Beyond regional specialization — using individualized networks to characterize cortical brain function during emotion processing
The human brain is the world’s greatest translator taking basic physical properties such as light and sound and transforming them into the most complex of human capacities. This translation arises via billions of nanoscopic transmissions that flow across a vast labyrinth of neurons. Once opaque, the immense neural maze of the brain is now recognized as being organized into coordinated yet discrete specialized systems each composed of distributed regions throughout the brain. It is the coordination within and between these systems that gives rise to our fundamental human capacities. Something that feels as simple to us as recognizing fear in another’s face requires not only specialized regions, but the integration across many specialized systems to seamlessly produce the conclusion: “Nessa is scared!” Understanding how the functional systems in the brain are organized and give rise to the fundamental human capacities, such as emotional face processing, is core to the field of cognitive neuroscience.
Cortical Brain Organization. In 2011, two independent research groups simultaneously published findings revealing that the human cerebral cortex can be reliably segregated into a set of functional networks by examining the correlation structure of spontaneous activity across with whole brain at rest, a method now known as functional connectivity analysis (Power et al., 2011; Yeo et al., 2011). This groundbreaking discovery confirmed earlier invasive studies in non-human primates, which suggested the cortical mantle is organized into distinct functional systems (Goldman-Rakic, 1988; Mesulam, 1990; Felleman & Van Essen, 1991). The discovery of functional connectivity analysis as a non-invasive technique that can reveal human brain organization was groundbreaking, as anatomical approaches alone had failed to illuminate meaningful organization in the higher-order association cortex (Frost & Goebel, 2011; Tahmasebi et al., 2012; Vazquez-Rodriguez et al., 2019; Fedorenko, 2020). Since this initial discovery, the existence of the cortical networks has proven robust across various analytical methods, including clustering, meta-analytic connectivity, edge detection, and multi-modal methods (Lashkari et al., 2010; Yeo et al., 2011; Eickhoff et al., 2011; Gordon et al., 2016; Laumann et al., 2015; Glasser et al., 2016; Schaefer et al., 2018; Baldassano et al., 2015; Blumensath et al., 2013; Smith et al., 2009). Furthermore, the identified networks correspond to known functional systems that have been studied in isolation, such as the visual and motor systems (Vincent et al., 2007; Patel et al., 2015), attentional systems (Fox et al., 2006; Vincent et al., 2008; Beckmann et al., 2005) and newly discovered higher-order systems, including a default system (Shulman et al., 1997; Buckner, 2008). In sum, the advent of functional connectivity analysis and the discovery of the large-scale cortical networks has provided a global framework for understanding the inherent functional organization of the human brain.
Measuring Individual Differences in Functional Connectivity. Since the initial discovery of the cortical networks, there has been a proliferation of research employing functional connectivity approaches to explore the organization and function of the brain. Work characterizing the topology of the networks has lead to the development of network parcellations (ie. network atlases that map the topology of the cortical networks) (Lashkari et al., 2010; Yeo et al., 2011; Eickhoff et al., 2011; Gordon et al., 2016; Laumann et al., 2015; Glasser et al., 2016; Schaefer et al., 2018; Baldassano et al., 2015). These network parcellations have been adopted by clinical and developmental researchers and used as to study how network specific properties such as average within-network connectivity (Fan et al., 2019; Karcher et al., 2019; Lydon-Staley et al., 2019; Yu et al., 2019) and graph theory measures such as modularity and global efficiency (Bullmore & Sporns, 2009; Power et al., 2010) vary as a function of factors such as age (Alarcón et al., 2018; Jalbrzikowski et al., 2019; Lopez et al., 2019; Satterthwaite et al., 2013; Sylvester et al., 2018) or the presence of psychopathology (Fan et al., 2019; Lydon-Staley et al., 2019; Reggente et al., 2018; Yu et al., 2019). This methodological approach to examining individual differences in brain organization has led to a significant amount of work suggesting the network function exhibits meaningful clinical and developmental differences. Though the use of network parcellations has become commonplace, the use of these parcellations, particularly in developmental and clinical samples, hinges on four fundamental assumptions: 1) the various parcellations are equally able to recover the networks of interest; 2) adult-derived parcellations well represent the networks in children’s brains; 3) network properties, such as within-network connectivity, are reliably measured across the different parcellations; and 4) parcellation selection does not impact the results with regard to individual differences in network properties. In project 1, I examined the fundamental assumptions that go overlooked when a priori parcellations are employed in clinical and developmental research (Bryce et al., 2021). I found that network properties such as average within network connectivity showed notable variability and poor consistency across various parcellations examined. Furthermore, parcellation selection meaningfully impacted the magnitude and significance of associations between functional connectivity and age, poverty, and cognitive function. These findings call into question work which depends on group-averaged a priori parcellations for network identification and suggest that alternative methods for the identification of cortical networks, that do not hinge on these primary assumptions, may be necessary.
A Shift to Precision Neuroscience. Until recently, functional connectivity research has relied on aggregating data across large samples to overcome fMRI signal limitations (Glasser et al., 2016; Gordon et al., 2016; Power et al., 2011; Schaefer et al., 2018; Yeo et al., 2011). Indeed, this approach was used to develop the network parcellations described above. While this work led to a fundamental framework for understanding brain organization,group-averaging methods obscure known variability in anatomy and function across individuals (Steinmetz et al., 1991; Rademacher et al., 1993; Michalka et al., 2015; Mueller et al., 2013). The blurring of data across idiosyncrasies has resulted in a coarse characterization of the cortex that may have resulted in a fundamental mischaracterization of functional organization (Salvo et al., 2021). Indeed, recent work that explored brain organization of a few deeply sampled individuals has revealed that the cortical networks exhibit highly idiosyncratic topographies across individuals and, further, the large-scale canonical networksobserved in group-averaged data actually comprise multiple distinct networks (Braga et al., 2019; Braga & Buckner, 2017; Gordon et al., 2017; Kong et al., 2018; Gordon et al., 2021). These findings have led to the rise of precision neuroscience, which employs individualized approaches to account for neuroanatomical differences, providing a more nuanced framework for understanding brain organization and function (Fedorenko, 2021; Salvo et al., 2021). Improved alignment between individualized network boundaries, relative to group-averaged networks, and task-evoked activity has been found across various tasks (Mennes et al., 2010; Tavor et al., 2016; Peer et al., 2015; Chong et al., 2017; Gordon et al., 2017; Glasser et al., 2016; Braga et al., 2020; DiNicola et al., 2020; Salvo et al., 2021). Furthermore, influential studies, such as DiNicola et al. (2020), have demonstrated discrete functional roles within networks identified in the individual that were previously thought to comprise a single large scale cortical network. These findings suggest that individualized network boundaries reveal critical functional principals that are lost in group-averaged data and provide an essential new framework for characterizing human brain function.
Canonical Emotion Processing Task. One of the tasks that has been used to illustrate improved correspondence between network boundaries and task-evoked activity when precision neuroscience approaches are used is a canonical emotion processing task (Hariri et al., 2000, 2002) that has been included in task batteries of massive data collection efforts such as the Human Connectome Project, the UK biobank, and the Dunedin Study (Van Essen et al., 2013; Poulton, Moffitt, & Silva, 2015; Bycroft et al., 2018). There has been significant investment into this task, as a vast literature has found associations between amygdala reactivity, the target neural substrate in this task, and factors such as psychopathology and experiences of life stress (Dannlowski et al., 2013; McCrory et al., 2011; van Harmelen et al., 2013; Weissman et al., 2020; Yurgekun-Todd et al., 2000; Thomas et al., 2000; Etkin & Wager, 2007; Whalen et al., 2002; Groenewold et al., 2013; Sicorello et al., 2020; Simmons et al., 2011). These findings have led to the widely shared hypothesis that amygdala reactivity might serve as a biomarker of risk for psychopathology (Herman and Cullinan, 1997; Kim et al., 2003; Murty et al., 2010; Pessoa and Ungerleider, 2004; Swartz et al., 2015). However, recent works examining the reliability of this task have revealed poor test-retest reliability of activity within the amygdala in this task, calling into question the validity of using the task to draw conclusion about individual differences in emotion processing (Elliott et al., 2020; Fröhner et al., 2019; Nord et al., 2017; Flournoy et al., 2023). Examining possible alternative approaches to characterizing activity could breathe new life into this widely used canonical task.
The emotion processing task in question is designed to elicit a robust response by employing a broad contrast between two simple task conditions, an emotionally salient condition of threatening faces, known to activate the amygdala, versus geometric shapes (Hariri et al., 2000, 2002). Given this broad contrast, widespread cortical activity is observed in this task alongside the expected amygdala response. This widespread cortical activity during emotion processing has been confirmed in meta-analytic studies of emotion processing that include this task (Fusar-Poli et al., 2009; Sabatinelli et al., 2011). While the primary target in this task is the amygdala, using an individualized network parcellation to examine the widespread cortical activity in this task could provide a new framework for characterizing the observed activity. In project 2, I explore a network-level characterization of neural activity during this canonical emotion processing task using an individualized parcellation approach. This study adopts individualized methods to examine network-level activity during emotion processing, focusing on individualized network-level recruitment as compared to standard region-level and group-averaged approaches. If individualized networks exhibit meaningful activity during this canonical task, it is possible that they may serve as alternative targets in studies of individual differences in emotion processing.
As mentioned, the activity in the amygdala (the task target) has poor test-retest reliability (Elliott et al., 2020; Fröhner et al., 2019; Nord et al., 2017; Flournoy et al., 2023). Shifting to an individualized network framework circumvents many of the sources of noise that plague the amygdala, and so may serve to improve the test-retest reliability of activity in this task. In project 3, I examine whether a shift to an individualized network-level framework improves the test-retest reliability of activity in this canonical task. If I do see improved test-retest reliability of activity within the individualized network relative to the amygdala, it could reposition this task as viable for the study of individual differences in emotion processing. Strong test-retest reliability of task-evoked activity is crucial for studies of individual differences, as it establishes the upper threshold for detecting valid associations. Following a test-retest reliability analysis, I will further examine whether network-level function is associated with individual differences in stress and psychopathology. Recent work has illustrated that emotion processing activity exhibits high within-person variability, which may be a major factor in the poor test-retest reliability observed in this task. Furthermore, they found that this variability can be explained by within-subject fluctuations in stress, mood and sleep (Flournoy, et al., 2023). Therefore, I will additionally examine associations between network-level function and within-person fluctuations in stress and psychopathology. If I observe improved test-retest reliability using individualized networks and also identify associations with factors like stress and psychopathology (either between or within-person), then these individualized methods could provide a novel framework for studying emotion processing and its connections to mental health and behavior.
Study Description & Methods. As relevant methods come up throughout the following pages, direct links to methodological details are provided.
FIRST UP: PROJECT 1 — Evaluating the Impact of Parcellation Selection on Connectivity Results
JUMP TO: PROJECT 2 — Characterizing network-level activity during a canonical emotion processing task
JUMP TO: PROJECT 3 — Exploring the potential use of individualized network approaches in studies of individual differences in emotion processing