Introduction

Sense of agency refers to the conscious experience of being in control of one’s actions [1]. Thus, its pervasive nature in conscious awareness of one’s interaction with the external world has attracted the attention of psychologists, neuroscientists, and philosophers. Despite being a key factor interlinked with our minimal self-hood [2], the experience associated with the feeling of agency over one’s actions is not as salient as one would expect. Thus, it remains challenging to study such a ‘phenomenally thin’ [3] observation.

In the last two decades, several attempts have been made to measure the sense of agency, which have been broadly categorized into two groups of measurement, i.e., explicit and implicit measures [4]. The explicit measures of agency rely on subjective self-reports and, thus, suffer from consistent cognitive bias in terms of misattribution of agency, especially for the actions with positive outcomes [5,6]. On the other hand, implicit measure attempts to quantify the feeling of agency without explicitly making one think over agency over the performed action. Following the seminal work of Patrick Haggard and colleagues [7], distortion in time perception in an ‘intentional binding’ setup has been reported to be linked with the implicit measure of sense of agency. The perceived compression of the time interval between a voluntary action and its perceived external sensory outcome is referred to as the intentional binding effect [7].

Many researchers have explored the connection between intentional binding and psychological or behavioral traits. Recent studies have shown that interoception, which is the awareness of the body’s internal physiological state, like the systole or diastole phase in the cardiac cycle, affects the sense of agency [8,9].

Studies have reported about the so-called ‘social sense of agency’ where one’s activities trigger the reactions of others, one’s actions together modulate their reactions, or their simple social presence affects one’s sense of agency [10]. In a work by Yoshie and Haggard [11], social-emotional modulation of sense of agency has been reported in which the authors identified a mechanism whereby routinely negative action outcomes could influence sense of agency. Some recent studies have also discussed the impact of social, affective, and other kinds of non-motoric cues in modulating the sense of agency [12–14]. The researchers in [12] found that emotions like fear or anger can affect the subjective sense of control over an action outcome, even though the actual causal relationship between action and outcome remains unchanged. According to [15], the emotional content of an intention is significant, indicating that implicit bias is heightened when we express a negative intention. Intentional binding has been influenced by beliefs in free will [16], obsessive-compulsive tendencies [17], and exposure to emotionally positive images [18]. It has also been affected by low or vulnerable-type narcissism [19,20] and the recall of episodic memories in which individuals felt a sense of power over others [21], as well as the activation of depressing memories [22].

It is worth mentioning that the impact of interventions such as self-reflection and meditation, which directly affect different aspects of one’s sense of self, on the implicit measure of sense of agency has not received much attention. Recently, Lush et al. [23] reported that mindfulness meditators exhibited stronger intentional binding (stronger compression between action and outcome) compared to subjects easily hypnotizable due to higher meta-cognitive access to one’s intentions [23,24]. On the contrary, Chiarella et al. [25] reported weakening intentional time binding for mindfulness meditators due to less pronounced attachment to self [25]. Thus, an absence of clear consensus over how self-reflection and contemplation affect the sense of agency calls for more studies in this direction. Moreover, Silvanto and Nagai [26] have recently proposed a theoretical framework emphasizing the role of interoception and the insula-anterior cingulate cortex network in integrating bodily signals with sensory and cognitive information, which is critical for vivid mental imagery and a robust sense of agency. Therefore, it is important to investigate the neural correlates of self-reflection and meditation in relation to implicit measures of agency. This also underscores the need to study how electroencephalogram (EEG)-based neural markers of interventions such as self-reflection relate to the implicit sense of agency.

In this study, we aimed to explore the effects of self-reflection upon sense of agency and the neural underpinnings, by examining two distinct modes: self-centered reflection and self-less reflection on life events.

The first mode, termed “self-centeredness", involved the self as a central reference point in various psychological activities, marked by biased self-interest and an inflated sense of importance. In contrast, the second aspect, termed “selflessness", was characterized by diminished self-centeredness and a reduced emphasis on the self’s significance. This latter mode of functioning was associated with altruism, kindness, empathy, and a weak distinction between self and others [27].

Thus, to gain a comprehensive understanding of these cognitive states, we examined traditional EEG features such as band powers and novel topological data analysis (TDA) [28,29] based measures like Hodge spectral entropy [30], persistent entropy, and persistent amplitude. We aimed to identify neural correlates associated with two types of self-reflection. Topological features of EEG time series have been successful in uncovering meaningful insights into cognitive processes, emotional states, and neurological disorders [30–34]. This approach transforms EEG time series data from every single electrode into a point cloud in a higher-dimensional space and extracts temporal topological features at various scales of filtration.

Our study found that self-centered reflection (SCR) on life events enhanced the overall sense of agency through increased judgment error in time perception, also known as total time binding (TTB), unlike a control group (CTR) without self-reflective activity. On the other hand, selfless reflection (SLR) reduced TTB, decreasing the sense of agency compared to the CTR group. The topological signatures and band power features of EEG signals exhibited significant differences between self-centered and selfless reflections. Our findings revealed a statistically significant negative correlation between the sense of agency and Hodge spectral entropy, which denoted the amount of topological information needed to describe the temporal connectivity patterns of a single EEG channel’s time series. Thus, a negative correlation indicated that a stronger sense of agency might be associated with less complex temporal connections of the neural processes. This could be a potential biomarker associated with different reflective mental states. We also explored another measure of temporal complexity known as multiscale entropy (MSE). We found that the MSE for self-centered reflection was lower than the selfless and control groups. Higher MSE values indicated greater signal complexity, which varies significantly in different neurological conditions [35,36].

The organization of this manuscript is as follows. We first delineate the experimental framework. Next, we briefly introduce the essential aspects of TDA and the definitions of different topological measures. We then present our findings and associated discussions. Finally, we conclude with a summary of our work.

Experimental framework

Experiment design

We employed a between-subject experimental design. During the experiment, participants were comfortably seated in a quiet room, 50 cm from a MacBook Pro (13.3-inch screen) with a screen resolution of 2560 x 1600 pixels and a refresh rate of 60 Hz, equipped with an Apple M2 processor. Participants were randomly assigned to one of three groups:

• SCR (16 males, 15 females) for 240 seconds,
• SLR (16 males, 15 females) for 240 seconds,
• CTR (15 males, 15 females) for 240 seconds.

Participants in the SCR group were asked to reflect on eight questions (30 seconds each) related to their past and future, focusing on personal successes and failures. These questions were as follows:

• Visualize the last time someone praised you a lot.
• Visualize the last time someone rebuked you harshly.
• Reflect on occasions when your hard efforts brought you great success.
• Reflect on occasions when you failed despite putting in significant effort.
• How do you perceive yourself when you stand before a mirror now?
• How do you imagine you will look when you stand before a mirror in 25 years?
• Contemplate your biggest current worry.
• Contemplate the anxieties you have about your future.

These questions were inspired by previous studies [41] investigating the effects of autobiographical recall on cognitive performance. Participants in the SLR group were asked to reflect on eight questions (30 seconds each) related to acts of selflessness. These questions were as follows:

• Think of a time when you felt grateful for your body as a gift from God.
• Recall a moment when you helped someone without expecting anything in return.
• Think of a situation when you shared something important with someone in need.
• Remember when you stayed calm in a difficult situation, trusting God’s plan.
• Recall a time when you prayed for guidance to act according to God’s will.
• Think of a time when you let go of your ego and embraced humility.
• Visualize contributing to others’ well-being in the future.
• Envision yourself sharing joy and wisdom, not for credit, but to enrich the lives of others.

These questions were originally part of a mantra-based CBT study [42] to invoke selflessness. Prior to formal data collection, both SCR and SLR prompt sets were pilot-tested on 12 volunteers representative of the target population; their structured feedback led to minor wording refinements that improved clarity, emotional resonance, and ecological validity.

The CTR underwent no specific contemplation and sat silently for 240 seconds. Appropriate event markers were recorded during different experiment phases using the Pylsl framework [43]. This passive resting-state condition served as a baseline for comparison with the active reflection conditions.

Intentional time binding.

Following the reflections above, each group underwent a cognitive experiment on intentional time binding to measure the judgment error in time perception. This cognitive task was popularised by Haggard et al. [7] to study the implicit measure of sense of agency. The task had four randomized conditions, i.e., action-only, tone-only, and two action-tone. Each condition included 20 trials, for a total of 80 trials. The block diagram for the experiment is given in Fig 1. The total duration of the task was around 15 minutes.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1. (Left) Overall pipeline of the experiment for the three groups.

A beep sound follows every reflection to indicate to open the eyes for the next task. An intentional time binding experiment is carried out for each group at the end. (Right) Schematic timeline of the intentional time binding experiment.

https://doi.org/10.1371/journal.pone.0335276.g001

In all the trials, participants were confronted with a clock featuring a red dot rotating along the edge. In the action-only condition, they pressed any key at will while the red dot rotated. After their key press, the red dot stopped its rotation after a random, short duration. After the red dot stopped rotating, participants clicked on the clock face at the location where they believed the red dot was positioned at the moment they pressed the key.

In the tone-only condition, participants waited for an audio tone. After hearing the tone, the red dot stopped rotating after a brief, random delay. Participants then clicked on the clock face at the location where they believed the red dot was positioned at the moment of the tone. These two conditions served as the baseline.

In action-tone conditions, the participants initiated actions by pressing a key at their discretion, followed by a 250-millisecond delay before an audio tone sounded. This condition had two variations: in one, participants clicked on the clock face where they thought a red dot was at the time of the key press; in the other, they clicked where they believed the red dot was at the time of the audio tone.

Participants also filled out a questionnaire that asked them to mark the number of erroneous markings they believed they made in each phase. The markings were based on their confidence in their own judgment about the red dot’s location. At the end of the experiment, perceived mental load and feedback about the overall experiment procedure were rated on a scale of 1 to 5. This information was used to assess if any participants should be excluded from the analysis due to excessively negative feedback, which might indicate a lack of engagement or difficulty in understanding the task. The responses from two participants were excluded from the time binding analysis on the basis of subjective reports that they had fallen asleep during the time binding task.

Theoretical framework

Topological features

The EEG time series data is embedded as a point cloud using Takens’s embedding theorem [48]. According to Takens’s embedding Theorem, we can construct vectors in a higher-dimensional space by taking successive points from the time series. Suppose we have a time series . We obtain the resulting vectors in the reconstructed m-dimensional space:

(1)

Here, m is the embedding dimension, and is the time delay. If the system is periodic, the point cloud might resemble a loop or closed curve. In the case of chaotic dynamics, the point cloud might exhibit a fractal-like, intricate structure.

In TDA, the primary component is constructing a simplicial complex. It consists of simplices such as points (0-simplex), edges (1-simplex), and triangles (2-simplex) etc. For a point cloud , we construct a simplicial complex using the Vietoris Rips (VR) algorithm (see Fig 2). A k-simplex represents a collection of (k + 1) points in P that are sufficiently close to each other, according to the specified distance threshold r. Given r > 0, we thus define the simplicial complex K_r by considering all simplices such that for all in . In the VR complex, r is the filtration parameter.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 2. Generating Vietoris-Rips (VR) Complexes from Point Cloud Data: Each point in the point cloud is constructed from an EEG time series using Takens’s embedding theorem.

Each point is enveloped by a growing sphere, which simultaneously and uniformly expands, forming k-simplices [47].

https://doi.org/10.1371/journal.pone.0335276.g002

One of the key variables in this analysis [30] is the so-called Betti numbers () corresponding to k^th homology dimension. counts the number of k-dimensional holes in the simplicial complex. counts the number of connected components, , the number of 1-dimensional holes or circles, and , the 2-dimensional holes or voids. Next, we present a few important topological features used in this work.

Persistent homology features.

Persistent homology accounts for the topological features that remain persistent across different values of filtration parameter r. They are as follows.

• Betti curve : The Betti curve, represented as , is a function that counts the multiplicity of points in the homology group associated with a filtration D_k at a specific filtration parameter r = s. This counting is constrained by the condition , where b_i denotes the r at which the feature appeared, also termed as ‘birth’ of the topological feature and d_i indicates its ‘death’ or disappearance of the feature. Consequently, the Betti curve provides insights into the persistence of topological features as the filtration parameter varies.
• Persistence entropy (): The persistence entropy quantifies the distribution of points in a persistence diagram (collection of persistence intervals D_k), reflecting the complexity of topological features. Let denote the set of birth–death pairs for the order homology group, with . The order persistence entropy is defined as(2)where
• Persistent amplitude (A_k): The persistent amplitude for the order homology group is defined on D_k, the set of persistent pairs, as a function , for which there exists a vectorization , where V is a normed space. The relationship between the amplitude and the vectorization is given by for all . For norm computations, we utilize the Wasserstein metric.

Hodge spectral entropy.

In information theory, entropy is a metric for the degree of randomness or uncertainty in a system. In TDA, one finds the eigenvalues of the so-called Hodge Laplacian L_[k] [30,49], a higher-dimensional analog of the graph Laplacian. Hodge spectral entropy is expressed as follows:

(3)

where denotes the eigenvalue of L_[n], is the damping factor, and N_n is the total number of eigenvalues of L_[n]. We used the gudhi [50] and hodgelaplacians [51] libraries to compute Hodge spectral entropy. Higher Hodge spectral entropy implies richer topological information content.

Conventional EEG features

Relative band power.

The power spectral density (PSD) is computed using Welch’s method with MNE. We estimate the relative band powers for each intervention in the six bands - delta (0.5 Hz, 5.0 Hz), theta (5.0 Hz, 8.0 Hz), alpha (8.0 Hz, 13.0 Hz), sigma (13.0 Hz, 16.0 Hz), beta (16.0, 30.0 Hz), gamma (30.0 Hz, 45.0 Hz) [52] by taking an average of PSD in each of these bands and dividing with the summation of averages over all the bands.

Multiscale entropy (MSE).

MSE facilitates the analysis of signal complexity across various timescales. When dealing with a one-dimensional discrete time series, , we generate successive coarse-grained time series, represented as , corresponding to the scale factor, . The original times series is divided into non-overlapping windows of length , and subsequently, data points within each window are averaged. In a more general sense, each element of a coarse-grained time series is computed using the equation

(4)

For the scale factor of one, the time series remains identical to the original. The length of each coarse-grained time series is equivalent to the original time series divided by the scale factor . Subsequently, the sample entropy (SampEn) measure is computed for each coarse-grained time series, constituting what we refer to as multiscale entropy () analysis.

The SampEn for a time series is calculated using a defined pattern length m and a tolerance level r. The formula is as follows:

(5)

Where is the number of sequences of length m that are similar, i.e., with a maximum difference between any two corresponding points less than or equal to r.

methodology provides a means to explore the complexity and regularity of data across multiple scales or levels of detail. We have used pyEntropy [53] library for computation.

Results

Effect on time binding

There was an increase in TTB during the intentional binding task in the SCR group compared to the SLR and CTR groups (see Table 1). It indicates that participants reported a larger difference between when they perceived an action or event occurred and when it actually occurred, suggesting a stronger perception of the event being closely related to their intentional actions or mental states after the SCR compared to the other conditions (SLR and CTR).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. Time binding results (in ms).

Action/Tone Binding quantifies the difference between baseline and agency-related errors in action and tone perception. Baseline errors are the average time discrepancies between reported and actual actions/tone over 20 trials in action-only or tone-only tasks. Agency-related errors measure the corresponding discrepancies during action-tone tasks. Total time binding reflects the difference between action and tone bindings. Values are presented as mean (SD) across subjects in each group

https://doi.org/10.1371/journal.pone.0335276.t001

The three binding effects - action, tone, and total binding were tested for normality for the three groups. Upon the Shapiro-Wilk test for normality of distributions, we found that the Tone Binding associated with Group SCR (p=0.9) and Group SLR (p=0.7) followed normal distributions. Also, the Action Binding effect for Group CTR (p=0.2) was normally distributed. The remaining data were not normally distributed. Thus, Mann Whitney U-test (non-parametric test) was used to compare the means of TTB (see Table 2).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 2. Omnibus Kruskal–Wallis test and Mann–Whitney post-hoc p-values for TTB.

Effect-size magnitude is interpreted using conventional cut-offs: small (.01–.06), medium (.06–.14), large (.14)

https://doi.org/10.1371/journal.pone.0335276.t002

The choice of single-tailed hypothesis was made in order to validate the hypothesis central to the study that SCR intervention enhances agency while SLR intervention reduces agency compared to the CTR condition. The mean (SD) for Action Binding after the SCR intervention was -40 (84) ms, following the SLR intervention was 25 (85) ms, and after the CTR condition was 6 (65) ms. In contrast, Tone Binding exhibited mean (SD) values of 49 (72) ms after the SCR intervention, 39 (66) ms following the SLR intervention, and 29 (65) ms after the CTR condition. No significant differences were observed between the groups for both action and tone binding based on the Mann-Whitney U test.

The mean (SD) for TTB in the SCR group was 123 (75) ms, in the SLR group was 51 (41) ms, and in the CTR group was 82 (54) ms. Statistical analysis revealed that TTB associated with the SCR intervention was significantly higher than the SLR intervention (), the SLR intervention was significantly lower than the CTR condition (), and the SCR intervention was significantly higher than the CTR condition (). The omnibus Kruskal–Wallis test confirmed a significant group difference (), with a large effect size ().

Also, the violin plots representing the distribution of TTB are shown in Fig 3. Differences in shape suggest variations in the distributions for the three groups. The medians showed a similar trend discussed above, confirming the hypothesis that the TTB of SCR is greater than CTR, which is greater than SLR. A broader violin shape for the SLR group centered around the median suggested significant data dispersion, particularly clustered around a relatively low median value.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 3. Violin plot representing the distribution of total time binding (in ms) for different groups.

Group SCR – Self-Centered Reflection, Group SLR - Selfless Reflection, Group CTR – Control.

https://doi.org/10.1371/journal.pone.0335276.g003

Effect on relative band powers

Box plots illustrating the various band powers across different channels for the three groups are presented in Fig 4. Among the six bands (delta, theta, alpha, sigma, beta, and gamma), the sigma, beta, and gamma bands had much lower relative band powers than the delta, theta, and alpha bands. For the relative alpha band power, the median for the SCR group was lower than that of the SLR and CTR groups in both TP and AF channels. Conversely, for theta bands, there was a higher median relative band power for the SCR group compared to the CTR group. In the case of the delta band, in the AF channel, group SLR displayed a substantially lower median band power than others. However, this was not the case in the TP channel.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 4. Box plot of the band powers for different groups.

https://doi.org/10.1371/journal.pone.0335276.g004

We performed the Kruskal–Wallis test to determine if the medians for the three groups were significantly different. In the TP channel, the higher frequency (>8 Hz) bands were found to be significantly different: alpha band (p = 0.03, , small–medium), sigma band (p = 0.01, , medium), beta band (p = 0.02, , medium), and gamma band (p = 0.0005, , large). For the AF channel, only the gamma band (p = 0.009, , medium) was found to be significantly different.

Effect on persistent homology features

• Betti curve : Fig 5 depicts the average Betti curves for the SCR, SLR, and CTR groups across all subjects, categorized by TP (top panel) and AF (bottom panel) channels. The shaded area represents the 95% confidence interval. The Betti numbers, , , and , correspond to the number of connected components, holes, and voids in the data, respectively. For , as r increased, the number of connected components decreased. Given the comparable sizes of the point clouds constructed from EEG time series data for all three groups, the curves showed similar initial values, although their slopes differed. SCR and CTR exhibited similar slopes, while SLR showed distinct slope. Additionally, the peaks of the higher-order Betti curves ( and ) revealed distinct patterns across groups and electrode positions (TP/AF).
• Persistence entropy (): Tables 3 and 4 present the mean persistence entropy and the corresponding Kruskal–Wallis p-values for each group in the TP and AF channels, respectively. All p-values were found to be less than 0.05, demonstrating statistical significance. The SLR group exhibited the highest entropy across all three groups and all homology dimensions in both TP and AF channels. For and , the values for the SCR and CTR groups were comparable; however, for , SCR showed higher values than CTR in both channels. The corresponding effect sizes were found to be large for all comparisons, indicating substantial group-level differences in entropy across conditions.
• Persistent amplitude (A_k): The mean persistence amplitude for each group in the TP and AF channels is also shown in Tables 3 and 4. All p-values were found to be less than 0.05, demonstrating statistical significance. Notably, the SLR group exhibited the lowest amplitude across all three groups for all homology dimensions in both the TP and AF channels. For all homology dimensions, the values for the SCR and CTR groups were similar in both channels. The effect sizes for these amplitude-based comparisons were also large in most cases, except for the AF channel where the effect size for A₁ was moderate.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 3. Results of the Kruskal-Wallis Test on persistent amplitude and persistent entropy measures for TP channel.

Effect sizes are reported using epsilon-squared ()

https://doi.org/10.1371/journal.pone.0335276.t003

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 4. Results of the Kruskal-Wallis Test on persistent amplitude and persistent entropy measures for AF channel.

Effect sizes are reported using epsilon-squared ()

https://doi.org/10.1371/journal.pone.0335276.t004

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 5. Betti curves averaged across subjects from SCR, SLR, and CTR groups.

The top (bottom) panel gives curves for the TP (AF) channel of the EEG. represents for k–th homology group.

https://doi.org/10.1371/journal.pone.0335276.g005

Effect on multiscale entropies

The multiscale entropies S_MS for the three groups in TP and AF channels are depicted in Fig 6. In the TP channel, we noted that at smaller scales, groups SLR and CTR exhibited identical sample entropies but differed from the SCR group. Conversely, at larger scales in the TP channel, groups SCR and SLR displayed identical sample entropies but differed from the CTR group. In the AF channel, a distinct separation was observed between group SCR and the other groups. In the AF channel, groups SLR and CTR were distinguishable either at very small or very large scales.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 6. Multiscale entropies for TP and AF channels.

https://doi.org/10.1371/journal.pone.0335276.g006

Effect on Hodge spectral entropies

Box plots depicting the Hodge Spectral Entropies across different channels for the three groups are shown in Fig 7. The labels correspond to the zeroth order, first order, and second order Hodge Spectral entropies. The TP channel showed a consistent upward trend in and from the SCR group to the CTR group to the SLR group, in contrast to the opposite trend observed in their corresponding TTB scores. A similar pattern was evident in the AF channel, with the exception that in group CTR had a slightly higher median compared to group SLR. In the case of , there was an increase from group SCR to group SLR in the TP channel but not in the AF channel.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 7. Box plot of the Hodge spectral entropies for different groups.

is the k^th order Hodge spectral entropy associated with the EEG signal.

https://doi.org/10.1371/journal.pone.0335276.g007

Additionally, the interquartile range in Fig 7 for group SCR was significantly smaller compared to the SLR and CTR groups across all three orders of entropy. When a Kruskal–Wallis test was performed, we found that the distributions of the three groups were significantly different in the TP channel. We obtained the following p-values: p = 0.002 for , p = 0.03 for , and p = 0.3 for . In the AF channel, statistically significant differences were observed for () and (p = 0.02), but not for (p = 0.2). The corresponding effect sizes () indicated a consistent trend, with showing the largest effect in both TP () and AF () channels, while yielded negligible effects () and reflected small to moderate effects (–0.08). We would like to emphasize that correction methods to control for multiple statistical comparisons were not employed to avoid the risk of inflated type I errors based on the recommendations for pilot studies [54].

Relation between the Hodge spectral entropy and Total Time Binding (TTB)

We found that the TTB was negatively correlated with second-order Hodge spectral entropy in TP and AF channels for all groups, as indicated in Table 5. All these correlations were significant (p < 0.05) across all groups. In the AF channels, while the correlations were negative for all three groups, only the SCR and SLR groups exhibited significance, not the CTR group. Interestingly, action binding and tone binding did not demonstrate such correlational effects. Additionally, the zeroth and first-order Hodge spectral entropy did not exhibit similar effects.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 5. Pearson correlations (p-values) between total time binding (TTB) and second-order Hodge spectral entropy .

Bold represents significant results (p < 0.05)

https://doi.org/10.1371/journal.pone.0335276.t005

Discussion

The primary objective of this study was to test whether different forms of self-reflection - self-centered reflection (SCR), selfless reflection (SLR), and control reflection (CTR)—influence the implicit sense of agency, and to identify the corresponding neural correlates. Consistent with our primary hypothesis, intentional time binding (TTB) differed significantly across groups, with SCR increasing and SLR decreasing the binding effect relative to CTR. These results indicate that the orientation of self-reflection dynamically alters the perceived coupling between actions and their outcomes. In line with our secondary hypothesis, distinct EEG signatures were observed across the reflection conditions, both in spectral band powers and in topological features. Furthermore, as hypothesized, Hodge spectral entropy was found to significantly correlate with TTB, suggesting a mechanistic link between topological neural features and the modulation of agency.

To explore how self-reflection alters the implicit sense of agency we have performed the intentional time-binding task post two kinds of self-reflective states, namely self-centered and selfless. To understand the complex interplay between neural activity and sense of agency, we examined various EEG features, including band powers and TDA-based metrics such as Betti curves, persistent entropy, persistent amplitude, and Hodge spectral entropy. Some of the key findings from this study were as follows:

• Total time binding (TTB), an implicit measure of the sense of agency, diminished for the group engaged in selfless reflection (SLR) compared to the control group (CTR). Conversely, compared to CTR, it increased for a group engaged in self-centered reflection (SCR) when considering retrospective and prospective life successes and failures. The result indicated that self-centered (selfless) reflection strengthens (weakens) the associative relationship between an action and its consequence as the doer.
• The Kruskal-Wallis test showed significant differences in the higher frequency bands between the three groups. In the TP channel, alpha, sigma, beta, and gamma bands had statistically significant differences, while in the AF channel, only the gamma band showed a significant difference.
• The distributions of the three groups differed significantly for persistence entropy and amplitude across up to second homology dimensions in both the TP and AF channels. Additionally, Hodge spectral entropy showed significant differences for the first and second homology dimensions in both channels.
• The second-order Hodge spectral entropy of EEG for the TP channels demonstrated a significant negative correlation with TTB for all three groups, indicating that the higher information flow between the higher and lower-order simplicial chains corresponded to a reduced intentional time-binding effect. Higher information flow may be interpreted as more pathways or connections through which topological information can flow or propagate within the simplicial complex and correlated with reducing the binding of actions and their temporal consequences.

Our first major finding was related to the effects of the self-reflective states on sense of agency. While various intentional time-binding tasks have been used to investigate the specific causes of the sense of agency, our experiment employed a canonical intentional binding paradigm, in which all actions produced specific outcomes after a consistent short time interval during the agency blocks. The sense of agency is typically explained through mechanisms such as the brain’s predictive processes [55,56], causal mechanisms [57], and postdictive mechanisms [55], which suggest that the brain rewrites history to strengthen the feeling of agency. The group receiving the SCR intervention showed a greater intentional time-binding effect than the CTR group. As we prompted the subjects to reflect on their past and future successes and failures in the SCR intervention, we likely fostered a sense of self-centeredness. Dambrun et al. [27] stated that overemphasis on the self arises primarily from self-centeredness, indicating a heightened sense that one’s own situation was more significant than others. In contrast, during the CTR or SLR conditions, the self-agency system may not have been as actively engaged, suggesting a potential link to the enhanced time-binding effects in the SCR group. Here, the self-agency system may have caused participants to feel they had caused the outcome, effectively contracting the time interval between action and result.

Recent theoretical advances by Silvanto and Nagai [26] provide a compelling neurocognitive framework relevant to our findings. They propose that vivid mental imagery and a robust sense of agency arise from the integration of interoceptive bodily signals with sensory and cognitive representations, mediated by the insula and anterior cingulate cortex. Within this framework, SCR likely engages this integrative network more strongly, anchoring mental imagery in bodily signals and thereby enhancing the implicit sense of agency measured by intentional time binding. Conversely, SLR condition may shift attention away from interoceptive bodily awareness, disrupting this integrative process and leading to a diminished sense of agency. Thus, Silvanto and Nagai’s model offers a neurobiological explanation for the differential effects of SCR and SLR on agency observed in our study. EEG studies of self-referential processing often focused on brain regions that overlap with the default mode network, especially the medial prefrontal cortex (MPFC) [58–60]. These areas were associated with self-awareness, mind-wandering, and social cognition. The MPFC’s influence on EEG activity in AF7 and AF8 was significant due to their proximity and neural connections. Self-referential processing was commonly examined in relation to the P300 event-related potential component, which reflected the brain’s ability to distinguish between self-related and other-related information. However, there are relatively few studies investigating the EEG topological features associated with self-reflective states. Alpha oscillations were prominently linked to spontaneous self-referential mentation, such as mind-wandering and internal thought processes. They tended to be involved when the brain was at rest, particularly during introspective or self-related thinking. Also, beta and gamma oscillations played important roles in various aspects of self-referential processes [58]. As observed in our case, a statistically significant difference in relative band powers for >8 Hz was quite in line with this. We observed significant differences in topological features across various reflection groups, replicating findings of our earlier work [30]. Consistent with the previous results, self-centered reflections led to lower first- and second-order Hodge spectral entropy than the CTR group (see Fig. 7).

Entropies have been used to study the effects of meditation [61–63]. It was suggested that low complexity in the brain corresponded to a limited range of possible configurations or a narrower scope of conscious experiences [61,64]. Conversely, high-entropy states were believed to encompass a wealth of phenomenological content, indicating more diverse and abundant conscious experiences [65,66]. The third significant finding related to Hodge spectral entropies suggested that, regardless of the intervention, second-order Hodge spectral entropy was negatively correlated with TTB (see Table 5). This implied that the enhanced phenomenological content of one’s experiences made one less prone to making judgment errors. Conversely, lower Hodge spectral entropy, due to a narrower scope of consciousness resulting from self-centeredness, caused one to make more judgment errors. This study provided initial evidence that topological features could be biomarkers for different cognitive states.

The multiscale entropy results (see Fig 6) suggested that distinctions between SCR and SLR interventions were observed at lower scales in temporoparietal brain regions. This indicated that the self-modulation due to SCR and SLR interventions could be observed at more granular timescales. Conversely, in the anteriofrontal regions, modulations were observed at coarser timescales, indicating that the effects of these interventions might have operated at different timescales in various brain regions.

We acknowledge a few limitations of this study. This study was not pre-registered because at the time of study initiation, there were no clear requirements mandating pre-registration for pilot studies. The primary objective of this pilot study was to generate hypotheses for future confirmatory research. All study methods, hypotheses, and analyses were developed prior to data analysis, and we have reported them transparently to maintain the integrity and reproducibility of our findings. Additionally, given the pilot nature of this study, we did not control for multiple statistical comparisons thereby avoiding the risk of inflated type I errors, as recommended by Moore et al. [54]. As with many feasibility-oriented EEG studies, we prioritized a streamlined setup to enable rapid data collection, and thus, our study employed only four EEG channels, limiting the scope of brain areas under investigation. A longitudinal study incorporating additional sessions and a high-density EEG system could reveal clearer effects. Furthermore, while we utilized a standard intentional time-binding protocol, a more diversified set of experimental variants is warranted to delve deeper into the mechanisms underlying the sense of agency. Though the guided reflection prompts were adapted from previous studies and piloted, future studies should validate these prompts to evaluate the effects of different types of reflections on sense of agencies. The short duration of reflection tasks and reliance on predefined self-reflective exercises might not fully capture the complexity of personal reflection processes. Expanding the scope of study with personalized tasks and integrating detailed psychological and behavioral assessments would offer a more nuanced view of self-reflection. Despite these limitations, our findings provide novel evidence that self-reflection modulates both behavioral and neural markers of the sense of agency, and that topological EEG features such as Hodge spectral entropy can serve as promising biomarkers for capturing these effects.

Conclusion

In conclusion, our study examined the impact of self-centered versus selfless reflections on the implicit sense of agency, measured by the compression of perceived time between voluntary actions and sensory outcomes. We found that self-centered reflections enhanced the associative binding between actions and consequences, increasing the sense of agency, while selfless reflections had the opposite effect. The study also explored the neural processes underlying selfless and self-centered reflections.

We found statistically significant differences among the self-reflective EEG states in relative band powers and topological neural markers such as persistent amplitude, persistent entropy, and Hodge spectral entropy. Our analysis of Hodge spectral entropy features revealed a significant negative correlation with the sense of agency. This negative correlation implied that higher entropy and increased topological complexity during self-reflective states were associated with lesser judgment errors in perceiving actions or consequences. Consequently, enhanced spectral information in the selfless reflection group was linked with a lesser sense of agency. The situation was exactly the opposite for the self-centered reflection group.

Thus, this study suggests that various self-reflective practices modulated the sense of agency, and the corresponding neural processes were characterized by rich topological signatures. We suggest that topological features like Hodge spectral entropy may serve as a valuable biomarker for different self-reflective and meditative states, acknowledging that interpretations can vary based on context and cognitive tasks. Our future work aims to provide further insights into the topological characteristics of the neural processes behind such cognitive states.