Rationale and conceptual considerations
We built a comprehensive and globally applicable inventory of all known plastics chemicals and developed an innovative framework combining hazard and grouping components with the aim to facilitate a transition towards safer plastics. We define plastic chemicals as all chemicals that can be present in plastic materials and products, including the polymer backbone, intentionally added substances (that is, starting substances, processing aids and additives) and NIASs (for example, impurities, unreacted intermediates, reaction by-products and degradation products). The fact that some plastic chemicals may have additional uses in non-plastic applications has no bearing on their inclusion. Contaminants sorbing to plastics during use or end-of-life are not considered plastic chemicals. See complete definitions in Supplementary Text 1.
To identify chemicals of concern in plastics, we developed and applied a hazard-based approach. In brief, we classified a plastic chemical as of concern on the basis of the criteria of persistence, bioaccumulation, mobility and toxicity. The rationale for taking this approach is threefold. First, we argue that plastics should not contain hazardous chemicals, to protect human health and the environment. Although this approach might be considered conservative, it enables scientists and businesses to improve the safety of plastics and proactively prevent potential harm in a timely and efficient manner. Second, we posit that exposure to most plastic chemicals is probable given that most are not covalently bound to the polymer. This is backed by empirical evidence showing that 83% of the 1,788 chemicals tested for release from plastic food contact materials indeed migrate into foodstuff17,25. Third, we reason that an alternative risk-based approach is neither efficient nor effective—and often infeasible—to identify chemicals of concern owing to numerous practical challenges, blind spots and high implementation costs that would unduly delay a transition to safer plastics.
Apart from the identification of chemicals of concern, we collected additional evidence on the use, presence and release of plastic chemicals. We use this information to link chemicals to specific polymer types, shed light on which compounds are intentionally added to plastics (commercial use) and infer their exposure potential. Although the last of these is only qualitative, it suggests that human and environmental exposures to these chemicals are probable, given that scientific studies demonstrate their migration or volatilization into foodstuff or environmental media.
We grouped plastic chemicals on the basis of their structure to efficiently address chemicals of concern as well as regrettable ‘drop-in’ substitutions (that is, replacing a chemical of concern with a structurally very similar compound). We assumed a chemical without hazard data is ‘guilty by association’ if it is part of a group that consists of many known chemicals of concern. This is supported by ample evidence available for certain groups, such as bisphenols50, as well as the fact that the structure of a chemical determines its hazards. In combination with the hazard-based approach, such chemical grouping offers a robust framework to improve the chemical safety of plastics in a timely and efficient manner.
Building the PlastChem inventory
We built the PlastChem inventory by: compiling and harmonizing information from seven prior databases; retrieving additional and updated information on hazard classifications and regulatory status from 15 authoritative sources (latest update August 2023) to identify chemicals of concern; integrating data on the use, detection and release of chemicals in and from plastics; and grouping the chemicals on the basis of their structure (Extended Data Fig. 1). All data engineering work was performed using R (R Consortium, 2023) and Microsoft Excel (see Data and Code availability).
Core of the PlastChem inventory
We compiled and harmonized information from prior databases that had identified chemicals in plastics, relying on six peer-reviewed and one regulatory source. As a first step, we retrieved lists of plastic chemicals from seven sources, including: the dataset of ref. 27; the database of Chemicals associated with Plastic Packaging (CPPdb, version 1)28; the European Union (EU) list of Authorized Substances Annex I, Plastic Food Regulation 10/2011/EU (PFCRdb)31; the Food Contact Chemicals database (FCCdb, version 5)29; the PlasticMAP dataset26; the dataset on Migrating and Extractable Food Contact Chemicals (FCCmigex, version February 2023)17; and the LitChemPlast dataset (version June 2023)25. Given the scale of the results, we deemed verifying the original sources not feasible and accepted the entries in the seven databases as plastic chemicals.
To identify chemical structures, compile data from various sources and assemble the inventory, we used CAS RNs or chemical names (when CAS RNs were invalid or unavailable) with the automatic Application Programming Interface services of PubChem. The information retrieved included the PubChem Compound Identification (CID), molecular formula, molecular weight, canonical SMILES, isomeric SMILES, InChI, InChIKey, IUPAC name, predicted octanol/water partition coefficient (XLogP), exact mass, monoisotopic mass, topological polar surface area, complexity and charge. CAS RNs were validated according to CAS guidelines51. Further details are available in Supplementary Text 2.
Despite thorough quality assurance and control, some uncertainties may remain in the inventory (Supplementary Text 2). For example, 400 duplicates were identified, originating from the use of multiple names for the same chemical or from non-unique CAS RNs that correspond to multiple substances. These duplicates are shown in an additional file in the inventory. Of these, 188 chemicals were manually selected by expert judgement for inclusion in the main inventory, as they were deemed the most likely matches for the corresponding CAS RNs. Although we validated and curated the CAS RNs to reduce duplication, additional duplicates may still exist.
Information on functions and regulation
Additional information on the known functions of plastic chemicals and their regulatory status is also included in the PlastChem inventory. Briefly, the functions were gathered from ref. 27, CPPdb28 and PlasticMAP26, aligned to a common terminology and assigned larger functional classes (starting substances, additives, processing aids and NIASs). See further details in Supplementary Text 3.
To integrate regulatory information, we compiled information from the Basel, Stockholm and Minamata Conventions, along with the Montreal Protocol, to identify the chemicals globally regulated under Multilateral Environmental Agreements at present. In addition, we also included regional and national lists of chemicals that were easily accessible, including chemicals regulated in the EU, Japan, the Republic of Korea and the USA, along with those substances regulated under the Rotterdam Convention. See further details in Supplementary Text 6.
Identification of chemicals of concern
Hazard information was collected from 15 regulatory sources, some of which aligned with the Globally Harmonized System of Classification and Labelling (Supplementary Table 4). This information is consistent with the principles outlined in the EU Chemicals Strategy for Sustainability52 and the updated Classification, Labelling and Packaging of Substances and Mixtures regulation36. We did not include hazard information from the scientific literature because relevant studies are either already synthesized in the regulatory sources or would require individual quality assessment, data extraction and analysis. Accordingly, we consider the regulatory sources used as authoritative.
We applied a stepwise process to harmonize the compiled hazard information and translate it into a single hazard score per chemical. First, chemicals with hazard data for a specific trait (for example, carcinogenicity) were assigned a score ranging from 0 to 2, with 0 indicating non-hazardous, 0.5 indicating less hazardous, 1 indicating hazardous, and 2 indicating very hazardous (Supplementary Table 5). Chemicals with hazard evaluations under development, postponed or pending received a score of 0.25 to reflect that these are undergoing assessment but have not been classified yet. A blank indicates that no information is available. In cases of multiple classifications per source, we used the highest scores for each trait, with regulatory information taking precedence over industry data. As a second step, we aggregated the individual scores per trait into a single score for each hazard criterion (P, B, M and T) using the highest score whenever multiple traits within a criterion had information, retaining the 0–2 scoring system as described above. Finally, the maximum score across the P, B, M and T criteria resulted in a unique hazard score for each chemical that we used to identify chemicals of concern when the score was ≥1, indicating that a chemical of concern fulfils at least one hazard criterion. Additionally, we calculated the sum of hazard scores across the P, B, M and T criteria, ranging from 0 to 8, and provide an evidence score reflecting the number of sources that classify a chemical as hazardous for the same criterion, ranging from 0 to 31. Further details are available in Supplementary Text 4.
Use, presence and release information
We compiled and harmonized data on the use, presence and release of chemicals in plastic polymers from FCCmigex17, PlasticMAP26 and LitChemPlast25. Presence data were derived from studies performing extraction experiments (for example, with solvents); release data came from studies performing migration experiments (for example, with food simulants). Chemicals with release data were assigned a score of 2, and those with only presence data received a score of 1. Data on the commercial use of plastic chemicals were sourced from PlasticMAP and given lower priority (score of 0.5 if no presence or release data are available). Chemicals with inconclusive information were scored 0.25, and substances that were analysed but not detected in experimental studies were scored 0. Finally, chemicals without data received a blank. See Supplementary Text 5 for further details.
Grouping and groups of concern
Plastic chemicals were grouped on the basis of their structures following two approaches (details in Supplementary Text 7). In brief, the first approach consisted of searching a predefined set of keys in the chemical names to identify inorganic compounds, organometallics and metalorganics, UVCBs, polymers and mixtures. This was followed by searching the name and chemical symbol of respective elements in the chemical names and SMILES to identify organohalogens, organophosphates, organosilicons and various chemicals containing certain metal(loid) elements. The second approach consisted of matching chemicals in the PlastChem inventory with existing lists of chemical groups (list in Supplementary Text 7). In addition, we applied expert judgement to identify PFASs from all organofluorine chemicals, as well as those PFASs and chlorinated paraffins regulated under the Stockholm Convention.
Groups of plastic chemicals were prioritized on the basis of the number of chemicals of concern they contain. As some groups of chemicals lack hazard data, our prioritization considers only groups with sufficient information. Starting with more than 100 initial groups, we isolated relevant groups by excluding groups that were too large and unspecific (for example, mixtures), groups that are regulated internationally, and groups with too few members. The final selection of prioritized groups was ranked on the basis of the proportion of known chemicals of concern they contain. This follows the rationale that groups consisting of many hazardous chemicals have more evidence for being concerning. Priority was assigned to groups for which ≥40% of the members are chemicals of concern or if additional scientific considerations pointed towards group-specific hazards (details in Supplementary Text 6 and Supplementary Table 9).
During the grouping, we also identified some technical challenges. Notably, no single automated tool is available to consistently group large numbers of chemicals. Instead, a combination of various techniques is required. For example, we used an algorithm53 to identify homologue series containing repeating units of –CH2–, –CF2–, –CF2O– or –CF2CF2O–, covering more than 2,000 substances. Additionally, it is possible to identify UVCBs, polymers and other mixtures using predefined keyword searches47. However, in both cases, the subsequent assignment into more specific groups (for example, PFASs) requires manual inspection and expert judgement. This is due to a lack of machine-processable structural identifiers for these substances. Therefore, the manual curation of grouping and certain entries may have introduced some inconsistencies in our inventory. More work is needed to improve the availability of unique structural identifiers for chemicals on the global market, and to develop cheminformatics tools for grouping.
