Neurophysiological Correlates of the Rubber Hand Illusion in Evoked and Oscillatory Brain Activity

The rubber hand illusion (RHI) allows insights into how the brain resolves conflicting multisensory information regarding body position and ownership. Previous neuroimaging studies have reported a variety of neurophysiological correlates of illusory hand ownership, with conflicting results likely originating from differences in experimental parameters and control conditions. Here, we overcome these limitations by using a fully automated and precisely-timed visuo-tactile stimulation setup to record evoked responses and oscillatory responses in the human EEG. Importantly, we relied on a combination of experimental conditions to rule out confounds of attention, body-stimulus position and stimulus duration, and on the combination of two control conditions to identify neurophysiological correlates of illusory hand ownership. In two separate experiments we observed a consistent illusion-related attenuation of ERPs around 330 ms over frontocentral electrodes, as well as decreases of frontal alpha and beta power during the illusion that could not be attributed to changes in attention, body-stimulus position or stimulus duration. Our results reveal neural correlates of illusory hand ownership in late and likely higher-order rather than early sensory processes, and support a role of premotor and possibly intraparietal areas in mediating illusory body ownership.


1
Introduction 24 Philosophy, psychology, and neuroscience continue to debate the sources and modulators of 25 conscious experience. The scientific study of consciousness has long been focussed on the visual 26 domain, but recent decades have seen a rise of interest in bodily self-consciousness and the 27 integration of bodily signals with other multisensory information (Jeannerod, 2007). Bodily self-28 consciousness refers to the integrated, pre-reflexive experience of being a self in a body and has been 29 related to tactile, vestibular, proprioceptive, as well as visual and motor information (Blanke, 2012; 30 Tsakiris and Haggard, 2005). One extensively investigated aspect of bodily self-consciousness is the 31 experience that our body and its parts belong to us and are distinguished from non-body objects and 32 other people's bodies, so-called body ownership. A widely used paradigm to study body ownership is 33 the rubber hand illusion (RHI; Botvinick, 2004) during which participants watch an artificial rubber 34 hand being stroked in synchrony with strokes on their own occluded hand. This synchronous visuo-35 tactile stimulation alters bodily experience as it induces the illusion that the rubber hand is one's own 36 hand. 37 Several functional magnetic resonance imaging (fMRI) studies have aimed to identify the neural 38 correlates of illusory hand ownership. The experience of illusory hand ownership has been linked to 39 activity in frontal brain regions, such as the premotor cortex (Bekrater-Bodmann et al., 2014; 40 Ehrsson, 2004;Petkova et al., 2011), occipito-temporal regions such as the extrastriate body area 41 (Limanowski et al., 2014), intraparietal areas (Petkova et al., 2011), the anterior insula (Limanowski 42 et al., 2014), and the temporoparietal junction (Guterstam et al., 2013). However, given the nature of 43 the fMRI signal, these studies have not been able to provide a functionally specific picture that 44 assigns these neural correlates to a specific part of the sensory-perceptual cascade, for example by 45 assigning the relevant neural activations to a specific latency following each repeat of the visuo-46 tactile stimulation. 47 Overcoming these limitations, several electroencephalographic(EEG) studies have aimed to reveal 48 the physiological correlates of illusory hand ownership at higher temporal precision. One such study 49 has described the relative attenuation of somatosensory-evoked responses during the Illusion about 50 55 ms after stimulus onset (Zeller et al., 2015). This attenuation was localized to the primary 51 somatosensory cortex and the anterior intraparietal sulcus, and was interpreted in the context of 52 predictive coding models as an attenuated precision of the relevant proprioceptive representations 53 that are required to solve the multisensory conflict induced by the RHI. However, another EEG study 54 using a similar experimental paradigm reported illusion-related changes in ERPs only at much longer 55 latencies of around 460 ms over frontal electrodes (Peled et al., 2003). As a result it remains unclear 56 whether neural correlates of the RHI include aspects of early sensory encoding, hence at shorter 57 latencies relative to stimulus onset, or mostly involve higher cognitive processes emerging at longer 58 latencies relative to the touch stimulus. 59 The lack of clear insights from the existing EEG studies on the RHI may in part result from the use of 60 different stimulation parameters and the use of distinct control conditions. Two widely used control 61 conditions for the rubber hand illusion are the Incongruent condition, in which the rubber hand is 62 placed as an anatomically incongruent angle, and the Real condition, in which the rubber hand is 63 absent and stimulation occurs on the real hand in view (Ehrsson, 2004 In experiment 1 five conditions were administered in a randomised order (Fig. 1) motor was positioned right below the dummy vibration motor on the top storey. The vertical distance 129 between the two motors was 10 cm. Besides this, the setting was identical to the Incongruent 130 condition. In both experiments participants were instructed to use their right hand to press the right arrow key 151 on a computer keyboard when they felt the onset of the illusion and the left arrow key when they lost 152 the feeling of the illusion. Participants sat with their gaze fixed on the LED and wore ear plugs 153 throughout the experiment to reduce the noise caused by the vibration motors. 154

EEG Recording 155
Experiments were performed in a darkened and electrically shielded room. EEG signals were 156 continuously recorded using an active 64 channel BioSemi (BioSemi, B.V., The Netherlands) system 157 with Ag-AgCl electrodes mounted on an elastic cap (BioSemi) according to the 10/20 system. Four 158 additional electrodes were placed at the outer canthi and below the eyes to obtain the electro-159 occulogram (EOG). Electrode offsets were kept below 25 mV. Data were acquired at a sampling rate 160 of 500 Hz using a low pass filter of 208 Hz. Patuzzi, 1999). To detect potential artefacts pertaining to remaining blinks or eye movements we 173 computed horizontal, vertical and radial EOG signals following established procedures (Hipp and 174 Siegel, 2013; Keren et al., 2010). We rejected trials on which the peak signal amplitude on any 175 electrode exceeded a level of ±75 V, or during which potential eye movements were detected based 176 on a threshold of 3 standard deviations above mean of the high-pass filtered EOGs using procedures 177 suggested by Keren et al. (2010). Together these criteria led to the rejection of 34±8 % of trials 178 (mean±SD) in Experiment 1 and of 25±11 % of trials (mean±SD) of trials in Experiment 2. For 179 further analysis the EEG signals were referenced to the common average reference. 180 Condition averages of the evoked responses (ERPs) and oscillatory power (see below) were 181 computed by randomly sampling the same number of stimulation events from each condition. This 182 was necessary as the number of available trials differed across conditions. Condition averages were 183 obtained by averaging 500 times trial-averages obtained from 80% of the minimally available 184 number of trials. 185 To analyse oscillatory activity we extracted single trial spectral power for alpha (8-12Hz) and beta 186 (13-25Hz) using a discrete Fourier transformation on sliding Hanning windows with a length of 200 187 ms. Power values in the range of 100 ms pre-stimulus and 350 ms post-stimulus were averaged 188 across trials. No baseline normalization was performed but within-subject statistical comparisons 189 were used (see below), which make the subtraction of a common baseline unnecessary. 190 In experiment 1 our primary aims were to determine ERP and oscillatory signatures of the illusion 191 and to compare these to ERP and time-frequency signatures of attentional and body-related 192 processes. While we expected to find significant differences between the Illusion vs. the two control 193 conditions, and in the Attention and Body-stimulus position contrasts, we had no prior expectations 194 about the timing and localisation of significant differences. We hence used spatio-temporal  based Permutation Analysis to detect significant condition differences (Maris and Oostenveld, 2007). 196 A two-tailed paired t-test was performed for each electrode, and the cluster statistic was defined as Hand under, Two hands, Real, which differed along the factors of Attention (focussed, divided) and 203 Body-stimulus position (visual stimulus on body, visual stimulus not on body) in a 2x2 design (Fig.  204 1). To obtain the contrasts for each factor we averaged over the respective conditions belonging to 205 each level and then compared the averages with a cluster permutation test. To calculate the 206 interaction of Attention and Body-stimulus position factors, that is the difference between the 207 differences between the means of one factor, across the levels of the other factor, we subtracted Two 208 hands from Real, and Incongruent from Hand under, and compared these differences with a cluster 209 permutation test. left parietal areas (Tsum = 490.9, p<0.05) compared to the Incongruent condition ( Fig. 2A). At 227 330 ms the Illusion condition showed lower amplitudes in frontocentral regions compared to the 228 Incongruent condition and this frontocentral negativity was centred around electrode FCz (Tsum = -229 404.4, p<0.05, Fig. 2A). Significant differences between the Illusion condition and the Real condition 230 emerged around 330ms and were also centred around electrode FCz (Tsum = -823.1, p<0.05; Fig.  231 2B). The respective ERPs at electrode FCz suggest that the illusion is characterized by a more 232 pronounced negativity of the evoked activity around 330ms in compared to the two control 233 conditions (Fig. 2C). 234 To better localize the illusion effect we determined those electrodes that were part of both significant 235 effects around 330 ms, i.e. which were part of the significant time-electrode clusters in the Illusion-236 Incongruent and Illusion-Real contrasts. The resulting electrodes comprised the medial central and 237 centrofrontal electrodes (Fig. 2D). 238 While the timing and location of the attention effects do not resemble the illusion effect, the 257 topography of significant effects in the body-stimulus position contrast closely resembles the 258 topography of the illusion effect (c.f. Fig. 2D). The electrodes consistently involved in both effects 259 comprised medial central and centrofrontal electrodes (Fig. 4C), making it possible that potentially 260 similar regions are involved in mediating the illusion and body-stimulus effects, but reflect these at 261 distinct latencies relative to the stimulus. 262 We found no significant differences in oscillatory responses in the attention and body-stimulus 263 position contrasts in either the alpha (8-12Hz) or beta band (13-25 Hz). 264

Illusion effect -ERPs 266
In the second experiment we compared the Illusion to the Incongruent condition while manipulating 267 the duration of the visuo-tactile stimulation. We then performed a repeated-measures ANOVA on the 268 ERP amplitudes at the time-electrode cluster identified by the illusion effect in experiment 1 (c.f. Fig.  269 2D) to test the effects of illusion and stimulus duration ( We studied the neurophysiological correlates of the rubber hand illusion using a fully automated and 284 precisely-timed visuo-tactile setup and a combination of experimental conditions. Across two studies 285 and two control conditions we reliably found an illusion-related attenuation of ERPs around 330ms 286 over frontocentral electrodes. This effect was not related to attention or body-stimulus position 287 confounds and was robust against changes in stimulus duration. We furthermore found that 288 oscillatory activity in the alpha and beta bands was reliably reduced during the illusion. We thereby 289 provide multiple neural markers of the RHI. 290

Illusion effects in evoked responses 291
Several previous EEG studies have aimed to understand the neural correlates and mechanisms 292 underlying the illusory percept of body ownership in the RHI. These studies compared the evoked 293 responses associated with the tactile stimulus on the participant's hand between conditions inducing 294 the illusion and control conditions. The rationale behind this approach is to see whether and how the 295 cortical representation of the tactile stimulus changes when its subjective location changes from the 296 actual hand to the rubber hand. Previous studies differed regarding the latency of such an illusion-297 correlate in ERPs, reporting either early effects around 55 ms (Zeller et al., 2015) or much later 298 effects around 460 ms (Peled et al., 2003). However, both studies relied on the manual stimulation by 299 a brush handled by an experimenter, whereby each individual brush stroke can differ in timing and 300 intensity. This variability in the sensory stimulus can be detrimental for measuring the timing and 301 shape of the respective sensory evoked responses. To overcome this problem we here designed an 302 automated setup that allows visuo-tactile stimulation with great temporal fidelity and consistency 303 across trials. Furthermore, we asked subjects to indicate the onset of the rubber hand illusion during 304 each trial and hence were able to include only those stimulation events in the analysis during which 305 subjects actually reported the presence of the RHI. To facilitate this we only considered participants 306 that had previously and reliably experienced ownership over a rubber hand and were familiar with the 307 sensations associated with onset and presence of the RHI as determined by a pilot session. 308 To establish neural correlates of the RHI a comparison of the illusion condition with a control 309 condition is required. Most previous ERP studies relied on the Incongruent condition in which the 310 rubber hand is placed at an anatomically incongruent angle, or relied on the Real condition in which 311 the rubber hand is absent and stimulation occurs on the real hand in view (Peled et al., 2003;Zeller et 312 al., 2015Zeller et 312 al., , 2016. Using only one control condition makes the implicit and critical assumption that the 313 illusion and control conditions differ only in a single factor, the presence of the subjective illusion. 314 Yet, closer inspection of these conditions suggests that these may differ by other factors as well, such 315 as focus of attention and body-stimulus position in the Incongruent condition, or the absence of a 316 rubber hand in the Real condition. We therefore relied on the combination of control conditions to 317 identify potential changes in evoked activity that are reliably associated with the illusion. The need 318 to consider multiple control conditions is further demonstrated by the observation that some 319 significant ERP effects were observed only in one of the two contrasts (c.f. Fig. 2). For example, the 320 Illusion-Incongruent difference revealed a significant effect around 150 ms, which was absent in the 321 Illusion-Real difference, and hence unlikely is a correlate of the subjective illusion. This suggests that 322 results on the neural correlates of illusory body ownership that were obtained using a single control 323 condition have to be considered with care. 324 We found neural activations that were reliably associated with the illusion only at longer latencies 325 (here 330ms) over frontocentral regions. Furthermore, this illusion effect did not interact with 326 changes in stimulus duration. Together this suggests that these activations do not reflect processes 327 related to early sensory encoding but rather reflect late and higher-order processes. We did not administer any behavioural or physiological measures to measure the RHI, such as 368 proprioceptive drift measurements or changes in body temperature. The reason for this was twofold. 369 Firstly, we relied on a subjective measure of the illusion, as it allowed for uninterrupted recording of 370 EEG data across all conditions. Secondly, our study aimed to identify the correlates of the ownership 371 aspect of the RHI. As shown recently, proprioceptive drift does not provide a reliable assessment this 372 ownership aspect (Rohde et al., 2011). Rather, subjective ownership and the proprioceptive drift can 373 be dissociated, with the latter measuring the spatial updating of the body in space rather than the 374 strength of ownership over the rubber hand itself. 375

Illusion, attention and body-stimulus position 376
We used additional control conditions to reliably dissociate the neural correlates of the RHI from 377 attention and body-stimulus position related activity. Specifically, we identified the timing and 378 location of attention / body-stimulus position related effects and compared these to the activations 379 revealed by the two statistical contrasts obtained from the Illusion. By comparing conditions where 380 the visual stimulus was near the body with conditions where the visual stimulus was far from the 381 body, we found body-stimulus position related processing to be associated with activity in 382 frontocentral areas around 180 ms. This is in line with previous studies investigating the influence of 383 proximity of hands and visual stimuli. For example, Reed et al. (2013) recorded ERPs during a visual 384 detection task in which the hand was placed near or kept far from the stimuli. Similar to the results of 385 the current study, they found increased negativity in the Nd1 component around 180 ms in the near 386 hand condition (see also Sambo and Forster, 2009). The timing of the body-stimulus position related 387 activity (~180 ms) was notably different from that of the illusion effect (~330 ms). This differentiates 388 the illusion effect from body-stimulus position related activity. However, the topography of the body-389 stimulus position related activity at 180 ms was highly similar to that of our illusion effect at 330 ms. 390 Thus, it is possible that both effects may emanate from the same cortical networks related to body 391 processing. Support for this comes from a study by Brozzoli et al. (2012) who measured BOLD 392 response while presenting participants with visual stimuli occurring next or distant from their hands. 393 Their results indicated increased activity in premotor and intraparietal cortices in the condition where 394 the stimulus was close to the hand compared to the condition where the stimulus was distant form the 395 hand. Similar results were obtained when the participant's hand was replaced by a rubber hand on 396 which the RHI was induced (Brozzoli et al., 2012). This suggests that both, the effects of body-397 stimulus position and the illusion may originate from processing in the intraparietal-premotor 398 network but do so at different latencies relative to stimulus onset, further corroborating that the ERP 399 correlates of the illusory percept reflect sensory integration processes in the parietal-premotor 400 network. 401 We found attention related activity in frontal and parietal regions around 100 ms and around 250 ms.