Obese patients after gastric bypass surgery have lower brain-hedonic responses to food than after gastric banding

Objectives Roux-en-Y gastric bypass (RYGB) has greater efficacy for weight loss in obese patients than gastric banding (BAND) surgery. We hypothesise that this may result from different effects on food hedonics via physiological changes secondary to distinct gut anatomy manipulations. Design We used functional MRI, eating behaviour and hormonal phenotyping to compare body mass index (BMI)-matched unoperated controls and patients after RYGB and BAND surgery for obesity. Results Obese patients after RYGB had lower brain-hedonic responses to food than patients after BAND surgery. RYGB patients had lower activation than BAND patients in brain reward systems, particularly to high-calorie foods, including the orbitofrontal cortex, amygdala, caudate nucleus, nucleus accumbens and hippocampus. This was associated with lower palatability and appeal of high-calorie foods and healthier eating behaviour, including less fat intake, in RYGB compared with BAND patients and/or BMI-matched unoperated controls. These differences were not explicable by differences in hunger or psychological traits between the surgical groups, but anorexigenic plasma gut hormones (GLP-1 and PYY), plasma bile acids and symptoms of dumping syndrome were increased in RYGB patients. Conclusions The identification of these differences in food hedonic responses as a result of altered gut anatomy/physiology provides a novel explanation for the more favourable long-term weight loss seen after RYGB than after BAND surgery, highlighting the importance of the gut–brain axis in the control of reward-based eating behaviour.


Supplemental Figures
. Subject characteristics of whole cohort Table S2. Characteristics of separate cohort of overweight/obese subjects used to create functional regions of interest in brain activation analysis Table S3. Spatial coordinates of whole brain activation for food > objects contrast in separate cohort of overweight/obese subjects. Table S4. Spatial coordinates of functional regions of interest in brain activation analysis Table S5. Psychological questionnaires from subjects in whole cohort Table S6. Spatial coordinates of whole brain comparison of activation to food between surgical groups Table S7. Region of interest activation during food evaluation and auditory-motor-visual control task  Figure S1. A priori functional regions of interest for reward system activation during food evaluation task Group activation in separate cohort of obese/overweight patients for any food (high-calorie or low-calorie) vs. object picture contrast. Activation is thresholded at voxel-wise FDR P<0.05, overlaid onto the average T1 scan for all subjects (n=24). A priori functional regions of interest (ROIs) are indicated: nucleus accumbens (NAcc, yellow), orbitofrontal cortex (OFC, light blue), caudate (Caud, dark blue), amygdala (Amy, green), anterior insula (Ins, magenta). Co-ordinates are given in standard MNI space. Figure S2. A priori functional regions of interest for auditory, motor and visual cortex activation during control task (A) Group activation maps of separate cohort of overweight/obese subjects overlaid with a priori anatomical regions of interest for control auditory-motor-visual task: auditory (red: listening to story) with bilateral posterior division of superior temporal gyrus (overlaid in yellow), motor task (green: button press) with left pre-central gyrus (overlaid in magenta), and visual (dark blue: flashing checkerboard) with lingual gyrus (overlaid in light blue). Activation is thresholded at voxelwise FDR P<0.05, overlaid onto the average T1 scan for all subjects (n=24). Co-ordinates are given in standard MNI space.
(B) Comparison of BOLD signal for auditory, motor and visual control task in a priori functional regions of interest between body mass index-matched unoperated controls (BMI-M, white), and obese patients after gastric banding (BAND, dotted) and gastric bypass (RYGB, striped) surgery, adjusting for age, gender and BMI. Data are presented as mean ± SEM. n=19-20 per group.

Figure S3. Plasma levels of bile acid sub-fractions, glucose and insulin
Comparison of plasma (A-F) bile acid sub-fractions (glycine, primary bile acid, deoxycholic bile acid), (G,H) glucose and (I,J) insulin levels. (A,C,E) levels during fMRI scan (area under curve (AUC) +70 to +150 mins), and (G,I) during fMRI scan (AUC +40 to +150 mins) between body mass indexmatched unoperated controls (BMI-M, white), and obese patients after gastric banding (BAND, dotted) and gastric bypass (RYGB, striped) surgery. (B,D,F,H,J) change in levels after ice-cream meal (ΔAUC +150 to +210 mins) in surgical groups. Data are presented as median and interquartile range. # P<0.05, ## P<0.01 vs. BMI-M; *P<0.05, **P<0.05, ***P<0.005 vs. BAND; n=20-21 per group. 8   Table S2. Characteristics of separate cohort of overweight/obese subjects used to create functional regions of interest in brain activation analysis.   Stereotactic coordinates (x, y, z given in standard MNI space) for peak voxel of group activation, adjusting for age, gender and BMI, thresholded at voxel-wise FDR P<0.05 (n=24).  Data included for the whole cohort. Data presented as mean ± SEM or median [interquartile range] for data that is not normally distributed, and (range), adjusted for age gender and BMI. a P value for overall comparison of averages or prevalence between groups.
Note that similar results were obtained when limiting the analysis to the scanned subjects only (data not shown). Abbreviations: BAND: gastric banding, BAS/BIS: Behavioural Activation and Inhibition Scale, BMI-M: body mass index matched, EPQ-R: Eysenck Personality Questionnaire, PANAS: Positive and Negative Affect Schedule, RYGB: gastric bypass. Stereotactic coordinates (x, y, z) for peak voxel of group activation for food category vs. objects, adjusted for age, gender and BMI, cluster thresholded at Z>2.1, FWE P<0.05 (n=20 per group), given in standard MNI space. Voxel-wise differences in BOLD activation between groups did not survive FDR P<0.05 correction. Average group activation in separate and combined a priori regions of interest (ROI) for food category vs. objects during food evaluation task, or auditory, motor or visual cortex during control task, adjusted for age, gender and BMI. Data presented as mean ± SEM and (range). a P value for overall comparison of averages between groups using ANOVA, with post-hoc comparison given beneath. b Contrasts with food pictures are compared to object pictures.
18 Table S8. Assessment of dumping syndrome in surgical groups. <0.001 RYGB > BAND Data presented as mean ± SEM or median [interquartile range] for data that is not normally distributed, and (range). a n=18-19 per group Δ heart rate and blood pressure: change between time points +150 and +210 min. Δ AUC for VAS: change in AUC between time points +150 to +210 min. Abbreviations: AUC: area under the curve BAND: gastric banding group, BMI-M: body mass index matched group, BP: blood pressure, mm: millimeters, RYGB: gastric bypass, VAS: visual analogue scale.

Participants
Obese patients who had previously undergone gastric bypass (RYGB) or gastric banding (BAND) surgery were recruited between June 2009 and June 2011 from the Imperial Weight Centre, Charing Cross Hospital, London, UK at follow-up clinics or through invitation letters. A BMImatched unoperated control group was recruited from the clinic or by public advertisement. The study was approved by the Local Research Ethics Committee, performed in accordance with the principles of the Declaration of Helsinki. All participants provided written informed consent.

Exclusion and inclusion criteria
Inclusion criteria for the study were: for surgical groups (i) loss of more than 8% of their total body weight since surgery, and (ii) surgery more than 2 months ago. All surgical procedures were performed by one of two surgeons (A.A. and T.O.), with RYGB as previously described (Olbers et al. 2003).
Exclusion criteria for the study were: (i) smoking, (ii) pregnancy or breast feeding, (iii) significant neurological, psychiatric or cardiovascular disease including addiction, stroke and epilepsy, other than previous depression, (iv) commencement of anti-depressants less than 6 months ago, (v) type 2 diabetes mellitus (T2DM) treated with agents other than metformin alone, (vi) type 1 diabetes mellitus.
Exclusion criteria for the scanning visit were: (i) inability to use right-handed button keypad, (ii) claustrophobia, (iii) shoulder width >58cm (inability to fit in scanner bore), (iv) metal implants which would preclude safe MRI scanning, (v) vegetarianism or veganism, (vi) reported gluten or lactose intolerance, and (vii) non-Western diet assessed by dietary record.

Patient characteristics
Eligible subjects attended an initial assessment visit during which they completed a medical history, physical examination and questionnaires to assess mood, psychological traits and eating behaviour. Medical notes were examined to ascertain pre-operative clinical information including body weight, presence of T2DM, and binge eating disorder (BED) from review by the clinic psychiatrist (S.S.) or psychologist, and calculation of obesity co-morbidity score using the Kings criteria (Aylwin & Al-Zaman 2008).
In line with standard policy of the obesity clinic, patients in this study had chosen themselves which surgical procedure to undergo. There was therefore no specific selection bias introduced by medical professionals as to which patients had which surgery, as there were no evidence based guidelines to inform bariatric procedure selection. However, in practice patients with T2DM tended to choose RYGB more often due to its more beneficial effects on glycemic control and T2DM resolution (Kashyap et al. 2010, Pournaras et al. 2012. There was therefore a significantly greater prevalence of T2DM and thus obesity co-morbidity score in the RYGB group, but no significant difference in current post-operative T2DM prevalence or other characteristics between surgical groups (see Table 1 and Table S1).

Psychological and eating behaviour questionnaires
The following questionnaires were completed at the initial assessment visit: 1. Dutch Eating Behaviour Questionnaire (DEBQ): to measure dietary restraint, emotional (e.g. stress-induced eating) and external (e.g. food palatability) influences on eating behavior (van Strien 1986).
3. Positive and Negative Affect Schedule (PANAS): to measure symptoms of positive and negative affect over the previous week, which have previously been correlated with fMRI responses to food pictures (Watson et al. 1988, Killgore & Yurgelun-Todd 2006.

Behavioural Activation / Behavioural Inhibition Scales (BAS/BIS): to measure punishment and reward sensitivity. BIS/BAS (reward responsiveness) scores have previously been correlated
with fMRI responses to food pictures (Carver & White 1994, Beaver et al. 2006.

Scanning visit protocol
On the day before scanning, subjects were instructed to avoid exercise and alcohol intake, to eat other symptoms were recorded at serial time points to measure hunger, pleasantness to eat, volume of food wanting to eat, fullness, sickness, sleepiness and stress (Flint et al. 2000, Blundell et al. 2010. The visit protocol is illustrated in Figure S1. Area under the curve (AUC) for VAS ratings were calculated from +40 to +150 mins to cover the period over the MRI scan in all three groups; and post-prandial changes in VAS ratings were calculated as delta AUC from baseline at +150 to +240 mins in the two surgical groups.

fMRI protocol
Patients were asked to refrain from strenuous exercise and alcohol the day before and day of the study. Patients were scanned for 1 hour starting between 11am and noon (Goldstone et al. 2009).
Female participants were scanned in first half phase of menstrual cycle (apart from one BMImatched control subject who was scanned on day 16 of her cycle) to avoid variations in reward responses including food over the menstrual cycle (Frank et al. 2010). Pregnancy was excluded at each visit.

fMRI confounding variables
There were no significant differences between the three groups in BMI, % body fat, time since last meal, sleep duration the night before the visit (Benedict et al. 2012, St-Onge et al. 2012, or positive or negative affect (PANAS) at the scanning visit (Table S10) (Killgore & Yurgelun-Todd 2006). During scanning there were no significant differences between the groups in absolute or relative head motion during the food evaluation or auditory-motor-visual fMRI tasks (Table S10).

Food picture evaluation fMRI paradigm
During the fMRI food picture paradigm, four types of colour photographs were presented in a block design split across two 9 minute, 192 volume runs: (1) 60 high-calorie foods (e.g. pizza, cakes and chocolate), (2) 60 low-calorie foods (e.g. salads, vegetables, fish), (3) 60 non-food related household objects (e.g. furniture, clothing) and (4) 180 Gaussian blurred images of the other pictures (as a low-level baseline), similar to those used previously (Goldstone et al. 2009). Food images were selected to represent familiar foods that are typical to the modern Western diet.
Pictures were obtained from freely available websites and the International Affective Picture System (IAPS, NIMH Center for the Study of Emotion and Attention, University of Florida, Gainesville, FL, USA). Food and object pictures were of similar luminosity and resolution.
Each run contained different pictures in 5 blocks each of high-calorie and low-calorie foods and objects interleaved with 31 blocks of blurred pictures (6 pictures per 18 secs) using one of four pseudorandom block orders with a randomized picture order within each block. Every image was displayed for 2500 ms, followed by a 500 ms inter-stimulus interval of a fixation cross. Each high-calorie food block consisted of equal numbers of foods containing chocolate, non-chocolate sweet and savory non-sweet foods (2 of each).
Images were viewed via a mirror mounted above an 8 channel RF head coil which displayed images from a projector using the IFIS image presentation system (In Vivo, Wurzburg, Germany) and ePrime 2 software (Psychology Software Tools Inc., Pittsburgh, PA, USA). Whilst each image was on display to subjects in the scanner, they were asked to immediately and simultaneously rate how 'appealing' each picture was to them using a 5 button hand-held keypad (1=not at all, 2=not really, 3=neutral, 4=a little, 5=a lot) (Goldstone et al. 2009). The appeal rating was thus made and recorded simultaneously with the stimulus presentation used for fMRI activation.
In our fMRI paradigm we studied the differences in BOLD activation to food pictures between surgical groups, rather than food receipt itself. fMRI paradigms with food pictures have been widely used to study human eating behavior (Carnell et al. 2012), and allow exposure to more complex, real-life food stimuli than can be achieved with the restricted nature of tastants such as milkshakes. Furthermore qualitatively similar correlations of fMRI responses to food pictures, anticipation of food receipt and actual food receipt have been reported (Stice et al. 2013).
Furthermore our study has demonstrated that greater activation of brain reward systems during evaluation of high-calorie food pictures is associated with greater palatability of high-calorie foods when actually consumed (see Results -Correlation between outcome measures).

Auditory-motor-visual control fMRI paradigm
A 6 min, 114-volume auditory-motor-visual (AMV) control task was performed. Over nine 33 second blocks, subjects performed two of each of the following tasks simultaneously: (i) listening to a story, (ii) tapping their right index finger once every second, or (iii) watching a 4Hz colour (yellow/blue) flashing checkerboard, with each task performed in 6 blocks, and instructions about whether to start or stop the motor task displayed for 3 seconds prior to each block.

fMRI analysis
The first 6 scans were discarded to allow for the BOLD signal to stabilize. The following preprocessing was applied: motion correction using MCFLIRT (Beckmann et al. 2003), fieldmapbased EPI unwarping using PRELUDE+FUGUE (Woolrich et al. 2004, Chang et al. 2012, non-brain removal using BET (Smith 2002), spatial smoothing using a Gaussian kernel of FWHM 6.0mm, grand-mean intensity normalization of the entire 4D dataset by a single multiplicative factor, and high pass temporal filtering (Gaussian-weighted least-squares straight line fitting, with sigma=100.0s).
Time-series statistical analysis was carried out using FILM with local autocorrelation correction including picture onsets, temporal derivative and motion parameters as co-variates. Two subjects (1 gastric bypass, 1 BMI-matched control) were excluded from fMRI analysis as their average relative motion over the food evaluation or control AMV fMRI tasks was greater than 0.5 mm/TR.
Registration to high resolution T1 structural and/or standard space images was carried out using FLIRT. Registration from high resolution structural to standard space was then further refined using FNIRT non-linear registration (Anderson et al. 2007b, Anderson et al. 2007a).
For the food pictures, higher level analysis was carried out using a fixed effect model to combine the two runs, by forcing the random effects variance to zero in FLAME (FMRIB's Local Analysis of Mixed Effects) to determine activation for the following contrasts: food > objects (high-calorie or low-calorie food), high-calorie food only > objects and low-calorie food only > objects (Beckmann et al. 2003, Woolrich et al. 2004. Similar time-series statistical analysis was performed for the single run AMV paradigm including the onsets of each task (auditory, motor and visual), with temporal derivative and motion parameters as co-variates, to contrast activation during performance of each task with that when the other tasks were being performed.
All higher-level analysis was carried out using FLAME (FMRIB's Local Analysis of Mixed Effects) stage 1 (Beckmann et al. 2003, Woolrich et al. 2004.

fMRI regions of interest
Functional regions of interest (fROIs) for the following areas: bilateral OFC, amygdala, nucleus accumbens, anterior insula and caudate nucleus ( Figure S2) were determined from a separate cohort of 24 overweight/obese subjects (Table S2) who underwent an identical protocol after fasting overnight. Higher level whole brain analysis was carried out with mixed effects analysis to identify those voxels which were significantly more activated at the group level, with correction for multiple comparisons made using false discovery rate (FDR) at P<0.05 for the food>objects contrast (high-calorie or low-calorie food minus objects) (Table S3). Similar functional localizers were made from this separate cohort for the control auditory, motor and visual tasks for bilateral superior posterior temporal gyrus (auditory), left pre-central gyrus (motor), bilateral lingual gyrus (visual) ( Figure S3, Table S3).
The functional anatomically constrained ROIs were obtained by masking these group activation maps with the a priori anatomical ROI. These were defined by the relevant bilateral ROIs from the cortical and subcortical structural Harvard FSL atlases thresholded at 10% probability. The OFC fROI included regions in the OFC and frontal pole with y > 22 and z < -6, since analysis of functional activation in this region demonstrated distinct bilateral clusters overlapping the anatomical Harvard atlas regions ( Figure S2). The insula mask was subdivided into the anterior insula (y > 4) (Chang et al. 2012).
The average (median) magnitude of bilateral BOLD activation within each a priori fROI was then extracted for each individual subject separately for any food, high-calorie food and low-calorie food (> object) contrasts using featquery in FSL, to measure the differences in activation between groups for the different picture categories, or different control auditory-motor-visual tasks.
Average BOLD activation for each of these contrasts within each ROI was then compared between groups outside FSL, adjusting for age, gender and BMI.

Food palatability
Ad libitum Hagen Daz™ vanilla or pralines and cream flavoured ice cream, was given to subjects in the operated groups in 50ml (43g) portions every 5 minutes and subjects were asked to eat until comfortably full (le Roux et al. 2007). Upon completion, they were asked to rate by VAS how 'pleasant' and 'sweet' the ice cream test meal was to eat. BMI-M control subjects did not have an ice cream test meal.

Dietary habits
Diet macronutrient composition was assessed using 3-day self-reported dietary records at home in the two surgical groups and analyzed using Dietplan6 (Foresfield Software Ltd., West Sussex, UK). GLP-1 7-36 amide and GLP-1 9-36 amide) and PYY (total PYY 1-36 and PYY 3-36 ) were assayed using established in-house radio-immunoassays (Allen et al. 1984, Kreymann et al. 1987. Plasma acyl ghrelin was measured by a two-site sandwich ELISA in a single run (Liu et al. 2008). Intra-assay coefficients of variation (CV) for gut hormones were <10%.

Metabolic, hormone and bile acid assays
Extraction of bile acids (BA) from plasma was performed as described previously. (Tagliacozzi et al. 2003) BA fractions were analysed using high-performance liquid chromatography (Jasco, Essex, UK) tandem mass spectrometry (Applied Biosystems, Cheshire, UK). The method was linear between 0.1 and 10 µmol/L for all BAs and their conjugates with CV of 1.5-6.8% at the lower limit of quantitation (0.1 µmol/L). The inter-assay CV was 3.6-8.0%.
Area under the curve (AUC) for metabolites and hormones were calculated from +40 to +150 mins, and for bile acids from +70 to +150 mins, to cover the period before and over the MRI scan in all three groups; and in the two surgical groups post-prandial changes in metabolites, hormones and bile acids were calculated as delta AUC from baseline at +150 to +210 mins per kCal ice cream eaten at lunch.

Dumping symptoms
The presence of symptoms of possible 'dumping syndrome' was assessed using change in nausea and sleepiness from before lunch to 1.5 hours after lunch (ΔAUC +150 to +240 mins), and change in physiological markers indicative of dumping syndrome, pulse and blood pressure, from before lunch to one hour after lunch (difference +150 to +210 min) (Ukleja 2005). In addition patients retrospectively completed two validated questionnaires to assess post-prandial symptoms of dumping (e.g. fainting, breathlessness, sleepiness, palpitations, headaches and nausea) in the 3 months following surgery (Sigstad 1970, Arts et al. 2009).

Role of funders
None of the funding sources have played a role in the collection, analysis, and interpretation or reporting of data or in the decision to submit data for publication.