We reviewed previously published human VOCs mixture PBPK models (Dennison et al., 2005; Haddad et al., 1999, 2000, 2001; Marchand et al., 2015, 2016; Tardif et al., 1995, 1997). We recoded the BTEX model using Berkeley Madonna software (see Supplementary) version 8.3 for Windows (University of California at Berkeley, California). The recoded PBPK model presented here adapts the original model developed in AcslX (Haddad et al., 1999) using human physiological parameter values given in Table 1 (Marchand et al., 2015; Tardif et al., 1995, 1997). This model was used to predict alveolar air concentration levels in human volunteers exposed for 7h to toluene (17ppm), ethylbenzene (33ppm), and m-xylene (33ppm) alone or in combination.

The model was assessed by comparing generated simulations to the published model and with human data sets (Tardif et al.,1997). We weighed the performance of the model by calculating the percent median absolute performance error (MAPE%) based on estimates of performance error (PE) between the recoded model and the experimental data (Ruiz et al., 2011), as follows: (1)PE=(Cmeasured−Cpredicted/Cpredicted)∗100%.

Therefore, MAPE% = median (|PE₁|, |PE₂|, |PE₃|……|PE_i|…. |PE_n|), where PE_i is the performance error measure of the model prediction and data at time point i. We assessed the measure for the model bias by considering the median of the true values of PE. This measure is calculated as MPE% = (PE₁, PE2, PE3…PE_i…. PE_n). The accuracy of the prediction was evaluated by percent root median-square performance error (RMSPE%), as follows: (2)RMSPE%=∑i=1nPE2n, where n is the total number of data points. We also calculated the correlation coefficient between C_measured and C_predicted.

Traditionally, a hazard index (HI) is calculated using estimated exposures and allowable levels in environmental media. The American Conference of Governmental Industrial Hygienists (ACGIH) Threshold Limit Values (TLVs), and Biological Exposure Indices (BEIs) are intended for use by industrial hygienists in making decisions regarding safe levels of exposure to various chemical substances and physical agents found in the workplace. The TLVs and BEIs represent conditions under which ACGIH believes that nearly all workers may be repeatedly exposed without adverse health effects. BEIs are guidance values for assessing biological monitoring results (https://www.acgih.org/tlv-bei-guidelines/policies-procedures-presentations/overview; last accessed December 2019). The human PBPK models enable calculation of biological hazard indexes (BHIs). A BHI is based on estimated internal dose rather than applied dose.

We used the human PBPK model mixture to calculate BHIs for 8-h exposures to varying simulated mixtures of the 3 chemicals (5–40ppm toluene, 10–50ppm m-xylene, and 10–50ppm ethylbenzene) (Haddad et al., 1999; Tardif et al., 1997). We used the following equation to calculate the BHIs: (3)BHI=∑i−1nSCiBEIi, where SC_i is the simulated venous blood concentration of the component chemical (i) and BEI_i is the biological exposure index or blood concentration of the component chemical in a healthy person (ie, at the TLV for toluene [50ppm], m-xylene [100ppm], and ethylbenzene [100ppm]). Simulated venous blood concentration (SC_i) values were predicted for the mixture components using the ternary mixture PBPK model. Individual chemical PBPK models were used to calculate the TLV-based BEIs. The BHIs were subsequently compared with exposure concentration-based HI values for each mixture, calculated as follows, where E_i is the exposure level of the chemical: (4)HI=∑i−1nEiTLVi.

Under Occupational Safety and Health Administration and American Conference of Governmental Industrial Hygienists guidelines, the mixture formula (unity calculation, UC) is used to evaluate exposures to mixtures of chemicals that cause similar toxicities (Dennison et al., 2005). Using this methodology described in the guidelines, the overall exposure is acceptable if exposures are reduced in proportion to the number of chemicals and their respective exposure limits. Most of the occupational exposure limits are derived from studies of humans or animals at rest, often using inhaled concentrations to estimate external exposures. For a variety of exposures to toluene, ethylbenzene, or xylene, UCs were performed, using the equation below (Dennison et al., 2005): (5)UC=∑UiOULi.

U_i is the exposure level of each chemical and OUL_i is the occupational expose limit for each chemical. We used the human PBPK model to examine the potential toxicity from coexposure to TEX under resting and working conditions. Instead of using inhaled concentrations to estimate external exposure, we utilized the PBPK model to determine the blood concentration of each chemical and in the UCs. The UCs were based on internal doses of each chemical instead of the traditional external exposure level, thus considering nonlinear PK and the influence of PK interactions on cumulative dose. The chemical effects are assumed to result from systemic absorption and exposure of internal organs to the blood and not from direct toxicity of inhaled concentration at the port of entry.

An upcoming area of research is to input the potential concentrations derived from high-throughput assays such as microarray data, into the existing PBPK model to back-extrapolate external exposure concentrations that would produce equivalent target tissue concentrations research. Through an interactive process both the PBPK modeling and the experimental testing can be improved and/or calibrated. We used the PBPK model to initially simulate the starting points of equivalent external exposure for controls, short- and long-term exposure groups as reported by Hong et al., to estimate potential distribution of blood concentrations (more reflective of internal estimated dose). Initially the simulation was performed for one-third TLVs doses. Following this, different scenarios for incremental changes in TLVs were simulated.

The CTD(http://ctdbase.org; last accessed December 2019) contains more than 1 million curated chemical-gene interactions and more than 15 million curated and inferred gene-disease interactions (Davis et al., 2017; Grondin et al., 2018). The National Institute of Environmental Health Sciences funds the database, which can be used to identify genes or environmental chemicals that commonly affect the proteins that they code for. These sets of common genes can be further filtered to only include genes that have been associated with a condition or disease. This approach uses bioinformatics to identify toxicogenomic interactions that are promising targets for further study. The CTD was used to identify (1) top curated genes for each of the 3 VOCs, toluene, xylene, and ethylbenzene, (2) genes in common to them, and (3) top interacting curated diseases in common to all the individually VOCs.

We used the CTD “VennViewer” function to compare the genes associated with each of the VOCs, TEX. Only genes with curated associations with each chemical were included. Genes with an inferred relationship or those with no known effects were excluded. The candidate genes associated with each of 3 chemicals were identified; these genes were then used as an input for the “Set Analyzer,” which identifies common genes that have an association with specific diseases or health conditions. The list of genes that were common amongst the 3 VOCs was imported to MetaCore (version 6.36 build 69400) software to perform an enrichment analysis (https://clarivate.com/cortellis/solutions/early-research-intelligence-solutions/; last accessed December 2019). MetaCore is a tool for functional analysis of different “omics” data, including gene expression data. One of MetaCore’s relevant applications for pathway analysis is the Enrichment Analysis Workflow, which allows the users to understand the biological impact of their data by visualizing the intersection of their data set to curated ontologies which are ranked by significance based on p value (hypergeometric mean calculation).

These curated ontologies (eg, pathway, maps, toxicity biomarkers, networks) are primarily derived from peer-reviewed literature and from the Gene Ontology lexicon.

Hong and colleagues reported the effects of TEX exposure in humans on gene expression data for short-term (workers exposed to VOCs for less than 10years), long-term (workers exposed to VOCs for more than 10years), and no exposure (control) (Hong et al., 2016). We downloaded their microarray data from the GEO Database (www.ncbi.nlm.nih.gov/geo/; last accessed December 2019) ID GSE68906, which describes the differentially expressed genes (DEGs) in human blood samples collected from industry workers at a factory and chemical production company in Ansan, Korea (Hong et al., 2016). The gene expression data were imported as experiment files into MetaCore, and genes showing statistically significant expression changes within the exposed groups were compared.

MetaCore’s system’s toxicology module, which utilizes organ-specific toxicity biomarkers curated for different ontologies, was used to perform enrichment analysis on lung-specific ontologies. We identified and rank cellular pathways and processes most impacted in this tissue by the gene expression changes in the exposed group.

Our enrichment analysis focused not only on the genes that were common to both long- and short-term exposures, but also on the genes that were unique to the long-term exposure.

Our CTD analysis identified a list of 24 genes common among TEX (Figure 3A). Individually, 826 genes were associated with toluene, 109 genes with ethylbenzene, and 55 genes with xylenes. We have also identified 2 conditions, potentially associated with the overlap of the curated diseases potentially cause by these 3 chemicals that include prenatal exposure delayed effects and micronuclei chromosomal defects (Figure 3B). To understand what pathways are impacted by the list of 24 genes, we performed pathway enrichment analysis using MetaCore. The names, symbols, and functions of the 24 genes are shown in Figure 3A.

Based on our enrichment analysis, a joint mechanism of action was proposed for the combined exposure to TEX mixture. The pathways map shows cross interactions between the downstream targets (Supplementary Figure 1). The analysis shows that the top score enriched pathway map is involved in neural cell death and mitochondrial dysfunction (Supplementary Figure 1). Ubiquitin with dinucleotide deletion acts as a strong inhibitor of the proteasome. As a result, the pool of available proteasomes becomes substantially depleted and not enough for proper degradation of tumor protein (p53), protein kinase C-alpha (PKC-alpha), and mutant Huntingtin. Slow degradation of PKC-alpha results in activation of N-methyl-D-aspartate subclass of ionotropic glutamate receptor (NMDA receptor), followed by mitochondrial dysfunction, alterations in neurotransmitter systems, and neuronal cell death (Fong et al., 2002; Zemskov and Nukina, 2003). In addition, impairing of proteasomal function leads to alterations in neurotransmitter pathways (Wang et al., 2008). These common activation processes can influence or activate each other. Although this computational systems biology-generated global pathway map cannot be considered as a proof of causal linkages without further experimental validation, it provides justification for the mechanistic hypothesis and contributes to new interpretation linking available published toxicology and disease information domains.

To understand the toxic impact of VOCs to the target organ lung, we used the system’s toxicology module of MetaCore to compare gene expression data collected from blood samples of individuals exposed for short- and long-term to TEX. The comparison analysis demonstrates that 236 genes are common between the short-term (1) and long-term exposure groups (2) gene expression (Figure 4A). It also shows that 242 genes are unique to short-term exposure (1, orange bar), and 225 genes are unique to long-term exposure (2, blue bar). These findings help explain the overall biological themes represented in each of the studied exposure group networks. The results of the enrichment analysis ranked pathway maps specifically curated for lung toxicity. Ranking was determined by the genes that were common to blood samples of individuals exposed to short- and long-term exposures (Figure 4B, blue stripe bars) and presented from highest log p value to the lowest. Orange and blue bars correspond to log p values that are unique to enrichment of genes from short- and long-term exposures, respectively.

The 2 top-scored lung toxicity pathways (maps with the lowest p values), based on the enrichment distribution sorted by “common” set, showed common themes related to processes involved in respiratory, developmental, and the immune systems (Figure 4B). The 2 top pathway maps are Ephrin signaling and CD28 signaling; these are involved in cell adhesion, cell proliferation, and immune response. Ephrin receptors are major participants in the regulation of cell movement, cell adhesion, and cell survival, whereas CD28 plays a key role in many T-cell processes related to immune signaling pathways.

The first top-scored map (corresponding to the lowest p value) shows Ephrin-mediated signaling (Figure 4B). Short- and long-term exposures upregulate the Ephrin A receptors, as indicated by the red thermometers (Nos 1 and 2, respectively) and downregulate Ephrin B receptors as indicated by the blue thermometers (Nos 1 and 2) as shown in Supplementary Figure 2. Ephrin B proteins have conserved cytoplasmic tyrosine residues that are phosphorylated upon interaction with an EphB receptor and may transduce signals that regulate a cellular response. However, long-term exposures downregulate gene paxillin (see thermometer No. 2, Supplementary Figure 2).

The second top-scored pathway map (second lowest p value), CD28 signaling pathway, is related to induction of immune responses (Supplementary Figure 3). The T-cells receive 2 sets of signals from antigen-presenting cells. The first signal is delivered via T-cell receptor complex (TCR), the second one proceeds via coreceptor CD28. T-cell receptor and CD28 are independent signaling units. However, CD28 amplifies signals triggered by TCR ligation. CD28 is a T-cell surface protein activated by interacting with the B-cell activation antigens CD80 and CD86. The CD28 and CTLA-4 signaling pathways are important participants in a very complex group of regulatory events that maintain immunologic homeostasis. Lung toxicity biomarkers for lung interstitial congestion and lung perforation are major processes in regulation of macrophage migration, suggesting inflammatory response plays a large role in modulation of lung toxicity.

When taking a closer look just at the unique long-term exposure data, the enrichment analysis shows some additional interesting results such as xenobiotic metabolism enzymes are impacted by long-term exposure (Supplementary Figure 4.). The top-scored pathway map (indicating by the lowest p value) for the long-term exposure data is the peroxisome proliferator-activated receptors (PPARs) transcription pathway (Supplementary Figure 4). The PPARs modulate the expression of genes involved in cell proliferation, lipid metabolism, and inflammation.

The top-scored map (corresponding to the lowest p value) for long-term exposure (Supplementary Figure 4) shows that Mitogen-activated protein enzymes (MEK3) are activated in the long-term exposure data set which is an upstream activator of PPARc (Supplementary Figure 4). PPARα activates CYPA13 via transcriptional regulation, as seen in the MetaCore interactions database. As seen in the Supplementary Figure 4, the protein kinase inhibitor of PPARα is downregulated. CYPA13, which is known to metabolize carcinogens in cigarette smoke, is upregulated, whereas CYP4F2 and CYP4F3 are downregulated, both of which are involved in inactivating the inflammatory mediator leukotriene B4. Downregulation of CYP4F2 and CYP4F3 would result in an increased inflammatory response via leukotriene B4.

This analysis identified 24 genes that are commonly regulated by each of the VOCs and provided insight into potential pathways where in vivo effects have been observed, while recognizing that TEX mixtures would not necessarily act on the same pathways in the same ways to result in additivity effects. The enrichment pathway analysis using the 24 common genes showed that TEX compounds impact the Ubiquitin pathway. This provides evidence that combined exposure of TEX could result in synergistic activation or deactivation of this pathway. The pathway seems relevant to neurological functions and may be relevant in establishing important adverse outcomes for TEX (Supplementary Figure 1).

Competitive metabolic inhibition is the most plausible mechanism of interaction at higher concentrations among the TEX components based on the PBPK studies, as well as in vitro and in vivo metabolism and toxicity studies for some of the binary component mixtures (Tardif et al., 1997). Therefore, due to the apparent lack of competitive metabolic interactions in TEX mixtures below approximately 20ppm of each component (one-third of TLVs), it is plausible that joint neurotoxic actions among the chemicals will be additive at environmental levels of exposure. Exposure to higher concentrations of TEX components (ie, above the threshold for metabolic inhibition) would be expected to lead to greater than additive increases in blood levels of parent compounds and, consequently, increased concern for neurotoxicity. However, it is unclear whether the PBPK model descriptions are adequate for predicting interactions from inhalation of TEX mixtures above approximately 200ppm of each component, because it has not been tested for these high doses (corresponding to the dose reducing the metabolic rate). At a certain dose, the enzymatic saturation is reached, reducing the metabolic rate formation and that may increase the toxicity of the parent compounds.

Studies that directly examined the joint toxic action of TEX chemicals on the nervous system are limited to a few human and animal inhalation studies of some binary mixtures of components. Neurotoxicity studies of the ternary mixtures provide agree with the predictions of the PBPK studies (ie, joint action is expected to be additive at TEX concentrations below approximately 20ppm of each component).

If the concentrations in blood are below the competitive inhibition threshold for each of the components, no change can be expected in the proposed pathways leading to toxicity. However, such changes in constant concentration below the threshold might have effects that will cause some health outcome at long-term exposure.

The CTD results highlighted interconnections, also evident from the independent data analysis from Hong et al. study for the up- and downregulated genes related to neurological diseases, embryonic development, cellular compromise, cellular adhesion, and cellular movement effects.

An expressed gene set list (up- and downregulated) at the average short- and long-term exposed group were subdivided into 9 cluster genes using a hierarchical clustering analysis (Hong et al., 2016). In addition, they reported the functional categories and annotations of these genes and created a gene function network that show how key components (consisting of 5 up- and 12 downregulated genes that also exhibit time-dependent changes in methylation) of different pathways interact using DAVID and Ingenuity tools.

By applying enrichment analysis on lung-specific ontologies, this study goes beyond just listing of DEGs to understand the mechanism of the biological impact on the lung tissue. Both approaches, enrichment analysis and gene set analysis are supposed to complement to take gene expression levels and leverage existing knowledge about a given organism to identify the underlying biological processes and mechanisms.

Using the system’s toxicology module of MetaCore, our study used enrichment analysis to understand the toxic impact of TEX to the target organ, lung. As mentioned in the “Materials and Methods” section, the enrichment analysis takes into consideration the number of DEGs that fall on a given curated ontology and uses hypergeometric mean to calculate the significance of this enrichment.

This analysis showed 2 top-scored lung toxicity pathways (maps with the lowest p values), based on the enrichment distribution sorted by “common” set (short- and long-term exposures), were Ephrin signaling and CD28 signaling. These pathways are involved in cell adhesion, cell proliferation, and immune response, suggesting that they may play a crucial role in lung toxicity. Consistent with this hypothesis, Ephrin A receptor stimulation has been previously shown to increase lung vascular permeability (Carpenter et al., 2012; Larson et al., 2008).

In addition, we performed a similar enrichment analysis for the genes that were to unique long-term exposure data, which suggests that xenobiotic metabolism enzymes are impacted by long-term exposures. The top-scored pathway was the PPARs transcription pathway (Supplementary Figure 4). Peroxisome proliferator-activated receptors are nuclear hormone receptors and are ligand-induced transcription factors and, whereas these were not sampled directly in the data set, many upstream interactors (MEK3, AKT, and PKA-reg) of these transcription factors were shown to be upregulated in the long-term exposure group. As shown in Supplementary Figure 4, PPARs play a crucial role in signaling of variety of different biological processes, including inflammation, and cancer. It is therefore not surprising that the PPARs have been explored as candidates for therapeutic intervention to treat variety of lung diseases such as asthma and non-small cell lung cancer (Banno et al., 2018; Li et al., 2011).

Based on our enrichment analyses using the gene lists identified from CTD and Hong et al. data, we hypothesize that exposure to TEX mixture may result in disruption of biological pathways such as Ubiquitin, Ephrins, CD28, and PPARs. Disruption of these proposed pathways can translate into adverse respiratory and neurological outcomes, depending on exposure durations. The different genes mapped to these biological pathways have been related to alterations of the neurotransmitter system including neural cell death, cell proliferation, inflammation, xenobiotic metabolism, signal transduction, and cell processes. These proposed pathways are based on experimental data from multiple laboratories and have not been examined sufficiently to date comprehensively by an integrated research laboratory. Future experimental evaluation of these pathway maps might lead to the development of new predictive markers of TEX effects that could translate into new disease prevention and clinical use strategies. Specific avenues of laboratory research might include, but not limited to, in vitro studies of target cell populations such as lung cells. Complimentary in vivo studies in rodents dosed with TEX mixture could be performed to determine if the observed in vitro study findings are observed after in vivo exposure and can be extrapolated to human exposure scenarios.

The enrichment analysis performed using CTD and the MetaCore analyses provides mechanistic insights of TEX mixture and could help the PBPK model to elucidate the biological relevant dosimetry. To use this approach a comprehensive characterization of exposure is needed, because humans are exposed to multiple chemicals from multiple sources through their daily activities.

The estimated blood concentrations from the recoded PBPK model represent a gold standard to understand or interpret the gene expression results driven health effects. Using the model, we generated different scenarios of TEX mixture to see the influence of dose and time in the blood concentration (Table 3). In the Hong et al. study, workers were exposed short- or long-term at doses we assumed were close to the TLV values. It is interesting to observe that each chemical from the mixture easily reaches a plateau (C_max) very quickly (approximatively few days). This suggests that the maximum blood concentrations will not change whether workers are exposed for a month or a year (Table 3). This observation also indicates that the metabolic interaction does not seem important if exposure occurs to the combination of these VOCs and seems to have a limiting impact on the blood concentration. In fact, another parameter that also influence the concentration of chemical in blood is the partition coefficient in tissue. This parameter for each tissue will influence not only the concentration of parent VOCs in blood, but also the availability of the parent VOCs to be metabolized.

Further research is needed to confirm or validate this proposed approach through experimental testing using internal doses estimated by the PBPK model.

The merit of this approach lies in advancement of mixtures risk assessment methodologies using the available current knowledge of the TEX mixtures toxicity, using contemporary alternative toxicity testing data, and applying computational methods. This work provides a Berkeley Madonna BTEX human PBPK model code to obtain a better dosimetry than external environmental exposures. Second, the enrichment analysis provides a possible pathway for the short- and long-term effects of TEX mixtures.