Computational Approach to Link Chemicals in Anthropogenic Smoke Particulate Matter with Toxicity

Abstract

A weighted chemical coexpression network analysis (WCCNA) was utilized to identify chemicals co-modulated to variable burning of anthropogenic materials and to link chemicals to biological responses (lung toxicity and mutagenicity). Polyaromatic hydrocarbons (PAHs) were co-modulated with increased concentrations in flaming smoke particulate matter (PM) from the burning of plastic-containing materials and showed significant association with increased neutrophil influx, cytokine levels, and mutagenicity. Inorganic elements were co-modulated with increased concentrations in flaming plywood and cardboard smoke PM and showed significant association with increased protein and albumin levels. This study shows the potential for using a computational network analysis to identify and prioritize hazardous chemical components within complex environmental mixtures and provides guidance on key chemical tracers required for intervention research to protect public health from the exposure.

Graphical Abstract

It is well recognized that smoke from open burning of anthropogenic materials such as plastic and plywood can be harmful to human health and the environment.¹ Anthropogenic smoke is emitted from not only burning of the solid portion of wastes (e.g., burn pits) and man-made materials² but also natural wildfires that spread into wildland-urban interface (WUI)³ communities with burning homes, businesses, and motor vehicles, among other items. Notably, chlorinated dioxins, furans, heavy metals, and per-/polyfluoroalkyl substances (PFAS) can be emitted from these fires, raising the possibility for more toxic emissions.^4,5 Moreover, due to the dynamic and chemical complexity of the smoke emissions, consequent health impacts following exposures vary greatly across different source materials.⁶ Thus, it is challenging to identify key chemical drivers of possible adverse health effects associated with anthropogenic smoke exposures.

Data-driven bioinformatics approaches using correlation-based network analysis can help overcome this challenge. Weighted gene coexpression network analysis (WGCNA) has been effectively used to identify the significant modules (clusters) and hub genes associated with biological phenotypes.⁷ This analysis also can be applied to high-dimensional chemistry data referred to as weighted chemical coexpression network analysis (WCCNA).^8–11 In this analysis, clustering is used to group chemicals that have a similar expression pattern in multiple smoke samples and analyze correlations between individual chemicals and biological responses. We previously used WCCNA to identify clusters (modules) of highly interconnected chemicals across biomass smoke particulate matter (PM) to determine links between the chemicals and lung toxicity end points associated with the smoke exposure.¹¹

To expand the application of WCCNA, we leveraged existing data from a recently published study from our group, where anthropogenic smoke emissions were generated, and their toxicity (lung toxicity and mutagenicity) was evaluated using in vivo and in vitro assays.¹² In this study, we used correlation-based network analysis to identify chemical sets co-modulated in association with anthropogenic emission sources and to better understand relationships between chemicals present in the smoke PM and associated toxicity responses.

Identifying Individual Chemicals Associated with Biological Responses.

To relate chemical components of the anthropogenic smoke PM to lung toxicity responses after exposures, we first characterized 53 chemicals across 8 different smoke PM samples and evaluated their exposure patterns through WCCNA in an unsupervised manner (Tables S1 and S2). We found that 5 different chemical groups (designated as “modules” to different colors) were co-modulated across PM exposures (Figure 1, Table S3). The turquoise module contained chemicals (mostly polyaromatic hydrocarbons (PAHs) and nitro-PAHs) that co-occurred at relatively high concentrations in flaming plastic smoke PM, while the blue, brown, and green modules contained chemicals (mostly inorganic elements) that co-occurred at relatively high concentrations in flaming plywood and cardboard smoke PM. Interestingly, inorganic elements co-occurred with carbonaceous matter (organic carbon) more often than PAHs. The flaming mixture smoke PM collected from burning of plywood, cardboard, and plastic contained relatively high concentrations of PAHs and inorganic and organic chemicals spread across the most modules. Because the smoldering smoke PM samples contained low levels of PAHs and inorganic elements on an equal mass basis compared to the flaming smoke,^12,13 they contained the lowest concentrations of chemicals across the modules.

Figure 1.

Chemical concentrations emitted in anthropogenic smoke PM across eight exposure types (left) and their correlations with lung toxicity outcomes (right), organized into WCCNA-derived modules. The number represents individual chemical components: 1, 1,4-naphthoquinone; 2, 1-naphthalenecarboxaldehyde; 3, 1-pyrenecarboxaldehyde; 4, 4-nitrobiphenyl; 5, 9-fluorenone; 6, 9-nitroanthracene; 7, acenaphthene; 8, acenaphthylene; 9, anthracene; 10, benzanthrone; 11, benzo[a]anthracene; 12, benzo[a]pyrene; 13, benzo[b]fluoranthene; 14, benzo[ghi]-perylene; 15, benzo[k]fluoranthene; 16, chrysene; 17, dibenzo[ah]anthracene; 18, elemental carbon; 19, fluoranthene; 20, fluorene; 21, indeno[123-cd]pyrene; 22, naphthalene; 23, phenanthrene; 24, pyrene; 25, Sb; 26, Ba; 27, benz[a]anthracene-7,12-quinone; 28, Ca; 29, Co; 30, Cr; 31, Cu; 32, Fe; 33, K; 34, Mg; 35, Ni; 36, organic carbon; 37, Pb; 38, Sr; 39, Zn; 40, Ag; 41, Mn; 42, Na; 43, P; 44, Sn; 45, Ti; 46, 1,8-naphthalic anhydride; 47, 9,10-anthraquinone; 48, Al; 49, Mo; 50, Si; 51, Bi; 52, Cd; 53, S.

We also utilized the WCCNA approach to identify chemical modules significantly correlated to increasing lung toxicity responses (Figure 1, Table S4). The turquoise module was identified as the chemical group showing the largest number of positive associations with increased neutrophil numbers (p = 0.03) and IL-6 (p = 0.03) levels (Table S4). However, the blue, brown, and yellow modules were also positively correlated with increased IL-6 (yellow: p = 0.01), MIP-2 (yellow: p = 0.05), LDH (yellow: p = 0.02), protein (p < 0.05), and albumin (brown and yellow: p < 0.05) levels in bronchoalveolar lavage fluid (BALF), suggesting that increased neutrophil influx was mainly triggered by PAHs (in the flaming plastic smoke) and that disruption of cell membrane integrity or vascular permeability in the lung was increased by inorganic elements (in the flaming plywood and cardboard) exposure. Overall, we identified groups of chemicals with similar concentration patterns in various combustion smoke that were positively or negatively correlated to lung toxicity outcomes. This mixture-based approach suggests that biological responses are largely in-line with the chemical structure or class rather than individual components which may act differently in the presence of additional chemicals.¹¹ Although this analysis provides valuable information on the potential joint effects of chemical mixtures in smoke, it can also be used to prioritize individual chemicals for their hazardous behavior.

The WCCNA approach was used to further identify which chemicals were most associated with the process of biological response to smoke exposure (Figures 1 and S1–S5). Specifically, 9-nitroanthracene (6 in Figure 1) and fluorene (20 in Figure 1) were identified at the highest relative concentrations within the flaming plastic smoke PM and as the main driver of the increased neutrophil number (p < 0.01; Table S5, red and blue lines in Figure S1A). They were, however, at relatively low concentrations within the flaming plywood and cardboard PM and were the minor factor increasing protein and albumin levels (p > 0.5; Table S5, red and green lines in Figures S1E and S2E, respectively). Instead, benz[a]anthracene-7,12-quinone, Cu, and Si (27, 31, and 50 in Figure 1, respectively) of the flaming plywood and cardboard PM were found to be major contributors to the protein and albumin leakage in the lung (p < 0.05; Table S5, purple, red, and orange lines in Figures S2E and S4E, respectively). This finding is consistent with previous studies that some PAHs (e.g., benzo[a]pyrene) can stimulate cells through the aryl hydrocarbon receptor (AhR),^14–16 resulting in an increased number of neutrophils in the lung. In addition, we previously reported that inorganic elements in the biomass smoke PM were significantly associated with increased protein and albumin levels in the lung.¹¹

Alongside the network analysis of the chemicals and lung toxicity responses, mutagenicity responses were also analyzed using the WCCNA approach (Tables S6 and S7). In this analysis, inorganic elements were excluded because mutagenicity outcomes were assessed from the exposure to extractable organic material (EOM) of the anthropogenic smoke PM.¹³ We characterized 27 chemicals across 10 different smoke PM and identified 1 chemical group (turquoise) co-modulated across PM exposures and 1 chemical group (gray) not considered to have any co-modulation patterns (Figure 2 and Table S8). Similar to the result described above, the turquoise module (mostly PAHs and nitro-PAHs) co-occurred at relatively high concentrations in the flaming plastic, mixture, and mixture/diesel smoke PM. Specifically the individual chemical-based analysis showed that 1-naphthalenecarboxaldehyde, 9-nitroanthracene, and acenaphthene (3, 7, and 8 in Figure 2, respectively) were identified at increased concentrations in those PM samples. However, 1,8-naphthalic anhydride and benz[a]anthracnene-7,12-quinone (2 and 11 in Figure 2, respectively) were the most abundant chemicals in the flaming plywood PM, and 4-nitrobiphenyl (26 in Figure 2) was found to be the major PAH in the flaming cardboard smoke PM. This suggests that open burning of plastic-containing materials generates more PAHs per unit mass PM than nonplastic combustions.

Figure 2.

Chemical concentrations emitted in anthropogenic smoke PM across ten exposure types and their correlations with mutagenicity outcomes, organized into WCCNA-derived modules. The number represents individual chemical components: 1, 1,4-naphthoquinone; 2, 1,8-naphthalic anhydride; 3, 1-naphthalenecarboxaldehyde; 4, 1-pyrenecarboxaldehyde; 5, 9,1-anthraquinone; 6, 9-fluorenone; 7, 9-nitroanthracene; 8, acenaphthene; 9, acenaphthylene; 10, anthracene; 11, benz[a]anthracene-7,12-quinone; 12, benzanthrone; 13, benzo[a]anthracene; 14, benzo[a]pyrene; 15, benzo[b]fluoranthene; 16, benzo[ghi]perylene; 17, benzo[k]fluoranthene; 18, chrysene; 19, dibenzo[ah]anthracene; 20, fluoranthene; 21, fluorene; 22, indeno[123-cd]pyrene; 23, naphthalene; 24, phenanthrene; 25, pyrene; 26, 4-nitrobiphenyl; 27, organic carbon.

We further identified mutagenic processes that were significantly correlated with chemical modules (Figures 2 and S6–S7). The turquoise module was identified as the most significantly positively correlated with mutagenicity outcomes in TA98 +S9 (p = 0.002) and TA100 +S9 (p < 0.001) strains and was also positively but not significantly associated with mutagenicity in TA98 −S9 (p = 0.2) (Table S9), suggesting that PAHs emitted from burning of the plastic-containing materials were primarily indirect-acting mutagens (requiring metabolic activation, +S9) in the Salmonella assay. The gray module was negatively correlated with mutagenicity in all 3 strains. Although the mixture-based analysis demonstrated that many PAHs in the flaming smoke PM did not significantly contribute to the mutagenicity in TA98 −S9, the individual-based analysis revealed that some oxygenated PAHs were uniquely correlated and involved in the processes of mutagenic response. Specifically, 1,8-naphthalic anhydride and benz[a]anthracene-7,12-quinone (2 and 11 in Figure 2, respectively) were identified to be major contributors to the mutagenicity in TA98 −S9 (p < 0.005; Table S10, red and blue lines in Figure S6B). This is consistent with our previous finding that nitro- or oxy-PAHs, which are direct-acting mutagens, played an important role in the mutagenicity in TA98 −S9.^12,13

Overall, in this study we used a computational network analysis with existing data sets to identify specific relationships between chemicals (classes or individuals) in anthropogenic smoke and associated biological (lung toxicity or mutagenicity) responses which could not be determined in traditional toxicity assessments. More importantly, as comprehensive chemistry and toxicity data sets from various traditional methods are becoming more accessible, computational modeling approaches show great promise to take advantage of publicly available data sets and prioritize toxic and hazardous chemicals for potential adverse health effects. Additionally, this approach can provide not only the toxic effects of individual chemical components (ranking hazardous chemicals) but also their effects in mixture combination environments which are more relevant to real-world chemical exposure situations. Taken together, we demonstrate that WCCNA can be used to better leverage information from chemical exposure and toxicology. Such approaches can also be used to help state and government officials make better decisions about protection of public health and the environment from complex exposures to harmful chemicals.