Sphingolipids as indicators of exposure to long-term air pollution: a targeted metabolomics approach Authors: Jeeyoung Kim, Eun Ju Lee, Ha Eun Song, Hyo Yeong Lee, Chin Kook Rhee, Sun-Young Kim, Hyun Ju Yoo, Woo Jin Kim Published: 2025 ABSTRACT Metabolomic changes related to air pollution have been previously reported, including sphingolipid metabolism; however, evidence from long-term exposures assessed with long-term exposure using targeted metabolomics remains limited. We profiled plasma sphingolipids and amino acids in 188 individuals from the chronic obstructive pulmonary disease (COPD) in Dusty Areas cohort. Annual average concentrations of particulate matter ≤ 10 μm (PM10, µg/m³) and nitrogen dioxide (NO2, ppb) were estimated for each participant’s residence using a national-scale universal kriging model integrating regulatory monitoring data and geographic predictors across South Korea. Multiple linear regression models, adjusting for demographic, clinical, and environmental covariates, evaluated associations between exposures and metabolites. Long-term PM10 exposure was nominally associated with higher sphingomyelins [SM(d18:1/18:0), SM(d18:1/18:1), and SM(d18:1/24:1)], whereas NO2 exposure was nominally associated with Cer(d18:1/24:0) and glutamate. PM10 was not significantly associated with amino acids. No main associations remained significant after false discovery rate (FDR) correction. In subgroup analyses, particularly among patients with COPD, SM(d18:1/16:0), SM(d18:1/18:1), SM(d18:1/24:1), and SM(d18:1/18:0) met the FDR threshold, highlighting robust and subgroup-specific associations. Although results provide novel insights into pollutant-specific metabolic pathways, confirmation in larger cohorts is needed.Supplementary InformationThe online version contains supplementary material available at 10.1038/s41598-025-33354-3. Introduction In industrialized settings, air pollution tends to stem from traffic, industrial combustion, and manufacturing processes, emitting particulate matter (PM1.0–PM10) and hazardous air pollutants such as nitrogen dioxide (NO2), sulfur dioxide (SO2), and ozone (O3)1–3. Chronic exposure to these pollutants disrupts various physiological processes, thereby contributing to the emergence of adverse health outcomes4–10. However, the mechanisms underlying these associations and their implications for human health remain poorly understood, necessitating in-depth research. Metabolomics has proven effective in the characterization of low-molecular-weight metabolites11. Identifying metabolomic changes can provide valuable insights into biological mechanisms and corresponding cellular responses perturbed by environmental exposure12. Epidemiological studies have linked short- and long-term air pollution exposure to the dysregulation of lipid metabolism (e.g., glycerophospholipids, sphingolipids, amino acids homeostasis, and energy pathways)13–15. However, the reliance un non-targeted metabolomics has generated vast and poorly annotated datasets, with more than 80% of significant features remaining unvalidated16. This outcome undermines the reliability of pathway analyses and highlights the need for hypothesis-driven targeted investigations to identify actionable metabolite signatures. Sphingolipid metabolism dysregulation—particularly ceramide and sphingomyelin imbalances—has emerged as a critical mediator of pollution-induced inflammatory and oxidative stress. Experimental studies have demonstrated that short- to medium-term air pollution exposure (days to months) elevate sphingolipid levels in human lung carcinoma cells and in mice (especially in the blood, aorta, and lungs)17–20. Furthermore, sphingolipid peroxidation products may mediate PM2.5-induced metabolic dysfunction across multiple diseases21, pointing to their potential role as mechanistic biomarkers. Clinically, sphingolipid perturbations are associated with immune cell activity in biofluids and are implicated in chronic lung diseases, such as chronic obstructive pulmonary disease (COPD). Amino acid metabolites are essential for protein synthesis, energy production, and signaling functions. Previous studies have reported associations between air pollution and altered amino acid concentrations, particularly alanine, threonine, and glutamate, although findings have been inconsistent22–24. Such disruptions—especially in pathways like the urea cycle—are linked to oxidative stress and inflammation, which may contribute to cardiopulmonary dysfunction25. These responses can vary by age and sex, highlighting the importance of considering population-specific vulnerability. Our previous study confirmed sphingomyelin and ceramide, alongside dysregulated amino acids, as intermediates linking metabolic dysfunction to the decline in lung function in patients with COPD. However, despite these mechanistic insights, population-based evidence regarding the impact of long-term air pollution exposure on sphingolipid homeostasis and amino acid metabolism in humans remains scarce. Here, we hypothesized that chronic air pollution exposure induces dysregulation of plasma sphingolipid and amino acid metabolism, modulated by individual risk factors (e.g., age and smoking status)9,26,27. Using metabolomics, we quantified sphingolipids species in a COPD in Dusty Areas (CODA) cohort study, adjusting for confounders including age, gender, body mass index (BMI), smoking, and disease status. We aimed to: (1) identify pollution-specific metabolic signatures and (2) elucidate how individual susceptibility influences these associations, thereby providing insights into the metabolic changes driven by air pollution. Materials and methods Study design and participants We used data and biological specimens from baseline examinations of the CODA study from 2012 to 2016; the detailed design was previously described by Hong et al.28. Briefly, participants were classified into two groups, COPD (n = 113) and non-COPD (n = 72), according to the standards of the Global Initiative for Chronic Obstructive Lung Disease (GOLD) criteria, which define COPD as a post-bronchodilator forced expiratory volume to forced vital capacity (FEV1/FVC) ratio < 0.70. A total of 185 participants (125 men and 60 women) aged 50–80 years, were enrolled in the CODA study. The aim of the CODA study was to investigate the clinical outcomes of Koreans with COPD living near cement plants in six regions, Gangneung (GN), Danyang (DY), Donghae (DH), Samcheok (SC), Yeongwol (YW), and Jecheon (JC). Participants with known or suspected cancer, or hospitalized within the preceding 3 months, were excluded. All participants provided written informed consent for physical examination, blood and urine sampling, spirometry, and computed tomography (CT. Demographic, medical history, and environmental exposure data were collected using standardized questionnaires. This study was approved by the Kangwon National University Hospital IRB (KNUH-2012-06-007–069, clinical trial number: not applicable) and was conducted in accordance with the Declaration of Helsinki. Estimation of air pollutant exposure concentrations We used a national-scale air pollution prediction model to estimate annual average exposure concentrations of PM10 and NO2 for each participant29. Individual residential addresses, obtained from the baseline survey, were geocoded and converted into geographic coordinates using GeoCoder-Xr software (Geoservices, Seoul, South Korea). These coordinates were then linked to the prediction model to assign exposure concentrations. The prediction model was developed using a universal kriging framework, which integrates regulatory monitoring data from 2001 to 2015 with geographic predictors to estimate annual mean concentrations at unmonitored locations across South Korea. More than 300 geographic variables were initially considered, including traffic density, demographics, land use, physical geography (e.g., elevation, slope), transportation facilities, emission sources, vegetation, and altitude30. Partial least squares regression was applied to identify the most relevant predictors. Model performance was evaluated using 5-fold cross-validation, yielding cross-validated R2 values of 0.45 for PM10 and 0.82 for NO2, with corresponding root mean square errors of 6.2 µg/m and 3.5 ppb, respectively. Although predictive accuracy for PM10 was moderate, this performance is consistent with values reported for other national-scale prediction models in Europe and the United States31–33. The relatively lower R2 for PM10 compared with NO2 reflects its greater spatial heterogeneity, which makes prediction inherently more challenging. By contrast, pollutants such as NO2 and PM2.5 typically achieve higher predictive accuracy because of their more homogeneous spatial distribution. Annual average concentrations of PM10 and NO2 were assigned to each participant by overlaying their geocoded residential coordinates onto the kriging-based prediction grid. Participants without sufficient residential address information could not be geocoded and were excluded from the exposure assessment. In principle, exposures could be assigned at the district centroid level when only district-level address information is available, but no such cases were included in this study34. For each participant, year-specific annual average concentrations were obtained during the study period (2012–2016) based on the timing of biospecimen collection, and these estimates were used in subsequent analyses. Target metabolomics The analytical procedures described below have in detailed previously35. Briefly, plasma samples were prepared using liquid–liquid extraction. The nonpolar phase was used for quantitation of sphingolipids and the polar phase was used to determine amino acids after chemical derivatization with phenyl isothiocyanate. A LC-MS/MS system equipped with a 1290 Infinity ultra-high performance liquid chromatography (Agilent Technologies, Palo Alto, CA, USA) and a Qtrap 5500 mass spectrometer (AB Sciex, Framingham, MA, USA) was used to profilemetabolites with multiple reaction monitoring in the positive-ion mode. The analysis included 40 metabolites: 7 ceramides (Cer), 5 sphingomyelins (SM), and 28 amino acids. Data analysis was performed using Analyst 1.5.1. Calibration curves ranged from 0.1 to 10 µM with R2 > 0.99. Calibration ranges were slightly modified depending on the analyte, and measurement results below the limit of quantitation were reported as zero. Commercial human plasma (Sigma-Aldrich, St. Louis, MO, USA) was used as a quality control (QC) sample. An equal number of samples from each area was allocated across the batches, and two QC samples were included in each batch. Batch correction was performed using QC samples, applying a normalization factor derived from the weighted mean of each analytical feature36 Most analytical features had a relative standard deviation of < 15%, with only a few exceptions. Statistical analyses Baseline characteristics of participants were analyzed using the Student’s t-test and chi-square test. Categorical variables were reported as frequencies and percentages [n (%)], whereas continuous variables were expressed as mean (standard deviation) [mean (SD)]. As sphingolipid and amino acid profiles were not normally distributed, metabolite intensities were log-transformed prior to analysis. Multiple linear regression models were used to evaluate associations between air pollutants and metabolites, adjusting for sex, age, BMI, residential area, COPD, smoking status, Charlson Comorbidity Index (CCI), year, average annual temperature, and season at the time of biospecimen collection. Sex was classified as men or women, and residential areas were categorized as Gangneung, Donghae, Samcheok, Yeongwol, Jecheon, or Danyang. Participants were grouped by COPD status, and smoking status was classified as nonsmoker, former smoker, or current smoker. CCI scores were categorized as 0, 1, 2, or ≥ 3. Year and season were treated as categorical variables, whereas age, BMI, and temperature were included as continuous covariates. Two participants had missing CCI data; therefore, multivariate imputation by chained equations (MICE) was performed using the R mice package. To address sex imbalance in the study population, inverse probability weighting (IPW) was applied in the final analysis using the R survey package. Air pollutant exposures were modeled as annual mean concentrations for each participant (PM10: µg/m³, NO2: ppb) without transformation. Metabolite levels were analyzed on the log-transformed intensity scale. Effect estimates were expressed as β coefficients with 95% confidence intervals (CIs), representing the change in log-transformed metabolite intensity per 1 µg/m³ increase in PM10 or per 1 ppb increase in NO2. For baseline characteristics (Table 1), statistical significance was defined as p < 0.05. For metabolite association analyses, raw p-values < 0.05 were considered nominally significant, and false discovery rate (FDR)-adjusted p-values (q-values) < 0.05 were considered statistically significant after multiple testing correction. Sensitivity analyses were performed by stratifying models according to sex, smoking status, and COPD status to assess the consistency of associations across subgroups. All statistical analyses were conducted using SAS version 9.4 (SAS Institute Inc., Cary, NC, USA), R version 4.2.3, and RStudio version 2023.03.0 + 386. Table 1Demographics of study participants.n (%) or Mean (SD)p-valueAlln = 188COPDNon-COPDn = 114(60.6%) n=74 (39.4%) Sex Women63 (33.51%)33 (28.95%)30 (40.54%)0.0999Men125 (66.49%)81 (71.05%)44 (59.46%)Age, years71.99 (6.53)72.96 (5.62)70.50 (7.51)0.0171 Study area GN13 (6.91%)8 (7.02%)5 (6.76%)0.7558DH5 (2.66%)3 (2.63%)2 (2.7%)SC27 (14.36%)20 (17.54%)7 (9.46%)YW81 (43.09%)47 (41.23%)34 (45.95%)JC47 (25.00%)28 (24.56%)19 (25.68%)DY15 (7.98%)8 (7.02%)7 (9.46%) Smoking status Current53 (28.19%)38 (33.33%)15 (20.27%)0.0272Former60 (31.91%)39 (34.21%)21 (28.38%)Never75 (39.89%)37 (32.46%)38 (51.35%)BMI, kg/m²23.11 (2.99)22.63 (2.88)23.86 (3.03)0.0056 CCI 0118 (62.77%)77 (67.54%)41 (55.41%)0.0848143 (22.87%)24 (21.05%)19 (25.68%)216 (8.51%)10 (8.77%)6 (8.11%)≥ 311 (5.85%)3 (2.63%)8 (10.81%) Season Summer15 (7.98%)8 (7.02%)7 (9.46%)0.3038Fall169 (89.89%)105 (92.11%)64 (86.49%)Winter4 (2.13%)1 (0.88%)3 (4.05%) Temperature (°C) 11.91 (0.91)12.01 (0.92)11.74 (0.86)0.0440*Data are presented as n (%) or mean ± SD. P-values were determined using Student’s t-test for continuous variables and the chi-square test for categorical variables. Bold values denote statistical significance at p < 0.05. GN, Gangneung; DH, Donghae; SC, Samcheok; YW, Yeongwol; JC, Jecheon; DY, Danyang; BMI, body mass index; CCI, Charlson Comorbidity Index. Table 1Demographics of study participants.n (%) or Mean (SD)p-valueAlln = 188COPDNon-COPDn = 114(60.6%) n=74 (39.4%) Sex Women63 (33.51%)33 (28.95%)30 (40.54%)0.0999Men125 (66.49%)81 (71.05%)44 (59.46%)Age, years71.99 (6.53)72.96 (5.62)70.50 (7.51)0.0171 Study area GN13 (6.91%)8 (7.02%)5 (6.76%)0.7558DH5 (2.66%)3 (2.63%)2 (2.7%)SC27 (14.36%)20 (17.54%)7 (9.46%)YW81 (43.09%)47 (41.23%)34 (45.95%)JC47 (25.00%)28 (24.56%)19 (25.68%)DY15 (7.98%)8 (7.02%)7 (9.46%) Smoking status Current53 (28.19%)38 (33.33%)15 (20.27%)0.0272Former60 (31.91%)39 (34.21%)21 (28.38%)Never75 (39.89%)37 (32.46%)38 (51.35%)BMI, kg/m²23.11 (2.99)22.63 (2.88)23.86 (3.03)0.0056 CCI 0118 (62.77%)77 (67.54%)41 (55.41%)0.0848143 (22.87%)24 (21.05%)19 (25.68%)216 (8.51%)10 (8.77%)6 (8.11%)≥ 311 (5.85%)3 (2.63%)8 (10.81%) Season Summer15 (7.98%)8 (7.02%)7 (9.46%)0.3038Fall169 (89.89%)105 (92.11%)64 (86.49%)Winter4 (2.13%)1 (0.88%)3 (4.05%) Temperature (°C) 11.91 (0.91)12.01 (0.92)11.74 (0.86)0.0440*Data are presented as n (%) or mean ± SD. P-values were determined using Student’s t-test for continuous variables and the chi-square test for categorical variables. Bold values denote statistical significance at p < 0.05. GN, Gangneung; DH, Donghae; SC, Samcheok; YW, Yeongwol; JC, Jecheon; DY, Danyang; BMI, body mass index; CCI, Charlson Comorbidity Index. Demographics of study participants. *Data are presented as n (%) or mean ± SD. P-values were determined using Student’s t-test for continuous variables and the chi-square test for categorical variables. Bold values denote statistical significance at p < 0.05. GN, Gangneung; DH, Donghae; SC, Samcheok; YW, Yeongwol; JC, Jecheon; DY, Danyang; BMI, body mass index; CCI, Charlson Comorbidity Index. Results Characteristics of study participants In total, 188 participants were included in the analysis, after excluding 12 individuals with missing data on confounding variables, such as personal PM10 and NO exposure concentrations, smoking status, BMI, and CCI (Fig. 1). Baseline characteristics of the study population are summarized in Table 1. The study population comprised 39.4% non-COPD (n = 74) and 60.6% COPD (n = 114) patients. Among them, 66.5% (n = 125) were male and 33.5% (n = 63) were female. The mean age and BMI of the participants was 71.99 years (SD = 6.53) and 23.11 kg/m2 (SD = 2.99), respectively. Among the participants, 39.89% (n = 75) were non-smokers, whereas 28.19% (n = 53) and 31.91% (n = 60) were former and current smokers, respectively. Among patients with comorbidities, CCI scores were distributed as follows: 0 (n = 118), 1 (n = 43), 2 (n = 16), and ≥ 3 (n = 11). Approximately 38% of participants had at least one comorbid condition alongside COPD. Using residential address-based air quality models, we estimated the concentrations of each pollutant for the study year. Air pollutant data were aligned with the year of blood sample collection. Table 2 summarizes the distribution of samples across years and the mean concentrations of PM10 and NO2 by exposure year. The long-term average exposures were 44.17 µg/m3 (SD = 2.8 µg/m3) for PM10 and 13.44 ppb (SD = 1.84 ppb) for NO2. Annual PM10 concentrations (2012–2016) were 45.13, 41.59, 40.86, 45.72, and 44.84 µg/m³ (SD = 1.97, 2.78, 2.97, 3.97, and 2.61 µg/m³, respectively). Annual NO concentrations were 13.24, 13.80, 13.81, 15.07, and 14.88 ppb (SD = 1.37, 3.27, 1.96, 0.42, and 1.29 ppb, respectively). The estimated exposure concentrations for both PM10 and NO2 exceeded the WHO annual thresholds of 15 µg/m³ and 10 µg/m³, respectively for all participants3. Fig. 1Flow chart of participant exclusion process in this study. Flow chart of participant exclusion process in this study. Table 2Characteristics of air pollutants.Year n PM10NO2Mean (SD)MinMedianMaxMean (SD)MinMedianMaxAll18844.17 (2.8)36.9744.8348.9913.44 (1.84)5.8513.3918.17201213445.13 (1.97)37.9345.3448.9913.24 (1.37)10.1613.2715.8220132741.59 (2.78)36.9741.6046.9213.8 (3.27)5.8514.6918.1720142040.86 (2.97)37.5739.9547.4513.81 (1.96)10.2614.5515.882015345.72 (3.97)41.1547.6348.3715.07 (0.42)14.6515.1015.482016444.84 (2.61)42.2444.8947.3314.88 (1.29)12.9715.3315.87Notes: Summary statistics for PM10 and NO2 concentrations are presented as n, mean (SD), minimum, median, and maximum. n, number of participants; SD, standard deviation; Min, minimum; Max, maximum; PM10, particulate matter ≤ 10 mm in aerodynamic diameter; NO2, nitrogen dioxide. Table 2Characteristics of air pollutants.Year n PM10NO2Mean (SD)MinMedianMaxMean (SD)MinMedianMaxAll18844.17 (2.8)36.9744.8348.9913.44 (1.84)5.8513.3918.17201213445.13 (1.97)37.9345.3448.9913.24 (1.37)10.1613.2715.8220132741.59 (2.78)36.9741.6046.9213.8 (3.27)5.8514.6918.1720142040.86 (2.97)37.5739.9547.4513.81 (1.96)10.2614.5515.882015345.72 (3.97)41.1547.6348.3715.07 (0.42)14.6515.1015.482016444.84 (2.61)42.2444.8947.3314.88 (1.29)12.9715.3315.87Notes: Summary statistics for PM10 and NO2 concentrations are presented as n, mean (SD), minimum, median, and maximum. n, number of participants; SD, standard deviation; Min, minimum; Max, maximum; PM10, particulate matter ≤ 10 mm in aerodynamic diameter; NO2, nitrogen dioxide. Characteristics of air pollutants. Notes: Summary statistics for PM10 and NO2 concentrations are presented as n, mean (SD), minimum, median, and maximum. n, number of participants; SD, standard deviation; Min, minimum; Max, maximum; PM10, particulate matter ≤ 10 mm in aerodynamic diameter; NO2, nitrogen dioxide. Effect of long-term exposure to air pollutants on plasma metabolites In the main analyses of the total cohort (Fig. 2, Supplementary TableS1), long-term PM10 exposure was positively associated with higher plasma levels of SM(d18:1/18:0), SM(d18:1/18:1), and SM(d18:1/24:1) (p = 0.0498, 0.0027, and 0.0428, respectively). Long-term NO2 exposure was positively associated with plasma Cer (d18:1/24:0) (p = 0.0232) and the amino acid glutamate (p = 0.0366). These associations were statistically significant at the nominal level (p < 0.05) but did not remain significant after FDR correction. No other amino acids showed significant associations with PM10 exposure. Fig. 2Forest plots presenting associations between long-term exposure to air pollutants and plasma metabolites. (A) Associations between annual average PM10 exposure and plasma sphingolipid and amino acid concentrations. (B) Associations between annual average NO2 exposure and plasma sphingolipid and amino acid concentrations. Exposure estimates were derived from a national-scale air pollution prediction model. Multivariable linear regression models were adjusted for age, sex, body mass index, smoking status, chronic obstructive pulmonary disease, Charlson Comorbidity Index, residential area, year, season, and temperature. Forest plots display β coefficients with 95% confidence intervals. Blue markers indicate nominally significant associations (p < 0.05), whereas black markers indicate non-significant associations. Abbreviations: Cer, ceramide; SM, sphingomyelin. Forest plots presenting associations between long-term exposure to air pollutants and plasma metabolites. (A) Associations between annual average PM10 exposure and plasma sphingolipid and amino acid concentrations. (B) Associations between annual average NO2 exposure and plasma sphingolipid and amino acid concentrations. Exposure estimates were derived from a national-scale air pollution prediction model. Multivariable linear regression models were adjusted for age, sex, body mass index, smoking status, chronic obstructive pulmonary disease, Charlson Comorbidity Index, residential area, year, season, and temperature. Forest plots display β coefficients with 95% confidence intervals. Blue markers indicate nominally significant associations (p < 0.05), whereas black markers indicate non-significant associations. Abbreviations: Cer, ceramide; SM, sphingomyelin. Sensitivity analyses stratified by sex, COPD status, and smoking status are presented in Fig. 3 and detailed in Supplementary Tables S2–S4. Consistent with the main analysis, subgroup analyses revealed that PM10 exposure was most strongly associated with sphingomyelins, particularly SM(d18:1/18:1) and SM(d18:1/24:1), in males, ever-smokers, and patients with COPD. Notably, SM(d18:1/16:0), SM(d18:1/18:0), SM(d18:1/18:1), and SM(d18:1/24:1) remained significant after FDR correction in the COPD subgroup, underscoring enhanced susceptibility in this population. In contrast, NO2 exposure primarily influenced amino acid metabolism, with kynurenine being significantly associated with male patients, as well as valine and isoleucine. Additional subgroup-specific associations were observed—such as ceramide (d18:1/18:0) in females, methionine, ornithine, and Cer (d18:1/24:0) in non-COPD individuals, and several amino acids (e.g., t4-OH-proline, α-aminoadipic acid, asymmetric dimethylarginine) in never-smokers—though most did not withstand FDR adjustment. Overall, these findings indicate that both sphingolipid and amino acid pathways may be differentially responsive to PM10 and NO2 in susceptible subgroups, Fig. 3Subgroup analyses of the associations between long-term air pollution exposure and plasma metabolites. (A) Results for PM10 exposure stratified by sex (men vs. women), smoking status (ever-smokers vs. never-smokers), and chronic obstructive pulmonary disease (COPD) status (COPD vs. non-COPD). (B) Results for NO2 exposure stratified by the same effect modifiers. Percent changes (95% confidence intervals) in metabolite levels are presented, with continuous variables replaced by categorical effect modifiers to assess subgroup-specific susceptibility. Each colored line indicates the effect estimate for a given subgroup, with corresponding p-values and p for interaction values displayed to the right. Statistically significant associations are highlighted. Abbreviations: Cer, ceramide; SM, sphingomyelin; ADMA, asymmetric dimethylarginine; alpha-AAA, alpha-aminoadipic acid. Subgroup analyses of the associations between long-term air pollution exposure and plasma metabolites. (A) Results for PM10 exposure stratified by sex (men vs. women), smoking status (ever-smokers vs. never-smokers), and chronic obstructive pulmonary disease (COPD) status (COPD vs. non-COPD). (B) Results for NO2 exposure stratified by the same effect modifiers. Percent changes (95% confidence intervals) in metabolite levels are presented, with continuous variables replaced by categorical effect modifiers to assess subgroup-specific susceptibility. Each colored line indicates the effect estimate for a given subgroup, with corresponding p-values and p for interaction values displayed to the right. Statistically significant associations are highlighted. Abbreviations: Cer, ceramide; SM, sphingomyelin; ADMA, asymmetric dimethylarginine; alpha-AAA, alpha-aminoadipic acid. Discussion This study identified long-term exposure to PM10 and NO2 as environmental factors associated with alterations in plasma sphingolipid metabolism, with PM10 positively associated with SM(d18:1/18:0), SM(d18:1/18:1), and SM(d18:1/24:1), and NO2associated with Cer(d18:1/24:0). Notably, NO2 exposure—commonly considered a marker of traffic-related air pollution—was also positively associated with glutamate, suggesting that traffic-related pollutants may influence amino acid metabolism. In contrast, PM₁₀ exposure was not significantly associated with any amino acids. These findings highlight that lipid and amino acid pathways may be influenced by long-term air pollution exposure, with stronger effects observed in vulnerable groups such as men, smokers, and patients with COPD. Alterations in lipid metabolism driven by oxidative stress and inflammation may underlie the adverse health effects of air pollution. Sphingolipids—critical for membrane integrity, cell signaling37, and immune regulation38—are particularly susceptible to disruption. In our study, chronic PM10 exposure showed suggestive positive associations with increased SM (d18:1/18:1) and SM (d18:0/24:1), likely reflecting a compensatory mechanism to stabilize membranes against oxidative damage. This aligns with previous reports linking air pollution to sphingolipid dysregulation13. PM-induced oxidative stress activates neutral and acidic sphingomyelinase (SMase), leading to ceramide accumulation. This promotes the flux of SM to Cer, which can lead to subsequent cellular damage. There is experimental evidence that particulate extracts activate SMase in an NADPH oxidase-dependent manner, thereby increasing ceramide concentration. This in turn induces apoptosis and inflammatory signaling pathways in airways and vascular cells39. SM–ceramide interconversion may serve as an important adaptive response to chronic pollutant exposure, but can also lead to pathological phenomena. Excessive ceramide accumulation induces mitochondrial membrane damage, promotes the production of reactive oxygen species, and increases the secretion of inflammatory cytokines, thereby forming a feed-forward loop of oxidative damage40,41. Recent metabolomic studies have demonstrated a significant correlation between short-term exposure to air pollution, increased plasma ceramide concentrations, and neutrophil activation, supporting the role of the oxidation–lipid axis in human populations42. These studies suggest that fine dust pollution can re-regulate sphingolipid metabolism through the redox-mediated SMase–ceramide signaling pathway, providing a biological basis for the PM10-associated changes in sphingomyelin observed in the present study population. For instance, the SCOPE study observed increased sphingolipids (e.g., ceramide, sphingomyelin) following short- and medium-term PM2.5 exposure, corroborating our PM10 findings and highlighting their connection to pathways such as lower- high-density lipoprotein cholesterol modulation38,43. Our study provides suggestive and novel evidence for an association between long-term NO2 exposure and plasma Cer (d18:1/24:0), a ceramide subclass not previously analyzed. Prior epidemiological studies have mainly focused on Cer(d18:1/16:0), Cer(d18:1/18:0), and Cer(d18:1/24:1), linking them to cardiometabolic outcomes and impaired lung function10,38. In contrast, the observed association between NO2 exposure and Cer(d18:1/24:0) points to distinct pollutant-specific metabolic pathways. Although prior research has focused on Cer(d18:1/x) subclasses in cardiometabolic and pulmonary disease, the specific role of Cer(d18:1/24:0) may mediate effects via mitochondrial dysfunction, membrane remodeling, or differential susceptibility to oxidative stress40,44. Zhou et al.45 reported that Cer(d18:1/24:0) modulated proliferation and invasion in cancer cells, suggesting bioactive functions beyond simple lipid accumulation. Moreover, Lachkar et al.46 highlighted how ceramide subclasses participate as metabolic intermediates within biosynthetic fluxes, indicating that altered desaturation or elongation steps may underlie pollutant-specific responses. Such disruption to lipid homeostasis may contribute to lung structural damage and mitochondrial dysfunction, observed in COPD, asthma, and cystic fibrosis. These results add to the growing body of evidence and emphasize the need for further mechanistic studies on ceramide dysregulation under chronic pollutant exposure. In addition to these lipid alterations, our study also identified significant associations with amino acid metabolites, particularly glutamate. Amino acids are critical to biological systems; however, owing to the complex chemical nature of air pollutants, high-throughput and sensitive approaches are required to elucidate the molecular mechanisms underlying amino acid responses to air-pollution47. In our main analysis, most amino acids did not show significant associations with long-term PM10 or NO2 exposure; however, we identified a positive association between NO2 exposure and glutamate. This finding suggests that glutamate may represent a particularly sensitive amino acid marker of traffic-related pollution. Previous untargeted metabolomics research has reported associations between NO2 exposure and perturbations in amino acid–related metabolic pathways, even if glutamate itself was not always highlighted individually13. More recently, large-scale metabolome-wide association studies have identified overlapping metabolite signatures associated with ambient air pollution, including disruptions in amino acid–related pathways, thereby supporting the plausibility of amino acid dysregulation under long-term NO2 exposure48. Mechanistically, oxidative stress may indirectly influence glutamate metabolism by altering the redox balance and mitochondrial function, leading to possible disruptions in the tricarboxylic acid cycle and nitrogen turnover49. Although the exact causal link between NO2 exposure and glutamate accumulation remains to be elucidated, this pathway provides a biologically plausible explanation connecting oxidative stress with amino acid dysregulation under chronic air pollution50. Further experimental studies are required to verify whether oxidative stress-mediated metabolic perturbations directly contribute to the observed Glu elevation during long-term exposure. Mechanistic studies also provide supporting evidence, with short-term exposure experiments showing perturbations in amino acid metabolism, including glutamate turnover, in response to pollution22,23. Taken together, while the majority of amino acids did not show consistent associations in our cohort, the observation of an NO2–glutamate link provides subclass-specific evidence of pollutant-induced amino acid disruption. This finding warrants replication in independent cohorts and mechanistic exploration into whether glutamate can serve as a biomarker of chronic NO2 exposure. Chronic activation of the SMase–ceramide pathway in patients with COPD and in smokers may exacerbate the sphingomyelin response to PM10 exposure. The existing literature reveals that both cigarette smoke and sustained inflammation enhance acidic SMase activity and ceramide synthesis, thereby facilitating apoptosis and driving tissue remodeling processes51,52. Notably, subgroup analyses in the present study indicated a heightened association between PM10 levels and sphingomyelin in COPD patients and smokers, aligning with the established hypothesis that oxidative stress triggers inflammatory ceramide signaling through lipid remodeling mechanisms. In sensitivity analyses stratified by sex, smoking status, and COPD, we observed stronger and more consistent associations than in the main cohort analyses. For PM₁₀, associations with sphingomyelin species (SM(d18:1/18:1) and SM(d18:1/24:1)) were more apparent in males, ever-smokers, and individuals with COPD, suggesting that chronic pollutant exposure may amplify membrane lipid remodeling in susceptible groups with higher background oxidative stress and inflammation10,53,54. This pattern is consistent with prior reports that particulate matter perturbs sphingolipid pathways and contributes to vascular and pulmonary injury25. For NO2 and contributes to vascular and pulmonary injuryer-smokers, and individuals with COPD, suggesting that chronic pollutant exposure may amplify membrane lipid remodeling in susceptible groups with higher background oxidative sother amino acid pathways in response to air pollution exposure, and elevated glutamate levels have been mechanistically linked to oxidative stress, mitochondrial dysfunction, and lung tissue remodeling in chronic respiratory diseases55. In females and non-COPD participants, PM10 exposure was associated with methionine metabolism and ceramide (d18:1/18:1), suggesting that specific metabolic pathways may be differentially responsive depending on clinical context. Notably, subgroup patterns revealed that demographic and clinical factors critically modulate the metabolic response to air pollution. Males showed stronger PM10-associated elevations in SM(d18:1/18:1) and SM(d18:0/24:1), while NO₂-exposed males exhibited greater shifts in Cer(d18:1/18:1), potentially owing to hormonal influences on oxidative stress and lipid turnover56, metabolic differences in adiposity and lipid handling57, or behavioral factors such as higher smoking prevalence58. Patients with COPD demonstrated heightened sphingolipid sensitivity to PM10 and NO2, likely reflecting preexisting metabolic dysregulation observed in COPD metabolomic profiling35,54, and smokers exhibited amplified pollutant–metabolite associations59, consistent with the hypothesis that smoking exacerbates susceptibility to the adverse effects of air pollution. Such selective responsiveness across metabolites is biologically plausible and highlights the need for pathway-level integration when interpreting pollutant–metabolite associations. Taken together, these subgroup findings support a vulnerability framework in which preexisting oxidative and inflammatory burden unmasks pollutant-related metabolic shifts—sphingolipid remodeling under PM10 exposure and amino acid perturbations under NO2, particularly glutamate. This study has several strengths. First, the inclusion of PM10 and NO2 exposure enabled a comprehensive assessment of pollutant-specific metabolic effects. Second, the targeted metabolomics approach allowed reliable quantification of sphingolipids and amino acids using authenticated standards and MS/MS transitions, reducing annotation uncertainties common in un-targeted workflows. Third, residential stability among participants improved the accuracy of long-term exposure estimates26. In addition, the nationwide kriging-based air pollution model applied in this study was developed and validated using regulatory monitoring data, and its performance (cross-validated R² = 0.45 for PM10 and 0.82 for NO2) was comparable to that of national-scale models in Europe and the United States. Taken together, these strengths provide confidence in the robustness of our approach. Nonetheless, some limitations should be acknowledged. The relatively small sample size, particularly in certain enrollment years, limited the statistical power to detect modest associations. Although we applied IPW to address sex imbalance partially and used multiple imputation for missing values, residual selection bias remains possible due to the predominance of older adults and the specific regional distribution of the cohort. Survival bias may also be present, as individuals most affected by pollution could be underrepresented in this older adult cohort. Furthermore, the absence of an external validation dataset restricts the generalizability of our findings. Although our exposure assessment relied on a validated nationwide kriging-based prediction model, the metabolomic associations themselves were not independently replicated. Finally, most associations did not remain significant after FDR correction, highlighting the need for cautious interpretation. Because this analysis was conducted in a retrospective COPD cohort with single-time biospecimen collection, causal inferences between air pollution exposure and plasma metabolite changes remain limited.Although spatial models for historical pollutant concentrations effectively captured spatial contrasts (as validated in prior work), temporal misalignment or measurement error may have biased exposure estimates. Furthermore, lack of PM2.5 data restricted multi-pollutant analysis60,61. Uncontrolled confounders—including dietary patterns, genetic variability, and self-reported lifestyle factors (e.g., smoking status inaccuracies)— may also have influenced the associations between metabolites and pollution. The study’s overrepresentation of older adults (≥ 65 years) limited extrapolation to younger populations, while the heightened sensitivity of COPD patients could bias outcomes. Although confounding variables were adjusted for, residual bias may persist due to unmeasured cumulative exposures or incomplete control of covariates. Future studies with larger, more diverse cohorts, independent replication datasets, and longitudinal designs are warranted to validate our observations and further explore the role of air pollution in metabolic dysregulation. Conclusions This study demonstratedthat chronic PM10 and NO2 exposure is associated with alterations in sphingolipid metabolism, with PM10 linked to elevated levels of SM(d18:1/18:0), SM(d18:1/18:1), and SM(d18:0/24:1), and NO2 associated with increased levels of Cer(d18:1/24:0) and glutamate.Subgroup analyses revealed heightened susceptibility among male, smokers, and patients with COPD, underscoring the influence of demographic and clinical factors on metabolic responses. These findings highlight the importance of sphingolipid profiling in high-risk populations to inform personalized prevention strategies. Future research should prioritize longitudinal assessments of Cer(d18:1/24:0) and glutamate as potential biomarkers of NO2-induced metabolic dysfunction, and explore intervention strategies for vulnerable subgroups. Supplementary Information Below is the link to the electronic supplementary material. Supplementary Material 1 Supplementary Material 1