Unique spatiotemporal fMRI dynamics in the awake mouse brain.

Human imaging studies have shown that spontaneous brain activity exhibits stereotypic spatiotemporal reorganization in awake, conscious conditions with respect to minimally conscious states. However, whether and how this phenomenon can be generalized to lower mammalian species remains unclear. Leveraging a robust protocol for resting-state fMRI (rsfMRI) mapping in non-anesthetized, head-fixed mice, we investigated functional network topography and dynamic structure of spontaneous brain activity in wakeful animals. We found that rsfMRI networks in the awake state, while anatomically comparable to those observed under anesthesia, are topologically configured to maximize interregional communication, departing from the underlying community structure of the mouse axonal connectome. We further report that rsfMRI activity in wakeful animals exhibits unique spatiotemporal dynamics characterized by a state-dependent, dominant occurrence of coactivation patterns encompassing a prominent participation of arousal-related forebrain nuclei and functional anti-coordination between visual-auditory and polymodal cortical areas. We finally show that rsfMRI dynamics in awake mice exhibits a stereotypical temporal structure, in which state-dominant coactivation patterns are configured as network attractors. These findings suggest that spontaneous brain activity in awake mice is critically shaped by state-specific involvement of basal forebrain arousal systems and document that its dynamic structure recapitulates distinctive, evolutionarily relevant principles that are predictive of conscious states in higher mammalian species.

Introduction

Recent years have seen an increased interest in the application of resting-state fMRI (rsfMRI) in physiologically accessible species., The use of these methods in the mouse has highlighted encouraging cross-species correspondences in the organization of functional networks,, offering novel opportunities to mechanistically probe the neural basis of brain-wide fMRI coupling and its breakdown in brain disorders.4, 5, 6

To ensure immobilization of animals during image acquisition and prevent motion-related artifacts, the vast majority of mouse rsfMRI studies to date have been carried out using light anesthesia., While the employed protocols have been shown to preserve the functional, and dynamic architecture of rsfMRI networks in this species, anesthetic agents can alter hemodynamic and neurovascular coupling,, or generate unwanted genetic or pharmacological interactions that can confound the mechanistic interpretation of rsfMRI signals. More importantly, the lack of well-characterized rsfMRI datasets in awake mice prevents a full understanding of how anesthesia-induced loss of consciousness affects the functional architecture of rsfMRI networks with respect to awake conditions in this species. This area of research is of especial significance, given the increasing interest in the study of how the dynamic structure of rsfMRI activity evolves and reconfigures as a function of brain state.

Influential investigations in wakeful and anesthetized primates and humans have uncovered hallmark dynamic features that are strongly biased toward high and low levels of consciousness, respectively. Specifically, loss of consciousness has been associated with a partial reorganization of long-range functional connectivity, disappearance of anticorrelated cortical states, and a repertoire of dynamic states dominated by rigid functional configurations tied to the underlying anatomical connectivity., By contrast, conscious wakefulness has been associated with greater global integration and interareal cross-talk, anticorrelations between the activity of different brain regions,, and a more flexible repertoire of functional brain configurations departing from anatomical constraints., Interestingly, some of these network features have been proposed to constitute a putative “signature of consciousness” that exhibits significant evolutionary conservation in primates and humans., However, it remains unclear whether and how analogous dynamic rules govern the spatiotemporal organization of spontaneous fMRI activity in the awake mouse.

Leveraging a robust protocol for rsfMRI in awake head-fixed mice, here, we provide a fine-grained description of the functional topography and dynamic structure of rsfMRI networks in wakeful animals. By comparing awake network features with those obtained in anesthetized states, we find that rsfMRI networks in wakeful mice are topologically configured to maximize integration between neural communities and exhibit topographically unique coactivation patterns encompassing a dominant involvement of arousal-related nuclei, anti-coordination between visual-auditory and default-mode network (DMN) areas, and idiosyncratic temporal transitions toward distinctive network attractor states. These results reveal that rsfMRI activity in awake mice is critically shaped by state-specific involvement of basal forebrain arousal systems and exhibits a stereotypic spatiotemporal structure that reconstitutes foundational principles of conscious network states in higher mammalian species.

Results

Implementation of resting-state fMRI in awake head-fixed mice

Prompted by the emerging use of head fixation to reduce motion-related artifacts in rodent neuroscience studies, we devised a simple protocol for awake mouse rsfMRI via the surgical implantation of plastic headposts, followed by gentle and progressive restraint onto a custom-made cradle (Figure 1). After recovery from surgery, we subjected mice to progressive operator handling and mock scanning sessions of increasing length with the aim to acclimate them to the procedure and minimize restraint-induced stress (Figure 1). Mice undergoing the habituation procedure did not experience substantial weight loss (Figures S1A–S1C) and exhibited a growth rate only marginally lower than that observed in co-housed control littermates not subjected to any experimental procedure (two-way ANOVA, group p = 0.39, session × group p = 0.15). We also measured plasma corticosterone levels in surgically implanted animals during the initial acclimation period (handling stage) and immediately after the first MRI scan at the end of habituation period. We next compared the resulting levels with those obtained in co-housed littermates under baseline conditions (no experimental procedure), or after a routine behavioral assessment (open field test [OFT]; Figures S1D and S1E). We found statistically higher levels of corticosterone in the fMRI awake group under both conditions (baseline, and rsfMRI or OFT, two-way ANOVA, group p = 0.02, session p < 0.001); however, no group × session interaction (p = 0.90), suggesting that the acclimation and scanning procedure is overall only marginally more stressful than routine behavioral testing. Moreover, a comparison of measured corticosterone levels with those previously reported by others under different conditions or manipulations (Figure S1E) revealed that the corticosterone levels associated with awake rsfMRI scanning were ca. 2- to 5-fold lower than those elicited by unhabituated acute immobilization, 2-fold lower than anesthesia-stimulated corticosterone release, and broadly comparable to the amounts elicited by natural circadian excursion, or induced by similar habituation protocols employed for awake MRI imaging in the mouse (Figure S1E). Importantly, mice who underwent an OFT after fMRI scanning did not reveal any stress-related phenotype when compared with co-housed littermates not subjected to any restraint or habituation procedures (Figure S1F; p > 0.31, all metrics). The results of these investigations suggest that the acclimation procedure is well tolerated and that stress response produced by our habituation and awake fMRI scanning procedure is marginal.

Figure 1

Acclimation protocol for awake rsfMRI

(A) Three-dimensional render of the custom mouse cradle used for rsfMRI acquisitions (Ugo Basile S.r.L).

(B) Plastic headpost used for head fixing.

(C) Restraint apparatus upon and coil positioning.

(D) Experimental timeline for habituation protocol. D1 habituation started 10 days after headpost surgery.

See also Figure S1.

rsfMRI networks in awake mice exhibit focal functional reconfiguration

Using this protocol, we acquired 32-min long rsfMRI timeseries in n = 10 awake C57Bl6/J male mice. To map the functional organization of rsfMRI networks in awake conditions, we systematically probed rsfMRI connectivity networks via seed-based correlation mapping. This analysis revealed robust interhemispheric and antero-posterior rsfMRI synchronization (i.e., “functional connectivity”), including the presence of distributed networks anatomically recapitulating rsfMRI systems previously described in lightly anesthetized mice.,, The identified systems include a DMN, a salience (insular-cingulate) network, a sensory-motor latero-cortical network (LCN), a visual-auditory latero-posterior network (LPN), plus a number of subcortical sub-systems, including dorsal (caudate putamen) and ventrostriatal networks (mesolimbic pathway), a dorsal hippocampal network, and a widely distributed olfactory and amygdaloid network (Figure 2).

Figure 2

rsfMRI connectivity networks in the awake mouse brain

Each panel represents the averaged seed-based correlation map across n = 10 awake subjects, thresholded to voxels with significant connectivity (two-tailed t test, p < 0.01, FWER cluster corrected with defining threshold T = 2.8). Seed location is indicated by red lettering (DMN, default-mode network; PFC, prefrontal cortex; Cg, cingulate cortex; Rs, retrosplenial cortex; TH, thalamus; CPu, caudate putamen; Ins, insula; dHC, dorsal hippocampus; Ent, entorhinal cortex; Au, auditory cortex; M1, motor cortex; SS, somatosensory cortex; V1, visual; BF, basal forebrain; Amy, amygdala; NAc, nucleus accumbens; HT, hypothalamus).

See also Figure S6.

While the functional organization of rsfMRI networks in awake mice appears to broadly reconstitute key architectural features previously mapped in anesthetized mice, focal or more subtle state-dependent differences in the topography and organization of rsfMRI network activity may uncover dynamic features and functional substrates representative (and predictive of) wakeful, conscious states in lower mammalian species. To investigate this aspect, we spatially compared rsfMRI network topography in our awake scans with those previously acquired in a separate cohort of age-matched anesthetized C57Bl6/J mice (n = 19). Interestingly, this comparison revealed a set of focal state-dependent differences in the extension and anatomical organization of rsfMRI connectivity networks (Figures 3A and 3B). First, we found that rsfMRI networks in awake animals exhibited robust functional anti-coordination between some of the probed regions and their long-range substrates. Anti-coordination was especially prominent between medial prefrontal cortex (PFC) and olfactory regions, as well as between visual-auditory areas and midline regions of the DMN. The observed anticorrelation was accompanied by a reduced spatial extension of the DMN in awake mice, where a clear segregation of midline corticolimbic and visuo-auditory postero-lateral portions of the DMN was apparent. Moreover, in awake mice ventral forebrain area (e.g., diagonal band, hypothalamus, and nucleus accumbens) were part of an extended highly synchronous network that exhibited only marginal functional coupling in anesthetized subjects. More nuanced network-specific differences in rsfMRI connectivity strength were also apparent, with evidence of reduced cortico-cortical coupling in the DMN and LCN, and stronger connectivity in visual-auditory and basal forebrain areas in awake versus anesthetized animals.

Figure 3

rsfMRI network topography in the awake and anesthetized mouse brain

(A) Group-averaged seed-based correlation maps in awake (n = 10, left) and halothane anesthetized (n = 19, right) mice, thresholded to voxels with significant connectivity (one sample t test, p < 0.01, cluster corrected with defining threshold T = 2.8).

(B) Between-group connectivity differences (two-sample t test, p < 0.01, FWER corrected with defining threshold T = 2.8).

(C) Group-averaged voxel-wise functional connectivity (FC) matrices for each condition, with voxels organized within axonal connectivity modules, and further partitions into network sub-modules. The sub-module corresponding to thalamic voxels within the LCN is not labeled (light-red block).

(D) Significant between condition FC differences at the voxel level (two-tailed two-sample t test, p < 0.05, FDR corrected) (DMN, default-mode network; LCN, latero-cortical network; HC, hippocampal network; OF-BF, olfactory-basal forebrain network; PFC, prefrontal cortex; Cg, cingulate cortex; Rs, retrosplenial cortex; TH, thalamus; CPu, caudate putamen; Ins, insula; dHC, dorsal hippocampus; Ent, entorhinal cortex; Au, auditory cortex; M1, primary motor; SS, somatosensory cortex; V1, visual; BF, basal forebrain; Amy, amygdala; NAc, nucleus accumbens; HT, hypothalamus).

See also Figures S2 and S3.

These results were corroborated by whole-brain voxel-wise mapping of interareal connectivity via correlation matrices (Figures 3C and 3D). To help anatomical interpretability of this analysis, we organized voxels into axonal connectivity modules recently identified in the mouse brain (Figure S2A). The resulting fine-grained partitioning (Figure 3D) confirmed that in awake animals (1) the DMN reconfigures into two segregated submodular division (midline and sensory-PLN); (2) connectivity within basal forebrain systems, where most ascending neuromodulatory systems are located, is dramatically increased; and (3) rsfMRI network activity exhibits higher functional anti-coordination (Figure S2B), uncovering relations between networks that are not present in anesthetized conditions, such as the inverse coupling between midline and PLN components of the DMN (Figures 3C and 3D). Corroborating these findings, an extension of our comparisons to a third cohort of mice (n = 14) anesthetized with a different anesthetic regimen (isoflurane plus medetomidine), e.g., isoflurane plus medetomidine, replicated these topographical features (Figures S2E–S2G), suggesting that they are not the result of a specific pharmacological mechanism but can more broadly reflect state-dependent reconfiguration following anesthesia-induced loss of responsiveness. Taken together, these investigations document that wakeful states in the mouse lead to a focal rsfMRI network reconfiguration dominated by increased basal forebrain-cortical coupling and anti-coordination between sensory and latero-posterior DMN associative cortices.

rsfMRI connectivity in awake mice shows increased between-network communication

The observation of the areas of negative correlation in the awake mouse brain is of great interest, as similar findings have been suggested to serve as a putative signature of fMRI network activity in conscious states in other mammalian species such as marmosets, macaques, and human., Importantly, these reports also showed that network configuration in anesthetized and wakeful animals may be characterized by different anatomical organization, with the unconscious state being more tied to the anatomical map and awake brain networks exhibiting a topographical departure from their underlying anatomical architecture. To investigate whether similar principles would apply to the mouse brain, we used a graph theoretical approach to probe the relationship between structural and functional connectome in wakeful and anesthetized animals. To this aim, we leveraged a recent anatomical partition of the voxel-wise mouse axonal connectome into four macro-communities that spatially reconstitute macroscopic network systems of the mouse brain, i.e., the DMN, the LCN, the hippocampus, and olfactory-basal forebrain (OF-BF) areas. A graphic representation of the functional connectome with respect to these axonal communities (Figure 4A) revealed that, departing from the modular partitioning of the axonal connectome, fMRI networks in awake subjects exhibit greater interareal communication than corresponding anesthetized state. In keeping with this notion, structure-function correspondence was significantly lower in awake animals compared with anesthetized subjects (p < 0.01, Mann-Whitney test; Figure 4B). Formal quantifications of rsfMRI network connectivity strength corroborated these qualitative observations (Figure 4C), revealing dramatically increased between-network connectivity in awake mice, a finding that was especially prominent between BF and cortico-hippocampal areas (p < 0.01, Mann-Whitney test, FDR corrected). Importantly, analogous features were observed when we contrasted awake rsfMRI data with those obtained in isoflurane-medetomidine anesthetized animals (Figures S2H–S2J), hence supporting a possible generalization of this finding to other anesthetic regimens. These findings recapitulate prior observations in conscious primates, suggesting that, departing from the underlying structure of the axonal connectome, in the awake mouse brain rsfMRI network activity topologically reconfigures to maximize cross-talk between cortical and subcortical neural systems.

Figure 4

Structure-function relationship in awake and anesthetized states

(A) Graphic representation of rsfMRI connectivity within and between previously described axonal modules of the mouse brain (DMN, LCN, HC, and OF-BF; Coletta et al.). Each cluster of nodes represents a subset of anatomically defined regions of interest within the corresponding module. Nodes have been empirically arranged to maximize figure legibility.

(B) Structure-function correspondence in awake (n = 10) and anesthetized mice (n = 19). Between-group differences were assessed with a Mann-Whitney test (p < 0.05).

(C) Quantification of within (diagonal) and between (off-diagonal) network functional connectivity (∗p < 0.05, ∗∗p < 0.01, ∗∗∗p < 0.001, ∗∗∗∗p < 0.0001, Mann-Whitney test, FDR corrected) (DMN, default-mode network; HC, hippocampus; OF-BF, olfactory and basal forebrain; LCN, latero-cortical network).

See also Figure S2.

rsfMRI dynamics in awake mice exhibits unique coactivation topography

Our investigations revealed focal functional reconfiguration in “static” (i.e., time-averaged) rsfMRI networks in awake conditions. Prompted by the identification of dynamic connectivity signatures of consciousness in primates and human,14, 15, 16 we hypothesized that the observed static network changes could similarly reflect state-dependent differences in the underlying dynamic structure of rsfMRI. To this aim, we decomposed rsfMRI activity into six dominant recurring coactivation patterns (CAPs),, as these have been recently shown to govern rsfMRI dynamics in the mammalian brain, and their dynamics has been found to be predictive of conscious states in humans. Further analyses showed that k = 6 provides a partition with high variance explained in concatenated as well as individual datasets, and that, with k > 6, higher-order clustering did not offer a gain in variance explained higher than 1% across all datasets (Figures S4A and S4B).

In keeping with previous investigations, all the highlighted CAP topographies encompassed recognizable network systems of the mouse brain (Figure 5A), and opposing spatial configurations, delineating three dominant CAP- anti-CAP pairs (CAP 1–2, 3–4, and 5–6; Figure 5C), capturing peaks and throughs of fluctuating BOLD activity. Interestingly, while the overall anatomical organization of CAPs appeared to be comparable in awake and anesthetized conditions, distinctive, state-dependent topographical differences in the anatomical organization of the corresponding BOLD coactivation patterns were apparent (Figure 5B). An especially notable finding was the observation of robust coactivation of arousal-related BF nuclei in CAPs 5–6 of awake animals. This spatial signature was associated with a weaker but extended coactivation of thalamic and latero-posterior visuo-auditory regions, which exhibited anti-coordinated peaks of BOLD activity in DMN and visual areas in awake state. In humans, analogous CAPs have been found to encompass similar network engagements (SMN+ CAP in Huang et al.). CAPs 3 and 4 in awake mice similarly showed enhanced fluctuations in BF areas, together with the diffuse involvement of striatal and thalamic substrates, reminiscent of patterns of widespread pan-cortical coactivation recently described in human CAPs (GN+ CAP in Huang et al.). Finally, CAPs 1 and 2 in awake mice were characterized by focally dampened peaks of BOLD activity in midline areas of the DMN and in the LCN, recapitulating topographies that are also observed in human (DMN+ CAPs in Huang et al.). These topographic differences show that rsfMRI dynamics in the awake state is characterized by a distinctive coactivation of arousal-related subcortical nuclei and additional network components (e.g., PLN and midline DMN) that we found to be differentially configured in the static connectome of awake and anesthetized animals (cf. Figure 2). This finding corroborates a tight link between CAP dynamics and the ensuing time-averaged network activity, suggesting that the functional architecture of the “static” rsfMRI connectome is critically shaped by its underlying coactivation structure.

Figure 5

Coactivation pattern (CAP) topography and occurrence in the awake and anesthetized mouse brain

(A) CAP topography in awake (n = 10, left) and anesthetized mice (halothane, n = 19, middle; t test, p < 0.01, FWER cluster corrected with defining threshold T = 2.8).

(B) Corresponding awake > anesthesia difference maps (t test, p < 0.01, FWER cluster corrected with defining threshold T = 2.8).

(C) Between CAP spatial similarity (Pearson’s correlation) in awake and anesthetized mice. Note the presence of clear CAP anti-CAP pairs.

(D) Quantification of CAP occurrence rates in awake and anesthetized mice (∗∗∗p < 0.001, ∗∗∗∗p < 0.0001, t test, FDR corrected) (PFC, prefrontal cortex; Cg, cingulate cortex; Rs, retrosplenial cortex; TH, thalamus; CPu, caudate putamen; Ins, insula; HC, hippocampus; Au, auditory cortex; M1, primary motor; SS, somatosensory cortex; V1, visual; BF, basal forebrain; Amy, amygdala; NAc, nucleus accumbens; HT, hypothalamus).

See also Figures S3 and S4.

Further supporting this notion, it has been recently shown that most variance in static rsfMRI connectivity is explained by a small fraction of fMRI frames exhibiting the highest cofluctuation amplitude. This observation suggests that CAPs might similarly represent peaks (and troughs) of BOLD activity that critically shape the structure of the static rsfMRI connectome. To test this hypothesis, we generated CAP cofluctuation matrices for both awake and anesthetized conditions (Figure S3) and compared their mean topography with that of the corresponding time-averaged, “static” functional connectivity. This comparison yielded a spatial correlation r = 0.79 (R2 = 0.62) and r = 0.78 (R2 = 0.61) for awake and anesthetized rsfMRI time series, respectively. Similarly high spatial correlation was found when the approach was applied to medetomidine-isoflurane-anesthetized animals (r = 0.69, R2 = 0.49). These results expand recent investigations by showing that the peaks of BOLD activity captured by CAPs account for a dominant fraction of variance in time-averaged static rsfMRI connectivity. They also suggest that CAPs, and their unique state-dependent functional configuration, crucially shape the topography and dynamics of rsfMRI networks mapped in awake and anesthesia-induced loss of responsiveness.

Prevalence of coactivation patterns distinguishes awake and anesthetized states

Human and primate research has shown that conscious states are associated with rsfMRI connectivity signatures characterized by the dominant occurrence of stereotypical functional configurations.14, 15, 16 These findings led us to investigate whether analogous changes in CAP dynamics could underlie the network reconfiguration observed in wakeful mice. To this aim, we first probed CAP oscillatory dynamics by quantifying the fractional amplitude of low-frequency fluctuations (fALFFs; Figure 6C) in the spectral power of each coactivation pattern’s (CAP’s) timeseries (Figures 6A and 6B). This analysis revealed that, although both wakeful and anesthetized states exhibit broadly comparable infraslow [0.01–0.03 Hz] dynamics, the power of these slow fluctuations was significantly higher under anesthesia (p < 0.05, Mann-Whitney test, all CAPs except CAP 4). We next computed the distribution of GS phases at the occurrence of each CAP within infraslow (0.01–0.03 Hz) fMRI GS cycles (30–100 s long), as previous investigations revealed that in anesthetized conditions CAP occurrence is phase locked to GS cycles. Interestingly, we found that CAP occurrence in awake states is also locked to GS infraslow cycles (Rayleigh test, p < 0.001, FDR corrected; Figure 6D). However, CAPs 3 and 4 did not exhibit opposite phase occurrence as seen in anesthetized animals but showed instead concordant preferred occurrence in antiphase with that of GS cycles. Moreover, phase occurrence of awake CAPs 5 and 6 was in antiphase with that of the corresponding CAPs in anesthetized animals. These results document that, while CAPs in both anesthetized and awake conditions occur at specific phases of GS fluctuations, significant state-dependent changes in the preferred phase occurrence of these coactivation patterns exist.

Figure 6

Infraslow dynamics and preferred occurrence of coactivation patterns (CAPs) in awake and anesthetized mice

(A) Group-averaged power spectral density of awake (red) and halothane anesthesia (blue) CAP time courses (mean ± SEM).

(B) Power spectral density of the global fMRI signal (GS, mean ± SEM).

(C) Quantification of the fractional amplitude of low-frequency fluctuations (fALFF) in CAPs and GS spectra, computed as the ratio between 0.01–0.03 Hz band-limited power and the full dynamic range band (0.01–0.1 Hz) for awake (red) and anesthetized mice (blue) (∗p < 0.05; ∗∗p < 0.01; ∗∗∗p < 0.001, two-sample t test, FDR corrected).

(D) Distribution of GS phases at the occurrence of each CAP showing significant deviations from circular uniformity in awake and anesthetized states (Rayleigh test, p < 0.05 Bonferroni corrected).

See also Figure S4.

Lastly, to assess whether any of the observed CAPs could be predominantly associated with a specific brain state, we compared CAP occurrence in awake mice and anesthetized animals (Figure 5D). This analysis revealed dramatic state-dependent differences in CAP prevalence. Specifically, occurrence of CAPs 1 and 2 (and to a much lower extent, CAPs 3 and 4) was significantly higher (ca. 2- to 5-fold) in anesthesia than awake states. Conversely, CAPs 5 and 6 were prominently more frequent (ca. 3-fold) in the awake state (p < 0.001, two-tailed, two-sample t test, FDR corrected). These result show that in the mouse prevalence of CAPs is critically different in awake and anesthetized states. Importantly, very similar CAP occurrences were observed in animals anesthetized with isoflurane-medetomidine (Figure S4G), supporting a possible generalization of this observation across anesthetic conditions. To further corroborate these findings, we assessed the effect of the framewise censoring used in our preprocessing step and performed our CAPs analyses without censoring high-motion fMRI frames. We found that no CAP was strictly associated to high-motion events (Figures S4D and S4E), and that their structure was virtually indistinguishable from CAPs obtained after censoring (Figure S4F). Moreover, framewise scrubbing did not affect CAP occurrence rate findings (Figures S4G and S4H). These results suggest that the observed changes in large-scale rsfMRI dynamics are not driven by motion-affected frames. Interestingly, it should also be noted that the CAPs that are prevalent in the awake state (CAPs 5 and 6) exhibit distinctive anti-coordinated engagement of visual-auditory regions (PLN) and DMN areas (Figures 5A and S3A). This finding recapitulates similar patterns of regional anti-coordination in higher mammalian species,, pointing at a putative evolutionarily conserved network signature predictive of wakeful, conscious states in the mammalian brain.

rsfMRI networks in awake mice exhibit unique temporal structure

Prompted by recent evidence of stereotypic CAP transition trajectories in conscious humans, we next investigated whether similarly different state-dependent temporal trajectories could be identified in awake and anesthetized mice. To unravel sequential transitions within the awake and anesthetized states, we modeled CAP time series as a Markov process including both the transition probability between CAPs as well as self-transitions (e.g., persistence probability). Consistent with the slow dynamics of CAPs, we found that all persistence probabilities in both awake and anesthetized conditions were highly significant when compared with a null-hypothesis distribution with randomly permuted CAP sequences (p < 0.001, all CAPs). In agreement with CAP occurrence results, between-state comparisons next showed that CAPs 1–2 and 3–4 exhibit higher persistence in the anesthetized state, while CAPs 5–6 show increased persistence in awake conditions (Figures 7A–7C, top row).

Figure 7

CAP transition probabilities and corresponding temporal trajectories

(A and B) Persistence (top row) and transition (off-diagonal) probability in awake (A, n = 10) and anesthetized (B, n = 19) mice. Significant preferred transitions are denoted with a red “+” sign. Directional transitions are depicted if the incoming transition is significant.

(C) Transition probability differences matrix. Statistically significant between-group differences are denoted with a “+” sign for transitions higher in awake, and conversely with a “−” sign, if higher in anesthesia.

(D and E) Graph representation of significant persistence (round self-connecting arrows) and transition (black arrow heads) probability for awake and anesthesia conditions, respectively. Significant transitions (p < 0.001, 1,000 surrogate time series) are depicted with an arrowhead, while directionally dominant transitions are denoted with a red circle.

(F) Significant persistence and transition probability differences between groups. Red arrows indicate higher transition probabilities in awake mice, and blue arrows in anesthetized mice.

(G and H) Entropy of Markov trajectories (HMT) in awake (G) and anesthetized mice (H). Higher entropy of a trajectory indicates lower accessibility of a CAP destination from a starting CAP.

(I) Between-group differences in HMT. Positive elements represent CAP trajectories with lower accessibility in the awake state as compared with anesthesia, while negative elements are trajectories with facilitated access in awake. Only paths from CAPs 2 and 6 to CAP 3 did not reach statistical significance at p < 0.001 (1,000 surrogate timeseries).

See also Figure S5.

To portray the temporal trajectory of CAP transitions in awake and anesthetized mice, we next computed, and graphically depicted, the corresponding transition probability matrices (Figures 7A–7F). These investigations revealed stereotypic CAP trajectories in awake and anesthetized conditions. Specifically, we found that in awake mice CAPs 5 and 6, besides being the most recurrent (Figure 5D) and persistent CAPs, were also configured as dominant attractors state, i.e., they are the most probable destination from all other CAPs. Conversely, temporal transitions under anesthesia where more distributed, with both pan-cortical CAPs 3–4 and DMN-LCN CAPs 1–2 (i.e., the two CAP pairs with the highest occurrence in this brain state), emerging as the most prominent CAP attractors (Figure 5D). Importantly, very similar temporal signatures were observed in mice anesthetized with medetomidine-isoflurane (Figures S5A–S5C), corroborating the robustness of this finding. Moreover, our results were confirmed after performing 500 split-half resampling of the sequences (r > 0.96; Figure S5G) and after comparing the group concatenated probabilities with each single mouse (r > 0.85 all mice, with one exception r = 0.69; Figure S5H). These additional analyses strongly corroborate the robustness and state-specific nature of the identified temporal transitions signatures identified.

To investigate whether the observed temporal transitions could be associated with state-dependent differences in the complexity of trajectories between each pair of CAPs, we computed the entropy of Markov trajectories (HMT) for each transition probability matrix. This metric provides an estimate of the “accessibility” of each CAP from another one: low descriptive complexity (i.e., entropy close to 0) from a starting point (initial CAP i) to its destination (final CAP j) indicates an almost deterministic direct path from CAP i to CAP j (high accessibility). In contrast, high entropy values (i.e., close to 1 bit) imply a higher uncertainty, and lower accessibility, as the trajectory encompasses random steps through different CAPs to reach its destination. Consistent with our CAP occurrence results and the temporal structures described above, our investigations revealed that CAPs 5–6 are the two temporal states with the highest accessibility in awake animals (Figures 7G and 7H) and conversely those exhibiting the lowest accessibility in anesthetized states. CAPs 1–2 and 3–4 required instead more elaborate transition trajectories in awake conditions but had higher accessibility in anesthetized animals (Figures 7G and 7H). Supporting the generalizability of our findings to other anesthetic regimens, a remarkably similar CAP transition signature was observed in isoflurane-medetomidine anesthetized mice (Figures S5D–S5F). Moreover, our results were robust after repeating the calculations both with 500 split-half resampling iterations of each dataset (Figure S5I) and at the single-mouse level (Figure S5J). Taken together, these results indicate that rsfMRI activity in the awake mouse exhibits stereotypic temporal trajectories and increased accessibility to a state-dominant network configuration (CAPs 5 and 6) characterized by a critical engagement of arousal-related BF areas, and anti-coordination between DMN and visual regions.

Discussion

Leveraging a robust protocol for rsfMRI mapping in awake mice, we investigated the functional architecture and dynamic organization of rsfMRI activity in wakeful and lightly anesthetized mice, with the aim of determining how the ensuing network architecture reconfigures as a function of the underlying brain state. Our investigations revealed that rsfMRI activity in awake mice undergoes focal topographic reconfiguration entailing the presence of network-specific regional anticorrelation, heightened connectivity in arousal-related BF areas and greatly increased between-network cross-talk, departing from the rigid structure of the axonal connectome. Notably, these changes were associated with remarkably distinct dynamic patterns of signal coordination as assessed with a deconstruction of rsfMRI activity into recurring coactivation patterns. Specifically, we found that rsfMRI activity in wakeful animals exhibits a largely dominant occurrence of CAPs encompassing arousal-related forebrain nuclei, as well as features recently described to be predictive of consciousness in higher mammalian species, such as transient anticorrelation between visual-auditory and DMN areas,, and a stereotypic temporal transitions driven by a set of more accessible and persistent temporal states. These results suggest that rsfMRI activity in awake mouse is critically shaped by a state-specific involvement of BF arousal systems and that its dynamic structure recapitulates evolutionarily relevant principles described in higher mammalian species.

Expanding initial attempts to map spontaneous fMRI activity in awake mice,, our work describes a protocol enabling the reliable implementation of awake rsfMRI imaging in this species and provides a fine-grained, comprehensive description of rsfMRI network organization and dynamics in wakeful mice. Our investigations show that the static, time-averaged architecture of rsfMRI networks in awake mice reconstitutes organizational principles previously observed in anesthetized conditions, including the presence of distributed systems such as a DMN, a LCN, and a salience-like network. Analogous broad topographic correspondences between anesthetized and awake conditions have been reported in other mammalian species, including rats,, primates,26, 27, 28 and humans,, underscoring a tight relationship between the general spatial structure of spontaneous fMRI activity and its underlying structural map.,

The focal topographic reconfiguration observed in awake state is of interest as it highlights functional substrates that are critically sensitive to the effect of general anesthesia and that as such may putatively govern a transition between stimulus-unresponsive to wakeful, conscious states in the rodent brain. In this respect, our observations of heightened functional connectivity in basal forebrain-hypothalamic areas and their increased cross-talk with cortical modules are fully consistent with the established role of these regions as key mediators of arousal and vigilance in the mammalian brain.32, 33, 34 As conscious perception relies upon the ability to integrate information across specialized communities of brain regions, our finding that rsfMRI network structure in awake states shows increased interareal crosstalk is important, as it reconstitutes in mice a dominant functional configuration that has been associated with conscious conditions in other species., This finding is also consistent with the postulates of prevailing theories of consciousness,, according to which functional networks that support awake, conscious states must exhibit global integration, evidenced in our data as greater functional coupling between cortical and subcortical network systems.

Our topographic investigations also corroborate the notion that during conscious wakefulness rsfMRI activity is distinctly characterized by the appearance of anticorrelation between the activity of different brain regions, a feature that is virtually absent in anesthetized conditions. It should be noted here that, because our rsfMRI time series were processed identically across all conditions and without regression of the global fMRI signal, this state-dependent change cannot be attributed to methodological artifacts. Analogous observations have been reported in primates, and humans, where they have been theoretically linked to the global neuronal workspace theory. According to this view, different streams of information compete to engage widespread networks of regions via the mutual inhibition of activity at different cortical areas, leading to anticorrelated dynamics. Within this framework, the observed segregation of medial corticolimbic and postero-lateral visuo-auditory cortical portions of the DMN in awake states is of interest, as it highlights a focal, state-dependent network reconfiguration occurring in awake rodents including rats, which, however, does not appear to have a direct correlate in higher mammalians, with the possible exception of New World primates. As this segregation affects a widely distributed community of monosynaptic connections,, we speculate it could reflect a dominant configuration aimed to enable increased cortical information capacity (i.e., the ensuing number of discriminable activity patterns), in the otherwise poorly differentiated rodent postero-lateral cortex.41, 42, 43 Such a functional segregation of the posterior and midline DMN components may not be necessary in higher mammalian species, owing to a larger and more specialized cortical differentiation.

Our deconstruction of awake rsfMRI activity in recurring CAPs revealed a unique dynamic structure recapitulating foundational principles of network organization in conscious primates and humans, including (1) the identification of a largely state-dominant CAPs characterized by a rich topographical organization, not rigidly anchored to the axonal structure, and by prominent anti-coordination between visual-auditory and DMN regions, and (2) stereotypic temporal trajectories in which state-specific CAPs are configured as dominant network attractors. By contrast, anesthetized state was associated with dominant temporal states that are topographically shaped by the underlying axonal structure with extensively integrated DMN and visual-auditory regions., These findings are consistent with prior theoretical conceptualizations of conscious network activity as a balanced recurrence of intermittent epochs of segregated and global brain synchronization. Above and beyond this, our results highlighted key state-dependent differences in the topographical organization of coactivation patterns, with evidence of distinctive involvement of BF, hypothalamic and thalamic areas in the awake, but not anesthetized, state. Owing to the tight link between CAP topography and the resulting patterns network activity, this finding is important as it suggests that rsfMRI network dynamics in awake states is critically shaped by BF and thalamic substrate areas, corroborating a previously postulated involvement of these regions to wakeful network organization. Importantly, our investigations of CAP topography show that state-specific differences in rsfMRI dynamics may be accompanied by significant state-specific topographical reconfiguration of the ensuing coactivation patterns, leading to the identification of functional substrates that may critically shape network activity. The future extension of our analytic approach to map state-dependent topographic differences in higher species is warranted to investigate the predictive and evolutionarily validity of our observations across species and consciousness states.

Importantly, our results provide key empirical data that might gage the predictive validity of current theoretical models of consciousness and help further develop and constrain emerging computational models of spontaneous brain dynamics in the mammalian brain. For example, the observation that CAPs and their oscillatory patterns are specifically reconfigured in conscious states might help refine current brain models based on the use of coupled oscillators. Furthermore, our empirical finding of increased divergence between functional connectivity and the underlying anatomical structure in awake conditions is broadly compatible with theoretical models describing the emergence of consciousness in terms or distribution or re-distribution of energy in re-entrently connected networks.45, 46, 47, 48, 49 The development of theoretical models that can predict the topographic and dynamic structure of brain-wide coactivation patterns in conjunction with targeted perturbations of the functional connectome might be key to rule in and rule out current hypotheses about wakefulness and the emergence of consciousness in the mammalian brain.

A strength of our approach is the use of a translatable readout that allowed us to relate the observed state-specific changes to analogous investigations in higher mammalian species. However, a few limitations need be recognized when extrapolating and comparing our findings across species and studies. First, the extent to which rsfMRI in habituated head-fixed mice and conceivably primates can recapitulate the quiet wakefulness state that characterize human rsfMRI studies remains unclear. Our corticosterone measurements are encouraging, as they show that stress response to our acclimation procedure was marginal. However, wakeful imaging in head-fixed rodents may possibly entail arousal states not directly comparable to those attainable in humans. rsfMRI studies encompassing concurrent pupil-tracking may crucially help relate some of the functional signatures we describe here to arousal-state fluctuations. Second, to ensure consistency with previously acquired datasets under halothane anesthesia, our investigations were carried out in male animals only. Future work is required to extend our results to female mice and pinpoint possible sex-related differences in the dynamics organization of brain networks across states. Third, the anesthetic regimens we used in our studies generally correspond to light anesthesia, and although they induce animal immobility and unresponsiveness, they might, at least conceivably, induce intermittent states of residual “consciousness.” The lack of internal state reporting in animals cautioned us to minimize use of “conscious state” when referring to the brain state associated with awake scanning. For this reason, all our references to conscious feature in this work are the results from the indirect comparison of our findings with corresponding investigations in human studies. Notwithstanding these limitations, the degree of correspondence between the dynamic signatures we found in the mouse and those highlighted by corresponding primate and human research is remarkable, supporting a putative extrapolation of these signatures across species. Finally, the significance and neural basis of the observed rsfMRI anticorrelation remain unclear, and future neurophysiological investigations will be required to establish whether this hemodynamic signature can be related to inhibitory neural engagement.

In conclusion, we describe a robust protocol for rsfMRI mapping in awake mice and report that the corresponding patterns of rsfMRI activity exhibit stereotypic spatiotemporal dynamics characterized by a dominant occurrence of coactivation patterns involving arousal-related nuclei, regional anticorrelation, and idiosyncratic temporal trajectories. These results suggest that dynamic structure of rsfMRI activity in the awake rodent brain recapitulates evolutionarily relevant principles predictive of conscious states in higher mammalian species and pave the way to the implementation of awake rsfMRI in this species.

STAR★Methods

Key resources table

Resource availability

Lead contact

Further information and requests for resources and reagents should be directed to and will be fulfilled by the Lead Contact, Alessandro Gozzi (alessandro.gozzi@iit.it).

Materials availability

Mouse head-holders and cradles are available at Ugo Basile (www.ugobasile.com).

Experimental model and subject details

In vivo experiments were conducted in accordance with the Italian law (DL 26/214, EU 63/2010, Ministero della Sanità, Roma) and with the National Institute of Health recommendations for the care and use of laboratory animals. The animal research protocols for this study were reviewed and approved by the Italian Ministry of Health and the animal care committee of Istituto Italiano di Tecnologia (IIT). All surgeries were performed under anesthesia.

Animals

Adult (< 6 months old) male C57BL/6J mice were used throughout the study. Mice were group housed in a 12:12 hours light-dark cycle in individually ventilated cages with access to food and water ad libitum and with temperature maintained at 21 ± 1 °C and humidity at 60 ± 10%.

Experimental groups and datasets

A first group of mice (n = 10, awake dataset) was subjected to the head-post surgery and habituation procedure and underwent awake fMRI image acquisitions as described below. The scans so obtained constitute the awake rsfMRI dataset we used throughout our study. Two additional groups of age matched male C57BL/6J mice were used as reference rsfMRI scans under anesthesia. The first group of animals (n = 19, halothane dataset) was previously scanned in our laboratory under shallow halothane anesthesia, 0.75%. In the present work we have used this dataset as benchmark reference anesthesia dataset for all comparisons with awake fMRI networks, because the employed anesthesia regimen is well characterized,, it is representative of the network architecture observed with different anesthesia regimens in rodents, it exhibits rich spatiotemporal dynamics by preserving spectral properties of BOLD fluctuations and it closely models the organization of the underlying axonal connectome.

To probe a possible generalizability of our findings to other anesthetic conditions, we imaged a second, separate group (n = 14) of mice under medetomidine-isoflurane anesthesia (0.05 mg/kg bolus and 0.1 mg/kg/h IV infusion, plus 0.5% isoflurane). While this anesthetic combination shifts the spectral components of BOLD fluctuations towards higher frequencies (hence departing from the characteristic 1/f power law distribution that characterizes awake and halothane rsfMRI datasets,), it nonetheless represents the most-widely anesthetic mixture used in the rodent imaging community,, and for this reason we used it as a supplementary reference dataset. All the imaged mice were bred in the same vivarium and scanned with the same MRI scanner and imaging protocol employed for the awake scans (see below).

An additional cohort of n = 14 male mice was used to assess the effect of habituation and scanning on acute and chronic stress levels, as assessed with blood corticosterone level measurements, body-weight tracking, and an open field behavioral test. Within this cohort, n = 7 randomly selected “awake fMRI” mice were subjected to the same head-post surgery, habituation and awake scanning as our main rsfMRI dataset. Control littermates (n = 7) were not implanted, habituated or fMRI scanned, but always left in their home cages with the exception of an initial in-cage blood corticosterone sampling (baseline, Figure S1) and an additional one after an open field behavioral test. This group of mice was used to obtain reference baseline corticosterone levels timed to those obtained in the experimental group. The actual blood sampling timepoints corresponded to day 15 post-surgery in the fMRI awake cohort (handling, Figure S1A), and day 39 post surgery, right after the first rsfMRI scan, which we carried out in the fMRI awake group only. Both sets of mice were then subjected to the open field test to assess behavioral phenotypes indicative of chronic stress. Blood sampling was carried out at the end of the handling procedure session (fMRI awake), or in the home cage (control), and at the end of the MRI session (for awake mice) or open field test (for control mice). Housing was organized such that each cage contained a mix of both control and awake mice.

Method details

Headpost surgery

Head fixation during awake imaging was achieved using custom-made headposts (Ugo Basile, Italy; Figure 1B). A surgical procedure was performed to adhere the headposts to the skull. During surgery mice were anesthetized with isoflurane (5% induction and 2% surgery) and head fixed on a stereotaxic apparatus (Stoelting Co.) while body temperature was maintained with a heating pad at 37 °C. The fur on the head was removed and the skin was triple scrubbed with 70% ethanol. Once it was confirmed that the surgical anesthetic plane was maintained, a vertical incision (from a few mm anterior to bregma to the posterior end of the skull up to the eye level) was cut and the skin on top of the skull was removed. The headpost was cemented to the skull with a two-step procedure: first, dental cement (Superbond C&B kit, Dental leader) was applied to the skull surface; the headpost was pressed gently on top until the cement was partially dried (1-2 minutes). A second cement (Paladur, Kulzur s.r.l) was then applied to the skull and the headpost and allowed to dry for 5 minutes. Dental cement was applied to the edges of the surrounding skin to ensure complete coverage of all exposed bone. No suturing was required because the skin was adhered to the bone with the cement. Once the procedure was complete, mice were allowed to recover in their home cages with repeated monitoring. The headposts were well tolerated and showed no deterioration nor detachment over a 10-week examination window. After surgery, head post implanted mice could be group-housed without apparent direct damage to other headposts.

Habituation

A custom-made MRI-compatible animal cradle (Ugo Basile, Italy; Figures 1A–1C) was designed such that the headposts could be secured to the cradle during scanning. A habituation protocol to acclimate the mice to immobilization and scanning was initiated 10–15 days after headpost surgery to allow sufficient time for recovery. The habituation timeline is depicted in Figure 1D, and was performed in two steps, entailing an initial stage of handling followed by extensive mock scanning. For the first 3 days (habituation day [D]1-3), 15 minutes daily were dedicated to acclimating the mice to the experimenter. On D4 and D5, mice were allowed to explore the cradle for 10 minutes, but on D5 the experimenter also grabbed and held the headpost by hand for a few seconds (during the 10 minutes) to acquaint the mouse with head fixation. Habituation to the mock scanner began on D8. During mock scanning, the headpost was secured to the cradle with plastic screws, a 3D printed copy of the RF-receive coil was placed over the head of the mouse, and the body was left unrestrained for the first two mock scanning habituation sessions. In the third mock scanning habituation session, the body of the mouse was gently restrained by taping its back to the cradle arc. Specifically, the mouse back was gently taped to the cradle in a way that allowed the animal to breathe comfortably and make small movements below the solid edges of the cradle. Its tail was also taped to the base of the cradle but its forepaws and hind paws were not. Cotton rolls were placed to reduce jaw and forepaw movement. A sound system played an audio recording of the EPI pulse sequence at increasing loudness to reach dB levels recorded in the scanner bore. Figure 1D shows the duration and sound level for each daily session of mock scanning (D8-D27).

Open field test (OFT)

The open field test was adapted from existing protocols. Testing was carried out in a square arena (44cm x 44cm x 44cm) with a homogenous 130 lux illumination subdivided in a central zone (24 cm width) and a peripheral zone (10 cm width). Mice were placed in the apparatus for a 20 minutes test period, during which total distance travelled (cm), immobile time (s), and central distance (m) were automatically scored by Anymaze software (Stoelting Co). Behavioral data were analyzed performing an unpaired t-test between groups.

Plasma corticosterone measurement

The overall procedure of blood collection was no longer than 5 min and all blood collections (and corresponding tests or scanning, where applicable) were performed between 9:00 and 12:00 h to minimize effect circadian excursion on corticosterone levels. Blood sampling was carried out using tail vein bleeding. The mouse tail was superficially cut with a sterile scalpel blade and blood was collected with EDTA tubes (Microvette® 200 K3 EDTA, Sarstedt, Nümbrecht, Germany). Pressure was gently applied to stop any bleeding before returning to the cage. All blood samples were centrifuged at 4000 g for 5 min at 4°C to obtain plasma, and stored at −80°C until further analysis. Blood corticosterone concentration was measured with an enzyme-linked immunosorbent assay (ELISA, Enzo Life Sciences, Inc., USA). All our reported corticosterone concentration levels refer to the amount of corticosterone in plasma, and are expressed in nanograms per milliliter (ng/ml).

rsfMRI acquisition

For awake scanning, the mouse was secured (using the headpost) to the custom-made MRI-compatible animal cradle and the body of the mouse was gently restrained by taping its back to the cradle arc as described in the habituation paragraph. For scanning under light anesthesia, mice were first deeply anesthetized with isoflurane (4% induction), intubated and artificial ventilated (90 BPM). In one group, anesthesia was then switched to halothane (0.75%). In a second group, a bolus of medetomidine (0.05 mg/kg) was given via tail vein cannulation before waiting 5 minutes and starting an infusion of medetomidine (0.1 mg/kg/h) with isoflurane reduced to 0.5%. In both cases, the rsfMRI acquisition started 30 minutes after the switch to light anesthesia.

All scans were acquired at the IIT laboratory in Rovereto (Italy) on a 7.0 Tesla MRI scanner (Bruker Biospin, Ettlingen) with a BGA-9 gradient set, a 72 mm birdcage transmit coil, and a four-channel (awake, halothane) or three-channel (medetomidine-isoflurane) solenoid receive coil. Awake and medetomidine-isoflurane rsfMRI scans were acquired using a single-shot echo planar imaging (EPI) sequence with the following parameters: TR/TE=1000/15 ms, flip angle=60°, matrix=100 x 100, FOV=2.3 x 2.3 cm, 18 coronal slices (voxel-size 230 x 230 x 600 μm), slice thickness=600 μm and 1920 time points, for a total time of 32 minutes. Mice under halothane anesthesia (n = 19) were scanned with a TR/TE=1200/15ms, flip angle=60°, matrix=100 x 100, 24 coronal slices (voxel-size 200 x 200 x 500 μm), for a total of 1600 time points, total acquisition time of 32 minutes as described in Gutierrez-Barragan et al. Figure S4C depicts a representative EPI frame from an awake scan. No animal was excluded from this study.

Data Preprocessing

Preprocessing of BOLD images (Figure S6A) was carried out as described in previous work., Briefly, the first 2 minutes of the time series were removed to account for thermal gradient equilibration. RsfMRI timeseries were then time despiked (3dDespike, AFNI), motion corrected (MCFLIRT, FSL), skull stripped (FAST, FSL) and spatially registered (ANTs registration suite) to an in-house mouse brain template with a spatial resolution of 0.23 x 0.23 x 0.6 mm3. To optimize awake fMRI timeseries preprocessing we systematically subjected rsfMRI time series to a set of increasingly stringent denoising and motion correction strategies involving the regression of either 7, 13 or 25 nuisance parameters (Figure S6A). These were: average cerebral spinal fluid signal plus 6, 12 or 24 motion parameters determined from the 3 translation and rotation parameters estimated during motion correction, their temporal derivatives and corresponding squared regressors. No global signal regression was employed as this procedure artificially introduces regional anticorrelation, and has been reported to reduce antero-posterior extension of fMRI connectivity in the mouse, hence decoupling structural and functional connectivity. In-scanner head motion was quantified via calculations of frame-wise displacement (FD). Average FD levels in awake conditions were comparable to those obtained in anesthetized animals (halothane) under artificial ventilation (p = 0.13, Student t test, Figure S6B). To rule out a contribution of residual head-motion in awake scans, we further introduced frame-wise fMRI scrubbing using a very stringent FD threshold (FD > 0.075 mm, compared to 0.1 mm 0.75 mm, cf. with 1 mm in Grandjean et al.). Quantifications of putatively motion-contaminated volumes at this stringent threshold revealed a predictably higher proportion of labelled frames in awake mice compared to anesthetized subjects (Figure S6B). However, the use of denoising pipelines of increased stringency, with or without strict frame scrubbing (Figure S6C), did not reveal any difference in interregional fMRI connectivity strength (p > 0.89, one way ANOVA, all regions Figure S6C), hence arguing against a significant contribution of in-scanner head-motion to our imaging findings. We nonetheless carried out all of our further analyses (both in awake and anesthetized mice) using the most stringent denoising pipeline assessed, i.e. regression of 25 nuisance parameters followed by strict FD > 0.075 mm volume censoring. The resulting time series were band-pass filtered (0.01-0.1 Hz band) and then spatially smoothed with a Gaussian kernel of 0.5 mm full width at half maximum.

Quantification and statistical analysis

Seed-based Functional Connectivity

To investigate the topography of functional networks in the awake mouse, we performed a seed-based functional connectivity (FC) investigation using predefined regions of interest (size 4 x 4 x 1 voxels) known to be network hubs in the anesthetized mouse brain., These were selected to include both cortical and subcortical systems such as the prefrontal (PFC) and retrosplenial cortex (RS) within the default-mode network (DMN); primary motor (M1) and somatosensory (S1) regions within the latero-cortical network (LCN); visual (VIS) and auditory (A) regions in posterolateral networks (PLN); the insula (Ins) within the salience network (SN); and other subcortical substrates within the hippocampus (HC), thalamus (TH), basal forebrain (BF), and striatum (ST). Canonical correlation mapping was done for each subject by computing the Pearson product-moment correlation coefficient (Pearson’s correlation, for short) between the average signals extracted from each seed, with that of each voxel. The spanned single-subject correlation maps were then transformed using Fisher’s r-to-z transform; averaged across all animals; and thresholded to significant connections with t-scores > 2.8 (two-tailed t-test, p < 0.01, FDR corrected). Averaged maps were then re-transformed to correlation values (r-scores). Group-level seed-based FC differences between awake and anesthetized groups were assessed by means of two tailed, two-sample t-test (p < 0.01, family-wise error cluster-corrected, with defining threshold of t > 2.8).

Voxel-wise Functional Connectivity Analyses

Voxel-wise FC matrices were computed for both awake and anesthetized mice. This was done by computing for each subject the Pearson’s correlation between each voxel and each other voxel in the brain (resulting in a 6843 x 6843 matrix). Single-subject FC matrices were Fisher transformed and averaged across animals in each group and re-transformed to r-scores. Comparison between groups was done by means of a two-sample t-test (two-tailed, p < 0.05, FDR corrected, critical p = 0.009). To avoid topographical bias and help interpretability of our findings, we organized voxels into axonal connectivity modules recently identified in the mouse brain. These include a default-mode (DMN); Latero-cortical (LCN), Hippocampus (HC), and Olfactory-Basal Forebrain (OF-BF) networks. Given the observation that some thalamic regions are associated to the axonal DMN module, and others to the LCN, we deliberately separated these into sub-modules for visualization purposes. For the same purposes, we also separated voxels within the DMN corresponding to midline cortico-limbic areas, and posterolateral visual-auditory ones. With this, we computed voxel-wise functional connectivity matrices, and averaged them within groups.

Our analyses using the structural connectome leveraged a high-resolution model of the mouse brain connectome (100 μm3) previously released by Knox and colleagues and recently resampled as described in Coletta et al. The Knox connectome is based on 428 viral microinjection experiments in C57BL/6J male mice obtained from the Allen Mouse Brain Connectivity Atlas (http://connectivity.brain-map.org/). The connectome data were derived from green fluorescent protein (eGFP)-labeled axonal projections that were then registered to the Allen Mouse Brain Atlas, and aggregated according to a voxel-wise interpolation model (Knox et al., 2018). In Coletta et al., we provide a comprehensive description of the computational steps implemented to make this resource computationally tractable and suited to the topological analyses described in the present manuscript. Briefly, resampling of the Knox et al. connectome was carried out by aggregating neighboring voxels according to a Voronoi diagram based on Euclidean distance between neighboring voxels (Figure S8 in Coletta et al.). The used Voronoi-based aggregation strategy entails the identification, for each voxel of the mouse connectome, of its 27 closest neighbors as per Euclidean distance, and the subsequent averaging of their connectivity profiles into a single value. The average spatial extension of the obtained Voronoi voxels in each plane corresponds to 242 μm × 323 μm × 336 μm in the x (sagittal), y (horizontal), and z (coronal) planes, respectively. We next registered the structural modules obtained from the down-sampled structural connectome into the EPI space, performing the structure function correspondence analysis at the voxel level.

Structure-function correspondence

Structure-function correspondence was calculated using communities defined by the structural connectome. In our prior work we found the identified macro-communities to spatially encompass four macroscopic network systems of the mouse brain that spatially correspond to analogous rsfMRI macro-networks, i.e. the DMN, the LCN, the hippocampus and olfactory-basal forebrain areas (see Figure S2A). To probe the relationship between the functional connectome and the underlying axonal structure we next computed structure-function correspondence (Figure 4B) as previously described:

Zw = within-network functional connectivity strength i.e. mean connectivity strength of edges between all pairs of nodes within the same structural network.

Zb = between-network functional connectivity strength i.e. mean connectivity strength of edges between all pairs of nodes that spanned two structural networks.

To depict how intra- and inter-modular communication changes across conditions, we mapped the top 10% strongest functional connections at the group level (absolute value, as in Chelini et al.) to obtain a network diagram (Figure 4A) using an anatomical parcellation derived from the Allen Brain Institute. We next quantified for each subject the mean functional connectivity for all the voxels within each module, as well as the average of between module connections, again retaining the top 10% strongest functional links at the subject level. Between-group differences were assessed using a Mann-Whitney test (p < 0.05, FDR corrected).

Whole brain co-activation patterns (CAPs) analysis

To investigate how rsfMRI dynamics is affected by wakefulness or anesthesia, we used the co-activation patterns approach. This method classifies fMRI volumes into clusters based on their spatial similarity, and the averaged frames within each cluster are then taken to represent recurrent patterns of rsfMRI BOLD co-activation. Specifically, following Huang et al., after censoring motion-contaminated frames (FD > 0.075 mm), we ran the k-means clustering algorithm (spatial correlation as distance metric, 500 iterations, 5 replications with different random initializations) on the concatenated data including all frames from all subjects in both awake and anesthetized groups.

We set k = 6 clusters for two main reasons. First, in previous work we verified that choosing k = 6 clusters effectively describes most rsfMRI dynamics across multiple halothane-anaesthetized mouse datasets, with over 60% of the variance explained within each dataset with a limited number of parameters, and diminishing returns using larger cluster numbers; high reproducibility of the clustering algorithm across random initializations; and high replicability across independently acquired datasets. Second, we further verified in this study that k = 6 also accounts for a large fraction of variance explained in the collective dataset of awake and anaesthetized mice, with k = 6 being in the “elbow” region of the variance explained as a function of the number of clusters. These results were supported by independent clustering analyses using the concatenated datasets and also clustering frames from each dataset independently (Figure S4A), showing that in each scenario, with k>6, higher-order clustering did not offer a gain in variance explained higher than 1% (Figure S4B).

The chosen number of clusters was also highly stable across different random initializations, with spatial correlations across CAPs obtained with different initializations greater than 0.99 for all CAPs. For each subject, the fMRI frames classified within a cluster were next voxel-wise averaged to create a single-subject CAP map. Within each group, these maps were then averaged for each CAP at the group level, and thresholded to voxels with significant mean BOLD co-activations or co-deactivations (two-tailed t-test, p < 0.01, FWER cluster-corrected, with defining threshold of t > 2.8). Between-group comparisons for each CAP were done by a voxel-wise two-sample t-test (two-tailed, p < 0.01, FWER cluster-corrected, with defining threshold of t > 2.8). Importantly, we repeated the k = 6 clustering without scrubbing fMRI frames, revealing CAP maps virtually indistinguishable from those obtained in scrubbed timeseries (Figure S4F). We also computed, for each mouse in each group, the proportion of fMRI frames flagged as high-motion (FD > 0.075 mm) that were also associated to a specific CAP. We then compared the mean of these proportions between CAPs for each group, revealing that no CAP was significantly associated to high-motion frames (Figure S4E).

For each mouse in each group, we computed the occurrence rate of each CAP as the proportion of frames associated to the respective cluster. Between group comparisons of CAP occurrence rates were performed using a two-tailed t-test (p < 0.05, FDR corrected). We then computed the spatial similarity (Pearson’s correlation) between group-averaged CAP maps to verify CAP mirroring motifs structure previously described. We further explored the relation of CAPs to FC by transforming, for each group, the mean co-activation map into a co-fluctuation matrix, multiplying each voxel’s mean BOLD intensity within a CAP with each other voxel’s. This procedure yields a voxel-wise representation of the concordant (or diverging) peaks of BOLD activity that characterize each CAP. Given the robust mirror structure of each CAP anti-CAP pair (cf. Figure 5C), the resulting cofluctuation matrices are characterized by virtually indistinguishable cofluctuation structure. To spatially link CAP cofluctuation structure to steady-state rsfMRI connectivity, these co-fluctuation matrices were group-averaged and their similarity with the static FC matrix was assessed with Pearson’s correlation. We also verified the robustness of our results by comparing CAP occurrence rate in awake and anesthetized mice before and after scrubbing high-motion frames, revealing largely comparable between-group differences (Figures S4G and S4H).”

CAP infraslow dynamics

To investigate the dynamics of each CAP, we generated instantaneous CAP-to-frame correlation time courses at the subject-level. Subsequently, the power spectrum of CAP time courses and the Global fMRI Signal (GS, i.e. the instantaneous average of all in-brain voxels) were computed and averaged for each frequency in each group for the 0.01–0.1 Hz range (Figures 5A and 5B). These CAP-to-frame correlation time courses are depicted in Figure S4D along with the frame-wise displacement trace, providing an example of how CAP time course peaks do not co-occur preferentially at FD peaks. To summarize the relative contribution of the infra-slow band to CAP dynamics and that of the Global fMRI Signal (GS), we further computed the fractional amplitude of low frequency fluctuation, or fALFF, as the proportion of power in the 0.01–0.03 Hz band. fALFF values were compared between groups with a Mann-Whitney test (p < 0.05, FDR corrected). This analysis, and the subsequent investigation of CAP occurrence within GS cycles described below, were limited to the sole halothane dataset, as the spectral properties of medetomidine-isoflurane anesthesia present a well-characterized spectral shift towards higher frequencies that is non representative of the 1/f-like spectral profile observed in awake conditions (cf. Figure 6).

To map CAP occurrence within the common temporal reference of the phase of the Global fMRI Signal (GS), we computed the instantaneous phase of the filtered (0.01-0.03 Hz) GS using the Hilbert Transform, and divided the trace into cycles of minimum 30 s and maximum 100 s. We sampled the GS phase values within cycles at each CAP’s occurrence, and to build phase occurrence distributions of CAPs, we retained only time frames when the normalized CAP time course at that instant was above 1 SD, hence ensuring the selection of frames that are reasonably well assigned to a specific CAP. We then built the distribution of GS phases at the occurrence of each CAP. Using MATLAB’s CircStats toolbox, we tested if the so obtained GS-phase distributions at each CAP’s occurrence deviated from circular uniformity (Rayleigh test, p < 0.001, FDR corrected).

Entropy of CAP Markov trajectories

To investigate the dynamics of transitions between CAPs, we defined for each group, a concatenated sequence of CAP occurrences across subjects, and computed the transition probability matrix as the probability of switching from a certain CAP at time to another CAP at time . Only transitions within the same subject were included. We first considered matrices in which we counted the auto-transitions , namely persistence probabilities., The off-diagonal , namely transition probabilities from CAP to CAP , were then assessed after building matrices from a sequence in which we removed repeating elements in order to control for autocorrelations given the CAP’s dwell time., The prevalence of a directional transition was computed by taking the difference between a transition probability from to and the transition probability from j to . We used the off-diagonal matrices to measure the entropy of Markov trajectories (HMT).,, This method computes the descriptive complexity of a trajectory between CAPs (in bits), where a higher complexity prescribes more information required to access an ending CAP from an initial CAP ; hence it is less accessible as it transitions to other CAPs before reaching its destination. Specifically, for a Markov Chain (MC) representing a CAP sequence with transition probability matrix P, we define the Entropy Rate per step as:where is the stationary distribution solving . We then define the Markov entropy of a trajectory from CAP to CAP as:Where , is the identity matrix, and is a matrix of stationary probabilities . Here is the matrix of single-step entropies (from CAP to any CAP ), and is a diagonal entropy matrix with trajectories from one CAP to itself , with zeros if .

Persistence probabilities were tested for significant deviations from those obtained from random sequences by generating 1000 permutations of CAP occurrence sequences at the subject level and then concatenating the sequences. Differences between each group’s persistence probability was assessed also with these random surrogates and comparing their group differences between auto-transitions with the real one. Transition probabilities between different CAPs, and the prevalent directional transitions were instead tested for significance above null elements of matrices built after randomly permuting the non-repeating sequences 1000 times at the subject level before concatenating them in each group (removing an element if ). These surrogates were also used to test the significance of between group differences; group-level entropies of Markov trajectories and between group differences.

To rule out the possibility that results were not defined by outlier subjects, we split the groups of subjects into two equal partitions 500 times, and computed the CAP persistence and transition probabilities as well as Markov entropy of the trajectories. Then, we assessed the similarity between the matrices from each split-half sample by means of correlations. We further tested our results at the single-subject level by re-computing for each subject, the persistence and transition probabilities, and comparing them with the results obtained from the concatenated sequences, again by means of correlations between the matrices.

REAGENT or RESOURCE	SOURCE	IDENTIFIER
Critical commercial assays
Corticosterone ELISA kit, USA	Enzo Life Sciences, Inc.	ADI-900-097
Deposited data
rsfMRI dataset #1, awake (10 mice)	This paper	Mendeley Data - http://dx.doi.org/10.17632/np2fx99hn6.2
rsfMRI dataset #2, anesthesia – halothane (19 mice)	This paper	Mendeley Data - Part 1: http://dx.doi.org/10.17632/354f8dc8xh.2. Part 2: http://dx.doi.org/10.17632/3tc72r2m4z.2
rsfMRI dataset #1, anesthesia – Medetomidine-Isoflurane (14 mice)	This paper	Mendeley Data - http://dx.doi.org/10.17632/fmb2fwb53f.2
Experimental models: Organisms/strains
Mouse: C57BL/6J	Charles River	C57BL/6J; RRID: IMSR_JAX:000664
Software and algorithms
MATLAB 2017b	MathWorks	https://www.mathworks.com; RRID: SCR_001622
Image clustering	MATLAB Statistics and Machine learning Toolbox	https://www.mathworks.com/help/stats/index.html; RRID: SCR_001622
Signal processing	MATLAB Signal Processing Toolbox	https://www.mathworks.com/help/signal/index.html; RRID: SCR_001622
Tools for NIfTI and ANALYZE image	MATLAB NIFTI Toolbox	https://it.mathworks.com/matlabcentral/fileexchange/8797-tools-for-nifti-and-analyze-image
Circular Statistics	MATLAB Circular Statistics Toolbox	https://www.mathworks.com/matlabcentral/fileexchange/10676-circular-statistics-toolboxdirectional-statisticsstatistics; RRID: SCR_016651
Entropy of Markov Trajectories	MATLAB Entropy of Markov Trajectories Toolbox	https://github.com/stdimitr/Entropy_of_Markov_Trajectories.git
FSL	FMRIB Software Library	https://fsl.fmrib.ox.ac.uk/fsl/fslwiki/; RRID: SCR_002823
GraphPad 9.2	Prism	www.graphpad.com; RRID:SCR_002798
SPM	Wellcome Department of Cognitive Neurology, London, UK	https://www.fil.ion.ucl.ac.uk/spm/; RRID: SCR_002823
AFNI	SSCC at NIMH	https://afni.nimh.nih.gov/; RRID: SCR_005927
Nilearn 0.5.0	Nilearn: statistics for neuroimaging with Python	https://nilearn.github.io/stable/index.html; RRID: SCR_001362
Python 3.5.6	Anaconda	https://www.anaconda.com/products/individual; RRID: SCR_018317
Gephi 0.9.2 Graph Visualization	Gephi	https://gephi.org/users/download/; RRID: SCR_004293
Resting state fMRI BASH preprocessing pipeline	Functional Neuroimaging Lab., CNCS, IIT	https://github.com/functional-neuroimaging/rsfMRI-preprocessing