Plastic Food Packaging from Five Countries Contains Endocrine- and Metabolism-Disrupting Chemicals

Plastics are complex chemical mixtures of polymers and various intentionally and nonintentionally added substances. Despite the well-established links between certain plastic chemicals (bisphenols and phthalates) and adverse health effects, the composition and toxicity of real-world mixtures of plastic chemicals are not well understood. To assess both, we analyzed the chemicals from 36 plastic food contact articles from five countries using nontarget high-resolution mass spectrometry and reporter-gene assays for four nuclear receptors that represent key components of the endocrine and metabolic system. We found that chemicals activating the pregnane X receptor (PXR), peroxisome proliferator receptor γ (PPARγ), estrogen receptor α (ERα), and inhibiting the androgen receptor (AR) are prevalent in plastic packaging. We detected up to 9936 chemical features in a single product and found that each product had a rather unique chemical fingerprint. To tackle this chemical complexity, we used stepwise partial least-squares regressions and prioritized and tentatively identified the chemical features associated with receptor activity. Our findings demonstrate that most plastic food packaging contains endocrine- and metabolism-disrupting chemicals. Since samples with fewer chemical features induce less toxicity, chemical simplification is key to producing safer plastic packaging.


Figures S1 to S10
Tables S1 to S13 Supplementary references Other supplementary materials for this manuscript include the following: Mass spectral data of all samples can be assessed under https://doi.org/10.18710/LZNLFXS1 Supporting methods and materials

S1.1 Determination of polymer type
Where not stated on the packaging, the polymer types of the FCA were determined with Fouriertransformed infrared spectroscopy (Cary 610-FTIR and Resolutions Pro software, Agilent) in attenuated total reflectance and transmission mode for the expanded polymers (LDPE 8, PS 5a and b).The polymer type was determined for the inside and outside of the samples.Spectra were acquired with a resolution of 4 cm -1 and 32 scans per spectrum and matched to the library of Primpke et al. 1 Differentiation between HDPE and LDPE products was done by determining the melting point with differential scanning calorimetry (TGA/DSC 1 thermogravimeter, Mettler Toledo) in the temperature range of 25-180 °C.The measurements were carried out in an N2 atmosphere with a 50 mL min -1 flow rate and a 10 K min -1 heating rate.

S1.3 Chemical analysis
For chromatographic separation a 38-min linear gradient program was used (Table S1) and compounds eluting between 1.5 and 36 min were included in the analysis.The flow rate was set to 0.2 mL min -1 , the column temperature was maintained at 55°C and the injection volume was 2 μL.After each sample a 7 min wash step with 100% methanol at 0.35 mL min -1 was included, followed by a 3 min equilibration to the initial column conditions.The collision energy ramped from 15 to 45 V, and the scan time was 0.3 s (details in Table S3).The analysis included all samples, PBs 1-4, the solvents used during extraction (methanol evaporated to DMSO), and 7 quality controls (pool of all PUR and PVC or PP, PE, PET, and PS samples).
Data treatment and compound identification was performed with the software Progenesis QI (Nonlinear Dynamics, version 3.0).The retention times were automatically aligned to one of the quality controls (pool of all samples) selected by the software.Peak picking was performed with automatic sensitivity to detect "fewer" peaks.The minimum chromatographic peak width was set to 0.05 min and the fragment sensitivity was set to 0.2% of the base peak.The program searched for common adducts (M+H, 2M+H, M+2H, M+H-H2O, M+Na, and 2M+Na), and deconvoluted the mass spectra.The resulting list of chemical features was exported to Microsoft Excel for Windows (Version 2021-2306).
The R package "pheatmap" 2 was used for the hierarchical cluster analysis (Euclidian distances) of the chemical features (Figure 2C and 4I).Euler diagrams were produced with the R package "eulerr" (Figure 3 and S4). 3 Table S2.Chromatographic parameters for the nontarget chemicals analysis.Table S3.Mass spectrometer parameters for the nontarget chemicals analysis.

S1.4 Compound identification, ToxCast and PubChem search
Identification was done with the Metascope algorithm in Progenesis QI.For the in silico fragmentation, four databases were used as previously described. 4In addition, we constructed a fifth database containing the plastic chemicals described by Wiesinger et al. 5 For their PlasticMap database, we converted the available 10 547 CASRNs to SMILES-codes (n = 6631) using the EPA CompTox batch search. 6The SMILES codes were then converted to CIDs (n = 6423) using the PubChem Identifier Exchange service 7 .CIDs were uploaded to PubChem to retrieve the chemical structures (n = 6414) as a sdf file which was subsequently used for the in silico fragmentation search.
Among the tentatively identified compounds 304 were present in the ToxCast data based on matching InChlKeys, IUPAC names or common names.We extracted their activity concentrations 50 (AC50) in 45 ToxCast assays that matched the receptors analyzed here (Table S4).ToxCast assays were selected based on the receptor gene names.The AC50 values were compared to the compound's highest peak area and ranked based on the AC50/area ratio, for all compounds with a peak area >100 (Table S9).A Spearman rank correlation between number of compounds activating a certain receptor and the samples effect concentrations was calculated with GraphPad Prism (9-10).
For the tentatively identified compounds among the ten features with the highest abundance per sample, an additional search for toxicity and usage information was done on PubChem, 8 and in the supplementary information from Wiesinger et.al. and Groh et al. 5,9 Table S4.Selected ToxCast assays used to retrieve information about PXR, PPARγ, estrogenic and anti-androgenic activity.Cytotoxic chemicals were least prevalent in our samples (Figure S3): Five of the FCA extracts induced significant cytotoxicity, including two freezer bags (LDPE 3 and 5), two hydration bladders (PUR 1 and 2), and a cling film (PVC 4).The two PUR hydration bladders induced the strongest cytotoxicity (EC20 0.05 and 0.06 mg/well).Cytotoxicity correlates significantly with the number of chemical features (Figure S4) but this correlation is mostly driven by the two PUR samples.Nonetheless, cytotoxicity is mostly thought of as a result of non-specific effects that will accumulate when more chemicals are present in a sample.

S2.3 Tentatively identified chemicals
Most of the chemicals were identified with the NORMAN Suspect List Exchange database (77% in the PE, PET, PP and PS samples and 91% in the PUR and PVC samples), 11 which is the most comprehensive database used in this study and includes about 70,000 compounds but is not specific to plastics.
Of the ten most abundant features per sample, we tentatively identified 69 compounds and retrieved use and toxicity information from Pubchem.Our analysis showed that 31 of these compounds were associated with plastics, including colorants, plasticizers, flame retardants, antioxidants, and other processing aids and further 12 are likely used in plastics (Table S8).Among the remaining features, 12 have no relevant use or toxicity data while 14 are not related to plastic and consist of cosmetics, pharmaceuticals, and fungicides among others (Table S8).Examples of plastic additives detected are erucamide (CAS 112-84-5) and octadecanamide (CAS124-26-5), which are used as colorants, fillers, and lubricants. 5These substances were found with high abundance in the three LDPE freezer bags (LDPE 2, 3 and 5).Further, we detected the NIAS ethylene terephthalate cyclic trimer (CAS 7441-32-9) with high abundance in five of the six PET samples.This compound is known to occur in PET. 9 Among the plastic additives, we detected several known toxic and/or persistent and bioaccumulative compounds with high abundances, such as triphenyl phosphate (  It should be noted that the number of chemicals reported here is not comprehensive as the measurements were only performed using positive ionization mode.This means that negatively ionizable compounds were not detected, possibly leading to an underestimation of the actual number of chemicals present.Yet, the largest fraction of plastic related chemicals is positively ionizable. 13One-third of the samples analyzed had previous food content, which can alter the chemical composition to varying extents, as demonstrated by a comparison of three products analyzed both with and without previous food content.This could result in an overestimation; however, no general trend of higher numbers was observed in samples with previous food content.
Furthermore, the four samples with the highest number of features did not have previous food content.
The low identification rate, as achieved in this study, is a known challenge, especially with liquid chromatographic methods.The scarcity of publicly available information on plastic chemicals paired with technical challenges of identification of LC-MS data hinders a more comprehensive identification. 14While there is much uncertainty regarding the use of chemicals in plastics, about 44% of the top ten most abundant features in our samples were not clearly related to plastics based on the sources we consulted.These chemicals may have originated from the food content, be formed or introduced during plastic production (e.g., NIAS), are used in plastic production but not known to the public, or simply be misidentified. 5,15,16 another study which used a nontarget liquid chromatography coupled to high-resolution mass spectrometry, 17 76% of the identified chemicals in food packaging were classified as NIAS.In accordance with this, Geueke et al. reported that 65% of chemicals in food contact materials, including plastics, are not listed for their use in food contact applications. 18Nonetheless, 56% of the ten most abundant features were identified as chemicals probably used in plastics, indicating the suitability of this nontarget approach.

FigureFigureFigure S3 .
Figure S1.Dose-response relationships of the reference compounds used in the reporter gene assays.Nicardipine was used for the pregnane X receptor (PXR, A), rosiglitazone for peroxisome proliferator receptor gamma (PPARγ, B), 17-β estradiol for estrogen receptor alpha (ERα, C), and flutamide for the anti-androgenic activity (Anti-AR, D). n ≥ 60.Note: LOD = limit of detection

Figure S4 .Figure S5 .
Figure S4.Correlation between biological effects (EC20/50) and the number of chemical features detected in the plastic food contact articles.The values in the boxes represent Spearman's rank correlation coefficient, the asterisks indicate that the correlation was statistically significant (* p < 0.05, ** p < 0.01.*** p < 0.001).

Figure S6 .
Figure S6.Overlap of chemical features in the samples made of the same polymer (A) HDPE, (B) LDPE, (C) PET, (D) PP, (E) PS, (F) PVC, (G) PUR and PUR-PE composite.An overlap of <1% is not shown.

S2. 4
Limitations of the chemical analysis and identifications.

S2. 6
Figure S7.Comparison of samples without (a) and with (b) previous food content on chemical composition (A) and (B-M) receptor activity.Bioassay data are derived from at least three independent experiments, with four technical replicates per concentration (n ≥ 12).

Figure S8 .
Figure S8.Effect of previous food content on receptor activity and cytotoxicity.EC20/50 values calculated from a minimum of three independent experiments with four technical replicates per concentration (n ≥ 12).Mann-Whitney tests do not indicate significant differences (p > 0.05).

Figure S9 .Figure S10 .
Figure S9.Impact of country of purchase on receptor activity and cytotoxicity.EC20/50 values calculated from a minimum of three independent experiments with four technical replicates per concentration (n ≥ 12).Kruskal Wallis tests and Dunn's multiple comparison tests do not indicate significant differences (p > 0.05).

Table S5 .
Distribution of chemical features across samples.

Table S6 .
Chemical features in the FCAs made of the same polymer, total number, common features and features unique to a single product.
Note: a including the composite made of PUR and PE.

Table S7 .
Number of features and identified compounds detected in the samples.

Table S8 .
Tentatively identified chemicals among the top ten features per sample with use and toxicity data retrieved from Groh et al. (2019) comprising plastic chemicals specific to food contact articles (A), Wiesinger et al. (2021) comprising intentionally added plastic chemicals (B), PubChem (C) and ToxCast (D).

Table S9 .
PXR, PPARγ, ERα agonists and AR antagonists tentatively identified in the plastic extracts.Compounds ranked based on AC50/peak area per receptor.First 20 compounds per receptor are listed for others see TableS15.

Table S10 .
Number of known, active chemicals detected in the samples compared to the experimentally derived effect concentrations.

Table S12 .
). Kruskal-Wallis tests with Dunn's multiple comparison tests do not indicate significant differences (p > 0.05).Performance of the PLS regression of chemical features (abundance) and receptor activity (normalized EC20/50 values) detected in PE, PET, PP and PS FCAs and comparison of cross-validated (CV) root mean squared errors (RMSE) and CV determination coefficient (R 2 ) of the initial model 0 and optimized model (VIP filtered).Number of features included in the optimized model and corresponding values for the validation based on 500 iterations with randomly chosen variables.Model 0 included 2993 features.TableS13.AC50 values of tentatively identified compounds in the optimized PLS regression models positively correlated with the receptor activity.