Sonsoles Alonso

To elucidate the neurobiological basis of cognition, which is dynamic and evolving, various methods have emerged to characterise time‐varying functional connectivity (FC) and track the temporal evolution of functional networks. However, given a selection of regions, many of these methods are based on modelling all possible pairwise connections, diluting a potential focus of interest on individual connections. This is the case with the hidden Markov model (HMM), which relies on region‐by‐region covariance matrices across all pairs of selected regions, assuming that fluctuations in FC occur across all investigated connections; that is, that all connections are locked to the same temporal pattern. To address this limitation, we introduce Targeted Time‐Varying FC (T‐TVFC), a variant of the HMM that explicitly models the temporal fluctuations between two sets of regions in a targeted fashion, rather than across the entire connectivity matrix. In this study, we apply T‐TVFC to both simulated and real‐world data. Specifically, we investigate thalamocortical connectivity, hypothesising distinct temporal signatures compared to corticocortical networks. Given the thalamus's role as a critical hub, thalamocortical connections might contain unique information about cognitive processing that could be overlooked in a coarser representation. We tested these hypotheses on high‐field functional magnetic resonance data from 60 participants engaged in a reasoning task with varying complexity levels. Our findings demonstrate that the time‐varying interactions captured by T‐TVFC contain task‐related information not detected by more traditional decompositions.

To elucidate the neurobiological basis of cognition, which is dynamic and evolving, various methods have emerged to characterise time-varying functional connectivity (FC) and track the temporal evolution of functional networks. However, given a selection of regions, many of these methods are based on modelling all possible pairwise connections, diluting a potential focus of interest on individual connections. This is the case with the hidden Markov model (HMM), which relies on region-by-region covariance matrices across all pairs of selected regions, assuming that fluctuations in FC occur across all investigated connections; that is, that all connections are locked to the same temporal pattern. To address this limitation, we introduce Targeted Time-Varying FC (T-TVFC), a variant of the HMM that explicitly models the temporal fluctuations between two sets of regions in a targeted fashion, rather than across the entire connectivity matrix. In this study, we apply T-TVFC to both simulated and real-world data. Specifically, we investigate thalamocortical connectivity, hypothesizing distinct temporal signatures compared to corticocortical networks. Given the thalamus' role as a critical hub, thalamocortical connections might contain unique information about cognitive processing that could be overlooked in a coarser representation. We tested these hypotheses on high-field functional magnetic resonance data from 60 participants engaged in a reasoning task with varying complexity levels. Our findings demonstrate that the time-varying interactions captured by T-TVFC contain task-related information not detected by more traditional decompositions.

Time-varying functional connectivity (FC) methods are used to map the spatiotemporal organization of brain activity. However, their estimation can be unstable, in the sense that different runs of the inference may yield different solutions. But to draw meaningful relations to behavior, estimates must be robust and reproducible. Here, we propose two solutions using the hidden Markov model (HMM) as a descriptive model of time-varying FC. The first, best ranked HMM, involves running the inference multiple times and selecting the best model based on a quantitative measure combining fitness and model complexity. The second, hierarchical-clustered HMM, generates stable cluster state time series by applying hierarchical clustering to the state time series obtained from multiple runs. Experimental results on fMRI and magnetoencephalography data demonstrate that these approaches substantially improve the stability of time-varying FC estimations. Overall, hierarchical-clustered HMM is preferred when the inference variability is high, while the best ranked HMM performs better otherwise.

Recurrence in major depressive disorder (MDD) is common, but neurobiological models capturing vulnerability for recurrences are scarce. Disturbances in multiple resting‐state networks have been linked to MDD, but most approaches focus on stable (vs. dynamic) network characteristics. We investigated how the brain's dynamical repertoire changes after patients transition from remission to recurrence of a new depressive episode. Sixty two drug‐free, MDD‐patients with ≥2 episodes underwent a baseline resting‐state fMRI scan when in remission. Over 30‐months follow‐up, 11 patients with a recurrence and 17 matched‐remitted MDD‐patients without a recurrence underwent a second fMRI scan. Recurrent patterns of functional connectivity were characterized by applying Leading Eigenvector Dynamics Analysis (LEiDA). Differences between baseline and follow‐up were identified for the 11 non‐remitted patients, while data from the 17 matched‐remitted patients was used as a validation dataset. After the transition into a depressive state, basal ganglia‐anterior cingulate cortex (ACC) and visuo‐attentional networks were detected significantly more often, whereas default mode network activity was found to have a longer duration. Additionally, the fMRI signal in the basal ganglia‐ACC areas underlying the reward network, were significantly less synchronized with the rest of the brain after recurrence (compared to a state of remission). No significant changes were observed in the matched‐remitted patients who were scanned twice while in remission. These findings characterize changes that may be associated with the transition from remission to recurrence and provide initial evidence of altered dynamical exploration of the brain's repertoire of functional networks when a recurrent depressive episode occurs.

Time-varying functional connectivity methods are used to map the spatiotemporal organization of brain activity. However, their estimation can be unstable, in the sense that different runs of the inference may yield different solutions. But to draw meaningful relations to behaviour, estimates must be robust and reproducible. Here, we propose two solutions using the Hidden Markov Model (HMM) as a descriptive model of time-varying FC. The first, best-ranked HMM, involves running the inference multiple times and selecting the best model based on a quantitative measure combining fitness and model complexity. The second, hierarchical clustered HMM, generates stable aggregated state timeseries by applying hierarchical clustering to the state timeseries obtained from multiple runs. Experimental results on fMRI and MEG data demonstrate that these approaches substantially improve the stability of time-varying FC estimations. Overall, hierarchical clustered HMM is preferred when the inference variability is high, while the best-ranked HMM performs better otherwise.

Introduction: Recurrence in major depression disorder (MDD) is common, but neurobiological models capturing vulnerability for recurrences are scarce. Disturbances in multiple resting-state networks have been linked to MDD, but most approaches focus on stable (vs. dynamic) network characteristics. We investigated how the brain’s dynamical repertoire changes after patients transition from remission to recurrence of a new depressive episode. Methods: Sixty drug-free, MDD-patients with ≥2 episodes underwent a baseline resting-state fMRI scan when in remission. Over 30-months follow-up, 11 patients with a recurrence and 17 matched-remitted MDD-patients without a recurrence underwent a second fMRI scan. Recurrent patterns of functional connectivity were characterized by applying leading eigenvector dynamics analysis (LEiDA). Differences between baseline and follow-up were identified for the 11 non-remitted patients, while data from the 17 matched-remitted patients was used as a validation dataset. Results: After the transition into a depressive state, the reward and a visuo-attentional networks were detected significantly more often, whereas default mode network activity was found to have a longer duration. Additionally, the fMRI signal in the areas underlying the reward network were significantly less synchronized with the rest of the brain after recurrence (compared to a state of remission). These changes were not observed in the matched-remitted patients who were scanned twice while in remission. Conclusion: These findings characterize the changes that are specifically associated with the transition from remission to recurrence and provide first evidence of increased segregation in the brain’s dynamical repertoire when a recurrent depressive episode occurs.

Brain function depends on the flexible and dynamic coordination of functional subsystems within distributed neural networks operating on multiple scales. Recent progress has been made in the characterization of functional connectivity (FC) at the whole-brain scale from a dynamic, rather than static, perspective, but its validity for cognitive sciences remains under debate. Here, we analyzed brain activity recorded with functional Magnetic Resonance Imaging from 71 healthy participants evaluated for depressive symptoms after a relationship breakup based on the conventional Major Depression Inventory (MDI). We compared both static and dynamic FC patterns between participants reporting high and low depressive symptoms. Between-group differences in static FC were estimated using a standard pipeline for network-based statistic (NBS). Additionally, FC was analyzed from a dynamic perspective by characterizing the occupancy, lifetime, and transition profiles of recurrent FC patterns. Recurrent FC patterns were defined by clustering the BOLD phase-locking patterns obtained using leading eigenvector dynamics analysis (LEiDA). NBS analysis revealed a brain subsystem exhibiting significantly lower within-subsystem correlation values in more depressed participants (high MDI). This subsystem predominantly comprised connections between regions of the default mode network (i.e., precuneus) and regions outside this network. On the other hand, LEiDA results showed that high MDI participants engaged more in a state connecting regions of the default mode, memory retrieval, and frontoparietal network (p-FDR = 0.012); and less in a state connecting mostly the visual and dorsal attention systems (p-FDR = 0.004). Although both our analyses on static and dynamic FC implicate the role of the precuneus in depressive symptoms, only including the temporal evolution of BOLD FC helped to disentangle over time the distinct configurations in which this region plays a role. This finding further indicates that a holistic understanding of brain function can only be gleaned if the temporal dynamics of FC is included.