Risk assessment of nitrate, nitrite and N-nitrosamines in food – Part 2

Intro This project is a part of the risk assessment of nitrate and nitrite for the Scientific Committee for Food and Environment (VKM). You can find this here: https://vkm.no/download/18.6eff4cf519e62f3ad8c388b3/1780377043253/Risk%20assessment%20of%20nitrate,%20nitrite,%20and%20nitrosamines%20-%20Part%202_Final_2026.pdf Dietary exposure assessment of nitrate, nitrite, and nitrosamines Dietary exposure to nitrate, nitrite, and nitrosamines was estimated using two-day dietary recall data from a Norwegian dietary survey (Norkost 4). The survey collected food consumption data on two non-consecutive days from each participant, with quantities reported in grams per day. Body weight was recorded to allow calculation of body weight-adjusted exposure (µg or ng per kg body weight per day). The same method were used to calculate the exposure for 4-, 9-, and 13-years old from Ungkost3. Food consumption data were organised by individual food items coded with survey-specific variable codes (V7001–V7041 in the demonstration dataset). A food code lookup table containing item codes, names, and descriptions was used to map individual food items to the relevant food groups for each substance. Matching was performed using keyword pattern searches on food codes and item descriptions. Concentration data for nitrate and nitrite (µg/g food) and nitrosamines (ng/g food) were compiled from publicly available sources, mainly from EFSA opinions. Concentrations were assumed to be lognormally distributed, and lognormal parameters (mu and sigma) were derived from summary statistics (mean, standard deviation). Exposure for each individual and food group was calculated using a probabilistic Monte Carlo simulation (MC_sim), drawing random concentrations from the lognormal distribution and multiplying by individual food consumption. Simulations were run separately for each food group and each dietary recall day. For nitrosamines, ten individual compounds were assessed: NDMA, NDEA, NMEA, NMOR (high carcinogenic potency) and NDBA, NDPA, NMA, NPYR, NPIP, NSAR (low potency, multiplied by a potency factor of 0.2 before summation). Compound-specific results were merged into a total nitrosamine exposure estimate expressed as NDMA equivalents. All analyses were implemented in R. The demonstration scripts use entirely fictional consumption data (30 individuals, 41 food variables) in place of the confidential survey data, while concentration data and analytical methods are identical to those used in the original analysis Files for exposure modeling The R codesCode/Exposure_nitrate_fiktiv_demo.RExposure_nitrite_fiktiv_demo.RExposure_nitrosamines_fiktiv_demo.RUtility_functions_nitrate_nitrite_nitrosamines_160126.R The data Nitrites_EFSA_2017.csvNitrites_EFSA_2017.xlsxall_nitrosamine_summaries.csvKroppsvekt_fiktiv.csvNorkost_4_dag1_fiktiv.csvNorkost_4_dag2_fiktiv.csvall_nitrosamine_summaries_fiktiv.csvcode_norkost4_fiktiv.csv Nitrate Nitrite & Nitrosamines modelling As a part of the project we looked for models that estimate the endogenous transformation between nitrate and nitrite, and also look for models that can estimate the conversion of nitrite to nitrosamines (NDMA used as a surrogate). Both Cheng et al 2021 and Zeilmaker et al 2010 have published models for endogenous conversion of nitrate (NO3) and nitrite (NO2). The model of Zeilmaker also seems to be able to convert the nitrate to the nitrosamine NDMA. However, the model structure seems very similar and Cheng et al also published the code in Berkley Madona. We have translated this model into R. EFSA used a model to translate endogenous nitrite to the nitrosamine NDMA, on page 103. This model has been implemented into R, and the script are in the Code folder.EFSA 2017 Risk assessment of nitrite TKTD model for nitrite formation from nitrate, and nitrosamines from nitrite. This R Markdown script implements a probabilistic toxicokinetic–toxicodynamic (TKTD) model for nitrite formation in adults, then derives formation of NDMA in the stomach. Data and setup The script loads rxode2, EnvStats, dplyr, ggplot2, and tidyr, then reads a CSV containing calculated exposure of nitrate by VKM, nitrate and nitrite exposure, sex, body weight, height, and blood volume. Core TKTD model The function calcul() defines parameter distributions for saliva flow, nitrate and nitrite absorption/elimination, distribution volumes, conversion between nitrate and nitrite in saliva, urinary elimination, endogenous nitrate production, and parameters for hemoglobin/methemoglobin kinetics. Using rxode2, it sets up a system of ODEs for nitrate and nitrite in absorption, central blood, saliva compartments, and for hemoglobin and methemoglobin, then simulates 48 hours after a single oral co‑dose of nitrate and nitrite given at time 100 h. Output of the TKTD model In calcul(), 1000 parameter sets are generated (Monte Carlo) per individual, the ODE system is solved, and the script extracts DD_Ni at time 148 h (48 h after dosing) as a metric of cumulative nitrite in the stomach region. The lapply over dataset_lb_sum$ni runs this model for each individual, returning a list of 1000 simulated values of the nitrite metric per person. NDMA formation and MoE Using the simulated nitrite metric, the script assumes all stomach nitrite reacts with dimethylamine (DMA) to form NDMA, with fixed stomach volume, daily DMA intake, and a nitrosation rate constant taken from guideline data. It computes a daily NDMA dose (mg/kg bw/day) for each simulation and individual, then divides a reference BMDL10 for NDMA (0.027 mg/kg bw/day) by these doses to obtain MoE values, again 1000 per person. Visualization and diagnostics Additional code (calcul_with_ts) re‑runs the TKTD model while keeping full time courses, then summarizes and plots median and 5–95% bands over time for central nitrate, nitrite, hemoglobin, and methemoglobin for a chosen individual (e.g., person 2). The script also produces violin/boxplots for the stomach nitrite metric, NDMA dose, and MoE per person, an overall MoE ECDF with a reference line at MoE = 10,000, and simple checks like plotting body weight versus median NDMA dose with labels Sensitivity analysis Sensitivity analysis are done given different parameters, additional functions are given for visualization. The R codesNiNa_kinetic_model_MoE-25022026_0_ni.RmdSurface_plot.R The datalb_sum3-VKM-ikke-dummy-0-Ni.csvExtracted_data_fig6.xlsx FoodChem PNCs StatModelling Estimation of predicted net change (PNC) factors quantifying processing-induced concentration changes between nitrates (NO₃⁻), nitrites (NO₂⁻) and nitrosamines (NAs) in foods under realistic processing and storage conditions. We present here the modelling framework used to estimate predicted net change (PNC) factors describing processing-induced concentration changes for the apparent NO₃⁻→NO₂⁻ and NO₂⁻→NAs pathways in different foods under realistic storage and heat-treatment conditions. The analysis was based on paired before–after concentration measurements, where each observation linked the initial and processed concentrations for the same food product and processing condition. Linear mixed-effects (LME) models were applied to estimate pathway-specific PNC factors, using process type, temperature and processing time as fixed effects and random intercepts for food products to account for product-specific variability. Bootstrap resampling was used to derive confidence intervals around the estimated PNC factors. The PNC factors do not represent true mechanistic conversion fractions, but empirical, exposure-oriented factors that quantify the observed net concentration change relative to the relevant precursor concentration. These factors can be applied to update occurrence concentrations after processing and to evaluate the potential impact on dietary exposure estimates. Although the available paired datasets are limited, the LME framework provides a stable and interpretable approach for estimating processing-related concentration changes and their uncertainty across food products and processing conditions. Further development on this model is performed at the repository FoodChem_PNCs_StatModelling