A chemical structure-based approach for estimating the added levels of flavourings to foods for the purpose of assessing consumer intake

ABSTRACT

Intake assessment and hazard profile of chemical substances are the two critical inputs in a safety assessment. Human intake assessment presents challenges that stem either from the absence of data or from numerous sources of variability and uncertainty, which have led regulators to adopt conservative approaches that inevitably overestimate intake. Refinements of intake assessments produce more realistic estimates and help prioritise areas of concern and better direct investment of resources. However, use levels (ULs), which represent the usual added amount of flavourings to food products, are the starting point for refined intake assessments, are data-intensive, and data availability is often a limitation. The work presented here was undertaken to investigate the use level patterns of substances used as flavourings in foods and to develop a systematic tool for data extrapolation based on chemical structure. The available dataset consists of use levels reported through eight industry surveys and hence are representative of industry uses rather than regulatory limits, which are higher by design and not realistic. A systematic statistical analysis was undertaken to determine whether the industry-reported UL data can be used to estimate use levels of flavouring substances belonging to the same chemical group for which such data are not available. Predictive modelling approaches were explored to evaluate relationships in the data and utilised additional variables relevant to technological considerations, such as volatility losses upon heat treatment, and Tanimoto index-based pair-wise structural similarity scores to determine whether more granular similarity information can reduce the within-group variability. The analyses indicated that the use levels of flavouring substances can reasonably be estimated based on the available data using chemical group classifications stratified by food category. Source of uncertainty and limitations are discussed.

KEYWORDS:

• Flavouring
• intake
• chemical structure
• use level

Acknowledgments

The authors thank Drs. Kevin Renskers, Ben Smith and Shim-mo Hayashi for their review and constructive input in the preparation of the manuscript. The authors also thank Michael Armesto and Vivian Lu for their support in assembling the data.

Declaration of Interests

This work was supported by the International Organization of the Flavor Industry. The authors declare that there are no conflicts of interest.

Supplemental material

Supplemental data for this article can be accessed on the publisher’s website.

Notes

1. The maximised survey-derived daily intake (MSDI) is a method for intake estimation based on the amount of a flavouring that disappears into the food supply of a defined population size over a specified period of time, and is typically based on the annual volume of a flavouring in the market for use in food. The conservative estimate assumes that the total volume disappearing in the market is consumed by only 10% of the population (consumers only).

2. The series of FEMA numbers from 2001 to 3250 excludes 21 FEMA numbers: 2016, 2111, 2126, 2168, 2226, 2227, 2405, 2433, 2524, 2566, 2606, 2758, 2786, 2990, 3006, 3015, 3208, 3211, 3216, 3234, 3248 for which there were no reported data.

3. FACET software was developed by Crème Global in collaboration with industry, government and academia as a harmonised methodology to monitor consumer exposure in Europe.

4. CG 3: α,β-Unsaturated (alkene or alkyne) straight-chain and branched-chain aliphatic primary alcohol/aldehyde/acid, acetal, and ester with ester containing α,β -unsaturated alcohol and acetal contains α,β -unsaturated alcohol or aldehyde. No aromatic or heteroaromatic moiety as a component of an ester or acetal. No difunctional oxygenised alcohol or acid in an ester or acetal.

5. CG 5: Saturated and unsaturated aliphatic secondary alcohol/ketones/ketals/esters with ester containing secondary alcohol. No aromatic or heteroaromatic moiety as a component of an alcohol, ketone, ester or ketal.

6. CG 8: Secondary alicyclic saturated and unsaturated alcohol/ketone/ketal/ester with and ketal containing alicyclic alcohol or ketone with ester or ketal containing secondary alicyclic alcohol. Ester/ketal may contain aliphatic acyclic or alicyclic acid component.

7. This baseline model attempts to estimate the (logarithm of) use level according to CG and tier-2 FC. If tier-2 training data is not available, then tier-1 data is used. If no appropriate FC data is available during training, the mean use level in the entire CG is used for prediction.

8. Comparing cross-validation results to those produced with held-out data is a useful diagnostic; during modelling, for example, a mismatch between the two measures identified a coding error.

9. If a substance appears multiple times in an FC – e.g., it is reported across multiple tier-2 FCs – then it is counted multiple times in the MAE because each entry is treated as a new prediction.

Additional information

Funding

This work was supported by the International Organization of the Flavor Industry.