Pace-Schott, E. F., Amole, M. C., Aue, T., Balconi, M., Bylsma, L. M., Critchley, H., Demaree, H. A., Friedman, B. H., Gooding, A. E. K., Gosseries, O., Jovanovic, T., Kirby, L. A. J., Kozlowska, K., Laureys, S., Lowe, L., Magee, K., Marin, M.-F., Merner, A. R., Robinson, J. L., & Smith, R. C. (2019). Physiological feelings. Neuroscience & Biobehavioral Reviews, 103(103), 267–304. https://www.sciencedirect.com/science/article/pii/S0149763418308674 (Pace-Schott et al., 2019)
Edward F. Pace-Schott a, Marlissa C. Amole b, Tatjana Aue c, Michela Balconi d, Lauren M. Bylsma b, Hugo Critchley e, Heath A. Demaree f, Bruce H. Friedman g, Anne Elizabeth Kotynski Gooding f, Olivia Gosseries h, Tanja Jovanovic i, Lauren A.J. Kirby j, Kasia Kozlowska k, Steven Laureys h, Leroy Lowe l, Kelsey Magee f, Marie-France Marin m, Amanda R. Merner f, Jennifer L. Robinson n, Robert C. Smith o, Derek P. Spangler p, Mark Van Overveld q, Michael B. VanElzakker r
a Harvard Medical School, Boston, MA, USA
b University of Pittsburgh, Pittsburgh, PA, USA
c University of Bern, Bern, Switzerland
d Catholic University of Milan, Milan, Italy
e University of Sussex, Sussex, UK
f Case Western Reserve University, Cleveland, OH, USA
g Virginia Tech, Blacksburg, VA, USA
h University of Liege, Liege, Belgium
i Emory University, Atlanta, GA, USA
j University of Texas at Tyler, Tyler, TX, USA
k University of Sydney, Sydney, Australia
l Neuroqualia (NGO), Truro, Nova Scotia, Canada
m Université du Québec à Montréal, Montreal, Canada
n Auburn University, Auburn, AL, USA
o Michigan State University, East Lansing, MI, USA
p United States Army Research Laboratory, Aberdeen, MD, USA
q Erasmus University Rotterdam, Rotterdam, the Netherlands
r Massachusetts General Hospital, Boston, MA, USA
Received 13 November 2018, Revised 27 March 2019, Accepted 3 May 2019, Available online 22 May 2019, Version of Record 28 June 2019.
What do these dates mean?
Publication milestones
The dates displayed for an article provide information on when various publication milestones were reached at the journal that has published the article. Where applicable, activities on preceding journals at which the article was previously under consideration are not shown (for instance submission, revisions, rejection).
The publication milestones include:
- Received: The date the article was originally submitted to the journal.
- Revised: The date the most recent revision of the article was submitted to the journal. Dates corresponding to intermediate revisions are not shown.
- Accepted: The date the article was accepted for publication in the journal.
- Available online: The date a version of the article was made available online in the journal.
- Version of Record: The date the finalized version of the article was made available in the journal.
More information on publishing policies can be found on the Publishing Ethics Policies page. View our Publishing with Elsevier: step-by-step page to learn more about the publishing process. For any questions on your own submission or other questions related to publishing an article, contact our Researcher support team.
Highlights
- •Peripheral physiological changes can be perceived as feelings via interoception.
- •Both perceived and unperceived interoceptive information contribute to emotions.
- •The forebrain can generate physiological feelings without actual peripheral events.
- •Interacting networks of afferent and efferent signals generate physiological feelings.
- •Physiological feelings contribute to survival, reproduction and emotion regulation.
Abstract
The role of peripheral physiology in the experience of emotion has been debated since the 19th century following the seminal proposal by William James that somatic responses to stimuli determine subjective emotion. Subsequent views have integrated the forebrain's ability to initiate, represent and simulate such physiological events. Modern affective neuroscience envisions an interacting network of “bottom-up” and “top-down” signaling in which the peripheral (PNS) and central nervous systems both receive and generate the experience of emotion. “Feelings” serves as a term for the perception of these physical changes whether emanating from actual somatic events or from the brain's representation of such. “Interoception” has come to represent the brain's receipt and representation of these actual and “virtual” somatic changes that may or may not enter conscious awareness but, nonetheless, influence feelings. Such information can originate from diverse sources including endocrine, immune and gastrointestinal systems as well as the PNS. We here examine physiological feelings from diverse perspectives including current and historical theories, evolution, neuroanatomy and physiology, development, regulatory processes, pathology and linguistics.
Keywords
Emotion, Feelings, Interoception, Somatic markers, Emotion regulation, Autonomic nervous system, Insula
Introduction
Conscious emotional experience is closely bound to changes in bodily sensations. Indeed, if one accepts the notion that consciousness is grounded in biological processes (Crick, 1994, Damasio, 1994, Pinker, 2018), emotional experience must, by its nature, be physiological. However, the direction of causality and the specificity of relationships between physiological processes and emotional experience have been topics of debate since at least the 19th century. This review focuses on interaction between subjective emotional experience and the body. Specifically, we address how emotional feelings are informed by physiological processes that are detected by peripheral organ ‘interoceptors’, encoded and transmitted from the peripheral nervous system (PNS), to be represented at specific levels of the central nervous system (CNS). We additionally address how emotional experience may encompass the prediction, as well as feedback, of physiological changes, through efferent responses from CNS to PNS that change bodily physiology as an obligatory component of an emotive response.
This work is undertaken as part of ‘The Human Affectome Project’, an initiative organized in 2016 by a non-profit organization called Neuroqualia. The project aims to produce a series of overarching reviews that summarize current knowledge about affective neuroscience and the language that we use to convey feelings and emotions. The projects comprise twelve teams, organized into a task force with the aim of developing a comprehensive and integrated model of affect to serve as a common focal point for future affective research. To that end, our team was specifically tasked to review neuroscience research on the relationship between subjective experiences of emotion and associated physiological processes. We were also asked to review the language that people use to express feelings that relate to physiological processes and consider whether or not feelings that people convey in language might inform the way we approach neuroscience research on these topics. An important reason to focus on the physiology of emotional experience and expression along with consideration of emotional language is the well-documented imprecision inherent in self-report instruments for such experiences (Harmon-Jones et al., 2016, Mauss and Robinson, 2009). Moreover, emotion is a multi-component process hence it is important to examine from different levels, including physiology, which is complementary to self-reported experience and behavior.
A unifying theme of this review is that the subjective experience of emotion is influenced, and often determined by CNS representations of afferent sensory input from peripheral organs and tissues – conventionally termed interoception (Craig, 2016, Critchley and Garfinkel, 2017, Damasio and Carvalho, 2013). We strive to follow Damasio and Carvalho's (Damasio and Carvalho, 2013) distinction between ‘emotion’ – an adaptive, patterned neural response akin to physiological drives such as hunger and thirst, and ‘feeling’ – internal sensations resulting from these patterned responses as well as from other interoceptive sources. However, we also acknowledge models of emotion that encompass experiential, physiological and behavioral dimensions as well as the notion of ‘construction’, in which conscious affect lies at a level below labeled emotions built from cognitive representations of context, physiology and prior experience (Barrett, 2017c). (Note that operational definitions are provided in Box 1 and nuanced distinctions are further elaborated by the Affectome Project in the “Linguistics” section of this review).
Box 1
Operational, working definitions of terms frequently used throughout this review that roughly follow those of Damasio and Carvalho (2013). Note that these definitions are influenced by what they seek to explain and they are not advanced as universal definitions. Additional terms are defined in relevant sections.
Feeling
Sensations perceived as emanating from inside the body that may originate from afferent information from peripheral receptors (including visceral, temperature and pain) that has been processed to varying degrees (e.g., in brainstem nuclei and thalamus), or that may originate from central representations of such bodily sensations in the brain itself.
Emotion
A programmed neural response evolved to serve an adaptive function by mobilizing specific neural activity in both the brain and periphery and by favoring certain behaviors. An emotional response can be evoked by, generate, or be shaped by specific feelings as well as by specific exteroceptive stimuli, cognitions or cognitive processes.
Mood
A valenced, tonic, persisting state during which valence-congruent phasic feelings may repeatedly arise or emotional responses may be evoked. Mood is distinguished from feelings and emotion by its duration whether or not it is tied to a specific stimulus.
Affective
Pertaining to a valenced feeling, mood or emotional response.
Afferent
A directional term referring to information flow, encoded as neural activity, from a more peripheral locus in the nervous system to a more central locus. For example, movement, at each step or in total, from a sensory receptor through the peripheral nervous system, spinal cord and/or brainstem nuclei and thalamus, to primary, unimodal and multimodal association cortices.
Efferent
A directional term referring to information flow from a more central locus in the nervous system to a more peripheral locus. For example, movement in the opposite direction along the above afferent pathway.
Interoception
Receipt, by the brain, of afferent sensory input from peripheral organs, tissues and physiological processes, or receipt of information from brain representations of such organs, tissues and processes by other regions of the brain. Interoception may be consciously perceived in which case it is very similar to feeling. Interoception may also affect feelings or generate peripheral physiological changes by evoking efferent signals below the level of consciousness.
Homeostasis
An optimum physiological state from which departures trigger automatic physiological and behavioral processes which seek to return physiological values to those of that state. Emotions have been hypothesized to have evolved, in part, to favor behaviors that promote return to homeostasis.
Emotion processing
Changes occurring in an emotional response once it has been evoked that may result from deliberate or automatic processes such as cognitive attributions that diminish or augment response, intrinsic physiology (e.g., activation then dissipation of stress response), active behavior (e.g., avoiding the evoking stimlus), etc.
Emotion regulation
A form of emotion processing, deliberate or automatic, that acts to augment or diminish the duration or intensity of an emotional response to a more manageable level. Regulation may occur before (e.g., avoidance, cognitive preparation) or during an emotional response (e.g., suppression, reattribution). Regulation may also result from low-level neural processes such as habituation and extinction.
Emotional perception
The degree of and nature of awareness that an emotional response has been evoked including, for example, continua from unawareness (alexithymia) to hyper-awareness, habitual biases toward one or the other end of a valence continuum.
Emotional experience
The subjective state when an emotional response has been evoked. May vary from unawareness to hyper-awareness of the ongoing response, or may be a transformation of response (e.g., somatization).
(Note feelings and moods can also be processed, regulated, perceived and experienced albeit by mechanisms that may differ from those directed toward an emotional response.)
Three additional aspects of interoception are hypothesized. First, CNS representations of interoceptive information need not reach the level of conscious awareness in order to profoundly influence the subjective phenomena of feelings and emotions (Wiens, 2005). Second, the CNS efferent responses that target the sources of afferent input, as well as the resulting re-afferent feedback, can contribute to the subjective experience of emotion (Damasio and Carvalho, 2013). Third, interoceptive feelings can be both generated and perceived within the CNS (Barrett and Simmons, 2015, Damasio, 1996, Seth and Friston, 2016).These latter two points extend the tight physiological definition of interoception beyond afferent sensory input and representation to inferential computational processes that ultimately support experience. Thus, recent models of interoception and emotion incorporate Bayesian concepts (e.g. predictive coding), wherein the brain makes sense of dynamic changes in a wealth of viscerosensory information by predicting likely causes of those changes: prior interoceptive experiences are deployed as predictions of probable interoceptive experience, to yield ‘prediction errors’, i.e., the difference between an expected internal state and afferent interoceptive data (Barrett and Simmons, 2015, Paulus and Yu, 2012, Saxe and Houlihan, 2017, Seth and Friston, 2016). One remarkable corollary of this perspective is that, when prediction errors are small, interoceptive experience may be generated largely by what our brain expects to feel rather than by actual afferent neural signals from the periphery (Barrett and Simmons, 2015, Seth and Friston, 2016). Also the ‘top-down’ efferent neural drive to the PNS to change the internal bodily state supports ‘active inference’, whereby prediction errors are minimized by changing the input (feedback) at source. These counterintuitive possibilities capture the key function of interoception in the maintenance of physiological homeostasis (Seth and Friston, 2016). Thus, centrally-generated predictions of physiological states are expressed in autonomic reflexes. Unpredicted interoceptive inputs generate prediction errors, which trigger both emotional feelings and homeostatic responses (Seth and Friston, 2016). Although this review will not describe in detail the theoretical and computational models of predictive coding, active inference or Bayesian analyses of the physiology of emotions, it is crucial to bear in mind that the reality of perceived emotions (including their interoceptive components) encompasses more than the direct representation of peripheral states and includes central predictions of peripheral physiology.
Another important, but difficult distinction is between emotion and mood (Beedie et al., 2005). Typically these are operationally distinguished by duration and sometimes by intensity or reactivity (Beedie et al., 2005, Kaplan et al., 2016), but consensus regarding cutoffs on any dimension is lacking. This review conceptualizes emotion as a phasic response, yet recognizes its overlap with the enduring phenomenon of mood (particularly when describing interoceptive influences of tonic or gradually changing physiological state including the fluctuating hormonal milieu, pain, sickness and inflammation). Indeed, mood states are perhaps the affective phenomena most strongly influenced by interoception and are recognizably a powerful biasing factor for the subjective experience and regulation of affective feelings in both normal emotion (Ritchie et al., 2009, Sereno et al., 2015) and psychopathology (Gilbert et al., 2013).
This review will not seek to provide an exhaustive account of the physiology of emotion or the phenomenon of interoception. Instead, we highlight a sample of important theories and discoveries with an appreciation for their historical precedents and clinical relevance. The following provides a brief overview:
Reductionist ‘peripheral’ theories of emotion, asserting that physiological responses precede emotions (James, 1884, Lange, 1922), and their rebuttal by ‘central’ theories asserting central determinants of emotion specificity and nonspecific PNS responses as epiphenomena (Bard, 1928, Cannon, 1928), established a scientific dialectic that persists to this day. However, recent neuroscientific evidence describes dynamic and often flexible interactions, wherein emotions generate efferent influences on peripheral physiology from central representations (including predictive active inference) and are, in turn, shaped by afferent information from the periphery (including prediction errors). These later theories also accommodate critical contributions of cognitive appraisals and psychosocial determinants in inferential predictive representation of emotion.
The evolutionary importance of emotions and their function in intra-specific communication was articulated in Darwin's prescient work, The Expression of the Emotions in Man and Animals (Darwin, 1872). Recognition of the evolutionary importance of emotions has helped drive the field of evolutionary psychology (Cosmides and Tooby, 2013).
Advances in human neuroscience have greatly informed our understanding of emotion with implications for all such emotion theories. Functional neuroimaging, particularly functional magnetic resonance imaging (fMRI), permits detailed examination of CNS representations of emotion. Moreover, meta-analytic approaches to fMRI data continue to highlight key commonalities underpinning emotional mechanisms across distinct studies (e.g. Phan et al., 2002, Phan et al., 2004, Yarkoni et al., 2011). Although fMRI provides excellent spatial resolution, it shows poor temporal resolution. Among other techniques for the study of emotion and interoception, neural recording (scalp and intracranial EEG; magnetoencephalography) using event related potentials (ERPs) provides convergent information on the millisecond scale about timing of relevant brain responses. ERPs are time domain measures which reflect changes in neural activation in response to specific stimuli. Frequency domain measures (i.e., neural oscillations at specific frequencies) can also be extracted from scalp-recorded EEG signals that may reflect important aspects of emotional processing, though thus far these have been primarily investigated in the context of response to rewarding stimuli (for review, see Glazer et al., 2018).
Modern studies of interoception as a discrete topic of neurocognitive inquiry were presaged by decades of research on the influence of the autonomic nervous system on emotion. Following influential reviews (Cameron, 2001, Craig, 2002, Craig, 2016, Critchley, 2005), the study of interoception in relation to normal emotion has blossomed (Barrett, 2017c, Craig, 2010, Critchley and Garfinkel, 2017, Kleckner et al., 2017, Krautwurst et al., 2016, Strigo and Craig, 2016). Three very different types of feelings clearly representative of those based on interoception – nociception, disgust and empathy – will be discussed. Abnormalities of interoception are now recognized to occupy a central role in the conceptualization of addiction and other psychiatric disorders (Gray and Critchley, 2007, Khalsa et al., 2018, Paulus and Stein, 2010, Paulus and Stewart, 2014, Paulus et al., 2009).
Peripheral systems that influence subjective feelings convey physiological information to the CNS via interoceptive mechanisms operating at both conscious and unconscious levels. This is the case both for classical stress systems in the HPA axis and sympathetic nervous system as well as other neuroendocrine systems including gonadal steroids and neuropeptides such as oxytocin. Interoceptive feelings are nowhere more apparent than in the human reward system manifesting both as “wanting” and “liking” (Berridge and Kringelbach, 2015, Paulus and Stewart, 2014). Like the endocrine system, the immune system's influence on emotional experience may result from long-latency, tonic processes, such as inflammation (Harrison et al., 2009). Another newly appreciated influence on emotion is the gastrointestinal system with its enteric nervous system and gut microbiota, where the vagus nerve is proposed as a potential pathway through which the brain and the gut may influences each other (Cryan and Dinan, 2012, Cryan and Dinan, 2015, Rook et al., 2014).
The ability of peripheral physiology to influence emotion below the level of conscious awareness raises the question of whether emotion can be experienced or expressed with absent or attenuated consciousness. This topic has been explored in the case of unresponsive wakefulness syndrome and the minimally conscious state. Emotion in the diminished states of consciousness during sleep and dreaming are also considered.
Regulation of emotion is inseparable from the ability to regulate physiological state. Both conscious and automatic emotion regulation is heavily reliant upon perception of interoceptive feedback. The vagus nerve represents one important conduit for afferent interoceptive information to flow from the viscera, via the brainstem, to the forebrain, as well as for efferent emotion regulatory mechanisms to influence visceral activity. Another key interoceptive afferent input to the CNS is the lamina 1 spinothalamic tract with its A-delta and C-type fibers (Craig, 2002, Craig, 2016). A key mechanism by which humans regulate emotion is the ability to tolerate, via compensatory behaviors, ‘unhealthy’ physiological and psychological states beyond their normative range of values (McEwen and Wingfield, 2003). Adaptive physiological regulation is put on hold to meet transient challenges. These allostatic responses within stress systems allow humans to temporarily tolerate abnormally elevated levels of physiological stress in support of concomitant emotive behaviors. However, when chronic, the emotional (e.g. anxiety) and physiological adjustments increase ‘allostatic load’ undermining long-term physiological and emotional health (McEwen, 1998).
Dramatic changes in the ability and strategies for emotion regulation occur across the lifespan. Notably, acquisition of basic emotion regulatory skills are among the earliest milestones in normally developing infants and toddlers. Emotion regulatory skills continue to develop into adolescence and early adulthood, paralleling brain changes, e.g. myelination of frontal projections. Interoception plays an early and sustained role in these processes, underpinning the subjective physiological experience of distress that demands regulation, and communicating the amelioration of that distress that signals successful regulation and can facilitate learning (by negative reinforcement). A social mechanism, overlooked until only recently – coherence of physiological responses in close dyadic relationships – supports emotional regulation of one or both members of the dyad, and directly facilitates emotion regulatory learning via interoceptive feedback (e.g. in a parent–infant dyad).
Human emotion regulation is aided by two primitive learning processes – extinction and habituation – that we share with nearly the entire animal kingdom. Arguably, these represent the neural foundation upon which more advanced (e.g. cognitive) processes of human emotion regulation are built (Delgado et al., 2008). These processes are accompanied by brain plasticity, and here the emotional regulatory function of sleep is gaining important recognition in affective neuroscience and psychiatry (Goldstein and Walker, 2014, Pace-Schott et al., 2015a, Pace-Schott et al., 2015b, Palmer and Alfano, 2017, Tempesta et al., 2017).
In psychopathology and neuropathologic syndromes, the normal relationship between physiology and emotion is typically disrupted and accompanied by altered interoception (Khalsa et al., 2018). A particularly compelling example is somatization, in which abnormal interoception can occur at multiple points in the pathway from peripheral sense organs to central representation. In contrast, psychological experience may become uncoupled from the physiological/behavioral expression of a specific emotion in neurologic disorders such as pseudobulbar affect. Similarly, in parasomnias, dissociation between different elements of full wakefulness sometimes allow emotional behaviors to be expressed automatically. Lastly, symptoms of neuropsychiatric disorders can reflect the interoceptive awareness of an abnormal physiological state (as in panic attack) or of a normal physiological response (e.g., grief) activated under inappropriate circumstances (e.g., major depressive episode).
Interoception plays a key role in each of the above normal and abnormal experiences of physiological emotion. The ways in which these experiences are labeled and communicated verbally can also vary dramatically among different languages and cultures. In the English language, some expressions of physiological feelings are unambiguous (hunger, thirst, pain, temperature). However some interoceptive feelings are extremely subtle, their description may rely upon metaphor or analogy (e.g., “butterflies in the stomach”), and many are difficult to map onto specific physiological events. It is likely that some of these reflect the constrained awareness and integration of bodily signals that nevertheless influence feelings, emotions and moods (Craig, 2002, Craig, 2009, Damasio and Carvalho, 2013). One goal of the Affectome Project's linguistic approach is to identify how language may reveal experiential feelings and associated physiological processes meriting further study.
2. Theories of emotion and physiology
When asked to define emotions, most researchers will include physiological terms. However, accounts disagree on the importance and nature of physiological changes associated with different emotions. Whether or not emotions are characterized by unique physiological changes (i.e., emotion-specific physiological response patterns) has been fervidly debated since the 19th century (e.g., James, 1884, Lange, 1922 vs. Bard, 1928, Cannon, 1928). Currently the predominant opinion is that somatovisceral and central nervous responses associated with an emotion serve to prepare situationally adaptive behavioral responses (e.g., Ohman and Mineka, 2001, Panksepp, 2000).
According to theorists endorsing the arousal concept (e.g., Duffy, 1972), emotion-specific response profiles do not exist. Instead, behavior and related psychological processes are suggested to be based on a unidimensional activation concept. Whereas it should be possible to measure psychological activation (i.e., emotional arousal) by using any single physiological variable (see, however, Duffy, 1972, for legitimate exceptions [e.g., during voluntary suppression of activity in specific muscles]), it would not be possible to draw qualitative inferences about the kind of predominating emotion. However, correlations between different psychophysiological measurements are often low or even nonexistent (see Fahrenberg and Foerster, 1982, Stemmler, 1992), thus questioning the arousal concept. Investigations of the specificity versus aspecificity of autonomic correlates of particular emotions have a rich history that is reviewed in detail in Section 5.
Other dimensionalists (e.g., Barrett, 2006, Bradley et al., 1992) do not assume fine-grained emotion-specific physiological response patterns either, but propose instead two (or three) strategic emotion dimensions (e.g., appetitive versus aversive action dispositions) that are assumed to be closely linked with physiological responses. Commonly, these emotion dimensions are related to activity in specific subcortical networks (appetitive system: e.g., nucleus accumbens, mesolimbic dopaminergic system; aversive system: e.g., amygdala; see also Davis, 1989, LeDoux, 2000). In addition, authors in this tradition admit the existence of so-called tactical aspects of emotions that consider contextual conditions during goal pursuit. Both strategic and tactical aspects of emotions would be supported by dynamic physiological response profiles well explained via the defense-cascade model (Bradley and Lang, 2007). Similar thoughts are at the basis of theoretical models proposed by Davidson (Davidson et al., 2000) and Gray (Gray and Mc Naughton, 2000).
In contrast, authors in the tradition of basic emotions (e.g., Ekman, 1992a) postulate the existence of a limited number of distinct emotions that are characterized by specific (i.e., qualitatively different) central nervous and somatovisceral response profiles. Some (e.g., Levenson, 2003) admit the possibility that certain basic emotions (e.g., joy) are characterized by only low or even no physiological specificity at all, because they entail no need for immediate survival-relevant responses. Basically, emotion-specific bodily response patterns are thought to physiologically prepare the organism for specific emotion-related actions, prioritizing response mobilization. Fear, for instance, initiates preparation of a flight response, whereas anger prepares the body to fight. To date, however, there is limited evidence for emotion-specificity in physiological responses, although there remains much interest in this question and its implications (e.g., Ax, 1953, Funkenstein, 1955, Kreibig, 2010, Stemmler, 1992). Whereas – taken separately – several studies seem to yield convincing evidence for the existence of emotion specificity, an overall comparison of the results stemming from different experiments shows that these are, on the whole, inconsistent (Cacioppo et al., 1997). To overcome this lack of consistency, Stemmler's (1992) component process model of somatovisceral organization posits simultaneous influences of emotion-relevant and contextual effects on human physiology. This shifts the idea of absolute emotion specificity from basic emotions (Ax, 1953, Ekman, 1992a, Lange, 1922) to context-deviation specificity. A careful control or examination of contextual influences would thus be an indispensable requirement for the identification of emotion-specific physiological changes.
A related problem for proposed emotion specificity in physiological responses is that even when confronted with the same overall experimental situation, different individuals might engage in divergent cognitive evaluations depending on prior experience and respective goals. Different evaluations of the same situation would – according to appraisal theorists (e.g., Lazarus, 1966) – lead to distinct physiological responses and feeling states. Thus, instead of manipulating a situation with respect to particular overall target emotions, various researchers engaged in a systematic manipulation of specific appraisal outcomes and studied their effects on physiology (e.g., Aue et al., 2007, Lazarus and Alfert, 1964, Pecchinenda and Smith, 1996, Tomaka et al., 1997).
Contemporary models increasingly consider emotion-associated central nervous response patterns (see also Section 4). Historically, the limbic system was conceived as an essential component of the emotional brain (e.g., Maclean, 1952, Papez, 1937). More recently, a largely subcortical neural network with the amygdala as a key region has been suggested to play a central role in the elicitation of fear (LeDoux, 2000, Ohman and Mineka, 2001; see also Adolphs et al., 1995, LeDoux and Brown, 2017). The amygdala projects to sensory cortices and both cortical and subcortical regions, including the hippocampus, to influence perception, attention, and memory (Armony and Dolan, 2002). In addition, via its connectivity with the thalamus and brain stem, the amygdala can initiate defensive responses, even before the cortical processing of detailed sensory information about threat (LeDoux, 2002). Because phobias persist despite the explicit knowledge that a feared object is harmless, the fear module has been proposed to be “impenetrable to conscious cognitive control” (Ohman and Mineka, 2001, p. 515).
The ventromedial prefrontal cortex (vmPFC) is suggested to support emotional experience through body–brain interactions. According to the somatic marker hypothesis (Damasio, 1996), human decisions are guided by anticipation of the affective consequences of specific actions, based on prior experience. This link between actions and anticipated feeling consequences relies on the vmPFC and is encoded as somatovisceral feelings through evoked physiological responses (the ‘body loop’), and their simulated representation within the brain in somatosensory and, especially, interoceptive (insular) cortices (the “as if loop”). Once established, somatic markers may be evoked predictively to weigh outcomes of different actions, thereby biasing their choices toward the best alternative (i.e., the one with the most desirable affective consequence). In this context, somatic markers, especially the “as if loop” are conceptually similar to interoceptive predictions (priors) envisaged by Bayesian theorists (Barrett and Simmons, 2015, Seth and Friston, 2016).
Recent Bayesian theories of emotion, including the Embodied Predictive Interoception Coding (EPIC) model (Barrett and Simmons, 2015; see also Seth, 2013, Seth and Friston, 2016) emphasize the role of cortex, notably the insula, in emotion generation. Transitional cortical architecture is invoked to support Bayesian representations and computations: pyramidal projection neurons in deep (V and VI) layers of agranular cortex (in peri-genual anterior cingulate cortex, vmPFC, anterior insula) generate predictions about physiological states (and simulations of their interoceptive corollaries). These predictions are sent, via cortico-cortical projections, to dysgranular posterior and mid-insular (primary interoceptive) cortex. Here, visceromotor simulations are compared to afferent interoceptive inputs from the thalamus within a rudimentary layer IV (absent in agranular cortices) which, in turn, generate prediction errors that are sent back to agranular cortices to refine interoceptive predictions (Barrett and Simmons, 2015). Iteration of this process allows interoceptive feelings to reflect changing physiological conditions yet, if prediction error is small, what is experienced as interoception may in fact reflect the visceromotor cortex's simulation of rather than afferent representation of the body's physiological state (Barrett and Simmons, 2015).
Cerebral asymmetry underpins theoretical proposals, including the hypothesized right lateralization of emotional processes (Schwartz et al., 1975), the valence hypothesis [left lateralization of positive emotions, right lateralization of negative emotions (Gur et al., 1994)], and the action tendency hypothesis [left lateralization of emotions associated with approach, right lateralization of emotions associated with withdrawal – in particular within prefrontal areas (Harmon-Jones and Allen, 1998)]. Meta-analysis of this topic (Wager et al., 2003), however, does not fully support any of these emotion-lateralization theories. Left-dominant activation of basal ganglia (including amygdala) during withdrawal-related negative emotions, for instance, contrasts with all theories mentioned above. What is more, the absence of reliable gender differences in cerebral lateralization of emotions needs additional consideration.
In summary, emotions are integrative phenomena that cannot be unequivocally captured by any of the above formulations, though most highlight interoceptive physiological contributions. Alternatives to the somatovisceral origin of emotional experience in the James–Lange theory typically propose central mechanisms for secondarily generating interoceptive experience. For example: (1) interoception-like experiences emanate from representations of the body in viscerosensory and somatosensory cortices via an “as-if” pathway (Damasio, 1996); (2) output of central generators of emotion (e.g., amygdala) may evoke peripheral events that are then perceived via interoception (LeDoux, 2000); or (3) cortically generated predictions of peripheral states generate subjective interoceptive experience after being compared with current afferent information from the periphery (Barrett and Simmons, 2015). The following section will consider how and why human emotion, including its interoceptive components, may have evolved.
3. Evolutionary considerations
According to Darwin's theory of evolution, variations in traits not aiding in perpetuation of a species will be selected against, and ultimately eliminated from the gene pool, whereas characteristics proving beneficial are likely to be passed down to subsequent generations (Darwin, 1859). Though most often considered in terms of physical traits and behaviors, Darwin's theory of evolution also applies to emotions (Darwin, 1872). Emotions serve an evolutionary purpose: their experience and expression enable adaptive reactions to survival- and reproduction-related threats and opportunities (Ekman, 1992b, Nesse, 1990, Ohman, 1986, Tooby and Cosmides, 1992). Darwin (1872) proposed the “principle of serviceable associated habits” asserting human emotional expression originates as an evolutionarily beneficial mode of intra-species communication of intentions.
To function effectively as a means of communication, emotions must be innate and culturally universal (Darwin, 1872). Notably, the way infants express emotions is consistent with adults (Izard et al., 1980). Likewise, mannerisms used by congenitally blind individuals to express emotions like happiness and pride do not differ from those in sighted individuals, despite never visually witnessing these behaviors (Galati et al., 1997, Matsumoto and Willingham, 2009). Across cultures, people experience and understand emotions similarly, even those without connection to the modern world (Ekman, 1984, Ekman, 1993, Ekman et al., 1969, Elfenbein and Ambady, 2002, Elfenbein and Ambady, 2003, Izard, 1971, Izard, 1994). Arguably these illustrations exist because emotional expression is not learned, but is rather genetically encoded by evolutionary processes.
In addition, because of a shared evolutionary history, primates’ emotional expressions mirror those exhibited by humans. Specifically, facial structures of humans and chimpanzees have functional similarities suggesting the capability to create similar facial expressions (Waller et al., 2006). Indeed, during positive interactions, non-human primates exhibit a “silent bared-teeth display” (reminiscent of a human smile) and “relaxed open-mouth display” (mimicking a human laugh) (Preuschoft, 1992). Likewise, humans share expressions of embarrassment with other mammals: face touching paired with downward head movements and eye gaze (Keltner, 1995, Keltner and Anderson, 2000, Keltner and Buswell, 1997). The universality of emotional behavior and expression supports the argument that emotion developed evolutionarily.
Evolutionary behaviors are those necessary for survival of a species. Eating and mating are essential behaviors that allow a species to survive. Increased dopamine (DA) release in mesolimbic reward circuitry is associated with approach behaviors that promote survival. For example, the presence of food triggers an increase in extracellular DA in the nucleus accumbens (Hernandez and Hoebel, 1988, Yoshida et al., 1992) and this increase in DA provides a chemical reward, encouraging future eating behavior. Similarly, increased DA is associated with sexual activity (Damsma et al., 1992, Pfaus et al., 1995) and, again, the pleasurable experience of increased DA triggers reward motivation, increasing potential for the behaviors to occur again.
Importantly, however, as an animal learns which stimuli predict the availability of reward, DA release advances in time to occur when such predictive signs appear rather than at reward delivery – an adaptation that facilitates reinforcement-based learning and promotes reward motivation and appetitive behaviors (Aggarwal et al., 2012, Glimcher, 2011). Reward motivation is particularly relevant to emotion. Specifically, positive high-approach motivating emotions like desire are associated with increased DA in the mesolimbic dopaminergic reward circuit, thereby promoting desire or wanting, and activating advancement toward a particular goal (Robinson et al., 2005). Research also suggests DA is actively involved in aversive motivation like escape behavior (Faure et al., 2008, Salamone, 1992). DA in the nucleus accumbens thus modulates the neocortical and subcortical areas involved in affective processes, influencing both approach and avoidance motor activity (Salamone, 1992). This “wanting” (incentive salience) system, which relies on the mesolimbic DA circuit, is distinct from neural systems that generate pleasure (“liking”) which relies on more diffuse neural loci activated by a variety of neurotransmitters (e.g., neuropeptides) but not DA (Berridge and Robinson, 2016).
Approach and withdrawal motivating emotions are associated with evolutionarily important outcomes (Darwin, 1872, Plutchik, 1980). Fear triggers sympathetic arousal, which promotes muscle activation and subsequent escape from predators (Frijda, 1986, Frijda, 2009, Frijda et al., 1989). Desire triggers approach motivation toward a desired object. Anger results in renewed efforts to keep desired objects, be they food, sexual partners, or otherwise (Plutchik, 1980). Increases in DA motivate approach (wanting) behavior toward eating, mating, protection of self and kin and other behaviors necessary for perpetuation of the species. The pleasure (liking) associated these beneficial behaviors positively reinforces them, ensuring their reoccurrence; resulting in both wanting and liking systems being transmitted to offspring and descendants.
Emotions associated with evolutionarily important outcomes are high in motivation to approach or withdraw. Emotions high in motivation, to approach – e.g., desire and anger – or withdraw – e.g., fear – both result in narrowed cognitive breadth (Harmon-Jones et al., 2013). Narrowed breadth associated with these emotions may make goals to approach or avoid more attainable. That is, by focusing attention on the target (e.g., a desired object or escape route), the individual is less likely to be distracted by irrelevant stimuli that may prevent successful goal completion (Harmon-Jones et al., 2017). If emotion and the associated cognitive narrowing are in fact evolutionarily developed processes, they should be well integrated. When a cognitive and an emotional process are integrated, they develop the capacity to influence each other bidirectionally (Simon, 1967). Indeed, the relationship between high approach motivating desire and attention is bidirectional, such that simply narrowing one's attentional scope increases approach motivation for desirable desserts (Kotynski and Demaree, 2017).
The adaptive significance for survival and reproduction is clear for interoceptive sensations associated with desire for and fulfillment of basic drives. Afferent information from interoception undoubtedly supports these basic drives such as osmoreception for thirst or metaboreception for hunger. Similarly, interoceptive signals indicating physiological deviation (prediction error) from homeostatic limits have obvious adaptive significance (e.g., thermoregulation) (Barrett and Simmons, 2015, Seth and Friston, 2016). However, the adaptive significance, in humans, of being able to sense and form predictive models of other varieties of interoceptive signals, for example of muscle tone being increased by anger or decreased during fear (freezing), may lie in the evolution of complex social behavior. As suggested by the somatic marker hypothesis (Damasio and Carvalho, 2013, Damasio, 1996), interoceptive information may bias behavior in the direction of the most advantageous outcome; in evolutionary terms being the behavior most likely to transmit ones genetic information to subsequent generations and survive long enough to do so. Here, subtler internal signals, both predictive and afferent from the PNS, may help individuals negotiate complex social behaviors (Otten et al., 2017), for example, those within a social dominance hierarchy.
In conclusion, though Darwin's theory of evolution is frequently considered in the context of physical traits or behaviors, it also explains humans’ adapted capacity for emotion (Darwin, 1872). The experience and expression of emotions are universal: similar in other mammals as well as humans of all ages across cultures. Emotion may have evolved to make evolutionarily beneficial actions like eating, mating, and protecting oneself more motivating, rewarding, and likely to occur. Emotions are therefore traits naturally selected for as they provide greater chances of perpetuation of the human species. The next section will begin consideration of CNS representations of emotional perception and expression.
4. Central representation of emotions
4.1. Neuroimaging
Representations of emotions in the CNS have been investigated using a variety of frameworks (Murphy et al., 2003). Single-system models of emotion – such as the limbic system (Maclean, 1952) and the right-hemisphere models (Adolphs et al., 1996, Borod et al., 1998, Borod et al., 2001, Heller and Nitschke, 1997) posit that one neurological system drives the experience and expression of all emotions. Dual-system models examine potential dissociable dichotomies such as valence or approach and withdrawal (Carver and Harmon-Jones, 2009, Davidson, 1984, Davidson, 1998, Davidson et al., 1990, Feldman Barrett and Russell, 1999, Lang et al., 1997, Posner et al., 2005, Russell, 2003, Schmidt and Schulkin, 2000). Multi-system models conceptualize emotions and their correlates as discrete (Lindquist et al., 2012). Recent advancements in functional neuroimaging have afforded the opportunity to examine the neurophysiological basis of emotions in vivo, allowing for greater delineation of the central mechanisms supporting affective processes.
Single-system models have largely been absorbed into more complex theories. For example, the limbic system hypothesis, proposed by (Maclean, 1952), holds that subcortical structures constitute one system that is responsible for the experience and expression of all emotion. In a review, Murphy et al. (2003) found almost no substantial support for this view. However, involvement of the classical components of the limbic-system in emotional processing and experiences is undeniable. Similarly, the right hemisphere hypothesis, which views the right hemisphere as more emotional in general (Adolphs et al., 1996, Borod et al., 1998, Borod et al., 2001, Heller and Nitschke, 1997), has received mixed support (Murphy et al., 2003). Other versions of the hypothesis are better supported, but are still inconclusive. Valence asymmetry (the right hemisphere is responsible for negative emotions and the left for positive) has also received only limited support (Lindquist et al., 2012, Lindquist et al., 2016, Murphy et al., 2003). Similarly, many failed to find support for a hemispheric dichotomy between emotion perception versus expression (Adolphs et al., 1996, Borod et al., 1998, Borod et al., 2001, Heller and Nitschke, 1997). Recent studies suggest that emotion may be functionally lateralized in specific structures, rather than the entire cerebrum (Beraha et al., 2012, Kober et al., 2008, Lindquist et al., 2012, Lindquist et al., 2016, Murphy et al., 2003).
Instead, some researchers advocate an approach-avoidance asymmetry model (Carver and Harmon-Jones, 2009, Carver et al., 2000, Davidson, 1998, Davidson et al., 1990, Lang et al., 1997, Schmidt and Schulkin, 2000). Murphy et al. (2003) found left-lateralization for approach emotions and bilateral symmetry for withdrawal emotions, but only in the cases of happiness and sadness and in anterior brain regions. However, individual differences exist as to whether anger is an approach or avoidance emotion (Carver and Harmon-Jones, 2009, Harmon-Jones, 2003). Thus, the approach-avoidance distinction holds for emotions such as happiness, compassion, and fear, but its limited support may be due to individual differences or varying induction procedures.
Multi-system models of emotion usually encompass basic emotions, which can be discrete or continuous. Basic emotions are cross-cultural, intuitively understood, and named across languages (Ekman, 1999). Common lists include anger, anxiety, disgust, fear, happiness, and sadness. This model has also received both support (Celeghin et al., 2017, Vytal and Hamann, 2010) and criticism, with strong, consistent evidence for specialized fear processing in the amygdala and disgust processing in the insula and globus pallidus (Murphy et al., 2003), but other emotions show mixed neural representation (for review, please see Lindquist et al., 2012). Moreover, basic emotions still vary along dimensions of valence and arousal (Posner et al., 2005). Reviews have consistently concluded that single-system models are too simple, dual-system models are too coarse, and that there may not be completely distinct neural profiles for each basic emotion (Lindquist et al., 2012, Lindquist et al., 2016, Lindquist et al., 2012, Murphy et al., 2003, Touroutoglou et al., 2015). Rather, emotions likely arise from interacting neural components, and may be differentiated by subcortical and cortical networks based on their evolutionary functions (Barrett and Satpute, 2013, Berntson et al., 2007, Citron et al., 2014, Lindquist et al., 2012, Lindquist et al., 2016, Pessoa and Adolphs, 2010).
Perhaps one of the most formidable advancements in affective neuroscience has been the leveraging of big data resources to test theories of emotion and to develop convergent evidence regarding underlying neural representations. For example, application of activation likelihood estimation (ALE), a robust meta-analytic technique that statistically tests for convergence of activation patterns in functional neuroimaging data, reveals consistent, dissociable activation of distinct areas for happiness, sadness, anger, fear, and disgust (Vytal and Hamann, 2010). Emotion specificity was confirmed by comparing emotion categories against each other. Kirby and Robinson (2017) replicated Vytal and Hamann's (2010) work using the BrainMap database and ten times the number of studies. Although evidence of distinct networks was found for some emotions, there were also brain regions that were consistently activated across all emotions, suggesting a multi-system model (Kirby and Robinson, 2017). Their results indicated more distributed subcortical- and prefrontal-based networks also observed elsewhere (Lindquist et al., 2012, Lindquist et al., 2016). Their findings were strongly right-lateralized and had strong contributions from classical limbic structures. They also found all emotions associated with activity in the right dorsolateral prefrontal cortex, suggesting it might serve as an emotional processing hub (Eickhoff et al., 2016, Kirby and Robinson, 2017, Robinson et al., 2010). Together, these meta-analyses seem to support, in part, elements of the right hemisphere hypothesis and provide limited support for a multi-system approach.
Other meta-analytic approaches have spurred new insights into affective processing. For example, Lindquist et al. (2012) framed their investigation in terms of locationism versus constructionism. Locationism ascribes specialized functions to discrete neural areas, whereas a constructionist view sees emotion as an emergent property of existing cognitive processes. They tested the locationist view in several regions mentioned in previous reviews, finding a lack of consistency and specificity across studies. They concluded that locationism is overly reductionist, and instead favored constructionism. Kober et al. (2008) eschewed theory and opted for a data-driven approach instead. They found six functional groups related to emotion, which they declined to label. Depending on meta-analytic method, idiosyncrasies of individual studies, particularly those with high numbers of participants, could potentially bias the meta-analyses and fail to reveal the true nature of the common activation across neuroimaging studies on a particular topic. Furthermore, the utility of Kober et al.’s (2008) data-driven approach without reference to socially-accepted categories is unclear, especially in the context of clinical and other applied work regarding emotions.
It is important to note that meta-analyses of neuroimaging studies have strong limitations. First, most of the analyses rely to some degree on database infrastructure, which may represent only a fraction of neuroimaging papers (Derrfuss and Mar, 2009), may be biased toward specific psychological processes (i.e., cognition), or may not have sufficiently specific information to adequately test affective processes (e.g., unclear operational definitions; the Human Affectome Project seeks to address this issue). This becomes particularly important in the discussion of emotion because different modalities can elicit varying degrees of emotion, and to our knowledge, there is no database that accounts for the dimensionality of emotional experience (i.e., valence and arousal). Second, the variety of presentations (e.g., visual versus auditory) may have substantial effects on the neural processing streams, especially as they relate to emotion (Pessoa and Adolphs, 2010). Thus, the very strengths of meta-analytic methods (i.e., ignoring study specific features to identify convergent patterns of activations) can also be their weaknesses when applied to emotions. Toward overcoming these limitations, Müller et al. (2018) have recently developed a list of ten rules to improve the integrity and generalizability of neuroimaging meta-analyses.
Reviews and meta-analyses have treated interoception separately from studies emotion (e.g., Critchley and Harrison, 2013, Schulz, 2016) although findings of recent investigations have converged (Adolfi et al., 2017, Craig, 2010, Critchley and Garfinkel, 2017, Singer et al., 2009). One recent meta-analysis showed convergence of areas linked to interoception, emotion regulation, and social cognition in the right anterior insula and neighboring lateral frontal areas, as well as the amygdala and basal ganglia (Adolfi et al., 2017; see also Singer et al., 2009). There is some evidence showing that the insula shows a postero-anterior gradient in successive stages of interoceptive processing (Craig, 2009, Namkung et al., 2017). Notably the agranular cortices (e.g., peri-genual anterior cingulate cortex, vmPFC, anterior insula) are hypothesized to generate predictions of interoceptive feelings that can, in turn, predict the physiological consequences of emotional states (Barrett and Simmons, 2015, Seth and Friston, 2016).
Big data approaches have provided, to date, the most comprehensive survey of CNS emotion representation. However, they rely heavily on the models of emotion used by researchers which may bias coding and testing of results in later meta-analyses. Additionally, procedural differences across studies abound. For example, emotion elicitation practices vary widely. They may activate any number of neural networks and cognitive processes. Until we have a large, comprehensive database with strong operational definitions, it will be difficult to conclusively understand the CNS representation of various emotions. Further compounding the issue, small brain structures such as the amygdala, can be particularly difficult to image. Choices in pre-processing steps such as spatial smoothing (Fransson et al., 2002) and motion correction (Johnstone et al., 2006) may change observed activation patterns. Inconsistency across studies could also be due to the use of different selection criteria and meta-analytic techniques, as reviewed in Kirby and Robinson (2017). In order to increase our understanding of the central representation of emotion, we need large-scale functional neuroimaging studies that encompass a variety of affective tasks.
Whereas fMRI studies provide cutting-edge spatial resolution for investigating emotion representation, they provide poor temporal resolution. Because emotions are a dynamic, unfolding process in response to triggering events, converging operations are needed to examine them more completely. Electrophysiological markers of emotion do just that: ERPs derived from encephalography (EEG) reflect neurophysiological responses immediately following triggering by an emotional stimulus. While MEG also has a similar temporal resolution to EEG and better spatial localization, it is limited in only being able to measure brain activity from sulci (and not cortical gyri), and little research focusing on MEG indices of emotion has been implemented to date. Therefore, we focus on ERP indices of emotion here.
4.2. Electrophysiological and peripheral measures of emotion
Before techniques of functional neuroimaging became widely available in the 1990s, classical psychophysiology of emotion had already created an immense body of knowledge on electrophysiological and peripheral measures of emotional experience as well as overt and covert emotional expression. Such measures, including time- and frequency-domain EEG, cardiovascular indices, electrodermal responses and facial electromyography including the startle response, are described in detail in four successive volumes since 1990 of Cacioppo, Tassinary and Berntson's Handbook of Psychophysiology (Cacioppo et al., 2017). Especially informative have been indices that directly reflect responses of the autonomic nervous system such as skin conductance (Critchley, 2002, Dawson et al., 2007) and high frequency heart rate variability, also termed respiratory sinus arrhythmia or RSA (Balzarotti et al., 2017, Electrophysiology, 1996), as indices of sympathetic and parasympathetic activity respectively. This is consistent with the notion that RSA reflects self-regulatory ability (e.g., Thayer and Lane, 2000). While historically sympathetic and parasympathetic indices have typically been examined independently, the importance of examining the dynamic interplay between sympathetic and parasympathetic regulation (such as balance or co-activation/co-inhibition) has been highlighted more recently as important for a comprehensive understanding of autonomic regulation, as these indices are not necessarily reciprocally controlled (Berntson et al., 1991, Sunagawa et al., 1998). Of similar importance are HPA axis (cortisol) and other biochemical stress biomarkers (e.g., Strahler et al., 2017). Peripheral and electrophysiological measures are of continued importance in understanding emotional behavior and have been exceedingly successful in combination with neuroimaging paradigms (e.g., Critchley and Harrison, 2013, Milad and Quirk, 2012). Some more recently adopted techniques for studying emotion in healthy subjects include measures of inflammation (see Section 9), genetic markers (e.g., Jonassen and Landro, 2014), diverse forms of experimental neurostimulation (e.g., Busch et al., 2013, Kuo and Nitsche, 2015) and complex emotion recognition algorithms (Mehta et al., 2018). Details of these diverse methodologies are beyond the scope of this review but well described in the above-cited and similar sources. Here we will focus on findings from EEG time-domain measures – the method most often used to examine brain responses to emotional stimuli with high temporal resolution. Frequency domain EEG measures of emotion are briefly considered but not the emerging field of EEG source localization that is also increasingly used in studies of emotion (see Pourtois et al., 2008).
4.2.1. Electrophysiological indices of emotion
Event-related potentials (ERPs) are time-domain scalp-recorded brain potentials that can provide temporally precise information regarding the processing of affective stimuli through the examination of the amplitudes and latencies of ERP components (Rugg and Coles, 1995). ERPs reflect neural changes in the milliseconds range that reflect rapid processing of affective stimuli, including perception, attending and orienting responses, which are critical aspects of emotional responding (Olofsson et al., 2008). Their high temporal precision allows for a detailed examination of the time course of an emotional response, providing greater ability to disentangle key processes. Several ERP components have been identified that are particularly relevant to the processing of motivationally salient information. ERP components of emotion processing are typically elicited in experimental paradigms involving viewing of emotional pictures or faces, sometimes including emotion regulation instructions, or in tasks involving anticipation of reward or reward related feedback. While these components do not map onto discrete emotions, they reflect mechanisms underlying processing of emotional information, such as motivational value.
The late positive potential (LPP) and early posterior negativity (EPN) are two components that have been examined in the context of neural processing of visual emotional stimuli. The EPN has been identified as the first cortical ERP component reflecting the facilitated processing of emotional stimuli at the early perceptual level (Schupp et al., 2004). The EPN develops at around 150 ms and is maximal between 250 and 300 ms after picture onset. The amplitude of the EPN is most pronounced for affective stimuli of high evolutionary significance, such as erotic images or pictures of mutilations (Schupp et al., 2004). Source analysis of the EPN amplitude identified a widespread network of temporo-parieto-occipital areas implicated in visual information processing (Junghofer et al., 2001). Furthermore, a recent fMRI study using rapid visual picture presentation (Junghofer et al., 2002) revealed increased activations by emotional pictures in occipital (occipital, lingual and fusiform gyrus, cuneus, calcarine), temporal (superior, mid- and inferior-temporal gyrus), and parietal (inferior and superior parietal, angular, supramarginal gyrus, precuneus) structures. The LPP is a positive deflection of the ERP signal most apparent around 400–600 ms following an emotional stimulus that is thought to reflect motivated attention to things of emotional significance to the individual (Brown et al., 2012, Cuthbert et al., 2000, Hajcak et al., 2010). The LPP is considered to be a useful neurophysiological measure for studying emotion and emotion regulation across the life span (Hajcak et al., 2010), and it has been repeatedly shown to be enhanced (increased in magnitude) in response to emotional (relative to neutral) stimuli (Hajcak et al., 2010). The LPP is generated in occipital and parietal cortices (Sabatinelli et al., 2007), which both receive projections from the amygdala, a region critical for emotional processing. Importantly, the LPP is influenced by directed instructions to utilize specific emotion regulation strategies, such that its amplitude is reduced following emotion regulation strategies aimed at reducing an emotional response (Foti et al., 2015, Hajcak and Nieuwenhuis, 2006).
Several ERP indices associated with reward anticipation and receipt have also been identified. The feedback negativity (FN) is a negative deflection of the ERP signal that peaks around 250 ms following the receipt of feedback or reward. The FN is considered to be an index of the function of a performance monitoring/evaluative system that rapidly assesses the motivational salience of positive and negative environmental feedback (Bress et al., 2013, Bress et al., 2012). The FN is thought to relate to flexible selection of actions aimed at pursuing rewards and is part of a reinforcement learning system used to adjust subsequent behavior (Bress et al., 2012). The FN is correlated with measures of reward sensitivity and is thought to reflect a binary evaluation of feedback as unfavorable or favorable (loss vs. gains) (Hajcak et al., 2007, Holroyd et al., 2003). Neuroimaging and source localization studies have shown that the FN is generated primarily by the anterior cingulate cortex, an important hub for integrating cognitive and affective processing that is also involved in emotion regulation and flexible responding (Foti et al., 2015). Concurrent ERP and fMRI studies have demonstrated that the FN amplitude is also associated with activity in the ventral striatum (Foti et al., 2011), another key area for reward processing (Liu et al., 2011).
The cue-P3 and the contingent negative variation (CNV) have been associated with aspects of reward anticipation (Goldstein et al., 2006, Pfabigan et al., 2014), where the cue-P3 reflects aspects of salience and attention during the reward anticipation process, and the CNV reflects more aspects of cognitive effort involved in the anticipation process. In the context of a reward task, the cue-P3 is a centroparietal positivity which emerges between 300 and 600 ms after an anticipation cue and its amplitude increases as a function of reinforcer magnitude (Broyd et al., 2012, Goldstein et al., 2006). The cue-P3 has been associated with neural activity in reward related regions, including the ventral striatum (Pfabigan et al., 2014), and is thought to reflect variation in context updating in working memory, such as updating whether a potential gain or loss is at stake in a monetary guessing task (Bonala and Jansen, 2012). The CNV is a negative-going potential shift that is primarily associated with anticipatory attention and preparation of effortful processes (Falkenstein et al., 2003, Gomez et al., 2007) and has been assumed to reflect neural activity within the thalamo–cortico–striatal network (Fan et al., 2007, Macar and Vidal, 2003).
Frequency based measures of EEG activity have not been used as extensively to study emotional responses, but there has been increasing interest in the utility of these measures, particularly in the area of emotional responses to rewarding stimuli (see Glazer et al., 2018 for review). For example, measures of alpha, beta, delta, or gamma neural oscillations in response to rewarding stimuli may provide complementary information to what ERP components such as the FN may provide. These oscillatory changes in the EEG signal in response to affective stimuli are also referred to as event related spectral perturbations (Makeig et al., 2004). Alpha de-synchronization in response to affective pictures has also been related to the LPP and is thought to provide complementary information about sustained attention to emotionally salient stimuli (for review, see Uusberg et al., 2013).
In studies of interoception, it is possible to measure experimentally-induced ERPs to stimuli targeting distinct organ systems (e.g., respiratory evoked potentials). However, ERP studies of interoceptive sensations have focused on the ‘natural’ heartbeat evoked potential (HEP), a positive potential occurring maximally in frontocentral derivations 250–350 ms following the R-wave in the ECG (Pollatos and Schandry, 2004). The HEP probably reflects afferent signals originating in vascular sensors (e.g., baroreceptors) transmitted rostrally via the vagus and glossopharyngeal nerves and processed in the insular and anterior cingulate cortices. The amplitude of the HEP is significantly greater in those who can more accurately detect their own heartbeat (Pollatos and Schandry, 2004). To the extent that changes in emotion reflect increased arousal, the HEP may index emotional state (Luft and Bhattacharya, 2015) Additionally, as is the case for accuracy of interoceptive heartbeat detection (Herbert et al., 2007, Pollatos et al., 2007), the HEP may be related to state and trait measures of emotion (Fukushima et al., 2011). Interoceptive information, arrives at the cortex via spinal Lamina 1 fibers (Craig, 2016) or afferent components of the vagus nerve (Berthoud and Neuhuber, 2000) both of which are weakly myelinated and hence conduct relatively slowly. Could such interoceptive afferents influence ERPs that can occur within 150 ms of a perceived exteroceptive stimulus? Although HEPs occur within 250–350 ms of the R-wave (Pollatos and Schandry, 2004), it is possible that interoceptive effects on more rapid response potentials, such as the auditory N1 (at 140–170 ms), may result from predictions of physiological states that originate within the CNS itself (Babo-Rebelo et al., 2016, van Elk et al., 2014).
The following section will address interoception in greater detail along with discussion of three emotion-related subjective experiences, the first two of which, pain and disgust, are defined by their interoceptive features and the third of which, empathy, may be a human capacity uniquely dependent on interoception. First, however, the important linkages between the autonomic nervous system, interoception and emotion must be considered.
5. Autonomic nervous system and emotion
In the last century, influential theories have emphasized the role of the autonomic nervous system (ANS) in emotion. Bodily manifestations are often central to the experience of emotion, such as feeling a strong heartbeat when fearful. The bodily aspects of emotion are largely mediated by the ANS, a collection of nerve cells/fibers projecting from the spinal cord to the viscera (e.g., organs, glands, blood vessels, airways). The correlation between emotion and autonomic activity is believed to have an evolutionary function (Darwin, 1872). By distributing metabolic and other physiological resources, autonomic changes in visceral activity support emotional behaviors with survival value (e.g., approach reward, avoid threat; Levenson, 2003, Tomkins, 1962). ANS activity also influences sensory-perceptual processes that drive the experience of emotion (Craig, 2003a, Damasio and Carvalho, 2013, James, 1884). The diverse actions of the ANS are achieved by its complex organization wherein each ANS branch has afferent fibers that carry visceral information to the brain (bottom-up) and efferent fibers that regulate the viscera based on brain activity (top-down). This construction allows for feedback loops that are critical to homeostasis, stress, and emotional experience (Benarroch, 1993, Chwalisz et al., 1988). Despite its complexity and debated place in emotion theories (Friedman and Thayer, 2018), the role of the ANS in emotion can be summarized with two themes: specificity and causation.
5.1. Specificity
If the ANS supports behaviors unique to a specific emotion, then emotions should differ in their characteristic patterning of ANS activity. This notion of autonomic specificity has been widely debated in the dominant theories of emotion. Adding further complexity to the matter, views on the autonomic specificity of emotions also raise the related issue of the causal relationship between emotional experience and ANS function.
William James (James, 1884) and Carl Lange (Lange, 1885/1912Lange, /, 1912Lange, 1885/1912) were the first to formulate that emotions have differentiable ANS patterns. This model, that came to be known as the “James–Lange” theory of emotion, has scaffolded decades of research on basic emotions. Here, motivationally relevant stimuli first elicit autonomic and bodily changes, which then lead to emotional experience. For physiology to elicit specific feelings, emotions should have unique patterns of ANS responses. Biologically-oriented theories of emotion embrace autonomic specificity because reliable activation of pre-programmed physiological patterns is efficient and conducive to survival (Ekman, 1992b, Levenson, 2003, Tooby and Cosmides, 1990). The notion that basic emotions have some degree of autonomic specificity is supported by rich experimental work (for review, see Kreibig, 2010) and multi-measure studies using statistical classification (Christie and Friedman, 2004, Kragel and Labar, 2013, Stephens et al., 2010).
A case has also been made for a lack of autonomic specificity (Barrett, 2006). This concept is not new, however. Cannon (1927) directly challenged James by stating that emotional feelings and ANS responses are independent, and that autonomic responses are diffuse, rather than patterned. Later, influential cognitive theories of emotion de-emphasized the centrality of autonomic function in favor of top-down appraisals (Ellsworth, 2013, Quigley and Barrett, 2014). For example, the two-factor theory emphasizes global autonomic “arousal” that leads to experience through the filter of socially constructed appraisals (Schachter and Singer, 1962).
Cognitive theories and their evidence do not hold a monolithic view of ANS specificity. Across the last four decades, studies have shown that appraisals of stressors can lead to differentiable patterns of ANS activity (Scherer, 1984). Of note, appraisal-generated patterns of ANS activity appear broader than those of the discrete emotions suggested by James (1884) and others (e.g., Ekman, 1992b). These diverse findings inspire the question: does ANS activity map onto discrete categories or onto broader dimensions related to appraisals and motivational significance? Studies in the past few decades support a role for both, implying a gradient of autonomic specificity where autonomic differentiation exists at multiple levels (e.g., Christie and Friedman, 2004, Nyklíček et al., 1997, Witvliet and Vrana, 1995). Obfuscating the matter, the degree of ANS specificity appears subject to contextual factors, individual differences, and the method of emotion elicitation (Cacioppo et al., 2000, Stemmler, 2003). Taken together, the ANS specificity of emotion conceivably exists on a continuum with the degree of specificity being determined by a constellation of neuropsychological and environmental variables that change over time. Such complexity in the coordination between ANS physiology and emotional state is possible and quantifiable in a dynamical systems model of emotion (Lewis, 2005, Thayer and Friedman, 1997). Here, the ANS rapidly “explores” many affective states in the state-space, and strong attractors (i.e., basic emotions) can pull the organism into a more programmed ANS pattern if multiple conditions are satisfied.
5.2. Causation
Does the ANS cause emotional experience in a bottom-up fashion, or are autonomic responses generated by top-down factors, such as affective experience and/or brain activity? Understanding the causal connections between these components is critical for a precise definition of emotion. For James and his successors in discrete emotion theory, specific physiological patterns cause emotional experience through autonomic afference to the brain (Friedman, 2010). These afferent pathways have empirical support, such that autonomic afference from the viscera influences perceptual activity in the brain (Damasio et al., 2000, Park et al., 2014). Afferent influences on emotional experience likely occur both inside and outside conscious awareness (Vuilleumier, 2005).
Cognitive appraisal models of emotion adopt a view opposite of the bottom-up perspective: cognitions about motivationally salient events cause ANS responses in a top-down manner (Blascovich and Tomaka, 1996, Lazarus and Folkman, 1984). The evidence used to support the top-down view is vast but mostly correlational. Experimental designs are rarely employed to substantiate that cognition causes emotion-related ANS responses. The top-down view nevertheless remains strong in cognitive neuroscience where the expression and regulation of emotion is putatively grounded in the frontal lobe regulation of subcortical emotion regions, including autonomic source nuclei (Roy et al., 2012, Wager et al., 2009). As with autonomic specificity, an “either–or” approach to reconciling the bottom-up vs. top-down issue is too simplistic. Functional models of central-autonomic relationships have emphasized both autonomic afference and efference as well as their dynamic interplay in emotion (Craig, 2002, Damasio, 1994, Thayer and Lane, 2000). Direct empirical evidence for afferent–efferent interaction is surprisingly sparse, however. Neuroimaging findings revealing relations between brain activity and ANS responses during emotion are often interpreted in terms of the brain causing autonomic responses – although the direction of causation here is suggestive due to correlational designs (Macefield et al., 2013, Roy et al., 2012, Thayer et al., 2012).
Empirical support for the afferent determination of emotion may be lacking because this question is one of cause-and-effect, ultimately requiring the tight controls afforded by experimentation. While it is clear that top-down factors and the ANS interact in emotion, the dynamics of such interactions are relatively unknown. To this end, sophisticated mathematical approaches can prove useful in unraveling the complex nonlinear dynamics that characterize brain–body relationships (Lewis, 2005, Thayer and Lane, 2000). Leveraging modeling approaches with experimental techniques (e.g., pharmacological blockade and lesions) might clarify the causal role of ANS responses in emotion.