Dynamic functional connectivity MRI allows for observation of correlations among brain areas without the assumption they are stable over time. This talk will discuss how concepts from steady-state functional connectivity can be expanded to account for dynamic changes, and how those changes could reshape our understanding of brain function at different levels of consciousness, during task performance, and in diseases. Current lines of thoughts on the possible mechanisms resulting in the non-stationarity of the signals will be discussed.
Dynamic fcMRI: Mechanisms & Applications
Functional connectivity MRI (fcMRI) studies the correlation patterns of BOLD signal fluctuations during an assumed steady state (Biswal, Yetkin et al. 1995). Nonetheless, it is known that the patterns of brain connectivity change at different sleep stages (Horovitz, Braun et al. 2009). For example, the frontal and parietal regions of the default mode network (DMN) disconnect during deep sleep (Horovitz, Braun et al. 2009), but they reconnect during REM sleep (Chow, Horovitz et al. 2013). The amplitude of the BOLD signal fluctuations in sensory areas correlates with the inverse index of wakefulness, a metric derived from the electroencephalography (EEG) (Horovitz, Fukunaga et al. 2008). When levels of consciousness are decreased by propofol, increases in frequency and decreases in long-range correlations are observed in the frontal lobe, thalamus and salience network (Tagliazucchi, Chialvo et al. 2016). Put together, these findings suggest the characteristics of the BOLD signal fluctuations and the long-range correlations might reflect brain intrinsic states. Models of effective connectivity among brain areas indicate their connectivity is altered by external stimuli (Horwitz, Warner et al. 2005). When subjects perform a task, underneath the activations, connectivity changes can be observed in the residual of the BOLD signal (Horovitz, Berman et al. 2010). These changes are also present on more complex tasks and at short time intervals. It has been shown that when subjects perform diverse activities such as math computations, watching a video, n-back, and memory tasks distinctive meta-stable states / quasi stable states that differ from rest can be derived from the fcMRI data (Gonzalez-Castillo, Saad et al. 2012). These defined “states” that can be recognized in just few seconds and can used for classification. Moreover, classification results are poor in low performer subjects, suggesting the connectivity pattern is indeed a marker of the task performance, in line with previous studies showing correlations within the central nodes of the DMN are present even during tasks, and, at the group level, the strength of their correlation is predictive of performance (Hampson, Driesen et al. 2006). While a typical fcMRI study is of the order of 5 to 10 minutes, it is clear now that the dynamic changes are of the order of seconds to minutes (Chang and Glover 2010). Exploring these dynamic changes the DMN shows varying correlation strength with areas related to attention and salience, in a similar fashion to changes observed during sleep and tasks. fcMRI is altered in a variety of disorders (see review (Fox and Greicius 2010)). Current studies are exploring the dynamic changes of the FC to better understand diseases. For example, fcMRI decreases in connectivity observed in patients with schizophrenia could be, in part, explained by the longer time these patients spend on the less connected states at the expenses of the dwell time on the states with cortico-subcortical connections. In patients with Alzheimer’s disease, dwelling time in different modularity configurations were observed using graph theory metrics (Jones, Vemuri et al. 2012). The observation of the fcMRI and its dynamics can provide useful information about behavior and disease characterization. While post-task connectivity changes could be explained as a top-down mechanism, a bottom-up model would explain how a change of neuronal firing at local level would affect areas connected to it. Perturbations to brain areas considered hubs, and alterations in neuromodulatory nuclei might have a particular effect in fcMRI dynamics (Hutchison, Culham et al. 2014). Indeed, to probe the possible mechanisms of these signal, much smaller spatial and temporal scales than achievable with fMRI are required, as it is required to understand the underpinnings of the phenomena at the cellular level. Electrophysiological recordings (He, Snyder et al. 2008), and computational modeling (Rabinovich and Varona 2011) have provided insights into the possible mechanisms. The current theory favors the concept of criticality (Chialvo and Bak 1999). A competitive theory argues that local oscillatory processes are the drivers of the connectivity, not criticality (Li, Bentley et al. 2015). In summary, while the mechanisms behind these dynamic patterns remain to be elucidated, dynamics fcMRI became a new tool to explore brain fluctuations. Dynamic changes have been observed at rest, during tasks and in disorders. Understanding the role of dynamic fcMRI in diseases, and the mechanisms that trigger the dynamic changes has the potential for discovering new biomarkers, and improving treatments.