Just a thought: How mind-wandering is represented in dynamic brain connectivity

Highlights

• Dynamic functional connectivity (dFC) reflects intrinsic functions and mind-wandering.

• Time-varying communication in default mode and other networks reflects mind-wandering.

• Mind-wandering involves several processes, each represented distinctly in dFC.

• Network-level dynamics better reflect mind-wandering than localized activity.

• Intrinsic neural activity must be accounted for in the study of mind-wandering.

Abstract

The neuroscience of mind-wandering has begun to flourish, with roles of brain regions and networks being defined for various components of spontaneous thought. However, most of brain activity does not represent immediately occurring thoughts. Instead, spontaneous, organized network activity largely reflects “intrinsic” functions that are unrelated to the current experience. There remains no consensus on how brain networks represent mind-wandering in parallel to functioning in other ongoing, predominantly unconscious processes. Commonly, in network analysis of functional neuroimaging data, functional connectivity (FC; correlated time series) between remote brain regions is considered over several minutes or longer. In contrast, dynamic functional connectivity (dFC) is a new, promising approach to characterizing spontaneous changes in neural network communication on the faster time-scale at which intra-individual fluctuations in thought contents may occur. Here I describe how a potential relationship between mind-wandering and FC has traditionally been considered in the literature, and I review methods and results pertaining to the study of the dFC-mind-wandering relationship. While acknowledging challenges to the dFC approach and to behaviorally capturing fluctuations in inner experiences, I describe a framework for describing spontaneous thoughts in terms of brain-network activity patterns that are comprised of connections weighted by time-varying relevance to conscious and unconscious processing. This perspective suggests preferential roles of certain anatomical communication avenues (e.g., via the default mode network) in mind-wandering, while also implying that a region's connectivity fluctuates over time in its immediate degree of relevance to conscious contents, ultimately allowing novelty and diversity of thought.

Introduction

“What a peculiar privilege has this little agitation of the brain which we call 'thought', that we must thus make it the model of the whole universe?” David Hume, 1779.

What constitutes a unitary ‘thought’? While there is no clear consensus, thought may be described on different temporal scales, ranging from near-instantaneous experiences to seconds-long streams of connected elements (discussed at length by James, 1890). A hallmark quality of much of our thought is spontaneity— fluctuations of content in the absence of conscious control and in independence from the immediate sensory environment.

Mind-wandering is a form of spontaneous thought that is relatively unconstrained and undirected by cognitive control (Christoff et al., 2016). The neuroscience of mind-wandering has recently begun to flourish, with roles of brain regions and networks being defined for various components of thought. However, often underappreciated is that most spontaneous brain activity does not represent immediately occurring thoughts. The likely miniscule fraction of neural activity that has “peculiar privilege”— as put by Hume— in representing current conscious experience, can only be fully understood when considered in tandem with other ongoing functions of the brain.

Brain activity is in constant fluctuation during all stages and states of life. While such ongoing fluctuations were acknowledged since some of the first electrophysiological measurements of neural activity (Caton, 1875, Berger, 1929), only recently has the study of spontaneous activity in the human brain become a mainstream topic of interest (reviewed by Fox and Raichle, 2007). Functional neuroimaging has provided a non-invasive window into brain activity, revealing that each part of the brain shows persistent functional connectivity (FC; correlated time series) with a wider network of regions, prompting increased emphasis on understanding the brain as a dynamic, complex network (Sporns, 2012).

“Resting state” fMRI, in which subjects typically lie awake in a task-free state and let the mind wander, has become a common and effective approach for identifying FC networks. In conventional analyses, single correlations of several minutes-long time series are performed. While suitable for identifying organized brain-network activity at a global level, a key component of mind-wandering is wandering itself (reviewed by Christoff et al., 2016)— processes mediating changes in conscious contents on second- or sub-second-level time-scales. Thus, methods and analyses that are sensitive to time-resolved brain-network fluctuations are needed to unpack the neural mechanisms of mind-wandering.

Dynamic functional connectivity (dFC) analysis (Hutchison et al., 2013a, Calhoun et al., 2014) is a new approach to characterizing spontaneous changes in network-level communication on time-scales that are close to the speed at which fluctuations in conscious contents may occur. Here I discuss how mind-wandering can be considered as a component of dFC. To provide context, I first review how a potential relationship between mind-wandering and conventional (static, or temporally-extended) resting-state FC has traditionally been considered in the literature. I then give a similar overview of initial dFC studies, and I explain methods that can be used to capture mind-wandering as well as studies that employed those approaches to directly investigate the dFC-mind-wandering relationship. Finally, I describe a framework for describing spontaneous thoughts in terms of dFC patterns and point to the pertinent future challenges.

Even though mind-wandering is highly common during task-free states, the presence of resting-state FC over time-scales of several minutes or longer (so-called “static FC”) was initially largely dismissed as being a reflection of mind-wandering. In a seminal, early fMRI study showing resting-state FC between right and left human sensorimotor cortex, Biswal et al. (1995) concluded “We believe that the functional connectivities demonstrated in the motor cortex are a general phenomena and not due to ‘imagined’ motor tasks.” In subsequent, early resting-state studies of other brain systems, involving both sensory and association cortices, authors similarly suggested no excepted link with mind-wandering or did not discuss the possible influence on FC (Lowe et al., 1998, Xiong et al., 1999, Cordes et al., 2000, Hampson et al., 2002). Thus, in contrast to earlier task-based neuroimaging studies of FC which had suggested that experimental conditions and cognitive state influence brain network configuration (Friston et al., 1993), resting-state FC was suggested to reflect state-independent spontaneous communication between brain regions mediated by direct or indirect anatomical projections (Xiong et al., 1999, Cordes et al., 2000).

In 2003, Greicius et al. adopted a more nuanced interpretation of resting-state FC that they reported within a brain system known as the default mode network (DMN). This network, including the posterior cingulate cortex (PCC), medial prefrontal cortex (mPFC) and lateral parietal regions, exhibited spatial overlap with areas that were shown to be more active during task-free states than during a wide range of states involving goal-directed task performance (Shulman et al., 1997, Raichle et al., 2001). Given the common occurrence of mind-wandering during task-free states, Greicius et al. suggested that “… the regions linked to the PCC in the default mode network are well-suited to support the mental processes presumed to be ongoing during the resting state,” whereas they suggested that other within-DMN connections may reflect “… more basic (possibly subconscious) processing necessary for calibrating affective and autonomic states.” This viewpoint therefore implies anatomical heterogeneity— even within a network— in the relationship between resting-state FC and ongoing conscious contents.

As the field of resting-state fMRI rapidly grew, whole-brain static FC was studied and compared across a wide range of contexts. Studies of altered or absent consciousness during sleep (Fukunaga et al., 2006, Horovitz et al., 2009), general anesthesia (Peltier et al., 2005, Vincent et al., 2007), sedation (Greicius et al., 2008), and vegetative state (Guldenmund et al., 2012), revealed that patterns of network FC, including that within the DMN, are modulated but are still often strongly persistent across states. Moreover, the topography of whole-brain static FC patterns during the resting state was found to be highly similar when a subject was performing a wide range of cognitive tasks (with some task-specific changes in FC strength) (Greicius and Menon, 2004, Fransson, 2006, Cole et al., 2014, Krienen et al., 2014, Bellana et al., 2017). Despite limitations of fMRI, intracranial human electrophysiology has confirmed that organized FC persists across rest, task and sleep states (He et al., 2008, Nir et al., 2008, Keller et al., 2013, Ramot et al., 2013, Foster et al., 2015, Foster et al., 2016, Hacker et al., 2017).

Taken together with additional supportive evidence (detailed by Fox and Raichle, 2007, Buckner et al., 2013), these findings establish that static FC predominantly reflects an ‘intrinsic’ functional architecture that is independent of immediately occurring spontaneous thoughts. Intrinsically generated activity, which includes spontaneous neuronal firing, excitability fluctuations and neurochemical processes, results in FC when there is coordination of such processes between brain sites that is not attributable to current inputs or outputs (reviewed by Fox and Raichle, 2007, Deco and Kringelbach, 2016). Interestingly, this notion of intrinsic FC is concordant with the initial interpretation of resting-state FC as a “general” phenomenon (Biswal et al., 1995). The interpretation can also be reconciled with earlier notions of brain function as being predominantly intrinsic, rather than reflexive to demands of the sensory environment (reviewed by Llinas, 1988, Buzsaki, 2006, Raichle, 2010), which is supported by the fact that the brain is only ∼2% of body mass but persistently consumes ∼20% of the body's energy. Intrinsic FC could serve a number of ‘housekeeping’ roles that involve unconscious processing, for instance the maintenance of structural integrity of connections, homeostatic functions, and information processing for interpreting and preparing to respond to incoming sensory input (reviewed by Fox and Raichle, 2007).

Given that static FC largely reflects intrinsic function, the exclusive focus on spontaneous cognition in relation to resting brain activity has been subject to critiques (Hugdahl et al., 2015, Raichle, 2015b). However, the notions of intrinsic activity and of conscious content reflected in resting FC are not incompatible with one another. While static FC patterns can reliably identify a stable, intrinsic functional architecture that is relatively reproducible within an individual (Finn et al., 2015, Laumann et al., 2015, Wang et al., 2015), behavioral interventions or training can result in subtle but significant FC changes (Waites et al., 2005, Lewis et al., 2009, Tambini et al., 2010, Kucyi et al., 2016b). Collectively, the evidence suggests that resting, static FC is a combined measure of an individual's trait-like characteristics with some degree of influence of current behavioral state.

Thus, for a researcher interested in the neural basis of mind-wandering, intrinsic activity can be considered as a large ‘noise’ component, whereas activity reflecting ongoing conscious contents can be considered as the (relatively weaker) signal. Accordingly, Fransson (2006) proposed the partitioning of resting-state FC (particularly in the context of the DMN), into two “conceptual layers”: one reflecting conscious and another reflecting unconscious (i.e., intrinsic) processing. The idea of conceptual layers is in line with the notion of regional heterogeneity relevant to ongoing conscious contents (Greicius et al., 2003) but also implies that the communication between a single pair of regions can subserve both intrinsic activity and ongoing mind-wandering, each to varying degrees at a given moment.

The state-independent persistence of intrinsic activity has implications for interpreting correlations that have been described of resting-state FC with post-scan individual differences in self-reported thoughts involving the past and future (Andrews-Hanna et al., 2010), social cognition (Marchetti et al., 2015), visual and auditory imagery (Doucet et al., 2012, Gorgolewski et al., 2014, Chou et al., 2017) and abstract experiences such as ‘discontinuity of mind’ (Stoffers et al., 2015). Such studies reveal how different within- and between-network connections are not each equally relevant to specific aspects of spontaneous thought. However, due to cross-sectional design, the relative influence of each FC conceptual layer cannot be easily inferred. For example, an individual's general propensity to engage in thoughts involving social cognition could largely be reflected in intrinsic FC, or specific thoughts about others that occurred throughout the resting-state scan could be reflected in changes of FC (i.e., trait cannot be dissociated from state). To gain insight into network-level activity underlying fluctuations of thought, changes in FC must be directly investigated within single individuals and on time-scales that are much faster than durations of several minutes.

Communication between brain regions fluctuates on multiple time-scales. While static FC studies inherently assume temporal stationarity over minutes, FC fluctuations on the second- and subsecond-level are of more interest when investigating intra-individual fluctuations in subjective experience. In typical dFC analysis of fMRI data, FC over the course of brain activity recordings lasting at least several minutes is extracted from different selected, shorter time-windows (sometimes with overlap among windows). A time course of fluctuations in FC can then be obtained for any given pair of regions [see Preti et al. (2016) for a recent summary of additional dFC methods].

The term “dFC” is generally reserved in the field for referring to spontaneous changes in FC (reviewed by Hutchison et al., 2013a). Thus, dFC analyses investigate distinct aspects of FC compared to those aspects studied in the context of actively driven tasks or to experimentally controlled cognitive states (“psychophysiological interaction”) (Friston et al., 1997). Dynamic FC studies usually involve an awake resting state, but time-varying FC during fluctuations in behavior and experience that are spontaneous can also be considered as dFC.

Initial dFC-fMRI studies in humans provided important insights into the temporal properties underlying commonly observed static FC phenomena. Using dFC analyses, Chang and Glover (2010) showed that between-network anticorrelations may be weaker than typical within-network positive correlations because of considerable variability over time in inter-network interactions. Moreover, others showed that whole-brain resting-state FC configurations fluctuate over time but that common network ‘states’ are intermittently revisited (Handwerker et al., 2012, Liu and Duyn, 2013, Allen et al., 2014).

The initial dFC-fMRI studies involved temporal windows on the order of tens of seconds, and some of the time-varying FC could be due to nuisance factors such as non-neural noise, in-scanner head motion, or limitations of analyses (Handwerker et al., 2012, Hindriks et al., 2015, Laumann et al., 2016). However, electrophysiological recordings (reviewed by Keilholz, 2014) and relationships with ongoing cognitive function (see Cohen, 2018) suggest a neural basis of BOLD dFC. Magnetoencephalography studies have also revealed spontaneous fluctuations of FC on the scale of milliseconds (de Pasquale et al., 2012, Baker et al., 2014). Thus, static FC, summarized over minutes, can be modeled and predicted from much faster variations in stability or flexibility of temporally dynamic network interactions (reviewed by Hutchison and Morton, 2016, Breakspear, 2017, Cabral et al., 2017).

A critical question arises from knowledge that dFC is ubiquitous: to what degree do spontaneous changes in network states reflect intrinsic activity, and to what degree do such changes reflect an individual's current behavioral state (e.g., mind-wandering)? While initial dFC studies did not include measures of spontaneous behavioral output, critical insights were gained from studies of dFC during unconscious states. Hutchison et al. (2013b) showed that both awake humans and anesthetized monkeys displayed strong fluctuations in BOLD FC of homologous (oculomotor) networks. This study demonstrated that there is a significant, intrinsic component of dFC—as anticipated by earlier static FC studies— and ruled out the possibility that dFC is solely explained by mind-wandering. However, subsequent studies revealed that dFC during wakefulness and under general anesthesia are not equivalent (Hutchison et al., 2014, Barttfeld et al., 2015). For example, dFC during wakefulness, compared to unconsciousness, was shown to be comprised of a richer repertoire of whole-brain network states (Barttfeld et al., 2015). In concordance, an increased repertoire of dFC states was found during a psilocybin-induced psychedelic experience that involves decreased constraints on ongoing thoughts (Tagliazucchi et al., 2014).

Taken together, while studies of altered consciousness indicate that dFC contains a strong intrinsic component, the state of full consciousness may have a ‘signature’ mode of network dynamics. Studies of sustained conscious states, however, do not give direct insight into the neural representation of ongoing changes in conscious contents. Mind-wandering that occurs at rest involves complex sequences of quickly changing inner experiences that cannot be easily inferred, and so different study approaches are needed.