Plasma Concentration of 36 (Poly)phenols and Prospective Body Weight Change in Participants from the EPIC Cohort Open Access

Phenols and polyphenols, referred to as (poly)phenols hereafter, are secondary metabolites from plants with bioactive properties [1]. More than 500 individual (poly)phenols have been identified from different plant foods in the human diet, with flavonoids and phenolic acids representing the major classes [1]. Fruits, vegetables, wholegrain cereals, olive oil, cocoa, tea, and coffee are some of the main dietary sources of (poly)phenols [2].

Several epidemiological studies estimating intakes through dietary questionnaires or urine biomarkers have reported associations between dietary (poly)phenols and body weight (BW) change [3‒6]. A French cohort study observed an association between higher intake of different (poly)phenols and a smaller increase in body mass index (BMI) and waist circumference [3]. An Italian cohort study concluded that higher flavonoid intake was associated with a lower BW [4]. Despite not assessing specifically anthropometric outcomes, three recent cohort studies have reported data on dietary (poly)phenols and central obesity as marker of metabolic syndrome (MetS). In a Danish cohort, dietary intake of total (poly)phenols, flavonoids, and phenolic acid was associated with lower odds of MetS, defined as the presence of at least three markers including high WC, high plasma triglycerides or HDL-cholesterol, high systolic or diastolic blood pressure, and/or high HbA1c [7]. Likewise, higher flavonoid consumption in Chinese adults [8] and dietary intake of the subclass flavonol in a Polish cohort [9] were found to be potentially protective against MetS, probably mainly impacting central obesity.

We recently investigated the associations between dietary intake of (poly)phenols (i.e., total, classes, subclasses, and individually) and 5-year BW change in the European Prospective Investigation into Cancer and Nutrition (EPIC) cohort [5, 6]. In general, intakes of most of the individual (poly)phenols and total flavonoids were inversely associated with 5-year BW change, except for (poly)phenols from coffee (i.e., hydroxycinnamic acids).

Our motivation to investigate the hypothesis of an inverse association between plasma (poly)phenols and BW gain is multifaceted. Previous research has suggested that dietary (poly)phenols may possess anti-obesity properties [3‒6, 8, 9], but reliance on self-reported dietary data can introduce bias [10]. To mitigate this and obtain a more objective measure of dietary intake, we turned to plasma biomarkers. These nutritional biomarkers offer a reliable quantification of (poly)phenol exposure, allowing for a more accurate assessment of their impact on BW changes. Notably, limited research has explored the relationship between plasma (poly)phenols and obesity markers, underscoring the need for a more comprehensive investigation. Utilizing data from approximately 800 participants with information on 36 well-characterized plasma (poly)phenols and BW change, our study aimed to elucidate the potential role of these (poly)phenol biomarkers in BW management.

The EPIC study is a multicentre prospective cohort that includes over half a million participants from 10 Western European countries [11]. It primarily aimed to evaluate the association between dietary and lifestyle factors and the risk of cancer and cancer-related diseases. Details on recruitment and study design have been published previously [11]. For the present study, a convenient sample of cancer-free controls was selected from two EPIC-nested case-control studies (colon and thyroid cancer studies) (n = 1,321) [12, 13]. Participants with no plasma (poly)phenol measurement (n = 430) and with no follow-up BW were excluded (n = 73). Data from Greece were not available at the time of this study, and therefore, Greek participants were excluded (n = 57). The final sample included 761 participants (online suppl. Fig. S1; for all online suppl. material, see https://doi.org/10.1159/000535803).

The majority of lifestyle factors were collected at recruitment through standardized questionnaires, and medical history was self-reported at baseline [11]. For this study, we included data on baseline age, sex, physical activity, education level, menopausal status, and alcohol consumption. We also included data on changes in smoking status at follow-up. Dietary data were collected at baseline through validated country-specific dietary questionnaires [14]. To estimate nutrient and total energy intake, the standardized EPIC Nutrient Database was used [15]. Baseline BW and height were collected by trained staff using standardized methods, except for centres from in France, Oxford, and Norway, where they were self-reported [11]. At follow-up, BW was self-reported in all centres, with the exception of Doetinchem and Cambridge, where it was measured [16]. The accuracy of self-reported BW was improved with prediction equations derived from EPIC-Oxford participants, in which BW at baseline was both measured and self-reported [17]. To calculate the 5-year BW change, baseline BW was subtracted from follow-up BW for each participant. The result was then divided by the number of follow-up years to obtain the annual BW change and multiplied by 5 to finally obtain the BW change over 5 years.

Plasma (poly)phenol measurements were available to participants from 19 EPIC centres (from Denmark, Germany, Italy, Norway, Spain, Sweden, the Netherlands, and the UK). Blood samples were collected at baseline following standardized procedures [18]. Plasma concentrations of (poly)phenols (nmol/L) were measured by a highly sensitive method based on differential isotope labelling with 12 and 13 C-dansyl chloride and an ultra-performance liquid chromatography-tandem mass spectrometry system (UPLC-MS/MS) with prior enzymatic hydrolysis. This method allowed the quantification with high accuracy and reproducibility of a large selection of compounds representative of the main classes of dietary (poly)phenols in low volumes of plasma [18]. Limits of quantification (LOQ) for the individual (poly)phenols varied between 0.11 nmol/L for apigenin and 44.4 nmol/L for quercetin. Intra-batch coefficients of variability (CV) varied between 2.3 and 9.0%, and inter-batch CV was <20% for the majority of (poly)phenols, except for quercetin (23.4%) and enterodiol (20.3%). Details about the laboratory methodology have been published elsewhere [18].

Baseline characteristics were calculated according to categories of BW change over 5 years. BW change was categorized into three groups: loss (<−1.5 kg/5 years), maintenance (from −1.5 to 1.5 kg/5 years), and gain (>1.5 kg/5 years). Data on 36 plasma (poly)phenols (nmol/L) spanning the majority of compounds found in the human diet were available for this study and were analysed as continuous variables. Right-skewed distributions were standardized using log2-transformations with zero values imputed with half the LOQ for each (poly)phenol. Therefore, results were expressed as doubling plasma (poly)phenol concentrations. Spearman rank coefficients were calculated to assess correlations between individual plasma (poly)phenol concentrations. Correlations were also evaluated between plasma (poly)phenol levels and the intake of the following 12 (poly)phenol-rich food groups: potatoes and other tubers, vegetables, legumes, cereals, fats and oils, condiments, fruits, nuts and seeds, olives, coffee, tea, and herbal teas.

Two general linear mixed models (GLMM) were fitted to evaluate the association between plasma (poly)phenols (nmol/L) and continuous BW change (kg/5 years). In addition, we fitted multinomial logistic regression models using BW change over 5 years as a categorical variable with three categories: BW maintenance (from −1.5 to 1.5 kg/5 years) as the reference category, BW loss (<−1.5 kg/5 years), and BW gain (>1.5 kg/5 years).

For model adjustment, we selected variables a priori [19‒23]: age (continuous, years), sex, baseline BMI (continuous, kg/m²), follow-up years (continuous), energy intake (continuous, kcal/day), fibre intake (continuous, g/day), vitamin C intake (continuous, mg/day), physical activity using the Cambridge index [24] (categories: inactive, moderately inactive, moderately active, active), menopausal status (categories: pre, post and peri-menopausal), change in smoking status at follow-up (categories: stable never smoker, stable current smoker, starter, quitter, former at follow-up), laboratory batch, plausibility of energy intake reporting (categories according to the ratio of reported energy intake (EI) to predicted basal metabolic rate (BMR): under (EI:BMR <1.14), plausible (1.14–2.1) and over reporters (>2.1), with the use of cut-off points proposed by Goldberg et al. [25]), education level (categories: none, primary, technical, secondary, longer), alcohol consumption (continuous, g/day), EPIC centres, and type of cancer project (thyroid/colon). Some of these variables were excluded after stepwise regression screening (i.e., menopausal status, education level, and alcohol consumption). Participants with missing values in categorical adjustment variables were placed in a separate category, while continuous adjustment variables presented no missing values. Aikaike Information Criterion was used to compare and select the best-fitting models. Restricted cubic splines were used to evaluate the non-linearity of the associations for continuous covariates. BMI, follow-up years, and energy intake showed a non-linear relationship with BW change. Thus, splines with 3 knots (percentile 10, 50, and 90) were included as covariates for these three variables. Knot positions were determined using the Harrell criteria [26].

For GLMM, in Model 1, age, sex, and baseline BMI were used as covariates. EPIC centres and type of cancer project (thyroid or colon cancer control groups) were included as random effects to control for potential confounding due to differences in follow-up procedures, questionnaire design, and blood sample collection and analysis. Model 2 was additionally adjusted for years of follow-up, physical activity level, change in smoking status at follow-up, laboratory batch, energy intake, and plausibility of energy intake reporting. To account for certain dietary variables that may be related to both exposure and outcome, we fitted a third model, adding intake of fibre (g/day) and intake of vitamin C (mg/day) as further confounding variables. A sensitivity analysis was performed for each model to exclude participants with chronic diseases at baseline (diabetes, stroke, or myocardial infarction, n = 92, leaving n = 669 included). We further fitted interactions between (poly)phenol concentrations and sex, baseline age, and smoking status at follow-up [27, 28]. p values for interactions were calculated using the likelihood ratio test. The adjustment variables for multinomial logistic regression models were the same as the GLMMs, except for EPIC centres and type of cancer project that was included here as fixed effects.

For all models, p values were adjusted by computing the false discovery rate (FDR) to control for multiple comparisons and considered statistically significant at an FDR q value <0.05. All statistical analyses were performed with R Statistical Software version 4.2.2 [29].

Table 1 shows characteristics of the 761 participants included in this study according to categories of BW change over 5 years. Overall, mean (SD) 5-year BW change was 0.6 (4.5) kg and ranged from − 18.2 to 21.8 kg. A total of 293 participants (38.6%) gained >1.5 kg of weight over 5 years, whereas 194 (25.5%) lost >1.5 kg/5 years. Baseline BW and BMI were higher in participants in the BW loss category. More than 50% of participants in the BW loss category had lower levels of education, whereas more than 50% of participants in the BW gain category presented higher educational levels. According to change in smoking status at follow-up, a high proportion of participants were classified as never smokers (48%), followed by former smokers (including never at baseline and former at follow-up, 23%). The majority of participants were females (77%), as the thyroid cancer control group included only female participants and represented 40% of the study population, while also 60% of the colon cancer group were females. Table 2 shows the median and 5th and 95th percentiles of plasma concentrations of (poly)phenols among participants. The highest plasma concentration values were observed for caffeic acid (median 367.0 nmol/L) and 4-hydroxy-phenylacetic acid (287.0 nmol/L), whereas the lowest concentrations were observed for equol (0.4 nmol/L) and enterodiol (1.0 nmol/L). Figure 1 illustrates the statistically significant (p ≤ 0.05) correlations between each of the plasma (poly)phenols and between plasma (poly)phenols and (poly)phenol-rich food groups. The strongest correlation was observed between 3,5-dihydroxy-benzoic acid and 3,5-dihydroxy-phenylpropionic acid (r = 0.90). Correlation coefficients between plasma (poly)phenols and food groups were mostly low (r < 0.40), only highlighting correlations for tea (epicatechin r = 0.46; 3,5-dihydroxy-benzoic acid r = 0.45) and coffee (ferulic acid r = 0.51). The strongest inverse correlations were identified between protocatechuic acid and caffeic acid (r = −0.52), while isoharmentin displayed notable inverse correlations with three compounds: 3,4-dihydroxy-phenilacetic acid (r = −0.52), gallic acid (r = −0.55), and 3,4-dihydroxy-phenilpropionic acid (r = −0.56).

We obtained nearly identical results for models 2 and 3. Consequently, we present the more extensively adjusted model, namely, model 3. Figure 2 shows results after modelling continuous 5-year BW change against log2-transformed plasma concentrations of (poly)phenols and adjusting for relevant confounders. Only plasma vanillic acid concentration appeared to be inversely associated with BW change in model 3 (beta per doubling concentration: −0.53 kg/5 years; 95% confidence interval [CI]: −0.99, −0.07), and a borderline association was observed for tyrosol (beta: −0.29 kg/5 years; 95% CI: −0.62, 0.00). It is worth mentioning that the majority of (poly)phenols displayed a suggestive trend towards inverse associations with BW gain. However, it is important to emphasize that, following the application of FDR adjustment, none of these associations reached statistical significance. Furthermore, similar trends were observed in the sensitivity analysis, which excluded participants with chronic diseases at recruitment (online suppl. Table S1). No statistically significant interactions were observed with respect to sex, age, or changes in smoking status at follow-up.

Table 3 shows the results from multinomial logistic regression analyses comparing BW loss versus BW maintenance (reference category). Most of the (poly)phenols showed a tendency towards BW loss. Plasma naringenin concentration was associated with BW loss in model 3 (odds ratio [OR]: 1.1; 95% CI: 1.0, 1.2). Additionally, plasma levels of ferulic acid (OR: 1.3; 95% CI: 1.0, 1.7), caffeic acid (OR: 1.7; 95% CI: 1.0, 3.0), and kaempferol (OR: 1.7; 95% CI: 1.0, 2.9) exhibited borderline associations. Table 4 presents results comparing BW gain with BW maintenance. We observed a pattern consistent with our previous analysis. Specifically, there were no statistically significant associations for BW gain, either before or after the application of the FDR correction, except for a suggestive inverse association for 3,5-dihydroxybenzoic acid (OR: 0.9; 95% CI: 0.8, 1.0).

The current study suggested a tendency towards BW maintenance or reduction over 5 years at higher concentrations of plasma (poly)phenols. We would like to draw attention to the potential protective associations observed between plasma concentrations of vanillic acid and naringenin and BW loss. It is important to note that while the results related to vanillic acid are derived from GLMMs, those concerning naringenin are derived from multinomial models. It is noteworthy that both vanillic acid and naringenin have been proposed as metabolites of dietary flavonoids, as discussed below. Given their shared potential dietary sources, it is plausible that they may exhibit similar behaviours. However, we must approach these associations with caution due to the complexity of the relationships involved.

Even though both GLMMs and multinomial models included the same covariates and followed the same tendency, results were different. One possible explanation for these differences might be that plasma concentrations do not necessarily represent the same compounds present in foods. Natural vanillic acid is a phenolic compound widely present in vanilla beans and in different fruits and grains [30], but it can also be a derivative metabolite from different precursors. For example, as a result of the host metabolism, intake of caffeic acid from coffee has been shown to generate a high increase in urine levels of vanillic acid in vivo [31]. Likewise, one study using animal models observed that after red wine consumption, which is naturally rich in flavonoids and phenolic acids, vanillic acid was found in animal urine together with other microbial metabolites [32]. One study in vivo showed that after intake of an oral dose of cyanidin 3-glucoside (an anthocyanin widely found in fruits, particularly berries), vanillic acid was present in both human urine and serum [33].

Naringenin occurs naturally in an inactive form as naringin (the glycoside form) and is converted into its active form (aglycone) by bacteria belonging to the gut microbiome [34]. Naringin is a component of the everyday human diet, mainly present in citrus fruits. The gut microbiota metabolizes naringin and breaks it down to naringenin which is absorbed in the gut [35]. One study using animal models observed that the major components in plasma and urine after 18 h of gastric gavage of naringenin were glucuronides as well as the colonic metabolite 3-(4-hydroxyphenyl) propionic acid [36]. Thus, compounds present in plasma are not necessarily classified as those precursors present in diet – while naringenin is classified as a flavonoid in diet, one of its major human gut microbial metabolites is 3-(4-hydroxyphenyl) propionic acid, classified as phenolic acid [37].

As far as we know, there is a lack of evidence on plasma (poly)phenols and obesity-related parameters. Previous cohort studies have investigated similar relationships but with dietary or urinary (poly)phenols. The SU.VI.MAX study [3] concluded that higher dietary (poly)phenol intake, particularly flavonoids and their subclasses, may help reduce BW gain. The MEAL study observed that higher flavonoid intake was associated with a lower BW [4]. In line with this, the large randomized PREDIMED study concluded that higher (poly)phenol intakes, objectively measured through total urinary (poly)phenol excretion, were inversely correlated with BW and obesity [38]. Bertoia et al. [39] observed, in three prospective cohorts, that higher intake of foods rich in flavonoids may contribute to BW maintenance and prevention of obesity. A longitudinal study from the Netherlands found that a higher intake of flavonols and flavones was associated with lower increase in BMI over 14 years in the general female population [40].

In comparing our current study to our previous work utilizing dietary (poly)phenol data from the EPIC-PANACEA cohort [5, 6], we observed few distinct differences in the findings. Our previous dietary-based research found a strong, statistically significant inverse link between (poly)phenol intake and BW gain. However, in this study using plasma (poly)phenol concentrations, associations were less pronounced and borderline statistically significant. Two key factors contribute to these disparities. Firstly, we had dietary data for a much larger subgroup (around 350,000) of the EPIC cohort, while plasma (poly)phenol analysis was limited to under 800 participants due to logistical constraints. This variation in sample size could affect our ability to detect associations. Secondly, using biomarkers like plasma (poly)phenols offers a more objective measure of diet but may not capture its full complexity due to individual differences in bioavailability and metabolism [41]. In contrast, dietary data provide a comprehensive view of dietary patterns [42]. In spite of these differences, both measurements (dietary data and nutritional biomarkers) provide evidence that (poly)phenols may have a protective impact in BW maintenance.

In addition, the discrepancies in explaining these associations can be attributed to differences in (poly)phenol bioavailability. In general, (poly)phenols show low bioavailability (<20%) which can be affected by several factors, including chemical structure (including sugar moiety), food matrix, interaction with other compounds, and individual conditions like intestinal activity, gut microbiota composition, sex, or age [41]. According to pharmacokinetic studies, (poly)phenols that are most absorbed in humans are isoflavones and phenolic acids, followed by catechins, flavanones, and quercetin glucosides. Low-molecular-weight (poly)phenols are easily absorbed through the gut barrier [43, 44]. When they reach the basal membrane of the enterocytes and when they enter the liver, they are conjugated to glucuronic acid, sulphate, and methyl groups to facilitate their transport and excretion and limit their potential toxicity [43, 44]. Conversely, large molecular weight (poly)phenols such as proanthocyanidins are poorly absorbed in the small intestine, and they reach the colon to be metabolized by the gut microbiota into lower molecular weight compounds, generally phenolic acids, which can be partially absorbed in the colon [45].

Increasing evidence supports the idea that the gut microbiome plays a key role in the relationship between (poly)phenols and metabolism, body fat, and obesity. It has been shown, for example, that diets high in (poly)phenol-rich whole plant-based foods improve gut microbiota profiles when compared to diets high in animal-based products [46]. Interactions between (poly)phenols and the gut microbiome have also been shown to decrease obesity-related conditions such as development of adipose tissue and obesity-induced inflammatory genes [47]. Plasma concentrations of different (poly)phenols may be a reflection of these metabolites resulting from dietary (poly)phenols-microbiome interactions [48].

The robust positive correlations observed between certain plasma (poly)phenols, such as the noteworthy correlation between 3,5-dihydroxy-benzoic acid and 3,5-dihydroxy-phenylpropionic acid, can be attributed to a complex interplay of dietary factors, metabolic processes, and individual variability. In a previous study assessing the pharmacokinetics of novel metabolites in urine, it was observed that these two compounds were new candidate biomarkers for whole-grain wheat and rye intake [49]. Therefore, both 3,5-dihydroxy-benzoic acid and 3,5-dihydroxy-phenylpropionic acid were proposed as potential biomarkers to increase the accuracy of whole-grain wheat and rye intake in epidemiology studies. These findings underscore the multifaceted nature of (poly)phenol metabolism and its dependence on dietary patterns, metabolic pathways, and individual characteristics, offering valuable insights for future research in (poly)phenol metabolism and its potential health implications. The strong correlations observed between certain compounds and tea and coffee may reflect some characteristics of dietary patterns in our population. Tea and coffee are frequently consumed by individuals as part of their daily routines [11]. Consistent and habitual consumption of these beverages may result in a continuous supply of (poly)phenols to the body, potentially leading to higher and steadier plasma levels over time [50].

As previously mentioned, one limitation of our investigation is the interpretation of human study data when working with these compounds. For example, as the occurrence of these metabolites in circulating blood is the result of digestive and hepatic activity, the complex interaction between (poly)phenols, individual gut microbiota, and host metabolism cannot be disregarded [1]. Another limitation is the relatively small sample size with available measurements of plasma (poly)phenols, particularly for subgroup analyses. The EPIC sampling characteristics do not allow the total generalizability of these findings to other populations. In addition, the fact that female participants constituted the vast majority of the study population means males were underrepresented. Concentrations of (poly)phenols were measured in single plasma samples at baseline; thus, intra-individual variations in circulating levels of these compounds were not available, which could also lead to attenuation of the observed associations [51]. Because of the observational nature of the study design, although we were able to adjust our models for relevant lifestyle and dietary covariates, the possibility of residual confounding cannot be ruled out.

A major strength of this study is its prospective and multicentre design, including participants from different European countries with wide variation in diet. In addition, we had data on a relatively large number of (poly)phenols (n = 36), spanning all major classes found in the human diet. While plasma (poly)phenols in this study may indicate a plant-based diet, it is important to consider a broader context. While we incorporated fibre and vitamin C intake into our model, we recognize a lack of detailed knowledge regarding the potential biological mechanisms connecting plasma (poly)phenols to BW loss. Consequently, attributing our findings solely to a healthier diet may be overly simplistic. Nevertheless, further research is needed to explore how (poly)phenols and plant-based compounds influence BW regulation. In epidemiological studies, prediction of (poly)phenol intake mainly relies on food records and composition tables, often failing to assess total intake accurately [52]. Thus, assessing the relationship between diet-related compounds and health outcomes can be challenging. The measurement of plasma concentrations of (poly)phenols represents an objective measurement of a snapshot of internal exposure to these compounds that could come directly from different dietary sources or their precursors [51]. This has been a way of reducing the potential bias from dietary questionnaire-based data. While the evidence on the determination of plasma (poly)phenols is not extensive, a recent review described the common analytical technique utilizing LC-MS/MS [53]. However, the use of differential isotope labelling in our methodology is an extra advantage. This methodology reinforces the robustness, sensitivity, and specificity of our analytical approach and its applicability to quantifying low plasma (poly)phenol concentrations [54].

In conclusion, this prospective investigation suggested a tendency towards 5-year BW loss or maintenance at higher plasma (poly)phenol concentrations at baseline. Among specific (poly)phenols, vanillic acid and naringenin can be highlighted, which may act as biomarkers of flavonoid intake. While these specific associations seemed promising, they did not withstand FDR correction, indicating the need for caution in interpreting these results. These findings are preliminary, and further studies using (poly)phenol biomarkers are needed to better understand the potential observed trends. This study may contribute to the identification of specific (poly)phenols for future mechanistic studies or clinical trials on obesity-related pathways in humans.

We thank all EPIC participants and staff for their contribution to the study. We also thank Bertrand Hémon and Leila Luján Barroso for their valuable help with the EPIC database. We also thank Aarhus University, Aarhus, Denmark; the University Medical Center Utrecht, Utrecht, The Netherlands; the National Institute for Public Health and the Environment (RIVM), Bilthoven, the Netherlands; and Oxford University, UK, for their contribution and ongoing support to the EPIC Study. NGF acknowledges support from the MRC Epidemiology Unit (MC_UU_00006/3) and from the NIHR Cambridge Biomedical Research Centre nutrition, obesity, endocrinology, and metabolism (NOME) theme (NIHR203312). She is an NIHR Senior Investigator. The views expressed are those of the authors and not necessarily those of the NIHR or the Department of Health and Social Care.

The study has been performed in accordance with the ethical standards laid down in the 1964 Declaration of Helsinki and its later amendments and has obtained ethical approval from participating centres and the IARC Ethics Committee (reference number 20-02). Written informed consent was obtained from all study participants.

The authors have no conflicts of interest to declare. Where authors are identified as personnel of the International Agency for Research on Cancer/World Health Organization, the authors alone are responsible for the views expressed in this article, and they do not necessarily represent the decisions, policies, or views of the International Agency for Research on Cancer/World Health Organization.

This study has been funded by the Institute of Health Carlos III through the grant PI18/00191, co-funded by the European Regional Development Fund, ERDF, as a way to build Europe. The PANACEA project was funded by the European Union in the framework of the Public Health Programme (project number: 2005328). The coordination of EPIC is financially supported by the International Agency for Research on Cancer (IARC) and also by the Department of Epidemiology and Biostatistics, School of Public Health, Imperial College London, which has additional infrastructure support provided by the NIHR Imperial Biomedical Research Centre (BRC). The national cohorts are supported by: Danish Cancer Society (Denmark); Ligue Contre le Cancer, Institut Gustave Roussy, Mutuelle Générale de l’Education Nationale, Institut National de la Santé et de la Recherche Médicale (INSERM) (France); German Cancer Aid, German Cancer Research Center (DKFZ), German Institute of Human Nutrition Potsdam-Rehbruecke (DIfE), Federal Ministry of Education and Research (BMBF) (Germany); Associazione Italiana per la Ricerca sul Cancro-AIRC-Italy, Compagnia di San Paolo and National Research Council (Italy); Dutch Ministry of Public Health, Welfare and Sports (VWS), Netherlands Cancer Registry (NKR), LK Research Funds, Dutch Prevention Funds, Dutch ZON (Zorg Onderzoek Nederland), World Cancer Research Fund (WCRF), Statistics Netherlands (The Netherlands); Health Research Fund (FIS) – Institute of Health Carlos III (ISCIII), Regional Governments of Andalucía, Asturias, Basque Country, Murcia and Navarra, and the Catalan Institute of Oncology – ICO (Spain); Swedish Cancer Society, Swedish Research Council and County Councils of Skåne and Västerbotten (Sweden); Cancer Research UK (14136 to EPIC-Norfolk; C8221/A29017 to EPIC-Oxford), Medical Research Council (1000143 to EPIC-Norfolk; MR/M012190/1 to EPIC-Oxford) (UK). We thank the CERCA Program/Generalitat de Catalunya for institutional support of IDIBELL. MG-L, EA-A, and RZ-R were supported by the PFIS (FI19/00185), “Sara Borrell” (CD20/00071), and “Miguel Servet II” (CPII20/00009) programmes, respectively, from the Institute of Health Carlos III (co-funded by the European Social Fund (ESF) investing in your future). The funders had no role in the study design, data collection and analysis, decision to publish, or preparation of the manuscript.

R.Z.-R. designed research and wrote paper – reviewed and edited; M.G.-L. conducted research and wrote paper – prepared original draft; A.K.E., A.T., J.A.R., S.S., C.C., V.K., T.J., M.B.S., A.O., F.P., R.T., L.M., G.M., G.S., M.W.L., M.B., C.L., M.C.-B., E.M.-M., S.C.-Y., M.G., P.A., I.J., J.H., N.G.F., H.F., M.M., C.D., A.K.H., E.K.A., and D.A. provided essential material; M.G.-L. and D.G.-F. performed statistical analysis; and E.A.-A., J.C., and M.G.-L. had primary responsibility for final content. All authors have read and agreed to the published version of the manuscript.

For information on how to submit an application for gaining access to EPIC data and/or biospecimens, please follow the instructions at http://epic.iarc.fr/access/index.php.