Neural correlates of interoception: Effects of interoceptive focus and relationship to dimensional measures of body awareness

Abstract

Interoception has been defined as the sensing of the physiological condition of the body, with interoceptive sensibility (IS) characterizing an individual's self‐reported awareness of internal sensation. IS is a multidimensional construct including not only the tendency to be aware of sensation but also how sensations are interpreted, regulated, and used to inform behavior, with different dimensions relating to different aspects of health and disease. Here we investigated neural mechanisms of interoception when healthy individuals attended to their heartbeat and skin temperature, and examined the relationship between neural activity during interoception and individual differences in self‐reported IS using the Multidimensional Scale of Interoceptive Awareness (MAIA). Consistent with prior work, interoception activated a network involving insula and sensorimotor regions but also including occipital, temporal, and prefrontal cortex. Differences based on interoceptive focus (heartbeat vs skin temperature) were found in insula, sensorimotor regions, occipital cortex, and limbic areas. Factor analysis of MAIA dimensions revealed 3 dissociable components of IS in our dataset, only one of which was related to neural activity during interoception. Reduced scores on the third factor, which reflected reduced ability to control attention to body sensation and increased tendency to distract from and worry about aversive sensations, was associated with greater activation in many of the same regions as those involved in interoception, including insula, sensorimotor, anterior cingulate, and temporal cortex. These data suggest that self‐rated interoceptive sensibility is related to altered activation in regions involved in monitoring body state, which has implications for disorders associated with abnormality of interoception. Hum Brain Mapp 38:6068–6082, 2017. © 2017 Wiley Periodicals, Inc.

INTRODUCTION

Interoception has been defined as the sensing of the physiological condition of the body [Craig, 2003]. Anatomical work in animals has shown that the majority of signals coming from visceral organs (e.g., from the heart, lungs, and stomach) reach the brain via cranial nerves IX and X (glossopharyngeal and vagus) and the dorsal horn of the spinal cord, whereas somatic information (e.g., from the skin, muscles, and joints) is relayed through the dorsal horn route [Cameron, 2001; Cervero and Connell, 1984; Craig, 2002, 2003; Critchley and Harrison, 2013; Schachter and Saper, 1998]. Afferents from both sources project to brainstem and midbrain structures, and largely by way of the thalamus, connect to cortical and limbic regions including the insula, somatosensory cortex, cingulate cortex, orbitofrontal cortex, and amygdala [Cameron, 2001; Craig, 2002; Critchley and Harrison, 2013]. In one of the first studies to investigate neural correlates of interoception in humans using functional magnetic resonance imaging (fMRI), Critchley et al. [2004] found greater activation of the insula, somatosensory cortex (postcentral gyrus and inferior parietal lobule), premotor regions, anterior cingulate cortex (ACC), and inferior frontal gyrus (IFG) when subjects performed a task where they judged whether their heartbeat was in sync with a tone or not. Subsequent imaging studies where subjects attended to heart, stomach, and breathing sensations have noted similar patterns, identifying fairly consistent activations in insula and sensorimotor regions [Avery et al., 2015; Caseras et al., 2013; Ernst et al., 2013; Farb et al., 2013; Pollatos et al., 2007b; Simmons et al., 2013; Terasawa et al., 2013a, 2013b; Tracy et al., 2007; Wiebking et al., 2014; Zaki et al., 2012] (see Schulz [2016] for meta‐analysis). These human fMRI studies are consistent with anatomical models, and indicate that top–down attention to the body activates some of the same cortical regions that receive bottom–up information from the periphery.

Within the field of interoception research, recent attention has been drawn to the important but sometimes overlooked distinction between the correct identification of body sensation or signals, or interoceptive accuracy (IA), and self‐reported awareness of or sensitivity to body sensations (“interoceptive sensibility,” IS) [Ceunen et al., 2013; Garfinkel and Critchley, 2013; Garfinkel et al., 2015]. Neuroimaging studies have linked greater IA (as measured by heartbeat detection accuracy) with increased activation of anterior insula, somatosensory cortex, precentral gyrus, and IFG when attending to body sensations [Caseras et al., 2013; Critchley et al., 2004; Kuehn et al., 2016; Pollatos et al., 2007b], suggesting that enhanced cortical processing may confer an advantage for the accurate detection of body signals. Recent functional connectivity studies have also shown that connectivity of the insula at rest [Chong et al., 2016] and during task (Kuehn et al., 2016] is related to IA measured outside of the scanner. Interestingly, objective measures of IA are not always correlated with subjective measures of IS [Ceunen et al., 2013; Forkmann et al., 2016; Garfinkel et al., 2015; Khalsa et al., 2008; Meessen et al., 2016; Mehling, 2016; Whitehead et al., 1977]. For example, depression is associated with reduced IA (heartbeat detection accuracy) [Limmer et al., 2015; Pollatos et al., 2009] but increased self‐reported IS [Grossi et al., 2017; Limmer et al., 2015], a dissociation that has also been found in autism spectrum disorder [Garfinkel et al., 2016]. These data highlight the important role of subjective appraisals of interoception in psychopathology, yet little is known about the neural correlates of self‐reported IS.

There are several ways that IS has been assessed in the literature. Garfinkel et al. [2015] asked participants to rate confidence in their perceived accuracy of responses during a heartbeat detection task. Trait‐based scales, such as the Body Perception Questionnaire (BPQ)‐Awareness subtest [Porges, 1993], the Body Awareness Questionnaire [Shields et al., 1989], and the Self Awareness Questionnaire (SAQ) [Longarzo et al., 2015] have been used to measure an individual's general tendency to notice and be aware of body sensations. Critchley et al. [2004] found a relationship between IS as measured with the BPQ and gray‐matter volume in the insula. Additionally, two studies have examined the relationship between IS and resting‐state functional connectivity, both using samples that included patients with hypothetical abnormalities of body sensitivity. Grossi et al. [2017] found a relationship between increased IS as measured with the SAQ and connectivity between ACC and orbitofrontal cortex (OFC) in a sample that included healthy adults and patients with illness anxiety disorder. Also using the SAQ, Longarzo et al. [2016] found a correlation between IS and connectivity between anterior insula and somatosensory regions in a sample that included healthy adults and patients with irritable bowel syndrome. Although the self‐report measures used in these previous studies tap into a core aspect of IS (i.e., the tendency to notice and be aware of sensation), interoception is a multifaceted construct that can be viewed as including not only the tendency to be aware of bodily sensation, but also how these sensations are interpreted, regulated, and used to inform behavior [Cali et al., 2015; Mehling, 2016; Mehling et al., 2012]. These other aspects of interoception are particularly interesting in light of the fact that some may be maladaptive while others are adaptive [Mehling, 2016; Mehling et al., 2012]. For example, excess worry about bodily sensations is associated with panic disorder [Clark, 1986; Rudaz et al., 2010; Salkovskis et al., 1996; Yoris et al., 2015] and illness anxiety/hypochondriasis [Abramowitz, 2005], whereas the ability to control attention to body sensation (e.g., as implemented through mindfulness) can be psychologically beneficial [Farb et al., 2015; Mehling, 2016]. A characterization of the neural correlates of IS that takes into account the multidimensional nature of interoception could thus inform our understanding of the relationship between body processing and mental and emotional health.

In this study, we used fMRI to measure neural activity in a group of healthy adults during performance of an interoception attention task. Although interoception has been described in various ways in the literature, our use of the term follows the definition put forth by several researchers [Cameron, 2001; Craig, 2003; Khalsa et al., 2009] as the sensing of the physiological condition of the body, including signals from the organs (visceral) and from the skin, muscles, and joints (somatic). Following this approach, interoception conditions in the task involved attending to heartbeat as well as skin temperature; exteroception control conditions involved attending to words on a computer screen. We note that this approach diverges from a competing conceptualization that defines interoception as the detection of visceral but not somatic signals [Critchley and Harrison, 2013; Schulz and Vogele, 2015]; the inclusion of both visceral and somatic conditions in this study can contribute to this debate by directly investigating the similarities and differences between these different types of body sensing. We examined the relationship between IS and brain function during interoception using the Multidimensional Assessment of Interoceptive Awareness [Mehling, 2016; Mehling et al., 2012], a self‐rated scale that measures different components of interoception. We hypothesized that insula and sensorimotor regions would be more activated to the interoception compared to exteroception conditions, replicating prior studies. We further predicted that different components of IS (e.g., tendency to notice sensations vs tendency to worry about aversive sensations) would be associated with dissociable neural circuits.

METHODS

Subjects and Procedure

Nineteen healthy volunteers were run in the study. None had current or previous Axis I diagnosis, as confirmed with the Mini International Neuropsychiatric Interview (M.I.N.I.) [Sheehan et al., 1998], and none were taking psychoactive medications. Measurement of trait IS utilized the Multidimensional Assessment of Interoceptive Awareness (MAIA) [Mehling, 2016; Mehling et al., 2012]. The MAIA includes 32 questions assessing 8 dimensions of IS identified through factor analysis: (1) tendency to notice or become aware of body sensations (4 questions), (2) tendency to not distract oneself from sensations of pain or discomfort (3 questions), (3) tendency to not worry about or experience emotional distress in response to sensations of pain or discomfort (3 questions), (4) ability to sustain and control attention to body sensations (7 questions), (5) awareness of the link between emotion and body sensations (5 questions), (6) ability to regulate negative emotion through attention to body sensations (4 questions), (7) tendency to listen to body sensations for insight into emotion and guide behavior (3 questions), and (8) tendency to experience the body as safe and trustworthy (3 questions). A major strength of the scale is that it has construct validity with other measures of body awareness, mindfulness, and emotion regulation but goes beyond prior work by assessing multiple different aspects of IS [Mehling, 2016; Mehling et al., 2012]. Subclinical anxiety and depression symptoms were evaluated using the Beck Anxiety Inventory (BAI [Beck et al., 1988]) and Beck Depression Inventory (BDI [Beck et al., 1961]).

This study was part of a larger double‐blind placebo‐controlled crossover drug challenge project where all subjects were scanned twice, once under a placebo condition and once under an active drug condition (ondansetron). Given that the aim of this article is to investigate neural correlates of interoception and relationships with dimensions of IS, which has not been previously reported, we only present findings for the “baseline” placebo condition and not for the drug effect (which will be reported elsewhere). The order of placebo and drug conditions were counterbalanced across subjects and occurred on different days ∼5–10 days apart. In the present sample, 9 subjects received placebo first and 10 subjects received placebo second (i.e., during the second time they completed the task). To control for any effects of ordering or task repetition on behavior and brain activation, the number of the placebo session (first or second) was used as a covariate to capture variance associated with differences due to ordering in all behavioral and fMRI analyses. A confirmatory group comparison between subjects who received placebo at the first session compared to subjects who received placebo at the second session revealed no significant differences in brain activation for our contrast of interest (interoception > exteroception, see below).

Interoceptive Attention Task

This task is based on prior fMRI studies of interoception where subjects are asked to attend to their body sensations [Avery et al., 2015; Ernst et al., 2013; Farb et al., 2013; Simmons et al., 2013; Wiebking et al., 2014], with some modifications. In the task, subjects monitor their heartbeat and skin temperature (interoception conditions) and the blinking rate and brightness of words on a screen (exteroception control conditions) in 15 s blocks. While lying in the scanner before the task begins, heart rate and skin temperature are monitored for a period of ∼5 min, during which time subjects see these values represented as traces on a computer screen via a Biopac interface. During the task, for heartbeat (HB) blocks (n = 12), subjects see the word “HEARTBEAT” in the center of a screen, during which time they are asked to silently count how many heartbeats they detect over the 15 s period. At the end of the block, they rate how many beats they counted on a 5‐point Likert scale (up to 4 s for the rating response) that is populated with values derived from the measurement taken right before the task (e.g., someone with a heart rate of 60 beats per minute before the task started would have a Likert scale populated with the numbers 13, 14, 15, 16, and 17). For skin temperature (ST) blocks (n = 12), subjects see the word “SKIN TEMP” on the screen, during which time they are asked to monitor whether their skin temperature is increasing, decreasing, or not changing, on average, over the 15 s period, which is also followed by a rating (1 = “decreased a lot,” 2 = “decreased a little,” 3 = “no change,” 4 = “increased a little,” 5 = “increased a lot”). For the blinking (BL) condition (n = 12), subjects silently count how many times the word “NOTEBOOK” blinks on and off the screen. The blinking rate of the word varies (6, 8, 10, or 15 blinks per 15 s block), with the different rates distributed in equal numbers across the experiment. After this block, subjects rate the number of blinks they counted (1 = “7 or less,” 2 = “8,” 3 = “9,” 4 = “10,” 5 = “11 or more”). Finally, for the brightness (BR) condition (n = 12), subjects determine whether the luminance of the word “COMPUTER” is increasing, decreasing, or staying the same, on average, over the 15 s period. The actual change in word brightness varies between all 3 options (increasing, decreasing, and no change) in equal numbers across the experiment. After this block, subjects rate the change using the same scale as is used for skin temperature ratings (i.e., 1 = “decreased a lot,” 2 = “decreased a little,” 3 = “no change,” 4 = “increased a little,” 5 = “increased a lot”). After each rating, there is an intertrial interval (ITI) consisting of a fixation crosshair before the next block begins. The ITI is jittered between 2 and 10 s, which includes a “baseline” jitter between 2 and 5 s (in 1 s increments) plus any difference in time between the rating response time and the 4 s response deadline (i.e., if a subject takes 1 s to make their rating, the remaining 3 s get added to the ITI for that trial). Block types are presented in pseudorandom order and are equally distributed across 3 runs. Heart rate and skin temperature are measured throughout the task using a probe on the middle finger. After task completion, subjects filled out a debriefing questionnaire where they rated how difficult they found each of the conditions (on a scale of 1 = “very easy” to 5 = “very difficult”) and how accurate they thought their ratings were for each condition (on a scale of 1 = “not accurate at all” to 5 = “very accurate”).

We included both the HB and ST conditions to engage mechanisms of body‐state sensing arising from both visceral and somatic sources. Although our intention was to average across these interoception conditions to identify correlations with IS that were not specific to one particular sensory pathway, for completeness, we do investigate differences between the focus on visceral interoception (heartbeat condition) versus somatic interoception (skin temperature condition). Furthermore, we used a variant of the mental tracking task [Schandry, 1981] where subjects attend to and silently count their heartbeats rather than using a different but well‐established heartbeat discrimination paradigm where subjects judge whether their heartbeat is in or out of sync with an externally presented tone [Critchley et al., 2004; Katkin et al., 2001; Khalsa et al., 2008; Sokol‐Hessner et al., 2015; Whitehead et al., 1977; Wiens et al., 2000]. We chose this mental tracking approach in order to focus neuroimaging analysis on activation related to detecting body sensations in the absence of any interfering external stimuli.

Neuroimaging Data Acquisition and Preprocessing

MRI scanning occurred on a Siemens 3T Skyra scanner. Functional BOLD data were acquired using a high‐resolution multiband accelerated echo planar sequence for full brain coverage (TR = 1000 ms, TE = 35 ms, flip angle = 60°, FOV = 228 mm, 70 slices, 2.1 mm thickness, no gap, acceleration factor: 7). A high‐resolution T1 MPRAGE structural image was acquired for co‐registration (TR = 2,400 ms; TE = 2.06 ms; flip angle = 8°; FOV = 256 mm, 0.8 mm thickness). Heart rate and skin temperature were acquired during scanning (Biopac Systems Inc., Goleta, CA). Preprocessing was performed using Statistical Parametric Mapping (SPM; http://www.fil.ion.ucl.ac.uk/spm) v.12 and python/shell scripts taken from the Human Connectome Project (HCP) preprocessing pipeline [Glasser et al., 2013]. The following preprocessing steps were conducted: gradient nonlinearity distortion correction (HCP python and shell scripts), realignment of functional images (SPM), EPI distortion correction (HCP shell script), co‐registration of functional images to structural image, normalization to MNI template, and spatial smoothing with a 6‐mm kernel (SPM).

Data Analysis

MAIA scores

Ratings on the MAIA were highly correlated across several subscales in our sample. As a result, we performed a factor analysis with orthogonal rotation (SPSS, v. 23) to reduce the number of dissociable dimensions of IS that existed in our dataset and avoid running redundant correlational analyses. Three factors were retained that showed eigenvalues >1 [Kaiser, 1960; Yong and Pearce, 2013]. Table 1 shows the factor loadings that were >0.5 for each of the 3 components. The dimensions of the MAIA that loaded onto Factor 1 included: tendency to notice body sensations, ability to regulate negative emotion through attention to sensation, awareness of the link between emotion and sensation, the ability to sustain and control attention to sensation, and the tendency to listen to the body for insight. Dimensions that loaded onto Factor 2 were: experiencing the body as safe and the tendency to listen to the body for insight. Dimensions that loaded onto Factor 3 were the ability to sustain and control attention to body sensations, the tendency to not distract oneself from painful or uncomfortable sensations, and the tendency to not worry about and experience emotional distress in relation to painful or uncomfortable sensations. Together these three factors accounted for 80.5% of variance in scores, with Factor 1 accounting for 48.3%, Factor 2 accounting for 16.9%, and Factor 3 accounting for 15.3% of the variance.

Table 1.

Multidimensional Assessment of Interoceptive Awareness (MAIA) factor loadings

	Factor
MAIA subscale	1	2	3
Noticing	0.91
Regulating negative emotion through body	0.89
Linking emotion and body	0.88
Attentional control to body	0.67		0.60
Trusting		0.93
Listening	0.64	0.66
Not distracting			0.68
Not worrying			0.60

The scale consists of 8 subscales: Noticing (tendency to notice bodily sensations); Regulating negative emotion through body (ability to regulate negative emotion through attention to bodily sensations); Linking emotion to body (awareness of the link between emotion and bodily sensations); Attentional control to body (ability to sustain and control attention to bodily sensations); Trusting (experiencing the body as safe and trustworthy); Listening (tendency to listen to bodily sensations for insight into emotion and guide behavior); Not distracting (tendency to not distract oneself from bodily sensations); and Not worrying (tendency to not worry about or experience emotional distress in response to bodily sensations). Only those factor loadings with absolute value >0.5 are shown.

Behavioral measures

The study design was optimized for fMRI scanning and not for behavioral measures of interoceptive accuracy. However, for completeness we do report on accuracy measures and their relationship with MAIA factor scores. For the HB condition, accuracy calculations utilized a formula frequently used for heartbeat tracking tasks [Cali et al., 2015; Chong et al., 2016; Dunn et al., 2012; Fustos et al., 2013; Garfinkel et al., 2015; Grabauskaite et al., 2017; Herbert et al., 2007; Mirams et al., 2012; Pollatos et al., 2007a, 2005; Werner et al., 2009, 2014]: 1‐(|number of actual heartbeats − number of reported heartbeats|/number of actual heartbeats). This formula yields a number between 0 and 1, with 1 representing perfect interoceptive accuracy and any values ≥ 0.85 thought to reflect high IA [Herbert et al., 2007; Pollatos et al., 2007a, 2005; Werner et al., 2009, 2014]. However, given that our task implementation used a multiple‐choice response for reporting heartbeats [Ernst et al., 2013], there was by design a limit on how much the response could deviate from the actual number of heartbeats, and thus IA appeared to be quite high when utilizing this formula. Taking this issue into consideration, we also computed the percentage of errors for the HB condition by comparing whether a rating response matched the actual heart rate (divided by four and rounded to the nearest whole number).

For the ST condition, we could not find a precedent from the literature to guide us in the best method for assessing accuracy. First, we determined the overall percentage of errors, which included responses of an “increase” in skin temperature (large or small) when there was an actual decrease in skin temperature, responses of a “decrease” (large or small) in skin temperature when there was an actual increase in skin temperature, and responses of “no change” in skin temperature when there was an actual increase or decrease in skin temperature. However, the highly sensitive finger monitoring device recorded changes in skin temperature for all blocks, even if these changes were quite small, and thus all “no change” ratings made by subjects were calculated as errors using this method. To avoid this issue, we conducted a secondary analysis where we excluded these trials and computed accuracy as the percentage of errors where subjects reported an increase in skin temperature when there was an actual decrease and vice versa (“crossover error” analysis). Accuracy for the exteroception (control) conditions was calculated as the percentage of errors for the BL and BR conditions.

Post‐task ratings of task difficulty and perceived accuracy (confidence that in‐scanner responses were correct) were examined in 2 × 2 repeated‐measures analyses‐of‐variance (ANOVAs) with attentional focus (interoception vs exteroception) and type of change being monitored (discrete [HB and BL] vs continuous [ST and BR]) as factors.

Neuroimaging measures

A general linear model (SPM v. 12) was used to specify regressors for HB, ST, BL, and BR blocks separately with duration set to block length (15 s). A regressor for the rating period was included to capture variance, with duration set equal to response time. Six motion parameters were also included to reduce error variance associated with residual movement following realignment. Scans showing movement spikes of over 3 mm translation or 2° rotation were excluded and interpolated using ArtRepair [Mazaika et al., 2007]. Imaging contrasts focused on comparisons of interoception > exteroception ([HB + ST] > [BL + BR], HB > BL, and ST > BR). To compare interoception conditions, our main analysis compared HB directly with ST after excluding activations where exteroception conditions differed (i.e., HB > ST after excluding areas where BL > BR), which allowed us to identify neural differences between the two types of interoception while removing more general effects of type of judgment (discrete vs continuous). In a secondary analysis (results presented in the supplement), we compared the two different interoception > exteroception contrasts directly to each other ([HB > BL] > [ST > BR] and [ST > BR] > [HB > BL]). All group analyses of condition effects used one‐sample t tests, with placebo session [(1, −1) corresponding to second and first sessions] as a covariate to capture variance related to this effect. For the analysis of correlations between brain activity and MAIA factor scores, multiple regressions were run with the 3 factor scores separately (using the placebo session as a covariate of no interest [1, −1]). Stringent correction for voxel‐wise multiple comparisons utilized permutation testing, as suggested by Eklund et al. [2016] and Cox et al. [2016], with threshold‐free cluster enhancement [Smith and Nichols, 2009] as implemented by palm software [Winkler et al., 2014], with a whole‐brain corrected family‐wise error rate of P < 0.05.

RESULTS

Behavioral

Accuracy measures

Heartbeat detection accuracy using the formula described above revealed very high overall accuracy in the group (mean value = 0.92, range 0.82–0.97). However, as explained, given that heartbeat ratings used multiple choice responses, these values would be expected to be higher than those found in other studies where subjects reported the number of heartbeats without options from which to choose [Dunn et al., 2012; Fustos et al., 2013; Garfinkel et al., 2015; Herbert et al., 2007; Mirams et al., 2012; Pollatos et al., 2007a, 2005; Werner et al., 2009, 2014]. Accordingly, the error analysis painted a slightly different picture, revealing a high overall percentage of errors (mean = 76.03%, range 41.67–100%). As expected, results from the formula and error analyses were strongly negatively correlated (r = −0.75, P < 0.001).

For skin temperature accuracy, the overall error percentage was quite high (mean percentage errors: 87.6%, range 58.33–100%). However, as explained, due to the high sensitivity of the temperature monitor and the relative coarseness of the rating scale, all reports of “no change” were coded as errors. The analysis of “crossover errors,” where an increasing temperature is reported as decreasing and vice versa, revealed a much smaller frequency of these types of errors (mean: 17.67%, range: 0–41.67).

As might be expected, accuracy was higher for the exteroception (control) conditions. Error percentage was 18.26% (range: 0–50%) for the BL condition and 19.66% (range: 0–41.67%) for the BR condition. A 2 × 2 ANOVA indicated that there were significantly more errors for interoception compared to exteroception conditions (F _1,17 = 40.04, P < 0.001), but no effects of type of change being monitored (discrete vs continuous) and no interaction between factors.

Post‐task ratings

As might be expected given the differences in accuracy, interoception conditions (HB and ST) were rated as more difficult than exteroception conditions (BL or BR) (F _1,17 = 9.03, P = 0.008, average HB difficulty rating = 3.2 ± 0.26, ST = 3.7 ± 0.35, BL = 1.3 ± 0.17, BR = 2.3 ± 0.23). There were no other main effects or interactions.

Parallel findings were obtained for ratings of perceived accuracy of task performance. There was a main effect of interoception vs. exteroception conditions such that subjects thought their responses were less accurate during interoception (F _1,17 = 7.55, P = 0.014; HB = 2.7 ± 0.23, ST = 2.4 ± 0.33, BL = 4.4 ± 0.16, ST = 3.6 ± 0.21), but no other main effects or interactions. There was a trend for perceived accuracy to be higher for the discrete change (HB and BL) compared to continuous change (ST and BR) conditions (F _1,17 = 2.99, P = 0.10).

Correlations with MAIA factor scores

No significant correlations were found between accuracy measurements and factor scores. However, HB error percentage was correlated with Factor 1 scores at trend level (r = −0.43, P = 0.075), indicating that subjects who made fewer errors when counting heartbeats had higher scores on this factor (higher Factor 1 scores were associated with: increased tendency to notice body sensations, regulate negative emotion through attention to the body, sustain and control attention to the body, listen to the body for insight, and link emotion to the body). Post‐task ratings of difficulty and accuracy for the four conditions were not correlated with MAIA factor scores.

Scores on the BDI were correlated with Factor 3 scores (r = −0.57, P = 0.013), indicating that those subjects with greater subclinical depression symptoms had lower scores on Factor 3 (reduced ability to control attention to the body and increased tendency to distract from and worry about aversive body sensations). The BAI was correlated with Factor 3 scores at trend level (r = −0.44, P = 0.069). The BDI and BAI were not significantly correlated with the other two factor scores.

Neuroimaging

Effects of interoception

The main comparison of interoception > exteroception ([HB + ST] > [BL + BR]) yielded widespread activation in bilateral insula, somatosensory regions (paracentral lobule, postcentral gyrus, inferior parietal lobule), premotor cortex (supplementary motor area [SMA], precentral gyrus), superior temporal cortex, and occipital cortex (Table 2, Fig. 1, and Supporting Information, Fig. 1), replicating and extending prior work [Avery et al., 2015; Caseras et al., 2013; Critchley et al., 2004; Ernst et al., 2013; Farb et al., 2013; Pollatos et al., 2007b; Simmons et al., 2013; Tracy et al., 2007; Wiebking et al., 2014; Zaki et al., 2012] (see Schulz [2016] for meta‐analysis). In separate comparisons of each interoception condition to its respective exteroception condition, there was overlap for the HB > BL and ST > BR contrasts in bilateral mid insula (defined here as voxels located between y = −10 and +10), left posterior insula (defined here as voxels located posterior to y = −10), left inferior regions of somatosensory cortex (postcentral gyrus, inferior parietal), left precentral gyrus, and bilateral superior temporal cortex (Table 2 and Fig. 2). However, there were also significant differences between the two interoception conditions in the statistical comparison between HB and ST (Table 3 and Fig. 3). After excluding regions where BL > BR, HB showed greater activity than ST in SMA and dorsal ACC, mid and anterior insula (anterior insula defined here as voxels located anterior to y = +10), and precentral and inferior frontal gyri. After excluding regions where BR > BL, ST showed greater activity than HB in right posterior insula, superior regions of somatosensory cortex (postcentral gyrus), occipital cortex, temporal regions, and limbic areas including hippocampus and amygdala (Fig. 3). Results from analyses of (HB > BL) versus (ST > BR), and HB versus ST without excluding differences between exteroception conditions, are shown in Supporting Information, Figure 2.

Table 2.

Brain regions showing significant differences between interoception and exteroception

	BA	k	X	y	z	P corr
Interoception > exteroception ([HB + ST] > [BL + BR])
Superior/middle temporal (L)	21, 22, 38, 41	29,267	−64	−30	6	0.0007
Superior temporal/inferior parietal (L)	22, 40, 41, 42		−60	−26	16	0.0007
Superior temporal/precentral (L)	22, 42, 43		−54	−12	0	0.0008
Posterior insula (L)	13		−38	−32	18	0.0008
Lingual/PG (B)	18, 19, 36		−12	−52	−8	0.0008
Cuneus/precuneus/cingulate/SMA (B)	6, 7, 18, 19, 23, 24, 31, 32		−6	−90	22	0.0008
Postcentral/precentral/paracentral (L)	2, 3, 4, 5, 6		−22	−48	64	0.0009
Insula (ant, mid, post)/OFC/middle frontal gyrus (L)	8, 9, 10, 13, 47		−34	4	2	0.001
Postcentral/precentral/paracentral (R)	2, 3, 4, 5, 6		26	−44	62	0.002
Superior temporal (R)	21, 22, 41, 42	2,723	66	−16	2	0.002
Insula (mid, post) (R)	13		38	−12	6	0.005
Insula (post)/postcentral (R)	13, 40, 41		46	−32	12	0.007
Temporal pole (R)	38		50	12	−22	0.04
Heartbeat (HB) > blinking (BL)
Postcentral/inferior parietal (L)	1, 2, 3, 13, 22, 40, 41, 42	1,017	−64	−28	20	0.01
Superior temporal/precentral (L)	6, 22	1,222	−52	2	−2	0.02
Inferior frontal (L)	44, 45		−56	10	10	0.02
Insula (mid, post) (L)	13		−42	−4	−2	0.03
Insula (ant) (L)	13, 47		−30	26	0	0.04
Superior temporal/precentral/insula (ant, mid)/OFC (R)	13, 22, 44, 45, 47	457	54	6	−2	0.03
Inferior/middle frontal (L)	46	108	−46	34	14	0.04
Middle frontal (L)	10	28	−36	54	12	0.05
Skin temperature (ST) > brightness (BR)
PG/fusiform (L)	36, 37	34,698	−32	−28	−20	0.0001
Superior occipital (L)	19, 39		−40	−80	24	0.0001
Cuneus (B)	17, 18, 19		12	−96	14	0.0001
Precuneus/PCC/paracentral (B)	7, 23, 29, 31		16	−48	64	0.0001
Superior occipital (R)	19		32	−88	22	0.0001
PG/lingual gyrus (R)	30, 36, 37		26	−56	−8	0.0001
Postcentral/precentral (R)	1, 2, 3, 4, 5, 6		22	−36	58	0.0001
Postcentral (L)	1, 2, 3, 4, 5		−28	−36	66	0.0006
VMPFC (B)	10, 11		−2	48	−2	0.001
Temporal/postcentral (L)	21, 22, 38, 40, 41, 42, 43		−56	−10	4	0.001
ACC (B)	24, 32, 33		2	28	16	0.002
Insula (mid, post) (L)	13		−36	−14	0	0.003
Middle/superior frontal (L)	8, 9		−34	26	40	0.004
Superior/middle temporal/precentral (R)	6, 21, 22, 41, 42, 43	2,901	60	2	4	0.01
Inferior temporal (R)	20, 38		62	−10	−26	0.01
Insula (post)/postcentral (R)	13, 40		44	−14	18	0.01
Amygdala (R)	38		32	4	−20	0.02
Angular gyrus (R)	39	171	50	−56	22	0.02
OFC (L)	47	26	−36	30	−16	0.04
Paracentral/SMA (L)	6	14	−12	−14	66	0.05
Conjunction of heartbeat > blinking and skin temperature > brightness ([HB > BL] ∩ [ST > BR])
Superior temporal/insula (mid, post), precentral (L)	6, 13, 22	487	−38	−10	−12	–
Inferior parietal/superior temporal/postcentral/insula (post) (L)	2, 13, 22, 40, 41, 42	610	−62	−38	10	–
Superior temporal/insula (mid) (R)	22	68	54	2	−8	–

Data are whole‐brain corrected at a family‐wise error rate of P < 0.05 using threshold‐free cluster enhancement and nonparametric permutation analysis.

HB, heartbeat; ST, skin temperature; BL, blinking; BR, brightness; PG, parahippocampal gyrus; SMA, supplementary motor area; ant, anterior; mid, middle; post, posterior; OFC, orbitofrontal cortex; PCC, posterior cingulate cortex; VMPFC, ventromedial prefrontal cortex; ACC, anterior cingulate cortex; ∩, intersection. BA, Brodmann's areas; k, cluster extent; P _corr, whole‐brain corrected P value; R, right, L, left, B, bilateral; coordinates are in MNI space.

Bar graphs of parameter estimates for each condition (extracted from 6‐mm‐radius spheres located around the listed coordinates) are displayed in Figure 1 for the seven areas highlighted in bold font for the interoception > exteroception contrast. Bar graphs of parameter estimates for all coordinates listed for this contrast are shown in Supporting Information, Figure 1.

Figure 1.

Effects of interoception > exteroception, averaged across conditions ([HB + ST] > [BL + BR], areas in red) and for each condition separately (HB > BL [orange] and ST > BR [blue]). Bar graphs show parameter estimates for each condition extracted from 6‐mm‐radius spheres located around coordinates listed (also highlighted in bold font in Table 2. HB, heartbeat; ST, skin temperature; BL, blinking; BR, brightness. Color bars represent whole‐brain‐corrected P values. [Color figure can be viewed at http://wileyonlinelibrary.com]

Figure 2.

Conjunction analysis showing areas where HB > BL and ST > BR. HB, heartbeat; ST, skin temperature; BL, blinking; BR, brightness. Coordinates for the areas displayed are listed in Table 2. [Color figure can be viewed at http://wileyonlinelibrary.com]

Table 3.

Brain regions showing significant differences between heartbeat and skin temperature conditions (exclusive of differences between exteroceptive control conditions)

	BA	k	x	y	z	P corr
Heartbeat > skin temperature exclusive of blinking > brightness ([HB > ST]\[BL > BR])
SMA/ACC (B)	6, 24, 32	994	6	2	60	0.004
Insula (ant, mid)/precentral/inferior frontal (R)	6, 13, 22, 44, 47	702	40	12	0	0.02
Insula (ant, mid)/precentral/inferior frontal (L)	6, 9, 13, 44	383	−42	9	0	0.03
Precentral (R)	6, 9	251	56	2	42	0.03
Precentral (L)	6	28	−50	−6	52	0.04
Skin temperature > heartbeat exclusive of brightness > blinking ([ST > HB]\[BR > BL])
Cuneus/lingual gyrus/posterior temporal/angular gyrus (B)	17, 23, 30, 39	19,465	−4	−66	8	0.0006
Precuneus/PCC (B)	23, 29, 30, 31		10	−56	34	0.001
Cuneus/superior occipital gyrus (B)	18, 19		20	−68	12	0.001
Fusiform/PG/hipp/amygdala (R)	20, 30, 36, 37		26	−40	−12	0.003
Fusiform/PG/hipp (L)	30, 36, 37		−30	−42	−14	0.005
Cerebellum (R)	n/a		24	−76	−34	0.008
Postcentral/precuneus/paracentral/precentral (R)	4, 5, 6, 7, 40		14	−50	66	0.009
Postcentral/paracentral (L)	1, 2, 3, 5, 7		−24	−40	72	0.01
Insula (post)/temporal (R)	13, 20, 21, 22, 43	776	34	−18	16	0.02
Middle temporal/temporal pole (R)	21	125	52	4	−34	0.03
Cerebellum (L)	n/a	32	−28	−78	−36	0.04

Data are whole‐brain corrected at a family‐wise error rate of P < 0.05 using threshold‐free cluster enhancement and nonparametric permutation analysis.

HB, heartbeat; ST, skin temperature; BL, blinking; BR, brightness; \, exclusive of; SMA, supplementary motor area; ACC, anterior cingulate cortex; ant, anterior; mid, middle; PCC, posterior cingulate cortex; PG, parahippocampal gyrus; hipp, hippocampus; post, posterior; BA, Brodmann's areas; k, cluster extent; P _corr, whole‐brain corrected P value; R, right, L, left, B, bilateral; coordinates are in MNI space.

Figure 3.

Comparisons between interoception conditions. The HB > ST contrast (areas in orange) displays results after excluding regions where BL > BR and the ST > HB contrast (areas in blue) displays results after excluding regions where BR > BL. HB, heartbeat; ST, skin temperature; BL, blinking; BR, brightness. Color bars represent whole‐brain‐corrected P values. Coordinates for the areas displayed are listed in Table 3. [Color figure can be viewed at http://wileyonlinelibrary.com]

Given that participants rated interoception as more difficult than exteroception and that error percentage was higher for interoception, we conducted post‐hoc analyses examining whole‐brain differences between interoception and exteroception after including difficulty ratings (Analysis 1) and error percentage (Analysis 2) for HB and ST as covariates. These post‐hoc analyses were performed on the overall interoception > exteroception contrast ([HB + ST] > [BL + BR]) as well as the individual comparisons (HB > BL and ST > BR). Results from these post‐hoc comparisons were almost exactly the same as for the main analyses (see Supporting Information, Fig. 3), suggesting that the findings were not related to differences in perceived difficulty or error percentage.

Correlations with MAIA factor scores

When examining correlations between MAIA factor scores and the interoception > exteroception contrast ([HB + ST] > [BL + BR]), there were no significant relationships with Factor 1 or 2 scores. Factor 3 scores were significantly negatively correlated with activation in occipital cortex, cingulate cortex, SMA and precentral gyrus, mid/anterior insula, and somatosensory regions (paracentral lobule and postcentral gyrus) (Table 4 and Fig. 4). The effects were in the direction of subjects with lower Factor 3 scores (i.e., subjects with reduced ability to sustain and control attention to sensations, greater tendency to distract from aversive sensations, and greater worry and distress about aversive sensations) showing increased activity in these regions. Several of the regions identified in this correlational analysis were also found for the interoception > exteroception group contrast (Fig. 1).

Table 4.

Brain regions showing significant correlations with Factor 3 scores of the Multidimensional Assessment of Interoceptive Awareness

	BA	k	x	y	z	P corr
Correlation with Factor 3 for interoception > exteroception ([HB + ST] > [BL + BR])
Cuneus/lingual gyrus/fusiform/precuneus/PCC/PG/cerebellum (B)*	7, 17, 18, 19, 23, 30, 31, 36, 37	6,850∧	16	−76	8	0.001
Precentral/postcentral (L)*	3, 4	128	−22	−24	58	0.020
Precuneus/paracentral/postcentral/precentral (R)*	4, 5, 6, 7, 31	312+∧	14	−36	52	0.022
ACC/SMA/mid cingulate (B)*	6, 24, 32, 33	378+∧	−6	10	26	0.024
ACC (rostral) (B)	24	82+∧	2	38	8	0.027
Paracentral (L)	4, 5	56	−8	−38	62	0.028
Mid cingulate/PCC (L)*	N/A	21∧	−12	−20	36	0.029
Mid cingulate/PCC (R)*	31	19∧	8	−30	40	0.030
Precuneus/paracentral (L)*	5, 7	51	−16	−42	48	0.030
Mid cingulate (B)*	24	30+∧	−4	−8	30	0.031
Superior temporal/insula (mid, ant) (L)*	13, 22, 38	114+∧	−52	0	−6	0.035
Middle/superior frontal (L)	6	55+∧	−26	6	50	0.040
Superior temporal/precentral (L)*	6, 22, 44	36+∧	−60	4	4	0.040
Middle frontal (R)	9	13∧	42	18	36	0.043
Precentral/postcentral (L)	6	66+∧	−60	0	24	0.044
SMA (L) *	6	21+	−2	−12	64	0.047

Data are whole‐brain corrected at a family‐wise error rate of P < 0.05 using threshold‐free cluster enhancement and nonparametric permutation analysis.

HB, heartbeat; ST, skin temperature; BL, blinking; BR, brightness. PCC, posterior cingulate cortex; PG, parahippocampal gyrus; ACC, anterior cingulate cortex; SMA, supplementary motor area. BA, Brodmann's areas; k, cluster extent; P _corr, whole‐brain corrected P value; R, right; L, left; B, bilateral; coordinates are in MNI space.

To identify diverse subclusters, the first 10 clusters listed were derived from one large cluster (k = 11,561) by thresholding at a whole‐brain corrected P ≤ 0.03. Areas that were activated for the interoception > exteroception contrast ([HB + ST] > [BL + BR]) are denoted with asterisks. Clusters where activation was correlated with Factor 3 scores at trend or greater significance (P < 0.10) within each interoception contrast separately are noted (+ for the HB > BL contrast and ^∧ for the ST > BR contrast). Areas highlighted in bold font are shown in Figure 4 scatterplots.

Figure 4.

Correlations between Factor 3 scores and brain activation for the interoception > exteroception contrast ([HB + ST] > [BL + BR]). Reduced Factor 3 scores (associated with reduced ability to sustain and control attention to sensation and greater distraction from and worry about aversive sensation) were linked to increased activity in the regions displayed. Color bar represents whole‐brain‐corrected P value. Scatterplots depict the negative relationship between Factor 3 score (x axis) and parameter estimates (beta values) for the interoception > exteroception contrast (y axis) (see Table 4, areas highlighted in bold font). Coordinates for the areas displayed are listed in Table 4. [Color figure can be viewed at http://wileyonlinelibrary.com]

To determine whether any of the 3 subscales that loaded onto Factor 3 (controlling attention to sensation; not distracting from sensation; not worrying about sensation) was preferentially driving these correlations, we compared correlation coefficients of the relationship between each region and each of these subscales in a pairwise fashion using Williams’ test for dependent correlations [Weaver and Wuensch, 2013; Williams, 1959]. There was a significant difference between correlation coefficients for the SMA cluster (k = 21 in Table 4), such that the negative correlation was stronger with the not‐distracting subscale than the not‐worrying subscale (t = 2.50, P =0.024). There were no other significant differences between subscales in the strength of the correlation between brain activity and subscale score.

Correlations for HB and ST separately

Given that there were differences between the two interoception conditions (HB vs ST), we sought to investigate whether the regions identified in the correlational analysis (Table 4 and Fig. 4) were driven preferentially by either condition. Parameter estimates for clusters in Table 4 were extracted for the contrasts of HB > BL and ST > BR and correlated, post‐hoc, with Factor 3 scores. For the HB > BL contrast, activation in clusters encompassing cingulate cortex and SMA (k = 378), rostral ACC (k = 82), superior temporal gyrus and insula (k = 114), middle and superior frontal gyri (k = 55), superior temporal and precentral gyri (k = 36), and precentral and postcentral gyri (k = 66) was significantly negatively correlated with Factor 3 scores (noted with crosshair symbol in Table 4). At trend level, activity in three additional clusters (precuneus, paracentral lobule, postcentral and precentral gyri, k = 312; mid cingulate, k = 30; and SMA, k = 21) was also negatively correlated with Factor 3 scores. For the ST > BR contrast, activation in clusters encompassing occipital cortex (k = 6850), precuneus, paracentral lobule, and postcentral and precentral gyri (k = 312), rostral ACC (k = 82), mid and posterior cingulate (k = 21), superior temporal and precentral gyri (k = 36), middle frontal gyrus (k = 13), and precentral and postcentral gyri (k = 66) was significantly negatively correlated with Factor 3 scores (noted with caret symbol in Table 4). At trend level, activity in five additional clusters (cingulate and SMA, k = 378; mid cingulate, k = 30; mid and posterior cingulate, k = 19; superior temporal gyrus and insula, k = 114; and superior and middle frontal gyri, k = 55) was also correlated negatively with Factor 3 scores.

To determine the extent to which regions correlating with Factor 3 scores were also those that were more active for interoception compared to exteroception, we performed a whole‐brain conjunction analysis for areas showing significant: (1) correlations with Factor 3 scores for the interoception > exteroception contrast, (2) activity for the HB > BL contrast, and (3) activity for the ST > BR contrast. Activation in one cluster encompassing regions of left superior temporal, mid insula, and precentral gyrus (x = −52, y = 2, z = −8, k = 108, Brodmann's Areas 6, 13, and 22) overlapped for all three analyses.

Role of depression severity on correlations

Given the correlation between BDI and Factor 3 scores, we conducted partial correlations between extracted parameter estimates from the interoception > exteroception contrast and Factor 3 scores, controlling for BDI score. All clusters shown in Table 4 remained significantly related to Factor 3 scores (all P values < 0.02), with the exception of the precuneus, paracentral, postcentral, and precentral gyri cluster (k = 312), which became marginally significant (r = −0.44, P = 0.07), and the occipital cluster (k = 6,850, r = −0.30, P = 0.23). Indeed, the occipital cluster was the only region from Table 4 that was also significantly related to BDI scores (r = 0.59, P = 0.008).

DISCUSSION

This study investigated the neural correlates of interoception and its relationship to dimensions of self‐rated interoceptive sensibility (IS) in a group of healthy adults. Findings revealed greater activation in several cortical regions for interoception compared to exteroception, including insula and sensorimotor regions (postcentral gyrus, inferior parietal lobule, paracentral lobule, precentral gyrus, SMA) and occipital cortex, temporal cortex, ACC, and lateral prefrontal regions. Several of these areas were also correlated with Factor 3 scores obtained through analysis of MAIA responses. These data replicate and extend prior investigations into the neural mechanisms of interoception, and indicate that certain dimensional aspects of IS are associated with differential neural activation when attending to body sensation.

In the comparison of interoception versus exteroception, we have replicated prior studies pointing to a critical role for insula and sensorimotor regions [Avery et al., 2015; Caseras et al., 2013; Critchley et al., 2004; Ernst et al., 2013; Farb et al., 2013; Pollatos et al., 2007b; Simmons et al., 2013; Tracy et al., 2007; Wiebking et al., 2014; Zaki et al., 2012]. Indeed, a very recent meta‐analysis revealed that attention to heartbeat most consistently activated bilateral insula as well as premotor regions (precentral gyrus and SMA) [Schulz, 2016]. We additionally found several other areas that showed greater activation for interoceptive than exteroceptive attention, including temporal cortex, dorsolateral prefrontal cortex, OFC, and occipital and medial parietal cortex, consistent with other studies [Avery et al., 2015; Farb et al., 2013; Pollatos et al., 2007b; Simmons et al., 2013; Terasawa et al., 2013a; Zaki et al., 2012]. When comparing each interoception condition (heartbeat and skin temperature) to exteroception separately, we found a more widespread network for skin temperature than heartbeat monitoring. Some of the differences in brain activation between heartbeat and skin temperature conditions could be related to differential pathways for visceral and somatic sensation. Prior fMRI studies comparing the effects of visceral and somatic stimulation on brain activity have found overlap in activation of the insula [Eickhoff et al., 2006; Hobday et al., 2001; Lotze et al., 2001] and inferior regions of somatosensory cortex near the parietal operculum [Hobday et al., 2001; Lotze et al., 2001], but variability in the extent of the involvement of superior regions of somatosensory cortex [Hobday et al., 2001], motor and premotor regions [Lotze et al., 2001], and ACC [Hobday et al., 2001]. The present findings are consistent with these prior studies in identifying overlap within mid and posterior insula and adjacent (inferior) somatosensory regions for the heartbeat and skin temperature conditions. Furthermore, similar to prior work [Hobday et al., 2001; Lotze et al., 2001], attention to somatic information (i.e., in the skin temperature condition) activated superior somatosensory cortex (in the postcentral gyrus) more than attention to visceral information (i.e., in the heartbeat condition). Interestingly, whereas both the heartbeat and skin temperature conditions showed similar activation in mid insula, the heartbeat condition showed greater activity in anterior insula and skin temperature showed greater activation in posterior insula.

Many other differences between the two interoception conditions were identified that have not been previously reported (e.g., in occipital cortex, temporal cortex, and limbic regions) and cannot be readily attributed to differences in anatomical pathways between somatic and visceral afferents. In particular, the robust occipital activation for the skin temperature condition was surprising given that visual stimuli for the skin temperature and heartbeat conditions were the same (i.e., words on a screen). Prior research has found regions of lateral occipital cortex (extrastriate body area and the fusiform body area) to be involved in body processing, not only when viewing images of the human body and body parts [Costantini et al., 2011; Taylor et al., 2007; Urgesi et al., 2007], but also when engaging in mental imagery of embodied self‐location [Arzy et al., 2006] and experiencing illusory body ownership [Limanowski et al., 2014]. Relevant for this study, greater activity in these regions has been found for whole‐body compared to body part representations [Taylor et al., 2007]. It is conceivable that the increased occipital activity for the skin temperature condition reflects the fact that monitoring skin temperature may entail a greater focus on the body as a whole than does monitoring heartbeat. Overall, the differences between the heartbeat and skin temperature conditions provide support for the notion that attention to visceral and somatic signals engages distinct neural processes, while the areas of overlap revealed in the conjunction analysis also point to some amount of commonality between the two types of body processing.

We also obtained responses on the MAIA, a self‐report measure of interoceptive sensibility that assesses multiple different aspects of interoception. Owing to the presence of significant interrelationships in our sample between the different subscales of the MAIA, we conducted a factor analysis that revealed 3 independent factors that accounted for the majority of the variance in scores. It is noteworthy that 2 out of the 3 dimensions that loaded onto Factor 3—the tendency to (not) worry or experience distress in response to painful or uncomfortable body sensations and the ability to sustain and control attention to body sensations—have been previously linked to psychiatric health and disease. In particular, increased worry about body sensations is associated with panic disorder [Clark, 1986; Rudaz et al., 2010; Salkovskis et al., 1996; Yoris et al., 2015] and illness anxiety/hypochondriasis [Abramowitz, 2005], whereas the ability to control attention to body sensation (e.g., such as in mindfulness meditation) has been shown to be psychologically beneficial [Farb et al., 2015; Mehling, 2016]. In our study, reduced Factor 3 scores (related to reduced control, increased distraction, and increased worry) were associated with higher subclinical depression severity (as measured by the BDI) and, at trend level, anxiety severity (as measured by the BAI), whereas other factor scores were not related to these symptoms. Indeed, prior work has found that greater distraction and worry on the MAIA (i.e., reduced scores on not‐distracting and not‐worrying subscales) are related to increased emotion susceptibility, trait anxiety, pain catastrophizing, and difficulties in emotion regulation [Cali et al., 2015; Mehling et al., 2012], whereas greater scores on the attentional control subscale are associated with reductions on these same measures [Cali et al., 2015; Mehling et al., 2012]. Interestingly, in prior work, scores on the not‐worry subscale were unrelated to a measure of general sensitivity to body sensations as assessed by the Body Consciousness Questionnaire [Mehling et al., 2012]. Our results are consistent with this finding, showing that the not‐worrying and noticing subscales of the MAIA loaded onto separate factors (Factor 1 and 3, respectively).

When examining relationships with brain functioning, Factor 3 was the only factor that was correlated with neural activity during interoception. Greater activity was associated with reduced Factor 3 scores (i.e., reduced attentional control and greater distraction and worry) in several regions including mid/anterior insula, precentral gyrus, postcentral gyrus, SMA, paracentral lobule, cingulate cortex, and occipital regions. With the exception of the occipital cluster, all brain areas remained negatively related to Factor 3 scores even after controlling for BDI score. Many of the regions that were correlated with Factor 3 scores for the interoception > exteroception contrast were correlated with both heartbeat and skin temperature conditions compared to their respective exteroceptive conditions, including areas in precentral and postcentral gyri, precuneus and paracentral lobule, cingulate and SMA, superior temporal gyrus and insula, and middle and superior frontal gyri. Despite these commonalities, certain regions were correlated with Factor 3 scores only for the skin temperature contrast, including occipital cortex and posterior cingulate. These were areas that were more activated for skin temperature than heartbeat conditions, suggesting that some of the neural correlates of IS are specific to type of interoceptive focus. A whole‐brain conjunction analysis of regions that were both correlated with Factor 3 scores and more active for heartbeat and skin temperature conditions compared to their respective exteroception conditions revealed a single cluster encompassing regions of superior temporal gyrus, mid insula, and precentral gyrus. These data suggest that interindividual differences in self‐rated IS related to increased distraction, increased worry, and reduced attentional control are associated with differential engagement of brain regions that are involved in monitoring body state.

Consistent with prior work [Garfinkel et al., 2015; Meessen et al., 2016; Mehling, 2016; Whitehead et al., 1977], our measures of interoceptive accuracy were not significantly related to IS factor scores. However, we did find that greater Factor 1 scores were associated with reduced errors when monitoring heartbeats at trend level. Similar effects have been reported by others, with greater accuracy on heartbeat detection tasks associated with increased scores on the attentional control subscale of the MAIA [Cali et al., 2015] and the Body Awareness Questionnaire [Chong et al., 2016]. The fact that IA in our task was related to Factor 1 scores and not the other factor scores raises the interesting possibility that certain dimensional aspects of IS may be more related to interoceptive accuracy than others.

There are limitations to this work that should be addressed in future studies. First, our study design was optimized for fMRI scanning and subjects made their accuracy rating using multiple choice options in the scanner, which likely increased accuracy overall, at least for the heartbeat condition. Although the investigation of interoceptive accuracy was not a main aim of this work, we do discuss the findings from these analyses, which should be interpreted with caution. Future work examining the relationship between interoceptive accuracy, sensibility, and brain function would benefit from comparing measures of IA obtained outside the scanner with dimensional measures of IS and brain function. Another limitation to this work is that visual stimuli were not exactly matched between interoception and exteroception conditions, and some of the occipital differences found for ST > BR contrast could be related to this difference. However, it should be noted that differences in visual stimuli between interoception and exteroception conditions would not alter the interpretation of the comparisons between heartbeat and skin temperature conditions (which were matched on visual stimuli but showed large occipital differences) or the analysis of correlations with MAIA factor scores (which focused on between‐subjects differences across equivalent task stimuli). Nevertheless, future work could improve on our task design by providing interoception probes that exactly matched exteroception probes. It should also be noted that some of the post‐hoc correlations with Factor 3 scores within the individual interoception contrasts (HB > BL and ST > BR) only reached trend‐level significance. Given that the correlations were significant at a whole‐brain corrected threshold in the overall interoception > exteroception contrast, we felt it was still informative to examine activation in these areas for the two different types of interoceptive focus despite the lack of correction for multiple comparisons in these post‐hoc analyses. Future work with more trials for both types of interoception will be necessary to confirm the present findings. Finally, our characterization of the heartbeat condition as involving “visceral” interoception and the skin temperature condition as involving “somatic” interoception involves assumptions that were not directly tested in this study. Prior work suggests that information regarding organs can be obtained via somatic pathways innervating nearby skin or muscle (such as monitoring heartbeat by attending to sensation arising from the chest wall) in addition to visceral pathways [Khalsa et al., 2009]. As such, interpretation of the differences between heartbeat and skin temperature conditions should take into account the fact that some amount of somatic processing may have been engaged during the heartbeat condition as well. Despite these limitations, this study takes the novel approach of linking neural mechanisms of interoception to dimensions of self‐rated interoceptive sensibility. Findings revealing altered activation of insula, sensorimotor regions, and cingulate cortex have implications for the understanding of psychological health and disease as it relates to the processing of body state.

Supporting information

Supporting Information Figures.