associated contrasts included in this meta-analysis, met the following inclusion criteria: (a) description of standard Talairach and Tournoux (1988), or MNI coordinates (to enable comparison of reported peak activation across studies); (b) samples composed of unmedicated and untrained healthy adults; (c) corrected thresholds of significance established at a whole-brain level; (d) measurement of regional cerebral blood flow (e.g., through PET) or blood oxygenation (e.g., through fMRI); and (e) tasks tapping core processes of interoception or body ownership without assessing related high-level processes or pursuing more specific goals (such as correlations and/or interactions with other psychological constructs or demographic features).
Unlike a recent meta-analysis on interoception (Schulz, 2016), the present one was not limited to cardioception and included studies with a variety of tasks. We adopted this inclusive criterion because our study's aim was to compare the awareness of bodily signals (without limiting these to cardiac activity) and the ability to self-recognize bodily information. Likewise, the meta-analysis on body ownership included a variety of tasks manipulating the sense of body ownership.
For body ownership, the key words used to search the databases were as follows: "fMRI" <OR>"functional neuroimaging" <OR>"PET" <OR>"neural basis" <OR>neural correlate <AND>"body ownership" <OR>"body disownership" <OR> "rubber hand illusion" <OR>"virtual hand illusion." Following this search, our data set included 286 participants drawn from 39 contrasts derived from 16 studies.

| Multilevel kernel density analysis
For this study, we used the multilevel kernel density analysis (MKDA; Kober et al., 2008;Kober & Wager, 2010;Wager, Lindquist, & Kaplan, 2007;Wager, Lindquist, Nichols, Kober, & Van Snellenberg, 2009), which allows one to detect robust effects and to establish the level of consistency across findings (Costafreda, 2009;Lindquist, Satpute, Wager, Weber, & Barrett, 2016;Schurz, Radua, Aichhorn, Richlan, & Perner, 2014). The technique reduces variability between studies in search of convergent findings across a set of studies, to average results to identify brain areas that show consistent activity (Liang, Zou, He, & Yang, 2013;van den Heuvel & Sporns, 2013). The MKDA emphasizes the multilevel hierarchy of the meta-analytic input data by treating activation peaks as nested within a given contrast. The resulting contrast maps-instead of individual peaks-are the unit of analysis and through this, two levels of analysis (within-contrasts and between contrasts) are created. Therefore, the MKDA summarizes consistency across studies rather than across peak coordinates. An important consequence of this approach is that any single study that reports a large number of nearby peaks (due to differences in style of how regional effects are reported, voxel size, thresholding, or low spatial smoothness in the data) cannot have an excessive weight on the final results (Wager et al., 2009; for further explanations see also Supporting Information).
More specifically, the MKDA performs multiple nested analyses on individual peaks by: (a) nesting peak activation coordinates within contrasts, and contrasts within studies; (b) modeling the variability across peaks within a contrast, rather than just counting all the peaks without taking into account from which contrast or study the peaks come from-so that true effect sizes are assumed to vary between studies; (c) assessing statistical significance density maps by comparing against the null hypothesis whereby activated regions are randomly distributed throughout the whole brain (Kober & Wager, 2010;Wager et al., 2009).
To this end, one contrast indicator map (CIM) is created for each contrast, by convolving a 10-mm spherical kernel around each peak reported in this contrast. The CIMs are averaged and weighted by the square root of the sample size of each contrast so that studies with fewer participants are given less weight, while reports with a larger number of participants are given more weight. Additionally, CIMs are weighted with 0.75 for fixed-effects versus 1.00 for random effects analysis to reduce the impact of fixed-effect studies (Wager et al., 2007(Wager et al., , 2009; see also Supporting Information).

The resulting Summary Density Map built up of the individual
CIMs reflects the proportion of contrasts yielding activations near each voxel. Hence, the crucial measure of interest is the number of contrasts that produced activation near a voxel, rather than the number of individual activation peaks.
The general null hypothesis states that peak coordinates of activated regions are randomly distributed across the gray matter of the standard brain. To identify voxels with activations that exceed the frequency expected by chance, a threshold derived from a Monte Carlo Simulation with 5,000 iterations per analysis was used. Building on this, MKDA identified maps of activated clusters according to a "height" or "extent" based threshold. The height-based threshold encloses voxels that have proportions of contrasts inside the 10 mm kernel regions that exceed the maximum expected over the entire brain by chance (p < .05, family wise error rate-FWER corrected). To determine a cluster extent-based threshold, the largest cluster of contiguous voxels was saved after each Monte Carlo iteration (with primary alpha levels of .001, .01, and .05) and secondary FWERcorrected for spatial extent at p < .05. Beyond that, a combined map of voxels meeting both criteria (height [p < .05] and extent [primary alpha level: p < .05]) was computed using SPM8 contiguity assessment procedures (see Figure 1; Kober et al., 2008;Schulz, 2016). In conclusion, the statistics reported in text and tables as the "z-value" (Kober et al., 2008) correspond to the proportion of contrasts that T A B L E 1 List of studies included on Body Ownership and Interoception included in the meta-analysis Note: Stereotactic coordinates for the most consistent clusters across all body ownership studies according to a "height" and "extent" based thresholds.
Height-based threshold encloses voxels that have proportions of contrasts inside the 10 mm kernel regions that exceed the maximum expected over the entire brain by chance (p < .05, family wise error Rate-FWER corrected). Extent-based threshold encloses contiguous voxels outside the 10 mm of the clusters for the height-based threshold that showed greater activation than would be expected at a given level of chance (p < .001), and which are secondary FWER-corrected for spatial extent at p < .05.
F I G U R E ity resulting in a cluster-size with a minimum of 10 voxels is reported.
The clusters were analyzed using the xjView toolbox (http://www. alivelearn.net/xjview/). Furthermore, we compared the activation of body ownership and interoception by subtraction analysis in MKDA: separate maps constructed for both domains were subtracted to yield difference maps.
The same procedure was employed during the Monte Carlo randomization to establish a threshold for significant differences.

| Body ownership
For studies employing tasks addressing the concept of body ownership, peak areas of high concordance were detected in the inferior temporal gyri spanning to the inferior occipital lobes (right, z = 0.2; left, z = 0.18). Further, we found bilateral clusters in the inferior parietal lobes also involving the postcentral gyri with cluster centers in the supramarginal gyri (right, z = 0.14; left, z = 0.17). Additional cluster centers were located in the precentral gyri (right, z = 0.15; left, Note: Stereotactic coordinates for the most consistent clusters across all interoception studies according to a "height" and "extent" based thresholds. Height-based threshold encloses voxels that have proportions of contrasts inside the 10 mm kernel regions that exceed the maximum expected over the entire brain by chance (p < .05, family wise error Rate-FWER corrected). Extent-based threshold encloses contiguous voxels outside the 10 mm of the clusters for the height-based threshold, which showed greater activation than would be expected at a given level of chance (p < .001), and which are secondary FWER-corrected for spatial extent at p < .05.). z = 0.15), the cerebellum (right cerebellar tonsil, z = 0.16; right declive, z = 0.09; left declive = 0.08), the right parietal inferior lobe (z = 0.15), the right superior parietal lobe (z = 0.12), and the insular cortices (right, z = 0.08; left, z = 0.06; see Table 2 and Figure 1).

| Contrast analyses between body ownership and interoception
The  A cluster-based analysis of such convergence revealed by the conjunction analysis (p < .05, FWER-corrected, Figure 2) unveiled two large clusters with their cluster centers in the right and left SMG, F I G U R E 2 Results of the cluster analysis on the overlap. Panel A shows axial brain slices of the overlapping areas mapped on a standard brain template. Numbers above the slices indicate z-coordinates. Maxstat is the mean z-value of both meta-analyses. Panel B shows the brain areas (following AAL atlas) covered by the respective clusters. Voxel-size is 2 x 2 x 2 mm; SMG, supramarginal gyrus respectively (left SMG: 1,077 voxels; z = 0.14; right SMG: 1,003 voxels; z = 0.12). 1 Additionally, three smaller clusters were found in the right hemisphere. The maximum activation of the respective cluster was hereby found in the precentral gyrus (182 voxels; z = 0.11), the superior temporal gyrus (28 voxels; z = 0.07), and the postcentral gyrus (12 voxels; z = 0.08; see Table 4).

| DISCUSSION
We continuously receive different signals from our body, coming from both the inside and outside multisensory channels. Are there specific cerebral areas permitting the convergence of the two domains? Evidence on the topic was scanty with suggestions, based on qualitative reviews of the available evidence, that these levels of multisensory integration might be supported by the insular, posterior parietal cortex (Moseley et al., 2012;Park & Blanke, 2019), while one explicit experiment on the matter pointed to ventral premotor cortex (Blefari et al., 2017).
Here, we instead quantified the available evidence on where the internal and external signals may be integrated and demonstrated the specific importance of the inferior parietal cortex of both hemispheres, comprising the parietal portion of the Rolandic operculum and insula, together with right-lateralized areas such as the postcentral, precentral, and superior temporal gyri. In what follows, we will discuss in an orderly manner the brain regions mainly associated with body ownership, interoception, and the areas of convergence between the two domains.

| Body ownership
Our meta-analysis identified a compound set of occipitotemporal, premotor, parietal, and insular regions contributing to the sense of body ownership, consistent with the nature of the triggering paradigms. In keeping with previous studies and a very recent meta-analysis, this network most likely integrates bodily signals across diverse sensory channels (Grivaz, Blanke, & Serino, 2017).
The areas with the most significant clusters were the right inferior Another area with high convergent activation across studies was the left SMG. This parietal region has been associated with the decoding of self-location (Guterstam et al., 2015) and perceiving limbs in space in a body-centered reference (Brozzoli et al., 2012). In particular, during the RHI paradigm, being the real hand position remapped onto a prosthetic hand, such remapping associated with activity in the posterior parietal cortex closely reflects changes in the position sense of the arm (Brozzoli et al., 2012). Furthermore, the SMG is a part of the neuronal network integrating visuotactile input applied to the hand , and other findings indicate that this region also receives proprioceptive inputs (Berlucchi & Vallar, 2018;Freedman & Ibos, 2018;Paulesu, Frackowiak, & Bottini, 1997;Whitlock, 2017).
Our results further suggest that the right precentral gyrus plays a role in body ownership. Such an area is supposed to be implicated in the experimental manipulation of the sense of ownership concerning peripersonal space remapping (Brozzoli et al., 2012), and the multisensory integration of tactile-proprioceptive and visual inputs (Ehrsson et al., 2005).
Lastly, our meta-analysis also highlights the importance of cerebellum in body ownership. In the RHI paradigm, it has been shown that the activity of the cerebellum correlates with the perceived strength of the illusion (Ehrsson et al., 2004;Gentile et al., 2013;Petkova et al., 2011). Furthermore, it also contributes to distinguishing sensory signals generated by the self or by others, as in the case of the sense of touch (Blakemore, Frith, & Wolpert, 2001). In a recent fMRI study on anosognosia for hemiplegia, right cerebellar activation was the only distinctive finding between the veridical appreciations of voluntary motion of the nonparalyzed hand, and the delusional belief of having moved the paralyzed hand (Gandola et al., 2014). This suggests that the cerebellum might have a role in "closing the loop" in a body and action monitoring system. Moreover, the cerebellum is involved in the processing of synchrony and congruence among the senses (Blakemore et al., 2001;Ito, 2000;Miall, Weir, Wolpert, & Stein, 1993), integrating signals coming from different multisensory channels. Indeed, it has been suggested that the cerebellum is crucial for the detection of corresponding multisensory signals and the formation of cross-modal predictions .
Interestingly enough, unlike a recent meta-analysis on body ownership (Grivaz et al., 2017), we did not find the insula as an area of convergent activation of body ownership paradigms, except when using a less conservative threshold (see Table 2). Of course, the insula is part of the somatosensory network (see review in Bottini et al., 1995;Paulesu et al., 1997) and of the pain matrix and it may participate to the motor awareness (Karnath, 2005). One may speculate that, because the insula is involved in somatosensory perception from ground up, the subtle body ownership paradigms (e.g., the RHI) are not sufficient to give a consistent activation of the structure. Indeed, the difference sought in the paradigm if between the integration of synchronous touches compared to asynchronous touches to the subjects' hand and visual stimuli given to the rubber hand. However, one may wonder why other meta-analyses found body-awareness clusters in that region; the only other reason, we can invoke is a methodological difference behind our studies. Unlike Grivaz et al. (2018), we did not consider peaks resulting from small-volume corrections or ROIs, and we used a more conservative statistical threshold (FWER p < .05).

| Interoception
The meta-analysis of the studies on interoception confirmed the involvement of brain areas that have entered into common neuroscientific discourse when referring to brain foundations of bodily feelings (Craig, 2002). The area with the clusters of maximal significance in terms of peak height and spatial extent was in the bilateral insulae, widely considered the primary interoceptive cortical area (Craig, 2002;Critchley, 2004;Pollatos et al., 2007). It has been suggested that the right (anterior) insula mediates explicit awareness of internal bodily processes (Craig, 2002;Critchley, Wiens, Rotshtein, Öhman, & Dolan, 2004), the laterality being associated with the nature of afferents. Stimuli such as air hunger, pain, cardiac perception, and visceral perception are primarily projected to the right anterior insula via vagal pathways (Craig, 2002). Instead, stimuli like the subjective sense of fullness (Stephan et al., 2003) are said to provide input primarily to the left anterior insula (Craig, 2002;Kelly et al., 2012). This is in part consistent with a recent cardiac interoception meta-analysis showing a right hemispheric dominance for cardioception (Schulz, 2016). Our meta-analysis was not limited to cardioception and included studies with a variety of tasks (e.g., thirst, air-hunger, attention to spontaneous sensations, soft touch, and gastric balloon distension), possibly explaining the bilateral distribution of the clusters.
We also found highly convergent activation in the cingulate cortex (mainly the middle and anterior portions), which is-along with the insula-involved in emotional, homeostatic/allostatic, sensorimotor, and cognitive functioning (Craig, 2002;Critchley et al., 2003;Critchley, Tang, Glaser, Butterworth, & Dolan, 2005;Devinsky, Morrell, & Vogt, 1995). Resting-state imaging showed the anterior insula to be functionally connected with the anterior cingulate cortex and the middle cingulate cortex, while the mid/posterior insula seems to be connected with the middle cingulate cortex (Taylor, Seminowicz, & Davis, 2009). Notably, in patients suffering from irritable bowel syndrome abnormal rectal-evoked, fMRI responses in both, the insula and cingulate cortex have been identified (Davis et al., 2008).
Furthermore, the insula and the cingulate cortex are part of the so-called salience system, which is said to detect relevant environmental changes regardless of the sensory modality employed in the task (Downar, Crawley, Mikulis, & Davis, 2002). This evidence points in the direction that conjoint activation of the cingulate cortex and the insula contributes to salience detection (Seeley et al., 2007) through task-set maintenance (Dosenbach et al., 2007;Dosenbach, Fair, Cohen, Schlaggar, & Petersen, 2008) and sustained focal attention (Nelson et al., 2010). Thus, anterior cingulate cortex may act as an integrative hub for interoceptive perception (Kleckner et al., 2017).

As different interoceptive dimensions have different neuroanatomical
correlates (García-Cordero et al., 2016), while the insula seems to be a critical region for interoception in general, anterior cingulate cortex involvement could be dependent on the type of interoceptive processing (Couto et al., 2015). Also related to interoception, we found clusters in the thalamus, a relay of the lamina-1-spino-thalamocortical as well as the vagal pathway projecting physiological signals to the interoceptive network (Craig, 2002).

| Body ownership > interoception
It is worth emphasizing that the occipitotemporal areas discussed above were more frequently associated with the body-ownership paradigms rather than with the interoception ones. This distinction sur-

| Interoception > body ownership
As for body ownership, we also found specific associations in the case of interoception. These localized primarily in the insulae bilaterally.
Interestingly, the insular cortices also receive somatosensory stimuli for exteroceptive stimulation (Bottini et al., 1995;Burton, Videen, & Raichle, 1993). Their modulation through caloric vestibular stimulation can reverse attentional hemianaesthesia (Bottini et al., 1995). The same stimulation can revert the pathological sense of disownership observed in somatoparaphrenia (Bisiach, Jisa, Oni, & Vallak, 1991;Rode et al., 1992;Salvato et al., 2016. The receptive fields of the somatosensory insular or retro-insular neurons are vast with responses for stimuli to either side of the body or both sides of the body (Paulesu et al., 1997). These anatomo-physiological considerations suggest a noncomplete separation between interoceptive and exteroceptive signals for the sake of the generation of a sense of ownership instead, they suggest a gradient from regions more, but not exclusively, concerned with interoception, like the insulae, to regions more concerned with exteroceptive stimulation and body ownership (see below).
A close look to Figure 1 suggests the existence of a functional anatomical pattern, with a latero-medial gradient for the processing of exteroceptive rather than interoceptive stimuli: medial activations were mainly linked to interoception, whereas lateral activations were shared, or dominated by exteroception. This pattern reminds a similar gradient for the default mode network (DMN) and dorsal attentional network (DAN), which are both part of the brain networks observed at rest. Interestingly, the DMN has been associated with selfreferential mental activity and internally oriented emotional processing (Buckner, Andrews-Hanna, & Schacter, 2008;Gusnard, Akbudak, Shulman, & Raichle, 2001), whereas the DAN is active during any externally directed cognitive process (Corbetta & Shulman, 2002). This scenario fits well with our findings indicating a possible implication of distinctive resting-state networks in interoceptive (DMN internally driven activity) and the exteroceptive (DAN externally driven activity) processing.
These networks are typically anticorrelated at rest (Gao & Lin, 2012). However, the convergence of the two networks in task-based activation patterns suggests that their anticorrelation is not an inevitable rule; the same observation suggests that indeed the two networks, or part of them, that is, the areas of convergence, may move in the same functional direction when a sense bodily self-awareness is generated. Admittedly, this remains a suggestion, though, as a meta-analysis, by its nature, can only define areas of anatomical convergence but not the timing of such convergence and functional coherence.

| Convergence between body ownership and interoception
The largest areas of shared "meta-analytical activations" were found in a bilateral cluster centered in the SMG. The most likely homolog of the human SMG is the monkey area 7b, which, to our knowledge, has never been associated with interoception. However, the same area has connections with the granular insula and more generally with the limbic system (Friedman, Murray, O'Neill, & Mishkin, 1986;Hyvärinen, 1982;Mesulam, Van Hoesen, Pandya, & Geschwind, 1977), the main region that we associated with interoception. Accordingly, it becomes less surprising our identification of the SMG as the cortical region for both body ownership and interoception. In humans, it has been dem-  (Heydrich et al., 2018).
They found that a late somatosensory evoked potential component (P45) reflected the illusory self-identification with a virtual body, demonstrating that the combination of interoceptive and exteroceptive signals modulate activity in the parietal somatosensory cortex (Heydrich et al., 2018).
The parietal cortex also seems to play a role in integrating physiological signals, such as body temperature. Recent studies have suggested the existence of a link between the sense of body ownership and body temperature (Kammers, Rose, & Haggard, 2011;Moseley et al., 2008;Salvato, Gandola, et al., 2018;Sedda, Tonin, Salvato, Gandola, & Bottini, 2016;Tieri, Gioia, Scandola, Pavone, & Aglioti, 2017). In healthy subjects, the transient, illusory incorporation of the fake hand in RHI paradigm has been associated with a decrease of the real hand temperature (Moseley et al., 2008). Furthermore, limb temperature modifications in healthy subjects may affect the strength of the illusion of ownership toward the rubber hand (Kammers et al., 2011).
Furthermore, it has been recently suggested that the subjective experience of an external event (e.g., spatial perception) would result from the neural responses to visceral activity such as heartbeats. The integration of visceral input with visual perception would provide visual content with a first-person perspective (Tallon-Baudry, Campana, Park, & Babo-Rebelo, 2018). Tallon-Baudry et al. (2018) identified a brain network that may be responsible for the interplay between neural signals generated by heartbeats and visual awareness: this includes the rTPJ, the same area found by us. Its damage or disconnection can lead to USN (see Chelazzi, Bisley, & Bartolomeo, 2018) but also to the so-called "out-of-body experience," a situation in which patients perceive themselves as observing their body from an extracorporeal perspective (Blanke, Landis, Spinelli, & Seeck, 2004).
Interestingly, left caloric vestibular stimulation (CVS), associated with insular activations, may induce a temporary remission of both symptoms, also modulating body temperature, as it was previously found in a single right brain-damaged patient suffering from chronic somatoparaphrenia: the restored sense of limb ownership following CVS was associated with an increase of the body temperature (Salvato, Gandola, et al., 2018). Notably, left CVS acts on the brain areas found here (Lopez et al., 2012;Zu Eulenburg, Caspers, Roski, & Eickhoff 2012), a fact that fits well with a role of these areas in creating a point of contact between a sense of body ownership and interoception.
Finally, it is also important to note that our results partially con-

| CONCLUSIONS AND FUTURE DIRECTIONS
In conclusion, our data indicated that external and internal signals might converge in the SMG bilaterally together with a right-lateralized set of areas such as the precentral, postcentral, and superior temporal gyri. These higher-order brain areas are involved in integrating multisensory signals, and in recalibrating information from different incoming channels and spatial frames of reference.
Although disorders of body ownership are on record since 1942 (Gerstmann, 1942), and disorders of interoception have been repeatedly described (Critchley & Garfinkel, 2018;Salvato, Mercurio, et al., 2018), it remains to be investigated whether a higher order deficit of the bodily self-awareness exists with specific deficits due to a perturbed integration of the two dimensions. Our results set the rationale for future neuropsychological and brain stimulation studies that may explore the contribution and weight of each area found in this meta-analysis in the integration of the bodily self-awareness.

CONFLICT OF INTEREST
The authors declare no conflict of interest.

DATA AVAILABILITY STATEMENT
The data that support the findings of this study are openly available (see Supporting Information).