Skip to main content
NCBI home page
PLOS Medicine logoLink to PLOS Medicine
. 2025 Apr 8;22(4):e1004570. doi: 10.1371/journal.pmed.1004570

Food additive mixtures and type 2 diabetes incidence: Results from the NutriNet-Santé prospective cohort

Marie Payen de la Garanderie 1,2,*, Anaïs Hasenbohler 1,2, Nicolas Dechamp 1, Guillaume Javaux 1,2, Fabien Szabo de Edelenyi 1, Cédric Agaësse 1, Alexandre De Sa 1, Laurent Bourhis 1, Raphaël Porcher 3, Fabrice Pierre 2,4, Xavier Coumoul 2,5, Emmanuelle Kesse-Guyot 1,2, Benjamin Allès 1, Léopold K Fezeu 1, Emmanuel Cosson 1,6, Sopio Tatulashvili 1,6, Inge Huybrechts 7, Serge Hercberg 1,2,8, Mélanie Deschasaux-Tanguy 1,2, Benoit Chassaing 2,9, Héloïse Rytter 9, Bernard Srour 1,2, Mathilde Touvier 1,2,*
Editor: Fumiaki Imamura10
PMCID: PMC11977966  PMID: 40198579

Abstract

Background

Mixtures of food additives are daily consumed worldwide by billions of people. So far, safety assessments have been performed substance by substance due to lack of data on the effect of multiexposure to combinations of additives. Our objective was to identify most common food additive mixtures, and investigate their associations with type 2 diabetes incidence in a large prospective cohort.

Methods and Findings

Participants (n = 108,643, mean follow-up =  7.7 years (standard deviation (SD) =  4.6), age =  42.5 years (SD =  14.6), 79.2% women) were adults from the French NutriNet-Santé cohort (2009–2023). Dietary intakes were assessed using repeated 24h-dietary records, including industrial food brands. Exposure to food additives was evaluated through multiple food composition databases and laboratory assays. Mixtures were identified through nonnegative matrix factorization (NMF), and associations with type 2 diabetes incidence were assessed using Cox models adjusted for potential socio-demographic, anthropometric, lifestyle and dietary confounders. A total of 1,131 participants were diagnosed with type 2 diabetes. Two out of the five identified food additive mixtures were associated with higher type 2 diabetes incidence: the first mixture included modified starches, pectin, guar gum, carrageenan, polyphosphates, potassium sorbates, curcumin, and xanthan gum (hazard ratio (HR)per an increment of 1SD of the NMF mixture score = 1.08 [1.02, 1.15], p = 0.006), and the other mixture included citric acid, sodium citrates, phosphoric acid, sulphite ammonia caramel, acesulfame-K, aspartame, sucralose, arabic gum, malic acid, carnauba wax, paprika extract, anthocyanins, guar gum, and pectin (HR = 1.13 [1.08,1.18], p < 0.001). No association was detected for the three remaining mixtures: HR =  0.98 [0.91, 1.06], p = 0.67; HR =  1.02 [0.94, 1.10], p = 0.68; and HR =  0.99 [0.92, 1.07], p = 0.78. Several synergistic and antagonist interactions between food additives were detected in exploratory analyses. Residual confounding as well as exposure or outcome misclassifications cannot be entirely ruled out and causality cannot be established based on this single observational study.

Conclusions

This study revealed positive associations between exposure to two widely consumed food additive mixtures and higher type 2 diabetes incidence. Further experimental research is needed to depict underlying mechanisms, including potential synergistic/antagonist effects. These findings suggest that a combination of food additives may be of interest to consider in safety assessments, and they support public health recommendations to limit nonessential additives.

Trial Registration

The NutriNet-Santé cohort is registered at clinicaltrials.gov (NCT03335644). https://clinicaltrials.gov/study/NCT03335644.

Author summary

Why was this study done?

  • Several experimental and epidemiological studies have suggested potential deleterious effects of some food additives widely used by the food industry to enhance the texture, shelf life, taste, and appearance of foods.

  • So far, research and safety evaluation of food additives has been conducted on a substance-by-substance basis, while in real-life, food additive mixtures are consumed by billions of people globally.

  • Some experimental studies have raised concerns about potential interactions between additives within mixtures and their potential impact on health, but human epidemiological data are lacking.

What did the researchers do and find?

  • In this large cohort of 108,643 French adults, five frequently consumed food additive mixtures were identified.

  • Two of them were associated with higher type 2 diabetes incidence, independently of the nutritional quality of the diet, and after adjustment for a wide range of potential confounders: the first mixture primarily consisted of emulsifiers, preservatives, and a dye, while the second mixture was characterized by acidifiers, acid regulators, dyes, artificial sweeteners, and emulsifiers.

  • Exploratory analyses suggested both synergistic and antagonist interactions between several food additives emblematic of these mixtures.

What do these findings mean?

  • To our knowledge, this study is the first to estimate the exposure to food additive mixtures in a large population-based cohort and investigate their link with type 2 diabetes incidence. These results suggest that food additives found in a wide variety of products and frequently consumed together may potentially represent a modifiable risk factor for type 2 diabetes prevention. They support public health recommendations to limit nonessential additives.

  • The potential synergisms and antagonisms may be of interest in future mechanistic investigations, to better understand the relative influence of individual additives and their interactions in the observed associations.

  • Main limitations include possible exposure and outcome measurement errors and the fact that causality cannot be established on the basis of this observational study alone.

Using data from the NutriNet-Santé cohort study, Marie Payen de la Garanderie and colleagues investigate potential interactions of food additives in five commonly consumed food additive mixtures and their association with type 2 diabetes incidence.

Introduction

Ultra-processed foods (UPF) are endemic in Western diets and represent between 15%–20% (e.g., in Columbia, Romania) to almost 60% (in the United States) of daily energy intake [1]. Mounting evidence from epidemiological and experimental studies suggests a deleterious impact of UPF on many health outcomes, in particular metabolic-related diseases [1,2]. Beyond their poorer nutritional quality on average, one of the hypotheses to explain the health effects of UPF is the large use of food additives by the industry [1].

The World Health Organization defines food additives as substances primarily added to foods on an industrial scale, for technical purposes (e.g., lengthened product’s shelf-life, improve texture, taste, color, and palatability) [3]. In Europe, > 300 food additives are authorized (e.g., emulsifiers, artificial sweeteners, colors, preservatives) and their use in food manufacturing is governed by European regulation EC/1333/2008. Their safety has previously been assessed by the European Food Safety Authority (EFSA), which proposed acceptable daily intake for some of them.

However, these evaluations were constrained by the available scientific evidence at the time, which was limited due to a lack of human data and a predominant focus on specific toxicological targets such as cytotoxicity and genotoxicity. Recent in vitro/in vivo experimental studies now suggest deleterious effects of some food additives on a wider spectrum of health outcomes, including metabolic disorders, chronic inflammation, and gut microbiota disruption leading to intestinal inflammation [4]. Moreover, the NutriNet-Santé cohort study, which collected unique detailed dietary exposure data, including commercial names and brands of industrial products, provided new human data insights, suggesting associations between dietary exposure to widely consumed food additives (e.g., some artificial sweeteners and emulsifiers) and higher incidence of several chronic diseases, in particular type 2 diabetes [5,6].

Another important gap to date has been that previous evaluations have not been able to account for potential interaction/synergistic effects when assessing the safety of additives due to a lack of data. Single UPF often contain mixtures of additives [7]. Moreover, diets rich in UPF lead to the consumption of food combinations that result in the ingestion of mixtures of food additives. These additives may interact through synergistic or antagonistic effects, potentially influencing metabolism and overall health [8,9]. In a recent in vitro study based on four human cell models, we observed toxicological effects food additive mixtures, beyond the effect of these substances alone [10].

To our knowledge, this study is the first to aim at identifying the main mixtures of food additives and studying their associations with type 2 diabetes incidence using the large prospective NutriNet-Santé cohort. An exploratory, secondary aim was to examine interactions between food additives in mixtures associated with type 2 diabetes incidence, to explore potential synergisms and/or antagonisms.

Methods

Study population

This study was conducted within the population-based NutriNet-Santé prospective e-cohort. This French study was launched on May 11th 2009 with an ongoing open enrollment of volunteers. Its main objective is to investigate the relationships between nutrition and health [11]. Participants are recruited through vast multimedia campaigns from the general population of French citizens aged ≥15 years with internet access. To enroll, they are required to create a personal account on the NutriNet-Santé web-based platform (https://etude-nutrinet-sante.fr/). All participants enrolled up to December 31st, 2023 were included in this study. Upon enrollment, participants are invited to provide detailed information by completing five questionnaires about their lifestyle and socio-demographic data (e.g., date of birth, sex, educational level, professional occupation, smoking status, number of children), health status (e.g., personal and family medical history, medical treatments), dietary habits (three nonconsecutive 24-h dietary records [12]), anthropometric data (e.g., height, weight), and physical activity level (7-day assessment via the International Physical Activity Questionnaire [IPAQ]) [13]. Ethnicity and religion were not recorded since these variables are considered sensitive data by the French law, which strictly regulate their collection in population-based studies.

Ethical approval

NutriNet-Santé is registered at ClinicalTrials.gov (NCT03335644), conducted according to the Declaration of Helsinki guidelines, and approved by the Institutional Review Board of the French Institute for Health and Medical Research (IRB-Inserm) and the “Commission Nationale de l’Informatique et des Libertés” (CNIL n°908450/n°909216). Each participant provides an electronic informed consent prior to enrollment.

Dietary data collection

At inclusion, and every 6 months thereafter, participants were invited to fill out three nonconsecutive days of 24-h dietary records, randomly assigned over a 2-week period, including 2 weekdays and 1 weekend day (to account for variability in the diet across the week and the seasons[14–16]. Details on the dietary data collection and energy under-reporter’s (i.e., participants who systematically reported implausibly low energy intakes) identification are provided in eMethod1 in S1 Appendix. We calculated daily dietary intakes for food additives, nutrients, energy and food groups as the mean intake from all 24-h dietary records available for each participant during their first two years of follow-up. The NOVA classification was applied to identify UPF and calculate their contribution to energy intake [17].

Food additive intakes

Intakes of food additives were quantified based on data provided by the participants’ dietary records, in which the commercial brand/name of the industrial products consumed were recorded. The presence/absence of each specific additive in each specific food was determined by a dynamic matching with several databases, considering the date of consumption (to account for reformulations across time). Multiple sources were used to retrieve food additive doses (including ad hoc laboratory assays and EFSA data). The detailed method for estimating food additive intakes was previously described [8]; more information is provided in eMethod2 in S1 Appendix. Table A in S1 Appendix displays the list of the 269 food additives ingested by the participants with corresponding EU codes. In order to obtain a reliable estimate of food additive exposure and to focus on those most likely to have substantial public health impact, only those consumed by at least 5% of the cohort were included in the mixture modeling.

Type 2 diabetes ascertainment

Type 2 diabetes was assessed using a multisource approach. Throughout the follow-up period, participants were invited to report any health events, medical treatments, and examinations via the biannual health questionnaires or at any time, directly via the health interface of their personal account. Furthermore, the NutriNet-Santé cohort was linked to the national health insurance system database in order to obtain additional information regarding medical treatments and consultations. Linkage to the French National Mortality Registry (CépiDC) enabled the identification of the occurrence and cause of death. Further details can be found in eMethod3 in S1 Appendix.

Statistical analyses

Among participants from the NutriNet-Santé cohort who completed at least two 24-h dietary records during their first 2 years of follow-up, we included those who were not under-energy reporters and who did not have any prevalent type 1 or 2 diabetes diagnosed before their enrollment in the cohort. Participants with a null food additive consumption (n = 80, 0.07%) were excluded to perform mixture analyses (flowchart of participants presented in Fig A in S1 Appendix). Food additive mixtures were identified using nonnegative matrix factorization (NMF). This size reduction technique was specifically adapted to sparse matrices containing positive values [18]. The Lee algorithm was selected for its ability to balance residual minimization and sparseness [19], while ensuring nonnegativity and effectively handling noise. The number of ranks (i.e., the number of NMF components to retain) was determined according to the method proposed by Brunet and colleagues [20], using the smallest number for which the cophenetic coefficient starts decreasing, as visualized on the consensus map (Table C.a. in S1 Appendix). The NMF was performed using the R package NMF [21] and the scores arising from the components (= the food additive mixtures) were scaled. Food additives with loading values ≥ │0.15│were considered as the most emblematic of each mixture. This point was only for description purposes, since all factor loadings are displayed in the result tables and all additives contributed to the mixture score calculation. More details on NMF method are provided in eMethod4 in S1 Appendix. Sensitivity analyses using different decomposition algorithms are shown in Table C.b. and c. We checked the stability of food additive mixture intakes over time by performing NMF analysis on two periods of 7.5 years each (corresponding to the median follow-up: period 1 = 2009–2016; period 2 = 2017–2024, in S1 Appendix, Table C.d.). Spearman correlations coefficients were computed to assess the links between NMF components with each other (Table D in S1 Appendix), and between NMF components and food group intakes (Table E in S1 Appendix). Detailed consumption of food groups by sex-specific quintiles of NMF scores for the mixtures associated with higher type 2 diabetes incidence are provided in Table F in S1 Appendix (added to the initial statistical analysis plan, in response to peer review comments).

The associations between the exposure to food additive mixtures (as reflected by their continuous NMF scores) and higher type 2 diabetes incidence were assessed using multivariable proportional hazard Cox models with age as the time scale. Participants contributed person-time to the models from their age at enrollment in the cohort until their age at the date of type 2 diabetes diagnosis, the date of type 1 diabetes diagnosis, the date of death, the date of last contact, or December 31st 2023, whichever occurred first. Since death and incident type 1 diabetes occurring during follow-up were considered as competing risks for type 2 diabetes, cause-specific Cox models were used. The proportional hazard assumption was checked by examining Schoenfeld residuals and the Grambsch and Therneau’s lack-of-fit test (Fig B in S1 Appendix), and the log-linearity of the associations was assessed using restricted cubic splines with three knots at the 27.5th, 50th, and 72.5th percentiles of each mixture score’s distribution [22]. Hazard ratios (HRs) and 95% confidence intervals (95% CIs) were computed for a standardized increment of one standard deviation (SD) of each mixture score. Increments are specified in the legend of Fig 1.

Fig 1. Associations between food additive mixtures and type 2 diabetes incidence, NutriNet-Santé cohort, 2009–2023 (n.

Fig 1

Open in a new tab

 = 108,643 participants; 1,131 incident cases). Abbreviations: HR, hazard ratio; CI, confidence interval. Mixtures of food additives were derived from nonnegative matrix factorization (NMF, eMethod4 in S1 Appendix). HRs were computed for increments of 1 standard deviation (SD) of each mixture score: mixture 1 (SD = 12.9), mixture 2 (SD = 8.0), mixture 3 (SD = 88.3), mixture 4 (SD = 20.1), mixture 5 (SD = 14.7). The identified mixtures correspond to food additive combinations derived from the nonnegative matrix factorization (NMF) procedure. Mixture 1 was mainly characterized by sodium carbonates, diphosphates, glycerol, ammonium carbonates, potassium carbonates, and sorbitol. Mixture 2 was mainly characterized by modified starches, pectin, guar gum, carrageenan, polyphosphates, potassium sorbates, curcumin, and xanthan gum. The additives that contributed most to mixture 3 were magnesium carbonates, riboflavin, alpha-tocopherol, and ammonium carbonates. Main contributors to mixture 4 were ammonium carbonates, sodium carbonates, diphosphates, alpha-tocopherol, mono and diacetyl tartaric acid esters of mono and diglycerides of fatty acids, magnesium carbonates, and lecithins. Finally, the main food additives characterizing mixture 5 were citric acid, sodium citrates, phosphoric acid, sulphite ammonia caramel, acesulfame-K, aspartame, sucralose, arabic gum, malic acid, carnauba wax, paprika extract, anthocyanins, guar gum, and pectin. Multivariable Cox proportional hazard models were adjusted for age (time-scale), sex, Body Mass Index (BMI, continuous, kg/m2), physical activity (categorical International Physical Activity Questionnaire (IPAQ) variable: high, moderate, low), smoking status (never smoked, former smoker, current smokers), number of smoked cigarettes in pack-years (continuous), educational level (did not complete secondary education/ up to two years of university studies/ bachelor degree or higher), family history of type 2 diabetes (yes/no), number of dietary records (continuous), socio-professional categories (farmer, craftsman/shopkeeper/entrepreneur, managerial staff/intellectual profession, intermediate profession, employee, manual worker, retired, unemployed, student, and other without professional activity), monthly household income per consumption unit (<1,200 €/month; 1,200–1,800 €/month; 1,800–2,700 €/month; >2,700 €/month), intakes of energy without alcohol (continuous, kcal/d), saturated fatty acids (continuous, g/d), sodium (continuous, mg/d), dietary fiber (continuous, g/d), alcohol (continuous, g/d), added sugars (continuous, g/d).

The main model was adjusted for a set of predefined risk factors of type 2 diabetes, i.e.,: age (time-scale), sex, Body Mass Index (BMI, continuous, kg/m2), physical activity (categorical IPAQ variable: high, moderate, low), smoking status (never smoked, former smoker, current smokers), number of smoked cigarettes in pack-years (continuous), educational level (did not complete secondary education/up to 2 years of university studies/bachelor degree or higher), socio-professional categories (farmer, craftsman/shopkeeper/entrepreneur, managerial staff/intellectual profession, intermediate profession, employee, manual worker, retired, unemployed, student, and other without professional activity—added following peer review comments), monthly income per household unit (<1,200 €/month; 1,200–1800 €/month; 1800–2,700 €/month; > 2,700 €/month—added following peer review comments), family history of type 2 diabetes (yes/no), number of dietary records (continuous), intakes of energy without alcohol (continuous, kcal/d), saturated fatty acids (continuous, g/d), sodium (continuous, mg/d), dietary fiber (continuous, g/d), alcohol (continuous, g/d), and added sugars (continuous, g/d). Sensitivity analyses are presented in eMethods4 and Table G in S1 Appendix, including a model further adjusted for an indicator of health-seeking behaviors and geographical region (coding available in S1 Appendix in footnotes to Table G, Model 6 and 7– added following peer review comments). Exposure coded as tertiles was also tested (Table H in S1 Appendix).

Following peer review comments, we also performed the following set of sensitivity analyses for mixtures 2 and 5: we computed the main model, stratified by an indicator of the nutritional quality of the diet, i.e., the Programme National Nutrition Santé-Guidelines Score 2 (PNNS-G2, below/above the sex-specific median, Table I) We also explored the hypothesis that beyond associations of individual additives with type 2 diabetes incidence, there may be an influences of the mixtures themselves. For this: (1) we adjusted the main Cox model for each food additive characteristic of the mixture, using the residual method (Table J in S1 Appendix); (2) two-by-two interactions between the key food additives contributing to each mixture were formally tested with food additive intakes standardized to mean = 0 and SD = 1 for interpretability (Table K in S1 Appendix). Last, we tested whether the mixtures found associated with type 2 diabetes incidence contributed to mediate the associations between the food groups most associated with these mixtures and incidence of type 2 diabetes risk, using the CMAVERSE R package (Table L in S1 Appendix). All statistical tests were two-sided, and p-values < 0.05 were considered statistically significant, except for interaction tests (less powerful), for which 0.1 was considered statistically significant. Analyses were conducted in R version 4.3.3, except for the Restricted Cubic Splines, which were implemented in SAS version 9.4. This study is reported as per the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) guideline (S1 Checklist in STROBE Checklist).

Results

Descriptive characteristics

A total of 108,643 participants from the NutriNet-Santé cohort were included in this study (Fig A in S1 Appendix), among which 79.2% were women. At baseline, the median age of the cohort was 41.2 years (25th–75th percentiles: 29.8–54.5 years). Among the overall cohort, 3.82% (n = 4,150) participants have died since their inclusion (2,411 in the present population study) and 9.5% dropped out because they did not want to receive any more questionnaires. Participants included in this study completed a median of 5 dietary records (25th–75th percentiles: 3–9). Their characteristics are detailed in Table 1. UPF (NOVA 4) accounted for a median of 33.8% (25th–75th percentiles: 25.2%−43.7%) of daily energy intake. Daily food additive intakes are described in Table A in S1 Appendix (mean, SD, median, percentage of consumers). A total of 75 food additives were consumed by at least 5% of the participants and were therefore included in NMF mixture analyses.

Table 1. Baseline characteristics of study participants, NutriNet-Santé cohort, 2009–2023 (N = 108,643).

Mean (SD) or N (%) Median [25th–75th percentiles]
Age (years) 42.5 (14.6) 41.2 [29.8, 54.5]
Sex, Female 86032 (79.2%) ..
BMI (kg/m2)* 23.6 (4.4) 22.7 [20.6, 25.5]
Family history of type 2 diabetesa,* 16994 (15.6%) ..
IPAQ physical activity level*
 High 30928 (28.5%) ..
 Moderate 40415 (37.2%) ..
 Low 22535 (20.7%%) ..
Smoking status*
 Never 54613 (50.3%) ..
 Former smoker 35205 (32.4%) ..
 Current 18502 (17.0%) ..
Educational level*
 Did not complete secondary education 18997 (17.5%) ..
 Up to two years of university studies 52740 (48.5%) ..
 Bachelor degree or higher 35863 (33.0%) ..
Socio-professional categories
 Farmer 333 (0.3%) ..
 Craftsman/shopkeeper/entrepreneur 1975 (1.8%) ..
 Managerial staff/intellectual profession 25527 (23.5%) ..
 Intermediate profession 18321 (16.9%) ..
 Employee 20158 (18.5%) ..
 Manual worker 1397 (1.3%) ..
 Retired 17473 (16.1%) ..
 Unemployed 6337 (5.8%) ..
 Student 6756 (6.2%) ..
 Other without professional activity 9956 (9.2%) ..
Monthly household income per consumption unit (euros)
 <1200 18878 (17.4%) ..
 1200–1800 26327 (24.2%) ..
 1800–2700 24900 (23.0%) ..
 >2700 25365 (23.3%) ..
Geographical region
North 11517 (10.6%) ..
North-East 12805 (11.8%) ..
West 13226 (12.2%) ..
Center 4104 (3.8%) ..
South-West 19660 (18.1%) ..
South-East 23026 (21.2%) ..
Ile-deFrance 20954 (19.3%) ..
French overseas territories and departments 1024 (0.9%) ..
Corsica 324 (0.3%) ..
Other 1992 (1.8%) ..
Energy intake without alcohol (kcal/day) 1840 (450) 1790 [1538, 2093]
Alcohol intake (g/day) 7.70 (11.7) 3.21 [0, 10.6]
Saturated fat intake (g/day) 33.2 (12.1) 31.9 [64.6, 95.5]
Sugar intake (g/day) 198 (57.4) 192 [159.6, 229.8]
Sodium (mg/day) 2720 (886) 2600 [2118, 3186]
Fiber intake (g/day) 19.6 (7.3) 18.5 [14.7, 23.2]
Fruits and vegetables intake (g/day) 464 (231) 440 [307.2, 590.1]
Red and processed meat intake (g/day) 75.6 (52.6) 68.8 [38.9, 103.6]
Total food additive intake (mg/day) 5730 (4400) 4770 [2791, 7510]
Open in a new tab

aIn first-degree relatives.

*

Missing values: BMI n =  3,000; Family history of type 2 diabetes n =  333; IPAQ physical activity level n =  14,765; Smoking status n =  323; Educational level n =  1,043; Socio-professional categories n =  410; Monthly household income per consumption unit n =  13,173, geographical region n = 11.

Abbreviations: SD,  standard deviation, N, number, IPAQ,  International Physical Activity Questionnaire, BMI,  body mass index.

Food additive mixtures derived by NMF

The NMF procedure identified five main food additive mixtures (Table B in S1 Appendix: full results). Table 2 summarizes the food additives that were the most emblematic of each mixture. Mixture 1 was mainly characterized by sodium carbonates, diphosphates, glycerol, ammonium carbonates, potassium carbonates, and sorbitol. Mixture 2 was characterized by modified starches, pectin, guar gum, carrageenan, polyphosphates, potassium sorbates, curcumin, and xanthan gum. The additives that contributed most to mixture 3 were magnesium carbonates, riboflavin, alpha-tocopherol, and ammonium carbonates. Main contributors to mixture 4 were ammonium carbonates, sodium carbonates, diphosphates, alpha-tocopherol, mono and diacetyl tartaric acid esters of mono and diglycerides of fatty acids, magnesium carbonates, and lecithins. Finally, the main food additives in mixture 5 were citric acid, sodium citrates, phosphoric acid, sulphite ammonia caramel, acesulfame-K, aspartame, sucralose, arabic gum, malic acid, carnauba wax, paprika extract, anthocyanins, guar gum, and pectin. Mean (SD) NMF scores of the participants for the five mixtures were as follows: mixture 1: 9.1 (12.9); mixture 2: 8.9 (8.0); mixture 3: 9.1 (88.3); mixture 4: 9.1 (20.1); mixture 5: 9.0 (14.7). NMF performed using different algorithms (Table C.b. and C.c. in S1 Appendix) and computed at the two periods (corresponding to the first and second halves of the cohort follow-up, Table C.d. in S1 Appendix) indicated overall robustness of the findings and stability of the mixtures over time.

Table 2. Food additive mixtures identified by nonnegative matrix factorization: loading values of main additive contributorsa, NutriNet-Santé cohort, 2009–2023b.

Mixture 1
Food additive/Loading value		 Mixture 2
Food additive/Loading value		 Mixture 3
Food additive/Loading value		 Mixture 4
Food additive/Loading value		 Mixture 5
Food additive/Loading value
E500 Sodium carbonates 0.99 Modified starches 0.99 E504 Magnesium carbonates 0.99 E503 Ammonium carbonates 0.99 E330 Citric acid 0.83
E450 Diphosphates 0.78 E440 Pectins 0.31 E101 Riboflavin 0.53 E500 Sodium carbonates 0.35 E331 Sodium citrates 0.63
E422 Glycerol 0.37 E412 Guar gum 0.26 E307 Alpha-tocopherol 0.24 E450 Diphosphates 0.30 E338 Phosphoric acid 0.59
E503 Ammonium carbonates 0.35 E407 Carrageenan 0.24 E503 Ammonium carbonates 0.17 E307 Alpha-tocopherol 0.28 E150d Sulphite ammonia caramel 0.59
E501 Potassium carbonates 0.17 E452 Polyphosphates 0.21 E472e DATEMc 0.18 E950 Acesulfame K 0.56
E420 Sorbitols 0.16 E202 Potassium sorbate 0.17 E504 Magnesium carbonates 0.17 E951 Aspartame 0.41
E100 Curcumin 0.16 E322 Lecithins 0.15 E955 Sucralose 0.25
E415 Xanthan gum 0.16 E414 Arabic gum 0.23
E296 Malic acid 0.19
E903 Carnauba wax 0.18
E160c Paprika extract, capsanthin, capsorubin 0.17
E163 Anthocyanins 0.15
E412 Guar gum 0.15
E440 Pectins 0.15
Open in a new tab

aThe loading values indicate the strength of the association between the specific food additive and each NMF mixture score.

bFor conciseness purposes, only food additives with NMF loading values ≥│0.15│ are displayed here. Full results are presented in Table C in S1 Appendix.

cMono and diacetyl tartaric acid esters of mono and diglycerides of fatty acid.

Overall, there was little correlation between the five mixtures, the highest being observed between mixtures 1 and 4: Spearman correlation coefficient ρ = 0.39 (Table D in S1 Appendix). Table E in S1 Appendix displays the correlations between additive mixtures and food group intakes. Detailed consumption data for specific food groups, stratified by sex-specific quintiles of participants, are additionally provided for mixtures 2 and 5 (Table F in S1 Appendix). Mixture 1 was correlated with cakes and biscuits (ρ = 0.35) as well as savory snacks (ρ = 0.18). The food groups most correlated with mixture 2 were broth (ρ = 0.40), dairy desserts (ρ = 0.22), and fats and sauces (ρ = 0.21). No specific food group as a whole correlated with mixture 3. Indeed, the additives contained in this mixture are used in foods that are frequently consumed, but distributed in an isolated way in multiple food groups (e.g., E504 magnesium carbonate in table-top salt, in certain brands of energy drinks, as well as in specific brands of chocolate cookies and cocoa powder, etc.). In line with the correlation between mixtures 1 and 4, mixture 4 was also correlated with savory snacks (ρ = 0.19) and cakes and biscuits (ρ = 0.18). The food groups most correlated with mixture 5 were artificially sweetened soft drinks (ρ = 0.41) and sugary drinks (ρ = 0.37).

Associations between food additive mixtures and type 2 diabetes incidence

A total of 1,131 incident type 2 diabetes were detected (mean follow-up duration = 7.7 years (SD 4.6)). Schoenfeld residuals (Fig B in S1 Appendix) did not show evidence for violation of the proportional hazard assumptions. Associations between food additive mixtures and type 2 diabetes incidence are outlined in Fig A in S1 Appendix. Mixture 2 (HRper an increment of 1SD = 1.08 [1.02,1.15], p = 0.006) and mixture 5 (HRper increment of 1SD = 1.13 [1.08,1.18], p < 0.001) were positively associated with higher type 2 diabetes incidence. No association with type 2 diabetes incidence was observed for mixtures 1, 3, and 4. Sensitivity analyses fully aligned with results from the main model, supporting the robustness of the findings (Table G in S1 Appendix). These models tested several further or modified adjustments (for prevalent metabolic disorders; for other mixtures - mutual adjustment; for Healthy and Western dietary patterns; for food groups instead of nutrients; for health-seeking behaviors; and for geographical region), and exclusion of cases diagnosed in the first 2 years of follow-up to challenge reverse causality. Restricted cubic spline plots confirmed the linearity of the observed associations for all mixtures except mixture 3 (Fig 2 for mixtures 2 and 5 and Fig C for all in S1 Appendix). Categorical analyses (tertiles) were also conducted and showed similar results (Table H in S1 Appendix).

Fig 2. Dose-response associations between food additive mixtures 2 and 5 and type 2 diabetes incidence, restricted cubic spline plots, NutriNet-Santé cohort, 2009-2023 (n.

Fig 2

Open in a new tab

 = 108,643 participants; 1,131 incident cases). Abbreviations: CL confidence limit. Mixture 2: P-value for nonlinearity = 0.4; Mixture 5: P-value for nonlinearity = 0.05. Mixture 2 was mainly characterized by modified starches, pectin, guar gum, carrageenan, polyphosphates, potassium sorbates, curcumin, and xanthan gum. Mixture 5 was mainly characterized by citric acid, sodium citrates, phosphoric acid, sulphite ammonia caramel, acesulfame-K, aspartame, sucralose, arabic gum, malic acid, carnauba wax, paprika extract, anthocyanins, guar gum, and pectin.

For mixtures 2 and 5: very similar associations were observed in both PNNS-GS2 strata (although borderline nonsignificant for mixture 2, likely due to halved statistical power) and no interaction was detected between the PNNS-GS2 score and the additive mixtures (p for interaction =  0.8 for mixture 2 and 0.9 for mixture 5), supporting an association between these mixtures and type 2 diabetes incidence, independent from the nutritional quality of the diet (Table I in S1 Appendix). Despite slight attenuations, the associations between mixtures 2 and 5 and type 2 diabetes incidence remained similar after adjustment for each food additive characteristic of the mixture using the residual method, suggesting that the associations were not strongly driven by a unique additive alone (Table J in S1 Appendix). Out of the 28 two-by-two interactions tested between the 8 additives most emblematic of mixture 2, 3 were detected (i.e., p for interaction < 0.1) with beta coefficient >0 (suggesting synergism), and 4 were detected with beta coefficient <0 (suggesting antagonism). For mixture 5, 91 interactions were tested between the 14 most emblematic additives: 6 were detected with beta coefficient > 0 and 4 with beta coefficient < 0 (Table K in S1 Appendix). Lastly, mediation analyses were conducted (Table L in S1 Appendix). Among food groups most correlated with additives mixtures, fats and sauces (correlated with mixture 2), artificially sweetened beverages, and sugary drinks (both correlated with mixture 5) were associated with higher type 2 diabetes incidence (Table L in S1 Appendix, Cox models). The mediation of the association between fats and sauces and type 2 diabetes by mixture 2 was modest (proportion of the association mediated = 18%, p-value = 0.09). Mixture 5 mediated 42% of the association between sugary drinks and type 2 diabetes (p <  0.001) and 52% of the association between artificially sweetened beverages and type 2 diabetes (p =  0.03).

Discussion

This study identified a positive association between two broadly ingested food additive mixtures and a higher incidence of type 2 diabetes. One of these mixtures (mixture 2) was primarily composed of several emulsifiers (modified starches; pectin; guar gum; carrageenan; polyphosphates; xanthan gum), in addition to a preservative (potassium sorbate), and a dye (curcumin), which are typically found in a variety of industrially-processed foods, including broth, dairy desserts, fats and sauces. The other mixture associated with type 2 diabetes incidence (mixture 5) was primarily composed of food additives found in artificially sweetened beverages and sugary drinks. These additives included acidifiers and acid regulators (citric acid; sodium citrates; phosphoric acid; malic acid), dyes (sulphite ammonia caramel, which is characteristic of cola sodas; anthocyanins; paprika extract), artificial sweeteners (acesulfame-K; aspartame; sucralose), and some emulsifiers (arabic gum; pectin; guar gum). Exploratory analyses suggested both synergistic and antagonist interactions between several food additives emblematic of these mixtures. In order to account for the nutritional quality of the diet and isolate as much as possible potential effects from food additives, all models were systematically adjusted for key nutritional intakes and no interaction was observed with an overall indicator of the nutritional quality of the diet. The associations between additive mixtures and type 2 diabetes incidence were therefore studied ‘all other things being equal’ in terms of intakes of sugar, saturated fat, energy, alcohol, and so forth.

To our knowledge, this study is the first to evaluate and detect positive associations between food additive mixtures and higher type 2 diabetes incidence in a large prospective cohort. Thus, direct comparison of our findings with previous epidemiological literature is not possible. However, our results can be put into perspective with those of previous epidemiological studies that have examined associations between single food additives and the incidence of type 2 diabetes, which, to our knowledge, are only available in the NutriNet-Santé cohort to date. Consistently, several of the food additives emblematic of mixtures 2 or 5 were associated with higher type 2 diabetes incidence in previous publications on emulsifiers and artificial sweeteners [5,6] (preservatives and dyes are currently under investigation): carrageenans (HRper increment of 100 mg per day = 1.03 [1.01,1.05], p < 0.0001), sodium citrate (HRper increment of 500 mg per day = 1.04 [1.01,1.07], p = 0.0080), guar gum (HRper increment of 500 mg per day = 1.11 [1.06,1.17], p < 0.0001), gum arabic (HRper increment of 1,000 mg per day = 1.03 [1.01,1.05], p = 0.013), xanthan gum (HRper increment of 500 mg per day = 1.08 [1.02,1.14], p = 0.013), aspartame (HRfor an increment of 100 mg/day = 1.26 (1.08,1.46), p = 0.003), acesulfame-K (HRfor an increment of 100 mg/day = 1.62 (1.12,2.33), p = 0.010).

In these previous papers published on food additives and type 2 diabetes incidence in NutriNet-Santé [5,6,23], the objective was to investigate one specific food additive (or food additive category), independently from the intake of other additives. Therefore, the models (main or sensitivity) were adjusted for the proportion of UPF in the diet (as an overall indicator of multiple additive exposure), and/or for other additives than the one studied (e.g., intake of all emulsifiers except the one studied). These analyses suggested potential effects of these additives (some artificial sweeteners, emulsifiers, nitrites) per se, consistent with in vivo/in vitro experimental/mechanistic data—see below [24–26]. In the latter, the effect of each specific additive is isolated and tested, thus, results cannot be attributed to a potential mixture effect. The present study is complementary. Our hypothesis was that beyond potential influence of individual additives, the mixtures themselves may play a role, resulting from interactions between food additives. Our results suggest that the associations between mixtures 2 and 5 and type 2 diabetes incidence were not entirely driven by any of the specific additives alone. Besides, several interactions have been detected between emblematic additives of mixtures 2 and 5. However, the number of detected interactions was limited compared to the overall number tested. Moreover, both synergistic, but also antagonist interactions were observed. The relative influence of individual additives versus their interactions should be explored in future mechanistic investigations. To our knowledge, only one human study previously explored the potential health impact of a food additive mixture [27]: a randomized control trial conducted in the UK observed that a 6-week exposure to a pre-defined dye mixture (i.e., carmoisine [E122], sunset yellow [E110], tartrazine [E102], ponceau 4R [E124], allura red AC [E129]; 20 or 30 mg for the 3-year-old children and 24.98 or 62.4 mg for 8/9-year-old children) and 45 mg of sodium benzoate preservative [E211] increased hyperactivity among 3-year-old and 8/9-year-old children.

Our results are supported by several experimental in vivo and in vitro studies that suggest deleterious effects for several food additives emblematic of mixtures 2 and 5 and thus could explain the associations with higher type 2 diabetes incidence found in this study. For example, guar gum [E412], present both in mixtures 2 and 5, has been pointed out for its alteration of gut microbiota composition leading to an elevation of pro-inflammatory markers and potential metabolic perturbations in a mouse model [28]. Evidence of the role of the gut microbiota’s contribution in type 2 diabetes mellitus is growing, especially through the alteration of glucose metabolism pathways [29]. Carrageenan [E407], an emulsifier predominant in mixture 2, was observed to impair glucose metabolism in mice and showed inflammatory properties [30,31], which may be involved in type 2 diabetes etiology. The role of artificial sweeteners such as the ones found in mixture 5 on gut microbiota perturbations has also been suggested [32]. In particular, acesulfame-K [E950] and sucralose [E955] were observed to shape the microbial populations and lead to dysbiosis [33,34], which in turn may enhance glucose intolerance and changes in host physiology in mice. This was supported by the observation of similarities in microbial populations from noncaloric artificial sweeteners consumers and patients with type 2 diabetes [35]. Besides, some experimental results suggest that different additives may interact and thus lead to synergistic or antagonist effects [27,36]. In particular, an in vitro study assessed the neurotoxicity of two food additive ‘mixtures (i.e., Brilliant Blue [E133] and l-glutamic acid [E621], or Quinoline Yellow [E104] and aspartame [E951]) and reported that their synergy was potentially more toxic than the effect of the individual compounds [37]. Behavior and brain changes have been reported in rats when exposed to food-colorant mixtures [38]. In an in vitro experiment, Meng and colleagues observed cytotoxic effects caused by multiple food additives (Sodium benzoate, Potassium sorbate, New red, Caffeine, Sodium saccharin, Acesulfame K, Aspartame, and Tartrazine, Sunset Yellow, Erythrosine, Amaranth, Ponceau 4R, Brilliant blue FCF, Allura red AC) at doses below the tolerable upper levels in beverages [9]. In a recent paper based on four human cell models, we observed cyto-/genotoxic effects of some additive mixtures beyond the ones observed for the substances alone [10]. Further experimental studies are needed to gain a deeper understanding of potential interaction (synergistic and antagonist) effects of food additive mixtures on metabolism and diabetes etiology.

The strengths of this study lie in its prospective design, large sample size, meticulous assessment of dietary intakes, and unique data on the exposure to a broad range of food additives. The NutriNet-Santé study stands out as the first, to our knowledge, to evaluate both qualitative and quantitative exposures to food additives, leveraging detailed brand-specific and repeated 24-h dietary records, links to multiple food composition databases (Observatoire de la Qualité de l’Alimentation [OQALI], Open Food Facts, Global New Products Database [GNPD], EFSA, and General Standard for Food Additives [GSFA]), ad-hoc laboratory assays, and dynamic matching to account for reformulations of industrial food items over time [8]. The studied mixtures are clinically relevant because they represent the ones to which consumers are most frequently exposed daily. Indeed, the approach was to derive NMF components based on the observation of real-life consumption data in this large-scale population study. Besides, associations remained stable across various sensitivity analyses and are supported by mechanistic plausibility.

Several limitations should be acknowledged. First, the observational nature of the design introduces inherent constraints and a single observational epidemiological study is not sufficient per se to establish causality. Despite extensive adjustments for confounding variables, including dietary, lifestyle, anthropometric, and socio-demographic factors, the potential for unmeasured and residual confounding persists. Race/ethnicity and religion were not available in the cohort due to a very restrictive ethical/legal regulation policy regarding the collection of these data in French epidemiological studies (specific authorizations needed). Second, exposure to food additives has not been validated against blood or urine assays due to lack of existing specific biomarkers so far. Classification errors on exposure (e.g., in products exempt from labeling requirements) or covariates could not be entirely ruled out. However, numerous methodological studies in e-epidemiology were conducted and published, to challenge the reliability of the information collected online in this cohort. These included comparisons with traditional data collection methods (e.g., paper-and-pencil or interviews by trained dietitians) and against gold standards, such as blood and urine biomarkers of nutritional intakes and investigator-measured height and weight [12,14–16,39,40]. High consistency was observed between online tool and standard methods, supporting the reliability of the data collected. The methodological studies even highlighted the advantage of online web questionnaires in increasing the quality of collected data and reducing input error and outliers thanks to integrated field controls and automated conditional skips. Several studies also showed that web-based data collection may even increase accuracy by lowering social desirability bias associated with face-to-face interviews [41–43]. It is merely impossible to guaranty a perfect measurement of exposure and covariates in any population-based epidemiological study in real-life setting. However, in etiological studies, the ability to rank participants with contrasted exposures between themselves is more important than reaching perfectly accurate absolute values. Besides, misclassification in exposure or covariates may have induced a nondifferential measurement error (identically in future cases and noncases given the prospective design), but although an overestimation cannot be excluded, it most probably led to an underestimation of the observed associations. Third, several food additives were ingested by a very low number of individuals and thus, could not be included in NMF mixture analyses. In addition, food additive intake may vary over time. Yet, NMF analyses performed at the two time periods showed an overall high stability of the five main mixtures across follow-up. Next, despite the multisource case ascertainment strategy (which combined self-reported disease and medication use, linkage to a national health insurance database, and fasting blood glucose measurements in a subsample), under-detection of certain undiagnosed type 2 diabetes cases could not be entirely ruled out. In France, the prevalence of undiagnosed diabetes cases is estimated around 1.7% (IC95% 1.1–2.4), higher in men (2.7% IC95% 1.4–4.0) than in women (0.9% IC95% 0.3–1.4) [44]. Last, the generalizability of our findings may be influenced by the cohort’s demographic characteristics, such as a higher proportion of women and a more health-conscious population. Therefore, caution is warranted when extrapolating our results to broader populations.

In conclusion, the results of this large population-based study revealed positive associations between widely consumed food additive mixtures and a higher incidence of type 2 diabetes. To our knowledge, these findings provide the first insight into the food additives that are frequently ingested together [due to their co-occurrence in industrially-processed food products or resulting from the co-ingestion of foods in dietary patterns) and how these additive mixtures may be involved in type 2 diabetes etiology. Further long-term observational epidemiological studies, as well as short-term interventions and pre-clinical experimental research are required to elucidate the underlying mechanisms and gain a deeper understanding of potential synergies and antagonisms between these food chemicals. These results suggest that it may be of interest to consider potential interaction/synergistic/antagonist effects when assessing the safety of food additives and call for a reevaluation of regulations governing their use by the food industry, with the aim of enhancing consumer protection. In the meantime, these findings provide support for the public health recommendation to limit exposure to UPF and their nonessential food additives.

Supporting information

S1 Checklist. STROBE checklist.

(DOC)

pmed.1004570.s001.doc (94.5KB, doc)
S1 Appendix. eMethods. eResults.

Fig A. Flowchart, NutriNet-Santé cohort, 2009-2023. Table A. Daily food additive intakes (mg/d) among study participants from the NutriNet-Santé cohort, 2009–2023 (N = 108,643)a,b. Table B. Food additive mixtures identified by nonnegative matrix factorizationa. Table C. Consensus map for rank number determination and sensitivity analyses using other decomposition algorithms in the NMF procedure. Table D. Spearman correlations between the five NMF food additive mixtures. Table E. Spearman correlations between NMF food additive mixtures and food group intakes. Table F. Food group consumption of participants according to sex-specific quintiles of mixtures 2 and 5a. Fig B. Correlations between Schoenfeld residuals and timescale (age, y) from multivariable Cox models between food additive mixtures and type 2 diabetes incidence, NutriNet-Santé cohort, 2009–2023 (n = 108,643). Fig C. Dose-response associations between food additive mixtures and type 2 diabetes incidence, restricted cubic spline plots, NutriNet-Santé cohort, 2009–2023 (n = 108,643 participants; 1,131 incident cases). Table G. Associations between food additive mixtures and type 2 diabetes incidence, NutriNet-Santé cohort, 2009–2023—Sensitivity analyses. Table H. Association between food additive mixtures coded as tertiles and incidence of type 2 diabetes, NutriNet-Santé cohort, 2009–2023—Sensitivity analyses. Table I. Associations between food additive mixtures and incidence of type 2 diabetes, stratified by the Programme National Nutrition Santé-Guidelines Score 2 (PNNS-GS2), NutriNet-Santé cohort, 2009-2023—Sensitivity analyses. Table J. Association between mixtures 2 and 5 and type 2 diabetes incidence adjusted for the key food additives contributing to each mixture (residual method), NutriNet-Santé cohort, 2009–2023 (n = 108,643 participants; 1,131 incident cases) —Sensitivity analyses. Table K. Interactions between the key food additives contributing to mixtures 2 and 5, NutriNet-Santé cohort, 2009–2023 (n = 108,643 participants; 1,131 incident cases). Table L. Mediation analyses.

(DOCX)

pmed.1004570.s002.docx (3MB, docx)

Acknowledgments

We thank Thi Hong Van Duong, Régis Gatibelza, Jagatjit Mohinder, and Aladi Timera (computer scientists), Selim Aloui (IT manager); Julien Allègre, Nathalie Arnault (data-manager/statisticians); Paola Yvroud (health event validator); Maria Gomes and Mirette Foham (participant support); Marine Ricau (Operational coordinator); Marie Ajanohun, Souad Hadji (administration and finance), and Nadia Khemache (Administrative manager), for their technical contribution to the NutriNet-Santé study. We also warmly thank all the volunteers of the NutriNet-Santé cohort.

IARC disclaimer: 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, policy or views of the International Agency for Research on Cancer/World Health Organization.

Abbreviations

EFSA

European Food Safety Authority

GNPD

Global New Products Database

HR

hazard ratio

IPAQ

International Physical Activity Questionnaire

GSFA

General Standard for Food Additives

NMF

nonnegative matrix factorization

OQALI

Observatoire de la Qualité de l’Alimentation

SD

standard deviation

UPF

ultra-processed foods

95% CIs

95% confidence intervals

Data Availability

Researchers from public institutions can submit a request to have access to the data for strict reproducibility analysis (systematically accepted) or for a new collaboration, including information on the institution and a brief description of the project to collaboration@etude-nutrinet-sante.fr. All requests will be reviewed by the steering committee of the NutriNet-Santé study. If the collaboration is accepted, a data access agreement will be necessary and appropriate authorizations from the competent administrative authorities may be needed. In accordance with existing regulations, no personal data will be accessible. R/SAS code is available without restrictions upon request at collaboration@etude-nutrinet-sante.fr.

Funding Statement

The NutriNet-Santé study was supported by the following public institutions : Ministère de la Santé, Santé Publique France, Institut National de la Santé et de la Recherche Médicale (INSERM), Institut National de la Recherche pour l’agriculture, l’alimentation et l’environnement (INRAE), Conservatoire National des Arts et Métiers (CNAM), and University Sorbonne Paris Nord. This project has received funding from the European Research Council (ERC) under the European Union’s Horizon 2020 research and innovation program (grant agreement No 864219, ADDITIVES), the French National Cancer Institute (INCa_14059), the French Ministry of Health (arrêté 29.11.19) and the IdEx Université de Paris (ANR-18-IDEX-0001), and a Bettencourt-Schueller Foundation Research Prize 2021. This project was awarded the NACRe (Network for Nutrition and Cancer Research) Partnership Label. BC’s laboratory is supported by a Starting Grant from the European Research Council (ERC) under the European Union’s Horizon 2020 research and innovation program (grant agreement No. ERC-2018-StG- 804135 INVADERS), and the national program “Microbiote” from INSERM. This work only reflects the authors' view, and the funders are not responsible for any use that may be made of the information it contains. Researchers were independent from funders. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.

References

  • 1.Touvier M, da Costa Louzada M, Mozaffarian D, Baker P, Juul F, Srour B. Ultra-processed foods and cardiometabolic health: public health policies to reduce consumption cannot wait. The BMJ. 2023;383:e075294. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 2.Lane M, Gamage E, Du S, Ashtree D, McGuinness A, Gauci S. Ultra-processed food exposure and adverse health outcomes: umbrella review of epidemiological meta-analyses. BMJ. 2024;384:e077310. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 3.World Health Organization (WHO). Food additives [Internet]. 2025. [cited 2024 Jun 12]. Available from: https://www.who.int/news-room/fact-sheets/detail/food-additives#:~:text=Food%20additives%20are%20substances%20primarily%20added%20to%20processed,be%20stored%2C%20or%20modify%20sensory%20properties%20of%20food [Google Scholar]
  • 4.Srour B, Kordahi MC, Bonazzi E, Deschasaux-Tanguy M, Touvier M, Chassaing B. Ultra-processed foods and human health: from epidemiological evidence to mechanistic insights. Lancet Gastroenterol Hepatol. 2022;7(12):1128–40. doi: 10.1016/S2468-1253(22)00169-8 [DOI] [PubMed] [Google Scholar]
  • 5.Debras C, Deschasaux-Tanguy M, Chazelas E, Sellem L, Druesne-Pecollo N, Esseddik Y. Artificial sweeteners and risk of type 2 diabetes in the prospective NutriNet-Santé cohort. Diabetes Care. 2023;46(9):1681–90. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 6.Salame C, Srour B, Viennois E, Huybrechts I, Hercberg S, Deschasaux-Tanguy M. Food emulsifiers and risk of type 2 diabetes: results from the prospective NutriNet-Santé cohort. Eur J Public Health. 2023;33(Supplement_2):ckad160.1248. doi: 10.1093/eurpub/ckad160.1248 [DOI] [Google Scholar]
  • 7.Chazelas E, Deschasaux M, Srour B, Kesse-Guyot E, Julia C, Alles B. Food additives: distribution and co-occurrence in 126,000 food products of the French market. Scientific Reports. 2020;10(1):3980. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 8.Chazelas E, Druesne-Pecollo N, Esseddik Y, de Edelenyi F, Agaesse C, De Sa A. Exposure to food additive mixtures in 106,000 French adults from the NutriNet-Santé cohort. Sci Rep. 2021;11(1):19680. doi: 10.1038/s41598-021-98880-5 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 9.Meng X, Yang Y, Zhang J, Su J, Li G, Liu S. Synergistic toxicity of some food additives used in non-alcoholic beverages on renal tubular epithelial cells. Cell Mol Biol Noisy--Gd Fr. 2023;69(10):207–16. [DOI] [PubMed] [Google Scholar]
  • 10.Recoules C, Touvier M, Pierre F, Audebert M. Evaluation of the toxic effects of food additives, alone or in mixture, in four human cell models. Food Chem Toxicol. 2025;196:115198. doi: 10.1016/j.fct.2024.115198 [DOI] [PubMed] [Google Scholar]
  • 11.Hercberg S, Castetbon K, Czernichow S, Malon A, Mejean C, Kesse E. The Nutrinet-Santé Study: a web-based prospective study on the relationship between nutrition and health and determinants of dietary patterns and nutritional status. BMC Public Health. 2010;10:242. doi: 10.1186/1471-2458-10-242 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 12.Touvier M, Méjean C, Kesse-Guyot E, Pollet C, Malon A, Castetbon K. Comparison between web-based and paper versions of a self-administered anthropometric questionnaire. Eur J Epidemiol. 2010;25(5):287–96. [DOI] [PubMed] [Google Scholar]
  • 13.Committee IR. Guidelines for data processing and analysis of the International Physical Activity Questionnaire (IPAQ)-short and long forms. http://www.ipaq.ki.se/scoring.pdf [Internet]. 2005. [cited 2023 Feb 26]; Available from: https://cir.nii.ac.jp/crid/1573950400545533440 [Google Scholar]
  • 14.Lassale C, Péneau S, Touvier M, Julia C, Galan P, Hercberg S, et al. Validity of web-based self-reported weight and height: results of the Nutrinet-Santé study. J Med Internet Res. 2013;15(8):e152. doi: 10.2196/jmir.2575 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 15.Lassale C, Castetbon K, Laporte F, Deschamps V, Vernay M, Camilleri G. Correlations between fruit, vegetables, fish, vitamins, and fatty acids estimated by web-based nonconsecutive dietary records and respective biomarkers of nutritional status. J Acad Nutr Diet. 2016;116(3):427-438.e5. [DOI] [PubMed] [Google Scholar]
  • 16.Lassale C, Castetbon K, Laporte F, Camilleri G, Deschamps V, Vernay M. Validation of a web-based, self-administered, non-consecutive-day dietary record tool against urinary biomarkers. Br J Nutr. 2015;113(6):953–62. [DOI] [PubMed] [Google Scholar]
  • 17.Martinez-Steele E, Khandpur N, Batis C, Bes-Rastrollo M, Bonaccio M, Cediel G, et al. Best practices for applying the Nova food classification system. Nat Food. 2023;4(6):445–8. doi: 10.1038/s43016-023-00779-w [DOI] [PubMed] [Google Scholar]
  • 18.Lee D, Seung H. Algorithms for non-negative matrix factorization. Adv Neural Inform Process Syst. 2001;13. [Google Scholar]
  • 19.Hoyer PO. Non-negative matrix factorization with sparseness constraints. J Mach Learn Res. 2004;5(9). [Google Scholar]
  • 20.Brunet J, Tamayo P, Golub T, Mesirov J. Metagenes and molecular pattern discovery using matrix factorization. Proc Natil Acad Sci. 2004;101(12):4164–9. doi: 10.1073/pnas.0308531101 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 21.Gaujoux R, Seoighe C. A flexible R package for nonnegative matrix factorization. BMC Bioinformatics. 2010;11:367. doi: 10.1186/1471-2105-11-367 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 22.Desquilbet L, Mariotti F. Dose-response analyses using restricted cubic spline functions in public health research. Stat Med. 2010;29(9):1037–57. doi: 10.1002/sim.3841 [DOI] [PubMed] [Google Scholar]
  • 23.Srour B, Chazelas E, Druesne-Pecollo N, Esseddik Y, de Edelenyi FS, Agaësse C, et al. Dietary exposure to nitrites and nitrates in association with type 2 diabetes risk: results from the NutriNet-Santé population-based cohort study. PLoS Med. 2023;20(1):e1004149. doi: 10.1371/journal.pmed.1004149 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 24.Suez J, Cohen Y, Valdés-Mas R, Mor U, Dori-Bachash M, Federici S. Personalized microbiome-driven effects of non-nutritive sweeteners on human glucose tolerance. Cell. 2022;185(18):3307-3328.e19. [DOI] [PubMed] [Google Scholar]
  • 25.Chassaing B, Van de Wiele T, De Bodt J, Marzorati M, Gewirtz AT. Dietary emulsifiers directly alter human microbiota composition and gene expression ex vivo potentiating intestinal inflammation. Gut. 2017;66(8):1414–27. doi: 10.1136/gutjnl-2016-313099 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 26.Biggelaar L, Sep S, Mari A, Ferrannini E, van Dongen M, Wijckmans N. Association of artificially sweetened and sugar-sweetened soft drinks with β-cell function, insulin sensitivity, and type 2 diabetes: the Maastricht Study. Eur J Nutr. 2020;59(4):1717–27. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 27.McCann D, Barrett A, Cooper A, Crumpler D, Dalen L, Grimshaw K, et al. Food additives and hyperactive behaviour in 3-year-old and 8/9-year-old children in the community: a randomised, double-blinded, placebo-controlled trial. Lancet. 2007;370(9598):1560–7. doi: 10.1016/S0140-6736(07)61306-3 [DOI] [PubMed] [Google Scholar]
  • 28.Paudel D, Nair DVT, Tian S, Hao F, Goand UK, Joseph G, et al. Dietary fiber guar gum-induced shift in gut microbiota metabolism and intestinal immune activity enhances susceptibility to colonic inflammation. Gut Microbes. 2024;16(1):2341457. doi: 10.1080/19490976.2024.2341457 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 29.Di Vincenzo F, Del Gaudio A, Petito V, Lopetuso LR, Scaldaferri F. Gut microbiota, intestinal permeability, and systemic inflammation: a narrative review. Intern Emerg Med. 2024;19(2):275–93. doi: 10.1007/s11739-023-03374-w [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 30.Bhattacharyya S, O-Sullivan I, Katyal S, Unterman T, Tobacman JK. Exposure to the common food additive carrageenan leads to glucose intolerance, insulin resistance and inhibition of insulin signalling in HepG2 cells and C57BL/6J mice. Diabetologia. 2012;55(1):194–203. doi: 10.1007/s00125-011-2333-z [DOI] [PubMed] [Google Scholar]
  • 31.Kimilu N, Gładyś-Cieszyńska K, Pieszko M, Mańkowska-Wierzbicka D, Folwarski M. Carrageenan in the diet: friend or foe for inflammatory bowel disease?. Nutrients. 2024;16(11):1780. doi: 10.3390/nu16111780 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 32.Richardson IL, Frese SA. Non-nutritive sweeteners and their impacts on the gut microbiome and host physiology. Front Nutr. 2022;9:988144. doi: 10.3389/fnut.2022.988144 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 33.Abou-Donia MB, El-Masry EM, Abdel-Rahman AA, McLendon RE, Schiffman SS. Splenda alters gut microflora and increases intestinal p-glycoprotein and cytochrome p-450 in male rats. J Toxicol Environ Health A. 2008;71(21):1415–29. doi: 10.1080/15287390802328630 [DOI] [PubMed] [Google Scholar]
  • 34.Hanawa Y, Higashiyama M, Kurihara C, Tanemoto R, Ito S, Mizoguchi A, et al. Acesulfame potassium induces dysbiosis and intestinal injury with enhanced lymphocyte migration to intestinal mucosa. J Gastroenterol Hepatol. 2021;36(11):3140–8. doi: 10.1111/jgh.15654 [DOI] [PubMed] [Google Scholar]
  • 35.Suez J, Korem T, Zeevi D, Zilberman-Schapira G, Thaiss CA, Maza O, et al. Artificial sweeteners induce glucose intolerance by altering the gut microbiota. Nature. 2014;514(7521):181–6. doi: 10.1038/nature13793 [DOI] [PubMed] [Google Scholar]
  • 36.Stevens L, Kuczek T, Burgess J, Stochelski M, Arnold L, Galland L. Mechanisms of behavioral, atopic, and other reactions to artificial food colors in children. Nutr Rev. 2013;71(5):268–81. [DOI] [PubMed] [Google Scholar]
  • 37.Lau K, McLean W, Williams D, Howard C. Synergistic interactions between commonly used food additives in a developmental neurotoxicity test. Toxicol Sci. 2006;90(1):178–87. [DOI] [PubMed] [Google Scholar]
  • 38.Doguc DK, Aylak F, Ilhan I, Kulac E, Gultekin F. Are there any remarkable effects of prenatal exposure to food colourings on neurobehaviour and learning process in rat offspring?. Nutr Neurosci. 2015;18(1):12–21. doi: 10.1179/1476830513Y.0000000095 [DOI] [PubMed] [Google Scholar]
  • 39.Touvier M, Kesse-Guyot E, Méjean C, Pollet C, Malon A, Castetbon K, et al. Comparison between an interactive web-based self-administered 24 h dietary record and an interview by a dietitian for large-scale epidemiological studies. Br J Nutr. 2011;105(7):1055–64. doi: 10.1017/S0007114510004617 [DOI] [PubMed] [Google Scholar]
  • 40.Vergnaud A, Touvier M, Méjean C, Kesse-Guyot E, Pollet C, Malon A. Agreement between web-based and paper versions of a socio-demographic questionnaire in the NutriNet-Santé study. Int J Public Health. 2011;56(4):407–17. [DOI] [PubMed] [Google Scholar]
  • 41.van Gelder MMHJ, Bretveld RW, Roeleveld N. Web-based questionnaires: the future in epidemiology?. Am J Epidemiol. 2010;172(11):1292–8. doi: 10.1093/aje/kwq291 [DOI] [PubMed] [Google Scholar]
  • 42.Joinson A. Social desirability, anonymity, and Internet-based questionnaires. Behav Res Methods Instrum Comput. 1999;31(3):433–8. doi: 10.3758/bf03200723 [DOI] [PubMed] [Google Scholar]
  • 43.Wang Y-C, Lee C-M, Lew-Ting C-Y, Hsiao CK, Chen D-R, Chen WJ. Survey of substance use among high school students in Taipei: web-based questionnaire versus paper-and-pencil questionnaire. J Adolesc Health. 2005;37(4):289–95. doi: 10.1016/j.jadohealth.2005.03.017 [DOI] [PubMed] [Google Scholar]
  • 44.Lailler G, Piffaretti C, Fuentes S, Nabe H, Oleko A, Cosson E. Prevalence of prediabetes and undiagnosed type 2 diabetes in France: results from the national survey ESTEBAN, 2014–2016. Diabetes Research and Clinical Practice. 2020;165:108252. doi: 10.1016/j.diabres.2020.108252 [DOI] [PubMed] [Google Scholar]

Decision Letter 0

Alexandra Tosun

12 Sep 2024

Dear Dr Payen de la Garanderie,

Thank you for submitting your manuscript entitled "Food additive mixtures and risk of type 2 diabetes: results from the NutriNet-Santé cohort" for consideration by PLOS Medicine.

Your manuscript has now been evaluated by the PLOS Medicine editorial staff and I am writing to let you know that we would like to send your submission out for external peer review.

However, before we can send your manuscript to reviewers, we need you to complete your submission by providing the metadata that is required for full assessment. To this end, please login to Editorial Manager where you will find the paper in the 'Submissions Needing Revisions' folder on your homepage. Please click 'Revise Submission' from the Action Links and complete all additional questions in the submission questionnaire.

Please re-submit your manuscript within two working days, i.e. by Sep 16 2024.

Login to Editorial Manager here: https://www.editorialmanager.com/pmedicine

Once your full submission is complete, your paper will undergo a series of checks in preparation for peer review. Once your manuscript has passed all checks it will be sent out for review.

Feel free to email me at atosun@plos.org or us at plosmedicine@plos.org if you have any queries relating to your submission.

Kind regards,

Alexandra Tosun, PhD

Associate Editor

PLOS Medicine

Decision Letter 1

Alexandra Tosun

25 Oct 2024

Dear Dr Payen de la Garanderie,

Many thanks for submitting your manuscript "Food additive mixtures and risk of type 2 diabetes: results from the NutriNet-Santé cohort" (PMEDICINE-D-24-03030R1) to PLOS Medicine. The paper has been reviewed by subject experts and a statistician; their comments are included below and can also be accessed here: [LINK]

As you will see, the reviewers found the study to be interesting, but raised several points for clarification. After discussing the paper with the editorial team and an academic editor with relevant expertise, I'm pleased to invite you to revise the paper in response to the reviewers' comments. Please note that the manuscript has generated extensive discussion among the team, so we would like to emphasize that the reviewers' and academic editor's comments should be carefully considered. We plan to send the revised paper to some or all of the original reviewers, and we cannot provide any guarantees at this stage regarding publication.

When you upload your revision, please include a point-by-point response that addresses all of the reviewer and editorial points, indicating the changes made in the manuscript and either an excerpt of the revised text or the location (eg: page and line number) where each change can be found. Please also be sure to check the general editorial comments at the end of this letter and include these in your point-by-point response. When you resubmit your paper, please include a clean version of the paper as the main article file and a version with changes tracked as a marked-up manuscript. It may also be helpful to check the guidelines for revised papers at http://journals.plos.org/plosmedicine/s/revising-your-manuscript for any that apply to your paper.

We ask that you submit your revision by Nov 15 2024. However, if this deadline is not feasible, please contact me by email, and we can discuss a suitable alternative.

Don't hesitate to contact me directly with any questions (atosun@plos.org).

Best regards,

Alexandra

Alexandra Tosun, PhD

Associate Editor

PLOS Medicine

atosun@plos.org

-----------------------------------------------------------

Comments from the academic editor:

The authors conducted an observational study relating 'mixtures' of food additives to incidence of type 2 diabetes.

The major and minor comments are supplied hereafter.

Major comments:

0. Some of the individual food additives were evaluated previously, as cited #5 (Lancet Diabetes Endocrinol, 2024) and were shown to be associated with type 2 diabetes incidence in the same population. Because of the associations, the novelty of this current study is weak. Also, two analytic studies from the same cohort could question whether single chemicals or a combination played a role in showing the positive association. The authors discussed it but did not address the question.

The previous study reported in Lancet D&E 2024 had the weakness of mutual confounding between food additives (emulsifiers), and then this study demonstrated it. It would be an important notion, but it remained unclear if the positive association was due to a single factor or a combination. Without such an in-depth investigation, this study seems weak.

The authors argued the "cocktail effect", but a specific analysis to test the presence is absent, and the method is too weak to demonstrate or argue the effect. The authors cannot rule out the possibility that a single agent drove the positive association. Even if the "cocktail effect" looked significant, measurement errors and confounding would remain concerning in the multivariable setting.

The authors should design their analyses to test the effect of multiple food additives above and beyond the individual ones. Of note, several food additives in the statistically identified "mixture" were confirmed to be associated with type 2 diabetes, so the confounded interpretations should be avoided. The authors seem aware of the research question (Page 14 in the Discussion) but not aware of proper methods, but any approaches need to be conducted with or without limitations/assumptions. At least, the authors should cite the Lancet D&E paper and re-justify this study to test the hypothesis of the interaction.

The "Cocktail" effect is not a technical term, and the authors should avoid it. What the authors examined must align with a hypothesis about an interaction. The authors should document their interest using such a known term.

1. The authors documented the limitation of residual confounding, but the effort to reduce it and the highlight seem insufficient. The authors collected the majority of the information through the Internet. Self-reports partly affected by health consciousness must cause differential measurement errors, such as those of body weight and height, as is often the case, particularly among women. Such uncharacterised errors could happen to many covariates, cause insufficient adjustment for those and cause further residual confounding.

When the authors described the study limitations, they argued, "Despite extensive adjustments for confounding variables, including dietary, lifestyle, anthropometric, and socio-demographic factors, ..."

This statement indicated that the authors had identified many covariates to adjust for in their analysis. However, socioeconomic status (SES) was not adjusted for well. It likely contributed to dietary behaviours, health consciousness, health-seeking behaviours, and others that influence health outcomes through numerous mechanisms. Despite the well-known confounding roles of SES, the authors included only a single three-level covariate of education history. Just three levels were unlikely to capture the SES. The authors must be aware of the diversity of the study population, including race/ethnic, regional, cultural, and economic diversity across France. The authors did not account for any of those, except for educational attainment. The authors should make a greater effort.

The authors' group may not have considered it in previous studies from the same cohort, but that would not matter. The authors should adjust for a reasonable amount of SES variables, such as household or individual incomes, occupations, race/ethnic status, and religion. (Otherwise, collider bias happens in the current analysis, causing unmeasured confounding.).

Health consciousness is also a potential confounder, as the authors noted. The authors should adjust for participation in clinical screening, access to dental care, and others.

Those covariates would matter in this study because the authors have identified weak positive associations. Weak positive confounders could elevate the likelihood of detecting false-positive associations.

2.

The authors investigated a linear combination of multiple food additives. However, this does not mean that the authors examined a synergistic or cocktail effect of the multiple food additives. Possibly, only a single food additive within a mixture was causal, whereas other additives had no effect. Taking the possibility, the authors should limit and tone down their conclusion as if there were a cocktail effect. Then, they should argue the need for experimental studies to identify whether only a single causal agent may play a role predominantly or a synergism may happen.

3.

The authors identified type 2 diabetes cases with both subjective (self-reports) and objective information. The approach is great but could miss undiagnosed type 2 diabetes, which could be present to an unignorable degree. Individuals with unrecognized diabetes cannot be identified by self-reports or the national registry. The authors should discuss it.

4.

For each of the "mixture" #2 and #5, the authors should fit a regression model including individual food additives that contribute to the mixture and the mixture itself. If the positive association of the mixture with T2D remains, that should be considered as evidence that the combination matters beyond individual effects.

Similarly, the authors should demonstrate whether or not the positive association was driven by single factor or multiple factors.

Basically, the authors should not give up differentiating between the potential of individual effects and the potential of the effect of the combination or interaction.

These analyses are essential for this manuscript, in addition to the maximal effort to adjust for socioeconomic status and health-seeking behaviours. For the latter confounding, the authors should identify if the participants joined any screening activities, and I hope those are available in their registration data / linked medical records.

5.

In Discussion, the authors should discuss whether the previous studies on emulsifiers and artificial sweeteners should be interpreted. It seems that mutual confounding between food additives substantially matter. The authors should explicitly state that previous findings should be interpreted more than published, with regards to the issue of mutual confounding.

Minor comments:

Risk and incidence should be distinguished.

The aggregates of the food additives are unclear in terms of their clinical relevance. The analyses conducted for the current eTable 4 should be elaborated more to identify food sources and dietary patterns. SES may better be carefully considered, too.

On page 6 and the other pages, the authors should not use "validated" to describe the validity of their measurements. No variable must have been measured with perfection. The authors should document how valid each measure was as much as possible. Given the aim of this work, the authors must be mindful of whether or not each measured variable could identify a between-individual difference rather than its absolute levels of exposure. Thus, the authors should document the quantitative abilities of the exposures and covariates (such as BMI) to rank individuals.

As mentioned above, the measurement error of BMI and behavioural factors were more concerning than usual because the authors recruited individuals over the Internet and collected self-reports of many variables.

Page 8:

The authors documented specifications in their dimension-reduction technique. Those selections are subject to sensitivity analyses.

-----------------------------------------------------------

Comments from the reviewers:

Reviewer #1: See attachment

Michael Dewey

Reviewer #2: The paper by de la Garanderie et al. Reports a very fancy prospective cohort analysis on the relationship between the consumption of 5 different food additive mixtures and type 2 diabetes incidence, conducted in the large NutriNetSanté cohort in France.

Investigators first characterized presence/absence and doses of several food additives by leveraging food records completed by participants, ingredient lists coupled with different databases on food additives. This method was previously described and published in Sci Rep.

They then used a data-driven approach (NMF) to identify 5 food additive mixtures. These mixtures represent "clusters" of food additive commonly consumed together due to concomitant presence in the same food for instance.

They then assessed using cox models the relationship between each mixture and T2D incidence. Two mixtures were found to be positively associated with T2D risk. The first was mostly composed of emulsifiers and the second was composed of food additives commonly present in SSB and ASB.

Analyses are comprehensive and the paper is well written.

I think this paper can become an important contribution to the public health discussion on the place of UPF in the diet.

My main concern is related to the separation between the effects of the food additive mixtures per se from the one of the food matrix per se, or even the diet pattern. As mentioned by the authors, the first mixture included additives present in a variety of UPF, including broth dairy desserts, fats and sauces - all foods that have been associated with higher risk of T2D in many cohorts. Likewise, the other group was SSB/ASB driven, which are also important dietary risk factors of T2D (mostly SSB). In an attempt to control for diet quality, authors adjusted for intakes of energy without alcohol (continuous, kcal/d), total saturated fatty acids (continuous, g/d), sodium (continuous, mg/d), dietary fibre (continuous, g/d), alcohol (continuous, g/d), added sugars (continuous, g/d), fruits and vegetables (g/d), dairy products (continuous, ml/day), red and processed meats (continuous, g/d).

I first question the concomitant presence of nutrients and foods in the model. Adjusting for saturated fat and dairy products and red meat seems inappropriate because these foods are the main sources of saturated fat. I would first suggest using two distinct models, one nutrient-based and the other food-based.

Even by doing so, the control for the food matrix will likely be imperfect because the food additive mixtures are found in a limited number of different foods. To address this issue, I would invite the authors to test whether the food additive mixtures mediate the relationship between the foods that are the main sources of these food additive and T2D. Such model would provide key information on the potential contribution of the food additive mixture on the causal pathway between the foods that contain these additive and T2D. These models should also include key nutrients know to drive the relationship between the foods and T2D risk. For instance, relative to SSB/ASB consumption, food mixture #5, and T2D: Is food mixture #5 mediating the relationship between SSB/ASB and T2D risk? If yes, how strong is the effect vs with the one of added sugar? If food mixture #5 is not found to be a mediator of the relationship between SSB/ASB and T2D, what are the implications relative to the paper conclusions? The same should be tested with key foods associated with food additive mixture #1.

Also, I would strongly recommend adding analyses stratified by diet quality. The two food additive mixtures associated with T2D are associated with UPF, and UPF consumption correlate with lower diet quality, it appears important to evaluate whether the detrimental relationship between the food additive mixtures and T2D remains valid at any level of diet quality. Such data are key for public health policies.

I acknowledge that my recommendations require significant statistical work, but my concerns echo to key questions in the public health debate on the place of UPF in the diet. Addressing these questions would allow significant advances in the UPF field.

Merci

Reviewer #3: The paper is quite interesting, and the authors have conducted a remarkable research. However, I have two minor points regarding the study's limitations:

1) The authors stated: "Next, the generalizability of our findings may be influenced by the cohort's demographic characteristics, such as a higher proportion of women and a more health-conscious population. Therefore, caution is warranted when extrapolating our results to broader populations." As the study aims to evaluate an association, the generalizability is not necessarily compromised by demographic characteristics (as in a prevalence st. If the authors believe there are specific biases affecting the associations, they should clearly identify these issues and explain how they might impact the results.

2) Regarding the final limitation: "Last, this study did not allow us to specifically investigate mechanistic synergies and/or antagonisms between the food additive chemicals characterizing the mixtures per se.". While the authors suggest experimental studies to address this limitation, it's worth noting that certain epidemiological statistical analyses can account for synergistic effects among variables. It would be beneficial if the authors could: a) Explain why this limitation is relevant to the current study. b) Suggest alternative epidemiological approaches that could provide deeper insights into these associations.

Any attachments provided with reviews can be seen via the following link: [LINK]

--------------------------------------------------------- ---

General editorial requests:

(Note: not all will apply to your paper, but please check each item carefully)

* We ask every co-author listed on the manuscript to fill in a contributing author statement, making sure to declare all competing interests. If any of the co-authors have not filled in the statement, we will remind them to do so when the paper is revised. If all statements are not completed in a timely fashion this could hold up the re-review process. If new competing interests are declared later in the revision process, this may also hold up the submission. Should there be a problem getting one of your co-authors to fill in a statement we will be in contact. Please do not add or remove authors without first discussing this with the handling editor. You can see our competing interests policy here: http://journals.plos.org/plosmedicine/s/competing-interests.

* Please upload any figures associated with your paper as individual TIF or EPS files with 300dpi resolution at resubmission; please read our figure guidelines for more information on our requirements: http://journals.plos.org/plosmedicine/s/figures. While revising your submission, please upload your figure files to the PACE digital diagnostic tool, https://pacev2.apexcovantage.com/. PACE helps ensure that figures meet PLOS requirements. To use PACE, you must first register as a user. Then, login and navigate to the UPLOAD tab, where you will find detailed instructions on how to use the tool. If you encounter any issues or have any questions when using PACE, please email us at PLOSMedicine@plos.org.

* Please ensure that the paper adheres to the PLOS Data Availability Policy (see http://journals.plos.org/plosmedicine/s/data-availability), which requires that all data underlying the study's findings be provided in a repository or as Supporting Information. For data residing with a third party, authors are required to provide instructions with contact information (web or email address) for obtaining the data. Please note that a study author cannot be the contact person for the data. PLOS journals do not allow statements supported by "data not shown" or "unpublished results." For such statements, authors must provide supporting data or cite public sources that include it.

* We expect all researchers with submissions to PLOS in which author-generated code underpins the findings in the manuscript to make all author-generated code available without restrictions upon publication of the work. In cases where code is central to the manuscript, we may require the code to be made available as a condition of publication. Authors are responsible for ensuring that the code is reusable and well documented. Please make any custom code available, either as part of your data deposition or as a supplementary file. Please add a sentence to your data availability statement regarding any code used in the study, e.g. "The code used in the analysis is available from Github [URL] and archived in Zenodo [DOI link]" Please review our guidelines at https://journals.plos.org/plosmedicine/s/materials-software-and-code-sharing and ensure that your code is shared in a way that follows best practice and facilitates reproducibility and reuse. Because Github depositions can be readily changed or deleted, we encourage you to make a permanent DOI'd copy (e.g. in Zenodo) and provide the URL.

* FINANCIAL DISCLOSURES: The funding statement should include: specific grant numbers, initials of authors who received each award, URLs to sponsors’ websites. Also, please state whether any sponsors or funders (other than the named authors) played any role in study design, data collection and analysis, the decision to publish, or preparation of the manuscript. If they had no role in the research, include this sentence: “The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.”

* COMPETING INTEREST: All authors must declare their relevant competing interests per the PLOS policy, which can be seen here: https://journals.plos.org/plosmedicine/s/competing-interests

For authors with ties to industry, please indicate whether any of the interests has a financial stake in the results of the current study.

* DATA AVAILABILITY: Please include the details provided under "Data and Resource Availability" in the Data Availability statement in the online submission.

FORMATTING - GENERAL

* Abstract: Please structure your abstract using the PLOS Medicine headings (Background, Methods and Findings, Conclusions). Please combine the Methods and Findings sections into one section.

* At this stage, we ask that you include a short, non-technical Author Summary of your research to make findings accessible to a wide audience that includes both scientists and non-scientists. The Author Summary should immediately follow the Abstract in your revised manuscript. This text is subject to editorial change and should be distinct from the scientific abstract. Ideally each sub-heading should contain 2-3 single sentence, concise bullet points containing the most salient points from your study. In the final bullet point of 'What Do These Findings Mean?', please include the main limitations of the study in non-technical language. Please see our author guidelines for more information: https://journals.plos.org/plosmedicine/s/revising-your-manuscript#loc-author-summary.

* Please express the main results with 95% CIs as well as p values. When reporting p values please report as p<0.001 and where higher as the exact p value p=0.002, for example. Throughout, suggest reporting statistical information as follows to improve clarity for the reader "22% (95% CI [13%,28%]; p</=)". Please be sure to define all numerical values at first use.

* Please include page numbers and line numbers in the manuscript file. Use continuous line numbers (do not restart the numbering on each page).

* Please cite the reference numbers in square brackets. Citations should precede punctuation.

FIGURES AND TABLES

* Please provide titles and legends for all figures and tables (including those in Supporting Information files).

* Please define all abbreviations used in each figure/table (including those in Supporting Information files).

* Please consider avoiding the use of red and green in order to make your figure more accessible to those with color blindness.

SUPPLEMENTARY MATERIAL

* Please note that supplementary material will be posted as supplied by the authors. Therefore, please amend it according to the relevant comments outlined here.

* Please cite your Supporting Information as outlined here: https://journals.plos.org/plosmedicine/s/supporting-information

REFERENCES

* PLOS uses the numbered citation (citation-sequence) method and first six authors, et al.

* Please ensure that journal name abbreviations match those found in the National Center for Biotechnology Information (NCBI) databases (http://www.ncbi.nlm.nih.gov/nlmcatalog/journals), and are appropriately formatted and capitalised.

* Where website addresses are cited, please include the complete URL and specify the date of access (e.g. [accessed: 12/06/2024]).

* Please also see https://journals.plos.org/plosmedicine/s/submission-guidelines#loc-references for further details on reference formatting.

STUDY TYPE-SPECIFIC REQUESTS

* Abstract: Please include the study design, population and setting, number of participants, years during which the study took place (enrollment and follow up), length of follow up, and main outcome measures.

* Please ensure that the study is reported according to the STROBE (or appropriate STOBE extension) guideline (available from: https://www.equator-network.org/reporting-guidelines/strobe) and include the completed STROBE (or STROBE extension) checklist as Supporting Information. Please add the following statement, or similar, to the Methods: "This study is reported as per the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) guideline (S1 Checklist)." When completing the checklist, please use section and paragraph numbers, rather than page numbers.

* For all observational studies, in the manuscript text, please indicate: (1) the specific hypotheses you intended to test, (2) the analytical methods by which you planned to test them, (3) the analyses you actually performed, and (4) when reported analyses differ from those that were planned, transparent explanations for differences that affect the reliability of the study's results. If a reported analysis was performed based on an interesting but unanticipated pattern in the data, please be clear that the analysis was data driven.

* Please state in the Methods section whether the study had a prospective protocol or analysis plan. If a prospective analysis plan (from your funding proposal, IRB or other ethics committee submission, study protocol, or other planning document written before analyzing the data) was used in designing the study, please include the relevant document(s) with your revised manuscript as a Supporting Information file to be published alongside your study and cite it in the Methods section. A legend for this file should be included at the end of your manuscript. If no such document exists, please make sure that the Methods section transparently describes when analyses were planned, and when/why any data-driven changes to analyses took place. Changes in the analysis, including those made in response to peer review comments, should be identified as such in the Methods section of the paper, with rationale.

Attachment

Submitted filename: garanderie.pdf

pmed.1004570.s004.pdf (54.1KB, pdf)

Decision Letter 2

Alexandra Tosun

31 Jan 2025

Dear Dr Payen de la Garanderie,

Many thanks for re-submitting your manuscript "Food additive mixtures and risk of type 2 diabetes: results from the NutriNet-Santé cohort" (PMEDICINE-D-24-03030R2) to PLOS Medicine. The paper has been seen again by one subject expert and the statistician; their comments are included below and can also be accessed here: [LINK]

Thank you for your detailed response to the reviewers' comments. As you will see, the reviewers are satisfied with your responses to their comments. However, there are a number of remaining comments from the Academic Editor that require further clarification. After discussing the paper with the editorial team, we ask you to carefully address the comments in a further revision. We plan to send the revised paper to some or all of the original reviewers.

When you upload your revision, please include a point-by-point response that addresses all of the reviewer and editorial points, indicating the changes made in the manuscript and either an excerpt of the revised text or the location (eg: page and line number) where each change can be found. Please also be sure to check the general editorial comments at the end of this letter and include these in your point-by-point response. When you resubmit your paper, please include a clean version of the paper as the main article file and a version with changes tracked as a marked-up manuscript. It may also be helpful to check the guidelines for revised papers at http://journals.plos.org/plosmedicine/s/revising-your-manuscript for any that apply to your paper.

We ask that you submit your revision by Feb 20 2025. However, if this deadline is not feasible, please contact me by email, and we can discuss a suitable alternative.

Don't hesitate to contact me directly with any questions (atosun@plos.org).

Best regards,

Alexandra

Alexandra Tosun, PhD

Associate Editor

PLOS Medicine

atosun@plos.org

-----------------------------------------------------------

Comments from the academic editor:

The authors addressed reviewers' comments provided previously, but some concerns remain.

Two major comments are supplied hereafter, followed by minor comments.

1.

The authors failed to demonstrate the interactions/synergisms well. The authors tested statistical interactions, showing p-values in eTable 11. Significant interactions do not mean synergism because antagonism between two particular additives could happen. The authors should display the estimates of regression coefficients (+ 95% CI) of two single terms and an interaction term, three in total, for each test for presenting eTable 11. Each food additive may be studentized to mean=0 and SD=1 for interpretability, after the authors preset means and SDs of the food additives.

The tests for interaction miss the directionality. That is one of the issues in the current presentation. Second, the authors found a lot of null findings for interactions. It is unclear if the interaction effects explain the large fraction of the observed associations. Given the exploratory nature of the interaction following the significant testing of each cluster, the authors should tone down their inferences about interactions/synergisms, at least in the abstract.

In the abstract, the authors should not use the words "interaction/synergistic effects", as this study did not primarily evaluate the effects but evaluated the potential effect of a combination or cluster of multiple correlated additives and then explored the interaction.

The following revisions should be made in the abstract:

Background:

The following sentence should be revised: "These findings suggest that potential interaction/synergistic effects of food additives should be considered in safety assessments..."

A revised sentence may be said, "These findings suggest that potential effects of multiple food additives should be considered in safety assessments..."

Results:

The authors should state that several interactions were indicated in their exploratory analyses for one of the clusters, with clear information on the directionality.

Discussion:

The authors stated, "This study revealed positive associations between exposure to two widely consumed food additive mixtures and increased type 2 diabetes risk. Further experimental research is needed to depict underlying mechanisms. These findings suggest that potential interaction/synergistic effects of food additives should be considered in safety assessments, and they support public health recommendations to limit non-essential additives."

Grammer is not right, partly. This sub-section should be revised, "This study revealed positive associations between exposure to two widely consumed food additive mixtures and higher type 2 diabetes incidence. Further experimental research is needed to depict underlying mechanisms, including potential interactions/synergistic effects. These findings suggest that a combination of food additives should be considered in safety assessments, and they support public health recommendations to limit non-essential additives."

2.

This cohort seemed to be based on participants across France. The authors should clarify the geographical diversity and how they addressed the potential confounding due to geosocial factors.

Some cities are by the Mediterranean Sea, while other cities are mountainous, by the Atlantic Ocean or by the Strait of Dover. Cultural diversity was present, as well as racial and geosocial diversity.

Despite that, the authors adjusted only for education levels and professional occupation as socioeconomic factors. The degree of residual confounding owing to the nationwide diversity may be substantial.

The limitation section appears to include the limited opportunity of collecting race/ethnicity/religion. The authors are encouraged to document any such restriction in the method section, so that readers can digest the results with such an important limitation kept in mind.

Minor comments:

Hazard ratios presented in the abstract are not interpretable because of no information on the unit of the exposure. Please clarify.

Line 116-117:

The authors stated, "These associations were not driven by a specific food additive, and several interactions between additives of the same mixture were detected." This interpretation is misleading. There may be single independent effects of several food additives. The effect of an interaction could be minor. Also, the effect of an interaction could be antagonistic, being outweighed by positive independent single effects. As implied above, the authors should give a sentence that implies multiple interpretations.

Line 127-128: It sounds too early to say "should be", given the presentation of eTable 11 with many negative results and without any replication or biologically plausible explanations. The authors may rephrase, "the potential interactions/synergism may be of interest in future mechanistic investigations and considerations in safety assessment".

The authors also have an issue with measurement errors that are common in any observational research of this kind, owing to the dietary assessment undertaken and food-additive databases used. As the authors noted in the discussion section, those limitations must have mattered. Accounting for multiple epidemiological issues, the authors should follow the suggestions given above.

Line 134: "a major hypothesis" should be rephrased to "one of the hypotheses". There is no evidence that the authors' argument is right.

Line 160-163:

After this statement of the primary aim, the authors may want to state their exploratory, secondary aim to examine interactions between food additives in a cluster associated with type 2 diabetes incidence.

Line 170:

If the recruitment was ongoing, there should be a calender time at which the authors formed a cohort for this specific study. For example, it is unlikely that the authors included participants recruited when this manuscript was being written up. The authors should clarify the calendar time when the participants in this specific study were recruited.

Line 188: "are" should be "were".

Line 189: "validated" should be taken out. There is no possibility that any measures of a dietary assessment tool are perfectly valid. What is appropriate is to present how valid main dietary variables would be for the specific hypothesis to test. In this study, the authors failed to provide such information, so "validated" should be taken out, and some degree of validity should not be indicated in this manuscript.

Line 228: "is" should be "was".

Line 341: "risk" should be incidence. The authors should review their text not to misuse risk and incidence. The authors evaluated time-to-event data and analyzed incidence, not risk. It may be acceptable sometimes to use "risk" when inferring a risk. However, when the authors document results derived from their analyses of incidence (N cases / person-time), they should use properly the term, "incidence". It would be shocking if the authors do not understand the difference between risk and incidence.

Line 358-359:

This description should be elaborated more in this text. The authors are interested in interactions/synergisms, but the descriptions of the resulst are overly simple, and no one can judge the strength of evidence or identifiy the numbers of null results and directionality of the potential interactions.

Line 387:

"supporting potential synergistic effects". As mentioned above, the directions of associations were not presented; it is unclear if synergisms were reasonably indicated. The authors should present results sufficiently so that the authors' arguments make sense.

Line 413-415: These hypotheses were not given in the introductory section. At the very least, the brief intention should be documented as suggested above. It is odd to see the authors clarify their hypothesis in the discussion section.

Line 415-419: Confirm those with the directions of the interaction estimates.

Line 449: "highlighting" should be rephrased to "indicating". As repeatedly implicated, the presence of an effect of a specific mixture does not necessarily mean an interaction/synergism.

Line 511: Misclassification of diabetes cases as non-cases, considering undiagnosed diabetes as non-cases, is likely to depend on socioeconomic status and health consciousness. Also, if the size of cohort is big, non-differential misclassification can be overwhelming. The authors should eliminate the argument as if the impact was negligible. (Basically, the authors misunderstood the nature of outcome misclassification. Some outcome misclassifications, such as low positive predictive value, may cause just a power issue, but not necessarily the case).

Line 517-518: Too weak argument. Take it out. Consistency with national averages should not happen when the assessment was valid, as this study recruited more women than men, for example. The authors do not need to supply weak arguments as no study is perfect anyway.

-----------------------------------------------------------

Comments from the reviewers:

Reviewer #1: The authors have addressed my points.

Michael Dewey

Reviewer #2: Thank you for addressing carefuly the comments raised previously. I have no further comments.

Any attachments provided with reviews can be seen via the following link: [LINK]

-----------------------------------------------------------

Requests from Editors:

TITLE

Please check the formatting of your title (Is there a line break following the colon?). Please check that the title is according to PLOS Medicine's style. Your title must be nondeclarative and not a question. It should begin with main concept if possible. "Effect of" should be used only if causality can be inferred, i.e., for an RCT. Please place the study design ("A randomized controlled trial," "A retrospective study," "A modelling study," etc.) in the subtitle (i.e., after a colon).

ABSTRACT

1) l.59ff: Please replace ‘y’ with ‘years’ or define ‘y’ at first use. We would prefer the first option.

2) l.59ff: Please define ‘SD’ at first use. Throughout the abstract and main text, please ensure that abbreviations, including statistical abbreviations, are defined the first time they are used.

3) ll.59-60: For clarity, we suggest changing the sentence to: “Participants (n=108,643, mean follow-up = 7.7 years (standard deviation (SD) = 4.6), age = 42.5 years (SD = 14.6), 79.2% women) were adults from the French NutriNet-Santé cohort (2009-2023).”

4) l.68: When reporting 95% CIs please separate upper and lower bounds with commas instead of hyphens as the latter can be confused with reporting of negative values. Please revise throughout the manuscript. Also, please remember to introduce abbreviations, such as ‘HR’.

5) We feel that for full transparency you should also report the results, i.e. HR values, for the three food mixtures that did not show an association with increased diabetes risk.

6) In the last sentence of the Abstract Methods and Findings section, please describe the main limitation(s) of the study's methodology.

7) Please ensure that all numbers presented in the abstract are present and identical to numbers presented in the main manuscript text.

9) Please include the important dependent variables that are adjusted for in the analyses.

AUTHOR SUMMARY

1) Please remove any numerical results from the Author Summary.

2) Please temper claims of primacy of results by stating, "to our knowledge" (or something similar) or remove ‘is the first’.

3) We suggest changing the order of the bullet points under ‘What do these findings mean?’:

• The study results suggest that food additives found in a wide variety of products and frequently consumed together may potentially represent a modifiable risk factor for type 2 diabetes prevention. They support public health recommendations to limit non-essential additives.

• Potential interaction/synergistic effects of food additives should be considered in future safety assessments.

• Confirmation by other epidemiological and experimental studies will be necessary to support causality of the observed associations.

4) Please note that in the final bullet point of ‘What Do These Findings Mean?’ the main limitations of the study should be included in non-technical language (i.e. the above might require further changes).

INTRODUCTION

1) If there has been a systematic review of the evidence related to your study (or you have conducted one), please refer to and reference that review and indicate whether it supports the need for your study.

2) l.136ff: Please ensure to provide references.

3) ll.153-155: We suggest changing to: “Another important gap to date has been that previous evaluations have not been able to account for potential interaction/synergistic effects when assessing the safety of additives due to a lack of data.”

4) l.155: Please use the abbreviation 'UPF' instead of 'ultra-processed foods' as you introduced the abbreviation early in the Introduction. Please revise throughout.

5) ll.158-160: Please revise for grammar/clarity.

6) l.161: Please temper claims of primacy of results by stating, "to our knowledge" (or something similar).

7) l.163: We suggest writing '...using the large prospective NutriNet-Santé cohort'.

METHODS AND RESULTS

1) ll.223-224: We suggest providing a brief explanation of ‘under-energy reporters’ (e.g. in parenthesis).

2) l.251: Please spell out ‘T1D’ as you have not introduced the abbreviation.

3) l.295: The terms gender and sex are not interchangeable (as discussed in https://www.who.int/health-topics/gender#tab=tab_1 ); please use the appropriate term (also in the Abstract).

4) l.295, we suggest changing this to "At baseline, the median age of the cohort was ..." to avoid that "their" refers only to females.

5) ll.298-299: “Participants completed a median of 5 dietary records (25th – 75th percentiles: 3-9).” – does this refer to the entire NutriNet-Santé cohort or to the participants specifically included in this study? Please revise the characteristics paragraph for clarity.

6) ll.324-317: We suggest splitting the sentence into two for clarity.

7) l.332: Please remove the word ‘famous’.

8) l.345: If you agree, we feel it is worth repeating here (briefly) the types of sensitivity analyses that have been carried out.

9) l.353ff: Your study is observational and therefore causality cannot be inferred. Please remove language that implies causality, such as effect. Please refer to associations instead. Please revise throughout (including the discussion).

10) l.354: Please note that you have not yet introduced the abbreviation 'T2D' and we would suggest that you continue to write 'type 2 diabetes'. Please revise throughout.

11) Table 1: Please ensure that all abbreviations below the table, such as BMI and SD, are defined. Please also refer to comment 3) and revise accordingly (sex versus gender).

12) Table 2: Please indicate the meaning of the values following the food additive in the table heading.

13) Figure 1: Please include a definition of the five mixtures in the figure description. Please be sure to define all abbreviations (SD, BMI, CI, etc.). Please add a title for the x-axis. Please replace the hyphens with commas.

14) Figure 2: The image quality is very poor and, for example, the y-axis title is barely readable. Please revise the axes titles. Please also add a definition of mixture 2 and 5.

15) Figures 1 and 2 appear rather small at the moment. Please revise, also with regard to the comment about image quality.

DISCUSSION

General guidance: Please present and organize the Discussion as follows: a short, clear summary of the article's findings; what the study adds to existing research and where and why the results may differ from previous research; strengths and limitations of the study; implications and next steps for research, clinical practice, and/or public policy; one-paragraph conclusion.

1) ll.431-434: Please ensure to indicate that these research findings result from experiments in mice. Please revise throughout.

2) l.443: “type 2 diabetes patients” - PLOS Medicine prefers the use of patient-centered language, e.g. patients with type 2 diabetes. Please revise throughout.

3) l.522: : Please temper claims of primacy of results by stating, "to our knowledge" (or something similar).

4) Please remove any subheadings.

REFERENCES

1) Where website addresses are cited, please use the word ‘accessed’ when specifying the date of access (e.g. [accessed: 12/06/2024]).

2) Please make sure to check and update the references where necessary (e.g. [9]).

SUPPLEMENTARY MATERIAL

1) Please ensure that all supplementary files are referenced in the main text.

2) In the published article, supporting information files are accessed only through a hyperlink attached to the captions. For this reason, you must list captions at the end of your manuscript file. You may include a caption within the supporting information file itself, as long as that caption is also provided in the manuscript file. Do not submit a separate caption file.

When SI files are contained with a single file:

Please label the file as ‘S1 Supporting Information’.

Please apply alphabetical labelling to each table and figure contained within the S1 file. For example, ‘Fig A’ to ‘Fig Z’ and ‘Table A’ to ‘Table Z’.

Plain text does not need to be labelled and can just be given a title as necessary. For example, ‘Statistical Analysis Plan’.

Please cite tables/figures as ‘Fig A in S1 Supporting Information’ and/or ‘Table A in S1 Supporting Information’, for example.

Please cite plain text as, ‘Statistical Analysis Plan in S1 Supporting Information’, for example.

When SI files are uploaded as separate files:

Please label tables as ‘S1 Table’ (so on) and figures as ‘S1 Fig’ (and so on).

Any additional documents (protocols/analysis plans etc.) can be labelled as ‘S1 Protocol’, for example. Please cite items as exactly as labelled.

------------------------------------------------------------

General Editorial Requests

1) We ask every co-author listed on the manuscript to fill in a contributing author statement. If any of the co-authors have not filled in the statement, we will remind them to do so when the paper is revised. If all statements are not completed in a timely fashion this could hold up the re-review process. Should there be a problem getting one of your co-authors to fill in a statement we will be in contact. YOU MUST NOT ADD OR REMOVE AUTHORS UNLESS YOU HAVE ALERTED THE EDITOR HANDLING THE MANUSCRIPT TO THE CHANGE AND THEY SPECIFICALLY HAVE AGREED TO IT.

2) Please ensure that the paper adheres to the PLOS Data Availability Policy (see http://journals.plos.org/plosmedicine/s/data-availability), which requires that all data underlying the study's findings be provided in a repository or as Supporting Information. For data residing with a third party, authors are required to provide instructions with contact information for obtaining the data. PLOS journals do not allow statements supported by "data not shown" or "unpublished results." For such statements, authors must provide supporting data or cite public sources that include it.

3) Please review your reference list to ensure that it is complete and correct. If you have cited papers that have been retracted, please include the rationale for doing so in the manuscript text, or remove these references and replace them with relevant current references. Any changes to the reference list should be mentioned in the rebuttal letter that accompanies your revised manuscript.

Decision Letter 3

Alexandra Tosun

28 Feb 2025

Dear Dr Payen de la Garanderie, 

On behalf of my colleagues and the Academic Editor, Fumiaki Imamura, I am pleased to inform you that we have agreed to publish your manuscript "Food additive mixtures and type 2 diabetes incidence: Results from the NutriNet-Santé prospective cohort" (PMEDICINE-D-24-03030R3) in PLOS Medicine.

I appreciate your thorough responses to the reviewers' and editors' comments throughout the editorial process. We look forward to publishing your manuscript, and editorially there are only a few remaining minor stylistic points that should be addressed prior to publication. We will carefully check whether the changes have been made. If you have any questions or concerns regarding these final requests, please feel free to contact me at atosun@plos.org.

Please see below the minor points that we request you respond to:

1) Table 1: Please change the baseline characteristic 'Women' to 'Sex, Female'.

2) Table 2: Please add footnote 'a' below the table before 'The loading values...'.

3) Figure 2: The image is still quite small. Please define the meaning of the dashed line in the image description (define 'CL').

Before your manuscript can be formally accepted you will need to complete some formatting changes, which you will receive in a follow up email (including the editorial points above). Please be aware that it may take several days for you to receive this email; during this time no action is required by you. Once you have received these formatting requests, please note that your manuscript will not be scheduled for publication until you have made the required changes.

In the meantime, please log into Editorial Manager at http://www.editorialmanager.com/pmedicine/, click the "Update My Information" link at the top of the page, and update your user information to ensure an efficient production process. 

PRESS

We frequently collaborate with press offices. If your institution or institutions have a press office, please notify them about your upcoming paper at this point, to enable them to help maximise its impact. If the press office is planning to promote your findings, we would be grateful if they could coordinate with medicinepress@plos.org. If you have not yet opted out of the early version process, we ask that you notify us immediately of any press plans so that we may do so on your behalf.

We also ask that you take this opportunity to read our Embargo Policy regarding the discussion, promotion and media coverage of work that is yet to be published by PLOS. As your manuscript is not yet published, it is bound by the conditions of our Embargo Policy. Please be aware that this policy is in place both to ensure that any press coverage of your article is fully substantiated and to provide a direct link between such coverage and the published work. For full details of our Embargo Policy, please visit http://www.plos.org/about/media-inquiries/embargo-policy/.

Thank you again for submitting to PLOS Medicine. We look forward to publishing your paper. 

Sincerely, 

Alexandra Tosun, PhD 

Associate Editor 

PLOS Medicine

Associated Data

    This section collects any data citations, data availability statements, or supplementary materials included in this article.

    Supplementary Materials

    S1 Checklist. STROBE checklist.

    (DOC)

    pmed.1004570.s001.doc (94.5KB, doc)
    S1 Appendix. eMethods. eResults.

    Fig A. Flowchart, NutriNet-Santé cohort, 2009-2023. Table A. Daily food additive intakes (mg/d) among study participants from the NutriNet-Santé cohort, 2009–2023 (N = 108,643)a,b. Table B. Food additive mixtures identified by nonnegative matrix factorizationa. Table C. Consensus map for rank number determination and sensitivity analyses using other decomposition algorithms in the NMF procedure. Table D. Spearman correlations between the five NMF food additive mixtures. Table E. Spearman correlations between NMF food additive mixtures and food group intakes. Table F. Food group consumption of participants according to sex-specific quintiles of mixtures 2 and 5a. Fig B. Correlations between Schoenfeld residuals and timescale (age, y) from multivariable Cox models between food additive mixtures and type 2 diabetes incidence, NutriNet-Santé cohort, 2009–2023 (n = 108,643). Fig C. Dose-response associations between food additive mixtures and type 2 diabetes incidence, restricted cubic spline plots, NutriNet-Santé cohort, 2009–2023 (n = 108,643 participants; 1,131 incident cases). Table G. Associations between food additive mixtures and type 2 diabetes incidence, NutriNet-Santé cohort, 2009–2023—Sensitivity analyses. Table H. Association between food additive mixtures coded as tertiles and incidence of type 2 diabetes, NutriNet-Santé cohort, 2009–2023—Sensitivity analyses. Table I. Associations between food additive mixtures and incidence of type 2 diabetes, stratified by the Programme National Nutrition Santé-Guidelines Score 2 (PNNS-GS2), NutriNet-Santé cohort, 2009-2023—Sensitivity analyses. Table J. Association between mixtures 2 and 5 and type 2 diabetes incidence adjusted for the key food additives contributing to each mixture (residual method), NutriNet-Santé cohort, 2009–2023 (n = 108,643 participants; 1,131 incident cases) —Sensitivity analyses. Table K. Interactions between the key food additives contributing to mixtures 2 and 5, NutriNet-Santé cohort, 2009–2023 (n = 108,643 participants; 1,131 incident cases). Table L. Mediation analyses.

    (DOCX)

    pmed.1004570.s002.docx (3MB, docx)
    Attachment

    Submitted filename: garanderie.pdf

    pmed.1004570.s004.pdf (54.1KB, pdf)
    Attachment

    Submitted filename: Response to reviewer.docx

    pmed.1004570.s006.docx (102.7KB, docx)
    Attachment

    Submitted filename: PMEDICINE-D-24-03030_R2 - Point-by-point responses.docx

    pmed.1004570.s007.docx (46.8KB, docx)

    Data Availability Statement

    Researchers from public institutions can submit a request to have access to the data for strict reproducibility analysis (systematically accepted) or for a new collaboration, including information on the institution and a brief description of the project to collaboration@etude-nutrinet-sante.fr. All requests will be reviewed by the steering committee of the NutriNet-Santé study. If the collaboration is accepted, a data access agreement will be necessary and appropriate authorizations from the competent administrative authorities may be needed. In accordance with existing regulations, no personal data will be accessible. R/SAS code is available without restrictions upon request at collaboration@etude-nutrinet-sante.fr.

    Articles from PLOS Medicine are provided here courtesy of PLOS

    RESOURCES