Supplementary Figure 1

Transcription

1 Supplementary Figure 1 Health and Taste rating behavior. Behavior in Health and Taste rating phases (n = 42). (a) When rating items for Healthiness, both groups rated the high-fat items lower (F(1,40) = 802.9, P = 4.3*10-28 ). AN additionally rated foods lower overall (F(1,40) = 10.10, P = 0.003). (b) When rating the items for Tastiness, AN rate food as less tasty (F(1,40) = 10.07, P = ), an effect that is more pronounced when rating high-fat foods (Group X Food Type interaction: F(1,40) = 5.21, P = 0.028). (c) Logistic regression of Health and Taste ratings on choice. Among the HCs, choice was more influenced by taste than by health ( 2 = 75.64, P < ), whereas AN choices were influenced by both taste and health ( 2 = 0.22, P = 0.64). (d) Multilevel linear regression of Taste on Health ratings showed that Healthiness and Tastiness were relatively independent in HC, whereas there was a

2 significantly greater association between Health and Taste ratings in AN than in HC (z = 5.3, P = 0.02). (e) HC slowed responses when rating Healthiness of high-fat foods, whereas AN slowed responses when rating Tastiness of high-fat foods (when adjusting for overall group differences in RT; Rating phase X Food type X Group interaction F(1,40) = 11.81, P = 0.001). (f) When asked directly, most HC found Taste easier to rate, whereas most AN found Health easier to rate (proportions significantly different, 2 (1, N = 40) = 10.15, P = 0.002). Data are mean s.e.m. (abe); Data are fixed effects coefficients s.e. (cd).

3 Supplementary Figure 2 Need and implementation of self-control. Assessing self-control use across participants (n = 42). (a) A conflict between Health and Taste ratings creates choice trials with opportunity for self-control (examples outlined in pink): Choosing healthy, less tasty options or not choosing unhealthy, tasty options reflects self-controlled choices. Health-Taste rating alignment leaves no opportunity for self-control (examples outlined in orange). (b) On trials with opportunity for selfcontrol, individuals with AN exercised self-control on a greater proportion of trials relative to Controls (t(33.88) = 4.89, P = ). However, individuals with AN had significantly fewer trials with opportunity for selfcontrol (t(40) = 3.1, P = 0.004; M HC = ; M AN = ). This was likely due to greater alignment between Healthiness and Tastiness across items in AN (See Supplementary Fig. 1d). Thus, changes in valuation according to Health and Taste in AN resulted in diminished need for self-control, suggesting an alternate strategy for making food decisions. (c) Response times when opting for or against self-controlled choices. As expected, HC slowed down when making self-controlled choices, whereas individuals with AN sped up (Group X Self-control use interaction: 2 (1, N = 41) = 4.78, P = 0.03; 1 AN participant had no valid trials for this analysis). Response times on trials where no self-control was needed are shown for comparison (orange bars). Thus, for AN, in contrast to HC, engaging self-control did not result in a response time cost. Note that the low number of trials involving self-control precluded fmri data analysis of self-control trials. Such analyses depend both on trials where self-control was required and deployed as well as trials where selfcontrol was required but not deployed. Only 16 participants (8 HC, 8 AN) had 4 or more trials in each selfcontrol bin. Data are mean s.e.m.

4 Supplementary Figure 3 Characterizing dorsal striatum responses. (a) To assess the specificity of the differences observed between HC and AN related to food choices, values were extracted from the cluster identified in the choice analysis (Fig. 2b). No significant differences between HC and AN were found for Health or Taste ratings in the dorsal striatum region that showed a significant difference in the Choice phase (Health: t(40) = 0.81, P = 0.42, n = 42; Taste: t(39) = 0.41, P = 0.68, n = 41). Thus, the difference between groups in the dorsal striatum was specific to the Choice phase. Note that this ROI analysis is independent as the region was identified based on activity in the Choice phase. Data are mean s.e.m. (b) Parametric analysis of choice ratings for low fat and high fat trials separately within the caudate cluster identified in contrast between AN and HC reported in the manuscript (Fig. 2b, n = 42). There was a significant difference between AN and HC for low fat (t(40) = 2.29, P = 0.028) but not for high fat (t(40) = 1.396, P = 0.170) trials. There were no significant differences between high and low fat trials within the HC (t(20) = -1.03, P = 0.32) or AN group (t(20) = 0.87, P = 0.40). Data are mean s.e.m. (c) Choice phase response bin analysis within the caudate cluster identified in contrast between AN and HC reported in the manuscript (Fig. 2b, n = 42). Data were extracted for each response bin (1-5), to further assess whether the response in the dorsal striatum increased with increasing decision strength in the Choice phase or instead reflected a binary Yes / No response to the test food over the Reference food. Data are mean s.e.m.

5 Supplementary Figure 4 Comparison of HC versus AN choice-related activity in the vmpfc. (a) BOLD activity in regions of the vmpfc correlated with trial-by-trial choice values in both HC (left) and AN (middle) groups, with no significant difference between them (right) (FWE-corrected P < 0.05 whole-brain, cluster-forming threshold Z > 2.3, n = 42). Results from whole-brain analysis in both groups are displayed. Images and coordinates in Montreal Neurological Institute (MNI) space and radiological orientation (Right = Left). To further probe the lack of group differences within vmpfc, we performed an independent analysis using ROIs identified in previous studies of value based choice (ROIs displayed in blue): (b) Hare et al.; MNI coordinates = [3 51 3] (t(40) = 0.23, P = 0.82), and (c) Bartra et al. meta-analysis (t(40) = 0.52, P = 0.61). Data are mean s.e.m.

6 Supplementary Figure 5 Rating phase vmpfc analyses We assessed vmpfc responses in the Health and Taste rating phases using the same region identified by Hare et al (2009); MNI coordinates = [3 51 3]. (a) There were no significant differences in BOLD response between the HC and the AN group in the Health (t(40) = 1.44, P = 0.16, n = 42) or Taste phase (t(39) = 0.26, P = 0.54, n = 41). Data are mean s.e.m. (b) BOLD signal during the Health and Choice phases were significantly correlated in the AN group (r = 0.44, P = 0.046); BOLD signal during Taste and Choice phases were significantly correlated in the HC group (r = 0.49, P = 0.024).

7 Supplementary Figure 6 Fronto-striatal connectivity on low-fat and high-fat food trials in Anorexia Nervosa. (a) Fronto-striatal connectivity for low-fat foods was significantly negatively correlated with food intake in AN (r(14) = 0.51, P = 0.04, n = 16), (b) whereas connectivity for high-fat foods showed a trend towards a positive correlation with real food intake (r(14) = 0.44, P = 0.09, n = 16). Robust regressions including the participants who binge-ate yielded the same pattern of associations between eating behavior and connectivity on low-fat food trials (t(17) = 2.04, P = 0.057, n = 19) and high-fat food trials (t(17) = 1.92, P = 0.072, n = 19).

8 Supplementary Table 1. Participant clinical characteristics. Healthy Controls (n = 21) Anorexia Nervosa (n = 21) Mean ± SD Mean ± SD t df P Age (years) 22.7± ± BMI (kg/m 2 ) 21.5± ± < Subtype Restricting / Binge-purge n = 10 / 11 Duration of Illness (years) 11.4±6.7 Education (years) 15.7± ± EDE-Q, Global Score 0.11± ± < TFEQ-R 5.2± ± < STAI-S 26.4± ± < POMS depression 0.8± ± < Multi-Item Meal Intake (kcal) 831± ± < Fat intake (% kcal) 37.2± ± < BMI = Body Mass Index, WTAR = Weschler Test of Adult Reading, EDE-Q = Eating Disorder Examination Questionnaire, TFEQ-R = Three Factor Eating Questionnaire, Restraint Scale, STAI-S = Spielberger Anxiety Index, State version; POMS = Profile of Mood States Levene s test indicated unequal variances, so degrees of freedom were adjusted accordingly.

9 Supplementary Table 2. Food task stimuli. Low fat foods High fat foods 1 1% milk American cheese 2 air-popped popcorn avocado 3 apple slices baby cheese 4 baked potato bagel and cream cheese 5 banana baguette with olive oil 6 broccoli and cauliflower brownie 7 celery and carrot sticks burger 8 cherries cheese nachos 9 cherry tomatoes chicken nuggets 10 corn on the cob cookies 11 cucumber slices cupcakes 12 Froot loops (with milk) doughnuts 13 grapes eggs 14 green beans french fries 15 grilled chicken strips fried eggs 16 kiwis granola (with milk) 17 lollipops grilled cheese sandwich 18 mushrooms Hershey Kisses 19 orange slices hotdog with mustard 20 peaches ice cream sundae 21 pickles M&Ms 22 plain bagel mac and cheese 23 pretzels mashed potato 24 raisin bran with milk mini muffins 25 raisins mozzarella sticks 26 rice cakes peanut butter 27 rigatoni pizza 28 rigatoni and sauce Reese s pieces 29 Saltines Ritz crackers 30 seaweed salad with chicken 31 skim milk steak 32 vegetable soup string cheese 33 Sour patch candies tacos 34 soy chips trail mix 35 strawberries humus and pita 36 sushi rolls whole milk 37 turkey sandwich popcorn 38 yogurt yogurt covered pretzels

10 Supplementary Table 3 (see also Fig. 2). Results from the whole-brain parametric analysis of choice ratings (P < 0.05, corrected, n = 42). Cluster Region Cluster size x y z Peak Z HC 1 R Cerebellum R Cerebellum R Occipital fusiform gyrus R Temporal occipital fusiform gyrus R Occipital fusiform gyrus R Lingual gyrus L Postcentral gyrus L Precentral gyrus L Precentral gyrus L Precentral gyrus L Postcentral gyrus L Precentral gyrus L Frontal medial cortex L Frontal pole L Frontal pole R Anterior cingulate gyrus R Paracingulate gyrus L Anterior cingulate gyrus L Superior frontal gyrus L Superior frontal gyrus L Superior frontal gyrus L Superior frontal gyrus L Superior frontal gyrus L Superior frontal gyrus AN 1 L Precentral gyrus L Precentral gyrus L Precentral gyrus L Precentral gyrus L Postcentral gyrus L Postcentral gyrus L Frontal medial cortex L Frontal pole R Frontal medial cortex L Frontal orbital cortex R Paracingulate gyrus R Caudate R Cerebellum R Cerebellum

11 R Lingual / Occipital fusiform gyrus R Cerebellum R Lingual gyrus R Lingual gyrus L Precuneus L Precuneus L Posterior cingulate gyrus Posterior cingulate gyrus L Precuneus L Precuneus L Supplementary motor cortex L Superior frontal gyrus Superior frontal gyrus L Superior frontal gyrus L Middle frontal gyrus L Anterior cingulate gyrus R Cerebellum R Cerebellum R Cerebellum R Cerebellum R Cerebellum R Cerebellum HC>AN 1 R Occipital fusiform gyrus L Temporal cccipital fusiform gyrus R Lingual gyrus L Occipital pole R Lingual gyrus R Occipital pole AN>HC No significant clusters For each cluster, the peak and 5 local maxima within the cluster are listed along with x-y-z locations in MNI space.

12 Supplementary Table 4 (see also Fig. 3). Results from the PPI analysis of connectivity for high versus low-fat foods between caudate and PFC including age as a covariate (P < 0.05, TFCE corrected, n = 42). Cluster Region Cluster size x y z Peak P HC>AN 1 L Paracingulate gyrus L Paracingulate gyrus R Frontal pole / Middle frontal gyrus R Frontal pole / Middle frontal gyrus R Frontal pole / Middle frontal gyrus AN>HC No significant clusters Clusters containing more than 10 voxels are shown. Whole-brain analyses revealed no clusters outside the PFC (FWE-corrected P < 0.05, clusterforming threshold Z > 2.3).