Transcranial direct current stimulation (tDCS) is a non-invasive brain stimulation tool suited to alter cortical excitability and activity via application of direct currents. Anodal tDCS over the motor cortex during rest has been shown to increase and cathodal tDCS to decrease excitability of this area [10], [11]. Interestingly, the after-effects of tDCS are NMDA receptor-dependent [12], [13], [14], thus sharing some similarities with LTP, and LTD, which resemble well-known neuroplastic alterations thought to underlie cognitive processes like learning and memory formation [15]. In line with these studies in humans, anodal tDCS over M1 was shown to promote synaptic plasticity in rat brain slices, producing synaptic LTP [16]. In accordance, anodal tDCS improves motor learning and non-dominant hand function in healthy subjects [17], [18], [19], as well as facilitate performance of motor skills in stroke patients with respective deficits [20], [21], [22].

In prevoious imaging studies the impact of tDCS over M1 have been studied at the large-scale level (i.e. studying whole brain interactions), where it has been reported that excitatory anodal tDCS is capable of modulating motor-task related cortico-cortical [23], [24], [25] and cortico-subcortical [26] functional circuits. However it has been not yet stablished how tDCS-induced neuroplasticity over M1 intra-regionally reorganizes its functional architecture.

The second step was to perform an analysis at the nodal level. Therefore, a threshold common for all subjects was selected as described in detail in the methods section. The maximum T that included at least 95% of the nodes in the largest cluster across all of the subjects was T=0.352. Therefore, we used this threshold to build the undirected graphs and subsequently generate volume/surface maps with each network metric (K, C and L) for all of the resting state data-sets before and after tDCS. For visualization purposes, the undirected graph representation for one of the correlation matrices thresholded at T=0.352 is shown in Figures S3 and S4. The ANOVAs calculated in the surface space revealed a significant interaction effect (stimulation*time) for the clustering coefficient (Talairach x=−40,y=−9,z=57, peak F-value=5.1, p<0.005; cluster size 439 mm²; figure 3C) and the characteristic path length (Talairach x=−39,y=−11,z=55, peak F-value=5.9, p<0.005; cluster size 392 mm²; figure 3D). These significant clusters are located in the dorsolateral BA4 and approximately belonging to the arm/hand area according to an fMRI M1 mapping carried out by Meier et al. [43]. With regard to the connectivity degree, we did not find any interaction effect (figure 3A). After Monte-Carlo cluster correction (P<0.05) to the paired t-tests statistical clustering coefficient maps after-before of each stimulation condition, we identified one positive cluster in the cathodal stimulation condition (Talairach x=−39,y=−11,z=56, peak t-value=5.2, P<0.001; cluster size 499 mm²; figure 3F). The After – Before contrasts applied to anodal and sham conditions did not show any significant cluster. A re-test paired t-test analysis to the contrast After_Cathodal – After_Sham also revealed a positive cluster located at approximately the same location of the ANOVA test (Talairach x=−39,y=−11,z=57, peak t-value=5.2, p<0.005; cluster size 468 mm²; figure 3E). T-tests applied to the baselines between conditions did not reveal any significant cluster. The same analysis was repeated for the paired t-tests on the characteristic path length maps. The After_Anodal – Before_Anodal contrast revealed a positive cluster (Talairach x=−37,y=−13,z=53, peak t-value=4.7, p<0.005; cluster size 520 mm²; figure 3H). The After – Before contrasts applied to cathodal and sham conditions did not show any significant cluster. A retest paired t-test analysis to the contrast After_Anodal – After_Sham also revealed a positive cluster located at approximately the same location of the ANOVA test (Talairach x=−40,y=−9,z=55, peak t-value=4.9, p<0.005; cluster size 451 mm²; figure 3G). T-tests applied to the baselines between conditions did not reveal any significant cluster.

In an exploratory analysis, we investigated whether tDCS applied over the left M1 may have resulted in functional connectivity changes of the contralateral M1. Thus, we repeated the whole analysis (MRI-fMRI pre-processing and graph theory) using the right BA4. However we did not find any significant alterations of functional connectivity with regard to this area (no significant cluster showed up using ANOVA and evaluating the interaction group(anodal, cathodal, sham) * time(pre- and post-tDCS) for any of the nodal network parameters used in the present study (nodal connectivity degree, characteristic path length and clustering coefficient)). This lack of tDCS-induced functional reorganization in the contralateral hemisphere might correlate with a study of Lang and colleagues [44], where exactly the same electrode size (5×7 cm), stimulation intensity (1 mA) and duration (10 min) were applied; however the investigators failed to find changes of cortical excitability in the opposite hemisphere (also the right M1) – the authors only observed some weak effect in interhemispheric inhibition. One possible explanation for this is that interhemispheric connections have higher thresholds than local cortico-cortical and cortico-spinal connections [45], hence a lack of contralateral excitability. Additionally, it might be also possible that the intensity of 1 mA is too weak to modulate transcallosal activity. These are important points that should be examined in future studies by combining functional reorganization evaluated with graph theory accompanied by electrophysiological measures.

In a post-hoc analysis, we investigated whether the tDCS-induced functional connectivity alterations observed in the C and L maps depended on the initial functional network metric of each subject. The average of the effect within the cluster (after-before within tDCS) was computed for each subject and then these values were linearly regressed against the baseline value of its respective network metric, e.g. the after-before mean values of the significant cluster found in the C Maps in the cathodal tDCS condition were regressed against the before cathodal C maps. The effect of a clustering coefficient increase following cathodal tDCS strongly correlated with the baseline C (P=0.0051; R²=0.46) (Figure 4A). The negative increase in the characteristic path length that was found after anodal tDCS also correlated positively with the baseline metric (P=0.002; R²=0.51) (Figure 4B). Regressions applied to the same clusters in the sham condition did not result in any significant correlation.

In a second post hoc analysis we investigated the reason to have such localized effect of tDCS in the dorsolateral BA4 region (Figures 3C–H). We hypothesized that the reason for the arm/hand region to be significantly altered by tDCS during rest is that this is the M1 region with the most efficient dynamic architecture. We mapped the nodes that communicate more efficiently independently from stimulation (i.e. before tDCS intervention) within the M1 network. To this end, the average L maps for all subjects and all before-tDCS fMRI scans were averaged. Nodes with the highest L_rand/Li values (i.e. nodes that communicate more efficiently within M1) were mapped over the flattened BA4. As an exploratory threshold we used the 15% of the voxels that showed the highest L_rand/Li values. The largest hub was located at the centre of the flattened BA4 area, which approximately represents the arm/hand area according to an fMRI M1 mapping carried out by Meier et al. [43] (peak value at Talairach x=−36,y=−15,z=55). Two additional smaller hubs were identified: one belonging to approximately the leg area (cluster at the top of the flattened BA4 (medial BA4); Talairach x=−36,y=−15,z=55); and a second cluster belonging approximately to the face/tongue area (cluster at the bottom-tight of the flattened BA4 (most lateral BA4); Talairach x=−36,y=−15,z=55) according to an fMRI M1 mapping carried out by Meier et al. [43] (Figure 5).

We tested the hypothesis that tDCS-induced neuroplasticity over M1 results in an alteration of associations among populations of M1 neurons, reflected in a change of its functional architecture. However, before it was important to show that the functional connections of M1 are organized in distributed and efficient way, rather than having a lattice organization. For stimulation-independent functional connectivity of M1 during rest, the results show that M1 has salient small-world properties, i.e. lambda≈1, gamma1, thus sigma1 [32], [41], indicative for highly efficient integration of both, localized and segregated information processing, in M1. If M1 functional organization would reflect a precise topography with discrete representations, we would have expected reduced segregated connectivity (i.e. lambda1 [32]), which is clearly not the case. This finding is important, because this kind of functional architecture should be relatively flexible for modification and map reorganization. In accordance, it has been shown that the control and learning of simple and complex voluntary movements emerge from M1 distributed networks rather than discrete representations [46].

Hence, the second step was to explore if tDCS-induced neuroplasticity of M1 is reflected in a modulation its functional architecture. At first instance, Figure 2 shows a lack of global alterations of functional connectivity of M1 induced by tDCS, i.e. the global M1 network parameters remained constant (for all studied thresholds no significant changes in the global mean connectivity degree K, as well as the small-world properties were obtained (Figure 2)). This means that the application of relatively weak constant currents (1 mA) over the scalp of healthy humans for few minutes still preserves the global functional structure of M1. However, when we compared the surface maps that contained the information of the nodal network metrics, we found significant changes at local clusters within M1 following tDCS. Cathodal stimulation induced an increase in the clustering coefficient at the dorsolateral BA4 – approximately the M1 arm/hand area. It should be noticed that this effect was not accompanied by a significant modulation in the connectivity degree (i.e. the number of functional connections did not significantly increased or decreased). This means that the strength of the functional connections at the local level was significantly increased by cathodal tDCS. This is an interesting finding considering that the primary mechanisms of the excitability shifts induced by tDCS are subthreshold neuronal membrane depolarization by anodal stimulation and membrane hyperpolarization by cathodal stimulation, and similar consecutive alterations of spontaneous cortical activity as shown directly in animal studies [47], [48], but also suggested for tDCS of the human motor cortex [13]. A likely explanation for this result might be that the local decrease of spontaneous activity induced by cathodal tDCS increased the signal to noise ratio (by inducing neuronal hyperpolarization) and consequently increased synchronization at the local level. This hypothesis might be supported by previous studies where cathodal tDCS targeted at other brain regions is suggested to increase signal to noise ratio e.g. the motion processing areas in the visual cortex [49]. Following this concept, we would expect that anodal tDCS might have induced a decrease in local synchronization at M1 level due to a reduction of the signal to noise ratio. Although for anodal tDCS we did not find any region where the clustering coefficient or connectivity degree significantly increased or decreased, the characteristic path length significantly decreased in a cluster centred – again – at approximately the arm/hand area of M1. This means that the nodes belonging to that cluster communicate more efficiently with the rest of the M1 network. Similar to cathodal tDCS-induced effects, the significant decrease in the characteristic path length induced by anodal tDCS was not accompanied by a significant increase in the connectivity degree. Therefore, the increase in efficiency does not depend on an increase in the number of functional connections, but is rather due to a reorganization of the functional network. Thus, our results provide for the first time evidence that the promotion of LTP-like plasticity induced by anodal tDCS [16] might be related to an efficient reorganization of the functional architecture of M1. Interestingly, a recent study shows that both anodal and cathodal tDCS over M1 induces a change in the generalization of the intrinsic coordinates of movement representations within M1 [50]. In that study the authors speculated that this change could result either from larger recruitment of the neurons during learning (increase in population number) or from a larger modification of the activity the respective neurons (increase in modulation), thus suggesting that the behaviourally quantified generalization patterns are due to tuning properties of neurons in specific networks within M1. Our results may in part confirm these hypotheses.

In a post-hoc analysis we investigated whether the tDCS-induced modifications of functional architecture may depend on the baseline functional organization of the identified clusters in Figure 3. We found that the effect of cathodal tDCS on the clustering coefficient strongly depends on the efficacy of the local connectedness before stimulation (Figure 4A). The same was true for the effects induced by anodal tDCS, where the baseline efficacy in the communication of the detected cluster in Figure 3C with the rest of M1 was positively correlated with anodal tDCS-induced alterations (Figure 4B). Interestingly, motor cortex plasticity induced by paired associative stimulation has been shown to be more effective in physically active than in sedentary individuals [51]. The investigators speculated that participation in regular physical activity may offer global benefits to motor cortex function that makes easier to enhance neuroplasticity. In principle accordance, the results of the present study suggest that the more efficient the functional architecture of M1 is at baseline, the more efficient the tDCS-induced functional modulations are. Following this concept, we hypothesized that the reason for the arm/hand region to be significantly altered by tDCS during rest is that this is the M1 region with the most efficient dynamic architecture. To test this hypothesis, we carried out a second post-hoc analysis, where we mapped the nodes that communicate more efficiently independently from stimulation (i.e. before tDCS intervention) within the M1 network. To this end, the average L maps for all subjects and all before-tDCS fMRI scans were averaged. Interestingly, we found that the regions where we found the significant tDCS-induced alterations belong to the largest M1 hub which is located at approximately the arm/hand region (Figure 5).

The substrate for plasticity induction within M1 is most likely a system of horizontal connections that spans M1, which may mediate the formation of associations among populations of M1 neurons and have been repeatedly shown in several studies to have the capacity for long-lasting synaptic modification [52], [53], [54]. Altogether, our results suggest that the mechanisms of the excitability shifts induced by tDCS (primary membrane hyper- and de-polarization by cathodal and anodal tDCS respectively, which results in NMDA receptor-dependent alterations of synaptic strength [12]) may take advantage of this M1 intrinsic circuitry, which supports the previously mentioned optimal conditions for network reorganization.

One of the most studied and well established outcomes of tDCS-induced neuroplasticity over M1 is the alteration of the size of motor evoked potentials (MEPs) as a measure of regional plasticity. Anodal tDCS increases and cathodal tDCS decreases the MEP size respectively, providing evidence for tDCS polarity-dependent neuroplasticity [10], [11]. The exact relation between this regional neuroplasticity and alterations of functional connectivity is so far unclear. Hereby, connectivity modulations induced by tDCS – as shown in the present study – might hint to a neuroplastic effect of tDCS on functional connectivity. However, this should be explored more directly in future studies. Additionally, it is important to notice that in the present study we evaluated the effects of tDCS alone – i.e. no motor training or learning was performed during or after the application of the stimulation. Thus it cannot be derived from the results of the study if tDCS improves motor learning due to its impact on functional connectivity. Since we have shown that tDCS reorganizes the functional architecture at the local level, and task-dependent alterations of functional connectivity have been demonstrated in other studies, this is however a tempting speculation which should be tested directly in future experiments.

Figure S1

Figure S2

Figure S3

Figure S4