Interoceptive Exposure Impacts Food-Cue Extinction in Adolescents With Low-Weight Eating Disorders: An fMRI Study

Abstract

Objective:

This randomized controlled study tested the effect of interoceptive exposure on anterior insula function and connectivity for the extinction of palatable and rotten food-cue associations in female adolescents with low weight eating disorders (LWED).

Method:

Thirty-nine female adolescents with LWED and 19 matched controls performed a food-related conditioning paradigm during functional magnetic resonance imaging (fMRI). Adolescents with LWED were then randomized to six sessions of either interoceptive exposure (n = 18) or family-based (n = 21) treatment, followed by a second fMRI. Whole brain activation and insula-driven connectivity for the extinction of palatable and rotten food-cue associations were compared between groups and changes over treatment were compared between the two therapies.

Results:

Adolescents with LWED exhibited diminished bilateral anterior insula activation for the extinction of palatable food-cue associations compared to controls (T_1,55 = 3.9–4.1, p values < .001; Hedges g = 0.47–0.55). Brief interoceptive exposure treatment increased left anterior insula activation for the extinction of palatable food-cue associations (T_1,37 = 5.10, p <.001; Hedges g = 1.59) and non-significantly improved palatability ratings for these associations during extinction compared to family-based treatment (β = −1.492, p =.087). There were no effects of group or therapy on connectivity or activation for rotten food-cue associations.

Conclusion:

These results suggest that targeting food avoidance in female adolescents with LWED using interoceptive exposure engages anterior insula regions that mediate the visceral sensation of disgust and may underlie the resistance to extinction. The findings present a window into possible pathophysiological mechanisms of anorexia nervosa and other LWED.

INTRODUCTION

There is a need for novel evidence-based approaches to the treatment of the pathological food avoidance that defines anorexia nervosa and other low weight eating disorders (LWED). These disorders pose a public health challenge, with some of the highest morbidity and mortality rates of any psychiatric domain^1,2 and little to no long-term benefit of intensive treatment on weight gain or psychiatric status.³ Many of these treatments are based on translational models of food avoidance that emphasize fear of weight gain,⁴ but cannot explain the visceral sensation of revulsion that distinguishes avoidance behavior.⁵ This interoceptive feeling of disgust is evaluative in nature and persists beyond conscious appraisal,⁶ which is central to the development of the maladaptive patterns of food avoidance associated with LWED⁷ and renders these maladaptive patterns resistant to reappraisal or extinction by diverting attention.⁸ The visceral disgust response that epitomizes food avoidance represents a potential therapeutic target for novel interventions that reduce avoidance behavior without relying on conscious appraisal (e.g., counterconditioning).

Interoceptive exposure targets conditioned disgust responses to internal bodily sensations and has been used to treat a range of anxiety and somatic disorders.^9,10 The technique was adapted to target the maladaptive disgust conditioning purported to underlie food avoidance in LWED; the modified intervention applies counterconditioning to shift the valence of food cues from aversive to appetitive in order to enhance tolerance for the visceral discomfort that accompanies food consumption.¹¹ The modified exposure therapy targets interoceptive circuits centered in anterior insula cortex that mediate the subjective experience of disgust.^12,13 Viscerosensory inputs to posterior and mid-insula are elaborated along the caudal-to-rostral axis of the cortex and integrated with emotional salience and reward signals in anterior insula.¹⁴ Counter-conditioning is believed to shift the valence of food cues encoded in anterior insula activation¹⁵ and connectivity with amygdala and ventral striatum.^16,17 Aberrant anterior insula activation for valence encoding has been linked to the negative valuation of food implicated in the pathophysiology of LWED^18–20 and may be responsive to exposure therapy.²¹ The impact of interoceptive exposure on valence encoding of food may be partially mediated by extinction circuits that include ventromedial prefrontal cortex (vmPFC).¹⁷

This randomized controlled study tested the effect of a brief trial of interoceptive exposure therapy on anterior insula function and connectivity during the acquisition and extinction of appetitive and aversive food-cue associations in female adolescents with LWED. Adolescents with LWED and matched healthy controls were scanned with functional magnetic resonance imaging (fMRI) while performing a food-related conditioning task; adolescents with LWED were scanned a second time following six sessions of interoceptive exposure or family-based treatment. Based on findings from studies of taste processing,^18,19 we predicted that anterior insula, amygdala, ventral striatum, and vmPFC activation and connectivity would be increased for the acquisition and extinction of aversive food-cue associations and decreased for the acquisition and extinction of appetitive associations in female adolescents with LWED compared to healthy controls. Further, the results of a pilot study suggested that interoceptive exposure would normalize anterior insula activation and connectivity for the acquisition and extinction of both appetitive and aversive food-cue associations.²¹

METHOD

Study Participants

This study was approved by the institutional review board of the Icahn School of Medicine at Mount Sinai. Written informed consent was obtained from all participants and their parents. Sixty-three adolescents assigned female sex at birth were recruited from the Mount Sinai Eating and Weight Disorders Program for a larger randomized trial (ClinicalTrials.gov identifier: NCT02795455). Sampling was restricted to individuals assigned female at birth to comply with the request of the sponsor’s scientific review to reduce potential sex-based variance. Fifty-eight of the adolescents were scanned with fMRI at baseline and data from 39 adolescents who successfully completed post-treatment fMRI were included in the present analyses. The remaining adolescents did not complete the procedures because two were no longer interested, four were hospitalized, four were discontinued due to the pandemic, five slept or did not respond during fMRI, and scans for four were started prematurely. These adolescents did not differ in age, sex, or eating disorder severity from study completers. Twenty-seven age-matched females with no history of eating disorders were recruited from the local community and 24 of these adolescents were scanned with fMRI at baseline; data from five controls were excluded for excessive motion during fMRI.

The adolescents were screened for eligibility by study staff during an evaluation that included a clinical intake and semi-structured interview of participants with the Eating Disorder Assessment for DSM-5 (EDA-5),²² sections of the Schedule for Affective Disorders and Schizophrenia for School-Age Children-Present and Lifetime Version (K-SADS),²³ and the Columbia-Suicide Severity Rating Scale (C-SSRS). Participants also completed the Center for Epidemiological Studies Depression Scale for Children (CES-DC),²⁴ Youth Eating Disorder Examination Questionnaire (Y-EDE-Q),²⁵ and Disgust Scale-Revised (DS-R).²⁶ Inclusion criteria for adolescents with LWED included: i) female; ii) aged 12 to 18 years; iii) seeking treatment; iv) failure to maintain greater than minimally low body weight based on body mass index for age percentile and growth trajectory; v) clinically significant food restriction endorsed on the EDA-5; and vi) permission from pediatrician or other medical provider to receive outpatient care. Twenty-six (66.6%) of the adolescents with LWED were diagnosed with the restricting subtype of anorexia nervosa, nine (23.1%) adolescents were diagnosed with atypical anorexia nervosa, and four (10.3%) met criteria for the binge-purge subtype of anorexia nervosa, based on responses to the EDA-5. Mean Y-EDE-Q Global Score for adolescents with LWED was 2.5 ± 1.3.

Controls were 12 to 18 year-old females with no history of eating problems. Exclusion criteria for all participants included: i) history of a major psychiatric disorder based on responses on the K-SADS; ii) history of a learning disorder or developmental disability; iii) report of active suicidal ideation on C-SSRS; iv) current substance abuse; v) medications that impact interoception (e.g., anti-emetics) or performance (e.g., psychostimulants); and vi) medical conditions that influence eating or weight (e.g., diabetes mellitus). Clinical and demographic characteristics of the sample are presented in Table 1. Adolescents with LWED were younger, less racially diverse, weighed less, had a lower body mass index (BMI) and BMI percentile, and were rated as more depressed on the CES-DC and more sensitive to disgust on the DS-R than controls.

TABLE 1.

Demographic and Clinical Profile of Adolescent Females With Low Weight Eating Disorders (LWED) and Adolescent Female Controls

Characteristic	LWED (n = 39)		Control (n = 19)		Statistic (df)	p
Demographics
	Mean	(SD)	Mean	(SD)
Age (years)	15.8	1.8	16.9	1.5	t(56) = 2.31	.02
Years of education	9.4	2.1	10.3	1.3	t(56) = 1.92	.06
	n	(%)	n	(%)
Race					χ2(3) = 9.40	.02
African American	2	5.1	1	5.2
Asian	2	5.1	6	31.6
Caucasian	30	76.9	8	42.1
Mixed	5	12.8	4	21.1
Hispanic ethnicity	5	12.8	4	21.1	χ2(1) = 0.66	.42
Clinical severity
	Mean	(SD)	Mean	(SD)
Weight (lbs)a	102.3	15.5	123.0	23.4	t(54) = 3.29	.002
Body mass indexa	17.4	1.8	20.5	3.2	t(54) = 3.49	<.001
Body mass index %ilea	15.7	16.3	42.1	32.5	t(54) = 3.03	.004
CES-DC Total score	47.6	11.1	27.2	7.5	t(56) = 8.06	<.001
DS-R Total score	58.4	16.7	48.8	15.8	t(56) = 2.09	.04

Note: CES-DC = Center for Epidemiological Studies Depression Scale for Children; DS-R = Disgust Scale-Revised.

^aMeasure was missing for one participant in each group.

Study Design

Participants who met all eligibility criteria for the study returned for a baseline fMRI scan. Adolescents with LWED were randomly assigned to either interoceptive exposure or family-based treatment for six sessions and a follow-up fMRI. Mean ± SD length of time between fMRI was 53 ± 14 days for interoceptive exposure and 46 ± 22 days for family-based treatment (t₃₇ = 1.28, p = .21). Adolescents assigned to interoceptive exposure were rated significantly higher on the CES-DC at baseline than those assigned to family-based treatment (t₃₇ = 2.22, p = .03), although both groups were rated in the at-risk range (39.7 ± 11.3 versus 32.1 ± 9.9).²⁴ The two subgroups of adolescents with LWED did not differ at baseline in EDA-5 diagnosis, age, years of education, racial diversity, ethnicity, weight, BMI, BMI percentile, or ratings on the DS-R (all p > .10).

Interoceptive exposure targeted visceral sensitivity to increase tolerance for internal experience (e.g., disgust) and is described in detail elsewhere.¹¹ The intervention consisted of one session dedicated to the therapist explaining the concept of disgust aversion and the relation to food avoidance. Five additional sessions used a shake of unknown content and quantity, counterconditioning of food/visceral cues (e.g., positively valanced music during meal-time), and the practice of distress tolerance skills to maximize engagement in the exposure exercise. Family-based therapy for LWED used a standard protocol of parent-led contingency management to motivate eating and discourage food avoidance.²⁷ Participants were maintained in continued care for up to 20 total sessions following the 6-session intervention.

Food-Related Conditioning Procedure

The conditioning procedure performed during fMRI scans was adapted from a previously published paradigm.²⁸ Parallel versions of the paradigm that differed in trial order and stimulus images were counterbalanced across adolescents with LWED; paradigm version was also counterbalanced across controls.

The conditioning procedure consisted of separate acquisition and extinction runs of 8 minutes and 40 seconds that each began and ended with 20 second fixation periods. Each run contained 24 trials that paired a conditioned stimulus (CS) and an unconditioned stimulus (UCS). The CS consisted of two neutral visual stimuli (e.g., blue and green squares) with identical luminance that were presented at fixation for 2 to 8 seconds before the UCS was displayed to the left and right of the CS for 4 seconds (Figure 1). The UCS in the acquisition phase consisted of two sets of 12 images that depicted palatable and rotten versions of the same high-calorie food items (e.g., burritos). The food items were chosen for palatability and photographed in color before and after a two-week rotting period. The UCS in the extinction phase consisted of pixelated versions of the same 24 pictures that were intended to elicit a neutral response. The association between CS and UCS was consistent across the procedure (e.g. blue square = rotten, green square = palatable). Following a 1 to 4 second period of fixation, participants had 2 seconds to rate the palatability of the CS-UCS pair as “Yummy”, “Neutral”, or “Yucky”. The inter-trial interval was jittered from 3 to 12 seconds.

FIGURE 1. Food-Related Conditioning Paradigm.

Note: Schematic illustration of single palatable and rotten conditioning trials. Both trials began with the conditioned stimulus (CS) presented at fixation for 2 to 8 seconds before the unconditioned stimulus (UCS) was displayed to the left and right of the CS for 4 seconds. Following a 1 to 4 second period of fixation (+), participants had 2 seconds to rate the palatability of the CS-UCS pair as “Yummy”, “Neutral”, or “Yucky”. The inter-trial interval consisted of a fixation period that was jittered from 3 to 12 seconds.

MRI Acquisition

All images were acquired in a Siemens 3-T MAGNETOM Skyra (Siemens Medical Systems) scanner with a 32-channel head coil. Functional T₂*-weighted images depicting the blood oxygenation level-dependent (BOLD) signal were obtained in separate acquisition and extinction runs of 260 volumes each using a multi-band gradient-echo echo-planar imaging sequence (repetition time (TR) = 2 seconds, echo time (TE) = 35 milliseconds, multiband factor = 3, flip angle = 76º, field of view = 230 mm, 64 × 64 matrix, 2.56 × 2.56-mm in-plane resolution, 60 2.56-mm axial slices). A high-resolution T₁-weighted anatomical image was acquired using a magnetization prepared rapid acquisition gradient echo (MPRAGE) sequence for anatomical reference (TR = 2.4 seconds, TE = 2.06 seconds, flip angle = 8°, 256 × 256 matrix, 0.9 × 0.9-mm in-plane resolution, 224 0.9-mm axial slices).

Image Preprocessing

Functional images were preprocessed and analyzed using SPM12 software (Wellcome Trust Center for Neuroimaging, London, England). Two whole-brain BOLD time series that represented the acquisition and extinction runs for each scan session were realigned to the first image of the acquisition run, co-registered to the corresponding MPRAGE anatomical image, spatially normalized to the Montreal Neurological Institute template using parameters estimated from the corresponding MPRAGE image, and spatially smoothed with an 8 mm full width at half maximum Gaussian kernel.

First-Level Modeling of Neural Activation and Connectivity

Separate subject-specific general linear models (GLM) were conducted to fit beta weights to boxcar regressors for palatable and rotten CS-UCS pairs in the acquisition and extinction runs, as well as regressors for the food ratings and six motion parameters of no interest,²⁹ all convolved with the default SPM hemodynamic response function.³⁰ Each CS-UCS pair was modeled as a boxcar regressor that extended from the onset of the CS to the offset of the CS-UCS pair. The neural effects of food-related conditioning were modeled by applying linear contrasts to parameter estimates for palatable and rotten CS-UCS pairs separately in the acquisition and extinction runs of the baseline scan, resulting in four contrast images for each participant. The impact of therapy on these neural effects were modeled by applying linear contrasts to parameter estimates for the CS-UCS pairs in the baseline versus post-therapy scans, which generated four more contrast images per participant.

The generalized psychophysiological interaction (gPPI) of anterior insula for food-related conditioning and the effect of treatment were analyzed separately with the CONN v22 toolbox using a region of interest (ROI)-to-ROI approach.³¹ The left and right anterior insula ROIs were defined using masks developed by Lin and colleagues,³² while left and right amygdala were defined using masks from the “aal.002” atlas³³ and ventral striatum and vmPFC were delineated using bilateral masks developed by Safron and colleagues³⁴ and Ginty and colleagues,³⁵ respectively. The first-level regressors of interest and covariates were extracted from the subject-level models of activation. BOLD signals originating in white matter and cerebrospinal fluid were extracted using an anatomical component-based noise correction method (aCompCor) and were entered as nuisance covariates in the first-level GLM. The residual BOLD time series were then weighted by the appropriate predictor to derive condition-specific time series for the palatable and rotten CS-UCS pairs separately in the acquisition and extinction runs in the baseline scan and in the baseline versus post-treatment scans. The resultant contrast images were entered into the group analyses described below.

Statistical Analysis

Palatability Ratings.

Cumulative ordinal regression models with logit link function were conducted using mixed effects models with random intercepts implemented with the lme4 v1.1–27.1 and ordinal v2019.12–10 packages in the R programming language, v4.1 (http://cran.us.r-project.org). Palatability ratings were coded as Yucky = −1, Neutral = 0, and Yummy = 1. Fixed effects included Condition (Palatable vs. Rotten), Phase (Acquisition vs. Extinction), and Group (LWED vs. Control) in the analysis of baseline ratings and Condition (Palatable vs. Rotten), Phase (Acquisition vs. Extinction), Session (Baseline vs. 6-weeks), and Treatment (Interoceptive Exposure vs. Family-Based) in the analysis of the effects of treatment on the ratings. The Condition × Phase × Group and Condition × Phase × Treatment × Session interactions were the primary contrasts of interest in the tests of baseline ratings and treatment, respectively.

Neural Activation and Connectivity.

Second-level group analyses were conducted with separate random-effects GLMs that included age as a covariate. Subject-level contrast images for food-related conditioning and the impact of treatment on this conditioning were entered into a series of one- and two-sample t tests that examined within-group and between-group effects in the effects of interest, respectively. The resultant voxel-wise statistical maps were thresholded for significance using a cluster-size algorithm that protects against false-positive results.³⁶ The height (intensity) threshold for the analyses were set at an uncorrected voxel-wise level of p < .001 and a contiguous-voxel threshold, k, that was estimated to correct for multiple voxel comparisons at a cluster-level of p < .05 based on random field theory.

RESULTS

Palatability Ratings

There was no significant Condition × Phase × Group interaction effect on baseline ratings of palatability (see Table S1, available online). Still, the lower frequency of “Yummy” responses to palatable CS-UCS pairs in the extinction phase than the acquisition phase (57.3% vs. 33.3%; p < .00001) confirmed that the task elicited the expected extinction response. There was also no significant Condition × Phase × Treatment × Session interaction effect on palatability ratings (see Table S2, available online). However, adolescents with LWED made significantly fewer Yuck responses to palatable CS-UCS pairs in the extinction phase following interoceptive exposure than family-based treatment (49.5% vs. 34.17%; p = .019).

Neural Activation and Connectivity

Acquisition of Food-Related Associations.

Adolescents with LWED and controls activated similar regions in visual, anterior insula, inferior frontal, premotor, and anterior cingulate cortices (ACC) for the acquisition of both palatable and rotten CS-UCS associations at baseline (see Tables S1 and S2, available online, for the coordinates, volume, and significance level of activations for the acquisition of palatable and rotten CS-UCS associations, respectively). Direct comparison of baseline activation in the two groups revealed significantly reduced visual, cerebellar, premotor, and ventrolateral prefrontal activation for the acquisition of palatable CS-UCS pairings in adolescents with LWED compared to controls (Figure 2A). Among adolescents with LWED, interoceptive exposure significantly reduced activation for the acquisition of palatable CS-UCS associations in left vmPFC (peak voxel MNI coordinates: x = −6, y = 32, z = - 10; volume = 414 mm³; T_1,37 = 4.50; p < .001; Hedges g = 1.39) and left ACC (peak voxel MNI coordinates: x = −8, y =36, z =20; volume = 178 mm³; T_1,37 = 5.51; p < .001; Hedges g = 1.30) compared to family-based treatment (Figure 3). Interoceptive exposure also reduced activation for learning palatable CS-UCS associations in bilateral ventral striatum (peak voxel MNI coordinates: x = −16, y = 10, z = −8; volume = 1,950 mm³; T_1,16 = 6.48; p < .001), although this reduction in activation did not differ significantly from that for family-based treatment (p > .001; Hedges g = 0.09). There was no effect of treatment on activation for learning rotten CS-UCS associations and no differences in baseline activation or connectivity between the treatment groups.

FIGURE 2. Group Differences in Baseline Activation for Food-Related Conditioning.

Note: (A) Adolescents with LWED showed significantly reduced activation for the acquisition of palatable CS-UCS associations in left cerebellum (top left; peak voxel MNI coordinates: x = - 2, y = −72, z = −14; T_1,55 = 5.03; p < .001; Hedges g = 0.97), right cerebellum (top right; peak voxel MNI coordinates: x = 10, y =−62, z = −50; T_1,55 = 4.62; p < .001; Hedges g = 1.14), left ventrolateral prefrontal cortex (bottom left; peak voxel MNI coordinates: x = −26, y = 54, z = 2; T_1,55 = 4.13; p < .001; Hedges g = 1.10), and left premotor cortex (bottom right/bottom left; peak voxel MNI coordinates: x = −52, y =2, z =2; T_1,55 = 4.17; p < .001; Hedges g = 0.98) compared to controls. (B) Activation for the extinction of palatable CS-UCS associations was significantly reduced in left precuneus (top left; peak voxel MNI coordinates: x = −12, y = −72, z =32; T_1,55 = 5.29; p < .001; Hedges g = 1.29), right precuneus (top right; peak voxel MNI coordinates: x = 20, y = −70, z = 28; T_1,55 = 5.49; p < .001; Hedges g = 1.36), left (bottom left; peak voxel MNI coordinates: x = −34, y = 12, z = 4; T_1,55 = 3.91; p < .001; Hedges g = 0.45) and right anterior insula cortex (bottom left/bottom right; peak voxel MNI coordinates: x = 44, y = 4, z = 2; T_1,55 = 4.10; p < .001; Hedges g = 0.53), and right supplementary motor area (bottom right; peak voxel MNI coordinates: x = 2, y = 4, z = 64; T_1,55 = 4.64; p < .001; Hedges g = 1.13) in adolescents with LWED compared to controls. The figures were thresholded at an uncorrected voxel-wise level of p < .001 and a cluster extent of 25 contiguous voxels. L= left; R= right.

FIGURE 3. Differential Treatment Effects on Activation for the Acquisition of Food-Cue Associations.

Note: (A) Interoceptive exposure reduced activation for the acquisition of palatable food-cue associations in left vmPFC (peak voxel MNI coordinates: x = −6, y = 32, z = −10; volume = 414 mm³; T_1,37 = 4.50; p < .001; Hedges g = 1.39) and left ACC (peak voxel MNI coordinates: x = −8, y =36, z =20; volume = 178 mm³; T_1,37 = 5.51; p < .001; Hedges g = 1.30) compared to family-based treatment in adolescent females with LWED. The figure weas thresholded at an uncorrected voxel-wise level of p < .001 and a cluster extent of 25 contiguous voxels. (B) Mean beta weights (β) plotted separately for the pre- and post-treatment scans to illustrate the group differences in left vmPFC and ACC activation for the acquisition of palatable food-cue pairs. Error bars represent SEM.

Adolescents with LWED and controls exhibited a similar pattern of interhemispheric interaction in anterior insula and ventral striatal interaction with left amygdala for the acquisition of food-related associations (see Figure S1, available online). Psychophysiological interactions of vmPFC with left amygdala and ventral striatum differed between groups for the acquisition of palatable and rotten food-cue pairs, although not significantly (all p > .01). Neither treatment had an impact on connectivity for the acquisition of food-related associations (all p > .05).

Extinction of Food-Related Associations.

Similar patterns of activation in anterior insula, inferior frontal, premotor, ACC, and visual cortices for the extinction of food-related associations were found in adolescents with LWED and controls at baseline (for details of the activations for the extinction of palatable and rotten CS-UCS associations, see Table S3 and Table S4, available online). Comparison of activation in the two groups revealed significantly reduced bilateral anterior insula, supplementary motor area, and precuneus activation for the extinction of palatable CS-UCS pairings in adolescents with LWED compared to controls (Figure 2B). There were no group differences in activation for the extinction of rotten CS-UCS associations. Interoceptive exposure significantly increased activation in the ventral-most portion of left anterior insula for the extinction of palatable food-cue associations in comparison to family-based treatment (peak voxel MNI coordinates: x = −28, y = 28, z = 2; volume = 448 mm³; T_1,37 = 5.10; p < 0.001; Hedges g = 1.59) (Figure 4). There was no effect of treatment on activation for the extinction of CS-rotten food associations and no differences in baseline activation or connectivity between the treatment groups.

FIGURE 4. Differential Treatment Effect on Anterior Insula Activation for the Extinction of Food-Cue Associations.

Note: (A) Interoceptive exposure increased left anterior insula activation for the extinction of palatable food-cue associations compared to family-based treatment in adolescent females with LWED (peak voxel MNI coordinates: x = −28, y = 28, z = 2; volume = 448 mm³; T_1,37 = 5.10; p < 0.001; Hedges g = 1.59). The figure weas thresholded at an uncorrected voxel-wise level of p < .001 and a cluster extent of 25 contiguous voxels. (B) Mean beta weights (β) plotted separately for the pre- and post-treatment scans to illustrate the group differences in left anterior insula activation for the extinction of palatable food-cue pairs. Error bars represent SEM.

The extinction of food-related associations at baseline prompted a similar pattern of interhemispheric anterior insula interaction and ventral striatal interaction with left amygdala in both adolescents with LWED and controls (see Figure S1, available online). Adolescents with LWED also showed left amygdala interactions with vmPFC and left anterior insula that were not seen in controls. However, these group differences in connectivity were not significant (all p > .05). There were also no effects of treatment on connectivity for the extinction of food-related associations (all p > .05).

DISCUSSION

Adolescents with LWED showed a distinctive pattern of diminished activation in visual, precuneus, supplementary motor, and premotor regions for both the acquisition and extinction of palatable food-cue associations compared to a more diverse sample of healthy controls. More specifically, the results of this study confirm both the hypothesized deficit in anterior insula activation for the extinction of appetitive food-cue associations in adolescents with LWED compared to controls and the increase in this extinction-related anterior insula activation following a brief trial of interoceptive exposure compared to an abbreviated family-based treatment. Interoceptive exposure may have also reduced activation for the acquisition of appetitive food-cue associations in left vmPFC, left ACC, and bilateral ventral striatum, although there were no similar effects on palatability ratings, and the findings in the latter region did not differ from the comparison treatment and may have reflected practice or other non-specific shared factors. These effects of interoceptive exposure were unlikely to be related to differences in ratings of depression between the treatment groups since both groups were rated in the clinically significant range. Finally, the conditioning paradigm revealed no meaningful group differences in baseline anterior insula-driven connectivity for food-related conditioning or treatment effects on this connectivity.

The results point to anterior insula cortex as a potential therapeutic target for novel interventions for food avoidance driven by visceral sensations of aversive disgust. The distinctive pattern of reduced baseline activation for palatable food-cue associations seen in adolescents with LWED suggests that they reduced or diverted attention from cues associated with images of edible high-calorie food.³⁷ The diversion of attention may render the maladaptive patterns of eating behavior resistant to extinction and could explain the deficient anterior insula activation for the extinction of appetitive food-cue associations in adolescents with LWED at baseline in the current study.^6,8 The reduced anterior insula activation may represent a failure to shift the valence of cues previously paired with images of edible food and is consistent with a previous report of blunted anterior insula responses to sucrose in women recovered from anorexia nervosa.¹⁹ The conditioning paradigm may not have been sufficiently challenging or the analytic methods sensitive enough to detect deficits in anterior insula-driven connectivity previously reported in adolescents with anorexia nervosa³⁸ and women recovered from the disorder.³⁹ Nevertheless, the results suggest that resistance to extinction of maladaptive food-cue associations in adolescents with LWED may be driven by a perseverative bias in valence encoding rather than deficits in prefrontal extinction circuits.¹⁷ The results need to be confirmed in a more diverse sample better matched to the healthy controls. Therapeutic interventions for disgust-based food avoidance will need to address these disordered conditioning processes without relying on conscious awareness, particularly given the report of deficits in the prefronto-amygdala functional networks for conscious reappraisal in women with anorexia nervosa.⁴⁰

The impact of interoceptive exposure on extinction-related anterior insula activation implicates the region in the therapeutic mechanisms for the counterconditioning that forms the basis of the therapy.¹¹ Counterconditioning is a form of evaluative learning in which behavior (i.e., eating) is modified by association with a stimulus that has an opposing valence (e.g., positive).⁴¹ This shift in valence triggers a prediction error signal,⁴² which may improve evaluative conditioning through increased activation of anterior insula networks for emotional salience.^15,43 The baseline response of these anterior networks to food images predicted response to food exposure therapy in women with anorexia nervosa.²¹ Thus, the increase in extinction-related anterior insula activation following interoceptive exposure in the current study suggests that counterconditioning of food cues and visceral sensations of eating may have improved the valence re-encoding of these cues during extinction. The improved evaluative conditioning was reflected in marginally better palatability ratings for appetitive food-cue pairs during extinction and indicates that counterconditioning may overcome the resistance of these maladaptive learned behaviors to extinction.⁸ However, the current study found no evidence of the strengthened anterior insula-driven connectivity with ventral striatum and amygdala previously reported during counterconditioning.^16,43

The seemingly inhibitory effect of interoceptive exposure on extinction circuits centered in vmPFC and ACC for the acquisition of appetitive food-cue associations may have also contributed to improvements in food-cue conditioning in adolescents with LWED. These mediofrontal regions are integral to the extinction of conditioned aversive associations,¹⁷ by selectively reducing excitatory inputs to the basolateral nucleus of amygdala⁴⁴ and the separate populations of neurons that encode positively and negatively valenced stimuli.⁴⁵ The reduced vmPFC activation following interoceptive exposure in the current study may have disinhibited the intra-amygdala circuits that form new conditioned associations.⁴⁶ However, interoceptive exposure had no corresponding effect on the rather limited behavioral ratings obtained during the study (i.e., palatability).

The unique focus of this study on interoceptive circuits in anterior insula cortex that mediate the visceral sensation of disgust, together with the use of an innovative exposure treatment to target these interoceptive circuits in food avoidance, present a window into possible pathophysiological mechanisms of anorexia nervosa and other LWED. Optimization of interoceptive exposure through higher salience cues (e.g., pictures and odors), longer duration of exposure intervention, or enhanced activation of visceral signals (e.g., barostat distention) are likely to further enhance the impact on anterior insula connectivity.

Data Sharing:

Deidentified participant data will be made available in the NIMH Data Archive https://nda.nih.gov/edit_collection.html?id=2489