The authors have declared that no competing interests exist.
Conceived and designed the experiments: MC GK MB. Performed the experiments: VG SB. Analyzed the data: VG SB MC. Contributed reagents/materials/analysis tools: GK MC. Wrote the paper: MB GK MC. Proofread and commented on the manuscript: GK MB SB VG CC MC.
Recording of slow spontaneous fluctuations at rest using functional magnetic resonance imaging (fMRI) allows distinct long-range cortical networks to be identified. The neuronal basis of connectivity as assessed by resting-state fMRI still needs to be fully clarified, considering that these signals are an indirect measure of neuronal activity, reflecting slow local variations in de-oxyhaemoglobin concentration. Here, we combined fMRI with multifocal transcranial magnetic stimulation (TMS), a technique that allows the investigation of the causal neurophysiological interactions occurring in specific cortico-cortical connections. We investigated whether the physiological properties of parieto-frontal circuits mapped with short-latency multifocal TMS at rest may have some relationship with the resting-state fMRI measures of specific resting-state functional networks (RSNs). Results showed that the activity of fast cortico-cortical physiological interactions occurring in the millisecond range correlated selectively with the coupling of fMRI slow oscillations within the same cortical areas that form part of the dorsal attention network, i.e., the attention system believed to be involved in reorientation of attention. We conclude that resting-state fMRI ongoing slow fluctuations likely reflect the interaction of underlying physiological cortico-cortical connections.
In the human brain complex networks rather than isolated cortical areas sub-serve specific brain functions, such as, for instance, movement, memory, or attention
In humans, simultaneous recording of electroencephalography (EEG) and fMRI at rest has returned significant correlations between the variations of band power of different cortical rhythms and BOLD fluctuations within specific brain networks
A unique opportunity to challenge the neurophysiological characteristics of functional connectivity of the brain at rest is provided by multifocal transcranial magnetic stimulation (TMS)
Here, we aimed to investigate whether the physiological properties of these circuits may have a relationship with the resting-state fMRI measures, hypothesizing that BOLD fluctuations reflect underlying neurophysiological interactions. Therefore, we used correlation analyses to verify whether, across subjects, specific associations exist between connectivity in parieto-frontal circuits assessed by multifocal TMS, and BOLD signal changes in well-known RSNs whose nodes overlap with the sites of TMS stimulation.
In total, 19 healthy volunteers (7 males and 12 females; age ranging from 21 to 36 years) took part in this study. The participants are a subsample of those enrolled for a previous study, in which we combined TMS and diffusion imaging data
All subjects underwent an MRI examination, obtained at 3 T (Magnetom Allegra, Siemens, Erlangen, Germany), including the following sequences: (1) dual echo turbo spin echo (TR:6190 ms, TE:12/109 ms, matrix:2563192, FOV:230×172.5, 48 contiguous slices, slice thickness:3 mm, total scan time:4 min); (2) three-dimensional modified driven equilibrium Fourier transform (MDEFT) scan (TR:1338 ms, TE:2.4 ms, TI:910 ms, flip angle:15°, matrix:256×224×176, in plane FOV:256×224 mm2, slice thickness:1 mm, total scan time:12 min); (3) T2* weighted echo planar imaging (EPI) sensitised to blood oxygenation level dependent imaging (BOLD) contrast (TR:2080 ms, TE:30 ms, 32 axial slices parallel to AC-PC line, matrix:64×64, pixel size:3×3 mm2, slice thickness:2.5 mm, flip angle:70°) for resting state fMRI. BOLD echo planar images were collected during rest for a 7 min and 20 s period, resulting in a total of 220 volumes. During this acquisition, subjects were instructed to keep their eyes closed, not to think of anything in particular, and not to fall asleep.
Data were pre-processed using Statistical Parametric Mapping (Wellcome Department of Imaging Neuroscience; SPM8), and in-house software implemented in Matlab (The Mathworks Inc, Natick, Massachussetts, USA). For each subject, the first four volumes of the fMRI series were discarded to allow for T1 equilibration effects. The pre-processing steps included correction for head motion, compensation for slice-dependent time shifts, normalization to the EPI template in MNI coordinates provided with SPM8, and smoothing with a 3D Gaussian Kernel with 8mm3 full-width at half maximum. Then, all images were filtered by a phase-insensitive band-pass filter (pass band 0.01-0.08 Hz) to reduce the effect of low frequency drift and high frequency physiological noise.
We employed independent component analysis (ICA) (group ICA for fMRI toolbox, GIFT,
A first test stimulus (TS) was applied over the hand motor areas of left M1 and was defined as the point in which stimulation evoked the largest motor evoked potentials (MEPs) from the contralateral first dorsal interosseous (FDI) muscle. Electromyographic (EMG) traces were recorded bilaterally from the FDI muscles using 9-mm diameter, Ag–AgCl surface cup electrodes. The active electrode was placed over the belly muscle, while the reference electrode was located over the metacarpophalangeal joint of the index finger. Responses were amplified using a Digitimer D360 amplifier (Digitimer Ltd, Welwyn Garden City, Hertfordshire, UK) through filters set at 20 Hz and 2 kHz with a sampling rate of 5 kHz, then recorded by a computer using SIGNAL software (Cambridge Electronic Devices, Cambridge, UK). The test stimulator for M1 was connected to a small custom-made figure-of-eight-shaped coil (external diameter 50 mm). The intensity of the TS was adjusted to evoke a MEP of approximately 1 mV peak to peak in the relaxed contralateral FDI muscles. To best activate the ipsilateral PPC-M1 facilitatory functional connection, a conditioning stimulus (CS1) applied over the PPC preceded the M1 TS by 5 ms, at an intensity of 90% of the ipsilateral resting motor threshold (RMT) (
A) Left posterior parietal cortex (PPC) TMS preceded left M1 TMS by 5 ms and induced increased MEP amplitude, indicating intra-hemispheric functional connectivity (blue arrow and blue column). When a third conditioning pulse was applied 10 ms earlier to contralateral PPC, the left intrahemispheric interaction was abolished, reflecting the activation of a transcallosal inhibitory pathway (red arrow and red column). B) Mean normalized MNI coordinates (x, y, z, mean±SD) of TMS PPC site were −48.2±4.8, –65.2±3.9, and 45.3±3.4 mm in the left hemisphere and 52.5±6.3, −60.2±4.7 in the right hemisphere. Mean MNI coordinates of left M1 were −30±3.3, −12±3.4, and 71±4.3.
To investigate the functional inter-hemispheric connectivity between the PPC areas, a third TMS pulse (CS2) was applied over the contralateral PPC (PPCCONTRA) 10 ms before delivery of the PPC pulse ipsilateral to M1 (PPCIPSI), and therefore 15 ms before the M1 TS
A within-subject ANOVA was applied on mean MEPs amplitude with condition as main factor (TS alone (MEP), PPCIPSI + TS, PPCCONTRA + PPCIPSI + TS) in order to verify the changes in the MEP amplitude following bifocal or trifocal TMS.
The across-subject correlation between fMRI and TMS data was assessed using a random-effect analysis in SPM8. Two models were set up for each of the 3 selected RSNs (DMN, DAN, and SMN). In the first model, intra-hemisperic PPCIPSI-M1 functional connectivity assessed by TMS (as defined above) was used as a regressor; in the second model the inter-hemispheric PPCCONTRA -PPCIPSI-M1 functional connectivity (as defined above) was used as a regressor. T contrasts were used to test the hypotheses of either positive or negative correlations.
The RSNs identified among the components extract by ICA are shown in
The default mode network (DMN) has been linked to self-oriented mental activity; the dorsal-attention network (DAN) has been associated with goal-directed stimulus-response selection; The visual network (VisNet) has been associated with visual processing; the auditory network (AudNet) has been associated to processing of auditory stimuli; the left-lateralized fronto-parietal network (LFPN) and the right-lateralized fronto-parietal network (RFPN) have been associated to memory functions; the sensory-motor network (SMN) comprises primary sensory-motor areas and the supplementary motor area. For the correlation analysis, we selected the DMN, the DAN, and the SMN, based on anatomical considerations. See text for further details.
As expected, conditioning TMS applied to the left PPC produced an increase of MEP recorded by TMS applied over left M1 in isolation (ANOVA condition main effect F(2,38) = 16.04; p = 0.00001; t = −2.12; p = 0.03), which implies the activation of intra-hemispheric facilitatory functional connectivity
The integration of TMS and fMRI data revealed a remarkable correlation across subjects between functional connectivity (both intra- and inter-hemispheric) as measured by multifocal TMS, and the degree of correlation of spontaneous activity within the DAN
In panel A and B, areas of correlation between the dorsal attention network (DAN) and bifocal TMS are shown in green, while correlations between the DAN and trifocal TMS are shown in red. Overlapping regions are in yellow. Panel B also shows the sites of TMS stimulation (blue dots). The scatterplots reported in the upper row of panel C show the mean cluster Z-score (indicating the strength of functional connectivity estimated by resting state fMRI) against the percentage change in MEP for the 3 highlighted regions. In the lower raw of panel C, the corresponding mean Z-score relative to the default mode network (DMN) are also plotted against the percentage change in MEP (bottom row), but correlations were not significant. The directions of correlation (direct or inverse) are reversed in the two cases (bifocal and trifocal TMS) as expected due to the opposite (excitatory or inhibitory) nature of the underling connections. R and p values are estimated post-hoc using Pearson's correlation coefficient. See text for further details.
Anatomical Location | MNI coordinates [x y z] | t-value | Cluster size [voxels] | p-value (FDR corrected) |
Bifocal TMS vs Left supramarginal gyrus/angular gyrus | −54 −50 32 | 6.48 | 260 | 0.001 |
Bifocal TMS vs Left Middle frontal gyrus | −42 26 46 | 4.54 | 13 | 0.15 |
Trifocal TMS vs Left supramarginal gyrus/angular gyrus | −56 −46 30 | 6.95 | 149 | 0.042 |
Trifocal TMS vs Right supramarginal gyrus/angular gyrus | 62 −48 30 | 6.66 | 112 | 0.042 |
Trifocal TMS vs Left middle frontal gyrus | −42 24 48 | 5.91 | 115 | 0.042 |
MNI = Montreal Neurological Institute; The p-values are corrected for multiple comparisons according to the false discovery rate (FDR) method.
In order to exclude the possibility that age and gender might have affected these results, we repeated the analysis, adding age and gender as covariates of no-interest. The results were consistent with those of the original analysis, although the statistical significance was reduced (remaining significant at p<0.1, FDR corrected at cluster level) due to the reduced number of degrees of freedom. A figure summarising the results of this analysis is provided on-line as
In order to further explore our hypothesis, we performed some further correlation analysis between the fMRI and the TMS data. Every subject's functional data underwent an additional step of pre-processing, consisting of the removal of potential sources of spurious variance, including: global temporal drift using a 3rd order polynomial fit, realignment parameters, and the signal averaged over whole brain voxels. The mean timeseries from the clusters found to be significantly associated with the DAN in the main analysis (left and right PPC and left premotor cortex) were extracted from the resulting images. The correlation coefficient between each pair of regions was estimated subject by subject using Pearson's formula. The resulting values (RLPPC-RPPC, RLPPC-PM, RRPPC-PM) were correlated across subject with the intra- and inter-hemispheric connectivity assessed using TMS.
No significant results were found. Correlation coefficients and p values are shown in
RLPPC-RPPC | RLPPC-PM | RRPPC-PM | |
Bifocal TMS | R = 0.24; P = 0.32 | R = 0.25; P = 0.31 | R = 0.29; P = 0.23 |
Trifocal TMS | R = −0.20; P = 0.42 | R = −0.35; P = 0.14 | R = −0.22; P = 0.36 |
Results of post-hoc correlation performed between TMS measures of connectivity and the average timeseries extracted for each of the regions of the dorsal attention network found to be significantly associated with TMS data using independent component analysis.
RLPPC-RPPC = Pearson's correlation coefficient between left and right posterior parietal cortex;
RLPPC-PM = Pearson's correlation coefficient between left posterior parietal cortex and left pre-motor cortex;
RRPPC-PM = Pearson's correlation coefficient between right posterior parietal cortex and left pre-motor cortex.
Our data indicate that the correlation structure of hemodynamic signals recorded by resting-state fMRI is tightly related to the physiological interactions tested by methods based on non-invasive cortical stimulation. Crucially, we report here a strong overlap between the cortical areas selected for the TMS, and the resting-state areas of BOLD signal correlation within the DAN, but not with other RSNs, namely the DMN and SMN. It should be highlighted that the fMRI analysis was performed in a completely data-driven fashion, and did not require any
The DAN is supposed to be involved in voluntary visual attentional control trough a large-scale distributed network formed by the frontal, parietal and visual cortices
Several previous studies have demonstrated that the fMRI RSNs signals correlate with EEG signals, suggesting that the different RSNs assemble through synchronization of electric activity as measured by EEG
EEG/fMRI investigations, reporting an association between slow hemodynamic changes and faster electrical oscillations (up to 80 Hz), indicate a link between the ongoing connections that can be detected at different temporal scales. Notably, studies based on micro-state EEG provide evidence for a further association between slow resting state fMRI oscillations and neural activity in the scale of hundreds-milliseconds
In this perspective, our findings reinforce the notion that the RSNs emerging from slow brain fluctuations are based on the neural activity of specific cortical substrates that operate at different time scales. These areas, belonging to the same network, might communicate with each other within different temporal scales, ranging from few seconds (timescale for resting-state fMRI oscillations) to few millisecond (timescale for TMS measurements).
A crucial issue that still remains to be clarified concerns how the activity of these connections actually contributes to specific mental activities
In the present study, although the correlations of physiological interaction and fMRI functional connectivity were strongly significant, they were in the reversed direction than predicted: the lower the Z values of functional connectivity, the higher the facilitation induced by ipsilateral PPC stimulation in the bi-focal condition and the lower the inhibition induced by contralateral PPC stimulation in the tri-focal condition. With this regards, it has to be noted that these physiological interactions involve complex polysynaptic pathways mediated by inhibitory interneurons that could easily reverse the relationship between the physiological activity and the functional connectivity. Therefore it is difficult to provide a solid hypothesis to explain such apparent discrepancy.
It is also worth commenting on the inconsistency between the results of ICA and of the seed-based analyses. While strong correlations were found between TMS and DAN when using ICA decomposition, we were unable to demonstrate associations between TMS data and RS functional connectivity estimated by correlation analysis (see
A limitation of the current study is that TMS and fMRI experiments were not simultaneously performed. It has indeed been argued that RSNs are non-stationary and may change within and between experimental sessions. In this view, the state of the subject might potentially alter the effects of TMS on EMG as well as on the connectivity of networks. For example, alertness, drowsiness, fatigue and so on could have considerable influence on the responsiveness as well as on the network activity and their spatial pattern.
In conclusion, we demonstrate that resting state fMRI may represent an effective method to investigate specific neurophysiological circuits and to quantify the degree of neuronal functional connectivity within them. We showed that combining multimodal TMS and resting state fMRI can effectively improve the characterization of the anatomo-functional properties of some crucial brain connections. Multifocal TMS, which is able to provide unique information concerning the inhibitory or facilitatory nature of a certain pathway with a sub-millisecond temporal precision, can be complemented by the high spatial resolution afforded by fMRI, thus allowing to detect more precisely the cortical nodes underlying functional connectivity. This approach could also strengthen the clinical use of resting state fMRI in several physiological and pathological conditions
(TIF)