Brain Responses to Real and Imagined Interpretation of Tonal Versus Atonal Music: A Study Based on Electroencephalographic Connectivity Networks

The participants were 12 healthy professional cellists with a mean age of 39.25 ± 6.56 (SD) (7 males, 5 females). All participants met the following requirements: 1) right-handed according to the Edinburgh questionnaire; 2) started conservatory studies between 5–7 years of age; 3) reached the degree of cello teacher between 18–21 years of age; 4) studied at music conservatories where there was no specific teaching of the techniques used in contemporary music; 5) were conservatory teachers and participated in chamber or orchestral ensembles; 6) their dedication to contemporary music always occurred after graduation and represented less than 10% of their repertoire; 7) had overall musical experience of more than 20 years; 8) received verbal and written information about the type of study and tasks to be performed, and 9) were provided with a document in advance to provide their consent as a precondition for their participation.

The Ethics Committee of the University of La Laguna, Tenerife, Spain, approved the protocol of the present study with the Code of Ethical Approval CEIBA2014-0098 under the ethical standards of the Declaration of Helsinki.

All cellists were given 20 minutes to train the two musical excerpts before performing the EEG recording during the interpretation task. The two excerpts were: 1) a 26-s baroque tonal excerpt from J. S. Bach's Sarabande (II Suite for solo cello); 2) “Sincro” by A. González, a transformation of the Sarabande composed by one of the authors with characteristics of contemporary atonal music and identical duration as the tonal excerpt (see scores in Figure 1). The Bach's Sarabande covers the first eight bars, which correspond in formal analysis to the first section (c. 1–8). The Sarabande begins with a unison D on a double string to show the desire for polyphony; it continues bifurcating both D, the upper D advances melodically with a trill that appears above E to create expectation towards its resolution in such a way that from the D and the E, it leads us to F, while the lower D behaves like a bass, establishing the basic functions of tonic-dominant-tonic. A tonic chord continues at the fall of the second measure, which is rhythmically emphasized by a dotted quarter note that favors the assimilation of the speech. The sensitive C♯ of the first section appears right at the end in a trill modulating in a diminished 7th interval from the B♭. The trill on C♯ in measure 4 marks a significant point of arrival to the dominant function that closes the first period. The trill appears to highlight the instability of the tonal function, with a resolution need; that is, on the dynamic function of the dominant before its resolution. As we have observed, the tonal structure is present in each of the notes of the excerpt that are conducted in a speech of expectation resolution, a typical characteristic of the tonal structure. The atonal “Sincro” begins with two D notes—one branching into free glissandi on the G string and the other with left-hand pizzicato—then continues with microtones in descending tone intervals to advance with a ponticello glissandi. The excerpt again intersperses left-hand pizzicatos to end with a downward bow overpressure in the same D as at the beginning. The indication of rhythm refers to the duration of the work: 26 seconds. The performer can distribute timing as they wish, although there is a written meter as a guide for its interpretation. There is no tonal structure, but there is a structure based on timbre that plays a fundamental role in its development, incorporating sound effects such as pizzicatos and overpressure. The properties and acoustic differences of both excerpts are reported by González et al. (2021). Our interest in choosing the interpretation of baroque music for the study lies in the fact that baroque is a style of tonal music that is highly trained in conservatories and was performed profusely in concert halls by the selected musicians. Therefore, this exposure has produced a high degree of cerebral plasticity of this style in the brain of the expert, in contrast to the interpretation of the proposed atonal music. In the introduction, we provided data and citations that report the correlation between brain plasticity and training; data that justify in some way the selection of musical excerpts made according to our objectives.

Each participant received the two musical excerpts to be interpreted one month in advance so that they could prepare their training and memorization.

The acquisition of EEG signals—in the interpreting and resting conditions—was performed inside an electrically isolated room. With the EEG electrodes cap on, participants remained seated with their eyes closed and covered by a sleep mask and they could hear the onset instructions of each condition through nearby loudspeakers (Figure 2A, 2B). EEG acquisition, preprocessing procedures, and graphical analysis mirror our previous article (González et al., 2021) and are briefly described in the following paragraphs.

A Nihon Kohden P electroencephalograph equipped with EEG recording software was used. EEG recordings were performed with a 19-electrode cap in the following positions: two frontal Fp1-2, two dorsal frontal F3-4, two ventral frontal F7-8, two temporal T3-4, two posterior temporal T5-6, two post-central parietal C3-4, two posterior parietal P3-4, two occipital O1-2, one interhemispheric frontal Fz, one post-central parietal Cz, and one posterior parietal Pz. Monopolar recordings were made following the standard 10-20 EEG recording system and initially referenced (online) to the mean of the reference electrodes (A1, A2, placed on the mastoids), and re-referenced (offline by program) with the reference at infinity and the potential at zero (see González et al., 2021). Signals were filtered online with a high-frequency cutoff at 80 Hz, a low cutoff at 0.05 Hz, a notch filter at 50 Hz, and sampled at 500 Hz. The impedance of the electrodes was controlled to remain within the 3–5 k-Ohm range. Two additional signals—ECG and abdominal respiratory movements—were recorded for subsequent artifact detection and visual episode selection.

In our study, we used a successive block experimental design. The whole experiment was performed in a single session in which the cellist had to start the real or imagined interpretation (Real-INT or Imag-INT) of the excerpt (tonal or atonal) when they received the command to do so (cue word) through the loudspeakers and went to rest after 26 s when they received the corresponding instruction. The process was repeated six times and the order of the excerpts to be interpreted was altered to avoid habituation according to the block sequence (resting-task) as follows: resting-atonal-resting-tonal-resting-tonal-resting-atonal-resting-tonal-resting-atonal-resting-atonal-resting-tonal-resting-atonal-resting-tonal-resting-tonal-resting-atonal. All digitized recordings were stored in the computer for further preprocessing and analysis. The recordings were made with the subject seated in a chair, holding the cello (including during Imag-INT), with eyes closed, and wearing a sleep mask in an acoustically and electrically isolated room (Faraday cage) with the lights off (Figure 2C, 2D). COGENT software in MATLAB language was used to successively present through the loudspeakers the command to start the interpretation (or go to rest) in parallel with the EEG recording. Before starting the experiment, the cellists were told to make the interpretation movements (during Real-INT) as smooth as they could to avoid sudden head movements and the consequent production of EEG artifacts. For the same reason, during Imag-INT they were asked to avoid blinking, swallowing, fidgeting, or muscle tensing. When EEG data from one session (10.4 m) were inadequate or excessively artifactual, the cellist was asked to repeat the session for a period of time thereafter.

Approximately 30 EEG episodes of 5 s after the 6 repetitions of each interpretation (tonal or atonal) and 60 episodes after the 12 resting conditions were available for each participant. EEG episodes corresponding to a subject and a condition (tonal, atonal, or resting) were first selected on screen by visual inspection. Those contaminated by artifacts were discarded, using the simultaneous ECG and respiratory signal as an aid. Next, using a MATLAB script, the remaining episodes were detrended and subsequently normalized to zero mean and unit variance. They were then sorted according to their level of stationarity, as described in a previous article (González et al., 2021). By this procedure, the twenty most stationary of all those available for each condition and cellist were selected. Next, and before the estimation of the interdependence between EEG channel pairs, the selected EEG episode were filtered using a zero-phase distortion finite impulse response (FIR) filter (filter order: 256) in the following five frequency bands (FBs): delta- δ (1-4 Hz), theta-θ (4-8 Hz), alpha- α (8-13 Hz), beta-β (14-30 Hz), and gamma-γ (30-48 Hz).

A phase synchronization (PS) index called phase-locking value (PLV) was used to estimate the functional connectivity (FC) between the 19 electrode-channels pairs for each FB. PLV seems to be sensitive to changes in the activity of deep sources and to the interdependence between them (Pereda et al., 2018) and, it has also been contrasted against other PS indices to assess possible changes in the topological organization of brain networks, using graph metrics in studying dysconnectivity in neurological disorders (Olejarczyk & Jernajczyk, 2017). The calculation of the PLV between two real-valued noisy signals (x, y) requires mathematical operations such as filtering (in the required EEG frequency band) and a Hilbert transform to obtain the phase signals of both x and y and the relative phase between them (PSxy). From these and after further phase manipulations the PLV is obtained (see operations in González et al., 2021). PLV is an undirected PS measure, therefore symmetric [PLV(x,y) = PLV(y,x)]. We use the MATLAB script for PLV calculation found in the HERMES toolbox (hermes.ctb.upm.es/). Two-second windows with 50% overlapping for each 5-second trial were used to obtain PLV. PLV values range from 0 (indicating the absence of PS, or a uniformly distributed PSxy) to 1 (indicating the existence of a total PS or that PSxy is constant).

The PS between two EEG signals may be contaminated by brain sources of noise common to both (such as the well-known volume conduction) or by properties of the signals that do not influence the possible statistical relationships between them. Therefore, checking the statistical significance of the interdependence (PS) provided by the PLV—as an index of FC—employing some test is considered necessary. Moreover, as we will comment later, the parameters of one graph are usually calculated using only a percentage of the set of inter-electrodes connectivities; that is, only PLVs that exceed a predetermined connectivity threshold are considered in the calculations. In this work, we use a surrogate data test to verify the real interdependence (PS) given by the PLV. First, for each EEG signal, 100 surrogate versions were obtained. Each surrogate retains all the individual properties of the original (amplitude distribution, power spectrum) but is independent by construction. The surrogate signals were generated using the twin surrogate algorithm (Romano et al., 2009; Thiel et al., 2007, 2008). The latter has been considered useful when some signals may have nonlinear features, which cannot be ruled out in the case of EEG data (Pereda et al., 2001). The thresholding operation is as follows: first, the PLV(k, j) index between two original EEG signals (xk - xj) from channels k and j is obtained and then 99 PLV indices are calculated between the original xk and each of the 99 surrogates (xjs) of the original xj [PLV(k, js): s = 1,.…99]. From there, the empirical distribution of PLV values is obtained under the null hypothesis of no PS. Finally, the original value of the index PLV (k, j) is considered significant, at the p < .01 level, if PLV(k, j) > PLV(k, js) for any s; when this does not occur, PLV(k, j) is set to zero. The previous operations are Repeated for the PLV of all EEG episodes of each situation and each FB and participant. For each participant, and each PLV (k, j) in a given FB, the average PLV of the set of trials (EEG episodes) selected in a condition (tonal, atonal, or resting) was computed. Therefore, for each cellist and FB, we had 19 x 18 (342) PLVs that were halved (171) given the symmetrical nature of the PLV.

The 19 electrodes/channels and the 171 connections/links between them constitute respectively the nodes and edges of the graph representing the EEG neural network in a given FB. Arranging the 19 electrodes in matrix form, we obtain the 19 x 19 adjacency matrix of the graph (A); in our context, the EEG connectivity matrix, whose elements a(i,j) constitute the functional connectivity values (connectivity weights) between all channel/signal pairs, satisfying: a(i,j) = a(j,i) and diagonal elements a(i,i) = a(j,j) =1 (i,j = 1,..,19). Two central topological indices of each EEG network were obtained for each node and then averaged for all nodes: node degree (DEG) as the number n of connections (a(i, j) ≠ 0) reaching a node and the node strength or intensity (STR) equal to the average connectivity of these connections (Σ a(i,j)/n, (i,j = 1,..,19)]). The topological inter-node organization of one EEG network/graph g was assessed through two measures using the A matrix.

The first measure is global efficiency (GE). GE(g) is a measure of the ability of the graph g to interconnect and transmit information between distant nodes and is defined as the average of the inverse of the shortest path length from each node to all other nodes [N nodes is calculated as GE(g) = 1 / N(N-1) * Σ 1 / d(i,j), (i,j = 1,..,19) where d(i,j) is the shortest path length between node i and node j in the graph g and is calculated as the smallest sum of edge/connections lengths throughout all possible paths from node i and node j (Achard & Bullmore, 2007; Latora & Marchiori, 2001)]. The length of a connection was considered as the reciprocal of their connectivity weight d(i,j) = 1/a(i,j), under the assumption that the distance between two nodes is inversely proportional to their connectivity.

The second measure is local efficiency (LE), which for a node i is defined as the GE of the node and calculated in the subgraph g(i) created by its neighbors. The LE of the graph (g) is the average of the LEs of all nodes. It is calculated by applying the same steps in the subgraph g(i) formed by the neighbors of node i. LE(g) = 1/N * Σ GE(g(i)) (i = 1,..,19). From the LE and GE measurements, a regular or lattice network is defined as one with high LE and low GE, a random network as one with low LE and high GE, and a small-world (SW) network would lie somewhere between a regular and a random network, with high LE and GE. To test whether a real network has SW structure, one of the proposed methods is to normalize the LE and GE in the real network, relative to those computed in matched random networks, i.e., by dividing the LE and GE by the corresponding mean [LE(r) or GE(r)] obtained from 100 random networks that retained the same number of nodes, connections, and degree distributions as the real brain networks (Liu et al., 2016; Ma et al., 2018; Maslov & Sneppen, 2002). Thus, a SW network would be characterized by having LE > LE(r), and a comparable GE ∼ GE(r). If we consider the normalized versions of NLE = LE/ LE(r) and NGE = GE/ GE(r), a network will have a SW structure if NLE>1 and NGE∼1. Young adult brain neural network structure has been considered as a paradigm of SW-like architecture, with functional connectivity consisting of clusters of brain regions with strong local connections to each other, along with some global connections between these clusters in the form of neural hubs (Meunier et al., 2010; Wu et al., 2013).

LE/GE parameters of brain networks structure may be influenced by the magnitude of the degree of the network and thus, in comparisons between groups/conditions, differences that may be found in LE/GE values could be biased by differences in degree. For this reason, it has been suggested to compute those parameters with different node degree thresholds and investigate for possible differences in network topology (Zalesky et al., 2010). In this study, LE and GE have been calculated using the connectivity matrix A resulting from applying the surrogate data test to the PLV (i,j) values between nodes which corrects or thresholds—as we have seen—the degree of the network according to the statistical significance of these PS values. From above, it seems necessary to consider changes in the degree of the graph when comparing different groups/situations as a function of alterations in the topological properties of the LE or GE graph. The MATLAB Brain Connectivity Toolbox (brain-connectivity-toolbox.net) was used to calculate the above graph indices from graph theory (see comments from Rubinov & Sporns, 2010).

We used the MANOVA test for repeated measures to test for statistical differences between the main and repeated factors (or significant interactions between them) in the global parameters of the graph. The Bonferroni test for post hoc comparisons was used. Here the repeated factors in each participant were: the 2 interpretations (INT) (Real-INT, Imag-INT), the 3 conditions (TAR) (tonal, atonal, and resting), and the 5 EEG frequency bands (FB). STATISTICA software was used for MANOVA statistical comparisons and graphical plotting of the results.

As a complement to the MANOVA test, a permutation statistical analysis of the nodal graph indices was performed. A permutation test is a method that adjusts p values in a way that controls for family-wise error rate (FWER) and was used here for comparisons between the nodal indices (DEG, STR, and NLE) of two graphs, and to construct topographical maps of their statistical significance for different conditions contrasts. Permutation tests were extracted via MATLAB. The proportions of statistically significant nodes between the different musical style conditions in each interpretation mode were compared using McNemar's nonparametric test. A modified version of MATLAB's “topoplot” function was used to map onto each graph the level of statistical significance of the different graph indices computed at each of its 19 nodes.

Another method of analysis that we used in this work was the NBS (Network Based Statistic) (Zalesky et al., 2010). This is a nonparametric statistical procedure based on the permutation of the compared groups or conditions (e.g., by t-test) in which the FWER correction is used to solve the problem of multiple comparisons in a graph/network to select those subnetworks formed by connections whose weights (in our case, the magnitude of the estimated functional connectivity) are significantly different between groups or conditions, regardless of whether they are strong or weak (De Vico Fallani et al., 2014). This method makes use of the original 19 x 19 matrix of connectivities of each participant (without thresholding) of each group/condition. The NBS toolbox manual (nitrc.org/projects/nbs/) instructs the analysis steps to be performed and briefly, they refer to: the preparation of the two matrices to be compared according to whether they correspond to paired or independent conditions/groups; the choice of the t-test (paired or unpaired) and its threshold value; the choice of the number of permutations to perform; the election of: 1) the component of the graph to measure (its extent, i.e., the number of connections or, its strength/intensity, i.e., the sum of t-test values) and, 2) the limit value of p for the FWER correction to limit the p values obtained from the set of permutations. The NBS application finally provides us with the value of p lower than the selected limit. That will allow us to confirm or reject the existence of a sub-network whose connections have a higher or lower functional connectivity with the other group or condition compared. With these data, we can plot the connectome in the analyzed contrast and condition. In the present work, we selected paired t-test, p < .05 for FWER correction, 1000 as permutation number, and extent as graph component. The results obtained by NBS were plotted using MATLAB. Thus, a probability/statistic plot of the EEG connectome subnet at each FB was obtained. Here we also made use of McNemar's nonparametric test to compare the percentage of statistically significant connections between subnet pairs of different FB during Real-INT or Imag-INT in each of the contrasts considered or between them.

The MANOVA test showed that the interpretation factor (INT) (Real-INT vs. Imag-INT) significantly affected EEG graph global parameters—strength (STR), F(1, 11) = 22.46, p = .000; degree (DEG), F(1, 11) = 23.76, p = .000; normalized local efficiency (NLE), F(1, 11) = 47.02, p = .000—but did not influence normalized global efficiency (NGE). STR and DEG were of lower magnitude during Imag-INT than during Real-INT (p < .001); the opposite was true for NLE (p < .001). These effects occurred independently of the repeated factor TAR (tonal, atonal, resting) and of the FBs (delta-δ, theta-θ, alpha-α, beta-β, gamma-γ). In addition, MANOVA results showed a significant INT*TAR interaction for STR, F(2, 10) = 4.42, p = .043; DEG, F(2, 10) = 8.75, p = .006; NLE, F(2, 10) = 11.23, p = .002. Paired post hoc tests showed that: 1) STR and DEG magnitudes were significantly greater during Real-INT than during Imag-INT but only for the tonal and atonal conditions, the opposite occurred for NLE but only for the atonal condition; 2) during Real-INT, paired comparisons between tonal, atonal, and resting showed significant results for STR with resting > tonal and resting > atonal, but not for DEG and NLE; 3) during Imag-INT, STR, and DEG magnitudes were greater in resting than during tonal or atonal conditions and the opposite for NLE; 4) also during Imag-INT, tonal vs. atonal differences were significant for NLE with tonal < atonal. The above INT*TAR results are summarized in Figure 3, where significant p levels for paired statistical comparisons are indicated with asterisks. Finally, MANOVA results showed no significant INT*FB interaction nor INT*TAR*FB, which means that the interactions or differences between the TAR conditions with the 2 types of INT interpretation are not altered or are similar—in statistical terms—in the different frequency bands (FB).

Later we will see that statistical analyses of multiple comparisons using permutations and those based on network-based statistics (NBS) reveal in tonal vs. resting and atonal vs. resting contrasts alterations in FB-dependent connectivity measures during both Real-INT and Imag-INT.

Figures 4 to 6 show the graphs as head maps—showing the 19 nodes considered—for the STR, DEG, and NLE parameters, respectively. They show (in colors) the level of statistical significance p and the magnitude of the contrast reached at each of the cortical nodes for tonal-resting and atonal-resting contrasts, during Real-INT and Imag-INT interpretations. The maps have been plotted for each of the EEG graphs corresponding to the FBs considered. The color bars indicate the parameters’ level of significance (horizontal color bar at the bottom) and the magnitude of the contrast (vertical color bar on the right side) at each of the nodes in the condition considered. Figures 4 to 6 depict the graph parameters at the nodal/regional level under each condition. In order to avoid creating two figures for each parameter of the selected graphs (one to express the magnitude of the contrast and another to express its level of statistical significance), we utilized the same color scale for both properties of the contrast, as they are closely proportional. Consequently, in these figures, the filled color of each node in each subplot simultaneously represents, on one hand, its level of statistical significance, with the value that can be extracted from the color scale of the horizontal bar at the bottom, and on the other hand, the magnitude of its contrast (positive or negative) with the value that can be deduced from the color scale of the vertical bar on the right. In the following text, for the sake of clarity, only nodes with a p < .005 level in each condition and contrast are noted in square brackets.

First, significant nodes for STR (Figure 4) at the different FBs considered were:

• During Real-INT: a) tonal-resting contrast: delta [T5, T6, O2, Fz, Cz]; theta [Fp1, Fp2, F7, F8, Fz, Cz, T3, T4, T5, T6, O1, O2]; alpha [F7, Cz, T3, T4, T5, T6, P4, O1; O2]; beta [Fp2, Fz, Cz, T3, T4, T5, T6, O1]; gamma [T3, O1, T4, Cz]; b) atonal-resting contrast: delta [Fz]; theta [T5, O1, T6, O2, Fz, Cz]; beta [T4]. During the Real-INT condition and tonal-resting contrast, the significant nodes percentage in the theta graph was significantly greater (p < .05) than in the delta and gamma graphs, while during atonal-resting contrast this percentage in the theta graph was greater (p < .05) than that in the delta, alpha, and gamma graphs. In addition, during Real-INT, the node percentage in the tonal-resting contrast was greater (p < .01) than in the atonal-resting contrast for the alpha graph only.

• During Imag-INT: a) tonal-resting contrast: [-]; b) atonal-resting contrast: delta [all nodes]; theta [all nodes except O1]; alpha [T3, P3, T5, O1, P4, Pz]; beta [Fp1, F3, T5, O1, P4, T6, Fz]; gamma [Fp1, Fp2, T6, Fz, Cz].

• During Imag-INT there was no difference in the percentage of altered nodes between the different pairs of graphs in either the tonal-resting or atonal-resting contrast; the differences between tonal-resting vs. atonal-resting were evident for all graphs.

Second, significant nodes for DEG (Figure 5) at the different FBs considered were:

• During Real-INT: a) tonal-resting contrast: theta [Cz]; b) atonal-resting contrast: [-]. During Imagined-INT: a) tonal-resting contrast: [-]; b) atonal-resting contrast: delta [Fp1, F3, F7, T3, P3, T5, O1, F4, F8, C4, T4, P4, T6, O2, Pz]; theta [F7, T3, P3, T5, Fp2, F8, C4, T4, T6, O2]; gamma [P3, T6.].

• For the DEG index and during Real-INT, there were no differences in the percentage of altered nodes between the different pairs of graphs, neither in the tonal-resting contrast nor in the atonal-resting contrast. Furthermore, no differences were found between tonal-resting vs. atonal-resting for any of the graphs.

• During Imag-INT and in the tonal-resting contrast there were no differences in the percentage of altered nodes between the different pairs of graphs while in the atonal-resting condition. The percentage of altered nodes in the delta graph was greater (p < .01) than in the alpha, beta, and gamma graphs and the same occurred with the theta graph. The differences between tonal-resting vs. atonal-resting were evident (p < .01) in the delta and theta graphs with the percentage being greater in the latter condition versus the former.

Third, significant nodes for NLE (Figure 6) at the different FBs considered were:

• For the NLE index (Figure 6) only node alterations were found during the Imag-INT and in the atonal-resting contrast. These occurred in the delta [Fp1, F3, F7, F8, T3, T5, T6, Pz, and O2], theta [Fz, Cz.T5, F8, T4, O2], and gamma [T5] graphs. The results regarding the percentage of significantly altered nodes were similar to those obtained for the DEG.

For each interpretation mode (Real-INT, Imag-INT) and contrast (tonal<resting, atonal<resting), the subnets that were significant (FDR < .05) in each EEG FB graph are shown in Figure 7. All the links/connections and nodes involved in each subnet have been plotted in each head graph. The links and nodes appear with different colors according to their t-test (TT) contrast level and degree (DEG) magnitudes (see Figure 7 legend). Below, for each indicated condition, the number (N) of links or nodes of each subnet are pointed out between square brackets. For clarity, only the five links having a strength (TT) greater than 4, and the nodes with a DEG greater than 6 have been expressly noted in the text below with its t-test level TT or DEG magnitude appearing in parenthesis.

• Real-INT and tonal < resting contrast.Links: delta [N = 13, T6Fz(5.4), FzPz(5.3), O2Cz(4.7), P4Fz(4.6), O2Fz(4.5)]; theta [N = 151, O2Fz(9.1), C3Cz(8.7), FzPz(8.2), O2Cz(7.7), P3Fz(7.7)]; alpha [N = 33, F7Cz(7.0), F7Fp2(6.3), Fp1F7(5.8), T3Fz(5.4), F7Fz(5.1)]; beta [N = 59, Fp2F4(8.8), Fp2T4(8.6), Fp2Cz(7.9), F4T4(7.5), F8Cz(7.3)]; gamma [N = 21, T3Cz(7.2), T3Fz(5.8), F8Cz(5.7), T4Cz(5.7), T3O1(5.1)]. Nodes: delta [N = 10, Fz(6)]; theta [N = 19, C3(18), C4(18), T3(18), Fz(18), Cz(18)]; alpha [N = 18, T3(11), O1(7), F7(7), T5(6)]; beta [N = 19, T4(18), Cz(12), T3(11), Fp2(9), Fz(9)]; gamma [N = 11, T4(7), Cz(7), T3(6)].

• Real-INT and atonal < resting contrast. Links: delta [N = 2, P4Fz(4.6), FzPz(4.3)]; theta [N = 28, C3Fz(5.3), T5O1(5.2), FzPz(5.1), O2Fz(4.9), O1Fz(4.8)]; alpha [N = 5, T5O1(5.2), T5O2(4.7), T5T6(4.2), T5Pz(4.2), P3T5(4.1)]; beta [N = 16, T3Cz(5.2), T5O1(5.0), Fp2Cz(5.0), T3Fz(4.8), T5T4(4.5)]; gamma [N = 6, T3Cz(5.3), O1T6(4.7), T3T6(4.7), T3Fz(4.5), T3O1(4.3)].

• Imag-INT and tonal < resting contrast. Links: delta (N = 1) [Fp2F8(4.1)]; theta (N = 2) [Fp2F8(4.1), F3F8(4.0)]. Nodes: None

• Imag-INT and atonal < resting contrast. Links: delta [N = 62, P3T6(5.0), T6Pz(4.9), F3O2(4.8), F4O2(4.8), O2Pz(4.8)]; theta [N = 64, T3C4(4.8), T3Pz(4.7), T3P4(4.6), C3T4(4.6), T3P3(4.6)]; alpha [N = 3, T5O1(4.3), P3T5(4.2), T5T6(4.0)]; beta [N = 4, Fp1Fp2(4.4), Fp2F4(4.3), F4Cz(4.1), Fp2F8(4.1)]; gamma [N = 21, T4Fz(4.8), T6Fz(4.6), FzPz(4.4), T4Cz(4.4), P3Fz(4.3)]. Nodes: delta [N = 18, T6(17), O2(14), F8(11), T4(10), Pz(10)]; theta [N = 18, P4(11), Pz(11), Fp2(10), C4(10), P3(10)]; gamma [N = 16, Fz(11)].

The results concerning the percentage of significant connections in the subnets during Real-INT in the tonal<resting contrast showed that in the theta subnetwork, this was higher (p < .01) than in the other FBs; in the atonal<resting contrast also the percentage was higher (p < .01) in the theta subnet than in the other FBs except for the beta subnet. In the alpha and theta subnets the percentage of affected connections was much higher (p < .01) in the tonal < Rest contrast than in the atonal<Rest contrast; in the other delta, beta, and gamma subnets this result also occurred but in lower proportion and significance (p < .05). During the Imag-INT condition, in the tonal<resting contrast the percentage did not differ between the subnets of the different FBs, but in the atonal<resting contrast the percentage was higher (p < .01) in the delta subnet than in the alpha and beta subnets. The latter result also occurred with the theta subnet. The percentage was lesser (p < .01) in the tonal<resting than in the atonal<resting contrast but only for subnets delta, theta, and gamma.

The global results concerning the parameters measured in the representative graph of our neural network model constructed from the EEG connectivity measures in each EEG frequency band (FBs) indicate that, regardless of the FBs, during the real interpretation (Real-INT) of tonal and atonal musical excerpts there is a decrease in network strength (STR) compared to resting (Figure3A). According to our functional connectivity measure, this means a global desynchronization compared to rest (dysconnectivity) during the Real-INT of both excerpts. This effect occurs without a parallel decrease in the graph/network degree (DEG, the mean number of node connections) (Figure 3B).

Although at the global level, we found no differences between both music styles and different FBs, when we analyzed the results of the parameters at the nodal level in the maps of EEG graphs and connectomes in the different FBs, the situation was different. Thus, the STR graphs/networks (Figure 4) in the tonal-resting contrast show that dysconnectivity during Real-INT affects a high proportion of nodes, especially visible in the network-theta and also in the alpha and beta networks; the most significantly affected nodes correspond to the dorsal areas close to the cortical motor areas and the ventral cortical ring circumscribed by frontal, temporal, and occipital nodes. In the atonal-resting contrast, the number of affected nodes and therefore the dysconnectivity is also greater in the theta-network than in the other networks; however, the proportion of affected nodes in atonal-resting is less than in tonal-resting but only for the alpha-network in which the frontal nodes of the ventral ring do not appear desynchronized. On the other hand, for both musical excerpts, the STR changes described above are independent of the number of connections per node (DEG) of each FBs network, except for the central area/node near the motor cortex of the theta network (Figure 5).

The EEG connectome results obtained from the NBS analysis (Figure 7) confirm, in general terms, those obtained from the STR node maps in the graph analysis (Figure 4). Indeed, the dysconnectivity differs according to its FB and is globally greater in tonal<resting than in atonal<resting contrasts. In addition, in both contrasts, dysconnectivity is especially prominent in the network-theta and the most notable difference between them occurs in the network-alpha. But in these maps (Figure 7) we can also infer the distribution and cortical location of the most affected connections in both contrasts, a circumstance that is not possible to infer from the nodal STR graphs (Figure 4). Connectome results show that the distribution and location of the most affected (Figure 7) connections in the contrasts are different according to the FB of the network considered.

A first logical reflection about the origin of the dysconnectivity reflected in the global EEG connectivity would be to attribute it to the cognitive activities, such as memory and decision-making, and to the motor and sensory activities necessary for the Real-INT of the two musical excerpts regardless of musical style. These functions as described in previous research (Criscuolo, et al., 2019; Herholz & Zatorre, 2012; Schlaug, 2015) would involve multiple neural networks. The functional connectivities between them give rise to the desynchronization found concerning the resting situation (dysconnectivity) in which brain neural activity and the level of connectivity and synchronization between its networks would logically be very different. In general, functional connectivity in the resting state is considered to play an important role in brain functions and may be altered in certain psychological pathologies (for review, see Woodward & Cascio, 2015). In schizophrenia patients, the resting functional connectivity decreases (greater dysconnectivity) concerning healthy subjects (Zalesky et al., 2010), although the direction of change in other psychological dysfunctions is unclear (Woodward & Cascio, 2015). Specifically, regarding synchronization, the controversy on the role of the underlying mechanisms during rest and how this synchronization decreases in a cognitive task has been highlighted (Deco et al., 2011) and in line with our results. On the other hand, our group has reported in a recent study on neuroaesthetics (using graphs and NBS), that during resting cognitive tasks (audio-visualization of a ballet video clip) lesser nodal synchronization/connectivity was obtained compared to control conditions (visualizing a blue screen) (González et al., 2023). Therefore, the desynchronization/disconnectivity we found seems to be in line with the latter work.

The contrast results of the graphs and connectomes show that the regional alterations of the disconnectivities in nodes and subnetworks depend on the frequency of the EEG network considered. Thus, in delta and theta networks, mainly dorsal anteroposterior connections are affected (prefrontal-central-parietal-occipital of the dorsal route), while in the alpha beta and gamma networks, the dysconnectivity primarily affects dorsal-ventral transverse connections, some interhemispheric. Among the latter, it is worth mentioning the dysconnectivity between the left temporal area and the precentral-central areas close to the motor cortex, clearly visible in tonal<resting in the gamma-network and in atonal<resting in the beta and gamma network (Figure 7). The fact that in both tonal<resting and atonal<resting contrasts the dysconnectivity is especially noticeable in network-theta connectome could be related to the fact that oscillations-theta are associated with brain functionalities such as working memory (Tesche & Karhu, 2000), episodic memory (Clouter et al., 2017; Lega et al., 2012), arousal level (Green & Arduini, 1954), or sensorimotor processing (Vanderwolf, 1969), which are sensory and cognitive activities clearly implicated during Real-INT of both tonal- and atonal-styles. However, the mechanism by which the theta oscillations between different cortical regions are desynchronized during such brain activities in real-INT cannot be inferred from our data. It is possible to venture that this is because the theta oscillations of the different cortical regions/electrodes are not equally enhanced/activated during Real-INT, which would imply somehow a dysconnectivity between certain connections—in our case between mainly dorsal anterior-posterior regions.

The major difference between tonal<resting and atonal<resting in both connectome and graph analysis was observed in the alpha-network. Indeed, in this network while in tonal<resting contrast, the dysconnectivity subnetwork affects connections between anterior nodes (frontal and temporal) and to connections between posterior nodes (temporal, parietal, occipital); however, during atonal<resting, the dysconnectivity only affects posterior parieto-occipital connections. This result could be explained by the fact that alpha waves are altered during certain activities performed by the musician during Real-INT, which are different according to the style of music played. Thus, for example, EEG-activity/power is known to attenuate or disappear during the interpretation of specific tasks that require high concentration (Moini & Piran, 2020) and disappear from the precentral motor area (the area where mirror neurons are located) in hand movements (Hobson et al., 2016). EEG-activity/power has also been reported to alter during body movement (Fujioka et al., 2009; Pavlidou et al., 2014), motor control of certain movements (Pollok et al., 2009), music processing, and music training (Doelling & Poeppel, 2015). The expert cellist’s musical interpretation investigated in this work is subordinated to the motor, cognitive, processing, and music training activities mentioned above. These functions have a different degree of involvement in tonal (more trained) compared to atonal (less trained) performance, which could justify the differences found in the alpha network.

Moreover, the extent of dysconnectivity in the alpha band during the Real-INT is greater in tonal than during atonal. This effect could be justified by the time spent by the musician for learning and training and, consequently, for the acquisition of automatism and motor control of limbs during the interpretation, which is greater for tonal than atonal styles (at least in the Western musical context) in line with the results reported above concerning alpha activity. The alpha connectome result is also observed for the theta band, although involving different connections (Figure 7).

Regarding the regions altered during Real-INT, it is known, from EEG or fMRI neuroimaging literature, that during Real-INT of different musical instruments, cortical areas of fronto-parietal network and especially motor areas (primary and supplementary) are involved/activated (González et al., 2020; Kristeva et al., 2003; Meister et al., 2004; Nirkko et al., 2000; Segado et al., 2018). Our EEG connectivity network maps (Figures 4, 5, 6, 7) reflect during Real-INT the participation of nodes and subnetworks located in precentral/central frontal regions close to the motor cortex. But also, temporal areas (close to the auditory cortex) and posterior temporo-occipital areas, (close to the sensory and visual cortex) whose extension, as we have pointed out, depends on the style of music played and the EEG-FB considered. On the other hand, and from a functional connectivity perspective, during the cellist’s Real-INT, it has been pointed out that an auditory-motor circuit coactivation appears (González et al., 2020). Moreover, this auditory-motor connectivity seems to be dependent on music training and also manifests itself during real listening (Lotze et al., 2003). Our EEG connectome results confirm the existence of such connectivity also during Real-INT of both styles, visible mainly in the beta and gamma subnetworks of the left hemisphere. In addition, in Real-INT other connections seem to be desynchronized (vs. resting), including certain dorsal fronto-parieto-occipital connections, clearly visible in the theta-network (to a lesser extent in the delta); also, some anterior and posterior ventro-dorsal connections observable in the alpha-network show desynchronization. In all of them, as we have pointed out, the dysconnectivity is of greater extent in the Real-INT of tonal compared to atonal.

It should be noted that increased training in music reading—as occurs in skilled musicians—improves the connectivity of the ventro-occipital-temporal visual cortex with more distant regions (Bouhali et al., 2020). This result could partially explain the changes already noted in the connectivity of these posterior areas, mainly in the alpha and theta networks during Real-INT of both styles. Regarding the tonal and atonal differences during the Real-INT, a greater presence of connectivity seeds in precentral motor cortical areas, central somatosensory, and auditory areas has been reported (González et al., 2021) during tonal than in any of these areas during the Real-INT of atonal. These results could explain the smaller extent of dysconnectivity observed in Real-INT of atonal in some of the above-mentioned EEG connectomes, mainly in the alpha and theta bands.

Despite the networks alterations caused by dysconnectivities, the latter did not modify (compared to the rest) the topological SW structure of the network. Our results concerning the topology of the EEG-CNs showed that during Real-INT neither the values of the normalized local information transfer efficiency (NLE) nor the corresponding global efficiency (NGE) were modified in any of the conditions (atonal, tonal, resting) (NLE, NGE) (Figure 3C, 3D). Indeed, NLE and NGE magnitudes remained during real interpretation within the limits defining the topological structure of a small world (SW) network in the three conditions considered (tonal, atonal, and resting) in the different FBs. Regarding the SW topology of the EEG-CN, we found in our resting musicians an evident/clear SW topology (NLE>1 NGE ∼1); that is, with a higher local communication efficiency (LE) and global integration (GE) similar to that of a random network. According to the SW concepts reported in the Method section, high local NLE efficiency suggests a topological organization indicative of segregated neural processing, as nodes in networks with high LE tend to share information effectively within their immediate local communities, facilitating effective segregated information processing in the network. The lifelong trained musician profile seems to facilitate their SW topology of network connectivity in participants with a high level of communication between nearby local networks and global integration (GE) factors necessary to respond to the constant decision-making involved in musical interpretation.

Since SW topology depends on the brain processes of integration (NGE) and segregation (NLE), we can conclude that these are not altered when moving from rest to interpretation. This result seems to be in contradiction with Di and Biswal's (2015) suggestion that the brain is more segregated (into subnetworks or subsystems) when the motor network is intrinsically activated, which we assume occurs during real interpretation. We do not have a plausible explanation for why the SW topology of the EEG-NC network does not change from rest to real interpretation. Perhaps we could suggest that since these are expert musicians with highly trained brains throughout their lives, the changes from rest to actual performance are not reflected in the EEG-CN topology.

During Imag-INT, we observed that there is, both during tonal and during atonal (regardless of the FBs), a global dysconnectivity affecting STR values, as occurs in Real-INT but of greater magnitude than the latter. But in Imag-INT, unlike Real-INT, there is also a reduction in DEG; that is, there is a global dysconnectivity that affects both the weight of nodal connectivity and the number of connections per node (DEG). In addition, the global DEG and STR magnitudes are comparatively lower than during Real-INT, especially during atonal.

During Imag-INT, graphs’ results for STR and DEG show no alterations in the tonal-resting contrast for all nodes in each FBs network (row 3 of Figures 4 and 5). Furthermore, in the EEG connectome, there were also few altered connections in the tonal-resting contrast in the different FBs networks (row 3 of Figure 7). Global difference between tonal and resting conditions (Figure 3A, 3B) does not translate into differences (in EEG connectivity) between the different recorded areas/nodes: they all undergo tonal vs. resting similar changes (row 3 of Figures 4 and 5). On the other hand, during the Imag-INT of atonal, there was during tonal a global dysconnectivity in strength (STR) and degree (DEG) (Figure 3A, 3B). However, during Imag-INT, global changes in atonal versus rest were observed in STR, DEG that were of different magnitude in certain nodes of the delta, theta, and gamma networks (row 4 of Figures 4 and 5) as well as in some connections of the connectome subnetworks of these FBs (row 4 of Figure 7).

Dysconnectivity results during Imag-INT could be related to the greater cognitive stress—free of motor interference—during Imag-INT versus Real-INT, especially during atonal excerpts, for which there is lower cognitive plasticity for musicians as mentioned above. It thus appears that recalling and interpreting the memorized excerpt in an imaginary way produces a greater global EEG dysconnectivity in strength and degree (in all FBs networks) than during real-motor interpretation, and this effect is more pronounced in the Imag-INT of atonal. In this regard, it should be noted that during Imag-INT, functional connectivity is enhanced in certain brain areas, and/or new networks/regions are activated that do not appear in the real interpretation (Herholz et al., 2012; Hou et al., 2017; Langhiem et al., 2002; Tanaka & Kirino, 2017, 2019), which could provide some justification for our results.

On the other hand, a possible explanation of the difference between tonal-resting and atonal-resting contrasts during Imag-INT could lie in the following fact noted earlier: the structure, syntax, and interpretative technique and dynamics of the baroque excerpt’s tonal style are ingrained in the musician’s brain through training and exposure since beginning as a student. Therefore, tonal-music style is sufficiently embedded in the musician’s brain, and this plasticity effect in the neural networks could be the cause—during Imag-INT of the tonal excerpt—of the absence of differences in the connectivity parameters maps between the different cortical EEG recording areas in any of the FBs networks considered. The contemporary atonal excerpt was written following the strategies and concepts of atonal style, which are different from tonal style and are not as prevalent in Western music school education and performance, thus resulting in less cognitive and sensorial rooting in the musician's brain.

In our paradigm during the Imag-INT of atonal excerpts (atonal vs. resting), alterations of EEG connectivity measures do appear in certain EEG regions, some of them close to the brain regions in which alterations in activation/functional connectivity were reported in the fMRI imaginary investigations. In those studies, the musical excerpts and sequences interpreted did not follow the memorization and training protocol followed in our study, and they were not interested in the differences between styles. Therefore, differences between the methodologies used prevent us from comparing in detail the results of our study. However, we did observe that the Imag-INT (in our case only in atonal) produces alterations in our EEG connectivity measurements/maps in cortical areas close to the cortex region of the angular gyrus (P3, Pz, P4) and to the central motor/premotor areas (Fz, Cz). Therefore, there is a certain degree of coincidence with the results found with the fMRI connectivity indices during Imag-INT in these regions (Tanaka & Kirino, 2017, 2019).

Unlike Real-INT, in Imag-INT there is an increase in the local efficiency (normalized) of information transfer (NLE) in tonal and atonal (vs. resting) and of greater magnitude in atonal than in tonal. This effect occurs without changing the global efficiency (NGE) (Figure 3C, 3D). Some existing work on this issue deals with the impairment of SW topology in psychiatric disorders and the transition from young adults to older people (see review by Liao et al., 2017). We only found precedents in this field in our work on music listening (tonal, atonal, and noise music) in musicians and nonmusicians with eyes closed at rest (a situation resembling that of our interpreters at rest) (González et al., 2021). Using the same methodology as in the present work, we found clear differences between both groups in the SW topology of the network, essentially due to important alterations in the NLE (more pronounced than in the NGE), mainly in the delta and theta EEG-CN networks. Although these results are not comparable to those obtained in the present study, we do observe that the parameters (especially the NLE) defining the topology of the SW network allow us to discriminate—as in our study—the perception of different musical structures. Indeed, we found that the (atonal) NLE was different from the (tonal) NLE in the theta network of nonmusicians (González et al., 2021). Also, in a recent neuro-aesthetics study with a similar methodology, we found a higher magnitude of the NLE during the control condition (viewing a blue screen at rest without sound) than during the cognitive task (audio-visualization at rest of a movie clip) (González et al., 2023). The results of the latter work, although in a different experimental paradigm, seem to be in line with our results on musical interpretation.

Our results indicate that this SW topological configuration is altered for rest (only in the NLE) with tonal and atonal interpretation but only during Imag-INT. As for why we saw this result only in Imag-INT: it is evident that during Imag-INT the cognitive effort in certain components (memory, visual perception, language processing, reasoning, etc.) is higher (or at least different) than during Real-INT. This would imply an enhancement of the SW topology of the EEG-CN according to Douw et al. (2011) and this is the case in our study where global NLE in Imag-INT is higher than NLE in Real-INT.

In summary, using our method of analysis (graph metrics and EEG connectome) contrasting real and imagined interpretation of tonal and atonal music, we have found that Imag-INT clearly discriminates the two styles of music (tonal and atonal) and more effectively than Real-INT does. Thus, for example, while at the global nodal level, Real-INT of both tonal and atonal does not alter the SW structure of the cortical EEG network. Imag-INT of both styles modifies it compared to resting and does so differently depending on whether tonal or atonal is played. This result, in our opinion, reflects the different brain plasticity of the musician in relation to styles. Moreover, the Imag-INT of atonal music—which is not usual in the standard musical repertoire—affects, among others, EEG areas close to the auditory and motor cortex. Therefore, the Imag-INT of atonal music may be of interest to deepen the kinesthetic training of the expert musician. In fact, this suggestion of training by means of Imag-INT—without specifying the style of music—has been suggested in previous studies (Theiler & Lippman, 1995; Bernardi et al., 2013).

We would like to highlight limitations to consider when extending the results found and the discussion we have carried out to other studies.

First, the interpretative differences characteristic of each musical style are intermingled with those attributable to the different training times that the expert musician dedicates to each style. This duality may be important when analyzing the differences between styles during Real-INT. However, we think that during Imag-INT, since the musician does not make use of body movement and, therefore, of dynamics and auditory-motor interpretative coupling, the differences between styles seem to depend more on the time dedicated to the training of the style considered.

Second, the musician’s movements during Real-INT, even if interpreted with some care, may interfere in some way with the EEG signal. We were careful to eliminate artifactual EEG episodes or to repeat the Real-INT experiment in cases where the EEG episodes were excessively altered.

Third, this study considers only the interpretation of excerpts in two specific styles: baroque, representative of tonal style, and contemporary, representative of atonal music. Other styles and paradigms should be investigated and contrasted when investigating the brain plasticity of musicians in different musical styles.

Fourth, when describing the alterations in the connectivity of the nodes of the graph representing the EEG connectivity network, although we sometimes refer to the cortical nomenclature close to the electrode location, we must keep in mind that the position of the electrodes may not align with the activity of the brain area below as is now commonly accepted.

Considering the limitations, it would be suitable to expand the study to more musical styles and orchestral instruments to render a more complete vision of professional musical interpretation applied to different styles. These findings have implications for the learning and development of professional musicians.

Below is a summary of the findings detailed in this paper.

During Real-INT there is a decrease in connectivity strength similar for both tonal and atonal styles; that is, a desynchronization versus rest or dysconnectivity visible globally from the averages of the graph parameters including all conditions and FBs. Also, the dysconnectivity in the different conditions and FBs can be observed through some of the graph node maps and connectome subnetwork connections.

During Real, the dysconnectivity affected multiple graph nodes and connectome connections, most notably in the theta band and similarly for tonal and atonal. However, in the alpha band, differences between styles were evident: the dysconnectivity was more extensive during tonal than during atonal.

During Real, the global topological structure of the connectivity's graphs resulted in of small-world (SW) type at rest and was not altered during Real-INT of tonal or atonal extracts.

Overall, Imag-INT resulted in a greater magnitude and extent of dysconnectivity than Real-INT, as it affected not only the strength of the network but also its degree. In addition, the SW structure was strengthened during Imag-INT versus Real-INT due to a greater normalized efficiency in local information transfer (NLE). These effects were more prominent during Imag-INT of fragment atonal than during its Real-INT. Furthermore, during Imag-INT of atonal the SW structure was different from that of tonal since NLE (atonal) > NLE (tonal).

During the Imag-INT of the tonal excerpt, the global dysconnectivity was not reflected in any of the network parameters and hardly in the connectomes. However, during the Imag-INT of excerpt atonal, changes in global dysconnectivity were visible—in the case of degree and local efficiency—in certain nodes of the delta, theta, and gamma band graphs/networks and in some subnetworks of the connectomes of the delta, theta, and gamma bands.

Both interpretation forms exhibit differences between styles that are more evident in Imag-INT than in Real-INT. In Real-INT, the differences are only perceptible in the alpha band network/connectome possibly because this EEG activity is closely related to the control of the musician's body movements, limbs, and state of concentration, which are functions with different degrees of involvement in both styles. In Imag-INT, the absence of the motor control mechanisms of the musician’s body without auditive stimuli and the musician's need for greater experience and memory of the style structure and its interpretative dynamics, make the responses to its interpretation clearly different in the two styles. While the Imag-INT of tonal style produces hardly any ostensible connectivity networks alterations those of atonal style are noticeable and clearly affect the delta, theta, and beta band networks and connectomes differently.

The authors declare that the research described in this article has been carried out without any financial or commercial implications that could potentially lead to a conflict of interest.

Author contributions are as follows—A. González: conception of the research, methods and design, and drafting of the manuscript; A.González, A.Gamundí, J.J.González: data analysis; A.González: data acquisition; J.J. González: recruitment of subjects and validation of their neurological statements; A.González, A.Gamundí, J.J.González: critical revision of the manuscript; each author read and approved the final version of the manuscript.