Hierarchical Brain Systems Support Multiple Representations of Valence and Mixed Affect

The Psychological Structure of Valence

Classically, positivity and negativity were conceived to comprise extremities on a single bipolar dimension of valence (Green, Goldman, & Salovey, 1993; Russell & Fehr, 1987; Smith & Ellsworth, 1985; Wundt, 1912/1924). According to this “bipolar” perspective, positive and negative affect oppose such that high negative affect is concurrently accompanied by low positive affect, and vice versa. It has been argued that apparent independencies between positive and negative affect disappear when statistical approaches to control measurement error are considered, with the emergence of an underlying bipolar affective structure (Green et al., 1993).

One limitation in the conceptualization of affect along a single bipolar dimension is precisely the requirement of positivity and negativity as opposing extremes. One consequence is a limited ability to account for apparent simultaneous processing of positive and negative evaluations (i.e., ambivalence). Yet, considerable evidence has been provided that such competition between positive and negative processing can occur (e.g., Bargh, Chaiken, Govender, & Pratto, 1992; Bassili, 1996; van Harreveld, van der Pligt, de Vries, Wenneker, & Verhue, 2004).

Alternative accounts to a bipolar perspective on the structure of valence have taken a “bivalent” perspective, emphasizing instead independent dimensions of negativity and positivity (Cacioppo & Berntson, 1994; Cacioppo et al., 1997; Kaplan, 1972; Lang, Bradley, & Cuthbert, 1998; Watson & Clark, 1997). This argument has taken multiple forms: from a delineation of motivational systems underlying appetitive and aversive response (e.g., Cacioppo & Berntson, 1994; Gray, 1990; Lang et al., 1998) to statistical approaches from which independent constructs for positivity and negativity emerge (e.g., Watson, Clark, & Tellegen, 1988). Similarly, theories of evaluative processes conceptualize evaluations as joint functions of positive and negative reactions to external stimuli, which may vary in terms of positive and negative activation separately (Cacioppo et al., 1997). If evaluations are a joint function of appetitive and aversive responses to information in the world, bipolar perspectives that require positive and negative responses to be reciprocally controlled cannot account for nonreciprocal contributions of positive and negative processing. Independent systems of positive and negative affect, on the other hand, are better suited to explain ambivalent (high activation of both positive and negative) and indifferent (low activation of both) evaluations.

However, one fallacy in arguments for independent dimensions of positivity and negativity (Watson & Clark, 1997) is that they assume bipolar models of valence imply a unidimensional affective space, though it has been demonstrated that a multidimensional affective space better fits the empirical data (Barrett & Russell, 1998). In an effort to reconcile this debate, circumplex models of affect propose an affective space of two dimensions, but consisting of a bipolar valence dimension and an orthogonal arousal dimension (Barrett & Russell, 1998; Russell, 1980). One strength of this conceptualization is that multiple emotional states can be described within a circumplex space of core affect, based on relative positions on arousal and valence axes. Much work has been elaborated on this idea, where psychological constructivist theories explain emotional states as emergent properties from a few core dimensions (Barrett, 2009), and recent evidence has provided support that interacting brain systems involved in basic mental operations can account for a variety of emergent emotional phenomena (Lindquist, Wager, Kober, Bliss-Moreau, & Barrett, 2012).

More recent perspectives have differentiated between levels of analysis in conceptualizing the structure of valence and its role toward evaluative processes (Shuman et al., 2013). Building from a component process model of appraisals (Scherer, 1984), this “levels of valence” approach allows different types of valence to coexist, each corresponding to different facets of appraisal dimensions such as goal conduciveness, self-congruence, and moral goodness. Because multiple valenced appraisals are not only simultaneously present in mental life but often interact, this perspective gives a detailed account of different routes to mixed emotional experiences, which have increasingly been demonstrated empirically (Larsen & McGraw, 2011; Larsen, McGraw, & Cacioppo, 2001; Scherer & Ceschi, 1997; Scherer, Wranik, Sangsue, Tran, & Scherer, 2004). Simultaneous positive and negative processing can unfold on different levels of valenced appraisals or can occur within the same type of appraisal. For example, playing sports can be experienced as unpleasant (e.g., muscle pain, tiredness) and simultaneously interpreted as goal-conducive (e.g., keeping fit), thus being appraised positively and negatively (i.e., ambivalent) on two different appraisal dimensions. Further, a particular action may be simultaneously goal-conducive and obstructive (i.e., the same appraisal dimension) if its consequences have implications for two competing goals (e.g., work–life balance). Critically, this multilevel approach to valence allows for integration of valenced features of specific evaluations (e.g., “microvalence”) and general valence that can be employed as “common currency” in an iterative evaluative process (see also Cunningham et al., 2013). This feature may be directly related to the concept of utility in neuroeconomic theory where it is used to compare between choices for decision-making (Kable & Glimcher, 2007; Rangel, Camerer, & Montague, 2008).

Finally, to provide a mechanism by which levels of analysis might interact, a framework has been proposed that describes the manner in which affectively charged evaluations dynamically co-occur and interact across time, with each level of processing iteratively influencing the others (Cunningham et al., 2013; Cunningham & Zelazo, 2007; Cunningham et al., 2007). Critically, the iterative reprocessing model (Cunningham et al., 2007) provides a brain-based account of how these psychological interactions could take place through the description of hierarchical brain organization, such that lower order processes are reciprocally influenced by higher order reflective processes to generate dynamic evaluations. Subsequent empirical work has demonstrated that discrete emotional labels can be captured by the interacting evaluations of active states through time (Kirkland & Cunningham, 2012).

Here we constrain the investigation of valence organization by first discussing the psychological research on ambivalence and then reviewing the neuroscience literature. Following from perspectives emphasizing hierarchical brain organization and differentiating affective processes at distinct levels of analysis, we consider whether the brain is able to simultaneously represent positive and negative information across large-scale brain systems (macro-level), between unique brain circuits defined by anatomical and functional connectivity across brain regions (meso-level), and within intraregional distributions of neuronal populations (micro-level). Can the consideration of brain organization provide some basis for explaining the simultaneous processing of negative and positive attributes of specific evaluations?

Ambivalence as the Simultaneous Processing of Positivity and Negativity

Ambivalence, the co-occurrence of positively and negatively valenced evaluations, is often brought up as an example of a mixed-valence response. When experiencing ambiguity (e.g., under uncertain gain/loss conditions in monetary gambles), positive and negative information must be simultaneously represented and integrated for adaptive decision-making (Hsu, Bhatt, Adolphs, Tranel, & Camerer, 2005). Such competition between positively and negatively valenced information has been demonstrated, for example, when tracking mouse trajectories during the decision process. During a forced-choice task, participants’ mouse movements on the computer screen demonstrated more “pull” toward the nonchosen option as well as more directional reversals over the horizontal and vertical axis when evaluating ambivalent attitude objects (i.e., x-flips and y-flips, respectively; Schneider et al., 2015). Even though these directional reverses are often interpreted as a sign of complexity (Freeman & Ambady, 2010), in this case, when forcing a single evaluative response from contradictory evaluations, horizontal reversals may also represent the simultaneous salience of positivity and negativity.

More cognitive resources are necessary in situations when positive and negative characteristics have to be integrated to derive a single evaluative response and ambivalence is experienced as an internal conflict (Newby-Clark, McGregor, & Zanna, 2002; van Harreveld, Nohlen, & Schneider, 2015). Critically, such evaluations are made flexibly and in line with current contextual demands. Even though evaluations can rely on generally stable (positive and negative) attitude representations, these representations are re-evaluated given the current context so that the same event may be evaluated positively or negatively in order to create adaptive responses (Cunningham et al., 2007). Along these lines, neuroimaging research on ambivalence has reported the engagement of a broad network of regions including the anterior cingulate cortex (ACC), frontopolar cortex, inferior frontal cortex, and ventromedial prefrontal cortex (vmPFC) (Cunningham, Johnson, Gatenby, Gore, & Banaji, 2003; Cunningham, Raye, & Johnson, 2004; Jung et al., 2008; Luttrell, Stillman, Hasinski, & Cunningham, in press; Nohlen, van Harreveld, Rotteveel, Lelieveld, & Crone, 2013). Especially in the context of dichotomous good/bad judgments, processing information that is characterized by ambivalent features creates cognitive conflict, a state that has repeatedly been linked to greater activation in the ACC (e.g., Botvinick, Braver, Barch, Carter, & Cohen, 2001; Carter & van Veen, 2007), for example in complex decision situations (Kitayama, Chua, Tompson, & Han, 2013; Pochon, Riis, Sanfey, Nystrom, & Cohen, 2008). Similarly, the ACC may play a critical role in using positive (i.e., rewarded) and negative (i.e., non-rewarded) information to guide decision-making (Kennerley, Walton, Behrens, Buckley, & Rushworth, 2006). In processing the ambivalent features of a stimulus, ACC activation may reflect difficulty in integrating conflicting information into a single response on a bipolar valence dimension. Crucially, the ACC signal that tracks uncertain responses guided by ambivalent features is distinct from neural markers related to certain responses guided by non-ambivalent information (Luttrell et al., in press).

However, the possibility of ambivalent evaluations does not necessarily map directly on positive and negative emotional responses. Whereas some studies have indicated that salient, ambivalent evaluations can result in stronger self-reported mixed emotions (Newby-Clark et al., 2002), others have reported that experiencing ambivalence results in negative affective responses (van Harreveld, Rutjens, Rotteveel, Nordgren, & van der Pligt, 2009). Even though evaluating something as simultaneously positive and negative is not the same as experiencing positive and negative emotions, we argue that valence underlies both these experiences. It is plausible that parallel processing of valence occurs across multiple levels, from perceptions of affective quality to the experience of emotion. We subsequently provide evidence supporting the existence of multiple valence circuits in the brain that interact to give rise to such complex affective responses.

Large-Scale Networks Support Specific and Generalized Valence

Human neuroimaging research has recently focused on the role of large-scale brain networks in cognition. We refer here to functionally interconnected brain regions that are widely distributed across cortical and subcortical structures, working in concert to support mental phenomena. Correspondingly, research at this level of resolution often relies on techniques that can probe spatially disparate regions simultaneously, such as functional magnetic resonance image (fMRI), positron emission tomography (PET), electroencephalography (EEG), and magnetoencephalography (MEG; e.g., Andrews-Hanna, Smallwood, & Spreng, 2014; Bressler & Menon, 2010), or on meta-analytical methods to integrate across multiple findings (Bartra, McGuire, & Kable, 2013; Kühn & Gallinat, 2012; Lindquist, Satpute, Wager, Weber, & Barrett, 2015; Liu, Hairston, Schrier, & Fan, 2011; Murphy, Nimmo-Smith, & Lawrence, 2003; Vytal & Hamann, 2010; Wager, Phan, Liberzon, & Taylor, 2003).

The logic at the macro-level analysis of valence processing in the brain is that if a single network of regions is found to respond monotonically to changes between positive, negative, and neutral stimuli, this would support a bipolar perspective of valence; alternatively, if spatially independent brain regions supported positive and negative information then a bivalent hypothesis would be supported. A third “affective workspace” hypothesis has been posited in which emergent valence is flexibly implemented by a set of core, generalized regions (Lindquist et al., 2015; Liu et al., 2011). According to this view, regions in the generalized valence network can be functionally selective to either positive or negative processing, but are able to fluctuate selectivity in a dynamic manner according to contextual demands, including both external (e.g., situational) and internal (e.g., goal-based) influences.

Largely, meta-analytic evidence has pointed towards this generalized workspace in the brain that flexibly supports positivity and negativity according to context (Barrett & Satpute, 2013; Lindquist et al., 2015; Liu et al., 2011; Murphy et al., 2003; but see Vytal & Hamann, 2010). Meta-analytical findings from studies that have focused on affective features such as valence, rather than discrete emotions, have converged on a generalized network central to both positive and negative processing that comprises cortical and subcortical regions involved in both higher order cognitive and early sensory processes: the anterior insula, anterior cingulate cortex (ACC), ventromedial prefrontal cortex (vmPFC), amygdala, ventral striatum (including the nucleus accumbens [NAcc]), thalamus, and occipitotemporal cortex (Lindquist et al., 2015; Liu et al., 2011). Intracranial EEG recordings in humans have also pointed to a distributed set of regions that support both positive and negative valence representations, including limbic/paralimbic, subcortical, and neocortial regions (see Guillory & Bujarski, 2014, for a review).

A similar generalized network of brain regions has been found to support positive and negative reward processing (Bartra et al., 2013; Clithero & Rangel, 2014; Liu et al., 2011; Sescousse, Caldú, Segura, & Dreher, 2013), with separable subcircuits within this general network supporting different stages of reward processing, including the anticipation (ACC, anterior insula, brainstem) and outcome (NAcc, medial orbitofrontal cortex [mOFC], amygdala) of rewards (Liu et al., 2011). However, this work also found separable circuits underlying positive (mOFC, PCC) and negative reward (ACC, anterior insula, lateral PFC). Together, the literature at the macro-level of analysis suggests a generalized, large-scale capacity in the brain for the processing of both positive and negative information; nonetheless, a deeper consideration of the neural circuitry underlying these processes at higher levels of spatial resolution may illuminate greater specificity in function.

Brain Circuits in Appetitive and Aversive Processes

At the meso-level of analysis, we consider brain circuits composed of a set of functionally connected regions that support a common psychological process. At this and finer levels of resolution, the substrates of positive and negative processing begin to become more specialized. One way in which research on the processing of valence has operationalized positivity and negativity is with paradigms that tap into reward (Baxter & Murray, 2002; Blood & Zatorre, 2001; Davis & Whalen, 2001; O’Doherty, Kringelbach, Rolls, Hornak, & Andrews, 2001) and threat (Davis & Whalen, 2001; LeDoux, 1998, 2000; Namburi, Al-Hasani, Calhoon, Bruchas, & Tye, 2015) processes, respectively.

The critical role of the amygdala as a hub for both reward- and threat-related processing circuits has been well established (e.g., Baxter & Murray, 2002; LeDoux, 2007). Though often discussed as an individual structure in human neuroimaging studies (e.g., Cunningham, Raye, & Johnson, 2005; Phelps et al., 2001), the amygdala comprises an extended system of several nuclei with unique cytoarchitecture and different connectivity patterns (Amunts et al., 2005; Sah, Faber, Lopez De Armentia, & Power, 2003) and include the lateral, basal, and accessory basal nuclei (together the basolateral amygdala [BLA]), the central (CeA) and medial nuclei (together the centromedial amygdala [CeM]), and the cortical or superficial nucleus (SFA) (Davis & Whalen, 2001).

We propose that these distinct but interconnected subnuclei of the extended amygdala take part in dissociable neural circuits that support different functions, such as approach and aversive processes (LeDoux, 2000). Indeed, recent neuroimaging work has focused on delineating the different subnuclei of the amygdala (Roy et al., 2009) and discovering the cognitive and social processes in which each partake (Bickart, Dickerson, & Barrett, 2014; Bickart, Hollenbeck, Barrett, & Dickerson, 2012). Regardless of valence, the amygdala serves as a central routing hub where the bottom-up stimulus-driven processing of information interacts with top-down influences, consistent with the psychological perspective that evaluative processes can shape how we perceive and construe incoming external stimuli (Cunningham & Brosch, 2012; Cunningham et al., 2007).

The central nucleus of the amygdala has been found to be a critical structure in the neural circuit underlying an automatic system of threat detection (Davis & Whalen, 2001), consistent with fMRI studies that found greater amygdala responses to negative affective states across stimulus modalities, such as faces (Anderson, Spencer, Fulbright, & Phelps, 2000), words (Isenberg et al., 1999), and odours (Zald & Pardo, 1997), and at multiple stages of processing from perception (Adolphs et al., 1999; Calder, Keane, Manes, Antoun, & Young, 2000) to evaluation (Aldophs & Tranel, 2003). Output signals from the central nucleus of the amygdala form an important pathway that supports the expression of affective responses to threat as well as associated physiological responses. Projection sites from the central nucleus include the periaqueductal gray (PAG) and the lateral and paraventricular nuclei of the hypothalamus (Rizvi, Ennis, Behbehani, & Shipley, 1991), which regulate defensive fight or flight behaviour (LeDoux, Iwata, Cicchetti, & Reis, 1988), as well as cholinergic systems of the forebrain (Everitt & Robbins, 1997; Holland & Gallagher, 1999; LeDoux, 2000).

On the contrary, different circuits support reward processes and corresponding approach behaviour, and the focus within the amygdala shifts to the basolateral complex (BLA; Baxter & Murray, 2002; Davis & Whalen, 2001). Again, the specific connectivity between brain regions forms a complex network that supports multiple facets of reward processing (Elliott, Friston, & Dolan, 2000; Haber & Knutson, 2010; Kirkland, Man, & Cunningham, 2014). Critically, dopamine neurons from the substantia nigra and ventral tegmental area project to the basal, lateral, and accessory basal nuclei of the amygdala as well as the NAcc core (Berridge & Robinson, 1998). From there, this mesolimbic dopaminergic circuit continues to the prefrontal regions such as the vmPFC, from which there are important feedback projections to the BLA and NAcc that support reward processes flexibly (Baxter & Murray, 2002; Price, Carmichael, & Drevets, 1996).

The reciprocal connectivity between the BLA and NAcc to the vmPFC is critical for guiding choice behaviour and response selection. To probe how the interaction between the two regions contributes to decision-making, a crossed-disconnection manipulation paradigm lesioned the amygdala and the contralateral OFC, and the connections between them were severed with a commisurotomy (Baxter, Parker, Lindner, Izquierdo, & Murray, 2000). The authors found that lesioned monkeys were no longer able to adjust choice behaviour given shifts in reward outcomes, but food preferences and motivated behaviour were not affected. These findings indicate that the stored information about reward value was no longer being updated, or alternatively, that successful updates of reward value were no longer integrated to inform response selection. The latter is consistent with neuropsychological findings that patients with damage to the vmPFC are unable to optimize choice selection to increase net monetary gain, indicating a failure in assessing the relative risks and benefits of choices (Bechara, Damasio, Damasio, & Anderson, 1994). Indeed, the authors have argued the similarity in behavioural outcome between BLA and OFC lesion patients may be attributable to the role of the amygdala in providing affective information to guide decision-making (Bechara, Damasio, Damasio, & Lee, 1999; Bechara, Damasio, Tranel, & Damasio, 1997). Feedback projections from prefrontal regions to the amygdala and the NAcc support the regulation of emotional states (Ochsner, Bunge, Gross, & Gabrieli, 2002; Ochsner & Gross, 2005; Ochsner, Silvers, & Buhle, 2012) and the regulation of reward-seeking behaviour (Haber & Knutson, 2010; Kober et al., 2010; Staudinger, Erk, & Walter, 2011). Together, the findings suggest a critical role of amygdala–OFC connectivity in supporting decision-making in face of changing reward values, whereby the BLA supports the representation of current value and the OFC integrates this information with previous learning to guide behaviour.

Altogether, the research we highlight in this section points to the idea that the processing of positive (rewarding) and negative (threatening) information at the meso-level of analysis may rely on distinct circuits that share several hub regions such as the amygdala. In the following section, we consider how the neural heterogeneity within a single region accounts for the functional diversity attributed to that region. We continue to examine the amygdala and look at the OFC as candidate regions where the neural coding allows for the representation of both positive and negative information.

Intraregional Distributed Networks and Parallel Processing

Within the amygdala, the intraregional connectivity between the lateral (LA) and central amygdala (CeA) comprises an input–output functional model, where incoming multimodal sensory and associated affective (i.e., rewarding or punishing) information are bound and integrated at the projection terminals of the dorsal LA (LeDoux, 2015). From here, communication continues to the intercalacted layer (ITC), an important layer of GABAergic inhibitory neurons situated between the CeM and BLA (Amano, Unal, & Paré, 2010). The ITC comprises an inhibitory “core” that dampens spontaneous cell activity. Subsequent projections from the ITC terminate at mostly inhibitory neurons in the CeA, from which information is carried forward to brainstem and hypothalamic targets via disinhibitory mechanisms, allowing for the expression of behavioural and physiological responses (see LeDoux, 2007, for a review).

Within the subregions of the amygdala there is converging evidence for the interspersed distribution of reward- and threat-encoding principal neurons. These two subpopulations of valence-specific neurons are spatially interspersed in the BLA (Shabel & Janak, 2009), and each form a distinct distributed circuit (Zhang et al., 2013). In addition to the distributed neuronal populations that encode specific valence, there is a subset of neurons in the BLA that are valence-general, instead encoding salience (Shabel & Janak, 2009). The flexibility apparent in amygdala functional networks is not due to a switch in valence specificity of specific neuronal populations, which become locked after associations are formed. Rather, flexible processing is supported by the connectivity of these valence networks within the amygdala to hippocampal subregions such as the dentate gyrus (DG), which carries contextual information that shapes response (Redondo et al., 2014).

Output targets of the basal and central nuclei are also differentiated, with the basal nucleus projecting to striatal and prefrontal regions and the central nucleus projecting to the hypothalamus, PAG, and modulatory neurotransmitter systems relevant for arousal (e.g., norepinephrine) and attention (e.g., acetylcholine), consistent with the distinct reward- and threat-relevant meso-level circuits described before. As such, this functional model describes an intraregional connectivity architecture that allows for valence-general processing and an extraregional connectivity profile that allows for valence-specific processing (LeDoux, 2007).

Similarly, within the mOFC/vmPFC, functional subdivisions and their contribution to distinct circuits in the brain have been highlighted in supporting various aspects of value-based decision processes, including the computation of values ascribed to stimuli across reward modalities and task demands (Clithero & Rangel, 2014) as well as different levels of reward abstraction (Kringelbach, 2005; Sescousse et al., 2013). Subdivisions of the OFC show different responses to various aspects of goal states; nevertheless, the region continues to provide a common neural basis for both appetitive and aversive goals (Kringelbach & Rolls, 2004). Medial regions have been linked to memory and learning processes as well as monitoring and regulation (Kringelbach, 2005), and the integration of stimulus–outcome associations across multiple modalities (Grabenhorst & Rolls, 2011). On the other hand, lateral subdivisions within the OFC are linked to the evaluation of rewarding and punishing properties. Anterior to posterior distinctions within the OFC track the abstractness of the input valence stimulus information, with posterior OFC regions serving as multimodal sensory integration hubs and anterior regions representing, integrating, and comparing increasing abstraction in reward contingencies (Cunningham, Kesek, & Mowrer, 2009; Cunningham & Zelazo, 2009; Kringelbach, 2005; Zelazo & Cunningham, 2007).

A recent study examined the question of whether negative and positive features of stimuli were represented in a continuous manner and where in the brain this representation occurred (Chikazoe, Lee, Kriegeskorte, & Anderson, 2014). Consistent with the meta-analytical findings arguing for a generalized affective workspace reviewed before, the authors found regions that supported the representation of both positive and negative features of stimuli, rather than independent regions supporting positivity versus negativity. Importantly, they distinguished between brain regions that supported stimulus modality-specific coding of valence such as the posterior temporal and insular cortices and those that supported supramodal representations of valence including the medial and lateral OFC. Population-level coding within the same region using multivariate representational similarity analyses was able to distinguish between positive and negative valence, consistent with animal electrophysiological work demonstrating that positivity- and negativity-specific neurons are interspersed within the OFC (Morrison & Salzman, 2009). Together, these findings suggest that populations of neurons within regions across the brain, including the OFC, support both positive and negative valence.

Together, the literatures across separate brain regions at the micro-level of analysis, and in particular the amygdala, point toward an efficient system that allows for simultaneous valence representations.

Integration

At the macro-level of analysis, large-scale systems of the brain are not dissociable into discrete positive and negative networks. Rather, a generalized network of brain regions together supports valence processing, and is able to flexibly tune towards positive and negative processes. Such a system is compatible with the aforementioned psychological model of mixed emotional experiences arising for the interplay between positive and negative evaluations. It is likely that such a system interacts with other large-scale brain networks that are the underpinning of the cognitive “ingredients” of subjective experience (Barrett, 2009), including systems related to episodic memory (Schacter, Addis, & Buckner, 2007), various forms of attentional processes (Corbetta & Shulman, 2002), perceptual decision-making (Heekeren, Marrett, & Ungerleider, 2008), and motor networks (Grillner, 2003).

The perspective on valence processing at the meso-level is analogous to the concept of a generalized “affective workspace” at the macro-level, in that both perspectives describe flexible neural systems that allow for the processing of positive and negative information. It is plausible that at this level of analysis, information about current goal states, represented primarily in the OFC, biases the manner in which incoming affective information is evaluated via infralimbic projections to the intercalated layer within the amygdala, thereby altering the neural “landscape” through which information passes.

Furthermore, while the evidence suggests that there is some degree of valence-specific processing in the amygdala, the perspective that distinct nuclei can be functionally ascribed to positivity and negativity arises from an incomplete model at a coarse spatial resolution. Consideration of the neuronal architectural nuances at the micro-level of analysis reveals a more complex picture of valence processing in the amygdala. Together, the evidence points to a neuronal architecture within the amygdala whereby distinct functional networks support positive and negative processing specifically, as well as a “core” set of neurons that encode both positivity and negativity.

Through our review of the neuroscience literature, we argue that the architecture of the brain allows for the simultaneous processing of both positive and negative information across multiple spatial scales. This has important implications for psychological theories of both the organization of valence as well as mixed evaluative processes. Evidence from the neuroscience literature converges with findings in the psychological research on ambivalence. Critically, we emphasize that while the brain has the capacity for simultaneous and mixed valence processing, how this organization is employed for affective experiences is largely context-dependent. We encourage further work probing the role of both external and internal contexts, such as situational demands and intrinsic goals, respectively, in shaping brain responses to affective information.