Characterising the short-term habituation of event-related evoked potentials

Fast-rising sensory events evoke a series of functionally heterogeneous event-related potentials (ERPs), which reflect the activity of both modality-specific and supramodal cortical generators overlapping in time and space. When stimuli are delivered at long and variable intervals (e.g., 10-15 seconds), supramodal components appear as a large negative-positive biphasic deflection maximal at the scalp vertex (vertex wave) and dominate over modality-specific components. Stimulus repetition at 1 Hz induces a strong habituation of these supramodal components, which largely reflect stimulus saliency and behavioural relevance. In contrast, the effect of stimulus repetition on lateralized modality-specific components has been less explored, and the few existing results are inconsistent. To comprehensively characterize how the different ERP waves habituate over time, we recorded the ERPs elicited by 60 identical somatosensory stimuli (activating either non-nociceptive Aβ or nociceptive Aδ afferents), delivered at 1 Hz to healthy human participants. We show that the well-described spatiotemporal sequence of modality-specific and supramodal ERP components elicited by the first stimulus of the series is largely preserved in the smaller-amplitude, habituated response elicited by the last stimuli of the series. We also modelled the single-trial amplitude of both the lateralised modality-specific wave (N1) and the centrally-distributed vertex waves (N2 and P2) elicited by the 60 stimuli. The vertex waves decayed monotonically, with a largest drop of response magnitude at the first stimulus repetition followed by much smaller decreases in subsequent repetitions. In contrast, the lateralised somatosensory waves did not habituate consistently across the block. New & Noteworthy We characterized the decay of event-related potentials (ERPs) elicited by identical fast-rising stimuli repeated at 1Hz. Our observations indicate that, although response amplitude is reduced by stimulus repetition, both non-habituated and habituated ERPs are obligatory contributed by both modality-specific lateralized components and supramodal vertex components. Only supramodal components decay sharply at the first stimulus repetition, whereas modality-specific components do not. This suggests that modality-specific and supramodal components of the ERP habituate to stimulus repetition at different timescales.


28
Fast-rising sensory events evoke a series of functionally heterogeneous event-related potentials (ERPs), 29 which reflect the activity of both modality-specific and supramodal cortical generators overlapping in 30 time and space. When stimuli are delivered at long and variable intervals (e.g., 10-15 seconds), 31 supramodal components appear as a large negative-positive biphasic deflection maximal at the scalp 32 vertex (vertex wave) and dominate over modality-specific components. Stimulus repetition at 1 Hz 33 induces a strong habituation of these supramodal components, which largely reflect stimulus saliency 34 and behavioural relevance. In contrast, the effect of stimulus repetition on lateralized modality-specific 35 components has been less explored, and the few existing results are inconsistent. To comprehensively 36 characterize how the different ERP waves habituate over time, we recorded the ERPs elicited by 60 37 identical somatosensory stimuli (activating either non-nociceptive Aβ or nociceptive Aδ afferents), 38 delivered at 1 Hz to healthy human participants. We show that the well-described spatiotemporal 39 sequence of modality-specific and supramodal ERP components elicited by the first stimulus of the 40 series is largely preserved in the smaller-amplitude, habituated response elicited by the last stimuli of the 41 series. We also modelled the single-trial amplitude of both the lateralised modality-specific wave (N1) 42 and the centrally-distributed vertex waves (N2 and P2) elicited by the 60 stimuli. The vertex waves 43 decayed monotonically, with a largest drop of response magnitude at the first stimulus repetition 44 followed by much smaller decreases in subsequent repetitions. In contrast, the lateralised 45 somatosensory waves did not habituate consistently across the block. (event-related potentials, ERPs). ERPs are functionally heterogeneous and reflect the activity of distinct 63 cortical generators overlapping in time and space (Sutton et al., 1965). Since these generators include 64 both primary sensory and associative cortical areas, the scalp distribution of the ERPs elicited by stimuli 65 of different modalities partly differs depending on the modality of the sensory input. However, when 66 elicited by isolated and intense fast-rising stimuli, the activity of generators reflecting supramodal neural 67 activities dominates over modality-specific activities . The scalp distribution of the 68 electroencephalogram (EEG) signal reflecting supramodal generators is virtually identical regardless of 69 the modality of the eliciting stimulus: it consists in a biphasic negative-positive deflection widespread 70 over the scalp and maximal at the vertex -often referred to as 'vertex wave' or 'vertex potential' 71 (Bancaud et al., 1953).

73
The vertex wave amplitude is maximal when fast-rising stimuli are presented using large and variable habituate over time is still missing. This is particularly important considering that previous studies 86 suggested that neural activity in different cortical regions may adapt to repeated stimulation at different 87 timescales: for instance, neural activity in associative regions elicited by trains of innocuous, 88 somatosensory stimuli decays faster than neural activity in sensory cortices (Forss et  We recorded EEG while delivering trains of 60 identical Aβ stimuli at 1 Hz. We characterized the ERP 103 habituation in three complementary ways. First, we compared the ERP amplitudes between the first 104 and last stimuli of the block. Second, we statistically assessed the presence of the main response 105 components in both the non-habituated ERP (i.e. the ERP elicited by the first stimulus of a series) and 106 the habituated ERP (i.e. the ERP elicited by later stimuli that elicit a stable, habituated response). The 107 rationale for this decision was the consistent observation that the amplitude of the main ERP waves 108 (i.e., vertex waves) decays only minimally after the first few stimulus repetitions (Ritter et al., 1968;109 Fruhstorfer et al., 1969;Fruhstorfer et al., 1970;Fruhstorfer, 1971;Greffrath et al., 2007;Mouraux et 110 al., 2013), a finding corroborated by the present results (Figures 1-4). Third, we fitted a number of 111 functions derived from previous studies (Fruhstorfer et al., 1970;Greffrath et al., 2007), to model the 112 time-profile of the decay of the supramodal negative and positive vertex waves. To cross-validate and 113 generalise our findings across different sensory pathways, we replicated the experiment in a separate 114 group of healthy participants, using radiant-heat stimuli that selectively activate skin nociceptors and 115 elicit sensations of Aδ-mediated pinprick pain. 116 117 118

120
Participants 121 122 Thirty-two healthy subjects (14 women) aged 19-31 years (mean ± SD: 23.6 ± 3.9) participated in the 123 study, after having given written informed consent. All experimental procedures were approved by the 124 ethics committee of University College London (2492/001).

126
Transcutaneous electrical stimulation of Aβ fibers 127 128 Innocuous stimulation of Aβ afferents consisted of square-wave pulses (100 µs duration), generated by 129 a constant current stimulator (DS7A, Digitimer, UK). Stimuli were delivered through a bipolar 130 electrode placed above the superficial radial nerve, and elicited a paresthetic sensation in the 131 corresponding innervation territory. Aβ detection thresholds were identified using the method of 132 ascending staircases, on the right hand. The detection threshold was defined as the average of the 133 lowest stimulus energy eliciting a sensation in 3 consecutive trials. Electrical stimuli were delivered at 134 approximately 300% of each individual's Aβ detection threshold. Stimulus intensity was slightly 135 adjusted to elicit sensations of comparable intensities on the left and right hands (mean ± SD, 17.4 ± 136 11.4 mA) and to make sure that the elicited sensation was never painful.

138
Cutaneous laser stimulation of Aδ and C fibers 139 140 Nociceptive stimuli were radiant heat pulses generated by an infrared neodymium:yttrium-aluminum-141 perovskite laser with a wavelength of 1.34 µm (Nd:YAP; Electronical Engineering, Italy). At this 142 wavelength, laser pulses excite Aδ and C nociceptive free nerve endings in the epidermis directly and 143 selectively, i.e. without coactivating touch-related Aβ fibers in the dermis (Bromm and Treede, 1984; Laser stimuli were delivered within a squared skin area (4 x 4 cm) centered on the dorsum of the hand, 147 encompassing the area in which the stimulation of Aβ afferents elicited the paraesthesia. The laser 148 beam was transmitted through an optic fiber, and its diameter at target site was set at ~6 mm by 149 focusing lenses. A visible He-Ne laser pointed to the stimulated area.

151
The method of ascending staircases used for identifying the detection threshold of Aβ stimuli was also 152 used to identify the detection threshold of Aδ stimuli. For the EEG recordings, the stimulus energy 153 was clearly above the activation threshold of Aδ fibers (0.53 ± 0.06 J/mm 2 ). This stimulus energy 154 elicited intense but tolerable pinprick pain sensations, of comparable intensities on the right and left 155 hands. Because variations in baseline skin temperature may modulate the intensity of the afferent 156 nociceptive input (Iannetti et al., 2004), an infrared thermometer was used to ensure that the hand 157 temperature varied no more than 1ºC across blocks. To avoid receptor fatigue or sensitization, the laser 158 beam was displaced after each stimulus by ~1 cm within the predefined stimulated area.

160
Experimental procedure 161 162 Participants sat comfortably with their hands resting on a table in front of them. They were instructed 163 to focus their attention on the stimuli, and fixate a yellow circular target (diameter: 1 cm) placed in 164 front of them at a distance of approximately 60 cm from their face. A black curtain blocked the view of 165 the hands. Throughout the experiment, white noise was played through headphones, to mask any 166 sound associated with the either type of somatosensory stimulation.

168
The experiment was performed on 32 participants, divided in two groups of 16 participants. One group 169 received electrical stimuli, and the other group received laser stimuli, using an identical procedure. Each 170 participant received the somatosensory stimuli in 10 blocks, separated by a 5-minute interval, during 171 which participants were allowed to rest. Each block consisted of 60 somatosensory stimuli delivered at 172 1 Hz: thus, each block lasted 1 minute. In each block, stimuli were delivered either to the right hand or 173 to the left hand. Right-and left-hand blocks were alternated. The order of blocks was balanced across 174 participants; half of the subjects started with a right-hand block, and the other half started with a left-175 hand block. At the end of each block, participants were asked to provide an average rating of perceived 176 stimulus intensity, using a numerical scale ranging from 0 ("no sensation") to 10 ("most intense 177 sensation"). This was done to ensure that the perceived intensity of the stimuli was similar across 178 blocks (rating variability (SD) across blocks: electrical stimuli, 0.2 ± 0.2; laser stimuli: 0.3 ± 0.4).

180
Electrophysiological recordings 181 182 EEG was recorded using 30 Ag-AgCl electrodes placed on the scalp according to the International 10- 1. Preprocessing. EEG data were preprocessed and analyzed using Letswave 6 194 (http://www.nocions.org/letswave/) and EEGLAB (https://sccn.ucsd.edu/eeglab/). Continuous 195 EEG data were band-pass filtered from 0.5 to 30 Hz using a Butterworth filter, segmented into epochs 196 using a time window ranging from -0.2 to 0.8 sec relative to the onset of each stimulus, and baseline 197 corrected using the interval from -0.2 to 0 sec as reference. Trials contaminated by large artefacts 198 (<10% in each block) were removed. Eye blinks and movements were corrected using a validated 199 method based on unconstrained Independent Component Analysis ("runica" algorithm of EEGLAB). 200 In all datasets, independent components related to eye movements showed a large EOG channel 201 contribution and a frontal scalp distribution. To allow averaging across blocks while preserving the 202 possibility of detecting lateralized EEG activity, scalp electrodes were flipped along the medio-lateral 203 axis for all signals recorded in response to left hand stimulation. 204 205 2. Statistical assessment of ERP components. To assess decay of the ERP response, we first statistically 206 compared the response amplitude at the beginning and at the end of the block (note that a more formal 207 quantification of the response habituation was provided by the modelling analyses described in the next 208 section). We performed four paired-sample t-tests to compare the amplitude of the response to the 1 st 209 stimulus vs the 2 nd stimulus, and to the 1 st vs the last (60 th ) stimulus. We also compared the amplitude 210 of the ERPs elicited by the 2 nd vs 3 rd stimulus, and 59 th vs 60 th stimulus. All t-tests were performed for 211 each time point of the ERP timecourse, and for all electrodes. Thus, this analysis yielded a scalp 212 distribution of t-values across time, for each modality. To account for multiple comparisons, significant 213 time points (p < 0.05) were clustered based on their temporal adjacency (cluster-level statistical 214 analysis). For each cluster, we calculated the pseudo-t statistic of the two conditions, estimated its 215 distribution by permutation testing (1000 times), and generated the bootstrap p values for testing the 216 null hypothesis that there were no differences in signal amplitude (Maris and Oostenveld, 2007). This 217 procedure identified the clusters in which the responses in two given conditions were significantly 218 different.

220
Second, we assessed the consistency of stimulus-evoked modulations of EEG amplitude across time, to 221 statistically evaluate whether EEG deflections in the post-stimulus time window (from 0 to +0.8 s) was 222 significantly greater than baseline. Specifically, we performed a one-sample t-test against zero (i.e. 223 against baseline) for each electrode and time point of the entire baseline-corrected waveform, using 224 cluster-level permutation testing. This analysis yielded a scalp distribution of t-values across time, and 225 was performed separately on the non-habituated ERP and on the habituated ERP of each modality. 226 The non-habituated ERP was derived by averaging all the responses elicited by the 1 st stimulus of all 227 blocks. The habituated ERP was derived by averaging the responses elicited by the 6 th to the 60 th stimuli 228 of all blocks. The decision of using these responses elicited by stimuli 6 th to 60 th as a proxy of the 229 habituated ERP was based on the observation that the amplitude of the main ERP waves decays only 230 minimally after the first 5 stimulus repetitions, as observed here (Figure 1-2, 4) where y is the peak amplitude of each given ERP wave, x is the trial number (from 1 to 60), e is the 250 Euler constant, and a, b, c are the parameters to be estimated using a least squares method. Model 251 fitting was performed on the signal averaged across subjects, rather than on each individual subject, 252 because the signal-to-noise ratio of the habituated N1 wave is very low at single-trial and single-subject 253 levels, making the extraction of N1 peaks not always reliable (Hu et al., 2010). We tested these specific 254 models of ERP decay given the previous evidence that the vertex wave decays sharply at the first The decay of the N2 and P2 vertex components was modelled separately, because these waves can be 258 independently modulated (Legrain et al., 2002;Hatem et al., 2007). To compare which model best fitted 259 the data, we calculated the adjusted r 2 and the Akaike information criterion of each model, corrected 260 for low sample size (AICc). The AIC is a relative estimate of the information lost in a given model, and 261 it allows a fair comparison between non-linear models of different complexity, i.e. even when they have 262 a different number of parameters. The lower the AIC, the better the model represents the measured 263 data. From the difference in AICc values, we calculated the probability that each model was correct, 264 with the probabilities summing to 100% (Burnham and Anderson, 2002). Finally, we tested for equal 265 variance of the residuals using the 'test for appropriate weighting', as implemented in Prism GraphPad 266 7.0. 267 268 269

271
Response waveforms and topographies 272 273 Group-average ERPs elicited by Aβ and Aδ stimuli are shown in Figures 1, 2 and 3. As expected, the 274 latency of Aδ-ERPs was longer than the latency of Aβ-ERPs, because Aδ fibers are thinly myelinated 275 and thus have slower conduction velocity than large-myelinated Aβ fibers (Mountcastle, 2005).  Figures S1-S2). 285 286 Figure 2 shows the lack of any significant amplitude decay of the lateralized waves (N1 and P4) elicited 287 by the 60 repeated somatosensory stimuli. In both stimulus modalities, these waves were much smaller 288 than the vertex waves, as expected (Valentini et al., 2012;Hu et al., 2014a). Paired t-tests showed no 289 consistent habituation of the N1 and P4 waves at the first repetition of either Aβ or Aδ stimuli (trial #1 290 vs #2), and no significant habituation of any lateralised wave between the first and last stimulus of the 291 block (trial #1 vs #60; Figure 2). Importantly, albeit small in amplitude, both the early N1 and the late 292  where x is the trial number. The AICc and the adjusted r 2 values for all models are reported in Table I: 311 the fitting of equation #2 gave the lowest AICc and the highest adjusted r 2 values. Furthermore, the 312 residuals of the winning models were homoscedastic (p > 0.999). In qualitative terms, winning model 313 #2 indicates that the amplitude of the examined peaks decays monotonically, with a fastest and 314 sharpest drop of response magnitude at the first stimulus repetition, followed by much smaller 315 decreases in the subsequent repetitions.

317
The fitting of the decay of the Aβ-N1 and Aδ-N1 waves was rather poor, possibly because of the low 318 signal-to-noise ration at single-trial level. The winning model was again equation #2, although the 319 coefficients of determination were low (adjusted r 2 = 0.28 for the Aβ-N1, and 0.33 for the Aδ-N1). 320 This indicates that a large proportion of the variance of the data was not described by the winning 321 model. Hence, none of the models that we chose a priori gave a sufficiently accurate description of the 322 data. Therefore, we explored post hoc the fitting of other models (linear, power, sigmoidal, logistic, 1-6 323 degrees polynomial, and Fourier functions), but also these models poorly explained the variability of 324 Aβ-N1 and Aδ-N1 peak amplitude across stimulus repetitions (all adjusted r 2 < ±0.1). Thus, we found 325 no consistent and reliable evidence of habituation of the N1 across the block. 326 327 328

330
In this study, we characterised the habituation of modality-specific and supramodal ERP components 331 elicited by 60 identical somatosensory stimuli (activating either Aβ non-nociceptive, or Aδ nociceptive 332 primary afferents) delivered at 1 Hz. Although the response amplitude was clearly reduced, the 333 spatiotemporal sequence of the ERP waves was overall preserved in the habituated response (Figures 3,  334 S1, S2). This was substantiated by point-by-point statistical analysis: both somatosensory-specific and 335 supramodal components typically observed in the ERP elicited by sporadic and unpredictable stimuli 336

Effect of stimulus repetition on supramodal ERP responses 343 344
The negative-positive vertex wave (VW) is the largest component of the EEG response elicited by 345 sudden sensory stimuli. Its high signal-to-noise ratio makes the VW amplitude measurable in single 346 trials, even when the response is habituated by stimulus repetition. Therefore, we were able to estimate 347 the amplitude of the negative (N2) and positive (P2) vertex waves for each of the 60 ERPs (Figure 4).

348
The decay of the negative and positive peaks was best modelled as follows: 349 where y is the peak amplitude of the N2 or P2 wave, x is the trial number, and a, b, c are the estimated 351 parameters. This indicates that the amplitude of both vertex waves decays monotonically, with a largest,  The similarity of the decay of the VW elicited by Aβ and Aδ stimuli (Figures 1, 3, 4)  results are informative with respect to this functional framework. Indeed, stimulus repetition did not 398 abolish the VW elicited by either Aβ or Aδ stimuli, although it reduced its amplitude already after the 399 first stimulus repetition. Therefore, even when stimulus saliency is reduced by contextual factors, there 400 is a residual activity of the VW generators, only minimally reduced after the first few stimulus 401 repetitions (Figures 1, 3, S1, S2). These findings point towards the existence of an obligatory VW 402 activity triggered by any sudden and detectable change in the environment, even when contextual 403 modulations minimize its behavioural relevance.

405
Extensive evidence from cell physiology indicates that neural habituation to repeated stimuli arises 406 from alterations of synaptic excitability. Even the simple gill-withdrawal reflex in Aplysia dramatically 407 habituates at the first stimulus repetition (Byrne et al., 1978), due to a decreased drive from the sensory 408 neurons onto follower motor neurons (Castellucci et al., 1970;Carew and Kandel, 1973). The temporal 409 profile of this short-term habituation follows a fast decay function (Carew and Kandel, 1973), strikingly 410 similar to that observed in this and other studies on the habituation of electrocortical responses in 411 humans (Fruhstorfer et al., 1970;Greffrath et al., 2007). These synaptic changes have been interpreted 412 as a hallmark of learning, and are central to the ability of the nervous system to adapt to environmental 413 events (Carew and Kandel, 1973). Interpreting the decay of neural responses as functionally relevant for 414 learning is not in contradiction with attentional interpretations: stimuli that are learned and recognized 415 are likely to require less attentional resources than novel stimuli, and stimuli that need to be learned are 416 typically more salient.

418 419
Effect of stimulus repetition on somatosensory lateralized responses 420 421 In somatosensory ERPs, the VW is both preceded and followed by other deflections of smaller 422 amplitude. These have a topographical distribution maximal over centro-parietal electrodes in the 423 hemisphere contralateral to hand stimulation. The earliest negative wave is usually referred to as N1 424

432
We showed that these modality-specific N1 and P4 responses are detectable not only in the response to 433 the first stimulus, but also in the habituated ERP response, as supported by the statistical assessment of 434 the scalp distribution of the ERP response elicited by both the first and the last stimuli of the series 435 ( Figure 3). This is important, given that a previous study using trains of intra-epidermal electrical 436 shocks at 1 Hz failed to observe any lateralized response . It is difficult to 437 reconcile these two different observations, and we can only speculate about why that previous 438 experiment failed to detect lateralised responses in the habituated response. One possibility is that intra-439 epidermal electrical stimulation causes a stronger peripheral and perceptual habituation, more 440 significant than for radiant heat stimulation .

442
Finally, our study does not provide evidence of any consistent effect of stimulus repetition on the 443 amplitude of somatosensory-specific waves (Figures 2 and 4). This might simply be due to the fact that 444 N1 and P4 waves have small amplitudes and poor signal-to-noise ratio at single trial level (Hu et al., 445 2010). However, a previous MEG study has also reported that neural activity originating from primary 446 somatosensory cortex is more resilient to stimulus repetition (2-Hz pneumatic stimulation of the 447 fingers and face): in other words, it decays to a less extent and more slowly than neural activity in 448 higher-order cortical regions, such as the posterior parietal cortex (Venkatesan et al., 2014).

450
In conclusion, our results provide a functional characterization of the decay of the different ERP 451 components when identical somatosensory stimuli are repeated at 1Hz. Fast-rising stimuli elicit ERPs 452 obligatory contributed by both modality-specific and supramodal neural activities, even when stimulus 453 repetitions minimize stimulus relevance. This indicates a fundamental and compulsory property of the 454 nervous system: its sensitivity to respond to sudden changes in the environment with a transient 455 synchronization of thalamocortical activity that manifests itself as widespread brain potentials 456 detectable in the human EEG. The large supramodal vertex waves decay sharply at the first stimulus 457 repetition, whereas smaller modality-specific waves do not appear to habituate consistently across 458 stimulus repetitions. These results suggests that modality-specific and supramodal components of the 459 ERP habituate to stimulus repetition at different timescales. 460  The asterisk (*) detonates the winning model for each wave and modality (lowest AICc and highest adjusted r 2 ).

Blind source separation using Probabilistic-ICA (pICA)
We decomposed non-habituated and habituated ERPs in functionally independent components, by performing a blind source separation with Independent Component Analysis ( When applied to multi-channel EEG recordings, unconstrained ICA separates the signals recorded on the scalp into a linear combination of independent components (ICs), each having a fixed scalp topography and a maximally independent time course. When ICA is unconstrained, the total number of ICs equals the total number of recording electrodes. If the number of ICs differs greatly from the actual number of independent sources contributing to the signal, this may constitute a critical problem (Beckmann and Smith, 2004). Indeed, if the number of ICs is much larger than the number of sources, ICs containing spurious activity will appear because of overfitting. On the contrary, if the number of ICs is much smaller than the number of sources, valuable information will be lost due to under-fitting. The problem of overfitting could be particularly important when unconstrained ICA is applied to averaged ERP waveforms. Because the averaging procedure cancels out sources of activity unrelated to the stimulus (e.g. ongoing EEG activity, muscular activity and noise), the number of independent sources present in the average waveform may be far smaller than the number of independent sources present in the original EEG signal.
These fundamental limitations can be addressed using pICA, in which the number of ICs is constrained to an effective estimate of the number of independent sources contributing to the original data (Beckmann and Smith, 2004). The number of independent sources was estimated using a method based on maximum likelihoods, and operating on the eigenvalues of a Principal Component Analysis (Rajan and Rayner, 1997).
pICA was conducted on the signals averaged across subjects, and was performed separately, for each of the two sensory modalities, on the non-habituated ERP (trial #1) and on the habituated ERP (average of trials #6-60). pICA was also conducted on the concatenated non-habituated and habituated ERP, i.e. on the average ERP waveform from trial #1 concatenated to the average ERP waveform from trials #6-60. The functional decomposition of Aβ and Aδ ERPs are presented in Figures S1 and S2. These figures show the topographic distribution of each independent component (IC), ranked according to the percentage of explained variance, together with their contribution to the ERPs at channel Cz and to the EEG global field power (GFP).
Aβ-ERPs. pICA identified five ICs in the ERP elicited by stimulus 1 (Figure S1A), and five ICs in the ERP elicited by the average of stimuli 6-60 ( Figure S1B).
In the ERP elicited by stimulus 1, IC #1 and #2 explained the majority of the P2 wave (GFP peak latencies: 241 and 293 ms), and were centrally distributed, with a maximum over the vertex electrodes. IC #3 and #4 explained the majority of the N2 wave (peak latencies: 113 and 143 ms), although their distribution was not fully centred on the scalp vertex. IC #5 contributed to both the earliest and the latest part of the ERP time-course, and likely reflected lateralised responses (peak latencies: 67 and 347 ms). Accordingly, its scalp distribution was contralateral to the stimulated hand. Both the scalp distribution and the timecourse of these components match well previous blind source separation of ERPs elicited by intense and isolated somatosensory stimuli .
In the ERP elicited by stimuli 6-60, IC #1 explained the majority of the P2 wave (GFP peak latency: 238 ms), whereas IC #2 explained the majority of the N2 wave (peak latency: 135 ms). IC #3 isolated EEG activities contralateral to the stimulated hand, occurring in both early and late time windows of the signal (peak latencies: 112 and 389 ms). Therefore, the neural activity isolated by IC #3 are likely to correspond to the N1 and P4 waves of somatosensory ERPs.
To further assess whether the same components contributed to the non-habituated and habituated ERP waves, we performed a pICA on the concatenated ERP waveforms elicited by stimulus 1 and by the average of stimuli 6-60. This analysis identified five ICs ( Figure S1C). IC #1 was symmetrically distributed over the vertex and explained the vast majority of the N2 wave in both the non-habituated and habituated response, as well as part of the P2 wave in the nonhabituated ERP. IC #2 was also maximal over Cz, and contributed to the majority of the P2 wave in the response elicited by both stimulus 1 and stimuli 6-60. No single IC unequivocally isolated neural activities corresponding to the lateralised N1 and P4 waves shown in Figure 2.
Aδ-ERPs. Probabilistic ICA identified six ICs from the EEG responses to stimulus 1, and four ICs from the EEG responses to the average of stimuli 6-60 ( Figure S2).
In the ERP elicited by stimulus 1, IC #1 and #2 had the typical topographical distribution of a vertex wave: IC #1 explained the majority of the P2 wave (GFP peak latency: 347 ms), and IC #2 explained the majority of the N2 wave (peak latency: 185 ms). IC #3 had a central-parietal distribution contralateral to the stimulated hand, and clearly reflected the late P4 wave observed in laser ERPs (peak latency: 423 ms) (Hu et al., 2014;Mancini et al., 2015). Probably because of the low signal-to-noise ratio consequent to the small number of stimulus repetitions, no IC unequivocally explained the early contralateral neural activity.
In the ERP elicited by stimuli 6-60, IC #1 was again centrally and symmetrically distributed and explained the majority of the P2 wave (peak latency: 307 ms). IC #2 had a contralateral centralparietal distribution contralateral to the stimulated hand, and explained the P4 wave (peak latency: 389 ms). ICs #3 had a maximal distribution over Cz and C3, and contributed to both the N1 and N2 waves (peak latency: 196 ms), possibly because of underfitting. IC #4 had a distribution contralateral to the stimulated hand and explained the early part of the N1 wave (peak latency: 138 ms).
pICA performed on the concatenated non-habituated and habituated ERP waves identified six ICs ( Figure S2C). IC #1 and #2 were symmetrically distributed over the vertex. They explained the majority of the P2 wave (IC #1) and N2 wave (IC #2) in the response elicited by both stimulus 1 and stimuli 6-60. IC #3 had a maximal distribution over Pz and P3, and explained the early N1 wave and part of the late P4 modality-specific waves, both in the non-habituated and habituated response. Finally, IC #4 had maximal distribution over C3, Cz, and C4, and explained a late positive wave in the response to both stimuli 1 and 6-60.
In conclusion, irrespectively of amplitude differences, the spatiotemporal pattern of the evoked brain activity was qualitatively similar not only for Aβ-ERPs and Aδ-ERPs (Treede et al., 1988), but also for the response elicited by stimulus 1 and stimuli 6-60.