Correction: Neural excitability and sensory input determine intensity perception with opposing directions in initial cortical responses

Perception of sensory information is determined by stimulus features (e.g., intensity) and instantaneous neural states (e.g., excitability). Commonly, it is assumed that both are reflected similarly in evoked brain potentials, that is, higher evoked activity leads to a stronger percept of a stimulus. We tested this assumption in a somatosensory discrimination task in humans, simultaneously assessing (i) single-trial excitatory post-synaptic currents inferred from short-latency somatosensory evoked potentials (SEP), (ii) pre-stimulus alpha oscillations (8-13 Hz), and (iii) peripheral nerve measures. Fluctuations of neural excitability shaped the perceived stimulus intensity already during the very first cortical response (at ∼20 ms) yet demonstrating opposite neural signatures as compared to the effect of presented stimulus intensity. We reconcile this discrepancy via a common framework based on modulations of electro-chemical membrane gradients linking neural states and responses, which calls for reconsidering conventional interpretations of brain potential magnitudes in stimulus intensity encoding.

associated with modulations of a stimulus´ percept in various sensory domains including the 23 visual (Busch et al., 2009;Iemi et al., 2017), auditory (Müller et al., 2013), and 24 somatosensory domain (Baumgarten et al., 2016;Craddock et al., 2017;Forschack et al., 25 2020). According to the baseline sensory excitability model (BSEM; Samaha et al., 2020), 26 higher alpha activity preceding a stimulus leads to a generally lower excitability level of the 27 neural system, resulting in a lower detection rate of near-threshold stimuli but no changes in 28 the discriminability of sensory stimuli. On a cellular level, such excitability modulations may 29 be reflected in changes of membrane potentials (Castro-Alamancos, 2009), which may occur 30 in an oscillatory manner (Lakatos et al., 2005) and shift the threshold for incoming sensory 31 information to be processed further downstream in the neural system. This notion has been 32 further supported by monkey studies showing that higher oscillatory activity within the alpha 33 band is associated with a lower neural firing rate (Bollimunta et al., 2011;Haegens et al., 34 2011). 35 However, it remains unclear up to now whether the influence of instantaneous 36 excitability on perceptual processes can be generalized to the intensity perception of stimuli 37 per se (i.e., beyond the sensory threshold) -which would have far-reaching implications for a 38 wide variety of studies in the field of perception. Moreover, if such modulation indeed occurs, 39 the question remains: At which stage of the neural response cascade do instantaneous Bi-variate relationships between pre-stimulus alpha amplitude, N20 peak amplitude, and perceived stimulus intensity. a) Time course of the amplitude of pre-stimulus alpha band activity (8-13 Hz) displayed by behavioral response categories. Note that for statistical analyses, pre-stimulus epochs were cut at -5 ms relative to stimulus onset before filtering the data in the alpha band (8)(9)(10)(11)(12)(13), in order to prevent contamination of the pre-stimulus window by stimulus-related activity. b) Change in perception bias (i.e., SDT parameter criterion c) from the lowest to the highest alpha amplitude quintile (as measured between -200 and -10 ms). c) SEP (tangential CCA component) sorted with respect to pre-stimulus alpha amplitude quintiles. Alpha quintiles were sorted in ascending order (i.e., 1 st quintile = lowest alpha amplitude). d) SEP (tangential CCA component) sorted according to behavioral response categories. e) Change in perception bias (i.e., SDT parameter criterion c) from the most to the least negative N20 peak amplitude quintile. All panels show the grand average across all participants (N=32). Shaded areas in panels a, c, and d, as well as error bars in panels b and e correspond to the standard errors of the mean based on the within-subject variances (Morey, 2008). Transparent circles in panels b and e reflect data of individual participants while black lines reflect the arithmetic mean on group level. component of the SEP reflect changes in instantaneous cortical excitability, a covariation of 119 these two measures should be expected (Stephani et al., 2020). Indeed, higher pre-stimulus intensity), as is it also becomes evident from the SEPs sorted by the behavioral response 142 categories (Fig. 2d). more negative) for higher stimulus intensities -thus showing an effect of opposite direction 146 on N20 amplitudes as compared to instantaneous cortical excitability. In order to disentangle 147 these effects of excitability and stimulus intensity, we examined their respective contributions 148 in a two-level structural equation model, with stimulus intensity, pre-stimulus alpha amplitude, 149 and N20 peak amplitude as predictors of perceived stimulus intensity on level 1 (within 150 subjects), and random intercepts as well as their variances on level 2 (between subjects), 151 including all single trials. Furthermore, we added the measures of compound nerve action 152 potentials of the median nerve (CNAP; Fig. 3a (CNAP) in response to the median nerve stimuli, measured at the inner side of the ipsilateral upper arm (shown for an exemplary subject). b) Grand average across participants (N=32) of the CNAP, displayed by stimulus and response types. c) Single trials of the compound muscle action potential (CMAP), measured at the M. abductor pollicis brevis (shown for an exemplary subject). d) Grand average across participants (N=32) of the CMAP, displayed by stimulus and response types. Shaded areas in panels b and d correspond to the standard errors of the mean based on the withinsubject variances (Morey, 2008).

166
On the one hand, both these peripheral measures should relate to stimulus intensity. 167 On the other, there should be no effect of CNAP and CMAP on N20 amplitudes, when 168 statistically controlling for stimulus intensity if the hypothesized fluctuations of excitability 169 emerge on a cortical level. Yet, stimulus-induced thumb twitches may influence the 170 participants´ intensity ratings of the stimuli (even though the stimulated hand was covered 171 with a paper box). The resulting two-level structural equation model (SEM; Fig. 4 Multi-level structural equation model of the interplay between pre-stimulus alpha activity, the initial cortical response (N20 component of the SEP), intensity of the presented stimuli, the peripheral control measures CMAP of the M. abductor pollicis brevis and CNAP of the median nerve, as well as the perceived intensity as reported by the participants. Effect paths were estimated between the manifest variables on level 1 (within participants). Latent variables on level 2 served to estimate the respective random intercepts as well as their between-subject variances according to the latent variable approach for multi-level models as implemented in Mplus.

) indicated
To evaluate the model fit, we compared a list of alternative models in-or excluding 175 relevant effect paths (Table 1). As indicated by Chi-Square Difference Tests, the log-176 likelihood of SEM 1 did not differ from those of SEMs 2-4. Seeking model parsimony, SEM 177 1 is preferred over SEMs 2-4 since the latter models included one more parameter each, while 178 fitting the data equally well. In comparison to SEMs 5-8, SEM 1 showed a significantly 179 higher log-likelihood suggesting a better model fit than these more parsimonious models. This is further supported by the AIC and BIC values which were altogether lowest for SEM 1. 181 Hence, we conclude that SEM 1 fitted our empirical data best. 182 The estimated path coefficients (Fig. 4) correspond well with above reported bivariate 183 relationships: When controlling for stimulus intensity, both higher pre-stimulus alpha 184 amplitudes and more negative N20 amplitudes were associated with a lower perceived 185 intensity (equivalent to a response bias as reflected in criterion c), as well as higher pre-186 stimulus alpha amplitudes co-occurred with more negative N20 amplitudes. In addition, the 187 SEM further dissociated the effects of stimulus intensity on early electrophysiological 188 measures and their respective effects on perceived stimulus intensity. Higher stimulus 189 intensity was associated with larger N20 amplitudes, which constitutes an effect of opposite 190 direction as compared to the N20-related excitability effect on perceived intensity. 191 Furthermore, higher stimulus intensity also led to larger amplitudes of CMAP and CNAP, due 192 to the physical difference in stimulation strength, as could be expected a priori. Additionally,193 larger CMAP amplitudes resulted in a higher perceived intensity, while no such effect was 194 observed for CNAP. Importantly, neither CMAP nor CNAP related to N20 amplitudes when 195 controlling for stimulus intensity. Thus, fluctuations in cortical processing were not driven by 196 peripheral variability. Finally, a substantial effect on the perceived intensity was found for 197 stimulus intensity. This was expected as the overall accuracy in the discrimination task was 198 about 70%. 199 Taken together, the SEM confirms the hypothesized influences of instantaneous 200 fluctuations of early somatosensory evoked potentials as well as pre-stimulus oscillatory 201 activity on the consciously accessible percept of a stimulus. Moreover, this analysis 202 demonstrates that stimulus intensity and cortical excitability, which in turn determines the 203 perceived stimulus intensity, show opposing effects on the amplitude of the early SEP. 204

Variability in thalamus-related activity is not related to behavioral responses 207
To examine further whether the observed neuronal effects on the perceived stimulus 208 intensity were of a cortical origin, we analyzed the EEG responses prior to the N20 potential. 209 In a sub-sample of 13 participants, the CCA decomposition provided a component that 210 showed a clear peak at 15 ms, characterized by a spatial pattern that suggested a deep, medial 211 source ( Fig. 5a & b). Most likely, this CCA component thus corresponds to the P15 potential 212 of the SEP, which is thought to reflect activity in the thalamus (Albe-Fessard et al., 1986). 213 The amplitude of this P15 component did not relate to the perceived stimulus intensity, as 214 examined with a random-intercept linear-mixed-effects model with perceived intensity as 215 dependent variable and P15 amplitude and stimulus intensity as predictors, β P15 = .008, 216 on perceived intensity. The post-hoc power analysis revealed a statistical power of 71.9%. 222 Therefore, we conclude that it is unlikely that the effect of N20 amplitudes on perceived 223 stimulus intensity was driven by thalamic variability and that the modulation of perceived 224 stimulus intensity emerges rather on the cortical level, reflecting instantaneous changes of 225 cortical excitability. 226 . c) Grand average (N=32) of later SEP components (extracted with the tangential-CCA filter in the frequency range from 0.5 to 45 Hz). The N140 is visible as a negative peak at around 149 ms. Larger N140 amplitudes are associated with higher perceived intensities. Shaded areas correspond to the standard errors of the mean based on the within-subject variances (Morey, 2008). component of the SEP, the N140. For this SEP component, a larger amplitude has typically 230 been associated with a stronger percept of the presented stimulus (e.g., Al et al., 2020;231 Schubert et al., 2006). Indeed, a comparable effect of N140 amplitude on perceived intensity 232 was also present in our data ( Using a somatosensory discrimination paradigm, we examined the modulation of 246 perceived stimulus intensity by instantaneous fluctuations of cortical excitability at initial 247 cortical processing. Both pre-stimulus alpha band activity and initial cortical evoked 248 responses were associated with a bias in intensity discrimination, suggesting that a lower 249 cortical excitability reduces the perceived intensity of sensory stimuli. Furthermore, we rule 250 out that variability in peripheral nerve activity accounted for these effects, in line with the notion of instantaneous excitability changes being intrinsic to cortical brain dynamics. 252 Intriguingly, elevated excitability and higher presented stimulus intensity resulted in opposing 253 amplitude effects on the initial stimulus-related response in the cortex, the N20 component of 254 the SEP. Based on the neurophysiological principles of the EEG generation, this finding may 255 be explained by a mechanistic link between pre-stimulus alpha activity and initial cortical 256 EPSPs through modulations of resting membrane potentials. 257

Fluctuations of cortical excitability affect the perceived stimulus intensity 258
In line with previous studies on the modulatory role of alpha oscillations on perceptual 259 processes (Craddock et al., 2017;Iemi et al., 2017), we found higher pre-stimulus alpha 260 amplitudes to be associated with a lower perceived intensity of somatosensory stimuli. This 261 was indicated both by an increased threshold (criterion c) of reporting a higher stimulus 262 intensity according to Signal Detection Theory (Green & Swets, 1966), and by the negative 263 relationship between pre-stimulus alpha amplitude and reported stimulus intensity in the 264 structural equation model. Moreover, sensory processing appeared to be modulated by 265 ongoing oscillatory activity already during initial cortical responses, as suggested by the 266 relations between pre-stimulus alpha activity and N20 amplitude, as well as N20 amplitude 267 and perceived stimulus intensity. The N20 component of the SEP reflects initial stimulus-268 related excitatory activity (i.e., excitatory post-synaptic potentials; EPSPs) resulting from the 269 first thalamo-cortical volley to the primary somatosensory cortex (Bruyns-Haylett et al., 2017;270 Nicholson Peterson et al., 1995;Wikström et al., 1996) and thus represents a direct measure 271 of cortical excitability (when keeping the stimulus intensity constant). The modulation of 272 perceived stimulus intensity therefore relates to a sensory bias at earliest possible cortical 273 processing, reflecting fluctuations of instantaneous neural excitability. 274 Furthermore, these findings demonstrate that effects of pre-stimulus oscillatory 275 activity on the processing of sensory stimuli are not restricted to near-threshold stimuli where 276 "weak") precluded the potential confounding effect of perceptual confidence, which has 281 recently been considered as an alternative explanation for pre-stimulus alpha effects on 282 perceptual biases (Benwell et al., 2017;Samaha et al., 2017). In our forced-choice paradigm, 283 different levels of perceptual confidence could not have influenced the intensity ratings since 284 the task was to distinguish two clearly perceptible stimuli, and not to report whether a 285 stimulus was perceived or not (as done in near-threshold paradigms). Thus, the current 286 findings unequivocally indicate -to the best of our knowledge for the first time -that pre-287 stimulus alpha oscillations affect the behavioral outcome via a modulation of the internally 288 represented stimulus intensity. 289 Opposing signatures of presented stimulus intensity and excitability in the early SEP 290 Following the hypothesis of higher alpha activity being associated with lower cortical 291 excitability (Jensen & Mazaheri, 2010;Klimesch et al., 2007;Samaha et al., 2020), it may 292 seem counter-intuitive that higher alpha amplitudes were associated with larger (i.e., more 293 negative) N20 amplitudes in our data. However, as we proposed recently (Stephani et al., 294 2020), this relationship may be explained by the neurophysiological mechanisms of EEG 295 generation. The generated voltage on the scalp, U, in our case relating to the N20 potential, 296 can be defined in the following way (Ilmoniemi & Sarvas, 2019;Kandel et al., 2000;Lopes 297 da Silva, 2004): 298 where denotes the sum of local primary post-synaptic currents due to the activation 299 of a given neuron, !"!"#$% the number of involved neurons, and the lead field coefficient 300 projecting source activity to the electrodes on the scalp. Since the spatial arrangement of the 301 neural generators and the EEG sensors was stable across stimulation events, LF reflects a 302 constant in the measurement of the N20 potential. In contrast, !"#$%!& should increase with 303 stimulus intensity since more nerve fibers are excited at stimulation site when applying 304 stimuli of higher currents. This should lead to an increase of SEP amplitude with stimulus 305 intensity, as reported in previous studies (Jousmäki & Forss, 1998;Klostermann et al., 1998) 306 and as was observed in the current dataset for cortical (Fig. 2d) as well as peripheral responses 307 ( Fig. 3b & 3d). For constant stimulus intensity, however, !"#$%!& is expected to stay 308 approximately constant and amplitudes in the EEG should primarily depend on , reflecting 309 excitatory post-synaptic currents (EPSCs) in case of the N20 component. Crucially, EPSCs 310 directly depend on the electro-chemical driving forces produced by the membrane potential. 311 When moving the membrane potential towards depolarization -a state of higher excitability -312 the electro-chemical driving force for further depolarizing inward trans-membrane currents is 313 decreased (Castro-Alamancos, 2009), which leads to smaller EPSCs (Deisz et al., 1991), and 314 should in turn result in smaller amplitudes of the scalp EEG. Assuming an inverse relationship 315 between the amplitude of alpha oscillations and neuronal excitability (as for example 316 indicated by a lower neural firing rate during higher alpha activity; Haegens et al., 2011), one 317 should hence rather expect decreased N20 amplitudes following low pre-stimulus alpha 318 activity. This is what was observed in our data when controlling for stimulus intensity (Fig. 4). 319 Moreover, the notion of smaller (i.e., less negative) N20 amplitudes reflecting a state of 320 higher excitability is corroborated by the behavioral data: When controlling for stimulus 321 intensity, we found less negative N20 amplitudes to be associated with higher perceived 322 stimulus intensity. 323 Taken together, our findings thus demonstrate that the intensity of the presented 324 stimulus and the degree of instantaneous neural excitability are jointly reflected in the early 325 potential, decreased N20 amplitudes appear to be associated with an increase in excitability 327 (which in turn lead to a higher perceived stimulus intensity). This challenges the prevailing, 328 prominent assumption that the amplitude of brain potentials, especially at early processing 329 stages, reflects the coding of the perceived stimulus intensity. Rather, our findings call for a 330 more differentiated view. Although the amplitude of early event-related potentials may 331 indeed reflect the size of the input (e.g., a stronger or weaker somatosensory stimulus), the 332 neural evaluation of this input (i.e., the perceived intensity), however, further depends on 333 internal neural states, such as neural excitability, which may even reverse the amplitude 334 effects of the input already at the earliest cortical processing stages. Crucially, our data are at 335 the same time consistent with previous studies on somatosensory processing at later stages, 336 where larger EEG potentials are typically associated with a stronger percept of a given 337 stimulus (e.g., Al et al., 2020;Schubert et al., 2006), as our analyses of the N140 component 338 showed. Thus, the present findings of opposing signatures of neural excitability and sensory 339 input appear to be a distinct characteristic of early cortical potentials, involving the first 340 bottom-up sensory processing. (We note, however, that the physiological interpretation of 341 amplitudes of later EEG potentials, such as the N140, is not as straightforward as described 342 above for the N20, since several distinct SEP components may interact (Auksztulewicz et al., 343 2012), and excitatory and inhibitory contributions cannot be readily distinguished.) 344

Origin of excitability fluctuations 345
To further narrow down the neuronal sources that eventually led to fluctuations of the 346 perceptual outcome, we controlled for peripheral nerve variability, extracted spatially well-347 defined EEG potentials, and examined subcortical activity. 348 Variability in afferent peripheral activity, as measured by compound nerve action 349 potentials (CNAP) at the upper arm, did not influence the perceived stimulus intensity when 350 controlling for stimulus intensity. However, a robust effect on the perceived stimulus intensity was observed for efferent peripheral activity, as measured by compound muscle action 352 potentials (CMAP) of the M. abductor pollicis brevis. This may be explained by differences in 353 proprioceptive sensations associated with the thumb twitches elicited by the stimulation, 354 whose extent may depend on changes of the prevailing muscle tonus. Importantly, neither the 355 CNAP nor the CMAP measure related to cortical excitability as measured by the N20 Another possibility is that already subcortical sources -particularly in the thalamus -365 may play a role in modulating sensory excitability and hence shape the perceptual outcome 366 (Kosciessa et al., 2020). Yet, our analyses of the thalamus-related P15 component did not 367 support this notion. Given the acceptable statistical power of these analyses, we conclude that 368 the modulation of perceived intensity in somatosensory stimulation has its neuronal origin at 369 the cortical level. 370 However, it remains an open question whether the observed excitability changes 371 reflect local or global neural dynamics. Although there is initial evidence that cortical 372 excitability may be organized temporally in a scale-free manner (Stephani et al., 2020), which 373 may reflect an embedding into global critical-state dynamics (Avramiea et al., 2020;Beggs & 374 Plenz, 2003;Palva et al., 2013), future work has to examine the spatial organization of 375 excitability more specifically across different somatotopic projections in primary sensory 376 areas as well as across diverse brain regions. 377 Both ongoing oscillatory alpha activity as well as amplitude fluctuations of the first 379 cortical response shape the perceived intensity of somatosensory stimuli. These effects most 380 likely reflect instantaneous changes of cortical excitability in the primary somatosensory 381 regions of the cortex, leading to a sensory bias which manifests already during the very first 382 cortical response. Challenging the common view of how the evaluation of stimulus intensity 383 is reflected in brain potentials, cortical excitability and the presented stimulus intensity were 384 associated with opposing effects on the early SEP. We argue that this disparity may be 385 Stimuli of two intensities were presented, in the following referred to as weak and strong 408 stimulus. The intensity of the weak stimulus was set to 1.2 times the motor threshold, leading 409 to a clearly visible thumb twitch for every stimulus. The individual motor threshold was 410 determined as the lowest intensity for which a thumb twitch was visible to the experimenter, 411 as determined by a staircase procedure. The intensity of the strong stimulus was adjusted 412 during training blocks prior to the experiment so that it was barely above the just-noticeable 413 difference, corresponding to a discrimination sensitivity of about d' = 1.5 according to Signal stimuli were only barely distinguishable (despite both being clearly perceivable), with average 416 intensities of 6.60 mA (SD = 1.62) and 7.93 mA (SD = 2.06), for the weak and strong 417 stimulus, respectively. 418 Procedure 419 During the experiment, participants were seated comfortably in a chair their hands 420 extended in front of them in the supinate position on a pillow. The left hand and wrist, to 421 which the stimulation electrodes were attached, was covered with a paper box in order to 422 prevent the participants to judge the stimulus intensity visually by the extent of thumb 423 twitches elicited by the stimulation. Weak and strong stimuli were presented with an equal 424 probability in a continuous, pseudo-randomized sequence with inter-stimulus intervals (ISI) 425 ranging from 1463 to 1563 ms (randomly drawn from a uniform distribution; 426 ISI average = 1513 ms). In total, 1000 stimuli were applied, divided into five blocks of 200 427 stimuli each with short breaks in between. Participants were to indicate after each stimulus 428 whether it was the weak or strong stimulus, by button press with their right index and middle 429 fingers as fast as possible. The button assignment for weak and strong stimulus was balanced 430 across participants. Furthermore, every sequence started with a weak stimulus in order to 431 provide an anchor point for the intensity judgments (participants were informed about this). 432 While performing the discrimination task, participants were instructed to fixate their gaze on a 433 cross on a computer screen in front of them. 434 Prior to the experiment, training blocks of 15 stimuli each were run in order to 435 familiarize the participants with the task and to individually adjust the intensity of the strong 436 stimulus so that a discrimination sensitivity of about d'=1.5 resulted (the intensity of the weak 437 stimulus was set at 1.2 times the motor threshold for all participants). On average across 438 participants, this procedure comprised 10.5 training blocks (SD=5.8). During these training 439 blocks, participants were provided with visual feedback of their response accuracy. No 440 information on task performance was given during the experimental blocks. 441 Data Acquisition 442 EEG data were recorded from 60 Ag/AgCl electrodes at a sampling rate of 5000 Hz 443 using an 80-channel EEG system (NeurOne, Bittium, Oulu, Finland) Butterworth filter forwards and backwards over the data to prevent phase shift (MATLAB 473 function filtfilt). As outlined in a previous study (Stephani et al., 2020), this filter allowed to 474 specifically focus on the N20-P35 complex of the SEP, which emerges from frequencies 475 above 35 Hz, and to omit contributions of later (slower) SEP potentials of no interest. 476 Additionally, this filter effectively served as baseline correction of the SEP since it removed 477 slow trends in the data, reaching an attenuation of 30 dB at 14 Hz, thus ensuring that 478 fluctuations in the SEP did not arise from fluctuations within slower frequencies (e.g., alpha 479 band activity). Subsequently, segments of the data that were distorted by muscle or non-480 biological artifacts were removed by visual inspection. After re-referencing to an average 481 reference, eye artefacts were removed using independent component analysis (Infomax ICA) 482 whose weights were calculated on the data band-pass filtered between 1 and 45 Hz (4 th order 483 Butterworth filter applied forwards and backwards). For SEP analysis, the data were 484 segmented into epochs from -100 to 600 ms relative to stimulus onset, resulting in about 995 485 trials on average per participant. EEG pre-processing was performed using EEGLAB 486 (Delorme & Makeig, 2004), and custom written scripts in MATLAB (The MathWorks Inc., 487 Natick, Massachusetts). 488 denoting the grand average of all trials. Since averaging cancels the background noise and 498 recovers the shared morphology of the SEP of interest among all the trials, the CCA 499 procedure resembles a template matching between the single trial signals and the template 500 time signature of the SEP of interest. The spatial filter ! provides us with a vector of 501 weights for mixing the channels of each single trial (i.e. !,!!" = ! ! ! ) and recovering their 502 underlying SEP. Therefore, ! can be interpreted as the spatial signature of the SEP of 503 interest across all single trials. The optimization problem of CCA can be solved using 504 eigenvalue decomposition. Therefore, multiple CCA spatial components can be extracted for 505 each subject, being the eigenvectors of the corresponding eigenvalue decomposition. Since we 506 are mainly interested in the early portion of the SEP, the two signal matrices and were 507 constructed using shorter segments from 5 to 80 ms post-stimulus. The extracted CCA spatial 508 filter was, however, applied to the whole-length epochs from -100 to 600 ms. The signal 509 resulting from mixing the single trial's channels using the CCA spatial filter ! , 510 i.e. !,!!" = ! ! ! , is called a CCA component of that trial. 511 The spatial activity pattern of each CCA component was computed by multiplying the 512 spatial filters ! by the covariance matrix of , as ( ) ! , in order to take the noise 513 structure of the data into account (Haufe et al., 2014). The CCA components whose spatial 514 patterns showed a pattern of a tangential dipole over the central sulcus (typical for the N20-515 P35 complex) were selected for further analyses and referred to as tangential CCA 516 components. Such a tangential CCA component was present in all subjects among the first 517 two CCA components with the maximum canonical correlation coefficients. Since CCA 518 solutions are insensitive to the polarity of the signal, we standardized the resulting tangential 519 CCA components by multiplying the spatial filter by a sign factor, in the way that the N20 520 potential always appeared as a negative peak in the SEP. 521 Furthermore, in a sub-sample of 13 subjects, a CCA component could be identified 522 among the first four CCA components, that showed a peak at around 15 ms post-stimulus 523 (presumably the P15 component of the SEP) and a spatial pattern that was characterized by a 524 central, outspread activation (in the following referred to as thalamic CCA component). Also 525 here, the CCA components were standardized so that the P15 always appeared as a positive 526 peak. 527 In order to additionally evaluate the later time course of the SEP (i.e., the lower and 528 later frequency content), the spatial filter of the tangential CCA component was applied to 529 EEG data temporally filtered between 0.5 and 45 Hz (apart from this, preprocessed in the 530 same way as described above). 531 SEP peak amplitudes and pre-stimulus oscillatory activity 532 N20 peak amplitudes were defined as the minimum value in single-trial SEPs of the 533 tangential CCA components ±2 ms around the latency of the N20 in the within-subject 534 average SEP. P15 amplitudes were measured from the thalamic CCA components as the 535 average amplitude in a time window ±1 ms around the latency of the P15 in the within-subject 536 average SEP. N140 amplitudes were measured from the low-frequency-filtered EEG (0.5 to 537 45 Hz), after application of the tangential CCA filter, as the average voltage in a time window 538 between 140 and 160 ms after stimulus onset. 539 To estimate the average amplitude of pre-stimulus alpha band activity, the data were 540 segmented from -500 to -5 ms relative to stimulus onset and band-pass filtered between 8 and 541 13 Hz, using a 4 th order Butterworth filter (applied forwards and backwards). In order to avoid 542 filter-related edge artifacts, the data segments were mirrored before filtering to both sides 543 (symmetric padding). Segmenting the data before filtering prevented any leakage from post-544 stimulus signals to the pre-stimulus time window. In order to examine pre-stimulus alpha 545 band activity of the same sources as of the SEP, the spatial filter of the tangential CCA 546 component was also applied to the pre-stimulus alpha data. Subsequently, the amplitude 547 envelope of the extracted alpha oscillations was computed by taking the absolute value of the 548 analytic signal, using Hilbert transform of the real-valued signal. To derive one pre-stimulus 549 alpha metric for every trial, amplitudes of the alpha envelope were averaged in the pre-550 stimulus time window of interest between -200 and -10 ms and log-transformed for 551 subsequent statistical analyses in order to approximate a normal distribution. 552

EEG source reconstruction 553
Sources of the EEG signal were reconstructed using lead field matrices based on 554 individual brain anatomies and individually measured electrode positions. Structural T1-555 weighted MRI images (MPRAGE) were segmented using the Freesurfer software 556 (http://surfer.nmr.mgh.harvard.edu/), and a three-shell boundary element model (BEM) based 557 on the segmented MRI was used to compute the lead field matrix with OpenMEEG (Gramfort 558 et al., 2010;Kybic et al., 2005). A template brain anatomy (ICBM152; Fonov et al., 2009) 559 was used for two subjects for whom no individual MRI scans were available. Additionally, 560 standard electrode positions were used for one subject for whom the 3D digitization of the 561 electrode positions was corrupted. The lead field matrices were inverted using the eLORETA 562 method (Pascual-Marqui, 2007), and sources were reconstructed for the spatial patterns of the 563 tangential CCA component of every subject. For group-level analysis, we projected the 564 individual source estimates onto the ICBM152 template anatomy using the spherical co-565 registration with the FSAverage template (Fischl et al., 1999) derived from Freesurfer. 566 Subsequently, the source estimates were averaged across subjects. Brainstorm (Tadel et al., 567 2011) was used for building individual head models and visualizing the source space data. 568 The MATLAB implementation of the eLORETA algorithm was derived from the MEG/EEG 569 Toolbox of Hamburg (METH; https://www.uke.de/english/departments-570 institutes/institutes/neurophysiology-and-pathophysiology/research/research-571 groups/index.html). 572

Processing of peripheral electrophysiological data (median nerve CNAP and thumb CMAP) 573
Analogously to the EEG data, stimulation artifacts were cut out and interpolated 574 between -2 to 4 ms relative to stimulus-onset using Piecewise Cubic Hermite Interpolating 575 Polynomials. To achieve a sufficient signal-to-noise ratio (SNR) of the short-latency CNAP 576 peak of only a few milliseconds duration on single-trial level, the data were high-pass filtered 577 at 70 Hz (4 th order Butterworth filter applied forwards and backwards). For the CMAP, no 578 further filtering was necessary given the naturally high SNR of muscle potentials (mV range). 579 Here, only a baseline correction was performed from -20 to -5 ms to account for slow 580 potential shifts. For the CNAP, single-trial peak amplitudes were extracted as the maximum 581 amplitude ±1 ms around the participant-specific latency of the CNAP peak that was found 582 between 5 and 9 ms in the within-participant averages. The CMAP was evaluated regarding 583 its peak-to-peak amplitude, which was defined as the difference between the minimum and 584 maximum amplitude measured ±1 ms around the participant-specific latencies of the negative 585 and positive peaks of the biphasic CMAP response (which were found between 5 and 11 ms 586 as well as 10 to 20 ms in the within-participant averages, respectively). quantified using sensitivity d', as calculated in the following way: 592 where !! corresponds to the inverse of the cumulative normal distribution, 593 " " ) to the probability of strong stimuli being rated as strong stimuli, and 594 " " ) to the probability of weak stimuli being rated as strong stimuli. 595 Response probabilities were calculated as the number of responses divided by the number of 596 stimuli of the respective categories. The response bias, criterion c, was calculated as follows: 597 = −0.5 * ( !! ( " " ))+ !! ( " " ))) . 598 According to Signal Detection Theory, sensitivity d' here represents the distance 599 between the distributions of the internal responses of the two stimuli, and thus reflects the 600 discriminability between strong and weak stimulus intensity. Criterion c reflects the internal 601 threshold above which a stimulus is rated as strong stimulus and below which a stimulus is 602 rated as weak stimulus, thus representing a general response bias. With respect to our data, a 603 higher criterion c therefore indicates a general tendency to report lower stimulus intensities. 604

Statistical analyses 605
To confirm that task accuracy was above chance level, we ran non-parametric 606 permutation tests (Crowley, 1992). Within each participant, we derived a null distribution of 607 chance-level performance by randomly remapping the behavioral responses with the 608 presented stimuli 100,000 times (Combrisson & Jerbi, 2015). The p value of the empirical approach: First, trials were sorted according to the amplitudes of the EEG measures. Next, the 614 SDT measures corresponding to the first and fifth quintile of the sorted trials were compared 615 using paired-sample t-tests. To quantify effect sizes, Cohen´s d was calculated as the mean 616 difference between the dependent samples divided by the standard deviation of differences 617 between the dependent samples. 618 The relationship between pre-stimulus alpha activity and the N20 component was 619 tested using a random-slope linear mixed effects model with pre-stimulus alpha amplitude as 620 predictor of N20 peak amplitude, and subject as random factor: 621 N20 peak amplitude ~ 1 + pre-stimulus alpha + (1 + pre-stimulus alpha | subject) .

622
The relationship between thalamus-related activity and intensity perception was tested 623 using a random-intercept linear-mixed-effects model with P15 amplitude and presented 624 stimulus intensity as predictors of perceived stimulus intensity, as well as subject as random 625 factor: 626 Perceived stimulus intensity ~ 1 + P15 amplitude + presented stimulus intensity + (1 | subject) .

627
Here, a logit link function was used to account for the dichotomous scale of perceived 628 stimulus intensity (note that we refrained from estimating a random slope for P15 amplitude 629 here given the small sample size of available data for thalamic activity). Analogously, we 630 analyzed the effect of N140 amplitude on perceived stimulus intensity, however now 631 including a random slope for N140 amplitude: Furthermore, we conducted a post-hoc power analysis to evaluate the probability of 639 finding an effect of P15 amplitude on perceived stimulus intensity if it was existent. For this, 640 we used Monte Carlo simulations with 1000 permutations based on the empirical dataset 641 (Green & MacLeod, 2016), assuming an effect size of β = .05, which is in the range of the 642 observed effect of N20 amplitude on perceived stimulus intensity. 643 In addition, the interrelation of pre-stimulus alpha activity, the N20 component of the 644 SEP, peripheral nerve activity as measured by CNAP and CMAP, the presented stimulus 645 intensity, as well as the perceived stimulus intensity were examined using confirmatory path 646 analysis based on multi-level structural equation modeling as implemented in the general 647 latent variable framework of Mplus (Muthén & Muthén, 1998. Pre-stimulus alpha 648 amplitude and presented stimulus intensity were included as exogenous variables, N20 peak 649 amplitude, CNAP amplitude, CMAP amplitude, and perceived stimulus intensity as 650 endogenous variables. The relationships contained in the hypothesized model are summarized 651 in Table 2. Trials with no behavioral response were excluded from the analysis. In total, 652 31,347 single trials were included in the SEM, with 979.6 trials on average per participant. 653 Model parameters were estimated using the MLR estimator provided by Mplus, a maximum-654 likelihood estimator robust to violations of the assumption of normally distributed data. A 655 logit link function was used to account for the dichotomous scale of perceived stimulus 656 intensity. The fit of the hypothesized model was examined comparing it to alternative models Level 1 (within participants): N20 amplitude ~ 1 + stimulus intensity + pre-stimulus alpha CNAP ~ 1 + stimulus intensity CMAP ~ 1 + stimulus intensity Perceived intensity ~ 1 + stimulus intensity + N20 amplitude + pre-stimulus alpha + CMAP Level 2 (between participants): N20 amplitude ~~ N20 amplitude CNAP ~~ CNAP CMAP ~~ CMAP Perceived intensity ~~ perceived intensity Pre-stimulus alpha ~~ pre-stimulus alpha Table 2. Relationships included in the hypothesized SEM. Level 1 equations reflect the withinparticipant effects between variables of interest. On level 2, only intercepts and variances of each variable were modelled; apart from stimulus intensity which only varied within participants by experimental design. coefficients based on Maximum Likelihood (ML). To derive a p value for the fixed-effect 671 coefficients, the denominator degrees of freedom were adjusted using Satterthwaite´s method

Data availability 678
The data that supports the findings of this study are available upon request from the 679 corresponding author (T.S.; stephani@cbs.mpg.de). The data cannot be made available in a 680 public repository due to the privacy policies for human biometric data according to the 681 European General Data Protection Regulation (GDPR). 682

Code availability 683
The custom-written code that was used for data processing and statistical analyses is 684 publicly available at https://osf.io/v9xa6/?view_only=3428139b7ab94824bac0eff0b4b92cc5.