Nested oscillations and brain connectivity during ... · 34 between brain regions (Corbetta and...

1

Nested oscillations and brain connectivity during 1

sequential stages of feature-based attention 2

3

4

5

Mattia F. Pagnotta1,*, David Pascucci1,2, Gijs Plomp1 6

7

8 1 Perceptual Networks Group, Department of Psychology, University of Fribourg, Fribourg, 9

Switzerland 10 2 Laboratory of Psychophysics, Brain Mind Institute, School of Life Sciences, École Polytechnique 11

Fédérale de Lausanne (EPFL), Lausanne, Switzerland 12

13

14 * Corresponding author 15

Email: [email protected] (M.F.P.) 16

17

.CC-BY-NC-ND 4.0 International licensepreprint (which was not certified by peer review) is the author/funder. It is made available under aThe copyright holder for thisthis version posted February 29, 2020. . https://doi.org/10.1101/2020.02.28.969253doi: bioRxiv preprint

https://doi.org/10.1101/2020.02.28.969253

http://creativecommons.org/licenses/by-nc-nd/4.0/

2

Abstract 18

Brain mechanisms of visual selective attention involve both local and network-level activity changes 19

at specific oscillatory rhythms, but their interplay remains poorly explored. Here, we investigate 20

anticipatory and reactive effects of feature-based attention using separate fMRI and EEG recordings, 21

while participants attended to one of two spatially overlapping visual features (motion and 22

orientation). We focused on EEG source analysis of local nested oscillations and on graph analysis of 23

connectivity changes in a network of fMRI-defined regions of interest, and characterized a cascade of 24

attentional effects and their interplay at multiple spatial scales. We discuss how the results may 25

reconcile several theories of selective attention, by showing how β rhythms support anticipatory 26

information routing through increased network efficiency and β-γ coupling in functionally specialized 27

regions (V1 for orientation, V5 for motion), while reactive α-band desynchronization patterns and 28

increased α-γ coupling in V1 and V5 mediate stimulus-evoked processing of task-relevant signals. 29


https://doi.org/10.1101/2020.02.28.969253


3

1. Introduction 30

Visual selective attention enables us to prioritize the processing of behaviorally relevant stimuli while 31

filtering out irrelevant ones. This fundamental function of the human brain is supported by changes in 32

activity patterns that occur at multiple spatial scales, from local neuronal circuits to global interactions 33

between brain regions (Corbetta and Shulman, 2002; Greenberg et al., 2010; Kastner and Ungerleider, 34

2000; Scolari et al., 2014; Serences and Yantis, 2007). A central role in this distributed brain function 35

is played by specific rhythms that coordinate selective processing both at the level of neuronal 36

ensembles and at the macro-scale of multiple brain regions, determining the enhancement and 37

propagation of attention-related signals or the downregulation of irrelevant activity (Antzoulatos and 38

Miller, 2014; Bonnefond and Jensen, 2015; Buschman et al., 2012; Buzsáki and Draguhn, 2004; 39

Canolty et al., 2007; Haegens et al., 2011a; Kopell et al., 2000; von Stein and Sarnthein, 2000). 40

A growing body of evidence suggests that distinct neuronal rhythms may serve distinct functional 41

roles in selective attention. Activity over an extended frequency range around the alpha rhythm (α, 7–42

14 Hz), for instance, has been widely linked to the prevention or inhibition of task-irrelevant signals 43

(Chelazzi et al., 2019; Foster and Awh, 2019; Van Diepen et al., 2019; but see Noonan et al., 2016; 44

Schroeder et al., 2018). A potential mechanism by which α rhythms would gate selective attention is 45

through the pulsed inhibition of ongoing cortical activity, by providing phasic bursts of inhibition that 46

suppress the functional processing and inter-areal communication in the gamma frequency range (γ, 47

above 30 Hz) (Jensen and Mazaheri, 2010). In support of this, several studies have recently 48

documented an inverse relationship between α-band feedback signaling from attentional control 49

regions and feedforward γ-band activity in sensory cortices, revealing top-down, attention-related α-50

modulations that interfere with local structures of α-γ phase-amplitude coupling (PAC) in sensory 51

areas (Bonnefond and Jensen, 2015; Haegens et al., 2011b; Mathewson et al., 2011; Mazaheri and 52

Jensen, 2010; Pascucci et al., 2018; Popov et al., 2017). At a relatively higher range of frequencies, 53

beta rhythms (β, 15–30 Hz) have been instead related to the instantiation of distributed task-relevant 54

representations that convey context- and content-specific information (Antzoulatos and Miller, 2016, 55

2014; Buschman et al., 2012; Richter et al., 2018; Spitzer and Haegens, 2017). At the same time, β-56

band activity has been also shown to promote feedforward and inter-areal γ-band synchronization 57

(Richter et al., 2017), while preventing interference and irrelevant attentional shifts (Fiebelkorn and 58

Kastner, 2019). Thus, distinct rhythms appear to fulfil separate but complementary roles in selective 59

attention, by differentially modulating local activity, structures of cross-frequency coupling and 60

neuronal communication in large-scale networks. 61

A further distinction exists as to whether specialized rhythms support the anticipatory (pre-stimulus) 62

and reactive (post-stimulus) components of selective attention. Pre-stimulus synchronization of α-band 63

activity has been mostly described in relation to anticipatory, proactive-like mechanisms to filter-out 64

irrelevant features and locations (Foxe and Snyder, 2011; Kelly et al., 2006; Oliveira et al., 2014; 65


https://doi.org/10.1101/2020.02.28.969253


4

Snyder and Foxe, 2010; Worden et al., 2000; but see Noonan et al., 2016), whereas post-stimulus, α-66

band event-related desynchronization (ERD) has been linked to the release from inhibition (Klimesch 67

et al., 2007), which in turn facilitates local and feedforward γ-band activity conveying task-relevant 68

signals (Bonnefond et al., 2017; Pascucci et al., 2018; Popov et al., 2017). Conversely, post-stimulus 69

increases of α-band activity may represent reactive suppression and the return to inhibitory states that 70

prevent task-irrelevant processing (Pascucci et al., 2018). In a similar vein, increases of β-band activity 71

have also been found in preparatory stages of processing (Schneider and Rose, 2016), as top-down 72

modulatory signals that relay task-specific behavioral context (Richter et al., 2018), or in more reactive 73

stages as feedback control mechanisms that facilitate the bottom-up communication of attended 74

stimuli (Bastos et al., 2015). Crucially, most of these phenomena have been investigated separately 75

and the exact nature of their interplay, as well as the complexity of local and large-scale neuronal 76

dynamics involved in selective attention, remains poorly understood. 77

In the present work, we leveraged different spatial scales of neuronal activity to characterize the 78

temporal dynamics of local and network-level changes during anticipatory and reactive stages of 79

attentional processing. At the local level, we focused on the phenomenon of nested oscillations 80

(Bonnefond et al., 2017), hypothesizing functional differences in the coupling between α-β and γ 81

rhythms depending on the anticipation and processing of relevant or irrelevant sensory signals. At the 82

network level, we exploited measures of directed connectivity among distributed attentional and 83

sensory areas, with the goal to relate large-scale functional interactions to local changes in nested 84

oscillations. 85

To this end, we acquired separate EEG and fMRI recordings while human participants were 86

performing basic visual tasks under different feature-based attention conditions. We presented two 87

spatially overlapping features (Baldauf and Desimone, 2014), motion and orientation, and we 88

manipulated participants attention toward each of them in separate blocks, probing both anticipatory 89

and reactive feature-specific attentional responses. Our results revealed a cascade of attentional effects 90

at multiple spatial scales, from anticipatory β-band coupling and connectivity changes involving 91

functionally specialized regions (e.g., V1 and V5), to distinct local and global post-stimulus dynamics 92

initiated by α-band desynchronization. 93


https://doi.org/10.1101/2020.02.28.969253


5

2. Results 94

2.1. Attention modulates visual-evoked potentials 95

The behavioral task required the participants to attend visual stimuli presented at central location 96

(Figure 1A), and to discriminate either the motion direction of signal-dots in the Random Dot 97

Kinematograms (RDK; motion discrimination task), or the off-vertical tilt of the Gabor (orientation 98

discrimination task). Hereafter, these two tasks conditions are named Attended-motion and Attended-99

orientation, respectively. In both Attended conditions, the stimuli contained both features (motion and 100

orientation). In a third task condition that served as control, participants were asked to report sporadic 101

color changes in the fixation spot, which rendered the same stimuli irrelevant for the task (Unattended 102

condition), without changing their physical characteristics. We calibrated task-difficulty beforehand 103

(see Methods) and did not observe differences in behavioral performance between the two 104

discrimination tasks. Percentage correct was 89.0% (SD=5.2) and 88.6% (SD=7.2) for Attended-105

motion and Attended-orientation respectively (t=0.19, p=0.8485), with corresponding reaction times 106

(RTs) of 687.5 ms (SD=159.1) and M=699.5 ms (SD=141.0; t=-0.62, p=0.5432). In the Unattended 107

control condition, participants correctly detected a color change in the fixation spot in 99.2% of target 108

trials (SD=1.4), with RT of 484.4 ms (SD=46.6). 109

110 Figure 1. Experimental paradigm and VEPs results. A) Schematic representation of trial structures for Attended (motion 111 or orientation discrimination) and Unattended (detection of color changes in the fixation spot). The white arrows on top of the 112 visual stimuli indicate the direction of coherent motion and were not present during the experiment. The red fixation spot is 113 enlarged for illustrative purpose, it did not change size during color changes. B) Show grand-average VEPs from three 114 electrode clusters: left occipito-temporal (lOT); centro-occipital (CO); right occipito-temporal (rOT). Electrode locations of 115 each cluster are highlighted in red in the corresponding scalp representation (on the left). The VEPs in the Attended-motion 116 (ATT-MOT) and Attended-orientation (ATT-ORI) conditions are shown in orange and purple, respectively; while, VEPs in 117 the Unattended condition (UNATT) are shown in blue. For each condition, the shading represents the standard error of the 118 mean. Green vertical shades represent time intervals that showed statistically significant results in the repeated-measures 119 ANOVA (pFDR<0.05). Results of repeated-measures ANOVA and of post-hoc analyses are provided in Table 1. C) The 120 figures show the scalp topographical distribution of the differences between Attended-motion and Unattended, at the time 121 intervals derived from the ANOVA results (top to bottom). 122


https://doi.org/10.1101/2020.02.28.969253


6

Attention has been shown to modulate the amplitude of stimulus-evoked P1 component, at around 100 123

ms post-stimulus onset (Hillyard et al., 1998; Pascucci et al., 2018; Zhang and Luck, 2009). 124

Comparing Attended conditions with the Unattended showed an attentional P1 increase over posterior 125

electrodes (Figure 1B). Post-hoc analyses revealed significant P1 amplitude increases for both 126

Attended-motion and Attended-orientation in left occipito-temporal (lOT) and centro-occipital (CO) 127

electrode clusters. In right occipito-temporal (rOT) electrodes, P1 amplitude was significantly higher 128

than Unattended only for Attended-orientation (Table 1). 129

Table 1. 130 VEPs analysis; results of repeated measures ANOVA and post-hoc tests. 131 Sensor-space macro area Time-interval F-statistic Significant post-hoc comparisons

left occipito-temporal (lOT)

86–106 ms p=0.0214 (F=4.83) ATT-MOT > UNATT: p=0.0043 (t=3.27)

ATT-ORI > UNATT: p=0.0100 (t=2.88)

182–200 ms p=0.0229 (F=4.52) ATT-ORI > UNATT: p=0.0059 (t=3.12)

482–500 ms p=0.0269 (F=4.20) ATT-ORI < UNATT: p=0.0084 (t=-2.96)

centro-occipital (CO)


ATT-ORI > UNATT: p=0.0017 (t=3.68)

420–500 ms p=0.0024 (F=17.32) ATT-MOT < UNATT: p<0.0001 (t=-5.03)

ATT-ORI < UNATT: p<0.0001 (t=-5.10)

right occipito-temporal (rOT)

84–96 ms p=0.0308 (F=3.89) ATT-ORI > UNATT: p=0.0034 (t=3.37)


444–494 ms p=0.0096 (F=6.26) ATT-ORI < UNATT: p=0.0011 (t=-3.89)

Note. ATT-MOT and ATT-ORI refer to the Attended-motion and Attended-orientation conditions, respectively. UNATT refers 132 to the Unattended condition. 133

The results further showed attentional modulations of the P2 component (around 200 ms after stimulus 134

onset), with significantly increased amplitudes for Attended-orientation in lOT (p=0.0059), and for 135

Attended-motion in rOT (p=0.0266). The effects replicate previous findings of P2 modulations for 136

target stimuli (Kanske et al., 2011; Maeno et al., 2004; Van Voorhis and Hillyard, 1977). At latencies 137

beyond 400 ms, significant differences in VEPs were mostly observed over CO, with decreased 138

amplitude for both Attended conditions (Figure 1B, Table 1). The sign and topographical distribution 139

of these late differences resembled the posterior selection negativity (Figure 1C) (Hillyard and Anllo-140

Vento, 1998). While selection negativity is usually observed from 150-200 ms after stimulus onset 141

with static stimuli (Daffner et al., 2012; Hillyard and Anllo-Vento, 1998; Pascucci et al., 2018), our 142

use of moving stimuli with multiple features may explain the late occurrence. 143

2.2. Attention dynamically modulates brain rhythms 144

To investigate how attention modulates brain rhythms before and after stimulus onset, we compared 145

spectral power distributions between attention conditions. Across the pre-stimulus interval (-300–0 146

ms), we found that attention significantly enhanced power spectra. The anticipatory increases for the 147

Attended conditions were most prominent in the β-band and distributed over lateral and occipito-148


https://doi.org/10.1101/2020.02.28.969253


7

temporal electrodes (Figure 2A). To localize these anticipatory effects to underlying cortical areas, we 149

next performed the power-spectral analysis on source-reconstructed signals. We used EEG source 150

imaging based on individual lead-fields to extract the signals from 22 cortical regions of interest 151

(ROIs) across the brain (Oostenveld et al., 2011; Rubega et al., 2019; Van Veen et al., 1997). The 152

ROIs were derived from fMRI cluster-peaks of significant differences, obtained with the same tasks 153

and participants (Figure 3, Table 2) (see Methods). The EEG source-space results showed that the 154

anticipatory β-band power increases in Attended conditions mainly originate from lateral visual areas 155

(Figure 2B). 156

157 Figure 2. Attentional modulation of brain rhythms: comparison between Attended-motion and Unattended. Results of 158 sensor-space power spectra analysis are shown for anticipatory (A) and reactive (C) time intervals. Each figure represents the 159 distribution of statistically significant differences (positive t-values for Attended-motion higher than Unattended, and vice 160 versa negative t-values) from cluster-based permutation (two-tailed t-test with p<0.05, 50000 permutations, and p<0.05 for 161 the permutation test), together with the topographical distributions of these differences in the α-band (bottom) and β-band 162 (top). Here and in the following, red indicate positive differences between Attended and Unattended conditions (i.e., 163 increases with attention), while blue indicates negative differences (i.e., decreases with attention). B) The polar histogram 164 represents the effect sizes for each ROI of source-space power differences between conditions in the β-band (15–30 Hz). D) 165 The figure shows the results of source-space time-varying power spectra analysis, with the distribution of statistically 166 significant differences from cluster-based permutation (two-tailed t-test with p<0.05, 50000 permutations, and p<0.05 for the 167 permutation test). The marginal plots show time- or frequency-collapsed distributions of significant differences in power 168 spectra. The polar histograms on the right represent the effect sizes for each ROI of power differences between conditions, in 169 the following frequency bands: α (5–12 Hz), β (15–30 Hz), and γ (45–100 Hz). Effect sizes were estimated using Cohen's d 170 (Cohen, 1992). Analogous results were obtained when comparing Attended-orientation to Unattended (Figure S1). 171


https://doi.org/10.1101/2020.02.28.969253


8

172 Figure 3. fMRI results and ROIs. A) Results of the voxel-wise t-test comparison across participants for the contrast 173 Attended vs. Unattended; positive and negative results from the comparison are shown in the red and blue color scales, 174 respectively. B) Results of the voxel-wise t-test comparison across participants for the contrast Attended vs. Rest; voxels with 175 higher activity in both Attended conditions compared to Rest are shown in the yellow color scale. C) Centroids of the 22 176 ROIs identified from the fMRI results. 177

Table 2. 178 Details about the 22 regions of interest (ROIs). 179 Anatomical region ROI abbreviation Coordinates of centroid (MNI-space)

x y z

left premotor area PM L -43.60 -6.40 41.20

right premotor area PM R 42.61 -6.00 47.13

left middle frontal gyrus MFG L -32.00 30.00 42.00

right middle frontal gyrus MFG R 34.29 28.29 42.29

left inferior frontal gyrus IFG L -45.00 27.13 6.75

right inferior frontal gyrus IFG R 44.29 18.29 25.14

supplementary motor area SMA -2.50 -0.17 59.67

middle cingulate cortex MCC 10.00 -24.00 42.00

left calcarine sulcus V1 L -4.29 -93.14 2.29

right calcarine sulcus V1 R 12.00 -92.00 2.00

left cuneus Cuneus L -6.67 -81.00 26.33

left secondary visual cortex V2 L -36.47 -83.06 -3.76

right secondary visual cortex V2 R 35.00 -86.00 -7.00

left fusiform gyrus FFG L -35.00 -62.67 -13.67

left gyrus postcentralis GPC L -33.33 -36.00 51.33

left superior parietal lobule SPL L -10.00 -50.29 45.14

right superior parietal lobule SPL R 18.00 -63.50 52.50

left inferior parietal lobule IPL L -55.33 -34.67 32.67

right inferior parietal lobule IPL R 49.41 -48.94 37.41

left middle temporal area MTG L -55.00 -56.00 13.00

right visual cortex V5 V5 R 53.33 -61.00 1.33

right middle temporal area MTG R 57.50 -37.25 -5.75

Note. In the column about ‘ROI abbreviation’: L and R stand for left and right, respectively. 180


https://doi.org/10.1101/2020.02.28.969253


9

In the post-stimulus interval, we found greater decreases in α and β-band power for the Attended 181

conditions across widespread electrodes (Figure 2C). These effects reflect an ERD (Minami et al., 182

2014; Pfurtscheller and Lopes da Silva, 1999; Schmiedt et al., 2014; Yordanova et al., 2001) that was 183

larger for attended than unattended stimuli (Mazaheri and Picton, 2005; Pascucci et al., 2018). We 184

used time-varying spectral analysis in the source-space and showed that significant differences 185

between Attended and Unattended conditions emerged shortly after stimulus onset, with stronger ERD 186

in the α and β-band for Attended in fronto-parietal regions (Figure 2D). 187

Taken together, the results showed that attention modulates local brain rhythms in distinct ways before 188

and after stimulus onset. Anticipatory effects involved attentional increase of β-band power in lateral 189

visual areas, while reactive effects involved stronger α and β-band ERD in parietal and frontal regions 190

due to attention. Spectral power changes occurred irrespective of the attended feature (motion or 191

orientation). These results suggest a dynamic reorganization of distributed activity in distinct 192

frequency bands, depending on whether a stimulus is attended or not. We further investigated the pre- 193

and post-stimulus effects of attentional processing using network and PAC analyses. 194

2.3. Anticipatory stage: increased large-scale network efficiency 195

To identify anticipatory network changes due to attention, we estimated the frequency-specific 196

directed connections between the 22 ROIs using a multivariate functional connectivity measure, the 197

information partial directed coherence (iPDC) (Baccalá and Sameshima, 2014; Takahashi et al., 2010). 198

To describe and compare the topological properties of the resulting networks in different frequencies, 199

we used two network measures: global efficiency and local efficiency (Latora and Marchiori, 2001; 200

Rubinov and Sporns, 2010). Given the ability of the iPDC to characterize both directionality and 201

directness of inter-areal connections and because of its information-theoretic foundation (Baccalá and 202

Sameshima, 2014; Takahashi et al., 2010), these functional connections can be seen as paths of 203

integration between network’s ROIs. The iPDC-derived network’s global efficiency represents the 204

frequency-specific level of global functional integration among all regions. Likewise, the network’s 205

local efficiency represents the level of functional integration within subgraphs of neighbor regions, 206

defined by strongest functional connections, while the local efficiency of each single ROI measures 207

the local integration within its own subgraph only (see Methods). 208

Topological network analysis revealed that attention to motion significantly increased both global and 209

local efficiency of the network in the β (15–21 Hz) and high-γ band (around 60 Hz) (Figure 4A). We 210

found analogous increases for Attended-orientation (Figure 4B), where statistically significant 211

differences in the γ-band spanned a larger frequency range (57–71 Hz) and we observed an additional 212

increase in global efficiency at 33-34 Hz. The results suggest that anticipatory network effects could 213

enhance large-scale network communication and information routing in the β and γ-band, by changing 214

network topology at these frequencies to favor functional integration both globally and within 215

functional subgraphs. 216


https://doi.org/10.1101/2020.02.28.969253


10

217 Figure 4. Anticipatory differences in network efficiency. Results for network’s global efficiency and local efficiency, for 218 the comparisons: A) Attended-motion vs. Unattended; B) Attended-orientation vs. Unattended. Differences between 219 conditions are evaluated as a function of frequency for global (left) and local efficiency (right). Gray shading represents the 220 standard error of the mean. Green vertical shades represent the frequencies that showed statistically significant results in a 221 two-tailed t-test (pFDR<0.05). 222

2.4. Anticipatory stage: local changes in β-γ coupling 223

Before stimulus onset we found that attention enhanced network functional integration in the β and γ-224

band. The activity from the two frequency bands may be integrated locally through mechanisms of 225

cross-frequency coupling (Bonnefond et al., 2017; Buzsáki, 2006; Canolty et al., 2006; Canolty and 226

Knight, 2010; Jensen and Colgin, 2007), we thus investigated whether attention modulates the within-227

region PAC (Martínez-Cancino et al., 2019; Penny et al., 2008). For this analysis, we selected 9 ROIs 228

in occipito-temporal cortex based on the results of power spectra analyses, and in line with previous 229

studies that reported modulations of low-frequency activity and cross-frequency coupling in occipital 230

cortex (Bastos et al., 2015; Bonnefond and Jensen, 2015; Pascucci et al., 2018; Schmiedt et al., 2014). 231

The results showed that attention to motion selectively increased β-γ PAC in region V5-R (fPHASE: 24–232

27 Hz; fAMPLITUDE: 54–80 Hz) (Figure 5A). For Attended-orientation, we found a significant β-γ PAC 233

increase at high amplitude frequencies (fPHASE: 23–25 Hz; fAMPLITUDE: 76–86 Hz) and decrease at lower 234

frequencies (fPHASE: 19–20 Hz; fAMPLITUDE: 44–56 Hz) in region V1-L, and increased PAC in region 235

MTG-L (fPHASE: 20–22 Hz; fAMPLITUDE: 46–62 Hz) (Figure 5B). Furthermore, we observed a significant 236

β-γ PAC increase in region V2-R and decrease in region V2-L, for both Attended conditions compared 237

to Unattended. The anticipatory PAC effects could not be attributed to power differences in the 238

signals, as shown by a control analysis based on stratification (Aru et al., 2015; Oostenveld et al., 239

2011) (Figure S2). In sum, while certain regions showed general attentional modulations of PAC (V2-240

L, V2-R), others showed effects that were specific of the attended feature: V5-R for motion, V1-L and 241

MTG-L for orientation. 242


https://doi.org/10.1101/2020.02.28.969253


11

243 Figure 5. Anticipatory differences in local PAC and correlation with network’s local changes. Differences in within-244 region anticipatory PAC are shown for each pair fPHASE– fAMPLITUDE, for Attended-motion compared to Unattended (A) and for 245 Attended-orientation compared to Unattended (B). The figure shows regions with significant differences from the cluster-246 based permutation comparison (two-tailed t-test with p<0.05, 50000 permutations, and p<0.05 for the permutation test), 247 which are highlighted as more opaque. Black areas indicate fPHASE–fAMPLITUDE pairs excluded from the analysis (see Methods). 248 Figures C–E report the within-region correlation between variations in PAC (ΔPAC) and variations in single-node’s local 249 efficiency (ΔElocal), for regions that showed a significant correlation (p<0.05): C) V2-R for Attended-motion; D) V5-R for 250 Attended-motion; E) V1-L for Attended-orientation. Each figure reports the Spearman’s rho and the p-value for testing the 251 null-hypothesis of no correlation (inside brackets). 252

2.5. A link between PAC and local efficiency modulations 253

Before stimulus onset, we found both enhanced large-scale network integration in the β and γ-band 254

(Figure 4) and increased β-γ PAC in specific occipito-temporal regions (Figure 5A–B). Therefore, we 255

next asked whether in these regions there is a linear relation between the local cross-frequency 256

coupling and their level of functional integration within the network. In each ROI, we evaluated the 257

relationship between changes of PAC (ΔPAC) and changes of its local efficiency (ΔElocal) due to 258

attention (see Methods). The results showed a significant negative correlation in region V2-R between 259

β-γ ΔPAC and β-band ΔElocal for the comparison between Attended-motion and Unattended (rho=-260

0.608, p=0.009, number of outliers = 1) (Figure 5C), which means that when β-band local efficiency 261

increased with attention, the local β-γ PAC decreased. For the same comparison, we found a reversed 262

relationship in region V5-R, where β-γ ΔPAC correlated positively to both β-band ΔElocal (rho=0.595, 263

p=0.011, number of outliers = 1) and γ-band ΔElocal (rho=0.525, p=0.027, number of outliers = 1) 264

(Figure 5D). These effects reveal that in V5-R when local efficiency increased with attention in the β 265

and γ-band, the local β-γ PAC increased too, across subjects. From comparing Attended-orientation 266


https://doi.org/10.1101/2020.02.28.969253


12

and Unattended, a significant positive correlation between β-γ ΔPAC and β-band ΔElocal was observed 267

in region V1-L (rho=0.519, p=0.024, number of outliers = 0) (Figure 5E). No statistically significant 268

effects were observed in the other ROIs tested. 269

It is worth highlighting that a significant positive correlation between β-γ ΔPAC and β-band ΔElocal 270

was found in V5-R when motion was attended and in early visual area V1-L when orientation was 271

attended, which are respectively known to be sensitive to motion coherence and Gabor’s orientation 272

(Ahlfors et al., 1999; Bosking et al., 1997; Koelewijn et al., 2011; Schoenfeld et al., 2007; Simoncelli 273

and Olshausen, 2001). The effects therefore suggest that, only in functionally specialized regions, the 274

anticipatory within-region increases of local cross-frequency coupling are linked to the increased 275

levels of functional integration among neighboring areas of that region. 276

2.6. Reactive stage: dynamic local changes in α-γ coupling 277

We found larger stimulus-evoked α and β-band desynchronization for attended stimuli (Figure 2). We 278

therefore investigated whether α-γ PAC and β-γ PAC were modulated by reactive attentional 279

processing. Our results showed significantly increased α-γ PAC for Attended-motion in regions V5-R 280

and MTG-R (Figure 6A). The α-γ PAC increase in V5-R started around 46 ms post-stimulus onset and 281

lasted up to approximately 100 ms, involving high γ-band frequencies (64–100 Hz). The significant α-282

γ PAC increase in MTG-R occurred between 66 and 124 ms, modulating frequencies in the range 72–283

100 Hz. For Attended-orientation, we found that α-γ PAC in V1-L was significantly increased 284

compared to Unattended, at latencies and frequencies in the ranges 86–130 ms and 62–100 Hz, 285

respectively (Figure 6B). A recent study showed a similar reactive increase in α-γ PAC using an 286

orientation discrimination task (Pascucci et al., 2018). No statistically significant effects were 287

observed in the other ROIs tested, and no correlations between changes of within-region α-γ PAC and 288

local efficiency were observed in post-stimulus time. 289

290 Figure 6. Stimulus-evoked differences in local α-γ PAC. The temporal evolution of differences in PAC are shown for the 291 coupling between α-band phase (10 Hz for Attended-motion compared to Unattended, and 8 Hz for Attended-orientation 292 compared to Unattended) and γ-band amplitude (50–100 Hz, in 2Hz-steps). Panel A) shows Attended-motion compared to 293 Unattended, B) Attended-orientation compared to Unattended. Result were obtained via cluster-based permutation (two-294 tailed t-test with p<0.05, 50000 permutations, and p<0.05 for the permutation test), statistically significant results are 295 highlighted as more opaque. 296

It should be noted that the reactive attentional increases in α-γ PAC involved the same functionally 297

specialized regions that showed anticipatory increases in β-γ PAC (V5-R for Attended-motion and V1-298

L for Attended-orientation), positively correlated at between-subject level with β-band ΔElocal (Figure 299


https://doi.org/10.1101/2020.02.28.969253


13

5). We did not find, however, any significant differences in stimulus-evoked β-γ PAC between 300

Attended and Unattended conditions. Taken together, this suggests that the anticipatory and reactive 301

stages of attentional processing may be distinctively mediated by β and α rhythms. 302

2.7. Reactive stage: fast dynamics of network efficiency 303

We next asked how attentional processing affects large-scale network topology after stimulus onset. 304

To dynamically investigate stimulus-evoked network changes, we employed time- and frequency-305

resolved estimates of functional connectivity (iPDC) between the 22 ROIs (Pascucci et al., 2019; 306

Takahashi et al., 2010). The results showed significantly diminished global and local efficiency of the 307

network for Attended-motion compared to Unattended in both α and β-band, and from latencies 308

around 270 ms and 300 ms, respectively (Figure 7). These low-frequency differences in network 309

topology were preceded by a significant increase for Attended-motion of both global and local 310

efficiency in the γ-band, at around 160 ms post-stimulus onset (Figure 7C). Similar results were 311

obtained by comparing Attended-orientation and Unattended, where we found significantly decreased 312

global and local efficiency of the network at low-frequencies (α and β-band), roughly from 240 ms 313

post-stimulus onset, preceded by a significant γ-band increase for Attended-orientation around 180 ms 314

post-stimulus onset in both efficiency measures (Figure S3). 315

316 Figure 7. Stimulus-evoked differences in network efficiency. Differences between Attended-motion and Unattended are 317 shown for three frequency bands, A) α-band, B) β-band, and C) γ-band. Left panels show network’s global efficiency, right 318 panels show network’s local efficiency. Gray shading represents the standard error of the mean. Green vertical shades 319 represent latencies that showed statistically significant results in the two-tailed t-test (pFDR<0.05). 320


https://doi.org/10.1101/2020.02.28.969253


14

Reactive network changes occurred in rapid succession: attention first enhanced efficiency measures 321

in the γ-band (~150–200 ms) and then reduced them in the α and β-band (after 240 ms post-stimulus 322

onset). The low-frequency effects at longer latencies indicate that the evoked α and β-band 323

desynchronization (Figure 2) co-occurs with network-level changes that reduce the functional 324

integration both globally and within subgraphs, favoring more local processing. 325

3. Discussion 326

In the present work, we sought to characterize local and large-scale neuronal dynamics during 327

anticipatory and evoked stages of feature-based selective attention. To this aim, we investigated local 328

cross-frequency coupling and brain-wide directed connectivity during a visual discrimination task that 329

required the attentional selection of one of two spatially overlapping features. Our results show distinct 330

patterns of nested oscillations and functional network topologies that precede and follow stimulus 331

onset. More precisely, we found evidence for a dominant role of β rhythms in the anticipatory stage, 332

followed by stimulus-evoked α-band desynchronization patterns and α-driven structures of nested 333

oscillations. The use of a control condition and the presentation of overlapping features at a single 334

location also allowed us to disentangle between general effects of stimulus relevance and specific 335

effects of the attended feature (motion or orientation). Figure 8 summarizes the findings, highlighting 336

the sequential evolution of local and network-level phenomena with respect to stimulus onset. 337

338 Figure 8. Temporal sequence of the effects induced by visual selective attention. Red (blue) patches indicate attentional 339 increases (decreases), as do upwards (↑) and downwards (↓) arrows. The abbreviation Pow refers to the spectral power. 340

As evident from Figure 8, feature-based selective attention was mediated by a stimulus-induced 341

reconfiguration of both local and large-scale neuronal interactions. Before stimulus onset, increased 342

levels of network integration in the β and γ-band were paralleled by local increases of β-γ coupling in 343

occipito-parietal regions, and the two phenomena correlated across participants. After stimulus onset, 344


https://doi.org/10.1101/2020.02.28.969253


15

brain activity underwent a marked reconfiguration, with robust desynchronization of α rhythms and 345

increased α-γ coupling in feature-selective regions (V5 and MTG for motion discrimination, V1 for 346

orientation discrimination). 347

The observed cascade of attention-related effects suggests that different mechanisms, involving the 348

interplay between large-scale dynamics and nested oscillations, serve distinct aspects of selective 349

attention. Pre-stimulus β-band activity may reflect an increase in functional communication among 350

regions involved in the upcoming task, with a consequent boost of high-frequency activity in 351

functionally specialized areas (Buzsáki and Wang, 2012). Interpreting the role of β rhythms in the 352

context of attention and cognitive processing, however, is not trivial, given the large body of work 353

traditionally relating these rhythms to motor control (Pfurtscheller et al., 1996; Pogosyan et al., 2009; 354

Sanes and Donoghue, 1993; Swann et al., 2009), sensorimotor functions (Lalo et al., 2007) and 355

working memory (Lundqvist et al., 2016; Schneider and Rose, 2016; Siegel et al., 2009). A recent line 356

of evidence has associated β rhythms with the maintenance of the status quo, that is, the preservation 357

of ongoing sensorimotor states and cognitive sets (Engel and Fries, 2010). Following this line, other 358

studies have hypothesized a role for β-band synchronization in establishing flexible networks of 359

neuronal ensembles that carry content- and task-specific representations (Antzoulatos and Miller, 360

2016, 2014; Buschman et al., 2012; Spitzer and Haegens, 2017). Our results appear in agreement with 361

this latter definition and extend it, by showing that large-scale network interactions associated with β 362

rhythms may control endogenous mechanisms of attention before stimulus onset, likely conveying 363

task-related information to downstream areas (e.g., the type of stimulus to attend/ignore, the type of 364

response required) (Richter et al., 2017). 365

It is relevant to note that, contrary to previous reports (see Chelazzi et al., 2019; Foster and Awh, 366

2019; Snyder and Foxe, 2010 for reviews and discussions), we found no evidence of anticipatory 367

modulations of α-band activity in our data. While, at first, this seems to support the alternative view of 368

α rhythms as a secondary type of inhibitory mechanism, observable only in relation to target 369

processing (e.g., at ipsilateral sides of relevant stimuli; Chelazzi et al., 2019; Foster and Awh, 2019), 370

the fact that our analysis always involved subtracting a control condition with identical stimulation 371

could have hindered anticipatory increases in α power reflecting aspecific preparatory stages and 372

dynamic reallocations of attentional resources (van Diepen and Mazaheri, 2017). 373

A clear modulation of α-band activity was evident shortly after stimulus onset. Alpha 374

desynchronization (Mazaheri and Picton, 2005; Pfurtscheller and Lopes da Silva, 1999) was indeed the 375

main component determining the observed network reconfiguration, accompanied by local increases 376

of α-γ coupling in relevant sensory regions (V5 and MTG for motion discrimination, V1 for 377

orientation discrimination). This latter phenomenon is in line with recent accounts hypothesizing a role 378

of α rhythms in aligning the phase of high-frequency oscillations carrying relevant processing 379

(Bonnefond et al., 2017; Pascucci et al., 2018). Under this view, the decrease of α power resulting 380

from the evoked desynchronization would allow longer windows of neuronal excitability, favoring 381


https://doi.org/10.1101/2020.02.28.969253


16

local increases and feedforward propagation of γ-band activity (Bonnefond et al., 2017). Thus, the co-382

occurrence of α desynchronization and α-γ coupling may underlie the emergence of short functional 383

windows where the release from inhibitory signals enhances sensory and task-relevant processing. In 384

this respect, our results are also in agreement with the gating-by-inhibition hypothesis that postulates a 385

similar inverse relationship between the degree of α-band synchronization and the level of local 386

neuronal excitability and γ-band activity (Bonnefond and Jensen, 2015; Haegens et al., 2011b; Jensen 387

and Mazaheri, 2010; Mathewson et al., 2011; Mazaheri and Jensen, 2010). 388

As a means to describe functional network properties, we used graph efficiency measures that indicate 389

how easily signals can travel between network nodes, i.e., the degree of local and global integration, 390

depending on the configuration of directed functional connections. The analysis of network efficiency 391

revealed a sequence of stimulus-evoked brain dynamics involving α-band activity: an initial rapid 392

desynchronization of α rhythms with increased α-γ coupling at the local level (~50–120 ms post-393

stimulus), followed by the rise of large-scale γ-band interactions, with enhanced network efficiency in 394

the γ range (150–200 ms post-stimulus). The transition between local and distributed γ-band activity, 395

and the increased efficiency of inter-areal high-frequency communication occurred at latencies that are 396

consistent with stimulus processing time and early attentional effects (Figure 1) (Hillyard et al., 1998; 397

Hillyard and Anllo-Vento, 1998), thus, they may reflect the dynamics by which attentional selection 398

renders stimulus content accessible to widespread circuits. 399

Our results also suggest distinct mechanisms for general and task-specific attentional effects. General 400

effects were mediated by pre-stimulus changes in spectral power across cortical regions and by 401

topological changes in the brain-wide network of functional interactions. These included increased 402

spectral power in the β-band over lateral-occipital areas and increased network efficiency in the β and 403

γ-band. The results therefore point to an aspecific enhancement of large-scale network communication 404

that facilitates information routing while preparing for an attended stimulus to appear (Buschman and 405

Kastner, 2015; Fries, 2015). After stimulus onset, we found that attention induced desynchronization 406

first in the α and then in β-band (Klimesch et al., 2007, 2001; Mazaheri and Picton, 2005; Pascucci et 407

al., 2018; Schmiedt et al., 2014). These stimulus-evoked effects co-occurred with a sequence of 408

frequency-specific changes in network topology, starting with enhanced network efficiency in the γ-409

band (~150–200 ms), followed by reduced efficiency in the α and β-band (after 240 ms). This later 410

decrease of network integration may suggest a return to more local and sparse computations after the 411

critical time of attention-mediated processing. 412

Task-specific attentional modulations involved more local phenomena, characterized by increased 413

coupling between lower frequencies and γ-band activity. These effects were selective for functionally 414

specialized regions (in V5-MT for motion, in V1 for orientation) during both pre- and post-stimulus 415

time: task-relevant regions showed both anticipatory increases in β-γ coupling and reactive increases 416

in α-γ coupling. Based on the well-known specialization of these regions in processing the two types 417

of stimulus’ features used (Ahlfors et al., 1999; Bosking et al., 1997; Koelewijn et al., 2011; 418


https://doi.org/10.1101/2020.02.28.969253


17

Schoenfeld et al., 2007; Simoncelli and Olshausen, 2001), these results support the view of PAC and 419

nested oscillations as regulatory mechanisms for the local analysis and feedforward communication of 420

relevant sensory signals (Jensen and Mazaheri, 2010; Pascucci et al., 2018). 421

In this work, we provide first evidence of a sequential relationship between anticipatory increases in β-422

γ PAC and stimulus-evoked α-γ coupling. These two local phenomena may be linked to precise 423

endogenous and exogenous components of attentional processing: pre-stimulus anticipatory effects 424

could promote the endogenous reactivation of task-specific content in sensory circuits (Spitzer and 425

Haegens, 2017), whereas post-stimulus coupling may reflect the enhanced processing of relevant 426

exogenous signals (e.g., target stimuli) (Bonnefond and Jensen, 2015; Jensen and Mazaheri, 2010). 427

Theories of communication based on nested oscillations propose that low-frequency carriers in the 428

theta (θ, 3–7 Hz), α or β-band establish inter-areal communication at larger scales, by mediating local 429

excitability (reflected by γ-band activity) (Bonnefond et al., 2017). Cross-frequency coupling at the 430

local level, therefore, would coordinate γ-band activity and information routing at different scales 431

(Buzsáki, 2006; Buzsáki and Wang, 2012; Canolty et al., 2006; Canolty and Knight, 2010; Jensen and 432

Colgin, 2007; Penny et al., 2008; Voytek et al., 2010). Here we show that local integration and 433

network efficiency at the carrier frequency can be also a determinant of local coupling and information 434

flow. This was evident in the task-specific anticipatory increase of β-band local efficiency, which 435

showed a positive relationship with the degree of β-γ PAC enhancement due to attention. Stimulus-436

evoked α-γ PAC, however, did not correlate with network properties, but was instead driven solely by 437

α-band desynchronization. This indicates a different mechanism that may emerge upon the release 438

from inhibition. Adding on the gating-by-inhibition framework (Mazaheri and Jensen, 2010), we 439

found that the effects of the release from inhibitory gating are selective to the upcoming tasks, 440

occurring exclusively in cortical regions encoding task-relevant features. A compelling question for 441

future investigations is whether these distinct mechanisms of local cross-frequency coupling 442

(anticipatory β-driven, reactive α-driven) underlie also the attentional selection based on whole-object 443

representations, for example of faces and houses via selective involvement of the fusiform face area 444

and parahippocampal place area, respectively. 445

For functional connectivity analyses, we employed state-of-the-art multivariate methods (Baccalá and 446

Sameshima, 2014), and a recently developed adaptive filter that was specifically designed for fast-447

changing signals like VEPs (Pascucci et al., 2019). While we considered a large-scale network of 22 448

brain-wide cortical regions for our analyses, one potential limitation is linked to unobserved common 449

inputs (Bastos and Schoffelen, 2016; Pagnotta et al., 2018a). Previous studies revealed that the 450

pulvinar and mediodorsal thalamus play a role in attentional control in non-human primates 451

(Fiebelkorn et al., 2019; Saalmann et al., 2012). We did not include the thalamus as a ROI, due to the 452

EEG intrinsic limitations in accurately reconstructing subcortical activity (although see Seeber et al., 453

2019). Hence, our results provide solely an account for the cortico-cortical interactions in mediating 454

selective attention. In the future, the use of invasive recordings in clinical populations and implanted 455


https://doi.org/10.1101/2020.02.28.969253


18

patients may provide further insights about the role of subcortical structures and thalamo-cortical 456

interactions in controlling attention and other cognitive functions. 457

4. Methods 458

4.1. Participants 459

Twenty healthy human participants (13 female; all right-handed; ages 19–34 y, M=23.25, SD=4.15) 460

with normal or corrected-to-normal vision (visual acuity across participants 1.10–1.63, M=1.40, 461

SD=0.19; Freiburg visual acuity test (Bach, 1996)) took part in the study for monetary compensation 462

(20.– CHF/hour). The study was performed in accordance with the Declaration of Helsinki on 463

“Medical Research Involving Human Subjects” and after approval by the responsible ethics committee 464

(Commission cantonale d'éthique de la recherche sur l'être humain, CER-VD). Written informed 465

consent was obtained from each participant prior to the experimental sessions. Data from one 466

participant (male, age = 22) were excluded because of excessive artifacts in the EEG. 467

4.2. Experimental design 468

Visual stimuli were Random Dot Kinematograms (RDK: 10° field-size, 1200 dots, 0.2° dot-size, 469

infinite dot-life, and 4°/s dot-speed), that were contrast-modulated through a Gaussian-windowed 470

sinusoidal grating (Gabor: 6° width at 3 SDs, and 0.5 cycles/° spatial frequency), such that the Gabor 471

determined the visible region of the RDK and its spatial pattern of contrast (Figure 1A). The 472

orientation of the Gabor ranged between 45° and -45° off-vertical, and phase varied randomly at every 473

trial. In the RDK, a subset of dots moved coherently either towards left or towards right (signal-dots), 474

while the remaining dots moved with random walks (noise-dots), in such a way that their direction 475

varied at every frame but their speed was kept constant. Stimuli were presented on a gray background 476

for 300 ms always at the same spatial location, around a central fixation spot that was constantly on 477

the screen (0.2° size, and (0,0,160) color in RGB digital 8-bit notation). The inter-trial interval (ITI) 478

was randomly varied between 800 and 1000 ms. The generation and presentation of the visual stimuli 479

was performed using a combination of in-house Python codes and PsychoPy Builder (Peirce et al., 480

2019). 481

The study consisted of three separate experimental sessions: one behavioral (approx. 45 minutes), one 482

EEG (approx. 60 minutes), and one session of fMRI (approx. 45 minutes). In all sessions, the 483

participants were asked to performed different tasks on the same type of visual stimuli. The behavioral 484

session comprised two different tasks in separate blocks (100 trials), requiring a two-choice response. 485

At the beginning of each block, participants were either instructed to report the perceived motion 486

direction in the RDK (left vs. right, motion discrimination task), or the off-vertical tilt of the Gabor 487

(left vs. right, orientation discrimination task), by pressing the corresponding key on a keyboard. The 488


https://doi.org/10.1101/2020.02.28.969253


19

aim of the behavioral session was to determine individual thresholds for coherent motion and 489

orientation discrimination, in order to level performance for the two tasks in the EEG and fMRI 490

sessions, keeping accuracy at 82% of correct responses for both tasks. Thresholds were estimated 491

using an adaptive staircase procedure (Watson and Pelli, 1983). We used starting values of 60% for 492

signal-dots and 4° for angle of orientation, while the task-irrelevant feature was kept constant during 493

the staircase procedure (40% of signal-dots in the orientation discrimination task and ±3° off-vertical 494

in the motion discrimination task). This allowed to obtain individual discrimination thresholds at the 495

same level of accuracy across participants, both for the motion discrimination task (signal-dots 496

percentages 10.61–89.82, M=44.50, SD=29.09) and for the orientation discrimination task (angles of 497

orientation 1.53°–4.30°, M=2.45°, SD=0.72°). These individually calibrated values were used to define 498

stimuli in the subsequent EEG and fMRI sessions. 499

During neuroimaging sessions, participants also performed an additional control task, where they had 500

to report sporadic color changes in the fixation spot lasting 200 ms (from dark blue to red, 30% of 501

trials; Figure 1A) by pressing a response button. To ensure that the timing of these color changes was 502

unpredictable, we presented them at any time during the trial, randomly. Trials in which the color 503

change overlapped with visual stimuli appearance were excluded from the successive analyses. This 504

design resulted in two conditions in which the RDK was task-relevant (motion and orientation 505

discrimination) and one condition where it was task-irrelevant (color change detection). We refer to 506

these different task conditions as Attended (Attended-motion and Attended-orientation) and 507

Unattended, indicating the type of attentional processing required for the visual stimuli. The EEG 508

session consisted of 160 trials divided in 4 blocks for each of task condition (12 blocks in total). Small 509

breaks were offered between blocks. The fMRI session followed a similar block structure, but with 510

less trials (48 trials per task condition). A fixed 12 s break was presented after each block, following 511

the instruction to “REST” (Rest period), with the exception of the last Rest period that lasted 60 s. 512

Task order was counterbalanced across participants in each of the three experimental sessions. 513

Participants had a limited time to respond (1500 ms), after which the response for that trial was 514

considered incorrect. 515

4.3. MRI materials and procedures 516

4.3.1. Data acquisition and preprocessing 517

MRI data were collected using a Discovery MR750 3.0T (GE Healthcare, Chicago, USA) at the 518

cantonal hospital, HFR Fribourg. First, functional data were collected with T2*-weighted EPI 519

sequence with 40 slices each, 3 mm slice-thickness, 0.3 mm spacing between slices, interleaved 520

bottom-up slice acquisition, anterior-to-posterior phase encoding direction, 2500 ms repetition rime 521

(TR), 30 ms echo time (TE), and 85° flip-angle. The first 4 volumes of each run were discarded from 522

analyses. Functional data acquisition was performed while each participant performed the visual 523

discrimination tasks. Behavioral responses were collected using an MRI-compatible fiber optic 524


https://doi.org/10.1101/2020.02.28.969253


20

response pad (Current Designs Inc., Philadelphia, USA). Stimuli were presented on a NordicNeuroLab 525

(Bergen, NO) MRI-compatible LCD monitor (32 inches diagonal size, 1920 x 1080 resolution, 405 526

c/m2 surface luminance, 4000:1 contrast, 60 Hz refresh rate, 6.5 ms response time), placed above the 527

scanner-bed at a distance of 244 cm from the participant’s eyes and visible to the participant through a 528

mirror placed on the head coil. Participants who needed optical correction (e.g., for myopia) wore 529

MRI-compatible glasses with the appropriate lenses. Next, an anatomical whole-head image was 530

acquired using a T1-weighted BRAVO 3D sequence with 280 slices, 1 mm isotropic voxels, 730 ms 531

TR, 2.8 ms TE, 9° flip-angle, 256 x 256 mm2 field of view (FOV), and 900 ms inversion time (TI). 532

The preprocessing of both functional and structural images was performed using the Statistical 533

Parametric Mapping (SPM) toolbox (Penny et al., 2011), in its SPM12 version (University College 534

London, London, UK; https://www.fil.ion.ucl.ac.uk/spm/). Functional images were first aligned to the 535

mean of each session using a two-pass realignment procedure for motion correction, and successively 536

a slice-timing correction was applied (Sladky et al., 2011). After realignment, the mean functional 537

image was coregistered to the anatomical image using the normalized mutual information as cost 538

function. The standard segmentation procedure in SPM12 was applied to obtain individual masks for 539

cerebrospinal fluid (CSF) and white matter (WM), which were used to extract the time courses of CSF 540

and WM signals for each participant. At last, all images were normalized to the Montreal Neurological 541

Institute and Hospital (MNI) stereotaxic space using a fourth-order B-spline interpolation, and 542

smoothed with a Gaussian filter (8 mm FWHM kernel). 543

4.3.2. fMRI statistical analysis 544

A two-stage approach based on a general linear model (GLM) was employed to analyze the functional 545

images (Friston et al., 1994). In this approach, the first-level analysis was implemented using a block 546

design with four regressors of interest, each modeled with a boxcar function convolved with the 547

canonical hemodynamic response function (HRF). Four regressors of interest were defined to model 548

the two Attended conditions (Attended-motion and Attended-orientation), the Unattended condition, 549

and Rest periods. The GLM included also a set of nuisance regressors that modeled the six motion 550

realignment parameters, the mean signals in CSF and WM, and a constant term. Finally, a high-pass 551

filter (200 s cutoff) was applied to the functional images time series, which allowed removing noise at 552

very low frequencies. At the first-level individual analysis, two contrasts were considered to obtain 553

task-related activation maps: i) “Attended vs. Unattended”; ii) “Attended vs. Rest”. The difference 554

between the two Attended conditions and the Unattended was computed for the first contrast, which 555

allowed identifying areas that were differentially active across task-relevant and task-irrelevant 556

conditions (two-tailed test). The second contrast was based on the difference between Attended 557

conditions and Rest period, which allowed identifying areas more strongly engaged by the tasks (one-558

tailed test). The second-level group analysis was implemented on the previously obtained statistical 559

maps and involved voxel-wise t-test comparison across participants. Here, participants were modeled 560


https://doi.org/10.1101/2020.02.28.969253


21

as random effects, and statistical significance was assessed at the group-level using an uncorrected 561

voxel-based threshold p<0.001, together with a minimum cluster-size of 5 voxels (Figure 3A–B). The 562

results of this procedure were used to extract ROIs for source-space EEG analyses (see below). 563

4.4. EEG materials and procedures 564

4.4.1. Data acquisition and preprocessing 565

EEG recordings were collected at the Department of Psychology, University of Fribourg. During the 566

EEG session, participants were sitting inside a dark shielded room, with their head leaning on a 567

chinrest positioned at 71 cm from a VIEWPixx /EEGTM (VPixx Technologies Inc., Saint-Bruno, CA) 568

LCD monitor (24 inches diagonal size, 1920 x 1080 resolution, 100 cd/m2 luminance, 120 Hz refresh 569

rate, 1 ms pixel response time). A 2-button RESPONSEPixx response box (VPixx Technologies Inc.) 570

was plugged into the monitor and used by the participants during the visual tasks. EEG data were 571

acquired using a 128-channel ActiveTwo EEG system (Biosemi, Amsterdam, NL). Data acquisition 572

was performed at a sampling rate of 1024 Hz, and signals were referenced to the Common Mode 573

Sense (CMS) active electrode, which together with the Driven Right Leg (DRL) passive electrode 574

formed a feedback loop in the system. This CMS/DRL loop allow to drive the common-mode voltage 575

as close as possible to the ADC reference voltage in the AD-box, and to obtain an extra 40 dB 576

common-mode rejection ratio at 50 Hz, when compared with using normal ground electrodes with 577

same impedance. At the end of the EEG session, the 3D coordinates of the positions of the electrodes 578

were localized for each participant using an ELPOS system (Zebris Medical GmbH, Isny im Allgäu, 579

DE). These individual electrodes’ positions were used for EEG source reconstruction procedure (see 580

section 4.5). 581

EEG preprocessing was performed using a combination of EEGLAB (Delorme et al., 2011; Delorme 582

and Makeig, 2004), its plugins, and in-house scripts implemented in MATLAB (The MathWorks Inc., 583

Natick, USA). Data were first downsampled to 500 Hz using an anti-aliasing filter with 125 Hz cutoff 584

frequency and 50 Hz transition bandwidth. Afterwards, the data were detrended using high-pass 585

filtering (1 Hz low-frequency cutoff), as implemented in the PREP plugin (Bigdely-Shamlo et al., 586

2015). Line noise and its harmonics were reduced using the adaptive filtering technique implemented 587

in the EEGLAB plugin CleanLine (https://www.nitrc.org/projects/cleanline), which allows identifying 588

and removing significant sinusoidal artifacts. Epochs were extracted using the time window -1500–589

1000 ms around stimulus onset. Noisy EEG channels were identified by visual inspection and 590

removed before proceeding to the subsequent preprocessing steps (number of removed channels across 591

participants 12–21, M=15.85, SD=2.64). Epochs contaminated by noise artifacts and eye movements 592

artifacts, e.g., eye blinks occurring within 500 ms from stimulus presentation (either before or after), 593

were also rejected by visual inspection (percentage of rejected epochs across participants 12.50–594

38.95%, M=23.01, SD=7.49). Decomposition of EEG data by independent component analysis (ICA) 595

was performed using the FastICA algorithm (Hyvärinen and Oja, 2000). ICA components identified as 596


https://doi.org/10.1101/2020.02.28.969253


22

eye artifacts or muscular activity artifacts were removed from the data (number of removed 597

components across participants 5–23, M=10.30, SD=4.09). As final preprocessing steps, noisy EEG 598

channels were interpolated using a spherical spline interpolation (Perrin et al., 1989); then, the signals 599

were re-referenced to the common average reference (Lehmann and Skrandies, 1980). We excluded 600

trials with incorrect responses, and the amount of trials across task conditions was balanced within-601

subjects (number of trials per task condition across participants 74–127, M=103.5, SD=15.3). This 602

choice was motivated by the fact that functional connectivity analysis can be sensitive to the amount 603

of trials (Astolfi et al., 2008; Toppi et al., 2012), and consequently an imbalance of trials between task 604

conditions may create spurious differences from comparing them. 605

4.4.2. Visual-evoked potentials and power spectra 606

The VEPs of each participant were obtained by separately averaging the epochs of the three task 607

conditions. VEPs analysis was confined to three clusters on sensor-space: centro-occipital (CO), left 608

occipito-temporal (lOT) and right occipito-temporal (rOT). Each electrode-cluster comprised 7 609

electrode sites from the 128-channel Biosemi cap, using the same subdivision adopted in Daffner et al. 610

(2012). Each trial was baseline corrected before computing VEPs, using the 200 ms pre-stimulus 611

interval as baseline. A repeated-measures ANOVA was employed to compare the three task conditions 612

in each of the pre-selected sensor-space macro-area. Statistical significance was assessed at alpha-level 613

of 0.05, using false discovery rate (FDR) correction for multiple comparisons (Benjamini and 614

Hochberg, 1995). In addition, a temporal stability criterion was imposed on the results so that only 615

significant results that exhibited stability over at least 5 time frames (10 ms) survived. Post-hoc 616

analysis was carried out using the Fisher–Hayter procedure (Hayter, 1986), in each of the time 617

intervals that showed a statistically significant difference from the repeated-measures ANOVA. 618

Power spectral density analysis was performed on sensor-space time series in two time windows of 619

interest: pre-stimulus (anticipatory) and post-stimulus (reactive). The anticipatory window was defined 620

as the last 300 ms before stimulus onset. The reactive window was selected as the interval 200–500 ms 621

after stimulus onset, because it is when strong stimulus-induced power changes are typically observed 622

at low-frequencies (Klimesch et al., 2007, 2001; Mazaheri and Picton, 2005; Schmiedt et al., 2014). 623

Power spectrum at each electrode location was estimated in the frequency range 1–100 Hz, using a 624

multitaper method based on orthogonal tapers given by Slepian sequences, whose complete 625

description and MATLAB implementation were provided in previous works (Mitra and Pesaran, 1999; 626

Pagnotta et al., 2018a, 2018b). In this approach, the time-bandwidth product (NW) regulates the trade-627

off between variance and bias of the spectral estimates. To accommodate for the different 628

characteristics of the spectral components at low and high frequencies, two distinct time-bandwidth 629

products were here used for frequencies up to 45 Hz (NW=3) and for frequencies above 45 Hz 630

(NW=16). Attended and Unattended conditions were compared by using a cluster-based permutation 631

approach over frequencies and EEG electrodes locations, with 30 mm maximum distance to define 632


https://doi.org/10.1101/2020.02.28.969253


23

neighboring electrodes. In this and every other analysis using a cluster-based permutation approach, to 633

compare Attended and Unattended conditions, we employed the same settings: two-tailed t-test 634

(p<0.05), 50000 permutations, and p<0.05 for the permutation test (Maris and Oostenveld, 2007). 635

4.5. EEG source reconstruction 636

Source reconstruction techniques enable to estimate and localize the EEG sources of brain electrical 637

activity, by solving first a forward problem and then an inverse problem (Michel et al., 2004). For the 638

forward problem, the volume conduction model of each participant was constructed using a boundary 639

element method (BEM) (Hamalainen and Sarvas, 1989), which employed the information about 640

boarder surfaces between three different tissue-types: scalp, skull and brain. These surfaces were 641

obtained from a segmentation procedure of the individual-participant anatomical MRI, using a 642

Gaussian kernel for smoothing (5 voxels FWHM). The participant-specific 3D electrode coordinates 643

(see section 4.4.1) were aligned to the head model of the same participant using an interactive re-644

alignment procedure, followed by a projection of electrodes onto the head surface. For group analyses 645

on source-reconstructed EEG data, we employed a template grid based on MNI template anatomical 646

MRI to define the solution points. The template comprised 1725 source-points (equivalent current 647

dipoles) with 10 mm spacing, and was constrained to be within the cortical gray matter. Individual 648

MRIs were warped to the template, and the inverse of this warping procedure was applied to the 649

template grid. This procedure provided participant-specific grids that were no longer regularly spaced, 650

but that became equivalent across participants in normalized MNI space, facilitating group analysis on 651

source-space. Finally, individual lead field matrices were computed considering an unconstrained-652

orientation forward operator, i.e., each solution point was modeled as three orthogonal equivalent 653

current dipoles placed at that location. 654

To solve the EEG inverse problem, we employed the linearly constrained minimum variance (LCMV) 655

beamformer (Van Veen et al., 1997). The LCMV belongs to the family of spatial filtering methods and 656

its implementation relies on the use on an estimate of the sensor-space covariance matrix, which was 657

here estimated from the time window -500–500 ms around stimulus onset. Source-reconstructed 658

signals were extracted from 22 cortical ROIs using an approach based on singular value 659

decomposition (SVD), estimating scalar-value time series that explain most of the variability across 660

dipoles in each ROI (Rubega et al., 2019). The ROIs were defined with an automatic procedure, 661

starting from the spatially segregated cluster-peaks obtained from the fMRI statistical analysis (Figure 662

3A–B). First, each fMRI cluster-peak was used as the center of a sphere of 15 mm radius. Then, all 663

solutions points inside the sphere were selected, and tissue-type consistency was verified using the 664

automated anatomical labeling (AAL2) atlas parcellations defined in MNI space, so that only the 665

solution points that fell within the atlas areas were kept. This procedure allowed us to uniquely 666

identify each ROI, in such a way that no solution points were shared across ROIs. Table 2 lists the 22 667

ROIs (Figure 3C–E). As final step, an approach based on innovations orthogonalization was used for 668


https://doi.org/10.1101/2020.02.28.969253


24

leakage correction (Pascual-Marqui et al., 2017). This orthogonalization approach allowed reducing 669

the detrimental effects on functional connectivity analyses of zero-lag cross-correlations, which are 670

due to instantaneous linear mixing between source-reconstructed signals (source leakage) and are 671

known to produce spurious functional connectivity estimates (Anzolin et al., 2019). All the steps of 672

EEG source reconstruction were implemented using in-house MATLAB codes and routines from 673

FieldTrip (Oostenveld et al., 2011) (Radboud University, Nijmegen, NL; 674

http://www.ru.nl/neuroimaging/fieldtrip). 675

4.6. Time-varying power spectra 676

To investigate local brain rhythms modulations depending on whether the stimuli were relevant or not, 677

time-varying power spectra were estimated from source-reconstructed signals of all ROIs using a 678

Morlet wavelet transform with central frequency parameter ω0 = 6, in combination with zero-padding 679

to solve the problem of edge effects (Torrence and Compo, 1998). Time-varying spectral estimates 680

were estimated in the frequency range 1–100 Hz. This procedure was performed for each participant 681

and task condition using the MATLAB codes made available in Pagnotta et al. (2018b), after 682

subtracting the ensemble mean across trials from single-trial data, which allows minimizing the 683

influence of evoked responses on power spectra (Kalcher and Pfurtscheller, 1995; Rajan et al., 2018). 684

A cluster-based permutation approach was used to compare time-varying power spectra of Attended 685

and Unattended conditions, over time frames (-50–600 ms around stimulus onset) and frequencies (1–686

100 Hz). 687

4.7. Local cross-frequency coupling 688

4.7.1. Anticipatory (pre-stimulus) cross-frequency coupling 689

Cross-frequency coupling was measured in terms of dependence between phase of low-frequency 690

oscillations and amplitude of high-frequency oscillations, called phase-amplitude coupling or PAC. 691

Among the available types of cross-frequency coupling, we employed the PAC because of its more 692

clear functional role and physiological plausibility (Canolty and Knight, 2010; Voytek et al., 2010). 693

Several approaches have been proposed to measure PAC (Canolty et al., 2006; Martínez-Cancino et 694

al., 2019; Penny et al., 2008; Tort et al., 2010; Voytek et al., 2013). We here employed the modulation 695

index based on GLM (Penny et al., 2008) to calculate within-region PAC before stimulus onset, using 696

the toolbox from Martínez-Cancino et al. (2019). The ensemble mean was preliminary subtracted from 697

single-trial data for each participant and task condition before PAC estimation. The analysis was 698

restricted to the subset of 9 occipito-temporal ROIs. The central frequency for the phase time series 699

(fPHASE) ranged from 4 Hz to 30 Hz in 1Hz-steps, and the higher central frequency for the amplitude 700

time series (fAMPLITUDE) ranged from 40 Hz to 100 Hz in 2Hz-steps. The bandwidth of the band-pass 701


https://doi.org/10.1101/2020.02.28.969253


25

filters to estimate the phase time series (BWPHASE) and the amplitude time series (BWAMPLITUDE) were 702

defined according to the following expressions: 703

𝐵𝑊!"#$%(𝐻𝑧) = 𝑓!"#$% − 1, 𝑓!"#$% + 1𝐵𝑊!"#$%&'()(𝐻𝑧) = 𝑓!"!"#$%&' − 𝑓!"#$% + 1 , 𝑓!"#$%&'() + 𝑓!"#$% + 1

(1)

Filtering was applied to the pre-stimulus interval (-300–0 ms), using two buffer windows of 1850 ms, 704

before and after the interval of interest. Each buffer window consisted of 600 ms of observed signal 705

and the rest zeros, such that filtering was applied every time to a 4000 ms window. A combination of 706

high-pass and low-pass finite impulse response (FIR) filters was employed to band-pass filter the data 707

in the frequency bands of interest. Next, the Hilbert transform was applied to extract the analytic 708

signals from which instantaneous low-frequency phase and high-frequency amplitude time series were 709

estimated, which allowed computing PAC in the time window of interest using the GLM-based 710

method. This procedure was repeated for every combination of fPHASE and fAMPLITUDE, in each trial. The 711

averages across trials of PAC estimates were compared over all pairs of fPHASE and fAMPLITUDE between 712

Attended and Unattended conditions, using a cluster-based permutation approach. We excluded from 713

the analysis all the fPHASE–fAMPLITUDE pairs for which BWAMPLITUDE overlapped with BWPHASE. 714

Since strong imbalances in the power spectra between conditions can lead to spurious PAC differences 715

(Aru et al., 2015), a control analysis based on a stratification procedure was performed separately for 716

each ROI using the ft_stratify function from FieldTrip (Oostenveld et al., 2011), with 1000 iterations. 717

At each iteration, this procedure selects subsets of trials from the two conditions with matched 718

distributions of power spectra across trials, for both phase and amplitude frequency bands. Balanced 719

numbers of trials between conditions were maintained during this procedure. The final estimates from 720

this control analysis were obtained as the median across iterations of the power-balanced estimates, 721

and these were compared between Attended and Unattended conditions using the cluster-based 722

permutation. The control analysis based on stratification was used to confirm the between-condition 723

differences in pre-stimulus PAC. 724

4.7.2. Reactive (post-stimulus) cross-frequency coupling 725

A measure of stimulus-evoked cross-frequency coupling was computed for each task condition using 726

the instantaneous phase-locking value (PLV) between the low-frequency phase and the phase of 727

amplitude-filtered high-frequency signal (Lachaux et al., 1999; Pascucci et al., 2018). As in the pre-728

stimulus analysis, the ensemble mean was preliminary subtracted from single-trial data for each 729

participant and task condition, and the estimation was performed on the 9 occipito-temporal ROIs. 730

Time-varying PLV estimates were obtained at each time frame by averaging the instantaneous 731

contributes across trials. The analysis was performed on the interval 0–500 ms after stimulus onset, 732

using two buffer windows (before/after), each of 1450 ms length (including 300 ms of observed 733

signal). Low-frequency ranges were separately defined in α and β-band, as fPHASE = 10 Hz and fPHASE = 734

20 Hz for the comparison between Attended-motion and Unattended, and as fPHASE = 8 Hz and fPHASE = 735


https://doi.org/10.1101/2020.02.28.969253


26

17 Hz for the comparison between Attended-orientation and Unattended. These values were selected 736

in data-driven way, on the basis of the task-specific results obtained by comparing time-varying power 737

spectra (Figure 2D and Figure S1D). For the amplitude time series, fAMPLITUDE ranged from 50 Hz to 738

100 Hz in 2Hz-steps. The bandwidths of the two band-pass filters were chosen using the expressions 739

in equation (1) (see section 4.7.1). The procedure of filtering (FIR-based) and analytic signals 740

extraction (using Hilbert transform) was carried out as previously done for the anticipatory PAC 741

analysis. A cluster-based permutation approach over time frames and amplitude high-frequencies was 742

employed to compare PLV estimates between Attended and Unattended conditions. 743

4.8. Large-scale network analyses 744

To estimate the functional interactions among brain regions we used the information form of partial 745

directed coherence (iPDC) (Takahashi et al., 2010). The iPDC is based on the notion of Granger-746

Geweke causality (Geweke, 1984; Granger, 1969), which relies on the concepts of temporal 747

precedence and statistical predictability between simultaneously recorded time-series, extended to the 748

spectral-domain (Seth et al., 2015). Computationally the iPDC is derived from a multivariate 749

autoregressive (MVAR) model with a certain order, which determines the number of past observations 750

included in the model. To derive iPDC estimates the MVAR coefficients are Fourier transformed, 751

appropriately weighted by noise covariance matrix contributes to overcome the scale-variance 752

problem, and finally normalized (Baccalá and Sameshima, 2014). The iPDC is a measure of the 753

partialized delayed functional interactions between brain regions, which is able to characterize the 754

interaction directionality between two regions (directed measure), as well as discerning direct from 755

indirect or mediated paths of connections (directness). More precisely, the iPDC from the j-th time 756

series to the i-th time series is equivalent to the squared coherence between the innovation process 757

associated with i and the partialized version of j, which via appropriate logarithmic expression and 758

integration provides the mutual information rates between these two processes (Baccalá and 759

Sameshima, 2014; Takahashi et al., 2010). In the present study, the iPDC was used as measure of the 760

frequency-specific, directed connections between ROIs, and all connectivity analyses were performed 761

on source-reconstructed time series, because analyses applied on EEG sensor-space time series do not 762

allow any meaningful interpretation in terms of interacting brain sources (Brunner et al., 2016; Van de 763

Steen et al., 2016). 764

4.8.1. Anticipatory (pre-stimulus) functional connectivity 765

To investigate the anticipatory network-level modulations of attention, we performed a directed 766

functional connectivity analysis on the pre-stimulus interval (-300–0 ms). Preprocessing included a 767

signal decimation by a factor of 2 (new sampling rate of 250 Hz), the removal of temporal mean and 768

ensemble mean from single-trial data to meet zero-mean assumptions, and the use of the augmented 769

Dickey–Fuller test for stationarity as implemented in the Oxford MFE Toolbox 770


https://doi.org/10.1101/2020.02.28.969253


27

(https://www.kevinsheppard.com/code/matlab/mfe-toolbox/). The optimal model order for each 771

participant was selected by first identifying the value that minimized the difference between power 772

spectra obtained from parametric MVAR-modeling and those obtained using a nonparametric 773

approach based on multitaper (see section 4.4.2), in each task condition separately, and then selecting 774

the maximum value across task conditions (model orders across participants 7–18, M=14.05, 775

SD=3.27). These individual-participants’ fixed values were then used for analysis. A stepwise least 776

squares MVAR model estimation was then performed on source-reconstructed signals to derive iPDC 777

estimates in the network of 22 ROIs, for each task condition. The iPDC estimates were computed in 778

the frequency range 1–100 Hz (in 1Hz-steps). 779

Measures from graph theory were used to characterize large-scale topological properties of the 780

frequency-specific, directed functional networks in the different task conditions (Rubinov and Sporns, 781

2010). Each ROI was considered as a node in the graph and two measures were estimated: global 782

efficiency and local efficiency. The network’s global efficiency is the average inverse shortest path 783

length and indicates the level of global functional integration in the network (Latora and Marchiori, 784

2001). The single node’s local efficiency is the global efficiency of the local subgraph comprising all 785

its immediate neighbors (Latora and Marchiori, 2001), and the network’s local efficiency is computed 786

as average efficiency of all local subgraphs, which represents the level of local integration within 787

subgraphs. High local efficiency suggests functional segregation within the network (Rubinov and 788

Sporns, 2010). Both global efficiency and local efficiency were computed for each participant and task 789

condition from the full matrix of iPDC estimates, i.e., from a fully-connected, weighted and directed 790

adjacency matrix. Subgraphs and nodes’ neighbors were thus determined by the interaction strengths: 791

the stronger the connection weight between two nodes, the closer they were (functionally). 792

When we consider global and local efficiency derived from iPDC-based graphs, we can talk about 793

frequency-specific levels of functional integration among brain regions, respectively globally and 794

within subgraphs, for two reasons. First, the iPDC has a clear interpretation in terms of between-795

processes mutual information rates, providing an association with the notion of information flow 796

(Baccalá and Sameshima, 2014; Takahashi et al., 2010). Second, thanks to its partialization approach, 797

the iPDC provides the directness of functional connections. It is therefore reasonable to characterize 798

the cascades of these direct connections as paths of integration between regions, unlike when using 799

classical measures such as cross-correlations and spectral coherence (Rubinov and Sporns, 2010). 800

Each graph measure was separately compared between Attended and Unattended conditions (two-801

tailed t-test, p<0.05, with FDR correction). Graph measures were derived using the Brain Connectivity 802

Toolbox (Rubinov and Sporns, 2010) (http://www.brain-connectivity-toolbox.net). All the other steps 803

of the functional connectivity analysis were performed using in-house scripts implemented in 804

MATLAB. 805


https://doi.org/10.1101/2020.02.28.969253


28

4.8.2. Reactive (post-stimulus) functional connectivity 806

To accommodate the non-stationarity of VEP signals and obtain time-varying functional interactions 807

among brain regions, the iPDC can be derived from a time-varying MVAR (tvMVAR) model of the 808

signals, which can be computed using adaptive algorithms (Arnold et al., 1998; Milde et al., 2010; 809

Pagnotta and Plomp, 2018). This procedure allows obtaining iPDC estimates over frequencies and 810

time frames. Time-varying iPDC estimates were computed in the interval -1500– 800 ms around 811

stimulus onset, for each participant and task condition, after decimating single-trial data by a factor of 812

two and subtracting their temporal and ensemble means. We performed tvMVAR modeling with the 813

Self-Tuning Optimized Kalman filter (STOK) algorithm (Pascucci et al., 2019), using a percentage of 814

variance explained of 99% for setting the filtering factor threshold (Hansen, 1987). Similarly to the 815

anticipatory connectivity analysis, the model order was selected by comparing parametric and 816

nonparametric power spectra in the post-stimulus interval 0–500 ms (model orders across participants 817

11–21, M=17.53, SD=2.76). 818

Measures of time-varying global efficiency and local efficiency were estimated in the time window -819

20–500 ms in three frequency bands of interest (α, β, and γ-band). For each frequency band, the time-820

frequency iPDC estimates were first averaged over frequency bins inside that band, then used to derive 821

time-varying global efficiency and local efficiency, and finally compared between Attended and 822

Unattended conditions (paired sample t-test, p<0.05 with FDR correction). For the comparison 823

between Attended-motion and Unattended, the frequency bands were selected as 8–12 Hz (α-band), 824

17–23 Hz (β-band), and 70–95 Hz (γ-band), based on the results on time-varying power spectra 825

(Figure 2D) and α-γ PAC (Figure 6A). For the comparison between Attended-orientation and 826

Unattended, the frequency bands were 6–10 Hz (α-band), 14–20 Hz (β-band), and 65–95 Hz (γ-band) 827

(Figure S1D and Figure 6B). 828

4.9. Correlation between local network properties and PAC 829

In cortical regions that showed significant anticipatory changes in PAC, we investigated whether there 830

was a relationship between the local PAC and the local network properties of that region, as revealed 831

by the single-node’s local efficiency. In each region, we identified the cluster of fPHASE–fAMPLITUDE pairs 832

that showed significant differences between Attended and Unattended conditions (Figure 5), which 833

allowed defining also the significant frequency ranges for fPHASE and fAMPLITUDE. For example, from the 834

comparison between Attended-motion and Unattended in region V5-R the significant cluster is 835

highlighted in opaque red in Figure 5A (β-γ PAC), and the two significant ranges are 24–27 Hz for 836

fPHASE (β-band) and 54–80 Hz for fAMPLITUDE (γ-band). Participant-specific PAC estimates were 837

averaged over the significant cluster, separately for Attended-motion and Unattended, and a variation 838

in PAC (ΔPAC) was computed as the difference between Attended-motion and Unattended, 839

normalized by the latter. In a similar way, but separately for the two significant ranges for fPHASE and 840

fAMPLITUDE (in the example, β-band and γ-band), participant-specific estimates of the region’s local 841


https://doi.org/10.1101/2020.02.28.969253


29

efficiency were averaged over each significant range, separately for Attended-motion and Unattended, 842

and a variation in single-node local efficiency (ΔElocal) was computed as the difference between 843

Attended-motion and Unattended, normalized by the latter. We estimated the correlation between 844

ΔPAC and ΔElocal across participants using the Spearman's rank correlation coefficient. Outliers were 845

removed using a robust regression with bisquare weighting function, as implemented in robustfit 846

function in MATLAB, on the basis of the studentized residuals (t-distribution with 17 degrees of 847

freedom, p<0.05). The same analysis was performed for the comparison between Attended-orientation 848

and Unattended, focusing on the set of regions that showed significant anticipatory differences in PAC 849

(Figure 5B). 850

Acknowledgements 851

This study was supported by the Swiss National Science Foundation (grant PP00P1_157420; G.P.). 852

The authors declare that the study has been conducted in the absence of any conflict of interest. 853

References 854

Ahlfors, S.P., Simpson, G.V., Dale, A.M., Belliveau, J.W., Liu, A.K., Korvenoja, A., Virtanen, J., 855 Huotilainen, M., Tootell, R.B.H., Aronen, H.J., Ilmoniemi, R.J., 1999. Spatiotemporal 856 Activity of a Cortical Network for Processing Visual Motion Revealed by MEG and fMRI. J. 857 Neurophysiol. 82, 2545–2555. https://doi.org/10.1152/jn.1999.82.5.2545 858

Antzoulatos, E.G., Miller, E.K., 2016. Synchronous beta rhythms of frontoparietal networks support 859 only behaviorally relevant representations. eLife 5, e17822. 860 https://doi.org/10.7554/eLife.17822 861

Antzoulatos, E.G., Miller, E.K., 2014. Increases in Functional Connectivity between Prefrontal Cortex 862 and Striatum during Category Learning. Neuron 83, 216–225. 863 https://doi.org/10.1016/j.neuron.2014.05.005 864

Anzolin, A., Presti, P., Van De Steen, F., Astolfi, L., Haufe, S., Marinazzo, D., 2019. Quantifying the 865 Effect of Demixing Approaches on Directed Connectivity Estimated Between Reconstructed 866 EEG Sources. Brain Topogr. 32, 655–674. https://doi.org/10.1007/s10548-019-00705-z 867

Arnold, M., Milner, X.H.R., Witte, H., Bauer, R., Braun, C., 1998. Adaptive AR modeling of 868 nonstationary time series by means of Kalman filtering. IEEE Trans. Biomed. Eng. 45, 553–869 562. https://doi.org/10.1109/10.668741 870

Aru, Juhan, Aru, Jaan, Priesemann, V., Wibral, M., Lana, L., Pipa, G., Singer, W., Vicente, R., 2015. 871 Untangling cross-frequency coupling in neuroscience. Curr. Opin. Neurobiol. 31, 51–61. 872 https://doi.org/10.1016/j.conb.2014.08.002 873

Astolfi, L., Cincotti, F., Mattia, D., De Vico Fallani, F., Tocci, A., Colosimo, A., Salinari, S., 874 Marciani, M.G., Hesse, W., Witte, H., others, 2008. Tracking the time-varying cortical 875 connectivity patterns by adaptive multivariate estimators. IEEE Trans. Biomed. Eng. 55, 902–876 913. https://doi.org/10.1109/TBME.2007.905419 877

Baccalá, L.A., Sameshima, K., 2014. Partial Directed Coherence, in: Methods in Brain Connectivity 878 Inference through Multivariate Time Series Analysis, Frontiers in Neuroengineering Series. 879 CRC Press, pp. 57–73. https://doi.org/10.1201/b16550-6 880

Bach, M., 1996. The Freiburg Visual Acuity Test–Automatic Measurement of Visual Acuity. Optom. 881 Vis. Sci. 73, 49–53. https://doi.org/10.1097/00006324-199601000-00008 882


https://doi.org/10.1101/2020.02.28.969253


30

Baldauf, D., Desimone, R., 2014. Neural Mechanisms of Object-Based Attention. Science 344, 424–883 427. https://doi.org/10.1126/science.1247003 884

Bastos, A.M., Schoffelen, J.-M., 2016. A Tutorial Review of Functional Connectivity Analysis 885 Methods and Their Interpretational Pitfalls. Front. Syst. Neurosci. 9. 886 https://doi.org/10.3389/fnsys.2015.00175 887

Bastos, A.M., Vezoli, J., Bosman, C.A., Schoffelen, J.-M., Oostenveld, R., Dowdall, J.R., De Weerd, 888 P., Kennedy, H., Fries, P., 2015. Visual Areas Exert Feedforward and Feedback Influences 889 through Distinct Frequency Channels. Neuron 85, 390–401. 890 https://doi.org/10.1016/j.neuron.2014.12.018 891

Benjamini, Y., Hochberg, Y., 1995. Controlling the False Discovery Rate: A Practical and Powerful 892 Approach to Multiple Testing. J. R. Stat. Soc. Ser. B Methodol. 57, 289–300. 893

Bigdely-Shamlo, N., Mullen, T., Kothe, C., Su, K.-M., Robbins, K.A., 2015. The PREP pipeline: 894 standardized preprocessing for large-scale EEG analysis. Front. Neuroinformatics 9. 895 https://doi.org/10.3389/fninf.2015.00016 896

Bonnefond, M., Jensen, O., 2015. Gamma Activity Coupled to Alpha Phase as a Mechanism for Top-897 Down Controlled Gating. PLOS ONE 10, e0128667. 898 https://doi.org/10.1371/journal.pone.0128667 899

Bonnefond, M., Kastner, S., Jensen, O., 2017. Communication between Brain Areas Based on Nested 900 Oscillations. eneuro 4, ENEURO.0153-16.2017. https://doi.org/10.1523/ENEURO.0153-901 16.2017 902

Bosking, W.H., Zhang, Y., Schofield, B., Fitzpatrick, D., 1997. Orientation Selectivity and the 903 Arrangement of Horizontal Connections in Tree Shrew Striate Cortex. J. Neurosci. 17, 2112–904 2127. https://doi.org/10.1523/JNEUROSCI.17-06-02112.1997 905

Brunner, C., Billinger, M., Seeber, M., Mullen, T.R., Makeig, S., 2016. Volume Conduction 906 Influences Scalp-Based Connectivity Estimates. Front. Comput. Neurosci. 10. 907 https://doi.org/10.3389/fncom.2016.00121 908

Buschman, T.J., Denovellis, E.L., Diogo, C., Bullock, D., Miller, E.K., 2012. Synchronous Oscillatory 909 Neural Ensembles for Rules in the Prefrontal Cortex. Neuron 76, 838–846. 910 https://doi.org/10.1016/j.neuron.2012.09.029 911

Buschman, T.J., Kastner, S., 2015. From Behavior to Neural Dynamics: An Integrated Theory of 912 Attention. Neuron 88, 127–144. https://doi.org/10.1016/j.neuron.2015.09.017 913

Buzsáki, G., 2006. Rhythms of the Brain. Oxford University Press. 914 https://doi.org/10.1093/acprof:oso/9780195301069.001.0001 915

Buzsáki, G., Draguhn, A., 2004. Neuronal Oscillations in Cortical Networks. Science 304, 1926–1929. 916 https://doi.org/10.1126/science.1099745 917

Buzsáki, G., Wang, X.-J., 2012. Mechanisms of Gamma Oscillations. Annu. Rev. Neurosci. 35, 203–918 225. https://doi.org/10.1146/annurev-neuro-062111-150444 919

Canolty, R.T., Edwards, E., Dalal, S.S., Soltani, M., Nagarajan, S.S., Kirsch, H.E., Berger, M.S., 920 Barbaro, N.M., Knight, R.T., 2006. High Gamma Power Is Phase-Locked to Theta 921 Oscillations in Human Neocortex. Science 313, 1626–1628. 922 https://doi.org/10.1126/science.1128115 923

Canolty, R.T., Knight, R.T., 2010. The functional role of cross-frequency coupling. Trends Cogn. Sci. 924 14, 506–515. https://doi.org/10.1016/j.tics.2010.09.001 925

Canolty, R.T., Soltani, M., Dalal, S.S., Edwards, E., Dronkers, N.F., Nagarajan, S.S., Kirsch, H.E., 926 Barbaro, N.M., Knight, R.T., 2007. Spatiotemporal dynamics of word processing in the human 927 brain. Front. Neurosci. 1, 185–196. https://doi.org/10.3389/neuro.01.1.1.014.2007 928

Chelazzi, L., Marini, F., Pascucci, D., Turatto, M., 2019. Getting rid of visual distractors: the why, 929 when, how, and where. Curr. Opin. Psychol. 29, 135–147. 930 https://doi.org/10.1016/j.copsyc.2019.02.004 931

Cohen, J., 1992. A power primer. Psychol. Bull. 112, 155–159. https://doi.org/10.1037/0033-932 2909.112.1.155 933

Corbetta, M., Shulman, G.L., 2002. Control of goal-directed and stimulus-driven attention in the brain. 934 Nat. Rev. Neurosci. 3, 201–215. https://doi.org/10.1038/nrn755 935


https://doi.org/10.1101/2020.02.28.969253


31

Daffner, K.R., Zhuravleva, T.Y., Sun, X., Tarbi, E.C., Haring, A.E., Rentz, D.M., Holcomb, P.J., 936 2012. Does modulation of selective attention to features reflect enhancement or suppression of 937 neural activity? Biol. Psychol. 89, 398–407. https://doi.org/10.1016/j.biopsycho.2011.12.002 938

Delorme, A., Makeig, S., 2004. EEGLAB: an open source toolbox for analysis of single-trial EEG 939 dynamics including independent component analysis. J. Neurosci. Methods 134, 9–21. 940 https://doi.org/10.1016/j.jneumeth.2003.10.009 941

Delorme, A., Mullen, T., Kothe, C., Akalin Acar, Z., Bigdely-Shamlo, N., Vankov, A., Makeig, S., 942 2011. EEGLAB, SIFT, NFT, BCILAB, and ERICA: New Tools for Advanced EEG 943 Processing. Comput. Intell. Neurosci. 2011, 1–12. https://doi.org/10.1155/2011/130714 944

Engel, A.K., Fries, P., 2010. Beta-band oscillations—signalling the status quo? Curr. Opin. Neurobiol. 945 20, 156–165. https://doi.org/10.1016/j.conb.2010.02.015 946

Fiebelkorn, I.C., Kastner, S., 2019. A Rhythmic Theory of Attention. Trends Cogn. Sci. 23, 87–101. 947 https://doi.org/10.1016/j.tics.2018.11.009 948

Fiebelkorn, I.C., Pinsk, M.A., Kastner, S., 2019. The mediodorsal pulvinar coordinates the macaque 949 fronto-parietal network during rhythmic spatial attention. Nat. Commun. 10, 215. 950 https://doi.org/10.1038/s41467-018-08151-4 951

Foster, J.J., Awh, E., 2019. The role of alpha oscillations in spatial attention: limited evidence for a 952 suppression account. Curr. Opin. Psychol. 29, 34–40. 953 https://doi.org/10.1016/j.copsyc.2018.11.001 954

Foxe, J.J., Snyder, A.C., 2011. The Role of Alpha-Band Brain Oscillations as a Sensory Suppression 955 Mechanism during Selective Attention. Front. Psychol. 2. 956 https://doi.org/10.3389/fpsyg.2011.00154 957

Fries, P., 2015. Rhythms for Cognition: Communication through Coherence. Neuron 88, 220–235. 958 https://doi.org/10.1016/j.neuron.2015.09.034 959

Friston, K.J., Holmes, A.P., Worsley, K.J., Poline, J.B., Frith, C.D., Frackowiak, R.S.J., 1994. 960 Statistical parametric maps in functional imaging: A general linear approach. Hum. Brain 961 Mapp. 2, 189–210. https://doi.org/10.1002/hbm.460020402 962

Geweke, J.F., 1984. Measures of conditional linear dependence and feedback between time series. J. 963 Am. Stat. Assoc. 79, 907–915. https://doi.org/10.2307/2288723 964

Granger, C.W.J., 1969. Investigating Causal Relations by Econometric Models and Cross-spectral 965 Methods. Econometrica 37, 424–438. https://doi.org/10.2307/1912791 966

Greenberg, A.S., Esterman, M., Wilson, D., Serences, J.T., Yantis, S., 2010. Control of Spatial and 967 Feature-Based Attention in Frontoparietal Cortex. J. Neurosci. 30, 14330–14339. 968 https://doi.org/10.1523/JNEUROSCI.4248-09.2010 969

Haegens, S., Handel, B.F., Jensen, O., 2011a. Top-Down Controlled Alpha Band Activity in 970 Somatosensory Areas Determines Behavioral Performance in a Discrimination Task. J. 971 Neurosci. 31, 5197–5204. https://doi.org/10.1523/JNEUROSCI.5199-10.2011 972

Haegens, S., Nácher, V., Luna, R., Romo, R., Jensen, O., 2011b. α-Oscillations in the monkey 973 sensorimotor network influence discrimination performance by rhythmical inhibition of 974 neuronal spiking. Proc. Natl. Acad. Sci. 108, 19377–19382. 975 https://doi.org/10.1073/pnas.1117190108 976

Hamalainen, M.S., Sarvas, J., 1989. Realistic conductivity geometry model of the human head for 977 interpretation of neuromagnetic data. IEEE Trans. Biomed. Eng. 36, 165–171. 978 https://doi.org/10.1109/10.16463 979

Hansen, P.C., 1987. The truncatedSVD as a method for regularization. BIT 27, 534–553. 980 https://doi.org/10.1007/BF01937276 981

Hayter, A.J., 1986. The Maximum Familywise Error Rate of Fisher’s Least Significant Difference 982 Test. J. Am. Stat. Assoc. 81, 1000–1004. https://doi.org/10.1080/01621459.1986.10478364 983

Hillyard, S.A., Anllo-Vento, L., 1998. Event-related brain potentials in the study of visual selective 984 attention. Proc. Natl. Acad. Sci. 95, 781–787. https://doi.org/10.1073/pnas.95.3.781 985

Hillyard, S.A., Vogel, E.K., Luck, S.J., 1998. Sensory gain control (amplification) as a mechanism of 986 selective attention: electrophysiological and neuroimaging evidence. Philos. Trans. R. Soc. 987 Lond. B. Biol. Sci. 353, 1257–1270. https://doi.org/10.1098/rstb.1998.0281 988

Hyvärinen, A., Oja, E., 2000. Independent component analysis: algorithms and applications. Neural 989 Netw. 13, 411–430. https://doi.org/10.1016/S0893-6080(00)00026-5 990


https://doi.org/10.1101/2020.02.28.969253


32

Jensen, O., Colgin, L.L., 2007. Cross-frequency coupling between neuronal oscillations. Trends Cogn. 991 Sci. 11, 267–269. https://doi.org/10.1016/j.tics.2007.05.003 992

Jensen, O., Mazaheri, A., 2010. Shaping Functional Architecture by Oscillatory Alpha Activity: 993 Gating by Inhibition. Front. Hum. Neurosci. 4. https://doi.org/10.3389/fnhum.2010.00186 994

Kalcher, J., Pfurtscheller, G., 1995. Discrimination between phase-locked and non-phase-locked 995 event-related EEG activity. Electroencephalogr. Clin. Neurophysiol. 94, 381–384. 996 https://doi.org/10.1016/0013-4694(95)00040-6 997

Kanske, P., Plitschka, J., Kotz, S.A., 2011. Attentional orienting towards emotion: P2 and N400 ERP 998 effects. Neuropsychologia 49, 3121–3129. 999 https://doi.org/10.1016/j.neuropsychologia.2011.07.022 1000

Kastner, S., Ungerleider, L.G., 2000. Mechanisms of Visual Attention in the Human Cortex. Annu. 1001 Rev. Neurosci. 23, 315–341. https://doi.org/10.1146/annurev.neuro.23.1.315 1002

Kelly, S.P., Lalor, E.C., Reilly, R.B., Foxe, J.J., 2006. Increases in Alpha Oscillatory Power Reflect an 1003 Active Retinotopic Mechanism for Distracter Suppression During Sustained Visuospatial 1004 Attention. J. Neurophysiol. 95, 3844–3851. https://doi.org/10.1152/jn.01234.2005 1005

Klimesch, W., Doppelmayr, M.M., Wimmer, H., Gruber, W.R., Röhm, D., Schwaiger, J., Hutzler, F., 1006 2001. Alpha and beta band power changes in normal and dyslexic children. Clin. 1007 Neurophysiol. 112, 1186–1195. https://doi.org/10.1016/S1388-2457(01)00543-0 1008

Klimesch, W., Sauseng, P., Hanslmayr, S., 2007. EEG alpha oscillations: The inhibition–timing 1009 hypothesis. Brain Res. Rev. 53, 63–88. https://doi.org/10.1016/j.brainresrev.2006.06.003 1010

Koelewijn, L., Dumont, J.R., Muthukumaraswamy, S.D., Rich, A.N., Singh, K.D., 2011. Induced and 1011 evoked neural correlates of orientation selectivity in human visual cortex. NeuroImage 54, 1012 2983–2993. https://doi.org/10.1016/j.neuroimage.2010.11.045 1013

Kopell, N., Ermentrout, G.B., Whittington, M.A., Traub, R.D., 2000. Gamma rhythms and beta 1014 rhythms have different synchronization properties. Proc. Natl. Acad. Sci. 97, 1867–1872. 1015 https://doi.org/10.1073/pnas.97.4.1867 1016

Lachaux, J.-P., Rodriguez, E., Martinerie, J., Varela, F.J., 1999. Measuring phase synchrony in brain 1017 signals. Hum. Brain Mapp. 8, 194–208. https://doi.org/10.1002/(SICI)1097-1018 0193(1999)8:4<194::AID-HBM4>3.0.CO;2-C 1019

Lalo, E., Gilbertson, T., Doyle, L., Di Lazzaro, V., Cioni, B., Brown, P., 2007. Phasic increases in 1020 cortical beta activity are associated with alterations in sensory processing in the human. Exp. 1021 Brain Res. 177, 146–146. https://doi.org/10.1007/s00221-006-0828-5 1022

Latora, V., Marchiori, M., 2001. Efficient Behavior of Small-World Networks. Phys. Rev. Lett. 87, 1023 198701. https://doi.org/10.1103/PhysRevLett.87.198701 1024

Lehmann, D., Skrandies, W., 1980. Reference-free identification of components of checkerboard-1025 evoked multichannel potential fields. Electroencephalogr. Clin. Neurophysiol. 48, 609–621. 1026 https://doi.org/10.1016/0013-4694(80)90419-8 1027

Lundqvist, M., Rose, J., Herman, P., Brincat, S.L., Buschman, T.J., Miller, E.K., 2016. Gamma and 1028 beta bursts underlie working memory. Neuron 90, 152–164. 1029

Maeno, T., Gjini, K., Iramina, K., Eto, F., Ueno, S., 2004. Event-related potential P2 derived from 1030 visual attention to the hemi-space. Source localization with LORETA. Int. Congr. Ser. 1270, 1031 262–265. https://doi.org/10.1016/j.ics.2004.04.034 1032

Maris, E., Oostenveld, R., 2007. Nonparametric statistical testing of EEG- and MEG-data. J. Neurosci. 1033 Methods 164, 177–190. https://doi.org/10.1016/j.jneumeth.2007.03.024 1034

Martínez-Cancino, R., Heng, J., Delorme, A., Kreutz-Delgado, K., Sotero, R.C., Makeig, S., 2019. 1035 Measuring transient phase-amplitude coupling using local mutual information. NeuroImage 1036 185, 361–378. https://doi.org/10.1016/j.neuroimage.2018.10.034 1037

Mathewson, K.E., Lleras, A., Beck, D.M., Fabiani, M., Ro, T., Gratton, G., 2011. Pulsed Out of 1038 Awareness: EEG Alpha Oscillations Represent a Pulsed-Inhibition of Ongoing Cortical 1039 Processing. Front. Psychol. 2. https://doi.org/10.3389/fpsyg.2011.00099 1040

Mazaheri, A., Jensen, O., 2010. Rhythmic pulsing: linking ongoing brain activity with evoked 1041 responses. Front. Hum. Neurosci. 4. https://doi.org/10.3389/fnhum.2010.00177 1042

Mazaheri, A., Picton, T.W., 2005. EEG spectral dynamics during discrimination of auditory and visual 1043 targets. Cogn. Brain Res. 24, 81–96. https://doi.org/10.1016/j.cogbrainres.2004.12.013 1044


https://doi.org/10.1101/2020.02.28.969253


33

Michel, C.M., Murray, M.M., Lantz, G., Gonzalez, S., Spinelli, L., Grave de Peralta, R., 2004. EEG 1045 source imaging. Clin. Neurophysiol. 115, 2195–2222. 1046 https://doi.org/10.1016/j.clinph.2004.06.001 1047

Milde, T., Leistritz, L., Astolfi, L., Miltner, W.H., Weiss, T., Babiloni, F., Witte, H., 2010. A new 1048 Kalman filter approach for the estimation of high-dimensional time-variant multivariate AR 1049 models and its application in analysis of laser-evoked brain potentials. Neuroimage 50, 960–1050 969. https://doi.org/10.1016/j.neuroimage.2009.12.110 1051

Minami, T., Noritake, Y., Nakauchi, S., 2014. Decreased beta-band activity is correlated with 1052 disambiguation of hidden figures. Neuropsychologia 56, 9–16. 1053 https://doi.org/10.1016/j.neuropsychologia.2013.12.026 1054

Mitra, P.P., Pesaran, B., 1999. Analysis of Dynamic Brain Imaging Data. Biophys. J. 76, 691–708. 1055 https://doi.org/10.1016/S0006-3495(99)77236-X 1056

Noonan, M.P., Adamian, N., Pike, A., Printzlau, F., Crittenden, B.M., Stokes, M.G., 2016. Distinct 1057 Mechanisms for Distractor Suppression and Target Facilitation. J. Neurosci. 36, 1797–1807. 1058 https://doi.org/10.1523/JNEUROSCI.2133-15.2016 1059

Oliveira, F.T.P., Hickey, C., McDonald, J.J., 2014. Proactive and Reactive Processes in the Medial 1060 Frontal Cortex: An Electrophysiological Study. PLoS ONE 9, e84351. 1061 https://doi.org/10.1371/journal.pone.0084351 1062

Oostenveld, R., Fries, P., Maris, E., Schoffelen, J.-M., 2011. FieldTrip: Open Source Software for 1063 Advanced Analysis of MEG, EEG, and Invasive Electrophysiological Data. Comput. Intell. 1064 Neurosci. 2011, 1–9. https://doi.org/10.1155/2011/156869 1065

Pagnotta, M.F., Dhamala, M., Plomp, G., 2018a. Benchmarking nonparametric Granger causality: 1066 Robustness against downsampling and influence of spectral decomposition parameters. 1067 NeuroImage 183, 478–494. https://doi.org/10.1016/j.neuroimage.2018.07.046 1068

Pagnotta, M.F., Dhamala, M., Plomp, G., 2018b. Assessing the performance of Granger–Geweke 1069 causality: Benchmark dataset and simulation framework. Data Brief 21, 833–851. 1070 https://doi.org/10.1016/j.dib.2018.10.034 1071

Pagnotta, M.F., Plomp, G., 2018. Time-varying MVAR algorithms for directed connectivity analysis: 1072 Critical comparison in simulations and benchmark EEG data. PLOS ONE 13, e0198846. 1073 https://doi.org/10.1371/journal.pone.0198846 1074

Pascual-Marqui, R., Biscay, R.J., Bosch-Bayard, J., Faber, P., Kinoshita, T., Kochi, K., Milz, P., 1075 Nishida, K., Yoshimura, M., 2017. Innovations orthogonalization: a solution to the major 1076 pitfalls of EEG/MEG “leakage correction.” https://doi.org/10.1101/178657 1077

Pascucci, D., Hervais-Adelman, A., Plomp, G., 2018. Gating by induced Α-Γ asynchrony in selective 1078 attention. Hum. Brain Mapp. 39, 3854–3870. https://doi.org/10.1002/hbm.24216 1079

Pascucci, D., Rubega, M., Plomp, G., 2019. Modeling time-varying brain networks with a self-tuning 1080 optimized Kalman filter (preprint). Neuroscience. https://doi.org/10.1101/856179 1081

Peirce, J., Gray, J.R., Simpson, S., MacAskill, M., Höchenberger, R., Sogo, H., Kastman, E., 1082 Lindeløv, J.K., 2019. PsychoPy2: Experiments in behavior made easy. Behav. Res. Methods 1083 51, 195–203. https://doi.org/10.3758/s13428-018-01193-y 1084

Penny, W.D., Duzel, E., Miller, K.J., Ojemann, J.G., 2008. Testing for nested oscillation. J. Neurosci. 1085 Methods 174, 50–61. https://doi.org/10.1016/j.jneumeth.2008.06.035 1086

Penny, W.D., Friston, K.J., Ashburner, J.T., Kiebel, S.J., Nichols, T.E., 2011. Statistical parametric 1087 mapping: the analysis of functional brain images. Elsevier. 1088

Perrin, F., Pernier, J., Bertrand, O., Echallier, J.F., 1989. Spherical splines for scalp potential and 1089 current density mapping. Electroencephalogr. Clin. Neurophysiol. 72, 184–187. 1090 https://doi.org/10.1016/0013-4694(89)90180-6 1091

Pfurtscheller, G., Lopes da Silva, F.H., 1999. Event-related EEG/MEG synchronization and 1092 desynchronization: basic principles. Clin. Neurophysiol. 110, 1842–1857. 1093 https://doi.org/10.1016/S1388-2457(99)00141-8 1094

Pfurtscheller, G., Stancák, A.A., Neuper, C., 1996. Post-movement beta synchronization. A correlate 1095 of an idling motor area? Electroencephalogr. Clin. Neurophysiol. 98, 281–293. 1096 https://doi.org/10.1016/0013-4694(95)00258-8 1097


https://doi.org/10.1101/2020.02.28.969253


34

Pogosyan, A., Gaynor, L.D., Eusebio, A., Brown, P., 2009. Boosting Cortical Activity at Beta-Band 1098 Frequencies Slows Movement in Humans. Curr. Biol. 19, 1637–1641. 1099 https://doi.org/10.1016/j.cub.2009.07.074 1100

Popov, T., Kastner, S., Jensen, O., 2017. FEF-Controlled Alpha Delay Activity Precedes Stimulus-1101 Induced Gamma-Band Activity in Visual Cortex. J. Neurosci. 37, 4117–4127. 1102 https://doi.org/10.1523/JNEUROSCI.3015-16.2017 1103

Rajan, A., Siegel, S.N., Liu, Y., Bengson, J., Mangun, G.R., Ding, M., 2018. Theta Oscillations Index 1104 Frontal Decision-Making and Mediate Reciprocal Frontal–Parietal Interactions in Willed 1105 Attention. Cereb. Cortex. https://doi.org/10.1093/cercor/bhy149 1106

Richter, C.G., Coppola, R., Bressler, S.L., 2018. Top-down beta oscillatory signaling conveys 1107 behavioral context in early visual cortex. Sci. Rep. 8. https://doi.org/10.1038/s41598-018-1108 25267-1 1109

Richter, C.G., Thompson, W.H., Bosman, C.A., Fries, P., 2017. Top-Down Beta Enhances Bottom-Up 1110 Gamma. J. Neurosci. 37, 6698–6711. https://doi.org/10.1523/JNEUROSCI.3771-16.2017 1111

Rubega, M., Carboni, M., Seeber, M., Pascucci, D., Tourbier, S., Toscano, G., Van Mierlo, P., 1112 Hagmann, P., Plomp, G., Vulliemoz, S., Michel, C.M., 2019. Estimating EEG Source Dipole 1113 Orientation Based on Singular-value Decomposition for Connectivity Analysis. Brain Topogr. 1114 https://doi.org/10.1007/s10548-018-0691-2 1115

Rubinov, M., Sporns, O., 2010. Complex network measures of brain connectivity: Uses and 1116 interpretations. NeuroImage 52, 1059–1069. https://doi.org/10.1016/j.neuroimage.2009.10.003 1117

Saalmann, Y.B., Pinsk, M.A., Wang, L., Li, X., Kastner, S., 2012. The Pulvinar Regulates Information 1118 Transmission Between Cortical Areas Based on Attention Demands. Science 337, 753–756. 1119 https://doi.org/10.1126/science.1223082 1120

Sanes, J.N., Donoghue, J.P., 1993. Oscillations in local field potentials of the primate motor cortex 1121 during voluntary movement. Proc. Natl. Acad. Sci. 90, 4470–4474. 1122 https://doi.org/10.1073/pnas.90.10.4470 1123

Schmiedt, J.T., Maier, A., Fries, P., Saunders, R.C., Leopold, D.A., Schmid, M.C., 2014. Beta 1124 Oscillation Dynamics in Extrastriate Cortex after Removal of Primary Visual Cortex. J. 1125 Neurosci. 34, 11857–11864. https://doi.org/10.1523/JNEUROSCI.0509-14.2014 1126

Schneider, S.L., Rose, M., 2016. Intention to encode boosts memory-related pre-stimulus EEG beta 1127 power. NeuroImage 125, 978–987. https://doi.org/10.1016/j.neuroimage.2015.11.024 1128

Schoenfeld, M.A., Hopf, J.-M., Martinez, A., Mai, H., Sattler, C., Gasde, A., Heinze, H.-J., Hillyard, 1129 S.A., 2007. Spatio-temporal Analysis of Feature-Based Attention. Cereb. Cortex 17, 2468–1130 2477. https://doi.org/10.1093/cercor/bhl154 1131

Schroeder, S.C.Y., Ball, F., Busch, N.A., 2018. The role of alpha oscillations in distractor inhibition 1132 during memory retention. Eur. J. Neurosci. 48, 2516–2526. https://doi.org/10.1111/ejn.13852 1133

Scolari, M., Ester, E.F., Serences, J.T., 2014. Feature- and Object-Based Attentional Modulation in the 1134 Human Visual System. Oxford University Press. 1135 https://doi.org/10.1093/oxfordhb/9780199675111.013.009 1136

Seeber, M., Cantonas, L.-M., Hoevels, M., Sesia, T., Visser-Vandewalle, V., Michel, C.M., 2019. 1137 Subcortical electrophysiological activity is detectable with high-density EEG source imaging. 1138 Nat. Commun. 10, 753. https://doi.org/10.1038/s41467-019-08725-w 1139

Serences, J.T., Yantis, S., 2007. Spatially Selective Representations of Voluntary and Stimulus-Driven 1140 Attentional Priority in Human Occipital, Parietal, and Frontal Cortex. Cereb. Cortex 17, 284–1141 293. https://doi.org/10.1093/cercor/bhj146 1142

Seth, A.K., Barrett, A.B., Barnett, L., 2015. Granger Causality Analysis in Neuroscience and 1143 Neuroimaging. J. Neurosci. 35, 3293–3297. https://doi.org/10.1523/JNEUROSCI.4399-1144 14.2015 1145

Siegel, M., Warden, M.R., Miller, E.K., 2009. Phase-dependent neuronal coding of objects in short-1146 term memory. Proc. Natl. Acad. Sci. 106, 21341–21346. 1147 https://doi.org/10.1073/pnas.0908193106 1148

Simoncelli, E.P., Olshausen, B.A., 2001. Natural Image Statistics and Neural Representation. Annu. 1149 Rev. Neurosci. 24, 1193–1216. https://doi.org/10.1146/annurev.neuro.24.1.1193 1150


https://doi.org/10.1101/2020.02.28.969253


35

Sladky, R., Friston, K.J., Tröstl, J., Cunnington, R., Moser, E., Windischberger, C., 2011. Slice-timing 1151 effects and their correction in functional MRI. NeuroImage 58, 588–594. 1152 https://doi.org/10.1016/j.neuroimage.2011.06.078 1153

Snyder, A.C., Foxe, J.J., 2010. Anticipatory attentional suppression of visual features indexed by 1154 oscillatory alpha-band power increases: a high-density electrical mapping study. J. Neurosci. 1155 30, 4024–4032. 1156

Spitzer, B., Haegens, S., 2017. Beyond the Status Quo: A Role for Beta Oscillations in Endogenous 1157 Content (Re)Activation. eneuro 4, ENEURO.0170-17.2017. 1158 https://doi.org/10.1523/ENEURO.0170-17.2017 1159

Swann, N., Tandon, N., Canolty, R., Ellmore, T.M., McEvoy, L.K., Dreyer, S., DiSano, M., Aron, 1160 A.R., 2009. Intracranial EEG Reveals a Time- and Frequency-Specific Role for the Right 1161 Inferior Frontal Gyrus and Primary Motor Cortex in Stopping Initiated Responses. J. Neurosci. 1162 29, 12675–12685. https://doi.org/10.1523/JNEUROSCI.3359-09.2009 1163

Takahashi, D.Y., Baccalá, L.A., Sameshima, K., 2010. Information theoretic interpretation of 1164 frequency domain connectivity measures. Biol. Cybern. 103, 463–469. 1165 https://doi.org/10.1007/s00422-010-0410-x 1166

Toppi, J., Babiloni, F., Vecchiato, G., De Vico Fallani, F., Mattia, D., Salinari, S., Bolton, T.A., 1167 Leistritz, L., Witte, H., Astolfi, L., 2012. Towards the time varying estimation of complex 1168 brain connectivity networks by means of a General Linear Kalman Filter approach, in: 1169 Engineering in Medicine and Biology Society (EMBC), 2012 Annual International 1170 Conference of the IEEE. IEEE, pp. 6192–6195. https://doi.org/10.1109/EMBC.2012.6347408 1171

Torrence, C., Compo, G.P., 1998. A Practical Guide to Wavelet Analysis. Bull. Am. Meteorol. Soc. 1172 79, 61–78. https://doi.org/10.1175/1520-0477(1998)079<0061:APGTWA>2.0.CO;2 1173

Tort, A.B.L., Komorowski, R., Eichenbaum, H., Kopell, N., 2010. Measuring Phase-Amplitude 1174 Coupling Between Neuronal Oscillations of Different Frequencies. J. Neurophysiol. 104, 1175 1195–1210. https://doi.org/10.1152/jn.00106.2010 1176

Van de Steen, F., Faes, L., Karahan, E., Songsiri, J., Valdes-Sosa, P.A., Marinazzo, D., 2016. Critical 1177 Comments on EEG Sensor Space Dynamical Connectivity Analysis. Brain Topogr. 1178 https://doi.org/10.1007/s10548-016-0538-7 1179

Van Diepen, R.M., Foxe, J.J., Mazaheri, A., 2019. The functional role of alpha-band activity in 1180 attentional processing: The current zeitgeist and future outlook. Curr. Opin. Psychol. 1181 https://doi.org/10.1016/j.copsyc.2019.03.015 1182

van Diepen, R.M., Mazaheri, A., 2017. Cross-sensory modulation of alpha oscillatory activity: 1183 suppression, idling, and default resource allocation. Eur. J. Neurosci. 45, 1431–1438. 1184 https://doi.org/10.1111/ejn.13570 1185

Van Veen, B.D., Van Drongelen, W., Yuchtman, M., Suzuki, A., 1997. Localization of brain electrical 1186 activity via linearly constrained minimum variance spatial filtering. IEEE Trans. Biomed. Eng. 1187 44, 867–880. https://doi.org/10.1109/10.623056 1188

Van Voorhis, S., Hillyard, S.A., 1977. Visual evoked potentials and selective attention to points in 1189 space. Percept. Psychophys. 22, 54–62. https://doi.org/10.3758/BF03206080 1190

von Stein, A., Sarnthein, J., 2000. Different frequencies for different scales of cortical integration: 1191 from local gamma to long range alpha/theta synchronization. Int. J. Psychophysiol. 38, 301–1192 313. https://doi.org/10.1016/S0167-8760(00)00172-0 1193

Voytek, B., Canolty, R.T., Shestyuk, A., Crone, N., Parvizi, J., Knight, R.T., 2010. Shifts in gamma 1194 phase–amplitude coupling frequency from theta to alpha over posterior cortex during visual 1195 tasks. Front. Hum. Neurosci. 4. https://doi.org/10.3389/fnhum.2010.00191 1196

Voytek, B., D’Esposito, M., Crone, N., Knight, R.T., 2013. A method for event-related 1197 phase/amplitude coupling. NeuroImage 64, 416–424. 1198 https://doi.org/10.1016/j.neuroimage.2012.09.023 1199

Watson, A.B., Pelli, D.G., 1983. Quest: A Bayesian adaptive psychometric method. Percept. 1200 Psychophys. 33, 113–120. https://doi.org/10.3758/BF03202828 1201

Worden, M.S., Foxe, J.J., Wang, N., Simpson, G.V., 2000. Anticipatory Biasing of Visuospatial 1202 Attention Indexed by Retinotopically Specific α-Bank Electroencephalography Increases over 1203 Occipital Cortex. J. Neurosci. 20, RC63–RC63. https://doi.org/10.1523/JNEUROSCI.20-06-1204 j0002.2000 1205


https://doi.org/10.1101/2020.02.28.969253


36

Yordanova, J., Kolev, V., Polich, J., 2001. P300 and alpha event-related desynchronization (ERD). 1206 Psychophysiology 38, 143–152. https://doi.org/10.1111/1469-8986.3810143 1207

Zhang, W., Luck, S.J., 2009. Feature-based attention modulates feedforward visual processing. Nat. 1208 Neurosci. 12, 24–25. https://doi.org/10.1038/nn.2223 1209

1210

1211


https://doi.org/10.1101/2020.02.28.969253


37

Supplementary figures 1212

1213

1214 Figure S1. Attentional modulation of brain rhythms: comparison between Attended-orientation and Unattended. 1215 Results of sensor-space power spectra analysis are shown for anticipatory (A) and reactive (C) time intervals. Each figure 1216 represents the distribution of statistically significant differences (positive t-values for Attended-orientation higher than 1217 Unattended, and vice versa negative t-values) from cluster-based permutation (two-tailed t-test with p<0.05, 50000 1218 permutations, and p<0.05 for the permutation test), together with the topographical distributions of these differences in the α-1219 band (bottom) and β-band (top). B) The polar histogram represents the effect sizes for each ROI of source-space power 1220 differences between conditions in the β-band (15–30 Hz). D) The figure shows the results of source-space time-varying 1221 power spectra analysis, with the distribution of statistically significant differences from cluster-based permutation (two-tailed 1222 t-test with p<0.05, 50000 permutations, and p<0.05 for the permutation test). The marginal plots show time- or frequency-1223 collapsed distributions of significant differences in power spectra. The polar histograms on the right represent the effect sizes 1224 for each ROI of power differences between conditions, in the following frequency bands: α (5–12 Hz), β (15–30 Hz), and γ 1225 (45–100 Hz). Effect sizes were estimated using Cohen's d (Cohen, 1992). 1226

1227


https://doi.org/10.1101/2020.02.28.969253


38

1228

1229 Figure S2. Anticipatory differences in local PAC: control analysis based on stratification (1000 iterations). Differences 1230 in within-region anticipatory PAC are shown for each pair fPHASE– fAMPLITUDE, for Attended-motion compared to Unattended 1231 (A) and for Attended-orientation compared to Unattended (B). The figure reports only the regions that showed significant 1232 differences from the cluster-based permutation comparison (two-tailed t-test with p<0.05, 50000 permutations, and p<0.05 1233 for the permutation test), which are highlighted as more opaque. The black areas indicate fPHASE–fAMPLITUDE pairs excluded 1234 from the analysis (see Methods). 1235

1236


https://doi.org/10.1101/2020.02.28.969253


39

1237

1238 Figure S3. Stimulus-evoked differences in network efficiency. Differences between Attended-orientation and Unattended 1239 are shown for three frequency bands, A) α-band, B) β-band, and C) γ-band. Left panels show global network efficiency, right 1240 panels local efficiency. Gray shading represents the standard error of the mean. Green vertical shades represent latencies that 1241 showed statistically significant results in the two-tailed t-test (pFDR<0.05). 1242

1243


https://doi.org/10.1101/2020.02.28.969253


Nested oscillations and brain connectivity during ... · 34 between brain regions (Corbetta and...

Documents

Transcript of Nested oscillations and brain connectivity during ... · 34 between brain regions (Corbetta and...