First person experience of threat modulates cortical network encoding human peripersonal space

Peripersonal space is the area directly surrounding the body, which supports object manipulation and social interaction, but is also critical for threat detection. In the monkey, ventral premotor and intraparietal cortex support initiation of defensive behavior. However, the brain network that underlies threat detection in human peripersonal space still awaits investigation. We combined fMRI measurements with a preceding virtual reality training from either first or third person perspective to manipulate whether approaching human threat was perceived as directed to oneself or another. We found that first person perspective increased body ownership and identification with the virtual victim. When threat was perceived as directed towards oneself, synchronization of brain activity in the human peripersonal brain network was enhanced and connectivity increased from premotor and intraparietal cortex towards superior parietal lobe. When this threat was nearby, synchronization also occurred in emotion-processing regions. Priming with third person perspective reduced synchronization of brain activity in the peripersonal space network and increased top-down modulation of visual areas. In conclusion, our results showed that after first person perspective training peripersonal space is remapped to the virtual victim, thereby causing the fronto-parietal network to predict intrusive actions towards the body and emotion-processing regions to signal nearby threat.

specifically activated by stimuli within the PPS, with matched visual and tactile receptive fields 9 (Rizzolatti et al., 1981a;Rizzolatti et al., 1981b;Colby et al., 1993;Graziano and Gross, 1993; Graziano 10 et al., 1994;Graziano et al., 1997;Duhamel et al., 1998;Graziano, 1999). The alignment between the 11 visual and tactile receptive fields allows for the encoding of multisensory space in a common Research in monkeys has found that these multisensory regions representing peripersonal space 19 support object manipulation (Maravita et al., 2003;Bremmer, 2005). More recently it has been 20 shown that electrical stimulation of the neurons in the macaque vPM and VIP causes the monkey to 21 display defensive behavior similar to sustained air puff evoked responses Graziano, 2003, 22 2004). These results suggest that the peripersonal space also functions as an area of safety around 23 the body and that the underlying brain regions allow for the initiation of defensive behavior towards 24 threats (Cooke and Graziano, 2003; Graziano and Cooke, 2006). In humans, behavioral work has 25 further supported the hypothesis that peripersonal space performs an important defensive function. 26 After a week, participants came back to the lab and followed the same procedure as during the first 1 session, but during this second session they viewed the VR scenario from the other perspective (e.g. 2 if they viewed it from third person perspective in the first session, then they viewed it from the first 3 person perspective in the second session). All other aspects of the session were identical. They also 4 filled out the VR questionnaire again at the end of the session and were debriefed about the 5 contents and meaning of the study. Again, the emotional state of the participants was assessed and 6 they were asked to contact the experimenter if they had any reoccurring thoughts or feelings, or 7 were otherwise affected by participating in this experiment. No participant reported to be distressed 8 by the experiment or have persisting thoughts or feelings about the experiment. 9

Design 11
The order of the VR training perspective (1PP vs. 3PP) between sessions was counterbalanced across 12 subjects, so that half of the males and half of the females had the session order 1PP-3PP and the 13 other halves had the opposite order. The 3D video that was watched during fMRI measurements was 14 identical in both sessions and was preceded and followed by 3 seconds of fixation.

Statistical analyses 4
Questionnaire analyses 5 The VR questionnaire contained questions relating to the subjective experience of the 3D social 6 threat video. The scores on the VR questionnaire were compared between sessions (Session; 1PP vs. 7 3PP) by conducting an ordinal logistic regression analysis, with Score as an independent variable and 8 Session as factor, thresholded at p < 0.05 (Fig. 2). From the VR experience questionnaire, we used the 9 scores on the question "To what extent have you experienced the situation as if it was real?", the 10 question "To what extent did you feel in the female body and lived the situation as if you were the 11 woman?" and the question "To what extent did you feel identified with the female body during the 12 experience?" to analyze, respectively, the perceived Plausibility, Body Ownership and Identification 13 during the viewing of the 3D social threat video. 14 15 Functional MRI pre-processing 16 The fMRI data were pre-processed and visualized using fMRI analysis and visualization software 17 BrainVoyager QX version 2.8.4 (Brain Innovation B.V., Maastricht, the Netherlands). Functional data 18 were corrected for head motion (3D motion correction, sinc interpolation), corrected for slice scan 19 time differences and temporally filtered (high pass, GLM-Fourier, 2 sines/cosines). For the ISC 20 analyses, it is recommended to spatially smooth the functional data with a Gaussian smoothing 21 kernel of slightly larger than double the original voxel size (Pajula and Tohka, 2014). Therefore, the 22 functional data was spatially smoothed using a Gaussian kernel with a FWHM of 5 mm. The 23 anatomical data were corrected for intensity inhomogeneity (Goebel et al., 2006) and transformed 24 into Talairach space (Talairach and Tournoux, 1988). The functional data were then aligned with the 25 anatomical data and transformed into the same space to create 4D volume time-courses (VTCs). probabilistic region reflects the cytoarchitectonic probability (10-100%) of belonging to that region. 8 We followed a procedure to obtain maximum probability maps as described in Eickhoff et al. (2006), 9 as these are thought to provide ROIs that best reflect the anatomical hypotheses. This meant that all 10 voxels in the ROI that were assigned to a certain area were set to "1" and the rest of the voxels were 11 set to "0". The ROIs were transformed from MNI space to Talairach space (as Talairach was used in 12 the other analyses). We extracted the Colin27 anatomical data to help verify the subsequent 13 transformations. In order to transform the ROIs and the anatomical data from MNI space to Talairach 14 space, we imported the ANALYZE files in BrainVoyager, flipped the x-axis to set the data to 15 radiological format, and rotated the data -90° in the x-axis and +90° in the y-axis to get a sagittal 16 orientation. Subsequently, we transformed the Colin27 anatomical data to Talairach space (Talairach was applied to each pairwise correlation in each voxel. Subsequently, a sum ZPF statistic for the 5 difference between the two conditions (1PP and 3PP) was calculated over all subject pairs of one 6 group and tested against the null hypothesis that each ZPF value comes from a distribution with zero 7 mean (no difference between 1PP and 3PP) using non-parametric permutation testing. The null 8 distribution was obtained by randomly flipping the sign of pairwise ZPF statistics before calculating 9 the sum ZPF statistic using 25000 permutations. Maximal and minimal statistics over the entire image 10 corresponding to each labeling were saved. The map was thresholded at α = 0.05 using the 11 permutation distribution of maximal statistic, which accounts for the multiple comparisons problem 12 by controlling the FWER (see Nichols and Holmes, 2002). regions, which give rise to false connections, these would be present in both conditions. By only 25 taking into account differences in connections between conditions we circumvent the above 1 mentioned problem. 2 3 1

Subjective experience of perspective 2
The results from the questionnaire analysis (Fig. 2) showed that first person perspective training 3 induced higher ratings of Plausibility, Body Ownership and Identification during the perception of 4 threat than the third person perspective training. We assessed the strength of perceived Plausibility, 5 Body Ownership and Identification that participants experienced while viewing the 3D video in the 6 MRI scanner by analyzing the VR questionnaires that were administrated at the end of each session. 7 All questions were scored on a 1 (Not at all) to 7 (Completely) Likert scale. The answer scores were 8 analyzed using an ordinal logistic regression analysis (see Methods section). 9 1PP session (Fig. 2C). The odds of having high Identification in the 1PP session is 3.398 (95% CI, 1.056 8 to 10.935) times that of the 3PP session (Wald χ2(1) = 4.208, p = 0.04). 9

Peripersonal space network involved during first person perspective induced threat perception 11
The results of the fMRI analyses confirmed our first hypothesis: the PPS network was more strongly 12 involved in first than third person perspective primed threat perception. We  Figure S1 and Table S1). First 22 person perspective priming induced higher ISC than third person perspective priming in left dorsal 23 and ventral PM, in left IPS, left SMG, bilateral SPL and bilateral PVC during perception of the 3D video 24 (Fig. 3, top). Additionally, after both first (Fig. 3, top) and third person perspective priming (Fig. 3,  25 bottom) differences in ISC were found in different areas of the bilateral PAC, left MT and right PSC. The results from the effective connectivity analyses did not confirm our second hypothesis -that 5 visual and somatosensory regions would send information to PM and IPS during first person 6 perspective primed threat perception. We calculated effective connectivity differences between the 7 two conditions in an identical set of regions using RFX ANCOVA analyses (see Methods, N = 20, 8 p[corrected] < 0.05). 9 Fig. 4. Differences in effective connectivity between first (1PP) and third person perspective (3PP) primed perception of an identical 3D threat video (RFX ANCOVA, N = 20, p[corrected] < 0.05). The arrows indicate the direction of the connectivity between regions that is unique for each condition. Abbreviations as in Fig. 3. 10 We found that during first person perspective primed threat perception directed connections from 11 PM, IPS, SMG and MT towards SPL were stronger (see Fig. 4 top). This suggests that information was 12 integrated in SPL. These findings are in line with the ISC results, which also emphasized changes in 1 these regions during the first person perspective session. Moreover, we found that bilateral PM 2 showed directed connections to many of the other regions in the network. Additionally, not shown in 3 Fig. 4, we found stronger directed connections from left PAC and left IPS to right ACC and from right 4 ACC to right AMG and left ACC after first person perspective training. Although we found that ISC was 5 reduced in the PPS network during third person perspective primed threat perception, the 6 connectivity results revealed a more complex situation (see Fig. 4 bottom). Contrary to the first 7 person perspective session, we find stronger directed connectivity from IPS to PM, but no integration 8 of information in posterior parietal cortex. Moreover, we found enhanced top-down connectivity 9 from PM, IPS, SMG, SPL and PSC towards visual areas PVC and MT. Additionally, not shown in Fig. 4,  10 we found a directed connection from left SPL to right ACC after third person perspective training.

Influence of perspective priming on the peripersonal space network 2
Our first hypothesis was that activation in the PPS network will be stronger when a threat is primed 3 to be in our own space (first person perspective) than in another's space (third person perspective). 4 The results of the ISC analyses indicated a clear effect of first versus third person perspective priming 5 on the neural activity in the PPS network. After first person perspective priming we found that all 6 regions of the PPS network, including PM, IPS, SMG, SPL and PSC, were more synchronized across 7 participants during PPS intrusion. This was not the case when participants were primed with a third 8 person perspective. We expected that information gathered during threat processing from PVC, PAC, 9 and PSC would converge via the IPS in the PM, as the PM should initiate the defensive responses. We 10 based this hypothesis on the fact that electrical stimulation of F4 and VIP in monkeys (human 11 homologues of superior vPM and ventral IPS) produces movements similar to defensive movements 12 followed by air-puffs Graziano, 2003, 2004). Research in humans also indicated the indicate that this network is more strongly involved during first than third person perspective primed 25 approaching threat perception.
After third person perspective training we observed that brain activity in the PPS network was not 1 consistent across participants. This indicates that after first person perspective training the PPS is 2 aligned with the virtual body and intrusion of this space synchronizes activity in the fronto-parietal 3 network, while after third person perspective training this does not occur. Indeed, our ISC findings 4 suggest that priming with a third person perspective does not activate the defensive PPS network 5 and this aspect of the results did not provide support for a shared PPS. However, the connections 6 between regions of the PPS network did show third person perspective specific modulations. We these findings. We found that brain activity is more synchronized across participants in AMG, INS and 11 ACC when threat is near, but only when the threat is perceived as directed to one-self (1PP 12 condition). We found no evidence for enhanced ISC for nearby threat in emotion processing regions