Twenty-four healthy volunteers (13 females), aged 21-27 years, (mean $\pm$ std $=$ 24.23 $\pm$ 1.63 years) took part to the study. They were recruited among students of the University of Bologna (Italy). Each participant had normal or corrected to normal vision and reported no medical or psychiatric illness. The study was approved by the local bioethical committee of the University of Bologna (file number 29146; year 2019) and written informed consent was obtained from all participants before the beginning of the experiment. All data were analyzed and reported anonymously.

The experiment was performed in a controlled laboratory environment. The participants underwent four consecutive experimental sessions, each lasting 5 minutes and separated by a short interval during which the subject, while sitting, could slightly move and speak. The participants comfortably seated facing a computer monitor about 50 cm far; they were instructed to perform the four sessions with eyes open, reducing at minimum eye, head and body movements. First, a resting state (R) was recorded for 5 minutes, while the participants stayed relaxed in front of a grey screen displayed on the monitor. This was used as the basal condition. The other three sessions corresponded to the three conditions of attention. The internal attention session (IntAtt) consisted in executing a mental arithmetic task throughout the 5-minute period, while a uniform grey screen was displayed on the monitor. The task was a mental serial subtraction in steps of seventeen starting from a given number. This kind of task required intentional orientation of attention to internal processing, and did not rely on any information provided externally. Note that no distracting visual input was delivered during this session since only the grey screen was continuously presented to the participant. The external attention session (ExtAtt) consisted in presenting to the participant a series of pictures on the monitor during the 5-minute period. Specifically, thirty emotionally neutral pictures extracted from the IAPS (International Affective Picture System) database [26] were used. The pictures were displayed on the monitor one after the other every 10 seconds, in a random order. No specific task was associated to pictures viewing. Therefore, during this session visual external attention of participants was captured by the pictures in a bottom-up fashion, as it may occur in a real-life sensory rich environment. The session of internal-external attention competition (IntExtAtt) consisted in the combination of the previous two conditions: during this 5-minute session, participants were required to perform the mental arithmetic task as in the IntAtt condition, while pictures were displayed on the monitor as in the ExtAtt condition. Therefore, in this session, the pictures acted as visual distractors that competed and interfered with the execution of the mental math task. This session simulated a common realistic condition such as when we are engaged in internally-oriented tasks (e.g. problem solution in the classroom, office, etc.) and we need to isolate from the sensory-rich surrounding to avoid the intrusions of task-distracting inputs.

The order of the three attentional sessions was randomized across participants. The participants were provided with the instructions for the math task only immediately before the beginning of the corresponding sessions, so they remained unaware of the task until its onset. These instructions also communicated to the participant that he/she would be required to report the final number achieved at the end of the session. The starting number for the mental serial subtraction was 2500 in the first of the two math task sessions (IntAtt or IntExtAtt); the final number reached by the participant in this session was then used as the starting point for the serial subtraction in the subsequent session. Furthermore, participants were not informed in advance of pictures presentation either in case of the ExtAtt or IntExtAtt session. The same 30 pictures were randomly presented during both the two sessions. The selected pictures had normative level of pleasure between 4.5 and 5.5 points on a 9-point scale. Neutral pictures were used to avoid the involvement of emotional factors.

As in previous studies using covert mental arithmetic [6, 7, 25], it was not possible to control the actual execution of the task during the sessions. However, all participants were aware that not complying with the received instructions would have compromised the study. Furthermore, requiring the participants to report the final reached number likely further motivated them to engage in the task. We also asked each participant at the end of the sessions if he/she actually performed the serial subtractions (otherwise, the recording would have been discarded). A confirmation of task engagement was obtained from all participants; furthermore, all participants reported that performing the task under picture presentation was more demanding.

EEG signals were recorded through a Neurowave System (Khymeia, Italy, Brainbox® EEG-1166 amplifier, Braintronics) using an elastic cap with 32 Ag/AgCl scalp electrodes (Fp1, Fp2, AF3, AF4, F7, F3, Fz, F4, F8, FC5, FC1, FC2, FC6, T7, C3, Cz, C4, T8, CP5, CP1, CP2, CP6, P7, P3, Pz, P4, P8, PO3, POz, PO4, O1, O2). The reference electrode was placed on the left earlobe and the ground electrode was located on the forehead. The right earlobe electrode was acquired too, for offline re-referencing. During each experimental session, EEG data were digitized in continuous recording mode at a sample frequency of 128 Hz and 16-bit resolution, and with the inclusion of a hardware notch filter eliminating line noise at 50 Hz. Thus, for each participant, four 5-minutes recordings were acquired. Finally, for each participant and each recording, the thirty-two EEG signals (+ the right earlobe signal) were exported in Matlab (R2019b, The MathWorks Inc., Natick MA) for further analysis. Firstly, data were re-referenced with respect to the average of the two earlobe signals and high-pass filtered at 0.75 Hz to eliminate the DC offset and slow drifts. Then, for each participant the following processing steps were applied.

2.3.5 Cortical regions of interest (ROI) - alpha and theta power computation at cortical ROI level

We selected some cortical regions of interest (ROIs) to focus both the power analysis and connectivity analysis. The selection was mostly based on a priori considerations about the brain conditions under investigation and on results of previous literature; only in one case the selection was data-driven. These choices are further commented in section Discussion. Two sets of ROIs were selected for the theta-band and alpha-band analysis, with some ROIs common to the two sets. Fig. 1 shows the ROIs used for the theta- and alpha-band analysis, in the three-dimensional cortical source space adopted for the solution of the EEG inverse problem (Section 2.3.4). The assignment of a voxel to a specific region is based on information provided by the software LORETA-KEY$\mathrm{\copyright }\mathrm{\circledR }$, which specifies the region each voxel belongs to; the Supplementary Material Section S1 provides a detailed description of how voxels of the source space were assigned to each ROI.

Fig. 1.

The regions of interest (ROIs) selected for the alpha-band analysis (Panel A) and theta-band analysis (Panel B) at the cortical level. The ROIs are depicted in the cortical source space used by the method (eLORETA) adopted for the solution of the EEG inverse problem. A realistic head model based on the MNI152 template was used, with the solution space restricted to the cortical gray matter and divided into 6239 voxels at 5 mm cubic spatial resolution. Both panels show the back view and top view of the surface of the cortical space and the view across transversal slices along the z axis. The x, y, z axes (MNI coordinates in mm) have orientation left to right (x), posterior to anterior (y) and bottom to top (z). The voxels depicted with the same color belong to the same region, as indicated by the legend: note that, for the sake of clarity, the same color was used for homologous regions in the two hemispheres (left and right) and, in case, for the medial portion too (e.g. LGCU). Some ROIs (ITG, LGCU) were analyzed in both bands, and for completeness, they are represented both in panel A and B (but with different colors).

Fig. 2 shows the scalp maps of theta power in each session, the scalp maps of theta power difference between each attentional condition and rest, and the results of the corresponding channel-wise statistical analysis. Fig. 3 displays the same information as to alpha power. According to results in Fig. 2, engagement in the mental math task (IntAtt) was associated with frontal midline theta increase. Conversely, in ExtAtt, theta power increased only at parieto-occipital electrodes. In IntExtAtt condition, i.e. when the mental task had to be performed against task-irrelevant pictures, frontal-midline theta showed a strong increase, overcoming that observed in IntAtt condition; furthermore, theta increased at posterior electrodes too. As to alpha power (Fig. 3), the ExtAtt induced a dramatic decrease of alpha activity especially at parieto-occipital sites, an effect well expected due to visual stimulation. In IntAtt condition, alpha power tended to increase at the more posterior sites, although significance was not reached at any electrode. Finally, results obtained in IntExtAtt are especially interesting: indeed, the same visual stimuli as in ExtAtt applied during the math task were associated with a much smaller alpha power decrease at posterior sites, with no statistical significance.

Fig. 2.

Scalp maps and channel-wise statistical analysis for theta power. The maps in the first and second column represent the theta power ($\mu$V${}^2$) averaged across all participants as a function of the experimental session (Rest, IntAtt, ExtAtt, IntExtAtt). The maps in the third column represent the theta power difference ($\mu$V${}^2$) between each attentional condition and the rest, averaged across all participants. Each scalp map was obtained by color coding the average theta power value at each electrode position in a 2D circular cartoon head (top view of the head, nose at the top) and using interpolation on a fine 67 $\times$ 67 grid. In the cartoon heads of the fourth column, the red markers denote the electrodes that showed statistically significant difference of theta power in the attentional condition compared to rest (P 0.05, one-tailed permutation-based t-test, P-values corrected for multiple comparisons, see procedure (a) in Section 2.3.7).

Fig. 3.

Scalp maps and channel-wise statistical analysis for alpha power. The maps in the first and second column represent the alpha power ($\mu$V${}^2$) averaged across all participants as a function of the experimental session (Rest, IntAtt, ExtAtt, IntExtAtt). The maps in the third column represent the alpha power difference ($\mu$V${}^2$) between each attentional condition and the rest, averaged across all participants. Each scalp map was obtained as described in Fig. 2. In the cartoon heads of the fourth column, the red markers denote the electrodes which showed statistically significant difference of alpha power in the attentional condition compared to rest (P 0.05, one-tailed permutation-based t-test, P-values corrected for multiple comparisons, see procedure (a) in Section 2.3.7).

A summary of the previous results is represented in Fig. 4, which displays the theta and alpha power changes within the two scalp macro-regions (anterior and posterior, see Section 2.3.3), as a function of the attentional condition. The bar plots emphasized a different pattern of theta activity at the anterior and posterior region. Anterior theta activity exhibited the trend to progressively increase across the three conditions and reached the largest value in the IntExtAtt condition, in agreement with its relation with internal attention. Posteriorly, theta increase occurred to a similar extent in the two conditions involving picture presentation (ExtAtt and IntExtAtt) and was absent in IntAtt condition. Alpha power at the posterior region confirmed a significant decreased in ExtAtt but not in IntExtAtt condition, and a tendency to increase (but with high variability) in the IntAtt condition.

Fig. 4.

Theta power and alpha power computed at the two scalp macro-regions (anterior and posterior) in each experimental condition, and normalized to the rest condition. Powers at the anterior and posterior regions were obtained via arithmetic average across frontal electrodes (Fp1, Fp2, AF3, AF4, F3, Fz, F4, FC1, FC2) and parieto-occipital electrodes (P3, Pz, P4, PO3, POz, PO4, O1, O2) respectively (see Section 2.3.3). Each bar represents the mean $\pm$ SEM across all participants. Within each bar plot, * denotes statistically significant difference between the attentional condition and rest condition (P 0.05, Bonferroni corrected for multiple comparisons, see procedure (b) in Section 2.3.7).

The comparisons between each attentional condition and rest at the level of voxels are presented in Fig. 5 as to theta power and in Fig. 6 as to alpha power. For each comparison, the voxel-by-voxel power difference is represented, together with the results of the voxel-wise statistical analysis. Table 1 indicates, for each attentional condition, which of the selected ROIs were significantly implicated in power change on the basis of the voxel-wise statistical analysis (see Supplementary Material Section S3 and Supplementary Tables 2-6 for the list of all cortical voxels significantly involved).

Fig. 5.

Voxel-wise comparison of theta power between each attentional condition and rest. For each attentional condition, the 3D top view of the cortex shows the theta power difference (($\mu$A/mm${}^2$)${}^2$) at each voxel between the attentional condition and the rest, averaged across all participants. The transversal slices across the z axis (oriented from bottom to top) display the results of the voxel-wise statistical analysis: the colored voxels correspond to significantly higher theta power in the attentional condition compared to rest (P 0.05, one-tailed permutation-based t-test corrected for multiple comparisons, see procedure (c) in Section 2.3.7), with the color scale corresponding to corrected t-values (t${}_{th}$ indicates the critical threshold at 5% probability). It is worth noticing that only the results of the one-tailed statistical analysis (testing attentional condition $>$ rest) are reported, since the test in the other direction did not provide in any significance.

Fig. 6.

Voxel-wise comparison of alpha power between each attentional condition and rest. For each attentional condition, the 3D top view of the cortex shows the alpha power difference (($\mu$A/mm${}^2$)${}^2$) at each voxel between the attentional condition and the rest, averaged across all participants. The transversal slices across the z axis (oriented from bottom to top) display the results of the voxel-wise statistical analysis: the colored voxels correspond to significantly lower alpha power in the attentional condition compared to rest (P 0.05, one-tailed permutation-based t-test corrected for multiple comparisons, see procedure (c) in Section 2.3.7), with the color scale corresponding to corrected t-values (t${}_{th}$ indicates the critical threshold at 5% probability). For the ExtAtt and IntExtAtt condition, only the results of the one-tailed statistical analysis (testing attentional condition rest) are reported, since the test in the other direction did not provide in any significance. For the IntAtt condition, one-tailed statistical analyses did not provide any significant results in either direction.

Theta power increase in ExtAtt and IntAtt was localized within a limited area of voxels in the posteromedial cortex and in the midline/left frontal cortex, respectively (Fig. 5). It is worth noticing that the cluster of statistically significant voxels in IntAtt include also the selected ACC ROI (Table 1). Furthermore, as anticipated in Section 2.3.5, the results in ExtAtt motivated the inclusion of the PCU ROI in our analysis. Indeed, the cluster exhibiting significant theta increase included also voxels belonging to this region (see Table 1 and Supplementary Table 2), and significant PCU theta increase was confirmed at the level of the entire ROI too (as shown later in Fig. 7 and discussed in Section 4.2). In IntExtAtt, significant theta increase exhibited a widespread distribution, mostly involving temporal regions (ITG too) and portion of the posteromedial cortex; a few ACC voxels were involved too. As to Alpha power (Fig. 6), the ExtAtt condition resulted in significant decrease in almost all voxels in the posterior-parietal, occipital and temporal lobes (including LGCU and ITG ROIs). In IntAtt, although alpha tended to increase at occipital regions, no significant difference was obtained at any voxel. However, the effect of internal attention on alpha power appeared evident when considering the IntExtAtt condition; at variance with ExtAtt condition, here only a small cortical cluster in the parietal cortex exhibited significant decrease.

Fig. 7.

Overall theta power (obtained by averaging across all voxels within the ROI) in each of the seven ROIs selected for the theta-band analysis, computed in each experimental condition and normalized to the rest condition. Each bar represents the mean $\pm$ SEM across all participants. Within each bar plot, * denotes statistically significant difference between the attentional condition and rest condition (P 0.05, Bonferroni corrected for multiple comparisons, see procedure (d) in Section 2.3.7).

The previous analysis is useful to obtain a picture of the power changes in each attentional condition at a fine-scale spatial resolution, and revealed that voxels in most of the a priori selected ROIs were significantly implicated. In order to assess the modifications of the overall power in each selected ROI (rather than at single voxel level), Figs. 7 and 8 display the modulation of the overall ROI power vs the attentional condition in the theta band and alpha band, respectively. A significant increase of theta activity was seen in ACC in the two conditions involving internal attention (IntAtt and IntExtAtt, Fig. 7), larger when internal attention competed with the external one, identifying the ACC as implicated in the frontal-midline theta increase observed at scalp level in these cases (see Fig. 2). The other ROIs exhibited a different pattern of modulation: theta increase was associated with conditions involving external attention (ExtAtt and IntExtAtt), especially in posterior regions (except LGCU${}_M$), where it reached statistical significance in both conditions. Analysis in the alpha band (Fig. 8) settled that ExtAtt was associated with a strong decrease in the occipital (LGCU) as well as ITG regions, while in IntExtAtt alpha power in these same ROIs did not significantly decrease or decreased only to a much lower extent (see LGCU${}_M$), confirming the tendency of alpha power to increase (although not significantly) in IntAtt.

Fig. 8.

Overall alpha power (obtained by averaging across all voxels within the ROI) in each of the seven ROIs selected for the alpha-band analysis, computed in each experimental condition and normalized to the rest condition. Each bar represents the mean $\pm$ SEM across all participants. Within each bar plot, * denotes statistically significant difference between the attentional condition and rest condition (P 0.05, Bonferroni corrected for multiple comparisons, see procedure (d) in Section 2.3.7).

Figs. 9 and 10 show the results of the GC analysis for the selected ROIs in the theta and alpha band. The colored arrows over the 3D cortical surfaces indicate the single connections that significantly modified (P 0.05 uncorrected) compared to rest, with the red and blue indicating an increase and decrease respectively. The bar plots sum up some interesting aspects, reporting the overall causal outflow from a ROI (or a set of homologue ROIs) to a set of other ROIs. The results can be summarized as follows.

Fig. 9.

Results of the spectral Granger Causality in the theta band between the selected ROIs. The arrows over the 3D cortical surfaces (top view of the cortex) indicate the connections that increased (red) or decreased (blue) in each attentional condition compared to rest. The displayed arrows correspond to significant connectivity changes at P 0.05 (uncorrected P-values, non-parametric permutation based t-test, see procedure (e) in Section 2.3.7) and the thickness of the line denotes three levels of significance, i.e. thinnest line: 0.01 P 0.05; middle line: 0.001 P 0.01; thickest line: P 0.001. The two bar plots display the overall causal influence that emerged from ACC and PCU, respectively, and targeted the set of indicated ROIs (LGCU includes the left, right and medial parts, and ITG includes the left and right part). Each bar represents the mean $\pm$ SEM across all participants. Within each bar plot, + denotes statistical comparison vs rest resulting in P 0.05 without Bonferroni correction, while * denotes statistical comparison vs rest resulting in P 0.05 with Bonferroni correction.

Fig. 10.

Results of the spectral Granger Causality in the alpha band between the selected ROIs. The arrows over the 3D cortical surfaces (top view of the cortex) indicate the connections that increased (red) or decreased (blue) in each attentional condition compared to rest. The displayed arrows correspond to connectivity changes at P 0.05 (uncorrected P-values, non-parametric permutation based t-test, see procedure (e) in Section 2.3.7) and the thickness of the line denotes three level of significance, i.e. thinnest line: 0.01 P 0.05; middle line: 0.001 P 0.01; thickest line: P-value 0.001. The bar plots display the overall causal influence, in both directions, between left and right MFG separately (MFG${}_L$ and MFG${}_R$) and total LGCU region (bilateral and medial). Each bar represents the mean $\pm$ SEM across all participants. Within each bar plot, + denotes statistical comparison vs rest resulting in P 0.05 without Bonferroni correction, while * denotes statistical comparison vs rest resulting in P 0.05 with Bonferroni correction.

Theta-Band (Fig. 9) - In ExtAtt, increased theta-band connectivity was localized posteriorly, with the PCU exerting a significantly greater influence towards the visual ROIs compared to rest; no increased influence from ACC was observed. In IntAtt, increased connectivity from ACC emerged, while the influence from PCU exerted a minor role. In IntExtAtt, a sort of summation of the previous two effects occurred, with an overall increase in connectivity among the ROIs, and with ACC and PCU exerting each an overall influence on visual areas (LGCU and ITG) to a similar extent as in IntAtt and ExtAtt respectively (see the bar-plots).

Alpha-Band (Fig. 10) - The ExtAtt was characterized by an overall dramatic decrease of causal influence, with a clear bottom-up arrangement. No relevant modifications in top-down connections emerged (see in particular MFG${}_L$, MFG${}_R\to$ LGCU). Conversely, IntAtt was associated with an increase of both bottom-up and top-down connectivity between visual areas (ITG, LGCU) and the frontal area, involving mostly the left MFG. Finally, IntExtAtt was characterized by a decrease in bottom-up connectivity, but to a much lower extent than in ExtAtt condition (see in particular the bar plot LGCU$\to$MFG${}_L$, MFG${}_R$), and still by an increase in top-down influence from the left MFG (bar plot MFG${}_L\to$LGCU).

Some interesting main effects were obtained already at the scalp level (Figs. 2, 3, 4). An original result is that theta activity was modulated differently in the anterior and posterior sites by the attentional conditions. At frontal sites theta increase was mainly associated with internal attention, a result well in line with previous studies [5, 9]; in particular, increase in scalp frontal midline theta was previously observed during mental arithmetic tasks both similar and different from the one adopted here [6, 7, 36]. This result at scalp level (together with the source-level results indicating the main involvement of anterior cingulate cortex in frontal theta increase) can represent a posterior validation that the participants were engaged in the math task execution during the two corresponding sessions. Interestingly, frontal theta increase was larger when the math was performed with the presence of visual distractors. This result matches the general observation of higher frontal theta associated with increased task demands, usually implemented by increasing memory load in working memory tasks or by increasing task complexity [4, 51]. However, we are not aware of other studies that have investigated and revealed such an effect when the task demand is modulated by the absence vs presence of external distractors rather than by the intrinsic complexity of the internal task (which remained unchanged in IntAtt and IntExtAtt conditions). Here we found that although the internal task remained unchanged, the need to increase the protection of ongoing internal processing from external sensory interference was associated with higher frontal theta; this could reflect the allocation of greater internally-oriented attentional resources or increased error monitoring and conflict resolution.

Theta activity at scalp posterior sites did not follow the same pattern as frontally. Posterior theta increased in conditions involving external attention, i.e. in association with picture presentation. Studies investigating posterior theta in relation to visual attention and visual stimulation are sparser. Most of them reported scalp occipital theta enhancement following or during visual stimulus presentation [52, 53, 54, 55], in agreement with our results. Conversely, the effect of visual attention on posterior theta is unclear and controversial results have been reported [53, 55]. Our results provided similar levels of posterior theta increase in ExtAtt (when likely more attention was allocated towards visual stimuli) and in IntExtAtt (when likely less attention was directed towards visual stimuli); a speculative hypothesis is that this posterior theta might signal basic visual processing not affected by attention.

At variance with posterior theta, posterior alpha exhibited a clear different modulation in ExtAtt condition vs IntExtAtt condition. Picture presentation alone induced a strong posterior alpha decrease, indicating engagement of the sensory system and enhanced visual processing [11]. When pictures accompanied the mental task, alpha power exhibited a much lower decrease assuming values not statistically different from the rest; this indicates a resistance against the visual interfering information. This effect may result from a push-pull interaction between two opposing mechanisms: bottom-up visual engagement reducing alpha power (as it emerged in ExtAtt) and task-oriented visual inhibition increasing alpha power. Indeed, an average tendency of alpha power to increase occurred in IntAtt condition, although not significant. While this mechanism appeared weak when performing the internal task alone (against a grey screen), its operational role emerged more clearly in the presence of the visual distractors, being able to contrast alpha reduction. In our recent paper [56], which investigated alpha power modulations in a variety of conditions, we found that performing a similar mental task while immersed in a (highly stimulating) virtual reality environment increased alpha power to the same level as in rest condition, whereas the virtual reality immersion alone induced a strong decrease. Here, we confirmed the previous result and proved that even when using simpler and less motivating visual distractors (neutral pictures on a monitor vs virtual reality scenario) the need to isolate from visual engagement induced an increase in alpha oscillations. Our conclusion appears in contrast with a recent study [57] that questions the role of alpha oscillations in inhibition of online distractions. These authors observed that alpha did not increase compared to baseline during the retention interval of a working memory task, in the presence of visual distractors maintained throughout the retention period; furthermore, strong distractors (more similar to the memorized representation) produced larger alpha power decrease than weak distractors. Based on these results, the authors suggested that alpha oscillations were ineffective for inhibition of sustained distraction in the investigated condition. A possible reconciling interpretation may take into account that alpha power during internal vs external competition results from two antagonistic mechanisms, an increase to insulate internal processing and a decrease induced by external stimulation; the former may just reduce but not overcome the latter. This effect could occur in particular when the distractors contain features similar to the internal representation and thus may have strong attention influence.

Via the analysis at source level, we first investigated the involvement of some key cortical regions in theta and alpha power modulation and then we explored modifications in the pattern of functional connections. Here, source reconstruction was not based on a high-density electrode montage but on a limited number of electrodes (32). Therefore, the results obtained on source activity and connectivity merit additional studies to obtain a more robust validation (see also Section 4.4).

For the alpha-band analysis, we took into consideration both lower-level (lingual gyrus and cuneus) and higher-level (inferior temporal gyrus) visual regions, based on their functional roles and of previous results in literature (see Section 1. Introduction and Section 2.3.5). The lingual gyrus and cuneus were considered together (LGCU) in order to reduce the number of ROIs, especially in view of the connectivity analysis, and favor a more straightforward interpretation of the results.

The whole-brain statistical analysis reveals that almost all voxels in the posterior regions are implicated in alpha modulations (Fig. 6). More specifically, alpha power modulation in the selected ROIs (Fig. 8) paralleled that observed posteriorly at scalp level. In particular, ExtAtt condition was characterized by a significant alpha power decrease in all visual ROIs; notably, the decrease was larger in earlier (LGCU) than superior (ITG) visual ROIs. No significant modulation of the alpha power in the visual ROIs was found in IntExtAtt compared to baseline. Similarly to our results, a recent study [10] reconstructing source activity from magnetoencephalography, found a marked alpha reduction, especially in occipital bilateral regions, during a prospective memory task that required elevated external attention and low internal attention. Conversely, they did not find significant changes of cortical alpha in a task that required lower external monitoring and high internal attention. Again, we interpret the absence of alpha changes in that task as resulting from the balance between the alpha increase to protect the internal representations and the alpha decrease still induced by the external sensory inputs.

Besides the visual ROIs, we also included the middle frontal gyrus in alpha band analysis, since this region, although did not exhibit a significant alpha-power change neither at the voxel-level nor at the overall ROI level, may be implicated in top-down influences in the alpha-band toward the visual ROIs, especially in conditions involving internal attention. Indeed, there is a general agreement that prioritizing or suppressing sensory processing in a goal-oriented manner are controlled by top-down signals from higher-level frontal cortical areas [58, 59], although no consensus has emerged yet on the specific sources of these signals. However, implication of the middle frontal gyrus (including the dorsolateral prefrontal cortex) in attention-dependent modulation of visual activity is supported both by fMRI and ERP studies [60, 61], and other EEG studies [24, 34, 35].

For the theta-band analysis, we included not only the visual ROIs (LGCU, ITG) but also two additional ROIs, the Anterior Cingulate Cortex (ACC) and the Precuneus (PCU). The ACC was a priori included as one of the main source of frontal midline theta associated with internally focused attention and cognitively demanding task [4, 9], including mental arithmetic [6, 7, 36]. Our results suggest that the frontal theta increase in our math task originated mainly from the ACC, and from this hub it might potentially extend to other frontal regions (see indeed right top cortical surface in Fig. 5 showing higher and broader theta increase in frontal cortex, where significant voxels were found too). Conversely, the Precuneus was selected a posteriori based on the statistical voxel-wise results in ExtAtt condition, which showed a significant involvement of this region (Fig. 5 left slice-view maps and Table 1). Accordingly, the analysis at overall ROI level (Fig. 7) showed significant power modulation in PCU, suggesting that PCU could be implicated in theta power and that activity could spread from it to the close cingulate cortex (where significant voxels were found too). However, interpreting the involvement of PCU in ExtAtt is not straightforward and our explanation remains at a highly speculative level. Indeed, the precuneus has been widely associated to internally oriented functions and memory processes (e.g. visual imagery, episodic or semantic memory retrieval) [37, 38]. A preliminary interpretation, requiring further support, can be linked to a few studies suggesting that PCU has also visual functions [39, 40] and may be implicated in general monitoring of external environment [62]. We can tentatively speculate that, at least in the conditions investigated here, theta increase in PCU and in the other visual regions (especially bilateral LGCU) might represent a basic visual processing uninfluenced by the amount of allocated attention (see the similar level attained in ExtAtt and IntExtAtt in these ROIs, Fig. 7). However, this interpretation remains uncertain and definitely requires further inquiries.

When considering internal attention alone, causal theta influences increased from ACC towards temporal and then posterior (PCU) regions (Fig. 9), in agreement with previous results showing the importance of fronto-temporal-parietal theta networks in mental arithmetic [7, 36]. Conversely, in the ExtAtt condition theta connectivity appeared much more localized and confined posteriorly. Furthermore, the overall theta connectivity increased in IntExtAtt, condition, with the two previous mechanisms operating simultaneously, and apparently not influencing reciprocally.

Mechanisms of alpha-band causal influences seemed to operate in opposite direction in external and internal attention (Fig. 10). ExtAtt was dominated by a decrease in alpha-band connections in a bottom-up fashion. On the contrary, IntAtt was characterized by an increasing trend of alpha-band connections in both directions. A few considerations can be drawn. First, the selected frontal ROI (MFG, particularly in the left side) appeared implicated in top-down alpha influences during the internal task. Second, this mechanism was accompanied by a second one, i.e. an increase in bottom-up alpha influences (Fig. 10 central top panel). It is worth noting that, besides top-down, also posterior-to-anterior information flow in alpha band has been reported to be involved in internally oriented tasks and conditions [63, 64]. This is in line also with our results in IntExtAtt condition where an increase in top-down connectivity was accompanied by a reduced decrease of bottom-up connectivity compared with ExtAtt. Overall, these results might indicate that both top-down and bottom-up communication in alpha band possibly operate to favor insulation of internal processing from ongoing external interference.

Some limitations and ideas for further studies are presented.

In our experimental protocol, we used pictures changing every 10 seconds to induce visual attention. The significant and marked decrease of alpha activity in posterior visual regions during ExtAtt condition indicates that visual attention of participants was actually captured by the pictures. However, we cannot totally exclude that mind-wandering occasionally occurred, e.g. at some points between one picture and the next, reducing visual attention. Increasing the rate of picture presentation (e.g. every 5 seconds or less) or watching a video, might increase the effect of external attention capture, with possible impact on alpha power both in ExtAtt and IntExtAtt conditions. This could be explored in future studies. Another point is that the two examined conditions (performing mental arithmetic vs watching pictures) differ, besides the direction of attention, also as to other variables, such as voluntary vs automatic allocation of attention, requiring vs not requiring manipulation of information, demanding vs not demanding task. These variables might have partially contributed to the observed EEG differences in oscillatory activity and connectivity. However, our hypothesis is a posteriori validated by the observation that, according to the existing literature, the obtained changes in alpha and theta mechanisms are mainly modulated by the direction of attention, i.e. are specific of attention orientation rather than being task-specific or being associated to other variables (see the Introduction where we emphasized how similar alpha or theta mechanisms operate in tasks even very different but that share the same direction of attention).

Another important aspect concerns the investigation of other rhythms together with theta and alpha, e.g. gamma rhythm (30-100 Hz). Indeed, enhanced gamma oscillations have been observed during high-level (internally oriented) mental processing (reading, emotion, math task, memory) especially in task-relevant areas [5, 6, 65], and also increased gamma oscillations in sensory cortices have often been linked with increased sensory attention [3]. Furthermore, theta-gamma coupling, with gamma cycles nested within a theta cycle and phase-locked to it, is considered an efficient scheme for implementing the simultaneous representation and manipulation of multiple items. In relation to our results, adding the exploration of gamma oscillations may help to understand the functional role of the posterior theta in the investigated conditions.

We acknowledge that the accuracy of source reconstruction improves with increasing sensor density and using individualized head models from MRI, while we used 32 electrodes and a head model template (a limitation however common to a great body of literature and even severer in cases when just 19 electrodes are used). However, here we were not interested in an exact cortical localization, and we did not focus on single significant voxels; instead, we considered overall regions of interest, to characterize the average behavior of an area. Therefore, we expect that some blurring and inaccuracies in source reconstruction derived by the adopted techniques may have had a tolerable impact. Furthermore, some reassurances come from results at source level in concordance with the existing literature (as reported previously). Nevertheless, our source-level results, both as to power and connectivity, definitely deserve further studies using high-density electrode montage (e.g. 64 electrodes) and possibly a larger sample size (see also below) to achieve higher reliability. This may be of particular relevance to attest our provisional result showing involvement of theta activity in PCU during visual attention; this result indeed does not find clear interpretation within the existing literature and further validation is required.

Another weakness, that may benefit from a larger number of electrodes and of participants, concerns the low statistical significance of the Granger Causality effects that did not survive multiple comparison correction in most of cases (Figs. 9 and 10) and revealed a high inter-individual variability. A certain variability was already visible at power level (especially in alpha activity in IntAtt and IntExtAtt); it is possible that some participants adopted different strategies for keeping internal concentration or learned automatized procedure to perform the task (reducing the effort) or may even be particularly susceptible to external influences. In future studies we can further increase the number of participants to identify a clearer tendency and also perform an analysis at single-subject level to possibly disclose different adopted strategies. This kind of analysis could further benefit from controlling participants’ performances in the mental task and/or their arithmetic proficiency, which we did not systematically assess in this study.