Nontarget Screening and Health Risk of Hazardous Volatile and Semivolatile Organic Compounds in Premade Food Using a Molecular Network Assisted Data Collation Workflow

Abstract

Premade foods have been popular in recent years, but compounds from foods and their packaging may be harmful. A nontargeted screening method was developed for analysis of volatile and semivolatile organic compounds. Fully automated data for molecular network visualization was developed to automatically identify natural substances and substances permitted for addition to food. The workflow tackles the inefficiency of manual processing for massive raw data, which successfully narrowed down over 1,800 raw compounds to 80 for final manual inspection. Hazardous compounds like polycyclic aromatic hydrocarbons, phthalate esters, bisphenol, and organophosphate ester were successfully identified. In addition, other functional chemicals like antioxidants or preservatives were also identified to exist in food, among other suspicious nonintentional compounds, and a lot of them were outside current suspect lists, posting for further concern. The overall intake risk was estimated to be low, but given the current trend toward stricter regulations on bisphenols, the detection of bisphenols in some food samples remains a cause for concern. During chemical prioritization, 1,2-diphenylcyclopropane (PAHs transform product), 3-acridinol, and 9-phenyl (resin making chemical) ranked high due to high toxicity. Nontargeted analysis, combined with our robust data mining and data trimming workflows, can serve as a powerful tool for food safety assessments.

1. Introduction

Oral intake is one of the most crucial pathways for chemicals to enter living organisms, with diet being the primary route. Consequently, food safety has consistently been a focal point in relevant research fields. The cumulative effect within natural food chains amplifies the total amount of environmental chemicals ingested by humans. There have been reports on intake risk on toxic elements through aquatic products, and organic substances like antibiotics, endocrine-disrupters, and per- and polyfluoroalkyl substances (PFAS) have also been proven to enter the human body via animal-derived foods through bioaccumulation. Food processing also leads to a variety of unexpected substances in the human body, like high amounts of microplastics introduced during manufacture, food additives, and harmful chemicals like furan and acrylamide which can also formulate during cooking. Precooked and easily served, premade foods have now become a convenient food option for today’s fast-paced society, taking their place in the retail market and takeout sector. Due to the common practice of extending shelf life in prepared foods through measures such as adding exogenous substances or undergoing preservation treatments, their complex manufacturing processes may introduce unintended harmful substances, necessitating careful attention to their safety.

Food Contact Materials (FCMs) like packaging are an important source of unintentional detrimental substances in food content, which can potentially increase intake risk. Processing aids have been shown to migrate into food. Phthalic acid esters (PAEs) and their derivatives, used as plasticizers, have been widely detected in plastic-packaged foods. Photoinitiators (such as benzophenones) in paper food packaging also pose migration risks to food. Bisphenol A (BPA) in used for the manufacture of epoxy resin packaging materials can significantly increase health risks if improperly handled during production. Meanwhile, premade food packaging is typically designed to be more durable than traditional food packaging to preserve their more complex contents, and facilitate direct heating with boiling water, and even open flame. Substances intentionally or unintentionally used during production, along with those generated subsequently, may migrate into the food contents, among which there can be overlooked substances which lead to undesirable health effects if ingested by humans.

In order to identify the presence of regulated substances and to ensure the safety of the FCMs used, a variety of methods have been employed, including testing with standard migration test procedures, using custom simulants or even real food, as well as the use of different pretreatment methods and different analytical instruments. For example, GC-MS and LC-MS have been utilized for the detection of bisphenol related compounds in packaged food. Yet, it takes time and labor when it comes to long-time and large-scale migration tests, and it usually constrains only a fraction of the chemicals each time, as target-oriented analyses have difficulties in monitoring beyond the range of target substances. In-silico prediction methods like predictive migration models and demographic risk screening have been developed for fast chemical risk assessments without the need for experiments; however, the uncertainty of under- or over-assessed risk has been the deficiency that is too difficult to improve. Nontarget analysis is becoming a balanced choice of both accuracy and efficiency, and is now widely used in food and FCM safety studies. LC-MS and GC-MS with the application of a custom or commercial database have become practical tools for nontargeted identification of substances within FCM extracts and food simulants; they have been proven to be effective in the identification of substances inside and outside the prioritized list. However, as most nontargeted analysis focused on simulated migration tests of FCMs, few report are available for premade food or ready-to-eat food. As this kind of food gets more and more popular, food safety regarding FCMs and food content might need further evaluation.

Here we apply nontarget analysis on a series of premade food with different package types using gas chromatography quadrupole time-of-flight mass spectrometry (GC-QTOF-MS) to identify the substances in them. Additionally, to overcome the cumbersome post data analysis generated by the nontargeted approach, we propose a filtering workflow using publicly available data and chemical structure similarity, aiming to focus on artificially introduced chemicals with ease.

2. Materials and Methods

2.1. Sample Preparation

Premade food items were purchased through the Internet, a total of 17 items, all within their shelf life, covering four types of common premade food packages, composite aluminum pack, aluminum container (available for direct open fire heating), plastic pack, and bowl; additionally, there is also can-packed food. Detailed information is listed in Table S1. All foods were prepared through boiling in water unopened for 20 min as instructed. Food content was homogenized, freeze-dried for 48 h, and stored for later investigation. Food samples were weighed before and after freeze-drying for subsequent conversion. The concentrations of chemical substances in food samples were calculated based on fresh weight. Packages were washed with clean water and cleaned with alcohol wipes to remove greasy food remains.

2.2. Sample Extraction, Cleanup and Instrument Analysis

Chemicals and reagents used for analysis are detailed in Text S1.

An amount of 1 g of freeze-dried food content was weighed and centrifuged with 30 mL of a 1:1 v/v mixture of n-hexane and dichloromethane three times for 20 min each time under 3000 r/min. Extracts were collected and evaporated to approximate 1 mL for later cleanup.

For packages, 1 g of material was cut into approximately 0.5 cm × 0.5 cm pieces and then ultrasonically extracted with the same solvent and time setting as used for food samples. The collected extract followed the same cleanup procedure as used for food samples.

An additional cleanup method was adopted and modified from Cheng et al. In brief, the 1 mL extract was added to a preconditioned gel permeation chromatography column (GPC), the column was eluted with 200 mL of a 1:1 v/v mixture of dichloromethane and n-hexane, and then 70–200 mL fractions were collected. The collected fractions were evaporated to almost dryness and then redissolved in 100 μL of n-hexane for future instrument analysis. Finishing samples were analyzed by using an Agilent 8890 GC instrument coupled to an Agilent 7250 QTOF-MS instrument (Agilent Technologies). Detailed instrument settings are consistent with those of Cheng et al. Briefly, a roughly 100 min program which increases the temperature of the oven from 60 to 310 °C in multiple steps is used, and mass spectrometry used 70 eV electron ionization mode.

2.3. Data Process, Quality Assurance and Quality Control

Raw data produced from the above instrument was first processed with Agilent Unknowns Analysis (version 10.1) software. A threshold of match score over 75 and retention index deviation within 5% were set for the initial candidate filtering, giving confidence level 2. Candidates further confirmed with standards available would be given confidence level 1, and quantified with a standard curve of concentrations between 0.1 and 10 ppm; for those of level 2 without standards, semiquantification was applied using standards of a similar one from the same chemical family (the chemical family is described in Results and Discussion). The confidence level of these chemicals is shown in Table S2. Calibration curves for level 1 chemicals were constructed by using 7 concentrations ranging from 100 ppb to 10 ppm; r ² values of those curves were 96.1%–99.9%. A procedure blank was included in each sample batch, and the target analytes were not detected in all blanks. Method detection limits (MDL) (listed in Table S2) were calculated as 3 times the standard deviation of 7 spiked blank samples; these blank samples used 1 g of diatomaceous earth as the blank material, and were spiked with the eight standard compounds. Additionally, to validate the accuracy of the method, three food and FCMs samples were selected at random, each spiked with the level 1 standards, and the recovery for spiked standards ranged from 83% to 105%. The detected substance in a real sample was only kept if its peak area was more than five times the area of the relevant peak in the blank chromatogram.

2.4. Risk Assessment and Chemical Prioritization

A preliminary risk assessment is done by comparing estimated daily intake (EDI) with the corresponding reference dose (RfD) value. EDI is estimated here as

$$\mathrm{EDI}=C_{\mathrm{food}}\times m_{\mathrm{food}}/\mathrm{BW}$$ 1

C _food is the concentration of chemicals in food, the intake of food (m _food) is set to 100 g per day for a human body weight (BW) of 60 kg, and individual RfD and reference are listed in Table S2.

To assess the overall potential risk, a prioritization approach is adopted to rank the importance of all detected chemicals. For health risks associated with oral exposure to chemicals, end points involving absorption, distribution, metabolism, excretion, and toxicity behavior are considered. However, due to incomplete existing pharmacological data, all end points are predicted with ADMETlab for consistency. Human intestinal absorption, volume of distribution, human liver clearance, and half-life of clearance are considered as the end point for adsorption, distribution, metabolism, and excretion, respectively. For toxicity, NR-AhR, NR-A, NR-ER, SR-ARE, hERG, carcinogenicity, respiratory toxicity, and human hepatotoxicity are considered, and the average score is used during toxicity ranking. All the end points are scaled and normalized, and scores in all domains are summed for overall ranking.

2.5. Data Collation Workflow and Chemical Filter Rules

To refine the results, we adopted the following filtering strategy. As shown in Figure , first, basic information on all chemicals was gathered automatically through a web-based service, including chemical class based on classyfire rules; information like usage in industry and food additives was obtained from PubChem. Referring to the consolidated information provided by PubChem, a lot of chemicals in our results are natural or food originated products, for both substances detected in food content and some detected in packages, which permeated from food content. These natural and food originated chemicals are not of our interest, as we are targeting human-induced compounds, both intentional and unintentional. However, a lot of chemicals cannot be defined due to the lack of information; therefore, we implied information completion on these unknowns based on chemical similarity.

1.

Workflow diagram.

A pipeline is coded in python, chemical information is retrieved, and classification is done automatically by using pubchempy, pyclassfire, and web service with modification. Python packages networkx, pyvis, and matplotlib are used for the plotting of the molecular network. Code is available online. With information consolidated, we define chemicals that have records in natural product databases like Lotus and FooDB, while those with no record in sections like manufacture or industry are labeled as natural and food-related products. Chemicals with only manufacture information are labeled as synthesized. Chemicals with both kinds information are labeled as mix-info, while chemicals with no indication of whether they are natural or synthesized are labeled as unknown. For each unknown chemical, the top 10 known substances with the highest similarity scores were collected based on similarity scores (a threshold of 0.75 is set in this study), and if all of these chemicals were natural, then the chemical is also labeled as natural. Data fills are performed for all chemicals multiple times and iterated until no further changes occurred.

As traditional fingerprints like Morgan fingerprint focus on a predefined substructure with a radius of 2 atoms, which is insufficient to describe the similarity of molecules with multiring like PAHs, we adopted a custom similarity score, consisting of three parts, i.e., the Tanimoto coefficient of the whole molecular fingerprint, the Tanimoto coefficient of the molecular Murcko scaffolding fingerprint, and the number of atoms with the largest common substructure between the molecules (normalized by the sum of the atomic numbers). Used fingerprints include Morgan fingerprint (extended-connectivity fingerprints (ecfp) with a radius of 2) and RDKit fingerprint.

3. Results and Discussion

3.1. Refined Data with Data Collation Workflow

A total of ∼1800 unique chemicals were detected based on the identification criteria mentioned in Materials and Methods. During information completion, using the original 340 known food-related substances, another 200 substances with information missing were labeled as food-related using the similarity strategy. As illustrated in Figure , the molecular network is constructed based on the similarity rules for better visualization; in the example shown in Figure (b1 and b2), information completion was able to successfully sort the entire community consisting of steroids and terpenoids as food-related based on the known ones.

2.

Molecular network visualization of all the chemicals before (a1) and after (a2) the information-filling strategy. Chemicals labeled food-related are colored in green, those that are synthetic-related are in red, those which have both types of information are in blue, and those which have no information are in gray. Zoomed in images (b1 and b2) show a community consisting of steroids and terpenoids.

Substances recognized as food-related, along with substances whose chemical class is related to fatty acyls, lignans, glycerolipids, lipids, and steroids, were removed during the data-trim. Further, as the purpose of this study focuses on the potential migration of chemicals from the package to food content, cross reference was applied, only chemicals occurring at least once in both food content and its corresponding package were kept for final manual inspection.

As a result of chemical removal and cross referencing, about 80 chemicals remained for further investigation. This proves the effectiveness of the workflow, and with its simplicity and reproducibility, we believe this method could be used for complex FCMs data investigation. And with further improvement and customization, it can help handle the large amounts of data produced by other types of nontarget analysis, saving both time and labor, while preserving reproducibility.

3.2. Identification and Risk Assessment

The information on the final 83 chemicals is listed in Figure S1, and is illustrated in Figure using a molecular network. As illustrated, communities of chemicals were formed, and among them, many can be distinguished based on structure and common knowledge (as circled in Figure ). We also retrieved some of the filtered ones for these chemical families for a better overview (as listed in Table and full information in SI as Table S3). As shown in Table and Table S3, chemicals are sorted and quantified and discussed in the following sections, along with toxicity evaluation performed using the method described in a previous section.

3.

Molecular network of the final chemical list.

1. Concentrations (min–max­(mean)) of Chemicals in Food and Packaging .

		Food (ng/g (fresh weight))					Package (ng/cm2)
name	cid	Aluminum plastic composite bag	Aluminum smooth wall container	Plastic bag	Plastic bowl	Metal can	Aluminum plastic composite bag	Aluminum smooth wall container	Plastic bag	Plastic bowl	Metal can
PAHs and derivatives
naphthalene	931	-	-	-	ND-6.8(6.8)	-	ND-0.6(0.4)	ND-0.9(0.9)	0.2–0.3(0.2)	ND-2.1(1.6)	-
phenanthrene	995	ND-6.5(5.9)	ND-7.2(6.7)	ND-6.4(5.1)	ND-7.1(6.1)	-	0.3–2.0(0.8)	ND-2.1(1.5)	ND-0.7(0.5)	1.5–5.4(3.4)	-
tetracene	7080	ND-5.8(5.8)	-	-	-	-	-	-	ND-0.3(0.3)	ND-2.3(1.7)	-
benzo[a]anthracene	5954	-	ND-5.5(5.5)	-	-	-	-	-	-	ND-2.7(2.7)	-
perylene	9142	-	-	-	ND-6.7(6.7)	-	-	-	ND-0.4(0.3)	-	2.5
triphenylene	9170	-	-	ND-4.0(4.0)	-	-	ND-0.5(0.5)	ND-1.2(1.2)	ND-0.4(0.4)	1.1–3.5(2.2)	3
pyrene	31423	4.9–6.7(5.9)	ND-6.0(5.9)	ND-6.1(4.6)	5.0–7.2(6.2)	6.2	ND-1.3(0.7)	0.7–2.0(1.1)	ND-0.5(0.5)	1.5–4.8(3.1)	4
1,2-dimethylnaphthalene	11317	ND-6.2(6.2)	-	-	-	-	ND-0.5(0.4)	-	ND-0.2(0.2)	ND-1.8(1.8)	-
9-methylidenefluorene	78147	ND-6.8(6.6)	-	-	ND-5.6(5.6)	-	-	ND-0.8(0.8)	-	-	-
9-methylanthracene	13068	-	ND-4.8(4.8)	-	-	-	-	ND-0.7(0.7)	ND-0.2(0.2)	ND-1.0(0.9)	-
9-propylanthracene	605801	-	ND-6.3(6.3)	-	-	-	-	-	-	-	-
PAEs and derivatives
dimethyl phthalate	8554	ND-5.7(5.5)	ND-7.4(6.2)	ND-4.6(4.1)	ND-9.5(9.5)	-	ND-16.3(5.1)	1.0–5.8(2.7)	0.3–4.8(1.5)	0.9–2.3(1.8)	1.6
diethyl phthalate	6781	-	-	-	ND-6.3(6.3)	-	ND-0.6(0.4)	0.7–0.9(0.8)	0.2–0.3(0.2)	0.7–1.9(1.3)	1.4
dibutyl phthalate	3026	ND-24.1(20.5)	-	-	-	-	1.8–17.6(6.8)	3.1–18.1(9.5)	0.5–2.5(1.2)	4.9–13.7(8.7)	2.5
diamyl phthalate	8561	-	-	ND-6.6(6.6)	-	-	-	-	-	-	-
diisobutyl phthalate	6782	ND-9.6(9.6)	-	-	-	-	0.3–1.6(0.7)	0.9–2.5(1.5)	0.2–0.4(0.3)	0.8–2.5(1.7)	1.5
bis(2-hydroxyethyl) phthalate	66539	ND-21.6(21.3)	ND-6.2(6.2)	ND-11.6(8.4)	-	-	ND-31.5(11.1)	ND-7.5(5.5)	ND-2.0(1.2)	ND-10.9(10.9)	-
phthalic acid, isobutyl 2-pentyl ester	6423867	ND-4.3(4.3)	-	-	-	-	-	-	-	-	-
bis(2-pentyl) phthalate	6423898	ND-5.3(5.3)	-	-	-	-	-	-	-	-	-
phthalic acid, hept-4-yl isobutyl ester	91720280	-	ND-5.1(5.1)	-	-	-	-	-	-	-	-
phthalic acid, di(hept-4-yl) ester	70621663	-	-	-	ND-4.6(4.6)	-	-	-	ND-0.1(0.1)	-	-
phthalic acid, 6-ethyl-3-octyl butyl ester	6423866	ND-4.2(4.2)	-	-	-	-	ND-0.4(0.3)	ND-1.0(0.9)	ND-0.3(0.3)	ND-1.7(1.7)	-
mono-(2-ethylhexyl)phthalate	20393	-	-	-	ND-4.4(4.4)	-	-	-	-	-	-
BPA and BADGE
bisphenol a	6623	-	-	-	-	2.3	-	-	-	-	1.5
tetramethylbisphenol a	79717	-	-	-	-	1.2	-	-	-	-	4.7
bisphenol a diglycidyl ether	2286	ND-117.8(69.2)	1.5–17.1(7.2)	-	-	1.8	ND-6.1(2.4)	0.7–3.0(1.5)	ND-0.4(0.4)	ND-5.2(2.3)	1.6
OPEs
triethyl phosphate	6535	-	-	ND-13.7(13.7)	-	-	0.9–2.1(1.4)	2.3–3.8(2.9)	ND-1.1(0.8)	ND-7.4(4.4)	-
diphenylphosphorylbenzene	13097	-	ND-18.2(18.2)	-	-	-	ND-1.6(1.3)	2.4–3.0(2.7)	-	ND-5.8(5.8)	-
triphenyl phosphate	8289	-	-	-	-	20.3	ND-1.4(1.1)	-	-	ND-10.2(10.2)	12.1

^aND is not detected, defined as signal-to-noise ratio < 100, or does not exceed 5 times areas in the blank.

3.2.1. PAHs and Derivatives

In our samples, only low-ring polycyclic aromatic hydrocarbons (PAHs) were found in both food and packages. In addition to the suspect lists, food contact chemicals database (FCCdb), and list of migrating and extractable food contact chemicals (FCCmigex), our workflow was able to retrieve another 4 PAHs outside these lists, i.e., four-ring PAHs (tetracene and benzo­[c]­phenanthrene), branched anthracene and fluorene. Pyrene (Pyr, 129–00–0) has the largest detection rate between food and its packages, with concentrations of ND-7.2 ng/g, followed by phenanthrene (Phe, 85–01–8) with concentrations up to 7.2 ng/g. 1,2-Dimethylnaphthalene (573–98–8) and naphthalene (NaP, 91–20–3) were also detected in quite a few packages, yet only in two food samples. Some were found in food, but not in their corresponding package, i.e., tetracene (5.8 ng/g, S2), benzo­[a]­anthracene (5.5 ng/g, S7), and 9-methylanthracene (4.8 ng/g, S9). There are some other PAHs and their derivatives that were found in packages only, like benzo­[c]­phenanthrene, undecaene, and benzo­[a]­pyrene; they are not listed here for further discussion. Total concentrations of all PAHs listed in Table are ND-21.0 ng/g for our samples, similar to other PAHs reported in food stuff, a total concentration of 8 PAHs (ND–11.1 ng/g) was reported in various fresh and processed foodstuffs, ND-12.0 ng/g in some major food groups, and ND–12.0 ng/g in some major food group. This concentration was lower than that in smoked and grilled food, as previously reported as up to 500 ng/g, likely because our sample food did not undergo such high PAH formation pathways during industrial mass production, or because methods have been taken to limit PAHs during food processing, like additives, packaging, ways of cooking. There is no certainty that PAHs have primary migratory origins from packaging, as PAHs can also form and transform during cooking processes at high temperatures, and may even originate from indigenous substances in the food ingredient resulting from bioaccumulation.

PAHs’ potential carcinogenic properties and associated health risks have been widely known. Regulations regarding their presence in food have also been implemented, like EC No 1881/2006 which limits PAHs in smoking and grilling related food, GB7104–94 in China for foodstuff, etc. There are no specific restrictions on PAHs in FCMs, but there are general restrictions on PAHs in customer products, like limitation on human contact goods by REACH and substantial components usable as food additives. In our sample, upon comparison with RfD, with the generally low toxicity of these low-ring PAHs and their derivatives, the food-related risk appears to be low.

3.2.2. PAEs and Derivatives

In our food sample, as Table shows, mutual detected phthalic acid esters (PAEs) are dimethyl phthalate (DMP, 131–11–3), diethyl phthalate (DEP, 84–66–2), diisobutyl phthalate (DIBP, 84–69–5), dibutyl phthalate (DBP, 84–74–2) and bis­(2-hydroxyethyl) phthalate (84–73–1). Besides those, there are some low-detection frequency PAEs and derivatives in some food samples, i.e., diamyl phthalate (DPP/DPNP, 131–18–0), bis­(2-pentyl) phthalate (75151–00–7), mono-(2-ethylhexyl)­phthalate (4376–20–9), and also many other miscellaneous ones detected only in packages. These uncommonly alkyl branched ones were reported for the first time if referencing with the fccmigex. And as PAEs could undergoing complex transformation like metabolism in plants and animals and changes during food processing, it can explain the detection of these PAEs’ derivatives with different branch compositions. Traditional PAEs detected in our samples are DMP, DEP, DPP and DBP; DMP is the most frequently detected PAE in food samples. The total concentrations of these traditional PAEs were from ND-39.4 ng/g, which is ND-9.5 ng/g for DMP, 16.9 and 24.1 ng/g for DNP in two samples, 6.3 ng/g for DEP in one sample, and 6.6 ng/g for DPP in one sample. Comparing with other study, a total concentration up to 19.4 ng/g (2.3 ng/g for DMP, 2.8 ng/g for DEP) was found in food. Another food survey reported DMP at 0–44.9 ng/g (mean 0.949 ng/g), DPP at 0–2090 ng/g (mean 19.2 ng/g), DEP at 0–102 ng/g (mean 2.18 ng/g), DNP at 0–105 ng/g (mean 10 ng/g), respectively. Like the case with PAHs, PAEs have been reported to be ubiquitous in the environment and can be introduced during food processing, so there is no conclusive evidence at this stage that PAEs in our samples migrated from their packaging.

Due to their widespread use in cosmetics, personal care products, and plasticizers, PAEs are now commonly found in various matrices. Concerning the chemical safety of PAEs, they has been under heavy restriction as to limit their influence on human health, for example, limitation in children’s products and drinking water. Since PAEs are very connected to material making, they have been listed in FCMs related regulations, specifying their usage and migration amount before marketing, and EU 2023/1442 sets specific migration limits on DBP, Bis­(2-ethylhexyl) phthalate (DEHP, 117–81–7) and other PAEs. For our sample, detected DBP levels were all within the regulated 0.05% limit. The EDIs for the above PAEs and their derivatives were found to be lower than their RfDs by the risk assessment, but their combined risk may be alarming. With health risks concerning those PAEs derivatives not yet available, their risk should be investigated in the future.

3.2.3. BPA and BADGE

BPA (80–05–7), tetramethyl bisphenol A (TmBPA, 5613–46–7) and bisphenol A diglycidyl ether (BADGE, 1675–54–3) were detected in our samples, among which BPA and TmBPA were only detected in the can packaging and its food contents. BADGE, an oligomer in BPA epoxy resin, was detected in all types of packaging yet only in foods packaged in aluminum and cans. The concentrations of BPA and TmBPA were 2.3 and 1.2 ng/g, respectively. BADGE ranged from ND-20.6 ng/g with one of the samples reaching an extreme high concentration of 117.8 ng/g; it is worth noting that BADGE’s concentrations in those packaged foods are much higher than that in the canned food, possibly due to the different composition of their materials. Comparing with other reports, BPA was detected with concentrations of 2–23 ng/g in PP container food and ND-57 ng/g in canned food. For BADGE, there are reports of concentrations ranging from 0.12 to 16.17 ng/g in canned food, along with other derivatives of BADGE like BADGE·2H2O (highest at 493.51 ng/g). Others reported high concentrations of 60–220 ng/g in canned fish, also with the presence of many derivatives. Since BADGE was detected in our samples, there could be some other BADGE derivatives as well, and this might require further monitoring. As can be seen from our results, free BPA can be an issue for FCMs using epoxy resins, but it is interesting to see that besides the can package, other packages have BADGE oligomers detected with no BPA detection, since free BPA can be removed by flushing and cleaning during manufacture, which can be important for future epoxy-coated FCMs production.

Infamous for its adverse effects as an endocrine disruptor like estrogenic activity, thyroid hormone action, etc., the restrictions of BPA have been constantly updated, tightening its usage. European Food Safety Authority (EFSA) reduced its TDI from 4 μg/kg BW/d down to a strict 0.2 ng/kg BW/d. Previously banned only in FCMs for baby products, as (EU) 2024/3190 enacted at the end of 2024, a complete prohibition of BPA in all FCMs has been enforced, urging safer alternatives to BPA and other bisphenol derivatives. Canned food in our samples did not comply with the latest EU Regulation 2024/3190, which prohibits any detectable level of BPA; still it was below the migration limit of 50 μg/kg set previously by the superseding EU 2018/213. Although the RfD values we used here were collected by the U.S. Environmental Protection Agency’s (US EPA) Integrated Risk Information System (IRIS), which instructed the EDI to be safe, this EDI is about 10 times more than the strict TDI set by the EFSA mentioned above, which could be showing human health risk upon consuming this kind of canned food. For BADGE, the concentration was safe considering the SML of 9 mg/kg set by EC No. 1895/2005. But recent research on BADGE and its derivatives has been reporting signs of their potential toxicity; their monitoring in package food and further health assessments might still be required.

3.2.4. OPEs

In our samples, organophosphate esters (OPEs) were detected, i.e., triethyl phosphate (TEP, 78–40–0), diphenylphosphorylbenzene (TPPO, 791–28–6), and triphenyl phosphate (TPhP, 115–86–6). They were detected in quite a lot of packages; however, the detection frequency in food was very low, with each substance being detected in only one sample and each in an individual one. Our result was similar to those levels detected in other reported packaged food, which were ND-46.2 ng/g for TEP, ND-4.22 ng/g for TPhP, while lower than those from some studies, where packaged foodstuff contained up to 266 ng/g for TEP, 390 ng/g for TPhP. Higher levels of OPE have been reported in packaged foods than in nonpackaged foods, and cases of FCMs migration tests confirmed its potential of migration. , Therefore it is possible that packages were a critical source responsible for the containment of OPEs in the food content of our samples. As for the high detection rate in packages, yet rarely in food content, it is possible that actions were taken during FCMs manufacture and food packaging, in order to minimize migration of OPEs.

As OPE production grows, its existence in the environment as anthropogenic organic contaminants has been revealed, bringing attention to its environmental and ecological risk. However, there is yet any specific regulation on OPEs in food. The risk was negligible for the level of OPEs detected in our sample compared with their corresponding RfD, but given the current status of OPEs, there should be a risk warning system and comprehensive regulation for them in food and FCMs areas.

3.2.5. Other Chemicals

As shown in Figure , we can see that in community A there are 2,4-di-tert-butylphenol (96–76–4), 2,6-di-tert-butyl-4-methylphenol (128–37–0), 2-tert-butyl-4-methoxyphenol (121–00–6) and 3-methyl-4-propan-2-ylphenol (3228–02–2), which share similar structures of phenol and tert-butyl. They all have usage records as antioxidants or preservatives in various areas, like paint, ink, and food, explaining their presence in our samples.

Community B consists of substances sharing structures like prop-1-enylbenzene and other substructures; a lot of them have functions as preservatives and flavoring agents. Of note, in one of them, ethyl (E)-3-(4-methoxyphenyl)­prop-2-enoate (24393–56–4) can be used as a UV absorber in food packaging. There have been increasing reports on UV filters. This set of chemicals has been detected in various environmental matrices, revealing their potential risks. As there is little analysis on UV filters in food packages and FCMs migration, further investigation on their occurrence and risk for food safety purposes is required.

Besides the above chemicals sorted by their basic structure, there are some spare chemicals, one of them being 1,3,5-triphenylcyclohexane (28336–57–4), detected in every food and package sample except for the one of the food content. It is reported as an impurity in polystyrene (PS) food containers, yet none of our package samples is PS. It is likely that it was derived from food ingredients or introduced from equipment or intermediate packaging during the production process. Oligomers, like ethylene terephthalate cyclic dimer (24388–68–9), were also detected, judging by the fact that ethylene terephthalate cyclic dimer is a PET oligomer and PET can be used as layer material such as outer paint surface and vinyl-based coating for the composite aluminum package, like the one used by our samples. It is possible that the cyclic dimer originats from the painted surface of package, through migration or adhered during package stocking.

Apart from the above discussion, we also convert the result in the form of total amount of chemicals between the package and food, shown in Figure S2, and we can see that, for most chemicals, the total amount in food was larger than that in package. This may suggest that more chemicals can migrate from packaging to food in the long term, or our sample has been stored long enough for most chemical to migrate. This may require more detailed migration tests on raw package materials.

As over 50% of the chemicals in this refined data appeared outside the suspect lists of fccdb and fccmigex, we can see that nontarget analysis using GC-MS and this workflow on general commodities are capable of identifying and screening common and uncommon substances.

3.3. Chemical Prioritization

As shown in Figure S3, the substance with the highest score is 4-methyl-1,2,3-trithiolane, for which there is little publicly available information, but given that it is a cyclic alkane containing sulfur atoms and that garlic-related compounds like hexasulfur (13798–23–7) and cyclooctasulfur (10544–50–0) were also detected in the same sample, it is likely that it also originated from food, similar for the top-ranked 4,7-diethyl-1,2,3,5,6-pentathiepane, since its toxicity score is not too high (it will not discussed in detail here). Second in the ranking, benzaldehyde (100–52–7), a multipurpose chemical agent and flavoring agent used in food, having an RfD of 0.1 mg/kg-day given by IRIS, could be considered as having low toxicity. Fifth in the ranking, 1,2-diphenylcyclopropane (29881–14–9), while no information is available in public databases, is mentioned by Lin et al., which could be related to the aromatic compound product of Phe. PAHs, OPEs discussed in previous sections, like Pyr, Phe, TPhP, TmBPA, were top-ranked for their general high toxicity. 3-Acridinol and 9-phenyl scored highest in toxicity and high in overall ranking but no information is available. However, its substructure 9-Phenylacridine (602–56–2) has been listed in HBM4 EU CECscreen which is a suspect screening list for Chemicals of Emerging Concern (CECs). Upon searching with SciFinder, there are patent records about its use as an initiator during photosensitive resin making, which could explain its detection in our samples. The above data analysis gives a general perspective by having a preliminary risk ranking of all the compounds; further confirmation and safety determination are needed in the future.

4. Conclusion

In this study, we performed nontargeted analyses on premade foods and their packaging using an automated workflow to identify potential hazardous volatile and semivolatile substances, the output of which greatly condensed data to meet our needs, and upon utilizing the molecular network, the readability of the data was improved and formulation of clusters sharing similar structures and functions helped to explain chemical profiles and occurrence. Families of PAEs, PAHs, BPA, and OPEs were detected, and further intake health assessment indicated low health risk, except for potential risk of BPA detected in canned food due to recently tightened restrictions. In addition to common substances, other rarely reported chemicals were also detected, and clusters with similar functions like antioxidants and UV absorbers were identified. As volatile and semivolatile compounds can be detected by GC-MS in this study, further works with LC-MS may be used for analyzing more compounds. The flexibility of this workflow can be used for postdata processing with other instrumentation. In general, this workflow has proven its efficiency in nontarget postdata analyses with limited manual intervention, and it can be used to monitor unexpected chemical substances in food content, providing initial data support to the safety of premade food. Premade food with different food content and packaging materials has been available in the market. While traditional target analysis has the possibility of overlooking potential risk, we believe our work can provide the first insight into food safety of premade food using nontarget analysis and data mining.

The code is available at https://github.com/jiyuanweng/ChemInfoCollect_MolNetPlot.

The Supporting Information is available free of charge at https://pubs.acs.org/doi/10.1021/envhealth.5c00515.

• Chemicals and reagents used in analysis; sample information; confidence level, MDL and RfD value of chemicals; concentrations of chemical in food and packaging samples; chemical information after filtering; area response comparison for chemicals; priority ranking of chemicals (PDF)

The authors declare no competing financial interest.