Predicting Empathy From Resting State Brain Connectivity: A Multivariate Approach

predicting trait empathy

Published in Frontiers in Integrative Neuroscience

Abstract: Recent task fMRI studies suggest that individual differences in trait empathy and empathic concern are mediated by patterns of connectivity between self-other resonance and top-down control networks that are stable across task demands. An untested implication of this hypothesis is that these stable patterns of connectivity should be visible even in the absence of empathy tasks. Using machine learning, we demonstrate that patterns of resting-state fMRI connectivity (i.e. the degree of synchronous BOLD activity across multiple cortical areas in the absence of explicit task demands) of resonance and control networks predict trait empathic concern (n = 58). Empathic concern was also predicted by connectivity patterns within the somatomotor network. These findings further support the role of resonance-control network interactions and of somatomotor function in our vicariously driven concern for others. Furthermore, a practical implication of these results is that it is possible to assess empathic predispositions in individuals without needing to perform conventional empathy assessments.

Brief Summary: We used a LASSO cross-validation scheme to asses the degree to which we could predict metrics of empathic behavior by leveraging an individual’s functional connectivity observed at rest within and across cortical networks. Only connectivity across networks stemming from ROIs responsible for resonance and control were able to successfully predict trait-empathy. The somatomotor network was also able to predict individual differences in perspective-taking.

resonance and control networks

Discussion Summary:

In this study, we tested two hypotheses:

(I) We hypothesized that participants’ empathic concern for others would be predicted by resting connectivity between our theory-driven and literature-derived resonance and control networks.

(II) We hypothesized that we could predict subcomponents of participants’ trait empathy from the within- and between-network resting connectivity of canonical resting state networks.

As hypothesized in (I), participants’ levels of empathic concern were predicted by patterns of connectivity within and across the resonance and control networks (when treated as a single network), supporting the hypothesis (put forth in Christov-Moore and Iacoboni, 2016 and supported by Christov-Moore et al., 2017a) that these systems (a) continuously interact in a characteristic fashion observable in the absence of pertinent task demands and (b) this interaction is a likely neural substrate of empathic concern for others. Our findings (along with the previous work that prompted this study) support a dynamic, integrated view of empathic function, based on complex patterns of interaction between resonance and control systems rather than simply a univariate measure of overall connectivity. Indeed, numerous studies have reported task-related changes in connectivity between resonance and control networks during passive observation of emotions or pain (Christov-Moore and Iacoboni, 2016), reciprocal imitation (Sperduti et al., 2014), tests of empathic accuracy (Zaki et al., 2009), and comprehension of others’ emotions (Spunt and Lieberman, 2013). Interestingly, Raz et al. (2014) found evidence for complex, context-dependent interactions between “simulation” and “theory-of-mind” networks (largely corresponding to what are defined here as resonance and control networks) during empathic experience (observing films depicting personal loss). This multivariate approach may help reconcile findings supporting an integrated view with activation (e.g. Van Overwalle and Baetens, 2009) or lesion studies that suggest dissociated systems (e.g. Shamay-Tsoory et al., 2009): Lesions (transient/induced or physical) may simply be altering a crucial node for a specific integrated network outcome, just as a hand injury may affect the ability to catch a ball more than a back injury, though catching-like activities typically rely on hands, arms, and the core operating in unison. Indeed, the complexity of these interactions may be an obstacle to their efficient detection by standard activation or univariate connectivity methods. By employing flexible machine learning methods that make few a priori assumptions about the patterns of intrinsic connectivity underlying individual differences, we may achieve a more comprehensive multivariate view of the possible network-level patterns of neural interaction that give rise to individual differences in empathic function. It is common within cognitive neuroscience to theorize first about psychological processes and then investigate the neural correlates of such processes. However, in an exceedingly complex system such as the brain, much could be gained by approaching the problem from the opposite direction, by investigating how psychological processes emerge from brain organization (Fox and Friston, 2012).

As for (II), empathic concern was predicted by the within-network connectivity of the somatomotor network. This result further supports an embodied, somatomotor foundation for our concern for others’ welfare, in line with numerous findings relating vicarious somatosensory activation to multiple forms of prosocial behavior (non-strategic generosity in economic games: Christov-Moore and Iacoboni, 2016; harm aversion in moral dilemmas: Christov-Moore et al., 2017b; donations to reduce pain in another: Gallo et al., 2018; helping behavior: Hein et al., 2011Masten et al., 2011; charitable donations: Ma et al., 2011). This also agrees with our recent finding that inferior premotor activation during observation of pain in others was predictive of participants’ later tendency to avoid inflicting harm in hypothetical moral dilemmas (Christov-Moore et al., 2017b). A major proposed subcomponent of empathy is fantasizing (Davis, 1983Clay and Iacoboni, 2011), our ability to take the perspective of absent or fictional characters and become correspondingly invested in their welfare. Perhaps we implicitly construct internal models of others (present or implied/hypothetical) using perceptual, affective, and motor experiences we associate with past experience, framed by others’ intentions, moral character, group affiliation, etc. This embodied model of the “other” and its contextual framing would likely be represented by interactions between resonance and control processes, thus shaping the relative utility of their welfare (Bechara and Damasio, 2005), and hence the positive and negative reward values assigned to the outcomes of decisions that can affect them (Fehr and Camerer, 2007).

A clinical avenue suggested by this study is the potential ability to predict empathic functioning in populations that might have difficulty performing empathy tasks or filling out questionnaires, either due to being less cooperative or less cognitively able, e.g. in populations such as those with schizophrenia, low functioning autism, intellectual disabilities, or traumatic brain injury. Individuals in these groups might have, in principle, intact inherent capability for normal-range empathy that could be impeded by other limitations such as verbal or non-verbal communication (autism) or disorganized thought processes (schizophrenia); thus it would help us know what reasonable outcomes in terms of social and interpersonal functioning could be expected to result from therapies that help with training to rehabilitate or improve empathy, ultimately in the interest of enhancing social competence and social cognition. Indeed, it may be pertinent to include measures of empathic function along with standardized, multisite resting-state scan protocols (like the Human Connectome Project), paving the way for a massive data-driven approach to produce models that can predict empathic function from the resting brain in many different populations.

%d bloggers like this: