Implicit agency is related to gamma power changes in an automatic imitation task

Often we have a feeling that we can control effects in the external world through our actions. The role of action processing associated with this implicit form of agency is still not clear. In this study, we used automatic imitation and electroencephalography to investigate neural oscillations associated with action processing and its possible contribution to implicit agency. Brain activity was recorded while participants performed actions (congruent or incongruent with a displayed finger movement) which subsequently triggered an outcome (a tone). We used a time estimation task to measure intentional binding (an index of implicit agency). We observed a decrease of alpha, beta and gamma power for congruent compared to incongruent actions and increased theta power for incongruent compared to congruent actions. Crucially, participants who showed greater intentional binding for congruent versus incongruent actions also presented greater gamma power differences. Alpha, beta and theta power were modulated by congruency but were unrelated to intentional binding. Our study suggests that an increased implicit agency for facilitated actions is associated with changes in gamma power. Our study also contributes to a characterization of neural oscillations in automatic imitation.


INTRODUCTION 51
Agency is the experience of being in control of our actions and their consequences (Haggard and Tsakiris, 52 2009). Two aspects of agency have been investigated so far, an implicit "feeling of agency" and a more 53 conscious "judgment of agency" (Synofzik et al., 2008(Synofzik et al., , 2013. Implicit agency is related to the causal 54 association between the actions we perform and the sensory consequences we perceive (Synofzik et al., 55 2008(Synofzik et al., 55 , 2013. This has been demonstrated in studies where voluntary (but not involuntary) actions and their 56 sensory consequences are perceived occurring closer in time than when they are isolated (Haggard et al.,57 2002; Moore and Obhi, 2012). This perceived time compression (bringing action and consequence closer) 58 is interpreted as the construction of a coherent conscious experience of our agency in time (Haggard, 2017;59 Haggard et al., 2002). This phenomenon is called intentional binding (IB) and it is said to reflect implicit 60 agency. A predominant view states that agency emerges from mechanisms that compare intended actions 61 and observed effects. However, a recent account suggests that mechanisms directly related to motor 62 commands and the execution of actions might play a major role in agency (Chambon and Haggard, 2012; agency. In this vein, fluency of actions can be also modulated by imitative behaviour. In a study conducted 65 by Vastano et al. (2017) agency was modulated with an automatic imitation task (Vastano et al., 2017(Vastano et al., , 2020. 66 In this task participants had to lift their index or middle finger in response to an imperative cue while 67 simultaneously observing a similar (congruent) or distinct (incongruent) finger movement of a mirrored 68 right-hand (Brass et al., 2000(Brass et al., , 2001Cracco et al., 2018). Movement execution was thus facilitated by 69 congruent and impeded by incongruent observed movements. Implicit agency was investigated measuring 70 IB. Participants were asked to report the time elapsed between their actions and subsequent outcome tones. 71 Intervals of time that were reported shorter relative to the real time intervals reflected a larger IB (Cravo et  with more time compression for congruent relative to incongruent actions. In a follow-up EEG study these 74 congruency effects correlated positively with late P300 changes and negatively with pre-response positivity 75 changes, suggesting that modulations of cognitive load and interference impact implicit agency (Vastano et 76 al. 2020). 77 78 Given the nature of automatic imitation it would be important to know if motor processes contribute to 79 implicit agency in this task. There is a set of candidate neural oscillations associated with action processing 80 that could be implicated in agency. Alpha band (8-12 Hz) oscillations generated in sensori-motor cortex are 81 modulated by the execution and observation of movements (Cochin et al., 1999;Koelewijn et al., 2008;82 delivered for 300 ms by means of headphones. Next, after a variable interval between 300 and 800 ms a 140 Visual Analogue Scale (VAS) appeared. The VAS ranged between 100 and 900 ms and has marks of 200 141 ms-intervals. In order to measure IB, participants were asked to estimate the interval of time between their 142 actions (key release) and the ensuing tone using the VAS. For this time estimation task participants used the 143 mouse to point in the VAS. They had a maximum of 5000 ms to give the answer. Finally, after a variable 144 inter-trial interval (1000, 2000 or 3000 ms) the next trial started. The critical manipulation in this experiment 145 is that the observed finger movements could result in a match or in a mismatch with the instructed finger 146 movement. In congruent (C) trials the participant moves the index finger and sees an index finger movement 147 (and similarly for middle fingers). In incongruent (IC) trials the participant moves the index finger and sees 148 a middle finger movement (or alternatively moves the middle finger and sees an index finger movement). 149 In addition, there was also a condition where the number was displayed but the fingers didn't move 150 (baseline; B). The experiment consisted of 240 randomized trials: 80 trials for each condition (congruent, 151 incongruent and baseline), each of which was composed by 26-28 trials for each interval (300, 400, and 500 152 ms). The experiment was divided in 4 smaller blocks of 60 trials each (20 trials in each congruent, 153 incongruent and baseline) to allow participants to rest between blocks. Before the main experiment the 154 participants were trained for the time estimation task and then were familiarized with the task. For the 155 training, participants listened two tones separated by 100 or 900 ms and then they were asked to indicate if 156 the time elapsed between the two tones were 100 or 900 ms. For this training they received a feedback 157 (correct or incorrect response). This short training phase (21 trials) aimed at capacitating the participants to 158 discriminate between 100 or 900 ms. For the sake of the manipulation in the main experiment the 159 participants were told the interval between their action and the tone was always random between 100 ms 160 and 900 ms. Following the training, the participants were familiarized with 30 randomized trials (15 for 161 each interval) identical to the main experiment. Once the instructions were clear the main experiment starts.     Mathworks, Inc, Natick, MA). The raw EEG data was loaded in EEGlab and filtered off-line.

ARTEFECT REJECTION 187
In a first step raw data was filtered between 0.5 and 40 Hz. Non-stereotyped artefacts were cleaned and bad 188 channels were detected by visual inspection. Bad channels (F6, FC6, T7, P2, and Iz across eight different 189 participants) were interpolated using spherical splines (Perrin et al., 1989). Next, stereotyped artefacts (such 190 as eye movements, eye blinks and muscle tension) were reduced by independent component analysis (ICA) 191 and the SemiAutomatic Selection of Independent Components (SASICA) toolbox (Chaumon et al., 2015), 192 removing no more than 3 components. ICA decompositions were performed separately on each subject over 193 all conditions and then saved. In a parallel step, raw data was not filtered. The same corrections regarding 194 non-stereotyped artefacts and bad electrodes done were re-applied to this dataset. Following this, the ICA 195 weights computed before were also re-applied to this dataset. Applying the pre-computed weights allowed  Greenhouse-Geisser correction was used to correct for sphericity violations when necessary. 231 232

RESULTS 233
In the automatic imitation task participants responded to a numerical cue that appeared between two fingers 234 while one finger simultaneously moved ( Figure 1). Following the response of the participants and after a 235 delay the participants listened to a tone and estimated the time elapsed between their actions and the tone 236 (this time estimation task allows us to measure IB). We set out to characterize the modulation of neural 237 oscillations occurring during automatic imitation to investigate its impact on IB. Before presenting these 238 results we will briefly present the behavioural results. 239 240 241

Performance and intentional binding increased for congruent actions 243
Behavioural results from this study have been reported in a previous publication (Vastano et al., 2020). 244 Accuracy was higher for the congruent condition (98.4%) followed by the baseline (97.1%) and the 245 incongruent (87.5%) condition. Similarly, reaction times were the fastest for the congruent condition (427 246 ± 63 ms, mean ± standard deviation) followed by the baseline (475 ± 70 ms) and incongruent (493 ± 78 ms) 247 conditions. Thus, as typically observed in the automatic imitation task, performance of the participants is 248 facilitated by congruent relative to the baseline and incongruent conditions. 249

250
In this task we evaluated whether time compression between actions and outcomes (the IB effect) would be 251 modulated by congruency. Participants performed a time estimation task following their responses to the 252 numeric cues. IB was measured as the difference between the participants' estimations and the real interval 253 durations (time judgements errors). Larger temporal compression reflects an increased IB effect. We 254 observed that the congruent condition (−72 ± 81 ms) led to significantly reduced time judgment errors than 255 the incongruent condition (−62 ± 74 ms), while a marginally significant difference was observed between 256 the congruent and the baseline (−61 ms ± 80 ms) conditions. This means that congruency in the automatic 257 imitation task can modulate IB levels. There is increased IB when the participants perform congruent 258 compared to incongruent actions. Interestingly, the IB levels produced by incongruent and baseline trials is 259 very similar, suggesting that facilitated actions can increase IB levels. We then examined whether gamma activity would be modulated by congruency. We found a significant

282
Gamma power differences correlates with intentional binding effects 283 Our second aim was to investigate whether gamma power changes were related to the IB effects. To assess  suggests that the link between gamma power and IB effects might be driven by a facilitatory effect. 315 However, conducting a robust correlation these effects become non-significant (r = 0.43; h = 0). Taking 316 altogether, we observe an overall effect where those participants that show greater time compression for 317 congruent versus incongruent actions also presented greater gamma power decreases. We also observe a 318 trend where participants that showed greater increased facilitatory effects in IB also showed greater gamma 319 power decreases. This trend is in favour of the notion that changes in gamma power on IB might be driven 320 by a facilitatory effect albeit the current data do not allow us to make a firm conclusion. 321

322
Decreased alpha and beta power for congruent actions 323 Next, we characterized changes in alpha and beta power in the automatic imitation task. We determined the 324 time-frequency windows where changes in alpha and beta occurred across all conditions. We observed a 325

357
Following this characterization, we investigated whether alpha and beta power changes were relevant for 358 IB effects. To assess this relationship, we performed the same correlations analyses described for gamma 359 power. We correlated differences in alpha and beta power with IB effects over participants. Alpha and beta 360 power differences did not correlated with the IB effects (alpha: r(26) = -0.03; p = 0.86; beta: r(26) = -0.29; 361 p = 0.14). 362 363

Increased theta power for incongruent actions 364
Lastly, we characterized changes in theta power in the automatic imitation task. We determined the time-365 frequency windows where changes in theta activity occurred across all conditions. We observed a peak of 366 theta power confined at FCz in the 4 and 8 Hz band between -0.2 and 0.1 s relative to response onset (see 367 Figure 6B, "All" conditions). Next, we examined whether theta power was modulated by congruency. We

381
Following this characterization, we investigated whether theta power changes were relevant for IB effects. 382 To assess this relationship, we performed the same correlations analyses described for gamma power. We 383 correlated differences in theta power with IB effects over participants. Theta power differences did not 384 correlated with the IB effects (r(26) = 0.21; p = 0.27). Altogether, alpha and beta power together with theta 385 power show no reliable relationship to IB. 386 387

DISCUSSION 388
We aimed at investigating the role of neural oscillations generated in sensory-motor regions in relation to 389 the implicit feeling of agency as well as how they are modulated by automatic imitation. We manipulated 390 the congruency between observed and planned actions and quantified the temporal compression (between 391 actions and tones) experienced by the participants (the intentional binding, IB, effect). Participants have 392 stronger IB when they performed congruent rather than incongruent actions in relation to the observed finger 393 movement. Furthermore, participants that showed greater IB for congruent versus incongruent actions 394 (congruency effect) also presented greater gamma power differences for congruent versus incongruent 395 actions. Increased power desynchronization in alpha and beta was observed for congruent relative to 396 incongruent trials, while increased power synchronization was observed in theta for incongruent relative to 397 congruent trials. These frequencies were modulated by congruency but were unrelated to IB. In the sections 398 below we will discuss the putative role of gamma oscillations for implicit agency and the significance of 399 our findings for understanding automatic imitation. 400

401
Implicit agency is related to gamma power changes 402 Our primary finding is that the implicit agency (as measured by IB) was associated with changes in gamma to manipulations of sensory feedback. In our study, we manipulated the congruency of executed and 406 observed finger movements, thus modifying the fluency of response selection. We expected to find 407 modulations across our three candidate neural oscillations (alpha, beta and gamma). However, we only 408 found that modulations in gamma power were associated with IB. The changes in alpha and beta power 409 were unrelated to IB and seem to reflect mirror activity (see below). Previous studies suggest that the motor 410 system is involved in the mental representations of actions performed by oneself and by others. However, 411 by its very nature these resonance mechanisms cannot realise the function of self/other distinction (Brass 412 and Heyes, 2005) which may be necessary for processes like agency. Thus, a possible interpretation is that 413 mirror activity does not contribute to agency in this task. This aspect needs to be investigated further. The 414 effect of congruency on gamma oscillations was associated with IB. Participants with greater congruency 415 effects in IB (i.e. greater time compression differences between congruent and incongruent actions) also 416 presented greater differences of gamma power between congruent and incongruent actions. One 417 interpretation is that gamma oscillations may have a role in agency possibly reflecting monitoring of actions. shown a decrease in the readiness potential for the congruent action (relative to the other conditions), 433 supporting the notion that congruency effects are driven by action facilitation (Deschrijver et al., 2017). It 434 is also worth noticing that in our previous study the correlation between the IB effect and late P300 changes 435 was also driven by a facilitatory effect (Vastano et al., 2020). Increased late P300 amplitudes for congruent 436 actions were interpreted as reflecting reduced cognitive load and accessible attentional resources. Together 437 our ERP and gamma band neural oscillations findings suggest that monitoring and executive attention 438 processes (both building blocks of executive functions) are affected by the congruency between observed 439 and executed actions and these processes in turn affect implicit agency. 440

441
One distinctive aspect of this study is worth mentioning. There is a strong theta response elicited for the 442 incongruent relative to the congruent condition. Theta responses have been typically associated with 443 cognitive control and executive processes including response inhibition (Cohen and Ridderinkhof, 2013;444 Gulbinaite et al., 2014;Nigbur et al., 2011). Although here we focused neural oscillations from sensorimotor 445 regions, theta oscillations could have also contributed to implicit agency insofar these are neural responses 446 that signal interference or conflict and could affect fluency (Cohen, 2014). We correlated the congruency 447 effects on IB with the theta effects but we did not find any relationship. This is surprising given that gamma 448 oscillations correlate with IB and we interpret that these effects are associated with action facilitation. 449 However, previous evidence shows that theta and gamma oscillations might reflect distinct aspects of 450 cognitive control. Isabella et al. (2015) used a modified go/no-go task that included a switching condition 451 and shows that theta oscillations were broadly related to cognitive control (rather than to response inhibition 452 or switching) while gamma oscillations were specifically associated with switching responses. The increase 453 of gamma power for the switch condition could be analogous to the incongruent condition in our automatic 454 imitation task. This again highlights the role that gamma oscillations can have for monitoring and updating 455 motor programs, in complement to theta oscillations. Future studies will need to further investigate this 456 interesting aspect of motor control. The role of gamma oscillations should also be related to its location in 457 the cortex. Gamma oscillations have been shown to be generated in primary motor cortices but also in brain 458 areas such as the supplementary motor area (SMA) and the premotor cortex (Ball et al., 2008;Szurhaj et al., 459 2005). In our study, gamma effects seem to be produced by central generators in the brain (although EEG 460 does not allow to draw strong conclusions). Interestingly, the SMA has an important role in voluntary 461 movements (Deiber et al., 1991(Deiber et al., , 1996 and timing perception (Wiener et al., 2010). In fact, activity in the 462 SMA is directly related to the time compression between voluntary actions and outcomes experienced by 463 participants in a time estimation task (Kühn et al., 2013). In sum, our study suggests that as we perform 464 actions we have an implicit feeling of agency that is sensitive to the fluency experienced in these actions. 465 Our brain is able to detect these changes through the implementation of monitoring processes involving fast 466 rhythms in the gamma band range. 467 468

Neural oscillations in automatic imitation 469
Our secondary finding is related to the modulation of neural oscillations by congruency in automatic 470 imitation. Importantly, this study is the first that investigates oscillatory pattern in this task. During this task 471 a participant responded to a visual cue while observing a finger movement produced by a mirrored hand in 472 the background and where the participant's response could be congruent or incongruent with the observed 473 finger movement (or "neutral" if no movement was observed). We identified strong movement-locked 474 responses in the alpha, beta and theta bands. Alpha oscillations at central brain areas have been typically 475 associated with perception of biological movement (Perry & Bentin 2009) and beta oscillations have been 476 related to motor preparation (Pfurtscheller et al., 1997), amongst other functions. In our task, alpha and beta 477 oscillations show increased desynchronization for the congruent relative to the incongruent condition, 478 possibly reflecting increased visuo-motor integration and processing for movements that match (relative to 479 movements that do not match) the observed actions. This effect could reflect the engagement of a mirror imitation. Future studies should aim at better characterize interactions between the distinct rhythmic 491 activities that occur in automatic imitation, for instance between gamma and theta oscillations (Jensen and 492 Colgin, 2007). 493 494 We must acknowledge some limitations of our study. First, unlike previous work on agency we focused on 495 evaluating implicit agency using a time estimation task. The relationship between the implicit feeling and 496 the judgement of agency is still not fully clarified (Synofzik et al., 2008) but see (Ebert and Wegner, 2010) 497 so we cannot make strong conclusions regarding the general mechanisms of agency. A fruitful avenue of 498 research when investigating the neural correlates of agency will be to include these two types of agency. 499 Second, since we focused on neural responses time-locked to the motor responses it could argued that 500 differences of neural activities between conditions are merely related to differences in motor parameters. 501 We think this is unlikely given that all conditions evoked a similar uplift finger movement and the difference 502 only concern the congruency between observed and executed actions. Future studies should include 503 additional measures of movements such as electromyography to better monitor these effects. Lastly, since 504 there is evidence that agency is impaired in psychopathologies such as schizophrenia (Frith, 2005