Task-based dynamic functional connectivity: Recent findings and open questions

Highlights

• Functional connectivity reshapes efficiently when switching between rest and task.

• Moment-to-moment FC can predict subsequent perceptual outcomes.

• Task-concurrent dynamic-FC metrics have significant behavioral relevance.

• Analytical and interpretational challenges of task dynamic-FC are discussed.

Abstract

The temporal evolution of functional connectivity (FC) within the confines of individual scans is nowadays often explored with functional neuroimaging. This is particularly true for resting-state; yet, FC-dynamics have also been investigated as subjects engage on numerous tasks. It is these research efforts that constitute the core of this survey. First, empirical observations on how FC differs between task and rest—independent of temporal scale—are reviewed, as they underscore how, despite overall preservation of network topography, the brain's FC does reconfigure in systematic ways to accommodate task demands. Next, reports on the relationships between instantaneous FC and perception/performance in subsequent trials are discussed. Similarly, research where different aspects of task-concurrent FC-dynamics are explored or utilized to predict ongoing mental states are also examined. The manuscript finishes with an incomplete list of challenges that hopefully fuels future work in this vibrant area of neuroscientific research. Overall, this review concludes that task-concurrent FC-dynamics, when properly characterized, are relevant to behavior, and that their translational value holds considerable promise.

Introduction

In neuroscience, functional connectivity (FC) usually refers to the degree of co-variation between spatially distributed signals emanating from the brain and recorded with different functional neuroimaging techniques such as functional magnetic resonance imaging (fMRI (Biswal et al., 1995)), electro-encephalography (EEG (Babiloni et al., 2005)), magneto-encephalography (MEG (Brookes et al., 2011)), functional near infrared spectroscopy (fNIRS (Lu et al., 2010)), and electrocorticography (ECoG (Antony et al., 2013)). FC studies are most commonly conducted under resting conditions (i.e., without any external stimulation or task demands), yet understanding how environmental stimuli and cognitive demands modulate FC is also the subject of rigorous research.

As explained elsewhere in this special issue [Ref Needed], rest-FC is known to be dynamic, with FC patterns evolving in biologically meaningful ways at temporal scales ranging from years—as it is the case with developmental FC changes (Dennis and Thompson, 2014)—to seconds (Chang and Glover, 2010). The same is true when tasks or external stimuli are present. For example, mastering motor skills over the course of weeks is accompanied by increased autonomy of sensorimotor systems and independence from cognitive control hubs (Bassett et al., 2015). At shorter temporal scales, FC patterns computed over tens of seconds contain sufficient information to determine the task with which subjects are actively engaged (Gonzalez-Castillo et al., 2015, Shirer et al., 2012). Yet, whether rest-FC and task-FC dynamics are two manifestations of the same underlying neuronal phenomenon, or the result of distinct ones is still under debate; particularly when referring to short temporal scales (e.g., seconds to minutes).

One interpretation is that short-term rest-FC dynamics are, at least partially, explained by unconstrained ongoing cognition unfolding as subjects are allegedly at rest (Doucet et al., 2011). As such, under this framework, dynamic FC at both rest and task are tightly related from a mechanistic perspective. This view is reinforced by research showing that resting state is characterized by at least seven different cognitive phenotypes—namely, discontinuity of mind, theory of mind, self, planning, sleepiness, comfort and somatic awareness (Diaz et al., 2013)); and that it is not uncommon for subjects to engage in a variety of self-paced tasks, such as inner-speech or memory retrieval, during a rest scan (Hurlburt et al., 2015). Moreover, consciousness is often envisioned as an incessant gush of cognitive processes (William James “stream of consciousness”), which are present during awake rest and hypothesized to manifest as dynamic changes in FC (Barttfeld et al., 2015).

Rest-FC dynamics have also been reported during sleep (Larson-Prior et al., 2009) and anesthesia (Hutchison et al., 2013, Keilholz et al., 2013). Although recent research has shown that the complexity of FC dynamics decreases in proportion to degree of consciousness (Amico et al., 2014, Barttfeld et al., 2015), the fact that some level of dynamic FC remains during such unconscious states has been cited as an argument to reject ongoing cognition as a primary driver of dynamic FC. One alternative hypothesis, as to the origin of resting FC dynamics, is that rather than embodying specific cognitive operations, dynamic FC is a manifestation of the brain continuously exploring an array of available cognitive architectures. And, that such pseudo-random exploration offers advantages in terms of response time and sensitivity to upcoming environmental stimuli (Deco et al., 2013). Computational simulations based on noise-driven exploration of this landscape of possible dynamic states have been able to produce dynamic FC behaviors like those observed empirically (Hansen et al., 2014); therefore, supporting the plausibility of this alternative role of dynamic FC.

Finally, using multivariate kurtosis—a forth order statistic proportional to the sampling variability of the covariance matrix (e.g., the connectivity matrix)—to evaluate the temporal stability of rest-FC, Laumann et al. (2016) argued that rest-FC dynamics are mostly an artefactual consequence of sampling variability and head motion, and that fluctuating sleep states are the only physiological factor leading to dynamic FC behaviors during rest. Similar viewpoints have been expressed by others using simulations and both fMRI (Hindriks et al., 2015) and EEG data (Hlinka and Hadrava, 2015). It is worth mentioning within the context of this review that, in that same study, Laumman et al. found that task performance did indeed modulate multivariate kurtosis (their test of non-stationarity), suggesting that external tasks can in fact alter short-term FC.

In summary, the origin of rest- and task-FC dynamics remains a matter of debate; as does the specific role, if any, that ongoing cognition plays in these phenomena. Yet, as we shall discuss in the remainder of this document, it is well established that task performance modulates FC at different temporal scales, including that of seconds to minutes. In section 2, we review our understanding of how FC, independently of temporal scale, differs between rest and task experiments. As we shall discuss, task performance modifies FC in limited, yet consistent ways. Section 3 focuses on how moment-to-moment estimates of FC can help predict behavioral responses to upcoming events (e.g., auditory stimuli, pain, etc.). This suggests that ongoing intrinsic connectivity can modulate perception and cognitive performance, stressing the importance of “healthy” short-term dynamics for efficient interaction with the environment. Section 4 then examines work where moment-to-moment FC computed during task epochs is used to predict the cognitive processes taking place. In contrast with section 3, which focuses on how pre-stim/task FC state constrains upcoming performance, section 4 discusses how task/stimulation modulates ongoing FC sufficiently as to allow moment-to-moment identification of what subjects are doing. The manuscript then concludes with some thoughts on current challenges and future work needed to better understand how cognitive demands shape dynamic FC in the human brain.