Exploration of the curated list of plastic additives

Plastic additives are chemicals that are added to plastics to achieve specific desirable properties in the end product (Pritchard 2012; Hahladakis et al. 2018). External stress on such products can cause the separation of these additives, thereby leading to their release into the environment and eventually posing risks to humans and ecosystems (Maddela et al. 2023). In this study, we curated a list of plastic additives from chemicals cataloged in the UNEP report titled ‘Chemicals in Plastics—A Technical Report’ (Table S2) (UNEP 2023) and explored their potential risks by systematically integrating the associated heterogenous biological endpoints within the context of adverse outcome pathway (AOP) framework (Fig. 1).

The UNEP report provides functional annotations for each of these chemicals based on two independent studies by Aurisano et al. (2021b) and Wiesinger et al. (2021). Notably, Wiesinger et al. observed that their text mining approach for identifying these functional annotations lacked context sensitivity, leading to some inaccuracies. Despite this limitation, the UNEP report remains the most comprehensive source cataloging chemicals found in plastics and their associated functions. Therefore, we relied on the functional annotations provided by the report to identify 6470 chemicals with reported functions, which we designate as 'plastic additives' in this study (Methods; Table S2).

Among the 6470 plastic additives, we observed that many chemicals (3217 of 6470) provide a variety of functions to the plastics, with colorants being the most frequently associated function (3675 of 6470) (Methods; Table S2). Furthermore, we observed that majority of plastic additives (4309 of 6470) are found in products made by different priority use sectors, of which 3963 additives are found in the use sector ‘Packaging, including food contact materials’ (Methods; Table S2).

Next, we relied on the United States high production volume (USHPV) chemical list (https://comptox.epa.gov/dashboard/chemical-lists/EPAHPV) and Organisation for Economic Co-operation and Development high production volume (OECD HPV) chemical list (https://hpvchemicals.oecd.org/ui/Search.aspx) and identified 2084 of 6470 plastic additives to be high production volume (HPV) chemicals (Table S2). Notably, among these HPV plastic additives, we found 154 additives to be known endocrine disrupting chemicals (EDCs) with experimental evidence for endocrine disruption in humans or rodents from DEDuCT (Karthikeyan et al. 2019, 2021), and 101 additives as potential carcinogens based on IARC monographs on identification of carcinogenic hazards to humans (https://monographs.iarc.who.int/list-of-classifications/) (Table S2). Furthermore, we observed that 215 additives are identified as substances of very high concern (SVHC) (https://echa.europa.eu/candidate-list-table) by European Chemicals Agency (ECHA) and 412 additives are prohibited for use as per REACH regulation (Table S2). Figure 2a shows the distribution of HPV, SVHC and REACH prohibited plastic additives across the ten priority use sectors.

Fig. 2

Identification of plastic additives in different chemical regulations and human biospecimens. a Heatmap depicting the presence of plastic additives from ten priority use sectors in chemical regulations. The number of the plastic additives from the priority use sector in each of the chemical regulations is denoted in the heatmap. b Heatmap depicting the presence of plastic additives from ten priority use sectors in different human biospecimens based on published exposure studies

Plastic additives are accumulated in various human biospecimens

Humans are exposed to various plastic additives via direct contact, inhalation or ingestion, which can eventually accumulate in different human tissues and potentially lead to various adverse health effects (Meeker et al. 2009; UNEP 2023). To explore the plastic additives detected in various human biospecimens, we relied on two databases namely, Tissue-specific Exposome Atlas (TExAs) (Ravichandran et al. 2021a) (https://cb.imsc.res.in/texas/) and Exposome–Explorer (Neveu et al. 2020) (http://exposome-explorer.iarc.fr/) which have compiled the presence of environmental chemicals as xenobiotics in different human tissues from published exposure studies. Although, these two databases have compiled information from limited human exposure studies, they have documented 204 of the 6470 plastic additives to be accumulated as xenobiotics in 37 different human biospecimens (Fig. 2b; Table S2). Moreover, we observed that plastic additives from nine of the ten priority use sectors have been documented as xenobiotics in human biospecimens namely, faeces, serum, urine, lung, placenta, and adipose tissue (Fig. 2b). Note, the use sector ‘Synthetic textiles’ comprised the least number of additives (6 chemicals) in our curated list of 6470 additives, and there are no published exposure studies wherein their presence was detected in different human biospecimens.

Stressor–AOP network for plastic additives

Stressor–AOP networks provide a panoramic visualization of the different AOPs associated with stressors of interest, and help in understanding the stressor-induced adverse biological effects (Aguayo-Orozco et al. 2019). In this study, we therefore constructed stressor–AOP network to understand the various adverse effects induced by plastic additives. First, we followed a systematic approach that involved data-centric integration of heterogenous toxicogenomics and biological endpoints data from five exposome-relevant resources namely, ToxCast, CTD, DEDuCT, NeurotoxKb and AOP–Wiki, and identified 688 KEs within AOP–Wiki to be associated with 1314 of the 6470 plastic additives (Methods; Table S7). Thereafter, we curated 328 high confidence AOPs within AOP–Wiki and mapped them to plastic additives if at least one KE within that AOP is associated with the plastic additive. Based on these plastic additive–AOP associations, we constructed a plastic additives-centric bipartite stressor–AOP network comprising two types of nodes namely, 1287 plastic additives and 322 high confidence AOPs, and 46,243 stressor–AOP links as edges between the two types of nodes, and we designate this bipartite network as plastic additives–AOP network. Notably, we observed that AOP–Wiki documented only 37 of the 1287 plastic additives in the constructed stressor–AOP network to be associated with 27 of the 322 high confidence AOPs in the network.

Next, we leveraged the KEs associated with plastic additives to compute the coverage score for the stressor–AOP links in the plastic additives–AOP network and observed that 20 plastic additives are associated with all the KEs (coverage score = 1.0) in 15 high confidence AOPs, and these stressor–AOP links were otherwise not documented in AOP–Wiki (Methods; Table S9). Moreover, we calculated the levels of relevance for the stressor–AOP links in the plastic additives–AOP network and observed that 27,189 links between 1155 plastic additives and 288 AOPs are classified as Level 1, 4236 links between 345 plastic additives and 241 AOPs are classified as Level 2, 14,187 links between 1152 plastic additives and 139 AOPs are classified as Level 3, and 631 links between 118 plastic additives and 98 AOPs are classified as Level 5 (Methods; Table S9). Note, the stressor–AOP links with Level 4 relevance were also satisfied by Level 5 criterion, and therefore, there are no stressor–AOP links with Level 4 relevance in the constructed network (Methods; Table S9).

Next, we relied on the standardized disease ontology provided in Disease Ontology (Schriml et al. 2022) database (https://disease-ontology.org/do) to classify the AOPs based on their adverse outcomes (AOs). Based on the standardized ontology, we classified 322 AOPs into 26 disease classes based on their AOs (Tables S9, S10). Note that 125 of the 322 AOPs could not be classified under any standardized ontology provided by Disease Ontology, and we therefore marked them as ‘unclassified’. Importantly, we observed that cancer is the most represented disease category comprising 40 of the 322 AOPs in the plastic additives–AOP network (Table S9). Finally, we have linked the plastic additives to their corresponding priority use sectors and the AOPs to their corresponding disease categories in the plastic additives–AOP network (Table S9).

We relied on the graph visualization software Cytoscape (Shannon et al. 2003) to visualize the plastic additives–AOP network for each of the 1287 plastic additives, and make them available on a dedicated website: https://cb.imsc.res.in/saopadditives/. In the website, the plastic additives are grouped based on their priority use sectors. For instance, the priority use sector ‘Toys and other children’s products’ consists of 162 plastic additives, 301 AOPs and 8300 stressor–AOP links, wherein 4696 links between 148 plastic additives and 265 AOPs are classified as Level 1, 1460 links between 75 plastic additives and 170 AOPs are classified as Level 2, 1928 links between 144 plastic additives 117 AOPs are classified as Level 3, and 216 links between 30 plastic additives and 58 AOPs are classified as Level 5. Figure 3 shows a portion of the plastic additives–AOP network, comprising Level 5 stressor–AOP links for plastic additives in the use sector ‘Toys and other children’s products’.

Fig. 3

Visualization of a stressor-centric AOP network for plastic additives in the priority use sector ‘Toys and other children’s products’, along with the disease classifications of adverse outcomes (AOs) in AOPs. In the stressor–AOP network, only edges or stressor–AOP links with Level 5 relevance are shown. Furthermore, the edges or stressor–AOP links are weighted based on their coverage score, i.e., the fraction of KEs within AOP that are linked with the plastic additives

In addition, the constructed plastic additives–AOP network can highlight the adverse outcomes induced by plastic additives across different use sectors. Figure 4 shows the disease categories linked with ten priority use sectors for plastic additives in the plastic additives–AOP network with Level 5 relevance, where cancer is the most represented disease category.

Fig. 4

Aggregate visualization of the disease categories associated with AOPs in plastic additives stressor–AOP network with Level 5 relevance for each of the ten priority use sectors

Stressor–AOP network reveals highly relevant AOPs associated with plastic additives

A stressor–AOP network can help identify most relevant AOPs associated with each stressor which can further highlight the complexity and diversity among toxicity pathways induced by that stressor (Knapen et al. 2018). Here, we considered stressor–AOP links from the constructed plastic additives–AOP network with Level 5 relevance and coverage score threshold of 0.4, and identified 107 of the 1287 plastic additives to be associated with 88 of the 322 AOPs through 526 stressor–AOP links (Methods; Table S9). Note the coverage score threshold of 0.4 denotes that at least 40% of the KEs in that AOP are linked with the stressor (Chai et al. 2021; Sahoo et al. 2024). Notably, 15 of these 107 plastic additives are associated with more than 10 AOPs (Table 1). Among these 15 plastic additives, 14 are documented as EDCs in DEDuCT, and 10 are documented as carcinogens in IARC monographs (Table S2).

Table 1 Top 15 plastic additives based on the associated number of highly relevant AOPs in the plastic additives–AOP network with Level 5 stressor–AOP links

Among the AOPs associated with these 15 plastic additives, we observed that majority of the AOPs are identified through our systematic data integrative approach. Notably, we observed that AOP:263, AOP:264, AOP:265, AOP:267, and AOP:268 are shared among all 15 plastic additives (Table 1). Moreover, we observed that these five AOPs share the same MIE ‘Decrease, Coupling of oxidative phosphorylation’ (KE:1446) and AO ‘Decrease, Growth’ (KE:1521), while AOP:263 titled ‘Uncoupling of oxidative phosphorylation leading to growth inhibition via decreased cell proliferation’ is endorsed by Working Group of the National Coordinators of the Test Guidelines Programme (WNT) and the Working Party on Hazard Assessment (WPHA) under the OECD AOP development programme.

An AOP network constructed from stressor-specific AOPs can highlight interactions among the associated AOPs, thereby aiding in the assessment of stressor-induced toxicity (Pollesch et al. 2019; Rugard et al. 2020; Chai et al. 2021; Pogrmic-Majkic et al. 2022; Yang et al. 2022; Sahoo et al. 2024). Among the 15 plastic additives, we observed that benzo[a]pyrene (28 associated AOPs), bisphenol A (27 associated AOPs), and bis(2-ethylhexyl) phthalate (19 associated AOPs) are the top three chemicals based on the number of associated AOPs (Table 1). Although benzo[a]pyrene has been annotated as a plastic additive in this study, evidence suggests that it is more likely a contaminant or a byproduct resulting from the use of extender oils or carbon black in plastic production (Lassen et al. 2011; Alassali et al. 2020; Wiesinger et al. 2021). Nonetheless, all these three chemicals are well-known pollutants. Therefore, we constructed AOP networks for each of these chemicals and explored their potential human-relevant and ecotoxicology-relevant toxicity pathways.

Exploration of toxicity pathways in AOP network constructed from benzo[a]pyrene-relevant AOPs

Benzo[a]pyrene (B[a]P or CAS:50-32-8) has the largest number of associated AOPs (28 AOPs), all of which were solely identified through our systematic data integrative approach (Table 1). These 28 AOPs are classified under various disease categories namely, cancer, gastrointestinal system disease, reproductive system disease, respiratory system disease, cognitive disorder, thoracic disease, and musculoskeletal system disease (Table S9). Previously, Yang et al. (2022) had constructed an AOP network for B[a]P-induced toxicity, but they relied only on CTD to identify KEs associated with B[a]P and focused only on B[a]P-induced male reproductive damages. Therefore, we relied on 28 AOPs associated with B[a]P-induced toxicity (which we designate as B[a]P–AOPs) and constructed an AOP network to explore various adverse effects associated with B[a]P.

Next, we computed the cumulative weight of evidence (WoE) for each of these 28 B[a]P–AOPs based on their KER information to assess the biological plausibility (Ravichandran et al. 2022b; Sahoo et al. 2024). We observed that 9 of these 28 B[a]P–AOPs have ‘High’ cumulative WoE and 6 B[a]P–AOPs have ‘Moderate’ cumulative WoE (Table S11). Moreover, we observed that many of these 28 B[a]P–AOPs are applicable across various species and developmental stages (Table S11). Figure 5 shows the undirected AOP network representation of the 28 B[a]P–AOPs, where nodes represent B[a]P–AOPs and the edges represent the existence of shared KEs between two AOPs. We observed that the B[a]P–AOPs form three connected components (with two or more AOPs) and two isolated nodes, where the largest connected component (labeled C1) comprises 18 B[a]P–AOPs.

Fig. 5

Undirected network of B[a]P–AOPs. Each node corresponds to B[a]P–AOP and an edge between two nodes denotes that the two AOPs share at least one KE. This undirected network has 3 connected components (with two or more nodes) which are labeled as C1, C2, and C3, and 2 isolated nodes

We constructed and visualized a directed AOP network to explore the interactions among the B[a]P–AOPs present in the largest connected component C1 (Fig. 6). We observed that the directed network comprised 66 unique KEs (including 7 MIEs and 11 AOs) and 99 unique KERs (Fig. 6; Table S12). Among the 66 KEs, 36 KEs were associated with B[a]P-induced toxicity through our systematic data integrative approach, of which 5 are MIEs and 10 are AOs (Fig. 6). Notably, we observed that the toxicity pathway originating from MIE ‘Activation, AhR’ (KE:18), passing through KEs ‘Altered gene expression, NRF2 dependent antioxidant pathway’ (KE:1917) and ‘Increase, Cell Proliferation’ (KE:870), and eventually terminating at AO ‘Lung Cancer’ (KE:1670) consists of 4 of the 36 KEs associated with B[a]P-induced toxicity (Fig. 6). Upon further inspection, we identified that this toxicity pathway was captured in the AOP:420 titled ‘Aryl hydrocarbon receptor activation leading to lung cancer through sustained NRF2 toxicity pathway’, which was systematically built and supported by extensive literature survey and experimental data on B[a]P (Jin et al. 2022).

Fig. 6

Directed network corresponding to the largest component in the undirected B[a]P–AOP network comprising 66 KEs and 99 KERs. Among the 66 KEs, 7 are categorized as MIEs (denoted as diamond), 11 are categorized as AOs (denoted as circle), and the remaining 48 are categorized as KEs (denoted as rounded square). The 36 KEs (including MIEs and AOs) associated with B[a]P are marked in ‘red’. In this figure, the 66 KEs are arranged vertically according to their level of biological organization (colour figure online)

Next, we computed different node-centric network measures to explore various features of this directed network. We observed that the MIE ‘Activation, AhR’ (KE:18) has the highest out-degree of 16, while the KE ‘Altered, Cardiovascular development/function’ (KE:317) and AO ‘N/A, Liver fibrosis’ (KE:344) have the highest in-degree of 5 (Table S12). The MIE ‘Increased, Reactive oxygen species’ (KE:1115) has the highest betweenness centrality value, denoting that several toxicity pathways are passing through it in this network (Fig. S5) (Villeneuve et al. 2018). The KEs ‘Induction, CYP1A2/CYP1A5’ (KE:850) and ‘Altered gene expression, NF-kB dependent Interleukin-6 pathway’ (KE:1921), and MIEs ‘Activation, AhR’ (KE:18) and ‘Increase, Reactive Oxygen Species production’ (KE:257) have the highest eccentricity, denoting that they are the most remotely placed KEs in this network (Fig. S6) (Takes and Kosters 2011).

Finally, we relied on artificial intelligence (AI) based tool, AOP-helpFinder (Jornod et al. 2022; Jaylet et al. 2023) (https://aop-helpfinder.u-paris-sciences.fr/), and Abstract Sifter (Baker et al. 2017) (https://comptox.epa.gov/dashboard/chemical/pubmed-abstract-sifter/) to screen published literature and manually identified novel associations between the B[a]P-induced toxicities and the remaining 30 KEs in the directed AOP network (Table S13). In addition, we compiled auxiliary evidence for the 36 KEs that were associated with B[a]P-induced toxicity through our systematic data integrative approach. Furthermore, we compiled information on the type of evidence and the reported toxicity dosage values of B[a]P exposure from these published evidence (Table S13). To conclude, we performed two case studies to explore both the human-relevant and ecotoxicology-relevant B[a]P-induced toxicity pathways from this directed AOP network.

Toxicity pathway linking B[a]P exposure to liver fibrosis in humans

Liver fibrosis, which results from chronic damage to the liver, is a characteristic of many chronic liver diseases (Bataller and Brenner 2005). Previously, exposome-based studies had found a significant association between environmental chemicals such as B[a]P and different liver diseases (Cheung et al. 2020; Barouki et al. 2023). Here, we observed an emergent B[a]P-induced toxicity pathway originating from MIE ‘Activation, AhR’ (KE:18) and terminating at AO ‘N/A, Liver fibrosis’ (KE:344). Therefore, we relied on this emergent toxicity pathway to understand the rationale behind B[a]P-induced liver fibrosis in humans.

Various in vivo and in vitro experiments in human cell lines and rodents had shown that B[a]P induces different downstream processes through the activation of AhR (Tsai et al. 2015; Lou et al. 2022; Jin et al. 2022; Almendarez-Reyna et al. 2023). Subsequently, B[a]P exposure has been observed to induce oxidative stress in cells through increased interleukin 6 (IL-6) production as a result of activated NF-κB signaling pathway (Malik et al. 2018; Jin et al. 2022; Zheng et al. 2022). The oxidative stress caused by B[a]P exposure has been studied as a cause for disruption of lysosomes, eventually leading to dysfunctional autophagy (Gorria et al. 2008; Li et al. 2023). Finally, it has been shown that B[a]P exposure can induce different fibrotic pathways, including dysfunctional autophagy, in human hepatic models (Yan et al. 2021; Hill and Wang 2022). In conclusion, by leveraging various published evidence, we were able to explore a potential toxicity pathway that links B[a]P-induced toxicity with liver fibrosis in humans.

Toxicity pathway linking B[a]P exposure to early life-stage mortality in aquatic organisms

B[a]P is found in large quantities in different aquatic environments due to various anthropogenic activities and waste discharges from both household and industries (Bukowska et al. 2022; Sathikumaran et al. 2022). B[a]P is a toxic pollutant, and drastically affects various aquatic organisms, including economically relevant fish (Nacci et al. 2002; Seemann et al. 2015; Sathikumaran et al. 2022). Here, we observed that the AOP titled ‘Aryl hydrocarbon receptor activation leading to early life stage mortality via sox9 repression induced impeded craniofacial development’ (AOP:455), with biological applicability for developmental effects in aquatic species, has been identified as a B[a]P–AOP (Table S11). Moreover, this B[a]P–AOP is part of the largest connected component, and is currently included in the OECD work plan (Table S3). Therefore, we relied on this AOP to understand the rationale behind B[a]P-induced ecotoxicological effects in aquatic organisms.

Independent in vivo experiments in zebrafish and clam have shown that B[a]P exposure alters gene expression patterns through activation of AhR and subsequent dimerization of AhR and ARNT in affected tissues (Bugiak and Weber 2009; Wang et al. 2020). It has been shown that B[a]P exposure in zebrafish facilitates the recruitment of AhR-dependent long noncoding RNA (slincR) to sox9b 5′ UTR, eventually repressing its transcription (Garcia et al. 2018). sox9b is an important transcription factor involved in chondrocyte differentiation during zebrafish development (Dalcq et al. 2012). Subsequently it has been shown that B[a]P exposure induces alteration in expression patterns of genes involved in chondrogenesis, thereby leading to improper craniofacial skeleton development and eventually early life stage mortality in zebrafish (He et al.