3.5 Analysis on the frequency band 0.01-0.1 Hz
3.5.1 EEG correlates of dFC states
The results obtained to identify the most different topography associated to adFCstate for the relative power on the fourEEGbands (delta [1,4] Hz, theta [4,8] Hz, alpha [8,12] Hz and beta [12,20] Hz), as well as the normalizedASImeasure are represented in Figure3.7.
Regarding the relative power, for the majority number of states the most different topographies were found for the relative alpha power, with some exceptions where the relative delta power registered the highest value. The relative theta and beta power always showed low Euclidean distance and similar to
Figure 3.7:Maximum Euclidean distance between most similar state topographies for each number of states using EEG relative power from different bands and ASI.
all number of states meaning they were not good to differentiate topographies of differentdFC states.
Regarding the highest value, it is observed for the relative alpha power for 14 states. The results for the relative delta power showed that the highest value was also for 14 states, however a similar value was obtained for 4 states.
In the case of the ASImeasure, the number of states with the most different state topography is 4,
followed by the 14 states, the same with the most different topography using relative alpha power.
In order to have a full perspective of the dFC states and the associated topographies, Figure 3.8 shows the mean of topographies of relative alpha (the band with highest differences) across time points with the same dFC label, for the range k = 3 to 15 states and the correspondent dFC state that is represented in the cortical space, where the values of the centroid vector of each state are used to scale the color of each brain area and the connections (blue links) are plotted between brain areas with a value
<−0.1, to highlight the network contrasting from the global mode. The negative value is used due to the initial definition where was defined that the positive elements corresponded to brain areas following the global mode, whereas negative elements were brain regions going in the opposite direction of the global mode.
Analyzing Figure3.8it is possible to observe that, aside from 3 states, in all other states the topogra- phies with overall high relative alpha power with a prominence in the central regions of the parietal lobe are associated to at least one of these two networks: one that resembles theDMN, with frontal areas, the posterior cingulate cortex and the left and right angular gyrus; another with frontal and parietal regions from the left and right hemispheres, resembling the frontoparietal network. However, the frontoparietal network is the one that has an overall greater power in most states that the others. There is also one moredFCstate associated with high alpha power, which appears only with 7 states but maintains until the highest number of states analyzed.
It may seem interesting that the maximum Euclidean distances found with relative band-power and ASIare in 4 and 14 states and these are the states where there is clearly one topography that is the most different from the others. However, one should recall that the methodology used was to try to find only one topography different from the others, penalizing the number of states where two equal topographies where associated to different states.
Regarding the 14 states, the time-course of state 11 is shown in figure3.9. The occurrence of this dFCstate is very localized in one specific subject, with some other occurrences in some other subjects (not all). Furthermore, the subject where this state mostly occurs is subject 6, which has an increased alpha power, so the result is biased due to this subject. Thus, in order to avoid the cases where adFC state has most of its occurrences in a single subject, which is more frequent as the number of states increases, a lower number of states will be chosen to characterize in terms of band-power topographies, as well as the dynamics of thedFCstates.
In order to choose a number of states without imposing a number of topographies with high alpha, as these are the most different, the mean value of each topography was computed to characterize each topography by a single value. Then the maximum value was taken in order to select states that have higher alpha than others from different number of states. The comparison between the maximum values obtained on each number of states (3 to 15) are shown in Figure3.10.
Figure 3.8:Cortical representation of the dFC state and associate mean topography of relative alpha power for dif- ferent numbers of states (from k=3, top, to k=15, bottom). The blue links in the cortical representations are plotted between brain regions with a value<−0.1, highlighting the network that contrasts with the global mode.
Figure 3.8:(Continued)
Figure 3.10shows 3 local maxima, for 9, 11 and 14 states. As seen before, the high value for 14 states is due to a state quite specific of a single subject. So, in order to avoid cases like this, which are more probable as the number of states increases, the lowest number of states where there is a local maximum was chosen, that is 9 states. From Figure3.8it is possible to see that state 9 has 3 states with relatively high alpha power (states 3, 5 and 6). Therefore, to check if these three states are also influenced by the spectral content of a specific subject, the time-courses are shown in Figure3.11.
These three states have several occurrences across all (or almost all) subjects, not being so specific and influenced by a single subject and being more representative of the group of subjects.
The mean topographies across time points with the same dFClabel for the delta and alpha bands are represented in Figure 3.12, as well as the standard error associated to each topography. The topographies for theta and beta can be found in AppendixB.
States 3 and 5 and 6, besides being associated with an increase in the overall alpha power with a
Figure 3.9:Time-course of dFC state 11 after k-means clustering with k=14. The black dashed lines set the sepa- ration between each subject’s time-points.
Figure 3.10:Comparison between the maximum mean of each state topography, for different number of states (3 to 15).
peak on Pz electrode and neighbours, they also have a decrease in delta power compared to the other states, which is expected as these bands are usually anti-correlated.
Due to the highlight of these three states, thefMRI dFCstate associated to it is presented in Figure 3.13, where the regions that belong to each network are represented. A higher value in an element of the vector means that brain region belongs more strongly to that network.
Figure 3.11:Time-course of dFC states 3, 5 and 6 after k-means clustering with k=9.
Figure 3.12:Mean topographies of relative delta and alpha power for 9 states, followed by the standard error of the topographies across TRs. For comparison purposes, the color scales used for mean and standard error are the same.
Regarding the network of state 3, it encompasses the middle orbitofrontal gyrus, gyrus rectus and olfactory cortex. State 5 is mostly formed by regions from the frontal lobe (superior frontal, superior medial frontal and medial orbitofrontal cortex), the anterior and posterior cingulate cortex, angular gyrus and with less strength the gyrus rectus, olfactory cortex and hippocampus. State 6 is mostly formed by
regions from the frontal lobe (medial orbitofrontal, inferior frontal and left precentral) and parietal lobe (inferior lobules and left angular gyrus) and also includes the left medial and inferior temporal gyrus.
Regarding the connection between the state similar to theDMNand the increased alpha power, it is, to a certain extent, consistent to a result reported by Allen and colleagues [8], where they found adFC state (state 3 in their case) that showed a peak in the alpha band in the EEGspectrum, was present during eyes open and had positive correlations betweenICs of theDMNand anti-correlation betwenn DMNandCCregions. However, the DMNregions were also slightly connected to visual ICs, which is not observed in our state. Besides, there were differences in the methodologies, as they usedICAto define theROIs instead of using theAALatlas to segment intoROIs. They also had different conditions on the experiment, since their subjects had their eyes open half of the time and then closed their eyes the other half, while in our study the subjects remained the whole scan with their eyes open. And, finally, they are only analyzing some electrodes (Cz and the combination of 01 and 02), whereas we are using all electrodes. Unfortunately, besides this study, to the best our knowledge, no further studies analysed associations betweendFCstates and electrophysiological correlates.
A couple studies examined electrophysiological correlates ofRSNs [5,6,86]. However, in these stud- ies theBOLDtime-course of theRSNis known and can be correlated with alpha oscillations, whereas in the case ofdFCstates, these only show information about the synchronization between brain regions, whether or not theBOLDsignal is increasing or decreasing. Thus, only an association between increase of alpha power (alpha synchronization) anddFCstates can be elicited from our results.
Concerning the dFC state similar to the frontoparietal network, there is not a state found by Allen and colleagues that has onlyICs from this network. The other state, formed by the middle orbitofrontal gyrus, gyrus rectus and olfactory cortex, is an interesting state as only few regions are not synchronized with the global mode. In fact, this highlights one of the main advantages of performing adFCanalysis and also the employment of an atlas (AAL) instead ofICAto chooseICs similar toRSNs. As we are not imposing the existence of only resting-state networks, or regions only belonging to them, but are letting the data dictate which brain regions exhibit functional connectivity a greater number of times throughout the scan, this allows to find electrophysiological correlates of brain states that otherwise could have been missed.
The reason these three states are associated with high alpha power mostly on channel Pz and surrounding channels may be related to the proposal of alpha oscillations as an ”idling rhythm”, i.e, a rhythm whose presence inactivates certain brain regions [87]. High levels of activity of the DMN have been associated to mind wandering and attention lapses [86], being a network that is active when there is no task. Klimesch and colleagues [88] found that alpha synchronization has an active role in performing functional inhibition of regions not relevant to task performance. So, it seems reasonable to have a connection between higher alpha power and higher connectivity between regions of theDMN.
Regarding the frontoparietal network, Sadaghiani and colleagues found that alpha synchronization, i.e, increased alpha power, was positively correlated to cognitive functions associated with the fron- toparietal network [89]. Furthermore, a recent study found that this network had a diminished expression after the administration of psilocybin [48]. EEG studies with psilocybin have shown decreased parieto- occipital alpha power [90]. Thus, the association of high alpha power with this network is consistent with these findings.
With respect to state 3, the functionally connected regions that oppose the global mode are regions that are also found connected in theDMN. It may suggest that sub-parts of this network are functionally connected at certain times and at other instants they may connected with other regions. This would, at a certain extent, explain the association of such network with high alpha power. However, further studies need to be performed to understand why high alpha power, mainly parieto-occipital alpha, is associated with these states.