Pre-movement sensorimotor oscillations shape the sense of agency by gating cortical connectivity

doi:10.21203/rs.3.rs-3220811/v1

Download PDF

Article

Pre-movement sensorimotor oscillations shape the sense of agency by gating cortical connectivity

https://doi.org/10.21203/rs.3.rs-3220811/v1

This work is licensed under a CC BY 4.0 License

Version 1

posted

You are reading this latest preprint version

Our sense of agency, the subjective experience of controlling our actions, is a crucial component of self-awareness and motor control. It is thought to originate from the comparison between intentions and actions across broad cortical network. However, the underlying neural mechanisms are still unknown. We hypothesized that oscillations in the theta-alpha range, thought to orchestrate long-range neural connectivity, may mediate sensorimotor comparisons. To test this, we manipulated the relation between intentions and actions in a tetraplegic user of a brain machine interface (BMI), decoding primary motor cortex (M1) activity to restore hand functionality. We found that the pre-movement phase of low-alpha oscillations in M1 predicted the participant’s agency judgements. Using EEG-BMI in healthy participants, we linked M1 and supplementary motor area (SMA) pre-movement alpha oscillations to agency ratings, and to changes in their functional connectivity with sensory areas. These findings argue for phase-driven gating as a key mechanism for sensorimotor integration and agency.

Biological sciences/Neuroscience/Sensorimotor processing

Biological sciences/Neuroscience/Sensory processing

The sense of agency refers to the subjective feeling of causing and controlling our actions ¹, which enables us to perceive ourselves as autonomous, self-governing agents. This feeling is fundamental to self-awareness ^1–3. Disturbances in the sense of agency are associated with psychiatric disorders, such as schizophrenia and autism ^4,5. Theoretical work and related experimental findings suggest that the sense of agency arises from the comparison between predicted and observed sensory outcomes of intended actions ^6,7. Due to its link with basic sensorimotor mechanisms, it is likely that the sense of agency does not merely accompany volitional motor control, but also directly contributes to it ^8,9. This makes it particularly important not only for natural closed-loop control, but also for artificial control loops such as in prosthetics and brain machine interfaces ^10,11.

To perform the sensorimotor comparisons which underlie the sense of agency, the brain must integrate pre- and post-movement signals at a large scale, orchestrating the information flow among functionally specialized but widely distributed brain regions. Consistent with this notion, a large fronto-parietal network, including the premotor cortex, the supplementary motor area (SMA), and the angular gyrus, has been implicated in the sense of agency ^12–17. However, the underlying brain-wide neural dynamics remain poorly understood. Most previous studies focused on post-movement processing related to agency (e.g., ¹⁸) while a separate line of research investigated pre-movement intentions (e.g., ^19,20). However, we still lack specific knowledge of the mechanisms leading to the integration of pre- and post-movement signals. Indeed, many previous neuroimaging studies lacked the temporal precision needed to resolve the fast dynamics of such process ^15,21. Other studies ^12,14, mainly utilizing invasive recordings, could gather temporally resolved information but lacked the spatial coverage to link it to large scale neural processes.

Mounting evidence indicates that theta (4–8 Hz) and alpha band (8–13 Hz) oscillations play a crucial role in coordinating long-range information exchange ^22,23. Their pre-stimulus phase and/or power have been shown to modulate subsequent perception ^24,25, neural connectivity ²⁶ and information integration across sensory modalities ²⁷. Moreover, pre-stimulus alpha oscillations are involved in perceptual anticipation ^28,29. Here, we tested the hypothesis that pre-movement neural oscillations may influence the sense of agency by conveying top-down predictions, which are necessary for the underlying sensorimotor comparisons.

We thus studied the interplay between sense of agency for hand movements and theta-alpha oscillations preceding them. In a set of three experiments, we used brain machine interfaces (BMIs) as tools to manipulate the relationship between read-out motor commands and sensory feedback. For Experiment 1 and 2, we enlisted the participation of a tetraplegic individual who is a proficient user of an intracranial BMI. This BMI system translates motor commands decoded from the M1 into functional hand movements through a neuromuscular electrical stimulation system (NMES, see Fig. 1). In Experiment 1, the participant was asked to explicitly rate his sense of agency for cued hand movements ¹¹. In Experiment 2, we employed an implicit measure of the sense of agency based on the subjective perception of the timing of self-initiated movements ³⁰. In Experiment 3, to investigate the crosstalk between motor areas and other brain regions, we implemented a similar design as in Experiment 1 in healthy participants using a scalp EEG-based BMI.

in Experiment 1, we found that the phase of pre-movement low alpha oscillations predicted explicit agency judgements. In Experiment 2, we showed that these oscillations modulated the temporal binding between intentions and actions, an implicit measure of the sense of agency ³⁰, and that the effect is specific for intentional movements. In Experiment 3, we extended our findings to whole brain recordings in heathy participants and found the strongest correlation between pre-movement oscillatory phase and sense of agency in the left, contralateral SMA. Additionally, movements starting close to the high agency oscillatory phase were followed by an increase in connectivity between the contralateral SMA and contralateral parietal, temporal and frontal regions. These results support phase coupling as a key mechanism for sensorimotor integration, linking the sense of agency with general theories about the rhythmic gating of brain-wide communication.

Experiment 1 – The phase of pre-movement low alpha oscillations in M1 predicts explicit agency judgements

Behavioural and neurophysiological results from Experiment 1 are extensively described in a previous report ¹¹. Here we focused on a subset of conditions to highlight the role of neural oscillations. The participant was instructed to plan and execute one of four possible hand movements (hand closing/opening, thumb flexion/extension) using the BMI prosthesis. Somatosensory feedback was manipulated by producing either the decoded hand movement (congruent feedback, S+) or the opposite hand movement (incongruent feedback, S-; e.g., flexion instead of extension, and vice versa) through NMES. By the same logic, visual feedback was concurrently manipulated by displaying either the decoded hand movement (V+) or the opposite hand movement (V-) by a virtual hand, superposed to the participant’s (hidden) real hand (Fig. 2a). All possible combinations of congruent and incongruent visual and somatosensory feedback (i.e.: V+/S+, V-/S-, V+/S-, V-/S+) were presented in a randomized order. Following each trial, the participant was asked to provide an agency judgement for the executed movement (Q1: “Was it you who generated the movement? Yes - No”).

Congruent (V+/S+) or incongruent (V-/S-) trials consistently elicited positive or negative agency judgements, respectively, yielding little variability besides that induced by feedback congruency (see Fig. 2b and ¹¹). To minimise the exogenous contribution of sensory afference and highlight the role of endogenous neural oscillations, we thus focused on conflicting feedback conditions (V+/S- and V-/S+), exhibiting a much weaker correlation between agency judgements and sensory feedback (McFadden’s R² in a logistic regression Q1 ~ feedback = 0.02).

We hypothesized that pre-movement theta and alpha band oscillations may modulate the sense of agency, due to their role in perceptual anticipation and functional connectivity for multisensory and sensorimotor integration ^32–34. We thus analysed the relationship between agency judgements and the phase and power of oscillations in the 4–13 Hz range in M1 LFP. The signal from all electrodes was averaged, reducing the number of comparisons almost a hundred-fold with negligible information loss at the low frequencies considered here (see Fig. S1). To assess the effect of oscillatory phase on agency judgements, we contrasted the instantaneous phase between trials with positive and negative (Q1 = Yes / No) agency judgements by using the phase opposition product ³⁵. This measure indexes the amount of clustering of phase angles for high and low agency trials around opposite phases. We found a significant (p = 0.0004) cluster of phase opposition spanning the 6–9 Hz range and peaking at about 8 Hz. The cluster lasted from 500 to 50 ms before movement onset (Fig. 2c-d; see Fig. S2a for confirmation of the pre-movement timing of the effect from a causal filter analysis). Phase angles at the maximal phase opposition time-frequency point (8 Hz, -256 ms) were clustered between π and π/2 for high agency trials, and between 0 and 3π/2 for low agency trials (Fig. 2e). The relationship between the 8 Hz phase and agency was similar in V+/S- and V-/S + trials, with positive agency judgements becoming increasingly frequent as phase angles approached the optimal phase in both conditions (Fig. 2f). In contrast, power in the 4–13 Hz range (up to 40 Hz) showed no significant difference between high and low agency trials (Fig. S3a). Phase opposition analyses at higher frequencies (up to 40 Hz) also yielded no significant results (see Fig. S4a).

The observed effect peaked at 8 Hz, at the boundary between the conventional theta (4–8 Hz) and alpha (8–13 Hz) frequency bands. However, in our implanted participant, the peak of the power spectrum, typically observed in the alpha band ³⁶, was rather low in frequency (6.2 Hz, see Fig. S5a). Additionally, movement-related desynchronisation was also observed in a similarly low frequency range. This suggests that the observed phase opposition occurred within the participant’s individual range of the sensorimotor mu rhythm, which is typically associated with the alpha frequency band. This deviation from the typical frequency range is not surprising as the mu rhythm is not stable across development ³⁷ and has been observed to be lower in frequency in patients with chronic paralysis ³⁸. Therefore, we refer to the frequency range of this effect as "low-alpha", and expect to observe it within the alpha range in healthy participants.

Experiment 2 – The phase of pre-movement low alpha oscillations predicts an implicit index of sense of agency

The previous analyses establish a relationship between the phase of pre-movement oscillations around 8 Hz and the sense of agency, as measured through an explicit judgement. We next aimed to confirm this relationship via a novel temporal binding paradigm (see ³⁰) designed to provide an implicit measure of the sense of agency ³⁹. As in the classic Libet paradigm ¹⁸, a rotating clock was displayed on a screen, and the participant was asked to report the position of the clock at the onset of a hand movement triggered by the NMES system (Fig. 3a). To validate this protocol as an implicit measure of agency, the experiment consisted of two sessions designed to induce a high and low sense of agency, respectively. In the former, the movement was triggered by the participant’s intention to move as decoded by the BMI system (voluntary condition). In the latter, the movement was randomly generated via the NMES system without motor intention (involuntary condition).

Previous studies have demonstrated a phenomenon called intentional binding, where the effects of intentional actions, such as a sound triggered by pressing a button, are perceived as occurring earlier than they actually do ³⁹. This anticipatory effect has been described for voluntary actions, and is considered to be a proxy of the sense of agency. In our unique BMI setup, we expected a similar binding effect between the internal motor intention and the delayed execution of the decoded movement, resulting in voluntary BMI-generated actions being perceived as occurring earlier than involuntary ones (see ³⁰ for an in-depth discussion). In line with this prediction, the participant perceived voluntary BMI-generated movements as occurring earlier than involuntary ones, relatively to their actual timing (median voluntary = -497.8 ± 299 ms interquartile range, median involuntary = -384 ± 185 ms, Wilcoxon p = 0.033, Fig. 3b). We thus utilized the amount of anticipation as an implicit index to stratify trials into high (movement perceived earlier), and low (movement perceived earlier later) agency, based on a median split.

To investigate the relationship between pre-movement phase and our implicit agency measure, we computed the phase opposition product between high and low implicit agency trials. We found a significant cluster of phase opposition ranging from 6 to 10 Hz (p = 0.0017), with the peak occurring at 8 Hz and at -342 ms relative to movement onset (Fig. 3c-d; see Fig. S2b for confirmation via causal filter). As depicted in Fig. 3e, the phase at 8 Hz/-256 ms of trials with high (low) implicit agency was qualitatively similar to trials with high (low) explicit agency in Experiment 1 (note that, to allow comparison with Experiment 1, we chose the same time-frequency point as in Fig. 2e and not the phase opposition peak). Analyses at higher frequencies (15–40 Hz) revealed another significant phase opposition cluster at around 30 Hz (Fig. S4b), which was not found in Experiment 1 (Fig. S4a). As for Experiment 1, no significant difference in power was found (Fig. S3b).

To rule out that our low alpha phase effect was merely a result of attentional or perceptual processes related to the timing judgment required by the Libet-like task, we ran the same analysis on the involuntary control condition. Here, the perceptual task is identical, but no voluntary action is required and thus no agency is expected. We found no association between pre-movement oscillations and perceived action timing in the involuntary control condition (Fig. S6).

As shown in Fig. 4a, averaged LFPs, filtered at 8 Hz, shared a qualitatively similar phase when comparing high (or low) agency judgements across Experiment 1 (explicit) and Experiment 2 (implicit). This was statistically confirmed by bootstrapping (Fig. 4b-c), and by computing the phase opposition between all trials with high agency (“Yes” answers in Experiment 1 and “early” perceived movements in Experiment 2) and low agency (“No” answers in Experiment 1 and “late” perceived movements in Experiment 2) (Fig. 4d). Therefore, the same pre-movement 8 Hz oscillatory phase predicted the participant’s sense of agency, as indexed both via an explicit measure based on subjective judgements and an implicit measure based on the temporal perception of voluntary movements (Fig. 4e).

Low alpha LFP oscillations capture modulations of M1 firing

Rhythms recorded with LFPs capture a multitude of neural phenomena, which may not be straightforward to interpret ⁴⁰. To better characterize the neural bases of the 8 Hz rhythm reflected in the LFP and modulating the sense of agency, we quantified the relationship between the phase of this rhythm and M1 spiking activity. We focused on data from Experiment 1 due to the larger number of trials per experimental session. We quantified the strength of the relation between level of firing and LFP oscillatory phase by computing the LFP-spike phase locking value (PLV, e.g., as in ⁴¹) for different frequencies in the 4–13 Hz range. We found that the PLV peaked at the LFP frequency of 8 Hz (Fig. 5a), indicating that firing activity was most strongly modulated by the phase of 8 Hz oscillations. Firing was about 6% higher when in the most favourable oscillatory phase of 8 Hz LFP oscillations than when in the least favourable (Fig. 5b). Moreover, the preferred LFP phase angles of individual units exhibited a clear clustering around 4π/3 (Fig. 5c). These results suggest that the 8 Hz LFP oscillations recorded in our experiments capture periodic fluctuations in M1 firing activity. To rule out that the overall level of firing per se, rather than its periodic component, influenced the sense of agency, we assessed whether the pre-movement global M1 firing rate was associated to agency judgements. As shown in Fig. 5d, this analysis revealed no significant difference in firing rates between high and low agency trials. These results suggest that the sense of agency is modulated by the specific 8 Hz periodic component of firing fluctuations, rather than the average firing rate itself.

Experiment 3 – pre-movement SMA and M1 alpha oscillations predict agency ratings in healthy participants using an EEG-BMI

Our results so far establish a relationship between pre-movement 8 Hz oscillations in M1 - the only recording site in our implanted participant - and his sense of agency. To investigate the potential contribution of areas beyond M1, in Experiment 3, we devised a conceptually similar, EEG-based version of Experiment 1. Thirty healthy participants were trained to use an EEG-BMI based on kinaesthetic motor imagery to trigger the movement (hand closing) of an anatomically congruent virtual hand on a screen. After each movement, they rated their sense of agency for the movement on scale from 1 to 9 (Fig. 6a-b). After verifying the validity of our setup through preliminary behavioural analyses (see Methods and Fig. S7), we examined whether the phase of pre-movement sensorimotor oscillations was associated with agency ratings. To localise the source of neural oscillations, we projected EEG activity to 114 cortical regions of interest (ROIs) through eLORETA⁴². For each ROI, we then contrasted the highest and lowest 33 %of agency ratings via the phase opposition product in the alpha (8–13 Hz) range in the 0.5 seconds preceding the movement. The alpha range was chosen to match the overall higher spectral peak in healthy participants compared to our implanted participant (see Experiment 1 and Fig. S5).

The cortical map of P-values for the alpha range phase opposition product based on agency ratings is shown in Fig. 6c. At the whole brain level, the most significant cortical region was localized in the posterior part of the left supplementary motor area, contralateral to the BMI movement (SMA, Fig. 6c-d). The significance of the effect in this region resisted false discovery rate (FDR) correction across all 114 ROIs (whole brain corrected p = 0.023, uncorrected p = 0.0002). This result was robust to the specific choice of the frequency range (Fig. S8a-b) and confirmed by a causal filter (Fig. S2c). When focusing the statistical analysis to a subset of 12 ROIs corresponding to motor and premotor areas (see Fig. S9), a second significant area (corrected p within motor-premotor areas = 0.019, uncorrected p = 0.0033, Fig. 6c-e) was found in the dorsal part of the left primary motor cortex. The SMA effect was relatively spread across the whole alpha band (Fig. 6d), possibly due to individual differences in the alpha peak, and peaked at 9 Hz, close to what observed in our implanted participant.

As the pre-stimulus alpha phase is known to influence visual perception ²⁵, we explored whether the observed effects on agency could be partially explained by perceptual effects by running an additional control analysis on the set of 26 ROIs corresponding to visual regions (see Fig. S9). No visual region survived FDR correction for multiple comparisons. Additionally, no significant difference in power (Fig. S3c-d) or phase at higher frequencies was found (Fig. S4c-d).

The optimal phase for agency is associated with increased alpha-band functional connectivity

Results from our three experiments showed that the pre-movement oscillatory state of motor areas is associated with the subjective sense of agency for a subsequent movement. Since agency is reported (and most likely experienced) post-movement, we searched for a trace of the pre-movement oscillatory phase in post-movement signals, possibly affecting the sense of agency. In line with theories about brain rhythms and communication ²³, previous studies ²⁶ have highlighted correlations between pre-stimulus phase in a given brain area and post-stimulus connectivity originating from that area. Thus, we searched for an association between pre-movement SMA oscillatory phase (the region showing the strongest phase effect) and post (and during) movement functional connectivity between SMA and the rest of the brain, as a putative source of modulation of subjective agency.

For each participant, we extracted the left, contralateral SMA phase at the time-frequency point in which the modulation of the sense of agency was strongest. We then selected trials in which the pre-movement oscillatory phase was close to the optimal phase for agency, and trials in which it was far from it (see Methods for details). We contrasted between these subsets of trials functional connectivity in the 4–45 Hz range, measured through the debiased weighted phase-lag index, WPLI ⁴³. Specifically, we studied how the pre-movement SMA oscillatory phase modulated the post-movement (0.2–1.2 s; see Methods) connectivity between SMA and the rest of the brain. When computing a global average of functional connections between SMA and all other cortical regions, we found that the optimal oscillatory phase was associated with an increase in connectivity in the 9–12 Hz range (corrected p = 0.005, Fig. 7a-b). To localise the source of this effect while reducing the degrees of freedom of our analysis, we repeated it after grouping brain regions by lobe and hemisphere, obtaining eight macro-regions (see Methods). Left (contralateral to the movement) frontal, temporal, and parietal areas survived Bonferroni correction (uncorrected p = 0.002, 0.0023, and 0.0063, respectively), confirming that the optimal oscillatory phase was associated with a widespread increase in connectivity from the left SMA (Fig. 7c). The same analysis using the left M1 as a seed revealed a similar pattern of connectivity changes (Fig. S10). Control analyses showed that the increase of connectivity was not simply due to an increase of power in SMA (Fig. S11) and could be reproduced with different methods to extract optimal phase angles (Fig. S12-S13).

Finally, we tested whether the observed connectivity changes were associated with a change in the directionality of functional connectivity. To do this, we studied the 9–12 Hz phase coherence of left SMA signals with time-shifted signals obtained from each of the three significant target regions (using the methodology set in ^44,45). The time shift at which the coherence is stronger indicates the direction in which functional connectivity is stronger. Stronger coherence for positive time shifts would support stronger correlation of left SMA oscillatory activity with future oscillatory activity of the target region (i.e., from SMA). Conversely, stronger correlation for negative time shift would support stronger correlation with past oscillatory activity of the target region (i.e., from target region). We found that values of phase coherence were higher at positive time shifts in trials with optimal phase, whereas coherence was stronger at negative time shifts in trials with non-optimal phase, (Fig. 7d). Time shifts of peak coherence were significantly different between optimal and non-optimal phase trials (Fig. 7e), regardless of the range of time lags used in the analysis (Fig. S14). This suggests that functional connectivity in trials with the optimal phase was enhanced more in the direction from SMA to temporal areas, compared to trials with non-optimal phase. To better localise this effect, we performed the same directionality analysis at a finer spatial resolution within the temporal lobe, at the scale of the original 114 ROIs used for source reconstruction. We found the region with the most significant (p = 0.0022, one tailed) directionality shift, again from SMA to the target area, to be localised in the posterior part of the temporal lobe (Fig. 7f).

We investigated the relationship between neural oscillations and the sense of agency for hand movements through invasive (Experiments 1 and 2, one tetraplegic participant) and non-invasive BMIs (Experiment 3, thirty healthy participants). BMIs offer an exceptional setting for studying the sense of agency, as they allow for the perturbation of the coherence between intentions and actions. Thus, BMIs permit to introduce nuances in the sense of agency for self-generated movements, which is typically very high and hard to manipulate under normal circumstances. In Experiment 1, we decoded motor intentions from an expert user of an implanted BMI and manipulated the visual and somatosensory congruency between intended actions and sensory feedback via virtual reality and NMES. The participant’s agency judgements were predicted by the phase of low alpha (8 Hz) LFP oscillations, ~ 250 ms prior to the movement (Fig. 2c-e).

Although most studies concur that both pre- and post-movement signals contribute to the sense of agency ^46,47, the exact processes underlying the integration of sensorimotor predictions and sensory feedback are not yet understood. Seminal studies showed that intentional movements are preceded by a slow negative deflection in scalp potentials above motor areas ⁴⁸, and that such potential precedes the reported timing of the conscious intention to move ¹⁹ (also see ²⁰ for a study focusing on stochastic fluctuations in these potentials). During this period, as motor intention builds up, sensory predictions are likely formulated. However, these studies focused solely on the efferent aspect of the “intentional chain” ⁴⁹, from intention to action without exploring its relationship with sensory reafference. Here, we showed that endogenous oscillatory activity, taking place before sensory feedback about the movement becomes available, modulates the subsequent sense of agency for the movement. This suggests that alpha band oscillations contribute to predictive components of the sense of agency.

In Experiment 2, we corroborated this finding using an implicit measure of agency based on the perceived timing of BMI movements in a Libet-like paradigm. The 8 Hz phase preceding movement onset predicted the anticipation of perceived movement timing (binding) that characterized voluntary BMI actions (Fig. 3c-e), here used as an implicit proxy of agency. Importantly, the optimal phase for explicitly (Experiment 1) and implicitly (Experiment 2) assessed sense of agency was the same (Fig. 4). This confirms and reinforces our finding that the pre-movement low alpha phase modulates the subsequent sense of agency. Furthermore, the implicit nature of our measure suggests that the effect relates to the genuine pre-reflexive experience of agency, rather than to metacognitive aspects of explicit judgements. Importantly, the pre-movement phase biased the implicit agency measure for voluntary movements, but not for involuntary ones. This implies that the activation of the intentional chain is necessary for sensorimotor oscillations to modulate the temporal binding between intention and action. Incidentally, this also rules out any explanation in terms of attentional or perceptual effects, as these would have equally affected time perception for involuntary movements.

When investigating the link between LFP oscillations and spiking activity, we found an 8 Hz peak in coupling between LFP and M1 spiking activity (Fig. 5a), in line with studies describing alpha band LFP-spike coupling in the non-human primate sensorimotor system ⁵⁰. This result offers an intriguing interpretation for the pre-movement phase effect. Studies in humans show that motor evoked potentials induced by a TMS pulse are modulated by the phase of sensorimotor oscillations ⁵¹. It is possible that also bursts of M1 activity that trigger the onset of spontaneous movements are more likely to occur during the most excitable phase of M1 LFPs. Then, BMI-generated movements occurring during the excitatory LFP phase may lead to higher agency because natural self-generated movements are more likely to happen at that time. This may underlie the association between pre-movement phase and sense of agency, which we consistently observed in Experiments 1, 2 and 3.

In Experiment 3, we extended our investigation to healthy participants, using EEG to study the interplay between pre-movement phase, sense of agency and whole brain neural dynamics. Once again, the phase of alpha-band pre-movement oscillations predicted agency ratings. We traced the most relevant correlation between pre-movement phase and agency ratings to the left (contralateral to the movement) SMA and M1 (Fig. 6c), extending of our finding to healthy participants and to a brain region, SMA, not monitored in our implanted participant. Several studies highlighted SMA as a key region for agency ⁵², and as one of the nodes showing the earliest activations in the generation of intentional movements ¹⁴. Such early pre-movement activations suggest that SMA may contribute especially to the predictive component of the sense of agency, an idea supported by neuroimaging ⁵³ and TMS ⁵⁴ studies. Here, we add to these findings by linking them to a specific alpha-band oscillatory mechanism.

We ran functional connectivity analyses to study the interplay between pre-movement oscillations in motor areas (i.e., the source of the phase-related agency modulation) and whole brain dynamics. These analyses revealed that BMI movements starting in the optimal phase for agency were associated with a higher alpha-band connectivity from SMA to the rest of the brain after movement onset (Fig. 7a-b). The observed changes in functional connectivity were especially pronounced between SMA and temporal-parietal regions, possibly encoding high level features of visual feedback (Fig. 7c). A similar pattern was observed using M1 as a seed. Additionally, the optimal phase was also associated with a change in directionality of functional connectivity between SMA and the posterior part of the temporal lobe (Fig. 7f), which has been implicated in the visual processing of hand movements ^55,56. It is tempting to speculate that the pre-movement SMA phase may gate the information exchange involved in sensorimotor comparisons between motor intentions (encoded in SMA) and visual information about their outcome (encoded in the temporal lobe). This would in turn modulate the amount of binding between intentions and actions and the associated measures of agency. Previous studies have proposed the parietal cortex^57,58 to be acting as a “sensorimotor comparator”, while other works pointed at the cerebellum⁵⁹ and the premotor cortex ⁶⁰. Our results possibly encompass these findings by suggesting a distributed architecture, wherein sensorimotor comparisons are performed through alpha-band communication between SMA and temporal-parietal regions.

Previous research has implicated alpha-band communication in the top-down, anticipatory modulation of sensory areas ^61–63. This is believed to contribute to sensory processing, e.g. by enhancing relevant stimuli ⁶² or optimizing detection performance ⁶⁴. Transposing this idea to the intentional chain, we propose that during the preparatory phase preceding a movement, premotor and motor areas like SMA and M1 send predictive signals through alpha-band connectivity to temporal and parietal sensory areas. These predictive signals likely carry information about the expected sensory consequences of the impending action, contributing not only to anticipate sensorimotor comparisons, but also to pre-select relevant sensory features for closed loop motor control (see for example ⁶⁵). The sense of agency could be then seen as the subjective correlate of selective communication between motor areas and sensory areas encoding features that are self-generated, and should thus be integrated with efferent commands to improve motor performance. Under this perspective, the functional relevance of the neural process leading to the sense of agency becomes apparent. Using invasive BMIs, we highlighted the key role of alpha-band sensorimotor oscillations in this process, and generalized this finding to more broadly implementable non-invasive BMIs or other forms of human-computer interactions. Our findings may gain further relevance in future prosthetics and brain machine interfaces, as the rapid advancement of decoding and actuation technologies may elevate cognitive aspects of motor control to the ultimate bottleneck to efficient and intuitive use.

By detecting sensorimotor contingencies, humans can differentiate between self and externally generated events, and the sense of agency is the experiential counterpart of such process. It has been argued that such mechanism underlies personal responsibility ⁶⁶ and even self-awareness, as the self would emerge as the agent of internally generated mental states ^1,67. Here, we provide novel evidence that alpha oscillations are related to the sense of agency, bolstering functional theories on the role of the alpha rhythm for brain connectivity, binding, and anticipation ^23,26,34,61. These findings pave the way to further mechanistic models of key cognitive processes, by applying oscillatory theories of brain communication ²³ to the causal binding between internal states and their consequences.

Experiment 1 and 2 - participant

The participant was a 27-year-old (at the time of recordings) male with quadriplegia at the C5/C6 level originating from a cervical spinal cord injury (SCI) dating to 8 years prior to data collection. He had a full range of motion in both shoulders and elbow flection and could perform twitches of wrist extension (1/5 and 2/5 strength on left and right wrists respectively). He had no motor function below C6. His proprioception was intact in the right upper limb/shoulder for internal through external rotation, forearm pronation through supination, and wrist flexion through extension. Proprioception at the level of metacarpal-phalangeal joints for all right-hand digits was impaired. He was enrolled in a pilot clinical trial (NCT01997125, Date: November 22, 2013) of a custom BMI system (Battelle Memorial Institute) to restore motor functionality of the upper limb following SCI. The BMI system required the implantation of a Utah microelectrode array (96 channels, 4.4 x 4.2 mm, 1.5 mm depth) in the hand region of the left primary motor cortex. Reference wires were placed subdurally. The target region was identified via pre-operative functional Magnetic Resonance Imaging as the patient was asked to attempt performing right hand movements. See the first description of the BMI system ³¹ for further details about the participant and surgical process.

Experiment 1 and 2 - BMI system

Neural data from the Utah array was sampled at 30kHz and band-pass filtered between 0.3Hz and 7.5kHz at the hardware level (3rd order Butterworth). The data were digitized in 100ms bins and analysed through custom MATLAB code. Before decoding, artefacts due to NMES were removed by blanking the signal over 3.5ms around the artefact, defined as a signal amplitude exceeding 500µV in at least 4 out of 12 randomly selected channels. Neural decoding was based on a non-linear Support Vector Machine ⁶⁸ (SVM). The SVM used 96 input features consisting of the mean wavelet power (MWP) for each channel and 100 ms bin. To obtain the MWPs, neural activity was decomposed into 11 wavelet scales (Daubechies wavelet, MATLAB), and the coefficients of wavelets 3–6, corresponding to the multi-unit frequency band spanning from 235 to 3.75kHz, were averaged for each channel. The decoder was re-trained before each experimental session, by asking the participant to attempt performing one of the four hand movements (HO, HC, TE, TF) in order to generate the training data. The decoding system achieved more than 90 %accuracy for all the four movements (see ³¹). The subject performed 7 blocks consisting of 3 repetitions per movement type each. The decoder output consisted of four numbers in the − 1/1 range, indicating the relative probability for each movement. A threshold of 0 was set for the selection of an intended movement, with the movement with the highest score prevailing if two or more classes exceeded the threshold. A custom-built Neuromuscular Electrical Stimulation (NMES) system was used to translate the decoded intentions into actual hand movements, by stimulating forearm muscles. The NMES system consisted of a circumferential forearm sleeve with 130 copper-coated electrodes, 12mm in diameter. The electrodes were disposed in an array, spaced at regular intervals (22mm longitudinally x 15mm transversely). Stimulation was delivered through rectangular pulses of 50Hz monophasic current (pulse width 500µs, amplitude 0-20mA). The stimulation patterns and intensity were re-calibrated at the beginning of each session in order to optimize the match with the participant’s intentions. Further details about the neural decoder and NMES system can be found in ³¹.

Experiment 1 – protocol

In Experiment 1, we manipulated the congruency between the participant’s motor intentions and sensory feedback and assessed how this affected his sense of agency. Each trial started with a verbal cue about the hand movement to be performed (“hand open”, “hand close”, “thumb extension”, “thumb flexion”), followed after a 2 second delay by a verbal “go” cue. The participant was instructed to start attempting the cued movement only at the “go”, without anticipating. During the following 4 seconds, the participant received visual and somatosensory feedback according to the decoded movement and the feedback congruency for that trial and sensory modality. Somatosensory feedback was delivered by eliciting the target movement through the NMES sleeve. Visual feedback was constituted by an animation of a virtual hand performing the target movement, displayed on a screen placed horizontally to cover the participant’s right hand. The hand model and the animation corresponded to the ones routinely used by the participant during BMI training sessions, and its size and position were adjusted to match the participant’s real hand. In trials with congruent somatosensory (and/or visual) feedback, the decoded movement was executed through NMES (or displayed in a virtual animation). In incongruent trials, the opposite movement was executed and/or displayed, replacing hand opening with hand closing, thumb extension with thumb flexion, and vice versa. Sensory feedback was only delivered when one of the output classes of the neural decoder reached the threshold of 0. In the 5–6 seconds after the sensory feedback phase, the participant answered two questions, Q1 and Q2, about his feeling of agency for the movement. The whole experiment consisted of five experimental sessions performed over different days, each consisting of four blocks of BMI training and four blocks of experiment, each lasting around 15 minutes. Each experimental block consisted of 32 trials, where each combination of V/S feedback and cued movement was repeated twice. Therefore, the grand total of trials was 640, 160 for each feedback condition.

Experiment 2 – protocol

The second experiment was part of a broader study ³⁰ aiming at investigating the effects of manipulating the intentionality chain composed of intentions, motor acts, and their consequences on the external world, based on a modification of Libet’s intentionality experiment ¹⁹. Here, we focused on two experimental sessions within that study to establish an implicit measure of sense of agency. In the first session (high agency), the participant was cued to perform one of two possible movements through the BMI system, HO and HC. While performing the movements, the participant observed a single hand clock on a computer screen, with numbers from 5 to 60, completing a full rotation in 2.56 seconds. Again, movements were triggered by the activation of the neural decoder, but the NMES was always activated congruently. Additionally, 300 ms after HC was executed, a 1000 Hz “beep” was produced, lasting 100 ms. No additional consequence followed HO execution. Since here we focused on pre-movement signals, we pooled trials from the two movements (see Fig. S15 for further analyses). The participant was instructed to pay attention to the location of the clock hand at the time of movement onset, and to report it at the end of the trial allowing us to measure the perceived timing of the action. Differently from Experiment 1, the movements were self-paced, meaning the participant was instructed to freely initiate the movement and encouraged to vary his waiting time, which should in any case exceed one full clock rotation. In the second session (low agency), the only difference was that the same movements were executed involuntarily, by randomly activating the NMES for HO or HC while the participant was instructed to remain at rest. Each session consisted of 80 trials, 40 per movement.

Experiment 3 - participants

Thirty healthy participants (13 females, age = 26.6 years old, SD = 3.99) took part in this study. All participants were right-handed and had no psychiatric or neurologic history. All participants were naive to the purpose of the experiment and gave written informed consent to take part in the study. They were remunerated for their time with 20 Swiss francs per hour. The Ethical approval for the project was granted by the Vaud canton ethical committee.

Experiment 3 - EEG acquisition

EEG data was collected via 64-channel EEG eego mylab amplifier (AntNeuro, Hengelo, Netherlands) with a sampling rate of 512 Hz, referenced the CPz electrode. Impedences were kept below 20 KOhm.

Experiment 3 - protocol

The experiment lasted a total of about 3 hours and consisted in two parts: BMI training-testing, and agency assessment.

- Part 1 – BMI training

After setting the EEG, participants sat at a table in a light and sound-controlled environment with their hands placed below a computer screen laying horizontally on the table. They had to alternate periods of rest and kinesthetic motor imagery of their right hand to train the BMI algorithm in 3 blocks of 30 trials.

For motor imagery, they were asked to imagine the physical sensation of their muscles contracting when they imagined squeezing their right hand. They had to imagine a continuous and strong squeezing effort, almost as if they were on the verge of making a movement, but without contracting the arm, shoulder, or face muscles. For rest, they were instructed to relax as much as possible and especially avoid thinking about movement. Participants were asked to keep a constant strategy throughout the training session. During each trial, a grey fixation cross appeared on the screen, showing one of two different cues in randomized order: Rest or Motor Imagery (MI). Then, a red bar started filling the fixation cross to signal the participant for how long he had to maintain the rest or motor imagery (4 seconds for each trial). Once the recordings were completed, the BMI algorithm was trained on the participant’s data to produce an individualized motor imagery decoder, while the participant had a 5-minute break. If the decoder’s cross-validated accuracy on the training data was below 0.6, another block of 30 trials was collected and training was restarted. Before moving on to the second part, it was verified that the participant was able to control the BMI. This step was also important to convince the participant that the BMI could truly detect whether he/she is performing rest or motor imagery. The participant performed 20 trials with either Rest or MI appearing on a gray fixation cross. This time, a red square moved left or right on top of the fixation cross displaying to the participant the decoder’s output. 20 trial blocks were repeated while adjusting the motor imagery threshold, until it would take approximately 3 to 6 seconds to reach it on successful trials. This had the aim to keep the task challenging but not impossible, while minimizing the number of unwanted activations (see also below).

- Part 2 – agency assessment

In the second and main part of the experiment, participants saw two virtual, real sized open hands overlapping with their real hands which were kept below the screen, participants performed motor imagery and provided agency ratings when the right hand closed. We chose to use sham BMI as the key condition, as it allows to reliably produce the illusion that the participant is controlling the virtual hand, while being less subject to fatigue and decoder variability than real BMI. This allowed to collect a greater and constant number of trials regardless of each participant’s proficiency with BMI. The illusion was made possible by the fact that in EEG-BMI delays are long, and a prolonged and continuous effort lasting several seconds is required to reach the decoding threshold. Therefore, even when truly controlling the BMI, the participant cannot predict the exact timing of the movement, and habituates to experience control for delayed and temporally unpredictable movements. If the movement is provided at a randomized delay comparable to the intrinsic delay of the BMI system, a sham BMI trial is hardly distinguishable from a real BMI one (hence the careful adjustment of decoding thresholds in the previous phase). Sham BMI trials were cued and constituted 2/3 of the trials, the remaining 1/3 of trials were self-paced and used real BMI, having the purpose to ensure participants remained unaware that part of the trials used sham BMI.

In sham BMI trials, participants saw a GO cue appearing after a randomized delay of 1.5–2.5 seconds (uniform), at which point they had to start motor imagery. After a Gaussian randomized delay of 4.25 ± 0.5 (mean ± SD) seconds (limited between 3 and 5.5 seconds), the right hand closed with a continuous, pre-determined movement lasting about 0.4 seconds irrespective of the decoder’s output. 1.4 seconds from the onset of the movement, they were asked to provide an agency rating “How much did you feel like you generated the movement? (From 1 to 9)". They reported their answer verbally and the experimenter entered it into the experimental software through a keyboard.

In real BMI trials no GO cue was displayed, and participants were instructed to start motor imagery whenever they wanted between about 1 and 5 from the onset of the trial. The right hand closed when the decoder reached the threshold. 1.4 seconds after the movement, subjects provided an agency rating like in cued trials. If the threshold was not reached within 10 seconds, the trial ended, and no rating was asked. Participants performed five blocks of 60 trials, separated by about 3 minutes of break. In total, 200 cued and 100 self-paced trials were collected.

Before the experiment, subjects were instructed to focus on pre-reflexive aspects of the control experience, and not to use cognitive reasoning to provide the ratings. They were also asked to try to provide variable ratings using all the available range, and to keep the mean rating around 5. When told that they were truly in control of the virtual hand only in self-paced trials, at the end of the experiment, all subjects reported that they were not aware of it.

Experiment 3 - BMI decoder

The Python based Neurodecode framework was used for EEG-BMI (https://github.com/dbdq/neurodecode). To obtain the classifier, a random forest was trained to discriminate Rest and MI trials from the training blocks (see ⁶⁹). 322 input features were used, consisting of average power values during each trial’s 4 seconds Rest or MI period in 23 frequency bands, equally spaced from 8 to 30 Hz, measured in 14 sensorimotor channels (a 3 by 5 grid from FC3 to Cp4 excluding the reference, CPz). 8-folds cross-validation was used to test decoding accuracy, using 75% of trials as train set and 25% as test set.

Experiment 1 and 2 - data pre-processing

For LFPs, the main focus of the present study, data pre-processing consisted of four steps: trial selection, artefact removal, down sampling, and epoching. Trial selection had the main goal of discarding trials in which the participant failed to generate any movement, or to activate the correct decoder. Therefore, we only kept trials in which the participant managed to keep the cued decoder above threshold for at list 600 ms (6 classifier bins). In Experiment 1, we additionally required that such movement happens after the “go” cue, and at least 1.5 seconds before the “stop”, in order to ensure a sufficient time window for epoching. Additionally, in Experiment 2, 5 HC trials from the high agency session had to be removed due to technical issues with the recording. After trial parsing, we retained 422 out of 640 trials for Experiment 1 (66%), and 61 out of 80 (76%) trials from the high agency session in Experiment 2 (all 80 trials were retained in the low agency session as the participant did not need to activate the decoder in this session). The parsing was relatively even across conditions of interest, with 114/160 for V+/S+, 93/160 for V+/S-, 117/160 for V-/S+, and 98/160 for V-/S-. Similarly, in Experiment 2, we retained 26/35 (HC) trials for the movement eliciting the sound, and 35/40 for the movement not eliciting the sound (HO). Artefact removal was performed before epoching, as done online for BMI decoding, with the difference that we applied an 8.7 ms blanking window, in order to be more conservative on oscillatory analyses. Then, the data was down sampled to 1000 Hz, using a Kaiser anti-aliasing kernel. Spiking activity was extracted through the Wave_clus spike detection and sorting algorithm⁷⁰ with default settings. For the detection, a threshold was set at four times the standard deviation of baseline noise. Spikes were clustered through the superparamagnetic clustering algorithm, allowing to remove spurious signals. Since data collection was done in different sessions spanning several weeks, and the number of units in each channel fluctuated across sessions, we did not attempt to match units across recording sessions, and considered each unit in each session separately (e.g.: in Fig. 5c). For both LFP and multiunit activity, the data was epoched by time-locking to the onset of sensory feedback.

Experiment 3 - data pre-processing

EEG analyses were performed using FieldTrip and custom MATLAB code. Data was filtered in the 0.5–45 Hz range, a 50 Hz notch was applied, and epochs going from − 2 to 2 seconds from movement onset were created in FieldTrip. At this stage, data was re-referenced from the original CPz reference to average reference. Bad epochs presenting movement or muscular artifacts were manually rejected, and bad channels were selected and removed prior to independent component analysis (ICA). ICA was then run on the already epoched data with FieldTrip, with bad epochs and channel removed. Components corresponding to ocular movements were manually identified and removed before projecting the signal back to electrode space. Bad channels were re-inserted at this stage by interpolating neighbouring channels.

Experiment 3 - source reconstruction and templates

Source reconstruction was performed in FieldTrip using a standard head template. The inverse solution was computed on a grid of 3835 points regularly spaced by 7.5 mm through eLORETA. Then, source space data was projected to a set of regions of interest derived from the Lausanne atlas ⁷¹ (scale 2, 129 ROIs), from which subcortical regions were excluded obtaining a total of 114 ROIs. To obtain the signal of each ROI, singular value decomposition was performed on all the signals from solution points falling within the ROI. For connectivity analyses, the dimensionality of EEG data was further reduced by grouping together ROIs by hemisphere (left or right) and lobe (frontal, parietal, temporal, occipital). Connectivity values were first computed at ROI level, and then averaged within each of eight macro-regions corresponding to the left and right portion of each of the four cerebral lobes (see Fig. S16).

Experiment 1 - Multiunit analyses

Analyses on multiunit activity consisted of the evaluation of the coherence between spiking activity in M1 and LFP oscillations at different frequencies. For each frequency, the spike-LFP phase-locking value (PLV) was calculated bsy extracting the phase vectors for LFP oscillations at each spike location (through the procedure described above), and then applying formula (1) to the ensemble of phase vectors. For this analysis, all units from all channels were pooled. The analysis was re-applied separately for each unit to obtain Fig. 5c.

Experiment 2 - statistical analysis of behavioural data

Due to the strongly non-normal distribution of responses with frequent outliers (Kolmogorov-Smirnov test p < .0001 in both voluntary and involuntary condition, K-statistic = 0.80 and 0.99 respectively), we used the median as a robust indicator of the subject’s perceived movement timing and performed a Wilcoxon rank sum test to compare the two conditions.

Phase opposition analyses

Instantaneous values for power and oscillatory phase were obtained by convolving the signal with Morelet wavelets over 10 linearly spaced frequencies between 4 and 13 Hz, setting the number of cycles at 2π (4π for EEG signals to compensate for the lower signal to noise ratio). Our main analysis focuses on quantifying phase opposition in time and frequency between conditions of interest (high-low levels of explicitly or implicitly assessed agency). As a measure of phase opposition, we use the phase opposition product (POP ³⁵). To compute the POP, first we compute the inter-trial phase coherence (ITC) for all the trials pooled together, and for the two conditions separately.

$${\text{I}\text{T}\text{C}}_{\text{A}\text{L}\text{L}}=\left|{\sum }_{\text{i}=1:\text{n}}{{\omega }}_{\text{i}}/\left|{{\omega }}_{\text{i}}\right|\right|/\text{n}$$

$${\text{I}\text{T}\text{C}}_{\text{A}}=\left|{\sum }_{\text{c}\text{o}\text{n}\text{d}\text{i}\text{t}\text{i}\text{o}\text{n}\text{A}}{{\omega }}_{\text{i}}/\left|{{\omega }}_{\text{i}}\right|\right|/{\text{n}}_{\text{A}}$$

$${\text{I}\text{T}\text{C}}_{\text{A}\text{L}\text{L}}=\left|{\sum }_{\text{c}\text{o}\text{n}\text{d}\text{i}\text{t}\text{i}\text{o}\text{n}\text{B}}{{\omega }}_{\text{i}}/\left|{{\omega }}_{\text{i}}\right|\right|/{\text{n}}_{\text{B}}$$

Then, the POP is simply obtained as follows

$$\text{P}\text{O}\text{P}={\text{I}\text{T}\text{C}}_{\text{A}}{\text{I}\text{T}\text{C}}_{\text{B}}-{\text{I}\text{T}\text{C}}_{\text{A}\text{L}\text{L}}^{2}$$

The underlying idea is that, if trials within a condition are clustered around some angle, and trials in the other condition are clustered around an opposed angle, then the inter-trial coherence within conditions is going to be higher than when pooling the trials together. Therefore, higher values of POP indicate a stronger phase opposition between conditions. Instantaneous values of oscillatory power were simply computed as the squared absolute value of the wavelet convolution. Since in our analyses we focus on low frequencies, which are expected to be highly coherent on the small spatial scales of a Utah array, all analyses were performed on the mean LFP across all channels for Experiment 1 and 2. The high coherence of oscillations in the 4–13 Hz range was furthermore confirmed by analyses shown in the Supplementary Information (Fig. S1). The analysis was performed independently for each channel or ROI in Experiment 3.

Experiment 1 and 2 - statistical analysis of phase opposition

Statistical analyses of the time-frequency distribution of POP values were performed through cluster-based permutation tests to address the multiple comparison problem ⁷². In order to run the permutations, we started by defining a suitable statistics for POP values, as with a single subject it is not possible to simply run a T-test on POP values across subjects. To the best of our knowledge, the analytical form for the null distribution of POP values is not known, so we obtained P-values by comparing true POP values to POP values obtained by shuffling the labels of high and low agency trials over 10000 permutations. The P-value was then simply given by counting the number of permutations with a higher POP value. The same procedure was applied to all permutations, so that a time-frequency map of P-values was available for each permutation. P-values obtained this way were then transformed into T-values, and a threshold of 2 (corresponding to p ~ 0.05) was set to define the clusters. The total value of each cluster was then defined as the sum of the T-values of all time-frequency points composing it. Importantly, the exact nature of the statistics used at this stage to define cluster scores does not influence the test’s ability to appropriately control for type I errors, as this is addressed by the permutations performed subsequently ⁷². The final P-value for each cluster was defined as the probability of finding a cluster with a larger score over the 10000 permutations. The analysis was performed over a 1s window ending at the time of movement onset.

Experiment 3 – statistical analysis of phase opposition and behaviour

As for intracranial data, the aim of the analysis is to determine whether the phase of pre-movement neural oscillations can predict agency ratings. Preliminary behavioural analyses showed that sham BMI and real BMI trials elicited comparable agency levels (mean sham BMI 5.75 ± 0.13 SEM, mean real BMI 5.57 ± 0.21 SEM, p = 0.33, Fig. S7). The analysis was then performed only on sham BMI trials, reducing variability due to decoder accuracy (see protocol for the rationale).

Further analyses showed that the delay between GO cue and movement significantly influenced agency ratings, with 17 out of 30 subjects showing a significant negative correlation, and 1 subject showing a significant positive correlation (see Fig. S17). To remove the effect of such exogenous factor, we regressed out the effect of delay on agency ratings within each participant before splitting trials and computing POP values. Then, we contrasted the highest and lowest 33% of detrended ratings of each participant, in order to exclude uninformative intermediate values. The POP analysis was performed similarly to the intracranial data, using permutations of trial labels to obtain P-values for each frequency and timepoint. The main difference in the statistical analysis depended on the presence of a pool of multiple subjects instead of a single participant. P-values at the group level were obtained by permuting trial labels within each subject and computing POP values, and then averaging the POP values across subjects for real and permuted data. Again, P-values were computed by counting the number of occurrences of larger average POP values in permuted data. The procedure was performed in the 8–13 Hz, -0.5-0 s range for ROI level data.

Experiment 3 - Determination of optimal phase angles and trials

The analysis was performed on ROI level data, focusing on the left SMA, where the phase driven modulation of agency ratings is strongest. First, we tested whether the optimal phase angles at the group-level highest POP time-frequency point (9 Hz, -304 ms) were comparable for all participants. For each participant, we computed the optimal phase angle by taking the average phase angle of high agency trials and averaging it with the flipped average phase angle of low agency trials, at 9 Hz and − 304 ms in the left SMA. We tested whether the distribution of such optimal phase angles was significantly different from a circular (i.e., random) distribution through a Raleigh test, and the test was not significant at p = 0.13. We concluded each subject may in principle have a different optimal phase angle. This is in line with what observed in similar works ²⁷ and may be exacerbated by individual differences in the frequency of the strongest effect. We then decided to define optimal phase angles (and trials) individually. For each subject, we first identified the frequency at which the strongest phase modulation of agency ratings was observed within the 8–13 Hz range (but see Fig. S8 for a robustness check with respect to the choice of such frequency range). Then, we extrapolated and averaged phase angles at that frequency in the 500 ms preceding the movement, and computed the optimal angle as described above.

Experiment 3 - Connectivity analyses

Connectivity analyses were performed in FieldTrip on ROI signals, using the debiased weighted phase-lag index (debiased WPLI) ⁴³. We focused on the temporal window from 0.2 to 1.2 seconds post-movement, spanning symmetrically the temporal interval between movement onset (0 s) and the behavioural response (the agency rating was requested at 1.4 s), where we expect phase related connectivity changes to be strongest. For each subject, trials were split in optimal or non-optimal, depending on whether the phase was closer than π/3 to the individual optimal phase, or to the non-optimal phase, selecting approximately 33 %of the trials within each group. Since optimal and non-optimal trials are selected at the individual level based on behavioural results, we excluded from further analyses 5 subjects showing no significant phase opposition in any time-frequency point in the 8–13 Hz, -0.5-0 s range, as the optimal phase angle cannot be meaningfully defined (see Fig. S12-S13 for alternative analyses on the whole sample or using alternative definitions of the optimal phase). Then, average connectivity values were computed for each subject and pair of ROIs for optimal and non-optimal trials, for 42 linearly spaced frequencies in the 4–45 Hz range. We then averaged connectivity values from the left SMA to all other regions and contrasted them between optimal and non-optimal trials by mean of paired t-tests and corrected for multiple comparisons across frequencies using a cluster-based permutation test. We analysed connectivity values in the frequency range which showed significant changes in this first analysis (9–12 Hz) separately for each of the eight macro-regions, by averaging all connections from the left SMA to all other ROIs belonging to each macro-region.

The directionality analysis was performed using custom MATLAB code, and using a method inspired by a previous work on hippocampal theta oscillations ⁴⁴. The key idea is to look at time-delayed correlations between the phase of oscillations of the seed region (left SMA) and of ROIs belonging to the target region, within the usual time-frequency range (0.2–1.2 s, 9–12 Hz). For each pair of analysed regions and frequency (9, 10, 11 and 12 Hz), we computed the difference of oscillatory phase angles (extracted through Morelet wavelets as described above), while introducing a variable delay in the signal of the target region. Then, we computed the inter trial coherence (ITC) of phase differences as a function of the temporal lag. The idea is that the delay at which maximal coherence is observed indicates the preferred direction of information flow. If the delay on the target region is positive, it means that the present of the seed region best predicts the future of the target, and that information flows from the former to the latter. The difference was computed separately for each temporal delay, trial, frequency and ROI of a macro-region, and subsequently averaged to keep only its dependency on temporal lag. To perform statistical analyses on the effect of pre-movement phase, we extracted the peak of the lag-ITC curve in optimal and non-optimal phase trials for each subject and compared them using a paired t-test.

Gallagher, S. Philosophical conceptions of the self: implications for cognitive science. Trends Cogn Sci 4, 14–21 (2000).
Synofzik, M., Vosgerau, G. & Newen, A. I move, therefore I am: A new theoretical framework to investigate agency and ownership. Conscious Cogn 17, 411–424 (2008).
Blanke, O. & Metzinger, T. Full-body illusions and minimal phenomenal selfhood. Trends Cogn Sci 13, 7–13 (2009).
Zalla, T. & Sperduti, M. The sense of agency in autism spectrum disorders: a dissociation between prospective and retrospective mechanisms? Front Psychol 6, 1–7 (2015).
Gallagher, S. & Trigg, D. Agency and anxiety: Delusions of control and loss of control in schizophrenia and agoraphobia. Front Hum Neurosci 10, 1–12 (2016).
Blakemore, S. J., Wolpert, D. & Frith, C. Why can’t you tickle yourself? Neuroreport 11, R11-6 (2000).
Frith, C. D., Blakemore, S. J. & Wolpert, D. M. Abnormalities in the awareness and control of action. Philosophical Transactions of the Royal Society B: Biological Sciences 355, 1771–1788 (2000).
Schwartz, A. B. Movement: How the Brain Communicates with the World. Cell 164, 1122–1135 (2016).
Wen, W. et al. Perception and control: individual difference in the sense of agency is associated with learnability in sensorimotor adaptation. Sci Rep 11, (2021).
Moore, J. W. What is the sense of agency and why does it matter? Front Psychol 7, 1–9 (2016).
Serino, A. et al. Sense of agency for intracortical brain–machine interfaces. Nat Hum Behav 6, 565–578 (2022).
Desmurget, M. & Sirigu, A. Conscious motor intention emerges in the inferior parietal lobule. Curr Opin Neurobiol 22, 1004–1011 (2012).
Fornia, L. et al. Direct electrical stimulation of the premotor cortex shuts down awareness of voluntary actions. Nat Commun 11, 705 (2020).
Fried, I., Mukamel, R. & Kreiman, G. Internally Generated Preactivation of Single Neurons in Human Medial Frontal Cortex Predicts Volition. Neuron 69, 548–562 (2011).
Chambon, V., Wenke, D., Fleming, S. M., Prinz, W. & Haggard, P. An Online Neural Substrate for Sense of Agency. Cerebral Cortex 23, 1031–1037 (2013).
Chambon, V., Moore, J. W. & Haggard, P. TMS stimulation over the inferior parietal cortex disrupts prospective sense of agency. Brain Struct Funct 220, 3627–3639 (2015).
Desmurget, M. et al. Movement Intention After Parietal Cortex Stimulation in Humans. Science 324, 811–813 (2009).
Farrer, C. & Frith, C. D. Experiencing Oneself vs Another Person as Being the Cause of an Action: The Neural Correlates of the Experience of Agency. Neuroimage 15, 596–603 (2002).
Libet, B., Gleason, C. A., Wright, E. W. & Pearl, D. K. Time of conscious intention to act in relation to onset of cerebral activity (readiness-potential). The unconscious initiation of a freely voluntary act. Brain 106 (Pt 3), 623–42 (1983).
Schurger, A., Sitt, J. D. & Dehaene, S. An accumulator model for spontaneous neural activity prior to self-initiated movement. Proceedings of the National Academy of Sciences 109, E2904–E2913 (2012).
Zito, G. A., Wiest, R. & Aybek, S. Neural correlates of sense of agency in motor control: A neuroimaging meta-analysis. PLoS One 15, e0234321 (2020).
Fries, P. A mechanism for cognitive dynamics: Neuronal communication through neuronal coherence. Trends Cogn Sci 9, 474–480 (2005).
Fries, P. Rhythms for Cognition: Communication through Coherence. Neuron 88, 220–35 (2015).
Ai, L. & Ro, T. The phase of prestimulus alpha oscillations affects tactile perception. J Neurophysiol 111, 1300–1307 (2014).
Busch, N. A., Dubois, J. & VanRullen, R. The phase of ongoing EEG oscillations predicts visual perception. Journal of Neuroscience 29, 7869–7876 (2009).
Hanslmayr, S., Volberg, G., Wimber, M., Dalal, S. S. & Greenlee, M. W. Prestimulus oscillatory phase at 7 Hz gates cortical information flow and visual perception. Current Biology 23, 2273–2278 (2013).
Rohe, T., Ehlis, A. C. & Noppeney, U. The neural dynamics of hierarchical Bayesian causal inference in multisensory perception. Nat Commun 10, 1–17 (2019).
Sherman, M. T., Kanai, R., Seth, A. K. & VanRullen, R. Rhythmic Influence of Top–Down Perceptual Priors in the Phase of Prestimulus Occipital Alpha Oscillations. J Cogn Neurosci 28, 1318–1330 (2016).
Samaha, J., Bauer, P., Cimaroli, S. & Postle, B. R. Top-down control of the phase of alpha-band oscillations as a mechanism for temporal prediction. Proceedings of the National Academy of Sciences 112, 8439–8444 (2015).
Noel, J.-P. et al. Human primary motor cortex indexes the onset of subjective intention in brain-machine-interface mediated actions. bioRxiv 2023.07.21.550067 (2023) doi:10.1101/2023.07.21.550067.
Bouton, C. E. et al. Restoring cortical control of functional movement in a human with quadriplegia. Nature 533, 247–250 (2016).
Tomassini, A., Ambrogioni, L., Medendorp, W. P. & Maris, E. Theta oscillations locked to intended actions rhythmically modulate perception. Elife 6, 1659–1668 (2017).
Thézé, R., Giraud, A.-L. & Mégevand, P. The phase of cortical oscillations determines the perceptual fate of visual cues in naturalistic audiovisual speech. Sci Adv 6, eabc6348 (2020).
Zhang, Y., Zhang, Y., Cai, P., Luo, H. & Fang, F. The causal role of α-oscillations in feature binding. Proceedings of the National Academy of Sciences 116, 17023–17028 (2019).
VanRullen, R. How to evaluate phase differences between trial groups in ongoing electrophysiological signals. Front Neurosci 10, 1–22 (2016).
Bazanova, O. M. & Vernon, D. Interpreting EEG alpha activity. Neuroscience and Biobehavioral Reviews vol. 44 Preprint at https://doi.org/10.1016/j.neubiorev.2013.05.007 (2014).
Berchicci, M. et al. Development of mu rhythm in infants and preschool children. Dev Neurosci 33, 130–43 (2011).
Foldes, S. T., Weber, D. J. & Collinger, J. L. Altered modulation of sensorimotor rhythms with chronic paralysis. J Neurophysiol 118, (2017).
Wenke, D. & Haggard, P. How voluntary actions modulate time perception. Exp Brain Res 196, 311–318 (2009).
Einevoll, G. T., Kayser, C., Logothetis, N. K. & Panzeri, S. Modelling and analysis of local field potentials for studying the function of cortical circuits. Nat Rev Neurosci 14, 770–785 (2013).
Martin, K. A. C. & Schröder, S. Phase Locking of Multiple Single Neurons to the Local Field Potential in Cat V1. Journal of Neuroscience 36, 2494–2502 (2016).
Pascual-Marqui, R. D. Discrete, 3D distributed, linear imaging methods of electric neuronal activity. Part 1: exact, zero error localization. arXiv: Mathematical Physics (2007).
Vinck, M., Oostenveld, R., van Wingerden, M., Battaglia, F. & Pennartz, C. M. A. An improved index of phase-synchronization for electrophysiological data in the presence of volume-conduction, noise and sample-size bias. Neuroimage 55, 1548–1565 (2011).
Siapas, A. G., Lubenov, E. V. & Wilson, M. A. Prefrontal Phase Locking to Hippocampal Theta Oscillations. Neuron 46, 141–151 (2005).
Eschenko, O., Magri, C., Panzeri, S. & Sara, S. J. Noradrenergic neurons of the locus coeruleus are phase locked to cortical up-down states during sleep. Cerebral Cortex 22, (2012).
Synofzik, M., Vosgerau, G. & Voss, M. The experience of agency: An interplay between prediction and postdiction. Front Psychol 4, 1–8 (2013).
Haggard, P. Sense of agency in the human brain. Nat Rev Neurosci 18, 196–207 (2017).
Kornhuber, H. H. & Deecke, L. Hirnpotentialänderungen bei Willkürbewegungen und passiven Bewegungen des Menschen: Bereitschaftspotential und reafferente Potentiale. Pflügers Arch 284, 1–17 (1965).
Astington, J. W. The child’s discovery of the mind. Developing child (1993).
Haegens, S., Nácher, V., Luna, R., Romo, R. & Jensen, O. α-Oscillations in the monkey sensorimotor network influence discrimination performance by rhythmical inhibition of neuronal spiking. Proceedings of the National Academy of Sciences 108, 19377–19382 (2011).
Desideri, D., Zrenner, C., Ziemann, U. & Belardinelli, P. Phase of sensorimotor μ-oscillation modulates cortical responses to transcranial magnetic stimulation of the human motor cortex. Journal of Physiology 597, 5671–5686 (2019).
Sperduti, M., Delaveau, P., Fossati, P. & Nadel, J. Different brain structures related to self- and external-agency attribution: a brief review and meta-analysis. Brain Struct Funct 216, 151–157 (2011).
Seghezzi, S. & Zapparoli, L. Predicting the Sensory Consequences of Self-Generated Actions: Pre-Supplementary Motor Area as Supra-Modal Hub in the Sense of Agency Experience. Brain Sci 10, 825 (2020).
Moore, J. W., Ruge, D., Wenke, D., Rothwell, J. & Haggard, P. Disrupting the experience of control in the human brain: Pre-supplementary motor area contributes to the sense of agency. Proceedings of the Royal Society B: Biological Sciences 277, 2503–2509 (2010).
Peelen, M. V., Wiggett, A. J. & Downing, P. E. Patterns of fMRI Activity Dissociate Overlapping Functional Brain Areas that Respond to Biological Motion. Neuron 49, 815–822 (2006).
Pelphrey, K. A., Morris, J. P., Michelich, C. R., Allison, T. & McCarthy, G. Functional Anatomy of Biological Motion Perception in Posterior Temporal Cortex: An fMRI Study of Eye, Mouth and Hand Movements. Cerebral Cortex 15, 1866–1876 (2005).
Wolpert, D. M., Ghahramani, Z. & Jordan, M. I. An Internal Model for Sensorimotor Integration. Science 269, 1880–1882 (1995).
Doricchi, F., Lasaponara, S., Pazzaglia, M. & Silvetti, M. Left and right temporal-parietal junctions (TPJs) as “match/mismatch” hedonic machines: A unifying account of TPJ function. Phys Life Rev 42, 56–92 (2022).
van Kemenade, B. M. et al. Distinct Roles for the Cerebellum, Angular Gyrus, and Middle Temporal Gyrus in Action–Feedback Monitoring. Cerebral Cortex 29, 1520–1531 (2019).
Berti, A. et al. Shared cortical anatomy for motor awareness and motor control. Science 309, 488–491 (2005).
Zanto, T. P., Rubens, M. T., Thangavel, A. & Gazzaley, A. Causal role of the prefrontal cortex in top-down modulation of visual processing and working memory. Nat Neurosci 14, 656–661 (2011).
van Ede, F., de Lange, F., Jensen, O. & Maris, E. Orienting Attention to an Upcoming Tactile Event Involves a Spatially and Temporally Specific Modulation of Sensorimotor Alpha- and Beta-Band Oscillations. Journal of Neuroscience 31, 2016–2024 (2011).
Buchholz, V. N., Jensen, O. & Medendorp, W. P. Different roles of alpha and beta band oscillations in anticipatory sensorimotor gating. Front Hum Neurosci 8, (2014).
Zanto, T. P., Chadick, J. Z. & Gazzaley, A. Anticipatory alpha phase influences visual working memory performance. Neuroimage 85, 794–802 (2014).
Sadaghiani, S. et al. Alpha-Band Phase Synchrony Is Related to Activity in the Fronto-Parietal Adaptive Control Network. Journal of Neuroscience 32, 14305–14310 (2012).
Moretto, G., Walsh, E. & Haggard, P. Experience of agency and sense of responsibility. Conscious Cogn 20, 1847–1854 (2011).
Arango-Muñoz, S. Joëlle Proust: The Philosophy of Metacognition: Mental Agency and Self-Awareness. Minds Mach (Dordr) 25, 297–300 (2015).
Cortes, C. & Vapnik, V. Support-Vector Networks. Mach Learn 20, 273–297 (1995).
Thenaisie, Y. et al. Principles of gait encoding in the subthalamic nucleus of people with Parkinson’s disease. Sci Transl Med 14, eabo1800 (2022).
Quiroga, R. Q., Nadasdy, Z. & Ben-Shaul, Y. Unsupervised spike detection and sorting with wavelets and superparamagnetic clustering. Neural Comput 16, 1661–1687 (2004).
Hagmann, P. et al. Mapping the structural core of human cerebral cortex. PLoS Biol 6, (2008).
Maris, E. & Oostenveld, R. Nonparametric statistical testing of EEG- and MEG-data. J Neurosci Methods 164, 177–190 (2007).

There is NO Competing Interest.

Supplementaryfigures.docx

Download PDF

Version 1

posted

You are reading this latest preprint version

Pre-movement sensorimotor oscillations shape the sense of agency by gating cortical connectivity

Status:

Version 1

Abstract

Figures

Introduction

Results

Low alpha LFP oscillations capture modulations of M1 firing

The optimal phase for agency is associated with increased alpha-band functional connectivity

Discussion

Methods

Experiment 1 and 2 - participant

Experiment 1 and 2 - BMI system

Experiment 1 – protocol

Experiment 2 – protocol

Experiment 3 - participants

Experiment 3 - EEG acquisition

Experiment 3 - protocol

- Part 1 – BMI training

- Part 2 – agency assessment

Experiment 3 - BMI decoder

Experiment 1 and 2 - data pre-processing

Experiment 3 - data pre-processing

Experiment 3 - source reconstruction and templates

Experiment 1 - Multiunit analyses

Experiment 2 - statistical analysis of behavioural data

Phase opposition analyses

Experiment 1 and 2 - statistical analysis of phase opposition

Experiment 3 – statistical analysis of phase opposition and behaviour

Experiment 3 - Determination of optimal phase angles and trials

Experiment 3 - Connectivity analyses

References

Additional Declarations

Supplementary Files

Status:

Version 1