The Anterior Cingulate Gyrus and Social Cognition: Tracking the Motivation of Others

Abstract

The anterior cingulate cortex (ACC) is implicated in a broad range of behaviors and cognitive processes, but it has been unclear what contribution, if any, the ACC makes to social behavior. We argue that anatomical and functional evidence suggests that a specific sub-region of ACC-in the gyrus (ACCg)-plays a crucial role in processing social information. We propose that the computational properties of the ACCg support a contribution to social cognition by estimating how motivated other individuals are and dynamically updating those estimates when further evidence suggests they have been erroneous. Notably this model, based on vicarious motivation and error processing, provides a unified account of neurophysiological and neuroimaging evidence that the ACCg is sensitive to costs, benefits, and errors during social interactions. Furthermore, it makes specific, testable predictions about a key mechanism that may underpin variability in socio-cognitive abilities in health and disease.

Figure 1

Anatomy of Cingulate Cortex (A) Location of the different regions of the cingulate cortex in the macaque (left) and human (right) brain. Regions in blue comprise the ACC gyrus, red lie in the sulcus (although they may also lie within the adjacent paracingulate gyrus and paracingulate sulcus when they are present). Note that these representative drawings are based on a composite of cytoarchitectonic atlases of the cingulate. The border between areas 24a/b has no gross anatomical landmark and also varies considerably between individuals (Vogt, 2009). Regions in green correspond to those typically referred to as dmPFC. Regions in dark blue are sometimes referred to as part of the vmPFC. Image taken from Rushworth et al. (2004). (B) Connectivity-based parcellation of the medial prefrontal cortex including ACCg (Neubert et al., 2015). The ACC gyrus regions 24a/b and 32pl are shown in light green and maroon, respectively. A distinction between 24a/b and 24a’/b, proposed on the basis of cytoarchitectonic criteria by Vogt (2009) is not always recognized. It was not possible to detect reliable differences between the connectivity profiles of these regions. The green box shows the resting connectivity strength of area 24a/b to other brain areas in both the human (top left: lateral surface; bottom left: medial wall) and macaque (top right: lateral surface; bottom right: medial wall) brain. The central spider plot shows the relative strength of connections between 24a/b and other brain areas in humans (red) and macaques (blue). The maroon box shows the same information for area 32pl. (C) A representative schematic of the connections of the ACCg. The arrows highlight the key systems the ACCg connects to and their putative roles in social cognition. The ACCg (areas 24a/b and 32pl) is connected to a broad set of regions engaged in reward processing, decision making, and social information processing, making it well placed to form part of a distributed network engaged in computing information about the reward-based behavior of others. Abbreviations: VTA, ventral tegmental area; hypoth, hypothalamus; hippoc, hippocampus; amygd/amy, amygdala; pallid, pallidum; Caud, caudate; ventrStr, ventral striatum; temPol, temporal pole; rsplC, retrosplenial cortex; PCC, posterior cingulate cortex; pIPL, posterior inferior parietal lobule; pIPS, posterior intraparietal sulcus; aIPS, anterior intraparietal sulcus; ParOP, parietal operculum; S1, primary somatosensory cortex; M1, primary motor cortex; SMA, supplementary motor area; stria, striatum.

Figure 2

Neurophysiological Recordings from Macaque ACCs, ACCg and OFC as Monkeys Performed a Modified Dictator Game The recipients of the reward could be self, other monkey, or neither (Chang et al., 2013a). (A) Left: recording sites on the medial wall, represented on a schematic of the different zones of the ACC. ACCg recordings were taken from a region that overlaps with both areas 24a/b and 32. Right: recording regions on a coronal plane of an MR image taken from Rushworth et al. (2004). (B) Example neurons from the ACCs (top) and ACCg (bottom). Shown are spiking activity profiles after monkeys made a reward allocation decision (aligned to Choice, left) and after the outcome was actually received (aligned to Reward, right). The ACCs neuron responded to “other” and “neither” after a choice and at the time the reward was delivered. By contrast, the ACCg neuron responded only to other reward and not to self or neither both at the time of choice and the receipt of the reward. (C) Proportion of neurons out of all neurons with a significant modulation to any decision or reward outcome that showed one of three potential reference frames. “Self- referenced” (red) neurons responded to rewarding outcomes referenced to self (whether self-received a reward or not) but did not distinguish between other or neither. These neurons could signal either received (predominantly in OFC) or foregone (predominantly in ACCs) reward for self. In blue are neurons that signaled reward for other only, and in purple are neurons that signaled reward for both self and other in a similar manner (but not to neither reward). The ACCg showed a significantly higher proportion of other-referenced and both-referenced neurons than either the ACCs or OFC, in which the majority of neurons were self-referenced.

Figure 3

ACCg Responses to Other-Oriented Information in fMRI Studies (A) Results from fMRI studies in humans that have compared the processing of similar information for self and other. In these studies, activity was present for others’ high versus low reward probability (Lockwood et al., 2015; top left), the net-value (reward-effort) of another’s actions but not one’s own (Apps and Ramnani, 2014; bottom left), responses to others’ high pain more than low pain (from the meta-analysis of Lamm et al., 2011; top right), and others’ prediction error (Apps et al., 2015a). The central panel shows the locations of the ACC gyrus (blue) and ACC sulcus (red). Strikingly, these results all appear to fall in the same region the ACCg. (B) The response of the ACCg to others’ prediction errors when the subject monitored another’s actions. The graph shows the peristimulus time histogram plot of the BOLD response following trials in which the other person would have a positive PE (light green triangles) or a negative PE (dark green circles). (C) Response in the ACCg at the time of the outcome of decisions during a strategic social interaction task covarying with a PE updating beliefs about how the valuations of another will change (Zhu et al., 2012). The right panel shows that the extent to which this region signaled a PE (i.e., the parameter estimates for the covariation with the BOLD response) correlated with the degree to which subjects’ decisions were influenced by the other person. (D) Activity in the ACCg (red) to the outcomes of another’s advice or the outcomes of one’s own decisions (green) during a social interaction task in which a subject learned either from their own outcomes but also received advice from another. The right panel shows that the variability in the ACCg response at the time of the outcome of each action to how “volatile” the advice of another was, correlated with how influenced subjects’ behavior was by the advice of the other person.

Figure 4

Variability and Pathology in ACCg Activity to Other-Oriented Information (A) Resting-state activity in the macaque ACCg (left) correlates with social network size (right; x axis: size of social group, y axis: strength of resting-state connectivity between ACCg and TPJ/STS) (Sallet et al., 2011). (B) The extent to which the ACCg (left) exclusively processes the expected value of others’ reward correlates with self-reported levels of emotion contagion (a component of empathy). The graph to the right shows the extent to which the ACCg signaled an interaction between probability of reward and agent identity (self versus other). In this graph, the interaction term (BOLD response) correlated with emotion contagion (trait empathy, x axis). Higher levels of empathy were related to greater specialization of the ACCg for processing others’ reward and not subjects’ own reward (Lockwood et al., 2015). (C) Differences in ACC activity between healthy controls and a high-functioning ASD group during a social interaction task (Chiu et al., 2008). Patterns of activity in controls and ASD (left) were different at a moment in time when subjects could predict the consequences of their actions on the reward that would be obtained by another. The response of the cingulate at that moment in time, correlated with ASD symptom severity (right). (D) Activity in the ACC to the pain of others (pictures of others’ in pain: y axis) is negatively correlated with callous (psychopathic/ICU-callous traits) traits (x axis) in children with conduct problems (Lockwood et al., 2013). Although responses in (C) and (D) were not localized specifically to the sulcus or gyrus, they do provide evidence that links together ACC and impairments in social behavior.

Figure 5

The MoVE Model (A) Prior to another’s movement, the value of a behavior for another individual can be estimated. From this it is possible to compute how motivated a specific individual will be following a specific cue associated with a particular behavior pattern. That is, an observer can weigh up the costs and benefits of a behavior for a specific other individual and vicariously estimate the expected “value” of a behavior for that individual (Vv). The benefit (Rv) for another reflects the motivational benefit of acting. Thus, value increases if the magnitude of a reward is higher, but also if magnitude of painful stimulation will be greater. Thus, the magnitude (salience) of the potential outcome, regardless of its valence, increases “value” in this model. The magnitude of the benefit is temporally discounted (kD), by the probability of the behavior succeeding (P) and weighed against the effort cost of exerting the action (ϕE). (B) These estimates are updated by two different categories of prediction error (PE). Vicarious outcome prediction errors (vOPEs) update expectations about the value of a behavior for another based on the presence or absence of a beneficial outcome (i.e., Rv, D, or P are updated); vicarious dynamic PEs (vDPEs), errors in the estimation of another’s motivation based on their movement kinematics and also update estimates of the motivation of another. These error signals can update the parameter estimates for another agent (i.e., K, θ, and ϕ), although we note that these idiosyncratic motivational parameters are not “stored” in the ACC, they influence estimates of how motivated a specific other individual may be. As a result, estimates of the motivation of another agent are updated by inferences about the internal states that guide another’s behavior. (C) There are multiple possible ways in which under- or over-estimation of parameters relating to the other individual can lead to an inaccurate estimate of how motivated the other individual will be, leading to inaccurate planning of one’s own behavioral response. Two examples are represented here graphically. First (left), we show that over-estimating how sensitive another will be (ϕ = 0.75) to the true effort (E) leads to a reduction in how motivated we infer another will be (light green) compared to their true level of motivation (dark green). Second (right), we show that under-estimating the probability (θP) that another will obtain a benefit leads to an under-estimation of their motivation (dark green) compared to their actual level of motivation (light green). Such changes illustrate how inaccurate parameter estimates can lead to misrepresentation of another’s motivation.