Ambient PM_2.5 species and ultrafine particle exposure and their differential metabolomic signatures

Abstract

Background:

The metabolomic signatures of short- and long-term exposure to PM_2.5 have been reported and linked to inflammation and oxidative stress. However, little is known about the relative contribution of the specific PM_2.5 species (hence sources) that drive these metabolomic signatures.

Objectives:

We aimed to determine the relative contribution of the different species of PM_2.5 exposure to the perturbed metabolic pathways related to changes in the plasma metabolome.

Methods:

We performed mass-spectrometry based metabolomic profiling of plasma samples among men from the Normative Aging Study to identify metabolic pathways associated with PM_2.5 species. The exposure windows included short-term (one, seven-, and thirty-day moving average) and long-term (one year moving average). We used linear mixed-effect regression with subject-specific intercepts while simultaneously adjusting for PM_2.5, NO₂, O₃, temperature, relative humidity, and covariates and correcting for multiple testing. We also used independent component analysis (ICA) to examine the relative contribution of patterns of PM_2.5 species.

Results:

Between 2000 and 2016, 456 men provided 648 blood samples, in which 1,158 metabolites were quantified. We chose 305 metabolites for the short-term and 288 metabolites for the long-term exposure in this analysis that were significantly associated (p-value <0.01) with PM_2.5 to include in our PM_2.5 species analysis. On average, men were 75.0 years old and their body mass index was 27.7 kg/m². Only 3% were current smokers. In the adjusted models, ultrafine particles (UFPs) were the most significant species of short-term PM_2.5 exposure followed by nickel, vanadium, potassium, silicon, and aluminum. Black carbon, vanadium, zinc, nickel, iron, copper, and selenium were the significant species of long-term PM_2.5 exposure. We identified several metabolic pathways perturbed with PM_2.5 species including glycerophospholipid, sphingolipid, and glutathione. These pathways are involved in inflammation and oxidative stress, immunity, and nucleic acid damage and repair. Results were overlapped with the ICA.

Conclusions:

We identified several significant perturbed plasma metabolites and metabolic pathways associated with exposure to PM_2.5 species. These species are associated with traffic, fuel oil, and wood smoke. This is the biggest study to report a metabolomic signature of PM_2.5 species’ exposure and the first to use ICA.

1. Introduction

Approximately 4.2 million deaths in 2015 were attributed to long-term exposure to ambient fine particulate matter ≤2.5 μm (PM_2.5)¹. This accounts for 7.6% of the total global mortality, thus making long-term exposure to PM_2.5 the fifth-ranked global risk factor¹. Inflammation and oxidative stress^2–5 are biological processes that have been linked to both short- and long-term exposures to PM_2.5 and are also associated with many adverse health outcomes such as pulmonary⁶, cardiovascular⁷, and neurological diseases⁸, as well as mortality^9–11. As there is need for the mechanistic understanding that may aid prevention or treatment, more studies have examined the metabolomic signatures using metabolomic profiling in the plasma for ambient PM_2.5 exposure including in this study population, the Normative Aging Study (NAS)^12–21. The plasma metabolome is a collection of biologically active chemicals in the plasma derived from both endogenous processes and exogenous exposures²². These signatures have been linked to metabolic pathways related to inflammation, oxidative stress, and DNA damage.

PM_2.5 is a complex mixture of organic compounds, elemental carbon, ions, and metal oxides. These PM_2.5 species reflect specific exposure sources and have different physicochemical and toxicological properties²³ and hence differences in their health effects^24,25. Although, there is a plethora of studies on the health effects of PM_2.5 exposure, there is still uncertainty about which species are most detrimental, and for which health outcomes. In addition, to date, no human studies have examined the different species of long-term PM_2.5 exposure and their independent metabolic signatures while adjusting for the main PM_2.5 mass and limited studies for the short-term exposure^12–20. Therefore, in this study we aimed to determine changes in the plasma metabolome related to the species of short- and long-term PM_2.5 exposure, while adjusting for the main PM_2.5 mass concentrations. Such adjustment assures that our results can be interpreted as differential effects of species. We hypothesized that PM_2.5 species of short- and long-term exposure can differentially contribute to differences in the plasma metabolome consistent with perturbations to different metabolic pathways. This will help us understand the differential contribution of the different species and their major sources compared to the average PM_2.5 on the plasma metabolome.

2. Materials and methods

2.1. Participants and study design

The NAS was established in 1963 as a longitudinal study of aging men in Boston, Massachusetts. Men (N=2,280), 21-80 years old between 1963 and 1970 and free of known chronic diseases enrolled in NAS and have been followed-up since²⁶. NAS was approved by the institutional review boards of the Department of Veterans Affairs and Harvard T.H. Chan School of Public Health. All participants provided written informed consent. Every 3-5 years men had physical examinations and laboratory tests. During follow-up visits, participants self-reported their demographic, lifestyle, and medical history, and dietary intake. In addition, at every visit, a 7ml fasting venous blood sample was collected from participants in a trace metal-free tube containing ethylenediaminetetraacetic acid. The samples were spun at 3,000 revolutions/ minute for 15 minutes. Plasma samples were placed in 1.8 ml Nunc tubes for long-term storage at −80°C. In the current analysis, 464 men contributed 659 visits/samples between 2000 and 2016 and had plasma metabolomics measured in these samples. Profiling of the samples collected before 2000 were not suitable due to their storage conditions. Only 8 men were non-white, therefore, we restricted this analysis to 456 white men who contributed 648 plasma samples.

2.2. Quantification of PM_2.5 species

We measured daily PM_2.5, and 14 of its species: ambient UFP concentrations [number of particles (PN)/cm³], black carbon (BC), sodium (Na), iron (Fe), aluminum (Al), silicon (Si), potassium (K), nickel (Ni), vanadium (V), sulfur (S), selenium (Se), lead (Pb), zinc (Zn), and titanium (Ti). We focused on these species as their levels were mostly above detection limits and representative of different PM_2.5 sources²⁷. All exposures were measured at a fixed monitoring site at the ambient air particle monitoring Harvard Supersite, located on the roof of the Countway Library of the Harvard Medical School in downtown Boston, MA and approximately 1 km from the examination center. Because study participants lived in greater Boston with a median distance of 20.8 km from the examination center²⁸, we considered the ambient air pollutant levels as surrogates of participants’ exposures. To estimate short-term exposure, we calculated the moving averages 30-days and 7-days prior, and on the day of each blood draw. To estimate long-term exposure, we calculated the 365-day moving average up to the visit when blood was drawn.

PM_2.5 (μg/m³) levels were measured hourly using a tapered element oscillation microbalance (Model 1400A, Rupprecht and Pastashnick)²⁹. BC (μg/m³) was measured using an aethalometer (Magee Scientific Inc., Berkeley, CA, USA). UFPs (diameter <100 nm) were measured as number of particles ((PN)/ cm³) using a condensation particle counter (CPC, model 3022A; TSI Inc., Shoreview, MN). The other twelve PM_2.5 species (μg/m³) (Na, Ni, Al, Si, S, K, V, Fe, Ti, Zn, Se, and Pb) were assessed using the Energy Dispersive X-ray Fluorescence Spectrometers (Epsilon 5, PANalytical, Almelo, Netherlands) from PM_2.5 filters.

2.3. Metabolomic Profiling

All samples were sent to the lab for analysis, at the same time. Metabolomic profiling was conducted using untargeted high-resolution Ultrahigh Performance Liquid Chromatography Coupled Tandem Mass Spectroscopy (UPLC-MS/MS) enabling the broadest coverage of the metabolome by Metabolon Inc. (Durham, NC, USA). Methods were described in detail previously³⁰. In summary, the four platforms were: (1) UPLC-MS/MS under positive ionization for early eluting metabolites, (2) UPLC-MS/MS under positive ionization for late eluting metabolites, (3) UPLC-MS/MS under negative ionization, and (4) UPLC-MS/MS, polar platform (negative ionization). The samples were randomized for profiling. In-house standards and quality control (QC) measures were followed³⁰. Metabolites were identified by their mass/charge ratio, retention time, and through a comparison to a library of purified known standards. Metabolites were quantified using area-under-the-curve (AUC) of the peak and processed according to our in-house standard QC pipeline³¹. In brief, for a given metabolite, missing values were imputed with half the minimum observed value for that metabolite. Metabolite levels were then log-transformed and pareto-scaled³⁰. Overall, 1,301 metabolites were profiled. We excluded 143 metabolites because they had an interquartile range of zero and considered uninformative leaving 1,158 metabolites . We previously examined and reported the associations between short- and long-term exposure to PM_2.5 as a whole with metabolites in the same population²¹. Therefore, for the PM_2.5 species in the current analysis, we included only the metabolites that were significantly (at level of p-value <0.01) associated with the PM_2.5 i.e., 305 metabolites for the short-term (with any of the 3 moving averages) and 288 metabolites for the long-term.

2.4. Statistical Analysis

We calculated descriptive statistics to summarize socio-demographics and lifestyle factors for the participants at the first visit and all visits. We examined whether PM_2.5 species of short-term (moving averages of 30-days, 7-days prior, and on same day of the visit), and long-term (365-day moving average) exposure were associated with differences in the levels of the metabolites while adjusting for the PM_2.5 mass concentrations. We used time-varying linear mixed-effect regression models (LMEM) with random participant-specific intercepts, accounting for the correlation of repeated measures within participants. We modeled the outcomes and the exposures at each exposure window as continuous variables. We adjusted for the same moving averages of NO₂ and O₃ (from local state monitors in the greater Boston area), ambient temperature and relative humidity (from the National Weather Service Station at Logan Airport (Boston, MA), located approximately 12 km from the examination site), age (years), body mass index (BMI, kg/m²), cigarette pack-years, alcohol intake (< or ≥2 drinks per day), socioeconomic status (income and years of education), and season (warm/cold) at each visit. We used the species levels adjusted for PM_2.5 mass method^32,33 where we simultaneously adjusted for the 14 species (multi-species models) in addition to the main PM_2.5 mass itself, to account for their correlated nature³². Since the coefficient of a species in a multiple regression estimates the effect of that species holding all other covariates constant, including PM_2.5 mass, the coefficient of a species estimates the effect of one unit increase in that species holding total mass constant, i.e., substituting for the other species. Hence it measures the relative effect of that species, with positive numbers indicating a larger effect per unit mass than average and negative coefficients indicating a smaller effect than average per unit mass.

Multi – species models:

$$E\left(Y_{ij}\right)={\beta }_0+{\mu }_i+{\beta }_{1-14}specie{s}_{ij}+{\beta }_{15}P{M}_{2.5ij}+{\beta }_{16}N{O}_{2ij}+{\beta }_{17}{O}_{3ij}+{\beta }_{18}Temperatur{e}_{ij}+{\beta }_{19-n}Covariate{s}_{ij}$$

where Y_ijis the metabolome level of subject i at visit j, β₀ is the fixed intercept, μ_iis the random intercept for subject i, constituents _ij are the moving averages of the 14 species (BC, UFP, Na, Ni, Al, Si, S, K, V, Fe, Ti, Zn, Se, and Pb) for subject i at visit j.

To account for multiple testing while also accounting for the high correlation between metabolites that are closely connected through interlinked biological pathways, we used the “number of effective/independent tests” (ENT) approach^34,35. We determined the number of principal components required to explain a given percentage of the variance in the metabolite data i.e., ENT. We used thresholds of 95% (ENT95%) and 99% (ENT99%) variance explained for both short- and long-term analyses. We calculated the adjusted p-value threshold as α/m where α denotes the nominal p-value threshold of 0.05, and m denotes the number of principal components at the given threshold and divided by 14 (PM_2.5 species) and for the short-term analysis divided further by 3 (moving averages).

In order to explore metabolomic profiles rather than individual metabolites, we next used an independent component analysis (ICA)³⁶ as an unsupervised technique to reduce the highly-correlated metabolites into five independent factors. Because, we previously showed that only ICA-factor 2 was significantly associated with PM_2.5 exposure²¹, we used only that factor in this analysis with PM_2.5 species. We then conducted LMEM where ICA-factor 2 was the outcome while exposures and covariates remained as above. For each ICA-factor, there is an attached weight for each metabolite to determine its individual contribution to that factor. We corrected for multiple testing in the ICA models by applying Benjamini-Hochberg false discovery rate (FDR) and set the false positive threshold as 5%³⁷ and then divided by 14 (PM_2.5 species) and for the short-term exposure further divided by 3 (moving averages).

ICA is a computational method for separating a multivariate signal into additive factors that are maximally independent. It is similar to its cousin, principal component analysis (PCA), and differs in that PCA divides multiple correlated variables into underlying factors that are uncorrelated, but not necessarily independent. For the latter to hold the factors all need to be Gaussian distributed. In contrast, ICA allows for factors that are non-Gaussian, and makes them independent, which involves more than just being uncorrelated³⁶. If the variables are Gaussian distributed then uncorrelated factors are also independent and the results should be the same, however for non-Gaussian distributed variables such as metabolites’ levels, that is not true.

As sensitivity analyses, we conducted LMEM models for each single PM_2.5 species at a time (single-species models) while adjusting for PM_2.5 mass and the other covariates as shown below. In this case, we were assessing the effect of a change in exposure to one species holding total mass constant, and hence substituting for a mix of other species.

Single – species models:

$$E\left(Y_{ij}\right)={\beta }_0+{\mu }_i+{\beta }_1{\mathrm{species}}_{ij}+{\beta }_{2}P{M}_{2.5ij}+{\beta }_{3}N{O}_{2ij}+{\beta }_4{O}_{3ij}+{\beta }_{5}Temperatur{e}_{ij}+{\beta }_{6-n}Covariate{s}_{ij}$$

where Y_ij is the metabolome level of subject i at visit j, β₀ is the fixed intercept, μ_i, is the random intercept for subject i, constituent _ij is the moving average of one species at a time (for subject i at visit j.

Analyses were conducted using R version 3.6.0 and SAS version 9.4 (SAS Institute Inc., Cary, NC, USA).

We then further conducted pathway analysis for the significant metabolites (at p-value <0.01) from the multi- species model and the 100 metabolites that had the highest contributing weights to the ICA-factor2. To do that, we used the ‘Pathway Analysis’ functionality in MetaboAnalyst 4.0, that accounts for both over-representation analyses (i.e., number of significant metabolites within a pathway) and pathway topology (i.e., influence of those metabolites to that given pathway)³⁸ and uses Human Metabolome Database (HMDB), and Kyoto Encyclopedia of Genes and Genomes (KEGG) databases to define the underlying pathways. We considered statistical significance for the pathways at level of p-value ≤ 0.1²¹ and considered additional noteworthy pathways if the impact score was ≥ 0.5 while p-value < 0.3.

3. Results

Between 2000 and 2016, 456 white men provided 648 plasma samples, in which 1,158 metabolites were quantified, passed our in-house QC pipeline, and we had access to all features. In this analysis, we included 305 and 288 metabolites for the short- and long-term exposure, respectively which were significantly associated (at level of p-value <0.01) with PM_2.5 when modeled alone without the individual PM_2.5 species²¹. Approximately 64% of the participants provided one plasma sample, 36% provided two or three samples. At the initial visit, participants were on average 75.0 years old, had a mean BMI of 27.7 kg/m², and had received an average of 15.1 years of education (Table 1). Only 3% were current smokers and 69% of the participants were former smokers. Approximately, 79% of the participants consumed <2 alcoholic drinks/day. Around 60% of the visits were in the warm season (April to September). Average air pollution and temperature measures remained relatively stable across the different exposure windows (Table 2)

Table 1.

Characteristics of the study population in the Normative Aging Study (2000 to 2016)

Demographic Characteristics	At first visit (2000 to 2013) (N=456)	All visits (2000 to 2016) (N=648)
Age, Mean (SD)	75.0 (6.70)	76.0 (6.68)
Body mass index (kg/m2), Mean (SD)	27.7 (4.19)	27.7 (4.27)
Years of education, Mean (SD)	15.1 (3.00)	15.0 (3.00)
Baseline annual income, thousands $US, Mean (SD)	8.61 (3.77)	8.64 (3.78)
Smoking
Never, N (%)	128 (28)	190 (29)
Current, N (%)	14 (3)	21 (3)
Former, N (%)	314 (69)	437 (67)
Pack-year smoked (years), Mean (SD)	21.4 (25.2)	20.95 (25.4)
Season
Cold: October - March, N (%)	189 (41)	259 (40)
Warm: April - September, N (%)	267 (59)	389 (60)
Alcohol consumption (≥2 drinks per day), N (%)	95 (21)	132 (20)

Abbreviations: N, number of participants or visits; SD, standard deviation; Kg, Kilogram; m, meters.).

Table 2.

Mean (SD) of the moving averages of air pollutant and temperature levels in all participants’ visits 2000 to 2016 (N=648)

Air pollutant and temperature moving averages	24-hours moving averages	7-days moving averages	30-days moving averages	365-days moving average
NO2 (ppb)	19.2 (5.82)	17.6 (3.85)	17.7 (3.25)	18.4 (2.35)
O3 (ppb)	24.7 (12.6)	25.1 (8.48)	25.4 (7.36)	24.0 (1.12)
Temperature (°C)	13.9 (8.66)	13.4 (7.89)	13.3 (7.62)	11.0 (0.60)
PM2.5 (μg/m3)	10.3 (6.04)	9.79 (3.69)	9.87 (2.70)	10.1 (1.31)
PM2.5 species
BC (μg/m3)	0.7796(0.3873)	0.702(0.2173)	0.7044(0.1691)	0.6979(0.1021)
UFPs (PN/ cm3)	20072.7(11296.97)	18960.54(9339.84)	19316.68(9090.84)	21624.36(6663.21)
Na (μg/m3)	0.2007(0.1388)	0.1944(0.0725)	0.1992(0.0442)	0.2012(0.0164)
Al (μg/m3)	0.057(0.0461)	0.0547(0.0306)	0.0539(0.0187)	0.051(0.0082)
Si (μg/m3)	0.0855(0.0752)	0.079(0.0529)	0.0772(0.0329)	0.073(0.0154)
S (μg/m3)	1.1394(0.9223)	1.0802(0.5317)	1.0939(0.3761)	1.067(0.1482)
K (μg/m3)	0.0369(0.0233)	0.0433(0.0599)	0.0417(0.0254)	0.0402(0.0033)
Cu (μg/m3)	0.0038(0.0041)	0.0037(0.0019)	0.0036(0.0008)	0.0035(0.0003)
V (μg/m3)	0.0029(0.0025)	0.0029(0.0016)	0.003(0.0014)	0.0034(0.0013)
Se (μg/m3)	0.0002(0.0006)	0.0002(0.0003)	0.0002(0.0001)	0.0002(0.00005)
Fe (μg/m3)	0.0754(0.0359)	0.0669(0.0187)	0.0661(0.0123)	0.0654(0.0095)
Ni (μg/m3)	0.0025(0.0023)	0.0024(0.0016)	0.0025(0.0016)	0.003(0.0014)
Zn (μg/m3)	0.0124(0.0095)	0.0111(0.0049)	0.0109(0.0035)	0.0114(0.0022)
Pb (μg/m3)	0.0054(0.0033)	0.0058(0.0017)	0.0057(0.001)	0.0057(0.0006)

Abbreviations: N, number of visits; SD, standard deviation; ppb, is part per billion; C, Celsius; PM_2.5, particulate matter ≤2.5 μm; NO₂, nitrogen dioxide; and O₃, ozone; PN: particle number (measure for ultrafine particles (UFPs)); BC: black carbon; sodium (Na); iron (Fe), aluminum (Al); silicon (Si); potassium (K); nickel (Ni); vanadium (V); sulfur (S); Selenium (Se); lead (Pb); zinc (Zn); and titanium (Ti).

Short-term exposures:

In the adjusted multi-species models while adjusting PM_2.5 mass concertation, at the 95%ENT level, short-term UFPs had the largest number of significant associations with individual metabolites (62 metabolites) including 26 metabolites were significant throughout the 3 moving averages, followed by Ni (14 metabolites, all of them with the 30-day window only), V (5 metabolites), K (3 metabolites), Si and Al (2 metabolites each) (Figure 1, Supplemental Figure 1, Supplemental Tables 1 and 2). Short-term exposure to Cu, Fe, Zn, Pb, Se, Na, S were not significantly associated with any metabolites i.e., they showed a similar effect to the average PM_2.5 mass. Similarly, in the ICA, mainly with the 30-day window, short-term exposure to UFPs, Ni, and V in addition to Na were significantly associated with ICA-factor 2 (Table 3). When we conducted the metabolic pathway analysis including the significant metabolites at p-value < 0.01 from the regression models, we identified 12 metabolic pathways perturbed with short-term exposure species. Glycerophospholipid and Butanoate metabolisms were significantly perturbed with short-term exposure to UFPs and Ni, Sphingolipid metabolism with UFPs, Ni, V, and K, Glutathione metabolism with UFPs, Ni, V, and Si, Beta-alanine metabolism with UFPs, Pyrimidine metabolism with UFPs and Si, Propanoate metabolism with UFPs and K, Purine metabolism with Ni, Al, and K, Arginine biosynthesis with V, Porphyrin and chlorophyll metabolisms with Ni and V, Neomycin, kanamycin and gentamicin biosynthesis with Si and Al, and Galactose metabolism with Al (Figure 2).

Figure 1. Volcano Plots presenting the adjusted associations between exposure to short-term (30-day moving average) and long-term (annual average) of PM_2.5 species with metabolomics.

These models were linear mixed-effect regression models (LMEM) with random participant-specific intercepts and simultaneously adjusted for the 14 PM_2.5 species, and PM_2.5 mass (multi-species models) for the same exposure window.

All models were adjusted for NO₂, O₃, temperature, and relative humidity for the same exposure window, age (years), body mass index (kg/m2), cigarette pack-years, alcohol intake (< or ≥2 drinks per day), socioeconomic status (income payment and years of education), and season (warm/cold).

The transverse dashed lines represent different statistical significance levels of p-values (from lower to upper): 0.05, 0.01, ENT95%, and ENT99% calculated for the short-term and long-term exposures separately.

Note the different scale of the axes.

Abbreviations: PN: particle number (measure for ultrafine particles (UFPs)); BC: black carbon; sodium (Na); iron (Fe), aluminum (Al); silicon (Si); potassium (K); nickel (Ni); vanadium (V); sulfur (S); Selenium (Se); lead (Pb); zinc (Zn); and titanium (Ti); log10, logarithmic base 10; ENT: Effective/independent number of tests; PM_2.5, particulate matter ≤2.5 μm; NO₂, nitrogen dioxide; and O₃, ozone.

Table 3.

The adjusted associations between different exposure windows of PM_2.5 species with the factor-2 of the independent component analysis (ICA_factor2) in the multi-species models.

	24 hour- window			7-day window			30-day window			365-day window
	Beta (SE)	P_value	FDR p_value	Beta (SE)	P_value	FDR p_value	Beta (SE)	P_value	FDR p_value	Beta (SE)	P_value	FDR p_value
BC	0.194(0.16)	0.23	0.99	0.293(0.379)	0.44	0.99	0.216(0.588)	0.71	0.99	3.935(1.729)	0.02	0.17
UFPs	−3.64E-05(5.24E-06)	2.95E-10	1.24E-08	−5.27E-05(8.49E-06)	1.23E-08	5.15E-07	−7.69E-05 (1.00E-05)	5.95E-12	2.50E-10	−2.97E-05(3.32E-05)	0.37	0.99
Zn	−11.46(5.88)	0.05	0.99	−17.47(11.12)	0.12	0.99	−51.97(16.57)	0.002	0.09	37.07(58.97)	0.53	0.99
Se	134.87(62)	0.03	0.99	231.36(146.08)	0.12	0.99	−168.79(295.73)	0.57	0.99	−1392.69(1016.79)	0.17	0.99
Ni	40.74(37.99)	0.29	0.99	29.45(62.18)	0.64	0.99	370.05(72.63)	1.41E-06	5.93E-05	681.15(320.7)	0.04	0.49
Na	−0.33(0.42)	0.44	0.99	1.68(0.76)	0.03	0.99	5.52(1.35)	8.21E-05	0.003	14.08(8.35)	0.09	0.66
Si	0.15(1.81)	0.93	0.99	−2.14(3.73)	0.57	0.99	17.49(7.41)	0.02	0.84	−17.43(22.02)	0.43	0.99
S	−0.08(0.09)	0.34	0.99	0.09(0.17)	0.61	0.99	0.74(0.32)	0.02	0.75	2.01(1.39)	0.15	0.99
Al	−1.76(3.13)	0.57	0.99	6.85(8.07)	0.40	0.99	−23.4(15.84)	0.14	0.99	−14.72(41.02)	0.72	0.99
Fe	2.73(2.28)	0.23	0.99	0.54(4.77)	0.91	0.99	−2.53(6.51)	0.70	0.99	106.69(30.83)	6.94E-04	0.01
Cu	15.07(10.89)	0.17	0.99	50.64(34.61)	0.15	0.99	−11.66(73.59)	0.87	0.99	563.26(359.79)	0.12	0.84
Pb	2.04(12.82)	0.87	0.99	29.71(27.68)	0.29	0.99	−90.1(48.9)	0.07	0.99	−417.07(313.04)	0.18	0.99
K	5.14(3.01)	0.09	0.99	−1.04(1.56)	0.51	0.99	11.06(3.58)	0.003	0.11	−23.2(19.29)	0.23	0.99
V	373.07(55.84)	2.97E-10	4.16E-09	−73.03(53.62)	0.18	0.99	−298.52(71.69)	6.14E-05	0.003	−512.83(221.03)	0.02	0.19

These models were linear mixed-effect regression models (LMEM) with random participant-specific intercepts and simultaneously adjusted for the 14 PM_2.5 species, and PM_2.5 mass (multi-species models) for the same exposure window.

All models were adjusted for NO₂, O₃, temperature, and relative humidity for the same exposure window, age (years), body mass index (kg/m2), cigarette pack-years, alcohol intake (< or ≥2 drinks per day), socioeconomic status (income payment and years of education), and season (warm/cold).

Beta (SE) and p-values presented are from ICA-Factor 2 analysis.

Abbreviations: ICA, independent component analysis; FDR, Benjamini-Hochberg false discovery rate; SE, standard error; PM_2.5, particulate matter ≤2.5 micrometers; NO₂, nitrogen dioxide; and O₃, ozone; PN: particle number (measure for ultrafine particles (UFPs)); BC: black carbon; sodium (Na); iron (Fe), aluminum (Al); silicon (Si); potassium (K); nickel (Ni); vanadium (V); sulfur (S); Selenium (Se); lead (Pb); zinc (Zn); and titanium (Ti).

Figure 2. Metabolic Pathways identified as enriched among the metabolites significantly associated with (A) Short-term exposure to PM_2.5 species and (B) Long-term exposure to PM_2.5 species.

Pathway analysis conducted using Metaboanalyst³⁸ is based on both over enrichment analysis, i.e. how many significant metabolites fall within a given pathway, and pathway topology analysis, i.e. how important those metabolites are to that pathway.

The figure shows significant and noteworthy metabolic pathways based on the enrichment p-value and the impact score which is based on the cumulative importance of all the significant metabolites within the pathway.

Abbreviations: PM_2.5, particulate matter ≤2.5 μm; UFPs: ultrafine particles; BC: black carbon;; sodium (Na); iron (Fe), aluminum (Al); silicon (Si); potassium (K); nickel (Ni); vanadium (V); sulfur (S); Selenium (Se); lead (Pb); zinc (Zn); and titanium (Ti); ENT: Effective/independent number of tests; PM_2.5, particulate matter ≤2.5 μm; NO₂, nitrogen dioxide; and O₃, ozone.

Long-term exposures:

In the adjusted multi-species models while adjusting PM_2.5 mass concertation, at the 95%ENT level, long-term BC had the largest number of significant associations with individual metabolites (12 metabolites), followed by V (11 metabolites), Zn (10 metabolites), Ni (8 metabolites), Fe (5 metabolites), Cu (4 metabolites), Se and Pb (2 metabolites each), and Si and Al (1 metabolite each) (Figure 1, Supplemental Figure 1, Supplemental Tables 1 and 2). Long-term exposure to UFPs, K, Na, and S were not significantly associated with any metabolites. In the ICA, only long-term exposure to Fe was significantly associated with ICA-factor 2 (Table 3). We identified 18 metabolic pathways perturbed with long-term exposure species. Glycerophospholipid metabolism was significantly perturbed with long-term exposure to Ni, Sphingolipid and Purine metabolisms with BC and Fe, Glutathione metabolism with BC, Ni, V, Zn, and Cu, Butanoate metabolism with Cu, Arginine biosynthesis with BC, Ni, and V, Arginine, proline, Histidine, Starch and sucrose metabolisms with Ni and V, Porphyrin and chlorophyll metabolisms with BC and Ni, Thiamine metabolism with Zn, Glycosylphosphatidylinositol (GPI)-anchor biosynthesis with Ni, V, and Zn, nitrogen metabolism and D-glutamine and D-glutamate metabolism with Cu, phosphatidylinositol signaling system and glycerolipid metabolism with Se, Neomycin, kanamycin and gentamicin biosynthesis and Ubiquinone and other terpenoid-quinone biosynthesis are noteworthy with V due to their high impact score (Figure 2). Among the long-term species that were associated significantly with metabolites, Pb, Si and Al were not associated with any metabolic pathways.

There was an overlap with the identified metabolic pathways associated with the individual species and the metabolic pathways perturbed by the highest 100 metabolites of the ICA-Factor 2²¹. The pathways that were associated with the 100 highest contributing metabolites to ICA-factor 2 were purine (p-value=0.02), sphingolipid (p-value=0.03), beta-alanine (p-value=0.06), and glycerophospholipid (p-value=0.1) metabolisms. Taurine and hypotaurine metabolism was also noteworthy due to high impact score of 0.71 for the ICA-factor2. Results were similar in the single-species models, but in these analyses a larger number of metabolites were statistically significant (data not shown) as well as more statistically significant species in the ICA analysis (Supplemental Tables 3).

4. Discussion

To our knowledge, this is the first study to examine the untargeted metabolomic signature of long-term exposure to the individual species of PM_2.5 and the biggest study for the short-term exposure, in which we report the differential effects of PM_2.5 species compared to the average effect of PM_2.5. We reported multiple significant associations between several PM_2.5 species and metabolites as well as several perturbed metabolic pathways. This is also the first study to use ICA to examine metabolic signature of air pollution studies.

In this study, we observed several significant associations between short-term species of PM_2.5 (UFPs, Ni, V, K, Si, and Al), while simultaneously adjusting for PM_2.5 mass concentration. We identified 12 metabolic pathways (glycerophospholipid, butanoate, sphingolipid, glutathione, beta-alanine, pyrimidine, propanoate, purine, galactose, and porphyrin and chlorophyll metabolisms, arginine biosynthesis, neomycin, kanamycin and gentamicin biosynthesis) associated with those species.

We observed several significant associations between long-term species of PM_2.5 (BC, Ni, V, Zn, Fe, Cu, Se, Pb, Si, and Al) while adjusting for PM_2.5 mass. We identified 18 unique metabolic pathways (glycerophospholipid, glycerolipid, butanoate, sphingolipid, glutathione, purine, thiamine, and porphyrin and chlorophyll, arginine and proline, histidine, starch and sucrose, phosphatidylinositol signaling system, D-glutamine and D-glutamate, and nitrogen metabolisms, Arginine biosynthesis, neomycin, kanamycin, gentamicin biosynthesis, glycosylphosphatidylinositol (GPI)-anchor biosynthesis, and ubiquinone and other terpenoid-quinone biosynthesis) associated with those species except Pb, Si, and Al.

Collectively, UFPs and BC had the largest number of individual metabolites significantly associated with short and long-term exposure, respectively. The main perturbed metabolic pathways were related to sphingolipid, glutathione, purine, arginine, and porphyrin and chlorophyll metabolisms.

The identified pathways were mainly related to inflammation, oxidative stress, immunity, and nucleic acid damage and repair. Inflammation and oxidative stress are biological processes that have been linked to both short- and long-term exposures to PM_2.5^2–5. PM species have been reported to be associated with cardiovascular disease morbidity³⁹ and mortality⁴⁰.

Inhaled UFPs can cross the lung epithelium and have been reported to enter the bloodstream directly from the lungs⁴¹. UFPs can induce local and systemic inflammation, leading to numerous adverse health outcomes including respiratory and cardiovascular morbidity and mortality^42–44. The iron-rich nanoparticles i.e., UFPs from air pollution were recently detected in the mitochondria within the hearts of people living in polluted cities, causing cardiac stress⁴⁵. The nano-size of the UFPs (diameter <100 nm) and hence higher surface area of exposure can dictate their toxicity that could be potentially even more than the bigger particles^46,47. Laser printer-emitted nanoparticles (similar to UFPs) previously were shown to induce molecular pathways and functions e.g., glycerophospholipid, linoleic acid metabolism, biotin, and pyrimidine pathways in rats⁴⁶. Conversely, the larger particles that cannot cross the lung epithelium, can induce inflammation in the lungs triggering a systemic response observed in the peripheral blood⁴⁸. Those together along with toxicological studies may suggest different biological mechanisms for the effects of the different species of PM_2.5 ⁴⁹. For example, BC contributes in the reactive oxygen species (ROS)-mediated DNA damage and hence oxidative stress and inflammatory pathways⁵⁰. Also, transition metals like Fe and Cu are redox-active where the metal ion can serve as both electron acceptor and donor in reactions to generate radical ions. Such metals can induce oxidative stress through redox cycling or quenching antioxidant capacity⁵¹. PM derived from oil combustion e.g., V may be associated with inflammation and endothelial dysfunction as measured by intercellular adhesion molecule-1 (ICAM-1) and vascular cell adhesion molecule-1 (VCAM-1), and it is likely that oxidative stress plays a role in these associations²⁷. Residual oil ash which contains these metals has been shown to be highly toxic in animal and in vitro studies.

The PM_2.5 species may serve as tracers for the exposure source. The major source of urban UFP is combustion⁵² with motor traffic the most important source⁵³. Atmospheric UFP formation was observed both in pristine and polluted areas, making exposure to UFPs in urban outdoor air ubiquitous⁵⁴. BC is the sooty black material emitted primarily from diesel engines and wood burning among other sources⁵⁵.

Storm water runoff from roadways has been reported to contain many metals including Zn, Fe, Cu, Ni, V, Pb, and Al^56,57 and other species including K, Na, and Se⁵⁸. Hence, these are components of road dust. Important contributors to element loads on roadway surfaces include tires (Zn^59,60), brakes (Cu, Pb, Zn:^60–62), outdoor yellow paint (Pb⁵⁹), and adjacent structures (Pb, Cu, and Zn⁶⁰). Si and Al are mostly from road dust and crustal materials. V and Ni are predominantly from fuel oil combustion^27,63. Wood burning and local pollution produces K and Cu. Na is mostly from sea salt. Regional sources related to emissions from power plants and other urban areas are dominated by S and likely reflects coal combustion. Pb, Al, and Fe also originates from regional pollution^27,63.

V and Ni share a similar emissions source (fuel oil) and they shared many of the associated metabolic pathways in our study. This could indicate that they were surrogates for other components of fuel oil emissions and this unmeasured component would be the true culprit to the changes in the metabolic pathways. Alternatively, they both could truly share the metabolic pathways as reported previously²⁷.

Si and Al originate from dust^27,63 and are in oxide form with strong bonds that are very unlikely to participate in chemical reactions i.e., inert and redox-inactive and unlikely to be the main culprit for the perturbed metabolic changes. However, because they shared a metabolic pathway in this study, they both could be surrogates for other suspended particles in the dust (mostly road traffic dust) that we did not measure, which could be the culprit for the detrimental effects. These could include the metals mentioned above, latex particles from tire wear, polycyclic aromatic hydrocarbons (PAH)’s etc. On the other hand, despite being inert, they can irritate the lung, and possibly cause external impacts though perturbing the autonomic nervous system. Al still could shift biological systems into oxidative stress⁶⁴.

On the other hand, except for glutathione metabolism, there was no overlap between the metabolic pathways associated with long-term Zn and Cu. While both of them originate mostly from traffic^27,63, it is unlikely they were surrogates for other unmeasured traffic-related exposures since they had different patterns of associations. This may suggest that these pathways could be truly associated with the Zn and Cu.

Although long-term Pb exposure was associated with two metabolites, we did not observe metabolic pathways that were perturbed. In addition, we did not observe significant associations with any metabolites with short-term Pb exposure i.e., Pb had similar effect of average PM_2.5. These results were expected given the regulations on Pb levels. Nevertheless, although Pb is redox-inactive, it could have participated in the depletion of antioxidants⁵¹.

Similarly, the metals that we did not observe significant associations with, this does not mean that they do not have effect on the metabolic pathways. Instead, this means that they did not have an effect that is significantly different from that of PM_2.5 mass itself.

Glycerophospholipids are the main component of biological membranes^65,66 and sphingolipids are implicated in membrane biology⁶⁷. Air pollution can generate or act directly as ROS that induce oxidative stress⁶⁸ and cell membranes can be one of the primary targets of ROS derived from air pollutants. Oxidative stress can induce the activation of phospholipase A2 (PLA2) that hydrolyzes phospholipid from the cell membrane to generate polyunsaturated free fatty acid and lyso-phospholipid⁶⁵. This metabolism has been reported in association to short-term traffic related air pollution in pregnant women^69,70 and exposure to cigarette smoke in mice, another source of particulate exposure⁷¹. We observed perturbations in sphingolipid metabolism that is related to inflammation and immunity^72,73 consistent with previous finding with short-term air pollution⁶⁹. Sphingolipid metabolism has been linked to diabetes, Alzheimer’s disease, and hepatocellular carcinoma^74,75, heart failure⁷⁶, and cancer⁷³.

We also observed perturbations in glutathione metabolism that is involved in xenobiotic-mediated oxidative stress^77,78 and has been reported before to be associated with air pollution⁷⁰. We observed perturbations in histidine and arginine metabolisms that are amino acids and have been reported to have anti-inflammatory effects. They are related to endothelial function, inflammation, and airway hyperresponsiveness, and previously have been reported to be inversely associated with elevated level of air pollution⁷⁹. Previous studies found that histidine was negatively associated with inflammation and oxidative stress^17,80–82.

We observed perturbations in propanoate metabolism that is involved in branched-chain amino acids metabolism and the short-chain fatty acid metabolism⁸³. Propanoate metabolism is also involved in the colon microbiome and gluconeogenesis in the liver and acetate achieving the highest systemic levels in blood⁸⁴. We observed perturbations in purine, pyrimidine, thiamine, and beta-alanine metabolisms; all related to nucleic acid metabolism, damage and repair⁸⁵ consistent with reported association with air pollution^17,86. We observed perturbations in butanoate involved in immune regulation⁸⁷ and this metabolite has been reported previously to be associated with long-term air pollution⁸⁸. Those metabolic signatures could be promising biomarkers for exposure to PM_2.5 species. These results warrant more research in the promising field of human metabolomics to confirm the specific pathways and metabolites perturbed by PM_2.5 species’ exposure.

Metabolomics has been increasingly applied to investigate associations with exposure to air pollution. Untargeted and targeted metabolomics strategies are increasingly used to characterize the overall metabolic changes and relevant metabolic and toxicological pathways in animal blood and urine^89–95. Lipid, purine, and amino acid metabolisms are common pathways that are perturbed with exposure to air pollution^89–95. The detrimental effect of air pollution to the human body is closely associated with lipid metabolism, and disturbed plasma lipid profiles have been observed in several cohorts including NAS with exposure to various air pollutants^15,21,96.

Results from an untargeted metabolomic study of three cohorts in Germany, revealed that changes in lysophosphatidylcholines (LPCs) in human plasma were associated with short-term exposure to air pollutants¹⁵. This result was consistent with the findings in animal and cell models, in which LPCs increased in air pollutant-exposed rats and alveolar type II cells compared with the control group⁹⁷. The study further determined that the observed increase may have been due to air pollutant-induced activation of PLA2, that catalyzes the hydrolysis of phosphatidylcholines to LPC⁹⁷. Using untargeted metabolomics approach metabolites and pathways were reported to be associated with both air pollutants and compromised health outcomes using human serum samples⁸⁸. It was found that linoleate metabolism and glycerophospholipid metabolism were common pathways linking UFP exposure to asthma; and glycosphingolipid metabolism linking UFP exposure to cardiovascular disease⁸⁸.

There are a limited number of studies that examined the metabolomic signature of the species of PM_2.5. Ladva et al.²⁰ studied metal exposure from in-vehicle air pollution and high-resolution metabolome; Atlanta Commuters Exposure (ACE–2). PM_2.5 species (BC, particle-bound PAH and UFPs as well as 2-hr integrated elemental metal (Al, Pb, and Fe), total organic carbon (OC), and water-soluble organic carbon (WSOC) concentrations were measured for 60 participants in-vehicle with repeated biological sampling of inflammation-related endpoints. They examined associations between short-term exposures (over 10-hr duration) with high-resolution targeted metabolome²⁰. The authors reported perturbations in lipid mediators of inflammation and nucleotide driven antioxidation, namely arachidonic acid, leukotriene, and tryptophan metabolism with short-term (10 hr) exposure to particulate metal exposures (Al, Pb, and Fe). The results were supported by overlap of the pathways (leukotriene metabolism) with changes in inflammatory cytokines and acute-phase protein in the same population. On the other hand, there was no significant associations observed with other measured species including UFPs. Perturbations of the arginine metabolome was also reported following exposures to traffic-related air pollution in a subset of the same study¹⁹. Others reported metabolic perturbations associated with air pollution exposure that are related to dysregulated inflammatory and oxidative stress, including leukotriene, vitamin E, glutathione, arginine, purine, pyrimidine, urea cycle/amino group, glycerophospholipid, tryptophan metabolism^20,98–100. These altered pathways could contribute to the tissue damage and systemic inflammation observed in response to air pollutant exposure. Nevertheless, while our results are generally consistent with the previous studies, there was no perfect overlap between our results and the results for the ACE-2 study. The differences could be explained that previous studies examined the species one at a time and not simultaneously and the different time window of exposures. For example, in the ACE-2, short-term exposure was measured over hours while we assessed the short-term exposure by moving averages over 1-day, 1-week and 1-month which could truly reflect different metabolic pathways over time. Also, our exposures are more aged particles than the fresh in vehicle particles in the ACE-2 study, and the toxicological profile of the particles could change with the chemical changes that occur with particle aging.

Perturbations in plasma metabolomes as a result of ambient air pollution exposures were reported using untargeted metabolomic profiling using repeated biological measures and detailed exposure characterization, including trace metal composition, to capture the dynamic nature of high-dimensional exposure and biological responses¹⁴. In this study, 31 healthy volunteers were exposed to ambient air pollution for 5 hr. Then, they measured exposure to PM, UFPs, absorbance/BC, elemental/organic carbon, trace metals, secondary inorganic components, endotoxin content, gaseous pollutants, and PM oxidative potential. They collected blood from the participants 2-hr before and 2- and 18-hr after exposure. They conducted untargeted metabolite profiling to monitor 3,873 metabolic features in 493 blood samples from the volunteers. Several robust associations between air pollutants and metabolic features after short-term exposures (hrs) were observed. The study reported significant eight pathways, among which glycerophospholipid, and arginine and proline metabolisms. The authors pointed toward a stronger effect of PM composition to biological response than PM mass concentration itself. Most features were associated with sulfate (SO₄), which is one the major species of PM. We did not observe any significant perturbed metabolic pathways with S (SO₄ = 3*S), however, as the authors noted SO₄ might also be a surrogate of unmeasured secondary aerosol species that may be biologically active¹⁴.

Our study had a few limitations that could inform future study designs. We estimated the exposure to outdoor air pollution using a central site, which may differ from personal exposure. This potential misclassification, however, may have led to underestimation of the observed association instead of creating spurious ones^101,102. Generalizability might be limited because the participants were predominantly elderly white males who were residents in a low-polluted environment. In addition, a given species may reflect a different relative contribution of sources in one area than another (e.g., emissions from industry vs. traffic). In addition, the chemical composition of PM_2.5 from a specific source may differ between different cities (e.g., traffic source affected by distribution of vehicle and fuel types and traffic patterns). Because this is one of the first human studies to report the metabolomic signatures and the differential contribution of the PM_2.5 species and metabolic expression could be sensitive to inter-personal variability¹⁰³, our results will need to be replicated in more diverse population and higher exposure levels. Despite the lack of another population to replicate our results, 36% of the participants had repeated measures partitioning of inter- and intra-participant variability of both exposures and outcomes over time, that may have provided analytical improvement over cross-sectional analyses for causal inference. Finally, although we measured a panel of 14 PM_2.5 species, we cannot exclude the possibility of residual confounding by other unmeasured exposures.

Our study had several strengths. This is the biggest human study to date to report the untargeted metabolomic signatures of PM_2.5 species and the first for the long-term exposure. This study is based in a well-characterized longitudinal study. NAS population were geographically stable and well followed-up since they were enrolled with >80% response rates. In addition, we used ICA which is a new statistical method in the metabolic studies that ensures independence if the factors even in the non-Gaussian variables compared to the more commonly used PCA. Finally, we conducted untargeted metabolome profiling which has the potential to increase metabolome coverage (increasing from several hundred markers to several thousands) and to reduce bias towards identifying well-studied metabolites¹⁴ if compared to targeted approaches.

5. Conclusions

This is biggest human study we are aware of that comprehensively examined the metabolomic signature for exposure to PM_2.5 species for both short and long-term exposure. Using a global and untargeted approach, we have identified several significant metabolites associated with different species of PM_2.5 short- and long-term exposure. In addition, we identified the metabolic pathways associated with those changes. These findings, although they need to be replicated, increase our understanding of the biological mechanisms of the different species, thus helping research targeted to prevention and/or treatment of their adverse health effects. These findings underscore the potential of metabolomics to provide novel ways to help better understand the health effects of air pollution on a metabolic pathway level. The implicated metabolic pathways could be potentially targeted in interventional studies for therapeutic purposes.

Highlights.

1. Metabolomic profiling in plasma is a powerful tool for mechanistic understanding of the impact of PM_2.5 exposure

2. Exposure to PM_2.5 has metabolomic signatures that are related to oxidative stress, immunity, and nucleic acid damage and repair

3. Little is known about the specific PM_2.5 species (and hence sources) that drive these metabolomic signatures

4. Particulate number (i.e., ultrafine particles) was the main significant species of short-term PM_2.5 exposure effects followed by nickel and vanadium then silicon, aluminum, and potassium

5. Black carbon, nickel, vanadium, zinc, copper, iron, and selenium were the main significant species of long-term PM_2.5 exposure effects