As already presented, there was no significant difference between control and left-HS group with regard to the results of WM recognition test. This finding is in agreement with Tudesco et al. (2010) , who did not detect WM deficits in patients with MTLE associated with HS by means of classic neuropsychological evaluation.
Separated EEG and fMRI data collected in the control group
The EEG results for the control group during the WM encoding period revealed theta oscillations (5-6 Hz) in the frontal area, with peak amplitude in the Fz electrode. This result agrees with some studies in the literature, which have assigned the so-called frontal midline theta rhythm (fmθ) to possible dipolar sources located in the dorsal portion of the anterior cingulated cortex and in the medial prefrontal cortex [15, 44–46], and related it to the performance on cognitive tasks that requires a high level of attention, such as WM tests [44, 45, 47]. Other studies have also associated this increase in theta oscillations in frontal region with the increase in working memory load [15, 17, 44], emphasizing the preponderant role of frontal lobe structures to the WM encoding and maintenance processes. However, it is important to highlight that we just analyzed the presence or absence of oscillatory rhythms during encoding and maintenance stages, but not the amplitude variations of these rhythms in relation to the amount of items loaded in the WM. On the contrary, WM maintenance did not show theta oscillations, being in disagreement with some other findings in the literature. Some authors detected theta oscillations , while other also correlated it to the number of items to be sustained in WM by normal controls [15, 17, 44].
On the other hand, the EEG results for both WM encoding and maintenance periods demonstrated the presence of alpha oscillations (9-10.5 Hz) in the parietal area, with peak amplitude in the Pz and P3 electrodes, slightly dislocated to the left cerebral hemisphere. According to Gevins et al. (1997) , the alpha rhythm is detected not only in the encoding but also in the maintenance stages of WM. However, in opposition to what happens with the theta rhythm, its amplitude tends to decrease with the increase of WM demand.
Although the majority of the studies have focused on the theta and alpha oscillations, some authors have also found delta [49, 50] and gamma [51, 52] oscillations in normal controls during working memory tasks.
With respect to the fMRI results for the control group during the WM encoding period, a positive BOLD signal was found in the bilateral > left frontal midline area, corresponding to the same region of theta rhythm peak location on the EEG recording. During WM maintenance, a positive BOLD was also detected in the same frontal midline area, although presenting decreased intensity and extension, corresponding to the disappearance of the fmθ on the EEG.
Likewise, a positive BOLD signal was identified in the bilateral > left and left parieto-occipital areas in the WM encoding and maintenance periods, respectively, coinciding with the alpha rhythm location on the EEG. Only in the encoding period, other positive BOLD signal sources were found in the left temporal region, as well as in more medial and deep cerebral structures, but without any corresponding rhythmic activity on the EEG. This lack of oscillatory rhythms in certain cerebral areas is probably due to the EEG low sensitivity to detect electrophysiological signals arising from medial/deep structures, such as putamen, insular cortex, and fusiform and lingual gyrus.
In both periods, there was co-occurrence of positive and negative BOLD signals, which were distributed in adjacent but not coincident cerebral regions. During the WM encoding, the distribution of these positive versus negative signals was more counterbalanced in terms of number and extension. Conversely, during the WM maintenance, there was a clear negative signal predominance that remained basically at the same locations, although with larger extension. Nonetheless, there is still a lot of controversy in the literature concerning the relationship between the WM stage (encoding or maintenance) and its associated BOLD signal (positive or negative). Some authors [53, 54] have related the maintenance period to the positive signal (“task positive network”), while others [15, 17, 22, 55] have associated it with the negative signal (“default mode network”). Maybe the positive BOLD signals detected during the encoding period are related to the memorizing effort (“task positive network”) of the selected items, whereas the more concentrated and restricted positive BOLD identified during the maintenance period is associated with the sustaining effort (“default mode network”). This sustaining effort may also involve more inhibitory mechanisms, which are expressed by more negative BOLD signals in the fMRI images.
Separated EEG and fMRI data collected in letf-HS group
In similarity with the control group, left-HS group EEG results revealed the presence of theta oscillations (5-7 Hz) in the frontal area during the WM encoding, but not in the WM maintenance period. Nevertheless, while these theta oscillations were more concentrated in Fz for the controls, they were more distributed between Fz and Cz for the patients during the WM encoding. In addition, similarly to the controls, the EEG results for the left-HS group demonstrated the presence of alpha oscillations (9-10 Hz) in the parietal region (mainly in Pz), during both WM encoding and maintenance periods. However, these alpha oscillations were more extensive and bilateral (P3 and P4) for the patients, whereas they were more left-sided (P3) for the controls. Taken together, these findings indicate a more widespread central theta and bilateral alpha oscillations for the left-HS group, probably to compensate for their left mesial temporal dysfunction.
In the left-HS group, in contrast to the control group, other frequencies oscillations around 17 Hz, 18 Hz and 20 Hz, consistent with the beta rhythm, were also observed in the frontal, central and parietal (Fz, F4, Cz, Pz, P3 and P4) areas in three patients during the WM encoding, and in two of them during the WM maintenance. In other studies using similar WM paradigms (Sternberg task), significant positive correlations between beta/gamma and the BOLD signal have been found more focally than the correlations in lower frequency bands (theta and alpha), which are localized in larger cortical networks, e. g. in the default mode network (DMN) . These local field potential oscillations in the beta/gamma range, highly correlated to the BOLD response, probably represent neuronal assembly dynamics cognitive processing . On the other hand, our finding of other oscillatory rhythms in the left-HS group may indicate a reorganization of the patients´ neuronal network involved in the WM task, as an attempt to overcome the left hippocampal system dysfunction.
As regards fMRI results for the left-HS group during the WM encoding period, a positive BOLD signal was found in the frontal midline area, matching with the frontal midline theta rhythm location on the EEG recording. During the WM maintenance, a decrease of this signal was observed, which was in accordance with the disappearance of the theta oscillations. Other frontal regions also presented positive signals, such as the lateral middle and inferior frontal and precentral gyrus. However, whereas these activations were bilateral > left for the controls, they were bilateral for the patients. Likewise, a positive BOLD was identified in the bilateral and left parieto-occipital areas in the WM encoding and maintenance periods, respectively, coinciding with the alpha rhythm location on the EEG. Nonetheless, in opposition to the controls who exhibited bilateral > left positive signals, the patients showed bilateral positive signals during the encoding period. Conjointly, these findings suggest a more bilateral engagement of frontal and parietal regions, without right or left predominance, or a greater involvement of right frontal and parietal structures for the encoding stage in the left-HS group. Finally, in contrast to controls, no positive BOLD signal was found in the temporal area during WM encoding or maintenance periods in left-HS patients, probably due to their hippocampal lesion.
Similarly to controls, there were: (1) coexistence of positive and negative BOLD signals, which were distributed in adjacent cerebral regions in both WM encoding and maintenance; and (2) the encoding period was characterized by a more counterbalanced distribution of positive versus negative signals, while the maintenance period was marked by a clear negative signal predominance. On the other hand, the most important difference between the two groups was that, while controls had exhibited positive and negative BOLD signals in temporal area for both WM encoding and maintenance, the patients had no positive and less negative signals in this cerebral region for encoding and maintenance periods, respectively. In general, these results are in line with other studies [57–60] indicating, from different experimental contexts, network reconfiguration induced by neuronal damage associated with MTLE.
The reduced patient sample may be seen as one of the drawbacks of the present study. Its main purpose, however, is to present evidence of changes in patterns of EEG and fMRI signals that may be associated with brain plasticity due to epilepsy. Certainly, a more extensive study, including right-HS patients, would be needed to further explore this finding. Another limitation comes from the coarse resolution of the EPI images acquired in this study. While some authors [61–63] have found bilateral symmetrical hippocampal activation in normal controls and bilateral-asymmetrical hippocampal activation in MTLE patients, we did not find clear hippocampal activations. This is most likely due to partial volume effects, since in low resolution images nonactivating white matter is averaged with the signal from hippocampal gray matter leading to loss of BOLD signal . Additionally, it should be added a restriction we had on the analysis of the EEG data. The EEGLAB software used in this study did not allow perform a quantitative group analysis, since it requires time and channels consistency for all individuals within the same group, which was not achieved. Thus, a quantitative analysis was performed individually using ICA decomposition, which resulted in values of frequencies found for each subject in each group. Despite this limitation, it was possible to find consistency in the results of individuals in each group.