Copenhagen University Hospital, Centre for Health and Society, Copenhagen, Denmark

Transcription

1 (2011) 35, & 2011 Macmillan Publishers Limited All rights reserved /11 ORIGINAL ARTICLE wwwnaturecom/ijo Intake of total, animal and plant protein and subsequent changes in weight or waist circumference in European men and women: the Diogenes project J Halkjær 1, A Olsen 1, K Overvad 2,3, MU Jakobsen 2,4, H Boeing 5, B Buijsse 5, D Palli 6, G Tognon 6, HDu 7,8, DL van der A 7, NG Forouhi 9, NJ Wareham 9, EJM Feskens 10, TIA Sørensen 11 and A Tjønneland 1 1 Danish Cancer Society, Institute of Cancer Epidemiology, Copenhagen, Denmark; 2 Department of Epidemiology, School of Public Health, Aarhus University, Aarhus, Denmark; 3 Department of Cardiology, Aarhus University Hospital, Aarhus, Denmark; 4 Department of Clinical Epidemiology, Aarhus University Hospital, Aalborg, Denmark; 5 Department of Epidemiology, German Institute of Human Nutrition, Potsdam, Germany; 6 Molecular and Nutritional Epidemiology Unit, Cancer Research and Prevention InstituteFISPO, Florence, Italy; 7 National Institute for Public Health and the Environment (RIVM), Bilthoven, The Netherlands; 8 Department of Human Biology, Nutrition and Toxicology Research Institute of Maastricht (NUTRIM), Maastricht, The Netherlands; 9 MRC Epidemiology Unit, Institute of Metabolic Science, Cambridge, UK; 10 Division of Human Nutrition, Wageningen University, Wageningen, The Netherlands and 11 Institute of Preventive Medicine, Copenhagen University Hospital, Centre for Health and Society, Copenhagen, Denmark Background: As protein is considered to increase thermogenesis and satiety more than other macronutrients, it may have beneficial effects on prevention of weight gain and weight maintenance Objective: The objective of this study is to assess the association between the amount and type of dietary protein, and subsequent changes in weight and waist circumference (WC) Methods: men and women from five countries participating in European Prospective Investigation into Cancer and Nutrition (EPIC) were followed for a mean of 65 years Associations between the intake of protein or subgroups of protein (from animal and plant sources) and changes in weight (g per year) or WC (cm per year) were investigated using gender and centre-specific multiple regression analyses Adjustments were made for other baseline dietary factors, baseline anthropometrics, demographic and lifestyle factors and follow-up time We used random effect meta-analyses to obtain pooled estimates across centres Results: Higher intake of total protein, and protein from animal sources was associated with subsequent weight gain for both genders, strongest among women, and the association was mainly attributable to protein from red and processed meat and poultry rather than from fish and dairy sources There was no overall association between intake of plant protein and subsequent changes in weight No clear overall associations between intakes of total protein or any of the subgroups and changes in WC were present The associations showed some heterogeneity between centres, but pooling of estimates was still considered justified Conclusion: A high intake of protein was not found associated with lower weight or waist gain in this observational study In contrast, protein from food items of animal origin, especially meat and poultry, seemed to be positively associated with long-term weight gain There were no clear associations for waist changes (2011) 35, ; doi:101038/ijo ; published online 7 December 2010 Keywords: protein intake; animal protein; plant protein; weight change; waist circumference change; cohort Introduction Correspondence: Dr J Halkjær, Danish Cancer Society, Institute of Cancer Epidemiology, Strandboulevarden 49, 2100 Copenhagen Ø, Denmark jytteh@cancerdk Received 26 January 2010; revised 20 October 2010; accepted 22 October 2010; published online 7 December 2010 The increasing prevalence of overweight and obesity is one of the main contributors to the growing burden of lifestylerelated chronic diseases, such as type 2 diabetes, heart disease and some cancers 1 It is, therefore, of great interest to

2 identify modifiable predictors of body weight gain, among them dietary factors As protein is considered to increase thermogenesis and satiety more than other macronutrients; attention has lately turned to its potential beneficial effects on prevention of weight gain and weight maintenance 2,3 A comprehensive review from 2009, studying the epidemiological evidence on the associations between diet (both at macronutrient level and food level) and subsequent weight gain and obesity, however, concluded that based on the selected literature Levels of protein intake, regardless of source, are not associated with subsequent excess weight gain or obesity although the results were inconsistent 4 Among the eight cohort studies found on adults, six studies did not find any significant associations between total protein intake and subsequent changes in weight or waist circumference (WC) One study found positive associations between higher protein intake and both gain in weight and waist-to-hip ratio in white, but not in black, men and women and one study found lower gain at the waist at higher intakes of protein Very few studies were identified on subtypes of protein (various animal and plant origins) so no clear conclusions could be drawn with regard to protein sources There is an obvious need for more large-scale prospective studies to investigate these associations We have, therefore, studied the association between dietary intake of total protein and protein from various animal and plant sources and subsequent changes in weight and WC in several cohorts from the European Prospective Investigation into Cancer and Nutrition (EPIC) participating in the diet, genes and obesity (Diogenes) project Materials and methods Study population We used data from six cohorts within five countries participating in the EPIC study: Denmark (Copenhagen and Aarhus with identical methods and hence, considered one cohort, ), Germany (Potsdam, Ger-Pot), Italy (Florence, ), The Netherlands (Doetinchem, and Amsterdam/Maastricht,, as two separate cohorts, because of the differences in data collection methods) and the UK (Norfolk, ) These cohorts are all population-based and included both men and women 5 From the initial participants across six cohorts, our final study population consisted of eligible participants ( women, 58%), after exclusions An aprioridecision was made to exclude the following participants: those with no follow-up data available (n ¼ ), women who were pregnant at either baseline or follow-up (n ¼ 133), those with missing data on diet at baseline (n ¼ 113), the top and bottom 1% of the ratio of total energy intake/predicted energy expenditure (n ¼ 1803), missing anthropometry (baseline or follow-up) or missing follow-up time (n ¼ 2022) or those with extreme anthropometric data (heighto130 cm, body mass index (BMI)o16 kg m 2, waisto40 cm or 4160 cm, and weight change 45 kg per year or WC change 47 cm per year, and unrealistic combinations of BMI and WC, n ¼ 331) or with baseline chronic disease (prevalent diabetes, cancer or cardiovascular disease, n ¼ 8512) The mean duration of follow-up ranged from 37 to 100 years in the six cohorts Dietary assessment Detailed dietary information at baseline was collected by country-specific food frequency questionnaires (FFQs) that asked about habitual intake of medium-sized serving of foods over the past year 5 Food intake in g per day was derived by multiplying the frequency of intake with portion sizes Energy intake was calculated by using national food composition tables Estimated daily protein intakes were calculated by multiplying the protein content of each food of the specific portion size by the frequency of consumption as stated on the FFQ using country-specific national food tables All reported foods were classified as follows: animal (defined as X95% animal origin); plant (defined as X95% plant origin); mixed origin; non-organic; or unknown quantities of animal/plant (as, for example, for ready-to-eat dishes and cakes without any clear declaration or containing ingredients of mixed or unknown origin) On the basis of this information it was possible to estimate the intake of protein of animal and plant origin For the relative few food items, in which the origin could not be identified, protein origin was classified as unknown In addition, as supplementary exploratory analyses, protein from animal sources was divided into protein from red or processed meat, poultry, fish and dairy products In a random sample of 6790 participants among the participants (constituting B7 8% of each EPIC cohort), dietary intake was also assessed by a standardized 24-h recall using EPIC-SOFT 6 Data from this method were used to account for differences between national FFQs and to reduce potential measurement error introduced by the FFQs 7,8 The Spearman s correlation coefficients between the FFQ and the 24-h recall for total, animal and plant protein ranged between 024 and 040, lowest for animal protein (024 men and 028 women) and highest for plant protein (038 men and 040 women) Estimates for men and women were very similar Calibrated dietary intake variables were derived from a linear calibration model that regressed dietary intake variables from the 24-h recall on the FFQ dietary values 9 Both observed and calibrated dietary protein intake data consisted of protein intake in kcal per day for total protein, and subgroups of protein divided into animal, plant or unknown origin We also used data for other macronutrients: carbohydrate, fat (in kcal per day) and alcohol (included in the following categories (0 to o01 g per day, X01 to o5g per day, X5 too15 g per day, X15 to o30 g per day, X30 to o60 g per day, X60 g per day) In the main 1105

3 1106 analyses we used the observed dietary data, whereas calibrated data were used in sensitivity analyses Assessment of anthropometry and changes in weight and WC At baseline, weight, height and WC were measured according to a pre-specified protocol by trained staff in all six cohorts, with participants wearing no shoes and either light indoor clothing (, and ) or underwear (, Ger-Pot and ) At follow-up, anthropometric measurements were performed according to an identical protocol as during baseline examination by staff in and or were self-reported by the participants in a follow-up questionnaire in the other cohorts The BMI was calculated by dividing weight (kg) by the square of height (m 2 ) We used observed anthropometric measures without applying any correction factors for clothing or self-reporting in the main analyses to give an accurate reflection of the available weight data across Europe The outcomes in the current analyses were annual change in body weight (g per year) or WC (cm per year) This was calculated for each participant by subtracting the baseline value from the follow-up value and dividing the difference by the duration of follow-up in years Statistical analysis We examined the distribution of dietary protein intake (exposure) and of annual weight and waist change (outcome) across the six cohorts in men and women, separately Next, we explored the association between total or subgroups of protein intake and annual weight and waist changes using the energy partition method, 10 which is a model not adjusted for total energy Estimates in a partition model are interpreted as the effect of increasing the total energy intake of protein for a fixed level of the remaining macronutrient included in the model In the model energy intake from protein, carbohydrate and fat was included as continuous variables in the models and the estimate for protein are presented per 150 kcal per day increments (equalling per 375 g of protein), whereas alcohol intake was included in categories (0 to o01 g per day, X01 to o5 g per day, X5 too15 g per day, X15 to o30 g per day, X30 to o60 g per day and X60 g per day) Further adjustments were made for baseline weight, height, WC (in the analyses of waist change only), age, follow-up time (all included as continuous variables), changes in smoking status between baseline and follow-up (stable smoker, stable non-smoker, start smoking and stop smoking), baseline physical activity index (categories of inactive, moderately inactive, moderately active and active), education (highest level achieved from categories of none/primary school, secondary school technical/professional qualification or university degree) and in women, also menopausal status and use of hormone replacement therapy at baseline (yes/no categories) If information was missing for the potential confounder variables an unknown category was created for each variable and included in the models Analyses on subgroups of protein were conducted with the three subgroups included together (estimates are presented per 150 kcal per day) instead of total protein and then adjusted for the same macronutrient and covariates as in the model for total protein Test of linearity of continuous variables included in the model was evaluated by linear splines with three knots placed at the quartiles of the distribution 11 Gender-specific linear regression models were constructed, to test initially for statistical interactions between the dietary protein exposures and cohort to define our analytical strategy For both weight and waist changes interactions with centre were present for total and animal protein (strongest among women), therefore, it was decided to run gender and centre-specific analyses Estimates (and 95% confidence intervals, 95% CIs) of the effect of the exposure were calculated within each cohort and gender group and displayed in forest plots A pooled estimate (and 95% CI) across cohorts was calculated using a random effects metaanalysis approach I 2 and a P-value for heterogeneity between cohorts were also reported 12 Interactions between protein intake and (1) age, (2) duration of follow-up, (3) baseline BMI and (4) changes in smoking behaviour was performed using data stratified on gender and cohort and with the relevant interaction term included The P-value for interaction terms was calculated using an F-test, and if Po001 in at least three out of the six cohorts the interaction was considered statistically significant The a priori strategy was to evaluate gender-specific associations However, the associations with neither animal, or plant, nor unknown protein differed significantly between men and women (according to the above mentioned criteria for interaction) Therefore, joint estimates for men and women are presented in addition to the gender stratified associations In the supplementary analyses, it was also tested whether main sources of animal protein (red and processed meat, poultry, fish and dairy) were differentially associated with weight and waist changes A series of sensitivity analyses, in which main analyses were repeated using calibrated dietary exposure data, were performed Furthermore the associations for animal and plant protein were evaluated after assuming the amount of unknown protein as either 100% plant or 100% animal protein Furthermore, we applied a correction for differences in body weight and WC across cohorts by subtracting 1 kg or 2 cm when measured in light clothing, and by applying a regression equation to predict weight or WC from self-report 13 To investigate the potential role of comorbidity during follow-up, participants with incident chronic diseases (diabetes, cancer and cardiovascular disease) were excluded from the analyses Finally, an energy substitution model was used to evaluate the degree at which the

4 Table 1 Baseline dietary intake (median and 5 95th percentiles) of macronutrients, and characteristics at baseline and follow-up of anthropometrics and covariates (median and 5 95th percentiles or percentages) for all men and women in the six centres combined a Women Median, (5 95th percentile)/ percentage Men Median, (5 95th percentile)/ percentage N Baseline age (years) 540, 360 to , 397 to 650 Follow-up time (years) 55, 35 to , 35 to 103 Weight (kg) Baseline 658, 521 to , 648 to 1021 Follow-up 670, 520 to , 650 to 1030 Yearly weight change (g per year) 1098, to , to BMI (kg m 2 ) Baseline 246, 200 to , 214 to 321 Follow-up 249, 200 to , 216 to 325 Waist (cm) Baseline 790, 660 to , 800 to 1110 Follow-up 850, 700 to , 837 to 1150 Yearly waist change 09, 11 to 33 06, 13 to 22 (cm per year) % Stable non-smokers b % University or higher education Physical activity, % active % HRT users 21 F % Postmenopausal 54 F Total energy (kcal per day) 1907, 1230 to , 1547 to 3538 Total protein (kcal per day) 317, 195 to , 243 to 587 Animal (kcal per day) 191, 91 to , 119 to 411 Plant (kcal per day) 93, 55 to , 67 to 182 Unknown (kcal per day) 25, 5 to 74 25, 4 to 80 Fat (kcal per day) 653, 375 to , 475 to 1338 Carbohydrate (kcal per day) 849, 510 to , 603 to 1586 Alcohol (g per day) 6, 0 to 38 17, 1 to 73 Abbreviations: BMI, body mass index; HRT, hormone replacement therapy a Characteristics for each of the six centres individually are available in online Supplementary Tables b Includes both stable never and stable ex-smoker at both baseline and follow-up choice of energy adjustment methodology used affected the results Analyses were performed using SAS software (version 82) SAS Institute Inc (Cary, NC, USA) and Stata (version 92) (STATAcorp, College Station, TX, USA) Results Table 1 shows the baseline dietary intake and characteristics of anthropometrics and potential confounders in the total population Details from each of the six centres are presented in Supplementary Online Material Tables 1 and 2 Associations between protein intake and changes in weight and WC An overview of the pooled estimates and 95% CI of the association between intake of total and subgroups of protein and changes in weight and WC based on the random effects meta-analysis across centres are presented in Table 2 Weight changes An overall association between higher daily intake of total or animal protein and subsequent gain in weight was present, for both men and women The strongest effect was seen among women, in whom a 150 kcal higher daily intake wasassociatedwithanyearlyweightincreaseof78g(95%ci 35 to 120) and 82 g (95% CI 41 to 124) for total and animal protein, respectively For men the yearly weight increase was 29 g (95% CI ¼ 1 to 59) and 30g (95% CI¼ 8 to 68), respectively (Table 2) Although most cohort-specific estimates were positive, some heterogeneity across centres was present, but the pooling of the estimates was still considered justified (Figure 1 and 2) For plant protein and protein from unknown origin no overall significant association with weight changes was present, and no clear indication of heterogeneity across centres was present either Changes in WC No overall associations between daily intake of total protein or subgroups of protein and 1107 Table 2 Associations (pooled estimates and 95% CI) between intake of total and subgroups of protein (per 150 kcal) and subsequent yearly changes in weight and WC from a random effects meta-analysis across centres a Weight changes (g per year) Waist changes (cm per year) Men Women All Men Women All Per 150 kcal Estimate (95% CI) Estimate (95% CI) Estimate (95% CI) Estimate (95% CI) Estimate (95% CI) Estimate (95% CI) Total protein 289 ( 08 to 586) 775 (353 to 1197) 529 (259 to 799) 002 ( 004 to 007) 003 ( 005 to 011) 002 ( 002 to 007) Animal 300 ( 84 to 684) 824 (408 to 1240) 562 (269 to 856) 001 ( 005 to 008) 004 ( 004 to 013) 003 ( 002 to 008) Plant 84 ( 929 to 762) 394 ( 170 to 958) 177 ( 326 to 680) 004 ( 005 to 014) 005 ( 017 to 007) 00 ( 007 to 008) Unknown 315 ( 390 to 1019) 479 ( 171 to 1128) 396 ( 76 to 868) 015 (002 to 028) 004 ( 011 to 018) 009 ( 00 to 019) Abbreviations: 95% CI, 95% confidence interval; WC, waist circumference a In a model, including baseline intake of fat, carbohydrate (in kcal) and alcohol intake (in categories), baseline age, weight, (WC in the WC change analyses), height, physical activity, educational level and, for women, menopausal status and hormone use, follow-up time and changes in smoking status between baseline and follow-up Two models were constructed: one including total protein and the remaining macronutrients and covariates and the other including the three protein subgroups (animal, plant and unknown) and the remaining macronutrients and covariates

5 1108 Study ID ES (95% CI) % Weight Men Subtotal (I-squared = 498%, p = 0076) Women Subtotal (I-squared = 770%, p = 0001) Overall (I-squared = 722%, p = 0000) NOTE: Weights are from random effects analysis 7587 (1332, 13842) 8738 (2535, 14940) 1937 (-3658, 7531) (-15119, 5563) -081 (-4106, 3943) 2296 (-455, 5048) 2887 (-084, 5858) (7565, 15055) (4797, 16739) (4896, 17528) (-032, 26432) 2402 (-1813, 6616) 2606 (-337, 5549) 7746 (3527, 11965) 5290 (2590, 7991) Figure 1 Association between intake of total protein (per 150 kcal) and subsequent yearly changes in weight (g per year) from a random effects meta-analysis across centres (in a model, including baseline intake of fat, carbohydrate (in kcal) and alcohol intake (in categories), baseline age, weight, height, physical activity, educational level, follow-up time and changes in smoking status between baseline and follow-up, and for women, baseline menopausal status and use of hormone replacement therapy) subsequent changes in WC were present, besides a borderline significant positive association for protein of unknown origin among men (Table 2) However, as for weight changes, some heterogeneity across centres was present for total and animal protein with estimates ranging from significantly positive to significantly inverse associations (Figures 3 and 4), whereas less heterogeneity was seen for protein of plant and unknown origin Interactions We tested for interactions between daily protein intake and baseline BMI (below or above 27 kg m 2 ), age (linear, below and above 60 years of age), follow-up time (linear) and changing smoking status on the association with both weight and waist changes None of the potential effect modifiers investigated altered the association between total, animal, plant or unknown protein and annual changes of weight and WC based on the criteria described in the method section (data not shown) Supplementary analyses of subtypes of animal protein The findings on animal proteins lead to suggestion of additional exploratory analyses on protein intake from a number of main animal food sources, such as red and processed meat, fish, poultry and dairy products These analyses showed that the positive association between animal protein and weight gain could mainly be ascribed to protein from red and processed meat and poultry, rather than to protein from fish and dairy products (Table 3) As for total animal protein none of the different animal protein sources were strongly associated to changes in WC; intake of protein from fish and dairy products tended to be inversely associated and the intake of red and processed meat seemed to be positively associated with changes in WC Sensitivity analyses Additional analyses were conducted, in which protein from unknown origin were considered as either solely belonging to animal or plant origin Including protein of unknown origin in either animal or plant protein did not change the overall estimates for either animal or plant protein, although it did slightly influence the individual estimates in some centres When conducting the analyses using a traditional energy substitution model, in which protein replaces intake of other energy sources (fixed total energy intake), our overall results were very similar to those obtained from the partition model Total and animal protein was still positively associated with weight gain and the association remained stronger among women than that among men The only

6 Study ID ES (95% CI) % Weight 1109 Men Subtotal (I-squared = 659%, p = 0012) Women Subtotal (I-squared = 733%, p = 0002) Overall (I-squared = 738%, p = 0000) NOTE: Weights are from random effects analysis 9656 (3240, 16071) 8887 (2332, 15441) 2263 (-3590, 8115) (-16954, 4628) (-6691, 2549) 2934 (131, 5737) 3003 (-835, 6841) (7966, 15661) (4793, 17457) (4579, 17671) (458, 27721) 3416 (-1509, 8341) 3091 (091, 6092) 8238 (4082, 12395) 5622 (2686, 8558) Figure 2 Association between intake of animal protein (per 150 kcal) and subsequent yearly changes in weight (g per year) from a random effects meta-analysis across centres (in a model, including baseline intake of plant and unknown protein, fat, carbohydrate (in kcal) and alcohol intake (in categories), baseline age, weight, height, physical activity, educational level, follow-up time and changes in smoking status between baseline and follow-up, and for women, baseline menopausal status and use of hormone replacement therapy) clear difference was that the non-significant increase in weight among women for higher daily intake of plant protein, now reached statistical significance For changes in WC, as in the partition model, no significant associations were present (data not shown) Exclusion of incident disease (diabetes, cancer and cardiovascular disease) between baseline and follow-up did not change the results and overall conclusions notably, beside a tendency towards a stronger positive association between plant protein and yearly weight change among women, although still without reaching significance Use of calibrated dietary data resulted in larger absolute estimates for weight changes but similar overall conclusions Using calibrated dietary data indicated positive overall associations between total and animal protein and waist changes for women, which were not seen in the analyses with observed dietary data This positive effect was to some extent driven by, where the association changed from 003 to 129 cm per year per 150 kcal higher daily total protein intake, whereas much less extreme changes were seen for the calibrated estimates in other centres Excluding from the random effect meta-analysis reduced the overall estimate and the heterogeneity across centres considerably, and no significant association was then present Finally, correcting the anthropometric data (adjustment for self-report and clothing) did not change the associations and conclusions Discussion In this prospective cohort study with participants from five European countries, total protein and protein from animal sources was positively associated with subsequent body weight change, strongest among women, whereas plant protein was not associated with weight changes The association with animal proteins could be attributed mainly to protein from red or processed meat and poultry rather than to fish and dairy products Intake of total or subgroups of protein was not associated with subsequent overall changes in WC The results were all stable in different sensitivity analyses and no clear indication of modifications of the associations was present, when testing selected potential effect modifiers The strengths of our study include its large-scale and multi-centre prospective study design (nearly participants from six cohorts across European countries with repeated anthropometric measurements), the inclusion of both men and women, the wide age range included and

7 1110 Study ID ES (95% CI) % Weight Men Subtotal (I-squared = 770%, p = 0001) Women Subtotal (I-squared = 877%, p = 0000) Overall (I-squared = 831%, p = 0000) NOTE: Weights are from random effects analysis 013 (005, 021) 010 (002, 019) -001 (-007, 006) -004 (-016, 008) -002 (-007, 003) -004 (-007, 000) 002 (-004, 007) 014 (009, 019) 003 (-005, 011) 002 (-006, 009) 012 (-006, 029) 001 (-004, 006) -009 (-014, -003) 003 (-005, 011) 002 (-002, 007) Figure 3 Association between intake of total protein (per 150 kcal) and subsequent yearly changes in WC (cm per year) from a random effects meta-analysis across centres (in a model, including baseline intake of fat, carbohydrate (in kcal) and alcohol intake (in categories), baseline age, weight, height, WC, physical activity, educational level, follow-up time and changes in smoking status between baseline and follow-up, and for women, baseline menopausal status and use of hormone replacement therapy) participants with widely varying intake and sources of dietary protein 5,14 The analyses were based on centrally standardized covariables and harmonized nutrient databases (EPIC nutritional database (ENDB)) 15 The detailed information about the animal and plant origin of all food items, from which intake of animal and plant protein has been estimated provided important extra knowledge to investigate the association between health outcomes and sources of protein The minor part of protein in all centres that could not be classified as either animal or plant origin, could hypothetically introduce bias Sensitivity analyses assuming that this remnant was either solely of plant or animal origin did not change the overall result for either plant or animal protein indicating that this was not the case Use of the FFQ for dietary assessment has the potential problems of possible under- or over-reporting of specific food intake However, this problem might be a bigger issue for fat intake than protein, as some studies indicate a presumed relatively greater underestimation of fat and/or carbohydrate, than of protein 16,17 We only analyzed dietary intake at one time point (baseline) and were consequently not able to take dietary changes during follow-up into account Repeated dietary information would be of interest However, studying concordant changes in both diet and body size, as done in some studies, 18 makes it difficult to separate the exposure from the possible effects and introduce difficulties due to a possible modification of the diet as a consequence of a change in weight or WC, so-called reverse causality Thus to study changes in diet followed by weight changes would optimally require at least three examination time points Weight and WC were self reported at follow-up instead of measured in four out of six study centres, potentially adding to the heterogeneity between centres In additional analyses, we corrected anthropometrics for clothing differences and self-reporting using methods previously developed in the EPIC study 13 These corrections did not change the study conclusions There was a considerable difference in follow-up time between the six cohorts, introducing the possibility of differential weight changes at varying follow-up time To overcome this difference, yearly weight changes were analyzed instead of absolute changes over the entire followup time and in addition we adjusted for duration of followup time in the analyses Some degree of heterogeneity across centres was present for the associations between protein and changes in weight or WC Thus, we presented the results for each cohort separately, and also calculated a pooled estimate of effect, using a random effects model that allowed us to take into account the between cohort heterogeneity, as well as allowing for different influences of the potential confounders

8 Study ID ES (95% CI) % Weight 1111 Men Subtotal (I-squared = 817%, p = 0000) Women Subtotal (I-squared = 880%, p = 0000) Overall (I-squared = 853%, p = 0000) NOTE: Weights are from random effects analysis 015 (007, 022) 011 (002, 020) -002 (-009, 005) -007 (-019, 006) -006 (-011, -000) -003 (-007, 001) 001 (-005, 008) 016 (011, 021) 004 (-004, 012) 002 (-006, 010) 014 (-004, 032) 001 (-005, 007) -008 (-013, -002) 004 (-004, 013) 003 (-002, 008) Figure 4 Association between intake of animal protein (per 150 kcal) and subsequent yearly changes in WC (cm per year) from a random effects meta-analysis across centres (in a model, including baseline intake of plant and unknown protein, fat, carbohydrate (in kcal) and alcohol intake (in categories), baseline age, weight, height, WC, physical activity, educational level, follow-up time and changes in smoking status between baseline and follow-up, and for women, baseline menopausal status and use of hormone replacement therapy) Table 3 Associations (pooled estimates and 95% CI) between intake of protein from main animal food sources (per 150 kcal) and subsequent yearly changes in weight and WC from a random effects meta-analysis across centres a Weight changes (g per year) Waist changes (cm per year) Men Women All Men Women All Per 150 kcal Estimate (95% CI) Estimate (95% CI) Estimate (95% CI) Estimate (95% CI) Estimate (95% CI) Estimate (95% CI) Protein from Red/processed meat 404 ( 105 to 914) 1169 (869 to 1470) 810 (468 to 1153) 006 ( 001 to 013) 015 (006 to 024) 011 (004 to 017) Poultry 723 ( 141 to 1586) 1609 (821 to 2397) 1149 (498 to 1800) 001 ( 013 to 015) 007 ( 006 to 019) 004 ( 005 to 013) Fish 58 ( 705 to 590) 896 (15 to 1778) 408 ( 169 to 985) 008 ( 016 to 00) 008 ( 021 to 006) 008 ( 015 to 00) Dairy 65 ( 528 to 658) 230 ( 374 to 833) 132 ( 262 to 527) 004 ( 008 to 001) 005 ( 011 to 001) 005 ( 008 to 002) Abbreviations: 95% CI, 95% confidence interval; WC, waist circumference a In a model, including baseline intake of fat, carbohydrate (in kcal) and alcohol intake (in categories), baseline age, weight, (WC in the WC change analyses), height, physical activity, educational level and, for women, menopausal status and hormone use, follow-up time and changes in smoking status between baseline and follow-up All protein subgroups were included in the same model The remaining protein intake was adjusted for by including a variable protein from other sources The protein variables sum up to total protein intake across cohorts 19 Reasons for the heterogeneity across centres are not clear However, the dietary sources of protein do vary in the different centres, 14 which might explain part of the differences In the present study the additional analyses on protein from different animal food sources did, to some content, confirm this Less heterogeneity across centres was generally seen here, in comparison to when total animal protein was considered (data not shown) Differences in measurement methods of the anthropometrics (self reported in four, whereas measured in two centres at follow-up) might also introduce heterogeneity, but associations in centres with measured data did not align more than associations based on self-reported data at follow-up Total protein and protein from various food sources, rather than intake of the whole food items as such, are studied in

9 1112 the present study It can therefore not be excluded that the other macronutrients in the protein-rich food items are partly responsible for the results found for protein A large effort was however done to carefully adjust the associations for the remaining macronutrients to prevent this, but residual confounding cannot be totally excluded Interpretation of the results needs to be done in the light of this The reasons for differences in the associations between dietary protein and changes in abdominal or general body size are not clear It could be that higher animal protein intake mainly is associated with overall weight gain, but not specifically to gain at the waist To be able to investigate this question in detail, accurate body composition data are necessary; unfortunately, such data do not exist in the present cohorts Another reason why associations were seen for weight, but not for waist changes, could be the mixed effects for different sources of animal protein on changes in WC, which ends up in no overall association between total animal protein and waist changes, whereas for weight changes the estimates are in the same direction Finally, the self-reported WC may be less precise than self-reported weight introducing a larger degree of measurement error and thereby, a dilution of a possible association In comparison to the present results, among the rather few previous studies one study also found a positive association between changes in weight and total protein intake in US white, but not in black, men and women, 20 and a longitudinal study showed that participants with a high body fatness (425% for men and 435% for women) at the age 36 years reported higher intake of protein at the age 32 and 36 years compared with participants with lower body fatness; no information about weight and WC changes was included 21 Most other studies found no associations 4 Different methodology as, for example, energy adjustment methods 10 and adjustments for other dietary factors and covariates, may contribute to the inconsistency of the results Even fewer studies have investigated the associations between protein intake and changes in WC or waist-to-hip ratio and for these studies both positive, inverse and no associations were seen 20,22,23 Protein from different dietary sources has hardly been studied making it impossible to compare the present results with other prospective cohort studies The present study did indicate that animal and plant protein were differently associated to weight and waist gain and that associations might differ for different sources of animal protein The reasons why protein from different sources may influence weight differently are still very speculative and need further investigation High intake of animal protein may reflect a western diet pattern high in intake of red meat, which has been related to weight gain 24 On the other hand protein from meat has been seen to produce a higher 24-h energy expenditure than protein from soy and is consequently assumed to be associated to lower weight gain than other sources of protein 25 This hypothesis is, however, based on a mechanistic study and it is unknown whether this applies in the long run in members of the free-living populations There are hypotheses suggesting that intake of dairy sources might prevent weight gain and produce a preferential loss of abdominal fat Here, the suggested mechanism has primarily been related to the high content of calcium, which may function synergistically in combination with bioactive compounds, such as angiotensin-converting enzyme inhibitors and the rich concentration of branched-chain amino acids 26 In conclusion, this study showed associations between higher daily intake of total protein and protein from different food items of animal origin and subsequent moderate weight gain, especially among women, whereas no clear overall associations were seen for waist changes The present study adds valuable information to the limited existing evidence on the associations between intake of different sources of protein and subsequent changes in weight and WC More studies investigating the different subtypes of protein and whether different associations are present for different body sites are still needed, together with studies investigating whether the effects can be ascribed to protein as such, or whether they are rather due to the complex of nutrients found in protein-rich food items Conflict of interest The authors declare no conflict of interest Acknowledgements The diet, obesity and genes (Diogenes) project is a pan European study within the EU Sixth Framework Programme for Research and Technological Development ( ) (FOOD-CT , This integrated programme was set up to target the issue of obesity problem from a dietary perspective, seeking new insights and new routes to prevention We thank the EPIC study investigators Further this work is part of the project Hepatic and Adipose Tissue and Functions in the Metabolic Syndrome (HEPADIP, which is supported by the European Commission as an Integrated Project under the Sixth Framework Programme (Contract LSHM-CT ) and the research program of the Danish Obesity Research Centre (DanORC, danorcdk), which is supported by the Danish Council for Strategic Research (Contract ) Author contributions Dr JH had full access to all of the data in this study and takes responsibility for the integrity of the data and the accuracy

10 of the data analysis The conception of the study emerged from the building of the core objectives of the Diogenes study, in which TIAS was involved NJW, AT, DP, KO, HB, EJMF and TIAS designed the study, contributed for acquiring data and providing funding, helped with the interpretation of the results and gave critical comments on the manuscript All authors helpfully contributed to the study and draft versions and accepted the final version of the manuscript Financial disclosures None of the authors have any financial disclosures, with the exception of TIAS (listed on person/tias/disclosureshtml) Role of the sponsors None of the study sponsors had a role in study design, in data collection, analysis or interpretation, in writing the report or in the decision to submit for publication Disclaimer The views expressed are those of the authors and should not be construed to represent the positions of any of the sponsors of the study References 1 Guh DP, Zhang W, Bansback N, Amarsi Z, Birmingham CL, Anis AH The incidence of co-morbidities related to obesity and overweight: a systematic review and meta-analysis BMC Public Health 2009; 9: 88 2 Halton TL, Hu FB The effects of high protein diets on thermogenesis, satiety and weight loss: a critical review J Am Coll Nutr 2004; 23: Westerterp-Plantenga MS, Nieuwenhuizen A, Tome D, Soenen S, Westerterp KR Dietary protein, weight loss, and weight maintenance Annu Rev Nutr 2009; 29: Summerbell CD, Douthwaite W, Whittaker V, Ells LJ, Hillier F, Smith S et al The association between diet and physical activity and subsequent excess weight gain and obesity assessed at 5 years of age or older: a systematic review of the epidemiological evidence Int J Obes (Lond) 2009; 33(Suppl 3): S1 S92 5 Riboli E, Hunt KJ, Slimani N, Ferrari P, Norat T, Fahey M et al European Prospective Investigation into Cancer and Nutrition (EPIC): study populations and data collection Public Health Nutr 2002; 5(Suppl): Slimani N, Kaaks R, Ferrari P, Casagrande C, Clavel-Chapelon F, Lotze G et al European Prospective Investigation into Cancer and Nutrition (EPIC) calibration study: rationale, design and population characteristics Public Health Nutr 2002; 5(Suppl): Rosner B, Gore R Measurement error correction in nutritional epidemiology based on individual foods, with application to the relation of diet to breast cancer Am J Epidemiol 2001; 154: Ferrari P, Kaaks R, Fahey MT, Slimani N, Day NE, Pera G et al Within- and between-cohort variation in measured macronutrient intakes, taking account of measurement errors, in the European Prospective Investigation into Cancer and Nutrition study Am J Epidemiol 2004; 160: Kaaks R, Riboli E, van Staveren W Calibration of dietary intake measurements in prospective cohort studies Am J Epidemiol 1995; 142: Willett WC Nutritional Epidemiology, 2nd edn, Oxford University Press: New York, USA, Greenland S Dose-response and trend analysis in epidemiology: alternatives to categorical analysis Epidemiology 1995; 6: Higgins JP, Thompson SG Quantifying heterogeneity in a meta-analysis Stat Med 2002; 21: Haftenberger M, Lahmann PH, Panico S, Gonzalez CA, Seidell JC, Boeing H et al Overweight, obesity and fat distribution in 50- to 64-year-old participants in the European Prospective Investigation into Cancer and Nutrition (EPIC) Public Health Nutr 2002; 5(Suppl): Halkjaer J, Olsen A, Bjerregaard LJ, Deharveng G, Tjonneland A, Welch AA et al Intake of total, animal and plant proteins, and their food sources in 10 countries in the European Prospective Investigation into Cancer and Nutrition Eur J Clin Nutr 2009; 63(Suppl 4): S16 S36 15 Slimani N, Deharveng G, Unwin I, Southgate DA, Vignat J, Skeie G et al The EPIC nutrient database project (ENDB): a first attempt to standardize nutrient databases across the 10 European countries participating in the EPIC study Eur J Clin Nutr 2007; 61: Heitmann BL, Lissner L Dietary underreporting by obese individualsfis it specific or nonspecific? BMJ 1995; 311: Heitmann BL, Lissner L, Osler M Do we eat less fat, or just report so? Int J Obes Relat Metab Disord 2000; 24: Koh-Banerjee P, Chu NF, Spiegelman D, Rosner B, Colditz G, Willett W et al Prospective study of the association of changes in dietary intake, physical activity, alcohol consumption, and smoking with 9-y gain in waist circumference among US men Am J Clin Nutr 2003; 78: DerSimonian R, Laird N Meta-analysis in clinical trials Control Clin Trials 1986; 7: Ludwig DS, Pereira MA, Kroenke CH, Hilner JE, Van Horn L, Slattery ML et al Dietary fiber, weight gain, and cardiovascular disease risk factors in young adults JAMA 1999; 282: Koppes LL, Boon N, Nooyens AC, van Mechelen W, Saris WH Macronutrient distribution over a period of 23 years in relation to energy intake and body fatness Br J Nutr 2009; 101: Halkjaer J, Tjonneland A, Thomsen BL, Overvad K, Sorensen TIA Intake of macronutrients as predictors of 5-y changes in waist circumference Am J Clin Nutr 2006; 84: Stookey JD, Adair LS, Popkin BM Do protein and energy intakes explain long-term changes in body composition? J Nutr Health Aging 2005; 9: Schulze MB, Fung TT, Manson JE, Willett WC, Hu FB Dietary patterns and changes in body weight in women Obesity (Silver Spring) 2006; 14: Mikkelsen PB, Toubro S, Astrup A Effect of fat-reduced diets on 24-h energy expenditure: comparisons between animal protein, vegetable protein, and carbohydrate Am J Clin Nutr 2000; 72: Zemel MB The role of dairy foods in weight management J Am Coll Nutr 2005; 24: 537S 546S 1113 Supplementary Information accompanies the paper on website (