Air Pollution Triggers Metabolomic Signature of Mitochondrial Dysfunction in Mice and Humans: Implications for Cardiovascular Disease Development

Abstract

Background:

Previous animal studies found that subchronic exposure to diesel exhaust (DE) induces hyperlipidemia, accompanied by upregulation of 12- and 15-lipoxygenase (LOX) pathways and hepatic mitochondrial dysfunction. However, the human relevance of these findings has not been established.

Methods:

We studied ApoE−/− mice exposed to DE or filtered air (FA) for 2 weeks, and 26 healthy adults who traveled from Los Angeles (PM_2.5: 14.4 μg/m³) to Beijing (PM_2.5: 67.6 μg/m³) for 10 weeks. Both mice and humans were previously found to have increased 12- and 15-LOX metabolite levels but normal HDL and total cholesterol level in the blood after air pollution exposure. In this study, we profiled blood metabolomics and lipidomics across multiple platforms in mice and humans, and conducted integrated data analyses to identify common metabolic pathways that were affected by air pollution, mechanistically related to oxidative stress and hyperlipidemia.

Results:

Enrichment analysis of overlapping metabolites detected in both mice and humans indicates that air pollution induced metabolic alterations in (1) dicarboxylic acids (DCAs), (2) acyl-carnitines (ACs), (3) tryptophan, (4) pyrimidine, and (5) lysine pathways. Although the metabolomic signatures of tryptophan, pyrimidine, and lysine metabolites differed between mice and humans, we observed consistent increases in long-chain DCAs and medium- to long-chain ACs, likely due to mitochondrial dysfunction as evidenced by impaired mitochondrial respiration by a Seahorse assay on livers from the same mice. In the human study, the changes of DCAs and ACs were significantly associated with increased lipid peroxidation products from 12- and 15-LOX pathways and exposure biomarkers for polycyclic aromatic hydrocarbons.

Conclusions:

We provide real-world human evidence supporting that mitochondrial dysfunction and impaired fatty acid oxidation are plausible mechanisms mediating the adverse early metabolic effects of air pollution.

Graphical Abstract

Introduction

Air pollution, particularly particulate matter with an aerodynamic diameter <2.5 μm (PM_2.5), is among the leading environmental risk factors of cardiovascular disease, responsible for 90.5 million disability-adjusted life years caused by cardiovascular diseases in 2023.¹ The cardiovascular effects of PM_2.5 may begin early in life and accumulate throughout lifetime, ultimately leading to adverse cardiovascular outcomes (e.g., metabolic disorders, ischemic heart disease, and death).^2–4 Thus, interventions to prevent early cardiovascular effects induced by air pollution could have long-term health benefits by altering the overall disease progression. A mechanistic understanding of air pollution’s early cardiovascular effects can help developing biomarkers to evaluate cardiovascular risk and inform on strategies to mitigate the risk.

We have previously conducted a series of mouse studies to examine the chronological order of cardiovascular effects induced by particulate matter pollution. We found that two-week exposure to diesel exhaust (DE) led to decreased paraoxonase 1 (PON1) activity, increased lipid peroxidation via 5-, 12-, and 15-lipoxygenase (LOX) pathways, development of dysfunctional high-density lipoprotein (HDL), and hepatic mitochondrial dysfunction.^5,6 Five-week exposure to ambient ultrafine particles also induced lipid peroxidation and HDL dysfunction together with larger atherosclerotic lesions.⁷ Although neither 2-week DE exposure nor 5-week ambient ultrafine particles caused changes in total cholesterol or HDL levels,^5,7 16-week exposure to DE led to hyperlipidemia and liver steatosis accompanied by lipid peroxidation via 12- and 15-LOX pathways and suggestive evidence for hepatic cell mitochondrial dysfunction.⁸ These results suggested that the sequential early cardiovascular effects of air pollution involves the progression of heightened oxidative stress via 12- and 15-LOX pathways to mitochondrial dysfunction and impaired fatty acid oxidation, ultimately leading to hyperlipidemia in mice. However, the human relevance of this mechanism remains not established.

In a previous natural experiment among 26 healthy Los Angeles residents who traveled to more-polluted Beijing for 10 weeks, we found that traveling to Beijing led to significantly decreased PON1 activities and increases in lipid peroxidation metabolites from LOX pathways as well, in association with increased exposure to polycyclic aromatic hydrocarbons (PAHs; a combustion-originated component of PM_2.5), while the levels of total cholesterol and HDL were unchanged.⁹ The cardiovascular phenotypes of these healthy adults following air pollution exposure are highly similar to those of ApoE^−/− mice exposed to DE for two weeks.^5,9 In this study, we utilized a mouse-to-human translational framework to identify metabolic pathways potentially linking lipid peroxidation induced by air pollution with subsequent cardiovascular events such as hyperlipidemia. By profiling the blood metabolomics in both mice and humans, we (1) identified common metabolic changes induced by air pollution; (2) confirmed the biological process driving the metabolic changes in experimental conditions using animals; and (3) explored the associations of metabolomic signatures with a panel of exposure and cardiovascular biomarkers in real-world settings in humans.

Materials and Methods

Data Availability

Data from this study are deposited in and publicly available through the Harvard Dataverse (https://doi.org/10.7910/DVN/VTYZFW).

Animal Experiments

ApoE^−/− (C57BL/6J background) mice were obtained from the Jackson Laboratory (Bar Harbor, ME), and were bred in the University of Washington South Lake Union animal facility. This study used blood samples collected from male mice after controlled exposure to DE or filtered air (FA) as described previously.⁵ Briefly, two experiments were conducted. In the first experiment, mice were assigned to three groups that were exposed to: (A) DE for 2 weeks (n= 12), (B) FA for 2 weeks as controls (n=13), and (C) DE for 2 weeks followed by FA for 1 additional week (n= 13). In the second experiment, mice were assigned to two groups that were exposed to almost identical conditions as groups A (n = 10) and B (n= 12) of the first experiment. In both experiments, mice were housed in a temperature- and humidity-controlled, specific pathogen-free environment with a 12-h light/dark cycle. At ~ 8 weeks of age, the mice were kept in standard mouse cages (up to 4 mice/cage) and moved into a self-enclosed “Biozone” facility connected to the DE generating system. DE exposures were controlled by opening or closing a valve to the “Biozone” resulting in minimal stress for animals during the exposure period. Exposures were started after a 7-day period of acclimatization in the “Biozone” and consisted in sessions of 6 hours/day and 5 sessions/week. DE was generated with a single cylinder diesel engine generator set with a maximum electrical power output of 5.5 kW (Model YDG5500EV-6EI) was titrated to achieve a concentration of ~ 250 μg/m³ PM_2.5 in the DE exposure system section, reaching an average PM_2.5 concentrations of 258 ± 39 and 253 ± 5 μg/m³ in the first and second animal experiments, respectively. Immediately following the last exposure session and ~ 8 hours following the last diet, blood was drawn from the retro-orbital sinus of each mouse. They were allowed access to water and standard rodent chow diet (PMI 5053, LabDiet) during non-exposure periods. Animal procedures were approved by the Animal Care and Use Committee of the University of Washington.

Natural Human Experiments.

The human study was based on 26 healthy young adults (14 women and 12 men) who traveled from Los Angeles (PM_2.5 levels ~ 14.4 μg/m³) to Beijing (PM_2.5 levels ~ 67.6 μg/m³) for ten weeks in a student-exchange program between the University of California, Los Angeles and Peking University in the summers of 2014 and 2015 as described previously.⁹ Compared to Los Angeles, the ambient levels of criteria gaseous pollutants were also drastically higher in Beijing, except for ozone’s levels that were only modestly higher in Beijing (Table S1).⁹ The average (standard deviation) age was 23.8 (5.6) years, and the body mass index was 21.6 (2.4) kg/m². All participants had no history of smoking, cardiometabolic diseases, or chronic inflammation during the previous six months. From each participant, we collected paired urine and peripheral blood samples before departing Los Angeles (LA-before), 6–8 weeks after arriving at Beijing, and 4–7 weeks after returning to Los Angeles (LA-after). All samples were collected in the morning between 8 a.m. and 10 a.m. after fasting for at least 8 h to minimize potential effects induced by circadian rhythms or diets. The study was approved by the institutional review boards of both University of California, Los Angeles and Peking University. All participants signed a written consent form upon enrollment.

Metabolomics Data Acquisition and Processing.

In the second animal experiment, nontargeted metabolomic analyses were performed on a subset of plasma samples (n=5 for each group) using the Metabolon platform. Briefly, samples were spiked with several internal standards for quality control purposes. Then protein precipitation was conducted with methanol under vigorous shaking followed by centrifugation. Five aliquots of the extracts were obtained for mass spectroscopy (MS) analyses under five conditions: (1) Reversed-phase liquid chromatography (LC) and positive electrospray ionization, (2) reversed-phase LC and negative electrospray ionization, (3) hydrophilic interaction LC and positive electrospray ionization, (4) hydrophilic interaction LC and negative electrospray ionization, and (5) gas chromatography MS.

In the human study, nontargeted metabolomic analyses were performed on all serum samples collected (n=72) from study participants at Peking University and University of California Davis West Coast Metabolomics Center.¹⁰ Briefly, samples were randomized in order and added a solvent mixture of methanol and methyl tertiary-butyl ether, followed by 10-s vortexing and 5-min shaking. Then water was added, and the mixture was vortexed for 20 s and was centrifuged. The supernatant solvent was separated for the reversed-phase LC-MS analysis while the bottom aqueous phase was collected for the hydrophilic interaction LC-MS analysis. Both analyses were conducted under both positive and negative electrospray ionization modes. The aqueous phase was also analyzed by a gas chromatography MS.

In both animal and human studies, the processing of raw metabolomics followed standard and validated procedures as described previously.^10,11 Since the metabolomic profiling of animal and human samples was conducted at different laboratories, this study only includes annotated metabolites with confirmed structure with authentic standards or putative chemical ID identified in either animal or human study. In the human study, we also retrieved data from our previous metabolomic profiling using LC-time-of-flight-MS that includes 50 annotated metabolites with confirmed structure with authentic standards.¹²

Accessorial Data.

We retrieved data collected from the same animal and human studies as previously reported by us, including (1) systemic cardiovascular biomarkers measured in the first animal experiment and in humans, namely total cholesterol, HDL, PON1 activity, 5-,12, and 15- hydroxyeicosatetraenoic acids (HETEs), 9- and 13-hydroxyoctadecadienoic acids (HODEs), 8-isoprostane, and acute phase proteins (i.e., C-reactive protein [CRP] in human and its mouse-equivalent serum amyloid protein);^5,9 (2) hepatic mitochondrial functional measures and liver metabolomics in the second animal experiment and transcriptomics in the first animal experiment;⁶ (3) urinary exposure biomarkers of 29 chemicals in the human study, including five PAHs,⁹ cotinine,¹³ bisphenol a,¹⁴ and 22 elements.¹⁵

Statistical Analysis.

Identification of Common Metabolic Alterations.

This analysis is based on a subset of metabolites that have a chemical ID (e.g., PubChem ID) and were detected in both animal and human studies. First, we conducted metabolite-specific analysis to calculate the fold-change of individual metabolites in response to air pollution. In the animal study, we used linear regression models in which the metabolite concentration is a function of exposure (FA and DE). In the human study, we used linear mixed-effects models in which the metabolite concentration is a function of study phase, treated as a categorical variable (LA-before, Beijing, and LA-after), with random intercepts of participants. In the human study, we previously observed no sex-specific differences in travel-induced changes in inflammatory and lipid peroxidation biomarkers.⁹ Therefore, data from both men and women were combined to increase statistical power. Second, the metabolite-specific fold-changes and p-values were used for enrichment analyses using the ChemRICH algorithm, which calculates the pathway-level p-values using the Kolmogorov–Smirnov test.¹⁶ We used the Metabolon sub-pathway ontology for both animal and human metabolomic data and the enrichment analyses for two datasets were performed separately. The subsequent analyses only focused on endogenous sub-pathways that were changes in both animals and humans.

Metabolomic Signatures in Associations with Other Factors.

This analysis focused on common endogenous sub-pathways influenced by air pollution and included all metabolites from these pathways if they were detected in either animal or human study. We examined metabolite-specific changes in animals (DE vs. FA) and humans (LA-before vs. Beijing vs. LA-after) using the same model as described above. In the animal study, we examined the association between individual metabolites and functional measures in liver and blood using linear regression models. In the human study, we examined the association between individual metabolites and urinary exposure biomarkers using mixed-effects models with random intercept of participants and study phases.

For all statistical analyses, we calculated the confidence intervals for the effect size without multiple-comparison adjustment, and used the Benjamini-Hochberg method to control an overall false discovery rate (FDR) of 5%. All statistical analyses were performed using the statistical software R with the ChemRICH script and lme4 and lmeTest packages (www.r-project.org).

Results

Cardiovascular Responses to Air Pollution in Animals and Humans.

We compared our previously reported cardiovascular effects of air pollution in mice versus in humans as summarized in Table S2.^5,9,17 The results indicate consistent increases in a panel of circulating lipid oxidative biomarkers consisting of various HETEs, HODEs, and 8-isoprostane. In addition, we have previously reported increased 5-HETE levels in the liver of DE-exposed animals due to activation of the 5-LOX pathway. In humans, there were increased 5-HETE concentrations in plasma without significant changes in arachidonic acid levels. These results indicate heightened lipid peroxidation induced by air pollution via lipoxygenase and/or non-enzymatic pathways in both mice and humans. Consistently across mice and humans, air pollution did not cause significant changes in circulating total or HDL cholesterol levels but decreased enzymatic activities for PON 1 in plasma. In mice, air pollution impaired HDL functionality. We observed significant increases in circulating levels of CRP and interleukin-8 (IL-8) in humans after traveling from Los Angeles to Beijing, while the serum amyloid A protein level remained unchanged following DE exposure in mice. Of note, presence of endotoxins have been confirmed in fine particles collected in Beijing as reported previously,¹⁸ which may contribute to the inflammatory responses in humans. In addition, the human participants were exposed to air pollution for a longer period (i.e., 6–8 weeks) as compared with mice (i.e., 2 weeks), which may explain why inflammatory responses were observed in humans but not in mice. In the human study, all travel-induced changes in oxidative and inflammatory biomarkers were reversed, at least partially, after participants returned to Los Angeles.

Common Metabolomic Responses to Air Pollution.

We focused on annotated metabolites with a confirmed structure by authentic standards or putative chemical ID, and identified 547 metabolites in mouse plasma and 1357 metabolites in human serum. Among these metabolites, 154 (mostly lipids, amino acids, and nucleotides) were commonly detected in both mouse and human specimens (Figure S1). We examined concentration changes of these 154 metabolites after 2-week exposure to DE (for animals) or traveling to Beijing (for humans) (Figure 1) and did not observe clear overlapping metabolic alterations between the two studies except for a common increase in deoxycholate levels. In the pathway enrichment analysis, several pathways affected by DE exposure in mice were related to lipid metabolism (Figure 2). In contrast, traveling from Los Angeles to Beijing led to changes in many pathways related to xenobiotic or amino acid metabolism (Figure 2). Besides air pollution levels, participants’ dietary patterns and exposure to other xenobiotics likely differed between Los Angeles and Beijing, which can explain the metabolic alteration of these pathways in humans. We identified three endogenous pathways that significantly changed in both animals and humans at FDR < 5% levels, namely dicarboxylic fatty acid (DCA) metabolism, tryptophan metabolism, and pyrimidine (uracil containing) metabolism (Figure 2). In addition, the metabolism of acylcarnitine (AC) and lysine was changed in animals at FDR < 5% levels and in humans at p<0.05 levels. Metabolites from these five pathways were further examined in the subsequent analyses.

Figure 1. Volcano plots on 154 common metabolites in animals (panel A, DE vs. FA) and humans (panel B, Beijing vs. LA-before).

Metabolite-specific fold-changes in animals (DE vs. FA) and humans (Beijing vs. LA-before) were calculated using linear regression models and mixed-effects models with random intercepts of participants, respectively. Black symbols indicate results for ACs and DCAs. Red and blue symbols indicate upregulated and downregulated metabolites (fold change > 2 and p < 0.05, excluding ACs and DCAs), respectively. Grey symbols indicate metabolites that were unchanged (fold change < 2 or p > 0.05, excluding ACs and DCAs). Number of mice: n = 5 /group. Number of human participants: n=26.

Figure 2. Pathway enrichment analysis of 154 common metabolites in animal plasma (A) and human serum (B) samples in response to subchronic air pollution exposure.

The p-value is from a Kolmogorov–Smirnov test in the enrichment analysis and the FDR is calculated using the Benjamini-Hochberg method. Number of mice: n = 5 /group. Number of human participants: n=26.

Metabolomic Signature of Overlapped Pathways.

We identified 20 metabolites from DCA pathways, 8 metabolites from tryptophan pathways, 8 metabolites from pyrimidine (uracil containing) pathways, 36 metabolites from AC pathways, and 5 metabolites from lysine pathways that were affected by DE in mice and PM_2.5 in humans. In general, concentration changes of tryptophan, lysine, and pyrimidine metabolites following exposure were in opposite directions in mice as compared with humans (Figures S2–S4). The change of DCAs and ACs was chain length-dependent. We found that circulating levels of long-chain DCAs and medium-to-long-chain ACs were increased in both mice and humans following air pollution exposure (Figure 3 and Table S3). The DE exposure increased mouse plasma concentrations of all the long-chain DCAs and medium-to-long-chain ACs, among which the increases in eicosanedioic acid and 6 ACs reached statistical significance at p<0.05 levels. After traveling from Los Angeles to Beijing, study participants had increased serum levels of long-chain DCAs (p<0.05 for all three DCAs), most medium- and long-chain ACs (p<0.05 for 5 ACs, FDR <5% for 3 ACs) (Figure 3). Serum concentrations of most of these DCAs and ACs returned to pre-travel levels, after participants returning to Los Angeles (Figure S5). Mice and humans did not exhibit consistent changes in short-chain ACs, short-/medium-chain DCAs, or hydroxy-/branched-DCAs after exposure to air pollution.

Figure 3. Changes of individual DCAs and ACs in animal plasma (A) and human serum (B) in response to exposure to air pollution.

Metabolite-specific fold-changes in animals (DE vs. FA) and humans (Beijing vs. LA-before) were calculated using linear regression models and mixed-effects models with random intercepts of participants, respectively. *: FDR < 5%; #: p<0.05 with exact values shown in Table S3. Number of mice: n = 5 /group. Number of human participants: n=26.

Plasma Fatty Acids and Hepatic Mitochondrial Dysfunction in Mice.

Among the same mice used in this study, we have previously reported that two-week DE exposure led to decreased hepatic mitochondrial respiration (Figure S6).⁶ Integrating transcriptomic and metabolomic data in the liver, we profiled the palmitic acid oxidation pathways and observed increased levels of AC (16:0), fumarate, malate, and oxaloacetate along with decreased citrate levels, due to impaired β-oxidation and slowdown of the tricarboxylic acid cycle (Figure S7) as previously reported.⁶ Furthermore, we compared the DE-induced changes in fatty acid metabolites in the liver and plasma (Figure 4A). In both plasma and liver, DE exposure led to greater increases in long-chain AC (16:0) and DCA (tetradecanedioic acid) as compared with medium-chain AC (6:0) and DCA (azelaic acid). The fold changes of plasma ACs and DCAs were positively correlated with those of liver fatty acids (r=0.71, p=0.051, Figure 4B). Furthermore, the fold changes of all the detected fatty acids in the plasma were also positively correlated with those in the liver (r=0.46, p=0.017, Figure 4C). These results suggest a significant contribution of hepatic lipid dysmetabolism to plasma fatty acid metabolites.

Figure 4. Changes of hepatic and plasma fatty acids (A) in mice exposed to DE, and the correlations of metabolite changes in liver and plasma for ACs and DCAs (B) and for all fatty acids (C).

Metabolite-specific fold-changes were calculated using linear regression models. Letters in panel B denotes individual metabolites listed in panel A. Number of mice: n = 5 /group.

DCAs/ACs and Biomarkers of Early Cardiovascular Effects in Humans.

We focused on three DCAs and five ACs that were increased after traveling from Los Angeles to Beijing, and examined their associations with cardiovascular biomarkers affected by subchronic air pollution exposure in both mice and humans. Both DCAs and ACs were positively associated with circulating levels of 9- and 13-HODEs and 12-HETE (Figure 5 and Table S4), all products of lipid peroxidation derived from 12/15-LOX pathways. In contrast, neither DCAs nor ACs were associated with 8-isoprotane (Figure 5 and Table S4) - a non-enzymatic lipid peroxidation product. In addition, various DCAs were positively associated with IL-8, soluble P-selectin (sCD62P), soluble intercellular adhesion molecule-1 (sICAM-1), and HDL but not with CRP, soluble CD40 ligand (sCD40L), PON1 activity, or total cholesterol (Figure S8).

Figure 5. Percentage changes of selected DCAs and ACs associated with IQR increases in logarithmic concentrations of circulating lipid peroxidation biomarkers (A-D) and urinary metabolites polycyclic aromatic hydrocarbons (E-H) among 26 healthy adults traveling between Los Angeles and Beijing.

Associations were tested by mixed-effect models with random intercepts of study participant and phase. Exact effect sizes and p-values were shown in Tables S4 and S6. *: FDR < 5%.

DCAs/ACs and Exposure Biomarkers in Humans.

We leveraged existing datasets of 29 urinary exposure biomarkers to identify key pollutants that were associated with changes in DCAs/ACs levels. Among 29 urinary exposure biomarkers, the metabolite levels of four PAHs (dibenzofuran [DBF], fluorene [FLU], phenanthrene [PHE], and pyrene [PYR]), cotinine, bisphenol A, strontium, tin, and lead were increased after traveling from Los Angeles to Beijing, while levels of chromium, bromine, and lithium were lower in Beijing (Table S5). We found that concentration changes in many DCAs and ACs were associated with urinary metabolites of PAHs, but not other exposure biomarkers (Figure S9). These urinary PAHs metabolites have been validated as exposure biomarkers of air pollution among the study participants in our previous studies.^19,20 Specifically, the changes of ACs were positively associated with urinary PAH metabolites (Figure 5 and Table S6) while the changes of DCAs were negatively associated with urinary PAH metabolites (Figure 5 and Table S6).

Discussion

This study provided a novel opportunity to study comparable metabolomic features of air pollution exposure in human and murine models. We observed remarkable similarities in air pollution induced metabolomic changes across the two models in spite of marked differences in their lipid and inflammatory status and mode of exposure (DE vs. PM_2.5). While the human study was subject to potential influences from factors that could not be controlled under real-world conditions, the mouse study was carried out under tightly controlled laboratory conditions. Herein, we focused on an early stage of cardiovascular effects when subchronic exposure to air pollution led to increased lipid peroxidation and impaired HDL functionality without evidence of dyslipidemia. Using untargeted metabolomic analysis, we identified consistent increases in long-chain DCAs and medium- to long-chain ACs in both animals and humans, likely resulting from mitochondrial dysfunction as evidenced by impaired mitochondrial respiration and β-oxidation in livers from exposed animals. These results support the involvement of impaired fatty acid metabolism and mitochondrial dysfunction in the early cardiovascular effects of air pollution.

Increased levels of DCAs and ACs in blood and urine have been widely documented among individuals with deficits in fatty acid β-oxidation, the major degradation pathway for DCAs and ACs.²¹ Additionally, impaired β-oxidation may further lead to increased ω-oxidation of mono-fatty acid that produces DCAs.²² Remarkably, we observed consistent increases in long-chain DCAs and medium- to long-chain ACs, rather than short-chain ACs/DCAs or very long-chain ACs. The β-oxidation of medium- and long-chain fatty acids is mainly handled by mitochondria while that of very long-chain fatty acids occurred mostly at the peroxisome.²² Particularly, carnitine palmitoyltransferases, which selectively synthesize medium-to-long chain ACs, mainly exist in mitochondria but not in peroxisome.²³ Therefore, the ACs signature indicated that the impaired β-oxidation most likely occurred at the mitochondria rather than the peroxisome. This may redirect fatty acid oxidation toward ω-oxidation, leading to the generation of DCAs. This was confirmed in our mouse study where DE exposure decreased the mitochondrial respiratory capacity and β-oxidation of livers from ApoE^−/− mice,⁶ which was recapitulated by treatment of HEPG2 cells or mitochondria isolated from C57BL/6 mouse livers with diesel exhaust particles.^8,24

Many environmental factors, besides air pollution, are likely to be different between Los Angeles and Beijing. Thus, we have conducted a comprehensive exposure assessment using 29 urinary biomarkers to identify key exposures that were associated with metabolomic responses. Among these exposure biomarkers, the metabolites of DBF, FLU, PHE, and PYR have been previously validated as air pollution exposure biomarkers among the same participants though that the levels of PHE and PYR metabolites were also influenced by secondhand smoke.¹³ During the travel, the changes of medium-to-long-chain ACs were associated with these PAH metabolites and therefore were likely attributable to exposure to combustion-originated air pollutants. This is causally supported by our animal study where two-week exposure to DE, rich in PAHs and other combustion products, induced accumulation of long-chain DCAs and medium-to-long-chain ACs in the blood and liver, accompanied by decreased hepatic mitochondrial respiration and β-oxidation.⁶ It was aligned with previous in-vitro experiments that the effects of DE and/or ultrafine particles on oxidative stress and mitochondrial dysfunction can be attributable to PAHs and quinone,^25,26 as well as previous epidemiological evidence associating air pollution^27–29 and/or PAHs³⁰ with decreased copy numbers of mitochondrial DNA.

It has been increasingly recognized in animal and cell models that mitochondrial dysfunction is implicated in the adverse health effects of air pollution. However, comparable evidence is scarce in humans largely due to the lack of biomarkers of mitochondrial dysfunction. Several epidemiologic studies have associated PM_2.5 and its components with decreased copy number^27–29 and methylation³¹ of mitochondrial DNA in the general population. The changes in mitochondrial DNA numbers and methylation have also been associated with altered heart rate variability and blood pressure.^32,33 Herein, we added novel metabolic evidence to support the use of long-chain DCAs and medium-to-long-chain ACs to detect and track air pollution-induced mitochondrial dysfunction. In both mice and healthy adults, the change in DCAs and ACs after air pollution exposure occurred without significant changes in circulating levels of total or HDL cholesterol. Plasma total triglycerides were not changed in mice following DE exposure but decreased in humans following from Los Angeles to Beijing.^5,34 Taken together, our findings support that the accumulation of long-chain DCAs and medium-to-long-chain ACs is among the early metabolic effects induced by subchronic air pollution exposure.

Our results provide mechanistic insights into how air pollution could induce mitochondrial dysfunction. In mice exposed to DE, the increase in DCAs and ACs as well as mitochondrial dysfunction were accompanied by increased bronchoalveolar lavage fluid and blood levels of a panel of lipid oxidative biomarkers including those from 12- and 15-LOX pathways (i.e., 9-HODE, 13-HODE, and 12-HETE) or non-enzymatic pathways (i.e., 8-isoprostane). While it has been reported that mitochondria-induced lipid peroxidation is typically through uncontrolled generation of reactive oxygen species which lead to the formation of 8-isoprotane,³⁵ it is not known whether mitochondrial dysfunction caused lipid peroxidation or viceversa.³⁶ In the human study, traveling from Los Angeles to Beijing led to greater increases in 12- and 15-LOX metabolites (by 47.3% to 998%) than 8-isoprostane (by 20.8%), which was potentially due to antioxidant responses that could have compensated non-enzymatic oxidation more effectively. Moreover, we did not observe associations of DCAs and ACs with 8-isoprostane levels. Instead, DCAs and ACs were significantly associated with 9-HODE, 13-HODE, and 12-HETE levels. Consistently, a mouse study showed that 10-day exposure to concentrated PM_2.5 increase circulating levels of ACs, DCAs, 9- and 13-HODE, which was abated by extracellular superoxide dismutase overexpression in the lung.³⁷ Recently, we also reported that 16-week exposure to DE led to 12-LOX upregulation in the livers of ApoE^−/− mice. This was replicated in DE-exposed HepG2 cells which also exhibited mitochondrial dysfunction.⁸ Administration of acetate prevented the DE-induced increase in 12-LOX expression and mitochondrial dysfunction in HepG2 cells.²⁴ Another study also found that blocking Alox12 and AMPK activation reversed vascular mitochondrial dysfunction developed by ApoE and LDLR double knockout mice in the C57BL/6 background.³⁸ Taken together, it is likely that 12/15-LOX-mediated lipid peroxidation contributed to mitochondrial dysfunction.

Our animal findings support the likely role of hepatic mitochondrial dysfunction in DE-induced dyslipidemia. However, mitochondrial dysfunction may also occur in extrahepatic tissues such as platelets, endothelium, or cardiomyocytes, which has previously been shown to contribute to the development of cardiovascular disease.^2,39 For example, a randomized trial of 38 healthy adults found that one-week use of air purifier decreases serum levels of epidermal growth factor, vascular endothelial growth factor-A and D-dimer which were partially mediated by increased oxidative stress.⁴⁰ This validated in vitro data where PM_2.5 treatment led to increased intracellular oxidative stress and mitochondrial dysfunction in human umbilical vein endothelial cells.⁴⁰ Other studies have associated ambient PM_2.5 levels with platelet mitochondrial dysfunction in healthy adults.⁴¹ Aligned with these findings, we found that plasma DCAs and ACs were positively associated with biomarkers of platelet activation (sCD62P) and endothelial dysfunction (sICAM-1) in the human study (Figure S8). Importantly, other studies have shown that air pollution induces mitochondrial dysfunction in the heart as well, impacting myocardial energetics,⁴² reducing cardiac contractility,⁴³, and exacerbating ischemia reperfusion in a mouse model of myocardial infarction.⁴⁴

Despite the consistency in pro-oxidative and ACs/DCAs effects of air pollution in both mice and humans, we observed pro-inflammatory effects only in humans (Table S2). This can explain the difference in the effects on tryptophan metabolism between mice and humans. Inflammation has been shown to activate kynurenic pathways of tryptophan metabolism through inducing indoleamine 2,3-dioxygenase.⁴⁵ In humans, we found that traveling from Los Angeles to Beijing led to significant increases in the kynurenine/tryptophan ratio. In mice, DE exposure led to increased metabolite ratios indicative of the activation of indole rather than kynurenine pathways (Figure S10). The lack of inflammatory responses in mice suggested that the DE-induced oxidative stress and mitochondrial dysfunction may be independent of inflammatory pathways.

This study has several strengths, including the harmonized protocols for metabolomic and biomarker analyses and data analysis across mice and human studies. Although metabolomic profiling for mice and human specimens was based on different platforms, we focused on chemicals with known identities to ensure that the results from different platforms were comparable and could cross-validate each other. In addition, there was a drastic contrast in subchronic exposure to PM_2.5 (4.7-fold) and PAHs (up to 9.0-fold) between Los Angeles and Beijing, enhancing the statistical power to detect an effect. Furthermore, we conducted biomarker-based exposure assessment in the human study, allowing us to account for participants’ activity patterns in relation to exposure at different micro-environments.

This study has several limitations. First, there were differences in the nature of air pollution that mice and humans were exposed to. Mice were exposed to particulate and gaseous pollutants from DE without ozone, while humans were exposed to real-world PM_2.5 and multiple gas pollutants, including ozone. Thus, our results mainly reflect common effects associated with combustion-originated air pollutants and did not rule out the effects of other pollutants on other pathways. Likewise, ApoE^−/− mice are genetically predisposed to hyperlipidemia and atherosclerosis development. DE induced extensive changes in lipid metabolism in ApoE^−/− mice that were not observed in healthy young adults. It is unclear whether similar changes could be developed by human subjects with morbid conditions that may increase susceptibility such as pre-existing diabetes, cardiovascular diseases or hyperlipidemia. Second, because we used different platforms to profile blood metabolomics in mouse and human specimens and conservatively select metabolites with confirmed identities, the number of overlapped metabolites (n=154) in both mice and humans was relatively small, which restricted the statistical power to identify common metabolic pathways. Of note, we observed consistent increases in deoxycholate in both mice and humans, suggesting that bile acid metabolism may also be implicated in air pollution’s metabolic effects. Additionally, we used different biospecimens for metabolomic profiling (mouse plasma vs. human serum), which may hinder the identification of common metabolites and pathways that are affected by clotting processes.⁴⁶ Third, the mouse experiments include only male mice and the results may not be generalizable to female mice. Nevertheless, we observed no sex differences in travel-induced changes in inflammatory and lipid peroxidation biomarkers in the human study.⁹ Last, although the mouse study results are consistent with the human study, this does not demonstrate causality in humans. The role of exposures beyond air pollution and/or mechanisms beyond activation of LOX pathways (e.g., microparticles and antiangiogenic changes)⁴⁷ cannot be ruled out.

In conclusion, we found consistent increases in circulating long-chain DCAs, and medium- to long-chain ACs, a signature of mitochondrial dysfunction, following air pollution exposure, through integrated metabolomic analyses of ApoE^−/− mice and healthy travelers to Beijing. Among travelers, the changes of DCAs and ACs were associated with 12- and 15-LOX metabolites. Consistently, ApoE^−/− mice exposed to DE also showed increased 12- and 15-LOX metabolites. Our results suggest that the axis of LOX-mediated lipid peroxidation, mitochondrial dysfunction, and impaired fatty acid oxidation may represent a human-relevant mechanism linking oxidative stress to lipid dysmetabolism.

Supplementary Material

• Figures S1–S10

• Tables S1–S6

• Major Resources Table

What are the Clinical Implications?

Air pollution is known to induce a trajectory of cardiometabolic effects, progressing from lipid peroxidation after acute exposure to dyslipidemia after prolonged exposure. This mouse-to-human translational study aims to identify mechanisms that link oxidative stress with dyslipidemia. Through integrated metabolomic analyses of ApoE^−/− mice exposed to diesel exhaust for two weeks, and 26 Los Angeles residents exposed to real-world air pollution in Beijing for 6–8 weeks, we found that air pollution led to common metabolic derangements, including consistent increases in long-chain dicarboxylic acids (DCAs), and medium- to long-chain acyl-carnitines (ACs), a signature of mitochondrial dysfunction which was confirmed in mice using a Seahorse assay on frozen liver sections. In the human study, the changes of DCAs and ACs were associated with increased lipid peroxidation products from 12- and 15-lipoxygenase pathways and exposure biomarkers of polycyclic aromatic hydrocarbons. Consistently, ApoE^−/− mice exposed to diesel exhaust also showed increased circulating levels of 12- and 15-lipoxygenase metabolites. Our results suggest that the axis of lipoxygenase-mediated lipid peroxidation, mitochondrial dysfunction, and impaired fatty acid oxidation represents a human-relevant mechanism linking oxidative stress to dyslipidemia.

Nonstandard Abbreviations and Acronyms

ACs

Acyl-carnitines

ApoE

Apolipoprotein E

CRP

C-reactive protein

DBF

Dibenzofuran

DCAs

Dicarboxylic acids

DE

Diesel exhaust

FDR

False discovery rate

FLU

Fluorene

HETE

Hydroxyeicosatetraenoic acids

HDL

High-density lipoprotein

HODE

Hydroxyoctadecadienoic acids

IL-8

Interleukin-8

FA

Filtered air

LC

Liquid chromatography

LOX

Lipoxygenase

MS

Mass spectrometry

PAHs

Polycyclic aromatic hydrocarbons

PHE

Phenanthrene

PYR

Pyrene

PON1

Paraoxonase 1

PM_2.5

Particulate matter with an aerodynamic diameter <2.5 μm

sCD40L

Soluble CD40 ligand

sCD62P

Soluble P-selectin

sICAM-1

Soluble intercellular adhesion molecule-1