Where is emotional feeling felt in the body? An integrative review

Contemporary research on “embodied emotion” emphasizes the role of the body in emotional feeling. The evidence base on interoception, arguably the most prominent strand of embodied emotion research, places emphasis on the cardiac, respiratory and gastrointestinal systems. In turn, interoception has evidence-based links with improved emotion regulation. Despite the focus on separate bodily systems, it is unclear whether particular interoceptive locations play a greater role in emotional feeling and emotion regulation. Further, according to Gross’ “process model”, the sooner that regulation of an emotion occurs, the better; hence, it is additionally important to identify the first body areas to activate. These issues are investigated in a two-stage integrative review. The first stage was preliminary, giving an overview of the evidence base to highlight the distribution of measured body areas. This indicated that 86% of publications (n = 88) measured cardiac activity, 26% measured the respiratory system, and six percent the gastrointestinal system. Given the emphasis placed on all three systems in interoception theory and research on emotion, this suggests a dearth of comprehensive findings pertaining to feeling locations. The second stage investigated the core issues of where emotional feelings are felt in the body and time-related implications for regulation. This was based on ten texts, which together suggested that the head, throat and chest are the most consistently detected locations across and within numerous emotional contexts. Caution is required, however, since–among other reasons discussed–measurement was not time-restricted in these latter publications, and direct physiological measurement was found in only a minority of cases.


Introduction
With the exception of a period dominated by behaviorist reductionism, subjective emotional experience has been a focal point in psychology since its inception as a separate discipline (with Wundt's studies on introspection). Emotional experiences are routinely captured using self-report tools and objective measures such as cognitive tasks, behavioral observation and physiological measurement. As an advance on the introspective approach to examining feelings, William James emphasized the intersection between subjective emotional feeling and the objectively measurable features of emotion, leading to what is now referred to as "embodied emotion". According to James, emotion is the feeling of bodily changes that arise immediately from "perception of the exciting fact" [1]. For example, there is the distinction between running from a bear because we are afraid, and being afraid because we run. Running is the response to the "exciting fact" (i.e., a bear) and feeling the overall embodied response is the emotion. According to James, then, emotion is not a separate state that causes bodily responses; rather, emotion consists of bodily responses. Modern scientists such as Damasio [2] have revitalized and extended James' perspective, with articulation of an account involving afferent feedback from bodily associations, which has resulted in the significance of the body in the contemporary psychological study of emotion. In particular, Damasio's "somatic marker hypothesis" suggests that some bodily responses are the translation of emotional information from the autonomic nervous system to conscious awareness, acting as "biasing" bodily signals to mark response options that require full processing [2]. Somatic markers boost attention, thereby amplifying environmental features (including James' bear). Hence, such markers act as an interface between subjective experiences and objective events.
One prominent construct for understanding the interface between subjectivity and objectivity, and thereby consistent with the embodied view of emotion, is interoception. This has been defined as "the sense of the physiological condition of the body" [3], which makes a vital contribution to emotional experience [4]. Arguably the most prominent articulation of interoception in the literature is that provided by Garfinkel and colleagues [5]. This articulation is also clearly consistent with the embodied emotion view of James and Damasio, with an emphasis in all cases on objective bodily changes providing the foundation for subjective emotional experience. According to Garfinkel and colleagues, interoception is decomposable into three aspects [5]: • interoceptive sensitivity (the ability to accurately detect physiological changes, using objective detection measures), • interoceptive sensibility (the self-reported subjective sense of physiological changes, measurable by tools such as sub-scales of the Multidimensional Assessment of Interoceptive Awareness, or MAIA; [6]), • metacognitive awareness (the degree of correspondence between measurements of sensitivity and sensibility).
Originally, "interoception" pertained to only the viscera [7] whereas it is now understood to pertain to locations throughout the body, but especially the viscera [8]: the "visceroceptive" processes of "cardioception" (interoception of the heart), respiratory interoception (interoception of respiration rate, inspiratory resistance etc., regarding the activity of the lungs and surrounding structures) and "gastroception" (interoception of the gastrointestinal tract). In relation to research on interoception and emotion, specifically, the three visceroceptive domains (heart, lungs, gut) have to date been prominent focal points, including within clinical studies such as those involving the viscera and panic attacks (see [9]). For example, in an overview of the literature on emotion and interoception, Critchley and Garfinkel [10] emphasised the three visceral systems, and particularly the dominance of the heart in interoceptive research. The latter is likely due in part to the availability of non-invasive heart measurement (e.g., electrocardiogram, or ECG, and self-report). There is, for example, the "Schandry task", which involves counting one's heartbeats within varied time frames for comparison with objective heartbeat measurements [11]. Greater cardioceptive ability is measurable using this task, where such ability has been linked to subjective emotional experiences, such as increased strategies and "response-focused" strategies [33]. Antecedent-focussed regulation is an early four-stage modulation sequence, prior to the embodied emotional case of "multi-componential syndromes" where numerous body parts and systems become active, and with responsefocussed regulation occurring after a response is underway. In terms of cognitive strategies, at least, the antecedent-focussed approach appears to have the best outcomes [33], which is likely due to greater efficiency, given that such strategies are implemented at a point prior to the need for responding to a full-blown emotional episode requiring greater cognitive resourcing [34,35]. One such strategy involves the attentional deployment stage of the "process model": either attending to or redirecting one's attention from the early genesis of an emotion/situation. In terms of interoception and its role in emotion regulation, then, it may be that an antecedent approach (e.g., a training in interoceptive awareness), has the best chance of success, with early stage conscious awareness of physiological change during an emotional response being preferable to a later stage escalation of bodily activation, in order for there to be successful adaptation.
Further, there may be clinical applications regarding interoceptive contributions to emotion regulation. Such applications may involve bringing underlying emotionally relevant signals into awareness for conscious processing [36]. There is, for instance, growing evidence that emotion dysregulation is a significant factor in suicidality and, more generally, on the negative consequences of emotional disconnection in relation to interoceptive deficits in people experiencing suicidality. Conversely, the benefits of reconnection with embodied emotion are also indicated in relation to suicide prevention [37]. There are numerous studies showing the psychological and clinical value of interoception, such as when combining interoception and dispositional mindfulness (e.g., [38]), or with Mindful Awareness in Body-oriented Therapy (MABT) (e.g., [39]), involving techniques such as the "body scan" (a sequential focus on body areas), which has been shown to improve interoception [40]. Some studies have, however, suggested that focusing on the body may not always be positive, such as when detection of sensations by interoception exacerbates social anxiety [41]. However, this may be due more to preexisting cognitive influences pertaining to worry about the body, rather than a problem of excessive interoception (see, for instance, [42]).
One challenge for any form of emotion regulation is that emotional responses develop rapidly, even those involving potentially sensed activation of the body. The timeframes within which responses unfold are evident in studies such as Codispoti, Mazzetti & Bradley [43], in which an unmasked emotional stimulus (from the International Affective Picture System, IAPS; [44]), produced reliable arousal levels (assessed using self-report and physiological measures) even at 25ms presentation. Even in a masked condition, similarly early signs of activation resulted after an 80ms presentation time [43]. Further, in the unmasked condition, arousal did not increase significantly beyond 150ms post-onset; subjective ratings were provided within a limited window of 10-second post-stimulus onset. Hence, subjective ratings reflected underlying physiological changes, all undertaken within tight timescales, providing evidence of early stage sensing of emotional changes in the body. Relatedly, another study using the IAPS [45] indicated the mean arousal rating required to elicit emotionally relevant physiological changes, in this case heart rate variability. A score of > = 4.5 was required, out of a maximum of 5 on the Self-Assessment Mannikin (SAM; [46]). Such research indicates, then, the possibility of eliciting genuinely embodied emotional experiences using standardized tools, experiences that appear possible during the earliest stages of emotional responses. The awareness of these early stage changes could provide the basis for regulation of an otherwise potentially rapid non-adaptive escalation of emotional responding.
Overall, then, the links between subjective emotional feeling, emotion regulation and the body have been well explored in recent years, which includes the rising importance of interoception, most prominently articulated in the tripartite form developed by Garfinkel and colleagues [5], as well as other forms such as the multidimensional account of Mehling and colleagues [6]. Independently of any one articulation are the differing forms of interoceptive practice; i.e., different focal points in and across the body. Rather than the "body scan" of mindfulness or some other generic body focus, there may be additional regulatory value in focusing on a specific part or parts of the body. To our knowledge, however, whilst interoceptive research implies the importance of bodily location with the emphasis on the heart, the gut, and lungs, whether these or other locations/bodily systems play distinctive roles in emotional experience has yet to be explored. Further, whilst there is a general sense that the heart dominates interoception research-and perhaps embodied emotion research more generally-the extent to which this occurs is unclear. With the evidence base pointing to channel-specific interoceptive ability, there is little sense in limiting focus to the cardiac system. Another implication of cardiac dominance is that with current measures such as the Schandry task having been questioned regarding their reliability (e.g., [47]) cardioceptive dominance may severely inhibit progress in interoception research. Finally, it is unclear which-if any-bodily location merits more attention regarding activation in emotional experience and how location may in turn relate to optimal emotion regulation.

Aims
The primary focus of this review is the relationship between emotional feelings and body locations, including the heart, lungs and gut, which have to date been prominent focal points in the field of interoception. The review seeks to identify publications addressing body locations including, but not limited to, visceral locations, the awareness of which could play a role in effective emotion regulation, especially as an early-stage response strategy. A prima facie "gold standard" publication would be one where at least a majority of overall body locations implicated in the literature (including all three visceroceptive systems) are measured within tight time-scales following the presentation of emotional stimuli, with accompanying self-report measurement for those same body areas to establish corresponding subjective sensations, thereby grounding sensation in objective activity. There may also be grounds for including self-report only studies with innovative designs that are able to capture subjective reports of genuine bodily sensations even without physiological measurement. These may, for instance, draw on the findings cited previously regarding brief stimuli presentation timescales and required arousal levels for physiological activation. In this way, whilst the current review emphasises the tripartite model of Garfinkel and colleagues [5], as a prominent viewpoint, no absolute claims are asserted about its accuracy. This is particularly important, given that the construct of interoception remains under development (see, e.g., [48]). Such development is indeed required, given findings on a lack of correspondence between interoceptive sensitivity and sensibility; i.e., high scores in the former may not yield high scores in the latter [5]. Indeed, Murphy, Catmur, and Bird [48] suggest that the lack of correspondence may relate to the distinction between attention and accuracy; in other words, it matters as much what is being measured in addition to how. Outside of these debates, the review is committed to the view that subjective experiences of emotion are grounded in bodily changes-this is a more fundamental commitment, grounded in the original view of James [1] and that of Damasio [2].
Secondarily, the review seeks to provide an initial overview of which body locations are routinely measured in emotion research, in cases where subjective feelings of those locations could occur, at least in principle. This is to set the scene for the primary aim by identifying the extent to which the literature is dominated by certain body locations [10] to the exclusion of others, with the potential for biasing embodied emotion research via a self-reinforcing process of limiting measurement to those locations (such as when driven by attempts to replicate). At the same time, this can contextualize the situation for the three visceroceptive systems that appear to make important theoretical and applied somatic contributions to emotion, specifically. Whilst not a comprehensive treatment of "where emotional feeling is", measuring all three systems in a study would at least provide a reasonably well-rounded assessment of embodied emotion, as a reflection of the current evidence base on interoception-emotion. Since all three systems can be measured non-invasively and with no a priori obstacle to concurrent measurement it seems likely such studies have been undertaken, but the extent to which this has happened is an open question.
Against the backdrop provided by this secondary aim, the key research questions of this review are: (i) Where is emotional feeling reported to be located in the body? (ii) In particular, which part(s) of the body is/are active in the earliest stages of emotional feeling?

Methods
Given the wide-ranging nature of the subject matter, with emotional responses occurring across several levels-i.e., subjective sensation, behavioural responding, physiological responses-this review includes a relatively wide range of subject areas. However, with the focus being bodily sensations (i.e., subjective feelings), purely neuroscientific studies are not in scope, although physiological studies, more generally, were included provided they pertained to potentially sensed body locations. In addition to relevance criteria, information relevant to the quality of research reports was extracted but not used to exclude studies. Given the diversity of publications in scope, only general standards for research quality were applicable (i.e., assessments of sample size, clarity of aims, clarity of outcome etc. Quality assessment was based on Jadad and colleagues [49]). With a requirement that studies were peer-reviewed publications, pre-existing quality assurance was, to some extent, already in place prior to assessment of individual texts.
The general prerequisites for inclusion were then as follows: • The subject area was limited to psychology/psychiatry; medicine/physiology.
• The selected studies were empirical (i.e., not philosophical, conceptual, theoretical or symbolic), including systematic reviews of empirical investigations.
• Embodied emotion was investigated (given the empirical requirement, this excluded an undefined role for the body, or meaning/symbol it represented).
• An explicit emotional stimulus was used to elicit emotional responses in a (potentially felt) bodily location as a measured outcome.
• Emotional stimulus was non-tactile, thereby not involving direct manipulation of the body area to be measured (i.e., to avoid measurement artefacts, in the context of current aims), such as cases of administering a drug causing tachycardia with heart activity as an outcome measure).
The review channels publications into two main groups. There is a first group of "general" texts, which had to meet the above criteria. These are intended to permit assessment of the extent to which certain body areas dominate embodied emotion research, and where others may be under-represented. Beyond this overview of the literature, there is then a second separate main group of "core" texts, which specifically focus on the key question of this review, of where emotional feelings/sensations manifest in the body, as an outcome measure, in addition to meeting the general relevance criteria. To be considered comprehensive, a set of essential locations are as follows: (i) heart/cardiac; (ii) lung/respiratory; (iii) gut/gastrointestinal; (iv) skin (i.e., due to widespread use of skin conductance and galvanic skin response in emotion research); (v) eye/face (i.e., widespread use of electromyography, startle/blink response, facial expressions), and (vi) head (i.e., as reported in a relevant study, [50], and due to the current authors' own data from a study related to the current review, during which participants were asked to indicate where they ". . .most tend to detect emotional feeling", with many indicating the head).
The search strategy involved four separate stages: (i) a large scale initial set of searches following collaboration with medical library specialist staff at the University of Otago Wellington (UOW) to develop bespoke sets of search terms for each of the databases accessed (databases accessed 20 th February 2020); (ii) a subsequent search of gut/gastrointestinal sources (for reasons stated in the results section; databases accessed 2 nd April 2020); (iii) searching reference lists to identify further sources; (iv) a final set of searches based on the measurement tools used in the near-final set of core texts (databases accessed 5 th June 2020). The databases accessed were: MEDLINE, PsycInfo, EMBASE, EBSCO (behavioural sciences), SCOPUS, the Cochrane Database of Systematic Reviews, and the Cochrane Central Register of Controlled Trials.
For the first stage of searches, each publication was sifted initially by title, followed by abstract, then by full text, with publications streamed into one of the groups of texts as appropriate. Those passing all three stages went forward to a final integrative/synthesis stage. For this main set of searches, a 10% check was undertaken by a colleague to ensure a consistent and appropriate application of text selection criteria. The second search stage used keywords and titles from the original search/sift, applying these across the same databases as previously, but accessing specifically gut-focussed publications. The third stage accessed abstracts and full texts from the reference lists of all core and publications measuring all three visceroceptive locations, to identify further core texts. After the first reference searches, the process was repeated until no additional core texts were identified. (This resulted in three reference list searches.) The fourth and final stage was based on the set of core texts, which (as stated in the results) formed three clear groupings, arranged around distinct types of self-report measurement: (i) body mapping, (ii) a "Scene Construction Questionnaire", and (iii) a "Visual Analogue Scale" used by authors in combination with a self-report checklist. On that basis, a final set of searches were undertaken, again using all databases from the original searches, involving many of the same keywords and title words used in the additional gut-focussed searches. The combinations of search terms used for each stage are provided in S1 Table. The resulting core texts were synthesised using the integrative review guidance of Whittemore and Knafl [51], which outlines a five-stage process of: (i) data reduction (i.e., coding extracted information, forming logical sub-groups); (ii) data display (e.g., matrices, graphs; charts); (iii) data comparison (i.e., a process of "constant comparison" to convert extracted data into systematic categories, forming patterns, themes, and variation in relationships); (iv) conclusions (i.e., higher levels of abstraction/generalization); and (v) verification (i.e., confirming conclusions using the primary sources).
For (i) data reduction, subgroups were formed based on a study's methodology (in our case, a questionnaire, a rating scale, or a "body map"). For (ii) data display, matrices were used in Excel with extracted data in separate columns, using tables to summarise extracted data. Texts then underwent (iii) the data comparison process of 'constant comparison' involving memo-writing within a separate column to combine and summarise data from remaining columns for each publication, identifying patterns/variation, before writing between publications to establish links. Using mind-maps, memos were linked to form more general statements for the collection of publications to arrive at (iv) final conclusions, supported in turn by reference to the primary sources, i.e., (v) verification. Where necessary, the integrative process also involved returning to the original publications to check for accuracy and to extract any additional information to inform each of the five stages.
The final outputs from this process for core texts are a main table displaying a summary of extracted material with an accompanying narrative, plus some additional tables responding to the various findings of a "body map" cluster of publications (for the latter, see S2 Table), and a separate table of those remaining publications using more traditional self-report tools pertaining to bodily sensations (S3 Table).

Results
The results of the first stage of searching are displayed in Fig 1 (the minimal dataset underlying all results is in S2). From an initial 1777 hits, 51 publications went through to the final list of texts. Following this search it was clear that, in comparison to the heart and respiratory system, far fewer publications than anticipated had undertaken any form of measurement of the gastrointestinal tract or of emotion related symptoms (e.g., stomach ache or nausea). It is for this reason that the second set of searches was undertaken to check whether the original search terms had been adequate for capturing gut-related studies. Together with the two further stages of searching, 47 further texts were added to the original 51 (see Fig 2), resulting in a total number of 98 texts (general and core). The majority of texts met a general quality standard (i.e., clarity of reporting: aims, analyses, experimental design, sampling criteria, findings; having undertaken a power calculation), although relatively few reported on sample size/ power calculations.

General texts
Of these 98 texts, 88 were classified as general texts (Table 1). A wide array of stimulus types were used across these general publications, including images (e.g., from established emotional stimulus sets, such as the IAPS); sounds (e.g., music, heartbeat); computer games and virtual reality (VR); social scenarios and stressful encounters (e.g., public speaking); films; and the use of startle probes in combination with other stimuli. Measured outcomes were largely as expected, with combined self-report and bodily/physiological locations (Fig 3) showing a predominance of heart/cardiac activity (88% of publications), followed by skin (63%), eye/face/ head combined (40%), and the respiratory system (32%). The respiratory system was rather better represented among studies where outcomes could not be included due to artefacts from tactile stimuli, mainly in studies using respiratory loads and hyperventilation as stimuli in the investigation of anxiety and panic disorder. Despite follow-up searches for the gut/gastrointestinal tract, the final proportion of publications where this was measured was just 15%. The pattern was similar when focussing on only bodily/physiological outcome measurements, but with the gut being yet less represented (falling to 6% of all publications). Overall, three of the general papers reported participants' self-reports on specific body locations as outcomes. These also recorded the corresponding body area using physiological measures, with correspondence identified between self-reported and physiological measures for skin activation [52,53], and heart activation [53,54].

Core texts
From the full set of searches, 10 publications were classified as core texts (see Table 2). Of these, 7 used "body mapping" tools, mostly the "emBODY" tool originating with Nummenmaa, Glerean, Hari & Hietane [50]. This is a computer-based pixel selection program that allows participants to register mouse clicks within a body outline (silhouette), one to record a sensation of bodily activation, and one to record deactivation. These silhouettes are combined to create heat maps, with warmer colours indicating increasing activation, cooler colours

PLOS ONE
indicating deactivation, and black indicating neutral. With one exception [139], all body mapping papers used the emBODY software or, in the case of child participants, subsequently transferred a pen and paper version to the program [140,141]. In place of the emBODY tool,   [59] Used "scripts" using the Scene Construction Questionnaire, which were played in a laboratory setting. One script was "stressful", another related to drug/alcohol addiction and the third was a neutral control. Used the "Behavioural Arousal Scale" (BAS) and self-report to measure outcomes Stressful "scripts" Self-report/behavioural observation (using BAS): respiratory, skin, muscles, head, gut, eye(tears) Measures taken during the session Psych/medic Chaplin [60] Used "scripts" using the Scene Construction Questionnaire, which were played to them in a laboratory setting. One script was "stressful", another related to drug/alcohol addiction and the third was a neutral control. Used the "Behavioural Observation Scale" (BOS) (involving muscle twitching, muscle tremor, restlessness, muscle tension, muscle ache, headache, quickened breathing, yawning, talking / facial movements, crying, sweating, and stomach / abdominal changes), blood pressure, heart rate) Stressful "scripts" Heart; self-report ( [110] Hypothesised that high interoceptive awareness will be associated with greater heart-rate reactivity and better emotional memory in a recognition task, using the Schandry task Images Heart Psych/ physiology Pollatos [111] Hypothesised that high interoceptive associated with greater heart rate deceleration, and arousal. Schandry was used to measure interoceptive sensitivity

PLOS ONE
Jung, Ryu, Lee, Wallraven & Chae [139] used "a bodily sensation map-emotion (BSM-E) application" (p. 5). Participants could similarly report where sensations were felt in their bodies and could leave areas blank where nothing was felt, but this tool did not appear to record levels of deactivation, only varying levels of activation. These authors also created "Regions of Interest" (ROI) formed from clusters of clicks, which formed part of their analysis.
Overview of body mapping studies. Overall, the core texts utilizing body mapping technology used a range of emotional stimuli: images, GIFs, music, films, words, and sentences. Two report having undertaken studies in English [141,146], including a cross-cultural study [147], one was carried out in an English-speaking country but without an explicit statement on the language used, two reported their procedures in English, including quotations of experimental materials delivered to participants, but in nations where English is not the native language [139,145], with the remainder having undertaken studies in Finnish [140], or Finnish, Swedish and Taiwanese across a sample [50]. All studies used emotional stimuli consistent with basic emotion categorisations (i.e., anger, fear, surprise, happiness, disgust and sadness), with some also including non-basic/complex emotions (i.e., anxiety, depression, pride etc.), and one using stimuli pertaining to social scenarios [145]. Two publications [141,145] also included distinctions between self and other, i.e., separate tasks focusing on first-person responses to emotional stimuli and another involving judgements of the mapped responses of  Following positive and negative film stimuli (two films lasting 9 mins each: "Raiders of the Lost Arc" and "Peanuts", which were expected to produce several affective states), participants self-rated on a Visual Analogue Scale (VAS) for 14 items, immediately before and after the end of each film. The body location items pertained to: the heart, respiratory system, the gut (i.e. tense stomach), skin, eyes (i.e., tears), face (i.e., hot face), head (i.e., rush of blood). Concurrent EMG, skin conductance level (SCL), respiratory activity and ECG measures. Self-ratings of mood (e.g., anxiety).
Activity in heart, respiratory system & gut were prominent self-reports. Negative films: "heart rate increases", "faster breathing", "tense stomach"; positive films: decreased "tense stomach". EMG (corrugator): increased activity for negative, decreases for positive. EMG  Provoked anger, fear, disgust, happiness, sadness, and neutral. The emotional nature of the task was disguised using a visual accuracy task, a high difficulty "masking task". For map drawing, the authors did not use the emBODY tool, but "a bodily sensation mapemotion (BSM-E) application". Participants reported where sensations were felt leaving areas blank where nothing was felt. "Regions of Interest" (ROI) were identified as emotionally relevant clusters: "circular area with a radius of 10 pixels from the peak points of significant bodily sensation clusters" All emotions were represented in the chest, which is consistent with the other findings, and again anger was linked to hands (as was fear, which was also active in lower legs and feet  Emotion word stimuli: "neutral", "fear", "anger", "disgust", "sadness", "happiness", "surprise", "anxiety", "love", "depression", "contempt", "pride", "shame", and "jealousy".

PLOS ONE
Where is emotional feeling felt in the body? An integrative review a second-person. None of these mapping studies required speeded responses within restricted timescales, which is perhaps a reflection of the focus on basic emotion categories, which may take time to unfold as multi-componential syndromes. Only one mapping study [139] undertook bodily/physiological measurement (the Schandry task), which was also the only core text to assess interoceptive sensitivity. Further, whilst the body mapping tool could be viewed as an attempt to measure this construct, no study assessed interoceptive sensibility using an established tool, such as the MAIA [6]. Hence, in general, measurement of interoception was overlooked. Overview of non-mapping studies. Of the remaining texts, two reported studies by the same authors [143,144] using the same self-report tool: a visual analogue scale (VAS) in combination with a 14-item list of bodily symptoms, including bodily locations. These symptoms had nuanced qualities; for instance, the list contains specific references to "shallow breathing", "tense stomach", "heart rate increasing" and "sweaty palms", as opposed to more generic indications of bodily changes with wider applicability. In turn, participants may have been less likely to report these more specific outcomes (e.g., noticing their heart rate, but not necessarily as "heart rate increasing"). Overall, the list of self-report items included the heart, respiratory system, the gut (stomach), the head/face (including the eyes/crying), the skin (e.g., sweating) and general somatic activity (e.g., trembling). Both studies recorded physiological changes after emotional stimuli were applied: one, a study of anxiety [143], recorded facial/eye changes via EMG, skin conductance, heart activity by ECG, and changes in respiration by a thermistor. The other study recorded only skin conductance and heart activity [144]. These were the only core texts to measure multiple physiological systems in combination with self-report. The emotional stimuli used were films (seemingly the same films across the two publications), which were intended to be either generically pleasant or unpleasant [143], or positive or suspenseful [144]. Again, no speeded responses appeared to have been required of participants within restricted time scales, with the film stimuli lasting for up to 10 minutes in each instance followed by self-reports within an undisclosed period. As was mostly the case in the wider context of the general texts, besides self-report options there was no reporting of gastrointestinal activity at the bodily/physiological level in either publication alongside the cardiac, skin, EMG and respiratory measurements. Finally, there is some crossover with the body mapping studies in terms of basic emotion categories, with some reporting on "mood", e.g., disgust and sadness, although these were not indexed to the stimuli themselves.
The one remaining study [142] focussed on stress, drug and alcohol dependency using a Scene Construction Questionnaire to develop stimuli, with participants recalling stressful and drug/alcohol scenarios (we focus on only the former here). Participants reported any bodily changes from a list of 67 sensations, involving multiple body areas relating to physical symptoms at general levels (e.g., "Heart") and broken down into more specific symptoms (e.g., "Heart pounds", "Heart beats faster", "Heart skips a beat" etc.). The majority of body locations are covered in the list of responses: head/face (including eyes/crying), throat, chest (heart and respiratory), gut (stomach), arms, hands, skin and overall somatic activity. No physiological recordings were taken, with all body-related data consisting of reported symptoms, again without a time restriction. Finally, unlike the mapping studies, there was no framing of the stimuli in line with basic emotion categories. Whilst outcomes from the list of sensations included some basic emotions, these were not systematically combined with body sensations to indicate multi-componential syndromes pertaining to those categories.
Body mapping findings. The body mapping studies tended to identify distinct maps for each emotion type (and for social scenarios [145]. These are suggested to be patterns of "net sensations" for each type [50]; i.e., a subjective summation of bodily changes in multiple areas during an emotional response. There were exceptions to this distinctiveness, with sub-groups of participants displaying reduced differentiation between emotion types, i.e., those with schizophrenia, especially regarding low arousal emotions [146], child participants [140,141], older participants [147] and, in the case of schizophrenia and older age, reduced levels of activation, even for high arousal emotions. For the single mapping study that included a physiological component [139] the resulting Schandry task scores were, overall, positively correlated with the magnitude of activation indicated in the ROIs, which may suggest that greater sensitivity results in stronger bodily sensations pertaining to emotion. The ability to correctly categorise the body maps of others was also found [50], and consistency between self/other maps [141,145]. In addition to the distinctiveness of maps for each emotion type-both within and between publications-these also displayed a level of consistency of activation/deactivation across types (e.g., [147]), as well as some inter-study inconsistencies regarding single emotion types across publications (see assessment below, Table 3).
Visual inspection of body maps. Since each map is intended to display a representation of a "net sensation", i.e., a summative subjective report, it should be possible to visually inspect each of these to arrive at conclusions on which body areas were deemed to be reliably activated or deactivated (or neutral) during an emotion. That is, the 'heat' (activation), 'coolness' (deactivation) and unchanged areas of heat maps putatively correspond to the 'heat' or 'coolness' of an inter-subjective embodied experience. Indeed, this is in large part the value of such a visualization, in addition to the statistical inferences that can be undertaken using pixel selection data. For this reason, we have summarised an attempt at visual inspection (see S2 Table, and  Table 3) as activation (A), deactivation (D) and neutrality (N) in areas for each emotion type, for each publication. This is intended to allow comparisons and contrasts between similar mapping studies regarding which body areas change in emotional responses. From there, the degree of consistent identification is indicated per body part (with A or D indicating a reported bodily change, and N no change), across body mapping texts, drawing on only the investigation of basic emotions (i.e., not complex emotions or social scenarios) from healthy adult samples, to allow comparability. Although basic emotion categories themselves are not the intended focus of this review, we present the results of these papers in terms of the authors' own categorisation. In any case, this may indicate common changes across emotion types, suggesting perhaps more fundamental body locations, which may also be foundational; i.e., always occurring during a response, regardless of duration or complexity.
To undertake this visual assessment, we have made bodily distinctions based, in part, on those made within the texts themselves: (i) head/face (changes anywhere within the boundary of the head outline, including eyes, cheeks etc.); (ii) chest (encompassing the area from the termination of the throat down to the horizontal line that cuts through the elbows on the body silhouettes); (iii) the throat (the area from the termination of the upper chest to immediately below the jawline); (iv) the upper arms (including the shoulders, down to the inward curve of  In some cases, the maps displayed only a trace change tending towards neutrality as opposed to clear shifts in subjective experience. Relevantly to this, Volynets, Glerean, Hietanen, Hari & Nummenmaa [147] state the need to consider the vast array of possible influences (including cultural) on emotional responding that inevitably leads to a degree of mapping variability (which they otherwise consider to be genuinely universal). Readers are, thereby, encouraged to focus more on the degree of commonality across contexts as opposed to relatively minor differences: ". . .our view of psychological universality refers to the degree of consistency of psychological phenomena across cultures, rather than naïve all-or-nothing conceptualization" ( [147], p. 8).
Therefore, in cases of trace change, the "T" is used to indicate a level of ambiguity between a trace of subjective experience and purely uninformative subjective noise. This then retains an observable (but minor) distinction between maps without forcing a conclusion on consistency due to a potentially arbitrary judgement. Further, we make no other distinctions regarding differences in the level of clear activation or deactivation (i.e., that do not approach a value of N). Finally, we have allowed for one discrepancy from the majority of texts per body part (i.e., one of the seven texts in each case). For example, if one body map for an emotion type displays neutrality with the remaining publications showing agreed activation in a body part, this remains classified as consistency. In this way, having taken ambiguity into account and allowing for a minor degree of error, any remaining discrepancies after implementing a relaxed standard may suggest a more significant departure from consistency.
Across the options, the head/face, chest and throat show the greatest degree of agreement throughout emotion types regarding changed activity following an emotional stimulus. The peripheral body parts show the most 'traces' of sensation. With the exception of the lower abdomen (where displayed changes were for the most part either a potential spill over from reported changes in the chest or were otherwise barely observable on the maps), it is the peripheral body parts that are less likely to show agreement across publications. (The groin was only consistently neutral, hence less relevant to this categorization.) If taking a simple unweighted average of the proportion of consistency across the two interpretative options, the results indicate an order of consistency that broadly reflects the distance of each body part from the head/face, a fact reflected under each option: body areas), followed by happiness and surprise (6/11), fear (5/11) and disgust (3/11). Alternatively, where "T" is interpreted as 'noise', surprise alone showed the greatest degree of agreement (10/11 areas), followed by disgust (9/11), anger (8/11), fear (6/11), happiness (4/11) and sadness (1/11).
Non-body mapping findings. The full results for active body locations reported by the three more traditional self-report studies are provided in S3 Table. Comparisons between these and the body mapping studies are limited by several factors, not least of which is the limited numbers of body areas reported on (i.e., 6 of the original 11), the use of more general emotional stimuli (i.e., "stress", "pleasant", "suspenseful etc.) in addition to contrasts between the self-report tools used.
Visual analogue scale studies. In the study of anxiety [143], self-reports indicated the head/ face ("blood rushing to the head"), heart, respiratory system and gut to be active in response to stimuli (with the gut self-reports indicating increasing or decreasing tension depending on whether stimuli were negative or positive, respectively), along with changes in skin conductance, changes in EMG and heart rate. Self-reported disgust was found to correlate with a "tense stomach", "shallow breathing" and a negative correlation of disgust with physiological measures of respiratory activity (in the high anxiety group). For self-reported anxiety in both the highly anxious and low anxious groups, there were accompanying self-reports of "heart rate increasing", changes in breathing and "tense stomach"; for the highly anxious group only, there were corresponding changes in actual respiratory activity. For "joy", the highly anxious group self-reported respiratory changes. There were also subjective-objective correlations between skin conductance measures and self-reports of "blood rushing to the head", "heart rate increasing", "breathing faster" and "tense stomach" (the high anxiety group only). The other study by the same authors [144] has similar self-reports of heart, respiratory, and facial changes but no gut changes, along with actual heart changes (ECG) and sweating palms, in response to "suspenseful" film stimuli.
Overview of core text findings. The connections between the 10 core publications consist, then, in the widespread findings on: (i) the head/face (by self-report supported by findings from EMG, in particular); (ii) the chest in general, with specific findings on the heart (by selfreport and ECG); (iii) specific findings on the respiratory system (self-report and thermistor recordings); and (iv) support for activation of the gut (but by self-report only, and which shows limited consistency across the body mapping studies). Self-reports on the throat, too, may indicate gut changes, but which may also be interpretable as respiratory changes, or both. Overall, some of the more pervasive changes displayed in the maps, such as in peripheral body parts, could relate to the correspondence found in the two "VAS" studies between skin conductance changes and self-reported activation in the head/face, heart, respiratory system, and to a lesser degree, the gut/stomach. Further, in the case of basic emotion categories, where these were reported in relation to body locations in the traditional self-report studies, selfreported disgust indicates some consistency regarding the gut (stomach) between the mapping studies and the study of anxiety (but less obviously so for the chest/respiratory system), for anxiety/fear regarding the heart (self-report) and the respiratory system (self-report and physiological), and for joy/happiness regarding self-reports of the chest/respiratory system. Finally, there was also the limited support from the few general texts cited (i.e., for subjective-objective concordance for the heart and skin).

Discussion
The findings from the wider collection of publications confirms the predominance of the heart in research involving measures of bodily changes in the investigation of human emotion. The respiratory system was perhaps not as widely investigated as anticipated, although many studies were excluded due to artefacts (e.g., an inspiratory load was the emotional stimulus, activity in the respiratory system was an outcome measure). Regarding the one remaining member of the visceroceptive trio-the gastrointestinal tract-in only a minority of cases was there measurement by self-report or by a physiological tool such as the electrogastrogram (EGG).

Lacking measurement of the gastrointestinal system
As noted by the authors of one of those cases [131] the gut has been relatively underresearched in emotion studies ("In psychophysiology, measures of electrodermal activity, cardiac function, facial EMG, and respiration have been used frequently to assess the emotional states of experimental participants. . .Remarkably, the gastrointestinal system has been almost completely neglected. . .", p. 70), which is despite the non-invasiveness and reliability of EGG (p. 71). Since publication of that paper, the situation has not vastly improved. This general lack of emphasis on measuring all three visceroceptive domains, which appear to be of particular relevance to embodied emotion, means that potentially useful findings on emotion regulation are being missed. With an accompanying scarcity of studies using both self-report and physiological responses, we found limited consideration of the concurrence of subjectivity and objectivity in emotion research. This was also true of the core texts, with only three instances of concurrence, again without objective measurement of the gut despite all 10 texts presenting the opportunity for gut-related self-reports. If there is an aim to investigate the bodily contributions to emotional experience, it is necessary to rely on some sort of self-report or subjective display, and to have this grounded somehow in factual bodily changes. This aim is consistent with the tripartite model of interoception [5] and embodied emotion research [1,2] more generally.

Lacking measurement of interoception
Relatedly, without physiological measurement there can be no measurement of interoceptive sensitivity, part of the tripartite model. The one exception among the core texts [139] used the Schandry task [11], which has been the target of recent criticism, with charges of pervasively undercounted heartbeats [148], and that the task may be influenced by pre-existing knowledge of heartbeat [149][150][151]. Indeed, in the case of Jung, Ryu, Lee, Wallraven & Chae [139], the reader is provided with only a scatterplot, without raw Schandry scores. If scores are substantially higher than the mean usually found in interoception studies that categorise participants as highly interoceptive (e.g., [110]), this could indicate improvements in performance via noninteroceptive influences. The remaining core texts measuring physiology were not overtly concerned with interoception (and, indeed, were published some time before the more recent explosion of interest in the construct).
A key finding of the review, therefore, is that-for the most part-studies of emotional feeling and body location were not grounded in interoceptive research, under any articulation. The body mapping authors variously refer to interoception (e.g., [139,142,144,146]) but, even where it was possible to measure the construct as articulated in the literature [5,6] there was little attempt to do so. Also, given that the body mapping tools have not been reported as validated using established interoceptive measurements (although one study [139] may be seen as an attempt to do so), there is the potential to undertake validation not only in relation to sensitivity measures, but also with established sensibility measures such as the Body Perception Questionnaire [152] and other articulations of interoception, such as the MAIA [6]. Another consequence of having no measures of sensitivity and validated sensibility is that a calculation of interoceptive awareness (i.e., a metacognitive construct [5]) is not possible. Whilst it cannot be assumed that sensitivity measurements will map onto sensibility measurements (e.g., [5]), any possibility of detecting correspondence evaporates if studies do not attempt to collect such data. It may be particularly important to gather data across different interoceptive channels, which may be a decisive factor if interoceptive sensitivity is indeed indexed to specific channels.

Using independently validated stimuli
Further, the wide variety of stimuli used across the core studies were for the most part adopted or developed by the authors themselves, thereby ensuring they reflected the basic emotion category each stimulus was intended to elicit. As indicated previously, there are established stimulus sets (e.g., the IAPS), which have previously undergone assessment of their ability to produce detectable physiological responses. Hence, whilst the mapping in particular is an innovative approach to self-report, it may be worthwhile using it with stimuli that have an a priori likelihood of producing physiological changes to thereby increase the likelihood that the body maps genuinely represent underlying physical changes. If there can be no accompanying physiological measurement, there would then at least be a stronger implication of objectivity. In addition, without independent validation, there is the risk that stimuli could inadvertently help to articulate a desired outcome. Indeed, this may relate to the measurement of basic emotions themselves.

The emphasis on basic emotions
With the stimuli being categorised as they are in many of the core texts, these may have the potential to act as primers for basic emotion outcomes. The majority of core papers were concerned with basic emotions, again with the body mapping papers in particular tending to categorize stimuli accordingly. In some cases [50,140]), there was yet more explicit grounding in basic emotion theory, with assertions that basic emotions have been established neurophysiologically, with the role of the body maps being to confirm this reality at a subjective level. However, the basic emotion view is controversial, with much in the way of counter-evidence.
There is, for instance, a research cluster involving the work of Lisa Feldman Barrett where, rather than conclusions on discrete emotion types, the evidence points to more fundamental biological features that interact with constructive processes to create emotional instances [153,154]. If the latter model is true, there would be no requirement for discrete bodily activation to match discrete subjective body maps.

Concept-based vs sensation-based interpretations of the body mapping texts
This latter suggestion pertains in particular to the verbal/linguistic stimuli used in the body mapping studies, including emotion words and statements, where in one case participants were provided with definitions of the words used [146]. Three of those seven studies were likely undertaken in English, including the cross-cultural study [147]. Whilst it was recognised that the body maps may be concept-based, culture-independent readouts [50], authors indicate that their use of multiple stimulus types, principally "non-verbal" stimuli (e.g., films without words), resulted in comparable body maps to those elicited using words. The argument here appears to be, then, that since there is parity between non-verbal and verbal stimuli this indicates that the sensation-based reports are not limited to concepts encapsulated in language.
However, there is also the possibility that this demonstration of parity between stimulus types illustrates that "non-verbal" stimuli draw on the same concept-based, culture-independent readouts as the explicitly linguistic stimuli. Hence, this may be why it was possible to provide a participant with an emotion word (e.g., "disgust"), which did not involve inducing emotion [50], and thereby find a similar map to the one produced when showing a disgusting film. To some extent, then, the stimulus types may articulate responses on the maps due to a shared "readout" or script. The sense of this suggestion is seen, for example, in maps that highlight the eyes in "sadness", the hands in "anger", and the whole body in "happiness" (see [50]). Whilst these responses are explicable, the explicability may, in large part, consist in participants reverting to established display rules for each emotion (i.e., crying when "sad", clenching fists in "anger", and our whole body "lighting up" when feeling "happy"), but without any such essential, universal, biological phenomena. We can, after all, feel sad without the urge to cry, and feel angry without the urge to form a fist; hence, the mapping suggests (although does not determine) a combination of factors: socio-cultural or universally applicable constructs, along with subjective sensations.
This concept-based interpretation may link to some of the other findings, such as the dominance of not only the face, which is understandable given the role of emotional facial displays and startle responses, but also the head. Nummenmaa, Glerean, Hari & Hietanen, [50] suggest that the level of agreement over activation in the head is in part due to "felt changes in the contents of mind triggered by the emotional events" (p. 648). This is, however, a different sense of "sensation" from that which the body maps are intended to display, which is explicitly a matter of capturing actual bodily feelings at a location, such as feeling one's heart increase. It would not seem possible to feel "changes in the contents of mind" in the head, in quite this way; to express so would appear to be drawing on a purely cognitive awareness of mental content coupled with an inference about the brain/mind being located in the head, rather than a sensation experienced in the head. It is not possible to separate out the head from the face in the mapping responses, but if head clicking was even in part due to the reasons stated by the authors, this should provide further reason for caution when interpreting the other responses. However, with indications of "blood rushing to the head" in the traditional self-report studies, supported by physiological measures, to some extent the maps may be drawing on this physical phenomenon, or similar.
The authors' interpretations of the maps produced by children [140,141], people with schizophrenia [146] and older participants [147] point to developmental or organic explanations, with body maps being viewed as essentially biological representations. Again, it may be said that to the extent there are relevant socio-cultural explanations, children will have limited access to the representations of emotion ("net sensations") that are so familiar to adults. Indeed, as "net sensations", these could be interpreted as "generalised instances of emotionby-situation"; hence, children will not have the experiential base to form such generalisations. This may be one explanation. Another may be more relevant to older participants and those with schizophrenia, who show less differentiation and activation in their maps. Rather than an interoceptive deficit, as suggested in the publications, it may be rather an issue of memory impairment, perhaps more obviously so in the case of older age, although memory impairment is also a symptom of schizophrenia and a side effect of psychoactive medication. Hence, impaired memory may result in reduced access to the situation-indexed array of concepts required to respond in predictable ways on the body maps. Incidentally, one group of authors [147] suggest that the apparently diminished interoceptive ability of older people fits with previous research indicating that older people have greater emotion regulation ability than younger. However, as indicated previously, there is a body of research pointing to greater interoceptive ability being vital for greater emotion regulation ability.
Another related issue pertains to the well-known difficulties of engaging in interoception, as indicated in relation to heartbeat detection. The benefits of interoceptive training for interoceptive performance have been shown in multiple studies. For example, Farb, Segal, Mayberg, Bean, McKeon, Fatima & Anderson [36] report an 8-week mindfulness intervention-an investigation of interoceptive sensations-finding that when participants had no mindfulness training there was a coupling of brain regions involved in sensory and conceptual processing. However, after training, these brain regions became decoupled, suggesting that whilst trained individuals can undertake interoception, those without training may struggle to distinguish sensations from sensation concepts [36]. With the body mapping participants seemingly undertaking no such training, there is again reason for caution before drawing definitive conclusions that the maps represent sensations rather than concepts.
A concept-based interpretation is also potentially relevant to the issue of early stage emotional responses, prior to full blown responses. The suggestion has been that embodied emotion research could contribute to the emotion regulation evidence base by identifying rapidly emerging bodily changes as a target for early stage emotion regulation. This regulation could be antecedent-focussed, a strategy in place prior to rapid changes, and therefore able to be effective prior to escalation into full-blown multi-componential syndromes. As indicated previously, in order to research this the timescale for applying a stimulus and recording a response must be brief, but the difficulty with the core texts is that the stimuli were delivered, it seems, over indefinitely long timescales (e.g., [50], or restricted but still many minutes long (e.g., [143,144]), and with no indication of speeded responses having been required. This increases the likelihood that participants could draw on rich appraisals of the stimuli, which, even in those cases of "non-verbal" stimuli, it is possible that the emotional categories to which these belong were clear to participants and, therefore, which appraisals were conceptually appropriate to apply (i.e., which "readout" to draw on) when responding. Indeed, Jung, Ryu, Lee, Wallraven & Chae [139], imply this influence of concepts (p. 10) regarding their use of masking.
Despite the critique thus far, there is a high degree of concordance found within studies that use different languages, from different cultures (see, e.g., [147]), and using multimodal stimuli (e.g., [50]), which retains reason to take seriously the prospect of universal locations of sensation. However, despite applying a relatively low threshold for consistency in a general assessment across the body mapping publications, we still identified a notable lack of agreement regarding which parts of the body change within each "basic emotion", particularly regarding peripheral body parts. Indeed, the only reliably high concordance for activation/ deactivation seems to be for the head/face, chest and throat, and with similarly variable consistency when comparing consensuses between emotion types. If the body maps are capturing underlying physiological changes we should perhaps expect greater consistency across studies. Or, perhaps it just is the case (as stated in [147]), that there are other influences on consistency that need to be taken into account. The more consistent body area changes may genuinely be universal, possibly as sensations that "represent a general arousal state resulting from autonomic activation" ( [142], p. 6), with the remainder being situationally variable and contextdependent. It is also interesting to note that, in our brief visual assessment, the further away a location on the map was from the head/face, the lower the level of consistency found within and across studies. Again, this could indicate something important about actual bodily activation (e.g., facial muscle movements are a core feature of emotional sensations), or that the maps tap into a concept-based articulation rather than actually mapping bodily sensations (i.e., the possibility that the head/face consistency displays mental "sensations"). The difficulty, then, is how to disentangle genuine sensation reports from concept-based reports without concurrent physiological measurement.
The more traditional self-report studies are only somewhat helpful in this respect, given they do not all provide extensive opportunities for self-report, do not clearly indicate attempts to reduce appraisals of bodily activity, do not report early stage time-stamped self-reports (hence reducing the influence of concepts), and do not objectively measure all self-reported body areas (e.g., the gut). Their value is, however, in supporting the possibility of a few fundamental bodily changes via concordance between self-report and same-location physiological measurement (including the physiological support for head-specific changes that can be sensed as opposed to a change in mental processing). Unfortunately, there is limited ability to make "basic emotion" comparisons with the mapping studies, given that different category names were used ("joy" in [143,144], as opposed to "happiness"; plus "anxiety" as opposed to "fear"), and mixed findings on disgust.
Referring back to the general texts that at least measured all three key visceroceptive domains does little to clarify this situation, with only six texts covering all three, and with a split between subjectivity and objectivity across these. Four of the six [60,106,116,129] collected but did not report data for all three domains, instead combining self-report measurements of body location into composite scores, and otherwise reported physiological measurement of the heart only. The remaining two texts [56,123] illustrated rather the converse situation: recording and reporting only physiological measurements of all three domains. In a study of adult participants, the first of these [56] recorded gastric changes using EGG, respiration with a resistive chest strap, and heart rate by ECG, also measuring RSA. They found no main effects in any domain, with some interactions involving the heart and for RSA but not for respiratory rate. Sternbach [123], however, found changes in all three domains, using an ingested magnet to measure gastric motility changes, a strain gauge around the chest for respiratory changes and heart measurement by ECG, with film clips used as stimuli for child participants. Hence, in the absence of corresponding self-report data, differing measurement tools regarding the gastrointestinal component, child vs. adult samples, bidirectional results, and with one study being over sixty years old, little can be concluded from these studies.

The future of embodied emotion research
Future research would benefit from, at least, incorporating the three key domains in investigations of embodied emotional feeling, whilst ensuring opportunities for subjective-objective correspondence (but noting difficulties elsewhere in identifying correspondence between interoceptive sensitivity and sensibility). Concurrent physiological measures would be essential in at least these domains, perhaps in combination with body mapping. On this general point, Nummenmaa, Glerean, Hari & Hietanen [50] suggest that future developments in physiological measurement may involve the option of using "whole-body O-H20 PET imaging" (p. 650), which would permit objective assessments matching the self-report responses in body mapping studies. If body maps are used, it would be important to validate these as providing genuine measurements of interoceptive sensibility (or, given, the ongoing development of interoception as a construct, whichever form this will ultimately take). As emphasized throughout, there would also be substantial benefit from focusing on time-limited emotional stimuli presentations, potentially also with the use of masking to thereby limit appraisal. Whilst something like a mask was used in one core text [139], seemingly paradoxically, these authors suggested that their masking prevented "emotional experience", per se, whilst at the same time they sought to address "emotional sensations". In general, to increase the chance of eliciting physiologically detectable bodily changes, validated emotional stimuli should be used (e.g., IAPS). An interoceptive training period could be applied in longitudinal studies, which may help to identify whether improvements in interoception result in greater concordance in body maps, potentially finding more general patterns of activation and deactivation that cut across "basic emotion" types.

Limitations
This review is subject to several limitations. Firstly, although four sets of literature searches were undertaken, including an initial set that used a search strategy developed with professional library staff, and across multiple databases, there remains the possibility of bias. The 2014 paper by Nummenmaa, Glerean, Hari & Hietanen [50] was already known to the authors prior to accessing the databases; hence, when this relevant paper was not identified during searches, this gave cause for concern. A related paper [146] was, however, identified using the same initial search strategy, which involved a clinical sample. There was, then, the possibility that the main search strategy tended to identify clinical studies, despite the search terms themselves not restricting in this way, thereby potentially missing a collection of relevant non-clinical studies.
The language restriction decision was based on there being limited resources for accessing and translating all non-English texts that otherwise satisfied the search criteria. This may have resulted in missed relevant texts. Also, in terms of applying criteria, there would ideally have been more than a 10% check of the text selection process; again, due to limitations on resources, this was not possible. Hence, this has increased the chances of error in applying criteria.
Regarding the visual inspection of body maps and subsequent comparisons and contrasts, there may have been room for interpretation regarding classifications of N, T, or perhaps borderline cases for A or D. For the most part, the maps clearly display areas of changed activity, with room for interpretation largely relating to activation in the hands, feet, legs and arms. Due, perhaps, to computer screen limitations, or the quality of online graphics, other visual "traces" of activity on maps may have been missed. However, the images inspected have been indicated in the foregoing text and are available within the cited publications. Also, providing an unweighted simple average of the consistency scores may not be valid, depending on the level of indeterminacy regarding whether one of the two interpretative options has greater 'weight'.
The review has emphasised the tripartite model of interoception developed by Garfinkel and colleagues [5]. This assumes that objective measures of interoception (sensitivity) are possible, and that these can correspond (as metacognitive awareness) to subjective measures of interoception (sensibility)-however, the extent to which this all occurs has been challenged in the literature. Hence, the standard imposed throughout the review regarding the ideal of studies providing objective and subjective measures, a standard in part predicated on the tripartite model, may be considered unnecessarily restrictive. However, this standard is also based on the more general aim of the review: to establish where in the body (an objective matter) emotions are felt (a subjective matter). Hence, even independently of the tripartite view of interoception, at the level of Williams James' original intuitions, subjective-objective correspondence remains a key focus, as opposed to being restricted to a single articulation of one possible construct (interoception) within the broader field of embodied emotion.

Conclusions
In the attempt to identify empirical studies on where emotional feeling is felt in the body, particularly during the early stages of an unfolding emotion, this review ultimately found no studies meeting a "gold standard". That is, no study undertook concurrent subjective self-report and objective bodily/physiological measurements pertaining to a comprehensive array of body areas including all three key visceroceptive domains; nor were there innovative self-report tools that could clearly approximate the value of physiological measurement.
The few general papers (providing a backdrop to the review's core texts) that did measure all three domains, did so only at the objective level, with varying outcomes. The dominance of heart measurement in the wider literature was confirmed, with perhaps greater dominance than might have been expected. Also confirmed was the widespread scarcity of gut measurement (whether self-report or physiological), which was perhaps more scarce than might have been expected. The gut may be a particularly important area for embodied emotion, given the role of the enteric nervous system, the gut biome and the vagus nerve [19,20].
Taken together, the core texts suggest that the face/head and chest (heart and respiratory activity) may be fundamental in embodied emotion. These were the most consistent selfreports, regardless of emotion type, reports that matched physiological activity in studies that also recorded facial movements (EMG), heart rate (ECG) and respiratory activity (using a thermistor). This does not, however, permit conclusions on early stage sensations, given the absence of early stage subjective reports concurrent with time stamped physiological changes. Further, the current review identified multiple issues with the core texts, which necessarily gives cause for caution when interpreting the self-report findings, in particular, as a matter of emotional sensation. Also, none of the core texts measured physiological changes in gut activity.
Overall, there is significant room for development in this research area, to ground subjective reports of emotional feeling in objective bodily changes, in keeping with the original insights of James [1], and then Damasio [2]. This has the potential to provide greater nuance for understanding location-based contributions to interoception, and for the continued development of clinical intervention tools, building on the evidence base addressing the connections between mindfulness and interoception. The notable lack of studies incorporating gastrointestinal measurement, in particular, despite physiological evidence supporting its role in emotion, should be a focal point in future studies by utilizing existing measurement tools (e.g., electrogastrograph, EGG).