Our study is based on participants from the Danish part of the MONICA study , the Diet, Cancer and Health (DCH) study  and the INTER99 study , each of which are described in the following.
This cohort consists of a random subset of 4,581 men and women born in 1922, 1932 in 1942 and 1952, who were selected from residents of 11 surrounding municipalities in the former Copenhagen County. In 1982–83, a total of 3,608 (78.8%) of these participants took part in a health examination, including measurement of BW, height, dietary intake and blood sampling. Five years later, during 1987–88, another invitation was sent to all living participants and 2,987 men and women participated in both the first and second health examinations . A total of 1,852 participants completed a seven-day food record in 1982–83 , 1,578 of these had complete information on covariates as well as repeated measures of BW and 1,426 of these had information on genetic variants.
For this study, we further excluded participants with prevalent cancer (n = 16), cardiovascular disease (n = 61) or self-reported diabetes (n = 20). The final study population consisted of 1,329 healthy participants with information on diet, genes, baseline and follow-up BW as well as information on potential confounders.
The diet cancer and health (DCH) study
During 1993–97, a total of 160,725 Danish men and women living in Copenhagen and Aarhus, between 50 and 64 years of age, born in Denmark and with no diagnosis of cancer registered in the Danish Cancer registry were invited to participate in DCH. Of these, a total of 57,053 (35%) accepted the invitation. These participants completed a lifestyle questionnaire, and a 192 item semi-quantitative food frequency questionnaire (FFQ) to assess the average intake of food during the last year. Furthermore, a follow up study was conducted during 1999 to 2002 which included self-administered questionnaires on diet, lifestyle and self-measured anthropometry.
Data used in this particular study is based on two samples of 1,200 BW gainers and 1,209 randomly selected control cohort/sample individuals. These participants had no cancer, cardiovascular disease or diabetes at baseline and follow-up, stable smoking habits, an annual weight gain not more than 5 kg/year, age <60 years at baseline and <65 years at follow-up. BW gainers were defined as the individuals who experienced the greatest degree of annual weight gain during follow-up. They were identified in gender stratified analysis by using the residuals from a regression model of ∆BW on baseline values of age, BW and height, smoking status (current/former/never smokers), and follow-up time. From these models, a total of 600 male and 600 female BW gainers were selected. The random sample was based on the complete cohort. The overlap between BW gainers and random sample was small (n = 79), thus the size of the remaining (non-overlapping) random sample (n = 1,130) almost equaled the number of weight gainers (12). Of the 2,330 individuals, we had information on diet, genes, changes in anthropometry and selected covariates on 2,167 men and women (2,128 in the analysis of ∆WC). However, 278 of these only had information on FTO (rs9939609).
A population-based randomized controlled trial (CT00289237, ClinicalTrials.gov) initiated in 1999. This year, an age- and sex- stratified random sample of 13,016 men and women born in 1939–40, 1944–45, 1949–50, 1954–55, 1959–60, 1964–65, 1969–70, living in 11 municipalities in the former Copenhagen County was drawn from the Civil Registration System and invited to a health examination. Of the 13,016 participants, a total of 12,934 were eligible for invitation, and, of these, 6,784 (52.5%) participated. Dietary intake was assessed through FFQ . Design and methods used in the study have been described in detail elsewhere [23, 27]. In 2004, all participants from the baseline examination were re-invited for a follow-up study where the baseline examination programme was repeated . Information on diet, genes, baseline and follow-up anthropometric measures as well as information on potential confounders was available on 4,574 subjects. For the present study, we excluded participants with prevalent cancer (n = 87), cardiovascular disease (n = 320) or self-reported diabetes (n = 94) and ended up with 4,073 participants (3,536 participants in the analysis of ∆WC).
All procedures in the three studies were in accordance with the Helsinki Declaration and all participants provided written informed consent.
Assessment of dietary intake
As described in detail elsewhere, information about the participants' dietary intake was collected using a validated seven-day food record in MONICA  and the same validated FFQ in DCH [29, 30] and INTER99 . From this information, daily consumption of foods and nutrients were calculated for each participant using the software program DANKOST in MONICA and FoodCalc in DCH and INTER99. Both DANKOST and FoodCalc are based on the official Danish food composition tables (http://www.foodcomp.dk). Participants' daily intake of ascorbic acid was then calculated and included in the analysis as a continuous variable (unit; mg/day).
Assessment of covariates
All participants reported information on smoking status (never smoked, ex-smoker or current smoker). Likewise, information was gathered about consumption of alcohol and included in the analysis as a continuous variable (MJ/day). Regarding physical activity, the MONICA participants were asked to classify themselves into one of four groups 1) Almost completely inactive: sedentary activities such as reading, watching television and going to the movies. 2) Some physical activity: at least 4 hours weekly including for example walking, cycling, construction work, bowling and table tennis. 3) Regular hard activity at least 3 hours weekly, including for example swimming, tennis and badminton etc. or heavy gardening. 4) Hard activity: elite sports such as swimming, soccer, badminton or long distance running several times a week. In DCH the questionnaire was used to obtain information on duration and types of physical activity. From this information the validated Cambridge Physical Activity Index was calculated by combining occupational physical activity with time spent on cycling and sport in summer and winter . Participants were then divided into four physical activity categories (inactive, moderately inactive, moderately active, and active). In INTER99 information on physical activity was based on two questions on commuting physical activity and leisure time physical activity. From these two questions, overall physical activity was calculated by summing response on commuting physical activity (converted into minutes per week using five day working week) and a leisure time physical activity variable (converted into minutes per week) . From this variable, overall physical activity was grouped into four categories <2 h/week, 2–3.9 h/week, 4–6.9 h/week and ≥7 h/week. Education was assessed with questions about years of regular schooling in all three cohorts and classified with respect to having a school education above or below the primary level. Finally, we included information on the participants’ age and gender, and the women reported whether they had entered menopause.
Assessment of anthropometric measures
At baseline, height was measured to the nearest 0.5 cm and BW to the nearest 0.1 kg in all three cohorts. Likewise, WC was measured horizontally midway between the lower rib margin and the iliac crest to the nearest 1 cm in DCH and INTER99. We did not have measures of WC on enough participants to include this measure in the analysis of the MONICA participants.
At follow-up, the baseline procedure was repeated for MONICA and INTER99 participants. Regarding the follow-up measures in DCH, the participants received a self-administrated questionnaire and reported their weight (kg) and WC (cm) measured at the level of the umbilicus using an enclosed paper measuring tape. A validation study was performed on 408 participants to compare measures of waist circumference obtained by technicians and by self-report, which showed that the self-reported WC at the level of the umbilicus was highly correlated with the technician-measured WC at the natural WC. The Spearman’s correlation coefficient was 0.87 in men and 0.88 in women .
From this information, we calculated change in BW as the difference between measures during 1982–83 and 1987–88 for MONICA, and change in BW and WC during 1993–97 and 1999–2002 for DCH and during 1999–2001 and 2004–06 for INTER99. From this we calculated ∆BW and ∆WC in each cohort by dividing the derived differences with the individual follow-up time in years.
SNP selection and genotyping
Through review of GWAS, we found 63 SNPs associated to different obesity related phenotypes [8–20, 35, 36], and 58 of these SNPs were consistently associated with BMI, WC or WHR [8–20]. In the present study, we included SNPs that were available in all three cohorts. Hence, we ended up with a total of 50 SNPs (Additional file 1: Table S1).
In MONICA and DCH, the SNPs were genotyped with the KASPar SNP Genotyping method. In MONICA, they had an average genotyping success rate of 98.3% (minimum 95.8%). In DCH, the average genotyping success rate was 97.8% and 185 replicate samples had a success rate above 98% and an error rate below 0.5%.
Finally, in INTER99 the SNPs were successfully genotyped using either the KASPar SNP Genotyping method, or through Human Cardio-metabo bead chip array (2 SNPs; rs7138803 and rs7647305) using Illumina Hi-Scan technology and GenomeStudio software (http://www.illumina.com/systems/hiscan.ilmn). The average genotyping success rate for the INTER99 study was 96.7% (minimum 94.7).
Genetic predisposition scores
For each subject, the 50 SNPs were coded 0, 1 or 2 according to number of obesity associated risk alleles. Four different SNP-scores were then calculated as indicators of genetic predisposition: A score of all 50 SNPs (range: 0 to 100), 33 BMI associated SNPs (range: 0 to 66), 6 WC associated SNPs (range: 0 to 12) and 14 WHR associated SNPs (range: 0 to 28), with higher scores indicating higher genetic predisposition to these specific traits.
Linear regression was used to examine the association between dietary ascorbic acid and subsequent ∆BW and ∆WC. First in a crude model with adjustments only for height and baseline measure of outcome, and then in a fully adjusted model where we also included age, sex, smoking status, education level, physical activity, menopausal status and alcohol consumption. Furthermore, to assess associations that were independent of ΔBW, the analysis with ΔWC as outcome was performed both with and without adjustment for concurrent ΔBW. The same procedure was applied using each of the four SNP-scores as the exposure variable. All continuous variables were evaluated by model control (investigating linearity of effects on used outcomes, consistency with a normal distribution and variance homogeneity).
To examine interaction between the four genetic predisposition scores and dietary ascorbic acid in relation to ΔBW or ΔWC, we correspondingly added the SNP-score variables as well as the interaction terms (SNP-score × ascorbic acid). After performing the analysis in the individual cohorts, the results were combined in a meta-analysis. We performed both fixed- and random effect models. The effect-estimates from the individual cohorts were weighted based on the inverses of their variances. Heterogeneity between the studies was assessed by so called Q tests and I2-values, where the latter measure indicates the amount of total variation explained by between-study variation  and was evaluated according to the following categories: no heterogeneity, I2: 0–25%; moderate heterogeneity, I2: 25–50%; significant heterogeneity, I2: 50–75%; and extreme heterogeneity, I2: 75–100%. Since the Q-tests showed no significant heterogeneity and the calculated I2-values were all below 50% we only presented results from the fixed effect models. Furthermore, results from the random effect models were almost identical to those from the fixed effect models.
Finally, with exploratory purposes, we performed the analysis of interactions between the 50 individual SNPs and dietary ascorbic acid in relation to ∆BW and ∆WC, with Bonferroni adjustment for multiple testing.
To further limit the possibility of confounding by energy intake, we performed supplementary analyses where we adjusted for total energy intake. Finally, in DCH we had information about intake of ascorbic acid from supplements, and in MONICA we had information on intake of multivitamins. Hence, in supplementary analyses we also adjusted for these variables.
As described, the DCH participants included in the present study consist of both a sample of BW gainers and a random sample. To maximize the study's statistical power, we decided to include both groups. However, we also performed separate analyses for the two groups.
Furthermore, INTER99 is a multifactorial lifestyle intervention, where the intervention group received a lifestyle counseling talk focusing on smoking, physical activity, diet and alcohol. Hence, in supplementary analyses, we further adjusted the INTER99 analyses for baseline intervention status.
P-values ≤ 0.05 were regarded as statistically significant. All analyses were performed using the statistical software package Stata 12 (StataCorp LP, College Station, Texas, USA; http://www.stata.com).