Length of PM2.5 exposure and alterations in the serum metabolome among women undergoing infertility treatment : Environmental Epidemiology

Secondary Logo

Journal Logo

Original Research Article

Length of PM2.5 exposure and alterations in the serum metabolome among women undergoing infertility treatment

Hood, Robert B.a,*; Liang, Donghaib; Tang, Ziyinb; Kloog, Itaic; Schwartz, Joeld,e,f; Laden, Francined,e,f; Jones, Deang; Gaskins, Audrey J.a

Author Information
Environmental Epidemiology 6(1):p e191, February 2022. | DOI: 10.1097/EE9.0000000000000191

Abstract

What this study adds

Both short- and long-term exposure to PM2.5 has been related to adverse health outcomes. However, the biological pathways underlying these health effects are largely unknown. We identified several unique serum metabolomic pathways associated with acute and chronic PM2.5 exposure. Major pathways associated with acute PM2.5 exposure included amino acid, energy, and lipid metabolism. Major pathways associated with chronic PM2.5 exposure included pro-inflammatory and anti-inflammatory pathways. Seven unique metabolites were identified with level-1 evidence.

Introduction

Fine particulate matter (PM2.5) air pollution is a complex mixture of liquid and solid particulates with a diameter of 2.5 micrometers or less and can penetrate deeply into the respiratory tract. PM2.5 has been linked to a myriad of negative health outcomes in humans including but not limited to respiratory diseases, cardiovascular diseases, neurological and mental health issues, adverse reproductive and pregnancy outcomes, and mortality.1–6 To better understand the underlying pathways between PM2.5 and these health outcomes, several studies have used both targeted and untargeted methods to investigate potential alterations in the human metabolome.7–17 Metabolomics is a relatively new field that focuses on global detection and relative quantification of small molecules, both endogenous and exogenous, in human tissues and fluids and evaluates how changes in these molecules are related to changes in exposures and disease states. Using metabolomics, several studies have observed alterations in pro-inflammatory and oxidative stress pathways when people are exposed to PM2.5.7,9,10,12–14,17,18 While they have yielded important findings, these studies are limited by the use of a singular time window for exposure to PM2.5.

There are several downsides to focusing solely on one time window of PM2.5 exposure. Chiefly, several studies examining the health effects of PM2.5 exposure have demonstrated that both short- and long-term exposure to PM2.5 are important but long-term exposure may elicit a greater or different response than short-term exposure. In studies of Medicare patients in New England, both acute (1–2 days) and chronic (1–7 years) measures of PM2.5 exposure were associated with increased hospital admissions and mortality but the effect estimates for chronic PM2.5 exposure were of greater magnitude than the acute exposures.19,20 Recently, the same logic has also held true in studies examining adverse pregnancy outcomes, specifically preterm birth. Among a cohort of pregnancies in China, daily exposure to PM2.5 in the one to six days prior to delivery and chronic exposure to PM2.5 throughout pregnancy were both associated with an increased risk of preterm birth; however, the magnitude of association gradually increased as the moving average of PM2.5 exposure expanded from one to 37 weeks prior to birth and the greatest effect estimate was observed for chronic exposure during the entire pregnancy.21 While both acute and chronic exposures may influence risk of preterm birth, the underlying biological mechanisms may differ. Several pathways have been proposed to underlie the association between both acute and chronic PM2.5 exposure and preterm birth including heightened oxidative stress, inflammation, and endocrine disruption.22

To our knowledge, there has only been one previous study that examined differences in the human metabolome during different windows of PM2.5 exposure. Among 197 Belgian mother-infant pairs, Martens et al. used targeted metabolomics to measure 37 oxylipins in neonatal cord blood plasma samples and related these to in utero PM2.5 exposures. Alterations in metabolites derived from the lipoxygenase pathway were only observed when examining total PM2.5 exposure during pregnancy or second-trimester PM2.5 exposure (but not first or third-trimester exposures).23 Martens et al., hypothesized that this difference may be due in part to the thinning barrier between the maternal and fetal blood supplies with increasing gestational age and with the increasing fetal capillaries size until week ten of gestation.23 These results support the hypothesis that the timing and duration of PM2.5 exposures are important to consider, particularly for outcomes that may have critical windows of susceptibility and both short- and long-term exposure-response relationships.

To expand on the limited literature, our study sought to investigate the similarities and differences in how varying durations of PM2.5 exposure may alter the serum metabolome. Specifically, we explored three acute, one intermediate, and one longer-term time window of PM2.5 exposure and their association with metabolic features and metabolomic pathways identified using untargeted metabolomics. Untargeted metabolomics allows for a greater examination of the metabolome rather than targeting a single pathway or class of metabolites. Understanding the changes in the metabolome across exposure windows may offer novel insight into how acute and chronic exposure to PM2.5 may lead to different disease states in humans and could lead to biomarkers for specific durations of exposure.

Materials and methods

Study population

The women included in our analysis were participants in the Environment and Reproductive Health (EARTH) study.24 Briefly, the EARTH study was a prospective cohort that enrolled couples seeking infertility evaluation and treatment at the Massachusetts General Hospital (MGH) Fertility Center. The goal of the study was to evaluate how environmental and dietary factors influence fertility. Upon enrollment, women completed questionnaires on demographics, medical history, environmental exposures, diet, lifestyle, and reproductive health. Participants’ height and weight were also measured via study staff to calculate body mass index (BMI; kg/m2). Women provided their residential address, initially for reimbursement purposes, but later these were used for geocoding and linking to environmental exposure data. The EARTH study was approved by the Human Studies Institutional Review Boards of MGH and the Harvard T.H. Chan School of Public Health (IRB No. 1999P008167). All study participants signed an informed consent after the study procedures were explained by research study staff.

In 2019, we randomly selected 200 women using a random number generator (from the 345 women with complete air pollution data who underwent a fresh, autologous assisted reproductive technology (ART) cycle between 2005 and 2015)25 for inclusion in a metabolomics sub-study. All of these women provided a non-fasting blood sample during controlled ovulation stimulation, between 2005 and 2015, which was used for metabolomic profiling. The blood samples were collected via venipuncture during a routine morning appointment (between 7 am and 10 am). Approximately 6-ml of blood was collected from each participant. Serum was centrifuged, aliquoted, and stored at −20°C initially before being transferred to Harvard for storage at −80°C.

Air pollution measures

We estimated individual ambient PM2.5 exposure by linking women’s geocoded residential address at enrollment to a spatiotemporal model of PM2.5 exposure at a 1 km2 spatial resolution.26 The validated hybrid model of ground-level PM2.5 concentrations used satellite-derived aerosol optical depth measurements, land use (e.g., measures of population density, elevation, traffic, percentages of land use, normalized difference vegetation index (NDVI), and point and source pollutant emissions), meteorological (e.g., air temperature, wind speed, daily visibility, sea-land pressure, and relative humidity) variables, and temporally resolved data on planetary boundary layer to estimate exposure.26 All data used for the PM2.5 model were publicly available and obtained from a variety of sources including satellites (aerosol optical depth data), the US Environmental Protection Agency (EPA) (monitoring data), the US Geological Survey National land Cover dataset (spatial data), the National Climatic Data Center (meteorological data), Moderate Resolution Imaging Spectroradiometer (MODIS) satellite NDVI (NDVI data), and the National Oceanic and Atmospheric Administration (planetary boundary layer). We derived daily estimated ambient PM2.5 concentrations starting three months prior to the date of blood collection. Air pollution exposures per day were averaged across the following windows: one day, two days, three days, two weeks, and three months prior to blood collection to examine short-term (one-three days), intermediate (two weeks), and longer-term (three months) exposures to PM2.5.

High-resolution metabolomics

Using established standard protocols,11–13 samples were treated with two volumes of acetonitrile and were centrifuged. Samples were analyzed in triplicate. Prepared samples were analyzed via liquid chromatography with high-resolution mass spectrometry (LC-HRMS) (Dionex Ultimate 3000 RSLCnano; Thermo Orbitrap Fusion; Thermo Fisher Scientific, Waltham, MA). We utilized two column types, C18 hydrophobic reversed-phase chromatography (C18 Neg) with negative electrospray ionization (ESI) and hydrophilic interaction liquid chromatography (HILIC) with positive ESI. In the C18 column, analyte separation was achieved using water, acetonitrile, and 10 mM ammonium acetate during the mobile phase with the following gradient elution: initial one minute period, 60% water, 35% acetonitrile, and 5% ammonium acetate, followed by a linear increase to 0% water, 95% acetonitrile, and 5% ammonium acetate at three minutes and held for the remaining two minutes. In the HILIC column, analyte separation was achieved using water, acetonitrile, and 2% formic acid during the mobile phase with the following gradient elution: initial one-and-a-half-minute period, 22.5% water, 75% acetonitrile, and 2.5% formic acid, followed by a linear increase to 75% water, 22.5% acetonitrile and 2.5% formic acid at 4 minutes and a final hold of 1 minute. The mobile phase flow rate was 0.35 mL/min for the first minute and was increased to 0.4 mL/min for the final four minutes for both columns. In the C18 column, the gradient elution started at 60% aqueous condition could miss metabolites separated between 100% and 60% aqueous. However, the HILIC column is generally better for the detection of these metabolites. We applied to columns to maximize metabolomic coverage.27–29 The sheath gas and auxiliary gas were set at 30 (arbitrary units) and 5 (arbitrary units) for the negative ESI, respectively. For the positive ESI, the sheath gas and the auxiliary gas were set at 45 (arbitrary units) and 25 (arbitrary units), respectively. The spray voltage was −3.0 kV for the negative ESI and 3.5 kV for the positive ESI. To ensure quality control and standardization, two controlled pooled reference plasma samples, NIST 195030 and pooled human plasma (Equitech Bio, Kerrville, TX), were included at the beginning and end of each batch. Using ProteoWizard, raw data were converted to.mzML files.31 Files were further abstracted using R package apLCMS modified by xMSanalyzer.32,33 We defined unique features (detected signals) using mass-to-charge ratio (m/z), retention time, and ion intensity. Features are unique metabolomic signals that have been detected but have not been identified by their chemical name. Features detected in less than 10% of samples were removed. Additionally, serum samples with a median coefficient of variation (CV) >30% and a Pearson correlation <0.7 among the technical replicates were not included in the analysis. We excluded these features because they had a low reproducibility across the replicates. Average intensity of the remaining features was log-transformed to allow for further analysis.

Statistical analysis

We followed the standard workflow for an untargeted metabolomics study (Supplemental Figure 1; https://links.lww.com/EE/A174). We used generalized linear models to evaluate the association between each metabolomic feature and PM2.5 exposure window. Models were fit using the following equation:

Yji=α+β1jPM2.5ik+β2jTempik+β3jAgei+β4jBMIi+β5jEducationi+β6jSmokingi+eij

In these models, Yji was the natural log of the intensity for feature j and participant i. PM2.5ik was woman i’s exposure to PM2.5 averaged over exposure window k. Similarly, Tempik was woman i’s exposure to ambient temperature over exposure window k. Daily ambient temperatures were derived from the Parameter-elevation Regressions on Independent Slopes Model (PRISM)34 and were averaged over the same windows as PM2.5 exposure. Finally, these models also included the woman’s age (Agei), body mass index (BMIi), education (Educationi), and smoking status (Smokingi). The (summand) Eij denotes the residual normal error. Covariates were selected based on a priori knowledge and biological relevance. We included ambient temperature and not season because these two variables were correlated and given changes in climate and weather, ambient temperature may be a better measure of time spent indoors and is more directly linked to fuel usage (for heating and cooling). Separate models were used for the HILIC [positive] and C18 [negative] columns. We identified significant features at increasingly stringent levels of statistical significance (P value: <0.05, <0.005, and <0.0005) which allowed us to select the most stringent significance level with interpretable results. Given the high number of statistical tests, we also corrected these raw p-values for multiple comparisons using the Benjamini–Hochberg false discovery rate (FDR) procedure at two thresholds (q-value: <0.20 and <0.05). In most cases, features with P < 0.005 were used. Analyses were conducted in R (v. 4.0.3, R Foundation for Statistical Computing, Vienna, Austria).

Metabolic pathway enrichment analysis and metabolite annotation

Pathway analysis was completed using Mummichog (v. 1.0.10) in Python (Python Software Foundation, Wilmington, DE) which has been described and validated elsewhere.35 Briefly Mummichog is an innovative bioinformatic tool that computes biological pathways from a feature list using m/z and retention time without prior metabolite identification. Mummichog computes an adjusted P value for each pathway by resampling the reference input file using a gamma distribution.35 We utilized a reference file for each technical column (C18 [negative] and HILIC [positive]) with the file consisting of features with a raw P value <0.005. Features with a raw P value <0.0005 and corrected q-values could not be used due to the lack of significant metabolites in the shorter-term exposure windows. We examined pathways with P values <0.05 in any of the five exposure windows and compared the significance and number of matched metabolites in each pathway. In this analysis, pathways could have the same number of overlapping features but different pathway P value because of the different number of significant underlying features in the reference files (e.g., C18 [Negative] 1-day: six significant features versus 3-months: 36 significant features). Heat maps were used to visually compare the pathways across each time window. A P value <0.05 was utilized for the pathway analysis since Mummichog computes an adjusted p-value and we utilized a stringent criterion for significant features (P < 0.005) so there was limited need to be more conservative than traditional statistical norms.

For metabolite confirmation, we selected the features that were significantly associated with any of the PM2.5 exposure windows (P < 0.005). We examined extracted ion chromatography for retention time, isotope patterns, and peak quality. Significant features with high-quality peaks were then compared to authentic standards from our laboratory that were analyzed with the same methods (level-1 evidence).36 Significant features were matched to authentic standards by comparing the m/z, retention time, and ion dissociation. For each identified metabolite, we used the Human Metabolome Database to determine their chemical superclass and class.

Results

Sample characteristics

The average age of women in our study was 34.8 years (standard deviation [SD]: 3.9) and the majority were white (86%; n = 171) (Supplemental Table 1; https://links.lww.com/EE/A174). Ninety-two percent had at least a college degree (n = 183) and 40% of the participants had an unexplained initial infertility diagnosis (n = 79). Demographic and clinical characteristics were similar between all eligible participants and those included in the metabolomics sub-study (Supplemental Table 1; https://links.lww.com/EE/A174).

The average 1-day PM2.5 exposure was 8.7 µg/m3 (SD: 4.0) while the average 3-month PM2.5 exposure was 9.0 µg/m3 (SD: 1.9) (Supplemental Table 2; https://links.lww.com/EE/A174). The correlation between 1-day and 2-day PM2.5 exposures was 0.89 while the correlation between 1-day and 3-month PM2.5 exposures was 0.26. Similar trends were observed across the exposure windows with windows closer together in time having higher correlations compared to windows further apart.

Significant features (P < 0.005)

We detected 10,803 and 12,968 unique features using the C18 [negative] and HILIC [positive] columns respectively (Table 1). Using the 1-day exposure window, 28 and 68 features were significantly associated (P value <0.005) with the 1-day exposure window using the C18 [negative] and HILIC [positive] columns, respectively. In contrast, 136 and 267 features were significantly associated (P value <0.005) with the 3-month exposure window using the C18 [negative] and HILIC [positive] columns, respectively. Additionally, when using the corrected q-values (<0.05), no features were significantly associated with the 1-day exposure window but 21 and 83 features were significantly associated with the 3-month exposure window in the C18 [negative] and HILIC [positive] columns, respectively. In general, as the exposure window lengthened the number of significant features increased and this trend held across the various levels of statistical significance.

Table 1. - Number of significant metabolomic features associated with different PM2.5 exposure windows among 200 women in the EARTH study from 2005 to 2015 in the Northeast United States.
Exposure Window C18 [Negative] (n = 10,803) HILIC [Positive] (n = 12,968)
Raw P values Corrected Q values Raw P values Corrected Q values
<0.005 <0.20 <0.05 <0.005 <0.20 <0.05
1 day 28 0 0 68 0 0
2 days 56 0 0 74 2 0
3 days 74 0 0 100 2 1
2 weeks 85 14 5 163 36 12
3 months 136 37 21 267 209 83

In total 267 and 484 unique features were significantly associated with at least one of the exposure windows in the C18 [negative] and HILIC [positive] columns, respectively (Figure 1). In the C18 column, the largest overlap of significant features occurred between the 2-week and 3-month exposure windows (n = 31) and the 2-day and 3-day exposure windows (n = 24). In the HILIC column, the largest overlaps again occurred between the 2-week and 3-month exposure windows (n = 46) and the 2-day and 3-day exposure windows (n = 26). Only four significant features were associated with all five exposure windows and all of these were detected in the HILIC [positive] column.

F1
Figure 1.:
Number of significant features associated with each PM2.5 exposure window among 200 women in the EARTH study from 2005 to 2015 in the Northeast United States using the C18 [Negative] and HILIC [Positive] technical columns.

Metabolic pathways

Using the C18 [negative] significant features (P < 0.005), 26 significant pathways were identified that were associated with one or more PM2.5 exposure windows. On the metabolite level, amino acids and inflammatory pathways had the most features identified using Mummichog across the exposure windows (Figure 2). In some instances, features identified by Mummichog were found in several pathways and this occurred in the 2-day, 3-day, 2-week, and 3-month exposure windows. Nine of the 26 pathways - D4&E4-neuroprostanes formation, hexose phosphorylation, nitrogen metabolism, parathio degradation, phosphatidylinositol phosphate metabolism, putative anti-inflammatory metabolites formed from eicosapentaenoic acid, tryptophan metabolism, valine, leucine, and isoleucine degradation, and xenobiotics metabolism were only associated with an acute exposure window (1–3 days prior to blood sample) but not the intermediate- or long-term exposure window (Table 2). An additional nine pathways including amino sugars metabolism, ascorbate and aldarate metabolism, beta-alanine metabolism, CoA catabolism, electron transport chain, glutamate metabolism glutathione metabolism, glycine, serine, alanine, and threonine metabolism, and histidine metabolism were only associated with the intermediate or long-term exposure windows but not the acute exposure windows. Four pathways were commonly altered across all or most (four out of five) exposure windows including arachidonic acid metabolism, arginine and proline metabolism, aspartate and asparagine metabolism, and leukotriene metabolism.

T2
Table 2.:
Pathways (Features with P values <0.005) associated with different exposure windows of PM2.5 among 200 women in the EARTH study from 2005 to 2015 in the Northeast United States using the C18 [Negative] technical column.
F2
Figure 2.:
Number of features linked to pathways using Mummichog and classification of pathways modified by PM2.5 Exposure in the C18 [Negative] Column among 200 women in the EARTH study from 2005 to 2015 in the Northeast United States. Each pie chart represents a single exposure window with the total number of features that matched to known metabolites in pathways identified using Mummichog. The colors for the pie charts represent the type of pathway a feature was found to be a part of, with some features being identified as a metabolite present in several pathways (denoted by the black color). Because of this, the total number of features will not add to the total number of matched features in Table 2. The numbers in the pie chart denote the number of features found in each type of pathway.

Using the HILIC [positive] significant features (P value <0.005), 20 pathways were significantly associated with one or more PM2.5 exposure windows. In contrast to the findings from the C18 [negative] column, the categories of the metabolomic pathways related to the acute and long-term exposure to PM2.5 in the HILIC [positive] column were strikingly different (Figure 3). Across the five exposure windows, lipid metabolism pathways generally had the highest number of features identified using Mummichog. In contrast, features involved in inflammatory pathways were most prominent in the 2-week and 3-month exposure windows. Features involved with amino acid metabolism pathways were uniquely associated with acute exposures. Generally, fewer features identified by Mummichog occurred across pathways. Nine pathways—carnitine shuttle, de novo fatty acid biosynthesis, di-unsaturated fatty acid beta-oxidation, fatty acid activation, fatty acid metabolism, histidine metabolism, mono-unsaturated fatty acid beta-oxidation, tryptophan metabolism, and vitamin E metabolism—were associated with at least one of the acute exposure windows but not the intermediate or long-term exposure windows (Table 3). Six pathways—arachidonic acid metabolism, leukotriene metabolism, nucleotide sugar metabolism, prostaglandin formation from arachidonate, putative anti-inflammatory metabolites—formed from eicosapentaenoic acid, and vitamin A (retinol) metabolism were associated with intermediate and long-term exposure windows but not the acute exposure windows. Three pathways, D4&E4-neuroprostanes formation, linoleate metabolism, and omega-3 fatty acid metabolism, were associated across all or most (four out of five) exposure windows in the HILIC [positive] column. Across both technical columns, tryptophan metabolism pathways were consistently associated with acute exposure to PM2.5.

T3
Table 3.:
Pathways (features with P values <0.005) associated with different exposure windows of PM2.5 among 200 women in the EARTH study from 2005 to 2015 in the Northeast United States using the HILIC [Positive] technical column.
F3
Figure 3.:
Number of features linked to pathways using Mummichog and classification of pathways modified by PM2.5 Exposure in the HILIC [Positive] Column among 200 women in the EARTH study from 2005 to 2015 in the Northeast United States. Each pie chart represents a single exposure window with the total number of features that matched to known metabolites in pathways identified using Mummichog. The colors for the pie charts represent the type of pathway a feature was found to be a part of, with some features being identified as a metabolite present in several pathways (denoted by the black color). Because of this, the total number of features will not add to the total number of matched features in Table 3. The numbers in the pie chart denote the number of features found in each type of pathway.

Metabolite identification

Using level-1 evidence, we identified seven unique metabolites that were significantly (P value <0.005) associated with various exposure windows. In the C18 [negative] column, one metabolite was associated with only the short-term exposure windows (glutamic acid) (Table 4). Glutamic acid was only associated with the 3-day exposure window. One metabolite, in the C18 [negative] column, was only associated with the intermediate- and long-term exposure (hypoxanthine). Three metabolites in the C18 [negative] column were associated with both the short- (e.g., 3-day) and intermediate- (e.g., 2-week) -term exposure windows (N-Acetyl-serine, N-Methyl-aspartic acid, and O-Acetyl-serine).

T4
Table 4.:
Significant (P < 0.005) metabolites identified using level 1 evidence and associated with different exposure windows of PM2.5 among 200 women in the EARTH study from 2005 to 2015 in the Northeast United States using the C18 [Negative] and HILIC [Positive] technical columns.

In the HILIC [positive] column, two identified metabolites, Bis(2-Ethylhexyl)Phthalate (DEHP) and Retinoic acid, were associated with both the intermediate- and long-term exposure window. Across the seven unique metabolites we identified, the most common superclass was organic acids and derivates (n = 4; 57.1%) (Supplemental Table 3; https://links.lww.com/EE/A174).

Discussion

Key findings

In this metabolomics study among women undergoing infertility treatment, shorter- versus longer-term PM2.5 exposure windows were largely associated with unique alterations in specific metabolites and metabolomic pathways while fewer pathways were common across all exposure windows. We identified 17 pathways solely associated with acute exposure to PM2.5 and 15 pathways solely associated with the chronic exposure to PM2.5. Only seven pathways were found to be commonly altered across the majority (four out of five) of exposure time windows. Furthermore, we were able to identify seven unique metabolites associated with PM2.5 exposures of varying duration, using level-1 evidence, several of which were involved in the observed pathways.

We identified 17 pathways (eight in C18 [negative], eight in HILIC [positive], and one overlapping pathway) associated with acute (e.g., 1-3 day) exposure to PM2.5, including many metabolites involved in amino acid metabolism, lipid metabolism, energy and nutrient metabolism, and free radical formation. Several studies have also identified many of these same pathways when studying acute exposure to air pollution.9,11,13,15 For amino acid metabolism, several studies have observed an association between acute PM2.5 exposure and tryptophan metabolism9,15 which we also observed. In addition to tryptophan, we also found alterations with histidine metabolism, and valine, leucine, and isoleucine degradation. Under normal circumstances, these amino acids and their metabolism are involved in numerous responses including immune response, cell signaling, and hormone formation.37–40 However, some of these amino acids have both antioxidant and pro-inflammatory metabolites and depending on which metabolites are upregulated, there could be serious consequences for the human body.41,42 Thus, acute PM2.5 exposure is concerning due to the potential damage from oxidative stress through these pathways. In addition to amino acid metabolism, six lipid metabolism pathways were associated with acute exposure to PM2.5. Of these six lipid metabolism pathways, only three, carnitine shuttle, de novo fatty acid biosynthesis, and fatty acid activation, have previously been linked to acute exposure to PM2.5.11 These pathways may be activated after acute exposure to PM2.5 as a means for the body to expend energy to repair itself from oxidative stress induced by short-term PM2.5exposure. Finally, several anti-inflammatory pathways were associated with acute exposure to PM2.5 including vitamin E metabolism and putative anti-inflammatory metabolites formed from eicosapentaenoic acid. Activation of these pathways is likely the body’s immediate defensive response to short-term PM2.5 exposure-creating antioxidants that will help the body combat an increase in oxidative stress.

We also observed 15 pathways (nine in C18 [negative] and six in HILIC [positive]) associated with intermediate- or longer-term exposure to PM2.5, which in our study was defined as average exposure over the past 2-weeks or 3-months. These 15 pathways included several pro-inflammatory pathways, energy pathways, and anti-inflammatory pathways. Among the pro-inflammatory pathways, both leukotriene metabolism and prostaglandin formation from arachidonic acid have been previously observed in relation to long-term exposure to air pollution.16,17 Interestingly, we also observed a relationship between long-term exposure to PM2.5 and arachidonic acid metabolism. Previously, this pathway has only been associated with short-term exposures to air pollution.9 These three pathways together indicate a large and likely sustained pro-inflammatory response with chronic PM2.5 exposure. Unlike previous long-term exposure window studies, we observed a relationship between vitamin A metabolism and putative anti-inflammatory metabolites from eicosapentaenoic acid. Interestingly, the putative anti-inflammatory pathway was associated with the acute exposure window in the C18 [negative] column whereas it was associated with the long-term exposure in the HILIC [positive] column. We also observed an association between long-term PM2.5 exposure and ascorbate and aldarate metabolism, another anti-inflammatory pathway, that has been previously identified by others.16 These three anti-inflammatory pathways taken together may be the body’s attempt to compensate for the sustained inflammatory response from the upregulated pro-inflammatory pathways in an attempt to maintain homeostasis. We again observed a relationship between PM2.5 exposure and energy metabolism pathways as well as amino acid metabolism pathways. Both the electron transport chain and the nucleotide sugar metabolism pathways were associated with long-term exposure to PM2.5; however, neither of these pathways have previously been associated with long-term PM2.5 exposure. The increased need for energy may be due in part to fuel cellular efforts to repair damages from oxidative stress. With regards to amino acid metabolism, we observed three pathways associated with long-term exposure, beta-alanine metabolism, glycine, serine, alanine, and threonine- and histidine metabolism. All three of these amino acid pathways have previously been associated with long-term exposure to PM2.5.16,17 Additionally, histidine metabolism was associated with long-term exposure to PM2.5 in the C18 [negative] column but was associated with acute exposure to PM2.5 in the HILIC [positive] column.

Across the five PM2.5 exposure windows, we observed seven pathways (four in C18 [negative] and three in HILIC [positive]) that were associated with at least four of the five exposure windows. Similar to the pathways associated with only the acute or long-term exposure, we again observed associations with anti-inflammatory pathways and lipid metabolism. Interestingly, studies of both acute-12,15 and long-term16 exposure to PM2.5 have observed alterations in the arginine and proline metabolism and aspartate and asparagine metabolism which adds credence to our finding of these pathways being associated with PM2.5 exposures of varying duration. We observed two lipid metabolism pathways, omega-3 fatty acid metabolism and linoleate metabolism associated across several windows of PM2.5 exposure. Thus far omega-3 fatty acid metabolism has only been associated with acute exposure to PM2.512 while linoleate metabolism has only been associated with long-term exposure to PM2.5.16,17,43 Lastly, we observed several inflammatory pathways that were associated with at least four of the five exposure windows including arachidonic acid metabolism, D4&E4-neuroprostanes formation, leukotriene metabolism, and prostaglandin formation from arachidonate. Three of these pathways, arachidonic acid, leukotriene, and prostaglandin formation, were only associated with the long-term exposure window in the HILIC [positive] column but in the C18 [negative] column these three pathways were commonly dysregulated across several PM2.5 exposure windows. In contrast, the D4&E4 pathway was associated with acute PM2.5 exposure in C18 [negative] column but was associated with several PM2.5 windows in the HILIC [positive] column.

Overall, we were able to identify seven unique metabolites using level-1 evidence. Similar to the pathway analysis, we observed differences in metabolites by PM2.5 duration of exposure. In the short-term windows, we identified one metabolite, glutamic acid. Glutamic acid is involved in several metabolomic pathways that were commonly altered with short-term PM2.5 exposure including arachidonic acid metabolism, arginine and proline metabolism, aspartate metabolism, and the urea cycle. We identified three metabolites, DEHP, retinoic acid, and hypoxanthine associated with both the intermediate- and long-term PM2.5 exposure windows. Retinoic acid is a part of vitamin A metabolism which is a pathway we observed being associated with long-term exposure to PM2.5. There were three metabolites commonly associated with 3-day and 2-week average exposure to PM2.5, N-Acetyl-serine, N-Methyl-aspartic acid, and O-Acetyl-Serine. N-Acetyl-serine, N-Methyl-aspartic acid, and O-Acetyl-serine are all types of amino acids derivatives and offer credence to our finding of amino acid pathways related to PM2.5 exposure. N-Methyl-aspartic acid is needed for normal synaptic transmission and plasticity but when overstimulated can be excitotoxic.44 The degree to which these metabolites can be used as biomarkers of short and long-term exposure to PM2.5 warrants further study.

Clinical and policy implications

We observed that exposure to PM2.5 of varying durations (from days to several months) led to different alterations in the serum metabolome of reproductive-aged women. The different alterations in the serum metabolome may explain the different health effects that have been observed when comparing acute versus long-term exposure to PM2.5. Our results may be particularly relevant for perinatal studies focused on pregnancy loss or pre-term birth where air pollution has been shown to have both acute and long-term adverse impacts and the biological mechanisms are largely unknown.6,21,45–47 Until further evidence is available, our results support the hypothesis that air pollution largely acts on different biological pathways when encountered acutely versus chronically and this may have important implications for future studies when determining the most biologically relevant time window to focus on. Additionally, to identify sensitive biomarkers of air pollution exposure, metabolomics can be a powerful hypothesis-generating tool. In this analysis, we highlight the specific metabolic features and pathways that are linked to short-term or long-term, or both, exposure windows, which can contribute to follow-up biomarker development studies.

Strengths and Limitations

Our study has several important limitations. First, this study utilized data collected from women residing in the Northeastern US who were undergoing infertility treatment which potentially limits the generalizability of our findings. The majority of our women were also white and of high socioeconomic status, which is typical of studies focusing on infertility clinic populations, but may limit the applicability of our findings to other race/ethnicities and socioeconomic status. Nevertheless, the results were consistent with many existing air pollution and metabolomic applications conducted in population-based and highly selected populations. Second, our exposure measure only captured ambient exposure to PM2.5 and did not capture indoor air pollution and occupational exposure to PM2.5. Because we were unable to include these exposures in our measure, women’s personal exposure to PM2.5 is likely misclassified. However, we have no reason to believe that this misclassification would be differential, thus the likely consequence is that our results are biased towards the null. In addition, PM2.5 exposure in this population is generally low in comparison to other regions of the world and therefore may not be generalizable to reproductive-aged women who live in areas with high exposure to PM2.5. Third, we were unable to separate the effect of the length of the time window and the effect of acute versus chronic PM2.5 exposure on the serum metabolome. The shorter time windows (1-, 2- and 3-day) could be subject to greater noise and variation when compared to the longer time windows (2-week and 3-month). In our results, we observed that as the length of time window declined, the number of significant features also decreased. Future studies using personal monitors would be the ideal way to address this limitation; however, conducting a study like this in a large representative sample remains expensive and challenging. Fourth, we utilized average PM2.5 exposure windows which could mask potentially important temporal variations. For example, in a 3-month exposure window, PM2.5 could rapidly rise and fall and this would be recorded as the same average as a 3-month exposure window that had a steady amount of PM2.5. By not accounting for these temporal variations, we could have missed out on identifying important effects on the serum metabolome. While we examined a range of acute and chronic exposure windows that were defined a priori, there could be other critical time windows of exposure that were not investigated in our study. Future studies should consider the advantages and disadvantages of using averaged air pollution exposure windows versus other methods that may account for temporal variation in air pollution exposure and select critical time windows using a more data-driven approach. Fifth, because Mummichog relies on the number of significant features to determine P values for each pathway, it is possible that some pathways with the same number of overlapping features were significant in one time window and not in another (for example, phosphatidylinositol phosphate metabolism in the C18 [neg] column). Additionally, because we are testing multiple pathways, it is possible that some of the metabolomic pathways were associated with time windows by chance. Because of these concerns, our pathway analysis results should be interpreted with caution and will need to be confirmed by additional studies. Sixth, because we utilized non-fasting blood samples, dietary factors could have influenced our results. However, it is unlikely that diet and PM2.5 exposure were related, which means that diet is unlikely to be a confounder. Additionally, we utilized a comprehensive metabolomic workflow that has been successful in analyzing non-fasting samples. In addition, we observed similar metabolomic alterations to other air pollution studies using fasting blood samples12 which may indicate that diet had a minimal impact on our results. Future studies should consider the potential difference in results that non-fasting and fasting blood samples could provide with regard to metabolomic analyses. Next, we attempted to adjust for the false positive rate. However, due to a lack of interpretable data for the short-term exposures, we were unable to use the most stringent, FDR corrected q-values. Therefore, our results should be carefully interpreted and will need to be confirmed by additional studies with larger samples sizes. Finally, the samples used in this analysis were stored for a long period of time at −80oC prior to analysis, which could have negatively impacted the quality of the sample. However, a review of studies investigating pre-analytic factors, found samples under this condition did not have any significant negative impacts on quality after 30 months of storage48; longer storage times have not been investigated so the impact on sample quality remains a question and should be investigated in future studies. Our study does have several strengths. We utilized a validated measure of ambient exposure to PM2.5 and used a standard protocol for metabolomics analysis including laboratory standards to confirm metabolite identification with level-1 evidence using these same protocols. Lastly, due to the prospective nature of the EARTH Study, we were able to adjust for several potential confounders including age, smoking status, education, and BMI.

Conclusion

In our study of reproductive-aged women, we found that short versus long-term exposure to ambient PM2.5 had differential impacts on the serum metabolome as many specific metabolites and metabolic pathways were only associated in the acute window or the long-term window, with very few being commonly altered across all time windows examined. Differences in pathways activated by PM2.5 exposure windows may explain how differences in health outcomes arise depending on the exposure windows utilized. Researchers should be aware that PM2.5 exposures of differing duration may lead to different biological responses in the human metabolome and should take this into consideration when planning and studying the health effects of PM2.5.

ACKNOWLEDGMENTS

We would like to thank all members of the EARTH study team, specifically our research nurse Jennifer B. Ford, senior research staff Ramace Dadd, the physicians and staff at Massachusetts General Hospital Fertility Center, and all the EARTH study participants.

References

1. Arias-Pérez RD, Taborda NA, Gómez DM, Narvaez JF, Porras J, Hernandez JC. Inflammatory effects of particulate matter air pollution. Environ Sci Pollut Res Int. 2020;27:42390–42404.
2. Fiordelisi A, Piscitelli P, Trimarco B, Coscioni E, Iaccarino G, Sorriento D. The mechanisms of air pollution and particulate matter in cardiovascular diseases. Heart Fail Rev. 2017;22:337–347.
3. Yang Y, Ruan Z, Wang X, et al. Short-term and long-term exposures to fine particulate matter constituents and health: a systematic review and meta-analysis. Environ Pollut. 2019;247:874–882.
4. Brunekreef B, Holgate ST. Air pollution and health. Lancet. 2002;360:1233–1242.
5. Yuan L, Zhang Y, Gao Y, Tian Y. Maternal fine particulate matter (PM2.5) exposure and adverse birth outcomes: an updated systematic review based on cohort studies. Environ Sci Pollut Res Int. 2019;26:13963–13983.
6. Li X, Huang S, Jiao A, et al. Association between ambient fine particulate matter and preterm birth or term low birth weight: an updated systematic review and meta-analysis. Environ Pollut. 2017;227:596–605.
7. Chen C, Li H, Niu Y, et al. Impact of short-term exposure to fine particulate matter air pollution on urinary metabolome: a randomized, double-blind, crossover trial. Environ Int. 2019;130:104878.
8. Huang Q, Hu D, Wang X, et al. The modification of indoor PM2.5 exposure to chronic obstructive pulmonary disease in Chinese elderly people: a meet-in-metabolite analysis. Environ Int. 2018;121(Pt 2):1243–1252.
9. Ladva CN, Golan R, Liang D, et al. Particulate metal exposures induce plasma metabolome changes in a commuter panel study. PLoS One. 2018;13:e0203468.
10. Li H, Cai J, Chen R, et al. Particulate matter exposure and stress hormone levels: a randomized, double-blind, crossover trial of air purification. Circulation. 2017;136:618–627.
11. Li Z, Liang D, Ye D, et al. Application of high-resolution metabolomics to identify biological pathways perturbed by traffic-related air pollution. Environ Res. 2021;193:110506.
12. Liang D, Moutinho JL, Golan R, et al. Use of high-resolution metabolomics for the identification of metabolic signals associated with traffic-related air pollution. Environ Int. 2018;120:145–154.
13. Liang D, Ladva CN, Golan R, et al. Perturbations of the arginine metabolome following exposures to traffic-related air pollution in a panel of commuters with and without asthma. Environ Int. 2019;127:503–513.
14. Mu L, Niu Z, Blair RH, et al. Metabolomics profiling before, during, and after the Beijing olympics: a panel study of within-individual differences during periods of high and low air pollution. Environ Health Perspect. 2019;127:57010.
15. Vlaanderen JJ, Janssen NA, Hoek G, et al. The impact of ambient air pollution on the human blood metabolome. Environ Res. 2017;156:341–348.
16. Walker DI, Lane KJ, Liu K, et al. Metabolomic assessment of exposure to near-highway ultrafine particles. J Expo Sci Environ Epidemiol. 2019;29:469–483.
17. Yan Q, Liew Z, Uppal K, et al. Maternal serum metabolome and traffic-related air pollution exposure in pregnancy. Environ Int. 2019;130:104872.
18. Nassan FL, Kelly RS, Kosheleva A, et al. Metabolomic signatures of the long-term exposure to air pollution and temperature. Environ Health. 2021;20:3.
19. Kloog I, Coull BA, Zanobetti A, Koutrakis P, Schwartz JD. Acute and chronic effects of particles on hospital admissions in New-England. PLoS One. 2012;7:e34664.
20. Shi L, Zanobetti A, Kloog I, et al. Low-Concentration PM2.5 and mortality: estimating acute and chronic effects in a population-based study. Environ Health Perspect. 2016;124:46–52.
21. Guan T, Xue T, Gao S, et al. Acute and chronic effects of ambient fine particulate matter on preterm births in Beijing, China: a time-series model. Sci Total Environ. 2019;650(Pt 2):1671–1677.
22. Li Z, Tang Y, Song X, Lazar L, Li Z, Zhao J. Impact of ambient PM2.5 on adverse birth outcome and potential molecular mechanism. Ecotoxicol Environ Saf. 2019;169:248–254.
23. Martens DS, Gouveia S, Madhloum N, et al. Neonatal cord blood oxylipins and exposure to particulate matter in the early-life environment: an ENVIRONAGE birth cohort study. Environ Health Perspect. 2017;125:691–698.
24. Messerlian C, Williams PL, Ford JB, et al.; EARTH Study Team. The Environment and Reproductive Health (EARTH) Study: a prospective preconception cohort. Hum Reprod Open. 2018;2018:hoy001.
25. Gaskins AJ, Fong KC, Abu Awad Y, et al. Time-varying exposure to air pollution and outcomes of in vitro fertilization among couples from a fertility clinic. Environ Health Perspect. 2019;127:77002.
26. Kloog I, Chudnovsky AA, Just AC, et al. A new hybrid spatio-temporal model for estimating daily multi-year PM2.5 concentrations across northeastern USA using high resolution aerosol optical depth data. Atmos Environ (1994). 2014;95:581–590.
27. Go YM, Walker DI, Liang Y, et al. Reference standardization for mass spectrometry and high-resolution metabolomics applications to exposome research. Toxicol Sci. 2015;148:531–543.
28. Liu KH, Nellis M, Uppal K, et al. Reference Standardization for quantification and harmonization of large-scale metabolomics. Anal Chem. 2020;92:8836–8844.
29. Chang CJ, Barr DB, Ryan PB, et al. Per- and polyfluoroalkyl substance (PFAS) exposure, maternal metabolomic perturbation, and fetal growth in African American women: a meet-in-the-middle approach. Environ Int. 2022;158:106964.
30. Simón-Manso Y, Lowenthal MS, Kilpatrick LE, et al. Metabolite profiling of a NIST Standard Reference Material for human plasma (SRM 1950): GC-MS, LC-MS, NMR, and clinical laboratory analyses, libraries, and web-based resources. Anal Chem. 2013;85:11725–11731.
31. Chambers MC, Maclean B, Burke R, et al. A cross-platform toolkit for mass spectrometry and proteomics. Nat Biotechnol. 2012;30:918–920.
32. Yu T, Park Y, Johnson JM, Jones DP. apLCMS–adaptive processing of high-resolution LC/MS data. Bioinformatics. 2009;25:1930–1936.
33. Uppal K, Soltow QA, Strobel FH, et al. xMSanalyzer: automated pipeline for improved feature detection and downstream analysis of large-scale, non-targeted metabolomics data. BMC Bioinformatics. 2013;14:15.
34. Group PC. Prism data set: recent years (Jan 1981-Nov 2018). In:2019.
35. Li S, Park Y, Duraisingham S, et al. Predicting network activity from high throughput metabolomics. PLoS Comput Biol. 2013;9:e1003123.
36. Sumner LW, Amberg A, Barrett D, et al. Proposed minimum reporting standards for chemical analysis Chemical Analysis Working Group (CAWG) Metabolomics Standards Initiative (MSI). Metabolomics. 2007;3:211–221.
37. Le Floc’h N, Otten W, Merlot E. Tryptophan metabolism, from nutrition to potential therapeutic applications. Amino Acids. 2011;41:1195–1205.
38. Nie C, He T, Zhang W, Zhang G, Ma X. Branched Chain amino acids: beyond nutrition metabolism. Int J Mol Sci. 2018;19:E954.
39. Brosnan ME, Brosnan JT. Histidine metabolism and function. J Nutr. 2020;150(Suppl 1):2570S–2575S.
40. Moro J, Tomé D, Schmidely P, Demersay TC, Azzout-Marniche D. Histidine: a systematic review on metabolism and physiological effects in human and different animal species. Nutrients. 2020;12:E1414.
41. Zhenyukh O, Civantos E, Ruiz-Ortega M, et al. High concentration of branched-chain amino acids promotes oxidative stress, inflammation and migration of human peripheral blood mononuclear cells via mTORC1 activation. Free Radic Biol Med. 2017;104:165–177.
42. Wang Q, Liu D, Song P, Zou MH. Tryptophan-kynurenine pathway is dysregulated in inflammation, and immune activation. Front Biosci (Landmark Ed). 2015;20:1116–1143.
43. Jeong A, Fiorito G, Keski-Rahkonen P, et al.; EXPOsOMICS Consortium. Perturbation of metabolic pathways mediates the association of air pollutants with asthma and cardiovascular diseases. Environ Int. 2018;119:334–345.
44. Vyklicky V, Korinek M, Smejkalova T, et al. Structure, function, and pharmacology of NMDA receptor channels. Physiol Res. 2014;63(Suppl 1):S191–S203.
45. Sheridan P, Ilango S, Bruckner TA, Wang Q, Basu R, Benmarhnia T. Ambient fine particulate matter and preterm birth in California: identification of critical exposure windows. Am J Epidemiol. 2019;188:1608–1615.
46. Liu C, Sun J, Liu Y, et al. Different exposure levels of fine particulate matter and preterm birth: a meta-analysis based on cohort studies. Environ Sci Pollut Res Int. 2017;24:17976–17984.
47. Grippo A, Zhang J, Chu L, et al. Air pollution exposure during pregnancy and spontaneous abortion and stillbirth. Rev Environ Health. 2018;33:247–264.
48. Stevens VL, Hoover E, Wang Y, Zanetti KA. Pre-Analytical factors that affect metabolite stability in human urine, plasma, and serum: a review. Metabolites. 2019;9:E156.
Keywords:

amino acid metabolism; anti-inflammatory; energy metabolism; lipid metabolism; PM2.5; pro-inflammatory; untargeted metabolomics

Supplemental Digital Content

Copyright © 2022 The Authors. Published by Wolters Kluwer Health, Inc. on behalf of The Environmental Epidemiology. All rights reserved.