Eating Behaviours and Food Cravings; Influence of Age, Sex, BMI and FTO Genotype

Previous studies indicate that eating behaviours and food cravings are associated with increased BMI and obesity. However, the interaction between these behaviours and other variables such as age, sex, BMI and genetics is complex. This study aimed to investigate the relationships between eating behaviours and food cravings, and to examine the influence of age, sex, body mass index (BMI) and fat mass and obesity-associated (FTO) genotype on these relationships. A total of 475 participants (252 female, 223 male, BMI: 25.82 ± 6.14 kg/m2, age: 30.65 ± 14.20 years) completed the revised 18-question version of the Three Factor Eating Questionnaire (TFEQ-R18) to assess cognitive restraint, uncontrolled eating and emotional eating, and the Food Cravings Inventory (FCI) to assess cravings for fatty food, sweet food, carbohydrates and fast food. DNA samples were genotyped for the rs9939609 polymorphism in the obesity-linked gene FTO. Questionnaire data was analysed for associations between the TFEQ-R18 and FCI subscales for the whole study group, and the group divided by sex, genotype and age (≤25 years versus >25 years). Finally, mediation analysis was used to explore the relationships between BMI, cognitive restraint and food cravings. FTO AA + AT genotype was associated with increased BMI, but not with differences in eating behavior scores or food craving scores; age was associated with increased BMI and decreases in food craving scores in which this effect was stronger in women compared to men. Increased cognitive restraint was associated with decreased food craving scores in the ≤25 years group. Mediation analysis demonstrated that in this group the association between BMI and reduced food cravings was mediated by cognitive restraint indicating that in this age group individuals use cognitive restraint to control their food cravings. The positive correlation between age and BMI confirms previous results but the findings of this study show that age, sex, FTO genotype and BMI have an influence on the relationships between eating behaviours and food cravings and that these variables interact.


Introduction
Obesity is a global problem; worldwide 39% of adults are classified as being overweight and 13% as having obesity [1]. It is now recognised that the development of obesity is influenced by a complex interaction between genetics and psychological factors, such as eating behaviours and food cravings.
Behaviours related to food intake which influence the frequency of eating, meal size, meal content, and attitude to meals are described as 'eating behaviours'. Eating behaviours can influence the amount

Aim
The aim of the current study is to investigate the interactions between eating behaviours, food cravings and BMI. Previous findings indicate that when considering the influence of eating behaviours and food cravings on obesity, it is important to take into account other variables that are known to also influence these measures including genetics, age and sex. It is also important consider which associations are mediated by other relevant variables. This study therefore (1) investigates the interactions between eating behaviours, food cravings and BMI (2) determines the influence of FTO genotype, sex and age on these interactions and (3) uses mediation analysis to explore the role of these mediators in the interactions.

Study Participants
Participants were recruited through advertisements at Sheffield Hallam University, Sheffield, UK. The participants included students at the university, staff at the university and friends and family of students at the university. Ethical clearance for the study was granted by the Ethics Committee of the Biomolecular Research Centre, Sheffield Hallam University and all volunteers provided written informed consent before taking part in the study. To be included in the study participants needed to be ≥18 years. People who were pregnant or breastfeeding, who had a history of serious psychiatric illness or were taking medication known to affect appetite or metabolism were excluded. A total of 475 participants (n = 252 F, n = 223 M), average BMI 25.82 ± 6.14 kg/m 2 , average age 30.65 ± 14.20, age range 18-81 were included in this study. 249 of the participants were aged 25 years or younger (≤25 years), with 226 participants aged over 25 years (>25 years). We were interested in studying eating behaviours at a stage in life before patterns had become fully established. We considered previous work which shows that life events such as co-habiting, marriage, securing a job and having children are associated with changes in eating patterns and weight gain [38][39][40]. We chose 25 years as the cut-off because for many in our study group up to this age would be prior to the aforementioned life events.

Anthropometry
Height was measured to the nearest 0.1 cm using a wall-mounted stadiometer (Seca, Hamburg, Germany). Participants stood barefoot with their heels together against a wooden back plate keeping their arms loosely by their side. The measurement was recorded when the head was in the Frankfort Plane i.e., a horizontal line between the lower orbits of the eyes and the external auditory meatus. Body mass was measured to the nearest 0.01 kg using a balance beam scale (Avery, Birmingham, UK). Participants wore light clothing, removed their shoes and jewelry and were told to remove anything from their pockets. BMI was calculated as the body mass in kilograms divided by the square of height in meters.

Eating Behaviours
Eating behaviours were measured using the revised 18 items Three Factor Eating Questionnaire (TFEQ-R18) [3]. This is a shortened version of the original 51-item TFEQ developed by Stunkard and Messick (1985) [41]. The questionnaire measures three different subscales of eating behaviour: (1) Cognitive restraint (CR) defined as the conscious restriction of food intake aimed to control body weight and/or to promote weight loss comprised of six items e.g., "I consciously hold back at meals in order not to gain weight". (2) Uncontrolled eating (UE) defined as the tendency to eat more than usual due to a loss of control over intake with a subjective feeling of hunger. Comprised of nine items e.g., "When I see a real delicacy, I often get so hungry that I have to eat right away". (3) Emotional eating (EE) defined as an inability to resist emotional cues, eating as a response to different negative emotions. Comprised of three items e.g., "When I feel blue, I often overeat". The 18 items are measured on a 4-point response scale (definitely true: 4, mostly true: 3, mostly false: 2, definitely false: (1) and items scores are summated into subscale scores: CR, UE and EE. Previous studies have reported that TFEQ-R18 has adequate internal consistency reliability coefficients for the three subscales, as well as for the whole questionnaire (0.75-0.87) [3].

Food Cravings
Food cravings were measured using the Food Craving Inventory (FCI), a self-report measure that assesses general and specific food cravings [42]. The FCI is a valid and reliable measure of specific food cravings in a subjective manner within the previous month. The questionnaire consists of 28 items measuring the frequency of cravings for a specific food item on a scale of 0-never to 4-almost every day. The FCI has four subscales: (1) high fat foods (fried chicken, sausages, gravy, fried fish, bacon, steak, sausage rolls), (2) sweets (brownies, cookies, chocolate, doughnuts, cakes ice-cream, sweets, sponge cake), (3) carbohydrates (breads, pancakes, biscuits, sandwich bread rice, baked potato, pasta, cereal, danish pastry, crisps) and (4) fast food (hamburger, French fries, pizza). A total score for each subscale is calculated by the summation of each food item score for each participant. The FCI has demonstrated acceptable internal consistency reliability and test-retest reliability in adults [42]. The inventory shows adequate internal consistency (Cronbach's a of 0.86 for the fats and sweets subscales, 0.84 for the starches/complex carbohydrates subscale, 0.76 for the fast-food subscale, and 0.93 for the inventory as a whole [42]. Further psychometric support for the FCI has been established in diverse community and clinical samples [43,44].

Genotyping
Buccal samples for genotyping were collected using Easicollect devices (GE Healthcare Life Sciences, Little Chalfont, UK). DNA was extracted from the FTA card contained in the device using a QIAamp DNA mini kit (QIAGEN, Manchester, UK) according to the manufacturer's instructions. Genotyping was performed using Taqman SNP genotyping assays (Applied Biosystems, Life Technologies, Ltd., Paisley, UK). FTO gene, SNP ID: rs9939609 was investigated in this study.

Statistical Analysis
Statistical Package for the Social Sciences (SPSS software version 24, IBM) was used for the statistical analysis with significance accepted if p < 0.05. Distribution of data was checked using the Kolmogorov-Smirnov Test and descriptive statistics were calculated. Pearson correlation was undertaken to explore association between variables including age, BMI, eating behaviours and food craving. T-tests were performed to assess gender and genotype differences in age, BMI, eating behaviours and food cravings. To explore relationships between variables mediation analysis with bootstrapping as recommended by Preacher and Hayes (2008) [45] was used; the SPSS macro provided by Hypotheses were tested in SPSS using PROCESS Model 59 [46], with using model 4,10,000 bootstrapped estimates and a 95 percent confidence interval. This model estimates a moderated indirect effect by evaluating the impact of the moderator on the indirect effect, as well as the predictive relationship to the direct effect. For the mediation model (Figures 1 and 2), BMI was the independent variables, eating behaviours was the mediator variable, and food craving was the dependent variable. Age, sex and FTO genotype were the covariates. For outcome data, effect sizes were calculated using Cohens d. Thresholds were set at 0.0-0.19 for a trivial effect, 0.2-0.49 for a small effect, 0.5-0.79 for a medium effect and ≥0.8 for a large effect [47,48].

Descriptive Statistics
Descriptive statistics for the primary variables for the overall sample, by sex, genotype and age are shown in Tables 1 and 2. There were no significant differences in mean BMI between men and women but there was a significant difference between carriers of the FTO TT genotype compared to those with AT + AA genotype (24.62 ± 5.29 versus 26.52 ± 6.55 kg/m 2 , p = 0.001). Table 1 also shows the results of mean food craving scores; women had higher mean scores for carbohydrate cravings compared to men (14.73 ± 5.99 versus 13.59 ± 5.74, p = 0.036). There were no differences in eating behaviour scores between men and women and no difference in scores for food cravings or eating behaviours between the different genotypes. BMI was higher in the >25 years group, but craving for fat and carbohydrate were higher in the ≤25 years group, as were scores for uncontrolled eating and emotional eating. All significance tests were two-tailed. M = mean score; SD = standard deviation; t = t-test; p = p-value; d = Cohen's d. * Denotes a significant difference between women and men. Table 3 shows that there is a positive relationship between age and BMI (r = 0.509 **) and an inverse relationship between age and fatty food cravings (r = −0.107 *), UE (r = −0.222 **) and EE (r = −0.184 *) for the whole sample. There is an inverse relationship between BMI and fatty food cravings (r = −0.180 **), and a positive relationship between BMI and CR (r = 0.128 **). Table 4 shows that there is an inverse relationship between CR and fatty food cravings (r = −0.259 **), sweet food cravings (r = −0.138 **) and fast food cravings (r = −0.167 **). There is a positive relationship between UE and carbohydrate cravings (r = 0.164 **) and there is an inverse relationship between EE and fatty food cravings (r = −0.197 **), and fast food cravings (r = −0.125 **).  Correlations of primary variables for overall population. All significance tests were two-tailed (* p < 0.05; ** p < 0.01). Tables 5-7 show the data dichotomised by sex (male and female), genotype (TT genotype and AA + AT genotype), and age (≤25 years and >25 years).
When exploring the differences between men and women, Table 5 shows that in women there is an inverse relationship between age and fat cravings (r = −0.157 *) and carbohydrate cravings (r = −0.195 **), in contrast to men where these relationships are not seen. In men there is an inverse relationship between BMI and CR (r = 0.210 **) in contrast to women where there is no relationship. In men there were also inverse relationships between CR, EE and fast food cravings which were not seen in women. Table 6 shows that FTO genotype significantly affects the relationships between age, BMI, eating behaviours and food cravings. The decline in emotional eating with age was greater in the AA + AT genotype group; in addition the inverse relationship between BMI and fat craving was stronger in this group compared to the TT group.
As there was such a strong effect of age on eating behaviours and food cravings, we investigated the participant group split by age. The results show differences between the ≤25 years and >25 years age groups. There is an inverse relationship between sweet cravings and EE in the younger group, whereas in the older group this relationship is positive. A similar difference is seen for the relationship between fat cravings and EE. In the younger group there are relationships between fast food cravings and CR and EE which are not present in the older group. Correlations of primary variables among women n = 252 presented above the diagonal, correlations of primary variables among men n = 223 are presented below the diagonal. All significance tests were two-tailed (* p < 0.05; ** p < 0.01). Correlations of primary variables among TT genotype n = 153 presented above the diagonal, correlations of primary variables among AA + AT genotype n = 322 are presented below the diagonal. All significance tests were two-tailed (* p < 0.05; ** p < 0.01).

Discussion
The aim of this study was to explore the relationships between eating behaviours and food cravings, and examine the influence of sex, BMI, age and FTO genotype on these relationships. We used the TFEQ-R18, FCI, and genotyped for the rs9939609 FTO polymorphism to study 475 individuals. We analysed data from this group split by sex, FTO genotype and age (≤25 years versus >25 years old). We then used mediation analysis to investigate possible mechanisms underlying the association between some of the relationships observed. The main findings are as follows.
Mean scores for the three subscales of the TFEQ-R18 and the four subscales of the FCI were similar to values previously reported. There were no differences between mean scores for BMI, eating behaviours or food cravings split by sex, with only cravings for carbohydrate being higher in women. This agrees with previous studies [28,49], but is in contrast with others which found higher scores for eating behavior subscales in women [30,50]. This difference may be due to the age of the groups Figure 2. CR mediates the relationship between BMI and cravings for sweet food in the ≤25 years group but not in the >25 years group (* p < 0.05; ** p < 0.01).

Discussion
The aim of this study was to explore the relationships between eating behaviours and food cravings, and examine the influence of sex, BMI, age and FTO genotype on these relationships. We used the TFEQ-R18, FCI, and genotyped for the rs9939609 FTO polymorphism to study 475 individuals. We analysed data from this group split by sex, FTO genotype and age (≤25 years versus >25 years old). We then used mediation analysis to investigate possible mechanisms underlying the association between some of the relationships observed. The main findings are as follows.
Mean scores for the three subscales of the TFEQ-R18 and the four subscales of the FCI were similar to values previously reported. There were no differences between mean scores for BMI, eating behaviours or food cravings split by sex, with only cravings for carbohydrate being higher in women. This agrees with previous studies [28,49], but is in contrast with others which found higher scores for eating behavior subscales in women [30,50]. This difference may be due to the age of the groups studied; the average age of our group was 31 years, whereas the average age of the participants in these two studies was 50 and 48 years respectively indicating that sex differences in eating behaviours may become more apparent with increasing age. When the group was analysed by genotype the AA + AT group had a mean BMI of 26.52 kg/m 2 compared to the TT genotype group who had a mean BMI of 24.62 kg/m 2 , this is consistent with the findings of previous studies which also found that carriers of the risk allele A had higher BMI scores compared to people with the TT genotype [13,51]. We did not find differences in TFEQ-R18 or FCI scores when the groups were split by genotype, which is in agreement with a recent study of children by Rivas et al. [52], which also did not find differences in subscale scores for the TFEQ-R18. This is in contrast to other findings which describe higher food cravings associated with the A allele in a group with an average age of 50 years [23]. These differences agree with our findings presented in Table 6 which demonstrate an influence of FTO genotype on age-related eating behaviours.
We report highly significant relationships between eating behaviours and food cravings, with BMI increasing with age, whilst cravings for fatty food decreased, and UE and EE decreased with age, in agreement with earlier studies [33][34][35]. The data presented in Table 2 also confirms this; when the two age groups were compared the scores for cravings for carbohydrates and fat, and the scores for emotional eating and uncontrolled eating were all higher in the ≤25 years group compared to the >25 years group. This may reflect changes in eating patterns associated with stages of life; with younger people tending to consume less healthy food [53,54], it may also be due in part to age-related changes in taste sensitivity [36,37]. We also found that increased BMI was associated with a decrease in cravings for fatty foods but an increase in cognitive restraint, confirming previous results by Johnson et al. who also proposed that high cognitive restraint in normal weight individuals increases the risk of overeating tendencies when restraint is relaxed, thus leading to further increases in BMI [55]. We found highly significant relationships between eating behaviours and food cravings; increased cognitive restraint correlated with lower cravings for fatty foods, sweet foods and fast foods, increased uncontrolled eating was associated with increased cravings for carbohydrates, and increased emotional eating was associated with lower cravings for fatty foods and fast foods. This is consistent with previous findings; decreased food cravings are associated with increased fMRI-FCR of brain regions that regulate executive control over ingestion [56] and with cognitive reappraisal strategies, in particular those focusing on the benefits of not eating unhealthy foods, this could potentially increase the ability of individuals with obesity to inhibit appetitive motivation and reduce unhealthy food intake [57].
When the group was split by sex analysis showed that the relationships between age and decreasing food cravings were stronger in women compared to men. However, the relationships between eating behaviours and food cravings were not affected by sex. FTO genotype influenced some of the relationships between BMI, age, eating behaviours and food cravings; in particular the inverse relationship between BMI and fatty food craving was stronger in the AA + AT genotype group compared to the TT group. We also saw an effect of the FTO genotype on the relationship between age and emotional eating, with age-related decline in this behavior only present in the AA + AT genotype group.
We split the group into a younger and older age group based on assumptions of life changes that could induce eating behaviours. We observed clear differences between the ≤25 years group compared to the >25 years group. The inverse relationships between CR and EE with food cravings for fatty, sweet and fast foods were much stronger in the ≤25 years group compared to the >25 years group. In the case of the relationship between sweet food cravings and emotional eating there was a negative correlation in the ≤25 years group, but in the >25 years group there was a significant positive correlation. There was also a strong relationship between age and BMI in the >25 years group but not in the ≤25 years group.
We used mediation analysis to investigate these relationships. This showed that in the ≤25 years group for both cravings for fatty food and cravings for sweet food CR mediated almost all of the relationships with BMI, demonstrating that it is very likely that the people in this group with higher BMI are using CR to suppress food cravings. This was in contrast to the >25 years group where this pattern was not seen. In the ≤25 years group sweet cravings were strongly influenced by sex, with a more pronounced inverse relationship between BMI and sweet cravings seen in women. When CR was first investigated as an eating behavior the strong correlation observed between CR and BMI led to the conclusion that CR was a type of unhealthy eating behavior [4], and that people who restrained were more susceptible to binge eating, thereby leading to increased BMI. Thus, CR as an eating behavior was seen to drive an increase in BMI. However, many early studies were carried out in participants with overweight or obesity and in older demographic groups; in these groups CR may indeed be driving an increase in BMI. In contrast the data presented here supports that in younger age groups BMI may drive CR rather than the other way round, a suggestion which has also been made in other papers [29,57]. We have observed the same association between CR and BMI seen in previous studies, but in the ≤25 years group this was particularly associated with a decrease in food cravings. We therefore propose that in people ≤25 years, who are perhaps more self-conscious about their weight, CR is used to suppress food cravings, particularly in people with a raised BMI, hence the association between BMI and CR. This may instill habitual eating behaviours, which in later life increases the susceptibility to the cycles of restraint and binge-eating associated in some older populations with increased BMI [58]. Evidence to support this theory is provided by Rocks et al. who studied undergraduate students and found that this group, with a similar age and background to our ≤25 years group, demonstrated high dissatisfaction with body weight and high rates of disordered eating behaviours [59]. An additional possibility is that BMI also mediates the effect of age on eating behaviours and food cravings.
Our study has limitations; although many of the participants were drawn from the same geographical area there may be demographic differences between the ≤25 years group who were mostly undergraduate students, and the >25 years group who were a mixture of staff, students and people from outside the university. In addition the lifestyle and eating habits of the ≤25 years group in our study may not be fully representative of the general population in this age category. As some of the participants were friends and family of the students at the university they may share similar tastes, food preferences and cultural background which may also have influenced the results.
We did not collect detailed data on socio-economic status, education, marriage status or ethnicity in the current study so in future studies these potential confounders will be considered. The division into two age groups with a cut-off of 25 years was based on previous work that shows that life events such co-habiting, employment and having children are associated with changes in eating behaviour [38][39][40]; in our study group the age of 25 reflects a point after which these life events start to have an effect, in future studies collection of data on these variables will allow for their inclusion in the analysis. We chose to examine the effect of FTO genotype as this has been reported to be the most influential gene on BMI, in the future investigating a wider range of target genes would potentially reveal more information about the influence of genetics on the variables studied. A limitation of the use of self-reported questionnaires is that participants may not always be honest when answering, and the degree to which this is an issue may vary with weight status. In future studies we plan to standardise the conditions under which the questionnaires were completed to minimise this effect, and to incorporate food frequency questionnaires and DEXA analysis into our study protocol.

Conclusions
In conclusion whilst the findings regarding the strong association between increasing BMI and age confirm previous reports the mediation analysis shows that sex, FTO genotype and BMI have an influence on the relationships between eating behaviours and food cravings and that these variables interact with notable differences between young and older age groups. Examining these interactions along with further polymorphisms linked to obesity could help identify those at risk of increasing BMI.