Metabolic effects of air pollution exposure and reversibility

Air pollution is the leading environmental cause of premature reversible death and disability in the world today (1). A large body of evidence implicates the component of air pollution containing particulate matter smaller than 2.5 μm (PM_2.5) in size in the development of cardiovascular disease risk factors such as hypertension, insulin resistance (IR), and type 2 diabetes mellitus (T2DM) (2, 3). Although the mechanisms by which inhalation of PM_2.5 induces IR and T2DM remain unclear, a constellation of responses including inflammation and redox stress have been implicated. To our knowledge, no studies have examined the metabolic, transcriptomic, and epigenetic effects of air pollution in comparison with other stressors such as a high-fat diet (HFD). Importantly, the reversibility of PM_2.5-induced transcriptional and epigenetic changes has not, to our knowledge, been examined. In this study, we evaluated the metabolic phenotypic, transcriptional, and epigenomic changes in response to exposure to concentrated ambient PM_2.5 (referred to hereafter as PM_2.5) and the response of these parameters to cessation of exposure.

Air pollution exposure. Three-week-old C57BL/6J mice on a chow diet were started on exposure to PM_2.5 for 14 weeks. Figure 1 provides a summary of the systemic effects of ambient air pollution on IR in PM_2.5-exposed male mice. PM_2.5 was delivered inhalationally (~10× ambient level/~60–120 μg/m³ for 6 hours/day, 5 days/week) using a versatile aerosol concentrator and enrichment system (Supplemental Figure 1 indicates the PM_2.5 concentrations during the exposure period [the mean PM_2.5 concentration was 80 μg/m³, ~10× ambient concentration], the meteorological conditions, and the elemental characterization of the air pollution exposure; supplemental material available online with this article; https://doi.org/10.1172/JCI137315DS1). Three-week-old C57BL/6J mice were fed a HFD for 14 weeks and served as a positive control for IR and associated transcriptomic and epigenomic changes.

Figure 1

Systemic effects of ambient air pollution (PM_2.5) on IR. (A) Experimental plan including a cessation period of 8 weeks following 14 weeks of exposure to FA or PM_2.5. (B) GTT (intraperitoneal administration of glucose, 2 g/kg body weight) and AUC (n = 12/group). (C) ITT (intraperitoneal administration of insulin, 0.75 U/kg body weight) and AUC (n = 12/group). (D) VO₂ consumption, (E) VCO₂ production, and (F) EE. Line graphs indicate averages of the day and night cycles over a 48-hour period, and bar graphs indicate the total value at the indicated time points (n = 6). (G) Representative PET image (whole-body ¹⁸FDG uptake) of FA- and PM_2.5-exposed mice (n = 4/group) and quantitative measurement of ¹⁸FDG uptake from BAT (value of region of interest [ROI]). Data are expressed as the mean ± SEM. *P < 0.05 relative to FA mice, as determined by Student’s t test or, as appropriate, 2-way ANOVA.

IR and metabolic dysfunction with air pollution exposure. PM_2.5 exposure resulted in alterations in glucose clearance compared with filtered air (FA) in male mice. Insulin responsiveness (Figure 1, B and C) was also altered by PM_2.5 exposure. We observed no difference in mean body weights between the groups (Supplemental Figure 2A). Metabolic cage experiments demonstrated a reduction in VO₂, VCO₂, and energy expenditure (EE) in PM_2.5-exposed mice during both light (Zeitgeber time 0–12 [ZT0–ZT12]) and dark (ZT12–ZT24) phases (Figure 1, D–F). PM_2.5 exposure reduced the respiratory quotient only at nighttime (Supplemental Figure 2D). PM_2.5-exposed mice had reduced glucose uptake in brown adipose tissue (BAT) as measured by ¹⁸FDG-PET (Figure 1G). The body mass of both male and female mice increased in response to a HFD, but not to PM_2.5 exposure (Supplemental Figure 2A). PM_2.5 induced abnormalities in glucose clearance and insulin responses that were comparable to those seen with a HFD, but only in males (Supplemental Figure 2, B and C). Given that phenotypic changes were seen only in male mice, we restricted further analysis to males. We acknowledge that it is entirely possible that there could still be conserved responses in males and females that diverge from the sexual dimorphic responses of glucose tolerance and insulin responsiveness. Hepatic cholesterol, but not plasma cholesterol, increased in the PM_2.5-exposed group, whereas lipid deposition and fibrosis were increased in both PM_2.5- and HFD-exposed groups (Supplemental Figure 3, A and B). Hepatic inflammation in PM_2.5-exposed and HFD-fed mice was characterized by an elevation of proinflammatory M1 genes (TNF-α, IL-6, and TLR4) and downregulation of antiinflammatory M2 genes (IL-10) (Supplemental Figure 3, C and D). We found that liver adiponectin levels were markedly downregulated in response to PM_2.5 exposure. In many of these genes, the degree of regulation and directionality in hepatic tissue was comparable to that seen with a HFD (Supplemental Figure 3, C and D). Hepatic glycogen levels were decreased in PM_2.5-exposed mice, albeit to a lesser degree, compared with levels in HFD-fed mice (Supplemental Figure 3E).

In summary, the phenotype encountered with chronic air pollution exposure was characterized by alterations in oxygen consumption and EE, hepatic inflammation, elevated triglycerides with evidence of hepatic steatosis, and glycogen depletion. Interestingly, mice exposed to air pollution did not develop adiposity, similar to the phenotype of lean IR that is common in many parts of Asia. Importantly, we observed a strong sexually dimorphic response, as only male mice developed IR.

Differential transcriptome and functional annotation of PM_2.5 versus HFD exposure. We performed RNA-Seq to quantify transcriptomes using RNA from liver, skeletal muscle, BAT, white adipose tissue (WAT), and hypothalamus from FA- and PM_2.5-exposed mice, which resulted in the identification of multiple differentially expressed genes (DEGs) (Figure 2A). Supplemental Figure 4, A–F, shows the results of our principal component analysis (PCA) according to the tissue of origin and exposure status. We identified a total of 589 DEGs for PM_2.5 exposure versus FA exposure, with 11 genes being differentially expressed in 2 tissue types (Venn diagram, Figure 2A, and hypergeometric test, Supplemental Table 2). The HFD appeared to induce large transcriptional changes across all tissue types (Figure 2B). In contrast, PM_2.5 changes were less pronounced, with the largest number of DEGs in BAT, followed by WAT and liver (Figure 2B). The most enriched gene ontology (GO) terms for the PM_2.5 transcriptome were related to cancer progression, cardiometabolic function, and circadian rhythm, and the corresponding pathways associated with these were inflammation, redox stress, metal ion transport, and glucose metabolism (Figure 2C). In contrast, a HFD induced changes in fatty acid biosynthesis, inflammation, gluconeogenesis, and lipid regulatory pathways. More detailed DEGs, functional annotations, and potential upstream and downstream regulators are summarized in Supplemental Figures 5–9. Supplemental Figure 4G depicts the number and directionality of DEGs in response to PM_2.5 exposure and a HFD when compared with FA exposure. In liver, WAT, hypothalamus, and muscle, the number of DEGs that were concordant was higher than the number of genes displaying a discordant response. We detected comparable numbers of discordant and concordant DEGs in BAT (Supplemental Figure 4G).

Figure 2

Comprehensive transcriptome analysis in insulin-responsive tissues. (A) Venn diagram depicting DEGs in various tissues from FA- and PM_2.5-exposed mice (FDR <0.05, logCPM >1). (B) Tissue-specific DEGs and overlapped genes in FA, PM_2.5, and HFD-exposed mice (n = 45). (C) Summary of enriched GO terms and functional pathways in FA, PM_2.5, and HFD transcriptomes. (D) Volcano plots compare a chow diet versus a HFD and FA versus PM_2.5 exposure, indicating selected DEGs of IR and stress response elements in various tissues from HFD-fed mice.

DEGs in insulin-responsive tissue and functional annotations. PM_2.5 exposure induced 42 DEGs in the liver, 35 of which overlapped with HFD-induced DEGs (Supplemental Figure 4G). GO analysis revealed that circadian rhythm genes in the liver ranked at the top (GO:0007623 circadian rhythms). These genes included transcriptional regulators (e.g., Dbp, Bhlhe41, Cry1, Per3, and Arntl/Bmal1). Although the core clock components cryptochrome 1 (Cry1) and Bmal1 (Arntl1) were downregulated with PM_2.5 exposure, the transcriptional repressor basic helix-loop-helix family member e41 (Bhlhe41) was upregulated (4). Ppp1r3g, a markedly downregulated gene with PM_2.5 exposure, belongs to a family of glycogen-targeting regulatory subunits (G subunits) that coordinate glycogen synthesis by targeting the catalytic subunit of PP1 to the glycogen particles and that activate glycogen synthase through PP1-mediated dephosphorylation (5). These findings were consistent with glycogen depletion in response to PM_2.5 exposure (Supplemental Figure 3D). Several G protein–coupled receptor proteins (C5ar1, Celsr1, Adgrv1) were downregulated in the liver, including the nuclear receptor protein Nr4a1 (Nur77) (Figure 2D). Previous studies have identified Nr4a1 as a transcriptional regulator of glucose utilization in the liver and macrophage polarization (6, 7). Rgs16, a regulator of G protein signaling shown to restrict the proinflammatory response of monocytes, was the most upregulated gene in the liver (8). In addition, Supplemental Figure 7E depicts the TGF-β signaling pathway and the corresponding PM_2.5- and HFD-specific DEGs. We also observed the downregulation of mTOR signaling DEGs (Supplemental Figure 8A, FDR <0.1). The enriched functional pathways and diseases for skeletal muscle and hypothalamus are summarized in Supplemental Figure 9. BAT, WAT, and liver tissues had the largest number of overlapping DEGs between PM_2.5- and HFD-exposed mice (Supplemental Figure 4G). In contrast, skeletal muscle and hypothalamus showed a smaller number of common DEGs in response to a HFD and PM_2.5. Enriched pathway terms specific for BAT DEGs upregulated specifically with PM_2.5 exposure included T2DM signaling and neuronal NOS (nNOS) signaling (Supplemental Figure 5B). We also observed downregulation of adipogenesis DEGs (Ccng2, Acly, Slc25a1, Angptl4) and upregulation of inflammatory response DEGs in gene set enrichment analysis of BAT (Supplemental Figure 5C). Protein phosphatase 1 regulatory subunit 3G (Ppp1r3b) in BAT, an analog of Ppp1r3g, was markedly downregulated with both PM_2.5 and HFD exposure (Figure 2D), whereas myostatin (Mstn), known to be linked to diabetes (9), was markedly upregulated in BAT (Figure 2D). Supplemental Figure 6 depicts functional links between the upregulated genes in WAT and their corresponding pathways.

Our results support epidemiologic and empirical observations linking air pollution with pathways that promote susceptibility to other noncommunicable diseases (1–3). Circadian rhythm alterations constitute a common denominator for the development of cancer as well as metabolic and cardiovascular diseases (10–12). Many circadian factors are classic epigenetic regulators, and, conversely, many metabolites affect epigenetic regulators and the epigenomic landscape (13, 14).

Changes in chromatin accessibility in response to PM_2.5 exposure and a HFD. Given that PM_2.5 exposure resulted in a distinct transcriptomic response (Figure 3A) similar to that elicited by a HFD, we hypothesized that DEGs may be regulated by epigenetic reprogramming driven by differential transcription factor binding (Supplemental Figure 10). Using ATAC-Seq, we evaluated genome-wide chromatin accessibility in the livers of PM_2.5-exposed mice and compared the results with those for HFD-fed mice (Figure 3B). We analyzed the ATAC-Seq data sets on the basis of fragment size distribution and identified uniquely reproducible peaks in the PM_2.5 and HFD groups (Supplemental Figure 11, A and B). We further identified differentially accessible regions (DARs) and found that 218 were common gain-of-accessibility regions and 124 were common loss-of-accessibility regions (Supplemental Figure 11C). We also explored the degree of concordance and discordance between transcriptomic and epigenomic alterations in response to a HFD and PM_2.5 exposure (Figure 3, B and C). The DARs generally corresponded with upregulated gene expression and, conversely, less accessible regions with downregulated gene expression. We summarized the CG/AT ratio on the promoter sites (Figure 3B) as a reflection of chromatin accessibility, corresponding to CpG-rich regions. Of the reproducible peaks that overlapped with both FA and PM_2.5, 1937 peaks were differentially accessible with PM_2.5 exposure (Supplemental Figure 11C). Among those peaks, we searched for PM_2.5-specific differential “distal” peaks, denoting potential distal regulatory sites (1–5 kb, 10 kb, 25 kb, 100 kb, 500 kb and >500 kb) (Figure 3C), and performed differential motif binding analysis (15) using publicly available ChIP-Seq data sets (16). We found the chromatin remodeler SMARCA5 (SWI/SNF-related matrix-associated actin-dependent regulator of chromatin subfamily A member 5) to be a differential binding protein in PM_2.5-exposed mice when compared with FA-exposed mice (Figure 3, D and E).

Figure 3

Liver transcriptomic and epigenomic changes. (A) Comparison of the expression levels of DEGs (enriched inflammatory and circadian genes) between PM_2.5- and HFD-exposed mice. (B) Integrative analysis using RNA-Seq and ATAC-Seq data sets and the corresponding CpG content in promoters. FC, fold change. (C) Open chromatin distribution across the 3 groups (FA, PM_2.5, and HFD). (D) Differential transcription factor (TF) binding analysis using the available ChIP-Seq database. MD, motif displacement. (E) ATAC-Seq enrichment on the SMARCA5 binding motifs. **P < 0.01, by Kolmogorov-Smirnov test.

Epigenetic reprogramming in response to environmental exposure may represent a critical buffer against an adverse health response through the regulation of gene expression and chromosome integrity (17). Circadian rhythm genes in the liver ranked at the top of the GO analysis in our study, with changes noted in core clock components and transcriptional regulators of circadian rhythm. Although the core clock components cryptochrome 1 (Cry1) and Bmal1 (Arntl1) were downregulated in response to PM_2.5, we found that the transcriptional repressor of the clock-Arntl/Bmal1 heterodimer basic helix-loop-helix family member e41 (Bhlhe41) was upregulated.

Cessation of PM_2.5 exposure leads to reversal of transcriptomic and epigenomic changes. After 8 weeks of cessation of PM_2.5 exposure, we noted an improvement in glucose tolerance and insulin sensitivity (Figure 4A). Given the degree of phenotypic variation in reversibility (Supplemental Figure 12, A and B), we were interested in the corresponding transcriptomic and epigenomic changes. Figure 4B shows the reversal of a number of highly regulated genes in liver and WAT, with a marked reversibility within 8 weeks of exposure cessation (Figure 4C and Supplemental Figure 13). We performed ATAC-Seq on liver to identify possible chromatin accessibility changes indicating a reversal mechanism after cessation of PM_2.5 exposure. Supplemental Figure 14A depicts DARs in response to PM_2.5 cessation in liver. In total, we discovered 3467 DARs that changed in the liver following PM_2.5 cessation. Analysis of the regulatory elements of the same DARs showed that cis-regulatory regions that change with cessation are highly enriched in pathways related to insulin action, resistance, gluconeogenesis, and metabolism (Supplemental Figure 14B). Supplemental Figure 14C shows protein-coding genes (n = 21,955) in the form of a heatmap of all proximal peaks (±1 kb) from the transcription start site (TSS), depicting changes in chromatin accessibility following reversal. Interestingly, SMARCA5 expression, which was shown to be regulated in response to PM_2.5 exposure, demonstrated reversibility with exposure cessation (Figure 4D). Two of the most highly regulated genes in the liver were Nr4a1 (Nur77) and Rgs16, which happen to be particularly relevant to the genesis of IR and inflammation (6, 18). The impact of cessation on the expression and regulatory elements of the most upregulated (Rgs16) and downregulated (Nr4a1) genes in the liver is depicted in Figure 4, E and F. Epigenomic changes with exposure cessation were concordant with the change in mRNA levels for several circadian genes, providing evidence for the reprogramming of transcription in their promoters such as the Rgs16 (Supplemental Figure 13B) and Bmal1 (Supplemental Figure 15) promoters. The reversible changes in circadian genes seen with cessation of exposure suggest a causality of air pollution exposure in circadian dysregulation.

Figure 4

Impact of exposure cessation versus continued PM_2.5 exposure in liver. (A) Comparison of GTTs and ITTs, as represented by the AUC, in FA-, PM_2.5-, and HFD-exposed mice at the indicated time points. Dashed line indicates the cessation of exposure at 14 weeks. (B) Heatmap of liver and WAT DEGs by week 8 after PM_2.5 exposure cessation (14–22 weeks). (C) Volcano plot of the WAT transcriptome at 14 weeks and scatter plot of the DEG fold change distribution showing a reversal of highly upregulated genes in PM_2.5-exposed mice. (D) Differential transcription factor binding analysis using ATAC-Seq data sets and the reversal of SMARCA5 binding profiles (PM_2.5 vs. cessation). (E) Quantitative assessment summarizing normalized transcript values for 2 select genes (Rgs16 and Nr4a1) across mouse samples. FPKM, fragments per kilobase per million mapped reads. (F) Summary of reversible DARs and genomic regions.

PM_2.5 exposure promoted substantial chromatin remodeling, especially at promoter and enhancer sites that were pliable, with cessation of exposure resulting in a reversal of changes in chromatin accessibility and of expression of transcripts, notably those involved in insulin action, circadian rhythm, and inflammation. An important additional finding was that expression of SMARCA5 (SWI/SNF complex) was regulated in response to PM_2.5 exposure and was reversible with exposure cessation. SWI/SNF complexes are a family of polymorphic, ATP-dependent chromatin remodeling complexes that are recruited to cis-regulatory elements such as promoters and enhancers, where they contribute to chromatin accessibility (19). Chromatin remodeling may be particularly important in the context of environmental exposures in which epigenomic changes may be required to buffer and regulate gene expression (17). A growing body of evidence shows that short- and short-term exposure to ambient PM_2.5 is associated with altered DNA methylation in specific genes related to inflammation, vascular endothelial dysfunction, and cytokine production and that these effects are implicated in the perturbation of circulating cytokines and fasting blood glucose levels (20–22).

Our study has limitations that must be acknowledged. We did not identify the precise epigenetic regulators that lead to metabolic dysfunction through genome-wide methylation analysis. In addition, histone modifications (e.g., H3K36me3, activation marks; H3K27me3, repression marks; H3K27ac, enhancer marks) and the 3D structure of the chromatin may also be involved in epigenetic reprogramming upon chronic PM_2.5 exposure. Our results will require validation in well-designed human studies and, if confirmed, may have important implications for interventions to reduce air pollution levels and exposure.

Further information can be found in the Supplemental Methods.

Data availability. Sequencing data from this study have been deposited in the NCBI’s Gene Expression Omnibus (GEO) database (GEO GSE145840).

Study approval. All animal procedures and experiments were approved by the IACUC of Case Western Reserve University.