Project 2: Introduction
Characterizing network-level activity during emotion processing using an individualized parcellation approach
One of the most significant contributions in the field of cognitive neuroscience over the past decade has been the discovery that the human cerebral cortex is organized into a series of discrete functional networks. This groundbreaking discovery was initially made in two independent studies, both published in 2011, that examined the correlation structure of spontaneous fluctuations in activity of the brain at rest, an approach now known broadly as resting-state functional connectivity analysis (Power, 2011; Yeo, 2011). These studies found that the spontaneous activity of the human cortex can be reliably parcellated into a series of distinct networks, each comprising a set of regions distributed across the cortical mantle. The existence of these distributed large-scale cortical networks has now been well replicated (Glasser et al., 2016; Gordon et al., 2016; Schaefer et al., 2018) and further has provided a fundamental framework for interrogating the functional architecture of the human brain. As the field has grown since this initial discovery, work exploring these networks has expanded beyond characterizing cortical organization using functional connectivity analysis to examinations of how activity within these networks are associated with cognition and behavior. By using network parcellations as units of analysis in studies of task-derived activity, instead of standard approaches that use region-level methods–such as whole brain analysis and meta-analytic ROIs, researchers are able to evaluate the functional contributions of the cortical networks (Glasser et al., 2016; Braga et al., 2020; DiNicola et al., 2020). In this study, I adopted such an approach and characterized network-level neural activity in a canonical emotion processing task.
The Rise of Precision Neuroscience. The first wave of research examining the functional architecture of the human brain using resting-state functional connectivity relied on data aggregated across large samples of individuals (aka. “group-averaged” approaches) (Glasser et al., 2016; Gordon et al., 2016; Power et al., 2011; Schaefer et al., 2018; Yeo, et al., 2011). Group-averaging is used to overcome signal-to-noise limitations inherent in the fMRI BOLD signal (Fedorenko, 2021). Though this early work has established a framework for understanding general principles of human brain organization, there is significant variability in both neuroanatomy as well as function across individuals (Steinmetz et al., 1991; Rademacher et al., 1993; Michalka et al., 2015; Mueller et al., 2013). Averaging data across subjects inherently blurs these individual-level differences, and has therefore resulted in a coarse characterization of cortical organization (Salvo et al., 2021). While group-average approaches have been crucial in establishing the presence of now canonical cortical networks (e.g., default network, dorsal attention network, fronto-parietal control network etc.) (Glasser et al., 2016; Gordon et al., 2016; Power et al., 2011; Schaefer et al., 2018; Yeo et al., 2011), recent empirical work suggests such approaches may fundamentally mischaracterize the functional architecture of the human brain. In a series of influential studies, researchers shifted from group-averaged approaches to examining functional architecture in a few deeply sampled individuals (Laumann et al, 2015; Poldrack et al., 2015; Wang et al., 2015; Xu et al., 2016; Braga & Buckner, 2017; Gordon et al., 2017; Gratton et al., 2018; Braga et al., 2019). This work has revealed that the cortical networks exhibit highly idiosyncratic topography across individuals (Braga et al., 2019; Braga & Buckner, 2017; Gordon et al., 2017; Kong et al., 2018; Gordon et al., 2021) and further has led to the discovery that many of the canonical association networks observed at a group level, actually comprise multiple distinct networks that are obscured in group-averaged data (Braga et al., 2019; Braga & Buckner, 2017). These discoveries have paved the way for a second wave of cognitive neuroscience work, now known as “precision neuroscience,” that aims to establish a more nuanced characterization of the functional architecture of the human brain by taking advantage of the precision gained by employing individualized approaches that account for individual differences in neuroanatomy and the topography of functional networks.
Individualized Network Topography and Function. The shift to precision neuroscience approaches has not only provided the opportunity to establish a more accurate characterization of the functional organization of the human brain, but has also created a new landscape for examining the association between the cortical networks and task-derived activity (Fedorenko, 2021; Salvo et al., 2021). Network-activity correspondence is assessed by examining the overlap between network boundaries, defined using functional connectivity approaches in resting-state data, with the activity patterns observed in response to carefully designed tasks within individuals. Researchers have found that individualized network approaches reinforce the correspondence between network boundaries and task-driven activity patterns across an array of tasks including a basic finger tapping tasks, emotional face processing tasks, relational reasoning tasks, language tasks, as well as high-order episodic projection and theory of mind tasks (Mennes et al., 2010; Tavor er 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, in an influential study, DiNicola et al., 2020 demonstrated discrete functional roles of two highly interdigitated networks (coined default network A and default network B), that were previously thought to be a single large-scale network. Previous work detailing the function of the canonical large-scale default network (identified in group-average data) have suggested a role for the network across a broad range of high-order capacities involved in internal mentation including autobiographical memory (Svoboda et al. 2006), future prospection (Schacter et al. 2012), social inference (Iacoboni et al. 2004; Mars et al. 2012; Schurz et al. 2014), and self-referential processing (D’Argembeau et al. 2005; Gusnard and Raichle 2001). Using an individualized parcellation approach DiNicola et al, created subject-specific network boundaries for a few highly sampled individuals. They then examined task-evoked activity within the boundaries of default network A and B during a series of high-order tasks involving episodic projection and theory of mind. They discovered that these two networks support dissociable functional capacities — whereby default network A is preferentially recruited during episodic projection tasks, while default network B is preferentially recruited during theory of mind tasks (DiNicola et al., 2020). Taken together, these key insights suggest that using individualized network boundaries to examine cortical activity reveals details lost in group averaged data and are therefore necessary for accurate characterizations of human brain function.
Individualized Network-Level Characterization of Canonical Emotion Processing Task. One of the earliest pieces of work illustrating that individualized network parcellations improve alignment with task-derived activity relative to group-based boundaries employed a canonical emotion processing task (Chong et al., 2017). Though Chong et al., showed improved correspondence between individualized network boundaries and task activity patterns in this task relative to group averaged parcellations, they did not fully characterize the task in terms of network-level recruitment. This canonical task has been a fundamental mainstay in the field, core to work examining individual differences in emotion processing of threat-relevant information (Hariri et al., 2000, 2002; Dannlowski et al., 2013; McCrory et al., 2011; van Harmelen et al., 2013; Weissman et al., 2020). This task has thus been included in numerous massive data collection efforts including the Human Connectome Project, the UK biobank, as well as the Dunedin Study (Van Essen et al., 2013; Poulton, Moffitt, & Silva, 2015; Bycroft et al., 2018). A vast literature using this and similar tasks has shown that individual differences in emotion processing are associated with 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). This work has primarily focused on the amygdala as the neural target in emotion processing, although numerous meta-analysis examining this and other similar task, additionally implicate a broad set of cortical regions in emotional processing of threat-relevant information, such as anterior insula, precuneus, posterior cingulate, medial prefrontal cortex, middle temporal gryus; inferior, middle, and superior frontal gyrus; as well as the fusiform (Fusar-Poli et al., 2009; Sabatinelli et al., 2011). These findings confirm emotion processing as a human capacity underpinned by widespread cortical engagement. Though region-based approaches, such as meta-analytic analyses and examining average whole brain activity, do show general overlap in regions implicated–most notably the amygdala and the fusiform, there is considerable variability across studies as to which regions are reliably recruited (Fusar-Poli et al., 2009; Sabatinelli et al., 2011). This lack of reliability across studies makes it challenging to clearly characterize the neural substrates that underpin the the processing of emotionally salient face stimuli, which involves numerous cognitive processes including face processing, identity processing, as well as detection of emotional salience (Elliott et al., 2020; Hariri et al., 2002; Somerville et al., 2018). Examining task-derived emotion processing activity in this canonical task at an individualized network-level may provide a unique opportunity to characterize activity within a more reliable and generalizable framework. The development of individualized networks is task independent, making them robust to differences in task design across studies. They thus circumvent reliability issues that plague region-level group-based approaches. Therefore, a network-level characterization of activity in this canonical task could shed light on our understanding of emotion processing by more comprehensively characterizing the cortical networks involved in this complex form of information processing and further provide an additional methodological framework that can be explored in studies that employ this canonical task.
Aim of Present Study. In the present study I aim to characterize the widespread cortical activity observed during a canonical emotion processing task within an individualized network framework. Using methods similar to those used in recent precision neuroscience approaches(e.g., DiNicola et al. 2020), I will examine the average emotion processing activity within each network as defined by individualized parcellations obtained for each participant using resting-state functional connectivity data. I will compare patterns of activity within these individualized network boundaries to a standard group-based region level approach by additionally examining meta-analytically derived regions of interest (ROIs), such as the amygdala and fusiform gyrus (Sabatinelli et al., 2011). I will also include an examination of emotion processing activity within a group-based network parcellation (Yeo et al., 2011), providing the opportunity to examine the difference in functional recruitment observed when employing individualized parcellations as compared to group-based parcellations. In sum, I aim to characterize the cortical activity observed in a canonical emotion processing task using three different boundary-based approaches — an individualized network approach, in contrast to a region-based analysis, as well as a group-average parcellation analysis. Examining network-level activation may provide a more comprehensive characterization of the cortical regions involved in this complex form of information processing than standard approaches that use ROIs focused on specific cortical regions. Furthermore, if I observe consistent functional recruitment at the network level in cortical networks not observed in group-averaged data, that would speak to the importance of taking an individualized approach to defining the network boundaries.