Research ArticleNEUROSCIENCE

Context-specific modulation of intrinsic coupling modes shapes multisensory processing

See allHide authors and affiliations

Science Advances  10 Apr 2019:
Vol. 5, no. 4, eaar7633
DOI: 10.1126/sciadv.aar7633


Intrinsically generated patterns of coupled neuronal activity are associated with the dynamics of specific brain states. Sensory inputs are extrinsic factors that can perturb these intrinsic coupling modes, creating a complex scenario in which forthcoming stimuli are processed. Studying this intrinsic-extrinsic interplay is necessary to better understand perceptual integration and selection. Here, we show that this interplay leads to a reconfiguration of functional cortical connectivity that acts as a mechanism to facilitate stimulus processing. Using audiovisual stimulation in anesthetized ferrets, we found that this reconfiguration of coupling modes is context specific, depending on long-term modulation by repetitive sensory inputs. These reconfigured coupling modes lead to changes in latencies and power of local field potential responses that support multisensory integration. Our study demonstrates that this interplay extends across multiple time scales and involves different types of intrinsic coupling. These results suggest a previously unknown large-scale mechanism that facilitates multisensory integration.


Large-scale patterns of intrinsically generated coupling are a hallmark of brain networks. These coupling patterns, which we term intrinsic coupling modes (ICMs) (1), have gained attention since they may have crucial importance for brain function (19). ICMs correspond to dynamic coupling patterns, which are not imposed by entrainment to an external stimulus or movement but emerge from the connectivity of cortical and subcortical networks (1). Substantial evidence suggests two different mechanisms for ICMs. One type (which we term phase ICMs) arises from phase coupling of band-limited oscillatory signals, whereas the other results from coupled aperiodic fluctuations of signal envelopes (hence, termed envelope ICMs) (1). ICMs occur both in ongoing activity and during tasks. Their impact has been observed in the modulation of sensory responses (1013), behavior, and perception (1418), and furthermore, ICMs play a key role in mediating effective and selective communication in networks of cortical and subcortical areas (1922).

Cortical activity is believed to arise from the complex interplay between the intrinsic dynamics and the continuous stream of sensory information impinging from the external environment (23, 24). Understanding this interplay is fundamental for elucidating the role of ICMs in information processing. In a classical approach, the global organization of neural activity, or brain state, has been viewed as the arena or playground on which extrinsic sensory inputs are processed (2, 2427). The opposite, that is, the effects of sensory stimuli on ongoing activity, is less well understood. Whereas Fiser et al. (5) found that, during visual stimulation, most of the spatiotemporal correlations in visual cortex are driven by the network and not by the stimulus, a study by He (28) using functional magnetic resonance imaging (fMRI) revealed a negative interaction between ongoing and evoked activity. Thus, the large-scale interaction between intrinsic and externally driven dynamics is still far from being understood.

By using a 64-channel electrocorticographic (ECoG) array, we study here the interplay between large-scale ICMs and extrinsically evoked activity during repeated multisensory stimulation in the anesthetized ferret. To elucidate this interplay, two complementary aspects were analyzed. On the one hand, we investigated how visual, auditory, and audiovisual stimuli modulated ICMs. On the other hand, we tested whether phase or envelope ICMs, extracted from prestimulus intervals, predicted stimulus-related multisensory effects. We found that prestimulus functional connectivity affects both response timing and power and that this effect extends over several frequency bands of cortical activity. This causal correlation is itself stimulus context dependent, suggesting an adaptation in functional connectivity giving rise to ICMs. Our results suggest that the role of this connectivity adaptation is to facilitate multisensory integration in the cortex.


General procedures related to surgical preparation and data acquisition have been described in detail elsewhere (22). Briefly, we recorded cortical activity in ferrets using custom-built ECoG arrays (29) with 64 electrodes distributed to cover the posterior cortex of the left hemisphere in the ferret (Fig. 1, A and B, and fig. S1, A and B). For the present study, we used a total of five adult female anesthetized ferrets (Mustela putorius) (see Materials and Methods). We collected data during stimulus presentation (see Materials and Methods) and from periods of ongoing activity. For the stimulus-related study, we used two classes of stimulus blocks: transient stimuli (clicks and flashes) and sustained complex stimuli (drifting Gabor patches and auditory ripples). This allowed us to determine whether possible changes in ICMs are stimulus specific. For statistical analysis, we grouped the responsive sites into auditory areas [primary auditory cortex (A1), anterior auditory field (AAF), anterior dorsal field (ADF), anterior ventral field (AVF), posterior suprasylvian field (PSF), and posterior pseudosylvian field (PPF)] and visual areas (17, 18, 19, 20, 21, and suprasylvian visual areas) (Fig. 1A).

Fig. 1 Effects of multisensory stimulation on response timing and power.

(A) Map of functional areas. A1, primary auditory cortex; AAF, anterior auditory field; PPF, posterior pseudosylvian field; PSF, posterior suprasylvian field; ADF, anterior dorsal field; AVF, anterior ventral field; VP, ventroposterior area; 3b, primary somatosensory cortex; S2, secondary somatosensory cortex; S3, tertiary somatosensory cortex; PPc, posterior parietal caudal; PPr, posterior parietal rostral; SSY, suprasylvian visual areas. (B) Schematic of the ECoG array placed on the left hemisphere of the ferret brain, covering most of the occipital, temporal, and parietal cortex. Red dots represent recording sites, and white circles represent the holes in the foil. lat, lateral sulcus; sss, suprasylvian sulcus; ps, pseudosylvian sulcus. (C) Topographic distribution of response amplitudes with unimodal (A: clicks and V: flashes) and bimodal (AV: simultaneous clicks and flashes) stimuli. Dark colors represent the amplitude of the strongest deflection in the event-related potential (ERP) within the first 80 ms after stimulus onset. (D) Topographic distribution of changes of total power in the alpha band in response to auditory (left), visual (middle), and audiovisual (right) stimuli. Power change was averaged over a stimulus time window (100 to 600 ms, relative to stimulus onset) and normalized to the prestimulus interval (−600 to −100 ms). (E) Latency reduction. Scatterplot of latencies of responses to unimodal versus bimodal stimulation. Blue circles represent data from electrodes that responded to unimodal visual stimuli but not unimodal auditory stimuli; red circles indicate electrodes that responded in the unimodal condition only to auditory stimuli. Sites that responded to both unimodal visual and unimodal auditory stimuli were not included. Top right: Probability distribution of the variation index (difference between latencies with unimodal and bimodal stimuli). Elements with low variation index are located along the diagonal. Bottom: Probability distribution of unimodal latencies. Left: Probability distribution of bimodal latencies. (F) Scatterplot of power enhancement (z score) during unimodal (ripples or drifting Gabor patches) and bimodal stimulation across responsive electrodes for theta band (4 to 8 Hz, green circles) and alpha band (8 to 16 Hz, black circles) for sustained stimuli. Continuous lines and filled diamonds correspond to visual areas, while dashed lines and open circles correspond to auditory areas. Top right: Normalized probability distribution of the variation index (difference between unimodal and bimodal power enhancements). Bottom and left: Panels showing probability distributions of power enhancement during unimodal and bimodal stimuli, respectively.

Multisensory interactions modulate response latency and power

First, to study the role of ICMs in stimulus processing, it was necessary to characterize the spatial distribution and the multisensory properties of cortical sensory responses in the ferret. Figure 1C shows the spatial distribution of amplitudes of event-related potentials (ERPs) to auditory, visual, and combined audiovisual stimulation in a single animal. Across animals, the topography of cortical responses was in agreement with previous studies on auditory (30, 31) and visual stimulation in the ferret (32). A total of 92 recorded sites responded to auditory stimuli across animals, while 118 sites responded to visual stimuli. To test for multisensory effects on response timing, we compared the ERP latencies between unimodal and bimodal stimuli (see Materials and Methods). Visual responses showed higher latency variability compared to auditory responses (Avar = 139 ms and Vvar = 45 ms; Brown-Forsythe test, P < 0.0001). Bimodal stimulation resulted in a significant latency reduction for visually responsive sites (mean, V = 40.2 ms and AV = 34.3 ms; sign test, P = 1.7 × 10−5) but not for sites with auditory responses (A = 24.6 ms and AV = 20.3 ms; sign test, P = 0.44). This effect was particularly pronounced for sites along the suprasylvian sulcus and a region at the pseudosylvian sulcus, which responded to visual stimuli, and has been suggested as the homolog of the cat’s anterior ectosylvian visual area located in the multisensory anterior ectosylvian sulcus (AES) (31, 33, 34).

In addition, we analyzed the spectral distribution of stimulus-related changes in normalized total power (see Materials and Methods). As shown by mean time–frequency spectrograms (fig. S2B), we observed stimulus-related changes in total power across a broad range of frequencies with all stimuli used. Figure 1D illustrates the topography of cortical stimulus-related total power in the alpha band (8 to 16 Hz) in a single animal. In general, although the spatial distribution of responses agreed well with known functional properties of cortical areas, the number of responsive sites in auditory areas varied across frequency bands (figs. S3 and S4).

Multisensory effects in local field potential (LFP) power responses were quantified by the difference AV − (A+V), with A, V, and AV being the LFP normalized total power in response to auditory, visual, and audiovisual stimulation, respectively. To test the effects of sensory stimulation, we used blocks with either sustained stimuli (ripples and/or gratings, R-G) or short transient stimuli (clicks and/or flashes, C-F; for details, see Materials and Methods). In transient stimulation blocks, multisensory effects in auditory areas were frequency dependent [one-way analysis of variance (ANOVA), F1,130 = 6.5, P < 0.001 with Bonferroni correction for frequency] and mostly dominated by subadditive effects across frequency bands (fig. S3E). Superadditive effects occurred more rarely at both low and high frequencies. Similarly, in visual areas, the subadditive effect dominated across both stimulus blocks and frequency bands (fig. S3F). In sustained stimulation blocks, multisensory effects were frequency dependent (Fig. 1F and fig. S3, G and H); in particular, the difference in power (AV − A-V) in alpha, beta, and gamma bands was significantly lower than zero (t test with Bonferroni correction, P < 0.001). The population distribution of multisensory responses and the topography of the multisensory effects across all frequency bands are shown in figs. S3 and S4. Together, our results show that multisensory effects on response power are dominated by subaddition, with less sites showing superadditivity. Similar observations have been made by Kayser et al. (35) in auditory cortex of monkeys. Moreover, our data show that the multisensory effects on the power of responses differ between the two types of stimulus blocks for both auditory and visual areas (one-way ANOVA, F1,334 = 6.6, P = 0.01 and F1,600 = 17.6, P < 0.0001, respectively).

Sensory stimuli have limited impact on ICMs

A key goal of our work was to study whether sensory stimulation modulates ICMs in the cortex. We tested this first for envelope ICMs, which were quantified by correlation of the signal amplitudes. For computing this coupling measure, we used the orthogonalized components of the respective signals to eliminate amplitude envelope correlations introduced by volume conduction (36).

We computed connectivity matrices for periods between 100 and 600 ms after stimulus onset for the stimulation blocks with sustained and those with transient stimulation. We reordered and grouped electrodes in the array according to their modular structure to build the connectivity matrix (Fig. 2). Matrices in Fig. 2 show the patterns of connectivity during the presentation of auditory, visual, and audiovisual stimulation. To evaluate whether these correlations represent functional connectivity, or rather artifactual correlations through common inputs, we compared these matrices with the connectivity obtained after shuffling trials within each stimulus modality. Connections that were not significantly different from these shuffle controls are marked with a cross (x) in the coupling matrices in Figs. 2B and 3 (A and B).

Fig. 2 Stimulus-related modulation of envelope ICMs.

Coupling matrices represent the strength of amplitude envelope correlations between functional areas. The labeling of rows and columns, as shown in the top left in (A), applies to all matrices in the figure. On the basis of the anatomical regions (abbreviations as in Fig. 1), these can be grouped in three modalities: visual, somatosensory parietal, and auditory regions (black dark squares in top left matrix). (A) Connectivity during stimulation with ripples (left column, A), drifting Gabor patches (second column, V), and simultaneous presentation of ripples and Gabor patches (third column, AV). Different rows represent different frequency bands. Note that matrices for the high gamma and the high frequency bands are not shown. (B) Connectivity during unimodal (A, V) and bimodal (AV) stimulation with clicks and flashes for different frequency bands (rows). Connections that were not significantly different from the shuffle controls are marked with an “x.” (C) Normalized degree (left) and betweenness (right) for connectivity matrices shown in (A). Each subplot represents the mean and standard error of the respective graph measure associated with auditory (red), visual (blue), and audiovisual (green) stimulation. *P < 0.05; ***P < 0.001. (D) Normalized degree and betweenness for connectivity matrices shown in (B).

Fig. 3 Stimulus-related modulation of phase ICMs.

(A) Matrices from the first to the third column represent the mean imaginary coherence between areas during stimulation with auditory ripples (A), drifting Gabors (V), or both (AV). (B) Connectivity during unimodal (A, V) and bimodal (AV) stimulation with clicks and flashes. In (A) and (B), connections that were not significantly different from the shuffle controls are marked with an “x” in the coupling matrices. (C) Normalized degree and betweenness connectivity matrices shown in (A). **P < 0.01; ***P < 0.001. (D) Normalized degree and betweenness for connectivity matrices shown in (B).

Strongest correlations appeared mostly within clusters of functionally related areas rather than across modalities. Furthermore, within blocks, the connectivity did not differ significantly between stimulation conditions (Fig. 2, A and B). In general, we observed higher envelope ICMs for low frequency (theta, alpha, and beta) compared to high frequency bands. Within the sustained block, the connectivity matrices did not differ significantly, on an area-by-area basis, from those in the prestimulation intervals (Mann-Whitney U test at a two-sided P < 0.05, Bonferroni-corrected) (Fig. 2A, cf. Fig. 4A). In this stimulation block, all connections were significantly stronger than the shuffled condition.

Fig. 4 Prestimulus connectivity matrices.

(A) Envelope ICMs. Left column shows the average matrix across animals of the prestimulus connectivity obtained during the sustained stimulation blocks (pre–R-G). Middle column shows the prestimulus connectivity matrix for the blocks with clicks and flashes (pre–C-F). Right column: Connectivity in the ongoing condition, i.e., in a recording block without intermittent sensory stimulation. Rows represent the frequency bands as defined in Materials and Methods. (B) Phase ICMs. Matrices show imaginary coherence in pre–R-G, pre–C-F, and ongoing activity blocks for the same recordings epochs as in (A).

We used graph theoretical measures to investigate whether sensory stimulation alters the topology of ICMs. In principle, this could take place in different ways, for example, through a global modulation in strength of connectivity or through topological reorganization of the functional network architecture. To quantify these effects, we computed the average connectivity strength (degree), prevalence of hubs (betweenness), and network segregation (clustering coefficient) (37). The strength of connectivity was not affected by stimulus modality (two-way ANOVA, P = 0.93); however, it did show a dependency across frequency bands (two-way ANOVA, F1,189 = 22.2, P = 0.04, Bonferroni-corrected for frequency). We observed the strongest correlations in the alpha-band, where degree was significantly higher than in the gamma bands (t test, P = 1.3 × 10−6 and P = 2.07 × 10−7, Bonferroni-corrected) and high frequency band (t test, P = 1.2 × 10−8, Bonferroni-corrected). In contrast, betweenness was affected neither by stimulus modality nor frequency bands (two-way ANOVA, P > 0.3 in both cases) (Fig. 2C). Last, stimulus block showed no effect in the network segregation across frequency bands (fig. S5).

Applying the same analysis to transient stimulation blocks, we observed, again, that connectivity was not influenced by sensory modality (Fig. 2B). As in the sustained block, coupling was higher within functional cortical systems. In the transient stimulation blocks, some connections were not significantly stronger than the shuffled condition, especially at high frequencies (Fig. 2B). However, these connections were rather weak, suggesting that functional coupling was generally not introduced through common inputs. Both graph measures in Fig. 2D did not appear to reflect the stimulus modality (two-way ANOVA, F1,107 = 0.11, P = 0.89). Here, again, average connectivity strength was frequency band specific (two-way ANOVA, F1,107 = 6.9, P < 0.01, Bonferroni-corrected). Last, a comparison between stimulus blocks showed a significant effect on connectivity strength (one-way ANOVA, F1,197 = 23.2, P < 10−5, Bonferroni-corrected) and on clustering properties (fig. S5).

In the same dataset, phase ICMs were quantified by using the imaginary part of the coherence (Fig. 3) (38). This analysis revealed connectivity patterns that strongly differed from those obtained by envelope correlation. We mainly found strongest phase ICMs in the theta and alpha bands. They were most prominent among auditory areas and, in the C-F blocks, also between visual and auditory regions. As for envelope ICMs, the impact of stimulation modality on phase ICMs was not significant neither in R-G nor C-F blocks (Fig. 3, C and D). In contrast, connectivity strength, betweenness, and clustering coefficient appeared to be suitable to detect differences between stimulation blocks (two-way ANOVA, F1,214 = 21.4, P < 0.001 and F1,214 = 17.1, P < 0.001, respectively). Furthermore, clustering coefficients (fig. S5) significantly differed between frequency bands (two-way ANOVA, P < 10−5) but not between stimulus modalities. As shown by cross (x) labeling in Fig. 3 (A and B), we also observed connections for which phase coupling did not differ significantly from shuffle controls. However, this only held for connections with very low values of imaginary coherence.

Together, our data consistently show that, during sensory stimulation, significant functional connectivity patterns can be observed, which exhibit spatial specificity and differ across frequency bands. However, these coupling patterns do not reflect the modality of the presented stimulus. A possible explanation for this unexpected result might be that the types of ICMs investigated here mostly reflect the underlying structural connectivity of the networks under study. Alternatively, they might be shaped by factors that influence functional connectivity on a longer time scale, such as changes in neuromodulatory systems, more slowly occurring state changes in the network, or contextual effects resulting from repeated sensory stimulation over extended time periods. In the latter case, a relevant parameter in our dataset was the stimulus block, i.e., the repeated exposure to sustained or transient stimuli.

Prestimulus ICMs are context specific

We therefore decided to test whether the connectivity patterns during intervals with no stimulation were dependent on the stimulus block they were embedded in. We hypothesized that, if these differed between blocks, then this would support the idea of modulation of connectivity through contextual effects. The time interval was the same duration as before (500 ms) but now taken from −600 to −100 ms relative to stimulus onset. We analyzed connectivity matrices for prestimulus intervals across stimulation blocks and frequency bands using the same connectivity measures as used before (Fig. 4). In general, we observed strong differences in the spatial patterning of envelope and phase ICMs. In most frequency bands, envelope ICMs showed a clustering of areas into functional systems. This was particularly pronounced, e.g., for the prestimulus intervals of sustained stimuli in the alpha band (Fig. 4A). For envelope ICMs, there were strong differences between the mean connectivity strengths of the two stimulus blocks, especially at low frequencies (t test, P < 10−8 for all cases, Bonferroni-corrected). In contrast, for phase ICMs, the mean strength in prestimulus intervals of the transient stimulation blocks were stronger for theta, alpha, and beta bands (t test, P < 10−5 in all cases) (Fig. 4B). However, for gamma bands and the high frequency band, coherence was higher preceding sustained stimuli than transient stimuli (t test, P < 10−5). For both envelope and phase ICMs, rescaling of the connectivity matrices by normalizing to the mean connectivity (fig. S6) emphasized the similarity of coupling patterns across frequency bands within each of the two ICM classes, suggesting that part of the difference emerges from a modulation of connectivity strengths. It should be noted that graph theoretical analysis for the prestimulus intervals generally yielded the same results as that for the stimulation epochs (Fig. 5).

Fig. 5 Graph theoretical analysis of prestimulus connectivity.

Mean degree, betweenness, and clustering coefficient were used to characterize the ICMs in intervals before stimulus onset. (A to C) Three measures for envelope ICMs during R-G stimuli (orange) and C-F blocks (blue) across frequency bands. Significant differences occur at low frequencies, in particular, theta, alpha, and beta bands. Asterisks represent P < 0.001. (D to F) The graph theoretical measures characterizing the prestimulus intervals in C-F and R-G blocks. The data presented are the means ± SD across animals.

We checked whether these differences in prestimulus connectivity were a consequence of differences in the power spectral distribution between prestimulation conditions. Figure S2A shows the mean power distributions of both stimulus conditions across animals and recording sites. Both stimulus conditions showed, on average, very similar distributions with no significant differences across the frequency ranges of interest.

The observations on the prestimulus blocks led us to investigate the global structure of connectivity patterns during long periods devoid of stimulation. In addition to the sensory stimulation recordings, in each animal, we recorded three to four periods of ongoing activity randomly positioned relative to the stimulation blocks. From these periods, which were of at least 15-min duration, we sampled 100 randomly distributed epochs of 500-ms duration to match the trial structure during the stimulation blocks. We did not extract epochs during the first 3 min of the ongoing activity blocks to minimize any possible effect resulting from preceding stimulation. The comparison of the connectivity matrices for ongoing activity blocks and prestimulus intervals during the stimulation blocks yielded a number of observations (Fig. 4). First, envelope ICMs in ongoing activity resembled, in terms of their strength, more closely those in presustained than in pretransient intervals. Within the auditory areas, the presustained connectivity showed a slight decorrelation relative to the ongoing condition (e.g., sustained mean = 0.08 versus ongoing mean = 0.150; t test, P < 10−6), which extended across all frequency bands (Fig. 4). In contrast, phase ICMs during ongoing activity showed larger similarity in strength to the pretransient condition at low frequencies. The analysis of the connectivity matrices after normalization to the mean connectivity in each matrix (fig. S6) revealed that, for envelope ICMs, the spatial pattern of coupling was qualitatively similar across the different blocks for some frequency bands (e.g., the low and high gamma band), whereas for other bands, differences were still apparent (e.g., the alpha band).

Together, our results indicate that it is highly probable that differences in connectivity between stimulus blocks are not the consequence of network responses to individual stimuli but the result of a more global change that may relate to the stimulation context on longer time scales. This suggests that, during the presentation of a train of stimuli, the brain undergoes a context-specific reconfiguration of functional networks that can differ from the spatial pattern of coupling modes that prevails during long periods without any sensory stimulation.

Prestimulus ICMs predict multisensory response effects

In the final step of our analysis, we investigated the role of these context-specific network reconfiguration. We hypothesized that the multisensory effects observed during stimulus processing, as reflected in the latency and power changes described above, might be a consequence of this functional reconfiguration. For both types of ICMs, we tested whether there was a correlation between the strength in the prestimulus connectivity and the change in latency (latV-latAV) observed in visually responsive sites (cf. Figs. 6, A and B; 1E). We included prestimulus connectivity between sites that responded only to visual stimulation and sites that responded exclusively to auditory stimuli. For these connections, which involved mostly auditory and early visual areas (fig. S7, A and B), we computed the Spearman correlation between prestimulus connectivity strength and the change in latencies during the subsequent response. We observed that this correlation was specific to both frequency bands and coupling measures (Fig. 6). During the blocks with transient sensory stimulation, beta and low gamma envelope ICMs induced a reduction of latencies in visual responses (P < 10−4, Pearson’s correlation with Bonferroni correction; Fig. 6A). Phase ICMs predicted a latency reduction mainly during the sustained stimulation blocks (Fig. 6B). Here, change in response latencies was positively correlated with the prestimulus phase coupling strength in alpha, beta, and low gamma bands (P < 10−3, Pearson’s correlation with Bonferroni correction). The overall comparison of these results across stimulus blocks and coupling measures showed that, except for the correlation in beta band, all significant correlations were either in one or the other coupling mode, suggesting that phase and envelope ICMs may reflect independent coupling mechanisms that potentially differ in the functional relevance.

Fig. 6 Prestimulus connectivity predicts multisensory effects on response timing and power.

(A) Correlation between envelope coupling and change in latencies (latV-latAV) for visually responsive sites in the R-G (orange) and C-F (blue) stimulation blocks. Asterisks represent significant correlations (P < 0.01), and error bars represent the standard error of the correlation. (B) Correlation of differences in latency with phase coupling. (C) Correlation between prestimulus envelope coupling and multisensory power enhancement observed with ripples and Gabor patches (R-G). Each element in the matrix represents the strength of the correlation between the prestimulus connectivity in a certain band and the stimulus-related power enhancement in a specific band. Hot colors represent positive correlations, and cold colors represent negative correlations. Nonsignificant correlations (P > 0.01) are masked. (D) Correlation matrix between prestimulus phase coupling and stimulus-related power changes with R-G. (E and F) Same analyses of the relation between prestimulus connectivity and multisensory power enhancement for the stimulation blocks with clicks and flashes (C-F).

Following a similar hypothesis, we studied whether the strength of prestimulus connectivity predicted multisensory effects on response power. Note that, in contrast to latency, both response power and connectivity can be defined in a spectrally specific manner. Therefore, results of this analysis are displayed as a matrix in which rows represent the response power frequency band and columns represent the prestimulus connectivity frequency band (Fig. 6, C to F). The results show specific patterns of effects differing across coupling modes, frequency bands, and stimulation blocks. In the sustained stimulation blocks, prestimulus envelope ICMs in the low and high gamma band were positively correlated with multisensory responses in the low gamma band (Fig. 6C) (Pearson’s correlation r = 0.6, P < 0.01, Bonferroni-corrected for number of elements of the matrix). In contrast, high-frequency envelope ICMs negatively correlated with multisensory response power changes (Fig. 6C). These effects were not observed in the blocks with transient sensory stimulation (Fig. 6E). However, the strongest correlations occurred for phase ICMs (Fig. 6, D and F). In particular, phase ICMs in frequency ranges between 4 and 64 Hz (theta to low gamma) correlated with multisensory effects at frequencies higher than 30 Hz in the sustained blocks. In the transient blocks, we observed a relation between alpha and beta phase ICMs and gamma band response effects (Fig. 6F). Substantial negative correlations occurred in the transient stimulation blocks between prestimulus phase coupling higher than beta and multisensory response effects in the alpha and beta ranges (Fig. 6F). For response frequencies >140 Hz, the correlations with broad band prestimulus connectivity were significant for both types of coupling modes in the sustained stimulation blocks. In contrast, transient stimulation blocks showed a very different pattern of correlations (Fig. 6, E and F), in which a predictive relation for high-frequency (>140 Hz) components was mostly absent. Together, these results suggest a causal effect of prestimulus large-scale ICMs on multisensory response power effects, in which functional connectivity between early sensory areas before the stimulus presentation seems to play an important role.


This study has investigated the interplay between intrinsically generated large-scale coupling and stimulus-related activity using an audiovisual paradigm in the ferret. We hypothesized that repetitive sensory stimulation reconfigurates the ICMs and that this reconfiguration, at the same time, affects the upcoming sensory stimulus processing. Our approach was to present auditory, visual, and audiovisual stimuli to study whether this reconfiguration could act as facilitator during multisensory integration. We tested this hypothesis in an anesthetized preparation to have a better control on the sensory inputs and to reduce effects of other factors or brain states such as arousal, attention, and motor preparation. We used an ECoG array to record brain activity under two different stimulus blocks, and during periods without any sensory stimulation, to study whether this possible reconfiguration of ICMs might also occur spontaneously or is guided by the context.

The analysis of the topography of responses to visual and auditory stimuli showed a power increase in response to unimodal stimuli, which spread across several frequency bands, including alpha, which is reduced in response to visual stimuli in awake conditions (39). Multisensory stimulation also produced changes in stimulus-related power that showed both enhancement and mostly suppression across frequency bands. Enhancement was dominant in visual areas (beta and low gamma), presumably area 20, and at high frequencies (>64 Hz) in the secondary auditory areas ADF and PPF, which possibly correspond to the borders between unimodal subregions within AES area in the cat (31, 33). Note that, because of the inverse effectiveness of multisensory integration (40), this distribution of suppression/enhancement may have been favored by the stimuli applied here, which were presented at moderately high intensities to assure the sensory responses (see Materials and Methods). The topographic maps of suppression and enhancement appeared to be stimulus block specific (35). In our study, the responses to auditory and visual stimuli were not exclusive to their associated functional pathways, but it was frequent to observe visual responses in auditory areas and vice versa. This cross-modal effect has been widely reported across the cortex in several functional areas and species (30, 35, 41, 42). Last, because of the relatively short interstimulus intervals, sensory habituation might be expected. However, effects of recent history in the network response were not observed (fig. S1). The combination of different stimulation modalities and the randomized interstimulus intervals may contribute to a decreased efficiency of habituation mechanisms.

A recurrent result was that latencies at visually responsive cortical sites were shortened by combined audiovisual stimulation compared to the unimodal visual condition. This was observed, in particular, for higher-order sensory regions around the suprasylvian sulcus and for regions around the pseudosylvian sulcus. We interpret this as the consequence of a collaborative interaction between both modalities. This reduction in latencies might be responsible for the speeding up of perceptual and motor responses to visual targets (4347). In our study, the presence of the visual stimuli did not affect the latency of auditory responses, as has been observed in awake monkeys (48). One possible reason for this may be that our recordings were done in the anesthetized preparation. Another reason why shortening in auditory responses were not observed may be related to the fact that visual and auditory stimuli were presented in synchrony. If occurring asynchronously, visual stimulation can reset the phase of ongoing auditory cortical oscillations, and the degree of asynchrony can regulate the enhancement or suppression of multisensory responses (35, 49).

Oscillatory dynamics and frequency-specific coupling across multiple temporal scales are important characteristics of functional networks in the brain. Two of the mechanisms that underlie ICMs are phase coupling of band-limited oscillatory signals (phase ICMs) and the correlated fluctuations of signal envelopes (envelope ICMs) (1). Our results demonstrate that envelope ICMs and phase ICMs differ strongly in their spatial patterning. Whereas envelope ICMs showed a clustering of areas into functional systems, phase ICMs tended to be higher for connections between, rather than within functional clusters of cortical areas. For envelope ICMs, the spatial topography of coupling patterns seemed to be similar across frequency bands, but the strength of coupling varied across frequency ranges. The mean correlation values of envelope ICMs were higher in theta, alpha, and beta bands than in higher frequency bands. Therefore, we observed a decay of average connectivity as frequency increases, and at least in blocks with sustained stimulation, this was significantly lower at frequencies above 140 Hz. Although not in such a substantial manner, we observed the same result in the transient stimulation blocks. This agrees with the idea that envelope coupling captures mostly correlated fluctuations at lower frequencies, in a manner similar to what has been observed by correlation of fMRI blood oxygen level–dependent signals (50, 51). This dependence of coupling strength on frequency bands was much less prominent for phase ICMs. This fact, together with the strong difference in patterns across the two coupling modes, supports the idea that these types of ICMs may mediate different aspects of the dynamic communication across brain areas. Whether they mutually interact or are independent needs to be further investigated.

To our surprise, the stimulus-related connectivity patterns did not appear to depend on stimulus modality, and they did not differ substantially from the prestimulus interval for any of the modalities and coupling modes. The comparison of connectivity matrices between stimulation blocks during the prestimulus windows revealed prominent differences, suggesting that, rather than the current stimulus itself, it was the global context, created by the prolonged repetition of similar stimuli, that strongly shaped ICMs. This suggests that the long-term stimulation context can induce specific network states reflected in the dynamics of large-scale coupling patterns. The comparison of the spatial patterns for envelope and phase ICMs revealed a clear dissociation of the two types of coupling modes. This suggests that network connectivity states can emerge independently through the coordinated coactivation of sensory areas on slower time scales (as reflected in envelope ICMs) or through phase alignment of neural signals (as reflected in phase ICMs). This was observed only with respect to functional connectivity and not regarding the local spectral properties of ongoing activity, which did not show significant differences across stimulation contexts. Our findings suggest a previously undescribed form of brain state reconfiguration that, in contrast to previous studies (12, 27), affects connectivity without altering the energy of local signals. This reconfiguration seems to occur as a result of prolonged exposure to certain sets of sensory stimuli, leading to a reshaping of the predominant patterns of both envelope and phase ICMs.

Such a context-dependent reconfiguration of functional networks, in turn, might have an impact on the processing of individual sensory stimuli. Previous studies have shown the impact of prestimulus states on multisensory interactions (13, 49, 52). We tested this by correlating envelope and phase ICMs between visual and auditory regions in prestimulus epochs with the multisensory effects that we observed on latency and power of local responses. This analysis revealed strong effects of prestimulus functional connectivity on multisensory stimulus processing. The impact of prestimulus connectivity on multisensory responses differed between ICM types and stimulus contexts. Thus, for instance, response latency reductions were predicted mainly by envelope ICMs in the transient stimulation blocks but by phase ICMs in the blocks with sustained stimulation. Multisensory effects on response power were predicted mainly by phase ICMs. The ICMs predicting timing and power changes during multisensory stimulation predominantly involved connections between primary auditory and visual areas, supporting the notion that reconfiguration of connectivity as a mechanism for multisensory integration may already take place at early cortical processing stages (53).

Our study suggests that long-term modulation by repetitive sensory inputs can lead to a context-specific shaping of ICMs. These reconfigured coupling modes, in turn, lead to changes in latencies and power of local field potential responses that support multisensory integration. Our results demonstrate that this reciprocal interplay between network dynamics and external signals extends across multiple frequency bands and involves different types of intrinsic coupling. Although the present study was done in an anesthetized preparation, we suggest that this large-scale contextual adaptation may provide a mechanism that effectively facilitates responses of the network to sensory stimulation, as observed previously in behavioral studies (54). This might be particularly suitable for processing of complex stimuli, which require adaptation to the statistics of the environmental signals. In conclusion, the present study sheds light on the relationships between large-scale ICMs on multiple time scales and cortical evoked responses and provides evidence for the basis of a large-scale facilitating mechanism for multisensory integration of complex stimuli.


Data presented in this study were collected from five adult female ferrets (M. putorius). All experiments were approved by the Hamburg state authority for animal welfare (BUG-Hamburg) and were performed in accordance with the guidelines of the German Animal Protection Law.


Animals were initially anesthetized with an injection of ketamine (15 mg/kg) and medetomidine (0.08 mg/kg). A glass tube was then placed in the trachea to allow artificial ventilation of the animal and to supply isoflurane anesthesia (0.4% isoflurane, 50:50 N2O/O2 mix). To prevent dehydration throughout experiments, a cannula was inserted into the femoral vein to deliver a continuous infusion of 0.9% NaCl, 0.5 % NaHCO3, and pancuronium bromide (60 μg/kg per hour). Body temperature was maintained at 38°C with a heating blanket controlled by measurement of the animal’s rectal temperature. Heart rate and end-tidal CO2 concentration were constantly monitored throughout the duration of experiments to maintain the state of the animal. Before fixing the animal’s head in the sterotaxic frame, a local anesthetic (Lidocaine, 10%) was applied to the ear bars. To keep the animal’s head in a stable position throughout the placement of the ECoG and the measurements, and to allow auditory and visual stimulation, a headpost was fixed with screws and dental acrylic to the frontal bone of the head. The temporalis muscle was folded back, so that a large craniotomy could be performed over the left posterior cortex. After careful removal of the dura, the cortex was covered with saline solution. The ECoG array was gently placed on the surface of the cortex, such that electrodes covered large areas of visual and auditory cortex (fig. S1). A small piece of artificial dura (Lyoplant; B. Braun Melsungen AG, Melsungen, Germany) was cut in the shape of the craniotomy and gently placed over the ECoG array. Subsequently, the removed piece of skull was placed over the artificial dura, and then, the craniotomy was resealed using silicon elastomer (World Precision Instruments, Sarasota, FL). To prevent exsiccation of the cornea, a contact lens was placed over the right eye, while the left eye was occluded to ensure monocular stimulation. Experiments typically lasted between 24 and 36 hours, after which the animal was deeply anesthetized with 5% isoflurane and euthanized with an overdose of potassium chloride.


During electrophysiological recordings, isoflurane level was maintained at 0.4%. Neural activity across the entire posterior cortex was recorded using a custom-designed micro ECoG array (55). ECoG arrays were polyimide based and consisted of 64 hexagonally spaced (1.5-mm interelectrode distance) platinum electrodes with a diameter of 250 μm (fig. S1). Cortical surface LFPs were referenced to a clamp placed on the deflected temporalis muscle. Signals were digitized at 1395.1 Hz (0.1-Hz high-pass and 357-Hz low-pass filters) and sampled with an AlphaLab SnR recording system (Alpha Omega Engineering, Nazareth, Israel). All analyses of neural data presented in this study were performed offline after the completion of experiments. To obtain the location of ECoG electrodes for each animal, the position of the ECoG on cortex was recorded by taking photographs with a Zeiss OPMI pico microscope. The position of ECoG grids was then projected onto a scaled illustration of a model ferret brain that contained a map of the functional specialization of all posterior cortical areas (20 in total) (30). Data from each electrode were then allocated to the cortical area directly underlying the corresponding ECoG contact. To avoid potential confounding effects of fluctuations in brain state, we kept the anesthesia level as constant as possible and presented the stimulation blocks in a randomized manner.

Sensory stimulation

To ensure controlled conditions for sensory stimulation, all experiments were carried out in a dark sound-attenuated anechoic chamber (Acoustair, Moerkapelle, Netherlands). Auditory and visual stimuli were generated using the Psychophysics Toolbox (56) in MATLAB (MathWorks Inc., Natick, MA). Stimuli were digitalized at 96 kHz and delivered with an RME soundcard (RME HDSPe AIO Intelligent Audio Solutions) through a Beyerdynamic T1 speaker located 15 cm from the animal’s right ear. Before performing any experiments, the sound delivery system was calibrated using a Brüel and Kjær (Brüel and Kjær, Nærum, Denmark) free-field 4939 microphone coupled to a Brüel and Kjær 2670 preamplifier and 2690 amplifier. Visual stimuli were presented on an LCD display (Samsung SyncMaster 2233; frame rate, 100 Hz) placed 28 cm in front of the animal.

We presented two types of stimulation blocks that strongly differed in the duration and structure of the presented stimuli: (i) A transient stimulation block comprised visual stimuli consisting of 10-ms flashes of a 14° by 14° white square located at the center of the monitor displayed on a black background and auditory stimuli consisting of 0.5-ms sound clicks presented at 65-dB sound pressure level (SPL). We presented both unimodal and bimodal conditions (with zero delay) in a randomized intermixed fashion with interstimulus intervals varying between 900 and 1100 ms. In total, 100 flashes, 100 clicks, and 100 simultaneous clicks and flashes stimuli were presented per stimulation block. (ii) A sustained stimulation block comprised drifting Gabor patches with a size of 14° (full width at half maximum), a spatial frequency of 2 cycles/deg, and a temporal frequency of 5°/s, drifting in four directions (0°, 90°, 180°, and 270°). For the auditory stimulation, we presented dynamic moving ripples, an acoustic analog of the visual gratings (57). Moving ripples are complex noise-like stimuli that consist of the sum of several sinusoidal amplitude-modulated tones. By adjusting the amplitude and phase of the tones, they obtain a sinusoidal spectral envelope (58). We created the ripples using 60 tones logarithmically spaced along the frequency axis between 0.5 and 32 kHz. The spectral envelope of the composite sound was then modulated as a single sinusoid along the frequency axis on a linear amplitude scale (spectral peak density Ω = 0.1 and 1 cycle per octave and peak drift speed w = 0.5 and 0.2 Hz in both directions). Acoustic signals were multiplied by a 0.5-ms cos2 onset and offset function and scaled to a target root mean square amplitude to generate and present stimuli at 65-dB SPL. In total, 100 Gabor patches, 100 ripple stimuli, and 100 simultaneous patches and ripples were presented per stimulation block.

Stimuli of the sustained blocks had a duration of 1000 ms with interstimulus intervals varying between 900 and 1100 ms. Last, to enable comparison of results obtained from prestimulation intervals during the two types of stimulus blocks with longer epochs not perturbed by sensory stimulation, we also recorded blocks of 10- to 15-min duration of ongoing activity. We recorded three to five of these periods per animal at the beginning of recording session and between blocks of stimulation.

Data analysis

All data analysis was performed using custom scripts in MATLAB (MathWorks). Artificial ventilation and heart activity introduce, in some cases, strong artifacts. For that reason, frequencies below 4 Hz were not considered for current analysis. Sensory response latencies were defined as the poststimulus time, where LFPs exceeded four times the SD of prestimulus activity. The amplitude of responses was quantified as the largest LFP deflection (positive or negative) in the first 80 ms after stimulus onset. Spectral components of neural activity following sensory stimulation were estimated by computing fast Fourier transforms (FFTs) on LFPs. FFTs were computed from 100 to 600 ms after stimulus to ensure that transient onset responses did not confound spectral estimates. Reflecting the natural segregation of cortical rhythms into distinct frequency ranges, spectral estimates were averaged within the theta (4 to 8 Hz), alpha (8 to 16 Hz), beta (16 to 32 Hz), low gamma (32 to 64 Hz), high gamma (64 to 128 Hz), and high frequency band (>140 Hz). FFT-derived spectral estimates were then used for computing total power of local responses. To this end, we normalized the stimulus-related total power to the prestimulus power, which was calculated between −600 and −100 ms relative to stimulus onset. To quantify multisensory effects on neural responses, we computed the raw differences between latencies and also between power of responses to unimodal and bimodal stimulation. For graphical representation of this effect (Fig. 1, E and F), we used a variation index that corresponds to the raw difference multiplied by 2/2.

Spectral coherence is a measure of the consistency of phase coupling between simultaneously recorded signals and comprises both real and imaginary components. We chose to quantify phase ICMs by taking only the imaginary part of coherence. This approach eliminates the contribution of volume-conducted signals to functional connectivity measures by exclusively considering nonzero phase–lagged signal components (38).

For analysis of envelope ICMs, amplitude envelopes were computed in a time-resolved fashion by band-pass filtering and then by Hilbert-transforming LFP signals. To avoid the detection of spurious envelope correlation due to the effects of shared noise or volume conduction, pairs of ECoG signals were orthogonalized before the computation of the amplitude envelope time series (36). Similar to imaginary coherence, this step removes zero phase–lagged components that are shared between simultaneously recorded signals, therefore ensuring that only nonzero phase–lagged signal components are considered for analysis. Following orthogonalization, amplitude envelopes for the time window of 100 to 600 ms after stimulus onset for all trials were computed. For each stimulus condition, amplitudes were concatenated across all trials. Amplitude correlation was then defined as the linear correlation coefficient of pairwise envelope time series.

Functional connectivity measures were grouped into matrices where each pixel represented the mean functional connectivity for all pairs of electrodes between specialized cortical areas (see below for functional cortical area parcellation). The graph theoretical analysis of connectivity matrices for phase and envelope ICMs was performed using the Brain Connectivity Toolbox (37). To quantify large-scale cortico-cortical connectivity, we computed average degree, betweenness centrality, and clustering coefficients of weighted functional connectivity matrices. The SD of graph theoretical measures was estimated by repeating analysis on randomly shuffled functional connectivity matrices (100 repetitions). Graph theoretical measures were then normalized by the mean and SD of randomly shuffled matrices.

We calculated the imaginary coherence and amplitude envelope correlation first on an electrode-to-electrode basis, resulting in a 64 × 64 matrix with all possible connections. We assigned to each electrode the area over it was located and then regrouped and averaged connectivity data in terms of pairs of areas, yielding a connectivity matrix in terms of a functional area representation (Fig. 1A). Note that, in this new representation, the number of electrodes per area varied across areas (from two to six electrodes). We took averages across animals only for areas that were recorded in all animals. Note that areas V1, AVF, and VP, displayed in Fig. 1, are not in the mean matrix because they were not recorded in all animals.

Last, we evaluated the potential artifactual connectivity introduced by the underlying volume conductivity of the tissue. To this end, we calculated both connectivity measures in two preprocessing conditions: first, using the filtered signals without re-referencing, and second, using local bipolar derivatives (re-referencing to closest neighboring electrode). Our assumption was that the volume conductivity increases connectivity, and, as a consequence, the preprocessing condition less affected by volume conduction shows lower values in both connectivity measures. We measured the mean connectivity between all first, second, and third neighboring pairs (fig. S8). On the basis of this analysis, in the current study, we used unreferenced signals.


Supplementary material for this article is available at

Fig. S1. Recording approach in the anesthetized ferret.

Fig. S2. Spectral properties of ongoing and stimulus-related activity.

Fig. S3. Fraction of responsive sites and multisensory responses.

Fig. S4. Topography of response power changes for sustained and transient stimuli.

Fig. S5. Clustering coefficient for stimulus-related connectivity.

Fig. S6. Contrast index of connectivity matrices during prestimulus conditions.

Fig. S7. Relation between prestimulus functional connectivity and multisensory effects.

Fig. S8. Grand mean of connectivity measures as a function of frequency.

This is an open-access article distributed under the terms of the Creative Commons Attribution-NonCommercial license, which permits use, distribution, and reproduction in any medium, so long as the resultant use is not for commercial advantage and provided the original work is properly cited.


Acknowledgments: We would like to thank D. Bystron for assistance with data acquisition, E. Fiedler for assembling the thin-film arrays with the connectors, and C. Moll and A. Gulberti for discussions and critical comments. Funding: This research was supported by funding from the German Research Foundation (GRK 1247/2, SFB 936/A2, SPP 1665/EN533/13-1, and SPP 2041/EN533/15-1 to A.K.E.) and the European Union (ERC-2010-AdG-269716 to A.K.E.). Author contributions: E.E.G.-L., I.S., F.P., G.E., and A.K.E. conceived and designed experiments. E.E.G.-L., I.S., and F.P. performed experiments. E.E.G.-L. analyzed data. G.E. wrote the animal ethics permission. T.S. designed and developed ECoG arrays. E.E.G.-L. and A.K.E. wrote the paper. All authors edited and approved the manuscript. Competing interests: The authors declare that they have no competing interests. Data and materials availability: All data needed to evaluate the conclusions in the paper are present in the paper and/or the Supplementary Materials. Additional data related to this paper may be requested from the authors.

Stay Connected to Science Advances

Navigate This Article